Hadis Morkoc¸ Handbook of Nitride Semiconductors and Devices
Related Titles Piprek, J. (ed.)
Nitride Semiconductor De...
82 downloads
1478 Views
15MB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
Hadis Morkoc¸ Handbook of Nitride Semiconductors and Devices
Related Titles Piprek, J. (ed.)
Nitride Semiconductor Devices: Principles and Simulation 2007 ISBN: 978-3-527-40667-8
Adachi, S.
Properties of Group-IV, III-V and II-VI Semiconductors 2005 ISBN: 978-0-470-09032-9
Ruterana, P., Albrecht, M., Neugebauer, J. (eds.)
Nitride Semiconductors Handbook on Materials and Devices 2003 ISBN: 978-3-527-40387-5
Ng, K. K.
Complete Guide to Semiconductor Devices 2002 ISBN: 978-0-471-20240-0
Hadis Morkoç
Handbook of Nitride Semiconductors and Devices Vol. 1: Materials Properties, Physics and Growth
The Author Prof. Dr. Hadis Morkoç Virginia Commonwealth University Dept. of Chemical Enigineering Richmond, VA USA Cover SPIESZDESIGN GbR, Neu-Ulm, Germany
All books published by Wiley-VCH are carefully produced. Nevertheless, authors, editors, and publisher do not warrant the information contained in these books, including this book, to be free of errors. Readers are advised to keep in mind that statements, data, illustrations, procedural details or other items may inadvertently be inaccurate. Library of Congress Card No.: applied for British Library Cataloguing-in-Publication Data A catalogue record for this book is available from the British Library. Bibliographic information published by the Deutsche Nationalbibliothek Die Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available in the Internet at . # 2008 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim All rights reserved (including those of translation into other languages). No part of this book may be reproduced in any form – by photoprinting, microfilm, or any other means – nor transmitted or translated into a machine language without written permission from the publishers. Registered names, trademarks, etc. used in this book, even when not specifically marked as such, are not to be considered unprotected by law. Typesetting Thomson Digital, Noida, India Printing betz-druck GmbH, Darmstadt Binding Litges & Dopf GmbH, Heppenheim Printed in the Federal Republic of Germany Printed on acid-free paper ISBN: 978-3-527-40837-5
V
Contents Preface
XIII
Color Tables XXI
1 1.1 1.2 1.2.1 1.2.2 1.2.3 1.3 1.3.1 1.3.2 1.3.3 1.3.4 1.4 1.4.1 1.4.2 1.4.3 1.4.4 1.4.5 1.5 1.5.1 1.5.2 1.5.3 1.5.4 1.5.5
General Properties of Nitrides 1 Introduction 1 Crystal Structure of Nitrides 1 Gallium Nitride 30 Chemical Properties of GaN 35 Mechanical Properties of GaN 35 Thermal Properties of GaN 47 Aluminum Nitride 62 Mechanical Properties of AlN 62 Thermal and Chemical Properties of AlN Electrical Properties of AlN 69 Brief Optical Properties of AlN 71 Indium Nitride 75 Crystal Structure of InN 77 Mechanical Properties of InN 77 Thermal Properties of InN 79 Brief Electrical Properties of InN 81 Brief Optical Properties of InN 84 Ternary and Quaternary Alloys 89 AlGaN Alloy 90 InGaN Alloy 92 InAlN Alloy 97 InAlGaN Quaternary Alloy 99 Dilute GaAs(N) 105 References 110
Handbook of Nitride Semiconductors and Devices. Vol. 1. Hadis Morkoç Copyright # 2008 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim ISBN: 978-3-527-40837-5
66
VI
Contents
2 2.1 2.1.1 2.1.2 2.1.3 2.1.4 2.1.5 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.8.1 2.8.2 2.9 2.9.1 2.9.2 2.10 2.10.1 2.10.2 2.10.3 2.10.4 2.10.5 2.10.6 2.10.7 2.10.8 2.11 2.11.1 2.11.2 2.11.3 2.12 2.12.1 2.12.2 2.12.3 2.12.3.1 2.12.3.2 2.12.3.3 2.12.4 2.12.4.1 2.12.4.2 2.12.5
Electronic Band Structure and Polarization Effects 131 Introduction 131 Band Structure Calculations 132 Plane Wave Expansion Method 134 Orthogonalized Plane Wave (OPW) Method 134 Pseudopotential Method 135 Augmented Plane Wave (APW) Method 135 Other Methods and a Review Pertinent to GaN 136 General Strain Considerations 154 Effect of Strain on the Band Structure of GaN 159 kp Theory and the Quasi-Cubic Model 160 Quasi-Cubic Approximation 167 Temperature Dependence of Wurtzite GaN Bandgap 169 Sphalerite (Zinc blende) GaN 172 AlN 176 Wurtzite AlN 177 Zinc Blende AlN 183 InN 184 Wurtzitic InN 185 Zinc Blende InN 200 Band Parameters for Dilute Nitrides 202 GaAsN 205 InAsN 208 InPN 209 InSbN 209 GaPN 210 GaInAsN 210 GaInPN 212 GaAsSbN 212 Confined States 212 Conduction Band 216 Valence Band 224 Exciton Binding Energy in Quantum Wells 227 Polarization Effects 230 Piezoelectric Polarization 236 Spontaneous Polarization 241 Nonlinearity of Polarization 245 Origin of the Nonlinearity 250 Nonlinearities in Spontaneous Polarization 253 Nonlinearities in Piezoelectric Polarization 256 Polarization in Heterostructures 264 Ga-Polarity Single AlGaN/GaN Interface 272 Ga-Polarity Single AlxIn1xN/GaN Interface 276 Polarization in Quantum Wells 278
Contents
2.12.5.1 2.12.5.2 2.12.6 2.12.7
Nonlinear Polarization in Quantum Wells 280 InGaN/GaN Quantum Wells 286 Effect of Dislocations on Piezoelectric Polarization Thermal Mismatch Induced Strain 290 References 299
3
Growth and Growth Methods for Nitride Semiconductors 323 Introduction 323 Substrates for Nitride Epitaxy 324 Conventional Substrates 326 Compliant Substrates 327 van der Waals Substrates 328 A Primer on Conventional Substrates and their Preparation for Growth 329 GaAs 329 A Primer on GaAs 330 Surface Preparation of GaAs for Epitaxy 331 Si 332 A Primer on Si 332 Surface Preparation of Si for Epitaxy 333 SiC 334 A Primer on SiC 334 Surface Preparation of SiC for Epitaxy 338 Sapphire 342 A Primer on Sapphire 343 Surface Preparation of Sapphire for Epitaxy 346 ZnO 350 A Primer on ZnO 351 Substrate Preparation for Epitaxy 353 LiGaO2 and LiAlO2 355 LiGaO2 Substrates 355 LiAlO2 Substrates 358 AlN and GaN 359 Seedless Growth of GaN 362 Seedless Growth of GaN by High Nitrogen Pressure Solution Growth (HNPSG) for Substrates 362 Seeded Growth of GaN by HNPSG Method for Substrates 363 Pertinent Surfaces of GaN 365 GaN Surface Preparation for Epitaxy 369 Other Substrates 371 GaN Epitaxial Relationship to Substrates 372 Epitaxial Relationship of GaN and AlN with Sapphire 372 Epitaxial Relationship of GaN and AlN with SiC 381 Epitaxial Relationship of GaN and AlN with Si 381 Epitaxial Relationship of GaN with ZnO 381
3.1 3.1.1 3.1.2 3.1.3 3.2 3.2.1 3.2.1.1 3.2.1.2 3.2.2 3.2.2.1 3.2.2.2 3.2.3 3.2.3.1 3.2.3.2 3.2.4 3.2.4.1 3.2.4.2 3.2.5 3.2.5.1 3.2.5.2 3.2.6 3.2.6.1 3.2.6.2 3.2.7 3.2.7.1 3.2.7.1.1 3.2.7.1.2 3.2.7.2 3.2.7.3 3.2.8 3.3 3.3.1 3.3.2 3.3.3 3.3.4
289
VII
VIII
Contents
3.3.5 3.4 3.4.1 3.4.1.1 3.4.1.2 3.4.1.3 3.4.1.3.1 3.4.1.3.2 3.4.1.3.3 3.4.1.3.4 3.4.1.3.5 3.4.1.3.6 3.4.2 3.4.2.1 3.4.2.2 3.4.2.3 3.4.2.4 3.4.2.5 3.4.2.6 3.4.2.7 3.4.2.8 3.4.2.9 3.4.2.9.1 3.4.2.9.2 3.4.2.9.3 3.4.2.9.4 3.5 3.5.1 3.5.1.1 3.5.1.2 3.5.2 3.5.3 3.5.3.1 3.5.3.2 3.5.3.2.1 3.5.3.2.2 3.5.3.2.3 3.5.3.2.4 3.5.3.2.5 3.5.3.3 3.5.3.4
Epitaxial Relationship of GaN with LiGaO2 and LiAlO2 and Perovskites 382 Nitride Growth Techniques 384 Vapor Phase Epitaxy 385 Hydride Vapor Phase Epitaxy 385 Organometalic Vapor Phase Epitaxy 393 Modeling of OMVPE Growth 398 Thermal Decomposition of GaN as it Relates to Growth 398 Ga and N Precursor Adsorption and Decomposition 400 Ga and N2 Desorption from the Surface 401 Ga and N Surface Diffusion 403 Kinetic Model: Balance Between Adsorption and Desorption 405 Comparison of Model with Growth Conditions for Surface Morphology 406 Molecular Beam Epitaxy 409 Adsorption 411 Desorption 412 Surface Diffusion 413 Incorporation 415 Decomposition 416 Reflection High-Energy Electron Diffraction 417 Plasma-Assisted MBE (PAMBE) or RF MBE, Primarily N Source 428 Reactive Ion MBE 435 Principles of RMBE and PAMBE Growth 436 Growth by RMBE 437 Growth by PAMBE 446 Which Species of N is Desirable? 451 The Effect of III/V Ratio and Substrate Temperature on Surface Morphology 455 The Art and Technology of Growth of Nitrides 462 Sources 467 HVPE Buffer Layers and Laser Liftoff 468 Benchmark HVPE Layers/Templates 472 Growth on GaAs Substrates 477 Growth on SiC: Nucleation Layers and GaN 479 Stacking and Interfacial Relationship 480 Nucleation Layers on SiC 482 High-Temperature AlN Nucleation Layers on SiC 486 Low-Temperature GaN Nucleation Layers on SiC 488 High-Temperature GaN Nucleation Layers on SiC 489 Alloy and Multiple Layer Nucleation Layers on SiC 491 Nucleation Layers on SiC by MBE 492 Substrate Misorientation and Domain Boundaries 495 Polarity of GaN on SiC 498
Contents
3.5.3.4.1 3.5.3.5 3.5.3.6 3.5.4 3.5.5 3.5.5.1 3.5.5.1.1 3.5.5.1.2 3.5.5.1.3 3.5.5.2 3.5.5.2.1 3.5.5.2.2 3.5.5.2.3 3.5.5.2.4 3.5.5.2.5 3.5.5.2.6 3.5.5.3 3.5.5.3.1 3.5.5.4 3.5.5.5 3.5.6 3.5.6.1 3.5.7 3.5.8 3.5.9 3.5.10 3.5.11 3.5.11.1 3.5.11.2 3.5.11.3 3.5.12 3.5.13 3.5.14 3.5.14.1 3.5.15 3.5.15.1 3.5.15.1.1 3.5.15.1.2 3.5.15.2 3.5.15.2.1 3.5.15.2.2 3.5.15.2.3 3.5.15.3
GaN Growth on SiC 499 Growth on Porous SiC 503 Zinc Blende Phase Growth 507 Growth on Si 507 Growth on Sapphire 512 OMVPE Low-Temperature Nucleation Buffer Layers 513 The Effect of V/III Ratio on Nucleation Buffer Layer 523 Effect of Epitaxial Growth Temperature 525 Effect of Process Pressure 525 Epitaxial Lateral Overgrowth 528 Selective Epitaxial Growth and Lateral Epitaxial Overgrowth with HVPE 537 Lateral Epitaxial Overgrowth on Si 539 Pendeo-Epitaxy 540 Pendeo-Epitaxy on SiC Substrates 542 Pendeo-Epitaxy on Silicon Substrates 544 Point Defect Distribution in ELO Grown GaN 558 Nanoheteroepitaxy and Nano-ELO 564 SiN and TiN Nanonets 569 Selective Growth Using W Masks 583 Low-Temperature Buffer Interlayer 584 Polarity and Surface Structure of GaN Layers, Particularly on Sapphire 586 MBE Buffer Layers 597 Growth on ZnO Substrates 598 Growth on LiGaO2 and LiAlO2 Substrates 603 Growth on GaN Templates 605 Growth on Spinel (MgAl2O4) 611 Growth on Non c-Plane Substrates 611 The a-Plane GaN Growth 613 Epitaxial Lateral Overgrowth of a-plane GaN 616 00) m-Plane GaN Growth 623 The (11 Growth of p-Type GaN 627 Growth of InN 629 Growth of AlN 638 Surface Reconstruction of AlN 642 Growth of Ternary and Quaternary Alloys 652 Growth of AlGaN 653 Growth of p-Type AlGaN 666 Ordering in AlGaN 668 Growth of InGaN 671 Doping of InGaN 678 Phase Separation in InGaN 679 Surface Reconstruction of InGaN 689 Growth of AlInN 695
IX
X
Contents
3.5.15.3.1 3.5.15.4 3.5.16 3.5.16.1 3.5.16.2 3.5.16.3 3.5.16.4 3.5.16.4.1 3.5.16.4.2 3.5.16.5 3.5.16.5.1 3.5.16.5.2 3.6
Miscibility Gap in InAlN 697 InGaAlN Quaternary Alloy 699 Growth of Quantum Dots 706 Quantum Dots by MBE 711 Quantum Dots by OMVPE 719 Quantum Dots by Other Techniques 723 Preparation and Properties of Nanostructures 725 Approaches for Synthesis 726 Vapor Phase Growth 726 Nanowires and Longitudinal Heterostructures 737 Coaxial Heterostructures 755 Nanotubes 756 Concluding Remarks 759 References 760
4
Extended and Point Defects, Doping, and Magnetism 817 Introduction 817 A Primer on Extended Defects 818 Dislocations 819 Misfit Dislocations 822 Threading Dislocations 822 Edge Dislocations 824 Screw Dislocations 828 Mixed Dislocations 836 Nanopipes or Hollow Pipes 840 Planar Defects: Domain Boundaries 844 Stacking Faults 851 Grain Boundaries 862 Electronic Structure of Extended Defects 863 Open Core Versus Close Core in Screw Dislocations 865 Edge and Mixed Dislocations 866 Simple Stacking Faults: Electrical Nature 884 TEM Analysis of High Nitrogen Pressure (HNP) Solution Growth (HNPSG) and HVPE-Grown GaN 886 Extended Defect Characterization 887 Pyramidal Defects 894 V-Shaped Defects (Pits) in InGaN Multiple Quantum Wells (MQWs) 905 Structural Defect Analysis by Chemical Etch Delineation 905 Structural Defect Observations with Surface Probes 910 Point Defects and Autodoping 917 Theoretical Studies of Point Defects in GaN 919 Hydrogen and Impurity Trapping at Extended Defects 924 Vacancies, Antisites, Interstitials, and Complexes 928 Vacancies 929
4.1 4.1.1 4.1.1.1 4.1.1.2 4.1.1.2.1 4.1.1.2.2 4.1.1.2.3 4.1.2 4.1.3 4.1.4 4.1.5 4.1.6 4.1.6.1 4.1.6.2 4.1.6.3 4.2 4.2.1 4.2.2 4.2.3 4.2.4 4.2.5 4.3 4.3.1 4.3.1.1 4.3.1.2 4.3.1.2.1
Contents
4.3.1.2.2 4.3.2 4.3.2.1 4.3.2.2 4.3.2.3 4.3.2.4 4.4 4.4.1 4.4.2 4.4.3 4.4.4 4.5 4.6 4.6.1 4.6.2 4.7 4.8 4.9 4.9.1.1 4.9.1.2 4.9.1.3 4.9.2 4.9.2.1 4.9.2.2 4.9.2.3 4.9.2.4 4.9.3 4.9.3.1 4.9.3.1.1 4.9.3.1.2 4.9.3.1.3 4.9.3.2 4.9.3.3 4.9.3.4 4.9.3.5 4.9.3.6 4.9.3.7 4.9.3.8 4.9.4 4.9.4.1 4.9.4.2 4.9.5 4.9.6
Interstitials and Antisite Defects 934 Complexes 935 Shallow Donor – Gallium Vacancy Complexes 936 Shallow Acceptor – Nitrogen Vacancy Complexes 936 Hydrogen-Related Complexes 937 Other Complexes 938 Defect Analysis by Deep-Level Transient Spectroscopy 938 Basics of DLTS 939 Applications of DLTS to GaN 948 Dispersion in DLTS of GaN 970 Applications of DLTS to AlGaN, In-Doped AlGaN, and InAlN 977 Minority Carrier Lifetime 977 Positron Annihilation 982 Vacancy Defects and Doping in Epitaxial GaN 983 Growth Kinetics and Thermal Behavior of Vacancy Defects in GaN 993 Fourier Transform Infrared (FTIR), Electron Paramagnetic Resonance, and Optical Detection of Magnetic Resonance 998 Role of Hydrogen 1004 Intentional Doping 1006 Shallow Donors 1007 Substitutional Acceptors 1007 Isoelectronic Impurities 1009 n-Type Doping with Silicon, Germanium, Selenium, and Oxygen 1010 Si Doping 1010 Ge Doping 1011 Se Doping 1012 p-Type Doping 1013 p-Type Doping and Codoping with Donors and Acceptors 1014 Magnesium Doping 1014 Codoping for Improving p-Type Conductivity 1019 Use of Superlattices for Improving p-Type Conductivity 1032 Role of Hydrogen and Defects in Mg-Doped GaN 1034 Beryllium Doping 1038 Mercury Doping 1040 Carbon Doping 1040 Zinc Doping 1042 Calcium Doping 1042 Cadmium Doping 1043 Other Acceptors in GaN 1043 Doping with Isoelectronic Impurities 1044 Arsenic Doping 1044 Phosphorus Doping 1045 Doping with Rare Earths 1045 Doping with Transition Metals and Rare Earths 1046
XI
XII
Contents
4.9.6.1 4.9.6.2 4.9.6.3 4.9.6.4 4.9.6.5 4.9.6.5.1 4.9.6.5.2 4.9.6.5.3 4.9.6.6 4.9.6.6.1 4.9.6.7 4.9.6.7.1 4.9.6.7.2 4.9.7 4.9.7.1 4.9.7.2 4.9.7.3 4.9.8 4.9.9 4.9.10 4.9.11 4.9.12 4.10 4.11
Manganese Doping for Electronic Properties 1046 Other TM Doping for Electronic Properties 1060 General Remarks About Dilute Magnetic Semiconductors 1063 General Remarks About Spintronics 1075 Theoretical Aspects of Dilute Magnetic Semiconductor 1082 Carrier – Single Magnetic Ion Interaction 1084 Interaction Between Magnetic Ions 1086 Zener, Mean Field, RKKY, and Ab Initio Treatments 1089 A Primer to Magnetotransport Measurements 1103 Faraday Rotation, Kerr Effect, and Magnetic Circular Dichroism (MCD) 1104 II–VI and GaAs-Based Dilute Magnetic Semiconductors 1123 II–VI-Based Dilute Magnetic Semiconductors 1124 III–V-Based DMS: (GaMn)As 1133 Experimental Results of TM-Doped GaN 1141 Magnetotransport Properties TM-Doped GaN 1141 Magnetic Properties of Mn-Doped GaN 1143 Magneto-Optical Measurements in TM-Doped GaN 1146 Magnetic, Structural, Optical, and Electrical Properties of Cr-Doped GaN 1156 Other TM and Rare Earth Doped GaN:(Co, Fe, V, Gd, and so on) 1163 TM-Doped Nanostructures 1166 Applications of Ferromagnetism and Representative Devices 1167 Summarizing Comments on Ferromagnetism 1186 Ion Implantation and Diffusion for Doping 1188 Summary 1190 References 1191 Index
1231
Appendix
1257
XIII
Preface This three-volume handbook represents the only comprehensive treatise on semiconductor and device fundamentals and technology under the overall umbrella of wide bandgap nitride semiconductors with comparison to GaAs when applicable. As it stands, the book is a reference book, a handbook, and a graduate text book all in one and would be beneficial to second-year graduate students majoring in semiconductor materials and devices, graduate research assistants conducting research in wide bandgap semiconductors, researchers and technologists, faculty members, program monitors, and managers. The philosophy of this endeavor is to present the material as much clearly and comprehensively as possible, so that there is very little need to refer to other sources to get a grasp of the subjects covered. Extreme effort has been expended to ensure that concepts and problems are treated starting with their fundamental basis so that the reader is not left hanging in thin air. Even though the treatise deals with GaN and related materials, the concepts and methods discussed are applicable to any semiconductor. The philosophy behind Nitride Semiconductors and Devices was to provide an adequate treatment of nitride semiconductors and devices as of 1997 to be quickly followed by a more complete treatment. As such, Nitride Semiconductors and Devices did not provide much of the background material for the reader and left many issues unanswered in part because they were not yet clear to the research community at that time. Since then, tremendous progress both in the science and engineering of nitrides and devices based on them has been made. While LEDs and lasers were progressing well even during the period when Nitride Semiconductors and Devices was written, tremendous progress has been made in FETs and detectors in addition to LEDs and lasers since then. LEDs went from display devices to illuminants for lighting of all kinds. Lasers are being implemented in the third generation of DVDs. The power amplifiers are producing several hundred watts of RF power per chip and the detectors and detector arrays operative in the solar-blind region of the spectrum have shown detectivities rivaling photomultiplier tubes. The bandgap of InN has been clarified which now stands near 0.7 eV. Nanostructures, which did not exist
Handbook of Nitride Semiconductors and Devices. Vol. 1. Hadis Morkoç Copyright # 2008 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim ISBN: 978-3-527-40837-5
XIV
Preface
during the period covered by Nitride Semiconductors and Devices, have since become available. The technological breakthroughs such as epitaxial lateral overgrowth, laser liftoff, and freestanding GaN were either not fully developed or did not exist, neither did the highly improved quantum structures and devices based on them. In the interim period since then, the surfaces of nitrides and substrate materials, point defects and doping, magnetic ion doping, processing, current conduction mechanisms, and optical processes in bulk and quantum structures have been more clearly understood and many misconceptions (particularly, those dealing with polarization) identified, removed and/or elucidated. The handbook takes advantage of the fundamental and technological developments for a thorough treatment of all aspects of nitride semiconductors. In addition, the fundamentals of materials physics and device physics that are provided are applicable to other semiconductors, particularly, wurtzitic direct bandgap semiconductors. The handbook presents a thorough treatment of the science, fundamentals, and technology of nitride semiconductors and devices in such a width and depth that the reader would seldom need to engage in time-consuming exploration of the literature to fill in gaps. Last but not the least, the handbook contains seamless treatments of fundamentals needed or relied on throughout the entire book. The following is a succinct odyssey through the content of the three-volume handbook. Volume 1, Chapter 1 discusses the properties of nitride-based semiconductors with plenty of tables for reference. Volume 1, Chapter 2 treats the band structure of III–V nitrides, theories applied to determining the band structure, features of each theory with a succinct discussion of each, band structure of dilute III–V semiconductors doped with N, strain and stress, deformation potentials, and in-depth discussion of piezo and spontaneous polarization with illustrative and instructive artwork. Volume 1, Chapter 3 encompasses substrates that have been and are used for growth of nitride semiconductors, mainly, structural and mechanical (thermal) properties of those substrates, surface structure of planes used for growth, and substrate preparation for growth. Orientation and properties of GaN grown on those substrates are discussed along with commonly used surface orientations of GaN. The discussion is laced with highly illustrative and illuminating images showing orientations of GaN resulting through growth on c-plane, a-plane, m-plane, and r-plane substrates whichever applicable and the properties of resulting layers provided. The treatment segues into the discussion of various growth methods used for nitrides taking into account the fundamentals of growth including the applicable surface-oriented processes, kinetics, and so on, involved. A good deal of growth details for both OMVPE and MBE, particularly, the latter including the fundamentals of in situ process monitoring instrumentation such as RHEED, and dynamics of growth processes occurring at the surface of the growing layer are given. Of paramount interest is the epitaxial lateral overgrowth (ELO) for defect reduction. In addition to standard single multistep ELO, highly attractive nanonetwork meshes used for ELO are also discussed. Specifics in terms of growth of binary, ternary, and quaternaries of nitride semiconductors are discussed. Finally, the methods used to grow nanoscale structures are treated in sufficient detail.
Preface
Volume 1, Chapter 4 focuses on defects, both extended and point, doping for conductivity modulation and also for rendering the semiconductor potentially ferromagnetic segueing into electrical, optical, and magnetic properties resulting in films, with sufficient background physics provided to grasp the material. A clear discussion of extended defects, including line defects, are discussed with a plethora of illustrative schematics and TEM images for an easy comprehension by anyone with solid-state physics background. An in-depth and comprehensive treatment of the electrical nature of extended defects is provided for a full understanding of the scope and effect of extended defects in nitride semiconductors, the basics of which can be applicable to other hexagonal materials. The point defects such as vacancies, antisites, and complexes are then discussed along with a discussion of the effect of H. This gives way to the methods used to analyze point defects such as deep level transient spectroscopy, carrier lifetime as pertained to defects, positron annihilation, Fourier transform IR, electron paramagnetic resonance, and optical detection of magnetic resonance and their application to nitride semiconductors. This is followed by an extensive discussion of n-type and p-type doping in GaN and related materials and developments chronicled when applicable. An in-depth treatment of triumphs and challenges along with codoping and other methods employed for achieving enhanced doping and the applicable theory has been provided. In addition, localization effects caused by heavy p-type doping are discussed. This gives way to doping of, mainly, GaN with transition elements with a good deal of optical properties encompassing internal transition energies related to ion and perturbations caused by crystal field in wurtzitic symmetry. To get the reader conditioned for ferromagnetism, a sufficient discussion of magnetism, ferromagnetism, and measurement techniques (magnetic, magneto transport, magneto optics with underlying theory) applied to discern such properties are given. This is followed by an indepth and often critical discussion of magnetic ion and rare earth-doped GaN, as well as of spintronics, often accompanied by examples for materials properties and devices from well-established ferromagnetic semiconductors such as Mn-doped GaN and Cr-doped ZnTe. Volume 2, Chapter 1 treats metal semiconductor structures and fabrication methods used for nitride-based devices. Following a comprehensive discussion of current conduction mechanisms in metal semiconductor contacts, which are applicable to any metal semiconductor system, specific applications to metal-GaN contacts are treated along with the theoretical analysis. This gives way to a discussion of ohmic contacts, their technology, and their characterization. In particular, an ample discussion of the determination of ohmic contact resistivity is provided. Then etching methods, both dry (plasma) and wet, photochemical, process damage, and implant isolation are discussed. Volume 2, Chapter 2 deals with determination of impurity and carrier concentrations and mobility mainly by temperature-dependent electrical measurements, such as Hall measurements. Charge balance equations, capacitance voltage measurements, and their intricacies are treated and used for nitride semiconductors, as well as a good deal of discussion of often brushed off degeneracy factors.
XV
XVI
Preface
Volume 2, Chapter 3 is perhaps one of the most comprehensive discussions of carrier transport in semiconductors with applications to GaN. After a discussion of scattering processes in physical and associated mathematical terms, the methods discussed are applied to GaN and other related binaries and ternaries with useful ranges of doping levels, compositions, and lattice temperatures. Comparisons with other semiconductors are also provided when applicable. This discussion segues into the discussion of carrier transport at high electric fields applicable to field-effect transistors, avalanche and pin (biased) photodiodes. This is followed by the measurement of mobility and associated details, which are often neglected in text and reference books. The discussion then flows into magnetotransport beyond that present in Hall measurements. Low, medium, and high magnetic field cases, albeit only normal to the surface of the epitaxial layers, determination of which is affected by the value of the mobility and various cases are treated. The treatise also includes cases where the relaxation time, if applicable, is energy-dependent and somewhat energy-independent. The discussion of the magnetotransport paves the way for a fundamental and reasonably extensive discussion of the Hall factor for each of the scattering mechanisms that often are not treated properly or are treated only in a cursory manner in many texts leading to confusion. After providing the necessary fundamental knowledge, the transport properties of GaN are discussed. This gives way to the discussion of various scattering mechanisms in two-dimensional systems that are relied on in high-performance FETs. For determining the mobility of each layer (in the case of multiple layer conduction), quantitative mobility spectrum analysis including both fundamentals and experimental data obtained in nitride semiconductors is discussed. The quantum Hall effect and fractional quantum Hall effect in general and as germane to GaN are discussed along with parameters such as the effective mass determined from such measurements. Volume 2, Chapter 4 is devoted to p–n junctions, beginning with the discussion of band lineups, particularly, in the binary pairs from the point of view of theoretically and experimentally measured values. Current conduction mechanisms, such as diffusion, generation-recombination, surface recombination, Poole–Frenkel, and hopping conductivity are discussed with sufficient detail. Avalanche multiplication, pertinent to the high-field region of FETs, and avalanche photodiodes, are discussedfollowed by discussions of the various homojunction and heterojunction diodes based on nitrides. Volume 2, Chapter 5 is perhaps the most comprehensive discussion of optical processes that can occur in a direct bandgap semiconductor and, in particular, in nitride-based semiconductors and heterostructures inclusive of 3, 2-, and 0-dimensional structures as well as optical nonlinearities. Following a treatment of photoluminescence basics, the discussion is opened up to the treatment of excitons, exciton polaritons, selection rules, and magneto-optical measurements followed by extrinsic transitions because of dopants/impurities and/or defects with energies ranging from the yellow and to the blue wavelength of the visible spectrum. Optical transitions in rare earth-doped GaN, optical properties of alloys, and quantum wells are then discussed with a good deal of depth, including localization effects and their possible sources particularly media containing InN. The discussion then leads to the
Preface
treatment of optical properties of quantum dots, intersubband transitions in GaNbased heterostructures, and, finally, the nonlinear optical properties in terms of second and third harmonic generation with illuminating graphics. Volume 3, Chapter 1 is devoted, in part, to the fundamentals of light emitting diodes, the perception of vision and color by human eye, methodologies used in conjunction with the chromaticity diagram and associated international standards in terms of color temperatures and color rendering index. Specific performances of various types of LEDs including UV varieties, current spreading or the lack of related specifics, analysis of the origin of transitions, and any effect of localization are discussed. A good deal of white light and lighting-related standards along with approaches employed by LED manufacturers to achieve white light for lighting applications is provided. The pertinent photon conversion schemes with sufficient specificity are also provided. Finally, the organic LEDs, as potential competitors for some applications of GaN-based LEDs are discussed in terms of fundamental processes that are in play and various approaches that are being explored for increased efficiency and operational lifetime. Volume 3, Chapter 2 focuses on lasers along with sufficient theory behind laser operation given. Following the primer to lasers along with an ample treatment based on Einsteins A and B coefficients and lasing condition, an analytical treatment of waveguiding followed by specifics for the GaN system and numerical simulations for determining the field distribution, loss, and gain cavity modes pertaining to semiconductor lasers are given. An ample fundamental treatment of spontaneous emission, stimulated emission, and absorptions and their interrelations in terms of Einsteins coefficients and occupation probabilities are given. This treatment segues into the extension of the gain discussion to a more realistic semiconductor with a complex valence band such as that of GaN. The results from numerical simulations of gain in GaN quantum wells are discussed, as well as various pathways for lasing such as electron-hole plasma and exciton-based pathways. Localization, which is very pervasive in semiconductors that are yet to be fully perfected, is discussed in the light of laser operation. Turning to experimental measurements, the method for gain measurement, use of various laser properties such as the delay on the onset of lasing with respect to the electrical pulse, dependence of laser threshold on cavity length to extract important parameters such as efficiency are discussed. The aforementioned discussions culminate in the treatment of performance of GaN-based lasers in the violet down to the UV region of the optical spectrum and applications of GaN-based lasers to DVDs along with a discussion of pertinent issues related to the density of storage. Volume 3, Chapter 3 treats field effect transistor fundamentals that are applicable to any semiconductor materials with points specific to GaN. The discussion primarily focuses on 2DEG channels formed at heterointerfaces and their use for FETs, including polarization effects. A succinct analytical model is provided for calculating the carrier densities at the interfaces for various scenarios and current voltage characteristics of FETs with several examples. Experimental performance of GaNbased FETs and amplifiers is then discussed followed by an in-depth analysis of anomalies in the current voltage characteristics owing to bulk and barrier states,
XVII
XVIII
Preface
including experimental methods and probes used for cataloging these anomalies. This is followed by the employment of field spreading gate plates and associated performance improvements. This segues into the discussion of noise both at the low-frequency end and high-frequency end with sufficient physics and practical approaches employed. The combined treatment of various low-frequency noise contributions as well as those at high frequencies along with their physical origin makes this treatment unique and provides an opportunity for those who are not specialists in noise to actually grasp the fundamentals and implications of low- and high-frequency noise. Discussion of high-power FETs would not be complete without a good discussion of heat dissipation and its physical pathways, which is made available. Unique to GaN is the awareness of the shortfall in the measured electron velocity as compared to the Monte Carlo simulation. Hot phonon effects responsible for this shortfall are uniquely discussed with sufficient theory and experimental data. A section devoted to reliability with specifics to GaN based high power HFETs is also provided. Finally, although GaN-based bipolar transistors are not all that attractive at this time, for completeness and the benefit of graduate students and others who are interested in such devices, the theory, mainly analytical, of the operation of heterojunction bipolar transistors is discussed along with available GaN based HBT data. Volume 3, Chapter 4 discusses optical detectors with special orientation toward UV and solar-blind detectors. Following a discussion of the fundamentals of photoconductive and photovoltaic detectors in terms of their photo response properties, a detailed discussion of the current voltage characteristic of the same, including all the possible current conduction mechanisms, is provided. Because noise and detectors are synonymous with each other, sources of the noise are discussed, followed by a discussion of quantum efficiency in photoconductors and p–n junction detectors. This is then followed by the discussion of vital characteristics such as responsivity and detectivity with an all too important treatment of the cases where the detectivity is limited by thermal noise, shot current noise, generation-recombination current noise, and background radiation limited noise (this is practically nonexistent in the solar-blind region except the man-made noise sources). A unique treatment of particulars associated with the detection in the UV and solar-blind region and requirements that must be satisfied by UV and solar-blind detectors, particularly, for the latter, is then provided. This leads the discussion to various UV detectors based on the GaN system, including the Si- and SiC-based ones for comparison. Among the nitride-based photodetectors, photoconductive variety as well as the metal-semiconductor, Schottky barrier, and homo- and heterojunction photodetectors are discussed along with their noise performance. Nearly solar-blind and truly solar-blind detectors including their design and performance are then discussed, which paves the way for the discussion of avalanche photodiodes based on GaN. Finally, the UV imagers using photodetectors arrays are treated. It is fair to state that I owe so much to so many, including my family members, friends, coworkers, colleagues, and those who contributed to the field of semiconductors in general and nitride semiconductors in particular, in my efforts to bring this manuscript to the service of readers. To this end, I thank my wife, Amy, and son,
Preface
Erol, for at least their understanding why I was not really there for them fully during the preparation of this manuscript, which took longer than most could ever realize. Also, without the support of VCU, with our Dean R. J. Mattauch, Assistant Dean Susan Younce, Department Chair A. Iyer, and my coworkers and students, it would not have been possible to pursue this endeavor. Special recognitions also go to Dr N. Izyumskaya for reading the entire manuscript for consistency in terms of figures, references, and so on, which had to have taken perseverance beyond that many could muster; Dr Ü. Özgür for being the bouncing board and proofing many parts of the book, particularly chapters dealing with optical processes, lasers and magnetism; my colleague P. Jena for reading and contributing to the band structure section; my coworker Professor M. Reshchikov for his contributions to the point defects and doping sections; Professor A. Baski for her expert assistance in obtaining microprobe images; Dr D. Huang for his many contributions to the quantum dots section; Dr Y-T Moon for his assistance in current crowding; C. Liu for her assistance with ferromagnetism; Prof. A. Teke for reading the chapter on detectors; Dr. R. Shimada for her contributions to the surface emitting laser section; Dr. J.-S. Lee for his help in updating the LED chapter; Dr Q. Wang for her help in generating the accurate ball and stick diagrams in Volume 1, Chapter 1; Dr V. Litvinov for calculating the energy levels in quantum wells; students Y. Fu, Q. Fan, X. Ni, and S. Chevtchenko for their contributions to various sections of the book with proofing equations, redoing calculations, and so on; and to J. Leach who took it upon himself to be the local expert in the latest in semiconductor and organic LEDs and helped with the chapter on LEDs and read the chapter on transport as well as proofread some of the other chapters and create the figures; Ms G. Esposito for reading a large portion of the text for English. Undergraduate students K. Ngandu, D. Lewis, B. D. Edmonds, and M. Mikkelson helped in reading various parts of the manuscript as well as helping with the artwork. Unbeknown to them, many graduate students who took classes from me helped in many immeasurable ways. Special thanks go to Professors R. M. Feenstra, A. Matulionis, A. Blumenau, P. Ruterana, G. P. Dimitrakopulos, P. Handel, K. T. Tsen, T. Yao, P. I. Cohen, S. Porowski, B. Monemar, B. Gil, P. Le Febvre, S. Chichibu, F. Tuomisto, C. Van de Walle, M. Schubert, F. Schubert, H. Temkin, S. Nikishin, L. Chernyak, J. Edgar, T. Myers, K. S. A. Butcher, O. Ambacher, V. Fiorentini, A. di Carlo, F. Bernardini, V. Fiorentini, M. Stutzmann, F. Pollak, C. Nguyen, S. Bedair, N. El-Masry, S. Fritsch, M. Grundman, J. Neugebauer, M. S. Shur, J. Bowers, J. C. Campbell, M. Razhegi, A. Nurmikko, M. A. Khan, J. Speck, S. Denbaars, R. J. Trew, A. Christon, G. Bilbro, H. Ohno, A. Hoffmann, B. Meyer, B. Wessels, N. Grandjean, and D. L. Rode; and Drs Z. Liliental-Weber, P. Klein, S. Binari, D. Koleske, J. Freitas, D, Johnstone, D. C. Look, Z.-Q. Fang, M. MacCartney, I. Grzegory, M. Reine, C. W. Litton, P. J. Schreiber, W. Walukiewicz, M. Manfra, O. Mitrofanov, J. Jasinski, V. Litvinov, Jan-Martin Wagner, K. Ando, H. Saito, C. Bundesmann, D. Florescu, H. O. Everitt, H. M. Ng, I. Vurgaftman, J. R. Meyer, J. D. Albrecht, C. A. Tran, S.-H. Wei, G. Dalpian, N. Onojima, A. Wickenden, B. Daudin, R. Korotkov, P. Parikh, D. Green, A. Hansen, P. Gibart, F. Omnes, M. G. Graford, M. Krames, R. Butte, and M. G. Ganchenkova for either reading sections of the book, providing unpublished data, or providing
XIX
XX
Preface
suggestions. Many more deserve a great deal of gratitude for willingly spending considerable time and effort to provide me with digital copies of figures and highquality images, but the available space does not allow for individual recognition. They are acknowledged in conjunction with the figures. In a broader sense, I benefited greatly from the counsel and support of Professor T. A. Tombrello of Caltech. I also would like to use this opportunity to recognize a few of the unsung heroes, namely, Dr Paul Maruska and Professor Marc Ilegems who truly started the epitaxy of nitrides with the hydride VPE technique independently, and Dr S. Yoshida and Professor T. Matsuoka for their pioneering work in AlGaN and InGaN, respectively. Richmond, VA January 2008
Hadis Morkoç
XXI
Color Tables
Figure 1.4 A stick-and-ball stacking model of crystals with (a, both top and bottom) 2H wurtzitic and (b, both top and bottom) 3C zinc blende polytypes. The bonds in an A-plane (1 1 2 0) are indicated with heavier lines to accentuate the stacking sequence. The figures on top depict the
three-dimensional view. The figures at the bottom indicate the projections on the (0 0 0 1) and (1 1 1) planes for wurtzitic and cubic phases, respectively. Note the rotation in the zinc blende case along the h1 1 1i direction. (This figure also appears on page 6.)
Handbook of Nitride Semiconductors and Devices. Vol. 1. Hadis Morkoc Copyright 2008 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim ISBN: 978-3-527-40837-5
XXII
Color Tables
Figure 1.9 The bandgaps of nitrides, substrates commonly used for nitrides, and other conventional semiconductors versus their lattice constants. (This figure also appears on page 12.)
Figure 1.23 An artists view of the scanning thermal microscope. Patterned after D.I. Florescu and F.H. Pollak. (This figure also appears on page 57.)
Color Tables
50
40
q D = 800 K
–1
–1
Molar specific heat, Cp (cal mol K )
q D = 500 K
30
20
Cp data, GaN q D = 500 K q D = 600 K
10
q D = 700 K q D = 800 K
0 0
200
400
600
800
1000
Temperature (K) Figure 1.26 Molar specific heat at constant pressure, Cp (cal mol1 K1), of GaN versus temperature. Open circles represent the experimental data. The solid lines are calculation based on the Debye model for Debye temperatures, yD, of 500, 600, 700, and 800 K. Unfortunately, it is difficult to discern a Debye
temperature that is effective over a wide temperature range because a large concentration of defects and impurities is present in GaN. However, a value of 600 K estimated by Slack is used commonly. The data are taken from Refs [215,216], as compiled in Ref. [88]. (This figure also appears on page 61.)
XXIII
Color Tables
50
40 Specific heat, Cp (J mol–1 K–1)
XXIV
30
20 Specific heat AlN (J mol–1 K–1)
800 K 850 K 900 K 950 K 1000 K 1050 K 1100 K
10
0
0
200
400
600
800
1000
Temperature (K) Figure 1.29 Molar specific heat at constant pressure, Cp (J mol1 K1, 1 cal ¼ 4.186 J), of AlN versus temperature. Open circles represent the experimental data. The solid lines are calculation based on the Debye model for Debye temperatures, yD, in the range of 800–1100 K
with 50 K increments. The data can be fit with Debye expression for yD ¼ 1000 K, which compares with 950 K reported by Slack et al. The data are taken from Ref. [88]. (This figure also appears on page 68.)
Color Tables
Figure 1.41 Bandgap versus composition for quaternary AlxInyGa1xyN (assumed InN bandgap ¼ 0.8 eV). (This figure also appears on page 101.)
Figure 1.42 Bandgap versus composition for quaternary AlxInyGa1xyN (assumed InN bandgap ¼ 1.9 eV). (This figure also appears on page 101.)
XXV
Color Tables
0.02
Spontaneous polarization (C m–2)
XXVI
+0.333 InxAl1–xN
InxGa1–xN
0.00
+0.193
–0.02
InxGa1–xN
–0.04
GaN
+0.095 –0.06
InxAl1–xN
InN AIxGa1–xN
+0.037 –0.08
Random alloy CH-like CP-like
AIN
AIN
–0.10 0
0.2
0.4
0.6
1
Molar fraction, x Figure 2.37 Spontaneous polarization versus the molar fraction in all three ternary nitride alloys. Circles, squares, and triangles represent random alloy, CH-/LZ-, and CP-like structures, respectively. The dashed/dotted lines (blue) with solid triangles are for the CP-like alloys, the dashed lines (green) with solid squares are for CH-like alloys, and solid lines (black) with filled
circles are for random alloys. The black dashed lines represent the data calculated using Vegards law. Numbers indicated in the figure are for CP and CH-/LZ-like ordered alloy bowing parameters in terms of C m2. Courtesy of F. Bernaridini and V. Fiorentini. (This figure also appears on page 249.)
Color Tables
25
AlN Si 6HSiC 4HSiC 3CSiC LiGaO2 Al2O3 MgO GaAs ZnO MgAl2O4 LiAlO2 ScMgAlO4 NdGaO3
Curvature (1 m–1)
20
15
10
5
0.0
–5 0.0
20
40
60
Thickness (μm) Figure 2.59 A compilation of the variation of thermal curvature, a measure of strain, in epitaxial GaN layers grown on different substrates with respect to layer thickness [492]. (This figure also appears on page 294.)
80
100
XXVII
XXVIII
Color Tables
Figure 2.59 (Continued )
Color Tables
5108
Stress on various substrates (Pa)
0
–5108
–1109 AlN Si 6HSiC 4HSiC 3CSiC LiGaO2 Al2O3 MgO GaAs ZnO MgAl2O4 LiAlO2 ScMgAlO4 NdGaO3
–1.5109
–2109
–2.5109 –3109 0100
210–5
410–5
610–5
810–5
Thickness (m) Figure 2.60 A compilation of residual thermal stresses in epitaxial GaN layer on different substrates with respect to layer thickness [492]. (This figure also appears on page 296.)
110–4
XXIX
XXX
Color Tables
Stress versus thickness of GaN/potential substrates 5.00E+08
0.00E+00 0
0.00002
0.00004
0.00006
0.00008
0.0001
AlN Si MgO
–5.00E+08
3C-SiC 6H-SiC 4H-SiC
Stress (Pa)
–1.00E+09
ZnO Al2O3 LiGaO2
–1.50E+09
MgAl2O4 GaAs NdGaO3** ScAlMgO**
–2.00E+09
LiAlO2 LSAT –2.50E+09
–3.00E+09
Thickness (m) Figure 2.60 (Continued )
Color Tables
Figure 3.1 The perspective view of the GaAs crystal (a) along [1 0 0] (1 · 1 · 1 unit), (b) [1 1 0] (2 · 2 · 2 units), and (c) [1 1 1] (2 · 2 · 2 units) directions [13]. (This figure also appears on page 330.)
Figure 3.2 The perspective view along (a) the [0 0 1], (b) [0 1 1], and (c) [1 1 1] directions of a Si cell. (This figure also appears on page 333.)
Figure 3.3 Tetragonal bonding of a carbon atom with its four nearest silicon neighbors. The bond lengths depicted with a and C–Si (the nearest neighbor distance) are approximately 3.08 and 1.89 Å, respectively. The right side is the three-dimensional structure of 2H-SiC structure. (This figure also appears on page 334.)
XXXI
XXXII
Color Tables
Figure 3.9 (a) Top view of the oxide structure on SiC ð0 0 0 1Þ. The Si2O3 silicate adlayer consisting of a honeycomb structure with SiOSi bonds. At the center of the hexagons, one carbon atom of the topmost substrate bilayer is visible [the dark shaded area indicates the (1 1; 1) unit cell and light shaded the
pffiffiffi pffiffiffi ð 3 3ÞR30 -unit cell]; (b) side view of the oxide structure on the SiC (0 0 0 1) in ð0 1 1 0Þ SiC projection. Linear SiOSi bonds connect the silicate layer and the underlying SiC substrate. Courtesy of N. Onojima (patterned after Ref. [40]). (This figure also appears on page 342.)
Figure 3.10 (a) Top view (projection on the Si-plane of the basal plane of SiC) and (b) side view of SiC after an in situ Ga exposure indicating of the lack of silicate adlayer. Courtesy of N. Onojima. (This figure also appears on page 342.)
Color Tables
Figure 3.12 The unit cell of sapphire: (a) rhombohedral unit cell; (b) hexagonal unit cell. Smaller spheres are for O and large ones are for Al [13]. (This figure also appears on page 344.)
Figure 3.13 Perspective views in (2 · 2 · 1) unit cells: (a) along the [0 0 0 1] direction in a rhombohedral unit cell; (b) along the (0 0 0 1) direction in hexagonal unit cell [13]. (This figure also appears on page 344.)
XXXIII
XXXIV
Color Tables
Side view of sapphire r-plane {1 0 1 2}
Oxygen
(a)
Al
Al2O3 [1 0 1 1]
Color Tables
a
2
m
C
b
m
LiGaO2
Lower O
Li
Upper O
Ga
Figure 3.24 Example of the exact fit of GaN atoms over the LiGaO2 lattice if there is no distortion. Courtesy H. Paul Maruska. (This figure also appears on page 356.)
3
Figure 3.20 (a) Sapphire r-plane stacking sequence showing O atoms in larger clear circles and Al atoms in smaller, filled circles. The salient feature is that each Al layer has an O layer above and below it. (b) The atomic arrangement on three layers (the uppermost one is O,
immediately below is Al and third layer down is another O layer) on the r-plane of sapphire. The lines are there just guides to eye and do not represent bonds. (This figure also appears on page 352.)
XXXV
Color Tables
LGO: orthorhombic a = 5.402 Å b = 6.372 Å c = 5.007 Å
c b
a
(a)
LGO: orthorhombic a = 5.402 Å b = 6.372 Å c = 5.007 Å
c b
a
(b)
Projection on c-plane ∆b = –0.19 % aG
aN
=3
.18
9Å
∆a = +1.1%
bLGO = 6.372 Å
XXXVI
∆a = +1.9 % ∆b = –1.1% (c)
aLGO = 5.402 Å
Figure 3.25 Structure of (a) orthorhombic LiGaO2 (LGO), (b) GaN, and (c) a detailed view of the relative orientation of GaN with respect to LGO. Courtesy of H. Paul Maruska. (This figure also appears on page 357.)
Color Tables
Figure 3.43 A schematic representation of a vertical OMVPE system employed at Virginia Commonwealth University along with a picture of the deposition chamber (a); a photograph of the reactor chamber of the same (b). (This figure also appears on page 394.)
XXXVII
XXXVIII
Color Tables
Figure 3.82 Cartoons illustrating the laser liftoff process sequence of 20 GaN membranes. (a) Laser lift-off of the GaN film coated with silicone elastomer and affixed onto a support template; (b) removal of sapphire following laser scanning; (c) deposition of an approximately
3 mm thick thermoplastic adhesive layer at 120 C; (d) peeling-off of the silicone elastomer. In the last step the GaN film is fully removed by dissolving the thermoplastic adhesive in acetone. Courtesy of M. Stutzmann, Ref. [382]. (This figure also appears on page 471.)
Figure 3.128 SEM and CL wavelength image of a cross section of HVPE ELO sample. Courtesy of J. Christen and A.G. Hoffmann, Ref. [695]. (This figure also appears on page 560.)
Color Tables
Figure 3.129 SEM (a) and CL wavelength (b) images of two different regions, coherently grown above the openings and overgrown above the SiO2 stripes. The growth in the windows (between the SiO2 stripes) and the wing
(coalesced regions over the SiO2 stripes) regions indicated in the schematic drawing (c) are clearly visible in the CL wavelength image. Courtesy of J. Christen and A.G. Hoffmann, Ref. [697]. (This figure also appears on page 561.)
XXXIX
XL
Color Tables
Figure 3.130 A m-Raman scans along c-axis of overgrowth GaN (in blue) and coherently grown GaN (in red): (a) free carrier density and (b) biaxial compressive stress. Courtesy of J. Christen and A. G. Hoffmann, Ref. [697]. (This figure also appears on page 562.)
Color Tables
Figure 3.160 Schematic representation of m-, a-, and c-planes of GaN. (This figure also appears on page 613.)
XLI
XLII
Color Tables
Figure 3.165 Cartoon of epitaxial lateral overgrowth (ELO) on a-plane GaN with growth along the [0 0 0 1] direction, representing the Gapolar growth front and along the ½0 0 0 1 direction, representing the N-polar growth front: (a) top view and (b) side view. The growth along the Ga-polar front is about a factor of g ¼ 3 times faster. Courtesy of VCU students Vishal Kasliwal
and Xianfeng Ni. (c) Left 30 mm · 30 mm AFM image for sample B. Right 4 mm · 4 mm AFM image near the window and N-polar wing boundary of sample B, showing different surface pit densities for the window and the wing. Courtesy of VCU students Vishal Kasliwal and Xianfeng Ni. (This figure also appears on page 618.)
Color Tables
Figure 3.168 (a) AFM and (b) NSOM scans from a 40 mm · 40 mm area of a-GaN ELO sample B [875]. (This figure also appears on page 620.)
XLIII
XLIV
Color Tables
Figure 3.249 (a) I–V characteristics of a core multishell (CMS) nanowire device, the top view of which is shown in the inset in the form of a field emission scanning electron microscopy image. Scale bar is 2 mm. (b) Optical microscopy images collected from around theopaque p-contact ofcore multishell nanowire LEDs with increasing In
concentration in the shell quantum well and in forward bias, showing purple, blue, cyan, green, and near yellow emission, respectively. (c) Normalizedelectroluminescencespectraobtained from five representative multicolor CMS nanowire LEDs. Courtesy of C.M. Lieber. (This figure also appears on page 747.)
Color Tables
Figure 3.249 (Continued )
XLV
Color Tables
(c)
1.0
0.8 Normalized intensity (au)
XLVI
0.6
0.4
0.2
0.0 300
400
500 Wavelength (nm)
Figure 3.249 (Continued )
600
700
Color Tables
Figure 4.8 Side view (projection onto the ð1 1 2 0Þ plane) of a relaxed and neutral screw dislocation: (a) full-core screw dislocation; (b) Ga-filled screw dislocation. Core of a full-core screw dislocation (discussed in greater detail in Figure 4.10 and the associated text) showing the
double helix of Ga bonds. The supercell is repeated twice in the [0 0 0 1] for clarity. Note that the bonds at the core are heavily distorted. Courtesy of Blumenau et al. (patterned after Ref. [29]). (This figure also appears on page 829.)
XLVII
XLVIII
Color Tables
Figure 4.36 Potential profile across the dislocation in n-type GaN deduced from the holographic phase map compared with theoretical profile (a). Courtesy of D. Cherns. False colored map showing phase shifts produced by edge dislocations viewed end-on in nominally undoped GaN. Contour lines emphasize dipole-like phase shifts near
dislocation cores (b). Line profile through indicated dislocation in (b) allows quantification of nominally undoped GaN of electric fields, yielding an estimated bound surface charge of 4 · 1011 e cm2 on either side of the defect (c). Courtesy of M. McCartney. (This figure also appears on page 874.)
Color Tables
2 1.5
Phase (rad)
1 0.5 0 –0.5 –1 –1.5 –2 –150
–100
(c)
–50 0 Distance (nm)
50
100
Figure 4.36 (Continued )
Figure 4.42 Contour plots of dislocation-induced electronic gap states for three edge dislocation configurations, namely, (a) fourcore, (b) full-core, and (c) open-core structures. The plots are obtained by calculating atomic geometries with DFT theory used as input to image simulations. Large (small) balls correspond to Ga (N) atoms [125]. (This figure also appears on page 883.)
150
XLIX
Color Tables
Mean implantation depth (µm)
0
0.09
0.26
0.50
0.79
1.13
HVPE GaN
0.50
10–14 µm
Mg-doped reference
0.49
S parameter
L
1 µm
36–39 µm
5 µm
49–68 µm
VGa
0.48
0.47
0.46 Defect free
0.45 0
5
10
15
20
Positron energy (keV) Figure 4.97 Ga vacancies and 1, 5, 10–14, 36–39, 49–68 mm thick HVPE GaN layers indicating increased S parameter, thus increased Ga vacancy concentration toward the GaN/Al2O3 interface in each of the films. A Mg-doped p-type sample with very low, if any, Ga vacancy is shown as the reference. Courtesy of K. Saarinen. (This figure also appears on page 986.)
25
Color Tables
Figure 4.118 Hole concentration versus ND where the acceptor–donor–acceptor complex model of Ref. [495] is shown with thin and bold lines for NA ¼ 1 · 1019 cm3 (i) and 1 · 1020 cm3 (ii), respectively. The optimum hole concentration where NA ¼ 2ND, as expected from the complex formation, is shown with diamonds. For comparative purpose, the simple compensation model which assumes a single
donor and (unpaired) acceptor is depicted with thin and thick lines for NA ¼ 1 · 1019 cm3 (iii) and 1 · 1020 (iv), respectively. The random pair model is also plotted with thin and thick lines for NA ¼ 1 · 1019 cm3 (v) and 1 · 1020 cm3 (vi), respectively. Discussions with Dr R. Korotkov are acknowledged. (This figure also appears on page 1024.)
LI
Color Tables
Figure 4.126 Proposed bond center and antibonding site incorporation of H in GaN and its passivation of Mg during growth (the Mg atom is directly below the H atom). In part courtesy of C. Van de Walle. (This figure also appears on page 1037.)
GaN: Mn Mnd projected PDOS (/eV cell)
LII
t+
e-
t+
e+ t-
GaP: Mn
e-
t+ e+
t+
ee+
t-
t+
tt+
GaSb: Mn
e-
e+ t–3
t-
t+
t-
GaAs: Mn
–4
t-
–2
t-
t+ –1
–F Energy (eV)
1
2
3
Figure 4.150 In Mn d projected partial density of states for a single Mn in GaN, GaP, GaAs, and GaSb, where the symmetry (t2 and e) as well as the spin (þ and ) have been indicated. The shaded region represents the t2þ states (after Ref. [699]). (This figure also appears on page 1097.)
Color Tables
3d ion d n-1 t-(d )
e-(d )
t+(d)
Mn on Ga site t-CFR
Anion dangling bonds V Ga3-
CFR
e-
t+CFR t+(p)
e+(d) VBM
CFR
e+
DBH
t-
t-(p)
DBH
t+
Figure 4.151 A schematic energy-level diagram for the levels (central panel) generated from the interaction between the crystal field and exchange-split levels on the 3d transition metal ion (left panel) with the anion dangling bond levels (right panel), when the TM d levels are energetically shallower than the dangling bond levels (after Ref. [700]). (This figure also appears on page 1098.)
LIII
LIV
Color Tables Figure 4.155 Interband transitions in GaAs selected because of its well-known band structure and also its well-established and wellcharacterized properties in terms of magnetic ion doped diluted magnetic semiconductors: (arrows indicate emission but the concept is just as applicable to transitions from the valence band subband to the conduction band as in absorption). (a) Schematic band structure of GaAs near the G point, the center of the Brillouin zone. As for the terms, Eg is the bandgap and DSO the spin–orbit splitting; CB, conduction band; HH, valence heavy hole; LH, light hole; SO, spin–orbit split-off subbands; G6,7,8 are the corresponding symmetries at the k ¼ 0 point representing conduction, HH, LH, spin–orbit (SO) bands, or, more precisely, the irreducible representations of the tetrahedron group Td (see, e.g., Ref. [721]). The terms s1/2 and p3/2 and p1/2 represent the conduction band (s-like) and valence band (p-like) type of orbitals. (b) Selection rules for interband transitions between the Jz projection of the angular momentum along z-direction, sublevels for circularly polarized light sþ(right-hand circular polarization or positive helicity that results from transitions between the Jz ¼ 1/2 conduction band states and Jz ¼ 3/2 heavy-hole states, and Jz ¼ þ1/2 conduction band and Jz ¼ 1/2 light-hole states), and s (left-hand circular polarization or negative helicity, which results from transitions between the Jz ¼ þ1/2 conduction band states and Jz ¼ þ3/2 heavy-hole states, and Jz ¼ 1/2 conduction band and Jz ¼ þ1/2 light-hole states). The numbers by each transition indicate the relative transition intensities, with respect to the light-hole subband to the conduction band (absorption or excitation of carriers to higher bands), or the conduction band to the light-hole subband transition (emission), which apply to both excitation and radiative recombination (depicted by the arrows). The circular polarization (s polarization) for light energies that would not excite the spin split-off band is ideally 50%, which becomes 0 if the spin–orbit split-off band is also excited. For completeness, the transitions between the Jz ¼ 1/2 conduction band states and Jz ¼ 1/2 light-hole states, and Jz ¼ þ1/2 conduction band states and Jz ¼ þ1/2 light-hole states, which are linearly polarized (p polarization), are also shown as depicted by two-way arrows in the figure. The
" transition probability or the emission intensity normalized to the Jz ¼ 1/2 conduction state to the Jz ¼1/2 state transitions (indicated with 1) are also indicated in numbers for GaAs. The circular polarization resulting from the conduction band to the heavy-hole states are three times more intense than the circular polarization resulting from the conduction band states to the light-hole valence band states. The linearly polarized transitions are twice as intense as the circular polarization involving light-hole states. (c) Removal of the valence band heavyand light-hole degeneracy by, for example, strain inducing either by lattice mismatch or by confinement in a quantum well, which increases the electron polarization to nearly 100%. Note that heavy- and light-hole states are no longer degenerate. Both the tensile (left) and compressive (right) in-plane biaxial strain cases are shown. The respective ratios of various transitions (oscillator strengths) have been assumed to be the same as in the relaxed case. Note that spin is indifferent to strain, which means that spin-up and spin-down states are moved in the same direction by strain, but not to magnetic field, as spin-up and spin-down states in a given band are split and moved in opposite directions as shown in (d). In part courtesy of W. Chen, Linko1ping University. (d) Removal of the valence band heavy- and light-hole degeneracy as well as splitting the spin-up and spin-down states by application of magnetic field. The total splitting is enhanced due to sp-d interaction in DMS materials in the form of xN0a<Sz> for the conduction band states, xN0b<Sz> for the HH and LH valence band states, and (1/3)xN0a<Sz> for the spin–orbit split-off band. Here, N0, x, a, b, <Sz> represent the number of cations per unit volume, mole fraction of magnetic ions, the product of Bohr magneton and the g factor for the respective bands, and average spin for each magnetic ion site, respectively. Note that magnetic field/magnetization causes Zeeman splitting, and direction of splitting either up or down in energy is spin dependent. If the semiconductor is ferromagnetic as is the case of GaMnAs, one can either couple polarized light to the symmetry/splitting allowed bands or cause polarized light emission by tuning the wavelength. (This figure also appears on page 1112.)
Color Tables
LV
LVI
Color Tables
Figure 4.155 (Continued )
Color Tables
(a) x = 0.053 B ⊥ plane
2K 55 K 25 K
100 K 125 K 300 K
0.00
0.5
Rsheet(kΩ)
R Hall (kΩ )
0.03
-5
25 K
2K 0.3 -5
-0.03
100 K 125 K 55 K
0.4
300 K 0
5
B (T)
0 B (T )
5
0.08
1/χ Hall(au)
(RHall R ) sheet s
(b) 0.06 0.04 0.02 0.00 0
100
200
300
T (K) 120 (c)
T c (K)
80
40
0 0.00
0.04 x
Figure 4.170 (a, top) Temperature dependence of the Hall resistance RHall for a 200 nm thick Ga0.947Mn0.053As sample for which direct magnetization measurements have been performed but not shown. The inset shows the temperature dependence of the sheet resistance Rsheet. (b, center) Temperature dependence of the saturation magnetization [RHall/Rsheet]S obtained using Arrott plots (solid circles) and inverse susceptibility 1/wHall (open circles), both
0.08 deduced from the transport data shown in (a). Solid lines depict [RHall/Rsheet]S and (c, bottom) 1/wHall (bottom, c) calculated using the mean field Brillouin theory with S ¼ 5/2 for the Mn spin and the Curie–Weiss law, respectively. The dependence of magnetic transition temperature TC on Mn composition as determined from the transport data. Courtesy of Ohno and Matsukura [777]. (This figure also appears on page 1140.)
LVII
LVIII
Color Tables
+1/2 CB
Jz = −1/2
σ+
σ+
σ− π
σ−
π
σ+ σ−
+3/2 HH
−3/2 −1/2
+1/2 LH +1/2 CR
−1/2 VB;HH Figure 4.172 The GCB , G7VB;LH 7 conduction band and G9 VB;CR (spin–orbit split-off band), and G7 (crystal field split off band) valance bands in wurtzitic GaN at the G point along with polarization (sþ right-hand and s left-hand circular polarizations) of various transitions between the conduction and valence band states in the presence of a magnetic field. (This figure also appears on page 1147.)
0.15
Cr: 1%
5K
0.05
Cr: 3%
0.05
Magnetization (emu g–1)
Magnetization (emu g–1)
0.1
0.2 μ B /Cr
Cr: 5%
Cr: 0.5%
0
–0.05 –0.1 –0.15 –10 000
Cr : 5% Cr :1%
0
H c =100 Oe
–0.05 –1000
0
1000
Magnetic Field (Oe)
–5000
0
5000
10 000
Magnetic Field (Oe) Figure 4.177 Magnetization curves for Cr-doped GaN in atomic concentrations of 0.5, 1, 3, and 5% up to a magnetic field normal to the surface of 10 000 Oe (1 T). Note that the film containing 5% Cr does not show any saturation magnetization in the range measured and
appears to be paramagnetic. The blow-up version near the origin indicates hysteresis for 1% Cr sample and a coercive field of 100 Oe. Courtesy of F. Hasegawa. (This figure also appears on page 1157.)
Color Tables
0.04
Magnetization (emu g–1)
H = 200(Oe)
Ferromagnetic
Cr: 1%
0.02
Paramagnetic + ferromagnetic
Cr: 3 %
Paramagnetic
Cr : 5%
0 0
50
100
150 200 250 Temperature (K)
300
Figure 4.178 Temperature dependence of magnetization for 1, 3, and 5% Cr-containing GaN. As indicated, the film with 1% Cr is consistent with ferromagnetic behavior. The films with 3 and 5% Cr exhibit a combination of ferromagnetic and paramagnetic behavior, and paramagnetic behavior, respectively. Courtesy of F. Hasegawa. (This figure also appears on page 1158.)
350
LIX
LX
Color Tables
VG < 0 R (kΩ) Hall
0.04
1
1.5 K 5K 10 K
0
20 K
0.02
RHall (kΩ)
–1 –0.5
0.0 B(T)
VG > 0
0.5
0.00
22.5 K
VG = 0V
–0.02
+125 V –125 V 0V –0.04 –1.0
–0.5
0.0
0.5
1.0
B (mT) Figure 4.186 Hall resistance RHall of an insulated gate (In,Mn)As field effect transistor at 22.5 K as a function of the magnetic field for three different gate voltages. RHall is proportional to the magnetization of the (In,Mn)As channel. Upper right inset shows the temperature dependence of
RHall. Left inset shows schematically the gate voltage control of the hole concentration and the corresponding change of the magnetic phase. Courtesy of Ohno et al. [855]. (This figure also appears on page 1172.)
Color Tables
Figure 4.187 Injection of spin-polarized holes into a light-emitting p–n diode using a ferromagnetic semiconductor (Ga,Mn)As. (a) Sample structure. Spin-polarized holes hþ travel through the nonmagnetic GaAs and recombine with spin-unpolarized electrons in the (In,Ga)As quantum well. I represents the current, and sþ represents circularly polarized light emitted from the edge of the quantum well. (b) Dependence of the polarization DP of the emitted light on the magnetic field B at temperatures of 6, 31, 52 K, the latter above the Curie value. The solid and hollow symbols represent the degree of polarization when the magnetic field is swept in the positive and negative directions, respectively.
The magnetic field was applied parallel to the surface along the easy axis of magnetization of the (Ga,Mn)As. The temperature dependence of the residual magnetization M in (Ga,Mn)As, where the degree of polarization of the zero magnetic field seen in the emitted light exhibits the same temperature dependence as the magnetization (not shown). Dependence on temperature for B ¼ 0 of the change in the relative remanent polarization, DP, (hollow circles) and magnetic moment measured by a SQUID magnetometer (solid circles). Courtesy of Ohno and coworkers [770]. (This figure also appears on page 1174.)
LXI
Color Tables
∆P 1.00
T=6K
Magnetization 6
16 K 0.75
4 0.50
31 K
2
0.25
52 K 0.00 0
20
40
60
Temperature (K) (c) Figure 4.187 (Continued )
80
100
Magnetization (10-5emu)
Relative polarization,∆P (%)
LXII
Color Tables
High resistance state
Low resistance state
(b) Figure 4.190 (a) Schematic representation of a spin valve, a normal metal straddled by two ferromagnetic metals. When the spins in ferromagnetic metals on either end are aligned parallel to each other, the system is in the lowresistance state top. When, for example, the spin of the FM metal on the right is flipped by a magnetic field, making the spins of the
ferromagnetic metals antiparallel, a high resistance state is attained. (b) Schematic representation of transport that is parallel to the plane of a layered magnetic metal sandwich structure for antialigned (upper figure – high resistance) and aligned (lower figure – low resistance) orientations. (This figure also appears on page 1180.)
LXIII
j1
1 General Properties of Nitrides Introduction
GaN as a representative of its binary cousins, InN and AlN, and their ternaries along with the quaternary, is considered one of the most important semiconductors after Si. It is no wonder that it finds ample applications in lighting and displays of all kinds, lasers, detectors, and high-power amplifiers. These applications stem from the excellent optical and electrical properties of nitride semiconductors. The parameters are imperative in determining the utility and applicability of this class of materials to devices, as will be made evident in this chapter and throughout the book. In this chapter, the structural, mechanical, thermal, chemical, electrical, and optical properties of GaN and its binary cousins as well as the substrates commonly used for nitride epitaxy are treated in a general sense for quick reference. The detailed properties associated with electrical and optical parameters and properties are discussed in chapters dealing with transport and optical processes in GaN and related alloys. Because GaN is used in the form of a thin film deposited on foreign substrates, meaning templates other than GaN, a discussion of heteroepitaxial thin films is of paramount importance. Consequently, the properties of nitride films intricately depend on substrates, inclusive of the inherent properties such as lattice constants and thermal expansion coefficients, and on the process-induced characteristics such as surface preparation and chemical and physical interactions at the surface. These too are discussed in the book.
1.1 Crystal Structure of Nitrides
Group III nitrides can be of crystalline structures: the wurtzite (Wz), zinc blende (ZB), and rock salt. Under ambient conditions, the thermodynamically stable structure is wurtzite for bulk AlN, GaN, and InN. The zinc blende structure for GaN and InN has been stabilized by epitaxial growth of thin films on {0 1 1} crystal planes of cubic substrates such as Si [1], SiC [2], MgO [3], and GaAs [4]. In these cases, the intrinsic tendency to form the Wz structure is overcome by the topological compatibility.
Handbook of Nitride Semiconductors and Devices. Vol. 1. Hadis Morkoc Copyright 2008 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim ISBN: 978-3-527-40837-5
j 1 General Properties of Nitrides
2
However, Wz structure could very likely be present at the extended defect sites. The rock salt form is possible only under high pressures and, therefore, is laboratory form of exercise. Let us now discuss the space groups for the various forms of nitrides. The rock salt, or NaCl, structure (with space group Fm3m in the Hermann–Mauguin notation and O5h in the Schoenflies notation) can be induced in AlN, GaN, and InN under very high pressures. The reason for this is that the reduction of the lattice dimensions causes the interionic Coulomb interaction to favor the ionicity over the covalent nature. The structural phase transition to rock salt structure was experimentally observed at the following estimated pressure values: 22.9 GPa (17 GPa in other estimates) for AlN [5], 52.2 GPa for GaN [6], and 12.1 GPa for InN [7]. Rock salt III nitrides cannot be produced by any epitaxial growth. The space grouping for the zinc blende structure is F43m in the Hermann– Mauguin notation and T 2d in the Schoenflies notation. The zinc blende structure has a cubic unit cell, containing four group III elements and four nitrogen elements. (Although the term zinc blende originated in compounds such as ZnS, which could be in cubic or hexagonal phase, it has been used ubiquitously for compound semiconductors with cubic symmetry. The correct term that should be used for the cubic phase of GaN is actually sphalerite. However, to be consistent with the usage throughout the literature, even at the expense of accuracy, the term zinc blende has been used in this book). The position of the atoms within the unit cell is identical to the diamond crystal structure. Both structures consist of two interpenetrating facecentered cubic sublattices, offset by one quarter of the distance along a body diagonal. Each atom in the structure may be viewed as positioned at the center of a tetrahedron, with its four nearest neighbors defining the four corners of the tetrahedron. The stacking sequence for the (1 1 1) close-packed planes in this structure is AaBbCc. Lowercase and uppercase letters stand for the two different kinds of constituents. The wurtzite structure has a hexagonal unit cell and thus two lattice constants, c and a. It contains six atoms of each type. The space grouping for the wurtzite structure is P63mc in the Hermann–Mauguin notation and C46v in the Schoenflies notation. The point group symmetry is 6 mm in the Hermann–Mauguin notation and C6v in the Schoenflies notation. The Wz structure consists of two interpenetrating hexagonal close-packed (hcp) sublattices, each with one type of atom, offset along the c-axis by 5/8 of the cell height (5c/8). The wurtzite and zinc blende structures are somewhat similar and yet different. In both cases, each group III atom is coordinated by four nitrogen atoms. Conversely, each nitrogen atom is coordinated by four group III atoms. The main difference between these two structures lies in the stacking sequence of closest packed diatomic planes. The Wz structure consists of alternating biatomic close-packed (0 0 0 1) planes of Ga and N pairs, thus the stacking sequence of the (0 0 0 1) plane is AaBbAa in the (0 0 0 1) direction. Although the main interest is in Wz GaN as opposed to zinc blende GaN, a description of stacking sequence of both GaN polytypes with the accepted Ramsdel notation is warranted, so is the stacking order of SiC polytypes that are relevant to GaN because they are used for substrates in GaN epitaxy. Therefore, a generic description of stacking in Wz semiconductors is given below. A comprehensive description of the tetrahedrally coordinated structures is imperative for a clear picture
1.1 Crystal Structure of Nitrides
j3
of nitride semiconductors, particularly the extended defects that are discussed in detail in Chapter 4. The bonds describe a tetrahedron denoted by T, which has one atom species at each of the three corners and the other atom species in its center [8,9]. The basal plane of this structure is defined by one face of the tetrahedron and the bond perpendicular to this plane defines the c-axis. A rotation of 180 around the caxis produces a twin variant denoted by T0 as shown in Figure 1.1a (left).
(a)
Atom a Atom b [0001]
T
T'
〈1120〉
{1100} In-plane bonds Out-of-plane bonds
(b)
A
T 3´
b
B
T1
a
A b
B
Figure 1.1 Representation of the tetrahedrally coordinated materials in the Ramsdel notation. (a) The two possible 0 0 tetrahedra. (b) The T1, T3, T1, T3, tetrahedral stacking composing the 2H sequence. Courtesy of Pierre Ruterana [9].
T 3´
T1
j 1 General Properties of Nitrides
4
The two variants (twins T and T0 ) are related to one another by mirror symmetry about one of the {1 1 0 0} m-planes. A tetrahedron can occupy one of the three possible positions in the basal plane. The representation of the tetrahedrally coordinated materials in the Ramsdel notation is shown in Figure 1.1a for two possible tetrahedra, one is the mirror image twin of the other with respect to the (1 1 0 0) m-plane. The single bonds are on the (1 1 2 0) plane, called the a-plane. The 0 0 0 layers of the tetrahedra can then be denoted by T1, T2, T3, and by T1 ; T2 ; T3 for its 0 0 twin. An example of T1 ; T ; T1 ; T3 stacking order representing 2H ordering as in wurtzitic GaN is shown in Figure 1.1b. The structure of nitride semiconductors and most relevant polytypes of SiC can be completely described by a combinatorial stacking of the aforementioned six tetrahedra layers. Naturally, not all the stacking sequences must obey the following two rules to keep a corner sharing structure, as such not all stacking orders are allowed: (i) A tetrahedron T can be followed by another one of the same kind with the 0 0 0 following subscript: T1T2T3, and inversely for the twin variant: T3 T2 T1 . (ii) A tetrahedron T1 must be followed by the twin variant of the preceding subscript: 0 0 T1 T3 , and inversely for its twin variant: T1 T2 . In the Ramsdel notation, the stacking order for the wurtzite structure corresponding to various polytypes can be denoted as . . . .
0
0
0
T1 T3 or T2 T1 or T3 T2 for the 2H polytype, which is also applicable to Wz nitride semiconductors; 0 0 T1T2T1 T3 for the 4H polytype; 0 0 0 T1T2T3T2 T1 T3 for the 6H polytype; 0 0 0 T1T2T3 or T3 T2 T1 for the 3C polytype.
The 3C, 4H, and 6H stacking sequences as well as 2H sequence on 6H sequence are discussed in Chapter 3. Recall that GaN crystallizes in the cubic structure (zinc blende or sphalerite, the latter being the correct term and the former being the one used universally) or in the more stable hexagonal structure (wurtzite). The anions (N3) form an hcp structure in which the cations (Ga3þ) occupy half of the tetrahedral sites. The structure of a unit cell of GaN projected along [0 0 0 1] is depicted schematically in Figure 1.2. The open symbols represent g sites that are occupied by nitrogen atoms; the Ga atoms are in the tetrahedral sites, b. These latter sites can either be at heights (3/8)c above (b1) or below (b2) N site, depending on the crystal polarity. A stick-and-ball representation of Ga-polarity and N-polarity Wz structure is depicted in Figure 1.3. The Wz and zinc blende structures differ only in the bond angle of the second nearest neighbor (Figure 1.4). As clearly shown, the stacking order of the Wz along the [0 0 0 1] c-direction is AaBb, meaning a mirror image but no in-plane rotation with the bond angles. In the zinc blende structure along the [1 1 1] direction, there is a 60 rotation that causes a stacking order of AaBbCc. The point with regard to rotation is illustrated in Figure 1.4b. The nomenclature for various commonly used planes of hexagonal semiconductors in two- and three-dimensional versions is presented in Figures 1.5 and 1.6. The Wz group III nitrides lack an
1.1 Crystal Structure of Nitrides
a γ1
β1
γ1
γ1
* γ1
β2
c
β1
γ2
β2 β1
γ1
u
β2
*
β2 β1
β2 β1
γ1
γ1
γ1
Figure 1.2 Schematic diagram showing the b1 and b2 tetrahedral sites of GaN unit cell. Starting with the assumption that N occupies the g sites, only one family of b sites can be simultaneously occupied by Ga atoms. Courtesy of Pierre Ruterana [9].
inversion plane perpendicular to the c-axis; thus, nitride surfaces have either a group III element (Al, Ga, or In) polarity (referred to as Ga-polarity) with a designation of (0 0 0 1) or (0 0 0 1)A plane or a N-polarity with a designation of (0 0 0 1) or (0 0 0 1)B plane. We will use the former notations for each. The distinction between these two directions is essential in nitrides because of their implications for the polarity of the polarization charge. Three surfaces and directions are of special importance in
Ga-polarity
N-polarity
[0001]
[0001]
Ga
N
Ga
N
Ga
N
Ga N
Ga N
Ga N
Ga N
Ga
N
Ga N
Ga
Ga
Ga
N
N
N
Ga
Ga
Ga
N
N
N
Ga
N
Ga
Ga
N
N
N
N
N
Ga
Ga
Ga
N
N
Ga
Ga
N Ga
Ga
N
N
N
N
Ga
Ga
Ga
N Ga
Figure 1.3 A stick-and-ball diagram of a hexagonal structure.
N Ga
N
N
Ga
Ga
j5
j 1 General Properties of Nitrides
6
C View normal to [0001] and [111]
B
B
A
A
Ga N
Wurtzitic
Zinc blende
View along [0001] and [111]
(a) Figure 1.4 A stick-and-ball stacking model of crystals with (a, both top and bottom) 2H wurtzitic and (b, both top and bottom) 3C zinc blende polytypes. The bonds in an A-plane (1 1 2 0) are indicated with heavier lines to accentuate the stacking sequence. The figures on top depict the three-dimensional view. The
(b) figures at the bottom indicate the projections on the (0 0 0 1) and (1 1 1) planes for wurtzitic and cubic phases, respectively. Note the rotation in the zinc blende case along the h1 1 1i direction. (Please find a color version of this figure on the color tables.)
nitrides, which are (0 0 0 1) c-, (1 1 2 0) a-, and (1 1 0 0) m-planes and the directions associated with them, h0 0 0 1i, h1120i, and h1100i as shown in Figure 1.7. The (0 0 0 1), or the basal plane, is the most commonly used surface for growth. The other two are important in that they represent the primary directions employed in reflection high-energy electron diffraction (RHEED) observations in molecular beam epitaxial growth, apart from being perpendicular to one another. They also represent the direction of stripes employed in the epitaxial lateral overgrowth (ELO), details of which are discussed in Section 3.5.5.2. The cohesive energy per bond in the wurtzite form is 2.88 eV (63.5 kcal mol1), 2.2 eV (48.5 kcal mol1), and 1.93 eV (42.5 kcal mol1) for AlN, GaN, and InN, respectively [10]. The calculated energy difference DEW-ZB between wurtzite and zinc blende lattice is small [11]: DEW-ZB ¼ 18.41 meV/atom for AlN, DEW-ZB ¼ 11.44 meV/atom for InN, and DEW-ZB ¼ 9.88 meV/atom for GaN. Wurtzite form is energetically preferable for all three nitrides compared to zinc blende, although the energy difference is small. The Wz structure can be represented by lattice parameters a in the basal plane and c in the perpendicular direction, and the internal parameter u, as shown in Figure 1.8.
1.1 Crystal Structure of Nitrides 〈1100〉
〈1120〉
(m)
(a )
o
1010
30
(m )
o
30
r
n n
n n
c
1100
0110
r
r n
2110
1120
n 0111 s (c)
1210 a 57
o
0114 d 1213 n
1012 r
2113 n
1101 s
1104 d 0001 c
1210 1213 n
o
.6
61
1123 n
c
(n) r
(r) r
n
n
1102 r
32.4
n
o
1100 m
a
a
a
1014 d
2113 n
0112 r 0110
1123 n
1011 s
(m) 2110
1120 1010
n
n
r
n
Common crystallographic planes in sapphire d Plane Miller name index spacing a m c r n s
(1120) (1010) (0001) (1102) (1123) (1011)
2.379 Å 1.375 Å 2.165 Å 1.740 Å 1.147 Å 1.961 Å
Angles between common planes (0001) ^ (1102) (0001) ^ (1123) (0001) ^ (1011) (0001) ^ (1121) (0001) ^ (1120) (0001) ^ (1010) (1120) ^ (1010)
c^r c^n c^s c^ c^a c^m a^m
57º 35' 61º 11' 72º 23' 79º 37' 90º 00' 90º 00' 30 00'
Figure 1.5 Labeling of planes in hexagonal symmetry ( for sapphire).
The u parameter is defined as the anion–cation bond length (also the nearest neighbor distance) divided by the c lattice parameter. The c parameter depicts the unit cell height. The wurtzite structure is a hexagonal close-packed lattice, comprising vertically oriented X–N units at the lattice sites. The basal plane lattice parameter (the edge length of the basal plane hexagon) is universally depicted by a and the axial lattice parameter, perpendicular to the basal plane, is universally described by c. The interatomic distance in the basic unit is described by the internal parameter u. In an ideal wurtzite structure represented by four touching hard spheres, pffiffiffiffiffiffiffiffi the values of the axial ratio and the internal parameter are c=a ¼ 8=3 ¼ 1:633 and u ¼ 3/8 ¼ 0.375, The crystallographic ! pffiffiffi respectively. pffiffiffi ! vectors of wurtzite are a ¼ að1=2; 3=2; 0Þ, b ¼ að1=2; 3=2; 0Þ, and ! c ¼ að0; the basis atoms are (0, 0, 0), (0, 0, uc), pffiffiffi0; c=aÞ. In Cartesian coordinates, pffiffiffi a(1=2; 3=6; c=2a), and a(1=2; 3=6; ½u þ 1=2c=a). Table 1.1 tabulates the calculated structural parameters a, c/a, and e1 ¼ u uideal for the III–V nitrides by three different groups [12–14]. In the case of Bernardini et al. [12], they optimized the structure within both the generalized gradient
j7
j 1 General Properties of Nitrides
8
(tuvw) coordinate system v
1010 2110
1120
0110
0111 s
1123 n
1012 r 2113 n
0114 d
1104 d 1210
t
0001 c 0112 r
1102 r 2113 n
1100 m
1100
1213 n
1213 n
1210 a
1101s
1014 d
1123 n
0110
1011 s 1120
2110
u
1010
Figure 1.6 A magnified view of labeling of planes in hexagonal symmetry in the (tuvw) coordinate system with w representing the unit vector in the c-direction. The lines are simply to show the symmetryonly. If the lines connecting m-points among each other and a-points among each other were to be interpreted as the projection of those
planes on the c-plane, the roles would be switched in that the lines connecting the m-points would actually represent the a-planes and lines connecting the a-points would actually represent the m-planes that are normal to the plane of the page.
(1120) a-plane [1120] v
(1100) m-plane [0110]
[1010]
[1210]
[1100]
[2110]
t
[2110]
[1010]
[1100]
Ga N
[0110] u [1210]
m-planes a-planes
[1120]
Figure 1.7 The orientations which are commonly used in nitrides, namely the (1 1 2 0) and (1 1 0 0) planes and associated directions are shown as projections on the (0 0 0 1) basal plane.
1.1 Crystal Structure of Nitrides
a
M
M
α c N M
M
b=u x c M
M
b1
N
[0001]
N
b '2
M
N
b 3' N
β b'
N
1
N
M M
M
Figure 1.8 Schematic representation of a wurtzitic metal nitride structure with lattice constants a in the basal plane and c in the basal direction, u parameter, which is expressed as the bond length or the nearest neighbor distance (b) divided by c (0.375 in ideal crystal), a and b (109.47 in ideal crystal) are the bond 0 0 0 angles, and b 1, b 2, and b 3, represent the three types of second nearest neighbor distances.
approximation (GGA) and local density approximation (LDA). The experimental data are from Leszczynski et al. [15]. In all Wz III nitrides, experimentally observed c/a ratios are smaller than ideal and it has been postulated that not being so would lead to the zinc blende phase [16]. There are two avenues that can lead to a deviation from ideal: changing the c/a ratio or changing the u value. It should be pointed out that a strong correlation exists between the c/a ratio and the u parameter so that when c/a decreases, the u parameter increases in a manner to keep the four tetrahedral distances nearly constant through a distortion of tetrahedral angles. For equal bond length to prevail, the following relation must hold: uð1=3Þða2 =c 2 Þ þ 1=4:
ð1:1Þ
Table 1.1 Structural parameters for GaN reported by Bechstedt,
Großner, and Furthm€ uller (BGF) [13] and by Wei and Zunger (WZ) [14] using the local density approximation (LDA).
BGF WZ BFV (LDA) BFV (GGA) Experimental data
a (Å)
c/a
e1 (103c/a)
3.150 3.189 3.131 3.197 3.1890
1.6310 1.6259 1.6301 1.6297 1.6263
6.5 1.8 1.6 1.9 2.0
However, Bernardini, Fiorentini, and Vanderbilt [12] employed both the LDA and GGA methods. Lattice constant a is given in Å and e1 in 103c/a.
j9
j 1 General Properties of Nitrides
10
The nearest neighbor bond length along the c-direction (expressed as b in Figure 1.8) and off c-axis (expressed as b1 in Figure 1.8) can be calculated as sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 1 2 1 a þ u c2 : ð1:2Þ b ¼ cu and b1 ¼ 3 2 In addition to the nearest neighbors, there are three types of second nearest 0 0 neighbors designated in Figure 1.8 as b 1 (one along the c-direction), b 2 (six of them), 0 and b 3 (three of them), which are given as [17] sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 4 2 1 0 0 0 2 a þ c2 u : b 1 ¼ cð1 uÞ; b 2 ¼ a2 þ ðucÞ ; and b 3 ¼ 3 2 ð1:3Þ The bond angles, a and b, are given by [17] "qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi # 1 a ¼ p=2 þ arccos 1 þ 3ðc=aÞ2 ð u þ 1=2Þ2 ; "qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
b ¼ 2 arcsin
2
4=3 þ 4ðc=aÞ ð u þ 1=2Þ
2
1
ð1:4Þ
# :
Table 1.2 tabulates the calculated as well as experimentally observed structural parameters discussed above, inclusive of the lattice parameters, the nearest and second nearest neighbor distances, and the bond angles for three end binaries, GaN, AlN, and InN. The distances are in terms of Å. The lattice parameters are commonly measured at room temperature by X-ray diffraction (XRD), which happens to be the most accurate one, using the Bragg law. In ternary compounds, the technique is also used for determining the composition; however, strain and relevant issues must be accounted for as the samples are in the Table 1.2 Calculated (for ideal crystal) and experimentally
observed structural parameters for GaN, AlN, and InN [17]. GaN
u a (Å) c/a b (Å) b1 (Å) 0 Â b 1 ðeÞ 0 Â b 2 ðeÞ 0 Â b 3 ðeÞ a b
AlN
InN
Ideal
Exp.
Ideal
Exp.
Ideal
Exp.
0.375 3.199 1.633 1.959 1.959 3.265 3.751 3.751 109.47 109.47
0.377 3.199 1.634 1.971 1.955 3.255 3.757 3.749 109.17 109.78
0.375 3.110 1.633 1.904 1.904 3.174 3.646 3.646 109.47 109.47
0.382 3.110 1.606 1.907 1.890 3.087 3.648 3.648 108.19 110.73
0.375 3.585 1.633 2.195 2.195 3.659 4.204 4.204 109.47 109.47
0.379 3.585 1.618 2.200 2.185 3.600 4.206 4.198 108.69 110.24
1.1 Crystal Structure of Nitrides
form of epitaxial layers on foreign substrates. The accuracy of X-ray diffraction and less than accurate knowledge of the elastic parameters together allow determination of the composition to only within about 1% molar fraction. In addition to composition, the lattice parameter can be affected by free charge, impurities, stress (strain), and temperature [18]. Because the c/a ratio correlates with the difference of the electronegativities of the two constituents, components with the greatest differences show the largest departure from the ideal c/a ratio [19]. For GaN, the c/a ratio and the value of u are measured as 1.627 (1.634 in Ref. [17]) and 0.377, respectively, which are close to the ideal values [20]. AlN deviates significantly from the ideal parameters: c/a ¼ 1.601 (1.606 in Ref. [17]) and u ¼ 0.382. Although the data for InN are not as reliable, values of u ¼ 0.379 and c/a ¼ 1.601 have been reported [17]. Inhomogeneities, strain, partial relaxation of strain, and high concentration of structural defects may distort the lattice constants from their intrinsic values and cause a wide dispersion among the reported values. Table 1.3 lists a comparison of measured and calculated lattice parameters reported for AlN, GaN, and InN crystallized in the wurtzite structure in more detail in terms of the specifics of the sample used for measurements and complements. The dispersion is even a greater concern in ternary and quaternaries, as compositional inhomogeneities, in addition to the aforementioned issues, cause an additional dispersion. The particulars of the ternaries are discussed in Section 1.5. The wurtzite polytypes of GaN, AlN, and InN form a continuous alloy system whose direct bandgaps range, according to data that adorned the literature for years, from 1.9 eV for InN, to 3.42 eV for GaN, and to 6.2 eV for AlN. A revisit of the InN bandgap indicates it to be about 0.78 eV [30] and the same for AlN is about 6 eV in which case the energy range covered would be about 0.7–6 eV. Thus, the III–V nitrides could potentially be fabricated into optical devices, which are active at wavelengths ranging from the red well to the ultraviolet. The bandgaps of nitrides, Table 1.3 Measured and calculated lattice constants of wurtzite AlN, GaN, and InN.
Compound
Sample
a (Å)
c (Å)
AlN
Bulk crystal [21] Powder [22] Epitaxial layer on SiC [23] Pseudopotential LDA [24] FP-LMTO LDA [25] Bulk crystal [18] Relaxed layer on sapphire [26] Powder [29] Relaxed layer on sapphire [27] GaN substrate – LEO [28] Pseudopotential LDA [24] FP-LMTO LDA [25] Powder [29] Pseudopotential LDA [24] FP-LMTO LDA [25]
3.1106 3.1130 3.110 3.06 3.084 3.189 3.1892 3.1893 3.1878 3.1896 3.162 3.17 3.538 3.501 3.53
4.9795 4.9816 4.980 4.91 4.948 5.1864 5.1850 5.1851 5.1854 5.1855 5.142 5.13 5.703 5.669 5.54
GaN
InN
LDA: local density approximation; FP-LMTO: full-potential linear muffin–tin orbital.
j11
j 1 General Properties of Nitrides
12
Figure 1.9 The bandgaps of nitrides, substrates commonly used for nitrides, and other conventional semiconductors versus their lattice constants. (Please find a color version of this figure on the color tables.)
substrates commonly used for nitrides, and other conventional semiconductors are shown in Figure 1.9 with respect to their lattice constants. All III nitrides have partially covalent and partially ionic bonds. The concept of fractional ionic character (FIC) is useful in interpreting many physical phenomena in the crystals [31,32]. The FIC may be defined for a binary compound AB as FIC ¼ jQ A Q B j=jQ A þ Q B j, where Q A and Q B are effective charges on atoms A and B. The FIC values range from zero for a covalent compound (each atom has four electrons) to 1 for an ionic compound (all eight electrons belong to the anion). Figure 1.10 displays the charge distribution along the AB bond for all three compounds. The arrow along the bond charge indicates the atomic boundaries in the crystals that are not always at the minimum of the line charge along the bond AB. This should be expected taking into account the partial covalent bond of the compounds, because only in the ionic crystals, the atomic boundary is clearly defined. Table 1.4 lists the calculated effective radii, rIII and rN, the effective charges, and FIC for AlN, GaN, and InN. The ionicity of AlN is high. This may explain the difficulties with AlN doping. It is well known that only covalent semiconductors or semiconductors with a large covalent component can form hydrogen-like shallow levels in the bandgap by substitution of a host atom with a neighbor with one more or one less electron. GaN and InN have a smaller than AlN but nearly equal ionicity. GaN was doped both p- and n-type. Thus, one can expect that InN can also be doped n- and p-type. To date, only n-type InN has been obtained because of high volatility of nitrogen and easiness of nitrogen vacancy formation that acts as a donor in this compound.
1.1 Crystal Structure of Nitrides 2.0
AlN
1.5 1.0 0.5 0.0 0.0
0.5
1.0 r(Al-N)(λ)
1.5
2.0
1.5
2.0
2.0
ρ(r) (electrons/a.u.3)
GaN 1.5 1.0 0.5 0.0 0.0
0.5
1.0 r(Ga-N)(λ)
2.0 InN 1.5 1.0 0.5 0.0 0.0
0.4
0.8
1.2
1.6
2.0
r(In-N)(λ) Figure 1.10 Charge density along the III–N bond in III nitride semiconductors.
The III-nitrides are commonly grown on mismatched substrates because of the lack of suitable native substrates. Thus, the epitaxial layers are strained during cool down, if they are sufficiently thick for them to relax at the growth temperature. The mechanical forces related to strain dramatically change the band structure of the epitaxial layers. The pressure dependence of the bandgap energy Eg can be expressed as Eg ¼ Eg(0) gP þ dP2, where Eg(0) is the bandgap of stress-free semiconductor, g and d are the Table 1.4 Calculated ionic radii (Å), effective charges (electrons), and the fractional ionic character (FIC) for III nitrides [32].
Compound
rIII (Å)
rN (Å)
QIII (e)
QN (e)
FIC
AlN GaN InN
0.8505 0.9340 1.0673
1.0523 1.0119 1.0673
1.12 1.98 1.83
6.88 6.02 6.17
0.72 0.51 0.54
j13
j 1 General Properties of Nitrides
14
3.8
Energy gap (eV)
3.7
3.6
3.5
3.4
3.3
-2
0
2
4
6
8
10
Pressure (GPa) Figure 1.11 Pressure dependence of the GaN energy gap, showing the typical sublinear character. Solid line represents the calculations of Christiansen and Gorczyca [35], which have been rigidly upshifted by 0.82 eV for a better fit with experiments, and the squares represent experimental results [6].
pressure coefficients, and P is the pressure. For GaN, g and d parameters are 4.2 · 103 and 1.8 · 105, respectively [33,34]. The bandgap is in terms of eVand the pressure is in terms of kbar. The pressure dependence has, in general, a sublinear character. The variation of the GaN energy gap with pressure, both theoretical [35] and experimental [6], is shown in Figure 1.11. The calculated pressure coefficients for III nitrides are given in Table 1.5. Parameters associated with mechanical properties of GaN in wurtzitic phase are tabulated in Tables 1.6 and 1.7, the latter dealing with the sound wave velocity. The same parameters for the zinc blende phase of GaN are tabulated in Tables 1.8 and 1.9.
Table 1.5 Calculated pressure coefficients for III nitrides including wurtzitic, zinc blende, and rock salt phases (g in units of meV GPa1 and d in units of meV GPa2) [35].
Zinc blende polytype
Wurtzite polytype
Rock salt polytype
Compound
G
d
g
d
g
d
AlN GaN InN
40 39 33
0.32 0.32 0.55
42 40 16
0.34 0.38 0.02
43 39 41
0.18 0.32 0.08
It should be noted that rock salt phases cannot be synthesized and exist only under high pressures beyond the phase transition point. g and d parameters with values of Eg ¼ Eg(0) þ gP þ dP2.
1.1 Crystal Structure of Nitrides Table 1.6 Parameters related to mechanical properties of wurtzitic GaN (in part after Ref. [36]).
Wurtzite polytype GaN
Parameter value/comments
Group of symmetry Molar volume, Vc (cm3 mol1) Molecular mass (g mol1) Density (g cm3)
C 46v (P63mc) 13.61 83.7267 6.11 or 6.15
Number of atoms in 1 cm3 Lattice constants
8.9 · 1022 a ¼ 3.1893 Å for powder, c ¼ 5.1851 Å for powdera 210 [38]b or 20.4 · 1011 dyn cm2 (204 GPa) 4 150 0.23 0.06 (0.198–0.37) from C parameters 15.5 1200–1700
Bulk modulus B (GPa) (compressibility1) dB/dP Youngs modulus (GPa) Poissons ratio, n or s0 (n ¼ C13/(C11 þ C12)) Knoops hardness (GPa) Surface microhardness (kg mm2) Nanoindentation hardness (GPa) Yield strength (GPa) Deformation potential, Eds C11 (GPa) C12 (GPa) C13 (GPa)
C33 (GPa) C44 (GPa)
10.8 at 300 K 0.1 at 1000 K 8.54 eV unscreened, 12 eV screened 390 15, 29.6 · 1011 dyn cm2 (296 GPa) 145 20, 13.0 · 1011 dyn cm2 (130 GPa) 106 20, 15.8 · 1011 dyn cm2 (158 GPa)
Comments/ references
The latter by Bougrov et al. [37]
The latter by Bougrov et al. [37]
[39] At 300 K 300 K, using Knoops pyramid test [36,40,41]
[38,42] The second set is from Ref. [43] The second set is calculated from the mean square displacement of the lattice atoms measured by X-ray diffraction
398 20, 26.7 · 1011 dyn cm2 (267 GPa) 105 10, 2.41 · 1011 dyn cm2 (241 GPa)
nh0 0 0 1i ¼ (Da/arelaz)/(Dc/crelax) or nh0 0 0 1i ¼ (Da/a0)/(Dc/c0) with Da ¼ ameas arelax and Dcmeas crelax. Conversion: 1 dyn cm2 ¼ 0.1 Pa (i.e., 1 GPa ¼ 1010 dyn cm2). For details of elastic constants and piezoelectric constants, see Tables 2.27 and 2.28, and at 300 K Bs ¼ 210 10 GPa, Bs ¼ [C33(C11 þ C12) 2(C13)2]/[C11 þ C12 þ 2C33 4C13]. a See Section 1.2.2 for details and lattice parameter for GaN on different substrates b Average of Voigt and Reuss bulk modulus.
j15
j 1 General Properties of Nitrides
16
Table 1.7 Wave propagation properties in wurtzitic GaN [36].
Wave propagation direction [1 0 0]
[0 0 1]
Wave character VL (longitudinal) VT (transverse, polarization along [0 0 1]) VT (transverse, polarization along [0 1 0]) VL (longitudinal) VT (transverse)
Expression for wave velocity
Wave velocity (in units of 105 cm s1)
(C11/r)1/2 (C44/r)1/2
7.96 4.13
(C11 C12)/2r)1/2
6.31
(C33/r)1/2 (C44/r)1/2
8.04 4.13
Parameters associated with thermal properties of GaN in wurtzitic and zinc blende phases (expected to be identical or nearly identical – treated to be identical here) are tabulated in Table 1.10. The parameters associated with electrical and optical properties of wurtzitic GaN are tabulated in Table 1.11. The same parameters associated with the zinc blende phase of GaN are tabulated in Table 1.12. Table 1.8 Parameters related to mechanical properties of zinc blende GaN (in part after Ref. [36]).
Zinc blende polytype GaN
Parameter value/comments
Group of symmetry Molar volume, Vc, na, or O (cm3 mol1) Molecular mass (g mol1) Density (g cm3) Number of atoms in 1 cm3 Lattice constant (Å) Bulk modulus, B (GPa)
2 Tp d ðF43mÞ ffiffiffi ð 3a2 cÞ=4 ¼ 2:28310 23 cm3
dB/dP Youngs modulus (GPa)
1.936 · 1023 6.15 8.9 · 1022 a ¼ 4.511 4.52 Bs ¼ 204 [36], 201 (theory) [45], 237 [46], 200 [47] 3.9, 4.3 181 [36]
Shear modulus, C 0 (GPa) Poissons ratio, n or s0
67 [36] 0.352 [36]
Knoops hardness Surface microhardness Nanoindentation hardness Yield strength Deformation potential, Eds C11 (GPa) C12 (GPa) C44 (GPa)
293 159 155
Bs ¼ [C33(C11 þ C12) 2(C13)2]/[C11 þ C12 þ 2C33 4C13] or Bs ¼
Comments/references
2
C12 Þ 2ðC 13 Þ Bs ¼ CC3311ðCþ11C12þ þ 2C 33 4C13
Y0 ¼ (C11 þ 2C12) (C11 C12)/(C11 þ C12) C0 ¼ (C11 C12)/2 n or s0 ¼ C13/ (C11 þ C12)
[42]
C 33 ðC 11 þ C 12 Þ 2ðC13 Þ2 . C11 þ C 12 þ 2C33 4C 13
1.1 Crystal Structure of Nitrides Table 1.9 Wave propagation properties in zinc blende GaN (after Ref. [36]).
Wave propagation direction
Wave character
Expression for wave velocity
Wave velocity (in units of 105 cm s1)
[1 0 0]
VL (longitudinal) VT (transverse)
(C11/r)1/2 (C44/r)1/2
6.9 5.02
[1 1 0]
VL (longitudinal) Vt//(transverse) V t? (transverse)
[(C11þCl2þ2C44)/2r]1/2 Vt// ¼ VT ¼ (C44/r)1/2 [(C11 C12)/2r]1/2
7.87 5.02 3.3
[1 1 1]
V l 0 V l
[(C11 þ 2C12 þ 4C44)/3r]1/2 [(C11 C12 þ C44)/3r]1/2
8.17 3.96
0
For the crystallographic directions, see Ref. [44].
Table 1.10 Parameters related to thermal properties of GaN, wurtzitic, and zinc blende phases are expected to be the in this respect with the exception of the first two parameters, which are for the wurtzitic phase (in part Ref. [36]).
GaN
Parameter value/comments
Comments/references
Temperature coefficient (eV K1) Thermal expansion (K1)
dEg/dT ¼ 6.0 · 104
Wurtzite structure only
Da/a ¼ 5.59 · 106, a|| ¼ aa ¼ 5.59 · 106 (wurtzite structure) [48]
Dc/c ¼ 3.17 · 106; for a plot versus temperature, see Ref. [49] (wurtzite structure only) For low dissociation material (106 cm2)
Thermal conductivity k (W cm1 K1) Debye temperature (K) Melting point ( C) Specific heat (J g1 C1) Thermal diffusivity (cm2 s1) Heat of formation, DH298 (kcal mol1) Heat of atomization, DH298 (kcal mol1) Heat of sublimation (kcal mol1) Heat capacity (J mol1 K1) Specific heat (J mol1 K1) (298 K < T < 1773 K) Enthalpy, DH0 (kcal mol1) Standard entropy of formation, DS0 (cal mol1 K1)
11.9 at 77 K, 2.3 at 300 K, 1.5 at 400 K 600 >1700 (at 2 kbar), 2500 (at tens of kbar) 0.49 0.43 26.4
[50] [37] [37]
203 72.4 0.5 35.4 at 300 K Cp ¼ 38.1 þ 8.96 · 103T
[51]
37.7 32.43
The specific heat Cp of Wz GaN at constant pressure for 298 K < T < 1773 K is Cp ¼ 38.1 þ 8.96 · 103T (J mol1 K1) [51].
j17
j 1 General Properties of Nitrides
18
Table 1.11 Parameters related to electrical and optical properties
of Wz GaN (in part after Refs [36,44]).
Wurtzite polytype GaN
Parameter value/comments
Bandgap energy, Eg (eV), direct Breakdown field (cm1) Electron affinity (eV) Energy separation between G and M–L valleys (eV)
3.42 at 300 K, 3.505 at 1.6 K
Energy separation between M–L valleys degeneracy (eV) Energy separation between G and A valleys (eV) Energy separation between A valley degeneracy (eV) Index of refraction
Dielectric constants (static) Dielectric constants (high frequency)
Optical LO phonon energy (meV) A1-LO, nA1(LO) (cm1)
Comments/ references
3–5 · 106 at 300 K 4.1 1.9 at 300 K
[53] [37] [37]
1 at 300 K 0.6 at 300 K
[52] [37]
0.6 at 300 K 1.3–2.1 at 300 K
[52] [37]
2 at 300 K 1 at 300 K
[52] [37]
0.2 at 300 K n (1 eV) ¼ 2.35 or 2.3 2.29, n (3.42 eV) ¼ 2.85 at 300 K (extrapolated to 0 eV), E?c interference method (the value for E||c is 1.5(2)% lower at 500 nm); also see energy dependence and long-wavelength value [54] 10.4 (E||c) 9.5 (E?c) 8.9 in c-direction (E||c) at 300 K 5.35 5.8 (E||c) at 300 K 5.35 (E?c) at 300 K 5.47 (E||c) 91.2
[52]
[55] [55] [37] [37] [56] [55] [37]
710–735
[57]
744
A1-TO, nA1(TO||) (cm1) E1-LO, nE1(LO? ) (cm 1)
533–534 741–742
[34] [58]
533 746
E1-TO, nE1(TO? ) (cm 1) E2 (low) (cm1) E2 (high) (cm1)
556–559 143–146 560–579
[59]
559
Reflectivity [55] Raman [56] Reflectivity [55] Raman [60]
1.1 Crystal Structure of Nitrides Table 1.11 (Continued)
Wurtzite polytype GaN
Parameter value/comments
Energy of spin–orbital splitting, Eso (meV)
11 (þ5, 2) at 300 K calculated from the values of energy gap Eg,dir (given in this table) 40 at 300 K
Energy of crystal-field splitting, Ecr (meV)
Effective electron mass, == me or me
Effective electron mass, me? or m? e Effective hole mass Effective hole masses (heavy), mhh
Effective hole masses (light)
Effective hole masses (split-off band), ms
Effective mass of density of state, mv Effective conduction band density of states (cm3) Effective valence band density of states (cm3) Electron mobility (cm2 V1 s1) Hole mobility (cm2 V1 s1) n-doping range (cm3)
22 (2), calculated from the values of energy gap Eg,dir (given in this table) 0.20m0 at 300 K
Comments/ references [61]
[37] [61] [37]
0.20m0 0.27m0 by Faraday rotation 0.138–0.2 0.20m0, 300 K; fit of reflectance spectrum
[62] [52]
0.15–0.23m0 0.8m0 at 300 K mhh ¼ 1.4m0 at 300 K
[52] [64] Calculated
==
mhhz ¼ mhh ¼ 1:1m0 at 300 K mhh? ¼ m? hh ¼ 1:6m0 at 300 K == mhh ¼ 1:1 2:007m0 m? hh ¼ 1:61 2:255m0 mlh ¼ 0.3m0 at 300 K ==
mlhz ¼ mlh ¼ 1:1m0 at 300 K m? lh ¼ mlh? ¼ 0:15m0 at 300 K == mlh ¼ 1:1 2:007m0 mlh? ¼ 0:14 0:261m0 msh ¼ 0.6m0 at 300 K ==
mshz ¼ mch ¼ 0:15m0 at 300 K msh? ¼ m? ch ¼ 1:1m0 at 300 K == msh? ¼ mch ¼ 0:12 0:16m0 ¼ 0:252 1:96m0 m? ch 1.4m0
[63]
[15] [70] [52] [52] Calculated [15] [70] [52] [52] Calculated [36] [70] [52] [52] [37]
2.3 · 1018 at 300 K
4.6 · 1019 at 300 K 1400 experimental at 300 K <20
50 000 at 20 K [65] At 300 K
1016 cm3–high 1019
(Continued )
j19
j 1 General Properties of Nitrides
20
Table 1.11 (Continued)
Wurtzite polytype GaN
Parameter value/comments
Comments/ references
p-doping range (cm3) Diffusion coefficient for electrons (cm2 s1) Diffusion coefficient for holes (cm2 s1)
1016 cm3–mid 1018 25
[36]
5, 26, 94
[36,53]
The details of the energies of high symmetry points compiled by Fritsch et al. [52] are given in Table 2.1. Dependence of the bandgap on hydrostatic pressure: Eg ¼ Eg(0) þ gP þ dP2, where Eg(0) is the bandgap of stress-free GaN. The values for g parameter are 39–42 meV GPa1 and the same for d parameter are 0.18 to 0.32 meV GPa2. For others, see Table 1.5.
The phonon energies are discussed later on in Table 1.27 in detail and effective masses are discussed in Chapter 2. More details of effective masses can be found in Table 2.9. More should be said about the dielectric constants. Electromagnetic theory indicates that for any longitudinal electromagnetic wave to propagate, the dielectric function e(o) must vanish. Doing so leads to [66] eðwÞ w2LO w2 ; ¼ eð¥Þ w2TO w2
ð1:5Þ
where oLO and oTO represent the transverse optical (TO) and longitudinal optical (LO) phonon frequencies and e(o) and e(1) represent the low and high (optical) frequency dielectric constants. The phonon branches associated with a wurtzitic symmetry are discussed in Section 1.2.2 dealing with the mechanical properties of GaN. When o ¼ oLO, the dielectric function vanishes, e(oLO) ¼ 0. Equation 1.5 can be expanded to the directional dependence of the dielectric function in nitrides in general and GaN in particular. In the direction parallel to the c-axis or the z-direction, from the G point to the A point, in the k-space, (with x, y representing the in-plane coordinates), the lowand high-frequency dielectric functions are related each with the help of A1(LO) and E1(TO) phonons through [67] e== ðwÞ ¼ e¥?
w2 w2== ðLOÞ w2 w2== ðTOÞ
:
ð1:6Þ
Likewise, Equation 1.5 can be expanded in the direction perpendicular to the c-axis or in the basal plane or the (x, y) plane, the z-direction (in k-space between the G point and M (1/2, 0, 0) or K (1/3, 1/3, 0) points), the low- and high-frequency dielectric functions are related each with the help of A1(TO) and E1(LO) phonons through e? ðwÞ ¼ e¥?
w2 w2? ðLOÞ ; w2 w2? ðTOÞ
ð1:7Þ
where ? and // indicate in the basal plane and along the c-direction, respectively.
1.1 Crystal Structure of Nitrides Table 1.12 Parameters related to electrical and optical properties
of zinc blende GaN (in part after Refs [36,44]). Zinc blende polytype GaN
Parameter/comments
Comments/references
Bandgap energy (eV) Breakdown field (V cm1) Index of refraction Dielectric constant (static)
3.2–3.28 at 300 K 5 · 106 n (at 3 eV) ¼ 2.9, 2.3 9.7 at 300 K == 9.2 by ð2e? 0 þ e0 Þ=3 of wurtzitic form 5.3 at 300 K
3.302 at low temperature [36]
Dielectric constant (high frequency) Energy separation between G and X valleys, EG (eV) Energy separation between G and L valleys, EL (eV) Spin–orbit splitting in valence band, Dso or Eso (eV) Effective electron mass, me Effective hole masses (heavy)
1.4 1.1 1.6–1.9 2 0.02 0.017 0.13m0 0.14m0 mhh ¼ 1.3m0, ½110 mhh ¼ 1:52m0 m[100] ¼ 0.8m0,
[37] [52] Using Lyddane–Sachs–Teller relation (e0/ehigh ¼ o2LO/ o2TO) [36] [52] [36] [52] At 300 K [36] [69] At 300 K [37] [52] At 300 K [36,70]
½100
mhh ¼ 0:84m0 m[111] ¼ 1.7m0, ½111
mhh ¼ 2:07m0 Effective hole masses (light)
Effective hole masses (split-off band), ms, mch, or mso
Effective conduction band density of states Effective valence band density of states Electron mobility (cm2 V1 s1) Hole mobility (cm2 V1 s1) Diffusion coefficient for electrons (cm2 s1) Diffusion coefficient for holes (cm2 s1) Electron affinity Optical LO phonon energy (meV)
Second set of figures are from Ref. [52], which are deemed more reliable
mlh ¼ 0.19m0, ½110 mlh ¼ 0:20m0 m[100] ¼ 0.21m0, ½100 mlh ¼ 0:22m0 m[111] ¼ 0.18m0, ½111 mlh ¼ 0:19m0 msh ¼ 0.33m0, mso ¼ 0.35m0 m[100] ¼ 0.33m0 m[111] ¼ 0.33m0 1.2 · 1018 cm3
At 300 K [37]
4.1 · 1019 cm3
At 300 K [37]
1000 at 300 K 350 at 300 K 25
[36] [36] [36]
9, 9.5, 32
[36]
4.1 eV 87.3
[37] At 300 K
The details of the energies of high symmetry points are given in Table 2.1.
j21
j 1 General Properties of Nitrides
22
For wurtzitic GaN, the various directional components of phonon frequencies are o?(LO) ! E1(LO) ¼ 91.8 meV, oz(LO) ¼ o//(LO) ! A1(LO) ¼ 91 meV, o?(TO) ! E1(TO) ¼ 69.3 meV, and oz(TO) ¼ o//(TO) ! A1(TO) ¼ 66 meV. In the z-direction (along the c-direction) and perpendicular to the z-direction (in basal plane), LO and TO phonons are not mixed. For any direction other than the in-plane and out-of-plane configurations, the LO and TO phonons mix and hybridize. For a given propagation direction with an angle y relative to the c-axis (0z), one finds three phonon branches. One is an ordinary TO phonon mode with atomic displacement in the (0xy) plane. The other two branches have a mixed TO and LO character and their dielectric functions are given by the solutions of [68] e== cos2 q þ e? sin2 q ¼ 0:
ð1:8Þ
Using the above relationship, the phonon energy as a function of the angle can easily be calculated. Doing so leads to the conclusion that the upper branch (LO-like) remains between A1(LO) and E1(LO) energies, whereas the lower branch (TO-like) remains between A1(TO) and E1(TO) energies. Therefore, the dispersion remains small compared to the LO–TO separation, owing to the relatively small cell asymmetry and the large ionicity of atomic bonds. A more important consequence of LO–TO mixing is that the TO-like mode becomes coupled to carriers whereas in the c-direction A1(LO) mode and in the basal plane E1(LO) phonons couple to the carriers. For the special case o ¼ 0 (or very small frequencies compared to the LO and TO phonon frequencies), the relationship between the optical and static dielectric constants reduces to the well-known Lyddane–Sach–Teller relationship eðwÞ w2LO ; ¼ eð¥Þ w2TO
ð1:9Þ
which will be used to determine the optical frequency dielectric constant from the knowledge of A1(LO) and A1(TO) phonon frequencies along the c-direction and E1(LO) and E1(TO) in the basal plane. This relationship is used very often. Parameters related to the energy bandgap, carrier mass, and mechanical properties of AlN have been determined [71–76]. Extensive data on all the binary and ternary band structure parameters can be found in Chapter 2. For example, additional parameters on the critical point energies for Wz GaN, AlN, and InN are given in Tables 2.1–2.3, respectively. Tables 2.4–2.6 list the critical point energies for ZB GaN, AlN, and InN, respectively. Effective masses and other band parameters for Wz GaN are listed in Table 2.9. Table 2.10 tabulates the Luttinger band parameters for ZB GaN. Table 2.14 lists the effective band parameters for Wz AlN, whereas the effective masses and band parameters for Wz AlN are tabulated in Table 2.15. Luttinger parameters for ZB AlN are listed in Table 2.16. The band parameters and effective masses for Wz InN are tabulated in Tables 2.19 and 1.20, respectively. Returning to the content of this chapter, parameters associated with the mechanical properties of AlN in wurtzitic and zinc blende phases are tabulated in Tables 1.13 and 1.14, respectively. The parameters related to the sound wave velocity in wurtzitic AlN are listed in Table 1.15.
1.1 Crystal Structure of Nitrides Table 1.13 Parameters related to mechanical properties of wurtzitic AlN (in part after Ref. [36]).
Wurtzite AlN
Parameter/comments
C46v (P63 mc) 9.58 · 1022 12.47 40.9882 3.28 3.255 g cm3 by X-ray 3.23 g cm3 by X-ray Lattice constants a ¼ 3.112 Å, c ¼ 4.979–4.982 Å Bulk modulus, B (GPa) 159.9–210.1, 21 · 1011 dyn cm2 (210 GPa) (Bs ¼ 210) dB/dP 5.2–6.3 374, 308 Youngs modulus, E or Y0 (GPa) 0.18–0.21 Poissons ratio, n or s0 {0001}, c-plane Poissons ratio s0 along the different crystallographic f1 1 2 0g, a-plane ðl ¼ h0001i; m ¼ h1 1 0 0iÞ directions f1 1 2 0g, a-plane ðl ¼ h1 1 0 0i; m ¼ h0001iÞ Knoops hardness (GPa) 10–14 at 300 K Nanoindentation hardness (GPa) 18 Yield strength (GPa) 0.3 at 1000 C Surface microhardness on basal 800 kg mm2 by 300 K, using Knoops pyramid test plane (0 0 0 1) 410 10 C11 (GPa)a 149 10 C12 (GPa)a 99 4 C13 (GPa)a 389 10 C33 (GPa)a 125 5 C44 (GPa)a Velocity of the longitudinal sound 10 127 m s1 waves, vl 6333 m s1 Velocity of the shear waves, vs
Comments/references
Group of symmetry Number of atoms in 1 cm3 Molar volume, Vc (cm3 mol1) Molecular mass (g mol1) Density (g cm3)
Longitudinal elastic modulus, Cl Shear elastic modulus, Cs
[77] [78] [78]
The latter from Ref. [79] 0.287 Ref. [80] 0 Ref. [80] 0.216 Ref. [80] [81] [82] [36,40,41] [42,83] Refer to Table 1.29 as well Refer to Table 1.29 as well Refer to Table 1.29 as well Refer to Table 1.29 as well [79] The sound velocities and related elastic module (experimental data)
334 GPa 131 GPa
Conversion: 1 dyn cm 2 ¼ 0.1 Pa (i.e., 1 GPa ¼ 1010 dyn cm2). See Table 1.29 for more details. For details of elastic constants and piezoelectric constants, see Tables 2.27 and 2.28. This expression is given already in conjunction with Table 1.6 and it is also Bs ¼ [C33(C11 þ C12) 2 (C13)2]/[C11 þ C12 þ 2C33 4C13]. Temperature derivatives of the elastic module: dln Cl/ dT ¼ 0.37 · 104 K1; dln Cs/dT ¼ 0.57 · 104 K1; dln Bs/dT ¼ 0.43 · 104 K1. a See Table 1.29 for a more in-depth treatment of these parameters.
Parameters associated with thermal properties of wurtzitic AlN are tabulated in Table 1.16. Parameters associated with electrical and optical properties of wurtzitic AlN are tabulated in Table 1.17. The same range of parameters associated with the zinc blende
j23
j 1 General Properties of Nitrides
24
Table 1.14 Parameters related to mechanical properties of zinc blende AlN.
Not much is known about the zinc blende phase
Zinc blende AlN Lattice constant (Å) Bandgap (eV) E Xg ðeVÞ E Lg ðeVÞ Bulk modulus, B (GPa) Youngs modulus (GPa)
a ¼ 4.38 5.4, indirect 4.9 9.3 228
Shear modulus (GPa) Poissons ratio, n or s0 C11 (GPa)
348
304
C12 (GPa) C44 (GPa) me ðGÞ ml ðXÞ mt ðXÞ
168 135 0.25 0.53 0.31
160 193
Comments/references
[84] [84] [84] Bs ¼ (C11þ2C12)/3 Y0 ¼ (C11 þ 2C12) · (C11 C12)/(C11 þ C12) C ¼ (C11 C12)/2 n or s0 ¼ C12/(C11þC12) The latter figures are from Ref. [84]
[84] [84] [84]
See Table 2.18 for more details.
phase of GaN is tabulated in Table 1.18. Mechanical, phonon, properties of epitaxial AlN (deposited on silicon and sapphire substrates at 325 K by ion beam assisted deposition (IBAD)) have been investigated by Ribeiro et al. Raman scattering measurements revealed interesting features related to the atomic composition and structure of the films [94]. Vibrational modes corresponding to 2TA(L) at 230 cm1, 2TA(X) at 304 cm1, 2TA(S) at 435 cm1, TO(G) at 520 cm1, TA(S) þ TO(S) at 615 cm1, accidental critical points at 670 and 825 cm1, 2TO(D) at 950 cm1, 2TO(L) at 980 cm1, 2TO(G) at 1085 cm1, 2TA(X) þ 2TO(G) at 1300 cm1, and 3TO(G) at 1450 cm1 have been observed. While identifying the vibrational modes, one should be wary of the peak at 2330 cm1 caused by the molecular nitrogen on Table 1.15 Acoustic wave propagation properties in wurtzite AlN [36].
Wave propagation direction [1 0 0]
[0 0 1]
Wave character VL (longitudinal) VT (transverse, polarization along [0 0 1]) VT (transverse, polarization along [0 1 0]) VL (longitudinal) VT (transverse)
For the crystallographic directions, see Ref. [44].
Expression for wave velocity (C11/r)1/2 (C44/r)1/2 (C11-C12)/2r)1/2 (C33/r)1/2 (C44/r)1/2
Wave velocity (in units of 105 cm s1) 11.27 6.22 6.36 10.97 6.22
1.1 Crystal Structure of Nitrides Table 1.16 Parameters related to thermal properties of wurtzitic AlN (in part after Ref. [36]).
Wurtzite polytype AlN
Value
Comments/references
Thermal expansion (K1)
Da/a ¼ a|| ¼ aa ¼ 4.2 · 106, Dc/c ¼ aort ¼ ac ¼ 5.3 · 106 Da/a ¼ 2.9 · 106, Dc/c ¼ 3.4 · 106 aort ¼ ac ¼ 5.27 · 106, a|| ¼ aa ¼ 4.15 · 106
[48,85–87]
Thermal conductivity (W cm1 K1) Thermal diffusivity (W cm1 C1) Debye temperature (K) Melting point (K)
Specific heat (J g1 C1) Thermal diffusivity (cm2 s1) Heat of formation, DH298 (kcal mol1) Heat of atomization, DH298 (kcal mol1) Free energy, DG298 (kcal mol1)
k ¼ 2.85–3.2 2.85 at 300 K 950, 1150 3273 3023 (between 100 and 500 atm of nitrogen) 3487 (2400 C at 30 bar) 0.6
[88] T ¼ 20–800 C. X-ray, epitaxial layers, by Sirota and Golodushko [89], also see Ref. [48] [90]; later results by Slack et al. [91] [90]
[92] [78]
See Figure 1.29 and the expressions below this table
1.47 64 209.7 68.15
For 293 < T < 1700 K, Da/a300 ¼ 8.679 · 102 þ 1.929 · 104T þ 3.400 · 107T2 7.969 · 1011T3. For 293 < T < 1700 K, Dc/c300 ¼ 7.006 · 102 þ 1.583 · 104T þ 2.719 · 107T2 5.834 · 1011T3. The specific heat Cp of AlN for constant pressure: for 300 < T < 1800 K, Cp ¼ 45.94 þ 3.347 · 103T 14.98 · 105T2 (J mol1 K1); for 1800 < T < 2700 K, Cp ¼ 37.34 þ 7.866 · 103T (J mol1 K1). After Ref. [93]. Optical emission measurements indicate the FXA transition at 6.023 with an associated binding energy of 63 meV, which sets the bandgap of Wz AlN at 6.086.
the surface of c-Si [95]. It is worth noting that, owing to the extremely weak Raman signal usually presented by AlN films, it is not uncommon to ascribe some of the features erroneously to AlN [96]. Conduction band first- and second-order pressure derivatives [36]: Eg ¼ Eg(0) þ 3.6 · 103P 1.7 · 106P2 (eV) [121] EM ¼ EM(0) þ 7.5 · 104P þ 1.0 · 106P2 (eV) EL ¼ EL(0) þ 8.0 · 104P þ 6.9 · 107P2 (eV) Ek ¼ Ek(0) þ 6.3 · 104P þ 1.7 · 106P2 (eV) where P is pressure in kbar.
j25
j 1 General Properties of Nitrides
26
Table 1.17 Parameters related to optical and electrical properties
of wurtzitic AlN [97–102] (in part from Ref. [36]). Wurtzite polytype AlN
Parameter
Comments/references
Bandgap energy (eV) From the dichroism of the absorption edge, it follows that the G1 0 state lies slightly above the G6 state (transition E||c (G1 0 v Glc) at lower energy than transition E ? c (G6v Glc)), both states being split by crystal field interaction [105]
6.026 at 300 K 6.2 eV at 300 K
[103,104] Excitonic contribution near direct edge [105]
6.23 at 77 K
Excitonic contribution near direct edge [105] Excitonic edge assuming exciton binding energy of 75 meV [106] With a free exciton binding energy of 63 meV [107,108]
6.28 at 300 K
6.086 at 5 K
Breakdown field (V cm1) dEg/dP (eV bar1) Conduction band energy separation between G and M–L valleys (eV) Conduction band energy separation between G and M–L valleys Conduction band energy separation between M–L valleys degeneracy (eV) Conduction band energy separation between G and K valleys (eV) Conduction band K valley degeneracy (eV)
Valence band energy of spin–orbital splitting, Eso (eV)
6.0 at 300 K 6.1 at 5 K 1.2–1.8 · 106 3.6 · 103 0.7 1
[36] [109,110] [78] [52]
0.6
[78]
0.2
[52]
1.0
[78]
0.7 2
[52] [78]; empirical pseudopotential calculations of Fritsch et al. [52] do not show degeneracy at this critical point
0.019 at 300 K 0.036
[108]
1.1 Crystal Structure of Nitrides Table 1.17 (Continued)
Wurtzite polytype AlN
Parameter
Comments/references
Valence band energy of crystal field splitting, Ecr (eV), G7 on top of G9 Effective conduction band density of states (cm3) Effective valence band density of states (cm3) Index of refraction Dielectric constant (static)
0.225
[108]
6.3 · 1018
[78]
4.8 · 1020
[78]
Dielectric constant (high frequency)
4.68 4.77 4.84 at 300 K 4.35 for E//c (modeling) 4.16 for E?c (experiment) 2.1–2.2 at 300 K
Infrared refractive index
Effective electron mass, me
Effective hole masses (heavy) == For kz direction mhz or mhh For kx direction mhx or m? hh
Effective hole masses (light) == For kz direction mlz or mlh For kx direction mlx or m? lh Effective hole masses (splitoff band) ==
For kz direction msoz or mch For kx direction msox or
n (3 eV) ¼ 2.15 0.05 9.14 at 300 K 7.34 8.5 0.2 at 300 K 9.32 for E//c (modeling) 7.76 for E?c (experiment) 4.6 at 300 K
m? ch
1.9–2.1 at 300 K 1.8–1.9 at 300 K 3 for E//c (modeling) 2.8 for E?c (experiment) 0.27 and 0.35m0 0.25–0.39m0 0.4m0 at 300 K == me ¼ 0:231 0:35m0 m? e ¼ 0:242 0:25m0 == mhh ¼ 3:53m0 at 300 K 2.02–3.13m0 at 300 K
By reflectivity [111] [78,112] [113] [113] [111] [78] Reflectivity [113] [113] Epitaxial films and monocrystal Polycrystalline films Amorphous films [87] [113] [113] [14,114,115] [107] [52]
m? hh ¼ 10:42m0 at 300 K == mhh ¼ 1:869 4:41m0 m? hh ¼ 2:18 11:14m0 3.53m0, 0.24m0 == mlh ¼ 1:869 4:41m0 m? lh ¼ 0:24 0:350m0 0.25m0 at 300 K
[116] [107]; from Mg binding energy [116] [52] [52] At 300 K [116] [52] [52] [116]
3.81m0 at 300 K == mch ¼ 0:209 0:27m0
[52]
m? ch
[52]
¼ 1:204 4:41m0
(Continued )
j27
j 1 General Properties of Nitrides
28
Table 1.17 (Continued)
Wurtzite polytype AlN
Parameter
Comments/references
Effective mass of density of state, mv Optical phonon energy (meV) nTO(E1) phonon wave number (cm1)b
7.26m0 at 300 K
[116]
nLO(E1) phonon wave number (cm1) nTO(A1) phonon wave number (cm1) nLO(A1) phonon wave number (cm1) n(E2) phonon wave number (cm1) nTO(E1) phonon wave number (cm1) nTO(A1) phonon wave number (cm1) nLO(E1) phonon wave number (cm1) nLO(A1) phonon wave number (cm1) n(1)(E2) phonon wave number (cm1) n(2)(E2) phonon wave number (cm1)
99.2 895
614
608
671.6
821
888.9
888
514
667.2
659.3
663
909
303a
426
657–673
First column [117]; second column [118]; third column [113] See Table 1.30 for more details
[87,111,119,120]
607–614 or 659–667 895–924 888–910 241–252 655–660
The details of the energies of high symmetry points are given in Table 2.2. See Table 1.30 for additional details for phonon wave numbers. More details of effective masses can be found in Table 3.15. Temperature dependence of energy gap: Eg ¼ Eg(0) – 1.799 · 103T2/(T þ 1462) (eV) by Guo and Yoshida [103]. a Room-temperature Raman, tentative. b For more details regarding vibrational modes, refer to Section 1.3.1.
Phase transition from the wurtzite phase to the rock salt structure (space group O5h ; lattice parameter 4.04 Å) takes place at the pressure of 17 GPa (173 kbar) [109,110]. Parameters associated with the electrical and optical properties of zinc blende AlN are listed in Table 1.18. For details regarding the Luttinger parameters for the valence band in zinc blende AlN, refer to Table 2.16. Parameters associated with the mechanical properties of wurtzitic InN are tabulated in Table 1.19. For wurtzite crystal structure, the surfaces of equal energy in G valley should be ellipsoids, but effective masses in the z-direction and perpendicular directions are estimated to be approximately the same.
1.1 Crystal Structure of Nitrides Table 1.18 Parameters related to optical and electrical properties of zinc blende AlN.
Zinc blende polytype of AlN
Value
Comments/references
Bandgap energy (eV)
All at 300 K [52]
Dielectric constant (static) Dielectric constant (high frequency) Energy separation between G and X valleys EG (eV) Energy separation between G and L valleys EL (eV) Spin–orbit splitting in valence band, Dso or Eso (eV) Deformation potential (eV) Effective electron mass, me
4.2; 6.0 5.8 All (theory) 9.56 4.46 0.7 0.5 2.3 3.9 0.019
[52] [84] [52] [84] [84]
9 0.23m0
[84] [52]
Effective hole masses (heavy)
mhh ¼ 1:02m0
Effective hole masses (light)
Effective hole masses (split-off band), ms, mch, or mso Luttinger parameter c1 Luttinger parameter c2 Luttinger parameter c3 a
½100
½111 mhh
¼ 2:64m0
½110 mhh
¼ 1:89m0
mlh
½100
¼ 0:37m0
½111 mlh
¼ 0:30m0
½110 mlh
¼ 0:32m0
a a
[52]
[52]
0.54m0
[52]
1.85 0.43 0.74
[52]
C. Persson, and A. Ferreira da Silva, Linear optical response of zinc-blende and wurtztie III-N (III ¼ B, AI, Ga, and In), Journal of Crystal Growth 305 pp. 408–413 (2007)
The parameters associated with thermal properties of wurtzitic InN are tabulated in Table 1.20. The specific heat Cp of InN at constant pressure for 298 K < T < 1273 K [51] is Cp ¼ 38.1 þ 1.21 · 102T (J mol1 K1). Refer to Table 1.31 for a detailed treatment of mechanical properties of InN. The parameters associated with electrical and optical properties of wurtzitic InN are tabulated in Table 1.21. Available parameters associated with the mechanical properties of zinc blende InN, primarily calculated, are tabulated in Table 1.22. Other parameters dealing with electrical and optical properties of zinc blende InN, primarily calculated, are listed in Table 1.23.
j29
j 1 General Properties of Nitrides
30
Table 1.19 Parameters related to mechanical properties of wurtzitic InN (in part after Ref. [36]).
Wurtzite InN
Value
Group of symmetry
C 46v (P63 mc)
Molar volume (cm3 mol1)
18.49
Molar mass (g mol1)
128.827
Density (g cm3)
6.89 6.98 6.81 a ¼ 3.548 a ¼ 3.5446 a ¼ 3.533 c ¼ 5.760 c ¼ 5.7034 c ¼ 5.693 165 140
Lattice constants (Å)
Bulk modulus B (GPa) dB/dP
3.8
Nanoindentation hardness(GPa)
11.2
Comments/references
Measured by displacement X-ray, 298.15 K [122] Epitaxial layers, X-ray [123]; 300 K [124] Epitaxial layers, X-ray [123]; 300 K [124]
[124]
[125]
Youngs modulus (GPa) Poissons ratio, n or s0 Knoops hardness (GPa) Deformation potential, Eds C11 (GPa) C12 (GPa) C13 (GPa) C33 (GPa) C44 (GPa)
Can be calculated using S parameters and Equation 2.10 0.82, 0.68 7.10 eV 223 115 92 224 48
C31 ¼ 70 205
Estimate [42] [42] [42] [42] [42]
Eg ¼ Eg(0) þ 3.3 · 102P (eV), where P is pressure in GPa [35,126]. For details of elastic constants and piezoelectric constants, see Table 2.19. Bs ¼ [C33(C11 þ C12) 2(C13)2]/ [C11 þ C12 þ 2C33 4C13].
1.2 Gallium Nitride
Despite the fact that GaN has been studied far more extensively than the other group III nitrides, further investigations are still needed to approach the level of understanding of technologically important materials such as Si and GaAs. GaN growth often suffers from large background n-type carrier concentrations because of native defects and, possibly, impurities. The lack of commercially available native substrates
1.2 Gallium Nitride Table 1.20 Parameters related to thermal properties of wurtzitic InN (in part after Ref. [36]).
Wurtzite polytype InN
Value
Temperature coefficient Thermal expansion
dEg/dT ¼ 1.8 · 104 eV K1 Da/a ¼ 2.70 · 106 K1; Dc/c ¼ 3.40 · 106 K1 Da/a ¼ 2.85 · 106 K1; Dc/c ¼ 3.75 · 106 K1 Da/a ¼ 3.15 · 106 K1; Dc/c ¼ 4.20 · 106 K1 Da/a ¼ 3.45 · 106 K1; Dc/c ¼ 4.80 · 106 K1 Da/a ¼ 3.70 · 106 K1; Dc/c ¼ 5.70 · 106 K1 aa ¼ 3.8 · 106 K1; ac ¼ 2.9 · 106 K1 0.8 0.2 W cm1 K1 0.45 W cm1 C1 1.76 W cm1 C1, 300 K (estimate for ideal InN) 4.6
Thermal conductivity
Heat of formation, DH298 (kcal mol1) (Wz) Heat of atomization, DH298 (kcal mol1) Melting point
Debye temperature Specific heat (J mol1 K1) Heat capacity, Cp (cal mol1 K1) Entropy, S0 (cal mol1 K1) TSFCw at formation, DH0f (kcal mol 1) TSFCa at formation, DS0f (kcal mol 1 K 1) TSFCw at formation, DG0f (kcal mol 1) TSFCw at fusion, DHm (kcal mol1) TSFCw at fusion, DSm (cal mol1 K1) N2 equilibrium vapor pressure
Comments/references
At 190 K At 260 K At 360 K At 460 K At 560 K [124] Estimate [127]
175 1373 K 2146 K, vapor pressure 105 bar at 1100–1200 C 660 K at 300 K 370 K at 0 K Cp ¼ 38.1 þ 1.21 · 102T 9.1 þ 2.9 · 103T 10.4 34.3, 30.5
[92]
25.3
Experimental 298.15 K
22.96
Experimental 298.15 K
14.0 10.19 cal mol1 K1 1 atm 105 atm
Theoretical Theoretical 800 K 1100 K
[124] [128] [51] 298–1273 K 298.15 K Experimental 298.15 K
a
TSFC: thermodynamic state function changes.
exacerbates the situation. These, together with the difficulties in obtaining p-type doping, and the arcane fabrication processes caused the early bottlenecks stymieing progress. Information available in the literature on many of the physical properties of GaN is in some cases still in the process of evolution, and naturally controversial. This
j31
j 1 General Properties of Nitrides
32
Table 1.21 Parameters related to electrical and optical properties
of wurtzitic InN (in part after Ref. [36]). Wurtzitic InN
Value
Comments/references
Bandgap energy, Eg (300 K)
1.89 eV, 1.5 eV, 0.78 eV
See Section 1.3.1 for an expan ded discussion
15.3 e0,ort ¼ 13.1
300 K [124] 300 K [128]
e0, || ¼ 14.4,
300 K [128]
8.4
300 K, using the Lyddane– Sachs–Teller relation (e0 =ehigh ¼ w2LO =w2TO ) [129,130] Heavily doped film, infrared reflectivity [131] [132] At 300 K [124] At 300 K and l ¼ 1.0 mm, interference method; n ¼ 3–1020 cm3 [131] At l ¼ 0.82 mm [131] At l ¼ 0.66 mm [131]
Electron affinity Dielectric constant (static) Dielectric constant (static, ordinary direction) Dielectric constant (static, extraordinary direction) Dielectric constant (high frequency)
9.3
Infrared refractive index
Energy separation between G and M–L valleys (eV) Energy separation between M–L valleys degeneracy Energy separation between G and A valleys (eV) Energy separation between A valley degeneracy Energy separation between G and G1 valleys (eV) Energy separation between G1 valley degeneracy (eV) Effective conduction band density of states Effective valence band density of states Valence band crystal field splitting, Ecr
5.8 2.9 2.56
2.93 3.12 Reported range: 2.80–3.05 2.9–3.9
300 K [124]
4.8 0.6
[52] 300 K [124]
0.7 0.7–2.7
[52] 300 K [124]
4.5 1
[52] 300 K [124]
0.6 1.1–2.6
[52] 300 K [124]
1
300 K [124]
9 · 1017 cm
300 K [124]
5.3 · 1019 cm3
300 K [124]
0.017 eV
300 K [124]
1.2 Gallium Nitride Table 1.21 (Continued)
Wurtzitic InN
Value
Comments/references
Valence band spin– orbital splitting, Eso Index of refraction
0.003 eV
300 K [124]
2.9 at 300 K 2.56 at 300 K (interference method; n ¼ 3–1020 cm3, l ¼ 1.0 mm) 2.93 3.12 0.11m0 == me ¼ 0:1 0:138m0 m? e ¼ 0:1 0:141m0 1.63m0 at 300 K 0.5m0 at 300 K == mhh ¼ 1:350 2:493m0 m? hh ¼ 1:410 2:661 m0 0.27m0 at 300 K == mlh ¼ 1:350 2:493m0 m? lh ¼ 0:11 0:196m0 0.65m0 at 300 K == mch ¼ 0:092 0:14m0 m? ch ¼ 0:202 3:422 1.65m0 at 300 K
[124] [131]
[131] [131] [133] [52] [52] [32,134,135] [136] [52] [52] [32,134,135] [52] [52] [32,134,135] [52] [52] [134,135]
73 at 300 K
[124]
Effective electron mass, me
Effective hole masses (heavy), mh
Effective hole masses (light), mlp Effective hole masses (split-off band), ms Effective mass of density of state, mv Optical LO phonon energy (meV)
The details of the energies of high symmetry points are given in Table 2.3. More details of effective masses can be found in Table 2.19.
Table 1.22 Available parameters for mechanical for zinc blende InN.
Zinc blende InN
Value
Lattice constant Density (g cm3) Bulk modulus (GPa) dB/dP Youngs modulus (GPa), Y0 or E
a ¼ 4.98 Å 6.97 138–155, 145.6 [42]
Shear modulus (GPa) Poissons ratio, n or s0 C11 (GPa) C12 (GPa) C44 (GPa)
187 125 86
Comments/references
Derived from X-ray data Bs ¼ (C11 þ 2C12)/3 3.9–4.0 Y0 ¼ (C11 þ 2C12) (C11 C12)/(C11 þ C12) C ¼ (C11 C12)/2 n or s0 ¼ C12/(C11 þ C12) See Ref. [42] See Table 1.31 for details
j33
j 1 General Properties of Nitrides
34
Table 1.23 Available electrical and optical properties of zinc blende InN, primarily calculated.
Zinc blende InN
Value
Comments/references
Bandgap energy, Eg (300 K)
2.2 eV
In the absence of any reliable data, the bandgap to a first extent can be assumed to be similar to that for Wz InN. See Ref. [137] for a detailed treatment
Dielectric constant
1.5–2.1 (theory) 0.2 eV below the Wz polytype 8.4
2.88 0.30 2.90 0.30 2.90 3.05 0.30 2.65 6.97 3
LWL == ð2e? 0 þ e0 Þ of wurtzitic By 3 form (the spur) Using Lyddane–Sachs–Teller relation (e0 =ehigh ¼ w2LO =w2TO ) Theory — Transmission interference Transmission interference NIRSR Derived from X-ray data [52]
2.6
[52]
0.006
[84]
0.13m0
[52]
12.45 Dielectric constant (high frequency) Refractive index at LWL at 600–800 nm at 900–1200 nm at 900–1200 nm at 620 nm Density (g cm3) Energy separation between G and X c1 valleys EG (eV) Energy separation between G and L valleys EL (eV) Spin–orbit splitting in valence band, Dso or Eso (eV) Effective electron mass, me Effective hole masses (heavy)
Effective hole masses (light)
Effective hole masses (split-off band), ms or mch or mso
[138]
½110 mhh ½100 mhh ½111 mhh ½110 mlh ½100 mlh ½111 mlh
¼ 2:12 m0
[52]
¼ 1:18 m0 ¼ 2:89 m0 ¼ 0:20 m0
[52]
¼ 0:21 m0
¼ 0:19 m0 0.36m0
[52]
LWL: long-wavelength limit; NIRSR: normal incidence reflectance of synchrotron radiation.
is in part a consequence of measurements being made on samples of widely varying quality. For this book, when possible we have disregarded the spurious determination. However, measurements are too few to yield a consensus, in which case the available data are simply reported.
1.2 Gallium Nitride
The burgeoning interest in nitrides has led to substantial improvements in the crystal growth and processing technologies, thus overcoming many difficulties encountered earlier. Consequently, a number of laboratories consistently obtained highquality GaN with room-temperature background electron concentrations as low as 5 · 1016 cm3. The successful development of approaches leading to p-type GaN has led to the demonstration of excellent p–n junction LEDs in the UV, violet, blue, green, and even yellow bands of the visible spectrum with brightness suitable for outdoor displays, CW lasers, and UV detectors, including the ones for the solar blind region. Moreover, power modulation doped field effect transistors (MODFETs) also generically referred to as heterojunction field effect transistors (HFETs) have been developed. What follows reports on the state of knowledge regarding the physical properties of GaN. 1.2.1 Chemical Properties of GaN
Since Johnson et al. [139] first synthesized GaN in 1932, a large body of information has repeatedly indicated that GaN is an exceedingly stable compound exhibiting significant hardness. It is this chemical stability at elevated temperatures combined with its hardness that has made GaN an attractive material for protective coatings. Moreover, owing to its wide energy bandgap, it is also an excellent candidate for device operation at high temperatures and caustic environments. Although the hardness may have initiated the interest in GaN, it is the excellent semiconducting features that have piqued the attention of researchers. While the thermal stability of GaN allows freedom of high-temperature processing, the chemical stability of GaN presents a technological challenge. Conventional wet etching techniques used in semiconductor processing have not been as successful for GaN device fabrication. For example, Maruska and Tietjen [140] reported that GaN is insoluble in H2O, acids, or bases at room temperature, but does dissolve in hot alkali solutions at very slow rates. Pankove [141] noted that GaN reacts with NaOH forming a GaOH layer on the surface and prohibiting wet etching of GaN. To circumvent this difficulty, he developed an electrolytic etching technique for GaN. Low-quality GaN has been etched at reasonably high rates in NaOH [142,143], H2SO4 [144], and H3PO4 [145–147]. Although these etches are extremely useful for identifying defects and estimating their densities in GaN films, they are not as useful for the fabrication of devices [148]. Well-established chemical etching processes do help for the device technology development, and the status of these processes in the case of GaN can be found in Volume 2, Chapter 1. Various dry etching processes reviewed by Mohammad et al. [149] and Pearton et al. [150] are promising possibilities and are discussed in Volume 2, Chapter 1. 1.2.2 Mechanical Properties of GaN
GaN has a molecular weight of 83.7267 g mol1 in the hexagonal wurtzite structure. The lattice constant of early samples of GaN showed a dependence on growth conditions, impurity concentration, and film stoichiometry [151]. These observations were
j35
j 1 General Properties of Nitrides
36
attributed to a high concentration of interstitial and bulk extended defects. A case in point is that the lattice constants of GaN grown with higher growth rates were found to be larger. When doped heavily with Zn [152] and Mg [153], a lattice expansion occurs because at high concentrations the group II element begins to occupy the lattice sites of the much smaller nitrogen atom. At room temperature, the lattice parameters of GaN platelets [18] prepared under high pressure at high temperatures with an electron concentration of 5 · 1019 cm3 are a ¼ 3.1890 0.0003 Å and c ¼ 5.1864 0.0001 Å. The freestanding GaN with electron concentration of about 1016 cm3, originally grown on sapphire (0 0 0 1) by hydride vapor phase epitaxy (HVPE) followed by liftoff, has lattice constants of a ¼ 3.2056 0.0002 Å and c ¼ 5.1949 0.0002 Å . For GaN powder, a and c values are in the range of 3.1893–3.190 and 5.1851–5.190 Å, respectively. Experimentally observed c/a ratio for GaN is 1.627, which compares well with 1.633 for the ideal case, and the u parameter calculated using Equation 1.1 is 0.367, which is very close to the ideal value of 0.375. For more established semiconductors with the extended defect concentration from low to very low, such as Si, GaAs, and so on, the effect of doping and free electrons on the lattice parameter has been investigated rather thoroughly. In bulk GaN grown by the high-pressure technique, the lattice expansion by donors with their associated free electrons has been investigated [18]. However, large concentration of defects and strain, which could be inhomogeneous, rendered the studies of this kind less reliable in GaN layers. In spite of this, the effect of Mg doping on the lattice parameter in thin films of GaN has been investigated. Lattice parameters as large as 3.220–5.200 Å for a and c values, respectively, albeit not in all samples with similar hole concentrations, have been reported [154]. For GaN bulk crystals grown with high-pressure techniques and heavily doped (a small percentage) with Mg, the a and c lattice parameters were measured to be 3.2822–5.3602 Å [155]. Suggestions have been made that the c parameter of implanted GaN layers increases after implantation and languishes after annealing [156]. However, the a parameter could not be precisely measured because sharp off-normal diffraction peaks are needed to determine this parameter accurately. For the zinc blende polytype, the calculated lattice constant, based on the measured GaN bond distance in Wz GaN, is a ¼ 4.503 Å. The measured value for this polytype varies between 4.49 and 4.55 Å, while that in Ref. [18] is 4.511 Å, indicating that the calculated result lies within the acceptable limits [157]. A high-pressure phase transition from the Wz to the rock salt structure has been predicted and observed experimentally. The transition point is 50 GPa and the experimental lattice constant in the rock salt phase is a0 ¼ 4.22 Å. This is slightly different from the theoretical result of a0 ¼ 4.098 Å obtained from first-principles nonlocal pseudopotential calculations [158]. Tables 1.6 and 1.10 compile some of the known properties of Wz GaN. Parameters associated with electrical and optical properties of Wz GaN are tabulated in Table 1.11. The same parameters associated with the zinc blende phase of GaN are tabulated in Table 1.12. The bulk modulus of Wz GaN, which is the inverse of compressibility, is an important material parameter. Various forms of X-ray diffraction with the sample being under pressure can be used to determine the lattice parameters. Once the
1.2 Gallium Nitride
lattice parameters are determined as a function of pressure, the pressure dependence of the unit cell volume can be obtained and fitted with an equation of state (EOS), such as the Murnaghans EOS [159], and based on the assumption that the bulk modulus has a linear dependence on the pressure: 0 ! 1=B BP 1þ ; B0 0
V ¼ V0
ð1:10Þ
where B0 and V0 represent the bulk modulus and the unit volume at ambient pressure, respectively, and B0 the derivative of B0 versus pressure. X-ray diffraction leads to the determination of the isothermal bulk modulus, whereas the Brillouin scattering leads to the adiabatic one. Nevertheless, in solids other than molecular solids, there is no measurable difference between the two thermodynamic quantities [160]. The bulk modulus (B) of Wz GaN has been calculated from first principles [161] and the first-principle orthogonalized linear combination of atomic orbitals (LCAO) method [158], leading to the values of 195 and 203 GPa, respectively. Another estimate for B is 190 GPa [158]. These figures compare well with the value of 194.6 GPa estimated from the elastic stiffness coefficient [79] and a measured value for 245 GPa [6]. The bulk modulus is related to the elastic constants through B¼
ðC 11 þ C12 ÞC33 2C 213 C11 þ C12 þ 2C33 4C13
ð1:11Þ
and the range of bulk modulus values so determined is from about 173 to 245 GPa [160]. Using the room-temperature elastic constants of single-crystal GaN calculated by Polian et al. [38] yields an adiabatic bulk modulus, both Voigt and Reuss averages, of 210 GPa [91]. Earlier experimental investigations of the elastic constants of Wz GaN were carried out by Savastenko and Sheleg [162] using X-ray diffraction in powdered GaN crystals. The estimates of the Poissons ratio from the early elastic coefficients (n ¼ C13/C11 þ C12) [162] and its measured [26] values of 0.372 (for nh0 0 0 1i ¼ (Da/arelax)/(Dc/crelax)) and 0.378 (for nh0 0 0 1i ¼ (Da/a0)/(Dc/c0)), respectively, are in good agreement (to avoid confusion the R value is defined as R ¼ 2C31/C33). The experiments were performed on GaN layers on sapphire substrates because of X-ray diffraction. However, the results obtained later point to a Poissons ratio of more near 0.2 as tabulated in Table 1.6 and depend on crystalline direction. The Poissons ratio for the ZB case can be calculated from the elastic coefficients for that polytype as n or s0 ¼ (C12/C11 þ C12) leading to values of about 0.352 as tabulated in Table 1.8. The Poissons ratio varies along different crystalline directions as tabulated in Table 1.13 for AlN. It should be noted that there is still some spread in the reported values of elastic stiffness coefficients, as discussed in detail in the polarization sections of Section 2.12. More importantly, Kisielowski et al. [39] pointed out that expression ðn 1Þh0 0 0 1i ¼ ðDa=arelaz Þ=ðDc=c relax Þ;
ð1:12Þ
j37
j 1 General Properties of Nitrides
38
where Da ¼ ameas arelax and Dc ¼ cmeas crelax, should be used to calculated the Poissons ratio, n. Doing so leads to a Poissons coefficient of nGaN ¼ 0.2–0.3. Chetverikova et al. [163] measured the Youngs modulus and Poissons ratio of their GaN films. From the elastic stiffness coefficients, Youngs modulus Eh0001i is estimated to be 150 GPa [157,162]. Sherwin and Drummond [164] predicted the elastic properties of ZB GaN on grounds of values for those Wz GaN samples reported by Savastenko and Sheleg [162]. The elastic stiffness coefficients and the bulk modulus are compiled in Table 1.24. Considering the wide spread in the reported data more commonly used figures are also shown. Wagner and Bechstedt [178] calculated the elastic coefficients of Wz GaN using a pseudopotential plane wave method and pointed out the discrepancies among the results from different calculations and measurements tabulated in Table 1.24. It is argued that reliable values produce 2C13/C33 ¼ 0.50–0.56 and n ¼ 0.20–0.21 [178]. The agreement between ab initio calculations [42,178] and some measure-
Table 1.24 Experimental and calculated elastic coefficients (Cii), bulk modulus (B) and its pressure derivative (dB/dP), and Youngs modulus (E or Y0) and (in GPa) of Wz GaN and ZB GaN (in part from Ref. [160]).
Technique
C11
C12
C13
C33
C44
B0
X-ray [162] XAS EDX ADX Brillouin [38] Brillouin Brillouin [166] Brillouin [28] Brillouin [169] Brillouin [170] Ultrasonic [171] Ultrasonic [165] Single crystal X-ray Most commonly used values PWPP [42] FP-LMTO [165] Kim [168] PWPP (Wagner) ZB GaN
296
130
158
267
24.1
195 245 188 237 210 180 204 192 175 192 173 208 207
390 374 365 373 315 373 377 370
145 106 135 141 118 141 160 145
380
110
367 396 431 C11 253–264
135 144 109 515–C11 153–165
106 70 114 80.4 96 80 114 110
398 379 381 387 324 387 209 390
105 101 109 94 88 94 81.4 90
B0
E 150
4 3.2 4.3 356 329 362 281 362 161 343
105 103 100 64 104
405 392 476 414
95 91 126 60–68
202 207 201 207 200–237
363 355 461 373 3.9–4.3
The room-temperature elastic constants of single-crystal GaN have been determined by Polian et al. [38] yielding an adiabatic bulk modulus, both Voigt and Reuss, averages, of 210 GPa. The term B0 ¼ dB/dP represents the derivative of B0 versus pressure. EDX: energy dispersive X-ray; ADX: angular dispersive X-ray diffraction; XAS: X-ray absorption spectroscopy; PWPP: plane wave pseudopotential; FP-LMTO: full-potential linear muffin–tin orbital.
1.2 Gallium Nitride
ments [38,165,166] is satisfactory. However, several calculations [167,168] and measurements [28,169–171] suffer from deviations in one or more of the values of elastic constants. The results from Savastenko and Sheleg [162] show excessive deviation for all the elastic constants and, therefore, should be avoided completely. The results from surface acoustic wave measurements of Deger et al. [165] on epitaxial epilayers have been corrected for piezoelectric stiffening and, therefore, are among the most reliable. The vibrational properties of nitrides can best be described within the realm of mechanical properties. These vibrations actually serve to polarize the unit cell [172]. Phonons can be discussed under mechanical and optical properties. Here an arbitrary decision has been made to lump them with the mechanical properties of the crystal. Using GaN as the default, a succinct discussion of vibrational modes, some of which are active Raman modes, some are active in infrared (IR) measurements, and some are optically inactive called the silent modes, is provided [173]. Vibrational modes, which go to the heart of the mechanical properties, are very sensitive to crystalline defects, strain, and dopant in that the phonon mode frequencies and their frequency broadening can be used to glean very crucial information about the semiconductor. The method can also be applied to heterostructures and strained systems. Electronic Raman measurements can be performed to study processes such as electron–phonon interaction in the CW or time-resolved schemes. Time-resolved Raman measurements as applied to hot electron and phonon processes under high electric fields have important implication regarding carrier velocities. A case in point regarding GaN is treated in this context in Volume 3, Chapter 3. The wurtzite crystal structure has the C 46v symmetry and the group theory predicts the existence of the zone center optical modes A1, 2B1, E1, and 2E2. In a more simplified manner, one can consider that the stacking order of the Wz polytype is AaBb while that for the ZB variety is AaBbCc. In addition, the unit cell length of the cubic structure along [1 1 1] is equal to the width of one unit bilayer, whereas that for the hexagonal structure along [0 0 0 1] is twice that amount. Consequently, the phonon dispersion of the hexagonal structure along [0 0 0 1] (G ! A in the Brillouin zone) is approximated by folding the phonon dispersion for the ZB structure along the [1 1 1] (G ! L) direction [174], as shown in Figure 1.12. Doing so reduces the TO phonon mode at the L point of the Brillouin zone in the zinc blende structure to the E2 mode at the G point of the Brillouin zone in the hexagonal structure. This vibrational mode is denoted as EH 2 with superscript H depicting the higher frequency branch of the E2 phonon mode. As indicated in the figure there is another E2 mode at a lower frequency labeled as EL2 . This has its genesis in zone folding of the transverse acoustic (TA) mode in the zinc blende structure. It should be noted that in the hexagonal structure there is anisotropy in the macroscopic electric field induced by polar phonons. As a result, both the TO and LO modes split into the axial (or A1) and planar (or E1) modes where atomic displacement occurs along the c-axis or perpendicular to the c-axis, respectively. This splitting is not shown in Figure 1.12 as it is very small, just a few meV, near zone center; phonon dispersion curves for GaN including the splitting of the A1 and E1 modes can be found in Volume 3, Figure 3.84.
j39
j 1 General Properties of Nitrides
40
Zinc blende
Wurtzitic A1
LO
H
B1
H
TO
E1 A
[0001]
Phonon frequency
E2
Γ
L
B1
LA
L
E2
TA
Γ
A
[111]
L
Wave vector Figure 1.12 Schematic depiction of the phonon dispersion curves for ZB and Wz structures. Also shown are the G and A points of the zone in relation to the real space hexagonal structure. Phonon branches along the [1 1 1] direction in the ZB structure are folded to approximate those
of the wurtzite structure along the [0 0 0 1] direction, because the unit cell length of the cubic structure along the [1 1 1] direction is equal to the width of one unit bilayer, while that for the hexagonal structure along the [0 0 0 1] directions is twice that amount. Patterned after Ref. [174].
As discussed below, in the context of hexagonal structures, group theory predicts eight sets of phonon normal modes at the G point, namely 2A1 þ 2E1 þ 2B1 þ 2E2. Among them, one set of A1 and E1 modes are acoustic, while the remaining six modes, namely A1 þ E1 þ 2B1 þ 2E2, are optical modes. As shown in Figure 1.12, one A1 and one B1 mode (BH 1 ) derive from a singly degenerate LO phonon branch of the zinc blende system by zone folding, whereas one E1 and one E2 mode (EH 2 ) derive from a doubly degenerate TO mode in the cubic system. The first-order phonon Raman scattering is due to phonons near the G point zone center, that is, with wave vector k 0, because of the momentum conservation rule in the light scattering process. Raman measurements typically are employed to probe the vibrational properties of semiconductors. When performed along the direction perpendicular to the c-axis or the (0 0 0 1) plane, the nomenclature used to describe this configuration is depicted as ZðXY; XYÞZ. Here, following Portos notation [175] A(B, C)D is used to describe the Raman geometry and polarization, where A and D
1.2 Gallium Nitride
represent the wave vector direction of the incoming and scattered light, respectively, whereas B and C represent the polarization of the incoming and scattered light. In Raman scattering, all the above-mentioned modes, with the exception of B1 modes, are optically active. Because of their polar nature, the A1 and E1 modes split into longitudinal optical (A1-LO and E1-LO) meaning beating along the c-axis, and transverse optical (A1-TO and E1-TO), meaning beating along the basal plane. To reiterate, the A1 and B1 modes give atomic displacements along the c-axis, while the others, E1 and E2, give atomic displacements perpendicular to the c-axis, meaning on the basal plane. Here, the A1 and E1 modes are both Raman and IR active whereas the two E2 modes are only Raman active and the two B1 modes are neither Raman nor IR active, meaning silent modes. In the ZðX Y; X YÞZ configuration, only the E12 (or EL2 or E2 low), E22 (or EH 2 or E2 high), and A1(LO) modes should be observable. In particular, in ZðX ; XÞZ and ZðY; YÞZ geometries, all three modes are observable, while in ZðX; YÞZ or ZðY; X ÞZ geometries only E2 modes are detected [175]. The details of the mode–Raman configuration relationship are tabulated in Table 1.25. Shown in Figure 1.13 are the modes in the Raman backscattered geometries in relation to hexagonal crystalline orientation that can be used to sense the various phonon modes indicated. The acoustic modes, which are simple translational modes, and the optical modes for wurtzite symmetry are shown in Figure 1.14. The calculated phonon dispersion curves [57] for GaN are shown in Figure 1.15. There is another way to describe the number of vibrational modes in zinc blende and wurtzitic structures, which is again based on symmetry arguments. In the wurtzite case [66], the number of atoms per unit cell s ¼ 4, and there are total of 12 modes, the details of which are tabulated in Table 1.26. This table also holds for the zinc blende polytypes with s ¼ 2. This implies a total of six modes in zinc blende as opposed to 12 in wurtzite, three of which are acoustical (1 LA and 2 TA) and the other three are optical (1 LO and 2 TO) branches. These phonon modes for a wurtzite symmetry, specifically the values for wurtzite GaN, are listed in Table 1.27 obtained from Refs [56,157,176,177] along with those obtained from first-principles pseudopotential calculations [161,178]. Also listed are TO and LO optical phonon wave numbers of ZB GaN [25,179].
Table 1.25 Raman measurement configuration needed to observe
the phonon modes in hexagonal nitrides. Mode
Configuration
A1(TO), E2 A1(TO) E1(TO) E1(TO), E1(LO) E2 E2 A1(LO), E2
X ðY; YÞX X ðZ; ZÞX X ðZ; YÞX X ðY; ZÞY X ðY; YÞZ ZðY; X ÞZ ZðY; YÞZ
j41
j 1 General Properties of Nitrides
42
Z(X,X)Z + Z(X,Y)Z L E2
H E2
A1(LO)
X(Z,Z)X
A1 (TO)
X(Z,Y)X
A1 (TO)
Quasi-E1 (LO) H
E2
Y Y
Z
X GaN Substrate
GaN Z
Substrate
c-axis
X
c-axis Figure 1.13 Schematic representation of two Raman configurations with incident and scattered light directions in the backscattering geometry for ZðX ; X ÞZ þ ZðX; YÞZ configuration to sense EL2 , EH 2 , and A1(LO) modes, X ðZ; ZÞX configuration to sense A1(TO) and quasi-E1(LO) modes, and X ðZ; YÞX configuration to sense A1(TO) and EH 2 modes.
Owing to the presence of elastic strain, thin epilayers have different phonon energies compared to the bulk samples. In the general case, strain can give rise to a shift and splitting of phonon modes. However, for uniaxial strain along the c-axis or biaxial strain in the c-plane, the crystal retains its hexagonal symmetry resulting in only a shift of the phonon frequencies. The strain effects on GaN optical phonon energies have been studied experimentally [181] and theoretically [178]. Within a perturbative approach, the change in the frequency of a given phonon l under symmetry-conserving stress can be expressed in terms of the two strain components, exx and ezz, representing perpendicular and parallel to the z-axis, respectively, as DOl ¼ 2 alexx þ blezz, where al and bl are the corresponding deformation potential constants. The derivation of strain values from the Raman measurements of phonon frequencies is straightforward, once the phonon deformation potentials are known. Davydov et al. [181] combined high-resolution X-ray diffraction (HRXRD) measurements and Raman spectroscopy results to determine the phonon deformation potentials in GaN epitaxial layers grown on 6H-SiC. The strain components were obtained by comparing the lattice constants of the epitaxial layers with those of the strain-free GaN (a0 ¼ 3.1880 Å, c0 ¼ 5.18 561 Å). The Raman spectrum of a strain-free 300 mm thick GaN layer grown on sapphire was used as a reference. Also, using their relation to the hydrostatic pressure coefficients [6] through the bulk elastic coefficients, the sets of phonon deformation potentials were derived for most of zone center optical phonons. As seen in Table 1.27, except for the A1-TO mode, the phonon energies and the deformation potentials reported by Davydov et al. [181] agree well with the pseudopotential plane wave calculations reported by Wagner and Bechstedt [178]. Wagner and Bechstedt argue that the published conversion factors
1.2 Gallium Nitride
A 1 mode
Z
Z
Z
B 1 (1) mode
N
N
N
Ga
Ga
Ga
N
N
N
Ga
Ga
Ga
X
X Z
B 1 (2) mode
Z
E 1 mode
X Z
E 2 (1) mode
N
E 2 (2) mode N
N Ga
Ga
Ga
N
N Ga
N
Ga
Ga
X
(a)
X
X
A mode 1 Z
Z
B1H
Z
L
B1
N
N
N
Ga
Ga
Ga
N
N
N
Ga
Ga
L
Z
Ga
X
X
Z
E2
H
E2
N
N
Ga
Ga
N
Ga
(b)
X
N
X
Figure 1.14 Atomic vibrations in Wz GaN. The larger atom represents Ga while the smaller one is for N. X ¼ (1 0 0), Y ¼ (0 1 0), and Z ¼ (0 0 1) represent the optical polarization directions: (a) for general wave vector and (b) for zone center phonons.
Ga
X
j43
j 1 General Properties of Nitrides
44
E1(LO)
100
E1 (LO)
GaN
700 B 1
B1 A1(LO)
600 E2
E2 E (TO) 1
80
A1(TO)
A1(TO)
500
60
400 B1
B1
40
300 200
Energy (meV)
Frequency (cm–1)
800
E2
E2
20
100 0
0 Γ
K
M
Γ
A
H
L
A
DOS
Reduced wave vector coordinate Figure 1.15 Calculated phonon dispersion curves and phonon density of states for hexagonal bulk GaN. The solid and dashed lines correspond to the L1(or T1) and L2 (or T2) irreducible representations (following Ref. [180]), respectively. Note how close the E1(LO) and A1(LO) modes are, making high-quality samples with sharp modes imperative for their experimental delineation [169].
Table 1.26 Acoustic and optical phonon modes in a crystal with wurtzite symmetry such as GaN, AlN, and InN, where s represents the number of atoms in the basis.
Mode type
Number of modes
LA TA Total acoustic modes LO TO All optical modes All modes
1 2 3 s1 2s 2 3s 3 3s
The s parameter for wurtzite symmetry is 4. This table is also applicable to the zinc blende case but with s ¼ 2.
Wz [157,177]
556–559 533–534 741–741 710–735 143–146 560–579
Phonon mode
E1-TO A1-TO E1-LO A1-LO EL2 EH 2 BL1 BH 1
(a)
558.5 532.5 745.0 737.0
Wz template [177] 558.8 531.8 741 734 144 567.6
[169]
Wz relaxed
568 540 757 748 142 576 337 713
Wz unstrained (calculated) [178]
bk 680 50 1290 80
80 35 920 60
ak 820 25 630 40
115 25 850 25
Raman [181]
717 640 775 664 75 742 334 661
ak
Calculated [178]
Deformation potentials – Wz
by Akasaki and Amano [157] (Wz) and Huang et al. [177] (Wz template). Also shown are the calculated values. (b) Measured phonon wave numbers (in units of cm1) for wurtzitic GaN. (c) Zinc blende phase phonon wave numbers for zinc blende GaN [179] (theory [59]).
Table 1.27 (a) Zone center optical phonon wave numbers (in units of cm 1) of GaN obtained from Raman scattering at 300 K compiled
591 695 703 695 4 715 275 941
bk
1.2 Gallium Nitride
j45
GaN on sapphire, about 50–70 mm thick, at 300 K.
LO(L) (cm1) TO(L) (cm1) LA(L) (cm1) TA(L) (cm1) TO (cm1) LO (cm1)
LO(G) (cm1) TO(G) (cm1) LO(X) (cm1) TO(X) (cm1) LA(X) (cm1) TA(X) (cm1) 748 [59] 562 [59] 639 [59] 558 [59] 286 [59] 207 [59]
Mode
Mode
(c)
E1-TO A1-TO E1-LO A1-LO EL2 EH 2 BL1 BH 1
Phonon mode
(b)
Table 1.27 (Continued)
675 [59] 554 [59] 296 [59] 144 [59] 558 [179] 730 [179]
558.8 531.8 741 741 144 567.6
560 [58] 750 [58]
Wz unstrained (measured) (collected in Refs [174,178], but based on Refs [6,169])
46
j 1 General Properties of Nitrides
1.2 Gallium Nitride
between the luminescence or Raman shifts and the corresponding biaxial stress are seldom directly measured data. They are either obtained using elastic constants or are constructed from deformation potentials, which have been obtained by means of additional hydrostatic pressure coefficients. Owing to these varying procedures and different sets of parameters used to extract the conversion coefficients from the raw experimental data, discrepancies in the experimental reports of deformation potentials are present. 1.2.3 Thermal Properties of GaN
The lattice parameter of semiconductors depends on temperature and is quantified by thermal expansion coefficient (TEC), which is defined as Da/a or aa and Dc/c or ac, for in-plane and out-of-plane configurations, respectively. It depends on stoichiometry, extended defects, and free-carrier concentration. As in the case of the lattice parameter, a scatter exists in TEC particularly for nitrides as they are grown on foreign substrates with different thermal and mechanical properties. Measurements made over the temperature range of 300–900 K indicate the mean coefficient of thermal expansion of GaN in the c-plane to be Da/a ¼ aa ¼ 5.59 · 106 K1. Similarly, measurements over the temperature ranges of 300–700 and 700–900 K, respectively, indicate the mean coefficient of thermal expansion in the c-direction to be Dc/c ¼ ac ¼ 3.17 · 106 and 7.75 · 106 K1, respectively [140]. Sheleg and Savastenko [49] reported a TEC near 600 K for perpendicular and parallel to the c-axis of 4.52 0.5 · 106 and 5.25 0.05 · 106 K1, respectively. Leszczynski and Walker [182] reported aa values of 3.1 and 6.2 106 K1 for the temperature ranges of 300–350 and 700–750 K, respectively. The ac values in the same temperature ranges, in order, were 2.8 and 6.1 106 K1. In a similar vein, GaN and other allied group III nitride semiconductors are grown at high temperatures and also subjected to increased junction temperatures during operation of devices such as amplifiers and light emitting devices. As such, the structures are subjected to thermal variations as well. In this context, it is imperative to have knowledge of the thermal expansion coefficients, which are termed as TEC. Assuming that these figure remain the same with temperature, the linear expansion coefficients for the a and c parameters are tabulated in Tables 1.10 and 1.28 for heteroepitaxial GaN. However, it is instructive to know the temperature dependence of these parameters, which is shown in Figure 1.16. Being grown on various substrates with different thermal expansion coefficients leads to different dependencies of the lattice parameter on temperature. Temperature dependence of GaN lattice parameter has been measured for a bulk crystal (grown at high pressure) with a high free-electron concentration (5 · 1019 cm3), a slightly strained homoepitaxial layer with a low free-electron concentration (about 1017 cm3), and a heteroepitaxial layer (also with a small electron concentration) on sapphire [88]. The results of such study are tabulated in Table 1.28. It can be seen that the bulk sample with a high free-electron concentration exhibits a thermal expansion that is about 3% higher as compared to the homoepitaxial layer.
j47
j 1 General Properties of Nitrides
48
7
Linear expansion coefficent, a (10–6 K–1)
GaN 6
a
5
a
4
3
2 200
100
300
400
500
600
700
800
Temperature (K) Figure 1.16 Wz GaN coefficient of linear thermal expansion versus temperature for basal plane (a||), a parameter, and out of the basal plane (a?), c parameter, directions [49].
As for the case of the heteroepitaxial layer on sapphire, the thermal expansion of the substrate affects the dependence of the lattice parameter on temperature. Various spectroscopic techniques, such as Auger electron spectroscopy, X-ray photoemission spectroscopy (XPS), and electron energy loss spectroscopy (EELS) have been very useful for the study of surface chemistry of GaN. Building on earlier investigations of the thermal stability of GaN by Johnson et al. [139] and employing Table 1.28 Lattice parameters for GaN samples at various temperatures (lattice parameters c were measured with accuracy of 0.0002 Å, lattice parameter a with accuracy of 0.0005 Å) [88].
T (K)
20 77 295 500 770
GaN bulk, n ¼ 5 · 1019 cm3
Homoepitaxial GaN on conductive GaN substrate
GaN on sapphire
c (Å)
a (Å)
c (Å)
a (Å)
c (Å)
a (Å)
5.1836 5.1838 5.1860 5.1885 5.1962
3.1867 3.1868 3.1881 3.1903 3.1945
5.1822 5.1824 5.1844 5.1870 5.1944
3.1867 3.1868 3.1881 3.1903 3.1945
5.1846 5.1865 5.1888 5.1952
3.1842 3.1859 3.1886 3.1941
1.2 Gallium Nitride
the aforementioned techniques, the thermal stability and dissociation of GaN have been examined further. As indicated earlier, the materials characteristics depend, to a large extent, on defects and impurities, which in turn depend somewhat on growth conditions. Because of this the materials, obtained from various sources, studied in various laboratories exhibit different characteristics. This led to inconsistent results from different laboratories. While some experimental studies on GaN stability conducted at high temperatures suggested that significant weight losses occur at temperatures as low as 750 C, others contradicted this proposition and suggested that no significant weight loss should occur even at a temperature of 1000 C. Sime and Margrave [183] followed the investigation by Johnson et al. [139] by studying the evaporation of GaN and Ga metal in the temperature range 900–1150 C under atmospheric pressure in N2, NH3, and H2 environments with an emphasis on the formation and decomposition equilibrium. The heat of evaporation was determined and the existence of (GaN)x polymers in the gas phase was suggested. Morimoto [184] and Furtado and Jacob [185] observed that GaN is less stable in an HCl or H2 atmosphere than in N2. Some controversy exists regarding the process steps that dictate the decomposition of GaN. Using mass spectroscopy, Gordienko et al. [186] noted that (GaN)2 dimers are the primary components of decomposition. Others [187,188] found only N2 þ and Ga þ to be the primary components in the vapor over GaN. On the basis of measurements of the apparent vapor pressure, Munir and Searcy [189] calculated the heat of sublimation of GaN to be 72.4 0.5 kcal mol 1. Thurmond and Logan [190] determined the equilibrium N2 pressure of GaN as a function of temperature by measuring the partial pressure ratios existing in a (H2CNH3) gas mixture streaming over Ga and GaN. Thermal stability of GaN was taken up later by Karpinski et al. [191] with a detailed investigation of the problem at high temperatures and under pressure up to 60 kbar by employing a tungsten carbide anvil cell activated by a gas pressure technique. The bond strength in gallium nitride is high with bonding energy of 9.12 eV/molecule [192], particularly as compared to the more conventional semiconductors such as GaAs, which has a bonding energy of 6.5 eV/atom pair. As a result, the free energy of GaN is very low in relation to the reference states of the free N and Ga atoms. However, the N2 molecule is also strongly bonded with 4.9 eV/atom. Therefore, the free energies of the constituents of GaN (Ga and N2) at their normal states are close to that of the GaN crystal as illustrated in Figure 1.17, where the free energy of GaN (1 mol) and the free energy of the sum of its constituents (Ga þ 1/2N2) are shown as a function of temperature and N2 pressure. As the temperature increases, the Gibbs free energy, G(T), of the constituents decreases faster than G (T) of the GaN crystal. More importantly, GaN becomes thermodynamically unstable at high temperatures. The crossing of G(T) curves determines the equilibrium temperature where GaN coexists with its constituents at a given N2 pressure. The application of pressure increases the free energy of the constituents more than G(T) of the GaN crystal, which causes the equilibrium point to shift to higher temperatures, increasing the range of GaN stability. The data on phase diagrams of GaN are limited and contradictory by reason of high melting temperatures (Tm) and high nitrogen dissociation pressures (Pdis N2 ). Dissociation
j49
j 1 General Properties of Nitrides
50
–600 Ga(s) + (1/2)N2(g)
p = 1 bar
Ga(l)+(1/2)N2(g)
p = 1 kbar
G(kJ mol–1)
p = 10 kbar –800 p = 20 kbar GaN(s)
–1000
–1200 0
400
800
1200
1600
2000
T(K) Figure 1.17 Gibbs free energy of GaN and its constituents as a function of temperature and pressure [192].
pressure of MN, where M stands for Al, Ga, and In, and N for nitrogen, is defined as the nitrogen pressure at the thermal equilibrium of the reaction [193]: MN(s) ¼ M(l) þ 1/2N2(g), where s, l, and g stand for solid, liquid, and gas states, respectively. Reported values for Pdis N2 for GaN [193] show large discrepancies [191,194]. Specifically, in the high-pressure range, the partial pressure, p, versus the inverse temperature, 1/T, curve of Karpinski et al. deviates markedly from the linear dependence proposed by Thurmond and Logan as shown in Figure 1.18. Despite the discrepancies, there is a good agreement in the Gibbs free energy with DG0 32.43T 3.77 · 104 700 cal mol1 for GaN synthesis between the two references. The value of enthalpy DH0 (37.7 kcal mol1) is in good agreement as well with that estimated by Madar et al. [194]. The stars in Figure 1.18 indicate the melting point of AlN at T M AlN ¼ 3487 K, GaN M at T M GaN ¼ 2791 K, and InN at T InN ¼ 2146 K. The GaN and InN melting points so indicated may underestimate the real values, as perhaps a sufficient overpressure was not maintained. Line fits correspond to 8.3 109 exp( 5.41 eV/kT), 1.5 · 1014 exp(3.28 eV/kT), and 7.9 · 1017 exp(2.78 eV/kT) bar for AlN, GaN, and InN, respectively. The data over the larger temperature range are those compiled by Ambacher [196]. The results of Madar et al. [194], Thurmond and Logan [190], and Karpinski et al. [191] are also shown in a limited temperature range. For GaN (see Figure 1.18), the nitrogen dissociation pressure equals 1 atm at approximately 850 C and 10 atm at 930 C. At 1250 C, GaN decomposed even under pressure of 10 000 bar of N2. The turning over of the partial pressure for GaN and InN at temperatures approaching the melting point may need to be reexamined. What is clear, however, is that GaN and particularly InN have very high partial pressures that make it imperative to maintain high fluxes of N during growth. It should, therefore, come as no surprise that the incorporation of nitrogen is not a trivial problem at high temperatures. For the pressures below equilibrium at a given temperature, the thermal dissociation occurs at a slow and apparently constant rate suggesting a diffusion-controlled
1.2 Gallium Nitride
Temperature (oC) 105
4700
1400
440
700
104 u Th
103
rm
InN
d an Lo
2
n ga
101
Ma
dis
d
PN (atm)
on
102
dar et a
AlN
l. Ka
100
n rpi et ski
GaN
al.
10–1 10–2 10–3
2
4
6
10 8 104 /T ( K–1)
Figure 1.18 Equilibrium N2 pressure over the MN(s) þ M(l) systems corresponding to GaN (M ¼ Al, Ga, In) reported and/or compiled by Slack and McNelly[195], Madar et al. [194], Thurmond and Logan [190], Karpinski et al. [191], and Ambacher [196]. The melting points of the three binaries are indicated by stars. The desorption activation energies, EMN, determined by straight line fits to the data points are 3.5 eV
12
14
(336 kJ mol1), 3.9 eV (379 kJ mol1), and 4.3 eV (414 kJ mol1) for InN, GaN, and AlN, respectively. Caution should be exercised as there is significant deviation from the activation line for GaN and InN. This may simply be a matter of not being able to maintain sufficient pressure on GaN and InN at very high temperatures to reach the real melting point.
process of dissociation. Expanded equilibrium vapor pressure data inclusive of GaAs and InP in addition to the three nitride binaries reported by Matsuoka [197] are shown in Figure 1.19. Melting points and other thermodynamic characteristics of III-N compounds are given in Tables 1.10, 1.16, and 1.20 as compiled by Popovici and Morkoc [198] as well as those collected from various sources as indicated in the pertinent tables. Investigations utilizing epitaxial thin films of GaN, as well as AlN and InN, have been conducted by Ambacher et al. [199], who heated the samples in vacuum and recorded the partial pressure of relevant gases with a quadrupole mass spectrometer. Desorption spectra were then analyzed [200] to find the binding energies of various desorbed species as well as the thermal stability of the sample for a given thermal treatment. As expected, the nitrogen partial pressure increases exponentially above TE ¼ 850 C for GaN underscoring the point that the decomposition temperature in vacuum is much lower than the melting point shown in Figure 1.18. The rate of
j51
j 1 General Properties of Nitrides
52
Temperature ( oC)
Equilibrium vapor pressure (atm)
2000
1000
500
100 InN 10–2
10–4
GaN
AlN GaAs
10–6
InP 10–8 0.0
0.5
1.0
1.5
2.0
1000/T (K–1) Figure 1.19 Equilibrium vapor pressure of N2 over AlN, GaN and InN, the sum of As2 and As4 over GaAs, and the sum of P2 and P4 over InP [197].
nitrogen evolution F(N) was set equal to the rate of decomposition, and the slope of ln [F(N)] versus 1/T gives the effective activation energy of the decomposition in vacuum as compared to those shown in Figure 1.18. The decomposition rate equals the desorption of one monolayer every second (FN ¼ 1.5 · 1015 cm2 s1) at 970 C, and the activation energy of the thermally induced decomposition is determined to be EMN ¼ 3.9 eV (379 kJ mol1) for GaN. Despite some disagreement, as mentioned above, investigations of the equilibrium nitrogen overpressure versus temperature, PN2 –T, for GaN [92,190,201], including a very complete and consistent set of data obtained by Karpinski et al. [191,202] have set the stage for bulk template growth as well as setting benchmarks for growth of GaN by nonequilibrium methods. Those authors employed direct synthesis and decomposition experiments and used the gas pressure technique (for pressures up to 20 kbar) and the high-pressure anvil method beyond the reach of gas pressure technique (up to 70 kbar). The results of these experiments are shown in Figure 1.20 [203,204]. The message in the form of N2 partial pressure is that one must stay below the decomposition curve. This means that the selection of GaN synthesis temperature directly depends on the pressure that the vessel can provide. For example, if a pressure of 20 kbar is all that is available, then the temperature should be kept below about 1660 C. For a review of the stability of GaN as well as the growth GaN templates, the reader is referred to Ref. [192]. As alluded to earlier, nitride semiconductors in general and GaN in particular are considered for high-power/high-temperature electronic and optoelectronic devices
1.2 Gallium Nitride
Temperature (K) 3
10
2 × 103 1.5 × 103 1.25 × 10
5
1×10 3
Pressure limit of 20 kbar
Pressure (bar)
104
103 Ga+1/2N 2
GaN
102
101
100
4
6
8
10
12
1/T (10 –4 K –1) Figure 1.20 N2 partial pressure as a function of temperature GaN. Ref. [192], originally in Refs [203,204].
where thermal dissipation is a key issue. Device applications assure that the thermal conductivity (k) is an extremely important material property. Thermal conductivity is a kinetic property determined by contributions from the vibrational, rotational, and electronic degrees of freedom, and as such it is related to the mechanical properties of the material. However, for convenience, this property is generally categorized under the thermal properties of nitrides in this book. The electronic thermal conductivity contribution is negligible for carrier concentrations 1019 cm3. The heat transport is predominantly determined by phonon–phonon Umklapp scattering, and phonon scattering by point and extended defects such as vacancies (inclusive of the lattice distortions caused by them), impurities, and isotope fluctuations (mass fluctuation) as elaborated on by Slack et al. [91]. For pure crystals, phonon–phonon scattering, which is ideally proportional to T1 above the Debye temperature, is the limiting process. The lattice contribution (phonon scattering) to the thermal conductivity, k, in a pure solid is obtained from the kinetic theory as [205] 1 k lattice ðTÞ ¼ vs C lattice ðTÞLðTÞ; 3
ð1:13Þ
where T is the temperature, vs is the velocity of sound (nearly independent of temperature), Clattice(T) is the lattice specific heat, and L(T) is the phonon mean free length. In nearly all materials, the thermal conductivity, k(T), first increases with temperature, reaches a maximum (kmax) at some characteristic temperature Tch, and
j53
j 1 General Properties of Nitrides
54
then decreases. At low temperatures, L is relatively long and is dominated by extrinsic effects such as defects and/or finite crystal size and Clattice(T) (T/yD)3, where yD is the Debye temperature. As the temperature increases, Clattice(T) begins to saturate and the intrinsic temperature-dependent Umklapp processes become dominant, leading to a decrease in L. The other contribution, the electronic contribution, to the thermal conductivity is negligible for carrier concentrations 1019 cm3. It can be expressed as [206] k electr ðTÞ ¼
p2 nk2B Ttelectr ; 3mc
ð1:14Þ
where n is the carrier density, kB is the Boltzmann constant, telectr is the scattering time of the electrons, and mc is the conduction band effective mass. The first measurements of k of GaN were by Sichel and Pankove [207] on bulk GaN (400 mm of material grown by HVPE) as a function of temperature (25–360 K): k ffi 1.3 W cm1 K1 (along the c-axis) at 300 K. This room-temperature value measured is a little smaller than the value of 1.7 W cm1 K1 predicted in 1973 [77] and much smaller than the k 4.10 W cm1 K1 calculated by Witek [208]. Using the elastic constants reported by Polian et al. [38], Slack et al. [91] calculated a Debye temperature of 650 K, which led to a more recent thermal conductivity at 300 K for GaN of k ¼ 2.27 W cm1 K1, assuming that there is no isotope scattering in GaN. This is very close to the measured values in high-quality freestanding GaN samples, the electrical properties of which are discussed in Volume 2, Chapter 3 and optical properties of which are discussed in Volume 2, Chapter 5. Using a steady-state four-probe method and a high-quality freestanding GaN template, Slack et al. [91] measured a value for k of 2.3 W cm1 K1 at room temperature, which increased to over 10 W cm1 K1 at 77 K. The method holds for four-probe thermal measurement, where the term four probe is analogous to the four-probe electrical measurement method. Namely, a heater is attached on the end of the sample – sandwiched in a copper clamp so that the heat flows through the entire width of the sample, not just the surface –supplying a heat current Q (analogous to an electrical current I). Two thermocouple junctions are attached along the length of the specimen by two little copper clamps separated by a distance L, the schematic representation of which is shown in Figure 1.21. The heat current Q creates a temperature gradient of DT across the wafer. The k value is calculated using k ¼ (P/DT)(L/A), where P is the power (¼voltage · current) supplied to the heater and A represents the cross-sectional area of the sample. Although the technique sounds simple, its accuracy depends very critically on making sure that the heat conduction is through the specimen and along the direction in which the temperature gradient is measured. To make certain that heat is transferred in the said direction only, the radiation losses must be minimized as well as making sure that the electrical wires used do not remove heat. To this end, the sample is placed in a turbo pumped vacuum to eliminate conduction and convection through the surrounding medium. Heat losses via conduction through the wires are minimized using long (10 cm), thin (<100 mm) wires of low thermal conductivity, typically chromel/constantan thermocouple and heater wires. Radiation losses are
1.2 Gallium Nitride
j55
Variable temperature copper finger Chromel Thermocouple wires
Thermocouple 2
Constantan Chromel
Δ ΔT
Thermocouple 1
low t f a He Copper Heater resistor
le mp Sa
r te de un
st
Thermocouple for absolute temperature measurement
Figure 1.21 Schematic representation of the four-probe thermoelectric measurement setup used to measure the thermal conductivity of freestanding GaN [91].
minimized by surrounding the sample with a heat shield anchored thermally to the cold tip of the cryostat. By carefully designing this shield, Slack et al. [91] were able to reduce the total heat loss to the order of 1–2 mW K1 at room temperature. Because the radiation losses follow T3 dependence, they die off rather quickly below room temperature. For a sample with a thermal conductivity of 1 W cm1 K1, cross section of 1 · 3 mm2, and thermocouple probe separation of 5 mm, the thermal conductance is about 60 mW K1, so the heat losses are less than 5%. Fortunately, for wide bandgap semiconductors such as GaN the thermal conductivities are high enough so that the heat conduction is mainly through the sample, reducing the measurement error. Just as a reference point, for samples of lower thermal conductance (either lower conductivity or thinner), the heat losses can become important near room temperature. Thus, the samples for lower thermal conductivity materials (e.g., glasses or thermoelectric alloys) usually need to be short with large cross-sectional areas. The temperature-dependent thermal conductivity so measured for freestanding GaN is shown in Figure 1.22. From that temperature dependence and assuming the heat dissipation is through acoustic phonons, a Debye temperature of yD 550 K was deduced, which compares with 650 K reported by Slack et al. [91]. As can be seen in Figure 1.22, the measured thermal conductivity of GaN in the temperature range of 80–300 K has a temperature power dependence of 1.22. This slope is typical of pure adamantine crystals below the Debye temperature indicating acoustic phonon transport where the phonon–phonon scattering is a combination of
j 1 General Properties of Nitrides
56
Thermal conductivity (W cm–1 K –1)
10
2
T –1.22
101 500 μm phonon mean free path 100
Sichel and Pankove
10
–1
1
10
100
1000
Temperature (K) Figure 1.22 The thermal conductivity of 200 mm thick freestanding GaN sample (Samsung) as a function of temperature. The dashed line indicates calculation using the boundary scattering limit for a phonon mean free path of 500 mm. Also shown is the T 1.22dependence between about 80 and 300 K, and earlier results of Sichel and Pankove [207] measured using a 400 mm HVPE sample. Courtesy of Slack and Morelli [91].
acoustic–acoustic and acoustic–optic interactions. This temperature dependence strongly suggests that the thermal conductivity depends mainly on intrinsic phonon–phonon scattering and not on phonon–impurity scattering. Keep in mind that the net electron concentration in the measured film is about 1016 cm3 and the hole concentration is in the 1015 cm3 range. In addition, the dislocation density is low, about 106 cm2. It should be pointed out that the thermal conductivity degrades with increased dislocation density, particularly above 107 cm2, with a slope of 0.4 W cm1 K1 per decade dropping down to slightly above 1 W cm1 K1 for a dislocation density of mid-109 cm2. The dashed curve at low temperatures in Figure 1.22 has been calculated for boundary scattering assuming a mean free path of 500 mm [91]. The mean free path is comparable with the average sample diameter, which indicates that impurity scattering in this region is not dominant either. The thermal conductivity of GaN is a tensor quantity and has two principal values k? and k|| perpendicular and parallel to the c-axis, respectively, that is, in-plane and out-of-plane values. The anisotropy in the sound velocity, which relates to the phonon propagation velocity, has been calculated by Polian et al. [38] in the form of n? ¼ 5.56 · 105 cm s1 and n|| ¼ 5.51 · 105 cm s1 for in-plane and out-of-plane directions, respectively. These values are smaller than the measured values, reported by Deger et al. [165], of n? ¼ 8.02 · 105 cm s1 and n|| ¼ 7.79 · 105 cm s1 for in-plane
1.2 Gallium Nitride
and out-of-plane directions, respectively. The in-plane and out-of-plane sound velocities reported in Ref [36] are tabulated in Tables 1.7 and 1.9 for wurtzitic and zinc blende phases of GaN. Because the difference is negligible and the anharmonicity producing the phonon–phonon scattering is not discernible, one can conclude that in-plane thermal conductivity measurements are a good representative of k in GaN [91]. A newer method, named the scanning thermal microscopy (SThM) [209], has been developed to measure thermal conductivity and is purported to provide nondestructive, absolute measurements with a high spatial/depth resolution of about 2–3 mm. Thermal imaging is achieved using a resistive thermal element incorporated at the end of a cantilever/AFM-type feedback as shown in Figure 1.23. The resistive tip forms one element of a Wheatstone bridge as shown in Figure 1.24. The spatial/depth resolution is estimated to be 2–3 mm for GaN and AlN. Upon contact with the sample, the tip tends to cooldown due to heat conduction into the sample, which is related to its thermal conductivity, k. The bridge circuit applies a compensating voltage (Uout) to maintain its target operating temperature. The feedback signal for constant resistance is a measure of the thermal conductivity of the material with which the tip is in contact, specifically V 2out is proportional to k because power dissipation is the mechanism here. Measurements of the absolute values of k are based on a calibration procedure. This simply comprises calibrating the feedback signal, V 2out , for a constant thermal element resistance against that for samples with known conductivities such as GaSb, GaAs, InP, Si, and Al metal, as shown in Figure 1.25. The influence of the surface roughness on the effective thermal conductivity is of concern. For a perfectly flat surface,
Figure 1.23 An artists view of the scanning thermal microscope. Patterned after D.I. Florescu and F.H. Pollak. (Please find a color version of this figure on the color tables.)
j57
j 1 General Properties of Nitrides
58
2 Vout ~ κ
R1
Tip
R2
Vout R probe
Sample
R control
(a)
(b)
Figure 1.24 (a) Wheatstone bridge arrangement in which the tip temperature is kept constant before and after contact with the material whose thermal conductivity is being measured. The feedback signal Uout is related to thermal
conductivity, k. A calibration against known samples such as Si, GaAs, GaP, and so on, leads to absolute values of k. (b) Schematic diagram of heat dissipation into the sample from the tip. Courtesy of D.I. Florescu and F.H. Pollak.
the contact between the probe tip (radius of curvature 1 fm) and the sample surface is very small. For rough surfaces, however, the tip could impinge on a valley- or hillocklike feature with the valley/hillock leading to increased/decreased thermal signal accompanied by a corresponding change in the measured effective thermal conductivity. 0.68 Al
0.66
V 2out
0.64 0.62 0.60 Si 0.58 InP 0.56 GaAs 0.54 0.0
GaSb 0.5
1.0
1.5 2.0 – 1 –1 k (W cm K )
2.5
Figure 1.25 The feedback signal, V 2out, which is a measure of the thermal conductivity of the material under test, for a constant thermal element resistance for samples with known conductivities such as GaSb, GaAs, InP, Si, and Al metal. Courtesy of D.I. Florescu and F.H. Pollak.
1.2 Gallium Nitride
The SThM method has been applied to the measurement of the room-temperature thermal conductivity on both fully and partially coalesced epitaxial lateral overgrown GaN/sapphire (0 0 0 1) samples [209]. As expected, a correlation between low threading dislocation density and high thermal conductivity values was established. The reduction in the thermal conductivity with increased dislocation density is expected as threading dislocations degrade the sound velocity and increase the phonon scattering in the material. In fact, due to the high defect concentrations in early films, the thermal conductivity value measured was 1.3 W cm1 K1 [207]. Using this method, the highest GaN k values, 2.0–2.1 W cm1 K1, were found in the regions of the samples that were laterally grown and thus contained the lowest density of threading dislocations. This compares with a value of 2.3 W cm1 K1 in a freestanding sample measured by the steady-state four-probe method discussed earlier. Even then, it falls short of the predictions by Witek [208]. An explanation for the dramatic increase from to k 1.3 W cm1 K1 for the early samples to 2.3 W cm1 K1for the freestanding sample, as iterated above is most likely related to extended defect concentration (Dd) and the differences in background doping. The effect of dislocation density on the thermal conductivity has been calculated by Kotchetkov et al. [210]. The dislocation density in the thick film measured by Sichel and Pankove was between 109 and 1010 cm2, while the freestanding sample exhibited densities of less than 106 cm2 near the top surface (Ga-polarity) and 107 cm2 near the bottom surface (N-polarity). Kotchetkov et al. showed that k remains fairly independent of Dd up to some characteristic after which it decreases about a factor of 2 for every decade of value Dchar d increase in Dd. The thermal conductivity has also been correlated to doping levels in HVPE n-GaN/sapphire (0 0 0 1) by SThM on two sets of samples [211,212]. In both sets of data, the thermal conductivity decreased linearly with log n, n being the electron concentration, the variation being about a factor of 2 decrease in k for every decade increase in n. Significantly, it was concluded that the decrease in the lattice contribution to k, due to increased phonon scattering from impurities and free electrons, predominates the increase in the electronic contribution. Also, a correlation between the film thickness and the improved thermal conductivity was found, which is consistent with the observed general reduction of both extended (dislocations) and point defects with film thickness [212]. The k values at 300 K before and after plasma-induced effects on a series of n-GaN/ sapphire (0 0 0 1) samples fabricated by HVPE were also measured [213]. The sample thicknesses were 50 5 mm and the carrier concentrations were 8 · 1016 cm3, as determined by Hall effect measurements. The thermal conductivity before treatment was found to be in the 1.70–1.75 W cm1 K1 range, similar to that previously reported for HVPE material with this carrier concentration and thickness [211,212]. The k value was reduced, however, when the samples were processed under constant Ar gas flow and pressure for a fixed period of time (5 min). The only variable processing parameter was the DC bias voltage (125–500 V). After the initial 125 V procedure, k exhibited a linear decrease with the DC voltage in the investigated range. At 125 V, the thermal conductivity was only slightly less (k 1.65 W cm1 K1)
j59
j 1 General Properties of Nitrides
60
than the untreated case. The values of k had dropped to 0.3 W cm1 K1 for the 500 V case. To a first extent, the temperature dependence of the specific heat of Wz GaN (Cp) at constant pressure can be expressed by phenomenological expression [51]. In this vein, the specific heat Cp of Wz GaN at constant pressure for 298 K < T < 1773 K can be expressed as CP ðTÞ ¼ 9:1 þ ð2:14 10 3 TÞðJ mol 1 K 1 Þ; Cp ¼ 38:1 þ 8:96 10 3 T ðcal mol 1 K 1 Þ ð1 cal ¼ 4:186 JÞ:
ð1:15Þ
However, this expression is very simplistic, as will be seen below. As already mentioned, free electrons (very effective at low temperatures), impurities, defects (inclusive of point defects), and lattice vibrations contribute to specific heat. If GaN with negligible free-electron concentration and defects were available, only the lattice contribution would be considered, which is also the case in texts [214]. The specific heat of Wz GaN has been studied by Koshchenko et al. [215] in the temperature range of 5–60 K and also by Demidienko et al. [216] in the temperature range of 55–300 K and discussed by Krukowski et al. [88,127]. The Debye expression for the temperature dependence of specific heat in a solid at a constant pressure (Cp) can be expressed as [214]
T Cp ¼ 18R qD
3 xðD : 0
x 4 ex ð ex
1Þ2
dx;
ð1:16Þ
where x D qD =T and R ¼ 8.3144 J mol1 K1 is the molar gas constant. The coefficient in front of the term R has been multiplied by 2 to take into account the two constituents making up the binary GaN. By fitting the measured temperature-dependent heat capacity to the Debye expression, one can obtain the Debye temperature yD specific to heat capacity. The experimental data of Demidienko et al. and Krukowski et al. are plotted in Figure 1.26. Also shown in the figure is the calculated specific heat using the Debye expression for Debye temperatures of 500, 600, 700, and 800 K. It is clear that the quality of the data and/or sample prevents attainment of a good fit between the experimental data and Equation 1.16. Consequently, a Debye temperature with sufficient accuracy cannot be determined. It is easier to extract a Debye temperature using data either near very low temperatures or well below the Debye temperature where the specific heat has a simple cubic dependence of temperature [214]: 3 T Cp ¼ 234R : ð1:17Þ qD Unfortunately, the GaN samples contain large densities of free carriers and defects that compromise the application of the Debye specific heat expression. Consequently, a good fit to the data is not obtained and the Debye temperature so extracted is not as dependable as desired. There is a spread in the reported Debye temperatures for GaN, yD, in the range of about 600 to 700 K. Slack [77] estimated a value of 600 K at 0 K by utilizing the more established Debye temperatures for BeO and AlN. This compares with 550 K deduced
1.2 Gallium Nitride
50
40
q D = 800 K
–1
–1
Molar specific heat, Cp (cal mol K )
q D = 500 K
30
20
Cp data, GaN q D = 500 K q D = 600 K
10
q D = 700 K q D = 800 K
0 0
200
400
600
800
1000
Temperature (K) Figure 1.26 Molar specific heat at constant pressure, Cp (cal mol1 K1), of GaN versus temperature. Open circles represent the experimental data. The solid lines are calculation based on the Debye model for Debye temperatures, yD, of 500, 600, 700, and 800 K. Unfortunately, it is difficult to discern a Debye temperature that is effective over a wide
temperature range because a large concentration of defects and impurities is present in GaN. However, a value of 600 K estimated by Slack is used commonly. The data are taken from Refs [215,216], as compiled in Ref. [88]. (Please find a color version of this figure on the color tables.)
from heat transfer due to acoustic phonons, as mentioned above. Because the samples used in these measurements contained defects and large density of free electrons, the dispersion among the data and Debye expression is attributed to defects at high temperatures and free electrons at low temperatures. Elastic properties of GaN can also be used to deduce the Debye temperature. In this vein, Raman scattering measurements yielded a Debye temperature of yD ¼ 650 K [38]. Calculations since the estimate of Slack [77] yielded a range of 620–690 K [88]. Thermodynamic properties of Wz GaN have been reported by Elwell and Elwell [217]. From the reaction GaðsÞ þ 1=2N2 ðgÞ ¼ GaNðsÞ;
ð1:18Þ
the heat of formation of Wz GaN was calculated to be DH298 K ¼ 26.4 kcal mol1 [217], or as the standard heat of formation DH ¼ 37.7 kcal mol1 [194]. The equilibrium vapor pressure of N2 over solid GaN has been found to be 10 MPa at 1368 K and 1 GPa at 1803 K [202]. A thorough description of the GaN phase diagram including the equilibrium vapor pressure of N2 over GaN as well as AlN and InN has been presented by Porowski and Grzegory [218] (1 cal ¼ 4.186 J).
j61
j 1 General Properties of Nitrides
62
1.3 Aluminum Nitride
AlN exhibits many useful mechanical and electronic properties. For example, hardness, high thermal conductivity, resistance to high temperature and caustic chemicals combined with, in noncrystalline form, a reasonable thermal match to Si and GaAs in somewhat relaxed terms, make AlN an attractive material for electronic packaging applications. The wide bandgap is also the reason for AlN to be touted as an insulating material in semiconductor device applications. Piezoelectric properties make AlN suitable for surface acoustic wave device applications [219]. However, the majority of interest in this semiconductor in the context of electronic and optoelectronic device arena stems from its ability to form alloys with GaN producing AlGaN and allowing the fabrication of AlGaN/GaN and AlGaN/ InGaN-based electronic and optical devices, the latter of which is active from the green wavelengths well to the ultraviolet. AlN also forms a crucial component of the nitride-based AlInGaN quaternary, which makes tuning of the bandgap independent of composition to some extent. This way, lattice-matched conditions to the underlying epitaxial structure can be maintained while being able to adjust the bandgap. AlN is not a particularly easy material to investigate because of the high reactivity of aluminum with oxygen in the growth vessel. Early measurements indicated that oxygen contaminated material can lead to errors in the energy bandgap and, depending on the extent of contamination, in the lattice constant. Only recently achieved contamination-free deposition environments coupled with advanced procedures have allowed researchers to consistently grow improved-quality AlN. Consequently, many of the physical properties of AlN have been reliably measured and bulk AlN synthesized. 1.3.1 Mechanical Properties of AlN
When crystallized in the hexagonal wurtzite structure, the AlN crystal has a molar mass of 40.9882 g mol1, restated for convenience. The cubic form is hard to obtain and thus will be ignored. The point group symmetry for the wurtzite structure in the Schoenflies notation is C46v (P63mc in the Hermann–Mauguin notation), restated for convenience. Reported lattice parameters range from 3.110 to 3.113 Å for the a parameter (3.1106 Å for bulk, 3.1130 Å for powder, and 3.110 Å for AlN on SiC), and from 4.978 to 4.982 Å for the c parameter. The c/a ratio thus varies between 1.600 and 1.602. The deviation from that of the ideal wurtzite crystal (c/a ¼ 1.633) is probably because of the lattice stability and ionicity. The u parameter for AlN is 0.3821, which is larger than the calculated value of 0.380 using Equation 1.1. This means that the interatomic distance and angles differ by 0.01 Å and 3 , respectively, from the ideal [16]. Whereas the metastable zinc blende polytype AlN has a value of a ¼ 4.38 Å [220], the rock salt structure has a value of a ¼ 4.043–4.045 Å at room temperature [5,221]. Table 1.13 summarizes some of the observed structural properties of AlN.
1.3 Aluminum Nitride
Early investigations of the elastic properties of AlN were carried out on sintered polycrystalline specimens, owing to the unavailability of large single crystals. This, however, paved the way to more refined measurements as single crystalline AlN became available. The measured bulk modulus B, which is related to elastic stiffness coefficients through Equation 1.11, and Youngs modulus Y0 or E are compiled in Table 1.29 along with the entire set of elastic stiffness coefficients. The latter were obtained by fitting the results of surface acoustic wave measurements made on epitaxial AlN films and by Brillouin scattering measurements made on an AlN single crystal [160]. Wagner and Bechstedt [178] suggested that reliable values for the elastic constants should produce 2C13/C33 ¼ 0.5–0.6 and n ¼ 0.18–0.21. The ab initio calculations [42,178] and some measurements [83,165] provide similar results. However, the values from some of the calculations [165,222] and measurements [169,223] should be used with caution because of the large deviations in one or more coefficients. Surface acoustic wave measurements of Deger et al. [165] are very reliable because they include the correction for piezoelectric stiffening. Bulk modulus values range from 159.9 GPa, measured by an ultrasonic method, to 237 GPa, measured by Brillouin scattering. The range for the same from calculations is 111–239 GPa [160]. Youngs modulus is measured as 374 GPa for single-crystal AlN [82] and 295 GPa for AlN thin films [224]. The hardness of AlN has been measured to be 12 GPa on the basal plane (0 0 0 1) using a Knoop diamond indenter [225]. Some anisotropy in Knoops hardness has been observed for the indent direction perpendicular to the c-axis with measured values in
Table 1.29 Experimental bulk modulus and elastic coefficients (in GPa) of AlN (from Ref. [160] and references therein).
Method
C11
C12
C13
C33
C44
B
Ultrasonics [223] Ultrasonic Ultrasonic [165] ADX Brillouin [83] Brillouin [169] EDX Hardness [82] PWPP [42] FP-LMTO [165] HF [222] PWPP [178]
345
125
120
395
118
201 159.9 209 207.9 210.1 237 185 220 207 218 231 210
Zinc blende HF [222]
410
140
100
390
120
410.5 419
148.5 177
98.9 140
388.5 392
124.6 110
396 398 464 C11 FIX THIS 348
137 140 149 538 168
108 127 116 113
373 382 409 370
116 96 128
135
228
B0
E or Y0
5.2
334 308 354
6.3 354 326 5.7 374 329 322 365 322 Y0 ¼ (C11 þ 2C12) · (C11 C12)/ (C11 þ C12)
ADX: angular dispersive X-ray diffraction; EDX: energy dispersive X-ray; PWPP: plane wave pseudopotential; FP-LMTO: full-potential linear muffin–tin orbital; HF: Hartree–Fock.
j63
j 1 General Properties of Nitrides
64
the range of 10–14 GPa [81]. More recent nanoindentation measurements on singlecrystal AlN revealed a hardness of 18 GPa [82]. The phonon structure of AlN has been the subject of numerous investigations. As in the case of GaN, the phonon dispersion spectrum of Wz AlN has 12 branches, 3 acoustic, and 9 optical ones [226]. LO and TO phonon energies have been obtained from fits to infrared reflectivity measurements, the results of which are tabulated in Table 1.30. Raman-active optical phonon modes belong to the A1, E1, and E2 group representations. Several Raman scattering studies on AlN have been conducted, and the measured phonon energies are listed in Table 1.30 along with the calculated values by Wagner and Bechstedt [178]. The shift of phonon energies with strain was studied experimentally by Gleize et al. [227] and theoretically by Wagner and Bechstedt [178]. Gleize et al. [227] investigated strained 500 nm thick AlN layers grown on 6H-SiC. Using the strain values deduced from high-resolution X-ray diffraction and phonon frequency shifts measured by micro-Raman spectroscopy, the deformation potentials were obtained for most of the zone center optical phonons of Wz AlN. The determination of the deformation potentials is based on the knowledge of the ideal equilibrium state of the material from which strains and phonon shifts are defined. Additionally, hydrostatic pressure coefficients and elastic constants of the bulk material are also needed for extracting the deformation potentials from the raw experimental data. The deviations in the published data originate from the fact that different sets of parameters are typically used for this purpose. Table 1.30 lists the phonon deformation potentials from Gleize et al. [227] along with the results from pseudopotential plane wave calculations of Wagner and Bechstedt [178], which produced slightly lower values. The frequencies or energies of vibrational modes are very sensitive to the strain state of the samples. Strain inhomogeneities and imperfections cause the linewidths in Raman observable mode to be broad reducing the accuracy of central frequency determination. Tischler and Freitas Jr [228] utilized freestanding and high-quality AlN oriented along the (0 0 0 1) plane, as characterized by X-ray with full width half maximum (FWHM) of 36–54 arcsec. Listed in Table 1.30 are also the data of Tischler and Freitas Jr, which by virtue of the high quality of sample should be used as standard. Tischler and Freitas Jr [228] also estimated the linewidths of the Raman modes by fitting the data with Lorentzian peaks from which the phonon decay times were deducedby relying on theuncertainty principle in the form ofDE= h ¼ G= h ¼ 1=t, where DE, G, and t represent the error bar for the energy and linewidths, and the phonon decay time, respectively. The phonon decay times so deduced are in the range of 0.7 ps for E1(LO) to 5.3 ps for E12. To get a flavor of the range of experiments regarding the mechanical properties, including vibrational phonon properties, of epitaxial AlN, a mention of the investigations on AlN deposited on silicon and sapphire substrates at 325 K by IBAD undertaken by Ribeiro et al. [230] is made. Raman scattering measurements revealed interesting vibrational features related to the atomic composition and structure of the films. Features related to crystalline (c-) Si and corresponding to 2TA(L) at 230 cm1, 2TA(X) at 304 cm1, 2TA(S) at 435 cm1, TO(G) at 520 cm1, TA(S) þ TO (S) at 615 cm1, accidental critical points at 670 and 825 cm1, 2TO(D) at
E1-TO A1-TO A1-LO E1-LO E12 ðE2 lowÞ E22 ðE2 highÞ
Mode
82.8 75.4 75.4 112.8 30.5 81.2
(meV) 667.2 608.5 888.9 909.6 246.1 655.1
(cm 1)
Wz (Raman) unstrained [228]
895–921
667–673 614–667
Ref. [157]
246 655
668 608 890
Ref. [229]
Wz (Raman range) (cm1)
Table 1.30 Optical phonon energies and phonon deformation potentials for AlN.
924
677 618
Wz unstrained (calculated) (cm 1) [178]
bk 901 145 904 163
ak 982 83 930 94
Raman [227]
744 394 808
835 776 867
bk
Calculated [178] ak
Deformation potentials
1.3 Aluminum Nitride
j65
j 1 General Properties of Nitrides
66
1000
AlN
800
800
600
600
400
400
200
200
0 Γ
K
M
Γ
A
DOS
Frequency(cm –1 )
Frequency(cm–1)
1000
0
Figure 1.27 Phonon dispersion curves and phonon density of states for Wz AlN reported in Ref. [234]. The data points have their roots in Ref. [83] and are discussed and compared with the ab initio calculations in Ref. [233].
950 cm1, 2TO(L) at 980 cm1, 2TO(G) at 1085 cm1, 2TA(X) þ 2TO (G) at 1300 cm1, and 3TO(G) at 1450 cm1 have been observed. A very narrow peak seen at 2330 cm1 has been attributed to molecular nitrogen on the surface of c-Si [231]. It is worth noting that because of the extremely weak Raman signal usually presented by low quality AlN films, some of the previously reported features have been erroneously identified [232]. Misidentification of some vibration modes could lead to incorrect interpretations of the crystalline quality of AlN films. As in the case of GaN, the acoustic modes are simple translational modes, whereas the optical modes for wurtzite symmetry are more complex as shown in Figure 1.12. The calculated phonon dispersion curves [233,234] along with the phonon density of states for wurtzitic AlN are shown in Figure 1.27. 1.3.2 Thermal and Chemical Properties of AlN
Single crystalline forms of this compound either in the epitaxial form or bulk form represent the focus of this treatment. In its most commonly available form, AlN is an extremely hard ceramic material with a melting point higher than 2000 C. In single crystalline form, the melting of AlN was measured to be 2750–2850 C at nitrogen pressures of 100 and 200 atm (or bar) [235]. The melting temperatures for various nitrides were also determined by Van Vechten [236], who made use of a semiempirical theory for electronegativity and concluded that the melting point of AlN is close to 3487 K. Slack and McNelly [195] calculated the N2 equilibrium pressures over liquid Al to be 1, 10, and 100 bar at 2563, 2815, and 3117 C, respectively, as shown in Figure 1.18 in the context of the GaN discussion.
1.3 Aluminum Nitride
6.00 AlN
Δa/a
Δ (×10 3 ) Δa/a, Δc/c Δ
5.00
Δc/c
4.00
3.00
2.00
1.00
0.00 200
400
600
800 T (K)
1000
1200
1400
Figure 1.28 Variation of the thermal expansion coefficient of AlN with temperature in the c-plane and in the c-direction [86,87,225].
Using X-ray techniques across a broad temperature range (77–1269 K), it was noted by Slack and Bartram [85] that the thermal expansion of AlN is isotropic with a roomtemperature value of 2.56 · 106 K1. The thermal expansion coefficients of AlN measured by Yim and Paff [237] have mean values of Da/a ¼ 4.2 · 106 K1 and Dc/c ¼ 5.3 106 K1. The dependence of the thermal expansion coefficient on temperature in the c-plane and in the c-direction is shown in Figure 1.28, which can be fitted by the following polynomials (for 293 < T < 1700 K): Da=a0 ¼ 8:679 10 2 þ 1:929 10 4 T þ 3:400 10 7 T 2 7:969 10 11 T 3
ð1:19Þ
and Dc=c 0 ¼ 7:006 10 2 þ 1:583 10 4 T þ 2:719 10 7 T 2 5:834 10 11 T 3 ;
ð1:20Þ
where a0 and c0 represent the 300 K lattice parameters. For AlN powder, Krukowski et al. [88] reported the expansion coefficient for the a parameter to be 2.9 · 106 K1, and the same for the c parameter to be 3.4 · 106 K1. The specific heat of AlN has been discussed extensively. Mah et al. [238] approximated the specific heat Cp of AlN in the temperature interval 298–1800 K as Cp ¼ 45:94 þ 3:347 10 3 T 14:98 10 5 T 2 J mol 1 K 1
ð1 cal ¼ 4:186 JÞ:
ð1:21Þ
j67
j 1 General Properties of Nitrides
68
The same for the higher temperature range of 1800–2700 K was approximated by Glushko et al. [239] as Cp ¼ 37:34 þ 7:86 10 3 T J mol 1 K 1
ð1 cal ¼ 4:186 JÞ
ð1:22Þ
relying on the following points: the specific heat Cp ¼ 51.5 J mol1 K1 at T ¼ 1800 K, and an estimated value, Cp ¼ 58.6 J mol1 K1 at T ¼ 2700 K, as outlined by Krukowski et al. [88]. Specific heat obtained from the above approximations coupled with the measured values for a constant pressure from the literature has been tabulated [88], a plot of which is shown in Figure 1.29 along with the calculated specific heat using Equation 1.16 for Debye temperature values of 800–1100 K with 50 K increments. The best fit between the data and the Debye specific heat expression for insulators
50
Specific heat, Cp (J mol–1 K–1)
40
30
20 Specific heat AlN (J mol–1 K–1)
800 K 850 K 900 K 950 K 1000 K 1050 K 1100 K
10
0
0
200
400
600
800
1000
Temperature (K) Figure 1.29 Molar specific heat at constant pressure, Cp (J mol1 K1, 1 cal ¼ 4.186 J), of AlN versus temperature. Open circles represent the experimental data. The solid lines are calculation based on the Debye model for Debye temperatures, yD, in the range of 800–1100 K
with 50 K increments. The data can be fit with Debye expression for yD ¼ 1000 K, which compares with 950 K reported by Slack et al. The data are taken from Ref. [88]. (Please find a color version of this figure on the color tables.)
1.3 Aluminum Nitride
indicates a Debye temperature of 1000 K, which is in good agreement with 950 K reported by Slack et al. [240]. Compared to the GaN figure, the Debye temperature so obtained for AlN appears more dependable owing to a much better fit. The equilibrium N2 vapor pressure above AlN is relatively low compared to that above GaN that makes AlN easier to be synthesized. The calculated temperatures at which the equilibrium N2 pressure reaches 1, 10, and 100 atm are 2836, 3088, and 3390 K, respectively [195]. Details of the thermodynamic properties of AlN can be found in Refs [87,88]. Similar to GaN, albeit to a lesser extent, AlN exhibits inertness to many chemical etches. A number of AlN etches have been reported in the literature. However, molten salts such as KOH or NaOH at elevated temperatures such as 50–100 C, lower than what is required for GaN by as much as 200 C, etch AlN at appreciable rates. The surface chemistry of AlN has been investigated by numerous techniques including Auger electron spectroscopy, XPS, ultraviolet photoemission spectroscopy (UPS), ultraviolet photoelectron spectroscopy, and electron spectroscopy. One of these investigations by Slack and McNelly [195] indicated that the AlN surface grows an oxide 50–100 Å thick when exposed to ambient air for about a day. However, this oxide layer was protective and resisted further decomposition of the AlN samples. Details can be found in Refs [87,157]. The thermal conductivity k of AlN at room temperature has been predicted as k ¼ 3.19 W cm1 K1 [77,241]. Values of k measured at 300 K are 2.5 [241] and 2.85 W cm1 K1 [85]. The predicted values are near 3.2 W cm1 K1 in an O-free simulated material but are based on measurements in AlN containing O [77]. A more recent prediction for the value in AlN is 5.4 W cm1 K1, which is much larger than any measured value [208]. The measured thermal conductivity as a function of temperature in bulk AlN containing some amount of O is plotted in Figure 1.30a and b. Also shown is a series of samples with estimated concentrations of O showing an overall reduction in the thermal conductivity with O contamination. The calculation results of Slack et al. for impurity- and defect-free AlN are shown as well. In the temperature range of interest where many of the devices would operate, the thermal conductivity in the sample containing the least amount of O assumes a T 1.25 dependence. In freestanding and 300–800 mm thick AlN samples grown by HVPE, originally on Si(1 1 1) substrates, values in the range of 3.0–3.3 W cm1 K1 were measured by the SThM method [242]. The dislocation density in these freestanding AlN templates was about 108 cm2 [243]. 1.3.3 Electrical Properties of AlN
Owing to the low intrinsic carrier concentration and the deep native defect and impurity energy levels, the electrical characterization of AlN has usually been limited to resistivity measurements. One such measurement by Kawabe and coworkers [244] on transparent AlN single crystals yielded resistivities r ¼ 1011–1013 O cm, a value consistent with other reports [245–247]. However, it was found that impure crystals,
j69
j 1 General Properties of Nitrides
70
Thermal conductivity (W cm–1 K–1)
103
(a)
AlN
"Pure AlN"-Theor
4 x 1019 cm–3
102
5 x 1019 cm–3 2 x 10 20 cm–3 3 x 10 20 cm–3
101 100
(J/mol-1 K-1)
10 –1 10 –2 10 –3 10 –1
10 0
10 1 10 2 Temperature (K)
10 3
10 4
102 "Pure AlN" 19
Thermalconductivity(W cm–1 K–1)
4 x 10
19
5 x 10
20
2 x 10
20
3 x 10
–3
cm
–3
cm
–3
cm
101
100 10 (b)
–3
cm
100
1000
Temperature (K)
Figure 1.30 (a) Thermal conductivity of singlecrystal AlN. The solid line alone indicates the theory whereas the others represent measurements of AlN with various concentrations of O. The lower the O concentration, the higher the thermal conductivity. The theoretical data are from
Ref. [77], the data for the lowest O concentration are from Ref. [241], and the data for the other samples are from Ref. [91]. (b) The thermal conductivity of AlN in a limited temperature range of common interest that underscores the detrimental effect of O on the thermal conductivity [240].
1.3 Aluminum Nitride
which exhibited a bluish color possibly because of the presence of Al2OC, have resistivities of r ¼ 103–105 O cm, much lower than those reported by Chu et al. [248], who were purportedly able to obtain both n- and p-type AlN by introducing Hg and Se, respectively. However, they failed to determine the net carrier concentrations owing to very high resistivities. The n-AlN films grown by Rutz [249] had a quite low resistivity (r ¼ 103 O cm), which is comparable to those of Kawabe et al. [81]. Although Rutz [249] did not determine the source of the electrons, Rutz et al. [250] observed an interesting transition in their AlN films in which the resistivity abruptly decreased by two orders of magnitude with an increase in the applied bias. This observation found applications to switchable resistive memory elements that are operated at 20 MHz. The insulating nature of these early films hindered meaningful studies of their electrical transport properties. With the availability of refined growth techniques, AlN, presently grown with much improved crystal quality, shows both n- and p-type conductions. This has rejuvenated efforts to measure both electron and hole Hall mobilities. Edwards et al. [251] and Kawabe et al. [252] carried out some Hall measurements in p-type AlN producing a very rough estimate of the hole mobility mp ¼ 14 cm2 V1 s1 at 290 K. The predictions for the Hall mobility in the entire range of AlGaN alloy including the binary end points are treated in Volume 2, Chapter 3. Not all the parameters needed for the calculations are known precisely, reducing the confidence in predicted values somewhat. As in the case for GaN, the roomtemperature mobility is dominated by the polar optical phonon scattering. 1.3.4 Brief Optical Properties of AlN
Investigations of the optical absorption coefficient, a, of AlN at room temperature were reviewed as early as 1976 by Slack and McNelly [195]. Harris and Youngman [253] have reviewed photoluminescence and cathodoluminescence characteristics of AlN. Because the AlN lattice has a very large affinity to oxygen, it is almost impossible to eliminate oxygen contamination in AlN that affects observations. Commercially available AlN powder is said to contain about 1–1.5 at.% oxygen. Some oxygen is dissolved in the AlN lattice, with the remainder forming an oxide coating on the surface of each powder grain. After irradiation with ultraviolet light, AlN doped with oxygen was found to emit a series of broad luminescence bands at near-ultraviolet frequencies at room temperature no matter whether the sample was powdered, single crystal, or sintered ceramic. Pacesova and Jastrabik [254] observed two broad luminescence lines centered in the vicinity of 3.0 and 4.2 eV and more than 0.5 eV wide for samples contaminated at about 1 to 1.5 at.% oxygen. Youngman and Harris [255] and Harris et al. [256] investigated the luminescence characteristics of polycrystalline-sintered AlN samples and noted a continuous shift of the peak position in the ultraviolet luminescence line as a function of oxygen content up to a critical concentration of about 0.75 at.%. The luminescence lines beyond this limit of oxygen concentration remained stationary.
j71
j 1 General Properties of Nitrides
72
Epitaxial layers of about 1 mm or less in thickness on sapphire substrates led to observation of donor bound excitons with phonon replicas and some deep emission [257]. Cathodoluminescence spectra obtained in reactive MBE (RMBE), using ammonia as the nitrogen source, show a sharp band edge peak, which is tentatively assigned to the optical recombination of a donor bound exciton (D0, X), sometimes accompanied by weak one and two longitudinal optical phonon replicas. Also observed is a broad band with maximum in the range 320–370 nm, although some variation of this low-energy peak has been reported. The 6 K CL spectrum of the near band edge region shows a sharp and strong peak at 2068 Å in addition to two weak peaks on its low-energy side, which are not observed for all samples. The symmetric [0 0 0 2] X-ray diffraction peak is narrow and the linewidth of the prominent near band edge CL peak is 23 meV. Bulk single crystalline AlN has also been investigated for its CL emission properties. In addition to the near band edge emission at about 6 eV, the ultraviolet oxygen luminescence peak at 4.0 eV was also observed. In the investigation of Youngman and Harris [255], 4.64 eV photons were used to specifically explore the below the band transitions such as the peak near 4 eV. That investigation showed a blue shift with increased O concentration in the peak under question. With the availability of highquality AlN substrates, presumably high-quality epitaxial AlN layers have been grown on them and characterized for their optical properties by CL at 5 K [107]. In this particular work, the CL measurements were carried out at different temperatures for a fixed electron beam energy density value. The beam current was held at 5 mA and the voltage at 10 kV. The energy density was about 700 W cm2. Some six emission lines between the energies of 5.98 and 6.03 eV have been delineated in the lowtemperature near-band edge emission spectra of those AlN films. In addition to unidentified neutral donor and possibly acceptor bound exciton lines, free exciton A and its excited state, and free exciton B were observed. The availability of A exciton ground and excited states led to a binding energy of 63 meV that gives a lowtemperature bandgap of 6.086 eV for this material. The provisionally accepted value for the bandgap of this material is 6.2 eV (which is questionable now that the lower value is supported by measurements performed in homoepitaxial layers). In terms of absorption, Yim et al. [246] characterized AlN by optical absorption determining the room-temperature bandgap to be direct with a value of 6.2 eV. It should be pointed out that this early figure is not consistent with measurements performed later on using high-quality samples as discussed in the previous paragraph. Perry and Rutz [258] performed temperature-dependent optical absorption and determined a bandgap of 6.28 eV at 5 K compared to their room-temperature value of 6.2 0.1 eV. We should point again that it has been lowered to slightly over 6 eV at low temperatures in high-quality samples. Several groups have reported comparable values whereas others have produced questionable values considerably below 6.2 eV, probably due to oxygen contamination or nonstoichiometry. In addition to the band edge absorption, a much lower energy absorption peak at 2.86 eV (although some variation in the peak position has been recorded from 2.8 to 2.9 eV) is likely owing to nitrogen vacancies or nonstoichiometry as proposed by Cox et al. [245]. Yim et al. [237] also observed a broad emission spectrum range of 2–3 eV with a peak at about 2.8 eV.
1.3 Aluminum Nitride
10
4
(0001) AlN Single crystals 300 K Nitrogen vacancy
Absorption coefficient α (cm–1)
3
10
Oxygen
Pastrnak and Roskovcova
2
10
1
10
0
10
1
2
4 3 Energy (eV)
5
6
Figure 1.31 Room-temperature absorption spectra of several AlN films of varying thicknesses whose principal absorption edge occurs at 6.2 eV. The bump near 4.5–4.8 eV in the data is attributed to the oxygen absorption bands. The one at about 2.8 eV is attributed to N vacancies [245–247,259–261]. The ones not marked as Pastrnak and Roskovcova [259] are from Slack et al. [91].
This peak does not correlatewith the presence of oxygen. The oxygen absorption region lies between 3.5 and 5.2 eV, as originally found by Pastrnak and Roskovcova [259–261]. The exact position of this particular peak appears to change with the oxygen content from 4.3 eV at low oxygen content to 4.8 eV at high oxygen content. The results of the studies on the room-temperature absorption coefficient, a, are shown in Figure 1.31 for three different crystals along with the results of Pastrnak and Roskovcova [259]. Low-temperature and room-temperature absorption data taken at 2 K in a thin film on double side polished sapphire are shown in Figure 1.32. In the only optical study of AlN impurities, Karel and coworkers [262–266] reported on the luminescence of Mg and rare earth centers in AlN. Measurements of the refractive index of AlN have been carried out in amorphous, polycrystalline, and single-crystal epitaxial thin films. The values of the refractive index, n, are in the range of 1.99–2.25 with several groups reporting n ¼ 2.15 0.05.
j73
j 1 General Properties of Nitrides
74
2 × 10 10
RT absorption2
1.5 × 1010
1 × 1010
5 × 10
9
0 1900
1950
2000
2050
2100
2150
2100
2150
Wavelength (Å)
(a)
3.5 × 10 9 3 × 10 9
2K absorption 2
2.5 × 10 9 2 × 10 9 1.5 × 10 9 1 × 10 9 5 × 10 8 0 1900 (b)
1950
2000 2050 Wavelength (Å)
Figure 1.32 Room-temperature (a) and 2 K (b) absorption spectrum of a thin AlN film grown on double side polished c-plane sapphire by RF MBE. The data were taken at the University of Pittsburgh by Song Bay and W.J. Choyke using a sample prepared in authors laboratory.
These values are found to increase with increasing structural order, varying between 1.8 and 1.9 for amorphous films, 1.9–2.1 for polycrystalline films, and 2.1–2.2 for single-crystal epitaxial films. The spectral and polarization dependence of the index of refraction has been measured and showed a near-constant refractive index in the wavelength range of 400–600 nm. Some of these measurements also indicate that, in the long-wavelength range, the dielectric constant of AlN (e0) lies in the range of
1.4 Indium Nitride
8.3–11.5, and that most of the values fall within e0 ¼ 8.5 0.2. Other measurements in the high-frequency range produced dielectric constants of 4.68 and e1 ¼ 4.84. AlN has also been examined for its potential for second harmonic generation. Synchrotron radiation studies of AlN single crystals up to 40 eV have been performed, which resulted in the observation of an 8 eV luminescence peak. The same peak was also found in vacuum–ultraviolet reflection measurements.
1.4 Indium Nitride
InN forms the third binary anchor of the nitride family and was first synthesized in 1938 [267]. Compared to the other two, GaN and AlN, it is very difficult to form in high quality because of an inherent reason. The disparity in size, electronegativity, and very high vapor pressure of N over In imposes intractable tasks on the crystal grower [197,268]. The very early attempts explored absorption properties of polycrystalline InN films grown by DC discharge [269], reactive cathodic sputtering [270], or RF sputtering [271], and somewhat later RF deposition [272]. Inclusive properties of InN grown by various methods have been discussed in a review [273]. Essentially, the early results suggested a bandgap Eg ¼ 1.8–2.0 eV at room temperature. Those techniques gradually gave way to more refined growth methods such as MBE and OMVPE. The stoichiometry, however, has always been a pestering issue and will always remain so, which appears to have been one of the sources of controversy as to InNs true bandgap following the longstanding value of about 1.89 eV. An inordinate number of reports adorned many reputable journals and filled the programs in technical meetings wherein researchers in great numbers argued that the true bandgap of InN is actually 0.7–0.8 eV. In fact, lower values such as 0.65 [274] and 0.67 eV [275] have also been reported. What appeared to be experimentally impeccable results made the theorists to reexamine and redebate their band structure calculations [138,276]. The previously accepted larger value for the bandgap was explained away by O contamination and Moss–Burstein shift [277] because of high electron concentrations inherent to this semiconductor. Excellent reviews chronicling the developments in InN and controversy surrounding its bandgap are available in the literature [278,279]. Owing in part to visceral difficulties touched upon above and its bandgap, regardless of the controversy, InN has received nowhere near the attention given to GaN and AlN. Reiterating, the problems with InN are difficulties in growing highquality crystalline InN samples, poor luminescence properties of InN, and the existence of alternative well-characterized semiconductors such as AlGaAs and (Ga, Al)AsP, which have energy bandgaps close to that of the old value of InN bandgap (1.89 eV) and InGaAs close to that of the new InN bandgap. Setting the bandgap aside, which is the holy grail of optoelectronic devices, InN possesses the largest roomtemperature mobility among all the nitride-based semiconductors. Predictions point to InN being a superior channel layer with its higher mobility for field effect transistors, a topic that will receive more coverage in the pages to follow.
j75
j 1 General Properties of Nitrides
76
Just when the data and calculations appear to indicate the bandgap of InN to be close to 0.7–0.8 eV [30], there came the reports casting doubt on the accuracy of the small bandgap. The controversy regarding the true bandgap of InN brought this semiconductor, from obscurity to controversy, in the words of Scott Butcher, who is one of the pioneers involved in InN studies of and instrumental in determining the 1.89 eV bandgap [278]. The controversy is actually not limited to just the bandgap. It spans the whole gambit of its properties including the lattice constant, the effective electron mass, not to mention the hole mass, which is simply left in the dark, the role of hydrogen and oxygen, nonstoichiometry-induced defects, and point and extended defects. Returning to the controversy surrounding the bandgap of InN, proponents of the smaller bandgap argue that the measured larger bandgap from absorption data is most likely skewed by Moss–Burstein shift and/orO contamination.Theyare alsoquick toargue that emission near 1.8–1.9 eV has not been observed casting doubt on its accuracy. The opponents of the larger bandgap put forward arguments ranging from Mie resonance owing to scattering or absorption of light in InN-containing clusters of metallic In [280] to In-rich nonstoichiometry-induced defects active near the 0.7–0.8 eV region [281]. In the efforts of the authors of Ref. [280], microcathodoluminescencestudies coupled with imaging of metallic In have shown that bright infrared emission at 0.7–0.8 eV arises in the close vicinity of In inclusions and is likely associated with surface states at the metal/InN interfaces. Employing thermally detected optical absorption measurements, Shubina et al. [280] suggested that a true bandgap near 1.5 eV, reserving a more accurate figure for the bandgap until after more detailed measurements are carried out. The presence of In inclusions would also make suspect the mobility and doping level data published for this material. In fact, metal inclusions placed by design inGaAs have been shown to skew the electron mobility [282]. Setting this issue aside for now and referring the reader to Chapter 2 for bandgap-related discussion and Chapter 3 for growth-related discussion, let us segue into the discussion of InN properties. As mentioned, the energy bandgap of InN corresponds to a portion of the electromagnetic spectrum in which alternative and well-developed semiconductor technologies are already available. Consequently, practical applications of InN are more or less restricted to its alloys with GaN and AlN or related heterostructures. The growth of high-quality InN and the enumeration of its fundamental physical properties remain for the present a purely scientific enterprise except of course their impact on the properties of the ternaries it makes with GaN and AlN. InN is not different from GaN and AlN in the sense that it suffers from the same lack of a suitable substrate material and, in particular, a high native defect concentration. All these have hindered its progress. In addition, because of its rather poor thermal stability InN cannot be grown at the high temperatures required by CVD growth processes. As InN rapidly dissociates at high temperatures, even as low as 600 C, an extraordinarily high nitrogen overpressure would be required to stabilize the material up to the melting point, which is practically impossible. The large disparity of the atomic radii of In and N is another factor that increases the difficulty in obtaining InN of good quality. Notwithstanding the aforementioned characterization, the seminal work of Tansley and Foley [271] first characterized many of the fundamental physical properties of InN.
1.4 Indium Nitride
Tables 1.19–1.22 list the physical properties of InN. It has proved difficult to grow high-quality single crystalline material that would enable detailed optical, structural, and electrical measurements to be performed. Given the fact that the growth of bulk single-crystal InN films using equilibrium techniques is unlikely, attention turned to the deposition of thin films using nonequilibrium techniques. All of the early data summarized below under various properties, unless otherwise specified, were obtained from highly conductive n-type polycrystalline InN grown by nonequilibrium techniques. There have been several studies that report rapid dissociation of InN at temperatures above 500 C. Because no high-quality InN has yet been grown, the resistance of the high-quality material to chemical etching is unknown. Successful etching of single crystalline InN films in a hot H3PO4 : H2SO4 solution has been measured as was surface oxidation. 1.4.1 Crystal Structure of InN
To reiterate, indium nitride normally crystallizes in the wurtzite (hexagonal) structure. The zinc blende (cubic) form has been reported to occur in films containing both polytypes. Because of the absence of good-quality single-crystal films, early studies dealing with the crystal structure of InN were limited to mainly less than ideal thin films, particularly the ordered polycrystalline films with crystallites in the thickness range of 50–500 nm. Basically, the measurements confirm that, although InN normally crystallizes in the Wz structure, it occasionally also crystallizes in the zinc blende (cubic) polytype. Thermal instability of InN forbids growth at high temperatures and a large lattice mismatch with most available substrates inevitably leads to less than perfect structural quality of the epitaxial films grown by any method although the electron mobility in MBE-grown layers have improved considerably. A study based on XRD analysis has provided important insights into the dependence of the structural properties on the degree of lattice mismatch and film thickness of InN [283] on three different kinds of substrates, namely sapphire, GaN, and AlN. Significant improvement in the structural quality of the InN films, which do suffer from the residual strain, was observed on GaN templates. Below a thickness of 1200 Å, the InN film is composed of grain islands with different crystal orientations. Above this thickness, screw dislocations are nucleated, relieving the strain and leading to a reduction of the observed mosaicity in the surface morphology as grains with the same orientation grow with film thickness exceeding 1200 Å. 1.4.2 Mechanical Properties of InN
The measured InN lattice parameters using powder technique are in the range of a ¼ 3.530–3.548 Å and c ¼ 5.704–5.960 Å with a consistent c/a ratio of about 1.615 0.008. This c/a ratio is close to the more optimistic value of 1.633 determined
j77
j 1 General Properties of Nitrides
78
from layers especially grown under significant precautions, best possible growth conditions, and presumably with reduced nitrogen vacancies [284]. Another value of the c/a ratio of 1.612 has also been reported using powder diffractometry [29]. There is no reliable experimental u parameter for InN. An examination of the reported data indicates an unacceptably large scatter. This is possibly caused by nitrogen deficiency because nitrogen atoms are closely packed in (0 0 0 1) planes. The single reported measurement yields a lattice constant of a0 ¼ 4.98 Å in zinc blende (cubic) form InN occurring in films containing both polytypes [284]. While the cubic polytype of InN yields a molecular cell volume of 30.9 Å3, the hexagonal polytype gives a molecular cell volume of 31.2 0.2 Å3. The experimental density of InN deduced from Archimedean displacement measurements is 6.89 g cm3 at 250 C [285]. This is comparable with 6.81 g cm3 estimated from X-ray data. In a hexagonal structure, the second-order elastic moduli are C11, C12, C13, C33, and C44. The only report on InN elastic coefficients is by Savastenko and Sheleg [162], but their results are lower than the values calculated by the linear muffin–tin orbital (LMTO) and the plane wave pseudopotential (PWPP) methods and are suggested to be completely unreliable [178]. Table 1.31 summarizes the measured and the calculated elastic coefficients for both Wz and ZB InN. Because these figures depend on the lattice constants that are within some 10%, values of other nitrides can be used as a first approximation when absolutely needed [271,286]. The bulk modulus has been calculated from first principles by the local-density approximation [287] and by the LMTO method [288] suggesting bulk modulus B ¼ 165 GPa. The results of other calculations for bulk modulus are shown in Table 1.31. Most of the properties of InN are tabulated in Tables 1.19–1.22. As in the cases of Wz GaN and Wz AlN, Wz InN has 12 phonon modes at the zone center (symmetry group: C6v), 3 acoustic and 9 optical ones with the acoustic
Table 1.31 Theoretical and experimental elastic coefficients and bulk modulus (in GPa) of the various forms of InN [160].
Method
Structure
C11
C12
C13
C33
C44
B
X-ray [162] LMTO [24] PWPP [42] PWPP [42] PWPP LMTO PWPP PWPP ADX LMTO PWPP PWPP
Wz Wz Wz ZB ZB Wz Wz ZB Wz Wz Wz ZB
190 271 223 187
104 124 115 125
121 94 92
182 200 224
10 46 48 86
139 141 145.6 155 165 166 138 125.5 165 139 140
dB/dP ¼ B0
4 3.8 3.9 12.7
ADX: angular dispersive X-ray diffraction; PWPP: plane wave pseudopotential; LMTO: linear muffin–tin orbital.
1.4 Indium Nitride
branches near zero at k ¼ 0. The infrared active modes are of the E1(LO), El(TO), A1(LO), and A1(TO) type. Raman spectroscopy studies [289,290] have yielded four optical phonons characteristic for InN with wave numbers 190 cm1 (E2), 400 cm1 (A1), 490 cm1 (E1), and 590 cm1 (E2) in InN layers grown by atomic layer epitaxy (ALE). Moreover, a TO mode has been observed at 478 cm1 (59.3 meV) by reflectance and 460 cm1 (57.1 meV) by transmission measurements [286]. From other reflectance data, the existence of a TO phonon mode at 478 cm1, consistent with Ref. [286], and an LO mode at 694 cm1 was deduced [291]. 1.4.3 Thermal Properties of InN
The linear thermal expansion coefficients measured at five different temperatures between 190 and 560 K [292] indicate that both along the parallel and perpendicular directions to the c-axis of InN these coefficients increase with increasing temperature. Thermal conductivity derived from the Leibfried–Schloman scaling parameter, assuming that the thermal conductivity is limited by intrinsic phonon–phonon scattering, is about 0.80 0.20 W cm1 K1. Using InN microcrystals prepared by microwave nitrogen plasma, the specific heat of InN has been measured with differential scanning calorimeter with a precision better than 1% [88,293]. The data have been fit to the Debye equation, Equation 1.16, as shown in Figure 1.33a. An expanded view of the experimental data over a temperature range of 150–300 K is given in Figure 1.33b. Below 200 K, a Debye temperature of 600 K appears to fit the data well, tabulated in Table 1.32. However, above 200 K a Debye temperature of 700 K fits the data better. This paradox indicates the poor quality of InN and significant contribution by nonvibrational modes, as the Debye theory is developed for a perfect dielectric. Others [88] argue that the Debye temperature of yD ¼ 660 K, albeit in a relatively small range, describes InN best. As shown in Figure 1.18, the melting point of InN, T M InN , is 2146 K and the line fit to the partial pressure data for N indicates a temperature dependence of 7.9 1017 exp ( 2.78 eV/kT) bar for InN. As can be seen in Figure 1.18, the nitrogen partial pressure increases exponentially above TE ¼ 630 C for InN, illustrating that the decomposition temperature in vacuum is much lower than the melting point achievable under high pressures. To determine the effective decomposition activation energy more precisely, the nitrogen flux was calculated from the measured nitrogen pressure [196,199]. The rate of nitrogen evolution F(N) is equal to the rate of decomposition and the slope of ln[F (N)] versus 1/T in Figure 1.18 gives the effective activation energy of decomposition in vacuum, EMN. The decomposition rate is that corresponding to desorption of one monolayer in 1 s, in other words, FN ¼ 1.5 · 1015 cm2 s1, at 795 C. The activation energy of the thermally induced decomposition is determined as EMN ¼ 3.5 eV (336 kJ mol1) for InN (M is for metal and N is for nitrogen). This indicates temperature limits for high-temperature or high-power devices. This together with the reported small values of InN bandgap does not bode well for power devices based on InN alone. Of course, the picture is different if this
j79
j 1 General Properties of Nitrides
80
50
–1
Cp (J mol–1 K )
40
30 –1
–1
Cp (J mol K ) T = 600K T = 700K T = 800K
20
10
0 0
(a)
200
400 600 Temperature (K)
800
1000
–1 Specific heat, C p (J mo l K–1)
40
35
30
25
20 150 (b)
200
250 Temperature (K)
300
Figure 1.33 (a) The specific heat of InN with experimental data points, albeit in a small range, and the 600, 700, and 800 K Debye temperature fits. Data from Refs [88,127]. (b) Experimental specific heat of InN in a temperature range of 150–300 K [88,127].
material is used in conjunction with GaN if the associated technological difficulties can be overcome. While the heat capacity of InN is 9.1 2.9 · 103 cal mol1 K1) at temperatures between 298 and 1273 K, the entropy is 10.4 cal mol1 K1 at 298.15 K. The
1.4 Indium Nitride Table 1.32 Specific heat, Cp, of InN at constant pressure [88,127].
T (K)
Cp (J mol1 K1)
153 163 173 183 193 203 213 223 233 243 253 263 273 283 293
25.38 26.54 27.96 29.12 30.15 31.18 31.95 32.59 33.50 34.26 35.17 35.81 36.97 37.61 38.65
equilibrium partial pressure of N2 above InN is about 1 atm at 800 K and it increases exponentially with temperature to 105 atm at 1100 K.
1.4.4 Brief Electrical Properties of InN
It is fair to state that reliable experimental data for the true electron mobility in InN is waiting to be obtained. As mentioned repeatedly, InN suffers from the lack of a suitable substrate material and high native defect concentrations that limit its quality. In addition, a large disparity of the atomic radii of In and N makes it more difficult to obtain InN of high quality. As a result, nitrogen vacancies are thought to lead to large background electron concentrations in InN. Because of all these factors, the electron mobilities obtained from various films vary very widely, as can be inferred from Table 1.33. The electron mobility in InN can be as high as 3000 cm2 V1 s1, perhaps even much higher, at room temperature [294]. A study of the electron mobility of InN as a function of the growth temperature indicates that the mobility of ultrahighvacuum electron cyclotron resonance radio-frequency magnetron sputtering (UHV ECR-RMS) grown InN can be as much as four times the mobility of conventionally grown (vacuum deposition) InN [295]. The dependence of the electron concentration and mobility on the InN film thickness grown on AlN and GaN buffer layers by plasma-enhanced MBE is shown in Figure 1.34. Hall measurements in InN films grown on AlN buffer layers [300] that are in turn grown on sapphire indicated electron mobility to be 1310 cm2 V1 s1 at room temperature [307] for MBE-grown and 830 cm2 V1 s1 for OMVPE-grown material [304]. As discussed in Section 3.5 in conjunction with growth-related issues,
j81
j 1 General Properties of Nitrides
82
Table 1.33 A compilation of electron mobilities obtained in InN on different substrates and for various deposition conditions in InN as compiled in part in Ref. [296].
n-type carrier concentration (cm3)
Carrier mobility (cm2 V1 s1)
5–8 · 1018
250 50
1020 3–10 · 1018
20 20–50
1–200 · 1018 2–80 · 1020 5 · 1018 6 · 1020 6 · 1016
3 35–50 20 2 2
7–70 · 1016 3 · 1016 at 150 K 1020
730–3980 5000 at 150 K 10
4.8 · 1020 1–8 · 1020 1–10 · 1020 2–3 · 1020
38 50 50 20–60
2 · 1020 1020ae 2.0 · 1020 5.98 · 1018
100 220a 100 363
Sapphire Sapphire Sapphire Sapphire, silicon, mica GaAs GaAs GaAs Glass
5.0 · 1019 3.0 · 1018
700 542
Sapphire Sapphire
8.8 · 1018 2–3 · 1018 1019
500 800 306
Sapphire Sapphire Glass, KBr
1.0 · 1019 4 · 1017
830 2100
Sapphire Sapphire
1.4 · 1018
1420
Sapphire
5 · 1016 (3 · 1016 at 150 K)b
2700 (5000 at 150 K)b
a b
Zinc blende polytype. Have not yet been reproduced by others.
Substrate
Deposition technique
Sapphire, silicon, various metals Sapphire Glass, fused quartz Fused quartz Sapphire Glass, NaCl Fused quartz Glass, silicon, 304 stainless steel Glass, silicon Glass, silicon Glass
Reactive sputtering Reactive evaporation Reactive sputtering Reactive sputtering CVD Reactive sputtering Cathodic sputtering RF ion plating Reactive sputtering Reactive sputtering Reactive DC magnetron sputtering Magnetron sputtering Plasma-assisted MOVPE MOVPE Reactive RF magnetron sputtering ECR-assisted MOMBE Plasma-assisted MBE ECR-MOMBE [297] Reactive RF magnetron sputtering [298] MOVPE [299] Migration-enhanced epitaxy [300] MOMBE [301] MBE [302] RF reactive ion sputtering [303] RF MBE [304] MBE on thick HVPE GaN [305] Plasma-assisted MBE [306] Sputtering [308]
10
21
104
10
20
103
10
19
102
10
18
10
InN on GaN
mobility electron conc.
InN on AlN
mobility electron conc.
17
10
1
10
2
Mobility (cm2 V-1 s-1)
Electron density (cm–3)
1.4 Indium Nitride
101
100 3
10 Thickness (nm)
10
4
Figure 1.34 Electron density and mobility as a function of InN thickness. Samples grown on a GaN buffer are indicated by diamond and square symbols, and on an AlN buffer are indicated by circles and inverted triangles. The diamond and inverted triangle symbols indicate the electron mobility, whereas the squares and circle symbols indicate the electron concentration, all measured at room temperature. Courtesy of W.J. Schaff.
room-temperature mobilities increased to 2000 cm2 V1 s1 in MBE-grown InN films. Some directions, such as the insertion of low-temperature intermediate layers or AlN buffer layers have been pointed out for improvement. Very high inadvertent donor concentrations, >1018 cm3, seem to be one of the major problems for further progress. ON and SiIn have been proposed to be the likely dominant defects responsible for high electron concentration based on their low formation energies. Additionally, H has been proposed [307] as the dominant impurity candidate for the state-of-the-art MBE-grown InN. Assuming that the measurements were performed flawlessly and interpreted, there clearly seems to be some way to go to reach the goal of a mobility of 2700 cm2 V1 s1 at an electron concentration of 5 · 1016 cm3 reported for RF reactive ion sputtered growth of InN nearly a few decades earlier [308]. Ensemble Monte Carlo calculations have been the popular tool to investigate the carrier velocity field characteristics theoretically. Although early application of this method to InN was done by OLeary et al. [309], a more detailed calculation based on the full details of the conduction band structure appeared later in a paper by Bellotti et al. [310]. A peak electron drift velocity of 4.2 · 107 cm s1 has been predicted at an electric field of 65 kV cm1, substantially higher than that in GaN,
j83
j 1 General Properties of Nitrides
84
with a noticeable anisotropy for field direction parallel or perpendicular to the basal plane. The velocity is seen to decrease to 3.4 · 107 cm s1 with an increase of field to 70 kV cm1 and reach a value of 2.0 · 107–1.0 · 107 cm s1 at the onset of impact ionization, for field directions parallel and perpendicular to the basal plane, respectively. The calculated low-field mobility is 3000 cm2 V1 s1. This study claims to report the first calculation of high-field electron transport in InN. Electron–phonon coupling would hurt the high-field velocity as it seems to be the case in GaN, see Volume 3, Chapter 3 for details. Another interesting aspect of electron transport is its transient behavior, which is relevant to short channel devices with dimensions smaller than 0.2 mm, where a significant overshoot is expected [311] to occur in the electron velocity over the steadystate drift velocity. In yet another calculation, Foutz et al. [311] found that of the three III nitride binaries, GaN, InN, and AlN, this overshoot is most pronounced in InN and occurs above a critical field of 65 kV cm1. The peak velocity at this field is 4.2 · 107 cm s1 and the velocity overshoot is retained over longer distances as compared to that for GaN and AlN. 1.4.5 Brief Optical Properties of InN
As mentioned in the opening statements for Section 1.4, while the bandgap value reported in the early stages of InN development dating back to as early as 1980s, a controversy developed as to the true value of the InN bandgap developed. This is detailed in Section 2.9.2. However, a brief treatment is provided here for completeness. Excellent reviews chronicle and detail the evolution of the controversy in InN [278,279]. A number of groups have described optical measurements performed on InN [25,313,314]. Early values of the room-temperature InN direct bandgap ranged from 1.7 to 2.07 eV. A value of 1.89 eV was measured by optical absorption by Tansley and Foley [308], who also measured the infrared absorption of InN and observed an unidentified donor level approximately 50–60 meV below the conduction band edge. Reflectance spectroscopy on single-crystal material, using synchrotron radiation over the range 2–20 eV[315], later extended to 130 eV [316], was performed to determine the optical parameters of InN. A few studies of the interband optical absorption performed on InN thin films deposited by sputtering techniques [271] and OMVPE [103], were found consistent with a fundamental energy gap of about 2 eV. However, weak photoluminescence peaks with energies ranging from 1.81 to 2.16 eV were observed for InN grown on Si substrates later on [317]. In one such case, an emission centered at 1.86 eV at a temperature below 20 K was seen, whereas the reflectance measurement showed a strong plasma reflection at 0.7 eV, corresponding to an effective mass of 0.12m0. Another branch of studies shows that in improved InN films a strong photoluminescence transition at energies around 1 eV [30,318,319] appears. Observing that the position of the photoluminescence energy agrees with the onset of strong absorption, the optical transition at about 1 eV has been attributed to the fundamental bandgap [319].
1.4 Indium Nitride
4
Assuming large gap Assuming small gap
E g (eV)
3
2
1
0 0 InN
20
40
60
80
Composition
100 In2 O3
Figure 1.35 Vegards law plot of InN–In2O3 pseudobinary alloy system with both 0.7 and 1.9 eV bandgaps shown (dashed lines and solid line, respectively) versus O composition reaching In2O3, which has a bandgap of 3.75 eV reported in Ref. [320]. Also shown is the calculated (using LCAO) bandgap with 10% oxygen if O-free bandgap is 0.7 eV [278,322].
Following the wide acceptance of the above-mentioned data by the community, a couple of crystal growth groups, often in collaborations with others who focused on characterization, began to question the nearly 1.9–2.0 eV bandgap, because the newer and supposedly more improved samples showed a strong emission at much smaller energies, primarily around 0.7–0.8 eV. Those latter groups argued that the early samples did not show efficient PL near the band edge and had to have been contaminated with O to support the earlier measurements. One should, however, keep in mind that In2O3 has a bandgap of 3.75 eV [320], and if the Vegards law is applicable it would take some 35% O in InN to boost its bandgap from say 0.8 to 1.9 eV, which is substantial, as shown in Figure 1.35. Also shown is the bandgap of InN for 10% O contamination assuming an O-free bandgap of 0.7 eV [321,322]. Not only large amounts O is needed, but also that O must form an alloy with InN. Clearly, the assignment of the 1.9 eV bandgap to O, on the premise that the bandgap for O-free InN is 0.7–0.8 eV, requires large amounts of O, which is not supported by experiments, as discussed in detail in Section 2.10.1. To account for the 1.9 eV measured bandgap, the proponents of the 0.7 eV bandgap also suggested that the former result could be accounted for by Moss–Burstein blue shift owing to high electron concentration. This effect relates to semiconductors where the electron concentration is larger than the density of states and the Fermi level is actually in the conduction band itself. The extent of the penetration naturally depends on the electron concentration. Consequently, the measured apparent optical gap is skewed upward. This possibility was suggested in 1974 by Trainor and Rose [323] and has been the topic of an extensive study using In2O3 as the model wide bandgap material [320]. What accompanies the Moss–Burstein shift is the
j85
j 1 General Properties of Nitrides
86
2.4
E G +E F (eV)
2.2 2.0 1.8 1.6 1.4 1.2 1.0 0.8 0.6
Moss–Burstein effect
0.4 10 16
10 17
10 18
10 20
10 19
a b c d e f g h i
10 21
–3
Carrier concentration (cm ) Figure 1.36 The measured bandgap of InN (combination of the bandgap and the Moss–Burstein shift) deduced from absorption measurements versus the carrier concentration reported by various groups. The solid line shows the expected blue shift in the bandgap because of Moss–Burstein shift. Solid and dashed lines
indicate nonparabolic and parabolic theories, respectively. (a) Ref. [341]; (b) Ref. [324]; (c) Refs [274,318]; (d) Ref. [331]; (e) Ref. [271]; (f) Refs [303,325]; (g) Refs [131,272,326,327]; (h) Ref. [328]; (i) Ref. [329]. Collated by Butcher and Tansley [278,279].
bandgap renormalization, which is a red shift, caused by tail states extending into the bandgap and counters the former effect to some extent. Here too there is a controversy in that the samples of older times did not indicate large Moss–Burstein effect, which has been attributed by the proponents of the 0.7 eV bandgap, as having been heavily reduced owing to band tail states. Butcher [278] collated the apparent bandgap (presumably a combination of the bandgap with its band tailing and Moss–Burstein effect) measured by optical absorption and the electron concentration data for a large group of samples as shown in Figure 1.36. Although an argument for Moss–Burstein effect can be made for samples with high carrier concentrations, the same cannot be applied to a good many samples exhibiting the large bandgap while having low carrier concentration. To be thorough, one should recognize that many of the InN samples contained in Figure 1.36 with low carrier concentrations are heavily compensated, casting some amount of uncertainty as well. Turning the argument around and assuming that the large bandgap data are more dependable, the low bandgap data could be explained with some defect level in In-rich material or by Mie resonant absorption due to In inclusions. These are discussed in Section 2.10.1. After discussing the controversy regarding the bandgap and making many references to the low bandgap measured in InN, let us briefly discuss the data obtained in layers grown by MBE under In-rich conditions, which exhibit the so-called small bandgap. The optical absorption data in InN, purportedly having 0.7–0.8 eV bandgap, show an onset at 0.78 eV not near 1.9–2.0 eV, as shown in
1.4 Indium Nitride
Figure 1.37a. The absorption coefficient increases gradually with increasing photon energy and at the photon energy of 1 eV it reaches a value of more than 104 cm1. This high value is consistent with an interband absorption in semiconductors. Moreover, the integrated PL intensity increased linearly with excitation intensity over three orders of magnitude, lending more credence to the notion that the observed nonsaturable peak relates to the fundamental interband transition. The absorption squared versus the energy plots used to obtain the apparent bandgap in a semiconductor with very high carrier concentration underestimates the bandgap owing to band tailing. Briot et al. [330,331] estimated a bandgap near 1.2 eV from the absorption data while taking the large carrier concentration of 1019 cm3 into account. This, however, does not account for the large discrepancy between the 0.7–0.8 eV group and 1.8–2.0 eV group. The free-electron concentration in this sample was measured by Hall effect to be 5 · 1018 cm3. Figure 1.37a also shows that the samples exhibit intense roomtemperature luminescence at energies close to the optical absorption edge. Additionally, the 77 K photoreflectance (PR) spectrum exhibits a transition feature at 0.8 eV with a shape characteristic for direct gap interband transitions. Consistent with the absorption data, no discernible change in the PR signal near 2 eV is seen. The simultaneous observations of the absorption edge and PL and PR features at nearly the same energy led Wu et al. [30] to argue that this energy position of 0.78 eV is the fundamental bandgap of InN. This value is very close to the fundamental gap for InN reported by the other group, Davydov et al. [319], in favor of the smaller bandgap for InN. Tsen et al. [332] studying nonequilibrium optical phonons in a high-quality single-crystal MBE-grown InN with picosecond Raman spectroscopy reached the conclusion that their results are not consistent with the large bandgap of InN. Using the possible phonon emission allowed by energy and momentum conservations for a range of excitation photon energies, they argued that their observations are consistent with the small bandgap of InN. The basis of the argument is that if the bandgap energy were 1.89 eV, no nonequilibrium phonons could be observed, contradicting the observation of Tsen et al. [332], details of which are discussed in Section 2.10.1. Figure 1.37b shows the room-temperature electron mobility, the peak energy of PL, and the transition energy determined by PR as functions of electron concentration, showing that the transition energies increase with increasing free-electron concentration. This indicates that the transitions from higher energy occupied states in the conduction band contribute significantly to the PL spectrum. The PL energy as a function of pressure shows a linear pressure coefficient of 0.6 meV kbar1, which is considerably smaller than that for other III–V compounds. For comparison, the pressure coefficient of GaN is 4.3 meV kbar1 [333], AlxGa1xN is 4.1 meV kbar1 for 0.12 < x < 0.6 [334], and GaAs is 11 meV kbar1 [335], as compiled by Wu et al. [30]. The sapphire substrate on which the InN layer is grown has a larger bulk modulus than InN, which will reduce hydrostatic pressure transmission to the InN film providing that the film remains coherently strained. Using experimental elastic constants for sapphire and theoretical elastic constants for InN, Wu et al. [30] estimated the correction factor for coherently strained InN on sapphire to be 1.45, leading to between 0.6 and 0.9 meV kbar1 for the bandgap change. More work is needed to shed light on this unusually low pressure dependence although
j87
j 1 General Properties of Nitrides
88
8
6
PL or PR signal
ab. at 300 K 5 4 3 2
PR at 77 K
Absorption coefficient (104 cm-1)
7
PL at 300 K
1
0.5
1
1.5
2
0 2.5
Energy (eV)
(a) 0.90
103 mobility at 300 K
Energy (eV)
102
0.80
PL at 300 K PL at 12 K
Mobility (cm2 V-1s-1)
0.85
101
0.75 PR at 77 K
100
0.70 1018
(b)
1019
1020
Electron concentration, n (cm–3)
Figure 1.37 (a) Optical absorption (300 K), PL (300 K), and PR (77 K) spectra of a typical InN sample. This sample is undoped with roomtemperature electron concentration of 5.48 · 1018 cm3. The spike on the PR spectrum at 0.97 eV is an artifact due to the light source used in the PR measurement. (b) Room-
temperature mobility, PL peak energy (300 and 12 K), and the critical energy determined by PR (77 K) as a function of free-electron concentration. The sample with n ¼ 1 · 1019 cm3 (indicated by an arrow) is the Ritsumeikan sample.
1.5 Ternary and Quaternary Alloys
studies of InGaN showing smaller pressure dependence as the InN concentration is increased are consistent with data on InN [336]. Tyagai et al. [337] performed reflection and transmission measurements in InN with electron concentration larger than 1020 cm3. They were able to estimate an effective mass of me ¼ 0:11m0 and an index of refraction of n ¼ 3.05 0.05, which is in reasonable agreement with the value measured by Tsen et al. [332]. The longwavelength limit of the refractive index was reported to be 2.88 0.15. The temperature dependence of the InN bandgap indicates a bandgap blue shift of 23 meV from 300 to 77 K [291,338]. Inushima et al. [132] reported the effective mass to be 0.24m0 in InN grown by UV-assisted atomic layer epitaxy under atmospheric pressure. Using infrared spectroscopic ellipsometry, Kasic et al. [339] arrived at a value of m ¼ 0.14m0 in an MBE-grown InN layer having an electron concentration of n ¼ 2.8 · 1019 cm3. Using surface reflection of extrinsic semiconductors in the infrared region by the free-carrier plasma, used earlier for GaN [340], Wu et al. [341] determined the effective mass of the free carriers to be 0.07m0 at the bottom of the conduction band. This method hinges on the knowledge of the plasma frequency, electron concentration, and optical dielectric constant through the relation ne2 m ¼ ; ð1:23Þ ee¥ w2P where all the terms have their usual meanings and wP represents the plasma frequency. Using e¥ ¼ 6:7 (compares with figures in the range of 5.8–9.3 tabulated in Table 1.21) and utilizing infrared reflection measurements for a series of samples with different carrier concentrations, Wu et al. [341] arrived at an electron effective mass value of 0.07m0 at the bottom of the conduction band. The temperature dependence of the bandgap of InN indicates a bandgap temperature coefficient of [342] ðdE g =dTÞ ¼ 1:8 10 4 eV K 1 :
ð1:24Þ
1.5 Ternary and Quaternary Alloys
Many important GaN-based devices involve heterostructures as the primary means of achieving an improved performance. Ternary alloys of wurtzite polytypes of GaN, AlN, and InN have been obtained in the continuous alloy systems whose direct bandgap ranges from the old value for InN of 1.9 eV (the new value is approximately 0.7 eV according to Ref. [30]) for InN to 6.2 eV for AlN (the new value is approximately 6 eV). For an in-depth understanding of the physical mechanisms that underlie their operations, the properties of these alloys need to be extensively studied. Many of these properties such as the energy bandgap, effective masses of the electrons and holes, and the dielectric constant depend on the alloy composition. Although measured data for these parameters in InGaN and InAlN have been obtained, they are still not very precise (for AlGaN see Ref. [157] and for InGaN see Ref. [343]). Yamasaki et al. [344] have reported on p-InGaN. More research is necessary to confirm
j89
j 1 General Properties of Nitrides
90
these conclusions regarding p-AlN and p-InGaN. AlN and GaN are slightly lattice mismatched (2.4%). It has been noted that, for many devices, only small amounts of AlN are needed in the GaN lattice to provide sufficient carrier and optical field confinements. 1.5.1 AlGaN Alloy
The ternary alloys of wurtzite and zinc blende polytypes of GaN with AlN form a continuous alloy system with a wide range of bandgap and a small change in the lattice constant. An accurate knowledge of the compositional dependence of the barrier as well as material is a requisite in attempts to analyze heterostructures in general and quantum wells (QWs) and superlattices in particular. The barriers can be formed of AlGaN or AlN, and while dependent on the barrier material, the wells can be formed of GaN or AlGaN layers. The compositional dependence of the lattice constant, the direct energy gap, and electrical and CL properties of the AlGaN alloys were measured by Yoshida et al. [345]. A similar investigation followed a few years later [346]. On the structural side, namely the calculated lattice parameter of this alloy, predictions indicate that Vegards law applies [347] (also reviewed in Ref. [17]): ˚ and aAlx Ga1 x N ¼ 3:1986 0:0891x A
˚: c Alx Ga1 x N ¼ 5:2262 0:2323xA ð1:25Þ
By bringing to bear various tools such as HRXRD, the experimental data for various AlGaN support the applicability of Vegards law in that the experi˚ and c Al Ga N ¼ mental data aAlx Ga1 x N ¼ ð3:189 0:002Þ ð0:086 0:004Þx A x 1x ˚ ð5:188 0:003Þ ð0:208 0:005Þx A are within about 2% of those predicted by linear interpolation, Vegards law. However, the bond lengths exhibit a nonlinear behavior, deviating from the virtual crystal approximation. Essentially, the nearest neighbor bond lengths are not as dependent on composition as might beexpected from the virtual crystal approximation. The ensuing investigations to pin down the compositional dependence of the bandgap of this important alloy continued with conflicting results. These are discussed below following the presentation of an empirical expression used to relate the bandgap to composition. The compositional dependence of the principal bandgap of AlxGa1xN can be calculated from the following empirical expression providing that the bowing parameter, b, is known accurately: E g ðxÞ ¼ xE g ðAlNÞ þ ð1 xÞE g ðGaNÞ bxð1 xÞ;
ð1:26Þ
where Eg(GaN) ¼ 3.4 eV, Eg(Al N) ¼ 6.1 eV, x is the AlN molar fraction, and b is the bowing parameter.
1.5 Ternary and Quaternary Alloys
An earlier compilation by Amano et al. [348] already pinpointed the discrepancy in the reported bowing parameters. For example, Yoshida et al. [349] concluded that, as the AlN mole fraction increases, the energy bandgap of AlxGa1xN deviates upward, implying a negative value for the bowing parameter b. This contrasts the data of Wickenden et al. [350] that support a vanishing bowing parameter b. Koide et al. [351] observed that the bowing parameter is positive and that the bandgap of the alloy deviates downward indicating a positive value for the bowing parameter. To determine the bowing parameter accurately, as investigations expanded [352–362] so did the dispersion in the bowing parameters ranging from 0.8 eV (upward bowing) to þ2.6 eV (downward bowing), as compiled by Yun et al. [363]. Much of this spread emanates from the likely dispersion in the quality of AlxGa1xN, thus erroneous determination of its bandgap and to a lesser extent its lattice parameter. Because the genesis of PL transitions could be nonintrinsic, a technique relying on absorption or modulated photoreflectance is more accurate in the determination of bandgap energy. Using X-ray and surface analytical techniques, such as secondary ion mass spectroscopy (SIMS) and Rutherford backscattering (RBS), for determining composition, reflectance, and absorption for bandgap for AlGaN layers spanning the entire compositional range, Yun et al. [363] revisited the bowing parameter. The results of this study, shown in Figure 1.38 in the form of AlGaN bandgap versus the composition, yield a bowing parameter of b ¼ 1.0 eV for the entire range of alloy compositions. In this figure, the solid line represents a least square fit to the data, which in turn are depicted by solid circles. X-ray diffraction peaks generally tend to be wider for alloy compositions around the midway point that is the most likely source of error in determining the bowing parameter. The situation is exacerbated by the fact that the data points near the middle of the compositional range determine the bowing parameter to a much larger extent, as near each of the binary ends the compositional variation approaches the linear line. It is still possible that as the quality of the films improves smaller bowing parameters could result. A bowing parameter as low as 0.7 has been reported. Hall measurements for n-Al0.09Ga0.91N demonstrated a carrier concentration of 5 · 1018 cm3 and a mobility of 35 cm2 V1 s1 at 300 K [364]. This measurement did not reveal any temperature-dependent mobility of n-A10.09Ga0.91N. Other Hall measurements [365] on Mg-doped p-Al0.08Ga0.92N grown by MOVPE, however, addressed the temperature dependence of the mobility [365]. They indicate that the hole mobility decreases with increasing temperature, reaching a value of about 9 cm2 V1 s1 for a doping density of 1.48 · 1019 cm3. This low mobility is ascribed to a high carrier concentration and the intergrain scattering present in the samples. While the lattice constant was studied, it was observed to be almost linearly dependent on the AlN mole fraction in AlGaN. Until recently, the resistivity of unintentionally doped AlGaN was believed to increase so rapidly with increasing AlN mole fraction that AlGaN became almost insulating for AlN mole fractions exceeding 20%. As the AlN mole fraction increased from 0 to 30%, the n-type carrier concentration dropped from 1020 to 1017 cm3 and the mobility increased from 10 to 30 cm2 V1 s1. An increase in
j91
j 1 General Properties of Nitrides
92
6.5
Bangap of AIGaN (eV)
6.0
5.5
5.0
b=0
b=1
4.5
4.0
3.5 0
0.2
0.4
0.6
0.8
1
Al composition (x) Figure 1.38 Experimental data of energy bandgap of AlGaN (0 x 1), plotted as a function of Al composition (solid circle), and the least squares fit (solid line) giving a bowing parameter of b ¼ 1.0 eV. The dashed line shows the case of zero bowing. As the quality of the near 50 : 50 alloy layers get better, giving rise to sharper X-ray and PR data, smaller bowing parameters may ensue. Bowing parameters as low as 0.7 have been reported.
the native defect ionization energies with increasing AlN may possibly be responsible for this variation. Our knowledge of the doping characteristics of AlGaN is still incomplete. For example, it is not known how the dopant atoms such as Si and Mg respond to the variation of the AlN mole fraction in AlGaN. However, it was suggested that as the AlN mole fraction increases, the dopant atom moves deeper into the forbidden energy bandgap. AlGaN with an AlN mole fraction as high as 50–60% may be doped by both n- and p-type impurity atoms. The ability to dope a high mole fraction AlGaN, especially when low-resistivity p-type material is required, is important because it may otherwise restrict the overall characteristics of devices such as laser diodes. A low AlN mole fraction in AlGaN has been considered sufficient for acceptable optical field confinement. However, this must be addressed before the potential of AlGaN with respect to the other wide bandgap semiconductors is fully realized. 1.5.2 InGaN Alloy
The growth of high-quality InN and an enumeration of its fundamental physical properties remain somewhat elusive as compared to the other alloy, AlGaN.
1.5 Ternary and Quaternary Alloys
Notwithstanding the difficulties in technology, InGaN is already an integral part of important device designs. InxGa1xN (x is the InN mole fraction) is not any less important than AlxGa1xN for the fabrication of electrical and optical devices, such as LEDs and lasers, which can emit in the violet or blue wavelength range. It can be a promising strained QW material for these devices, but added complexities such as the phase separation and other inhomogeneities make the determination of the bandgap of InGaN versus the composition a very difficult task. The first growth of single crystalline InGaN by MOVPE was realized by Nagatomo et al. in 1989 [366] and Matsuoka et al. [367], followed by Yoshimoto et al. in 1991 [368]. Since then, considerable effort has been expended worldwide on this material, as it is responsible for emission in the near-UV, violet, blue, and green colors of the optical spectrum. High-efficiency blue and green LEDs utilizing InGaN active layers are commercially available. This material, however, is not as easy to grow because of the high vapor pressure of N on In and also mismatch between the large In atom and the small N atom. To mitigate this problem, V/III ratios in excess of 20 000, increasing with InN mole fraction, as well as reduced growth temperatures are employed. Matsuoka et al. [369] discovered that lowering the growth temperature to 500 C from nominal temperatures such as 800 C increased the In content in the layers, but at the expense of reduced quality. Efforts to increase the In concentration by raising the indium precursor temperature or the carrier gas flow rate resulted in the degradation of the structural and surface morphology so much that In droplets were formed on the surface [370]. The great disparity between Ga and In could lead to issues such as phase separation and instabilities. In this vein, Ho and Stringfellow [371] investigated the temperature dependence of the binodal and spinodal boundaries in the InGaN system with a modified valence force field model. The calculation of the extent of the miscibility gap yielded an equilibrium InN mole fraction in GaN of less than 6% at 800 C [371]. In the annealing experiments in argon ambient, the phase separation in an InxGa1xN alloy with x 0.1 was observed at temperatures between 600 and 700 C [372], pointing to the large region of solid immiscibility of these alloys. However, under nonequilibrium growth conditions, InxGa1xN layers were grown in the entire range of compositions. But, the decomposition into two phases upon annealing of the InxGa1xN alloys (x ¼ 0.11 and x ¼ 0.29) at 600 and 700 C was observed pointing to the existence of the miscibility gap. For some alloys with x ¼ 0.6, the phase separation could not be observed at 600 C. Above 800 C, the alloy samples with x ¼ 0.1 actively evaporated from the substrate. These results suggest that the solid solutions are grown in metastable conditions and decomposed under annealing conditions. Koukitu et al. [373] performed a thermodynamical analysis of InGaN alloys grown by MOCVD. They found that in contrast to other III–III–V alloy systems where the solid composition is a linear function of the molar ratio of the group III metalorganic precursors at constant partial pressure of group V gas, the solid composition of InGaN deviates significantly from a linear function at high substrate temperatures. Kawaguchi et al. [374] reported a so-called InGaN composition pulling effect in which the indium fraction is smaller during the initial stages of growth but increases with increasing growth thickness. This observation was to a first extent independent
j93
j 1 General Properties of Nitrides
94
of the underlying layer, GaN or AlGaN. The authors suggested that this effect is caused by strain caused by the lattice mismatch at the interface. They found that a larger lattice mismatch between InGaN and the bottom epitaxial layers was accompanied by a larger change in the In content. What one can glean from this is that the indium distribution mechanism in InGaN alloy is caused by the lattice deformation because of the lattice mismatch. With increasing thickness, the lattice strain is relaxed owing to the formation of structural defects and roughness, which weakens the composition pulling effect. Other substrates have also been used for InGaN growth. It has been reported that the crystalline quality of InGaN is superior when grown with the composition that lattice matches ZnO substrate to that grown directly on (0 0 0 1) sapphire substrate [368,369]. In the same investigations, it was observed that InGaN films grown on sapphire substrates using GaN as buffer layers exhibited much better optical properties than InGaN films grown directly on sapphire substrates [375]. For a given set of growth conditions, an increase of InN in InGaN can be achieved by reducing the hydrogen flow [376]. As in the case of AlGaN, the calculated lattice parameter of this alloy follows Vegards law [347] (also reviewed in Ref. [17]): ˚ aInx Ga1 x N ¼ 3:1986 þ 0:3862x A
and
˚ c Inx Ga1 x N ¼ 5:2262 þ 0:574x A: ð1:27Þ
By bringing to bear various tools such as HRXRD, the experimental data for various AlGaN support the applicability of Vegards law in that the experimental data ˚ and c Al Ga N ¼ ð5:195 0:002Þþ aAlx Ga1x N ¼ ð3:560 0:019Þþð0:449 0:019Þx A x 1x ˚ are within about 2% of that predicted by linear interpolation, ð0:512 0:006Þx A Vegards law. As in the case of AlGaN but to a larger extent, the bond lengths exhibit a nonlinear behavior, deviating from the virtual crystal approximation. Essentially, the nearest neighbor bond lengths are not as dependent on composition as might be expected from the virtual crystal approximation. The compositional dependence of InGaN bandgap is a crucial parameter in designs of any heterostructure utilizing it. As such, the topic has attracted a number of theoretical [377–382] and experimental (to be discussed below) investigations and reports. Similar to the case of AlGaN, the energy bandgap of InxGa1xN over 0 x 1 can be expressed by the following empirical expression: g
g
g
E Inx Ga1 x N ¼ xE InN þ ð1 xÞE GaN bInGaN xð1 xÞ ¼ 0:7x þ 3:4ð1 xÞ bInGaN xð1 xÞ eV; g
g
ð1:28Þ
where E GaN ¼ 3:40 eV and E InN ¼ 1:9 eV or near 0.7 eV. For the nomenclature GaxIn1xN, the terms x and 1 x in Equation 1.28 must be interchanged. Another point of caution is that the sign in front of the bowing parameter is changed to positive in some reports.When a comparison ismade, the sign of theb parametermust bechanged. An earlier investigation of InGaN bowing parameter for alloys with small concentrations of InN by Nakamura et al. [383] led to a bowing parameter of 1.0, which is in disagreement with the value of 3.2 reported by Amano et al. [348], who took into
1.5 Ternary and Quaternary Alloys
consideration the strain and piezoelectric fields as well. It should be mentioned that these reports dealt with the Ga-rich side of the alloy. To obtain a fit over a large range of compositions, a composition-dependent bowing parameter has been suggested. As the InGaN is grown on GaN, there are many complicating factors, such as the piezoelectric effect and the nonuniform strain; the impact of the former can be made negligible by growing thick films. Moreover, compositional inhomogeneities due to partial phase separation are present. If the strain caused by the lattice mismatch were uniform, it would be compressive due to the InN lattice constant being 11% larger than that of GaN with an accompanying blue shift of the band edge. Herein lies the dilemma faced by the experimentalists. Growing thick films could minimize the extent of strain and the piezoelectric effect. However, this is nearly an intractable proposition. The relaxation value of the lattice constant and the origin of the optical transitions must be known accurately to determine the bandgap versus composition dependence. Absorption and/or reflection measurements, provided that the absorption edge is sharp, are more useful in determining the bandgap but again require thick and/or high-quality films. Detailed X-ray reciprocal-space mapping undertaken by Amano et al. [348] purportedly indicated that InGaN wells and even somewhat thicker InGaN layers grown on GaN buffer layers are coherently strained; a conclusion reached by the observation that the in-plane lattice constants of GaN and InGaN match. At the same time, though, the layer thicknesses well exceeded the calculated critical values. When a bandgap of 1.9 eV for InN is assumed as the end point value for InN in regard to InGaN ternary, large and/or more than one bowing parameter is required to fit the compositional dependence of the bandgap energy. For example, a bowing parameter of 2.5 eV was obtained from optical absorption measurements and a value of 4.4 eV was obtained from the position of the emission peaks [384]. Nagatomo et al. [366] noted that the InxGa1xN lattice constant varies linearly with the In mole fraction up to at least x ¼ 0.42, but it violates the Vegards law for x > 0.42, which may be caused by erroneous determination of the composition and illustrates well the problem at hand. Even additional investigations did not agree on the exact value of the bowing parameter. For example, a value for bInGaN ¼ 3.9 0.5 eV was reported when 0.9 eV was used for the InN gap, but the bowing parameter had to be increased to 5.1 0.4 eV when 1.9 eV was used for the InN gap [385]. Using the bandgap determined by PL, a bowing parameter of 4.5 eV was also reported [386]. However, when reflectivity measurements together with PL data corrected for Stokes shift were used, bInGaN ¼ 2.5 0.7 eV was obtained for 0.9 eV bandgap of InN and bInGaN ¼ 3.5 0.7 eV for 1.9 eV InN bandgap. Optical transmission measurements led to a bowing parameter of 8.4 eV [387]. At least one theoretical effort resulted in a bowing parameter of 1.2 eV [377]. In fact, linear bandgap dependence on composition with a slope of 3.57 eV for up to 25% InN content has also been reported [388]. Linear dependence with a slope of 4.1 eV for InN mole fraction, x < 0.12, has been reported in another publication as well [389]. Wu et al. [390] visited the bandgap dependence of InGaN on composition by considering 0.8 eV for the bandgap of the end binaryInN. Figure1.39 shows the composition dependence of the bandgap of InGaN, determined by using photomodulated transmission [391] and optical absorption [392] measurements, as a
j95
j 1 General Properties of Nitrides
96
Lattice constant (Å) 3 .2
Energy gap (eV)
4
3.3
3 .4
3 .5
3 .6
abs. 300 K PL PT Old data Calorimetric
GaN
3 2
InN
1 0 0.0
0.2
0.4
0.6
0.8
1.0
Composition (x) Figure 1.39 PL peak energy and bandgap of InGaN determined by optical absorption as a function of composition, as compiled in Ref. [390], including previously reported data for InN. The solid curve shows the fit to the bandgap energies (determined by absorption and phototransmission) using a bowing parameter b ¼ 1.43 eV [390].
function of GaN fraction. The data near the GaN binary end include those reported by Pereria et al. [392], Shan et al. [390], and ODonnel et al. [393]. Care was taken by observing the dependence of squared absorption coefficient on light probe energy and seeing nearly a linear dependence to gain confidence on the measured bandgap and also confirming the values by bandgaps determined by photomodulated transmission measurements. The slight deviation from linearity near the InN end of the ternary has been attributed to the nonparabolicity of the conduction band caused by the k p interaction between the G6 symmetry conduction band and the G8 symmetry valence bands. As shown by the solid curve in Figure 1.39, the compositional dependence of the bandgap in the entire composition range can be well fit by a bowing parameter of b ¼ 1.43 eV. Shown in Figure 1.39 with dashed line is the fit to the empirical expression using energy of 1.9 eV for InN and bowing parameter of 2.63 eV to demonstrate that it does represent the Ga-rich side of the compositions well. However, the bowing parameter of 1.43 eV that is good for the entire compositional range is the one that utilizes 0.77 eV for the bandgap of InN. In an investigation with a different set of objectives, Yoshimoto et al. [368] studied the effect of growth conditions on the carrier concentration and transport properties of InxGa1xN. They observed that if the deposition temperature is increased from 500 to 900 C, InxGa1xN grown on sapphire with x 0.2 suffers from a reduction in carrier concentration from 1020 to 1018 cm3, but gains from an increase in the carrier mobility from less than 10 to 100 cm2 V1 s1. The same group later noted that this trend does not change if the films are grown on ZnO substrates instead of sapphire [369]. They could achieve good InGaN material with In mole fractions as
1.5 Ternary and Quaternary Alloys
large as 23%. Nakamura and Mukai [394] discovered that the film quality of InxGa1xN could be significantly improved if these films are grown on high-quality GaN films. Thus, from the reports cited above it may be concluded that the major challenge for obtaining high-mobility InGaN is to find a compromise in the growth temperature, because InN is unstable at typical GaN deposition temperatures. This growth temperature would undoubtedly be a function of the dopant atoms, as well as the method (MBE, OMVPE, etc.) used for the growth. This is evident from a study by Nakamura et al., who have since expanded the study of InGaN employing Si [395] and Cd [396] as dopants. A review of various transport properties of GaInN and AlInN by Bryden and Kistenmacher [296] is available but predates the bandgap reconsideration of InN; the growth and mobility of p-GaInN is discussed by Yamasaki et al. [344]. 1.5.3 InAlN Alloy
In1xAlxN is an important compound that can provide a lattice-matched barrier to GaN, low fraction AlGaN, and InGaN, and consequently lattice-matched AlInN/ AlGaN or AlInN/InGaN heterostructures. Although there was some discrepancy as to which composition really lattice-matched GaN, continued improvement in layer quality and persistence narrowed the In composition for matching. Compositions In0.29Al0.71N and In0.17Al0.83N have been reported as matching, but the value around the latter composition is gaining more acceptance [397]. The growth and electrical properties of this semiconductor have not yet been as extensively studied compared to the other two ternaries, particularly AlGaN, as the growth of this ternary is also challenging because of diverse thermal stability, lattice constant, and cohesive energy of AlN and InN. Moreover, thermal instability resulting from, for example, the spinodal phase separation phenomenon, which is more of an issue in Al1xInxN than in InyGa1yN [398], must be considered. Despite the above-mentioned difficulties, lattice matching and the lack of crack formation when AlGaN is replaced with InAlN in distributed Bragg reflectors (DBRs) and other structures requiring relatively thicker layers are more than enough to pursue this material. In fact, light emitters, field effect transistors, and DBRs, as mentioned, using InAlN barriers as opposed to AlGaN are gaining considerable momentum. As in the case of AlGaN and InGaN, the calculated lattice parameter of this alloy follows Vegards law [347] (also reviewed in Ref. [17]) as ˚ aAlx In1 x N ¼ 3:58480 4753x A
˚ and c Alx In1 x N ¼ 5:8002 0:8063x A: ð1:29Þ
By utilizing various tools such as HRXRD, the experimental data for various AlInN layers support the applicability of Vegards law in that the experimental data ˚ and c ¼ ð5:713 0:014Þ aAlx In1 x N ¼ ð3:560 0:019Þ ð0:449 0:019Þx A ˚ ð0:745 0:024Þx A are within about 2% of that predicted by linear interpolation, the Vegard law. As in the case of AlGaN and InGaN, the bond lengths exhibit a
j97
j 1 General Properties of Nitrides
98
nonlinear behavior, deviating from the virtual crystal approximation. Essentially, the nearest neighbor bond lengths are not as dependent on composition as might be expected from the virtual crystal approximation. Early experimental results [288] for the bandgap of In0.29Al0.71N, which was thought lattice matched to GaN, indicate that this alloy has an energy gap of 3.34 eV at low temperatures (the room-temperature value is actually closer to 4.5 eV) that is even below that for GaN. The estimations by Wright and Nelson [399] that followed pointed to a bandgap of about 5 eV for the zinc blende variety. The accompanying bowing parameter reported by Wright and Nelson is 2.53 eV at the time when the larger InN bandgap was accepted. Naturally, when the bandgap is in question the bowing parameter is even more in question. As in the case of AlGaN and InGaN, the compositional dependence of the bandgap of AlInN can be expressed with the following empirical expression using a bowing parameter, bAlInN, as g
g
g
E Alx InN ¼ xE AlN þ ð1 xÞE InN bAlInN xð1 xÞ ¼ 6:1x þ 0:7ð1 xÞ bAlInN xð1 xÞ eV:
ð1:30Þ
In addition to the aforementioned calculations, experimental data for the bowing parameter, b, exist. Using a bandgap of 6.2 eV for AlN (the new figure is closer to 6 eV), the values that have been reported encompass b ¼ 3.1 eV deduced by fitting the bandgap of this alloy determined by PL [400], b ¼ 2.384 eV by absorption measurements but by using 2.0 eV for the bandgap of InN and 5.9 eV for AlN [401], and b ¼ 5.4 eV in a review where 1.95 eV was used for InN bandgap [17]. Scaling the bandgap of AlN to about 6.0 eV would reduce the bowing parameter a little. Despite the scattered data, reasonably useful bandgap variation of AlInN with composition can be obtained as shown in Figure 1.40. Kim et al. [404] deposited thin AlInN films with X-ray rocking curve FWHM values between 10 and 20 arcmin. They observed an increase of In content in AlInN of up to 8% by lowering the substrate temperature to 600 C. A further reduction of substrate temperature during OMVPE is not useful because of the needed efficient pyrolysis of ammonia. Yamaguchi et al. [405] also reported on OMVPE growth of AlInN on GaN templates that were in turn deposited on low-temperature AlN nucleation layers on cplane sapphire. In macroscopic sense, the alloys grown were not phase separated and the bandgap variation followed the compositional variations in the InN composition range of 19–44%. From the square of the absorption coefficient versus E–Eg, the bandgap of the alloy was determined. Starosta [406] and later Kubota et al. [407] grew InAlN alloy by radio frequency (RF) sputtering. Kistenmacher et al. [408], however, used the RF magnetron sputtering (RF MS) from a composite metal target to grow InAlN at 300 C. It was observed that the energy bandgap E of this semiconductor varies between 2.0 eV (this is supposed to represent the InN binary end point, which assumes the old and incorrect value) and 6.20 eV (this too represents the old value for the bandgap of AlN with the new figure being approximately 6 eV) for x between 0 and 1 [407]. The carrier concentration and the mobility of In1xAlxN for x ¼ 0.04 were 2 · 1020 cm3 and 35 cm2 V1 s1, respectively, and for x ¼ 0.25 were 8 · 1019 cm3 and 2 cm2 V1 s1,
1.5 Ternary and Quaternary Alloys
Lattice constant (Å) 7 a b c d e
AlN
6
Energy gap (eV)
5
f g h i j k l
4 3 2
InN
1 Eg=6(1–x)+0.7x–3.1x(1–x)
0 0.0
0.2
0.4
0.6
0.8
1.0
Composition (x) Figure 1.40 Dependence of bandgap of the InAlN alloy on composition. Unless otherwise stated, the measurement temperature is room temperature. The solid line between the 0.8 eV gap of InN and 6 eV of AlN is deemed as being reasonably accurate. (a) Absorption; (b) RT PL;
(c) RT absorption [404]; (d) Ref. [401]; (e) absorption, poly [416]; (f) absorption [402]; (g) a theory [403]; (h) RT PL and CL [400]; (i) 8 K optical reflection [400]; (j) RT absorption [405]; (k) RT PL [405]; (l) fit to Eg ¼ 6(1 x) þ 0.7 x 3.1 x (1 x). In part courtesy of Wladek Walukiewicz.
respectively [296]. Thus, the mobility decreases substantially with an increase in the Al mole fraction because the structure of the InAlN approaches the structure of the insulating AlN. 1.5.4 InAlGaN Quaternary Alloy
By alloying InN together with GaN and AlN, the bandgap of the resulting alloy(s) can be increased from 1.9 eV (or near 0.7 eV if we use the updated InN bandgap) to a value of 6.2 eV (or 6 eV if we use the updated value), which is critical for making highefficiency visible light sources and detectors. In addition, the bandgap of this quaternary can be changed while keeping the lattice constant matched to GaN [409,410]. In quaternary alloys of nitrides, the N atoms constitute anion sublattice, whereas group III elements (In, Ga, Al) constitute the cation sublattice. Use of this quaternary material allows almost independent control of the bandgap and thus the band offset in AlInGaN-based heterostructures. However, among other difficulties brought about by the four-component system, the optimal growth temperature is important to optimize and control, as aluminum-based compounds generally require higher growth temperatures and In-based ones require lower
j99
j 1 General Properties of Nitrides
100
temperatures. Higher temperatures are also desirable for reducing the O incorporation in the growing film as oxides of Ga and In desorb from the surface. The growth temperature will therefore govern the limits of In and Al incorporation into the AlGaInN quaternary alloy [409]. The quaternary alloy (Ga1xAlx)In1yN is expected to exist in the entire composition range 0 < x < 1 and 0 < y < 1. Unfortunately, as in the case of the InGaN alloy, incorporation of indium in these quaternary alloys is not easy. To prevent InN dissociation, InGaN crystals were originally grown at low temperatures (about 500 C) [411], which also applies to InGaAlN. The use of a high nitrogen flux rate allowed the high-temperature (800 C) growth of high-quality InGaN and InGaAlN films on (0 0 0 1) sapphire substrates. Note that the incorporation of indium into InGaN film is strongly dependent on the flow rate, N/III ratio, and growth temperature in an OMVPE environment. The incorporation efficiency of indium decreases with increasing growth temperatures. Observations made in the case of InGaN should be applicable to In incorporation in quaternary nitrides. Ryu et al. [412] reported on the optical emission in this quaternary system and AlInGaN/AlInGaN multiple quantum wells grown by pulsed metalorganic chemical vapor deposition. Strong blue shift with excitation intensity was observed in both the quaternary layers and quantum wells that was attributed to localization. This would imply the inhomogeneous nature of the structures and/or presence of band tail states indicative of early stages of material development and/or serious technological problems involved. The relationships between composition and bandgap (or lattice constant) can be predicted by the equation below, which was originally developed for the InGaAsP system [413]. Qðx; y; zÞ ¼
xyT 12 ðð1 x þ yÞ=2Þ þ yzT 23 ðð1 y þ zÞ=2Þ þ zxT 31 ðð1 z þ xÞ=2Þ ; xy þ yz þ zx
T ij ðaÞ ¼ aBj þ ð1 aÞBi þ að1 aÞbij : The parameters x, y, and z represent the composition of GaN, InN, and AlN. If GaN, InN, and AlN are represented by 1, 2, and 3, T12 would represent GaxInyN. Further, the term T12 can be expressed as T 12 ðaÞ ¼ aB2 þ ð1 aÞB1 þ að1 aÞb12 , where b12 is the bowing parameter for the GaxInyN alloy and a ¼ ð1 x þ yÞ=2 or ð1 x þ yÞ=2 or ð1 z þ xÞ=2 is the effective molar fraction for GaInN, InAlN, and AlGaN, respectively, B2 the bandgap of InN, and B1 is the bandgap of GaN. Similar expressions can be constructed for T23 and T31 by appropriate permutations. An alternative approach is discussed in conjunction with Equation 1.31. The results of these calculations for the bandgap and lattice constant dependence on composition are shown in the three-dimensional diagrams of Figures 1.41–1.43. An empirical expression similar to that used for the ternaries can also be constructed for the quaternary as g
g
g
g
E Alx Iny Ga1 x y N ¼ xE AlN þ yE InN þ ð1 x yÞE GaN bAlGaN xð1 xÞ bInGaN yð1 yÞ;
ð1:31Þ
1.5 Ternary and Quaternary Alloys
Figure 1.41 Bandgap versus composition for quaternary AlxInyGa1xyN (assumed InN bandgap ¼ 0.8 eV). (Please find a color version of this figure on the color tables.)
Figure 1.42 Bandgap versus composition for quaternary AlxInyGa1xyN (assumed InN bandgap ¼ 1.9 eV). (Please find a color version of this figure on the color tables.)
j101
j 1 General Properties of Nitrides
102
Figure 1.43 Lattice constant a versus composition for quaternary AlxInyGa1xyN.
where the first three parameters on the right-hand side of the equation are contributions by the binaries to the extent of their presence in the lattice, the third term represents the bowing contribution related to Al, and the last term depicts the bowing contribution due to In. The bowing parameters, bAlGaN and bInGaN, indicated in Equation 1.31 are the same as those discussed in conjunction with InGaN and AlInN. As such, the values are the same. The parameters x, y, and z represent the molar fraction of binaries in the quaternary. After discussing all three ternary alloys of the nitride semiconductor family, the bandgap (both in terms of energy and also corresponding air wavelength) versus the lattice parameter is shown in Figure 1.44 for convenience. The discussion of alloys individually up to this point paves the way to a collective discussion of alloys in terms of structural parameters for a rapid observation of trends. This discussion would be of special value particularly for the least discussed of alloys, InAlN. Let us first discuss the structural properties such as the lattice constants and bond lengths, and angles of nitride semiconductor alloys, following the discussion in Sections 1.5.1–1.5.3 and that surrounding Figure 1.8, Equation 1.3, and Table 1.2. Following Ref. [17], the lattice parameter calculated using Equation 1.25 (for AlGaN), Equation 1.27 (for InGaN), and Equation 1.29 (for InAlN) can be used to calculate the lattice constants for the three ternaries for all compositions and compared with experiments for AlGaN [414], InGaN [415], and InAlN [416], as shown in Figure 1.45.
1.5 Ternary and Quaternary Alloys
Figure 1.44 The bandgap versus the lattice parameter for AlGaN, InGaN, and InAlN using bowing parameters in the same order, 1, 1.43, and 3.1 eV, and bandgap values of 6 eV for AlN, 3.4 eV for GaN, and 0.8 eV for InN. The lattice constants used for the binary AlN, GaN, and InN are 3.11, 3.199, and 3.585 Å, respectively.
Following the case for the binaries tabulated in Table 1.2 and discussed from a theoretical point of view in Refs [347,417], and the experimental points of view in Refs [414] (for AlGaN), [415] (for InGaN), and [416] (for AlxIn1xN), the cell parameter, u, has been calculated for randomly distributed A0.5B0.5N (here A and B represent the metal components forming the alloy) alloys by the theoretical approach of Ref. [418], the pertinent parts of which are succinctly discussed in Section 1.1. The internal cell parameter can be approximately expressed by the quadratic equation uAx B1 x N ¼ xuAN þ ð1 xÞuBN bAB xð1 xÞ;
ð1:32Þ
where bAB is the bowing parameter defined as bAB ¼ 2Y AN þ 2Y BN 4Y A0:5 B0:5 N :
ð1:33Þ
j103
j 1 General Properties of Nitrides
104
3.8
(a)
Lattice constant, a (x) (Å)
3.6
Alx In1–xN Inx Ga1–xN
3.4
Alx Ga1–x N
3.2
Theory (T = 0 K) Experiment (T = 300 K) 3.0 0
0.2
0.6 0.4 Molar fraction, x
0.8
6.0
1.0
(b)
5.8
Alx In1–xN
Lattice constant c(x) (Å)
Inx Ga1–xN 5.6
5.4
Alx Ga1–x N
5.2
5.0 Theory (T = 0 K) Experiment (T = 300 K) 4.8 0
0.2
0.4
0.6
0.8
1.0
Molar fraction, x Figure 1.45 (a) The a(x) lattice parameter and (b) the c(x) lattice parameter for random ternary alloys of AlxGa1xN, InxGa1xN, and AlxIn1xN as measured by HRXRD at room temperature (solid lines) and the calculated values using Equation 1.25 (for AlGaN) and Equation 1.27 (for
InGaN) for T ¼ 0 K (dashed lines). The agreement between calculations and measured lattice constants is better than 2% over the entire range of compositions, compiled in Ref. [17] utilizing Refs [414,415]. Courtesy of O. Ambacher.
1.5 Ternary and Quaternary Alloys
The internal cell parameters for each of the three alloys then are uAlx Ga1 x N ¼ 0:3819x þ 0:3772ð1 xÞ 0:0032xð1 xÞ; uInx Ga1 x N ¼ 0:3793x þ 0:3772ð1 xÞ 0:0057xð1 xÞ; uAlx In1 x N ¼ 0:3819x þ 0:3793ð1 xÞ 0:0086xð1 xÞ:
ð1:34Þ
The structural and other polarization related parameters of ternaries do not follow a linear relationship of the composition, as discussed in detail in Section 2.7. The nonlinearity in question for an alloy, AxB1xN, where A and B represent the metal components, is approximated by quadratic equations of the form [418] Y Ax B1 x N ¼ xY AN þ ð1 xÞY BN bAB xð1 xÞ;
ð1:35Þ
where Y represents any parameter, namely the lattice constant, the u parameter or polarization, and the bowing parameter is defined in Equation 1.33. As in the case of binaries discussed in Section 1.1, the cell parameter, u, and the c/a ratio do not follow the ideal crystal values for the three ternaries of nitride semiconductors. They are shown for the three ternaries for varying composition in Figure 1.46. Similar to the binaries, tabulated in Table 1.2 in conjunction with Figure 1.8, the aforementioned two parameters, the nearest and the second neighbor bond lengths, as well as the bond angles have been calculated for the three ternaries and those associated with 50% alloys are tabulated in Table 1.34. As displayed in Figure 1.46, the cell internal parameter increases as one goes from GaN to InN and, more significantly, to AlN. The nonlinear dependencies on the alloy composition are described by a bowing parameter, bAB, whose values are 0.0032, 0.0057, and 0.0086 for AlxGa1xN, InxGa1xN, and AlxIn1xN, respectively. The bowing parameter increases from AlxGa1xN to InxGa1xN, and continues on to AlxIn1xN. It is worth noting that the bowing parameter is negative for all the three ternaries, the average cell internal parameter of the same alloys is always above the ideal value of 0.375. If the lattice constants scale linearly with the alloy composition but the internal parameter does not, the bond angles and/or the bond lengths of the real and the virtual crystal must depend nonlinearly on the alloy composition. The average nearest neighbor bond lengths (b and b1, see Figure 1.8 for a graphical description) and bond angles (see Figure 1.8 for a graphical description) calculated by using Equations 1.2–1.4 are shown in Figure 1.47a and b and listed in Tables 1.2 and 1.34. The average cation–anion distances to the nearest and second nearest neighbors scale nearly linearly with alloy composition for AlxGa1xN, InxGa1xN, and AlxIn1xN. The average bond length along the c-axis is 0.7–0.9% longer than the nearest neighbor bonds in the direction of the basal plane (Figure 1.47a). 1.5.5 Dilute GaAs(N)
When small amounts of N and As are incorporated into GaAs and GaN lattices, respectively, a large negative bandgap bowing parameter results. Consequently, with
j105
j 1 General Properties of Nitrides
106
very small amounts of N in the GaAs lattice, its bandgap can be made very small, to a point where 1.3 mm lasers and 1.5 mm lasers if In and Sb are also added to the lattice can all be built with GaAs technology. Anomalously large bandgap bowing parameters exhibited by GaAsN and GaNAs are caused by large chemical and size 1.64 (a)
Ideal
1.63
c(x)/a(x) ratio
1.62
Inx Ga1–xN
Alx Ga1–xN
1.61
Alx In1–xN
1.60 Theory (T = 0 K) Experiment (T = 300 K)
1.59 0
0.2
0.4
1.0
0.8
0.6
molar fraction, x 0.383 (b)
cell-internal parameter u(x)
0.381
Alx Ga1–x N
b = –0.0032
0.379
Alx In1–xN
b = 0.0086
Inx Ga1–xN
0.377
b = –0.0057
Ideal
0.375
0
0.2
0.6 0.4 molar fraction, x
0.8
1.0
1.5 Ternary and Quaternary Alloys Table 1.34 Calculated cell internal parameter, a lattice parameter,
c/a ratio, cation–anion distance between the nearest and second nearest neighbors, and bond angles (given in degrees) for the three ternary random alloys in the virtual crystal limit with a composition of 50%.
u a (Å) c/a b (Å) b1 (Å) 0 b1 (Å) 0 b2 (Å) 0 b3 (Å) a b
Al0.5Ga0.5N
In0.5Ga0.5N
In0.5Al0.5N
0.379 3.154 1.620 1.935 1.924 3.175 3.701 3.694 108.80 110.14
0.377 3.392 1.625 2.078 2.073 3.436 3.977 3.975 109.13 109.81
0.378 3.347 1.612 2.042 2.041 3.354 3.921 3.920 108.76 110.18
The distance is in Å and the angles are in degrees [17].
~
differences between As and N [419–422]. Dependence of the bandgap energy in GaAsN and InPN on nitrogen content is shown in Figure 1.48. To a first extent, the dashed lines originating from both GaN end (in which case small amounts of As are added to GaN) and GaAs end (in which case small amounts of N are added to GaAs) represent the bandgap dependence of GaNAs. However, one must keep in mind that for both GaAsN and InPN the simple treatment behind the aforementioned statement fails and that the arrows shown in the figure indicate the boundaries of the regions where the gap dependence on composition may be predicted with any accuracy. Also shown is the bandgap variation with composition for other commonly used ternaries. The thicker vertical line through GaAs represents the bandgap attainable with GaInAsN, at least in theory, while maintaining lattice matching to GaAs. The decrease in the lattice constant caused by N can be compensated with In added to the lattice. The potential of covering a large range of bandgap energies on GaAs substrates has attracted a great deal of interest in this material system. In fact, the first laser containing N was an InGaAs(N) active layer one. Owing in part to
Figure 1.46 (a) The c/a ratio for the three random ternary alloys determined by HRXRD at room temperature (solid lines) and calculated using Equation 1.24 for T ¼ 0 K (dashed lines). The measured and calculated data confirm that the c/a ratios of Wz InGaN, AlGaN, and AlInN crystals are always less than the value of 1.633 for ideal hexagonal crystal. (b) The cell internal parameter, u, for three random AlGaN, InGaN, and AlInN alloys calculated using the quadratic
Equation 1.35. The nonlinearity of the internal cell internal parameter in its compositional dependence can be described by a negative bowing parameter b. This bowing parameter is 0.0032, 0.0057, and 0.0086 for AlxGa1xN, InxGa1xN, and AlxIn1xN, respectively, as indicated in the figure as well. The u parameter of the ternaries is always larger than 0.375 that is the value for an ideal hexagonal crystal [17]. Courtesy of O. Ambacher.
j107
j 1 General Properties of Nitrides
108
(a) M-Nc1
InN
2.2
InN
Metal-N average bond length (A)
M-Nb1
2.1 Inx Ga1–xN Alx In1–xN
2.0 Alx Ga1–xN
M-Nc1
GaN b=M-Nc1 b1= M-Nb1
1.9
0
0.2
0.4
M-Nb1 AlN 0.6
0.8
1.0
Molar fraction, x
111
T=0K
(b)
AlN
Bond angle of virtual lattice (deg)
InN
Alx In1-x N
110
InN
Alx Ga1-x N GaN
Inx Ga1-x N
ideal: α = β =109.47 ο
Inx Ga1-x N 109
β α
Alx Ga1-x N Alx In1-x N
AlN 108
0
0.2
0.8 0.6 0.4 Molar fraction, x
Figure 1.47 (a) The compositional dependence of the average nearest neighbor bond lengths, b and b1 (see Figure 1.8 for a graphical description) in the virtual crystal limit for the metal–nitrogen bonds along the c-axis (solid line) and off c-axis (dashed line). (b) The compositional dependence of the average bond angles a (dashed lines) and b (solid lines) of random
1.0
AlxGa1xN, InxGa1xN, and AlxIn1xN alloys (see Figure 1.8 for a graphical description). Clearly, the average bond angles deviate noticeably from the ideal hexagonal crystal for which a ¼ b ¼ 109.47. Moreover, the deviation increases from GaN to InN and continues onto AlN in a nonlinear fashion [17]. Courtesy of O. Ambacher.
1.5 Ternary and Quaternary Alloys
Γ valley energy gap (eV)
6
AlN
zinc blende T=0K
5 4
AlP AlAs
3
GaP
GaN
2 InN 1
AlSb
GaAs GaAsN InPN
0 4.5
5.0
5.5
InP GaSb InSb InAs 6.0
6.5
Lattice constant (A) Figure 1.48 Direct G valley energy gap as a function of lattice constant for the zinc blende form of 12 III–V binary compound semiconductors (filled circles) and some of their random ternary alloys (lines connecting the solid circles) at zero temperature. The energy gaps for certain ternaries such as AlAsP, InAsN, GaAsN, InPN, and GaPN are extended into regions where
no experimental data have been reported. For GaAsN and InPN, the arrows indicate the boundaries of the regions where the gap dependence on composition may be predicted with any accuracy, patterned after Ref. [423] with necessary changes, particularly the one reflecting the small bandgap of InN.
extreme nonequilibrium conditions employed for growth, MBE is the dominant growth approach for dilute arsenides with nitrogen. The critical issues are compositional control, incorporation of more than a small percentage of N, doping inefficiency, and layer quality. The situation is exacerbated on all fronts when the N concentration is increased for achieving 1.5 mm wavelength of emission. Postgrowth annealing is often employed to improve the crystal quality and/or to increase Si dopant incorporation, however, at the expense of blue shift in the bandgap. While GaAsN is chosen here for the present discussion, there are many other dilute nitride semiconductors as discussed in Section 2.11 in conjunction with band parameters. As alluded to earlier, the chemical and size differences between the N and As atoms are the challenges facing experimentalists. In addition, the generation of atomic nitrogen, although not that different from the technology required for hexagonal GaN growth [424], deserves some attention. While basic mismatch between N and As can be dealt with by growing the layers under nonequilibrium conditions, the issue of atomic nitrogen can be handled by compact RF sources that have seen a good degree of improvement lately. By adjusting the RF power and pressure in the cell, one can tailor the source to produce mostly the atomic species by optimizing the emission at 745 nm of wavelength. Note that the substrate and most of the structure are zinc blende and, consequently, the dilute material assimilates and assumes the same crystalline structure. The desired nitrogen concentrations are in the range of 1–10% for red shifting the transitions out to as long as 1.55 mm. Larger growth rates lead to a reduced incorporation of N in the lattice. Similarly, higher growth temperatures lead to the same. Consequently, when 1.55 mm wavelength material is desired, lower
j109
j 1 General Properties of Nitrides
110
growth rates must be employed as well as lower growth temperatures. At substrate temperatures of 500 C or below, if very large As overpressure is employed, incorporation of N is limited because the flux of atomic nitrogen is small. However, atomic nitrogen is very reactive and, therefore, compositional control should be much better as compared to quaternaries relying on P and As (InGaAsP). As expected, owing to dissimilarities of N and As, the luminescence properties of GaInNAs degrade rapidly with increasing nitrogen concentration. Employing remedies such as postgrowth annealing enhances the luminescence efficiency of GaInNAs. However, this enhancement is accompanied by a blue shift in the transition in bulk and quantum well materials. Nitrogen and possibly In diffusion out of GaInAsN are responsible for the observed luminescence shift to shorter wavelengths. For completeness, a one-paragraph discussion of device issues will be made in conjunction with the edge emitting and vertical cavity lasers operating at 1.3 and 1.5 mm portion of the optical spectrum, although other applications such as heterojunction bipolar transistors are possible. Several groups have reported lasers operating at 1.3 mm region [425–442], where the silica-based fiber dispersion is zero, and 1.5 mm region [443–447] (albeit with addition of Sb to the lattice as the quality required for laser operation for InGaAsN layers cannot be obtained), where the loss is low, again for the silica-based fibers. Both are intended for telecommunication purposes. Even 8 W [448] and 12 W [449] CW operation has been reported. Highspeed testing of these lasers has also been performed [450] with data transmission rates as high as Gbit s1 having been achieved already [451]. For interconnects and high-speed data links, vertical cavity surface emitting lasers (VCSELs) have received a great deal of attention. Now that dilute nitrides are becoming potential candidates for long-wavelength lasers, efforts are under way to explore VCSELs in this material system as well [452,453].
References 1 Lei, T., Fanciulli, M., Molnar, R.J., Moustakas, T.D., Graham, R.J. and Scanlon, J. (1991) Applied Physics Letters, 59, 944. 2 Paisley, M.J., Sitar, Z., Posthill, J.B. and Davis, R.F. (1989) Journal of Vacuum Science & Technology, 7, 701. 3 Powell, R.C., Lee, N.E., Kim, Y.W. and Greene, J.E. (1993) Journal of Applied Physics, 73, 189. 4 Mizita, M., Fujieda, S., Matsumoto, Y. and Kawamura, T. (1986) Japanese Journal of Applied Physics, 25, L945. 5 Xia, Q., Xia, H. and Ruoff, A.L. (1993) Journal of Applied Physics, 73, 8198.
6 Perlin, P., Jauberthie-Carillon, C., Itie, J.P., San Miguel, A., Grzegory, I. and Polian, A. (1992) Physical Review B: Condensed Matter, 45, 83. 7 Ueno, M., Yoshida, M., Onodera, A., Shimommura, O. and Takemura, K. (1994) Physical Review B: Condensed Matter, 49, 14. 8 Pirouz, P. and Yang, J.W. (1993) Ultramicroscopy, 51, 189. 9 Ruterana, P., Sanchez, A.M. and Nouet, G. (2003) Extended defects in wurtzite GaN layers: atomic structure, formation and interaction mechanisms, in Nitride Semiconductors – Handbook on Materials
References
10
11
12
13
14 15
16
17
18
and Devices (eds P. Ruterana, M. Albrecht and J. Neugebauer), Wiley-VCH Verlag GmbH, Weinheim, Germany. Harris, W.A. (1980) Electronic Structure and Properties of Solids, Dover, NY, pp. 174–179. Yeh, C.-Y., Lu, Z.W., Froyen, S. and Zunger, A. (1992) Physical Review B: Condensed Matter, 46, 10086. Bernardini, F., Fiorentini, V. and Vanderbilt, D. (2001) Accurate calculation of polarization-related quantities in semiconductors. Physical Review B: Condensed Matter, 63, 193–201. Bechstedt, F., Großner, U. and Furthm€ uller, J. (2000) Physical Review B: Condensed Matter, 62, 8003. Wei, S.-H. and Zunger, A. (1996) Applied Physics Letters, 69, 2719. Leszczynski, M., Teisseyre, H., Suski, T., Grzegory, I., Bockowski, M., Jun, J., Porowski, S., Pakula, K., Baranowski, J.M., Foxon, C.T. and Cheng, T.S. (1996) Lattice parameters of gallium nitride. Applied Physics Letters, 69, 73. Leszczynski, M. (1999) Common crystal structure of the group III-nitrides, in Properties, Processing and Applications of Gallium Nitride and Related Semiconductors (eds J.H. Edgar, S. Strite, I. Akasaki, H. Amano and C. Wetzel), EMIS Data Review Series, No. 23, INSPEC, The Institution of Electrical Engineers, Stevenage, UK, pp. 3–5. Ambacher, O., Majewski, J., Miskys, C., Link, A., Hermann, M., Eickhoff, M., Stutzmann, M., Bernardini, F., Fiorentini, V., Tilak, V., Schaff, B. and Eastman, L.F. (2002) Pyroelectric properties of Al(In)GaN/GaN heteroand quantum well structures. Journal of Physics: Condensed Matter, 14, 3399–3434. Leszczynski, M., Suski, T., Perlin, P., Teisseyre, H., Grzegory, I., Bockowski, M., Jun, J., Porowski, S., Pakula, K., Baranowski, J.M., Foxon, C.T. and Cheng, T.S. (1996) Applied Physics Letters, 69, 73.
19 Jeffery, G.A., Parry, G.S. and Mozzi, R.L. (1956) Journal of Chemical Physics, 25, 1024. 20 Schulz, H. and Theimann, K.H. (1977) Solid State Communications, 23, 815. 21 Tanaka, M., Nakahata, S., Sogabe, K., Nakata, H. and Tabioka, M. (1997) Japanese Journal of Applied Physics, 36, L1062. 22 Angerer, H., Brunner, D., Freudenberg, F., Ambacher, O., Stutzmann, M., H€opler, R., Metzger, T., Born, E., Dollinger, G., Bergmaier, A., Karsch, S. and K€orner, H.-J. (1997) Applied Physics Letters, 71, 1504. 23 Domagala, J., Leszczynski, M., Prystawko, P., Suski, T., Langer, R., Barski, A. and Bremser, M. (1999) Journal of Alloys and Compounds, 286, 284. 24 Kim, K., Lambrecht, W.R.L. and Segall, B. (1996) Physical Review B: Condensed Matter, 53, 16310. 25 Wright, A.F. and Nelson, J.S. (1995) Physical Review B: Condensed Matter, 51, 7866. 26 Detchprohm, T., Hiramatsu, K., Itoh, K. and Akasaki, I. (1992) Japanese Journal of Applied Physics, 31, L1454. 27 Leszczynski, M., Teisseyre, H., Suski, T., Grzegory, I., Bockowski, M., Jun, J., Porowski, S. and Major, J. (1995) Journal of Physics D: Applied Physics, 69, A149. 28 Deguchi, T., Ichiryu, D., Toshikawa, K., Sekiguchi, K., Sota, T., Matsuo, R., Azuhata, T., Yamaguchi, M., Yagi, T., Chichibu, S. and Nakamura, S. (1999) Journal of Applied Physics, 86, 1860. 29 Paszkowicz, W. (1999) Powder Diffraction, 14, 258. 30 Wu, J., Walukiewicz, W., Yu, K.M., Ager, J.W., III, Haller, E.E., Lu, H., Schaff, W.J., Saito, Y. and Nanishi, Y. (2003) Applied Physics Letters, 80, 3967. 31 Phillips, J.C. (1973) Bonds and Bands in Semiconductors, Academic Press, New York, p. 31. 32 Xu, Y.-N. and Ching, W.Y. (1993) Electronic, optical, and structural properties of some wurtzite crystals.
j111
j 1 General Properties of Nitrides
112
33
34
35
36
37
38
39
40
41
42
Physical Review B: Condensed Matter, 48, 4335–4350. Morkoc, H., Strite, S., Gao, G.B., Lin, M.E., Sverdlov, B. and Burns, M. (1994) Large-band-gap SiC, III–V nitride, and II–VI ZnSe-based semiconductor device technologies. Journal of Applied Physics, 76 (3), 1363–1398. Akasaki, I. and Amano, H. (1994) Properties of Group III Nitrides (ed. J.H. Edgar), EMIS Data Review Series, No. 11, INSPEC, The Institution of Electrical Engineers, Stevenage, UK, pp. 30–34. Christensen, N.E. and Gorczyca, I. (1994) Optical and structural properties of III–V nitrides under pressure. Physical Review B: Condensed Matter, 50, 4397–4415. Levinshtein, M., Rumyantsev, S. and Shur, M. (eds) (1996/1999) Handbook Series on Semiconductor Parameters, vols 1 and 2, World Scientific, London. Bougrov, V., Levinshtein, M.E., Rumyantsev, S.L. and Zubrilov, A. (2001) Properties of Advanced Semiconductor Materials GaN, AlN, InN, BN, SiC, SiGe (eds M.E. Levinshtein, S.L. Rumyantsev and M.S. Shur), John Wiley & Sons, Inc., New York, pp. 1–30. Polian, A., Grimsditch, M. and Grzegory, I. (1996) Journal of Applied Physics, 79, 3343–3344. Kisielowski, C., Kr€ uger, J., Ruvimov, S., Suski, T., Ager, J.W., III, Jones, E., Liliental-Weber, Z., Rubin, M., Weber, E.R., Bremser, M.D. and Davis, R.F. (1996) Physical Review B: Condensed Matter, 54, 17745. Nikolaev, V., Shpeizman, V. and Smirnov, B. (1998) The Second Russian Workshop on GaN, InN, and AIN-Structures and Devices, June 2, St Petersburg Technical University, St Petersburg, Russia. Drory, M.D., Ager, J.W., Suski, T., Grzegory, I. and Porowski, S. (1996) Hardness and fracture toughness of bulk single crystal gallium nitride. Applied Physics Letters, 69 (26), 4044–4046. Wright, A.F. (1997) Elastic properties of zinc-blende and wurtzite AlN, GaN, and
43
44
45
46
47
48
49
50
51
52
53
InN. Journal of Applied Physics, 82 (6), 2833–2839. Madelung, O. (ed.) (1991) Semiconductor: Group IV Elements and III–V Compound, Data in Science and Technology (series ed. R. Poerschke), Springer, Berlin. Truell, R., Elbaum, C. and Chick, B.B. (1969) Ultrasonic Methods in Solid State Physics, Academic Press, New York. Kim, K., Lambrecht, W.R.L. and Segall, B. (1994) Physical Review B: Condensed Matter, 50, 1502. Ueno, M., Yoshida, M., Onodera, A., Shimomura, O. and Takemura, K. (1994) Physical Review B: Condensed Matter, 49, 14. Leszczynski, M., Podlasin, S. and Suski, T. (1993) Journal of Applied Crystallography, 32, 1528. Qian, W., Skowronski, M. and Rohrer, G.R. (1996) Structural defects and their relationship to nucleation of GaN thin films, in III-Nitride, SiC, and Diamond Materials for Electronic Devices, Materials Research Society Symposium Proceedings, vol. 423 (eds D.K. Gaskill, C.D. Brandt and R.J. Nemanich), Pittsburgh, PA, pp. 475–486. Sheleg, A.U. and Savastenko, V.A. (1977) Vestsi Akademii Nauk, Seria. Fizika i Matematika Nauk, 126. Porowski, S. (1997) Materials Science & Engineering B: Solid State Materials for Advanced Technology, 44, 407–413. Barin, I., Knacke, O. and Kubaschewski, O. (1977) Thermochemical Properties of Inorganic Substances, Springer, Berlin. Fritsch, D., Schmidt, H. and Grundmann, M. (2003) Band-structure pseudopotential calculation of zinc-blende and wurtzite AlN, GaN, and InN. Physical Review B: Condensed Matter, 67, 235205. Chow, T.P. and Ghezzo, M.J. (1996) SiC power devices, in III-Nitride, SiC, and Diamond Materials for Electronic Devices, Materials Research Society Symposium Proceedings, vol. 423 (eds D.K. Gaskill, C.D. Brandt and R.J. Nemanich), Pittsburgh, PA, pp. 69–73.
References 54 Ejder, E. (1971) Physica Status Solidi a: Applied Research, 6, 445–448. 55 Barker, A.S., Jr and Ilegems, M. (1973) Infrared lattice vibrations and freeelectron dispersion in GaN. Physical Review B: Condensed Matter, 7, 743. 56 Manchon, D.D., Jr, Barker, A.S., Jr, Dean, P.J. and Zetterstrom, R.B. (1970) Solid State Communications, 8, 1227. 57 Siegle, H., Kaczmarczyk, G., Filippidis, L., Litvinchuk, L., Hoffmann, A. and Thornsen, C. (1997) Zone-boundary phonons in hexagonal and cubic GaN. Physical Review B: Condensed Matter, 55 (11), 7000–7004. 58 Karch, K., Wagner, J.-M. and Bechstedt, F. (1998) Ab initio study of structural, dielectric, and dynamical properties of GaN. Physical Review B: Condensed Matter, 57, 7043–7049. 59 Zi, J., Wan, X., Wei, G., Zhang, K. and Xie, X. (1996) Journal of Physics: Condensed Matter, 8, 6323–6328. 60 Lemos, V., Arguello, C.A. and Leite, R.C.C. (1972) Solid State Communications, 11, 1351. 61 Dingle, R. and Ilegems, M. (1971) Solid State Communications, 9, 175. 62 Rheinlander, A. and Neumann, H. (1974) Physica Status Solidi b: Basic Research, 64, K123. 63 Bloom, S., Harbeke, G., Meier, E. and Ortenburger, I.B. (1974) Physica Status Solidi b: Basic Research, 66, 161–168. 64 Pankove, J.I., Bloom, S. and Harbeke, G. (1975) RCA Review, 36, 163. 65 Look, D.C. and Sizelove, J.R. (2001) Predicted mobility in bulk GaN. Applied Physics Letters, 79 (8), 1133–1135. 66 Stroscio, M.A. and Dutta, M. (2001) Phonons in Nanostructures, Cambridge University Press. 67 Duboz, J.Y. (2002) Hot photoluminescence in GaN: carrier energy relaxation and hot phonon effects. Journal of Applied Physics, 92 (8), 4312–4319. 68 Lee, B.C., Kim, K.W., Dutta, M. and Stroscio, M.A. (1997) Physical Review B: Condensed Matter, 56, 997.
69 Ramirez-Flores, G., Navarro-Contreras, H., Lastras-Martinez, A., Powell, R.C. and Greene, J.E. (1994) Physical Review B: Condensed Matter, 50, 8433. 70 Fan, J.W., Li, M.F., Chong, T.C. and Xia, J.B. (1996) Electronic properties of zincblende GaN, AlN, and their alloys Ga1 xAlxN. Journal of Applied Physics, 79 (1), 188–194. 71 Chang, W.Y. and Harmon, B.N. (1986) Physical Review B: Condensed Matter, 34, 5305. 72 Hejda, B. and Hauptmanova, K. (1969) Physica Status Solidi, 36, K95. 73 Blum, S. (1971) Journal of Physics and Chemistry of Solids, 32, 2027. 74 Jones, D. and Lettington, A.H. (1972) Solid State Communications, 11, 701. 75 Kobayasi, A., Sankey, O.F., Volz, S.M. and Dow, J.D. (1983) Physical Review B: Condensed Matter, 28, 935. 76 Huang, M.Z. and Ching, W.Y. (1985) Journal of Physics and Chemistry of Solids, 46, 977. 77 Slack, G.A. (1973) Journal of Physics and Chemistry of Solids, 34, 321–335. 78 Goldberg, Y. (2001) Properties of Advanced Semiconductor Materials GaN, AlN, InN, BN, SiC, SiGe (eds M.E. Levinshtein, S.L. Rumyantsev and M.S. Shur), John Wiley & Sons, Inc., New York, pp. 31–47. 79 Gerlich, D., Dole, S.L. and Slack, G.A. (1986) Journal of Physics and Chemistry of Solids, 47, 437. 80 Thokala, R. and Chaudhuri, J. (1995) Thin Solid Films, 266 (2), 189–191. 81 Kawabi, K., Tredgold, R.H. and Inuishi, Y. (1967) Electrical Engineering in Japan, 87, 62. 82 Yonenaga, I., Shima, T. and Sluiter, M.H.F. (2002) Japanese Journal of Applied Physics, 41, 4620. 83 McNeil, E., Grimsditch, M. and French, R.H. (1993) Journal of the American Ceramic Society, 76 (5), 1132–1136. 84 Vurgaftman, I. and Meyer, J.R. (2003) Band parameters for nitrogen-containing semiconductors. Journal of Applied Physics, 94 (6), 3675–3696.
j113
j 1 General Properties of Nitrides
114
85 Slack, G.A. and Bartram, S.F. (1975) Journal of Applied Physics, 46 (1), 89–89. 86 Touloukian, Y.S., Kirby, R.K., Taylor, R.E. and Lee, T.Y.R. (eds) (1977) Thermophysical Properties of Matter, vol. 13, Plenum Press, New York. 87 Meng, W.J. (1994) Properties of Group III Nitrides (ed. J.H. Edgar), IEE EMIS Data Review Series, No. 11, INSPEC, The Institution of Electrical Engineers, Stevenage, UK, pp. 22–29. 88 Krukowski, S., Leszczynski, M. and Porowski, S. (1999) Thermal properties of the group III nitrides, in Properties, Processing and Applications of Gallium Nitride and Related Semiconductors (eds J.H. Edgar, S. Strite, I. Akasaki, H. Amano and C. Wetzel), EMIS Data Review Series, No. 23, INSPEC, The Institution of Electrical Engineers, Stevenage, UK, pp. 21–28. 89 Sirota, N.N. and Golodushko, V.Z. (1974) Tezisy Dokl., Vses Konf. Khi., Svyazi Poluprovdn. Polumetallakh 5th, p. 98. 90 Slack, G.A., Tanzilli, R.A., Pohl, R.O. and Vandersande, J.W. (1987) Journal of Physics and Chemistry of Solids, 48 (7), 641–647. 91 Slack, G., Schowalter, L., Rojo, J., Morelli, D. and Freitas, J. (2002) Proceedings of the International Workshop on Bulk Nitrides, May 2002, Amazonas, Brazil, Journal of Crystal Growth, 246 (3–4), 287–298; For an application of the method to synthetic diamond, see Morelli, D.T., Beetz, C.P. and Perry, T.A. (1988) Thermal conductivity of synthetic diamond films. Journal of Applied Physics, 64 (6), 3063–3066. 92 MacChesney, J.B., Bridenbaugh, P.M. and OConnor, P.B. (1970) Materials Research Bulletin, 5, 783. 93 Koshchenko, V.I., Grinberg, Ya.Kh. and Demidienko, A.F. (1984) Inorganic Materials, 20 (11), 1550–1553. 94 Ribeiro, C.T.M., Alvarez, F. and Zanatta, A.R. (2002) Applied Physics Letters, 81, 1005.
95 See, for example, Colthup, N.B., Daly, L.H. and Wiberley, S.E. (1990) Infrared and Raman Spectroscopy, Academic Press, San Diego, FL. 96 Ren, Z.M., Lu, Y.F., Ni, H.Q., Liew, T.Y.F., Cheong, B.A., Chow, S.K., Ng, M.L. and Wang, J.P. (2000) Journal of Applied Physics, 88, 7346. 97 Chang, W.Y. and Harmon, B.N. (1986) Physical Review B: Condensed Matter, 34, 5305. 98 Hejda, B. and Hauptmanova, K. (1969) Physica Status Solidi, 36, K95. 99 Blum, S. (1971) Journal of Physics and Chemistry of Solids, 32, 2027. 100 Jones, D. and Lettington, A.H. (1972) Solid State Communications, 11, 701. 101 Kobayashi, A., Sankey, O.F., Volz, S.M. and Dow, J.D. (1983) Physical Review B: Condensed Matter, 28, 935. 102 Huang, M.Z. and Ching, W.Y. (1985) Journal of Physics and Chemistry of Solids, 46, 977. 103 Guo, Q. and Yoshida, A. (1994) Japanese Journal of Applied Physics, Part 1: Regular Papers, Short Notes & Review Papers, 33 (5A), 2453–2456. 104 Teisseyre, H., Perlin, P., Suski, T., Grzegory, I., Porowski, S., Jun, J., Pietraszko, A. and Moustakas, T.D. (1994) Temperature dependence of the energy gap in GaN bulk single crystals and epitaxial layer. Journal of Applied Physics, 76 (4), 2429–2434. 105 Yamashita, H., Fukui, K., Misawa, S. and Yoshida, S. (1979) Journal of Applied Physics, 50, 896. 106 Roskovcova, L. and Pastrnak, J. (1980) Czechoslovak Journal of Physics B, 30, 586. 107 Silveira, E., Freitas, J.A., Jr, Kneissl, M., Treat, D.W., Johnson, N.M., Slack, G.A. and Schowalter, L.J. (2004) Nearbandedge cathodoluminescence of an AlN homoepitaxial film. Applied Physics Letters, 84 (18), 3501–3503. 108 Silveira, E., Freitas, J.A., Jr, Glembocki, O.J., Slack, G.A. and Schowalter, L.J. (2005) Excitonic structure of bulk AlN from optical reflectivity and
References
109 110
111
112 113
114
115
116
117
118 119
120
121
cathodoluminescence measurements. Physical Review B: Condensed Matter, 71, 041201(R)-1–041201(R)-4. Gorczyca, I. and Christensen, N.E. (1993) Physica B, 185, 410–414. Gorczyca, I., Svane, A. and Christensen, N.E. (1997) Internet Journal of Nitride Semiconductor Research, 2, article 18. Collins, A.T., Lightowlers, E.C. and Dean, P.J. (1967) Physical Review, 158 (3), 833–838. Chu, T.L. and Keln, R.W., Jr (1975) Journal of the Electrochemical Society, 122, 995. Moore, W.J., Freitas, J.A., Jr, Holm, R.T., Kovalenkov, O. and Dmitriev, V. (2005) Infrared dielectric function of wurtzite aluminum nitride. Applied Physics Letters, 86, 141912-1–141912-3. Suzuki, M. and Uenoyama, T. (1995) Physical Review B: Condensed Matter, 52, 8132. Kim, K., Lambrecht, W.R.L. and Segall, B. (1997) Physical Review B: Condensed Matter, 56, 7363. Suzuki, M. and Uenoyama, T. (1996) Strain effect on electronic and optical properties of GaN/AlGaN quantum-well lasers. Journal of Applied Physics, 80 (12), 6868–6874. Sanjurjo, J.A., Lopez-Cruz, E., Vogi, P. and Cardona, M. (1983) Dependence on volume of the phonon frequencies and the IR effective charges of several III–V semiconductors. Physical Review B: Condensed Matter, 28, 4579. Carlone, C., Lakin, K.M. and Shanks, H.R. (1984) Journal of Applied Physics, 55, 4010. MacMillan, M.F., Devaty, R.P. and Choyke, W.J. (1993) Infrared reflectance of thin aluminum nitride films on various substrates. Applied Physics Letters, 62 (7), 750–752. Perlin, P., Polian, A. and Suski, T. (1993) Raman-scattering studies of aluminum nitride at high pressure. Physical Review B: Condensed Matter, 47 (5), 2874–2877. Van Camp, P.E., Van Doren, V.E. and Devreese, J.T. (1991) High-pressure properties of wurtzite- and rocksalt-type
122
123
124
125
126
127
128
129
130 131
aluminum nitride. Physical Review B: Condensed Matter, 44 (16), 9056–9059. Pearson, W.B. (1967) A Handbook of Lattice Spacings and Structures of Metals and Alloys, Pergamon Press, Oxford. Pichugin, I.G. and Tiachala, M. (1978) Izvestia Akademii Nauk SSSR, Neorganicheskie Materialy, 14, 175. Zubrilov, A. (2001) Properties of Advanced Semiconductor Materials GaN, AlN, InN, BN, SiC, SiGe (eds M.E. Levinshtein, S.L. Rumyantsev and M.S. Shur), John Wiley & Sons, Inc., New York, pp. 49–66. Edgar, J.H., Wei, C.H., Smith, D.T., Kistenmacher, T.J. and Bryden, W.A. (1997) Journal of Materials Science, 8, 307. Perlin, P., Iota, V., Weinstein, B.A., Wisniewski, P., Suski, T., Eliseev, P.G. and Osinski, M. (1997) Influence of pressure on photoluminescence and electroluminescence in GaN/InGaN/ AlGaN quantum wells. Applied Physics Letters, 70, 2993–2995. Krukowski, S., Witek, A., Adamczyk, J., Jun, J., Bockowski, M., Grzegory, I., Lucznik, B., Nowak, G., Wroblewski, M., Presz, A., Gierlotka, S., Stelmach, S., Palosz, B., Porowski, S. and Zinn, P. (1998) Thermal properties of indium nitride. Journal of Physics and Chemistry of Solids, 59, 289–295. Davydov, V.Yu., Emtsev, V.V., Goncharuk, A.N., Smirnov, A.N., Petrikov, V.D., Mamutin, V.V., Vekshin, V.A., Ivanov, S.V., Smirnov, M.B. and Inushima, T. (1999) Experimental and theoretical studies of phonons in hexagonal InN. Applied Physics Letters, 75, 3297–3299. Tansley, T.L. (1994) Properties of Group III Nitrides (ed. J.H. Edgar), INSPEC, London, p. 39. Sobolev, V.V. and Zlobina, M.A. (1999) Semiconductors, 33, 395. Tyagai, V.A., Evstigneev, A.M., Krasiko, A.N., Andreeva, A.F. and Malakhov, V.Ya. (1977) Soviet Physics: Semiconductors, 11, 1257–1259; Fizika i Tekhnika Poluprovodnikov, 1977, 11, 2142 (in Russian).
j115
j 1 General Properties of Nitrides
116
132 Inushima, T., Shiraishi, T. and Davydov, V.Yu. (1999) Phonon structure of InN grown by atomic layer epitaxy. Solid State Communications, 110 (9), 491–495. 133 Lambrecht, W.R. and Segall, B. (1993) Anomalous band-gap behavior and phase stability of c-BN-diamond alloys. Physical Review B: Condensed Matter, 47, 9289–9296. 134 Yeo, Y.C., Chong, T.C. and Li, M.F. (1998) Electronic band structures and effectivemass parameters of wurtzite GaN and InN. Journal of Applied Physics, 83, 1429–1436. 135 Pugh, S.K., Dugdale, D.J., Brand, S. and Abram, R.A. (1999) Semiconductor Science and Technology, 14, 23–31. 136 Foley, C.P. and Tansley, T.L. (1986) Pseudopotential band structure of indium nitride. Physical Review B: Condensed Matter, 33, 1430. 137 Yeh, C.-Y., Lu, Z.W., Froyen, S. and Zunger, A. (1992) Zinc-blende-wurtzite polytypism in semiconductors. Physical Review B: Condensed Matter, 46, 10086–10097. 138 Bechstedt, F. and Furthm€ uller, J. (2002) Do we know the fundamental energy gap of InN? Journal of Crystal Growth, 246 (3–4), 315–319. 139 Johnson, W.C., Parson, J.B. and Crew, M.C. (1932) The Journal of Physical Chemistry, 36, 2561. 140 Maruska, H.P. and Tietjen, J.J. (1969) Applied Physics Letters, 15, 327. 141 Pankove, J.I. (1972) Journal of the Electrochemical Society, 119, 1110. 142 Chu, T.L. (1971) Journal of the Electrochemical Society, 118, 1200. 143 Lakshmi, E. (1981) Thin Solid Films, 83, L137. 144 Morimoto, Y. (1974) Journal of the Electrochemical Society, 121, 1383. 145 Shintani, A. and Minagawa, S. (1976) Journal of the Electrochemical Society, 123, 706. 146 Itoh, K., Amano, H., Hiramatsu, K. and Akasaki, I. (1991) Japanese Journal of Applied Physics, 30, 1604.
147 Ito, K., Hiramatsu, K., Amano, H. and Akasaki, I. (1990) Journal of Crystal Growth, 104, 533. 148 Visconti, P., Reshchikov, M.A., Jones, K.M., Wang, D.F., Cingolani, R., Morkoc, H., Molnar, R.J. and Smith, D.J. (2001) Highly selective photoelectro-chemical etching of nitride materials for defect investigation and device fabrication. Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 19 (4), 1328–1333. 149 Mohammad, S.N., Salvador, A. and Morkoc, H. (1995) Proceedings of the IEEE, 83, 1306. 150 Pearton, S.J., Zolper, J.C., Shul, R.J. and Ren, F. (1999) GaN: processing, defects, and devices. Journal of Applied Physics, 86, 1. 151 Maruska, H.P., Anderson, L.J. and Stevenson, D.A. (1974) Journal of the Electrochemical Society, 121, 1202. 152 Lagerstedt, O. and Monemar, B. (1979) Physical Review B: Condensed Matter, 19, 3064. 153 Maruska, H.P., Anderson, L.J. and Stevenson, D.A. (1974) Journal of the Electrochemical Society, 121, 1202. 154 Kr€ uger, J., Sudhir, G.S., Corlatan, D., Cho, Y., Kim, Y., Klockenbrink, R., Ruvimov, S., Liliental-Weber, Z., Kisielowski, C., Rubin, M., Weber, E.R., McDermott, B., Pittman, R. and Gertner, E.R. (1997) Proceedings of the Fall 97 Meeting of the Materials Research Society, Boston. 155 Leszczynski, M., Suski, T., Domagala, J. and Prystawko, P. (1999) Lattice parameters of the group III-nitrides, in Properties, Processing and Applications of Gallium Nitride and Related Semiconductors (eds J.H. Edgar, S. Strite, I. Akasaki, H. Amano and C. Wetzel), EMIS Data Review Series, No. 23, INSPEC, The Institution of Electrical Engineers, Stevenage, UK, pp. 6–10. 156 Liu, C., Bensching, B., Volz, K. and Rauschenbach, B. (1997) Applied Physics Letters, 71, 2313.
References 157 Akasaki, I. and Amano, H. (1994) Properties of Group III Nitrides (ed. J.H. Edgar), IEE EMIS Data Review Series, INSPEC, The Institution of Electrical Engineers, Stevenage, UK, p. 222. 158 Van Camp, P.E., Van Doren, V.E. and Devreese, J.T. (1992) Solid State Communications, 81, 23. 159 Murnaghan, F.D. (1944) Proceedings of the National Academy of Sciences of the United States of America, 30, 244. 160 Polian, A. (1999) Mechanical properties of the group III nitrides, in Properties, Processing and Applications of Gallium Nitride and Related Semiconductors (eds J.H. Edgar, S. Strite, I. Akasaki, H. Amano and C. Wetzel), EMIS Data Review Series, No. 23, INSPEC, The Institution of Electrical Engineers, Stevenage, UK, pp. 11–20. 161 Miwa, K. and Fukumoto, A. (1993) Physical Review B: Condensed Matter, 48, 7897. 162 Savastenko, V.A. and Sheleg, A.U. (1978) Physica Status Solidi a: Applied Research, 48, K135. 163 Chetverikova, I.F., Chukichev, M.V. and Rastorguev, L.N. (1986) Inorganic Materials, 22, 53. 164 Sherwin, M.E. and Drummond, T.J. (1991) Journal of Applied Physics, 69, 8423. 165 Deger, C., Born, E., Angerer, H., Ambacher, O., Stutzmann, M., Hornsteiner, J., Riha, E. and Fischerauer, G. (1998) Applied Physics Letters, 72, 2400. 166 Yamaguchi, M., Yagi, T., Azuhata, T., Sota, T., Suzuki, K., Chichibu, S. and Nakamura, S. (1997) Journal of Physics: Condensed Matter, 9, 241. 167 Kim, K.W., Lambert, W.R.L. and Segall, B. (1996) Physical Review B: Condensed Matter, 53, 16310. 168 Kim, K.W., Lambert, W.R.L. and Segall, B. (1994) Physical Review B: Condensed Matter, 50, 1502. 169 Davydov, V.Yu., Kitaev, Yu.E., Goncharuk, I.N., Smirnov, A.N., Graul, J., Semchinova, O., Uffmann, D., Smirnov, M.B., Mirgorodsky, A.P. and Evarestov,
170
171
172
173
174
175
176
177
178
179
180
R.A. (1998) Physical Review B: Condensed Matter, 58, 12899. Yamaguchi, M., Yagi, T., Sota, T., Deguchi, T., Shimada, K. and Nakamura, S. (1999) Journal of Applied Physics, 85, 8502. Schwarz, R.B., Khachaturyan, K. and Weber, E.R. (1997) Applied Physics Letters, 70, 1122. Bergman, L., Dutta, M. and Nemanich, R.J. (2000) Raman scattering spectroscopy and analyses of III–V nitride based materials, in Raman Scattering in Material Science, Springer, Berlin, Chapter 7. Cardona, M. (1982) Light Scattering in Solids II, Springer Topics in Applied Physics, vol. 50 (eds M. Cardona and G. G€ untherodt), Springer, Berlin, pp. 19–178. Harima, H. (2002) Properties of GaN and related compounds studied by means of Raman scattering. Journal of Physics: Condensed Matter, 14, R967–R993. Arguello, C.A., Rousseau, D.L. and Porto, S.P.S. (1969) Physical Review B: Condensed Matter, 181, 1351. Cingolani, A., Ferrara, M., Lugara, M. and Scamarcio, G. (1986) Solid State Communications, 58, 823. Huang, D., Yun, F., Visconti, P., Reshchikov, M.A., Wang, D., Morkoc, H., Rode, D.L., Farina, L.A., Kurdak,C., Tsen, K.T., Park, S.S. and Lee, K.Y. (2001) Hall mobility and carrier concentration in GaN free-standing templates grown by hydride vapor phase epitaxy with high quality. Solid State Electronics, 45 (5), 711–715. Wagner, J.M. and Bechstedt, F. (2002) Physical Review B: Condensed Matter, 66, 115202. Miyoshi, S., Onabe, K., Ohkouchi, N., Yaguchi, H., Ito, R., Fukutsu, S. and Shraki, Y. (1992) Journal of Crystal Growth, 124, 439. Miller, S.C. and Love, W.F. (1967) Tables of Irreducible Representations of Space Groups and Corepresentations of Magnetic Space Groups, Pruett, Boulder, CO.
j117
j 1 General Properties of Nitrides
118
181 Davydov, V.Yu., Averkiev, N.S., Goncharuk, I.N., Nelson, D.K., Nikitina, I.P., Polkovnikov, A.S., Smirnov, A.N., Jacobson, M.A. and Semchinova, O.K. (1997) Journal of Applied Physics, 82, 5097. 182 Leszczynski, M. and Walker, J.F. (1993) Applied Physics Letters, 62, 1484–1487. 183 Sime, J.R. and Margrave, J.L. (1956) The Journal of Physical Chemistry, 60, 810. 184 Morimoto, Y. (1974) Journal of the Electrochemical Society, 121, 1383. 185 Furtado, M. and Jacob, G. (1983) Journal of Crystal Growth, 64, 257. 186 Gordienko, S.P., Samsonov, G.V. and Fesenko, V.V. (1964) Soviet Journal of Physical Chemistry, 38, 1620. 187 Munir, Z.A. and Searcy, A.W. (1965) Journal of Chemical Physics, 42, 4223. 188 Groh, R., Gerey, G., Bartha, L. and Pankove, J.I. (1974) Physica Status Solidi a: Applied Research, 26, 353. 189 Munir, Z.A. and Searcy, A.W. (1965) Journal of Chemical Physics, 42, 4223. 190 Thurmond, C.D. and Logan, R.A. (1972) Journal of the Electrochemical Society, 119, 622. 191 Karpinski, J., Jun, J. and Porowski, S. (1984) Journal of Crystal Growth, 66, 1. 192 Grzegory, I., Krukowski, S., Leszczynski, M., Perlin, P., Suski, T. and Porowski, S. High pressure crystallization of GaN, in Nitride Semiconductors – Handbook on Materials and Devices (eds P. Ruterana, M. Albrecht and J. Neugebauer), Wiley-VCH Verlag GmbH, Weinheim, Germany. 193 Sasaki, T. and Matsuoka, T. (1995) Journal of Applied Physics, 77, 192. 194 Madar, R., Jacob, G., Hallais, J. and Fruchart, R. (1975) Journal of Crystal Growth, 31, 197. 195 Slack, G.A. and McNelly, T.F. (1976) Journal of Crystal Growth, 34, 263. 196 Ambacher, O. (1998) Growth and applications of group III-nitrides. Journal of Physics D: Applied Physics, 31, 2653. 197 Matsuoka, T. (2004) Progress in nitride semiconductors from GaN to InN – MOVPE growth and characteristics.
198
199
200
201
202 203
204
205
206
207
208
209
Superlattices and Microstructures, 37 (1), 19–32. Popovici, G. and Morkoc, H. (1999) Growth and doping of and defects in IIInitrides, in GaN and Related Materials II: Optoelectronic Properties of Semiconductors and Superlattices (ed. S.J. Pearton, series ed. M.O. Manaresh), vol. 7, Gordon and Breach, Amsterdam, pp. 93–172. Ambacher, O., Brandt, M.S., Dimitrov, R., Metzger, T., Stutzmann, M., Fischer, R.A., Miehr, A., Bergmaier, A. and Dollinger, G. (1996) Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 14, 3532. Redhead, P.A. (1962) Vacuum, 12, 203; Redhead, P.A. (1962) Journal of Crystal Growth, 9, 158. Madar, R., Jacob, G., Hallais, J. and Fruchart, R. (1975) Journal of Crystal Growth, 31, 197. Karpinski, J. and Porowski, S. (1984) Journal of Crystal Growth, 66, 11. Glushko, W.P. (ed.) (1979) Termodinamiczeskije swojstwa indiwidualnych weszczestw, Nauka, Moscow (in Russian). Zinowiev, W.E. (1989) Teplofiziczeskije swojstwa metallow pri wysokich temperaturach, Metallurgia, Moscow (in Russian). Bhandari, C.M. and Rowe, D.M. (1988) Thermal Conduction in Semiconductors, John Wiley & Sons, Inc., New York. Kittel, C. (1986) Introduction to Solid State Physics, 6th edn, John Wiley & Sons, Inc., New York, p. 150. Sichel, E.K. and Pankove, J.I. (1977) Journal of Physics and Chemistry of Solids, 38, 330. Witek, A. (1998) Some aspects of thermal conductivity of isotopically pure diamond – a comparison with nitrides. Diamond & Related Materials, 7 (7), 962–964. Florescu, D.I., Asnin, V.M., Pollak, F.H., Jones, A.M., Ramer, J.C., Schurman, M.J. and Ferguson, I. (2000) Applied Physics Letters, 77, 1464.
References 210 Kotchetkov, D., Zouand, J., Balandin, A., Florescu, D.I. and Pollak, F.H. (2001) Effect of dislocations on thermal conductivity of GaN. Applied Physics Letters, 79 (26), 4316–4318. 211 Florescu, D.I., Asnin, V.M., Pollak, F.H. and Molnar, R.J. (2000) Materials Research Society Symposium Proceedings, 595, 3.89.1. 212 Florescu, D.I., Asnin, V.M., Pollak, F.H., Molnar, R.J. and Wood, C.E.C. (2000) Journal of Applied Physics, 88, 3295. 213 Florescu, D.I., Pollak, F.H., Lanford, W.B., Khan, F., Adesida, I. and Molnar, R.J. (2000) Plasma-induced effects on the thermal conductivity of hydride vapor phase epitaxy grown n-GaN/sapphire (0001). Materials Research Society Symposium Proceedings, 639, G11.57. 214 Kittel, C. Introduction to Solid State Physics, 4th edn, John Wiley & Sons, Inc., New York, p. 215. 215 Koshchenko, V.I., Demidienko, A.F., Sabanova, L.D., Yachmenev, V.E., Gran, V.E. and Radchenko, A.E. (1979) Inorganic Materials, 15, 1329–1330 (translation of Izv. Akad. Nauk SSSR, Neorg. Mater., 1979, 15, 1686–1687). 216 Demidienko, A.F., Koshchenko, V.I., Sabanova, L.D. and Gran, V.E. (1975) Russian Journal of Physical Chemistry, 49, 1585. 217 Elwell, D. and Elwell, M.M. (1988) Progress in Crystal Growth and Characterization of Materials, 17, 53. 218 Porowski, S. and Grzegory, I. (1994) Properties of Group III Nitrides (ed. J.H. Edgar), IEE EMIS Data Review Series, No. 11, INSPEC, The Institution of Electrical Engineers, Stevenage, UK, pp. 71, 76, 83. 219 Oliner, A.A. (ed.) (1978) Acoustic Surface Waves, Topics in Applied Physics, vol. 24, Springer, Berlin. 220 Petrov, I., Mojab, E., Powell, R., Greene, J., Hultman, L. and Sundgren, J.E. (1992) Applied Physics Letters, 60, 2491. 221 Vollstadt, H., Ito, E., Akaishi, M., Akimoto, S. and Fukunaga, O. (1990)
222
223
224
225 226 227
228
229
230
231
232
233 234
235
Proceedings of the Japan Academy, Series B, 66, 7. Ruiz, E., Alvarez, S. and Alemany, P. (1994) Physical Review B: Condensed Matter, 49, 7115. Tsubouchi, K. and Mikoshiba, N. (1985) Zero-temperature coefficient SAW devices on AlN epitaxial films. IEEE Transactions on Sonics and Ultrasonics, 32, 634–644. Tsubouchi, T., Sugai, K. and Mikoshiba, N. (1981) Ultrosonic Symposium Proceedings, IEEE, New York, p. 375. Taylor, K.M. and Len, C. (1960) Journal of the Electrochemical Society, 107, 308. Cline, C.F. and Kalm, J.S. (1963) Journal of the Electrochemical Society, 110, 773. Gleize, J., Renucci, M.A., Frandon, J., Bellett-Amalric, E. and Daudin, B. (2003) Journal of Applied Physics, 93, 2065. Tischler, J.G. and Freitas, J.A., Jr (2004) Anharmonic decay of phonons in strain free wurtzite AlN. Applied Physics Letters, 85, 1943. Bergman, L., Alexson, D., Murphy, P.L., Nemanich, R.J., Dutta, M., Stroscio, M.A., Balkas, C., Shin, H. and Davis, R.F. (1999) Physical Review B: Condensed Matter, 59, 12977. Ribeiro, C.T.M., Alvarez, F. and Zanatta, A.R. (2002) Applied Physics Letters, 81, 1005. See, for example, Colthup, N.B., Daly, L.H. and Wiberley, S.E. (1990) Infrared and Raman Spectroscopy, Academic Press, San Diego, FL. Ren, Z.M., Lu, Y.F., Ni, H.Q., Liew, T.Y.F., Cheong, B.A., Chow, S.K., Ng, M.L. and Wang, J.P. (2000) Journal of Applied Physics, 88, 7346. Karch, K. and Bechstedt, F. (1997) Physical Review B: Condensed Matter, 56, 7404. T€ ut€ unc€ u, H.M. and Srivastava, G.P. (2000) Phonons in zinc-blende and wurtzite phases of GaN, AlN, and BN with the adiabatic bond-charge model. Physical Review B: Condensed Matter, 62 (8), 5028–5235. Class, W. (1968) NASA Report CR-1171.
j119
j 1 General Properties of Nitrides
120
236 Van Vechten, J.A. (1973) Physical Review B: Condensed Matter, 7, 1479. 237 Yim, W.M. and Paff, R.J. (1974) Journal of Applied Physics, 45, 1456. 238 Mah, A.D., King, E.G., Weller, W.W. and Christensen, A.U. (1961) Bureau of Mines Report of Investigations, 5716, 18. 239 Glushko, V.P., Gurevich, L.V., Bergman, G.A., Weitz, I.V., Medvedev, V.A., Chachkurov, G.A. and Yungman, V.S. (1979) Thermodinamicheskiie swoistwa indiwidualnych weshchestw, vol. 1, Nauka, Moscow, pp. 164–165. 240 Slack, G.A., Tanzilli, R.A., Pohl, R.O. and Vandersande, J.W. (1987) Journal of Physics and Chemistry of Solids, 48, 641. 241 Slack, G.A. and McNelly, T.F. (1977) Journal of Crystal Growth, 42, 560 (AlN from pellet drop, sublimation, and final growth). 242 Florescu, D.I., Asnin, V.M. and Pollak, F.H. (2001) Thermal conductivity of GaN and AlN. Compound Semiconductor, 7 (2), 62. 243 Nikolaev, A., Nikitina, I., Zubrilov, A., Mynbaeva, M., Melnik, Y. and Dmitriev, V. (2000) Materials Research Society Symposium Proceedings, 595, 6.5.1. 244 Edwards, J., Kawabe, K., Stevens, G. and Tredgold, R.H. (1965) Solid State Communications, 3, 99. 245 Cox, G.A., Cummins, D.O., Kawabe, K. and Tredgold, R.H. (1967) Journal of Physics and Chemistry of Solids, 28, 543. 246 Yim, W.M., Stotko, E.J., Zanzucchi, P.J., Pankove, J.I., Ettenberg, M. and Gilbert, S.L. (1973) Journal of Applied Physics, 44, 292. 247 Yoshida, S., Misawa, S., Fujii, Y., Takada, S., Hayakawa, H., Gonda, S. and Itoh, A. (1979) Journal of Vacuum Science & Technology, 16, 990. 248 Chu, T.L., Ing, D.W. and Noreika, A.J. (1967) Solid State Electronics, 10, 1023. 249 Rutz, R.F. (1976) Applied Physics Letters, 28, 379. 250 Rutz, R.F., Harrison, E.P. and Cuome, J.J. (1973) IBM Journal of Research, 17, 61.
251 Edwards, J., Kawabe, K., Stevens, G. and Tredgold, R.H. (1965) Solid State Communications, 3, 99. 252 Kawabe, K., Tredgold, R.H. and Inuishi, Y. (1967) Electrical Engineering in Japan, 87, 62. 253 Harris, J.H. and Youngman, R.A. (1994) Properties of Group III Nitrides (ed. J.H. Edgar), IEE EMIS Data Review Series, No. 11, INSPEC, The Institution of Electrical Engineers, Stevenage, UK, p. 203. 254 Pacesova, S. and Jastrabik, L. (1979) Czechoslovak Journal of Physics B, 29, 913. 255 Youngman, R.A. and Harris, J.H. (1990) Journal of the American Ceramic Society, 73, 3238. 256 Harris, J.H., Youngman, R.A. and Teller, R.G. (1990) Journal of Materials Research, 5, 1763. 257 Shishkin, Y., Devaty, R.P., Choyke, W.J., Feng Yun, King, T. and Morkoc, H., (2001) Near bandedge cathodoluminescence studies of AlN films: dependence on MBE growth conditions. International Conference on Nitride Semiconductors, Denver, CO, July; Physica Status Solidi a: Applied Research, 188 (2), 591–594. 258 Perry, P.B. and Rutz, R.F. (1978) Applied Physics Letters, 33, 319. 259 Pastrnak, J. and Roskovcova, L. (1968) Physica Status Solidi, 26, 591. 260 Pastrnak, J. and Souckova, L. (1963) Physica Status Solidi, 9, K71. 261 Pastrnak, J. and Roskovcova, L. (1965) Physica Status Solidi, 11, K73. 262 Karel, F., Pastrnak, J., Hejduk, J. and Losik, V. (1966) Physica Status Solidi, 15, 693. 263 Karel, F. and Pastrnak, J. (1969) Czechoslovak Journal of Physics B, 19, 78. 264 Karel, F. and Pastrnak, J. (1970) Czechoslovak Journal of Physics B, 20, 46. 265 Karel, F. and Mares, J. (1972) Czechoslovak Journal of Physics B, 22, 847. 266 Karel, F. and Mares, J. (1973) Czechoslovak Journal of Physics B, 23, 652. 267 Juza, R. and Hahn, H. (1938) Zeitschrift f€ ur Anorganishce und Allgemeine Chemie, 239, 282.
References 268 Matsuoka, T. (2004) MOVPE growth and characteristics of nitride semiconductors from GaN to InN, in Advanced Materials in Electronics (ed. Q. Guo), Research Signpost, pp. 46–83. 269 Osamura, K., Naka, S. and Murakami, Y. (1975) Journal of Applied Physics, 46, 3432. 270 Puychevrier, N. and Menoret, M. (1976) Thin Solid Films, 36, 141. 271 Tansley, T.L. and Foley, C.P. (1986) Journal of Applied Physics, 59, 3241. 272 Westra, K.L., Lawson, R.P.W. and Brett, M.J. (1988) Journal of Vacuum Science & Technology A: Vacuum Surfaces and Films, 6, 1730. 273 Bhuiyan, A.G., Hashimoto, A. and Yamamoto, A. (2003) Indium nitride (InN): a review on growth, characterization, and properties. Journal of Applied Physics, 94 (5), 2779–2808. 274 Davydov, V.Yu., Klochikhin, A.A., Emtsev, V.V., Ivanov, S.V., Vekshin, V.V., Bechstedt, F., Furthm€ uller, J., Harima, H., Mudryi, A.V., Hashimoto, A., Yamamoto, A., Aderhold, J., Graul, J. and Haller, E.E. (2002) Physica Status Solidi b: Basic Research, 230, R4. 275 Araki, T., Saito, Y., Yamaguchi, T., Kurouchi, M., Nanishi, Y. and Naoi, H. (2004) Radio frequency-molecular beam epitaxial growth of InN epitaxial films on (0 0 0 1) sapphire and their properties. Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 22 (4), 2139–2143. 276 Fritsch, D., Schmidt, H. and Grundmann, M. (2004) Band dispersion relations of zinc-blende and wurtzite InN. Physical Review B: Condensed Matter, 69, 165204. 277 Wu, J. and Walukiewicz, W. (2004) Band gaps of InN and group III-nitride alloys. Superlattices and Microstructures. 34, 63–75. 278 Butcher, K.S.A. (2004) InN, a historic review – from obscurity to controversy, in Advanced Materials in Electronics (ed. Q. Guo), Research Signpost. 279 Butcher, K.S.A. and Tansley, T.L. (2005) InN, latest development and a review of
280
281
282
283
284
285
286
287
288
the band-gap controversy. Superlattices and Microstructures, 38 (1), 1–37. Shubina, T.V., Ivanov, S.V., Jmerik, V.N., Solnyshkov, D.D., Vekshin, V.A., Kopev, P.S., Vasson, A., Leymarie, J., Kavokin, A., Amano, H., Shimono, K., Kasic, A. and Monemar, B. (2004) Mie resonances, infrared emission, and the band gap of InN. Physical Review Letters, 92 (11), 117407–117410. Butcher, K.S.A., Wintrebert-Fouquet, M., Chen, P.P., Timmers, H. and Sherestha, S.K. (2003) Detailed analysis of the absorption data for InN. Presented at the International Symposium on Point Defects and Nonstoichiometry, March 2003, Sendai, Japan; Material Science in Semiconductor Processing, 2003, 6, 351–354. Wolfe, C.M. and Stillman, G.E. (1975) Apparent mobility enhancement in inhomogeneous crystals, in Semiconductors and Semimetals, vol. 10 (eds R.K. Willardson and A.C. Beer), Academic Press, New York, pp. 175–220. Yamaguchi, S., Kariya, M., Nitta, S., Takeuchi, T., Wetzel, W., Amano, H. and Akasaki, I., (1999) Structural properties of InN on GaN grown by metalorganic vapor-phase epitaxy. Journal of Applied Physics, 85 (11), 7682–7688. Strite, S., Chandrasekhar, D., Smith, D.J., Sariel, J., Chen, H., Teraguchi, N. and Morkoc, H. (1993) Journal of Crystal Growth, 127, 204. Hahn, H. and Juza, R. (1940) Zeitschrift f€ ur Anorganishce und Allgemeine Chemie, 244, 111. Tansley, T.L. (1994) Properties of Group III Nitrides, (ed. J.H. Edgar), IEE EMIS Data Review Series, No. 11, INSPEC, The Institution of Electrical Engineers, Stevenage, UK, p. 35. van Camp, P.E., van Doren, V.E. and Devreese, J.T. (1990) Physical Review B: Condensed Matter, 41, 1598. Kubota, K., Kobayashi, Y. and Fujimoto, K. (1989) Journal of Applied Physics, 66, 2984.
j121
j 1 General Properties of Nitrides
122
289 Inushima, T., Yaguchi, T., Nagase, A., Iso, A., Shiraishi, T. and Ooya, S. (1995) Optical and electrical properties of InN grown by the atomic layer epitaxy. 7th International Conference on Indium Phosphide and Related Materials (Cat. No. 95CH35720), IEEE, New York, pp. 187–190. 290 Inushima, T., Yaguchi, T., Nagase, A., Iso, A. and Shiraishi, T. (1996) Investigation of optical properties of InN grown by atomic layer epitaxy. Presented at the International Meeting on Silicon Carbide and Related Materials, 1995, Kyoto, Japan, IOP Publishing, Bristol, UK, pp. 971–974. 291 Osamura, K., Naka, S. and Murakami, Y. (1975) Journal of Applied Physics, 46, 3432. 292 Sheleg, A.V. and Savastenko, V.A. (1976) Vestsi Akademii Nauk USSR, Seria Fiziki i Matematika Nauk, 3, 126. 293 Krukowski, S., Witek, A., Adamczyk, J., Jun, J., Bockowski, M., Grzegory, I., Lucznik, B., Nowak, G., Wroblewski, M., Presz, A., Gierlotka, S., Stelmach, S., Palosz, B., Porowski, S. and Zinn, P. (1998) Journal of Physics and Chemistry of Solids, 59, 289–295. 294 Tansley, T.L. and Foley, C.P. (1985) Proceedings of the 3rd International Conference on Semiinsulating III–V Materials, Warm Springs, OR, 1984 (ed. J.S. Blakemore), Shiva, London. 295 Bryden, W.R., Ecelberger, S.A., Hawley, M.E. and Kistenmacher, T.J. (1994) Diamond SiC, Nitride Wide Bandgap Semiconductors (eds C.H. Carter Jr, G. Gildenblat, S. Nakamura and R.J. Nemanich), Materials Research Society, Pittsburgh, PA; Materials Research Society Symposium Proceedings, 339, 497. 296 Bryden, W.R. and Kistenmacher, T.J. (1994) Properties of Group X Nitrides (ed. J.H. Edgar), IEE EMIS Data Review Series, No. 11, INSPEC, The Institution of Electrical Engineers, Stevenage, UK, pp. 117–118. 297 Abernathy, C.R., Pearton, S.J., Ren, F. and Wisk, P.W. (1993) Journal of Vacuum
298 299
300
301
302
303 304
305
306 307
308 309
Science & Technology B: Microelectronics and Nanometer Structures, 11, 179. Maruyama, T. and Morishita, T. (1994) Journal of Applied Physics, 76, 5809. Yamaguchi, S., Kariya, M., Nitta, S., Takeuchi, T., Wetzel, C., Amano, H. and Akasaki, I. (1999) Journal of Applied Physics, 85, 7682. Lu, H., Schaff, W.J., Hwang, J., Wu, H., Yeo, W., Pharkya, A. and Eastman, L.F. (2000) Applied Physics Letters, 77, 2548. Aderhold, J., Davydov, V.Yu., Fedler, F., Klausing, H., Mistele, D., Rotter, T., Semchinova, O., Stemmer, J. and Graul, J. (2001) Journal of Crystal Growth, 222, 701. Lu, H., Schaff, W.J., Hwang, J., Wu, H., Koley, G. and Eastman, L.F. (2001) Applied Physics Letters, 79, 1489. Motlan, Goldys, E.M. and Tansley, T.L. (2002) Journal of Crystal Growth, 241, 165. Saito, Y., Yamaguchi, T., Kanazawa, H., Kano, K., Araki, T., Nanishi, Y., Teraguchi, N. and Suzuki, A. (2002) Journal of Crystal Growth, 237–239, 1017. Lu, H., Schaff, W.J., Eastman, L.F., Wu, J., Walukiewicz, W., Yu, K.M., Auger, J.W., III, Haller, E.E. and Ambacher, O. (2002) Conference Digest of the 44th Electronic Materials Conference, Santa Barbara, p. 2; Schaff, W.J., Lu, H., Eastman, L.F., Walukiewicz, W., Yu, K.M., Keller, S., Kurtz, S., Keyes, B. and Gevilas, L. (2004) Electrical properties of InN grown by molecular beam epitaxy. Fall 2004 ECS Meeting. State-of-the-Art Program on Compound Semiconductors XLI and Nitride and Wide Bandgap Semiconductors for Sensors, Photonics, and Electronics, vol. 2004-06 (eds V.-H. Ng and A.G. Baca). Higashiwaki, M. and Matsui, T. (2003) Journal of Crystal Growth, 252, 128. Look, D.C., Lu, H., Schaff, W.J., Jasinski, J. and Liliental-Weber, Z. (2002) Applied Physics Letters, 80, 258. Tansley, T.L. and Foley, C.P. (1984) Electronics Letters, 20, 1066. OLeary, S.K., Foutz, B.E., Shur, M.S., Bhapkar, U.V. and Eastman, L.F. (1998) Journal of Applied Physics, 83, 826.
References 310 Bellotti, E., Doshi, B.K., Brennan, K.F., Albrecht, J.D. and Paul Ruden, P. (1999) Ensemble Monte Carlo study of electron transport in wurtzite InN. Journal of Applied Physics, 85, 916. 311 Foutz, B.E., Eastman, L.F., Bhapkar, U.V. and Shur, M.S. (1997) Comparison of high field electron transport in GaN and GaAs. Applied Physics Letters, 70, 2849–2851. 312 Foutz, B.E., OLeary, S.K., Shur, M.S. and Eastman, L.F. (1999) Transient electron transport in wurtzite GaN, InN, and AlN. Journal of Applied Physics, 85, 7727–7734. 313 Wakahara, A., Tsuchiya, T. and Yoshida, A. (1990) Journal of Crystal Growth, 99, 385. 314 Kubota, K., Kobayashi, Y. and Fujimoto, K. (1989) Journal of Applied Physics, 66, 2984–2988. 315 Guo, Q., Kato, O., Fujisawa, M. and Yoshida, A. (1992) Optical constants of indium nitride. Solid State Communications, 83 (9), 721–723. 316 Guo, Q., Ogawa, H., and Yoshida, A., (1996) Optical properties of indium nitride in vacuum ultraviolet region. Journal of Electron Spectroscopy & Related Phenomena, 79, 9. 317 Yodo, T., Yona, H., Ando, H., Nosei, D. and Harada, Y. (2002) Applied Physics Letters, 80, 968. 318 Inushima, T., Mamutin, V.V., Vekshin, V.A., Ivanov, S.V., Sakon, T., Motokawa, M. and Ohoya, S. (2001) Journal of Crystal Growth, 481, 227–228. 319 Davydov, V.Yu., Klochikhin, A.A., Seisyan, R.P., Emtsev, V.V., Ivanov, S.V., Bechstedt, F., Furthmuller, J., Harima, H., Mudryi, A.V., Aderhold, J., Semchinova, O. and Graul, J. (2002) Physica Status Solidi b: Basic Research, 229, R1. 320 Hamberg, I. and Granqvist, C.G. (1986) Journal of Applied Physics Reviews, 60, R123. 321 Alexandrov, D., Butcher, K.S.A. and Wintrebert-Fouquet, M. (2004) Absorption and photoluminescence features caused by defects in InN. Journal of Crystal Growth, 269 (1), 77–86.
322 Alexandrov, D., Butcher, K.S.A. and Wintrebert-Fouquet, M. (2004) Energy band gaps of InN containing oxygen and of the InxAl1xN interface layer formed during InN film growth. Journal of Vacuum Science & Technology A: Vacuum Surfaces and Films, 22, 954–961. 323 Trainor, J.W. and Rose, K. (1974) Journal of Electronic Materials, 3, 821. 324 Haddad, D.B., Takur, J.S., Naik, V.M., Auner, G.W., Naik, R. and Wenger, L.E. (2003) Materials Research Society Symposium Proceedings, 743, 701. 325 Hovel, H.J. and Cuomo, J.I. (1972) Applied Physics Letters, 20, 71. 326 Nataraian, B.R., Eltoukhy, A.H., Greene, J.E. and Barr, T.L. (1980) Thin Solid Films, 69, 201. 327 Sullivan, B.T., Parsons, R.R., Westra, K.L. and Brett, M.J. (1988) Journal of Applied Physics, 64, 4144. 328 Wintrebert-Fouquet, M., Butcher, K.S.A. and Chen, P.P.-T. (2003) InN grown by remote plasma enhanced chemical vapour deposition. Presented at the First International InN Workshop, November 16–20, Fremantle, Australia. 329 Butcher, K.S.A., Wintrebert-Fouquet, M., Chen, P.P.-T., Tansley, T.L. and Srikeaw, S. (2002) Materials Research Society Symposium Proceedings, 693, 341. 330 Demangeot, F., Frandon, J., Pinquier, C., Caumont, M., Briot, O., Maleyre, B., ClurRuffenach, S. and Gil, B. (2003) Raman scattering in large single indium nitride dots: correlation between morphology and strain. Physical Review B: Condensed Matter, 68, 245308. 331 Briot, O., Maleyre, B., Ruffenach, S., Pinquier, C., Demangeot, F., Frandon, J. and Gil, B. (2003) Absorption and Raman scattering processes in InN films and dots. Presented at the 1st International InN Workshop, November 16–20, Fremantle, Australia. 332 Tsen, K.T., Liang, W., Ferry, D.K., Hai Lu, Schaff, W.J., Özg€ ur, Ü., Fu, Y., Moon, Y.T., Yun, F., Morkoc, H. and Everitt, H.O. (2005) Optical studies of carrier dynamics
j123
j 1 General Properties of Nitrides
124
333
334
335 336
337
338
339
340
341
and non-equilibrium optical phonons in nitride-based wide bandgap semiconductors. Superlattices and Microstructures, 38 (2), 77–114. Shan, W., Song, J.J., Feng, Z.C., Schurman, M. and Stall, R.A. (1997) Applied Physics Letters, 71, 2433. Shan, W., Ager, J.W., III, Yu, K.M., Walukiewicz, W., Haller, E.E., Martin, M.C., McKinney, W.R. and Yang, W. (1999) Journal of Applied Physics, 85, 8505. Wolford, D.J. and Bradley, J.A. (1985) Solid State Communications, 53, 1069. Perlin, P., Gorczyca, I., Suski, T., Wisniewski, P., Lepkowski, S., Christansen, N.E., Svane, A., Hansen, M., DenBaars, S.P., Damilano, D., Grandjean, N. and Massies, S.P. (2001) Physical Review B: Condensed Matter, 64, 115319. Tyagai, V.A., Snitko, O.V., Evstigneev, A.M. and Krasiko, A.N. (1981) Physica Status Solidi b: Basic Research, 103, 589. Davydov, V., Klochikhin, A., Ivanov, S., Aderhold, J. and Yamamoto, A. (2003) Growth and properties of InN, in Nitride Semiconductors – Handbook on Materials and Devices, (eds. P. Ruterana, M. Albrecht and J. Neugebauer), Wiley-VCH Verlag GmbH, Weinheim. Kasic, A., Schebert, M., Saito, Y., Nanishi, Y. and Wagner, G. (2002) Effective electron mass and phonon modes in n-type hexagonal InN. Physical Review B: Condensed Matter, 65, 115206. Perlin, P., Litwin-Staszewska, E., Suchanek, B., Knap, W., Camassel, J., Suski, T., Piotrzkowski, R., Grzegory, I., Porowski, S., Kaminska, E. and Chervin, J.C. (1996) Determination of the effective mass of GaN from infrared reflectivity and Hall effect. Applied Physics Letters, 68 (8), 1114–1116. Wu, J., Walukiewicz, W., Shall, W., Yu, K.M., Ager, J.W., III, Haller, E.E., Lu, H. and Schaff, W.J. (2002) Effects of the narrow band gap on the properties of InN. Physical Review B: Condensed Matter, 66, 201403.
342 Osamura, K., Naka, S. and Murakami, Y. (1975) Journal of Applied Physics, 46, 3432. 343 Matsuoka, T. (1994) Properties of Group III Nitrides (ed. J.H. Edgar), IEE EMIS Data Review Series, No. 11, INSPEC, The Institution of Electrical Engineers, Stevenage, UK, pp. 231–238. 344 Yamasaki, S., Asami, S., Shibata, N., Koike, M., Manabe, K., Tanaka, T., Amano, H. and Akasaki, I. (1995) Applied Physics Letters, 66, 1112. 345 Yoshida, S., Misawa, S. and Gonda, S. (1982) Journal of Applied Physics, 53, 6844. 346 Khan, M.R.H., Koide, Y., Itoh, H., Sawaki, N. and Akasaki, I. (1986) Solid State Communications, 60, 509. 347 Zoroddu, A., Bernardini, F., Ruggerone, P. and Fiorentini, V. (2001) Firstprinciples prediction of structure, energetics, formation enthalpy, elastic constants, polarization, and piezoelectric constants of AlN, GaN, and InN: comparison of local and gradientcorrected density-functional theory. Physical Review B: Condensed Matter, 64 (45), 208. 348 Amano, H., Takeuchi, T., Sota, S., Sakai, H. and Akasaki, I. (1997) III–V Nitrides (eds. F.A. Ponce, T.D. Moustakas, I. Akasaki and B. Monemar), Materials Research Society Symposium Proceedings, 449, 1143–1150. 349 Yoshida, S., Misawa, S. and Gonda, S. (1983) Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 1, 250. 350 Wickenden, D.K., Bargeron, C.B., Bryden, W.A., Miragliova, J. and Kistenmacher, T.J. (1994) Applied Physics Letters, 65, 2024. 351 Koide, Y., Rob, H., Khan, M.R.H., Hiramatsu, K., Sawaki, N. and Akasaki, I. (1987) Journal of Applied Physics, 61, 4540. 352 Khan, M.A., Skogman, R.A., Schulze, R.G. and Gershenzon, M. (1983) Applied Physics Letters, 43, 492. 353 Korkotashvili, G.A., Pikhtin, A.N., Pichugin, I.G. and Tsaregorodtsev, A.M.
References
354
355
356
357
358
359
360
361
362
363
364
365
(1984) Soviet Physics: Semiconductors, 18, 913. Wethkamp, T., Wilmers, K., Esser, N., Richter, W., Ambacher, O., Angerer, H., Jungk, G., Johnson, R.L. and Cardona, M. (1998) Thin Solid Films, 313, 745. Paduano, Q.S., Weyburne, D.W., Bouthillette, L.O. and Alexander, M.N. (2002) 7th International Workshop on Wide Bandgap Nitrides, Abstract Book, Richmond, VA, 10–14 March, p. 21. Nikishin, S.A., Faleev, N.N., Zubrilov, A.S., Antipov, V.G. and Temkin, H. (2000) Applied Physics Letters, 76, 3028. Shan, W., Ager, J.W., III, Yu, K.M., Walukiewicz, W., Haller, E.E., Martin, M.C., McKinney, W.R. and Yang, W. (1999) Journal of Applied Physics, 85, 8505. Özg€ ur, Ü., Webb-Wood, G., Everitt, H.O., Yun, F. and Morkoc, H. (2001) Applied Physics Letters, 79, 4103. Wagner, J., Obloh, H., Kunzer, M., Maier, M., Kohler, K. and Johs, B. (2000) Journal of Applied Physics, 89, 2779. Jiang, H., Zhao, G.Y., Ishikawa, H., Egawa, T., Jimbo, T. and Umeno, M. (2001) Journal of Applied Physics, 89, 1046. Ochalski, T.J., Gil, B., Lefebvre, P., Grandjean, M., Leroux, M., Massies, J., Makamura, S. and Morkoc, H. (1999) Applied Physics Letters, 74, 3353. Lee, S.R., Lee, S.R., Wright, A.F., Crawford, M.H., Petersen, G.A., Han, J. and Biefeld, R.M. (1999) Applied Physics Letters, 74, 3344. Yun, F., Reshchikov, M.A., He, L., King, T., Morkoc, H., Novak, S.W. and Wei, L., (2002) Energy band bowing parameter in AlxGa1xN alloys. Journal of Applied Physics: Rapid Communications, 92, 4837–4839. Khan, M.A., Van Hove, J.M., Kuznia, J.N. and Olson, D.T. (1991) Applied Physics Letters, 58, 2408. Tanaka, T., Watanabe, A., Amano, H., Kobayashi, Y., Akasaki, I., Yamazaki, S. and Koike, M. (1994) Applied Physics Letters, 65, 593.
366 Nagatomo, T., Kuboyama, T., Minamino, H. and Omoto, O. (1989) Japanese Journal of Applied Physics, 28, L1334. 367 Matsuoka, T., Tanaka, H., Sasaki, T. and Katsui, K. (1990) Proceedings of the 16th International Symposium on GaAs and Related Compounds, 1989, Karuizawa, Institute of Physics Conference Series, vol. 106, Institute of Physics, Bristol, p. 141. 368 Yoshimoto, N., Matsuoka, T., Sasaki, T. and Katsui, A. (1991) Applied Physics Letters, 59, 2251. 369 Matsuoka, T., Yoshimoto, N., Sasaki, T. and Katsui, A. (1992) Journal of Electronic Materials, 21, 157. 370 Shimizu, M., Hiramatsu, K. and Sawaki, N. (1994) Journal of Crystal Growth, 145, 209. 371 Ho, I.-H. and Stringfellow, G.B. (1996) Applied Physics Letters, 69, 2701. 372 Osamura, K., Naka, S. and Murakami, Y. (1975) Journal of Applied Physics, 46, 3432. 373 Koukitu, A., Takahashi, N., Taki, T. and Seki, H. (1996) Japanese Journal of Applied Physics, 35, L673. 374 Kawaguchi, Y., Shimizu, M., Hiramatsu, K. and Sawaki, N. (1997) Materials Research Society Symposium Proceedings, 449, 89. 375 Nakamura, S. and Mukai, T. (1992) Japanese Journal of Applied Physics, 31, L1457. 376 Piner, E.L., Behbehani, M.K., El-Mastry, N.A., McIntosh, F.G., Roberts, J.C., Boutros, K.S. and Bedair, S.M. (1997) Applied Physics Letters, 70, 461. 377 Wright, A.F., Leung, K. and van Schilfgaarde, M. (2001) Effects of biaxial strain and chemical ordering on the band gap of InGaN. Applied Physics Letters, 78 (2), 189–191. 378 Ferhat, M. and Bechstedt, F. (2002) Firstprinciples calculations of gap bowing in InxGa1xN and InxAl1xN alloys: relation to structural and thermodynamic properties. Physical Review B: Condensed Matter, 65, 0752131–0752137.
j125
j 1 General Properties of Nitrides
126
379 S€ okeland, F., Rohlfing, M., Kr€ uger, P. and Pollmann, J. (2003)Density functional and quasiparticle band-structure calculations for GaxAl1xN and GaxIn1xN alloys. Physical Review B: Condensed Matter, 68, 075203-1–075203-11. 380 Pugh, S.K., Dugdale, D.J., Brand, S. and Abram, R.A. (1999) Journal of Applied Physics, 86, 3768. 381 Teles, L.K., Furthm€ uller, J., Scolfaro, L.M.R., Leite, J.R. and Bechstedt, F. (2000) Physical Review B: Condensed Matter, 62, 2475. 382 Teles, L.K., Furthm€ uller, J., Scolfaro, L.M.R., Leite, J.R. and Bechstedt, F. (2001) Influence of composition fluctuations and strain on gap bowing in InxGa1xN. Physical Review B: Condensed Matter, 63, 085204. 383 Nakamura, S. and Mukai, J. (1995) Journal of Vacuum Science & Technology A: Vacuum Surfaces and Films, 13, 6844. 384 ODonnell, K.P., Martin, R.W., TragerCowan, C., White, M.E., Esona, K., Deatcher, C., Middleton, P.G., Jacobs, K., van der Stricht, W., Merlet, C., Gil, B., Vantomme, A. and Mosselmans, J.F.W. (2001) Materials Science & Engineering B: Solid State Materials for Advanced Technology, 82, 194. 385 Monroy, E., Gogneau, N., Enjalbert, F., Fossard, F., Jalabert, D., Bellet-Amalric, E., Dang, L.S. and Daudin, B. (2003) Molecular-beam epitaxial growth and characterization of quaternary III-nitride compounds. Journal of Applied Physics, 94 (5), 3121–3127. 386 Aumer, M.E., LeBoeuf, S.F., McIntosh, F.G. and Bedair, S.M. (1999) Applied Physics Letters, 75, 3315. 387 Dimakis, E., Georgakilas, A., Androulidaki, M., Tsagaraki, K., Kittler, G., Kalaitzakis, F., Cengher, D., BelletAmalric, E., Jalabert, D. and Pelekanos, N.T. (2003) Plasma-assisted MBE growth of quaternary InAlGaN quantum well heterostructures with room temperature luminescence. Journal of Crystal Growth, 251 (1–4), 476–480.
388 Pereira, S., Correia, M.R., Monteiro, T., Pereira, E., Alves, E., Sequeira, A.D. and Franco, N. (2001) Compositional dependence of the strain-free optical band gap in InxGa1xN layers. Applied Physics Letters, 78 (15), 2137–2139. 389 McCluskey, M.D., Van der Walle, C.G., Romano, L.T., Krusor, B.S. and Johnson, N.M. (2003) Effect of composition on the band gap of strained InxGa1xN alloys. Journal of Applied Physics, 93, 4340–4342. 390 Wu, J., Walukiewicz, W., Yu, K.M., Ager, J.W., III, Haller, E.E., Hai Lu and Schaff, W.J. (2002) Small bandgap bowing in In1 xGaxN alloys. Applied Physics Letters, 80, 4741. 391 Shan, W., Walukiewicz, W., Haller, E.E., Little, B.D., Song, J.J., McCluskey, M.D., Johnson, N.M., Feng, Z.C., Schurman, M. and Stall, R.A. (1998) Journal of Applied Physics, 84, 4452. 392 Pereira, S., Correia, M.R., Monteiro, T., Pereira, E., Alves, E., Sequeira, A.D. and Franco, N. (2001) Applied Physics Letters, 78, 2137. 393 ODonnell, K.P., Martin, R.W., TragerCowan, C., White, M.E., Esona, K., Deatcher, C., Middleton, P.G., Jacobs, K., van der Stricht, W., Merlet, C., Gil, B., Vantomme, A. and Mosselmans, J.F.W. (2001) Materials Science & Engineering B: Solid State Materials for Advanced Technology, 82, 194. 394 Nakamura, S. and Mukai, T. (1992) Japanese Journal of Applied Physics, 3 (1), L1457. 395 Nakamura, S., Mukai, T. and Seno, M. (1993) Japanese Journal of Applied Physics, 31, L16. 396 Nakamura, S., Iwasa, N. and Nagahama, S. (1993) Japanese Journal of Applied Physics, 32, L338–341. 397 See, for example, Butte, R., Carlin, J.-F., Feltin, E., Gonschorek, M., Nicolay, S., Christmann, G., Simeonov, D., Castiglia, A., Dorsaz, J., Buehlmann, H.J., Christopoulos, S., Baldassarri, G., von Hogersthal, H., Grundy, A.J.D., Mosca, M., Pinquier, C., Py, M.A., Demangeot, F.,
References
398
399
400
401
402 403
404
405
406
Frandon, J., Lagoudakis, P.G., Baumberg, J.J. and Grandjean, N. (2007) Current status of AlInN layers lattice-matched to GaN for photonics and Electronics. Journal of Physics D: Applied Physics, 40, 1–17. Nakamura, S., Senoh, M., Nagahata, S., Iwasa, N., Yamada, T., Matsushita, T., Hiyoku, H. and Sugimoto, Y. (1996) Japanese Journal of Applied Physics, Part 2: Letters, 35 L74. Wright, A.F. and Nelson, J.S. (1995) Firstprinciples calculations for zinc-blende AlInN alloys. Applied Physics Letters, 66 (25), 3465–3467. Onuma, T., Chichibu, S.F., Uchinuma, Y., Sota, T., Yamaguchi, S., Kamiyama, S., Amano, H. and Akasaki, I. (2003) Recombination dynamics of localized excitons in Al1 xInxN epitaxial films on GaN templates grown by metalorganic vapor phase epitaxy. Journal of Applied Physics, 94 (4), 2449–2453. Lukitsch, M.J., Danylyuk, Y.V., Naik, V.M., Huang, C., Auner, G.W., Rimai, L. and Naik, R. (2001) Optical and electrical properties of Al1 xInxN films grown by plasma source molecular-beam epitaxy. Applied Physics Letters, 79 (5), 632–634. Guo, Q., Ogawa, H. and Yoshida, A. (1995) Journal of Crystal Growth, 146, 462. Goano, M., Bellotti, E., Ghillino, E., Garetto, C., Ghione, G. and Brennen, K.F. (2000) Journal of Applied Physics, 88, 6476. Kim, K.S., Saxler, A., Kung, P., Razeghi, R. and Lim, K.Y. (1997) Determination of the band-gap energy of Al1 xInxN grown by metal–organic chemical-vapor deposition. Applied Physics Letters, 71 (6), 800–802. Yamaguchi, S., Kariya, M., Nitta, S., Takeuchi, T., Wetzel, C., Amano, H. and Akasaki, I. (1998) Observation of photoluminescence from Al1 xInxN heteroepitaxial films grown by metalorganic vapor phase epitaxy. Applied Physics Letters, 73, 830. Starosta, K. (1981) Physica Status Solidi a: Applied Research, 68, K55–K57.
407 Kubota, K., Kobayashi, Y. and Fujimoto, K. (1989) Journal of Applied Physics, 66, 2984. 408 Kistenmacher, T.J., Ecelberger, S.A. and Bryden, W.A. (1993) Journal of Applied Physics, 74, 1684. 409 Bedair, S.M., McIntosh, F.G., Roberts, J.C., Piner, E.L., Boutros, K.S. and El-Masry, N.A. (1997) Journal of Crystal Growth, 178, 32. 410 Mohammad, S.N., Salvador, A. and Morkoc, H. (1995) Emerging GaN based devices. Proceedings of the IEEE, 83, 1306–1355. 411 Nagamoto, T., Kuboyama, T., Minamino, H. and Omoto, O. (1989) Japanese Journal of Applied Physics, 28, L1334. 412 Ryu, M-.Y., Chen, C.Q., Kuokstis, E., Yang, J.W., Simin, G. and Asif Khan, M. (2002) Luminescence mechanisms in quaternary AlxInyGa1xyN materials. Applied Physics Letters, 80 (20), 3730–3733. 413 Williams, C.K., Glisson, T.H., Hauser, J.R. and Littlejohn, M.A. (1978) Journal of Electronic Materials, 7, 639. 414 Angerer, H., Brunner, D., Freudenberg, F., Ambacher, O., Stutzmann, M., H€opler, R., Metzger, T., Born, E., Dollinger, G., Bergmaier, A., Karsch, S. and K€orner, H.-J. (1997) Determination of the Al mole fraction and the band gap bowing of epitaxial AlxGa1xN films. Applied Physics Letters, 71 (11), 1504–1506. 415 G€orgens, L., Ambacher, O., Stutzmann, M., Miskys, C., Scholz, F. and Off, J. (2000) Characterization of InGaN thin films using high-resolution X-ray diffraction. Applied Physics Letters, 76 (5), 577–579. 416 Peng, T., Piprek, J., Qiu, G., Olowolafe, J.O., Unruh, K.M., Swann, C.P. and Schubert, E.F. (1997) Band gap bowing and refractive index spectra of polycrystalline AlxIn1xN films deposited by sputtering. Applied Physics Letters, 71 (17), 2439–2441. 417 Bellaiche, L., Wei, S.-H. and Zunger, A. (1997) Bond-length distribution in tetrahedral versus octahedral semiconductor alloys: the case of
j127
j 1 General Properties of Nitrides
128
418
419
420
421
422
423
424
425
426
427
Ga1 xInxN. Physical Review B: Condensed Matter, 56 (21), 13872–13877. Bernardini, F. and Fiorentini, V. (2001) Nonlinear macroscopic polarization in III–V nitride alloys. Physical Review B: Condensed Matter, 64, 085207. Bellaiche, L. (1999) Band gaps of latticematched (Ga,In)(As,N) alloys. Applied Physics Letters, 75 (17), 2578–2580. Bellaiche, L., Wei, S.-H. and Zunger, A. (1997) Band gaps of GaPN and GaAsN alloys. Applied Physics Letters, 70, 3558. Wang, L.-W. (2001) Large-scale localdensity-approximation band gapcorrected GaAsN calculations. Applied Physics Letters, 78 (11), 1565–1567. Kent, P.R.C. and Zunger, A. (2001) Evolution of III–V nitride alloy electronic structure: the localized to delocalized transition. Physical Review Letters, 86 (12), 2613–2616. Vurgaftman, I., Meyer, J.R. and RamMohan, L.-R. (2001) Band parameters for III–V compound semiconductors and their alloys. Journal of Applied Physics, 89 (11), 5815–5875. Morkoc, H. (2001) III-nitride semiconductor growth by MBE: recent issues. Journal of Materials Science, 12, 677–695. Kondow, M., Kitatani, T., Nakatsuka, S., Larson, M.C., Nakahara, K., Yazawa, Y., Okai, M. and Uomi, K. (1997) GaInNAs: a novel material for long wavelength semiconductor lasers. IEEE Journal of Selected Topics in Quantum Electronics, 3, 719–730. Harris, J.S., Jr (2000) Tunable longwavelength vertical-cavity lasers: the engine of next generation optical networks? IEEE Journal of Selected Topics in Quantum Electronics, 6 (6), 1145–1160. Kitatani, T., Nakahara, K., Kondow, M., Uomi, K. and Tanaka, T. (2000) A 1.3-mm GaInNAs/GaAs single-quantum-well laser diode with a high characteristic temperature over 200 K. Japanese Journal of Applied Physics, Part 2: Letters, 39 (2A), L86–L87.
428 Yang, K., Hains, C.P. and Cheng, J.L. (2000) Efficient continuous-wave lasing operation of a narrow-stripe oxideconfined GaInNAs–GaAs multiquantumwell laser grown by MOCVD. IEEE Photonics Technology Letters, 12 (1), 7–9. 429 Kondow, M., Kitatani, T., Nakahara, K. and Tanaka, T. (1999) A 1.3-mm GaInNAs laser diode with a lifetime of over 1000 hours. Japanese Journal of Applied Physics, Part 2: Letters, 38 (12A), L1355–L1356. 430 Li, N.Y., Hains, C.P.,Yang, K., Lu, J., Cheng, J. and Li, P.W. (1999) Organometallic vapor phase epitaxy growth and optical characteristics of almost 1.2 mm GaInNAs three-quantum-well laser diodes. Applied Physics Letters, 75 (8), 1051–1053. 431 Sato, S. and Satoh, S. (1998) Roomtemperature pulsed operation of strained GaInNAs/GaAs double quantum well laser diode grown by metal organic chemical vapour deposition. Electronics Letters, 34 (15), 1495–1497. 432 Kitatani, T., Kondow, M., Nakahara, K., Larson, M.C. and Uomi, K. (1998) Temperature dependence of the threshold current and the lasing wavelength in 1.3-mm GaInNAs/GaAs single quantum well laser diode. Optical Review, 5 (2), 69–71. 433 Nakatsuka, S., Kondow, M., Kitatani, T., Yazawa, Y. and Okai, M. (1998) Index-guide GaInNAs laser diode for optical communications. Japanese Journal of Applied Physics, Part 1: Regular Papers, Short Notes & Review Papers, 37 (3B), 1380–1382. 434 Ougazzaden, A., Bouchoule, S., Mereuta, A., Rao, E.V.K. and Decobert, J. (1999) Room temperature laser operation of bulk InGaAsN/GaAs structures grown by APMOVPE using N2 as carrier gas. Electronics Letters, 35 (6), 474–475. 435 Fischer, M., Reinhardt, M. and Forchel, A. (2000) High temperature operation of GaInAsN laser diodes in the 1.3 mm regime. 58th Device Research Conference, IEEE, pp. 119–120. 436 Shimizu, H., Kumada, K., Uchiyama, S. and Kasukawa, A. (2000) High
References
437
438
439
440
441
442
443
444
performance CW 1.26 mm GaInNAsSbSQW and 1.2 mm GaInAsSb-SQW ridge lasers. Electronics Letters, 36 (20), 1701–1703. Setiagung, C., Shimizu, H., Ikenaga, Y., Kumada, K. and Kasukawa, A. (2003) Very low threshold current density of 1.3mm-range GaInNAsSb–GaNAs 3 and 5 QWs lasers. IEEE Journal of Selected Topics in Quantum Electronics, 9 (5), 1209–1213. Bank, S.R., Yuen, H.B., Bae, H., Wistey, M.A. and Harris, J.S. (2006) Overannealing effects in GaInNAs(Sb) alloys and their importance to laser applications. Applied Physics Letters, 88, article 221115. Gollub, D., Moses, S., Fischer, M. and Forchel, A. (2003) 1.42 mm continuouswave operation of GaInNAs laser diodes. Electronics Letters, 39 (10), 777–778. Ikenaga, Y., Miyamoto, T., Makino, S., Kageyama, T., Arai, M., Koyama, F. and Iga, K. (2002) 1.4 mm GaInNAs/GaAs quantum well laser grown by chemical beam epitaxy. Japanese Journal of Applied Physics, 41, 664–665. Tansu, N. and Mawst, L.J. (2002) Lowthreshold strain-compensated InGaAs(N) (l ¼ 1.19–1.31 mm) quantum well lasers. IEEE Photonics Technology Letters, 14 (4), 444–446. Peng, C.S., Jouhti, T., Laukkanen, P., Pavelescu, E.-M., Konttinen, J., Li, W. and Pessa, M. (2002) 1.32-mm GaInNAs-GaAs laser with a low threshold current density. IEEE Photonics Technology Letters, 14 (3), 275–277. Yang, X., Jurkovic, M.J., Heroux, J.B. and Wang, W.I. (1999) Molecular beam epitaxial growth of InGaAsN:Sb/GaAs quantum wells for long-wavelength semiconductor lasers. Applied Physics Letters, 75 (2), 178–180. Yang, X., Heroux, J.B., Mei, L.F. and Wang, W.I. (2001) InGaAsNSb–GaAs quantum wells for 1.55 mm lasers grown by molecular-beam epitaxy. Applied Physics Letters, 78 (26), 4068–4070.
445 Vurgaftman, I., Meyer, J.R., Tansu, N. and Mawst, L.J. (2003) (In)GaAsN–GaAsSb type-II W quantum-well lasers for emission at l ¼ 1.55 mm. Applied Physics Letters, 83 (14), 2742–2744. 446 Bank, S.R., Wistey, M.A., Goddard, L.L., Yuen, H.B., Lordi, V. and Harris, J.S., Jr (2004) Low-threshold continuous-wave 1.5-mm GaInNAsSb lasers grown on GaAs. IEEE Journal of Quantum Electronics, 40 (6), 656–664. 447 Fischer, M., Reinhardt, M. and Forchel, A. (2000) Room-temperature operation of GaInAsN/GaAs laser diodes in the 1.5 mm range. Conference Digest, 2000 IEEE 17th International Semiconductor Laser Conference (Cat. No. 00CH37092), IEEE, Piscataway, NJ, pp. 115–116. 448 Livshits, D.A., Egorov, Yu.A. and Riechert, H. (2000) 8 W continuous wave operation of InGaAsN lasers at 1.3 mm. Electronics Letters, 36 (16), 1381–1382. 449 Bugge, F., Erbert, G., Fricke, J., Gramlich, S., Staske, R., Wenzel, H., Zeimer, U. and Weyers, M. (2001) 12 W continuous-wave diode lasers at 1120 nm with InGaAs quantum wells. Applied Physics Letters, 79 (13), 1965–1967. 450 Reinhardt, M., Fischer, M., Kamp, M. and Forchel, A. (2000) 7.8 GHz small-signal modulation bandwidth of 1.3 mm DQW GaInAsN/GaAs laser diodes. Electronics Letters, 36 (12), 1025–1026. 451 Steinle, G., Mederer, F., Kicherer, M., Michalzik, R., Kristen, G., Egorov, A.Y., Riechert, H., Wolf, H.D. and Ebeling, K.J. (2001) Data transmission up to 10 Gbit/s with1.3 mmwavelengthInGaAsNVCSELs. Electronics Letters, 37 (10), 632–634. 452 Fischer, M., Reinhardt, M. and Forchel, A. (2000) A monolithic GaInAsN verticalcavity surface-emitting laser for the 1.3-mm regime. IEEE Photonics Technology Letters, 12 (10), 1313–1315. 453 Schneider, H.C., Fischer, A.J., Chow, W.W. and Klem, J.F. (2001) Temperature dependence of laser threshold in an In GaAsN vertical-cavity surface-emitting laser. Applied Physics Letters, 78 (22), 3391–3393.
j129
j131
2 Electronic Band Structure and Polarization Effects Introduction
The band structure of a given semiconductor is pivotal in determining its potential utility. Consequently, an accurate knowledge of the band structure is critical if the semiconductor in question is to be incorporated in the family of materials considered for serious investigations and device applications. The group III–V nitrides are no exception, and it is their direct bandgap nature and the size of the energy gap that spurred the interest in nitride semiconductors. Nitride semiconductors can be classified into two groups. One group pertains to stoichiometric systems where N represents 50% of the constituents while the other half is made of metal constituents. These stoichiometric nitrides come in wurtzitic and zinc blende (ZB) forms. The other class of nitrides is the dilute compound semiconductors, wherein very small amounts of N are added to the lattice for remarkably large negative bowing of the bandgap, making these dilute nitride systems compete for longer wavelength applications. For example, the bandgap of GaAs can be extended to 1.3 mm applications. Likewise, the bandgap of InGaAs coherently grown on GaAs can be extended with dilute amounts of N in the lattice to be a contender for 1.5 mm applications, which has been the domain of In0.53Ga0.47As lattice matched to InP. The impact of dilute nitrides is that, in at least the aforementioned example, what used to be the domain of InP-based technology can be met by GaAs technology with untold consequences in terms of not only technology but also the cost of that technology. A number of researchers have published band structure calculations for both wurtzite (Wz) and zinc blende GaN, AlN, and InN. To make matters more interesting, the bandgap of InN transmogrified from 1.9 eV downward to about 0.7 eV between the first edition and the current one. It is argued that the first set of bandgap measurements might have been conducted in films containing a large amount of O, which could have caused an upward shift in the measured data. The initial estimate of the 1.9 eV bandgap of InN, in addition to creating confusion concerning the nature and applications of InN, caused uncertainties in the bandgap of the InGaN ternary as well. The situation is exacerbated by inhomogeneities in composition and strain as well as poor sample quality. The situation in fact transformed into one in which
Handbook of Nitride Semiconductors and Devices. Vol. 1. Hadis Morkoç Copyright 2008 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim ISBN: 978-3-527-40837-5
j 2 Electronic Band Structure and Polarization Effects
132
multiple bowing parameters, depending on the InN composition in the alloy, were proposed. It can now be said that, with the small bandgap of about 0.8 eV for InN as agreed on, only a single bowing parameter can indeed account for the data dating back to 1980s where incompatibilities with the 1.9 eV InN bandgap were reported [1]. The first Wz GaN band structure found through a pseudopotential method led to a 3.5 eV direct bandgap. The band structure for ZB GaN has been obtained by a firstprinciples technique within the local density functional framework with a direct bandgap of 3.40 eV and a lattice constant of 4.50 Å. A treatise of the band structure in bulk and quantum wells (QWs) with and without strain for conventional nitrides, GaN, InN, and AlN, along with their alloys in the wurtzite and zinc blende form, and dilute nitrides (all of which are zinc blende ternaries and quaternaries, for example, GaAsN and GaInAsN) is discussed. Complete and consistent parameter sets are provided with tabulations of the direct and indirect energy gaps and spin–orbit and crystal field splitting. The alloy bowing parameters, electron and hole effective masses, deformation potentials, elastic constants, and piezoelectric (PE) and spontaneous polarization coefficients are given in many tables in Chapter 1. However, the basis of spontaneous and piezoelectric polarization effects and their practical impact on single- and dual-interface heterostructures are discussed in this chapter following the band structure discussion. The heterostructure band offsets are discussed in Volume 2, Chapter 4. The temperature dependence of the bandgap parameter of nitrides inclusive of the case of band tailing, primarily GaN, is discussed in this chapter as well as in Volume 2, Chapter 5 in conjunction with its optical properties.
2.1 Band Structure Calculations
A glossary of band structure calculations will be given before delving into the specific calculations employed to determine the band structure of nitride semiconductors. This is not an all-inclusive treatment of the field, but it is a compact treatment of the salient features of methods used to illustrate and/or calculate the band structure. The first exposure of many students to band structure calculations is that of a free-electron or nearly free electron approach in a periodic lattice [2–4]. In this one-electron model, the periodic potential can be thought of as arising from the periodic charge distribution associated with ion cores that are situated at the lattice sites. To expand the picture to include many electrons, an average constant potential contribution is added to account for all the other electrons in the system. This problem was first considered by Felix Bloch [5]. In its simplest form, the wave function representing the electrons in a periodic potential would be composed of the product of a plane wave function representing an electron in free space and a function representing the periodicity of the crystal. In the nearly free electron approach, the effect of the crystalline potential on the electronic structure is considered to be weak, and the energy levels of the electrons have little resemblance with those of the atoms. The allowed energies occur in bands of allowed states separated by forbidden energy regions (gaps). Within an allowed and
2.1 Band Structure Calculations
mainly occupied energy band, the electron motion is in many ways similar to that of free particles with an appropriate charge and effective mass. Whether the crystal is an insulator or conductor depends on whether the states within a band or set of bands are completely filled or partially empty. The crystal is considered a semiconductor if the gap between a filled band and the empty band is small, the exact value of which has changed over the years. For example, a gap of 3 eV was taken to be associated with insulators, which is smaller than the gap of GaN. The one-electron picture is an approximation and does not take into consideration processes such as electron– electron interaction, which is neglected other than what is convoluted in the average potential. To understand the basic aspects of band structure, it is instructive to consider an infinite periodic one-dimensional square well potential, which forms the basis for the Kronig–Penney approach [6]. This approximation leads to an exact solution of the Schr€odinger equation. Even though the square well potential approximation is very crude, it serves to illustrate explicitly many important characteristic features of the electron behavior in periodic lattices. In contrast to the free-electron approximation in which the potential energy of the electron is assumed to be small in comparison to its total energy, the opposite is assumed in another commonly practiced approximation, called the tight binding (TB) approximation. Specifically, it is assumed that the potential energy of the electrons accounts nearly for all of the total energy in the case of which the allowed energy bands are narrow compared to the forbidden ones. Unlike the free-electron model, the electronic wave functions are more or less localized around the atoms. Thus, the interaction between neighboring atoms is relatively weak, and the wave functions and the allowed energy levels of the crystal as a whole resemble the wave function and energy levels of isolated atoms. In a sense, an electron associated with an atom is assumed to remain in an orbit associated with that atom, and these orbitals are combined linearly in a form to be consistent with the Bloch–Floquent theorem to represent a state running throughout the crystal. Again, each orbital is localized on a particular atom. As such the results are very sensitive to overlap integrals, which in turn are sensitive to the details of the orbitals outside the cores and to the lattice spacing. Naturally, in this approach, electrons are not affected by atoms more than a single atomic spacing away. The choice of whether the free electron or the tight binding method is good depends on the particular crystal. In fact, in some crystals neither of these is good. The tight binding approach was assumed by Bloch in his original discussion of energy bands. If there is an appreciable interatomic interaction, the tight binding approach must use a linear combination of atomic orbitals (LCAO), in which a quantum mechanical variational procedure is employed to find the combination of s, p, and d orbitals that correspond to the lowest energy in the system. Not surprisingly, the tight binding methods are successful when the effect of the periodic core potential is quite large. This is true, for example, when the band is derived from the 3d states in the first series of transition elements. In these elements, the 3d states are partially filled and one or two electrons are present in 4s subshells. When atoms of these elements are brought together to make a solid, the interaction between the 4s states is very strong while the overlap between the partially filled inner shells is rather weak.
j133
j 2 Electronic Band Structure and Polarization Effects
134
There are other band structure methods that provide much improvement over the nearly free electron model and the tight binding model. These methods rely on choosing an appropriate basis for the electrons that represent electron behavior both inside and outside the atomic sphere. For example, while a single plane wave may be adequate to represent electron wave function in the interstitial space, to account for the rapid change in the function near the core region, a combination of a large number of plane waves would be necessary. Because the core functions are described in terms of radial functions and spherical harmonics, all unknown wave functions of crystals can be expanded in a set of known functions such as plane waves, radial functions, and spherical harmonics. The various methods for band structure calculations, therefore, differ in the initial choice of boundary conditions that the wave functions must satisfy. A brief qualitative discussion of these methods is given in the following sections. 2.1.1 Plane Wave Expansion Method
Here, the Bloch wave is expanded as a linear combination of plane waves (LCOPW), namely, ! ! ! X ! ð Þ:r Ck ðKÞei k þ k : ð2:1Þ Fk ð r Þ ¼ !
k
The combination coefficients Ck(K) are determined by solving the determinantal equation " # X ! ! h2 ðk þ KÞ2 0 E K jCk ðKÞj2 dK;K 0 þ Ck ðKÞCk ðK ÞVð k k 0 Þ ¼ 0; 2m 0 K;K
!
! 0
ð2:2Þ
where EK is the energy eigenvalue and Vð k k Þ is the crystalline potential. Although this is a simple method, its practical implementation is difficult, as it requires a large number of plane waves to represent the behavior of electrons near a core region. Consequently, the convergence in the eigenvalues is poor and it requires solving a large determinantal equation. 2.1.2 Orthogonalized Plane Wave (OPW) Method
The method originally proposed by Herring and Hill [7] and later discussed by Woodruff [8] considers the Bloch function to be a linear combination of OPW basis. The OPW basis consists of a plane wave orthogonalized to the atomic core functions, such that the electron behaves like a core electron while inside the core and like a plane wave while in the interstitial region. The OPW basis can be written as ! ! X 1 ! ! X k ð r Þ ¼ pffiffiffiffiffiffiffiffi ei k : r mkj Fkj ð r Þ; ð2:3Þ NW j
2.1 Band Structure Calculations
where Fkj are Ðthe core wave functions for constant j. The term mkj is evaluated by ! ! requiring that Fkj ð r ÞX k ð r Þd3 r ¼ 0mX , where W is the volume of the Wigner–Seitz cell, and N is the number of atoms. The Bloch function is then expressed as a linear combination of the OPWs, and the energy eigenvalues are computed by solving the appropriate determinantal equation. This method has been applied to band structure calculation of metallic and nonmetallic solids. 2.1.3 Pseudopotential Method
Phillips and Kleinman [9] later demonstrated that it is possible to rewrite the crystal potential that includes contributions from the core and valence electrons in such a way that the Bloch function can be written in terms of a linear combination of planes as in the plane wave expansion method without compromising the convergence advantages of the OPW method [10,11]. The crystal potential thus obtained is called the pseudopotential. As mentioned earlier, the orbitals between the cores (outer regions) are smooth where the wave functions are somewhat like plane waves. Near the core regions, however, the wave functions are complicated by the strong and rapidly varying potential. The orthogonality requirement causes nodes (zeros) in the wave function in the core region. The weak potential experienced by the electrons in the outer region can be treated as a perturbation that mixes the plane wave components (strongly only at the Brillouin zone (BZ) boundaries). While the orbitals are not like plane waves near the cores and potentials vary rapidly, it is argued that what goes on near the core is irrelevant to the dependence of the energy on the wave vector. The energy wave vector dependence can be calculated by applying the Hamiltonian operator to an orbital at any point in space, which when applied to the outer regions, the picture would be that of nearly free electron energy. The actual potential energy in the core region can be represented by an effective potential energy called the pseudopotential that gives the same wave functions outside the core regions as the actual potential. Surprisingly, the pseudopotential is nearly zero in the core region, which is arrived at by experience with such potentials as well as theoretical considerations. Although use of these pseudopotentials may lead to incorrect wave functions, doing so can, with acceptable accuracy, indicate how the energy varies. 2.1.4 Augmented Plane Wave (APW) Method
This method proposed by Slater [12] also makes use of the fact that the wave function inside the core behaves like atomic functions and outside the core like plane waves as in the OPW method. The difference lies in how one applies the boundary conditions. Unlike the OPW method where the wave function outside and inside the core are matched by the Schmidt orthogonalization condition, in the APW scheme one expands the wave functions outside the core region (r ri) by a set of plane waves and
j135
j 2 Electronic Band Structure and Polarization Effects
136
inside the core region (r ri) by a sum of spherical waves, namely, !! X ! Fk ð r Þ ¼ a0 Qðr r i Þei k : r þ alm Qðr i rÞY lm ðq; fÞRl ðE; rÞ; l;m
where Y is the step function with Y(x) ¼ 1 for x 0 and Y(x) ¼ 0 for x < 0. The ! coefficients alm are chosen in such a way that the wave function Fk ð r Þ is continuous across the sphere of radius ri. The term Rl(E, r) represents the solution of the radial equation. Slater [12] proposed a reasonable approximation to the potential with a constant potential outside the cores and an ordinary atomic potential inside a sphere surrounding each ion core. The Schr€odinger equation is then solved separately for each of the two regions and solutions are matched over the spherical boundaries between them, and within each core the wave function can be expanded in spherical harmonics. The solution outside the cores is a superposition of plane waves. The combination gives this method its name, the augmented plane waves. 2.1.5 Other Methods and a Review Pertinent to GaN
Another class of calculations relies on what is called the all-electron approach. Here, one develops potentials within the sphere around the core region and uses their energy derivatives to expand the wave functions in a manner to match those outside the sphere. Among the methods relying on this premise are linear muffin-tin orbital (LMTO) and linearized augmented plane wave (LAPW) methods. Essentially, in the LMTO method, the crystal is divided into nonoverlapping muffin-tin spheres surrounding the atomic sites and an interstitial region outside of spheres. Inside the muffin-tin sphere, the potential is assumed spherically symmetric while in the interstitial region the potential is assumed constant or slowly varying. The early application of LMTO method made use of what is called the atomic sphere approximation (ASA), in which the crystal potential is treated as a superposition of weakly overlapping spherical potentials centered around lattice sites but in such a way as to fill the space and the total volume of muffin-tin sphere is the same as the atomic volume. The potential is also assumed to be spherically symmetric inside each muffin-tin sphere. Additionally, the kinetic energy of the basis function in the interstitial regions is restricted to be a constant and that constant is typically assumed to be zero in the calculations. In open structures, it is customary to include spheres centered on interstitial sites. It is worth noting that the potential so determined is very close to the true full potential and provide a much better representation of the potentials than the original muffin-tin potentials that are constant in the interstitial regions and spherical within nonoverlapping spheres. The limited precision, 0.1 eV, prevents this method from being applied satisfactorily to the very intricate portions of the band structures. For example, very small differences in the total energy between the zinc blende and wurtzitic forms of GaN are a troublesome point for these ASA-LMTO calculations. Likewise, the crystal field splitting at the valence band maxima presents similar problems.
2.1 Band Structure Calculations
The LMTO method has advantages such as using a minimal basis for computational efficiency and thus allowing large unit cell calculations. It also treats all elements in the same manner and is accurate due to an augmentation procedure, which gives the wave function a reasonably correct shape near the nuclei. The method also uses atom-centered basis functions of well-defined angular momentum. The full potential LMTO (FP-LMTO) calculations are all fully relativistic electrons with the shape approximation to the charge density or the potential. As in the LMTO, the crystal is divided into nonoverlapping muffin-tin spheres and interstitial regions outside the spheres. Then, the wave function is presented differently in those two regions in that inside the muffin-tin spheres, the basis functions are as in the LMTO-ASA method and are of Bloch sum of linear muffin-tin orbitals and the kinetic energy is not restricted to zero in the interstitial regions. All of the other above methods rely on our knowledge of the potential in which the electrons inside a crystal move. This potential results from the interaction between electrons, electrons and nuclei, and between nuclei. This presents a many body problem that cannot be solved exactly. Thus, approximations are needed. One of these techniques developed by Hohenberg and Kohn and Sham, known as the density functional theory (DFT) [13,14], has shown a great promise in treating energy bands of solids. DFT can be construed as a self-consistent field method for attaining the crystal potential. Over the years, many band structure calculations evolved to include not only the electronic properties but also binding energies, through which the lattice constant, bulk modulus, elastic constants, and vibrational properties such as phonon frequencies can be computed. The DFT method is well suited for calculating the aforementioned parameters. The basic quantity is the electronic charge density, and the total energy is expressed as functional of this density. The total energy expression as a function of density includes exchange and correlation effects in an average electron gas like manner [15], which is termed as the local density approximation (LDA). Most common form expresses the charge density in terms of occupied oneelectron wave functions, the corresponding eigenvalues of which constitute the band structure [16]. Calculations of the self-energies are best done by the DFT-LDA method within the GW approximation. The nomenclature is adopted from the original paper of Lars Hedin [17] for one-electron Greens function and Hedin and Lundqvist [18] for screened Coulomb interaction. This approximation represents the first term in a perturbation expansion of the self-energy, which can also be viewed as the screened Hartree–Fock (H–F) theory. It should be pointed out that in semiconductors the GW method differs from the band structure determined by LDA chiefly due to a shift of the conduction band relative to the valence band. Consequently, the fundamental bandgap is underestimated. The genesis for Kohn–Sham [15] local DFT lies in lowering the conduction band in a manner that is k-value dependent, as has amply been pointed out by Sham and Schl€ uter [19] and Perdew and Levy [20] in back-to-back articles. It is for this reason that DFT-based calculations are supplemented by a shift of the calculated fundamental gap inspired by a body of experimental data. Most band structure calculations for GaN and related materials rely on the DFT-LDA approach. Computational development of the GW approximation was later accomplished by a
j137
j 2 Electronic Band Structure and Polarization Effects
138
series of authors, namely, Hyberstsen and Louie [21], Godby, Schl€ uter, and Sham [22], and even later on by Aryasetiawan and Gunnarsson [23,24]. Let us now discuss the band structure of some specific nitride semiconductors. This semiconductor family can exist in wurtzite and zinc blende crystal polytypes with the Wz phase being the stable and widely used form. While both the Wz and ZB polytypes of GaN have been given in quite some detail, the ZB varieties of AlN and InN are not thermodynamically stable and reports are very sketchy. In fact, there are predictions that ZB AlN has an indirect bandgap [25]. A review of the band structure calculation and the methodologies used for both polytypes of all group III nitrides, BN, AlN, GaN, and InN, has been given by Lambrecht and Segall [16]. Calculations of electronic and optical properties of Wz GaN and related structures have been undertaken over many years [16,26] with more of them emerging continually, as nitride-based devices become more popular. Methods such as the ab initio, tight binding [27], pseudopotential, ASA LMTO, a general treatment of which has been treated by Andersen [28], LCAO, LAPW, full potential linearized augmented plane wave (FP-LAPW) method such as that reported by Wimmer et al. [29] for calculating the electronic band structure of Wz semiconductors within the LDA [15,16], DFT, and GW methods have been employed to calculate the energy bands for both wurtzite and zinc blende GaN, InN, and AlN bulk materials. Christensen and Gorczyca [30,31] utilized the ASA-LMTO method for group III-nitrides mainly GaN and AlN, albeit in relation to their behavior under pressure. The ASA-LMTO method was also applied by Lambrecht and Segall [32] to contrast and compare the nature of the direct and indirect bandgap of various ZB and Wz nitrides and, in particular, in terms of the directness or indirectness of the bandgap. These methods change in their capabilities to varying degrees. Early versions of pseudopotential calculations did not include the contribution by 3d-electrons because of the difficulty of attaining convergence in-plane wave expansions due to the deep N pseudopotential [33]. A mixed basis set was later used to overcome this apparent shortcoming [34] but with the consequence that ZB GaN would be lower in energy, meaning favored over the Wz phase that is inconsistent with experiments, and also was not confirmed with well-converged plane wave calculations performed for GaN by Yeh et al. [35] and InAlN [36]. Yeh et al. [35] focused on the issue of polymorphism for a large number of semiconductors including AlN, GaN, InN, AlP, AlAs, GaP, GaAs, ZnS, ZnSe, ZnTe, CdS, C, and Si, and using the local density formalism (LDF), developed a simple scaling at T ¼ 0 that systematizes the energy difference (DE LDF W-ZB ) between the ZB and Wz forms. This energy difference was found ~ to be linearly dependent on the atomistic orbital radii coordinate RðA; BÞ that depends only on the properties of the free atoms A and B, making up the binary compound AB. Of special interest for the topic under discussion is that Yeh et al. [35] found that DE LDF W-ZB ðABÞ for GaN is 9.9 meV/atom, which is within 0.7 meV/atom of the calculations by Van Camp et al. [37] and within 0.3 meV/atom of the calculations by Min et al. [34]. The energy difference, DE LDF W-ZB ðABÞ, for all three nitride binaries are negative when this quantity is scaled with differences in tetrahedral radii and Paulings electronegativity, implying that the equilibrium state of all three nitride binaries is the Wurtzitic form.
2.1 Band Structure Calculations
The issue of d-electrons is an important one, as in Ga 3d, in relation to pseudopotential and all-electron calculations, specifically, to know whether these calculations can handle these bands. Because both In and Ga are heavy and have d cores, Ga 3d and In 4d states overlap with the deep N 2s states with serious implications about bonding and band structure. The way in which the d-electrons are treated as core orbitals with or without nonlinear core corrections or as valence states in pseudopotential methods caused some confusion, which has been the topic of some discussion [32,38]. In the case of Fiorentini et al. [38], the structural and electronic properties, albeit cubic GaN, were studied within the local density approximation by the full potential linear muffin-tin orbitals method, wherein the Ga 3d-electrons were treated as band states with no shape approximation to the potential and charge density. Owing to the resonance of Ga 3d-states with nitrogen 2s states, the cation d bands were found not to be inert, and features unusual for a III–V compound were found in the lower part of the valence band as well as in the valence charge density and density of states. Additional full and frozen (T ¼ 0) overlapped core calculations performed for GaN, ZnS, GaAs, and Si (all cubic) showed that an explicit description of closed-shell interaction has a noticeable effect on the cohesive properties of GaN. The resulting energy resonance causes the Ga 3d-electrons to strongly hybridize with both the upper and lower valence band s and p levels. Such hybridization is predicted to have a profound influence on the GaN properties, including quantities such as the bandgap, the lattice constant, acceptor levels, and valence band heterojunction offsets. Because Al has no 3d core states, there is no hybridization between the cation d states and the N 2s states. In short, the band structure and cohesive properties of GaN are very sensitive to the cation d bands. On the pseudopotential side, Wright and Nelson [36] provided a framework in which accurate calculations treating the Ga 3d- and In 4d-electrons explicitly as valence states were performed by extending the plane wave cutoff to 240 Ry to ensure convergence. An interesting observation is that while d-electrons are important in bonding, they appear as separate states in considering quasi-particle excitations in photoemission experiments [39]. Somewhat of a side note but with legitimate relevance, it has been predicted in the cases of ZnS and ZnSe that potential acceptors, such as Cu, whose d-electrons are resonant with the lower valence band, are repelled by the d-hybridized upper valence band, resulting in a deep level. Impurities without d-electron resonance form shallow acceptors. Mg has no d-electrons and turns out to be sufficiently shallow for roomtemperature p-type doping of GaN. On the contrary, Zn, Cd, and Hg, which all have d-electrons, form deep levels in GaN [40]. Further insight is warranted before conclusive statements can be made with certainty as, for example, photoemission data show the N 2s to be well below the Ga 3d band. Pseudopotential calculations can also be applied to defects [41] and surfaces [42], as has been done for GaN. The calculations relating to surfaces are discussed in Section 3.2.7.2 in reasonable detail. Likewise, the calculations in relation to defects are treated in Section 4.3.1. Other applications of the calculations discussed above are for the determination of dielectric properties and susceptibility [43] and vibrational properties [44].
j139
j 2 Electronic Band Structure and Polarization Effects
140
Calculations in the Hartree–Fock approximation rather than the local density approximation have also been performed for nitride semiconductors, notably GaN [45] and AlN [46,47]. While LDA calculations underestimate the bandgap, the opposite is the case for Hartree–Fock calculations. In the H–F method, exchange is treated exactly but the correlations are fully ignored. Specific to the case of GaN, its total energy as a function of unit cell volume has been calculated for the wurtzite, zinc blende, and rock salt phases by the ab initio all-electron periodic Hartree–Fock method by Pandey et al. [45]. In this case, the gallium 3d levels were treated as fully relaxed band states, and the internal parameters c/a and u in the wurtzite phase were optimized. The calculated transition pressure between the wurtzite and rock salt phases were found to be about 52 GPa at the Hartree–Fock level and about 35 GPa at the correlated level. The calculated electronic structure shows strong hybridization of Ga 3d and N 2s states with the ordering as Ga 3d–N 2s–N 2p in all the phases. The results indicate the bandgap to be direct at G in the wurtzite and zinc blende phases and indirect in the high-pressure rock salt phase where the valence band maximum is shifted away from the G point. The electronic structure of Wz AlN has been investigated by means of periodic ab initio Hartree–Fock calculations for the purpose of calculating the binding energy, lattice parameters (a, c), and the internal coordinate or parameter (u) [47]. The values of the bulk modulus, its pressure derivative, the optical phonon frequencies at the center of the Brillouin zone, and the full set of elastic constants have been calculated and compared with experimental data. When ab initio Hartree–Fock calculations were used to determine the electronic structure of AlN in high pressure, the rock salt phase resulted [46]. In this phase, the calculated lattice constant is 3.982 Å with the bulk modulus of 329 GPa. As in the case of GaN, the rock salt phase is predicted to be indirect at the X point with a gap of 8.9 eV. Moreover, the bonding is essentially ionic between Al and N. The direct gap shows a stronger linear dependence on pressure with a pressure derivative of 68 meV GPa1 compared to that of the indirect X-valley gap with a pressure derivative of 31.7 meV GPa1. It should be emphasized that the rock salt phase is favored to exist under high pressure, and as such throughout this book and literature, nitrides are spoken of as if they are wurtzitic with GaN being cubic also when grown away from thermodynamic equilibrium conditions on cubic substrates along h0 0 1i directions. As mentioned above, the electronic properties of nitride semiconductors can more accurately be calculated using first-principles techniques like density functional theory [25] within the Greens function theory with the characteristic GW approximation of the exchange correlation self-energy [17]. These calculations have been applied to Wz and ZB GaN and AlN by Rubio et al. [25] and to ZB GaN by Palummo et al. [48]. The computational complexity of the full GW method is prohibitive for applications to complex systems with large number of atoms, such as surfaces, interfaces, and clusters. However, it should be mentioned that efficient simplified version of the GW method has been reported to reduce the central processing unit (CPU) time by a factor of 100 (in conjunction with semiconductors Si, GaAs, AlAs, and ZnSe) [49,50]. Unless the simplified GW method is used [48,49], the full GW method is typically limited to simple systems, for example, elemental or binary semiconductors.
2.1 Band Structure Calculations
The GW calculations are reasonably consistent with each other and also with experiments in many cases. In the calculations of Rubio et al. [25], the ab initio pseudopotential method within the local density approximation and the quasi-particle approach have been employed to determine the electronic properties of both Wz and ZB phases of AlN and GaN. The quasi-particle band structure energies were calculated using a model dielectric matrix for the evaluation of the electron self-energy. In the zinc blende structure, AlN was predicted to be indirect (G to X) with (4.9 eV) and that GaN to be direct with 3.1 eV at the G point, the latter in good agreement with absorption experiments on cubic GaN, showing the bandgap to be 3.2–3.3 eV. In the calculations of Palummo et al. [48], models of diagonal and off-diagonal screening with LDA-RPA full calculations in cubic GaN were considered. Simplified GW calculations relying on these models were also compared with full GW calculations. At the time empirical pseudopotential calculations were not available, necessitating ab initio RPA calculations to be done within the DFT-LDA approach. These calculations have already been used for obtaining the full GW band structure of GaN [51]. It should be mentioned that with respect to pure LDA results, the valence band shifts down and the conduction band shifts up, resulting in larger bandgap estimation. The amount of downward shift of the valence bands increases with the increase in energy below the valence band maximum. The N2s states are about 1.5–2 eV more than the valence band maximum. Moreover, the bottom of the N2p valence band, the character of which is of a mixture of N2p cations, shifts by an amount of about 0.5 eV more than the maximum. The absolute shift of the valence band maximum is a problem in GW theories, which is also the case with this method for very established materials such as Si. This seems to stem from the choice parameterization used for the LDA starting point of the calculations. The GW method changes the gaps of GaN and AlN by 1 and 2 eV, respectively [16]. The conduction band correction is on the order of 0.1 eV across the k-points and specific states. This figure appears to increase with the increase in energy dealing with higher conduction bands, awaiting further refinement following any comparison with experiments when accurate measurements become available. Unlike the full GW calculations, the TB approach provides an attractive possibility for an extension of the system size accessible to electronic structure calculations with atomic resolution. TB calculations have been applied to nitride-based systems [52–54]. Moreover, the Slater–Koster parameters transferable between the ZB and Wz crystal phases have been treated [27]. There has been a plethora of reports regarding band structure calculations in nitride semiconductors. The calculated band structures of Wz GaN, AlN, and InN are exhibited in Figure 2.1. For semblance of completeness, results from empirical pseudopotential method are also included here. Fritsch et al. [55] investigated the electronic band structure of both the wurtzite and zinc blende group III nitride semiconductors GaN, AlN, and InN within the empirical pseudopotential approach. Using ionic model potentials and a static dielectric screening function derived, the cationic and anionic model potential parameters were obtained from the zinc blende GaN, AlN, and InN experimental data. Using these model potentials, Fritsch et al. [55] calculated the band structure of group III nitrides in both the wurtzite and zinc blende
j141
j 2 Electronic Band Structure and Polarization Effects
142
(a)
GaN
12 10
3
1 1
3
6
Energy ( eV )
6
1
8
3 1
2
4
3
2
1 1 6 5
0 –2
2
4 3 2
3
–4
3 1
–6
1 3 1
3
–8 –10
3
–12
3
–14 –16
3
1 3
3
1 3
A
S H
R L T U
P K
M
1 M
K
M U L
T
(b) 6
3
3
10
H
P K
4 1
3
8
3 3 1
1
6
S
A
R
AlN
12
3
2
1
4 Energy (eV)
3
4
3
2 1 5
4 3 1
3
3 1
6
0 3
–2
2 3
–4
1
–6 –8
A
S
Γ
R L T U
–10 3
–12 –14 M
K
T
3
1
1
3
M U
H P K
M
L
Figure 2.1 Calculated band structures of (a) wurtzite GaN, (b) AlN, and (c) InN in the LDA within the FP-LMTO method at the experimental lattice constant and optimized u-value. The first Brillouin zone is also shown for convenience [16].
R
A
S
H P K
2.1 Band Structure Calculations
(c)
8
6
3 1 3 2
6 4
1 1
1
3 1 1
3
2 Energy (eV)
InN 4 3
1
2
10
1 6
0 2
–2
5
4
1
3 2
3 1 3
–4 –6
3
3
1 3 1
A
S R L T U
–8 3
–10
3
1 3
–12 3
–14 M
K
1 3 T
H P K
M
3 1 M U
L
R
A
S
H P K
Figure 2.1 (Continued )
form, recognizing the necessity of including the anisotropy of wurtzite crystals in the screening function. The band structures so calculated for wurtzitic GaN, AlN, and InN are shown in Figure 2.2. The same for the zinc blende variety is shown in Figure 2.3. It should be noted at the outset that all these binary materials, including alloy compounds obtained by combinations of these binaries, are wide direct bandgap semiconductors in both crystal phases, except zinc blende AlN that is expected to have an indirect gap with the conduction band minimum being at the X valley. Due to the lack of reliable experimental data, many details of these studies must be improved to provide an accurate band description. Approaches such as the kp model [56,57] including strain have been employed to calculate the valence band structure of Wz GaN [56,58]. First-principles calculations of effective mass parameters and valence band structures in bulk and confined systems with and without strain, utilizing the FP-LAPW method [59–61,66] and envelope function formalism for valence bands in wurtzite quantum wells, have been undertaken [62]. The wurtzite structure has a hexagonal unit cell and thus two lattice constants, c and a. It contains six atoms of each type. The space group for the wurtzite structure is P63mc (C46v ) [63]. The wurtzite structure consists of two interpenetrating hexagonal close-packed (HCP) sublattices, each with one type of atoms, offset along the c-axis by 5/8 of the cell height (5c/8). The zinc blende structure has a cubic unit cell, containing four group III elements and four nitrogen elements. The space group for the zinc blende structure is T 2d : F 43m. The position of the atoms within the unit cell is identical to the diamond crystal structure. Both structures consist of two
j143
j 2 Electronic Band Structure and Polarization Effects
144
interpenetrating face-centered cubic sublattices, offset by one quarter of the distance along a body diagonal. Each atom in the structure may be viewed as positioned at the center of a tetrahedron with its four nearest neighbors defining the four corners of the tetrahedron. The zinc blende and wurtzite structures are similar. In both cases, each
Energy (eV)
10
5
0
–5
A
R
L U M
Σ
Γ Δ A
S
H P K
T
Γ
(a)
15
Energy (ev)
10
5
0
–5
A
R
L U M
Σ
Γ Δ A
S
(b) Figure 2.2 Band structure of (a) wurtzitic GaN, (b) wurtzitic AlN, and (c) wurtzitic InN along high-symmetry lines in the Brillouin zone calculated within the empirical pseudopotential method (EPM), using ionic model potentials obtained experimentally from zinc blend varieties. Courtesy of Daniel Fritsch et al. [55].
H P K
T
Γ
2.1 Band Structure Calculations
j145
Energy (eV)
10
5
0
–5
A
R
L U M
Σ
Γ Δ A
S
H P K
Γ
T
(c) Figure 2.2 (Continued )
Energy (eV)
15 L3
10
K1 X3
L1
5
X1 L3
0
X5 X3
L2
–5
L
Λ
Γ
Δ
X
(a) Figure 2.3 Band structure of (a) zinc blende GaN, (b) zinc blende AlN, and (c) zinc blende InN along high-symmetry lines in the Brillouin zone calculated within the empirical pseudopotential method (EPM), using ionic model potentials obtained experimentally. Courtesy of Daniel Fritsch et al. [55].
K1
K2 K1 K1 UK
Σ
Γ
j 2 Electronic Band Structure and Polarization Effects
146
20
15 Energy (eV)
L3 10
X1
5
0
L3
–5
L2
10
K1
K2
X5
K1
X3 Λ
L
(b)
K1
X3
L1
Γ
Δ
K1 X
U,K
Σ
Γ
L3
Energy (eV)
K1 X3
5
X1
K1
L1 0
K2
L3
X5
K1
L2
–5 L
K1
X3 Λ
Γ
Δ
X
U, K
Σ
Γ
(c) Figure 2.3 (Continued )
group III atom is coordinated by four nitrogen atoms. Conversely, each nitrogen atom is coordinated by four group III atoms. The main difference between these two structures lies in the stacking sequence of closest packed diatomic planes. For the wurtzite structure, the stacking sequence of the (0 0 0 1) planes is ABABAB in the h0 0 0 1idirection. For the zinc blende structure, the stacking sequence of the (1 1 1) planes is ABCABC in the (1 1 1) direction.
2.1 Band Structure Calculations
The structure and the first Brillouin zone of a wurtzite and zinc blende crystal along with the irreducible wedges, calculated using the LDA within the FP-LMTO method at the experimental lattice constant and optimized u value, are displayed in Figure 2.4a and b, respectively. In a crystal with Wz symmetry, the conduction band kz
So
A o
R
Δo
T o
Γ
H L o
o
S’
oP
U K
o
o
ky
T’
Σ
M
kx (a)
kz
L U Q
S Z
kx
K
X
ky
W
(b) Figure 2.4 (a) Structure and the first Brillouin zone of a wurtzite crystal. Schematics of the irreducible wedges of Wz structure, indicating the high-symmetry points and lines, are also shown. The Umin point of the Wz phase is located on the M–L line at two-thirds a distance away
from the M point. (b) Structure and the first Brillouin zone of a zinc blende crystal. Schematics of the irreducible wedges of ZB structure indicating the high-symmetry points and lines are also shown.
j147
j 2 Electronic Band Structure and Polarization Effects
148
Conduction band
E gA E0
E0
C6v
or
5
9
A
6
7
7
7
Valence band
9
, J = 3/2
Valence band
B cr
15
so 15
1 7
ZB
W Crystal field cr = so
W Spin orbit
7
C
W
8 , J = 1/2
Crystal field
1
=
ZB
ZB Spin orbit cr = 0 so
=
2
=
3
Figure 2.5 Schematic representation of the splitting of the valence band in Wz crystals due to crystal field and spin–obit interaction. From left to right, the crystal field splitting is considered first. From right to left, the spin–orbit splitting is considered first. Regardless of which is considered first, the end result is the same in that there are three valence bands that are sufficiently close to one another for band mixing to be nonnegligible.
wave functions are formed of the atomic s orbitals, which transform the G point congruent with the G7 representation of the space group C 46v , The upper valence band states are constructed out of appropriate linear combinations of products of p3-like (px-, py-, and pz-like) orbitals with spin functions. Under the influence of the crystal field and spin–orbit interactions, the hallmark of the wurtzite structure, the sixfold degenerate G15 level associated with the cubic system, splits into Gv9 , upper Gv7 and lower Gv7 levels (Figure 2.5). Figure 2.6 shows the dispersion of the uppermost valence and conduction band structures in Wz GaN and ZB GaN ((a) near the G band for Wz, (b) inclusive of M, L, and A minima in Wz, and (c) inclusive of G, L, and X minima in ZB GaN). The influence of the crystal field splitting, which is present only in the wurtzite structure, transforms the semiconductor from ZB to Wz, which is represented in the section on the left-hand side in Figure 2.5. The crystal field splits the G15 band of the ZB structure into G5 and G1, states of the wurtzite structure. These two states are further split into Gv9 , upper Gv7, and lower Gv7 levels by spin–orbit interactions. Application of the spin–orbit splitting, from right to left, splits the G15 band of the ZB crystal into G8 and G7 states while the crystal possesses the zinc blende symmetry. Application of a crystal field further splits these states into Gv9 , upper Gv7, and lower Gv7 levels, and the crystal now possesses the wurtzite symmetry.
2.1 Band Structure Calculations
E(k)
c
Γ7
Γ9 :HH Δ1 Γ7:LH Γ7:ΧΗ
kz
(a)
Γ
Figure 2.6 (a) Schematic representation of the G point valence and conduction bands in crystal with wurtzite symmetry, such as GaN, where the spin–orbit splitting leads to the bands labeled as HH and LH. The one caused by splitting due to crystal field is labeled as CH [59,60]. (b) Schematic representation of the band diagram for Wurtzite GaN showing the separation between the G, A, and M–L band symmetry points at 300 K. The values with respect to the top of the valence band are EG ¼ 3.4 eV, EM–L ¼ 4.5–5.3 eV, EA ¼ 4.7–5.5 eV,
kx , ky Eso ¼ 0.008 eV, Ecr ¼ 0.04 eV [56]. The values of EG ¼ 6 eV, EM–L ¼ 7 eV, and EA ¼ 8 eV are given by Fritsch et al. [55]. (c) Schematic representation of the band diagram for zinc blende GaN showing the separation between the G, X, and L band symmetry points at 300 K. The values with respect to the top of the valence band are EG ¼ 3.2 eV, EL ¼ 4.8–5.1 eV, Ex ¼ 4.6 eV, and Eso ¼ 0.02 eV. Note that in the ZB structure, the valence band is degenerate [56]. The values of EG ¼ 3.2 eV, EL ¼ 5.1 eV, EX ¼ 4.3 eV are given by Fritsch et al. [55].
j149
j 2 Electronic Band Structure and Polarization Effects
150
Energy
A valley
M and L valleys
valley
EA
EM–L
E Ecr
HH band LH band
kz
k x,y
Split–off band (b) Energy L valleys X valley
EL
valley
EX E <1 0 0>
Eso
HH band
<111>
LH band Split–off band (c) Figure 2.6 (Continued )
Literature values of the calculated and experimental critical point transition energies for wurtzitic GaN, AlN, and InN are tabulated in Tables 2.1–2.3. Literature values of the calculated and experimental critical point transition energies for zinc blende GaN, AlN, and InN are tabulated in Tables 2.4–2.6. As shown in Figure 2.6a for wurtzitic GaN, the hole effective masses of the three uppermost valence bands Gv9 , Gv7 , and Gv7 exhibit large k-dependence. The bands are labeled as HH (heavy Gv9 ), LH (light, upper Gv7 spin–orbit split) and CH (Gv7 , crystal field split). The mass of the Gv9 band is heavy in all k-directions, whereas that of the upper Gv7 is relatively light in the x- and y-planes but heavy in the z-direction. That of the lower Gv7 is light in the x- and y-planes, but it is heavy along the z-direction
2.1 Band Structure Calculations Table 2.1 Literature values of calculated and experimental critical point transition energies for wurtzitic GaN [55].
A (eV) Parameter
Anisotropic
Isotropic
B (eV)
M v2 Mc1 Mv4 Mc1 Mv4 Mc3 Gv6 Gc1 Gv3 Gv6 Gv1 Gv6 Gv5 Gv6 Gv5 Gc3 Gv5 Gc6 Hv3 Hc3 K v3 K c2 K v2 K c2
7.67 6.07 7.68 3.47 6.97 0.043 1.00 5.96 10.74 8.06 8.54 8.68
7.67 6.07 7.68 3.47 6.94 0.023 1.00 5.96 10.74 8.07 8.55 8.68
8.26 6.61 7.69 3.50 6.80 0.021
9.0 9.43 10.10
C (eV)
3.0 7.0 0.0 1.0 5.9 11.1 8.3 7.9
D (exp) (eV) 7.05 7.0a 7.05 3.6, 3.44a, 3.50a 7.0a 0.022a 5.3 9.4 7.9a 7.65, 9.0a
The term aniso represents the values derived using a band structure calculation with anisotropically screened model potentials, whereas the term iso describes a comparative band structure calculation on the basis of isotropically screened model potentials using an averaged e0 value by taking the spur of the dielectric tensor. A: empirical pseudopotential calculation from Ref. [55]; B: ab initio pseudopotential calculation within local density approximation from Ref. [25]; C: LCAO within local density approximation from Ref. [64]; D: unless stated otherwise, the experimental values are taken from Ref. [65]. a Experimental values taken from Ref. [66].
(c-direction). Two different definitions are prevalent in the literature. The G6, Gl pair has been used in Refs [63,71] and the G5, G1 pair in Refs [62,72,73]; for background information on group theory and symmetries in physics, see Refs [74,75]. We should mention that a carryover habit from the zinc blende nomenclature is still used for wurtzite symmetry by referring to the crystal field split-off band with the nomenclature SO as if it is the spin–orbit split-off band, because it happens to be the farthest from the HH band. In the zinc blende symmetry, the crystal field splitting is nonexistent, making the top of the valence band degenerate, and the spin–orbit splitting is large. Portions of this book, unfortunately, participate in the misuse of this nomenclature. Shown in Figure 2.6b are the most pertinent bands near the zone center and A, M, and L valleys. The same for zinc blende GaN is shown in Figure 2.6c. Without the spin–orbit interaction, the valence band would consist of three doubly degenerate bands: HH, LH, and CH bands. The spin–orbit interaction removes this degeneracy and yields six bands. Some band calculations based on the empirical pseudopotential method (EPM) [66] or the empirical tight binding method (ETBM) [76] have shown this splitting to be about 10 meV near the G point, which is comparable to the energy separation of the split-off band in GaN. In early attempts, the general Hamiltonian in kp theory included the spin–orbit interaction,
j151
j 2 Electronic Band Structure and Polarization Effects
152
Table 2.2 Literature values of calculated and experimental critical point transition energies for wurtzitic AlN [55].
A (eV) Parameter
Anisotropic
Isotropic
B (eV)
M v2 Mc1 M v4 Mc1 M v4 Mc3 Gv6 Gc1 Gv3 Gv6 Gv1 Gv6 Gv5 Gv6 Gv5 Gc3 Gv5 Gc6 Hv3 Hc3 K v3 K c2 K v2 K c2
9.56 7.88 8.81 6.11 6.44 0.13 1.04 8.95 12.99 10.10 9.43 9.59
9.54 7.87 8.83 6.11 6.41 0.16 1.03 8.94 12.97 10.91 9.43 9.57
10.0 8.3 8.5 6.0 6.7 0.2 0.9 9.4 14.0 10.5 9.6 9.7
C (exp) (eV)
6.29
8.02 14.00 10.39
The term aniso represents the values derived using a band structure calculation with anisotropically screened model potentials, whereas the term iso describes a comparative band structure calculation on the basis of isotropically screened model potentials using an averaged e0 value by taking the spur of the dielectric tensor. A: empirical pseudopotential calculation from Ref. [55]; B: ab initio pseudopotential calculation within local density approximation from Ref. [25]; C: experimental values taken from Ref. [67].
but it took a while for it to be applied to the calculations of the band structure of wurtzite materials such as GaN. Naturally, the band structures of wurtzitic and zinc blende polytypes are very distinct due to the differences in underlying symmetries. For the zinc blende case, the three Luttinger parameters and the spin–orbit splitting provide a minimal description of the valence band structure. Moreover, the energy gap and interband coupling strength are also required for complete parameterization of both the conduction and valence bands. The split-off hole mass can be treated as an independent parameter within the commonly used eight-band kp model. The increase in the electron effective mass due to interactions with higher conduction bands can be included via the F parameter (see Section 2.2 for details) [72,77–79]. The set of band parameters needed to describe the wurtzite lattice must be augmented due to its lower symmetry. Neglecting the effect of valence band and upper conduction bands on the electron effective mass allows one to omit the interband matrix element and the F parameter. Owing to the reduced symmetry, the electron mass can display a rather weak anisotropy. In contrast, a full description of the valence band in the wurtzite polytype band structure requires both the spin–orbit splitting, Dso, and the crystal field splitting, Dcr, along with the seven so-called A parameters. Analogous to the Luttinger parameters in zinc blende materials, the latter parameterizes the hole masses along the different directions. Figure 2.6a and b highlights the differences in wurtzite and zinc blende varieties in terms of their band structure.
2.1 Band Structure Calculations Table 2.3 Literature values of calculated and experimental critical point transition energies for wurtzitic InN [55].
A (eV) Parameter
Anisotropic
Isotropic
B (eV)
d (exp) (eV)
M v2 Mc1 Mv4 Mc1 Mv4 Mc3 Gv6 Gc1 Gv3 Gv6 Gv1 Gv6 Gv5 Gv6 Gv5 Gc3 Gv5 Gc6 Hv3 Hc3 K v3 K c2 K v2 K c2
7.30 5.94 6.71 2.58 5.63 0.214 0.90 5.22 10.16 7.34 8.13 8.60
7.23 5.88 6.70 2.59 5.50 0.084 0.89 5.18 10.12 7.36 8.12 8.50
6.65 5.05 5.80 2.04 5.77 0.017 1.05 4.65 8.74 6.51 7.38 7.20
7.3, 4.95a 7.3 2.11, 2.0a;b
5.0, 5.5, 4.7a 8.8, 8.9a 5.4a 7.3, 7.2a
The term aniso represents the values derived using a band structure calculation with anisotropically screened model potentials, whereas the term iso describes a comparative band structure calculation on the basis of isotropically screened model potentials using an averaged e0 value by taking the spur of the dielectric tensor. A: empirical pseudopotential calculation from Ref. [55]; B: empirical pseudopotential calculation from Ref. [66]. a Experimental values taken from Ref. [68]. b It should be pointed out that the values of bandgap values listed in this table compares with 1.8–2.1 eV values reported during the early stages of InN development and 0.7–0.8 eV reported later on. For an in-depth discussion of this seemingly controversial bandgap determined experimentally, the reader is referred to Section 2.9.1.
Table 2.4 High-symmetry point energies in zinc blende GaN in
reference to the top of the valence band for cases where spin–orbit effects are neglected (included) [55]. Parameter G15 c Gc1 Gv15 X c3 X c1 X v5 X v3 Lc3 Lc1 Lv3 Lv2
(Gc7 ) (Gc6 ) (Gv8 ) (X c7 ) (X c6 ) (X v7 ) (X v6 ) (Lc4;5 ) (Lc6 ) (Lv4;5 ) (Lv6 )
A (eV)
B (eV)
C (eV)
10.098 3.308 0.000 6.010 4.428 2.459 6.294 10.416 5.149 0.834 6.812
10.300 3.383 0.000 6.805 4.571 2.693 6.149 9.916 5.636 0.931 6.743
10.248 3.213 0.000 6.265 4.585 2.086 5.923 10.606 5.510 0.772 6.644
A: empirical pseudopotential calculation from Ref. [55]; B: empirical pseudopotential calculation from Ref. [69]; B: empirical pseudopotential calculation from Ref. [70].
j153
j 2 Electronic Band Structure and Polarization Effects
154
Table 2.5 High-symmetry point energies in zinc blende AlN in
reference to the top of the valence band for cases where spin–orbit effects are neglected (included) [55]. Parameter G15 c Gc1 Gv15 X c3 X c1 X v5 X v3 Lc3 Lc1 Lv3 Lv2
(Gc7 ) (Gc6 ) (Gv8 ) (X c7 ) (X c6 ) (X v7 ) (X v6 ) (Lc4;5 ) (Lc6 ) (Lv4;5 ) (Lv6 )
A (eV)
B (eV)
12.579 5.840 0.000 8.794 5.346 2.315 5.388 12.202 8.264 0.718 6.251
13.406 5.936 0.000 10.661 5.102 2.337 5.262 12.014 9.423 0.728 6.179
A: empirical pseudopotential calculation from Ref. [55]; B: empirical pseudopotential calculation from Ref. [69].
Table 2.6 High-symmetry point energies in zinc blende InN in
reference to the top of the valence band for cases where spin–orbit effects are neglected (included) [55]. Parameter G15 c Gc1 Gv15 X c3 X c1 X v5 X v3 Lc3 Lc1 Lv3 Lv2
(Gc7 ) (Gc6 ) (Gv8 ) (X c7 ) (X c6 ) (X v7 ) (X v6 ) (Lc4;5 ) (Lc6 ) (Lv4;5 ) (Lv6 )
A (eV)
B (eV)
9.722 2.112 0.000 6.416 5.187 1.555 4.303 10.168 4.733 0.480 4.667
10.193 1.939 0.000 7.392 2.509 1.408 4.400 8.060 5.818 0.456 5.200
A: empirical pseudopotential calculation from Ref. [55]; B: empirical pseudopotential calculation from Ref. [69].
2.2 General Strain Considerations
Strain–stress relationship or Hookes law can be used to describe the deformation of a crystal ekl, due to external or internal forces or stresses sij, X sij ¼ Cijkl ekl ; ð2:4Þ k;l
2.2 General Strain Considerations
where Cijkl is the fourth ranked elastic tensor and represents the elastic stiffness coefficients in different directions in the crystal, which due to the C6v symmetry can be reduced to a 6 · 6 matrix using the Voigt notation: xx ! 1, yy ! 2, zz ! 3, yz, zy ! 4, zx, xz ! 5, xy, yx ! 6. The elements of the elastic tensor can be rewritten as Cijkl ¼ Cmn with i, j, k, l ¼ x, y, z and m, n ¼ 1, . . . , 6. With this notation, Hookes law can be reduced to si ¼
X
Cij ej :
ð2:5Þ
j
or as treated in Ref. [80] for C6v symmetry, we have 2
3 2 sxx C11 6 syy 7 6 C12 6 7 6 6 szz 7 6 C13 6 7 6 6 sxy 7 ¼ 6 0 6 7 6 4 syz 5 4 0 szx 0
C12 C22 C13 0 0 0
C13 C13 C33 0 0 0
0 0 0 C44 0 0
0 0 0 0 C55 0
32 3 0 exx 6 7 0 7 76 eyy 7 6 7 0 76 ezz 7 7; 6 7 0 7 76 exy 7 0 54 eyz 5 ezx C66
ð2:6Þ
with C66 ¼ C11 2 C12 . If the crystal is strained in the (0 0 0 1) plane, and allowed to expand and constrict in the [0 0 0 1] direction, the szz ¼ sxy ¼ syz ¼ szx ¼ 0, sxx 6¼ 0, and syy 6¼ 0, and the strain tensor has only three nonvanishing terms (with C11 ¼ C22), namely,
sxx ; syy szz
¼
C11 þ C12 2C13
C 13 C 33
exx ; eyy ; ezz
ð2:7Þ
with exx ¼ eyy ¼ a a0a0 and ezz ¼ c c0c0 ¼ CC1333 ðexx þ eyy Þ, the latter of which describes the Poisson effect, and a and a0 and c and c0 represent the in-plane and out-ofplane lattice constants of the epitaxial layer and the relaxed buffer (substrate), respectively. The above assumes that the in-plane strain in x- and y-directions is identical, namely, exx ¼ eyy. When the crystal is uniaxially strained in the (0 0 0 1) c-plane and free to expand and constrict in all other directions, szz is the only nonvanishing stress term, and the strain tensor is reduced to 1 eyy C 12 C33 C213 ð2:8Þ ¼ 2 exx : ezz C13 C11 C33 C 11 C13 C12 C13 If the growth is performed on the ð1 1 0 0Þm-plane, meaning the growth plane is the xz-plane with the growth direction being along the y-axis, the in-plane strain anisotropy dictates that exx 6¼ eyy . The out-of-plane stress, syy ¼ 0, which when utilized in the stress–strain relationship of Equation 2.6 leads to eyy ¼
C12 exx þ C13 ezz : C11
ð2:9Þ
If the growth is performed on the ð1 1 2 0Þa-plane, meaning the growth plane is the yz-plane with growth direction being along the x-axis, the in-plane strain anisotropy
j155
j 2 Electronic Band Structure and Polarization Effects
156
dictates that exx 6¼ eyy. The out-of-plane stress, sxx ¼ 0, which when utilized in the stress–strain relationship of Equation 2.6 leads to exx ¼
C12 eyy þ C13 ezz : C 11
ð2:10Þ
The values of the elastic stiffness coefficients for GaN have been measured by Sheleg and Savastenko [81] and reproduced in Ref. [59]. These data are listed in Table 1.24. The inversion of the 6 · 6 elastic constant matrix in Equation 2.6 leads to the elastic compliance constants as follows: S11 ¼
C11 C 33 C213 ; ðC 11 C12 Þ½C33 ðC 11 þ C12 Þ 2C213
S12 ¼
C12 C 33 C213 ; ðC 11 C12 Þ½C33 ðC 11 þ C12 Þ 2C213
S13 ¼
C13 ; ½C33 ðC11 þ C 12 Þ 2C 213
S33 ¼
C11 þ C12 ; ½C33 ðC11 þ C 12 Þ 2C 213
S44 ¼
1 : C44
ð2:11Þ
Through the use of the aforementioned compliance constants, a very useful figure of merit can be determined, which in turn would lead to the directional hardness and the reciprocal Youngs modulus as a function of orientation to the crystal axis. For 0 hexagonal symmetry, the reciprocal Youngs modulus S11 along an arbitrary direction at an angle y with respect to the [0 0 0 1] axis is given by [82–84] 0
S11 ¼ S11 sin4 q þ S33 cos4 q þ ðS44 þ 2S13 Þsin2 qcos2 q:
ð2:12Þ
0
In Figure 2.7a and b, polar plots of S 11 as a function of the direction in reference to the [0 0 0 1] axis and for directions along the basal plane are shown for InN, GaN, and AlN binaries. Clearly, AlN and GaN are harder than InN by more than a factor of 2. The hardness of AlN is almost isotropic, whereas that for GaN and InN exhibit some preferential softness along the [0 0 0 1] and ½2 1 1 0 axes. Of paramount interest here is that the hardness of all the wurtzitic binaries is isotropic in the basal plane. This takes on a special meaning as the strain, with growth along the c-axis, caused in epitaxial heterostructures by mismatch lattice and thermal mismatch is along the basal plane. Lack of any force in the growth direction and the fact that the crystal can relax freely in this direction leads to a biaxial strain e1 ¼ e2, which in turn causes stresses s1 ¼ s2
with s3 ¼ 0:
The internal strain is defined by the variation of the internal parameter under strain, (u u0)/u0. In the limit of small deviations from the equilibrium, Hookes law
2.2 General Strain Considerations
[1 0 1 0] 8 S 11 [10 –12 m 2 /N–1]
(a)
6 In N 4
GaN
2 AlN –8
–6
–4
–2
0
0
[1 2 1 0] 2
4
6
8
–2 –4 –6 - –8 [0 0 0 1] 8
S 11 [10 –12 m 2 / N–1]
(b)
6 InN
4
GaN
2
AlN –8
–6
–4
–2
[2 1 1 0]
q 0
0
2
4
6
8
–2 –4 –6 - –8 Figure 2.7 (a) The reciprocal Youngs moduli, S0 11, in the basal plane of InN, GaN, and AlN indicating AlN and GaN are harder than InN to be harder by more than a factor of 2. Of paramount importance, the hardness of the wurtzitic crystals is isotropic in the basal plane. (b) The reciprocal
Youngs modulus, S0 11, along an arbitrary direction making an angle y with respect to the c-axis in the [0 0 0 1] direction. In basal plane the stiffness of AlN is isotropic whereas GaN and InN show preferential softness along the [0 0 0 1] and [2 1 1 0] directions [84].
gives the corresponding diagonal stress tensor s with the elements [85] sxx ¼ syy ¼ ðC11 þ C12 Þexx þ C13 ezz ; szz ¼ 2C13 exx þ C33 ezz :
ð2:13Þ
In Equation 2.13, four of the five independent stiffness constants Cij of the wurtzite crystal are involved. The modifications of Equation 2.13 by the built-in electric field due to the spontaneous and piezoelectric polarization are neglected, as the effect is small.
j157
j 2 Electronic Band Structure and Polarization Effects
158
In the case of uniaxial stress, for example, along the c-direction, there is an elastic relaxation of the lattice in the c-plane. The ratio of the resulting in-plane strain to the deformation along the stress direction is expressed by the Poissons ratio, which in general can be anisotropic. For the wurtzite lattice subjected to a uniaxial stress szz parallel to the c-axis, sxx ¼ syy ¼ 0 holds. Then Equation 2.13 gives the relation exx ¼ ½C13 =ðC 11 þ C12 Þezz ¼ nezz ;
ð2:14Þ
with n ¼ C13 =ðC 11 þ C12 Þ being the Poissons ratio. The uniaxial stress correlates with strain along the direction of the stress by the Youngs modulus E as szz ¼ Eezz and thus E ¼ C33
2C 213 : C11 þ C 12
ð2:15Þ
A homogeneous biaxial stress in the basal plane is described by a constant force in the plane with sxx ¼ syy and vanishing force in the c-direction szz ¼ 0. The Hookes law of Equation 2.13 leads to a relationship between axial and basal plane strain components as ezz ¼ RBexx, which reproduces Equation 2.7 with the biaxial relaxation coefficient being RB ¼
2C13 : C33
ð2:16Þ
The biaxial relaxation coefficient is also referred to as simplylater in this chapter in the polarization section. The in-plane stress–strain relationship using the biaxial modulus is sxx ¼ Yexx, which leads to Y ¼ C11 þ C12
2C 213 : C33
ð2:17Þ
The strain–stress relationship along the c-axis is sxx ¼ ðY=RB Þezz :
ð2:18Þ
We can then relate the Young modules E (is also commonly described by nomenclature as Yo) to biaxial modulus Y as E¼
C33 Y 2n or E ¼ B Y: C11 þ C12 R
ð2:19Þ
Equation 2.13 can now be expressed as sxx ¼
E ezz : 2n
ð2:20Þ
In the case of hydrostatic pressure sxx ¼ syy ¼ szz ;
ð2:21Þ
and from the Hookes law ezz ¼ RH exx ;
ð2:22Þ
2.3 Effect of Strain on the Band Structure of GaN
with RH expressed as RH ¼
C11 þ C12 2C13 : C33 C 13
To calculate elastic stiffness constants, Wagner and Bechstedt [85] considered C11 þ C12 as an independent quantity and made use of the relation between the elastic constants and isothermal bulk modulus, B0 (nomenclature Bs is also used): B0 ¼
ðC11 þ C12 ÞC13 2C213 : C11 þ C12 þ 2C 33 4C13
ð2:23Þ
Equation 2.22 can be obtained from Equation 2.13 with the aid of linearized relation DP/B0 ¼ DV/V0 ¼ 2(e11 þ e33) or Dp/B0 ¼ DV/V0 ¼ 2(e11 þ e33), where DV is the variation of volume with pressure and V0 is the static volume. The values of the bulk modulus B0 have been obtained by fitting the Vinet equation of state to the calculated dependence of energy on volume [86,87], using 1 Y ¼ 2 þ 1=2 ð2:24Þ 4 RB B 0 ; v and C13 ¼ Y=½ð1=vÞ RB :
ð2:25Þ
Wagner and Bechstedt [85] obtained the absolute values of the elastic stiffness coefficients. The strain- and stress-related issues represent the cornerstone of the discussion on piezoelectric polarization, and the above discussed section provides sufficient material to embark on the discussion of polarization issues in nitride semiconductor heterostructures.
2.3 Effect of Strain on the Band Structure of GaN
The strain in conventional group III–V semiconductors has been a much desired feature for its beneficial effects [88]. In the world of GaN, however, it is not necessarily a desirable commodity but could be construed as a nemesis brought upon by the lack of lattice- and thermal-matched substrates and uncomfortably large lattice and thermal mismatch with its ternaries. It is therefore imperative that strain effects be considered. Figure 2.8 exhibits the valence band structure of GaN in the x- and y-planes under biaxial compressive strain and uniaxial strain in the c-plane with the direction of strain as in Figure 2.8c. There are no major changes in the HH, LH, and CH bands, other than crystal splitting becoming larger, with the hole effective mass remaining heavy, the density of states staying high, and the crystal symmetry remaining the same, C46v . In contrast, the uniaxial strain in the c-plane causes an anisotropic energy splitting in the x- and y-planes, which leads to a symmetry lowering from C 46v to C2v . When a compressive uniaxial strain is induced along the y-direction, the HH band in the x-direction and the LH band in the y-direction move to higher energies. This causes a
j159
j 2 Electronic Band Structure and Polarization Effects
160
(a)
(b)
E
E
(c) Biaxial eff so
eff cr
Uniaxial z
kx
ky
kx
ky
y x
Figure 2.8 The valence band structure of GaN under (a) biaxial strain in the c-plane, (b) uniaxial strain in the c-plane, and (c) schematic of the particulars of the strain [59,89].
reduction in the density of states. A tensile uniaxial strain along the x-direction has the same effect. On the contrary, when a tensile uniaxial strain is induced along the ydirection, the HH band in the x-direction and the LH bands along the y-direction move to lower energies. This causes a reduction in the density of states. A compressive strain along the x-direction has the same effect [89].
2.4 kp Theory and the Quasi-Cubic Model
The conduction and valence bands of nitride semiconductors are comprised of s- and p-like states, respectively. Unlike the conventional ZB III-N semiconductors and the lack of a high degree of symmetry, the crystal field present removes the degeneracy at the top of the conduction band. Moreover, unlike the ZB case, the spin–orbit splitting is very small and makes all three bands in the valence band closely situated in energy. Consequently, the three valence bands and the conduction band must be considered in unison and this makes the use of an 8 · 8 kp Hamiltonian imperative. Because the bandgaps of nitrides are very large, the coupling between the conduction and valence bands can be treated as a second-order perturbation, which allows the 8 · 8 Hamiltonian to be split into one 6 · 6 Hamiltonian dealing with the valence band and another 2 · 2 dealing with the conduction band [59]. As indicated above, the conduction band is made of s-like states, which means that it can be treated as parabolic with the dispersion relation EðkÞ ¼ E c0 þ
2 k2z h
==
2mc
þ
2 ðk2x þ k2y Þ h 2m? c
== þ a? c ðexx þ eyy Þ þ ac ðezz Þ;
ð2:26Þ
==
where a? c and ac represent the in-plane and out-of-plane deformation potentials, respectively. For an isotropic parabolic conduction band, Equation 2.26 reduces to EðkÞ ¼ E c0 þ
2 k2 h þ ac e: 2mc
ð2:27Þ
2.4 kp Theory and the Quasi-Cubic Model
Ec0 is the conduction band energy at the k ¼ 0 point, e is the strain, and ac is the deformation potential for the conduction band. The other terms have their usual meanings. It should be pointed out that we are dealing with a linear system. Using the basis jY 11 ">; jY 11 #>; jY 10 ">; jY 10 #>; jY 11 ">; jY 11 #>, the 6 6 Hamiltonians can be expressed as 0 1 F 0 H 0 K 0 B 0 G D H 0 K C B C B H D l 0 I 0 C B C; ð2:28Þ B 0 H 0 l D I C B C @ K 0 I D G 0 A 0 K 0 I 0 F where F, G, l, D, H, I, and K are defined as (two forms are given by Ren et al. [90] and Albrecht et al. [91]) F ¼ D1 þ D2 þ l þ q; G ¼ D1 D2 þ l þ q; l ¼ A1 k2z þ A2 ðk2x þ k2y Þ þ D1 ezz þ D2 ðexx þ eyy Þ; q ¼ A3 k2z þ A4 ðk2x þ k2y Þ þ D3 ezz þ D4 ðexx þ eyy Þ; H ¼ iA6 kz ðk2x þ k2y Þ1=2 A7 ðk2x þ k2y Þ1=2 ; H ¼ ðiA6 kz A7 Þðkx þ iky Þ þ iD6 ðexz þ ieyz Þ; I ¼ iA6 kz ðk2x þ k2y Þ1=2 þ A7 ðk2x þ k2y Þ1=2 ; I ¼ ðiA6 kz þ A7 Þðkx þ iky Þ þ iD6 ðexz þ ieyz Þ; K ¼ A5 ðk2x þ k2y Þ; K ¼p A5ffiffiffiðkx þ iky Þ2 þ D5 ðexx eyy þ i2exy Þ; D ¼ 2D3 ;
½91 ½92 ½91 ½92 ½91 ½92
ð2:29Þ
where Ai is the valence band parameter corresponding to the Luttinger parameters in the ZB system, D1 and D2,3 are the crystal field and spin–orbit splitting energies (D2, ¼ D3 ¼ Dso), Di parameters represent the deformation potentials for the valence band, and exx, eyy, and ezz are the strain tensors (also referred to as e11, e22, and e33 in this book and many other publications). Both forms used by Albrecht et al. [91] and Ren et al. [90] for H, I, and K parameters are given. The former uses the basis and Hamiltonian in the (kx þ iky) form and contains all of the shear strain information. The latter does the phase rotation to get two 3 · 3 and the relevant matrix elements are then functions only of k-transverse and all of the directional phase information has been compiled onto the basis states by a unitary transformation. For presentation of the pseudomorphic strain, either form would be just as easy to use. Another important point is that in the case where the shear strain is nonzero, such as when uniaxial in-plane strain, the D5 and D6 terms must be included as represented by the Albrecht et al. [91] notation above. If only biaxial strain is considered, the shear terms of the strain tensor are zero and thus the D5 and D6 terms vanish. In the latter case, the representation by Ren et al. [90] and others similar to it hold. A good description of pertinent issues is discussed in Ref. [62]. Similar to the cubic system, the 6 · 6 matrix can be block diagonalized into two 3 · 3 matrices, and this can considerably simplify the band structure calculation. It should
j161
j 2 Electronic Band Structure and Polarization Effects
162
be mentioned that A7 can be assumed nearly zero due to symmetry considerations. Chuang and Chang [58] derived the two 3 · 3 Hamiltonians when A7 is neglected and obtained three doubly degenerate bands. Ren et al. [90] investigated the effect of A7 parameter on the valence band dispersion in wurtzitic crystals such as GaN. In fact, Ren et al. [90] argued that theories forwarded by Chuang and Chang [58] replicates that of Bir and Pikus [72] reported decades earlier. Choosing A7 ¼ 0 reduces the results to that of Chuang and Chang [58] and Sirenko et al. [62]. To underscore the effect of A7 parameter, Ren et al. [90] compiled data from empirical pseudopotential method (EPM) [67]. The same method was also applied to wurtzitic and zinc blende phases of all the three binaries of nitrides by Fritsch et al. [55]. Ren et al. [90] calculated the band structure for GaN for values of A7 ¼ 93.7 meV Å and 0 as, shown in Figure 2.9. Clearly, inclusion of the A7 parameter results in a much better fitting between the kp theory and EPM calculations. From the (a) 0.00
Wz GaN
–0.01
HH
Energy (eV)
–0.02 LH –0.03
–0.04 CH
–0.05 0.0
0.050
0.100
0.15
Wave vector, kx,( k// ) (1 Å–1) Figure 2.9 (a) Valence band structures of wurtzite GaN with the kp theory fitting including the spin–orbit interaction with A7 ¼ 93.7 meVÅ in the solid line. The dash-dotted line is the result of fitting with A7 ¼ 0. The empirical pseudopotential method (EPM) calculation data (o) are from Ref. [67] ([90]). (b) Valence band
structure of wurtzite GaN, using the parameters recommended by Ren et al. Courtesy of I. Vurgaftman and J. Meyer and Ref. [90]. (c) The same using the parameters recommended in Ref. [152]. The dashed lines represent the case for A7 ¼ 0. Courtesy of I. Vurgaftman and J. Meyer and Ref. [90].
2.4 kp Theory and the Quasi-Cubic Model
(b) 0.00
Wz GaN
HH
–0.01
Energy(eV)
LH –0.02
–0.03
–0.04 CH –0.05
–0.06 –0.15
–0.10 k z , (k )
–0.050
0.0
0.050
0.100
0.15
-1 Wave vector (1 Å ) kx, (k//)
(c) 0.00
Wz GaN (c)
HH
–0.01 LH
Energy(eV)
–0.02
–0.03
–0.04 CH –0.05
–0.06 –0.10
–0.15
k z , (k ) Figure 2.9 (Continued )
–0.050
0.0
0.050
Wave vector (1 Å-1)
kx, (k//)
0.100
0.15
j163
j 2 Electronic Band Structure and Polarization Effects
164
Table 2.7 Fitted splitting energies and Luttinger-like parameters
for the valence band of the wurtzite GaN: Di parameters are in unit of meV, and the Rydberg terms are in units of h2 =2m0 except for A7, which is in unit of meV Å. D1
D2
D3
A1
A2
A3
A4
A5
A6
A7
21.1
3.61
3.61
7.21
0.440
6.68
3.46
3.40
4.9
93.7
quality of the fit, one can argue that the 93.7 value for the Luttinger-like parameter A7 is a good one. The other parameters giving this fit are listed in Table 2.7. In addition to the cycled parametric values of A parameters by kp theorists, a good extraction of A1–A7 parameters from empirical pseudopotential method band structures of AlN, GaN, and InN can be found in Ref. [55]. Vurgaftman and Meyer [152] followed a nearly identical path to that of Ren et al. [90] but calculated the band structure in both kx and ky crystal directions, as shown in Figure 2.9. The dashed lines, as in the case of the paper by Ren et al. [90], correspond to the case representing the case of A7 ¼ 0. In addition, Vurgaftman and Meyer [152] and Vurgaftman et al. [92] mentioned the calculation of the valence band structure for GaN using parameters representing the properties of GaN that they prefer, the results of which are also shown in Figure 2.9 for comparison. The effective masses are calculated using the parallel and perpendicular hole masses m//and m? that can be expressed in terms of their dependence on the Luttinger-like parameters Ai as follows: ==
m0 =mhh ¼ ðA1 þ A3 Þ; ==
m0 =mlh ¼ ðA1 þ A3 Þ; == m0 =mso
ð2:30Þ
¼ A1 :
?
Here, m represents the mass in the (kx, ky) plane, which means kz ¼ 0 and m0 =m? hh ¼ ðA2 þ A4 A5 Þ; 2 m0 =m? lh ¼ ðA2 þ A4 A5 Þ 2A7 =jD1 j ;
m0 =mh? so
¼
ð2:31Þ
A2 þ 2A27 =jD1 j;
where m//is along the kz-direction (kx ¼ ky ¼ 0) and m? in the (kx, ky) plane, which means kz ¼ 0. The effective masses are calculated using the parallel and perpendicular hole masses m//and m? together with Luttinger-like parameters using Equations 2.30 and 2.31. As indicated in the schematic of Figure 2.5, both the spin–orbit and the crystal field splitting affect the structure of the valence band in wurtzitic crystals [93]. Typically, the relevant parameters are correlated to one another as Dso ¼ 3D2 ¼ 3D3, in spite of the fact that a small D2/D3 anisotropy has sometimes been reported [94,95] and Dcr ¼ D1. Experimentally, the splitting parameters are obtained from the energy differences of the A, B, and C free excitons, which have nonlinear dependencies on the various splittings [96]. It should be pointed out that the nomenclature for the three valence
2.4 kp Theory and the Quasi-Cubic Model
bands for hexagonal system is A, B, and C for HH, LH, and SO (CH) bands when including A7 terms, because spin splitting and strain can significantly alter as to which band of eigenstates is heavy or light at various k-values, particularly in the c-plane. An early experimental undertaking by Dingle et al. [97] led to Dcr ¼ 22 meV and Dso ¼ 11 meV. An analysis by Gil et al. [98] yielded Dcr ¼ 10 meV and Dso ¼ 18 meV. Chuang and Chang [58] attempting to rederive these parameter from the same data but with what was termed as a more precise description of the effect of strain on the valence band edge energies arrived at values of Dcr ¼ 16 meV and Dso ¼ 12 meV. Reynolds et al. [99] obtained Dcr ¼ 25 meV and Dso ¼ 17 meV from a fit to exciton energies, with A and B determined by photoluminescence (PL) and C determined by reflection but with a geometry not fully ideal in terms in that some error is introduced in the value of C exciton energy. Again, using exciton energies values of Dcr ¼ 22 meV and Dso ¼ 15 meV were obtained by Shikanai et al. [100], Dcr ¼ 37.5 meV and Dso ¼ 12 meV by Chen et al. [101], Dcr ¼ 9 meV and Dso ¼ 20 meV by Korona et al. [102], and Dcr ¼ 9–13 meV and Dso ¼ 17–18 meV by Campo et al. [103] and Julier et al. [104]. The values of Dcr ¼ 10 meV and Dso ¼ 17 meV were determined by both Edwards et al. [105] and Yamaguchi et al. [96]. Noticeable is one of the smallest reported crystal field splittings to date, Dcr ¼ 9 meV, along with Dso ffi 18 meV reported by Rodina et al. [106] based on detailed experimental investigation. On the theoretical side, an ab initio calculation by Wei and Zunger [107] overestimates the crystal field splitting Dcr ¼ 42 meV, but arrives at a Dso ¼ 13 meV that agrees well with the experimental data. Suzuki et al. [56] reported Dcr ¼ 40 meV and Dso ¼ 8 meV or Dso ¼ 3D2 ¼ 3D3 1.16 mRy and Dcr ¼ D1 ¼ 5.36 mRy for these splittings. Many firstprinciples calculations focusing on the valence band splitting are available in the literature [58,90,67,154,108]. The experimental data, however, appear to converge on the splittings Dcr ffi 10 meV and Dso ffi 17 meV, as suggested by Vurgaftman and Meyer and tabulated in Table 2.8 [152]. The spin splitting of the valence band of wurtzitic GaN can be determined via the A7 parameter, as shown in Figure 2.9a, which is derived by Vurgaftman and Meyer [152] assuming the parameters of Ren et al. [90]. On the contrary, Figure 2.9b is derived by Vurgaftman and Meyer [152] using A parameters from Ren et al. [90] combined with what is believed to be the more representative spin–orbit and crystal field splittings. Modification of the A parameters alone, but with the corrected values of the splitting energies, does not allow the recovery of the band structure resembling that shown in Figure 2.9a. This simply implies that the field is not yet settled on a set of reliable parameters and more refinement is needed for the most appropriate values of the A parameters to be arrived. To be sure six distinct valence band deformation potentials, as well as the strain tensor and the overall hydrostatic deformation potential, are necessary to describe the band structure of GaN under strain. In the cubic approximation, these can be expressed in terms of the more familiar av, b, and d potentials [93]. In terms of the calculation and combined calculation and measurement efforts, Christensen and Gorczyca [31] reported a hydrostatic deformation potential a ¼ 7.8 eV, which has been shown to agree well with 8.16 eV obtained by Gil et al. [98]. On the contrary, a somewhat lower value of a ¼ 6.9 eV was attained through an ab initio calculation by
j165
j 2 Electronic Band Structure and Polarization Effects
166
Table 2.8 Recommended band structure parameters for wurtzitic GaN from Ref. [152].
Parameter
Value
Parameter
Value
Parameter
Value
Eg (eV, low temperature) a (meV K1) b (K)
3.510
A1
7.21
D1 (eV)
3.7
0.909 (1 in [119]) 830 (1100 in [119]) 10 17 0.20 0.20 4.9 11.3
A2 A3
0.44 6.68
D2 (eV) D3 (eV)
A4 A5 A6 A7 (meV Å) d13 (pm V1) d33 (pm V1) d15 (pm V1) Psp (C m2)
3.46 3.40 4.90 93.7 1.6b 3.1b 3.1b 0.034
D4 (eV) D5 (eV) D6 (eV) c11 (GPa) c12 (GPa) c13 (GPa) c33 (GPa) c44 (GPa)
Dcr (meV) Dso (meV) == me =m0 m? e =m0 a1 (eV) a2 (eV)
4.5 8.2 4.1 4.0 5.5 390 145 106 398 105
367a 135a 103a 405a 95a
See Tables 2.27 and 2.28 for details related to the elastic constants, piezoelectric constants, and spontaneous polarization charge. Any dispersion among the tables is a reflection of the uncertainty in the available parameters. See Volume 2, Chapter 5 for an extended discussion of Varshni parameters. a The second column figures for the Cii parameters are from Table 2.28 where a more expanded list of elastic coefficients is given. b Table 2.28 provides additional data on d-parameters.
Kim et al. [109]. Noting that the hydrostatic potential is anisotropic, owing to the reduced symmetry of the wurtzite crystal, Wagner and Bechstedt [85] calculated values of 4.09 and 8.87 eV for the two hydrostatic interband deformation potentials. On the transport side, fits to the experimental mobility data [110,111] yielded a conduction band deformation potential approaching 9 eV, a topic discussed in some detail in Volume 2, Chapter 3. Using pressure-dependent optical transition energies with pressure, Shan et al. [112] reported a1 ffi 6.5 eV and a2 ffi 11.8 eV and uniaxial deformation potentials b1 ffi 5.3 eV and b2 ffi 2.7 eV for the two hydrostatic interband components. Employing photoreflectance measurements on compressively strained M-plane GaN films (grown along the h1 0 1 0i direction) grown on g-LiAlO2 (1 0 0), Ghosh et al. [113] reported a1 ¼ 3.1 eV with a2 ¼ 11.2 eV. Again, using M-planeoriented GaN layers, Gil and Alemu [114] obtained a1 ¼ 5.22 eV with a2 ¼ 10.8 eV. Vurgaftman and Meyer [152] recommend a set of a1 ¼ 4.9 eV and a2 ¼ 11.3 eV, which represents an average of all the measured values. Numerous sets of valence band deformation potentials have been derived from both first-principles calculations [58,60,61,115,116] and fits to experimental data [94,98,100,96,112–114,117]. The dispersion among the reported data is unacceptable, which calls for further work to resolve the discrepancies and converge on accurate parameters. If one were to average the deformation potentials that are most widely quoted, values of D1 ¼ 3.7 eV, D2 ¼ 4.5 eV, D3 ¼ 8.2 eV, D4 ¼ 4.1 eV, D5 ¼ 4.0 eV, and D6 ¼ 5.5 eV, which satisfy the quasi-cubic approximation that is about to be discussed [58], are obtained. The deformation potential values for GaN are
2.5 Quasi-Cubic Approximation
tabulated in Table 2.8, and their effect on optical transitions are discussed in detail in Volume 2, Chapter 5.
2.5 Quasi-Cubic Approximation
The genesis of the quasi-cubic approximation relies on the fact that the Wz and ZB structures are both tetrahedrally coordinated and hence are closely related. The nearest neighbor coordination is the same for Wz and ZB structures but differs at the next nearest neighbor positions. The basal plane (0 0 0 1) of the Wz structure corresponds to one of the (1 1 1) planes of the ZB. When the in-plane hexagons are lined up in Wz and ZB structures, the Wz [0 0 0 1], ½1 1 2 0, and ½1 1 0 0 planes are parallel to the ZB [1 1 1], ½1 0 1, and ½1 2 1 planes, respectively. This, in turn, leads to correlations between the symmetry direction and the k-points for the two polytypes. There are, however, twice as many atoms in the Wz unit cell as there are in the ZB one. In addition to the band structure similarities between the doubled ZB and Wz structures, one can establish a correlation between the Luttinger parameters in the ZB system and parameters of interest in the Wz system by taking the z-axis along the [1 1 1] direction and the x- and y-axes along the ½1 1 2 and ½ 1 1 0 directions. For details regarding the symmetry relations between the ZB and Wz polytypes, refer to Refs [32,118]. Doing so leads to D2 ¼ D3 ; A1 ¼ A2 þ 2A4 ;pffiffiffi A3 ¼ 2A4 ¼ 2A6 4A5 ; A7 ¼ 0; D1 ¼ D2 þ 2D4 ;pffiffiffi D3 ¼ 2D4 ¼ 2D6 4D5 :
ð2:32Þ
The A parameters can be related to the classical Luttinger parameters gi through A1 A2 A3 A4 A5 A6
¼ ðg 1 þ 4g 3 Þ; ¼ ðg 1 2g 3 Þ; ¼ 6g 3 ; ¼ 2g 3 ; ¼ p ðg ffiffi2ffi þ 2g 3 Þ; ¼ 2ð2g 2 þ g 3 Þ:
ð2:33Þ
The calculated values of the spin–orbit and crystal field splitting parameters with those deduced from the observation of A, B, and C excitons are listed in Tables 2.8 and 2.9, which will be presented shortly. The calculations agree well in terms of the spin–orbit splitting, but the theoretical crystal field splitting is much too large compared to experimental data. The discrepancy may be due to the unaccounted residual strain and strain inhomogeneities present in GaN films, as well as the inaccuracy of the parameter values. The debate will probably continue until strainfree or homogeneously strained films can be prepared.
j167
j 2 Electronic Band Structure and Polarization Effects
168
Table 2.9 Effective masses and band parameters for wurtzitic GaN.
Parameter Aniso ==
me m? e == mhh == mlh == mch m? hh m? lh m? ch A1 A2 A3 A4 A5 A6 A7 D1
0.138 0.151 2.000 2.000 0.130 2.255 0.191 0.567 7.692 0.575 7.192 2.855 2.986 3.360 0.160 0.043
Iso
A
B
0.138 0.151 2.007 2.007 0.130 2.249 0.261 0.317 7.698 0.600 7.200 2.816 2.971 3.312 0.171 0.023
0.20 0.18 1.10 1.10 0.15 1.65 0.15 1.10 6.56 0.91 5.65 2.83 3.13 4.86
0.20 0.18 1.76 1.76 0.16 1.61 0.14 1.04 6.27 0.96 5.70 2.84 3.18
a
0.039
C
0.14 0.15 1.479 1.479 0.130 1.592 0.299 0.252 7.706 0.597 7.030 3.076 3.045 4.000 a 0.194 0.038 0.022
D
E
0.14 0.15 1.453 1.453 0.125 1.595 0.236 0.289 7.979 0.581 7.291 3.289 3.243 4.281 0.179 0.022
0.19 0.17 1.76 1.76 0.14 1.69 0.14 1.76 7.14 0.57 6.57 3.30 3.28
F
0.19 0.17 1.96 1.96 0.14 1.87 0.14 1.96 7.24 0.51 6.73 3.36 3.35 4.72 0 0 0.021 0.021
G
H
I
0.19 0.19 0.23 0.19 1.89 2.00 1.96 1.89 2.00 1.96 0.12 0.16 0.16 2.00 2.04 1.20 0.15 0.18 0.16 0.59 1.49 1.96 7.21 6.4 6.36 0.44 0.50 0.51 6.68 5.9 5.85 3.46 2.55 2.92 3.40 2.56 2.60 4.9 3.06 3.21 0.094 0.108 0 0.021 0.036
Effective masses in units of free-electron mass m0, Luttinger-like parameters Ai (i ¼ 1, . . ., 6) in units of h2 =2m0 , and A7 in units of eV Å . The crystal field splitting energy D1 is given in units of meV. The term aniso represents the values derived using a band structure calculation with anisotropically screened model potentials, whereas the term iso describes a comparative band structure calculation on the basis of isotropically screened model potentials using an averaged e0 value by taking the spur of the dielectric == tensor. Here, me and m? e represent the effective electron masses along and perpendicular to the c-axis [55]. Anisotropically screened and isotropically screened values are from Ref. [55]. A: FP-LAPW band structure calculations are from Ref. [56], and effective mass parameters are obtained through a 3D fitting procedure within cubic approximation; B: FP-LAPW band structure calculations are from Ref. [56], and effective mass parameters are obtained by direct line fit; C: Ai from Ref. [153] obtained through a Monte Carlo fitting procedure to the band structure and effective masses calculated using Equations 2.30 and 2.31; D: direct kp calculations for Ai from Ref. [153] and effective masses calculated using Equations 2.30 and 2.31; E: effective masses and Ai from Ref. [67] obtained through a line fit to the band structure; F: direct kp calculation in a 3D fit from Ref. [67]; G: Ai obtained through a direct fit from Ref. [90] and effective masses calculated using Equations 2.30 and 2.31; H: direct fit of Ai to firstprinciples band structure calculations from Ref. [154]; I: Ai and effective masses obtained in the quasicubic model from zinc blende parameters from Ref. [154]. a A7 in the range of 0.136 eV Å has been set to zero.
The parameters mentioned in bandgap-related discussion for wurtzitic GaN are tabulated in Table 2.8 [152] for the wurtzitic phase GaN. All conventional nitridesinthe wurtzite phase exhibit a directenergy gap, and the next satellite conduction valley, which is the M valley, is some 2 [31]to 5 eV [56]higher than the G valley. In addition to the one made available here, the wurtzite indirect gap related issues have been amply discussed in the literature, some of which can be found in Refs [31,53,75,120–122]. Sufficing it to state the bottom of the conduction band in GaN can be well approximated by a parabolic dispersion relation, although a slight anisotropy is expected due to the reduced lattice symmetry [56]. In many devices, the pertinent property of the band structure is the region near the bottom of the conduction band, which can be represented to a great deal with effective mass. In relatively early
2.6 Temperature Dependence of Wurtzite GaN Bandgap
experimental studies in GaN grown by hydride VPE, Barker and Ilegems [123] obtained an electron effective mass of mn ¼ 0.20m0 from reflectivity measurements. Again early on, Rheinlander and Neumann [124] inferred 0.24–0.29m0 for the effective mass from a Faraday-rotation investigation of heavily n-doped GaN. Using heavily doped samples, which was the norm then, and fits to the thermoelectric power, Sidorov et al. [125] obtainedelectroneffective masses of0.1–0.28m0, dependingon whatprimary scattering channel was assumed. For a review of early investigations of these and other properties, thereaderisreferredtoreviewsfromthe1970s,suchastheonebyPankoveetal.[126]and that by Kesamanly [127]. Congruent with the increased activity in GaN, fuelled by the device demonstrations, particularly LEDs and later on lasers, a substantial body of work has since produced more precise estimations of the electron mass. Among them are the works by Meyer et al. [128] and Witowski et al. [129] who obtained masses of 0.236m0 and 0.222m0, respectively, utilizing shallow donor transition energies; the latter is with the smallest error bars quoted in the literature (0.2%). Underscoring the importance of the polaron correction, which is about 8% in GaN and comes about because of the strong polar nature of GaN, Drechsler et al. [130] derived a bare mass of 0.20m0 from cyclotron resonance data. A similar result was obtained by Perlin et al. [131] using infrared reflectivityandHalleffectmeasurements,whichalsoledtoananisotropyoflessthan1%. For comparison, a slightly larger dressed mass of 0.23m0 has been obtained by Wang et al. [132] and Knap et al. [133]. A small downward correction may be necessary in the former,astheelectronswereconfinedataninterface.Thelatterauthors,however,appear to have corrected for that effect. Using n-type bulk GaN, which does not require an appreciable correction that is needed in confined systems and employing infrared ellipsometry measurements, Kasic et al. [134] reported slightly anisotropic electron masses of 0.237 0.006m0 and 0.228 0.008m0 along the two axes. Again, using modulation-doped structures, a series of authors, Elhamri et al. [135], Saxler et al. [136], Wong et al. [137], Wang et al. [138], and Hang et al. [139], also reported on the effective mass, the values of masses for which ranged from 0.18m0 to 0.23m0 from Shubnikov–de Haas data. Elhamri et al. [135] suggested that strain effects, which are somewhat difficult to be certain of, could have compromised somewhat the masses reported in some of these reports. A value of 0.20m0 is very commonly used for the bare electron effective mass and 0.22m0 for the experimentally relevant dressed mass. A more in-depth discussion of the cyclotron and Shubnikov–de Haas measurements can be found in Volume 2, Chapter 3. This bare mass figure of 0.20m0 agrees reasonably well with a number of estimates based on theory, as outlined in a list in Ref. [140]. Owing to the large uncertainty, no attempt is made to specify an F parameter for wurtzite GaN. However, the interband matrix element may be obtained from the relation between the electron mass and the relevant zone center energies [92].
2.6 Temperature Dependence of Wurtzite GaN Bandgap
The temperature dependence of the bandgap in semiconductors is often described by an imperial expression (assuming no localization)
j169
j 2 Electronic Band Structure and Polarization Effects
170
EðTÞ ¼ Eð0Þ aT 2 =ðb þ TÞ:
ð2:34Þ
In the case of localization, which can also be construed as band tail effect, the temperature dependence deviates from the above equation. In the framework of the band tail model and Gaussian-like distribution of the density of states for the conduction and valence band, the temperature-dependent emission energy could be described by the following modified expression [141], which is based on a model developed for Stokes shift in GaAs/AlGaAs quantum wells [142]. EðTÞ ¼ Eð0Þ ½aT 2 =ðb þ TÞ ½s2 =ðkTÞ;
ð2:35Þ
where the last term represents the localization component with s indicating the extent of localization or band tailing, which is nearly imperative for In-containing alloys. The values of the parameters a (in units of energy over temperature) and b (in units of temperature), for wurtzitic GaN, are listed in Volume 2, Table 5.1. Although a detailed discussion of these parameters in very high-quality samples is deferred to Volume 2, Chapter 5, the evolution of them is discussed here to give the reader a flavor that when the sample quality is under question, the fits to experiments could lead to varying if not erroneous parameters. In concert with this approach, the spread in the values of a and b for A exciton is also discussed in the text surrounding Volume 2, Table 5.3. Varshni parameters have been deduced from the measured variation of the A, B, and C excitonic energies with temperature early on by Monemar [143] with Varshni parameters of a ¼ 5.08 · 104 meV T1 and b ¼ 996 T, the sign for the latter of which is contradictory to the agreed upon values deduced from high-quality samples. In chronological order of the reports, using optical absorption measurements on bulk single crystals and also epitaxial layers grown on sapphire, Teisseyre et al. [144] reported a ¼ 0.939–1.08 meV K1 and b ¼ 745–772 K; note the positive sign of b. Shan et al. [145] reported a ¼ 0.832 meV K1 and b ¼ 836 K deduced from the temperature variation of the A exciton resonance. Petalas et al. [146] determined the Varshni parameters to be a ¼ 0.858 meV K1 and b ¼ 700 K using spectroscopic ellipsometry. Relying on PL measurements, Salvador et al. [147] obtained a ¼ 0.732 meV K1 and b ¼ 700 K. Using absorption measurements, Manasreh [148] reported a ¼ 0.566–1.156 meV K1 and b ¼ 738–1187 K on samples grown by MBE and OMVPE. Using a variation of electroreflectance, the contactless electroreflectance, Li et al. [149] led to a ¼ 1.28 meV K1 and b ¼ 1190 K for the A exciton transition energy. Utilizing PL spectra of excitonic transitions, Zubrilov et al. [150] reported values of a ¼ 0.74 meV K1 and b ¼ 600 K based on exciton luminescence spectra. PL data of free and bound excitons were fitted by Reynolds et al. [151] to a modified Varshni-like form that resulted in a ¼ 0.5 meV K1 and b ¼ 1060. Some of the dispersion in the reported values can be attributed to the difficulty in identifying and resolving various excitonic transitions. Vurgaftman and Meyer [152], averaging what they term as more credible results, recommend a ¼ 0.909 meV K1 and b ¼ 830 K. A detailed discussion of the temperature dependence of the bandgap of GaN along with the Varshni parameters for all three excitons (A, B, and C) are listed in Volume 2, Table 5.3, where A exciton related parameters are a ¼ 1 meV K1 and b ¼ 1100 K.
2.6 Temperature Dependence of Wurtzite GaN Bandgap
The complete Ai parameters calculated by Fritsch et al. [55] as well as others deduced from alternative methods are tabulated in Table 2.9. In addition, a compilation of the dispersion in the effective mass for both the conduction band and various valence bands as obtained by various computational methods as well as parameters used in the description of the bandgap for wurtzitic GaN, particularly, in the context of empirical pseudopotential method, as described in Ref. [55], are also included in Table 2.9. Suzuki and Uenoyama [59] have determined the deformation potentials by the fullpotential linearized augmented plane wave (FLAPW) calculations. The values recommended for GaN by Vurgaftman and Meyer [152] are tabulated in Table 2.8. In the calculations of Suzuki and Uenoyama [59], biaxial and uniaxial strains have been introduced and reduced shifts in the G point energy to which a linear fit in terms of strain was obtained. From the linearfit,the deformation potential values for thevalence band were deduced. The figures obtained from the quasi-cubic approximation are listed too. The good agreement between the calculated values and those determined from the quasi-cubic model is strikingly obvious, which is indicative of the excellence of the quasi-cubic approximation. Assuming an approximately spherical potential in the neighborhood of the N atoms, of the two spin states, the higher energy is in the one in which the electron spin and the orbital angular momentum are parallel. This result is also anticipated on the basis of the atomic spin–orbit splitting in which the P1/2 state is known to have energy higher than the P3/2 state. The contributions of spin–orbit interaction and the crystal field perturbation to the experimentally observed splittings (E1,2 and E2,3) have been calculated with different linear combination of atomic orbitals (LCAO) approximations [155,156]. The large effective mass and the small dielectric constant of GaN, relative to more conventional group III–V semiconductors, lead to relatively large exciton binding energies and make excitons, together with large exciton recombination rates, clearly observable even at room temperature. The bottom of the conduction band of GaN is predominantly formed from the s levels of Ga, and the upper valence band states from the p levels of N. Even though sophisticated methods have been introduced and discussed on these pages, the method of Hopfield and Thomas [157], which treats the wurtzite energy levels as a perturbation to the zinc blende structure is discussed briefly, as it provides a physical picture of band splitting in the valence band. Using the quasi-cubic model of Hopfield [98], one obtains E 1 ¼ 0; v" ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi # u dþD u dþD 2 2 t E2 ¼ dD ; þ 2 2 3 ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi v" # u 2 dþD u d þ D 2 dD ; t E3 ¼ 2 2 3
ð2:36Þ
ð2:37Þ
ð2:38Þ
j171
j 2 Electronic Band Structure and Polarization Effects
172
where Dcr and Dso represent the contributions of uniaxial field and spin–orbit interactions, respectively, to the splittings E1,2 and E2,3.
2.7 Sphalerite (Zinc blende) GaN
The zinc blende GaN crystal consists of two interpenetrating face-centered cubic lattices, one having a group III element atom, for example, Ga, and the other a group V element atom, for example, N. The matrix element of the momentum operator between the conduction and valence bands has been expressed by Kane [92,158] in terms of a single parameter P whose value is termed as EP, which is an important parameter. The other parameter of importance is the F parameter, again defined by Kane, which comes about from a second-order perturbation theory and takes into account the higher order band contributions to the conduction band. Their importance aside [92] EP and F are inherently difficult to determine accurately, due to the fact that the remote band effects can be calculated but are not measurable quantities. One experimental technique relies on measuring the effective g factor, which is not as influenced by remote bands as the effective mass. A number of experimental and theoretical studies have determined energy gaps for zinc blende GaN [69,70,146,159–163]. The term gaps is used, as there seems to be a dispersion in the reported values. Typically, the excitonic transitions [164–166] observed in low-temperature PL is used to infer the bandgap, provided the exciton binding energy is known, which in this case is 26.5 meV. Although, low-temperature bandgaps ranging from 3.2 to 3.5 eV have been measured, most of them tend to be between 3.29 and 3.35 eV. It is therefore reasonable to use a low-temperature bandgap of 3.3 eV for zinc blende GaN. A reasonable figure for the room-temperature fundamental bandgap is 3.2 eV, although the range of 3.2–3.3 eV stated in Chapter 1 remains. As in the case of wurtzitic GaN, the temperature dependence of the energy gap was also studied for zinc blende GaN, examples of which can be found in the works of Petalas et al. [146] and Ramirez-Flores et al. [167]. Both group of authors found b ¼ 600 K (using the more reliable model 1 in Ref. [146]), but the a parameters differed somewhat. Not having any real basis for selecting one or the other, the average value of 0.593 meV K1 is considered the default value. Although the indirect gap energies have not been measured, for a calculation by Fan et al. [69], the X-valley and L-valley minima had been put at 1.19 and 2.26 eV above the G valley, respectively. These compare with an earlier prediction by Suzuki et al. [56] of about 1.4 and 1.6 eV, respectively, as shown in Figure 2.6c. Ramirez-Flores et al. [167] measured the spin–orbit splitting in zinc blende GaN to be 17 meV. Electron spin resonance measurements on zinc blende GaN determined an electron effective mass of 0.15m0 [168], which may represent the only experimental results, and the value is similar to the G-valley masses derived from first-principles calculations by Chow et al. [169] and Fan et al. [69]. Effective masses of ml ¼ 0:5m0 and mt ¼ 0:3m0 have been calculated for the X valleys in GaN [70], which are similar to the theoretical results of Fan et al. [69].
2.7 Sphalerite (Zinc blende) GaN
0.00
HH
ZB GaN
LH
–0.01
Energy (eV)
j173
–0.02
SO
–0.03
–0.04
–0.05
–0.06 –0.15
–0.10 [1 1 1]
–0.050
0.0
0.050
Wave vector (1 Å-1)
0.100 [0 0 1]
Figure 2.10 Valence band structure of zinc blende GaN [152].
The valence band of zinc blende GaN has been the topic of various theoretical efforts, and the E–k diagram by Vurgaftman and Meyer [152] is show in Figure 2.10. Although the hole effective masses in zinc blende GaN have apparently not been measured, a number of theoretical predictions of Luttinger parameters are available in the literature [69,70,164–166,168,169,425,170,171]. Once the Luttinger parameters are known, the full picture in terms of the hole effective masses can be determined. First, it should be pointed out that in polar semiconductors such as the III–Vcompounds in general and GaN in particular, it is the nonresonant polaron [172] mass that is actually measured. The polaron mass exceeds the bare electron mass by about 1–2%, the exact value of which depends on the strength of the electron–phonon interaction. Because the band structure is governed by the bare electron mass, this is the quantity that is typically reported whenever available. At the valence band edge, the heavy hole (hh) effective masses in the different crystallographic directions are related to the free mass by the Luttinger parameters in the following manner [92]: mzhh ¼
m0 2m0 m0 ½110 ½111 ; mlh ¼ ; mlh ¼ : g 1 2g 2 2g 1 g 2 3g 3 g 1 2g 3
ð2:39Þ
0.15
j 2 Electronic Band Structure and Polarization Effects
174
Here, the z-direction is perpendicular to the growth plane of (0 0 1). These expressions described by Equation 2.39 show the relationship of the Luttinger parameters to the hh effective masses that can typically be measured in a more direct manner. The light hole (lh) and so hole effective masses are given by mzlh ¼
m0 2m0 m0 ½110 ½111 ; mlh ¼ ; mlh ¼ ; g 1 þ 2g 2 2g 1 þ g 2 þ 3g 3 g 1 þ 2g 3
m0 E P Dso ¼ g1 : mso 3E g ðE g þ Dso Þ
ð2:40Þ
ð2:41Þ
Equation 2.41, which relates the split-off hole mass to the Luttinger parameters, should in principle contain an additional parameter to account for the effects of remote bands that is analogous to the F parameter [92], but the remote bands are not necessarily the same ones that cause the largest correction to the electron mass. Due to the dominance of the wurtzitic GaN, insufficient effort and thus data exist for zinc blende GaN, which is also true for even the well-investigated III–V materials to describe the effect of the interaction with remote bands on the split-off hole mass quantitatively. To restate, although the hole effective masses in zinc blende GaN have apparently not been measured, a number of theoretical predictions of Luttinger parameters are available in the literature. The values are based on averages of the heavy-hole and lighthole masses along [0 0 1], as well as the degree of anisotropy in g3–g2. Doing so leads to the parameter set as g1 ¼ 2.70, g2 ¼ 0.76, and g3 ¼ 1.11. Similarly, averaging all the reported split-off masses [69,70,163,171,173] leads to mso ¼ 0:29m0 . In its simplest form, the Luttinger parameters can be used to quickly determine the effective masses in various valence bands both in equilibrium and also under biaxial strain. In fact, with biaxial strain, the valence band degeneracy can be removed, and most strikingly the heavy-hole in-plane mass can be made smaller by compressing strain, a notion that has been exploited in the InGaAs/GaAs system very successfully. An average of the two theoretical values for EP in zinc blende GaN [163,173] yields EP ¼ 25.0 eV, which in turn implies F ¼ 0.95. Caution is advised because these values have not been verified experimentally. In conjunction with calculations of the electronic band structure of binary nitrides and specifically effective masses in the valence band, Fritsch et al. [55] also arrive at the Luttinger-like kp parameters by empirical fits for the effective masses at the G point. These Luttinger-like parameters for the valence band of zinc blende GaN are listed in Table 2.10. Those for zinc blende AlN and InN will be given in Sections 2.8.2 Sections 2.9.2. The Ai parameters transformed from the Luttinger parameters obtained with the help of the quasi-cubic approximation can be found in Tables 2.8 and 2.9. It has been argued that the Ai parameters calculated in this manner are in good agreement with the calculated values that have been used to support the value and validity of the quasi-cubic approximation, which greatly simplifies the calculations. Fritsch et al. [55] obtained the valence band effective masses for zinc blende binaries by solving the eigenvalues of the kp matrix while taking the spin–orbit interaction into account. The effective masses so calculated for the conduction and valence bands, the latter involving the light and heavy holes, as well as the spin–orbit
2.7 Sphalerite (Zinc blende) GaN Table 2.10 Luttinger parameters g1, g2, and g3 for zinc blende GaN obtained from a fit along the [1 1 0] direction along with those available in the literature, as compiled in Ref. [55].
Parameter
A
B
C
D
E
c1 c2 c3
2.89 0.85 1.20
2.96 0.90 1.20
2.70 0.76 1.07
3.07 0.86 1.26
2.67 0.75 1.10
A: empirical pseudopotential calculation by Fritsch et al. [55]; B: self-consistent FP-LAPW method within local density approximation from Ref. [171]; C: first-principles band structure calculations from Ref. [170]; D: empirical pseudopotential calculation from Ref. [174]; E: recommended values taken from Ref. [152].
split-off mass, and the anisotropy taken into account are listed in Table 2.11 for zinc blende GaN. The data contain those obtained by full potential linearized plane waves (FP-LAPW), empirical pseudopotential method (EPM) calculations and those calculated with Luttinger parameters employing ðm0 =mhh=lh Þ½100 ¼ g 1 2g 2 ; ðm0 =mhh=lh Þ½111 ¼ g 1 2g 3 ; 2g g 2 g 3 ðm0 =mhh=lh Þ½1100 ¼ 1 ; 2
ð2:42Þ
where subscripts hh and lh represent the heavy-hole and light-hole effective masses, respectively. The spin–orbit split-off-hole effective mass, mso, is isotropic in all the three directions and is given by ðm0 =mso Þ ¼ g 1 :
ð2:43Þ
Table 2.11 Effective masses for electrons (e), heavy holes (hh), light holes (lh), and spin–orbit split-off (so) holes in units of the free-electron mass m0 along the [1 0 0], [1 1 1], and [1 1 0] directions for zinc blende GaN. [100]
[100]
[111]
[111]
[110]
[110]
Reference
me
mhh
mlh
mhh
mlh
mhh
mlh
mso
A B C D E F
0.14 0.14 0.13 0.17 0.15 0.12
0.84 0.86 0.76 0.85 0.85 1.34
0.22 0.21 0.21 0.24 0.24 0.70
2.07 2.09 1.93 1.79 2.13 1.06
0.19 0.19 0.18 0.21 0.21 0.63
1.52 1.65 1.51 1.40 1.55 1.44
0.20 0.19 0.19 0.21 0.21 0.58
0.35 0.30 0.32 0.37 0.29 0.20
Compiled by Fritsch et al. [55]. A: after Ref. [55]; B: self-consistent FP-LAPW method within local density approximation from Ref. [171]; C: empirical pseudopotential calculation from Ref. [69]; D: calculated from Luttinger parameters from Ref. [170], using Equations 2.42 and 2.43; E: calculated from recommended Luttinger parameters from Ref. [152], using Equations 2.42 and 2.43; F: empirical pseudopotential calculation from Ref. [70].
j175
j 2 Electronic Band Structure and Polarization Effects
176
Because various calculations place the hydrostatic deformation potential for zinc blende GaN in the range 6.4 to 8.5 eV [31,69,70,104,169,175], for the lack of a better choice, the average value of the hydrostatic deformation potential a ¼ 7.4 eV is chosen. It should be mentioned that due to the nature of the atomic bonding in III–V materials, the bandgap increases for a compressive strain. Under positive hydrostatic pressure that produces negative strain because the lattice constant gets smaller, the change in the bandgap corresponding to a change in volume, DV/V ¼ (exx þ exx þ ezz), given by DEg ¼ a(exx þ exx þ ezz) is positive, which necessitates the sign of the deformation potential to be negative. The deformation potential is the sum of the conduction and valence band deformation potentials, a ¼ ac þ av. Under pressure, the conduction band edge is believed to move upward in energy while the valence band moves downward, with most of the change being in the conduction band edge. Similar to the total bandgap change, the change in valence band under pressure can be described as DEg ¼ av(exx þ exx þ ezz). Naturally, it is much easier to measure the change in the total bandgap due to strain than its component effects on the conduction and valence bands. The valence band deformation potential value reported by Wei and Zunger [176], namely, av ¼ 0.69 eV, is suggested. The sign convention for av is adopted from Vurgaftman et al. [92]. To restate, the shrinking volume and negative strain cause the valence band to move down. The values reported in the literature are in the range of 0.69 to 13.6 eV. The suggested value is consistent with the expectation that most of the strain shift should occur in the conduction band. As for the shear deformation potentials b and d, the same is applied, which yields a suggestion that b ¼ 2.0 eV with the full range of reported values being 1.6 to 3.6 eV. The recommendation d ¼ 3.7 eV is an average of the published results from Ohtoshi et al. [175], Van de Walle and Neugebauer [177], and Binggeli et al. [178]. No experimental confirmations of any of these deformation potentials for zinc blende GaN appear to exist. Turning to elastic constants, the values of C11 ¼ 293 GPa, C12 ¼ 159 GPa, and C44 ¼ 155 GPa have been taken from the theoretical calculations of Wright [179]. Very similar sets have also been calculated by Kim et al. [109,180] and Bechstedt et al. [181]. For more details of elastic constants for all three binaries, refer to Tables 2.25–2.27 and 2.28 that will follow later on in this chapter. The parameters in conjunction with the band structure for zinc blende GaN are compiled in Table 2.12.
2.8 AlN
AlN forms the larger bandgap binary used in conjunction with GaN for increasing the bandgap for heterostructures. As in the case of GaN, AlN also has wurtzitic and zinc blende polytypes, the latter being very unstable and hard to synthesize. Owing to increasing interest in solar blind devices and expectations that larger AlGaN with large mole fractions of AlN would have large breakdown properties, this material has been steadily gaining interest. It should also be mentioned that the N overpressure on Al is the smallest among those over Ga and In, paving the way for equilibrium growth of AlN bulk crystals, albeit not without O contamination.
2.8 AlN Table 2.12 Parameters associated with the band structure for zinc blende GaN.
Parameter alc (Å) at T ¼ 300 K E Gg (eV) a (G) (meV K1) b (G) (K) E Xg ðeVÞ a (X) (K) b (X) (meV K1) E Lg ðeVÞ a (L) (meV K1) b (L) (K)
Value 4.50 3.3 0.593 600 4.52 0.593 600 5.59 0.593 600
Parameter
Value
Parameter
Value
Dso (eV) me ðGÞ ml ðXÞ mt ðXÞ c1 c2 c3 mso
0.017 0.15 0.5 0.3 2.70 0.76 1.11 0.29
EP (eV) F VBO (eV) ac (eV) av (eV) b (eV) d (eV) c11 (GPa) c12 (GPa) c44 (GPa)
25.0 0.95 2.64 6.71 0.69 2.0 3.7 293 159 155
Bandgaps are for low temperature [152].
2.8.1 Wurtzite AlN
Wurtzite AlN is a direct bandgap semiconductor with a bandgap near 6.1 eV and still considered to be semiconductor. The zinc blende polytype is not stable with a predicted indirect bandgap, as will be discussed in the next section. The AlN derives its technological importance from providing the large bandgap binary component of the AlGaN alloy, which is commonly employed both in optoelectronic and electronic devices based on the GaN semiconductor system. Early on, the absorption measurements carried out by Yim et al. [182] and later by Perry and Rutz [183] indicated a large energy gap for wurtzite AlN of 6.28 eVat 5 K to 6.2 eVat room temperature. The actual figures are converging at values about 0.1 eV below those contained in those early reports. As for the dependence of the bandgap on temperature, the Varshni parameters of a ¼ 1.799 meV K1 and b ¼ 1462 K were reported by Guo and Yoshida [184], who also found the low-temperature gap to be 6.13 eV, which is similar to that reported by Vispute et al. [185], a value closer to the values observed of late. Tang et al. [186] resolved what they believed to be the free or shallow impurity bound exciton in their cathodoluminescence (CL) data, at an energy of 6.11 eV at 300 K. Brunner et al. reported a variation from 6.19 eV at 7 K to 6.13 eV at 300 K [187]. A group of same authors Wethkamp et al. [188] used spectroscopic ellipsometry and determined that the energy gap varies from 6.20 eV at 120 K to 6.13 eV at 300 K. Kuokstis et al. [189] resolved a low-temperature free-exciton transition at 6.07 eV. Guo et al. [190] reported the temperature dependence of the reflectance spectra, while fitting it to the Bose–Einstein expression. Using the low-temperature data in conjunction with the Varshni parameters of Guo and Yoshida [184], leads to an intermediate value of 6.23 eV for the lowtemperature bandgap. The Varshni parameters reported by Brunner et al. [187] indicate no significant divergence from GaN for the entire AlGaN alloy composition range, which may bring the accuracy of the data into question. As in any semiconductor, the quality and strain nature of the films can alter the results. The availability
j177
j 2 Electronic Band Structure and Polarization Effects
178
of high-quality homoepitaxial AlN with presumably no strain has shed the much needed light onto the issues surrounding the actual bandgap of AlN [191]. However, even then O contamination could cause the near-band emission peak observed for shift, as it has done so for high-quality bulk substrates reported by Slack et al. [192]. In the experiments of Silviera et al. [191], the epitaxial layer was 0.5 mm, plenty considering the small penetration depth of 10 keV electrons used for the CL experiments, and the substrate was 287 mm thick with O concentrations of about 5 · 1019 cm3 as measured by Neutron activation. The homoepitaxial AlN films have been grown by organometallic vapor-phase epitaxy on the single-crystal AlN substrates and efforts were undertaken to reduce the oxygen content of the film. The lowtemperature near band edge CL spectrum of the AlN film is shown in Figure 2.11. The open squares correspond to the experimental data obtained at 5 K, while the full line representing the best fit to the experimental data using Corinthian line shapes with transitions are indicated with dashed lines. The assignments shown are a result of thermal quenching behavior. The full widths at half maxima (FWHM) of the narrowest emission line at 6.023 eV is about 1.0 meV and was reported to be perhaps limited by the slit size used during the experiment. A measurement of the 253.65 nm emission line of a low-pressure Hg lamp using the same slits size set resulted in a FWHM of about 0.7 meV, as tabulated in Table 2.13. Figure 2.12 shows the temperature-dependent CL spectra for the AlN film. A rapid decrease in the intensities of the four peaks initially observed between 5.98 and 6.01 eV with the increase in temperature is evident, which is consistent with recombination processes involving excitons bound to shallow neutral centers. The peaks at 6.023 and 6.036 eV remain intense with increasing temperature, which lets them gain the free exciton A (FXA) and free exciton B (FXB) assignments, respectively,
Figure 2.11 High-resolution CL spectrum of an AlN homoepitaxial film. The full line represents our best fit using Lorentzian line shapes, and the dashed lines are the transitions composing the full line [191].
2.8 AlN Table 2.13 Energy positions, full widths at half maxima (FWHM),
and preliminary assignments associated with the transitions shown in Figure 2.11 [191]. Energy (eV)
FWHM (meV)
Assignment
5.98 6.000 6.008 6.01 6.023 6.036
49.4 11.0 1.5 44.0 1.0 8.0
A07 X A D027 X A D017 X A D017 X B FXA FXB
due to their large binding energies. The line around 6.07 eV, shown as FX2A , is some 2 orders of magnitude weaker than the most intense bands observed in the spectrum. On the basis of the similarity in the luminescence spectra of both GaN and AlN, the peak at 6.07 eV is attributed to the first excited state of the FXA. This assignment allows an estimation of the FXA binding energy as 63 meV, which is about twice the value for GaN. This leads to an estimated low-temperature bandgap of AlN of 6.086 eV (the sum of the FXA energy and its binding energy). Returning to the band structure of AlN, of considerable significance, the crystal field splitting in wurtzitic AlN is believed to be negative. The ramification of this is that the topmost valence band is the crystal hole band. Calculations have yielded a range of crystal field splittings, namely, Dcr ¼ 58 meV by Suzuki et al. [56], Dcr ¼ 217 meV by Wei and Zunger [107],Dcr ¼ 176 meV by Shimada et al. [193], and Dcr ¼ 244 meV by Wagner and Bechstedt [85]. Moreover, splittings of Dcr ¼ 104 and 169 meV were obtained from first-principles and semiempirical pseudopotential calculations, respectively, by Pugh et al. [163] and Dcr ¼ 215 meV by Kim et al. [154]. Averaging all of the available theoretical crystal field splittings, one obtains a value of Dcr ¼ 169 meV. Silveira et al. [194], using optical reflectance data performed on a- and c-plane bulk AlN and a quasi-cubic model developed for the wurzite crystal structure, determined the crystal field splitting to be D ¼ 225 meV. Note that the negative sign for the crystal
Figure 2.12 Temperature-dependent CL spectra of 0.5 mm thick AlN film deposited on 287 mm bulk AlN substrate [191].
j179
j 2 Electronic Band Structure and Polarization Effects
180
field splitting has important implications, namely, that the G7 valence band is on the top of G9 valence band, which is opposite of that in GaN. As for the spin–orbit splitting, the literature values range from 11 [163] to 20 meV [124]. Silveira et al. [194] again using the optical reflectance spectra in bulk AlN determined the spin–orbit splitting energy to be d ¼ 36 meV. In view of the experiments in high-quality bulk AlN, the value of 36 meV is recommended even though it is much larger than the calculated value of 19 meV recommended by Wei and Zunger [107]. As in the case of GaN, the region of the energy band near the bottom of the conduction band, as it manifests itself in devices, can be represented by the effective mass. The same of course applies for the top of the valence band. A number of investigators have calculated the AlN electron effective mass [154,124,163,195,196], with the prediction that it displays a greater anisotropy than that for wurtzitic == GaN [56]. The bare mass values of m? e ¼ 0:30m0 and m e ¼ 0:32m0 obtained by averaging the available theoretical masses may represent a good set of default values as this stage. It should again be underscored that further experimental studies are needed to verify the calculations. As for the valence band, a number of theoretical sets of valence band parameters are available [56,58,154]. There is an apparent disagreement in the signs for A5 and A6 among these reports, which may be irrelevant, because only absolute values of those parameters enter the Hamiltonian [154,163]. The A parameters given by Kim et al. [154] are suggested because the crystal field and spin–orbit splittings reported by these authors are closest to the ones suggested here. The hydrostatic deformation potential for wurtzite AlN has been reported to be in the range of 7.1 and 9.5 eV [31,180], which is consistent with the observation that the bandgap pressure coefficients for AlGaN alloys have little dependence on composition, as reported by Shan et al. [197]. The calculated values of a1 ¼ 3.4 eV and a2 ¼ 11.8 meV reported by Wagner and Bechstedt [85] are assumed to represent the material well. Theoretical values are also available for a few of the valence band deformation potentials such as D3 ¼ 9.6 eV, and D4 ¼ 4.8 eV [180]. However, the complete set D1 ¼ 17.1 eV, D2 ¼ 7.9 eV, D3 ¼ 8.8 eV, D4 ¼ 3.9 eV, D5 ¼ 3.4 eV, and D6 ¼ 3.4 eV with the last value derived using the quasi-cubic approximation presented by Shimada et al. [193] can be used in the absence of any other reliable data. For mere availability reasons and few other issues, the mechanical properties of AlN have seen a good deal of experimental activity very early on, which later was followed by theory. Tsubouchi et al. [198], McNeil et al. [199], and Deger et al. [200] measured the elastic constants of wurtzitic AlN. A good many theoretical papers have also been reported [154,193,201–204]. The values suggested by Wright [179], who also provided a detailed discussion of their expected accuracy, namely, C11 ¼ 396 GPa, C12 ¼ 137 GPa, C13 ¼ 108 GPa, C33 ¼ 373 GPa, and C44 116 GPa, are recommended. Several piezoelectric coefficients [205,206] for early AlN at least in part can be found in Ref. [207]. The result for d33 ¼ 5.6 pm V1 reported in Ref. [207] is in reasonably good agreement with the previous determinations but differs somewhat from d33 ¼ 5.1 pm V1 measured by Lueng et al. [208]. While these experiments focused on only d33, both d33 and d13 can be determined from first-principles
2.8 AlN Table 2.14 Recommended band structure parameters for wurtzitic AlN [152].
Parameter
Value
Parameter
Value
Parameter
Value
Eg (eV, low temperature) a (meV K1) b (K) Dcr (meV) Dso (meV) == me =m0 m? e =m0 a1 (eV) a2 (eV)
6.077 [194] 1.799, 0.9a 1462, 1000a 225 [194] 36 [194] 0.32 0.30 3.4 11.8
A1 A2 A3 A4 A5 A6 A7 (meV Å) d13 (pm V1) d33 (pm V1) d15 (pm V1) Psp (C m2)
3.86 0.25 3.58 1.32 1.47 1.64 0 (default) 2.1 5.4 3.6 0.090
D1 (eV) D2 (eV) D3 (eV) D4 (eV) D5 (eV) D6 (eV) C11 (GPa) C12 (GPa) C13 (GPa) C33 (GPa) C44 (GPa)
17.1 7.9 8.8 3.9 3.4 3.4 396 137 108 373 116
See Tables 2.27 and 2.28 for details related to the elastic constants, piezoelectric constants, and spontaneous polarization charge. Any dispersion among the tables is a reflection of the uncertainty in the available parameters. Note that the G7 valence band is above the G9 valence band, which is opposite of GaN. It is also similar to that in ZnO, which is somewhat controversial. See Zinc Oxide: Materials Preparation, Properties, and Devices, by H. Morkoç and Ü. Özg€ ur, Wiley (2008) regarding the valence band ordering in ZnO. a Obtained using the temperature dependence of the A exciton energy reported in Ref. [194].
calculation [193,209–212]. The recent theoretical values of Bernardini and Fiorentini, d33 ¼ 5.4 pm V1 and d13 ¼ 2.1 pm V1 [210], are suggested although the elastic coefficients given in that reference are somewhat larger than the recommended values. On the basis of recent measurements [206,207] and a calculation [210] of the shear piezoelectric coefficient, Vurgaftman and Meyer [152] recommend d15 ¼ 3.6 pm V1. The parameters concerning the bandgap-related issues for wurtzitic AlN recommended by Vurgaftman and Meyer [152] are tabulated in Table 2.14. A compilation of the dispersion in the effective mass for both the conduction band and various valence bands as obtained by various computational methods as well as parameters used in the description of the bandgap for wurtzitic InN, particularly, in the context of empirical pseudopotential method, as described in Ref. [55], are tabulated in Table 2.15. Even though a detailed discussion of polarization is reserved (Section 2.12), a succinct treatment of the topic is given here as it is relevant to the topic under discussion. The difference between the GaN and AlN spontaneous polarizations strongly causes a net polarization at the interface between the two materials that extends to the GaN/AlGaN interfaces as well. This charge, which is bound, influences the band profiles and energy levels in GaN/AlN quantum heterostructures. Although rigorous calculations [84,181,209,213] of the spontaneous polarization Psp(AlN) have produced results spanning a fairly broad range, from 0.09 to 0.12 C m2, values for the difference Psp(AlN)Psp(GaN) have tended to be more consistent, with most falling between 0.046 and 0.056 C m2. Experimentally, for some time the majority of workers on the GaN/AlGaN system reported somewhat smaller Psp(AlN)Psp(GaN). For example, Leroux et al. [214,215] derived 0.051 < Psp < 0.036 C m2 for AlN. A study of the charging of GaN/AlGaN field effect transistors led to a similar
j181
j 2 Electronic Band Structure and Polarization Effects
182
Table 2.15 Effective masses and band parameters for wurtzitic AlN.
Parameter ==
aniso
iso
A
B
C
D
E
F
me m? e
0.231 0.242
0.232 0.242
0.33 0.25
0.33 0.25
0.24 0.25
0.24 0.25
0.35
0.33
mhh
2.370
2.382
3.68
3.53
1.949
1.869
3.53
4.41
2.370
2.382
3.68
3.53
1.949
1.869
3.53
4.41
0.209 3.058 0.285 1.204 4.789 0.550 4.368 1.511 1.734 1.816 0.134 0.128
0.209 3.040 0.287 1.157 4.794 0.571 4.374 1.484 1.726 1.788 0.153 0.160
0.25 10.42 0.24 3.81 4.06 0.26 3.78 1.86 2.02
0.229 2.584 0.350 0.709 4.367 0.518 3.854 1.549 1.680 2.103 0.204 0.093
0.212 2.421 0.252 1.484 4.711 0.476 4.176 1.816 1.879 2.355 0.096 0.093
0.26 11.14 0.33 4.05 3.86 0.25 3.58 1.32 1.47 1.64
0.27 2.18 0.29 4.41 3.74 0.23 3.51 1.76 1.52 1.83 0
==
== mlh == mch m? hh m? lh m? ch
A1 A2 A3 A4 A5 A6 A7 D1
0.25 6.33 0.25 3.68 3.95 0.27 3.68 1.84 1.95 2.91 0 0.059
0 0.059
0.215
Effective masses in units of free-electron mass m0, Luttinger-like parameters Ai (i ¼ 1, . . ., 6) in units of h2 =2m0 , and A7 in units of eV Å. The crystal field splitting energy D1 is given in units of meV. The term aniso represents the values derived using a band structure calculation with anisotropically screened model potentials, whereas the term iso describes a comparative band structure calculation on the basis of isotropically screened model potentials using an averaged e0 value by taking the spur of the dielectric tensor [55]. Anisotropically screened and isotropically screened values are from Ref. [55]. A: FP-LAPW band structure calculations are from Ref. [170], and effective mass parameters are obtained through a 3D fitting procedure within cubic approximation; B: FP-LAPW band structure calculations are from Ref. [170], and effective mass parameters are obtained by a direct line fit; C: Ai from Ref. [153] obtained through a Monte Carlo fitting procedure to the band structure and effective masses calculated using Equations 2.30 and 2.31; D: direct kp calculations for Ai from Ref. [153] and effective masses obtained from Ai using Equations 2.30 and 2.31; E: direct fit of Ai to first-principles band structures from Ref. [154]; F: Ai and effective masses obtained in the quasi-cubic model from zinc blende parameters from Ref. [154].
conclusion [216], and Hogg et al. [217] were able to fit their luminescence data by assuming negligible spontaneous polarization. Park and Chuang [218] required Psp ¼ 0.040 C m2 to reproduce their GaN/AlGaN quantum well data. On the contrary, Cingolani et al. [219] reported good agreement with experiment using a higher value derived from the original Bernardini et al. [209] calculations. A significant step toward resolving this discrepancy has been the realization that the AlGaN spontaneous polarization cannot be linearly interpolated between the values at the binary end points [220–222]. In combination with an improved nonlinear strain treatment of the piezoelectric effect, the discrepancy between theory and experiment for GaN/AlGaN quantum wells has been largely eliminated [85]. For additional details, see Section 3.14 and the text dealing with Tables 2.25, 2.27 and 2.28. We adopt Psp ¼ 0.090 C m2 as the recommended value for AlN, in conjunction with Psp(GaN) ¼ 0.034 C m2. The recommended band structure parameters for wurtzite AlN are compiled in Table 2.14.
2.8 AlN
2.8.2 Zinc Blende AlN
Only a handful of purportedly successful growths of zinc blende AlN on zinc blend substrates, such as GaAs and 3-C SiC, and Si substrates following low-temperature zinc blende GaN buffer layers have been reported [223–226]. Consequently, much of the discussion here relies primarily on theoretical projections and so do the parameter set recommended for this polytype. The only quantitative experimental study of the bandgap indicated a G-valley indirect gap of 5.34 eV at room temperature [224]. Assuming that the Varshni parameters for the wurtzitic AlN hold for the zinc blende polytype, the aforementioned room-temperature bandgap translates to a low-temperature gap of 5.4 eV. Vurgaftman and Meyer [152] recommend 4.9 and 9.3 eV for the X- and L-valley gaps, respectively [31,69,163]. The spin–orbit splitting is expected to be nearly the same as in wurtzite AlN at 19 meV [154,107,171,227]. Averaging the theoretical results from a number of different publications [69,154,163,170,173], one arrives at a G-valley effective mass of 0.25m0. The longitudinal and transverse masses for the X valley have been predicted to be 0.53m0 and 0.31m0, respectively [69]. If the method used previously for the GaN is applied to zinc blende AlN, one arrives at recommended Luttinger parameters of g1 ¼ 1.92, g2 ¼ 0.47, and g3 ¼ 0.85, and mso 0.47m0 [69,154,170,173]. These as well as the other literature values of the Luttinger parameters are listed in Table 2.16. Fritsch et al. [55] calculated the effective masses for conduction and valence bands, the latter involving the light and heavy holes, as well as the spin–orbit split-off mass, which with the anisotropy taken into account are listed in Table 2.17 for zinc blende AlN. The data contain those obtained by FP-LAPW, EPM calculations and those calculated with Luttinger parameters (Equations 2.42 and 2.43). When the calculated [163,173] values for the momentum matrix are averaged, a value of EP ¼ 27.1 eV (with F ¼ 101) is obtained. Hydrostatic deformation potentials of 9.0 eV [31] and 9.8 eV [69] have been reported. The deformation potential values, a ¼ 9.4 eV, av ¼ 4.9 eV [69,107], b ¼ 1.7 eV [69,177,178], and d ¼ 5.5 eV [180,177,178], have been suggested [152]. The elastic constants of C11 ¼ 304 GPa, C12 ¼ 160 GPa, and C44 ¼ 193 GPa calculated by Wright [179] are Table 2.16 Luttinger parameters g1, g2, and g3 for zinc blende AlN obtained from a fit along the [1 1 0] direction along with those available in the literature, as compiled in Ref. [55].
Parameter
A
B
C
D
E
c1 c2 c3
1.85 0.43 0.74
1.54 0.42 0.64
1.50 0.39 0.62
1.91 0.48 0.74
1.92 0.47 0.85
A: empirical pseudopotential calculation by Ref. [55]; B: self-consistent FP-LAPW method within local density approximation from Ref. [171]; C: first principles band structure calculations from Ref. [170]; D: empirical pseudopotential calculation from Ref. [174]; E: recommended values taken from Ref. [152].
j183
j 2 Electronic Band Structure and Polarization Effects
184
Table 2.17 Effective masses for electrons (e), heavy holes (hh), light holes (lh), and spin–orbit split-off (so) holes in units of the free-electron mass m0 along the [1 0 0], [1 1 1], and [1 1 0] directions for zinc blende AlN. [100]
[100]
[111]
[111]
[110]
[110]
Reference
me
mhh
mlh
mhh
mlh
mhh
mlh
mso
A B C D E
0.23 0.28 0.21 0.30 0.25
1.02 1.44 1.05 1.39 1.02
0.37 0.42 0.35 0.44 0.35
2.64 4.24 2.73 3.85 4.55
0.30 0.36 0.30 0.36 0.28
1.89 3.03 2.16 2.67 2.44
0.32 0.37 0.31 0.38 0.29
0.54 0.63 0.51 0.67 0.47
Compiled by Fritsch et al. [55]. A: after Ref. [55]; B: self-consistent FP-LAPW method within local density approximation from Ref. [171]; C: empirical pseudopotential calculation from Ref. [69]; D: calculated from Luttinger parameters from Ref. [170], using Equations 2.42 and 2.43; E: calculated from recommended Luttinger parameters from Ref. [152], using Equations 2.42 and 2.43.
similar to the sets quoted in other theoretical works [180,181,228] and are therefore suggested. These and other band structure parameters recommended for zinc blende AlN are tabulated in Table 2.18 [152].
2.9 InN
As in the case of AlN, the interest in InN has so far been not necessary because of its properties, but because of the InGaN alloy that is used in lasers and LEDs operative in the visible and violet regions of the optical spectrum. In fact, if and when the technological issues are overcome, the InGaN channel FETs may also be superior to the GaN channel varieties [229], some details of which are discussed in Volume 3,
Table 2.18 Parameters associated with the band structure for zinc blende AlN.
Parameter
Value
Parameter
Value
Parameter
Value
alc (Å) at T ¼ 300 K E Gg ðeVÞ a (G) (meV K1) b (G) (K) E Xg ðeVÞ a (X) (K) b (X) (meV K1) E Lg ðeVÞ a (L) (meV K1) b (L) (K)
4.38 5.4 0.593 600 4.9 0.593 600 9.3 0.593 600
Dso (eV) me ðGÞ ml ðXÞ mt ðXÞ c1 c2 c3 mso
0.019 0.25 0.53 0.31 1.92 0.47 0.85 0.47
EP (eV) F VBO (eV) ac (eV) av (eV) b (eV) d (eV) c11 (GPa) c12 (GPa) c44 (GPa)
27.1 1.01 3.44 4.5 4.9 1.7 5.5 304 160 193
Bandgaps are for low temperature [152].
2.9 InN
Chapter 3. The properties, particularly the fundamental parameters of InGaN for a given composition, depend very much on the InN parameters, particularly its bandgap. It is therefore important to understand the properties of bulk InN in its wurtzitic form. InN, however, is not all that easy, even considering the general difficulties encountered in the nitride semiconductor system, to synthesize. The somewhat intractable problem with InN is the enormous difference in the ionic size of its constituent atoms in that the atomic radii for In and N are largely different, which leads to highly distorted interatomic distances, interatomic bonding charges, tendency to form metallic clusters of the group III constituent, and inhomogeneous strain. All of these could, in principle, lead to pronounced anomalies in all the properties of InN, inclusive of measured bandgap and nature of defects. To make matters worse, the InN layers are grown at best on GaN epitaxial layer with large lattice mismatch (lm), aggravating many of the aforementioned problems. In spite of all these, progress is steadily made. It should be added that after having been accepted as the bandgap of InN, the 1.98 eV figure came under new scrutiny in that a plethora of reports concluded the actual bandgap to be 0.7–0.8 eV. Just at a time when a good many got convinced of the newer data, questions have been raised about the models along with heightened level of scrutiny of the new low-bandgap data. Assuming that interpretation of experiments pointing to the small bandgap figures are impeccable, theories also begin to be developed, even though long-standing understanding such as cation rule would be broken by the small bandgap figure. 2.9.1 Wurtzitic InN
The bandgap of InN has been a point of controversy dating back to the early days of InGaN development [1,230,231]. Early absorption studies on sputtered thin films concluded that the InN bandgap is in the 1.7–2.2 eV range [232–236]. However, no band-to-band PL could be observed in the samples prepared by sputtering in early developmental stages or later on in films grown by OMVPE and MBE. A review [237] of various crystal growth related issues and resultant properties as well as a proceedings [238] of a meeting devoted to debating these issues is available in the literature. In contrast to earlier reports of no near band edge emission, Davydov et al. [239–242] and others [243–245] reported near band edge emission but at much lower energies near 0.6–0.8 eV, depending on the report. Also see the comment, Ref. [246], on Ref. [239] and reply to that comment [247]. These reports relied to various extents on absorption, photoluminescence, and photoluminescence excitation experiments that showed the experimental evidence for the bandgap of InN to be overwhelmingly in the range of 0.7–0.8 eV and recommended a zero-temperature gap of 0.78 eV [248] and 692 2 meV [249], and in fact values as low as 0.65 eV [239] and 0.67 eV [250] have been reported as well. InN films of N-polarity grown by RF molecular beam epitaxy exhibited a bandgap value of about 0.7–0.75 eV as measured by optical transmission and reflection measurements [251–256]. The PL emission
j185
j 2 Electronic Band Structure and Polarization Effects
186
line appeared at 0.7 eV with no shift in energy between 300 and 77 K. These particular samples also exhibited room-temperature electron mobilities in the range of m ¼ 1750–2000 cm2 V1 s1 and an electron concentration in the range of n ¼ 2–3 · 1018 cm3 at room temperature. Some details about the growth of these films are provided in Section 3.5.13. The polycrystalline or nanocrystalline nature of those early thin films associated with high electron densities and low mobilities led the proponents of the small bandgap for InN to suggest that those early films most likely contained a good deal of oxygen coupled with Moss–Burstein effect that could push the apparent bandgap upward. Another possible explanation for the dispersion in the reported values of the bandgap and also in support of the smaller bandgap may have to do with blue shift caused by any quantum size effects. At least the correct trend has been established by Lan et al. [257] who reported PL emission at 1.9 eV in nanorods of diameter between 30 and 50 nm (dubbed the brown InN), whereas 0.766 eV emission (measured at 20 K) was observed in rods with a diameter in the range of 50–100 nm (dubbed the black InN). Further refinement of the work led to the observation that the samples with fine (10 nm) or containing very high carrier concentrations exhibit the visible emission. In fact, both IR and visible peaks could be observed in the same sample when the samples show bimodal distribution of grain size [258]. The InN nanorods catalytically formed in the upstream portion of the substrate were of the brown type while those downstream were of the black variety. The Au catalyst that floats on the top of the InN nanorods as the growth progresses were shown to be encapsulated with In2O3, which is somewhat unexpected and may be made possible owing to close epitaxial relationship between wurtzitic InN and In2O3, indicating the participation of O in the catalytic process. The source of O, in this case, was attributed to residual O in the reactor as well as the quartz tube used. Furthermore, the black InN photon emission in a PL experiment quenched above 150 K while that from the brown InN exhibited strong emission at room temperature, albeit broad. The variation in the observed emission wavelength has been attributed to a variety of sources including O incorporation, Moss–Burstein shift due to high electron concentration, and quantum size effect. The quantum size effect would require diameters of less than 5 nm that is much larger than the 20–50 nm brown InN rods, which cannot therefore explain the bandgap shift to 1.9 eV from about 0.7 eV. It should be stated that the characteristic PL peak in O-implanted InN is different from the broad visible peak. The possibility of InN quantum dot formation in InGaN as being responsible for bandgap variations has not been of as much use in that even intentional dot formation, albeit limited to possibly only one report, did not lead to any blue shift [259]. The results are not precise but speak to the trend and demonstrate that great care must be taken to make sure that all the samples are grown under identical conditions, and doping levels are the same and not very large. Another issue of paramount importance that comes into the equation is the degree of crystallinity of the InN films. For example, Anderson et al. [260] produced polycrystalline InN with 0.8 eV luminescence present but did not identify the size of polycrystalline grains.
2.9 InN
Proponents of the earlier and larger bandgap for InN bring on the table several arguments undermining the validity of the low-bandgap figure. Among theme is that the absorption squared versus energy plots used to obtain the apparent bandgap in a semiconductor with very high carrier concentration underestimates the bandgap due to band tailing [261]. In addition, the 0.7–0.8 eV peak ascribed by some to the bandgap of InN is attributed to defects caused by nonstoichiometry of the films, which are grown under extremely In reach conditions and far away from equilibrium condition [261]. Moreover, The Moss–Burstein blue shift used by proponents of the small bandgap to account for the large bandgap reported earlier is not consistent among all the samples in that the sample with low carrier concentrations in the past confirmed the large bandgap of InN, albeit they were most likely heavily compensated. Finally, the opponents of the larger bandgap argue that the observed transition at 0.7–0.8 eV is due to Mie resonance that in turn is due to scattering or absorption of light in InNcontaining clusters of metallic In, which may have been mistaken for the low bandgap, as only the In cluster containing samples do show the 0.7–0.8 eV peak [262]. Let us now discuss the evolution of InN, and in particular, its perceived and admittedly controversial perceived bandgap. The early attempts to produce InN relied on not well-developed methods, as compared to modern crystal growth techniques, and as such produced mostly powder and nanopowders [263–265]. Naturally, the early work on thermodynamic decomposition studies and X-ray diffraction (advantageous for power diffractometry due to the nature of the films) were performed on powder films. Even in these very powder samples, there was some observed variation in the color of the material in that some have been noted to be black or blackbrown [264–266] as opposed to the deep red that one would expect for the 1.8 eV bandgap originally estimated [265]. For a witty historical account of these early efforts through the controversy, the reader is referred to Refs [267,268]. Below a discussion of the role of oxygen, possible defects that might be responsible for the 0.7 eV emission, the Moss–Burstein shift, particulars of carrier absorption, and Mie resonance is given in an effort to give an overall appreciation of issues causing the controversy in determining the bandgap parameter of InN. Focusing on the issue of O, because In2O3 has a bandgap of 3.75 eV [269], the argument goes that if InN films are heavily O contaminated, the bandgap would be pushed upward. This is one of the points used by the proponents of the small bandgap. However, as shown in Figure 1.34, to bring the bandgap from 0.7 to 1.9 eV would require some 30% O in the film assuming that the Vegards law holds. It should also be noted that the stated O must form an alloy for increasing the bandgap not just as surface contamination or inclusion of O at grain boundaries. One of the samples used to support the InN–In2O3 alloy formation was an RF-sputtered film with poor mobility and high carrier concentration [270]. Auger analysis used to determine the O content in the aforementioned layer is sensitive to only O and cannot determine whether it is alloyed in the semiconductor InN. In addition, any depth profiling accompanied by that technique requires argon ion etching, which is known to result in severe nitrogen loss and lead to overestimation of the atomic oxygen concentration owing to the recycling of sputtered oxygen on the film surface coupled with the strong bond between oxygen and surface indium [271]. An alternative method, Rutherford backscattering (RBS), was
j187
j 2 Electronic Band Structure and Polarization Effects
188
also used by Davydov et al. [270] with the sample mounted on glass, which in consideration of the sample thickness may not have allowed a definitive determination whether the O signal is due to O in InN or the glass substrate used. Nevertheless, the O content so measured was 20%, which is to be contrasted to 10% obtained [272] from measurements performed in similar layers using elastic recoil, which is accurate for elemental analysis. It should again be mentioned that elemental O in the film, even if present in stated quantities, is not the same as InN:In2O3 alloy, which would increase the bandgap, not to mention the fact that such alloying is not favored by temperatures employed in RF sputter deposited InN. One can surmise that amorphous InON, NO2 and surface hydroxide species are implicated as being responsible for the 1.8–2 eV bandgap reported. These species are only as surface species, and there has not as yet been any evidence reported of any other form of InN–In2O3 alloy species during the growth of InN, as can be discerned from Figure 1.34 that 10% or even 20% oxygen could not account for the bandgap of 1.9 eV if a bandgap of 0.7 eV is assumed for InN. Figure 1.34 clearly indicates that an alloy with about 37 at.% oxygen, which translates to 44% In2O3, would be needed to provide a bandgap of 2.0 eV if the bandgap of pure InN is 0.7 eV. Consequently, 10% (at.) oxygen, even if all were alloyed with InN, would account for a blue shift of about only 0.3 eV, much less than some 1 eV for consistency. If the bandgap of the alloy follows a bowing parameter, unless it is positive, which is unlikely, the blue shift caused by alloyed O would even be smaller. This view is supported by the results of Yoshimoto et al. [273] that reported a bandgap value of 1.8–2.0 eV for MBE InN grown on quartz with 3% atomic oxygen present in the film. Evidently, this level of O contamination does not prevent one from arriving at the longstanding bandgap of InN, as any blue shift caused by this level of elemental O even if all is in the alloy form does amount to much. The role of O in InN had been investigated early on in sputtered films. Among them is the work of Westra et al. [274] who produced InN with carrier concentrations between 7 · 1019 and 2 · l020 cm3 and mobility of 4–10 cm2 V1 s1. Rutherford backscattering data indicated 11% atomic oxygen, and indium-to-nitrogen ratios slightly above a value of 1. No evidence of oxygen or oxynitride phases was observed in the X-ray diffraction spectra, which led those authors to propose that oxygen is in the form of an amorphous indium oxynitride, similar to the observations of Foley and Lyngdal [275]. In this case the structure would maintain the stoichiometry, while the oxygen would not be detected by X-ray diffraction. Contribution from NO2 has been observed for samples containing higher oxygen content in a polycrystalline InN matrix with no evidence for O in X-ray diffraction or infrared absorption spectra [271], wherein grain boundaries were proposed to be the host for O. Raman data for sputtered InN with 10% atomic oxygen concentration reveals only InN-related phonon peaks, implying the lack of alloy formation [272,276]. The moral of the aforementioned discussion is that any O present in at least polycrystalline InN that has been examined with the role of O in mind did not seem to be incorporated as an alloy, which explicitly leads to the conclusion that could not contribute to the bandgap. An in-depth discussion of the topic can be found in Ref. [267]. Moss–Burstein effect has been forwarded as a plausible argument for observing 1.8–2 eV absorption if the 0.7 eV bandgap is assumed. This shift occurs when carrier
2.9 InN
concentration is above the Mott critical density, meaning larger than the conduction band density, that is, the Fermi level lies in the conduction band. In such a case, electrons fill the bottom of the conduction band so that the apparent bandgap measured by optical absorption is increased by an amount of extension of the Fermi level. As a result, extra photon energy is required to excite electrons from the top of the valence band to the Fermi-level position within the conduction band suggested by Trainor and Rose [277]. But the required large electron concentrations produces bandgap renormalization, which is in competition with the Moss–Burstein effect. High doping concentrations cause band tailing effect, which acts to reduce the apparent bandgap. Layers of InN with large variation in electron concentration have been reported along with the measured bandgap, as compiled in Figure 1.35, as function of electron concentration. The measured bandgap is the convolution of Moss–Burstein shift and bandgap renormalization culminating in the measurement of EG þ EF, the EF being measured with respect to the conduction band edge. If the plot is limited to MBE films only, the data appear to support the argument for small bandgap and variation attributed to Moss–Burstein shift. However, when all the data in films prepared by any growth method in aggregate are plotted, there seems to be quite a scatter, including large bandgap associated with relatively lower impurity levels. The data basically do not show any trend in that a good many of the samples with high and low electron concentration both do exhibit the large bandgap value. The critical data having to do with the samples of relatively lower electron concentrations are associated with compensated samples. This is an important issue in that material compositional, or stain-related nonuniformities may also hamper efforts to determine the bandgap accurately. A critical view may raise question about the quality of the samples. What is clear is that the bandgap of the nondegenerate InN is still unknown and will require considerably more investigation [272]. What about some yet unknown defect being responsible for or the source of the 0.7 eV emission and absorption? Ask Butcher [267] who considered this possibility and provided arguments backed by experimental data in support of it. Butcher [267] suggested that the evidence for a 0.7 eV bandgap is also consistent with the presence of a 0.7 eV deep-level trap in that the available absorption data for the best published apparently 0.7 eV material (without the Moss–Burstein shift due to low carrier concentrations) exhibit an energy dependence consistent with a deep-level trap of |si-like orbital symmetry and was inconsistent with direct band-to-band transitions. Butcher [267] also made the case that the slope dependence of absorption coefficient plots is consistent with variations in the density of such a deep-level trap. Further, the 0.7 eV material showed that emission to be emanating from regions of indium-rich aggregates [270]. If one wants to be skeptical, one could argue that it is likely that the said emission could be a result of surface states at the metal–semiconductor interface rather than being associated with the InN band edge. It was also shown that material grown without such aggregates had an absorption edge nearer 1.4 eV [278]. The early InN data were produced by N-rich material, whereas the 0.7 eV material is grown under In-rich conditions and the same probably holds for OMVPE-grown layers. The early samples grown under N-rich conditions exhibited a higher than usual unit cell
j189
j 2 Electronic Band Structure and Polarization Effects
190
volume [279]. Such variation in the unit cell volume with stoichiometry would be consistent with that observed for GaN [280]. Being forced to grow InN at very low temperatures, typically below 550 C, with the caveat that the temperature measurements are not absolute, the driving force for desorption of excess In is not all that potent. The potency of In droplet formation has been noted by Yamaguchi et al. [281] who indicated that as little as a 20 C increase in temperature from optimum growth conditions can result in indium droplets. Clearly the stoichiometry, specifically In inclusion issue, is something that would need to be dealt with. With the evidence for indium-rich aggregates in the InN matrix and suggestion that they are responsible for the 0.7 eV luminescence properties of InN, the data of Wu et al. [282] has been discussed under a different light [267]. The samples investigated by Wu et al. had undergone an irradiation with 2 MeV protons to a dose of 2.23 · 1014 protons cm2 that resulted in a factor of 2 increase in the intensity of 0.7 eV PL peak, which was interpreted as radiation hardness. This increase in PL peak intensity can be ascribed to increased density of radiative defects, as has been well documented for III–V materials by Lang [283]. This is typically accompanied by band-edge photoluminescence that is either quenched or left unperturbed by radiation damage. Another interpretation of the observations of Wu et al. [282] might therefore be an increase in radiation-induced defects although not being promoted [267]. An issue is the fact that InN is particularly susceptible to nitrogen loss when bombarded with ions [284–287], leading to possible indium-rich aggregates. The aforementioned discussion lays the groundwork for a plausible connection between the 0.7 eV peak and In-rich aggregates, but does not quite attempt to discuss the nature of that emission. Shubina et al. [288] and also Ivanov et al. [278] attempted to do just the same, meaning shed some light into the nature of the 0.7 eV emission. The efforts of those authors utilizing microcathodoluminescence studies coupled with imaging of metallic In have shown that bright infrared emission at 0.7–0.8 eV arises in the close vicinity of In inclusions and is likely associated with surface states at the metal–InN interfaces. Employing thermally detected optical absorption (TDOA) measurements, Shubina et al. [288] suggested a bandgap near 1.5 eV, reserving a more definitive judgment until after more accurate measurements could be performed. Shubina et al. [288] have actually broadened the range of samples, examining various substrates including sapphire by choosing two representative sets of InN epilayers grown by both plasma-assisted molecular beam epitaxy and organometallic VPE methods. The dominant IR emission in these samples were observed to be in the range of 0.7–0.8 eV, independent of the growth technique used to prepare them and of excitation, such as optical with different laser lines, or by an electron beam in conjunction with a CL performed at 5 K in an analytical scanning electron microscope. No correlation between the IR emission and the electron concentration, which ranged from 2.1 up to 8 : 4 · 1019 cm3 (determined from IR ellipsometry measurements using an effective electron mass of me ¼ 0 : 11m0) [234] and measurement temperature was discernable, consistent with other reports [289]. Not all the 18 samples studied emitted light, and all the samples emitting IR PL emission did contain In-rich aggregates. Analytical microscopy sensitive to atomic weight in
2.9 InN
backscattered electrons (BSE) geometry, and energy dispersive X-ray (EDX) analysis along with CL were employed to establish a definite correlation between the In-rich aggregates and IR PL emission. Again, the bright 0.7–0.8 eV IR emission in both MBE and OMVPE sets of the samples was found to be associated with the In aggregates. Total optical extinction losses in a semiconductor matrix with metallic clusters have been established. In addition to the interband absorption in the matrix, those losses contain two additional components, namely, a bipolar absorption of radiation energy and its conversion into heat in small particles, and resonant scattering on plasmon excitations, which is important for larger particles [290]. Both characteristic components have been observed in the optical spectra obtained by Shubina et al. [288]. Those authors also observed emission in the range of 0.8–1.4 eV that was attributed to the scattered background signal, the root cause of which was most likely associated with nonchromatic spontaneous emission of the semiconductor laser used or the fluorescence of all optical components at high excitation power levels. Owing to electron beam excitation, the scattered signal was absent in the CL spectrum. Unless the aforementioned spurious signals are accounted for, erroneous conclusions could be drawn. Citing the absence of such signals, Shubina et al. [288] employed thermally detected optical absorption technique performed at 0.35 K. The method is based on the detection of a small increase in the sample temperature caused by phonons produced by nonradiative recombination processes as a result of optical absorption and bipolar absorption of light in In-rich aggregates. The TDOA spectra contained additional peaks below the principle absorption edge in films containing In-rich aggregates but not others, owing to absorption within those aggregates. The sharpness of the observed feature is related to the resonance at the extremely low measurement temperature that prevents thermal broadening induced by electron acoustic phonon scattering [288]. A natural conclusion of the aforementioned observations is that the strong IR absorbance is most likely associated with Mie [291] resonances due to scattering or absorption of light in InN-containing In aggregates or metallic In clusters. In the Mie theory [291], the extinction losses for a metallic sphere depend on the complex dielectric functions of both the matrix material, e, which is InN in this case, and metal, em, which is In in this case. Consequently, the resonance energy of the In clusters in InN matrix with a high-frequency dielectric constant of e0 ¼ 8.4 (based on stoichiometric InN, which is the case here as the InN films studied were In-rich) is considerably smaller than that in vacuum. For a treatment of the dielectric constant in InN, the reader is referred to Ref. [292]. Shubina et al. [288] argued that the In inclusions are predominantly formed either between columns or initiated at the interface with the substrate. Not knowing the shape and density of those clusters accurately, the authors employed a model developed for nonspherical metal particles in an absorbing matrix [293], which is based on the Maxwell–Garnett approximation for an effective medium [294], to demonstrate that the resonance absorption energy in the InN–In composite can shift down to the IR range depending on the In content and the shape of the In clusters. A lack of accurate knowledge of the InN complex dielectric constant and the specifics of In clusters prevented the authors to arrive at a
j191
j 2 Electronic Band Structure and Polarization Effects
192
shape and density of In clusters responsible for the IR absorption. Suffice it to say that given the available parametric data, it is very plausible that the observations are related to Mie resonance absorption caused by In-rich aggregates in the InN matrix. Clearly, additional experiments are imperative and will surely be available in due time. Even with the reported available data and analysis, Shubina et al. [288] argue that it is unlikely that the bandgap of InN is at 0.7–0.8 eV range. Confirming to a large extent the results obtained in InN reported in 1980s, using epitaxially grown wurtzite InN, Guo and Yoshida [184] measured low-temperature and room-temperature gaps of 1.994 and 1.97 eV, respectively, along with Varshni parameters of a ¼ 0.245 meV K1 and b ¼ 624 K. Estimates of the crystal field splitting in wurtzite InN range from 17 to 301 meV [66,107,163]. A value of 40 meV can be adopted. On the basis of the calculation, spin–orbit splittings vary from 1 to 13 meV [107,163], but a value of Dso ¼ 5 meV is recommended by Vurgaftman and Meyer [152]. Ironically, the small bandgap of InN actually goes against the long held cation rule, in which for isovalent, meaning common-cation semiconductors, the direct gap at the G point increases as the anion atomic number decreases. This implies, for example, that the bandgap of InN should be larger than that of InP, which is 1.4 eV, which is consistent with the values of 1.5 eV and higher reported for InN. It should, however, be stated that the breakdown of the common-cation rule is not unusual in ionic semiconductors. This is articulated in an interesting observation where in Nag [246] pointed out that this gap is unusually low in the context of trends exhibited by other semiconductor materials. Fore example, the bandgap of ZnO is also smaller than that of ZnS. This unexpected behavior has been attributed to two effects. Wei et al. [295] argued that a much lower 2s atomic orbital energy of N (18.49 eV) compared with P (14.09 eV) and other group V elements lowers the conduction band minimum at the G point. Moreover, the smaller bandgap deformation potential of InN (3.7 eV) in comparison to InP (5.9 eV) weakens the atomic size effect. The atomic size effect is the one that forms the basis of the common-cation rule. Similarly, the effect of the lower orbital energy of O as compared to the other group VI elements such as S is responsible for ZnO bandgap being smaller than that of ZnS. The controversy on the theory side matches that of the experiments. It could be argued that in the end carefully thought of experimental evidence will carry the day and theories with appropriate approximations will be developed to support the general direction of experiments. Tsen et al. [296] studying nonequilibrium optical phonons in a high-quality singlecrystal MBE-grown InN with picosecond Raman spectroscopy reached the conclusion that their results are not consistent with the large bandgap of InN. The basis for this argument is as follows: above gap photons cause creation of electron and hole pairs that very quickly relax toward the bottom/top of conduction/valence band by emitting phonons. For wurtzitic InN, using me ¼ 0:14 me [297], mh ¼ 1:63 me [298], hwL ¼ 2:34 eV, and Eg ¼ 1.89 eV gives an index of refraction n ¼ 3.0 (measured by Tsen et al. [296]) and hwLO ¼ 73:4 meV (corresponding to A1(LO) phonon mode energy); the phonon wave vector in the experiments of Tsen et al. [296] is q ¼ 2nkL 7.08 · 107 m1, where kL is wave vector of the excitation laser with photon energy
2.9 InN
2.34 eV. Owing to a much larger associated hole mass, the phonons emitted by holes can be ignored. In a parabolic band, excess electron energy is given by mh DE ¼ ðhwL E g Þx ffi 0:41 eV; ð2:44Þ me þ mh which is about five times the LO phonon energy hwLO ¼ 73:4 meV, which means that the energetic electrons are capable of emitting five LO phonons during their thermalization to the bottom of the conduction band. However, because of energy and momentum conservation during the electron–LO phonon interaction, there exists a range of LO phonon wave vectors that electrons can emit. As depicted in Volume 2, Figure 3.7, for an electron with wave vector ~ ke and excess energy DE, the minimum and maximum LO phonon wave vectors it can emit are given by [299] pffiffiffiffiffiffiffiffiffi 2me pffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð DE DE hwLO Þ; kmin ðor qmin Þ ¼ h and kmax ðor qmax Þ ¼
pffiffiffiffiffiffiffiffiffi 2me pffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð DE þ DE hwLO Þ: h
ð2:45Þ
Because of the nature of E–k relationship of the electron, the lower the electron energy, the larger kmin (also referred to as qmin) and the smaller kmax (also referred to as qmax) are. Therefore, at some electron energy, kmin of the LO phonon will be larger than the wave vector probed by the Raman scattering experiment, which is q ¼ 7.08 · 107 m1. When this occurs, the energetic electrons can no longer emit LO phonons with wave vectors detectable in the Raman scattering experiments. These conditions corresponds to kmin ffi 1.1 · 108 m1 and kmax ffi 2.4 · 109 m1 in the experiments of Tsen et al. [296]. Although the effect of nonparabolicity of the conduction band for InN is currently not known, the general trend is that the effective electron mass increases with the increase in energy in the conduction band for a typical semiconductor. By taking the nonparabolicity into account, the detectable kmin can be revised as being larger than 1.1 · 108 m1. Consequently, no nonequilibrium A1(LO) phonon population should be detected in the Raman experiments of Tsen et al. [296] when the effect of nonparabolicity of the conduction band is considered. This means that if the bandgap energy of InN were 1.89 eV, there would be no detectable nonequilibrium A1(LO) phonon population with the excitation laser photon energy of 2.34 eV. This contradicts the fact that nonequilibrium A1(LO) phonons have been observed and argues against the large InN bandgap. A similar argument can be applied to the scenario that the bandgap of InN is ffi0.8 eV, in the case of which the same laser photon energy (2.34 eV) can excite the electrons up in the conduction band so that it would take 6 A1(LO) phonons to be emitted for full thermalization. No A1(LO) phonons should be detected with 1.96 eV laser excitation. Both of the above arguments are consistent with the experimental observations. Earlier, it was stated that the proponents of the larger bandgap InN argue that the small bandgap reported for InN could be due to incorrect attribution of the deep-level emission to the band-edge emission as defects capture electrons or
j193
j 2 Electronic Band Structure and Polarization Effects
194
holes and emit photons (through a radiative relaxation process) of lower energy than the bandgap. This capture process can also emit phonons through a nonradiative relaxation process. Let us for a moment suppose that the 0.8 eV luminescence reported in the literature is due to such a capture process by some unknown deep-level defect(s) in InN. If so, the observations of Tsen et al. [296] of nonequilibrium A1(LO) phonons in InN with excitation laser having photon energy 2.34 eV suggests that nonradiative relaxation processes also play a role in the capture. However, due to deep-level defects having very localized wave functions, their momenta are widely spread due to the uncertainty principle. This suggests that electron–phonon interaction during the capture process does not need to conserve momentum. Therefore, phonon wave vectors of almost every magnitude can be emitted. In other words, the defect-model predicts that if one detects nonequilibrium A1(LO) phonons with 532 nm excitation then one should also detect nonequilibrium A1(LO) phonons with 634 nm excitation. This is in contradiction with the experimental observation of Tsen et al. [296]. In short, the nonequilibrium phonon experiments are consistent with the 0.8 eV bandgap. In a renewed effort to determine the fundamental gap of InN, Arnaudov et al. [249] studied the shape and energy position of near band edge photoluminescence spectra of InN epitaxial layers with different doping levels. They implied that samples with high doping concentration have been used to infer the bandgap of InN without properly accounting for the effects of band filling, band nonparabolicity, and electron–electron and electron–impurity interactions. In addition to usual difficulties associated with highly doped samples, another aggravating factor is a clear lack of convergence in the value of the electron effective mass. Properly accounting for the aforementioned effects, Arnaudov et al. suggest a bandgap of Eg ¼ 692 2 meV for an effective mass at the conduction band minimum mn0 ¼ 0.042m0. They also argue that the value of Eg ¼ 0.69 eV reported in Ref. [270] extracted from the absorption and photoluminescence spectra of samples with carrier concentration n > 6 · 1018 cm3 within the band-to-band recombination model while taking into account the Burstein–Moss shift and bandgap renormalization due to many-body effects agrees with the data; the value used for the effective mass at the bottom of the conduction band, mn0 ¼ 0.1m0, is inconsistent with the universal Kanes relation (mn0; Eg). This relation predicts a bandgap energy of 1.7 eV for mn0 ¼ 0.1m0, and to obtain near to 0.7 eV bandgap one must use a much smaller effective mass of mn0 ¼ 0.042m0. Interestingly, the same value of Eg ¼ 0.69 eV was obtained by fitting the absorption spectra, this time in a lightly doped sample with n ¼ 3.5 · 1017 cm3, with a sigmoidal function that includes only the band tailing effect and does not involve any value for the effective mass [300]. Taking advantage of improved crystalline quality and utilizing samples with electron concentrations in the range of 7.7 · 1017–6 · 1018 cm3 Arnaudov et al. [249] undertook the task of determining the bandgap on InN from optical data but with the interpretation of the emission spectra in such highly conducting layers in terms of the free-electron recombination band (FERB) model, which has been previously reported in the context of GaAs [301,302], InP [303], InSb [304], and GaN [305]. By analyzing not only the emission energy position but also the shape of the spectra simultaneously and taking into consideration the specifics associated with both high and low electron concentrations, Arnaudov
2.9 InN
E
1.0
Ec
0.8 PL intensity (au)
n = 1.7 × 1018cm–3
E F = E Fn Degenerate band tails G* = E
0.6
Ev
0.4
n = 6 × 1018 cm–3
Fp
gn ,gp
0.2
Eg -G2*Eg-G1* Eg EF2
0.0 0.5
0.7
0.6
EF 3 0.8
Energy (eV) Figure 2.13 Experimental, depicted with symbols, and calculated, depicted with solid lines, PL spectra of samples with 1.7 · 1018 and 6.0 · 1018 electron concentrations as measured by Hall effect. The inset schematically depicts the recombination mechanism between the
degenerate electrons in the conduction band DOS to the level G* in the valence band tails as relied on in modeling. The energy positions representing the best fits for Eg, EF, and the intrinsic, or the unperturbed, bottom of the conduction band Eg–G* are also indicated [249].
et al. [249] are able to determine the fundamental bandgap for the electron effective mass in InN as Eg ¼ 692 2 meV for an effective mass at the bottom of the conduction band mn0 ¼ 0.042m0, which is consistent with Kanes relation. PL spectra of two samples with electron concentrations of 1.7 · 1018 cm3 6 · 1018 cm3, as determined by Hall measurements, by Arnodov et al. [249] are shown in Figure 2.13 (points)). Both samples exhibit a broad emission band with a maximum at 685 meV and 705 meV, respectively, the former for the lower doped and the latter for the higher doped sample due to a larger Burstein–Moss shift. Noteworthy, however, is that the emission band of the sample with a higher Hall concentration is broader and more asymmetric, and its PL peak is at a higher energy due to the larger Burstein–Moss shift. The low-energy side of the spectral band of the sample with higher electron concentration shifts to lower energy compared to the sample with lower Hall concentration, which tends to narrow the apparent optical bandgap, consistent with emission spectra from highly doped semiconductors. In addition, these observations are characteristic of free to bound recombination of degenerate conduction electrons with nonequilibrium valence holes in the valence band tail, as shown in the inset of Figure 2.13 [301–305]. Moreover, the shape of the emission bands follows the energy distribution of electrons in the conduction band, all the while their energy positions are determined by the interplay of the equilibrium Burstein–Moss shift (blue shift) and the effective bandgap renormalization (red shift).
j195
j 2 Electronic Band Structure and Polarization Effects
196
To deduce the bandgap energy Arnaudov et al. [249] fit the experimental emission spectra with that obtained from the general expression for the intensity versus the photon energy I(hn) given by (neglecting the energy dependence of the probability for radiative transitions, as was done for heavily doped GaN [305]) ð¥ ð¥ IðhnÞ g n ðE Fn Þf n ðE n E Fn Þg p ðE p Þf p ðE p E Fp ÞdðE n E p hnÞdE n dE p ; 00
ð2:46Þ where gn(En) and gp(Ep) are the density of states in the conduction and valence bands at electron and hole energies En and Ep, respectively. The terms fn and fp represent the associated nonequilibrium Fermi–Dirac functions, and EFn and EFp are the quasiFermi levels for electrons and holes. The conduction band density of states gn(En), as well as the electron effective mass mn(En), can be calculated in the framework of Kanes two-band kp model as described in Ref. [300]. Although the bandgap information is implicit in the En Ep term of the delta Dirac function, the bandgap term Eg is sometimes explicitly subtracted from En Ep for emphasis, as done by Arnaudov et al. [249]. The reader is referred to Volume 3, Chapter 2 for a detailed discussion of spontaneous emission intensity calculations. Because the nonequilibrium holes are situated in a relatively narrow energy window deep in the band tails, they do not significantly affect the spectral distribution of the emitted light. Therefore, the quantity gn(En) fn(En EFn) in Equation 2.46 roughly reproduces the shape of the FERB shown in the upper part of the inset in Figure 2.13. Likewise, the term gp(Ep) fp(Ep EFp) in Equation 2.46 determines primarily the energy position of the emission band associated with the unperturbed fundamental bandgap Eg via Equation 2.47, as depicted in the lower part of the inset in Figure 2.13. It should be pointed out that the FERB model includes a calculation of the spectral shape as well as analytical renormalization of the bandgap due to the presence of ionized impurities. In this case, the energy positions of both the low- and high-energy slopes of the emission band are sensitive to the electron concentration induced by ionized impurities. In conjunction with the nonequilibrium Fermi–Dirac function of electrons, fn, Arnodov et al. [249] used the Fermi level EFn ffi EF corrected for the temperature of electrons y, which can differ from the lattice temperature T, electron–electron and electron–impurity interactions [305], and the nonparabolicity of the conduction band density of states (DOS), which can be determined in framework of the two-band kp model. The valence band DOS gp(Ep) is replaced by a Gaussian, determining the tails deep in the bandgap through the root mean square (rms) impurity potential G as detailed in Ref [305]: pffiffiffi 4pe2 ðN i R3s Þ1=2 ; G¼2 p eRs with Rs ffi
ð2:47Þ
aBe e h2 ðna3Be Þ 1=6 and aBe ffi ; 2 4pe2 4pmn0
where the terms e represent in order the electric permittivity, Rs the Thomas–Fermi screening length, Ni ¼ [(1 þ K)/(1 K)]n the total ionized impurity concentration,
2.9 InN
K the compensation ratio, n the extrinsic electron concentration, and aBe the effective Bohr radius of electrons. The value of Rs is smaller than aBe and thus the equilibrium and nonequilibrium degenerate electrons are free above the bottom of the conduction band [305]. The situation with the nonequilibrium holes is quite different in that in III–V materials the effective Bohr radius aBh is much smaller than Rs due to the relatively large hole effective mass mp. Moreover, holes are classically localized, at least for not extremely high impurity concentrations and high temperatures [306], at the potential minima of the valence band tails near the thermal equilibrium level pffiffiffi G ¼ E v þ 2G kT=2. As shown for heavily doped GaN [305], the level G* plays the role of the quasi-Fermi level in the nonquasi-equilibrium recombination FERB model. Thus, we can replace the value of EFp in the Fermi–Dirac function for holes fp can be replaced by G*, meaning one case set EFp ¼ G*. To comment on the electron effective mass, the FERB emission spectra of samples with 1.7 · 1018 and 6.0 · 1018 cm3 Hall electron concentration can be calculated with varying n, y, and K and using the value of mn0 ¼ 0.042m0 suggested in Ref. [307]. For the relative static permittivity, a value of e ¼ 14.61 was used by Arnaudov et al. [249]. Assuming a zero-compensation ratio, which is a good first-order approximation, the best fits of the spectra are obtained with bandgap values of 690 and 692 meV, for the sample with 1.7 · 1018 cm3 electron concentration and the sample with 6.0 · 1018 cm3 electron concentration, respectively. To reconcile the small difference in the aforementioned bandgap values, a small compensation ratio of K ¼ 0.06 for sample with 1.7 · 1018 cm3 electron concentration and K-value of 0.01 for sample with 6.0 · 1018 cm3 electron concentration was introduced. Doing so resulted in a bandgap valued of Eg ¼ 692 meV for both samples. The best fit values for the two samples are shown in Figure 2.13 (solid lines). It should be noted that best fit values of the electron concentration nopt are noticeably lower than those deduced by the measured Hall effect. This dispersion may perhaps be related to the inhomogeneities in the films. In spite of this both represent degenerate cases because the Motts transition concentration is estimated to be about nMott ¼ 5 · 1016 cm3 for mn0 ¼ 0.042 m0. The calculated curves agree very well with the experimental spectra, with the exception for the low-energy range. In this region, an additional contribution from a deeper emission center could in principle be possible, which is not included in the model. In the high-energy portion of the spectrum, the accuracy is more reliable. To summarize the above discussion, using the shape as well as the energy position of the near band edge PL spectra of InN epitaxial layers with different doping levels, Arnaudov et al. [249] concluded that the radiative transition is between the degenerate electrons in the conduction band and nonequilibrium holes in the valence band tails and that the fundamental bandgap of InN is Eg ¼ 692 2 meV for an effective mass at the conduction band minimum of 0.042m0, which is consistent with the Kane model. The optical transmission and reflection data obtained from a high-quality InN film grown by MBE with N-polarity support the smaller bandgap figures of InN as shown in Figure 2.14. The growth details and transport properties of InN layers similar to the one that led to the data presented in Figure 2.14 are discussed in Section 3.5.13.
j197
j 2 Electronic Band Structure and Polarization Effects
198
70
100
80
65
60
60
40
55
20
Reflection (%)
Transmission (%)
E g = 0.7– 0.75 eV
50
0 45 1.0
1.5
2.0
2.5
3.0
3.5
Wavelength (μm) Figure 2.14 Optical reflection and transmission data obtained in an N-polarity InN film grown by MBE indicative of a bandgap between 0.7 and 0.75 eV. Courtesy of A. Yoshikawa.
As for the true value of bandgap of wurtzitic InN, although the data for small bandgap data are convincing and the arguments for the small bandgap are compelling, for some reason, some controversy remains. This controversy is expected to evaporate when and if the large bandgap observed is earlier and some InN samples are explained satisfactorily. We should reiterate that the data in high-quality samples converge on the small bandgap. However, there is still some dispersion in the exact bandgap value that seems to be between 0.65 and 0.8 eV. In this chapter, the arguments for both the large and small bandgap as well as the pitfalls for each in terms of the reliability of the data and a historical review are provided for the reader to be abreast with the conflicting issues surrounding the matter. Turning our attention momentarily to other electronic properties affected by the band structure, measurements of the electron effective mass in wurtzitic InN produced values of 0.11m0 [234], 0.12m0 [308], and 0.14m0 [297], as well as 0.24m0, for the mass perpendicular to the c-axis [309]. Kasic et al. [297] used infrared spectroscopic ellipsometry and micro-Raman scattering to study vibrational and electronic properties of wurtzitic in 0.22 mm thick InN layers grown by RF MBE, as well as Hall effect measurements, and arrived at the isotropically averaged effective electron mass of 0.14m0. The mass value of 0.14m0 closely matches at least one theoretical projection [53]. It should be mentioned that all the InN films used for these investigations featured very high electron concentrations, which are endemic to InN, in the 1018 cm3 or higher, and causes the Fermi level to degenerate well in the conduction. Consequently, any nonparabolicity in the conduction band would affect the effective mass measurements. However, the realization that the InN bandgap is narrower than that previously thought prompted a reexamination of the effective mass issue also [310]. Accounting for the substantial nonparabolicity that can cause
2.9 InN
an overestimate of the mass because high doping leads to a band-edge effective mass of 0.07m0, which is what is recommended here as was done by Vurgaftman and Meyer [152]. Turning our attention to holes, valence band mass parameters have been calculated by Yeo et al. [67] using the empirical pseudopotential method, and also by Pugh et al. [163] and Dugdale et al. [153] using more or less the same technique. The results of the first two investigations are quite similar, Pugh et al. [163] employed three different levels of computation comprising first-principles total energy calculations, semiempirical pseudopotential calculations and kp calculations. Band structures were obtained from each method in a consistent manner and were used to provide effective masses and kp parameters. These parameters are useful in investigating the electronic structure of alloys and quantum well heterostructures. These valence band parameters are the recommended values with the caveat in that the lower InN energy gap may require a downward revision of the light-hole mass. The parameters concerning the bandgap-related issues for wurtzitic InN recommended by Vurgaftman and Meyer [152] are tabulated in Table 2.19. A compilation of the dispersion in the effective mass for both the conduction band and various valence bands as obtained by various computational methods, as well as parameters used in the description of the bandgap for wurtzitic InN, particularly, in the context of empirical pseudopotential method, as described in Ref. [55], are tabulated in Table 2.20. Christensen and Gorczyca [31] predicted a hydrostatic deformation potential of 4.1 eV for wurtzite InN, which compares to a smaller value of 2.8 eV calculated by Kim etal. [180].Inthe absence of any predilection forany of the tworeports, averaging the two leads to a ¼ 3.5 eV. In the absence of any calculations of the valence band deformation potentials, appropriating the parameter set specified above for GaN could be a good default at this point. The elastic constants measured by Sheleg and Savastenko [81], early on more refined values arrivedby calculations suchasthe set reported by Wright [179], are available, which are C11 ¼ 223 GPa, C12 ¼ 115 GPa, C13 ¼ 92 GPa, Table 2.19 Recommended band structure parameters for wurtzitic InN [152].
Parameter
Value
Parameter
Value
Parameter
Value
Eg (eV, low temperature) a (meV K1) b (K) Dcr (meV) Dso (meV) == me =m0 m? e =m0 a1 (eV) a2 (eV)
1.5–1.8 0.245 624 0.040 0.005 0.07 0.07 3.5 3.5
A1 A2 A3 A4 A5 A6 A7 (meV Å) d13 (pm V1) d33 (pm V1) d15 (pm V1) Psp(C m2)
8.21 0.68 7.57 5.23 5.11 5.96 0 (default) 3.5 7.6 5.5 0.042 (0.041)
D1 (eV) D2 (eV) D3 (eV) D4 (eV) D5 (eV) D6 (eV) c11 (GPa) c12 (GPa) c13 (GPa) c33 (GPa) c44 (GPa)
3.7 4.5 8.2 4.1 4.0 5.5 223 115 92 224 48
See Tables 2.27 and 2.28 for details related to the elastic constants, piezoelectric constants, and spontaneous polarization charge. Any dispersion among the tables is a reflection of the uncertainty in the available parameters.
j199
j 2 Electronic Band Structure and Polarization Effects
200
Table 2.20 Effective masses and band parameters for wurtzitic InN.
Parameter ==
me m? e == mhh == mlh == mch m? hh m? lh m? ch A1 A2 A3 A4 A5 A6 A7 D1
anisotropic
isotropic
A
B
C
D
0.138 0.141 2.438 2.438 0.140 2.661 0.148 3.422 7.156 0.244 6.746 3.340 3.208 4.303 0.072 0.214
0.137 0.140 2.493 2.493 0.137 2.599 0.157 1.446 7.298 0.441 6.896 3.064 3.120 3.948 0.103 0.084
0.11 0.10 1.56 1.56 0.10 1.68 0.11 1.39 9.62 0.72 8.97 4.22 4.35
0.11 0.10 1.67 1.67 0.10 1.61 0.11 1.67 9.28 0.60 8.68 4.34 4.32 6.08 0
0.10 0.10 1.431 1.431 0.106 1.410 0.196 0.209 9.470 0.641 8.771 4.332 4.264 5.546 0.278 0.0375
0.10 0.10 1.350 1.350 0.092 1.449 0.165 0.202 10.841 0.651 10.100 4.864 4.825 6.556 0.283 0.0375
0
Effective masses in units of free-electron mass m0, Luttinger-like parameters Ai (i ¼ 1, . . ., 6) in units of h2 =2m0 , and A7 in units of eV Å. The crystal field splitting energy D1 is given in units of meV. The term aniso represents the values derived using a band structure calculation with anisotropically screened model potentials, whereas the term iso describes a comparative band structure calculation on the basis of isotropically screened model potentials using an averaged e0 value by taking the spur of the dielectric tensor [55]. Anisotropically screened and isotropically screened values are from Ref. [55]. A: effective masses and Ai are from Ref. [67] obtained through a line fit to the band structure; B: direct kp calculation in a 3D fit from Ref. [67]; C: Ai from Ref. [153] obtained through a Monte Carlo fitting procedure to the band structure and effective masses calculated using Equations 2.30 and 2.31; D: direct kp calculations for Ai from Ref. [153] and effective masses calculated using Equations 2.30 and 2.31.
C33 ¼ 224 GPa, and C44 ¼ 48 GPa. Other sets of parameters calculated by Kim et al. [180] and Davydov [201] are also available in the literature. For a more detailed discussion of elastic and piezoelectric coefficients as well as polarization issue, refer to Section 2.12 and Tables 2.25–2.27 and 2.28. Owing to the fact that the piezoelectric coefficients in InN have apparently not been measured, for consistency the theoretical values of Bernardini and Fiorentini [210], d33 ¼ 7.6 pm V1, d13 ¼ 3.5 pm V1, and d15 ¼ 5.5 pm V1, are suggested. In spite of the fact that the spontaneous polarization data for GaN/GaInN structures are not as conclusive as one would like at this point, most likely owing to relatively poor material quality, the value Psp(InN) ¼ 0.042 C m2 is consistent with a thorough comparison of experiment and theory [84]. Recommended band structure parameters for wurtzite InN are compiled in Table 2.19 [152]. 2.9.2 Zinc Blende InN
Experiments on zinc blend InN are very rare although this polytype has been reported [311]. The bulk of the reports comprise theoretical estimates of its band
2.9 InN Table 2.21 Luttinger parameters g1, g2, and g3 for zinc blende InN
obtained from a fit along the [1 1 0] direction along with those available in the literature, as compiled in Ref. [55]. Parameter
A
B
c1 c2 c3
7 0.97 1.22
3.27 1.26 1.63
A: empirical pseudopotential calculation by Fritsch et al. [55]; B: recommended values taken from Ref. [152].
parameters. The zinc blende variety has been projected to have a direct band alignment, with G-, X-, and L-valley gaps of 1.94, 2.51, and 5.82 eV, respectively [70]. However, this particular calculation was performed before the controversy in the bandgap of wurtzitic InN. If the arguments available in the literature and presented in the Section 2.9.1 dealing with wurtzitic InN were to hold in favor of the large bandgap, then the aforementioned bandgap values will be more credible. For the valence band, spin–orbit splittings in the range 3–13 meV have been projected [107,171,227]. Vurgaftman and Meyer [152] recommends the 5 meV value among them. As in the case of the bandgap, the effective mass value for the wurtzitic polytype is recommended for the zinc blende variety, that is, 0.07m0. As mentioned in the previous section, this value is arrived at after the nonparabolicity effects are accounted for. To reiterate, the range of values reported for the Wz InN is 0.10–0.14m0 [70,163,173]. The longitudinal and transverse masses for the X valley have been calculated to be 0.48m0 and 0.27m0, respectively [70]. The recommended Luttinger parameter set by Vurgaftman and Meyer [152] is g1 ¼ 3.72, g2 ¼ 1.26, and g3 ¼ 1.63, which is derived from the work of Pugh et al. [163], and the split-off mass is chosen to be mso ¼ 0.3m0 [70,163]. These as well as the other literature values of the Luttinger parameters for zinc blende InN are listed in Table 2.21. Fritsch et al. [55] calculated the effective masses for conduction and valence bands, the latter involving the light and heavy holes, as well as the spin–orbit split-off mass, which along with the anisotropy taken into account are listed in Table 2.22 for zinc blende InN. The data contain those obtained by FP-LAPW, EPM calculations, and those calculated with Luttinger parameters Equations 2.42 and 2.43. The EP parameter value given by Meney et al. [173] is more likely because the alternative value by Pugh et al. [163] implies too large a value for F. The resulting parameter set is EP ¼ 17.2 eV and F ¼ 4.36. For the hydrostatic deformation potential, an average value of 3.35 eV from the theoretical [31,180] candidates of 2.2 to 4.85 eV has been recommended [152]. The valence band deformation potentials listed in Table 2.19 are compiled from the calculations of Wei and Zunger [107], Kim et al. [180], Tadjer et al. [70], and Van de Walle and Neugebauer [177] are av ¼ 0.7 eV, b ¼ 1.2 eV, and d ¼ 9.3 eV. Elastic constants of C11 ¼ 187 GPa, C12 ¼ 125 GPa, and C44 ¼ 86 GPa are assumed from the calculations of Wright [179],
j201
j 2 Electronic Band Structure and Polarization Effects
202
Table 2.22 Effective masses for electrons (e), heavy holes (hh), light holes (lh), and spin–orbit split-off holes (so) in units of the free-electron mass m0 along the [1 0 0], [1 1 1], and [1 1 0] directions for zinc blende InN. [100]
[100]
[111]
[111]
[110]
[110]
Reference
me
mhh
mlh
mhh
mlh
mhh
mlh
mso
A B C
0.13 0.12 0.10
1.18 0.83 2.18
0.21 0.16 0.89
2.89 0.83 2.29
0.19 0.16 0.93
2.12 1.55 3.10
0.20 0.15 0.79
0.36 0.30 0.30
Compiled by Fritsch et al. [55]. A: after Ref. [55]; B: calculated from recommended Luttinger parameters from Ref. [152], using Equations 2.42 and 2.43; C: empirical pseudopotential calculation from Ref. [70].
Table 2.23 Parameters associated with the band structure for zinc
blende InN bandgaps are for low temperature [152]. Parameter
Value
Parameter
Value
Parameter
Value
alc (Å) at T ¼ 300 K E Gg ðeVÞ a (G) (meV K1) b (G) (K) E Xg ðeVÞ a (X) (K) b (X) (meV K1) E Lg ðeVÞ a (L) (meV K1) b (L) (K)
4.98 0.78 1.5–1.8 0.245 624 2.51 0.245 624 5.82 0.245 624
Dso (eV) me ðGÞ ml ðX Þ mt ðX Þ c1 c2 c3 mso
0.005 0.07 0.48 0.27 3.72 1.26 1.63 0.3
EP (eV) F VBO (eV) ac (eV) av (eV) b (eV) d (eV) c11 (GPa) c12 (GPa) c44 (GPa)
17.2 4.36 2.34 2.65 0.7 1.2 9.3 187 125 86
which are similar to other calculated sets [180,181]. The recommended band structure parameters for zinc blende InN are compiled in Table 2.23.
2.10 Band Parameters for Dilute Nitrides
Dilute nitrides can be described as conventional III–V compound semiconductors, wherein a small N fraction on the order of a small percentage is added. Addition of even very small amounts of N causes substantial changes to the bandgap and the lattice contact of the host compound to the point that standard bowing parameters for the bandgap variation linear interpolation of lattice constants between the host material and zinc blende GaN or GaInN do not apply. A single bowing parameter is inadequate even if the goal is only to describe the energy gap for a relatively wide range of compositions [312]. The technological driving force is really compelling in
2.10 Band Parameters for Dilute Nitrides
that wavelengths of interest for short and long haul communications systems can be obtained on GaAs technology by adding small fractions of N into the GaAs or InGaAs lattice. The motivation here is in materials incorporating only a small percentage of nitrogen [313], because it is highly questionable whether more than 10–16% N can be incorporated stably. Even in cases when one can, the quality is very inferior. Annealing techniques employed to improve layer quality end up causing N segregation, reducing the fraction in the bulk. Having mentioned this, it should also be pointed out that N isoelectronic doping of compound GaP, which has an indirect bandgap, for LEDs [314] predates the flurry of activity aimed at imparting substantial changes in the band structure, which is the topic of this section. The band structure of dilute nitride compound semiconductors has been reviewed in the literature by Vurgaftman and Meyer [152]. The treatise here follows the same philosophy in that the N-containing GaAs followed InP is treated before segueing onto other compound semiconductors such as those based on Sb anion. The properties of the conduction band in dilute nitride semiconductors can be described in terms of the band anticrossing (BAC) model [315]. It should be mentioned that a small percentage of N usually has little effect on the valence bands. This twoparameter anticrossing model can be spun formally in terms of the many-impurity Anderson model within the coherent potential approximation. It can also be thought of as the interaction between a single, spatially localized N level and the conduction band of the underlying traditional As-, P-, Sb-based compound semiconductors. Alternatively, Lindsay et al. [316] predicted the identical fundamental bandgap by assuming that the interaction involves a weighted average of perturbed upper states as opposed to a single N level. If one neglects the effect on the valence bands completely, the energy dispersion relation for the two-coupled bands within the BAC model can be expressed as qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1 C E ðkÞ ¼ ð2:48Þ E ðkÞ þ E N ½E C ðkÞ þ E N 2 þ 4xV 2 : 2 Here, EC(k) is the conduction band dispersion of the nominal nonnitride semiconductor (e.g., GaAs in the case of GaAsN), EN is the position of the nitrogen isoelectronic impurity level in the nonnitride semiconductor, V is the interaction potential between the two bands, and x is the N mole fraction. The energy dispersion relation for the twocoupled conduction bands in GaAs0.99N0.01 showing the characteristic anticrossing is plotted in Figure 2.15. As can be gleaned from the band structure treatment for standard nitride semiconductors, any temperature dependence arises from the shift of the conduction band dispersion EC(k) not that of the valence band. In the dilute semiconductor case also, the bandgap is assumed to follow the Varshni formula of Equations 2.34 and 2.35, but with EN taken temperature independent. One consequence of this assumption is a considerable weakening of the variation of the fundamental energy with temperature [317]. The weak shift of EN with applied pressure necessitates a fresh look at the deformation theory. The strain dependence of the E transitions can be determined by substituting the applicable deformation potentials of the host nonnitride semiconductor to obtain EC(k), followed by deriving E (k) from Equation 2.48. The implementation of Equation 2.48 requires input of
j203
j 2 Electronic Band Structure and Polarization Effects
204
2.4
T = 300 K
GaAs0.99N0.01
Energy (eV)
2.2
E+
2.0
1.8 EN 1.6
EC E–
1.4
1.2 –0.15
–0.10
–0.050
0.0
0.050
0.100
0.15
Wave vector (1 Å–1) Figure 2.15 Conduction band dispersion relations for GaAs0.99N0.01 at 300 K from the band anticrossing model (BAC) [319] (solid curves). For comparison, the unperturbed GaAs conduction band and the position of the nitrogen level are shown as the dotted and dashed curves, respectively [152].
band parameters of the host semiconductor. Although, we do not reproduce the nonnitride parameters in this work to preserve space, all of the required values are tabulated in a review [92]. The more detailed kp theory can also be applied. The BAC model can be extended to treat spin-doubled conduction bands, valence band, and nitrogen impurity bands by modifying the eight-band kp theory [318–321]. Coupling of the nitrogen band to the X and L valleys have also been introduced [322,323] but at the expense of additional complexity, which may not be warranted. Unless compelled, it is preferable to stick to the simple two-parameter fit of Equation 2.48. It has been pointed out that the fixed position of the nitrogen level with respect to the vacuum level implies a tandem shift with the valence band maximum of the host nonnitride material. To account for the experimental observation of nonnegligible deviations from referencing to the valence band offset, a separate nitrogen level for each host material is typically specified. Vurgaftman and Meyer [152] suggest that the valence band offset for an unstrained dilute nitride be set equal to that of the host semiconductor. As mentioned above, within the realm of the BAC model, the primary effects of the nitrogen are on the conduction band. Further,
2.10 Band Parameters for Dilute Nitrides
even the 10-band model does not shift the valence band maximum in the absence of strain or quantum confinement, while of course influencing the hole dispersion relations. Although a finite type-I or type-II offset in strained structures have been reported on the basis of experiments, they are not sufficiently compelling to deviate from a null offset relative to the host in the absence of strain. Despite its simplicity, Equation 2.48 provides a basis for describing material properties, such as the fundamental energy gap, which are governed by the transition from E to the top of the valence band, the temperature dependence of the gap, the electron effective mass, and the characteristics of the upper band Eþ. It must be kept in mind that within the theory of Lindsay et al. [316], there is not necessarily a single well-defined Eþ band. Somewhat in similar vein, the extent to which the BAC representation may be considered fundamentally realistic is still a matter of active discussion [315,324]. Clearly, the BAC model considers only a single nitrogen level on a substitutional lattice site or a narrow impurity band formed from such levels. In the process, it neglects mixing with the L and X valleys, and more complex nitrogen behavior in the semiconductor such as the formation of nitrogen pairs and clusters. In contrast, the more complicated pseudopotential calculations that consider some of these issues are computationally demanding [325–327]. In addition, the numerical results do not lend themselves to simple formalism such as Equation 2.48. Buttressing the simpler BAC approach is the discovery by Lindsay et al. [316] that the verifiable prediction of the dependence of the bandgap on the N content may be unaffected by generalizing to a multiplicity of higher lying states. Let us now turn our attention to specific dilute nitride semiconductor systems. 2.10.1 GaAsN
In conjunction with early investigations motivated by visible LED development, it has been known that small quantities of nitrogen in GaAs and GaP form deep-level impurities [314]. However, it has only been after the advent of nitride semiconductors that also paved the way for investigating traditional compound semiconductors, such as GaAs with N content definitely beyond the quantities ( 1% or more) used for doping experiments [328,329]. On the flip side of this, reports [330] of the incorporation of small amounts of As into GaN are neither common nor easy. The presence of As in GaN has been stated to cause modified surface reconstruction and/or act as surfactant or be a source of dopant impurities. Even though care must be exercised, PL measurements can be used to determine the bandgap, provided the relative position of the particular transition energy is known, and have been applied to GaAsN with N fractions up to 1.5% in an effort to determine the dependence of the energy gap on N composition [331]. As compared to other ternaries, a large bandgap bowing parameter of 18 eV was found, which for small compositions is equivalent to a linear model [332]. Early theoretical studies projected bandgap bowing parameters based primarily on the dilute nitride semiconductor with large N content [333–341]. Although the large bowing parameter was originally supposed to produce a semimetallic overlap at intermediate
j205
j 2 Electronic Band Structure and Polarization Effects
206
compositions [336], more detailed investigations that followed led to reduced bowing parameter with increasing composition [338,339], with experimental backing based on investigations of Bi and Tu [342], who studied N compositions as large as 15%. In discord with the above-mentioned linear dependence of the bandgap for small N fractions, subsequent investigations pointed to a highly nonlinear reduction in the energy gap for small N compositions [343–349]. Another noteworthy discovery was the realization of a significant weakening of the temperature and pressure dependencies of the bandgap for GaAsN and also GaInAsN with small In fractions [346,350,351]. In aggregate, it is clear that a simple bowing approximation could not adequately describe the GaAsN alloy. In this vein, Shan et al. [352] proposed the band anticrossing model and confirmed a weak pressure dependence for the nitrogen band transitions with a deduced deformation potential of 1.2 eV. To put matters in context, the density functional calculation of Jones et al. [353,354] also predicted reduced pressure dependence without invoking the BAC model. Additional report supporting the BAC model has been the finding by Skierbiszewski et al. and others pointing to a significantly heavy electron mass in GaInAsN [355–361]. On the anticorrelation side is another set of measurements by Young et al. [362] who found a reduction in the effective mass with increasing N content, in direct conflict with the BAC model that predicts an increase even at the zone center. Increasing mass with increasing N content, however, appears to hold. Consistent with the overall expected behavior of dilute nitride semiconductors based on the BAC model, the temperature dependence of the bandgap has been confirmed to be notably weaker in GaAsN than in GaAs [317,363]. In a similar vein, GaAsN film electroreflectance [364] resolved both the E and Eþ transitions, the band description of which can be seen in Figure 2.15. Extending to a ternary, the bandgap reduction was also observed in nitrogen-implanted Al0.27Ga0.73As samples [365]. The transition between the doped and alloyed materials was studied by Zhang et al. [366,367]. Here, doped implies quantities manifesting themselves as dopants without radical changes incurring on the host material. The alloy material implies that the N concentration is sufficiently high to cause substantial changes in the host material. Zhang et al. observed evidence for impurity banding at N concentrations as small as 0.1% N. Zhang et al. also proposed an alternative method for the characterization of the bandgap energy, which is not based on the BAC model [368]. This transition point was quantified as 0.2% by Klar et al. [369]. Figure 2.16 depicts the fundamental energy bandgap that is between the valence band maximum and the E conduction band minimum, as a function of N fraction, x, for GaAs1xNx at 300 K. Also shown is a curve with a constant bowing parameter of 18 eV (dotted line) along with another incorporating a variable bowing parameter of (20.4–100x) eV, as suggested in a review by Vurgaftman et al. [92] (dashed line). Ostentatiously, the BAC model predicts a substantially higher energy gap beyond the N fraction of 1.5%. The available experimental data (points in Figure 2.16), compiled in Ref. [315], show much better consistency with the BAC model than with either of the two utilizing bowing parameters. It should be reiterated that GaAs1xNx alloys with x > 5% become increasingly difficult to grow while retaining quality. As such, compositions with large N fraction
2.10 Band Parameters for Dilute Nitrides
GaAs 1.4
Bandgapenergy (eV)
GaAs1–xNx 1.2
1.0
BAC 0.8
C = 20.4 –100 xeV C = 18 eV Experimental data
0.6 0.00
0.01
0.02
0.03
0.04
0.05
N mole fraction, x Figure 2.16 Energy of the fundamental bandgap in GaAsN as a function of nitrogen concentration x (a) from the BAC model (solid curve), (b) using the variable bowing parameter from the review of Vurgaftman et al. [92] (dashed curve), and (c) using a constant bowing parameter (dotted curve). For comparison, the available experimental data as compiled in Ref. [315] are also plotted (circles) [92,152].
may not have the technological significance [313]. On the theory side, no upper limit on the N compositions beyond which the BAC model becomes invalid has been reported. Despite some spread in the reported values for the two primary parameters EN, 1.65–1.71 eV, referenced with respect to the GaAs valence band maximum, and V (2.3–2.7 eV), the values reported by Shan et al. [352], namely, EN ¼ 1.65 eV, V ¼ 2.7 eV, are suggested. In discord with the conventional BAC model, which assumes that the addition of N has little effect on the valence bands, two reports noted a larger than expected heavy-/ light-hole splitting in GaAsN containing a small percentage nitrogen [370,371]. This implies a strong bowing in the valence band shear deformation potential b, although the increase in the deformation potential is inconsistent in these two reports. Expanding their earlier work, Egorov et al. [372] again observed strain-induced splitting of light-hole and heavy-hole bands of tensile-strained GaAsN. The observed dependence of the bandgap in unstrained GaAsN on the nitrogen content differed substantially from that predicted by the theory assuming that the bandgap in GaAsN
j207
j 2 Electronic Band Structure and Polarization Effects
208
can be reduced to zero. In cases like this when there is discrepancy, the default position is to underscore the importance of additional investigations. An issue of importance in device applications is the nature and magnitude of the band alignment at a GaAs/GaAsN heterojunction. While the BAC model implies that the GaAsN conduction band minimum must lie below that in GaAs, it is not clear whether the valence band maximum in GaAsN should exhibit any relative shift. Various predictions disagree on this issue [334,341]. On the experimental side, X-ray photoelectron spectroscopy data suggest a type-II band alignment [373] but with quite large error bars on the valence band offset (VBO). This type of band alignment was supported by PL measurements [374]. Follow-up investigations utilizing not just optical but electrical characterizations as well have concluded a definite type-I alignment [375–379]. A point of discord may lie in the fact that the built-in strain was not completely relaxed in any of the heterostructures covered in the aforementioned investigations. Egorov et al. [377] accounted for strain effects and deduced that the band offset for unstrained GaAs0.98N0.02 with respect to GaAs. The value of the valence band discontinuity between GaAs and GaAsN0.02 at 18 K was evaluated to be 15 5 meV, taking into account the bandgap of GaAsN alloy [372]. On the contrary Egorov et al. [379] reported the band alignment of InxGa1xAs/GaAsN heterojunctions to be type I or type II, depending on the In content x, a point that needs to be confirmed by additional measurements. The BAC model parameters recommended by Vurgaftman and Meyer [152] for GaAsN and all of the other dilute nitrides for which information is available are summarized in Table 2.24. 2.10.2 InAsN
Another binary compounded with N is InAsN, which garnered early theoretical interest [334,380], followed by tight binding calculation focused on the effects of nitrogen clustering in the alloy [381]. Experimental investigations of this dilute nitride have been reported [382–386]. Measurements of the electron effective mass in this alloy indicated a large increase [380,390], analogous to that in GaInAsN [355].
Table 2.24 Band anticrossing (BAC) model parameters for some
of the dilute nitride semiconductors [152]. Parameters
EN w.r.t. VBM (eV)
V (eV)
GaAsN InAsN Ga1xInxAsN GaPN InPN Ga1xInxPN InSbN
1.65 1.44 1.65(1 x) þ 1.44x 0.38x(1 x) 2.18 1.79 2.18(1 x) þ 1.79x 0.65
2.7 2.0 2.7(1 x) þ 2.0x 3.5x(1 x) 3.05 3.0 3.05(1 x) þ 3.0x 3.3x(1 x) 3.0
2.10 Band Parameters for Dilute Nitrides
While the authors [385,386] appear to argue that the BAC model could not account for any increase greater than doubling of the mass in the nitrogen-free host material, upon closer examination of the model it becomes evident that their view is contradicted, and also by the results available for GaInAsN [355], clearly displaying a similarly large increase in mass. Consequently, it is reasonable to assume that the BAC model is applicable to InAsN. At the same time, however, it should be stated that the available information is somewhat incomplete and future investigations may alter this assertion. Extracting the position of the nitrogen level with respect to the valence band maximum in the host InAs from the valence band offsets tabulated in a review article [92], a value for EN ¼ 1.44 eV is recommended. The measurements of Naoi et al. [382] are consistent with values for the potential V ranging between 1.9 and 2.3 eV. A round figure of V ¼ 2.0 eV could be a good default. 2.10.3 InPN
A few experimental studies of InP1xNx are available in the literature [387,388]. Because the bandgaps and valence band offsets of GaAs and InP differ only slightly, it would follow that the BAC model would apply equally well to InPN. Yu et al. [388] derived band structure parameters in which the GaAs/InP VBO was assumed to be 0.35 eV. However, Vurgaftman et al. [92] recommended a value of 0.14 eV, but considering the result EN ¼ 2.0 eV by Yu et al. led them to update their recommended value to EN ¼ 1.79 eV, both being with respect to the valence band maximum of InP. [92] A reexamination of the coupling potential that is most consistent with the data of Yu et al. [388] led Vurgaftman et al. [92] to recommend V ¼ 3.0 eV. 2.10.4 InSbN
Even though InSb has the smallest bandgap for the conventional III–V binaries, addition of N has attracted some attention because of the potential that even longer wavelengths, beyond those accessible by InSb, can be accessed. Murdin et al. [389– 391] made experimental observations that the effective mass in InSb1xNx increases despite a considerable reduction in the bandgap, which is consistent with all the other dilute nitride semiconductors discussed above. The authors derived EN ¼ 0.65 eV. This compares with EN ¼ 0.85 eV above the top of the valence band. At this stage, the experimental values are recommended as default. The same authors also reported V ¼ 2.2 eV and supplemented the minimal BAC dispersion relation of Equation 2.48 with an additional shift of the level of nitrogen position with increasing N fraction [389,390]. It should, however, be pointed out that it appears that their value for V underestimates the observed bandgap decrease [389]. This led Vurgaftman et al. [152] to recommend V ¼ 3.0 eV, which should yield better consistency with the available data, albeit little. Because the effect of nitrogen on the band structure of InSbN is significant in terms of the light-hole dispersion, the full 10-band kp model may be required for any realistic calculations.
j209
j 2 Electronic Band Structure and Polarization Effects
210
2.10.5 GaPN
The early LED [314] developments utilized N-doped GaP because of the unique band structure of this binary. It is an indirect gap semiconductor with both X and L valleys lower in energy than the G point. Nitrogen acts as an isoelectronic impurity in GaP and has been employed as the active material of visible LEDs [314] until the advent of other conventional direct bandgap and III–V and nitride-based varieties. Initial studies of GaP with an alloy-like concentration of N were reported by Baillargeon et al. [392,393]. Miyoshi et al. [394] investigated the transmogrification of the GaPN luminescence spectrum with increasing N content and were able to observe the emission from excitons bound to nitrogen pairs for x < 0.5%. Bi and Tu [395] reported GaP1xNx with x as large as 16% using gas-source molecular beam epitaxy. A number of theoretical reports predicted that GaPN retains its indirect gap nature up to relatively large (beyond dilute) N concentrations [334,339,341]. Shan et al. [396,397] reexamined this view of indirect bandgap within the BAC model, and in the process found out that the anticrossing between G-valley states and the N impurity band moves E below the X valley for arbitrarily small values of x. They derived V ¼ 3.05 eV employing the well-established value of EN ¼ 2.18 eV, relative to the valence band maximum. This places the nitrogen level slightly below the conduction X valley. As in the case of GaAsN whose band structure is less dependent on pressure and temperature than its host material GaAs, GaPN too sports a decrease in the pressure dependence of the fundamental transition with a deformation potential of 1.2 eV [396]. The temperature dependence of the fundamental gap also sees a reduction [398,399]. As in the case of all the dilute nitrides, a large increase in the electron effective mass is observed [400]. Congruous with popular LED materials for decades, strong luminescence for small N fractions [401,402], occurring despite the indirect gap of the GaP host, is another feature of this material. The evolution of transitions due to isolated N centers, N pairs, and N clusters that hasbeenobservedforx < 1%,andthemixingwithX-andL-valleystates,however,cannot be described by the simple BAC model [403–405]. For example, Buyanova et al. observed a sudden reduction in the radiative lifetime of the fundamental transition for x > 0.5%, which they attributed to an effective indirect-to-direct crossover [406]. The wealth of phenomena reported for GaPN can perhaps beexplainedwith a more flexible theoretical approach such as the supercell pseudopotential formulations of Kent and Zunger [324– 326]. Even though the accuracy of the BAC model is more limited in GaPN than in the other dilute nitride alloys due to the proximity of the X valley and the plethora of complex experimental observations for intermediate compositions, the parameter set EN ¼ 2.18 eV and V ¼ 3.05 eV of Shan et al. [398] can be used as default values. 2.10.6 GaInAsN
Let us now extend our treatment of the effect of N to ternary alloys. Having established BAC parameters for GaAs1xNx and InAs1xNx, they need to be determined for the
2.10 Band Parameters for Dilute Nitrides
Ga1yInyAs1xNx alloy whose host material is InGaAs. Owing to applications in long haul communications systems, most of the technological interest among all the dilute nitrides has so far focused on this quaternary. Additional impetus can be found in its applications to solar cells and photovoltaics that can be grown on GaAs substrates rather than InP, the former having a more established and less expensive technology base. Addition of N in InGaN narrows the bandgap for In concentrations, while at the same time providing tensile strain compensation far less than those required in conventional GaInAs quantum wells that are under compressive strain, countering the composition-induced reduction in the bandgap. Moreover, the wavelengths near 1.55 mm are not accessible with standard InGaAs-based quantum wells on GaAs substrates, which necessitates the use of InP substrates, which allow employment of larger concentrations of In. Incorporation of N in InGaAs can mitigate the situation and allow the use of GaAs substrates even for the extended wavelength for long haul fiber-based communications systems. Specific to the discussion of the bandgap, the applicability of the BAC model to the case of GaInAsN is quite well established [320,321,351,352,355,357] in that much of what is reported for the dilute nitrated binaries discussed above has also been found to be applicable to this quaternary. Investigations employing very low In fractions [351,352,355], for example, on the order of 10%, typically found no differences of significance from GaAsN except of course the decrease in the InGaAs energy gap [92]. For materials with larger In fractions, prepared on both GaAs [407,408] and InP [409] substrates, Zhukov et al. [410] proposed an alternative model. Pan et al. [359] took EN to be independent of the In concentration and employed V ¼ 2.5 eV. Although Choulis et al. [320,321,411] utilized the same assumption with respect to EN, their value for the coupling potential was considerably lower: V ¼ 1.675 eV. A similar value (V ¼ 1.7 eV) was independently deduced by Polimeni et al. [408] for In compositions ranging from 25 to 41%. In contrast, Sun et al. [412] found that, depending on the particular transition between the conduction and valence subbands, V in the range of 2.8–3.0 eV was imperative to account for the luminescence data for InGaAsN/GaAs quantum wells having an In content of 27.2%. The finding of a smaller bandgap reduction in GaInAsN than in GaAsN with N fraction is in fact expected and is due to the ordering of the nitrogen atoms in the InGaAs matrix [413,414]. There is some experimental evidence for carrier localization in the presence of both In and N in the quaternary alloy [415,416]. In one investigation [416] a series of five distinct transitions, which were attributed to five different environments for the N atom in the alloy, have been reported. Vurgaftman et al. [92] attempted to provide the best available parameters for the InGaAsN quaternary, while maintaining consistency with the parameters recommended above for GaAsN and InAsN. Accordingly, the position of the nitrogen band EN should be determined from the shift of the valence band offset in GaInAs [92]. This should include the small, yet nonnegligible, bandgap bowing with composition. This leads to a smooth variation of EN between 1.65 eV (for GaAsN) and 1.44 eV (for InAsN) and is consistent with the intuitive expectation, the basis of which is in no contradiction with any definitive experiments in that the position of the nitrogen level should not vary with respect to vacuum. Within this framework, Vurgaftman et al. [92] proposed a
j211
j 2 Electronic Band Structure and Polarization Effects
212
bowing of the coupling potential V in the Ga1yInyAsN alloy: V ¼ 2.7(1 y) þ 2.0y 3.5y(1 y) eV. Admittedly, this parameterization does not fully agree with the experimental results for this quaternary [320,321,351,352,355,359,408]. But, it is nearly consistent with the median values and the recommended binary end points. As in the case of the ternaries, any strain must be added to the host semiconductor properties in the basic BAC model while employing these parameters. 2.10.7 GaInPN
It must be mentioned at the get go that there are not as much data available in this quaternary system. One report of the energy gap in bulk Ga0.46In0.54PN is available in the literature [417]. According to estimates based on the VBO dependence in GaInP, the quaternary alloy under discussion represents a special case in that one would expect a proximity of the nitrogen energy level and the host conduction band edge, perhaps within as little as 10–20 meV [92]. If one presumes, for the sake of discussion, that the BAC model remains valid in this limit, the reported energy gaps allow one to obtain a coupling potential in the 2.1–2.3 eV range. Considering that this is considerably smaller than the recommended values for both InPN and GaPN, by analogy with GaInAsN one could surmise a bowing of the Ga1yInyPN interaction potential: V ¼ 3.05(1 x) þ 3.0x–3.3x(1 x) eV. It is noteworthy that the bowing parameters derived for the two quaternaries are quite similar. It should again be underscored that further studies are needed to confirm and/or update this value. 2.10.8 GaAsSbN
In addition to the mixed cation ternary host materials, mixed anion host materials also present opportunities. Among them is the alloy host GaAsSb that produces the quaternary alloy GaAs1xySbyNx [418–420]. This quaternary has the potential for reaching long wavelengths also on GaAs substrates. Unfortunately, because of the sparse nature of the data for this material, and the total lack of any reports on GaSb1xNx, one can only conjecture that the procedure recommended above for GaInAsN be followed. The material-specific parameters, that is, the constant V ¼ 2.7 eV at least for Sb fractions 20% should be assumed.
2.11 Confined States
If the physical size of the semiconductor in any direction is comparable to the de Broglie wavelength for electrons in electronic processes and exciton diameter in lowtemperature optical processes, the size effects become important. The constriction can form one side, two sides, and three sides in the case of which the system represents three-dimensional, two-dimensional, one-dimensional, and zero-dimensional, as
2.11 Confined States
Figure 2.17 Schematic representation of three-dimensional, twodimensional, one-dimensional, and zero-dimensional systems in real space.
shown in Figure 2.17. The two-dimensional, one-dimensional, and zero-dimensional correspond to quantum wells, wires, and dots, respectively. From the transport point of view, when the size of the well in a quantum well structure is comparable to the de Broglie wavelength in the semiconductor forming the well, the conduction and valence bands are modified noticeably in that the density of states in both the conduction and valence bands are discretized. From the excitonic optical transitions point of view, the length scale is the Bohr radius. This picture is depicted schematically in Figure 2.17a and b for a wurtzite semiconductor bulk (3D system), quantum well (2D system), quantum wire (1D system), and quantum dot (0D system). Moreover, equally important is the modification of these energies in the semiconductor caused by strain effects. In short, eigenstates of strained-layer superlattices require the consideration of both strain and quantum size effects. Eigenstates applicable to the early versions of compound semiconductor based quantum wells without strain were modeled by the envelope function formalism of Bastard [421]. If this method is to be used, one has to calculate the strain-dependent bandgap first. Bastards formalism can then be utilized to determine the transition energies in quantum wells, which must be added to the strain component, as shown by Marzin [422]. Marzin developed a method for calculating the bandgap of a strained cubic semiconductor that, when employed in conjunction with the envelope function approximation, leads to eigenstates in quantized structures of strained systems (Figure 2.18). In Bastards method revised by Marzin, an 8 · 8 Kane Hamiltonian matrix [423] is employed to give an accurate description of a strained quantum structure. The
j213
j 2 Electronic Band Structure and Polarization Effects
214
Figure 2.18 Schematic representation of (a) a GaN/AlGaN multiple quantum well structure and the conduction and valence band edges at the G point, and (b) valence and conduction band diagrams of the same with the confined states indicated.
size of the Hamiltonian is justified because there are three valence bands, in addition to the conduction band, with spin-up and spin-down for each band. Similar to cubic semiconductors, the wurtzite phase also has one conduction band and three valence bands, heavy- and light-hole states as well as spin–orbit bands, each with spin-up and spin-down, necessitating the use of an 8 · 8 Hamiltonian. The strain effects in the GaN system have been treated by Gil et al. [98,424] and others, as reviewed in Ref. [59]. However, until the time when uniformly strained films can be grown, correlations to experiments will remain weak and reduce the level of confidence in the predictions especially and the analysis as a whole. This is particularly true for InGaN wells, as there is also phase separation to deal with, in addition to nonuniform strain. We shall once again emphasize that there exist many discrepancies concerning the determination of some important parameters for the relevant bulk material properties such as hole masses, bandgap energy bowing parameters, shear and deformation potentials, and band offsets that are necessary for determining and understanding the properties of confined structures. Calculations, using ab initio methods, have been applied to estimate the unknown but necessary parameters for the band structure calculations. Later on, Sirenko et al. [62] have performed envelope function calculations of the valence band in wurtzite quantum wells following the formalism of Rashba– Sheka–Pikus (RSP) developed for bulk wurtzite semiconductors. Employing a 6 · 6 Luttinger–Kohn model, Ahn [425] studied the effect of a very strong spin–orbit split-off band coupling on the valence band structure of GaN-based materials. Considering that the spin–orbit band is extremely important for GaN because of its very narrow
2.11 Confined States
spin–orbit splitting (10 meV), the spin–orbit split-off coupling was taken into consideration in the calculations. In addition, it was assumed that the electrons in QW are confined by the conduction band offset (DEc) and the holes by the valence band offset (DEv), the values of which are also of some controversy. When the size of the well is comparable to the Bohr radius of an exciton in the semiconductor forming the well, the exciton transition energies are modified. The Bohr radius is given by a0 ¼
4pes h2 ; m q2
ð2:49Þ
where es is the dielectric constant of the semiconductor. In GaN, due primarily to its large effective mass, this radius is about 28 Å, necessitating very small physical dimensions before noticeable quantization can occur. In an experimental quantum well, the wave function is constricted along the growth direction, which we shall term as the z-direction. In the orthogonal directions, the system is free. The problem is similar to that of a vibrating string with the ends held stationary. The vibration wavelengths are given by ln ¼
2L n
with n ¼ 1; 2; 3;
ð2:50Þ
where L represents the length of the string. In a quantum well, the rapidly varying Bloch waves will be affected by the barriers in the z-direction, and the effect can be lumped into an envelope function that is slowly varying. The wave function can be expressed as [49,426] Cn ðk? ; zÞ ¼
X n
f n ðk? ; zÞu n expðjk? r ? Þ;
ð2:51Þ
where fn(k?, z) and uv represent the envelope and Bloch functions, respectively, and n is the subband index. The summation is performed for spin-up and spin-down of the conduction band where the value of six is assumed for the three valence bands. The wave function expression must be solved for the conduction band and the valence band with the envelope function satisfying Schr€ odingers equation for the particular potential barrier height [59]. For the case where the conduction band is almost s-like, the G7 state suffices. However, for the valence band, due to band mixing, a 6 · 6 Hamiltonian including all the three uppermost valence bands must be used. Moreover, if strain is present, which is the case in almost all nitride-based structures, the Hamiltonian must include the effect of strain as well. To complicate matters further, the piezoelectric effect induces large electric fields at the heterointerfaces, particularly, in samples utilizing InGaN wells. The envelope function must satisfy X h q Hnn0 k? ; þ V n ðzÞdnn þ Hvn0 ðeÞ yv;n0 ðk? ; zÞ j qz ð2:52Þ v0 ¼ E n ðk? Þyv;n0 ðk? ; zÞ;
j215
j 2 Electronic Band Structure and Polarization Effects
216
where m ¼ 1, 2 for the conduction band state and n ¼ 1, . . . , 6 for the valence band state. The details of how the dispersion of the conduction and valence band states can be found in Ref. [59]. Suffice it to say that the conduction band is formed of nearly slike states and can be considered nearly parabolic, ameliorating the confined state calculations, as will be discussed in the next section. In finding the total energy between the confined conduction and valence band states, one may assume that the majority of the contribution is due to the confinement energy in the conduction band because of the large disparity between the electron and hole effective masses in favor of the conduction band. However, when gain in semiconductor lasers is considered, the dispersion of valence band states must be taken into consideration. 2.11.1 Conduction Band
If the potential barrier is infinitely high, the wave vector in the z-direction will be quantized and assumes the discrete values of 2p np ; n ¼ 1; 2; 3; ð2:53Þ ¼ kzn ¼ ln L where L is the thickness of the quantum well. Assuming a parabolic band structure that satisfactorily describes the s-like conduction band, the confinement energy can be expressed as h2 k2zn h2 np2 ¼ n ¼ 1; 2; 3: ð2:54Þ DE conf ¼ 2m 2m L Taking into account the energy dispersion relationship in the x- and y-directions for a parabolic conduction band, we have h2 np2 þ k2x þ k2y n ¼ 1; 2; 3; ð2:55Þ E ¼ Ec þ 2m L with n ¼ 1, 2, 3, and Ec representing the conduction band edge. For a true one-dimensional wire along the x-direction, discretization along the ydirection would also occur in addition to the z-direction, giving rise to confined energy states (with complete confinement) of h2 np2 E ¼ Ec þ þ k2x þ k2y ; n; m ¼ 1; 2; 3 . . . : ð2:56Þ 2m L If we consider a semiconductor whose constant energy surface for conduction band in k-space is a sphere, such as the case in GaN, the volume of that sphere and the number of available states in k-space are proportional to k3 in terms of momentum and E3/2 in terms of energy, as shown in Figure 2.19a. The density of states per unit energy associated with that system is proportional to E1/2, again as shown in Figure 2.19a. The area and the number of available states in k-space in an ideal system confined in one direction only (representing quantum wells), which is often the z or the growth direction, is proportional to k2 or E, as shown in Figure 2.19b. The
2.11 Confined States
density of states in this case is given by m =ph2 and forms a staircase as shown in Figure 2.19b. If we continue and place a confinement in the x-direction in addition to the z-direction, which represents quantum wires, the line length and the number of available states in k-space is proportional to k in terms of momentum and E1/2 in terms of energy as shown in Figure 2.19c. The corresponding density of states takes the dependence of E1/2, again as shown in Figure 2.19c. If confinement is imposed in all three directions, which represent the pseudoatomic or quantum dot state, the energy is discretized in all directions and the resultant density of states takes a deltalike function in energy, as shown in Figure 2.19d. The case for quantum dots where
(a)
kz
dN/dE ~E 1/2 3
N ~ k (N-E3/2)
ky
E
kx
(b) kz dN/dE N ~ k 2(N-E)
~Constant
ky E kx Figure 2.19 Constant energy surfaces and density of states for 3D in (a) , 2D in (b) , 1D in (c) , and 0D in (d) systems, respectively. The constant energy surface are represented by a sphere, circle, a line, and a point in 3D, 2D, 1D, and 0D systems, respectively, in semiconductor such as GaN conduction band that has a spherical constant energy surface in 3D.
j217
j 2 Electronic Band Structure and Polarization Effects
218
(c)
kz Bulk dN/dE ~ E–1/2 N ~ k(N-E 1/2 )
n=4
n=3 n=2 n=1
ky
E
kx
(d)
dN/dE ~ δ (E)
kz
Bulk n=4
n=3 n=2 n=1
ky E
kx Figure 2.19 (Continued )
physical dimension in all directions are scaled to the size of the wave function or smaller is treated in Volume 2, Chapter 5. Considering quantum wells formed in growth along the z-direction, the confinement occurs in the same direction and thus the wave vector is quantized. The component of the wave vector experiencing confinement is typically referred to as the z-component, out-of-plane component, or simply kz. In-plane components of the wave vector kx and ky, however, are not quantized and the usual energy momentum dispersion (E–k) diagram would apply. A pictorial description of the E–k diagram in a two-dimensional system with confinement along the z-direction for two quantum energy levels is shown in Figure 2.20. If the barrier potential is large but not infinite, the wave function outside the well decays exponentially, which is called the evanescent wave. No analytical solution exits for the subband energies when the potential barrier is not infinitely high. Graphical solutions treated in many textbooks on quantum mechanics and numerical solutions as well do exist. The problem is made even more complicated in semiconductors in
2.11 Confined States
E
E
E2 E1 π/Lz 2π /L z
kx ,k y
kz
kx Figure 2.20 Energy momentum dispersion relation in a twodimensional system with confinement along the z-direction. Shown on the left is the three-dimensional view while that on the right represent two slices depicting the E–kin-plane relationships for quantum levels.
that not only is the barrier not infinite but also the barrier and well materials do not have the same carrier mass. In this case, the boundary condition must be changed from the continuity of the derivative of the wave function in the z-direction to the continuity of the particle flux in the z-direction, that is, 1 qyB 1 qyW ¼ ; ð2:57Þ mB qz mW qz at Z L/2, assuming the origin of the z-axis to be in the middle of the well. The terms mB and mW represent the effective masses in the barrier and well materials, respectively. Likewise, fnB and fnW depict the envelope wave functions [427] in the barrier and well materials, respectively. The solution for the subband energies can be computed numerically. The conduction band minimum for GaN as well as AlN is at the zone center and twofold degenerate. The confinement energies for the GaN/AlGaN quantum wells can reasonably be estimated by means of the envelope function approximation in the same manner as that extensively used for the GaAs/AlGaAs material system [427,428]. Following the Weisbuch and Vinter notation, the low-lying conduction electron state can be represented by [427,428] X CðrÞ ¼ ½expðik? ðrÞucj ðrÞf n ðzÞ; ð2:58Þ j¼W;B j
where W and B represent the well and barrier materials, uc ðrÞ is the conduction band zone center Bloch wave function of GaN or AlGaN, fn(z) is a slowly varying envelope function, k? is the transverse (in-plane) wave vector, is the envelop wave function j [427], and the growth direction is along the z-axis. Because uc ðrÞ is the same for GaN and Al(Ga)N, Schr€odingers equation reduces to h2 q2 þ VðzÞ f n ðzÞ ¼ E n f n ðzÞ: ð2:59Þ 2m0 qz2
j219
j 2 Electronic Band Structure and Polarization Effects
220
In the above equation, m*(z) is the corresponding effective mass of the conduction electron, V(z) represents the profile of the minimum of the conduction band along the growth direction, and E(z) is the confinement energy. Assuming no doping in either regions, V(z) has a rectangular well-like profile. The solution of the confinement energies is similar to a particle in a box problem. The boundary conditions are 1 df n ðzÞ that fn(z) and mðzÞ dz be continuous across the interface. The latter is necessary to ensure the conservation of the particle current.
Al x Ga1–x N/GaN, x = 0.2, L B = L W = 2 nm
0.25
Energy (eV)
0.20 0.15 0.10 0.05 0.00
–2
–1
0
1
2
3
4
Distance (nm) (a) Figure 2.21 (a) The conduction band edge potential profile for a representative Al0.2Ga0.8N/ GaN single quantum well with a barrier and well thickness of 2 nm (or 20 Å) each for a Ga polar sample with the [0 0 0 1] direction pointing to the left. Polarization charge causes deviation from the square well and no screening due to free carriers is accounted for. The scales for the vertical and horizontal axes are in terms of eV and nm, respectively. Courtesy of V. Litvinov. (b) The conduction band edge potential profile for a representative Al0.3Ga0.7N/GaN single quantum well with a barrier and well thickness of 2 and 4 nm, respectively for a Ga-polar sample with the [0 0 0 1] direction pointing to the right.
Polarization charge causes deviation from the square well and no screening due to free carriers is accounted for r. The scales for the vertical and horizontal axes are in terms of eV and nm, respectively. Courtesy of V. Litvinov. (c) The conduction band edge potential profile for a representative In0.3Ga0.7N/GaN single quantum well with a barrier and well thickness of 2 and 4 nm, respectively, for a Ga-polar sample with the [0 0 0 1] direction pointing to the right. Polarization charge causes deviation from the square well and no screening due to free carriers is accounted for. The scales for the vertical and horizontal axes are in terms of eV and nm, respectively. Courtesy of V. Litvinov.
2.11 Confined States
Al x Ga1-x N/GaN, x = 0.3, LB = 4 nm, L W = 2 nm
0.5
Energy (eV)
0.4 0.3 0.2 0.1 0.0 –4
–2
0
2
4
Distance (nm)
(b)
In x Ga
1–x
N/GaN, x = 0.3, LB = 4 nm, LW = 2 nm
1.4 1.2
Energy (eV)
1.0 0.8 0.6 0.4 0.2 0.0 –4 (c)
–2
0
2
4
Distance (nm)
Figure 2.21 (Continued )
The case of quantum wells in the nitride system requires not just confinement effects due to barriers but also polarization-induced charge. The latter tends to distort the band profile due to the induced electric field. The details of the polarization charge and its effect on the quantum wells can be found in Section 2.12.5 and Volume 2,
j221
j 2 Electronic Band Structure and Polarization Effects
222
Chapter 5. Essentially, the conduction and valence band edges get skewed due to the field induced by polarization, as shown in Figure 2.21a and b for representative Al0.2Ga0.8N/GaN (Ga-polar) and Al0.3Ga0.7N/GaN (Ga-polar) single quantum well structures, the former with barrier and well thicknesses of 2 nm (or 20 Å) each, with [0 0 0 1] direction pointing to the left, and the latter with barrier and well thickness of 2 and 4 nm and [0 0 0 1] direction pointing to the right. The same for a representative In0.3Ga0.7N/GaN single quantum well with a barrier and well thickness of 2 and 4 nm, respectively, for a Ga-polar sample with the [0 0 0 1] direction pointing to the right is shown in Figure 2.21c. The band profile depicts the case with no screening. Free carrier induced screening would alter the profile. The calculated eigenstates, only the ground levels exist, for Al0.1Ga0.9N/GaN and Al0.2Ga0.8N/GaN as a function of quantum well thickness for a barrier thickness of 2 nm are shown in Figure 2.22. Also shown in dashed lines are the energy levels without polarization such as the case on a-plane (nonpolar surface) quantum wells discussed in Volume 3, Chapter 1. The calculated eigenstates for Al0.3Ga0.7N/GaN ground and the first excited states (only two states are available) for a barrier thickness of 4 nm as a function of quantum well thickness are shown in Figure 2.23. For increased activity in deeper UV devices, both
0.30
x = 0.1, level 1 x = 0.2, level 1 x = 0.2, level 2 LB = 2 nm
Alx Ga1–x N/GaN Dashed lines: without polarization
0.25
Energy (eV)
0.20
0.15
0.10
0.05
0.00 1
2
3
4
5
LW (nm) Figure 2.22 Calculated eigenstates, only the ground levels exist, for Al0.1Ga0.9N/GaN (lower curve) and Al0.2Ga0.8N/GaN (upper curve) for a barrier width of 2 nm as a function of quantum well thickness. Also shown in dashed lines are the energy levels without polarization such as the case on a-plane (nonpolar surface) quantum wells. Courtesy of V. Litvinov.
6
7
2.11 Confined States
in terms of emitters and detectors, it is necessary to consider structures where AlxGa1xN is used as the active part of the device in the form of a quantum well. Choosing AlN, arbitrarily, as the barrier for AlxGa1xN, the eigenstates for two representative cases are presented. Also shown in dashed lines are the energy levels without polarization such as the case on a-plane (nonpolar surface) quantum wells discussed in Volume 3, Chapter 1. Shown in Figure 2.24 are the ground and first excited states as a function of well width for an AlN/Al0.4Ga0.6N quantum having a barrier width of 2 nm. It should be noted that only the ground and first excited states are available. Also shown in dashed lines are the energy levels without polarization such as the case on a-plane (nonpolar surface) quantum wells discussed in Volume 3, Chapter 1. The same structure but with AlN/Al0.5Ga0.5N, representing nearly the solar blind region of the Suns spectrum, is shown in Figure 2.25. Also shown in dashed lines are the energy levels without polarization, such as the case of a-plane (nonpolar surface) quantum wells discussed in Volume 3, Chapter 1. For the visible or nearly visible part of the spectrum, InGaN quantum wells typically with GaN barriers are employed. This is in part due to not only benefits gained by heterostructure devices but also technological reasons, the latter due to the decomposition of thick InGaN layers and inordinate amounts of ammonia required
0.45 0.40 0.35
AlxGa1–x N/GaN
level 1 level 2 L B = 4 nm, x = 0.3 Dashed lines: without polarization
Energy (eV)
0.30 0.25 0.20 0.15 0.10 0.05 0.00 1
2
3
4 LW ( nm )
Figure 2.23 Calculated eigenstates for Al0.3Ga0.7N/GaN for ground and the first excited states (only two states are available) for a barrier thickness of 4 nm as a function of quantum well thickness. Also shown in dashed lines are the energy levels without polarization such as the case on a-plane (nonpolar surface) quantum wells. Courtesy of V. Litvinov.
5
6
j223
j 2 Electronic Band Structure and Polarization Effects
224
1.0
level 1 level 2 x = 0.4 L B = 2 nm
AlN/Al x Ga 1–x N Dashed lines: without polarization
Energy (eV)
0.8
0.6
0.4
0.2
0.0 1.0
1.5
2.0
2.5
3.0
3.5
4.0
L W (nm) Figure 2.24 Calculated ground and first excited states (only two states are available) as a function of well width for an AlN/ Al0.4Ga0.6N quantum having a barrier width of 2 nm. Also shown in dashed lines are the energy levels without polarization such as the case on a-plane (nonpolar surface) quantum wells. Courtesy of V. Litvinov.
to grow thick layers. In this vein, the calculated ground state (for x ¼ 0.1) and ground and excited states (for x ¼ 0.2) as a function of well width for an InxGa1xN/GaN quantum well having a barrier thickness of 2 nm are presented in Figure 2.26. 2.11.2 Valence Band
In an attempt to determine the valence band subband structure, Suzuki and Uenoyama [59] calculated the band discontinuities from first-principles calculations and found them to be 0.11 and 0.43 eV for the valence and conduction bands of GaN/ Al0.2Ga0.8N. The elastic stiffness constants, taken from prior experimental data, employed for GaN were, in units of 1011 dyn cm2, 29.6, 13.0, 15.8, 26.7, and 2.41 for C11, C12, C13, C33, and C44, respectively (see Table 1.24). For illustrative purposes, the valence band structure in unstrained 30 and 50 Å GaN/Al0.2Ga0.8N quantum wells is exhibited in Figures 2.27a and 2.28 where the strain due to the lattice and thermal mismatch are neglected. Bandgap discontinuities of 0.11 and 0.43 eV were adopted for the valence and conduction bands, respectively. The confinement energies in deep wells are inversely proportional to the effective mass in the growth direction and
2.11 Confined States
1.4 AlN/Al x Ga1–xN 1.2
Dashed lines: without polarization
level 1 level 2 x = 0.5 LB = 2 nm
Energy (eV)
1.0
0.8
0.6
0.4
0.2
0.0 1.0
1.5
2.0
2.5
3.0
3.5
4.0
LW (nm) Figure 2.25 Calculated ground and first excited states (only two states are available) as a function of well width for an AlN/ Al0.5Ga0.5N quantum well having a barrier width of 2 nm, representing nearly the solar blind region of the Suns spectrum. Also shown in dashed lines are the energy levels without polarization such as the case on a-plane (nonpolar surface) quantum wells. Courtesy of V. Litvinov.
directly proportional to the square of the well length. The HH and LH bands are not coupled. Consequently, the HH band can be construed as parabolic with little, if any, change in strain. This is because the C6v crystal symmetry of the bulk remains. The upper pffiffiffi (LH) and lower (CH) bands are coupled with a constant coupling coefficient 2D3 , which means that coupling of these bands is independent of kz and strain. The effective masses of the HH and LH bands are too heavy to cause substantial confinement energy, whereas the CH band with a lighter mass causes more split due to quantization and, in a sense, makes the crystal splitting (Dcr) larger. For comparison, in ZB structures the coupling between the light hole and the spin–orbit band is dependent on kZ and thus the bands change considerably with strain, and the in-plane heavy-hole mass becomes light and the in-plane light-hole mass becomes heavy. The effect of strain on the quantum well was also considered by Suzuki and Uenoyama [60], assuming coherently strained quantum wells. Consequently, the inplane lattice constant of the layers is assumed to be that of the substrate with a substantial effect on the subband structure of the quantum wells. The beneficial effects of strain in electronic and optoelectronic devices based on ZB crystals have
j225
j 2 Electronic Band Structure and Polarization Effects
226
0.8 0.7
Dashed lines: without polarization
In x Ga1–x N/GaN
x = 0.1 x = 0.2, first x = 0.2, second L B = 2 nm
Energy (eV)
0.6 0.5 0.4 0.3 0.2 0.1 0.0 1
2
3
4
5
6
LW (nm) Figure 2.26 Calculated ground state (for x ¼ 0.1) and ground and excited states (for x ¼ 0.2) as a function of well width for an InxGa1xN/GaN quantum wells having a barrier width of 2 nm. For x ¼ 0.2, the ground and the first excited states are available. Also shown in dashed lines are the energy levels without polarization, such as the case on a-plane (nonpolar surface) quantum wells. Courtesy of V. Litvinov.
been documented well [88]. This and the fact that there is some lattice mismatch between GaN and its ternaries make it imperative that the effect of strain on the properties of Wz quantum wells be considered. Figures 2.27b and 2.28b exhibit the valence band subband structure for 30 and 50 Å GaN/Al0.2Ga0.8N quantum wells. The biaxial strain was assumed to be 0.5% in the (0 0 0 1) c-plane. Superimposed is the valence band structure for 30 and 50 Å Wz GaN/Al0.2Ga0.8N quantum wells with 0.5% tensile strain in the c-plane. Results indicate that the wells get deeper for compressive strain and shallower for tensile strain. The density of states at the valence band maximum gets smaller for the compressive and larger for tensile strain. However, the change is very small as the symmetry remains as in the bulk with no further removal of the degeneracy. It must be mentioned though that uniaxial strain along the x- or the y-direction only causes the HH band to move to a higher energy and leads to a reduced density of states. Consequently, the effects of this type of strain resemble those in bulk GaN. In a sense, the Q well alone does not result in any special characteristic that would lead to much improved results for lasers [429]. Valence band confinement energies for relaxed and 0.5% compressively strained GaN/Al0.2Ga0.8N quantum wells in the c-plane are depicted in Figure 2.29. The data
2.11 Confined States
50
L = 30 Å
100
No strain
50
0
–50
0 Energy (meV)
Energy (meV)
HH1 LH1 LH2 –100 HH2 –150
–200 0
–50
L = 30 Å
j227
With 0.5 % compressive strain HH1
LH1
LH2 HH2
–100 –150
20 10 15 5 kx ,ky ,Wave number (× 106 cm –1)
–200 0
(a)
20 10 15 5 kx ,ky ,Wave number (× 10 6 cm –1) (b)
Figure 2.27 Plot of the upper valence band structure (HH and LH bands) in a 30 Å GaN/Al0.2Ga0.8N quantum well with bandgap discontinuities of 0.11 and 0.43 eV for the valence and conduction bands, respectively; (a) without strain and (b) with 0.5 compressive strain in the c-plane [60].
have been deduced from the calculations of Suzuki and Uenoyama [59] for 30, 40, 50, and 60 Å well thicknesses and band discontinuities of 0.11 and 0.43 eV for the valence and conduction bands, respectively. The other parameters utilized can be found in the tables presented in Chapter 1. 2.11.3 Exciton Binding Energy in Quantum Wells
Exciton binding energies in reduced dimensional systems vary from those of the bulk [430]. In the GaN/AlGaN system, one would expect the binding energy to go up, as the confinement gets stronger. If the well thickness is continually reduced, at some point the overlap with AlGaN becomes very noticeable. Toward zero well thickness, the binding energy should approach the value of AlGaN. Considering the strong localization and possible role of excitons in optical processes even at room temperature, it is imperative that the binding energy be known. Bigenwald et al. [431] considered the very problem of exciton binding energy in a GaN/Al0.2Ga0.8N system with results leading to the obvious conclusion that the effect of confinement is large and cannot be ignored. They calculated the exciton binding energies and oscillator strengths with the formalism developed [432] prior to the aforementioned investigation by applying a two-parameter trial function. Due to the anisotropy of the structure, the dielectric constant was globalized and the particle masses were weighted with the
j 2 Electronic Band Structure and Polarization Effects
228
50
100
L = 50 Å
Strain free
L = 50 Å HH1
50
0
0.5 % Compressive strain LH1
E n e r g y (m e V )
HH1 –50 HH2 –100 LH1
0
LH2
HH2
–50 –100
HH3
LH2
–150
–150 –200 0
–200 0
5
10
15
20
5
10
15
20
k x ,k y ,Wave number (× 106 cm–1)
kx,ky,Wave number (× 106 cm–1) (a)
(b)
Figure 2.28 Plot of the upper valence band structure (HH and LH bands) in a 50 Å GaN/Al0.2Ga0.8N quantum well with bandgap discontinuities of 0.11 and 0.43 eV for the valence and conduction bands, respectively; (a) without strain and (b) with a compressive strain of 0.5 in the c-plane [60].
0
HH LH CH
100
CH 1
Strained b ul k GaN
50
LH 2 Bulk Al 0.2Ga 0.8N
C onfin em ent en ergy (m eV)
LH 1
HH 2 CH 2
( 9 )1 HH 1 ( 9 )2 HH 2
HH 1
150 HH LH CH
200 0
50 Well width (Å)
Figure 2.29 Valence band confinement energies versus the thickness of the well in GaN/Al0.2Ga0.8N quantum wells. Courtesy of Professor Bernard Gil.
100
(
) 71 1
LH 1
(
71 )2
LH 2
( (
) CH 1 72 1 ) 72 2 CH 2
2.11 Confined States
j229
40
Exciton binding energy (meV)
35 B Exciton A Exciton 30
25
C Exciton 20
15 0
20
40
60
80
GaN Q Well thickness (Å) Figure 2.30 A, B, and C exciton binding energies as a function of the well width in a GaN/ Al0.2Ga0.8N system. Note that the exciton binding energies reported for GaN range from about 20 to 30 meV. The likely value of A exciton
binding energy in GaN is 21 meV. If so, the absolute values of the binding energies shown should be treated with caution. However, the trend with quantum well thickness holds, which is the reason for inclusion of this figure [431].
probability densities. The A, B, and C exciton binding energies computed as a function of GaN well thickness are shown in Figure 2.30. Two essential points stand out. The first point is that the G9v and G17v states are confined in the well as is the electron state. The largest binding energy corresponds to a well thickness of L 15 Å, which is when the spreading of both electron and hole functions in the barrier area minimal. The second point is that the relatively light G27v hole state leaks out of the well (for L < 100 Å) so that the electron–hole pair has a small binding energy and is almost constant for 20 < L < 100 Å. Caution should be exercised in applying the calculations of Ref. [432] for well thicknesses larger than about 100 Å, three times the Bohr radius. In GaN-based systems, the terms quantum well and superlattices have been used very liberally in that structures with well thicknesses well in access of the Bohr radius are referred to as quantum wells. The term superlattice requires that barriers are penetrable by the wave function, and further, the wave functions in adjoining wells overlap and form the superlattice minibands. If these standards are strictly applied at this point in time, there may not be much to discuss. Consequently, a conscious decision was made to treat many heterostructures with reduced dimensions, in at least one direction, as quantum-confined structures.
100
j 2 Electronic Band Structure and Polarization Effects
230
2.12 Polarization Effects
Solids are different from vacuum in that they respond to electric fields present or applied. There are three forms of polarization that are present at the atomic level [2]. One is due to partial or complete alignment of dipole moments of polar molecules with the electric field. When atoms forming the solid are different, as in binary, ternary, and quaternary semiconductors, and they have different electronegativities, any asymmetrical molecule has a permanent dipole moment, a process referred to as dipole orientation or paraelectric response. This component cannot keep pace with varying electric field above about 1010 Hz, causing a drop in the real part of the dielectric constant and a jump in the imaginary part (the loss part – appreciable loss factor). The loss is caused by dipoles attempting to respond to the field but seriously lagging in phase. At higher frequencies, the dipoles cannot follow the field and the effect is negligible including the loss factor. Electric field paves the way for rotation of dipole to participate in the electric displacement and align the dipoles collectively, barring thermal disturbances. In a completely or partially ionic solid, dipoles can be induced by relative motion of positive and negative ions under the influence of electric field, causing what is termed as the ionic polarization. Similar to the case of dipole orientation, this process cannot respond to frequencies above 1013 Hz (Reststrahlen frequency), causing yet an additional drop in the dielectric constant and a surge in the loss factor due to the phase lag. Again, as the frequency is increased further, above the Reststrahlen frequency, ionic motion cannot respond to the field and the loss factor due to this process diminishes. The third kind, which occurs in every dielectric, is called the electronic polarization. This is caused by displacement of electrons in an atom relative to the nucleus under the influence of electric field, in a sense deforming the electron shells. The electronic polarization remains at frequencies above Reststrahlen frequency, making the real part of the dielectric constant larger than unity. This process too cannot follow the field above 1015 Hz, above which the dielectric constant of the solid becomes very close to unity. Group III–V nitride semiconductors exhibit highly pronounced polarization effects. Semiconductor nitrides lack center of inversion symmetry and exhibit piezoelectric effects [209] when strained along h0 0 0 1i. Piezoelectric coefficients in nitrides are almost an order of magnitude larger than in many of the traditional group III–V semiconductors [209,433–437]. The strain-induced piezoelectric and spontaneous polarization charges have profound effects on device structures. The piezoelectric effect has two components. One is due to lattice mismatch (misfit) strain while the other is due to thermal strain (ts) caused by the thermal expansion coefficient difference between the substrate and the epitaxial layers. The low symmetry in nitrides, specifically, the lack of center of inversion symmetry present in zinc blende structure, may be interpreted as some sort of nonideality, which is not the case. Nonvanishing spontaneous polarization is allowed in an ideal wurtzite structure [212,438]. This spontaneous polarization is noteworthy, particularly when heterointerfaces between two nitride semiconductors with varying electronegativity are involved. This manifests itself as a polarization charge at heterointerfaces.
2.12 Polarization Effects
Spontaneous polarization was only understood fully not too long ago by King-Smith and Vanderbilt [439] and Resta et al. [440]. In heterojunction devices such as modulation-doped field effect transistors (MODFETs) where strain and heterointerfaces are present, the polarization charge is present and is inextricably connected to free carriers, which are indeed present. As such, polarization charge affects device operation in all nitride-based devices, particularly HFETs, and thus must be taken into consideration in device design unless nonpolar surfaces such as the a-plane are used. The quality of films on nonpolar surface has not kept pace with those on polar basal plane, which make the topic of discussion quite relevant. As mentioned above, polarization charge arises from two sources: piezoelectric effects and the difference in spontaneous polarization between AlGaN and GaN, even in the absence of strain. These charges exit in all compound semiconductors to varying degrees unless self-cancelled by the symmetry of the particular orientation under consideration such as the nonpolar surfaces/interfaces. In relative terms, spontaneous polarization is larger than the piezoelectric polarization in AlGaN/GaN-based structures. In the case of InGaN/GaN structures, spontaneous polarization is relatively small but not as small as the earlier predictions called for, but still noteworthy, as spontaneous polarizations in GaN and InN are not as different from one another. However, the strain-induced piezoelectric polarization can be sizable. If and when defect-associated relaxation occurs reducing the strain in the films, the strength of the piezoelectric polarization is lowered. Spontaneous polarization and piezoelectric polarization affect the band diagram of heterostructures. The effects are very large and can easily obscure the engineered designs. Polarization is dependent on the polarity of the crystal, namely, whether the bonds along the c-direction are from cation sites to anion sites or visa versa. The convention is that the [0 0 0 1] axis points from the face of the N-plane to the Ga-plane and marks the positive z-direction. In other words, when the bonds along the c-direction (single bonds) are from cation (Ga) to anion (N) atoms, the polarity is said to be the Ga-polarity, and the direction of the bonds from Ga to N along the c-direction marks the [0 0 0 1] direction, which is generally taken to be the þz-direction. By a similar argument, when the bonds along the c-direction (single bonds) are from anion (N) to cation (Ga) atoms, the polarity is said to be the N-polarity, and the direction of the bonds from N to Ga along the c-direction marks the direction, which is generally taken to be the z-direction. To shed further light, the Ga-polarity means that if one were to cut the perfect solid along the c-plane where one breaks only a single bond, one would end up with a Ga-terminated surface. The Ga- and N-polarity of a model GaN crystal is shown in Figure 1.3. A schematic representation of the spontaneous polarization in a model GaN/AlN/GaN wurtzitic crystal is shown in Figure 2.31. The spontaneous polarization Pspont (also commonly referred to as P0) in a solid has not always been well defined, although much better understanding of it has been emerging. Only those differences in P between two phases that can be linked by an adiabatic transformation that maintains the insulating nature of the system throughout are well defined. For example, one phase can be considered unstrained and the other strained. Vanderbilt proved that the polarization difference DP between the
j231
j 2 Electronic Band Structure and Polarization Effects
232
[0 0 0 1] Axis
P0AlN
P0GaN N
N
Ga
Ga
Al
Al
N
N
N
Al
Al
N
N
Ga
Ga
N
P0
P0 GaN
P0GaN
P0AlN
AlN
Figure 2.31 Schematic depicting the convention used for determining the polarity and crystalline direction in wurtzitic nitride films. The diagram shows the case for a Ga-polarity film with its characteristic bonds parallel to the c-axis (horizontal in the figure) going from the cation (Ga or Al) to the anion (N). The spontaneous polarization components P0Ga and P0AlN for a periodic GaN/AlN structure are also indicated with that for AlN having a larger magnitude. The
GaN
spontaneous polarization is negative and thus points in the [0 0 0 1] direction. Caution must be exercised here as there is no long-range polarization field, just that it is limited to the interface. The polarization in AlN is larger in magnitude than in GaN. There exists a difference in polarization at the interface, DP0 pointing in direction for both GaN/AlN the [0 0 0 1] interfaces. The Born factor is defined in Equation 2.82 [433].
wurtzite and zinc blende phases could be calculated by considering an interface between the two phases and by defining Pspont to be zero in the zinc blende phase. In short, by calculating the integral of a quantum mechanical Berry phase along a line in the Brillouin zone from one end to the other in the bulk wurtzite symmetry leads to polarization P with respect to that in zinc blende (which is zero by definition because zinc blende is cubic and cannot have a spontaneous polarization in an infinite bulk periodic crystal). The Berry phase actually represents an overlap integral between the periodic part of the Bloch function at k and a neighboring k-point, k0 . Zorroddu et al. [212] and Bernardini et al. [433] showed that the charges accumulating at each interface in a self-consistent calculation can be obtained from the DP of the two bulk layers forming the heterointerface. The relation between the charge and P follows basically from Gausss law. The bound charge density rb ¼ ÏP. This means that across an abrupt interface with P1 on one side and P2 on the other side, one gets P P2 P 1 ¼ i rs (surface charge density at the interface with the appropriate signs). Even though it is overly simplistic, a graphical picture of polarization due to strain (piezo component) and heterointerfaces (spontaneous component), the latter is in the case of different ionicity, can be obtained, which is helpful. Shown in Figure 2.32 is a ball-and-stick diagram of a tetrahedral bond between Ga and N in Ga- and N-polarity configurations, showing the polarization vector due to the electron cloud being closer to the N atoms. Actually, the cumulative polarization due to the triply bonded atoms is along the direction of the single bond. The in-plane and vertical components of polarization due to pairs of atoms cancel one another if the tetrahedron is ideal.
2.12 Polarization Effects
Ga Polarity
[0 0 0 1]
N Polarity
N
[0 0 0 1]
P0 P0 P0
P0zr
P0
P0
Ga
Ga P0
P0zr
P0
P0
N
Ga
P0
N
j233
Ga
P0
N
Ga
Figure 2.32 Ball-and-stick configuration of an ideal GaN tetrahedron with proper c/a ratio and internal parameter u for both Ga and N polarity in a relaxed state.
However, when a Ga-polarity film is under homogeneous in-plane tensile strain, the cumulative z-component, [0 0 0 1] direction, of the polarization associated with the triple bonds decreases, causing a net polarization that would be along the ½0 0 0 1 direction, as shown Figure 2.33. In a nitrogen-polarity film, the same occurs except that the net polarization would be in the opposite, [0 0 0 1], direction. When an inplane and homogeneous compressive strain is present, the net polarization would be in the [0 0 0 1] direction in the Ga-polarity case and ½0 0 0 1 direction in the N-polarity case, as shown in Figure 2.34. Ga Polarity
[0 0 0 1]
N Polarity
N
P0
P0 P0 P0
P0zr
P0
P0 Ga
N
[0 0 0 1]
F
P0
P0
N
N
Ga
Ga P0 Ga
Figure 2.33 Ball-and-stick configuration of a GaN tetrahedron for both Ga and N polarity with a homogeneous in-plane tensile strain showing a net polarization in the [0 0 0 1] direction for Ga-polarity and [0 0 0 1] polarization for N polarity.
P0
P0
P0zr
P0
Ga
F
j 2 Electronic Band Structure and Polarization Effects
234
Ga Polarity
[0 0 0 1]
N Polarity
N
[0 0 0 1]
Ga
P0 P0 P0
P0zr
P0
Ga P0
N N
P0
P0 P0zr
P0
P0
N
Ga Ga
P0
Ga Figure 2.34 Ball-and-stick configuration of a GaN tetrahedron for both Ga and N polarity with a homogeneous in-plane compressive direction for Ga strain showing a net polarization in the [0 0 0 1] polarity and [0 0 0 1] polarization for N polarity.
The same graphical argument can be used to attain a mental image of spontaneous polarization at heterointerfaces as well. For example, if we were to construct two tetrahedra, one representing a GaN bilayer and another on top of it representing an AlN bilayer, the top N atom shown in Figure 2.32 for the Ga-polarity configuration would make triple bonds with it. Because AlN is more electronegative than GaN, the net component of the polarization vector in the [0 0 0 1] direction in triply bonded N with Al is larger in amplitude than in the GaN tetrahedron, and there would be a net interfacial polarization in the ½0 0 0 1 direction even without strain. In short, the source of piezoelectric polarization is strain in an electronegative binary. That for spontaneous polarization is the change in electronegativity across an interface such as the AlN and GaN interface. Substrates upon which nitride films are grown lack the wurtzitic symmetry of nitrides. Consequently, the polarity of the films may not be uniform, as schematically depicted in Figure 4.21 where the section on the left is of Ga-polarity and the section on the right is of N-polarity, representing a Holt-type inversion domain. In this type of inversion domain, the wrong type of bonds, for example, GaGa and NN, are formed at the boundary projected on (1 1 2 0) plane. The structural and electrical details of the inversion domains observed and investigated in GaN can be found in Section 4.1.3. Inversion domains combined with any strain in nitride-based films lead to flipping piezoelectric fields with untold adverse effects on the characterization of nitride films in general and the polarization effect in particular, and on the exploitation of nitride semiconductors for devices. Such flipping fields would also cause much increased scattering of carriers, as they traverse in the c-plane. Having made the case, it should be mentioned that if proper measures are taken, the Gapolarity films grown by OMVPE and to a lesser extent by MBE are nearly or completely inversion domain boundary free even on sapphire substrates. However,
2.12 Polarization Effects
in the case of MBE, unless optimum AlN buffer layers are employed or N-polarity films are grown by incorporation of GaN initiation layers, inversion domain boundaries do occur, sometimes in high concentrations, see Section 3.5.6. The magnitude of the polarization charge converted into number of electrons can be in the mid-1013 cm2 level for AlN/GaN heterointerfaces, which is huge by any standard. For comparison, the interface charge in the GaAs/AlGaAs system is used for MODFETs in less than 10% of this figure. An excellent review of the polarization effects can be found in Ref. [440]. The magnitude of the polarization charge is tabulated in Table 2.25 along with elastic coefficients taken from a series of publications by Bernardini, Fiorentini, and Vanderbilt. The data in bold are those reported in an earlier publication [433] and the remaining data points are taken from a later publication [212]. Following the initial reports of piezoelectric and spontaneous polarization [441], the authors returned to the topic [212] as the values of the initial parameters were not consistent with other reports [181]. Bernardini et al. [442] reanalyzed the polarization as obtained using the Berry phase method within two different DFT exchange correlation schemes. Specifically, the authors used the Vienna ab initio simulation package (VASP) and the pseudopotentials provided therewith, as in Ref. [180]. The newer calculations were carried out using both the generalized gradient corrected local density approximation (GGA) to density Table 2.25 Elastic constants and spontaneous polarization charge
in nitride semiconductors.
2
e33 (C m ) LDA GGA e31 (C m2) LDA GGA p e31 LDA GGA C33 (GPa) GGA C31 (GPa) GGA P0 (C m2) LDA GGA P0 (C m2), ideal wurtzite structure R ¼ C31/C33 LDA GGA e31 (C13/C33)e33 (C m2)
AlN
GaN
InN
1.46 1.8 1.5 0.60 0.64 0.53 0.74 0.62 377 94 0.081 0.10 0.090 0.032 0.578 0.499 0.86
0.73 0.86 0.67 0.49 0.44 0.34 0.47 0.37 354 68 0.029 0.032 0.034 0.018 0.40 0.384 0.68
0.97 1.09 0.81 0.57 0.52 0.41 0.56 0.45 205 70 0.032 0.041 0.042 0.017 0.755 0.783 0.90
The data in bold are associated with DFT in the generalized gradient approximation (GGA) that are more accurate than others reported prior. Moreover, the resultant predictions are in relatively better agreement with experimental data as well as the bowing parameters observed in polarization charge in alloys [212,442]. e31 and e33 are piezoelectric constants. C m2 is Coulomb p per square meter. C31 and C33 are elastic stiffness coefficient or elastic constants. e31 is the proper piezoelectric constant. The data in bold, which are from Refs [212,442], are recommended. The remaining data are from Refs [212,433,442].
j235
j 2 Electronic Band Structure and Polarization Effects
236
functional theory in the Perdew–Wang PW91 version, and the LDA in the Ceperley– Alder–Perdew–Zunger form. For a glossary of these approximations within the DFT method, the reader is referred to Section 2.15. Ultrasoft potentials were used treating Ga and In d-electrons as valence at a conservative cutoff of 325 eV. Finally, the reciprocal space summation was done on a (8 8 8) Monkhorst–Pack mesh. The results of the refined GGA calculations in terms of spontaneous polarization, piezoelectric, and elastic constants so calculated are tabulated in Table 2.25 along with earlier calculations. The two sets of data, the earlier and refined, are within 10% agreement. The results of more refined LDA calculations are also provided. For deference to earlier work, the piezoelectric constants, albeit an incomplete list dealing with e14, for GaN were estimated theoretically for cubic GaN [443] and used in early investigation of piezoelectric effects in GaN [444] and deduced from the mobility data [445], which is indirect particularly in samples containing many scatterers. For comparison, the interface charge in the GaAs/AlGaAs system is used for MODFETs (or HFET) in less than 10% of this figure. An excellent review of the polarization effects can be found in Ref. [446]. 2.12.1 Piezoelectric Polarization
In a polarizable medium, the displacement vector can be expressed in terms of two components due to the dielectric nature of the medium and the polarizability nature of the medium as [447] !
!
!
!
!
!
D ¼ eE þ 4pP in cgs and D ¼ eE þ P in mks units; !
!
!
$ !
$
!
ð2:60Þ
where E and P represent the electric field and polarization vectors. Considering only the piezoelectric component, the piezoelectric polarization vector is given by [448] PPE ¼ e e ;
ð2:61Þ
where e and e are the piezoelectric and the stress tensors. To gain a quantitative understanding of the piezoelectric polarization, the piezoelectric tensor, which is defined as the derivative of the polarization with respect to strain, must be considered. In hexagonal P63mc symmetry, piezoelectric polarization is related to strain through the piezoelectric tensor (ei,j) as [449] 2 3 exx 36 eyy 7 2 3 2 7 Px 0 0 0 0 e15 0 6 6 ezz 7 5 4 Py 5 ¼ 4 0 6 7: 0 0 e24 0 0 6 ð2:62Þ eyz 7 7 Pz 0 0 6 e31 e31 e33 0 4 exz 5 exy Note that e24 ¼ e15 for hexagonal symmetry that reduces to 2 3 2 3 e15 exz Px 4 Py 5 ¼ 4 ðe24 ¼ e15 Þeyz 5; e31 ðexx þ eyy Þ þ e33 ezz Pz
ð2:63Þ
2.12 Polarization Effects
where Pi, eij, and eij represent the electric polarization, electric piezoelectric coefficient, and strain, respectively. It is clear from Equation 2.63 that the piezoelectric polarization along the ½1 100mdirection Px ¼ e15exz ¼ 0, along the ½1120a-direction Py ¼ e15eyz ¼ 0, and along the [0 0 0 1] c-direction Pz ¼ e31(exx þ eyy) þ e33ezz. The strain components given in Equation 2.7 are repeated here for convenience, without shear and eyz ¼ ezx ¼ exy ¼ 0. If only a biaxial strain is present, the in-plain strain can be calculated by relative difference in the in-plane lattice constants of the epitaxial layer and template (buffer or substrate) through a a0 : ð2:64Þ exx eyy e11 e22 ¼ a0 Here, abuffer a0 and aepi a represent the relaxed (equilibrium) in-plane lattice constants of the buffer layer or the substrate, depending on layers and their thicknesses, and of the epitaxial layer of interest, the strained epitaxial layer, respectively. The nomenclature as or a0 for the substrate (buffer) and ae or a for the epitaxial layer is also used. The expression for the out-of-plane strain is e33 ezz ¼
c c0 : c0
ð2:65Þ
Similarly, c0 and c represent the relaxed and the out-of-plane lattice parameters, which would correspond to the buffer layer and epitaxial layer, respectively. In case the in-plane strain is anisotropic, e11 6¼ e22. As it may have already become obvious, the nomenclature used in the literatures for parameters surrounding strain and piezoelectric and elastic constants vary in that, for example, exx and e11 are interchangeably used. The two-way transformation of x 1, y 2, z 3 can be used to convert from one nomenclature to the other. Likewise, PPE, pz pz PPE, PPE are commonly used in literature interchangeably to depict 3 , P 3 , and P piezoelectric polarization. If the subscript 3 is also used as in P PE 3 , it specifically indicates that for biaxial in-plane strain, the only nonvanishing polarization is along the c-axis. Even in cases where the subscript is not employed, the underlying assumption is that the polarization is along the c-direction, as growth of nitride semiconductor structures is performed predominantly on the basal plane. In the same vein, P0, Psp, and Psp indicate spontaneous polarization along the c-axis. DPsp and DP0 both represent differential spontaneous polarization at a heterointerface such as an AlN/GaN interface, a topic which follows the discussion of piezoelectric polarization. The components of the piezoelectric polarization tensor given by Equation 2.62 pz can be expressed in terms of a summation, using Pi instead of PPE, as X pz Pi ¼ eij ej with i ¼ 1; 2; 3 and j ¼ 1; . . . ; 6; ð2:66Þ j pz
where P i is the ith component of the piezoelectric polarization. The wurtzite symmetry reduces the number of independent components of the elastic tensor e to three, namely, e15, e31, and e33. The third independent component of the piezoelectric tensor, e15, is related to the polarization induced by a shear strain that is not applicable to the epitaxial growth schemes employed. The index 3 corresponds
j237
j 2 Electronic Band Structure and Polarization Effects
238
to the direction of the c-axis. It is clear that the piezoelectric properties of the Wz structures are somewhat more complicated. If we restrict ourselves to structures with growth along the [0 0 0 1] direction or z-direction, or along the c-axis, only the e31 and e33 components need to be considered. The piezoelectric polarization in the [0 0 0 1] direction can be obtained by setting i ¼ 3. The electric polarization component in the c-direction, which is designated by z- in the above nomenclature, is given by Pz ¼ e31 exx þ e31 exx þ e33 ezz ¼ 2e31 exx þ e33 ezz :
ð2:67Þ
For isotropic basal plane strain, the strain components exx e?/2 and thus Equation 2.67 can be written as pe
Pz P 3 ¼ e31 e? þ e33 ezz :
ð2:68Þ
In hexagonal symmetry, strain in the z-direction can be expressed in terms of the basal plane strain e? through the use of Poissons ratio, which is expressed in terms of the elastic coefficients Cij as ezz ¼ 2(C13/C33)exx ¼ (C13/C33)e? (see Equations 2.7, 2.14–2.16 for specifics). In the case of externally applied pressure in addition to mismatch strain, the out-of-plane strain can be related to the in-plane strain through ezz ¼ [(p þ 2C13exx)]/C33, where p is the magnitude of compressive pressure (in the same unit as the elastic coefficients). In terms of the nomenclature again, it should also be noted that e1 e11 exx and e3 e33 ezz in the other notation used in the literature and also in this text. C13 Pz ¼ e31 e33 ð2:69Þ e? : C33 pz
3 The z-component of the electric polarization is also referred to as P PE 3 , P 3 , and P pz . Piezoelectric polarization is also described in terms of piezoelectric moduli in the literature, which is treated here for completeness. In terms of piezoelectric moduli, dij, which are related to the piezoelectric constants by X eij ¼ dik Ckj with i ¼ 1; 2; 3 and j ¼ 1; . . . ; 6 and k ¼ 1; . . . ; 6: ð2:70Þ k
Using Equation 2.70 in Equation 2.66 and strain–stress relationship (stress is equal to the product of elastic constant and strain in a tensor form), the piezoelectric polarization can be expressed in terms of piezoelectric moduli as X pz dij sj with i ¼ 1; 2; 3 and j ¼ 1; . . . ; 6: ð2:71Þ Pi ¼ j
Symmetry considerations lead to d31 ¼ d32, d15 ¼ d24 and all other components dij ¼ 0, and thus Equation 2.71 reduces to a set of three equations: 1 1 pz pz P1 ¼ d15 s5 ; P2 ¼ d15 s4 ; 2 2
pz
and P 3 ¼ d31 ðs1 þ s2 Þ þ d33 s3 :
ð2:72Þ
For biaxial strain, which is the case with epitaxial layers, additional conditions are imposed in that s1 ¼ s2, s3 ¼ 0. Moreover, the shear stresses are negligible, which leads to s4 ¼ s5 ¼ 0. Consequently, in cases primarily applicable to epitaxial layers grown along the c-direction, the piezoelectric polarization is left with only one
2.12 Polarization Effects
nonvanishing component, which is in the growth direction and is given by, using Equation 2.72, pz
P3 ¼ 2d31 s1 :
ð2:73Þ
Utilizing stress–strain relationship C2 s1 ¼ e1 C 11 þ C12 2 13 ; C33
ð2:74Þ
we obtain C2 pz P3 ¼ 2d31 e1 C 11 þ C12 2 13 ; C33
ð2:75Þ
pz
where P 3 represents the piezoelectric polarization along the c-direction and similar to Equation 2.69, which expresses the same in terms of piezoelectric constants as opposed to piezoelectric moduli. For hexagonal crystals, the relations between piezoelectric constants and piezoelectric moduli expressed in Equation 2.70 can be reduced to e31 ¼ e32 ¼ C11 d31 þ C12 d32 þ C13 d33 ¼ ðC11 þ C 12 Þd31 þ C13 d33 ; e33 ¼ 2C13 d31 þ C33 d33 ; e15 ¼ e24 ¼ C 44 d15 ; eij ¼ 0 for all other components:
ð2:76Þ
The nonvanishing component of the piezoelectric polarization of Equation 2.66 due to only the biaxial strain is for i ¼ 3 and is given by, which recovers Equation 2.67, pz
P3 ¼ e1 e31 þ e2 e32 þ e3 e33 ¼ 2e1 e31 þ e3 e33 ¼ e? e31 þ e3 e33 because e31 ¼ e32 : ð2:77Þ Steps can be taken to express Equation 2.77 in the form of Equations 2.68 and 2.69, which are straightforward and therefore not presented. We have so far focused on lattice mismatch induced strain. However, the thermal expansion coefficients of the layers used to compose many of the heterostructures are different, which upon cooling from growth temperature could lead to thermalinduced strain. A larger effect in this vein, however, is that caused by differences in the thermal expansion coefficient between the substrate and epitaxial stack used. In that case, the piezo component would have two parts, namely, the lm or misfit strain and by the thermal strain leading to Ppe ¼ Plm þ Pts. Another issue that must be considered is that the electric field induced due to strain (piezoelectric field) in adjacent layers of a heterostructure comprised of A and B (i.e., A for AlGaN, B for GaN, and A,B for AlGaN/GaN) is sp
pe
E A;B ¼ E A;B þ E A;B :
ð2:78Þ
If, for example, A is composed of a ternary, then the linear interpolation for both sp and pe polarizations can be used to a first extent. Again, the nonlinearities are discussed later on in Section 2.12.3.
j239
j 2 Electronic Band Structure and Polarization Effects
240
Another relevant but not discussed nearly as much topic is that the properties that cause the piezoelectric polarization can also lead to pyroelectric effects. Such phenomena can be rather important in nitride-based devices, as the junction temperature is high by the nature of the applications such as lasers and high-power amplifiers. Consequently, the thermally induced electric field, pyroelectric effect, would most likely be present [450] with consequences similar to those ascribed to polarization effects. If one considers the distortion of the u parameter as well, the piezoelectric polarization can be expanded as qP 3 ¼
qP3 qP3 qP3 ða a0 Þ þ ðc c 0 Þ þ ðu u0 Þ: qa qc qu
ð2:79Þ
The internal u parameter is defined as the average value of the projection of the connecting vector of a nitrogen atom with its first neighbor in the ð0 0 0 1Þ direction along this same direction. The three parameters, a, c, and u, are not independent of each other. If the partial derivatives in Equation 2.79 are known, following Ref. [209], one can write for the two piezoelectric constants qP3 4qc 0 qu þ pffiffiffi 2 Z ; qc qc 3a0
ð2:80Þ
e31 ¼
a0 qP 3 2q qu þ pffiffiffi Z ; 2 qa qa 3a0
ð2:81Þ
Z ¼
pffiffiffi 2 3a0 qP3 ¼ ZT3 ; 4q qu
ð2:82Þ
e33 ¼ c 0
where
is the axial component of the Born, or the transverse component of the charge tensor ZT3 . Various structural parameters that are useful in treating the polarization issue in nitride semiconductors are tabulated in Table 2.26.
Table 2.26 Structural parameters of wurtzitic AlN, GaN, and InN
reported in Ref. [209] and updated with DFT in the GGA approximation in Ref. [212]. c0/a0
a0 (Å) Ref. [209] GGA [212] Exp. AlN 5.814 GaN 6.040 InN 6.660
3.1095 3.1986 3.5848
u0
Ref. [209] GGA [212] Exp.
3.1106 1.6190 3.1890 1.6336 3.538 1.6270
1.6060 1.632 1.6180
Ref. [209] GGA [212] Exp.
1.6008 0.380 1.6263 0.376 1.6119 0.377
0.3798 0.3762 0.377
0.3821 0.377
It should be mentioned that GGA produced data for structural as well as polarization-related parameters, see Table 2.25, are in better agreement with refined experimental data.
2.12 Polarization Effects
In Equation 2.80, it is implicit that the vector connecting the cation with the anion has a modulus uc associated with the internal cell parameter and points in the direction of the c-axis. The first term in Equations 2.80 and 2.81 signifies the term called the clamped-ion term, and represents the effect of the strain on the electronic structure. The second term represents the effect of internal strain on the polarization. The derivatives of u with respect to c and a in Equations 2.80 and 2.81 are related to the strain derivatives of u through c0du/dc ¼ du/de3 and a0du/da ¼ 2du/de1. In addition to binaries, the nitride heterojunction system utilizes ternary and to a lesser extent quaternary alloys as well. Knowing the piezoelectric parameters of the end binary points is generally sufficient, to a first order, to discern parameters for more complex alloys. For example, in the case of AlxGa1xN, the piezoelectric polarization vector expression, using linear interpolation within the framework of Vegards law, can be described as [448] $
$
!
Ppe ¼ ½x e AlN þ ð1 xÞ e GaN e ðxÞ:
ð2:83Þ
The same argument can be extended to piezoelectric polarization in quaternary alloys such as AlxInyGa1xyN in a similar fashion as $
$
$
$
!
Ppe ¼ ½x e AlN þ ð1 xÞ e InN þ ð1 x yÞ e GaN e ðxÞ:
ð2:84Þ
The linear interpolation is very convenient and does give reasonably accurate values. However, as will be discussed later in this section, while the Vegards law applies to the alloys, the polarization charge itself is not a linear function of composition [220,221]. 2.12.2 Spontaneous Polarization
Spontaneous polarization calculated for the binary nitride semiconductors are tabulated in Table 2.25. For ternary and quaternary alloys, the simplest approach is to use a linear combination of the binary end points, taking into account that the mole fraction can be used under the auspices of the Vegards law. However, this linear interpolation falls short of agreeing with the experimental variation of spontaneous polarization with respect to the mole fraction. Consequently, nonlinear models have been developed that are discussed later in Section 2.12.3. For now, the linear interpolation is applied for simplicity in which the spontaneous polarization in quaternary alloys such as AlxInyGa1xyN can be expressed as sp þ yP sp sp sp ðx; yÞ ¼ x P P InN þ ð1 x yÞP GaN : AlN
ð2:85Þ
The ternary cases can be obtained by simply setting either x or y to zero. This is again predicated on the assumption that polarization charge obeys Vegards law, as shown below. Later in this section, the nonlinearity involved is discussed. Using the GGA calculation results for the spontaneous polarization and linear interpolation, as in Equation 2.95, for ternaries one gets
j241
j 2 Electronic Band Structure and Polarization Effects
242
sp
PAlx Ga1 x N ¼ 0:09x 0:034ð1 xÞ; sp PInx Ga1 x N ¼ 0:042x 0:034ð1 xÞ; sp PAlx In1 x N ¼ 0:09x 0:042ð1 xÞ:
ð2:86Þ
The total polarization charge inclusive of spontaneous and piezoelectric must be considered in dealing with heterojunctions such as quantum wells, quantum wires, and quantum dots. The case of the quantum wells and dots are treated in Volume 2, Chapter 5. The case of the modulation-doped structures is treated in Volume 3, Chapter 3. Therefore, the total polarization involving layers A (AlxGa1xN) and B (GaN) in contact is Ptotal ¼ P sp þ Ppe :
ð2:87Þ
2
Spontaneous polarization P (C m ), Born Z or the transverse component of the charge tensor ZT3, piezoelectric constants (C m 2), elastic constants (GPa), and the ratio R ¼ 2C31/C33 of wurtzitic nitrides, as obtained in the LDA and GGA approximation [212] are tabulated in Table 2.27. Also, tabulated following the elastic ðpÞ constants e33 and e31 is e31 , which is the applicable piezoelectric constant in the context of experiments dealing with current flow across the sample. The constant e31 is relevant to systems in depolarizing fields such as nitride nanostructures [448]. A comprehensive table including experimental and calculated values of elastic compliance, elastic constants, and piezoelectric constants as well as the Poisson number of wurtzite binary group III nitrides at room temperature are shown in Table 2.28. Table 2.27 Spontaneous polarization, Born effective charges, Z*
(in units of e), piezoelectric constants, dynamical charges, elastic constants (GPa), and the ratio R ¼ 2C31/C33 of wurtzitic nitrides, as obtained with DFT calculations in the LDA and GGA approximation.
AlN LDA LDA [179] GGA GaN LDA [179] LDA GGA InN LDA LDA [179] GGA
P (C m2)
Z
e33 (C m2)
e31 (C m2)
«(p) 31 (C m 2)
C33 (GPa)
0.100
2.652
1.80
0.64
0.74
384 373
111 108
0.578 0.579
0.090
2.653
1.50
0.53
0.62
377
94
0.499
0.032
2.51
0.86
0.44
0.47
415
83
0.400
0.034
2.67
0.67
0.34
0.37
405 354
103 68
0.508 0.384
0.041
3.045
1.09
0.52
0.56
233 224
88 92
0.755 0.821
0.042
3.105
0.81
0.41
0.45
205
70
0.683
The last column reports the proper e31 piezoelectric constant [212].
C31 (GPa)
R¼ 2C31/C33
0.22c
367 135 103 405 95 0.52 3.267 1.043 0.566 2.757 10.53
Theory [179]
1.253 2.291 1.579
0.34 0.67
0.38
68 354
Theory [209,212]
GaN
396 137 108 373 116 0.58 2.993 0.868 0.615 3.037 8.621
370a 145a 110a 390a 90a 0.56 3.326 1.118 0.623 2.915 11.11
0.30d
Theory [179]
Experiment
2.298 5.352 2.069
0.53 1.50
0.50
94 377
Theory [212,433]
AlN
a
For an expanded list of elastic constants, see tables in Chapter 1 under mechanical properties [84]. Ref. [200]. b Ref. [198]. c Shur, M.S., Bykhovski, A.D. and Gaska, R. (1999) MRS Internet Journal of Nitride Semiconductor Research, S41, G16. d Ref. [437].
C11 (GPa) C12 (GPa) C13 (GPa) C33 (GPa) C44 (GPa) n(0001) S11 (1012 N m2) S12 (1012 N m2) S13 (1012 N m2) S33 (1012 N m2) S44 (1012 N m2) e31 (C m2) e33 (C m2) e24 ¼ e15 (C m2) d31 (1012 C m2 Pa) d33 (1012 C m2 Pa) d15 (1012 C m2 Pa)
Parameter
of wurtzite binary group III nitrides at room temperature (theory [179,212,433]).
410a 140a 100a 390a 120a 0.51 2.854 0.849 0.514 2.828 8.333 0.58b 1.55b 0.48b 2.65 5.53 4.08
Experiment 223 115 92 224 48 0.82 6.535 2.724 1.565 5.750 20.83
Theory [179]
Table 2.28 Experimental and predicted elastic compliance, elastic constants, and piezoelectric constants as well as the Poisson number
3.147 6.201 2.292
0.41 0.81
0.68
70 205
Theory [212,433]
InN
2.12 Polarization Effects
j243
j 2 Electronic Band Structure and Polarization Effects
244
In heterojunctions containing donors and acceptors and shallow defects, the associated free carriers within the Fermi statistics diffuse to the semiconductor with the smaller bandgap where they are confined, due to potential barriers, to potential minima. The resulting charge separation due to free carriers causes an internal electric field, screening field, which is represented by the first term in Equation 2.60. In addition, an electric field can also be induced by the application of an external voltage such as done through the use of Schottky barriers, metal oxide semiconductor structures, and p–n junctions. Spontaneous and strain-induced piezoelectric polarization can influence the final status of the interfacial free-charge density in these heterostructures. Any shallow defects (induced fields would change the ionization ratio), free carriers, and surface contacts must be included for a complete treatment. Let us now calculate the polarization charge for model heterojunctions using the linear interpolation method. For AlxGa1xN coherently strained on a relaxed GaN substrate, the strain e? is expected to be proportional to x and given by e? ¼ 2 (aGaN aAlGaN)/aAlGaN, which is 0.051x and is tensile. The piezoelectric polarization is then Ppiezo ¼ 0.0464x, that is, pointing in the ½0 0 0 1 direction. The corresponding difference in spontaneous polarization between AlxGa1xN and GaN is also expected to be proportional to x, the AlN mole fraction, and is given by DPspon ¼ 0.056x. Consequently, the two are in the same direction for this particular orientation and are comparable in magnitude. This treatment assumes that the polarization charge scales linearly with alloy composition, which does not necessarily hold but is used for simplicity. The matter is discussed below in more detail. The total polarization for AlN–GaN interface, which is defined in this case as the sum of the piezoelectric polarization and the differential polarization charge is 0.102x. Note that these are all in C m2 and that 1 C m2 ¼ 0.624 · 1015 electrons cm2. Thus, for x on the order of 0.1, we are dealing with total polarization charge of the order of mid-1012 cm2. In case the ternary AlxIn1xN is used for the barrier, the composition of Al0.82In0.18N can be grown lattice matched to GaN [451] and the piezoelectric polarization vanishes. For lower Al concentrations, that is, x < 0.82, the piezoelectric polarization increases due to the increase in biaxial compressive strain. For higher Al concentrations, that is, x > 0.82, the layer is under tensile strain and the piezoelectric polarization becomes negative. For a coherently strained InxGa1xN layer on relaxed GaN, the difference in spontaneous polarization is much smaller, DPspon ¼ 0.012x. Furthermore, the InxGa1xN layer on GaN would be under compressive strain e? ¼ 0.203x and Ppiezo ¼ þ0.139x. Here, the piezoelectric polarization dominates and is opposite in direction to the spontaneous polarization charge but even larger in absolute magnitude. Unlike for tensile-strained AlxGa1xN or AlxIn1xN (for large x-values) on GaN layer, wherein the piezoelectric and the spontaneous polarization are negative and point in the same direction, thus they add up, the spontaneous and piezoelectric polarizations oppose one another for compressively strained InxGa1xN or AlxIn1xN (for small x-values) layers. To calculate the differential spontaneous and piezoelectric polarization associated with alloys, one can to a first order employ a
2.12 Polarization Effects
linear interpolation for the spontaneous polarization and piezoelectric and elastic constants from the binary compounds [452]. In addition to nonlinear behavior of polarization that is discussed in Section 3.14.3, there are technological issues that must be considered. Some further words of caution about the above estimates based on linear interpolation are needed. If the AlGaN layers are not pseudomorphic but partially relaxed (by misfit dislocations for example), then the piezoelectric effect would be reduced but the spontaneous polarization would still be present. Finally, if domains with inverted polarity exist, the overall polarization effects may be washed out. Also note that in an inverted structure with nitrogen (N) polarity toward the surface, it may be possible to create a two-dimensional hole gas (2DHG) at the AlGaN/substrate GaN interface, provided that free holes are available. However, if an n-type GaN layer is placed on top, a 2DEG may form on top of the AlGaN layer. Polarization effects and devices are inextricable. In devices with a large concentration of free carriers, the polarization charge would be screened. In devices where the modulation of charge is the basis of operation such as the MODFETs, a detailed accounting of all polarization charge must be undertaken. In considering a Normal MODFET (N-MODFET) structure where the larger bandgap AlGaN donor layer is deposited on top of a GaN channel layer, both the spontaneous polarization and the piezoelectric polarization must be accounted for. For an N-MODFET structure with Ga-polarity, the potential will slope down from the surface of the AlGaN layer toward the AlGaN–GaN interface. The topic is discussed in some detail for both AlGaN/GaN and AlxIn1xN/GaN structures in Section 2.12.4.1 and for the InGaN/GaN heterojunction system in Section 2.12.4.2. 2.12.3 Nonlinearity of Polarization
In the treatment above and in the literature for quite sometime, a linear interpolation from the binary end points were used to deal with both piezoelectric and spontaneous components of polarization. Despite the linear interpolation method being reasonably successful in getting the figures to a first extent, as done in conjunction with Equation 2.86, discrepancies with experiments were noted. Thus, efforts continued to refine the polarization figures and, while in the process, entertain whether a bowing parameter such as the case in bandgap of ternaries, while the lattice parameter obeys the Vegards law, that is, linear interpolation, could also be considered for polarization. This forms the basis of the treatment of the problem by Bernardini and Fiorentini [220,221,453] and others [454,455]. The nonlinearity is quite pronounced in AlInN and InGaN alloys for which the binary constituents are very largely lattice mismatched. Spontaneous polarization bowing strongly depends on the microscopic nature of the alloy. What is more is that chemical ordering in the form of short period superlattices may increase the bowing up to a factor of five. Similarly, the piezoelectric polarization is also nonlinear. However, in random alloys, this nonlinearity is entirely due to a nonlinear strain dependence on piezoelectric polarization in pure end binary compounds while the
j245
j 2 Electronic Band Structure and Polarization Effects
246
piezoelectric coefficients follow Vegards law. On the contrary, in chemically ordered InGaN and AlInN alloys piezoelectric coefficients deviate from Vegards law, this effect reduces the strength of the piezoelectric polarization up to 38% of its value in AlInN alloy. However, linear extrapolation is simple and gets reasonably accurate figures. The details of this nonlinearity are discussed below. To reiterate, discrepancies between the experimental data and the theory using linear interpolation led Bernardini and Fiorentini to consider a bowing parameter for polarization in conjunction with ternaries, and it follows that this applies to quaternaries as well. For freestanding alloys, meaning relaxed for this purpose, and assuming that the ternary nitride alloys have random microscopic structure with no strain, the values of the lattice constants within the realm of Vegards law are given by aAlx Ga1 x N ¼ xaAlN þ ð1 xÞaGaN ; c Alx Ga1 x N ¼ xc AlN þ ð1 xÞcGaN :
ð2:88Þ
The spontaneous polarization Psp of an alloy, for example, AlxGa1xN, can be expressed in terms of the polarization values of the binary constituents, AlN and GaN, in the realm of nonlinearities (i.e., non-Vegard behavior). The polarization in an alloy can be described in a generic measurable quantity to a first approximation by a parabolic model involving a bowing parameter, similar to that used for the bandgap of alloys. In conjunction with nonlinear polarization effects in alloys, Bernardini and Fiorentini [453] considered ordered structures to get at the spontaneous polarization across the entire composition range by calculating it for compositions of 0.25, 0.5, and 0.75 in addition to the binary end points. The chalcopyrite-like (CH) structure, used for the 0.5 composition, is formed by each anion site being surrounded by two cations of one species and two of the other, with the overall condition of conforming to periodical (2 · 2 · 2) wurtzite supercell. This structure is highly symmetric, as there are only two kinds of inequivalent anion sites, differing in the orientation of the neighbors but not in their chemical identity. Among the possible ordered structures, CH is in a sense the most homogeneous for a given composition. A further useful ordered structure considered is a luzonite-like (LZ) structure, used for the 0.25 and 0.75 points, resembling zinc blende based alloys. In this structure, each nitrogen atom is surrounded by three cations of one species and one of the other: in a sense, this is the analog to the CH structure for molar fractions x ¼ 0.25 and 0.75. In the comparison between CH and random structures, the latter used for the 0.5 composition could provide insight into the effect of randomness versus ordering without the biases due to specific superlattice ordering, as in the CuPt (CP) structure used for the 0.5 composition point. Luzonite-like structure can provide values of the polarization in the intermediate molar fractions. Shown in Figure 2.35 is a comparison of calculated equilibrium basal and axial lattice parameters a and c for three binaries and their alloys with those determined in the realm of Vegards law (dashed lines). The agreement between the equilibrium a and c lattice parameters so calculated and those determined using Vegards law is quite good. The dependence of the polarization on composition is then the same as that on the lattice parameter(s), modulo a multiplicative factor.
2.12 Polarization Effects
5.9 InN
c-Lattice constant (Å)
5.7 In x G
a
1–x
N
5.5
In x
A l 1 –x
N
GaN 5.3
5.1
A lx G a
AlN
1 –x N
AlN
4.9 0
0.2
0.4
0.6
1.0
0.8
Molar fraction, x InN
a-Lattice constant (Å)
3.6 In x G
a 1 –x
N In x
3.4
A l 1 –x
N
GaN
A lx G a
3.2
1 –x
N
AlN
AlN 3.0 0
0.2
0.4
0.6
0.8
1.0
Molar fraction, x
Random alloy CH-like CP-like Figure 2.35 The basal plane lattice parameter a and axial lattice parameter c of wurtzite nitride binaries and alloys directly calculated versus those determined by Vegards law. The open circles denote the random alloy with 0.5 molar fraction. The dashed lines are Vegards law. Courtesy of F. Bernaridini and V. Fiorentini.
Shown in Figure 2.36 are again the spontaneous polarization values for the aforementioned ternaries in the freestanding strain-free form. The solid lines represent interpolations utilizing Equation 2.95 for AlInN, the binary points of which were determined using 32-supercell calculations. The dashed lines represent the simple Vegards law based interpolations and the numbers indicate the bowing parameters in terms of C m2.
j247
j 2 Electronic Band Structure and Polarization Effects
248
–0.02 In x Ga 1–x N
GaN
Spontaneous polarization (C m–2)
+0.038 InN –0.04 +0.071
+0.019 –0.06
In x Al1–x N Alx Ga1–xN
–0.08
AlN
AlN
–0.10 0
0.2
0.4
0.6
0.8
1
Molar fraction, x Figure 2.36 Spontaneous polarization versus molar fraction in freestanding (strain free) random nitride alloys (solid circles). The solid lines represent the results of the bowing model of Equation 2.95. The dashed lines are determined by Vegards law. The numbers shown in the figure are the bowing parameters expressed in units of C m2. Courtesy of F. Bernaridini and V. Fiorentini.
Depicted in Figure 2.37 are the spontaneous polarization values versus the molar fraction. The solid circles, squares, and triangles represent the values for the random alloy, CH-/LZ-, and CP-like alloys, respectively. The dashed lines represent the values calculated using Vegards law. The numbers shown in the figure represent the bowing parameters for the CH-/LZ- and CP-like ordered alloys. The CH and LZ calculations can be used to verify efficacy of the interpolation model inclusive of the bowing parameter. The ordered LZ structure is analogous to CH for molar fractions of 0.25 and 0.75. The extent to which the LZ values deviate from those calculated by Equation 2.95 and the CH (x ¼ 0.5) value indicates whether or not nonparabolicity occurs in the P(x) relation for CH-like order. Because the polarization of the CH structure behaves qualitatively as that of the random structure (Figure 2.37), the conclusions drawn for CH are applicable to the random phase. Fortuitously, the values of the polarization calculated for the LZ structures at molar fractions 0.25 and 0.75, shown in Figure 2.37, are very close to the parabolic curve for InGaN and AlGaN calculated using the quadratic expression depicted in Equation 2.95, paving the way for the use of the analytical (quadratic) expression for polarization calculations. However, for AlInN, the calculated values are somewhat
2.12 Polarization Effects
Spontaneous polarization (C m–2)
0.02
+0.333 InxAl1–xN
InxGa1–xN
0.00
+0.193
–0.02
InxGa1–xN
–0.04
GaN
+0.095 –0.06
InxAl1–xN
InN AIxGa1–xN
+0.037 –0.08
Random alloy CH-like CP-like
AIN
AIN
–0.10 0
0.2
0.4
0.6
1
Molar fraction, x Figure 2.37 Spontaneous polarization versus the molar fraction in all three ternary nitride alloys. Circles, squares, and triangles represent random alloy, CH-/LZ-, and CP-like structures, respectively. The dashed/dotted lines (blue) with solid triangles are for the CP-like alloys, the dashed lines (green) with solid squares are for CH-like alloys, and solid lines (black) with filled
circles are for random alloys. The black dashed lines represent the data calculated using Vegards law. Numbers indicated in the figure are for CP and CH-/LZ-like ordered alloy bowing parameters in terms of C m2. Courtesy of F. Bernaridini and V. Fiorentini. (Please find a color version of this figure on the color tables.)
above the quadratic relation for x ¼ 0.25 and below it for x ¼ 0.75, indicating some nonparabolicity in polarization. Specifically, the bowing is higher for low In concentration in AlInN. This nonparabolicity is relatively modest, of order 10%, as compared to the quadratic nonlinearity Equation 2.95 for AlInN. One can then conclude that the analytical (quadratic) expression would predict polarization in AlGaN and InGaN fairly accurately and also for AlInN but with about 10% accuracy. To understand the physical origin of the spontaneous polarization bowing, Bernardini and Fiorentini [453] decomposed the spontaneous polarization into three distinct components on the basis of their genesis, namely, the internal structural and bond alternation, volume deformation, and disorder. The internal structural and bond alternation (strain) can be caused by varying cation–anion bond lengths. The volume deformation can be due to compression or dilation of the bulk binaries from their original equilibrium lattice constants to the alloy values. The disorder effect is due to the random distribution of the chemical elements on the cation sites. Bernardini and Fiorentini [453] showed that in ordered alloys the structural
j249
j 2 Electronic Band Structure and Polarization Effects
250
contribution is dominant, the volume deformation accounts for one-third of the bowing found in random alloys, and the effect of disorder appears insignificant in terms of its effect on the bowing of spontaneous polarization. 2.12.3.1 Origin of the Nonlinearity To understand the origin of the structural contribution to the nonlinear behavior, Bernardini and Fiorentini [453] considered that while the bandgap is a scalar, the polarization is a vector of defined direction. The basal plane normal in wurtzite structures is the [0 0 0 1] direction. Thus, the bond length and angle alternation will affect the polarization bowing only if it changes the projection of the bond length along the c-axis, which is the [0 0 0 1] axis. This is consistent with the notion that the polarization in pure binaries is strongly affected by the relative displacement of the cation and anion sublattice sites in the [0 0 0 1] direction [441]. Also consistent is that there is a clear correlation between the u parameter of the wurtzite structure, which is the bond length along the singular polar (or pyroelectric) axis, and the value of the polarization. Shown in Figure 2.38 is the calculated spontaneous polarization of freestanding x ¼ 0.5 alloys of AlGaN, InGaN, and AlInN versus the average internal 0.02 Random CH – like CP – like Spontaneous polarization (C m–2)
0.00 InGaN AlInN – 0.02
– 0.04
AlGaN – 0.06
0.380
0.378
0.376
0.374
Average u lattice parameter Figure 2.38 Spontaneous polarization versus the average internal parameter u in AlGaN, InGaN, and InAlN ternary alloys. Open circles, squares, and triangles, refer to random, CH-like, and CPlike structures, respectively. Courtesy of F. Bernaridini and V. Fiorentini.
0.372
0.370
2.12 Polarization Effects
parameter u. The internal u parameter is defined as the average value of the projection of the connecting vector of a nitrogen atom with its first neighbor in the ð0 0 0 1Þ direction along this same direction. This is in the realm of the wurtzite structure convention in which each anion is situated at the (0, 0, u) Cartesian point from the cations and all the vertical bonds between Ga and N point along [0 0 0 1] direction. This definition can also be used for random phase alloys in spite of the displacement from the ideal sites. Figure 2.38 shows that for a given alloy composition, the spontaneous polarization of relaxed (freestanding) nitride alloys of different microscopic structure depends linearly on the average internal u parameter of the alloy structure. This indicates that spontaneous polarization differences between alloys of the same composition are primarily due to structural and bond alternation effects; disorder appears to have a negligible influence. The structural and bond alternation effects discussed above can also shed light on the random and CP phases in AlGaN having almost the same average u, hence, nearly the same polarization. In InGaN, the random alloy has a larger u than the CH phase, while the opposite is true for AlInN and AlGaN accounting for the CH versus random bowing behavior in InGaN being opposite of that in AlInN and InGaN. Moreover, the large bowing of CP-ordered AlInN and InGaN is consistent with the very large deviation of the average u as compared to the random and CH-like structures. If the internal strain were the only source of polarization bowing, all of the points in Figure 2.38 would fall on the same straight line. This not being the case suggests that another factor related to the chemical identity of the constituents plays a role of some importance and brings to the role of volume deformation. To investigate this role, Bernardini and Fiorentini [453] set up a model based on the polarization in a constrained ideal wurtzite structure in which only the a parameter is a variable, whereas c and u are fixed atp the determined by maximal sphere packing, ffiffiffiffiffiffiffivalues ffi namely, u ¼ 0.375 and c=a ¼ 8=3. Each nitrogen atom is then surrounded by four equidistant cations, meaning all bonds have the same length for a given lattice constant a. By design, the bond alternation caused by perturbation in the internal parameter u would not play any role, and the effects of chemical identity of the constituents can be easily distinguished. To continue to tackle the aspect of volume deformation, Bernardini and Fiorentini [453] assumed that Vegards law holds for the lattice constant a. Naturally, this establishes a linear relationship between the composition and the lattice constant. This segues into the calculation of the polarization in each of the binary nitrides in their ideal structure as a function of the lattice constant a(x). Finally, they could express the alloy polarization as a composition-weighted Vegard-like average of the polarizations of the binary end points (through a reduction of Equation 2.85) as aðxÞ Psp ðAlx Ga1 x NÞ ¼ xPaðxÞ sp ðAlNÞ þ ð1 xÞP sp ðGaNÞ:
ð2:89Þ
In this approach, any nonlinearity must have its origin in the different response of polarization to perturbations in a(x), hence, to hydrostatic compression. To illustrate this point, shown in Figure 2.39 are calculated polarizations in the ideal wurtzite structure for the three ternary alloys, polarization of the binaries in the ideal
j251
j 2 Electronic Band Structure and Polarization Effects
252
InGaN
Spontaneous polarization (C m–2)
–0.015 GaN
InN
–0.020 AlInN –0.025
AlGaN
AlN GaN InN
–0.030 AlN –0.035 3.0
3.1
3.2
3.3
3.4
3.5
3.6
3.7
Equilibrium lattice constant (Å) Figure 2.39 Spontaneous polarization versus the lattice constant in ideal wurtzite structures. Solid circles depict the values of binary compounds and random ternary alloys. Open circles, squares, and triangles represent the polarization calculated as a function of the lattice constant in bulk AlN, GaN, and InN. Solid lines correspond to Vegard interpolations based on
the ideal binaries under hydrostatic pressure. The dashed lines represent the Vegard interpolation of the polarization using the values for the binaries at equilibrium. The data show that the polarization in the alloys is a direct result of the hydrostatic pressure and thus volume deformation. Courtesy of F. Bernaridini and V. Fiorentini.
structure, and polarization interpolated by the Vegard interpolation of Equation 2.89. It can be seen that the calculated values of polarization and those interpolated by Vegards prediction agree well. Because the Vegard interpolation intrinsically account for the volume deformation, the origin of the volume deformation component of the nonlinearity and its large values in In-containing alloys become clear. Essentially, this has its genesis in the fact that polarization decreases with hydrostatic pressure in AlN and GaN, while it increases in InN. Also to be noted, polarizations in the ideal structure are between 35 and 50% of their values in freestanding (relaxed) alloys, and despite the absence of bond alternation, which is designed for the purpose separating the components in effect, the bowing is still very large. The above model dealing with the effect of strain on polarization provides the basis for developing an expression of the bowing parameter bmodel for an ideal wurtzite structure alloy as a function of the polarization response to hydrostatic pressure (for the model case of AlGaN): 0 1 qP qP GaN AlN A AlGaN @ bmodel ¼ ðaGaN aAlN Þ qa qa a¼að1=2Þ 0 1 ð2:90Þ 2 2 1 q P q P GaN AlN A 2@ þ ðaGaN aAlN Þ : qa2 qa2 4 a¼að1=2Þ
2.12 Polarization Effects
The agreement of the latter expression with the b resulting from a fit to the calculated values is very good (e.g., for the extreme case of AlInN, bmodel 20.0225 C m2, while from direct calculation, we get 20.0208 C m2). On the basis of the model, it is now understandable that the AlGaN bowing is pretty moderate because the region of interest is small (3.1–3.2 Å) and the responses to hydrostatic pressure of AlN and GaN are similar. On the contrary, in the large range 3.1–3.6 Å, AlN and InN have opposite behavior, when the huge bowing is found in AlInN alloys. The same goes, although to a lesser extent, for the InGaN alloys. 2.12.3.2 Nonlinearities in Spontaneous Polarization Any nonlinearity in the spontaneous polarization can be treated by using a bowing parameter as commonly employed in interpolating the bandgap of an alloy from the binary point with the help of a bowing parameter. In this vein, the spontaneous polarization for a ternary Psp (AxB1xN) with A and B representing the metal components and N representing nitrogen is given by [84,220,221] sp
sp
sp
PAx B1 x N ¼ xP AN þ ð1 xÞP BN bAB xð1 xÞ: sp
ð2:91Þ
sp
PAN and PBN are the spontaneous polarization terms for the end binaries forming the alloy. The bowing parameter is as defined bAB ¼ 2PAN þ 2P BN 4P A0:5 B0:5 N ;
ð2:92Þ
which requires only the knowledge of the polarization of the ternary alloy at the midpoint, that is, molar fraction x ¼ 0.5. Knowledge of the bowing parameter from Equation 2.97 would lead to the determination of the spontaneous polarization at any composition. For AlxGa1xN, Equations 2.91 and 2.92 take the form sp
sp
sp
PAlx Ga1 x N ¼ xPAlN þ ð1 xÞPGaN bAlx Ga1 x N xð1 xÞ;
ð2:93Þ
bAlx Ga1 x N ¼ 2P AlN þ 2P GaN 4P Al0:5 Ga0:5 N :
ð2:94Þ
with
The first two terms in Equation 2.93 are the usual linear interpolation terms between the binary constituents. The third term, quadratic, represents the nonlinearity. Higher order terms are neglected because their contribution is estimated to be less than 10%. Using the numerical GGA values in Table 2.25 for the spontaneous polarization in AlN and GaN and the bowing parameter for random alloy AlGaN given in Refs [220,221] leads to sp
PAlx Ga1 x N ¼ 0:09x 0:034ð1 xÞ þ 0:0191xð1 xÞ; and sp
sp
sp
PInx Ga1 x N ¼ xPInN þ ð1 xÞPGaN bInx Ga1 x N xð1 xÞ;
j253
j 2 Electronic Band Structure and Polarization Effects
254
with bInx Ga1 x N ¼ 2P InN þ 2PGaN 4PIn0:5 Ga0:5 N : Again, using the numerical GGA values in Table 2.25 and the bowing parameter for random alloy InxGa1xN given in Refs [220,221] leads to sp
PInx Ga1 x N ¼ 0:042x 0:034ð1 xÞ þ 0:0378xð1 xÞ; and sp
sp
sp
PAlx In1 x N ¼ xP AlN þ ð1 xÞP InN bAlx In1 x N xð1 xÞ; with bAlx In1 x N ¼ 2PAlN þ 2PInN 4P Al0:5 In0:5 N: Using the numerical GGA values in Table 2.25 and the bowing parameter for random alloy InxAl1xN given in Refs [220,221] leads to sp
PAlx In1 x N ¼ 0:090x 0:042ð1 xÞ þ 0:0709xð1 xÞ:
ð2:95Þ
In calculating the bowing parameters for the three ternaries mentioned above, Bernardini and Fiorentini [453] used a 32-atom supercell for both alloys and binary nitrides for spontaneous polarization for the end binaries and ternaries. These calculations, while being more efficient in terms of computer time, are not as accurate as those reported in Refs [220,221,441]. In the above treatment, the bowing parameters are taken from Ref. [453], while the spontaneous polarization for the end binaries is taken from Ref. [441]. Relying simply on the 32-atom supercell calculations, the bowing parameters and related spontaneous polarization figures for ordered alloys such as CuPt-like ordered alloy (CP-like), chalcopyrite-like ordered alloy (CH-like), luzonite-like ordered alloy (LZ-like) have also been obtained by Bernardini and Fiorentini [453]. In the series of tables below, the results of such calculations for the aforementioned ordered alloys are given for completeness. The binary figures in terms of spontaneous polarization and the lattice parameter resulting from the 32-atom supercell calculations are also given for consistency. Again, more accurate binary data exist in Refs [441]. For convenience, the lattice parameter (a), spontaneous polarization (Psp), and bowing parameter (bAB) for the three ternary nitride alloys in the form of random, ordered chalcopyrite (CH-like), ordered luzonite (LZ-like), and CuPt-ordered alloy (CP-like) are tabulated in Tables 2.29–2.35, in addition to the data shown in Figures 2.35–2.37.
Table 2.29 The lattice parameter (a) and spontaneous polarization (Psp) for AlN, GaN, and InN determined by 32-atom supercell calculations by Bernardini and Fiorentini.
AlN a (Å) Psp (C m2)
3.1058 0.0897
GaN 3.1956 0.0336
InN 3.5802 0.0434
2.12 Polarization Effects Table 2.30 The lattice parameter (a), spontaneous polarization (Psp), and the bowing parameter (bAB) for random alloy ternaries with a molar fraction of x ¼ 0.5 determined by 32-atom supercell calculations by Bernardini and Fiorentini [453].
R 50%
Al0.5Ga0.5N
In0.5Ga0.5N
Al0.5In0.5N
a (Å) Psp (C m2) bAB (C m2)
3.1500 0.0569 þ0.0191
3.3872 0.0290 þ0.0378
3.3352 0.0488 þ0.0709
Table 2.31 The lattice parameter (a), spontaneous polarization (Psp), and the bowing parameter (bAB) for CuPt ordered alloy (CPlike) with a molar fraction of x ¼ 0.5 determined by 32-atom supercell calculations by Bernardini and Fiorentini [453].
CP 50%
Al0.5Ga0.5N
In0.5Ga0.5N
Al0.5In0.5N
a (Å) Psp (C m2) bAB (C m2)
3.1489 0.0573 þ0.0176
3.3884 þ0.0098 þ0.1934
3.3222 þ0.0168 þ0.3336
Table 2.32 The lattice parameter (a) and spontaneous polarization (Psp) for the luzonite (LZ-like) and chalcopyrite (CH-like) alloy with a molar fraction of x ¼ 0.25 determined by 32-atom supercell calculations by Bernardini and Fiorentini [453].
LZ 25%
Al0.25Ga0.75N
In0.25Ga0.75N
Al0.25In0.75N
a (Å) Psp (C m2)
3.1724 0.0413
3.2920 0.0323
3.4510 0.0385
Table 2.33 The lattice parameter (a) and spontaneous polarization
(Psp) for the luzonite (LZ-like) and chalcopyrite (CH-like) alloy with a molar fraction of x ¼ 0.5 determined by 32-atom supercell calculations by Bernardini and Fiorentini [453]. CH 50%
Al0.5Ga0.5N
In0.5Ga0.5N
Al0.5In0.5N
a (Å) Psp (C m2) bAB (C m2)
3.1474 0.0523 þ0.0374
3.3949 0.0328 þ0.0226
3.3369 0.0427 þ0.0952
j255
j 2 Electronic Band Structure and Polarization Effects
256
Table 2.34 The lattice parameter (a) and spontaneous polarization
(Psp) for the luzonite (LZ-like) and chalcopyrite (CH-like) alloy with a molar fraction of x ¼ 0.75 determined by 32-atom supercell calculations by Bernardini and Fiorentini [453]. LZ 75%
Al0.75Ga0.25N
In0.75GaN0.25
Al0.75In0.25N
a (Å) Psp (C m2)
3.1276 0.0690
3.4828 0.0366
3.2146 0.0564
Table 2.35 The bowing parameter (a) for the random alloy and luzonite (LZ-like), chalcopyrite (CH-like), and CuPt ordered alloys (CP-like) calculated by Bernardini and Fiorentini [453].
Random CH and LZ CP
AlGaN
InGaN
AlInN
þ0.019 þ0.037 þ0.018
þ0.038 þ0.023 þ0.193
þ0.071 þ0.095 þ0.333
2.12.3.3 Nonlinearities in Piezoelectric Polarization The total polarization at heterointerfaces is the sum of spontaneous and piezoelectric polarization. Having treated the nonlinearity in spontaneous polarization, attention must be turned to piezoelectric polarization, specifically, its nonlinear dependence on composition in alloys. Some if not all of the components of the heterojunctions are grown pseudomorphically and are, therefore, under strain on the (0 0 0 1) axis. The ensuing symmetry-conserving strain causes a change in polarization that amounts to a piezoelectric polarization. The aim here is to show that piezoelectricity in nitride alloys is nonlinear, and that this nonlinearity is due to a pure bulk effect with its nonlinear behavior of bulk binary piezoelectric constants versus symmetry-conserving strain. The model heterostructure considered by Bernardini and Fiorentini [453] is a coherently strained alloy grown on a relaxed binary buffer layer (bulk for this purpose) in which the in-plane lattice parameter is aalloy ¼ aGaN. The piezoelectric component is the difference between the total polarization to be obtained and the spontaneous polarization discussed above. Shown in Figure 2.40 is the piezoelectric polarization as a function of the alloy composition. Symbols, which have similar designation as in the spontaneous polarization, represent the calculated polarizations for AlGaN, InGaN, and InAlN alloys as a function of compositions. After the publication of Ref. [453], to reconcile a discrepancy with a paper by Al-Yacoub and Bellaiche [454] who showed that CuPt-like ordering in In0.5Ga0.5N wurtzite-structure alloys causes a sizable deviation of the piezoelectric constants from Vegards like behavior. Bernardini and Fiorentini [220,221] revisited the piezoelectric polarization in nitride alloys. In fact, Figure 2.40 contained here represents the updated calculations. The error in question stems from the fact that the polarization values for the strained and unstrained alloys were subtracted
Piezoelectric polarization (C m–2)
2.12 Polarization Effects
InN
Vegard's Random alloy CH,LZ CP
0.2
0.1
InGaN AlInN GaN
0.0 AlGaN AlN
AlN – 0.1 0
0.2
0.4
0.6
0.8
1
Alloy molar fraction Figure 2.40 Piezoelectric component of the macroscopic polarization in ternary nitride alloys epitaxially strained on a relaxed GaN layer (template). Open symbols represent the directly calculated values for random alloy (circles), CHlike and LZ-like (squares), and CP-like (triangles)
structures, respectively. Dashed lines represent the prediction of linear piezoelectricity, while the solid lines are the prediction of Equation 2.96 using the nonlinear bulk polarization as shown in Figure 2.41. Courtesy of F. Bernaridini and V. Fiorentini.
correctly, but values calculated with different k-point meshes were used instead of those for the same k-point [454,455]. It is clear that, contrary to the spontaneous component of polarization discussed above, the piezoelectric polarization component hardly depends on the microscopic structure of the alloy. One might then ask whether the piezoelectric polarization of the alloy can be reproduced by a Vegard-like model interpolated from the binaries in the form pe
pe
pe
PAlGaN ðxÞ ¼ xPAlN ½eðxÞ þ ð1 xÞP GaN ½eðxÞ; pe P AlN ½eðxÞ
ð2:96Þ
pe PGaN ½eðxÞ
and represent the strain-dependent bulk piezoelectric where polarization of the binary end points. With obvious permutations, this expression can be constructed for InGaN and InAlN ternaries as well. To a first approximation, one may calculate the piezoelectric polarization of the binary compounds for symmetry conserving in-plane and axial strains as pe
PAlN ¼ e33 e3 þ 2e31 e1 :
ð2:97Þ
The piezoelectric constants e can be calculated for the equilibrium state of the binary, AN, and as such they do not depend on strain. The dashed lines in Figure 2.40 represent the piezoelectric term as computed from the above relations using the piezoelectric constants computed for the binaries [433]. The Vegards law of Equation 2.96 when combined with Equation 2.97 clearly fails to reproduce the calculated polarization and misses the strong nonlinearity of the piezoelectric term evident in
j257
j 2 Electronic Band Structure and Polarization Effects
258
Figure 2.40. This is due to a valid nonlinearity of the bulk piezoelectricity of the binary constituents, which is of nonstructural origin. It should be stated that bowing due to the microscopic structure of the alloys is negligible. The argument forwarded by Bernardini and Fiorentini [453] is that they calculate the piezoelectric polarization as a function of the basal strain for AlN, GaN, and InN while optimizing all structural parameters. The results depicted in Figure 2.41 clearly indicate that the piezoelectric polarization of the binaries is an appreciably nonlinear function of the lattice parameter a, which is related to basal strain. Because all lattice parameters closely follow Vegards law, the nonlinearity cannot be related to deviations from linearity in the structure. Bernardini and Fiorentini [453] substitute the nonlinear piezoelectric polarization computed for the binaries into the Vegard interpolation, Equation 2.96 In doing so, they obtain excellent agreement with the polarization calculated directly for the alloys as shown with solid lines in Figure 2.40. This led them to conclude that the nonlinearity in bulk piezoelectricity dominates over any effects related to disorder, structure, bond length and angle alternation, and so on. Importantly, they concluded that Vegards law still holds in calculating the piezoelectric polarization of the III–V nitrides alloys, provided that the nonlinearity of the bulk piezoelectric of the constituents is accounted for. Serendipitously, this means that the piezoelectric polarization of any nitride alloy at any strain can be found by noting the value for x (the composition), followed by calculating the basal plane strain, e(x) from Vegards law, and Ppe from Equation 2.96 using the nonlinear piezoelectric polarization of the binaries (Figure 2.41). This approach is of paramount value in the modeling of nitride heterostructures, especially those with high In content, and AlInN alloys. It should be noted that nonlinearities in the calculated piezoelectric constants of AlN and GaN have also been reported by Shimada et al. [456] but in the realm of
Piezoelectric polarization (C m–2)
0.4
AlN GaN InN
0.3
AlN InN
0.2
GaN 0.1
0.0 0.00
– 0.05
– 0.10
– 0.15
Basal strain Figure 2.41 Piezoelectric polarization in binary nitrides as a function of basal strain (symbols and solid lines) compared to linear piezoelectricity prediction (dashed lines). The c- and u-lattice parameters are optimized for each strain. Courtesy of F. Bernaridini and V. Fiorentini.
2.12 Polarization Effects
volume-conserving strain. A direct quantitative comparison with the results of Shimada et al. and Bernardini and Fiorentini is not possible, because the latter seeks to optimize the volume of the cell so that the stress component along [0 0 0 1] vanishes, which is more appropriate for epitaxial structures. In spite of this substantial difference, there is a common trend between the derivative of the piezoelectric polarization shown in Figure 2.41 (i.e., an effective piezoconstant) and the values of the piezoelectric constants reported in Ref. [456]. The nonlinear piezoelectricity of the binaries can be described by the relations (in C m2) [84] pz
PAlN ¼ 1:808e þ 5:624e2 for e < 0;
pz
P AlN ¼ 1:808e 7:888e2 for e>0;
pz
PGaN ¼ 0:918e þ 9:541e2 ; pz
PInN ¼ 1:373e þ 7:559e2 : ð2:98Þ It is important again to note that the nonlinearity in the bulk piezoelectricity exceeds any effects related to disorder or bond alternation, which have been taken into account [220,221]. The calculation of the piezoelectric polarization of an AxB1xN alloy for any level of strain would proceed with first calculating the strain e ¼ e(x) for a given molar fraction, x, using the Vegards law, and the piezoelectric polarization by pz
pz
pz
PAx B1 x N ¼ xP AN ðeÞ þ ð1 xÞPBN ðeÞ; pz
ð2:99Þ
pz
where PAN ðeÞ and PBN ðeÞ are the end binary strain dependent piezoelectric polarizations that can be calculated for a given strain, eðxÞ, using Equation 2.98. Application of this process to each of the three ternaries for all the possible cases of ternaries is as follows. Using Equation 2.96 and linear interpolation for the elastic constants that are tabulated in Table 2.28 and strain determined from Equation 2.64, the piezoelectric polarization between a given ternary and binary can be calculated, as represented below (in C m 2) [84]. pz
PAlx Ga1 x N=GaN ¼ 0:0525x þ 0:0282xð1 xÞ; pz P Alx Ga1 x N=AlN ¼ 0:026x þ 0:0282ð1 xÞ; pz PAlx Ga1 x N=InN ¼ 0:28x 0:113ð1 xÞ þ 0:042xð1 xÞ:
ð2:100Þ
pz
PInx Ga1 x N=GaN ¼ 0:148x þ 0:0424xð1 xÞ; pz PInx Ga1 x N=AlN ¼ 0:182x þ 0:026ð1 xÞ 0:0456xð1 xÞ; pz P Inx Ga1 x N=InN ¼ 0:113ð1 xÞ þ 0:0276xð1 xÞ:
ð2:101Þ
pz
P Alx In1 x N=GaN ¼ 0:0525x þ 0:148ð1 xÞ þ 0:0938xð1 xÞ; pz PAlx In1 x N=AlN ¼ 0:182ð1 xÞ þ 0:092xð1 xÞ; pz PAlx In1 x N=InN ¼ 0:028x þ 0:104xð1 xÞ:
ð2:102Þ
The calculated values are a function of molar fraction for the ternaries for the epitaxial layer and template combinations represented in Figure 2.42, assuming that the template is fully relaxed and the epitaxial layer is completely coherently strained
j259
j 2 Electronic Band Structure and Polarization Effects
260
with strain relaxation. For partial relaxation, unless the degree to which the relaxation that occurs is known, the calculations cannot be made. The extent of relaxation just depends on whether strain-relieving defects propagate from the template to the epitaxial layers. In addition, the effect of cooldown-induced thermal mismatch strain due to cooling from the growth temperature down to the operating temperature of the structure must also be considered. As stated, the results shown are for a perfect system with fully relaxed template and fully strained epitaxial layer on top of it. We should point out that for a particular pair with a particular composition, Al0.82In0.18N/ GaN heterostructure, there is a perfect lattice match; thus, the misfit-induced piezoelectric polarization is equal to zero. It should also be pointed out that other
0.2
Al x In1–x N/AlN In x Ga1-x N/AlN
b = 0.092
Piezoelectric polarization (C m–2)
b = – 0.046
0.1
A1 x Ga 1–x N/AlN b = 0.025 0.0
–0.1 0 (a)
0.2
0.4
0.6
0.8
1.0
Molar fraction, x
Figure 2.42 (a) Piezoelectric polarization of fully and coherently strained ternary alloys on fully relaxed AlN templates in the case of which the ensuing positive piezoelectric polarization and the negative spontaneous polarization are antiparallel. The bowing parameters describing the nonlinearity in the compositional dependence of Ppz are also indicated. (b) The piezoelectric polarization of coherently strained ternary alloys on fully relaxed GaN template. Note that Al0.82In0.18N/GaN heterojunction (indicated with an arrow) is lattice matched, thus
strain and piezoelectric polarization vanishes. (c) Also note that other experiments indicate the lattice matching composition for AlInN on GaN where the piezoelectric polarization charge vanishes to be different. The piezoelectric polarization of coherently strained ternary alloys on fully relaxed InN template. For ternary alloys grown on InN, the negative piezoelectric polarization and the spontaneous polarization are parallel and oriented to along the c-axis. Courtesy of O. Ambacher.
2.12 Polarization Effects 0.2
Piezoelectric polarization (C m–2)
Al x In1–x N/AlN b = 0.094
In x Ga 1–x N/GaN b = – 0.042
0.1
0.0
A1 x Ga 1–x N/GaN b = 0.028
–0.1
0
0.2
(b)
0.8
0.6
0.4
1.0
Molar fraction, x
0.0
Al x In 1–x N/InN b = 0.104
Piezoelectric polarization (C m–2)
In x Ga 1–x N/InN b = – 0.028 –0.1
–0.2
Al x Ga1–xN/InN b = 0.042
–0.3 0
(c) Figure 2.42 (Continued )
0.2
0.4
0.6
Molar fraction, x
0.8
1.0
j261
j 2 Electronic Band Structure and Polarization Effects
262
Nonlinear
InN 0.4
Piezoelectric polarization, P3 (C m–2)
AlN GaN 0.2
0.0
AlN
–0.2
–0.4
–0.2
–0.1
0.0
0.1
0.2
Strain, ε pz
Figure 2.43 The piezoelectric polarization (P 3 ) of wurtzitic binary group III nitrides under biaxial tensile or compressive strain calculated using Equations 2.98 and 2.99, the premise of which relies on nonlinearity as depicted in the figure. Only the AlN case is shown for positive strain consistent with Equation 2.98. In addition, the values of polarization exceed those of the linear model expressed in Equation 2.69. Courtesy of O. Ambacher.
experiments indicate the lattice matching composition for AlInN on GaN, where the piezoelectric polarization charge vanishes, to be different. The piezoelectric polarization versus strain, ignoring nonlinearity, can be calculated using Equation 2.69 for both compressive and tensile biaxial strain for all the epitaxial layer/template configurations. Use of Equations 2.98 and 2.99, however, takes the nonlinearity discussed above into consideration, as shown in Figure 2.43. As expected, the piezoelectric polarization for a given strain increases from GaN to InN and AlN. Significantly, higher piezoelectric polarizations, especially in cases of high strains (high In concentrations), result when nonlinearities in the piezoelectric charge are considered. The piezoelectric polarization associated with coherently strained random ternary AlGaN, InGaN, and AlInN alloys grown on GaN templates is shown in Figure 2.44. To reiterate, the piezoelectric polarization is nonlinear with respect to the alloy
2.12 Polarization Effects
Piezoelectric polarization, P3 (C m–2)
0.2
Al x In1–x N
In x Ga 1–x N
0.1
0.0
Al xGa1–x N/
0
0.2
0.4
0.6
0.8
1.0
Molar fraction, x Figure 2.44 The piezoelectric polarization of random coherent ternary alloys on relaxed GaN templates. Those calculated using Equation 2.69 and linear interpolations of the physical properties such as the piezoelectric constants and elastic coefficients (exy , Cxy) for the relevant binaries are shown in dashed lines. Calculations considering the nonlinearity in the piezoelectric polarization in terms of strain reflected in
Equations 2.98 and 2.99 are shown in solid lines. For alloys under high biaxial strain such as those containing high In concentration, the piezoelectric polarization is underestimated by the linear approach. It should be pointed out that the lattice matching composition for AlInN on GaN where the piezoelectric polarization charge vanishes is converging on 16–18% In in the InxAl1xN lattice. Courtesy of O. Ambacher.
composition, spanning a range of 0.461 to 0.315 C m2, the end points corresponding to coherently strained AlN on InN and InN on AlN, respectively. A point of side interest, albeit not practical, is that in alloys with high indium concentration, leading to high strain, the piezoelectric polarization is larger than that of the spontaneous polarization. As mentioned earlier in conjunction with Figure 2.42 that Al0.82In0.18N can be grown on GaN in a lattice-matched form making the piezoelectric polarization for this heterostructure vanish. To underline the point, the nonlinear model described by Equations 2.98 and 2.99 leads to larger piezoelectric polarization charge than linear interpolation where strain, elastic constants, and piezoelectric constants are assumed to vary linearly with composition.
j263
j 2 Electronic Band Structure and Polarization Effects
264
2.12.4 Polarization in Heterostructures
Investigation of heterostructures such as quantum well and single heterointerfaces such as those present in modulation-doped structures are ideal platforms for putting to test the knowledge base of spontaneous and piezoelectric polarization. Depending on the polarity of the sample, that is, Ga or N, the order of growth, meaning GaN on AlGaN or the other way around, and the buffer layer that determines to a first extent whether the barrier, the well, or both are in strain, the piezoelectric and spontaneous polarization charge add or subtract. Regardless of this, the resultant band bending leads to a red shift in energy due to the presence of electric field, commonly referred to as quantumconfined Stark effect (QCSE), and the ensuing deformation of the wave functions being pushed to opposite ends of the interface leads to reduction of radiative efficiency, increase in lifetime, and also blue shift with injection that is screening. In parallel, if the well size is small enough, carrier confinement induces a blue shift. Consequently, the red shift induced by polarization and the blue shift induced by localization compete in determining the transition energy. The optical transitions both in CW and timeresolved forms are discussed in detail in Volume 2, Chapter 5. Here, the manifestation of polarization charge in heterostructures is treated. We should mention that the electric field resulting from the polarization in GaN would cause a large Stark shift in optical measurements and reduce the effective bandgap in, for example, quantum wells (see Volume 2, Chapter 5), which is of paramount importance in optoelectronic devices. The field induced has been calculated in the context of the GaAs-based superlattices, which show notable Stark shift when grown along the [1 1 1] direction [457]. Stark shift can be screened on a length scale of the order of the Debye length ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pThe ekT=q2 n by injecting free carriers in the GaN layer(s), as is the case in LEDs, lasers, and PL experiments with large excitation intensities. In such a case, the strain-induced field causes carrier separation that, in turn, causes a field that opposes the strain field. Strain-induced polarization in such a heterostructure can also lead to a net electric field that can be measured as a voltage drop across the sample. The impact of the Stark shift and concomitant screening has been considered for laser gain also [458]. Fiorentini et al. [448] and Bernardini and Fiorentini [459] demonstrated that for an alternating sequence of wells (w) and barriers (b), the total electric field in the well can be calculated by recognizing that the normal components of the displacement vector D of Equation 2.60 are continuous (provided that there is no interface charge). total eW E W þ 4pPtotal W ¼ DW ¼ DB ¼ eB E B þ 4pP B
ð2:103Þ
in cgs units (for MKS remove the 4p terms). Utilizing Equation 2.103 together with the periodicity-imposed equality of voltage drops being equal but with opposite sign LW E W ¼ LB E B : One arrives at
ð2:104Þ
2.12 Polarization Effects total E W ¼ 4pLB ðPtotal W P B Þ=ðLW eB þ LB eW Þ;
or
total E W ¼ 4pðPtotal W P B Þ=ðLW =LB eB þ eW Þ;
total E B ¼ 4pLW ðPtotal B P W Þ=ðLB eW þ LW eB Þ; 0 1 L B total @ or E B ¼ 4pðPtotal þ eB A: B P W Þ= LW eW
ð2:105Þ
where eW and eB are the dielectric constant of the well and barrier layers, respectively. Likewise, LW and LB represent the well and barrier thicknesses. The same notation is used for polarization also. Both W and B superscripts and subscripts relate to wells and barriers, respectively. The total polarization term can be changed to piezoelectric term and spontaneous polarization term for cases when only the former or the latter is in effect, respectively. total Explicit in Equation 2.105 is that whenever Ptotal W „P B there will be an electric field. Due to strain or screening, the electric field will be present in both the well and the barrier. If we limit ourselves to piezoelectric polarization and somehow achieve relaxed heterostructures, the piezoelectric polarization charge and therefore the field will be zero. The electric field in wells and barriers in a quantum well has two components, one from spontaneous and the other from piezoelectric polarization. If the thickness of the wells and barrier are the same, the field in wells and barriers are related to each other as sp
pe
sp
pe
E W þ E W ¼ E W ¼ E B ðE B þ E B Þ;
ð2:106Þ
where EW and EB represent the electric field in the well and barrier material. The superscripts indicate the field due to spontaneous and the other from piezoelectric polarization. Additional comments that can be made are that if lattice-matched AlInGaN alloy is used, the piezoelectric component in that material is eliminated. However, the spontaneous component would still be present. Another point to be recognized is that the piezoelectric field induced in InGaN and AlGaN layers of the same composition of In in the former and Al in latter, if grown on relaxed GaN, is larger in the former because of the larger lattice mismatch between GaN and InN as compared to that between GaN and AlN. The spontaneous polarization in alloys can be found using Equations 2.85 and 2.86 together with values listed for the binary end points in Table 2.25. The spontaneous polarization so formulated together with piezoelectric polarization of Equation 2.77 (and also Equation 2.83) would allow the computation of total polarization charge. This must be done for both the barrier and well material. The electric field can then be calculated using Equation 2.105. Any free carrier present in the well and barrier regions as well as those injected by optical and/or electrical means tend to screen the polarization-induced field. A complete picture can be obtained by solving self-consistently a set of effective mass theory or tight binding theory and simultaneously the Schr€ odinger–Poisson equation [460], which is discussed in Section 2.1.14. Combining the ab initio calculations
j265
j 2 Electronic Band Structure and Polarization Effects
266
of Fiorentini and Schr€odinger–Poisson solver of Di Carlo allowed the authors to calculate the charge distribution and field profiles in quantum wells in the presence of free carriers [448,461]. Poissons equation is solved using the boundary condition that the electric field is zero at the ends of the simulated regions. This corresponds to LW reaching the infinity limit in Equation 2.105. The potential thus obtained across the structure is plugged into the tight binding Schr€ odinger equation, which is solved to obtain energies and wave functions [462]. The new quasi-Fermi levels are then calculated, which then lead to carrier concentrations, and the procedure is iterated until self-consistency is obtained. A case of interest for FETs and parts of quantum wells is one that features a ternary on the surface where polarization charge would exist. The same is true for the interface between an AxB1xN and a GaN heterostructure. The characteristic of this charge is that it changes abruptly, leading to a fixed two-dimensional polarization charge density s on the surface and at the interface, which is given by sp
pz
sAx B1 x N ¼ P Ax B1 x N þ P Ax B1 x N sp
pz
on the surface and sp
pz
sAx B1 x N=GaN ¼ ðPGaN þ P GaN Þ ðP Ax B1 x N þ P Ax B1 x N Þ at the interface; ð2:107Þ respectively. Figure 2.45 shows the polarization-induced surface and interface sheet density s/q for relaxed and coherently strained binary nitrides and ternary–binary AxB1xN/GaN interfaces, such as AlxGa1xN/GaN. The spontaneous polarization surface charge density on relaxed InN, GaN, and AlN layers is 2.62 · 1013 cm2, 2.12 · 1013 cm2, and 5.62 · 1013 cm2, respectively (there is no piezoelectric charge in this case). If a biaxial compressive strain of e ¼ 0.002 is present, the surface charge is reduced to 0.72 · 1013 cm2 · 1013, 0.74 · 1013, and 3.22 · 1013 cm2, as can be deduced from Figure 2.45a. For compressively strained InN, GaN, and AlN at the level of e ¼ 0.025, 0.030, and 0.045, respectively, the piezoelectric polarization fully compensates the spontaneous polarization and thus no polarization-induced field should exist. As can be seen from Figure 2.45a, polarization-induced surface charge is reduced by compressive and increased by tensile strain. For fully relaxed layers grown on the c-axis on a substrate, the spontaneous polarization-induced surface charge is negative for Ga-polarity and positive for N-polarity samples. The bound surface charge can be screened by oppositely charged surface defects and charges adsorbed on the surface. If this screening were incomplete, the carrier concentration profiles in the crystals would be different. Consequently, surface band bending, as well as that of interface, would be affected by polarization, and further such band bending would be dependent of the polarity of the sample. Experimental results are not sufficiently complete to draw a definitive conclusion. For coherently strained Ga-face AlxGa1xN/GaN (for 0 < x 1) and AlxIn1xN/ GaN (for 0.71 < x 1 interfaces), the polarization-induced interface charge is calculated to be positive as shown in Figure 2.45b. For both ternary and binary interfaces, the bound polarization charge increases nonlinearly with composition, x,
2.12 Polarization Effects
up to 7.06 · 1013 cm2. However, for Ga-face InxGa1xN/GaN (for the entire compositional range, 0 < x 1) and AlxIn1xN/GaN (or 0 < x 0.71), the polarization charge is calculated to be negative and nonlinear with respect to composition. At the limit for the former, InN/GaN structure, a bound sheet density of 14.4 · 1013 cm2 is calculated. Upon screening in n-type heterostructure, a positive polarization sheet 4
3 AlN Bound surface density s /q (1014 cm–2)
2
1
0
AlN
–1 InN –2
–3 GaN –4 –0.2
–0.1
0.0
0.1
0.2
Strain, ε (a) Figure 2.45 (a) In-plane biaxial strain dependence of surface polarization (piezoelectric plus spontaneous) charge for wurtzitic GaN, InN, and AlN. As in the case of Figure 2.43, the AlN case is with positive strain. (b) Bound interface charge density for coherent AlxGa1xN, InxGa1xN, and AlxIn1xN grown on relaxed Ga-polarity GaN. Polarization-induced
bound interface charge (positive in n-type samples and negative in p-type samples) is screened by free electrons in n-type and holes in p-type samples, respectively, leading to twodimensional gas. Note that the interface charge can be converted into carrier concentration by dividing it with the electronic charge of q ¼ 1.602 · 1019 C. Courtesy of O. Ambacher.
j267
j 2 Electronic Band Structure and Polarization Effects
268
2.0 Ga-Polarity on GaN
2DEG
Bound interface density s /q (1014cm–2)
1.5
1.0 AlN 0.5
N A l xG a1 –x
GaN 0.0
–0.5 In
x Ga
N
I n 1 –x
Al x
–1.0
1– x
N
InN
–1.5
InN 2DHG
–2.0 0
0.2
0.4
0.6
0.8
1.0
Molar fraction, x (b) Figure 2.45 (Continued )
charge, which is bound, leads to a 2DEG with a sheet carrier concentration close to the concentration of the bound interface density þs/e. Now that a conceptual issues related to polarization has been discussed, the attention can now be turned to calculating the sheet charge distribution and band structure at a single heterointerface typically used for MODFET. The available data indicate that GaN channel devices are the only viable ones, even though In-containing channels may someday work well as well. For the moment, the likely GaN channel MODFET structures utilize AlxGa1xN, AlxIn1xN, or perhaps the quaternary. The criterion is that the bandgap of AxB1xN ternary must be larger than that of GaN. As discussed often, the AlxIn1xN ternary lattice matches GaN; however, the reported values for lattice matching composition vary. For example, AlN molar fraction of 71% (corresponds to InN molar fraction of 29%) and 82–83% (corresponds to InN molar fraction of 18–17%, which is the more likely lattice matching composition) have been reported to lattice match GaN [451]. To treat the MODFET interface charge problem,
2.12 Polarization Effects
the bandgap of nitride semiconductor alloys, which are discussed in Section 1.5 in detail, must be known. For completeness, the expressions for the compositional dependence of the alloy bandgap are repeated below. It should be pointed out that the bandgap bowing parameter data bAx B1 x N for AlGaN are converging onto a value of nearly 1 eV even though early figures spanned from 0.8 eV (upward bending) to þ 2.6 eV (downward bending). The data for InGaN are fluidic also in part because of reasons having to do with difficulties in determining the composition and bandgap, detailed in Section 1.5.2, and also lack of high-quality samples with composition midway in the alloy range. For example, optical reflectivity measurements together with PL data corrected for Stokes shift led to a bInGaN ¼ 2.5 0.7 eV for 0.9 eV bandgap of InN and bInGaN ¼ 3.5 0.7 eV for 1.9 eV InN bandgap [463]. PL data alone in another report states the bowing parameter to be 4.5 eV [464]. The figure determined by optical transmission measurements is 8.4 eV [465], while the theory indicates 1.2 eV [466]. Other experimental values for lower InN molar fraction end of the ternary are near null, meaning linear variation of the bandgap [467,468]. The reported data in aggregate, taking into consideration the small bandgap of InN, leads to a value of 2.53, as discussed in Section 1.5.2. The value for the bowing parameter in AlInN is somewhat too controversial. For example, one particular theoretical report points the bowing parameter to be 2.53 eV [469] while a value of 3.1 eV was determined by fitting the bandgap of this alloy determined by photoreflection [470], 2.384 eV by absorption measurements but by using 2.0 eV for the bandgap of InN, and 5.9 eV for AlN [471] and 5.4 eV in a review article where 1.95 eV was used for InN bandgap [84]. However, when all the available data are considered in aggregate with 0.7 eV bandgap for InN, a bowing parameter of about 3 eV appears to be a very good value. g
g
g
E Alx Ga1 x N ¼ xE AlN þ ð1 xÞE GaN bAlGaN xð1 xÞ ¼ 6:1x þ 3:42ð1 xÞ xð1 xÞ eV; g E Inx Ga1 x N
g
g
¼ xE InN þ ð1 xÞE GaN bInGaN xð1 xÞ ¼ 6:1x þ 0:7ð1 xÞ 1:43xð1 xÞ eV;
g
ð2:108Þ
g
E Alx InN ¼ xE AlN þ ð1 xÞE EInN bAlInN xð1 xÞ ¼ 6:1x þ 0:7ð1 xÞ bAlInN xð1 xÞ eV: The presence of displacement or polarization gradient leads to a volume charge density rv given by !
!
!
r: D ¼ r:ðe E þ P total Þ ¼ rv
!
in MKS units: For cgs; add 4p before P : ð2:109Þ
In a one-dimensional system, which is considered here along the c-axis, and utilizing E ¼ dV/dz, Equation 2.109 can be rewritten as the Poissons equation in the form of dD=dz ¼ d=dzð eðzÞdV=dz þ P total ðzÞÞ ¼ qrz ¼ q½N Dþ ðzÞ þ pðzÞ nðzÞ N A ðzÞ:
ð2:110Þ
j269
j 2 Electronic Band Structure and Polarization Effects
270
The term on the right-hand side of Equation 2.110 represents the volume density of net charge, and the position-dependent quantities D, e, and V are the displacement field, dielectric constant, and potential, respectively. The term Ptotal is the position-dependent total transverse polarization (along the c-axis and perpendicular to the interfaces). N Dþ ðzÞ and N A ðzÞ represent the ionized donor and acceptor concentrations, and p(z) and n(z) represent the hole and electron concentrations, respectively. The effects of composition, polarization, and free-carrier screening are thus fully included. With the knowledge of the band profile for both the valence band and the conduction band, one can determine the electronic states in the heterostructure by solving the Schr€odinger (Equation 2.59) equation, as a function of the spatial coordinate z. In the effective mass approximation, one needs to solve the following eigenvalues problem: 2 d 1 dCi ðzÞ h þ V c ðzÞCi ðzÞ ¼ E i Ci ðzÞ; 2m0 dz mz dz
ð2:111Þ
for which the appropriate choice of the effective masses in the conduction band and in the valence band is done by preliminary calculation using the tight binding approximation. The solution of Equation 2.111 determines the eigenstates Ei in the conduction band and in the valence band, and the corresponding eigenfunctions Ci. It is imperative to note that free carriers cannot eradicate the polarization charge, which is bound and invariable (unless structural changes are made), but one can screen it to a degree determined by carrier concentration. Likewise, polarization charge is bound charge and cannot by itself be the source of free carriers, but it can cause a redistribution of free carriers that would tend to screen the polarization charge. In addition to the polarization charge related parameters that must be known to solve the Schr€odinger and Poisson equations, one must also know the dielectric constants, doping levels, effective masses, and band discontinuities, in addition to compositions and the layer thicknesses used. The band discontinuity issue is discussed in Volume 2, Chapter 1, but suffice it to say that the bandgap of the ternaries is found from Equation 2.108 with the appropriate bowing parameter, and a certain fraction of the total bandgap difference is assigned to the conduction band and the rest to the valence band. To mitigate the process, the band alignment for all the nitride heterostructures is of type I, which means that the larger bandgap straddles the smaller bandgap one. Typically, 60–70% of the total band discontinuity is assigned to the conduction band. A linear interpolation for the effective masses, given in Chapter 1, in ternaries from binary end points can be used. While the tables provided in Chapter 1 list the dielectric constants for binaries, knowledge of them leads to the ternary dielectric constants via linear interpolation. Relative dielectric constants of 10.28, 10.31, and 14.61 (compares with 15.1 presented in Chapter 1) for dielectric constants in GaN, AlN, and InN, respectively, have been calculated [472]. The GaN dielectric constant is nominally larger than that of AlN, but in this particular investigation the data are as listed above. Following the so calculated figures and applying the linear interpolation leads to dielectric constant for ternaries.
2.12 Polarization Effects
eAlx Ga1 x N =e0 ¼ 10:28 þ 0:03x; eInx Ga1 x N =e0 ¼ 10:28 þ 4:33x; eAlx In1 x N =e0 ¼ 14:61 4:33x: Using the values for the binary dielectric constants tabulated in Chapter 1, the linear interpolation scheme for the ternary dielectric constants lead to eAlx Ga1 x N =e0 ¼ 10:4 1:9x; eInx Ga1 x N =e0 ¼ 10:4 þ 4:9x; eAlx In1 x N =e0 ¼ 15:3 4:9x:
ð2:112Þ
In dealing with devices such as MODFETs where a Schottky barrier is placed on top, the Poissons equation, through boundary conditions, is affected by the metal barrier height as well as the applied bias on the metal. The Schottky barrier height for the ternaries can also be interpolated from the binary end points. It should also be noted that the barrier height depends on the particular metal used, the details of which is discussed in Volume 2, Chapter 1. Because the Fermi level on the surface of GaN is not fully pinned in that heavier metals with large work functions lead to larger barrier heights, it is plausible and natural to assume that the barrier height would increase with an increase in AlN content and decreases with InN content. However, for InN the Fermi level is most likely in the conduction band already, as in InSb and InAs. Limited studies of metal contact potential on nitride semiconductors make it difficult to state a molar fraction dependence of this parameter. However, Ti and Ni contacts on Al0.15Ga0.85N have been studied and were compared to those on GaN. The barriers heights for Ti and Ni increased on Al0.15Ga0.85N [473]. Comparing the data deduced for Ni using I–V, C–V, and photoemission methods, barrier heights of 0.95 (0.84 eV if not corrected for the nonideal ideality factor), 0.96, and 0.91 eV, respectively, have been obtained. For Al0.15Ga0.85N, the same figures are 1.25 (1.03 eV if not corrected for the nonideal ideality factor), 1.26, and 1.28 in order, using the same methods. These figures represent an increase of about 0.3 eV in barrier height for Al0.15Ga0.85N over GaN. Assuming that Ni is used for Schottky barriers, the barrier height for GaN is 0.95 eV, and more boldly reported figures for one mole fraction do actually represent a figure consistent with a linear interpolation, one can express the molar fraction dependence of the barrier height on AlxGa1xN as fAlx Ga1 x N ¼ 0:95 þ 2xV:
ð2:113Þ
Following Ref. [84], the linear interpolation for the other two ternaries modified for barrier height on GaN used here are fInx Ga1 x N ¼ 0:95 0:36xV; fAlx In1 x N ¼ 0:59 þ 1:36xV:
ð2:114Þ
These expressions lead to barrier heights (qf or ef) of 0.95, 2.95, and 0.59 eV on GaN, AlN, and InN, respectively. With the exception of the value for GaN, the rest is really speculative at this point.
j271
j 2 Electronic Band Structure and Polarization Effects
272
2.12.4.1 Ga-Polarity Single AlGaN/GaN Interface Returning to single-interface structures, such as those used in MODFETs (or HFETs), the total interface charge, for example, at a gated AlxGa1xN and GaN interface, rs, for an n-type case would be sum of total polarization charge rp and free-carrier charge ns. er e0 E Fi DE C ; ð2:115Þ rs ¼ rp ns ¼ rp V G fB V p2 þ qdAlGaN q
where VG is the applied gate bias in terms of V, fB is the Schottky barrier height in terms of V, on AlxGa1xN, EFi is the Fermi level in GaN at the interface with respect to the edge of conduction at the interface, DEC is the conduction band discontinuity between the ternary AlxGa1xN and GaN, and Vp2 is the voltage drop across the doped AlGaN, which in turn is given by V p2 ¼ qN d d2d =2er e0 where Nd is the donor concentration in AlxGa1xN, all assumed to be ionized, dd is the thickness of the doped AlGaN, and ere0 is the dielectric constant of AlxGa1xN. Vp2 is negative for depleting voltage in the case of which it would add to the Schottky barrier height. We should point out at this stage that the form of Equation 2.115 is good for determining the sheet carrier concentration at the interface, but when used for FETs both the polarization and the charge induced by doping must be lumped into the threshold or off voltage. For details see Volume 3, Chapter 3. For a p-type semiconductor, the semiconductor charge as well as the reference for the Fermi level and the band discontinuity should be that at the valence band edges [88,474]. The effect of any undoped AlxGa1xN layer designated as having a thickness of di is small because its thickness is several nanometers and is neglected in Equation 2.115. This setback layer was originally employed by the author in the GaAs system to further reduce remote Coulomb scattering for increased mobility. A detailed treatment including the effect of di can be found in Ref. [474]. For an undoped AlGaN/GaN heterostructure, the Vp2 term can be set to zero. In this case the boundary conditions for potential or the Fermi level are made consistent with the Schottky barrier height on the Alx Ga1 x N surface. In the bulk of the structure, due the special nature of the quasi-triangular barrier at the interface, shown in Figure 2.46, the Fermi level is generally taken to be near the midgap of the smaller bandgap material, which in this case is GaN. As in the case of quantum wells, the one-dimensional Schr€ odinger–Poisson solver can be iteratively used to determine the band and carrier profile (simultaneous solution of Equations 2.110 and 2.111 in an iterative mode either in the tight binding realm or the effective mass approximation, but in the current example it is in the effective mass realm). The bound charge can be represented by a thin and heavily doped interfacial layer, the thickness of which is about 1 nm or less, keeping in mind that the total charge associated with such a fictitious layer must be equal to the bound charge. In one investigation, a thickness for this fictitious layer of 0.6 nm was used. To reiterate, to solve this pair of equations boundary conditions at the interface and surface as well as the structural parameters must be known. In a typical undoped AlGaN/GaN MODFET, the GaN buffer layer is unintentionally doped with a level of ND 1016 cm3, and the thickness of this buffer layer spans 1–3 mm. The barrier thickness spans 10–20 nm. Because the structure is not intentionally doped,
2.12 Polarization Effects
Ec
AlGaN
GaN
di
qφB
Ec ΔEc
-qVG
E1 E0
EF
EFi
dd
x=0 Figure 2.46 Conduction band edge of what is generically referred to as a modulation-doped structure based on the AlGaN/GaN system. The origin of the interface charge is aggregate due to polarization and free carriers. In Ga-polarity samples, AlGaN grown on GaN produces polarization charge (due to spontaneous polarization and piezoelectric polarization
x because AlGaN is under tensile strain), causing accumulation of electrons at the interface in addition to any electron donated by the any donor impurities, intentional or unintentional, in AlGaN. The diagram is shown for a doped AlGaN, the doped portion of which is indicated by dd and the undoped part is indicated by di.
it is assumed that unintentional impurities in the form of contamination or native defects, such as O, Si, VN, are responsible for supplying the free carriers, which is measured with the accompanying assumption that it is sufficient to completely screen the positive polarization charge, þr/q, at the interface [476]. The influence of doping specifically in the case of AlGaN/GaN has been treated, for example, by Chu et al. [475]. In short, the sheet charge at the interface of such an undoped structure is dominated by the polarization-induced charge. On Ga-polarity surfaces, this charge increases with increasing AlN molar fraction in the barrier because both the piezo and spontaneous components of the polarization charge increase, assuming of course that the barrier is coherently strained. The conduction band and electron concentration profiles for an undoped Ga-face Al0.3Ga0.7N/GaN (30 nm/ 2 mm) heterostructure with an Ni Schottky contact on top are shown in Figure 2.47a. An electric field strength of about 0.4 MV cm1 in the barrier and a sheet electron concentration of 1.2 · 1013 cm2 are induced by polarization. The bound sheet density and 2DEG sheet carrier density induced by polarization in heterostructures (identical to that shown in Figure 2.47a with the exception that the alloy compositions in the barrier has been changed) are shown in Figure 2.47b and c with solid lines for both the linear and nonlinear polarization cases. The underlying assumptions are that the GaN buffer layer is relaxed, the barriers are coherently strained, and the physical parameters of importance (Cij, eij, and Psp) linearly scale from binaries to ternaries.
j273
j 2 Electronic Band Structure and Polarization Effects
274
The estimated error depicted by the gray area in Figure 2.47c is primarily due to uncertainties in the barrier thickness and the conduction band offsets. The sheet carrier concentrations of 2DEGs confined in an AlxGa1xN/GaN heterostructure for x ¼ 0.5 have been measured by C–V profiling using Ti/Al ohmic and Ni Schottky contacts [84]. The highest measured and calculated sheet carrier concentration for 2.0
+s E = 0.41 MV cm–1
1.5
2DEG
1.0
Energy (eV)
13
Ns = 1.2 10 cm –2
eΦ
1.23 eV
0.5
CB
0
Δ
0.30 eV
–0.5 (a)
0
10
ΔE
EF
c
0.38 eV
E0 = 0.17 eV
30 40 20 Depth (nm)
50
Figure 2.47 (a) Self-consistent calculation of the Schr€ odinger–Poisson equations for the conduction band edge and the electron density profile for an undoped Ga-polarity Al0.3Ga0.7N (30 nm)/GaN (2 mm) single-interface heterostructure. Also shown are the Schottky metal on the surface and the polarizationinduced surface and interface charges. (b) The polarization-induced bound interface charge density as a function of the alloy composition in the barrier for Ga-polarity AlxGa1xN/GaN heterostructures for the case of relaxed buffer layer and coherently strained barrier layer. The upper solid line for sheet charge corresponds to the case of linear interpolation of physical parameters (Cij, eij, and Psp) from the binary compounds. The lower solid line for the sheet charge, on the contrary, corresponds to the case of nonlinear extension of polarization from binary end points. The dashed lines depict the lagging sheet density with increasing molar fraction, x, due to partial relaxation, which is
60 accounted for by the measured degree of barrier relaxation into account. (c) 2DEG sheet carrier concentrations ns in unintentionally doped Gaface AlxGa1xN/GaN heterostructures (with 30 nm AlGaN and 2000 nm GaN – the structure referred to in (b)) as obtained by C–V profiling versus alloy composition of the barrier (open symbols), compared with the theoretical predictions for the bound interface charge r/q calculated using (i) a linear interpolation between the macroscopic polarizations of the binary compounds (upper solid line) and (ii) the nonlinear piezoelectric and spontaneous polarization (lower solid line). The sheet charge of the 2DEGs is then calculated as in (ii), considering in addition the depletion by a Ni Schottky contact (dotted line). The dashed lines account for the experimentally observed strain relaxation of the barrier for x > about 0.4. Courtesy of O. Ambacher. The related details can be found in Refs [476,84,479].
2.12 Polarization Effects
1014
Bound interface density σ/q(cm–2)
Linear interpolation
Coherently strained
1013
Nonlinear
1012
0
0.4
0.2
(b)
0.6
0.8
1.0
Molar fraction, x
3.5
Sheet carrier concentration ns (1013 cm–2)
Relaxed
Ga-face Alx Ga1–x N/GaN (30/2000 nm)
3.0
Pseudomorphic
2.5 2.0 Partly relaxed
Linear interpolation
1.5 1.0 0.5
Nonlinear
0 (c)
0
0.1
0.3 0.4 0.2 Molar fraction, x
0.5
0.6
Figure 2.47 (Continued )
AlxGa1xN/GaN heterostructures is 2 · 1013 cm2 for x ¼ 0.37, as higher AlN molar fractions in the alloy for 30 nm barrier causes a partial relaxation, lowering the piezo contribution [476]. The extrapolation, indicated by dashed lines in Figure 2.47c, represents the case when the mole fraction is sufficiently large to cause some degree of relaxation. As expected, the sheet density decreases nonlinearly with reducing mole fraction for the 30 nm barrier modeled. Among the two curve that bend over, the upper one is the total calculated polarization charge and the lower one is the
j275
j 2 Electronic Band Structure and Polarization Effects
276
calculated screening charge using the one-dimensional Schr€ odinger–Poisson equations. When full or partial relaxation occurs, the piezo component of the polarization charge is reduced while the spontaneous component remains unchanged. The experimental data along with the error bars are shown in open circles. Overall, there is then a reduction of the charge. Likewise, the screening sheet carrier concentration is also reduced. The obvious conclusion that can be made is that a much better agreement is attained between calculations and experiments when the nonlinear polarization is used. The case of the nonlinear polarization is stronger in the 2DEG case than it is in the quantum well case due to the direct nature of the measurement. Ridley et al. [477] presented an empirical expression relating the sheet carrier concentration for a nominally undoped AlGaN/GaN single-interface heterostructure having barrier thicknesses of greater than 15 nm and AlN molar fractions over 6%. ns ðxÞ ¼ ½ 0:169 þ 2:61x þ 4:50x 2 1013 cm 2 :
ð2:116Þ
Naturally, the sheet charge becomes affected by parameters other than the AlN molar fraction when the barriers are made much thinner, as can be deduced from Equation 2.115. Using an expression similar to Equation 2.115, Ambacher et al. [84] plotted the dependence of the sheet density of barrier thickness for several AlN molar fractions, namely, x ¼ 0.15, 0.30, and 0.45. A priori, it is clear that beyond a certain thickness of the barrier, the density should saturate even if coherent strain prevails, as shown Figure 2.48. Ambacher et al. [84] also measured the sheet density for a set of samples with x ¼ 0.3 by C–V profiling for barrier thicknesses spanning the range of 1–50 nm, and presented the data together with those from other reports. 2.12.4.2 Ga-Polarity Single AlxIn1xN/GaN Interface This system, if for nothing else, is of importance because the entire structure can be lattice matched, leading to vanishing piezoelectric polarization and thus allowing one to investigate and probe only the spontaneous polarization in undoped structures. Doing so would enhance our confidence in spontaneous polarization calculations, as there are fewer parameters to be determined and thus less uncertainty. For high concentrations of Al (for x > 0.6), the bandgap of AlxIn1xN is larger than that of GaN, and if used in conjunction with Ga-polarity GaN, a 2DEG would result due to polarization or doping in AlxIn1xN or both. For lattice-matched conditions where the x-value is about 0.82, the bandgap of the AlxIn1xN alloy is about 4.7 eV to produce sufficiently large band discontinuity with GaN, imperative for MODFET structures. Ambacher et al. [84] prepared coherently strained AlxIn1xN (50 nm)/ GaN (540 nm) single-interface heterostructures with Al concentrations between 0.78 and 0.88, and applied X-ray diffraction to determine the structural state of GaN, meaning relaxed or strained (GaN grown on sapphire is typically under compressive strain due to thermal mismatch). Through the in-plane and out-of-plane strain (which can both be measured using reciprocal space X-ray diffraction mapping) or the lattice constants, the mole fraction of the barrier and, through repeated attempts, the mole fraction giving rise to lattice matching conditions can be determined. Ambacher et al. [84] determined the lattice matching composition of AlxIn1xN to
2.12 Polarization Effects
10
14
Sheet carrier concentration ns (cm–2)
x = 0.45
10
13
= 0.3
= 0.15
1012
Ga-face AlGaN/GaN
10
11
0
10
20
30
40
50
60
d AlGaN (nm) Figure 2.48 Barrier thickness, d, dependence of sheet density in nominally undoped and coherently strained AlxGa1xN/GaN heterointerfaces for x ¼ 0.15, 0.30, and 0.45 (solid lines). Experimental data available for x ¼ 0.3 measured by C–V profiling for barrier thicknesses spanning 1 and 50 nm, representing an aggregate from several reports, are shown [84].
GaN to be x ¼ (0.83 0.01), as shown in Figure 2.49, which is compared with other experimental values of 0.82–0.83 [451]. The weak compressive strain in GaN was determined to be e ¼ 1.9 · 103, which would result in a piezoelectric polarization of 1.5 · 103 Cm2. The residual strain in GaN would lead to a bound sheet density of only 1012 cm2, which is much smaller than the 1013 cm2 electron sheet density. This implies explicitly that interface charge is dominated by the gradient in the spontaneous polarization across the GaN/AlxIn1xN interface, as show in Figure 2.50, which shows the calculated bound sheet density induced by a gradient in spontaneous polarization (upper dashed line) and strain-induced piezoelectric (lower solid line) for a range of compositions near the lattice matching conditions and the resultant 2DEG sheet carrier density (upper solid line). Also shown are the experimental 2DEG densities obtained by C–V measurements. The polarization and thus the sheet density have been obtained by a linear interpolation of the physical parameters (Cij, eij, and PSP) of the binary compounds but by taking the nonlinearity of Ppz and PSP into account. The calculations confirm the obvious in that the measured high electron sheet density can be accounted for by the spontaneous polarization charge.
j277
j 2 Electronic Band Structure and Polarization Effects
278
6
cAlInN (x) 4 aAlInN (x)
Strain, e (10–3)
2 Tensile strain 0 Compressive strain –2
–4
Lattice matched –6 0.75
0.85 0.80 Molar fraction, x
Figure 2.49 Strain in AlxIn1xN barrier along the c-axis and on the basal plane as determined from the measured lattice constants, cAlxIn1xN and cAlxIn1xN by high-resolution X-ray diffraction reciprocal mapping. AlxIn1xN for x ¼ 0.83 can be
0.90 grown lattice matched to GaN, leading to vanishing piezoelectric polarization due to misfit. However, residual strain, due to thermal mismatch between GaN and the substrate, can still induce piezoelectric polarization [84].
Note that for Ga-polarity samples and AlN compositions less than the lattice match conditions, the AlxIn1xN barrier layer is under compressive strain in-plane and therefore the piezo-induced polarization is opposite in sign to that for spontaneous polarization. When the AlN molar fraction exceeds that for matching condition, the in-plane strain is tensile in the case of which the two polarization charges are additive. Near the lattice matching conditions, the calculated sheet density is about 2.95 · 1013 cm2 and falls above the measured values. The reason for this discrepancy could be related to samples themselves, either in the form of defects near the interface, surface adsorbates, and/or inhomogeneities caused by In, and underestimation of the bowing parameters associated with the polarization charge. 2.12.5 Polarization in Quantum Wells
For multiple interface heterostructures, the sheet carrier density and barrier thickness, as well as the width quantum wells, are of interest because the total potential drop across the structure is directly proportional to the product of polarization field and well width in constant field approximation if free-carrier screening is neglected.
2.12 Polarization Effects
4
s sp e (P )
2
n (x) S
1
0
–1
–2 0.75
s z (P ) E
0.80
Lattice matched
Sheet carrier concentration (1013cm–2)
3
0.85
0.90
Molar fraction, x Figure 2.50 Compositional dependence of the spontaneous polarization (dashed line) and piezoelectric polarization (lower solid line) induced charge density reduced to interface sheet charge in an AlxIn1xN/GaN heterojunction using the nonlinear interpolation for polarization discussed in the text. For AlN molar fractions below the lattice matching
conditions, the two polarization charges oppose and above the lattice matching value, they add. The calculated sheet electron concentration (upper solid line) and the measured sheet electron density are also shown, the latter of which has been obtained by both Hall effect (solid symbols) and C–V profiling (open symbols) [84].
This issue goes to the heart of lasers, particularly in the stages as the gain is built up, in that the well widths greater than about 5 nm are not used. At higher injection levels, the polarization-induced field is screened pretty much [458]. To illustrate the point, the conduction band profile of a 10 nm GaN/In0.2Ga0.8N quantum well as calculated by Della Salla et al. [461] is shown in Figure 2.51 for several sheet densities. Even at a substantial sheet density of n2D ¼ 5 · 1012 cm2, a nearly uniform electrostatic field of strength 2.5 MV cm1 is still present in the well. One needs to increase n2D to 5 · 1013 cm2 before recovering the quasi-field-free shape of the quantum well that is needed for lasers. This is achieved much earlier in thinner quantum wells. In wells, the electrons and holes are indeed spatially separated by the polarization field, but the free carrier induced field acts to cancel the polarization field, which is efficient for high sheet densities. This reestablishes the efficient electron–hole recombination. Time-resolved PL (TRPL) is a wonderful way of seeing the effect of polarization. Due to band bending induced by polarization, when the optical excitation pulse is turned off, the free-carrier density goes down and so does the screening. As a result, a
j279
j 2 Electronic Band Structure and Polarization Effects
280
[0 0 0 1]
Conduction band edge energy (eV)
4.0
GaN/In0.2Ga0.8N
3.5 3.0 2.5 2.0
n2 D = 1013cm–2 0.5 1.0 2.0 5.0
1.5 1.0 0.5
0
10
20 30 Depth (nm)
40
50
Figure 2.51 Conduction band profile of a 10 nm GaN/In0.2Ga0.8N quantum well for various levels of free carriers ranging from 5 · 1012 cm2 to 5 · 1013 cm2 either present by doping or injection. The effect of polarization is nearly all but wiped out for the largest sheet electron concentration [461].
red shift accompanied with increased carrier lifetime due to lowering of the overlap between the electron and hole wave function occurs, as they are pushed to the opposing end of each well [478]. For details, the reader is referred to Volume 2, Chapter 5. 2.12.5.1 Nonlinear Polarization in Quantum Wells Asinthecaseofbulk binariesand alloys, the polarization issue inheterostructuresneeds a revisit to consider the nonlinearities discussed above in the context of binary and alloy nitride bulk layers. Assuming that the ternary nitride alloys have random microscopic structure, the spontaneous polarization of random ternary nitride alloys, in unit C m2, has been expressed by Fiorentini et al. [479] to the second order in the composition parameter x, as expressed in Equation 2.95 but repeated here for convenience. sp
PAlx Ga1 x N ¼ 0:09x 0:034ð1 xÞ þ 0:019xð1 xÞ; sp
P Inx Ga1 x N ¼ 0:042x 0:034ð1 xÞ þ 0:038xð1 xÞ; sp P Alx In1 x N
ð2:117Þ
¼ 0:09x 0:042ð1 xÞ þ 0:071xð1 xÞ:
The first two terms in all three equations indicated in Equation 2.117 are the usual linear interpolation between the binary compounds represented by Equation 2.86. However, the third term is the so-called bowing term encompassing the quadratic nonlinearity as in the case of the bandgap bowing parameter discussed in Section 1.5. The coefficient of the third term is the bowing parameter discussed in conjunction with Equations 2.95 and 2.92. Higher order terms are neglected, but their effect was estimated to be smaller than 10% in the worst case being the AlInN alloy [453].
2.12 Polarization Effects
For piezoelectric polarization, it was shown [453] in conjunction with Equation 2.96 that Vegards law holds provided that the appreciable nonlinearity of the bulk piezopolarization of the component binaries as a function of strain is accounted for. Doing so has led, in general, to a good agreement with experimental results [480]. For a model AlxGa1xN alloy, the piezoelectric polarization can be related to the binary end points using the Vegards law, but recognizing that the terms for the bulk binaries must contain the nonlinear terms as indicated in Equation 2.96. Such polarizations can be expressed accurately and compactly (in units of C m2) as pe
PAlN ¼ 1:808e þ 5:624e2 pe
PAlN ¼ 1:808e 7:888e2 pe PGaN ¼ pe PInN ¼
for e < 0; for e > 0;
0:918e þ 9:541e ; 2
ð2:118Þ
1:373e þ 7:559e2 ;
as a function of the basal strain of the alloy layer in question, with a(x) and asubs or a0 as the lattice constants of the unstrained alloy at composition x and of the relaxed buffer layer or the substrate. eðxÞ ¼ ½asubst aðxÞ=aðxÞ:
ð2:119Þ
In the case of pseudomorphic growth on GaN buffer layers, basal strain e can be calculated directly from the lattice constants, which are found to follow Vegards law as a function of composition (depending on the lattice constants used): aAlx Ga1 x N ðxÞ ¼ aGaN xðaGaN aAlN Þ ¼ 0:31986 0:00891x nm; aInx Ga1 x N ðxÞ ¼ aGaN þ xðaInN aGaN Þ ¼ 0:31986 þ 0:03862x nm; aAlx In1 x N ðxÞ ¼ aInN xðaInN aAlN Þ ¼ 0:35848 0:04753x nm:
ð2:120Þ
The combination of Equations 2.118–2.120 provide a convenient way of determining the polarization dependence on basal strain. The coefficients in Equation 2.118 are related (not equal) to piezoelectric constants and come about from the ab initio calculations [453]. The polarization charge values calculated using Equations 2.118–2.120 for heterostructures can be used together with a self-consistent Schr€ odinger–Poisson solver based, for example, on effective mass theory (not as accurate but efficient) or tight binding (more accurate but computation intensive) to determine field and charge distribution in the entire heterostructure as well as the effect of free carriers [448,460]. Typically, for a given structure or set of structures, two classes of observable interest can be simulated and compared with experiment. Experimental confirmation for the nonlinear theory can be garnered from quantum wells by probing the red shift for a given sized well and its dependence on excitation, and measuring the magnitude and gate voltage dependence of the interface sheet carrier concentration at an AlGaN–GaN interface with the aid of the one-dimensional self-consistent simultaneous solution of Schr€ odinger and Poisson equations in the effective mass approximation. Additional confidence can be estimated by repeating these experiments for these structures where, for example,
j281
j 2 Electronic Band Structure and Polarization Effects
282
only the barrier mole fraction is changed [479]. Fortunately, the results are very sensitive to the values of the polarization in the different layers, and therefore the built-in field or the screening charge. The particular bowing parameters in the polarization expression yielding the best agreement with experiment can be assumed to be valid. The point should be made that the C–V measurements are probably the most direct and optical shifts the least direct. The C–V data and the simulations based on the theories described above agree very well [479]. The attempt here is to show that inclusion of the nonlinear effects in calculations lead to a better agreement between the experiments and theory in quantum wells and AlGaN/GaN heterointerfaces. In terms of the quantum confined stark effect (QCSE), the photoluminescence energy of AlxGa1xN/GaN QWs with well thicknesses of 1,1.5, 2.5, 4, 6, and 8 nm, and with barrier alloy composition of x ¼ 0.08, 0.13, 0.17, and 0.27, respectively, reported in Ref. [481] were used to determine the polarization-induced electric field. The field determined by measuring red shift from the PL transition energies versus the mole fraction in the barrier of quantum wells as well as calculated values from the polarization charge determined using the linear and the nonlinear approach are shown in Figure 2.52. In both types of calculations, the electric field in the GaN QWs was obtained by solving self-consistently the coupled Schr€ odinger–Poisson equations including polarization-induced interface charges. In the first approach the polarization of the AlGaN barriers that are assumed under tensile coherent strain, as the bulk buffer layers were GaN, is determined by a linear interpolation between the elastic 2.5 Linear interpolation
Electric field (MV cm–1)
2.0
Nonlinear approach
1.5
1.0
Ga-polar AlxGa1–xN/GaN QWs
0.5
0 0
0.1 0.2 0.3 0.4 0.5 0.6
AIxGa1–x N molar fraction, x Figure 2.52 Polarization-induced electric fields in high resistivity and Ga-face AlxGa1xN/GaN MQWs versus the alloy composition of the barrier. The upper line represents the field predicted by linear interpolation of binary compound polarization. The lower line represents the calculated field using the nonlinear polarization concept. Open circles
are deduced, via self-consistent effective mass calculations, from the polarization-induced Stark shift of excitonic recombination reported in Ref. [481] via a self-consistent effective mass calculation. In addition, the experimental data published by Langer et al. [482] and Kim et al. [483] are also shown in filled circles [479].
2.12 Polarization Effects
and piezoelectric constants and the spontaneous polarizations of the binary compounds AlN and GaN. In the second approach, the nonlinearity of the polarization of AlGaN as described by Equations 2.117 and 2.118 was taken into consideration. Clearly, the electric field calculated including the nonlinearity in polarization versus barrier alloy composition does much better at reproducing the experimental data taken from Ref. [481]. In addition, the experimental data published by Langer et al. [482] and Kim et al. [483] are also shown in filled circles. A more convincing arguments for nonlinearity in polarization in the context of polarization charge and resultant red shift in the spectra is made by pressuredependent measurements of the transition energies as presented by Vaschenko et al. [484]. They considered a quantum well system with background unintentional dopants and excitation-induced carriers in the case of which the field deviates from that indicated in Equation 2.105. total ðPtotal Vs W PB Þ þ r þ LW þ LB LW eB þ LB eW 1 d LW 1 d LB qN D þ ; 2 2 eW eB
E W ¼ LB
ð2:121Þ
where eW,B is the permittivity of the GaN wells and AlxGa1xN barriers, respectively, (assumed to be independent of pressure in this work), LW,B are the cumulative thicknesses of the wells and the barriers in the MQW structure, r is the total twodimensional photogenerated charge density in the wells, Vs is the surface barrier potential determined as in Ref. [485], ND ¼ 1017 cm3 is the assumed background doping concentration based on bulk GaN layers grown by MBE, which was used to produce the structure, and d is the distance from the barrier–buffer interface to the well where the field is calculated. The first term in Equation 2.121 is similar to Equation 2.105 but the former having an additional charge r (the total two-dimensional photogenerated charge) in the first term in addition to second and third terms. The second term represents the field due to the surface barrier potential which neglects the effect of dielectric discontinuity between AlGaN and GaN, and the third terms represents the effect of dopant on the field. If Vs, r, and ND were made to go to zero, Equation 2.121 reduces to Equation 2.105. To find PW PB as a function of pressure, the experimentally measured PL peak energy variation, which was assumed to be representing the n ¼ 1 electron and heavyhole transitions, with well width was fitted to the calculated dependence of the same transition, The nonlinear behavior in this case is revealed by determining the PW PB (barrier–well polarization difference) as a function of applied hydrostatic pressure. Figure 2.53 shows the fit to the measured PL peak energies in samples in which the AlN mole fraction varied as 0.2, 0.5, and 0.8 at a pressure of 5 GPa. The solid lines are calculations with the polarization-induced field, whereas the dashed ones are without such field. The experimental data are shown in open symbols. The good agreement between the calculations where PW PB was treated as fitting parameter and the experiment underscores the crucial nature of the built-in electric field in the determination of the well width dependence of the ground-state energy.
j283
j 2 Electronic Band Structure and Polarization Effects
284
4.2 4.0 3.8
Photon energy (eV)
3.6 3.4 3.2 3.0 2.8 2.6
x = 0.2 x = 0.5
2.4
x = 0.8
2.2 1.0
1.5
2.0
2.5
3.0
3.5
4.0
4.5
5.0
5.5
Well width (nm) Figure 2.53 Well width dependence of the PL peak energy at 5 GPa. Open symbols correspond to the experimental points and the solid lines depict the fits to the experimental data obtained with PW PB as an adjustable parameter. The dashed lines correspond to the e1 hh1 transitions calculated assuming zero field [484].
The polarization difference PW PB and resulting electric field in the 2.9 nm wells for the x ¼ 0.5 MQW sample, as determined by the procedure described above, is presented in Figure 2.54. For the x ¼ 05, and the others that are not shown here, PW PB noticeably increases with pressure, resulting in an increase of the built-in field in accordance with the mole fraction in the barrier reaching 0.76 MV cm1 in the samples with x ¼ 0.5 at 8 GPa. Figure 2.54 also shows the calculated results of the pressure dependence of PW PB in that the solid lines correspond to PW PB calculated with the linear polarization [448]. This linear polarization model overestimates the values of PW PB at atmospheric pressure and underestimates the pressure dependence as compared to experiments. The dashed-dotted lines represent calculations where only the volume-conserving strain dependence of the GaN and AlN piezoelectric coefficients is taken into account [456]. Clearly, this model agrees better with the experimental data than the linear polarization model that is consistent with the conclusions of Perlin et al. [486], where the pressure dependence of PL in GaN/Al0.13Ga0.87N QWs was found to be adequately described by the volumeconserving strain dependence of the piezoelectric coefficients. Lastly, the dashed line
2.12 Polarization Effects
0.058
3.6
0.056
0.052
3.2
0.050 3.0 0.048 2.8
0.046 0.044
Built in electric field, MV/cm
Polarization difference (Pw-PB), (cm–2)
3.4 0.054
2.6
0.042 2.4
0.040
T = 35 K x = 0.5
0.038 0
2
4
6
8
10
2.2
Pressure (GPa) Figure 2.54 Pressure dependence of PW PB and corresponding electric field in 2.9 nm AlxGa1xN/GaN MQWs with x ¼ 0.5. The open circles represent the calculated points obtained from the fit to the PL data as shown in Figure 2.53. The solid line corresponds to PW PB calculated with the linear polarization that overestimates the data obtained from experiments at zero pressures and underestimates the slope of the pressure dependence of polarization. The dashed-dotted
line shows the calculations, where only the volume-conserving strain dependence of the GaN and AlN piezoelectric coefficients is considered, which overestimates the differential polarization but does very well in terms of the pressure dependence of the differential polarization. Lastly, the dashed line shows the result of calculations using the nonlinear polarization behavior with the distinctly best overall fit to the experimentally determined values. The dotted line is a guide to eye [484].
shows the result of calculations where the nonlinear behavior of both the spontaneous and piezoelectric polarizations had been taken into account using the results of Bernardini and Fiorentini [220,221]. Only the change in piezoelectric polarization due to hydrostatic compression of the ideal crystal and that due to the increase in the internal parameter u with pressure is considered here [86]. The spontaneous polarization bowing was included at p ¼ 0 [220,221]. Although this is only an approximation of the theory developed in Refs [220,221], this model also predicts the slope of the pressure dependence of PW PB significantly larger than that of the linear model. In short, the added value of pressure dependence of differential polarization causes this parameter to be very sensitive to which polarization picture is employed for the transition energies in AlGaN/GaN MQWs to the point that one can clearly state that the nonlinear dependence of the piezoelectric polarization in GaN and AlN unequivocally predicts the experimental data best.
j285
j 2 Electronic Band Structure and Polarization Effects
286
2.12.5.2 InGaN/GaN Quantum Wells Relatively better transport properties of InN as compared to GaN and possible translation of the same to InxGa1xN are one of the draws for considering InxGa1xN channels [229] for MODFETs, discussed briefly in Volume 3, Chapter 3. However, experimental results have so far been disappointing. On the contrary, InxGa1xN proved to be the magical material for its high radiative recombination efficiency for optical emitters. As detailed in Volume 3, Chapters 1 and 2, all the highperformance optical emitters feature InxGa1xN in one form or another in their active regions. Unlike the FETcases where GaN is used as the active layer and also the buffer layer where the buffer layer is high resistivity, the conductivity of InxGa1xN is high, which rules out its use as the buffer layer in an InxGa1xN channel FET. Consequently, even if InxGa1xN were to be considered for the active layer, it must be grown on GaN buffer layers. As for the barrier layer, it can be made of GaN or some composition of AlxGa1xN. In any case, the InxGa1xN layer would be straddled by large bandgap material on both sides with the resultant single quantum well structure. Consequently, polarization in single-well InxGa1xN quantum wells must be considered as discussed here. In a single quantum well structure, in addition to interface sheet charge that contains information on polarization and can be measured by electrical means, these structures also offer an additional avenue to probe the polarization-induced charge through quantum-confined Stark shift. While the optical properties of InxGa1xN quantum wells are discussed in great detail in Volume 2, Chapter 5, including the associated QCSE and any Stokes shift, a discussion of single quantum well for the sole purpose of polarization effects is provided here for completeness. Ambacher et al. [84] reviewed the issue in conjunction with nominally undoped, n-type GaN/In0.13Ga0.87N/GaN structures with Ga-polarity, where the width of the quantum well dInx Ga1 x N varied between sp 0.9 and 54 nm. The spontaneous polarization of the InGaN layer, P In0:13 Ga0:87 N , 2 is 0.031 Cm and points in the ½0 0 0 1 direction and the piezoelectric polarization is calculated to be 0.016 Cm 2, which is antiparallel to the spontaneous polarization because the InxGa1xN quantum well is under compressive strain. The bound charge at the GaN–InGaN interface near the surface is positive and that at the lower interface is negative for Ga-polarity sample due the compressive strain that InxGa1xN is under. This implies that the electron accumulation caused by screening would occur at the interface near the surface in n-type samples. If p-type samples were considered, a hole accumulation would occur at the other interface. The total polarization-induced interface sheet density is then given by sp
pz
sp
pz
ðPGaN þ P GaN Þ ðPInx Ga1 x N þ PInx Ga1 x N Þ:
ð2:122Þ
Recognizing that GaN is relaxed in this case, the samples are grown on GaN buffer layers that are presumed to be relaxed and any residual strain is neglected. This leads pz to PGaN ¼ 0, and substituting the numerical values, one gets for the total polarization ð 0:034 þ 0Þ ð 0:031 þ 0:016Þ Cm 2 ¼ q1:18 1013 Ccm 2 ;
or 1:18 1013 electrons cm 2 :
2.12 Polarization Effects
InGaN SQW x = 0.13
d = 54 nm
ns = 5 × 1012 cm–2
22
10
26 nm
Electron concentration (cm–3)
2.0 ×
× 10
1012
10
10 16 12 nm 2.6 × 1011
18
16
10
10
20
1018
2
20
10
10
10
22
1014
Electron concentration (cm-3)
10
10 12
4.3 nm 3.6 × 1011
+s –s
14
12
10
1
10
2
2
× 10
10
3
10 4
Depth (nm) Figure 2.55 Electron concentration profiles unintentionally doped, n-type GaN/In0.13Ga0.87N/GaN QWs having well widths of 4.3, 12, 26, and 54 nm, respectively, as deduced from C–V depth profiling [84].
The GaN/In0.13Ga0.87N/GaN the heterostructures with GaN top layer and In0.13Ga0.87N quantum well with thicknesses of 130 and 20 nm, respectively, have been examined by Ambacher et al. [84], and their electron profiles, as determined by C–V measurements, are shown in Figure 2.55. For well widths less than 4 nm, any electron accumulation was not observed, which implies failure to screen the bound charge fully. In fact, as the quantum well thickness was increased from 4.3 to 54 nm, the sheet carrier concentration increases from ns ¼ 3.6 · 1010 to 5 · 1012 cm2. The increase in 2DEG sheet carrier concentration with well width, which follows a 2.5 power of well width, is much faster than the well width (the volume), which may be attributed to the system not being in equilibrium in terms of screening for thinner wells. Overall, the sheet density is about half of what is expected even in thick quantum wells. The discrepancy between predictions and experiments raises an interesting question if the polarization-induced charge is fully screened by electrons and ionized donors. It is clear that an electric field causes band bending, the total extent of which scales with thickness, which in turn causes a red shift in radiative recombination transition energy. Assuming that the polarization-induced charge at the free surface and the GaN–substrate interface are fully screened, an electric field in the constant
j287
j 2 Electronic Band Structure and Polarization Effects
288
field approximation forms, which can be expressed as E well ¼
ptotal e0 ðeInGaN r
1Þ
¼
total P total GaN P InGaN : InGaN e0 ðer 1Þ
ð2:123Þ
The terms here have their usual meanings. The electric field has been predicted to be 2.2 MV cm1 for In0.13Ga0.87N QWs, which causes band bending to the extent of turning otherwise square quantum well potential distribution to a triangular distribution in the constant field approximation. The resultant red shift in transition energy, Stark shift, can be related to the polarization charge through the field as g
E InGaN E energy ¼ qE well W þ
9phqE W pffiffiffi 8 2
2=3
1 mnInGaN
þ
1 p
mInGaN
1=3 ; ð2:124Þ
where mnInGaN
p and mInGaN
are the electron and hole effective masses in InGaN. For an InN mole fraction of 0.13, these values will not be much different from those for GaN. W represents the InGaN well width. This would provide an additional means for determining the polarization charge with an all-optical method such as photoluminescence. Extreme care must then be exercised to be certain about the InN composition and its bandgap as well as making sure that the optical transitions observed are accurately related to the band edge, meaning their nature must be known. The experiments must also be conducted at low injection levels as to not screen the charge, and high injection levels to screen the charge. If screening is full and low injection levels is truly low, then the difference in energy between the very low and very high injection levels would represent the red shift due to the polarization charge, as was done by Ambacher et al. [84] who also determined the bandgap of InGaN by reflection measurements such as spectroscopic ellipsometry and room-temperature PL using an excitation energy of 3.41 eV, only absorbed by the InGaN well. The QW region was pumped with high-intensity light to generate a high density of electron–hole pairs to fully screen the polarization-induced charge, thus the field. For In0.13Ga0.87N QWs with widths of greater than 26 nm, a bandgap of 2.902 0.012 eV was measured, which is consistent with the value for bulk (2.946 eV). The spectroscopic ellipsometry and PL data show a monotonic increase in energy, reaching a value of 3.21 eV for a well width of 0.9 nm, as the Stark shift diminishes with decrease in well width while the quantum confinement increases, as depicted in Figure 2.56. The PL spectra measured by 3.81 eVexcitation with a pump power density of 103 Wcm2 yield a PL peak position that is increasingly shifted to lower energies if the QW width is increased from 0.9 to 5.3 nm (Stark effect). Knowledge of this red shift in the PL peak position together with Equation 2.124 allows one to calculate the electric field strength as 0.83 MV cm1. This field corresponds to a polarization-induced bound charge density of about 5 · 1012 cm2 and compares well with the data of 0.62 MV cm1 and charge of 3.7 · 1012 cm2 for In0.12Ga0.88N/GaN MQWs by Wetzel et al. [487]. As can be deduced from Figure 2.56, the polarization-induced red shift (Stark shift) is not notable for QWs wider than 26 nm where the bandgap of InGaN and the PL
2.12 Polarization Effects
3.4
3.2 g
Energy (eV)
3.0
2.8
E InGaN
660 meV
E 0 e, 0 h
2.6 5 nm
2.4
2.2 10 –1
10 1 10 0 InGaN SQW thickness (nm)
Figure 2.56 The dependence of the effective bandgap (without confinement effects) of In0.13Ga0.87N used in QWs (solid line) and the energy of the radiative recombination in QWs assumed to be between electron ground state (E0e) and hole ground state (E0h) on well width. The effective bandgap is determined by
10 2 spectroscopic ellipsometry and RT PL with 3.41 eV excitation (high and low intensity for PL to account for the Stark shift). Also noted in the figure is that a change of 5 nm in the well thickness leads to a change of 660 meV change in the QW emission energy [84].
peak position agree within 40 meV. It also appears that polarization-induced charged may either be fully screened, negating the constant field approximation throughout the quantum well, or be the notorious In fluctuations discussed in details in Chapter 3 in terms of growth, Chapter 4 in terms of its effects on optical transitions, and Volume 3, Chapters 1 and 2 in terms of its effects on optical emitters. 2.12.6 Effect of Dislocations on Piezoelectric Polarization
To fully consider the true nature of III-Nitride epilayers, the effect of dislocations on piezoelectric polarization must be included. The case does hardly needs to be made owing to the high density of dislocation (edge, screw, and mixed) that alters the strain distribution and thus the piezoelectric polarization. In this vein, Shi et al. [488] calculated the piezoelectric polarization around c-oriented screw and edge dislocations in Wz GaN and found that polarization around screw dislocations (having Burgers vector h0 0 0 1i) has no z-component, which is similar to a magnetic field around a conducting line, so there is no charge induced either at the core or around the screw dislocation. In the case of edge dislocations (having Burgers vector 1/3h1 2 0i), in which the strain field is compressive on one side and tensile on
j289
j 2 Electronic Band Structure and Polarization Effects
290
(0 0 0 1) c-plane
-
Qs
+
Pe Ps bs be
+
-
Figure 2.57 Schematic diagram illustrating the dislocation geometry, associated polarizations and charge densities [488].
the other, calculations show that piezoelectric polarization has only the z-component and its divergence vanishes at the core and around the dislocation. But at the interface, it results in an effective surface charge for the difference in polarization across the interface, as shown in Figure 2.57. It was estimated that for a c-oriented edge dislocation the charge density could reach 1011 e cm2 within 0.1 mm of the core. On the experimental side, electron holography was applied to investigate the built-in field caused by polarization [489] and charge distribution at the dislocation [490]. In undoped GaN, the holography results confirm that all dislocations are negatively charged and the line charge densities are calculated as 1 and 0.3 for screw and edge dislocations, respectively. Cai and Ponce [491] argued that screw dislocations always have relatively higher charge density. The electrical activity associated with extended defects is discussed in considerable detail in Section 4.1.6. We should be cognizant of the fact that not only extended defects in general cause local strain and therefore inhomogeneous strain but they also attract and often trap impurities, point defects, and free charge. Moreover, the strain component affects the charge through the piezoelectric component. 2.12.7 Thermal Mismatch Induced Strain
Having made the case that strain induced by both lattice mismatch and also by thermal mismatch plays a profound role in polarization, let us discuss the thermal mismatch case in some detail because the lattice mismatch component got good deal of coverage already. Within the realm of thermal mismatch, the dominant component is that introduced by the nonnative substrates used. The substrates used are many in number and kind, but the dominant ones are sapphire and SiC both of which introduce sizable thermal mismatch. An incomplete list of
2.12 Polarization Effects
substrates used, in addition to the ones cited, include other substrates, GaN, AlN, g-LiAlO2, b-LiGaO2, NdGaO3, Si, GaAs, MgO, ZnO, ScAlMgO4, MgAl2O4, and (La, Sr)(Al,Ta)O3. For a complete discussion refer to Chapter 4 and for a complete compilation refer to Ref. [492]. In a nut shell, the thermal stress is relatively small if GaN layers are grown on AlN, SiC, ScAlMgO4, Si, GaAs, and ZnO. However, it is relatively larger when all other substrates are used. The stress in GaN is compressive for all the substrates substrate except Si and SiC, which due to their small expansion coefficients give rise to tensile strain, which is notorious for layer cracking. The overall stress remains the same for nominally thick GaN layers when a sapphire substrate is used with or without an AlN buffer layer but reduces by an order when a 6H-SiC substrate is used with an AlN buffer layer. In these pages, the treatment reported in Refs [492,493], which followed the model of Olsen and Ettenberg [494], for an arbitrary stack of epitaxial layers on a substrate is followed. The three-layer heterostructure stack layer model used to study the thermal-induced strain is shown in Figure 2.58 with length L, width W, Youngs modulus Ei, layer thickness ti, moments Mi, coefficient of thermal expansion a, forces Fi, strain e, and curvature k. The index i ¼ 1 represents the substrate and i ¼ 2, 3, 4, . . . represents the epitaxial layers.
Figure 2.58 (a) A set of unstrained platelets of thicknesses ti (i ¼ 1, 2, 3, . . .) used to construct the composite layer. (b) A case depicting the films and the substrate to be in contact, representing the epitaxial growth of two different films on a substrate with a larger thermal
expansion coefficient (positive bending) as compared to the films. The dimensions are characterized by length L (into the page), width W, and thickness ti. The terms Ei, Mi, and Fi represent the Youngs modulus, moments (bending forces), and forces, respectively.
j291
j 2 Electronic Band Structure and Polarization Effects
292
For one epitaxial layer with i ¼ 1 representing the substrate and i ¼ 2 representing the epitaxial layer on it, one can surmise that from equilibrium of force, the total force must vanish as F 1 þ F 2 ¼ 0:
ð2:125Þ
Similarly, equilibrium of moments leads to
Wk t1 t2 ¼ 0: ðE 1 t31 þ E 2 t32 Þ þ F 1 þ F 2 t1 þ 2 2 12
ð2:126Þ
The strain at the interface between the film and the substrate in terms of mechanical parameters can be expressed as e¼
F2 F1 ðt1 þ t2 Þk : E 2 t2 L E 1 t1 L 2
ð2:127Þ
The strain is also determined from the difference between the coefficients of thermal expansion (CTE) of the substrate and the film multiplied by the difference between growth and room temperature, DT. e ¼ DTða1 a2 Þ:
ð2:128Þ
Considering a negligibly small bending stress, the one-dimensional stress in the ith epitaxial layer is then taken as constant and given by si ð1DÞ ¼
Fi ; ti W
ð2:129Þ
where si(1D) is the one-dimensional stress. By assuming a spherical bending for a square sample (meaning L W), the twodimensional stress can be deduced from the one-dimensional stress expression of Equation 2.129 as follows: si ð2DÞ ¼
si ð1DÞ
ð1 vÞ 1
;
ð2:130Þ
where si(2D) and n are the two-dimensional stress and Poissons ratio, respectively. For two epitaxial layers the equilibrium of forces Equation 2.125, the equilibrium of moments Equation 2.126, strain in terms of mechanical parameters Equation 2.127, strain in terms of the difference in CTE, and differential temperature Equation 2.128 take the form of F 1 þ F 2 þ F 3 ¼ 0;
ð2:131Þ
Wk t1 t2 t3 þ F 3 t1 þ t2 ¼ 0; ðE 1 t31 þ E 2 t32 þ E 3 t33 Þ þ F 1 þ F 2 t1 þ 2 2 2 12 ð2:132Þ e1 ¼
F2 F1 ðt1 þ t2 Þk ¼ DTða1 a2 Þ; E 2 t2 L E 1 t1 L 2
ð2:133Þ
2.12 Polarization Effects
e2 ¼
F3 F2 ðt2 þ t3 Þk ¼ DTða2 a3 Þ; E 3 t3 L E 2 t2 L 2
ð2:134Þ
where e1 and e2 represent the strain between the first epilayer and substrate and the second and first epilayers, respectively. Then si(2D), the two-dimensional stress in the epilayer i, is given as si ð2DÞ ¼
ðF i =ti WÞ
ð1 nÞ 1
:
ð2:135Þ
To calculate the misfit strain and to some extent CTE mismatch, the epitaxial relationship between the epitaxial layer and the substrate must be known, which is known simply as epitaxial relationship. That relationship between GaN and a variety of other substrates are discussed in Section 3.3. In spite of this, a succinct description is provided here as a part of the present topic on stress and strain for continuity and to help support the discussion on polarization. For sapphire substrates, orientations of (0 0 0 1), ð0 1 1 0Þ, ð2 1 1 0Þ, and ð0 1 1 2Þ (basically c-, m-, a-, and r-planes) have been used, see Table 3.6. The largest lattice mismatch, 33%, is between (0 0 0 1) GaN and ð2 1 1 0Þ sapphire along the ½0 1 1 0==½0 1 1 0 in-plane direction. The smallest mismatch, 1.19%, is between ð2 1 1 0Þ GaN and ð0 1 1 2Þ sapphire (r-plane) along the ½0 0 0 1==½0 1 1 1 in-plane direction. This orientation produces a-plane sapphire on r-plane sapphire and is discussed at some length in Sections 3.3.1 and 3.5.11. From the view point of lattice mismatch alone, the r-plane of sapphire is predicted to be most suitable for GaN growth. A thermal strain of 0.18% exists in GaN (0 0 0 1) when grown on sapphire (0 0 0 1), which is compressive. The next common substrate used to grow GaN is various polytypes of SiC. The epitaxial relationship on SiC substrates of hexagonal symmetry is tabulated in Table 3.7. The lattice parameter misfits between GaN and 6H-SiC, 3C-SiC, and 4H-SiC are very close to each other, namely, 3.48%, 3.46%, and 3.50%, respectively, and the corresponding thermal strains are 0.01, 0.09, and 0.03, respectively. The thermal strain in GaN on SiC is tensile and notorious for causing cracks in the epitaxial layers, beginning at about 2 mm thickness. In case AlN substrates would become available, growth on that substrate is also considered. Lattice misfit between GaN and AlN is only 2.41% and the thermal strain is 0.06% and 4.08% for epitaxial relationships (0 0 0 1)//(0 0 0 1) and ð1 1 2 0Þ==ð1 1 2 0Þ, respectively. From the thermal strain point of view, growth on c-plane AlN is preferred. GaN and ZnO share the stacking order and close lattice parameter, the epitaxial relationship for which is provided in Section 3.3.4. GaN can be grown on (0 0 0 1) ZnO, with the lattice parameter mismatch being only 1.97% and the misfit strain 0.21%. Some oxides have also been explored because of the small lattice misfit with GaN they provide. The epitaxial relationship between GaN and LiAlO2 is ð1 1 0 0Þ==ð1 0 0Þ with a lattice misfit of 0.31% along the [0 0 0 1]//[0 1 0] in-plane direction and a misfit strain of 0.41%. Some details can be found in Section 3.3.5. The structure of LiGaO2 is similar to the wurtzite structure, but because Li and Ga atoms have different ionic radii, the crystal has orthorhombic structure. Figure 3.36 shows the
j293
j 2 Electronic Band Structure and Polarization Effects
294
transformation of the hexagonal unit cell of GaN to an orthorhombic cell, which has lattice parameters close to that of LiGaO2. Similarly, Table 3.10 shows the corresponding lattice parameters of GaN before and after the transformation to orthorhombic cell as well as that of orthorhombic LiGaO2 unit cell. The misfit along [1 0 0] is 2.13%, with a misfit strain of 0.54%. Hexagonal (0 0 0 1) GaN has been prepared on (0 0 0 1) ScAlMgO4. The lattice parameter mismatch between the ½1 0 1 0 GaN and ½1 0 1 0 ScAlMgO4 is 1.49% and the misfit strain is 0.15%. Wurtzite (0 0 0 1) GaN films have been grown on cubic (1 1 1) spinel (MgAl2O4) substrates, as detailed in Section 3.5.10. The lattice parameter mismatch between the ½1 1 2 0 GaN and ½1 1 0 MgAl2O4 is 11.55% and the misfit strain is 0.06%. Zinc blende GaN was grown on (0 0 1) MgO with an epitaxial misfit of 6.99% and a misfit strain of 1.19%. Zinc blende GaN was grown on (0 0 1) MgO with an epitaxial misfit of 6.99% and a misfit strain of 1.19%. Mixed perovskite (La,Sr)(Al,Ta)O3 (LSAT) that is grown by Czochralski method could also be a promising substrate for GaN epitaxial layers and are reported. The lattice parameter 25
AlN Si 6HSiC 4HSiC 3CSiC LiGaO2 Al2O3 MgO GaAs ZnO MgAl2O4 LiAlO2 ScMgAlO4 NdGaO3
Curvature (1 m–1)
20
15
10
5
0.0
–5 0.0
20
40
60
Thickness (μm) Figure 2.59 A compilation of the variation of thermal curvature, a measure of strain, in epitaxial GaN layers grown on different substrates with respect to layer thickness [492]. (Please find a color version of this figure on the color tables.)
80
100
2.12 Polarization Effects
Figure 2.59 (Continued )
mismatch between ½1 1 2 0 GaN//LSAT ½1 1 0 is 16.64% and the corresponding misfit strain is 0.04%. Cubic GaN films can be epitaxially grown onto (0 0 1) Si and hexagonal polytype on (1 1 1) Si with a lattice misfit of 16.93% and misfit strain of 0.19%. The epitaxial relationship between GaN andSi is discussed in Section 3.3.3 and tabulated in Table 3.8. As for the case on GaAs, GaN with (0 0 0 1) orientation can be grown on (1 1 1) GaAs with an in-plane lattice arrangement of ½1 1 2 0==½1 1 0. The corresponding lattice misfit is 20.19% and the misfit strain 0.07%. Perovskite oxide substrates have also been considered as substrates for GaN and related structures in an effort to perhaps find a better match and/or utilize the
j295
j 2 Electronic Band Structure and Polarization Effects
296
nonlinear optical properties of perovskite oxide along with what GaN has to offer. The lattice mismatch of GaN to NdGaO3 has been calculated by assuming a perovskite cell of NdGaO3 with lattice parameters a, b, and c each being equal to 3.86 Å. Next, a new unit cell is constructed with a0 and b0 , where a0 and b0 are the diagonals of the old perovskite cell, as shown in Figure 3.37. The c0 -axis of the new cell is parallel to the caxis of the perovskite cell, but its length is doubled. In a sense this transforms a perovskite unit cell to a tetragonal unit cell. Accordingly, (1 0 0) plane becomes ð1 1 0Þ plane and (0 0 1) becomes (0 0 1). The corresponding lattice misfit is 1.72% and the misfit strain is 0.66%. Owing to the difference in stress between the thin film and substrate, the composite of film and substrate bends, which is the basis for many predigital thermometers and temperature controllers with spiral elements. The strain can be deduced from bending radius, for example, by optical means even during growth at elevated temperature, to monitor the evolution of stress and in attempts to reduce stress by epitaxial heterojunction layer design as performed on SiC substrates, see Section 3.5.3. It is therefore imperative to establish the relationship between the radius of curvature and the difference in the strain between the film and substrate, as discussed in the model of Olsen and Ettenberg [494]. In the calculation of curvature and stress instead of an average CTE over the entire range of temperature, the variation CTE with temperature has been considered for accuracy [492]. The final 5108
Stress on various substrates (Pa)
0
–5108
–1109 AlN Si 6HSiC 4HSiC 3CSiC LiGaO2 Al2O3 MgO GaAs ZnO MgAl2O4 LiAlO2 ScMgAlO4 NdGaO3
–1.5109
–2109
–2.5109 –3109 0100
210–5
410–5
610–5
810–5
Thickness (m) Figure 2.60 A compilation of residual thermal stresses in epitaxial GaN layer on different substrates with respect to layer thickness [492]. (Please find a color version of this figure on the color tables.)
110–4
2.12 Polarization Effects
j297
Stress versus thickness of GaN/potential substrates 5.00E+08
0.00E+00 0
0.00002
0.00004
0.00006
0.00008
0.0001
AlN Si MgO
–5.00E+08
3C-SiC 6H-SiC 4H-SiC
Stress (Pa)
–1.00E+09
ZnO Al2O3 LiGaO2
–1.50E+09
MgAl2O4 GaAs NdGaO3** ScAlMgO**
–2.00E+09
LiAlO2 LSAT –2.50E+09
–3.00E+09
Thickness (m) Figure 2.60 (Continued )
curvature and stress is then the integrated values from growth temperature to room temperature. Figures 2.59 and 2.60 show the curvature and the residual thermal stresses in epitaxial GaN layers grown on various substrates with respect to layer thickness, respectively. The parameters associated with each of the substrate used as well as the growth temperatures are tabulated in Table 2.36. An inspection of Figure 2.60 leads to the conclusion that the thermal stress is relatively small when GaN is grown on AlN, SiC, ScAlMgO4, Si, GaAs, and ZnO and much higher when NdGaO3 and MgO substrates are used. In all other cases (Al2O3, LiAlO2, LiGaO2, MgAl2O4, LSAT) the values are intermediate. Of paramount importance, the thermal stress is tensile in GaN when grown on Si and SiC, whereas in all other cases it is
j 2 Electronic Band Structure and Polarization Effects
298
Table 2.36 Properties and residual thermal stress of 1 mm epitaxial GaN film with other III-N compounds and substrates [492].
Substrate
Melting point ( C)
GaN AlN a-Al2O3 6H-SiC 3C-SiC 4H-SiC c-LiAlO2 b-LiGaO2 Si GaAs NdGaO3 MgO ZnO ScAlMgO4 MgAl2O4
>1700 at 2 kbar 2400 C at 30 bar 2030 2700 sublimes 1825 sublimes 2797 1700 1595 1415 1238 1600 2852 1975 2130
CTE (·106 C) (room temperature)
Growth temperature of GaN by grown OMVPE ( C)
aa ¼ 4.997, ac ¼ 4.481, a ¼ 5.45 aa ¼ 5.411 aa ¼ 8.31, ac ¼ 8.5 aa ¼ 4.76, ac ¼ 4.46 4.5 [76] 4.75 aa ¼ 12.1 aa ¼ 10.1, ab ¼ 21.1, ac ¼ 13.6 3.9 6.7 aa ¼ 11.9, ab ¼ 6.6, ac ¼ 5.8 13.9 aa ¼ 6.9, ac ¼ 4.75 aa ¼ 6.2, ac ¼ 12.2 8.7
950–1050 450–1040 950–1100 1000 600 600–1000 600 700 810 450–800 700 650 1000
compressive in nature, which leads to serious cracking issues. This is caused by CTE in SiC and Si being much smaller than that for GaN. As a result, during cooldown GaN is not freely permitted to reduce its lattice constant to the extent GaN naturally would like to, and consequently, the film remains under tensile strain, which causes cracks when the film thickness is about 2 mm or larger. Focusing on the most commonly used substrates for GaN and related epitaxy, Figures 2.61 and 2.62 show variation of the residual thermal stress in GaN versus the 7 × 107
Stress (Pa)
6 × 107 5 × 107 4 × 107 3 × 107 2 × 107 0
20
40
60
80
Thickness (μm) Figure 2.61 Residual thermal stress in GaN for a dual layer GaN/ AlN (0.1 mm) structure on 6H-SiC with respect to GaN thickness [492].
100
References
–0.2
Stress (GPa)
–0.4 –0.6 –0.8 –1.0 –1.2 0
20
60 40 Thickness (μm)
80
100
Figure 2.62 Residual thermal stress in GaN for dual layer GaN/ AlN (0.1 mm) structure on sapphire (Al2O3) with respect to GaN thickness [492].
thickness of GaN with an AlN buffer layer of thickness of 0.1 mm for dual-layer GaN/ AlN on 6H-SiC and GaN/AlN on Al2O3 heterostructures, respectively. There is very little change in thermal stress when GaN is grown on Al2O3 with or without a buffer layer of AlN. On the contrary, the stress decreases by an order by using a buffer layer of AlN while growing GaN on 6H-SiC. In summary, GaN layers grown on substrates such as AlN, SiC, ScAlMgO4, Si, GaAs, and ZnO has a residual thermal stress that is smaller by a factor of two or more as compared to the cases when the GaN layers are grown on other substrates. Moreover and very pivotally, the thermal stress is tensile in nature when grown on Si and SiC substrates whereas in all other cases it is compressive. The tensile residual strain in GaN grown on SiC and Si is notorious for cracking, and multiheterolayer buffer structures are used to deal with the problem, but not without limitations on layers thicknesses for a given substrate temperature used.
References 1 Matsuoka, T., Tanka, H., Sasaki, T. and Katsui, A. (1989) Wide-gap semiconductor (In,Ga)N. Proceedings of the Symposium on GaAs and Related Compounds, Karuizawa, Japan, Institute of Physics Conference Series 106, Chapter 3, pp. 141–146. 2 Blakemore, J.S. (1985) Solid State Physics, 2nd edn, Cambridge University Press. 3 McKelvey, J.P. (1986) Solid State Physics, 4th edn, Krieger Publishing Company. 4 Ashcroft, N.W. and Mermin, N.D. (1976) Solid State Physics, Holt-ReinhartWinston.
5 Bloch, F. (1928) Zeitschrift fur Physik, 52, 555. 6 Kronig, R.deL. and Penney, W.G. (1931) Proceedings of the Royal Society A, 130, 499. 7 Herring, C. and Hill, A.C. (1940) Physical Review, 58, 1332. 8 Woodruff, T.O. (1957) Solid State Physics, 4, 367. 9 Phillips, J.C. and Kleinman, L. (1959) New method for calculating wave functions in crystals and molecules. Physical Review, 116 (2), 287–294. 10 Cohen, M.L. and Heine, V. (1970) The fitting of pseudopotentials to experimental
j299
j 2 Electronic Band Structure and Polarization Effects
300
11 12 13
14 15
16
17
18
19
20
21
22
data and their subsequent application, in Solid State Physics: Advances in Research and Applications (eds H. Ehrenreich, F. Seitz and D. Turnbull), Academic Press, London, UK, pp. 37–248. Harrison, W.A. (1980) Solid State Theory, Dover. Slater, J.C. (1937) Physical Review, 51, 846. Hohenberg, P. and Kohn, W. (1964) Inhomogeneous electron gas. Physical Review, 136, B864–B871. Dreizler, R.M. and Gross, E.K.U. (1996) Density Functional Theory, Springer. Kohn, W. and Sham, L.J. (1965) Selfconsistent equations including exchange and correlation effects. Physical Review, 140, A1133–A1138. Lambrecht, W.R.L. and Segall, B. (1998) Band structure of the group-III nitrides, in Gallium Nitride, Semiconductors and Semimetals, vol. 50 (eds J.I. Pankove and T.D. Moustakas), Academic Press, San Diego, CA, pp. 369–402. Hedin, L. (1965) New method for calculating the one-particle Greens function with application to the electrongas problem. Physical Review, 139, A796–A823. Hedin, L. and Lundqvist, S. (1969) Solid State Physics: Advances in Research and Applications, vol. 23 (eds F. Seitz, D. Turnbull and H. Ehrenreich), Academic Press, New York, p. 1. Sham, L.J. and Schl€ uter, M. (1983) Density-functional theory of the energy gap. Physical Review Letters, 51 (20), 1888–1891. Perdew, J.P. and Levy, M. (1983) Physical content of the exact Kohn–Sham orbital energies: band gaps and derivative discontinuities. Physical Review Letters, 51 (20), 1884–1887. Hybertsen, M.S. and Louie, S.G. (1986) Electron correlation in semiconductors and insulators: band gaps and quasiparticle energies. Physical Review B: Condensed Matter, 34 (8), 5390–5413. Godby, R.W., Schl€ uter, M. and Sham, L.J. (1988) Self-energy operators and
23
24
25
26
27
28
29
30
31
32
exchange-correlation potentials in semiconductors. Physical Review B: Condensed Matter, 37 (17), 10159–10175. Aryasetiawan, F. and Gunnarsson, O. (1994) Product-basis method for calculating dielectric matrices. Physical Review B: Condensed Matter, 49 (23), 16214–16222. Aryasetiawan, F. and Gunnarsson, O. (1994) Linear-muffin-tin-orbital method with multiple orbitals per L channel. Physical Review B: Condensed Matter, 49 (11), 7219–7232. Rubio, A., Corkill, J.L., Cohen, M.L., Shirley, E.L. and Louie, S.G. (1993) Quasiparticle band structure of AlN and GaN. Physical Review B: Condensed Matter, 48 (16), 11810–11816. Lambrecht, W.R.L., Segall, B., Rife, J., Hunter, W.R. and Wickenden, D.K. (1995) Physical Review B: Condensed Matter, 51, 13516. Jancu, J.-M., Bassani, F., Della Sala, F. and Scholz, R. (2002) Transferable tightbinding parametrization for the group-III nitrides. Applied Physics Letters, 81 (25), 4838–4841. Andersen, O.K. (1975) Linear methods in band theory. Physical Review B: Condensed Matter, 12, 3060. Wimmer, E., Krakauer, H., Weinert, M. and Freeman, A. (1981) Physical Review B: Condensed Matter, 24, 864. Christensen, N.E. and Gorczyca, I. (1993) Calculated structural phase transitions of aluminum nitride under pressure. Physical Review B: Condensed Matter, 47 (8), 4307–4314. Christensen, N.E. and Gorczyca, I. (1994) Optical and structural properties of III–V nitrides under pressure. Physical Review B: Condensed Matter, 50 (7), 4397– 4415. Lambrecht, W.R.L. and Segall, B. (1995) General remarks and notations on the band structure of pure group III nitrides, in Group III Nitrides (ed. J.H. Edgar), EMIS Data Review Series, No. 11, IEE, London, p. 125.
References 33 Van Camp, P.E., Van Doren, V.E. and Devreese, J.T. (1991) High-pressure properties of wurtzite- and rocksalt-type aluminum nitride. Physical Review B: Condensed Matter, 44 (16), 9056–9059. 34 Min, B.J., Chan, C.T. and Ho, K.M. (1992) First-principles total-energy calculation of gallium nitride. Physical Review B: Condensed Matter, 45 (3), 1159–1162. 35 Yeh, C.-Y., Lu, Z.W., Froyen, S. and Zunger, A. (1992) Zinc-blende–wurtzite polytypism in semiconductors. Physical Review B: Condensed Matter, 46 (16), 10086–10097. 36 Wright, A.F. and Nelson, J.S. (1995) Firstprinciples calculations for zinc-blende AlInN alloys. Applied Physics Letters, 66 (25), 3465–3467. 37 Van Camp, P.E., Van Doren, V.E. and Devreese, J.T. (1992) Solid State Communications, 81, 23. 38 Fiorentini, V., Methfessel, M. and Scheffler, M. (1993) Electronic and structural properties of GaN by the fullpotential linear muffin-tin orbitals method: the role of the d electrons. Physical Review B: Condensed Matter, 47 (20), 13353–13362. 39 Lambrecht, W.R., Segall, B., Strite, S., Martin, G., Agarwal, A., Morkoç, H. and Rockett, A. (1994) X-ray photoelectron spectroscopy and theory of the valence band and semicore Ga 3d states in GaN. Physical Review B: Condensed Matter, 50 (19), 14155–14160. 40 Morkoç, H., Strite, S., Gao, G.B., Lin, M.E., Sverdlov, B. and Burns, M. (1994) Journal of Applied Physics, 76, 1363–1398. 41 Boguslawski, P., Briggs, E.L. and Bernholc, J. (1995) Native defects in gallium nitride. Physical Review B: Condensed Matter, 51 (23), 17255–17258. 42 Northrup, J.E. and Neugebauer, J. (1996) Theory of GaN ð1010Þ and ð1120Þ surfaces. Physical Review B: Condensed Matter, 53 (16), R10477–R10480. 43 Chen, J., Levine, Z.H. and Wilkins, J.W. (1995) Calculated second-harmonic susceptibilities of BN, AlN, and
44
45
46
47
48
49
50
51
52
53
GaN. Applied Physics Letters, 66 (9), 1129–1131. Miwa, K. and Fukumoto, A. (1993) Firstprinciples calculation of the structural, electronic, and vibrational properties of gallium nitride and aluminum nitride. Physical Review B: Condensed Matter, 48 (11), 7897–7902. Pandey, R., Jaffe, J.E. and Harrison, N.M. (1994) Ab initio study of high pressure phase transition in GaN. Journal of Physics and Chemistry of Solids, 55 (11), 1357–13561. Pandey, R., Sutjianto, A., Seel, M. and Jaffe, J.E. (1993) Electronic structure of high pressure phase of AlN. Journal of Materials Research, 8 (8), 1922–1927. Ruiz, E., Alvarez, S. and Alemany, P. (1994) Electronic structure and properties of AlN. Physical Review B: Condensed Matter, 49 (11), 7115–7123. Palummo, M., Del Sole, R., reining, L., Bechstedt, F. and Cappellini, G. (1995) Screening models and simplified GW approaches: Si & GaN as test cases. Solid State Communications, 95 (6), 393–398. Bechstedt, F. and Del Sole, R. (1988) Analytical treatment of band-gap underestimates in the local-density approximation. Physical Review B: Condensed Matter, 38 (11), 7710–7716. Bechstedt, F., Del Sole, R., Cappellini, G. and Reining, L. (1992) An efficient method for calculating quasiparticle energies in semiconductors. Solid State Communications, 84 (7), 765–770. Palummo, M., Reining, L., Godby, R.W., Bertoni, C.M. and Bornsen, N. (1994) Electronic structure of cubic GaN with self-energy corrections. Europhysics Letters, 26 (8), 607–612. Della Sala, F., Di Carlo, A., Lugli, P., Bernardini, F., Scholz, R. and Jancu, J.M. (1999) Applied Physics Letters, 74, 2002. Yang, T., Nakajima, S. and Sakai, S. (1995) Japanese Journal of Applied Physics, 34, 5912.
j301
j 2 Electronic Band Structure and Polarization Effects
302
54 Jogai, B. (1998) Effect of in-plane biaxial strains on the band structure of wurtzite GaN. Physical Review B: Condensed Matter, 57, 2382. 55 Fritsch, D., Schmidt, H. and Grundmann, M. (2003) Band-structure pseudopotential calculation of zinc-blende and wurtzite AlN, GaN, and InN. Physical Review B: Condensed Matter, 67, 235205. 56 Suzuki, M., Uenoyama, T. and Yanase, A. (1995) First-principles calculations of effective-mass parameters of AlN and GaN. Physical Review B: Condensed Matter, 52 (11), 8132–8139. 57 Suzuki, M. and Uenoyama, T. (1995) Firstprinciples calculation of effective mass parameters of gallium nitride. Japanese Journal of Applied Physics, Part 1: Regular Papers, Short Notes & Review Papers, 34 (7A), 3442–3446. 58 Chuang, S.L. and Chang, C.S. (1996) kp method for strained wurtzite semiconductors. Physical Review B: Condensed Matter, 54 (4), 2491–2504. 59 Suzuki, M. and Uenoyama, T. (1998) Electronic and optical properties of GaN based quantum wells, in Group III Nitride Semiconductor Compounds: Physics and Applications (ed. B. Gil), Clarendon, Oxford. 60 Suzuki, M. and Uenoyama, T. (1996) Japanese Journal of Applied Physics, 35, 1420. 61 Kamiyama, S., Ohnaka, K., Suzuki, M. and Uenoyama, T. (1995) Japanese Journal of Applied Physics, 34, L821. 62 Sirenko, Yu.M., Jeon, J.B., Kim, K.W., Littlejohn, M.A. and Stroscio, M.A. (1996) Physical Review B: Condensed Matter, 53, 1997. 63 Vainshtein, B.K. (1994) Fundamentals of Crystals: Modern Crystallography, 2nd edn, vol. 1, Springer, Berlin. 64 Zhao, G.L., Bagayoko, D. and Williams, T.D. (1999) Physical Review B: Condensed Matter, 60, 1563. 65 Madelung, O., Schulz, M. and Weiss, H. (eds) (1982) Numerical Data and Functional Relationships in Science and
66
67
68
69
70
71 72
73
74
75
76 77
78
Technology, Landolt-B€ornstein New Series, Group III, vol. 17, part a, Springer, New York. Yeo, Y.C., Chong, T.C. and Li, M.F. (1998) Electronic band structures and effectivemass parameters of wurtzite GaN and InN. Journal of Applied Physics, 83 (3), 1429–1436. Loughin, S., French, R.H., Ching, W.Y., Xu, Y.N. and Slack, G.A. (1993) Applied Physics Letters, 63, 1182. Foley, C.P. and Tansley, T.L. (1986) Physical Review B: Condensed Matter, 33, 1430. Fan, W.J., Li, M.F., Chong, T.C. and Xia, J.B. (1996) Journal of Applied Physics, 79, 188. Tadjer, A., Abbar, B., Rezki, M., Aourag, H. and Certier, M. (1999) Journal of Physics and Chemistry of Solids, 60, 419. Pikus, G. (1962) Soviet Physics: JETP, 14, 1075. Bir, G.L. and Pikus, G.E. (1974) Symmetry and Strain-Induced Effects in Semiconductors, John Wiley & Sons, Inc., New York. Monemar, B., Bergman, J.P. and Buyanova, I.A. (1997) Optical characterization of GaN and related material, in GaN and Related Material (ed. S.J. Pearton), Gordon and Breach, New York, pp. 85–140. Ludwig, W. and Falter, C. (1996) Symmetries in Physics. Group Theory Applied to Physical Problems, Springer Series in Solid-State Sciences, 2nd edn, vol. 64, Springer, Berlin. Inui, T., Tanabe, Y. and Onodera, Y. (1996) Group Theory and Its Applications in Physics, Springer Series Solid-State Sciences, 2nd edn, vol. 78, Springer, Berlin. Jogai, B. (1998) Solid State Communications, 107, 345. Jones, H. (1960) The Theory of Brillouin Zones and Electronic States in Crystals, North-Holland, Amsterdam. Yu, P.Y. and Cardona, M. (1999) Fundamentals of Semiconductors, 2nd edn, Springer, Berlin.
References 79 Bassani, F. and Parravicini Pastori, G. (1975) Electronic States and Optical Transitions in Solids, Pergamon, New York. 80 Nye, J.F. (1964) Physical Properties of Crystals, Oxford University Press; Nye, J.F. (1985) Physical Properties of Crystals: Their Representation by Tensors and Matrices, Clarendon, Oxford; Nye, J.F. (1998) Physical Properties of Crystals: Their Representation by Tensors and Matrices, Oxford University Press, New York. 81 Sheleg, A.U. and Savastenko, V.A. (1979) Izvestiya Akademii Nauk USSR, Neorganicheskie Materialy, 15, 1598. 82 Nye, J.F. (1985) Physical Properties of Crystals: Their Representation by Tensors and Matrices, Clarendon, Oxford. 83 Jogai, B. (2001) Three-dimensional strain field calculations in multiple InN/AlN wurtzite quantum dots. Journal of Applied Physics, 90 (2), 699–704. 84 Ambacher, O., Majewski, J., Miskys, C., Link, A., Hermann, M., Eickhoff, M., Stutzmann, M., Bernardini, F., Fiorentini, V., Tilak, V., Schaff, B. and Eastman, L.F. (2002) Pyroelectric properties of Al(In)GaN/GaN hetero- and quantum well structures. Journal of Physics: Condensed Matter, 14, 3399–3434. 85 Wagner, J.-M. and Bechstedt, F. (2002) Properties of strained wurtzite GaN and AlN: ab initio studies. Physical Review B: Condensed Matter, 66, 115202–115222. 86 Wagner, J.-M. and Bechstedt, F. (2000) Physical Review B: Condensed Matter, 62, 4526. 87 Vinet, P., Ferrante, J., Smith, J.R. and Rose, J.H. (1986) The Journal of Physical Chemistry, 19, L467. 88 Morkoç, H., Sverdlov, B. and Gao, G.B. (1993) Strained layer heterostructures and their applications to MODFETs, HBTs and lasers. Proceeding of IEEE, 81 (4), 492–556. 89 Suzuki, M. and Uenoyama, T. (1996) Strain effect on electronic and optical properties of GaN/AlGaN quantum-well lasers. Journal of Applied Physics, 80 (12), 6868–6874.
90 Ren, G.B., Liu, Y.M. and Blood, P. (1999) Valence-band structure of wurtzite GaN including the spin–orbit interaction. Applied Physics Letters, 74 (8), 117–119. 91 Albrecht, J.D., Ruden, P.P. and Reinecke, T.L. (2002) Hole scattering near the valence band edge in wurtzite gallium nitride. Journal of Applied Physics, 92 (7), 3803–3814. 92 Vurgaftman, I., Meyer, J.R. and RamMohan, L.-R. (2001) Band parameters for III–V compound semiconductors and their alloys. Journal of Applied Physics, 89 (11), 5815–5875. 93 Sirenko, Yu.M., Jeon, J.B., Lee, B.C., Kim, K.W., Littlejohn, M.A., Stroscio, M.A. and Iafrate, G.J. (1997) Hole scattering and optical transitions in wide-band-gap nitrides: wurtzite and zinc-blende structures. Physical Review B: Condensed Matter, 55 (7), 4360–4375. 94 Yamaguchi, A.A., Mochizuki, Y., Sunakawa, H. and Usui, A. (1998) Determination of valence band splitting parameters in GaN. Journal of Applied Physics, 83 (8), 4542–4544. 95 Rodina, V. and Meyer, B.K. (2001) Anisotropy of conduction band g values and interband momentum matrix elements in wurtzite GaN. Physical Review B: Condensed Matter, 64, 245209. 96 Hopfield, J.J. (1960) Journal of Physics and Chemistry of Solids, 15, 97. 97 Dingle, R., Sell, D.D., Stokowski, S.E. and Ilegems, M. (1971) Absorption, reflectance, and luminescence of GaN epitaxial layers. Physical Review B: Condensed Matter, 4 (4), 1211–1218. 98 Gil, B., Briot, O. and Aulombard, R.-L. (1995) Valence-band physics and the optical properties of GaN epilayers grown onto sapphire with wurtzite symmetry. Physical Review B: Condensed Matter, 52 (24), R17028–R17031. 99 Reynolds, D.C., Look, D.C., Kim, W., Aktas, Ö., Botchkarev, A., Salvador, A., Morkoç, H. and Talwar, D.N. (1996) Ground and excited state exciton spectra from GaN grown by molecular beam
j303
j 2 Electronic Band Structure and Polarization Effects
304
100
101
102
103
104
105
106
107
epitaxy. Journal of Applied Physics, 80 (1), 594–596. Shikanai, A., Azuhata, T., Sota, T., Chichibu, S., Kuramata, A., Horino, K. and Nakamura, S. (1997) Biaxial strain dependence of exciton resonance energies in wurtzite GaN. Journal of Applied Physics, 81 (1), 417–424. Chen, G.D., Smith, M., Lin, J.Y., Jiang, H.X., Wei, S.-H., Khan, M.A. and Sun, C.J. (1996) Fundamental optical transitions in GaN. Applied Physics Letters, 68 (20), 2784–2786. Korona, K.P., Wysmolek, A., Paluka, K., Stepniewski, R., Baranowski, J.M., Grzegory, I., Lucznik, B., Wroblewski, M. and Porowski, S. (1996) Exciton region reflectance of homoepitaxial GaN layers. Applied Physics Letters, 69 (6), 788–790. Campo, J., Julier, M., Coquillat, D., Lascaray, J.P., Scalbert, D. and Briot, O. (1997) Zeeman splittings of excitonic transitions at the Gamma point in wurtzite GaN: a magnetoreflectance investigation. Physical Review B: Condensed Matter, 56 (12), R7108–R7111. Julier, M., Campo, J., Gil, B., Lascaray, J.P. and Nakamura, S. (1998) Determination of the spin-exchange interaction constant in wurtzite GaN. Physical Review B: Condensed Matter, 57 (12), R6791–R6794. Edwards, N.V., Yoo, S.D., Bremser, M.D., Weeks, T.W., Jr, Nam, O.H., Davis, R.F., Liu, H., Stall, R.A., Horton, M.N., Perkins, N.R., Kuech, T.F. and Aspnes, D.E. (1997) Variation of GaN valence bands with biaxial stress and quantification of residual stress. Applied Physics Letters, 70 (15), 2001–2003. Rodina, A.V., Dietrich, M., G€oldner, A., Eckey, L., Hoffmann, A., Efros, Al.L., Rosen, M. and Meyer, B.K. (2002) Free excitons in wurtzite GaN. Physical Review B: Condensed Matter, 64, 115204. Wei, S.-H. and Zunger, A. (1996) Valence band splittings and band offsets of AlN, GaN, and InN. Applied Physics Letters, 69 (18), 2719–2721.
108 Yeo, Y.C., Chong, T.C. and Li, M.F. (1998) Electronic band structures and effectivemass parameters of wurtzite GaN and InN. Journal of Applied Physics, 83 (3), 1429–1436. 109 Kim, K., Lambrecht, W.R.L. and Segall, B. (1996) Elastic constants and related properties of tetrahedrally bonded BN, AlN, GaN, and InN. Physical Review B: Condensed Matter, 53 (24), 16310–16326. 110 Hsu, L. and Walukiewicz, W. (1997) Electron mobility in AlxGa1xN/GaN heterostructures. Physical Review B: Condensed Matter, 56 (3), 1520–1528. 111 Knap, W., Borovitskaya, E., Shur, M.S., Hsu, L., Walukiewicz, W., Frayssinet, E., Lorenzini, P., Grandjean, N., Skierbiszewski, C., Prystawko, P., Leszczynski, M. and Grzegory, I. (2002) Acoustic phonon scattering of twodimensional electrons in GaN/AlGaN heterostructures. Applied Physics Letters, 80 (7), 1228–1230. 112 Shan, W., Hauenstein, R.J., Fischer, A.J., Song, J.J., Perry, W.G., Bremser, M.D., Davis, R.F. and Goldenberg, B. (1996) Strain effects on excitonic transitions in GaN: deformation potentials. Physical Review B: Condensed Matter, 54 (19), 13460–13463. 113 Ghosh, S., Waltereit, P., Brandt, O., Grahn, H.T. and Ploog, K.H. (2002) Electronic band structure of wurtzite GaN under biaxial strain in the M plane investigated with photoreflectance spectroscopy. Physical Review B: Condensed Matter, 65, 075202. 114 Gil, B. and Alemu, A. (1997) Optical anisotropy of excitons in strained GaN epilayers grown along the h1010i direction. Physical Review B: Condensed Matter, 56 (19), 12446–12453. 115 Ohtoshi, T., Niwa, A. and Kuroda, T. (1997) Dependence of optical gain on crystal orientation in wurtzite–GaN strained quantum-well lasers. Journal of Applied Physics, 82, (4), 1518–1520. 116 Suzuki, M. and Uenoyama, T. (1996) Strain effect on electronic and optical
References
117
118
119
120
121
122
123
124
125
126
properties of GaN/AlGaN quantum-well lasers. Journal of Applied Physics, 80, 6868–6874. Alemu, A., Gil, B., Julier, M. and Nakamura, S. (1998) Optical properties of wurtzite GaN epilayers grown on A-plane sapphire. Physical Review B: Condensed Matter, 57 (7), 3761–3764. Lambrecht, W.R.L. and Segall, B. (1995) General remarks and notations on the band structure of pure group III nitrides, in Group III Nitrides (ed. J.H. Edgar), EMIS Data Review Series, No. 11, IEE, London, p. 125. Huang, Y.S., Pollak, F.H., Park, S.S., Lee, K.Y. and Morkoç, H. (2003) Contactless electroreflectance investigation, in the range of 20 K < T < 300 K of freestanding wurtzite GaN prepared by hydride-vaporphase-epitaxy. Journal of Applied Physics, 94 (2), 899–903. Logothetidis, S., Petalas, J., Cardona, M. and Moustakas, T.D. (1994) Optical properties and temperature dependence of the interband transitions of cubic and hexagonal GaN. Physical Review B: Condensed Matter, 50 (24), 18017–18029. Yang, Z. and Xu, Z. (1996) Electronic and optical properties of unstrained and strained wurtzite GaN. Physical Review B: Condensed Matter, 54 (24), 17577–17584. Bouhafs, B. and Lakdja, A. and Ruterana, P. (2003) First-principles calculations of the structural and electronic properties of the III-nitrides-based superlattices. Physica A, 17 (1–4), 235–237. Barker, A.S., Jr and Illegems, M. (1973) Infrared lattice vibrations and freeelectron dispersion in GaN. Physical Review B: Condensed Matter, 7 (2), 743–750. Rheinlander, B. and Neumann, H. (1974) Physica Status Solidi b: Basic Research, 64, K123. Sidorov, V.G., Sveshkova, L.S., Shagalov, M.D. and Shalabutov, Yu.K. (1976) Soviet Physics: Semiconductors, 10, 1309. Pankove, J.I., Bloom, S. and Harbeke, G. (1975) RCA Review, 36, 163.
127 Kesamanly, F.P. (1974) Soviet Physics: Semiconductors, 8, 147. 128 Meyer, B.K., Volm, D., Graber, A., Alt, H.C., Detchprohm, T., Amano, A. and Akasaki, I. (1995) Shallow donors in GaN – the binding energy and the electron effective mass. Solid State Communications, 95, 597–600. 129 Witowski, A.M., Pakula, K., Baranowski, J.M., Sadowski, M.L. and Wyder, P. (1999) Electron effective mass in hexagonal GaN. Applied Physics Letters, 75 (26), 4154–4155. 130 Drechsler, M., Hoffman, D.M., Meyer, B.K., Detchprohm, T., Amano, H. and Akasaki, I. (1995) Determination of the conduction band electron effective mass in hexagonal GaN. Japanese Journal of Applied Physics, Part 2: Letters, 34 (9B), L1178–L1179. 131 Perlin, P., Litwin-Staszewska, E., Suchanek, B., Knap, W., Camassel, J., Suski, T., Piotrzkowski, R., Grzegory, I., Porowski, S., Kaminska, E. and Chervin, J.C. (1996) Determination of the effective mass of GaN from infrared reflectivity and Hall effect. Applied Physics Letters, 68 (8), 1114–1116. 132 Wang, Y.J., Kaplan, R., Ng, H.K., Doverspike, K., Gaskill, D.K., Ikedo, T., Akasaki, I. and Amono, H. (1996) Magneto-optical studies of GaN and GaN/ AlxGa1xN: donor Zeeman spectroscopy and two-dimensional electron gas cyclotron resonance. Journal of Applied Physics, 79 (10), 8007–8010. 133 Knap, W., Contreras, S., Alause, H., Skierbiszewski, C., Camassel, J., Dyakonov, M., Robert, J.L., Yang, J., Chen, Q., Asif Khan, M., Sadowski, M.L., Huant, S., Yang, F.H., Goiran, M., Leotin, J. and Shur, M.S. (1997) Cyclotron resonance and quantum Hall effect studies of the two-dimensional electron gas confined at the GaN/AlGaN interface. Applied Physics Letters, 70 (16), 2123–2125. 134 Kasic, A., Schubert, M., Einfeldt, S., Hommel, D. and Tiwald, T.E. (2000) Free-carrier and phonon properties of n- and p-type hexagonal GaN films
j305
j 2 Electronic Band Structure and Polarization Effects
306
135
136
137
138
139
140
141
measured by infrared ellipsometry. Physical Review B: Condensed Matter, 62 (11), 7365–7377. Elhamri, S., Newrock, R.S., Mast, D.B., Ahoujja, M., Mitchel, W.C., Redwing, J.M., Tischler, M.A. and Flynn, J.S. (1998) Al0.15Ga0.85N/GaN heterostructures: effective mass and scattering times. Physical Review B: Condensed Matter, 57 (3), 1374–1377. Saxler, A., Debray, P., Perrin, R., Elhamri, S., Mitchel, W.C., Elsass, C.R., Smorchkova, I.P., Heying, B., Haus, E., Fini, P., Ibbetson, J.P., Keller, S., Petroff, P.M., DenBaars, S.P., Mishra, U.K. and Speck, J.S. (2000) Characterization of an AlGaN/GaN two-dimensional electron gas structure. Journal of Applied Physics, 87 (1), 369–374. Wong, L.W., Cai, S.J., Li, R., Wang, K., Jiang, H.W. and Chen, M. (1998) Magnetotransport study on the twodimensional electron gas in AlGaN/GaN heterostructures. Applied Physics Letters, 73 (10), 1391–1393. Wang, T., Bai, J., Sakai, S., Ohno, Y. and Ohno, H. (2000) Magnetotransport studies of AlGaN/GaN heterostructures grown on sapphire substrates: effective mass and scattering time. Applied Physics Letters, 76 (19), 2737–2739. Hang, D.R., Liang, C.-T., Huang, C.F., Chang, Y.H., Chen, Y.F., Jiang, H.X. and Lin, J.Y. (2001) Effective mass of twodimensional electron gas in an Al0.2Ga0.8N/GaN heterojunction. Applied Physics Letters, 79 (1), 66–68. Goano, M., Bellotti, E., Ghillino, E., Ghione, G. and Brennan, K.F. (2000) Band structure nonlocal pseudopotential calculation of the III-nitride wurtzite phase materials system. Part I. Binary compounds GaN, AlN, and InN. Journal of Applied Physics, 88 (11), 6467–6475. Eliseev, P.G., Perlin, P., Lee, J. and Osinski, M. (1997) Blue, temperatureinduced shift and band-tail emission in InGaN-based light sources. Applied Physics Letters, 71 (5), 569–571.
142 Gurioli, M., Vinattieri, A., Pastor, J.M. and Colocci, M. (1994) Exciton thermalization in quantum-well structures. Physical Review B: Condensed Matter, 50, 11817–11826. 143 Monemar, B. (1974) Fundamental energy gap of GaN from photoluminescence excitation spectra. Physical Review B: Condensed Matter, 10 (2), 676–681. 144 Teisseyre, H., Perlin, P., Suski, T., Grzegory, I., Porowski, S., Jun, J., Pietraszko, A. and Moustakas, T.D. (1994) Temperature dependence of the energy gap in GaN bulk single crystals and epitaxial layer. Journal of Applied Physics, 76 (4), 2429–2434. 145 Shan, W., Schmidt, T.J., Yang, X.H., Hwang, S.J., Song, J.J. and Goldenberg, B. (1995) Temperature dependence of interband transitions in GaN grown by metalorganic chemical vapor deposition. Applied Physics Letters, 66 (8), 985–987. 146 Petalas, J., Logothetidis, S., Boultadakis, S., Alouani, M. and Wills, J.M. (1995) Optical and electronic-structure study of cubic and hexagonal GaN thin films. Physical Review B: Condensed Matter, 52 (11), 8082–8091. 147 Salvador, A., Liu, G., Kim, W., Aktas, O., Botchkarev, A. and Morkoç, H. (1995) Properties of a Si doped GaN/AlGaN single quantum well. Applied Physics Letters, 67, 3322–3324. 148 Manasreh, M.O. (1996) Optical absorption near the band edge in GaN grown by metalorganic chemical-vapor deposition. Physical Review B: Condensed Matter, 53 (24), 16425–16428. 149 Li, C.F., Huang, Y.S., Malikova, L. and Pollak, F.H. (1997) Temperature dependence of the energies and broadening parameters of the interband excitonic transitions in wurtzite GaN. Physical Review B: Condensed Matter, 55 (15), 9251–9254. 150 Zubrilov, A.S., Melnik, Yu.V., Nikolaev, A.E., Jacobson, M.A., Nelson, D.K. and Dmitriev, V.A. (1999) Optical properties of
References
151
152
153
154
155
156 157 158
159
160
161
162
gallium nitride bulk crystals grown by chloride vapor phase epitaxy. Semiconductors, 33 (10), 1067–1071. Reynolds, D.C., Hoelscher, J., Litton, C.W. and Collins, T.C. (2002) Temperature dependence of free excitons in GaN. Journal of Applied Physics, 92 (9), 5596–5598. Vurgaftman, I. and Meyer, J.R. (2003) Band parameters for nitrogen-containing semiconductors. Journal of Applied Physics, 94 (6), 3675–3696. Dugdale, D.J., Brand, S. and Abram, R.A. (2000) Direct calculation of kp parameters for wurtzite AlN, GaN, and InN. Physical Review B: Condensed Matter, 61 (19), 12933–12938. Kim, K., Lambrecht, W.R.L., Segall, B. and van Schilfgaarde, M. (1997) Effective masses and valence-band splittings in GaN and AlN. Physical Review B: Condensed Matter, 56 (12), 7363–7375. Kawabe, K., Tredgold, R.H. and Inuishi, Y. (1967) Electrical Engineering in Japan, 87, 62. Balkanski, M. and de Cloizeaux, J. (1960) Journal de Physique et le Radium, 21, 825. Hopfield, J.J. and Thomas, D.G. (1963) Physical Review, 132, 563. Kane, E.O. (1958) Journal of Physics and Chemistry of Solids, 6, 236; Kane, E.O. (1966) Semiconductors and Semimetals, vol. 1 (eds R.K. Willardson and A.C., Beer), Academic Press; Kane, E.O. (1982) Handbook on Semiconductors, vol. 1 (ed. W. Paul), North-Holland, New York, p. 193. Bloom, S., Harbeke, G., Meier, E. and Ortenburger, I.B. (1974) Physica Status Solidi b: Basic Research, 66, 161. Strite, S., Ran, J., Li, Z., Salvador, A., Chen, H., Smith, D.J., Choyke, W.J. and Morkoç, H. (1991) Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 9, 1924. Sitar, Z., Paisley, M.J., Ruan, J., Choyke, J.W. and Davis, R.F. (1992) Journal of Materials Science Letters, 11, 261. Okumura, H., Yoshida, S. and Okahisa, T. (1994) Applied Physics Letters, 64, 2997.
163 Pugh, S.K., Dugdale, D.J., Brand, S. and Abram, R.A. (1999) Electronic structure calculations on nitride semiconductors. Semiconductor Science and Technology, 14, 23–31. 164 Menniger, J., Jahn, U., Brandt, O., Yang, H. and Ploog, K. (1996) Physical Review B: Condensed Matter, 53, 1881. 165 Wu, J., Yaguchi, H., Onable, K. and Ito, R. (1997) Applied Physics Letters, 71, 2067. 166 Xu, D., Yang, H., Li, J.B., Zhao, D.G., Li, S.F., Zhuang, S.M., Wu, R.H., Chen, Y. and Li, G.H. (2000) Applied Physics Letters, 76, 3025. 167 Ramirez-Flores, G., Navarro-Contreras, H., Lastras-Martinez, A., Powell, R.C. and Greene, J.E. (1994) Physical Review B: Condensed Matter, 50, 8433. 168 Fanciulli, M., Lei, T. and Moustakas, T.D. (1993) Conduction-electron spin resonance in zinc-blende GaN thin films. Physical Review B: Condensed Matter, 48 (20), 15144–15147. 169 Chow, W.W., Wright, A.F. and Nelson, J.S. (1996) Theoretical study of room temperature optical gain in GaN strained quantum wells. Applied Physics Letters, 68 (3), 296–298. 170 Suzuki, M. and Uenoyama, T. (1996) Optical gain and crystal symmetry in III–V nitride lasers. Applied Physics Letters, 69 (22), 3378–3380. 171 Ramos, L.E., Teles, L.K., Scolfaro, L.M.R., Castineira, J.L.P., Rosa, A.L. and Leite, J.R. (2001) Structural, electronic, and effective-mass properties of silicon and zinc-blende group-III nitride semiconductor compounds. Physical Review B: Condensed Matter, 63, 165210. 172 Pfeffer, P. and Zawadzki, W. (1988) Resonant and nonresonant polarons in semiconductors. Physical Review B: Condensed Matter, 37 (5), 2695–2698. 173 Meney, T., OReilly, E.P. and Adams, A.R. (1996) Optical gain in wide bandgap GaN quantum well lasers. Semiconductor Science and Technology, 11, 897–903.
j307
j 2 Electronic Band Structure and Polarization Effects
308
174 Fan, W.J., Li, M.F., Chong, T.C. and Xia, J.B. (1996) Solid State Communications, 97, 381. 175 Ohtoshi, T., Niwa, A. and Kuroda, T. (1998) Orientation dependence of optical gain in zincblende-GaN strainedquantum-well lasers. IEEE Journal of Selected Topics in Quantum Electronics, 4 (3), 527–530. 176 Wei, S.-H. and Zunger, A. (1999) Predicted band-gap pressure coefficients of all diamond and zinc-blende semiconductors: chemical trends. Physical Review B: Condensed Matter, 60 (8), 5404–5411. 177 Van de Walle, C.G. and Neugebauer, J. (1997) Small valence-band offsets at GaN/ InGaN heterojunctions. Applied Physics Letters, 70 (19), 2577–2579. 178 Binggeli, N., Ferrara, P. and Baldereschi, A. (2001) Band-offset trends in nitride heterojunctions. Physical Review B: Condensed Matter, 63, 245306. 179 Wright, A.F. (1997) Elastic properties of zinc-blende and wurtzite AlN, GaN, and InN. Journal of Applied Physics, 82 (6), 2833–2839. 180 Kim, K., Lambrecht, W.R.L. and Segall, B. (1997) Erratum: Elastic constants and related properties of tetrahedrally bonded BN, AlN, GaN, and InN [Phys. Rev. B 53, 16310 (1996)]. Physical Review B: Condensed Matter, 56 (11), 7018. 181 Bechstedt, F., Grossner, U. and Furthm€ uller, J. (2000) Dynamics and polarization of group-III nitride lattices: a first-principles study. Physical Review B: Condensed Matter, 62 (12), 8003–8011. 182 Yim, W.M., Stofko, E.J., Zanzucchi, P.J., Pankove, J.I., Ettenberg, M. and Gilbert, S.L. (1973) Journal of Applied Physics, 44, 292. 183 Perry, P.B. and Rutz, R.F. (1978) The optical absorption edge of single-crystal AlN prepared by a close-spaced vapor process. Applied Physics Letters, 33 (4), 319–321. 184 Guo, Q. and Yoshida, A. (1994) Japanese Journal of Applied Physics, Part 1: Regular
185
186
187
188
189
190
191
192
193
Papers, Short Notes & Review Papers, 33, 2454. Vispute, R.D., Wu, H. and Narayan, J. (1995) High quality epitaxial aluminum nitride layers on sapphire by pulsed laser deposition. Applied Physics Letters, 67 (11), 1549–1551. Tang, X., Hossain, F., Wongchotigul, K. and Spencer, M.G. (1998) Near band-edge transition in aluminum nitride thin films grown by metal organic chemical vapor deposition. Applied Physics Letters, 72 (12), 1501–1503. Brunner, D., Angerer, H., Bustarret, E., Freudenberg, F., Hoepler, R., Dimitrov, R., Ambacher, O. and Stutzmann, M. (1997) Optical constants of epitaxial AlGaN films and their temperature dependence. Journal of Applied Physics, 82 (10), 5090–5096. Wethkamp, T., Wilmers, K., Cobet, C., Esser, N., Richter, W., Ambacher, O., Stutzmann, M. and Cardona, M. (1999) Physical Review B: Condensed Matter, 59, 1845. Kuokstis, E., Zhang, J., Fareed, Q., Yang, J.W., Simin, G., Asif Khan, M., Gaska, R., Shur, M., Rojo, C. and Schowalter, L. (2002) Near-band-edge photoluminescence of wurtzite-type AlN. Applied Physics Letters, 81 (15), 2755–2757. Guo, Q., Nishio, M., Ogawa, H. and Yoshida, A. (2001) Temperature effect on the electronic structure of AlN. Physical Review B: Condensed Matter, 64, 113105. Silveira, E., Freitas, J.A., Jr, Kneissl, M., Treat, D.W., Johnson, N.M., Slack, G.A. and Schowalter, L.J. (2004) Nearbandedge cathodoluminescence of an AlN homoepitaxial film. Applied Physics Letters, 84 (18), 3501–3503. Slack, G.A., Schowalter, L.J., Morelli, D. and Freitas, J.A., Jr (2002) Journal of Crystal Growth, 246, 287. Shimada, K., Sota, T. and Suzuki, K. (1998) First-principles study on electronic and elastic properties of BN, AlN, and GaN. Journal of Applied Physics, 84, (9), 4951–4958.
References 194 Silveira, E., Freitas, J.A., Jr, Glembocki, O.J., Slack, G.A. and Schowalter, L.J. (2005) Excitonic structure of bulk AlN from optical reflectivity and cathodoluminescence measurements. Physical Review B: Condensed Matter, 71, 041201(R)-1–041201(R)-4. 195 Oshikiri, M., Aryasetiawan, F., Imanaka, Y. and Kido, G. (2002) Physical Review B: Condensed Matter, 66, 125204. 196 Persson, C., Sernelius, B.E., Ferreira da Silva, A., Ahuja, R. and Johansson, B. (2001) Journal of Physics: Condensed Matter, 13, 8915. 197 Shan, W., Ager, J.W., III, Walukiewicz, W., Haller, E.E., Little, B.D., Song, J.J., Schurman, M., Feng, Z.C., Stall, R.A. and Goldenberg, B. (1998) Near-band-edge photoluminescence emission in AlxGa1xN under high pressure. Applied Physics Letters, 72 (18), 2274–2276. 198 Tsubouchi, K., Sugai, K. and Mikoshiba, N. (1981) Ultrasonics Symposium, vol. 1 (ed. B.R. McAvoy), IEEE, New York, p. 375. 199 McNeil, L.E., Grimditch, M. and French, R.H. (1993) Journal of the American Ceramic Society, 76, 1132. 200 Deger, C., Born, E., Angerer, H., Ambacher, O., Stutzmann, M., Hornsteiner, J., Riha, E. and Fischerauer, G. (1998) Sound velocity of AlxGa1xN thin films obtained by surface acousticwave measurements. Applied Physics Letters, 72 (19), 2400–2402. 201 Davydov, S.Yu. and Davydov, S.Yu. (2002) Evaluation of physical parameters for the group III nitrates: BN, AlN, GaN, and InN. Semiconductors, 36 (1), 41–44. 202 Davydov, V.Yu., Kitaev, Yu.E., Goncharuk, I.N., Smirnov, A.N., Graul, J., Semchinova, O., Uffmann, D., Smirnov, M.B., Mirgorodsky, A.P. and Evarestov, R.A. (1998) Phonon dispersion and Raman scattering in hexagonal GaN and AlN. Physical Review B: Condensed Matter, 58 (19), 12899–12907. 203 Ruiz, E., Alvarez, S. and Alemany, P. (1994) Electronic structure and properties
204
205 206
207
208
209
210
211
212
213
of AlN. Physical Review B: Condensed Matter, 49 (11), 7115–7123. Kato, R. and Hama, J. (1994) Firstprinciples calculation of the elastic stiffness tensor of aluminium nitride under high pressure. Journal of Physics: Condensed Matter, 6, 7617–7632. Hutson, A.R. (1963) US Patent No. 3,090,876, May 21. Tsubouchi, K., Sugai, K. and Mikoshiba, N. (1982) Proceedings of the IEEE Ultrasonic Symposium, IEEE, New York, p. 340. Muensit, S., Goldys, E.M. and Guy, I.L. (1999) Shear piezoelectric coefficients of gallium nitride and aluminum nitride. Applied Physics Letters, 75 (25), 3965–3967. Lueng, C.M., Chan, H.W., Surya, C. and Choy, C.L. (2000) Piezoelectric coefficient of aluminum nitride and gallium nitride. Journal of Applied Physics, 88 (9), 5360–5363. Bernardini, F., Fiorentini, V. and Vanderbilt, D. (1997) Spontaneous polarization and piezoelectric constants in III–V nitrides. Physical Review B: Condensed Matter, 56, R10024. Bernardini, F. and Fiorentini, V. (2002) First-principles calculation of the $ piezoelectric tensor d of III–V nitrides. Applied Physics Letters, 80 (22), 4145–4147. Kamiya, T. (1996) Japanese Journal of Applied Physics, Part 1: Regular Papers, Short Notes & Review Papers, 35, 4421. Zoroddu, A., Bernardini, F., Ruggerone, P. and Fiorentini, V. (2001) Firstprinciples prediction of structure, energetics, formation enthalpy, elastic constants, polarization, and piezoelectric constants of AlN, GaN, and InN: comparison of local and gradientcorrected density-functional theory. Physical Review B: Condensed Matter, 64, 45208. Nardelli, M.B., Rapcewicz, K., Bernholc, J. (1997) Strain effects on the interface properties of nitride semiconductors. Physical Review B: Condensed Matter, 55 (12), R7323–R7326.
j309
j 2 Electronic Band Structure and Polarization Effects
310
214 Leroux, M., Grandjean, N., Massies, J., Gil, B., Lefebvre, P. and Bigenwald, P. (1999) Physical Review B: Condensed Matter, 60, 1496. 215 Grandjean, N., Damilano, B., Dalmasso, S., Leroux, M., Laugt, M. and Massies, J. (1999) Journal of Applied Physics, 86, 3714. 216 Garrido, J.A., Sanchez-Rojas, J.L., Jimenez, A., Munoz, E., Omnes, F. and Gibart, P. (1999) Applied Physics Letters, 75, 2407. 217 Hogg, R.A., Norman, C.E., Shields, A.J., Pepper, M. and Iizuka, M. (2000) Applied Physics Letters, 76, 1428. 218 Park, S.-H. and Chuang, S.-L. (2000) Applied Physics Letters, 76, 1981. 219 Cingolani, R., Botchkarev, A., Tang, H., Morkoç, H., Coliı, G., Lomascolo, M., Di Carlo, A. and Lugli, P. (2000) Spontaneous polarization and piezoelectric field in GaN/Al0.15Ga0.85N quantum wells: impact on the optical spectra. Physical Review B: Condensed Matter, 61 (4), 2711–2715. 220 Bernardini, F. and Fiorentini, V. (2001) Nonlinear macroscopic polarization in III–V nitride alloys. Physical Review B: Condensed Matter, 64, 085207. 221 Bernardini, F. and Fiorentini, V. (2002) Erratum: Nonlinear macroscopic polarization in III–V nitride alloys [Phys. Rev. B 64, 085207 (2001)]. Physical Review B: Condensed Matter, 65, 129903(E). 222 Fiorentini, V., Bernardini, F. and Ambacher, O. (2002) Evidence for nonlinear macroscopic polarization in III–V nitride alloy heterostructures. Applied Physics Letters, 80 (7), 1204–1206. 223 Petrov, I., Mojab, E., Powell, R.C., Greene, J.E., Hultman, L. and Sundgren, J.-E. (1992) Synthesis of metastable epitaxial zinc-blende-structure AlN by solid-state reaction. Applied Physics Letters, 60 (20), 2491–2493. 224 Gerthsen, D., Neubauer, B., Dieker, Ch., th, H. (1999) Lantier, R., Rizzi, A. and Lu Molecular beam epitaxy (MBE) growth and structural properties of GaN and AlN on 3C-SiC(0 0 1) substrates. Journal of Crystal Growth, 200, 353–361.
225 Thompson, M.P., Auner, G.W., Zheleva, T.S., Jones, K.A., Simko, S.J. and Hilfiker, J.N. (2001) Deposition factors and band gap of zinc-blende AlN. Journal of Applied Physics, 89 (6), 3331–3336. 226 Lebedev, V., Cimalla, V., Kaiser, U., Foerster, Ch., Pezoldt, J., Biskupek, J. and Ambacher, O. (2005) Effect of nanoscale surface morphology on the phase stability of 3C-AlN films on Si(1 1 1). Journal of Applied Physics, 97, 114306–1. 227 Cardona, M. and Christensen, N.E. (2000) Spin-orbit splittings in AlN, GaN and InN. Solid State Communications, 116 (8), 421–425. 228 Karch, K. and Bechstedt, F. (1997) Ab initio lattice dynamics of BN and AlN: covalent versus ionic forces. Physical Review B: Condensed Matter, 56 (12), 7404–7415. 229 Foutz, B.E., Oleary, S.K., Shur, M.S. and Eastman, L.F. (1999) Journal of Applied Physics, 85, 7727. 230 Matsuoka, T., Okamoto, H. and Nakao, M. (2003) Growth of wurtzite InN using MOVPE and its optical characteristics. Physica Status Solidi c, 0 (7), 2806– 2809. 231 Matsuoka, T. (2004) MOVPE growth and characteristics of nitride semiconductors from GaN to InN, in Advanced Materials in Electronics (ed. Q. Guo), pp. 46–83. 232 Osamura, K., Naka, S. and Murakami, Y. (1975) Preparation and optical properties of Ga1 xInxN thin films. Journal of Applied Physics, 46 (8), 3432–3437. 233 Puychevrier, N. and Menoret, M. (1976) Thin Solid Films, 36, 141. 234 Tyagai, V.A., Evstigneev, A.M., Krasiko, A.N., Andreeva, A.F. and Malakhov, V.Ya. (1977) Soviet Physics: Semiconductors, 11, 1257. 235 Tansley, T.L. and Foley, C.P. (1986) Journal of Applied Physics, 59, 3241. 236 Westra, K.L. and Brett, M.J. (1990) Thin Solid Films, 192, 227. 237 Bhuiyan, A.G., Hashimoto, A. and Yamamoto, A. (2003) Indium nitride (InN): a review on growth,
References
238
239
240
241
242
243
244
characterization, and properties. Journal of Applied Physiology, 94 (5), 2779–2808. See Butcher, K.S. (ed.) (2006) Proceedings of the 2nd ONR International Indium Nitride Workshop Kailua-Kona, January 9–13, 2005, Hawaii, Journal of Crystal Growth, 288 (2), 217–304. Davydov, V.Yu., Klochikhin, A.A., Emtsev, V.V., Ivanov, S.V., Vekshin, V.V., Bechstedt, F., Furthm€ uller, J., Harima, H., Mudryi, A.V., Hashimoto, A., Yamamoto, A., Aderhold, J., Graul, J. and Haller, E.E. (2002) Band gap of InN and In-rich InxGa1xN alloys (0.36 < x < 1). Physica Status Solidi b: Basic Research, 230 (2), R4–R6. Davydov, V.Yu., Klochikhin, A.A., Seisyan, R.P., Emtsev, V.V., Ivanov, S.V., Bechstedt, F., Furthmuller, J., Harima, H., Mudryi, A.V., Aderhold, J., Semchinova, O. and Graul, J. (2002) Physica Status Solidi b: Basic Research, 229, R1. Davydov, V.Yu., Klochikhin, A.A., Emtsev, V.V., Kurdyukov, D.A., Ivanov, S.V., Vekshin, V.A., Bechstedt, F., Furthm€ uller, J., Aderhold, J., Graul, J., Mudryi, A.V., Harima, H., Hashimoto, A., Yamamoto, A. and Haller, E.E. (2002) Band gap of hexagonal InN and InGaN alloys. Physica Status Solidi b: Basic Research, 234 (3), 787–795. Davydov, V.Yu., Klochikhin, A.A., Seisyan, R.P., Emtsev, V.V., Ivanov, S.V., Bechstedt, F., Furthm€ uller, J., Harima, H., Mudryi, A.V., Aderhold, J., Semchinova, O. and Graul, J. (2002) Absorption and emission of hexagonal InN. Evidence of narrow fundamental band gap. Physica Status Solidi b: Basic Research, 229 (3), R1–R3. Wu, J., Walukiewicz, W.W., Yu, K.M., Ager, J.W., III, Haller, E.E., Lu, H., Schaff, W.J., Saito, Y. and Nanishi, Y. (2002) Unusual properties of the fundamental band gap of InN. Applied Physics Letters, 80 (21), 3967–3969. Matsuoka, T., Okamoto, H., Nakao, M., Harima, H. and Kurimoto, E. (2002) Optical bandgap energy of wurtzite InN. Applied Physics Letters, 81 (7), 1246–1248.
245 Gwo, S., Wu, C.-L., Shen, C.-H., Chang, W.-H., Hsu, T.M., Wang, J.-S. and Hsu, J.-T. (2004) Heteroepitaxial growth of wurtzite InN films on Si (1 1 1) exhibiting strong near-infrared photoluminescence at room temperature. Applied Physics Letters, 84 (19), 3765–3767. 246 Nag, B.R. (2002) Comment on band gap of InN and In-rich InxGa1xN alloys (0.36 < x < 1). Physica Status Solidi b: Basic Research, 233 (3), R8–R9. 247 Davydov, V.Yu., Klochikhin, A.A., Emtsev, V.V., Bechstedt, F., Mudryi, A.V. and Haller, E.E. (2002) Reply to Comment on band gap of InN and In-rich InxGa1xN alloys (0.36 < x < 1). Physica Status Solidi b: Basic Research, 233 (3), R10–R11. 248 Inushima, T., Mamutin, V.V., Vekshin, V.A., Ivanov, S.V., Sakon, T., Motokawa, M. and Ohoya, S. (2001) Physical properties of InN with the band gap energy of 1.1 eV. Journal of Crystal Growth, 227–228, 481–485. 249 Arnaudov, B., Paskova, T., Paskov, P.P., Magnusson, B., Valcheva, E., Monemar, B., Lu, H., Schaff, W.J., Amano, H. and Akasaki, I. (2004) Energy position of nearband-edge emission spectra of InN epitaxial layers with different doping levels. Physical Review B: Condensed Matter, 69, 115216. 250 Araki, T., Saito, Y., Yamaguchi, T., Kurouchi, M., Nanishi, Y. and Naoi, H. (2004) Radio frequency-molecular beam epitaxial growth of InN epitaxial films on (0 0 0 1) sapphire and their properties. Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 22 (4), 2139–2143. 251 Xu, K., Terashima, W., Hata, T., Hashimoto, N., Yoshitani, M., Cao, B., Ishitani, Y. and Yoshikawa, A. (2003) Comparative study of InN growth on Gaand N-polarity GaN templates by molecular-beam epitaxy. Physica Status Solidi c, 0 (7), 2814–2817. 252 Xu, K. and Yoshikawa, A. (2003) Effects of film polarities on InN growth by
j311
j 2 Electronic Band Structure and Polarization Effects
312
253
254
255
256
257
258
259
260
261
molecular-beam epitaxy. Applied Physics Letters, 83 (2), 251–253. Yoshikowa, A., Xu, K., Hashimoto, N., Terashima, W., Yoshitani, M., Choe, S. and Ishitani, Y. (2003) Effects of film polarities on InN growth by RF-molecular beam epitaxy, MRS 8th Wide-Bandgap IIINitride Workshop, September 29–October 1, Richmond, VA. Xu, K. and Yoshikawa, A. (2003) Effects of film polarities on InN growth by molecular-beam epitaxy. Applied Physics Letters, 83 (2), 251–253. Ishitani, Y., Xu, K., Terashima, W., Masuyama, H., Yoshitani, M., Hashimoto, N., Che, S.B. and Yoshikawa, A. (2004) Temperature dependence of the optical properties for InN films grown by RF-MBE. GaN and Related Alloys – 2003 Symposium, Warrendale, PA, USA, Material Research Society Symposium Proceedings, 798, 207–212. Ishitani, Y., Masuyama, H., Terashima, W., Yoshitani, M., Hashimoto, N., Che, S.B. and Yoshikawa, A. (2005) Bandgap energy of InN and its temperature dependence. Physica Status Solidi c, 2 (7), 2276–2276. Lan, Z.H., Wang, W.M., Sun, C.L., Shi, S.C., Hsu, C.W., Chen, T.T., Chen, K.H., Chen, C.C., Chen, Y.F. and Chen, L.C. (2004) Growth mechanism, structure and IR photoluminescence studies of indium nitride nanorods. Journal of Crystal Growth, 269, 87–94. Chen, L.C. Center for Condensed Matter Sciences, Academia Sinica, private communication. ODonnell, K.P., Pereira1 S., Martin, R.W., Edwards, P.R., Tobin, M.J. and Mosselmans, J.F.W. (2003) Physica Status Solidi a: Applied Research, 195, 532. Anderson, P.A., Lee, T.E., Kendrick, C.E., Diehl, W., Kinsey, R.J., Kennedy, V.J., Markwitz, A., Reeves, R.J. and Durbin, S.M. (2004) Proceedings of the SPIE, 5277, 90. Butcher, K.S.A., Wintrebert-Fouquet, M., Chen, P.P., Timmers, H. and Sherestha,
262
263
264
265
266
267
268
269 270
S.K. (2003) Detailed analysis of the absorption data for InN. Presented at the International Symposium on Point Defects and Nonstoichiometry, March 2003, Sendai, Japan, Material Science in Semiconductor Processing, 6, 351. Shubina, T.V., Ivanov, S.V., Jmerik, V.N., Solnyshkov, D.D., Vekshin, V.A., Kopev, P.S., Vasson, A., Leymarie, J., Kavokin, A., Amano, H., Shimono, K., Kasic, A. and Monemar, B. (2004) Mie resonances, infrared emission, and the band gap of InN. Physical Review Letters, 92 (11), 117407–117407. Juza, R. and Hahn, H. (1938) Zeitschrift f€ ur Anorganishe und Allegemeine Chemie, 239, 282. Hahn, H. and Juza, R. (1940) Zeitschrift f€ ur Anorganishe und Allegemeine Chemie, 244, III. MacChesney, J.B., Bridenbaugh, P.M., OConnor, P.B. (1970) Materials Research Bulletin, 5, 783. Tomizawa, J., Takaim, O., Tsujikawa, S. and Goto, S. (1983) Reactive evaporation technique assisted by low temperature plasma for preparation of ultrafine nitride particles. Proceedings of the International Ion Engineering Congress, Institute of Electrical Engineering, Kyoto, Japan, p. 1411. Butcher, K.S.A. (2004) InN, a historic review – from obscurity to controversy, in Advanced Materials in Electronics, Research Signpost (ed. Q. Guo), p. 1 Butcher, K.S.A. and Tansley, T.L. (2005) InN, latest development and a review of the band-gap controversy. Superlattices and Microstructures, 38 (1), 1–37. Hamberg, I. and Granqvist, C.G. (1986) Journal of Applied Physics, 60, R123. Davydov, V.Yu., Klochikhin, A.A., Emtsev, V.V., Kurdyukov, D.A., Ivanov, S.V., Vekshin, V.A., Bechstedt, F., Furthm€ uller, J., Aderhold, J., Graul, J., Mudryi, A.V., Harima, H., Hashimoto, A., Yamamoto, A. and Haller, E.E. (2002) Physica Status Solidi b: Basic Research, 234, 787.
References 271 Kumar, S., Motlan, L.M. Tansley, T.L. (1996) Japanese Journal of Applied Physics, 35, 2261–2267. 272 Butcher, K.S.A., Wintrebert-Fouquet, M., Chen, P.P.-T., Tansley, T.L., Dou, H., Shrestha, S.K., Timmers, H., Kuball, M., Prince, K.E. and Bradby, J.E. (2004) Nitrogen rich indium nitride. Journal of Applied Physics, 95, 6124. 273 Yoshimoto, M., Vamamoto, V. and Saraie, J. (2003) Fabrication of InN/Si heterojunctions with rectifying characteristics. Physica Status Solidi c, 0, 2794–2797. 274 Westra, K.L., Lawson, R.P.W. and Brett, M.J. (1988) Journal of Vacuum Science & Technology A: Vacuum Surfaces and Films, 6, 1730. 275 Foley, C.P. and Lyngdal, J. (1987) Journal of Vacuum Science & Technology A: Vacuum Surfaces and Films, 5, 1708. 276 Butcher, K.S.A., Dou, H., Goldys, E.M., Tansley, T.L. and Srikeaw, S. (2002) Ultraviolet Raman and optical transmission studies of rf sputtered indium nitride. Physica Status Solidi c, 0, 373–376. 277 Trainor, J.W. and Rose, K. (1974) Journal of Electronic Materials, 3, 821. 278 Ivanov, S.V., Jmerik, V.N., Shubina, T.V., Vekshin, V.A., Kopev, P.S. and Monemar, B. (2003) Plasma assisted MBE growth and characterisation of InN on sapphire. Presentation at the 1st International InN Workshop, November 16–20, Fremantle, Australia. 279 Paszkowicz, W., Cerny, R. and Krukowski, S. (2003) Powder Diffraction, 18, 114. 280 Lagerstedt, O. and Monemar, B. (1979) Physical Review B: Condensed Matter, 19, 3064. 281 Yamaguchi, T., Saito, Y., Kano, K., Araki, T., Teraguchi, N., Suzuki, A. and Nanishi, Y. (2002) Study of epitaxial relationship in InN growth on sapphire (0 0 0 1) by RFMBE. Physica Status Solidi b: Basic Research, 228, 17–20. 282 Walukiewicz, J.Wu, Yu, K.M., Shan, W., Ager, J.W., III, Hailer, E.E., Lu, H., Schaff,
283
284
285
286
287
288
289 290
291 292
W.J., Metzger, W.K. and Kurtz, S. (2003) Journal of Applied Physics, 94, 6477. Lang, D.V. (1977) Review of RadiationInduced Detects in III–V Compounds, Institute of Physics Conference Series 31, Chapter 1, p. 70. Shrestha, S.K., Timmers, H., Butcher, K.S.A. and Wintrebert-Fouquet, M. (2004) Accurate stoichiometric analysis of polycrystalline indium nitride films with elastic recoil detection. Current Applied Physics, 4, 237–240. Butcher, K.S.A., Wintrebert-Fouquet, M., Chen P.P.-T., Prince, K.E., Timmers, H., Shrestha, S.K., Shubina, T.V., Ivanov, S.V., Wuhrer, R., Phillips, M.R. and Monemar, B. (2005) Non-stoichiometry and nonhomogeneity in InN. 2004 International Workshop on Nitride Semiconductors, Pittsburgh, Physica Status Solidi c, 2 (7), 2263–2266. Shrestha, S.K., Butcher, K.S.A., Wintrebert-Fouquet, M. and Timmers, H. (2004) Reliable ERD analysis of group-III nitrides despite severe nitrogen depletion. Nuclear Instruments and Methods in Physics Research B, 219–220, 686–692. Kosiba, R., Ecke, G., Cimalla, V., Spieb, L., Krischok, S., Schaefer, J.A., Ambacher, O. and Schaff, W.J. (2004) Sputter depth profiling of InN. Nuclear Instruments and Methods in Physics Research B, 215, 486. Shubina, T.V., Ivanov, S.V., Jmerik, V.N., Solnyshkov, D.D., Vekshin, V.A., Kopev, P.S., Vasson, A., Leymarie, J., Kavokin, A., Amano, H., Shimono, K., Kasic, A. and Monemar, B. (2004) Mie resonances infrared emission, and the band gap of InN. Physical Review Letters, 92 (11), 117407–117407. Xu, K. and Yoshikawa, A. (2003) Applied Physics Letters, 83 251. Kreibig, U. and Vollmer, M. (1995) Optical Properties of Metal Clusters, Springer, Berlin. Mie, G. (1908) Annals of Physics (Leipzig), 25, 377. Sobolev, V.V. and Zlobina, M.A. (1999) Semiconductors, 33, 395.
j313
j 2 Electronic Band Structure and Polarization Effects
314
293 Nolte, D.D. (1994) Journal of Applied Physics, 76, 3740. 294 Maxwell-Garnett, J.C. (1904) Philosophical Transactions of the Royal Society of London, 203, 385. 295 Wei, S.-H., Nie, X., Batyrev, I.G. and Zhang, S.B. (2003) Breakdown of the band-gap-common-cation rule: the origin of the small band gap of InN. Physical Review B: Condensed Matter, 67, 165209. 296 Tsen, K.T., Liang, W., Ferry, D.K., Lu, H., Schaff, W.J., Özg€ ur, Ü., Fu, Y., Moon, Y.T., Yun, F., Morkoç, H., and Everitt, H.O. (2005) Optical studies of carrier dynamics and non-equilibrium optical phonons in nitride-based wide bandgap semiconductors. Superlattices and Microstructures, 38 (2), 77–77. 297 Kasic, A., Schubert, M., Saito, Y., Nanishi, Y. and Wagner, G. (2002) Effective electron mass and phonon modes in ntype hexagonal InN. Physical Review B: Condensed Matter, 65, 115206. 298 Xu, Y.-N. and Ching, W.Y. (1993) Physical Review B: Condensed Matter, 48, 4335. 299 Tsen, K.T. (2001) Ultrafast Physical Processes in Semiconductors, Semiconductors and Semimetals, vol. 67 (ed. K.T. Tsen), Academic Press, New York. 300 Wu, J., Walukiewicz, W., Shan, W., Yu, K.M., Ager, J.W., III, Li, S.X., Haller, E.E., Lu, H. and Schaff, W.J. (2003) Journal of Applied Physics, 94, 4457. 301 Arnaudov, G., Vilkotskii, V.A., Domanevskii, D.S., Evtimova, S.K. and Tkachev, V.D. (1977) Fizika i Tekhnika Poluprovodnikov, 11, 1799; Soviet Physics: Semiconductors, 1977, 11, 1054. 302 De-Sheng, J., Makita, Y., Ploog, K. and Queisser, H.J. (1982) Journal of Applied Physics, 53, 999. 303 Bugajski, M. and Lewandovski, W. (1985) Journal of Applied Physics, 57, 521. 304 Averkiev, N.S., Kalinin, B.N., Losev, A.V., Rogachev, A.A. and Filipenko, A.S. (1990) Physica Status Solidi a: Applied Research, 121, K129. 305 Arnaudov, B., Paskova, T., Goldis, E.M., Evtimova, S. and Monemar, B. (2001)
306
307
308
309
310
311
312 313 314 315
316 317 318
319 320
Physical Review B: Condensed Matter, 64, 045213. See, for example, Shklovskii, B.I. and Efros, A.L. (1984) Electronic Properties of Doped Semiconductors, Springer, Berlin, Chapter 11. Nag, B.R. (2003) On the band gap of indium nitride. Physica Status Solidi b: Basic Research, 237 (2), R1–R2. Inushima, T., Yaguchi, T., Nagase, A., Iso, A., Shiraishi, T. and Ooya, S. (1995) Proceedings of 7th International Conference on InP and Related Materials, p. 187. Inushima, T., Shiraishi, T. and Davydov, V.Y. (1999) Solid State Communications, 110, 491. Wu, J., Walukiewicz, W., Shan, W., Yu, K.M., Ager, J.W., III, Haller, E.E., Lu, H. and Schaff, W.J. (2002) Effects of the narrow band gap on the properties of InN. Physical Review B: Condensed Matter, 66, 201403. Strite, S. and Morkoç, H. (1992) GaN, AlN and InN: a review. Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 10 (4), 1237–1266. Tisch, U., Finkman, E. and Salzman, J. (2002) Applied Physics Letters, 81, 463. Tomic, S. and OReilly, E.P. (2003) IEEE Photonics Technology Letters, 15, 6. Groves, W.O., Herzog, A.H. and Craford, M.G. (1971) Applied Physics Letters, 19, 184. Wu, J., Shan, W. and Walukiewicz, W. (2002) Semiconductor Science and Technology, 17, 860. Lindsay, A., Tomic, S. and OReilly, E.P. (2003) Solid-State Electronics, 47, 443. Suemune, I., Uesugi, K. and Walukiewicz, W. (2000) Applied Physics Letters, 77, 3021. OReilly, E.P. and Lindsay, A. (1999) Physica Status Solidi b: Basic Research, 216, 131. Lindsay, A. and OReilly, E.P. (1999) Solid State Communications, 112, 443. Choulis, S.A., Tomic, S., OReilly, E.P. and Hosea, T.J.C. (2003) Solid State Communications, 125, 155.
References 321 Choulis, S.A., Hosea, T.J.C., Tomic, S., Kamal-Saadi, M., Adams, A.R., OReilly, E.P., Weinstein, B.A. and Klar, P.J. (2002) Physical Review B: Condensed Matter, 66, 165321. 322 Gil, B. (2000) Solid State Communications, 114, 623. 323 Shtinkov, N., Desjardins, P. and Masut, R.A. (2003) Physical Review B: Condensed Matter, 67, 081202. 324 Kent, P.R.C., Bellaiche, L. and Zunger, A. (2002) Semiconductor Science and Technology, 17, 851. 325 Kent, P.R.C. and Zunger, A. (2001) Physical Review Letters, 86, 2613. 326 Kent, P.R.C. and Zunger, A. (2001) Physical Review B: Condensed Matter, 64, 115208. 327 Szwacki, G. and Boguslawski, P. (2001) Physical Review B: Condensed Matter, 64, 161201. 328 Weyers, M., Sato, M. and Ando, H. (1992) Japanese Journal of Applied Physics, Part 2: Letters, 31, L853. 329 Weyers, M. and Sato, M. (1993) Applied Physics Letters, 62, 1396. 330 Kuroiwa, R., Asahi, H., Asami, K., Kim, S.-J., Iwata, K. and Gonda, S. (1998) Applied Physics Letters, 73, 2630. 331 Kondow, M., Uomi, K., Hosomi, K. and Mozume, T. (1994) Japanese Journal of Applied Physics, Part 2: Letters, 33, L1056. 332 Pozina, G., Ivanov, I., Monemar, B., Thordson, J.V. and Andersson, T.G. (1998) Journal of Applied Physics, 84, 3830. 333 Bellaiche, L. and Zunger, A. (1998) Physical Review B: Condensed Matter, 57, 4425. 334 Sakai, S., Ueta, Y. and Terauchi, Y. (1993) Japanese Journal of Applied Physics, Part 1: Regular Papers, Short Notes & Review Papers, 32, 4413. 335 Wei, S.-H. and Zunger, A. (1996) Physical Review Letters, 76, 664. 336 Neugebauer, J. and Van de Walle, C.G. (1995) Physical Review B: Condensed Matter, 51, 10568. 337 Rubio, A. and Cohen, M.L. (1995) Physical Review B: Condensed Matter, 51, 4343.
338 Bellaiche, L., Wei, S.-H. and Zunger, A. (1996) Physical Review B: Condensed Matter, 54, 17568. 339 Bellaiche, L., Wei, S.-H. and Zunger, A. (1997) Applied Physics Letters, 70, 3558. 340 Mattila, T., Wei, S.-H. and Zunger, A. (1999) Physical Review B: Condensed Matter, 60, R11245. 341 Bellaiche, L., Wei, S.-H. and Zunger, A. (1997) Physical Review B: Condensed Matter, 56, 10233. 342 Bi, W.G. and Tu, C.W. (1997) Applied Physics Letters, 70, 1608. 343 Salzman, J. and Temkin, H. (1997) Materials Science & Engineering B: Solid State Materials for Advanced Technology, 50, 148. 344 Ougazzaden, A., Le Bellego, Y., Rao, E.V.K., Juhel, M., Leprince, L. and Patriarche, G. (1997) Applied Physics Letters, 70, 2861. 345 Makimoto, T., Saito, H., Nishida, T. and Kobayashi, N. (1997) Applied Physics Letters, 70, 2984. 346 Malikova, L., Pollak, F.H. and Bhat, R. (1998) Journal of Electronic Materials, 27, 484. 347 Uesugi, K., Morooka, N. and Suemune, I. (1999) Applied Physics Letters, 74, 1254. 348 Sik, J., Schubert, M., Leibiger, G., Gottschalch, V., Kirpal, G. and Humlicek, J. (2000) Applied Physics Letters, 76, 2859. 349 Kozhevnikov, M., Narayanamurti, V., Reddy, C.V., Xin, H.P., Tu, C.W., Mascarenhas, A. and Zhang, Y. (2000) Physical Review B: Condensed Matter, 61, R7861. 350 Uesugi, K., Suemune, I., Hasegawa, K., Akutagawa, T. and Nakamura, T. (2000) Applied Physics Letters, 76, 1285. 351 Perlin, P., Subramanya, S.G., Mars, D.E., Kruger, J., Shapiro, N.A., Siegle, H. and Weber, E.R. (1998) Applied Physics Letters, 73, 3703. 352 Shan, W., Walukiewicz, W., Ager, J.W., III, Haller, E.E., Geisz, J.F., Friedman, D.J., Olson, J.M. and Kurtz, S.R. (1999) Physical Review Letters, 82, 1221.
j315
j 2 Electronic Band Structure and Polarization Effects
316
353 Jones, E.D., Modine, N.A., Allerman, A.A., Kurtz, S.R., Wright, A.F., Tozer, S.T. and Wei, X. (1999) Physical Review B: Condensed Matter, 60, 4430. 354 Kurtz, S.R., Modine, N.A., Jones, E.D., Allerman, A.A. and Klem, J.F. (2002) Semiconductor Science and Technology, 17, 843. 355 Skierbiszewski, C., Perlin, P., Wisniewski, P., Knap, W., Suski, T., Walukiewicz, W., Shan, W., Yu, K.M., Ager, J.W., Haller, E.E., Geisz, J.F. and Olson, J.M. (2000) Large, nitrogeninduced increase of the electron effective mass in InyGa1yNxAs1x. Applied Physics Letters, 76 (17), 2409. 356 Skierbiszewski, C., Perlin, P., Wisniewski, P., Suski, T., Walukiewicz, W., Shan, W., Ager, J.W., Haller, E.E., Geisz, J.F., Friedman, D.J., Olson, J.M. and Kurtz, S.R. (1999) Effect of nitrogeninduced modification of the conduction band structure on electron transport in GaAsN alloys. Physica Status Solidi b: Basic Research, 216, 135. 357 Yu, K.M., Walukiewicz, W., Shan, W., Ager, J.W., III, Wu, J., Haller, E.E., Geisz, J.F., Friedman, D.J. and Olson, J.M. (2000) Physical Review B: Condensed Matter, 61, R13337. 358 Hai, P.N., Chen, W.M., Buyanova, I.A., Xin, H.P. and Tu, C.W. (2000) Applied Physics Letters, 77, 1843. 359 Pan, Z., Li, L.H., Lin, Y.W., Sun, B.Q., Jiang, D.S. and Ge, W.K. (2001) Applied Physics Letters, 78, 2217. 360 Buyanova, I.A., Chen, W.M. and Tu, C.W. (2002) Semiconductor Science and Technology, 17, 815. 361 Young, D.L., Geisz, J.F. and Coutts, T.J. (2003) Applied Physics Letters, 82, 1236. 362 Young, L., Geisz, J.F. and Coutts, T.J. (2003) Applied Physics Letters, 82, 1236. 363 Chtourou, R., Bousbih, F., Ben Bouzid, S., Charfi, F.F., Harmand, J.C., Ungaro, G. and Largeau, L. (2002) Applied Physics Letters, 80, 2075. 364 Perkins, J.D., Mascarenhas, A., Zhang, Y., Geisz, J.F., Friedman, D.J., Olson, J.M.
365
366
367
368
369
370
371 372
373
374
375
376
and Kurtz, S.R. (1999) Physical Review Letters, 82, 3312. Shan, W., Yu, K.M., Walukiewicz, W., Ager, J.W., III, Haller, E.E. and Ridgway, M.C. (1999) Applied Physics Letters, 75, 1410. Zhang, Y., Mascarenhas, A., Xin, H.P. and Tu, C.W. (2000) Physical Review B: Condensed Matter, 61, 7479. Zhang, Y., Mascarenhas, A., Geisz, J.F., Xin, H.P. and Tu, C.W. (2001) Physical Review B: Condensed Matter, 63, 085205. Zhang, Y., Mascarenhas, A., Xin, H.P. and Tu, C.W. (2001) Physical Review B: Condensed Matter, 63, 161303. Klar, P.J., Gruning, H., Heimbrodt, W., Koch, J., Hohnsdorf, F., Stolz, W., Vicente, P.M.A. and Camassel, J. (2000) Applied Physics Letters, 76, 3439. Zhang, Y., Mascarenhas, A., Xin, H.P. and Tu, C.W. (2000) Physical Review B: Condensed Matter, 61, 4433. Ya, M.H., Chen, Y.F. and Huang, Y.S. (2002) Journal of Applied Physics, 92, 1446. Egorov, A.Yu., Odnoblyudov, V.A., Mamutin, V.V., Zhukov, A.E., Tsatsulnikov, A.F., Kryzhanovskaya, N.V., Ustinov, V.M., Hong, Y.G. and Tu, C.W. (2003) Valence band structure of GaAsN compounds and band-edge lineup in GaAs/GaAsN/InGaAs heterostructures. Journal of Crystal Growth, 251 (1–4), 417–421. Kitatani, T., Kondow, M., Kikawa, T., Yazawa, Y., Okai, M. and Uomi, K. (1999) Japanese Journal of Applied Physics, Part 1: Regular Papers, Short Notes & Review Papers, 38, 5003. Sun, B.Q., Jiang, D.S., Luo, X.D., Xu, Z.Y., Pan, Z., Li, L.H. and Wu, R.H. (2000) Applied Physics Letters, 76, 2862. Buyanova, I.A., Pozina, G., Hai, P.N., Chen, W.M., Xin, H.P. and Tu, C.W. (2000) Physical Review B: Condensed Matter, 63, 033303. Krispin, P., Spruytte, S.G., Harris, J.S. and Ploog, K.H. (2001) Journal of Applied Physics, 90, 2405.
References 377 Egorov, A.Yu., Odnoblyudov, V.A., Krizhanovskaya, N.V., Mamutin, V.V. and Ustinov, V.M. (2002) Semiconductors, 36, 1355. 378 Klar, P.J., Gr€ uning, H., Heimbrodt, W., Weiser, G., Koch, J., Volz, K., Stolz, W., Koch, S.W., Tomic, S., Choulis, S.A., Hosea, T.J.C., OReilly, E.P., Hofmann M., Hader, J. and Moloney, J.V. (2002) Interband transitions of quantum wells and device structures containing Ga(N, As) and (Ga, In)(N, As). Semiconductor Science and Technology, 17, 830–830. 379 Egorov, A.Yu., Odnoblyudov, V.A., Krizhanovskaya, N.V., Mamutin, V.V. and Ustinov, V.M. (2002) Band-edge line-up in GaAs/GaAsN/InGaAs heterostructures. Fizika i Tekhnika Poluprovodnikov, 36 (12), 1440–1444. 380 Yang, T., Nakajima, S. and Sakai, S. (1997) Japanese Journal of Applied Physics, Part 2: Letters, 36, L320. 381 Tit, N. and Dharma-wardana, M.W.C. (2000) Applied Physics Letters, 76, 3576. 382 Naoi, H., Naoi, Y. and Sakai, S. (1997) Solid-State Electronics, 41, 319. 383 Beresford, R., Stevens, K.S. and Schwatzman, A.F. (1998) Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 16, 1293. 384 Fan, J.C., Hung, W.K., Chen, Y.F., Wang, J.S. and Lin, H.H. (2000) Physical Review B: Condensed Matter, 62, 10990. 385 Hung, W.K., Cho, K.S., Chern, M.Y., Chen, Y.F., Shih, D.K., Lin, H.H., Lu, C.C. and Yang, T.R. (2002) Applied Physics Letters, 80, 796. 386 Hang, D.R., Huang, C.F., Hung, W.K., Chang, Y.H., Chen, J.C., Yang, H.C., Chen, Y.F., Shih, D.K., Chu, T.Y. and Lin, H.H. (2002) Shubnikov–de Haas oscillations of two-dimensional electron gas in an InAsN/InGaAs single quantum well. Semiconductor Science and Technology, 17, 999–999. 387 Bi, W.G. and Tu, C.W. (1996) Journal of Applied Physics, 80, 1934. 388 Yu, K.M., Walukiewicz, W., Wu, J., Beeman, J.W., Ager, J.W., III, Haller, E.E.,
389
390
391
392
393
394
395 396
397
398
399
400
Shan, W., Xin, H.P. and Tu, C.W. (2001) Applied Physics Letters, 78, 1077. Murdin, B.N., Kamal-Saadi, M., Lindsay, A., OReilly, E.P., Adams, A.R., Nott, G.J., Crowder, J.G., Pidgeon, C.R., Bradley, I.V., Wells, J.-P.R., Burke, T., Johnson, A.D. and Ashley, T. (2001) Auger recombination in long-wavelength infrared InNxSb1x alloys. Applied Physics Letters, 78 (11), 1568–1568. Murdin, B.N., Adams, A.R., Murzyn, P., Pidgeon, R., Bradley, I.V., Wells, J.-P.R., Matsuda, Y.H., Miura, N., Burke, T. and Johnson, A.D. (2002) Band anticrossing in dilute InNxSb1x. Applied Physics Letters, 81 (2), 256–256. Ashley, T., Burke, T.M., Price, G.J., Adams, A.R., Andreev, A., Murdin, B.N., OReilly, E.P. and Pidgeon, C.R. (2003) Solid-State Electronics, 47, 387. Baillargeon, J.N., Cheng, K.Y., Hofler, G.E., Pearah, P.J. and Hsieh, K.C. (1992) Applied Physics Letters, 60, 2540. Liu, X., Bishop, S.G., Baillargeon, J.N. and Cheng, K.Y. (1993) Applied Physics Letters, 63, 208. Miyoshi, S., Yaguchi, H., Onabe, K., Ito, R. and Shiraki, Y. (1994) Journal of Crystal Growth, 145, 87. Bi, W.G. and Tu, C.W. (1996) Applied Physics Letters, 69, 3710. Shan, W., Walukiewicz, W., Yu, K.M., Wu, J., Ager, J.W., III, Haller, E.E., Xin, H.P. and Tu, C.W. (2000) Applied Physics Letters, 76, 3251. Wu, J., Walukiewicz, W., Yu, K.M., Ager, J.W., III, Haller, E.E., Hong, Y.G., Xin, H.P. and Tu, C.W. (2002) Physical Review B: Condensed Matter, 65, 241303. Rudko, G.Yu., Buyanova, I.A., Chen, W.M., Xin, H.P. and Tu, C.W. (2002) Applied Physics Letters, 81, 3984. Rudko, G.Yu., Buyanova, I.A., Chen, W.M., Xin, H.P. and Tu, C.W. (2003) SolidState Electronics, 47, 493. Xin, H.P. and Tu, C.W. (2000) Applied Physics Letters, 77, 2180.
j317
j 2 Electronic Band Structure and Polarization Effects
318
401 Xin, H.P., Tu, C.W., Zhang, Y. and Mascarenhas, A. (2000) Applied Physics Letters, 76, 1267. 402 Xin, H.P., Welty, R.J. and Tu, C.W. (2000) Applied Physics Letters, 77, 1946. 403 Zhang, Y., Fluegel, B., Mascarenhas, A., Xin, H.P. and Tu, C.W. (2000) Physical Review B: Condensed Matter, 62, 4493. 404 Leibiger, G., Gottschalch, V., Schubert, M., Benndorf, G. and Schwabe, R. (2002) Physical Review B: Condensed Matter, 65, 245207. 405 Reddy, C.V., Martinez, R.E., II, Narayanamurti, V., Xin, H.P. and Tu, C.W. (2002) Physical Review B: Condensed Matter, 66, 235313. 406 Buyanova, I.A., Pozina, G., Bergman, J.P., Chen, W.M., Xin, H.P. and Tu, C.W. (2002) Applied Physics Letters, 81, 52. 407 Sun, B.Q., Jiang, D.S., Pan, Z., Li, L.H. and Wu, R.H. (2000) Applied Physics Letters, 77, 4148. 408 Polimeni, A., Capizzi, M., Geddo, M., Fischer, M., Reinhardt, M. and Forchel, A. (2001) Physical Review B: Condensed Matter, 63, 195320. 409 Gokhale, M.R., Wei, J., Wang, H. and Forrest, S.R. (1999) Applied Physics Letters, 74, 1287. 410 Zhukov, A.E., Kovsh, A.R., Semenova, E.S., Ustinov, V.M., Wei, L., Wang, J.-S. and Chi, J.Y. (2002) Semiconductors, 36, 899. 411 Choulis, S.A., Hosea, T.J.C., Tomic, S., Kamal-Saadi, M., Weinstein, B.A., OReilly, E.P., Adams, A.R. and Klar, P.J. (2003) Physica Status Solidi b: Basic Research, 235, 384. 412 Sun, H.D., Dawson, M.D., Othman, M., Yong, J.C.L., Rorison, J.M., Gilet, P., Grenouillet, L. and Million, A. (2003) Applied Physics Letters, 82, 376. 413 Bellaiche, L. (1999) Applied Physics Letters, 75, 2578. 414 Al-Yacoub, A. and Bellaiche, L. (2000) Physical Review B: Condensed Matter, 62, 10847. 415 Pinault, M.-A. and Tournie, E. (2001) Applied Physics Letters, 78, 1562.
416 Klar, P.J., Gr€ uning, H., Koch, J., Sch€afer, S., Volz, K., Stolz, W. and Heimbrodt, W. (2001) Physical Review B: Condensed Matter, 64, 121203. 417 Hong, Y.G., Andre, R. and Tu, C.W. (2001) Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 19, 1413. 418 Ungaro, G., Le Roux, G., Teissier, R. and Harmand, J.C. (1999) Electronics Letters, 35, 1246. 419 Harmand, J.C., Ungaro, G., Largeau, L. and Le Roux, G. (2000) Applied Physics Letters, 77, 2482. 420 Harmand, J.-C., Caliman, A., Rao, E.V.K., Largeau, L., Ramos, J., Teissier, R., Travers, L., Ungaro, G., Theys, B. and Dias, I.F.L. (2003) Semiconductor Science and Technology, 17, 778. 421 Bastard, G. (1982) Physical Review B: Condensed Matter, 25, 7584. 422 Marzin, J.Y. (1986) Strained superlattices, in Heterojunction and Semiconductor Superlattices (eds A. Allan, G. Bastard, N. Boccara, M. Lannoo and M. Voos), Springer, Berlin, pp. 161–176. 423 Pollak, F.H. and Cardona, M. (1968) Physical Review B: Condensed Matter, 172, 816. 424 Gil, B., Hamdani, F. and Morkoç, H. (1996) Oscillator strengths for band to band optical processes in GaN epilayers. Physical Review B: Condensed Matter, 54 (11), R7678. 425 Ahn, D. (1996) Optical gain of InGaP and cubic GaN quantum-well lasers with very strong spin–orbit coupling. Journal of Applied Physics, 79 (10), 7731. 426 Uenoyama, T. and Suzuki, M. (1995) Applied Physics Letters, 67, 2527. 427 Weisbusch, C. and Vinter, B. (1991) Quantum Semiconductor Structures: Fundamentals and Applications, Academic Press, San Diego, CA. 428 For a review of electronic states in semiconductor quantum wells, see Bastard, G. and Brum, J.A. (1986) IEEE Journal of Selected Topics in Quantum Electronics, 22, 1625.
References 429 Jeon, J.B., Lee, B.C., Sirenko, Yu.M., Kim, K.W. and Littlejohn, M.A. (1997) Journal of Applied Physics, 82, 386. 430 Ivchenko, E.L. and Pikus, G.E. (1997) Superlattices and other Heterostructures, Springer Series in Solid-State Sciences, 2nd edn, vol. 110, Springer, Berlin. 431 Bigenwald, P., Christol, P., Konczewicz, L., Testud, P. and Gil, B. (1997) Presented at EMRS, Strassburg, France. 432 Bigenwald, P. and Gil, B. (1994) Solid State Communications, 91, 33. 433 Bykhovski, A., Gelmont, B. and Shur, M. (1993) Applied Physics Letters, 63, 2243. 434 Bykhovski, A., Gelmont, B. and Shur, M. (1993) Journal of Applied Physics, 74, 6734. 435 Bykhovski, D., Gelmont, B.L. and Shur, M.S. (1997) Journal of Applied Physics, 81, 6332. 436 Gualtieri, J.G., Kosinski, J.A. and Ballato, A. (1994) IEEE Transactions on Ultrasonics Ferroelectrics and Frequency Control, 41, 53. 437 OClock, G.D. and Duffy, M.T. (1973) Applied Physics Letters, 23, 55. 438 Burns, G. (1985) Solid State Physics, Academic Press, New York, pp. 88–92. 439 King-Smith, R.D. and Vanderbilt, D. (1993) Theory of polarization of crystalline solids. Physical Review B: Condensed Matter, 47 (3), 1651–1651. 440 Resta, R. (1994) Reviews of Modern Physics, 66, 899. 441 Bernardini, F., Fiorentini, V. and Vanderbilt, D. (2001) Accurate calculation of polarization-related quantities in semiconductors. Physical Review B: Condensed Matter, 63, 193201. 442 Bernardini, F., Fiorentini, V. and Vanderbilt, D. (2001) Accurate calculation of polarization-related quantities in semiconductors. Physical Review B: Condensed Matter, 63, 193201. 443 Littlejohn, M.A., Hauser, J.R. and Glisson, T.H. (1975) Applied Physics Letters, 26, 625. 444 Bykhovski, A., Gelmont, B. and Shur, M. (1995) Journal of Applied Physics, 77, 1616. 445 Shur, M., Gelmont, B. and Khan, A. (1996) Journal of Electronic Materials, 25, 777.
446 For a review, see Resta, R. (1994) Macroscopic polarization in crystalline dielectrics: the geometric phase approach. Reviews of Modern Physics, 66, 899, and references therein. 447 See, for example, Kittel, C. (1996) Introduction to Solid State Physics, John Wiley & Sons, Inc; Beam, W. (1965) Electronics of Solids, 7th edn, McGrawHill. 448 Fiorentini, V., Della Sala, F., Di Carlo, A. and Lugli, P. (1999) Effects of macroscopic polarization in III–V nitride multiple quantum wells. Physical Review B: Condensed Matter, 60, 8849–8858. 449 Rode, D.L. (1970) Physical Review B: Condensed Matter, 2, 4036. 450 Bykhovski, A.D., Kaminski, V.V., Shur, M.S., Chen, Q.C. and Khan, M.A. (1996) Applied Physics Letters, 69, 3254. 451 Yamaguchi, S., Kariya, M., Nitta, S., Takeuchi, T., Wetzel, C., Amano, H. and Akasaki, I. (1998) Observation of photoluminescence from Al1 xInxN heteroepitaxial films grown by metalorganic vapor phase epitaxy. Applied Physics Letters, 73, 830. 452 Ambacher, O., Smart, J., Shealy, J.R., Weimann, N.G., Chu, K., Murphy, M., Schaff, W.J., Eastman, L.F., Dimitrov, R., Wittmer, L., Stutzmann, M., Rieger, W. and Hilsenbeck, J. (1999) Journal of Applied Physics, 85, 3222. 453 Bernardini, F. and Fiorentini, V. (2002) Nonlinear behavior of spontaneous and piezoelectric polarization. International Workshop on Physics of Light-Matter Coupling in Nitrides (PLMCN-1), September 26–29, 2001, Rome, Italy, Physica Status Solidi a: Applied Research, 190 (1), 65–73. 454 Al-Yacoub, A., Bellaiche, L. and Wei, S.-H. (2002) Piezoelectric coefficients of complex semiconductor alloys: the case of Ga1 xInxN. Physical Review Letters, 89 (5), 057601. 455 Al-Yacoub, A. and Bellaiche, L. (2001) Piezoelectricity of ordered (Ga0.5In0.5)N alloys. Applied Physics Letters, 79 (14), 2166.
j319
j 2 Electronic Band Structure and Polarization Effects
320
456 Shimada, K., Sota, T., Suzuki, K. and Okumura, H. (1998) Japanese Journal of Applied Physics, Part 2: Letters, 37, L1421. 457 Smith, D.L. and Mailhiot, C. (1988) Journal of Applied Physics, 63, 2717. 458 Wang, J., Jeon, J.B., Sirenko, Yu.M. and Kim, K.W. (1997) IEEE Photonics Technology Letters, 9, 728. 459 Bernardini, F. and Fiorentini, V. (1998) Macroscopic polarization and band offsets at nitride heterojunctions. Physical Review B: Condensed Matter, 57, R9427. 460 Di Carlo, A., Della Sala, F., Lugli, P., Fiorentini, V. and Bernardini, F. (2000) Doping screening of polarization fields in nitride heterostructures. Applied Physics Letters, 76 (26), 3950. 461 Della Sala, F., Di Carlo, A., Lugli, P., Bernardini, F., Fiorentini, V., Scholz, R. and Jancu, J.-M. (1999) Free-carrier screening of polarization fields in wurtzite GaN/InGaN laser structures. Applied Physics Letters, 74 (14), 2002. 462 Di Carlo, A. (1998) Tightbinding Approach to Computational Materials Science (eds P.E.A. Turchi, A. Gonis and L Colombo), Materials Research Society, Pittsburgh, Materials Research Society Symposium Proceedings, 491, 389. 463 Monroy, E., Gogneau, N., Enjalbert, F., Fossard, F., Jalabert, D., Bellet-Amalric, E., Dang, L.S. and Daudin, B. (2003) Molecular-beam epitaxial growth and characterization of quaternary III-nitride compounds. Journal of Applied Physiology, 94 (5), 3121. 464 Aumer, M.E., LeBoeuf, S.F., McIntosh, F.G. and Bedair, S.M. (1999) Applied Physics Letters, 75, 3315. 465 Dimakis, E., Georgakilas, A., Androulidaki, M., Tsagaraki, K., Kittler, G., Kalaitzakis, F., Cengher, D., BelletAmalric, E., Jalabert, D. and Pelekanos, N.T. (2003) Plasma-assisted MBE growth of quaternary InAlGaN quantum well heterostructures with room temperature luminescence. Journal of Crystal Growth, 251 (1–4), 476.
466 Wright, A.F., Leung, K. and van Schilfgaarde, M. (2001) Effects of biaxial strain and chemical ordering on the band gap of InGaN. Applied Physics Letters, 78 (2), 189. 467 Pereira, S., Correia, M.R., Monteiro, T., Pereira, E., Alves, E., Sequeira, A.D. and Franco, N. (2001) Compositional dependence of the strain-free optical band gap in InxGa1xN layers. Applied Physics Letters, 78 (15), 2137. 468 McCluskey, M.D., Van der Walle, C.G., Romano, L.T., Krusor, B.S. and Johnson, N.M. (2003) Effect of composition on the band gap of strained InxGa1xN alloys. Journal of Applied Physics, 93, 4340–4342. 469 Wright, A.F. and Nelson, J.S. (1995) Firstprinciples calculations for zinc-blende AlInN alloys. Applied Physics Letters, 66 (25), 3465. 470 Onuma, T., Chichibu, S.F., Uchinuma, Y., Sota, T., Yamaguchi, S., Kamiyama, S., Amano, H. and Akasaki, I. (2003) Recombination dynamics of localized excitons in Al1 xInxN epitaxial films on GaN templates grown by metalorganic vapor phase epitaxy. Journal of Applied Physics, 94 (4), 2449. 471 Lukitsch, M.J., Danylyuk, Y.V., Naik, V.M., Huang, C., Auner, G.W., Rimai, L. and Naik, R. (2001) Optical and electrical properties of Al1 xInxN films grown by plasma source molecular-beam epitaxy. Applied Physics Letters, 79 (5), 632. 472 Bernardini, F., Fiorentini, V. and Vanderbilt, D. (1997) Polarization-based calculation of the dielectric tensor of polar crystals. Physical Review Letters, 79 (20), 3958–3958. 473 Yu, L.S., Qiao, D.J., Xing, Q.J., Lau, S.S., Boutros, K.S. and Redwing, J.M. (1998) Ni and Ti Schottky barriers on n-AlGaN grown on SiC substrates. Applied Physics Letters, 73 (2), 238. 474 Morkoç, H., Ünl€ u, H. and Ji, G. (1991) Principles and Technology of MODFETS, vol. II, John Wiley & Sons, Ltd, Chichester, UK, p. 317.
References 475 Chu, R.M., Zhou, Y.G., Zheng, Y.D., Han, P., Shen, B. and Gu, S.L. (2001) Influence of doping on the two-dimensional electron gas distribution in AlGaN/GaN heterostructure transistors. Applied Physics Letters, 79 (14), 2270. 476 Ambacher, O., Foutz, B., Smart, J., Shealy, J.R., Weimann, N.G., Chu, K., Murphy, M., Sierakowski, A.J., Schaff, W.J., Eastman, L.F., Dimitrov, R., Mitchell, A. and Stutzmann, M. (2000) Journal of Applied Physics, 87, 334. 477 Ridley, B.K., Ambacher, O. and Eastman, L.F. (2000) Semiconductor Science and Technology, 15, 270. 478 Reale, A., Massari, G., Di Carlo, A., Lugli, P., Vinattieri, A., Alderighi, D., Colocci, M., Semond, F., Grandjean, N. and Massies, J. (2003) Comprehensive description of the dynamical screening of the internal electric fields of AlGaN/GaN quantum wells in time-resolved photoluminescence experiments. Journal of Applied Physics, 93 (1), 400. 479 Fiorentini, V., Bernardini, F. and Ambacher, O. (2002) Evidence for nonlinear macroscopic polarization in III–V nitride alloy heterostructures. Applied Physics Letters, 80 (7), 1204. 480 G€ orgens, L., Ambacher, O., Stutzmann, M., Miskys, C., Scholz, F. and Off, J. (2000) Applied Physics Letters, 76, 577. 481 Grandjean, N., Damilano, B., Dalmasso, S., Leroux, M., La€ ugt, M. and Massies, J. (1999) Journal of Applied Physics, 86, 3714. 482 Langer, R., Simon, J., Ortiz, V., Pelekanos, N.T., Barski, A., Andre, R. and Godlewski, M. (1999) Applied Physics Letters, 74, 3827. 483 Kim, H.S., Lin, J.Y., Jiang, X.H., Chow, W.W., Botchkarev, A. and Morkoç, H. (1998) Applied Physics Letters, 73, 3426. 484 Vaschenko, G., Patel, D., Menoni, C.S., Ng, H.M. and Cho, A.Y. (2002) Nonlinear
485
486
487
488
489
490
491
492
493
494
macroscopic polarization in GaN/ AlxGa1xN quantum wells. Applied Physics Letters, 80 (22), 4211. Simon, J., Langer, R., Barski, A., Zervos, M. and Pelekanos, N.T. (2001) Physica Status Solidi a: Applied Research, 188, 867. Perlin, P., Suski, T., Lepkowski, S., Teisseyre, H., Grandjean, N. and Massies, J. (2001) Physica Status Solidi a: Applied Research, 188, 839. Wetzel, C., Takeuchi, T., Amano, H. and Akasaki, I. (1999) Japanese Journal of Applied Physics, 38, L163. Shi, C., Asbeck, P.M. and Yu, E.T. (1999) Piezoelectric polarization associated with dislocations in wurtzite GaN. Applied Physics Letters, 74, 573. Cartney, M.R., Ponce, F.A., Cai, J. and Bour, D.P. (2000) Mapping electrostatic potential across an AlGaN/InGaN/AlGaN diode by electron holography. Applied Physics Letters, 76, 3055. Cherns, D., Jiao, C.G., Mokhtari, H., Cai, J. and Ponce, F.A. (2002) Physica Status Solidi b: Basic Research, 234, 924. Cai, J. and Ponce, F.A. (2002) Determination by electron holography of the electronic charge distribution at the threading dislocation in epitaxial GaN. Physica Status Solidi a: Applied Research, 192, 407. Barghout, K. and Chaudhuri, J. (2004) Calculation of residual thermal stress in GaN epitaxial layers grown on technologically important substrates. Journal of Materials Science, 39 (18), 5817–5823. Morkoç, H., Ünl€ u, H. and Ji, G. (1991) Fundamentals and Technology of MODFETs, vols I and II, John Wiley & Sons, Ltd, Chichester, UK. Olsen, G.H. and Ettenberg, M. (1977) Journal of Applied Physics, 48, 2543.
j321
j323
3 Growth and Growth Methods for Nitride Semiconductors Introduction
Although the synthesis of GaN goes to back more than a half century, there are several pivotal developments, which, in the opinion of the author, are responsible for laying the technological framework and paving the way for the tremendous commercial and scientific interest in nitrides. They are as follows: synthesis of AlN by Tiede et al. [1], synthesis of GaN through the reaction of Ga, and ammonia by Johnson et al. [2] synthesis of InN by Juza and Hahn [3], epitaxial deposition of GaN using the hydride VPE technique by Maruska and Tietjen [4] employment of nucleation buffer layers by Amano et al. [5] and Yoshida et al. [6] and achievement of p-type GaN by Akasaki et al. [7]. A more recent development that paved the way for all the commercial activity is the preparation of high-quality InGaN by Nakamura et al. [8], which followed the synthesis of InGaN by Osamura et al. [9]. Nearly every crystal-growth technique, substrate-type, and orientation has been tried in an effort to grow high-quality group III–V nitride thin films. Increasingly, researchers have successfully taken advantage of the hydride vapor phase epitaxy (HVPE), organometallic vapor phase epitaxy (OMVPE), and molecular beam epitaxy (MBE) techniques, which have yielded greatly improved film quality. All of these epitaxial methods must contend with two main problems: the lack of native GaN substrates and difficulty with nitrogen incorporation and concomitant high ammonia flow rates needed particularly for In-containing nitride semiconductors. A major drawback of GaN is that native substrates are not yet available in large quantities. This is, in part, owing to the low solubility of nitrogen in bulk Ga and the high vapor pressure of nitrogen over GaN at the growth temperature of bulk crystals. The best alternatives now lie in development of sapphire, SiC, or AlN substrates. Interest in AlN substrates has increased recently owing to the closer lattice match over sapphire, matched stacking order, and high thermal conductivity. These factors make AlN one of the best choices for growth of detectors requiring high AlN content AlGaN and backside illumination. Early work on producing bulk AlN looked promising [10]. The problem of nitrogen is endemic in epitaxial deposition techniques as well. Regardless of the growth method employed, the major difficulty in growing group III Handbook of Nitride Semiconductors and Devices. Vol. 1. Hadis Morkoc Copyright 2008 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim ISBN: 978-3-527-40837-5
j 3 Growth and Growth Methods for Nitride Semiconductors
324
nitrides arises from the need to incorporate stoichiometric quantities of nitrogen into the film. This is accomplished in vapor phase processes at high substrate temperatures by decomposing a nitrogen-containing molecule, such as ammonia, on the substrate surface. It should be noted that as the N vapor pressure increases, going from AlN to GaN and then to InN, the ammonia flow rate must be increased reaching tens of liters per minute for InGaN and V/III ratios of 200 000 for InN. It can also be accomplished at lower temperatures in MBE growth by increasing the reactivity of nitrogen through remote plasma excitation or ionization or using ammonia as the reactive nitrogen source. A great deal of effort has been spent in trying to overcome the problems arising from lack of GaN substrates. Some of the best device results have been achieved through use of buffer layers such as thick HVPE, including freestanding, variety, or lateral epitaxial overgrowth variety.
3.1 Substrates for Nitride Epitaxy
As there is lack of a commercial native substrate, a plethora of substrates have been employed in the growth of GaN films [11–14]. Recognizing the thermodynamic bottleneck with regard to native substrates, unconventional methods employing various implementations of compliant substrates and lateral epitaxy have been explored. The most promising results on more conventional substrates so far have been obtained on sapphire and SiC, with SiC making substantial inroads. Also coming on the scene are thick freestanding GaN templates grown by HVPE and then separated from the sapphire substrates. GaN, AlN, and InN have been grown primarily on (0 0 0 1) sapphire but also on the ð2 1 3 1Þ, ð1 1 0 1Þ, ð1 1 0 2Þ, and ð1 1 2 0Þ surfaces. In addition, III–V nitrides have been grown on Si, NaCl, GaP, InP, SiC, W, ZnO, MgAl2O4, TiO2, and MgO. Other substrates as well have been used for nitride growth, including Hf and LiAlO2, and LiGaO2. Table 3.1 is a compilation of the lattice parameters and thermal characteristics of a number of prospective substrate materials for nitride growth. Lattice-mismatched substrates lead to a substantial density of misfit and threading dislocations (in the range of 108 and 1010 cm2), though selective epitaxy followed by coalescence, which goes by many names such as epitaxial lateral overgrowth (ELO), lateral epitaxial overgrowth (LEO), and epitaxial lateral overgrowth (ELOG), is a promising method for reducing dislocations down to 106 cm2. In comparison, the extended defect densities are essentially zero for silicon homoepitaxy and 102–104 cm2 for gallium arsenide homoepitaxy [13]. Additional crystalline defects besetting the layers include inversion domain boundaries (IDBs) and stacking faults. Such defects are either directly or indirectly responsible for the creation of nonradiative recombination centers, which manifest themselves as energy states in the forbidden energy bandgap reducing the quantum efficiency as well as producing scattering centers. Other adverse effects of structural and point defects are that impurities diffuse more readily along threading dislocations, and carrier transport is
3.104 3.189 4.758
AlN (hexagonal) GaN (hexagonal) Al2O3 (sapphire) (rhombohedral) 4H-SiC (hexagonal) 6H SiC (hexagonal) ZnO (hexagonal) ScAlMgO4 (hexagonal) c-LiAlO2 (tetragonal) LiGaO2 (orthorhombic) MgAl2O4 (cubic/spinel) Si (cubic) GaAs (cubic) b-SiC (cubic) MgO (cubic/rock salt) 6.372
—
Conventional b (Å)
10.053 15.1123 5.2065 25.195 6.2679 5.407
4.966 5.175 12.991
c (Å)
3.1340 6.372
3.0817 3.2496
3.104 3.19 2.747
Matched a (Å)
1.5 0.5 4.9
4.9 0.3–0.4
3.2 2.3 0.3–0.5
Thermal conductivity, j (W cm1 K1)
10.5
4.2, 4.68 4.75, 2.9 6.2, 12.2 7.1, 15 a ¼ 6, b ¼ 9, 7 7.45 3.59 6
4.2, 5.3 5.59, 3.17 7.5, 8.5
Da/a, Dc/c (·106 K1)
In part from Landolt-B€ ornstein, vol. 17, Springer, New York, 1982. ZnO data are from W. Harsch of Eagle Picher.
3.073 3.0817 3.2496 3.246 5.1687 5.402 8.083 5.4301 5.6533 4.36 4.216
a (Å)
Crystal
Table 3.1 Lattice parameters and thermal characteristics of a number of the prospective substrate materials for nitride growth and their lattice mismatch with GaN.
P63mc P63mc R3m P41212 Pna21 Fd3m Fd3m F 43m F 43m Fm3m
P63mc P63mc R3c
Space group
3.63% 3.36% þ1.9% þ1.8% 1.7% 0.18%
2.7% 0% 49% (13%)
Mismatch
3.1 Substrates for Nitride Epitaxy
j325
j 3 Growth and Growth Methods for Nitride Semiconductors
326
either impeded, as in lateral transport, or aided, as in vertical transport. The high density of defects also leads to boundary-limited transport making the important basic parameters such as diffusion constant and mobility nearly impossible to measure. The extended defects in nitrides also lead to inhomogeneities in electric potential because of the high piezoelectric constants of GaN. Electrically active defects induced either directly or indirectly by extended defects cause excess leakage that is detrimental to both optical devices (in the form of dark current in detectors and reduced quantum efficiency in emitters) and electrical devices (in the form of increased gate current leakage and output conductance in field-effect transistors). Details regarding the piezoelectric properties can be found in Chapter 2. Lattice-mismatched substrates are commonly used at present in spite of efforts to produce GaN [15,16] and AlN [10] bulk materials. Despite the lack of matched substrates, remarkable progress in the growth of high-quality epitaxial III nitride films has been achieved by a variety of methods such as hydride vapor phase epitaxy (inorganic VPE or hydride VPE also goes with the acronym HVPE) [11,17], OMVPE [18], and reactive molecular beam epitaxy (RMBE) [19,20]. Moreover, thick freestanding GaN templates for further epitaxy have been prepared by HVPE [21]. By far the most frequently used methods are the VPE methods with heterojunction capability. The most versatile among the VPE methods is the metalorganic chemical vapor deposition (OMVPE). OMVPE is the primary method employed in the investigation and production of optoelectronic devices, such as LEDs and lasers, albeit the quality of MBE films comes close to that grown by OMVPE. Electronic devices with higher quality interfaces are achieved principally with OMVPE and MBE. Inorganic VPE was the first method used to grow epitaxial III-N semiconductors but was nearly abandoned [22]. The technique, however, got revived recently by growing very high-quality and thick GaN buffer layers and templates [23] for the growth of device structures using MBE or OMVPE [11]. Efforts are underway to expand the method to the growth of AlGaN. Below, a discussion of the class of substrates that have been explored is followed by a discussion of the properties of and processing steps for the conventional substrates before growth for each of these substrates. 3.1.1 Conventional Substrates
GaN, as the most studied member of the semiconducting group III nitrides, has been grown on many substrates. Many of the major problems that have hindered the progress in GaN and related semiconductors can be traced back to the lack of a suitable substrate material that is lattice and thermally matched to GaN. Lattice mismatch is responsible for stacking faults and dislocations. Thermal mismatch causes the epilayer to crack on cooling. Specifically, the semiconductors GaN, AlN, and InN have been grown primarily on sapphire, most commonly in the c (0 0 0 1) orientation but also on the a- ð1 1 2 0Þ and R- ð1 1 0 2Þ planes [14]. Growth on a-plane produced c-plane GaN, but growth on R-plane sapphire produces a nonpolar, a-plane
3.1 Substrates for Nitride Epitaxy
GaN. In addition, the group III–V nitrides have been grown on SiC, ZnO, MgAl2O4, Si, GaAs, MgO, NaCl, W, and TiO2. As high-resistivity SiC substrates became available, it became the proffered substrate for transistor work, in part owing to its high thermal conductivity. In addition, freestanding GaN prepared by HVPE is at or nearing production capacity primarily for low threshold lasers that need low-defect material. Some of the suitable substrate materials have become commercially available only recently. Almost all the group III–V nitride semiconductors have been deposited on sapphire despite its poor structural and thermal match to the nitrides. The preference for sapphire substrates can be ascribed to its wide availability, hexagonal symmetry, and ease of handling and pregrowth cleaning. Sapphire is also stable at high temperatures (1000 C) required for epitaxial growth using the various CVD techniques commonly employed for GaN growth. Owing to thermal and lattice mismatches between sapphire and the group III–V nitrides, it is necessary to grow a thick epilayer to obtain good-quality material. 3.1.2 Compliant Substrates
When large mismatch exists between an epilayer and its substrate, the misfit is typically accommodated by the introduction of misfit dislocations at the interface, which are accompanied by threading dislocation segments in the epilayer. To overcome this problem, a compliant substrate is used in high-misfit systems. The role of the compliant substrate is to accommodate the large mismatch either by plastic deformation of the compliant substrate in a manner that avoids the formation of dislocations in the heteroepitaxial film or by homogeneous elastic strain of the threading dislocation, which also avoids formation of the threading dislocations. Both mechanisms are facilitated by a compliant substrate whose stiffness constants are well below those of the epilayer and the supporting bulk substrate. As for the former mechanism, the soft and thin nature of the compliant substrate energetically favors the capture of dislocations resulting from the mismatch by the substrate rather than by the stiffer epilayer, thus paving the way for predilection toward misfit accommodation (MA) by homogeneous elastic strain rather than misfit dislocations. An attractive approach is to insert a pillarlike interfacial layer that is capable of accommodating thermal strain, the effectiveness of which depends on the height of the pillars and the size of the wafer. An effort has been made to find a universal substrate onto which any epitaxial layer can be grown with a very low density of structural defects. Some experimental success has been achieved, but only in specific cases, since 1991 when Lo and colleagues [24,25] introduced the basic idea of a compliant substrate. Different kinds of epitaxial layers have been grown on compliant substrates of GaAs twist bonded to bulk substrates of GaAs. Among them are InGaP, In0.22Ga0.78As, GaSb, and InSb where the respective misfits with GaAs are 1, 1.5, 8, and 15%, respectively [26,27], with positive figures denoting tensile stresses in the layer on GaAs. The use of a thin compliant substrate is not limited to the epitaxial growth of III–V compounds but may also be found in SiGe grown on a thin compliant
j327
j 3 Growth and Growth Methods for Nitride Semiconductors
328
substrate of Si on a viscous SiO2 layer [28]. Other approaches pursue compliancy by twist bonding, that is, deposition of a low melting temperature interlayer. Twist bonding Si to SOI provides a high density of interlayer dislocations that can elastically accommodate misfit between the compliant substrate and the heteroepitaxial film. Otherwise, if the bonding between the compliant layer and substrate is strong, an array of screw dislocations form, transforming to edge/screw dislocations upon deposition of a lattice-mismatched film by elastic deformation. Various explanations have been offered to understand the mechanisms of misfit accommodation [29]. The term compliant is used liberally to describe an approach or a set of approaches to grow lattice mismatch materials where the substrate or some interface layer accommodates the mismatch by expansion/contraction or generating defects within itself as opposed to epitaxial layers. Specifically, the concept behind the latter case is to force the defects caused by mismatch to propagate into the substrate as opposed to the epilayer, requiring generation of misfit dislocations in the thin, weakly bound template layer rather than the growing epitaxial layer. In case of nitride, the initial approach proposed was that Si on an insulator be utilized for GaN growth. Here, a thin Si layer on silicon dioxide, which, in turn, is on Si would be employed. The downside is that the quality of GaN on compliant Si has been poor at best. The modified approach to overcome this barrier is to carbonize the Si to convert it to SiC. If one utilizes the (1 1 1) orientation, one would then get the wurtzitic phase of GaN. To be specific, Yang et al. [30] suggested that before the beginning of the carbonization process, SiO2, Si, and C need to be deposited successively on a Si substrate. By exposing the new composite substrate to a flux of acetylene or carbon particles at 900 C, a thin layer (less than 50 nm) of Si (on SiO2) will be partially or completelyconverted into SiC. GaNis then grownonthis SiC. Again, the problemhere, setting aside the problems associated with the growth on SiC, is that SiC so formed is not contiguous and is extremely defective both in terms of bulk and surface structural properties. In addition, air gaps form beneath the layer surface. Consequently, this technique has not yet lived up to the original proposal and expectations. Efforts still continue to exploit this approach despite the lack of progress so far. Compliance based on expansion and/or contraction is a very neat idea and may be workable for small-sized wafers. However, it is impractical for larger wafers. For example, if the lattice mismatch between the compliant substrate is 4% (epitaxial layer having the larger lattice constant) and the wafer is 50 mm, an expansion in the substrate required for producing defect-free epitaxial layer is 2 mm, which is substantial and unlikely. 3.1.3 van der Waals Substrates
To get around the lattice-mismatch problem, a new growth method called van der Waals epitaxy has been proposed [31], which delivers strain-free films. In this approach, the substrate and epitaxial film are separated by an intermediate epitaxial two-dimensional (2D) buffer material such as MoS2, WS2, or other materials such as II–VI (ZnTe) or III–VI compounds (GaSe, InSe, etc.) having weak van der Waals
3.2 A Primer on Conventional Substrates and their Preparation for Growth
bonding to the substrate and the film. Strain from lattice mismatch between the epitaxial film and the substrate is completely relieved in the region between the layer and the buffer. As in the case of the compliant substrate scheme, this approach has not been very successfully applied to nitrides.
3.2 A Primer on Conventional Substrates and their Preparation for Growth
A substrate is like the foundation of a building. As such, substrate preparation deserves the mostattention.Though thedetailsof the proceduresemployed vary fromone growth method to the next, a chemical preparation before loading into the growth reactor is common. In the OMVPE technique, this is followed by either a simple heat treatment or a combination of heat treatment with gas-phase etching, where temperatures for heat treatment in the vicinity of 1200 C are possible. In the case of vacuum-deposition techniques where it is not always possible to achieve sufficiently high temperatures, dry processing techniques are employed. One of the dry processing techniques is utilizing ECR remote plasma etching with a mixture of hydrogen and helium, as discussed below. The purpose of the He gas is to take advantage of its energetic metastable states with long mean free paths. In addition to a clean surface, the goal is to get as flat a surface as possible, because the nitride stacking order, AaBbAaBb, is different from the stacking order found in most of the substrates under investigation. The exception is ZnO whose stacking order matches that of the nitrides. Because the atomic steps on the (0 0 0 1) surface are of the bilayer type, the surface terraces would have the same surface polarity so that stacking mismatch boundaries (SMB) can be avoided. The surfaces of the substrates used have to be prepared for epitaxial growth, a process that includes degreasing followed by chemical etching when possible. Surfaces of as-received sapphire and SiC substrates contain mechanical polishing damage that must be removed. Chemical etches are not yet available for this purpose. Consequently, a high-temperature treatment under a controlled environment is employed, as will be discussed below. The degreasing procedure, which is the first step for growth, for Si, sapphire, SiC, ZnO, LiGaO2 and LiAlO2, and GaN and AlN, whose specifics will be discussed below, is the same. The substrate is first dipped in a solution of trichloroethane (TCE) kept at 300 C, for 5 min. It is then rinsed for 3 min each in acetone and methanol. This is followed by a 3-min rinse in deionized (DI) water. The above process is repeated three times to complete the degreasing process. The substrates are then etched, which is a substrate dependent procedure. Following degreasing, a variety of substrate-specific methods are employed to get various substrates growth ready as discussed below. 3.2.1 GaAs
GaAs, as a substrate for GaN epitaxy, is justified on the premise of obtaining pure (wurtzite-free) zinc blende GaN on GaAs(1 0 0), attaining thick wurtzite GaN films on
j329
j 3 Growth and Growth Methods for Nitride Semiconductors
330
GaAs(1 1 1) substrates, and dilute GaAsN films for infrared applications. The bulk of the research on GaAs based nitrides is on (1 0 0) surface for cubic GaN. GaAs(1 0 0) is one of the few semiconductor substrates on which metastable zinc blende GaN epitaxial films readily form, and many researchers have investigated the best ways of avoiding any inclusion of the wurtzite polytype into these films. GaAs is much more readily wet etched than any of the other substrates used for nitride epitaxy, which also makes GaN films easier to separate from GaAs than sapphire. Thus, GaAs(1 1 1) substrates are considered a template for creating freestanding thick GaN films for subsequent epitaxy. Because the decomposition rate of GaAs in NH3 or an ultrahigh vacuum (UHV) rapidly increases at temperatures above 700 C, a deposition process including multiple temperatures is required. Unless the substrate temperature is high, the maximum growth rate (GR) attainable is limited. Moreover, even a small amount of GaAs decomposition could be detrimental, as surface roughening or faceting enhances the onset of mixed polarity growth. Because MBE is capable of depositing epitaxial GaN films at a lower temperature as compared to vapor phase methods, it has been more commonly employed in this respect. The maximum allowed temperature could be increased once the GaAs substrate is completely encased with GaN deposited at a low temperature (LT), thereby making OMVPE and HVPE more viable. 3.2.1.1 A Primer on GaAs GaAs has the zinc blende crystal structure with the symmetry group of F 43m. Figure 3.1 displays the perspective view of the GaAs crystal along [1 0 0], [1 1 0], and [1 1 1] directions. Table 3.2 lists the physical, chemical, thermal, mechanical, and optical properties of GaAs important for the GaN epitaxy. For additional details, see Liu and Edgar [13]. GaAs substrates are grown with either liquid encapsulated Czochralski (LEC) or vertical gradient freeze (VGF) methods. GaAs has seen a spectacular improvement over a period of two decades in that crystal defects, impurities, and micro-inhomogeneities have been reduced. In fact, GaAs wafers with diameters greater than
Figure 3.1 The perspective view of the GaAs crystal (a) along [1 0 0] (1 · 1 · 1 unit), (b) [1 1 0] (2 · 2 · 2 units), and (c) [1 1 1] (2 · 2 · 2 units) directions [13]. (Please find a color version of this figure on the color tables.)
3.2 A Primer on Conventional Substrates and their Preparation for Growth Table 3.2 Properties of GaAs at room temperature (partially after Ref. [13]).
Parameter
Value
Lattice constant (Å) Density (g cm3) Melting point ( C) Heat capacity (J g1 K1) Thermal conductivity (W cm1 K1) Thermal diffusivity (cm2 s1) Thermal expansion (linear) (·106 K1) Percent change in lattice (300–1200 K) Bulk modulus (GPa) Youngs modulus (GPa) Poissons ratio Refractive index Relative dielectric constant Electrical resistivity (undoped)
5.6536 5.32 1240 0.327 0.45 0.26 6.03 Da/a0 ¼ 0.5876 75.0 85.5 0.31 3.66 near band edge e0 ¼ 13.1 1.0 · 104 O cm, nonstoichiometric defect compensated
150 mm and with various doping types and concentrations are commercially available. Silicon and tellurium are common n-type dopants with resultant electron concentrations in the range of 1016–1018 cm3. On the contrary, zinc is the standard p-type dopant with resultant hole concentrations in the range of 1018–1019 cm3. Commercial LEC GaAs has a typical etch pit density (EPD) of less than 104 cm2 and an electron mobility greater than 4000 cm2 V1 s1. But, VGF grown GaAs offers a lower defect density, with EPD less than 103 cm2. Both (1 0 0) and (1 1 1) types of GaAs with different vicinal degrees are available. 3.2.1.2 Surface Preparation of GaAs for Epitaxy The substrate preparation, following degreasing, includes etching in acid such as H2SO4 : H2O2 : H2O. After rinsing, the surface can be treated in dilute HF for H passivation of the surface. The H passivation layer can be desorbed in the deposition chamber. However, owing to the well-advanced nature of GaAs technology, epi-ready substrates with protective oxides are commercially available. Once in the deposition reactor, the oxide layer can be removed by thermal desorption. Direct nucleation of GaN on GaAs is difficult owing to great chemical and mechanical mismatch with GaN (the large lattice match of about 20% caused by 5.65 Å vs. 4.51 Å for zinc blende GaN), but it is a task mitigated somewhat with the deposition of a GaAs prelayer. Zinc blende GaN on (2 · 4) GaAs can be grown by first establishing an atomically smooth GaAs surface with minimized step density. This is accomplished by depositing a GaAs prelayer, on the order of 100 nm to as high as 1 mm, and choosing appropriate nucleation conditions. Without a GaAs prelayer, the more stable wurtzite polytype of GaN grows preferentially once (1 1 1) facets of the GaN film are generated on the rough substrates. The content of hexagonal GaN phase
j331
j 3 Growth and Growth Methods for Nitride Semiconductors
332
is drastically reduced with an epitaxial GaAs prelayer in MBE growth. The growth temperature for a GaAs prelayer is typically about 600 C. It is imperative that the GaAs surface is nitridated prior to GaN growth for chemical and mechanical transitioning. Nitridation of the GaAs substrate results in a thin nitride surface layer, which provides a cubic template for growth, improving the quality of the GaN layer and suppressing GaAs decomposition at higher substrate temperatures needed for GaN epitaxy. GaAs substrates are nitridated either by exposure to nitrogen plasma (MBE) or by annealing in ammonia (OMVPE and RMBE). At temperatures below 200 C, the nitridation is hindered by kinetic limitations. At temperatures above 600 C, simultaneous etching of the surface may occur along with the nitridation process. Nitridation of GaAs(0 0 1) does not take place homogeneously but proceeds along {1 1 1} facets into the underlying GaAs layer. Complete nitridation can lead to a highly facetted interface between the GaN layer and GaAs substrate, which, in turn, leads to the nucleation of the wurtzite phase and could lead to polycrystalline GaN. Using the highest growth rate possible to quickly bury the interface or maintaining an As flux during growth of the first few monolayers of GaN helps to prevent the deterioration of the GaAs surface. A buffer layer of GaN at low temperature following the initial nitridation generally improves the eventual quality. AlN is not used as buffer because its zinc blende polytype is extremely difficult to nucleate. 3.2.2 Si
Si is the most perfected and least expensive substrate that is available in sizes up to 300 mm. Unlike GaAs, ZnO, and a few others, silicon has good thermal stability under conditions used for GaN epitaxy. However, Wz GaN and AlN grown on Si(1 1 1) are highly defective. The incentives for using Si substrates remain high, however, and good progress in reducing the defect density by using epitaxial lateral overgrowth or pendeo-epitaxy has been reported. 3.2.2.1 A Primer on Si Si has a diamond-lattice structure with the space group of Fd 3m (No. 227) and can be thought of as two interpenetrating fcc sublattices with one sublattice displaced from the other by one quarter of the distance along a body diagonal of the cube (i.e., the pffiffiffi displacement of a 3=4, where a ¼ 5.43102 Å is the lattice constant). Each atom in the lattice is surrounded by four equidistant nearest neighbors that lie at the corners of a tetrahedron. Figure 3.2 illustrates the perspective view along the [0 0 1], [0 1 1], and [1 1 1] directions of a Si cell. Table 3.3 lists physical, chemical, thermal, mechanical, and optical properties of Si at room temperature. Single crystalline ingots are produced by the Czochralski (CZ) method (over 85% of silicon crystals are grown by this method) or the Float Zone (FZ) method, used mostly for purification. These ingots eventually become thin Si wafers through the processes of shaping, slicing, lapping, etching, polishing, and cleaning. Impurities
3.2 A Primer on Conventional Substrates and their Preparation for Growth
Figure 3.2 The perspective view along (a) the [0 0 1], (b) [0 1 1], and (c) [1 1 1] directions of a Si cell. (Please find a color version of this figure on the color tables.)
can be added directly to the melt to create p-type and n-type silicon. The only discernible half-drawback of the CZ method is that oxygen (typical at a level of 1018 cm3) and carbon (typical at a level of 1016 cm3) can be incorporated because of the reduction of the quartz crucible and contamination by graphite fixtures. It should be noted that these impurities are not without benefits in that oxygen increases the yield strength or acts as internal getter to tie up metallic contaminants. The FZ method does not use any crucible, and thus the impurity level is markedly reduced, making it easier to grow high-resistivity material. For more details, refer to Ref. [13]. 3.2.2.2 Surface Preparation of Si for Epitaxy For wurtzitic GaN growth, (1 1 1) plane Si is used. The (0 0 1) surface is also used for cubic GaN growth, albeit in only a few cases. As-received Si surface is already
Table 3.3 Properties of Si at room temperature (partially after Ref. [13]).
Parameter
Value
Lattice constant (Å) Density (g cm3) Melting point ( C) Heat capacity (J g1 K1) Thermal conductivity (W cm1 K1) Thermal diffusivity (cm2 s1) Thermal expansion (linear) Percent change in lattice (298–1311 K) Shear modulus (GPa) Bulk modulus (GPa) Youngs modulus (GPa) Poissons ratio Refractive index Relative dielectric constant Electrical resistivity (undoped)
5.43102 2.3290 1410 0.70 1.56 0.86 2.616 · 106 K1 Da/a0 ¼ 0.3995 680 97.74 165.6 0.218 3.42 e0 ¼ 11.8 Up to 50 kO cm
j333
j 3 Growth and Growth Methods for Nitride Semiconductors
334
excellent and removal of only a very thin surface layer, using the RCA etch followed by hydrogenation of surface dangling bonds, is sufficient. This is accomplished by immersing Si for 10 min in a 1 : 1 : 5 solution of HCl : H2O2 : H2O kept at 60 C, which grows a porous oxide, followed by a rinse in deionized water. The resulting oxide layer is then removed by dipping the substrate in a 10 : 1 solution of H2O : HF for 20 s. The hydrogenation process takes place through a short exposure of the wafer to an HF solution. 3.2.3 SiC
The cohesive bond strength of SiC is so large that it was once considered an element under the name of carborendum. Owing to its large thermal conductivity and dearth of defect causing in-plane rotation of GaN with respect to SiC lattice, and availability of high-resistivity substrates, SiC is continually gaining recognition as a very viable substrate for epitaxy for both optical and electronic devices. Much of the early drawbacks having to do with pre-epitaxy surface preparation, micropipes, size, and, to an extent, cost issues have been mitigated to the point that some commercial LEDs and almost all of high power field-effect transistors utilize nitride heterostructures on SiC. 3.2.3.1 A Primer on SiC A basic unit of crystalline SiC is a covalently bonded tetrahedron of C atoms with a Si atom at its center or vice versa, that is, either SiC4 or CSi4, as illustrated in Figure 3.3. Variation in the stacking order of SiC along the c-direction leads to more than 250
Figure 3.3 Tetragonal bonding of a carbon atom with its four nearest silicon neighbors. The bond lengths depicted with a and C–Si (the nearest neighbor distance) are approximately 3.08 and 1.89 Å, respectively. The right side is the three-dimensional structure of 2H-SiC structure. (Please find a color version of this figure on the color tables.)
3.2 A Primer on Conventional Substrates and their Preparation for Growth
A B C
Carbon Si Base
B A
C
A
B
C
A
C
A
A
C
B
B
B
A
A
B A
A 3C
A
2H
4H
6H
Figure 3.4 Stacking sequence of cubic and three polytypes of wurtzitic SiC.
polytypes, of which a few prominent ones are shown in Figure 3.4. (A basic discussion of stacking is given in Section 1.1.) By observing the SiC crystal from the side, the stacking sequence can be projected as in Figure 3.5. The distance a between neighboring silicon or carbon atoms is approximately 3.08 Å for all polytypes. The height of the unit cell c varies with the different polytypes, as tabulated in Table 3.4. Consequently, the c/a ratio varies from polytype to polytype but is always close to the ideal for a close packed structure. This ratio is approximately 1.641, 3.271, and 4.908 for the 2H-, 4H-, and 6H-SiC polytypes, respectively, whereas theffiffiffiffiffiffiffi equivapffiffiffiffiffiffiffiffi p ffi lent ideal ratios for these polytypes are 1.633, 3.266, and 4.899 ( 8=3 , 2 8=3 , and pffiffiffiffiffiffiffiffi 3 8=3), respectively [13]. Each polytype has a unique set of electronic and optical properties. The bandgaps at liquid helium temperature of the different polytypes range between 2.39 eV for 3CSiC and 3.33 eV for the 2H-SiC polytype. The two most important polytypes as substrates for GaN epitaxy, 6H-SiC and 4H-SiC, have bandgaps at liquid helium temperature of 3.02 and 3.27 eV, respectively. The hexagonal polytypes of SiC, such as 4H- and 6H-SiC, belong to the same space group, P63mc (No. 186), as wurtzite GaN. The most studied substrates for GaN epitaxy are the 3C-SiC/Si(1 0 0) and 6H-SiC, as these polytypes have been the most readily prepared or commercially available for the longest time. With 4H-SiC now commercially available, its use will become more common. Table 3.4 shows the physical, chemical, thermal, mechanical, and optical properties of SiC at room temperature. The thermal expansion coefficient of SiC in c- and a-planes as a function of temperature is shown in Figure 3.6. Bulk SiC crystals are produced by sublimation in the modified Lely process, developed by Tairov and Tsvetkov [33], which employs a SiC seed crystal for the control of polytype and orientation. Growth is achieved by the vapor transport of Si, Si2C, and SiC2 driven by a temperature difference in an argon atmosphere in a graphite, tantalum, or tantalum carbide crucible at 20–500 Torr and at about 2200 C. The 4H-SiC and 6H-SiC(0 0 0 1) varieties both on- and off-axis (typically 3.5 for 6HSiC and 8 for 4H-SiC), silicon and carbon face, are available in sizes up to 100 mm in
j335
j 3 Growth and Growth Methods for Nitride Semiconductors
336
Figure 3.5 Views of the ½1 1 2 0 planes for the 3C-, 2H-, 4H-, and 6H-SiC polytypes.
diameter. Screw dislocations occur in high densities and depending on the magnitude of their Burgers vector, the core of a screw dislocation can be hollow (nanopipes or micropipes) or closed and run through the entire wafer. Hollow core screw dislocations take place when the Burgers vector (b) is two or more times the c-lattice constant (c) for 6H-SiC or three times the lattice constant for 4H-SiC. Because the best micropipe density has been reduced to about 1 cm2, research has shifted on to closed-core screw dislocations, which occur in densities of approximately 103–104 cm2. Both 4H-SiC and 6H-SiC wafers are available in low resistivity nand p-type forms with concentrations in the range of 1015–1019 cm3. The resistivities for n- and p-type material for the aforementioned doping range lie in the range 0.01–0.10 and 1–10 O cm, respectively. Interest in semi-insulating SiC is driven by
3.2 A Primer on Conventional Substrates and their Preparation for Growth Table 3.4 Properties of SiC at room temperature (after Ref. [13]).
Parameter
Polytype
Value
Lattice constant (Å)
3C 2H 4H 6H 3C 2H 6H 3C 6H 3C 4H 6H 3C 6H 6H 3C 3C Ceramic 3C 2H 4H 6H 3C 6H
a ¼ 4.3596 a ¼ 3.0753, c ¼ 5.0480 a ¼ 3.0730, c ¼ 10.053 a ¼ 3.0806, c ¼ 15.1173 3.166 3.214 3.211 2793 0.71 3.2 3.7 3.8 3.9 4.46 for a-axis, 4.16 for c-axis Da/a0 ¼ 0.4781, Dc/c0 ¼ 0.4976 Da/a0 ¼ 0.5140 440 0.183–0.192 2.6916 at l ¼ 498 nm 2.6686 at l ¼ 500 nm 2.6980 at l ¼ 498 nm 2.6894 at l ¼ 498 nm e(0) ¼ 9.75, e(1) ¼ 6.52 e(0) ¼ 9.66, e(1) ¼ 6.52 ? c-axis e(0) ¼ 10.3, e(1) ¼ 6.70 || c-axis 102–103, higher in V doped
Density (g cm3)
Melting point ( C) Heat capacity (J g1 K1) Thermal conductivity (W cm1 K1) Linear thermal expansion coefficient (·106 K1) Percent change in lattice (300–1400 K) Youngs modulus (GPa) Poissons ratio Refractive index (ordinary ray)
Dielectric constant
Electrical resistivity (undoped)
4H
Thermal expansion coefficient (%)
1.8
α−SiC
1.6 1.4 1.2 1.0 0.8 0.6 0.4 0.2 0.0 –0.2 0
200 400
600
800 1000 1200 1400 1600 1800 2000 2200 2400 2600 2800
Temperature (K) Figure 3.6 Thermal expansion coefficient of SiC in c and a planes as a function of temperature [32].
j337
j 3 Growth and Growth Methods for Nitride Semiconductors
338
FETs. The resistivities close to 1010 O cm in both V-doped and intrinsically compensated SiC are possible. 3.2.3.2 Surface Preparation of SiC for Epitaxy The surface preparation of SiC prior to deposition takes on a different meaning in MBE as opposed to OMVPE. Lacking high temperature capability and H, MBE growth relies on ex situ cleaning procedures. In contrast, SiC surfaces can be epi readied in situ during growth by vapor phase. However, one of the best, if not the best, approach involves gas phase preparation of SiC surface in which the surface is exposed to in H and/or HCl gases at high temperatures, the details of which are discussed below. Unless SiC substrates are H polished a priori or in the growth reactor, as can be done in HVPE or OMVPE methods, it is recommended that approximately 3 mm from the surface be removed in a hot KOH solution (300–350 C). If the substrate quality is not high, the etch rate in defective regions is high and smooth surfaces do not follow. Assuming that the previous step is successful, it is followed by a DI rinse for 3 min and the wafer is blow dried by N2. The SiC substrate is then subjected to a series of oxidation and passivation procedure. The substrate is immersed for 5 min in a 5 : 3 : 3 solution of HCl : H2O2 : H2O at 60 C, followed by a 30 s DI rinse. The resulting oxide layer is then removed by dipping the substrate, for 20 s, in a 10 : 1 solution of H2O : HF. This procedure is repeated several times (three to five times) after which the substrate should not be exposed to the atmosphere for longer than 30 min, otherwise another oxidation–passivation procedure would be required. Skipping the chemical etching step leaves the surface with residual damage from mechanical polishing. As mentioned above, techniques have been developed to remove the surface damage by plasma or vapor etching. One such technique is the mechanical chemical polish, which has made substantial progress [34] (commercial service is available from Novasic), and the other is etching in H and/or Cl environments at very high temperatures. The surface morphology of the as-received SiC substrate that underwent a standard mechanical chemical polish (MCP) contains much surface and subsurface damage as characterized by many grit scratches shown in Figure 3.7a. Special MCP procedures have been developed to obtain smoother surfaces (see Ref. [34]), two example of which obtained by Eagle Picher and Novasic are shown in Figure 3.7b and c. Schottky barrier diodes on as-received and Novasic MCP-treated substrates indicated as much as four orders of magnitude reduction in the reverse bias current, albeit with a considerable nonuniformity. Another method that leads to atomically smooth SiC surfaces and much improved GaN overlayers, in terms of reduced extended defect concentration as shown by plan Figure 3.7 (a) AFM image of as-received SiC surface following a standard mechanical chemical polish. Image size is 10 mm · 10 mm and vertical scale is 50 nm. Note the presence of scratches; (b) AFM image of SiC surface after a mechanical chemical polish performed at Eagle
" Picher. Note that scratches are no longer present. Image 10 mm · 10 mm, vertical 5 nm; (c) a 2 mm · 2 mm AFM image of a Cree 6H-SiC wafer MCP polished by Novasic showing a root mean square roughness of 0.134 nm.
3.2 A Primer on Conventional Substrates and their Preparation for Growth
j339
j 3 Growth and Growth Methods for Nitride Semiconductors
340
view transmission electron microscopy (TEM) images, is the high temperature H annealing [35]. Those investigating various issues dealing with SiC have recently developed and exploited the in situ hydrogen etching [36]. H2, H2 þ HCl, or H2 þ C3H8 etching at temperatures between 1300 and 1550 C removes the scratches caused by mechanical polishing [37]. This is more effective than wet HF etching of SiO2 after oxidation. Other wet etching techniques in bases such as molten salts (Na2O2, NaOH, KOH, etc. at temperatures approaching or at 500 C) reveal the defect features of SiC surface and thus are not suitable for surface preparation for epitaxy. Even though nitridation has been shown to improve the smoothness of SiC substrate owing to a combination of nitrogen chemisorption and etching at 1050 C in NH3 flow for 30 min, the utility of long nitridation processes is questioned because of SixNy formation. An atomic force microscopy (AFM) image of a SiC substrate polished by a hightemperature H treatment (1500 C), similar to that reported in Ref. [38], and used for GaN growth in Ref. [35], is shown in Figure 3.8, which clearly shows well-ordered and unbroken atomic terraces indicative of superb surface quality. Similar results can be obtained by a H treatment in a typical SiC growth reactor at about 1500 C. A light follow-up etch in molten KOH solution ensures atomically smooth terraced surfaces if the H polishing steps are not ideal. As previously mentioned, vacuum deposition equipment, such as the one used in MBE, is not compatible with the high-temperature H or HCl treatment, but remote plasma etching techniques can be employed to at least remove the damaged surface layer if H-etched samples are not available. To circumvent the need for high temperatures and exotic treatments incompatible with conventional MBE setups, a preparation procedure adapted from conventional Si technology, and augmented by H plasma cleaning, has been shown to work for SiC. In the first step of this procedure reported by Lin et al. [39], the surface is hydrogen passivated using an HF dip before
Figure 3.8 AFMimageofSiCsurfaceaftera5 min1600 C hydrogen polishing step. Note that scratches are no longer present. The step height seen in a full c-direction lattice parameter for 6H-SiC.
3.2 A Primer on Conventional Substrates and their Preparation for Growth
being introduced into vacuum. In the second step, the substrate is treated with hydrogen plasma, which reduces the CO level (oxygen–carbon bonding) to a value below the X-ray photoemission detection limit. Detailed investigation of SiC surface after H anneal have been undertaken [40] using such techniques as low-energy electron diffraction (LEED) experiments and Auger electron spectroscopy (AES) in addition to reflection high-energy electron diffraction (RHEED). The emphasis is to determine the surface chemical and structural properties taking it beyond what may be needed simply for epitaxy but looking at it with the precision needed for MOS-like structures with AlN/SiC composite. When (0 0 0 1) SiC samples subjected to a H polish at 1500 C for 5 min in flow 3000 scum are introduced into the UHV system, they exhibited a 2 ffiffi pffiffiaffi Hp ffi ( 3 3) R30 LEED pattern with bright and sharp superstructure spots. For comparison on the C-polarity surfaces, ð0 0 0 1Þ, no background was observed, which indicates a high degree of order. On the (0 0 0 1) surfaces, however, always a faint background was visible. The ratio between the average intensities of fractional and integer order beams was above 0.5 for both surfaces, indicating strong surface reconstruction in both cases. The typical Auger spectra displayed a strong OKLL peak in addition to the typical SiLVV and CKLL peaks [40]. The SiLVV signal on the ð0 0 0 1Þ surface showed both bulk-related peak at 90 eV and a feature at around 65 eV attributed to oxygen-bonded silicon. The SiLVV signal of the (0 0 0 1) surface is more complex as expected because of differently coordinated Si for both surfaces. The LEED pattern on the ð0 0 0 1Þ surface changed to (3 3) structure following a 30-min annealing at 1050 pffiffiffi C.pIn ffiffiffi addition, both the oxygen and the Si–O AES signals vanished. The ( 3 3) R30 structure remained for the (0 0 0 1) surface after a 30-min annealing at 1000 C. However, OKLL peak disappeared and SiLVV signal returned the bulklike shape. The simultaneous structural evolution and oxygen removal from both surfaces are indicative oxide on the surface, which was pffiffiffi of pffiffisilicon ffi confirmed by LEED analysis of the ( 3 3) R30 phase of the presence of Si2O3 overlayer, referred to as the honeycomb silicate adlayer, on the SiC surface. The honeycomb silicate adlayer is formed by two Si atoms per unit cell and each oxygen atom connects two of the Si atoms completing a ring-type structure, as shown in Figure 3.9 pffiffiffi p ffiffiffi depicting the top view or the projection on the c-plane. The silicate-related ( 3 3) R30 structure on the (0 0 0 1) surface, shown as a side view, in Figure 3.9 is identically arranged with Si atoms also oriented toward the substrate. Interestingly, the silicate layer and substrate are linearly bridged by oxygen, Si–O–Si, not connected via Si–Si bonds. This simply implies that the surface as it stands is not optimum for nitride growth. It is possible that in OMVPE environment, the silicate adlayer may be removed, a statement that cannot be unequivocally made for MBE growth. In fact, the growth of AlN on such SiC surface leads to three-dimensional (3D) growth and, as discussed earlier, exposureto Gaathightemperature removes thesilicate layer [41–44]. Treating the surface of ex situ HCl-treated (1300 C) SiC with in situ Ga spray, Onojima et al. [41] obtained SiC surfaces free of the silicate adlayer, as shown schematically in Figure 3.10, and were able to obtain a 2D AlN growth. Owing pffiffiffi topGa ffiffiffi deposition on the SiC surface and subsequent flash-off, an oxygen-free ( 3 3) R30 surface structure was achieved and initial 2D growth with an evident RHEED
j341
j 3 Growth and Growth Methods for Nitride Semiconductors
342
[1 1 0 0]
(a)
[1 1 2 0] [0 0 0 1]
Top Si Second Si (b) Silicate adlayer
C Top O
SiC surface
Second O
Figure 3.9 (a) Top view of the oxide structure on SiC ð0 0 0 1Þ. The Si2O3 silicate adlayer consisting of a honeycomb structure with SiOSi bonds. At the center of the hexagons, one carbon atom of the topmost substrate bilayer is visible [the dark shaded area indicates the (1 1; 1) unit cell and light shaded the
pffiffiffi pffiffiffi ð 3 3ÞR30 -unit cell]; (b) side view of the oxide structure on the SiC (0 0 0 1) in ð0 1 1 0Þ SiC projection. Linear SiOSi bonds connect the silicate layer and the underlying SiC substrate. Courtesy of N. Onojima (patterned after Ref. [40]). (Please find a color version of this figure on the color tables.)
intensity oscillation was demonstrated. The initial growth mode of AlN closely correlated with the crystalline quality of AlN layer. Figure 3.11 shows the RHEED images of HCl-treated surface with a silicate adlayer, HF treated surface with has residual O, and finally in situ Ga spray treated SiC, of O on the surface for the ½1 1 2 0 and ½1 1 0 0 azimuths. Note the pffiffiffi which pffiffiffi is void ( 3 3) R30 RHEED surface reconstruction for the HCl-treated surface, 1 · 1 pffiffiffi pffiffiffi RHEED reconstruction, and again ( 3 3) R30 RHEED surface reconstruction for the in situ Ga spray treated surface. 3.2.4 Sapphire
Owing to its relatively low cost, availability in large area, and continual improvement in its quality, both in terms of bulk and surface properties, sapphire has become the [1 1 0 0] [1 1 2 0] [0 0 0 1]
(a) Si
(b) 1/3 ML Si ad atom
ad-Si C
SiC surface Figure 3.10 (a) Top view (projection on the Si-plane of the basal plane of SiC) and (b) side view of SiC after an in situ Ga exposure indicating of the lack of silicate adlayer. Courtesy of N. Onojima. (Please find a color version of this figure on the color tables.)
3.2 A Primer on Conventional Substrates and their Preparation for Growth
Figure 3.11 RHEED images along the ½1120 and ½1100 azimuths for HCl, HF,pand in ffiffisitu pffiffiSi-polarity ffi pffiffiffi SiC surface ffiffiffi p ffi Ga spray treated indicating ð 3 3ÞR30 , 1 · 1, and ð 3 3ÞR30 , respectively. Courtesy of N. Onojima.
dominant substrate material for epitaxy. Although there are other reasons, sapphire is transparent for most of the bandgaps of nitride alloys; thus, it affords certain benefits in detectors, for example, for back illumination and in LEDs, for lack of absorption. 3.2.4.1 A Primer on Sapphire Sapphire has the space group of R3c (No. 167), as provided in the International Tables for Crystallography, and is primarily of ionic bond nature. It can be represented by both rhombohedral unit cells, with volume 84.929 Å3, and hexagonal unit cell, with volume 254.792 Å3, which is displayed in Figure 3.12 [13]. In the rhombohedral unit cell there are 10 ions in total, 4 Al3 þ ions and 6 O2 ions. The hexagonal unit cell has 30 ions in all, 12 Al3 þ ions and 18 O2 ions. Oxygen is located at (x, y, z) ¼ (0.306, 0, 0.25). If this position is approximated to (x, y, z) (1/3, 0, 1/4), the anion framework forms an hcp lattice with a ¼ 0.476 nm and c ¼ 1.299 nm. The unit cell described by Miller–Bravais indices consists of six close-packed (0 0 0 1) planes of O2 ions sandwiching 12 planes of Al3þ ions that occupy two thirds of the available octahedral voids created by the O2 ions. An Al3þ ion is located at (x, y, z) ¼ (0, 0, 0.352) instead of (0, 0, 1/3), thus the cations are shifted by 0.025 nm along the c-axis from the ideal octahedral sites. The oxygen ion is larger than the aluminum ion by a factor of about 3 in terms of its radius; therefore, the steps on the substrate are limited to those in the oxygen sublattice, leading to step heights in multiples of c/6 (d(0006) 0.216 nm). The (0 0 0 1) Al2O3 surfaces are oxygen terminated and present steps along f1 1 2 0g and f1 1 0 0g planes [45]. Two crystallographically equivalent surfaces are related by a symmetry operation of the space group. Along the [0 0 0 1] direction, A–A or B–B surfaces are separated by c/3, 2c/3, and c steps. Steps separating two A surfaces are noted as A–A, and c/3 steps of height c/6, c/2, or 5c/6
j343
j 3 Growth and Growth Methods for Nitride Semiconductors
344
separate the two surfaces related by a glide symmetry operator. Such steps are dubbed demi-steps and are noted as A–B, c/6 [46,47]. The unreconstructed basal c-plane perspective views for both unit cells are given in Figure 3.13 [13], where the cell boxes are polyhedra. A schematic representation of sapphire unit cell indicating the six O layers in the unit cell is shown in Figure 3.14. The oxygen ions form a pseudohexagonal lattice. The small Al ions occupy the octahedral sites. The labeling of planes and directions in the context of sapphire substrates are shown in Figure 1.5. Properties of sapphire are provided in Table 3.5 [13]. All common surfaces employed for GaN epitaxy including the (0 0 0 1) and ð1 1 0 0Þ are nonpolar. Thus, the polarity control on sapphire depends on the particulars of growth conditions employed with the ominous inversion domain formation always a possibility. Because
Figure 3.13 Perspective views in (2 · 2 · 1) unit cells: (a) along the [0 0 0 1] direction in a rhombohedral unit cell; (b) along the
3.2 A Primer on Conventional Substrates and their Preparation for Growth 0.287 nm
0.252 nm
b B
O 2–
Al3+
a A
0.052 nm
b
0.0797 nm
C
0.1358 nm
a B b
0.1441 nm
0.1661 nm [0 0 0 1]
A a
[1 0 1 0]
Figure 3.14 A schematic diagram of the Al2O3 sapphire unit cell, there are six oxygen layers in the unit cell, the distances between the various atomic layers change as shown in the figure. The oxygen ions form a pseudohexagonal lattice. The small Al ions occupy the octahedral sites. Courtesy of P. Ruterana and Ref. [47]. Table 3.5 Properties of sapphire (in part after Ref. [13] and references therein).
Parameter
Value
Condition
Lattice constant (Å) Melting point ( C) Density (g cm3) Thermal expansion coefficient (K1)
a ¼ 4.765, c ¼ 10.2982 2030 3.98 6.66 · 106 || c-axis 9.03 · 106 || c-axis 5.0 · 106 ? c-axis a/a0 ¼ 0.83, c/c0 ¼ 0.892
20 C
Percent change in lattice constants with DT Thermal conductivity (W cm1 K1) Heat capacity (J K1 mol1) Youngs modulus (GPa)
Tensile strength (MPa) Poissons ratio Hardness: Knoop nanoindentation (GPa) Energy band gap (eV) Resistivity (O cm)
20 C 20–50 C 20–1000 C 20–1000 C 293–1300 K
0.23 || c-axis 0.25 || a-axis 77.9 452–460 in [0 0 0 1] direction, 352–484 in ½1 1 2 0 direction 190 0.25–0.30 23.9 2.0
296 K 299 K 298 K
8.1–8.6 >1011
Experimental value 300 K
300 K 300 K 300 K
j345
j 3 Growth and Growth Methods for Nitride Semiconductors Thermal expansion coefficient (%)
346
2.4 2.2 2.0 1.8
Polycrystalline
1.6 1.4 1.2 1.0 0.8 0.6 0.4 0.2 0.0 –0.2
0
c-axis
Al2O 3
a-axis Si
200
400
600
800
100
1200
1400 1600
1800 2000
Temperature (K) Figure 3.15 Thermal expansion coefficient of sapphire on the cplane (along the a-axis) and along the c-axis, and that of Si as a function of temperature [48].
of heteroepitaxy of GaN on sapphire, it is useful to display the thermal expansion coefficient of sapphire as done in Figure 3.15. 3.2.4.2 Surface Preparation of Sapphire for Epitaxy As-received sapphire substrates contain scratches caused by mechanical polishing with root mean square (RMS) roughness values between 0.8 and 2.1 nm over l mm2 areas. Wet chemical etches such as phosphoric acid (H3PO4), sulfuric–phosphoric acid combination (H2SO4–H3PO4), fluorinated and chlorofluorinated hydrocarbons, tetrafluorosulfur (SF4), and sulfur hexafluoride (SF6) have been employed. None of these techniques, however, produces a surface free of damage and scratches. For MBE growth, which does not allow in situ cleaning of the surface in H at high temperatures, a 3 : 1 solution of H2SO4 : H3PO4 is used as the etchant. The substrate is dipped in this solution and kept at 300 C for 20 min. This is followed by a rinse in DI water for 3 min. Although the hot etching removes some material, the resultant surface still bears the scratches caused by mechanical polishing. However, the surface becomes flatter after etching, with the RMS roughness being reduced from 0.323 nm in image to 0.211 nm in image in one case. A high-temperature annealing technique after wet chemical etching of mechanically polished sapphire substrates has been shown to result in atomically smooth surfaces [49]. Figure 3.16 shows the AFM images of two c-plane sapphire surfaces, (a) before and (b) after the chemical etching. In an OMVPE or HVEP environment, the typical process is to simply heat the sapphire under flowing hydrogen at temperatures between 1000 and 1100 C. This process etches sapphire slightly leading to the formation of hexagonal pits if there are residual amounts of gallium left in the reactor from prior runs. The crystal quality of subsequently deposited GaN films was insensitive to the presence of these pits [13]. To eliminate surface damage altogether, a high-temperature annealing step has been employed, which gives rise to atomically smooth surfaces. A very high
3.2 A Primer on Conventional Substrates and their Preparation for Growth
Figure 3.16 (a) AFM image of an as-received sapphire substrate. Note the scratches caused by mechanical chemical polishing; (b) AFM image of a sapphire substrate after a 180 C etch in sulfuric/ phosphoric acid. Some improvements are apparent, but the scratches remain and are accentuated to some extent. Image size 2 mm · 2 mm.
temperature annealing investigation of sapphire substrates was recently undertaken. Annealing experiments in air at 1000, 1100, 1200, 1300, and 1380 C (the ceiling of the furnace employed) for 30- and 60-min periods were conducted to determine the best conditions with the aid of AFM images of the finished surface. This was followed by observation of RHEED patterns once in an MBE system. A small, but progressive, improvement was observed in the reduction of scratches up to 1300 C. However, annealing at 1380 C for 1 h led to scratch-free and smooth surfaces to the point where the only noticeable feature in AFM images were the atomic steps about 0.15 nm in height. AFM images indicated that annealing at
j347
j 3 Growth and Growth Methods for Nitride Semiconductors
348
Figure 3.17 An AFM image of sapphire following a 1380 C–1 h annealing in atmosphere. Atomically flat surface is clearly visible. Atomic step heights are about 0.15 nm, which represent the only roughness in the image. The diagonal lines, from left to right, are the artifacts of AFM.
1380 C for 1 h leads to atomically smooth surfaces as shown in Figure 3.17. An atomically smooth surface is maintained after nitridation as well. RHEED images typically show extended and bright rods associated with sapphire at temperatures as low as 600 C during the ramp-up as shown in Figure 3.18.
Figure 3.18 A RHEED image at about 800 C of an annealed sapphire at 1380 C for 1 h (½1 1 2 0 azimuth). Clear streaky RHEED pattern observed at temperatures as low as 600 C indicates that the high temperature annealing step produces clean epiready surfaces. Without the annealing procedure, the RHEED images are not as clear and elongated and not reproducible.
3.2 A Primer on Conventional Substrates and their Preparation for Growth
Sapphire is nitridated by exposing it to nitrogen plasmas or thermally cracked ammonia (a practice that has been abandoned) in MBE reactors or by to ammonia/ hydrogen gas mixtures in OMVPE reactors. Sapphire substrates that have not undergone a heat treatment in O, as described above, exhibit the polishing damage in the form of random scratches when they undergo MBE-like pregrowth in situ heat treatment followed by an exposure to ammonia [50]. Considering the stacking-order mismatch between sapphire and nitrides, these features are likely to have deleterious effects on growth. A low density (108 cm2) of surface outgrowths was observed after 30 min of nitridation. The presence of surface damage does not appear to have influenced the formation of protrusions. There is no clear correlation between the positions at which the protrusions have formed and the local surface topography. Uchida et al. [51] observed similar protrusions after 5 min of nitridation at 1050 C in an OMVPE system, but at a much higher density than that observed in the MBE process. It is likely that a combination of a higher substrate temperature and the background ammonia pressure promotes a more rapid nitridation reaction leading to a higher density of protrusions in OMVPE-grown samples. Although what is reported for these particular samples may hold, it should again be noted that atomically smooth surfaces following annealing in air, as described above, do not show discernible change after nitridation. Noting that AlOxN1x would be unstable at the nitridation temperatures employed, the nitridation of sapphire should result in the formation of AlN. In fact, in the MBE process, this can be observed with RHEED in that the pattern associated with the ½1 1 2 0 azimuth of sapphire gives way to the pattern associated with the ½1 1 2 0 azimuth of AlN but with a 30 rotation to minimize misfit strain. Nitridation has direct consequences in the quality of the low-temperature buffer layer and final layer(s). The ultimate test whether nitridation was done properly requires going through the buffer and final layer growth. The particulars of the final layer are then used to draw conclusions about the nitridation process as was done by Wickenden et al. [52], the details of which are discussed in Section 3.5.5.1. In the study of Uchida et al. [51], nitridation was reported to occur very rapidly for times less than 3 min and then slowed considerably. For short nitridation times, of <3 min, the surface was reported to be relatively smooth, but stress-induced protrusions developed for longer nitridation times. The density of the protrusions increased with time, making the surface progressively rougher. Uchida et al. [51] posed the argument that nitridation produced an amorphous AlNxO1x layer via the exchange of oxygen atoms from sapphire and nitrogen atoms from ammonia. However, this amorphous layer was not seen in subsequent TEM micrographs after the deposition of a 4.0 mm thick GaN layer, which they attributed to diffusion of N and O atoms into the crystalline layer. X-ray photoelectron spectroscopy (XPS) analyses were carried out to confirm the incorporation of nitrogen atoms into the sapphire substrate during the nitridation process. Sapphire substrates exposed to ammonia even for as little as 1 min did indeed show the 1s nitrogen peak; the same is expected from RF nitrogen exposure as well. This indicates that nitrogen atoms react with the surface. The downshifting of the oxygen 1s line by about 0.25 eV, relative to the spectra obtained from a bare sapphire sample, and the sample nitrided for 1 min
j349
j 3 Growth and Growth Methods for Nitride Semiconductors
350
suggests the generation of a significant bonding between oxygen and nitrogen atoms in the nitrided layer [50]. The RHEED observations during nitridation indicate that AlN forms on sapphire during nitridation and its orientation along the c-direction is rotated by 30 to accommodate the strain caused by lattice mismatch. The duration of nitridation is an important factor in terms of the quality of the eventual GaN layer. Increase in nitridation times from 60 to 400 s has been reported [53] to result in mobility reduction in GaN films, the overall features of which are consistent with the study of Kim et al. [50,54]. The nitridation process, both in OMVPE and MBE, has a great impact on the subsequent layers. In MBE, if nitridation is done well, which can clearly be observed by RHEED image changing from sapphire to AlN, the subsequent AlN layer is of high quality. In OMVPE, the picture is more complex in that nitridation is followed by the low-temperature buffer layer that is then annealed prior to the commencement of growth of subsequent layers. The structure of the low-temperature buffer layer and the change following annealing are such that the morphology of the subsequently deposited GaN buffer layer changes from rough to highly faceted (possibly mixed with the zinc blende structure) and to smooth films of the wurtzite structure. The details of the lowtemperature buffer and other buffer layers as well as other issues related to GaN growth in general are discussed in Section 3.5.5.1. The calculated lattice mismatch between the basal GaN before the in-plane rotation and the basal plane of sapphire is about 49%. However, the actual lattice mismatch of nitride layers with sapphire is reduced by the rotation of the nitride lattice with respect to the substrate unit cell by 30 . Consequently, the lattice mismatch is reduced to 13% to AlN, 16% to GaN, and 29% to InN. This large mismatch would cause even the very thin layers to be fully relaxed at growth temperatures. When the samples are cooled down after the growth, a residual thermal strain is created. In general, films grown on the basal plane show either very little or none of the cubic GaN phase. Among the faces of sapphire that have been utilized for nitrides are the c-, a-, and r-planes. The stacking configurations perpendicular to the c- and a-planes are displayed in Figure 3.19a and b. The atomic arrangements on the c- and a-planes of sapphire are also shown in Figure 3.19. Figure 3.20 depicts the stacking arrangement perpendicular to the r-plane (a) and the atomic arrangement on the r-plane of sapphire (b). 3.2.5 ZnO
Unlike sapphire and polytypes of SiC that are available, ZnO and GaN have a common stacking order and a small lattice mismatch as shown in Table 3.1. In addition, high-quality substrates, nearly 2 in. in diameter, are available. ZnO is up against the fact that most laboratories select sapphire substrates with SiC making inroads. The reasons are that the layers grown on sapphire and SiC have in many cases better quality, and sapphire is available up to 6 in. in diameter and is inexpensive. The temporary problems plaguing the SiC approach are high cost, fluctuating
3.2 A Primer on Conventional Substrates and their Preparation for Growth
Basal or c-plane sapphire
[0 0 0 1]
O O O Al Al O O O Al Al O O O Al Oxygen
(a)
Al
a-plane (1 1 2 0) sapphire
[1 1 2 0]
O O Al Al O
O Oxygen Al
(b)
Figure 3.19 (a) The atomic arrangement and stacking order (left) and the top view (right) of the c-plane sapphire. (b) The atomic arrangement and stacking order (left) and the top view (right) of the a-plane sapphire.
quality, and poor surface finish. A more permanent, fundamental problem is the stacking-order mismatch on SiC and sapphire. As mentioned earlier, ZnO has the desired stacking order and a reasonably close lattice match to GaN. Moreover, as in the case of LiGaO2, it can be selectively removed from GaN followed by GaN being transferred to a template with good thermal conductivity. 3.2.5.1 A Primer on ZnO The thermal expansion coefficients and the lattice constants of ZnO are Da/a ¼ 4.8 · 106 K1, Dc/c ¼ 2.9 · 106 K1, a ¼ 3.2426 Å, and c ¼ 5.1948 Å, respectively. The thermal expansion coefficient of ZnO in the c-plane as a function of temperature
j351
j 3 Growth and Growth Methods for Nitride Semiconductors
352
Side view of sapphire r-plane {1 0 1 2}
Oxygen
(a)
Al
Al2O3 [1 0 1 1]
3.2 A Primer on Conventional Substrates and their Preparation for Growth
1.2
Thermal expansion coefficient (%)
ZnO
1.0
a-axis
0.8 Polycrystalline
0.6 0.4 c-axis
0.2 0.0 –0.1 0
200
400
600
800
1000 1200 1400 1600 1800 2000
Temperature (K) Figure 3.21 Thermal expansion coefficient of ZnO on the c-plane (along the a-xais), along the c-axis as a function of temperature; the thermal expansion coefficients of polycrystalline ZnO [48].
is depicted in Figure 3.21. Also shown are the thermal expansion coefficients of polycrystalline ZnO. High-quality ZnO substrates have recently become available in laboratory quantities. While stable in air and O environments at temperatures as high as 900 C, perhaps even higher, exposure to ammonia etches ZnO even at temperatures as low as 600 C. It is believed that atomic hydrogen reacts with O forming a volatile water vapor. The Zn metal is removed from the surface by evaporation. 3.2.5.2 Substrate Preparation for Epitaxy The chemical preparation of ZnO involves a 5 min acetone bath followed by the same procedure in methanol while using ultrasound agitation to remove particulates. The wafers are then rinsed in deionized water, followed by blow drying with filtered nitrogen prior to introduction into the growth chamber. The details of the growth processes are given in Section 3.5.7. ZnO is similar to the SiC and sapphire substrates and yet very different. It shares very similar problems, chiefly the scratches caused by mechanical polishing. 3 Figure 3.20 (a) Sapphire r-plane stacking sequence showing O atoms in larger clear circles and Al atoms in smaller, filled circles. The salient feature is that each Al layer has an O layer above and below it. (b) The atomic arrangement on three layers (the uppermost one is O,
immediately below is Al and third layer down is another O layer) on the r-plane of sapphire. The lines are there just guides to eye and do not represent bonds. (Please find a color version of this figure on the color tables.)
j353
j 3 Growth and Growth Methods for Nitride Semiconductors
354
Figure 3.22 An AFM image of a ZnO surface after a 3-h annealing procedure in air at Ta ¼ 900 C. The RMS is about 0.1 nm. The image is 500 nm · 500 nm.
Chemical etches have not yet been developed to deal with this problem. In fact, it may not even be possible to accomplish this task with chemical etches. However, as in the case of sapphire, annealing in oxygen appears to lead to improved surfaces over mechanically polished substrate. Annealing at 900 C for 3 h in air leads to atomic bilayer steps, which is about the best that can be achieved, as shown in Figure 3.22. Annealing in air at 850 and 950 C also led to very smooth surfaces [55]. This implies that the process is not very temperature sensitive and that reproducible results should be obtained without extreme control over the temperature employed. In an MBE system, the RHEED patterns also bear this out by showing sharp 1 · 1 diffraction rods during the ramp-up of the temperature, as shown in Figure 3.23.
Figure 3.23 RHEED image of ZnO taken at 780 C (½1 1 2 0 azimuth.
3.2 A Primer on Conventional Substrates and their Preparation for Growth
3.2.6 LiGaO2 and LiAlO2
There are other nontraditional substrates that are beginning to appear; among them are LiGaO2 (lithium gallium oxide (LGO), which is orthorhombic with space group Pna21 No. 33), and LiAlO2 (lithium aluminum oxide, which is tetragonal with space group P41212). 3.2.6.1 LiGaO2 Substrates Lithium gallate (LiGaO2) is the most closely lattice-matched substrate currently being considered for GaN heteroepitaxy, with an average lattice constant mismatch of only 0.9% in the basal plane. It is therefore expected that the quality of the very thin GaN films on LiGaO2 would be better than those on other substrates discussed so far. There is some experimental evidence to this effect. The GaN films grown on LiGaO2 are of Ga polarity. Although LiGaO2 is easily etched, which is useful for transferring the GaN film to another substrate, it is not easily etched to a smooth surface for epitaxy. As in the case of ZnO, etching for a few tens of seconds in H3PO4 removes the surface damage caused by mechanical polishing and reduces the short-range roughness. The main disadvantages of LiGaO2 are its low thermal stability under OMVPE growth conditions, low thermal conductivity, high thermal expansion coefficients, and electrical insulation. However, when GaN layers of about 300 mm are grown by HVPE followed by the removal of LiGaO2, the thermal and electrical conductivity of LiGaO2 are not relevant. The structure of LiGaO2 is similar to the wurtzitic structure, but because Li and Ga have different ionic radii, the crystal has an orthorhombic structure. The atomic arrangement in the (0 0 1) face is hexagonal, which promotes the epitaxial growth of (0 0 0 1) GaN, so that the epitaxial relationship (0 0 0 1) GaN/(0 0 1) LiGaO2 is expected. The crystal structure deviates slightly from the hexagonal symmetry because of the need to accommodate two different metallic atoms of Ga and Li. For the visualization of LiGaO2, one can think of it as the I–III–V analogue of ZnO, with one half of the Zn replaced by Li and the other half by Ga. The metal atoms are ordered, alternating in the [0 1 0] direction between Li layers and Ga layers. Unlike ZnO, LiGaO2 melts under atmospheric pressure, and hence large single crystals can be pulled from the melt using the Czochralski method. The orthorhombic lattice dimensions of LiGaO2 are a ¼ 5.402 Å, b ¼ 6.372 Å, and c ¼ 5.007 Å. The distance between the nearest cations in LiGaO2 is in the range of 3.133–3.189 Å, while the distance between nearest anions is in the range pffiffiffiof 3.021–3.251 Å. For comparison, the basal plane wurtzite lattice parameter h ¼ a= 3 ¼ b=2. The main virtue of LiGaO2 is that it has a very small lattice constant mismatch with GaN (averaging 0.9% at room temperature: 1.9% in the a-direction and 0.19% in the b-direction). The orientation relationship between GaN and LiGaO2 has been studied by several groups, yielding ½1 1 2 0 GaN||[0 1 0] LiGaO2 and [0 0 0 1] GaN||[0 0 1]LiGaO2. The atomic relationship between GaN and LiGaO2 in the ideal case is shown in Figures 3.24 and 3.25. Anion (oxygen) or cation (gallium and lithium) termination is available on the (0 0 1) substrate surfaces. As is the case in polar surfaces in all compound substrates,
j355
j 3 Growth and Growth Methods for Nitride Semiconductors
356
a
2
m
C
b
m
LiGaO2
Lower O
Li
Upper O
Ga
Figure 3.24 Example of the exact fit of GaN atoms over the LiGaO2 lattice if there is no distortion. Courtesy H. Paul Maruska. (Please find a color version of this figure on the color tables.)
the A-face (oxygen-terminated surface) can be etched easily, whereas the B-face (metal-terminated surface) is difficult to etch in an aqueous solution of nitric acid (H2O : HNO3 ¼ 1 : 1). This is attributed to surface oxygen atoms having a dangling bond with two electrons, whereas surface metal atoms have no dangling bonds (hard to etch). A maximum etch rate of 0.25 mm min1 for the cation face was obtained at pH 9.2 and temperature 50 C. The etch is selective in relation to GaN, which means that the entire LGO substrate could be chemically removed without affecting the GaN layer. As-received LGO surfaces contain scratches and pits as a result of polishing damage and crystalline defects, as shown in Figure 3.26. However, following a chemical cleaning in acetone, methanol, and DI water and etching in H3PO4 for 10 s
3.2 A Primer on Conventional Substrates and their Preparation for Growth
LGO: orthorhombic a = 5.402 Å b = 6.372 Å c = 5.007 Å
c
(a)
b
a
LGO: orthorhombic a = 5.402 Å b = 6.372 Å c = 5.007 Å
c
(b)
b
a
Projection on c-plane Δb = –0.19 % a
Ga
N
=3
.18
9Å
bLGO = 6.372 Å
Δa = +1.1%
Δa = +1.9 % Δb = –1.1% (c)
aLGO = 5.402 Å
Figure 3.25 Structure of (a) orthorhombic LiGaO2 (LGO), (b) GaN, and (c) a detailed view of the relative orientation of GaN with respect to LGO. Courtesy of H. Paul Maruska. (Please find a color version of this figure on the color tables.)
j357
j 3 Growth and Growth Methods for Nitride Semiconductors
358
Figure 3.26 AFM images of LiGaO2 (LGO): (a) as-received, (b) after etching in H3PO4 for 10 s at 80 C, and (c) H3PO4 for 20 s at 80 C.
at 80 C, the surface is improved, with more improvement resulting when the etching time is increased to 20 s as shown in Figure 3.26b and c, with reduced pits and shortrange roughness. Compared to the as-received LGO, which exhibited a RMS roughness of 1.12 nm, a 20-s etching reduces this value to 0.78 nm. The RHEED analysis also shows a marked improvement in the sharpness and intensity of the reconstruction rods. The bandgap of LGO is 5.6 eV. The Youngs modulus of LGO is about 150 GPa and its density is 4.175 g cm3. The hardness is 7.5–10 GPa, as measured by nanoindentation. Because LiGaO2 is asymmetric along the [1 0 0] and [0 1 0] directions, the coefficients along these two directions are different. The thermal expansion coefficients are Da/a ¼ 6 · 106 K1, Db/b ¼ 9 · 106 K1, and Dc/c ¼ 7 · 106 K1. Optically, it is biaxial with a transparency ranging from 0.3 to 6 mm and refractive indices of na ¼ 1.7617, nb ¼ 1.7311, and nc ¼ 1.7589 at 620 nm. The crystal has no natural cleavage planes. The structure of interest is the only stable form of the compound in the range from room temperature to its melting point. Because LiGaO2 melts congruently at 1585 C, the requirements for its crystal growth are relatively modest – no high-temperature or high-pressure apparatus is necessary. Therefore, good-quality wafers are available with relatively large diameters at a price comparable to that of sapphire. Crystals up to 50 mm in diameter and 200 mm in length have been produced at pull rates of 2–5 mm h1 by the Czochralski method from the mixture of Li2CO3 and Ga2O3 [56]. Crystals have been grown in [1 0 0], [0 1 0], and [0 0 1] orientations. 3.2.6.2 LiAlO2 Substrates The lattice constants of LiAlO2 are a ¼ 5.1687 Å and c ¼ 6.2679 Å with a density of 2.615 and hardness of 8 GPa, as determined by nanoindentation. The thermal expansion coefficients are Da ¼ 7.1 · 106 K1, and will cause a compressive strain in GaN g/cm3 and Dc ¼ 15 · 106 K1. Optically, it is uniaxial with transparency ranging from 0.2 to 4 mm and refractive indices of ne ¼ 1.6014 and n0 ¼ 1.6197 at
3.2 A Primer on Conventional Substrates and their Preparation for Growth
633 nm. The crystal has no natural cleavage planes requiring, as in the case of LGO, the facet development with a chemically assisted ion etching or chemical mechanical polishing. Of course, the former is better in terms of performance and cost reduction. The structure of interest is the high temperature form (or g form) of the compound. It melts congruently around 1700 C and is stable at room temperature. Single crystals can be grown by the conventional Czochralsky melt pulling method. The epitaxy relationships between GaN and LiAlO2 are expected to be ð0 1 1 0Þ GaN/(1 0 0) LiAlO2 with ½2 1 1 0 GaN/[0 0 1] LiAlO2. Unlike Al2O3 and 6H-SiC substrates with very smooth surfaces, except the scratches caused by mechanical chemical polishing, the LiAlO2 substrate exhibited a wavelike surface with equidistant grooves about 10 nm deep, which could have originated from the mechanical surface polishing. Because the preferred growth direction of the GaN epitaxial film is [0 0 0 1], the match along the a-axis is more critical for the film deposition than the c-axis. Considering the a-axis lattice constant, it is obvious that LiGaO2 is preferred for GaN and LiAlO2 for Al1xGaxN. Another important factor is that because these two crystals have exactly the same structure as GaN, the growth orientation may not have to be limited just to the c-axis or the [0 0 0 1] direction. In fact, epitaxial growth can be achieved at any orientation. The degree of lattice matching may vary slightly depending on the structure of epitaxial nitrides in exact orientation. 3.2.7 AlN and GaN
It is well known that nitride semiconductors do not enjoy native substrates, notwithstanding considerable efforts to produce them. The main impediment is the large vapor pressure of N on AlN, GaN, and InN, in ascending order, coupled with a low solubility of N in the molten metal at reasonable temperatures and pressures. It is thus imperative to consider the phase diagrams of these binaries. Shown in Figure 1.18 are the partial pressures of N2 over AlN and Al liquid as a function of temperature, as determined by Slack and McNelly [57], on GaN by Karpinski et al. [58], and on InN by Porowski and Grzegory [59]. The calculated values for the N2 vapor pressure on AlN by Slack and McNelly are 1, 10, and 100 atm at about 2550, 2800, and 3120 C, respectively. The GaN data, however, tell a different story in that the partial pressure of N2 is very high, necessitating high-pressure experiments to collect data. Karpinski et al. employed a tungsten carbide anvil cell and pressures of up to 60 kbar to collect the data presented. The GaN data deviate from the calculations of Thurmond and Logan [60] most noticeably at high temperatures. The equilibrium N2 pressure data were calculated from the measurement of the equilibrium ammonia pressure over GaN assuming N2 to be an ideal gas and thus would predict a linear dependence in the log vs. 1/Tscale. The partial pressure data alone are indicative that AlN would be easier to synthesize than GaN. In fact, Slack and McNelly [10] obtained growth rates in the millimeters per hour range under very moderate pressures with a duration of growth determined by the reaction rate of Al with a tungsten crucible. The same partial pressure data also indicate the difficulties associated with an epitaxial
j359
j 3 Growth and Growth Methods for Nitride Semiconductors
360
deposition going from AlN to GaN and then to InN requiring an increasing amount of nitrogen overpressure necessary to avoid decomposition. The data of the phase diagrams of GaN, AlN, and InN are limited and contradictory by reason of the high melting temperatures (TM) and the high nitrogen dissociation pressures (P dis N2 ), particularly for InN and to a lesser extent for GaN. Dissociation pressure of MN, where M represents a metal species such as Al, Ga, or In, is defined as the nitrogen pressure at the thermal equilibrium of the reaction [61] MClðgÞ þ NH3 ðgÞ ! MNðsÞ þ HClðgÞ þ H2 ðgÞ;
ð3:1Þ
where s, l, and g indicate solid, liquid, and gas state, respectively. Reported values for P dis N2 of GaN are plotted in Figures 1.18 and 1.19. Examining the available data, Sasaki and Matsuoka [61] concluded that the data of Madar et al. [62] and Karpinski et al. [58] are most reliable. The data of Karpinski et al. are therefore shown. As can be deduced from Figure 1.18, the nitrogen dissociation pressure equals 1 atm at approximately 850 C and 10 atm at 930 C. At 1250 C, GaN is unstable and decomposes even under a pressure of 10 000 bar of N2. The case of InN is even more problematic at the decomposition temperature of about 700 C [59]. It should therefore come as no surprise that the incorporation of nitrogen at high temperatures is a nontrivial problem at best. For pressures below equilibrium at fixed temperatures, the thermal dissociation occurs at a slow and apparently constant rate, suggesting a diffusive process of dissociation. Despite the fact that equilibrium growth of InN appears nearly impossible, InGaN grows at temperatures in the range of 700–900 C with very high amounts of reactive nitrogen on the surface. The difficulties of growing bulk GaN and especially InN are alleviated somewhat with AlN synthesis. Judging from the N pressure at equilibrium, AlN is the closest to successful production of all the group III–V nitride semiconductors under discussion for bulk growth in the conventional sense. The most satisfactory method of growing highpurity AlN is the one in which AlN itself is the starting material. Slack and McNelly [57,63] employed a technique in which high-purity AlN is produced through an intermediate AlN powder formed by utilizing AlF3. Sublimation converts the AlN powder into single crystals in a closed tungsten crucible or in an open tube with a gas flow. The main problem with this growth technique is perhaps the surface oxidation of the powder owing to the strong reactivity of oxygen and aluminum. If this oxidation is minimized, then AlN could be produced with only 100 ppm of oxygen and with lower amounts of other impurities. The purest AlN prepared by Slack and McNelly employing a tungsten crucible had 350 ppm oxygen. However, when W or Re was used for the crucible, very little contamination of the AIN with metal impurities was found. Furthermore, the crystals grown in W or Re crucibles generally showed uniform amber color indicating that, indeed, both oxygen and carbon contaminations were scrupulously minimized. GaN is more difficult to grow than AlN, but not as difficult as InN. A much lower equilibrium pressure of N on AlN, on the contrary, lends itself to the growth of this alloy utilizing the sublimation technique. Samples with dimensions of some 4 mm diameter and 12 mm length have already been prepared [64] and used to determine much of the AIN thermal data discussed in this
3.2 A Primer on Conventional Substrates and their Preparation for Growth
chapter. The deposition temperature is around 2250 C. Deterioration of the W boat determines the size of the crystal. The thermodynamic properties of GaN [65], in particular its melting conditions, are so extreme that the application of the common growth methods utilizing stoichiometric solutions is technically impossible. To increase the solubility of N in Ga melt and also to reduce N desorption, nitride crystal growth can be attempted under high N2 pressures as has been done by Leszczynski et al. [66,67], Porowski et al. [68,69], and Grzegory et al. [70,71]. The GaN templates are crystallized in gas pressure vessels with volumes up to 1500 cm3 and with a workable crucible volume of 50–100 cm3. The high pressure– high temperature reactor consists of a pressure chamber and a multizone furnace. It also has features for in situ annealing in vacuum and electronics for stabilizing and programming of pressure and temperature. The walls of the vessel are water cooled considering the high temperatures involved. The pressure in the chamber is stabilized to a precision better than 10 bar. The temperature is measured by an array of thermocouples in the furnace. This allows a stabilization of temperature to 0.2 C and programmable variations of the temperature distribution in the crucible. GaN crystals presented are grown from nitrogen solutions in pure liquid gallium or in Ga alloyed with 0.2–0.5 at.% of Mg or Be at pressures in the range of 10–20 kbar and temperatures of 1400–1600 C. Magnesium and beryllium, being the most efficient acceptors in GaN, are added to the growth solutions to obtain p-type crystals. Supersaturation in the growth solution is obtained by a deliberate application of temperature gradient of 2–20 C cm1 along the long axis of the crucible. This approach assures a continuous flow of nitrogen from the hotter part of the solution to the cooler parts. Crystallization experiments performed without intentional seeding resulted in crystals nucleating spontaneously on the internal surfaces of polycrystalline GaN crusts in the cooler zone of the solution. Typical duration of the growth process was 120–150 h. Leszczynski et al. [15], Porowski et al. [16], and Grzegory et al. [70] performed nitride crystal growth from a solution under high N2 pressure. The growth experiments under high pressure were carried out in a gas pressure chamber of 30 mm internal diameter with a furnace dimension of 14 mm (1500 C) or 10 mm (1800 C) and with a boron nitride (BN) crucible containing Al, Ga, or In. The temperature was stabilized to a precision better than 10 C. Measures were made to optimize the pressure range for growth. With this optimization, the crystals grew only at a pressure for which the nitride was stable over the entire temperature range. In the case of GaN, single crystals were grown from a solution in liquid Ga under a N2 pressure of 8–17 kbar at temperatures ranging between 1300 and 1600 C. The quasi-linear temperature gradient in the process, which spanned over 5–24 h, was 30–100 C cm1. The nucleation and growth of single-crystal GaN took place through the process of dissociation and transport of thin polycrystalline GaN film deposited on a Ga surface into the cooler part of the crucible. The technique has also been applied to AlN and InN as well, with GaN being the most successful. At high N2 pressure, the synthesis rate of AlN is high and that of InN is extremely low. The rate of AlN growth is so high that, at a pressure lower than
j361
j 3 Growth and Growth Methods for Nitride Semiconductors
362
6.5 kbar, thermal explosion takes place during heating of a bulk Al sample. Owing to a low stability, the crystallization rate of AlN at 1600–1800 C is marginally low (0.5 mm h1 in the literature). On the contrary, owing to kinetic (low-temperature) and thermodynamical (low-stability) barriers, crystal-growth experiments of InN result in very small crystallites (5–50 mm), particularly when grown by slow cooling of the system from the temperatures exceeding the stability of InN. A wide variety of GaN, AlN, and InN crystals result from changes in temperature and pressure used during growth. 3.2.7.1
Seedless Growth of GaN
3.2.7.1.1 Seedless Growth of GaN by High Nitrogen Pressure Solution Growth (HNPSG) for Substrates The GaN crystals grown by the high nitrogen pressure solution method without seeding are wurtzitic and mainly have the form of hexagonal platelets. They grow at a rate of just below 0.1 mm h1, along the h1 0 1 0i like directions and have nearly perfect morphology indicating a stable layer-by-layer growth mode. As one can deduce from the form of the crystals, the growth is strongly anisotropic in that it is much faster (about 100 times) in the c-plane, as can be deduced from the size of crystals produced. This holds true for supersaturations corresponding to an average growth rate in h1 0 1 0i directions of 0.05–0.1 mm h 1. The templates are transparent with flat mirrorlike faces. The behavior of the crystals remains the same for solutions containing Mg or Be. The average size of crystals, grown without any intentional seeding, scales with the diameter of the high-pressure reactor. A photograph of a slice of one such sample prepared at UNIPRES in Warsaw [72] is displayed in Figure 3.27. Growth rate anisotropy in favor of enhanced rate along the c-plane is desirable for acceptable size templates. However, this at the expense of a much reduced growth rate in the c-direction leads to only template growth as opposed to a boule growth, which would allow many substrates to be sliced. The vertical growth rate in the c-direction can
Figure 3.27 GaN crystals grown in high pressure chambers of different sizes. The grid size corresponds to 1 mm · 1 mm. The schematic cross section of the hexagonal platelet is shown [71].
3.2 A Primer on Conventional Substrates and their Preparation for Growth
be increased by increasing the supersaturation of the Ga melt, but this has the undesirable effect of unstable needlelike forms. The supersaturation of the solution is determined primarily by the temperature of growth and its gradient, mass transport mechanisms in Ga, and by the local competing processes such as neighboring crystals because we are dealing with seedless, or spontaneous, growth. For large GaN templates to be obtained without accelerated growth near the edges and corners, it is imperative that the supersaturation be attained. It should be noted that for extremely high supersaturation, edge nucleation on the hexagonal faces of GaN platelets often occurs, which seeds the unstable growth on that particular face. The tendency toward unstable growth is stronger for the Ga-face for undoped templates and N-face for doped templates (with sufficient Mg to render the crystal p-type), resulting in rough surfaces leaving the unaffected face mirrorlike. However, if Mg concentration is not high, the aforementioned instability has no effect in that the N-face remains mirrorlike [73]. The instability is characterized as macrosteps, periodic inclusions of solvent or cellular growth structures. The polarity of the crystal surfaces can be identified by etching in hot alkali solutions, because the Ga-polar surface is inert to etching, whereas the N-polar one etches very rapidly. The validity of the etching method has been verified by performing convergent beam electron diffraction (CBED) [74] and XPS [75] measurements in the same samples. The topic is further discussed in Section 4.2, in conjunction with extended defects. The low doping level, with no bearing on stability, may be related to the position of the Fermi level that may influence the microscopic nature of growing surfaces, which is consistent with ab initio calculations and underscores the importance of both native and impurity-related point defect formation in GaN [76,77]. This would presuppose that the instabilities mentioned above may have their genesis in point defect formation for which there is no direct evidence as yet. Summarizing, the seedless growth method leads to rates of 0.05–0.1 mm h1 along the h1 0 1 0i directions and the crystal is morphologically stable. However, the crystal is morphologically stable along the h1 0 2 0i directions even for higher growth rates. This explicates that the size of the stable platelets would depend on the volume of the solution and growth time. One can argue that increasing the volume of the crucible and growth time might lead to scaling up the size of the crystal. 3.2.7.1.2 Seeded Growth of GaN by HNPSG Method for Substrates To circumvent at least some of the scalability-related issues, seeded growth can be attempted using the templates prepared by the seedless method. However, seeding growth along the c-axis seems to be much more challenging because the observed growth rates are small and the growth tends toward instability [71]. Parameters such as seed preparation and the configuration of the experiment in the context of N flow could be optimized to enhance the in-plane or out-of-plane growth rates. For example, the macrosteps at the edges result from exposure to higher nitrogen fluxes, because the supply is from the hotter part of the solution. This can be alleviated by configuring the seed to melt relation, which can also lead to much more uniform supersaturation across the growth front. Other complications result from constitutional supercooling of the solution, caused by a very low temperature gradient and high thermal and low
j363
j 3 Growth and Growth Methods for Nitride Semiconductors
364
solute diffusivity [78]. This leads to the creation of a depleted zone at the growth front, which cannot be remedied because the concentration cannot be recovered by diffusion owing to low solute diffusivity. N-type GaN platelets have been used for seeded crystallization to suppress the cellular growth on the Ga-polar (0 0 0 1) surface. The N-polar ð0 0 0 1Þ surfaces have also been used for comparison. Large positive temperature gradients of the order of 100 C cm 1 have been applied at a nominal average crystallization temperature of about 1500 C. After 20–50 h growth, the substrates with the new crystals were removed from the solution and investigated [71]. Essentially, the new growth, which was a hillock 6 mm in diameter, was transparent, colorless, and void of periodic and cellular structures. The dominant feature was the propagation of the macrosteps from the hillock center. Growth on the N-polar surface was mainly by the propagation of the macrosteps [71]. Several growth centers were also noted for similar experimental conditions. It is still not clear if this is related to differences in surface nucleation mechanisms for the different polarities or due to inadequate surface preparation. The average growth rates in these experiments were between 4 and 8 mm h1 for both polarities, which depends on supersaturation at the growth front, which, in turn, is a function of temperature gradient and the amount of liquid Ga layer on the substrate. Microscopic observation of cross sections of the samples indicates a stable growth in terms of continuity of the newly grown material, that is, inclusions of the solvent and/or voids were not observed. The interface between template (80–100 mm thick) and the newly grown crystals was not noticeable, indicating that the surface preparation and the wetting procedures were adequate. The X-ray diffraction (XRD) analysis indicated similar quality for the template and overgrowth. Future efforts should include crystallization experiments for finding the optimum configuration for a stable growth along the h0 0 0 1i directions. Of the particular issues remaining to be addressed adequately are step bunching and its reduction, and uniform supersaturation across growing surface. The uniform supersaturation can be achieved by reducing the radial temperature gradient and/or decreasing the width of the Ga layer over the substrate. Summarizing, an examination of the crystal morphology indicates that the crystal shape and size depend on the pressure, the temperature range, and the supersaturation during growth. For pressures and temperatures lying deeply in the GaN stability field (e.g., higher pressure and lower temperature), the crystals are hexagonal prisms elongated in the c-direction. Under conditions close to the equilibrium curve, the dominating shape of the crystals is a hexagonal platelet. The crystals grown slowly (slower than 0.1 mm h1) at smaller temperature gradients, exhibited higher crystalline quality. Typical full width at half maximum (FWHM) of the X-ray rocking curves for (0 0 4) Cu Ka. reflections are 23–32 arcsec. Probably owing to the nonuniform distribution of nitrogen in the solution across the growing crystal face, the quality of GaN crystals deteriorates with increasing growth rate (high supersaturation) and with increasing dimensions of the crystals. This is apparent especially when the size of the face becomes comparable to the size of the crucible. The deterioration of quality of 5–10 mm crystals grown at a rate of
3.2 A Primer on Conventional Substrates and their Preparation for Growth
0.5–1 mm h1 was evidenced by the broadening of the X-ray rocking curve. GaN substrates are not available in large quantities. Extremely high N2 equilibrium pressure on GaN will most likely preclude even pseudoconventional growth methods to fall short of producing GaN substrates. Bulk GaN substrates have been prepared under high pressures (12–20 kbar) and temperatures (1200–1600 C). Despite complications and challenges, the high-pressure method is capable of growing thicker crystals, which can be sliced into platelets, as shown in Figure 3.28. 3.2.7.2 Pertinent Surfaces of GaN Improve surface morphology of epitaxial layers, metal–semiconductor contacts, processing and effect of surface features on transport, and optical properties of GaN, requires an understanding of surface structure and growth processes on an atomic scale. For the technologically relevant surfaces, that is, polar (0 0 0 1), ð0 0 0 1Þ and nonpolar ð1 1 0 1Þ, ð1 1 0 0Þ, ð1 1 2 0Þ, experimental and theoretical investigations have led to considerable understanding of their atomic geometry, the driving forces leading to these structures, and their effect on adatom kinetics, the latter is discussed in Section 3.4.2. The polar hexagonal {0 0 0 1} basal plane is the most common surface on which to grow. Even though other surfaces have been explored for growth, the structures grown on the basal plane have exhibited the best performance. Among the basal planes, the Ga-polarity surface (0 0 0 1) is more stable and can be produced with smoother surfaces than the N-polarity ð0 0 0 1Þ surface. Figure 3.29 shows the schematic representations of top view of the ideal truncated GaN(0 0 0 1) surface as presented by Neugebauer [79]. The dashed lines depict the boundary of a (2 2) unit cell. High symmetry adsorbate sites are also marked as H3 (hollow) and T4 (atop a second layer atom). The side views along the ð1 1 0 0Þm-plane and ð1 1 2 0Þa-plane are shown in Figures 3.31 and 3.32.
Figure 3.28 GaN substrate crystal, obtained by slicing of the new material, grown by directional crystallization along c-axis; (a) before slicing where the bottom ledge corresponds to the original template as shown in the cross-sectional artistic rendition and (b) one sliced template. The grid is 1 mm · 1 mm [71].
j365
j 3 Growth and Growth Methods for Nitride Semiconductors
366
T4
H3
Ga N
Figure 3.29 Schematic representations of top view of the ideal truncated GaN (0 0 0 1) surface. The dashed lines depict the boundary of a (2 · 2) unit cell. High-symmetry adsorbate sites are also marked as H3 (hollow) and T4 (atop a second layer atom). The darker circles depict Ga atoms, and the lighter circles show N atoms, respectively. The side views along the ð1 1 0 0Þm-plane and ð1 1 2 0Þa-plane are shown in Figures 3.31 and 3.32.
Experimentally, the most commonly observed reconstruction on (0 0 0 1) GaN is (2 · 2), which corresponds to a stable growth front, yielding high-quality thin films, as discussed in Section 3.5.6. Because this reconstruction has not been observed on ð0 0 0 1Þ GaN, its presence is considered the domain of (0 0 01) GaN leading to inversion domains, which are discussed in Section 4.1.3. Other reconstructions observed by RHEED and STM on this surface are (1 2), (4 4), (5 5), and (6 4), details of which are discussed in Section 3.5.6. The ideal truncated GaN(0 0 0 1) surface with Ga in the top surface layer is shown in Figure 3.29. Tetrahedral bonds are such that each Ga surface atom has one dangling bond and is bonded to three N atoms in the second layer below. Adding one Ga or N layer on this surface would be called a Ga or N adlayer structure and adding/removing a Ga atom would be called Ga adatom/vacancy. Theoretically, this surface has been studied by several researchers employing plane wave pseudopotential methods and ab initio tight-binding methods [79]. All studies are centered on (1 1) and (2 2) structures, specifically adatoms on H3 and T4 sites, trimers, and Ga adlayers on various sites. The calculated relative formation energies are such that the (2 · 2) H3 N adatom model (with N at the He hollow site) is energetically the most stable under N-rich conditions. The (2 · 2) Ga-vacancy model gives only slightly higher energy. In contrast, under Ga-rich conditions, the (2 · 2) T4 Ga adatom structure (Ga adatom atop the atom in the second layer below) is favored. Under N-rich conditions, the Ga-vacancy and N adatom (on an H3 site) are the stable surfaces with energies close to one another. In the case of Ga-rich conditions, the Ga adatom (on a T4 site) is the lower energy structure and is thus preferred. As for the ð0 0 0 1Þ N-face, four dominant reconstructions, (1 1), (3 3), (6 6), and c(6 · 12) listed in order of increasing surface coverage, have been identified, discussed in detail in Section 3.5.6. For N-rich conditions on this surface, a (2 · 2) H3
3.2 A Primer on Conventional Substrates and their Preparation for Growth
Ga adatom model is found to be most stable. In the case of Ga-rich conditions, a (1 · 1) adlayer structure is energetically favored. In the stable (1 · 1) model, a full monolayer (ML) of Ga atoms is situated directly atop the N atoms, with the Ga–N bond length equal to 1.97 Å (compared to 1.94 Å in bulk GaN). The Ga–Ga separation in the adlayer (3.19 Å) is considerably larger than a typical Ga–Ga separation of 2.7 Å in bulk Ga. It should be pointed out that this is a completely novel structure with no known analogue among other semiconductor surfaces. In fact, this structure violates most of the empirical rules used to describe semiconductor surfaces in that it clearly disobeys electron counting and maximizes the number of dangling bonds. One immediate conclusion that can be drawn is that the surface must be metallic, which has indeed been observed by scanning tunneling spectroscopy (STS) [79]. Regarding the (3 · 3) surface, structural models with one, two, or three additional Ga adatoms on (or in) the Ga adlayer have been considered. One additional Ga adatom is the best model for the observed (3 · 3) reconstruction where the extra Ga atom resides only 0.9 Å above the adlayer plane. If no lateral relaxation is allowed, the Ga adatom must be positioned 1.8 Å above the adlayer to preserve a reasonable Ga–Ga distance. However, the extremely large inward relaxation of the adatom is enabled by a 0.5 Å lateral relaxation of the nearest-neighbor Ga-adlayer atoms. This then allows the adatom to be situated much closer to the adlayer plane, which stabilizes the structure, adopting the nomenclature that it is an in-plane adatom model. The GaN ð1 1 0 1Þ surface has been observed as a sidewall facet in lateral epitaxial overgrowth (ELO), which is used for threading defect reduction and thus deserves discussion. The ELO method is discussed in Section 3.5.5.2. These surfaces have also been found as sidewalls in the inverted pyramid defects, which form at the termination of some threading dislocations at the GaN surface during growth of In-containing alloys [79]. Northrup et al. [80] have theoretically investigated this surface by employing first-principles calculations. The schematic representation of a stable surface structure under Ga-rich conditions is given in Figure 3.30. It consists of Ga atoms in two distinct types of sites in the surface layer. These sites have been labeled as B2 and T1 sites. The atoms in B2 sites are bonded to two N atoms in the layer below. However, the atoms in T1 sites are bonded to one N atom each in the layer below, as in the (1 · 1) Ga-adlayer structure existing on the ð1 1 0 1Þ surface [79]. The structure stable under N-rich conditions contains one Ga atom per cell bonded in an H3 site. This adatom is bonded to three N atoms in the layer below. In terms of nonpolar surfaces, the ð1 1 0 0Þ and ð1 1 2 0Þ are important as they do not exhibit the polarization endemic in layers grown on polar surfaces, see Section 2.12 for a discussion of polarization. Growth on these nonpolar surfaces has progressed well, some details of which are discussed in Section 3.5.11. These surfaces also represent the sidewalls of processed (0 0 0 1) surfaces as well as being central planes in ELO. A schematic model of the GaN ð1 1 0 0Þ surface is shown in Figure 3.31. On the ð1 1 0 0Þ surface, an equal number of threefold-coordinated Ga and N atoms exist in the top surface layer. This allows charge neutrality without changes in stoichiometry or reconstruction, hence the term nonpolar. The Ga and N atoms form an array of Ga–N dimers. The ð1 1 0 0Þ surface has been studied by Northrup et al. [81] and Filippetti et al. [82] and a review by Neugebauer [79] is available that employs density–functional
j367
j 3 Growth and Growth Methods for Nitride Semiconductors
368
(0 0 0 1) c-plane B 2 site
T1 site –
(1 0 1 1) facet B 2 site
Ga N
T1 site
Figure 3.30 Schematic representation of the GaN ð1101Þ Gaadlayer surface. Surface sites T1 and B2 are sites in which a Ga atom makes one or two bonds with N atoms. The darker circles depict Ga atoms and the lighter circles N atoms. Patterned after Ref. [79].
theory calculations. The two main relaxation mechanisms in effect are a contraction of the GaN bonds by 6% accompanied by a slight buckling rehybridization with N atoms. The N atoms tend to adopt a p3 configuration, whereas Ga atoms adopt a sp2 configuration. The bond rotation angle is 7% and the surface energy is 118 meV Å2. The structure of the ð1 1 2 0Þ surface can be construed as a chain of threefoldcoordinated Ga and N atoms, as shown in Figure 3.32. In each of the unit cells, there are four surface atoms, namely two Ga and two N atoms. This surface also has been
Ga N
Figure 3.31 Schematic top view of the Wz GaN ð1 1 0 0Þ surface. Larger circles depict the atoms in the first layer and the smaller ones portray those in the second layer. The dashed lines outline the boundary of a unit cell (5.179 Å 3.171 Å). Patterned after Ref. [79].
3.2 A Primer on Conventional Substrates and their Preparation for Growth
[0 0 0 1] Ga N Figure 3.32 Schematic top view of the Wz GaN ð1 1 2 0Þ surface. The dashed lines outline the boundary of a unit cell (5.493 Å 5.179 Å). The smaller circles denote atoms in the second layer. Patterned after Ref. [79].
investigated with density–functional theory calculations [81]. The calculated Ga–N bond lengths in the surface chain are 1.85 Å (cis) and 1.87 Å (trans), representing a contraction of approximately 4–5% compared to bulk. Similar to the ð1 1 0 0Þ surface, the Ga atoms relax toward an sp2 configuration. The bond rotation angle is 7 and the surface energy is slightly larger than that for the ð1 1 0 0Þ surface by 123 meV Å 2. Contrary to the ð1 1 0 0Þ surface, the structure formed by replacing surface N atoms with Ga atoms is not energetically favored even under Ga-rich conditions. 3.2.7.3 GaN Surface Preparation for Epitaxy A very crucial step in the homoepitaxial growth of semiconductors is the attainment of an atomically clean and smooth substrate surface, without which even severe extended defects would form at the interface between the template and epitaxial overlayer. Available methods for cleaning the surface prior to epitaxy are characterized as external and internal ones. Ex-situ processes attempt to remove any metallic and organic contaminants from the surface. However, these methods always leave a thin oxide layer on the surface (of thickness similar to the native oxide), which must be removed before growth. This is typically done in situ in the epitaxial reactor. In the vapor phase techniques, the thin oxide is removed during ramp to growth temperature under a reducing environment such as H. However, removal within the MBE UHV system (in situ) usually includes a preliminary heating for outgassing of adsorbed gases in a preparation chamber and a final heating for oxide removal and preparation of an atomically clean surface within the growth chamber [83].
j369
j 3 Growth and Growth Methods for Nitride Semiconductors
370
Overviews of the ex-situ and in-situ preparation of a GaN surface are found in Refs [84,85]. Because high-quality GaN epitaxy by MBE, and perhaps by other techniques as well, requires growth on GaN templates prepared by vapor phase techniques, it is necessary to discuss this issue of GaN surface cleaning in the context of MBE for the moment. Commonly, a GaN surface that has been exposed to the air is contaminated with oxygen and carbon [84–86], which must be removed before epitaxy. XPS can be used to quantify the surface contaminants. Such studies on GaN surface oxide led to the observation of two components of the O 1s peak centered at 531.3–531.6 and 532.4–532.7 eV, respectively [85–87]. The O 1s component centered at 531.3 eV, which can be removed in HCl [86], NH4OH [87], HF, and UV/O3 (ozone treatment) is attributed to stoichiometric Ga2O3, although the oxide may also contain some component of mixed oxynitride of Ga, which is also soluble in alkali solutions. The origin of the high-energy O 1s peak at 532.7 eV, which of course can be removed with ion etching, is not yet clear. Prabhakaran et al. [87] reported only that it could be removed by argon ion sputtering while the 531.3 eV O 1s peak remained. Ar ion etching was also reported to remove the C 1s peak that, concurrently with the highenergy 532.7 eV O 1s peak, paved the way for speculation that C and O are the components of C–O bonding [86]. The high-energy O peak has been attributed to OH species (hydroxides) by King et al. [85]. This oxide component was dominant on GaN surfaces after solvent cleaning and prior to UV/O3 exposure. The carbon contamination on the GaN surface, which manifests itself with a main C 1s XPS peak at 285.7 eV, is indicative of a mixture of C–O and C–H bonding [85]. Shifting of the C 1s peak to the lower binding energy of 283.7 eV after sputter cleaning led Prabhakaran et al. [87] to suggest that this is atomic carbon and could get incorporated into the epitaxial layer. Shalish et al. [86] have observed a C 1s component at 289.9 eV, which was attributed to the carboxylic group (COOH), and a peak at 287.7 eV, which was assigned to C–Cl bonds, appearing after etching in HCl. These observations [86,87] point to the necessity of surface preparation with HCl or HF, or possibly NH4OH to remove any native oxide from the surface prior to introducing the sample into the growth system. As for C contamination, the UV/O3 oxidation treatment was found most useful [85]. After chemical preparation, the sample is still exposed to air before loading which brings us to in situ treatment. The C and O contamination could not be removed at temperatures (900–1000 C) even higher than the decomposition temperature (800 C) [84,85]. This means that annealing of GaN must be conducted in N or Ga flux to maintain the stoichiometry of the surface [85]. The difficulty associated with oxide desorption has been attributed to the strength of the Ga–N bonds [85] in that oxides probably desorb as either Ga–O or N–O species instead of O2 and this requires breaking the strong Ga–N bonds. Most of the C–O-bonded carbon was reported to desorb at temperatures between 500 and 600 C leaving behind only C–H-bonded carbon, which apparently desorbs at much higher temperatures [85]. Efficient removal of C components from the surface of GaN has been successfully conducted with NH3 and Ga fluxes at 800–900 C in the authors laboratories and elsewhere [84,85]. King et al. [85] concluded that atomically clean surfaces obtained by annealing in NH3
3.2 A Primer on Conventional Substrates and their Preparation for Growth
at 800 C exhibited 2 · 2 reconstruction in LEED. However, Bermudez et al. [84] found only partial removal of oxygen after annealing in NH3 up to 900 C. They attributed this to the fact that the reaction Ga2 O3 þ 2NH3 ! 2GaN þ 3H2 O
ð3:2Þ
is endothermic by 0.79 eV/O atom; hence, reduction of Ga2O3 by NH3 may require temperatures above 900 C, while the reaction 3C þ 4NH3 ! 3CH4 þ 2N2
ð3:3Þ
is exothermic by about 6.7 eV/C atom. In-situ RHEED observations of GaN surfaces that have been exposed to air show the absence of any amorphous native oxide layer on the surface, in contrast to that for the conventional arsenide-based semiconductors [88]. Although the GaN surface adsorbs oxygen and other impurities, owing to the large number of unsaturated dangling bonds [89], the oxidation is self limiting due to the chemical inertness of GaN. Thus, it is possible to grow GaN overlayers by MBE after an ex situ and in situ preparation consistent with the above discussion, a recipe for which is summarized below, to the point that the interface between the epitaxial overlayer and underlying template could not even be detected with TEM. Depending on the history of the template, the procedure used at VCU consists of degreasing followed by HCl or HF : H2O treatment, which, in turn, is followed by a few minutes treatment in boiling aqua regia. After rinsing and drying, the sample is loaded into the introduction chamber of MBE. 3.2.8 Other Substrates
Other substrates such as Hf, MgO, NaCl, W, and TiO2, and so on are not discussed here as they do not represent any sustained effort. However, spinel at one point got a good deal of attention and has been discussed here. Lasers employing films grown on the c-plane of sapphire utilize etched cavities because sapphire does not cleave well. The cleaving process is further complicated in that the GaN epilayers are rotated (about 30 ) with respect to the underlying sapphire substrate making it impossible to align the cleavage plane of GaN and sapphire. For this reason, Nakamura et al. [90,91] explored lasers grown on the a-plane of sapphire where the c-plane of GaN aligns normal to the substrate surface. The facets in GaN and underlying sapphire are oriented on the a-plane and c-plane, respectively. Although the quality does not compare to that on the c-plane, improved cavity formation along the m-plane (1 0 1 0) or a-plane (1 1 2 0) outweighs its reduced material quality. Moreover, laser structures on (1 1 1) MgAl2O4 (spinel substrates), which lead to wurtzite GaN along the c-plane have also been explored for injection-laser experiments [91]. Spinel cleaves along the (1 0 0) plane, inclined to the surface with cleavage following the m-plane of GaN about where the epilayer is reached. Even though the facet quality in this scheme is the best among the aforementioned approaches, material quality degradation is too severe to pull it ahead of the other approaches. It is, however, very clear that a substrate with
j371
j 3 Growth and Growth Methods for Nitride Semiconductors
372
good cleavage characteristics and on which GaN can be grown without rotation is desperately needed. Considerable progress in etched facets have been made to circumvent this problem, as discussed in Volume 3, Chapter 2.
3.3 GaN Epitaxial Relationship to Substrates
Owing to the lack of native substrates for GaN epitaxy, GaN is grown on foreign substrates, often with large lattice mismatch. Consequently, internal strain minimization, among other considerations, leads to an atomic arrangement different from that of the substrate material. The degree of this difference is substrate dependent, as discussed below. 3.3.1 Epitaxial Relationship of GaN and AlN with Sapphire
Sapphire is an ubiquitous substrate on which to grow any semiconductor, and GaN is no exception. Sapphire remains the most frequently used substrate for group III nitride epitaxial growth owing to its low cost, the availability of 3-in. diameter crystals of good quality, its transparent nature, its stability at high temperatures, and a fairly mature technology for nitride growth. Before delving into the orientational relationship between the GaN epilayer and underlying substrate, taken sapphire as default, a discussion of two notations and translation between the two is warranted. Here it should be mentioned that Miller– Bravais indices containing four indices are the most common and straightforward. However, any plane and direction can be described with three indices for the hexagonal system. These two systems are referred to as Miller indices (h k l) and Miller (h j k l) indices. The latter can be obtained from the former by adding the first two indices of the former, changing the sign and assigning it as the third Miller index in the (h j k l) notation. For example, (1 1 0) in (h k l) notation would translate to ð1 1 2 0Þ in the (h j k l) notation. Similarly, (1 0 0) would translate to ð1 0 1 0Þ and (1 0 0) would translate to ð1 0 1 0Þ. Of interest are the commonly used X-ray diffraction peaks of (0 0 2) and (1 0 2), which in the (h j k l) notation would be (0 0 0 2) and ð1 0 1 2Þ. The index in the c-direction does not change, leading to a translation from (1 0 2) in the (h k l) notation to ð1 0 1 2Þ in the (h j k l) notation. In the (h j k l) notation, a dot is often placed after the second Miller index to indicate that the crystal symmetry under question is hexagonal. The orientation order of the GaN films grown on the main sapphire planes {basal, c-plane (0 0 0 1), a-plane ð1 1 2 0Þ, and R-plane ð1 1 0 2Þ} by ECR-MBE has been studied in great detail [92–94]. The epitaxial relationship between GaN and sapphire is insensitive to the method of growth in that both MBE and OMVPE layers exhibit the same relationship. A few examples of the film/substrate epitaxial relationships are (0 0 0 1) GaN || (0 0 0 1) Al2O3 with ½2 1:0 GaN || ½1 1:0 Al2O3 and ½1 1:0 GaN || ½1 2:0 Al2O3, ð2 1:0Þ GaN || (0 1.2) Al2O3 with [0 0.1] GaN || ½0 1:1 Al2O3 and ½0 1:0 GaN ||
3.3 GaN Epitaxial Relationship to Substrates Table 3.6 Crystallographic relationship between GaN films and
sapphire substrates ([13] and references therein).
Miller indices Crystal plane (h j k l) (h k l) c a m r
(0 0 0 1) ð1 1 2 0Þ ð1 0 1 0Þ ð1 0 1 2Þ
(0 0 1) (1 1 0) (1 0 0) (1 0 2)
GaN plane || sapphire GaN direction || surface plane (h j k l) sapphire direction (h j k l)
(0 0 0 1) always ð0 0 0 1Þ or ð1 0 1 0Þ ð1 0 1 3Þ or ð1 2 1 2Þ ð1 1 2 0Þ or ð1 2 1 0Þ
½1 2 1 0 jj½1 1 0 0, ½1 2 1 0 jj½1 1 0 0 ½1 1 2 0 jj ½1 0 0 0, ½1 1 2 0 jj ½0 0 0 3 ½1 2 1 0 jj½0 0 0 1, ½1 0 1 0 jj ½1 2 1 0 ½0 0 0 1 jj½1 1 0 1, ½0 0 0 1 jj½1 0 1 1, ½1 1 0 0 jj ½1 1 2 0; or ½1 1 2 0 jj ½1 1 0 2
The data on AlN on sapphire are somewhat limited, but the available data indicate AlN exhibits similar behavior to that of GaN.
½2 1:0 Al2O3 (Refs [10–12] in [95]). The aforementioned orientational relationships in the (h j k l) notation become (0 0 0 1) GaN || (0 0 0 1) Al2O3 with ½2 1 1 0 GaN || ½1 1 0 0 Al2O3 and ½1 1 0 0 GaN || [1 2 1 0] Al2O3, ð2 1 1 0Þ GaN || ð0 1 1 2Þ Al2O3 with [0 0 0 1] GaN || ½0 1 1 1 Al2O3 and ½0 1 1 0 GaN || ½2 1 1 0 Al2O3. The second set deals with a-plane GaN on r-plane sapphire, which can also be expressed as the epitaxial relationship of ð1 1 2 0Þa-plane GaN on ð1 1 0 2Þr-plane sapphire with ½1 1 2 0GaNjj½1 1 0 2sapphire, ½0 0 0 1GaNjj½ 1 1 0 1sapphire, and ½ 1 1 0 0 GaNjj½1 1 2 0sapphire owing to hexagonal symmetry. A tabular representation of epitaxial relationships of GaN grown on various planes of sapphire is given in Table 3.6. The calculated lattice mismatch between the basal GaN and the basal sapphire plane is larger than 30%. However, the actual mismatch is smaller (16%), because the small cell of Al atoms on the basal sapphire plane is oriented 30 away from the larger sapphire unit cell. This smaller lattice mismatch can be calculated by adopting the model explained in Figure 3.33. pffiffiffi 3awGaN asapphire ¼ 0:16: ð3:4Þ asapphire It is on this plane that the best films have been grown with relatively small in-plane and out-of-plane misorientations. In general, films on this plane show either none or nearly none of the cubic GaN phase. GaN grown on the ð1 1 2 0Þa-plane turns out to be (0 0 0 1) oriented and anisotropically compressed. The effect of this uniaxial stress on the laser performance is discussed in Volume 3, Chapter 2. The in-plane relationship of GaN and sapphire is depicted in Figure 3.34 where the ½1 1 2 0 direction of GaN is aligned with the [0 0 0 1] direction of sapphire. In this orientation, the bulk positions of both the substrate and the GaN cations lie along the sapphire [0 0 0 1] direction. The mismatch between the substrate and the film is given by c sapphire 4awGaN ¼ 0:02; c sapphire
ð3:5Þ
j373
j 3 Growth and Growth Methods for Nitride Semiconductors
374
GaN[1 2 1 0] Sapphire[1 1 0 0]
GaN[1 0 1 0] Sapphire[1 1 2 0]
Al N GaN cell Sapphire cell Figure 3.33 Projection of bulk basal plane sapphire and GaN cation positions for the observed epitaxial growth orientation. The circles mark Al-atom positions and the dashed lines show the sapphire basal plane unit cells. The open circles mark the N-atom positions and solid lines show the GaN basal plane unit cell. The Al atoms on the sapphire plane sit at positions approximately 0. 5 Å above and below the plane position [92].
and for the GaN ½1 1 0 0 direction parallel to the sapphire ½1 1 0 0 direction by c sapphire 1:5awGaN ¼ 0:005: c sapphire
ð3:6Þ
In one investigation [96], the range of the growth conditions leading to good films on the a-plane was found to be wider than that on the c-plane. Another impetus for exploring growth on the a-plane is the relative ease with which the sapphire could be cleaved along the weak single bond in the c-plane. The GaN film on the top would cleave along the ð1 1 2 0Þ a-plane or the ð1 1 0 0Þ m-plane depending on the in-plane rotation of GaN with respect to sapphire. The aforementioned discussion is somewhat academic in that GaN grown on a-plane still has [0 0 0 1] direction normal to the surface. However, growth on r-plane with OMVPE leads to a-plane GaN, the details of which are discussed in Section 3.5.11. GaN films have also been grown on the r-face f1 1 0 2g of sapphire purportedly to achieve a lattice mismatch smaller than on the c-plane sapphire. Films grown on the r-face has been reported to assume an orientation similar to f2 1 1 0g. The arrangement in the case of the ð1102Þ face of sapphire and ð2 1 1 0Þ of GaN is depicted in Figure 3.35a. Although of no immediate impact on the topic under discussion, Figure 3.35b provides an image for us to gain some familiarity with the r-plane of GaN in relation to the basic hexagonal lattice structure. The lattice
3.3 GaN Epitaxial Relationship to Substrates
GaN[1 1 2 0] Sapphire[0 0 0 1]
GaN[1 1 0 0] Sapphire[1 1 0 0]
Al N GaN cell Sapphire cell Figure 3.34 Projection of bulk a-plane sapphire and basal plane GaN cation positions for the observed epitaxial growth orientation. The solid circles mark the Al-atom positions and the dashed lines show the sapphire a-plan unit cells. The open circles mark the N-atom positions and the solid lines show the GaN basal plane unit cell. The mismatch along the GaN ½1 1 2 0 direction
parallel to sapphire [0 0 0 1] direction is 2%. The mismatch along the GaN ½1100 direction parallel to sapphire ½1 1 0 0 direction is – 5%. The lines drawn through the N atoms are for underscoring hexagonal symmetry, not the bonds as the bonds are formed between the Ga atoms on plane above the nitrogen plane and N atoms [92].
mismatch between the ½1101 direction of sapphire and the [0 0 0 1] direction of GaN parallel to the sapphire ½1101 direction is equal to r ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 3c GaN 3a2sapphire þ c 2sapphire rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ¼ 0:01: ð3:7Þ 3a2sapphire þ c 2sapphire In the case when the ½1 1 0 0 direction of GaN is parallel to the sapphire ð0 1 2 0Þ direction, the lattice mismatch is pffiffiffi awGaN ðasapphire = 3Þ pffiffiffi ¼ 0:16: ð3:8Þ ðasapphire = 3Þ
j375
j 3 Growth and Growth Methods for Nitride Semiconductors
376
GaN[1 1 0 0]
4.76 Å
Sapphire[1 1 2 0]
GaN[0 0 0 1]
5.12 Å N
Sapphire[1 1 0 1]
Ga Al
(a)
Figure 3.35 (a) Projection of bulk r-plane sapphire and a-plane GaN anion and cation positions for the observed epitaxial growth orientation. During growth it is assumed that the top O layer desorbs leaving behind the Al layer with its nearly square unit cells (dimensions are given in terms of angstroms). The figure should be treated with some caution as the exact placement of N atoms belonging to the GaN cell has been done somewhat arbitrarily. Calculations that can shed some light on the preferred position of Ga and N atoms with respect to Al are lacking. Because the a-plane of GaN is nonpolar, only the N and Ga atoms on the same plane are shown. The Ga atoms in the first layer would be lacking a lattice site to bond, making dangling bonds very plausible in addition to sever
distortion of the lattice as even the N atoms in the GaN unit cells not to align everywhere with the underlying Al atoms. Even though some N atoms appear to vertically align with Al below, that bond should in reality not be what is loosely referred as the long bond as that would imply the c-direction to be out of the plane, which is not the case. However, the image provides a reasonably good picture of how the a-plane GaN lattice is stacked on the r-plane sapphire. The lines connecting Al atoms represent only the unit cell not the bonds. For a 3D view of a plane GaN on r-plane sapphire, see Figure 3.161. (b) Although it is not of an immediate application, this figure illustrates the r-plane of GaN with its associated directions. In addition, the figure on the lower left indicates the placement of the r-plane in a hexagonal cell.
3.3 GaN Epitaxial Relationship to Substrates
One unit cell r-plane, GaN[1 1 0 2]
GaN[1 1 2 0] [1 1 0 1] N
c
Ga View direction Bars represent Ga–N bonds
r-plane (1 1 0 2)
c/2 b (b)
a
Figure 3.35 (Continued)
The mismatch along the [0 0 0 1] direction of GaN parallel to the ½ 1 1 0 1 direction of sapphire is 1%, which is much smaller than the 16% mismatch along ½1 1 0 0 direction of GaN parallel to the ½1 1 2 0 direction of sapphire. Growth on the r-face exhibits ridgelike features that allow relaxation of the mismatch. It is assumed, as in the case of c-plane of sapphire, the topmost O layer is desorbed and the Al layer of sapphire is then exposed. Another mechanism accomplishing the same is also implicitly assumed. A closer look, however, indicates that while the above argument about lattice mismatch would hold for only the Al–N bonds at the corners of the unit cell of sapphire, shown as rectangles in Figure 3.35a, it does not necessarily hold for the Al–N bond at the center of the unit cell. In addition, the Ga atoms that must be on the same plane as N in GaN (true for a nonpolar a-plane that is the topic of discussion here) will not have one of its bonds satisfied and would be left dangling. We must hasten to state that the representation in Figure 3.35a is a very simple one intended only to give the reader a first-order glimpse as to how the a-plane GaN might be organized on the r-plane sapphire with no consideration to energy minimization. This awaits calculations for additional insight, which is missing. More details on the r-plane sapphire can be found in Figure 3.20. The lattice mismatch between GaN and all the other substrates, inclusive of various epitaxial relationships, is tabulated in Table 3.7 for both completeness and convenience. The discussion here details the intricacies of lattice mismatch as affected by the epitaxial relationship driven by strain minimization brought about in the first place by the lattice-mismatched substrates.
j377
[0 0 0 1]//[0 0 0 1]
ð1 1 2 0Þ=ð1 1 2 0Þ (0 0 0 1)/(0 0 0 1)
Zinc blende (a ¼ 4.33)
Hexagonal (a ¼ 4.7589, c ¼ 12.991)
a-Al2O3
ð2 1 1 0Þ=ð0 1 1 2Þ
ð0 0 0 1Þ=ð2 1 1 0Þ
ð0 1 1 3Þ=ð0 1 1 0Þ
½1 0 1 0==½1 0 1 0
(0 0 0 1)/(0 0 0 1)
Wurtzite (a ¼ 3.1129, c ¼ 4.9819)
AlN
—
Zinc blende (a ¼ 4.511)
½0 0 0 1==½0 1 1 1
½0 1 1 0==½2 1 1 0
½2 1 1 0==½0 0 0 1
½0 1 1 0==½0 1 1 0
½2 1 1 0==½0 0 0 1
½0 3 3 2==½2 1 1 0
½0 1 1 0==½2 1 1 0
½21 10==½0110
—
—
—
0.18
0.06
—
—
Lattice misfit (%) Thermal strain (%) (growth temperature) (1000–25 C)
aGaN aAlN ¼ 2:41 2.35 aAlN c GaN c AlN ¼ 4:08 c AlN pffiffiffi 2aGaN 3aAl2 O3 pffiffiffi ¼ 22:65 22.83 pffiffiffi 3aAl2 O3 3aGaN aAl2 O3 ¼ 16:02 aAl2 O3 xaGaN c Al2 O3 ¼ 6:09; c Al2 O3 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffi where x ¼ c 2 þ ð2 3aÞ2 4aGaN c Al2 O3 ¼ 1:85 pffiffiffi c Al2 O3 pffiffiffi 3aGaN 3aAl2 O3 pffiffiffi ¼ 33:01 3aAl2 O3 pffiffiffi 3aGaN aAl2 O3 ¼ 16:02 aAl2 O3 4aGaN c Al2 O3 ¼ 1:85 c Al2 O3 3cGaN x Al2 O3 ¼ 1:19; x Al2 O3 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffi where x ¼ c2 þ ð2 3aÞ2
—
—
—
Wurtzite (a ¼ 3.1878, c ¼ 5.185)
GaN
—
In-plane direction Lattice misfit (%) Epitaxial relationship (GaN || substrate) (room temperature)
Lattice parameter (Å)
Crystal
substrates inclusive of various epitaxial relationships between GaN and substrates mentioned.
Table 3.7 Room-temperature lattice mismatch of GaN with other III-N compounds and
378
j 3 Growth and Growth Methods for Nitride Semiconductors
Zinc blende (a ¼ 4.35997)
Hexagonal (a ¼ 3.07997, c ¼ 10.083) (0 0 0 1)/(0 0 0 1)
3C-SiC
4H-SiC
Orthorhombic (0 0 0 1)/(0 0 1) (a ¼ 5.4063, b ¼ 6.3786, c ¼ 5.0129)
Orthorhombic (a ¼ 5.428, b ¼ 5.498, c ¼ 7.710)
Zinc blende (a ¼ 5.4309)
Zinc blende (a ¼ 5.652)
b-LiGaO2
NdGaO3
Si
GaAs (0 0 1)/(0 0 1)
(0 0 0 1)/(1 1 1)
(0 0 1)/(0 0 1)
(0 0 1)/(1 1 1)
(0 0 0 1)/(1 1 1)
(0 0 0 1)/(1 0 1)
(0 0 0 1)/(0 1 1)
(0 0 0 1)/(0 0 1)
Tetragonal (a ¼ 5.169, c ¼ 6.282)
g-LiAlO2
ð1 1 0 0Þ=ð1 0 0Þ
(0 0 1)/(0 0 1)
Hexagonal (a ¼ 3.0806, c ¼ 15.1173) (0 0 0 1)/(0 0 0 1)
6H-SiC
[0 1 0]//[0 1 0]
aGaN a6H-SiC ½1 0 1 0==½1 0 1 0 ¼ 3:48 a6H- Sic aGaN a3C-SiC ¼ 3:46 [0 1 0]//[0 1 0] a3C-SiC aGaN a4H--SiC ¼ 3:50 ½1 1 2 0==½1 1 2 0 a4H--SiC c GaN aLiAlO2 [0 0 0 1]//[0 1 0] ¼ 0:31 pffiffiffi aLiAlO2 3aGaN aLiAlO2 ¼ 6:8 ½1 0 1 0==½0 1 0 pffiffiffi aLiAlO2 3aGaN - aLiGaO2 ½1 0 1 0==½1 0 0 ¼ 0:13 aLiGaO2 ða-directionÞ p ffiffiffi 3aGaN bLiGaO2 ¼ 13:43 ½1 0 1 0==½0 1 0 bLiGaO2 ðb-directionÞ p ffiffiffi 3aGaN aNdGaO3 ½1 0 1 0==½1 0 0 ¼ 1:72 pffiffiffi aNdGaO3 3 a b GaN NdGaO3 ¼ 0:43 ½1 0 1 0==½0 1 0 bNdGaO pffiffiffi3 2a 2 a GaN Si pffiffiffi ¼ 16:99 ½1 0 2 0==½1 1 0 2aSi a a GaN Si ¼ 16:93 ½110==½101 aSi pffiffiffi 2aGaN 2aGaAs pffiffiffi ¼ 20:19 [0 1 0]//[0 1 0] 2aGaAsi aGaN aGaAs ¼ 20:19 ½1 0 2 0==½1 1 0 aGaAs
In-plane direction Lattice misfit (%) Epitaxial relationship (GaN || substrate) (room temperature)
Lattice parameter (Å)
Crystal
20.26
16.8
1.06
15.1 (b-direction)
1.59 (a-direction)
0.1
3.53
3.55
3.49
(Continued)
0.07
0.19
0.66
1.67 (b-direction)
0.54 (a-direction)
0.41
0.03
0.09
0.01
Lattice misfit (%) Thermal strain (%) (growth temperature) (1000–25 C)
3.3 GaN Epitaxial Relationship to Substrates
j379
Zinc blende (a ¼ 4.216)
Wurtzite (a ¼ 3.252, c ¼ 5.213)
MgO
ZnO
(0 0 0 1)/(1 1 1)
Mixed perovskite (a ¼ 7.730)
LSAT
aGaN aMgO ¼ 6:99 pffiffiffi aMgO pffiffiffi 3aGaN 2aMgO pffiffiffi ¼ 7:99 ½1 0 1 0==½1 1 0 2aMgO a a GaN ZnO ¼ 1:97 ½1 0 2 0==½1 0 2 0 aZnO aGaN - aScAlMgO4 ½1 0 2 0==½1 0 2 0 ¼ 1:49 aScAlMgO pffiffi4ffi 4a 2 a GaN MgAl 2 O4 pffiffiffi ½1 0 2 0==½1 1 0 ¼ 11:55 2ap ffiffiffi 2 O4 MgAl 4a 2 a GaN LSAT pffiffiffi ¼ 16:64 ½1 0 2 0==½1 1 0 2aLSAT
[0 1 0]/[0 1 0]
5.8
16.68
11.61
1.64
0.04
0.06
0.15
0.21
1.19
Lattice misfit (%) Thermal strain (%) (growth temperature) (1000–25 C)
2.18
The growth temperature for each of the cases is tabulated in Table 2.36. Courtesy of J. Chaudhuri, Wichita State University.
(0 0 0 1)/(1 1 1)
Spinel (a ¼ 8.083)
(0 0 0 1)/(0 0 0 1)
(0 0 0 1)/(0 0 0 1)
(0 0 0 1)/(1 1 1)
(0 0 1)/(0 0 1)
In-plane direction Lattice misfit (%) Epitaxial relationship (GaN || substrate) (room temperature)
MgAl2O4
ScAlMgO4 Hexagonal (a ¼ 3.236, c ¼ 25.15)
Lattice parameter (Å)
Crystal
Table 3.7 (Continued)
380
j 3 Growth and Growth Methods for Nitride Semiconductors
3.3 GaN Epitaxial Relationship to Substrates
3.3.2 Epitaxial Relationship of GaN and AlN with SiC
Both AlN and GaN deposit on hexagonal SiC substrates in a simple epitaxial relationship, that is, [0 0 0 2]III-N//[0 0 0 6]SiC and ½1 1 2 0III-N ==½1 1 2 0SiC . Using ð1 1 0 0Þ 6H-SiC substrates (not readily available commercially), Horino et al. [97] established the epitaxial relationship as ð1 1 0 0ÞGaN==ð1 1 0 0ÞSiC and ½1 1 2 0 III-N==½1 1 2 0SiC. 3.3.3 Epitaxial Relationship of GaN and AlN with Si
The epitaxial relationship of GaN on Si is GaN h0 0 0 1i||Sih1 1 1i and h2 1 1 0i GaNjjh0 1 1iSi. If the wurtzitic phase is produced on (0 0 1) Si, then the epitaxial relationship is (0 0 0 1) GaN || (0 0 1) Si and ½1 1 2 0 GaN || [1 1 0] Si. AlN on Si(1 1 1) showed two epitaxial relationships [13,98] namely AlN(0 0 0 1) ½2 1 1 0 || Si(1 1 1) ½0 2 2 for deposition temperatures greater than 650 C and AlN(0 0 0 1) ½1 1 0 0 || Si(1 1 1) ½0 2 2. These two relationships correspond to the (4 4) and (7 7) lattices, respectively. AlN grown at low temperatures of 400–600 C by MBE shows the relationship of AlN(0 0 0 1) ½0 1 1 0 || Si(1 1 1) ½1 1 2 [99]. These results are given in Table 3.8 and the atomic distance mismatches in Table 3.9. Figure 3.36 shows the orientational relationship between wurtzite AlN and the underlying Si (0 0 1) substrate. GaN deposited on Si(1 1 1) with an AlN buffer layer shows Ga polarity, and GaN grown directly on Si(1 1 1) shows N polarity. For details, refer to Refs [13,100,101], assuming the epitaxial relationship between AlN and Si to be the same as that between GaN and Si. 3.3.4 Epitaxial Relationship of GaN with ZnO
Epitaxial relationships between GaN and ZnO are expected to be [0 0 0 1] of GaN being parallel to [0 0 0 1] of ZnO. Because the lattice mismatch between GaN and ZnO in the Table 3.8 Epitaxial relationship of Wz GaN and AlN grown on Si
when buffer conditions are such that wurtzitic form of GaN or AlN is produced. Si crystal plane Miller indices (h j k l) or (h k l)
GaN plane || Si plane in the first column
GaN direction || Si direction
(1 1 1) (0 0 1) if Wz GaN is produced (1 1 1) if Wz AlN is produced (Ts > 650 C)
(0 0 0 1) (0 0 0 1) (0 0 0 1) (0 0 0 1) (0 0 0 1)
½2 1 1 0 jj ½0 1 1 ½1 1 2 0 jj ½1 1 0 ½2 1 1 0jj½0 2 2 ½1 0 1 0jj½0 2 2 ½0 1 1 0jj½1 1 2
(1 1 1) if Wz AlN is produced (Ts ¼ 400–600 C)
j381
j 3 Growth and Growth Methods for Nitride Semiconductors
382
Table 3.9 Epitaxial relationship of GaN grown on Si with the
atomic distance mismatch (ADM) and the extended atomic distance mismatch (EADM).
Epilayer/substrate ð0 0 1Þh-GaN/(1 0 0)Si
(1 0 0)c-GaN/(1 0 0)Si
(0 0 1)h-GaN/(1 1 1)Si
Atomic distance (Å) of epilayer/substrate [0 1 0]Ga–Ga/[0 1 0] Si–Si 3.186 5.431 [2 1 0]Ga–Ga/[1 0 0] Si–Si 5.518 5.431 [0 1 0]Ga–Ga/[0 1 0] Si–Si 4.500 5.431 [0 1 0]Ga–Ga/[1 1 0] Si–Si 3.186 3.84
ADM (%)
EADM (%) of epilayer/substrate
41.3
(Ga–Ga) · 5/(Si–Si) · 3
1.6
2.5 (Ga–Ga) · 1/(Si–Si) · 1
17.1
1.6 (Ga–Ga) · 6/(Si–Si) · 5
17.0
0.571 (Ga–Ga) · 6/(Si–Si) · 5 0.438
For details, refer to Refs [13,100,101].
c-plane is very small, under 2%, no discernible in-plane rotation for strain reduction is expected. However, if ZnO is deposited on sapphire followed by GaN deposition, the sapphire substrate and both epitaxial layers are oriented toward each other by a 30 rotation of the unit cell. That is, the in-plane epitaxial layer and substrate are in the form of ZnO, GaN ½1 0 1 0 k Al2O3½1 1 2 0, GaN ½1 1 0 0k Al2O3½1 1 2 1 0, GaN or ZnO ½2 1 1 0 k Al2O3½1 1 0 0, all of which indicates a 30 rotation of the epitaxial layer with respect to sapphire substrate. 3.3.5 Epitaxial Relationship of GaN with LiGaO2 and LiAlO2 and Perovskites
Some oxides have also been explored because of the small lattice misfit with GaN they provide, with LiGaO2 and LiAlO2 being the ones, particularly the latter, generating a good deal of interest. The epitaxial relationships between GaN and LiGaO2 are expected to be ½1 1 2 0 GaN || [0 1 0] LiGaO2 and [0 0 0 1] GaN || [0 0 1] LiGaO2. On the contrary, the epitaxial relationships between GaN and LiAlO2 are expected to 1 1 0 GaN//[0 0 1] LiAlO2, ð1 1 0 0Þ GaN// be ð0 1 1 0Þ GaN//(1 0 0) LiAlO2 with ½2 g-LiAlO2 (1 0 0) with ½1 1 0 0GaNjj½1 0 0LiAlO2 and ½1 1 2 0GaNjj½0 0 1LiAlO2 . The lattice misfit strain between GaN and LiAlO2 with ð1 1 0 0Þ=ð1 0 0Þ and along the [0 0 0 1]//[0 1 0] in-plane is 0.41%. The lattice structure of LiGaO2 is similar to the wurtzite structure. However, owing to Li and Ga atoms having different ionic radii, the crystal has orthorhombic structure [102]. Figure 3.37 shows the transformation of the hexagonal unit cell of GaN to an orthorhombic cell that has lattice parameters close to that of LiGaO2 (Table 3.10). Perovskite oxides have also been employed in the growth of GaN in an effort to attain a better match compared to more conventional substrates or for applications
3.3 GaN Epitaxial Relationship to Substrates
1
2
Dangling bonds
Dangling bonds Si top layer Si first underlayer [0 1 0] 2
〈2 1 1 0〉
Si second underlayer Al or N sublattice
[1 0 0] [0 0 1]//[ 0 0 0 1] 〈1 1 1 0〉
Figure 3.36 Atomic arrangement for the heteroepitaxial nucleation of 2H-AlN on the Si(0 0 1) surface. The two AlN domains with a 30 rotation are formed on neighboring terraces (1) and (2), separated by a single atomic step boundary according to the Si dangling bond directions. Refs [13,101].
where the properties of nitride semiconductors and electro-optic and nonlinear optics properties of perovskites could be offered in the same template/stack. Among the perovskite is NdGaO3 with its orthorhombic unit cell. The lattice mismatch of GaN with NdGaO3 has been calculated [102] by assuming a perovskite cell of NdGaO3 with lattice parameters a, b, and c each equal to 3.86 Å. This is followed by creating a new unit cell with a0 and b0 where a0 and b0 are the diagonals of the old perovskite cell, as shown in Figure 3.38. The c0 -axis of the new cell is naturally parallel to the c-axis of the perovskite cell but its length is doubled. Basically, this operation transforms a perovskite unit cell to a tetragonal unit cell representing GaN, as tabulated in
j383
j 3 Growth and Growth Methods for Nitride Semiconductors
384
GaN b a a'
b'
Figure 3.37 Transformation of a hexagonal unit cell to an orthorhombic unit cell. Courtesy of J. Chaudhuri and Ref. [102].
Table 3.11. Accordingly, (1 0 0) plane becomes ð1 1 0Þ plane and (0 0 1) becomes (0 0 1). The corresponding lattice misfit is 1.72% and misfit strain is 0.66%.
3.4 Nitride Growth Techniques
HVPE, OMVPE (inorganic VPE), RMBE, and bulk crystal growth from Ga solution are the main growth methods used for nitrides. By far the most frequently used methods are the variants of VPE methods. Although HVPE is used to produce thick GaN layers, including those thick enough to be self-supporting once peeled from the sapphire substrate, OMVPE produces sharp heterojunctions for devices. With the exception of FETs, OMVPE is the primary method employed in the investigation and production of optoelectronic devices, such as LEDs and lasers, albeit the quality of MBE films grown on HVPE buffers is slightly better than of those grown by OMVPE. Inorganic VPE was the first method used to grow epitaxial III-N semiconductors, but was nearly abandoned. The technique, however, was revived recently by growing very high quality and thick buffer layers and templates for the growth of device structures by MBE and OMVPE.
Table 3.10 Transformation of GaN hexagonal unit cell to an orthorhombic unit cell [102].
GaN original hexagonal cell with a and b lattice parameters
GaN transformed orthorhombic cell with lattice parameters a0 and b0
LiGaO2 orthorhombic cell
a ¼ 3.189 Å b ¼ 3.189 Å c ¼ 5.185 Å
pffiffiffi a ¼ 3a ¼ 5:52 Å 0 b ¼ 2b ¼ 6.38 Å c0 ¼ 5.185 · 1 Å
a ¼ 5.4063 Å b ¼ 6.3786 Å c ¼ 5.0129 Å
0
3.4 Nitride Growth Techniques
a
b'
a'
b
Figure 3.38 Transformation of a perovskite unit cell to a tetragonal unit cell. Courtesy of J. Chaudhuri and Ref. [102].
3.4.1 Vapor Phase Epitaxy
VPE has long been employed for the growth of many semiconductor structures. With ongoing source developments and improved reactor designs, this technique has become very powerful, particularly for GaN and related materials. Growth from the vapor phase is categorized on the basis of the sources used. If the sources are inorganic in nature, the term inorganic vapor phase epitaxy is used. This too can be subdivided on the basis of the sources used. For example, if a hydride source is used for the group V element, the term hydride vapor phase epitaxy is applied. If at least some of the sources are organic in nature, the terms organometallic vapor phase eptixay, organometallic chemical vapor deposition, metalorganic chemical vapor deposition, or metalorganic vapor phase epitaxy are employed. 3.4.1.1 Hydride Vapor Phase Epitaxy The genesis of the HVPE growth method can be traced to its wide use in silicon and conventional III–V semiconductors. In the early 1960s, the development of the halide precursor techniques applied to Si and Ge provided the foundation for their subsequent application to the growth of GaAs, which was coming to eminence. The HVPE method has since played an important role in the growth of III–V
Table 3.11 Transformation of a perovskite unit cell to a tetragonal unit cell.
Lattice parameters a b c
Perovskite unit cell
New unit cell
Final unit cell
3.86 Å 3.86 Å 3.86 Å
pffiffiffi a= ¼ 3:86pffiffi2ffi = b ¼ 3:86 2 c0 ¼ 3.86 · 2
a0 ¼ 5.43 Å b0 ¼ 5.43 Å c0 ¼ 7.72 Å
Courtesy of J. Chaudhuri [102].
j385
j 3 Growth and Growth Methods for Nitride Semiconductors
386
semiconductors. In fact, it has the unique distinction of being the first method to produce AlN [103] and, as reported by Maruska et al. [17], to produce single crystalline GaN with quality sufficient to launch the first stages of GaN technology that gained so much prominence in the 1980s. Typical thicknesses for these deposits were in the range of 50–150 mm. The advantage of this technique is that it is conducive to the growth of thick buffer layers at high growth rates on any available substrates to be used as templates for OMVPE and MBE growth, particularly the latter, of high-quality heterostructures with relatively low defect concentrations. A comprehensive treatment of this method can be found in Ref. [104]. A brief historical treatment and a succinct review will be given. After the report of Maruska et al. [17], Wickenden et al. [105] reported on GaN deposition on a-SiC and a-Al2O3, meaning wurtzitic varieties, in 1971, and Ilegems [106] obtained 100–200 mm thick single-crystalline GaN layers on sapphire substrates in 1972. Continuing on, Shintani et al. [107] investigated in 1974 the effects of the important growth parameters, such as the position of the substrate in view of the gas flow dynamics in the reactor, the reactant gas flow rate, and the substrate temperature, on the epitaxial growth rate of GaN on (0 0 0 1) sapphire substrates. Sano et al. [108] considered the influence of the surface sapphire orientation on the growth rate in 1976. In 1977, Madar et al. [109] and Jacob et al. [110] achieved doped GaN with n-type conductivity on sapphire substrates. A study on the growth rate of GaN in hydrogen as well as in inert gas ambience was undertaken by Seifert et al. [111] in 1981, which resulted in growth rates up to 800 mm h1. Owing to difficulties associated with uniform seeding of GaN on sapphire, coalescence of islands in a timely manner and resultant high n-type background doping (typically 1019 cm3), large defect concentrations, and inability to produce p-type GaN for light emitters, this technique was largely abandoned in the early 1980s, although Maruska et al. [112] later showed that Zn and Mg doping could be achieved by the simultaneous evaporation of the dopant source in the HCl stream. Since the first report of Maruska and Tietjen [17] many reports on GaN growth [113–124] became available in the literature. Further, several others [103,125,126] have extended this method to the growth of high-quality AlN. But when low-temperature nucleation buffer layers were employed in the context of OMVPE, as discussed in Section 3.5.5.1, followed shortly thereafter by reports of p-type conductivity in GaN, the HVPE method reappeared [127–129] because, in part, of its improved ability to grow thick GaN films with relatively low defect concentration and new techniques for nucleation layers (NLs). Further improvements resulted from lateral epitaxial overgrowth on patterned SiO2 masks [130–132], the concept of which is discussed in Section 3.5.5.2. Freestanding GaN substrates have been prepared by Kim et al. [133] and Melnik et al. [134], which eventually culminated in the production of very high quality freestanding GaN templates prepared at Samsung Advanced Institute of Technology [135], the detailed characteristics of which are discussed in Section 3.5.1.2. Despite these remarkable achievements, the nitride materials still suffer from a very high defect density because of the lattice mismatch between the nitrides and all the available foreign substrates. Recent twostep processes employing low-temperature GaN buffer layers [136] and techniques
3.4 Nitride Growth Techniques
for substrate removal [135,137] have shown good-quality materials with very promising characteristics. The development of the HVPE technique combined with other techniques for producing GaN templates may be the key to resolve the high density defect issue in the III nitride device technology. In HVPE, the group III precursors are chlorides formed by flowing hydrogen chloride gas over the liquid metal in a quartz tube. The group V precursors are hydrides fed into the reaction chamber by a separate quartz line in order to avoid premature reaction with molten source metal. For GaN growth, the chloride and hydride precursors are GaCl, which is formed by reacting Cl from HCl gas with molten Ga, and NH3, respectively. The GaCl in vapor phase is transported to the deposition zone by a carrier gas, which can be hydrogen, and/or an inert gas. The pressure inside the reaction chamber is kept at the atmospheric pressure. The reactor walls are made of high-purity quartz tube. For high-quality semiconductor films to result, the gases are of electronic quality with purity better than 1 ppm for contaminants. For nitrogen or hydrogen, each impurity is in concentration below 1 ppm. For the other gases, the total concentration of all the impurities is below 1 ppm. The metallic sources employed are of 7N, meaning they have a purity of 99.99999%. A multiple zone furnace is used as the metal source zone and substrate temperatures are different. In particular, the zone containing the metallic gallium, the central zone where the gases are homogeneously mixed, and the zone where the substrate resides and the deposition takes place are all kept at different temperatures, as shown schematically in Figure 3.39. The three-temperature zone allows one to independently set the partial pressure of each species, such as the chlorides and hydrides, and explore optimum growth conditions systematically. The influence of parametric variations in the vapor phase composition on the growth rate becomes relatively easier. The same applies to physical processes taking place during a growth. In brief, the vapor phase composition depends on the metallic source efficiency, the ammonia decomposition, and the flow of various gases introduced into the reactor. The vapor phase composition, the partial pressures of the various reactive gaseous species, and the temperature of the three zones in the reactor determine the growth rate and the solid composition of the epitaxial layer if ternaries are attempted. Obviously, the growth takes place in a thermodynamical equilibrium and a wide range of conditions can be applied. If a large flow of HCl is not introduced into the reactor, the Source zone
Mixing zone
Deposition zone
NH3
NH3
N2/H2+HCl
N2/H2+HCl GaCl
N2/H2+HCl add
Substrate
GaCl HCl H2/N2/NH3
Ga source
Bypass line
Bypass
Three-zone furnace Figure 3.39 Schematic diagram of a three-zone HVPE reactor, which utilizes Ga and ammonia sources, used for nitride growth. Patterned after Ref. [104].
Exhaust
j387
j 3 Growth and Growth Methods for Nitride Semiconductors
388
use of NH3 leads to initial conditions far from thermodynamical equilibrium. The growth rates from a low of 1 mm h1 to a high of more than 100 mm h1 can be achieved [104]. In HVPE as employed by Maruska and Tietjen [17] for GaN, HCl vapor flowing over a Ga melt causes the formation of GaCl, which is transported downstream. On the substrate surface, the GaCl reacts with NH3 and leads to GaN through the following chemical reaction: 2GaðlÞ þ 2HClðgÞ ! 2GaClðgÞ þ H2 ðgÞ; GaClðgÞ þ NH3 ðgÞ ! GaNðsÞ þ HClðgÞ þ H2 ðgÞ;
ð3:9Þ
where g, l, and s depict gaseous, liquid, and solid species, respectively. The efficiency of the first reaction had been estimated by Ban [138] to be about 99.5%. Using the GaAs analogy [139], two thermodynamic reaction pathways leading to the deposition of GaN [140] can be forwarded: GaClðgÞ þNH3 ðgÞ()GaN þ HClðgÞ þ H2 ðgÞ; 3GaClðgÞ þ 2NH3 ðgÞ()2GaN þ GaCl3 ðgÞ þ 3H2 ðgÞ:
ð3:10Þ
The gaseous species in the reactor are GaCl, GaCl3, HCl, NH3, H2, and either H2 or some other inert gas is used as carrier gas. Barin [141] used thermodynamical data to calculate the reactions of Equations 3.9 and 3.10. For additional details, see Ref. [104]. The supply of GaCl is controlled by the Ga boat temperature and the flow rates of the HCl gas and the carrier gas. The reactor in this method is heated by a two-zone resistance furnace with the region containing the Ga boat being kept at a different temperature than the region housing the substrate for reaction, as shown in Figure 3.39. The Ga zone temperature impacts the growth rate a great deal. The Ga source is held at a constant temperature between 850 and 900 C. The reaction efficiency of HCl with Ga is near unity. Although dependent on the reactor itself, the typical flow rates are about tens of sccm for HCl, 1 l min1 for NH3 and 2 l min1 for the carrier gas. GaN films can be grown with rates up to 1 mm h1 on Al2O3 substrates at atmospheric pressure. However, those high growth rates deplete the Ga source material rather quickly and lead to very rough surfaces with columnar growth. The substrate zone temperature varies between 1050 and 1200 C. At lower substrate zone temperatures, the growth rate decreases exponentially owing to the decreasing pyrolysis efficiency of the GaCl and NH3. On the contrary, at higher substrate zone temperatures, thermally induced decomposition of reactants reduces the growth rate. The hydrogen ambience also aids this reduction in that competing processes such as GaHx would take place. The main issue of concern associated with HVPE and other nitride growth methods is that the initial nucleation layer on sapphire substrates determines to a larger extent the material properties of the subsequent epitaxial layer than in other methods [142–144]. The nucleation or the prelayer is typically deposited using a GaCl or NH3 pretreatment consisting of flowing GaCl or NH3 over the sapphire surface prior to the initiation of growth at high temperatures. In other cases, a ZnO wetting layer is used [127,145,146]. Nitridation of sapphire has been mentioned as a means to improve HVPE materials by several groups [147,148].
3.4 Nitride Growth Techniques
Thermal dissociation of group V species during HVPE of GaAs leads to the formation of As2 or As4 molecules, which remain volatile and chemically reactive and thus participate in the film growth. However, in the case of GaN, by-products of decomposed NH3 are N2 and H2, and N2 molecules that are stable and unreactive at the temperature of interest. Other forms of N species such as NCl3 are not considered, as they are explosive. Moreover, there is a strong thermodynamic driving force for forming parasitic gas-phase reactions, which cause deposition on the walls, making the growth mechanism difficult to unravel. The process also tends to produce large amounts of NH4Cl, GaCl3, and GaCl3-NH3, which condense and clog the exhaust lines unless they are heated to sufficiently high temperatures (>150 C). Exchange reactions with the hot quartz walls of the reactor make it difficult to use HVPE for aluminum- and magnesium-bearing compounds required for AlGaN growth and efficient p-doping, respectively. Endemic to MBE and not considered that important in the case of vapor phase deposition techniques until the advent of GaN is the concept of kinetics, involving adsorption and desorption discussed in detail for OMVPE in Section 3.4.1.3 and for MBE in Section 3.4.2. Adsorption and desorption processes depend on the kinetics of the gaseous species on the substrate surface and the diffusion kinetics of the adatoms or admolecules before arriving at the incorporation sites dubbed half-crystal or K sites (see Figure 3.40) [104]. To put it simply, kinetics consists of complex but simplified approaches that can be considered to get insight into the growth resulting from two superficial diffusion flows, namely, the NGa admolecules and the other from the NGaCl admolecules. Near the step edges, only the NGa flow directly leads to incorporation through desorption of chlorine from the NGaCl. Because GaN growth relies on GaCl, which is identical to that in GaAs, much of the GaN work benefited from systematic investigations performed for GaAs. Among the studies of GaAs growth were those reported by Shaw [149,150] in a series of articles including a systematic measure of the growth rate of GaAs on {0 0 1}, {1 1 1}A, {1 1 1}B, and {1 1 0} surfaces (using hydrogen as carrier gas and the chloride Vapor phase AB
AB+B Vapor phase followed by diffusion
Adsorption followed by diffusion
B* A
AB*
Dissociation A A*
Su bstra
A
te * Adsorbed species
Surface diffusion Figure 3.40 Kinetics processes occurring from the growth by vapor phase. Patterned after Ref. [104].
j389
j 3 Growth and Growth Methods for Nitride Semiconductors
390
method) as a function of temperature and the GaCl partial pressure. The nitride growth rate in HVPE exhibits an increase followed by a decrease in the growth rate with decrease in substrate temperature. The latter was found consistent with Langmuir GaCl adsorption isotherm together with a kinetic model based on GaCl adsorption [151,152]. In this treatment, the lateral interaction between GaCl adsorbed molecules and the Cl desorption by hydrogen of adsorbed GaCl molecules (dubbed the H2 mechanism) are invoked to account for the slight decrease in the GaAs growth rate at low temperatures when the GaCl partial pressure is increased. Hollan and Schiller [153,154] and Hollan et al. computed surface diffusion, a component of kinetics at play, by fitting the experimental data and computed values as a function of the substrate orientation, and of Gentner [155] on 6 off {0 0 1} GaAs substrates. In the latter study, both atmospheric and reduced pressures in hydrogen and helium carrier gases were considered. These studies were also expanded to include AsCl3 instead of HCl gas. A desorption mechanism for two adsorbed chlorine atoms by GaCl in GaCl3 was, therefore, considered with an intermediate GaCl3 adsorption step [139]. This mechanism is called the GaCl3 mechanism. Cadoret [156] developed a model involving adsorption and desorption of species on the surface of the substrates. Among the adsorbing species considered are NH3 molecules, adsorption of N atoms resulting from NH3 decomposition, and adsorption of GaCl molecules on N atoms. They follow the reactions: V þ NH3 ðgÞ()NH3 ; 3 NH3 ()N þ H2 ðgÞ; 2
ð3:11Þ
N þ GaClðgÞ()NGaCl; where (g) depicts the gas-phase species and V is a vacant site. Cadoret also considered two desorption mechanisms of chlorine: that is, desorption in HCl vapor molecules following a surface reaction with H2 and desorption in GaCl3 vapor molecules following adsorption of a GaCl molecule on two GaCl underlying molecules, which is schematically shown in Figure 3.41. The representation is based on the premise that the substrate surface, which is sapphire, is terminated with Al atoms on which N atoms bond leading to a Ga polarity sample. As in the GaAs model [139], the processes follow the reactions: 2NGaCl þ H2 ðgÞ()2NGa þ 2ClH; NGa ClH()NGa þ HClðgÞ:
ð3:12Þ
The two mechanisms are labeled H2 and GaCl3 mechanisms. They can be treated by means of a one-monolayer model of adsorption on a (0 0 0 1) Ga or Al surface. The adsorbed species are NH3 molecules, N atoms, NGaCl, NGa–ClH, and 2N Ga–GaCl3 molecules. The one-monolayer adsorption model and the Bragg–Williams approximation are used to simplify the problem. The number of activated molecules involved in the reactions described by Equations 3.11 and 3.12 as well as possible intermediate states of hydrogen desorption from NH3 are neglected. GaCl adsorption on a Ga adatom [157,158] is assumed to be negligible because it would lead to antisite positions following chlorine desorption or would act as a simple inhibitor of the deposition
3.4 Nitride Growth Techniques
Ga N
H
H Cl
Cl
Ga
Ga
N
N
Cl Ga
NH3 N
N
N
V
(a)
Cl Ga
Ga N
N
Cl
Cl
Ga
Ga
N
Cl Ga N
NH 3 N
N
V
(b) Figure 3.41 Schematic steps of adsorption and desorption processes involved in the (a) H2 mechanism and (b) GaCl3 mechanism. Patterned after Ref. [104].
process that is not necessarily observed in GaAs. The two overall reactions corresponding to the H2 and GaCl3 growth mechanisms can be written as V þ NH3 ðgÞ þ GaClðgÞ()NGa þ HClðgÞ þ H2 ðgÞ; 2V þ 2NH3 ðgÞ þ 3GaClðgÞ()2NGa þ GaCl3 ðgÞ þ 3H2 ðgÞ:
ð3:13Þ
The adsorption and desorption processes including their flux, potential barriers associated with their activation barriers, are somewhat involved and are beyond the scope of the present treatment. An in-depth treatment of the topic can be found in Ref. [104]. In addition to adsorption and desorption, quantification dealing with growth by HVPE also involves mass transport, which is then followed by growth. In the mass transport case, partial pressures of reactant species are treated that, depending on local conditions and the substrate area, can lead to production or depletion of species. Coupled with the treatment of mass transport, the growth phase, which involves equilibrium between the vapor and substrate surface, must be treated. To get a handle on the problem, superficial flow of diffusive molecules toward step edges, which are assumed to be monomolecular in height, and surface coverage of vacant sites, their distribution over the substrate surface (assumed uniform), are considered. So is whether the growth is based on GaCl3 or H2 growth mechanisms. When the terrace width is small enough to not hinder surface diffusion, the parameters of importance are the adsorption energies of GaCl(g), HCl(g), and GaCl3(g). Further details can be found in Ref. [104]. Those who are more focused on the growth- and properties-related issues involving GaN by MBE are referred to Ref. [159].
j391
j 3 Growth and Growth Methods for Nitride Semiconductors
392
Summarizing, the growth process of GaN by HVPE could be analyzed by a combination of thermodynamical and kinetic considerations. The ample experimental data available in the literature regarding the growth rate on exact and misoriented {0 0 1} surfaces including those measured by Seifert [111] in He and H2 environments on 3 off (0 0.1), in the hkl configuration and (0 0 0 1) in the hjkl configuration, GaN by HVPE allowed for modeling and understanding the physical processes involved in growth. Relative kinetics considerations would indicate that the mass transfer is larger for GaN than for GaAs. Additionally, the high supersaturation, in effect, generally leads to important parasitic GaN deposition before the substrate zone, or the deposition zone, that reduces and even could potentially negate the relative supersaturation. The accuracy with which the growth rate is measured and controlled is the key to the study of the thermodynamics of the system. The variants of HVPE have also been explored and used for growth of GaN and its ternaries with AlN. Among them is a modified VPE process, dubbed the sublimation sandwich method (SSM), which was reported by Wetzel et al. [160] and Fischer et al. [161] (Figure 3.42). Initially, the GaN films were grown from metallic Ga and ammonia on (0 0 0 1) 6H-SiC, using a modification of the sandwich method described previously by Vodakov et al. [162]. In this approach, the quartz reactor contains a Ga cell for each substrate for multiwafer processing with one ammonia stream only. The gap between the substrate and the Ga source is typically about 5 mm. The ammonia flow rate through the gap is very high, 25–50 l m1 at atmospheric pressure. Under these conditions, there is an effective mass transport of Ga vapor and nitrogen to the surface of the substrates. At growth temperatures between 1170 and 1270 C, GaN layers were obtained at growth rates of up to 0.3 mm h1. Many other variations of this approach, including [163–171] gaseous sources such as GaCl, GaCl2, GaCl3, Ga(C2H5)2Cl, GaCl2NH3, AlC3, AlBr3, InCl3, and GaBr3 for group III element(s) as reactants for NH3, have been used. Pastrnak et al. [172] chose to react N2 with GaCl3, AlCl3, and InCl3 in their CVD process. Dryburgh [173] grew RF Coils Water
Graphite Substrate NH3
ΔT
Ga Graphite Water
Figure 3.42 Schematic diagram of a proximity HVPE vessel for the growth of GaN at very high growth rates, approaching 0.3 mm h1. It is dubbed the SSM.
Quartz
3.4 Nitride Growth Techniques
AlN from AlSe and N2. By introducing PH3, Igarashi et al. [174] achieved several percent P incorporation in GaN. The technique became popular because of the highquality buffer layers on the freestanding GaN templates for epitaxy by heterostructure deposition systems such as OMVPE and MBE [175]. The transport and optical properties of HVPE-grown films are discussed in Volume 2, Chapters 3 and 5, respectively. 3.4.1.2 Organometalic Vapor Phase Epitaxy High-quality epitaxial III-N films and heterostructures for devices have been accomplished by OMVPE technique. Manasevit et al. [176] applied this technique to the deposition of GaN and AlN in 1971. Using triethylgallium (TEG) and ammonia (NH3) as source gases for group III and V species, respectively, the authors obtained c-axis oriented films on sapphire (0 0 0 1) and on 6H-SiC(0 0 0 1) substrates. MIS-like LEDs followed, albeit they relied on deep states induced by Zn and suffered from very low efficiencies because of their poor crystalline quality. The development of LT buffer layers addressed the quality issue some [177]. The technique improved over the years to the point that undoped GaN films with a low background carrier concentration of 5 · 1016 cm3 and with an X-ray symmetric peak FWHM of 30 arcsec have been grown [178]. The X-ray data should be treated with caution, as the symmetric peak is not as sensitive to the edge dislocation as the asymmetric peak. OMVPE has been used for the development of LEDs [179], lasers [180], transistors [181], and detectors [182]. The best OMVPE reactors for group III nitride film growth incorporate laminar flow at high operating pressures and separate inlets for the nitride precursors and ammonia to minimize predeposition reactions. A successful, two-flow OMVPE reactor is shown in Figure 3.43 [183]. The main flow composed of reactant gases with a high velocity is directed through the nozzle parallel to a rotating substrate. The subflow gas composed of nitrogen and hydrogen is directed perpendicular to the substrate. The purpose of the flow normal to the substrate surface is to bring the reactant gases in contact with the substrate and to suppress thermal convection effects. Hydrogen is the carrier gas of choice. A rotating susceptor was used to enhance uniformity of the deposited films. If one goes with the premise that smallest rocking curve half width implies an all-around good quality, GaN films can claim this quality. These films with one of the narrowest rocking curves with FWHM values of 37 arcsec (values even under 30 arsec have been obtained) were grown with a modified EMCORE GS 3200 UTM reactor. It should be stated, however, that the X-ray data based on the symmetric diffraction peak are not a critical measure of sample quality necessarily. For a more complete analysis, one should also inspect the asymmetric peak, which is sensitive to edge dislocations. This reactor generally incorporates separate inlets for ammonia and the nitride precursor, all are normal to the substrate surface, which rotates at speeds over 1000 rpm, and a laminar flow cell to assure a uniform growth [212]. OMVPE reactors incorporating new concepts have been designed to grow layers at lower temperatures. (It should be mentioned that the motivation for lower temperature growth spawned from the perceived need to minimize the loss of nitrogen from
j393
j 3 Growth and Growth Methods for Nitride Semiconductors
394
Figure 3.43 A schematic representation of a vertical OMVPE system employed at Virginia Commonwealth University along with a picture of the deposition chamber (a); a photograph of the reactor chamber of the same (b). (Please find a color version of this figure on the color tables.)
the surface. However, later it became evident that high temperatures are needed to mobilize treading dislocations, as they are useful for reducing dislocation density and facilitating lateral growth. For the sake of completeness, a discussion of relatively lower temperature growth is provided.) These technologies utilize an activated form of nitrogen to lower deposition temperatures of group III nitrides. That these technologies are interesting is apparent, for example, from the deposition of polycrystalline and amorphous GaN films at temperatures lower than 350 C by plasma-enhanced CVD. Epitaxial GaN and AlN have been grown by variants of methods activating nitrogen, such as laser-assisted CVD, remote plasma enhanced CVD, atomic layer epitaxy with
3.4 Nitride Growth Techniques
NH3 cracked by a hot filament, with ammonia catalytically decomposed, photoassisted CVD, and ECR plasma-assisted CVD. However, none of these approaches has been able to produce material comparable in quality with the standard OMVPE systems and, consequently, they did not really become players in the field. We alluded to the genesis of the lower quality above. As for the mechanism involved, growth of nitride semiconductors by OMVPE relies on the transport of organometallic precursor gases, hydrides for the nitrogen source, and reacting them on or near the surface of a heated substrate. The deposition is through pyrolysis. The underlying chemical mechanisms are complex and involve a set of gas phase and surface reactions. Although OMVPE has long been assumed to be a thermodynamically equilibrium process, nitride OMVPE processes may involve kinetics as well. The fundamental understanding of the processes involved is still evolving and, as such, the reaction mechanism and the related kinetic rate parameters are poorly understood. The deposition of epitaxial nitride layers by OMVPE involves the reaction of metalcontaining In, Ga, or Al gases with ammonia, NH3. Commonly, the metal-containing gases are trimethylgallium ((CH3)3Ga), trimethylindium ((CH3)3In), or trimethylaluminum ((CH3)3Al). Radicals, reactive by most definitions, react in the gas phase with donors containing acidic hydrogen, such as NH3, and form adducts. The key here is to eliminate the unwanted radicals by forming stable molecules followed by their removal from the reaction region. The analysis of the mechanisms involved in the OMVPE process clearly indicates that any precursor must balance the requirements of volatility and stability, which often counter each other, to be transported to the surface and decomposed for deposition. To put it another way, these precursors must have appropriate reactivity to decompose thermally into the desired solid and to generate readily removable gaseous side products. Ideally, the precursors should be nonpyrophoric, water and oxygen insensitive, noncorrosive, and nontoxic. The trialkyls, trimethylgallium (TMG) [184] and triethylgallium (TEG), trimethylaluminum (TMA) [185], trimethylindium (TMI) [186], and others are usually used as III metal precursors. Ammonia (NH3) [187], hydrazine (N2H4) [142,188,189], monomethylhydrazine (CH3)N2H2 [190,191], and dimethylhydrazine (CH3)2N2H2 have all been used as nitrogen precursors with varying degrees of success. Although trialkyl compounds (TMA, TMG, TMI, etc.) are pyrophoric and extremely water and oxygen sensitive, and ammonia is highly corrosive, much of the best material grown today is produced by conventional OMVPE by reacting these compounds with NH3 at substrate temperatures close to 1000 C [192–203]. Investigators have reacted TMG [204,205], TEG, and GaCl [206,207] with NH3 plasma. Sheng et al. [208] reacted TMA and NH3 in the presence of hydrogen plasma. Wakahara et al. [209] grew InN by reacting TMI with microwave-activated N2. Eremin et al. [210] used nitrogen to transport metallic Ga to the reaction zone where it was reacted with active nitrogen. These commonly employed precursors at least satisfy the criteria of sufficient volatility and appropriate reactivity. During the growth of nitrides by employing trialkyl precursors, adduct formation between ammonia and TMA and TMG is well documented. Usually mixing at room temperature, adduct
j395
j 3 Growth and Growth Methods for Nitride Semiconductors
396
formation between TMG or TEG and ammonia is complete in less than 0.2 s. The resulting adduct Ga(CH3)3-NH3 has a vapor pressure of 0.92 Torr at room temperature, while the vapor pressure of Ga(C2H5)3 : NH3 is much lower. To counter early beliefs that stability of ammonia and required relatively high growth temperature the use of other more volatile nitrogen sources were explored. For example, Fujieda et al. [142] replaced NH3 with N2H4 and observed that a significantly smaller amount of N2H4 was required to maintain the same growth rate. However, they also noted that the CVD growth rate was limited by the decomposition of TMG, thus limiting the benefits of N2H4. Matloubian and Gershenzon [211] used TMA and NH3 and a substrate temperature range of 673–1473 K to grow nitrides. Single-crystal AlN films were obtained only at 1473 K. Dupuie and Gulari [212] reported that the presence of a hot filament near the substrate increased the growth rate of AlN grown with TMA and NH3 by two orders of magnitude. However, the use of a hot filament immediately raises concerns about residual contamination, most prominently oxygen, which was not addressed by the author. A case in point illustrating the elimination of unwanted radicals by forming stable molecules followed by their removal from the reaction region is AlN growth from mixtures of methyl alkyls that may proceed by the formation of an intermediate gas phase adduct (CH3AlNH3), followed by the elimination of CH4. The exact path may be that coadsorption of (CH3)3Al and NH3 at room temperature generates surface adduct species such as ((CH3)2AlNH3) and adsorbed NH3 [213]. As the substrate temperature is raised above 320 C, the appearance of vibrational bands corresponding to AlN indicates the formation of extended (Al–N) networks on the surface. These Al(NH2)2Al species finally eliminate H2 at the surface to form AlN [214]. The possible chemical reactions in the process are (a stands for adsorbate on the surface and g stands for gas-phase product):
and
2ðCH3 Þ2 Al:NH3 ðaÞ ! CH3 AlðNH2 Þ2 AlCH3 ðaÞ þ 2CH4 ðgÞ; CH3 AlðNH2 Þ2 AlCH3 ðaÞ ! AlðNHÞ2 Al þ 2CH4 ðgÞ;
ð3:14Þ
AlðNHÞ2 AlðaÞ ! 2AlNðaÞ þ H2 ðgÞ: As for GaN, investigations are relatively limited, but it would be fair to assume that processes similar to that with AlN growth are most likely in place. Adducts of Ga compounds are weaker electron acceptors than the corresponding Al adducts, and therefore these adducts may not be abundant owing to redissociation in the hot zone. It may be because of this that successful GaN growth by OMVPE requires very large V/III ratios, which favor adduct formation. Thermal stability of NH3, although low compared to that of N2, could be partially responsible for the use of high substrate temperatures, typically above 550 C for InN and above 1000 C for GaN and AlN. The high growth temperature necessitated by the process itself, associated with high nitrogen vapor pressure over GaN, lead to the inevitable nitrogen loss from the nitride film. This may also be the path to carbon contamination from the decomposition of the organic radical during metalorganic pyrolysis. The loss of nitrogen can be
3.4 Nitride Growth Techniques
alleviated by using high V/III gas ratios during the deposition, particularly for InGaN (e.g., >2000 : 1). Assuming that high substrate temperatures represent a problem in relation to ammonia, which seems reasonable particularly in early days, various alternative approaches can be and have been taken. One approach is to use alternative nitrogen precursors that are thermally less stable than NH3. Hydrazine (N2H4), which is a larger and less stable molecule, has been used in combination with TMA to deposit AlN at temperatures as low as 220 C [189]. However, hydrazine is toxic, unstable, and not as pure as NH3. Consequently, a compromise between quality and substrate temperature must be made. Researchers took the quality/purity as the primary parameter and stayed with NH3. More recently, other nitrogen sources such as tertbutylamine (t-BuNH2) [215], isopropylamine (i-PrNH2), and trimethylsilylazide (TMeSiN3) have been used with TMA or t-Bu3Al to deposit AlN films at lower substrate temperatures (400–600 C) and reduced V/III gas ratios (5 : 1–70 : 1) [216]. However, the deposited films were invariably contaminated with high levels of residual carbon (up to 11 at.%). Hydrogen and, to a lesser extent, nitrogen are predominantly used as the transport gas. They can influence the chemical reaction mechanism of Et3M or (CH3)3M in the gas phase by changing the reaction temperature of the metalorganic compounds or the concentration of reaction products. Hydrogen at the surface of the growing film can influence the growth rate and the structural properties [217,218]. To obtain a basic understanding of the role of hydrogen in GaN growth, the possible sources of hydrogen and the influence of hydrogen on the chemical reaction mechanism in the gas phase and at the surface are unique and important issues in the context of OMVPE. Pyrolysis of highly concentrated NH3 in the presence of H2 as the carrier gas results in a high concentration of molecular and atomic hydrogen near the substrate surface. Because the growth temperatures above 900 C are employed, which are higher than the decomposition temperature of the metalorganic compounds and their hydrocarbon ligands, the conditions for the desired bond breaking between the metal atom and the methyl or ethyl groups of the precursors are in place. However, the same can also lead to pyrolysis of the hydrocarbons with incorporation of hydrogen and carbon into the films. In this vein, decomposition of (CH3)3Ga (TMG) and Et3Ga in hydrogen and nitrogen atmospheres using a quadrupole mass analyzer has been investigated [219]. The decomposition reaction of the metalorganic precursors was found to be strongly affected by the presence of molecular hydrogen. The decomposition of (CH3)3Ga occurs at 400 and 500 C in H2 and N2, respectively. Similarly, the decomposition of Et3M occurs at 260 and 300 C in H2 and N2, respectively. Clearly, molecular hydrogen reduces the reaction temperature. The reaction mechanisms involve hydrolysis for (CH3)3Ga in H2, homolytic fission for (CH3)3Ga in N2, and b-elimination for Et3M in both H2 and N2. By changing the reaction temperature and the reaction mechanism of the metalorganic precursor, the partial pressure of hydrogen affects the deposition rate of GaN and therefore the structural properties of the resulting film, especially at low growth temperatures [220].
j397
j 3 Growth and Growth Methods for Nitride Semiconductors
398
3.4.1.3 Modeling of OMVPE Growth The GaN kinetic model with the associated details, to the extent necessary, of the GaN thermal decomposition, precursor adsorption, and decomposition, Ga and N2 desorption rates, and surface diffusion rates is discussed next to identify the important chemical steps for GaN growth. This will aid in developing the kinetic model. 3.4.1.3.1 Thermal Decomposition of GaN as it Relates to Growth The rule of the thumb for modeling the growth of conventional compound semiconductors is that MBE is a nonequilibrium growth with several activation barriers in place and, as such, a kinetic model is better suited. On the contrary, OMVPE is an equilibrium process and as such, a thermodynamic model is appropriate. However, nitrides are different in that temperatures employed are high and barriers such as NH3 dissociation are in play. Consequently, a kinetics-based model for OMVPE is appropriate as reported by Koleske et al. [221]. The group III nitrides have strong chemical bonds with the Ga–N bond strength being estimated near 4.2 eV, which is comparable to C–C bond strength of 3.8 eV in diamond. In addition, the group III nitride bonds are more ionic (31–40%) (excluding BN) than those in other III–V semiconductors (<8%). Despite a strong bonding, GaN decomposes above 800 C. As in the case of the other III–V semiconductors, the heat of formation of GaN is small and similar to that of InP and AlAs, indicating that GaN decomposes easily. The GaN decomposition temperature is important for OMVPE or OMVPE GaN growth because this temperature is about 200 C less than the growth temperature. As is clear from the overall discussion of growth issues, the growth of GaN on foreign substrates is complicated in that it involves at least three steps, namely, nitridation, nucleation layer growth, and the growth of the final structure. Before we delve into the kinetic model, a few words on the GaN decomposition is warranted. GaN does not melt congruently at pressures typically used for OMVPE or MBE growth, but decomposes at above 800 C at atmospheric pressure and at lower temperature in vacuum. It was initially postulated that GaN decomposed into N2 and Ga, which was eventually observed using mass spectroscopy. In addition, the desorption of small GaN clusters has been postulated and later observed as GaNþ and Ga2N2þ ions, again with mass spectroscopy. Both thermogravimetric and mass spectroscopic techniques have been used to measure the kinetics of GaN thermal decomposition, with both measurements resulting in an activation energy (EA) of 3.1 eV, which compares with 3.6 eV determined in an MBE environment [222]. Actually, there is a wide spread in the reported activation barrier to desorption, which spans from 0.34 to 3.62 eV and depends strongly on the annealing conditions [223]. The reported activation barriers appear to group into four distinct figures, which may imply a somewhat complex desorption process and, perhaps, a lack of attention to the temperature, pressure, and environment, such as H2 or N2. GaN decomposition temperatures reported are in the range of 400–900 C, which may well be explained by pressure, temperature, and gas flow (i.e., H2 or N2), which plausibly change the GaN decomposition rates and mechanisms.
3.4 Nitride Growth Techniques
Early on, several different mechanisms were proposed to explain GaN decomposition. These included decomposition into gaseous Ga and nitrogen [224] as in Equation 3.15, liquid Ga and nitrogen as in Equation 3.16, and sublimation of GaN into a diatomic or polymeric product [225] as in Equation 3.17. 2GaNðsÞ ! 2GaðgÞ þ N2 ðgÞ
ðdecompositionÞ;
ð3:15Þ
2GaNðsÞ ! 2GaðlÞ þ N2 ðgÞ ! 2GaðgÞ þ N2 ðgÞ ðdesorptionÞ;
ð3:16Þ
2GaNðsÞ ! GaNðgÞ or ½GaNx ðgÞ ðsublimationÞ:
ð3:17Þ
As compiled by Koleske et al. [223], N2 evolution has been observed using mass spectroscopy along with GaNþ and Ga2N2þ species. Occasionally, Ga droplets 20–30 mm in size have been observed on the GaN surface, indicating that the GaN decomposition rate, kGaN, can exceed the Ga desorption rate, kGa. As mentioned above, the onset of GaN decomposition in terms of the temperature is lower in H2 as compared to inert environments, such as N2, Ar, and vacuum. Hydrogen could assist decomposition by the reverse GaN synthesis reaction, that is, the reformation of NH3 via 2GaNðsÞ þ 3H2 ! 2GaðlÞ þ 2NH3 ðgÞ:
ð3:18Þ
Measured values of the activation energy, EA, and the preexponential factor, A0, for GaN decomposition in H2 under a pressure of less than 76 Torr, in N2, and in vacuum are listed in Table 3.12. The span of EA is 2.7–3.93 eV. Surprisingly, this is lower than the EA of 4.7 eV for GaAs decomposition, despite the stronger bond strength in GaN (4.1 eV) compared to GaAs (2.0 eV). When GaN layers are annealed in H2 at higher pressures and lower temperatures, the EA decreases (Table 3.12) down to about EA of 1.8 eV. In H2 at higher pressures and higher temperatures, even lower values of EA, down to 0.38 eV have been measured. These lower EA values for decomposition in H2 suggest hydrogen aids in N removal from the surface, possibly by NH3 formation, as in Equation 3.18. Because GaN desorption is an integral part of growth and adsorption and decomposition/desorption occur simultaneously, it is important that they are treated side by side. To make sure that the GaN decomposition rates measured under vacuum conditions are applicable to OMVPE pressures, as mentioned above, Koleske et al.[221,223] measured GaN decomposition under flowing H2 at a pressure of 50 Torr. Upon heating a 2 mm thick GaN film at 1030 C for 10 min, about 40% of the GaN film was gone and the remaining GaN film was thinner and covered with Ga droplets. From this measurement, a GaN decomposition rate of about 1 · 1017–3 · 1017 cm2 s1 was estimated, which compares well with a calculated rate of 1 · 1017 cm2 s1 for the thermogravimetric determination at 1030 C, suggesting that the GaN decomposition rate is similar in vacuum and OMVPE pressures. It should also be mentioned that the activation energy, EA, for GaN decomposition is only slightly larger than the EA of 2.8 eV for Ga desorption from liquid Ga.
j399
j 3 Growth and Growth Methods for Nitride Semiconductors
400
Table 3.12 Arrhenius parameters for kinetic reactions along with notes compiled in Ref. [221].
Process
Preexponent
EA (eV)
Comments
Decomposition GaAs GaN GaN
— 4 · 1029 cm2 s1 5 · 1028 cm2 s1
4.7 3.1 3.1
RHEED oscillation measurement Thermogravimetry measurement Mass spectrometry measurement
Ga desorption from Liquid Ga — GaAs — GaN — GaN — GaN 1.0 · 1028 cm2 s1
2.8 2.5 2.76 2.2 2.69
Review paper on liquid elements Modulated molecular beam technique RHEED study of hexagonal GaN Growth rate versus growth temperature Ga dosed, RHEED pattern recovery
N desorption from GaN 2.0 · 1044 cm2 s1
6.1
N dosed, RHEED pattern recovery
Surface diffusion on GaN Ga on GaAs 3.3 · 105 cm2 s1 Ga on GaN 0.007 cm2 s1 N on GaN — GaN —
1.3 2.48 >3 1.45
V/III desorption ratio 1.93 · 1016 kN/kGa
3.41
RHEED oscillations on vicinal surfaces Ga dosed, RHEED pattern recovery Estimate based on present work RHEED pattern recovery Calculation by Koleske et al. [223]
Note that a GaN decomposition activation energy of 3.6 eV has also been reported in Ref. [222].
3.4.1.3.2 Ga and N Precursor Adsorption and Decomposition Adsorption and decomposition processes are described in detail in conjunction with MBE growth of GaN in Section 3.4.2. However, the particulars of these processes, although the basics are the same, in an OMVPE environment are discussed here. Several factors, such as the precursor flux, the gas-phase diffusion rate, the sticking coefficient, the number of available surface sites, and the precursor decomposition rate, contribute to the concentration of active growth species on the surface. For the particular kinetics-based model, it will be assumed that, once adsorbed, the Ga and N precursors decompose completely to produce adsorbed Ga and N atoms and that precursor sticking coefficients are near unity. The assumption that TMG and TEG undergo full dissociation to Ga atoms is supported by several studies in that nearly complete pyrolysis of TMG for T > 500 C or TEG for T > 300 C in H2 or N2 occurs. In flowing NH3, measurable and complete TMG decompositions are observed for T > 525 C and T > 625 C, respectively. TMG and TEG have been shown to partially decompose in the gas phase but at a reduced pressure (e.g., 0.1 atm), which is parasitic. However, most of the decomposition occurs on the surface rather than in the gas phase. To ensure that the decomposition takes place on the surface, gas manifolds are designed to keep the precursors cool until they come in contact with the surface. The sticking coefficients for TMG and TEG are large because of the accommodation ratio, which increases as the molecular weight increases. TMG forms a strong adduct bond with NH3, which would further
3.4 Nitride Growth Techniques
increase the accommodation ratio and therefore the sticking probability. TMG conversion to metallic Ga has been observed on a sapphire surface at 525 C at atmospheric pressure, suggesting that metallorganic Ga precursors efficiently stick and completely dissociate to metallic Ga on GaN and sapphire surfaces. The ability to grow low temperature nucleation layers at substrate temperatures not much higher than 500 C is also indicative of this conversion. Although not much is known about NH3 adsorption and dissociation, the general assumption is that sticking of NH3 to a Ga-terminated surface should be large. Moreover, the sticking of NH3 may depend on the N surface coverage in a fashion similar to the reduced sticking of other group V hydrides on group V terminated surfaces. Once adsorbed on the surface, NH3 dissociates and forms adsorbed H and NHx<3 species. The latter either further dissociates to adsorbed N and H atoms or combines with adsorbed H to reform NH3. It has been shown that the active species that nitridates the surface is the N–H fragment, and that the large kinetic barrier associated with the decomposition of NH3 on the surface may be related to loosing the second or third H from NH3. Because H desorbs from the surface at temperatures as low as 300 C after NH3 exposure, the steady state to which H coverages at temperatures used for growth in OMVPE is small. It should also be noted that NH3 dissociation at RMBE and some OMVPE growth temperatures should be marginal, unless some sort of catalysis is in effect, in which case the rate of dissociation should depend strongly on the catalyst, such as metallic Ga on sapphire. Several investigations, in fact, showed a dramatic increase in the NH3 decomposition rate as the temperature of the catalyst was increased [226], from few percent at 900 C up to 40–50% at 1050 C. In the context of the present model, NH3 dissociation to N atoms will be assumed complete, with ramifications of this assumption given. 3.4.1.3.3 Ga and N2 Desorption from the Surface Growth is a complex process in that adsorption, dissociation, and desorption take place simultaneously at many levels. Because OMVPE does not lend itself well for the studies involving these processes, a good deal of reliance on MBE studies is called for. Studies to gain insight into these processes, particularly the desorption rates of Ga and N, are conducted by means of monitoring such processes as functions of substrate temperature. The N and Ga desorption rates, kN and kGa, respectively, can be plotted as functions of inverse temperature, from the slope of which activation energies can be obtained. Such an exercise in one study led to the Ga desorption activation barrier, EA, of 2.69 eV (among various reports, it ranged between 2.2 and 2.76 eV) as compared to the Ga desorption EA of 2.8 eV from liquid Ga and GaAs and GaN decomposition activation energy of 3.1–3.6 eV. Because the measured EA for Ga desorption from GaN, GaAs, and liquid Ga are similar and the liquid phase of Ga covers a range extending from 30 to 2403 C, it is reasonable to assume that the Ga desorption mechanism from GaN does not change for the temperatures employed in OMVPE growth. This justifies the use of kinetic parameters regarding the OMVPE growth, particularly considering the growth conditions typically employed. Using the line-of-sight quadrupole mass spectrometry as a quantitative in situ method, Koblm€ uller et al. [227] undertook an extensive investigation of Ga adsorption
j401
j 3 Growth and Growth Methods for Nitride Semiconductors
402
and desorption on both Ga-polar (0 0 0 1) and N-polar ð0 0 0 1Þ surfaces. Monitoring the desorbing Ga atoms, two characteristic desorption regimes (exponential and steady-state) were observed that were attributed to the formation of a thin equilibrium Ga adlayer and Ga droplets on top of it. The Ga adlayer coverage differs substantially between the two surface polarities, being 1.1 monolayers on the N-polar ð0 0 0 1Þ GaN and 2.4 monolayers on the Ga-polar (0 0 0 1) GaN. Additional temperature-dependent measurements of the surface lifetime of Ga adatoms revealed fundamental differences in the adsorbate–substrate binding energetics both for the Ga adlayers on Gaand N-polar surfaces and for the Ga droplets. In the above-mentioned experiments, Ga atoms were allowed to impinge on the substrate in the absence of nitrogen being impingent on the surface. The amount of Ga adlayer was controlled by varying the length of time that the Ga effusion cell shutter remained open. After deposition, the shutter was closed and a quantitative in situ method, the line-of-sight quadrupole mass spectrometry, was used to monitor the impinging of Ga atoms. Desorption energies were determined by the dependence of the decay time of the desorption process on substrate temperature. A desorption energy of 3.1 eV was reported for Ga evaporating from Ga droplets. The desorption energies (or activation energy for desorption) for Ga from the top Ga adlayer and bottom Ga adlayer were 3.7 and 4.9 eV, respectively. In comparison, the single Ga adlayer on the ð0 0 0 1Þ GaN surface exhibited a desorption energy of 3.7 eV, which is comparable to that from the top Ga adlayer on the Ga-polar surface. Moreover, on both surface polarities, additional Ga surplus was found to lead to the accumulation of metallic Ga droplets on the top of the adlayer. Further, Ga bound to droplets was found to remain in the weakest bound state of all possible adatom surface arrangements, resulting in the lowest activation energy for Ga desorption of EA ¼ 3.1 0.2 eV. This value is consistent with decomposition activation energy of 3.1–3.6 eV mentioned above and agrees well with the evaporation energy from liquid Ga. In addition, this activation for desorption was not found to depend on the underlying GaN surface polarity or the surface roughness. Turning our attention to N, it is typical that the group V element desorbs as a dimer (V2), implying that the desorption rate is limited by the surface diffusion of two group V atoms. There are some reports that N desorption is of the first order. For N to follow first-order desorption kinetics would mean that the surface diffusion may play less of a role in the desorption mechanism than in the surface structure. For example, the first-order desorption kinetics have been measured for H2 desorption from Si(1 0 0) and found to be associated with ordering of H on the Si(1 0 0) dimers. Another plausible mechanism is that once an N atom diffuses near another N atom, the N2 molecule would immediately form and desorb from the surface because of the highly exothermic nature of the N2 formation. This would make the N2 desorption first order in terms of N coverage. However, at higher temperatures (>720 C), the kinetic order may change and this cannot be ruled out. Of paramount pertinence to OMVPE growth is that the desorption rates of N and Ga become equal at 780 C (one monolayer per second or 1.14 · 1015 atoms cm2 s1), and thus, at the growth temperatures commonly used in OMVPE, the N desorption rate is 1000–10 000 times that of Ga. At about 1030 C, a reasonable
3.4 Nitride Growth Techniques
OMVPE growth temperature, the estimated GaN decomposition rate is about 3 · 1017 cm2 s1, which coincides well with kGa. Assuming that desorption occurs only from the top layer, the desorption limited surface residence times, tD, can be calculated by dividing the surface coverage (1.14 · 1015 cm2) by the desorption rates, kGa and kN, for Ga and N, respectively. Most studies indicate tD,Ga ¼ 1 s at temperatures in the range of 750–780 C, which is larger than for Ga on GaAs(1 1 1). 3.4.1.3.4 Ga and N Surface Diffusion At high growth temperatures and during growth where the surface diffusion length is larger than the terrace width, atoms move by surface diffusion from the terrace to the step edge for incorporation into the growing lattice, which is called the step-flow growth. As the diffusion length decreases, the surface morphology changes from step flow to islanded growth on the terraces. These two regimes have been established in MBE growth of GaAs and other conventional compound semiconductors. If thermal desorption is large, the diffusion length may be dramatically shortened. Consequently, diffusion lengths at growth temperatures must be evaluated. This can simply be done by assuming that the surface diffusion follows an Arrhenius form and that the mechanism does not change with temperature. In these descriptions and the following Einstein equation, the diffusion length, lS, is given by [228] pffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð3:19Þ lS ¼ DS tS ¼ DS0 expð E SD =kTÞtS ;
where DS is the diffusion coefficient,tS is the lifetime of the diffusion event, DS0 is the temperature-independent diffusion coefficient, ESD is the diffusion activation energy, and kB is the Boltzmanns constant. Table 3.12 lists representative values of DS0 and ESD for GaN and GaAs. The values of tS can be in two limiting values, which are the lifetime before desorption (tD), which depends on the desorption rate and the lifetime before lattice incorporation (tI), which depends on the GR. The estimate lS, Ga for both limiting values of tS will be presented here. Using the DS,Ga reported in the literature [229] and tD,Ga calculated in Section 3.4.1.3.3, Koleske et al. [221] estimated lS,Ga using Equation 3.19. The temperature dependence of lS,Ga is plotted as a solid line in Figure 3.44, which indicates that lS,Ga decreases slightly as the temperature increases. This is unexpected because the temperature increases as lS increases. The decrease in lS,Ga at higher temperatures takes place because the increase in DS,Ga is counterbalanced by the Ga desorption rate, which results in a reduced tD. As shown in Figure 3.44, lS,Ga is less than 1.2 nm for temperatures ranging from 600 to 1200 C. If the Ga coverage exceeds one monolayer, lS,Ga would increase and approach the lS,Ga value for Ga on liquid Ga metal. However, Ga droplet buildup on the surface during growth is not advantageous for maintaining the desired smooth surface morphology (in contrast to MBE). Therefore, lS,Ga is about 1 nm if the Ga desorption rate from the surface is large. It must be remembered that at OMVPE growth pressures (30–770 Torr), Ga desorption appears to be suppressed, suggesting that the Ga surface lifetime, tS,Ga, may be substantially longer than the desorption limited lifetime. The limit, of course, is the incorporation lifetime (tI) into the growing lattice. The value of tI depends inversely on the growth rate, as the slower the growth rate the larger tI. To
j403
j 3 Growth and Growth Methods for Nitride Semiconductors
404
102 3
Diffusion length, (nm)
10 30
λ S,Ga = (D S,Gaτl)1/2
101
100
nm min–1
100 λ S,Ga = (D,S,GaτD)1/2
10–1 600
700
800
900
1000
1100
1200
Temperature (ºC) Figure 3.44 Estimates of the Ga diffusion length, lS,Ga, on GaN. The solid line is a calculation of lS,Ga based on tD (desorption limit) and the dashed lines are calculations of lS,Ga based on tI (growth rate limit) for growth rates of 3, 10, 30, and 100 nm min1 [223].
show the dependence of tI on the growth rate, lS,Ga is plotted in Figure 3.44 as open segmented lines for four different GR ranging from 30 to 1000 Å min1. For this calculation, tI was determined by dividing the thickness per monolayer (actually bilayer, which is 2.583 Å and represents one half of the lattice parameter along the c-direction) by the growth rate. Several observations can be gleaned from Figure 3.44. The term lS,Ga increases with temperature and this temperature dependence is solely owing to the temperature dependence of DS,Ga because tI is determined by the GR. As the GR decreases, both tI and lS,Ga increase, and consequently, the Ga diffusivity increases, which should aid in the growth of a more ordered lattice because the number of adatoms that incorporate is increased. Enhancements in GaN crystalline quality have been attributed to reduced growth rates for the migration-enhanced epitaxy (MEE) approach where the surface is dosed with an alternating sequence of N and Ga exposures with dwell times between exposures in an effort to increase lS,Ga. At 1050 C, the estimated values of lS,Ga using tI (no desorption regime or complete incorporation) are about 10–100 times larger than the estimated values of lS,Ga based on tS,Ga (high desorption regime or growth rate limited case). For GaN growth by OMVPE, practical growth rates are typically >150 Å min1 (about 1 mm h1). At a growth rate of 300 Å min1, lS,Ga is about 11.3 nm at 1050 C. Turning our attention to the other species N, its surface diffusion length (lS,N) should be smaller than lS,Ga. Unlike Ga, no evidence for N diffusion was observed in
3.4 Nitride Growth Techniques
the RHEED intensity analysis in MBE when radio frequency (RF) activated nitrogen was used. This is not necessarily applicable to NH3 as RHEED oscillations have been observed. Assuming RF-activated nitrogen and N do not migrate outside the unit cell during the RHEED pattern, the recovery time, DS,N, must be <2.7 · 1017 cm2 s1 at 700 C, which is 40 times less than DS,Ga at 700 C. If the N lifetime is similar to the Ga lifetime, as might be expected in the case of NH3 but only as an upper limit, lS,N at 700 C is about six times less than lS,Ga. It is however very likely that lS,N is much less than lS,Ga, which is supported by ab initio calculations [230]. Even at atmospheric pressure, N2 desorbs from GaN above 800 C, suggesting that for T > 800 C, once N migrates to next to an adjacent N, N2 forms and desorbs. If the surface migration of atomic Ga and N is limited, partially decomposed reactants may diffuse more readily on the surface. This may be particularly important for OMVPE growth, because partially decomposed precursors that are weakly bound to the surface may more freely diffuse across the surface. Weaker N–N or N–Ga bonds between partially dehydrogenated NH3 and the GaN surface may form with these species diffusing more readily than Ga or N atoms. Therefore, the mean diffusion rates for Ga and N species on the surface may be substantially larger than the values calculated here using diffusion constants measured by MBE in a vacuum environment. 3.4.1.3.5 Kinetic Model: Balance Between Adsorption and Desorption Using the temperature dependence of desorption rates for N and Ga, kN and kGa, reported in Ref. [229] for an MBE environment, the ratio kN/kGa can be plotted and when done, a temperature dependence of kN/kGa ¼ 1.9205 · 1016 exp[39 563/T (K)] results. Using the kinetic values, kN/kGa is about 1000 at T ¼ 1020 C and about 0.1 at T ¼ 725 C. By choice of temperatures, these values of kN/kGa are comparable with the V/III ratios used for OMVPE and MBE growth, respectively. In addition to MBE analogue, if one also relies on the models developed for GaAs [231], one can then construct a kinetic model for GaN grown by OMVPE where activation barriers play an important role as done by Koleske et al. [221]. In this scenario, the Ga and N surface coverage, yGa and yN can be related to the incident fluxes, F, and desorption rates, k, as follows:
dqGa ¼ sGa F Ga kGa qGa vG ðqGa ; qN Þ for Ga; dt
ð3:20Þ
dqN ¼ sN F N mkN qm N vG ðqGa ; qN Þ for N; dt
ð3:21Þ
and
where s is the precursor sticking coefficient, m is the kinetic order of desorption, and vG is the growth rate. The Ga and N precursors decompose to atomic species (as discussed in Section 3.4.1.3.1 and SGa ¼ 1. For the NH3 sticking coefficient, SN, we assume that NH3 does not stick to surface sites already occupied by N. The value of SN will therefore depend on both yN and the NH3 decomposition mechanism. If two adjacent sites are necessary for the sticking and dissociation, then SN ¼ (1 yN)2; however, if only one site is required for NH3 adsorption, which is followed by
j405
j 3 Growth and Growth Methods for Nitride Semiconductors
406
dissociation, then SN ¼ (1 yN). The growth rate, vG, is then given by vG ðqGa ; qN Þ ¼ g GaN qGa qN dGaN ð1 qN Þ;
ð3:22Þ
where g is the incorporation rate of the surface atoms into the crystal and d is the decomposition rate of atoms from the crystal to the surface (see Section 3.4.1.3.1). The incorporation rate (the first term on the right-hand side) depends on the thermal activation of the adatoms to overcome the kinetic barriers for bonding to step edges and surface diffusion. Because both the incorporation rate and the decomposition rate depend on combinations of activation barriers (kinetic factors) [231], both increase exponentially with temperature. For growth, the incorporation rate, the gGaNyGayN term, is slightly larger than the decomposition rate, the dGaN ð1 qN Þ term. For GaAs growth, the incorporation rate is maximum near yGa ¼ yN with yGa and yN both less than 0.25 ML [231]. Above the temperature where the preceding conditions apply, the desorption rate approaches the incorporation rate, resulting in a rapid decrease in vG. Clearly, having these two rates as large as possible will lead to improved ordering of the lattice, resulting in a near equilibrium growth. Under steady state, the change in surface coverage (left sides of Equations 3.20 and 3.21) is zero, which leads to sGa F Ga ¼ kGa qGa
and
sN F N ¼ mkN ðqN Þm :
ð3:23Þ
To achieve the condition where yGa ¼ yN, the ratio of incident flux and desorption flux for the two components must be equal, or sGa F Ga =kGa ¼ ðsN F N =mkN Þ1=m :
ð3:24Þ
As discussed in Section 3.4.1.3.2, N2 desorption from GaN has been observed to be a first-order process, which means that m ¼ 1. Assuming that SGa ¼ 1 and near equilibrium, and yN is small, sN ¼ ð1 qN Þx 1 results. Using these relations, Equation 3.24 becomes, F N =kN ¼ F Ga =kGa
or after rearranging
F N =F Ga ¼ kN =kGa :
ð3:25Þ
Equation 3.25 implies that for near equilibrium growth where yGa ¼ yN, the ratio of input fluxes, FN/FGa, must match the ratio of the desorption fluxes, kN/kGa. Because the N surface diffusion rate is nearly zero (see Section 3.4.2 for details regarding growth under N-rich conditions), the N incorporation rate will depend on the number of times N atoms are adsorbed and desorbed on the surface. For growth to take place, vG > 0, g GaN qGa qN > dGaN ð1 qN Þ, kN > kG, and FGa > kGa, which lead to F N =F Ga > kN =kGa :
ð3:26Þ
Because kN/kGa depends exponentially on the temperature, the choice of FN/FGa also depends exponentially on the temperature. 3.4.1.3.6 Comparison of Model with Growth Conditions for Surface Morphology It is beneficial to relate values of these terms to various growth processes. For MBE growth of GaN, growth temperature, TG ¼ 600–800 C, and the films are grown in either
3.4 Nitride Growth Techniques
Ga-rich conditions, that is,FN/FGa < 1,or nitrogen-rich conditions, that is, FN/FGa about 1–10. For HVPE and OMVPE growth, TG ¼ 800–1100 C and FN/FGa ranges from 100 to 10 000, depending on the growth temperature. It is also useful to relate the predicted condition for optimized growth, that is, FN/FGa ¼ kN/kGa to the growth temperature, TG, V/III ratio, and the eventually resultant material quality. Data for this comparison were taken from the literature from several sources to determine the applicability of the growth model forwarded [221]. Generally, the data support the model and show how the V/III ratio and TG must be selected for the growth of high-quality GaN. For OMVPE growth on sapphire with a nucleation layer, if the V/III ratio is too low or TG is too high, both of which lead to the condition FN/FGa < kN/kGa, Ga droplets form to produce a rough surface morphology. Conversely, when FN/FGa > kN/kGa or quite N rich, the film morphology is smooth. An example of the change from smooth to rough morphology as affected by the V/III ratio for a constant growth temperature is shown in Figure 3.45a, where GaN was grown at a fixed TG using different V/III ratios. The filled circles denote GaN films with a smooth morphology, while the open circles denote a film with a rough morphology. Pictures of the two surface morphologies are also shown in Figure 3.45a, along with the kN/kG line. Note that the kN/kGa line properly depicts growth conditions that result in either smooth, that is, FN/FGa > kN/kG, or rough, that is, FN/FGa < kN/kGa, surface morphology. To further illustrate this, the values of the TG and V/III ratios used by many researchers are plotted in Figure 3.45b for GaN grown on either AlN or GaN nucleation layers. Again the open squares denote growth conditions leading to rough morphology, whereas the filled circles denote growth conditions where either smooth surface morphology was obtained or optimized films, judged by electronic properties, were grown. For each of the data points, the V/III ratio and TG were assumed to be reported correctly. Note that all but three of the filled circles are above the kN/kGa line and that all of the open circles fall below the kN/kGa line. Knowledge of the growth process would pave the way for optimizing growth conditions to obtain lower free carrier concentrations and higher mobilities, narrower XRD peaks, and intense photoluminescence (PL) by Koleske et al. [221]. Optimized growth temperature values and V/III ratios taken from five different references are shown in Figure 3.46. For example, the growth conditions leading to optimum electron mobility (500 cm2 V1 s1) occur at TG ¼ 1040 C and a V/III ratio of 1590. This condition is very close to the kN/kGa line shown in Figure 3.46. For minimum symmetric X-ray linewidth, the conditions TG ¼ 1025 C and a V/III ratio ¼ 1240 were used, which is slightly above the kN/kGa line. However, asymmetric diffraction linewidth must be observed to draw conclusions about the quality, as this diffraction is sensitive to in-plane extended defects. If the ratio of the band edge emission and yellow emission is used as a quality indicator in terms of radiative recombination, TG ¼ 1080 C and a V/III ratio of 5130 conditions had the best figures. This too is very close to the kN/kGa line. However, the ratio monitored strongly depends on the excitation intensity and further extrapolations based on this ratio alone should be handled with caution. The kinetic model presented above brings to the foreground the importance of the growth parameters. Specifically, the control of V/III ratio and TG is essential for the
j407
j 3 Growth and Growth Methods for Nitride Semiconductors
408
Figure 3.45 (a) Values of the V/III ratios for a growth temperature of above 1040 C. The inset pictures represent the surface morphology for the growth conditions plotted. The filled circles denote smooth morphology and the open circle denotes a rough morphology. The line is the ratio of kN to kGa. (b) Values of the growth temperature
and V/III ratio for smooth surface morphology films or a film grown with optimized growth temperature and V/III ratio (filled circles) and rough morphology films (open squares). Data were taken from 39 sources, which are listed in Ref. [221] where the details can be found. The line is the ratio of kN to kGa. Courtesy of D. Koleske.
reproducible growth of optimized material. Available mass flow controllers (MFC) used in OMVPE systems control flows to within 0.5–1%, which translates to a V/III ratio control to within 2%. The temperature control near 1000–1100 C is more challenging, and there is a greater uncertainty in both setting TG and measuring it. Temperature nonuniformities across the substrate are common; especially for larger substrates despite great strides made in uniform heat coupling and substrate rotation. This temperature nonuniformity leads to nonuniformities in the composition and the physical properties of alloys in the following manner. For growth at 1000 C, a 1 C temperature difference shifts the kN/kGa ratio by 15 out of a total kN/ kGa of 611, which represents a 2.5% change, and at 1100 C, the kN/kGa shifts by 123 out of the total kN/kGa of 5880, representing a 2.1% change. If the wafer center is 10 C
3.4 Nitride Growth Techniques
V/III ratio
104
Conditions for optimized GaN growth
103
k N/k Ga line
102
900
1000
1100
Temperature (ºC) Figure 3.46 Optimized growth temperature and V/III ratio values taken from five different references (see Ref. [221] for details). The kN/kGa line is also plotted. Note the good correlation between the optimized growth conditions and the kN/kGa line.
cooler than the edge for TG ¼ 1040 C, the kN/kGa ratio will be 25% larger at the edge than at the center. These examples demonstrate that to have uniform material properties over the entire wafer, the TG needs to be accurately and reproducibly controlled [221]. Before we delve into the world of MBE, which is next, a discussion for historical content is provided. In spite of the early assumption that CVD process using ammonia led to N-deficient layers and plasma-activated N would lead to better films, further research and ensuing high-quality layers proved this hypothesis to be invalid. For the sake of completeness, a few words are shared here on the topic. Although trialkyl compounds (TMA, TMG, TMI, etc.) are pyrophoric and extremely water and oxygen sensitive, and ammonia is highly corrosive, much of the best material grown today is produced by conventional OMVPE by reacting these trialkyl compounds with NH3 at substrate temperatures in the vicinity of 1000 C [196,232–238]. Temperatures in excess of 800 C are required to obtain single crystalline high-quality GaN films, and the GaN films with the best electrical and optical properties are grown at or around 1050 C. At substrate temperatures exceeding 1100 C, the dissociation of GaN results in voids in the grown layer. A similar situation is observed also for AlN film growth. 3.4.2 Molecular Beam Epitaxy
In this section, a succinct overall view of growth by MBE is given followed by an indepth discussion of physical processes that take place in MBE growth environments. In MBE technique, thin films are formed in vacuum on a heated substrate through various reactions between thermal molecular beams of the constituent elements and
j409
j 3 Growth and Growth Methods for Nitride Semiconductors
410
the surface species on the substrate [239]. The composition of the epilayer and its doping level depend on the arrival rates of the constituent elements and dopants [240]. Therefore, MBE growth is carried out under conditions governed primarily by the kinetics, rather than by mass transfer [241]. A thorough understanding of the growth kinetics, especially the surface processes of growth, is therefore critical. MBE is an extremely versatile technique for preparing thin semiconductor heterostructures owing to the control over the growth parameters that it offers, and the inherent in-situ monitoring capability. As mentioned earlier, thin films are formed on a heated substrate through various reactions between thermal molecular beams (atomic beams in the case of RF-activated nitrogen) of the constituent elements participated by the surface species on the substrate originating from the substrate itself by surface or bulk contamination. The composition of the epilayer and its doping level depend on the arrival rates of the constituent elements and dopants, respectively. The typical growth rate of 1 mm h1, or slightly more than one monolayer per second (ML s1), is sufficiently low to allow for surface migration of the impinging species on the surface. In the case of growth along the h1 1 1i for cubic and c-directions for wurtzitic systems, one monolayer constitutes a bilayer. As MBE growth occurs under conditions that are governed primarily by the kinetics, rather than by mass transfer, this allows the preparation of many different structures that are otherwise not possible to attain. This is in contrast to OMVPE of conventional compound semiconductors, such as GaAs. However, in the case of GaN, even the OMVPE has elements of kinetics [242]. In nitride growth by MBE, the metal species are provided by Ga, In, and Al metal sources, the dopants are provided by pure Si for n-type and Mg for p-type using conventional Knudsen effusion cells that are heated to sufficient temperatures for the desired growth rate, composition, and doping levels. On the contrary, nitrogen gas is one of the least reactive gases because of its large molecular cohesive energy (946.04 kJ mol ¼ 9.8 eV per N2 molecule). Because of the triple bond between the two nitrogen atoms, dissociation of one molecule into two reactive nitrogen atoms requires a large amount of energy, which cannot be provided by thermal means. In a plasma environment, however, and at reduced pressures, a significant dissociation of the nitrogen molecules takes place. Atomic nitrogen is reactive even at room temperature and bonds with many metals. Consequently, group III nitrides can be grown by plasma-assisted molecular beam epitaxy, where the plasma-induced fragmentation of nitrogen molecules is combined with the evaporation of metal atoms from effusion cells. In this vein, MBE growth of GaN has been reported by electron cyclotron resonance microwave plasma assisted molecular beam epitaxy (ECR-MBE). Several laboratories in the past have attempted RMBE growth in which N2 or NH3 was decomposed on the substrate surface [243–247]. The sequence of processes taking place during growth by MBE is adsorption, desorption, surface diffusion, incorporation and decomposition. All of these processes are in effect in many ways compete with each other during growth by MBE. Adsorption can be summed as the atoms or molecules impinging on the substrate surface and sticking by overcoming an activation barrier. Desorption, on the contrary, is a process in which the species that are not incorporated into the crystal lattice leave the substrate surface by thermal vibration. During surface diffusion, imperative for
3.4 Nitride Growth Techniques
Impinging atoms
Lattice site
Desorption
Surface diffusion
Adsorption Surface diffusion Decomposition
Substrate, T s Overgrowth Incorporation into lattice
Surface nucleation (aggregation)
Interdiffusion
Figure 3.47 Surface processes during the MBE growth: adsorption, desorption, surface diffusion, lattice incorporation, and decomposition.
growth, the constituent atoms or molecules diffuse on the substrate surface to find the low-energy crystal sites for incorporation. During the incorporation phase, the constituent atoms or molecules enter the crystal lattice of the substrate or the epilayer already grown by attaching to a dangling bond, vacancy, step edge, and so on. Owing to high temperatures involved, albeit much lower than those employed in OMVPE, during decomposition, the atoms in the crystal lattice leave the surface by breaking the bond. These processes are schematically shown in Figure 3.47. Let us discuss these five processes that govern growth by MBE. 3.4.2.1 Adsorption Impinging gas molecules (atoms) condense on the surface and, depending on the strength of the interaction between the adsorbate and the surface, can be adsorped by physisorption (weak) or chemisorption (strong) [248]. Physisorption represents weak adsorbate–surface interaction because of van der Waals forces with typical binding energies on the order of 10–100 meV. Owing to the weak interaction, physisorbed atoms or molecules do not disturb the structural environment near the adsorption sites to a significant degree. In chemisorption, on the contrary, an adsorbate forms strong chemical bonds with the substrate atoms with typical binding energies on the order of 1–10 eV, thus changing the adsorbates chemical state.
j411
j 3 Growth and Growth Methods for Nitride Semiconductors
412
The adsorbate coverage characterizes the surface concentration of adsorbed species expressed in monolayer units. The coverage is a relative value associated with a given substrate. It can be converted to an absolute surface density of atoms. Considering the kinetic approach in the case of a uniform solid surface exposed to an adsorbing gas, the adsorption rate is defined as [248] p r a ¼ sf ðQÞexpðE act =kB T s Þ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ; 2pmkB T s
ð3:27Þ
where p is the partial pressure of the adsorbing gas; s is the condensation coefficient responsible for the effects of the orientation and the energy accommodation of the adsorbed molecules; f(Y) is a coverage-dependent function that describes the probability of finding an adsorption site; exp(Eact/kBT) is the temperature-dependent Boltzmann term associated with the energetics of the activated adsorption. Finite equilibrium Ga adlayer coverage has been reported for typical substrate temperatures and Ga fluxes [249]. For large Ga fluxes, up to 2.5 0.2 monolayers of Ga are adsorbed on the GaN surface. For higher Ga fluxes, Ga droplets are formed [250]. At typical growth temperatures used in MBE, Ga adlayer does not condense into a reconstruction but behaves like a liquid film [251]. The reported height of the Ga adlayer is about 3.88 Å, as measured by scanning tunneling microscopy (STM), which corresponds to 1.9 ML [252]. For lower Ga fluxes, a discontinuous transition to Ga monolayer equilibrium coverage is found, followed by a continuous decrease toward zero coverage. The practical implications of this process in terms of GaN growth are discussed in Section 3.4.1.3.1. 3.4.2.2 Desorption Desorption is a process in which the adsorbate species gain sufficient thermal energy to escape from the adsorption well and leave the surface. The probability of desorption depends on the bonding strength of the particular atom to the surface. Bonding energies are different for different materials and the strength of a bond is expressed in terms of the amount of energy needed to break it. In the kinetic approach, desorption is described in terms of a desorption rate, rdes, which represents the number of species desorbed from unit surface area per unit time, and can be expressed as [248]
r des ¼ s f ðQÞexpð E des =kB T s Þ;
ð3:28Þ
where f (Y) describes the coverage dependence and s is the desorption coefficient standing for steric and surface mobility factors. For desorption to occur, the adsorbed species must overcome a barrier called the desorption energy, Edes. In case of activated chemisorption, the desorption energy is the sum of the binding energy in the chemisorbed state and the activation energy for adsorption, Edes ¼ Eads þ Eact. In case of nonactivated chemisorption, the desorption energy is simply the binding energy in the chemisorbed state, Edes ¼ Eads. The lifetime of an adsorbate as a function of temperature is needed for studying the desorption energy. The lifetime t is defined as the average time spent by the adsorbate on the surface marked from the time of adsorption to the time of desorption and
3.4 Nitride Growth Techniques
obeys an Arrhenius dependence in the form [250]: E des t ¼ t0 exp : kB T s
ð3:29Þ
Equations 3.28 and 3.29 describe the desorption process to be quite sensitive to temperature. For low Ts, the lifetime of adsorbates is sufficient to point that the desorption process can be neglected. For intermediate temperatures, the growth rate is determined through the competing processes of adsorption and desorption. However, for sufficiently high Ts, the desorption rate can be greater than the deposition rate, in which case evaporation rather than deposition would take place. Specific to the case of GaN, Ga desorption has been investigated by mass spectrometry [250,253] and RHEED [254,255] with consistency lacking. The measured activation energies reported for Ga desorption are in the range of 0.4–5.1 eV [250–255]. For N, which is usually desorbed as a dimer (N2), the desorption rate is limited by the surface diffusion of two N atoms. Some reports indicate this to be a first-order process [256], meaning that the surface diffusion may play less of a role in the desorption mechanism than the surface structure. Another plausible mechanism is that once an N atom diffuses near another N atom, the N2 molecule would immediately form owing in part to the large N–N binding energy and desorbs from the surface because of the highly exothermic nature of the N2 formation [257]. The practical implications of these processes in terms of GaN growth are discussed in Section 3.4.1.3.1. 3.4.2.3 Surface Diffusion As mentioned earlier, surface diffusion describes the motion of adsorbates (atoms or molecules) on the substrate (film) surface. The motion of adsorbates that become mobile because of thermal activation is described as a random walk. When a concentration gradient is present, this random walk motion of many particles results in their net diffusion being opposite to the gradient direction in a macroscopic sense. The microscopic view of surface diffusion is an activated process that is also affected by factors such as interaction between diffusing adsorbates, formation of surface phases, and defects. For an atom on the surface to diffuse to the next lattice position, it must overcome the lattice potential between the two neighboring positions. This activation energy required for diffusion, E0, is the microscopic origin of the lattice potential. The average length of diffusion, ls, in a unit time interval is a very important parameter to characterize the diffusion process and has an exponential temperature dependence given by Equation 3.19, called the Einstein equation [248]. At high growth temperatures where the surface diffusion length is larger than the terrace width, atoms move by surface diffusion from the terrace to the step edge to be incorporated into the growing lattice, which is called the step-flow growth. As the diffusion length decreases, for example, by reduced temperature, atoms meet and nucleate new islands on the terraces before reaching the edge of an existing island or a step edge. Moreover, islands can nucleate on top of the existing islands, a process that leads to rough surface formation and 3D growth. In the intermediate region, the
j413
j 3 Growth and Growth Methods for Nitride Semiconductors
414
(0 0 0 1)
Top view (c-plane)
(0 0 0 1)
hcp
Diffusion path
Bridge On top
N layer
Ga2L fcc Ga1L
Ga layer
Side view (a-plane)
Ga1L N Ga2L (a)
(b)
Figure 3.48 Schematic diagram of diffusion paths on the GaN (0 0 0 1) (a) and (0 0 0 1) (b) surfaces. The side views as in the a-plane for each are also shown [256].
islands do form on terraces, but the diffusion length is sufficiently long for adsorbates to diffuse to the edges of these islands and incorporate there, which leads to smooth surface also. In this mode of growth, the RHEED intensity oscillations are observed in that the intensity is maximum when the islands expand to cover the terrace and the intensity is minimum when the coverage is 50%. Specific to the case of GaN, diffusivity for Ga and N adatoms on GaN surface is different. Zywietz et al. [256] calculated the surface potential energy for Ga and N adatoms on GaN(0 0 0 1) surface. The simulations led to the presence of two transition sites, as shown in Figure 3.48. For Ga diffusion, the lower energy transition site is the bridge position (0.4 eV) and the higher energy site is the on-top position (>3 eV). However, for N adatoms, the barrier for bridge diffusion is 1.4 eV and the barrier for the on-top position is similar to Ga. Significantly lower diffusion barrier for Ga relative to N results in Ga being very mobile at typical growth temperatures, whereas the diffusion of N is slower by orders of magnitude. Further, the presence of excess N strongly increases the Ga diffusion barrier from 0.4 to 1.8 eV. The very divergent surface mobilities of Ga and N adatoms have serious consequences. In the Ga-rich regime, the Ga adatoms are highly mobile and a step-flow mode results in 2D growth. Furthermore, if excess Ga adatoms are present on the surface, N adatoms can be efficiently incorporated because the probability of
3.4 Nitride Growth Techniques
fast-moving Ga adatoms capturing N atoms is high. The presence of a Ga bilayer on the surface in Ga-rich growth conditions reduces the lateral diffusion barrier from about 1.3 to about 0.5 eV, paving the way for growth of smooth layers at low temperatures [258]. For details on the Ga-bilayer issue, refer to Section 3.5.6. However, N-rich growth conditions show roughly a five times higher diffusion barrier for Ga. A rough surface can, therefore, be expected for N-rich conditions, consistent with experimental observations. 3.4.2.4 Incorporation In the incorporation process, the molecules or atoms bond to the crystal through various reactions between the constituent elements and the surface species on the substrate. This process is controlled by the interplay of thermodynamics and kinetics. The general trends in film growth are understood within the thermodynamic approach in terms of the relative surface and interface energies. However, film growth by MBE is a nonequilibrium kinetic process in which rate-limiting steps affect the growth mode [241]. There are three modes of growth/incorporation that come to bear in general, namely, Frank–van der Merve (FM), Stranski–Krastanov (SK), and Volmer–Weber (VW). These three processes with graphics are discussed in Section 3.5.16, which pertains to quantum dots. Suffice it to say, the Frank–van der Merve mode is a layer-by-layer process. Each layer is fully completed before the next layer starts to grow, and it is strictly a two-dimensional growth mode. The Volmer–Weber mode is an island growth mode. Three-dimensional islands nucleate and grow directly on the substrate surface, which is typically seen in metals unless very low – even below room temperature – deposition conditions are employed. The Stranski–Krastanov mode is a layer-plus-island growth mode and represents the intermediate case between the FM and VW growth modes. After the formation of a complete two-dimensional layer of a few monolayer thickness (the exact value of which depends on the local strain), the growth of a three-dimensional layer (islands) takes place. The occurrences of growth have their genesis in the competition between the surface desorption and the surface diffusion. Because the desorption and diffusion processes are noticeably affected by the deposition rate, the surface condition and temperature, the growth mode, and the epilayer surface can be controlled by choosing a proper III/V ratio and a substrate temperature. Several studies [259,260] have revealed that not all incident Ga atoms are incorporated into the growing GaN at the usual growth temperature in PAMBE (650–750 C), even when an excess of N flux is present (III/V flux ratio <1). Guha et al. [250] studied the incorporation ratio of Ga during GaN growth under N-rich conditions by monitoring the reflected Ga signals detected by the mass spectrometer and found that the Ga incorporation rate is sensitive to the growth temperature. At low growth temperatures, the lifetime of Ga adatoms is sufficiently long so that almost all Ga atoms encounter N adatoms and are incorporated into the crystal, resulting in a near unity incorporation ratio. As the growth temperature increases, the residence time of a Ga adatom on the surface decreases. Thus, the probability of Ga adatoms encountering N adatoms decreases and their incorporation ratio drops.
j415
j 3 Growth and Growth Methods for Nitride Semiconductors
416
3.4.2.5 Decomposition As motioned previously, when the desorption rate is greater than the incorporation rate, the film would begin to decompose. In compound semiconductors such as GaN, one compound splits into two (or more pieces) as the bonds between the constituents is broken. The III nitrides have chemical bonds with the Ga–N bond strength estimated near 2.2 eV [261]. In addition, III nitride bonds have a sizeable ionic bond (31–40%) nature (excluding BN) than other III–V semiconductors (<8%) [262]. A lower ionic bond would have resulted in larger covalent bond with much larger bond strength. As in the case of other III–V semiconductors, the heat of formation of GaN is small and similar to that of InP and AlAs, indicating that GaN easily decomposes. In this vein, GaN does not melt congruently at pressures typically used in MBE growth, but decomposes above 800 C at atmospheric pressure and at lower temperatures in vacuum. Guha et al. [250] reported a decomposition rate of three to four monolayers per minute at 830 C. Grandjean et al. [222] found that the decomposition rate is nearly zero below 750 C, but increases rapidly above 800 C, and reaches 1 mm h1 at 850 C. This means that it may be impossible to grow GaN at high temperatures and that the growth temperature should be kept below 800 C in relation to growth in vacuum. Several different mechanisms have been proposed to explain GaN decomposition. These include decomposition into gaseous Ga and nitrogen, liquid Ga and nitrogen, and sublimation of GaN into a diatomic or polymeric product. One would surmise that these critical temperatures could be altered to some extent by the overpressure of N or N-containing reactive species used for growth, even though reports indicate that, to a large extent, GaN decomposition rate is similar in vacuum and at OMVPE pressures. However, a good deal of dispersion in the data allow us to assume that GaN is no different and that group V overpressure should have some affect on the rate of decomposition. In fact, GaN growth temperature with OMVPE is well in excess of 1050 C, and if the vacuum experiments regarding decomposition rates were to hold, it would be nearly impossible to achieve deposition. To make some sense of all these sometime competing processes, involving adsorption, desorption, surface diffusion, incorporation, and decomposition, it is compelling to summarize. The conventional III–V semiconductors, and nitrides are no exception, are grown under conditions (substrate temperature and system pressure) in which the III–V compounds and the gaseous environment with which they are in contact are thermodynamically stable. For high-quality compound semiconductor growth, the optimum substrate temperatures used are in the range of 1/2 to 2/3 of the melting temperature of the semiconductor. In the case of III nitrides, however, synthesis occurs, particularly with MBE, at temperatures significantly below 1/2 of the predicted melting temperature. Specific to GaN, the substrate temperature during MBE growth is about 20 to 35% (600–900 C) of the GaN melting temperature (TM 2500 C). Thus, GaN growth takes place in kinetic regime and the surface processes are critical in determining the quality. The synthesis of GaN involves a metastable growth process, which is controlled by a competition between the forward reaction (incorporation of Ga and N into the film and GaN epilayer forms) and the
3.4 Nitride Growth Techniques
reverse reaction (decomposition of GaN). The forward reaction depends on the arrival rates of Ga atoms and activated nitrogen species on the surface, as well as the substrate temperature, whereas the reverse reaction is strongly affected only by the substrate temperature. For a net growth to take place, the rate of GaN formation must be larger than the rate of decomposition. The remaining three additional processes determining the eventual interface morphology of a growing film include deposition, desorption, and surface diffusion. Their relative importance depends on the microscopic properties of the interface, the bonding energies, and the diffusion barriers. The experimentally controllable parameters are the arrival rates of group III and V compounds and their ratios and substrate temperature. By tuning these parameters, a variety of morphologies can be achieved, ranging from layer-by-layer growth with an essentially smooth interface to a rough self-affined surface. 3.4.2.6 Reflection High-Energy Electron Diffraction By far the most commonly used characterization tool in MBE growth is RHEED [263– 265]. In established materials such as GaAs, this method is capable of providing a window on processes taking place on the surface of the growing film at the atomic scale. A discussion of RHEED, which is synonymous with MBE, is therefore warranted. RHEED is a technique for probing the surface structure of solids in UHV. The method is based on diffraction of high-energy electrons from a clean and well-ordered surface. Electrons with high energy (typically 5–100 keV) and a grazing angle of incidence (typically <3 ) scatter elastically from only the first few atomic layers, making the method very surface sensitive. Compared to other diffraction methods, this glancing incidence angle geometry allows effusion cells to be placed in such a way that allows near normal radiation toward the substrates and at the same time allows real-time surface monitoring during growth [266]. The electron beam incident on the surface is also reflected following the laws of classical optics, the image of which on the phosphor screen is called the specular reflection spot. The intensity of this spot can be used to glean important information about the degree of smoothness or roughness of the surface. The growth regime in which growth initiates with islands on terraces, which eventually grow and coalesce, the intensity of the specular beam spot undergoes periodic oscillations as the growth proceeds, the period of which represents the time taken to grow one monolayer of material. This is used very successfully to monitor and even control, through a feedback loop, the growth of quantum wells and superlattices requiring ultimate thickness accuracy. As for diffraction, if electrons diffract only from the first atomic layer of a perfectly flat and ordered surface, the three-dimensional reciprocal lattice points degenerate into parallel infinite rods. In the Ewald construction that results from such an ideal case, the intersection of the Ewald sphere, whose radius is much larger than the interrod spacing for typical RHEED energies, consists only of a series of points placed on a half circle as seen in the phosphor screen depicted in Figure 3.49. In reality, however, thermal vibrations and imperfections cause the reciprocal lattice rods to have a finite thickness with additional contribution from the divergence and
j417
j 3 Growth and Growth Methods for Nitride Semiconductors
418
y Incident e-beam Diffracted e-beams k0
x k' Sample
z
Phosphor screen Figure 3.49 An artists view of the major elements of RHEED a system showing the incident and reflected beam wavevectors (k0 and k0 and resulting RHEED spots on the phosphor screen from an ideal surface wherein diffraction only from the top monolayer is assumed.
dispersion of the electron beam. Atomic steps associated with islands are the main source of broadening of the diffraction rods. In the case of GaAs, which has the subject of most of the RHEED-related investigations and ground-breaking research, the pattern starts off as a semicircle of dots and broadens into streaks as growth begins. This is not observed when the surface is rough and/or the growth mode is of the step-flow type. Therefore, even diffraction from a very flat semiconductor surface results in a diffraction pattern consisting of a series of streaks with modulated intensity rather than points expected from an ideal case. If the surface is not flat, many electrons will be transmitted through surface asperities (see the upper sketch in the inset of Figure 3.49) and scattered in different directions, resulting in a RHEED pattern comprising many spotty features. Therefore, the most obvious information gleaned from RHEED images pertains to the flatness of the surface. It also follows that diffraction from an amorphous surface such as an oxide on the semiconductor sample surface gives rise to no diffraction pattern at all and results in only a diffused background. This is pivotal in monitoring the thermal desorption of the oxide layer when the sample is initially heated prior to the commencement of the MBE growth. ! The geometric theory [265], in which diffraction of a plane wave (of wave vector k) by a single crystal is assumed, is the simplest one that can be used to describe RHEED. Before delving into RHEED, it would be instructive to review two-dimensional periodic lattice in real and k-space. Reciprocal lattice representations of periodic structures are fundamental to analytic treatments of phenomena like crystal diffraction and electron energy levels in solids. To analyze the results of this RHEED
3.4 Nitride Growth Techniques
experiment, you will need the two-dimensional version of the reciprocal lattice, which is sometimes called the reciprocal net. The surface of a crystalline material is nothing more than a two-dimensional diffraction grating in which atoms are arranged ! ! according to repetition of a unit mesh with basis vectors ! a1 ¼ ax and ! a2 ¼ ay, where ! ! x and y are the unit vectors in their respective directions and a is the spacing between the surface atoms in both the x- and y-directions. This real net has a reciprocal lattice whose basis vectors are given by !
b2 ¼
2p ! x a
!
and b2 ¼
2p ! y: a
ð3:30Þ !
!! ik r
wave e will A set of points in space located by a set of lattice vectors R and a plane ! have the periodicity condition satisfied for arbitrary wave vector k. The periodicity ! conditions are satisfied only for a set of wave vectors K , called the reciprocal lattice, which yields plane waves with periodicity of the lattice. Mathematically, the periodicity condition can be expressed as ! !
eiK ðr
!
þR Þ
!!
¼ ei k r ; !!
ð3:31Þ !
which leads to eiK R ¼ 1, which, in turn, must be must be true for all lattice vectors R, and it is this relationship that defines the reciprocal lattice. In kinematical scattering theory [267], the maximum value for scattering amplitude is obtained when the incident beam wave vector k0 and the diffracted beam wave vector k0 differ by the reciprocal lattice wave vector K of the surface atoms. In this theory, constructive interference will occur, provided the change in the wave vector is a vector of the reciprocal space, meaning diffraction results when the Laue condition [268] is satisfied, that is, !
!
!
k0 k0 ¼ K ; ! 0
! k0
ð3:32Þ
where k and are the wave vectors for the diffracted and the incident beams, ! respectively, and K is the reciprocal lattice vector. In the special case of elastic scattering, the amplitudes of the diffracted and incident electron beam wave vectors0 0 ! ! ! are equal, that is, jk j ¼ j k0 j. This condition is satisfied by an infinite number of k vectors pointing in all directions, which is the origin of the Ewald sphere. The origin ! of the Ewald sphere is same as that of the k0 , and its radius is equal to the absolute ! value of k0 . The Bragg diffraction and/or the Lau treatment indicate that the beam intensity decreases rapidly for Dk 6¼ K. Consequently, when the diffraction condition, Dk ¼ K, is satisfied, the bright spots appear on the phosphor screen, as shown in Figure 3.50. This can be visualized with the aid of a 3D Ewald sphere [268,269] of radius k, as depicted in Figure 3.51a–c. Only when the sphere passes through a reciprocal lattice point, the difference between incident and diffracted beam wave vectors is a reciprocal lattice vector K and the intensity of diffraction is the highest. Considering that the reciprocal lattice separation is inversely proportional to distances in real space, a 2D plane in real space should be a one-dimensional (1D) line in reciprocal space. As a result, a real and smooth 2D surface should be a set
j419
j 3 Growth and Growth Methods for Nitride Semiconductors
420
Rod
k–k'
R
Origin
Possible k–k' wave vectors
k–k'
k–k' R
Plan view of 2D ordered atoms
Side view of the 2D ordered atoms
Figure 3.50 The plan view of two-dimensional ordered atoms representing the surface atoms in a RHEED experiment with a ! real surface vector of r . The figure on the right side also shows the side view of the real space network and schematic description of the origination of the lattice rods. Adapted from Ref. [270].
of infinitely long rods in the reciprocal space (Figure 3.51), leading to a streaky RHEED pattern. For a 3D rough surface, the RHEED pattern becomes spotty (Figure 3.51). The electron energy determines the wavelength and the number of Lau rings as well as rod spacing observed on the phosphor screen. The magnitude of the wave vector is given as sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 2p 1 E ¼ 2m0 E þ ; ð3:33Þ k0 ¼ l h c where m0 is the electron rest mass and E is the accelerating energy. Equation 3.33 can be simplified and the wavelength can be expressed as [270] sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi h 150:4 12:26 ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi q l¼ ffi ffi pffiffiffi ðlin Awhen E is in eVÞ: 6 2 Eð1þ1:95x10 Þ 2 E 2m E þðE =cÞ 0
ð3:34Þ For a primary beam energy of 5–50 KeV, l values ranging from 0.17 to 0.055 Å result [271]. The theory discussed above is good for a first-order understanding of RHEED but falls short of describing the mechanisms involved. More advanced theories [265,272] treat the interaction between the incident wave and the scatterer quantum mechanically by solving the Schr€odinger equation for the wave function of the scattered wave
3.4 Nitride Growth Techniques !
yðr Þ (this wave function is the single-electron wave function that is a superposition of incident and scattered wave functions) given an effective potential U(r): !
!
ðr2 þ Uðr Þ þ k20 Þyðr Þ ¼ 0:
ð3:35Þ
Using Greens method, the solution for the above equation could be written as ! 1 ð exp ik j !r r! 0 j ! ! 0 ! ! ! Uðr 0 Þyðr 0 Þ dr 0 : yðr Þ ¼ exp ik0 r þ 4p jr! ?r! 0j
Figure 3.51 (a) The 3D Ewald sphere configuration; (b) a real 2D smooth surface in the reciprocal space.
ð3:36Þ
j421
j 3 Growth and Growth Methods for Nitride Semiconductors
422
k(–1,0)
k(0,0)
k in
(0,0,0)
(b) Figure 3.51 (Continued)
The first term in the above equation represents the incident plane wave, whereas the second term represents the scattered wave. In RHEED experiments, one is interested in the value of y(r) evaluated at large distances compared0 to the dimensions !0 ! ! ! of the scatterer, that is, jr r j r; hence, for large jr r j (or for r ! 1), Equation 3.36 can be written as ! ! ! e i k 0 r ð ! 0 ! ! ! ! ! e i k r Uðr 0 Þyðr 0 Þdr 0 : yðr Þ ¼ exp ik0 r þ 4pr !
So, the scattering amplitude has the form ð ! 0 !0 ik r 1 ! ! ! ! 0 Fðk Þ ¼ e Uðr 0 Þyðr 0 Þdr 0 : 4p
ð3:37Þ
ð3:38Þ
3.4 Nitride Growth Techniques
To discover the intensity distribution in the diffraction pattern, the effective potential should be assumed and the scattering amplitude integral should be evaluated. If the Born approximation is used, that is, the wave function inside the crystal is assumed to be that of the incident wave, Equation 3.38 can be written as ð !0 ! ! 0 1 ! ! ð3:39Þ FðK Þ ¼ e i k r Uðr 0 Þd r 0 ; 4p !
!
!
with the scattering vector k 0 k0 ¼ K . In other words, the scattering amplitude is the Fourier transform of the scattering potential. A number of theoretical calculations have been made to calculate the scattering amplitude, assuming functional forms for U(r) that take into account the periodicity of the crystal lattice. The main problem with the kinematical treatment is the oversimplified assumption that the wave function at the scatterer equals that of the incident plane wave, because this assumption overlooks the mutual interaction between the crystal and the incident electron beam. The kinematic theory ignores the ! details of the potential nearly entirely. If one substitutes for Uðr 0 Þ a sum of atomic potentials, then an equation for the relative diffracted intensity can be obtained as X ! I ¼ Fðs!Þ2 ¼ f 2 i; j exp½is ðr!i r!j Þ; ð3:40Þ where f is a scattering factor for identical scatterers that depends on the momentum ! transfer s . The potential is hidden in f and the position of the beams depends on the sum of the exponentials. It is kinematic (as opposed to dynamic) as the positions just depend on momentum and energy conservation. For the kinematic theory, the potential is included in the Born approximation in f, the scattering factor. However, it is just not included properly because the higher order terms of the Born approximation and, therefore, multiple scattering are neglected. Only in the case of weak surface scattering, when one can choose the diffraction condition in such a way that there is only one dominant diffracted beam, can kinematic theory be a good approximation. Because the mean free path of elastically scattered electrons in a crystal is shorter than that of inelastically scattered electrons, multiple scattering of electrons takes place. For the quantitative analysis of electron diffraction intensities, however, it is not sufficient because the strong interaction between the crystal and incident electrons changes the intensity of the total beam. A more elaborate theory, dynamical theory, has been introduced to deal with the diffraction problem without the oversimplification inherent to the kinematical theory. Dynamic scattering includes the potential and the main feature of dynamical theory is that it includes more terms of the Born approximation, taking multiple scattering into account. For electron scattering from a crystal, the wave within the crystal may be represented by a sum of plane waves yðr 0 Þ ¼
N X l¼0
!0 !0 yi ðr 0 Þe ik r :
ð3:41Þ
j423
j 3 Growth and Growth Methods for Nitride Semiconductors
424
Substituting Equation 3.41 into Equation 3.37 and using the Laue condition, k1 k0 ¼ 2ad, the scattering amplitude could be written as f ðkÞ ¼
N X
f l ðK l Þ ¼
l¼0
N 1 X 4p l¼0
ð
yl ðr 0 Þ e iðK k0 2ad Uðr 0 Þ dr 0 :
ð3:42Þ
scatterer
Again, to find the scattering amplitude, the integrals in Equation 3.42 have to be solved. This difficulty is the reason why dynamical theory is seldom used for analyzing RHEED data. There are different simplified methods for dynamical calculation of the intensity of RHEED, which face similar difficulties and result in similar accuracies. All these methods decompose the wave function and the crystal potential into their Fourier components. Because real crystal surfaces include surface relaxation and reconstruction, the atomic arrangement perpendicular to the surface is not periodic, whereas the atomic arrangement parallel to the surface is. Therefore, the three-dimensional Fourier expansion near the surface is not adequate for RHEED calculations. The ! ! crystal potential Uðr Þ and the wave function yðr Þ are expanded in a two-dimensional Fourier series in the direction parallel to the surface as !
Uðr Þ ¼
N X
! !
r== ; U n ðzÞeiR n
ð3:43Þ
n¼0
!
yðr Þ ¼
N X
yn ðzÞ e
h ! ! i ! i k0== þ Rn r ==
;
ð3:44Þ
n¼0 !
!
where Rn is the reciprocal rod vector, k0== is the component of the incident wave vector parallel to the surface, and the direction of z is normal to the surface. Substituting Equations 3.43 and 3.44 into Equation 3.35, one can obtain a second-order differential equation in the matrix form as d2 YðzÞ þ AðzÞYðzÞ ¼ 0; dz2
ð3:45Þ !
!
where (C(z))n ¼ Cn(z) and ðAðzÞÞnm ¼ ½k20 ðk 0== þ Rn Þ2 d nm þ U n m ðzÞ. A(z) is an N · N matrix for the N-beam case, and a key issue in dynamical theory is what number of beams must be included so that the calculation results would not change significantly if further beams were included. Further analysis involves reduction of Equation 3.45 by defining new coefficients in terms of cn(z) to a first-order homogeneous differential equation d 0 FðzÞ þ iA ðzÞFðzÞ ¼ 0; dz
ð3:46Þ
where A0 (z) is now a 2N · 2N matrix. After carefully choosing incident angles and determining diffracted beams, one can obtain the eigenvalues and eigenvectors of the transfer matrix A0 (z) and solve Equation 3.46 for the simple case in which a crystal potential is assumed to be constant in the direction normal to the surface and periodic
3.4 Nitride Growth Techniques Gate valve
Computer
Substrate CCD camera and optics
RHEED gun Video cable RHEED power supply
Camera control cable
Shutters
Liquid nitrogen shroud Shutter control
Figure 3.52 Schematic diagram of the RHEED technique in an MBE system.
in directions parallel to the surface. These results for the single-slice case may then be used for the multislice case in which the crystal is divided into multiple thin slices parallel to the surface where within each slice the potential is constant in the z-direction. By applying appropriate boundary conditions and matching them at each interface, solutions for F(z) can be obtained and RHEED intensities can be determined. Further details can be found in Refs. [264,265,270,272,273]. A typical RHEED system consists of an electron gun, a phosphor screen, and an image-processing system. Figure 3.52 shows the typical design for a RHEED system in an MBE system. Typical acceleration voltages range from 10 to 30 keV. This high energy necessary to image a reciprocal space into the relatively small solid angle of the phosphor screen has the additional advantage of reducing the influence of stray fields. The electron beam interacts with the sample at an angle that is typically in the range of 0.5–2.5 . The diffracted electron intensity pattern is converted into visible light by a phosphor screen. A charge-coupled device (CCD) camera external to the UHV system is used to record the RHEED pattern. The real-time still images and video of the CCD output can be displayed on a computer screen. The computer can also be used to analyze the RHEED pattern as well as to monit RHEED oscillations, which, in turn, can be used to control the shutters for precise growth of heterostructures. Having learned about the principle of RHEED and rods produced on the phosphor screen, we are now well motivated to learn how the image on the screen and the surface atoms correlate. If the surface atoms on a clean surface were to be situated in their bulk positions, the ensuing RHEED pattern rods, corresponding to (1,0), (0,0), and (1,0) diffraction rods, would be separated by the reciprocal of the separation between the atomic rows (or 2p per row separation depending on deflection of
j425
j 3 Growth and Growth Methods for Nitride Semiconductors
426
reciprocal space). These rods are called the bulk rods, and the pattern would change when the sample is rotated about its normal. For a bcc or fcc lattice on (1 0 0) like surfaces, owing to symmetry 90 rotation later, the original image is recovered. In RHEED terms, this is called the (1 · 1) surface construction, and it is important to specify the direction of the electron beam in relation to the crystallographic directions of the lattice. In an hexagonal close packed (hcp) lattice, the symmetry considerations are such that the pattern would repeat itself after every 60 , and this too would be termed (1 · 1) surface construction. The arrangement of surface atoms in their bulk positions in hcp symmetry can be gleaned from Figure 1.7. If surface atoms, whose surface coverage can be less than a monolayer, from beyond the bulk spacing would reconstruct, with additional RHEED rods appearing between the above-mentioned bulk diffraction rods. Owing to rotational (about the normal to the crystal surface) asymmetry in reconstruction, the number of additional lines would vary if the sample is rotated. A series of commonly seen surface reconstructions for (1 · 1), c(2 · 2), and p(2 · 2) on a (1 0 0) surface in fcc and bcc lattices is shown in Figure 3.53. For example, in the 2 · 2 reconstruction, an additional rod half way between the bulk rods in both azimuths, which are normal to each other, would appear. Symmetry of different surfaces is different. As an example, p(2 · 1) and c(2 · 2) surface reconstructions on fcc (1 1 0) plane are shown in Figure 3.54. The other commonly observed surface reconstructions on fcc (1 1 0) surfaces, namely, (2 · 4) and c(2 · 4), are shown in Figure 3.55. This surface is similar to ð1 1 2 0Þ surface in the hexagonal (hcp) system.
p( 1 x 1)
c(2 x 2)
p(2 x 2)
fcc (1 0 0) and bcc (1 0 0) Figure 3.53 Various (1 0 0) surface reconstructions corresponding to (1 · 1), c(2 · 2), and p(2 · 2) for fcc and bcc lattices. For convenience, the surface atoms participating in the said reconstruction are shown as smaller and shaded solid circles and those in bulk position are shown as open circles. The heavier lines indicate, one for each, the surface atoms participating in the surface reconstruction.
3.4 Nitride Growth Techniques
p(2 x 1)
c(2 x 2) fcc (1 1 0)
Figure 3.54 Various (110) surface reconstructions corresponding to p(2 · 1) and c(2 · 2) for an fcc lattice. For convenience, the surface atoms participating in the said reconstruction are shown as smaller and shaded solid circles and those in bulk position are shown as open circles. The heavier lines indicate, one for each, the surface atoms participating in the surface reconstruction.
(2 x 4)
c(2 x 4)
fcc (1 1 0) Figure 3.55 The (2 · 4) and c(2 · 4) on (1 1 0) like surfaces for an fcc lattice. For convenience, the surface atoms participating in the said reconstruction are shown as smaller and shaded solid circles and those in bulk position are shown as open circles. The heavier lines indicate, one for each, the surface atoms participating in the surface reconstruction.
j427
j 3 Growth and Growth Methods for Nitride Semiconductors
428
p(1 x 1)
p(2 x 2)
3x 3 r30º
fcc (1 1 1) , hcp (0 0 0 1) pffiffiffi pffiffiffi Figure 3.56 The (1 · 1), p(2 · 2), and 3 3 30 on fcc (1 1 1) and hcp (0 0 0 1) surfaces. For convenience the surface atoms participating in the said reconstruction are shown as smaller and shaded solid circles and those in bulk position are shown as open circles. The heavier lines indicate, one for each, the surface atoms participating in the surface reconstruction.
The atomic arrangement on the (1 1 1) fcc is similar to that on the (0 0 0 1) surface in the hexagonal (hcp) system. The (1 · 1) reconstruction, which is seen in Ga polarity GaN temperatures, and (2 · 2), which is seen during cooldown, and pffiffiffi at pffiffigrowth ffi 3 3 30 seen in AlN growth, discussed in detail in Section 3.5.14, are shown in Figure 3.56. A good deal of GaN growth is pursued on Si(1 1 1) substrates. The cleaning procedure for Si before epitaxy is discussed in Section 3.2.2.2. For this reason alone, the surface reconstruction of a clean Si(1 1 1) surface [274] is shown in Figure 3.57, which shows both an artistic view (a) and a scanning tunneling microscopy image (b). 3.4.2.7 Plasma-Assisted MBE (PAMBE) or RF MBE, Primarily N Source An MBE deposition system for nitrides consists essentially of a conventional MBE chamber, but with added equipment such as a compact RF source or a compact ECR source (which more or less gave way to RF relying on inductively coupled plasma (ICP)) and ammonia, as shown in Figure 3.58. These compact RF nitrogen sources are small enough to be mounted on one of the effusion cell port flanges. Some systems are equipped with both RF and ammonia sources such as the ones available in authors laboratory for maximum versatility. RF sources operate at several frequencies such as 13.56 MHz, paving the way for compact sources, which if desired produce mostly neutral N radicals with very little if any of the ionic nitrogen. The composition of the N plasma is important and its affect on layer quality and the required growth conditions are discussed later in this section. As alluded to earlier, the reactive nitrogen can also be supplied to the surface from an ammonia
3.4 Nitride Growth Techniques
Top view
B
A C
First layer adatoms Second layer Second layer, rest atoms Third layer, including dimers Fourth layer A
Side view B
C
(a)
Figure 3.57 (a) Atomic structure of the 7 · 7 surface reconstruction seen on clean Si(1 1 1) surfaces in the dimer adatom stacking (DAS) fault model. The unit cell contains two triangles with six adatoms per triangle. The stacking sequence in the left triangular subunit is faulted and that in the other triangular region is normal. This is by far the most complicated surface reconstruction and involves four layers. In the top view the atoms closer to the surface are indicated with the largest filled circles and so on. The atoms on the fourth layer are the bulk atoms and the rest are those of the reconstructed structure. The model contains dimers (depicted
with smaller open circles along the two triangles), a stacking fault layer (indicated as second layer atoms in the left triangular subunit), and adatoms on the top layer (depicted as the first layer adatoms). In the side view, atoms on the ð1 0 1Þd lattice plane along the long diagonal of the 7 7 unit cell are represented with larger circles than those behind them. The observation of this surface reconstruction is used as pivotal test whether the Si surface is clean and ready for epitaxy. Patterned after Ref. [274]. (b) A scanning tunneling miscopy image of a clean 7 7 surface reconstruction of Si(1 1 1). Courtesy of R.M. Feenstra.
j429
j 3 Growth and Growth Methods for Nitride Semiconductors
430
source [275,276]. Various ICP-activated RF sources, such as EPI (Veeco) Vacuum Products model Uni-Bulb, Oxford Applied Research models MPD21, CARS-25, and HD25, SVT Associates, Addon, Eiko Co., and Taisei Industry are used for PAMBE, details of which in terms of N species can be found in Ref. [83], but caution should be exercised in that the characteristic of these sources have changed continually with time owing to improvements. The PAMBE method possesses the basic nonequilibrium features of the molecular beam epitaxial growth technique and as such is well suited for growth of nitrides with large lattice mismatch with the substrate, large vapor pressure of N over the metal component, and so on. InN is one that provides the most challenge in this regard and PAMBE is reasonably well suited for its growth; refer to Section 3.5.15 for details. It should be briefly mentioned that highly In-rich conditions are used for InN growth, resulting in N-polarity. Owing to N2 activation by plasma discharge, the intensity of activated nitrogen flux does not depend on the growth temperature, TG, which is also described by Ts, which is the substrate temperature, allowing the operator to choose the optimum temperature independently in a wide range in regard to adatom surface diffusion and stoichiometry conditions required. It may also be that moderate levels of ionic N present in at least some of the plasma sources could be beneficial to growth at very low temperatures such as those used for InN. In a PAMBE system, the RF nitrogen flux through a plasma source is controlled best with a flow meter with typical flow rates giving rise to a system pressure of
RHEED gun Substrate
Liquid nitrogen shroud
Effusion cells
Optical spectrometer
RF in Cooling water in
Vent
Shutters
Mass flow controller
Phosphor screen
Manometer CCD camera
Purifier N2 in
Figure 3.58 Schematic diagram of an RF MBE system used for nitride growth at Virginia Commonwealth University.
Gate valve
3.4 Nitride Growth Techniques
anywhere between 2 · 106 and 1 · 104 Torr, the actual value being dependent on the particulars of the system in terms of pumping arrangement and system temperature. The RF plasma sources can be operated at power levels between 90 and 500 W and pressures of 10–100 Torr in the plasma chamber itself. Well-designed RF sources can have cracking efficiencies as high as 10%, though in most operating conditions, the efficiency is about 1%. In both ammonia and RF-activated growth, typical pressures are in the low 105 Torr regime, which can be construed as a nearly collision-free growth mode. In newer RF sources, the emission wavelength of the plasma concurs with the assertion that only the nonionic species with low energies, about 2 eV, are generated in large quantities. This energy is well below the 24 eV damage threshold energy predicted for GaN [277]. The same, however, does not necessarily hold for ECR sources in which the kinetic energy of the nitrogen ions generated can be sufficiently high and could result in surface atom displacement [278]. The plasma source in PAMBE activates chemically inert ground-state nitrogen molecules ðN2 ðX 1 Sgþ; v ¼ 0ÞÞ by converting them into a combination of atomic nitrogen and metastable nitrogen molecule and ionic nitrogen atom and molecule. Spherically, the activated species contain nitrogen atoms, N, (N ¼ 4 S; 2 P; 2 D) and vibrationally excited states (N2 ðvÞ ¼ N2 ðX 1 Sgþ ; v„0Þ), electronically excited states 00 (N 2 ¼ N 2 ðA3 Suþ ; vÞ and N2 ¼ N2 ða Sgþ Þ), ionized molecules (N2 þ ¼ N2 þ þ 2 þ ðX Su ; wÞ), ionized atoms (N ), and neutral atoms at different states, with ion energies up to tens of electron volts [279,280]. The metastable nitrogen molecules Q Q Q are also designated A3 Suþ ; B3 g ; a1 g ; C3 u [281–283]. Both N and N2 ðA3 Suþ Þ have potential energies considerably higher than the required Gibbs free energy for the GaN synthesis [283]. In short, the species in the activated nitrogen flux include nitrogen atoms (N), excited molecules (N2 þ ), and ionic species (N2 þ , N þ ) [284]. Among the N2 þ species, the N2 ðA3 Suþ Þ variety has sufficiently long lifetime to reach the substrate and is therefore considered the metastable molecular species participating in growth [285]. The cracking efficiency, the rate of transforming inert N2 into active nitrogen species, can be as high as 10% in well-designed RF nitrogen sources, though the efficiency is about 1% in most operating conditions. N2 is activated by plasma discharge, which depends on the RF plasma power and is not affected by the growth temperature Ts [286]. Nitrogen atoms have very long lifetimes in the plasma chamber, on the order of seconds, owing to the fact that the formation of a nitrogen molecule by the collision of two nitrogen atoms requires an additional interaction with a third body, such as the source walls [287]. This would lead to an enrichment of the plasma with nitrogen atoms until a steady state is reached which includes collision with the walls. Modern RF sources are indeed capable of producing reactive nitrogen-rich plasma with atomic nitrogen leading to very high GaN growth rates [288–290]. A comparison of different models of RF sources indicates that the increase in growth rates is proportional to the available amount of atomic nitrogen [288,289]. These results establish the importance of nitrogen atoms in addition to the N2 ðA3 Suþ Þ metastable
j431
j 3 Growth and Growth Methods for Nitride Semiconductors
432
molecules in the growth process [288]. The big picture involving the RF plasma sources indicates that the temperature of the walls of the RF discharge zone dramatically affects the ion content of the plasma, which decreases to a very low level at temperatures above room temperature [289]. The ion content is also reduced with reduction in the size of the holes in the PBN aperture [282,291], whereas their number can be increased to achieve a sufficient flux of reactive nitrogen species. Finally, the experimental results suggest that the input power and gas flow may be used to control the composition of the nitrogen plasma beam. The levels of atomic nitrogen increase with increase in the input power and decrease in the N2 gas flow [292,293]. However, a similar behavior reported for ionic species [285] is explained by enhanced recombination at high nitrogen pressures in the plasma, owing to an increased probability of collisions. Given the plethora of options for sources producing reactive N through plasma activation, a number of studies have been undertaken to determine the most suitable option for PAMBE. In general, ECR sources produce mainly excited molecular species, consisting of second positive series excited N2þ (metastable molecular nitrogen) and ionized N2þ (ionized nitrogen molecule) with little atomic nitrogen. On the contrary, the early versions of RF sources typically produced high densities of atomic nitrogen and excited molecular nitrogen of the first positive series [294], as confirmed by mass spectroscopy analysis [295]. Later varieties of RF sources could indeed produce mainly metastable molecular nitrogen, which has been reported to be the preferred species for growth, as discussed in Section 3.4.2.9.3. Optical emission spectra of RF sources are very useful in determining, to a good extent, the nitrogen species responsible for growth. Shown in Figure 3.59a and b are the emission spectra from two representative RF plasma sources operating at an input RF power of 450 W and system pressures of, depending on the system employed, 5 · 106 Torr (for Addon source) and 5 · 105 Torr (for Veeco-EPI). Relative to ECR sources, the RF source emission is primarily neutral and consists of peaks associated with molecular (first-positive and second-positive series) and atomic species. The first- and second-positive molecular nitrogen emissions occur near orange–red and UV wavelengths, respectively. The atomic nitrogen related emission, however, occurs near red (infrared (IR)) wavelengths of 745, 821, and 868 nm. One particular source, labeled as 1 in Figure 3.59, exhibits the strongest emission lines associated with atomic species as well as the first-positive molecular species. This source is capable of producing GaN growth rates of more than 1 mm h1 at substrate temperatures as high as 800 C. As mentioned previously, N atoms and excited N2 þ molecules are the species that primarily participate in the growth process. A detailed discussion of particulars of each of the available sources has been provided by Georgakilas et al. [83]. RF sources are generally void of optical emission transitions related to molecular ions. However, a N2þ transition at 391 nm was observed for low-pressure discharge. The ionized atomic N lines at approximately 776 nm (much suppressed in later varieties) and 854 nm were observed in a SVTA RF source, whereas a N þ line at 648.2 nm was observed in an Oxford HD25 RF source. Moreover, mass spectroscopy experiments showed that the atomic ion flux was typically two to three times larger than that of the molecular ions for an Oxford
3.4 Nitride Growth Techniques Atomic nitrogen
2000
750 nm
Intensity (counts)
Molecular N2
1600
Second positive molecular series
870 nm
First positive molecular series 590 nm
1200
357 nm
820 nm 580 nm
650 nm
800 400 0
-6
Addon: 450 W 6 x 10 Torr
200
400
600
(a)
800
1000
1200
1400
Wavelength (nm) Veeco RF power = 450 W
750 nm
Intensity (au)
870 nm
820 nm 650 nm 580 nm
200
(b)
400
600
800
1000
1200
Wavelength (nm)
Figure 3.59 (a) Optical spectra obtained from a nitrogen RF source manufactured by Addon using two difference detectors with overlapping spectral responsivities (190–860 nm for one and 640–1300 nm for the other), together covering the entire spectral range of interest. Note that molecular (both second and first positive series) and atomic nitrogen related transitions are seen (RF power: 450 W and the nitrogen flow is such that a pressure of 106 Torr is obtained).
(b) Optical spectra of a source manufactured by Veeco under similar conditions. The aperture plate is equipped with 200 holes, each about 250 mm in diameter. Note the absence of second positive molecular series and the higher intensity of the first positive molecular series being much smaller than that for atomic nitrogen. The intricacies of radicals of molecular nitrogen versus atomic nitrogen being desirable are discussed in Section 3.4.2.9.3.
CARS-25 source [298]. The reader should be cautioned that these sources continually evolve and the statements made here may or may not apply to some later varieties. By comparing the peak intensities of various species in the spectra, the relative composition of the radio frequency plasma nitrogen source can be obtained. Figures 3.60 and 3.61 show the relationships between the intensity of atomic nitrogen
j433
j 3 Growth and Growth Methods for Nitride Semiconductors
434
5000 745 peak
Intensity (au)
4000
868 peak
3000
821 peak
2000
1000
0
250
300
350
400
450
RF power (W)
Figure 3.60 Intensity of nitrogen RF spectrum peaks versus RF power. The system pressure is 9 · 106 Torr.
peaks with RF power and system pressure, respectively. These results suggest that the input power and nitrogen flow could be used to control the relative composition of the nitrogen plasma. Figure 3.60 shows that the intensity of atomic nitrogen increases linearly with input power. Figure 3.61 shows that the intensity of atomic nitrogen changes with the system pressure, which is controlled by nitrogen gas flow. As a further note, the intensity of atomic nitrogen increases with the nitrogen gas flow when the chamber background pressure is low. However, when the pressure is above 3.0 · 105 Torr, the intensity of atomic nitrogen peaks starts to decrease. A similar trend has also been reported for the ionic species. This behavior is explained by enhanced recombination at high nitrogen pressures in the plasma, owing to increased probability of collisions.
Intensity (au)
868 peak 745 peak
821 peak
10
20
30
40
50
–6
System pressure (10 Torr) Figure 3.61 Intensity of nitrogen RF spectrum peaks versus system pressure. The system pressure here should be treated as flow rate whose actual value depends on the system geometry and pumping speed. The RF power is 350 W.
60
3.4 Nitride Growth Techniques
Although practically not used anymore, early deposition of GaN in the realm of MBE was performed with the help of ECR sources, which are notorious for generating a good deal of ionic N species, which could cause damage. To determine the detrimental effects of the higher ion content of an ECR source on the GaN material properties, studies comparing ECR and RF-activated nitrogen sources have been undertaken [296]. In one such study, higher quality GaN layers were consistently grown with an Oxford MPD21 RF source at growth rates higher than those with an ASTeX ECR source [294]. The crystal damage in layers grown with the ECR source was reported to be comparable with that associated with ion implantation [297]. Optical admittance spectroscopy revealed distinct transitions in n-type GaN grown with an RF source as opposed to broad bands in the ECR grown samples, attributed to potential fluctuations caused by structural inhomogeneities induced by the ECR source [296]. Finally, yellow PL emission at about 2.2 eV appeared in the photoluminescence spectra of layers grown with an Irie ECR source, whereas it could be reduced in layers grown with an Eiko RF source [295]. All these GaN layers produced with ECR sources are consistent with point defects induced by plasma damage in the GaN layers [296]. Nitrogen atoms have five valence electrons and bond with all group III elements without any potential barrier [283,298,299]. The N2þ excited molecules interact with a solid-state surface through an exothermic dissociative chemisorption mechanism [300]. Therefore, N atoms and excited N 2þ molecules participate in III-N growth by RF-assisted PAMBE [281,298,301]. In contrast, ionic species of plasmaactivated nitrogen flux (N2þ , Nþ ) are responsible for ion-surface processes, such as adatom–vacancy pair generation, displacement of atoms, embedding of N2 molecules, and surface atom sputtering [302,303]. These are believed to have adverse effects on the epitaxial film quality, which in extreme cases could lead to even polycrystalline growth [304]. The damage threshold for GaN has been estimated at 18–24 eV [305]. However, the threshold ion density and doses of the energetic species for generating point defects and dislocations are not yet available. On the basis of the bond strength, one can surmise that the threshold ion energy would be larger for AlN and smaller for InN as compared to GaN. In spite of this, early attempts to grow GaN in vacuum environments utilized N2 and Nþ ions because of their availability [93]. In addition, NH3 þ ions with 30 eV energy and 3 1014 ions cm 2 s 1 current density were also utilized in ion-enhanced gas source MBE [306]. One should, however, be cautious as even a small flux (1012 cm 2 s 1) of relatively high-energy N2 ions (>20 eV) could cause surface damage. 3.4.2.8 Reactive Ion MBE Several laboratories have attempted MBE growth of the group III nitrides using nonplasma-based growth techniques. The most successful of these techniques utilized low-energy nitrogen ions from a Kaufmann ion source to grow GaN of a quality comparable with that grown with ECR sources. Several laboratories in the past have tried reactive ion molecular beam epitaxy in which N2 or NH3 was decomposed on the substrate surface [307,308]. However, decent growth rates and material quality were not obtained at MBE substrate temperatures. Two possible extensions of the
j435
j 3 Growth and Growth Methods for Nitride Semiconductors
436
original RMBE approach, which were thought to reduce the required substrate temperature, are the application of hydrazine and cracked NH3 as the source gas, which could be accomplished by a commercial hydride cracker. Hydrazine is believed to increase the reactivity of the nitrogen-containing species. Much effort on thermal cracking of ammonia turned out to be misguided for several reasons. Thermal cracking requires high temperatures with its expected complications that reduce this approach to a laboratory curiosity. In addition, the little atomic N produced recombines readily through collisions forming very inert and stable N2. Needless to say, this approach is not seriously pursued anymore. The use of hydrazine and the likes thereof raise safety concerns and is a source of contamination from the water vapor. Ammonia cracked only at the surface of the substrate is the successful approach and has been employed in various laboratories, as will be outlined later. High-quality layers grown on sapphire substrates by MBE permitted the fabrication of a variety of devices, such as Schottky diodes [309] and MODFETs [310,311], multiple GaN/AlGaN and InGaN/AlGaN quantum wells [312,313], and the observation of photo-pumped-stimulated emission at 300 K with a low pumping intensity [314]. In time, all MBE-grown lasers with ammonia N source on freestanding HVPE templates were prepared [315]. Growth by MBE differs greatly on the basis of whether layers are grown on a foreign substrate such as Si, sapphire, SiC, and ZnO, or on GaN, either in the form of a template or a bulk platelet. Growth on a foreign substrate generally requires high-temperature growth as low-temperature growths generally lead to good surface morphologies and large concentrations of edge dislocations propagating along the c-direction. Although substrate temperature measurements are not reliable, what is implied here is that temperature of 800 C and above would be high, whereas 650–750 C would be medium. Note that growth on sapphire and other foreign substrates may commence with high- or low-temperature nitridation and low-temperature buffer layers [316]. For growth on already existing GaN templates, the issues are very different in that the dislocation reduction measures and growth nucleation on a foreign substrate are not the main concerns. What becomes important then is the optimization of parameters such as group V/III ratio and growth temperature for optimizing properties such as mobility [317]. In this case, it is universally agreed that medium-temperature growth under appropriately Ga-rich conditions leads to best surface morphologies and mobilities [318]. 3.4.2.9 Principles of RMBE and PAMBE Growth The hallmark of the MBE approach, as applied to conventional group IV and group III–V semiconductors, is that it provides very reliable laboratory investigations of basic growth processes. The case of nitrides contradicts this long-standing status somewhat, in that little work has been published on modeling of group III-N growth. The lack of reliable data on surface migration rates, nucleation parameters, and sticking coefficients of group III and N atoms to different substrate surfaces, as well as other thermodynamic and kinetic parameters, makes quantitative estimates of growth parameters difficult. To overcome kinetic barriers in effect in the pursuit of epitaxy involving conventional compound semiconductors, the optimum substrate temperatures employed
3.4 Nitride Growth Techniques
fall in the range 1/2 to 2/3 of the melting temperature of the semiconductor (see Section 1.2.3 for details). In the case of III nitrides, however, synthesis occurs, particularly with MBE, under conditions in which the substrate temperature is much below the empirically acquired figure of 1/2–2/3 of the melting point, and thus, GaN growth is not attempted under thermodynamical equilibrium conditions [301]. MBE growth conditions employed for GaN epitaxy fall within an area of the phase diagram where the liquid Ga and molecular nitrogen gas have the lowest free energies [301]. As mentioned above, the Ga diffusion barrier makes the species mobile. The metastable growth process involves a balance between GaN formation (forward reaction such as those depicted by Equations 3.54–3.56) and GaN decomposition (reverse reaction such as those depicted by Equations 3.15–3.17) [301]. In order for net growth to take place, the rate of GaN formation must be larger than the rate of its decomposition, which is the case under the growth conditions employed, as the kinetic barrier for decomposition is large owing to the strong metal–N bonds at the surface (see Table 3.12). Practical limitations on the maximum reactive nitrogen flux attainable are such that the substrate temperature of GaN during MBE growth is about 20–35% (600–900 C) of GaN melting temperature (TM 2500 C, see Section 1.2.3 for details). Sections 3.4.2.7 and Sections 3.4.2.8 discuss the first-order guiding principles for reactive MBE growth with ammonia as the nitrogen source and PAMBE with plasma-activated nitrogen as the source. 3.4.2.9.1 Growth by RMBE Ammonia was initially thought not to be compatible with MBE systems as it reacts with metals such as copper used in the flange and valve seals, which is a simple problem and has been dealt with effectively. The other concern is that hydrogen released from ammonia makes hot metal components fragile/brittle. This has also been dealt with effectively as the amount of ammonia used is small and proper shielding methods can be employed to mitigate the potential damage. Another apprehension about NH3 is its reactivity with oxygen. This does not seem to be a problem as the vacuum systems are UHV type and the oxygen background is negligible. The technique utilizing ammonia as a nitrogen source has been termed the reactive MBE (RMBE). Basically, the RF nitrogen plasma source shown in Figure 3.58 gets replaced with an ammonia injector maintained at sufficiently high temperature to avoid ammonia condensation as the shroud walls are cold. In RMBE, ammonia is decomposed only on the surface of the substrate by pyrolysis. As such, the lower end of the temperature range where this technique can be utilized is about 600 C, albeit with very low growth rates. Temperatures somewhat below this can be used, but the ammonia flow rate must be increased by some 10-fold as done in growing initiation layers on sapphire substrates. Growth rates of about 1–3 mm h1 can be obtained at substrate temperatures between 750 and 850 C while maintaining a high quality. This growth temperature is much lower than the growth temperature in OMVPE. Kim et al. [319] showed that RMBE can produce films of a very good quality. Although applicable to all substrates, this particular study was performed on sapphire substrates. The actual growth is preceded with a nitridation and a thin AlN buffer layer. In this approach, the substrate temperatures varied between 610 and 820 C. Growth
j437
j 3 Growth and Growth Methods for Nitride Semiconductors
438
Cracking efficiency of NH3 (%)
5 4 3 2 1 0 200
300
400
500
600
700
800
900
Temperature (ºC) Figure 3.62 Cracking efficiency of ammonia as a function of substrate temperature [320].
rates up to 2.9 mm h1 were obtained for temperatures between 725 and 820 C. These growth rates are quite comparable with those produced by conventional OMVPE. Studies of the growth kinetics have been carried out by employing various V/III ratios and substrate temperatures. It was found that the layer quality increases and compressive stress decreases as the N/Ga ratio increases. It should be pointed out that incorporation of residual impurities as well as the creation of a high density of native defects is made less efficient at higher N/Ga ratios. The growth mechanism of GaN in RMBE involves thermal cracking of ammonia and surface reaction of N with Ga atoms to form GaN. Monatomic nitrogen becomes available on the film surface by pyrolysis of the ammonia in a temperature-dependent process. The efficiency of ammonia dissociation (shown in Figure 3.62) has been measured directly by using a mass spectrometer while varying the substrate temperature [320]. The growth rate as a function of ammonia flow rate was also measured at a substrate temperature of 830 C, as will be discussed shortly. The efficiency of ammonia dissociation increases slowly above 600 C being only about 4%. Not all of nitrogen of course is incorporated into the film, and above 700 C, GaN decomposition becomes a consideration contributing to the nitrogen radical flux. In light of this fact, the growth rate versus ammonia flow becomes a more important parameter. Group III–V limited growth regimes can be found from the growth rate parameter as shown above [320]. The growth rate of GaN was greatest at a nozzle temperature of 300 C and then decreased at a nozzle temperature of 400 C for a 100% ammonia beam. Seeded and anti-seeded beams did not attain as high growth rates. A nitrogen incorporation efficiency of 12% was derived for the supersonic beams [326]. Nitrogen incorporation efficiency reported by others include 16% [321], 10% [322], and 4% [320], as shown in Figure 3.62. The nitrogen incorporation efficiency for low-energy ammonia from a leak valve was derived as 26%, suggesting that dissociative chemisorption of ammonia on GaN(0 0 0 1) occurs via a precursor-mediated pathway. Because the
3.4 Nitride Growth Techniques
ammonia sticking coefficient does not increase with increasing kinetic energy, it is concluded that ammonia dissociative chemisorption is inactivated [326]. Substrate temperature and incident Ga flux affect the amount of Ga available on the surface for incorporation into the film. For constant ammonia pressure, there is a substrate temperature below which Ga will condense on the surface and above which all the incident Ga will desorb. This temperature is clearly defined and has been measured by RHEED and desorption mass spectroscopy (DMS). If a surface is exposed to ammonia in the absence of Ga and then exposed to Ga in the absence of incident ammonia, Ga will react to form GaN, and hydrogen will be desorbed. If the Ga flux is continued after all nitrogen is consumed, Ga coverage will occur. Below the condensation temperature, Ga multilayers and/or Ga droplets will form. Above the condensation temperature, less than a monolayer of weakly bound Ga (sitting on preferential sites of the underlying GaN lattice) will accumulate until the Ga flux is stopped, at which time it will desorb [323]. Adsorption of ammonia and Ga behaves differently for N- and Ga-face polarity films. N-face polarity has two types of adsorption sites: strongly bonded Ga on a nitrided surface and weakly bonded Ga on a gallided surface. Ga-face polarity only adsorbs weakly bonded Ga [324]. Ga-polarity films have been observed to give both (1 · 1) and (2 · 2) reconstructions by RHEED depending on the Ga-to-ammonia flux ratio where the transition between them is activated at 2.76 eV. It is concluded that growth on Ga-face polarity occurs by step flow because Ga adsorption is only detected when there is a coincident ammonia flux. Furthermore, no RHEED intensity oscillations could be observed for growth on Ga-face polarity [324]. Exposure of a nitrided N-face polarity film to less than a monolayer of Ga flux results in featureless steps. When this surface is exposed to ammonia again, irregularly shaped islands form [324]. The growth rate can be limited by weakly adsorbed Ga in the regime with excess Ga on the surface. A Ga-terminated surface with featureless terraces can be achieved by quenching after the Ga shutter is closed. Held et al. [325] have concluded that Ga accumulation reduces the growth rate in the ammonia-limited regime by blocking ammonia chemisorption sites. Additionally, GaN surface morphology is determined by the growth regime. Ga-limited growth displays rough highly faceted surfaces with hillocks as will be discussed shortly. On the contrary, ammonia-limited growth produces smoother, nonfaceted films. It is suggested that surface roughness under Ga-limited growth is because of a Ga adatom diffusion length of only a few atomic diameters. Stochastic roughening, originating from statistical fluctuations in the Ga arrival flux, will dominate under these conditions. In ammonia-limited conditions, the V/III ratio affects the surface morphology. As the V/III ratio is decreased, the surface roughness decreases eventually forming pits and then Ga droplets [326]. To investigate the growth kinetics [319], the substrate temperatures were varied in the range of 700–820 C with an ammonia flow rate of 45 sccm while keeping the Ga flux at about 3 · 1015 cm2 s1. These early investigations of growth rate and its relationship with ammonia flow were followed by additional experiments [320], all of which clearly indicate that growth rate for constant Ga flux at a given temperature increases initially with ammonia flow and saturates, as displayed in Figure 3.63. Although the scatter in the data is quite substantial, a weak trend of increase in the
j439
j 3 Growth and Growth Methods for Nitride Semiconductors
440
1.4 NH3-limited Ga-limited
Growth rate (µL–1)
1.2 1.0 0.8 0.6 0.4
T growth = 830ºC 0
10
20
30
40
50
NH3 flux (sccm) Figure 3.63 Variation of the growth rate with the ammonia flow at a substrate temperature of 830 C. The growth rate increases with ammonia flow rate initially and saturates, indicating of ammonia limited growth rate for lower ammonia flow rates and Ga limited growth rate for higher ammonia flow rates. After Ref. [320].
growth rate with increased ammonia flow rate before reaching saturation is apparent. This is similar to the case of plasma-assisted MBE as discussed in 3.5.b.3.ii and shown in Figure 3.69. It can be argued that, at low ammonia flow rates, the Ga desorption is responsible for the reduced growth rate, implying ammonia flow limited growth regime, which represents the Ga-rich growth conditions. It should be mentioned that these growth temperatures are sufficiently high for efficient Ga desorption unless Ga is adsorbed through bonding with N. At high ammonia flow rates, the growth rate is limited by the Ga arrival rate, which is responsible for the apparent saturation. This regime is the Ga-limited regime or N-rich growth conditions. At much lower growth temperatures, for example, 700 C, not shown, the growth rate is much smaller because of the lower ammonia cracking efficiency, supporting the Ga-desorption argument. A GaN equivalent flux can be calculated by dividing the product of the GaN density, growth rate, and Avogadros number by the molecular weight of GaN. Incident ammonia flux is the ammonia throughput divided by the area of the ammonia beam at the sample position. The nitrogen incorporation efficiency can be calculated by dividing the equivalent GaN flux by the incident ammonia flux [326]. The effective nitrogen flux is determined by the incident ammonia flux and the sticking coefficient for dissociative chemisorption (aNH3 ). The Ga sticking coefficient is very close to unity, so the effective Ga flux is approximately equal to the incident flux [326]. Ammonia incident kinetic energy affects the GaN surface morphology indirectly via changes in the V/III flux ratio. This has been studied by using ammonia-seeded supersonic molecular beams as a monoenergetic source of ammonia and group III precursors for MBE growth. The seeded molecular beam allows the incident kinetic energy of the precursors to be precisely controlled using a heated nozzle. The growth
3.4 Nitride Growth Techniques
rate should be enhanced if ammonia dissociative chemisorption is direct and activated. If an activation barrier to chemisorption exists, then the substrate temperature can be lowered as incident kinetic energy of the precursor is used to overcome the barrier. However, if ammonia chemisorption is inactivated and occurs via a precursor-mediated channel, the growth rate is expected to decrease with increasing ammonia kinetic energy [326]. Based on the assumption that no desorption takes place from chemisorbed states, a growth kinetics study of GaN grown by reactive ion molecular beam epitaxy (RIMBE), which is also applicable to RMBE with some modifications, was carried out by Powell et al. [93]. In this model, competitive contributions (associated with site selection) of collision-induced dissociation followed by chemisorption of nitrogen and ion-stimulated Ga desorption were considered to be the primary mechanisms that affect the growth. The model was successful in demonstrating that, at a constant substrate temperature and a constant nitrogen-to-Ga flux ratio (J N2 þ J Ga ), the growth rate decreases with increasing ion energy. However, the model could not account for the experimentally observed drop in the growth rate at high substrate temperatures. For an RMBE study, the same model can be used with some modifications in parameters and activation energies for the surface-state transitions. The equations describing the rate of change of the number of physisorbed and chemisorbed Ga atoms on the surface may be given as 0
dnGa =dt ¼ J Ga ðnGa =tÞ knGa ð1 qÞ þ k N s q;
ð3:47Þ
N s ðdq=dtÞ ¼ knGa ð1 qÞ k0 N s q f q J NH3 ;
ð3:48Þ
where k is the forward reaction rate constant from the physisorbed state to the chemisorbed state, k0 is the backward reaction rate constant from the chemisorbed state to the physisorbed state, t (describes Ns) is the average residence time of free Ga prior to thermal desorption, J NH3 is the ammonia flux, and f is the fraction of ammonia that is thermally cracked and reacts with Ga to form GaN on the growing surface. To correctly describe the nature of growth, the last term on the right-hand side of Equation 3.48 is modified to f qJ NH3 for RMBE from 2 f q J NH3 , originally formulated for RIMBE. The multiplication factor of 2 was reduced to 1 because each ammonia molecule in RMBE can provide only one nitrogen atom. The parameters used in Equations 3.47 and 3.48 are defined as k ¼ n0 expð E p =kB TÞ;
ð3:49Þ
1=t ¼ n0 expð E d =kB TÞ;
ð3:50Þ
0
k ¼ n0 expð E c =kB TÞ;
ð3:51Þ
f ¼ expð E p =kcr TÞ:
ð3:52Þ
All parameters other than those defined by Equations 3.49–3.52 are the same as those used in Powells model. The solution of Equations 3.47 and 3.48 (at steady state with the same values for the various parameters, as given in Powells model) predicts
j441
j 3 Growth and Growth Methods for Nitride Semiconductors
442
a decreasing growth rate as the substrate temperature increases from 650 to 850 C for a constant Ga flux and a constant V/III ratio. Another calculation for a constant Ga flux (3 · 1015 cm2 s1) and a constant substrate temperature (800 C) reveals an increasing growth rate as the V/III ratio becomes larger. These results are exhibited in Figure 3.64a and b. Jones et al. [327] and Evans et al. [328] reported on an empirical value for Ga desorption in the growth of GaN with gas-source MBE (GSMBE), which essentially follows the same growth scheme as used in the present experiment. The desorption 0.5 (a)
Growth rate (µm h–1)
0.4
0.3
0.2
JNH3 /JGa = 4.3
0.1
15
JGa = 3.096 x 10
0 650
700
–2 –1
cm s
750
800
850
Substrate temperature (ºC) 1.4 (b)
Growth rate (µm h–1)
1.2 1 0.8 0.6 0.4 T s = 800 ºC 15 –2 –1 JGa = 3.096 x 10 cm s
0.2 0
1
10
100 V/III ratio (JNH3/JGa)
1000
Figure 3.64 Calculated growth rate from the simple precursormediated chemisorption model. (a) Growth rate change with substrate temperature (V/III ¼ 4.3, JGa ¼ 3.096 · 1015 cm2 sl) (b). Growth rate change with V/III ratio (Ts ¼ 800 C, JGa ¼ 3.096 · 1015 cm2 s1). The solid line indicates the calculations whereas the circle represents the measured data.
10 4
3.4 Nitride Growth Techniques
energy for Ga atoms obtained by temperature-programmed desorption (TPD) was published to be 1.9 0.2 eV. In addition, Burns et al. [329] measured the thermal desorption energy of Ga on Al2O3 by means of a thermal desorption spectroscopy (TDS) technique and deduced a value of 2.05 eV. Between these two measured values, the 1.9 eV, according to Evans et al. [319] is deemed more applicable because of similarities in the growth schemes employed in the RMBE investigations. The growth rate has been calculated with 1.0 and 0.25 eV, respectively, for the parameters Ec and Ecr . The activation energy for the transition from the physisorbed state to the chemisorbed state was chosen to be zero because the sticking coefficient of Ga is generally considered to be 1. In Figure 3.64, the calculated growth rate data for 12 and 25 sccm of ammonia flow are compared with experiments. Even though there are differences in magnitude between the experimental data and the calculated values for the 12 sccm ammonia case, the calculated data follow the same qualitative trend of growth rate versus substrate temperature as the experimental data. More specifically, the calculated results also present growth rates increasing with temperature because of more efficient thermal cracking of ammonia in the temperature range where Ga desorption is not the limiting factor. Considering that the effective amount of ammonia molecules is proportional to the ammonia flux and assuming that the cracking rate depends only on temperature, it is quite reasonable to expect that the growth rate of the sample grown with the lowest ammonia flux (4 sccm) would decrease rather drastically with decreasing substrate temperature. That this is the case becomes apparent from Figure 3.64 for ammonia flow rates below 12 sccm. If the substrate temperature was not sufficiently high to produce as much or more reactive ammonia than that necessitated by the given surface concentration of Ga, the growth rate would not be determined by the ammonia flow rate. Instead, it would be determined by the Ga population on the surface and would remain constant with changes in ammonia flow rate for a given temperature, unless there is a change in Ga flux. Additional experiments carried out at 800 C point to the growth rate remaining more or less constant, independent of the ammonia flux. However, one can conclude that the growth rate is determined by the ammonia flow rate if the substrate temperature is not sufficiently high to crack substantial amounts of the ammonia supplied. In the present case, this is a temperature below 750 C. It is to be noted that Ga desorption from the surface becomes the limiting factor at higher substrate temperatures. Consequently, the growth rate decreases despite more efficient ammonia cracking. For example, a significant drop in growth rate was found at 820 C for the sample grown with 25 sccm ammonia (Figure 3.64). Furthermore, in this growth regime, the growth rate is constant at a specific temperature, regardless of the ammonia flux unless it is lower than the Ga flux. The trend in growth rates was observed using several different flow rates of ammonia. When the substrate temperature was increased to 850 C, the growth rate practically became nil. The incorporation kinetics of gallium during GSMBE of GaN using elemental Ga and NH3 gas as source materials was studied by Jenny et al. [330]. Desorption mass spectroscopy (DMS) was applied to perform in situ quantitative measurements of GaN formation, Ga desorption, and Ga surface accumulation during growth. The
j443
j 3 Growth and Growth Methods for Nitride Semiconductors
444
110 100
s (Ga)
NH3 (Ji) (x10–7 Torr BEP)
90 0.9
80
GaN Growth
0.6
70
0.3
60 50
GaN growth + Ga desorption
40 30
0.1
20
GaN growth + Ga desorption + Ga accumulation
10 0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
–1
J i (Ga) (ML s ) Figure 3.65 Plot of the Ga incorporation ratio, s(Ga), contours as a function of Ga and NH3 fluxes. Labeled in the figure are the three primary regimes for Ga adatoms. The area above the heavy line represents the GaN growth regime. The dashed line delineates the regimes of GaN growth þ Ga desorption and GaN growth þ desorption þ accumulation. Reference [330].
rate of formation of GaN was reported as a function of incident Ga flux (0.1–0.75 ML s1) and incident NH3 flux (1 · 107 to 3 · 105 Torr beam equivalent pressure (BEP)) at a growth temperature of 725 C. Three distinct growth regimes (Figure 3.65) were observed: (1) GaN is formed where all of the incident Ga flux is consumed; (2) part of the incident Ga atoms are consumed to form GaN, whereas the excess is desorbed by the surface; and (3) GaN formation coexists with desorption and surface accumulation of Ga. It was generally found that the Ga surface accumulation is inhibited by increasing the rate of incidence of NH3 and/or by decreasing the rate of incidence of Ga at this temperature. In addition, the order of the reaction between Ga and NH3 is determined to be unity, and this supports the validity of the Ga þ NH3 ! GaN þ ð3=2ÞH2 reaction. If the cracked-ammonia flux on the growing surface is insufficient as compared to the Ga flux, the surplus Ga will thermally desorb as substrate temperatures are high, in the vicinity of 800 C. Consequently, those Ga atoms that do not participate in the formation of GaN or desorb tend to migrate to surface sites with minimum surface energy. This could give rise to Ga clusters on the surface that appear to serve as preferred nucleation sites. It is then reasonable to argue that the larger population of Ga atoms near the cluster sites causes a locally accelerated growth. The immediate result of this is the appearance of huge hexagonal hillocks on the film surface. The proposed process is depicted schematically in Figure 3.66. The conclusion that can be drawn is that if the Ga-to-ammonia flux ratio is high, hillocks are likely to form on
3.4 Nitride Growth Techniques Large Ga flux Small NH3 flux
Growing surface
Clustering of surface Ga atoms Gathering of reactive ammonia by the concentration gradient of Ga on the surface
Hexagonal hillock formation
Appearance of highly strained area on the top of hillocks due to locally enhanced growth rate
Figure 3.66 A proposed pathway to hillock formation in the growth of GaN.
the surface. These features are confirmed with scanning electron images of GaN surfaces. A noticeable feature of hexagonal hillocks is that their tops appear to be composed of many overlapping triangles and etch pits. At higher temperatures and ammonia flow rates, the tips of the hillocks are etched giving rise to flatter surfaces. Growth on OMVPE templates with pits from screw dislocations has been shown to produce spiral hillocks, see Figure 4.11. The step edges on the hillocks have a terrace width that decreases in proportion to the supersaturation of ammonia. Terrace width was found to be independent of the Ga flux. It was found that depending on the growth rate, hillock growth can be minimized by using low ammonia supersaturation independent of the Ga flux or the desorbing H2 [331]. One growth model proposed suggests that hydrogen bonded with nitrogen adatoms suppresses their desorption from the GaN(0 0 0 1) surface. This model also shows that blocking of ammonia adsorption sites by Ga adatoms and ammonia radicals leads to unusual behavior of the growth rate as a function of substrate temperature and incident Ga flux for Garich conditions. Under this Ga-rich condition at a substrate temperature above 800 C, the surface is nearly free of adsorbed species. Under N-rich conditions at a low substrate temperature range, there was still the growth of GaN and the surface was found to be covered predominantly with ammonia radicals up to 1000 C [332]. RHEED was used to show the dependence of the substrate temperature on the growth mode. Under excess ammonia conditions giving a V/III ratio between 3 and 5, the RHEED oscillations were measured at substrate temperatures of 500, 600, 700, and 800 C. Contrary to some other experiments, well-defined oscillations were obtained at 500 and 600 C, but these were damped out at 700 C and no oscillations
j445
j 3 Growth and Growth Methods for Nitride Semiconductors
446
were obtained at 800 C. This shows that two-dimensional layer-by-layer growth can be obtained at low temperatures, but it changes to step-flow growth mode at higher temperatures [333]. Consistent with the notion that tips of facets can be etched for a smoother surface when high ammonia flow rates are used, Grandjean et al. [334] reported that the smoothest surface in their results had an RMS roughness value of 3.6 nm on a 5 · 5 mm2 AFM scan. This film was grown with an ammonia flow of 50 sccm corresponding to a BEP of 6.4 · 106 Torr. As the ammonia flow decreased, the film roughness increased. The film with the ammonia flow of 50 sccm also gave the highest PL intensity and the lowest carrier concentration. It was shown by SIMS that the oxygen and silicon concentrations might be the source of high background carrier concentrations because the concentration of these elements decreased with an increase in the ammonia flow of up to 50 sccm. Substrate temperatures can be 80–90 C higher when using ammonia as the nitrogen source rather than atomic nitrogen from plasma. Under N-rich conditions, blocking of nitrogen adsorption sites by ammonia radicals becomes essential and results in additional thermal stability of the GaN. Crystal growth carried out at a higher temperature is expected to enhance the overall properties of GaN [332]. This has been exhibited experimentally by Webb et al. [335] where optimal electron mobility was found at 900 C. The optimal substrate temperature for Ga incorporation was found to be 820 C at large ammonia overpressure owing to the competition between thermally activated pyrolysis of ammonia and re-evaporation of Ga from the surface [336]. GaN films grown by ammonia MBE on GaN buffer layers on sapphire were found to depend mostly on the temperature of the buffer. The optimal buffer growth temperature was 500 C for a 25 nm thick buffer. An annealing step of 900 C for 10 min was performed before growing 2–4 mm thick GaN films at 830 C. These conditions gave optimal PL and X-ray linewidths [337]. High-mobility GaN was achieved using substrate temperatures from 860 to 920 C and an ammonia flow rate of 50 sscm. These films were grown on an AlN buffer layer. The optimum mobility was found at a substrate temperature of 900 C and film thickness of 2 mm. It is attributed to increased grain size and defect reduction. Carbon doping was also employed to reduce the background carrier concentration to approximately 1.5 · 1017 cm3 [335]. 3.4.2.9.2 Growth by PAMBE As mentioned in Section 3.4.1.2 in relation to OMVPE and Section 3.4.2 in relation to MBE, the growth process is a balance between incorporation and decomposition processes with conditions chosen so that the adsorption rate is greater than the desorption rate. It is a general notion that without the desorption process, if the growth temperature, for example, is very low, the quality of the resultant layers is not good. Here, this dynamic and competing process will be discussed in the framework of PAMBE. It should, however, be clear that the same also applies to RMBE. The evaporation of GaN has been studied by heating in vacuum. Guha et al. [250] reported a decomposition rate of three to four ML per minute at 830 C, whereas Grandjean et al. [222] utilized reverse RHEED oscillations as a measure of GaN evaporation to show that the decomposition rate is nearly zero below
3.4 Nitride Growth Techniques
Ga incorporation ratio
1.0 0.8 0.6 0.4 0.2 0.0 8.8
9.6
9.2
10 4/T (K –1)
10.0
10.4
Figure 3.67 Measured incorporation ratio behavior for Ga during GaN growth as a function of reciprocal growth temperature. The Ga arrival rate was 0.6 ML s1 and the active nitrogen flux was 0.8 ML s1 (Figure 3.2 from Ref. [250], reprinted with permission from Guha et al., Appl. Phys. Lett. 69, 2879 (1996). Copyright 1996, American Institute of Physics).
750 C, increases rapidly above 800 C, and reaches 1 mm h1 at 850 C. On the surface, this may imply that the GaN decomposition is not an important consideration when the substrate temperature is below 750 C. However, GaN evaporation, meaning breaking already formed GaN bonds, and desorption and adsorption of impingent Ga and various N species are two different entities. As alluded to in Section 3.4.2.1, Guha et al. [250] investigated the lifetime of Ga adatoms on the GaN(0 0 0 1) surface with no impingent active nitrogen flux. A mass spectrometer mounted in direct line of sight to the wafer was used to monitor the desorbed Ga (mass 69) signal after each pulse of Ga. The resultant Ga lifetimes were in the range of 0.6–5 ns in the 685–750 C range and the activation energy for Ga desorption was E 2.2 0.2 eV, which is similar to the value of 2.5 eV for Ga desorption from a GaAs(1 1 1) surface [338] (also see Table 3.12). Incorporation ratio of Ga into the growing epitaxial GaN was also studied [250] by monitoring the reflected Ga signal detected by the mass spectrometer, the results of which are shown in Figure 3.67. The incorporation ratio was defined as 0
I½R0 RðTÞ R0 ;
ð3:53Þ
where R(T) is the reflected Ga signal at temperature T and R0 represents the total impingent Ga flux measured by observing the reflected signal at a high enough temperature where all of the incident Ga was desorbed. In Section 3.4.2, we discussed the general principles of adsorption, desorption, diffusion, and decomposition that are applicable to any growth, particularly growth in vacuum such as MBE. The implications of these processes as they apply to GaN by MBE are discussed here. The data presented in Figure 3.67 show that in the presence of active nitrogen flux, Ga adatoms have a finite surface lifetime during which they can diffuse on the surface through either incorporation or desorption [250]. It should be mentioned that this description is not applicable to growth conditions where two
j447
j 3 Growth and Growth Methods for Nitride Semiconductors
448
monolayers of Ga are present on the surface, in which case the Ga surface diffusion barrier is low and N surface diffusion barrier is high. Naturally for any growth to occur, the activation energy for desorption must be greater than the activation energy for surface diffusion, as discussed in Section 3.4.2.3. With increasing growth temperature, the residence time of Ga adatoms on the surface decreases. Consequently, the probability of encountering nitrogen adatoms for incorporation decreases, causing a drop in the incorporation ratio of Ga. At sufficiently low temperatures, the residence time of Ga adatoms is long enough for almost all of them to encounter nitrogen adatoms and incorporate into the epitaxial film. Although this results in a near unity incorporation ratio, the layer quality will not be as high as the selection process, which would provide a threshold for bonding, and impurity and defect incorporation is reduced. In addition to the work cited previously [250], investigation of the Ga desorption during growth of GaN has been reported by other groups [339–341]. Of particular relevance is the study by Hacke et al. [339] who studied the appearance of the 2 · 2 reconstruction on a GaN(0 0 0 1) surface with varying Ga flux and substrate temperature. Transition from 2 · 1 to 1 · 1 surface reconstruction was observed when Ga adatom density on the surface was increased. An increase in the substrate temperature required an exponential increase in the Ga flux to maintain the surface in the transitional region. This implies that the residence time of Ga adatoms on the surface is reduced; in general agreement with the data of Figure 3.67. As-determined activation energies for Ga desorption were between 2.22 and 3.25 eV with a weighted mean of 2.76 eV, which is similar to that of evaporation of metallic Ga. Although this results in a near unity incorporation ratio, the layer quality will not be as high as in the case for the selection process, which would provide a threshold for bonding and be accompanied by reduced impurity and defect incorporation. GaN growth by MBE can be conducted under Ga-limited (N-rich) conditions in which the growth rate is determined by the ratio of Ga arrival and desorption rates and under N-limited (Ga-rich) conditions in which the growth rate is determined by the ratio of arrival and desorption of reactive nitrogen on the surface at a given temperature. If one is far from the transition region, the growth rate can be made nearly independent of substrate temperature in a wide range, as has been amply discussed [342,343]. It follows then that the growth rate increases monotonically with increasing flux of reactive nitrogen species (JN) under N-limited growth conditions. In this growth regime for sufficiently low substrate temperatures (where Ga and N desorption is negligible), the fluxes of Ga and N satisfy JGa/JN > 1. Unless the excess Ga is thermally desorbed from the surface (GaN does not decompose) by employing a proper growth temperature, Ga droplets are formed on the surface, as shown in Figure 3.68. In the N-stable regime, the GaN growth rate is limited by the Ga flux, JGa, and naturally the growth rate increases monotonically with the Ga flux. This regime represents the N-rich growth conditions and occurs when JGa/JN < 1 for a given substrate temperature. This regime generally leads to surfaces with pits that gather the extended defects leading to relatively rougher surfaces as compared to that in Garich conditions. The details of the nature of defects and optical properties of each type
3.4 Nitride Growth Techniques
Figure 3.68 SEM micrograph showing the formation of Ga droplets with a density of 5 · 104 cm2 on the surface of a GaN film that was grown under Ga-rich growth conditions.
are discussed primarily in Volume 1, Chapter 4 and Volume 2, Chapter 5, the latter in terms of the optical manifestation of defects. In an effort to gain insight into the kinetics of growth, Myoung et al. [340] have applied a precursor-mediated model with underlying physical processes remaining essentially the same as those by Guha et al. [250]. In this case, the Ga atoms arriving at the surface are assumed to be adsorbed into a mobile and weakly bound precursory state. These precursory Ga adatoms migrate on the surface, some being chemisorbed and some thermally desorbed. When the Ga atoms encounter atomic nitrogen in a chemisorbed state, the forward reaction occurs and GaN forms. Desorption from the chemisorbed state and decomposition of GaN were neglected, as the plasma sources used species lacking high kinetic energy, leading to the growth kinetics being determined by the JGa/JN ratio. Proposed model for physisorption of molecular N. A theoretical model for the MBE growth that accounts for a physisorption precursor of molecular nitrogen was proposed for the analysis of group III nitrides by Averyanova et al. [344]. The kinetics of nitrogen evaporation were found to be an essential factor influencing the MBE growth process. The high thermal stability of nitrides was explained to be related to the desorption kinetics, resulting in a low value of the evaporation coefficient. Evaporation coefficients as functions of temperature were extracted from the experimental Langmuir evaporation data of GaN and AlN. The process parameter dependent growth rate and the transition to extra liquid-phase formation during the GaN MBE were calculated. To gain insight into the critical role of the III/V flux ratio, Zywietz et al. [256] studied the diffusion of Ga and N atoms on (0 0 0 1) and GaNð0 0 0 1Þ surfaces using totalenergy density-functional theory, which led to Ga adatoms having a diffusion barrier approximately four times lower than that of N adatoms on Ga-terminated (0 0 0 1) and ð0 0 0 1Þ surfaces. This is because the adsorbed Ga atoms interact with the substrate by delocalized metallic Ga–Ga bonds, which are weak because Ga melts at 27 C, and
j449
j 3 Growth and Growth Methods for Nitride Semiconductors
450
the adatoms assimilate the behavior of a liquid film on the surface. On the contrary, although the N adatoms could readily evaporate as N2 molecules, they can be kinetically stabilized at the surface owing to their low surface mobility because migration is required for N2 formation. Consequently, the surface could be extensively coated with excess N adatoms under N-rich growth conditions causing significant reduction in the mobility of Ga atoms because strong Ga–N bonds would have to be broken for Ga migration. Zywietz et al. [256] stated that the diffusion barriers so determined for ideal surfaces should be considered as lower limits of the effective diffusion barriers on real surfaces with defects, impurities, and steps. As expected, the III/V flux ratio and substrate temperature play an important role in determining the surface adlayer and hence the adatom mobility on the growth surface [256]. The Ga-rich growth conditions result in excess N, albeit a small amount, which is not yet incorporated on the surface and the Ga adatoms are abundant and highly mobile at the substrate temperatures employed. This results in a 2D growth by a step-flow mode and leads to planarization of the surface and a reduction in stacking faults and point defects when the films are grown on GaN templates. Abundance of Ga on the surface also leads to efficient incorporation of N adatoms because the probability of Ga atoms capturing and bonding to N atoms is much higher than N atoms forming molecules and desorbing from the surface [256]. N-rich growth conditions lead to surfaces with an abundance of N and a significantly shorter Ga diffusion length. Reduction of the diffusion length to less than the mean distance between the available bonding sites would lead to roughening of the surface. Moreover, adatoms may be trapped at the wrong sites, which would favor the nucleation of stacking faults. It can then be argued that slightly Ga-rich conditions are expected to be optimal for the growth of GaN. The expected influence of surface diffusion properties of Ga and N on the surface morphology of GaN grown under Gaand N-rich conditions is discussed in Section 3.4.2.9.4. The discussions above indicate that for a given Ga flux, the growth rate increases monotonically with increasing N flux until the growth rate is limited by the Ga flux and saturates, as depicted in Figure 3.69 [345] in which the GaN growth rate is plotted as a function of Ga temperature representing the Ga flux, JGa, for three different N fluxes, JN. In terms of the dilute nitrides, kinetic modeling of microscopic processes during ECR microwave plasma-assisted molecular beam epitaxial growth of GaN/GaAsbased heterostructures was done by Bandic et al. [346]. Microscopic growth processes associated with GaN/GaAs molecular beam epitaxy were examined through the introduction of a first-order kinetic model, which has been applied to ECR microwave plasma-assisted MBE (ECR-MBE) growth of a set of delta GaNyAs1y/GaAs strainedlayer superlattices (SLS). The delta SLS consists of nitrided GaAs monolayers separated by GaAs spacers and exhibits a strong decrease in y with increasing T over the range 540–580 C. This y(T) dependence is quantitatively explained in terms of microscopic anion exchange and thermally activated N surface desorption and surface-segregation processes.
3.4 Nitride Growth Techniques -5
–1
Growth rate (μm h )
Ts = 700ºC
P = 0.7 x 10 Torr -5 P = 1.0 x 10 Torr -5 P = 1.5 x 10 Torr -5 P = 3.0 x 10 Torr
0.6
0.4
N-rich condition 0.2
Ga-rich condition 0.0 1020
1040
1060
1080
1100
1120
1140
1160
1180
Ga cell temperature (ºC) Figure 3.69 The growth rate of GaN at 700 C versus Ga cell temperature with the system pressure at 0.7 · 105 Torr (diamond); 1.0 · 105 Torr (circle); 1.5 · 105 Torr (triangle) and 3.0 · 105 Torr (square). The solid line indicates the stoichiometric condition (JGa/JN 1). The N-rich regime (JGa/JN < 1) is situated to the left side of the solid line; while the Ga-rich regime (JGa/JN > 1) is on the right side.
3.4.2.9.3 Which Species of N is Desirable? Let us now turn our attention to an issue touched upon earlier regarding the intricate nature of the extent of participation by various N species simultaneously provided by RF sources. Of particular interest to nitride growth by MBE, the effect of these N species on the growth of mainly GaN and to some extent on InN has been investigated [83,347]. Specifically, a comparative discussion of relative roles of N and N2 (likely to be the N2 ðA3 Suþ Þ species) species has been treated in a comprehensive manner in conjunction with GaN growth [348]. Although it is clear that thermodynamic equilibrium models cannot be used, quantitative information concerning the kinetics of interactions of the active nitrogen species with an evolving growth surface through different mechanisms is not available. A complex character of such interactions can be explained by a combined action of closely related chemical (such as bond breaking) and physical (such as collisions) effects, yielding a subtle intertwining of activated beam characteristics with growth parameters such as surface temperature and III/N flux ratio. The proposed chemical pathways for the growth process have been reported as follows [301]:
GaðlÞ þ 1=2N2 ! GaNðsÞ;
ð3:54Þ
where N2 represents the reactive and neutral molecules such as N2 ðA3 Suþ Þ, N2 Q Q Q ðB3 g Þ, N2 ða1 g Þ, and N2 ðC3 u Þ. Considering the lifetime only, the first species is thought to participate in growth following the reaction: GaðlÞ þ 1=2N2 þ ! GaNðsÞ;
ð3:55Þ
j451
j 3 Growth and Growth Methods for Nitride Semiconductors
452
where N2 þ ¼2
Pþ g
represents the reactive and ionized molecules, and
GaðlÞ þ N ! GaNðsÞ;
ð3:56Þ
where N ¼ P ; D represents the reactive N atoms, and 2 2
GaðlÞ þ N ! GaNðsÞ;
ð3:57Þ
where N ¼ S. All of the above processes could take place simultaneously with each contributing to the eventual growth rate determined by the relative concentration of each of these species in the plasma. Let us now attempt to treat whether the atomic N or N2 is better for growth, a topic that has been addressed by VanMil et al. [348]. Normally, determining the effect of various active nitrogen species on layer growth is complicated when RF plasma source is used because, as discussed above, these sources typically produce a complex mixture of active nitrogen superimposed on a background of presumed inert molecular nitrogen. However, there seems to be a discernible difference in the composition of N species produced by different sources. Capitalizing on this, those authors used an Oxford source, which predominantly produced atomic nitrogen and EPI source with an aperture plate having 400 holes with relatively smaller diameter, which produced predominantly radicals of molecular nitrogen. In this particular study, OMVPE grown buffer layer on (0 0 0 1) sapphire was used for the growth. These templates were mounted with In–Sn to the substrate block that was in contact with the monitoring thermocouple. A nitrogen flow rate of 0.85 sccm and RF power of 200 W employed led to a growth rate of 0.25 mm h1 when decomposition was negligible, meaning the growth temperature was sufficiently low. The efficacy of the particular species was determined by monitoring the growth rate as a function of temperature extended to the regime where notable decomposition would occur, as displayed in Figure 3.70. The growth rate in the case of atomic nitrogen began to rapidly decrease slightly above 650 C, whereas that for molecular radicals began to drop only very slightly at about 720 C. Even at 800 C, a growth rate of about 0.14 mm h1 was attained with molecular nitrogen, which is about 100 C higher than that for atomic nitrogen for the same growth rate. On all MBE-grown N-polarity samples, the temperature at which the growth rate began to decrease increased with a growth rate of about 0.17 mm h1 attainable at a high temperature of 850 C. The Arrhenius-type plot of the growth rate (growth rate vs. the inverse of substrate temperature) indicated a decomposition activation energy of 3.1 eV, which is the same as that for GaN decomposition in vacuum. Of particular importance is that both ionic and neutral atomic nitrogen can participate both in the growth and in the decomposition of GaN. One decomposition pathway might be through capturing of a nitrogen atom from the GaN to form N2, which would then desorb. Although the rate constants for the reactions involving atomic nitrogen are not known, free energy considerations for such reactions are favorable [349]. This might very well be the reason for the relatively poor efficiency for growth with atomic nitrogen. In short, competition between growth, surface decomposition, and adsorbed nitrogen capture appears to limit the efficacy of atomic nitrogen. In contrast, metastable molecular nitrogen would be much less likely to 4
Growth rate (μm h–1)
3.4 Nitride Growth Techniques
0.1
Atomic N N polar Metastable N2 Ga polar N polar
650
700
750
800
850
Substrate temperature (ºC) Figure 3.70 Dependence of the GaN growth rate on the substrate temperature for different primary nitrogen species, atomic nitrogen and molecular nitrogen radicals, and surface polarity. The predicted temperature dependence based on the measured vacuum decomposition rate for GaN with an activation energy of 3.1 eV is shown with a solid line for reference. Courtesy of B.L. van Mil and T.H. Myers, Ref. [348].
result in such a decomposition pathway, hence the increased upper limit on growth temperature using metastables. The incident Ga flux required to maintain Ga-stable growth increases rapidly above 750 C for Ga-polarity GaN with a slope comparable to Ga desorption rate above liquid Ga. In the case of N-polarity samples, the temperature at which a higher Ga flux begins to be required is shifted to higher substrate temperatures, implying N-polar material reflects a tighter bonding of Ga on this surface. These data illustrate the need for increased Ga overpressure with substrate temperature for maintaining twodimensional growth, suggesting that Ga desorption may also play a role in thermal decomposition during growth. These data indicate that increase in the Ga overpressure might lead to decrease in decomposition effects in GaN at normal growth temperatures. This may very well represent an additional benefit of growth under an increased Ga flux. Studies by VanMil et al. [348] of growth rate as a function of temperature suggest that the GaN surface is most likely attacked by atomic nitrogen above 700 C while attempting to form molecular nitrogen with nitrogen from GaN. Therefore, growth with neutral metastable molecular nitrogen would results in a significantly reduced thermal decomposition rate. In contrast, more ready decomposition with atomic N might lead to N vacancies and other ailments, which would all together lead to inferior carrier transport as borne out by the data obtained in these samples.
j453
j 3 Growth and Growth Methods for Nitride Semiconductors
454
0.35 E
0.30
Growth rate (µm h–1)
0.25 0.20
A
B
C
D
0.15 F 0.10 0.05 0.00 700
720
740
760
780
800
820
840
Tsubstrate (ºC) Figure 3.71 Growth rate as a function of growth temperature for a fixed Ga flux of 1 · 106 Torr BEP (beam equivalent pressure) along with electrical properties: (300 K mobility [cm2 V1 s1] and carrier concentration [cm3], peak low temperature mobility [cm2 V1 s1]): A: semi-insulating; B: mRT ¼ 200 (mmax-LT ¼ 200) –
n ¼ 1017; C: mRT ¼ 550 (mmax-LT ¼ 1150) – n ¼ 1015; D: mRT ¼ 320 (mmax-LT ¼ 660) – n ¼ 2 · 1017; E: mRT ¼ 560 (mmax-LT ¼ 1160) – n ¼ 8 · 1016 (Ga flux was increased to retain the lower temperature value); F: semi-insulating. Courtesy of B.L. van Mil and T.H. Myers, Ref. [348].
Follow-up studies also gave insight into the transport characteristics of the GaN films grown with molecular nitrogen radicals. A series of experiments investigating the growth rate for a fixed N flux versus Ga flux and also versus substrate temperature discovered that films grown below 660 C are semi-insulating. As shown in Figure 3.71, when the substrate temperature is increased to somewhat above 730 C, the room temperature electron concentration became 1017 cm3 and the mobility reached 200 cm2 V1 s1. Upon further increase in the growth temperature to 750 C, the room temperature electron concentration reduced to 1015 cm3 and the mobility increased to 550 cm2 V1 s1 (the maximum mobility for this sample reached 1130 cm2 V1 s1 at low temperatures). Continued increase in substrate temperature to 775 C led to electron concentration of 2 · 1017 cm3 and mobility of 320 cm2 V1 s1 (with a low temperature maximum of 660 cm2 V1 s1). A 820 C substrate temperature led to semi-insulating behavior. When the Ga flux was increased to retain the lower temperature value while growing at 810 C substrate temperature, electron concentration and mobility data of 8 · 1016 cm3 and 560 cm2 V1 s1 (with a low temperature maximum of 1160 cm2 V1 s1), respectively, resulted. The experiments with a constant nitrogen flux and a substrate temperature of 750 C while varying the Ga flux led to the typical increase in growth rate with an increase in the Ga flux up to a point beyond which the growth rate saturated corresponding to nitrogen-limited growth. At the turn over, see Figure 3.72 where the growth rate begins to saturate, the film was semi-insulating. As the Ga flux was
3.4 Nitride Growth Techniques
0.30
0.25
Growth rate (µm h–1)
0.20
A
B
C
D
0.15
0.10
0.05
0.00 0
2
4
6
8
10
12
14
16
–7
Ga flux (10 Torr BEP) Figure 3.72 Growth rate versus Ga-flux for Gapolar growth below the onset of significant thermal decomposition along with electrical properties for layers grown at 750 C. The units are electron mobility [cm2 V1 s1], maximum low temperature mobility [cm2 V1 s1] and
carrier concentration [cm3]): A: semi-insulating; B: mRT = 35 – n = 4 · 1016; C: mRT = 550 (mmax15 LT = 1130) – n = 10 ; D: mRT = 490 (mmax-LT = 1020) – n = 6 · 1016. Courtesy of B.L. van Mil and T. H. Myers, Ref. [348].
increased, the mobility increased reaching a maximum of 550 cm2 V1 s1 at room temperature. Further increase in Ga flux led to a reduction in mobility. The various growth regimes in play here are discussed in more detail in Section 3.4.2.9.4. 3.4.2.9.4 The Effect of III/V Ratio and Substrate Temperature on Surface Morphology Section 3.4.2.9.4 touches upon the role of the III/V ratio on the growth kinetics on GaN(0 0 0 1) surface. It is timely to discuss how parameters such as the flux ratio and growth temperature affect the growth and resultant layer quality. This issue has been a topic of extensive reviews [79,83,350,351]. Under Ga-stabilized conditions, layer-bylayer growth of GaN and relatively high-quality GaN growth on GaN templates have been discussed if samples are grown with RF N [317]. Typically, there are four growth regimes in terms of the resultant surface morphology, which can best be investigated if the samples are grown on templates with smooth surface morphologies such as those obtained by HVPE and OMVPE, as follows:
(a) N-rich growth regime: excess N covers the GaN surface (JGa/JN < 1). (b) Transition growth regime: presumably stoichiometric condition. The surface coverage of Ga or N can be neglected (JGa/JN 1). (c) Ga-rich growth regime: Ga adlayer covers the GaN surface (JGa/JN > 1). (d) Ga-droplet growth regime: Ga droplet formation on top of the Ga adlayer (JGa/ JN >> 1).
j455
j 3 Growth and Growth Methods for Nitride Semiconductors
456
4
III/V ratio
3 Ga-droplets
Ga-rich
2 Transition 1 N-rich 0
600650700750800 Growth temperature (ºC)
Figure 3.73 Surface phase diagram for PAMBE GaN homoepitaxy for a fixed N flux and varying Ga flux to vary the III/V ratio. The N rich, transition, Ga rich, and Ga-droplet regions are roughly indicated.
These four regimes are qualitatively, and somewhat subjectively, shown in Figure 3.73. Tarsa et al. [352] pointed out the significance of the III/V flux ratio (JGa/JN) to the surface morphology, structure, and optical properties of GaN layers grown on OMVPE-grown GaN templates. With low JGa/JN (N-stable growth), a granular surface morphology with a tilted columnar structure with a high density of stacking faults was obtained. However, with high JGa/JN (Ga-stable growth), a smooth surface morphology with characteristic spiral growth hillocks and some Ga droplets was obtained. The spiral growth hillocks have been argued to comprise bilayer height steps and terraces as a result of step-flow growth around mixed edge/screw dislocations [353,354]. Because two steps are pinned at each mixed dislocation, the spiral hillocks are comprised of two interlocking spiral ramps [355]. The RHEED studies indicate the transition between smooth and rough surface morphology in the homoepitaxial growth of GaN on GaN templates with varying flux ratio. For constant JN, there is a critical JGa at which the RHEED pattern transits between streaky (smooth surface) and spotty (rough surface) for a given growth temperature. The critical JGa is independent of substrate temperature below approximately 700 C (the measured value may change from system to system as measuring the substrate temperature accurately is not an easy task) and rises above this temperature because of Ga desorption. Figure 3.74a and b shows the typical RHEED and AFM data, respectively, for GaN grown in the N-rich regime. The Bragg spots always appear in the RHEED pattern during growth and the surface morphology is characterized by small islands in AFM images. These results indicate that the surface is very rough with an RMS roughness of 5.0 nm. The theoretical considerations of Zywietz et al. [256] show a low Ga adatom mobility on N-rich GaN(0 0 0 1) surfaces and a significantly shorter Ga diffusion length than under Ga-rich conditions. Low adatom mobility would lead
3.4 Nitride Growth Techniques
Figure 3.74 Typical (a) RHEED results and AFM (b) results for GaN grown in the N-rich regime.
to a high island nucleation rate and to multilayer growth, consistent with the rough GaN surface observed by RHEED and AFM. In the transition regime, the Bragg spots appear with weak streaky lines between them in the RHEED pattern (Figure 3.75) and the surface morphology is characterized by inverted pyramids in AFM (Figure 3.75b). The origin of these inverted pyramids is still unclear, but they appear to be similar to the V-shaped defects observed for GaN growth, which are defined by f1 0 1 1g facets. It is worth noting that the surface morphology of GaN layers grown in this regime strongly depends on the substrate temperature. At higher growth temperatures, the pyramids start to merge. When the substrate temperature is higher than 750 C, the surface becomes flat with deep pits. During growth under Ga-rich conditions, the RHEED pattern remains streaky (Figure 3.76). The surface morphology characterized by atomically flat terraces
Figure 3.75 Typical (a) RHEED results and AFM (b) results for GaN grown in the transition regime.
j457
j 3 Growth and Growth Methods for Nitride Semiconductors
458
Figure 3.76 Typical (a) RHEED results and AFM (b) results for GaN grown in the Ga-rich regime.
(Figure 3.76) is much smoother than that obtained in the N-rich and transition regimes. The RMS roughness in the Ga-rich case is less than 1.0 nm. Under this growth condition, the Ga adatoms are abundant and highly mobile at the substrate temperatures employed, resulting in a 2D growth mode and a smooth surface. There are some small hexagonal pits on the surface, with a density of 108–109 cm2, when the temperature is higher than 750 C. A higher desorption rate of Ga at this temperature may cause this phenomenon because higher substrate temperatures shift the growth toward the lower III/V regime [342]. Under the Ga-droplet regime, a weak but streaky RHEED pattern is obtained during the growth as shown in Figure 3.77. The surface morphology in this regime is smooth (Figure 3.77). In contrast to the sample grown in the Ga-rich regime, no
Figure 3.77 Typical (a) RHEED results and AFM (b) results for GaN grown in the Ga-droplet regime.
3.4 Nitride Growth Techniques
hexagonal pits are observed in large-area images. However, some spiral hillock features, attributed to bilayer height steps and terraces as a result of step-flow growth around mixed edge/screw dislocations, are visible on the surface. The typical size of the macroscopic Ga droplets observed on the sample is 3–5 mm at a density of 105 cm2, as determined by optical microscopy. The issue of the Ga-adlayer coverage has been illuminated by experimental investigations [350,356] of the adsorption and desorption of Ga on the GaN (0 0 0 1) surface by analyzing the RHEED specular spot transients during Ga deposition, Ga desorption, and Ga incorporation of consumption by exposure to N. A pseudooscillatory transient with duration affected by the Ga cell temperature (TGa) was observed after opening the Ga cell shutter. This is most likely owing to sudden heat loss caused by shutter opening and concomitant effort by the temperature control system to compensate for it. This issue was surfaced in MBE of GaAs and measures dealing with sudden heat loss were undertaken to minimize the problem. After closing the Ga shutter and opening the N shutter, a RHEED transient approximately symmetric to the adsorption transient but shorter in duration was observed. The time span of the desorption transient (tdes) increased initially with the Ga deposition time (tads) but remained constant (saturated) above a certain value of tads (7 s in the experiment [356]), even a long deposition time, tads ¼ 1 h, was employed. This saturation was attributed to the formation of a dynamically stable Ga adlayer. The thickness (d) of the Ga adlayer could be determined by the expression: d¼
r GaN ; 1=tN 1=tdes
ð3:58Þ
where rGaN is the GaN growth rate under N-limited conditions and tN is the duration of the transient for Ga adlayer consumption under N flux. The measurements were made for several substrate temperatures (Ts) in the 700–740 C range and in all cases a Ga film thickness of d ¼ 2.7 0.3 ML was determined. This is comparable with the predicted [357] Ga bilayer that contains 2.3 ML of Ga, in terms of GaN surface atomic density. Likewise, Ga adlayer coverage during GaN growth has also been investigated by Adelman et al. [350] by measuring the tdes of the excess Ga after interrupting the GaN growth by shuttering off both the Ga and N fluxes. From the evolution of the RHEED transient, four growth regimes were identified, as previously shown in Figure 3.73. RHEED transients exhibiting full oscillations were observed only in regime c – the Garich region but not quite in the regime with Ga droplets. Then the four regimes were associated with different Ga adlayer coverages on the GaN surface: (a) N-rich regime with no Ga accumulation on the surface; (b) slightly Ga-rich regime with Ga coverage perhaps less than a bilayer; (c) Ga-rich regime with approximately a Ga bilayer present on the surface; (d) very Ga-rich regime with appearance of Ga droplets. Based on these results, the authors constructed a phase diagram of the Ga adlayer coverage during GaN growth as a function of impinging Ga flux and substrate temperature (Ts) for a fixed active N flux, which is shown in Figure 3.73. Within the regime c, which represents the Ga-rich case with a Ga bilayer on the surface, GaN film growth proceeds within the realm of the step-flow mode, the
j459
j 3 Growth and Growth Methods for Nitride Semiconductors
460
justification for which has its roots in smooth surfaces with about 40 nm terraces, the observation of spirals induced by screw dislocations and the absence of RHEED oscillations. Adelman et al. [350] pointed out that there is a growth window in which smooth GaN surface without Ga accumulation can be attained. Additional investigations along these lines that shed more light on the mechanisms likely in play are available in the literature [358–360]. The surface morphology of MBE GaN films depends on whether GaN templates or some other substrate is used. If grown on templates, the correlation between the growth conditions and surface morphology is reasonably well defined. When GaN templates are used, the films grown under N-rich conditions exhibit rough surfaces with large density of pits, whereas those grown under Ga-rich conditions are smoother with characteristic spiral hillocks. The films grown in the intermediate regime exhibit large areas of flat surfaces separated by trenches. As the Ga flux is increased, those trenches give way to pits and if the Ga flux continues to be increased, the pits give way to spiral hillocks [317,318]. When other substrates are used, the surface morphology even under Ga-rich conditions is not as smooth as those on GaN templates grown under similar conditions with Ga droplets and pit being present simultaneously. The genesis of surface pits (depressions) has been correlated to extended defects [318]. The pits in the N-rich regime correspond to pooling and termination of mixed and pure edge threading dislocations (TDs). The larger, deeper pits correlate to mixed dislocations and the shallower pits correspond to edge dislocations. Both dislocation types are also pooled at the large pits (trenches) observed in the films grown within the intermediate regime. These large pits appeared to be associated with groupings of TDs, which were predominantly edge TDs forming low angle grain boundaries. Consequently, aside from the surface morphology, N-rich conditions have the benefit of orienting and pooling the dislocations where the probability of dislocation annihilation is increased. Consequently, when grown directly on sapphire or other foreign substrates, N-rich regime leads to lower overall dislocation density [361]. Very Ga-rich regime, on the contrary, does not necessarily have this mechanism in place to this extent and the resulting films have much higher density of pure edge dislocations as compared to N-rich samples as shown in the TEM images of Figure 3.78. Note, however, that when grown on GaN templates, the extended defect structure is to a first extent determined by the template, Ga-rich conditions and lead to smooth surface morphologies, making this regime more attractive. In an effort to gain better insight into the aforementioned mechanisms, a theoretical analysis of the dislocation-mediated surface morphology (pinned steps, surface depressions, spiral hillocks) of GaN layers, grown by MBE or OMVPE, has been undertaken by Heying et al. [355], using the theories of Burton, Cabrera, and Frank. The theoretical results of Zywietz et al. [256] can be brought to bear in to shed some light on the particulars of Ga- and N-rich growth results. An increase in the Ga adlayer coverage should increase the adatom surface diffusion and reduce the surface pit density, which can be completely eliminated in the Ga-droplet regime if it is grown on GaN templates prepared by HVPE or OMVPE. The high-resistivity of the films grown in the N-stable regime on GaN templates could be attributed to the formation
3.4 Nitride Growth Techniques
Figure 3.78 TEM bright-field images of a 1.9 mm thick film grown with a relatively low Ga/N ratio (R 1.1). (a) Cross-sectional image revealing surface undulation with amplitude 50 nm. Pairs of threading dislocations were observed to combine into one dislocation (black arrows). Dislocations also annihilated each other by joining into a half loop (white arrows). (b) and (c) Plan view images revealing dislocations
remaining mostly at the valleys of the morphology (bright areas). Image (c) was acquired from the same sample area as in (b) but with a dynamical, two-beam condition that results in thickness contours that reflect the film morphology. Dislocation density is about 1 · 109 cm2 at the top part of the film. Courtesy of R.M. Feenstra and Ref. [361].
j461
j 3 Growth and Growth Methods for Nitride Semiconductors
462
of compensating centers (point defects or point defect complexes) that trap free carriers. For reference, films grown under N-rich conditions on other substrates exhibit n-type conductivity at least at or near 1017 cm3. An increase in the Ga-adlayer coverage, as is the case in the intermediate regime, is expected to suppress the formation of these compensating centers resulting in an increase in the mobility and electron concentration. However, the optimum properties are achieved at the highest Ga flux within the intermediate regime (at the boundary between intermediate and Ga-droplet regimes). The degradation of the electrical properties within the Gadroplet regime is attributed [317] to the inferior quality of the material that is grown underneath the Ga droplets [353]. Techniques such as periodic desorption of excess Ga can be employed to combat the issue of inferior quality layer growth underneath the Ga droplets. It should be mentioned that these arguments are mere plausible mechanisms that could account for the results. More extensive data and experiments are needed to illuminate the underlying mechanisms. It should also be pointed out again that the above discussion applies primarily to the growth on GaN templates prepared by variants of vapor phase epitaxy.
3.5 The Art and Technology of Growth of Nitrides
GaN samples with the best transport and structural properties have so far been produced with HVPE. TEM, Hall measurements, secondary ion mass spectroscopy, photoluminescence, and deep level transient spectroscopy are among the techniques used to quantify the critical parameters in HVPE-grown thick GaN layers, some of which are freestanding in that they have been removed by laser liftoff (LLO) from the sapphire substrates on which they were grown. The method of X-ray diffraction is commonly used for quantifying the quality of GaN layers. It is worthwhile to say a few words about the various geometries used to glean information about the lattice constant and/or the quality of the layers, and to develop a firm grasp of the data presented as well as intricacies associated with various diffraction geometries. XRD is a powerful tool for investigating the crystalline structure of materials and had an omnipresence particularly during early developments of crystalline material and dates back to 1911 when XRD patterns of rock salt were obtained [362]. Singlecrystal diffraction and powder diffraction dominated the early employment of this method [363]. Single-crystal diffraction for structural analysis reveals the crystal and molecular structure of inorganic, organic, and biological compounds. The powder diffraction method is used to determine the crystal orientation and the scattered intensity when more than one phase is present in the sample. For the case in hand, which is the III nitride semiconductor epilayers, XRD is mainly used to evaluate the quality of the film, determine the mole fraction of alloys, and investigate the thickness and fine structure of materials with superlattice structures [364,365]. X-rays used in diffraction have wavelengths of the order of 0.5–2.5 Å, which is close to the spacing of atoms in crystals. Because atoms are arranged periodically in a lattice, diffraction occurs when X-rays are coherently scattered by them and the crystal
3.5 The Art and Technology of Growth of Nitrides
structure is revealed through appropriate analyses. Braggs law is the simplest and most useful description of crystal diffraction: nl ¼ 2d sin q;
ð3:59Þ
where n is an integer representing the order of diffraction, l is the X-ray wavelength, d is the interplanar spacing of the diffracting planes, and y is the angle of incidence. Details about the crystal structure can be obtained from the maxima in an XRD pattern using Braggs law. Figure 3.79a shows the typical schematic diagram of an XRD system with a special point made in regard to the terminology used for various scattering geometries (Figure 3.79b). An X-ray source, a beam conditioner, and a detector are the necessary elements of an X-ray system. X-rays are produced in an X-ray tube by accelerating a high-energy electron beam toward a metal anode, resulting in characteristic X-ray transitions in the anode. The beam conditioners are used to collimate and monochromate the X-rays. The specimen is mounted on a sample holder that can rotate along three axes: Phi (F), Psi (C), and Omega (O). A detector is used to collect the diffracting X-rays, where the detector can move through the Bragg angle while scanning. For determining the out-of-plane lattice constant, the y 2y scan is employed, refer to Figure 3.79 for details. In this case, the magnitude of Dk varies while maintaining its orientation relative to sample normal, shown in Figure 3.80. The resulting data of intensity versus 2y show peaks for Dk values satisfying the diffraction condition. This scan mode is used to identify the lattice constant and ultimately determine the mole fraction of alloys. It is also used to investigate the thickness and fine structure of materials comprising superlattices that some use to calibrate layer thicknesses if sufficiently smooth interfaces are obtained to allow observation of the satellite peaks. For determining the layer quality, the y o scan or proverbially called rocking curve is employed. In this case, the orientation of Dk, which represents the difference between the incident wave vector k0 and the diffracted vector k0 , varies relative to the sample normal while maintaining its magnitude, as shown in Figure 3.80b. When diffraction intensity versus o (sample angle) is plotted, the full width at half maximum is a measure of sample quality. In the (0 0 2) diffraction (the hkl notation equivalent to (0 0 0 2) diffraction in the hjkl notation), the dispersion relates to the degree of tilting with respect to the basal plane. In the (1 0 4) and (1 0 2) (the hkl notation equivalent to ð1 0 1 4Þ and ð1 0 1 2Þ diffractions in the hjkl notation) diffraction geometries, the half width is sensitive to the in-plane rotation as well as perfect edge dislocations. In other words, the width of the rocking curve is a direct measure of the range of orientations present in the irradiated area of the crystal, as each subgrain of the crystal is probed while the crystal is rotated. As such, it can be manipulated by narrowing the X-ray slit widths, which serve to reduce the radiated area. Relatively thin layers in terms of HVPE standards and thick in terms of the OMVPE and MBE standards, on the order of 10 mm, show symmetric and asymmetric X-ray diffraction peaks of about the same width, near 5–10 arcmin. Because these samples
j463
j 3 Growth and Growth Methods for Nitride Semiconductors
464
Tilt Psi (Ψ)
Rotation Phi(φ)
2θ
s am
Detector
For rocking curve
pl e
Omega (Ω) Teta (θ)
Detector axis Beam conditioner
(a)
[0 0 0 1] Tilt
[0 0 0 1]
[0 0 0 1] Tilt
Incident beam
Tilt
Sapphire substrate
[0 0 0 1] (0 0
0 2)
Roc kin gd for ire tilt mea ction sure men t
Twist (0 0 0 1) Rocking direction for twist measurement
)
(1010) 10 0 (1
[1 0
[1 1 2 0]
10
] Twist
(b) Figure 3.79 Schematic diagram of the XRD technique (a). Schematic of an X-ray diffraction configuration for a typical GaN layer with (0 0 0 2) diffraction geometry being sensitive to the parallelinity of the basal plane and (1 0 1 0) diffraction being sensitive to twisting of the columns indicated (b).
[1 1 2 0] [1 1 2 0]
3.5 The Art and Technology of Growth of Nitrides
Δk
k'
k0
2θ
θ
(b)
Δk Sample normal
k0
k'
ω (a) Figure 3.80 Schematic configuration of XRD. (a) y/2y scan: the magnitude of Dk varies while maintaining its orientation relative to the sample normal; (b) y/o scan: the orientation of Dk varies relative to the sample normal while maintaining its magnitude.
represent a benchmark for GaN, their properties are treated, but not before a few words on X-ray diffraction analysis itself. The OMVPE and MBE grown films with very narrow out-of-plane rocking curves with FWHM as low as 37 arcsec have been reported. Conventional wisdom would imply that the X-ray rocking curve FWHM of the GaN(0 0 2) in hkl and (0 0 0 2) in hjkl convention peak is a good barometer of quality. This conclusion is based on a good deal of experience with conventional semiconductors grown homoepitaxially. In conventional semiconductor technology it is typical that the rocking curves become significantly broadened at high densities of dislocations (generally >105 cm2).
j465
j 3 Growth and Growth Methods for Nitride Semiconductors
466
However, the GaN films with narrow rocking curves contain threading dislocations which are predominantly pure edge types along the c, axis. The specific threading dislocation geometry leads to distortions of only the specific crystallographic planes. Pure edge dislocations distort only the (hkl) planes with either h or k being non-zero. Rocking curves on off-axis (hkl) planes are broadened while symmetric (0 0 1) rocking curves are insensitive to the pure edge threading dislocation content in the film. Screw threading dislocations with [111] directions have a pure shear strain field that distorts all (hkl) planes with l non zero [366,367]. Therefore, the rocking-curve widths for the off-axis reflections are more reliable indicators of the structural quality of GaN films. The properties of GaN film grown on sapphire differ for the in-plane and outof-plane in terms of structural features. The measured in-plane coherence lengths are smaller and the rocking-curve widths are larger than in the plane-normal direction. It implies that optical and electrical properties are anisotropic along the film plane and plane-normal directions. The in-plane structure measurements are more closely related to the electronic mobility and optical properties than those in the planenormal direction. This premise is supported a high-resolution X-ray investigations [368]. Films with very sharp (45 arcsec) out-of-plane rocking curves show poor electrical and optical properties, which is consistent with an in-plane or asymmetric X-ray diffraction analysis. On the contrary, the films with about 5 arcmin out-of-plane rocking curves reveal much better optical and electrical properties; this is again consistent with the excellent in-plane X-ray diffraction. Judging from the out-of-plane XRD data and the cross-sectional TEM images of the same films, it appears that the planar edge dislocations do not really affect the symmetric out-of-plane diffraction and sharp out-of-plane diffraction is not necessarily a sign of enhanced overall quality. Owing, in part, to the heteroepitaxial nature of GaN growth on foreign substrates with the accompanying chemical and lattice mismatch, some kind of a buffer layer is imperative for achieving high-quality GaN. In the case of OMVPE, the buffer layer or the nucleation layer is deposited at relatively low temperatures (550–650 C) compared to that for epitaxial film growth. In the MBE case, however, AlN buffer layers are grown at higher temperatures, generally between 700 and 900 C. The low-temperature buffer layer can protect the GaAs surface at relatively high growth temperatures (>700 C) and improve the quality of the zinc blende phase in addition to enhancing GaN nucleation and initiating growth. The thickness and growth temperature of this low-temperature buffer layer have a remarkable effect on the crystalline quality of the wurtzitic and zinc blende polytypes of GaN in MBE or OMVPE growth. Lowtemperature buffer layers grown by OMVPE are of polycrystalline, mixed polytype nature, which improves upon annealing under H2 atmosphere during the ramping process to growth temperature, as discussed in Section 3.5.3 on SiC and Section 3.5.5.1 on sapphire substrates. Until the advent of buffer layers, the quality of GaN films grown directly on sapphire was very poor and the background concentrations were in some cases as high as 1020 cm3 [5,6]. The poor crystalline quality was characterized by broad X-ray rocking curves of typically 10 arcmin and uneven surface morphologies that contained hillocks. A low-temperature AlN or GaN buffer layer improves the GaN layer quality substantially. Although the low-temperature AlN and GaN buffer layers were
3.5 The Art and Technology of Growth of Nitrides
initially used solely for OMVPE, their modified implementation in the MBE growth process proved highly beneficial too. 3.5.1 Sources
Trimethylgallium for metal and high-purity ammonia for nitrogen are the most commonly used ingredients. They are transported to the growth chamber. The gas manifold typically features fast switching of the group III elements and dopants and permits separate injection of ammonia. Nominally, c-plane [0 0 0 1] oriented sapphire substrates are used, though other orientations of sapphire and alternative substrates are sometimes employed, too. For a hydrogen flow of 40 sccm through the TMG bubbler (10 C), an NH3 flow of 3.0 standard liter per minute (slm) and hydrogen make up the flow of 1.5 slm; the GaN growth rate was 2 mm h1. Sources commonly employed for Al and In are TMA and TMI, whereas for n- and p-type dopants they are silane and Cp2 Mg. Before the employment of low-temperature nucleation buffer layers, to be discussed in Section 3.5.5.1, it was commonly accepted that the inability to provide sufficient quantities of nitrogen was the bottleneck deserving particular attention [369]. Consequently, much of the early efforts dealt with exploring ways to address this issue of the high vapor pressure of N2 on GaN. Several investigators substituted the more reactive hydrazine in favor of ammonia [143,370]. Moreover, attempts have been made to use Ga(C2H5)3 NH3, which already has a Ga–N bond, for the GaN source [371]. Although no evidence to justify the approach exits, plasma excitation of ammonia in a CVD growth environment was attempted. Suitable precursors are those that possess good reactivity, thorough pyrolysis, and transportability. As mentioned earlier, the precursors, ideally, should be nonpyrophoric, water and oxygen insensitive, noncorrosive, and nontoxic. TMG and TEG are very popular for Ga, though GaCl has also been tried. TMI and TMA are the commonly employed sources of In and Al, respectively. Ammonia (NH3) is the unchallenged source of nitrogen, as it is reasonably pure and stable. TMI and TMA are the commonly used sources of In and Al, respectively. TMA and NH3 have been reacted in the presence of hydrogen plasma to grow AlN. TMI has been utilized with microwave-activated N2 to grow InN. Even metallic Ga transported with nitrogen to the reaction zone where it is allowed to react with active nitrogen has been tried. During the growth of nitrides employing trialkyl precursors, adduct formation between ammonia and TMA and TMG has been well documented. Usually when mixing at room temperature, the adduct formation between TMG or TEG and ammonia is completed in less than 0.2 s. The vapor pressure of the resulting adduct Ga(CH3)3–NH3 is 0.92 Torr at room temperature, whereas that of Ga(C2H5)3 : NH3 is much lower (for a review, see Ref. [370]). Deposition of InN is nearly an intractable problem even for low-temperature deposition processes such as MBE let alone OMVPE, which is a much higher temperature process. Needless to say that InN is a very challenging problem for OMVPE because the decomposition temperature is close to the minimum temperature
j467
j 3 Growth and Growth Methods for Nitride Semiconductors
468
for efficient thermal activation of ammonia, about 550 C (N–H bond strength, 3.9 eV), and pyrolysis of the organometallic source [188,372]. There are two reports of alternative single-source precursors for InN. The solid-state pyrolysis of polymeric [In (NH2)3] [373] and air-stable, nonpyrophoric, and volatile single-source precursor (N3)Ga[(CH2)3N(CH3)2]2 [374]. Growth rates between 0.3 and 5 mm h1 were obtained at the substrate temperature range of 300–450 C and a reactor pressure of 2 · 104 Torr [374]. As mentioned earlier in Section 3.4.2.7, plasma excitation of the nitrogen species has not proven to be necessary in CVD growth. Although trialkyl compounds (TMA, TMG, TMI, etc.) are pyrophoric and extremely water and oxygen sensitive and ammonia is highly corrosive, much of the best material grown today has been produced by conventional OMVPE by reacting these trialkyl compounds with NH3 at substrate temperatures close to 1000 C. Temperatures in excess of 800 C are required to obtain single crystalline high-quality GaN films, with the best GaN films grown at 1050 C. At substrate temperatures exceeding 1100 C, the dissociation of GaN results in voids in the grown layer. A similar situation has also been observed for AlN film growth. To overcome this difficulty, N2H4 has been employed instead of NH3, with the conclusion that a significantly smaller amount of N2H4 was required to maintain the same growth rate. 3.5.1.1 HVPE Buffer Layers and Laser Liftoff The primary types of layers grown by HVPE are GaN, though the method has been extended to include AlGaN as well, which is either directly grown on sapphire by this technique, in which case the layer quality is not very good, or a thin ZnO layer is deposited on sapphire prior to the deposition of GaN. The GaN deposition temperatures far exceed the temperature up to which ZnO is thermally stable. Sputtered deposited ZnO on sapphire in either the Ar or O2 atmosphere was successfully used, by Detchprohm et al. [375], prior to the deposition of GaN to demonstrate improved material as compared to direct deposition of GaN on sapphire. In the same vein, Molnar et al. [11,376] also used the HVPE technique to grow thick GaN templates by employing ZnO as a sacrificial buffer layer. To achieve such properties, the sapphire substrates were treated in situ with GaCl or ZnO [377]. GaN films of sufficient thicknesses to be self-supporting templates were removed from sapphire by laser liftoff [137]. Another group headed by Park et al. [21] at Samsung Advanced Institute of Technology has been producing freestanding templates with this technique as well. These layers and templates exhibit the lowest impurity concentration with the highest electron mobilities of any GaN so far. The deposition temperature is generally about 1050 C with growth rates of 15 mm h1. The results obtained thus far indicate that high-quality GaN films and templates can be grown on sapphire and other substrates by HVPE. These layers and templates served as the basis for further growth on them by MBE and OMVPE with excellent results. As such, the technique is very useful, the extent of which could be expanded if high-resistivity films could also be obtained. In this vein, HVPE compensated with Zn exhibited high resistivities, the properties of which are discussed in Volume 2, Chapter 5 in terms of optical processes, as the electrical measurements are difficult to
3.5 The Art and Technology of Growth of Nitrides
interpret. Owing to the high temperatures employed, it has not yet been possible to grow InGaN ternary with this method. Similarly, p-type doping has so far been lacking, which means that HVPE technique alone is not in a position to produce device structures requiring p–n junctions. Although Si and O incorporation from the walls has been an issue, the high quality of the recent layers indicates that this issue is not as serious as it used to be. Gu et al. [378] extensively studied the role of ZnO while depositing GaN by the HVPE method. They observed that the thin ZnO prelayer evaporates from the sapphire surface at the GaN growth temperatures because ZnO is thermodynamically unstable, but not before a chemical reaction occurs. The XPS analysis performed by Gu et al. [378] showed traces of ZnAl2O4 after a simulated thermal cycle. One can conclude that ZnAl2O4 so formed may act as nucleation centers for GaN growth. The carrier concentration of GaN grown on ZnO (20 nm) coated sapphire was reduced by several orders of magnitude, as compared to those grown on bare sapphire, to 1 · 1016 cm3. Elaborate buffer layers or templates are grown with the HVPE technique also, among which is the ELO for defect reduction. The topic is discussed in sufficient detail in conjunction with OMVPE in Section 3.4.1.2, the basics of which also apply to HVPE. Nevertheless, for flow and compactness, the particulars of ELO associated with HVPE are discussed in the same section. As briefly mentioned in Section 3.4.1.1, HVPE with its large growth rate is well suited for thick layers to the point where freestanding GaN can be obtained when the layer is removed from the substrate. Sacrificial buffer layers such as AlN between the substrate and the bulk of the layer followed by selective etching in, for example, KOH has not been very successful because of chemical diffusion, stagnation, and complications with growth as HVPE is much better suited for growth of GaN. Another approach that does not rely on chemical processes but makes use of residual thermal strain that serves to self-detach the layer from its sapphire substrate during cooldown has been developed [379]. However, a rather cumbersome pretreatment of the wafer and the need for lithography make the method less desirable. Although the LLO [137] method has been comparatively successful, farming of large area wafers has proved difficult. In the LLO process, a GaN layer is separated by irradiation of the substrate–film interface through the substrate with high-power pulsed laser at a wavelength to which the substrate is transparent, but is strongly absorbed in the GaN layer. The third harmonic of a Nd : YAG laser [137,380] or excimer lasers [381] meet these requirements. The absorption of such high-intensity laser pulses causes a rapid thermal decomposition of the irradiated GaN interfacial layer into metallic Ga and gaseous N2. An extensive review [382] on the topic is available and the process is described with a schematic diagram shown in Figure 3.81. As briefly touched upon, the epitaxial layers are mechanically very fragile and break into pieces during the LLO process if performed as described in Figure 3.81. This makes it imperative that the layers be supported during liftoff. The elastic modulus of the supporting layer must be compatible with that of GaN, must be flexible and easily removable, and must handle the temperature used during liftoff, that is, 650 C. To remove the veil around this suspense, strictly speaking that material is not available.
j469
j 3 Growth and Growth Methods for Nitride Semiconductors
470
Laser pulse Sapphire screen Thermal decomposition
Scanning Al2O3 Substrate
N2
Nitride layer
Liquid Ga
Hot plate Figure 3.81 Cartoon depicting the LLO process in which high-intensity laser pulses area allowed to enter the sample from the sapphire substrate side and thermally decompose a thin GaN-layer at the substrate interface due to strong absorption. The shock waves resulting from the
explosive generation of nitrogen gas during pulse are damped by placing the GaN-sample into sapphire powder. A hot plate can be used to raise the substrate temperature during the process to mitigate some of the accumulated thermal strain. Courtesy of M. Stutzmann, Ref. [382].
However, crack-freesamplesweredelaminatedbyfixingthe samplestotwo-component epoxy resin prior to laser liftoff. The downside of this method is that it is difficult to separate the resinfromtheGaNfilmafter sapphire removalbecause ofits highchemical stability. Although the epoxy can be dissolved with dichloromethane or in boiling water after several minutes, it still represents a complication. A more practical method, which is chronicled in Figure 3.82, has been used to separate GaN from its substrate, albeit with some cleavage lines appearing. In this process, the top surface of the film on sapphire is coated with the thermoresistant (up to 300 C) silicone to an approximately 3 mm thick support layer. The film is then affixed to a metallic plate with double-sided adhesive tape to stabilize the flexible silicone elastomer. Following laser scanning at room temperature, the sapphire substrate is removed. Then, the film with the elastometer affixed is removed with the aid of an approximately 3 mm thick transparent thermoplastic adhesive, as illustrated in Figure 3.82d. The strain in thick HVPE films, which tend to crack the wafer upon laser liftoff, or any other separation method for that matter, limits the size of the freestanding GaN available for further epitaxy. Gogova et al. [383,384] and Kasic et al. [385] have reported on the characteristics of approximately 300 mm thick crack-free GaN wafers initially grown on a 2-in. (0 0 0 1) sapphire substrate buffered with a 2 mm thick GaN layer grown by organometallic vapor phase epitaxy. What is interesting about these wafers is that during the cooldown from the growth temperature, the strain was relaxed by inducing cracks in the sapphire substrate instead, allowing strain-free bulklike GaN wafer to be attained. The in-plane homogeneity of the wafer was monitored by lowtemperature photoluminescence mapping. The position of the main near-bandgap photoluminescence line and the phonon spectra obtained from infrared spectroscopic ellipsometry showed that the 2-in. crack-free GaN is virtually strain free over a
3.5 The Art and Technology of Growth of Nitrides
Figure 3.82 Cartoons illustrating the laser liftoff process sequence of 20 GaN membranes. (a) Laser lift-off of the GaN film coated with silicone elastomer and affixed onto a support template; (b) removal of sapphire following laser scanning; (c) deposition of an approximately 3 mm thick thermoplastic adhesive layer at
120 C; (d) peeling-off of the silicone elastomer. In the last step the GaN film is fully removed by dissolving the thermoplastic adhesive in acetone. Courtesy of M. Stutzmann, Ref. [382]. (Please find a color version of this figure on the color tables.)
diameter of approximately 4 cm. The dislocation density determined from cathodoluminescence images was 2 : 0 · 107 cm2 on the Ga-polar face. The full width at half maximum value of XRD o-scan of the freestanding GaN was 248 arcsec for the ð1 0 1 4Þ reflection. Wafer bonding in conjunction with laser liftoff has been developed, which also has the added advantage of reducing current crowding, discussed in Volume 3, Chapter 1 in the context of quasi-planar device process owing to the insulating nature of sapphire. Wong et al. [386] have used a Pd–In bonding to bond GaN to Si, whereas Jasinski et al. [387] have bonded GaN to GaAs by direct wafer fusion in which a 2 mm thick GaN film grown on sapphire was fused to a GaAs substrate in nitrogen ambience by applying uniaxial pressure of 2 MPa at temperatures of 550 and 750 C. Separation of HVPE GaN layers from sapphire produced freestanding GaN templates, paving the way for producing high-quality film growth by both OMVPE and MBE. In addition, the freestanding templates provided high-quality samples for investigating transport and optical properties of GaN with accuracy and confidence that were not possible with previous layers. The structural properties
j471
j 3 Growth and Growth Methods for Nitride Semiconductors
472
of the benchmark thick HVPE layers still affixed to their sapphire substrates and freestanding varieties are discussed below. 3.5.1.2 Benchmark HVPE Layers/Templates For structural analyses, transmission and cross-sectional TEM [388,389] (also with varying thicknesses to investigate the evolution of extended defects as the growth progressed) and X-ray diffraction measurements were carried out. In some cases, extended defect delineating etches were employed for quicker analysis. The dominant defects present in the layers are the threading dislocations with the density of misfit dislocations at the layer–substrate interface being about 2 · 1013 cm2. Brightfield (BF) TEM images, recorded under multibeam conditions to image all dislocations with different Burgers vectors, were used to estimate the density of these dislocations with resultant figures to be of the order of about 1010 cm2 near the layer–substrate interface. The density of threading dislocations gradually decreased away from the interface and for the 55 mm thick layer it reached a value of about 108 cm2 near the surface. The threading dislocation densities are plotted as a function of distance from the interface in Figure 4.45. A gradual decrease in density of these dislocations with increase in the distance from the substrate shown on this plot indicates a gradual improvement in layer quality with thickness. The relative number of different types of threading dislocations (edge, screw, and mixed) were investigated and two types of dark field images with (0 0 0 2) and ð 2 1 1 0Þ type reflections were recorded. In the first type of images, only the screw and mixed dislocations are visible, whereas in the second ones, only the edge and mixed dislocations are observed. The results indicate that the edge, screw, and mixed dislocations are present in comparable densities. A freestanding GaN template [390] was obtained from Samsung and was grown by HVPE on sapphire to a thickness of 300 mm and separated from the sapphire substrate by laser-induced liftoff [137]. The GaN layer was then mechanically polished and dry etched on the Ga-face to obtain a smooth nearly epi-ready surface, whereas the N-face was only mechanically polished. The polarity on the two sides of the GaN template was determined by the well-established method of convergent beam electron diffraction (CBED). Because GaN is non-centro-symmetric, the difference in the intensity distribution within (0 0 0 2) and ð0 0 0 2Þ diffraction discs in the convergent beam electron diffraction pattern can be used to determine polarity of the sample, taking into account the effect of the sample thickness on the intensity distribution. The CBED patterns obtained on the side previously next to the substrate indicate that it is of ½0 0 0 1 N-polarity, which means that a long bond along the c-axis is from N to Ga. The polarity determination by CBED is consistent with chemical etching experiments in which the N-face is etched very rapidly in hot phosphoric acid (H3PO4). In addition, Schottky barriers fabricated on this surface exhibited a much reduced Schottky barrier height (0.75 eV versus 1.27 eV on the Ga-face) [391], only after some 30–40 mm of the material was removed by mechanical polishing followed by chemical etching to remove the damage caused by the first mechanical polish. An in-depth treatment of current conduction in Schottky barriers is provided in Volume 2, Chapter 1.
3.5 The Art and Technology of Growth of Nitrides
Figure 3.83 Bright field TEM micrographs of a cross-section sample near the N-face side for the g-vectors perpendicular (a) and parallel (b) to the c-axis. Note that both dislocations are visible in both images. Courtesy of J. Jasinski and Z. Liliental-Weber.
The density of threading dislocations determined from the plan view sample was estimated to be about 4 1 · 107 cm2. These threading dislocations were observed in cross-sectional view also. Few of them are clearly visible in bright-field images as shown in Figure 3.83. The density of these dislocations determined from cross section was found to be about 3 1 · 107 cm2, which is in good agreement, within experimental error, with that obtained from the plan view sample. For comparison, a density of about 1 · 107 cm2 was obtained by etching the N-face in H3PO4 for 15 s at 160 C followed by counting the etch pits on several images, details of which are discussed in Section 4.2.4. Most of these threading dislocations are of mixed Burgers vectors because they are visible in bright-field images with g-vector parallel and perpendicular to the c-axis (Figure 3.83). However, caution is warranted with such a conclusion because of the very low statistics (very few dislocations observed within the electron transparent area). The plan view TEM studies for the Ga-face side revealed a surface with an estimated dislocation density of much less than 1 · 107 cm2; however, owing to the very low statistics, there is a relatively large uncertainty for this estimation. Hot H3PO4 defect revealing process followed by several AFM images up to 50 mm · 50 mm indicated a dislocation count of about 5 · 105 cm2. The
j473
j 3 Growth and Growth Methods for Nitride Semiconductors
474
significantly lower dislocation density of the Ga-face side with respect to that near the N-face was most likely because of dislocation interaction within the layer, which is remarkable because reduction through dislocation interaction has not been very successful in GaN. In standard HVPE GaN layers [392,388], the dislocation density is several orders of magnitude larger and indicates a very high structural quality of the freestanding GaN template. High-resolution X-ray rocking curves from the HVPE samples were measured by a Philips XPert MRD system equipped with a four-crystal Ge(2 2 0) monochromator and Cu Ka1 line of X-ray source. The instrumental resolution was better than 10 arcsec in the diffraction geometry in effect. The X-ray diffraction rocking curves from both the Ga- and N-faces were examined for different X-ray beam slit widths, from 2 down to 0.02 mm, which correspond proportionally to the spot size on the sample owing to the highly collimated nature of the beam. About 10 mm thick HVPE epitaxial layers on sapphire substrates exhibited (0 0 0 2) peaks with FWHM values ranging from 5.8 to 9.3 arcmin. The (1 0 4) in hkl and ð1 0 1 4Þ in the hjkl convention asymmetric peak FWHM values were narrower and ranged from 3.9 to 5.2. As mentioned above, the in-plane structure is more closely related to electron mobility and, to some extent, to optical properties than those in the plane-normal direction, suggesting that the asymmetric diffraction, as opposed to out-of-plane symmetric diffraction, should be weighed more heavily. A crystallographic analysis by high-resolution X-ray diffraction (HRXRD) rocking curves (o-scans) with different slit widths of a GaN template [393] shows a very narrow FWHM on the Ga-face of the freestanding template, down to 69 arcsec for (0 0 0 2) reflection (at a slit width of 20 mm), and 103 arcsec for the ð1 0 1 4Þ diffraction (at a slit width of 100 mm as the signal gets to be very small for smaller slit widths and much longer data collection times are required). The FWHM for the N-face is 160 arcsec for (0 0 0 2) direction (at a slit width of 20 mm) and 140 arcsec for the ð1 0 1 4Þ diffraction (at a slit width of 100 mm). The superior quality of the Ga-face over that of the N-face agrees well with the large deviation of extended defect densities as determined from the etching experiments. Comparing with the reported X-ray data [394,395] of HVPE GaN, the FWHMs of which are typically in the range of 5–10 arcmin for the (0 0 0 2) peak and 4–5 arcmin for the ð1 0 1 4Þ peak, the density of both types of dislocations are dramatically reduced in the freestanding GaN template. With the increase in the slit width, not only is the FWHM increased, but also a nonGaussian multipeak feature emerges. Specifically, when the slit size is increased from 0.02 to 2 mm, the FWHM of the (0 0 0 2) peak on the Ga-face increases from 69 arcsec to 20.6 arcmin; when the slit size is increased from 0.1 to 2 mm, the FWHM of the (1 0 4) (ð1 0 1 4Þ in the hjkl convention) peak increased from 103 arcsec to 24 arcmin. A more improved Samsung GaN template exhibited much better X-ray diffraction characteristics. The FWHM of (0 0 2) ((0 0 0 2) in the hjkl convention), (1 0 2) (ð1 0 1 2Þ in the hjkl convention), and (1 0 4) (ð1 0 1 4Þ in hjkl convention) peaks were 53, 137, and 54 arcsec, respectively. Source beam slit widths of 20 mm for the [0 0 2] diffraction and 50 mm for the (1 0 4) and (1 0 2) diffractions were used, as the signal intensity for these asymmetric peaks was smaller. This sort of broadening has been attributed to the tilt and twist [367] that results in submillimeter scale mosaic
3.5 The Art and Technology of Growth of Nitrides
spread and causes the observed dependence on the slit size. This seems at first not plausible as the X-ray diffraction system uses a highly collimated Cu Ka1 point source. Instead, the low defect concentration observed by TEM and defect delineating etches in these templates point to another source, namely bowing [396]. This is despite the fact that GaN was removed from sapphire and a casual observer could indeed make the assumption that it is relaxed with no cause for bowing. Although the exact genesis is not known at this time, bowing could also be caused by the surface mechanical polish, which affects both Ga- and N-faces. By assuming the FWHM at beam width of 0.02 mm to be the intrinsic broadening of GaN, an upper limit of the bowing radius of 1.20 m or perhaps slightly larger was estimated. It has been reported [137] that a bowing radius of about 0.8 m was found in the HVPE-grown GaN thick films (275 mm) before separation from sapphire substrate and a bowing radius of about 4 m after separation. As mentioned above, the sample underwent a mechanical polishing, which could be the reason for the relatively large bowing. The genesis of impurities is the chemicals used and/or introduced during the deposition process. The latter may be because of the deposition environment and impurities present in the source material used. In addition, the presence of many point defects creates ionized centers as well. Among the chemical impurities are elements such as O, Si, and C, with C receiving the attention during the early development of GaN, which gave way to O and Si as time went on. The nature of carbon in GaN is not well understood. In addition to being substitutional and well-behaved impurities, these species can also form complexes with native defects, complicating the analyses. Because the conductivity is of paramount importance for devices, it is imperative that impurity incorporation in GaN is examined. However, it is agreed that carbon compensates donors and that the resistivity of the sample can be changed by varying the carbon concentration. This is achieved in an OMVPE environment by changing the system pressure used for growth, lower pressures leading to higher carbon incorporation (therefore to higher resistivities) and higher pressures leading to lower carbon incorporation and therefore lower resistivities [397]. Secondary ion mass spectroscopy (SIMS) measurements demonstrate that it is difficult to reduce common impurities, such as C, Si, and O, much below the 1017 cm3 level in thick GaN layers, 1016 cm3 in freestanding templates, and somewhat worse in thin epitaxial layers grown directly on lattice-mismatched substrates. The picture is somewhat better in films grown by MBE and OMVPE on HVPE templates. The SIMS data can shed light on whether the measured donor and acceptor concentrations, by Hall measurements, are due to impurities or native defects or both. The SIMS technique is refined and calibrated to the point where chemical impurity concentrations down to about 1015 cm3 can be quantitatively determined. In Figures 3.84 and 3.85, SIMS profiles in an HVPE layer on sapphire and a freestanding GaN template (for both Ga and N faces) are shown for O, C, and Si. In the HVPE layer, both O and Si concentrations drop rapidly away from the surface, owing in part to the artifact of the technique and in part to condensates on the surface, down to about 1017 cm3 for Si and a high 1016 cm3 for O. It should be mentioned that there is a rise in O concentration near the interface with the substrate owing to O out-diffusion [398].
j475
j 3 Growth and Growth Methods for Nitride Semiconductors
476
Figure 3.84 SIMS profiles in a HVPE GaN layer on sapphire substrate showing the depth distributions of O and Si concentrations. The average n and true n (from Hall measurements) in HVPE GaN on Al2O3 are also shown. VGa-PA and n-ECV indicate positron annihilation
Concentr ation (cm -3)
10
10
10
10
10
10
determined Ga vacancy concentration and electropolish capacitance–voltage determined electron concentration, respectively. Courtesy of Steve Novak of Evans East and D.C. Look of Wright State University.
21
N face
20
O
19
C
18
Si
17
Ga face
C O Si
16
0
1
2 Depth ( µm)
3
Figure 3.85 SIMS profiles in a HVPE GaN template showing the depth distributions of O, C, and Si concentrations from both Nand Ga- faces with Ga-face figures being below mid-1016 cm3. Courtesy of Steve Novak of Evans East and D.C. Look of Wright State University.
3.5 The Art and Technology of Growth of Nitrides
The Ga-face profile in the Samsung template indicates levels below mid 1016 cm3 for all three impurities. The picture is different for the N-side, however, as this side was juxtaposed to the substrate during growth and was mechanically polished after laser separation. The impurity concentration is some 1–3 orders of magnitude higher than is the case for the Ga-face. Particularly, the concentration of O and C is high, albeit some drop occurs deeper in the film. Data suggest that impurity incorporation depends on the structural quality of the film. 3.5.2 Growth on GaAs Substrates
It is abundantly clear that wurtzitic nitride polytypes dominate the wide bandgap nitride field. However, initial investigations of cubic GaN have taken place on GaAs substrates, particularly with MBE, owing to its availability and ease with which its surface can be cleaned with a combination of well-developed ex situ and in situ methods. This eventually gave way for the activity on GaAs to be primarily focused on dilute nitrides in GaAs and other conventional III–V compound hosts as discussed in Section 2.10. As for the wide bandgap cubic nitrides, the motivations are twofold. With superior electron and, particularly, hole mobilities and isotropic properties owing to the cubic symmetry and high optical gain in stimulated emission, the zinc blende phase is desirable over the wurtzitic phase. However, the sample quality is not comparable to the wurtzitic phase because of to a larger density of extended defects and their electrical activity, leading to a shift in the motivation to dilute N cubic III–V systems such as GaAsN, InGaAsN, and so on, where small quantities of N added to the lattice reduce the bandgap considerably, paving the way for GaAs-based emitters. For example, bandgap emission in the 1.3–1.5 mm range, which is very important in telecommunications, can be obtained in InGaAs : N lattice-matched or nearly lattice matched to the well-established GaAs technology. To underscore again, while a good deal of progress has been made with semiconductor laser reports in dilute GaAs : N and InGaAs : N, the zinc blende variety of GaN has been beset by the difficulty in producing low-defect content material. Several substrates discussed in this chapter such as GaAs, Si, 3C-SiC, GaP, and MgO can be used to grow zinc blende GaN, but gallium arsenide (GaAs) has emerged as the most widely used substrate for zinc blende GaN epitaxy because it is well developed and large-area substrates are commercially available. The GaAs(1 0 0) surface is used for the zinc blende phase, whereas the (1 1 1) face produces the wurtzitic phase. If the wurtzitic phase can be made pure, thick GaN layers can be grown on GaAs, which can then be selectively removed with ease to obtain freestanding GaN templates. Even a small amount of GaAs decomposition has adverse effects on zinc blende GaN epitaxy, as surface roughening or faceting enhances the onset of wurtzite growth. Because of this, MBE, which is a relatively lower temperature process, has been more commonly employed than OMVPE or HVPE. Some success has been achieved in increasing the maximum allowed temperature by encapsulating the GaAs substrate with GaN deposited at a low temperature, thereby making growth on GaAs by OMVPE and HVPE more viable.
j477
j 3 Growth and Growth Methods for Nitride Semiconductors
478
It is customary that the initial deposition of GaN is carried out at relatively low temperatures, typically in the range of 550–650 C, compared to that for epitaxial film growth. This GaN layer deposited at relatively lower temperatures can protect the GaAs surface during the actual growth, which takes place at relatively high growth temperatures, typically >700 C. Moreover, it does improve the quality of the zinc blende phase, in addition to enhancing GaN nucleation and initiating growth. Highresolution TEM and electron diffraction measurements show that the as-deposited GaN buffer layer at 550 C by HVPE on GaAs substrates is polycrystalline but becomes a single crystal after thermal annealing at 850 C for 10 min [13]. The thickness and growth temperature of this initial buffer layer has a pivotal role in the crystalline quality of the zinc blende GaN. In an MBE growth system with a radio frequency plasma discharge source, a two monolayer thick interfacial layer deposited at 600 C resulted in better quality without any wurtzite phase. In OMVPE experiments, a low-temperature GaN or AlN buffer layer was also grown in a temperature range of 550–575 C for several minutes up to a thickness of 20 nm, followed by growth of the main layer in a temperature range of 600–900 C. The buffer layer was also found to lead to two-dimensional GaN growth, in addition to protecting the GaAs substrate. As in the case of, and even worse than, GaN on sapphire, the large difference in lattice constants of GaN and GaAs causes a high density of extended defects, among which are the stacking faults. The MBE growth reported by Strite et al. [399] and OMVPE growth by Bae et al. [400] concluded that the majority of defects originated from disordered regions at the GaN/GaAs interface and propagated along {1 1 1} planes and their extent and nature depended on the N/Ga flux ratio with the nearstoichiometric condition bringing about the best result. The MBE method, as opposed to the OMVPE method, results in good crystal quality as expected, because it is a nonequilibrium growth method affording lower substrate temperatures to be employed, which prevent dissociation of GaAs. Although not well studied, the GaN growth temperature affects the phase of the epitaxial layer, with high temperatures (>850 C) leading to wurtzitic GaN. HVPE has been used to grow thick wurtzite GaN at high temperatures (1000 C) on GaAs(1 1 1)A with the GaN buffer layer grown at 850 C toward freestanding GaN. Although not studied in detail, atomically flat GaAs surfaces drastically suppress the wurtzite GaN phase. Nagayama et al. [401], who employed OMVPE, reported that the samples grown on GaAs(1 0 0) surfaces tilted toward ½1 1 0 showed an enhancement of the wurtzite domain on the ð1 1 1Þ face, whereas the wurtzite domain on the ð1 1 1Þ face was suppressed. The sample grown on the (1 0 0) surface tilted toward [1 1 0] showed equal levels of wurtzite domains on both the ð1 1 1Þ and ð 1 1 1Þ faces. The generation of the wurtzite domains was suggested to be suppressed by exposure of the ð1 1 1Þ or ð1 1 1Þ faces by the reduced thermal damage of the substrate surface. With a low-temperature GaN buffer layer on 2 miscut GaAs(1 0 0) by HVPE, Yang et al. [402] obtained a GaN epitaxy film containing mainly the zinc blende phase. This 2 miscut GaAs(1 0 0) without nitridation also led to a pure zinc blende GaN thin film deposited by MBE [403]. This may be owing to the arrangement of five GaN lattice spacings coinciding with four GaAs(1 0 0) lattice spacings. The zinc blende GaN
3.5 The Art and Technology of Growth of Nitrides
growth on patterned GaAs(1 0 0) substrates has also been investigated by employing OMVPE [404]. It may be that lateral epitaxial overgrowth is in play here reducing the dislocation density greatly, as has been the case on patterned sapphire and 6H-SiC substrates. The incorporation of wurtzite GaN was reduced by using ½0 1 1 oriented stripes to suppress the formation of (1 1 1)B facets. As mentioned earlier, one of the motivations for growth on GaAs is the potential for thick wurtzitic GaN followed by the removal of GaAs for a freestanding GaN template. To this end, GaAs(1 1 1) substrates are used for wurtzite GaN growth as opposed to GaAs(1 0 0) substrates, which are used for zinc blende GaN growth. Both GaAs(1 1 1) A and GaAsð1 1 1ÞB faces can result in single wurtzite GaN in MBE growth [405] and HVPE, the latter having been used for producing freestanding wurtzite GaN [406– 408]. Numerous pinholes were found to form in the GaN film when grown on a GaAs (1 1 1)B surface, possibly because of arsenic desorption, whereas a GaN layer with a mirrorlike surface was grown on a (1 1 1)A substrate [409]. 3.5.3 Growth on SiC: Nucleation Layers and GaN
As substrates, 6H- and 4H-SiC offer the advantage that the c-plane lattice constants, with <1% lattice mismatch for AlN, and 3.49 and 3.53% for 6H- and 4H varieties, respectively, for the basal plane of GaN. The thermal expansion coefficients associated with both polytypes of SiC are closer to those of GaN and AlN than in the case of sapphire, although tensile residual strain ensues, which is troublesome. Moreover, the 4H- and 6H polytypes of SiC are polar crystals in such a manner that N bond to Si is stronger on the Si-face of SiC than that of Al or Ga, which paves the way for obtaining solely Ga-polarity GaN if the surface of SiC is prepared well for epitaxy. Furthermore, the in-plane crystalline alignment, as well as the out-of-plane, of GaN is the same as that of the substrate, as opposed to the 30 in-plane rotation encountered on sapphire with the associated disorder and troublesome boundaries between the inevitable domains. Moreover, the thermal conductivity of SiC is much higher than that of sapphire and nearly twice that of GaN. The caveat with SiC is that the residual thermal mismatch is tensile, as in the case of growth on Si with all the other substrates giving rise to compressive strain (see Section 2.12.7), and GaN films tend to crack at a thickness of about 2 mm. The availability of both conducting and high-resistivity substrates bodes well for SiC to produce vertical conduction as in LEDs and lasers (avoids current crowding effects) and high-resistivity substrates as in FETs. Already, the best LEDs, lasers, and MODFETs (or HFETs) [410] are produced on SiC substrates. The early drawback of SiC surfaces being damaged during polishing has been eliminated with better SiC mechanical polishing techniques coupled with H, HCl etching techniques as discussed in detail in Section 3.2.3.2. Devices such as LEDs, injection lasers, and modulation-doped structures are reported on SiC, as detailed in Volume 3, Chapters 1–3. The early problems plaguing growth on SiC had to do with surface preparation and quality of SiC available. Associated with the surface preparation is the stacking order disparity between nitrides and SiC substrates, applicable to both 6H- and 4H-SiC
j479
j 3 Growth and Growth Methods for Nitride Semiconductors
480
varieties. Let us discuss the issues involved in depositing nitrides on SiC with a focus on stacking order. An examination of the stacking shows that 6H-SiC is two thirds cubic and one third wurtzitic and the 4H variety is one half cubic and one half wurtzitic, whereas nitrides are fully wurtzitic with stacking order AaBbAa along the cdirection, and in the cubic form, the stacking order is AaBbCcAa [411]. The particulars of stacking-related issues on SiC are discussed in Section 3.5.3.1. To achieve complete GaN film coverage on SiC, some sort of nucleation layer must be employed. The buffer layer takes on a new meaning also on SiC as it is also used for strain management to avoid cracking to the extent possible as discussed later in this section. The ability to form facets with somewhat more ease than with sapphire makes deposition of nucleation layers on SiC relatively easier, but again with the perennial issue of achieving complete coalescence. The nature and the deposition conditions of this nucleation layer differ significantly whether OMVPE or MBE is employed. In the former, both GaN and AlN nucleation layers can be used; the former having been attempted both at low and high temperatures, with the latter giving better results. The deposition temperature for both GaN and AlN nucleation layers in OMVPE is reasonably high (in the vicinity of 1000 C), as we will discuss in Section 3.5.3.2. However, in the case of MBE, the behavior of GaN and AlN directly on SiC is drastically different. Although AlN can be grown by MBE at very high temperatures for MBE standards (even comparable to those employed in OMVPE growth), the growth temperature of GaN directly on SiC in MBE must be kept below a certain level, about 800 C, to avoid complete desorption. 3.5.3.1 Stacking and Interfacial Relationship The discussion of nucleation and stacking of GaN on SiC critically depends on the surface atomic structure of SiC, which is discussed in Section 3.2.3.1. To gain an understanding of AlN and GaN growth, which have 2H stacking, on SiC, which has either 4H or 6H stacking, we must describe the possible surface structure of SiC and the atomic arrangement of AlN on SiC, with particular emphasis on the interface between the two. Detailed structural and electrical properties of stacking faults are discussed in Sections 4.1.4 and 4.1.6, respectively. Here, AlN is chosen because it is nearly lattice matched to SiC, thereby simplifying the picture for a clearer presentation. Unless the steps on 4H- and 6H-SiC are four bilayer and six bilayer steps, respectively, defects will form in nitride layers. In this vein, if we consider an interface step as shown in Figure 3.86, we will note that the step shown delineates two different surface terminations of the substrate, indicated as terrace 1 and 2. For a quantitative description of the step, the step height and the position of the surface termination in the stacking sequence of the substrate need to be well described [47]. The two crystals grown on the surface terminations ! associated with terrace!1 and 2 can be related to each other by a displacement vector d . If the displacement d relating the two surface terminations does not correspond to a symmetry operation of nitrides, a defect is ! generated. If so, the d vector can be reduced to the fault vector of one of the possible stacking faults. Furthermore, as shown in Figure 3.87, the epitaxy could begin at wurtzite or sphalerite positions on each of the two terraces leading to four possibilities, namely, two wurtzite interfaces, two sphalerite interfaces, and two cases with one
3.5 The Art and Technology of Growth of Nitrides
Terrace 1 d
A
C
Terrace 2
Si
Figure 3.86 Interface step and displacement vector at a (0 0 0 1) SiC surface; the area marked A under terrace 1 faces a purely wurtzite portion on terrace 2, it is considered as a faulted wurtzite sequence. Courtesy of Pierre Ruterana, Ref. [47].
2H-AlN
Interface 6H-SiC Sphalerite
Wurtzite
Si
Al
C
N
Figure 3.87 Wurtzite and sphalerite positions for the growth of AlN on (0 0 0 1) SiC. Courtesy of Pierre Ruterana, Ref. [47]. !
sphalerite and one wurtzite interface. The first two configurations do not modify d , whereas the two latter ones introduce a twin component. ! In case the displacement vector d does not correspond to a symmetry operation of epitaxial nitride layers at the interface, the area A can be considered a faulted sequence of the 2H polytype. In this case, a combination of stacking faults of the 2H polytype can result. The sum of the fault vectors of these defects can always be reduced to that of one of the three possible stacking faults, namely, I1, I2, and E of the hexagonal structure (for I1: 1/6 h3 2 2 0i, for I2: 1/3 h0 1 1 0i, or for E: 1/2 h0 00 1i), see Section 4.1.4 for a detailed description of stacking faults. Therefore, the dislocation character of all steps can always be identified as and reduced to one of these three vectors [47]. If we assume that all possible step configurations have the same probability of occurrence, we can then estimate the probability of the different defects that would occur. The results when a 2H polytype, such as GaN, is deposited on 6H-SiC, 4H-SiC, and 3C-SiC polytypes are summarized in Table 3.13 for wurtzite and general interfaces (sphalerite and/or wurtzite).
j481
j 3 Growth and Growth Methods for Nitride Semiconductors
482
Table 3.13 Electrical properties for Si-doped GaN films grown at 350 Torr on 6H- (on axis and vicinal) and 4H-SiC.
SiC substrate
AlN temperature ( C)
nRT (cm3)
nCV (cm3)
lRT (cm2 V1 s1)
n77 K (cm3)
l77 K (cm2 V1 s1)
6H 6H off 3.5 4H
1080 1080 980
8.9 · 1016 4.6 · 1016 9.7 · 1016
7 · 1016 2 · 1016 —
876 932 884
2.6 · 1016 1.3 · 1016 2.7 · 1016
1888 2438 1895
The SiC substrate used and the AlN growth temperature are listed in columns 1 and 2. The Hall electron concentration at room temperature and 77 K are listed in columns 3 and 6, along with the electron concentration determined by CV measurements listed in column 4. The electron mobility at 300 and 77 K are listed in columns 5 and 7. Courtesy of D. Koleske [434].
Let us now discuss the 2H/6H and 2H/4H interfaces. For the former, the results show that almost one step in two is of the type I1 and 25% of the steps do not generate defects. The E type steps occur only when the sphalerite interface is taken into account; this probability is lower than 10%. For the latter, only the single and four bilayer step, tetrahedra, are considered. For wurtzite interfaces, only I1 type steps occur and one step in two does not exhibit a dislocation character. The I2 and E type steps occur only when sphalerite interfaces are present. Considering the case of a 2H nitride semiconductor on 3C-SiC, a step height of six tetrahedra is sufficient to attain coincidence between the two lattices, namely, the SiC substrate and the nitride epitaxial layer. In this case, in contrast to the 4H-SiC or 6H-SiC substrates, the position of the step in the stacking sequence of the substrate is not critical for the defect type and the only height that does not lead to a dislocation character for the step is of six tetrahedra [47]. In short, although a substrate such as SiC may present a good lattice match and the same in-plane symmetry with the deposited material, numerous defects because of the structural differences can be generated at the interface. 3.5.3.2 Nucleation Layers on SiC Although not as difficult a step as in the case of growth on sapphire in terms of initiating growth on SiC, still the nucleation layer represents a critical step in growth of nitrides on SiC as it has implications not only on the quality of the subsequent GaN layer but, more importantly, also on cracking. Typically, GaN layers of more than 2 mm thick crack upon cooling down from the reactor and/or during processing. The cracking problem, therefore, dictated the developments in nucleation layers on SiC. Relevant to the case is the somewhat lack of in situ diagnostic capabilities in OMVPE to allow in situ buffer morphology observations. Suffice it to say that thin buffer layers of GaN on SiC, particularly the high-temperature varieties, do not fully coalesce. This is also applicable to MBE growth even for AlN unless proper surface treatment procedures are employed (see Section 3.2.3.2 for surface preparation of SiC for epitaxy) and suitable growth temperatures must be employed for a two-dimensional growth, as discussed in Sections 3.5.3 and 3.2.3.2.
3.5 The Art and Technology of Growth of Nitrides
Cracking of GaN layers is caused by thermal mismatch induced tensile strain and exacerbated by the better registry, for example, lack of in-plane rotation, between GaN or AlN and SiC, as compared to sapphire, between the film and the substrate. Cracking problem is an issue also on sapphire substrates, but it occurs for thicker layers as compared to those on SiC. Applicable to growth both on sapphire and SiC is the plastic relaxation (allowing the epitaxial layer to crack) followed by regrowth that relies on epitaxial lateral overgrowth [412]. The plastic relaxation method has also been applied to AlGaN on GaN [413] wherein thick AlGaN layers also crack, which has been a difficult problem in lasers as thick cladding layers of AlGaN are required for mode confinement. Various techniques have been developed to ameliorate stress and allow deposition of crack-free >4 mm thick AlGaN layers on sapphire. LT AlGaN interlayers have proven effective in suppressing cracks during AlGaN growth on a GaN template, but the quality of low-temperature AlGaN layers is not as good as those grown at high temperatures. The coefficient of thermal expansion (CTE) mismatch between the III nitrides and SiC leads to a calculated tensile stress of approximately 0.4 GPa for GaN and approximately 0.8 GPa for AlN upon cooldown from approximately 1100 C, typical growth temperature, down to 25 C, which is typically cited as a cause for cracking of AlGaN layers on SiC [414]. Additional details of residual stress on SiC and other substrates can be found in Section 2.12.7. Owing to the smaller thermal expansion coefficient for SiC than for the Wz GaN, the GaN films on SiC typically are under tensile strain, with the biaxial stress being, depending on the growth temperature, as large as 1 GPa. Optical and Raman spectroscopy studies of OMVPE-grown and MBE-grown GaN on SiC confirm they are under tensile stress, whereas those on sapphire substrates are under compression [415]. This suggests that the mismatch of the thermal expansion coefficient is the dominant factor determining the stress in the GaN films on SiC substrate. The residual tensile biaxial strain causes a red shift of the excitonic transition energy and reduces the valence band splitting as compared to strain-free bulk GaN [416]. Raman and photoluminescence measurements [417] indicated a red shift of 2.7 0.3 cm1/ GPa of the E 22 phonon and 20 3 meV/GPa of the excitonic transitions. The total measured stress at an arbitrary temperature is given by [418]: s ¼ si þ sth ;
ð3:60Þ
where si is the intrinsic stress, because of the lattice mismatch, and sth is the thermal stress given by sth ¼
E2 ða1 a2 ÞðT T 0 Þ; 1 v2
ð3:61Þ
where a1 and a2 are the thermal expansion coefficients of substrate and GaN films, respectively, and E2 is the Youngs modulus, and n2 is the Poissons ratio of the film. Temperature T0 is a reference temperature, and for the purpose in hand, T0 is the growth temperature. Depending on the growth temperature, the tensile thermal strain Da/a should be about 0.14% (about 0.08% for Dc/c) [419] and could be a factor of 10 smaller if the growth temperature is about 1000 C. However, the lattice mismatch of GaN is not
j483
j 3 Growth and Growth Methods for Nitride Semiconductors
484
very large as mentioned above (about 3.1%), and for many samples, the relaxation is only partial. Perry et al. [420] and Nikitina et al. [421] reported that the a and c lattice parameters for GaN on SiC are in the range of 3.1945–3.1850 and 5.1812–5.190 Å, respectively. The former values for both a and c lattice parameters lead to an a-plane strain of 0.26% and a c-plane strain of 0.07%. The latter values for both a and c lattice parameters lead to an a-plane and c-plane strains of 0.03 and 0.09%, respectively. For the ternary compounds where the lattice mismatch depends on the composition (13.5% for InN, 1% for AlN w.r.t. SiC), the balanced nature of the thermal strain and residual mismatch strain can be different, a topic that has not been fully explored. Unlike growth on sapphire, which will be discussed below, the nitride semiconductors grown on SiC reproduce hexagonal cells of the substrate with no rotation with respect to the substrate. The strain in GaN films is also affected by the nature of extended defects; and in the spirit of Ref. [422], a strain mechanism associated with the presence of defects has been introduced by Ahmad et al. [423]. According to this study, the dominant defects that correlate to stress relaxation in GaN films are the threading dislocations, which introduce a modeled free volume into the crystal. Assuming a strain in direct proportion to n, that is, eTD ¼ L20 n;
ð3:62Þ
where L0 is a length parameter related to the range surrounding a TD defect over which stress relaxation occurs. In this model, strain relaxation because of defects affects the area of L20 . The term n is the dislocation density in question. The strain in the epilayer perpendicular to the growth c-axis is then ð1 n2 Þ ea ¼ ð1 L20 nÞ 1 þ s 1; ð3:63Þ E2 where s is the biaxial stress. From this, and using de/dT ¼ (a1 a2), which is consistent with Equation 3.61, we obtain ds E 2 ða1 a2 Þ E2 ða1 a2 Þð1 þ L20 nÞ; ¼ dT 1 n2 ð1 L20 nÞ 1 n2
ð3:64Þ
where the approximation holds whenever L20 n 1. As in Equation 3.61, assuming the expansion coefficients to be nearly constant over the temperature range of interest, Equations 3.63 and 3.64 reduce to the in-plane strain given by ea ¼ sð1 n2 Þ=E 2 and Equation 3.61, respectively, when n ¼ 0. The differences in the sign for ds/dT are because of the ordering of the thermal expansion coefficients discussed in relation to Equation 3.61, namely, compressive strain leading to positive sign (such as that on sapphire) and tensile strain leading to negative sign (such as that on SiC or Si). The model described accurately preserves the signs of the stress and expected stress–temperature coefficient (ds/dT) owing to the effects of either tensile or compressive stress on the open volume associated with TDs. Thus, a linear dependence between the stress–temperature coefficient and n was suggested [423]. Related to this biaxial strain, a Poisson ratio of v ¼ 0.18 was determined by Perry et al. [420]. Other reports suggest that the picture is more complex and the sign of the
3.5 The Art and Technology of Growth of Nitrides
strain may be thickness dependent [424]. Attempts have been made to explain the conflicting strain results, which have been attributed to the influence of the AlN nucleation layers on the growth mode and strain relief. Waltereit et al. [425] observed that the strain of GaN layers on SiCwas determined by the growth mode,which, in turn, was governed to a larger extent by the degree of wetting of the substrate than by the lattice mismatch. They suggested that the tensile strain, which favors the formation of cracks in the GaN layer, can be avoided by growing GaN epitaxial layers on thin and thus coherently strained AlN nucleation layers. Other buffer methods such as AlGaN and also alternating layers of GaN/AlN can be used to mitigate the cracking issue. The particulars of thenucleation layers dependon whether the interface is active for current conduction. In vertical devices, this is the case and graded AlGaN layers and/or very thin alternating layers would have to be used. Additional sources affecting stress, thus cracking, result from changes in film morphology during growth. This, however, is rather complex and its effect is not known precisely. Attempts have been made to investigate stress by in situ optical curvature measurements during OMVPE growth of GaN, AlN, and AlGaN [426,427]. In situ stress measurements during deposition of AlGaN layers on LT AlN nucleation layers on sapphire revealed stress to be initially compressive, before segueing into relatively stress-free film with increasing thickness [428]. On the contrary, AlGaN on LT GaN nucleation layers started with tensile strain, similar to growth on thick GaN. The sign of the initial stress in both cases was ascribed to epitaxial mismatch. Incorporation of LT AlGaN interlayers has been reported to reduce the tensile strain during growth of AlGaN on thick GaN [429,430] owing to a partial relaxation of strain redefining, which modifies the in-plane lattice parameter. The overall message here is that by designing the nucleation layer and structural parameters, strain state could be mitigated to allow thicker layer thicknesses to be obtained before cracking occurs. Because cracking is a more serious problem on SiC (as well as on Si and AlGaN) because of thermally induced tensile strain, much of the nucleation layer growth on SiC is discussed in the context of cracking. Even though the low-temperature GaN initiation layers in OMVPE have been attempted on SiC, they have not been all that successful in terms of quality. The same holds for lowtemperature AlN nucleation layers. In contrast, both GaN and AlN (also AlGaN) hightemperature nucleation layers in OMVPE have been successful. In MBE, the situation is different in that GaN initiation layers must be performed at moderate temperatures with MBE standards, whereas AlN initiation layers could be achieved at moderate to high temperatures. High-temperature AlN nucleation layers on SiC in OMVPE growth, and to some extent in MBE growth, may be more reproducible compared to a low-temperature AlN nucleation layer typically used on sapphire. However, as in the case of the nucleation layers on sapphire, attainment of the proper AlN or GaN NL grain structure, with the eventual goal of complete coalescence, is crucial for high-quality GaN top layers [431] and the growth of high-resistivity GaN is needed as buffer layers for FETs [432]. As mentioned in Section 3.2.3.2, unless the damage caused by mechanical polishing on SiC surface is removed, the layers grown on SiC may not be comparable with those on sapphire, as was the case until the advent of better surface preparation techniques.
j485
j 3 Growth and Growth Methods for Nitride Semiconductors
486
The analysis of George et al. [433] of the OMVPE-grown GaN layers on SiC and sapphire suggested that the subsequent GaN quality primarily depends the nucleation buffer layer, which, in turn, is determined by some important factors such as interfacial bonding rather than the lattice mismatch. Koleske et al. [434] found that the dependence on the substrate temperature of the nucleation layer was of paramount importance and electron mobilities of 876, 884, and 932 cm2 V1 s1 were obtained on 6H-SiC, 4H-SiC, and 3.5 off-axis 6H-SiC, respectively, by optimizing that temperature to 300 K. An issue of interest is the way the LT AlN layers grow on SiC. An AlN buffer layer deposited on sapphire at low temperature is likely to form large grains during the annealing temperature ramp, which improves the structural quality of the buffer layer and thus the structural quality of the subsequently deposited layer. But the sequentially deposited GaN films on AlN on SiC have inferior quality compared to those on AlN buffer layer deposited at high temperature (HT). This observation is inconsistent with layers grown on sapphire where annealed LT buffer layers result in higher GaN quality than that of HT buffer layer in the case of OMVPE [435]. Similarly, the postgrowth thermal processing of thin (200 nm) AlN films grown by MBE at high temperature, in the range of 1200–1400 C, has shown to markedly improve the quality [436]. To restate, unlike sapphire substrates, the successful nucleation layers, both AlN and GaN varieties, on SiC substrates are deposited at relatively high temperatures in an OMVPE environment. In the effort by Koleske et al. [434], the SiC substrates were subjected to an in situ treatment at 1200 C for 10 min in 2 standard liters per minute (slm) H2 at 50 Torr. Next, approximately 100 nm of AlN was grown at substrate temperatures ranging from 925 to 1190 C using 1.2 mmol min1triethylaluminum (TEA), 1.25 slm NH3, and 1 slm H2 at a system press sure of 50 Torr. Koleske et al. [434] converged on an AlN NL deposited at 1080 C on-axis and 3.5 off-axis 6H-SiC, while an AlN NL temperature of 980 C was used for 4H-SiC.The surface topology as inspected by AFM of the AlN NL grown at 1080 C exhibited smaller AlN grains on the 6H-SiC than those on 4H-SiC, suggesting that the AlN morphology influences GaN film formation and subsequent electron mobility. Cross-sectional transmission electron microscopy (XTEM) measurements revealed the absence of screw dislocations in the AlN and a low screw dislocation density near the AlN/GaN (the layer grown on top of the AlN nucleation layer) interface, which would in part account for the high electron mobilities achieved in these films. 3.5.3.2.1 High-Temperature AlN Nucleation Layers on SiC It is a common experience that the AlN nucleation layer growth temperature influences optical properties and electron mobility (termed mRT at room temperature). Consequently, two growth temperatures, namely, 980 and 1080 C, were explored for each type of SiC substrate in the work of Koleske et al. [434]. For comparative purposes, a 100 nm thick AlN layer was grown at 1080 C on 6H- and 4H-SiC simultaneously. Clear differences are apparent in the AlN morphology as shown in the AFM images in Figure 3.88. In Figure 3.88, the AlN grain size (approximately 100 nm) grown on 6H-SiC is approximately half the size of the AlN grain size (approximately 200 nm) on 4H-SiC. Moreover, the AlN grains on
3.5 The Art and Technology of Growth of Nitrides
Figure 3.88 AFM images of the AlN film growth on (a) 6H-SiC from CREE and (b) 4H-SiC from Sterling. Both AlN films were grown to a thickness of 100 nm at 1080 C in the same growth run. The AFM vertical scale for (a) is 15 nm and (b) 10 nm per division. Courtesy of D. Koleske and after Ref. [434].
6H-SiC are more rounded and rougher (z height scale ¼ 15 nm), whereas the AlN grains on 4H-SiC have flatter tops as indicated by the smaller z height scale ¼ 10 nm in Figure 3.88. A point that was made in Section 3.2.3.2 in relation to surface preparation is in effect here also in that a major difference between the 6H- and 4H-SiC lies in the surface polish, where the 6H-SiC had visible polishing scratches, whereas the 4H-SiC had a smoother surface morphology. The epitaxial GaN layer exhibiting the highest mobility was achieved for the AlN nucleation layer grown at 1080 C (mRT ¼ 738 cm2 V1 s1) on 6H-SiC and at 980 C (mRT ¼ 743 cm2 V1 s1) on 4H-SiC [434]. The growth temperature dependence on 6H-SiC was more notable than on 4H-SiC, which in part may be a reflection of technological issues regarding substrates and their surface preparation. For lower AlN growth temperatures on 4H-SiC, the room temperature carrier concentration, nRT, is 1.5 · 1017 cm3 and decreases gradually to 7 · 1016 cm3 as the AlN growth temperature is increased. Because identical GaN growth and Si doping conditions were used on the AlN nucleation layers, the decrease in nRT with increase in the AlN growth temperature suggests increased compensation rather than a reduction in unintentional impurities because a decrease in mobility accompanies a decrease in electron concentration. In another report, which deals with transport but uses the FET performance and X-ray diffraction as a measure, the effect of the particulars of AlN/SiC nucleation layers on the quality of subsequent GaN/AlGaN FET structures was assessed. Green et al. [437] used high-temperature AlN nucleation layers grown between 1030 and 1080 C. Variation of the nucleation temperature, V/III ratio, and thickness were cited to have a dramatic effect on the balance between edge, screw, and mixed character dislocation densities as inferred rather indirectly from the X-ray data. The character of dislocations was established primarily through comparison of rocking curves associated with symmetric (0 0 0 2) and asymmetric ð1 0 1 2Þ and ð2 0 2 1Þ diffractions
j487
j 3 Growth and Growth Methods for Nitride Semiconductors
488
and, moreover, the screw dislocation component, as measured by the width of the symmetric X-ray diffraction peak widths. Green et al. [437] constructed three series of samples to optimize the AlN NL focusing on reactant flow rate, substrate temperature, and nucleation layer thickness. To probe the effect of reactant flow, three samples were grown with NH3 flow/TMA flows of 30 slm/10 sccm, 60 slm/20 sccm, and 90 slm/10 sccm, respectively, while keeping the substrate temperature at 1080 C and the layer thickness at 80 nm. The substrate temperatures of 1030, 1050, 1070, and 1080 C were employed to determine the effect of the deposition temperature, while keeping the NH3 flow/TMA flows at 30 slm/10 sccm and the layer thickness at 80 nm. The layer thicknesses of 40, 80, and 160 nm were employed to test the effect of the nucleation layer thickness while keeping the NH3 flow/TMA flows at 30 slm/10 sccm and deposition temperature at 1070 C. Increasing the NH3 flow rate decreased the growth rate of AlN, despite increasing TMA flow that points to gas-phase reactions that cause a reduction in the TMA delivery to the substrate surface. As is always the case in nucleation layer growth by OMVPE under suitable conditions, increasing the thickness of AlN layer led to smoother AlN nucleation layer surface. Decreasing both increasing AlN thickness and nucleation temperature led to an increase in the edge dislocation density, as measured by broadening of the asymmetric X-ray diffraction peak. Moreover, the screw dislocation component, as measured by the width of the symmetric X-ray diffraction peak, was relatively insensitive to the parameters studied. Further, the edge dislocation density had an impact on the trapping characteristics in FETs, which also manifested itself as a decrease in the drain saturation current under thermal stress. Among the samples with varying deposition temperature, the layer deposited at 1070 C had the smoothest surface (the NH3 flow/ TMA flows at 30 slm/10 sccm and the layer thickness is 80 nm). The lowest temperature sample, 1030 C, was very granular, the grain size of which increased when the deposition temperature was raised to 1050 C. For a deposition temperature of 1080 C, the grain size decreased when the density increased. In terms of the thickness series, the nucleation layer with a thickness of 160 nm exhibited the best surface morphology with nearly complete coalescence. The sample with 40 nm thickness was very granular, giving way to reasonably complete coalescence as the thickness was increased to 80 nm (the NH3 flow/TMA flows at 30 slm/10 sccm and deposition temperature at 1070 C). As is always the case with OMVPE, atomically smooth GaN surface resulted when several microns of the material was grown on the above-mentioned nucleation layers. This is a fluidic field and is in constant flux. The trend is to deposit the nucleation layer at as high a temperature as possible while tolerating discontinuous (incomplete) layer coverage. The subsequent layer growth conditions are then optimized to achieve coalescence rapidly. The structural and therefore electrical/optical properties thus obtained are superior to those grown on what is still called the high-temperature nucleation layers but performed at relatively lower temperatures. 3.5.3.2.2 Low-Temperature GaN Nucleation Layers on SiC As mentioned above, low-temperature AlN and GaN nucleation layers have not been as successful on SiC as the high temperature varieties. Nevertheless, attempts have been made to optimize
3.5 The Art and Technology of Growth of Nitrides
surface morphology of low-temperature GaN nucleation layers on the vicinal surfaces of 6H-SiC(0 0 0 1) in an OMVPE environment [438]. In addition, cross-sectional transmission electron microscopy, Hall measurement, and photoluminescence spectra were obtained in the subsequent GaN layers to draw conclusions about the effect of the nucleation layer. The optimum growth condition for a GaN nucleation layer on SiC(0 0 0 1) was determined to be 1 min of growth at 550 C within the scope of low-temperature GaN nucleation layers. Low-temperature GaN nucleation layers growth at 565 C on SiC attempted in authors laboratory displayed featureless surface as opposed to a nanoporous surface typically observed for high-temperature varieties. XRD analysis led to >20 and >30 arcmin HWFM values for (0 0 2) and (1 0 2) X-ray rocking measurements, respectively. The subsequent high-temperature GaN epilayer grown on this buffer layer had good surface morphologies but exhibited only marginally improved X-ray data. It should be stated that the low-temperature buffer layer/SiC (i.e., 600 C) leads to high defective interface between the epilayer and the substrate that precludes any device application relying on the quality of that interface. Presented only for completeness, it is clear that the low-temperature GaN and also AlN nucleation layers on SiC are not successful as compared to the high-temperature varieties. 3.5.3.2.3 High-Temperature GaN Nucleation Layers on SiC Unlike the low-temperature GaN nucleation layers, high-temperature nucleation layers of the GaN, AlGaN, and AlN varieties have all been successful and which one is used depends very much on the particulars of the device performance. If GaN is grown using typical growth parameters including the substrate temperature, for example, 1080 C, the resultant GaN nucleation is 3D, which leads to the formation of large islands presenting hexagonal truncated shape with f1 1 0 1g lateral facets and a top {0 0 0 1} facet [439]. Lahreche et al. [439] developed a three-step growth process for growing high-quality mirrorlike GaN layers with only GaN layers in the structure, the growth sequence of which is as follows: During the first step, a thin 3D GaN layer (100–130 nm) is deposited at a substrate temperature in the range of 980–1080 C. The second step involves growth interruption during which the 3D GaN layer is annealed at 1080 C under an ammonia flux, which resulted in a smooth surface. The third step involves the growth of main GaN layer at 1080 C. In situ optical reflectivity (higher intensity corresponding to smoother surfaces) as well as ex-situ SEM images were used to monitor and investigate the surface morphologies of the resultant films. Figure 3.89 shows the SEM images of the as-deposited 100 nm nucleation layer whose topological features can be described as 3D growth and that of a similar nucleation layer after the annealing in ammonia flow at 1080 C. In authors laboratory, high-temperature GaN initiation layers have been developed for growth of GaN on SiC. Figure 3.90 shows SEM images of (a) 100 nm HT GaN buffer layer grown on H2-annealed 6H-SiC and (b) 5 mm thick HT GaN epilayer grown on the 100 nm HT GaN buffer. The growth details are as follows: 3 min annealing at 950 C under H2 and NH3 and then 100 nm GaN buffer layer growth at 950 C (V/III 2000 and pressure 30 Torr) and then GaN epilayer growth at 1050 C
j489
j 3 Growth and Growth Methods for Nitride Semiconductors
490
Figure 3.89 SEM image of a 100 nm GaN nucleation layer as grown on 6H-SiC substrate at 980 C (a) and that of a similar layer but after annealed under ammonia flow at 1080 C during growth interruption (b). After Ref. [439].
(V/III 2000 and pressure 30 Torr). The crystalline quality of 100 nm GaN buffer layer was 5 and 13.2 arcmin for the (0 0 2) and (1 0 2) X-ray rocking FWHM, respectively. The surface of 100 nm GaN HT buffer layer shows clear signs of incomplete coalescence. The surface morphology can be controlled by the growth conditions such as pressure, temperature, V/III ratio, and thickness. The 5 mm GaN epilayer showed 4.7 and 8.5 arcmin for the (0 0 2) and (1 0 2) X-ray rocking FWHM, respectively. Figure 3.90b showed the crack generated on the GaN surface. The initiation layer of similar thickness to the one shown in Figure 3.90a has been grown at 980 C by Lahreche et al. [439] and shown not to be contiguous either, but after annealing under ammonia flow at 1080 C, the layer became smoother and importantly configured. It is therefore reasonable to conclude that this annealing occurs automatically prior to the growth of the higher temperature main GaN growth, as is the case in Figure 3.90b.
3.5 The Art and Technology of Growth of Nitrides
Figure 3.90 SEM images of (a) 100 nm HT GaN buffer grown on 6H-SiC and (b) 5 mm thick HT GaN epilayer grown on the 100 nm HT GaN buffer, which shows cracks thermal mismatch.
3.5.3.2.4 Alloy and Multiple Layer Nucleation Layers on SiC Having considered GaN nucleation layers, let us now consider AlGaN and AlN and combination of GaN, AlGaN, and AlN layers, high-temperature nucleation layers. Poisson et al. [440] used FET performance as a means of comparing GaN and AlGaN nucleation layers deposited at about 985 C to delineate as to which one is better for that particular application. A maximum drain current of approximately 1 A mm1 has been measured for devices with GaN nucleation layers as compared to 1.3 A mm1 obtained for devices with AlGaN nucleation layers. Moreover, in this single-step process, TEM analyses revealed that the GaN/SiC interfaces grown at 985 C are sharp with steps originating from the substrate misorientation, but not so when grown at higher growth temperatures >990 C, indicating a three-dimensional growth process. In response to the cracking of GaN films associated with growth on foreign substrates, particularly SiC substrates, because the lattice mismatch and thermal mismatch parameters are endemic and cannot be changed, Acord et al. [441] investigated the effects of growth conditions on stress evolution during OMVPE growth of AlGaN on SiC using in situ stress measurements based on wafer bowing measurements [442], dubbed multibeam optical stress sensor (MOSS) system [443]. An initial compressive stress was measured during AlGaN growth on thin AlN buffer layers, which evolved with film thickness into a tensile growth strain. The thickness at which the transition occurred was found to depend on the buffer layer growth conditions and on the Al fraction used in AlGaN. Increase in the AlN buffer layer V/III ratio from 750 to 10 600 led to an increase in the AlGaN initial compressive stress from 1.9 to 8.7 GPa, which reduced the film strain in the end being tensile. GaN was not observed to transition into tensile growth stress within the 3.8 m thickness grown, whereas increasing the Al fraction ushered in a more rapid change in tensile growth. The observed stress evolution has been explained in terms of an initial compressive growth stress that changes to a final tensile stress owing to morphological evolution during growth. Both the effects of V/II ratio and the composition of AlxGa1xN in the range of 0 x 0.63 were investigated in terms of stress evolution. In terms of the effect of the nucleation layer/buffer layer structure, GaN grown on AlN (80 nm)/SiC exhibited compressive strain (negative slope in the plot of stress thickness product versus layer GaN thickness) with increasing GaN thickness, which
j491
j 3 Growth and Growth Methods for Nitride Semiconductors
492
gradually changed with the GaN top layer thickness reaching a stress thickness product value of 1 GPa mm at 4 mm GaN layer thickness. GaN grown on Al0.35Ga0.65N/GaN (70 nm)/AlN (80 nm)/SiC was under tensile strain and exhibited a thickness product value of 2 GPa mm at only 1 mm GaN layer thickness. On the contrary, GaN layers grown on Al0.38Ga0.62N/AlN (80 nm)/SiC begin with compressive stress, which gradually begins to change slope, indicating heading toward tensile strain, reaching nearly zero stress at 1 mm. The lattice mismatch between the nucleation/buffer layer and GaN epitaxial layer determines the initial sign of the stress. The implication here is that the strain balance is in effect here and might provide a means of designing a nucleation layer buffer combination with minimal stress if not zero. Acord et al. [441] also examined the effect of the V/III ratio, which was varied between 750 and 10 600, during the growth of the AlN buffer layer on the initial incremental stress and transition point (the thickness at which the incremental stress is zero) during AlGaN growth. The AlxGa1xN epilayer composition was maintained at x ffi 0.44 (strain em ¼ 1.3%), and film growth continued until a tensile incremental stress was observed. The choice of the alloy mole fraction was made to coincide with the solar-blind region (280 nm) of the optical spectrum. During Al0.44Ga0.56N growth on 80 nm AlN/SiC, transition point thickness (at which the stress is zero) increased from 0.3 mm for a V/III ratio of 750 to 1.5 mm for a V/III ratio of 10 600. The gist of this discussion is that both the layer structure and the V/III ratio, the latter affecting the topological properties of the layer surfaces, are important parameters for reducing strain/stress in the overall structure for avoiding cracks. 3.5.3.2.5 Nucleation Layers on SiC by MBE Buffer layers deposited by MBE represent a good laboratory to examine the intricacies of the initial states of growth on SiC, which is intricately dependent on SiC surface preparation before introducing the sample into the MBE chamber and in situ processes once in the vacuum system. Owing to vastly divergent surface mobility and desorption rates of Ga and Al from SiC surface, the conditions used for growth of these binaries are vastly different. It should also be mentioned that AlN on SiC has been considered as a nearly lattice-matched dielectric layer for MISEFTs and not simply a nucleation layer on which to grow [44]. Assuming atomically smooth and chemically clean SiC surface, GaN grown under Ga-rich conditions at temperatures near approximately 750 C, which varies from system to system because of inaccuracies in the substrate temperature measurements, assumes a 3D growth mode as illustrated in Figure 3.91a. If the deposition temperatures is approximately at or below 650 C, a 2D-like growth is obtained as shown in Figure 3.91b. While the GaN in the 2D mode can be assumed coherent up to the relaxation thickness, defects generate readily afterwards. At these low temperatures, which are accompanied by Ga-rich conditions, edge dislocations are generally abundant, as shown in Figure 3.78. Raising the growth temperature at the expense of walking toward the 3D mode could reduce the total density of dislocations. In addition, the 3D mode would eventually give way to coalescence at which point defects are formed and quasi-2D mode would then ensue. If the substrate temperature is raised beyond 800 C, again depending on the system owing to the complexity
3.5 The Art and Technology of Growth of Nitrides
GaN or AlN 3D nuclei Compressive strain
Defects
SiC substrate (a)
AlN 2D nuclei
Compressive strain
Coherent coalescence
SiC substrate (b)
Interface wetting
Figure 3.91 (a) Schematic representation of GaN growth on SiC (0 0 0 1) indicative of 3D islandic growth caused either by higher than optimum growth temperature in the case of GaN and silicate adsorbate in the case of AlN. Defects depicted by circles form when coalescence place; (b) A 2D growth results when the optimum
deposition temperature is used in the case of GaN and silicate free surface is obtained by Ga deposition at 600 C followed by silicate layer decomposition during ramp-up to 1000 C growth temperature in the case of AlN. Courtesy of N. Onojima.
in measuring the substrate temperature, no growth would occur. Strain also could play a role in the deposition kinetics as has been pointed out [425]. The case for AlN is different in many ways in that much higher temperatures can be used as Al evaporation and desorption are not as severe as in the case of GaN. Despite this stated difference, if the surface of SiC is not free of the silicate adsorbate layer, see Section 3.2.3.2 for details, a 3D nucleation would still occur similar to the depiction in Figure 3.91a where the lines indicate the crystal lattice and divergence in the vertical direction indicates relaxation of the epitaxial layer as it has a larger lattice constant than AlN (holds for GaN as well). However, if the surface of SiC is O free as a result of decomposition by Ga flux (Ga is typically deposited at low temperatures such as 600 C and the substrate temperature is ramped to the deposition temperature of 1000 C during which the silicate adsorbate is fully removed [44,444]), 2D growth commences in islands and coalescence occurs leading to a coherent growth that is maintained up to the critical thickness. The lattice mismatch between AlN and SiC is very small and if 2D growth conditions are used, coherent films up to the critical thickness can be obtained, beyond which the film relaxes to its natural lattice constant. This transition can be
j493
j 3 Growth and Growth Methods for Nitride Semiconductors
494
monitored by RHEED rod spacing as the growth progresses and/or by measuring the lattice parameter of several films below at or near and beyond the critical thickness. The transition region is usually not abrupt as many factors go into determining when the epitaxial layer should relax. For example, the growth temperature and the SiC substrate surface preparation have been shown to affect the relaxation process. About 30 and 20 nm critical thicknesses have been obtained at 900 and 1000 C, respectively. As for SiC surface preparation, c-lattice parameter measured as a function of thickness for AlN grown on SiC prepared by HCl at 1300 C in a CVD system, HF dip, prior to both the introduction into the deposition system and in situ Ga-assisted desorption of the silicate adsorbate. As shown in Figure 3.92, the HCl treatment alone led to 3D growth mode and smallest average layer thickness to reach relaxation, the HF treatment also led to 3D growth and average thickness slightly more than the critical thickness, and the Ga treatment led to 2D growth that led to a graded transition to not reaching complete relaxation even at 300 nm. The structural properties of AlN layers as determined by (0 0 0 2) diffraction, which is sensitive to screw dislocations, and ð0 1 1 4Þ diffraction, which is sensitive to edge dislocations, have also been examined. As shown in Figure 3.93, the (0 0 0 2) diffraction width, which is sensitive also to the c-plane order, gets worse with layer thickness, with HCl-only treatment leading to 3D growth showing the steepest worsening. The in situ Ga-assisted desorption of the silicate adsorbate is the best in that the degradation of the (0 0 0 2) diffraction peak is minimal with increasing thickness. The ð0 1 1 4Þ bears the same message in that HCl treatment alone is the worst, HF dip is next, and Ga-aided surface cleaning leads to narrowest peak, which remains low with increasing thickness. Measured by HRXRD
c-axis lattice constant (Å)
5.01
Coherent 5.012 Å
5.00
Ga(2D) HCl(3D)
4.99
HF(3D)
AlN bulk 4.98
4.97 0 10
4.982 Å
1
2
10 10 AlN layer thickness (nm)
3
10
Figure 3.92 Dependence of AlN c-axis lattice constant on layer thickness for three different cases; namely, 1300 C surface exposure of SiC to HCl gas, HF dip of the substrates, and in situ Gaassisted desorption of the silicate layer remaining on the surface. Courtesy of N. Onojima.
3.5 The Art and Technology of Growth of Nitrides
800
FWHM (arcsec)
XRC(0 0 0 2)
600
HCl(3D)
400
Ga(2D) 200 0 0 10
10
1
10
2
10
3
AlN layer thickness (nm)
FWHM (arcsec)
3000 XRC(0 1 1 4)
HCl(3D) 2000
1000
0 0 10 (a)
HF(3D)
Ga(2D)
10
1
10
2
10
3
AlN layer thickness (nm)
Figure 3.93 The half width at full maximum of the (0 0 0 2) diffraction, which is sensitive to screw dislocation and c-plane ordering across the wafer (a); and ð1 0 1 4Þ diffraction that is sensitive to edge dislocation for AlN (b) with varying thickness grown on SiC prepared with HCl gas, HF dip and Ga-assisted desorption of the silicate layer from the surface. Courtesy of N. Onojima.
To complete the characterization, surface morphology, as imaged by AFM, for an AlN sample grown to a thickness of 30 nm on a substrate cleaned with Ga-assisted silicate removal was taken. In addition, the same was taken for a film thickness of 290 nm, again grown on SiC subjected to Ga-assisted removal of the silicate adsorbate. Both images are shown in Figure 3.94 where the thinner films indicate fully 2D growth replicating the atomic steps on the substrate surface. The thicker layer, on the contrary, shows a step-flow growth pattern. The dark spots are associated with dislocation terminations. 3.5.3.3 Substrate Misorientation and Domain Boundaries Vicinal SiC substrates, misoriented from the (0 0 0 1) plane, can enhance twodimensional growth of epitaxial GaN by reducing the distance that atoms must diffuse on the substrate surface and adatom surface lifetime before incorporation, though results contrary to this premise have been reported [445]. This reduces the
j495
j 3 Growth and Growth Methods for Nitride Semiconductors
496
Figure 3.94 Initial AlN layer grown at 1000 C is coherent and within the realm of 2D growth mode replicating the atomic steps on the SiC substrate surface (a on the left), which gives way to step flow growth as the deposition progresses (b, on the right). Courtesy of N. Onojima.
adatom lifetime on the surface before incorporation as well. In addition, a misorientation by 4 can relieve the stress at the interface between the SiC substrate and GaN films [457]. In addition to edge, screw, and mixed dislocations, which propagate throughout the epitaxial layers, defects such as IDBs and double positioning boundaries (DPB) could also be present in GaN unless grown on a substrate with similar stacking order. The growth of GaN on SiC substrates will be used below as the forum to discuss IDBs and DPBs. It should be mentioned that SMB (a special kind of double positioning boundary) are inevitable because of the steps on the substrate surface. If special care was taken to initiate growth with only one species and the substrate surface did not contain any steps, then IDBs would not form. However, substrates invariably contain steps and, even with single-species initiation, the lattice inverts across each single-step substrate. For example, in 6H-SiC, the stacking sequence is AaBbCcAaCcBb, so three stacking sequences would be possible for wurtzite GaN: (1) substrate . . .AaBbCc leading to BbCcBb GaN; (2) substrate . . .BbCcAa leading to CcAaCc GaN; and (3) substrate CcBbAa leading to BbAaBb GaN. All of these leading to vertical defects, between differently stacked domains in wurtzite GaN; they are named SMB. IDBs can be avoided by ensuring that the steps on 6H-SiC and 4H-SiC are six and four bilayer steps, respectively, which can be done with H polishing and possibly followed by a slight KOH etch. A high-temperature (1300–1500 C) HCl etching to accomplish the same has been reported by Powell et al. [36] and later by Nakamura et al. [446].On 6H-SiC, SMBs would always be created for two- and four-bilayer steps, and with a 2/3 possibility for one-, three-, and five-bilayer steps, but never for sixbilayer steps [447]. The SMBs are results of individual island growth and coalescence. SMBs must be considered as inevitable during growth of GaN on nonisomorphic substrates such as SiC.
3.5 The Art and Technology of Growth of Nitrides
Consider a 6H-SiC substrate region terminated as . . .AaBbCc and an adjacent region separated by a bilayer step terminated as . . .AaBbCcAa. With subsequent wurtzite GaN growth, the stacking sequence BbCcBbCc would form over the first region, whereas the stacking sequence CcAaCcAa would form over the second region and inevitably lead to an SMB, as illustrated by the arrow labeled S1 in Figure 3.95. Two important points must be made. First, a surface bilayer step on a wurtzite substrate would not lead to SMB because subsequent wurtzite growth would have the same stacking sequence on both sides of the bilayer step. Second, 4H- and 6H-SiC are
Figure 3.95 (a) High-resolution electron micrograph of wurtzite GaN grown by plasma-enhanced MBE on 6H-SiC showing substrate steps and an associated SMB. (b) Cross-sectional ball and stick model of wurtzite GaN on 6H SiC. Steps on SiC surface such as S1 are likely to create stacking mismatch boundaries whereas others such as S2 will not.
j497
j 3 Growth and Growth Methods for Nitride Semiconductors
498
a mixture of cubic and hexagonal stacking and not every substrate bilayer step would lead to an SMB. For 6H-SiC, consider the substrate terminations . . .AaBbCcAa and . . .AaBbCcAaCc separated by a bilayer step. The first would lead to CcAaCcAa GaN, whereas the second would lead to AaCcAaCc GaN. Clearly, there would be no SMB at this substrate step, as illustrated by the arrow labeled S2 in Figure 3.95. The nature and particulars of SMBs and IDBs are discussed in Section 4.1.3. Wurtzite nitride growth on sapphire is more complex because the exact interface arrangement of atoms is not known and sapphire has a coordination of atoms completely different from GaN. However, if we suppose that nitrogen atoms attach in the same way to each uppermost Al atom of the substrate during the so-called nitridation of the sapphire surface, then each step on the sapphire substrate has a high probability of producing an SMB owing to the position change of Al atoms in successive basal planes. One may then suggest that every bilayer surface step, on a close-packed nonisomorphic substrate of wurtzite GaN, has a high probability of creating an SMB threading defect during wurtzite GaN epitaxy. The above-discussed viewpoint provides a fresh perspective on the notorious problem of finding a good substrate for GaN growth. For the reason of stacking-order match and close lattice match to GaN, it is tempting to state that a ZnO substrate would be superior to a ZnO buffer layer on sapphire for subsequent GaN growth. Because none of the presently known growth methods on available substrates completely eliminates the problem of SMBs in wurtzite GaN, we will briefly consider possible deleterious effects of these defects on the properties of epitaxial films. An SMB originating from a linear step is essentially two dimensional and the equivalent density of defects averaged over the GaN volume could be as high as 1018–1019 cm3 if we estimate the linear density of these defects from TEM micrographs. The geometry of atoms and unsaturated bonds at SMB is not presently known. Extensive theoretical work is needed to understand the electronic character of these defects. 3.5.3.4 Polarity of GaN on SiC Growth on the Si-face of SiC(0 0 0 1) tends to produce Ga-polarity GaN regardless of whether AlN or GaN buffer layers are used. Polarity control afforded by SiC substrates could be a key advantage of SiC as compared to sapphire on which both Ga- and N-polarity GaN can be deposited, depending on the choice of the buffer layer. In addition, mixed polarity layers could also result, as is the case in predominantly N-polarity GaN on sapphire. The fact that the Si–N bond is stronger than any Si–Ga bond leads to GaN initiating with the N-layer first, leading to Ga-polarity GaN. The polarity of GaN on SiC has been studied by a number of techniques with some inconsistencies in the literature. The argument that Ga-face GaN grows on Si-face SiC and N-face GaN grows on C-face SiC could form the basis of argument [448]. Sasaki and Matsuoka [449] concluded that the epitaxial GaN layers on (0 0 0 1)Si and (0 0 0 1)C SiC were terminated with nitrogen and gallium, respectively. Most results are consistent with the standard framework, though the antistandard results still deserve serious attention. The stronger Si–N bond argument is supported by the electronic structure calculations of the (0 0 0 1) interface, which indicate that the stronger bonds are at
3.5 The Art and Technology of Growth of Nitrides
the Si–N interface [450]. Along the same lines, Ren and Dow [451,452] considered the lattice arrangement at the interface between SiC and GaN employing a tight-binding model. They concluded that the GaN grown on a C-terminated SiC(0 0 0 1) substrate has a local lattice mismatch of 6%, whereas the GaN grown on the Si-terminated surface has local lattice mismatch of <3%. By comparing the high-resolution electron microscopy image and simulated image of the atomic arrangements at a GaN/SiC interface, Stirman et al. [453] concluded that the atomic arrangement at the GaN/SiC interface most likely consists of N bonded with Si. However, they also argued that some Ga bonded to C, by replacing the surface Si with Ga as shown schematically in Figure 3.96, might be present to maintain the charge balance. 3.5.3.4.1 GaN Growth on SiC Despite the cost of SiC and owing to the continually improving quality, development of surface preparation methods, high thermal conductivity, and availability of high-resistivity substrates paved the way for a good deal of device structures to be explored on SiC. This imperative is complete for FETs in that all the high-performance FETs are built in nitride heterostructures on SiC substrates. Let us now discuss some general aspects and examples of GaN growth on SiC substrates with the associated optical and/or electrical properties. When AlN buffer layers are used, the large lattice mismatch between the buffer layer and the subsequent GaN is sufficient to generate large concentrations of structural defects. To ameliorate this GaN on SiC Ga
Ga
Ga
N
N Ga
Ga
N
N Si
N
N
N
Ga N
C Si C
Si
C Si
Ga N
Si
Si
C
C
Ga
C
C
N
N
C
(a) Figure 3.96 (a) With ideal SiC/GaN interface (b). Possible atomic rearrangements for GaN/SiC interface with (0 0 0 1)-oriented substrate and epilayer, with allowance for some intermixing such as N atom replaced by C atom to maintain charge balance (patterned after Ref. [453]).
N Ga N
Si
C Si
Ga
C
Si
C
N
N Ga
Ga
Si
Si
N Ga
Ga
N Interface Si
C
Ga
N
N
N
Ga
Ga
Ga
Ga
Ga
N Ga
Si
N
Ga
N
Ga
Ga
Si
C Si
C
Si C
(b)
C
C
j499
j 3 Growth and Growth Methods for Nitride Semiconductors
500
situation by shifting more of the mismatch to the buffer substrate interface, AlxGa(1x)N buffer layers have been employed. Lin et al. [454] found the GaN quality improved with an Al0.08Ga0.92N buffer layer grown by low-pressure OMVPE. The GaN epitaxial layer deposited on that buffer exhibited an electron mobility and carrier concentration of 612 cm2 V1 s1 and 1.3 · 1017 cm3 (at 300 K), respectively. It has long been suspected that nitridation of the Si-face of SiC(0 0 0 1), if not prevented, could lead to SiN formation, which would prevent the successful growth of GaN. Consequently, growth on SiC commences with a very short exposure to N or nearly simultaneous exposure to the metal and nitrogen in the MBE growth case. Because Si–N bonds are much stronger than the Si–Ga bonds, the growth is likely to begin with N forming single bonds with underlying Si, leading to a Ga-polarity GaN. Convergent beam electron diffraction investigations do indeed indicate that growth on the Si-face of SiC results in Ga-polarity samples [455]. When nitrides are grown on SiC by OMVPE, the growth generally is initiated with AlN or AlGaN at temperatures as high as 1200 C, which requires much lower nitrogen flows than that for GaN. Most common polytypes of silicon carbide, for example, 4H and 6H, have been used for nitride heteroepitaxy as doing so provides several advantages over sapphire, among which are a smaller lattice constant mismatch (3.1%) for [0 0 0 1] oriented films and a much higher thermal conductivity (4.9 W cm1 K1 or lower depending on the impurity and defect concentration). With SiC, conductive substrates are available and electrical contacts to the backside of the substrate are possible, thus simplifying the vertical device structures compared to sapphire substrates. On SiC, GaN epilayers do not rotate around the c-axis and assume the in-plane orientation of the SiC substrate, making registry of planes and cleaved facet formation in lasers easier. Although the (0 0 0 1) surface is available with both carbon and silicon polarities, the Si polarity is more conducive for epitaxy and leads to Ga-polarity samples. Gallium nitride epitaxy directly on SiC is problematic in that low temperatures must be employed for continuous coverage and/or AlN AlGaN buffer layers must be used. This that appears to be a disadvantage can be used to advantage in selective epitaxy on substrates patterned with AlN. Even though the lattice constant mismatch for SiC is smaller than that for sapphire, it is still sufficiently large to cause a large density of defects to form in the GaN layers to the point that extended defect concentration may not improve. SiCs thermal expansion coefficient is lessthan that of AlN or GaN, thus the films are typically under biaxial tensile tension at room temperature. Atomically smooth and damagefree surfaces require additional processing steps, which will be described below. What is more, the cost of silicon carbide substrates is high and currently relatively few manufacturers produce single-crystal SiC. For some heterojunction GaN/SiC devices [456], it is highly desirable to grow GaN directly on SiC without an AlN buffer layer because it will inhibit carrier injection across the AlN/SiC interface owing to perhaps the insulating nature of AlN and the large bandgap discontinuity with SiC and GaN. However, strain/stress issues and ensuing cracking problem must also be kept in full view. Direct nucleation of GaN has been applied for the growth on SiC substrate in MBE growth [457] and OMVPE growth [439,458,459]. Both moderate-temperature and low-temperature GaN nucleation accompanied by a sequential high-temperature annealing have been applied.
3.5 The Art and Technology of Growth of Nitrides
As mentioned in conjunction with Figure 3.91, a good deal of MBE experiments indicate that GaN buffer layers must be grown at relatively low temperatures for wetting. The AlN buffer layers, however, are grown at elevated temperatures such as 1000 C for best results, as discussed in Section 3.5.3.2. In OMVPE, however, the situation is different in that relatively high temperatures, compared to that used for GaN initiation layers on sapphire, can be used for growth initiation. Following initial experiments, the remaining major problem of HT GaN buffer on SiC is the unavoidable crack generation of subsequent GaN epilayer, which is assuaged somewhat by use of AlN buffer layers. Let us discuss the effect of the deposition conditions of the AlN nucleation layer on the transport properties of the subsequent GaN layers, as performed by Koleske et al. [434]. In that particular investigation, following the deposition of the hightemperature AlN nucleation layer at 1080 C, the growth temperature, Ts, was lowered to 1030 C for the growth of GaN using 21 mmol min1 and 1.25 slm NH3 and 1 slm H2 at 250 or 350 Torr. Following the growth of 1 mm of undoped GaN, 1 mm thick Si-doped GaN layers were grown. Using electrical isolation from the n-type SiC substrates (high-resistivity SiC substrates are available albeit at a much larger cost), a 1000 Å thick AlN film was grown on 4H-SiC, which showed breakdown voltages >70 V, giving an experimental breakdown field of >700 kV mm1, paving the way for Hall measurements, conducted at voltages below 5 V, in the subsequent GaN layers grown on AlN buffer layers on SiC. For a comparative analysis, a set of GaN layers were grown on AlN nucleation layers where the AlN nucleation layer growth temperature was varied. Specifically, the GaN films were grown at a system pressure of 250 Torr on approximately 1000 Å thick AlN nucleation layers. For the four higher AlN growth Ts, the GaN was grown on 6H- and 4H-SiC pieces simultaneously, whereas at the lower two Ts, GaN was only grown on 4H-SiC. The electrical characteristics of three additional GaN films grown at 350 Torr by Koleske et al. [434] are listed in Table 3.14. On the on-axis and 3.5 off-axis vicinal 6HSiC substrates, an AlN growth temperature of 1080 C was employed, whereas on the 4H-SiC, an AlN growth temperature of 980 C was used. To lower the electron concentration, nRT, less Si doping was also used in case it was not limited by unintentional impurities. CV measurements for GaN on 6H-SiC showed a carrier concentration, nCV, of 7 · 1016 cm3, close to nRT of 8.9 · 1016 cm3. On the vicinal 6H-SiC, an electron concentration, nCV, of 2 · 1016 cm3 was measured, which is less Table 3.14 XRD and RT PL linewidths and TRPL decay constants
and amplitude ratios (at 200 mJ cm 2 excitation density) for GaN thin layer samples grown with single (s-SiN) and double (d-SiN) SiN layers on low-temperature GaN buffer layers.
Control s-SiNx d-SiNx
10 K PL D0X (meV)
XRD ð0 0 0 2Þ; ð1 0 1 2Þ ðarcminÞ
s1, s2 (ns)
A2/A1
17.7 17.1 13.0
10.7, 26 10.9, 15.6 9.6, 12
0.04, 0.15 0.09, 0.18 0.10, 0.29
0.03 0.27 0.61
The time-resolved data fit biexponential decay function: A1 exp(t/s1) þ A2 exp(t/s2) well.
j501
j 3 Growth and Growth Methods for Nitride Semiconductors
502
than the measured nRT of 4.6 · 1016 cm3. The difference between nCV and nRT may be due to the assumed Si-doped layer thickness of 1 mm used to calculate nRT. From plan view TEM, 8 · 108 cm2 edge dislocations and 2 · 108 cm2 screw dislocations were deduced for the GaN film grown on 6H-SiC (first row of Table 3.14). Because the electron mobilities continue to rise as the sample temperature is lowered, reaching roughly values that are about a factor of 2 larger at 77 K than at 300 K, one can assume that ionized impurity concentration and/or scatterers caused by extended defects have been lowered in concentration [460]. TEM investigations have been conducted to directly view extended defects in GaN as affected by the AlN nucleation layer deposition temperature [434]. Both g-vector scattering configurations, parallel to ½0 1 1 0 and [0 0 0 2], were employed to discern edge and screw dislocations, respectively. For details regarding edge and screw dislocations, see Section 4.1. The general observation was that grains on SiC substrates are better aligned than on sapphire. Specifically, a strong reduction in dislocation density near the GaN/AlN interface was observed for SiC substrates as compared to sapphire substrates, inclusive of a reduction in the screw dislocation component as well. Higher magnification imaging conditions suggest that edge dislocations are likely the result of discrete AlN nuclei forming on the SiC surface. No contrast for screw dislocations was observed, which may in part explain the lower screw dislocation density in the GaN near the GaN/AlN interface. GaN or AlN nucleation layers are usually grown on SiC substrates during MBE growth of GaN-based structures [461]. A discussion of GaN and AlN in the context of nucleation layers is given in Section 3.5.3.2. Here, the emphasis is on layers grown on nucleation layers. As in any growth case, the minimal defect density, the highest mobility, and the best PL characteristics of GaN/SiC epitaxial layers are obtained on the optimized GaN buffer layers. The stress-related phenomena in GaN films grown on SiC are much smaller than those in films grown on sapphire [422]. The defect structure of GaN films grown with and without an AlN nucleation layer on 6H-SiC by ECR MBE have been studied by Smith et al. [462]. Threading defects are identified as double position boundaries originating at the substrate–film interfaces. The density of these defects has been found to correlate with the smoothness of the surface. Even early on, GaN films grown on SiC exhibited higher mobilities (as high as 580 cm2 V1 s1) than those grown on sapphire (230 cm2 V1 s1), AlN buffer layers were used in both cases [39]. The electron mobility of 580 cm2 V1 s1 of film grown on SiC speaks well for the potential of GaN on Si, as hot gas etching of SiC produces atomically flat and damage-free surfaces (discussed in Section 3.2.3.2). Growth of GaN by MBE is much simpler on SiC than on sapphire once proper surface cleaning procedures have been worked out. GaN on SiC has recently been grown in the authors laboratory using both RF and ammonia for N source. Beginning with RF for N source, the SiC substrate shows a well-defined 1 · 1 pattern indicative of the efficacy of H polishing. Once the thermal desorption process is complete, the substrate temperature is reduced to the growth temperature, which is about 650 C. At that point, both Ga and N sources were directed to the substrate with fluxes clearly placing the surface under very Ga-rich conditions. During the first few seconds of the growth, the RHEED pattern is of a very well defined 3 · 3 pattern. Soon
3.5 The Art and Technology of Growth of Nitrides
afterwards, the pattern evolves to a 1 · 1 pattern, which is typical of GaN. Upon cooling after growth, the pattern gives way to a 2 · 2 pattern. Layers grown with ammonia show different RHEED images initially. At first, the pattern is spotty, indicative of 3D growth, which evolves into a typical 1 · 1 pattern in a short period. Postgrowth evaluation generally includes AFM, X-ray, and PL investigations. Generally, the surface is characterized as smooth overall, but broken steps indicate structural defects. GaN layers were also grown in H-polished SiC using ammonia. The XRD measurements for ammonia-grown GaN indicated (0 0 2) (it is (0 0 0 2) in the hjkl configuration) symmetric and (1 0 2) (it is (1 0 1 2) in the hjkl configuration) asymmetric peak FWHM values of 3.81 and 10.5 arcmin, respectively, despite the 0.23 mm thickness of the layer. The PL measurements indicated half width values of about 10 meV and an efficiency of 0.3% at 30 K, which is very good for an unintentionally doped GaN layer of this thickness. 3.5.3.5 Growth on Porous SiC With the advent of porous SiC (PSiC) substrates, which have been reported to lead to SiC epitaxial layers with better quality, as compared to on standard SiC substrates, attempts with both OMVPE and MBE have been made to utilize this method for GaN growth. Standard SiC substrates are made porous in an anodization process, which requires conducting substrates or layers in modified vessel for anodization. The motivation is that the nanosized pores would pave the way for epitaxial lateral overgrowth at the nanoscale with improved results over the conventional ELO. This approach has its genesis in SiC epitaxial growth on PSiC [463] and has been extended to GaN growth on PSiC [464]. Applicable to layers grown by both MBE and OMVPE, optical and electrical properties of SiC grown on PSiC have shown improvement over those grown on standard SiC substrates. However, GaN layers on PSiC did not show decidedly improved electrical, structural, and optical properties, though marginal improvements in all three categories have been shown using ammonia MBE. However, growth on PSiC may have fuelled the debate on nanolateral epitaxy, which will be discussed briefly here. This method has been conceptually and experimentally shown to relieve strain and may in fact cause dislocation bending as well [465]. The technique, dubbed nanoheteroepitaxy (NHE) has in theory been shown to provide 3D stress relief mechanisms when an epilayer is nucleated on a compliant patterned substrate with an array of nanoscale islands. Consequently, the approach can significantly reduce the strain energy in the epilayer and extend the critical thickness dramatically. Calculations show that with 10–100 nm patterning, which is achievable with advanced lithography, one can eliminate mismatch dislocations from heterojunctions that are mismatched by as much as 4.2%, which is in the realm of GaN on SiC. The details are presented in Section 3.5.5.3. The nanometer-scale pores on 6H-SiC substrate are formed by anodization in hydrofluoric acid under UV illumination [466]. In this process, platinum is used as the cathode, whereas the substrate serves as the anode. The anodized substrates usually show very few pores on the surfaces most of the pores are buried under the socalled skin layer, though this may change in that as-prepared samples may have pores on the surface. Figure 3.97a is the cross-sectional TEM image of an as-prepared PSiC
j503
j 3 Growth and Growth Methods for Nitride Semiconductors
504
Figure 3.97 (a) Cross-sectional TEM image showing the porous structure of anodized 6H-SiC substrate. The skin layer is present which blocks most pores from reaching the surface. The depth of the pore region is 2.8 mm, and the two different sizes of pores are 10 and 100 nm, respectively. (b) AFM surface morphology of PSiC substrate after H annealing for 1 min at 1700 C, which served to remove the skin layer.
substrate, showing such a skin layer. The pores with sizes of about 10 and 100 nm each underneath the skin layer extends 2.8 mm into the substrate. Most of the pores start from the substrate surface and penetrate into the substrate like cones, the opening angle of which depends on the anodization parameters. The skin layer can be removed by etching in wet or dry chemistry such as H annealing at 1700 C. Figure 3.97b is an AFM image of a PSiC surface after removal of the skin layer, which causes the pore dimensions to enlarge. GaN films were grown under Ga-rich conditions by MBE using NH3 as a reactive nitrogen source with four samples, A, B, C, and D investigated [467]. The first sample, A, is grown on standard (0 0 0 1) 6H-SiC substrate and the second sample, B, is grown on PSiC(0 0 0 1) substrate with a skin layer of 60 nm. The thickness of GaN in samples A and B, which were grown without any buffer layers, was small (0.11–0.15 mm). The third and fourth samples were thicker (0.77–1.40 mm) and were grown with PSiC skin layer for sample C and without PSiC skin layer for sample D.
3.5 The Art and Technology of Growth of Nitrides
Both samples C and D were grown with a thin layer of AlN buffer (40 nm) between the GaN epilayer and the PSiC substrate. The substrate of sample C had an 8 miscut toward the ð1 1 2 0Þ orientation. GaN growth temperature was about 650 C for growth on PSiC substrate and the NH3 flow rate was fixed at 10 sccm. Nominally identical growth parameters were employed, which were optimized for GaN/6H-SiC growth. H annealing at 1700 C for 1 min performs the removal of the skin layer [468]. TEM images for samples A, thin GaN layer (0.15 mm) grown on standard 6H-SiC substrate, and B, GaN layer grown on PSiC substrate with skin layer, showed dislocation densities of 1 · 1010 and 7 · 109 cm2 near the top GaN surface, respectively. The strain condition of GaN with respect to SiC can be obtained by measuring the relative lattice constant from an electron diffraction pattern that includes both GaN and SiC reflections. The in-plane lattice mismatch for samples A and B are therefore estimated to be Da/|a| ¼ 3.06% and Da/|a| ¼ 3.25%. Considering the in-plane lattice mismatch between GaN and SiC crystals (bulk material) is Da/|a| ¼ 3.48%, we found the degree of strain relaxation to be 87 and 93% for samples A and B, respectively. It is clear that for the same thickness of GaN layer, it is easier to achieve lattice and/or thermal relaxation when grown on PSiC substrate, even with the presence of a skin layer. The effect of the skin layer is significant on the dislocation distribution, as shown in Figure 3.98. Sample C, which is GaN grown on PSiC with a skin layer, exhibits a dislocation density of about 5 · 109 cm2 for a thickness up to 1.40 mm, as can be seen in Figure 3.98. However, when the skin layer (nonporous layer) was removed from the PSiC surface, the dislocation density near the top of a 0.77 mm GaN reduced to 1 · 109 cm2. If the same layer was grown to a thickness of 1.40 mm, it can be predicted that the dislocation density could have been further reduced. The effect of the H annealing/etching at high temperature can be clearly observed in that many pores extend to the surface where the AlN buffer layer and GaN epilayer were grown, though the interface is much roughened because of the presence of many pores being exposed on the surface after etching. Some of the exposed pores are seen to be backfilled with Al and/or Ga droplets during the MBE growth (dark area inside pores). For samples C and D, the electron diffraction patterns indicate 100% relaxation based on the measured value of Da/|a| ¼ 3.5% for both. As discussed in Ref. [469], when Da/|a| exceeds the strain-free value (3.48%), the GaN film is likely to be slightly under tensile strain. It is also possible that the lattice constant of PSiC is slightly different from SiC, as is the case with porous Si material. A tentative growth mechanism can be suggested, though speculative at this juncture, for GaN growth on PSiC. At the beginning of growth, AlN buffer grows on the nonporous regions of SiC, including possibly the sidewalls of SiC pores, leading to columns of AlN. This is then followed by GaN growth on top of the AlN columns. This is supported by the observation of open tubes in the initial stages of GaN growth, especially on the tilted PSiC substrates [469,470]. As the growth progresses, lateral epitaxy is enhanced, and the open tubes are closed at the upper part of the GaN layer. These tubes may serve as a relief mechanism for strain caused by the lattice and thermal mismatch during growth. In addition, some threading dislocations start to bend, merge, and bury in the mid-layer.
j505
j 3 Growth and Growth Methods for Nitride Semiconductors
506
Figure 3.98 Cross-sectional TEM images for (a) GaN layer (1.40 mm—sample C) grown on PSiC substrate with skin layer, and (b) GaN layer (0.77 mm—sample D) grown on PSiC substrate after the skin layer is removed by high temperature (1700 C) annealing for 1 min. The insets show the electron diffraction patterns for GaN layer and the PSiC substrates.
The crystalline quality of GaN grown on PSiC substrate was assessed by HRXRD rocking curve (o-scan). The best FWHM of the (0 0 0 2) rocking curve diffraction obtained was 3.3 arcmin and that of ð1 0 1 2Þ was 13.7 arcmin. This implies that device-quality GaN film can be achieved in thin layers grown on PSiC substrate. Moreover, low-temperature PL measurements indicate good quality of GaN films grown on PSiC substrates. The PL spectra measured at 15 K for samples B, C, and D showed the best FWHM of GaN excitonic peak to be 9.5 meV for the GaN grown on PSiC with the skin layer removed (sample D), whereas GaN films with the skin layer exhibit FWHMs of 14–16 meV (samples B and C). The quantum efficiency of PL from thicker layers (samples C and D, both with and without skin layer) is quite high (about 1.6%), whereas that from the thin layer is low (about 0.1%), indicating a reduction in nonradiative defects (presumably dislocations) in thicker layers. The features of the PL spectra for all three samples are similar in that they include excitonic emission,
3.5 The Art and Technology of Growth of Nitrides
shallow donor–acceptor band with the main peak at about 3.26 eV, and a very weak yellow luminescence band at about 2.2–2.3 eV. The highest optical quality of sample D from PL spectra agrees with the lowest dislocation density (1 · 109 cm2) of this GaN layer grown on top of the porous SiC with skin layer removed. 3.5.3.6 Zinc Blende Phase Growth Owing to the cost and size restrictions associated with SiC substrates, a substantial impetus exists to grow GaN on Si substrates. To ameliorate chemical and lattice mismatch issues, 3C-SiC(0 0 1) films are generally deposited on Si(0 0 1) before cubic GaN deposition. Specifically the lattice constant mismatch of 3.4% between 3C-SiC and zinc blende GaN and the excellent thermal stability of 3C-SiC are the driving forces. However, 3C-SiC/Si(0 0 1) substrates generally suffer from very poor crystal quality and rough surfaces, both of which tend to promote the nucleation of wurtzite GaN. Furthermore, zinc blende AlN is very difficult to stabilize. Since wurtzite AlN is strongly favored regardless of the substrate or its orientation, it is not effective as a buffer for growth of zinc blende GaN. These complications and successful device demonstrations in Wz GaN gave way to the exploration of Wz GaN on SiC, which, in turn, is formed on Si. The particular approach used for this particular combination is discussed in conjunction with Pendeo-epitaxy on Si discussed in Section 3.5.5.2.3. Getting back to the zinc blende GaN on SiC, a 3C-SiC(0 0 1) film of several microns thick was grown on Si(0 0 1) by CVD and used to grow zinc blende GaN epilayers by several groups [471]. The crystal structure of GaN in OMVPE growth is greatly affected by both the growth temperature and the V/III ratio [472]. More growths of GaN on 3C-SiC have been accomplished by MBE [471,473–475]. Daudin et al. [476] found that both the initial substrate roughness and the V/III ratio affect the wurtzite/ zinc blende ratio in MBE-deposited films, whereas a N-rich condition was found detrimental to the growth of the zinc blende phase. The observations were explained by the assumption that MBE growth of zinc blende GaN is mainly governed by the impingent and overabundant, as compared to Ga, active N flux, which directly determines the mean free path of gallium adatoms. Thin 3C-SiC film with thickness of several nanometers produced by directly carbonizing the Si(1 0 0) surface has been used in MBE growth [477,478], and it was also found, as in the case of Wz GaN growth by MBE, that a Ga-rich growth condition is beneficial for growing films with flatter surfaces [477]. The quality of zinc blende GaN films is normally worse than that of wurtzite films, as determined by X-ray diffraction, TEM, PL, and other characterization methods. More discussion of zinc blende GaN growth can be found in the section dealing with growth on GaAs substrates. 3.5.4 Growth on Si
Si substrates are very attractive not only because of their high quality, availability, and low cost but also the distant possibility of integration of Si-based electronics with wide bandgap semiconductor devices. Both zinc blende and wurtzite GaN epilayers have been grown on Si(0 0 1) by MBE [479], OMVPE [480], or HVPE [481]. It should,
j507
j 3 Growth and Growth Methods for Nitride Semiconductors
508
however, be mentioned that dismal quality of films along with cracking owing to the tensile residual thermal stress makes the standard growth of GaN directly on Si undesirable. Although still not as good as those on Wz SiC and sapphire, growth on 3CSiC(1 1 1) either deposited or formed on Si by carbonization and growth with the aid of epitaxial lateral overgrowth have attracted attention. In fact, very competitive AlGaN/ GaN MODFETs have been reported on Si substrates as detailed in Volume 3, Chapter 3. The lateral overgrowth method directly on Si is discussed in Section 3.5.5.2. Its variant pendeo-epitaxy on 3C SiC/(1 1 1) Si is discussed in Section 3.5.5.2.3. Nitridation as carried out on sapphire is not performed because of the possibility of either stoichiometric or nonstoichiometric SiN formation (Si3N4 is its stoichiometric form). Several growth initiation techniques can be employed. One approach utilizes the growth of a thin AlN layer directly on Si followed by GaN deposition. GaN deposition directly on Si is not generally performed in MBE as Ga evaporates rather readily from the Si surface at typical MBE deposition temperatures. The other method involves the deposition of a sub- to one monolayer of Al deposition at about 650 C followed by nitridation to convert it to AlN, which is then followed by regular deposition of AlN. Cubic GaN is generally grown on Si(1 0 0) substrates [482], whereas Si(1 1 1) substrates are employed for the wurtzitic GaN growth [483,484]. Both phases, wurtzitic and zinc blende, are frequently detected, accompanied by a large number of extended defects such as dislocations, stacking faults, and twins. GaN grown on Si (0 0 1) is predominantly cubic, but not without mixed polarity. In one experiment, a 30 nm GaN low-temperature buffer layer was employed to accommodate the 17% lattice mismatch between the film and the substrate by a combination of misoriented domains and misfit dislocations [485]. Beyond the buffer layer, highly oriented domains separated by inversion domain boundaries are found by TEM. Stacking faults, microtwins, and localized regions of wurtzitic structure are major defects in the film. The films grown on Si(1 1 1) have a predominantly wurtzitic structure with a presence of twinned cubic phase. The GaN films grown directly on Si(0 0 1) are generally phase mixed, containing both zinc blende and wurtzite modifications or only wurtzite phase and, depending on the buffer layer details, two different rotation alignments with different orientations of the wurtzite phase (with c-direction of GaN being on the (0 0 1) surface of Si or r-plane of GaN being oriented parallel to the Si surface) [486]. Owing to the large lattice mismatch, it is difficult to grow pure zinc blende phase GaN directly on the (0 0 1)Si, which is aggravated by an amorphous SixNy layer that may form at the GaN/Si interface. To mitigate this problem, a 3C-SiC thick layer produced by CVD or a thin layer produced by carbonization is commonly used. The GaN epitaxy on 3C-SiC/Si is discussed in conjunction with pendeo-epitaxy in Section 3.5.5.2.3. The utilization of other buffer layers, such as g-Al2O3, AlN, or AlxGa1xN, resulted in the wurtzite phase growth, as will be discussed below [13]. The Si(1 1 1) orientation is generally the preferred substrate plane for GaN epitaxy. The quality of GaN on these substrates has been improved to the point where devices have been demonstrated. For example, ultraviolet LEDs grown by MBE [487] and OMVPE [488], Schottky-barrier ultraviolet detectors [489], and field-effect
3.5 The Art and Technology of Growth of Nitrides
transistors [490], in the form of Si channel with AlN as the gate dielectric [491], have been attained. Because Si is a nonisomorphic substrate to GaN, the surface step could possibly lead to an SMB, as discussed in section 3.5.3.3 on the SiC substrate. Martin et al. [492] showed that surface preparations reducing irregularities on the substrate surface greatly improved the crystalline and luminescent quality of a GaN epilayer grown by MBE. Although this was carried out on Si(1 1 1) with ECR MBE, the gist of this work applies to Si(0 0 1) also. Wide atomically flat terraces were created by etching using a solution of 7 : 1 NH4F : HF, greatly increasing the distance between surface steps and thus decreasing their density. The latter reduced the number of stacking mismatch boundaries of the epitaxial GaN films. Moreover, to overcome the problem of a large difference in lattice parameters and the strength of the Si–N bond, different buffer layers such as AlAs, oxidized AlAs, GaAs, AlN, LT GaN, ZnO, and 3C-SiC(1 1 1) have been implemented [13]. In addition, to avoid the formation of silicon nitride, it is a common practice in MBE growth to initiate the AlN buffer layer growth by exposing the Si substrate to the flux of aluminum and ammonia alternatively for a short time or by briefly exposing to RF nitrogen followed by Al. These techniques help suppress the formation of excessive SiNx at the interface and result in a 2D growth mode of GaN. Despite the 23.4% misfit in the AlN/Si system, optimized AlN films grown on a (7 · 7) reconstructed Si surface made it possible to grow GaN. In OMVPE growth, a hightemperature AlN buffer layer is much better than a low-temperature one and it is necessary to enhance 2D growth for GaN epitaxy [493,494]. In the case of MBE, the RHEED pattern obtained on GaN grown on Si followed by an AlN buffer layer displays the typical 1 · 1 pattern at the growth temperature, which transforms to a 2 · 2 pattern upon cooldown, indicative of a Ga-polarity sample, implying that N forms the long bonds to Si [495]. The XRD measurements indicated (0 0 2) symmetric and (1 0 2) asymmetric peak FWHM values of 15.9 and 10.9 arcmin, respectively. PL measurements indicated half width values of about 19 meV at 30 K. Alternative substrates, such as compliant silicon-on-insulator (SOI) substrates and porous silicon (PS) have been utilized for GaN growth both by OMVPE and MBE [13]. The premise of these substrates is that they are expected to improve the quality of epitaxial GaN layers by releasing the strain and absorbing the generated threading dislocations in the thin Si overlay of the SOI substrate in much the same way as in SiGe on Si. Moreover, selective area growth has also been applied by the use of stripepatterned and dot-patterned Si(1 1 1) by OMVPE [496] and MBE [497] in an effort to grow low-defect GaN heterostructures. Despite the progress made in all types of nitride growth on Si, the GaN layers on Si are in general inferior to those grown on sapphire, SiC, and SiC-coated Si. The advantages of Si, which are always the same no matter what the application is, are high thermal conductivity, availability in large numbers and in high quality, and Si microelectronics. Epitaxial relationship between GaN or AlN and Si is discussed in Section 3.3.3. Having discussed the issues dealing with surface preparation, interface mismatch, and so on, let us now turn our attention to thermal cracking that characterizes GaN growth on Si. The thermal expansion coefficient of GaN is much larger than that of Si, see Chapter 1. The resultant tensile strain leads to cracking when the film thickness
j509
j 3 Growth and Growth Methods for Nitride Semiconductors
510
exceeds a critical value (usually 1 mm) [498], as slip system is not available [413,499]. The primary cracking directions are ½1 1 2 0, ½ 12 1 0, ½ 2 1 1 0 with resultant cleavage plane being f1 1 0 0g [500]. Another difficulty involving Si substrates is that at high temperatures, a strong chemical reaction between Ga and Si takes place, namely, meltback etching, which leads to a deterioration of substrate and epilayer [501,502]. The meltback etching, once initiated, cannot be stopped and causes rough surfaces with deep voids in the substrate. A third relevant point is the nitridation of Si surface when it is exposed to ammonia, leading to the formation of amorphous Si3N4 [165]. However, this process is also reported to be self-limiting when a SixNy thickness of a few nanometers is reached [503]. The point is that special precautions must be taken to prevent surface nitridation of Si during growth of nitrides by OMVPE. As could be garnered from the above paragraph, most research relevant to GaN on Si revolves around finding ways to produce crack-free GaN on Si with thickness >1 mm. Naturally, the nucleation layers have received a great deal of attention in this regard. Typically, AlN buffer layers, grown at a high temperature of 1100 C, represent the most commonly employed scheme to grow GaN on Si [504–506]. Follstaedt et al. [504] obtained GaN epitaxy on Si(1 1 1) by using a 30 nm AlN buffer layer grown at 1080 C, with the subsequent GaN growth at 1060 C. The TEM images showed that AlN buffer layer was continuous, implying that AlN can wet silicon substrate very well. Lahreche et al. [505] found that the best AlN growth temperature is approximately 1060 C. In most cases, GaN grown using just a single hightemperature AlN buffer layer produces cracks during cooling down to room temperature. Zamir et al. [506] discovered that AlN should be grown at temperatures well above 760 C, and as close as possible to 1100 C, to get a more aligned GaN and a smooth GaN surface for short growth times, meaning thin films. Low-temperature AlN buffer has successfully been used to reduce stress in GaN on sapphire by Amano et al. [507] and Dadgar et al. [508] who used low-temperature AlN buffer to eliminate cracks in GaN on Si. The crack density was reduced to zero from an original density of 240 mm2. Additionally, the FWHM value of X-ray rocking curve for ð2 0 2 4Þ improved from approximately 270 to 65 arcsec. Blasing et al. [509] explained the mechanism of stress reduction in GaN with a low-temperature AlN buffer layer arguing that the buffer layers grown at high temperatures are pseudomorphic, whereas those grown at lower temperatures are relaxed. Therefore, AlGaN or GaN layers grown on a low-temperature AlN interlayer grow under compressive interlayer-induced strain that compensates the thermal strain. The stress in the GaN layer depends on the growth temperature, which is likely to control the degree of AlN interlayer relaxation. To avoid surface nitridation of silicon substrate, a few monolayers of Al predeposition prior to AlN growth are usually necessary [510,511]. A lack of amorphous SixNy characterizes the interface between AlN and Si when such an AlN initiation layers is employed. Besides low-temperature AlN buffer, other methods have also been applied to obtain crack-free GaN on Si. Feltin et al. [512,513] employed GaN/AlN superlattices as a buffer layer to reduce tensile stress in the epilayer thus reduce the tendency to crack and achieved a relatively thick (2.5 mm) GaN without any cracks. A side benefit of the
3.5 The Art and Technology of Growth of Nitrides
AlN/GaN superlattice buffer composite is reduction of threading dislocation density from 1.5 · 1010 cm2 for a sample without superlattices to 2.5 · 109 cm2 for the best uncracked sample with four stacks of superlattices. Marchand et al. [514] reported on the growth of crack-free GaN, which was under compressive stress, using graded AlN-to-GaN buffer with a total buffer thickness of 800 nm. In contrast, the GaN buffer layers grown with a high-temperature AlN buffers showed a tensile stress and exhibited cracking in film. Additionally, HfN buffer layers have also been used to grow GaN on Si [498]. Compared to AlN, HfN is a better diffusion barrier, which can prevent diffusion of Si to epilayer, and has a closer lattice match to GaN. Crack-free GaN with a thickness of 1.2 mm has been grown on both (1 1 1) and HfN/Si(0 0 1) surfaces. Another method developed to obtain a crack-free thick GaN on Si, other than using various kinds of buffer layers, is by patterning the silicon substrate. This can be done by either masking or etching the substrate. The concept here is not to avoid cracks completely but to guide them to the masked or etched parts of the substrate. As long as the patterned regions are smaller than the average crack–crack distance, no cracking of the GaN layers should occur because the substrate is softer. A Si3N4 or SiO2 mask layer deposited directly on the substrate can be used to produce the pattern needed. Because the dielectrics are amorphous and have very long migration lengths, no GaN layer is typically grown on them as evidenced by no or a little GaN observed on such a mask. A related approach, which does not require a mask, is to etch trenches sufficiently deep to avoid the formation of continuous film during growth. In this vein, Zamir et al. [515] used patterned Si substrates, formed by reactive ion etching (RIE) square mesas, and obtained 14 mm ·14 mm crack-free GaN grown by OMVPE. Krost et al. [500] used square-patterned Si(1 1 1) aligned along Si ½1 1 0 and Si ½1 1 2 with SixNy masks, prepared by standard photolithography and wet chemical etching, to obtain crack-free 3.6 mm thick GaN in the window regions on Si. In their samples, cracks can only be observed in the SixNy stripes separating the square areas. The epitaxial lateral overgrowth method has also been applied to growth on Si substrates, as discussed in Section 3.5.5.2. Building on the earlier brief mention of this, let us now give a succinct review of GaN-based devices on Si. The FET results on Si substrates are discussed in sufficient detail in Volume 3, Chapter 3. In terms of LEDs, green InGaN LED on Si using AlN/GaN superlattice as buffer layer by MOVPE have been reported [516–518]. The turn-on voltage of the LED was 6.8 V and the operating voltage was 10.7 V at 20 mA. The electroluminescence peaked at 508 nm, with a full width at half maximum of 52 nm. An optical output power of 6 mW was achieved for an applied current of 20 mA. Dadgar et al. [519] employed in-situ SixNy mask to improve the crystalline quality of GaN and grow InGaN/GaN multiquantum well LED. By using the combination of low-temperature AlN buffer and in-situ SixNy mask, a 2.8 mm thick diode structure was grown using OMVPE. In current–voltage measurements, turn-on voltage of 2.5–2.8 V and a series resistance of 55 O were observed for a vertically contacted diode. The light output power was 152 mW at a current of 20 mA, with a wavelength being 455 nm, which is not comparable to that obtained on sapphire.
j511
j 3 Growth and Growth Methods for Nitride Semiconductors
512
3.5.5 Growth on Sapphire
At the time of writing this book, OMVPE is the workhorse of GaN and related materials growth, even though the growth mechanisms are not well understood. In this section, OMVPE growth and related issues will be discussed. The discussion of the growth by MBE on sapphire will be limited to the effect of buffer layers on the polarity of the resultant films. As mentioned in Section 3.4.1.1, substrate temperatures well in excess of 900 C are required to obtain single crystalline high-quality GaN films, and the GaN films with the best electrical and optical properties are grown at 1050 C or even higher. Such high temperatures in H2 and NH3 ambience provide highly formidable challenges. Metalalkyls need to be kept below pyrolysis temperatures until just before the reaction zone: ammonia and metalalkyls would have to be kept separate until just before the reaction zone as well. High growth temperatures associated with the OMVPE process require that the substrates used do not decompose at the deposition temperatures. Sapphire, Si, and SiC are among the substrates that meet this criterion. The high N/Ga flux ratio used to minimize the nitrogen loss is another issue. The high temperatures have a contradictory effect on the crystal quality. On one hand, the surface mobility of atoms and thus the quality of grown film is higher for higher substrate temperatures. On the other hand, high vapor pressure of N over Ga and particularly In brings the growth process very close to the dissociation temperature. Not as serious, but postgrowth cooling introduces more strain and thus more structural defects may be introduced during cooling. At substrate temperatures exceeding 1100 C, the dissociation of GaN results in voids in the growth layer. Use of dimethylhydrazine (DMHy) [520] as a nitrogen source permits lower deposition temperature, because DMHy decomposes at temperatures (500 C) lower than that for ammonia (800 C). The film quality is not high enough, however. Rocking curve FWHM of 150–250 arcsec and electron concentration and mobility of 5 · 019 cm3 and 48 cm2 V1 s1, respectively, have been obtained. Early on, nitrogen vacancies had been widely considered to be the primary cause of large background electron concentration in GaN. However, this gave way to other concepts such as O doping. In any case, an improvement in crystal growth must involve more complete incorporation of nitrogen into the crystal lattice, as well as reducing the likelihood of impurity incorporation. In an effort to achieve this complete incorporation of nitrogen into the crystal lattice, several investigators [143,521,522] substituted the more reactive hydrazine in favor of ammonia. Andrews and Littlejohn [523] tried Ga(C2H5)3NH3, which already has a Ga–N bond, as a source material. Others used plasma excitation of nitrogen in CVD growth environment, which proved to be the most popular method of increasing the reactivity of nitrogen. In the CVD growth approach, several variations have been attempted in the presence of plasma. Suitable precursors in the metalorganic CVD (OMVPE) growth must exhibit sufficient volatility and stability to be transported to the surface. These precursors should also have appropriate reactivity to decompose
3.5 The Art and Technology of Growth of Nitrides
thermally into the desired solid and to generate readily removable gaseous side products. Ideally, the precursors should be nonpyrophoric, water and oxygen insensitive, noncorrosive, and nontoxic. In contrast to the MBE growth, plasma excitation of the nitrogen species has not proven necessary in CVD growth. In spite of the fact that trialkyl compounds (TMA, TMG, TMI, etc.) are pyrophoric and extremely water and oxygen sensitive and that ammonia is highly corrosive, much of the best material grown today is produced by conventional OMVPE by reacting these trialkyl compounds with NH3 at substrate temperatures in the vicinity of 1000–1100 C [196,524–533]. The reactive nature of constituents makes it imperative that special materials are used for any component that experiences high temperatures. This is especially true for reactors where the substrate heating is done by resistive means internal to the deposition system. Although there is a notable dependence on the reactor and associated temperature measurement methods, deposition temperatures in excess of 800 C are required just to obtain single crystalline GaN films and near or above 1050 C for high-quality GaN films. At substrate temperatures exceeding 1100 C, the dissociation of GaN results in voids in the growth layer. A similar situation is observed for AlN film growth. 3.5.5.1 OMVPE Low-Temperature Nucleation Buffer Layers As in the case of HVPE, if GaN is grown directly on sapphire, the quality of the crystals grown by OMVPE is not as good as that grown on small GaN templates because of large lattice and thermal expansion coefficient mismatch and mostly nonideal nucleation. The degradation of the quality is evident by a wide X-ray rocking curve (10 arcmin typically), rough surface (containing hillocks), high electron concentration (up to 1019 cm3), and considerable yellow emission in the luminescence spectra. To improve the GaN layer quality, low-temperature AlN or GaN nucleation layers are usually grown following the nitridation process (which may not require a separate and deliberate step) [534,535]. Although the low-temperature AlN and GaN buffer layers (initiation or nucleation layer terminology is more descriptive and less confusing as subsequent GaN layer may also be called buffer layer in a device structure) were initially used solely for OMVPE, their use in the MBE process was also shown to be highly beneficial. The use of nucleation layers has become an indispensable step for obtaining good-quality GaN and AlGaN films. However, lowtemperature initiation layers have given way to moderately high-temperature initiation layers both on SiC and sapphire, the exact value of which depends on whether AlN or GaN initiation layer is employed. Although low-temperature layers for enhanced nucleation were explored much earlier, a milestone was achieved in 1986 when Amano et al. [536] reported the use of low-temperature buffer layers to enhance the surface morphology and electrical and optical properties of GaN. This was accomplished by depositing about 50 nm thick AlN film on c-sapphire as a nucleation layer. Until then, hexagonal GaN grown by OMVPE was deposited directly on sapphire substrates and suffered from low quality, which manifested itself with high electron concentrations approaching 1020 cm3. In the low-temperature buffer layer approach, as reported by Amano et al., an AlN layer is first deposited at 600 C, or thereabouts, followed by a high-temperature annealing
j513
j 3 Growth and Growth Methods for Nitride Semiconductors
514
in H2 atmosphere, and finally, the GaN layer is grown at 1000 C. The surface morphology of the GaN film grown in this manner was improved to an RMS roughness of about 1 nm and was also free of cracks. GaN films grown on low-temperature AlN nucleation layers displayed much improved quality as measured by X-ray diffraction and photoluminescence. The FWHM of the double-crystal X-ray rocking curve (0 0 0 2), which measures ordering along the c-direction only, obtained for optimum conditions was 110 arcsec [537]. The 4 K photoluminescence spectrum measurement exhibited a sharp donor-bound exciton transition with a FWHM value of about 1 meV. This was accompanied by the observation of free exciton related transitions in some of the samples. As for the transport properties, the electron concentration was reduced to about 1017 cm3 with an associated mobility of about 600 cm2 V1 s1 at room temperature, a first for that time. Such relatively low electron concentration moved the GaN layers from the degenerate state to the nondegenerate state and temperature dependency of the carrier concentration and electron mobility emerged. Consequently, the interpretation of residual donors and scattering mechanisms began. Important process parameters in the pursuit of the best nucleation layers in random order are as follows: (i) Growth rate of the nucleation layer [538–540]. (ii) Thermal annealing of the GaN buffer layer of optimal thickness [541–544]. (iii) Possible misorientation of the sapphire substrate. In one investigation, it was found that a slight misorientation of the sapphire by about 0.17 significantly improves the surface morphology [545]. (iv) Composition of the carrier gas. Optimizing the V/III ratio during the nucleation phase is relevant for the achievement of mid-108 cm2 dislocation density [546]. An optimal combination of H2 and N2 carrier gases reduces the residual strain and improves the quality of GaN [547]. Some of the above-mentioned parameters are discussed in some detail below. Here, we will discuss the low-temperature nucleation layers followed by moderate-temperature nucleation layers. The time sequence employed during all stages of GaN growth by the group of I. Akasaki is depicted in Figure 3.99 with the low-temperature buffer layer growth temperature of 600 C. This must be contrasted to a temperature range of 480–600 C employed at various laboratories for the low-temperature AlN and GaN nucleation layer growth on sapphire substrates. As demonstrated by the nucleation layer deposition temperature, the details may vary from laboratory to laboratory, but the essential elements of this time sequence are fairly standard. The purpose of the lowtemperature nucleation layer is to optimize the transition between the sapphire substrate and the GaN device layers. The benefits include a high nucleation density and no need to follow the surface arrangement of the sapphire substrate. The latter is to allow the nitride layer to assume its own structure with as few defects as possible by appropriate twisting, tilting, and prismatic growth; hardly the goal of homoepitaxy. The crystalline structure of this low-temperature buffer layer is still shrouded in controversy. What can be stated is that this layer contains cubic and wurtzitic phases
3.5 The Art and Technology of Growth of Nitrides
Sapphire/AlN/GaN
H2
NH 3
TMA
TMG or TEG
Cool down
GaN growth
RT
LT AlN growth
600 Nitridation
Substrate temperature (ºC)
1100 1000
Time (au) Figure 3.99 The time sequence employed in the growth of group-III nitride heterostructures at Meijo University. Courtesy of I. Akasaki, Meijo University.
as well as much disorder. Although very different in nature from the OMVPE case, some sort of buffer layer is also employed in the MBE approach. Unlike the OMVPE case, in the MBE approach, no matter how low the substrate temperature is, the registry between the buffer and the underlying substrate is strong, which is not desirable in nitride heteroepitaxy. The initial stages of growth are very critical for obtaining heteroepitaxy and the quality of the resulting film. In general, the deposition can be characterized by three cases, namely 2D layer-by-layer mode, 3D island mode, and mixed (M) mode, which is layer-by-layer followed by island formation. The first mode results in a smooth surface, whereas the next two have been stated to lead to rough surfaces and lowquality epitaxial layers unless one deals with quantum dot like structures. However, the picture is more complex and depends very much on whether GaN or AlN nucleation layers are employed. In the case of GaN nucleation layers, during ramp-up in H2 and NH3, the regions at the boundaries would most likely be etched away,
j515
j 3 Growth and Growth Methods for Nitride Semiconductors
516
leaving behind possibly the ordered material only. In this vein, the investigations by Lorenz et al. [548] and Narayanan et al. [549] have demonstrated that as-deposited lowtemperature GaN nucleation layers are fairly rough. However, results of Gonsalves et al. [550] indicate that low-temperature AlN nucleation layers are relatively smooth and strain may exist at the AlN/sapphire interface. When these nucleation layers are heated to high temperature for growth, rounded islands develop that are essentially dislocation free. These regions are thought to be the nucleation sites for growth during the high-temperature deposition of GaN and epitaxial lateral overgrowth might also aid the process to lead to coalesced and smooth layers in the end. The growth mode is determined by many parameters, such as interfacial energy of solid and vapor phase and interfacial energy of vapor phase and substrate, which, in turn, depends on the growth temperature, the bond strength and the bond lengths of the substrate and the overgrowth material, the rate of impinging species (flux), surface migration rates of reactants, supersaturation of the gas phase, and the size of the critical nuclei, among others. Because the roughness of the nucleation layer is larger at higher growth temperatures, the epitaxial growth as a rule is divided into two steps: smooth thin (20 nm) buffer layer grown at low and moderate temperatures and main layer grown at higher temperatures. Attempts were made to determine the evolution of the low-temperature AlN nucleation layer and the subsequent GaN overlayer. TEM analyses showed that, as expected, GaN is defective particularly at the AlN/GaN interface and that GaN grew in a columnar fashion [551]. The crystalline structure of the low-temperature buffer layer is still shrouded in controversy. The AlN buffer layer has amorphouslike structure at the deposition temperature of 600 C, but during the ramping process, it is crystallized and exhibits the well known columnar structure [551], as shown in the first panel of Figure 3.100. Cross-sectional transmission electron microscopy images of the highly defective initial GaN layer display features similar to those of the AlN buffer layer, suggesting columnar fine crystals as shown in the artistic rendition of Figure 3.100, second panel. Each GaN column is probably grown from a GaN nucleus formed on top of each columnar AlN region. Consequently, a high-density nucleation occurs owing to the high density of the AlN columns. The columns have disordered orientations and as the base of the columns gets larger, the number of columns emanating at the growth surface gradually decreases. The crystalline structure supports the prismatic growth, leading to a general alignment along the c-direction with some twist and tilt. The relative tilt and twist between these columns decrease as the layer thickness increases, showing that the film continues to evolve toward a more ordered structure as growth proceeds, as shown in Figure 3.101 [552]. The subgrain boundaries are more likely the primary defected regions of the epilayer. The small value of strain broadening observed for (0 0 0 2) and ð1 0 1 2Þ reflection indicates that the crystal quality is high within each GaN columns. During the stage where the trapezoidal crystals are formed, as the front area of the column increases by the geometric selection, the growth front follows the c-face, as shown in Figure 3.100 (third from the top). The changes in surface morphology during the early growth stage of GaN deposition [551] is schematically illustrated in
3.5 The Art and Technology of Growth of Nitrides
AlN buffer layer AlN α-Al2O3
Prismatic growth GaN AlN Island formation and expansion GaN AlN Island coalescing GaN AlN Complete coverage
Column boundary
Good region Transition region ~150 nm Faulty region ~50 nm AlN buffer ~ 50 nm Figure 3.100 Artistic rendition of the salient structural features as the low-temperature AlN nucleation layer followed by ramping to high temperature (about 1000–1100 C) and GaN growth evolve. The schematic is based on the analysis of the cross sectional TEM images. Patterned after [551].
Figure 3.100. The stage depicted in Figure 3.100a corresponds to the generation of trapezoidal islands where the preferred orientation begins to be established. Subsequently, lateral growth and coalescing of the islands occur in stages as shown in Figure 3.100b and c. The trapezoidal crystals grow at a higher rate in a transverse direction, as shown in the schematic of the growth front (panel 4) in Figure 3.100, and the islands coalescence. Finally, because the crystallographic directions of all islands agree well with each other with a few degrees of disorder, a smooth GaN layer with a small number of defects results, as seen in Figure 3.100 (fifth from the top). The critical importance of the deposition rate of the low-temperature buffer layer is further illustrated by the FWHM of the (0 0 0 2) XRD peak for 5 mm thick GaN epilayers grown on these buffer layers. In one experiment [144], the measured FWHM values were 180, 250, and 320 arcsec for 30, 70, and 90 Å min1 growth rates, respectively. These data clearly demonstrate a direct correlation between the crystallinity of the buffer layer and the crystallinity of the 5 mm GaN films grown on top of these buffer layers.
j517
j 3 Growth and Growth Methods for Nitride Semiconductors
518
[0 0 0 1]
[0 0 0 1]
Tilt
Tilt
[0 0 0 1] Tilt
[1 1 2 0]
Sapphire substrate Twist
[1 1 2 0] [1 1 2 0]
(a) Figure 3.101 Artistic view of the microstructure of GaN on sapphire. (a) Side view showing relative tilt of [0 0 0 1] directions between columns. (b) Plan view showing relative twist of columns, ½112n directions. After Ref. [552].
Figure 3.102 Surface morphology of GaN layers, as observed by scanning electron microscopy (SEM) grown on annealed lowtemperature AlN buffer as the growth evolves, panels a–d correspond to 3, 5, 10 and 60 min, respectively. Patterned after [551].
(b)
3.5 The Art and Technology of Growth of Nitrides
To appreciate the implications of the data, it is imperative that the origin of the XRD peak broadening is clarified. It should be pointed out that this is applicable to any layer grown by any method. To a first extent, the broadening is a measure of the disorder in the alignment of the c-axis among the columns and the inhomogeneous strain. HRXRD can discriminate between broadening of an X-ray peak owing to strain (a range of crystal-plane spacing) and broadening owing to the mosaicity (a range of misorientation for a particular set of crystal planes). Broadening because of strain turns out to be less than about 30 arcsec, leaving the major influence of the broadening to the mosaicity. The mosaicity broadening decreases with increasing layer thickness. Thus, the trend toward a common crystal orientation, which was observed during buffer-layer annealing, continues to be operative even after several micrometers of GaN growth. The relative constancy of the strain broadening is attributed to an inclined slip plane in GaN grown on [0 0 0 1] sapphire, which is purported to preclude annihilation of dislocations. Continuing on with the AlN nucleation layers but bringing to bear additional diagnostic tools, Zhao et al. [553] and Zhang et al. [554] have studied the dependence of the properties of subsequent GaN epilayers on low-temperature AlN nucleation layers using in situ optical reflectivity. They have shown that high-quality GaN epilayers can be achieved when the AlN nucleation layers with an optimal thickness undergo long annealing time. The 20 nm thick AlN nucleation layers grown at 600 C and annealed during the 1000 s temperature ramp to GaN growth temperature resulted in GaN of acceptable quality as judged by the broadening of (0 0 0 2) symmetric and ð1 0 1 2Þ asymmetric X-ray diffraction peaks. The traces of in situ optical reflectivity measurements were found to reveal that the initial stages of the growth process of GaN epilayers are affected by the annealing time and the thickness of the AlN nucleation layer. These authors have also found that an AlN nucleation layer with rough surface led to degraded quality of GaN epilayers. Cross-sectional transmission electron microscopy images show that a region of about 50 nm of the GaN layer immediately atop the AlN nucleation layer contains a large concentration of defects. Above this defect-laden region, another zone that has a number of trapezoid crystals could be delineated. The density of defects in this 150 nm zone is much lower than that in the interfacial zone. The remaining region on top exhibits a rapid decline in defects, eventually down to defect densities in the range of 108–1010 cm2 depending on the substrates used and growth parameters applied. Following the use of relatively low-temperature nucleation layers for nearly a decade, moderate-temperature nucleation layers of AlN gained considerable interest because of improved device performance. In one such example and with the use of two-step high-temperature AlN nucleation layers on sapphire, Ohba and Iida [555] obtained lasers operating at 413 nm with typical threshold densities of 7.6 kA cm2. Although these possibly unoptimized results are comparable to the best devices fabricated using low-temperature AlN nucleation layers, the simplicity of the device structure employed points to the better quality layers obtainable with high-temperature AlN nucleation layers. One of the difficulties associated with high-temperature nucleation layers is the pitted surface morphologies. Although the subsequent GaN layers tend to
j519
j 3 Growth and Growth Methods for Nitride Semiconductors
520
Figure 3.103 Topographical AFM image of a100 nm thick hightemperature AlN buffer layer grown on sapphire substrate at 1070 C, which is not sufficient for complete coalescence.
coalesce over the pits, it is believed that pit formation represents inherent problems with the growth and have a net adverse affect on the subsequently grown GaN layer. The surface topology of a 100 nm thick HT AlN buffer layer grown at T ¼ 1070 C, 30 Torr, H2 ¼ 6420 sccm, NH3 ¼ 7.2 sccm, and TMA ¼ 21.2 mmol min1 that resulted in a growth rate of 4 nm min1 is shown in Figure 3.103. A nanoporous surface with a RMS roughness of 0.29 nm over a 5 mm · 5 mm area is apparent. Ohba and Sato [556] reported on the details of the two-step high-temperature AlN nucleation layer process wherein relatively small V/III ratios were employed to eliminate pits. Specifically, the first-step nucleation layer step was performed at 1200 C followed by the second step at 1270 C, though temperatures as high as 1450 C have been explored. Use of a very low V/III ratio of 1.5 during the first step was deemed responsible for producing a fully two-dimensional AlN during the second step. Both slightly lower and higher V/III ratios, namely, 1.2 and 4.0, led to microcrystalline islands. A nearly pit-free smooth surface was obtained after the second-step growth at 1270 C. Second-step growth temperatures as high as 1450 C did not lead to better surfaces. Although it would seem that further improvement could be realized by even higher second-step growth temperatures, the premature vapor phase reactions are problematic. In short, among the parameter space of group V/III ratios of 1.2, 1.5, and 4 (performed at both 1200 and 1270 C for the second step) and temperatures of 1200, 1270, and 1450 C for the second step, the 1.5 and 1270 C combination produced the best surface morphologies, as judged by AFM images. The authors attributed the improvements in the two-step growth in part to the suppression of the substrate nitridation at the initial stages of growth, which seems to counter the other efforts where nitridation is deliberately performed. Typically,
3.5 The Art and Technology of Growth of Nitrides
because the thermal annealing takes place in NH3 environment, there is no need for special nitridation steps of sapphire prior to growth. The concept of low temperature AlN buffer layers has been extended by Nakamura [557] to include GaN. One may argue that the large lattice mismatch between the low-temperature AlN nucleation layer and the following GaN layer would cause generation of new defect and that the use of low-temperature GaN would remove the cause for this generation. The criterion whether a GaN or AlN nucleation layer should be used is not that simple in that nucleation characteristics of these two types are very different and AlN can be deposited at moderately high temperature with benefits to be garnered. As in the case of AlN, the crystalline quality of the GaN nucleation layer is initially poor but improves while heating to the growth temperature of the epitaxial GaN layer, commonly above 1000 C [558]. Although each process step along the way can be monitored in situ by MBE technique to the extent allowed by the availability of analytical tools installed, in a method like standard OMVPE, testing can be performed only after the entire layer structure has been grown and the sample taken out of the system. Having stated this, one must add that optical surface reflectometry can be used to determine how smooth the surface is to the extent that can be discerned considering the wavelength of the light used. Typically, critical and sequential steps such as nitridation, low-temperature buffer, high-temperature annealing, and layer growth must all be carried out before the final structure could be characterized. Only by careful and systematic layer growth experiments can any inferring back to each step be done. The solace, however, is that the growth can be terminated after the nucleation layer and its properties have been examined. AFM and X-ray analysis methods are used to determine the quality of nucleation layers. With the former, the surface morphology and the effect of annealing on it can be investigated. The latter provides data about the crystallinity of the nucleation layer. Typically, the low-temperature nucleation layers on sapphire are too disordered to yield useful X-ray data. In terms of the recipe used for the GaN nucleation layers at Virginia Commonwealth University, they were grown at approximately 550 C under different growth pressures, thicknesses, and V/III ratios. One recipe requires growth at 560 C, with V/III ratio of 2500, and pressure of 30 Torr. Experiments with varying nucleation layer thickness (grown with a V/III ratio of 2000 corresponding to a TMGa flow rate of 22.4 mmol min1) indicated a steady increase in the roughness up about 30 nm, beyond which the roughness saturated. The affect of the V/III ratio on the roughness is characterized by an initial smoothing followed by roughening. In the V/III ratio range of 1500–4000, the smoothest nucleation layers were obtained for a ratio of 2500. Hersee et al. [144] investigated the affect of the deposition temperature on the surface morphology of the GaN nucleation layer, as the quality of the surface is linked to the eventual quality of the subsequent high-temperature GaN layers. AFM images of nominally 20 nm thick, low-temperature GaN nucleation layers grown at 550 C showed the islanding growth characteristic of low-temperature AlN nucleation layers wherein the islandic nature diffused with that deposited at 480 C. To gain insight into the benefits of high-temperature annealing and the nature of the surface prior to the deposition of the main layers, AFM images of the 550 and 480 C buffer layers
j521
j 3 Growth and Growth Methods for Nitride Semiconductors
522
were taken after ramping the substrate temperature to 1025 C. The 550 C buffer showed larger, well-developed islands that were approximately 200–300 Å high. The buffer grown at 480 C was also modified by annealing but retained a much smoother morphology after ramping, with a typical peak to valley roughness of 10 nm. A comparison of the buffer layers deposited with 3 and 7 nm min1 showed that larger islands are formed for the higher temperatures and lower rates, which imply the mechanism to be kinetically limited island growth. The crystallinity of thin nucleation layers can be probed with XRD. The GaN(0 0 0 2) reflection obtained from one such annealed layer reveals broad X-ray rocking curve peaks with a sharper superimposed peak [144]. The broadening is consistent with the misorientation of the islands. The sharp superimposed peak at the center is indicative of a preferred orientation, which can be influenced by the rate of temperature ramp and annealing time to some extent. Using AFM images and XRD analyses as probes, the effect of a ramping procedure with durations between 0 and 20 min is such that the island size increases with ramping duration. The XRD measurements show the narrow high-intensity peak component growing in lieu of the broader part of the peak, indicating a more closely matched orientation. These measurements reveal that the GaN buffer layers initially consist of a dense array of islands (as in the case of AlN buffer layers [559]) whose size and crystallinity are limited by kinetics and affected by deposition temperature, growth rate, and ramp duration. Although generally not considered as an enabling technology and somewhat system dependent, Fuke et al. [560] and Wickenden et al. [52] focused on the effect of various nitridation conditions on the final epitaxial layers, which is also intricately connected to the nucleation layer, which early on went with nomenclature of lowtemperature buffer layer. As mentioned in Section 3.5.5, some practitioners, as a matter of a standard procedure, nitridate sapphire substrates before attempting a low-temperature nucleation layer. In the study of Wickenden et al. [52], a series of experiments to investigate the affect of nitridation temperature, process pressure, and V/III ratio used for the GaN nucleation layer on the main GaN film morphology and transport properties was performed, albeit on a-plane sapphire. To be valid, the low-temperature buffer layer or the NL, and epitaxial layer conditions were kept the same. Following the nucleation layer (the low-temperature buffer layer), the substrate temperature was ramped to the growth temperature in 5 min, and an epitaxial GaN film was grown just for 1 min. In one experiment, the nitridation temperature was changed from very low to high, namely 675–1065 C while using an NH3 flow rate of 2 slm1. In the second experiment, the NH3 flow rate was changed down to 0.5 slm1. TEM analysis of this epitaxial GaN on ð1 1 2 0Þ sapphire, instead of the c-plane, which is standard, nitridated at 675 and 1065 for 10 min revealed that the 1065 C treatment led to improved alignment of the crystalline grains to the sapphire substrate and to each other. In the case of a 10 min nitridation at 675 C, epitaxial GaN film showed features with a large degree of in-plane disorder, indicative of a mosaic structure. An unexpected 30 rotation of the normal orientation of the GaN film relative to the sapphire substrate, which is normal on c-plane sapphire, was also observed with the GaN ½1120==Al2 O3 ½1 1 0 0. When the nitridation was performed
3.5 The Art and Technology of Growth of Nitrides
for 1 min at 1065 C followed by a 9 min decrease in temperature to the lowtemperature buffer layer nucleation temperature, oriented films with GaN ½1 1 0 0==Al2 O3 ½1 1 0 0 resulted, and in-plane misorientation decreased. Increasing the nitridation temperature to 1065 C under these high-temperature exposure conditions and using 2 slm 1 NH3 flow and 1 min nucleation layer growth time (established under conditions of 550 C nitridation) caused a delay in the onset of nucleation of the nucleation layer and thus in its thickness. Truncated hexagonal features up to 50 mm in size were observed in subsequently grown 2 mm thick epitaxial GaN films. Specular epitaxial films resulted after increasing the NL growth time under the higher temperature nitridation conditions. This is in agreement with the work of Fuke et al. [560] and suggests that the nucleus density is severely restricted under conditions of high-temperature nitridation and insufficient NL thickness. When 0.5 slm1 ammonia flow rate was used for nitridation, GaN growth produced well-oriented hexagonal features of approximately 2 mm size. The GaN films showed mobilities of approximately 300 cm2 V1 s1 when the substrate temperature was set to an intermediate temperature of 800 C over a period of 30 min after nitridation and prior to NL growth. 3.5.5.1.1 The Effect of V/III Ratio on Nucleation Buffer Layer When the NL was done using TMG flow ¼ 40.2 mmol min1 and V/III ratio ¼ 2100 at a temperature of approximately 550 C, the NL features were of indefinite shape, poorly aligned, and of greater density than considered optimum for high-quality growth [52]. However, when the GaN film was grown using nearly identical NL and epitaxial growth conditions, but at a temperature of 1065 C after nitridation, the film formed after 1 min of epitaxial growth displayed features that are better aligned and more hexagonal in character. When NH3 flow was reduced to yield a V/III ratio of 1500, the flow reduction was compensated by an increase in the H2 carrier gas flow to keep the total process flow in the reactor constant, the density of the nuclei was reduced, which may allow subsequent epitaxial growth to form larger grains, with associated reduction in the formation of threading dislocations. Too high a nucleation density is believed to lead to a greater overlap of the crystallites at the early stages of growth and a loss of the hexagonal morphology. Shown in Figure 3.104 is a series of AFM images of the GaN layers grown for 1 min only, representing 550 C nitridation with high V/III ratio, 1065 C nitridation with medium to high V/III ratio in three steps. As can be seen, high V/III ratios during the NL growth leads to high density of nuclei, which is not desirable. Optimized growth in this set of experiments using 43 mmol min1 TMG flow was found at a V/III ratio of 520. Combined optimization of sapphire nitridation and NL V/III ratio reduced Nd–Na in epitaxial GaN films by an order of magnitude, from 1018 to 1 · 1017 cm3, and improved the mobility of GaN : Si films by a factor of three or more to 300 cm2 V1 s1 regime for n ¼ 3 · 1017 cm3. This improvement in transport properties is attributed to improved grain alignment and larger grain size in the epitaxial GaN film owing to control of the nuclei density. Just about every group conducting research and development of GaN and related structures has reported on the low-temperature nucleation buffer layers and/or their
j523
j 3 Growth and Growth Methods for Nitride Semiconductors
524
Figure 3.104 AFM images of 3 mm · 3 mm areas of 1 min GaN growths, under varying conditions of nitridation, NL TMG flow rate, and NL V/III ratio; (a) 550 C nitridation, TMG ¼ 40 mmol min1, V/III ¼ 2100, (b) 1065 C nitridation,
TMG ¼ 43 mmol min1, V/III ¼ 1500, (c) 1065 C nitridation, TMG ¼ 43 mmol min1, V/III ¼ 2100, (d) 1065 C nitridation, TMG ¼ 43 mmol min1, V/III ¼ 2600. Thermal ramp and epitaxial growth conditions were held constant for all films [52].
impact on the eventual semiconductor nitride structure down to the point of the grain structure [432]. Consistent with the work of Wickenden et al. [52], Yi et al. [561] reported that buffer layers grown at or below 500 C are composed of islands and smooth regions that are misaligned. However, the layers grown at or above 650 C are of islandic nature and presumably better aligned as compared to the aforementioned ones. Buffer layers grown at about 470 C showed 3.2 eV PL peaks, which could be associated with cubic GaN. However, a peak at that particular energy is also observed in wurtzitic GaN and often associated with Zn impurity, as discussed in Volume 2, Chapter 5. However, when the buffer layer is grown at 500 C, the PL peak blue shifts to that of wurtzitic GaN but a broad defect peak around 2.5 eV also appears. When the buffer layer is grown at about 650 C, the 2.5 eV peak associated with defects did not appear in the linear intensity PL scan of Yi et al. [561], which most likely means that its intensity was reduced. However, linear scale is not the best format to characterize weaker yellow peaks.
3.5 The Art and Technology of Growth of Nitrides
As elaborated on by Wickenden et al. [52], the ramping time after the buffer growth is critical in that small and misaligned islands are removed during this process. Sugiura et al. [562] investigated the effect of ramping time from the buffer growth temperature of 550 C to the growth temperature of 1100 C. Provided that Sugiura et al. [562] were able to maintain all the other growth parameters exactly the same in the series of layers investigated, the mobility (increased from about 550 cm2 V1 s1 for 12 min), the symmetric X-ray diffraction peak (FWHM decreased from about 6 to about 3.5 arcmin), and PL were all optimized in a 5 mm thick GaN film after a 12 min ramping time under H2. The best surface morphology was obtained when the temperature was ramped to between 850 and 900 C over 1 min and retained there for 10 min followed by ramping to 1100 C over 1 min again for the final GaN layer growth. Multiple low-temperature layers can be inserted for improved quality, which is discussed in Section 3.5.5.5. In another study, the effect of the thickness of the lowtemperature buffer layer on a 4 mm thick GaN layer was investigated in terms of the FWHM of the symmetric X-ray diffraction peak. For a 20 nm, that peak width was reported to be about 5 arcmin with a steep rise on either side of the 20 nm thickness [563]. Other studies concluded that about 40 nm thickness was optimum. In another investigation, the effect of the deposition rate during the low-temperature buffer layer was investigated [564]. Electron mobility in the top GaN layer increased from about 54 cm2 V1 s1 for a deposition rate of 4.56 nm min1 to 539 cm2 V1 s1 for a deposition rate of 18.3 nm min1. The electron concentration for the same two deposition conditions improved from 2 · 1018 to about 8 · 1016 cm3. Moreover, the asymmetric X-ray diffraction peak (1 0 2) (in the hkl notation and ð1 0 1 2Þ in the hjkl notation), dropped from about 12 arcmin to just under 7 arcmin, and the dislocation density near the surface dropped from 6 · 109 to about 2 · 109 cm2 for the same parametric change. It should be mentioned that the exact thickness of the optimal low-temperature buffer layer may depend on the particulars of the growth, such as the deposition rate, temperature, and V/III ratio, and of the buffer layer. 3.5.5.1.2 Effect of Epitaxial Growth Temperature Experiments done on ELO, discussed in Section 3.5.5.2, of GaN over a patterned SiO2 layer has demonstrated that increased temperature enhances lateral growth of epitaxial GaN in [2 1 0] (in the hkl notation and ½2 1 3 0 in the hjkl notation) directions. In one set of experiments, Wickenden et al. employed substrate temperatures of Tg ¼ 1030 and 1065 C, the latter leadingtomoreoptimizedepitaxialGaNfilms.Theenhancedlateralgrowthoftheinitial epitaxial growth on this template, owing to the higher temperature employed, may have allowed the formation of larger grains, and paved the way for increased mobility in the GaN film grown on top [52]. The V/III ratio for the epitaxial GaN films grown at 1065 C was increased to 2775, from the V/III ratio of 2100 used for Tg ¼ 1030 C, to compensate for any increased nitrogen desorption during growth at increased temperature. 3.5.5.1.3 Effect of Process Pressure The morphology and electron transport properties of the GaN layers critically depend on the reactor pressure [52]. Figure 3.105 illustrates the morphological differences of three GaN films grown using varying conditions of pressure. In all cases, nitridation was performed under identical
j525
j 3 Growth and Growth Methods for Nitride Semiconductors
526
Figure 3.105 AFM images of 1 mm · 1 mm areas (upper view) and 7 mm · 7 mm areas (lower view) of 10 min GaN growth under varying pressure conditions, holding nitridation and NL growth conditions constant; (a) thermal ramp and epitaxial growth at 150 Torr, (b) thermal ramp at 76 Torr, epitaxial growth at 150 Torr, (c) thermal ramp and epitaxial growth at 76 Torr [52].
optimized conditions of 0.5 slm1 NH3 with a nitridation temperature of 800 C. The NL was grown under identical optimized conditions of TNL ¼ 550 C, a TMG molar flow of 43 mmol min1, V/III ratio ¼ 520, and P ¼ 76 Torr. Following NL, the substrate temperature was ramped to the growth temperature, Tg, over a 30 min period, with the NH3 flow adjusted to the level used for epitaxial growth. In each case, the epitaxial GaN film was grown for 10 min at Tg ¼ 1065 C, with a TMG molar flow of 32 mmol min1 and V/III ratio ¼ 2775. The pressure of the ramp from TNL to Tg and the epitaxial growth pressure were varied, being done at either 76 or 150 Torr. The film illustrated in Figure 3.105a was grown under the same conditions, which in other studies led to an electron mobility of 550 cm2 V1 s1, and displayed well aligned, large grains and a dislocation density of 108 cm2. After the growth of the NL at 76 Torr, the NH3 flow was increased and the chamber reactor pressure was reset to 150 Torr for the thermal ramp to the growth temperature, Tg. The epitaxial growth was also performed at 150 Torr. The epitaxial growth comprised of features of the form of distinct hexagonal morphology and is of a fairly narrow size distribution at the stage of growth just prior to the epitaxial film coalescence. The high-resolution (1 mm · 1 mm) AFM image of the top of one of the features indicates step-flow growth, with a RMS roughness of 3 Å over alength of 1000 Å, indicative of biatomic stepheights. Very few, if any, threading dislocations are seen in the film at this stage of growth, supporting the reduced dislocation density observed in the 1.5 mm thick GaN film. This growth morphology closely resembles that reported for ELO growth of GaN films.
3.5 The Art and Technology of Growth of Nitrides
In contrast, Figure 3.105c illustrates film growth where both the thermal ramp and the epitaxial growth were done at a pressure of 76 Torr. The larger-scale AFM image of the film indicates reduced roughness compared to the film illustrated in Figure 3.105a, which was grown under higher pressure conditions. Again, the epitaxial GaN film appears to have coalesced prematurely. The AFM image of a 1 mm · 1 mm area shows a large number of step terminations and threading dislocations. Thick epitaxial GaN films grown under these conditions typically displayed mobilities on the order of 300 cm2 V1 s1. Figure 3.105b illustrates a film in which the epitaxial growth was done at a pressure of 150 Torr, as in Figure 3.105a, subsequent to a thermal ramp that was done at a pressure of 76 Torr. The nature of the epitaxial film features appear to be similar to that of the film in Figure 3.105a, although with a greater size distribution. The AFM analysis over a 1 mm · 1 mm area on the top of one of these features reveals the presence of an intermediate number of step terminations and threading dislocations, possibly caused by nonoptimal film coalescence. In the way of analyzing the results, it appears that the increased pressure influences the transformation the GaN nuclei during the thermal ramp from the nucleation layer step to the growth of the active layer. As mentioned in Section 3.4.1.3.1, at sufficiently high temperatures such as those employed during ramping, GaN decomposes and its decomposition rate in H2 is enhanced at pressures greater than 100 Torr at temperatures near 900 C. In this case, Ga droplets form. The comparison of the growth morphologies illustrated in Figure 3.105a and b supports the notion that increased pressure during the thermal ramp enhances the decomposition rate of GaN and results in the decomposition of smaller nuclei. This enhanced decomposition and the associated enhanced gallium diffusion may facilitate the lateral expansion of large (also low-density) nuclei of a relatively narrow size distribution. As shown in Figure 3.105a, subsequent high-temperature epitaxial growth on these large nuclei is conducive for coalescence with improved registry, resulting in larger-than-average grain size and fewer extended defect density than that observed in the films shown in Figure 3.105b and c. Increased pressure also appears to cause a reduction in the epitaxial GaN growth rate, which is consistent with increased GaN decomposition, and an optimum GaN decomposition may promote a step-flow type of growth mode, as illustrated in Figure 3.105a, and would result in lower point defect density. At lower pressures, the continuous, random formation and growth of GaN nuclei may be responsible for the premature coalescence in the initial epitaxial growth seen in Figure 3.105c. This could cause the increased density of step terminations, threading dislocations, and subsequent reduced mobility in thicker films grown at lower pressure. In short, the combined effects of optimized nuclei formation and suppressed renucleation during epitaxial growth at higher pressures result in the increased quality of the films. Other approaches such as Indium doping, high-pressure growth, and three-step growth have been explored to aid the epitaxial process. It is well known that Indium acts as a surfactant in the growth of GaN in addition to its similar role in other conventional compound semiconductors. Naturally, In incorporates in the lattice if temperature is lowered. However, at temperatures typical of GaN growth, In
j527
j 3 Growth and Growth Methods for Nitride Semiconductors
528
incorporation is nil. Despite the fact that In does not incorporate in the lattice when GaN growth temperature is employed, presence of In-containing species in the vapor phase and the role of In on the growing surface cause a decrease in the threading dislocation density. In addition, the PL linewidth gets narrower and the free exciton recombination lifetime increases. Moreover, the effect of In in strain relief should also be considered [565–568]. Additional benefits of the presence of In in the vapor include the reported reduction in the background electron concentration [569]. Further improvements in the crystal quality can be garnered by in situ annealing during the early stage of growth [570]. In some OMVPE experiments, growth at pressures exceeding the atmospheric pressure (up to 1.6 atm) has been shown to reduce the etch pitch density (EPD), which is a measure of defect concentration [571]. A three-step process that involved an AlN layer grown by atomic layer epitaxy (ALE) and the standard two-step growth have been shown to lead to high-quality templates with TD densities in the low 108 cm2 range [572]. No matter what technology or combination thereof is used, the TD density in GaN is limited to about mid 107 cm2. In MBE, the insertion of multistacked AlN/GaN quantum dots decreases the TD density significantly, but the same limitation still applies. The basic mechanism involved is believed to be the termination of TDs in the quantum dots [573]. Further reduction in dislocation density requires steps aimed at reducing defects such as epitaxial lateral overgrowth and preferably hundreds of microns thick GaN layers grown by HVPE. The effect of the growth rate of the GaN nucleation layer on the growth of a hightemperature GaN layer has been studied with the aid of in situ normal incidence reflectance also [574]. The lateral growth and coalescence of a high-temperature GaN layer was found to be enhanced as the growth rate of the nucleation layer was increased. However, the measurement of (1 0 2) hk circle scan using X-ray diffraction and transport measurements showed that the in-plane structural quality of GaN deteriorated with increased growth rate of the nucleation layer. These results suggest that the nucleation sites are well oriented on sapphire for low nucleation layer growth rates but that the lateral growth and coalescence of the GaN layer are hindered owing to the limited surface diffusion of adatoms at a fast growth rate. 3.5.5.2 Epitaxial Lateral Overgrowth To reiterate, heteroepitaxial deposition of GaN on low-temperature GaN or AlN buffer layers on Al2O3 and SiC substrates, or other mismatched substrates, results in films containing high dislocation densities (108–1010 cm2). As discussed in detail so far in this chapter and regardless of the deposition method employed, this high concentration of dislocations result from the lattice mismatch between the buffer layer and the film and/or the buffer layer and the substrate. Without any doubt, the high defect concentration limits device performance. Although the intermediate stages of the ELO process cause inhomogeneous impurity incorporation and stress distribution it produces high-quality GaN, with threading dislocation (TD) densities in the mid 106 cm2, line widths of the near bandgap recombination peaks in low-temperature PL below 1 meV and deep electron traps concentration below 1014 cm3 (compared to mid 1015 cm3 in standard GaN). Numerous modifications of ELO, which also goes
3.5 The Art and Technology of Growth of Nitrides
by ELO or ELOG, with the pseudomaskless, meaning the dielectric mask is removed prior to the second epitaxial step, version of it dubbed pendeo-epitaxy, are discussed in this section. A complete review of the topic including inhomogeneous strain distribution and its effects can be found in Ref. [575]. Papers reviewing the method in regard to standard OMVPE ELO [576,577] and on Si(1 1 1) can be found in the literature [578]. The genesis of ELO, also selective area growth (SAG), which refers to the same concept, can be traced to Si [579] and other conventional semiconductors prior to its use in GaN, which broke the bottleneck in relation to GaN-based laser longevities. The method relies on the growth of GaN on windows opened in a dielectric material such as SiO2 followed by lateral extension and coalescence. The growth conditions are optimized to support the lateral growth until coalescence. The ELO method has been demonstrated in material systems such as Si [579–581], GaAs on Si [582,583], InGaAs on GaAs [584], and InP on Si [585]. It really relies on growth anisotropy in the form of different growth rates on different crystallographic planes. In the case of GaAs, which is a predecessor of GaN, the surface kinetics showed about a difference of about two order of magnitude between the growth rates of {111}Ga and f1 1 1gAs surfaces. Growth anisotropy in OMVPE also occurs but to a limited extent, as evidenced by growth on preferentially etched or masked GaAs substrates or on GaAs spheres [586]. In OMVPE, growth anisotropy occurs as a result of preferential incorporation of diffusing molecular species on different crystallographic surfaces. This is in spite of uniform flows, as incorporation occurs after kinetically controlled diffusion of adsorbed species on the surface. OMVPE is a process occurring at very high supersaturation and the growth rate is controlled by the diffusion of active species through a boundary layer. After the lithographically defined opening in the dielectric mask is filled with growth, epitaxial lateral overgrowth occurs owing to combined effects of selective epitaxy and growth anisotropy. Selective epitaxial growth of GaN on sapphire was reported as early as 1994 [587]. The technique ELO of GaN layers using patterned SiO2 has been reported [588]. This was followed by a transmission electron microscopic study that revealed that these overgrown regions of pyramids contained much lower density of dislocations [589]. Using the ELO technique, Nakamura reported laser diode lifetimes of about 10 000 h at room temperature [590]. This is substantial in that the longevity of laser diodes hovered around 300 h prior to ELO. The first step in the ELO process is the growth of a few micron thick GaN layer on sapphire, SiC, or on SiC/Si followed by deposition of a dielectric (SiO2 or SiN) mask by CVD or PECVD. A set of parallel stripes, separated by window areas is then opened in the mask using standard photolithography. During the initial regrowth, either in OMVPE [589], HVPE [591], or sublimation growth [592,593], selective area epitaxy is achieved in the windows without any nucleation on the dielectric mask. If the growth parameters are then chosen correctly, and once the GaN growing film reaches the top of the stripes, epitaxial lateral growth takes place over the mask and finally leads to a full coalescence culminating in a smooth surface. The basic concept is shown in Figure 3.106, which of course has the benefit of filtering of the defects. Above the windows, the microstructure of the underlying GaN template is reproduced with its
j529
j 3 Growth and Growth Methods for Nitride Semiconductors
530
Mask Mask
Mask
Mask
Substrate
(a) Single-step ELO
Mask
Mask
GaN template Mask
Mask
substrate
(b) Two-step ELO
Figure 3.106 Schematic representation of single step (one step) ELO (a) and two-step ELO (b). The vertical lines represent threading dislocations.
defect structure to a first extent. In the laterally grown regions (wings, what used to be the area covered by the mask), however, the material contains much fewer defects, with defects concentrated at the coalesced regions of the wings, as growth over the mask area is seeded by the sidewalls of the epitaxial layer growing out in the open windows. This is the one-step ELO. In two-step ELO, an additional masking and regrowth is performed in a such as way that what used to be unmasked areas are masked, which in theory would lead to uniformly extended defect-free material. Zheleva et al. [589] analyzed the structural defects in these GaN pyramids. In the region over the windows, the dislocations have a density comparable to that of the GaN seed layer as mentioned previously. Most of the extended dislocations propagate in the growth direction through the GaN. The dislocation density is drastically reduced within a pyramidal volume, which ends at approximately one third of the pyramid height. The part of the pyramidal region above the masked area contains a much reduced dislocation density. In one particular investigation, substrate templates for the lateral epitaxy studies are prepared by depositing 1–2 mm thick GaN layers at 1000 C on high-temperature (1100 C) AlN buffer layers, which, in turn, were grown on 6H-SiC substrates, or lowtemperature AlN or GaN nucleation layers. A SiO2 mask layer (thickness ¼ 1000 Å) is subsequently deposited on GaN/AlN/6H-SiC(0 0 0 1), assuming SiC is the substrate and patterned by standard photolithography techniques and etching in a buffered HF solution. In one reported case [587], the pattern consisted of 3 mm wide, parallel stripe openings with a 7 mm pitch, which were oriented along the h1 1 2 0i and h1 1 0 0i directions in each GaN film. Prior to lateral overgrowth, the patterned samples were dipped in a 50% buffered HCl solution to clean the underlying GaN layer. The lateral overgrowth of GaN was achieved at 1000–1100 C and 45 Torr. Triethylgallium (13–39 mmol min 1) and NH3 (1500 sccm) precursors were used in combination with a 3000 sccm of H2 diluent. The second lateral epitaxial overgrowth was conducted on the first laterally grown layer via the repetition of SiO2 deposition, lithography, and lateral epitaxy. Figure 3.107 shows the representative cross-sectional SEM images of two GaN stripes along h1 1 2 0i and h1 1 0 0i directions selectively grown for 60 min. Truncated
3.5 The Art and Technology of Growth of Nitrides
Figure 3.107 Scanning electron micrographs showing the morphologies of GaN layers grown on stripe openings oriented along (a) < 1 1 2 0> and (b) < 1 1 0 0> directions. Courtesy of R. Davis.
triangular stripes that have ð1 1 0 0Þ slanted facets and a narrow (0 0 0 1) top facet were observed for window openings along the h1 1 2 0i direction. Rectangular stripes with a (0 0 0 1) top facet, ð1 1 2 0Þ vertical side facet, and a ð1 1 0 0Þ slanted facet appear to have been developed in samples grown on stripes along the h1 1 0 0i direction. SEM observations of GaN stripes for different growth durations of up to 3 min revealed similar morphologies regardless of stripe orientation. The amount of lateral growth exhibited a strong dependence on stripe orientation. Results obtained under various growth conditions showed that the lateral growth rate perpendicular to the h1 1 0 0i direction was much faster than those perpendicular to the h1 1 2 0i direction. Dependence of morphological development on the window orientation has been attributed to the stability of the crystallographic planes in the GaN structure. Stripes oriented along the h1 1 2 0i direction always had wide ð1 1 0 0Þ slanted facets and either a very narrow or no (0 0 0 1) top facet, depending on the growth conditions. The stability of the ð1 1 0 1Þ plane of the wurtzitic GaN and the associated low growth rate of this plane are thought to be reasons for the observations. As shown in Figure 3.107b, the f1 1 0 1g planes of the h1 1 0 0i oriented stripes were wavy, which points to the coexistence of several Miller index planes. It is believed that competitive growth of selected f1 1 0 1g planes occurs during the deposition, which causes these planes to become unstable and their growth rate to increase relative to that of the ð1 1 0 1Þ plane of stripes oriented along h1 1 2 0i. The lateral overgrowth rates and the equilibrium facets developed in the process are strong functions of both the crystallographic orientation of the stripe openings and the growth parameters, such as the temperature, V/III ratio in the vapor phase, reactor pressure, composition of the carrier gas (H2, N2), and the fill factor (defined as the ratio of the stripe opening and the pattern period) [594–598]. Because the lateral expansion is easier when the stripes are aligned along one of the h1 1 0 0iGaN directions [599], the morphology gradually changes from triangular stripes to that with rectangular cross section with (0 0 0 1) top facet and f1 1 2 0g sidewalls as the growth temperature increases [595]. In contrast, when the stripes are along the h1 1 2 0i direction, the extent of the lateral overgrowth is limited by a slow growth rate along the f1 1 0 1g facets, which are the most stable facets in GaN. In short, the lateral
j531
j 3 Growth and Growth Methods for Nitride Semiconductors
532
0i growth rate along the h1 1 0 0i stripes is much faster than that along the h1 1 2 stripes [594,600,601]. Hiramatsu et al. [598] investigated the morphology of ELO GaN using stripes along the h1 1 2 0i or h1 1 0 0i directions as a function of the pressure (P) and temperature (Ts). For stripes along h1 1 2 0i, triangular stripes with f1 1 0 1g facets are formed independent of the pressure and temperature. For stripes along h1 1 0 0i, the final shape turned out to be dependent significantly on the growth conditions. This is illustrated in Figure 3.108, which displays the morphologies of ELO structures as a
Figure 3.108 (a) SEM images of lateral epitaxial overgrowth of GaN using stripes oriented along the < 1 1 0 0> direction at different temperatures and pressures; (b) schematic evolution of the morphology of ELO of GaN using stripes oriented along the <1 1 0 0> direction as a function of temperature and pressures [598].
3.5 The Art and Technology of Growth of Nitrides
function of temperature and pressure. The diagram can be divided in four regimes as follows (shown schematically in Figure 3.108): (i) which depicts the case for Ts < 925 C, the morphology is poor. (ii) the sidewall of the triangular stripes have f1 1 2 2g facets, as in i, but the surface is smooth. (iii) which depicts the case for increasing Ts or decreasing P as compared to regime ii, the sidewalls evolve from f1 1 2 2g facets to vertical f1 1 2 0g prismatice planes, as previously reported by Nam et al. [588,602], as a function of Ts. (iv) which depicts the case for even lower P or higher Ts, the (0 0 0 1) surface becomes rough. An artistic view of the overall evolution of the ELO process from the beginning to the end is shown in Figure 3.109. Eventually, flat surfaces are obtained when full coalescence extending from the f1 1 2 0g prismatic planes is obtained, which is possible when the ratio of the lateral and vertical growth rates is sufficiently high. In addition to increase in the growth temperature (>1100 C) [603] to promote the anisotropic growth rates, introduction of (CH3Cp)2Mg in the vapor phase allows one to maintain the growth temperature in the conventional range of 1000–1100 C, which has been employed [604]. It has been shown that the addition of Mg in the vapor phase enhances the ratio of the growth rates along the h1 1 0 0i direction and over windows along the h0 0 0 1i direction, respectively. Beyond the orientation of the stripes, temperature, and pressure, the composition of the carrier gas (H2 versus N2), and in particular the flow rate of the metalalkyl, which in this case is TEG, influences the lateral overgrowth, the growth selectivity, and the morphological evolution of the GaN stripes as shown in Figure 3.110. Using stripes along h1 1 0 0i direction, Tadatomo [596] and Kawaguchi [605–607] concluded that H2 as a carrier gas produces smooth surfaces, but with a low lateral growth rate. On the contrary, N2 enhances the lateral growth rate, but the surface quality is poor. A compromise of a mixture of H2 and N2, 1 : 1, allows acceptable growth rates and surface morphologies. As for the reactant gas composition, an increase in the flow increased the growth rate of the stripes in both the lateral and the vertical directions. However, the lateral/vertical growth rate ratio decreased from 1.7 at a TEG flow rate of 13 mmol min 1 to 0.86 at 39 mmol min 1. The considerable increase in the concentration of the Ga species on the surface may sufficiently impede their diffusion to the f1 1 0 1g planes such that chemisorption and GaN growth occur more readily on the (0 0 0 1) plane. The morphologies of the GaN layers were also a strong function of the growth temperatures. Stripes grown at 1000 C possessed a truncated triangular shape. This morphology gradually changed to the rectangular cross section as the growth temperature was increased. Modulation of the V/III ratio can also be employed to control the morphology of overgrown stripes. This was done by Zhang [608] using NH3 flow modulation. This can be implemented by introducing controlled interruptions of the NH3 flow in the growth process. The lateral growth increases with the duration of the flow interruption. It is possible to tune the growth conditions to enhance the lateral growth and to get full coalescence. Fareed et al. [609] also proposed this flow modulation approach
j533
j 3 Growth and Growth Methods for Nitride Semiconductors
534
Ga precursor Ga diffusion SiO 2 GaN – <1 1 2 0>
C-sapphire Ga precursor Ga diffusion
Selective growth GaN C-sapphire
High-growth rate GaN C-sapphire {1101} facet
Slow-growth rate
GaN C-sapphire
Lateral overgrowth
GaN C-sapphire Figure 3.109 Schematic representation of the evolution of the ELO growth process. At first the growth initiates at the boundary between the mask and GaN followed by eventual expansion of the grown region to cover the entire opening in the mask. This is followed by trapezoidal growth, which is in turn followed by lateral growth.
for the ELO on 6H-SiC with SiO2 mask stripes patterned along the h1 1 2 0i direction. Using different NH3 flow-off times, an evolution of the morphology from triangular stripes with f1 1 0 1g facets to rectangular f1 1 0 0g facets was observed. The rectangular shape was obtained for a 5 s NH3 flow interruption in a 8 s cycle, which corresponds to a lateral-to-vertical growth rate ratio of 4 : 1. It should be noted that this evolution is very similar to the one displayed in region II of Figure 3.108.
3.5 The Art and Technology of Growth of Nitrides
Figure 3.110 Scanning electron micrographs of the < 1 1 0 0> oriented GaN stripes grown at TEG flow rates of (a) 13 mmol min 1 and (b) 39 mmol min 1 for 60 min. Courtesy of R. Davis.
In early ELO experiments, the lateral growth was induced from the inception of the regrowth to achieve coalescence and flat surface simultaneously rather rapidly. With this process, low defect density GaN was limited to the region of lateral expansion above the mask, excluding the coalescence boundaries [610], whereas the GaN above the window stripes replicates the defect density of the GaN template. The situation is, however, more complex in that dislocation half loops are possibly created in the single-step ELO technology. Cross-sectional TEM images along the ½1 0 1 0 zone axis (images projected along the direction of the stripes) for ELO GaN indicate that lateral expansions can be induced by a growth temperature of 1120 C with smooth and full coalescence resulting in 1.75 mm thick films. The dislocations under the mask terminate at the amorphous mask, but they propagate above the windows in the top layer. In one-step ELO, at the interface between two laterally grown GaN where the wings merge, extended defects such as threading dislocations and dislocation half loops are generated. In wing regions but away from the interfaces, the overgrowth does not adhere to the amorphous mask and voids can be observed between the mask and the overgrown GaN. These voids do not deteriorate the top material quality or the flatness of the resulting film. As a side remark, an AFM scan of a one-step ELO GaN on Si (1 1 1) revealed that regions grown in the openings of the mask have high TD densities; wings have almost no observable TDs and fully coalesced boundaries. For ELO with stripes along the h1 1 0 0i direction, threading dislocations originating from the GaN template layer propagate to the surface of the regrown GaN and the dislocation density in the overgrown GaN above the mask is lower. With stripes along the h1 1 2 0i direction, ELO stripes exhibit a triangular cross section causing the dislocations originating from the GaN template to bend at 90 when they encounter the growing f1 1 0 1g facets. The threading dislocations are also bent in triangular stripes with f1 1 2 2g lateral facets obtained with h1 1 0 0i stripes when growth conditions correspond to region II in Figure 3.108. The bending mechanism of threading dislocations was observed to cause a sizeable reduction in threading dislocation density in the pyramids obtained by selective area epitaxy (SAE) [589]. The same behavior has also been observed during mass transport in GaN [611], the evidence for which comes from the investigation of trench stripes patterned on the
j535
j 3 Growth and Growth Methods for Nitride Semiconductors
536
0i direction followed by annealing at surface of an OMVPE GaN along the h1 1 2 1100 C under N2 þ þ NH3. During this process, Ga atoms diffuse on the convex part of the surface and are then incorporated at the concave part. Following the formation of f1 1 0 1g facets, higher index facets appear and the trench is gradually buried with ensuing dislocation-free regions. As in the ELO process, threading dislocations bend at 90 when they reach the f1 1 0 1g facets and minimize the free energy of the system by bending toward the free surface. Gibart et al. [575] schematically described the dislocation bending in both one-step and two-step ELO processes in conjunction with (CH3Cp)2Mg-augmented ELO, which allows reduction in growth temperature [604]. The first step in the process is the growth of GaN at 1040 C as in the standard ELO process [575]. The growth rate Gc of the top (0 0 0 1) facet (here c stands for growth along the c-direction) is higher than the growth rate Gs of the inclined f1 1 2 2g lateral facets (here s stands for sidewall). During this stage, the growth rate ratio of the (0 0 0 1) plane is Gc/Gs ¼ 3. This first step is pursued until the top facet vanishes completely. The formation of the apex is critical. At this stage, the cross-sectional shape of the ELO GaN is triangular with f1 2 1 2g facets as shown in Figure 3.111. Then, in the second step, the lateral growth is favored (the growth rate of the lateral facets is higher than the growth rate of the top facet) by either introducing (CH3Cp)2Mg in the vapor phase at the same temperature or by increasing the temperature up to 1120 C until complete coalescence and smoothing of the surface is achieved. TEM and AFM microstructural studies on ELO GaN obtained from coalescence and bonding of f1 1 2 0g sidewalls, initiated from h1 1 0 0i stripes, indicate that threading dislocation densities are below 5 106 cm 2 and that the ELO GaN is free Convergingtrajectories of{1 1 2 2}facetedgeson (0 0 0 1)
B'
B
{1 1 2 2}facets SiN x ,SiO 2 Mask
q
a
C'
A'
O
A
C
GaN template
Substrate Figure 3.111 Schematic representation of the one-step ELO process at the end of the first step (vertical growth rate higher than lateral growth rate). The shape is triangular with f1 2 1 2g facet. y is the angle between [0 0 0 1] and ½1 1 2 2, GC the growth rate of the (0 0 0 1) plane, GS the
growth rate of ð1 1 2 2Þ, GC/GS ¼ 3, depends on the anisotropy of the growth. Red lines represent the behavior of threading dislocations inside the triangular stripes. Courtesy of P. Gibart and after Ref. [575].
3.5 The Art and Technology of Growth of Nitrides
of mixed-character dislocations [612,613]. This is to be contrasted to GaN grown in the windows in the mask that has the same mixed and pure edge threading dislocation density as the GaN template layer. A few pure edge dislocations having a line direction on the basal plane along the h1 1 0 0i direction appear above the edges of the mask. The inhomogeneous nature of ELO GaN both topographically and materially, such as patterned stripes, substrate (sapphire, SiC), dielectric mask (SiO2 or SiNx), and GaN all with different thermal expansion coefficients, warrants a deeper investigation of strain and stress distribution as well as impurity and point defect distribution. The magnitude and spatial distribution or stress has been modeled for the system GaN/AlN 6H-SiC with SiO2 mask using finite-element analysis [614]. It was shown that localized compressive stress fields up to 3 GPa occur in the vicinity of the GaN/ dielectric interface and the edge of ELO GaN. Even though ELO GaN has lower dislocation density than uniformly grown GaN, internal stress fields are believed to assist in the generation of horizontal dislocations [615]. Plan view TEM images on the topmost part of an ELO structure also reveal arrays of dislocations between the coherent and wing regions, which to a first extent run along the stripe direction [616,617]. Similar in-plane dislocations have been observed midway over the mask at the coalesced region. These dislocations have segments close to the boundary, which bend, in the form of half loops into the wings [616]. These dislocations may be caused by shear stress developed during and after growth owing to thermal coefficient mismatches between the three different materials alluded to above. To gain some insight, Hacke et al. [618] investigated ELO GaN/SiC by cathodoluminescence (CL) and observed nonradiative lines at 30 with respect to the h1 1 0 0i stripes, which are attributed to screw dislocations forming loops. Where the wings meet, a pinhole with f1 1 0 1g facets is often observed. Misorientation owing, for example, to sagging and also mismatch in the registry between the two coalescing GaN wings also result in grain boundaries with much larger tilt and twist components than those observed in GaN/sapphire. These tilts have been observed for both OMVPE and HVPE ELO GaN with either h1 1 0 0i or h1 1 2 0i oriented stripes [612,619]. X-ray diffraction, which can discern the degree of tilts or sagging, which are normally on the order of 1 , has been instrumental in fine tuning the growth conditions to lower tilting. Small tilt angles, 0.1 , have been achieved by either increasing the thickness [620] or tailoring growth conditions of uncoalesced stripes to get a low wing tilt and then quickly achieve coalescence in a second step by changing the growth conditions [621]. 3.5.5.2.1 Selective Epitaxial Growth and Lateral Epitaxial Overgrowth with HVPE One of the advantages of HVPE is its high growth rate (30–100 mm h1), which is suitable for producing bulk GaN templates including freestanding varieties. However, in GaN films on sapphire that are 20 mm thick, cracks occur because of strain. Selective area epitaxy/ELO reduces and changes the strain distribution, which could have a bearing on cracking. Lateral epitaxial overgrowth, was demonstrated Kato et al. [622] who showed that f1 1 0 1g facets appear in the grown GaN in the oxide stripe openings are along the h1 1 2 0i direction. Fully coalesced layers were eventually
j537
j 3 Growth and Growth Methods for Nitride Semiconductors
538
obtained from extending laterally from the stripes along the h1 1 2 0i direction with widths ranging from 1 to 4 mm and a period of 7 mm after several tens of microns of growth [623]. As in the case of OMVPE, the dislocation density was reduced from 109–1010 cm 2 to 5 107 cm 2 and continued to decrease for larger GaN thicknesses, with the best figure of 8 106 cm 2 obtained for a 560 mm thick layer [624]. As expected, nearly all the dislocations propagated into the HVPE layer from the OMVPE template, which contained 109 cm 2 dislocations consisting mainly of pure edge, with some mixed and very few screw dislocations, in the windows with no new ones caused by the regrowth process. As in ELO, dislocations originating from the openings are bent at 90 and afterwards propagate laterally, never reaching the surface [591]. Moreover, the dislocations originating from the OMVPE template are piled-up on the mask, which was described with the coined term facet-initiated ELO (FIELO). Coalescence of the wing regions in HVPE, as in OMVPE ELO, does not occur in the same manner for stripes oriented along the h1 1 0 0i and h1 1 2 0i directions. [623,625–627]. Triangular stripes with f1 1 0 1g side facets are obtained when the stripe windows are along the h1 1 2 0i direction and rectangular stripes with f1 1 2 0g facets are obtained when the stripe windows are along the h1 1 0 0i direction. The ELO growth with windows along the h1 1 2 0i direction does not result in a flat surface unless a very thick layer is grown. On the contrary, when the stripes are along the h1 1 0 0i direction, smooth surfaces can be obtained [628]. For stripes along h1 1 0 0i, the facet structure not only depends on the temperature but also of the carrier gas, H2 þ N2 mixture. For pure N2, the threading dislocations do not experience the 90 bending observed in OMVPE ELO, whereas a mixture of N2 þ H2, 1 : 1, carrier gas results in a more efficient reduction of threading dislocations via the 90 bending process [629]. More detailed studies of defects in HVPE ELO GaN [619,630] revealed two types of defects, called D1 and D2, produced with stripes along the h1 1 2 0i direction. The D1 defect originates from the center of the mask where two lateral overgrowing GaN layers coalesce in the wing region. The D2 defect originates at the two edges at the junction of the mask and the window regions. The D1 defect consists of two groups of dislocations, an array of dislocation segments along the h0 0 0 1i direction and those running vertically along the h0 0 0 1i direction. The D2 defects consist of arrays of dislocations parallel to the h1 1 2 0i direction that form a tilt boundary, where the tilt angle between two neighboring GaN layers was estimated to be 1 . Defects close to the mask (similar to D2) were seen and assumed to originate from shear stress [631]. It has been experimentally demonstrated that GaN grows under a constant tensile stress on sapphire by OMVPE at 1000 C [632]. As the stripes extend vertically and laterally, dislocations and the onset of c-axis tilting may be generated at the expanding boundaries. Finite-element method simulations of the stress distribution shows that at high temperatures, the tensile stress in the GaN seed layer (template) and the thermal stress from the mask result in a high shear stress at the growth facets. The relaxation of this shear stress is assumed to be the driving force for generating dislocations and the c-axis tilting during lateral growth [631]. It should be emphasized that the higher growth rate in HVPE ELO compared to OMVPE is most likely responsible for D2 dislocations and more pronounced c-axis tilting.
3.5 The Art and Technology of Growth of Nitrides
If diethylchlorine gallium (C2H5)2GaCl is used as Ga source in an OMVPE reactor, GaCl is produced at the growth interface, and the applicable chemistry is somewhat similar to that of HVPE. The basic features expected of HVPE are therefore observed [633,634], which may pave the way for high growth rates to be achieved in OMVPE. Indeed, this has been experimentally observed in that the growth rate was found to increase with the growth temperature, with the vertical rate also increasing with decreasing V/III ratio. The commonly observed tilt of the c-axis, endemic to both CVD and MBE techniques, decreases with decreasing lateral-to-vertical growth rate ratio [635]. TEM investigations of a 16 mm thick ELO GaN obtained by this method show that threading dislocations are coherent with the substrate above the windows, which is consistent with ELO by any method, threading dislocations in the coalescence region in the wings, a V-shaped complex of dislocation loops, and longitudinal dislocations that run over the windows after experiencing a 90 bending. Threading dislocations in the ELO GaN bend away from the c-axis (most of them at 90 ) and, as they reach the coalescence plane in the wings, most of them bend again to thread to the surface. On the coalescence plane that is normal to the c-plane, in addition to threading dislocations, a wall of dislocations parallel to the c-plane appears with Burgers vectors in the basal plane, or the c-plane. Rectangular loops on the ð1 1 0 0Þ plane produce a V-shaped arrangement of edge dislocations, which are centered on each coalescence plane. These loops are formed at the coalescence boundaries and propagate into the ELO GaN, details of which can be found in Ref. [634]. 3.5.5.2.2 Lateral Epitaxial Overgrowth on Si The bulk and early work on ELO technology was developed in conjunction with sapphire and 6H-SiC substrates. The dielectric mask is either SiO2 or SiNx. The diffusion and incorporation of O and inhomogeneous stress distribution somehow limits the quality of the ELO GaN. For freestanding GaN, the removal of sapphire is difficult and removal of SiC original substrate has not been yet reported. The abundantly available high-quality Si as a substrate is very attractive for the epitaxial growth of GaN because of cost reasons and the relative ease with which the substrates can be removed, albeit at the expense of having to deal with cracks, which occur during cooldown. ELO GaN [594,606,636,637] as well as GaN SAG [638–641] has been implemented on Si(1 1 1). Because GaN growth directly on Si is unsuccessful, an intermediate AlN or AlGaN layer must first be grown [641]. Growing GaN directly on Si has the additional disadvantage, in addition to efficient desorption of Ga from Si surface, that the thermal mismatch between Si and GaN produces cracks for, depending on the growth temperature, the GaN thickness exceeding approximately 2 mm. More elegantly, incorporation of AlGaN/GaN multilayers releases most of the strain and allows successful ELO GaN to follow [642]. Despite the incorporation of ELO, the quality of ELO GaN on Si(1 1 1) in terms of threading dislocation density is inferior to the quality of ELO/sapphire or ELO/SiC [643]. Borrowing from the GaAs and AlAs extensive research carried out in conjunction with GaAs on Si [644], Kobayashi et al. [645] started the growth process with AlAs and then GaAs on Si(1 1 1). Stripes can then be formed on GaAs using standard photolithography. Application of a selective oxidation converts AlAs to AlOx and GaAs
j539
j 3 Growth and Growth Methods for Nitride Semiconductors
540
to Ga2O3 producing a Ga2O3 template on AlOx. The remaining GaAs is then removed. GaN growth is subsequently initiated only on the Ga2O3 template and results in a lateral growth over the AlOx stripe. A variant of this approach, but one that relies on SiC on Si followed pendeo-epitaxy, which is discussed next in Section 3.5.5.2.3, has also been used. 3.5.5.2.3 Pendeo-Epitaxy A variant of ELO is termed pendeo (from the Latin, to hang or be suspended) epitaxy (PE). This method relies fully on selective growth of GaN on GaN stripes on SiC or Si substrates [646,647]. The method has the advantage of eliminating the SiO2 on which the conventional ELO relies. In this approach, the etched columnar GaN posts are capped with a mask layer, following which GaN growth proceeds laterally and vertically until it coalesces between and over the mask on top of the columns, thereby creating a continuous layer as schematically shown in Figure 3.112. Compared to conventional ELO, the advantages of PE are as follows: (i) growth initiates from a GaN facet different from (0 0 0 1) and (ii) the substrate can be used as mask in the case of 6H-SiC substrate, taking advantage of the selective nature of growth. At a certain temperature window, no growth occurs on SiC, whereas normal growth takes place over and on the edges of the GaN posts (columns). In short, the growth does not initiate through open windows on the (0 0 0 1) surface of the GaN seed layer; instead, it is forced to selectively begin on the sidewalls of a tailored microstructure comprised of forms previously etched into this seed layer. Continuation of the pendeo-epitaxial growth of GaN layer until coalescence over and between these forms results in a layer of lower defect density GaN. This process has also been extended to include Si(1 1 1) substrates after the conversion of its surface with 3C-SiC layer, which eventually is situated between the AlN nucleation layer and the Si(1 1 1) substrate. The seed microstructures are rectangular stripes oriented along the h1 1 0 0i direction and exhibit f1 1 2 0g facets [648]. (b)
(a) SiNx mask Seed GaN AlN 6H-SiC
(c)
PE GaN
(d)
SiNx mask Seed GaN AlN 6H-SiC Figure 3.112 Schematic representation of different steps for growth in the technique of PE: (a) GaN template, (b) etched GaN columns prior to growth, (c) partial growth of PE GaN showing growth only from the sidewalls, and (d) fully coalesced growth of GaN [648] Courtesy of R. Davis.
3.5 The Art and Technology of Growth of Nitrides
The PE method also has some of the maladies of conventional ELO such as voids that are formed above the mask. Additional disadvantages of ELO, which to some extent plague PE, are that coalescence boundaries are formed between two laterally growing fronts owing to misregistry in terms of their bases not having the long-range coherence owing to defects in the template layer. As a consequence, a wavy surface morphology is formed. Tilting of 0.2 is also present in GaN grown over the mask. In spite of the fact the mask on top of the GaN seed posts stops the threading dislocations that would otherwise propagate into the top layer, it is also responsible for the formation of coalescence boundaries and tilt in overgrown GaN. Investigating pendeo-epitaxy of GaN on sapphire, Kim et al. [649] identified two types of coalescence boundaries in the masked and window regions, as well as thin voids on the SiO2 mask, whereas triangular voids were formed in the coalescence region on sapphire. Actually, dislocations are generated at both coalescence regions (over the mask and in between the posts) as well as at both sides of the mask edges, as observed by Sakai [620]. Use of submicron seed posts, which can be defined by electron beam lithography, and mask suppression resulted in the elimination of voids above the seed, as shown in Figure 3.113. This is actually on a trajectory with nano-heteroepitaxy and nano-ELO discussed in Section 3.5.5.3. TEM revealed a strong reduction of the threading dislocation density, with no dislocations generated in the coalescence region and some threading dislocations propagating vertically from the GaN seed post (Figure 3.113). This combined with no tilt observed represents a significant improvement in the standard pendeo-epitaxy method [650,651]. As mentioned, pendeo-epitaxy has been explored on on-axis 6H-SiC(0 0 0 1) and on-axis Si(1 1 1) substrates. In the former, each seed layer consisted of a 1 mm thick GaN film grown on a 100 nm thick AlN buffer layer previously deposited on a 6H-SiC (0 0 0 1) substrate. In the growth on the Si substrates, a 1 mm 3C-SiC(1 1 1) film was initially grown on a very thin 3C-SiC(1 1 1) layer produced by conversion of the Si (1 1 1) surface at 1360 C for 90 s by reaction with C3H8, with H2 as the carrier gas. The film was subsequently achieved by simultaneously decreasing the flow rate of the C3H8/H2 mixture and introducing a SiH4/H2 mixture. Both the conversion step and the SiC film deposition were achieved using a cold-wall atmospheric pressure chemical vapor deposition (APCVD) reactor. A 100 nm thick AlN buffer layer and
Figure 3.113 (a) SEM image of GaN grown over the masked GaN posts and laterally, (b) SEM image of GaN grown on the unmasked submicron post and laterally, (c) TEM image of (b) [651].
j541
j 3 Growth and Growth Methods for Nitride Semiconductors
542
a 1 mm GaN seed layer were subsequently deposited in the manner described above for the 6H-SiC substrates. Following the growth of the base GaN layer either on 6H-SiC- or 3C-SiC-coated Si, a 100 nm Si3N4 mask layer and a Ni layer (as a mask resistant to nitride etches) were deposited on the seed layers through plasma-enhanced CVD and e-beam evaporation, respectively. The Ni layer was patterned by lithography and sputtering and the underlying Si3N4 and nitride layers were patterned by ICP etching in windows. The nickel mask layer was removed in a wet etch, leaving the Si3N4 mask in place on top of the stripes, as shown in Figure 3.114. It is critical that the etching is complete all the way down to SiC. The stripes used were of rectangular shape oriented along the ½1 1 0 0 direction, thus providing a sequence of parallel GaN sidewalls (nominally ð1 1 2 0Þ faces). The windows were 2 and 3 mm wide and window-to-window separations were 3 and 7 mm. Immediately prior to pendeo-epitaxial growth, the patterned samples were dipped in an acid solution to remove surface contaminants from the walls of the underlying GaN seed structures. Three primary stages associated with the pendeo-epitaxy are worth noting: (i) initiation of lateral homoepitaxy from the sidewalls of the GaN seed, (ii) vertical growth, and (iii) lateral growth over the Si3N4 mask covering the raised stripes. Pendeo-epitaxial growth of GaN was achieved within the temperature range of 1050–1100 C using the same pressure and V/III ratio as used for the deposition of the GaN seed layer. 3.5.5.2.4 Pendeo-Epitaxy on SiC Substrates The pendeo-epitaxial phenomenon is made possible by the initiation of growth from a GaN-face other than the (0 0 0 1) and the use of the substrate (in this case SiC) as a pseudomask. By capping the seed-forms with a growth mask, the GaN is limited to grow initially and selectively only on the GaN sidewalls. As in the case of conventional ELO, no growth occurs on the Si3N4 mask. What is unique is that no deposition occurs on the exposed SiC surface areas if higher growth temperatures are employed to enhance lateral growth. In this scenario, the Gaand N-containing species are more likely to either diffuse along the surface or evaporate from both the Si3N4 mask and the silicon carbide substrate, as confirmed by the cross-sectional SEM image of Figure 3.115 where the newly deposited GaN has grown truly suspended (pendeo) from the sidewalls of the GaN seed. Following
Figure 3.114 Schematic representation of pendeo-epitaxial growth from GaN sidewalls and over a silicon nitride mask. Courtesy of R. Davis.
3.5 The Art and Technology of Growth of Nitrides
Figure 3.115 Cross-sectional SEM of a GaN/Al10Ga90N pendeoepitaxial growth structure on SiC showing coalescence over the seed mask. Courtesy of R. Davis.
enhanced lateral growth, vertical growth occurs from the advancing (0 0 0 1) face of the laterally growing GaN. Once the vertical growth becomes extended to a height greater than the silicon nitride mask, the epitaxial growth coalesces as in the case of conventional ELO. A cross-sectional TEM micrograph showing a typical pendeo-epitaxial growth structure is shown in Figure 3.116. Threading dislocations extending into the GaN seed structure, originating from the GaN/AlN and AlN/SiC interfaces, are clearly visible. The Si3N4 mask acts as a barrier to direct vertical growth of GaN and thus propagation of these defects into the laterally overgrown pendeo-epitaxial film. Preliminary analyses of the GaN seed/GaN PE and the AlN/GaN PE interfaces revealed evidence of the lateral propagation of the defects; however, there is yet no evidence that the defects reach the (0 0 0 1) surface where device layers will be grown.
Figure 3.116 Cross-sectional TEM of a GaN pendeo-epitaxial structure showing confinement of threading dislocations under the seed mask and a reduction of defects in the regrowth. Courtesy of R. Davis.
j543
j 3 Growth and Growth Methods for Nitride Semiconductors
544
As in the case of ELO, there is a significant reduction in the defect density in the regrown areas. Compared to conventional ELO, the ratio between the lateral and vertical growth rate in PE on SiC was reported to be four times larger for PE [652]. For the GaN/AlN/ 6H-SiC system, the curvature or bowing of the wafer observed in ELO resulting from residual stress is significantly reduced in PE. With the same basic assumptions as in ELO, Zheleva et al. [653] calculated the stress distribution in PE and found that it ranges from 3.8 GPa compressive stress in the GaN column to 0.6 GPa tensile stress in the window region. Speaking of inhomogeneous strain, the crystallographic tilt of the wings in maskless pendeo-epitaxy was found to result from the strain relaxation of the starting stripe along the c-axis. Finite-element simulation showed that the c-plane stress is partially relieved within the wing regions [654,655]. In a somewhat similar method, the lateral overgrowth was initiated from trenches (termed LOFT for lateral overgrowth from trenches) [656]. In this approach, the trenches are etched into the GaN thin film after which the bottom of the trenches and the top surfaces of the mesas are masked with SiO2. This is then followed by GaN regrowth. As in the case of pendeo-epitaxy, GaN regrown from the sidewalls has a much lower dislocation density because the threading dislocations propagate in the direction perpendicular to the surface. Compared to conventional ELO, this LOFT method allows a significant reduction in the threading dislocation density over the entire wafer using only one step. The threading dislocation density in the top part of the regrown GaN was reduced to 6 · 107 cm2 from the density of 8 · 109 cm2.in coherent layers. The residual threading dislocations either leaked from the sidewalls of the trenches or were generated in the regions where two growth fronts merged. Yet another variant of ELO and PE concepts was demonstrated and dubbed the air-bridged ELO [657–659], which takes advantage of the basic principles of both methods. In this method, the GaN starting epilayer is grooved along the h1 1 0 0i directions using standard photolithography and etching. This is followed by the deposition of silicon nitride on the etched bottom of the trenches and on the sidewalls of the trapezoidal ridge stripes as shown schematically as well as SEM cross-sectional images in Figure 3.117. Regrowth initiates at the stripes and extends laterally forming ELO GaN with f1 1 2 0g facets, which finally coalesce. A gap is formed above the mask, thereby providing nearly freestanding GaN. The tilt angle of the c-axis relative to the seed GaN is only 0.08 . Most likely, the void between the dielectric silicon nitride and ELO GaN prevents stress development as theoretically predicted by Zheleva et al. [653]. However, a high density of threading dislocations still remains above the seed regions. 3.5.5.2.5 Pendeo-Epitaxy on Silicon Substrates The approach of PE on Si suffers from the three-dimensional nucleation and growth of GaN islands caused by the combination of significant mismatches in lattice parameters, the higher surface energy of GaN, and the chemical reactivity of Si with the reactants in the growth environment. To address the above concerns, Davis et al. [660] have developed a process similar to those used for growth of GaN on 6H-SiC (0 0 0 1) but replaced the 6H-SiC substrate with a 3C-SiC(1 1 1) transition layer grown on a Si(1 1 1) substrate.
3.5 The Art and Technology of Growth of Nitrides
Figure 3.117 (a) Schematic representation of the air-bridged ELO, (b) SEM cross-sectional image of the early stages of the regrowth, markers are AlGaN layer inserted for that purpose. The enlarged view of the outlined region is shown just below it, and (c) after full coalescence. Courtesy of A. Ishibashi and Ref. [659].
The atomic arrangement of the (1 1 1) plane of 3C-SiC is equivalent to the (0 0 0 1) plane of 6H-SiC; this facilitates the sequential deposition of a high-temperature 2HAlN(0 0 0 1) buffer layer of sufficient quality for the GaN seed layer. The rest of the growth and processing steps are similar to that of PE of GaN on 6H-SiC that we just discussed. For the initial demonstrations of PE growth of GaN films on silicon, 0.5 and 2 mm thick 3C-SiC(1 1 1) layers were deposited on 50 mm diameter and 250 mm thick converted Si(1 1 1) substrates. All subsequent research described below used the 2.0 mm barrier layer.
j545
j 3 Growth and Growth Methods for Nitride Semiconductors
546
Figure 3.118 Cross-sectional SEM micrograph of a coalesced PE GaN epilayer deposited on a 3C-SiC/Si (111) substrate. Courtesy of R. Davis.
Figure 3.118 shows a cross-sectional SEM micrograph of a PE GaN layer grown laterally and vertically from raised GaN stripes etched in a GaN/AlN/3C-SiC/Si(1 1 1) substrate and over the silicon nitride mask atop each stripe. As in the case of ELO, tilting or sagging of the wing region as they coalesce is common to this method as well [661]. This tilt manifests itself by the additional peaks observed in XRD data. In PE GaN on Si with 3C-SiC interlayers, the XRD spectrum taken along the ½1 1 0 0 direction, parallel to the stripes, consisted of one peak. However, the spectrum along the ½1 1 2 0 direction, perpendicular to the stripes, exhibited two superimposed peaks, separated by a tilt of 0.2 . In contrast to ELO, the PE GaN regions that coalesce contain two sets of coalesced growth fronts; namely, over the trenches and over the masks. A more sensitive method of evaluating the tilt in PE-grown GaN, such as TEM diffraction, is needed to determine if tilting is present in both of these areas. Analyses of such a study from small areas of coalesced regions in a trench and near the coalesced region over the silicon nitride mask, revealed no evidence of tilt in the laterally grown material over the trenches. However, significant tilt was observed over the masked regions of the GaN seed. Lateral epitaxial overgrowth of any kind, including the pendeo-epitaxy variety, require that the samples be taken out of the growth vessel, processed and returned to the reactor, which is cumbersome and costly. Even then, the method does not necessarily address the cracking issue when GaN is under tensile strain caused by residual thermal strain or by the choice of heteroepitaxial under layers. The former occurs when GaN is grown on SiC and Si, whose thermal expansion parameters are smaller than that of GaN. The latter occurs when GaN is grown on thick and relaxed AlGaN layers such as the cladding layers used in lasers. Moon et al. [412] reported on a method where both defect reduction and strain relaxation issues are simultaneously addressed. The method involves allowing the GaN layer to cooldown and crack followed by returning to the growth temperature, deposition of in situ SiNx and in situ
3.5 The Art and Technology of Growth of Nitrides
etching in H2 and further growth of GaN. In the process, the growth is initiated at crack sites and followed by lateral epitaxy. Because this is similar to pendeo-epitaxy in some ways and takes advantage of cracks, the method is dubbed the crack-assisted pendeo-epitaxy (CAPE). The evolution of the growth examined after a brief deposition of GaN on cracks indicated layer growth commencing at the cracks, which act as onedimensional wires, and growth front advancing laterally to lead to coalescence. TEM characterization in the cross-sectional and plan view configurations confirm threading dislocation reduction with plan view images indicating defect concentration as low as low 107 cm2 in some regions and mid 108 cm2 in others. Because the 107 cm2 figure is very low for GaN, which without growth on patterned templates is somewhat uniformly defective in the range of 109 cm2 or higher, lateral epitaxial growth is assumed to play an important role. The results are preliminary and the method is potentially very attractive. From the fitting the fast and slowing decaying biexponential PL decay curve, a radiative recombination time of slightly over 1 ns has been obtained. Two-Step Epitaxial Lateral Overgrowth An obvious extension of the single-step ELO is the two-step ELO. This of course requires, an additional lithographical step following a second growth interruption/exposure to environmental conditions of the GaN template. Because the lateral overgrowth depends on parameters such as temperature, carrier gas, V/III ratio, as in the case of single ELO, several two-step processes have been conceived. Two-step ELO is initiated with a low V/III ratio for producing smooth vertical f1 1 2 0g sidewalls and continued with an increased V/III ratio for increasing the lateral growth [597]. However, even though such a process produces a smooth morphology, threading dislocations are still present in the coherent growth region. To avoid this drawback, other technologies such as double-layer ELO have been proposed. In this alternative ELO, the mask is rotated by 60 (or 90 ) after the first ELO layer, allowing the growth of stripes along another equivalent h1 1 0 0i direction. This results in the reduction of dislocations over the windows. Using basically the same concepts to eliminate the remaining threading dislocations in the coherent region, a technology in which the second mask has a lateral offset from the first array of openings has been proposed by Davis [662], Sugiura [663], and Ikeda [664]. Specifically, the first ELO GaN utilizes stripe openings along the h1 1 0 0i direction, after which the second mask covers all the coherent GaN from the first growth step, where the threading dislocation density remains high, thus eliminating the propagation of threading dislocations into the second ELO layer. If all worked ideally, the entire final layer would have a low dislocation density. Another original two-step ELO process was recently developed [665] that allows a larger defect reduction, especially above the window stripes [665–668]. This process takes advantage of the phenomena of dislocation bending that has been observed by several authors [130,616] and already discussed. In the first step of this process, a vertical expansion is favored, whereas a lateral expansion is promoted in the second step. Hiramatsu et al. [669] conceived a process dubbed the FACELO for facetcontrolled epitaxial lateral overgrowth, which in principle is the same as the 2S-ELO, and a P, T, V/III diagram as shown in Figure 3.108 to optimize the shape of the lateral
j547
j 3 Growth and Growth Methods for Nitride Semiconductors
548
overgrown GaN for optimum bending of dislocations. In this approach, no tilt of the c-axis or the ELO layer was reported. Gibart et al. [575] described the dislocation bending and the growth conditions required to do so for each step for the two-step ELO processes in conjunction with (CH3Cp)2Mg augmented ELO, which allows reduction in growth temperature [604] as well. As mentioned in conjunction with the one-step ELO, the first few steps in the two-step ELO process are identical to those in the one-step process as described in Section 3.5.5.2 dealing with discussion surrounding Figure 3.111 in that the growth of GaN is conducted at 1040 C, with the vertical growth rate being initially larger than that along the sidewalls, and so on [575]. The second set of steps is different in the twostep ELO as described with the aid of Figure 3.119, which is a schematic representation of the two stages of the two-step (2S)-ELO process. At the end of the first growth, where the vertical growth rate is higher than the lateral one, the second growth stage is initiated where with the aid of growth at high temperatures or introduction of (CH3Cp)2Mg, the lateral growth rate is made larger than the vertical one, or Gs > Gc, which paves the way for flattening. In the case of Gs/Gc 2, coalescence boundaries are formed together with voids. The red lines represent the behavior of threading dislocations up to the coalescence boundaries. Figure 3.120 is a typical image of a 2S-ELO sample but at the end of the first step, viewed in cross section along the ½1 0 1 0 zone axis. The second step that is centered around the lateral growth was induced by the introduction of (CH3Cp)2Mg in the vapor phase [604]. The resulting microstructure is similar to the case when the lateral growth is induced by the much higher growth temperature 1120 C. Full coalescence and smoothing of the surface required 1.5 h, resulting in a thickness of 12 mm. Once full coalescence is achieved, the behavior of the dislocations is different than in
Converging (1) followed by diverging trajectories of (0 0 0 1) facet edges
(2)
(1)
Mask GaN template Figure 3.119 Schematic representation of the two stages of the two-step (2S-)-ELO process during the second ELO step performed either at relatively higher temperature or at typical growth temperatures but with introduction of (MeCp)2Mg. The evolution of growth is such that
for GS > GC flattening occurs and for GS/GC 2, coalescence boundaries are formed together with voids. The gray lines represent the threading dislocations up to the coalescence boundaries. Courtesy of P. Gibart and Ref. [575].
3.5 The Art and Technology of Growth of Nitrides
Figure 3.120 (a) Cross section along the ½1 0 1 0 zone axis of a 2SELO film at the end of the first step. Dashed lines join the dislocation bending points. (b) Schematic representation of 2SELO. Dotted lines represent the shape of the ELO material at different stages of the first step of the 2S-ELO process. Broken black lines join the successive edges of the top C facet. Solid black lines represent dislocations. Courtesy of P. Gibart and Ref. [575].
standard one-step ELO. Specifically, the dislocations under the masks are also blocked when they encountered the mask. A close examination of the dislocation propagation behavior shows that the dislocations above the window first propagate vertically, as in the standard ELO. However, they afterwards bend by 90 to assume a direction in the (0 0 0 1) basal plane. It should be mentioned that this behavior is observed regardless of whether the type of the dislocations are, edge, screw, or mixed (a, c, or a þ c. refer to Chapter 4 for description of each of the three dislocations). Some of the dislocations adopt an intermediate inclined direction before bending again to assume a direction lying in the basal plane. Broken lines in Figure 3.120 indicate the points at which the dislocations bend. The dislocations close to the edge of the mask bend first. A
j549
j 3 Growth and Growth Methods for Nitride Semiconductors
550
vertical growth rate higher than the lateral growth rate leads to intermediate overgrowth whose material shapes are represented by dotted lines in Figure 3.120. As soon as a vertical dislocation line gets sufficiently close to the edge between the top and inclined facets, it bends and adopts a new direction in the basal plane. This general picture can be best understood in the frame of image dislocation and image forces (in much the same way as charge and image charge), as already proposed [616] and treated in books [670]. A dislocation generates a stress field, which should be compensated when the dislocation line gets sufficiently close to surfaces for it to leave the free surfaces stress free. An image dislocation is a virtual dislocation located outside the material, which generates a stress field compensating at the free surface the stress field of the actual inner dislocation. These image dislocations induce attractive forces acting on inner dislocations. In the experiments of Gibart et al. [575], owing to symmetry considerations, the image force has no effect on the dislocations when TDs are perpendicular to the (0 0 0 1) basal surface without lateral facets within the interaction range. However, the dislocation lines may lie in the basal plane when the threading dislocations are close enough to an inclined facet. In fact, during the first step of 2S-ELO, the lateral inclined facet progressively moves toward the dislocation lines, reaching first to the dislocation lines that are closest to the edge of the mask followed by reaching to the dislocations situated just in the center of the window. The dislocations situated in the middle of the window may not bend and propagate vertically to the surface because interactions with both the free surfaces on the left and right, ð1 2 1 2Þ and ð1 2 1 2Þ, may compensate each other. The 90 bending of dislocations could be the result of the general principle of minimization of the free enthalpy of the system. As such during growth, dislocations follow a direction leading to minimum enthalpy. As the line energy of a dislocation depends also on its character, the energy of a screw being the lowest, bending at 90 of an edge eventually produces a screw dislocation, or introduces a screw component, thus lowering the enthalpy of the system. Bending of dislocations results in a discernible reduction of their density in the upper part of the film provided that the thickness is greater than the height of the pyramids formed at the end of the first growth step. After bending, most of the dislocations have a line parallel to ½1 2 1 0, which extends to the coalescence boundary with the overgrown GaN advancing from the adjacent stripe. Defects therefore accumulate at the boundaries. Dislocations at the interfaces manifest themselves in several ways as has been reviewed by Vennegues et al. [666]. Voids are clearly visible over the mask, as discussed in the next paragraph, and have a basal triangular shape with an extension over their top apex. Most of the lateral dislocations are blocked when they encounter such a void and therefore do not extend further. Other possible manifestations of these lateral dislocations when they reach the coalescence boundary have been shown to bend down to the void resulting in their termination therein, or bend up in the boundary and thread up to the surface. Single ELO GaN layers with a thickness of 2 mm were obtained using 3 mm wide stripe openings with 10 mm pitches and oriented along the h1 1 0 0i direction (Figure 3.121a). The growth parameters were temperature of 1100 C and a TEG flow rate of 26 mmol min 1. Atomic force microscopy showed that the surfaces of the
3.5 The Art and Technology of Growth of Nitrides
Figure 3.121 Cross-sectional and surface SEM micrographs of the first, (a) and (b), and second, (c) and (d), coalesced GaN layers, respectively, grown on 3 mm wide and 7 mm spaced stripe openings oriented along < 1 1 0 0>. Courtesy of R. Davis.
ELO GaN had a terraced structure with an average step height of 0.32 nm. As in the single ELO, one must deal with the voids formed in two-step ELO at the coalescence interface. If the voids were to form in the dislocated regions and cause the dislocations to move laterally owing to local strain modification, they would be beneficial. However, at the coalesced region, their presence point to problems with proper indexing of the lattice. As an example of these undesirable voids, each black spot in the overgrown double ELO GaN layers shown in Figure 3.121a and c is a subsurface void that forms when two growth fronts coalesce. These voids were most often observed using the lateral growth conditions wherein rectangular stripes having vertical f1 1 2 0g side facets developed. The morphologies of the finished surfaces of single and double ELO layers imaged by SEM are shown in Figure 3.121b and d. Surface morphology of the second overgrown layer was comparable to the first layer. Cracks were occasionally observed along the coalesced interface under selected growth conditions, probably owing to the thermal mismatch between the GaN layers and the SiO2 mask. The cross-sectional TEM image of Figure 3.122 shows a typical laterally overgrown GaN. Threading dislocations, originating from the GaN/AlN buffer layer interface, propagate to the top surface of the regrown GaN layer within the window regions of the mask. The dislocation density within these regions, calculated from the plan view TEM micrograph is approximately 109 cm2. By contrast, however, additional microstructural studies of the ELO regions showed much fewer dislocations. Cross-sectional TEM observation of the double ELO sample in the micrograph presented in Figure 3.123 shows that a very low density of dislocations parallel to the (0 0 0 1) plane, formed via bending of threading dislocation, exists in the first and second ELO GaN layers on the SiO2 masks. The second SiO2 mask is slightly misaligned relative to the first. These results suggest that very low defect density GaN layers can be fabricated by precise alignment of the mask in the second lithographic process.
j551
j 3 Growth and Growth Methods for Nitride Semiconductors
552
Figure 3.122 Cross-sectional TEM micrograph of a section of a laterally overgrown epitaxial GaN layer on a SiO2 mask region. Courtesy of R. Davis.
The dislocation densities in ELO GaN are low enough so that questionable figures could be arrived at by TEM, which, in this context, led to report of figures [599] in the range of 104–10 5 cm2, which are very optimistic. The etching methods reported in Section 3.5.1.2 dealing with HVPE samples, and also in Section 4.2.4 are much more reliable, although they too could be incorrect unless extreme care is taken and many calibration samples are utilized. Ultimately, when sufficient progress is made, electrical and optical properties and to some extent electrically active point defect density would take the place of extended defect density. In this particular report, the work was done with stripes oriented in the ½1 1 0 0 direction to yield a large lateral growth rate/vertical growth rate ratio. The stripe spacing was varied to give ratios of open width and patterned period of 0.1–0.5. The ELO GaN was bound by the (0 0 0 1) facet on top and by vertical ½1 1 2 0 sidewalls on the edges, which showed a lateral growth rate of up to 6 mm h 1. Patterns with 10 mm stripes and a ratio of open width and patterned period of 0.5 enabled full coalescence of the overgrown GaN film after 90 min of growth. TEM and AFM (counting the pit density where the bilayer steps are terminated/broken) observations indicated that the density of mixed character dislocations reaching the surface of the ELO GaN and leaving a dark spot in the AFM image is in the 104–105 cm2 range. To draw the contrast between the overgrown region and the region directly over the template GaN, a cross-sectional TEM image and a plan view TEM image around the vicinity of a stripe are shown in Figures 3.122 and 3.124, respectively. Clearly, the ELO process is effective in reducing the extended defect density. As in the case of one-step ELO, CL has been used copiously to image the radiative recombination on and near the surface. Dislocations with a screw component are expected to cause deep states in the gap and act as nonradiative recombination centers in GaN and therefore lead to contrast in CL images. Assuming a one-to-one
3.5 The Art and Technology of Growth of Nitrides
Figure 3.123 Cross-sectional TEM micrograph of a section of the second lateral grown GaN layers. Courtesy of R. Davis.
correspondence between the CL dark spots and the threading dislocations, the difference in CL images taken at 90 K of one-step and two-step ELOs is striking [575], as illustrated in Figure 3.125. Figure 3.125a shows the top view monochromatic CL mapping at l ¼ 358 nm, which corresponds to the free exciton A of a 2S-ELO GaN sample. Figure 3.125b shows a 1S-ELO GaN sample. To make the comparison more meaningful, the conditions used to grow the top GaN, which is what is probed with CL, have been kept identical. For both samples, full coalescence was achieved and no topography contrast was observed in secondary electron images. Therefore, stark differences noted in the CL images of Figure 3.125 are attributable only to the nature of two-step and one-step ELO. In the images, the dislocations pointing up appear as
j553
j 3 Growth and Growth Methods for Nitride Semiconductors
554
Figure 3.124 Plan view TEM micrograph of a section in the vicinity of the lateral grown GaN and that grown on the GaN template below. Courtesy of R. Davis.
black points, and it can be seen that they are organized along the lines parallel to the stripes, consistent with dislocation bending and pooling in the coalescence plane arguments. Lines with a high density of dislocations (white arrows) alternate with lines with a lower density of dislocations (black arrows) and are separated by 5 mm, which corresponds to half the stripe period. The highly defective and lightly defective lines correspond to the coalescence boundaries and the center of the windows, respectively. Filtering of extended defects through dislocation bending is therefore not completely effective and dislocations located just at the center of the window, which meet the apex of the pyramid at the end of the first stage of the 2S-ELO, may not bend and may propagate up to the surface. Getting back to the CL image, a few dislocations are present between the lines as well. The density of defects average over the entire surface as determined by CL imaging is 1.7 · 107 cm2. If only the areas between the coalescence boundaries, spaced 10 mm apart, are considered, the density is 107 cm2. The two-step process reduces the threading dislocation density by at least one order of magnitude compared with the standard non-ELO-grown GaN layers.
Figure 3.125 CL map of GaN layers grown with the two-step (a), and one-step ELO processes (b). The scale bar is 20 mm. Courtesy of P. Gibart and Ref. [575].
3.5 The Art and Technology of Growth of Nitrides
Figure 3.126 UV CL monochromatic image recorded at 90 K on a bevel edge of a 2S-ELO GaN. The luminescent triangles correspond to laterally overgrown material whereas the dark triangle corresponds to the coherent part. Coalescence boundaries and TDs are running parallel to the c-plane after bending. In the top surface, the dark spot corresponds to merging dislocations. Courtesy of P. Gibart and Ref. [575].
To get an insight into the evolution, as layer growth progresses, of dislocations, cross-sectional CL experiments in a two-step ELO sample have been undertaken [671]. Images displayed in Figure 3.126 are for coalescence using (CH3Cp)2Mg, but similar features are obtained when coalescence is performed by the use of increased substrate temperature. In addition to imaging, CL spectra in cross-sectional mode were also obtained for an in-depth analysis [671]. The spectra, which are not shown, indicate that the growth in the windows is defective as judged from a lack of near band edge emission and strong yellow line (YL) emission. At the lateral edges of the stripes (GaN grown on the striped windows) exhibited broad near band edge emission. In the region where the lateral epitaxy is enhanced by addition of (CH3Cp)2Mg, donor– acceptor pair band emission was observed. It should be pointed out that Si and O contamination coupled with Mg contamination from the mask and vapor lead to impurity incorporation, which could adversely affect the devices such as FETs. The unintentionally doped top layer exhibited sharp excitonic transitions with free and bound excitons visible with a line width of 0.7 eV, which degrades to 3–4 meV when ELO is not employed [575]. Even though the research effort on 2S-ELO in HVPE has been less intensive as compared to OMVPE, the 2S-ELO technology is achievable in HVPE, which is motivated by the potential of HVPE providing templates for further growth by other methods such as OMVPE and MBE. The basic mechanisms in effect for HVPE are, however, not completely identical to those for OMVPE and the experimental conditions for achieving the growth anisotropy are significantly different. There are also similarities in that the morphology is controlled by the composition of the carrier gas. Starting with windows aligned in the h1 1 0 0i direction and under N2 at 1050 C, the
j555
j 3 Growth and Growth Methods for Nitride Semiconductors
556
equilibrium shape of the overgrowth leads to trapezoidal stripes with (0 0 0 1) top faces and f1 1 2 2g lateral facets. When the carrier gas is switched to H2, the lateral growth rate is significantly reduced and triangular stripes with f1 1 2 2g lateral facets are obtained. This means that a mixture N2 : H2 carrier gas would lead to pyramidal nucleation islands during the initial states of growth. In a second step, pure N2 carrier gas, which enhances lateral growth, can be employed. With two-step ELO, complete coalescence without pits and planarization have been obtained, which are in Figure 3.127 as cross-sectional SEM images GaN layers grown with one-step and two-step ELOs. The pits present along the coalescence boundaries in one-step ELO are absent in the two-step ELO case. In addition, the two-step ELO process produces planar surfaces [672,673]. Figure 3.127c shows cross-sectional SEM image of a 60 mm 2S-ELO HVPE layer obtained by carrying out the first growth step with a mixture of H2 þ N2 as carrier gas (using 5 mm windows and 10 mm mask stripes): growth proceeds until the growth
Figure 3.127 Comparison of coalescence on onestep (a) and two-step (b) ELO performed by HVPE. Note that the one-step ELO leads to a high number of pits along the coalescence boundary while those pits are absent in the two-step ELO with the additional benefit of good planarization
obtained, (c) Cross-sectional SEM images of a 60 mm thick HVPE-grown GaN layer showing the initial coalescence followed by an enhanced lateral growth by switching to pure N2 carrier gas leading to planar growth (openings, 5 mm, period 15 mm) Courtesy of V. Wagner and M. Ilegems (Ref. [673]).
3.5 The Art and Technology of Growth of Nitrides
front from two adjacent f1 1 2 2g facets meet. In the second growth step, pure nitrogen carrier gas is used as to achieve full coalescence and a flat surface, see Figure 3.127. The nature of threading dislocations in HVPE two-step ELO is somewhat similar to that observed in 2S-ELO in OMVPE in that they first propagate from the template. They then bend 90 and propagate horizontally only to be terminated at the coalescence boundaries. In some cases, the 90 bending occurs in two steps as determined by TEM investigations. Once again the wavelength resolved crosssectional CL allows an in-depth evaluation of the evolution of the crystallographic quality and the density of incorporated impurities. The seed region and the part grown in the windows on the template show about the same optical and crystallographic quality as characterized by relatively narrow near bandgap emission. However, the laterally overgrown region exhibits a broad blue-shifted emission but with a high-integrated intensity, the origin of which may be associated with impurities. Three-Step Growth with ELO Template It has been demonstrated that several steps led to ELO GaN with wider usable surfaces, free of emerging TDs. Even better quality can be foreseen with more than two steps at the expense of higher complexity. Nagahama et al. [674] implemented a three-step ELO process in an effort to further reduce the threading dislocation density. The effort led to densities slightly above the mid 105 cm2 point, and thereby allowed the fabrication of LDs lasting 15 000 h with an output power of 30 mW. In this 3S-ELO, a standard OMVPE ELO GaN is first grown as discussed. On top of this ELO GaN, a 200 mm thick HVPE layer is grown. The sapphire substrate is then removed by mechanical polishing to get a 150 mm freestanding GaN layer. Finally, in the last step, another ELO GaN is grown on top of this self-supported GaN, leading to a threading dislocation density of 7 · 105 cm2 in the top layer on which the laser structure is built. 1S-ELO on 2S-ELO with the mask in the second step located exactly above the first mask (to eliminate coalescence ELO boundaries) further reduces the threading dislocation density [675]. The inverted version, the 2S-ELO on 1S-ELO produces similar results [676]. An AFM image of the top surface of a 3S-ELO GaN, where the threading dislocation density over the entire surface is 1 · 107 cm2, showed very few threading dislocations between the coalescence boundaries. In another approach, starting from grooved stripe substrate as in Ref. [681–683]. Ishida et al. [677] employed two regrowths on periodically grooved structures and in the process reduced the density of threading dislocations down to 6.3 · 106 cm2. Once again, the mechanism of dislocation reduction is bending. These improvements, however, come at the expense of complicated processes, which increase the cost of templates. Maskless Epitaxial Lateral Overgrowth As discussed, two epitaxial growth steps, one masking step and one photolithography step, are required in one-step ELO. A simpler ELO process that does not significantly degrade the final quality of the ELO GaN is of considerable interest. In a method for which the regrowth step is not necessary, the ELO process is initiated directly on the SiC or sapphire substrates patterned with SiNx mask [678]. The AlGaN nucleation layer is deposited, which wets the exposed substrate
j557
j 3 Growth and Growth Methods for Nitride Semiconductors
558
but not the silicon nitride mask, followed by GaN, which proceeds both vertically and laterally. AFM studies indicate that the threading dislocation densities are in the mid 108 cm2 in the coherent region and 106 cm2 in the ELO region. Further effort along these lines led to the use of textured sapphire substrates [679]. In this particular process, mesas are formed on the substrate by etching prior to just one growth sequence. Growth takes place laterally to produce GaN cantilevers that finally coalesce. Cantilever epitaxy (CE) results in a significant reduction of threading dislocations, with densities below 107 cm2 above the support mesa. Threading dislocations in the edge region bend and further extend laterally into the cantilever. Compared to the conventional ELO, the dislocation density at the coalesced front and at the mesa is lower. To date, this method presents the limitation of the 1S-ELO. Despite a significant decrease in dislocations reached by CE, the wing tilt of approximately 0.8 was observed for CE of GaN on SiC and Si(1 1 1) [680]. With further improvements, this process may produce low dislocation GaN at a lower cost than the conventional ELO. An alternative method, which relies on periodically grooved substrates (sapphire or SiC) with stripes parallel to the h1 1 0 0i GaN direction, was proposed by Sano et al. [681–683]. On these grooved substrates, a fast vertical growth rate produces triangular stripes with f1 1 2 2g facets as in the ELO process that finally coalesce. Low dislocation density GaN was produced with this method, with nearly similar crystallographic quality to that produced by conventional ELO. The coalesced boundaries also produce high densities of threading dislocations. In a slightly modified version, Strittmatter [684] utilized textured Si(1 1 1) substrates with grooves parallel to Sih1 1 0i. In such a case, GaN grows on the bottom of the grooves and then the V/III ratio is increased to enhance the lateral growth rate and obtain full coalescence. As usual, the wing areas exhibit improved optical and crystallographic quality. An ELO process in which nucleation is performed on a sapphire substrate patterned with a SiO2 mask was proposed to further simplify the process, thereby restricting nucleation and further lateral growth within the windows. Direct lateral overgrowth occurs on these SiO2 patterned sapphire substrates. In spite of achieving ELO quality, the random nucleation does not allow smooth coalesced films [685]. 3.5.5.2.6 Point Defect Distribution in ELO Grown GaN In standard ELO, two regions are clearly identified: the overgrown region over the mask and the coherent growth in the windows. The structural data provided clearly show that ELO reduces the threading dislocation density as well as tilting of the c-axis and twisting of the cplane, with improvement being more dramatic in the wing regions as quantified by Xray measurements conducted in ELO GaN grown by both OMVPE and HVPE [686]. In OMVPE, the tilting and twisting strongly depend on the stripe orientation of the pattern (h1 1 2 0i or h1 1 0 0i), which is closely related to the difference in the growth process. However, the extended defect density is much lower in the wing regions than that over the seed template in the windows. Therefore, an assessment of point defects and impurities necessitate position-sensitive tools such as micro-Raman, micro-PL (m-PL) both CW and time-resolved [628], and cathodoluminescence (CL). The CL method is extensively used to map the luminescence efficiency in the layer and thus gain some insight into the spatial distribution of radiative recombination
3.5 The Art and Technology of Growth of Nitrides
efficiency. Higher efficiency is associated with less nonradiative defects and the converse is true for more nonradiative defects. For correlation to structural defects, it is also assumed that structural defects cause nonradiative recombination centers, the exact nature for which is complex and is discussed in Section 4.3. This can be done with emission at a given wavelength (monochromatic) or all wavelengths (panchromatic). In addition, excitation electron energy can be changed to affect the absorption depth, and so on. However, waveguiding of the emitted light as well as secondary excitations both in terms of electron-initiated excitation and photon-initiated excitation are among the processes that cause some uncertainty in the collected data. In one such investigation, the monochromatic CL image at 365 nm reveals that in the coherent region above the windows, the CL shows mottled luminescence contrast as usually observed in uniformly grown OMVPE GaN. On the contrary, the ELO material extending on both sides of the seed region exhibits fewer nonradiative regions, but the coalesced boundaries are characterized by dark CL lines [687]. The vertical and horizontal propagation of threading dislocations were investigated by Rosner et al. [688], who intercalated a GaInN single quantum well (SQW) grown with the ELO method and obtained a 421 nm CL mapping including depth analysis, showing that the lateral defects do not propagate into the upper part of the film. Stripes Along the h1 1 0 0i Direction in HVPE ELO For ELO GaN on AlN/6H-SiC with stripes along the h1 1 0 0i direction, CL imaging of uncoalesced stripes shed light on the threading dislocation propagation and impurity incorporation [689]. When the overgrown GaN has the form of trapezoidal cross section (such as that shown in Figure 3.108 region II), the fast vertical growth rate results in the incorporation of a high density of defects and produces a strong YL emission in the coherent region, whereas in the triangular part, a blue emission is clearly visible. In the region with rectangular cross section (such as that shown in Figure 3.108 region III), most of the YL emission originates from the region between the stripes. PL measurements show an additional band at 3.4643 eV, which are most likely linked to a donor incorporated in the ELO process, and a blue-shifted emission in relation to the underlying GaN. These observations show that the ELO GaN exhibits improved material quality and reduced biaxial strain, as judged by point defect activity as well. On GaN pyramidal structures partially coalesced along ½1 1 0 0 direction, spatially resolved CL demonstrates that the coalescence region exhibits stronger and more uniform luminescence than the pyramidal sidewalls [690]. Moreover, these results are consistent with Raman scattering data where a reduced linewidth and a slight shift of the E2 phonon in the ELO region are observed [689]. It should be mentioned that stress affects Raman frequencies in addition to defect generation and propagation, such as horizontal dislocations [691], which makes stress analysis an integral part of ELO. ELO GaN results from a growth process involving different materials (dielectric mask, substrate) with different lattice parameters and thermal coefficients and involves several thermal cycles. Therefore, such a system generates inhomogeneous stress. The magnitude and spatial distribution of stress have been modeled using finite-element analysis for the system GaN/AlN-6HSiC with SiO2 mask [692]. It was shown that the edge of ELO GaN in the vicinity of the
j559
j 3 Growth and Growth Methods for Nitride Semiconductors
560
Figure 3.128 SEM and CL wavelength image of a cross section of HVPE ELO sample. Courtesy of J. Christen and A.G. Hoffmann, Ref. [695]. (Please find a color version of this figure on the color tables.)
GaN/dielectric interface could be under localized compressive stress fields up to 3 GPa. The stress inhomogeneities are endemic to ELO no matter which growth method is employed. The stress analysis for HVPE ELO of GaN was carried out using simulations based on continuum elastic theory. The lattice mismatch a was adjusted to reproduce the biaxial stress in the GaN template layer and was taken to be zero for the amorphous SiO2 mask. The calculated biaxial stress, sxx, distribution agrees very well with the experimental values deduced from the m-Raman data. The ELO GaN is under biaxially compressing stress; sxx relaxes rapidly through the window to 0.32 GPa and to 0.45 GPa when stripes are along the h1 1 2 0i and h1 1 0 0i directions, respectively. Along the mask, the ELO GaN is also under biaxial stress [693]. Other efforts of stress analysis in the pyramids led to the deduction of compressive stress in the GaN buffer template layer, tensile stress in facets, and full relaxation in the mid to upper parts of the pyramids [694]. Kaschner et al. [695] investigated ELO with stripes along the h1 1 0 0i direction with cross-sectional CL and m-Raman. Cross-sectional CL mapping and SEM images are shown in Figure 3.128. The coherent region grown in the windows exhibits a monochromatically uniform rectangular region dominated by CL emission at 3.467 eV. The CL spectra of this coherent part exhibits sharp D0X recombination at 3.467 eV. As can be deduced from Figure 3.128, the overgrown region is red shifted, which is similar to the case with the stripes along h1 1 2 0i, which will be discussed below. It is postulated that this red shift results from incorporation of impurities in the coalesced region. The TRPL measurements support the basic conclusions deduced from CL and m-Raman. Decay time constants in the coherent region of samples over the windows reach about 220 ps, indicating high-quality material, which is also supported by m-PL measurements, the spectra of which are dominated by two peaks D0X
3.5 The Art and Technology of Growth of Nitrides
Figure 3.129 SEM (a) and CL wavelength (b) images of two different regions, coherently grown above the openings and overgrown above the SiO2 stripes. The growth in the windows (between the SiO2 stripes) and the wing (coalesced regions over the SiO2 stripes) regions
indicated in the schematic drawing (c) are clearly visible in the CL wavelength image. Courtesy of J. Christen and A.G. Hoffmann, Ref. [697]. (Please find a color version of this figure on the color tables.)
and XA. Both peaks show a weak intensity on the coherent region and a high intensity in the wings. From the analysis of the cross-sectional m-PL, it is concluded that the incorporation of donors is enhanced in the wing regions before full coalescence, but once GaN coalesces over the mask, the incorporation of donors is suppressed [696]. Stripes Along the h1 1 2 0i Direction in HVPE ELO In addition to planar CL mapping, cross-sectional mapping of ELO with stripes along the h1 1 2 0i direction has been conducted [697] for a detailed investigation and evolution of nonradiative defects. As can be seen in Figure 3.129, the coherently grown GaN over the windows in the mask exhibits a homogeneous emission at 3.463 eV. On the contrary, the ELO region, the wing region, exhibits an inhomogeneous and blue-shifted emission around 3.483 eV. Conversely, a red-shifted (3.425 eV) CL peak dominates at the coalescence boundaries. A CLwavelength image and local CL spectra from the ELO regions and the coherently grown GaN were also taken by Betram et al. [697]. The spectra in the coherent growth region show sharp excitonic lines associated with the free exciton X, D0X (two donors) and A0X. With increasing distance from the substrate into the layer, an 8 meV blue shift was observed for all lines. In the ELO region, broad and blue-shifted CL emission was observed. To complement the CL measurements, m-Raman scattering experiments were carried out on the same spots [697]. These experiments showed that E2 mode is directly related to the local strain. The free carrier concentration deduced from the LO phonon–plasmon coupled modes (LPP) indicated that the free carrier density in the overgrown wing region increased to about 1019 cm3 above the buffer and remains constant in the growth direction, as shown in Figure 3.130a. In the coherent region, it starts at a low level (<1018 cm3) above the apex and a jump up to 1.3 · 1019 cm3 is observed. It is worth noticing that these data is in full agreement with CL mapping.
j561
j 3 Growth and Growth Methods for Nitride Semiconductors
562
Figure 3.130 A m-Raman scans along c-axis of overgrowth GaN (in blue) and coherently grown GaN (in red): (a) free carrier density and (b) biaxial compressive stress. Courtesy of J. Christen and A. G. Hoffmann, Ref. [697]. (Please find a color version of this figure on the color tables.)
3.5 The Art and Technology of Growth of Nitrides
These results show that the distribution of residual donors is inhomogeneous and that strong impurity incorporation occurs in the ELO region. The local distribution of biaxial stress is deduced from the E2 Raman mode as shown in Figure 3.130b. The short point is that m-Raman reveals inhomogeneities. Other m-Raman spectroscopy measurements of the A1 (LO) phonon line indicated a residual doping level of 1 · 1017 cm3, the source of which was assumed to be the SiNx mask [698]. The confocal m-Raman spectroscopy allows mapping of the free carrier concentration. Specifically, the free carrier concentration n can be obtained from the scan of the LO phonon–plasma (LPP) coupled mode. In the triangular stripes, n was found to reach 1020 cm3, whereas n 1019 cm3 in the coalescence region, where the two laterally growing GaN meet, was located above the triangular stripes [699]. These features were attributed to enhanced incorporation of O and Si, the source of which is most likely the SiO2 mask, in the coherent part over the windows. Additionally, Kelvin microscopy investigations were used to suggest that compensating acceptors, which could be VGa, are also incorporated in the triangular stripes [700]. Local probe analysis in ELO GaN for stripes oriented in the h1 1 2 0i direction can be summarized as follows: .
In the coherent grown regions for growth times long enough for (0 0 0 1) facet to disappear, most of the threading dislocations undergo a 90 bending and the top part of the coherent triangle is virtually dislocation free. The CL spectra become extremely narrow, figures for which are not included in this text but can be found in Ref. [697].
.
In the ELO regions in the wings, lower quality is evidenced, despite lower extended defect concentration, most likely because of enhanced impurity incorporation (Figure 3.130a, for details refer to Ref. [697]).
.
In the coalescence region, the CL spectra are wide as well as red shifted due to the generation of defects at the coalescence boundary (spectrum c in Ref. [697]).
The effect of the ELO process on point defects, specifically the nature of how they express themselves in the YL band is discussed in Volume 2, Chapter 5. Because of the relatively known distribution of dislocations, the ELO GaN provides a very good laboratory to find, if any, correlation between the dislocations and the YL luminescence, which captured a good deal of interest. This problem can be studied with spatially resolved PL, and/or CL, as has been done by Gibart et al. [701] using the hexagonal openings and partially coalesced pyramids. The band edge PL spectrum is strong at the edge of the overgrowth and the lowest from the center. Conversely, the YL intensity is by one order of magnitude the strongest at the center of the overgrowth region where the dislocation density is lower, but impurity contamination owing to O (the source of which could be sapphire, as SiNx masks also lead to O in SIMS, which has been reported by Popovici et al. [702], and the mask) and Si may be high. SIMS measurement in 2S-ELO samples indicated [575] that even with SiNx mask, O incorporation during the first growth step as well as in the template is on the order of several 1018 cm3, the source of which is sapphire, which decreases significantly below 1018 cm3 during the second step. Inside the triangular stripes, the incorporation of O is slightly less than on the laterally overgrown part, which can be explained
j563
j 3 Growth and Growth Methods for Nitride Semiconductors
564
as follows: After the growth of triangular stripes, the growth temperature is increased up to 1150 C for enhanced lateral growth and high incorporation of oxygen, approximately 2 · 1018 cm3, results from growth on f1 1 2 2g facets. Facet- and polarity-dependent incorporation of O in GaN is known. Above the apex, (0 0 0 1) facets reappear and O is decreased owing to both temperature and facet dependence incorporation. CL spectra taken from an array of GaN islands obtained by SAE show that the yellow emission comes from the pyramid apex [703]. Close examination of the dislocation distribution in the pyramid and YL led to the conclusion that YL emission does not originate from dislocations. From top view and cross-sectional wavelength-resolved CL of SAE GaN on Si(1 1 1), Cho et al. [704] arrived at similar conclusions in that the band edge emission is nonuniform and is highest at the top of the apex and over the lateral area, whereas the YL emission originates from the part of the pyramid grown directly on the template in the windows. As is discussed in Volume 2, Chapter 1 in detail, the YL emissions are associated with transitions to Ga vacancy complexes and should not be used necessarily as a measure of degraded quality GaN. Quite on the contrary, GaN templates with low dislocation densities show strong YL emission, which is also true for samples that have high quantum efficiencies. Elsner et al. [705] suggested that the stress field in the core of dislocations could act as a trap for impurities such as oxygen or intrinsic defects such as VGa, and thereby enhance VGa–O complex formation, which are most likely responsible for yellow emission. 3.5.5.3 Nanoheteroepitaxy and Nano-ELO This NHE scheme, which is shown schematically in Figure 3.131, has the potential of eliminating defects more efficiently than other ELO techniques in that coalescence should take place when in the realm of nanodimension with ensuing overall strain relaxation for minimal defect propagations. The pitch, meaning the center-to-center distance of the array, as well as the dot size is important, and the deposition conditions would have to be optimized for a given set of dimensions. The nanoheteroepitaxy of
Coalescence
b a Defect region Silicon islands
GaN Void 40 nm
SiO2
80–300 nm
360–900 nm
SOI
Silicon Figure 3.131 Patterned SOI substrate with islands created in the upper Si layer. The selective growth of GaN is also illustrated. A defected region is located at the heterointerface. Point a corresponds to selective growth of islands on a patterned substrate. Point b corresponds to the case where lateral epitaxy has been performed already [706].
3.5 The Art and Technology of Growth of Nitrides
GaN on patterned h1 1 1i oriented silicon-on-insulator (SOI) substrates has been reported by organometallic vapor phase epitaxy [706]. Transmission electron microscopy reveals that the defect concentration decays rapidly away from the heterointerface, congruous with the predictions [465]. The melting point of the nanoscale islands was found to be significantly reduced, which enhanced substrate compliance and further reduced the strain energy in the GaN epitaxial layer. The NHE approach begins with patterning a substrate into a two-dimensional array of 10–300 nm sized nucleation islands. This is then followed by selectively growing epitaxial material vertically on the islands in the case where the bare substrate is used. In the case of SiC as the substrate, the AlN array is generated on AlN on SiC with lithography followed by etching. These processes are followed by lateral overgrowth to attain coalescence as shown in Figure 3.131. In contrast to current epitaxial lateral overgrowth and pendeo approaches, where the patterning is on the scale of 1–10 mm, the nanoscale patterning in NHE allows enhanced substrate compliance mechanisms to operate. For example, in addition to strain partitioning similar to that found in planar compliant structures, stress in the NHE sample decays exponentially away from the heterointerface with a characteristic decay length proportional to the diameter of the island. Strain partitioning and stress decay interact synergistically to significantly lower the strain energy in nanostructural lattice-mismatched material systems. NHE theory predicts that mismatch dislocation formation can be eliminated from materials systems with a lattice mismatch in the range of 0–4%. Defects are probably unavoidable in material systems with larger lattice mismatch such as GaN on Si (20% lattice mismatch); however, as we show below, NHE results in a significant reduction in the local defect density in nanostructural GaN/Si islands. In addition to exploiting three-dimensional stress relief mechanisms made possible by the geometrical effects, nanostructuring can also enable other stress relief effects and impede defect propagation owing to the large surface-to-volume ratio afforded by nanoscale structures. The case in point is that in material systems where the epitaxial material has a larger melting point than that of the substrate, active compliance can be enhanced owing to softening of the nanoscale substrate islands at growth temperatures while remaining well below the substrate bulk melting point. Recent results on nanophase materials have shown that the materials properties (inclusive of the melting point) of nanoscale particles are significantly modified [707]. In the context of NHE, this reduced melting point can be taken advantage of during growth to soften the silicon islands (i.e., reduce their elastic modulus) and enhance the substrate compliance. Evidence for this has been previously presented [706]. Active compliance has been incorporated in the strain partitioning equations by adjusting the elastic compliance ratio, K, given by [706] K ¼ Yepi ð1 nsub Þ=½Y sub ð1 nepi ÞÞ;
ð3:65Þ
where Y is Youngs modulus and n is the Poissons ratio in the epitaxial material and the substrate materials as depicted by its subscripts. At temperatures well below the actual melting point of the epitaxial and substrate materials, the Youngs moduli of
j565
j 3 Growth and Growth Methods for Nitride Semiconductors
566
1.0
Normalized interfacial strain, e/eT
Interfacial epilayer strain
0.8
0.6
0.4 Interfacial substrate strain
0.2
K= 1K K= 5K
0.0 10
-1
10
0
10
1
10
2
Epitaxial layer thickness (nm) Figure 3.132 Normalized interfacial strain in the epilayer and substrate versus epilayer thickness. Without active compliance (solid lines) the final amount of strain in the epilayer and substrate are similar. With active compliance (dashed lines) much more of the total strain is transferred to the substrate [706].
these materials are that for their respective bulk values. For growth temperatures near the reduced melting point of the substrate island, Ysub is reduced, which causes K to be increased above its bulk value, K0. The effect of active compliance on strain partitioning is illustrated in Figure 3.132, which shows the effect that increasing K has on the normalized strain, e/eT, in the epitaxial and substrate layers, where eT is the total mismatch strain. It is not clear by how much the value of K can be increased beyond K0. To illustrate the active compliance effect, however, an increased value of 5 K0 was arbitrarily chosen for comparison in Figure 3.132. The graph shows that active compliance enhances strain partitioning by increasing the partitioning at a given thickness and increasing the eventual degree of strain partitioning. Moreover, active compliance acts to directly lower the total strain energy in the system when Ysub is reduced because strain energy is proportional to the Youngs modulus [708]. The dot size is a pivotal parameter as it affects both the elastic modulus, through the actual melting point of the substrate island, and the characteristic decay length of the strain energy. XTEM micrographs of GaN growth on two silicon islands, one 80 nm in diameter and the other 280 nm in diameter, show that the defects are concentrated near the GaN/Si heterointerface and the defect density decays rapidly and nearly vanish at approximately 20–50 nm from the interface. The general shape of the strain contours observed indicates that the strain originates at the GaN/Si
3.5 The Art and Technology of Growth of Nitrides
heterointerface. The contour lines broaden and separate, away from the heterointerface, indicating a decaying strain field in agreement with the predictions. Highresolution XTEM revealed stacking faults in the GaN near the interface and a highly defected silicon layer at the interface. Segmented silicon lattice planes lying within the 2.0-nm thick defected layer indicated that the layer is composed of silicon, implying that much of the misfit strain is accommodated by the silicon [706]. A cross between nanoheteroepitaxy and ELO process is the nano-ELO process. Unlike ELO, where the dimensions of stripes for growth are in the microns, in this method, the windows for growth are nanometer sized and are expected to eliminate many of the shortcomings of ELO as described below: Following the formation of nanosized patterns, either stripe or dot, the nanosized growth windows allows for prismatic GaN growth with thickness at the top below the critical thickness allowing extended defect-free material to be obtained. This is contingent upon the AlN layer being extended defect free, which should be possible when grown on nearly latticematched SiC substrates and provided that the substrate is defect free. This is schematically shown in Figure 3.133a up to the point of prismatic GaN growth. (a)
(1 1 0 1)
(1 1 0 1) GaN selective growth
AlN
SiC substrate
Figure 3.133 (a) Artistic rendition of the nanoELO process through the selective growth sequence. (b) Artistic rendition of the nano-ELO process through the entire sequence. The triangular diagram on the right shows a TEM cross-sectional image of selective growth of GaN
over windows in SiO2 with extended defects. The nanotriangular sections in the nano-ELO case are expected to be defect free because of small dots or stripes size and well-defined prismatic planes limiting the extent of vertical growth.
j567
j 3 Growth and Growth Methods for Nitride Semiconductors
568
Figure 3.134 AFM image, left, of 150 nm diameter AlN dots formed on SiC by reactive ion etching. SEM image, right, of selectively grown GaN on AlN nanodots by OMVPE (courtesy of Steve Hersee and Steve Brueck).
Shown in Figure 3.133b is the completed process where growth conditions are changed to favor lateral growth and coalescence followed by a thicker GaN growth after the process shown in Figure 3.133a. The actual images of selectively grown GaN prismatic dots with ð1 1 0 1Þ prismatic planes exposed are shown Figure 3.134. The one on the left shows the AFM image of 150 nm AlN dots with 500 nm center-to-center distance on SiC. The image on the right is an SEM view of GaN after the selective epitaxy of GaN on AlN pillars on a 6HSiC substrate. As mentioned earlier, the prismatic GaN grown in nano-ELO should be defect free because of its small thickness and three-dimensional strain-minimizing shape. This is in contrast to the conventional ELO where the base for the selective growth is large and simple geometry with ð1 1 0 1Þ prismatic planes indicates the layer thickness to reach a tip is beyond the coherence limit, and therefore is defect laden as shown on the side of Figure 3.133b. Following prismatic growth, the growth temperature can be increased as in conventional ELO for enhanced lateral epitaxial growth for coalescence. Assuming successful coalescence, the GaN top layer at some point would have to relax in which case the misfit strain would have to be transferred to AlN pillar causing them to expand in plane at the GaN/AlN interface. However, depending on the lateral dimension, some of this expansion may be provided by SiC, if SiC pillars also are formed during the etching process forming the AlN pillars. This is aided by the presence of voids between the pillars. It is therefore reasonable to expect that at least for small-area wafers, top layers with reduced defect density could be obtained. Preliminary results confirm this assertion, as shown in Figure 3.135, but additional investigations and statistics will have to be performed before this technique can be adopted. The image on the left is an AFM view of 150 nm AlN pillars formed on 6H-SiC by reactive ion etching. The center-to-center distance is about 500 nm. Eventually, pillar size could be reduced to 30 nm or less with center-to-center distance as low as 60 nm or so. The lithographic methods that can be employed could be laser interferometric and electron beam writing in nature. The latter is rather slow, as the dimensions get smaller. However, if a master stamp could be produced by using either of the two methods, a
3.5 The Art and Technology of Growth of Nitrides
Figure 3.135 AFM images of OMVPE-overgrown GaN on nanoAlN dots after having been subjected to various KOH-based defect delineation procedures along with defect densities counted for each of the three cases.
stamping method (nanoimprint lithography) could be used to transfer the pattern to substrates. Other methods are also explored for this purpose. Preliminary investigations by defect delineation etches, discussed in Section 4.2.4, showed defect densities in coalesced films under low 107 cm2, which is remarkable considering a thickness of 2 mm for the coalesced GaN film. However, TEM investigations as well as a larger sample base must be established before this figure can be taken seriously. 3.5.5.3.1 SiN and TiN Nanonets Methods similar to those mentioned above, but not requiring lithography, have also been developed. However, it should be stated that lithographically defined patterns allow one to tailor the growth conditions for proper coalescence followed by growth across the entire wafer. In these methods, SiN or TiN nanomesh through which the GaN is nucleated followed by lateral epitaxial growth have been employed [709–715]. A schematic view of the process in the form of dual SiNx blocking layer and in the context of SiC substrates is shown in Figure 3.136.
Figure 3.136 Schematic diagram of the structure of GaN epitaxy by OMVPE on a 6H-SiC substrate with dual SiNx defect blocking layers.
j569
j 3 Growth and Growth Methods for Nitride Semiconductors
570
In the case of SiNx mesh, SiH4, and NH3 gases were simultaneously introduced during OMVPE growth at a low temperature to form and shape the SiNx mesh before the growth of an initial low-temperature GaN buffer layer [709,710,714]. The threading dislocation originating from the interface between low-temperature buffer layer and high-temperature GaN overlayer decreased considerably from 7 · 108 cm2 in the conventional OMVPE grown layers. Atomic force microscopy images indicate that introducing SiH4 and NH3 gases at a low temperature changes the surface morphology of GaN, which possibly may enhance the lateral growth that is most likely responsible for decreased dislocation density. It is well known that SiH4 and NH3 react above 700 C to form SiNx. It is also well known that SiNx is etched in H at elevated temperatures comparable to temperatures used for this particular approach. It is very likely that SiN is of porous nature, which paves the way for nucleation in the openings followed by lateral growth. At first sight, the mechanism in effect might be related to an early work, mainly reported by Tanaka and coworkers [716–719], dealing with quantum dot formation with the aid of Si antisurfactant. The self-assembly of GaN QDs was reported to occur in this small lattice-mismatched system by exposing AlxGa1xN surface to Si during the growth. The Si is from tetraethylsilane [Si(C2H5)4: TESi: 0.041 mmol] (TESi) and carried by H2. The presence of Si and ammonia in the system are required for the formation of SiNx. However, additional research lead to the mechanism responsible for dot formation evolving from Si antisurfactant to SiN with holes for selective growth. The results about to be reported here supports the nucleation from pores formed in SiNx rather than any antisurfactant effect. The steps involved with SiNx mesh are as follows: after initial annealing of sapphire substrate, the temperature is decreased to 500 C and SiH4 (20 sccm) and NH4 (5 slm) gases are introduced simultaneously for certain length of time depending on the sample for growth at 30 Torr pressure. The SiNx deposition time evolved, first in the range of 50–150 s, which later got extended to 6 min. In addition, the chamber pressure also got increased to 200 Torr in conjunctions with SiNx deposited for a period up to 6 min in experiments that followed. The basic method has already been applied to LED structures emitting at 355 and 348 nm where a factor of two improvement in the output power resulted owing to SiNx mesh, everything else being equal [720,721]. A systematic investigation of SiNx nanomesh in the form or single and dual blocking layers have been undertaken by OMVPE initially on SiC substrates [722–724]. However, growth optimization further led to the conclusion that the longer the SiNx deposition the better the sample quality is, particularly when accompanied with 200 Torr chamber pressure. Deposition times up to 6 min have been explored with 6 min samples leading to radiative recombination times of about 2.5 ns and 5 min films leading to dislocation densities below mid 107 cm2, which is remarkable in 5.5 mm films. The longer the SiNx deposition time, fewer the nucleation sites and more difficult it is to coalesce the films. The results are that threading dislocation density has been reduced to varying degrees depending on the buffer layers used. SiNx was found to be most effective in reducing the edge-type dislocations by an order of magnitude, supported by a reduction in half width of the ð1 0 1 2Þ X-ray diffraction peak. For SiNx nanomesh (used in conjunction with low pressure, 30 Torr initiation layer) about to be discussed both high-temperature and low-temperature GaN
3.5 The Art and Technology of Growth of Nitrides
initiation layers were used. The SiNx nanomesh was more effective with lowtemperature nucleation layers as they tend to have higher dislocation count in comparison with the high-temperature variety. Interestingly, in some cases, the edge dislocation tended to bundle up at the grain boundaries often forming arrays of short dislocation leaving the central part of the grains to be much higher quality. This may in fact be the reason for notable improvement in the minority carrier lifetime observed with each added SiNx nanometer [725]. In addition to defect reduction, SiNx was noted to be also useful in mitigating the surface crack problem commonly experienced in GaN grown on SiC. Shown in Figure 3.137 are the cross-sectional TEM micrographs of the control sample (without SiNx) and the sample with double SiNx layers, both grown on HT GaN nucleation layers on SiC by OMVPE. Most of the threading dislocations originating from the buffer layer in the control sample propagate and reach the top GaN surface. On the contrary, in the sample with a SiNx nonporous network, a large number of threading dislocations are blocked and buried in the GaN vicinal to the SiNx porous layer. A close inspection of the cross-sectional image of Figure 3.137b clearly indicates dislocation consolidation mediated by SiNx. Change of directions of the Burgers vectors, bunching and re-routing of dislocations, and even direct filtration of threading dislocations by the SiNx masks are evident. As a result, a large portion of the threading dislocations has been filtered out by the first SiNx interlayer. With optimized growth, however, and 5 min SiNx template deposition, the dislocation filtering has been made more effective as shown in Figure 3.137c [726]. To get a more accurate account of dislocations reaching the surface, plan view TEM images were obtained with the added advantage that details specific to each type of dislocation could be garnered. Shown in Figure 3.138 are the plan view TEM micrographs taken from the three GaN samples grown on HT GaN buffer layers. The end-on screw-type dislocations (marked as s) show strong, characteristic contrast aligned with the imaging reflection vector, whereas the pure edge type (marked as e) and mixed dislocations have much weaker contrast owing to smaller Burgers vectors and strain-relieving array configurations. Figure 3.138a associated with the control sample displays a high density of edge dislocations (1.2 · 109 cm2) for the GaN grown without SiNx interlayer (control sample). With the use of a single SiNx interlayer (Figure 3.138b), the edge dislocation density is reduced by a factor of 3 to 3.7 · 108 cm2. The second SiNx layer (Figure 3.138c), however, shows a limited effect in further reducing the edge dislocation density (3.6 · 108 cm2). However, the time-resolved luminescence measurements indicated considerable improvement with the second SiNx indicating that although dislocation density is not reduced substantially, the nonradiative recombination center density is [725,727]. This may be because of dislocations being confined to grain boundaries to a first extent leaving the main part of grain boundaries of relatively much higher quality. It may also be because of the electrical character of dislocations change. An additional advantage of SiNx insertion layers is that the crack density, caused by residual tensile thermal strain is reduced. With further optimization of growth [728], however, and 6 min SiNx template deposition, the dislocation density has been substantially reduced to 4.4 · 107 cm2 for screw and 1.7 · 107 cm2 for edge type dislocations (Figure 3.138d).
j571
j 3 Growth and Growth Methods for Nitride Semiconductors
572
3.5 The Art and Technology of Growth of Nitrides
Although time-resolved PL (TRPL) measurements are very instructive in gaining an understanding of the carrier recombination dynamics with direct implications to light-emitting and detecting devices and indirectly to others, reports on the TRPL lifetimes of excess carriers in GaN at 300 K are limited. To cite a few, Kwon et al. [729] reported a biexponential decay with 150 and 740 ps time constants for high-quality Si-doped OMVPE-grown GaN/sapphire. Decay times between 205 and 530 ps have been reported for thick HVPE-grown templates [730–732], and figures ranging from 445 ps [733] to 506 ps [732] have been reported for homoepitaxially grown GaN layers. In addition, Chichibu et al. [734] and Izumi et al. [735] reported biexponential decays with lifetimes (t1, t2) of (130, 400) ps and (80, 459) ps, respectively, for GaN/sapphire films grown using ELO. However, the same groups obtained longer biexponential TRPL lifetimes of (98, 722) ps [734] and (130, 860) ps [735] for bulk GaN. In ELO GaN epitaxial layers prepared in authors laboratory, a radiative recombination time of 0.6–0.9 ns have been measured. € ur et al. [725] reported on the minority carrier lifetime in GaN films grown with Ozg€ SiNx nanonet over low-temperature and high-temperature grown nucleation layer as well as high-temperature AlN nucleation layers. To remind the reader, the LT GaN nucleation layer while providing good coalescence owing to the high-density but small nucleates lead to poorer films. This is rectified somewhat by the hightemperature GaN nucleation layers but at the expense of difficulties associated with full coverage of the surface during the nucleation layer stage. High-temperature AlN nucleation layers cover the surface fairly well while providing high-quality overgrown GaN layers. GaN grown with no (control), single, and dual SiNx blocking layers were € ur et al. [725]. The decay times investigated in terms of X-ray, CW PL, and TRPL by Ozg€ for all the samples in the TRPL experiments could well be characterized by a biexponential decay function: A1 expð t=t1 Þ þ A2 expð t=t2 Þ In a series of tables that follow, namely, Tables 3.15–3.17, the CW PL, symmetric and asymmetric X-ray peak half width data, t1, t2, and amplitudes associated with decays with t1 and t2 character, are tabulated for control, single SiN blocking layer (s-SiNx), and double SiN blocking layer (d-SiNx) in samples with LT GaN nucleation buffer layers, HT GaN nucleation buffer layers, and HT AlN nucleation buffer layers. Basically, the structures with LT GaN nucleation layers see steady improvement as one goes from the control to the single-SiNx then onto the double-SiNx layer sample. With HT GaN nucleation layer with better starting quality, there is still improvement with s-SiN but not as much with d-SiN, which may represent an issue with that particular sample. With high-temperature AlN nucleation layer, we have best overall results even for the control with steady improvement each time a SiNx blocking layer is introduced (see Table 3.15). With optimized SiNx, the radiative recombination lifetime has been improved to about 2.5 ns in a 5.5 mm thick GaN with double SiNx (4.5 min þ 4 min) 3 Figure 3.137 Cross-sectional TEM micrographs of GaN overlayers grown on HT-GaN nucleation layers (a) without SiNx (control sample), and (b) with double SiNx nanonet layer grown by OMVPE on 6H-SiC substrates, (c) cross-sectional TEM image of an optimized 5.5 mm thick GaN with a single 5 min SiNx nanotemplate grown on sapphire.
j573
j 3 Growth and Growth Methods for Nitride Semiconductors
574
3.5 The Art and Technology of Growth of Nitrides Table 3.15 XRD and 10 K PL linewidths and TRPL decay constants
and amplitude ratios (at 200 mJ cm 2 excitation density) for GaN thin layer samples grown with single (s-SiN) and double (d-SiN) SiN layers on high-temperature GaN buffer layers.
Control s-SiNx d-SiNx Improved d-SiNx (4.5 þ 4 min)
10 K PL D0X (meV)
XRD ð0 0 0 2Þ; ð1 0 1 2Þ ðarcminÞ
s1, s2 (ns)
A2/A1
3.9a 4.2a 3.6a 2.5
4.7, 8.5 3.6, 4.9 4.3, 5.2 4.2, 3.51
0.04, 0.54 0.16, 0.45 0.10, 0.85 0.494, 2.488
0.04 0.30 0.62 0.876
The time-resolved data fit biexponential decay function A1 expð t=t1 Þ þ A2 expð t=t2 Þ well. a The PL FWHM values are obtained by fitting to the high-energy side of the PL spectra, because the PL peaks show too much broadening on the low-energy side.
nanonet layers and in single SiNx (6 min) nanonet layers that are not completely coalesced. This compares with 0.85 ns in double but thin SiNx mesh samples. Also important is the A2/A1 ratio of 0.876 in the completely coalesced double SiNx layer as compared to 0.62 in thin SiNx nanonet samples. Excellent surface morphology with well-resolved biatomic steps with corresponding roughness has been obtained in films with up to 5 min SiNx nanonet layer. Further optimization is needed for longer SiNx deposition times. In addition, current–voltage measurements made in GaN layers grown with 4.5 min SiNx nanonet improved Schottky characteristics with barrier height for Ni/Au reaching 1.2 Vand increasing the breakdown voltage from about 50 V to nearly 250 V. The deep level concentration also improved as discussed in Section 4.4. A variant of SiNx, TiN was used to grow GaN with HVPE for improved quality, which also has the added benefit that the void between the initial nucleation centers Table 3.16 XRD and 10 K PL linewidths and TRPL decay constants
and amplitude ratios (at 200 mJ cm 2 excitation density) for GaN thin layer samples grown with single (s-SiN) and double (d-SiN) SiN layers on high-temperature AlN buffer layers.
Control s-SiN d-SiN
10 K PL D0X (meV)
XRD ð0 0 0 2Þ; ð1 0 1 2Þ ðarcminÞ
s1, s2 (ns)
A2/A1
4.3 2.3 2.3
8.2, 7.9 7.8, 8.4 7.5, 4.4
0.13, 0.32 0.20, 0.45 0.29, 0.77
0.06 0.42 0.38
The time-resolved data fit biexponential decay function A1 expð t=t1 Þ þ A2 expð t=t2 Þ well.
3 Figure 3.138 Plan view TEM micrographs of GaN overlayers grown on HT-GaN nucleation layers on 6H-SiC substrates by OMVPE: (a) without SiNx (control sample), (b) with single SiNx, (c) with dual SiNx. The term e represents
for edge-type dislocations while s is for screwtype dislocations, and (d) optimized 5.5 mm thick GaN with a single 6 min SiNx nanotemplate grown on sapphire having 4.4 · 107 cm2 screw, and 1.7 · 107 cm2 edge type dislocation density.
j575
j 3 Growth and Growth Methods for Nitride Semiconductors
576
Table 3.17 XRD and 10 K PL linewidths and TRPL decay constants
and amplitude ratios (at 200 mJ cm 2 excitation density) for GaN thin layer samples grown on TiN templates. Sample Ti anneal time in NH3
15 min 30 min 45 min 60 min Control Freestanding GaN
FWHMFXA (meV)
FWHMD0X (meV)
XRD XRD (0 0 0 2) ð1 0 1 2Þ (arcmin) (arcmin) s1 (ns)
3.22 0.15 3.88 0.03 4.00 0.03 3.40 0.03 3.09 0.05
3.90 0.07 4.52 0.05 5.27 0.12 3.24 0.03 3.15 0.05
6.43 5.83 5.09 4.97 5.60
6.11 4.86 4.61 4.57 7.11
0.47 0.01 0.39 0.01 0.45 0.01 0.41 0.01 0.13 0.01 0.34 0.01
s2 (ns)
A2/A1
1.86 0.02 1.32 0.01 1.68 0.01 1.54 0.01 0.30 0.01 1.73 0.02
0.59 0.58 0.54 0.52 0.36 0.33
Data for the control sample with no TiN layer and a bulk GaN sample are also included.
paves the way to easily separate the epitaxial layer from the substrate for a freestanding GaN [711]. The FWHM of (0 0 0 2) and ð1 0 1 0Þ peaks in the X-ray rocking curves in these 45 mm wafers were 60 and 92 arcsec, respectively. The dislocation density was evaluated at 5 106 cm 2 by etch pit delineation process. The void-assisted separation (VAS) process starts with a GaN template having a thickness of around 300 nm that is prepared on (0 0 0 1) 2-in. diameter sapphire substrate with aid of a low-temperature buffer layer using atmospheric pressure OMVPE. This is followed by a 20-nm-thick Ti layer deposited by vacuum evaporation. This GaN template with Ti is annealed in a gas mixture of 80% H2 þ 20% NH3 at 1060 C for 30 min. The last step is the growing of a 300-mm thick GaN layer on this annealed template using an atmospheric HVPE reactor utilizing GaCl and NH3 as gallium and nitrogen sources, respectively. As is typical in HVPE, GaCl is formed in situ upstream in the reaction region by the reaction of liquid Ga and HCl gas at 850 C. Typical partial pressures of GaCl and NH3 are 7 10 3 and 6.3 10 2 atm, respectively. The temperature of the growth zone is kept constant at 1060 C during the growth. The N2 and H2 gases and their mixture are used as carrier gases. The resulting growth rate is 75–120 mm h 1 under these growth conditions. The n-type doping is controlled by varying the flow rate of SiH2Cl2. Figure 3.139 shows SEM images near the GaN/TiN/GaN interfaces grown with varying carrier gas compositions, namely (a) 100% H2, (b) 10% H2 þ 90% N2, and (c) 100% N2. The density of the voids largely depended on the composition of the carrier gas. When 100% H2 was used as the carrier gas, the voids in the template GaN layer were completely refilled by HVPE GaN and few voids remained at the boundary, as shown in Figure 3.139a and the GaN layer grown under this growth condition could not be separated. On the contrary, voids existed at both sides of the TiN nanonet when the N2/H2 mixed carrier gas was used, as shown in Figure 3.139b. Almost all the voids on top of the TiN nano-mesh were contained inside the islands implying that they formed after the islands. In addition, most of them were accompanied by the voids under the TiN nano-mesh. The voids both on top and under the TiN nano-mesh were enlarged
3.5 The Art and Technology of Growth of Nitrides
Figure 3.139 SEM images around the GaN/TiN/GaN boundaries of the samples grown under different carrier gas compositions. (a) 100% H2, (b) 10% H2 þ 90% N2, (c) 100% N2 [711].
when 100% N2 carrier gas was used as depicted in Figure 3.139c. Miyake et al. [736] reported that the lateral growth rate of HVPE GaN in N2 carrier gas is larger than that in the H2 carrier gas. The large lateral growth rate would result in many voids remaining unburied in the GaN layer under the TiN nanonet as shown in Figure 3.139c. However, further investigation is necessary to clarify the mechanism of the formation of the voids on the TiN nanonet. Although GaN layers grown under growth conditions both (b) and (c) separated from the substrate, many cracks tended to be generated in the GaN layer grown in the growth condition (c). Therefore, growth condition (b) was chosen for the thick GaN growth in this study. A weak force must be applied at the growth interface to induce separation after the cooling process of the HVPE growth. Fu et al. [715] extended the TiN nanoporous blocking method to OMVPE growth of GaN on sapphire substrates. In this particular investigation, the process commenced with the growth of a 0.7 mm GaN buffer on a sapphire to be used as the template. A 20-nm thick Ti film was then deposited on the GaN(0 0 0 1) surface by e-beam evaporation. This step was followed by in situ annealing of the Ti film above 1000 C in a mixture of NH3 and H2 gases to form a TiN network. The annealing step and the subsequent GaN growth were carried out in a low-pressure custom-designed vertical
j577
j 3 Growth and Growth Methods for Nitride Semiconductors
578
OMVPE reactor. A relatively thick layer of GaN was then grown on the TiN network at 1030 C, with a constant TMGa flow rate of 78 mmol min1 and a NH3 flow rate of 7.6 l min1. For comparison, a control GaN layer was grown on another piece from the same 0.7 mm GaN template using identical growth conditions but without the TiN network. The concept behind the process is that GaN growth on these TiN networks originates as GaN islands at the microscopic windows of the discontinuous TiN network. Subsequent GaN growth continues from the islands by lateral and vertical expansion, leading eventually to complete coalescence. Figure 3.140a and b shows cross-sectional, bright-field TEM images for samples T63 (13 mm thick and annealed for 30 min in NH3 : H2 ¼ 1 : 1 at 1050 C) and T68 (7.5 mm thick and annealed for 60 min in NH3 : H2 ¼ 1 : 3 at 1050 C) grown on TiN network, respectively. For comparison, Figure 3.140c shows a cross-sectional TEM image of a 5.2-mm thick, control GaN, which has no discernible dislocation reduction above the initial GaN template. Thin and extended surface voids are formed above the discontinuous TiN layer, which may help alleviate interfacial stress and lead to enhanced quality of the coalesced regions [737]. Dislocation reduction is likely to result from TiN layer blocking the dislocation and those that penetrate through the TiN windows change their propagation direction and extend laterally. Additional dislocation reduction then may occur by means of dislocations with different but suitable Burgers vectors recombining to form a single dislocation, or possibly annihilating each other by forming dislocation loops. The degree to which the dislocation density is reduced can be made by plan view TEM micrographs shown in Figure 3.141. The plan view image of the 13-mm thick GaN control layer grown without TiN, shows a high density of edge/mixed dislocation arrays (1.5 · 109 cm2) as marked by e, and a much lower density of isolated end-on screw dislocations (1.3 · 108 cm2) as marked by s. The image of sample T63 in Figure 3.141b shows 10· reduction in edge/mixed dislocations (1.6 · 108 cm2) and 2· reduction in screw dislocations (0.7 · 108 cm2). Sample T68 (Figure 3.141c) has even fewer edge dislocations (0.9 · 108 cm2) and more screw dislocations (1.4 · 108 cm2) than sample T63, consistent with the observations in Figure 3.140a and b. The TiN network is very effective in reducing the density of edge/mixed dislocations but not the screw dislocations, as theirnumber is already small (about 10% of the total dislocations) and they are not bent by the TiN network (Figure 3.140b). The effect of TiN nanonet was found to be more profound on the minority carrier € ur et al. [738] noted lifetime that would be indicated by TEM defect investigation. Ozg€ that the room temperature decay time deduced from biexponential fits to the timeresolved photoluminescence data is 1.86 ns measured for a TiN network sample is slightly longer than that for a 200 mm thick high-quality freestanding GaN. The relative magnitude of the slow-decaying component to the fast-decaying component is also about twice that for the bulk GaN sample and the samples without TiN network. Figure 3.142 shows the TRPL data for the sample grown with TiN nanonet layer that was annealed in NH3 for 15 min (nitridation process), the control sample, and
3.5 The Art and Technology of Growth of Nitrides
Figure 3.140 Cross-sectional electron micrographs of GaN layers grown on TiN networks in (a) T63 which is 13 mm thick and annealed for 30 min in NH3 : H2 ¼ 1 : 1 at 1050 C, (b) T68 that is 7.5 mm thick was annealed for 60 min in NH3 : H2 ¼ 1 : 3, at 1050 C, and (c) represents the control without any TiN network.
j579
j 3 Growth and Growth Methods for Nitride Semiconductors
580
Figure 3.141 Plan view electron micrographs showing the top of (a) a 13 mm thick GaN control sample grown without TiN, (b) T63 (13 mm thick and annealed for 30 min in NH3 : H2 ¼ 1 : 1 at 1050 C), and (c) T68 (7.5 mm thick and annealed for 60 min in NH3 : H2 ¼ 1 : 3 at 1050 C).
a very high-quality HVPE-grown 200 mm thick freestanding bulk layer. The decays for all the samples could well be characterized by a biexponential decay function: A1 exp(t/t1) þ A2 exp(t/t2). Table 3.17 summarizes the decay constants and the amplitude ratios (A2/A1) obtained from the fits using the Levenberg–Marquardt
3.5 The Art and Technology of Growth of Nitrides
Figure 3.141 (Continued )
algorithm, as well as the symmetric and asymmetric X-ray diffraction peak halfwidths. The 15 min nitridation sample exhibits the longest decay time among the samples investigated, and particularly the large (A2/A1 > 0.5) magnitude for the slowdecaying components underscores the increased radiative efficiency in the sample with TiN nanonet formed by 15 min annealing. Although detailed deep level transient spectroscopy (DLTS) analyses (see Section 4.4 for description of the method and its applications to GaN) for samples grown using TiN and SiNx nanonets are not yet available, preliminary data for SiNx (whose deposition time was varied at increases of 0.5 min between 4 and 6 min) have been obtained and compared to standard that for OMVPE samples and standard ELO samples. Specifically, two samples with single (5 and 6 min) and one with double (5 þ 5 min) SiNx nanonetwork interlayers along with a standard ELO sample and a control sample without ELO have been investigated and compared to each other [728]. For each sample, the bias voltage in DLTS measurements was chosen to probe the top 200–300 nm from the surface, and the pulses amplitude and width were adjusted to fill electron traps, and the results obtained are shown in Figure 3.143. The dominant trap A for all the layers under study has a peak at 325 K in the DLTS spectra. The activation energy for this trap determined from Arrhenius plots varied from 0.55 to 0.58 eV for different layers with the capture cross sections being in the low 1015 cm2 range. This trap is similar to E2 (0.58 eV) [739], D2 (0.60 eV) [740], and B (0.62 eV) [741] traps commonly observed in HVPE-, OMVPE- and MBE-grown n-GaN layers and has been suggested to relate to substitutional nitrogen atom on the Ga site [739], in addition to being dependent of the type of Ga source used for growth, as it disappeared with TEG [742]. Clearly, these are all indirect deductions and the origin of this trap still remains unclear.
j581
j 3 Growth and Growth Methods for Nitride Semiconductors
582
Figure 3.142 Normalized time-resolved PL spectra for the TiN nanonet grown GaN sample with 15 min nitridation (annealing) time, control sample, and bulk GaN. The solid lines are biexponential fits to the data.
Figure 3.143 Comparison of the DLTS spectra associated with the topmost GaN : Si layers (n 0.5–1 · 1017 cm3) grown on standard GaN template (reference), templates with single 5 and 6 min SiNx nanonetwork, that with double SiNx nanonetwork (5 min þ 5 min), and standard ELO technique. The rate window for all spectra is 120 s1 [728].
3.5 The Art and Technology of Growth of Nitrides
Peak B (inset of Figure 3.143) located at 155 K corresponds to a trap with activation energy 0.21–0.28 eV, which has also been observed by others [739,741]. The concentration of this trap was nearly two orders of magnitude lower than that of the dominant trap A for all the samples investigated. Trap B concentration was the highest in the reference sample and the lowest in the GaN layer with 6 min SiNx. It has been suggested [743] that trap B is related to screw dislocations, which is also supported by the trend observed among the samples investigated here. As the data indicate, the lowest trap concentration is seen in the sample with 6 min of SiNx followed by the standard ELO. As expected, the highest trap densities are seen in the sample on the standard GaN template. 3.5.5.4 Selective Growth Using W Masks Selectiveareagrowth(SAG)andlateralepitaxialovergrowth(ELO)ofGaNwithtungsten (W) masks [744] and WNx masks [745] using OMVPE and HVPE have also been investigated. The WNx mask was employed to prevent dissolution of the underlying GaN layer owing to the Wcatalytic effect. The WNx mask was produced by nitridation of the W film using NH3, which is already present in the reactor, at temperatures higher than 600 C. Thermal stability of WNx is good and the WNx/n-GaN contact forms a Schottky type. The selectivity of the GaN growth on the W mask as well as the control SiO2 mask was found to be excellent for both OMVPE and HVPE. The ELO GaN layers were successfully obtained by HVPE on the stripe patterns along the h1 1 0 0i crystal axis with the W mask as well as the SiO2 mask. No voids between the SiO2 mask and the overgrown GaN layer were observed. In contrast, there were triangular voids between the W mask and the overgrown layer. The surface of the ELO GaN layer was quite uniform for both mask materials. In the case of OMVPE, the ELO layers on the W mask and SiO2 masks are similar for stripes oriented along the h1 1 2 0i and h1 1 0 0i directions. In other words, no voids were observed between the W or SiO2 mask and the overgrown GaN layer by using OMVPE. As in the case of SiO2 masks, W and WNx masks led to growth patterns where triangular growth results for stripes along the h1 1 2 0i direction and truncated triangular growth results for stripes along the h1 1 0 0i direction, as shown schematically in Figure 3.144. The details of the experimental methods employed are as follows: [745] The growth experiments were performed using atmospheric pressure HVPE and OMVPE systems on 3.0–4.5 mm thick (0 0 0 1) GaN layers grown on the c-plane (0 0 0 1) of sapphire substrates. As usual, a LT buffer layer was first grown by OMVPE on the substrate. A 120 nm thick W film was deposited on the GaN surface by RF sputtering at room temperature. Stripe windows 10 mm wide with a periodicity of 20 mm were formed on the W film with conventional photolithography and wet chemical etching in H2O2 at room temperature. In the case of HVPE, GaCl and NH3 were used as the source gases and N2 was used as the carrier gas. The flow rates of HCl and NH3 were 10 sccm and 0.5 l min1, respectively. The growth temperature was 1090 C. In the case of OMVPE, TMG and NH3 were used as the source gases and H2 was used as the carrier gas. The flow rates of TMG and NH3 were 18.7 mmol min1 and 2.5 l min1, respectively. The growth temperature was 1060 C. The nitridation of W was
j583
j 3 Growth and Growth Methods for Nitride Semiconductors
584
W or SiO 2
Stripes along <1 1 2 0>
W or SiO 2
Stripes along <1 1 0 0> Figure 3.144 Schematic representation of GaN grown in windows opened in W or SiO2 masks. The triangular growth results for stripes along the < 1 1 2 0> direction and truncated triangular growth results for stripes along the < 1 1 0 0> direction. Patterned after a figure provided by K. Hiramatsu and Ref. [745].
performed at 600–950 C in NH3 in H2 or N2 ambience [745]. To remove any O from the surface of the W film, the substrate was annealed in an H2 ambience at 400 C for 10 min. Then, the substrate was heated to the nitridation temperature in H2 and NH3 ambience. Following the nitridation of W, the ELO of GaN was attempted by a twostep growth at 950 and 1050 C at a low pressure (LP) of 300 Torr. The two-step process was to prevent the dissociation of GaN in contact with W by completing the first stage at a lower temperature, which was 950 C for 30 min in this case. In the second step, to bury the W mask easily, a high-temperature (1050 C) growth was performed for 90 min. 3.5.5.5 Low-Temperature Buffer Interlayer In parallel to ELO, other techniques, such as those that do not require exposure to atmosphere part way through the process and photolithography, have been explored. Among them is low-temperature interlayer [746]. The LT buffer layers inserted periodically a few times reduce the dislocation density in the top layer. The LT buffer layers are grown using growth conditions very similar, if not identical, to the standard LT buffer layers. After each LT buffer layer, the structure is automatically annealed in H2 prior to the growth of the high-temperature layers. In the process, the propagating defects collide with the new interfacial defects generated in LT insertion buffer layer and annihilate one another. Defect concentrations have been lowered to about the same value available by ELO using this method. Recently, this method has been extended to grow AlGaN on GaN layers without the notorious cracking effect observed in tensile strained AlGaN on GaN.[746]. The problem is very serious for nitride lasers as the cladding layer thickness is limited by cracking. The reduced thickness employed to avoid cracking causes leakage of the optical field to the GaN buffer layer with deleterious effects.
3.5 The Art and Technology of Growth of Nitrides
The insertion of a low-temperature GaN or AlN interlayer between high-temperature grown GaN reduces the threading dislocation densities down to the low 108 cm2 range [747–750].TEM observations showed a significant reduction of dislocations with a screw component [434], which were found to terminate at the interlayer [751,748]. Similar features were observed in GaN growth using dimethylhydrazine as the nitrogen precursor [752,753]. The mechanisms of reduction in dislocation density are attributed to those in effect in Section 3.5.5.2 dealing with ELO technology. Figure 3.145a and b shows images of two GaN layers grown by OMVPE after they have been subjected to a defect staining etching. Corresponding schematic diagrams of the grown layers are also shown. Figure 3.145a shows a sample grown only on a LT buffer layer, which, in turn, is grown on sapphire. While Figure 3.145b shows an additional LT buffer layer inserted in the main GaN layer. As the images indicate, the etch pit density dropped from about 108 to about 5 · 106 cm2 when an additional LT buffer layer was inserted [754]. LT AlN insertion buffer layers have been examined with a particular attention to electrical transport and optical properties of the resulting GaN layer on top [434]. It was shown that as the number of AlN IL/HT GaN layers increased, the electron mobility increased in the top Si-doped GaN layer, nearly as much as by a factor of two, from 440 to 725 cm2 V1 s1. The dependence of the electron mobility on temperature
Figure 3.145 Images of two GaN layers grown by OMVPE after they have been subjected to a defect staining etch, (a) is for a sample grown only on an LT buffer layer, which in turn is grown on sapphire and (b) has an additional LT buffer layer inserted in the main GaN layer. Courtesy of M. Koike.
j585
j 3 Growth and Growth Methods for Nitride Semiconductors
586
had the characteristic peak at about 100 K before falling off toward lower temperatures. This implies that the mobility was not dominated by any two-dimensional electron gas (2DEG), which could form at each of the interfaces between the AlN layer and the underlying GaN layer and that the measured values do indeed show an improvement in the electrical quality of GaN with LT AlN insertion. Cross-sectional transmission electron microscopy images were remarkable in that a significant reduction in the screw dislocation density for GaN films grown on the AlN IL/HT GaN layers was obtained. This is consistent with the assertion that, in GaN, edge dislocation core energy is minimal along the c-direction and it is unlikely that the edge dislocations would be affected by these AlN insertion layers. Of course, the situation is different with screw dislocations. As for the XRD investigation, the symmetric and off-axis linewidths increased as the number of AlN IL/HT GaN layers increased, indicating a greater relative misalignment of the adjacent HT GaN layers. An interesting observation about the difference between LT AlN and GaN insertion layers is the notably different residual strain in GaN [746]. It was shown that the inplane biaxial stress thickness product increased with each successive repetition of the LT GaN insertion. Further, the HT GaN layers were under tensile strain. On the contrary, the picture with LT AlN buffer insertion remained the same with each insertion of LT AlN buffer layers. Threading dislocations (TDs) were also measured using plan view TEM analysis. A reduction of TDs with each insertion of a LT buffer layer is observed, from about mid 109 cm2 in HT GaN with no LT insertion buffers to about mid to high 107 cm2 when some six LT buffer layers were inserted. Although there were some variations, the TD density did not seem to be that dependent on the type of LT buffer, meaning GaN or AlN is used [746]. Consistent with the structural properties with reduced screw dislocations, electrical transport properties also could be improved by inserting multiple low-temperature buffer layers of the AlN type [755]. For Si-doped GaN films grown on five AlN interlayers, the room temperature electron mobility, mRT, was 725 cm2 V1 s1 and the room temperature electron concentration, nRT, was 1.47 · 1017 cm3, both of which are similar to the electrical properties for GaN grown on both 6H- and 4H-SiC, but without multiple buffer layer insertion. The dependence of the electron mobility on temperature was such that conduction by any two-dimensional electron gas, which would give erroneously high electron mobilities falsely assumed to be for the bulk, was not in effect, as the mobility decreased when the temperature was lowered below about 70–100 K. An increase in mobility accompanied by a decrease in screw dislocations suggest that screw-type dislocations may cause scatterers as well as compensating n-type donors in GaN. 3.5.6 Polarity and Surface Structure of GaN Layers, Particularly on Sapphire
Because GaN does not share the same atomic stacking order as many of the substrates on which it is grown, the crystal direction [0 0 0 1] of GaN film, the direction of the long bond along the c-axis from Ga to N atoms, can be either parallel or antiparallel to the growth direction [756]. The epilayer in the former case is conventionally referred to as
3.5 The Art and Technology of Growth of Nitrides
having Ga-polarity or Ga-face, whereas the latter has N-polarity or N-face [757]. Investigations have shown that these two polar films have vastly differing growth and surface properties. For examples, a Ga-face is typically smoother than a N-face. For MBE growth near stoichiometric conditions, the growth rate of N-polar domains may be slightly lower than that of Ga-polar matrix, leading to the formation of pits with inversion domains at their centers. A p-type doping by Mg is easier in Ga-polar films, whereas C, O, Si, and other residual impurities are more likely to incorporate into Npolar films. The Ga-face is also more stable than a N-face against wet chemical etching. The photoluminescence spectra, the Pt/GaN Schottky barrier, the band discontinuities, and the two-dimensional charges in GaN/AlGaN heterostructures are all affected by the polarity of the structures. Different polarities can be identified by convergent beam electron diffraction, hemispherically scanned X-ray photoelectron diffraction [758], convergent beam electron diffraction [758,759], coaxial impact collision ion scattering spectroscopy, and X-ray standing wave method [760,761]. The polarity impacts the dopant incorporation also in that the N-face is more amenable to incorporation [762]. Readily available to an MBE grower is the RHEED capability, which allows the determination of the polarity of some of the films by inspecting the surface reconstruction during cooldown [763,764]. With more insight and control over the MBE growth process, reports emerged as to which growth parameters lead to which polarity. It has been implied that GaN buffer layers grown at 700 C lead to N-polarity with poor layer quality and that the N-polarity samples are characterized with much higher background concentration and overall inferior quality [765]. Others reported N-polarity when grown directly on sapphire and Ga-polarity when grown on an AlN buffer layer [766,767]. Surface charge sensitive electric force microscopy for determining the polarity of the sample and its distribution on the surface has also been reported [768]. Furthermore, a close correlation between the morphology and the polarity has been established for both as-grown and etched GaN surfaces [769,770], making it possible to determine the film polarity using more easily available techniques such as AFM. Wet chemical etching has been used to determine the polarity in GaN. Particularly, NaOH-, KOH-, and H3PO4-based solutions at different temperatures have been demonstrated to attack N-polar surface, whereas the morphology of Ga-polar surface remained unchanged [758]. For device applications, an understanding and control of the crystal polarity in the epitaxial growth is essential. For this purpose, the effects of the substrate nitridation [771], buffer layer materials, and growth conditions [772], such as III/V ratio, have been investigated. The change in film polarity by Mg doping was also observed [773,774]. It has been demonstrated that, in the case of OMVPE on sapphire substrates, either GaN or AlN low-temperature buffer layers lead to Ga-polar films [757]. For MBE growth, however, the published results show that AlN buffer layers commonly lead to Ga-polar films, whereas GaN buffer layers lead to N-polar films [772]. Although there are suggestions that low-temperature GaN buffer layers may increase Ga domains or lead to Ga-polar films in some cases [771,775], a correlation between the film polarity and buffer layer growth is still not very well established.
j587
j 3 Growth and Growth Methods for Nitride Semiconductors
588
MBE with its inherent control over the growth parameters can be used to interrogate certain structural and electrical processes in the crystal. One such topic is the polarity of the films because the c-plane of sapphire is a polar surface and GaN does not share the stacking order with sapphire. Consequently, GaN grown on sapphire could either be terminated with Ga ((0 0 0 1) or (0 0 0 1)A-face) or N (ð0 0 0 1Þ or (0 0 0 1)B-face) [776,777]. Being non-centro-symmetric owing to its wurtzitic and ionic structure, nitrides exhibit large piezoelectric effects when under stress along the c-direction. Moreover, spontaneous polarization charges also appear at the heterointerfaces owing to the different degrees of ionicity of the various binary and ternary nitrides [778]. The signs of spontaneous polarization and piezoelectric polarization depend on the polarity of the film. This charge and its sign must be known and controlled in electronic devices, particularly in modulation-doped FETs. For example, electric field caused by polarization effects can increase or decrease interfacial free carrier concentration causing the gate potential needed to vary drastically [779]. The polarity of the film can also have an impact on the effective band discontinuities [780]. GaN layers grown by MBE can be either N- or Ga-polarity, and each can be grown under Ga- and N-rich conditions. In the case of growth with ammonia as the nitrogen source and in conjunction with high-temperature growth >800 C, very nitrogen rich conditions lead to better quality films in all respects, that is, PL, X-ray diffraction, and electron mobility. This is in part because of simultaneous removal of material, which leads to the elimination of regions that are not of as high quality. In addition, the tips of the clusters are partially etched away, which leads to smoother surfaces. Systematic RHEED investigations supported by LEED, scanning tunneling microscopy (STM), and theory [764,763,781,357] mapped the reconstructions of GaN(0 0 0 1) surface that may appear on each GaN{0 0 0 1} surface. The 2 · 2, 5 · 5, 6 · 4, and 1 · 1 reconstructions were identified for the Ga-face and the 1 · 1, 3 · 3, 6 · 6, and c(6 · 12) reconstructions were identified for the N-face [764]. Moreover, a 1 · 2 reconstruction and on occasion 3 · 2 and 2 · 3 RHEED reconstructions were also observed for the Ga-face, which was supported by AES, and first principles total energy calculations were used to identify possible atomic structures [781]. Smith et al. [764] have constructed schematic diagrams for the surface reconstructions observed on the N and Ga faces and corresponding RHEED patterns, as viewed along the ½1 1 2 0 azimuth, which are shown in Figure 3.146a and b, respectively. The N-polarity GaN is prepared by nucleating and growing GaN directly on sapphire using RFMBE. The 1 · 1 reconstruction depicted in Figure 3.146 was produced by heating the as-grown film surface to high temperature (800 C) to remove excess Ga adatoms. The 3 · 3, 6 · 6, and c(6 · 12) reconstructions, which are sustainable only below 300 C (Figure 3.146a), were obtained by depositing submonolayer quantities of Ga onto the surface. When the surface temperature is increased, the structures undergo reversible order–disorder phase transitions and the nonintegral RHEED features disappear. The Ga-polarity GaN is obtained by growth on AlN buffer layers, as will be discussed in the next section. However, the work depicted in Figure 3.146b utilized Ga-polarity OMVPE-grown GaN templates on sapphire [764]. During growth at growth
3.5 The Art and Technology of Growth of Nitrides
Figure 3.146 Schematic diagrams along with RHEED images illustrating the coverage and temperature dependence of the RHEED reconstructions on the: (a) N face and (b) Ga face. Ga coverage increases from left to right in both diagrams. Temperatures given correspond
to either order–disorder transitions or annealing transitions. The crosshatched regions indicate either mixed or intermediate phases. RHEED patterns for both the Ga and N face, as viewed along the ½1 1 2 0 azimuth, are also shown. Courtesy of R.M. Feenstra and Ref. [764].
temperatures, a streaky 1 · 1 RHEED pattern is observed, which upon cooling, transforms to a pseudo 1 · 1 at 350 C. When viewed along the ½1 1 2 0 azimuth, satellite lines adjacent to the integral order lines accompanied this 1 1 structure. Additional RHEED patterns from this structure can be found in Ref. [782]. When the
j589
j 3 Growth and Growth Methods for Nitride Semiconductors
590
surface exhibiting the 1 1 reconstruction is annealed at 750 Cto remove excess Ga atoms, the RHEED pattern changes to a 1 2 (not shown in Figure 3.146a), with a weak half order streak. Deposition of Ga onto the 1 2 surface at room temperature removes the half orderstreak and nofractional orderstreaks appear. However, by depositing half ML of Ga and treating the surface to 700 C followed by cooling, the 5 5 reconstruction is formed. Further deposition of an additional half ML of Ga and annealing to 700 C followed by cooling results in a 6 4 reconstruction. The 6 4 reconstruction undergoes a reversible phase transition at 250 C, whereas the 5 5 structure is stable up to 700 C. Deposition of about one ML of Ga on the surface exhibiting the 6 4 reconstruction, followed by a rapid annealing at 700 C, results also in the 1 1 structure.[763] The 2 2 RHEED pattern results when the 5 5 surface is annealed at 600 C or nitridated at 600 C. The dependence of polarity on growth conditions, particularly the type of buffer layers used, in the context of RF MBE growth on c-plane of sapphire substrates has been investigated [783]. It should be mentioned that a low-temperature nitridation, in combination with low-temperature buffer growth, has been reported to lead to interfaces void of cubic GaN and improved quality [784]. Four groups of samples were grown and investigated. The first and second sets utilized GaN buffer layers grown at about 500 and about 800 C, respectively. The third and fourth groups utilized AlN buffer layers grown at about 500 and 890–920 C, respectively. Following the buffer layers, typically 1 mm thick GaN layers were grown at a substrate temperature between 720 and 850 C with growth rates in the range of 300–1000 nm h1 under N-limited (Ga-rich) conditions. In addition to the in-situ RHEED images, AFM, X-ray diffraction, and hot H3PO4 at 160 C were employed to confirm the polarity assignment. Layers with high-temperature GaN buffer layers (around 770 C or higher) invariably turned out to be of N-polarity regardless of whether a static or graded substrate temperature was employed during the buffer growth. Upon cooling, the RHEED pattern indicated only the bulk 1 · 1 structure, though others have reported higher order reconstruction [764]. Conversely, layers with AlN buffers grown in the temperature range of 880–960 C (some are not included in the data presented here) with thicknesses in the range of 8–35 nm and growth rates of 40–60 nm h1 led to Gapolarity. Consequently, a (2 · 2) RHEED pattern was observed upon cooldown at temperatures ranging between 280 and 650 C depending on the V/III ratio employed for the mail layer. The low substrate temperature buffer growth, Ts 650 C but primarily around 500 C, resulted in layers with either polarity with either GaN or AlN buffer layers. The GaN structures with 100–150 nm thick GaN buffer layers at a growth rate of about 600 nm h1 led to Ga-polarity with the characteristic 2 · 2 pattern upon cooldown after the entire structure was completed. However, when the thickness of the buffer layer was reduced to 30–60 nm, keeping the growth rate constant, the layers turned out to be of mixed polarity with a faint 2 · 2 reconstruction observed upon cooldown. When about 110–220 nm thick buffer layers grown at 500 C with 220 nm h1 growth rate were used, the resultant layers were of N-polarity with only the 1 · 1 reconstruction observed during cooldown.
3.5 The Art and Technology of Growth of Nitrides
The AlN buffer layers grown at Ts 650 C, but primarily around 500 C, exhibited Ga- or N-polarity depending on the growth conditions. When 10–15 nm thick buffer layers grown at a rate of 60 nm h1 were employed, Ga-polarity resulted. However, when 2.5–22 nm thick buffer layers were employed with a growth rate of 15–25 nm h1, N-polarity resulted. The typical surface morphologies of as-grown Ga-polar films with different buffer layers are presented in Figure 3.147. In this case, a high-temperature AlN buffer layer tended to result in a smooth, but pitted layer (Figure 3.147a), consistent with the group III/V ratio employed. Higher group III/V ratios generally led to disappearance of the pits. A low-temperature AlN buffer layer grown at a high rate led to a Ga-polar surface morphology with irregular stepped terraces, often with pits and/or a rough surface (Figure 3.147b). When a low-temperature GaN buffer layer grown at high rate was used, a similar morphology to that shown in Figure 3.147 b with a more drastic variation in terrace height and shape was observed (Figure 3.147c). The surface morphologies of as-grown N-polar films with different buffer layers are presented in Figure 3.148. With a high-temperature GaN buffer layer, the film morphology is that of noncoalesced columns (Figure 3.148a). In general, smoother morphologies with stepped terraces were found when a low-temperature AlN buffer layer grown at a low rate was used (Figure 3.148b). Using low-temperature GaN buffer layers grown at a low rate, the morphologies vary from extremely rough surfaces with noncoalesced columns to a surface shown in Figure 3.148c, where very tall columns and terraces are separated by deep troughs. The simple model that explains the N-polarity with GaN buffer layer calls for Ga to form [0 0 0 1] or the long bond to the O-surface of sapphire. In the same vein, the Gapolarity results with AlN buffer layers when O leaves the surface and N forms the [0 0 0 1] or the long bond with Al of sapphire. This simple model, although consistent with high-temperature buffer layers, does not explain our results with low-temperature AlN and GaN buffer layers, which led to either polarity depending on the growth conditions. Detailed investigations are necessary to gain an insight into the mechanisms involved. High growth rates mainly leading to Ga-polarity and low growth rates to N-polarity would indicate that there must be some atomic exchange or interaction that may be suppressed or promoted by large growth rates, depending on the case. Experimental and theoretical investigations [764,785] led to possible models of the above-mentioned surface structures (see Ref. [83] for a detailed account). It has been pointed out that reconstructions on both Ga- and N-face GaN{0 0 0 1} surfaces consist of metallic layers of Ga, bonded to the GaN, unlike other semiconductor surfaces that prefer to be nonmetallic because the opening of a surface gap provides a mechanism for lowering the energy of the system [785]. Moreover, the separation of Ga atoms in bulk GaN along the [0 0 0 1] direction is typically 2.7–2.8 Å, which is smaller than the lattice constant of GaN, 3.19 Å, and should not favor the metallic surface layers. In spite of this, the ð0 0 0 1Þ N surface with the 1 1 reconstruction was found to consist of one ML of Ga atoms (adlayer) located on sites above the N atoms [786]. Charge transfer from Ga to N atoms has been proposed to stabilize Ga- on N-face because the Ga atoms in the adlayer are slightly positively charged so that there is a Coulombic repulsion between them [785]. RHEED intensity variations in combination with
j591
j 3 Growth and Growth Methods for Nitride Semiconductors
592
Figure 3.147 AFM images of Ga-polarity samples obtained under three different buffer layer and growth conditions: hightemperature AlN buffer layer (a), low-temperature AlN buffer layer (b), and low-temperature GaN buffer layer (c).
3.5 The Art and Technology of Growth of Nitrides
Figure 3.148 AFM imges of N-polarity samples obtained under three different buffer layer and growth conditions: hightemperature GaN buffer layer (a), low-temperature AlN buffer layer (b), and low-temperature GaN buffer layer (c). The rough surface morphology can be improved by growing the top GaN layer at lower temperature such as 720 C instead of 800 C, which was the case here.
j593
j 3 Growth and Growth Methods for Nitride Semiconductors
594
0i DMS led Held et al. [777] to confirm the stability of Ga adlayer on gallied h0 1 1 surface. The 3 3, 6 6, and c(6 12) reconstructions are formed by additional Ga adatoms weakly bound on top of the 1 1 Ga adlayer. In particular, the 3 3 structure has one additional atom per 3 3 cell compared with the 1 1 [786]. A lateral relaxation of the nearest-neighbor Ga adlayer atoms allows the adatom to move much closer to the adlayer plane, which aids in stabilizing the structure [786]. The 3 3, 6 6, and c(6 12) reconstructions are characterized by reversible order–disorder phase transitions at 300 C, owing to the weak bonding of the Ga adatoms. Turning our attention to the Ga(0 0 0 1) surface of GaN under N-rich conditions, it is terminated with an arrangement of adatoms, with each adatom forming three bonds with the underlying Ga atoms [785]. Most likely N and Ga atoms participate as adatoms, as can be inferred from the relative surface energies calculated by Northrup et al. [357], which are shown in Figure 3.149. The theoretical calculations also point to the stability of the 2 · 2 N and 2 · 2 Ga adatom structures, as described in Refs [787– 790]. STM observations showed that the 2 · 2 reconstructed surface is well ordered, but only in small domains, which is consistent with the half-order diffraction lines in RHEED not being very sharp [791]. Because the 2 · 2 structure is formed by nitridation, it is most plausible that it is associated with N adatoms at the H3 hollow site, supported by theory [763,789]. It is also plausible that nitridation might convert the surface from a Ga-rich structure to a 3.0 GaN(0 0 0 1)
2.5
Energy (eV/1× 1)
2.0 Laterally contracted Ga monolayer
1.5 1.0 0.5 0.0 –0.5
2 × 2 N adatom
2 × 2 Ga adatom
–1.0 Laterally contracted Ga bilayer –1.5 –2.0 –1.2
–1.0
–0.8
–0.6
–0.4
μ Ga - μ Ga(bulk) (eV) Figure 3.149 Relative energies of the surfaces are plotted as a function of the Ga chemical potential. For Ga-rich conditions the most stable structure is the laterally contracted Ga bilayer. (Courtesy of J. Northrup and Ref. [357].
–0.2
0.0
0.2
3.5 The Art and Technology of Growth of Nitrides
less Ga-rich structure, with the 2 · 2 reconstruction being associated with the Ga adatoms. If so, the calculations indicate that the Ga adatoms prefer the T4 site, right above the second layer N atoms. Regarding the 5 · 5 reconstruction, the experimental evidence suggests adatoms being at H3 and T4 sites with three adatoms and a dangling bond site at a Ga rest atom per each 5 · 5 unit cell [763]. However, the 6 · 4 structure was reported to be quite complicated in origin and is likely to involve more than just adatoms and/or vacancies on the surface [763]. In the Ga-rich limit, the pseudo 1 · 1 structure of the (0 0 0 1) surface contains a fluidlike overlayer of Ga [357,763,790]. Motivated by RHEED and LEED diffraction observations and modeling of the AES Ga/N peak intensities, the premise that there are two monolayers of Ga on top of the Ga-terminated GaN bilayer has been forwarded. In this case, the Ga layers assume spacing close to their bulk values so that they form an incommensurate structure on the surface. Moreover, utilizing first principles total energy calculations, Northrup et al. [357] provided the theoretical backing (Figure 3.149) in favor of a laterally contracted mobile bilayer structure p ffiffiffi pffiffiffi containing 2.3 ML of Ga, as shown in Figure 3.150. Modeling utilized a 3 3 cell, which allowed pffiffiffi one to model hexagonal Ga adlayers with a reduced Ga–Ga spacing of ac ¼ ð 3=2Þa1·1 ¼ 2.75 Å (because a1·1 ¼ 3.17 Å is the in-plane spacing of Ga atoms on the 1 · 1 reconstructed ideal surface). The lattice vectors of the overlayer were rotated by 30 with respect to those of the substrate. In the structure depicted in Figure 3.150, layer 1 contains three atoms and layer 0 contains
0.16 Å
ac
a1×1
Ga-layer 0
z
01
z
12
T1 sites Ga-layer 1
a1×1 Ga-layer 2 N-layer 3
N
Ga
Figure 3.150 A schematic representation of a laterally contracted Ga bilayer above a Gaterminated (0 0 0 1) substrate. The average separations between layers are z12 ¼ 2.54 Å and z01 ¼ 2.3 7Å. The filled and open circles in layer 0 represent a time-averaged image of the Ga atoms. The filled circles in layer 0 correspond to
the positions at a particular time. The timeaveraged vertical corrugation of layer 0 is approximately 0.16 Å. Note: In this projection the laterally contracted monolayer (layer 0) has been rotated by 30 for ease of viewing. Courtesy of J. Northrup et al. and Ref. [357].
j595
j 3 Growth and Growth Methods for Nitride Semiconductors
596
pffiffiffi pffiffiffi four atoms in each ( 3 3) cell. In the Ga-rich limit, the energy of this laterally contracted Ga bilayer structure was the lowest among all the structures investigated, as shown in Figure 3.149 inclusive of the special case of the Ga adatom structure by 0.25 eV/1 1. Several inequivalent registries of layer 0 relative to layer 1 were considered pffiffiffi pffiffiffi within a tolerance of about 0.02 eV/1 1. However, the energy of the ( 3 3) laterally contracted bilayer (LCB) structure was independent of the registry with the substrate. This is consistent with sufficiently mobile Ga atoms and also accounts for the observation of a 1 1 corrugation pattern in STM [763]. The Ga-bilayer model associated with the stable structure of the Ga-rich (0 0 0 1) surface has been supported by another research group [350,792]. In these investigations, the amount of excess Ga on the surface was measured by observing the transients in the intensity of the specular RHEED beam. However, confirmation of the incommensurate bilayer structure [785] remains, because STM imaging of the (0 0 0 1) surface did not yield the expected incommensurate pattern (i.e., series of fringes). Moreover, the observed atomic spacing corresponds to the 1 · 1 spacing (3.19 Å) as opposed to a smaller value by 1/6 or 1/12 as implied by the diffraction experiments [782]. There have been several other investigations of the two GaN {0 0 0 1} surfaces inclusive of a discussion of theoretical calculations of Ref. [793]. For a detailed account, the reader is referred to Ref. [79]. studies pointed to additional surface pffiffiffi These pffiffiffiffiffi reconstructions, such as the 10 · 10, 5 3 2 13 and 4 4 for the (0 0 0 1) face and pffiffiffi pffiffiffi 6 8, 7 7 and 2 3 for the N-face. A 2 pffiffiffi pffiffiffi 2 reconstruction has been also reported for a ð0 0 0 1Þ surface [794] and ( 3 3) R30 for a (0 0 0 1) surface [795]. The observation of a 2 · 1 RHEED pattern on a ð0 0 0 1Þ surface has been attributed to a poorly ordered 2 2 reconstructed surface, in which the twofold periodicity is more easily seen with the electron beam along the ½1 1 2 0 azimuth [794]. Another important issue that could have a bearing on the surface reconstruction is that of contaminants. Several reports have indicated the sensitivity of the surface reconstruction on the presence of contaminants. A 4 · 4 reconstruction on the Npolar surface has been attributed to oxygen or arsenic contamination, the latter associated with systems that have previously been used for arsenide semiconductors [795,796]. A body of experimental work attributed the 2 · 2 reconstruction [797,798] to impurities such as As or Mg. It was shown that a partial pressure of 109 Torr of arsenic modified the growth kinetics significantly (As behaves as a surfactant) and induced a 2 · 2 surface reconstruction during growth of GaN (0 0 0 1) [799]. The proposed model of this 2 · 2 structure consists of one As adatom per 2 · 2 unit cell [799,89]. Moreover, adding small amounts of Ga to the arsenicinduced 2 · 2 surface resulted in the formation of 4 · 4 and 5 · 5 reconstructions [781], in agreement with Ref. [788]. A 4 · 4 reconstruction was also observed at low temperature in Ref. [797]. A 2 · 2 reconstruction is also observed on the (0 0 0 1) surface during MBE growth with NH3 gas source [777] and could be pffiffirelated ffi pffiffiffi to the adsorption of H atoms.[787]. A case related to InGaN growth, a stable ( 3 3) R30 reconstruction (1 · 3 RHEED pattern) on the GaN(0 0 0 1) surface has been induced by In atoms after sufficient nitridation [83]. The same reconstruction has also been observed on InGaN surface under N-rich conditions [800].
3.5 The Art and Technology of Growth of Nitrides
The affect of in situ annealing on the surface structure has also been investigated using such techniques as LEED. The LEED patterns of clean GaN{0 0 0 1} 1 · 1 surfaces, if heated to temperatures above 830 C, have exhibited splitting of the normal order spots into circular sextets (multiplets of six spots) as a function of the primary electron energy [801]. These have been attributed to faceting of the surface [84] and alternatively the splitting may be owing to oppositely oriented, regular step arrays in the ½2 1 1 0, ½1 1 2 0, and ½1 2 1 0 directions on the GaN{0 0 0 1} surfaces [801]. The multiplets of sharp spots evolve from weakly diffused rings as the annealing temperature was increased above 830 C, which was attributed to the formation of straight step edges oriented perpendicular to the ½2 1 1 0, ½1 1 2 0, and ½ 12 1 0 directions, meaning rotated by 60 each. A quantitative exercise led to terrace widths of 11.0 1.0 Å and step heights of 5.25 0.2 Å. This surface structure may be the result of the development of growth spirals around the screw dislocations by thermal etching of the surface in much the same way as the development of growth spirals observed under Ga-rich conditions [782,785]. 3.5.6.1 MBE Buffer Layers As in the case of OMVPE, growth on sapphire with MBE requires a short period of nitridation of the sapphire substrate before the buffer layer growth, generally between 10 and 30 min, or until the RHEED image indicates a transition from sapphire to AlN with the accompanying 30 rotation. During the nitridation process, a thin AlO1xNx film may form on the substrate surface [802] transforming Al2O3, in a natural way, through AlO1xNx into AlN, and the epitaxial films grow on AlN, which has nearly the same lattice constant as GaN (mismatch 2–4% in c-plane). This is why the best films until now have been obtained on sapphire with an AlN buffer layer. It is obvious that nitridation parameters can greatly influence the properties of the epitaxial layer [803,804]. A set of RHEED patterns showing the nitridation stage and subsequent growth of AlN is shown in Figure 3.151. The other buffer layer employed is GaN, either a low-temperature or a high-temperature variety. In general, AlN and GaN buffer layers grown at relatively high temperatures lead to Ga- and
Figure 3.151 (a) RHEED image taken along the ½1 1 2 0 direction of AlN, rotated about 30 with respect to that of sapphire, during nitridation of sapphire; (b) RHEED image taken along the ½1 1 2 0 direction of AlN during AlN growth on sapphire.
j597
j 3 Growth and Growth Methods for Nitride Semiconductors
598
Figure 3.152 (a) RHEED imaging of GaN deposited on a nitridated sapphire, taken along ½1 1 2 0 direction at 800 C and represents N polarity of the film; (b) RHEED image of AlN grown on GaN of Figure 3.151 taken at 800 C along the ½1 1 2 0 azimuth, and represents a N-polarity film. The bright spots represent the specular reflection associated with each diffraction.
N-polarity films, respectively [805]. Those grown on buffers prepared at lower temperatures can lead to either polarity if grown on AlN buffer layers, depending on the particulars, as discussed in Section 3.5.6. When grown on low-temperature GaN buffer layers, the film has N-polarity. Shown in Figure 3.152 are RHEED images of GaN buffer layers deposited directly on nitrided sapphire and AlN grown on that GaN buffer layer. These images are for an N-polarity film and upon cooling do not show the characteristic 2 · 2 pattern observed in Ga-polarity films. Instead, it shows a 1 · 1 pattern, although a 3 · 3 diffraction should have been observed. A 1 · 1 pattern can also be observed for a Ga-polarity film for a range of group III/V ratios, but under the conditions used, Ga-polarity films lead to strong 2 · 2 patterns. Unlike Ga-polarity films where the edge dislocations are dominant and present in concentrations above 109 cm2, N-polarity films have much fewer extended defect concentrations, and the concentrations of screw, edge, and mixed dislocations are comparable. Moreover, the N-polarity films exhibit wider symmetric diffraction peaks and higher photoluminescence efficiency, as compared to the Ga-polarity ones, which indicates that the edge dislocations are not benign, the details of which are discussed in Volume 2, Chapter 5 in terms of optical properties and in Chapter 4 in terms of electrical nature of dislocations. 3.5.7 Growth on ZnO Substrates
ZnO is considered a promising substrate (yet to really be implemented, but quality of GaN on ZnO has been improving steadily. Because GaN can be separated from ZnO easily, this method is considered by some as an attractive freestanding GaN template preparation method) for the III-N semiconductors because it has a close match for GaN c- and a-planes and an identical stacking order with GaN [806,807], and GaN follows the in- and out-plane orientations of ZnO. In the context of GaN, ZnO has been used in two capacities. One is the buffer layer, particularly for HVPE growth of
3.5 The Art and Technology of Growth of Nitrides
GaN, and the other as a substrate. Several techniques have been employed and are being developed to grow nitride-based compounds on different kinds of substrates. Vispute et al. [808] prepared GaN layers using ZnO as a buffer layer by the pulsed KrF excimer laser deposition technique with a wavelength of 248 nm and pulse duration of 25 ns ablating a ZnO (99.99%) target. The focused beam energy and pulse repetition rate were 2–3 J cm2 and 5–15 Hz (20 ns pulse duration), respectively. The substrate temperature for the deposition of ZnO was in the range of 300–800 C. The background oxygen pressure was in the range of 105 to 102 Torr. In a similar fashion, GaN films have also been deposited on ZnO/sapphire at 850 C under a background NH3 pressure of 106 to 105 Torr. Similarly, but employing a Nd : YAG pulsed laser, with a wave length of 266 nm, a fluency of 0.8 J cm2 per pulse and a repetition rate of 5 Hz, Wang et al. [809] also prepared GaN layers on sapphire using ZnO buffer layers, all deposited at 800 C. Uedo et al. [810] grew thick GaN layers on pulsed laser deposited ZnO buffer layers on sapphire by the HVPE technique. The growth was performed at 1000 C at the V/III ratio of 500 and a rate of 10 mm h1. Detchprohm et al. [375], and Molnar et al. [11] used sputtered deposited ZnO on sapphire as buffer layers for GaN epitaxy. In the former case, a 10 cm diameter ZnO ceramic disk was used as target, and Ar or O2 was employed as sputtering gas with a chamber pressure of 0.13 and 0.4 Torr for Ar and O2, respectively. The discharging input and reflecting powers were 200 and 20 W, respectively, for both gases. The ZnO deposition rates were in the range of 150 and 5 nm h1 for Ar and O2, respectively, and the deposition took place at room temperature. In exploring GaN deposition by the HVPE on ZnO-coated sapphire, Gu et al. [811] observed that the thin ZnO prelayer, which is thermodynamically unstable at the temperatures employed and reducing gases present, disappeared. To quantify this obvious conclusion, two 20 nm thick ZnO samples were deposited on sapphire. One was annealed at 1050 C for 10 min in N2 ambience and the other was not investigated. The unannealed one was dipped into HCl to dissolve the ZnO film. The XPS analysis showed no residues of ZnO or any other compound formed with sapphire. However, the annealed one, after dipping in HCl, showed traces of ZnAl2O4 by XPS analysis, as compared to that of a standard compound such as ZnAl2O4. One can conclude that there might be formation of ZnAl2O4, which may act as nucleation centers for GaN growth, which is not yet corroborated. Hamdani et al. [812] deposited GaN films on ZnO substrates in an MBE environment by reacting Ga and NH3 at 760 C and employing either GaN or AlN as the buffer layer. During the growth, the chamber pressure was maintained at 2–5 · 105 Torr. Matsuoka et al. [813] used degreased and etched ZnO substrates (the etching rate of the O-face ZnO is faster than that of the Zn-face by about one order of magnitude) to grow thin GaN films with OMVPE. Until attaining the required growth temperature, a nitrogen flow was maintained in the chamber, then ammonia and group III sources were allowed into the chamber. A vertical cold walled OMVPE reactor was used for the growth of GaInN and InGaAlN epitaxial layers. As expected, the incorporation of In increased when the substrate temperature was reduced from 800 to 500 C, which is because of the much higher vapor pressure of In compared to that of gallium. A typical ammonia flow rate was 20 l min1. The growth
j599
j 3 Growth and Growth Methods for Nitride Semiconductors
600
temperature was 500–700 C and the V/III flow rate ratio was maintained at 20 000. At 800 C, the flow rate of TEG was kept 17 mmol min1, which resulted in a V/III ratio of 4000. An XRD analysis showed that the FWHM of the rocking curve of the (0 0 0 2) diffraction values are 0.21, 0.20, and 0.52 for GaN/ZnO/Si(1 1 1), GaN/ZnO/ sapphire (0 0 0 1), and ZnO/sapphire, respectively [814]. The FWHM of the (0 0 0 2) reflection for ZnO films was 0.17 for a substrate temperature of 750 C and O2 pressure of 105 to 104 Torr. When ZnO on sapphire was used for GaN growth, the sapphire substrate and the epitaxial layer (ZnO and GaN) are oriented with each other by a 30 rotation of the unit cells. The in-plane epitaxial layer and substrate are in the form of ZnO½1 0 1 0kAl2O3½1 1 2 0, similar to AlN and GaN on sapphire. Wang et al. [809] reported that the FWHM of rocking curve of the (0 0 0 2) diffraction for GaN/ZnO/sapphire decreased from 0.45 to 0.22 with increasing deposition chamber pressure from 105 to 1 Torr. Similarly, this value decreased with increasing substrate temperature. The GaN films deposited at >500, 700, and 800 C under 0.1 Torr in N2 ambience showed amorphous, polycrystalline, and epitaxial nature, respectively. Ueda et al. [810] observed that the RHEED showed ring pattern indicating that a polycrystalline film had formed by a reaction between ammonia and the single-crystal ZnO. The (0 0 0 2) reflection was predominant when the GaN films were grown with ZnO buffer layer. An additional ð1 1 2 1Þ reflection was observed along with (0 0 0 2). In this sample, sharp rocklike three-dimensional structures were observed. The GaN layers grown on the O-face of ZnO exhibited a lack of tilting with respect to the substrate, in contrast to that on sapphire where noticeable tilting is observed. By observing the positions of the (0 0 0 2) diffraction peaks of GaN and ZnO for two X-ray incident beam directions that are 180 apart, one can determine the relative tilt of GaN with respect to ZnO. The data shown in Figure 3.153 clearly show that the direction of the incident beam did not affect the position of the GaN and ZnO diffractions. Consequently, one can argue that the GaN film is not tilted with respect to the ZnO substrate on which it was grown. Optical processes in GaN on ZnO have attracted a good deal of attention. Because ZnO and GaN transitions are close in energy, and there may be overlap in the PL spectra, a short discussion of ZnO PL spectra is warranted. ZnO substrates have improved dramatically, to the point where very sharp X-ray and PL peaks are possible now, with PL spectra showing low FWHM of 0.55 meV for the exciton bound to neutral donor peak (D X) at 3.3597 eV, as shown in Figure 3.154. Other sharp peaks shown in the inset of this figure are related to excited states and excitons bound to different donors. Identification of the exciton structure of ZnO is quite controversial in the literature, but suffice it to say that the main peak at 3.36 eV is repeated three times on the low-energy tail of exciton emission at energies that are multiples of the LO phonon energy (about 71 meV). Another sharp peak was observed at 3.3206 eV with LO phonon replicas at 3.2505 and 3.180 eV, which could be attributed to an exciton bound to acceptor. From defect-related features, only the broad band with the maximum at about 2.4 eV (green band) resolved. The low-temperature PL spectrum
3.5 The Art and Technology of Growth of Nitrides
ZnO (0 0 0 2)
Δα (a)
GaN (0 0 0 2)
Δα (b)
15.6
15.7
15.8
15.9
16.0
Diffraction Angle ( 2θ ) Figure 3.153 X-ray rocking curves for a GaN film grown on the Oface of ZnO with ammonia MBE. The solid and dotted lines associated with the (0002) diffraction peaks of GaN and ZnO are for two X-ray incident beam directions that are 180 apart. Because GaN and ZnO peaks do not show any directional dependence, one can conclude that the relative tilt of GaN with respect to ZnO is negligible.
of the Cermet sample is nearly identical to that of the Hanscom sample. However, the FWHM of 3.36 eV peak is lower by approximately three times that of Hanscom sample. The broad defect related band is observed at about 2.1 eV instead of the green band. In the Eagle Picher sample, the PL spectra from two faces (Zn and O) are compared at room temperature, which demonstrated total identity. Residual strain owing to the mismatch of thermal expansion coefficients of substrates and GaN layers is of special importance as it impacts optical, vibrational, properties, and, to some extent, the electrical properties of GaN. To this end, Hamdani et al. [812] demonstrated the relative benefit of GaN directly grown on ZnO as opposed to that grown on sapphire and SiC. X-ray and optical measurements indicated that GaN layers grown on sapphire substrate undergo a compressive strain and GaN layers grown on SiC substrate are under tensile strain. An accurate calculation of the temperature variation of the thermal strain (eth) owing to the difference in the thermal expansion coefficients between GaN and the substrate used is given by: eth ðTÞ ¼ ½Dal ðTÞ Das ðTÞ=as ;
ð3:66Þ
j601
j 3 Growth and Growth Methods for Nitride Semiconductors
602
10
PL Intensity (au)
10
10
10
5
T = 10 K Bulk ZnO
4
3
2
3.355 10
10
3.36
3.365
1
0
2
2.4
2.8
3.2
3.6
Photon Energy (eV) Figure 3.154 PL spectrum from the Hanscom ZnO sample at 10 K. The inset enlarges fine structure in the near-band edge region.
where Dal(T) and Das(T) are the integrals representing the variation of the lattice parameter in the range of temperature between the growth temperature and the room temperature for the GaN layers and the substrate, respectively. Figure 3.155 shows the variation of the calculated thermal strain eth with growth temperature for different heterostructures, using the temperature dependence of the thermal expansion coefficient presented in Chapter 1. Owing to the close thermal expansion between GaN and ZnO, it is important to note that the thermal strain in GaN/ZnO is about half that of the GaN/SiC and AlN/ZnO heterostructures, established from Figure 3.155. The data show that the cracking of thick GaN, particularly AlGaN, layers observed on SiC substrates could be avoided in the case where ZnO is used as a substrate. Secondly, AlN is not very suitable to be used as buffer layer in the growth on ZnO substrate. We shall mention that the thermal strain eth has negative values for GaN/ZnO and AlN/ZnO indicating a compressive strain and positive values for GaN/SiC and AlN/SiC heterostructures, indicating a tensile strain. This is congruent with the shift of the energy position of free excitons observed in both reflectivity and photoluminescence spectra obtained in some of these heterostructures. It is important to note that the thermal strain increases linearly with increasing growth temperature, indicating that the MBE-grown samples undergo less thermal strain compared to the OMVPE-grown samples, which are grown at higher temperature.
3.5 The Art and Technology of Growth of Nitrides
0.15 GaN/SiC
0.050 0.0 GaN/ZnO
—0.050 —0.10
AlN/ZnO
—0.15 —0.20 200
400
600
800
Compressive strain
Thermal strainε th(%)
0.10
Tensile strain
0.20
1000 1200 1400
Growth Temperature (°C) Figure 3.155 Variation of the thermal strain with the growth temperature calculated for GaN/ZnO, AlN/ZnO and GaN/SiC heterostructures.
3.5.8 Growth on LiGaO2 and LiAlO2 Substrates
As alluded to earlier, in Section 3.2.6.1, the structure of LiGaO2 is similar to the wurtzitic structure, but because Li and Ga have different ionic radii, the crystal has orthorhombic structure. The atomic arrangement in the (0 0 1) face is hexagonal, which promotes the epitaxial growth of GaN(0 0 0 1), so that the epitaxial relationship GaN(0 0 0 1)/LiGaO2(0 0 1) is expected; the in-plane relationship is ½1 1 2 0 GaN|| [0 1 0]LiGaO2. The distance between the nearest cations in LiGaO2 is in the range of 3.133–3.189 Å, whereas the distance between nearest anions is in the range of 3.021–3.251 Å. The lattice mismatch between GaN(0 0 0 1) and LiGaO2(0 0 1) is then only 1–2%. Even though there are difficulties with surface preparation and stability of the material in the presence of H and high temperatures, not to mention the poor thermal conductivity, efforts have been expended to take advantage of this material for substrates. Promising results from material growths on lithium gallate (LGO) including aluminum, gallium, and indium nitride alloys by MBE [815] as well as OMVPE [816] growth have been reported. Yun et al. [817] examined the in-plane atomic arrangement of GaN with respect to LGO. To do this, two-dimensional c–j scans with 2y fixed at the ð1 0 1 2Þ orientation of GaN; the resultant pole figure is shown in Figure 3.156. From the contour plot of c–j space, six intensity maxima corresponding to GaN asymmetric ð1 0 1 2Þ
j603
j 3 Growth and Growth Methods for Nitride Semiconductors
604
Figure 3.156 XRD pole figure of GaN epilayer grown on LiGaO2 substrate, with 2y fixed at GaN ð1 0 1 2Þ orientation. Six intensity maxima are from GaN (1 0 2) diffraction. No rotation relative to the substrate is observed. The random patterns are from the background.
orientation are observed at 60 equiangularly. No lattice rotation within the equatorial plane relative to the LGO substrate is observed. This is significant in that it shows close lattice matching (Da/|a| < 1%) and may pave the way for high-quality epitaxial c-plane GaN films. Structural qualities of the films were evaluated by performing o-scan (rocking curve) in both symmetric (0 0 0 2) and the asymmetric ð1 0 1 2Þ directions. The FWHMs of 4.5 and 21.2 arcmin have been obtained on GaN/LGO for (0 0 0 2) and ð1 0 1 2Þ, respectively, in approximately 1 mm thick films [817]. The photoluminescence measurements of GaN on LGO showed a sizeable shallow donor–acceptor pair band in the range 3.1–3.3 eV in addition to the band edge emission. In the case of MBE, second-order reconstructions of the GaN surface on LGO, indicating smooth and well-ordered Ga-polarity surfaces have been reported. From a structural point of view, the X-ray reciprocal space mapping indicated higher structural quality than GaN on SiC and sapphire available in that particular laboratory, whereas TEM data indicate 6 · 108 cm2 threading dislocation density [815]. However, GaN on sapphire and SiC, in general, prepared in other laboratories are superior to this particular report of GaN on LGO. By improving the surface finish of LGO by
3.5 The Art and Technology of Growth of Nitrides
polishing, the authors noted linear, as opposed to spiral, step-flow growth of AlGaN/ GaN during MBE growth. In this mode of growth, monolayer terraces are observed through AFM with 30 nm of Al0.25Ga0.75N on a 2.4 mm thick GaN layer, in contrast to the spiral step-flow growth observed previously for MBE-grown GaN samples on OMVPE buffer layers [818]. Additionally, the dislocation density derived from the small pits in the AFM image result in a threading dislocation density of 4–5 · 108 cm3, which is in close agreement with the TEM data presented above. The FWHM of o 2y X-ray diffraction rocking curves as low as 85 arcsec for a 1 mm thick film were measured with the h0 0 0 4i reflection. However, one must be careful of this reflection as it is sensitive mostly to screw dislocations, which are miniscule in density compared to edge dislocations. Asymmetric reflections, such as the [102] reflection, would be a better gauge of the quality of the film. With OMVPE growth, it was reported that the films grown at 1000 C peeled off as soon as they were in contact with water vapor in the atmosphere. Moreover, hydrogen attacks LiGaO2, thus nitrogen must be used as a carrier gas in OMVPE growth of GaN to achieve high structural, high-quality GaN on LiGaO2. With the controlled low growth temperature (850 C) and extremely low input partial pressures of hydrogen, Duan et al. [819] successfully deposited GaN on both domains of LiGaO2 by OMVPE without the problem of peeling off. Fischer and coworkers [820] used LiAlO2 substrates with 1.5% lattice mismatch and 21% mismatch in the thermal expansion coefficients, respectively. As mentioned previously, in conjunction with substrate issues, the epitaxy relationships between GaN and LiAlO2 are expected to be ð0 1 1 0Þ=ð1 0 0Þ LiAlO2 with ð2 1 1 0Þ==ð0 0 1Þ LiAlO2. Unlike Al2O3 and 6H-SiC substrates with very smooth surfaces, the LiAlO2 substrate exhibited a wavelike surface with equidistant grooves about 10 nm deep, which could have originated from the mechanical surface polishing. Because of the high surface roughness, the interface between GaN and LiAlO2 was highly defective. Three-dimensional growth by OMVPE of monocrystalline GaN thin films on bLiGaO2 substrates was realized by Kung et al. [816]. The GaN layers were grown by OMVPE at temperatures between 600 and 1000 C, at deposition rates of about 0.7 mm h1. The samples had smooth surfaces and the X-ray rocking curve was as narrow as 300 arcsec for substrate temperatures of 900 C. All GaN layers were n-type (n about 1020 cm3) as determined by room-temperature Hall measurements. The electron mobility was about 10 cm2 V1 s1. The high electron concentrations may be because of the incorporation of oxygen, resulting from the decomposition of LiGaO2 at elevated substrate temperatures. This effect would probably be reduced by the use of low-temperature buffer layers. 3.5.9 Growth on GaN Templates
In the absence of large area GaN substrates in the conventional sense, the basic research of the homoepitaxial GaN growth has been on small templates of GaN platelets grown from the liquid phase under high hydrostatic pressure and at high temperatures. Growth was performed both by OMVPE [821–823] and MBE [824,825]
j605
j 3 Growth and Growth Methods for Nitride Semiconductors
606
on the aforementioned GaN platelets. In addition, GaN epitaxial layers have been grown by MBE on OMVPE [317] and HVPE [826] prepared templates and by OMVPE on HVPE-prepared templates [827]. More on this topic will be given shortly. Such studies serve to establish benchmark values for the optoelectronic properties of thin GaN films on native substrates. Both p- and n-type dopings were achieved, and a p–n junction was realized by OMVPE on GaN crystals grown from Ga solution. The deposition temperature was 1000–1050 C. The width of rocking curves for epitaxial layers was practically the same as that of the substrate. The photoluminescence spectra were dominated by exciton emission at low temperatures, pointing to high quality of the epitaxial layers. There are, in general, two possible approaches for suitable substrates for homoepitaxial growth: The desirable one is to grow a large bulk crystal, cut it, and polish the slices. This method is widely used for conventional semiconductors. It cannot be applied easily to GaN bulk growth, however, because only small pieces of GaN can be grown only at high temperatures and very high pressures (tens of kbar), taking into account that GaN begins to decompose at 800 C. The second possibility that has been explored is the growth of thick GaN films on foreign substrates (sapphire, SiC, or Si). It is well known that the quality of the heteroepitaxial film improves as the thickness increases. Very popular epitaxial deposition methods, such as OMVPE and MBE, have very slow growth rates, a few microns per hour at best, and cannot be used for the growth of thick films. As discussed earlier, the inorganic CVD method has high growth rates (up to 100 mm h1 or larger). Thus, it is argued that the best way to grow homoepitaxial GaN film is to use a two-stage growth, a thick GaN substrate grown by inorganic CVD in 1–2 h followed by the device layer grown by OMVPE or MBE. Early efforts of homoepitaxy relied on growth by MBE on buffer layers grown by OMVPE or HVPE. Owing to smooth surfaces obtained by these two vapor techniques and avoidance of complications brought about by heteroepitaxy, atomically smooth layers could be grown by MBE. This permitted delineation of regimes leading to smooth and rough surfaces, namely Ga-rich and N-rich growth conditions. We should mention that prior to GaN, growth by MBE was always conducted under slightly group V rich conditions. As mentioned previously, homoepitaxy also includes growth on GaN templates or on epitaxial layers grown by HVPE or OMVPE. In those cases, in addition to misfit dislocations, other threading defects originating at the GaN substrate interface, such as edge dislocations, screw dislocations, and mixed dislocations, propagate to the surface unless they run into one another and loop, and their Burgers vectors cancel leading to dislocation looping [828]. GaN and related heterostructures have been grown by MBE on GaN/sapphire templates prepared by OMVPE [318]. In this particular work, the samples grown under nitrogen-rich conditions are not conductive, whereas the others are and have room temperature mobilities approaching 1200 cm2 V1 s1. The mobility increases as the Ga flux increases and turns over above a certain group III/V ratio where the Ga droplets begin to form. The regions where the Ga droplets form are of lower quality and degrade the overall mobility measured. The optimum properties are reached
3.5 The Art and Technology of Growth of Nitrides
when this Ga-adlayer coverage is maximized without the formation of Ga droplets (i.e., at the highest Ga flux within the intermediate regime). Extending GaN growth to heterostructures, the AlGaN/GaN 2DEG system has been prepared on GaN templates. Samples over a broad range of electron densities, ranging from ns ¼ 6.9 · 1011 to 1.1 · 1013 cm2 were grown with the best mobility of 53 300 cm2 V1 s1 at a density of 2.8 · 1012 cm2, and temperature T ¼ 4.2 K [829]. Magnetotransport studies on these samples display exceptionally clean signatures of the quantum Hall effect [830–832]. The investigation of the dependence of the 2DEG mobility on the carrier concentration suggests that the low-temperature mobility in these AlGaN/GaN heterostructures is currently limited by the interplay between the charged dislocation scattering and the interface roughness. The typical MBE overgrowth consisted of a 0.5 mm undoped GaN buffer layer capped by approximately 30 nm of Al0.09Ga0.91N. Changing the thickness of the Al0.09Ga0.91N caused a variation of the sheet carrier concentration (formed by electrons released from defect centers on or near the surface of Al0.09Ga0.91N), which is the screening charge in response to polarization charges present at the interface. Recently, these mobility figures were extended to about 73 000 cm2 V1 s1 and beyond by MBE growth on HVPE GaN layers using a RF N source and also on GaN bulk platelets using ammonia as the reactive nitrogen source. A nominally undoped 1.5 mm layer was recently grown by MBE on the Ga-face of a freestanding GaN template [833]. The template, in turn, was grown by HVPE on a cplane sapphire substrate and separated from the substrate by laser liftoff. Before the overgrowth, the GaN template was mechanically polished, dry etched, and finally etched in molten KOH. A 30 nm thick undoped AlxGa1xN cap layer with x ¼ 12% has been deposited on top of the 1.5 mm GaN layer. Part of the 10 mm · 10 mm surface of the sample was mechanically shuttered during the MBE process, so that both the GaN substrate and the MBE-overgrown layer could be studied under the same experimental conditions. Steady-state PL was excited with a He–Cd laser (325 nm), dispersed by a 1200 rules per millimeter grating in a 0.5 m monochromator and detected by a photomultiplier tube. The best resolution of the PL setup was about 0.2 meV, the photon energy was calibrated with a mercury lamp accounting for the refraction index of air (1.0003). The temperature was varied from 15 to 300 K in the closed cycle cryostat. The excitation density in the range of 104 to 100 W cm2 was obtained by using unfocused (2 mm diameter) and focused (0.1 mm diameter) laser beam attenuated with neutral density filters. Structural properties of the GaN template and overgrown MBE epilayer were characterized by high-resolution X-ray diffraction and AFM. The FWHM of the oscan (rocking curve) for the [0 0 2], [1 0 2], and [1 0 4] directions are 53, 145, and 54 arcsec, respectively, for the MBE-grown area and 52, 137, and 42 arcsec, respectively, for the GaN substrate. The similar characteristics of the substrate alone and the substrate with MBE overgrowth indicate similar crystal quality of GaN both in the substrate and overgrown layer. A comprehensive characterization of the quality of the GaN templates has been reported elsewhere [834] and will not be repeated here. Suffice it to say that the template quality is unmatched in terms of transport and
j607
j 3 Growth and Growth Methods for Nitride Semiconductors
608
optical properties, and extended and point defects. The surfaces of both substrate and MBE-overgrown layer are very smooth, with the root mean square roughness of about 0.2 nm in the 2 mm · 2 mm AFM images. The homoepitaxial growth of GaN has recently been achieved by using GaN(0 0 0 1) bulk substrate [835,836]. No buffer layer was used for the growth. The GaN bulk substrate, which was used for the growth, was cleaned first by boiling in aqua regia for 10 min and then in organic liquid with an ultrasonic cleaner prior to the growth. A horizontal OMVPE reactor was employed at atmospheric pressure to grow the GaN layer. TMG and ammonia were the source gases and H2 was the carrier gas. For the growth of Mg-doped GaN, bis-cyclopentadienyl magnesium was utilized. To protect the GaN substrate from the escape of nitrogen at high temperatures (>800 C), ammonia was fed into the reactor before the beginning of the substrate heating. The as-grown n-GaN films thus obtained were found to be high-quality single-crystal films with good surface morphology. The as-grown GaN : Mg films showed strong blue cathodoluminescence and photoluminescence spectra peaking at 445 nm even without LEEBI treatment or thermal annealing. GaN homoepitaxial layers grown by OMVPE on the highly conductive GaN bulk crystals grown at high hydrostatic pressure exhibit smaller free-electron concentrations in the layers in contrast to the substrates, which had about 2.5 · 1019 cm3 of free electrons [837]. In X-ray diffraction [0 0 0 2] peaks for the substrate and epitaxial layers had rocking curves with half widths of about 20 arcsec. The photoluminescence spectrum exhibited by the epitaxial layer was less than 1 meV wide. GaN layers grown by OMVPE on freestanding GaN templates grown by HVPE also showed excellent properties. X-ray reflections, with a FWHM of as low as 20 arcsec, were obtained. The dislocation density was determined to be 2 · 107 cm2. The lattice mismatch between the GaN substrate and the homoepitaxial layer was below 3 · 105 and the PL linewidth was about 0.5 meV [838]. Much more refined experiments followed these early attempts, which resulted in GaN with excellent properties as determined by photoluminescence [839–841]. One could realize unstrained GaN layers with dislocation densities comparable with that in the template on which the layers are grown. In the case of platelets prepared by the high-pressure technique, this dislocation density is several orders of magnitude lower than the best conventional heteroepitaxy. Through the use of dry etching techniques for surface preparation, a pathway was blazed to achieve high crystal quality in the overgrown epitaxial film [839]. The layers so grown reveal an exceptional optical quality as determined by a reduction of the low-temperature PL linewidth from 5 to 0.1 meV and a reduced symmetric XRD diffraction rocking curve width from 400 to 20 arcsec. The latter is not surprising as the epitaxial layer replicates the structural properties of the template unless the surface preparation is not optimum. Narrow PL linewidths paved the way for observing fine structure of the donor-bound exciton line at 3.471 eV with five fine features inclusive of the excited states of free excitons. Additionally, all three free excitons as well as their excited states were visible in the photoluminescence spectrum at 2 K. A PL scan taken at 2 K is shown in Figure 3.157. Moreover, InGaN/GaN multiple quantum well (MQW) structures as well as GaN p–n and InGaN/GaN double heterostructure LEDs on GaN bulk
3.5 The Art and Technology of Growth of Nitrides
n=1
8. (Do,X An = 1) 9. (D o,X C
PL Intensity (au)
104
o
3. (Ao,X An = 1)
n=1
4. (D ,X A
n=1
o
n=1
5. (D ,X B
n=1
) 6. (XA
n=1
) 11. (X C
) 10. (X A 4
)
) 12. (XBn = 1) 13. (X Cn = 1)
o
(A ,X An = 1): 0.09 meV
PL and PR @2K 3
4
3 6
103
5 2
1 2
7
2
10
8
9
10 11 12
3.44
Reflectance (au)
o
n=1 ) 2e1. (D o,XAn = 1)2e- 2. (D ,X B
3.46
3.48
3.50
13
1 3.52
Energy (eV) Figure 3.157 Photoluminescence and reflectance scans of a homoepitaxial GaN layer grown on GaN platelet grown by the high-pressure technique. Courtesy of M. Kamp.
single-crystal substrates have also been prepared [839]. These particular LEDs were reported to be twice as bright as their counterparts grown directly on sapphire. In addition, they exhibited improved high power characteristics, which are attributed to enhanced crystal quality and increased p-type doping. Although very preliminary, high-quality GaN layers have also been grown on freestanding templates, which, in turn, are prepared by HVPE. The PL spectrum of an undoped GaN epilayer grown by OMVPE on such a template is shown in Figure 3.158 for a wide range of photon energies. Besides exciton-related peaks, which will be considered later, the spectrum contains two broad bands: a YL band with a maximum at about 2.3 eV and a blue luminescence (BL) band with a maximum at about 3 eV. The YL band is the omnipresent feature in PL spectra of n-type GaN grown by different techniques and it is most commonly attributed to a structural defect, namely a complex of the gallium vacancy with oxygen or silicon atom [842]. The BL band is most probably related to the surface states of GaN [843]. Unlike the BL often observed in HVPE- and OMVPE-grown GaN [844], the BL in the sample strongly bleached with the laser exposure time, very similar to the behavior observed earlier [843]. The bleaching is attributed to photo-assisted desorption of oxygen atoms from the GaN surface. Note that with increasing excitation intensity, the relative contribution of the YL and BL bands decreases because of the saturation of the corresponding defects with holes, and at highest excitation density, these bands can be barely detected in the spectrum (intensity is five orders of magnitude weaker than the peak intensity of the main exciton peak).
j609 n=1
) 7. (X B
)
j 3 Growth and Growth Methods for Nitride Semiconductors
610
10
9
T = 15 K
Sample
PL Intensity (au)
410 H01 10
8
10
7
10
6
3.44
3.45
3.46
3.47
3.48
3.49
3.5
Photon energy (eV) Figure 3.158 Excitonic spectrum of the undoped epilayer (sample H01) and freestanding substrate (sample 410). Excitation density is 100 W cm2 (focused laser beam).
Excitonic spectrum of the undoped GaN epilayer is very similar to the spectrum of the freestanding GaN template, except some difference in relative intensities of some peaks. The peak at 3.4673 eV is attributed to the A-exciton bound to unidentified shallow acceptor (A0 ; X An¼1 ) [845]. The most intense peaks at 3.4720 and 3.4728 eV are attributed to the A-exciton bound to two neutral shallow donors, D1 and D2: (D01 ; X An¼1 ) and (D02 ; X An¼1 ). The FWHM of these peaks is 0.6 meV at 15 K and their positions are the same with an accuracy of 0.2 meV for the substrate and epilayer. The peaks at 3.4758 and 3.4766 eV, having the same intensity ratio and energy separation 0 n¼1 as the (D01 ; X n¼1 A ) and (D2 ; X A ) peaks, are attributed to the B-exciton bound to the D1 0 n¼1 and D2 donors: (D1 ; X B ) and (D02 ; X Bn¼1 ), in agreement with Refs [845,846]. Two other peaks, also related to donor-bound excitons (DBE), are observed at 3.4475 and 3.4512 eV and attributed to the so-called two-electron transitions: 0 n¼1 ðD01 ; X n¼1 A Þ2e and ðD2 ; X A Þ2e . This type of the DBE recombination, first observed in GaP [847], involves radiative recombination of one electron with a hole leaving the neutral donor with second electron in an excited state. In the effective-mass approximation, the donor excitation energy from the ground to the n ¼ 2 state is three fourth the donor-binding energy ED. Consequently, from the energy separation between the principal DBE line and the associated two-electron satellite, we can find the binding energies of two shallow donors: E D1 ¼ 4=3 24:5 ¼ 32:6 meV and E D2 ¼ 4=3 21:6 ¼ 28:8 meV. It appears reasonable that a weak peak at 3.443 eV is the two-electron transition related to the second excited state (n ¼ 3) of the D1 donor because its separation from the (D01 ; X An¼1 ) peak is 8/9 32.6 ¼ 29.0 meV. Free excitons (FE) related to the A and B valence bands (X An¼1 and X Bn¼1 ) are identified at about 3.479 and 3.484 eV, respectively (the X Bn¼1 peak is seen as a shoulder). From the separation between the FE and DBE peaks, we find the binding
3.5 The Art and Technology of Growth of Nitrides
energies of the DBEs related to the D1 and D2 donors as 7.0 and 6.2 meV, respectively. According to empirical Haynes rule, the binding energy of the DBE is proportional to the binding energy of the corresponding donor. The proportionality constant (a) for the D1 and D2 donors is found to be 0.215, which is close to the result of Meyer (a ¼ 0.2 0.01) [846]. A peak at 3.4983 eV is attributed to the n ¼ 2 excited state of the A exciton (X An¼2 ), in agreement with Refs [848,849]. From the energy positions of the X An¼1 and X An¼2 peaks (3.479 and 3.4983 eV, respectively), we find the A-exciton binding energy in the hydrogen model as 25.7 meV and the bandgap Eg ¼ 3.5047 eV. In short, optical studies of GaN grown on GaN templates, both by MBE and OMVPE, have been conducted. Submillielectron volt luminescence linewidths allowed the delineation of many excitonic transitions involving intrinsic transitions and their excited states and excitons bound to impurities such as donors with two electron transitions. Consequently, exciton-binding energies as well donor-binding energies could be determined accurately. The details can be found in Volume 2, Chapter 5. 3.5.10 Growth on Spinel (MgAl2O4)
Cubic MgA12O4 has a spinel-type structure (Fd3m) with the oxygen atoms forming a face centered cubic sublattice and Mg and Al atoms occupying the tetrahedral and octahedral sites, respectively. Lattice mismatch with GaN, Dd/d is 10%. The crystals are stable at the GaN growth temperature. The spinel substrate has an advantage over the sapphire substrate for obtaining mirror laser facets by cleaving [90,850,851]. GaN crystals were grown on (1 0 0) and (1 1 1) oriented MgA12O4 substrates by OMVPE. As in the case of sapphire substrates, low-temperature buffer layers are employed followed by a few micrometers thick GaN films grown at about 1000 C. As expected, GaN films grown on (1 1 1) substrates are wurtzitic single crystals. The crystallinity of the films on spinel is not comparable to that on sapphire. The only attractive feature of spinel is the cleaved facet prospect, which with the recent advances on sapphire and SiC is becoming a moot point. 3.5.11 Growth on Non c-Plane Substrates
The layers and heterostructures on polar c-plane orientation are associated with polarization-induced charge, owing to strain and in the case of heterostructures to compositional gradient, which, for example, causes electric field to be present in quantum wells and barriers and consequently leads to a quantum confined stark effect (QCSE) [852]. This also manifests itself as a red shift in emission energy such as in LEDs, the degree of which depends on the injection level with higher injection leading to a blue shift (attempting to negate the red shift) [853] (see Figure 3.159). Although this polarization has been used to induce 2DEG at the interfaces such as that between AlGaN and GaN for field-effect transistors without introducing intentional dopants, the same field causes a spatial separation of electrons and holes in quantum wells of LED, thereby increasing the radiative lifetime owing to spatial
j611
j 3 Growth and Growth Methods for Nitride Semiconductors
612
4.0
(a)
3.8
3.6 Energy (eV)
3.6 Energy (eV)
(b)
3.7
3.35 eV 0.4
3.5
3.49 eV
0.2 0.1
0.2
0.0 0.0 0
5
10 15
20 25
Thickness (nm)
0
5
10 15
20 25
Thickness (nm)
Figure 3.159 Calculated band profiles in (5 nm GaN)/(10 nm Al0.1Ga0.9N) quantum wells by self-consistent effective mass Schr€ odinger–Poisson calculations. (a) The very large electrostatic fields in the [0 0 0 1] orientation result in a quantum confined Stark effect and poor electron-hole overlap. (b) The ½1 1 0 0 orientation is free of electrostatic fields, thus true flat-band conditions are established. Patterned after Ref. [853].
separation of carriers [854] and reducing the quantum efficiency [855]. The increased carrier lifetime and reduced recombination efficiency are more noticeable at low injection levels. At high injection levels, screening reduces the field and thus increases the wave function overlap between the electron and hole states. Moreover, the existence of spontaneous and piezoelectric polarization effects can induce large densities of free electrons or in principle holes (has not been observed irrefutably) at nitrides interface, which can sometimes be much larger than that induced by intentional doping. This can therefore reduce our ability to control the free carrier concentration at the interface. Tocircumventthisproblem,onecangrowcubicratherthanhexagonalGaN,whichis nonpolar along the cubic [0 0 1] direction, and therefore can avoid strong polarizationinduced electric field in heterointerfaces. Unfortunately, cubic GaN is metastable and also very difficult to achieve with quality comparable with that obtained for the wurtzite phase. Seemingly, more a attractive approach is to explore the a- or m-plane hexagonal GaN structures rather than c-plane GaN. The GaN layers with nonpolar orientation are termed nonpolar GaN, because the c-axis is parallel to the substrate surface and there would then be no polarization-induced electric field at nitrides interfaces. Figure 3.160 shows schematically the m-, a-, and c-plane of GaN. These planes are perpendicular to each other. However, as always, the case 1 problem is traded with another. Although growth GaN on SiC follows the substrate orientation (e.g., a-plane GaN results on aplaneSiCandm-planeGaNresultsonm-planeSiC),thesameisnottrueforsapphire.As discussed early on in this chapter, GaN on a-plane sapphire by MBE and OMVPE is low quality c-plane. Toobtain a-plane GaNonsapphire r-plane sapphire isused. Inaddition, growthonm-planesapphiredoesnotleadtoc-directiontobealongthesubstratesurface
3.5 The Art and Technology of Growth of Nitrides
Figure 3.160 Schematic representation of m-, a-, and c-planes of GaN. (Please find a color version of this figure on the color tables.)
plane. Instead, it is off the plane, leading to semipolar as opposed to nonpolar surfaces. Thegenesisofthesepeculiaritiesonsapphireisnotwellknownandrequiresagooddeal of investigation. Degrading device performance, however, is that the stacking fault formation energy in a-plane GaN is very low leading to high concentration of stacking faults that are radiative recombination killers. In addition, In incorporation of InN into the GaN lattice is difficult. Further, the LEDs fabrication on this plane provides optical powers in the submilliwatt range as opposed to 20 mW range obtained for c-plane. InN incorporation in m-plane GaN is much easier and LED power levels close to 1 mW are attainedinthedevelopmentandproductionlaboratories[856]andtensofmicrowattsin journal publications [857]. However, the forward turn on voltage is high (which may havetodowiththelowp-typedoping)andsurfaceisrelativelyrough,2 nm,asopposedto atomically smooth on c-plane, although turn-on voltages of about 3.2 V have been reported [857]. To improve the surface morphology, tilted substrates are used. The researchisitsearlystagesandmoredevelopmentsareneededforadefinitivestatement. 3.5.11.1 The a-Plane GaN Growth Usually ð1 1 2 0Þa-plane GaN is grown on ð1 1 0 2Þr-plane sapphire, a-plane SiC or aplane g-LiAlO2. Before the c-plane GaN gained the dominant status in GaN research, some early research has been devoted to the growth of a-plane GaN using OMVPE [858–860] and MBE [861–864]. In fact, growth of GaN by MBE on c-, r-, a-, and m-plane sapphire substrates has been explored in a systematic study that included the investigation of A1 and E1 phonon modes by Raman scattering and excitonic transitions by PL and photoreflectance on all the aforementioned surfaces
j613
j 3 Growth and Growth Methods for Nitride Semiconductors
614
[865]. But the crystalline quality was not so good on non-c-plane sapphire so the potential for device fabrication could not be realized. Owing to the maturity and to a great extent the built-in polarization issues of c-plane GaN, the growth of nonpolar GaN, that is, a-plane or m-plane GaN, has gained considerable interest. The epitaxial relationship of ð1 1 2 0Þa-plane GaN on ð1 1 0 2Þr-plane sapphire is ½1 1 2 0 GaNjj½1 1 0 2 sapphire, ½0 0 0 1 GaNjj½ 1 1 0 1 sapphire, and ½ 1 1 0 0 GaNjj ½1 1 2 0 sapphire [866] (refer to Figure 3.161). Craven et al. [866] have grown a-plane GaN on r-plane sapphire using OMVPE. Their cross-sectional TEM (see Figure 3.162a) results show that there was a large density of threading dislocations (TDs) originating at the sapphire/GaN interface with line directions parallel to the growth direction ½1 1 2 0. The TD density determined by plan view TEM (see Figure 3.162b) was 2.6 1010 cm 2. Pure screw dislocations will have Burgers vectors aligned along the growth direction (b ¼ ½1 1 2 0), whereas pure edge dislocations will have b ¼ ½0 0 0 1. In plan view TEM, stacking faults have also been found, with a direction parallel to [0 0 0 1]. In SEM images (see Figure 3.163), after full coalescence, we can observe surface striations that are uniformly aligned parallel to the GaN [0 0 0 1] direction [867]. The growth in c-direction is larger than in the ð1 1 0 0Þ direction, accounting for the elongated features to be aligned in the [0 0 0 1] direction. Striations may also point to some stacking faults. Omega rocking curves were measured for GaN on-axis ð1 1 2 0Þ and off-axis ð1 0 1 1Þ planes showing anisotropy with FWHM values of 0.29 –0.52 and 0.29 –0.43 ,
0
1]
Sapphire [0 0 0 1]
N
[0
0
GaN [0 0 0 1]
Ga
Sapphire [0 1 1 1]
c/2
GaN [1 1 0 0] Sapphire [2 1 1 0]
Figure 3.161 Epitaxial relationship between a-plane GaN and rplane sapphire. The hexagon lying on the r-plane of sapphire depict the GaN crystal. Note the a-plane is formed on r-plane sapphire with huge lattice distortions.
3.5 The Art and Technology of Growth of Nitrides
Figure 3.162 Cross-sectional TEM image (a) and plan view (b) TEM image of a-plane GaN on r-plane sapphire. The TDs in image (a) have a common line direction parallel to ½1 1 2 0. The stacking faults in image (b) have a common faulting plane parallel to the (0 0 01). Courtesy of J.S. Speck. Ref. [866].
respectively. Compared to c-plane GaN, the crystalline quality for a-plane GaN is not very good and needs to be optimized further for devices purpose. Several research groups have grown a-plane AlGaN/GaN MQWs on r-plane sapphire and made comparative studies with that of c-plane structures, showing the absence of polarization-induced electric filed in nonpolar GaN [868]. Ng [869] has grown GaN/AlGaN MQWs on r-plane sapphire using MBE technique. Roomtemperature photoluminescence results show that the photoluminescence intensity was 20–30 times higher for the ð1 1 2 0Þ MQWs compared to the (0 0 0 1) MQWs. This particularly holds at low injection levels as high injection levels screen the polarization-induced charge, the details of which are discussed in Volume 2,
Figure 3.163 SEM surface morphological evolution during early stage of a-plane GaN growth: (a) LT GaN buffer, (b) 20 nm GaN, (c) 50 nm GaN, (d) 100 nm GaN, (e) 300 nm GaN, (f) 1.5 mm fully coalesced GaN. Note that c direction is the stripe direction shown in (f), and all samples here have their c directions parallel to each other.
j615
j 3 Growth and Growth Methods for Nitride Semiconductors
616
Chapter 5. The fact that the peak transition energy, as a function of well width for the ð1 1 2 0Þ MQWs, followed the trend for rectangular potential profiles indicated the absence of built-in electrostatic fields. Craven et al. [870] confirmed the advantage of nonpolar GaN. The PL emission energy from a-plane MQWs followed a square well trend modeled using self-consistent Poisson–Schr€ odinger calculations, whereas the c-plane MQW emission showed a significant red shift with increasing well width, which is attributed to the quantum-confined Stark effect. Despite a higher dislocation density, the a-plane MQWs exhibit enhanced recombination efficiency as compared to the c-plane wells because well emission is no longer observed for c-plane wells wider than 50 Å. Because p-type doping (Mg coefficient) may be sensitive to polarity, p-type doping of a-plane GaN has been carried out for the first time by Armitage et al. [871]. The electron mobility in a-plane GaN : Si was only 18 cm2 V1 s1, compared to 200 cm2 V1 s1 for c-plane epilayers of similar Si doping level. On the contrary, the maximum conductivity of p-type a-plane GaN : Mg was as high as was typically found for c-plane GaN:Mg. A maximum hole concentration of 6 · 1017 cm3 was found, with mobility of 2 cm2 V1 s1. A comparison suggested that the Mg sticking coefficient may be higher in ð1 1 2 0Þ GaN than in ð0 0 0 1Þ GaN. So the control of Mg doping will be easier in a-plane than in c-plane epilayers. Their result basically proved that sufficiently high p- and n-type doping can be achieved for applications in light-emitting diodes. Chakraborty et al. [872] also reported their research about ptype doping of a-plane GaN using OMVPE technique. Their maximum hole concentration achieved was 6.8 · 1017 cm3. They also found that higher Mg incorporation could be achieved with increasing growth rate, higher V/III ratio, and lower growth temperature. In-plane anisotropy is an important characteristic for a-plane GaN when compared to c-plane GaN. In the surface morphology of a-plane GaN, several groups [873,874] have observed [0 0 0 1] oriented stripes features on the surface of a-plane GaN and discussed the anisotropic properties of GaN. Wang et al. [873] attributed the stripe features to the unequal growth rates along ½1 1 0 0 and [0 0 0 1] directions. The overall growth rate in [0 0 0 1] direction was higher than that in ½1 1 0 0 direction, resulting in a growth front with stripe features. The FWHM values of XRD o-scan were different with different azimuth angle, which is defined with respect to the [0 0 0 1] direction (see Figure 3.164). Li et al. [874] observed in-plane anisotropy of electronic property of a-plane GaN grown by OMVPE, in addition to the crystallographic anisotropy. The electron mobilities in the [0 0 0 1] and ½1 1 0 0 directions were 10 and 6 cm2 V 1 s 1, 18 respectively, when doped to 1.0 10 cm 3. 3.5.11.2 Epitaxial Lateral Overgrowth of a-plane GaN Although a variety of techniques have been demonstrated to grow a-plane GaN, the crystalline quality should be improved further for the fabrication of optoelectronic and other devices. ELO is a good technique in this direction. The lateral epitaxy on aplane GaN is interesting in that windows in the mask material are made parallel to ½1 1 0 0 and the growth emanating in the window regions extends laterally along the relatively fast growth [0 0 0 1] direction, representing the Ga-polar growth front or
3.5 The Art and Technology of Growth of Nitrides
Template
Intensity (a.u.)
5
(a)
Azimuth 0o
4 30o 3
60o
2
90o
1 0 -1.0
-0.5
0.0
0.5
1.0
Δω (o) 0.6
Template
(b)
FWHM (o)
0.5
0.4
0.3
0.2 0
300 100 200 o Azimuth angle ( )
400
Figure 3.164 XRD ð1 1 2 0Þ Omega scan of a-plane GaN with different azimuth angles (a), and the FWHM values of ð1 1 2 0Þ Omega scan as a function of azimuth angles (b). Courtesy of M.A. Khan and Ref. [873].
½0 0 0 1 direction, representing the N-polar growth front. These two fronts advance toward one another and meet somewhere on the masked region forming the coalescence front, as shown in Figure 3.165a–c. Also to be noted in the figure is that Ga-front advances a factor of 3 faster than the N-front leading to shift in the coalescence point from the center line of the masked region. A scan of the yellow optical emission band across the wafer by near field scanning optical microcopy (NSOM) in a 20 mm 20 mm area is also consistent with the topographic images obtained by AFM in that the N-face front grows at about one third the rate as compared to the Ga-face front (Figure 3.165c). Note the window region in the NSOM image is dark and the dark line running parallel to the stripes in the wing region corresponds to the coalescence front.
j617
j 3 Growth and Growth Methods for Nitride Semiconductors
618
Figure 3.165 Cartoon of epitaxial lateral overgrowth (ELO) on a-plane GaN with growth along the [0 0 0 1] direction, representing the Gapolar growth front and along the ½0 0 0 1 direction, representing the N-polar growth front: (a) top view and (b) side view. The growth along the Ga-polar front is about a factor of g ¼ 3 times faster. Courtesy of VCU students Vishal Kasliwal
and Xianfeng Ni. (c) Left 30 mm · 30 mm AFM image for sample B. Right 4 mm · 4 mm AFM image near the window and N-polar wing boundary of sample B, showing different surface pit densities for the window and the wing. Courtesy of VCU students Vishal Kasliwal and Xianfeng Ni. (Please find a color version of this figure on the color tables.)
3.5 The Art and Technology of Growth of Nitrides
As in the case of c-plane GaN ELO, wing tilt is observed that can be reduced by adopting a two-stage growth where lower substrate temperature is employed for the first stage with relatively low lateral-to-vertical growth rate ratio [875]. This is followed by increased temperature growth that enhances the lateral growth rate over the vertical one. Because the gap to the bridged laterally is smaller, the height difference between the advancing Ga- and N-fronts at the coalescence boundary is smaller as shown in Figure 3.166. The wing tilt is also influenced by any miscut of the r-plane sapphire and deviations of the template surface from the exact a-plane GaN template. XRD rocking curves have been obtained with three different f angles, where f is the angle of rotation about the sample surface normal and is defined as 0 when the projection of incident X-ray beam is parallel to the SiO2 mask stripes, top determine the wing tilts. As shown in Figure 3.167a and b, for f ¼ 0 , only one diffraction peak from the a-plane of GaN can be observed for both samples, with a FWHM of 0.40 and 0.19 for samples A (single-step growth) and B (two-step growth), respectively. For f ¼ 90 (the projection of incident X-ray beam is normal to the mask stripes), however, samples A and B exhibit two and three peaks, respectively, the order of which is reversed for f ¼ 270 . The strongest peak observed for sample A is that from the Ga-polar wings, since the area of the Ga-polar wing is much larger than that of the N-polar wing and the window. The observed tilt angle of 0.86 is much larger than 0.25 obtained from the large angle convergent beam electron diffraction (LACBED), which may be attributed to the local nature of LACBED. The peak with the strongest intensity, shown in
Figure 3.166 Schematics of the height difference between two neighboring wings on a-plane GaN (not in exact proportion), with Ga- to N-polar wing width ratios of (a) 5 : 1 and (b) 1.6 : 1. (c) Schematic showing the inclination of the growth planes due to the 1.05 miscut of the r-plane sapphire toward its [0 0 0 1] c-axis [875].
j619
j 3 Growth and Growth Methods for Nitride Semiconductors
620
Figure 3.167 XRD rocking curve data for (a) sample A and (b) sample B with different f angles. The dashed lines correspond to the multiple Gaussian fits to the rocking curve data [875].
Figure3.167b, is again fromthe Ga-polar wings (0.44 tilt), while the central oneis from the crystal plane in windows and third is from the N-polar wings (0.37 tilt). NSOM measurements have been carried out on sample B. Figure 3.168a and b shows the AFM and NSOM images, respectively, taken from a 40 mm · 40 mm area. The window and wing regions and meeting fronts are clearly distinguishable in the NSOM image of Figure 3.168b where the windows appear dark straddled by bright wings. The narrow and the wide bright wings are because of N- and Ga-polar wings, respectively. Part of the observed PL intensity degradation at the meeting fronts may be because of an artifact associated with a probe-size-related effect, as a weak drop in intensity has been observed at the meeting fronts in the near field reflection measurements. However, the PL intensity variations in the window and the wing regions are inherent to the samples. The increased PL intensity in the wing regions suggests improved l quality by defect reduction. After their demonstration of epitaxial growth of a-plane GaN on r-plane sapphire, Craven et al. [876] reported laterally overgrown a-plane GaN, as demonstrated in Figure 3.169 in the form of a cross-sectional TEM image. They found that the
Figure 3.168 (a) AFM and (b) NSOM scans from a 40 mm · 40 mm area of a-GaN ELO sample B [875]. (Please find a color version of this figure on the color tables.)
3.5 The Art and Technology of Growth of Nitrides
Figure 3.169 Cross-sectional bright field TEM images of laterally overgrown a-plane GaN near the edge of SiO2 pattern (encompassing both window and overgrown regions) for ½ 1 1 00 (a and b the latter being an expanded view) and [0 0 0 1] (c) stripes. The dislocation lines are seen to bend from the window region into the
overgrown region for stripes aligned along [0 0 0 1]; whereas no dislocation bending is observed for ½1 1 0 0 stripes. The diffraction conditions are g ¼ ½1 1 2 0 for (a) g ¼ 0 0 0 6 for (b) and g ¼ 0 1 1 0 for (c). Courtesy of J.S. Speck and Ref. [876].
optimum stripe orientation of SiO2 mask is that along the ½ 1 1 0 0 direction. Reminiscence of pendeo-epitaxy, the a-plane GaN can be fully removed and growth can be initiated from resultant walls ((0 0 0 1) on one side and ð0 0 0 1Þ plane on the other) to bypass seed growth on relatively low-quality a-plane GaN. Threading dislocating densities by this method in the range of 106–107 cm2, which are large, have been reported [877]. The evolution of growth is different from that on c-plane [875]. Shown in Figure 3.170 is the evolution of an overgrown GaN sample as viewed by plan view and cross-sectional SEM images after 2 and 5 h of growth. As displayed in Figure 3.168, the PL intensity is also much stronger in the wing regions of ELO a-plane GaN samples than that of the a-plane GaN template. From AFM and NSOM data, it is clear that a significant reduction in dislocation density in the wing region occurs. Laterally overgrown a-plane GaN on r-plane sapphire has also been reported by Chen et al. [878] also achieved fully coalesced overgrown GaN with smooth surface with a root mean square roughness as low as 4.6 Å for a 5 mm · 5 mm sample. Using an improved selective area lateral epitaxy for the growth of a-plane GaN [879], the crystallinity was further improved, with the FWHM value of XRD rocking curve for ½1 1 0 0 decreased from 0.17 to 0.09 . Conductive AFM (CAFM) can be used to produce premature current conduction maps over the surface area (see Section 4.2.5 for a short discussion of the method). Again the point should be made that the device formed by the CAFM tip (Schottky barrier) and the ohmic contacts on the sample must be biased in such a manner as to
j621
j 3 Growth and Growth Methods for Nitride Semiconductors
622
Figure 3.170 SEM images of ELO a-plane GaN samples on rplane sapphire. (a) Plan view and (b) cross-sectional SEM image for an a-plane GaN ELO sample after 2 h of growth at 1000 C. (c) Plan view and (d) cross-sectional SEM images for this sample after additional 3 h growth at 1050 C.
have only the defective (point defects) region(s) to conduct. Under this premise, the region over the windows (presumed to be more defective) and the line of coalescence (over the mask) would conduct first. The complicating factors involve much faster growth along the [0 0 0 1] direction as compared to the ½1 1 0 0 direction during the growth. This is true of course during the buffer layer growth with its own set of complications for natural lateral growth for complete coalescence. Therefore, when ELO is performed on top of this buffer layer, the ELO imposed surface features are added on top of the buffer features with its sets of point defect arrangement making the analysis of already complicated CAFM. In this spirit, we show topographic and CAFM images of an a-plane ELO GaN grown on r-plane sapphire substrate in Figure 3.171 for a reverse bias voltage of 12 V. The meeting front as well as the window regions appear darker than the wing regions. Assuming that the results are reproducible, this would imply that the defective regions are very highly resistive, which is why the lack of current, whereas the wing region is conductive. This goes counter to the underlying premise that the premature conduction in lower quality films should appear. Improvements in the growth procedure led to films where the meeting fronts and the windows represent the premature current conduction paths. As in the case of the growth on c-plane, ELO improves the layer quality, particularly when attention paid to growth conditions that affect the growth rates of Ga- and N-polarity growth fronts. When a two-stage growth process is used for the ELO process, the first promoting vertical growth and the second promoting lateral growth, much reduced wing tilts and thus improved quality has been obtained [880]. TRPL measurements conducted on standard a-plane GaN indicated a radiative recombination lifetime of about or less than 50 ps (system response). With one-stage ELO (two-stage leads to much better films) the radiative recombination lifetime improved to about 200 ps, but still much lower than about 2 ns observed in very best layers grown with ELO using nanonetworks. Koida et al. [881] confirmed the advantage of laterally grown nonpolar GaN owing to the absence of polarization-induced electric fields in quantum wells and higher crystalline quality. Their results also showed that the AlGaN/GaN MQWs grown on
3.5 The Art and Technology of Growth of Nitrides
Figure 3.171 AFM topographic image (a) and conductive AFM current map (b) forward direction and (c reverse direction) of an ELO GaN layer having 4 mm windows separated by 10 mm SiO2 masked stripes perpendicular to the [0 0 0 1] direction. Due to the experimental setup and the
instrument, there is a large voltage drop extrinsic to the tip-sample device, and the dark and bright spots in forward and reverse bias directions, respectively, correspond to larger values of the current. Courtesy of VCU students Chris Moore and Xianfeng Ni.
GaN prepared by ELO can have higher quantum efficiency compared to conventional a-plane MQWs. The above-mentioned reports show that ELO technique is a very promising way to reasonably grow high-quality a-plane GaN for the purpose of highperformance optoelectronic devices. In addition to using r-plane sapphire substrate for the growth of a-plane GaN, it can also be grown on a-plane SiC substrate. Craven et al. [882] reported the successful growth of a-plane GaN growth on a-plane SiC. The crystalline relationship between the substrate and GaN was [0 0 0 1]SiC||[0 0 0 1]GaN, and ½1 1 0 0SiCjj½1 1 0 0GaN. 3.5.11.3 The ð1 1 0 0Þ m-Plane GaN Growth Growth of GaN m-plane is of interest as the lattice mismatch on the face with m-plane sapphire is relatively small (3%) and it is nonpolar. In this vein, the growth of GaN on on-axis m-plane ð1 1 0 0Þ sapphire, m-plane sapphire rotated around the c-axis by as much as 20 , g-LiAlO2, and m-plane SiC [883,884] has been pursued. Although the surface roughness on the order of 2 nm has been obtained on on-axis m-plane sapphire (albeit without any observation of atomic steps that are characteristic of cplane GaN), the tilt of the wafer surface from the m-plane around the c-axis has been reported to improve the quality and surface morphology of GaN [885]. In spite of the fact that Matsuoka and Hagiwara [885] reported that the tilt angles of 15 and 20 led to single-crystalline GaN growth, growth experiments since then indicated not only
j623
j 3 Growth and Growth Methods for Nitride Semiconductors
624
single crystalline GaN but also reasonable LED performance for growth on on-axis mplane [856]. However, growth on exact ð1 1 0 0Þm-plane sapphire does not lead to cdirection of GaN to lie on the m-plane and leads to not fully nonpolar but partially polar (semipolar) surfaces [886]. Baker et al. [886] reported planar GaN films of ð1 0 1 3Þ and ð1 1 2 2Þ grown on ð1 0 1 0Þm-plane sapphire. The in-plane epitaxial relationship for ð1 0 1 3Þ GaN was reported to be ½3 0 3 2GaNjj½1 2 1 0sapphire and ½1 2 1 0GaNjj½0 0 0 1sapphire, whereas that for ð1 1 2 2Þ GaN was ½1 1 2 1GaNjj½0 0 0 1sapphire and ½1 1 0 0GaNjj½1 2 1 0sapphire. The ð1 1 2 2Þ films, however, were reported to have N-face sense polarity and a threading dislocation density of 9 108 cm 2. On the contrary, the ð1 1 2 2Þ films were noted to have Ga-face sense polarity and a threading dislocation density of 2 1010 cm 2. The basal plane stacking fault density was recorded at 2 105 cm 2 for both orientations. The investigation by Ni et al. [887], however, indicated that only the semipolar ð1 1 2 2Þ GaN film orientation results in nominally on-axis ð1 0 1 0Þm-plane sapphire substrates, regardless of the III/V ratio used. HRXRD results indicate a preferred ð1 1 2 2ÞGaNjjð1 0 1 0Þsapphire orientation as shown in Figure 3.172a, which depicts the on-axis 2y–o scans supporting the aforementioned epitaxial alignment. The inset of the figure suggests that ð0 0 0 2ÞGaN is oriented 180 away from ð0 1 1 2Þsapphire. The GaNð1 1 2 2Þ plane forms an angle of 58.4 with ð0 0 0 2ÞGaN, whereas the sapphire ð0 1 1 2Þ plane forms an angle of 32.4 with the sapphire ð1 0 1 0Þm-plane. After combining these results, one determines the epitaxial relationships for GaN on m-plane sapphire, as shown in Figure 3.172b, namely, ð1 1 2 2ÞGaN jjð1 0 1 0Þsapphire, ½1 0 1 0GaNjj½1 2 1 0 sapphire, ½1 2 1 1GaNjj½0 0 0 1sapphire. While on the topic of ð1 1 2 2Þ plane GaN, InGaN/GaN MWQ LED structures have actually been grown and investigated on ð1 1 2 2Þ oriented bulk GaN substrates (produced by cutting thick HVPE GaN templates) [888] in which case the epitaxial layer follows the orientation of the substrate [889]. Free A excitons were reported to dominate the PL spectrum at 10 K and with a weaker but sharp doublet emission associated with neutral donor-bound excitons. The PL decay obtained at 428 nm was fitted with the double exponential decay form with characteristic lifetimes of 46 and 142 ps at 10 K. These values are two orders of magnitude shorter than those in coriented QWs and signify not only the reduced polarization induced internal field but also increased nonradiative processes. The E field of emission from GaN bulk and MQWs was found to be polarized parallel to the ½1 1 0 0 direction, with polarization degrees of 0.46 and 0.69 , respectively, owing to the low crystal symmetry and dispersion in crystalline orientation. Packaged LEDs in epoxy with 320 mm 320 mm area exhibited output power and external quantum efficiency (EQE) 1.76 mW and 3.0%, respectively, for the blue LED, 1.91 mW and 4.1% for the green LED, and 0.54 mW and 1.3% for the amber LED at 20 mA. At 200 mA, the output powers were 19.0 mW (blue), 13.4 mW (green), and 1.9 mW (amber) with associated EQEs of 4.0% at 140 mA (blue), 4.9% at 0.2 mA (green), and 1.6% at 1 mA (amber). The details of GaN-based LEDs fabricated on c-plane are discussed in Volume 3, Chapter 1. In the investigation of Matsuoka and Hagiwara [885], the GaN layers grown on the surface tilted 15 from the m-plane around the c-axis exhibited a very smooth surfaces
3.5 The Art and Technology of Growth of Nitrides
Figure 3.172 (a) XRD 2y–o scan for GaN grown on m-plane sapphire, with the inset showing the off-axis f scans with different c tilt angles (i.e., pole figure) for GaN (0 0 0 2) and sapphire ð0 1 1 2Þ of GaN ð1 1 2 2Þ on m-plane sapphire. (b) A schematic representation depicting the epitaxial relationship derived from XRD measurements for ð1 1 2 2Þ GaN on m-plane sapphire.
and pure near band edge emission with strong PL intensity. The dislocation density in this GaN was reported to be 50% higher than in GaN grown on c-plane, which is expected to improve with further developments. The variation of the surface morphology of GaN on on-axis and vicinal m-plane sapphire grown by OMVPE is shown in Figure 3.173. To reduce the defect density, ELO has been carried out using HVPE [890], resulting in basal plane stacking fault and threading dislocation densities of less than 3 · 103 cm1 (from 105 cm1) and5 · 106 cm2 (from109 cm2),respectively, in the Ga-face (0 0 0 1) wing regions of the ELO material. These values are better than those reported for the a-plane GaN, making the m-plane GaN very promising for device applications. It should be mentioned that m-plane ELO GaN films were grown on MBE m-plane GaN films, which, in turn, are grown on either g-LiAlO2(1 0 0) or 6H-SiCð1 1 0 0Þ substrates, as the c-direction of the GaN films grown on ð1 1 0 0Þ sapphire does not lie on the ð1 1 0 0Þ m-plane of sapphire. The growth experiments were carried out in a three-zone horizontal directed-flow HVPE reactor. Moreover, unlike the ELO process on c- and
j625
j 3 Growth and Growth Methods for Nitride Semiconductors
626
Figure 3.173 Variation of the GaN surface morphology grown by OMVPE on sapphire with the surface tilted 0 , 8 , 15 , and 20 from m-plane around the c-axis of sapphire as observed with a differential interference optical microscope. Courtesy of T. Matsuoka.
a-plane GaN, X-ray diffraction experiments did not detect any wing tilt. Vertical growth rates ranged from 40 to 300 mm h1 at substrate temperatures of 860–1060 C. Figure 3.174a shows a plan view SEM image for the m-plane ELO sample with SiO2 stripes oriented along the a-axis of the sapphire substrate obtained by Ni et al. [887]. The cross-sectional SEM image of Figure 3.174b shows that the wings for this mask orientation are inclined by 32 with respect to the substrate plane with well-defined aand c-plane surfaces. The growth along the c-axis advances faster than others. Therefore, the observation of inclined wings is consistent with the epitaxial relationships shown in Figure 3.172, which suggest a 32.4 angle between the c-axis of GaN and c-axis of sapphire. The upwardly inclined wings are suggested to be because of the Ga-polar (0 0 0 1) wings, since their growth fronts (GaN c-plane) are smooth. Because of the large incline angle, only the Ga-polar wings extend, whereas the Npolar wing growth is stymied by the template. With further growth, the Ga-polar wings advanced along the c-axis of GaN with negligible growth along the a-axis. The m-plane GaN growth has also been reported on g-LiAlO2(1 0 0) substrate rather than sapphire substrate. The epitaxial relationship between the m-plane GaN and LiAlO2 substrate is ½1 1 0 0GaNjj½1 0 0LiAlO2 and ½1 1 2 0GaNjj½0 0 1LiAlO2 [891]. Waltereit et al. [853] reported on their results concerning the successful growth of m-plane GaN on g-LiAlO2(1 0 0), which was free of polarization-induced electric field. Sun et al. [891] from the same research institute reported that the stoichiometric ratio of Ga and N for the GaN nucleation layer would influence the phase of subsequently grown nonpolar GaN layers. Chen et al.[878,879] reported their result on growth of m-plane GaN on freestanding ð1 1 0 0Þ GaN template, which was prepared by HVPE on g-LiAlO2(1 0 0) substrate. They also showed that the emission of AlGaN/GaN MQWs grown on this m-plane GaN template did not exhibit any quantum-confined stark effect. Researchers also observed in-plane anisotropic optical properties in m-plane GaN on g-LiAlO2(1 0 0) substrate. Ghosh et al. [892] studied the polarization dependence of absorption, reflectance, and photoreflectance spectra of compressively strained
3.5 The Art and Technology of Growth of Nitrides
Figure 3.174 (a) Plan view, (b) cross-sectional SEM, and (c) crosssectional TEM images of GaN ELO sample with SiO2 stripes oriented along the a-axis of sapphire. (d) Plan view, (e) crosssectional SEM, and (c) cross-sectional TEM images of the GaN ELO sample with SiO2 stripes oriented along the c-axis of sapphire.
m-plane GaN. They observed different optical energy bandgaps when varying the measurement angle with respect to the c-axis [0 0 0 1], which was attributed to the influence of in-plane compressive strain. Moreover, in-plane anisotropy of PL intensity for m-plane AlGaN/GaN MQWs was found by Rau et al. [893] and Kuokstis [894]. Sun et al. [895] studied the polarization anisotropy of PL for m-plane InGaN/GaN MQWs and found similar anisotropic phenomenon. The details of optical processes in m-plane GaN are discussed in Volume 2, Chapter 5. 3.5.12 Growth of p-Type GaN
Until 1989, efforts to achieve p-type GaN led to compensated high-resistivity material. Activation of Mg doping for p-type conductivity has been responsible for placing GaN and related materials in a commanding position as light emitters [896,897]. An inordinate number of early investigations were directed toward the potential of Zn as
j627
j 3 Growth and Growth Methods for Nitride Semiconductors
628
a p-type dopant in GaN. These investigations demonstrated that Zn impurities effectively compensate the semiconductor and result in high-resistivity material. All Zn-doped samples exhibited the commonly observed 2.8 eV emission with heavily doped samples, which exhibit room-temperature peaks at 1.8–1.9, 2.2, and 2.5–2.6 eV. A number of researchers have investigated Mg doping that is known to effectively compensate GaN. As the Mg concentration is increased, the Mg-related emission peak broadens and shifts to lower energies. Attempts to dope GaN with Cd failed to produce p-type conductivity, even though a Cd-related peak at 2.85 eV was observed. As for Be, early attempts led to compensated material as well while yielding high-resistivity GaN. Later efforts led to possibly p-type material but the hole concentrations obtained were too low to be measured by the Hall effect, although p–n junctions have been made, leaving Mg the only practical p-type dopant. Hg doping was also investigated with no electrical measurements. Hg, too, leads to optical transitions at energies much lower than the band edge, about 0.4–0.8 eVabove the valence band edge. Carbon has also been tried and produced only deep-level transitions reaching even the yellow region with no electrical conductivity. Li atoms have gone the same way and produced only compensated material with optical transitions of about 750 meV below the band edge [369]. A common point is that all p-type dopants including Mg, particularly at high concentrations, can lead to donorlike levels that act to compensate the semiconductor. In a sense, unsuccessful attempts end up generating defects and only when the undoped layers have a low background donor concentration, which is also applicable to the case of MBE growth, can the available data point to active p-type doping. The deep transitions have, in fact, been exploited to produce blue emitters before reliable p-type doping was realized. First achieving and then controlling p-type doping represented a formidable challenge for semiconductor nitride researchers. The first breakthrough came when Amano et al. [898] converted compensated Mg-doped GaN into conductive p-type material by low energy electron beam irradiation (LEEBI). Nakamura et al. [897] improved upon these results using LEEBI to achieve GaN with p ¼ 3 · 1018 cm3 and a resistivity of 0.2 O cm with the follow-up of thermal annealing at 700 C under an N2 ambience, which converted the material to p-type equally well. Owing to the high binding energy (150–200 meV) of Mg, acceptor activation ratios of only 102 to 103 are typically achieved and require Mg chemical concentrations in the 1020 cm3 range. However, OMVPE samples require postgrowth activation, the MBE-grown samples do not. Hydrogen has long been suspected of passivating Mg atoms and is driven off in the subsequent annealing treatment. The general issue of p-type doping in general and Mg doping in particular is discussed in Section 4.9.3. Theoretical insights have provided a plausible explanation for the success of the Mg acceptor in GaN compared to the other group II materials, which continue to compensate GaN even after LEEBI or annealing. Owing to the strong binding of the nitrogen anion, group III nitrides are considerably more ionic than typical group III semiconductors. Their calculated band structures resemble II–VI semiconductors in many ways, including a large splitting between upper and lower valence bands (LVBs). The Ga(3d) core-level energies have been predicted and observed to
3.5 The Art and Technology of Growth of Nitrides
overlap in energy with the N(2s)-like LVB states as a result of the LVBs being deeper in GaN than in GaAs, which is a more typical group III semiconductor. The resulting energy resonance causes the Ga(3d) electrons to strongly hybridize with both the upper and lower valence band s- and p-levels. Such hybridization is predicted to have a profound influence on GaN properties including such quantities as the energy bandgap, lattice constant, acceptor levels, and valence heterojunction offsets. It is known, in the cases of ZnS and ZnSe, that potential acceptors such as Cu have the p–d acceptor level raised because of antibonding repulsion between Cu(3d) and Se(4p) [S (3p)] resulting in a deeper level, whereas impurities without d-electron resonances form shallow acceptors. Mg has no d-electrons and turns out to be sufficiently shallow for room-temperature p-type doping of GaN. On the contrary, Zn, Cd, and Hg, all of which have d-electrons, form deep levels in GaN. These observations are consistent with theory. Enhanced hole concentrations can be obtained by codoping, such as the Mg–O pair, the details of which can be found in Chapter 4. 3.5.13 Growth of InN
InN is pivotal in the triad of the group III nitride system and holds the key for full exploitation of the group III nitride system for optical emitters, as it is the binary needed to form alloys with transition energies in the visible. This being the case, the bandgap of InN, particularly for considerable InN molar fractions, must be known accurately for predicting the resultant transition energy on the basis of the knowledge of alloy composition. The issue of bandgap has been controversial with earlier data pointing to a bandgap of 1.9 eV and the later data suggesting this parameter to be near 0.7 eV. The details regarding the bandgap-related issues are discussed in Section 2.9. In this section, the focus is placed primarily on growth-related issues. In part because of the low dissociation temperature (beginning at temperatures as low as approximately 500 C at low pressures), growth of high-quality InN films has proven difficult. The nitrogen equilibrium vapor pressure over InN is many orders of magnitude higher than that over both AlN and GaN (Figure 1.6). To circumvent this difficulty, a number of options, including low-temperature (less than 600 C) deposition, various growth techniques, such as reactive evaporation, ion plating, reactive RF sputtering, reactive magnetron sputtering, vapor phase epitaxy, microwave-excited OMVPE, laser-assisted CVD, halogen transport (HVPE), pulsed laser deposition (PLD), and MBE, have been used [369,899]. Unlike the case of GaN wherein the better quality films result inclusive of the surface morphology with Ga-polarity samples, properties of N-polarity samples InN are better. Most of the InN growths by MBE have been performed at below 500 C, for example, 475 C for In-polarity samples and below 600 C, for example, 575 C for Npolarity samples [900,901]. Essentially, the highest growth temperature that can be tolerated would lead to a better film. In this vein, the best films are obtained with high V/III ratios, as well as high indium resulting in growth rates over 1 mm h1 in MBE growth. The high V/III ratio is to avoid In droplet formation on the surface. However, from the point of view of In adatom surface migration during growth, In-rich
j629
j 3 Growth and Growth Methods for Nitride Semiconductors
630
conditions are preferable. Therefore, the optimum point is a compromise between the two requirements that culminate in a surface stoichiometry that is slightly N rich. With N-polarity, the growth temperature can be nearly 100 C higher than that for Inpolarity growth. An example for the growth schedule for both the In and N polar InN films is shown in Figure 3.176. In early developments and in particular for films produced by sputtering, short thermal annealings (30 min, 450–500 C) have been found useful for improving the crystallinity of the as-deposited InN films. Particularly true for early varieties, InN films were generally polycrystalline with agglomerates of small columnar grains having various degrees of texture and epitaxy. The improvement in the crystalline quality is because of a rearrangement of these crystallites in the film. The electrical properties of the as-deposited InN films are dominated by conduction between the grains. Buffer layers used for the growth also play a role. For example, application of an AlN buffer layer on a sapphire substrate significantly alters the granular polycrystalline growth mode of InN and results in substantial improvements in the growth morphology and in electrical properties. However, the electrical transport properties are still dominated by intergranular interactions. It is well known that annealing helps mainly when the starting sample quality is not really all that good. There is no substitute in compound semiconductors to employ the best method and growth conditions to produce the highest quality film by growth alone. This is what transpired with efforts focusing on growth by MBE and OMVPE. In what follows, the growth and properties of InN by first MBE followed by OMVPE and finally by other techniques are discussed, keeping in mind that MBE and OMVPE produced films are the best available. In one effort, reasonably high-quality epilayers of InN on (0 0 0 1) sapphire substrates have [902] been achieved by radio frequency plasma excited (RF-MBE) by using low-temperature grown intermediate layers. Nitridation was carried out at 550 C for 1 h with a nitrogen flow rate of 1 sccm and RF plasma power of 300 W. An InN buffer layer was deposited at 300 C with a nitrogen flow rate of 2 sccm and RF plasma power of 330 W. The InN epilayers were grown at 550 C for 1 h with a nitrogen flow rate of 2 sccm and RF plasma power of 240 W. This was followed by an intermediate layer at 300 C with the same conditions as for the buffer layer and the process was repeated three times. SEM observations of the final epilayer showed uniform surface morphology. Layers with a thickness of 600 nm and with a carrier density of 1.0 · 1019 cm3 and a reasonably high electron mobility of 830 cm2 V1 s1 have been reported. More details of electron transport in much improved InN layers can be found in Section 1.4.4. Fairly high-quality InN films have also been grown on Si(1 1 1) substrates using a sequence of processes whereby Si (7 · 7) structure was obtained followed by nitridation of the Si surface to form a few monolayers of Si3N4, which was followed by the growth on an AlN buffer layer. On top of this buffer layer, InN layer was grown all by RF-MBE. The interesting features of this approach is that lattice constant of layers/templates that are juxtaposed are indexed well in terms of the lattice constant. For example, a 2 : 1 indexing is provided by the Si : Si3N4 pair, a 5 : 4 in-plane lattice constant indexing is provided by the AlN : Si pair, and a 8 : 9 lattice constant indexing
3.5 The Art and Technology of Growth of Nitrides
Figure 3.175 Reflection high energy electron diffraction patterns obtained on (a) clean (1 1 1) surface indicating the characteristic 7 · 7 reconstruction, (b) b-Si3N4 with 8 · 8 reconstruction, (c) AlN with 3 · 3 recontruction, and (d) InN with 1 · 1 reconstruction. Courtesy of S. Gwo.
is provided by the InN : AlN pair. By using reflection high energy electron diffraction, images of which are shown in Figure 3.175, and cross-sectional transmission electron microscopy, Wu et al. [903] have noted the pseudomorphic temperature to commensurate lattice transition within the first monolayer of growth in switching growth from template to the layer in the aforementioned pairs. This resulted in an abrupt heterojunction at the atomic scale. This avenue of indexing among the otherwise lattice mismatched pairs allows the formation of commensurate and two-dimensional layer with minimal strain and reduced propagating dislocations. Use of AlN buffer layers on sapphire substrates is a good alternative for growth of good-quality InN layers by all MBE growth of In polarity samples with improved structural and electrical properties. An improved surface morphology, a monotonic
j631
j 3 Growth and Growth Methods for Nitride Semiconductors
(a)
Time (au)
(b)
500 nm N-polar GaN
Nitridation
~600 °C Med./temp. N-polar GaN
MBE In-polar InN
Sapphire thermal treatment
~500 °C
800 °C
Temperature (au)
800 °C MBE 500 nm Ga-polar GaN
OMVPE Ga-polar GaN:Thermal
Temperature (au)
632
InN:N-polar
Time (au)
Figure 3.176 Growth schedule employed in the preparation of Inpolar (a) and N-polar (b) InN on Ga-polar OMVPE-grown GaN and N-polar MBE-grown GaN buffer layers, respectively. Courtesy of A. Yoshikawa.
increase in the Hall mobility, and a decrease in the electron density result with increasing thickness of the AlN buffer layer. A Hall mobility of <800 cm2 V1 s1 with a carrier concentration of 2–3 · 1018 cm3 at room temperature has been reported [904] for an InN layer thickness of 0.1 mm. These layers have a quality comparable to that obtained by the migration-enhanced epitaxy variant of MBE [905] based on alternate supply of pure In atoms and nitrogen plasma. As in the case of GaN [348], the N-polar InN growth rate remained higher than that for In-polar InN at or above the maximum temperatures employed for growth. This is very important as it is well known that the higher the growth temperatures, while maintaining good surface morphologies, the better the quality of the resultant films. It must be underscored that the option of N-polarity makes it possible to extend the growth temperature by nearly 100 C, which is quite large considering the terminally activated exponential nature of processes involved. The growth rate as a function of temperature for both In-polar and N-polar InN by RFMBE is shown in Figure 3.177. Note that there are two different sets of N-polar growth runs marked as 1 and 2, the former representing the same growth conditions as those used for the In-polar growth series, which is also shown. The shaded area for the N-polarity film represents the stoichiometric region with optimal quality and void of In droplets. As clearly seen in Figure 3.177, with increasing substrate temperature, InN tends to decompose, and if continued, eventually only the In droplets would remain on the surface. However, even at these high temperatures, if the N-flux is further increased, one can reduce the appearance of In droplets and thus the useful epitaxial growth temperature, the region wherein high-quality epitaxy is attained can be extended to higher temperatures by some 100 C. Therefore, it is fair to state that at higher temperatures, the V/III beam flux ratio must be sufficiently high compared to those at low temperatures. This is necessary for sustained and stable growth under an effective surface stoichiometry to be slightly higher than unity. To reiterate, stoichiometric growth results for a given set of N and In arrival rates at a given temperature. Increasing the In flux beyond this point could result in In droplets and perhaps even In inclusions in the films, which have been linked to the small bandgaps measured, as discussed in Section 2.9.1. Increasing the N flux beyond that point leads to N-rich growth conditions with consequences such as poorer electron mobilities.
3.5 The Art and Technology of Growth of Nitrides
3.0 N polarity In polarity
2
2.0
1
0.5
0.0 350
400
450
500
550
In droplets
1.0
Best quality
1.5
Best quality Best quality
InN growth rate (Ås–1)
2.5
600
Substrate temperature (ºC) Figure 3.177 Growth rate versus growth temperature for one In-polar (open circles) and two N-polar (solid circles) InN. The series of N-polar films marked with one 1 is prepared under otherwise identical conditions with In-polar films shown. The shaded areas show the region wherein InN growth runs were executed for optimal results. The solid lines represent a guide to eye. Courtesy of A. Yoshikawa.
Figure 3.178 provides a qualitative view of the stoichiometric boundary for InN indicated by a dashed line. To the left of this In-rich boundary and depending on the extent and the temperature, In droplet conditions would ensue. To the right of this boundary, N-rich conditions prevail. The shaded area shows the region where In droplet free growth can be obtained. Furthermore, the boundary for In-droplet free area for the high beam flux region deviates from the stoichiometric line (dashed line in Figure 3.178) to the N-rich side. This area corresponds to the high-growth rate region and represents almost the highest substrate temperature that can be employed. Naturally, to counter higher desorption rates endemic to high growth temperatures while obtaining reasonably high growth rates at higher end temperatures, the N-III ratio must be higher than that for nearly unity stoichiometric conditions used for lower temperatures. In terms of the surface reconstruction, the RHEED patterns observed on InN surfaces is somewhat similar to those observed on GaN surface. The RHEED patterns on InN taken along the ½1 0 1 0 azimuth are shown in Figure 3.179. The pattern in the left panel is for In-polarity, whereas that in the right panel is for N-polarity. The 3 pattern along the ½1 0 1 0 azimuth for N-polarity growth represents a fine surface reconstruction, which is a clear indication that smooth surfaces are indeed possible [906]. For a description of this particular surface reconstruction, refer to
j633
j 3 Growth and Growth Methods for Nitride Semiconductors In beam flux (atomic intensity)
634
Stoichiometric GaN for reference GaN @2.4 μm h–1 and InN @1.3 μm h–1
In-droplet free area
N beam flux (atomic intensity) Figure 3.178 A qualitative view of the GaN and InN stoichiometric growth boundary in RF MBE. To the left of the dashed line, which marks the stoichiometric boundary, In rich and depending on the extent and temperature In droplet conditions would ensue. To the right of the
boundary, N-rich conditions prevail. The shaded area shows the region where In-droplet free growth can be obtained. The fact that this region sags below the stoichiometric boundary is due to the mechanical limitations imposed on In flux. Courtesy of A. Yoshikawa.
Figure 3.179 RHEED patterns on InN taken along the ½1 0 1 0 azimuth for In polarity (left panel) and N-polarity (right panel). The 3 pattern along the ½1 0 1 0 azimuth for the N-polarity growth. Courtesy of A. Yoshikawa.
Section 3.5.6 devoted in part to the fundamentals of RHEED. The representative AFM images of In- and N-polarity InN films grown at 470 and 580 C, respectively, are presented in Figure 3.180. Although both surfaces were very flat as can be judged from the height scale of about 3.5 nm, the morphological features vary dramatically. Clear growth steps could be seen in the image for the N-polar surface, opposite to the case for GaN. This implies that the growth temperature is high to the point that In is able to diffuse to the ledge edges before incorporation [907]. The In flux intensity had to be lower for In-polar surfaces than for the N-polar ones for the same N-plasma conditions to achieve the smooth surface morphologies shown in Figure 3.180. Although not shown, the surface of N-polar InN grown at 550 C exhibited pin holes, whereas that grown at 580 C was void of such pin holes with fully coalesced surface. In terms of the structural characterization, the FWHMs of N-polar InN(0 0 2) and (1 0 2) were about 3.7 and 16 arcmin, respectively, comparable with the N-polarity
3.5 The Art and Technology of Growth of Nitrides
Figure 3.180 AFM images of InN films grown with In-polarity on OMVPE-GaN at 470 C (a), and grown with N polarity on MBEN-polar GaN at 580 C. The N-polar InN grown at 580 C shows atomic terraces that are indicative of the step flow like growth mechanism. Courtesy of A. Yoshikawa and Ref. [906].
MBE GaN template used. On the contrary, the FWHMs of In-polarity InN(0 0 2) and (1 0 2) were about 10 and 16.7 arcmin, respectively. The (0 0 2) diffraction associated with the In polar sample is much more inferior to the high-quality OMVPE GaN template. Moreover, the (0 0 2) diffraction rocking curves with In-polar films can also be characterized with a higher background (about two order of magnitude) than those of N-polar films most likely because of more random distribution of grains, particularly the associated tilt because the (0 0 2) diffraction is sensitive to any tilting of the c-plane. The reduction in the excess signal for the (1 0 2) diffraction, although present, was not as notable, as this diffraction is sensitive to dislocations. The dislocation density in N-polar samples is on the order of 109 cm2. The roomtemperature Hall measurements of N-polar InN films resulted in a maximum electron mobility in the range of 500–1200 cm2 V1 s1 (with a maximum of 2000 cm2 V1 s1 for an 8 mm thick film and electron concentrations of (1–5) · 1018 cm3). The typical electron concentration obtained in In-polar films grown under optimum growth conditions was in the range of (1–10) · 1018 cm3 and the highest electron mobility for 1.5–1.65 mm thick epilayers was 1600 cm2 V1 s1 and the average values are about 1200 cm2 V1 s1. More details of electron transport in much improved InN layers can be found in Section 1.4.4. The differences in electrical properties for films grown with different polarity were thought to be mainly because of the differences in crystal quality. The photoluminescence of the InN films peaked at 0.697 eV. Optical reflection and transmission measurements were also made in an effort to determine the bandgap of InN that resulted in a room temperature bandgap of about 0.70–0.75 eV, details of which can be found in Section 2.9.1. Turning our attention to OMVPE, the main issues for growth of InN with this method include the temperature of growth, the flow rate of gases, the III/V ratio, and
j635
j 3 Growth and Growth Methods for Nitride Semiconductors
636
the nature of the substrate and preparation of its surface prior to growth. Growth temperatures ranging from 325 to 600 C have been tried with different degrees of success, the main concern being In droplet formation. Temperature of 600 C is the highest that is used to grow InN without forming droplet in plasma-assisted OMVPE [908]. Plasma-assisted growth with TMI and NH3 as the precursors and microwave-excited nitrogen as the reactive nitrogen source have been the two main modes for OMVPE growth. Growth of InN epilayers has been attempted by the OMVPE technique on a variety of substrates including sapphire, MgAl2O4, GaAs, GaP, and GaN with varying degrees of success. Although GaAs(1 1 1)B substrate provided the best surface morphology, the InN layers are reported to have a higher carrier density and a lower electron mobility even under optimal conditions than those grown on a-sapphire (0 0 0 1) for which n 5 · 1019 cm3 and Hall mobility mH 300 cm2 V1 s1 were achieved with growth at atmospheric pressure [909]. Nitridation of the sapphire surface (using microwave-excited nitrogen as the reactive nitrogen source for both nitridation and growth) is found [910] to result in better crystalline quality of the InN layer as assessed by X-ray rocking curve analysis (FWHM of 140 arcsec for (0 0 0 2) diffraction). However, the quality depends strongly on the temperature of nitridation. InN films with an even superior structural quality (XRC FWHM of 96 arcsec) have been grown on sapphire by OMVPE at a temperature of 375 C under a high V/III ratio [911]. In a series of publications, Matsuoka et al. [912] and Matsuoka [913] reported on 0.1–0.2 mm thick InN samples grown by atmospheric pressure OMVPE on 1.6 mm thick GaN, which, in turn, was grown on sapphire (0 0 0 1), primarily in an effort to determine a set of growth parameters leading to InN growth and also determining its optical bandgap energy. In this particular study, the sapphire substrate was cleaned in hydrogen atmosphere at 1050 C followed by nitridation in ammonia gas for 5 min. The growth commenced with a 20 nm thick GaN nucleation layer at 550 C followed by annealing at 1020 C for 10 min. On this annealed nucleation layer, a 1.1 mm thick GaN buffer layer was grown at 1010 C. The typical X-ray diffraction spectral FWHM for (0 0 0 2) diffraction of the GaN buffer layer was about 50 and 20 arcsec in the o- and o 2y scans, respectively. The InN layers under investigation were then grown in a temperature window between 500 and 600 C and with a TMI flow rate of 1.0–6.0 mmol min1 under an NH3 flow rate of 15 slm. One of the main problems is that ammonia decomposition rate is less than 0.1% in this growth temperature range. To ameliorate the situation somewhat, N carrier gas, instead of hydrogen was used in an effort to promote ammonia decomposition. InN film with a thickness of less than 0.2 mm were successively grown for 8 h, which translates to a growth rate of only 0.025 mm h1, which is comparable with growth rates over 1 mm h1 obtained for MBE growth of InN. The smallest value in X-ray diffraction was 37.7 and 8.8 arcmin in the o- and o 2y scans, respectively, in great part because of thin layers. The reason why these values are quite large in comparison with GaN is that the InN film is too thin. Matsuoka et al. [912] also established the growth phase diagram for InN in an OMVPE environment. The relationship between the growth conditions and the
3.5 The Art and Technology of Growth of Nitrides
625 No growth
X
600
A
Growth temperature (ºC)
C
575 Single crystal
550 B
525
500
In precipates A
475 1
2
3 4 TMI flow rate (μmol min–1)
5
Figure 3.181 Surface growth phase diagram for InN prepared by OMVPE. Ref. [912] and courtesy of T. Matsuoka.
resulting InN grown on GaN is shown in Figure 3.181. As highlighted in the figure, InN grown in region B showed no spectra from metal-like indium in X-ray diffraction, but region A did show In precipitates. No film growth was observed if conditions representing region C were used. The samples in region B were brownish in color. Their surface morphology could be characterized with very smooth surfaces with no indium droplets. For region B, the flow ratio of ammonia and TMI (V/III ratio) was 6.6 · 105. This sample was confirmed to have high crystalline quality by confocal micro-Raman scattering [914]. Although details of the band structure are covered in Section 2.9.1, a bandgap value in the range of 0.8–1.0 eV was deduced by optical absorption and 0.7–0.8 eV was deduced by photoluminescence measurements. Metal-organic MBE (MOMBE) has also [915] resulted in successful growth of goodquality InN layers on sapphire with a careful choice of growth conditions yielding layers with electron concentrations as low as 8.8 · 1018 cm3. The lattice-mismatch dependence of the substrate, which is a key factor in determining the crystal quality, has been studied by using OMVPE growth on GaN, AlN, and sapphire in the same system. Growth on GaN showed the best crystalline quality. InN layers with a thickness of 2400 Å grown on GaN have been reported [916] with a Hall mobility of 700 cm2 V1 s1 even at carrier densities as low as 5 · 1019 cm3. Variants of the sputtering technique have been applied to the growth of InN epitaxial layers with varying degrees of success. In 1970s, RF magnetron sputtering was the most used method applied to the growth on InN, although a somewhat novel
j637
j 3 Growth and Growth Methods for Nitride Semiconductors
638
method based on UHV electron cyclotron resonance (ECR) assisted reactive sputtering technique [917] has also been introduced with encouraging results. RF magnetron sputtering has even been used to deposit InN films on glass substrates [918] with n-type conductivity corresponding to an electron concentration of 1020 cm3. Hall mobilities ranging from 18 to 115 cm2 V1 s1 have been obtained with the substrate temperature ranging from 100 to 500 C. Growth of InN on GaAs has also been achieved [919] by this method, although with poor inplane uniformity, at temperatures as low as 100 C. Nitridation in nitrogen plasma before growth improves the in-plane orientation significantly, leading to singlecrystal InN. For growth on (0 0 0 1) sapphire substrates, epitaxial grains are obtained only at growth temperatures ranging from 200 to 525 C. The ECR-assisted magnetron sputtering has been reported to lead to enhanced Hall mobility (from 50 to 80 cm2 V1 s1), a decrease in carrier concentration by a factor of two, and a reduction in homogeneous strain [917]. Apart from the principal techniques discussed above, CVD, laser-assisted CVD, and HVPE are among the other methods investigated for InN crystal growth. Halide vapor phase epitaxy was first used [920] for the deposition of InN films using InCl as the In source and NH3 as the nitrogen source in 1994. It has recently been used to grow InN layers on sapphire substrates with a GaN buffer layer, based on InCl3 and NH3 as the source materials for In and N, respectively. Reproducible InN epitaxial layers have been obtained [921] at growth temperatures as high as 750 C. Laserassisted CVD has been reported [922] to grow thin (20 Å or so) layers of InN on Si (1 0 0) substrates at temperatures in the range 300–700 K. 3.5.14 Growth of AlN
Early developments of nitride semiconductors relied on AlN more than the other two binary end points in the AlGaInN system, in part because of the large cohesive energy of Al and N bonds, as discussed in Section 1.3. In addition to being a large bandgap semiconductor, AlN is a pivotal component of the AlGaN ternary and is necessary in all the heterostructure devices fabricated in nitrides. In one particular application, in the form of solar-blind detectors, single-crystalline AlN can serve as an ideal substrate, as back illuminated structures are best suited for large mole AlxGa1xN alloys with Al mole fractions in excess of x ¼ 0.45. With its large low temperature direct bandgap of about 6.1 eV and about 6 eV at room temperature (the widest direct gap among all III–V nitrides [923]), high thermal conductivity and hardness, and high resistance to chemicals [924], AlN has many attractive features [925]. AlN layers have been grown on a variety of substrates, sapphire and SiC being the most common ones. In both case, particularly in the case of SiC, the substrate preparation for epitaxy is crucially important, as discussed in Section 3.2.3.2, where a high-temperature H etching method is discussed. Yamada et al. [926] reported on the nucleation and growth kinetics of AlN films on atomically smooth 6H-SiC(0 0 0 1) surfaces, which were obtained by HCl etching at elevated temperatures prior to growth. The surface morphology and the defect density of the resulting AlN films on
3.5 The Art and Technology of Growth of Nitrides
such surfaces were significantly improved, compared to those on as-received SiC surfaces, owing to elimination of defects, enhanced diffusion, and reduced incoherent boundaries at the coalescence regions of the AlN islands. AlN nuclei on the asreceived SiC surface were crystallographically misaligned, which induced incoherent boundaries at the coalescence stage. The coalescence of nuclei and a smaller aspect ratio (height/width) of each nucleus are observed on the etched SiC sample. A threedimensional AlN nucleus was formed within a limited area (30 nm in diameter) where the SiC surface is atomically smooth, implying the presence of a kinetic barrier for adatoms at step edges such as the Schwoebel barrier [927]. In contrast, AlN nuclei on the etched surface have a larger size probably because of the absence of kinetic barriers to adsorption, which suppress the diffusion length of adatoms. Crosssectional BF TEM micrographs of each AlN film indicate strong and localized contrasts, which are attributed to threading dislocations and planar defects at the island coalescence regions, in AlN grown on as-received SiC. On the contrary, very small features characterized samples grown on the etched SiC substrate, indicative of a higher quality AlN film. Growth of AlN on sapphire substrates has been undertaken by Heffelfinger et al. [928], who focused on the initial stages of AlN growth on (0 0 0 1) oriented Al2O3 substrates by MBE using conventional and high-resolution TEM. For film thickness 25 nm, AlN forms islands of varying alignments with respect to the Al2O3 substrate. Some of the AlN islands were found to be well aligned with the AlN½1 12 0jjAl2 O3 ½1 0 1 0 and AlN(0 0 0 1) and Al2O3(0 0 0 1), which matches closed-packed planes and directions as in the case of GaN on sapphire, as discussed in detail in Section 3.4.1. Other islands exhibited either an alignment of one set of planes, that is, grains with the ð0 1 0 1ÞAlN==ð1 1 2 0Þ Al2O3 alignment, or misalignment with respect to the Al2O3 substrate. As the AlN film thickness is increased, the film becomes continuous and the closed-packed planes and directions of the film and substrate are aligned for the majority of the film. Islands of AlN with an alignment other than this predominant orientation disturb the growth near the AlN/Al2O3 interface and create displacements along the AlN[0 0 0 1] direction in overlying AlN grains. These misaligned AlN grains cause planar defects to form in the epitaxial AlN films, as shown in Figure 3.182. Figure 3.183 is a schematic representation that shows how oriented growth along the [0 0 0 1] AlN direction (arrows) would lead to the observed microstructure. Epitaxial relationship between AlN and sapphire is expected to be identical to that between GaN and sapphire, in fact many of the GaN layers used for determining the epitaxial relationship have been grown in AlN buffer layers, which, in turn, have been grown on sapphire and the subsequent GaN layer takes the orientation of AlN. Table 3.6 summarizes the observed orientational relationships between GaN (AlN) and Al2O3 substrate. Further details can also be found in Table 3.7. As for AlN by MBE, King et al. [929] studied the mechanisms of AlN growth on GaN via ammonia RMBE utilizing X-ray photoelectron spectroscopy, low-energy electron diffraction, and Auger electron spectroscopy. Essentially, the growth of AlN on GaN occurred via a layer-by-layer growth within the 750–900 C temperature range investigated, as is consistent with growth of any semiconductor where the
j639
j 3 Growth and Growth Methods for Nitride Semiconductors
640
Figure 3.182 High-resolution TEM image of the Al2O3/AlN interface. Marker G indicates two small AlN grains that have an alignment other than the primary orientation, that is, the (0 0 0 1) AlN planes are not parallel to the (0 0 0 1) Al2O3 planes. Marker D indicates a shift in the AlN fringes between two adjacent AlN grains [928].
[0 0 0 1] AlN Figure 3.183 Schematic representation of the AlN grains and their evolution, from top to bottom, on the (0 0 0 1)-oriented Al2O3 surface. Patterned after Ref. [928].
3.5 The Art and Technology of Growth of Nitrides
epilayer has a smaller lattice constant than the template underneath. AlN on Si(1 1 1) by ammonia RMBE has been studied by Nikishin et al. [930]. The transition from (7 · 7) to (1 · 1) silicon surface reconstruction at 800 C was used for in situ calibration of the substrate temperature, which is typical among the growers by MBE. The initial deposition of a monolayer or even less of Al, at 830–890 C, formed a nucleation layer for the growth of AlN. The Al layer may have also reduced islands of SiNx that might have formed on the silicon surface prior to the onset of epitaxial growth owing to background NH3. The transition to two-dimensional growth mode occurred following an initial AlN thickness of 7 nm, if the growth conditions were optimum. Owing to piezoelectric and nonlinear optical properties, AlN is of interest all by itself and in that context for specialty devices such as surface acoustic wave devices. To this end, Assouar et al. [931] explored deposition of reactive dc magnetron sputtering of Al on Si substrates as a function of N2 concentration in Ar–N2 gas mixtures. XRD of 2 mm films shows best crystalline quality of AlN films between 60% and 80% of N2 in the Ar/N2 mixture, but with a columnar structure textured in (0 0 2) orientation corresponding to hexagonal wurtzite structure with a c-axis perpendicular to the surface. Films synthesized with 75% N2 showed higher peak intensity of AlN(0 0 2) diffraction and higher resistivity. AFM analysis of AlN films demonstrate a low roughness at approximately 3 nm. Deposition by ion-assisted reactive dc magnetron sputter deposition of AlN has been reported as well [932]. The low-energy ionassisted growth (Ei ¼ 17–27 eV) resulted in an increasing surface mobility, promoting domain-boundary annihilation, and epitaxial growth. Domain widths increased from 42 to 135 nm and strained-layer epitaxy was observed in this energy range. For Ei > 52 eV, an amorphous interfacial layer of AlN was formed on the SiC, which inhibited epitaxial growth. Employment of UHV conditions and very pure nitrogen sputtering gas yielded reduced impurity levels to the extent that O contamination was 3.5 · 1018 cm3. The RMS surface roughness and surface feature size of AlN films as a function of ion-assisted energy (Ei) during the growth were measured with AFM with the largest RMS roughness value of 10 resulting for an ion energy of 20 eV. Large Al–N binding energy allowed Ohta et al. [933] to pursue pulsed-laser deposition of epitaxial AlN films at room temperature on nearly lattice-matched (Mn, Zn) Fe2O4(1 1 1) substrates. In situ RHEED observations have shown that the growth starts in the two-dimensional mode followed by a transition to the threedimensional mode when the film thickness reaches about 2 nm. The heterointerface between AlN and (Mn, Zn) Fe2O4 was determined to be abrupt and approximately 90% of the lattice mismatch was released at the interface owing to the introduction of crystalline defects such as misfit dislocations. Differences in thermal expansion coefficients of the epilayer material, in this case AlN, and the substrate lead to residual strain upon cooling down from growth temperature. Using Raman, Liu et al. [934] analyzed stress in the AlN/SiC (AlN was deposited by sublimation). The stress distribution in AlN crystals seeded on 6H-SiC was modeled using Equations 3.67 and 3.68 based on the sketch shown in Figure 3.184.
j641
j 3 Growth and Growth Methods for Nitride Semiconductors
642
R L AlN
t1 t2
W
6H-SiC y Figure 3.184 Sketch of AlN/6H-SiC structure after cooling down. Patterned after Ref. [934].
The thermal strain is expressed as ðl ei ¼ ½ðal ah Þ=ah ¼ ai dT:
ð3:67Þ
h
The stress along the y-axis is expressed as Fi E ti þ y : sy ¼ ti w R 2
ð3:68Þ
The theoretical frequency shifts of the E1 (transverse optical) mode calculated from model-predicted stress were in good agreement with experimental values taken along the edges of crystal samples. The stress was linearly distributed along the depth of the samples and changed from compressive at the growing surface to tensile at the interface between AlN and SiC for a thickness range of several hundred micrometers. Large tensile stresses, up to 0.6 GPa, were detected in the AlN at the interface. It is predicted that the AlN on 6H-SiC must be at least 2 mm thick to prevent it from cracking while cooling down the sample from a growth temperature of 2000 C. In addition to the thermal and mechanical, electrical, and optical properties discussed in Section 1.3, a good deal of investigations exists on the electronic manifestation of defects that are assumed to be similar to those observed for GaN to a first extent except the energy levels of course discussed in Sections 4.1.6 and 4.3. 3.5.14.1 Surface Reconstruction of AlN MBE of AlN, as in the case of other two main binaries, lends itself to the investigation of surface reconstruction examined by in situ RHEED. The investigation can also be aided by vacuum connected scanning tunneling microscopy (STM), LEED, and AES. The detailed atomic arrangements for most of the reconstructions on both (0 0 0 1) and ð0 0 0 1Þ surfaces have been determined by a combination of STM data and firstprinciples theoretical analysis, as discussed in Section 3.5.14.1. A unique feature of the GaN surfaces is that they can be terminated by excess cation species, that is, the Ga
3.5 The Art and Technology of Growth of Nitrides
atoms. Termination by N atoms is energetically unfavorable, because those N atoms prefer to form N2 followed by desorption from the surface. Against a large body of work on identifying surface arrangement of GaN, the AlN cousin has not received as much attention. Several studies have reported the symmetry of AlN surface reconstructions as seen by RHEED [935,936]. Symmetries of 1 1, 2 2, and 2 6 are commonly reported, and a sequence consisting of 1 3, 3 3, and 6 6 is reported by several groups. It should be noted that for RHEED a surface with hexagonal pffiffiffion p ffiffiffi symmetry, a 1 3 pattern usually is indicative of a 3 3 R30 symmetry. This comes about from the usual notation for RHEED from a surface with hexagonal symmetry, namely, listing the number of fractional-order streaks plus one for the integer-order streak observed with the electron beam along 2 1 1 0 and 0 1 10 directions, respectively. The latter is not a primitive lattice vector so that a pffiffiffi pffiffiffi 3 3 R30 reconstruction leads to a 1 3 RHEED pattern [937]. AlN films investigated by Lee et al. [937] were grown by PAMBE on approximately 2 mm thick OMVPE GaN templates. Depositing AlN on top of 0.5 mm thick GaN film by MBE on the OMVPE template led to rough surface morphology for AlN thicknesses greater than about 10 nm, presumably because of the lattice mismatch. To alleviate this problem, 5–10 periods of approximately 1 nm thick AlN layers followed by approximately 2 nm thick GaN layers, doped with Si, were deposited at 750–800 C, in an attempt to gradually grade the lattice constant to that of AlN. An undoped AlN layer with thickness of about 10 nm was then deposited on top, which yielded an overall flatter morphology. In a few instances AlN was directly deposited on SiC and a flat morphology was also achieved in that case (lattice mismatch of þ1.0% of AlN relative to SiC). Low Al coverage (as in N-rich growth conditions), moderate Al coverage (as in metal-rich growth conditions), and high Al coverage samples have been investigated. For obtaining low Al coverage or nitrogen-rich conditions, Lee et al. [937] heated an as-grown surface to 700–750 C and exposing it to the N plasma for a period of about 20 min. During this time, the 1 · 1 RHEED pattern of the surface stays streaky and sharp. Upon cooling down under the N plasma, a brightening of the RHEED pattern is seen when the surface passes through the melting point of Al (660 C). At that temperature, the plasma is turned off to avoid surface roughening. The resulting RHEED pattern is 1 · 1 with sharp streaks, which is referred to it as 1 · 1 nitrided by Lee et al. [937]. The Al content of such a surface can vary considerably as confirmed by AES (not shown). In some cases, this procedure of forming a 1 · 1 nitrided surface with low Al coverage in a 1 · 3 pattern, as shown in Figure 3.185, which is pffiffiffi presults ffiffiffi indicative of a 3 3 R30 surface reconstruction. This surface reconstruction can also be obtained from a 1 1 nitrided surface, with saturated N content, by careful Al deposition and annealing. pffiffiffi pffiffiffi It should be kept in mind that the range of Al coverages over which the 3 3 R30 surface is formed is quite narrow. A 2 2 N rich surface reconstruction has been occasionally observed when the MBE growth is performed under Al-poor conditions, that is, with reduced Al flux compared to the cases discussed above. A sharp, streaky 1 1 RHEED pattern is still found during growth. If the growth is terminated by closing the Al shutter followed by cooldown under the N-plasma, the 2 2 pattern forms. The 1 1 nitrided, 2 2, or
j643
j 3 Growth and Growth Methods for Nitride Semiconductors
644
Figure 3.185 RHEED patterns for low Al coverage samples displaying 1 · 3 pattern, along the (a) h0 1 1 0i azimuth and (b) along the h2 1 1 0i azimuth. The tick marks in (a) indicate the 3 spacing. Courtesy of R.M. Feenstra and Ref. [937].
pffiffiffi pffiffiffi 3 3R30 surfaces were not investigated by STM in detail. However, a few attempts at studying the 1 1 nitrided surface indicated a disordered surface arrangement. Figure 3.186 shows a terraced surface, in the moderate Al coverage category, consisting mainly of the long-period structure consisting of a hexagonal array of corrugation maxima whose separation is 25 1 nm. Some deviations and defects in the stacking arrangement of are evident. Moreover, another reconstruction can be seen on the terrace appearing at the lower left-hand corner of the image. A hexagonal array of corrugation maxima, with separation between maxima of 10 1 Å is found. The orientation of the short-period and long-period corrugation are identical in that they are not rotated by 30 relative to each other. The RHEED pattern during growth was 1 · 1 under Al-rich growth conditions and persisted during cooldown when the growth was terminated by simultaneously turning off the Al and N sources. However, on occasion, a 2 · 6 RHEED pattern appearing at a temperature of 500 C could result from this growth termination. This pattern can also be obtained by a postgrowth annealing at 800 C for about 10 min. Longer annealing results in a 1 · 1 pattern. Surfaces displaying the 2 · 2 pattern had surface reconstructions as seen by STM of the type shown in Figure 3.186. Typical 2 · 6 RHEED patterns with sixfold streaks along the h0 1 1 0i azimuth and half order streaks along the h2 1 1 0i azimuth are shown in Figure 3.187a and b. Additionally, a relatively intense satellite fringe for both azimuths was always seen at wave vectors larger than that of the first-order streaks, as indicated by the white arrows in Figure 3.187a and b. LEED investigations were performed for additional insight into the surface structure, the results of which are shown in Figure 3.187c. pffiffiffi The six first-order spots are apparent, with spacing relative to the origin of b ¼ 2=ðpffiffiffiffi 3ffiaÞ along h0 1 1 0i directions. At smaller wave vectors, additional spots with b=ð2 3Þspacing along both h2 1 1p0i h0ffiffiffi1 1 0i directions are visible. This arrangement is a result of ffiffiffi andp an underlying 2 3 2 3 R30 symmetry and leads to the conclusion that the basic symmetry of the surface reconstructions is rotated by 30 relative to the underlying 1 1 lattice. This observation helps clarify the symmetries of the structures observed inpSTM images. The 10 1 Å terrace spacing is consistent with either 3 3 or ffiffiffi p ffiffiffi 2 3 2 3R30 . However, the latter is the correct one based on the LEED.
3.5 The Art and Technology of Growth of Nitrides
Figure 3.186 STM images of a surface with multiple reconstructions, obtained with sample voltages of (a) þ3 V and (b) þ2 V after a background subtraction routine with which the background is formed by averaging the image over a window of 3 nm · 3 nm. The surface morphology consists mainly of five terraces,
pffiffiffi pffiffiffi labeled A–E. A region of 2 3 2 3 R30 structure is seen in terrace of pffiffiEffi andpremainder ffiffiffi the terraces consists of 5 3 5 3 R30 structure. Distortions of the stacking sequence are marked by the dashed lines and arrows. Courtesy of R.M. Feenstra and Ref. [937].
pffiffiffi pffiffiffi A pffiffi2ffi 3p ffiffiffi 2 3R30 surface periodicity produces a 2 6 pattern in RHEED. The 3 3 R30 periodicity produces a 1 3 RHEED pattern. For the long-period with a period of 25 3 Å, the surface structure is pffiffiffi pstructure ffiffiffi identified as 5 3 5 3 R30 , the genesis of which is probably because of a surface lattice that is contracted and slightly rotated with respect to the underlying AlN in such a way that the Al atom density in this surface lattice is close to the bulk Al. Additional splitting of the LEED spots is apparent in Figure 3.187c and are shown in an expanded view in Figure 3.187d, where the first-order LEED spot is labeled by A.
j645
j 3 Growth and Growth Methods for Nitride Semiconductors
646
Figure 3.187 Diffraction patterns from a moderately Al-rich AlN (0 0 0 1) surface. (a) Reverse-contrast RHEED pattern with electron beam along a h0 1 1 0i direction. Black tic marks indicate the location of 6 streaks. White arrows mark additional streaks seen at higher wavevectors. (b) Reverse-contrast RHEED pattern, with electron beam along a h2 1 1 0i
direction. White arrows mark the location of streaks occurring at high wavevector. (c) LEED pattern, acquired at 80 eV electron energy. The left-most first-order peak is slightly distorted due to a stray light reflection in the optical system used to acquire the image. (d) Expanded view of first-order reflection and satellite peaks. Courtesy of R.M. Feenstra and Ref. [937].
The satellite spots labeled B–F, oriented at approximately 30 intervals around a ring of radius (0.125 0.007)b surround the first-order spot. The radius of this ring agrees well with the observed spacing of the satellite fringes in the RHEED pattern of Figure 3.187b, which corresponds to a cut through the first-order spots of the LEED pattern and have a spacing (13 1)% larger than the first-order fringe spacing. Moreover, the inverse of this radius, 21.5 1.2 nm, is inpagreement with the spacing of the ffiffiffi corrugation lines seen in the STM images ð 3=2Þð25 3Þ nm ¼ 22 3 nm. The reconstructions were found to depend on the Al coverage, which can vary considerably because continued Al deposition at substrate temperatures below about 750 C produces thick and flat films on the surface. The Al content of our surfaces was estimated from the AES measurements. It should be pointed out that in the case of Ga on GaN where any excess surface Ga above about two ML condenses into droplets. For Al coverage of two to three monolayers (ML ¼ 1.19 · 1015 atoms cm2) a characteristic 2 · 6 RHEED pattern was observed. Invariant line intensities for the 2 · 6 patterns from surface to surface and across a given surface points to the
3.5 The Art and Technology of Growth of Nitrides
pffiffiffi pffiffiffi pffiffiffi pffiffiffi presence of a mixture of 2 3 2 3 R30 and 5 3 5 3 R30 phases that was seen in STM images. Guided by pdensity pseudopotential ffiffiffi pfunctional ffiffiffi pffiffiffi pffiffiffi calculations, Lee et al. [937] argued that the 2 3 2 3 R30 and the 5 3 5 3 R30 is probably incorporating a laterally contracted monolayer or bilayer structure containing a 4/3 ML Al layer that is contracted and rotated by 30 so that it fits the underlying 1 1 lattice. For Al films with thickness greater than a few ML, a characteristic 1 1 RHEED pattern was observed but with the diffraction streak spacing of about 6% greater than 1 1 AlN spacing. This implies a contraction of the lattice such that the Al atom density is close to that of bulk Al. For lower Al coverages additional reconstructions with symmetry 2 2 and 1 3 was also observed. During metal-rich growth conditions, high Al coverage occurs readily that can be simulated by depositing pure Al at a substrate temperature of approximately 150 C onto the 2 · 6 surface discussed above or onto a 1 · 1 nitrided surface about to be discussed [937]. Figure 3.188a shows an STM image of a surface that is covered by this type of thick (5 ML) Al layer. Unlike the case of moderate Al coverage, highmagnification STM images generally do not reveal any atomic corrugation on the surface, except for an occasional and very weak corrugation with an approximate 1· spacing can be seen, as in Figure 3.188b. This type of surface displays sharp firstorder RHEED, streaks, located at distinctly larger wave vectors than for a 1 · 1 AlN surface as illustrated in Figure 3.188c, which is referred to as 1 · 1-Al, with the observed streaks being located at wave vectors (6 1)% larger than that for the AlN 1 · 1 surface. Thermal annealing at temperatures above about 750 C for 5–10 min effectively removes the thick Al film from the surface but leaves micron-sized Al droplets [938]. The RHEED pattern of Figure 3.188d was acquired at the elevated temperature from such an annealed surface. With sufficient annealing time, the Al/N Auger ratio returns to a value close to that of the 2 · 6 structure and the RHEED pattern displays the 2 · 6 symmetry, as shown in Figure 3.188e. However, when the film was annealed for an insufficient time, a 2 · 6 structure resulted as shown in Figure 3.188f. The pattern displays streaks at the 1· positions of AlN and neighboring streaks located at wave vectors (10 1)% larger than the 1· spacing. The wave vector of these latter streaks is thus intermediate between those of the 1 · 1 Al pattern (Figure 3.188c) and those of the 2 · 6 pattern (Figure 3.188e or Figure 3.187b). LEED analyses were also attempted for high Al coverage sample in an attempt to determine the nature of the surface structure as was done for moderate coverage that did not prove to be straightforward in that LEED displayed a threefold splitting of the spots that may be attributed to nearly hexagonal (1 1 1) oriented Al layers. It should also be pointed out that LEED samples not only the surface layer but also a few subsurface layers. The occurrence of nearly hexagonal, (1 1 1) oriented Al layers in the film is consistent with the diffraction data in that a uniform (6 1)% contraction is observed in the first-order streak in RHEED. The observed LEED pattern probably arises from multiple domains of distorted Al(1 1 1) layers, and based on the similarity pffiffiffi pofffiffiffi their respective diffraction patterns, Lee et al. [937] propose that the 5 3 5 3 R30 structure may serve as a template for subsequent growth of the thin Al film. First-principles total energy calculations have been performed for a large number of possible AlN(0 0 0 1) surface reconstructions to gauge the likelihood of the surface
j647
j 3 Growth and Growth Methods for Nitride Semiconductors
648
Figure 3.188 (a) Large-scale STM image of AlN surface covered with an Al film obtained with a sample voltage ¼ 4 V and a gray scale range ¼ 12 Å. (b) Higher magnification view of the same surface acquired at a sample voltage ¼ 0.5 V and gray scale range ¼ 0.3 Å. (c–f) show the RHEED images along the h2 1 1 0i azimuth with (c) taken at 20 C (surface covered by 5 ML thick Al film) and (d) taken at
750 C (15 min anneal of the same surface as in (c)) and (e) taken at 20 C, following cooldown as the same surface as in (d); (f) taken at 20 C, following cooldown of a similar surface as in (d), which was annealed for only 10 min. White solid lines indicate the location of the first-order AlN streaks and dashed lines indicate the first-order streaks associated with the Al overlayer as in (c). Courtesy of R.M. Feenstra and Ref. [937].
reconstruction observed experimentally [937]. The calculations employ the local density functional theory and the electron–ion interaction is treated using firstprinciples pseudopotentials [939,940]. In previous studies of the AlN(0 0 0 1) surface, the focus was on structures having doubly occupied N dangling bonds and empty Al
3.5 The Art and Technology of Growth of Nitrides
dangling bonds and the guiding principle for the choice of structures was to satisfy the electron-counting rule (ECR) [941], which typically necessitates a 2 · 2 reconstruction on III–V (0 0 0 1) surfaces. In the light of the experimental observations discussed above, the theory pffiffiffi was pffiffiffi augmented and extended by performing calculations for structures having 3 3 symmetry and comparing the energies of these pffiffiffi with pffiffiffi the lowest energy 2 2 structures found previously. As discussed above, a 3 3 symmetry is observed experimentally in some cases. Moreover, this type of unit cell allows us to consider the laterally contracted metal adlayer structures that have been shown to be stable on GaN(0 0 0 1) surfaces under Ga-rich growth conditions [942]. As we are about to discuss below, metallic laterally contracted Al adlayer structures are found to be energetically favorable under Al-rich conditions in comparison to the standard 2 2 structures so that in this respect there is a similarity between AlN and GaN. The supercell employed in the calculations consists of four layers of Al and four layers of N. A layer of pseudo-hydrogen atoms having three fourth the charge was also employed to passivate the back side of the slab. The (0 0 0 1) surface on the opposite side of the H-terminated surface may be decorated by Al or N adatoms or even one or two pffiffiffi adlayers pffiffiffi of Al may be present. The plane wave cutoff was chosen at 50 Ry. For the 3 3 and 2 2 structures, a mesh of six special k-points is employed to sample the Brillouin zone. The relative energies are calculated as a function of the chemical potential of Al, which varies between a lower limit of mAl ¼ mAlðbulkÞ DH and an upper limit of mAl ¼ mAlðbulkÞ as discussed in the literature [941,943]. For AlN the formation energy from bulk Al and molecular N2 (at zero temperature) is DH¼ 3.3 eV [941]. The calculated energies for a subset of the structures considered in this work are shown in Figure 3.189. For a good portion of chemical potentials on the N-rich side of the phase diagram, the most stable structure is the 2 · 2 N adatom model, in which the adatom occupies an H3 site [941]. As the Al chemical potential is increased, the 2 · 2 Al T4 adatom structure becomes stable in a small region of the chemical potential space [941]. The 2 · 2 structure observed experimentally could in principle to either of these pffiffiffi correspond pffiffiffi possibilities. As for the Adatom structures having 3 3 symmetry, both the N-H3 and Al-T4 structures are energetically unfavorable for all conditions, as seen in Figure 3.189. Because these structures do not contain rest atoms, the electron-counting pffiffiffi pffiffiffi rule is not satisfied (as doing so would require subsurface donors). Thus, the 3 3 adatom are energetically unstable on impurity-free surfaces, including pffiffiffi structures pffiffiffi pffiffiAlN ffi p ffiffiffi the 3 3 Al vacancy structure and a number of other 3 3 structures. With increasing Al chemical potential, metallic structures with more than one ML of Al eventually become more pffiffiffi p ffiffiffi stable than those in the 2 · 2 Al adatom model. One such model is of 3 3 symmetry and contains 4/3 ML of Al above the ideal surface that could be described as a laterally contracted monolayer (LCM). pffiffiffiIn this model, the hexagonal lattice of Al withplattice vectors is reduced by a factor 3=2 and ffiffiffi pffiffiffi rotated by 30 to be in registry with a 3 3 cell of the substrate. This structure is stable with respect to the Al T4 adatom structure forpmffiffiffiAl >m 0:29 eV as shown pAlðbulkÞ ffiffiffi in Figure 3.189. A schematic representation of the 3 3 LCM model is shown in Figure 3.190.
j649
j 3 Growth and Growth Methods for Nitride Semiconductors
650
Figure pffiffiffi3.189 pffiffiffi Relative energies for several 2 · 2 and 3 3 surface reconstructions of the AlN (0 0 0 1) surface plotted as a function of the chemical potential of Al, which is relative to that of the 1 1 relaxed ideal surface. The zero of the Al chemical potential corresponds to bulk fcc Al. In the N-rich limit, where mAl is lower than the maximum value by 3.3 eV, the 2 · 2 N adatom model has the lowest surface energy. As the chemical potential is increased the 2 · 2 Al adatom model becomes lower in energy, and
pffiffiffi pffiffiffi then the 3 3 LCM pffiffiffistructure pffiffiffi is preferred. In the Al-rich limit, the 3 3 LCB structure is preferred by a small amount. The numbers in parentheses refer to the excess (or deficit) in the number of Al atoms relative to the ideal 1 1 surface, per 1 1 unit cell. Many possible models (not shown here) can be excluded on the basis of calculations performed in Ref. [941]. Courtesy of J. Northrup and R.M. Feenstra and Ref. [937].
In the high Al-coverage case, where mAl ¼ mAlðbulkÞ , it is slightly energetically favorable to add another layer of Al to the surface and form a laterally contracted bilayer structure with 7/3 ML ofpexcess ffiffiffi pAl, ffiffiffi the model structure of which is shown in Figure 3.190b and c. In each 3 3 cell, there are four atoms in the laterally contracted top layer and three atoms in T1 sites in the layer below. One of the atoms in the top layer is directly above an Al atom in the T1 layer below, whereas the other three being in bridge sites. The top layer then exhibits substantial corrugation: The atoms above the T1 sites reside approximately 0.53 Å above the other three atoms in the top layer. The 1 1 Al adlayer structures having one or two MLs of Al are tensile strained because the Al–Al spacing in the adlayer is too large, which can be illustrated with two examples. Compared to a 1 1 adlayer structure with one ML of Al in T1 sites, the Al–Al spacing in the 4/3 ML laterally contracted monolayer is reduced by pffiffiffi pffiffiffi structure 13% and the energy also is lowered by 0:59 eV= 3 3 cell in the Al-rich limit. This holds when comparing structures with two adlayers also. The 7/3 ML laterally contracted bilayer structure is more pffiffiffi than pffiffiffi stable a two ML structure, with adlayers in T1 and T4 registry, by 0:27 eV= 3 3 cell in the Al-rich limit. The calculations demonstrate that the 2 · 2 N-H3 adatom model is relatively stable over a substantial region of the chemical potential space and is, therefore, considered a plausible model for the 2 · 2 structure seen in N-rich conditions. The most likely
3.5 The Art and Technology of Growth of Nitrides
Figure 3.190 Schematic top views of the LCM and LCB models, respectively. Open and solid circles represent Al and N atoms, respectively with the diameter of the circles representative of the height of the atoms relative to the surface, and the layer numbers are depicted in parentheses. In each case, layer 1 is the outermost layer. (c) A 3D rendition of a side view of the LCB model where gray and black spheres
represent Al and N atoms, respectively. A projection slightly rotated from h0 1 1 0i is illustrates the vertical corrugation of the Al adlayers where the Al atoms in layer 2 are positioned directly above those in layer 3. The Al atoms in layer 1 that are positioned directly above those in layers 2 and 3 are higher than the other atoms in layer 1 by approximately 0.5 Å. Courtesy of J. Northrup and R.M. Feenstra and Ref. [937].
pffiffiffi pffiffiffi model to account for the observation of a 3 3 symmetry is the 4/3 ML laterally contracted monolayer. In very Al-rich conditions, the metallic adlayer structures having more than one ML of excess Al are energetically favorable and these adlayers are stabilized by a contraction offfiffiffi the Al–Al spacing to relieve tensile stress. pffiffiffi p Experimentally observed 2 3 2 3 R30 structure at this coverage may be formed from the 7/3 ML structure by an appropriate buckling of the surface atoms or
j651
j 3 Growth and Growth Methods for Nitride Semiconductors
652
alternatively by a more complex rearrangement. In any case, of either pffiffiffia 30 protation ffiffiffi the first or second layers (or both) seems likely.p As for the 5 3 5 3 R30 structure ffiffiffi pffiffiffi observed at Al coverage close to that of the 2 3 2 3 R30 , this structure arises from an approximately 4 rotation and approximately 9% contraction of the surface lattice of Al, which produces a surface Al–Al separation close to that of unstrained Al (1 1 1) planes. The unit cell of the resulting structure is too large for the first-principles theory to deal with. Finally, at higher Al coverages, the formation of Al layers with a contracted lattice compared pffiffiffi topAlN ffiffiffi is observed. The nature of this distortion appears to follow that of the 5 3 5 3 R30 structure, the net result being that the Al–Al separation in each plane of the film is close to bulk Al. To summarize, forpAl-poor 2 · 2 and 1 · 3 RHEED patterns, the latter may ffiffiffi pffiffisurfaces, ffi be associated with a 3 3 R30 reconstruction, pffiffiffi pffiffiffi are observed. For moderately Alpffiffiffi pffiffiffi rich surfaces, 2 3 2 3 R30 and 5 3 5 3 R30 reconstructions are observed. The former produces a 2 6 RHEED pattern and the latter then adds some satellite features and complex intensity variations (as discerned from diffraction spot splitting) in that pattern. For even larger Al surface coverage, flat films of epitaxial Al are found to form on the surface, with a characteristic 1 1-Al RHEED pattern with fringe spacing slightly expanded relative to the 1 1 AlN spacing and a LEED pattern with threefold spot pattern.
3.5.15 Growth of Ternary and Quaternary Alloys
By alloying InN together with GaN and AlN, the bandgap of the resulting alloy(s) can be increased from 0.8 to 6.2 eV, which is critical for making high-efficiency visible light sources, ultraviolet detectors, and high-performance electronic devices. The quaternary alloys (InGaAlN) have the added advantage that their lattice constants can be tuned, at least to ZnO, whereas its bandgap can be adjusted to meet the need set forth by the device design. Detection of missiles requires solar-blind detectors, meaning those that do not get affected by the suns radiation. Nitride-based ultraviolet detectors employing high fractions of AlN would not be blinded by the large infrared, visible, and UV components of the sunlight, but would be able to detect missiles by detecting their plumes. Incorporation of indium in these alloys is not easy, even though the InN molar fraction in InGaN has been pushed almost to 70%. To prevent InN dissociation, the InGaN crystal was originally grown by vacuum deposition at low temperatures (about 500 C). However, the use of a high nitrogen flux rate allowed the high-temperature (800 C) growth of high-quality InGaN and InGaAlN films on (0 0 0 1) sapphire substrates. It was noted that the incorporation of indium strongly depends on the flow rate ratio of TMI, TMA, and TMG (or TEG), with larger flow rates of TMG producing larger InN mole fractions. The indium incorporation efficiency decreases quite rapidly with increasing growth temperatures to values above 500 C. The crystalline quality of InGaN was observed to be superior when grown on the well-matched ZnO substrate to that grown on a bare (0 0 0 1) sapphire substrate. Employment of a buffer layer provides further improvements in the crystalline quality of InGaN. As in the case of GaN, InGaN films grown on
3.5 The Art and Technology of Growth of Nitrides
sapphire substrates with low-temperature GaN buffer layers exhibit much better optical properties. This has to do with the nucleation issues on the substrate. The pathways to deposition of high-quality AlGaN layers are similar to those for GaN with the exception that higher deposition temperatures can be employed. One concern is the reactivity of AIN and oxygen. The higher the temperature, the lower the incorporation of oxygen, the source of which is attributed to ammonia. Despite the above-mentioned complications, high-quality InGaN/AlGaN, InGaN/GaN, and InxGa1xN/InGal-yN superlattices have been grown. 3.5.15.1 Growth of AlGaN The AlxGa1xN alloy forms the barrier for all the GaN-based devices and as such is of paramount interest. In all the devices, the AlxGa1xN alloy constitutes a pivotal component and determines such properties as carrier and light confinement and sheet carrier density at heterointerfaces. With the advent of short wavelength sources and detectors, as discussed in Volume 3, Chapters 1–3, AlGaN is increasingly taking the role of active emission and absorption medium as well. The bandgap of this material, which is direct, lies in a typical range of 3.42 (x ¼ 0) to 6.0 (x ¼ 1) at T ¼ 300 K. Although a number of these devices have been reported, there are still serious issues to be addressed. Among them is the deteriorating radiative recombination efficiency with increasing Al content [944] and control of the mole fraction. The influence of growth conditions on the aluminum mole fraction and the quality of AlxGa1xN film have been studied in some detail [945,946], as will be expanded upon below. The MBE growth of AlxGa1xN on sapphire is less well understood, except for some regularities of the growth established for the MBE growth employing electron cyclotron resonance-microwave plasma source of nitrogen [947]. In this section, mainly the growth issues and structural properties of AlGaN are addressed, leaving the optical properties for Volume 2, Chapter 5. The OMVPE growth of AlGaN is often accompanied by parasitic gas-phase chemical reactions, which not only diminishes the group III deposition efficiency but also makes it difficult to control the alloy composition. Such chemical mechanisms proposed usually are initiated by the formation of adducts between the group III metalorganic and NH3. Mihopoulos [948] proposed a mechanism for AlGaN OMVPE growth that mainly focused on the eventual formation of dimer and trimer particles large enough to be kept from the surface by thermophoretic forces. This proposal sufficiently explained the reduction in AlN growth rate at typical OMVPE conditions and was confirmed by Creighton [949] who directly observed the formation of gasphase nanoparticles during AlN, GaN, and AlGaN OMVPE growth. Keeping the gas stream cool and/or allowing Al- and N-containing gases to be in contact only near the sample surface reduces this pregrowth interaction. Aluminum nitride films can be grown through the reaction of AlCl3 with NH3 or ammonolysis of AlCl3NH3 [950]. In this pyrolytic process, the reactant gases are supplied to the reaction zone either separately or in the form of the compound AlCl3NH3 in appropriate composition. In early varieties, vertical reactors and substrate temperatures in the range of 700–1300 C were used for pyrolytic decomposition of reactant species, which resulted in deposition of AlN films. Evaporation
j653
j 3 Growth and Growth Methods for Nitride Semiconductors
654
temperatures for AlCl3 and AlCl3 NH3 are 80–140 and 180–400 C, respectively, which are conveniently accessible. Deposition rates in the range of 0.6–6 mm h1 have been attained. Following the binary growth, AlxGa1xN (x ¼ 0.45) layers were grown on sapphire substrates at 1050 C by Baranov et al. [951]. A mixture of GaCl, AlCl, and ammonia was introduced to obtain the ternary material. The growth was initiated with GaN, which was then followed by the deposition of AlGaN. Simply raising the temperature of the Al source from 600 to 750 C, increased the Al content. All undoped AlGaN samples showed n-type conductivity with temperature-independent electron concentrations and mobilities between 77 and 300 K. With increasing Al content up to values of x ¼ 0 : 4 or greater, the electron concentration decreased from 5 · 1019 to 1 · 1019 cm3 and the mobility from 100 to 10 cm2 V1 s1. The results obtained thus far indicate that high-quality GaN films and templates can be grown on sapphire and other substrates by HVPE. These layers and templates served the basis for further growth on them by MBE and OMVPE. As such, the technique is very useful. The extent to which the method can be useful can be expanded if high-resistivity films can also be obtained. Owing to the high temperatures employed, it has not yet been possible to grow InGaN ternary with this method. Similarly, p-type doping has so far been lacking, which means that the HVPE technique alone is not in a position to produce device structures requiring p–n junctions. Although Si and O incorporation from the walls has been an issue, the high quality of the recent layers indicates that this issue is not as serious as it used to be. Traditionally, AlGaN layers by MBE were grown under N abundance. In that case, relative fluxes of Ga and Al would determine the mole fraction. For GaN, however, Ga abundant, in which case the N flux determines the growth rate, and N abundant, in which case the Ga flux determines the growth rate, conditions have been applied. Attempts to control the Al mole fraction in AlGaN were undertaken by varying the Al cell temperature (thus the Ga flux), Ga cell temperature, both Ga and Al fluxes with the fixed Al/Ga ratio, and Al/Ga ratio with the fixed total flux. However, a new realm of metal-rich growth of the AlGaN ternary has been applied that resulted in AlGaN with high radiative efficiency over the entire composition range. He et al. [952] reported on growth and properties of AlxGa1xN epilayers deposited on (0 0 0 1) sapphire substrates by MBE under Ga-rich conditions for a wide range of aluminum compositions (0.13–0.92). Some details are as follows: An overabundant Ga flux (two Ga sources) and moderate N pressure were kept to maintain N-limited (Ga-rich) growth conditions. The Al mole fraction in the alloy was controlled by fixing the Al and Ga source temperatures and varying the N flow. Variation of the nitrogen partial pressure in the MBE chamber in the range of 2.7 · 106 to 9 · 106 Torr during growth provided AlxGa1xN with almost a complete range of x. Care, however, must be exercised because of the temperature-dependent Ga desorption rate from the surface. This means that at higher temperatures, the AlN component in AlGaN will increase even when the cell temperatures and N flux are kept the same, necessitating a good control over the substrate temperature. Shown in Figure 3.191, the dependence of Al mole fraction in AlxGa1xN epilayers grown under Ga-rich conditions at constant substrate temperature versus the MBE
3.5 The Art and Technology of Growth of Nitrides
Aluminum mole fraction, x
1.0
0.8
0.6
0.4
0.2
–6
2.0 x 10
–6
4.0 x 10
–6
6.0 x 10
–6
8.0 x 10
–5
1.0 x 10
Pressure (Torr) Figure 3.191 Aluminum mole fraction x in AlxGa1x N alloys as a function of nitrogen plasma pressure under metal-rich condition.
chamber pressure, which represents the N flux. The Al mole fraction is about inversely proportional to the N flow and the Al concentration increases rapidly when the pressure is lower than 4 · 106 Torr. This phenomenon can be explained by preferential formation of the AlN component in the AlxGa1xN alloy under Ga-rich growth conditions [947]. Indeed, because the Al–N bond is stronger than the Ga–N bond, it is reasonable to surmise that the available N would preferentially be used to form AlN. The remaining N would then form the GaN bonds for the alloy. Consequently, the N arrival rate would determine the mole fraction in the presence of an abundant amount of Ga on the surface in some range of substrate temperatures. As expected, under these conditions, the Al mole fraction in AlxGa1xN remained the same even with increasing temperature of the gallium sources. Keeping the pressure of nitrogen plasma fixed, the aluminum mole fraction in AlxGa1xN is a function of aluminum flux, or the source temperature, and it could not be changed when the temperature of Ga sources was varied as shown Figure 3.192. Implicit in Figure 3.192 is that the higher the aluminum source temperature or flux, the higher the aluminum mole fraction x in AlxGa1xN alloys. This can be inferred from observing that the Al mole fraction is increased by reducing the N flux, which leads to reduced contribution from Ga, and thus the growth rate diminishes. PL intensity for the samples grown under Ga-rich conditions generally quenched at higher sample temperatures, as compared to the N-rich samples. Quantum efficiency of PL at 15 K was markedly higher for the AlxGa1xN layers grown under Ga-rich conditions (3–48%) compared to the layers grown under N-rich conditions (1–10%). These values in turn are much higher than radiative efficiencies obtained by us for a large set of GaN layers (Ga polarity) grown in similar conditions: 0.01–0.3% for Ga-rich and 0.1–2% for N-rich GaN. Such improvement of radiative efficiency in AlxGa1xN, especially under the Ga-rich conditions points to reduced density of dislocations, although carrier confinement cannot be excluded.
j655
j 3 Growth and Growth Methods for Nitride Semiconductors
656
0.5
Growth rate (Πm h–1)
0.4
0.3
0.2
0.1
0.0 0.0
0.2
0.4
0.6
0.8
1.0
Aluminum mole fraction, x Figure 3.192 Growth rate of AlxGa1x N alloys as a function of mole fraction and indirectly (in conjunction with Figure 3.191) nitrogen-plasma pressure under Ga-rich conditions. The growth conditions are identical to those used for Figure 3.191.
TEM was used to characterize defects and the microstructure of AlGaN layers grown under N- and Ga-rich conditions [953]. The AlxGa1xN layers investigated had nominal thicknesses in the range of 0.5–1 mm, Al mole fraction x in the range of 0.10–0.25, and were grown on (0 0 0 1) sapphire substrates by MBE. The dominant defects present in films grown under N-rich conditions were threading dislocations as determined from multibeam imaging conditions. The TEM bright-field images of Figure 3.193a–c associated with a sample of N-rich conditions show the density of dislocations in the vicinity of the substrate to decrease from 5 · 1010 cm2 at the interface with the substrate to 2 · 1010 cm2 at the surface. To estimate the relative distribution of different types of dislocations (edge, screw, and mixed), bright-filed images were recorded under two beam conditions for a g-vector parallel to the [0 0 0 2] and ½1 1 2 0 directions, respectively. A pair of such images is shown in Figure 3.193b and c. In the image recorded for a g-vector parallel to the [0 0 0 2] direction, the edge dislocations are out of contrast, whereas the screw dislocations are out of contrast in the image recorded for a g-vector parallel to the ½1 1 2 0 direction. However, mixed-type dislocations appear in both images. In the case of layers grown under N-rich conditions, the majority (95%) of all dislocations were out of contrast in images recorded for a g-vector parallel to the [0 0 0 2] direction meaning that they were pure edge type. The microstructure and defect distribution in layers grown under Ga-rich conditions differed significantly. First, these layers had a rough surface compared to layers grown under N-rich conditions (see Figure 3.193d). Moreover, despite similar densities of threading dislocations in layers grown under Ga-rich conditions
3.5 The Art and Technology of Growth of Nitrides
Figure 3.193 TEM Bright-field images of layer grown under Ga-rich (left panels)/N-rich (right panels) conditions recorded under (a)/(d) multibeam condition, (b)/(e) two-beam condition with g-vector parallel to [0 0 0 2] direction, and (c)/(f) two-beam condition with g-vector parallel to ð1 0 1 0Þ direction.
(7 · 1010–8 · 1010 cm2 in the vicinity of the substrate and 5 · 109 cm2 at the layer surface) to those measured for layers grown under N-rich conditions, there was a significant difference in their type distribution. In layers grown under Ga-rich conditions, edge dislocations are again the majority (70%), although other dislocations are also in significant densities (see Figure 3.193e and f). Iliopoulos et al. [954] reported the growth kinetics of AlGaN films by plasmaassisted MBE on (0 0 0 1) sapphire substrates at 750–800 C. The incorporation probabilities of Al and Ga were determined by measuring the growth rate and composition of the films. In both the N-rich and Ga-rich conditions, the incorporation probability of Al is unity for the entire investigated temperature range. The incorporation probability of Ga kept constant and equaled 0.75 at a substrate temperature of 750 C only under the N-rich growth condition. The temperature dependence of the incorporation probability in this regime has an activation energy of 2.88 eV, which is consistent with Ga desorption from the surface. In the group III rich growth regime, the incorporation probability of Ga decreases monotonically with group III fluxes owing to the competition with Al for the available active nitrogen atoms. In this regime, the GaN phase growth rate was determined by the capture probability of the available active nitrogen from the gallium surface adlayer.
j657
j 3 Growth and Growth Methods for Nitride Semiconductors
658
As is the case for all ternaries and quaternaries in all compound semiconductors, high-quality layers containing large fractions of Al or In are difficult to grow, which is particularly so in AlGaN. Although the crystalline quality of AlGaN on sapphire can be improved by using a LT insertion layer, the quality progressively worsens with increasing AlN molar fraction [955]. However, the crystalline quality can be significantly improved if the AlGaN layer is grown on a high-quality GaN layer. However, one has to be mindful of the high-density crack network, which originates from the tensile stress induced by the lattice mismatch between AlGaN and GaN when the thickness of AlGaN exceeds a critical value [956]. The growth can be interrupted after some thickness beyond the critical thickness, the sample be allowed to cool and crack by plastic relaxation, and growth is then resumed [413]. Cracking of AlGaN layers for higher compositions and for thicker layers is a serious problem. Recall, the origin of dislocations in AlGaN is the lattice misfit and the different thermal expansion coefficients between GaN and AlN, which result in stress in AlGaN epilayers. Einfeldt et al. [957] studied the relaxation of tensile strain in AlGaN layers of different compositions grown on GaN/sapphire. Extended crack channels along the h2 1 1 0i directions are formed, as shown in Figure 3.194, if the Al content exceeds a critical value that decreases with increasing layer thickness. The surface gap at the crack was found to increase linearly with stress. Annealing of the AlGaN sample above the growth temperature introduced additional tensile stress upon cooldown owing to the mismatch in thermal expansion coefficients. This tensile stress in AlGaN layers was relieved not only by the formation of additional cracks but also by the extension of these cracks. Other authors also studied the generation of dislocations and cracks in AlGaN epilayers [958,959]. In one such case, high-resolution X-ray diffractometry was employed to investigate the mechanism of stress reduction [960]. It was found that the tensile stress decreased with decreasing interlayer growth temperature, as shown in Figure 3.195. From reciprocal space maps it could be observed that the AlN interlayers grown at high temperatures were pseudomorphic, whereas interlayers grown at low temperatures were strain relaxed. Therefore, the AlGaN layers grown on a low-temperature AlN
δo
b t w
(a)
(b)
(c)
Figure 3.194 Artistic view of various cracking modes of tensilestressed AlGaN film. The crack type (a) propagates toward the interface; (b) channels across the interface; and (c) along the plane of the substrate as if to peel the layer. Patterned after Ref. [957].
3.5 The Art and Technology of Growth of Nitrides
1.4 1.2
Tensile stress (GPa)
1.0 0.8 0.6 0.4 0.2 0.0 500
600
700 800 900 1000 AlN growth temperature(ºC)
1100
Figure 3.195 Room temperature average tensile stress in 1–3 mm GaN layers grown on Si(1 1 1) substrates determined from the in-plane a-lattice constant by high resolution X-ray diffraction measurements. The only LT AlN interlayer was inserted at indicated growth temperatures following a 0.5 mm GaN [960].
grow under compressive interlayer-induced strain. The reduction of mismatchinduced tensile stress and suppression of crack formation during growth of AlxGa1xN was also realized by inserting a low-temperature AlGaN interlayer between GaN and AlxGa1xN [961]. The interlayer was found to mediate the elastic tensile mismatch between the adjacent layers and extend the critical thickness before the onset of cracking. The use of LT insertion layers (interlayers) and their effect on overlaying GaN is discussed in Section 3.5.5.5. In addition to the stress and associated cracking issue, another major problem with AlGaN is the presence of a large density of defects that play a detrimental role in the performance and reliability of AlGaN-based devices [962–964]. Thick, heavily doped and high-quality AlGaN layers are essential to developing GaN-based devices. For dislocation reduction, several approaches centered on buffer and interlayers are employed [965]. Amano et al. and others [966–969] reported the growth of low dislocation density AlGaN on GaN underlayers combining the low-temperature AlN interlayer and the trenched substrates technique. But for deep UV LEDs, GaN buffers greatly decrease the light extraction efficiency from the substrate side and does not allow light propagation in back illuminated detectors. Another approach relying on AlN/AlGaN superlattices, which is common in the GaAs-based systems, has been employed for reducing the threading dislocation density and growing thick AlGaN on sapphire [970]. Doing so allowed, crack free 3 mm thick AlxGa1xN layers to be achieved. A thin AlN interlayer grown at low temperatures, LT-interlayer, is often incorporated to attain crack-free AlGaN layers on GaN on sapphire [971]. Inserting the LT interlayer reduces the tensile stress during growth and the also the number of threading dislocations with a screw component. An added benefit of LT interlayer, as in the case of GaN growth, is that the quality of AlGaN also improved with LT buffer
j659
j 3 Growth and Growth Methods for Nitride Semiconductors
660
insertion as demonstrated by Amano et al. [972]. Although the crystalline quality of AlGaN on sapphire improves by using an LT buffer layer, it progressively degrades with increasing AlN content. When the underlying GaN is improved, so did the AlGaN quality, but a network of high-density cracks caused by the residual thermal tensile stress is generated when the thickness of AlGaN exceeds a critical value. This problem is alleviated by LT AlN insertion. Amano et al. [972] inserted a LT AlN between the underlying GaN layer and the upper AlGaN layer. The LT interlayer reduces the tensile stress and threading dislocations that have screw components. Figure 3.196 shows the difference of the grown-in stress of the Al0.18Ga0.82N on GaN (a) with and (b) without the LT AlN interlayer. Nearly strain-free AlGaN could be grown on the LT interlayer, whereas relaxation occurs during a growth of Al0.18Ga0.82N, which is confirmed by the steep decrease of the stress and thickness product in Figure 3.196. It is important to emphasize that the crystalline quality of 500
Refraction intensity
Al0.18Ga0.82N
1
0.05(GPa)
300
0 200
100
–1 1.4 μm
1.7 μm
Stress×thickness(GPa×µm)
LT-AlN
GaN 400
0 3000
(a)
4000
5000
6000
7000
Growth time (s)
Al0.18Ga0.82N
GaN
LT-AlN insersion layer LT-AlN nucleation layer
Sapphire (b) Figure 3.196 (a) In situ measured stress and thickness product and reflectivity for AlGaN/ GaN heterostructure grown on sapphire with LT insertion layer. (b) Schematic diagram of the heterostructure with the insertion layer. (c) In situ measured stress and thickness product and
reflectivity for AlGaN/GaN heterostructure grown on sapphire without the insertion layer. (d) Schematic diagram of the heterostructure without the insertion layer. Courtesy of H. Amano and Ref. [972].
3.5 The Art and Technology of Growth of Nitrides
Al0.18Ga0.82N
GaN 1.6 (GPa)
1.15 (GPa)
Refraction intensity
400
1
300 0
0.62(GPa)
200
100
–1 1.3 µm
1.4 µm
0
3000
(c)
Stress×thickness (GPa× µm)
500
4000 Growth time (s)
5000
Al0.18Ga0.82N GaN
LT-AlN nucleation layer
Sapphire (d) Figure 3.196 (Continued)
this AlGaN is much superior to that grown on sapphire covered with only one LT buffer layer directly on sapphire, as confirmed by TEM. The reduction of the density of screw and mixed threading dislocations leads to a reduction in the leakage current in solar-blind UV photoconductors and pin photodiodes confirming that these defects act as a current leakage path. The fact that one LT interlayer is effective paved the way for the use of periodically spaced LT insertion layers [434,973]. The LT interlayer process was also used for a distributed Bragg reflector (DBR) mirror based on an AlGaN/GaN multilayered structure. For an expansive treatment of the topic, se Ref. [974]. The disadvantage of the LT interlayer technique is that it increases the density of edge dislocations, which also act as nonradiative recombination centers, as discussed in Volume 1, Chapter 4 and Volume 2, Chapter 5. Therefore, the fabrication of highly luminescent AlGaN is difficult. The lateral epitaxial overgrowth used to reduce the dislocation density, discussed in Section 3.5.5.2, cannot simply be adopted in the growth of AlGaN, particularly with a high AlN molar fraction, because of the deposition of polycrystalline islands on the dielectric mask. Another method of growing low dislocation density AlGaN is to use grooved GaN [972]. However, LT interlayers are still necessary to alleviate the cracking problem.
j661
j 3 Growth and Growth Methods for Nitride Semiconductors
662
The best overall results are obtained when the LT interlayer in used in conjunction with grooved templates underneath, as shown in Figure 3.197a and b. It should be emphasized that the LT interlayer is essential for alleviating fracture of the layer. Figure 3.197c shows the cross-sectional TEM image of the AlGaN layer grown on grooved GaN with LT AlN interlayer. Figure 3.198 shows the low-temperature CL image of an Al0.25Ga0.75N layer grown on grooved GaN covered with the LT interlayer. Several dark spots are visible in the grooved region, whereas it is entirely dark in the terraced regions, which implies that edge dislocations present in the terraced region act as nonradiative recombination centers. The upper limit for the density of the dark spots on the grooved region is around 107 cm2. Therefore, a reduction in the threading dislocation density by as many as two orders of magnitude can be achieved. Amano [974] also reported that the efficiency of the AlGaN/GaN MQW with low dislocation density on grooved and LT interlayers is comparable to that of the GaN/GaInN MQW. Without the grooved LT interlayer employment, the AlGaN/GaN variety is much inferior, a point that was reflected in the performance of UV LEDs. The growth processes for grooved and LT interlayer concepts are rather complex because etching and regrowth are unavoidable. In addition, the underlying GaN layer is necessary for growth but acts as an absorption layer for UV emission, which is not desirable. To alleviate the absorption by the GaN template, a process that utilizes a grooved substrate has been proposed [975–977]. In this process, grooves are initially formed on the surface of sapphire, SiC, or Si substrates. AlGaN is then grown utilizing an LTnucleation buffer layer of AlN. The grooves should be sufficiently deep so that the laterally grown AlGaN initiating on the terraces will coalesce and bridge over the grooved region before the AlGaN initiated on the troughs reaches the laterally grown AlGaN. The practical advantages of this process is that it does not require etching or regrowth after the growth of GaN. As alluded to above, the absence of a GaN layer eliminates absorption of the UV emission. The thermal stress caused by the differences in the thermal expansion coefficients between AlGaN and the substrate used can also be reduced. To gain some insight into the behavior of threading dislocations and the mechanism of stress relief, cross-sectional TEM observations were conducted for GaN/ AlGaN/IL-AlN/GaN [978]. Figure 3.199a and b shows dark-field (DF) images of a specimen with GaN layer thickness of 580 nm, taken with the reflection of g||0002 and ½0 1 1 0, respectively. Using the well-known selection rule that indicates that screw dislocations are out of contrast for g b ¼ 0 diffraction conditions, and in contrast for g ¼ (0 0 0 2) diffraction conditions, Burgers vector b can be determined. Here g is the diffraction vector. In the image recorded for g-vector parallel to the [0 0 0 2] direction, the edge dislocations are out of contrast, whereas the screw dislocations are out of contrast in the image recorded for a g-vector parallel to the ½0 1 1 0 direction, see Section 4.1.1.2 for a description. TDs appearing in Figure 3.199a are of screw and mixed type, and those in Figure 3.199b are of edge type or mixed type. Because Figure 3.199b happened to be taken with the ð0 1 1 0Þ reflection, there can be hidden TDs that disappear from Figure 3.199a and b. Because most of the TDs in Figure 3.199a appear also in Figure 3.199b, they are identified to be mixed type,
3.5 The Art and Technology of Growth of Nitrides
Figure 3.197 (a) SEM image and the schematic structure of Al0.19Ga0.81N on grooved GaN with low temperature deposited AlN interlayer. (b) SEM image and the schematic structure of Al0.19Ga0.81N on grooved GaN without low temperature deposited AlN interlayer. (c) Cross-sectional TEM image of the sample shown in (a). Courtesy of H. Amano.
j663
j 3 Growth and Growth Methods for Nitride Semiconductors
664
Figure 3.198 CL image, taken at cryogenic temperature, of Al0.25Ga0.75N grown on grooved GaN. Courtesy of H. Amano and Ref. [972].
whereas those that appear only in Figure 3.199b are of edge type. One can see that TDs that are formed at the interface of GaN/LT GaN are mostly of mixed type with a few of edge type. The density of TDs in the GaN layer was roughly estimated to be of the order of 108 cm 2. The TDs of mixed type penetrate the AlGaN layer and propagate up to the top surface. This indicates that the IL-AlN has no effect in the termination of mixed type TDs. Figure 3.199b clearly indicates that many edge type TDs are generated at the interface of IL-AlN and propagate into the AlGaN layer. The mechanism of the preferential formation of edge type TDs at the IL-AlN is not yet clear. The density of the TDs is estimated to be above 109 cm 2 in the AlGaN layer but decreases sharply in the top GaN layer. Al segregation was observed around the threading dislocations in Al0.1Ga0.9N and Al0.3Ga0.7N layers grown on 6H-SiC by OMVPE by TEM [979]. Up to 70% more Al concentration was found around dislocations than those regions free of dislocations. The Al-depleted regions were observed to be within a few nanometers of the dislocation lines. Pecz et al. [980] also reported the segregated Al in the middle of each V pit that is on the surface of AlGaN grown on GaN by OMVPE. The origin and morphological evolution of a self-terminating V-shaped defect in AlGaN/GaN superlattices was studied by TEM [981]. These V-shaped defects, which generally originated from pits on the top plane of the undoped GaN layers with a critical thickness below 500 nm and deposited with inclined superlattice structure during growth of superlattice, could terminate in the superlattice region after propagating to a distance of about 800–1000 nm. The structural variations of the defects could be attributed to the different growth rates between the surrounding matrix and the domains within the defects at various growth stages. Intentional impurities provided in high concentrations are known to help reduce the structural and point defects, as has been demonstrated in GaAs and GaN using, for example, Si and to some extent isoelectronic species such as In in low concentrations as to still maintain the binary nature of GaN. However, the case with Mg is
3.5 The Art and Technology of Growth of Nitrides
Figure 3.199 Dark-field TEM images of GaN (580 nm)/ Al028Ga0.72N/IL-AlN/GaN/LT GaN/sapphire for (a) g ¼ 0 0 0 2 and (b) g==0 1 1 0. Courtesy of H. Amano and Ref. [978].
complicated in that low concentrations follow the aforementioned trends, whereas the high concentrations introduce structural defects dubbed the pyramidal or V defects as discussed in Section 4.2.3. In this vein, while leaving the details to the aforementioned section, Cho et al. [982] reported the influence of Mg doping on structural defects in Al0.13Ga0.87N layers grown on sapphire substrates by OMVPE. The reduction of dislocation density occurred up to the Mg precursor flow rate of 0.103 mmol min1. However, the reduction from 4 · 109 cm2 (undoped) to 1.3 · 109 cm2 (moderately Mg doped) is not very significant and may therefore not represent a true trend. The inversion domain boundaries present in the films were observed to have horizontally multifaceted shapes. The vertical type inversion domain boundaries were also observed in Al0.13Ga0.87N grown with low Mg precursor flow rate. In addition, isoelectronic In doping was also used to improve the crystallinity of AlGaN. Nakamura et al. [983] reported the structural analysis of
j665
j 3 Growth and Growth Methods for Nitride Semiconductors
666
Si-doped AlGaN/GaN multiquantum wells. The effect of Si doping on the structural properties of AlGaN/GaN multiquantum well layers grown on GaN was investigated. V-shaped defects were formed by Si doping. PL intensity gradually increased with Si doping. But when the Si concentration exceeded 4.2 · 1019 cm3, PL intensity rapidly decreased with the formation of V-shaped defect. See Section 4.2.3 for description of V defects. For UV emitters and detectors, highly conductive n-type AlGaN alloys with high Al contents are indispensable. But, AlxGa1xN alloys with high x are very difficult to grow. For silicon-doped AlxGa1xN, the conductivity decreases by six orders of magnitude from x ¼ 0.2 to x ¼ 1. Undoped AlxGa1xN alloys with high x (x > 0.4) generally have high resistivity, possibly because of the sharp increase of the carrier localization energy around x ¼ 0.4. Although n-type AlxGa1xN (x ¼ 0.58, conductivity ¼ 0.08 O1 cm1) with Si doping and n-type AlxGa1xN (conductivity 5 O1 cm1) with indium–silicon codoping were realized [984,985], highly conductive n-type AlxGa1xN alloys with Al contents as high as 0.6–0.7 are still needed. Nam et al. [986] reported the OMVPE growth of highly conductive n-type Al0.65Ga0.35N with a conductivity of 6.7 O1 cm1 (free electron concentration 2.1 · 1018 cm3 and mobility of 20 cm2 V1 s1 at room temperature). Their experimental results also revealed that the conductivity of AlxGa1xN alloys continuously increases with an increase in Si doping level for a fixed Al content value and that a critical Si doping level of about 1 · 1018 cm3 is needed to render insulating AlxGa1xN (x > 0.4) n-type. Hall-effect measurements show that the activation energy of the Si donor increases linearly from 0.02 eV in GaN to 0.32 eV in AlN [987]. Yoshida et al. [988] noted a decrease in the free carrier concentration from 1020 cm3 in GaN to 1017 cm3 for AlGaN with x ¼ 0.3 in nominally undoped films grown by MBE. Steude et al. [989] obtained the localization energy Eloc ¼ 0.08 eV and the donorbinding energy ED ¼ 0.4 eV from Eloc ¼ aED, which showed that in high-quality undoped AlGaN films with room temperature free carrier concentrations below 1017 cm3, Si is residual donor. 3.5.15.1.1 Growth of p-Type AlGaN In p–n junction devices such as LEDs and lasers, p-type layers are imperative. Although the device design in HBTs and detectors could be manipulated to rely only on p-type GaN, this additional restriction is not welcome. This being the cases, efforts have been undertaken to produce p-type AlGaN. Although applicable to all ternaries in general, to reduce the parasitic reactions between metalorganic sources, particularly with TMA and NH3, a dualflow channel reactor was employed [990]. For activating Mg in AlGaN, flash-lamp annealing under atmospheric pressure nitrogen flow at 1140 C for 60 s was used. One problem with p-type nitrides, progressively getting worse as the AlN mole fraction increases, is the poor ohmic contact, which makes measurements difficult and less accurate. Measured temperature dependencies of the hole concentration indicate that the hole concentration drops off with decreasing temperature from about 1018 cm3 at 500 K to about 1013 cm3 at 125 K. Follow-up studies of Mg-doped AlxGa1xN have resulted in p-type conductivity for Al mole fractions up to 27% [991–995]. Essentially, the binding energy of Mg progressively gets larger leading to lower and lower hole
3.5 The Art and Technology of Growth of Nitrides
0.38
10 19
0.36 10
Mg activation energy, EA (eV)
0.32
P(cm –3)
0.34
x=0.27 EA=0.31eV
17
10 15
0.30 0.28
10 13
2
0.26
3
4 5 1000/T(K –1)
6
0.24 0.22 0.20 0.18 0.16
GaN
0.14 0.0
0.1 0.2 Al content, x
Figure 3.200 Activation energies of Mg acceptors in Mg-doped p-type AlxGa1xN as a function of Al content x. Solid squares, solid and hollow circles, and triangles are data from Refs [993–995], respectively, all obtained by Hall measurements. Open circles indicate data
0.3 obtained by PL measurements from Ref [995]. The inset shows measured temperature dependence of Hall concentration p in the Mgdoped p-type Al0.27Ga0.73N sample from which EA ¼ 0.310 eV was obtained [995].
concentrations as the Al mole fraction is increased, making it difficult for device applications requiring p-doping and large mole fractions. Among them are the UV detectors and lasers. Figure 3.200 shows activation energies of Mg acceptors in Mgdoped p-type AlxGa1xN as a function of Al content x. For x ¼ 0.27, activation energy increases from 150 meV in GaN to 310 meV in Al0.27Ga0.73N as determined by variable temperature Hall measurements that include screening. From the measured EA versus x in Mg-doped p-type AlxGa1xN, the resistivity versus x can be estimated as follows [994]: rðAlx Ga1 x N Þ
¼ r0 exp½E A ðAlx Ga1 x N Þ=kT ; ¼ r0 expf½ðE A ðGaNÞ þ DE A =kT Þg; ¼ rðGaNÞÞexpðDE A =kT Þ;
ð3:69Þ
where DE A ¼ E A ðAlx Ga1 x N Þ E A ðGaNÞ. A typical p-type GaN has a resistivity of about 1.0 W-cm, which, using Equation 3.69, leads to an activation energy and a resistivity of 0.4 eV and 2.2 104 W-cm, respectively, for the Al0.45Ga0.55N alloy, which
j667
j 3 Growth and Growth Methods for Nitride Semiconductors
668
is needed for solar-blind detectors. Obviously, the problem of high resistivity is of paramount concern for solar-blind detectors and deeper UV emitters. 3.5.15.1.2 Ordering in AlGaN Because the misfit between AlN and GaN is only 2.5%, one would expect a stable growth of AlGaN with good control of composition. However, the atomic long-range ordering along the c-axis had also been observed in the form of AlN/GaN (1 : 1), Al0.25Ga0.75N (3 : 1) and Al0.16Ga0.84N (10 : 2) [996]. These three chemical orderings were found to coexist in the same layers and lead to inhomogeneous composition in AlGaN films. Korakakis et al. [997] first reported the observation of long-range order in AlGaN films grown by MBE on sapphire and 6H-SiC. In XRD studies of AlGaN films grown under nitrogen-rich conditions, other than the (0 0 0 2) and (0 0 0 4) diffraction peaks expected in hexagonal AlGaN, they observed the normally forbidden diffraction peaks (0 0 0 1), (0 0 0 3), and (0 0 0 5), as shown in Figure 3.201. The presence of such normally forbidden diffraction peaks indicates the long-term atomic ordering existing in AlGaN. AlGaN films with different Al contents were investigated, the relative intensity of these peaks was found to be largest for Al content in the 30–50% range, in qualitative agreement with expectations for an ordered structure of ideal Al0.5Ga0.5N stoichiometry. The maximum ordering was found to occur in Al0.5Ga0.5N, just as expected. The average size of the ordered domains in the films was found to be within a factor of 4 of the films thickness. Iliopoulos et al. [947] found that the ratio of group III and group V fluxes influences the relative incorporation of gallium and aluminum in AlGaN. Different types of spontaneously formed superlattice structures with periodicities of 2, 7, and 12 ML were found. Figure 3.202 is a cross-sectional TEM image with selected area diffrac-
Al0.25Ga0.75N
Al0.25Ga 0.75N ×1 (0 0 0 2)
Intensity (au)
×30 (0 0 0 1)
GaN x1 (0 0 0 2)
17.0
17.5
34.5
35.0
2θ (degrees) Figure 3.201 y-2y XRD scan of a bilayer GaN–Al0.25Ga0.75N grown on c-plane sapphire. Courtesy of T.D. Moustakas and Ref. [997].
3.5 The Art and Technology of Growth of Nitrides
Figure 3.202 Cross-sectional TEM and SAD pattern of an Al0.89Ga0.11N film grown under group III-rich condition. Courtesy of T.D. Moustakas and Ref. [947].
tion (SAD) pattern of an Al0.89Ga0.11N film grown under group III rich condition. The SAD indicated a more complex superlattice than that reported earlier by the same group [997]. Figure 3.203 is an on-axis y 2y XRD spectra of three AlGaN films with different III/V ratios and Al mole fraction, but other growth conditions were similar. AlGaN (AlN mole fraction is 55 2% by RBS and 45.5% by XRD) grown under N-rich condition (III/V ¼ 0.6) showed a long range atomic order consisting of a superlattice
Figure 3.203 On-axis XRD spectra of the three AlGaN films grown with different III/V ratios and Al mole fractions, 1, 0.9, and 0.6 for samples A (89% Al mole fraction), B (74% Al mole fraction) and C (55% Al mole fraction), respectively. Courtesy of T.D. Moustakas and Ref. [947].
j669
j 3 Growth and Growth Methods for Nitride Semiconductors
670
with 2 ML periodicity reported previously [997]. AlGaN films grown under group III rich conditions showed a number of additional peaks, suggesting superlattice structures. By fitting the data to Voight functions, it was concluded that AlGaN grown under group III rich conditions has two intermixed superstructures with different periodicities of 7 and 12 ML. The origin and the exact structure of these spontaneously formed superlattices have not yet been established. In Iliopouloss work [947], the three kinds of superlattice structures were found in different AlGaN layers grown under different conditions. However, Ruterana et al. [996] reported the coexistence of different types of ordering in the same AlGaN film. The sample they investigated was grown by OMVPE with good optoelectronic properties. A close examination by diffraction analysis as shown in Figure 3.204 showed the AlGaN sample had the usual 1 : 1 (Ga : Al) ordered domains [996] and the 3 : 1 ordering, which was reported in InGaN [1027]. Moreover, another type of ordering is shown to take place in which a series of five GaN cells and one AlN form the new supercell (i.e., 10 GaN monolayers and 2 AlN monolayers). The average composition of such ordered phase corresponds to Al0.66Ga83.4N and the electron diffraction analysis was made inside a 14% Al composition layer. The occurrence of such ordered phases gives rise to compounds with lower symmetry than the parent wurtzite GaN with the same space group (P3m1). However, a more accurate spatial distribution and the structural relationships between the ordered phases and the matrix is needed, requiring further investigation in an effort to understand formation mechanisms involved.
Figure 3.204 Diffraction patterns with superlattice spots. (a) 1 : 1 ordered phase along the ½1 0 1 0 zone axis, g1//[0 0 0 1], g2 ==½1 1 2 0; (b) 3 : 1 with three spots between 0 0 0 0 and 0 0 0 2, along ½1 0 1 0 zone axis; and (c) superlattice spots, d is approximately 3.1 nm, diffraction pattern along the ½1 1 2 0 direction, g1: [0 0 0 1], g 2 ==½1 0 1 0. Courtesy of P. Ruterana and Ref. [996].
3.5 The Art and Technology of Growth of Nitrides
Figure 3.204 (Continued )
3.5.15.2 Growth of InGaN The growth of GaInN alloys has proven to be relatively more difficult. The large difference in interatomic spacing between GaN and InN and high nitrogen pressure over InN due to N volatility are the causes that give rise to a solid phase miscibility gap. Growth at high temperature (800 C) results in higher crystalline quality, but the amount of InN in the alloy is low and very high V/III ratios, meaning primarily high NH3 flow rates, are required. It should be mentioned that the flow rate for a given
j671
j 3 Growth and Growth Methods for Nitride Semiconductors
672
V/III ration is system dependent. In authors laboratory the ammonia flow rate is about 10 l min1. Conversely, growth at lower temperatures (500 C) increases the InN concentration but at the expense of low crystalline quality [998]. Increasing In pressure in the vapor results in the formation of In droplets on the surface [999]. The early attempts for realizing single crystalline InGaN by OMVPE were made by Nagatomo et al. in 1989 [1000] and Matsuoka et al. [1001] again in 1989, and followed by Yoshimoto et al. in 1991 [1002]. Since then, considerable work has been expended worldwide. Matsuoka et al. [1003] discovered that lowering the growth temperature to 500 C from nominal temperatures, such as 800 C, increased the In content in the layers, but at the expense of reduced quality. Efforts to increase the In concentration by raising the indium precursor temperature or the carrier gas flow rate resulted in degradation of the structural and surface morphology so much so that In droplets were formed on the surface, which is common to all the other growth methods, including sputtering [1004]. Because InGaN layers are typically used as thin active layers and are generally straddled by GaN and/or AlGaN layers, and gas source consumption is considerable, only thin layers are grown in the form of quantum wells. Consequently, the bulk of the reports in this material system deal with quantum wells. It is therefore instructive to discuss the aspects of InGaN quantum wells in the context of growth related issues. Defects and in particular V defects garnered a good deal of attention as reflected by number of publications. In an effort to map the quality of InGaN layers Reed et al. [1005] prepared 5, 10, and 15% InGaN double heterostructures by OMVPE and studied the PL, X-ray, and Hall mobilities to determine the best conditions for InGaN layers. The structures consisted of a LT GaN buffer(550 C) followed by 1 mm GaN then InGaN growth at 800 C (5%), 775 C (10%), and 750 C (16%), and finally a 100 nm GaN cap at 1000 C (5–10%) and 880 C (16%). The premise of this particular exercise is that there is a critical range of parameters, namely the thickness and intertwined with it the In molar fraction, within which GaN/InGaN structures display the best electrical and optical properties. The electron mobility increases with increasing InGaN thickness up to 100 nm for 5% and 50 nm for 10% In molar fraction, beyond which that parameter degrades. The authors suggest that all of these results can be related to the strain/relaxation state of the InGaN material as well as continued relaxation during any thermal loads encountered by the samples even after the MQW has been capped. Moving from the dimensional parameters associated with the structures to the growth-related parameters, Piscopiello et al. [1006] investigated the effects of H2/N2 ratio as carrier gas during InGaN growth. Their OMVPE growth on sapphire commenced with a 50 nm nucleation layer formed at 560 C followed by 1 mm GaN grown at 1150 C and finally 100 nm InGaN grown at 790 C under H2 and N2/H2 (4 : 1). While the surface roughness between the two is about the same, in the area of a few nanometers, a large difference is in the indium incorporation, which increases as the N2/H2 ratio increases, was observed. However, the defect density and structure are reported to be very similar between the two types of samples. Focusing on the effects of growth-related parameters, Kim et al. [1007] set out to determine the effects of temperature and carrier gas on the quantum barriers in the MQW. They did so by using thermally precracked NH3 and growing a 30 nm LT
3.5 The Art and Technology of Growth of Nitrides
buffer (500 C) followed by 2 mm GaN (1080 C at 300 Torr) then a five period In0.28Ga0.72N (3 nm)/GaN (7 nm) MQW with InGaN at 700–740 C and GaN at 840–880 C. Those authors explored constant growth temperature under N2 ambience for both the wells and barriers, ramping up the temperature without capping for the barriers under N2 ambience, ramping up the temperature with capping for the barriers under N2 ambience, and ramping up the temperature with capping for the barriers under H2 ambience. The results show that the sample with H2 ambience for the capping layer had the best PL results but a higher XRD value. The authors also suggest optimization of capping layers as they saw degradation of PL if the cap was over 1 nm. This degradation would more likely be due to temperature loads of the cap layer instead of layer thickness according picture presented by Reed et al. [1005] regarding the critical layer thickness. Along similar lines, Lefebvre et al. [1008] attempted to determine the effects on PL of barriers versus QWs and AlGaN caps versus GaN caps. Their MBE growth on sapphire consisted of a 1–5 mm GaN buffer (800 C) followed by 15–20% InGaN (550 C) with a GaN or AlGaN cap. The samples investigated had 2 nm thick quantum wells with either GaN or Al0.14Ga0.86N barriers bulk and/or surface barriers. The buried and surface GaN barrier layer thicknesses of 800 and 40 nm, respectively, were employed. In another set, buried GaN barrier layers of 500 nm and Al0.14Ga0.86N surface barrier of 30 nm were employed. The authors found increased PL intensity at higher temperatures for the layers with the AlGaN cap versus the GaN cap. They attribute this to interdiffusion of aluminum atoms introducing potential barriers, reducing the in-plane mobility of carriers between different recombinations centers and toward nonradiative centers. The optical emission seen was attributed primarily to localized electron–hole pairs even at room temperature. Furthermore, the authors did not note any benefit because of quantum boxes over quantum boxes because the localization was smaller than the quantum box dimensions. Ting et al. [1009] studied the evolution of defects and related morphology in InGaN/ GaN MQWs as well as subsequent device characteristics. The growth was conducted by OMVPE on sapphire with a 300 Å GaN buffer layer (grown at 520 C) followed by 3 mm n-GaN (4 · 1018 cm3) then 1 mm unintentionally doped GaN (1050 C). The growth continued with a 2500 Å n-GaN regrowth burying layer followed by a 20 Å GaN barrier layer and finally a five period 30 Å In0.3Ga0.7N/20 Å GaN MQWs. Single quantum well (QW) structures were also grown and studied and included a 120 Å GaN barrier on top. Much of the analysis relied on AFM imaging and LED device fabrication and testing. Applying the best growth conditions based on the materials characterization, the authors created a thermally robust 525 nm LED. It was concluded that inclusion formation being evident only in InGaN layers could not be the result of LT GaN growth, but that the inclusions continued to grow during GaN periods of the multiple quantum well structure. The authors also observed that adding H2 or ramping the temperature during the GaN barrier layer growth would planarize the InGaN surface and improve barrier layer properties. The authors believe the V defect is a result of transitioning from InGaN 3D growth to GaN 2D growth and it is this interface that is the most important because InGaN layers seem to have 3D growth no matter how good the GaN
j673
j 3 Growth and Growth Methods for Nitride Semiconductors
674
layer underneath looks. The PL on the MQWs shows that the structures grown with H2 used in the barrier show three times less thermal degradation when subjected to subsequent thermal processes than the ones with N2 used in the barrier layer. Named after their shapre V defects are fairly common in InGaN and InGaNcontaining quantum wells. Many studies have been devoted to the investigation of these defects including their genesis. Among them, Sharma et al. [1010] also discussed V defects found in InGaN/GaN MQWs including their formation and chemical composition. These V-shaped voids, which have also been discussed in the same context by Cho et al. [1011], should not be confused with pyramidal defect observed in Mg-doped GaN and AlGaN layers, a topic discussed in Section 4.2.3. The samples used for this investigation were grown by OMVPE using low-temperature GaN initiation layers. The structures investigated in the effort of Sharma et al. were grown using a typical two-step procedure as follows: a 50 nm nucleation layer was deposited at a susceptor temperature 560 C followed by 150 nm of undoped GaN. Then a 2.8 mm layer of n-type Si-doped GaN was deposited at 1065 C. The MQW was composed of five periods of InGaN/GaN grown at 790 and 950 C, respectively. Finally, the entire structure was capped by a 150 nm p-type Mg-doped GaN layer grown at 1045 C. Energy filtered cross-sectional TEM chemical mapping of the quantum well structures was used to concluded that the V defects nucleate at dislocations in the first layer and propagate through the entire active layers that can be filled with p-type GaN. Cho et al. [1011] reported, however, that the V pits can be nucleated at a stacking mismatch boundary, which, in turn, is caused by a stacking fault. A TEM image of the threading dislocation induced V defect is shown in Figure 3.205, whereas models for threading dislocation nucleated and stacking mismatch boundary nucleated models for Vdefects are shown in Figure 3.206. Further, TEM and annular dark field imaging (ADF) studies of the samples show these V defects to be approximately 100–200 nm apart along the MQW. Moreover, it has been stated that each V defect incorporates a pure edge ðb ¼ 1=3 h1 1 2 0iÞ dislocation, which runs through the apex of the V defect up to the free surface. The lack of growth where these dislocations are is most likely because of preferential evaporation of reactive species at the dislocation site at the particular temperature dictated by InGaN growth. However, growth of the p-type layer at a higher temperature, 1045 C, enhances reactive specie diffusion at the surface ensuing lateral growth that together leads to filling of the V groove defect as shown in Figure 3.206. Along the same lines, Kim et al. [1012] prepared MQWs and also thicker (bulk) InGaN to study V defects, indium mole ratio, and thickness. The authors prepared bulk (considered bulk in terms of exciton radius and de Broglie wavelength) GaN by OMVPE and MBE on OMVPE templates followed by a MQW structure grown by OMVPE on sapphire. It consisted of a 25 nm buffer layer (grown at 550 C) followed by 1.5 mm GaN grown at 700 C and then a 2 : 3 thickness ratio, 5 period InGaN/GaN MQW grown at 800 C. Relying on TEM images, Kim et al. [1012] have argued that only mixed and screw dislocations terminate in V defects on the surface, with pure edge having no surface characteristic. Therefore, Kim et al. [1012] disagree with Sharma et al. [1010] in regard to the type of dislocation that initiates a V-defect site. Kim et al. [1012] observed a clear degradation in surface quality as the In incorporation increased. This poor morphology has been also linked to increased strain in the InGaN layer due to increasing indium
3.5 The Art and Technology of Growth of Nitrides
Indium concentration (%)
15
10
5
0 (c)
0
20
40
60
80
Profile width (nm) Figure 3.205 Energy filtered TEM (EFTEM) indium ratio map (a) and gallium elemental map (b), which show that OMVPE-grown InGaN quantum wells end abruptly at the V-defect boundary. Quantitative EFTEM from a line profile across the MQW (b) shows a small variation in the indium concentration between the wells (c). The V-defects can be filled with the p-type GaN cap layer. Courtesy of C. Humphreys and Ref. [1010].
mole ratio and increasing thickness. Poor XRD results for the thicker layers with higher indium incorporation have been attributed to phase segregation in the InGaN. Other substrates were also used for InGaN growth. It was reported that the crystalline quality of InGaN was superior when grown with the composition that lattice matches ZnO substrate to that grown on bare (0 0 0 1) sapphire sub-
j675
j 3 Growth and Growth Methods for Nitride Semiconductors
676
Filled V-pit V-pit
V-pit p-GaN
SMB SF n-GaN
Threading dislocation
InGaN GaN SMB: stacking mismatch boundary SF: stacking fault Figure 3.206 A model for the formation of the V-defect nucleated at a threading dislocation defect site or due to a combination of a stacking fault (SF) and a SMB. Growth of p-type GaN has been reported to fill the V-shaped pit. GaN. Patterned after Refs [1010,1011].
strate [1002,1003]. In the same investigations, it was observed that InGaN films grown on sapphire substrates using GaN as buffer layers exhibited much better optical properties than InGaN films grown directly on sapphire substrates [1013]. For a given set of growth conditions, an increase of InN in InGaN can be achieved by reducing the hydrogen flow [1014]. The overarching issue in In containing ternary and quaternary nitride semiconductors is the phase separation and compositional instability caused by great disparities between In and other group III metals used, such as between atomic radii, equilibrium N vapor pressure over the metal, surface diffusion barrier, and other aspects. Although the wrenching issues in this vein are left to be discussed in Section 3.5.15.2.2, a glossary is given here. In this vein, Ho and Stringfellow [1015] investigated the temperature dependence of the binodal and spinodal boundaries in the InGaN system with a modified valence force field model. The calculation of the extent of the miscibility gap yielded an equilibrium InN mole fraction in GaN of less than 6% at 800 C [1015]. In the annealing experiments in argon ambience, the phase separation of an InxGa1x N alloy with x 0.1 was observed at 600 and 700 C [1016], pointing to the large region of solid immiscibility of these alloys. However, under nonequilibrium growth conditions, Ga1x InxN layers were grown in the entire range of compositions. However, the decomposition into two phases upon annealing of the InxGa1xN alloys (x ¼ 0.11 and x ¼ 0.29) at 600 and 700 C was observed pointing to the existence of the miscibility gap. For some alloys with x ¼ 0.6, the phase separation could not be observed at 600 C. Above 800 C, the alloy samples with x ¼ 0.1 actively evaporated from the substrate. These results suggest that the solid solutions are grown in metastable conditions and decompose under annealing conditions.
3.5 The Art and Technology of Growth of Nitrides
Koukitu et al. [1017] performed a thermodynamical analysis of InGaN alloys grown by OMVPE. They found that in contrast to other III–III–V alloy systems where the solid composition is a linear function of the molar ratio of the group III metalorganic precursors at constant partial pressure of group V gas, the solid composition of InGaN deviates significantly from a linear function at high substrate temperatures. Kawaguchi et al. [1018] reported on the so-called InGaN composition pulling effect in which the indium fraction is smaller during the initial stages of growth but increases with increasing growth thickness. This observation was to a first extent independent of the underlying layer whether it was GaN or AlGaN. The authors suggested that this effect is caused by strain due to the lattice mismatch at the interface. It was found that a larger lattice mismatch between InGaN and the bottom epitaxial layers was accompanied by a larger change in the In content. What one can glean from this is that the indium distribution mechanism in InGaN alloy is caused by the lattice deformation due to the lattice mismatch. With increasing thickness, the lattice strain is relaxed owing to the formation of structural defects, which weakens the compositional pulling. As always, progress is made this time on the front of better understanding processes involved in the growth of In-containing ternaries. Reaction pathways in effect during growth of In-based alloys have been investigated with good progress [1019]. Such insight led to the attainment of InGaN films with high InN molar fraction and improved structural and optical properties. It is to be noted that the phase separation is a real concern in this alloy, particularly for high InN molar fraction. Using a unique susceptor, Bedair et al. [1019] reported In0.1Ga0.9N layers with symmetric X-ray FWHM values of as little as 6 arcmin. The PL was characterized as being intense band edge type with occasional deep level transitions. In this particular approach, dubbed molecular stream epitaxy (MSE), the substrate is first exposed to a mixed gas composition inclusive of group III and V species. The sample is then rotated away from the impinging gas along the axis of the susceptor. After a complete rotation, the substrate is again exposed to reactant gases. The state of affairs is such that presence of In brings about compositional nonuniformities and localization, which makes the picture highly complex. Although higher temperatures and ensuing inhomogeneities have been mastered to extract efficient light from InGaN for LEDs and lasers, the same is not helpful for electron transport. Consequently, the expected relatively higher mobilities visceral to InGaN have not materialized. The electrical properties of InGaN quantum wells, in the form similar to single-period modulation doped or undoped single heterostructures have also been investigated in structures grown by both MBE and OMVPE. Because the implications are more pertinent to FETs, the some details are discussed in Volume 2, Chapter 5. In an effort to enhance the quality of InGaN in MQW structures, Pozina et al. [1020] employed a mass transport overgrowth technique to grow high-quality InGaN/GaN MQWs with good optical properties. The OMVPE growth on sapphire started with an AlN buffer followed by 7 mm unintentionally GaN. Micrometer size trenches (3, 5, 10 mm) were then patterned using RIE and the sample was annealed at 1100 C in ammonia and N2 ambience resulting in mass transport overgrowth. Finally, 35 Å
j677
j 3 Growth and Growth Methods for Nitride Semiconductors
678
InGaN/105 Å GaN (Si doped to 1018) three-period MQWs were grown. The CL images taken by Pozina et al. [1020] suggest improved optical and physical properties of the mass transport regions. These improvements include a dislocation density reduction from >108 to <107 based on CL, and the narrowest InGaN/GaN MQW PL line reported as of the year 2001 with FWHM of 40 meV. The authors also reported an additional peak about 100 meV below the main peak that they attributed to incomplete charge compensation in the barrier creating a strong potential gradient across the MQWs, causing higher carrier filling in one of the quantum wells. The conclusions that can be drawn from the above discussion in terms of growth is that higher temperatures lead to improved optical properties and yet high vapor pressure of InN scuffles the high temperature. It then becomes an issue of how high In and ammonia flow rate one is willing to employ for higher temperature growth. The enhanced optical properties at higher temperatures might in fact be because of increased compositional inhomogeneities that serve well for low-level injections, the advantage of which would wane at higher injection levels. In addition, compositional inhomogeneities in a highly piezoelectric material such as InGaN cannot possibly be construed as helpful to mobility, which may be in part responsible for lack of demonstration of increased mobility in InGaN as compared to GaN. 3.5.15.2.1 Doping of InGaN As in the case of GaN, achieving n-type doping in InGaN is simple and in fact the unintentionally doped InGaN contains sufficient donorlike impurities and/or defects to produce sufficient electron concentration to obviate the need for external doping. Nevertheless, attempts have been made to dope InGaN with shallow donors, shallow acceptors such as Mg and deep impurities such as Zn. Doping by Zn introduces highly radiative deep centers and thus provides one more parameter for wavelength tuning. However, the doping level is not large enough to prevent light emitting diodes from exhibiting blue shift at higher injection levels as the Zn-related centers begin to saturate. In one such investigation, Tong et al. [1021] employed PL to view the effect of Zn and Si doping as well as the effect of temperature ramping in InGaN layers. The layers were grown by OMVPE on sapphire with a 20 nm GaN low temperature initiation layer (550 C) followed by a GaN epilayer (1060 C) and finally an InGaN layer (760–1060 C). In the particular work growth of InGaN with greater than 25% In at temperatures above 780 C met with difficulty due to high vapor pressure of InN. The dilemma is that the luminescence intensity severely degrades with lower temperature, that is, 20 times lower PL for layers grown at 760 C versus at 820 C. Therefore, Zn was used as dopant to obtain additional red shift of the emission to obtain blue light emission. The authors state that Zn also increased the PL intensity by as much as 30 times as compared to undoped InGaN. Moreover, Si was also shown to increase the PL intensity by as much as 13 times, but there is a critical level that if crossed will result in degradation of the PL. In another investigation, Wang et al. [1022] determined the effects of Si doping in InGaN/GaN MQWs on the PL and mobility. The employed OMVPE growth on sapphire consisted of a 25 nm GaN buffer (450 C) then a 1.5 mm unintentionally doped GaN (1075 C) followed by a 2.5 nm In0.13Ga0.87N/7.5 nm GaN : Si MQW (700 C and SiH4 varied 3, 5, 10, 15 sccm) and finally a 100 nm GaN cap (700 C). An
3.5 The Art and Technology of Growth of Nitrides
undoped MQW was also grown for reference. The authors state that their observations indicate a localization effect in the slightly doped Si samples and that this localization leads to increased PL and mobility values, whereas the former is understood but the latter is counter institutive. If the only purpose of Si doping is to enhance the point defect nature of the films, the authors recommend only slight doping based on the deterioration of the mobility and PL intensity with increasing SiH4 flow rate. This recommendation is in general concord with the results of Tong et al. [1021] especially in regards to the effect of Si doping on the PL. Turning our attention to p-type InGaN, following the painstakingly slow accomplishment of p-GaN, the p-type nitride semiconductor family quickly expanded to include AlGaN with AlN mole fractions of about 50% and InGaN with the InN mole fraction somewhat lower than AlN in the AlGaN alloy. Mg-doped InxGa1xN (x ¼ 0.09) films were grown at 800 C by Yamasaki et al. [1023], again by employing OMVPE and low-temperature AlN buffer layers. TMG, TMA, TMI, and Cp2 Mg were used for Ga, Al, In, and Mg source gases, respectively, and NH3 for nitrogen. After annealing at 800 C for 5 min in a nitrogen ambient gas at atmospheric pressure, the layers showed p-type conduction. Hall-effect measurements in the temperature range of 90–500 K indicate that the hole concentration initially decreases rather rapidly, and then almost linearly with decreasing temperature down to T ¼ 143 K, before saturating. The measured hole concentration, if dependable, was about 5 · 1018 cm3 at 500 K and dropped to 8 · 1016 cm3 at 100 K. The measured Hall mobility lip first increased and then decreased with increasing temperature. It shows a peak value of about 40 cm2 V1 s1 at 150 K and a value of 10 cm2 V1 s1 at 300 K. By fitting the theoretical expression of hole concentration as a function of the measurement temperatures, the activation energy of the Mg acceptor, EA, was estimated to be 204 meV above the valence band. This is in the same ballpark as that in GaN. 3.5.15.2.2 Phase Separation in InGaN The solid phase immiscibility in InGaN, which occurs due to large difference in interatomic spacing (with lattice mismatch 11.25% between the two end binaries), and other physical properties such as thermodynamical and chemical stabilities, between GaN and InN, has been studied both experimentally ortheoretically. Ontheexperimental side,ordering, clusters, phase separation, and growth instabilities in InGaN system have been reported by many researchers [1024–1027]. The accompanying special variation in composition naturally affects the optical properties [1028]. Depending on the problem in hand, phase separation can be helpful or hurtful. For example, localization caused by compositional variations InGaN, particularly when the quality of this ternary was not as advanced, was useful in obtaining strong optical emission. However, this benefit holds true only within reason and only for a range on InN molar fractions. For example, the operation wavelength of LD for years were restricted to 430 nm but not reaching to longer wavelength because of not obtaining sufficient optical gain with higher In content due to intolerable inhomogeneous broadening in the optical gain spectrum [1029,1030]. Turning our attention to transport, phase separation has adverse effects in that the carrier mobility is hindered because of boundaries as well as strain inhomogeneities that induce piezoelectric fields scattering carriers. Excess phase separation, in addition
j679
j 3 Growth and Growth Methods for Nitride Semiconductors
680
to the aforementioned gain broadening, only causes In-rich precipitates to form, which is hurtful. Unlike the carrier localization, In-rich precipitates are structural defects that can generate misfit dislocations at the boundaries or contain In metal rich regions, resulting in the degradation of device performance. As in the case of AlGaN, growth concerns and structural properties, particularly phase separation, are discussed in this section, leaving the treatment of optical properties to Volume 2, Chapter 5. The compositional decomposition in the extreme case of full phase separation may even lead to InN inclusions in InxGa1xN epilayers grown on (0 0 0 1) c-plane sapphire substrates, as has been reported experimentally [1031]. Strain and compositional fluctuations affect every parameter of the nitride alloys including the miscibility. Theoretical studies of the InxGa1xN alloys performed within the framework of the valenceforcefield(VFF)approximationindicateashiftofthemiscibilitygapintothearea of higher InN concentrations [1032] (Figure 3.210) or lower InN concentration [1033] depending on the details of the albeit simplified calculations. Structurally the first- and second-nearest-neighbor distances are assumed affected by strain [1033–1035]. Congruent with the above prelude, one often encounters that to incorporate the desired amount of cation into the host crystal, the growth temperature must be lowered at the expense of crystalline quality. This was noted by Matsuoka et al. [1036] in their attempt to grow InGaN by OMVPE wherein the growth temperature had to be lowered from 800 to 500 C to increase In incorporation. Attempts to increase the In incorporation by increasing the amount of In provided to the surface met with In droplets [1037]. Phase separation was noted as early as 1975 in InGaN samples after annealing in argon ambience at various temperatures below 700 C [1038]. It is therefore necessary to discuss the thermodynamic stability of nitride alloys, particularly those containing In, as those are the ones where the constituent binaries differ substantially in their structural parameters. It is now well known that the nitrides are not fully miscible, that is, there are strong indications for a miscibility gap [1039–1041]. Under ambient conditions, GaN, AlN, and InN crystallize in the hexagonal wurtzite structure. From a theoretical standpoint, the numerical tools have been more readily developed for cubic semiconductors more so then wurtzitic varieties. This holds in the context of the current topic in terms of the phase separation. Focusing on the cubic system is considered acceptable in terms of the issues related to phase separation. Moreover, successful growth of cubic (c-) zinc blende AlN [1042] and InN [1043] bulk materials by means of molecular-beam epitaxy has been reported and could provide the means for a comparison. It should also be noted that although not as common as the InGaN variety, the ternary c-InxAl1xN thick has been deposited [1044–1047]. As in the case of InGaN, the large difference in the equilibrium lattice constants of InN and AlN results in a considerable internal strain, which is considered a major component of the tendency for the In-containing nitrides such as InGaN, InAlN, and AlInGaN alloys to phase separate. A synonymous statement, which is often seen in the literature, is that the large difference in interatomic spacing between GaN and InN, and AlN and InN gives rise to a solid phase miscibility gap. It should be noted that the boundaries between unstable and metastable regions are important in the mixing of some binary compounds. This boundary is expressed with the spinodal isotherms.
3.5 The Art and Technology of Growth of Nitrides
This isotherm can be calculated as the inflection points for the Helmholtz free-energy surface plotted as a function of fractional compositions. Ho and Stringfellow [1015] calculated the temperature dependence of the binodal and spinodal lines in the Ga1xInxN employing a modified VFF model where the lattice is allowed to relax beyond the first nearest neighbor. The calculations indicated the strain energy to decrease until approximately the sixth nearest neighbor within the framework of the dilute limit. An interaction parameter of 5.98 kcal mol1 was found assuming a symmetric, regular solution like composition dependence of the enthalpy of mixing. At a typical growth temperature of 800 C, as far as the OMVPE is concerned, the solubility of In in GaN is calculated to be less than 6%. In the calculations, the crystal symmetry of both GaN and InN was taken to be that of zinc blende, recognizing that the wurtzitic form is the polytype that is available and used. Ho and Stringfellow [1015] assumed that these two structures are different only in the atomic stacking sequence along the direction normal to their corresponding closepacked planes which are (1 1 1) in zinc blende and (0 0 0 1) in wurtzitic forms, and further the equilibrium bond lengths and force constants are the same. Under these assumptions, those authors found the interaction parameters for the two polytypes to differ by only about 1%. Consequently, the two structures were considered nearly equivalent in the strain energy calculations and the results obtained for the zinc blende form should closely approximate those for the wurtzitic InGaN system. Shown in Figure 3.207 are the calculated, by Ho and Stringfellow [1015], binodal and spinodal compositions versus temperature over the entire composition range using the average value of solution interaction parameter of 55.98 kcal mol1. The critical temperature is found to be near 1250 C, which exceeds the melting point of
1200 Binodal
Temperature (ºC)
1000 800 600 Spinodal
400 200 0 0.0 GaN
0.2
0.4
0.6
Indium fraction, x
0.8
1.0 InN
Figure 3.207 Calculated binodal and spinodal, solid and dashed curves, respectively, for the InxGa1xN system, assuming a constant average value for the solid phase interaction parameter [1015].
j681
j 3 Growth and Growth Methods for Nitride Semiconductors
682
InN. At the typical y maximum growth temperature of 800 C used for InxGa1xN, the solubility of InN in GaN is calculated to be less than 6%. The results discussed above demonstrate that, in general, the InxGa1xN alloy is unstable at the temperatures typically used for epitaxial growth. This conclusion is supported by the results of annealing experiments of Osamura et al. [1038] who reported phase separation in InxGa1xN with x > 0.1 following annealing in an argon ambience at 600 and 700 C. The large region of solid immiscibility may in part be responsible for the difficulties involved in the epitaxial growth of these alloys. Rigorous first principles thermodynamics calculations with a focus on phase separation in the InGaN have been undertaken by Teles et al. [1040]. The calculations are based on the generalized quasichemical approach to disorder and composition effects and a pseudopotential plane wave approximation for the total energy. Those authors generalized the cluster treatment to study the influence of biaxial strain, which pointed to a remarkable suppression of phase separation in InxGa1xN. These are consistent with experiments. The calculations of Teles et al. [1040] led to the conclusion that a broad miscibility gap exits for growth temperatures around 1000 K for unstrained InxGa1xN alloys. Those authors also calculated a critical temperature of 1295 K for phase separation, which is some 250 K lower than that calculated by Ho and Stringfellow [1015]. However, the resulting structural properties, such as the dependence of the lattice constant on composition, the bond lengths and the second nearest neighbor distances are in agreement with other calculations or measurements, and that the structural parameters for intermediate In molar fractions would exhibit notable fluctuations. Fortuitously, the biaxial strain has far reaching consequences in terms of the miscibility behavior of the alloys in that miscibility gap is reduced [1040]. This is particularly the case for cubic InxGa1xN, which is conjectured to represent wurtzitic alloy in this respect, as compared to AlxIn1xN. Recall that AlxGa1xN does not exhibit miscibility gap and as such the effect of strain is a mute point. However, for the InxGa1xN case, the higher the InN molar fraction, the more noticeable the effect. Invariably, the region of spontaneous decomposition is reduced by strain. By considering an inhomogeneous strain distribution over the clusters contributing to the ternary alloy, Teles et al. [1040] sought to show that a regular behavior of the structural and elastic properties are obtained. In this limit, the miscibility gap and the critical temperature are remarkably reduced. For coherent layers, that is, the in-plane lattice constants of the underlying binary compounds, the phase separation is even completely suppressed. As the in-plane strain increased from the relaxed case to e// ¼ 0.01 to 0.05, the excess free energy is decreased by the elastic energy due to the buildup strain. Consequently, the miscibility gap decreased with increased strain as shown in the T–x phase diagram of Figure 3.208. For a biaxial strain of 21%, the strain-induced change in the critical temperature is not notable, except that the phase diagram becomes only slightly more asymmetric owing to small elastic energy. However, because the strain energy is proportional to the square of the in-plane strain e// further increases as the critical temperature decreases and reaches a low of 120 K for 25% strain. Strain also reduces the miscibility gap as well as the region of spontaneous decomposition, particularly for larger In molar fractions. In short, the in-plane compressive strain suppresses the phase separation.
3.5 The Art and Technology of Growth of Nitrides
1600
Binodal Spinodal
ε = – 0.001
1400
Temperature (K)
1200
1000 ε = – 0.05
800
600
400
200
0 0.0
0.2
0.4
0.6
0.8
1.0
Composition, x Figure 3.208 T–x phase diagram for homogeneously strained InxGa1xN. Heavy and light lines depict the binodal decomposition curves while the corresponding dashed ones depict the spinodal decomposition curves [1040].
There are indications that the assumption of an equal strain in all microclusters of the alloy is not suitable. Teles et al. [1040] observed that the c12/c11 elastic stiffness constant ratio is nonlinear versus composition for both tensile and a compressive strain of 1%. It should be mentioned that the arithmetic average of the two curves gives an almost linear variation of the ratio c12/c11 with composition between the values known for GaN and InN. Therefore, an accurate representation requires the inhomogeneous strain distribution in the alloy on a microscopic length scale be considered. Congruent with one of the basic assumptions that each cluster takes its own volume, Teles et al. [1040] assumed a coherently strained alloy, which means that each cluster j is strained according to the change in the in-plane lattice constant a//. The in-plane strain in each cluster is therefore given by ej// ¼ (a// aj)/aj. Consider two inhomogeneous strain cases: one representing the coherent growth of the ternary InxGa1xN alloy on a GaN substrate with a// ¼ aGaN and the other on InN substrate with a// ¼ aInN. The thermodynamic potential taking inhomogeneous strain into consideration is shown in Figure 3.209. The inhomogeneous strain distribution, albeit assumed in this model, substantially affects the mixing free energy of the system by changing primarily the mixing enthalpy.
j683
j 3 Growth and Growth Methods for Nitride Semiconductors
684
Mixing free energy, ΔF (eV/pair)
0.0
on GaN on InN
–0.01
T (K)
–0.02
700
–0.03
950
–0.04 700
–0.05 –0.06 –0.07 0.0
950
0.2
0.4
0.6
0.8
1.0
Composition, x Figure 3.209 Compositional dependence of the mixing free energy of inhomogeneously strained InxGa1xN alloys for T ¼ 700 and 950 K. The biaxial strain is defined by a// ¼ aGaN and depicted with solid lines or in the case of growth on InN a// ¼ aInN and depicted with dashed lines [1040].
Ho and Stringfellow [1015] demonstrated, using a valence-force-field model, that at growth temperature of 800 C, the solubility of In in GaN is less than 6%, as shown in Figure 3.207 assuming of course that the thermodynamic equilibrium holds. Equilibrium conditions do not really apply to a large extent to MBE and apply more readily to OMVPE grown layers. Regardless of the extent, phase separation is synonymous with InGaN. Ho and Stringfellow found a critical temperature of 1250 C for phase separation of InGaN, which is higher than the melting point of InN. It should be noted that at nominal growth temperatures, the alloy is unstable over the entire composition. With strain, however, as shown comparatively in Figure 3.210a and b, the solid solubility of the alloy can be increased remarkably to a value of about 60% at 800 C from the 6% value effective for the relaxed case [1032]. Behbehani et al. [1048] observed phase separation in InxGa1xN samples grown by metal organic chemical vapor deposition (OMVPE). For x ¼ 0.49, phase separation and c-plane ordering were observed by both XRD and SAD analyses. A typical transmission electron micrograph image for a nominally In0.49Ga0.51N and its SAD image are shown in Figure 3.211 where the dark regions contain high InN composition and lower percentage of InN regions showing ordering. Extra spots in the SAD image are due to ordering and some corresponding to In0.25Ga0.75N. Samples with less than x ¼ 0.25, did not show phase separation. Surprisingly, in thin samples, for x ¼ 0.21 extra spots were observed by SAD analysis; however, thick layers did not show any sign of phase separation. The SAD pattern revealed that both In0.5Ga0.5N and In0.21Ga0.79N samples showed ordering pattern. In general, InGaN samples containing a high percentage of InN show ordering [1015]. The extra spots might be due to defects. The nonideal diffraction patterns were observed, which are
3.5 The Art and Technology of Growth of Nitrides
(a)
Binodal
Spinodal
1600
Temperature (K)
1200
Relaxed InGaN
800
400
0 0.0
0.2
GaN
0.4
0.6
0.8
1.0
InN
Indium fraction, x
1200 (b) 1000
Wurtzite to c-axis
interface
Temperature (K)
800 600
Binodal
Strained InGaN
400 200 0 0.0
Spinodal
0.2
0.4
0.6
0.8
1.0
InN
GaN Indium fraction, x
Figure 3.210 Phase diagram of InN–GaN system (Binodal–solid line, spinodal–dashed line); (a) for relaxed InGaN and (b) for strained InGaN where an asymmetric shift to of the equilibrium InN fraction in the alloy to higher values occurs [1032].
due to the interference of the 1 : 1 and 1 : 3 ordered phases for x ¼ 0.25. The ordered phase has stacking Ga and In atoms with ratio of 3 : 1 along the c-axis (Ga–Ga– Ga–In–Ga–Ga–Ga–In–Ga–). The authors constructed theoretical diffraction patterns for 1 : 3 and 1 : 1 orderings, as shown in Figure 3.212. Simulated lattice images by Fourier transforms indicated that the interference of the 1 : 1 (left) and 1 : 3 (right) ordered phases showed the nonideal diffraction patterns A 2–111 twin plane with 35 rotation around ½0 1 1 0 [1048]. Sing et al. [1049] studied the segregation of secondary phases in InGaN thin layers and heterostructures grown by MBE. The InGaN layers (x > 0.30) showed phase separation by both XRD and optical absorption analysis in that InN, In0.37Ga0.63N,
j685
j 3 Growth and Growth Methods for Nitride Semiconductors
686
Figure 3.211 (a) TEM image of a nominally In0.49Ga0.51N layer where the dark regions contain high InN composition and lower percentage of InN regions showed ordering; (b) electron diffraction pattern of a nominally In0.49Ga0.51N layer, which shows extra spots due to ordering and corresponding to In0.25Ga0.75N. Courtesy of N. El-Masry and Ref. [1048].
Ga In Ga In
Ga In Ga Ga Ga In Ga Ga
50% In
Ga
Ga: In
Ga: In
1:1
3:1
One unit cells 50% InGaN
Two unit cells 25% InGaN
Figure 3.212 Theoretical diffraction pattern of 1 : 3 (right) and 1 : 1 (left) ordering where the smaller and extra spots are due to ordering along with schematically layered structure for each case. Courtesy of N. El-Masry and Ref. [1048].
25% In
3.5 The Art and Technology of Growth of Nitrides
GaN
104
In0.37Ga0.83N
103
Intensity (au)
InN
102
101
100
–8000 –6000 –4000
–2000
0
2000 4000
6000
Δθ (arcsec) Figure 3.213 X-ray rocking curve of In0.37Ga0.63N epilayers in which InN is phase separated [1049].
and GaN phases were observed in the X-ray data, as shown in Figure 3.213 . However, in an InGaN/GaN heterostructure, only the In0.53Ga0.47N and GaN phases were observed by X-ray rocking curve, as shown in Figure 3.214. Note that the heterostructure suppressed the segregation of InN in the InGaN layers. The optical absorption studies showed a bandgap of 2.0 eV for In0.37Ga0.63N. However, no phase separation was observed by TEM analysis owing to high density of defects. In the double heterostructures, no InN phase separation was detected because the strainassociated thin InGaN layers could prevent phase separation by stabilizing the alloy, as shown in Figure 3.210. El-Masry et al. [1050] extensively studied phase separation in two different types of samples grown at 710 C, which contained 20 and 49% of InN, respectively. The layers grown at 710 C showed In0.2Ga0.8N and GaN phases, whereas those that contained more than 40% of InN in InGaN, showed the intended In0.49Ga0.51N and extra 95, 36, and 14% InN multiphase structures by XRD analysis. Samples grown with 40% InN showed single crystalline nature. SAD pattern also indicated that the layers exhibited single-crystal nature with InN compositions of 10, 28, and 49. InN with 10% phase did not show any sign of extra phases, however, the remaining 28 and 49% InN phase was mixed with some other phases indicating extra phases. In the 49% InN sample, tweed nature was observed because of spinodal decomposition showing pseudobinary system of GaN–InN. The solubility of In in GaN is 6% at 800 C, whereas for spinodal composition, it is 22% at the same growth temperature. Moon et al. [1051] studied the suppression of In segregation by use of MQW structures. The segregated In on the surface of the well disturbs the abruptness of the
j687
j 3 Growth and Growth Methods for Nitride Semiconductors
688
GaN
105
Intensity (au)
104
103
In0.53Ga0.47N 102
101
100
–8000 –6000 –4000
–2000
0
2000
4000
6000
Δθ (arcsec) Figure 3.214 X-ray rocking curve of In0.53Ga0.47N/GaN double heterostructure, it shows no sign of InN phase [1049].
interface between the well and barrier and diffuses into the barrier. As-grown films exhibited abrupt and flat interfaces, as revealed by a TEM analysis, the sharpness of which degraded as the interruption time was increased. From XRD analysis, the In composition for InGaN layers in InGaN/GaN MQW was measured 27, 25, 22, and 16% for the growth interruption times of 0, 0.5, 1, and 2 min, respectively. H2 purging eliminates impurities such as C, O, and H at the interface. Again using MQW structures, Moon et al. [1052] reported that randomly distributed indium-segregated regions, which were formed near the upper interface of the InGaN well layers during the subsequent high-temperature growth process were found to act as nonradiative recombination centers and these could be effectively removed by introducing H2 gas during the growth interruption period. McCluskey et al. [1053] observed phase separation in InGaN/GaN quantum well structures grown by OMVPE, which consisted of In0.27Ga0.73N (20 Å)/GaN between a 4 mm GaN : Si layer on sapphire and a 0.2 mm thick GaN : Mg layer. The phase separation was found in the QW structure after annealing at a temperature of 950 C for 40 h, which is related to In-rich InGaN phases in the active region. Optical absorption studies indicated the absorption threshold at 2.65 eV corresponding to InGaN phase with x ¼ 0.35 in the annealed sample. However, XRD analysis showed slightly higher In composition in the sample. In TEM analysis, voids and precipitates were seen. Moire fringes were observed in the In-rich InGaN precipitates, which are due to the difference between the lattice constants of the precipitates and neighboring regions. Energy dispersive X-ray (EDX) analysis revealed that in the precipitates, the In content is 40%, which is much higher than the surrounding areas.
3.5 The Art and Technology of Growth of Nitrides
In another investigation of MQW structures, Narukawa et al. [1054] observed a high density of dark spots in the well region by TEM analysis with a diameter in the range of 2–5 nm, with most being 3 nm. The origin of dark spots was from selfassembled isotropic dotlike structures with an areal density of dots of 5 · 1011 cm2. The observed In composition was higher than that in the neighboring well region. These observations are very similar to those by Chichibu et al. [1055] in In0.2Ga0.8N for which the experimental artifacts were ruled out. The arisen quantum dots in the well are due to compositional fluctuations because of the intrinsic nature of the InGaN alloy unlike interface fluctuations in the GaAs/AlAs and CdSe/ZnSe quantum wells. Cho et al. [1056] also studied In clusters in the QW structures. TEM analysis showed that as-grown samples exhibited a strong lateral variation in contrast to growth interrupted samples, owing to the presence of indium clusters in the QWs. Similarly, energy filtered TEM (EFTEM) analysis showed clear clusters of Indium in the as-grown samples, as compared to that of growth-interrupted samples. Line scan In composition profiles showed the In compositional fluctuation to be between 5 and 40% in the as-grown samples, which provided direct evidence of the strong indium clustering. In the growth interruption case, samples with 31% composition (with 5 s interruption) and 24% composition (with 30 s interruption), the In compositional fluctuation could not be detected. This does not, however, mean that fluctuations do not occur with growth interruption. It simply means that fluctuations are not as extensive and interruption suppresses them. Again employing TEM, Ruterana et al. [1057] investigated the different strain inhomogeneities introduced by compositional inhomogeneities in InGaN QWs. The OMVPE growth on sapphire included thick GaN buffers and nominally 2.4 nm InGaN QWs. The In content and thus the compositional inhomogeneities were therefore determine, as shown in Figure 3.215. Those authors have found differing thickness in QWs with each subsequent QW being thinner from 3.4–1.7 nm. They also reported nominal In incorporation of 17% but strong segregation with areas between 10 and 45% based on TEM analysis. The overriding conclusion was that the indium has segregated into quantum dots (or quantum boxes) with 2 nm height and 3 nm width. 3.5.15.2.3 Surface Reconstruction of InGaN Unlike its GaN counterpart, InGaN surface has not been heavily reported on. However, sufficient surface studies exist to draw conclusions on the surface reconstruction of In rich InGaN surface and its surface morphology when grown by RF MBE [1058]. In addition to the standard questions dealing with morphology of N- and Ga-polarity films, the question of In from the point of view of surfactant must also be considered. As in the case of GaN, InGaN grown on the metal-polarity surfaces tend to be smoother and In acts as a surfactant. Again, as in the case of GaN, the N-polarity films are rough in nature. Metal-rich metal-polarity (0 0 0 1) (cation-polarity) surface tends to rough surface as one deviates from the metal rich toward the N-rich p conditions. ffiffiffi pffiffiffi Near the smooth/ rough transition of InGaN(0 0 0 1) growth, a bright ð 3 3Þ has been observed. The same surface structure is also observed when the metal flux is turned off while
j689
j 3 Growth and Growth Methods for Nitride Semiconductors
690
Figure 3.215 (a) TEM image of a sample with three InGaN quantum wells with a nominal In molar fraction of 17% and thickness of 2.4 nm. The arrow tips show the local regions with contrast change corresponding to variations in strain. (b) Typical HREM image of the QWs in area close to edge encompassing the maximum thickness of 7 nm. Courtesy of P. Ruterana and Ref. [1057].
pffiffiffi pffiffiffi the N flux is maintained at the growth temperature. The ð 3 3Þ reconstruction in fact persists during cooldown at temperatures as low as 300 C. Chen et al. [1058] noted that the In atoms occupy the top two surface layers when Ga-polarity is used, see Figure 3.216. Based on theoretical results for the surface structures, they argue that strain arising from the presence of In in the second (S3) layer leads to the formation of small pits on the surface. Inside, and at the border of (a)
A1 S3
(0 0 0 1)
(b)
(0 0 0 1)
In N Ga
Figure 3.216 Basic structure of In-rich InGaN surfaces: (a) (0 0 0 1), assuming full coverage of the top hexagonal layer by In, and (b) ð0 0 0 1Þ. The A1 and S3 layers are indicated in (a) [1058].
3.5 The Art and Technology of Growth of Nitrides
the pits, the surface concentration of indium is found to be larger than that far from the pits, which may lead to inhomogeneous indium incorporation in the bulk film. Typically, the InGaN films are grown on GaN buffer layers, which in turn are grown on generally SiC or sapphire substrates. The growth procedures employed for GaN are detailed in Section 3.5. Due to the high vapor pressure of N on InN, relatively large, in relation to Ga, desorption rate of In, and relatively small binding energy of In with N, as compared to Ga and N, the InGaN films are grown at relatively lower temperatures, typically between 550 and 650 C. These substrate temperatures represent a reduction of about 100 C, in some cases higher, as compared to GaN. In the authors laboratory, the molar fraction of InGaN is controlled with control of In and Ga with abundance of N. In this growth mode, the affinity of Ga and N is such that the Ga flux on the surface is satisfied with N first and the remaining N is available for In for InN formation. The substrate temperature is such that accumulation of In on the surface, beyond the surface layers, is avoided. The molar fraction is dependent of the substrate temperature and higher the substrate temperature, the larger the In flux must be to maintain the same In molar fraction. The general issue of surface In coverage has been addressed well by Chen et al. [1058] by employing a combination of Auger and scanning tunneling microscope (STM) probes. As discussed in Section 3.4.2.9.4, the Ga-rich growth conditions on (0 0 0 1) surface leads to smoother surface in MBE growth of GaN than N-rich. In addition, the ð0 0 0 1Þ surfaces are always relatively rough regardless of whether Ga- or N-rich conditions are used. In the case of InGaN, the role of In on the surface morphology must also be considered. For device structures requiring smooth heterointerfaces, smooth surfaces may be required in which case, as in the case of GaN, it is imperative that conditions leading to smooth InGaN surfaces are found and the smooth to rough transition point determined. Widmann et al. [1059] reported that indium atoms on the surface act as a surfactant, keeping the growth in the smooth regime when the gallium flux is slightly reduced below the transition flux. Feenstra et al. [1060] investigating N-polarity InGaN on GaN ð0 0 0 1Þ did not observe such a surfactant effect on this nominally rough surface. To gain insight on this issue on both the cation and anion surfaces, Chen et al. [1058] undertook a detailed study on the smooth/ rough transition of InGaN growth on both the ð0 0 0 1Þ and (0 0 0 1) faces, the results of which are shown in Figure 3.217. In experiments which dealt with both metal-polar and N-polar surfaces, the nitrogen flux was kept constant, followed by exposing the surface to a certain indium flux while adjusting the gallium flux to determine the smooth to rough transition point. The dashed lines in Figure 3.217 show where the total metal flux (indium gallium) is constant. Figure 3.217a shows the case for the (0 0 0 1) face where it was found that when the indium flux is applied the gallium flux can be greatly reduced (by an amount considerably greater than that of the added indium flux) before the growth transition to a rough. This is interpreted as evidence that indium on the (0 0 0 1) surface serves as a surfactant, consistent with the observations of Widmann et al. [1059]. In contrast, Figure 3.217b shows that for the ð0 0 0 1Þ face, even when a large indium flux is applied, the gallium flux can only be reduced slightly before the growth becomes rough.
j691
j 3 Growth and Growth Methods for Nitride Semiconductors
692
4 (0 0 0 1)
2 Smooth Smooth
1
In flux (1014 cm–2s–1)
In flux (1014 cm–2s–1)
(a)
0
(0 0 0 1)
(b)
0 3
Smooth Rough 2
1
Rough 0
1
2
Rough
3
Ga flux (1014 cm–2s–1)
0
1
2
3
Ga flux (1014 cm–2s–1) Figure 3.217 Smooth/rough transition on (a) (0 0 0 1) cation face (metal-polar) and (b) ð0 0 0 1Þ anion face (N polar) for a fixed nitrogen flux and a substrate temperature of 600 C. Experimental data are shown with dots representing the experimental data, each with an
error bar. A dashed line is drawn in each figure (-1 slope) for comparison denoting the line with constant total metal flux. To the right of the transition lines, demarked by solid lines, the growth is smooth [1058].
As mentioned above, the competition between Ga and In is really lopsided in favor of Ga so much so that under typical Ga rich conditions In incorporation is not possible. This disparity is made worse because the indium atoms tend to segregate to the surface so that when there is abundance of gallium, the indium atoms do not incorporate. To incorporate In, the gallium flux must be lower than the transition flux of gallium in the absence of indium [1061,1062]. Consequently, to grow InGaN with significant indium content, requires that growth on the (0 0 0 1) face must occur in the smooth regime, whereas it can take place in the rough growth regime for the ð0 0 0 1Þ face. In concert with GaN, the (0 0 0 1) surface is better suited for better quality InGaN growth. The ensuing smooth surface on this polarity bodes well with faster lateral surface diffusion (up to a limit, as excessive diffusion is not desirable either) during growth. Additional insight can be gleaned from the surfactant effect on the (0 0 0 1) surface in that more nitrogen would have to desorb from the surface when the gallium flux is reduced to maintain stoichiometric growth. Moreover, now the total metal flux is lower than that during GaN growth without indium. This increased desorption might result from either higher nitrogen surface diffusivity or higher nitrogen surface concentration. In either case, the nitrogen atoms have more opportunity to meet and form molecules and then leave the surface. For GaN growth, it was pointed out that nitrogen accumulation leads to paffiffiffi higher pffiffiffi diffusion barrier and hence a rough growth [256]. With a nitrogen-rich ð 3 3Þ reconstruction, however, smooth surfaces can be obtained, indicating that indium lowers the diffusion barrier even when the surface has high nitrogen concentration. In PA MBE of GaN surface reconstruction is generally not observed during growth. Although a 2 2 reconstruction occurring during growth [1063–1065] has been reported, this has later been attributed to arsenic present in the growth
3.5 The Art and Technology of Growth of Nitrides
Figure 3.218 RHEED pattern along the ½1 1 0 0 azimuth on InGaN at the growth temperature of 540 C; (a) with In and Ga fluxes as well as the N flux on and (b) with metal fluxes off and N flux on. The InGaN layer was grown under N-rich conditions.
environment [1066]. It should be mentioned that for growth ammonia as the nitrogen source, a different, intrinsic 2 2 structure occurs [1067]. At least in the case of Lin et al. [1063], the system used for GaN was previously used for GaAs and it is conceivable the As p atoms have been present. Unlike GaN, when grown on ffiffiffi pcould ffiffiffi (0 0 0 1) surface, a ð 3 3Þ surface reconstruction is observed in authors laboratory as well as that reported in by Chen et al. [1058] with no As present in the environment. In fact, in the case of the authors laboratory, the system used for this experiment had never been exposed to As. While only a 1 · 1 structure is observed when metal and N fluxes are on, when the metal fluxes are turned off, a 3x reconstruction occurs as shown in Figure 3.218. It should be stressed that the Nrich growth conditions were employed during the InGaN growth on GaN templates, which were grown under Ga-richpconditions. ffiffiffi pffiffiffi Further, the GaN barrier layers grown on InGaN quantum well with ð 3 3Þ surface reconstruction maintained the pffiffiffi pffiffiffi ð 3 3Þ surface reconstruction even at the growth temperature with metal fluxes off but N flux on. Figure 3.218 shows the corresponding RHEED pattern, which is a 3x structure when viewed along the ½1 1 0 0 azimuth and 1x structure along the ½1 1 2 0 azimuth, the latter not shown. According to Chen et al. [1058], this reconstruction is observed when the Ga flux is near or below the transition point between rough and smooth growth in the presence of indium. In addition, Chen et al. [1058] have varied the growth conditions and surface stoichiometry in an effort to determine the geometrical structure of the surface reconstruction. In a series of experiments, indium atoms were deposited onto a (0 0 0 1) smooth GaN surface up topaffiffiffifewp monolayers (1 ML ¼ 1.14 · 1015 atoms cm2) ffiffiffi 1 at a rate of 0.5–2 ML min . No ð 3 3Þ reconstruction was observed during the deposition nor was it observed when the same process was repeated with indium deposition that pffiffiffi pffiffiffi carried out at the growth temperature. These observations indicate pffiffiffi p ffiffiffi ð 3 3Þ reconstruction is not due to an adlayer of indium with ð 3 3Þ structure on top of a gallium-terminated GaN surface. However,pif RF ffiffiffi thepnitrogen ffiffiffi plasma source was turned on when indium was deposited the ð 3 3Þ appeared very quickly. The same result occurred in the nitrogen-rich growthpregime ffiffiffi pffiffiffiwhen indium was deposited during the GaN growth. For the latter case, the ð 3 3Þ was clearly seen when about 1/10 ML of indium was deposited, the intensity reaching a
j693
j 3 Growth and Growth Methods for Nitride Semiconductors
694
maximum when 1/3 ML of indium was deposited. If the gallium flux is well above the smooth to rough transition point, the intensity of the reconstruction gradually decreases and finally disappears when more indium is deposited. Moreover, when GaN is grown in the regime (without indium) followed by deposition of pffiffiffigallium-rich pffiffiffi indium atoms, a ð 3 3 Þ reconstruction does not result. Chen et al. [1058] found pffiffiffi pffiffiffi the ð 3 3Þ reconstruction to persist when the nitrogen plasma is onpwhile ffiffiffi pthe ffiffiffi metal fluxes are off. Also, deposition of In on GaN surface did not lead to a ð 3 3Þ pffiffiffi pffiffiffi reconstruction with N source off, but a ð 3 3Þ reconstruction was observed when the N source is turned on. It can, pffiffitherefore, ffi pffiffiffi be stated that N-rich conditions are required for the formation of the ð 3 3Þ structure. In short, it can be stated that pffiffiffi pffiffiffi the ð 3 3Þ reconstruction forms under N-rich conditions and contains about 1/3 ML of indium p atoms. ffiffiffi pffiffiItffi is worth noting that growth can still be smooth with the nitrogen-rich ð 3 3Þ reconstruction, which implies relatively low surface diffusion barriers for this structure. Chen et al. [1058] also employed scanning tunneling microscopy to study the structure of the InGaN(0 0 0 1) surface, as shown in Figure 3.219. The film shown in Figure 3.219 was grown at 610 C, with Ga and In flux rates of 1.7 · 1014 and 4.8 · 1013 cm2 s1, respectively, well above the smooth to rough transition line. From prior theoretical work by the same group one expects the possible presence of surface structures containing In atoms either in the top surface layer or in the top two layers [1068], as illustrated in Figure 3.216a. In Figure 3.219 in the lower part of the image a region of uniform 1 · 1 corrugation is visible. The structure of this region is
Figure 3.219 STM image of InGaN (0 0 0 1) surface containing 0.9 0.2 ML of indium. Image was acquired at a sample voltage of 11.25 V, and with tunnel current of 0.075 nA. Gray scale range is 0.5 Å. Different surface regions are labeled A1 and A1 þ S3 [1058].
3.5 The Art and Technology of Growth of Nitrides
assigned to having In only in the top layer, which is referred to as the A1 phase. The bright corrugation maxima observed elsewhere in the image are attributed to In atoms in the S3 layer. The height of the observed bright maxima, typically 0.2 Å above the nominal height of the 1 · 1 region, is consistent with theoretical result of 0.30 Å for the change in surface height caused by a second layer In atom. The small black pits appearing on the surface appear dark (lower surface height) for both positive and negative sample bias voltage, indicating that they are some type of surface vacancy island. Another salient feature in Figure 3.219 is that the indium concentration in the second layer is higher around pits. This inhomogeneous surface In concentration may contribute to the formation of the widely observed indium compositional fluctuation in the bulk InGaN discussed in detail in Volume 2, Chapter 5. The In compositional fluctuation could be grown in from the inhomogeneous surface, caused by the processes of In surface segregation [1068], and formation of a strained surface layer owing to lattice mismatch between InN and GaN. This strain is even larger than typical from bulk properties of these binaries because the In surface concentration is much higher than its bulk concentration. 3.5.15.3 Growth of AlInN Interest in AlInN stems from its tunable bandgap by compositional change during deposition, as in the case of InGaN with the added benefit that the Al0.82In0.18N composition the alloy lattice matches GaN. The alloy for this composition has a bandgap of 4.7 eV, which is larger than that of GaN, and forms a barrier layer for modulation-doped structures. Lattice matching has added consequences in polar semiconductors such as nitrides in that piezoelectric-induced polarization due to mismatch vanishes leaving only the spontaneous polarization owing to compositional grading across the interface of an Al0.82In0.18N/GaN heterojunction. Similar to the case of InGaN, the bandgap and particularly the bandgap bowing parameter of this ternary remained elusive and confusing for quite some time owing in part to the incorrect assignment of the InN fundamental gap. Details of the bowing parameter and related issues are discussed in detail in Section 1.5.3. Moreover, the particulars of the polarization charge involving this ternary together with GaN are treated in Section 2.12. The discussion here will be limited to growth issue. The potential applications of Al1xInxN include, but not limited to cladding layers that are lattice matched to GaN or as an active layer for LEDs and LDs emitting in the ultraviolet (UV) to infrared (IR) region. This alloy as in the case of other nitride semiconductors has been grown primarily by either OMVPE or MBE. The disparity between InN and AlN in terms of atomic radii and binding energy of N does not set the stage for an easy task. The relatively high equilibrium pressure of N on InN necessitates the relatively low temperatures be used for growth. However, the early growth efforts relied on sputtering such as magnetron RF sputtering [1069]. On the sputtering side, thin films of III–V nitride semiconductors (AlN, GaN, InN), mixed-crystalline films (AlxIn1xN), as well as multilayered films (GaN/InN)n were grown by RF magnetron sputtering at low substrate temperatures below 500 C [1070]. These films were characterized by X-ray diffraction, Raman scattering,
j695
j 3 Growth and Growth Methods for Nitride Semiconductors
696
optical absorption, and electrical measurements. Dependence of the bandgap energy of AlxIn1xN on composition x was determined. On the OMVPE side, ternary AlInN thin films were grown using, in one case [1071], a horizontal atmospheric pressure OMVPE reactor. In contrast to extensive efforts invested in Ga1–yInyN by OMVPE and Al1-zGazN for optical device applications such as light-emitting diodes (LEDs) and laser diodes (LDs), efforts, particularly early on, expended for AlxIn1xN has been sparse. This is to a great extent owing to the difficulties experienced in the growth of high-quality AlxIn1xN. This ternary is also characterized by its thermodynamic instability. The unstable region in mixing has been calculated from the free energy of mixing using the strictly regular solution model [1072]. From this calculation, one can conclude that the ternary InAlN always shows, InGaN sometimes shows, and GaAlN never shows unstable mixing region at temperatures below 3000 C. As is typical, the precursors for Al and In metals were TMA and TMI, respectively, and ammonia (NH3) for N source. To reduce any parasitic chemical reactions, the TMA, TMI, and NH3 were mixed at the entrance of the reactor chamber. The substrate temperature range, as determined by measuring the temperature of the susceptor on which the substrate rests during growth, was 650–900 C. Sapphire (Al2O3) substrates with different orientations have been used. In another growth experiment with OMVPE, Yamaguchi et al. investigated growth-related processes as well as the determination of the optical properties of this alloy [1073]. In this investigation, Yamaguchi et al. prepared Al1xInxN films covering a composition range of 0.14 · 0.58 on GaN buffer layers by OMVPE on c-plane sapphire substrates. The growth temperature was 720 C. The FWHM in 2y obtained from an o–2y scan of the (0 0 0 2) XRD peak (denoted by D2yc) increased from about 10 arcmin for AlN to over 60 arcmin for the largest In fraction (x ¼ 0.58). The dispersion in 2y is indicative of the degree of distribution of the lattice constant c, namely, the fluctuation of the alloy composition x in AlxIn1xN. The FWHM in o obtained from an o-scan of the symmetric (0 0 0 2) peak (Y-axis, denoted by Doc) ranged from about 10 arcmin for x ¼ 0 and 40 arcmin for x ¼ 0.58. The dispersion in the o-scan of the ð1 0 1 0Þ peak (R-axis, denoted by Doa) range from 25 arcmin for AlN (x ¼ 0) to about 40 arcmin for x ¼ 0.58. The spreads depicted by Doc and Doa indicate the degree of tilting and twisting components of the crystalline mosaicity, respectively. Lukitsch et al. [1074] prepared epitaxial AlxIn1xN thin films with 0 x 1 by plasma source molecular beam epitaxy on sapphire (0 0 0 1) substrates at a low temperature of 375 C. In-plane orientation of AlxIn1xN on c-plane sapphire was investigated by both reflection high-energy electron diffraction and X-ray diffraction measurements with the result being nitride [0 0 0 1]||sapphire[0 0 0 1] and nitrideh0 1 1 0i||sapphire h2 1 1 0i. This means that, as in the case of GaN, the c-plane of nitride lattice is rotated by 30 with respect to the c-plane of sapphire lattice to reduce the lattice mismatch. But, the broad and diffused spots in RHEED images indicate that the nitride films have some degree of crystalline mosaicity, both along the normal and in the plane of the film. The degree of crystalline mosaicity and the compositional fluctuation was reported to increase with decreasing x (increasing InN fraction). Similarly, XRD FWHM,
3.5 The Art and Technology of Growth of Nitrides
D(2y) in y 2y scan and D(o) in rocking curve scans, both increased with increasing InN composition. The D(2y) widening is indicative of the degree of distribution of lattice constant c, which has been construed as varied alloy composition. The D(o) widening indicates the dispersion in the degree of tilt of the c-axis throughout the mosaic structure of the film. These observations are similar to the high-temperature (720 C) grown AlxIn1xN films on GaN by organometallic vapor phase epitaxy [1075]. The direct energy bandgap, determined using optical transmission and reflection measurements, of the alloy was used to comment on the bowing parameter discussed in Section 1.5.3. Resistivity and Hall-effect measurements indicated n-type conduction with carrier concentrations n 1019 cm3 for In-rich alloys and n 1010 cm3 for Al-rich alloys. 3.5.15.3.1 Miscibility Gap in InAlN Phase separation in InGaN has been covered with more rigor than in AlInN and AlGaInN alloys, in part because of the critical role InGaN plays in light emitters. There are a few theoretical studies of the thermodynamic stability of the AlInN alloy and all of them use a very simplified model. Matsuoka [1076] expanded the material system of interest in terms of decomposition to include Al in addition to Ga-containing alloys formed with InN, namely the wurtzite quaternary alloy system. The unstable region in mixing was calculated from the AlxGa1-xyInyN free energy of mixing using the strictly regular solution model. The interaction parameter used in the calculation of Matsuoka was obtained using the delta-latticeparameter method with the result that AlInN always, InGaN sometimes and AlGaN hardly ever, has an unstable mixing region at the temperature below 3000 C. Unlike GaN, relatively less is known about the other binaries and much less about the ternaries and quaternary. The least is known about AlInN and the quaternary, particularly on the theoretical side concerning electronic properties [1077,1078]. To make matters more challenging, the experimental results, progressively less so, are often somewhat contradictory [1039,1079–1081]. For structural properties of AlxIn1xN, there is not much theoretical work which discusses, in an equal footing, both a reasonably sized model supercell and the statistics of the alloy. Rigorous calculations are also available in regard to thermodynamic properties of the AlxIn1xN alloy. Teles et al. [1082] present a rigorous theoretical study of thermodynamic, structural and electronic properties of ternary c-AlxIn1xN bulk nitride alloys. The calculations performed for AlxIn1xN are based on an ab initio pseudopotential plane wave method, within the framework of the density-functional theory (DFT) and the local-density approximation (LDA), the so-called Vienna ab initio simulation package (VASP), and a generalization of the quasi-chemical approach combined with a cluster expansion of the thermodynamic potentials, as was performed for AlGaN and InGaN [1082]. It has been demonstrated that this model is able to successfully describe the physical properties of group III nitride alloys, even including those involving boron, such as BGaN and BAlN [1082–1085]. Focusing on the topicathand, weare goingto focusinparticular onthephasediagram of thec-InxAl1xN alloy, its chemical bonds, through first- and second nearest neighbor distances and bond angles and the bandgap behavior as functions of the alloy composition x. The reader is referred to the paper by Teles et al. [1086] for a more in-depth treatment, but suffice it to state that the mixing free-energy DF is calculated as a
j697
j 3 Growth and Growth Methods for Nitride Semiconductors
698
function of x and T at a fixed pressure, which allows us to access the temperature–composition (T–x) phase diagram and obtain the critical temperature for the miscibility. As in the case of the calculations of Ho and Stringfellow [1015], the zinc blende system is modeled as it represents well the thermodynamic properties of the wurtzitic system. The calculations of the mixing free energies were carried out by combining the cluster expansion method within the framework of the generalized quasichemical approximation (GQCA) and the self-consistent total-energy pseudopotential (the VASP) calculations. In the calculations of Teles et al. [1086], eight-atom supercells have been used as the basic clusters to describe the fully relaxed alloys. The structure of each cluster has been optimized in terms of its lattice constant by minimization of the total energy. The calculations indicated a propensity for substantial deviation from randomness toward phase separation. The T–x phase diagram for the InxAl1xN ternary alloys calculated within the GQCA and the ab initio total-energy method is shown in Figure 3.220, where the binodal and spinodal curves are depicted with solid and broken lines, respectively. Note that the spinodal curve marks the equilibrium solubility limit, that is, the miscibility gap. The behavior displayed in Figure 3.220 is similar to that observed in InGaN due to a rigorous thermodynamics calculations by Teles et al. [1087]. One salient exception is that the critical temperature for AlInN, Tcrit ¼ 1485 K, is higher than that for InGaN. The results depicted in Figure 3.220 indicate that for typical growth temperatures,
1600 Inx Al1–xN Binodal
1400
Temperature (K)
1200 Spinodal
1000 800 600 400 200 0 0.0
0.2
0.6 0.4 Composition, x
0.8
Figure 3.220 Calculated T–x phase diagram for unstrained c–InxAl1xN. Binodal curve: solid line; spinodal curve: dashed line [1086].
1.0
3.5 The Art and Technology of Growth of Nitrides
phase separation for a wide range of compositions should be observed if grown under thermodynamical equilibrium. Considering a growth temperature of, for example, T ¼ 1000 K, a large decomposition tendency is clearly seen for In molar fraction between 15 and 79%. This result is in good agreement with experimental findings that show a tendency of phase separation for x > 0.17 [1039]. 3.5.15.4 InGaAlN Quaternary Alloy This quaternary material is explored for the simple reason that its bandgap can be changed while maintaining lattice match to GaN, which would pave the way to reduced residual stress and thus associated piezoelectric effects as well as alleviating cracking problems in high mole fraction and/or thick AlGaN layers used in MODFETs and lasers. Use of this quaternary material allows almost independent control of the band offset in AlInGaN-based heterostructures. With the advent of UV emitters and detectors, near or in the solar-blind regions, details of which are discussed in Volume 3, Chapter 4, interest in the AlxGa1xyInyN quaternary increased. In this vein, LEDs emitting at 280 nm have been demonstrated using AlGaInN/AlN superlattices [1088]. Moreover, laser diodes emitting at 366.4 nm have also been demonstrated using AlGaInN quantum wells as active media [1089]. AlGaInN quaternary alloys potential might reduce the differential dilatation coefficient in heterostructures, which could be an important advantage in epitaxial growth. The carrier localization induced by In can also lead to higher quantum efficiencies as opposed to AlGaN for wavelengths shorter than that associated with the fundamental gap of GaN, details of which are discussed in Volume 2, Chapter 5. However, the growth of quaternary AlGaInN is extremely challenging due to the different bond lengths and decomposition temperatures of the binary compounds making up the quaternary, namely, AlN, GaN, and InN. Further, the vastly different surface mobility and desorption temperature of the impinging species make matters even more complex. The optimal growth temperature is paradoxical, as aluminumbased compounds generally require higher growth temperatures and In-based ones require lower temperatures. Higher temperatures are also desirable for reducing the O incorporation in the growing film as oxides of Ga and In desorb from the surface. Higher temperatures are also desirable for reducing the O incorporation in the growing film as oxides of Ga and In desorb from the surface. The growth temperature will therefore govern the limits of In and Al incorporation into the AlGaInN quaternary alloy [1090]. Furthermore, there are fundamental issues such as immiscibility, which have been predicted for the InN–AlGaN system [1072], and must be dealt with. The feasibility of AlGaInN has been demonstrated using OMVPE [1091,1092]. This quaternary has also been grown by MBE, which brings into the arena additional strengths, particularly in fundamental investigations of growth and surface structure [1093–1095]. A detailed analysis of MBE growth kinetics and a systematic study of material properties as a function of composition has been reported using plasma-assisted MBE for the controlled growth of quaternary AlxGa1xyInyN [1096]. Monroy et al. [1096] prepared AlxGa1xyInyN (0 < x < 0.5, 0 < y < 0.2) epilayers with thicknesses in the range of 0.2–0.7 mm by MBE on GaN templates, 2–4 mm thick
j699
j 3 Growth and Growth Methods for Nitride Semiconductors
700
that were grown on c-sapphire by OMVPE. The growth rate was fixed to 0.20 ML s1 for all the alloy compositions. Prior to AlGaInN growth, a thin (10 nm) GaN buffer layer was deposited at 730 C. Quaternary compounds were then grown at substrate temperatures in the range of 590 and 650 C. The growth kinetics was analyzed in situ by RHEED. To remind the reader, the GaN and AlN layers are typically grown in the 700–800 and 750–900 C ranges, respectively, but In desorbs at temperatures above 550 C. The upper limit for the growth temperature of AlxGa1xyInyN is thus determined by In incorporation in AlGaN while maintaining quality that sets the window to 650–610 C. Because the incorporation of N is more favorable adjacent to Ga- and Al-occupied sites, to get In incorporation the sum of Ga and Al fluxes, FGa þ FAl, must be less than the nitrogen flux, FN, while the excess metal is provided by the In flux. In other words, the compositional control is achieved by controlling the fluxes of Al, Ga, and N. The N flux control is not as well developed as the other when using RF nitrogen sources, unless special precautions are taken into consideration. Monroy et al. [1096] determined the adequate amount of In flux by wetting of the AlGaN surface in such a way to produce stable one ML thick In layer [1094,1097]. Evolution of RHEED specular beam intensity for AlGaN and also Al0.36Ga0.61In0.03N (Al, Ga, and N fluxes are the same for growth of both compounds) was monitored, shown in Figure 3.221, which shows some transient oscillations assigned to the formation of the In film at the surface. The growth is said to proceed with N diffusing through the In film. As the intensity oscillations indicate, the growth proceeds layer by layer. The difference in the growth rates measured from the oscillation period for the quaternary and the ternary compounds, 0.245 versus 0.235 ML s1, respectively, can be used to determine In incorporation. After the initial transient, the RHEED intensity remains stable throughout the growth, indicating the lack of excess In accumulation beyond the one ML stable In layer. This adsorbed In film can be removed from the surface by a postgrowth thermal cycle of 1 min at a substrate temperature of 650 C. The final RHEED image, shown in the inset of Figure 3.221, consists of a streaky pattern, indicative of 2D growth. If the substrate temperature is decreased abruptly at the end of the growth, the In film segregating at the surface does not have sufficient time to desorb. Monroy et al. [1096] also investigated the composition of the quaternary alloy versus Al molar fraction (recall that Al and then Ga bonds with N must be satisfied first before In bonds are served and increasing Al and Ga fluxes would lead to reduced In incorporation everything else being the same) by Rutherford backscattering spectrometry (RBS). To determine the maximum amount of In content that can be incorporated, AlGaInN layers have been grown under In-rich conditions at a fixed substrate temperature and Al flux. When the Ga flux is reduced, the RHEED intensity oscillation indicates no change in the growth rate indicating that In replaces Ga. Naturally, if pushed further, the growth rate decreases because of In incorporation dynamics at that particular substrate temperature. However, the system is still in the In-rich regime and only a limited amount of In is incorporated with the rest segregating to the surface. Figure 3.222 presents the maximum In content measured by RBS as a function of substrate temperature and Al molar fraction. The decrease in
3.5 The Art and Technology of Growth of Nitrides
Figure 3.221 Variation of the RHEED intensity with time when starting the growth of ternary and quaternary alloys at a substrate temperature Ts ¼ 640 C, using the same Al, Ga, and N fluxes. In the inset, photograph of the final RHEED pattern of AlGaInN after 2 h of growth (360 nm). Courtesy of B. Daudin and Ref. [1096].
In incorporation with increasing temperature is common to In-containing nitrides inclusive of InGaN and AlGaInN across the growth methods, both MBE [1098,1099] and OMVPE [1100]. In addition to the temperature, In incorporation also decreases with increasing Al molar fractions at all substrate temperatures, which contradicts a
In molar fraction (%)
50
T s = 610 ºC T s = 630 ºC T s = 650 ºC
40 30 20 10 0 0
20
40 60 Al molar fraction (%)
80
100
Figure 3.222 Maximum In incorporation in the quaternary compound as a function of the Al mole fraction and the substrate temperature TS. Courtesy of B. Daudin and Ref. [1096].
j701
j 3 Growth and Growth Methods for Nitride Semiconductors
702
report on the same material deposited by OMVPE [1101]. The impetus force for In segregation and limitation of its incorporation for both InGaN of AlGaInN requires further research, although this has been attributed to re-evaporation of adsorbed In atoms, this does not apply to the case as there is an equilibrium monolayer of In. It has also been argued that the immiscibility of InN and AlGaN is responsible for limit for In incorporation. A standing model, which accounts for the nature of surface segregation, is preferential segregation that occurs during growth of ternary and quaternary III–V semiconductor alloys and drives one of the column III elements involved in the surface [1102]. This process reflects the competition between strain and/or binding energies (which promotes the driving of certain atoms to the surface) and entropic factors that favor the incorporation of a fraction of these atoms in growing layers that are successively buried below the surface. Body of evidence in arsenides, phosphides, and nitrides suggests that the direction and extent of the segregation process follows the In > Ga > Al order, whereas in antimonides Al tends to segregate (Al > Ga). Specific to the case of AlGaInN quaternary alloys, exchange of In with Ga or Al is energetically favorable due to the different binding energies of, in ascending order, InN (8 eV), GaN (9 eV), and AlN (12 eV) [1103]. It is therefore argued, as alluded to above, the larger binding energy of AlN as compared to that of GaN is most likely the genesis for In segregation with increased Al. Moreover, elastic strain also would favor In segregation because In–N bond is longer than both Ga–N and Al–N. Strain may also be involved. The structural analysis of the quaternary alloy was carried out using HRXRD. The (0 0 0 1) oriented III nitride epilayers present a columnar structure characterized by the average tilt and twist of the columnar domains typical of nitride semiconductors. The broadening of symmetric (000) o-scans is sensitive to the grain size and the column tilt such as the rotation of columns out of the growth axis and has been attributed to screw dislocations with Burgers vector b ¼ h0 0 0 1i. The FWHM of the (0 0 0 2) o-scans of AlxGa1xyInyN lied in the 300–500 arcsec range, which is comparable with 370 50 arcsec obtained in OMVPE GaN templates. This implies that the mosaicity of the quaternary layer is mostly determined by the substrate quality. In contrast, FWHM of the (0 0 0 2) y 2y scans of the alloy increases with Al content (independent of the In content) from 150 to 300 arcsec and compares with 35 4 arcsec for the GaN template. It should be noted that broadening, caused by alloy disorder with increasing Al content, is similar to that obtained in AlxGa1xN alloys grown with the help of In surfactant [258]. It appears therefore that the broadening is mainly caused by Al. The X-ray diffraction broadening for the quaternary layers obtained by MBE are similar to those reported in quaternary layers grown by OMVPE, with compositional range of 0 < x < 0.2 [1104] and 0 < y < 0.06 [1105]. Additionally, the quaternary layers exhibit low-sheet-resistance that could be of importance in reducing the current crowding in deep-ultraviolet lightemitting diodes over sapphire substrates, but it has not been based on fact as yet [1106]. Aumer et al. [1107] reported the OMVPE growth of the quaternary alloy AlInGaN and argued that it is desirable to grow quaternary films at temperatures greater than
3.5 The Art and Technology of Growth of Nitrides
855 C to suppress deep level emissions in the room-temperature photoluminescence using two sets of layers. One set was a sequence of relaxed thick films (430 nm) grown with 11% InN and different aluminum compositions. The second set of films comprises strained thin films (60 nm) with 6–10% InN and up to 26% AlN molar fractions. First, a 35 nm AlN buffer layer is grown by atomic layer epitaxy (ALE), followed by standard OMVPE growth of 2 mm of GaN at which point the growth of the quaternary alloy is initiated. The relationships between composition and bandgap (or lattice constant) can be predicted following a routine that was originally developed for the InGaAsP system [1108], as discussed in Section 1.5.4 and Figures 3.140–3.142. Here, the experimental efforts dealing with bandgap determination of the quaternary material is discussed. The bandgaps of the strained and relaxed quaternary films are plotted versus the aluminum content in Figure 3.223a and b.The bandgap variation with aluminum for
3.30 (a)
3.28
Bandgap (eV)
3.25 3.23 3.20 3.18 3.15 3.13 3.10
0
0.02
0.04 0.06 x (%AlN)
0.08
0.1
3.50 (b)
Bandgap (eV)
3.40 3.30 3.20 3.10 3.00 2.90 2.80
0
0.05
0.1
0.15
0.2
0.25
0.3
x (%AlN) Figure 3.223 (a) Bandgap versus x for thin films of AlxInyGa1xyN (triangles, y ¼ 0.06; squares, y ¼ 0.08). (b) Bandgap for AlxIn0.08Ga0.92xN (triangles, thick films; squares, thin films; solid line, theoretical Eg using bAlGaN ¼ 1, bInGaN ¼ 4.5, and bAlInN ¼ 5). Courtesy of S.M. Bedair and Ref. [1107].
j703
j 3 Growth and Growth Methods for Nitride Semiconductors
704
thin films with less than 9% indium is nearly linear. Thin films with that composition experience biaxial strain from the underlying GaN. As more aluminum is added to the films, two competing effects alter the bandgap. Increasing aluminum content results in films with a larger bandgap; however, the smaller lattice constant reduces the biaxial strain in the thin film, thereby reducing the strain-related blue shift of the bandgap. These effects are observed by differences in the bandgap of thin and thick films. Chen et al. [1109] reported on the growth of quaternary AlxInyGa1-x-yN layers using a slight variant of OMVPE-based pulsed ALE. TMA, TMI, and TEG were pulsed during the growth at a temperature of 750 C while keeping the ammonia flow rate constant throughout the growth. It was found that the existence of ammonia increases three-dimensional growth, which results in increasing the density of localized tail states because of stronger alloy compositional fluctuations. The layers exhibited intense room temperature photoluminescence spanning from 320 to 350 nm. Lima et al. [1110] reported on the AlInGaN growth on a GaN buffer layer by plasmainduced MBE. They obtained different alloy compositions by varying the growth temperature with constant Al, In, Ga, and N fluxes because the incorporation rate of In and, to a lesser extent, Ga depends on the substrate temperature, decreasing with increasing substrate temperature. The In content in the alloy, measured by Rutherford backscattering spectroscopy, increased from 0.4 to 14.5% when the substrate temperature was decreased from 775 to 665 C. X-ray reciprocal space maps of asymmetric AlInGaN reflections were used to measure the lattice constants and to verify the lattice match between the quaternary alloy and the underlying GaN buffer layer. They found that the alloys with lower In concentrations, that is, grown at a higher temperature, have a smaller lattice constant. The samples were also characterized for their optical emission characteristics [1111] and as expected, the samples grown at higher temperatures, with lower InN component, showed stronger photoluminescence. The following expression can be used to find the lattice constant of the quaternary from the composition and binary lattice constants. aAlx Iny Ga1 x y ¼ xaAlN þ yaInN þ ð1 x yÞaGaN :
ð3:70Þ
This formula is based on the assumption that a solid solution of the binary constituents is present in the quaternary alloy. Having discussed the two ternaries, InGaN and AlInN, which are prone to phase separation, attention can be directed to the lone quaternary system in the nitride semiconductors. Matsuoka [1076] regarded the In1-xyGaxAlyN alloy as a pseudoternary system because mixing of In, Ga, and Al atoms occurs at one of the sublattices. That author employed the strictly regular solution approximation to calculate the immiscibility gap, a method originally applied to a ternary mixture by Meijering [1112] with the assumption of pairwise interactions among nearestneighbor atoms and random mixing of the constituent atoms in a single phase. In this approach, the calculation for the miscibility gap can be focused on the calculation of binodal isotherms.
3.5 The Art and Technology of Growth of Nitrides
AlN
0.8
0.6 3 00 0 oC
0.4
oC 0 1 00 0 0 0 o C 2
Y
0.2
InN
0.2
0.6
0.4
0.8
GaN
X Figure 3.224 Calculated spinodal isotherms for In1xy GaxAlyN system at 0–3000 C for the constant K of 1.15 · 107 cal mol1 Å2.5 [1076].
The calculated spinodal isotherms of In1-x-yGaxAlyN for 0–3000 C are shown in Figure 3.224 for a reasonable value of the parameter K, which enter into the interaction parameter. The regular interaction parameter that enters in the calculation of the molar free energy of mixing is dependent in a complex form on the lattice parameters of the binaries forming the ternary and a constant T, which has the unit of cal/mol Å2.5, for details see [1015,1076]. The region surrounded by the InN–AlN line and isotherms shows the unstable composition of In1-x-yGaxAlyN at the growth method under the equilibrium conditions. As clearly seen, the ternary InAlN alloy always has an unstable region. The AlGaN, however, could be grown at all compositions without instability because the isotherms are parallel to the GaN–AlN line. For InGaN, the unstable composition appears to depends on growth temperature. The tendency toward instability for InAlN and InGaN is determined by the value of each interaction parameter, namely, the difference in lattice constants between the binary compounds. In the calculation mentioned above, only the constant K that enters into the calculation interaction parameter, which, in turn, enters into the determination of the free energy of mixing is uncertain. Therefore, in discussing the unstable mixing region of In1-x-yGaxAlyN, it is imperative that one is clear in regard to the dependence of spinodal isotherms on the K value. Figure 3.225 shows the calculated spinodal isotherms of In1-xyGaxAlyN at 1000 C for the K value of 0.6 · 107–1.0 · 107 cal mol1 Å2.5. As can be seen in the figure, the unstable mixing region becomes narrower as the K value decreases. In particular, this value affects the instability of the InGaN ternary alloy.
j705
j 3 Growth and Growth Methods for Nitride Semiconductors
706
AlN 1000 ºC
0.8
0.6
Y 0.4
0.6 0.8
0.2
1.0
InN
0.2
0.4
0.6
0.8
GaN
X Figure 3.225 Dependence of calculated spinodal isotherms for In1xyGaxAlyN system at 1000 C on the constant K (cal mol1 Å2.5) [1076].
3.5.16 Growth of Quantum Dots
Semiconductor quantum dots (QDs) have been investigated very extensively [1113,1114]. As compared to bulk (3D) materials and quantum well (QW) (2D) structures, QD is the prototype of a zero-dimensional system, which, in the limit, spatially localizes the electrons, and the energy is fully quantized, similar to the atomic system, as shown in Figure 2.17. In addition to quantization, the quantum dots have other benefits such as localization that leads to reduced internal quantum yield degradation. Gerard et al. [1115] pointed out that once the carriers are captured by QDs, they become strongly localized and their migration toward nonradiative recombination centers is made difficult. Furthermore, the increased localization gives rise to increased radiative recombination rates, which brings one to the expected low threshold for lasers that have already been experimentally observed in the InGaAs system. A flurry of interest in low-dimensional GaN and other III nitrides is in part due to a desire to develop new optoelectronic devices with improved quality and wider applications. Development of the light emitters with QDs, for example, is expected to have a lower threshold current in LDs and a higher thermal stability [1116]. The quantization causes the electronic density of states near the bandgap to be higher than in 3D and 2D systems, leading to a higher probability for optical transitions. Furthermore, the electron localization may dramatically reduce the scattering of electrons by bulk defects and does reduce the rate of nonradiative recombination. In the GaN system, additional impetus is due to dots tending to nucleate because of reduced strain and increased nucleation probability, where
3.5 The Art and Technology of Growth of Nitrides
the extended defects are then driven to the surface of dots for more effective annihilation. The size of QDs and of its distribution as well as density over the wafer are important parameters. The typical value of size is on the order of a few nanometer, which necessitates a large assembly of QDs rather than a single one, although probes are continually developed to interrogate individual dots. The fluctuation in dot size produces an inhomogeneous broadening in quantized energy levels and may destroy the very fundamental properties expected from a single QD. The random rather than well-ordered distribution may destroy the coherence of the optical and electronic waves propagating through the structure. Similar to other semiconductor heterostructures, the surface or interface of the QDs must also be free of defects (see recent chapters for details) [1117,1118]. Otherwise, the surface/interface may become the effective scattering center for electrons. In metal dots, minimization of the chemical potential leads to merging of small dots with large ones, which leads to convergence to more or less uniform dots on the surface. Initial variation of the dots size obviously is caused by disorder in the system. In semiconductor dots, however, the surface mobility and mass is not available, and other methods, such as strain and periodic topological features such atomic steps on vicinal substrates, must be used to the extent possible to accomplish this. The holy grail of quantum dots is the nature of the confined states and the resultant density of states. If we consider a semiconductor whose constant energy surface for conduction band in k-space is a sphere, such as the case in GaN, the volume of that sphere in k-space is proportional to k3 in terms of momentum and E3/2 in terms of energy as shown in Figure 2.19a. The density of states associated with this system is proportional to E1/2, again as shown in Figure 2.19a. The area in k-space in an ideal system confined in one direction only (representing quantum wells), which is often the z or the growth direction, is proportional to k2 or E, as shown in Figure 2.19b. The density of states in this case is given by m =ph2 and forms a staircase as shown in Figure 2.19b. If we continue and place a confinement in the x-direction in addition to the z-direction, which represents quantum wires, the line length in k-space is proportional to k in terms of momentum and E1/2 in terms of energy as shown in Figure 2.19c. The corresponding density of states depends on E1/2, again as shown in Figure 2.19c. If confinement is imposed in all three directions, which represent the pseudoatomic or quantum dot state, the energy is discretized in all direction and the resultant density of states takes on a deltalike function in energy, as shown in Figure 2.19d. The quantum confinement effect shifts the bandgap to a higher energy. This shift, called the confinement energy, depends on the size and shape as well as the material properties of both QDs and surrounding matrix. Here, we estimate the confinement energies for two simplified cases: a plate or disk and a sphere. Assuming an infinite barrier, the confinement energy of the ground state for an electron in a rectangular box is given by ! h2 1 1 1 ; ð3:71Þ þ þ E¼ 8 mx d2x my d2y mz d2z
j707
j 3 Growth and Growth Methods for Nitride Semiconductors
708
where dj (j ¼ x, y, and z) are the dimensions of the box and mj are the electron mass in j-directions. For a plate- or disklike dot, in which the in-plane sizes dx and dy are much larger than the height dz ¼ d, the confinement energy is simplified to h2/(8mzd2). The shift in the bandgap is calculated using 1/mz ¼ 1/me,z þ 1/mh,z, where me,z and mh,z are, respectively, the effective masses of the electrons and holes in III nitride along the z-direction. For a cubic box of size d, the confinement energy is given by the same expression, h2/(8mzd2), but with 1/mz replaced by (2/me,t þ 2/mh,t þ 1/me,z þ 1/mh,z), where me,t and mh,t are the transverse effective masses of the electrons and holes. In Figure 9.173, the confinement energies as a function of d for a plate and a sphere have been plotted [1119]. In both cases, the effective masses of 0.22m0 for electrons and 1.1m0 for holes were assumed. For small dots of a few nanometer in size, the confinement energy is very sensitive to the dot size in that when the size decreases from 10 to 2 nm, the confinement energy changes from 20 to more than 1 eV depending on the shape of the dots. The curves shown in Figure 9.176 represent good lower and upper bounds for the effective confinement energy in GaN QDs, both free and embedded in an AlN matrix, if the size is not too small. For electrons, the barrier height in GaN/AlN interface (0.75 DEg or 2.1 eV) [1120] is high and the mass anisotropy is small. For holes, the barrier is lower (0.7 eV) and the mass anisotropy is larger and the effect on confinement energy is much reduced due to the large mass. For a specific GaN QD, the actual energy shift from the bulk value is expected to be within those two bounds mentioned earlier. In the case of self-assembled QDs, the platelike results may be more suitable because the aspect ration is very large. In other cases, such as GaN nanocrystallines, spherelike shape may be more appropriate. To attain the advantages of the QDs, many requirements must be met in material preparation. The most important one is the size and distribution of QDs. Depending on the material and the dot shape, the maximum size should be near or less than some characteristic length of the electrons in the bulk III nitride, such as exciton radius. The typical value is on the order of a few nanometer. With such a small size, the practical applications are thus often associated with a large assembly of QDs rather than a single one. This implies that the size uniformity of the dot assembly is critical. The fluctuation in dot size produces an inhomogeneous broadening in quantized energy levels and may destroy the very properties expected from a single QD. In addition to the size uniformity, the spatial position of each QD is also important in many applications. The random rather than well-ordered distribution may damage the coherence of the optical and electronic waves propagating through the system. Similar to other semiconductor heterostructures, the surface or interface of the QDs must also be free of defects. Otherwise, the surface/interface may become the effective scattering centers to electrons. Fabrication of QD assembly with small and uniform size, attainment of high density, well-ordered placement, and production of defect-free material remain as great challenges today in any semiconductor system, especially in III nitride materials. The majority of III nitride QDs are grown by MBE and metalorganic chemical vapor deposition (OMVPE). Due to the lack of a suitable material both lattice and thermally matched to GaN, III nitride heterostructures are commonly grown on
3.5 The Art and Technology of Growth of Nitrides
foreign substrates, as mentioned above [1121]. Sapphire (a-Al2O3), 6H-SiC(0 0 0 1), or 4H-SiC, Si(1 1 1) and cubic 3C-SiC(0 0 1) (for cubic GaN QDs) have been used for growth and GaN by HVPE is used as a template for further MBE growth. All of these mean that suitable buffer layers must first be grown before quantum dot growth can be attempted. In the case of GaN dots driven by strain, either AlGaN or AlN buffer layers must be used. The composition of the ternary and the thickness of these layers depend on the level of strain desired, as thin barrier layers tend not to relax fully. In the case of InGaN dots, GaN underlayer becomes an option also. In addition to strain driven dots, antisurfactants such as Si (the exact mechanism of which has evolved quite substantially in the case of GaN and is now believed to be due to SiNx formation [1122]) and lithographic processes have been explored. In the latter, lithography in conjunction with anisotropic nature of growth associated with OMVPE has been employed to further reduce the dimensions to what is closer to Bohr radius. There are many other methods among which are as colloidal forms and inorganic matrices. In this paper, however, only GaN dots driven by strain caused by underlying AlN layers will be discussed. Growth of GaN self-assembled QDs on AlGaN, with the aid of a sub- to monolayer Si layer, which are then covered with AlGaN has been reported by Tanaka et al. [1123] and Shen et al. [1124]. Other approaches have been reported by Widmann et al. [1125,1126] and Damilano et al. [1127] who used AlN layers that provide a larger lattice mismatch to GaN than AlGaN, and in turn provide the impetus for a 3D growth. In addition, the surface topology of AlN is smoother, which removes the surface features from being the nucleation sites for dots. Dots have been demonstrated on 6H-SiC [1123,1124] and sapphire (0 0 0 1) [1125,1127]. Blue-light emission has been reported from such QD structures [1127]. By changing the size of the quantum dots, one can in fact tune the color of the emission owing to large polarization-induced band bending [1127]. However, the wavelength would depend on the injection level, as injected carriers tend to screen the polarization charge. The epitaxial growth by MBE and OMVPE is essentially a nonequilibrium process. However, it is very useful to categorize it into three different modes as in the equilibrium theory. The typical growth modes in any epitaxial deposition are Frank–van der Merwe (FV or FM), which results in 2D growth, desired for epitaxial growth with lateral uniformity [1128,1129]. The other mode is VW mode, which results in 3D growth from the get go and typically occurs in metals. The third mode is the Stranski–Krastanov (SK) mode for which strain is the driving force. After an initial 2D growth, called the wetting layer, the built-in strain drives the system to a 2D growth for strain minimization. Obviously, the wetting layer must be grown on a buffer whose lattice constant is smaller than that of the dot material. The SK mode is the exploited for the growth of semiconductor QDs. The investigation of InAs QDs grown on GaAs shows that the 2D to 3D transition in SK mode is in fact a first-order phase transition [1130]. Schematic representation of the aforementioned processes as shown in Figure 3.226. On vicinal substrates that often consist of a staircase of equally spaced steps in tilting direction, one may expect the approximate 2D analogues of all the 3D equilibrium growth modes, for example, FV, SK, and VW shown in Figure 3.227.
j709
j 3 Growth and Growth Methods for Nitride Semiconductors
710
(a)
(b)
(c)
Frank–van der Merwe (FV)
Volmer–Weber (VW)(VW) Volmer-Weber
Stranski–Krastanov (SK) Stranski-Krastanov (SK)
Figure 3.226 Various growth modes occurring in general on exactly oriented templates. (a) FV mode, layer-by-layer 2D growth mode; (b) Volmer–Weber mode applicable mostly to metals and leads to 3D growth, (c) Stranski–Krastanov mode, which is driven by strain and is typical of semiconductors.
Which mode would be in effect is determined by the balance between the interstep energy (laterally between the film and the substrate steps, equivalent to interfacial energy in 3D) and energies of film steps and substrate steps (equivalent to surface energies in 3D) [1131]. For homooepitaxy on a vicinal substrate (Figure 3.227a), one would expect ideal step-flow growth. In the case of heteroepitaxy, there is misfit strain, and depending on the extent of the strain, different modes occur. For example, with wetting layer (stripe in this case), Figure 3.227b the VM mode occurs, and without the wetting stripe, the SK mode occurs (Figure 3.227c). The wetting stripe corresponds to the wetting layer in the 3D case. Experimentally, the growth mode depends not only on the materials of both the epilayer and substrate but also on the growth conditions such as substrate temperature and flow rates of various sources. Essentially, it is the result of a competition between the kinetic energy of adatoms and the free energies of bulk, surfaces, and interface. For a lattice-matched system, in the limit of equilibrium growth, the layer-by-layer growth is favored if the energy of the substrate surface is higher than the sum of the epilayer surface energy and the interface energy. Island or 3D growth can be realized by changing the surface and interface energies. In a lattice-mismatched system, the bulk elastic energy in the epilayer induced by strain plays an important role. Because it increases with layer thickness, a strain
3.5 The Art and Technology of Growth of Nitrides
(a)
(b)
(c)
Frank–van der Merwe (FV)
Volmer–Weber (VW)
Stranski–Krastanov (SK)
Figure 3.227 2D analogues of 3D equilibrium growth mode, which occurs during step flow growth on vicinal substrates. (a) FV mode, rowby-row growth, which is termed the step flow growth. In the case of GaAs, there will be no RHEED oscillations observed in this mode.
Continuous stripes formed under ideal step flow. (b) VM mode, island growth. (c) SK mode, row-by-row followed by island growth. The island being square in footprint is just an assumption for simplicity.
relaxation is expected when the layer thickness is increased beyond a critical value. In fact, the SK mode is only observed when the epilayer is subject to compressive strain. In this case, the stress field tends to force the adatoms to coalesce. The strain energy can be partially released by the formation of islands through elastic relaxation, without any dislocations of the islands. When the epilayer is subject to tensile strain, the growth will continue to be 2D and the strain energy is released through plastic relaxation with the creation of dislocations. The spontaneous growth of QDs by either 3D or SK mode is known as self-organized or selfassembled growth. The methods used to produce GaN and InxGa1xN QDs quantum dots are discussed below. 3.5.16.1 Quantum Dots by MBE Both RF nitrogen plasma [1125,1126,1132–1135] and ammonia (NH3) [1127,1136] have been used as nitrogen sources during the growth of quantum dots. A thin (10–100 nm) AlN buffer layer is grown on substrates followed by the active layer, which can be either a single layer of GaN QDs or repeated layers of GaN QDs separated by AlN or ternary AlGaN spacer layers. The latter is necessary for many practical applications requiring higher QD density. In this case, the AlN is used to not only isolate the adjacent QD layer but also provide a flat surface for the growth of the following QD layer so that topological features do not impede dot formation driven by
j711
j 3 Growth and Growth Methods for Nitride Semiconductors
712
strain. The top layer of QDs may or may not be capped by AlN depending on the measurement to be performed. Extensive investigation of GaN QDs grown by MBE has been carried out by Daudin and coworkers [1125,1126,1132–1134,1137,1138]. Typically, after substrate nitridation, a thin (10–30 nm) AlN buffer layer was grown at a temperature between 500 and 550 C, followed by a thick (0.2–1.5 mm) AlN layer grown at a higher temperature of 650–730 C [1138]. Sometimes the growth of the thick AlN layer was preceded by a thick (2 mm) GaN buffer layer [1132]. The GaN QDs were grown on AlN by depositing two to four MLs of GaN at temperatures ranging from 680 to 730 C. Due to the 2.5% lattice mismatch between GaN and AlN, under the growth conditions that were used, the growth follows a SK mode. After the 2D growth of a GaN wetting layer (about 2 MLs), 3D growth follows, and the GaN QDs are formed [1132]. It was found that the growth mode is sensitive to the substrate temperature. At growth temperatures below 620 C, the growth was purely 2D. Only at the elevated temperatures (680–730 C) did the growth transitions from 2D to 3D, that is, the SK mode, take place [1132]. RHEED is a powerful method for real-time observation of dot formation. The intensity, shape, and rod spacing can be used to monitor the transition from wetting layer to 3D growth where the dots form and evolution of the lattice constant as shown in Figure 3.228. The self-assembled GaN QDs have a disklike shape, or more accurately, a truncated pyramid with a hexagonal base, with a base diameter a few times larger than the height. The dot size and density depend on growth condition, deposition time, as well as postgrowth treatment. AFM has been widely used to image the general morphology of the QDs that are not covered by any capping layer. Figure 3.229 shows the typical AFM images of the GaN QDs grown on AlN at three different temperatures near 700 C [1125]. The dot density is higher than 1011 cm2 and decreases with growth temperature. The density can be effectively reduced through a postgrowth reorganization, called the ripening effect [1132], after GaN growth is finished. During this period, the sample was exposed to N plasma and kept at high temperature for approximately 50 s. Figure 3.230 clearly demonstrates the ripening effect. In this particular case, as compared to the sample without 50 s ripening process, the dot density reduced from 5 · 1011 to 5 · 1010 cm2, whereas the average size (height/ diameter) increased from 2/20 to 5/25 nm. A detailed investigation, with the help of RHEED, shows a similar ripening effect when the samples are either exposed to nitrogen or held at high temperature, typically comparable to the growth temperature, in vacuum [1125]. At low temperature, however, the dot size and density remain unchanged under the nitrogen plasma. The reason was suggested that the Ga diffusion on the surface might be inhibited in the presence of nitrogen [1125]. The detailed structure of GaN QDs has been investigated by high-resolution transmission electron microscopy (HRTEM) [1135,1137]. Figure 3.231a gives an example of the HRTEM image of a GaN dot embedded in AlN, taken along the ½0 1 1 0 axis. A 3D schematic view derived from the HRTEM analysis is shown in Figure 3.231c. The sample is composed of repeated layers of about 2 GaN MLs and thick AlN layers. The analysis of the HRTEM image reveals the following results: (1)
3.5 The Art and Technology of Growth of Nitrides
Figure 3.228 Evolution of lattice parameter and the RHEED images during quantum dot formation. Constructed using data by B. Daudin and coworkers.
The QD has the shape of a truncated pyramid of hexagonal base with a ¼ 30 . The dimensions in this particular case are measured to be 3.3 nm in height and 15.3 nm in diameter. (2) The QD is fully strained and is dislocation free. (3) A wetting layer of
Figure 3.229 200 nm · 200 nm AFM images from GaN quantum dots grown on AlN surface at (a) 725, (b) 705, and (c) 685 C. The growth follows SK mode. The dot density decreases with growth temperature [1125].
j713
j 3 Growth and Growth Methods for Nitride Semiconductors
714
Figure 3.230 (a) AFM image of smooth AlN surface. (b) GaN quantum dots formed by depositing the equivalent of four GaN monolayers on the smooth AlN surface immediately followed by cooling under vacuum. (c) GaN quantum dots formed by depositing the
equivalent of two GaN monolayers on the smooth AlN surface immediately followed by exposure to N plasma for 50 s. The structure reorganization or ripening effect is observed in (c) [1132].
two ML GaN is demonstrated. Only a small amount of Ga and Al atoms (15% in this case) is exchanged across the GaN/AlN interface. It has been established earlier that the strained islands such as InAs grown in successive layers separated by a spacer layer such as GaAs could lead to vertical correlation if the thickness of the spacer layer is appropriate [1139,1140]. The driving force for such vertical self-organization is schematically shown in Figure 3.232 [1139]. Islands in the first layer produce a tensile strain in the spacer above the islands, whereas little stress exists in the spacer away from the islands. Indium adatoms impinging on the surface would be driven by the strained field to accumulate on the top of the islands where they can achieve an energetically lower state due to lower lattice mismatch between the new islands and the spacer. The vertical correlation of self-assembled multilayer QDs was also demonstrated in GaN/AlN systems [1125,1126]. A HRTEM image is shown in Figure 3.233 [1132]. It reveals such a correlation for an AlN spacer of 8 nm. For a thicker AlN layer of 20 nm, no vertical correlation is observed [1132]. In addition to the strain-induced vertical correlation of GaN QD arrays, a correlation between the QD growth and the threading edge dislocations propagating in AlN has been noted [1134]. The conventional TEM and HRTEM images shown in Figure 3.234 demonstrate that the GaN QDs may be more likely to form adjacent to the edge dislocations. In this experiment, the dislocation density in the thick AlN layer is 1.8 · 1011 cm2, comparable with the density of GaN QDs (1.1 · 1011 cm2). The strain field near the edge dislocation favors the nucleation of QDs where the AlN lattice is stretched and the mismatch to GaN is smaller. If the dislocation density is high, as in this case, the vertical correlation of the QDs may be disturbed by the presence of a dislocation line that is slightly inclined. Instead of following the vertical positions of the QDs in the previous layer, the QDs seem rather likely to follow the dislocation line. This effect may be unimportant if the dislocation density is lower than the QD density.
3.5 The Art and Technology of Growth of Nitrides
The growth of GaN QDs on Si(1 1 1) [1127,1141] and SiC(0 0 1) [1142] substrates by MBE were also reported. The purpose of growing GaN QDs on Si substrates is mainly for the integration of light-emitting devices with Si technology. The growth processes
Figure 3.231 (a) HRTEM image of a GaN dot, taken along the ½0 1 1 0 axis. The region of the truncated pyramidal dot is outlined. (b) Fourier filtered image of (a) obtained by using all the (0 0 0 1) frequencies except those belonging to AlN. (c) Schematic view of the dot within the
cross-sectional sample as deduced from fits. Note that only half of the pyramidal dot is within the cross-sectional thickness. (d) Experimental and simulated interplanar distance profiles of the dot [1137].
j715
j 3 Growth and Growth Methods for Nitride Semiconductors
716
Impinging In flux
d Surface stress field GaAs spacer
(2)
(1)
Stressed region I
II Low or no stress region
ls InAs island GaAs (1 0 0)
l
z o
x
Figure 3.232 A schematic representation showing the two major processes for the In adatom migration on the stressed surface using the InAs/GaAs system as an example: (1) directional diffusion under mechanochemical
InAs wetting layer potential gradient contributing toward vertical self-organization and (2) largely symmetric thermal migration in regions from the islands contributing to initiation of new islands not vertically aligned with islands below [1139].
in the former case are essentially the same with those on sapphire substrates. By controlling the size of the GaN QDs in AlN matrix, intense RT PL with different colors from blue to orange as well as white were demonstrated [1127]. When GaN is grown on 3C-SiC(0 0 1) surface, the QDs with cubic rather than hexagonal structure can be obtained. The zinc blende GaN islands were formed on AlN buffer by RF MBE with
Figure 3.233 HRTEM image, taken along the ½0 1 1 0 direction of a superlattice of GaN dots capped by AlN. Because of the low magnification of the printed image, the atomic columns are not seen although they are present. The vertical correlation of the GaN dots is evident. The two-dimensional GaN wetting layer is also clearly visible. Note the dislocation line running through the column of dots at the left-hand side [1125].
3.5 The Art and Technology of Growth of Nitrides
an average height of 1.6 nm and a diameter of 13 nm. The island density is 1.3 · 1011 cm2. In addition to SK mode, the 3D growth of GaN QDs on AlxGa1xN was possible by using a so-called antisurfactant Si [1136]. In this experiment, a smooth AlxGa1xN layer was prepared on 6H-SiC(0 0 0 1) by OMVPE and used as the substrate for MBE regrowth. The GaN QDs were grown by MBE in which NH3 was used as the N source and CH3SiH3 was used as the Si source. The AlxGa1xN surface was exposed to Si flux before the GaN growth and the NH3 flow was stopped for this step. The subsequent GaN growth was carried out with and without introducing Si on two different samples. In the sample without Si flux during the GaN growth, the growth was twodimensional and the streaky RHEED patterns were observed. With Si flow, a change of GaN growth mode from 2D to 3D was observed and the RHEED patterns turned out to be spotty. Formation of GaN QDs was confirmed by AFM. The dot density could be changed by the variation of the Si flux and the growth temperature. The dot density decreased by a factor of 103 and the dot sizes increased from 4/50 to 10/200 nm by raising temperature from 660 to 740 C [1136]. More investigations on the antisurfactant growth scheme by OMVPE will be presented in the next subsection. For applications to light-emitting devices, the InxGa1xN alloy is more frequently used as the active layer. The bandgap and the emitting wavelength are easily modified by alloy composition. The quantum efficiency of light emission from InxGa1xN QWs is usually higher than that in GaN. As compared to GaN QDs, however, fewer investigations have been published on the growth of InxGa1xN QDs. The fluctuation in alloy composition or phase segregation during the growth may complicate the growth and the origin of light emission. Using a conventional MBE with RF plasma source, the growth of InxGa1xN QDs on GaN in SK mode was demonstrated [1143]. In this experiment, the substrate was a 2 mm thick GaN layer grown by OMVPE on sapphire. The growth parameters were monitored by RHEED and the fluctuation in In mole fraction was estimated to be 3%. At the substrate temperature of 580 C, a layer-by-layer growth of In0.35Ga0.65N on GaN was observed during the first 1.7 ML deposition. Beyond 1.7 ML, the growth mode was changed from 2D to 3D and the InxGa1xN islands were formed. The AFM image of the InxGa1xN surface with five ML deposition shows a high island density of 1011 cm3. The average diameter and height of the islands are 27 and 2.9 nm, respectively. This investigation shows that the transition of 2D–3D growth can be realized for In content from 18 to 100%. For an In content below 18%, the growth mode remains 2D. The multiple layers of InxGa1xN QDs can also be formed by the overgrowth of a GaN layer of 4–5 nm, which smoothes the surface. The InxGa1xN QDs grown by SK mode can also be realized by MBE using NH3 as the nitrogen source [1144]. Before the growth of InxGa1xN, a GaN buffer of a few micrometers was grown at 820 C. InxGa1xN was then grown at temperatures from 530 to 570 C and a growth rate of 0.1–0.2 mm h1. The In composition was kept at 0.15, which is larger than the critical value of 2D–3D transition (0.12) determined in the same experiments [1145]. The 2D-to-3D transition was observed after the deposition of four to five ML (11 Å) of In0.15Ga0.85N. The average island size is about 35 nm in diameter and 4 nm in height. The island density was approximately
j717
j 3 Growth and Growth Methods for Nitride Semiconductors
718
3.5 The Art and Technology of Growth of Nitrides
5 · 1010 cm3 and greater than the dislocation density in the GaN buffer layer, which was approximately 5 · 109 cm3. A typical surface morphology of InxGa1xN QDs on GaN imaged by AFM is shown in Figure 3.235. As compared to GaN, the density and the sizes of InxGa1xN QDs are more difficult to control in growth. Although AFM and RHEED investigations have been performed, no detailed lattice structures have been imaged by HRTEM for InxGa1xN QDs. 3.5.16.2 Quantum Dots by OMVPE The III sources used in the OMVPE growth of III nitrides are TMG, TMA, and TMI, carried by nitrogen (N2) or hydrogen (H2) gas. The nitrogen source is ammonia (NH3). The self-organized growth of GaN QDs by OMVPE was first reported by Dmitriev et al. [1146] who grew GaN QDs directly on 6H-SiC substrates. In this case, the lattice mismatch between the GaN and SiC is large enough to lead to the island growth. The main contributions of the growth of GaN QDs on AlxGa1xN (x < 0.2) by OMVPE are by Tanaka and coworkers [1147–1150]. They developed a method called the antisurfactant, which can change the growth mode from 2D to 3D. The selfassembly of GaN QDs is realized in this small lattice-mismatched system by exposing the AlxGa1xN surface to Si during growth. The Si is from tetraethyl-silane [Si (C2H5)4: TESi: 0.041 mmol] (TESi) and carried by H2. Samples were grown on Si-face of 6H-SiC(0 0 0 1) substrates [1147–1150]. Typically, after depositing a thin (1.5 nm) AlN buffer layer, a thick (0.6 mm) AlxGa1xN cladding layer was grown. The Al content x varied from 0.07 to 0.2. GaN was then grown on the top of this AlxGa1xN with a short supply (5 s) of TMG and NH3 during which TESi may or may not have been used. The QDs may be covered by a 60 nm AlxGa1xN layer for optical studies or left without a capping layer for AFM studies. If TESi is not supplied, a step-flow growth of GaN with a smooth surface is observed. Only when the AlxGa1xN surface is exposed to TESi are the GaN QDs effectively grown. A transformation of the surface morphology with and without Si exposure is shown in Figure 3.236. The step-flow growth observed without TESi flux (Figure 3.236c) was explained by a fairly small lattice mismatch between GaN and AlxGa1xN (0.37% for x ¼ 0.15). Under the exposure of small Si dose, large GaN islands were formed. These islands transformed into isolated small dots under a higher Si flux (Figure 3.236a and b). The dot density in this case could be controlled from 107 to 1011cm2 by changing the TESi flow rate, growth temperature, growth time, and alloy composition. The density was found to be very sensitive to the growth temperature, varying by a factor of 3 Figure 3.234 (a) Weak-beam image of a cross section of the GaN QDs in the AlN matrix ðg ¼ f2; 1; 1; 0gÞ. Only the dislocations with a Burgers vector along the c ¼ [0 0 0 1] direction are visible and most of are of screw type. (b) A weakbeam image with g ¼ (0 0 0 1) of the same area. Only the dislocations with Burgers vector of the form 1=3 < 2; 1; 1; 0> (in-plane) are visible. (c) A
0> HRTEM image. The slightly off-axis < 0; 1; 1; first eight QD layers of the samples can be seen. Note that the shape of the QDs in the first three QDs layers above the thick AlN layers are less well defined than the shape of the QDs of other layers. Traces of edge dislocations are outlined by dark arrows for convenience. As clearly seen, the QDs are vertically aligned [1134].
j719
j 3 Growth and Growth Methods for Nitride Semiconductors
720
Figure 3.235 A 1 mm · 1 mm AFM image of nonburied selfassembled InxGa1x N islands grown on GaN by MBE [1144].
Figure 3.236 AFM image of GaN quantum dots assembled on an AlxGa1x N surface using TESi as an antisurfactant, (a) plane view; (b) birds eye view. (c) An AFM image of GaN grown on AlxGa1xN surface without TESi doping, showing a step-flow growth [1147].
103 between 1060 and 1100 C. Figure 3.237 illustrates the AFM images from a typical set of GaN QDs grown on AlxGa1xN surface (x ¼ 0.2) [1150]. The hexagonal-shaped GaN dots have an average height of 6 nm and diameter of 40 nm. The dot thickness/diameter ratio could be changed from 1/6 to 1/2 by varying the growth temperature and Si dose [1147,1148]. The dot density is 3 · 109 cm2, more than one
3.5 The Art and Technology of Growth of Nitrides
Figure 3.237 Cross-sectional HRTEM image of uncapped GaN quantum dots grown on an AlxGa1xN surface. The upper bright part is glue used in the HRTEM sample preparation [1150].
order of magnitude lower than the GaN QDs grown by MBE. For a fixed growth temperature (Ts ¼ 1080 C) the densities of 5 · 109 and 5 · 108 cm2 were obtained with a TESi doping rate of 44 and 176 nmol min1, respectively. By increasing the GaN growth time from 5 to 50 s, the dot size was changed from 6/40 to 100/120 nm. The PL and the stimulated emission [1151] from the similar QDs will be discussed later. Using a similar method, Hirayama et al. [1152] fabricated InxGa1xN (x from 0.22 to 0.52) QDs on an AlxGa1xN (x ¼ 0.12) surface. A two-layer buffer structure was used. First a 300 nm Al0.24Ga0.76N layer and then a 100 nm Al0.12Ga0.88N layer were grown on a SiC substrate, both at 1100 C. Prior to InxGa1xN growth, a small amount of Si antisurfactant was deposited at 1120 C. Then the temperature was cooled to 800 C for QD growth. The dot density was as high as 1011 cm2 and decreased with increasing Si dose. The average dot height and diameter were 5 and 10 nm, respectively, as determined from AFM images. The microscopic mechanism of antisurfactant in the growth of GaN QDs is not well understood. Incorporation of Si in the growth process is assumed to change the surface energy of the AlxGa1xN layer so that the growth mode is modified. An effective surfactant usually raises the surface energy so that a 2D layer-by-layer growth is favored. For GaN growth, the opposite is assumed. Incorporation of Si is assumed to reduce the surface energy of AlxGa1xN and increase the diffusion length of the adatoms. As a result, the adatoms are more likely to coalesce in order to reduce the total energy. InxGa1xN QDs can also be grown on GaN by OMVPE without using antisurfactant [1153]. The equipment is an atmospheric pressure two-flow system with a horizontal quartz reactor. A GaN buffer layer was first grown on (0 0 0 1) sapphire at a temperature of 1075 Cand a V/III ratio of 2000. The InxGa1xN portion was grown at a reduced temperature of 700 C with a growth time 10 s. The growth rate is estimated to be 0.17 nm s1. The thickness of InxGa1xN is about 10 MLs. The average In composition is low (x < 0.1). AFM images show the density of the QDs increasing with growth time. If the growth time is short (6.4 ML), two kinds of QDs, bigger and smaller sizes, were observed. The bigger QDs have a diameter of 15.5 nm and height of 5.4 nm. The smaller QDs have a diameter of 9.3 nm and height of 4.2 nm. When the growth temperature is increased, the dot density decreases monotonically. The important difference in this case from the typical SK mode is the formation of the InxGa1xN QDs even at a long growth time (19 ML). Thus, the formation of the QDs may be mainly attributed to phase segregation rather than strain-induced coalescence.
j721
j 3 Growth and Growth Methods for Nitride Semiconductors
722
Formation of QD-like structures in semiconductor alloys and QWs induced by alloy fluctuation or phase segregation has a great effect on the material properties [1154]. The In-rich clusters in InxGa1xN QWs were suggested to be the origin of high luminescence efficiency in InxGa1xN/GaN LEDs [1155]. Existence of alloy fluctuation/phase separation in InxGa1xN grown on GaN by OMVPE was confirmed by HRTEM [1154]. The spherical QDs were observed in HRTEM images of a 280 nm thick In0.22Ga0.78N layer. A typical dot consists of a core and a surrounding strain zone. The lattice parameters inside the core are slightly larger than that in the surrounding matrix and approach those for InN. The size estimated from an HRTEM image is in the range of 1.5–3 nm. A method very different from self-assembly is selective growth of QDs. InxGa1xN QDs have been grown on Si-patterned GaN/sapphire substrates by OMVPE [1156,1157]. As shown in Figure 3.238, a Si film of 50 nm was first deposited onto the surface of GaN/sapphire substrate by electron beam evaporation at room temperature. Nanoscale circular windows were then opened in the Si mask by focused ion-beam irradiation followed by photoassisted wet chemical etching. GaN/ InxGa1xN multilayers were finally epitaxially grown on the GaN plinths with shapes
Figure 3.238 Upper panel: Schematic diagram of the focused-ion-beam/OMVPE process used for the fabrication of InxGa1xN/GaN nanostructures. Sixty percent of the Si layer was sputtered by focused-ion-beam whereas the rest was removed by photo-assisted-wet etching. Five periods of InxGa1xN/GaN QDs were then selectively grown on GaN plinths exhibiting a
small density of dislocations. Lower left panel: A top SEM view of GaN plinths laterally overgrown on circular windows with diameters of 600 nm (upper row) and 300 nm (lower row). The regrown GaN has a hexagonal pyramid shape with six f1 1 0 1g side facets. Lower right panel: Schematic drawing of InxGa1xN quantum dot structures [1156,1157].
3.5 The Art and Technology of Growth of Nitrides
of hexagonal pyramids. Both InxGa1xN QWs and QDs were formed in the structure but the QDs appeared only on the top of the pyramids. As compared to self-organized growth, the selective growth on pattered substrates could in principle provide a better way of controlling the position, size, and density of the QDs. Ripening after the initial formation of quantum dots has also been employed to shape the dots. In addition, Ga spraying techniques followed by nitridation and conversion to GaN have been used in the authors laboratory to achieve a high density of dots. Spraying at temperatures where Ga migration occurs leads to ball formation. 3.5.16.3 Quantum Dots by Other Techniques In addition to MBE and OMVPE, growth of GaN QDs by other techniques was also reported. Goodwin et al. [1158] have fabricated nanocrystalline GaN by reactive laser ablation of pure Ga metal in a high-purity N2 atmosphere. The samples were collected from the surface of a membrane filter and then thermally annealed at 800 C in a high-purity ammonia atmosphere. TEM dark field images show a log-normal size distribution with a mean diameter of 12 nm and a standard deviation of 8 nm. Selected-area electron diffraction pattern confirms the hexagonal phase. The quantum confinement effect was observed from the blue shift of the size-selective PL and PLE spectra. Nanocrystalline GaN thin films were also fabricated recently on quartz substrates by RF sputtering using GaAs as a target material at a nitrogen pressure of 3.5 · 105 bar [1159]. The average particle size of the nanocrystalline GaN increased from 3 to 16 nm when the substrate temperature was raised from 400 to 550 C. Crystalline GaN particles can be synthesized by simple inorganic reactions at various temperatures. Well et al. [1160,1161] reported a method of nanosized GaN synthesis by pyrolysis of gallium imide {Ga(NH)3/2}n at high temperatures. Dimeric amidogallium [Ga2N(CH3)2]6 was first synthesized by mixing anhydrous GaCl3 with LiN(CH3)2 in hexane. This dimer was then used to prepare polymetric {Ga(NH)3/2}n by reacting [Ga2N(CH3)2]6 with gaseous NH3 at room temperature for 24 h. The GaN QDs were prepared underammoniaflow by slowlyheating{Ga(NH)3/2}n in trioctylamine to 360 C, cooling it to 220 C to add and stir a mixture of trioctylamine and hexadecylamine, and finally cooling it to room temperature. Formation of isolated spherical QDs in colloidal GaN solution was confirmed by TEM images, as shown in Figure 3.239 [1162]. The image reveals that GaN has a zinc blende structure with dots formed having diameters ranging from 2.3 to 4.5 nm. The absorption and PL peaks were observed to shift to a higher energy as compared to bulk GaN. In addition to GaN, an AlGaN nanoparticle/polymer composite was also synthesized using a similar method and the microstructure of zinc blende QDs was confirmed by HRTEM [1163]. Optical properties of dots are discussed in Volume 2, Chapter 5. Because colloidal formation of quantum dots has been very successful in other materials such as conventional III–Vs and particularly II–VIs with aspect ratios approaching a real sphere, a compelling case can at least be made to make mention ofGaNdots producedbythis wetchemicalmethod. Micic etal. [1164] reported on the use of this colloidal chemistry to synthesize GaN quantum dots. The process begins by heating a GaN precursor, polymeric gallium imide, {Ga(NH)3/2}n, which was prepared by the reaction of dimeric amidogallium with ammonia at room temperature, in
j723
j 3 Growth and Growth Methods for Nitride Semiconductors
724
Figure 3.239 TEM image of GaN QDs taken in bright field. The particles are well dispersed and not agglomerated. Top panel shows low magnification of QDs and some linear alignment. Bottom two right panels show high magnification and lattice fringes of QD oriented with the h1 1 1i axis in the plane of the micrograph. Bottom left panel shows electron diffraction pattern of GaN QDs indicating zinc blende structure [1162].
trioctylamine at 360 C for 1 day to produce GaN nanocrystals. The resultant GaN particles were separated, purified and partially dispersed in a nonpolar solvent that culminated in transparent colloidal solutions that consisted of individual GaN particles. The GaN nanocrystals had a spherical shape and mean diameter of about 30 2 Å, unlike very unfavorable aspect ratios that result in epitaxial methods. The spectroscopic behavior of colloidal transparent dispersion has been investigated and showed slightly
3.5 The Art and Technology of Growth of Nitrides
blue shifted near band edge emission, indicating quantum confinement. The PL spectrum recorded at 10 K (excited at 310 nm) showed band edge emission with several emission peaks in the range between 3.2 and 3.8 eV, whereas the PL excitation spectrum showed two excited-state transitions at higher energies. With the advent on nano the scope of the game changed and nano implicitly meant all kinds of nanowires and tubes, and so on, as if dots were not nano in dimension. Below, we will follow the great masses, mainly the chemists, and discuss nanowires and the methods used to produce them under the auspices of nanostructures that do not include quantum dots. 3.5.16.4 Preparation and Properties of Nanostructures The usual justification used for exploring nanostructures is that they exhibit electronic and optical properties that are novel, primarily owing to confinement owing to dimensions that are comparable to intrinsic process lengths in semiconductors, although many nanostructures reported in nitrides are not sufficiently small to exhibit confinement in optical experiments. In some cases, the goal is to produce less defective material by limiting the contact area of the growing nanostructure with the mismatched substrate and/or cause the nanostructure to form freely. Just to develop a sense of relative dimensions, the length scales of the well-known processes in semiconductors are illustrated in Figure 3.240. In optical processes, the exciton Bohr radius is one of the most if not the most critical dimension. Other processes such as phonon and electron mean free paths, the Debye length (depends on doping level as well as the material) and the exciton diffusion length for certain polymers are in the range of 1–500 nm. It is hoped that Critical magnetic single domain size Exciton diffusion length in polymers Phonon mean free length
Exciton Bohr radius Debye length Fermi wavelength in metals
10–1
100
101
102
103
Characteristic length (nm) Figure 3.240 Characteristic lengths of a few of the well-known processes occurring in semiconductors at 300 K. The length scales of over a few tens of nanometers are accessible by advanced lithography. Below that length scale chemical processes are used, however, the production is not yet conducive to semiconductor fabrication processes that revolutionized our world through high performance electronics with reduced cost per function.
j725
j 3 Growth and Growth Methods for Nitride Semiconductors
726
synthesized nanowires with 5–100 nm in diameter might allow access to the realm of mesoscopic phenomena. These structures are often beyond the realm of lithography, albeit the limit for lithography has steadily reduced to the point where dimensions near 100 nm are easily producedandarationalforsynthesizingthemmustbedevelopedfortheirself-formation such as the quantum dots discussed in Section 3.5.16. The blue shift of the bandgap expected from an ideal confined system in the simplest sense is given by Equation 3.71, which predictsthefigurefor a cubic quantum box model to beh2/(8md2), with d being the dimension of the cubes edge. In practice, the blue shift is of the form 1/dn (1 n 2). More detailed calculations for the blue shift in a 2D plate, a sphere with a diameter of d, with d varied, are shown in Figure 9.173. The cubic box falls in between the 2D plate and the sphere. An InP-based study [1165] indicated the blue shift to be DEg 1/d1.35 for dots and DEg 1/d1.45 for wires. The calculations mentioned assume no band bending due to surface states and the accurate extraction of the confined system size from the blue shift would require that the surface band bending be taken into account. The bandgap tunability with a controlled size variation would have applications in emitters, detectors, andsolarcells.Interestingstructures such asperiodic quantumwells along thenanorods or wiresashasbeen implemented inZnO wellsseparated byZn0.8Mg0.2Obarriers[1166] and coresheathnanowires[1167] implemented inGaN/Al0.75Ga0.25N in thesame veinas quantum dots coated with a larger bandgap barriers for enhanced quantum efficiency by removing the high surface recombination velocity out of the picture. 3.5.16.4.1 Approaches for Synthesis The reader would be well served with a discussion of the merits and limitations of often used processes for producing nanowires. A very good review of the field in the context of group III nitrides can be found in Ref. [1168]. In addition to desired ability to produce high-quality material, pivotal issues to look for are the understanding of the one-dimensional growth process based on kinetic and thermodynamic rationale and predictability of the process as well as being applicable to a wide variety of materials systems [1169]. Nanowire growth, in general, comes about by instilling and taking advantage of anisotropic growth that often times can be promoted by templates having at least one-dimensional morphologies. Introduction of liquid/solid interface in the process to reduce the symmetry of the seed and use of an appropriate capping reagent to control the growth rates of various facets of the seed are among the methods garnered. Many of the early methods did not have the requisite attributes such as reproducibility, uniformity, scalability, cost effectiveness, and the knowledge of the basis for growth. It is of course instructive to know a priori whether core/sheath (coaxial) or longitudinal heterostructures are desired as end product. Let us now give a succinct review of methods used to produce nanostructures, among which are growth from the vapor phase and vapor–liquid–solid (VLS), inclusive of self-catalytic and vapor phase varieties, vapor–solid (VS) growth, and confined chemical reaction such as that utilizing carbon nanotubes for preparation. 3.5.16.4.2 Vapor Phase Growth This is by far the most successful and most commonly employed method for producing low-dimensional structures, the term that was popular before the nano craze, such as whiskers (was considered a failure by some of the
3.5 The Art and Technology of Growth of Nitrides
pioneers who stumbled on them), nanorods, and nanowires. Numerous methods have been developed for precursors to make this process possible, including laser ablation, chemical vapor deposition, chemical vapor transport methods, molecular beam epitaxy, and sputtering. As in any growth mechanism, measures must be taken to avoid or suppress parasitic secondary nucleation events, leaving only the nanowire synthesis as the dominant process. Vapor–Liquid–Solid Process As the name suggests, this is a process wherein constituents in the vapor phase diffuse through a metal catalyst and deposit underneath as a solid. During theentireprocess,thecatalystrides asliquidontop ofthegrowthfront.The catalyst is involved during growth but does not incorporate inthe final product. This is by far the most widely used method among the vapor-based processes in generating large quantities of single crystalline nanowires. Typically, the process takes place in a two- or three-zone furnace, allowing independent temperature control of the source and substrate regions. The source isnaturally kept at a higher temperature than the substrate thatis downstream, asshown inFigure 3.241.The group IIIelements, mainly Ga or In in the case of GaN or InGaN, in the form of foil or powder is placed in an alumina boat that providesvaporsourceuponheating.Thesubstrateisplaceddownstreaminacolderzone of the furnace after precoating it with a catalysts. The substrate could be preloaded with catalyticmetalstrip/powders orprecoatedwithathinlayerofcatalyticmetalfilmbyvapor transport of by evaporation in a separate vacuum system. Introducing catalysts via vapor transport on a patterned substrate by either shadow masking or lithography offers the opportunity for selective area growth. A variant of this process is one wherein the catalytic growth may also be attained with source materials in a solution form. This process is thus termed as solution–liquid–solid (SLS) growth. The VLS process dates back to 1960s, albeit with dimen-
Three-zone furnace
Quartz tube
Flow controller
NH3
Powder
O-ring joint
Substrate
Bubbler Figure 3.241 Schematic representation of a VLS reactor for the growth of GaN-based nanowires and nanorods. The setup can be used in either the single-boat configuration in which the sample is collected on a substrate placed on top of the crucible containing reactants, or the two-boat configuration in which the sample is collected on a separate substrate which is placed downstream of the reactants (shown).
j727
j 3 Growth and Growth Methods for Nitride Semiconductors
728
sions in micron and submicron regimes, when it was applied to produce whiskers [1170,1171], which, following the development of the thermodynamic and kinetic basis [1172], paved the way for generating nanowires and nanorods from a rich variety of inorganic materials [1173–1175]. Even with improved techniques that culminated in the reduction of dimensions and production of a plethora of shapes, the control of the orientation, size, nucleation sites (unless aided by lithography) with sufficient precision remains challenging. In addition, the catalyst used for nucleating and feeding the growth front remains with the nanorods (nanowires) and must somehow be removed. To circumvent this problem altogether, catalyst-free methods have been developed, mainly MBE and OMVPE like approaches, which will be discussed shortly. Let us take the example of Ge nanorod growth, as it provides a very good example of the process, because of the manner in which the rods are produced, to shed some light on the evolution of the growth processes in VLS epitaxy. Using an environmental transmission electron microscope in the realm of Ge nanorod growth with Au as the liquid catalyst, Wu and Yang [1176] monitored the VLS growth in real time. They found that the process commences with the dissolution of gaseous reactants into nanosized liquid droplets of a catalyst metal, which pave the way to nucleation and growth of single-crystalline rods and then to wires. The liquid droplets, whose sizes remain essentially unchanged during the entire process of wire growth, are pivotal in initiating and maintaining growth. The evolution of the VLS process is illustrated in Figure 3.242, with the growth of a Ge nanowire observed by in situ TEM. Based on the Ge–Au binary phase diagram, Ge and Au form liquid alloys when the temperature is raised above the eutectic point (361 C). Once the liquid droplet is supersaturated with Ge, nanowire growth will start to occur at the solid–liquid interface. The
Figure 3.242 In situ TEM images recorded in an environmental transmission electron microscope during the VLS process of Ge nanowire growth using Au dots as catalyst. (a) Au nanoclusters in solid state form at 500 C; (b) alloying is initiated at 800 C with Au still existing
in mainly in a solid state; (c) liquid Au/Ge alloy formation; (d) commencing of the nucleation of Ge nanocrystal on the alloy surface; (e) Ge nanocrystal elongates with further Ge condensation, and (f) eventual forming of a wire. Courtesy of P. Yang and Ref. [1169].
3.5 The Art and Technology of Growth of Nitrides
establishment of the symmetry-breaking solid–liquid interface is the key step for the pseudo-one-dimensional nanocrystal growth in this process, whereas the stoichiometry and lattice symmetry of the semiconductor material systems are less relevant. One of the main challenges for catalyst assisted growth has to do with the catalyst itself that is to produce a uniform dispersion and yet maintain the small and uniform size of the catalyst. In general, the correlation in the size of the catalytic nanoparticles and resulting diameters of the nanowires or nanorods is reasonably good. In other words, in the VLS process the diameter of each nanowire is to a large extent determined by the size of the catalyst particle. It then follows that smaller catalyst islands yield thinner nanowires and dimensional control is strictly based on the control on the initial size of the catalyst used. Having said that it should be noted that the average diameter of the nanowires (nanorods) is typically larger than that of the catalyst itself. Poorly dispersed catalytic nanoparticles and thin film coating often lead to significant agglomeration of catalysts during heating. In addition to the agglomeration problem of the catalyst, the nonuniform distribution of the constituent group III elements and/or N in the catalytic reaction can also cause a broad diameter distribution of the nanowires. Because the process temperature for nanowires is relatively high, considerable self- and surface-diffusion could occur. To circumvent the diffusion issues, diffusion barrier material and the catalyst are mixed, a process that is effective in preventing agglomeration of the catalyst. In this vein, reaction of Ga and SiO2 mixtures with NH3 in the presence of the Fe2O3 catalyst supported by Al2O3 have been used to produce GaN nanowires [1177]. The addition of SiO2 is thought to reduce the melting temperature of the catalyst by forming FeSi2 as well as enhancing the production of high-pressure Ga2O gas through formation of a eutectic Ga–SiO2 or the following reaction: 4Gaðs; lÞ þ SiO2 ðsÞ ! 2Ga2 OðgÞ þ SiðsÞ:
ð3:72Þ
In this case, the catalysts continuously dissolve the gaseous Ga2O and NH3, leading to the formation of GaN as discussed in Section 3.5.16.4.2. In this process, although Al2O3 does not participate directly in the catalytic reaction, together with Fe2O3, it seems to prevent the agglomeration of the Fe-containing catalytic droplets, paving the way for GaN nanowire formation with a much smaller diameter (10–50 nm) than those produced with Fe2O3 only (80–200 nm) [1168]. In addition to the agglomeration problem of the catalyst mentioned above, obviously the nonuniform distribution of the constituent group III elements and/ or N in the catalytic reaction can also cause a dispersion in the diameter of the nanowires. Group III elements are typically placed in a boat at the hot zone in the furnace close to the gas inlet and the substrate on which the rods are to form is placed in the colder section downstream. The distance between the source and substrate typically has an impact on the dispersion in the diameter, with diameter decreasing with increasing distance. This leads to larger diameter forms at the leading edge of the substrate than the trailing edge. In this respect, a long furnace with a large source to substrate distance would reduce the diameter variation across the substrate.
j729
j 3 Growth and Growth Methods for Nitride Semiconductors
730
Obviously, the deposition (growth) time is also a factor in determining the nanowire diameter and length. In addition to solid sources such as powders used for group III elements, gas sources as in the standard vapor phase deposition techniques have also been employed. Although increasing the complexity of the experimental setup, this feature provides a better control over the group III element supply and also paves the way for heterostructure growth. An example is the combined use of NH3 gas for N, and ferrocene (C10H10Fe) and gallium dimethylamide (Ga2[N(CH3)2]6) for catalyst and group III element, respectively [1178]. By this method, GaN nanorods with diameters of 15–70 nm and lengths of 3–30 mm have been produced. However, due to the presence of carbon in the organometallic sources used, carbon nanotubes were also formed at process temperatures of 900 and 1000 C. Other organometallic sources [1179] for Ga such as (CH3COCHCOCH3)3Ga have also been explored for GaN nanowire growth at temperatures as low as 620 C. The correlation between the catalyst size and resulting rod size has been demonstrated in Si and GaP nanowires in that any specific size could be obtained by controlling the diameter of monodispersed gold colloids serving as the catalyst [1180,1181]. Analogous to the thickness in films grown by MBE and OMVPE, the length of the wire is proportional to the growth time. Because VLS lives or dies with the catalyst, the selection of catalyst is of paramount importance. In the selection criterion of the catalyst, the main requirement, in addition to availability in sizes desired, is that the catalyst must be capable of forming an alloy with the target material, ideally an eutectic compound, which brings about the pseudobinary phase diagram between the metal catalyst and the solid material of interest. It should be mentioned that VLS cannot be applied to the production of metal nanocrystals because of alloy formation between the metal and catalyst. In terms of the semiconductor nanocrystal growth by VLS, one of the major drawbacks, at least potentially, is the contamination of semiconductor nanocrystal by the metal catalyst. There has been considerable activity in GaN and to a lesser extent AlN based nanorods and nanowires in this respect. InN, however, did not received as much attention due to the high vapor pressure of N over In and thus low thermal stability that requires the growth temperature to be low (<600 C) and inordinate amounts of N to be used. Consequently, there are only a few InN nanowire reports wherein either a catalyst [1182] or single molecular precursors exhibiting a low decomposition temperature [1183] has been used for forming the nanowires at temperatures around 500 C, which is below its dissociation temperature and is safe. Self-Catalytic VLS In the production of binary and even more complex ternary nanowires produced with VLS, it is possible for one of these elements to serve as the catalyst, thus the nomenclature self-catalytic. Stach et al. [1184] used an environmental TEM to observe self-catalytic growth of GaN nanowires by heating a GaN thinfilm in a vacuum of 107 Torr. It is known that GaN decomposes substantially at temperatures of about 850 C in high vacuum via a process [224] described in Equation 3.15. Also, the congruent sublimation of GaN to the diatomic or polymeric
3.5 The Art and Technology of Growth of Nitrides
vapor species has been predicted [225] and observed [223], as described in Equation 3.17. Stach et al. [1184] observed that initially decomposition of the GaN film leads to the formation of isolated liquid Ga nanoparticles. The vapor species, composed of the atomic nitrogen and diatomic or polymeric GaN, then proceed to redissolve into the Ga droplets and initiate the VLS nanowire growth after supersaturating the metal and establishing a liquid–Ga/solid–GaN interface. Each step in the VLS process has been observed by Stach et al. [1184] in their environmental microscope, the evolution of which is shown in Figure 3.243. Essentially, alloying of the Ga droplet with the nitrogen-rich vapor species, the nucleation of the nanowire liquid–metal interface, and the subsequent axial nanowire growth could all be seen. The only advantage of the self-catalytic process is that it has the potential of avoiding contamination from foreign metal atoms typically used in the typical VLS process. The self-catalytic behavior has been reported when the direct reaction of Ga with NH3 or direct evaporation of GaN was used to produce GaN nanowires [1175,1185]. The drawback of this process is that the control of the nanowire lengths and diameters has yet to be demonstrated.
Figure 3.243 A series of video frames grabbed from the observations of GaN decomposition at about 1050 C in an environmental microscope chronicling the real-time GaN nanowire formation. The number at the bottom left corner of each frame indicates the time in the format of second : millisecond. Courtesy of P. Yang and Ref. [1184].
j731
j 3 Growth and Growth Methods for Nitride Semiconductors
732
Vapor–Solid Method In methods requiring a catalyst for growth, preparation, and removal of the templates or catalysts cannot be done by just a few simple steps in most cases and thus are not deemed cost effective. Therefore, growth techniques that do not require any catalyst and template are highly desirable. In this vein, vapor–solid growth method has been employed and proved to be an effective catalyst-free and template-free method for synthesizing whiskers. As can be garnered, both VLS and VS techniques involve vapor phase deposition. In the VS method, the chemical species are transported toward the surface of a substrate and are adsorbed on the substrate surface. This is followed by nucleation and whisker growth that is accompanied by the elimination of gaseous by-products. In contrast, in the VLS method a liquid-phase catalyst is involved for adsorbing gas reactants and forming eutectics. Consequently, the diameter of the whiskers is controlled by the size of the catalyst, whereas the diameter is determined by the surface migration of gas-phase species for the VS growth. What is common in both methods is that there exists a large difference in growth rate along different crystal orientations that is responsible for whisker growth. The highly anisotropic growth (growth rate along the c-direction being relatively large) may be inherent in the wurtzite III-N and II-VI systems, which offers an opportunity for forming one-dimensional nanostructures in a wide range of process conditions. Moreover, cubic III-N (which is not thermodynamically favorable over the wurtzite phase) nanowires have also been produced using this type of technique. Unlike the standard VLS process, because in the VS method the vapor species are introduced into the vessel by either gas sources (CVD in particular OMVPE for nitrides) or by evaporation (MBE) in which the growth takes place, it would be possible to produce nanowire heterostructures with compositional gradients and dopants. This will also bring the technology used for nanowire production in line with that of mainstream semiconductor deposition methods such as OMVPE and MBE. Interpreting vapor species very liberally to include the laser ablation processes as well, GaN-based structures have been produced by laser ablation, chemical vapor transport, MBE, and OMVPE, the latter two are discussed copiously in previous sections of this chapter. Of these methods, OMVPE and MBE appear to have the flexibility needed for producing GaN-based nanowires with controlled dopant and composition. This is partly because of the similarity of the precursor chemistries for the constituent and dopant atoms in the case of OMVPE and control over the effusion of N and Ga species as well the substrate temperature.There have been numerous reports on one-dimensional nanostructure formation from vapor phase precursors in the absence of a metal catalyst by evaporation [1186]. Because no catalyst is involved, it would be appropriate to call this method as the vapor–solid method. Thermodynamic and kinetic considerations would suggest that the formation of nanowires could possibly be through a variety of processes, among which are the anisotropic growth mechanism, Franks screw dislocation mechanism [1187], and self-catalysis. The anisotropic growth mechanism might be due to the preferential reactivity and binding of gas phase reactants along certain specific crystal facets and also the predilection for a system to minimize surface energies. In the dislocation and defect-induced growth process, specific defects such as screw dislocations are known
3.5 The Art and Technology of Growth of Nitrides
to enhance reactivity or nucleation and therefore deposition of gas phase reactants would take place first at these defects [1169]. Other variants such as oxide-assisted growth on nanowires have been reported as well [1188]. Single volatile molecular precursors for nitride semiconductor nanostructures have been developed on the basis of volatile group III amide, azide, and hydrazide compounds and have been investigated in the realm of organometallic chemical vapor deposition (OMCVD) for nitride nanostructures [374,1189,1190]. For example, Parala et al. reported InN whiskers prepared by CVD under very specific conditions using the single molecule precursor (Azido[bis(3-dimethylamino)-propyl]indium) (azin), [N3In[(CH2)3N(CH3)2]2] [1183]. The azin precursor, which was synthesized first by slightly modifying the synthetic procedure [374], was placed in a reservoir and nitrogen was used as a carrier gas with ammonia as an additional reactive gas for growing InN at a substrate temperature of 500 C. InN whiskers resulted when bare cplane sapphire [Al2O3(0 0 0 1)] was used. When the deposition was conducted on nitridated sapphire substrates, for example, forming an AlN buffer layer with a reported thickness of the order of 20–50 nm (on the high side for nitridation only), dense and preferentially oriented thin films were formed instead. Again with azin single molecular precursors, InN nanowires have been attained after about 15 min growth, which is much shorter than that normally reported in CVD-based catalytic growth and other vapor–solid methods. It should also be noted that the InN fibers produced by the pyrolysis of azin contain In droplets at their tips. Although conventional metal catalysts such as Fe, Ni, and Au were not introduced deliberately, it is plausible that indium was produced at the beginning of azin pyrolysis, which later on served as catalyst for further fiber growth. This self-catalyst behavior is very similar to that observed by Buhro et al. [1191] in the formation of InN nanowires by SLS method in the presence of H2NN(CH3)2. As could be garnered, many of these proposed vapor solid growth mechanisms are not fully understood in terms of the underlying thermodynamic and kinetic fundamentals that might in effect be responsible for one-dimensional growth. In spite of this, many materials with interesting morphologies have been produced with the variants of VLS method. Among the examples are nanoribbons (of ZnO, SnO2, In2O3, and CdO) having rectangular cross sections and with typical thicknesses ranging from 30 to 300 nm. These were produced by simply evaporating commercial metal oxide powders at elevated temperatures [1192]. In terms of the size, the nanoribbons typically had thicknesses ranging from 30 to 300 nm, width-to-thickness ratios of 5–10, and lengths up to several millimeters [1186]. The method has also been utilized to form ZnO tetrapods and comblike morphologies [1193,1194]. MBE method normally offers uniform growth over the entire substrate but only under conditions that promote uniform growth. Taking advantage of the very high sticking coefficient of N and high evaporation rate of Ga at very high substrate temperatures, nanorodlike growth mode can be induced for GaN in an MBE environment [1195]. In this approach, the typical sapphire cleaning and nitridation processes (other substrates such as Si have also been used) are adhered to. However, the GaN growth conditions used deviate from the standard growth conditions. Using an RF source for reactive nitrogen, substrate temperatures at or slightly above 800 C,
j733
j 3 Growth and Growth Methods for Nitride Semiconductors
734
which is some 200 C higher than the nominal temperatures for growth under Garich conditions resulting in uniform layers and very high nitrogen fluxes vertically aligned [0 0 0 1] oriented GaN nanorods, can be produced. The Ga temperature can be optimized under a fix N flux to produce nanowires with diameters as low as 40 nm. The demonstrated advantage of these rods is that high quality can be obtained owing to a minimum contact area with the lattice mismatched sapphire or Si substrates and reduced strain. The advantage of MBE over some of the other methods discussed so far is that the vertical compositional variations can be introduced to obtain quantum confinements and also observe quantum-confined Stark shift owing to highly polar nature of nitrides along the [0 0 0 1] direction. We should reiterate that nanowires are allowed to relax more so than in the case of two-dimensional growth and thus the piezo-induced polarization is minimized. The compositional manipulation can be used to produce wavelengths from the InGaN well material covering nearly the entire visible spectrum. The details of the growth issues as well as results are discussed in Section 3.5.16.5. Nanowire Growth in Solution Growth from solution would have the obvious advantage of a low-temperature process that is involved and thus the lower cost. However, control and reproducibility often required for electronics and integration possible with other processes, which might be necessary, might prove difficult if not impossible with this method. Nevertheless, for completeness, a few words are shared here to describe the method along with a few applications. The process has been used to produce one-dimensional nanostructures with high yields in the gram scale by selective capping mechanisms. Molecular capping agents are believed to play a significant role in the kinetic control of the nanocrystal growth by preferentially adsorbing to specific crystal faces on which the growth is normally inhibited [1169]. The aforementioned selective capping has been demonstrated by Sun et al. [1196] in silver nanowires using poly(vinyl pyrolidone) (PVP) as a capping agent. The presence of PVP paves the way for most of the silver particles to participate in the growth of nanowires with uniform diameter. It is plausible that PVP selectively binds to the {1 0 0} facets of silver, freeing the {1 1 1} facets to be available for growth. As a demonstration, Sun et al. [1196] functionalized their nanowires following growth with a dithiol compound followed by addition of gold nanoparticles to the solution. The gold nanoparticles were found to be bonded only to the end {1 1 1} caps, thereby showing only the dithiol adhesion on the end caps and not the {1 0 0} faces owing to the preferential bonding of the PVP to these faces. Nanowires of silver with diameters in the range of 30–60 nm and lengths up to 50 mm have been produced. The aforementioned process is amenable to metal-based system. The growth of semiconductor nanowires, such as challenging InN nanostructures, has been produced with a similar method, SLS [1197]. By this method, InN can be grown at 203 C owing to nanometer-sized metal droplets serving as catalyst for fiber formation. The precursors, iPr2InN3 (1a) and tBu2InN3 (1b), have been prepared in one-pot procedures from the corresponding trialkylindanes via diakylmethoxyindane intermediaries. Reactions of the precursors 1a or 1b with the mild reductant 1,1dimethylhydrazine, H2NN(CH3)2, have been stated to produce crystalline InN.
3.5 The Art and Technology of Growth of Nitrides
Thermolysis of 1a in refluxing diisopropylbenzene (203 C) with H2NN(CH3)2 led to crystallites with a coherent length of 18 nm, as determined by XRD data. On the contrary, the same process but without H2NN(CH3)2 led to amorphous structure. Meanwhile, thermolysis of 1b both with and without H2NN(CH3)2 produced InN crystallites, albeit with the coherence lengths of 12 and 7 nm with and without H2NN (CH3)2, respectively. To gain an appreciation for the breath of the method, ZnO microrods have been attained by hydrolysis of zinc salts in the presence of amines [1198]. In a similar vein, Greene et al. [1199] using hexamethylenetetramine as a structural director, produced dense arrays of ZnO nanowires in aqueous solution with diameters of 30–100 nm and lengths of 2–10 mm. Most significantly, these oriented nanowires can be prepared on any substrate. One of the advantages of this growth process is that a majority of the nanowires in the array are in direct contact with the substrate and a continuous pathway for carrier transport is available. The major disadvantage is that an empirical trial-and-error approach is typically used to choose most capping agents. Confined Chemical Reactions: Template-Based Methods For the synthesis of GaN and AlN nanorods, materials that provide confined space in nanometer scale, such as carbon nanotubes [1182], GaAs nanocolumns [1200], and alumina membranes produced by anodic etching [1201] have been used [1168]. Let us now discuss these approaches briefly, a detailed review for this method and other relevant material for nanostructure preparation can be found in Ref. [1168]. Using carbon nanotubes (CNT), Han et al. [1202] reacted Ga2O with NH3 to prepare GaN nanowires with diameters between 4 and 50 nm and lengths of up to 25 mm, which are similar to those of the original carbon nanotubes. The starting material for Ga2O was a 4 : 1 molar mixture of Ga and Ga2O3 powder over which the Ga2O vapor pressure is generated through the following reaction upon heating [1203]:
Ga2 O3 ðsÞ þ 4GaðlÞ ! 3Ga2 OðgÞ:
ð3:73Þ
It should be appreciated that the pressure of the Ga2O vapor generated through the above reaction is about 1 Torr at 900 C and 7.2 Torr at 1000 C [1204]. In Hans [1202] experiment, the Ga–Ga2O3 powder mixture was first placed in an alumina crucible and its top covered with a porous alumina plate. In such an arrangement, the Ga2O vapor generated from the Ga–Ga2O3 powder mixture flows toward the CNTs through the porous alumina plate and react with the CNTs and NH3 gas. The main reaction in the formation of GaN nanorods is: 2Ga2 OðgÞ þ CðsÞ þ 4NH3 ðgÞ ! 4GaNðsÞ þ H2 OðgÞ þ COðgÞ þ 5H2 ðgÞ: ð3:74Þ It should be noted that in the absence of the Ga–Ga2O3 powder mixture no change in the CNTs was observed under the identical process condition. Furthermore, without the presence of CNTs GaN powders were produced by the reaction between gaseous Ga2O and ammonia [1205]: Ga2 OðgÞ þ 2NH3 ðgÞ ! 2GaNðsÞ þ H2 OðgÞ þ 2H2 ðgÞ:
ð3:75Þ
j735
j 3 Growth and Growth Methods for Nitride Semiconductors
736
These experimental results provide further support for the CNT-confined reaction for the formation of GaN nanorods through reaction defined by Equation 3.74. Similar to the GaN case, AlN nanowires have also been prepared at 950–1200 C using confined environment provided by CNTs [1206,1207]. Unfortunately though, the Al2O powder is somewhat difficult to prepare at low reaction temperatures (e.g., below 1000 C) because it is difficult to reduce Al2O3 by Al or carbon due to the high positive change of Gibbs free energy. Meanwhile, Al and NH3 may react with each other at 900 C through the following reaction [1168]: 2AlðsÞ þ 2NH3 ðgÞ ! 2AlNðsÞ þ 3H2 ðgÞ:
ð3:76Þ
This reaction is highly exothermic and may result in substantial local heating, which, in turn, would promote the following reactions that reduce the Al2O3 together with C from CNTs or with Al through the reactions: Al2 O3 ðsÞ þ 2CðsÞ ! Al2 OðgÞ þ 2COðgÞ;
ð3:77Þ
Al2 O3 ðsÞ þ 4AlðsÞ ! 3Al2 OðgÞ:
ð3:78Þ
Finally, the AlN nanowires can be produced according to the following reaction: 2Al2 OðgÞ þ CðsÞ þ 4NH3 ðgÞ ! 4AlNðsÞ þ COðgÞ þ H2 OðgÞ þ 5H2 ðgÞ: ð3:79Þ Although the end result is the production of AlN, this process is quite different from that defined in Equation 3.74 for the production of GaN. In addition to confinement by CNTs discussed above, anodized Al membrane confined reactions have also be exploited to produce nanowires [1208–1211]. As a side note, these alumina membranes have been used for selective ion transport, molecular filtration, and drug separation. The anodic alumina membrane is a self-ordered and formed (in that no lithography is required for their formation) nanochannel material formed by anodization of Al in an appropriate acid solution [1212,1213]. The anodic alumina membrane has hexagonal ordered pores with diameters ranging from 10 to 200 nm with typical channel depth of tens of microns and channel density in the range 1010–1012 cm2. The very high aspect ratio is indeed difficult to achieve with conventional lithographic/etching techniques. The control of the pore size and depth and the uniformity of the depth are among the critical parameters that are often difficult to control. The capillary effect of the anodic nanochannels has been suggested to facilitate the formation of nanowires. Owing to these desirable geometric features and chemical stability, the nanochannel membrane-based method has been employed for fabricating various kinds of ordered nanowire arrays [1214–1217]. Being relatively established and not requiring expensive starting equipment and budget, the anodic alumina membrane became a popular choice of template. Ordered crystalline GaN nanowires embedded in the nanochannels of anodic alumina membrane were achieved by a direct reaction of Ga and NH3 [1218], through the following reaction path: 2GaðgÞ þ 2NH3 ! 2GaNðsÞ þ 3H2 ðgÞ:
ð3:80Þ
3.5 The Art and Technology of Growth of Nitrides
To make this reaction work, the metallic Ga inside the crucible is placed in the hightemperature zone and under ammonia flow the temperature of the crucible is increased to or above 900 C. Below this temperature, hardly any GaN nanowires are formed. To increase the yield of GaN nanowires, the Ga2O3 powder can also be added to the Ga metal so that gaseous Ga2O is generated following reaction described by Equation 3.73. In this case, the gaseous Ga2O and NH3 can react directly, leading to the formation of GaN via the reaction described by Equation 3.75. If the confined reaction is preformed with the addition of a catalyst, then the gaseous Ga2O and NH3 can continuously be dissolved into the catalyst, leading to the formation of GaN via vapor–liquid–solid mechanism discussed in Section 3.5.16.4.3. In practice though, the nanochannel-confined formation of the GaN nanowires reported in the literature has not solely been achieved by direct reactions but rather through the assistance of some catalysts. 3.5.16.5 Nanowires and Longitudinal Heterostructures As mentioned, CNT-confined reaction of Ga2O vapor was reacted with NH3 to form GaN nanorods that mimic the dimensions of CNTs [1202]. Similarly, albeit with more complex reaction paths, AlN nanorods have also been produced using confined reaction provided by CNTs [1206,1207]. As for the use of aluminum membrane, Cheng et al. [1201,1219] have employed metallic indium by angle evaporation on one side face of the anodic alumina membrane to form indium nanoparticles prior to the reaction processes. With indium nanoparticles present, it is believed that the vapor–liquid–solid growth mechanism causes the formation of the GaN nanowires. The as-synthesized products have also been shown to exhibit phase pure hexagonal wurtzite GaN structure, indicating that the indium nanoparticles only act as a catalyst to form ordered GaN nanowires. However, the indium nanoparticles on the side face of the alumina membrane as a catalyst cannot assure growth for the GaN nanowire to take place exclusively inside the nanochannels of the membrane. Overgrowth of the nanowires outside of the uniform nanochannels of the membrane and extension from thereon have been observed. In contrast, electrochemically deposited In has been shown to be more advantageous [1220]. The anodic alumina membranes as well as the carbon nanotubes do indeed act as a template and control the shape and size of the nanowires. Only powder products are produced by the above-mentioned reactions without the templates. Despite the slight dispersion in the diameter, a direct proportionality between the diameter of the nitride nanowires and the outer diameter of the carbon nanotubes (or the channel size of the membrane) is typically observed for samples fabricated in a rather wide range of process temperatures and reaction times ranging from 30 to 200 min [1206]. However, the impact and effect of the template are eventually lost if the reaction temperature is too high. For the case of AlN, the carbon nanotubes did not work as well at temperatures above 1500 C, at which point Al and NH3 compose directly into AlN fibers with much larger diameters [1221–1223]. While the template-based techniques are effective in producing nanowires with controllable size, the inability to isolate the product and then maintain the orientation alignment after removing the membrane does seriously hamper its wide application.
j737
j 3 Growth and Growth Methods for Nitride Semiconductors
738
Changing our attention to VLS, using catalytic reaction of gallium with ammonia, Chen et al. [1224] synthesized GaN nanowires using a two-boat system, the upstream one containing the source material while the downstream one containing the substrate that would be decorated with the catalyst. At the process temperature, the group III elements become molten droplets and get transported by vapor downstream to the substrate containing the catalyst. It was found that the diameter dispersion of the nanowires collected downstream strongly depended on the distance between the group III source material and the catalyst [1224]. The diameter of the nanowires so produced was observed to decrease with increasing source to substrate distance, presumably due to the depletion of the reactants. The diameters of the nanowires were found to be primarily in the range of 50–150 nm when the catalyst/substrate was placed at a distance of about 1 cm from the boat containing the group III elements upstream. In contrast, the diameters of the nanowires were significantly reduced to 20–50 nm when the distance between the two was increased to 10 cm. It should be mentioned that in a single-boat setup, the distance between the source and the substrate that is immediately above is constant, which leads to nanowires with uniform diameter. In addition to the experimental setup, the total reaction time is yet another critical factor in determining the diameter and length of the resulting nanowires. Chen et al. [1224] have investigated the structure and morphology of the GaN nanowires produced with a reaction time varying from 3 to 48 h, keeping the other process conditions invariant [1224]. For example, a reaction time of 3 h produced a large number of short rodlike structures with diameters of several hundred nanometers. For a period of 3–12 h, the rods continuously grew in length along the axial direction to form wirelike structures while their diameters decreased. However, when the reaction time was made longer than 12 h, the diameters of nanowires dramatically increased. Further, unusual shapes of GaN bulk crystals were observed when the total reaction time exceeded 48 h. In terms of the optical properties, thermal quenching of PL was much reduced in nanowires compared to the bulk and five first-order Raman modes were observed. The peaks were broadened owing to what was reported to be confinement and additional modes not seen in the bulk were observed. It should be mentioned that unless there exists a surface induced potential, the diameters mentioned above are not sufficiently small to lead to discernible confinement in GaN. If heterostructure growth is incorporated into the nanowire process, owing to two degrees of confinement by the geometry of the wire and in the growth direction by the barrier, zero- dimensional electronic systems can be obtained. The process involves using a single one-dimensional growth mechanism that can be easily switched between the constituents of the well and barrier materials. Owing to the VLS growth mechanism being readily amenable for such control, most of the reported data involve longitudinal heterostructure synthesis by this approach. Switching to more standard nomenclature used by device quality heterostructure producers, MBE and OMVPE are very conducive methods for integrating nanowhisker growth with compositional and doping modulation along the growth direction being possible. One can also lump the pulsed laser deposition in this category but the method is not
3.5 The Art and Technology of Growth of Nitrides
as commonly used for semiconductor structures. In the following paragraphs, a few examples of heterostructures produced by these techniques in the nanowire form are discussed. For example, Wu et al. [1225] produced longitudinal Si/SiGe heterostructures using a hybrid pulsed laser ablation/chemical vapor deposition (PLA-CVD) process. In this process, Si and Ge vapor sources could be independently controlled and alternately delivered into the VLS nanowire growth system. As a result, singlecrystalline nanowires containing the Si/SiGe superlattice structure were obtained. Figure 3.244 illustrates a TEM image of two such nanowires in the bright-field mode where the dark stripes delineate the SiGe alloy because its cross section is larger than that of Si. Similarly, GaAs/GaP [1226] and InAs/InP [1227] heterostructured nanowires and ZnSe/CdSe [1228] superlattice nanowires have also been produced. Moreover, the VLS process has also been applied to p–n junctions, coupled quantum dot structures, and heterostructured unipolar and bipolar transistors based on individual nanowires. Methods not requiring an external catalyst, which will be referred to as non-VLS methods for the lack of a better term, have also been used to synthesize longitudinal
Figure 3.244 TEM image of two Si/SiGe superlattice nanowires. Courtesy of P. Yang and Ref. [1225], also reproduced in Ref. [1169].
j739
j 3 Growth and Growth Methods for Nitride Semiconductors
740
nanowire heterostructures. However, they are limited to the growth of metal nanostructures, such as striped Ag/Au nanowires reported by Keating and Natan [1229] and Au/Co nanowire superlattices reported by Valizadeh et al. [1230] Functional device structures in the GaN system have been reported by Kim et al. [1231] using a chemical vapor transport vapor–solid process by introducing Cp2Mg to achieve the p-type dopant. In the same vein Cheng et al. [1232] incorporated Mg doping into GaN nanowires with 35 nm, fabricated by the VLS method. Turning on the Mg doping midway through the synthesis resulted in p–n junction nanowires that showed rectification. A nanowire was successfully placed between two Au pads on a SiO2 template for electrical measurements. Most wires grow such that the crystalline c-axis is normal to the long axis of the nanowire. In another report by Nam et al. [1233] nanostructures were grown by thermal reaction of gallium oxide and ammonia with the resultant morphology being dependent on the ammonia flow rate. Nanowires and polyhedral crystals and nanobelts were observed for ammonia flow rates of 75 and 175 sccm, respectively. The growth orientation of most of the smooth nanowires, as well as the nano belts, was perpendicular to the c-axis (h0 0 0 1i). However, the corrugated nanowires and the large polyhedra grew parallel to h0 0 0 1i. A model has been proposed to account for these variations in morphology and growth orientation in terms of the Ga/N ratio and the different characteristic length of {0 0 0 1} polar surface in the different nanostructures. Cheng et al. [1234] reported ordered nanowires synthesized in anodic porous alumina membrane through a gas reaction of Ga2O vapor in a constant ammonia atmosphere at 900 C. Atomic force microscopy, X-ray diffraction, transmission electron microscopy, and high-resolution electron microscopy were used to determine that the orderly nanostructure consisted of polycrystalline GaN nanoparticles with a hexagonal wurtzite structure of a diameter of about 10–20 nm. In terms of the optical properties, strong visible light emission was observed. Utilizing self-organized nanometer-sized holes as mask appearing in InGaN layer (occurring during InGaN growth in conjunction with the OMVPE process by modifying the growth parameters), Wang et al. [1235] attained high-density GaN nanowire arrays. Optical investigation by RTPL indicated strong emission from an n-GaN nanowire array at 367 nm, the near band edge emission wavelength for n-type GaN. Kuykendall et al. [1236] synthesized GaN nanowires via metal-initiated OMVPE on silicon, c-plane, and a-plane sapphire substrates. The wires were formed via the VLS mechanism with gold, iron, or nickel as catalysts with widths of 15–200 nm. TEM investigations indicated the orientation of the wires to be predominantly along the [2 1 0] or [1 1 0] direction. Wires growing along the [2 1 0] orientation were found to have triangular cross sections. Transport measurements confirmed that the wires were n-type and had an electron mobility of 65 cm2 V1 s1. Khanderi et al. [1237] prepared GaN nanopillars, nanorods, and nanowires by OMVPE on c-plane sapphire substrates using the single-molecule precursors (SMPs) bisazido (dimethylaminopropyl) gallium (BAZIGA, 1) and its ethyl derivative bisazido (diethylaminopropyl) gallium (E-BAZIGA, 2) in a horizontal and vertical stagnation flow cold-wall reactor. At a given growth temperature, H2 flow rate and
3.5 The Art and Technology of Growth of Nitrides
the total pressure were found to influence the morphology, changing it from selforganized nanopillars to randomly oriented nanowires through dense and ordered nanorods. The GaN nanopillars grown by SMPs 1 and 2 in the presence of N2 obeyed autocatalytic VLS growth and exhibited a surface distribution density of 3.8 · 1010–4.5 1012 cm2. On the contrary, GaN nanorods and nanowires were produced in the presence of various amounts of H2, and their surface distribution density, in the case of the nanorods, depended on the H2 flow. The nanostructures were easily detached from the substrate and dispersed in organic solvents and showed strong photoluminescence in the near-UV region (2.7–3.0 eV). Others [1238] also synthesized GaN nanowires by a catalytic chemical vapor deposition method. The nanowires with hexagonal single-crystalline structure had diameters of 10–50 nm and lengths of tens of micrometers and exhibited UV bands at 3.481 and 3.285 eV in low-temperature PL measurements associated with donorbound excitons and donor–acceptor pairs, respectively. The blue shift of UV bands was attributed to quantum confinement effects in the thin GaN nanowires that had diameters of about 11 nm. As for the field emission properties, the turn-on field of nanowires was 8.5 V mm1 and the current density was about 0.2 mA cm2 at 17.5 V mm1, which is sufficient for the applications of field emission displays and vacuum microelectronic devices. Optically pumped ultraviolet–blue stimulated emission in single-monocrystalline GaN nanowires has been reported [1239,1240]. Characterization methods such as near-field and far-field optical microscopy as well as the standard spectral measurements were used to analyze the waveguide mode structure and spectral properties of the resultant radiation at room temperature. From a series of investigations dealing with the optical spectra, Johnson et al. [1239] determined the dependence of the optical spectra and optical emission intensity on the pump power and made a case for lasing action in an isolated single GaN nanowire. The pump power dependence of the emission intensity showed the usual spontaneous emission region, stimulated emission region, and finally nearly a saturated behavior for pump intensities above 1.4 mJ cm2 due to gain pinning. Because the wire length cannot be changed, the standard method of monitoring the threshold current dependence on cavity length could not be used to determine the gain accurately. However, using the standard gain expression in terms of the cavity length and mirror reflectivities, the gain was estimated between 400 and 1000 cm1. A far-field image of a single nanowire under 3 Jcm2 optical pumping is shown in Figure 3.245. Gradecak et al. [1240] also reported optically pumped room-temperature stimulated emission in GaN nanowires with triangular cross sections grown by OMVPE along the nonpolar h1 1 2 0i direction. The wires were single-crystalline in terms of their structure and triangular in terms of their cross-sectional shape. Optical excitation of these wires led to thresholds for stimulated emission of 22 kW cm 2, lowering substantially the previously reported GaN nanowires. Using cleaved ends of nanowires as mirrors for the resonant cavity in high-temperature grown nanowires, the authors observed spontaneous emission from both ends of the wire as well as its body at low optical excitation intensities as seen in Figure 3.246a for an excitation optical power density of 4 kW cm 2. Upon increasing the excitation intensity, however, the
j741
j 3 Growth and Growth Methods for Nitride Semiconductors
742
Figure 3.245 Image of an individual and isolated GaN nanowire what was reported to be laser in action. The sample was backilluminated with a lamp to show the nanowire. The nanowire was excited with about 3 mJ cm2. The color emission is associated with laser light emanating from the ends of the nanowire. Courtesy of R. Saykally and Ref. [1239].
emission leaking through the body gave way to facet emission as shown in Figure 3.246b and c for excitation power densities of 17 kW cm 2 and 66 kW cm 2, respectively, indicative of waveguiding features of the nanowire against a lower refractive index air surrounding. Of particular interest is that a spectra from the end of the nanowire at higher excitation densities exhibit a red shift with well-defined Fabry–Perot cavity modes as seen in Figure 3.246d. The mode spacing Dl0 for a cavity with length L is given by Dl0 ¼ l02 /[2L(nr l0(dnr/dl0))], where nr is the GaN index of refraction and dn/dl is the first-order dispersion of the index of refraction (for l0 ¼ 378 nm,18 nr ¼ 2.64, and dnr/dl0 ¼ 0.0077 nm1), see Volume 3, Chapter 2, Equation 2.36 and its derivation for details. It is interesting to note that the mode spacing expression, describing the experimentally observed cavity mode spacing, unlike many of the quantum well lasers discussed in Volume 3, Chapter 2 where the mode spacing corresponding to cavity lengths shorter than fabricated ones has been attributed to lateral morphological features of the grown structures. Gradecak et al. [1240] also estimated the quality factor Q of their n-type GaN nanowire cavities to be in the 500–700 range, where Q ¼ l0/Dl0 and Dl0 is the FWHM of the cavity mode, the accuracy of which may have been affected by the spectrometer resolution. Three-dimensional finite difference time domain (FDTD) calculations were performed for freestanding triangular and cylindrical GaN nanowires with the same cross-sectional area and length in an effort to discern the effect of the triangular cross-sectional nature of the nanowires used. Both geometries turned out to have modes with quality factors of 1000–1500 higher than those estimated from experimental data or macroscopic models that do not take into account the finite size of the nanowire. The same calculations also indicate comparatively better wavelength stability of the quality factor for triangular versus cylindrical nanowires. It should, however, be mentioned that semiconductor lasers, particularly gas lasers, have much better Q factors. The Q factor in lasers is so high that it loses its impact and is thus replaced with finesse factor.
3.5 The Art and Technology of Growth of Nitrides
Figure 3.246 Excitation intensity-dependent PL obtained in an n-type high temperature grown (995 C which is close to typical GaN growth temperatures by OMVPE) GaN nanowire. The top row images a single nanowire excited with (a) 4, (b) 17, and (c) 66 kW cm2. (d) PL spectra
recorded at the end of 13 mm long nanowire for 4 and 18 kW cm2 excitation intensities, showing Fabry–Perot cavity modes with average mode spacing of 1.1 nm for the higher excitation intensity. Courtesy of C.M. Lieber.
Electrooptic modulators are an integral part of the optical systems, as they provide optical switching and have advantages over active switching/modulation of semiconductor laser to avoid wavelength instabilities and chirping. Among the assortments of devices made from nanowires are electrooptic modulators [1241].
j743
j 3 Growth and Growth Methods for Nitride Semiconductors
744
Greytak et al. [1241] achieved electric field modulation of visible and ultraviolet nanoscale lasers consisting of single CdS or GaN nanowires using integral microfabricated electrodes. Modulation of laser emission intensity was achieved with no detectable change in the laser wavelength. The devices could also be operated below the lasing threshold to modulate the intensity of light propagating within the nanowire waveguide. Studies of the electric field dependence in devices of varied geometries indicate that the modulation is due to the electroabsorption mechanism. The stimulated emission spectrum shown in Figure 3.247a exhibits multimode features with a dominant peak at 373 nm. An important feature of the modulator is that the intensity modulation occurs without a change in the peak positions as the applied voltage is varied. Detailed measurements of modulation M versus V [Figure 3.247b] carried out below the lasing threshold show that the edge emission intensity varies linearly with voltage over the 45 V range investigated. These results show that 20% modulation can be readily achieved above or below the threshold for lasing in these GaN nanowire electrooptic modulators. In an effort to not only demonstrate light emission but also to do so with increased efficiency, Qian et al. [1242] reported the growth and characterization of core/ multishell nanowire radial heterostructures and their application as efficient and synthetically tunable multicolor sources. The structures were prepared by OMVPE with an n-GaN core and InxGa1xN/GaN/p-AlGaN/p-GaN shells. Being the smallest bandgap semiconductor in the entire structure, emission takes place in the InxGa1xN layer, the composition of which as well as its thickness can be changed for wavelength adjustability. The schematic triangular structure along with the band structure (neglecting band bending) of the nanowire is shown in Figure 3.248. Cross-sectional transmission electron microscopy studies revealed that the core/ multishell nanowires are dislocation-free single crystals with a triangular morphology. Energy-dispersive X-ray spectroscopy clearly shows shells with distinct chemical compositions. For carrier injection metal contacts were formed separately onto the p-type outer shell and n-type core at the ends of individual nanowires (inset, Figure 3.249a). The I–V characteristics showed typical p–n diode characteristics with a turn-on voltage of around 3.5 V, as displayed in Figure 3.249a. Electroluminescence (EL) images collected from the nanowire LEDs with In mole fraction in the range of 1 to 40% showed the expected red shift with increasing In as seen from colored images of Figure 3.249b. Normalized EL spectra collected from five representative CMS nanowires with varying In composition in the InGaN shell exhibited emissions peaking at 367, 412, 459, 510, and 577 nm, consistent with band edge emission from InxGa1xN with In compositions of 1, 10, 20, 25, and 35%, respectively. The estimated external quantum efficiencies (QEs) in the CMS nanowire LEDs was on the order of 5.8% at 440 nm and 3.9% at 540 nm. Notably, these values are several times better than the best QEs reported for nanoscale LEDs and are marching toward the InGaN-based SQW thin film LEDs at similar emission wavelengths. Optimization of the structural design is likely to lead to increased efficiency, which might bring the efficiency numbers close to nitride-based LEDs that are commercially available.
3.5 The Art and Technology of Growth of Nitrides
30 (b)
Modulation (%)
20 10 0 –10 –20 –30 –45
–30
–15 15 0 Voltage (V)
Figure 3.247 (a) GaN nanowire emission spectrum above stimulated emission threshold, with þ45 V bias and without bias applied. Insets: spectra recorded below threshold: top, superimposed PL and photo image of device (scale bar is 10 mm). Bottom, edge emission
30
45
spectra for three different bias values, þ45, 0, and -45 V below threshold (b) Modulation M versus V, below threshold. For this device, ttop ¼ 100 nm, tbot ¼ 50 nm. All data were recorded at room temperature. Courtesy of C.M. Lieber.
j745
j 3 Growth and Growth Methods for Nitride Semiconductors
746
n-GaN Inx Ga1–xN
(a)
GaN p-AlGaN p-GaN
p-AlGaN
p-GaN GaN
Inx Ga1–xN n-GaN EC
(b)
EV Figure 3.248 Cross-sectional view of a core multishell (CMS) nanowire structure and the associated schematic energy band diagram neglecting band bending. Courtesy of C.M. Lieber.
Kim et al. [1243] fabricated a Schottky junction diode with GaN nanowire synthesized by chemical vapor deposition and studied its electrical transport properties. Al and Ti/Au were used as a Schottky barrier and ohmic contact, respectively. The measured current–voltage characteristics exhibited clear rectifying behavior and no reverse bias breakdown was observed up to the measured voltage, 5 V. Huang et al. [1244] carried the GaN nanowires one step further by fabricating FETs in individual nanowires in much the same fashion as those built in carbon nanotubes. Gate-dependent electrical transport measurements confirmed a con-
Figure 3.249 (a) I–V characteristics of a core multishell (CMS) nanowire device, the top view of which is shown in the inset in the form of a field emission scanning electron microscopy image. Scale bar is 2 mm. (b) Optical microscopy images collected from around theopaque p-contact ofcore multishell nanowire LEDs with increasing In
" concentration in the shell quantum well and in forward bias, showing purple, blue, cyan, green, and near yellow emission, respectively. (c) Normalizedelectroluminescencespectraobtained from five representative multicolor CMS nanowire LEDs. Courtesy of C.M. Lieber. (Please find a color version of this figure on the color tables.)
3.5 The Art and Technology of Growth of Nitrides
Figure 3.249 (Continued )
j747
j 3 Growth and Growth Methods for Nitride Semiconductors
748
Figure 3.249 (Continued )
3.5 The Art and Technology of Growth of Nitrides
(c)
1.0
Normalized intensity (au)
0.8
0.6
0.4
0.2
0.0 300
400
500
600
700
Wavelength (nm) Figure 3.249 (Continued )
ductance modulation by more than 3 orders of magnitude. Electron mobilities determined for the GaN NW FETs, which were estimated from the transconductance, were as high as 650 cm2 V1 s1. FETs, conveniently provided by the long length of NWs, where the output current is dominated by the electron mobility, have been used to deduce the electron mobility along a single-walled carbon nanotubes (SWCNs) [1245]. Specifically, transistors with channel lengths exceeding 300 mm have been used to extract the drift mobility to be 79 000 cm2 V1 s1 with an estimate of the intrinsic mobility >100 000 cm2 V1 s1 at room temperature, although largely temperature insensitive. It should be pointed out that a microscopic theory to explain the behavior of the intrinsic mobility is lacking which also applies to GaN before one should attempt to read too much into the 650 cm2 V1 s1 mobility figure. The GaN NW FETs studied by Huang et al. [1244], a cartoon of which is shown in Figure 3.250b, were prepared by dispersing a suspension of GaN NWs in ethanol onto the surface of an oxidized silicon substrate with 1–10 Ocm resistivity and 600 nm thick SiO2. Field emission (FE) SEM image of a GaN nanowire FETwhere the scale bar in white is 2 mm is shown in Figure 3.250a. The conducting silicon below was used as the back gate. Source and drain electrodes were formed by electron beam lithography followed by Ti/Au (50/70 nm) metallization. The current–voltage measurements were performed at room temperature. Figure 3.250c shows a set of typical current versus source-drain voltage characteristics obtained from a single GaN NWFET for different gate voltages (Vg). The two-terminal I–Vsd characteristics are all linear, indicating ohmic behavior of metal contacts. The transfer characteristics in
j749
j 3 Growth and Growth Methods for Nitride Semiconductors
750
Figure 3.250 (a) Field emission (FE)-SEM image of a GaN nanowire FET where the scale bar in white is 2 mm. (b) Schematic diagram of a nanowire FET (NW FET) with the inset showing the. (c) Gate-dependent I–Vds characteristics obtained from a 17.6 nm diameter GaN NW. The gate voltage steps for each I–Vds curve are
indicated. (d) I–Vg data recorded for values of Vds between 0.1 and 1 V with the inset showing the conductance G versus the gate voltage. The vertical current scale is in terms of microamperes. Courtesy of C.M. Lieber and Ref. [1244].
the form of I versus Vg for a GaN NW device for different source-drain voltages (Figure 3.250d) are consistent with an n-channel metal oxide semiconductor FET. Moreover, the conductance modulation of the GaN NW FET exceeds 3 orders of magnitude over a gate voltage range of 8 to þ6 V.
3.5 The Art and Technology of Growth of Nitrides
(d)
G (nS)
I ds (mA)
1.5
Vds = 1.0 V
103
2
0.9
102
0.8
101
0.7
100 –5
0
5
0.6
Vg (V)
1
0.5 0.4 0.3
0.5
0.2 0.1
0 –8
–4
0 V g (V)
4
8
Figure 3.250 (Continued )
High-sensitivity UV photodetectors using GaN nanowires have also been reported [1246]. One of the major bottlenecks for all the nanowires and dots is the electrical contact issue. In this particular approach, Chen et al. [1246] prepared nanowires by growth between two previously prepared posts. Prior to nanowire growth, the arrays of highly n-doped GaN(0 0 0 1) and Ni film electrodes were patterned on the sapphire (0 0 0 1) substrate via photolithography, liftoff, and reactive ion etching (RIE) methods established for epi-GaN layers [1247], and GaN etched down to the sapphire substrate for electrical isolation between the posts. This was followed by the sputter deposition of a 10 nm Au film on the entire patterned substrate as the catalyst for subsequent nanowire growth. During the nanowire process, the thin Au catalyst layer is in the form of nanoparticles dispersed discretely on the substrate. The nanowire growth was achieved in a chemical vapor deposition (CVD) furnace at 800–1050 C and 760 Torr using NH3 and Ga as source reagents. With some degree of control over the deposition parameters, GaN nanowires were symmetrically with a 30 tilt angle on the sidewalls of the GaN {2 1 1 0} facets. Fortuitously, a large number of nanowires could grow between the electrode posts. The typical diameter and length of the nanowires were 40 20 nm and 10 2 mm, respectively. The photoconductive responsivity (see Volume 3, Chapter 4 for a detailed discussion of responsivity) measured for 1.0 V bias was about 3 · 105 A/W at 4.0 eV. When the bias was increased to 10 V, the photo responsivity increased reaching a value >106 cm3, which is remarkable. An analysis of the data indicated photoconductive gain of >107. The high gain is indicative of carrier lifetime being much longer than the transit time. Chen et al. [1246] observed that the photoresponsivity scaled with the number of wire bridging the posts thus arguing that the photoconductivity originates in the wires. Interestingly, the conductivity was noted to depend on the surface conditions, particularly exposure to
j751
j 3 Growth and Growth Methods for Nitride Semiconductors
752
air or oxygen, and heating in vacuum that desorbs the oxygen. The extremely high gains were then attributed to O causing n-type doping at the surface and bending the conduction band downward (toward the surface) which is inconsistent with other published work as to the effect of O on standard GaN surface (see Volume 2, Chapter 1). Nanowires of semiconductors for which lattice-matched substrates are not available, as is the case for GaN, offer an avenue for obtaining nearly structurally defectfree material. This is owing in part to the limited area of registry required and also the ability of the nanowires to relax without having to generate extended defects. The latter is also beneficial in that strain-induced polarization along the polar direction can be minimized, an advantage for heterojunction systems such as green InGaN LEDs in a GaN matrix, although presence of large strain has also been reported. Consequently, growth on both Si and sapphire substrates has been investigated. In one investigation, Tu et al. [1248] prepared GaN pillars on Si(1 1 1) substrates by MBE without any catalyst. Depending on the growth parameters, the reported nanorods exhibited lateral dimensions in the range of 10 and 800 nm, and lengths between 50 and 3 mm. The tops of the nanorods are typically flat and hexagonal shaped. In the study of Tu et al. [1248], an extra PL peak at 363 nm originating from nanorods was observed at 66 K, which was attributed to the surface states, although a basis for which was not really provided. Micro-Raman spectroscopy on a single nanorod revealed E1 and E2 modes at 559.0 and 567.4 cm1, respectively, indicating a straininduced shift. In another investigation [1249], wurtzite single-crystal GaN nanocolumns were grown by plasma-assisted molecular beam epitaxy on both Si(0 0 1) and (1 1 1) substrates. Nanocolumns with diameters in the range of 20–40 nm were nearly extended defect free and grew along the [0 0 0 1] c-direction on Si substrates with (0 0 1) and (1 1 1) surfaces. Note that in uniform growth experiments, cubic GaN typically results on (0 0 1) Si and GaAs substrates. The formation of wurtzite GaN is yet another proof that if not bounded by the substrate, wurtzite phase is the preferred polytype with c-direction being the dominant growth direction. Photoluminescence measurements in nanocolumns were reported to exhibit intense and narrow excitonic emissions. Raman scattering data showed that the nanocolumns are strain free, unlike the results of Tu et al. [1248] Self-assembled columnar AlGaN/GaN nanocavities having an active region of GaN quantum disks embedded in an AlGaN nanocolumn and straddled by top and bottom AlN/GaN Bragg mirrors have also been prepared [1250]. Unlike the DBRs in uniform growth cases, the nanocavity straddled with DBRs exhibited no cracks or detectable extended defects owing to the relaxation of the films and the large free surface to volume ratio in nanocolumns. The emission from the active region was observed to be within the peak stop band of DBR, which resulted from their ability to tune the Al content and the GaN disk thickness. Quantum confinement effects depending on both the disk thickness and the inhomogeneous strain distribution within the disks have been clearly observed. Switching on to sapphire substrates, self-formed nanocolumns have been produced with RF MBE when grown under very N-rich growth conditions (with Ga-polarity) at high substrate temperatures [1195]. In this catalyst-free and selforganization process, the Ga flux, determined by the cell temperature for a given cell
3.5 The Art and Technology of Growth of Nitrides
and deposition reactor, was optimized for a fixed N flux to reduce the column diameter to 40–45 nm. The side and top views of GaN nanorods as prepared are shown in Figure 3.251. The GaN nanocolumn growth was also followed by deposition of GaN/Al0.18Ga0.82N multiquantum disk (MQD) with 10 pairs of GaN(6 nm)/ Al0.18Ga0.82N(9 nm) multilayer structure. The blue shift in a PL peak wavelength at room temperature was observed for the MQD sample, probably due to the quantum-size effect on the growth direction. GaN nanocolumns with a GaN/AlN superlattice (SL) region have been grown by RF plasma assisted MBE [1251]. Owing to mainly the lower defect concentration in the nanostructures, the PL peak intensity of the GaN/AlN SL nanocolumns was 300–500 times stronger than that of conventional 3.75 mm thick GaN grown by OMVPE with a dislocation density of 3–5109 cm2. The peak wavelengths of the GaN (10.2 ML)/AlN (15.2 ML) SL and GaN (7.7 ML)/AlN (12.4 ML) SL were observed at 420 and 380 nm, respectively. The calculated transition wavelength taking the polarization-induced field agreed well with experimental value, suggesting that GaN/AlN SL nanocolumns involve a large built-in electrostatic field of about 5.8 MV cm1. As has been suggested, if one can assume that the strain-induced polarization is minimized owing to relaxation of the films, this field is then mainly due to the spontaneous polarization. The integrated PL intensity was increased by a factor of 2.2 in going from contiguous to columnar structures. Switching over to functional devices, GaN/GaN multiple quantum disk (MQD) nanocolumn LEDs have been fabricated on n-type Si(1 1 1) substrates [1252]. The samples were grown with RF-MBE under N-rich conditions at high temperatures, nominally leading to columnar growth. In the growth, after subjecting the substrate surface to the Ga beam, RF plasma excited nitrogen was provided on the surface to form GaN dots, both at 530 C. This was followed by columnar growth. Upon completion of the columnar structure, the growth conditions were changed to more typical ones for growth of a Mg-doped p-type GaN layer. The columns were observed to widen to form trapezoidal forms during the p-layer growth with the overall p-layer having relatively larger grains. The nanocolumn LED showed rectifying I–V characteristics with a typical turn-on voltage of 2.5–3.0 V at room temperature [1253]. Electroluminescence was recorded through semitransparent electrodes with emission from green (530 nm) to red (645 nm) observed depending on the InN mole fraction mainly. This process was later expanded to include demonstration of LEDs across the entire visible spectrum. Of particular interest is the carrier lifetime in InGaN quantum well, as part of these nanocolumns were about 10 ns for an InN composition corresponding to green LEDs, whereas in contiguous quantum wells this was 80 ns because of a much larger quantumconfined Stark effect in the latter structure caused by an additional strain-induced electric field. The long lifetimes cause reduction in the light intensity that can be obtained in LEDs, particularly the green color, and provide the motivation for exploring nonpolar (a-plane) and partially polar (m-plane rotated along the c-axis) orientations in GaN. The a-plane structures suffer from large stacking faults due to low formation energy and thus radiative recombination is weak. This is exacerbated by the difficulty to incorporate InN in concentrations needed for visible
j753
j 3 Growth and Growth Methods for Nitride Semiconductors
754
Figure 3.251 (a) Side and (b) top views of GaN nanorods prepared by RF MBE on (0 0 0 1) sapphire by growing under high N fluxes (still Ga polarity owing to the AlN nucleation layer on sapphire) and high growth temperatures, for example, 800 C. Courtesy of K. Kishino.
3.5 The Art and Technology of Growth of Nitrides
LEDs in a-plane GaN. The m-plane GaN is more conducive for InN incorporation, and the emission intensity, although much better than on the a-plane, still lags behind that obtained on c-plane GaN. Perhaps with advances on the preparation of these structures, the emission intensity can be increased to make this approach viable. Stimulated emission with a relatively low optical threshold excitation power density has been reported by Kikuchi et al. [1254] for GaN nanocolumns grown on (0 0 0 1) sapphire substrates by RF plasma assisted molecular beam epitaxy. The threshold excitation power density in the photopumped excitation experiment using a 355 nm Nd : YAG laser was 198 kW cm2 at room temperature. For the lower excitation conditions, as in PL experiments, using a 325 nm He–Cd laser, the spontaneous emission peak was observed at 363.2 nm and the intensity was some 2030 times stronger than that from a 3.7 mm thick OMVPE-grown GaN film with a dislocation density of 35109 cm2. 3.5.16.5.1 Coaxial Heterostructures Coaxial nanowires are both fundamentally interesting and have technological potential in that the surface recombination velocity for emission from the core material is much reduced by avoiding exposure of the core to atmospheric conditions. Coaxial structures can be fabricated by coating an array of nanowires with a conformal layer of a second material. The coating method chosen should allow excellent uniformity and control of the sheath thickness. Cladding nanowires with amorphous layers of SiO2 or carbon is reasonably straightforward to synthesize as routinely demonstrated in the literature. The same is true for colloidal quantum dots formed in group II–VI semiconductor heterostructures. In this vein, self-organized GaN quantum wire UV lasers have been demonstrated [1167]. Similarly, synthesis of GaN/Al0.75Ga0.25N (Figure 3.252), ZnO/GaN [1255], and Si/Ge [1256] core–sheath structures using a chemical vapor transport method has been reported as well. Obviously the core and sheath materials must be compatible, as in the case for any viable heterostructure system. Qian et al. [1257] reported a new and general strategy that makes possible the synthesis of well-defined doped core/shell/shell (CSS) nanowire heterostructures for efficient carrier injection in the context of photonic devices. The n-GaN/InGaN/p-GaN CSS nanowire structures were grown by OMVPE, which begins with a VLS process mediated by an initial metal nanocluster (Ni in this case). PL data showed that the optical properties could be controlled by the CSS structure with strong emission from the InGaN shell centered at 448 nm. Optical devices made by simultaneously contacting the n-type core and outer p-type shell of the CSS nanowires demonstrated LED operation with bright blue emission from the InGaN shell. An SEM image of Si-doped GaN nanowires obtained following axial elongation, shown in Figure 3.253a, indicates a high yield of uniform nanowire cores. The lengths of the nanowire cores depend directly on growth time and those used in this study were 10–20 mm. A bright-field TEM image of an n-type GaN core (Figure 3.253b) shows that the nanowire diameter is essentially the same as the nickel nanocluster diameter, as shown with the arrow and expected for the VLS process. Corresponding TEM images of n-GaN/InGaN CS nanowires shown in Figure 3.253c
j755
j 3 Growth and Growth Methods for Nitride Semiconductors
756
Figure 3.252 Transmission electron microscopy image of a GaN/ AlGaN core–sheath nanowire. Courtesy of P. Yang and Ref. [1169].
demonstrate that the overall diameters increased following InGaN shell growth and that the nanowire has a triangular cross section. EDX mapping shown in the inset of Figure 3.253c illustrates that nanoclusters at the CS nanowire ends are nickel (red) (catalyst material), whereas the larger core–shell nanowire structure contains Ga, In, and N. Large-scale wurtzite GaN nanowires and nanotubes were grown by He et al. [1258] through direct reaction of metal gallium vapor with flowing ammonia in a horizontal oven in a temperature window of 850–900 C. The cylindrical structures so produced were up to 500 mm in length with diameters between 26 and 100 nm. The size of the nanowires was reported to depend on the temperature and the NH3 flow rate. In a somewhat different vein, it must be mentioned that multifunctional structures are also possible that harness optical, ferromagnetic, ferroelectric, and superconducting properties of materials but in the form of nanowires and processes to synthesize such structures have been demonstrated [1259]. For example, TiO2 nanoribbons were used as substrates for the thin-film growth of TiO2 (ferroelectric behavior), transition metal doped TiO2 (ferromagnetic behavior), and ZnO (light emission and detection) using pulsed laser deposition (PLD). 3.5.16.5.2 Nanotubes The epitaxial core–sheath method can be used to synthesize single-crystalline nanotube materials by simply dissolving the inner core. This synthetic approach requires that the core and sheath materials exist in epitaxial registry and possess vastly differing response to wet chemical etches. This epitaxial casting strategy has been applied by Goldberger et al. [1255] to GaN nanotubes with
3.5 The Art and Technology of Growth of Nitrides
Figure 3.253 (a) SEM image of n-GaN nanowires grown using nickel nanoclusters as catalyst with the scale bar being 20 mm. (b) Bright-field TEM image of n-GaN nanowire core showing its diameter of the nanowire to be the same as the diameter of the corresponding nanocluster; the white arrow highlights the
nanocluster. The scale bar is 50 nm. (c) Brightfield TEM image of n-GaN/InGaN CS nanowire wherein the white arrow highlights the Ni nanocluster. Scale bar is 50 nm. Inset: EDX map of nickel (red) and gallium (blue) for this same nanowire. Courtesy of C.M. Lieber and Ref. [1257].
inner diameters of 30–200 nm and wall thicknesses of 5–50 nm. In this approach, hexagonal ZnO nanowires were used as templates for the epitaxial overgrowth of thin GaN layers in an OMVPE reactor. The ZnO core regions were subsequently removed by simple thermal reduction and evaporation in a NH3/H2 mixture. The resultant set of GaN nanotubes is shown in the TEM image of Figure 3.254. In the nitride family
j757
j 3 Growth and Growth Methods for Nitride Semiconductors
758
Figure 3.254 Transmission electron microscopy image of a set of single-crystalline GaN nanotubes prepared using the epitaxial casting methodology in which ZnO nanorod templates on which the GaN sheath was deposited is removed thermally in an OMVPE reactor which is also used for GaN growth. Courtesy of P. Yang.
nanotubular AlN has also been reported [1260]. Not directly related but relevant in the larger context of nanorods, amorphous SiO2 nanotubes have been synthesized using silicon nanowire templates that were then etched away with XeF2 to yield silica tubes [1261]. A method using arc discharge in a nitrogen atmosphere for synthesizing large quantities of GaN–carbon composite nanotubes and GaN nanorods has also been reported [1262]. The reaction was attained by a dc arc discharge between a graphite anode filled with a mixture of GaN, graphite, and nickel powders and a graphite cathode in nitrogen atmosphere. The resulting GaN was rodlike fillings in the composite tubes and the isolated GaN nanorods had diameters in the range of 7–45 nm and lengths of up to 40 mm. The outer graphitic shells of the composite carbon nanotubes had thicknesses in the range of 1–8 nm. The authors stressed the
3.6 Concluding Remarks
crucial role played by nitrogen atmosphere for the growth of the GaN nanorods fillings and the individual GaN nanorods.
3.6 Concluding Remarks
In conclusion, since the last edition, the technology and science of growth, strongly coupled to a plethora of characterization methods, have come a long way. Although, the structural and electronic properties of GaN are not yet near perfection, most of the lagging behavior is thought to be a result of not having high-quality lattice-matched substrates at the disposal of the grower. It is temporarily accepted that armed with current understanding of the growth mechanisms and technological developments the growers are sufficiently poised to prepare high-quality films. The growers can at least bring the quality to the point where impurities introduced from the sources and/or the environment take a reasonably dominant role so that they can be made the point of focus. In contrast, growth of the other two binaries, AlN and InN, despite a good deal of effort on InN, is lagging behind that of GaN, which, of course, is moving forward. The case of InN is a celebrated one owing to drastically different atomic radii of In and N atoms, and high vapor pressure at temperatures well below the melting temperature make this binary a difficult one. Very large ammonia flow rates are needed to grow InN in the face of exceedingly high growth temperature sensitivity. It now appears that the bandgap of InN is in the sub 1 eV range and high bandgap measured in the past and still measured in films deposited by techniques such as VLS from powders might cumulatively have something to do with the size of crystallites (confinement), doping level (Moss–Burstein effect) and presence of O (compositional shift due to InO). Understanding and technology of the dopant issues in InN and AlN binaries are really in a nascent state, particularly p-type doping, which neither of these two binaries has accomplished. There may be fundamental barriers to attaining p-type in InN as surface exhibits electron accumulation and defects are decidedly donor type in addition to impurities and the acceptor level is very deep in AlN. As for the ternaries, growth and doping issues particularly with high fractions of In and Al in the GaN lattice still are in need of much more progress. In fact, the inhomogeneities present in InGaN have not received as much attention as they perhaps warrant in part because they are to a large extent responsible for good LED performance. As demand and expectations for higher performance LEDs grow, inhomogeneities and strain and compositional gradient induced polarization issues are likely to present impediments. In fact, exploration of nonpolar surfaces is indicative of this. Naturally, the aforementioned issues are much more pivotal for lasers. Getting back to the opening statement of the concluding remarks, the growth in general has come a long way and a vast literature basis is now available as could easily be garnered from the size of this chapter which ballooned well beyond that originally envisioned.
j759
j 3 Growth and Growth Methods for Nitride Semiconductors
760
References 1 Tiede, E., Thimann, M. and Sensse, K. (1928) Chemische Berichte, 61, 1568. 2 Johnson, W.C., Parson, J.B. and Crew, M.C. (1932) Journal of Physical Chemistry, 36, 2561. 3 Juza, R. and Hahn, H. (1938) Zeitschrift fur Anorganische und Allgemeine Chemie, 239, 282. 4 Maruska, H.P. and Tietjen, J.J. (1969) Applied Physics Letters, 15, 327. 5 Amano, H., Sawaki, N., Akasaki, I. and Toyoda, Y. (1986) Applied Physics Letters, 48, 353. 6 Yoshida, S., Misawa, S. and Gonda, S. (1983) Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 1, 250. 7 Akasaki, I., Amano, H., Kito, M. and Hiramatsu, K. (1991) Journal of Luminescence, 48–49, 666. 8 Nakamura, S. and Mukai, T. (1992) Japanese Journal of Applied Physics, 31, L1457–LI459. 9 Osamura, K., Nakajima, K., Murakami, Y., Shingu, P.H. and Ohtsuki, A. (1972) Solid State Communications, 11, 617. 10 Slack, G.A. and McNelly, T.F. (1976) Journal of Crystal Growth, 34, 263. 11 Molnar, R.J., Goetz, W., Romano, L.T. and Johnson, N.M. (1997) Journal of Crystal Growth, 178, 147. 12 Popovici, G. and Morkoc, H. (1999) Growth and doping of and defects in IIInitrides, in GaN and Related Materials II: Optoelectronic Properties of Semiconductors and Superlattices, vol. 7 (ed. S.J. Pearton, series ed. M.O. Manaresh), Gordon and Breach, Amsterdam, pp. 93–172. 13 Liu,L.andEdgar,J.H.(2002)Substratesfor gallium nitride epitaxy. Materials Science & Engineering R: Reports, 37, 61–127. 14 Strite, S. and Morkoc, H. (1992) GaN, AlN and InN: a review. Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 10, 1237. 15 Leszczynski, M., Grzegory, I. and Bockowski, M. (1993) Journal of Crystal
16
17 18
19
20
21
22
23
24
25
Growth, 126, 601; Porowski, S. (1995) Acta Physica Polonica A, 87, 295. Porowski, S., Grzegory, I. and Jun, J. (1989) High Pressure Chemical Synthesis (eds J. Jurczak and B. Baranowski), Elsevier, Amsterdam, p. 21. Maruska, H.P. and Tietjen, J.J. (1969) Applied Physics Letters, 15, 327. Akasaki, I. and Amano, H. (1996) International Symposium on Blue Lasers and Light Emitting Diodes, Chiba University, Japan, pp. 11–16. Popovici, G., Mohammad, S.N. and Morkoc, H. (1998) Deposition and properties of III-nitrides by molecular beam epitaxy, in Group III Nitride Semiconductor Compounds: Physics and Applications (ed. B. Gil), Clarendon Press, Oxford. € Botchkarev, A. and Kim, W., Aktas, O., Morkoc, H. (1996) Reactive molecular beam epitaxy of wurtzite GaN grown with ammonia: material characteristics and growth kinetics. Journal of Applied Physiology, 79 (10), 7657–7666. Park, S.S., Park, I.-W. and Choh, S.H. (2000) Japanese Journal of Applied Physics, 39, L1141. Detchprohm, T., Amano, H., Hiramatsu, K. and Akasaki, I. (1993) Journal of Crystal Growth, 128, 384. Huang, D., Yun, F., Visconti, P., Reshchikov, M.A., Wang, D., Morkoc, H., ., Tsen, Rode, D.L., Farina, L.A., Kurdak, C K.T., Park, S.S. and Lee, K.Y. (2001) Hall mobility and carrier concentration in GaN free-standing templates grown by hydride vapor phase epitaxy with high quality. Solid State Electronics, 45 (5), 711–715. Lo, Y.H. (1991) New approach to grow pseudomorphic structures over the critical thickness. Applied Physics Letters, 59 (18), 2311–2313. Teng, D. and Lo, Y.H. (1993) Dynamic model for pseudomorphic structures grown on compliant substrates: an approach to extend the critical thickness. Applied Physics Letters, 62 (1), 43–45.
References 26 Ejeckam, F.E., Lo, Y.H., Subramanian, S., Hou, H.Q. and Hammons, B.E. (1997) Applied Physics Letters, 70, 1685. 27 Ejeckham, F.E., Seaford, M.L., Lo, Y.H., Hou, H.Q. and Hammons, B.E. (1997) Applied Physics Letters, 71, 776. 28 Powell, A.R., Iyer, S.S. and LeGoues, F.K. (1996) Applied Physics Letters, 64, 173. 29 Jesser, W.A., van der Merwe, J.H. and Stoop, P.M. (1999) Misfit accommodation by compliant substrates. Journal of Applied Physics, 85 (4), 2129–2139. 30 Yang, Z., Guarin, F., Tao, I.W. and Wang, W.I. (1995) Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 13, 789. 31 Pearsall, T.P. (1996) III-Vs Review, 9, 38. 32 Touloukina, Y.S., Kirby, R.K., Taylor, R.E. and Lee, T.Y. (eds) (1977) Thermal Expansion, Thermal Properties of Matter, vol. 13, Plenum, New York. 33 Tairov, Y.M. and Tsvetkov, V.F. (1978) Journal of Crystal Growth, 43, 209. 34 Cantwell, G., ZN Technology, Inc. (2001), private communication. 35 Ruterana, P., Vermaut, P., Nouet, G., Salvador, A. and Morkoc, H. (1997) Surface treatment and layer structure in 2H-GaN grown on the (0 0 0 1)Si surface of 6H-SiC by MBE. MRS Internet Journal of Nitride Semiconductor Research, 2, 42. 36 Powell, J.A., Larkin, D.J. and Trunek, A.J. (1998) Use of gaseous etching for the characterization of structural defects in silicon carbide single crystals, in Silicon Carbide, III-Nitrides, and Related Materials, vols 264–268 (eds G. Pensl, H. Morkoc, B. Monemar and E. Janzen), Trans Tech Publications, Sweden, pp. 421–424. 37 Powell, J.A., Larkin, D.J., Neudeck, P.G., Yang, J.W. and Pirouz, P. (1994) Investigation of defects in epitaxial 3C-SiC, 4H-SiC and 6H-SiC films grown on SiC substrates, in Silicon Carbide and Related Materials (eds M.G. Spencer,
38
39
40
41
42
43
44
R.P. Devaty and J.A. Edmond), IOP Publishing, Bristol, pp. 161–164. Lee, C.D., Ramachandran, V., Sagar, A., Feenstra, R.M., Greve, D.W., Sarney, W.L., Salamanca-Riba, L., Look, D.C., Bai, B.S., Choyke, W.J. and Devaty, R.P. (2001) Properties of GaN epitaxial layers grown on 6H-SiC(0 0 0 1) by plasmaassisted molecular beam epitaxy. IEEE Journal of Electronic Materials, 30 (3), 162–169. Lin, M.E., Strite, S., Agarwal, A., Salvador, A., Zhou, G.L., Teraguchi, N., Rockett, A. and Morkoc, H. (1993) GaN grown on hydrogen plasma cleaned 6H-SiC substrates. Applied Physics Letters, 62 (7), 702–704. Bernhardt, J., Schardt, J., Starke, U. and Heinz, K. (1999) Epitaxially ideal oxide–semiconductor interfaces: silicate adlayers on hexagonal (0 0 0 1) and (0 0 0 1) SiC surfaces. Applied Physics Letters, 74 (8), 1084–1086. Onojima, N., Suda, J., Kimoto, T. and Matsunami, H. (2003) Impact of SiC surface control on initial growth mode and crystalline quality of AlN grown by molecular-beam epitaxy. Physica Status Solidi c, 0 (7), 2529–2532. Onojima, N., Suda, J. and Matsunami, H. (2003) High-quality AlN by initial layerby-layer growth on surface-controlled 4H-SiC(0 0 0 1) substrate. Japanese Journal of Applied Physics, Part 2: Letters, 42 (5A), L445–L447. Suda, J., Miura, K., Honaga, M., Nishi, Y., Onojima, N. and Matsunami, H. (2002) Effects of 6H-SiC surface reconstruction on lattice relaxation of AlN buffer layers in molecular-beam epitaxial growth of GaN. Applied Physics Letters, 81 (27), 5141–1543. Onojima, N., Suda, J. and Matsunami, H. (2002) Molecular-beam epitaxial growth of insulating AlN on surfacecontrolled 6H-SiC substrate by HCl gas etching. Applied Physics Letters, 80 (1), 76–78.
j761
j 3 Growth and Growth Methods for Nitride Semiconductors
762
45 Susnitzky, D.W. and Carter, C.B. (1986) Journal of the American Ceramic Society, 69, 217. 46 Pond, R.C., Gowers, J.P. and Joyce, B.A. (1985) Surface Science, 152, 1191. 47 Ruterana, P., Sanchez, A.M. and Nouet, G. (2003) Extended defects in wurtzite GaN layers: atomic structure, formation and interaction mechanisms, in Nitride Semiconductors – Handbook on Materials and Devices (eds P. Ruterana, M. Albrecht and J. Neugebauer), Wiley-VCH Verlag GmbH, Weinheim, Germany. 48 Touloukina, Y.S., Kirby, R.K., Taylor, R.E. and T.Y. Lee (eds) (1977) Thermal Expansion, Thermal Properties of Matter, vol. 13, Plenum, New York. 49 Cui, J., Sun, A., Reshichkov, M., Yun, F., Baski, A. and Morkoc, H. (2000) MRS Internet Journal. The URL for the front page is http://nsr.mij.mrs.org/5/7/. 50 Kim, W., Yeadon, M., Botchkarev, A.E., Mohammad, S.N., Gibson, J.M. and Morkoc, H. (1997) Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 15, 921. 51 Uchida, K., Watanabe, A., Yano, F., Koguchi, M., Tanaka, T. and Minagawa, S. (1996) Journal of Applied Physiology, 79, 3487. 52 Wickenden, A.E., Koleske, D.D., Henry, R.L., Gorman, R.J., Culbertson, J.C. and Twigg, M.E. (1999) The impact of nitridation and nucleation layer process conditions on morphology and electron transport in GaN epitaxial films. Journal of Electronic Materials, 28, 301. 53 Keller,S.,Keller,B.P.,Wu,Y.-F.,Heying,B., Kapolnek, D., Speck, J.S., Mishra, U.K. and DenBaars, S.P. (1996) Influence of sapphire nitridation on properties of gallium nitride grown by metalorganic chemical vapor deposition. Applied Physics Letters, 68, 1525. 54 Kim, K.S., Lim, K.Y. and Lee, H.J. (1999) The effects of nitridation on properties of GaN grown in sapphire substrate by metal-organic chemical vapor
55 56 57 58 59
60
61 62
63 64 65 66
67
68
69 70
71
deposition. Semiconductor Science and Technology, 14, 557. Nause, J., Yun, F. and Morkoc, H., unpublished. Maruska, P.H. (2002) private communication. Slack, G.A. and McNelly, T.F. (1976) Journal of Crystal Growth, 34, 263. Karpinski, J., Jun, J. and Porowski, S. (1984) Journal of Crystal Growth, 66, 1. Porowski, S. and Grzegory, I. 1994 Phase diagram of InN, in Properties of Group III Nitrides (ed. J.H. Edgar), INSPEC, London, pp. 71–88. Thurmond, C.D. and Logan, R.A. (1972) Journal of the Electrochemical Society, 119, 622. Sasaki, T. and Matsuoka, T. (1995) Journal of Applied Physiology, 77, 192. Madar, R., Jacob, G., Hallais, J. and Fruchart, F. (1975) Journal of Crystal Growth, 31, 197. Slack, G.A. and McNelly, T.F. (1977) Journal of Crystal Growth, 42, 560. Slack, G. and McNelly, (1977) Journal of Crystal Growth, 42, 560. Porowski, S. and Grzegory, I. (1997) Journal of Crystal Growth, 178, 174. Leszczynski, M., Grzegory, I. and Bockowski, M. (1993) Journal of Crystal Growth, 126, 601; Porowski, S. (1995) Acta Physica Polonica A, 87, 295. Leszczynski, M., Grzegory, I. and Bockowski, M. (1993) Journal of Crystal Growth, 126, 601. Porowski, S., Grzegory, I. and Jun, J. (1989) High Pressure Chemical Synthesis (eds J. Jurczak and B. Baranowski), Elsevier, Amsterdam, p. 21. Porowski, S. (1995) Acta Physica Polonica A, 87, 295. Grzegory, I., Jun, J., Bockowski, M., Krukowski, S., Wroblewski, M., Lucznik, B. and Porowski, S. (1995) Journal of Physics and Chemistry of Solids, 56, 639. Grzegory, I., Krukowski, S., Leszczynski, M., Perlin, P., Suski, T. and Porowski, S. (2003) High pressure crystallization
References
72
73
74
75
76
77
78
79
80
81
82
of GaN, in Nitride Semiconductors – Handbook on Materials and Devices (eds P. Ruterana, M. Albrecht and J. Neugebauer), Wiley-VCH Verlag GmbH, Weinheim, Germany. Leszczynski, M., Grzgory, I. and Bokowski, M. (1993) X-ray examination of GaN single crystals grown at high hydrostatic pressure. Journal of Crystal Growth, 126, 601. Suski, T., Litwin-Staszewska, E., Perlin, P., Wisniewski, P., Teisseyre, H., Grzegory, I., Bockowski, M., Porowski, S., Saarinen, K. and Nissila, J. (2001) Journal of Crystal Growth, 230, 368. Liliental-Weber, Z. (1999) EMIS Data Review Series, No. 23, INSPEC, The Institution of Electrical Engineers, London, p. 230. Seelmann-Eggebert, M., Weyher, J.L., Obloh, H., Zimmermann, H., Rar, A. and Porowski, S. (2000) Journal of Crystal Growth, 210, 151. Neugebauer, J. and Van de Walle, C.G. (1994) Physical Review B: Condensed Matter, 50, 8067. Boguslawski, P., Briggs, E.L. and Bernholc, J. (1995) Physical Review B: Condensed Matter, 51, 17255. Tiller, W.A., Jackson, K.A., Rutter, J.W. and Chalmers, B. (1953) Acta Metallurgica, 1, 428. Neugebauer, J. (2003) Surface structure and adatom kinetics of group-III nitrides, in Nitride Semiconductors – Handbook on Materials and Devices (eds P. Ruterana, M. Albrecht and J. Neugebauer), Wiley-VCH Verlag GmbH, Weinheim, Germany. Northrup, J. and Romano, L. (1999) Neugebauer, J. Applied Physics Letters, 74, 2319. Northrup, J.E. and Neugebauer, J. (1996) Physical Review B: Condensed Matter, 53 (16), R10477. Filippetti, A., Fiorentini, V., Cappellini, G. and Bosin, A. Physical Review B: Condensed Matter, 59 (12), 8062.
83 Georgakilas, A., Ng, H.M. and Komninou, P. (2003) Plasma assisted molecular beam epitaxy of III–V nitrides, in Nitride Semiconductors – Handbook on Materials and Devices (eds P. Ruterana, M. Albrecht and J. Neugebauer), Wiley-VCH Verlag GmbH, Weinheim, Germany. 84 Bermudez, V.M., Koleske, D.D. and Wickenden, A.E. (1998) Applied Surface Science, 126, 69. 85 King, S.W., Barnak, J.R., Bremser, M.D., Tracy, K.M., Ronning, C., Davis, R.F. and Nemanich, R.J. (1998) Journal of Applied Physiology, 84, 5248. 86 Shalish, I., Shapira, Y., Burnstein, L. and Salzman, J. (2001) Journal of Applied Physiology, 89, 390. 87 Prabhakaran, K., Andersson, T.G. and Nozawa, K. (1996) Applied Physics Letters, 69, 3212. 88 Gupta, V.K., Wamsley, C.C., Koch, M.W. and Wicks, G.W. (1999) Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 17, 1249. 89 Zywietz, T.K., Neugebauer, J. and Scheffler, M. (1999) Applied Physics Letters, 74, 1695. 90 Nakamura, S., Senoh, M., Nagahama, S., Iwasa, N., Yamada, T., Matsushita, T., Kiyoku, I. and Sugimoto, Y. (1996) Japanese Journal of Applied Physics, 35, L217. 91 Nakamura, S., Senoh, M., Nagahama, S., Iwasa, N., Yamada, T., Matsushita, T., Kiyoku, I. and Sugimoto, Y. (1996) Applied Physics Letters, 68, 2105. 92 Lei, T., Ludwig, K.F., Jr and Moustakas, T. (1993) Journal of Applied Physiology, 74, 4430. 93 Powell, R.C., Lee, N.E., Kim, Y.-W. and Green, J. (1993) Journal of Applied Physiology, 73, 189. 94 Tansley, T.L., Goldys, E.M., Godlewski, M., Zhou, B. and Zuo, H.Y. (1997) The contribution of defects to the electrical and optical properties of GaN, in GaN and Related Materials (ed. S. Pearton),
j763
j 3 Growth and Growth Methods for Nitride Semiconductors
764
95
96
97
98
99
100
101
102
103 104
Gordon and Breach, Amsterdam, pp. 233–295. Leszczynski, M. (1999) Common Crystal Structure of the Group III-Nitrides (ed. J.H. Edgar), IEE EMIS Data Review Series, No. 23, INSPEC, London, p. 6. Doverspike, K., Rowland, L.B., Gaskill, D.K. and Freitas, J.A., Jr (1995) Journal of Electronic Materials, 24, 269. Horino, K., Kuramata, A., Domen, K., Soejima, R. and Tanahasji, T. (1996) Growth of (1100) oriented GaN on (1100) 6H-SiC by metalorganic vapor phase epitaxy. International Symposium on Blue laser and Light Emitting Diodes, p.530. Bourret, A., Barski, A., Rouviere, J.L., Renaud, G. and Barbier, A. (1998) Growth of aluminum nitride on (1 1 1) silicon: microstructure and interface structure. Journal of Applied Physiology, 83 (4), 2003. Auner, G.W., Jin, F., Nail, V.M. and Naik, R. (1999) Microstructure of low temperature grown AlN thin films on Si (1 1 1). Journal of Applied Physiology, 85 (11), 7879. Ohsato, H. and Razeghi, M. (1977) Growth models of GaN thin films based on crystal chemistry: hexagonal and cubic GaN on Si substrates. Proceedings of SPIE, 2999, 288. Lebedev, V., Jinschek, J., Kaiser, U., Schoroter, B., Richter, W. and KrauÔlich, J. (2000) Epitaxial relationship in the AlN/Si(0 0 1) heterosystem. Applied Physics Letters, 76 (15), 2029. Barghout, K. and Chaudhuri, J. (2004) Calculation of residual thermal stress in GaN epitaxial layers grown on different substrates. Journal of Materials Science, 39 (18), 5817–5823. Chu, T.L., Ing, D.W. and Noreika, A.J. (1968) Electrochemical Technology, 6, 56. Trassoudaine, A., Cadoret, R. and Aujol, E. (2003) The growth of gallium nitride by hydride vapour phase epitaxy (HVPE), in Nitride Semiconductors – Handbook on Materials and Devices (eds P. Ruterana,
105
106
107 108 109
110 111 112
113
114
115
116 117
118
119
120
121
M. Albrecht and J. Neugebauer), WileyVCH Verlag GmbH, Weinheim, Germany. Wickenden, D.K., Faulkner, K.R. and Brander, R.W. (1971) Journal of Crystal Growth, 9, 158. Ilegems, M. (1972) Vapour epitaxy of gallium nitride. Journal of Crystal Growth, 13–14, 360–364. Shintani, A. and Minagawa, S. (1974) Journal of Crystal Growth, 22, 1. Sano, M. and Aoki, M. (1976) Japanese Journal of Applied Physics, 15, 1943. Madar, R., Michel, D., Jacob, G. and Boulou, M. (1977) Journal of Crystal Growth, 40, 239. Jacob, G., Boulou, M. and Furtado, M. (1977) Journal of Crystal Growth, 42, 136. Seifert, W., Fitzl, G. and Butter, E. (1981) Journal of Crystal Growth, 52, 257. Maruska, H.P., Rhines, W.C. and Stevenson, D.A. (1972) Materials Research Bulletin, 7, 777. Wickenden, D.K., Faulkner, K.R., Brander, R.W. and Isherwood, B.J. (1971) Journal of Crystal Growth, 9, 158. Burns, G., Dacol, F., Marinace, J.C., Scott, B.A. and Burstein, E. (1973) Applied Physics Letters, 22, 356. Seifert, W. and Tempel, A. (1974) Physica Status Solidi a: Applied Research, 23, K39. Monemar, B. (1974) Physical Review B: Condensed Matter, 10, 676. Matsumoto, T., Sano, M. and Aoki, M. (1974) Japanese Journal of Applied Physics, 13, 373. Suleimanov, Y.M., Pichugin, I.G. and Marasina, L.A. (1974) Soviet Physics: Semiconductors, 8, 537. Ejder, E. and Fagerstr€om, P.O. (1975) Journal of Physics and Chemistry of Solids, 36, 289. Crouch, R.K., Debnam, W.J. and Fripp, A.L. (1978) Journal of Materials Science, 13, 2358. Fremunt, R., Cerny, P., Kohout, J., Rosicka, V. and B€ urger, A. (1981) Crystal Research and Technology, 16, 1257.
References 122 Adonin, A.S., Evmenenko, V.A., Mikhailov, L.N. and Ryabtsev, N.G. (1982) Inorganic Materials, 17, 1187. 123 Kuznetsov, A.V., Galstyan, V.G., Muratova, V.I. and Chaplygin, G.V. (1983) Soviet Microelectronics, 11, 214. 124 Naniwae, K., Itoh, S., Amano, H., Itoh, K., Hiramatsu, K. and Akasaki, I. (1990) Journal of Crystal Growth, 99, 381. 125 Yim, W.M., Stofko, E.J., Zanzucchi, P.J., Pankove, J.I., Ettenberg, M. and Gilbert, S.L. (1973) Journal of Applied Physiology, 44, 292. 126 Chu, T.L. and Kelm, R.W., Jr (1975) Journal of the Electrochemical Society, 122, 995. 127 Detchprohm, T., Hiramatsu, K., Amano, H. and Akasaki, I. (1992) Applied Physics Letters, 61, 2688. 128 Hiramatsu, K., Detchprohm, T. and Akasaki, I. (1993) Japanese Journal of Applied Physics, 32, 1528. 129 Molnar, R.J., Nichols, K.B., Maki, P., Brown, E.R. and Melngailis, I. (1995) Proceedings of the Material Research Society Symposium, Boston, p.479. 130 Usui, A., Sunakawa, H., Sakai, A. and Yamaguchi, A. (1997) Japanese Journal of Applied Physics, 36, L899. 131 Shibata, T., Sone, H., Yahashi, K., Yamaguchi, M., Hiramatsu, K., Sawaki, N. and Itoh, N. (1998) Journal of Crystal Growth, 189–190, 67. 132 Nataf, G., Beaumont, B., Bouille, A., Haffouz, S., Vaille, M. and Gibart, P. (1998) Journal of Crystal Growth, 192, 73. 133 Kim, S.T., Lee, Y.J., Moon, D.C., Hong, C.H. and Yoo, T.K. (1998) Journal of Crystal Growth, 194, 37. 134 Melnik, Y., Nikolaev, A., Nikitina, L., Vassilevski, K. and Dimitriev, V. (1998) Proceedings of the Material Research Society Symposium, Boston, p.269. 135 Park, S.S., Park, I.W. and Choh, S.H. (2000) Proceedings of the IWN2000 Conference, Nagoya, p.60. 136 Tavernier, P.R., Etzkom, E.V., Wang, Y. and Ckarke, D.R. (2000) Applied Physics Letters, 77, 1804.
137 Kelly, M.K., Vaudo, R.P., Phanse, V.M., G€ orgens, L., Ambacher, O. and Stutzmann, M. (1999) Japanese Journal of Applied Physics, 38, L217. 138 Ban, V.S. (1972) Journal of the Electrochemical Society, 119, 761. 139 Cadoret, R. and Gil-Lafon, E. (1997) Journal de Physique I, 7, 889. 140 Cadoret, R. (1999) Journal of Crystal Growth, 205, 123. 141 Barin, I. (1993) Thermodynamical Data of Pure Substances, Wiley-VCH Verlag GmbH, Weinheim, Germany. 142 Fujieda, S., Mizuta, M. and Matsumoto, Y. (1987) Japanese Journal of Applied Physics, 26, 2067. 143 Okumura, H., Misawa, S. and Yoshida, S. (1991) Applied Physics Letters, 59, 1058. 144 Hersee, S.D., Ramer, J.C. and Malloy, K.J. (1997) MRS Bulletin, 22, 45. 145 Molnar, R.J., G€otz, W., Romano, L.T. and Johnson, N.M. (1997) Journal of Crystal Growth, 178, 147. 146 Trassoudaine, A., Aujol, E., Disseix, P., Castelluci, D. and Cadoret, R. (1999) Physica Status Solidi a: Applied Research, 176, 425. 147 Arnaudov, B., Paskova, T., Goldys, E.M., Yakimova, R., Evtimova, S., Ivanov, I.G., Henry, A. and Monemar, B. (1999) Journal of Applied Physiology, 85, 7888. 148 Gu, S., Zhang, R., Shi, Y., Zheng, Y., Zhang, L., Dwikusuma, F. and Kuech, T.F. (2001) Journal of Crystal Growth, 231, 342. 149 Shaw, D.W. (1968) Journal of the Electrochemical Society, 115, 405. 150 Shaw, D.W. (1970) Journal of the Electrochemical Society, 117, 683. 151 Cadoret, R. (1980) Current Topics in Materials Science, vol. 5 (ed. E. Kaldis), North-Holland, Amsterdam. 152 Theeten, J.B., Hollan, L. and Cadoret, R. (1977) Crystal Growth and Materials (eds E. Kaldis and H.J. Scheel), NorthHolland, Amsterdam. 153 Hollan, L. and Schiller, C. (1972) Journal of Crystal Growth, 13–14, 319.
j765
j 3 Growth and Growth Methods for Nitride Semiconductors
766
154 Hollan, L., Durand, J.M. and Cadoret, R. (1977) Journal of the Electrochemical Society, 124, 135. 155 Gentner, J.L., Bernard, C. and Cadoret, R. (1982) Journal of Crystal Growth, 56, 332. 156 Cadoret, R. (1999) Journal of Crystal Growth, 205, 123. 157 Korec, J. and Heyen, M. (1982) Journal of Crystal Growth, 60, 297. 158 Jurgenssen, H., Korec, J., Heyen, M. and Balk, P. (1984) Journal of Crystal Growth, 66, 73. 159 Molnar, R.J. (1999) Hydride vapor phase eptitaxial growth of III–V nitrides, in Gallium Nitride (GaN) II, Semiconductors and Semimetals, vol. 57 (eds J.I. Pankove and T.D. Moustakas), Academic Press, pp.1–28. 160 Wetzel, C., Volm, D., Meyer, B.K., Pressel, K., Nilsson, S., Mokhov, E.N. and Baranov, P.G. (1994) Applied Physics Letters, 65, 1033. 161 Fischer, S, Wetzel, C., Hansen, W.L., Bourret-Chourchesne, E.D., Meyer, B.K. and Haller, E.E. (1996) Applied Physics Letters, 69, 2716. 162 Vodakov, Y.A., Karklina, M.I., Mokhov, E.N. and Roenkov, A.D. (1980) Inorganic Materials, 17, 537. 163 For gaseous GaCl approach, see Baranov, B., D€aweritz, L., Gutan, V.B., Jungk, G., Neumann,H.and Raidt,H.(1978)Physica Status Solidi a: Applied Research, 49, 629. 164 Liu, S.S. and Stevenson, D.A. (1978) Journal of the Electrochemical Society, 125, 1161. 165 Butter, E., Fitzl, G., Hirsch, D., Leonhardt, G., Seifert, W. and Preschel, G. (1979) Thin Solid Films, 59, 25. 166 Seifert, W., Franzheld, R., Butter, E., Subotta, H. and Riede, V. (1983) Crystal Research and Technology, 18, 383. 167 Neumann, H., Seifert, W., Staudte, M. and Zehe, A. (1974) Kristall und Technik, 9, K69. 168 For gaseous GaCl2, GaCl3 and Ga (C2H5)2Cl sources, see (a) Troost, D., Baier, H.U., Berger, A. and M€onch, W.
169
170
171
172
173 174
175
(1991) Surface Science, 242, 324; (b) Born, P.J. and Robertson, D.S. (1980) Journal of Materials Science, 15, 3003; (c) Nickl, J.J., Jus, W. and Bertinger, R. (1975) Materials Research Bulletin, 10, 1097. For gaseous GaCl2NH3 sources, see Marasina, L.A., Pikhtin, A.N., Pichugin, I.G. and Solomonov, A.V. (1976) Physica Status Solidi a: Applied Research, 38, 753. For gaseous AlCl2 sources, see (a) Callaghan, M.P., Patterson, E., Richards, B.P. and Wallace, C.A. (1974) Journal of Crystal Growth, 22, 85; (b) Akasaki, I. and Hashimoto, H. (1967) Solid State Communications, 5, 851; (c) Chu, T.L., Ing, D.W. and Noreika, A.J. (1967) Solid State Electronics, 10, 1023; (d) Irene, E.A., Silvestri, V.J. and Woolhouse, G.R. (1975) Journal of Electronic Materials, 4, 409; (e) Bauer, J., Biste, L. and Bolze, D. (1977) Physica Status Solidi a: Applied Research, 39, 173. For gaseous AlBr3, InCl3, and GaBr3 sources, see (a) Someno, Y., Sasaki, M. and Hirai, T. (1991) Japanese Journal of Applied Physics, 30, 1792; (b) Marasina, L.A., Pichugin, I.G. and Tlaczala, M. (1977) Kristall und Technik, 12, 541; (c) Chu, T.L. (1971) Journal of the Electrochemical Society, 118, 1200; (d) Morimoto, Y., Uchiho, K. and Ushio, S. (1973) Journal of the Electrochemical Society, 120, 1783;(e) Aleksandrov, S.E., Kovalgin, A.Y. and Krasovitskii, D.M. (1994) Russian Journal of Applied Chemistry, 67, 663. (a) Pastrnak, J. and Souckova, L. (1963) Physica Status Solidi, 11, K71; (b) Pastrnak, J. and Roskovcova, L. (1965) Physica Status Solidi, 9, K73. Dryburgh, P.M. (1989) Journal of Crystal Growth, 94, 23. Igarashi, O. and Okada, Y. (1985) Japanese Journal of Applied Physics, 24, L792;Igarashi, O. (1988) Japanese Journal of Applied Physics, 27, 790. Molnar, R.J., Goetz, W., Romano, L.T. and Johnson, N.M. (1997) Growth of
References
176
177
178
179
180
181
182
183
184
185
186 187
gallium nitride by hydride vapor-phase epitaxy. Journal of Crystal Growth, 178 (1–2), 147–156. Manasevit, H.M., Erdman, F.M. and Simpson, W.I. (1971) Journal of the Electrochemical Society, 118, 1864. Amano, H., Sawaki, N., Akasaki, I. and Toyoda, Y. (1986) Applied Physics Letters, 48, 353. Yuan, C., Salagaj, T., Gurary, A., Zavadski, P., Chen, C.S., Kroll, W., Stall, R.A., Li, Y., Schurman, M., Hwang, C.Y., Mayo, W.E., Lu, Y., Pearton, S.J., Krishnankutty, S. and Kolbas, R.M. (1995) Journal of the Electrochemical Society, 142, L163. Nakamura, S., Mukai, T. and Senoh, M. (1994) Journal of Applied Physics, 76, 8189. Nakamura, S., Senoh, M., Magahama, S., Iwasha, N., Yamada, T., Matsushita, T., Kioku, H. and Sugimoto, Y. (1996) Japanese Journal of Applied Physics, L74, 1998. Khan, M.A., Kuznia, J.N., Bhattarai, A.R. and Olson, D.T. (1993) Applied Physics Letters, 62, 1786. Binet, F., Duboz, J.Y., Rosencer, E., Scholz, F. and Harle, V. (1996) Applied Physics Letters, 69, 1002. Nakamura, S., Harada, Y. and Seno, M. (1991) Applied Physics Letters, 58, 2021. See, for example, Wickenden, D.K., Miragliotta, J.A., Bryden, W.A. and Kistnmacher, T.J. (1994) Journal of Applied Physiology, 75, 7585; Knights, J.C. and Lujan, R.A. (1978) Journal of Applied Physiology, 49, 1291; Wickenden, A.E., Wickenden, D.K. and Kistenmacher, T.J. (1994) Journal of Applied Physiology, 75, 5367. Matloubian, M. and Gershenzon, M. (1985) Journal of Electronic Materials, 40, 633–644. Wakahara, A. and Yoshida, A. (1989) Applied Physics Letters, 54, 709. Sheng, T.Y., Yu, Z.Q. and Collins, G.J. (1988) Applied Physics Letters, 52, 576.
188 Gaskill, D.K., Bottka, N. and Lin, M.C. (1986) Applied Physics Letters, 48, 1449. 189 Mizuta, M., Fujieda, S., Jitsukawa, T. and Matsumoto, Y. (1987) Proceedings of the International Symposium on GaAs and Related Compounds, 1986, Las Vegas, NV, Adam Hilger, Bristol. 190 Miyoshi, S., Onabe, K., Ohkouchi, N., Yaguchi, H. and Ito, R. (1992) Journal of Crystal Growth, 124, 439. 191 Ishii, M., Minami, T., Miyata, T., Karaki, S. and Takata, S. (1996) Institute of Physics Conference Series, 142, 899. 192 Manasevit, H.M., Erdmann, F.M. and Simpson, W.I. (1971) Journal of the Electrochemical Society, 118, 1864. 193 Pizzarello, F.A. and Coker, J.E. (1975) Journal of Electronic Materials, 4, 25. 194 OClock, G.D., Jr and Duffy, M.T. (1973) Applied Physics Letters, 23, 55. 195 Liu, J.K., Lakin, K.M. and Wang, K.L. (1975) Journal of Applied Physiology, 46, 3703. 196 Morita, M., Uesugi, N., Isogai, S., Tsubouchi, K. and Mikoshiba, N. (1981) Japanese Journal of Applied Physics, 20, 17. 197 Khan, M.A., Skogman, R.A., Schulze, R.G. and Gershenzon, M. (1983) Applied Physics Letters, 42, 430. 198 Hashimoto, M., Amano, H., Sawaki, N. and Akasaki, I. (1984) Journal of Crystal Growth, 68, 163. 199 Matloubian, M. and Gershenzon, M. (1985) Journal of Electronic Materials, 14, 633. 200 Sasaki, T. and Matsuoka, T. (1988) Journal of Applied Physiology, 64, 4531. 201 Nagatomo, T., Kuboyama, T., Minamino, H. and Omoto, O. (1989) Japanese Journal of Applied Physics, 28, L1334. 202 Khan, M.A., Kuznia, J.N., Van Hove, J.M., Olsen, D.T., Krishnankutty, S. and Kolbas, R.M. (1991) Applied Physics Letters, 58, 526. 203 Nakamura, S., Harada, Y. and Seno, M. (1991) Applied Physics Letters, 58, 2021. 204 See, for example, Wickenden, D.K., Miragliotta, J.A., Bryden, W.A. and
j767
j 3 Growth and Growth Methods for Nitride Semiconductors
768
205
206 207
208 209 210
211
212 213
214
215
216
217
218
Kistnmachr, T.J. (1994) Journal of Applied Physiology, 75, 7585. Knights, J.C. and Lujan, R.A. (1978) Journal of Applied Physiology, 49, 1291; Wickenden, A.E., Wickenden, D.K. and Kistenmacher, T.J. (1994) Journal of Applied Physiology, 75, 5367. Shiloh, M. and Gutman, J. (1973) Materials Research Bulletin, 8, 711. Lappa, R., Glowacki, G. and Galkowski, S. (1976) Thin Solid Films, 32, 73. Sheng, T.Y., Yu, Z.Q. and Collins, G.J. (1988) Applied Physics Letters, 52, 576. Wakahara, A. and Yoshida, A. (1989) Applied Physics Letters, 54, 709. Eremin, E.N., Nekrasov, L.I., Rubtsova, E.A., Belova, V.M., Ivanter, V.L., Zakharov, L.N. and Petukhov, L.N. (1982) Russian Journal of Physical Chemistry, 56, 788. Matloubian, M. and Gershenzon, M. (1985) Journal of Electronic Materials, 40, 633–644. Dupuie, J.L. and Gulari, E. (1991) Applied Physics Letters, 59, 549. Liu, H.H, Bertolet, D.C. and Rogers, J.W. (1994) The surface chemistry of aluminum nitride MOCVD on alumina using trimethylaluminum and ammonia precursors. Surface Science, 320, 145. Jones, A.C., Whitehouse, C.R. and Roberts, J.S. (1995) Chemical Vapor Deposition, 1, 65. Jones, A.C., Auld, J., Rushworth, S.A., Houlton, D.J. and Critchlow, G.W. (1994) Journal of Materials Chemistry, 4, 1591, Jones, A.C., Auld, J., Rushworth, S.A., Williams, E.W., Haycock, P.W., Tang, C.C. and Critchlow, G.W. (1994) Advanced Materials, 6, 6. Takeda, F., Mori, T. and Takahashi, T. (1981) Japanese Journal of Applied Physiology, 20, L169. Lee, H.C., Lee, K.Y., Yong, Y.J., Lee, J.Y. and Kim, H. (1995) Thin Solid Films, 271, 50.
219 Yoshida, M., Watanabe, H. and Uesugi, F. (1985) Journal of the Electrochemical Society, 132, 677. 220 Ambacher, O. (1998) Journal of Physics D: Applied Physics, 31, 2653. 221 Koleske, D.D., Wickenden, A.E., Henry, R.L., DeSisto, W.J. and Gorman, R.J. (1998) A growth model for GaN with comparison to structural, optical and electrical properties. Journal of Applied Physiology, 84, 1998–2010. 222 Grandjean, N., Massies, J., Semond, F., Karpov, S.Yu. and Talalaev, R.A. (1999) GaN evaporation in molecular-beam epitaxy environment. Applied Physics Letters, 74 (13), 1854–1856. 223 Koleske, D.D., Wickenden, A.E., Henry, R.L., Culbertson, J.C. and Twigg, M.E. (2001) GaN decomposition in H2 and N2 at OMVPE temperatures and pressures. Journal of Crystal Growth, 223, 466–483. 224 LVov, B.V. (2000) Kinetics and mechanism of thermal decomposition of GaN. Thermochimica Acta, 360, 85–91. 225 Munir, Z.A. and Searcy, A.W. (1965) Activation energy for the sublimation of gallium nitride. Journal of Chemical Physics, 42, 4223–4228. 226 Liu, S.S. and Stevenson, D.A. (1978) Journal of the Electrochemical Society, 125, 1161. 227 Koblm€ uller, G., Averbeck, R., Tiechert, H. and Pongratz, P. (2004) Direct observation of different equilibrium Ga adlayer coverages and their desorption kinetics on GaN (0 0 0 1) and (0 0 0 1) surfaces. Physical Review B: Condensed Matter, 69, 035325. 228 Feynman, R.P., Leighton, R.B. and Sands, M. (1966) The Feynman Lectures on Physics, vol. 1, Addison-Wesley, Reading, MA, pp. 41–49. 229 Brandt, O., Yang, H. and Ploog, K.H. (1996) Physical Review B: Condensed Matter, 54, 4432. 230 The H2N–NH2 bond energy of 3 eV (Weast, R.C. (1974) CRC Handbook of Chemistry and Physics, 55th edn, CRC
References
231
232
233 234
235
236
237
238 239
240
241
Press, Cleveland) is less than the estimated Ga–N bond energy (4.2 eV). Weaker N–N bonds (i.e., <3 eV) are formed when each N is bonded to more electronegative elements than N (e.g., O or F). Karpov, S.Yu. and Maiorov, M.A. (1997) Surface Science, 393, 108; Karpov, S.Yu. and Maiorov, M.A. (1997) Technical Physics Letters, 23, 38. Manasevit, H.M., Erdmann, F.M. and Simpson, W.I. (1971) Journal of the Electrochemical Society, 118, 1864; See also Pizzarello, F.A. and Coker, J.E. (1975) Journal of Electronic Materials, 4, 25. OClock, G.D., Jr and Duffy, M.T. (1973) Applied Physics Letters, 23, 55. Liu, J.K., Lakin, K.M. and Wang, K.L. (1975) Journal of Applied Physiology, 46, 3703. Khan, M.A., Skogman, R.A., Schulze, R.G. and Gershenzon, M. (1983) Applied Physics Letters, 42, 430. Hashimoto, M., Amano, H., Sawaki, N. and Akasaki, I. (1984) Journal of Crystal Growth, 68, 163. Matloubian, M. and Gershenzon, M. (1985) Journal of Electronic Materials, 14, 633; See also Sasaki, T. and Matsuoka, T. (1988) Journal of Applied Physiology, 64, 4531; Nagatomo, T., Kuboyama, T., Minamino, H. and Omoto, O. (1989) Japanese Journal of Applied Physics, 28, L1334; Khan, M.A., Kuznia, J.N., Van Hove, J.M., Olsen, D.T., Krishnankutty, S. and Kolbas, R.M. (1991) Applied Physics Letters, 58, 526. Nakamura, S., Harada, Y. and Seno, M. (1991) Applied Physics Letters, 58, 2021. Shchukin, V.A., Ledentsov, N.N. and Bimberg, D. (2004) Epitaxy of Nanostructures, Springer, Berlin. Ledentsov, N.N. (1999) Growth Processes and Surface Phase Equilibria in Molecular Beam Epitaxy, Springer, Berlin. Herman, M.A. and Sitter, H. (1989) Molecular Beam Epitaxy: Fundamental and Current Status, Springer, Berlin.
242 Koleske, D.D., Wiekenden, A.E., Henry, R.L., DeSisto, W.J. and Gorman, R.J. (1998) Journal of Applied Physics, 84 (4), 1998–2010. 243 Yoshida, S., Misawa, S. and Itoh, A. (1975) Applied Physics Letters, 26, 461. 244 Winsztal, S., Wauk, B., Majewska-Minor, H. and Niemyski, T. (1976) Thin Solid Films, 32, 251. 245 Elliott, K.R. and Grant, R.W. (1984) Rockwell Project Final Report MRDC41116.2FR. 246 Baier, H.U. and M€onch, W. (1990) Journal of Applied Physiology, 68, 586. 247 Yoshida, S., Misawa, S. and Gonda, S. (1983) Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 1, 250. 248 Oura, K., Lifshits, V.G., Saranin, A.A., Zotov, A.V. and Katayama, M. (2003) Surface Science, Springer, Berlin. 249 Adelmann, C., Brault, J., Mula, G., Daudin, B., Lymperakis, L. and Neugebauer, J. (2003) Physical Review B: Condensed Matter, 67, 165419. 250 Guha, S., Bojarczuk, N.A. and (1996) Kisker, D.W. Applied Physics Letters, 69, 2879. 251 Brandt, O., Sun, Y.J., Daweritz, L. and Ploog, K.H. (2004) Physical Review B: Condensed Matter, 69, 165326. 252 Xue, Q., Xue, O.K., Bakhtizin, R.Z., Hasegawa, Y., Tsong, I.S.T., Sakurai, T. and Ohno, T. (1999) Physical Review B: Condensed Matter, 59, 12604. 253 Koblm€ uller, G., Averbeck, R., Tiechert, H. and Pongratz, P. (2004) Physical Review B: Condensed Matter, 69, 035325. 254 Adelmann, C., Brault, J., Jalabert, D., Gentile, P., Mariette, H., Mula, G. and Daudin, B. (2002) Journal of Applied Physics, 91, 9638. 255 Zheng, L.X., Xie, M.H. and Tong, S.Y. (2000) Physical Review B: Condensed Matter, 61, 4890. 256 Zywietz, T., Neugebauer, J. and Scheffler, M. (1998) Applied Physics Letters, 73, 487.
j769
j 3 Growth and Growth Methods for Nitride Semiconductors
770
257 Myoung, J.M., Gluschenkov, O., Kim, K. and Kim, S. (1999) Journal of Vacuum Science & Technology A: Vacuum Surfaces and Films, 17, 3019. 258 Neugebauer, J., Zywietz, T.K., Scheffler, M., Northrup, J.E., Chen, H. and Feenstra, R.M. (2003) Adatom kintecs on and below the surface: the existence of new diffusion model. Physical Review Letters, 90, 056101. 259 Myoung, J.M., Gluschenkov, O., Kim, K. and Kim, S. (1999) Journal of Vacuum Science & Technology A: Vacuum Surfaces and Films, 17, 3019. 260 Iliopoulos, E. and Moustakas, T.D. (2002) Applied Physics Letters, 81, 295. 261 Harrison, W.A. (1980) Electronic Structure and the Properties of Solid, Dover, New York. 262 Phillips, J.C. (1973) Bonds and Bands in Semiconductors, Academic Press, New York. 263 Peng, L.-M., Dudarev, S.L. and Whelan, M.J. (2004) High-Energy Electron Diffraction and Microscopy, Oxford University Press. 264 Ichimiya, A. and Cohen, P.I. (2004) Reflection High-Energy Electron Diffraction, Cambridge University Press. 265 Braun, W. (1999) Applied RHEED, Springer. 266 Woodruff, D.P. and Delchar, T.A. (1994) Modern Techniques of Surface Science, Cambridge University Press. 267 OConnor, D.J., Sexton, B.A. and Smart, R.St.C. (2003) Surface Analysis Methods in Materials Science, Springer, Berlin. 268 See, for example, Ashcroft, N.W. and Mermin, N.D. (1976) Solid State Physics, Saunders, Chapter 6. 269 Kittel, C. (1986) Introduction to Solid State Physics, John Wiley & Sons, Inc. 270 Mahan, J.E., Geib, K.M., Robinson, G.Y. and Long, R.G. (1990) A review of the geometrical fundamentals of reflection high-energy electron diffraction with application to silicon surfaces. Journal of Vacuum Science & Technology A: Vacuum Surfaces and Films, 8 (5), 3692.
271 Brundle, R.C. Evans, C.A., Jr and Wilson, S. (1992) Encyclopedia of Materials Characterization Butterworth–Heinemann. 272 Wang, Z.L. (1996) Reflection Electron Microscopy and Spectroscopy for Surface Analysis, Cambridge Press. 273 Bauer, E. (1969) Reflection electron diffraction (RED), in Techniques for the Direct Observation of Structure and Imperfections. Part 2. Techniques of Metals Research, vol. II (ed. R.F. Bunshah), Interscience, New York, p. 501, 274 Takayanagi, K., Tanishiro, Y., Takahashi, M. and Takahashi, S. (1985) Structural analysis of Si(1 1 1)7 7 by UV-transmission electron diffraction microscopy. Journal of Vacuum Science & Technology A: Vacuum Surfaces and Films, 3 (3), 1502–1506. € Botchkarev, A. and 275 Kim, W., Aktas, O., Morkoc, H. (1996) Journal of Applied Physiology, 79 (10), 7657–7666. 276 Gotoh, H., Suga, T., Suzuki, H. and Kimata, M. (1981) Japanese Journal of Applied Physics, 20, L545. 277 Boer, K.W. (1990) Survey of Semiconductor Physics, vol. 1, Van Nostrand Reinhold, New York. 278 Botchkarev, A., Salvador, A. and Sverdlov, B., Myoung, J. and Morkoc, H. (1995) Journal of Applied Physiology, 77, 4455. 279 Wright, A.N. and Winkler, C.A. (1968) Active Nitrogen, Academic Press, New York. 280 Slovetskii, D.I. (1980) Mechanism of Chemical Reactions in Nonequilibrium Plasma, Nauka, Moscow. 281 Jordan, D.C., Tsong, I.S.T., Smith, D.J., Wilkens, B.J. and Doak, R.B. (2000) Applied Physics Letters, 77, 3030. 282 Myers, T.H., Millecchia, M.R., Ptak, A.J., Ziemer, K.S. and Stinespring, C.D. (1999) Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 17, 1232.
References 283 Foxon, C.T., Cheng, T.S., Hooper, S.E., Jenkins, L.C., Orton, J.W., Lacklison, D.E., Novikov, S.V., Popova, T.B. and Tretyakov, V.V. (1996) Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 14, 2346. 284 Boenig, H.V. (1982) Plasma Science and Technology, Cornell University Press, Ithaca. 285 Ohkawa, K., Ueno, A. and Mitsuyu, T. (1992) Journal of Crystal Growth, 117, 375. 286 Carins, R.A. (1985) Plasma Physics, Blackie, Glasgow. 287 Morkoc, H. (1998) Wurtzite GaN based heterostructures by molecular beam epitaxy. IEEE Journal of Selected Topics in Quantum Electronics, 4 (3), 537–549. 288 Johnson, M.A.L., Yu, Z., Boney, C., Hughes, W.C., Cook, J.W., Jr, Schetzina, J.F., Zhao, H., Skromme, B.J. and Edmond, J.A. (1997) Mater Research Society Symposium Proceedings, 449, 215. 289 Blant, A.V., Hughes, O.H., Cheng, T.S., Novikov, S.V. and Foxon, C.T. (2000) Plasma Sources Science & Technology, 9, 12. 290 Komninou, Ph., Kehagias, Th., Karakostas, Th., Nouet, G., Ruterana, P., Amimer, K., Mikroulisand, S. and Georgakilas, A. (2001) Mater Research Society Symposium Proceedings, 639, G3.47.1. 291 Hoke, W.E., Lemonias, P.J. and Weir, D.G. (1991) Journal of Crystal Growth, 111, 1024. 292 Van Hove, J.M., Cosmini, G.J., Nelson, E., Wowchak, A.M. and Chow, P.P. (1995) Journal of Crystal Growth, 150, 908. 293 Vaudo, R.P., Kook, J.W., Jr and Schetzina, J.F. (1994) Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 12, 1232. 294 Hughes, W.C., Rowland, W.H., Jr, Johnson, M.A.L., Fujita, S., Cook, J.W., Jr, Schetzina, J.F., Ren, J. and Edmond, J.A. (1995) Journal of Vacuum Science &
295 296
297
298
299
300
301 302 303
304
305
306
307
Technology B: Microelectronics and Nanometer Structures, 13, 1571. Cho, S.-H., Okumura, H. and Akimoto, K. (2000) Applied Physics Letters, 76, 3861. Krtschil, A., Witte, H., Lisker, M., Christen, J., Birkle, U., Einfeldt, S. and Hommel, D. (1999) Applied Physics Letters, 74, 2032. Kichner, V., Heinke, H., Birkle, U., Einfeldt, S., Hommel, D., Selke, H. and Ryder, P.L. (1998) Physical Review B: Condensed Matter, 8, 15749. Ptak, A.J., Millecchia, M.R., Myers, T.H., Ziemer, K.S. and Stinespring, C.D. (1999) Applied Physics Letters, 74, 3836. Hughes, W.C., Rowaland, W.H., Jr, Johnson, M.A.L., Fujita, S., Cook, J.W., Jr, Schetzina, J.F., Ren, J. and Edmond, J.A. (1995) Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 13, 1571. Ptak, A.J., Ziemer, K.S., Millecchia, M.R., Stinespring, C.D. and Myers, T.H. (1999) MRS Internet Journal of Nitride Semiconductor Research, 4S1, G3.10. Newman, N. (1997) Journal of Crystal Growth, 178, 102. Keudell, A. (2000) Plasma Sources Science & Technology, 9, 455. Lifshitz, Y., Kasi, S.R., Rabalais, J.W. and Ekstein, W. (1990) Physical Review B: Condensed Matter, 41, 10468. Myers, T.H., Millecchia, M.R., Ptak, A.J., Ziemer, K.S. and Stinespring, C.D. (1999) Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 17, 1654. Davis, R.F., Paisley, M.J., Sitar, Z., Kester, D.J., Ailey, K.S., Linthicum, K., Rowland, L.B., Tanaka, S. and Kern, R.S. (1997) Journal of Crystal Growth, 178, 87. Headrick, R.L., Kycia, S., Woll, A.R., Brock, J.D. and Ramana Murty, M.V. (1998) Physical Review B: Condensed Matter, 58, 4818. Yoshida, S., Misawa, S. and Itoh, A. (1975) Applied Physics Letters, 26, 461.
j771
j 3 Growth and Growth Methods for Nitride Semiconductors
772
308 Winsztal, S., Wauk, B., Majewska-Minor, H., Niemyski, T. (1976) Thin Solid Films, 32, 251. € Fan, Z., Botchkarov, A., 309 Aktas, O., Mohammad, S.N., Roth, M., Jenkins, T., Kehias, L. and Morkoc, H. (1997) IEEE Electron Device Letters, 18, 293. 310 Mohammad, S.N., Salvador, A. and Morkoc, H. (1995) Emerging GaN based devices. Proceedings of IEEE, 83, 1306. 311 Morkoc, H. (1998) Beyond SiC! III–V nitride based heterostructures and devices, in SiC Materials and Devices, vol. 52 (ed. Y.S. Park), Academic Press, San Diego, FL, pp. 307–394. 312 Smith, M., Lin, J.Y., Jiang, H.X., Salvador, A., Botchkarev, A., Kim, W. and Morkoc, H. (1996) Applied Physics Letters, 69, 2453. 313 Singh, R., Doppalaudi, D. and Moustakas, T.D. (1996) Applied Physics Letters, 69, 2388. 314 Schmidt, T.J., Yang, X.H., Shan, W., Song, J.J., Salvador, A., Kim, W., Aktas, € Botchkarev, A. and Morkoc, H. O., (1996) Applied Physics Letters, 68, 1820. 315 Hooper, S.E., Kauer, M., Bousquet, V., Johnson, K., Barnes, J.M. and Heffernan, J. (2004) InGaN multiple quantum well laser diodes grown by molecular beam epitaxy. IEE Electronics Letters, 40 (1), 33–34. 316 Widmann, F., Feuillet, G., Daudin, B. and Rouviere, J.L. (1999) Journal of Applied Physiology, 85 (3), 1550–1555, and references therein. 317 Heying, B., Smorchkova, I., Poblenz, C., Elsass, C., Fini, P., DenBaars, S., Mishra, U. and Speck, J.S. (2000) Applied Physics Letters, 77, 2885. 318 Heying, B., Averbeck, R., Chen, L.F., Haus, E., Riechert, H. and Speck, J.S. (2000) Journal of Applied Physiology, 88, 1855. € Botchkarev, A.E., 319 Kim, W., Aktas, O., Salvador, A., Mohammad, S.N. and Morkoc, H. (1994) Journal of Applied Physiology, 79, 7657.
320 Mesrine, M., Grandjean, N. and Massies, J. (1998) Applied Physics Letters, 72, 350. 321 Torres, V. (1998) Ph.D. Dissertation, North Carolina State University. 322 Held, R., Crawford, D.E., Johnston, A.M., Dabiran, A.M. and Cohen, P.I. (1998) Surface Review and Letters, 5, 913. 323 Held, R., Crawford, D.E., Johnston, A.M., Dabiran, A.M. and Cohen, P.I. (1997) Journal of Electronic Materials, 26, 272. 324 Held, R., Nowak, G., Ishaug, B.E., Seutter, S.M., Parkhomovsky, A., Dabiran, A.M., Cohen, P.I., Grzegory, I. and Porowski, S. (1999) Journal of Applied Physiology, 85, 7697. 325 Held, R., Ishaug, B.E., Parkhomovsky, A., Dabiran, A.M. and Cohen, P.I. Journal of Applied Physiology, (2000) 87, 1219. 326 McGinnis, A.J., Thomson, D., Davis, R.F., Chen, E., Michel, A. and Lamb, H.H. (2001) Surface Science, 494, 28. 327 Jones, C.R., Ting, Lei, Kaspi, R. and Evans, K.R. (1996) Gallium incorporation kinetics during GSMBE of GaN, gallium nitride and related materials. 1st International Symposium of the Materials Research Society, pp.141–144. 328 Evans, K.R., Lei, T., Kaspi, R. and Jones, C.R. (1995) Presented at Topical Workshop on III–V Nitrides, Nagoya, Japan. 329 Burns, R.P., Gabriel, K.A. and Pierce, D.E. (1993) Journal of the American Ceramic Society, 76, 273. 330 Jenny, J.R., Kaspi, R., Jones, C.R. and Evans, K.R. (1997) Journal of Crystal Growth, 175–176, 89. 331 Parkhomovsky, A., Dabiran, A.M., Benjaminsson, B. and Cohen, P.I. (2001) Applied Physics Letters, 78, 2315. 332 Karpov, Yu.S., Talalaev, R.A., Makarov, Yu.N., Grandjean, N., Massies, J. and Damilano, B. (2000) Surface Science, 450, 191.
References 333 Grandjean, N. and Massies, J. (1997) Applied Physics Letters, 71, 1816. 334 Grandjean, N., Leroux, M., Massies, J., € gt, M. (1999) Mesrine, M. and Lau Japanese Journal of Applied Physics, 38, 618. 335 Webb, J.B., Tang, H., Bardwell, J.A., Moisa, S., Peters, C. and MacElwee, T. (2001) Journal of Crystal Growth, 230, 584. 336 Fu, W. and Venkat, R. (2000) Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 18, 1867. 337 Grandjean, N., Leroux, M., La€vgt, M. and Massies, J. (1997) Applied Physics Letters, 71, 240. 338 Arthur, J.R. (1968) Journal of Applied Physiology, 39, 4032. 339 Hacke, P., Feuillet, G., Okumura, H. and Yoshida, S. (1996) Applied Physics Letters, 69, 2507. 340 Myoung, J.M., Gluschenkov, O., Kim, K. and Kim, S. (1999) Journal of Vacuum Science & Technology A: Vacuum Surfaces and Films, 17, 3019. 341 Iliopoulos, E. and Moustakas, T.D. (2002) Applied Physics Letters, 81, 295. 342 Riechert, H., Averbeck, R., Graber, A., Schienle, M., StrauÔ, U. and Tews, H. (1997) Materials Research Society Symposium Proceedings, 449, 149. 343 Kamp, M. and Riechert, H. (1999) Properties, Processing and Applications of Gallium Nitride and Related Semiconductors (ed. J.H. Edgar et al.), EMIS Data Review Series No. 23, INSPEC, IEE, UK, pp. 426–440. 344 Averyanova, M.V., Karpov, Yu.S., Makarov, Yu.N., Przhevalskii, I.N., Ramm, M.S. and Talalaev, R.A. (1996) MRS Internet Journal of Nitride Semiconductor Research, 1, 31. 345 Botchkarov, A., Salvador, A., Sverdlov, B., Myoung, J. and Morkoc, H. (1995) Journal of Applied Physiology, 77, 4455. 346 Bandic, Z.Z., Hauenstein, R.J., OSteen, M.L. and McGill, T.C. (1996) Applied Physics Letters, 68, 1510.
347 Davydov, V., Klochikhin, A., Ivanov, S., Aderhold, J. and Yamamoto, A. (2003) Growth and properties of InN, in Nitride Semiconductors – Handbook on Materials and Devices (eds P. Ruterana, M. Albrecht and J. Neugebauer), WileyVCH Verlag GmbH, Weinheim, Germany. 348 VanMil, B.L., Guo, H., Holbert, L.J., Lee, K., Myers, T.H., Liu, T. and Korakakis, D. (2004) High temperature limitations for GaN growth by RF-plasma assisted molecular beam epitaxy: effects of active nitrogen species, surface polarity, hydrogen and excess Ga-overpressure. Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 22 (4), 2149–2154. 349 Newman, N. (1998) Thermochemistry of III–N Semiconductors, Semiconductors and Semimetals, vol. 50, Academic Press, New York, pp. 55–101. 350 Adelmann, C., Brault, J., MartinezGuerrero, E., Mula, G., Mariete, H., Dang, L.S. and Daudin, B. (2001) Physica Status Solidi a: Applied Research, 188, 575. 351 Moustakas, T.D. (1996) Materials Reseach Society Symposium Proceedings, 395, 111. 352 Tarsa, E.J., Heying, B., Wu, X.H., Fini, P., DenBaars, S.P. and Speck, J.S. (1997) Journal of Applied Physiology, 82, 5472. 353 Kruse, C., Einfeldt, S., B€ottcher, T., Hommel, D., Rudloff, D. and Christen, J. (2001) Applied Physics Letters, 78, 3827. 354 Parkhomovsky, A., Dabiran, A.M., Benjaminsson, B. and Cohen, P.I. (2001) Applied Physics Letters, 78, 2315. 355 Heying, B., Tarsa, E.J., Elsass, C.R., Fini, P., DenBaars, S.P. and Speck, J.S. (1999) Journal of Applied Physiology, 85, 6470. 356 Mula, G., Adelmann, C., Moehl, S., Oullier, J. and Daudin, B. (2001) Physical Review B: Condensed Matter, 64, 195406–1. 357 Northrup, J.E., Neugebauer, J., Feenstra, R.M. and Smith, A.R. (2000) Physical Review B: Condensed Matter, 61, 9932.
j773
j 3 Growth and Growth Methods for Nitride Semiconductors
774
358 Xie, M.H., Seutter, S.M., Zhu, W.K., Zheng, L.X., Wu, H. and Tong, S.Y. (1999) Physical Review Letters, 82, 2749. 359 Zheng, L.X., Xie, M.H., Seutter, S.M., Cheung, S.H. and Tong, S.Y. (2000) Physical Review Letters, 85, 2352. 360 Xie, M.H., Cheung, S.H., Zheng, L.X., Ng, Y.F., Wu, H., Ohtani, N. and Tong, S.Y. (2000) Physical Review B: Condensed Matter, 61, 9983. 361 Lee, C.D., Sagar, A., Feenstra, R.M., Inoki, C.K., Kuan, T.S., Sarney, W.L., and Salamanca-Riba, L. (2001) Role of Gaflux in dislocation reduction in GaN films grown on SiC (0 0 0 1). Applied Physics Letters, 79, 3428. 362 Keith Bowen, B.K.T.D. (1998) High Resolution X-Ray Diffractometry and Topography, Taylor & Francis. 363 Warren, B.E. (1969) X-Ray Diffraction, Addison-Wesley. 364 Vickers, M.J.K.M.E. and Smeeton, T.M. (2003) Journal of Applied Physiology, 94, 1565. 365 Blakeslee, A.S.A.A.E. (1973) Journal of Applied Crystallography, 6, 19. 366 Heying, B., Wu, X.H., Keller, S., Li, Y., Keller, B., DenBaars, S.P. and Speck, J.S. (1996) Applied Physics Letters, 68, 643. 367 Metzger, T., Hopler, P., Born, E., Ambacher, O., Stutzmann, M., Stommer, R., Schuster, M., Gobel, H., Christiansen, S., Albrecht, M. and Strunk, H.P. (1998) Philosophical Magazine A, 77, 1013. 368 Zhu, Q., Botchkarev, A., Kim, W., Aktas, € Sverdlov, B.N. and Morkoc, H. (1996) O., Applied Physics Letters, 68, 1141. 369 Mohammad, S.N. and Morkoc, H. (1996) Progress in Quantum Electronics, 20, 361. 370 Fujieda, S., Mizuta, M. and Matsumoto, Y. (1987) Japanese Journal of Applied Physics, 26, 2067. 371 Andrews, J.E. and Littlejohn, M.A. (1975) Journal of the Electrochemical Society, 122, 1273. 372 Beaumont, B., Gibart, P. and Faurie, J.P. (1995) Journal of Crystal Growth, 156, 140.
373 Purdy, A.P. (1994) Inorganic Chemistry, 33, 282. 374 Fischer, R.A., Miehr, A., Metzger, T., Born, E., Ambacher, O., Angerer, H. and Dimitrov, R. (1996) Low-temperature OMCVD of InN thin films from the novel air-stable single-molecule precursor azido{bis[(3-dimethylamino) propyl]}indium,(N3)In[(CH2)3NMe2]2, Chemistry of Materials, 8 (7), 1356–1359. 375 Detchprohm, T., Hiramatsu, K., Amano, H. and Akasaki, I. (1992) Applied Physics Letters, 61, 2688. 376 Goetz, W., Romano, L.T., Walker, J., Johnson, N.M. and Molnar, R.J. (1998) Applied Physics Letters, 72, 1214. 377 Naniwae, K., Ithoh, S.K., Amano, H., Ithoh, K., Hiramatsu, K. and Akasaki, I. (1990) Journal of Crystal Growth, 99, 381. 378 Gu, S., Zhang, R., Sun, J., Zhang, L. and Kuech, T.F. (2000) Applied Physics Letters, 76, 3454. 379 Tomita, K., Kachi, T., Nagai, S., Kojima, A., Yamasaki, S. and Koike, M. (2002) Physica Status Solidi a: Applied Research, 194, 563. 380 Kelly, M., Ambacher, O., Stutzmann, M., Brandt, M., Dimitrov, R. and Handschuh, R. (2003) US Patent 6,559,075 B1. 381 Wong, W.S., Sands, T., Cheung, N.W., Kneissl, M., Bour, D.P., Mei, P., Romano, L.T. and Johnson, N.M. (1999) Applied Physics Letters, 75, 1360. 382 Miskys, C.R., Kelly, M.K., Ambacher, O. and Stutzmann, M. (2003) Freestanding GaN-substrates and devices. Physica Status Solidi c, 0 (6), 1627–1650. 383 Gogova, D., Kasic, A., Larsson, H., Pecz, B., Yakimova, R., Ivanov, I.G. and Monemar, B. (2004) Characteristization of high-quality free-standing GaN grown by HVPE. Physica Scripta, T114, 18–21. 384 Gogova, D., Larsson, H., Kasic, A., Yazdi, G.R., Ivanov, I., Yakimova, R., Monemar, B., Aujol, E., Frayssinet, E., Faurie, J.-P., Beaumont, B. and Gibart, P. (2005) High-Quality 200 bulk-like free-standing GaN grown by hydridevapour phase
References
385
386
387
388
389
390
391
392
393
epitaxy on a Si-doped metal organic vapour phase epitaxial GaN template with an ultra low dislocation density. Japanese Journal of Applied Physics, 44 (3), 1181–1185. Kasic, A., Gogova, D., Larsson, H., Ivanov, I., Hemmingsson, C., Yakimova, R., Monemar, B. and Heuken, M. (2005) Characterization of crack-free relaxed GaN grown on 200 sapphire, Journal of Applied Physiology, 98, 073525. Wong, W.S., Sands, T., Cheung, N.W., Kneissl, M., Bour, D.P., Mei, P., Romano, L.T. and Johnson, N.M. (2000) Applied Physics Letters, 77, 2822. Jasinski, J., Liliental-Weber, Z., Estrada, S. and Hu, E. (2002) Applied Physics Letters, 81, 3152. Fang, Z.-Q., Look, D.C., Jasinski, J., Benamara, M., Liliental-Weber, Z, Saarinen, K. and Molnar, R.J. (2000) Evolution of deep centers in GaN grown by hydride vapor phase epitaxy. Applied Physics Letters, 78, 332. Detchprohm, T., Hiramatsu, K., Amano, H. and Akasaki, I. (1992) Applied Physics Letters, 61, 2688. Jasinski, J., Swider, W., Liliental-Weber, Z., Visconti, P., Jones, K.M., Reshchikov, M.A., Yun, F., Morkoc, H., Park, S.S. and Lee, K.Y. (2001) Characterization of freestanding HVPE GaN. Applied Physics Letters, 78, 2297–2299. Fang, Z.-Q., Look, D.C., Visconti, P., Wang, D.-F., Lu, C.-Z., Yun, F., Morkoc, H., Park, S.S. and Lee, K.Y. (2001) Applied Physics Letters, 78 (15), 2178–2181. Chernyak, L., Osinsky, A., Nootz, G., Schulte, A., Jasinski, J., Benamara, M., Liliental-Weber, Z., Look, D.C. and Molnar, R.J. (2000) Applied Physics Letters, 77, 875. Yun, F., Reshchikov, M.A., Jones, K.M., Visconti, P., Morkoc, H., Park, S.S. and Lee, K.Y. (2000) Electrical, structural, and optical characterization of free-standing GaN template grown by hydride vapor phase epitaxy. Solid State Electronics, 44, 2225.
394 Romano, L.T., Krusor, B.S. and Molnar, R.J. (1997) Applied Physics Letters, 71, 2283. 395 Wong, W.S., Sands, T. and Cheung, N.W. (1998) Applied Physics Letters, 72, 599. 396 Leszczynski, M., Suski, T., Teisseyre, H., Perlin, P., Grzegory, I., Jun, J., Porowski, S. and Moustakas, T.D. (1994) Journal of Applied Physiology, 76, 4909. 397 Wickenden, A.E., Koleske, D.D., Henry, R.L., Twigg, M.E. and Fatemi, M. (2004) Resistivity control in unintentionally doped GaN films grown by MOCVD. Journal of Crystal Growth, 260, 54–62. 398 Popovici, G., Kim, W., Botchkarev, A., Tang, H., Solomon, J. and Morkoc, H. (1997) Impurity contamination of GaN epitaxial films from sapphire, SiC, and ZnO substrates. Applied Physics Letters, 71, 3385–3387. 399 Strite, S., Ruan, J., Li, Z., Chen, H., Smith, D.J., Choyke, W.J. and Morkocc, H. (1991) An investigation of the properties of cubic GaN grown on GaAs by plasma assisted molecular-beam epitaxy. Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 9 (4), 1924. 400 Bae, I., Seong, T., Park, Y. and Kim, E.K. (1999) Structural study of GaN grown on (0 0 1) GaAs by organometallic vapor phase epitaxy. Journal of Electronic Materials, 28 (7), 873. 401 Nagayama, A., Katatama, R., Nakadan, N., Miwa, K., Yaguchi, H., Wu, J., Onabe, K. and Shiraki, Y. (1999) Substrate misorientation dependence of the hexagonal phase inclusion in cubic GaN films grown by metalorganic vapor phase epitaxy. Physica Status Solidi a: Applied Research, 176, 513. 402 Yang, C., Wu, M., Lee, C. and Chi, G. (1999) X-ray diffraction characterization of epitaxial zinc-blende GaN films on a miscut GaAs(0 0 1) substrates using the hydride vapor-phase epitaxy method. Journal of Crystal Growth, 206, 8.
j775
j 3 Growth and Growth Methods for Nitride Semiconductors
776
403 Amimer, K., Georgakilas, A., Tsagaraki, K., Androulidaki, M., Cengher, D., Toth, L. and Pecz, B. (2000) Single-crystal hexagonal and cubic GaN growth directly on vicinal (0 0 1) GaAs substrates by molecular-beam epitaxy. Applied Physics Letters, 76 (18), 2580. 404 Wu, J., Kudo, M., Nagayama, A., Yaguchi, H., Onabe, K. and Shiraki, Y. (1999) Selective growth of cubic GaN on patterned GaAs(1 0 0) substrates by metalorganic vapor phase epitaxy. Physica Status Solidi a: Applied Research, 176, 557. 405 Xin, Y., Brown, P.D., Boothroyd, C.B., Preston, A.R., Humphreys, C.J., Cheng, T.S., Foxton, C.T., Andrianov, A.V. and Orton, J.W. (1996) TEM and PL characterization of MBE-growth epitaxial GaN/GaAs. Materials Research Society Symposium Proceedings, 423, 311. 406 Kumagai, V., Murakami, H., Koukitu, A., Takemoto, K. and Seki, H. (2000) Growth of thick hexagonal GaN layer on GaAs (1 1 1)A surfaces for freestanding GaN by metalorganic hydrogen chloride vapor phase epitaxy. Japanese Journal of Applied Physics, 39, L703. 407 Belkov, V.V., Botnaryuk, V.M., Fedorov, L.M., Goncharuk, I.N., Novikov, S.V., Ulin, V.P., Zhilyaev, Yu.V., Cheng, T.S., Jeffs, N.J., Foxton, C.T., Katsavets, N.I. and Harrison, I. (1998) Hydride vapour phase epitaxy on GaN on molecular beam epitaxial GaN substrates. Journal of Crystal Growth, 187, 29. 408 Hasegawa, F., Minami, M., Sunaba, K. and Suemasu, T. (1999) One possibility of obtaining bulk GaN: halide VPE growth at 1000 C on GaAs(1 1 1) substrates. IEICE Transactions on Electronics, E83-C (4), 633. 409 Kumagai, Y., Koukitu, A. and Seki, H. (2000) Investigation of substrate orientation dependence for the growth of GaN on GaAs (1 1 1)A and (1 1 1)B surfaces by metalorganic hydrogen chloride vapor-phase epitaxy. Japanese Journal of Applied Physics, 39, L149.
410 Micovic, M., Kurdoghlian, A., Janke, P., Hashimoto, P., Wong, D.W.S., Moon, J.S., McCray, L. and Nguyen, C. (2001) IEEE Transactions on Education, 48 (3), 591–596. 411 Sverdlov, B., Martin, G.A., Morkoc, H. and Smith, D.J. (1995) Applied Physics Letters, 67, 2063. 412 Moon, Y.T., Liu, C., Xie, J., Ni, X., Fu, Y., Biyikli, N., Morkoc, H., Zhou, L. and Smith, D.J. (2006) in situ pendeo epitaxy of GaN using heteroepitaxial AlGaN/ GaN cracks. Applied Physics Letters, 89, 024103. 413 Bethoux, J.-M., Vennéguès, P., Natali, F., Feltin, E., Tottereau, O., Nataf, G., De Mierry, P. and Semond, F. (2003) Growth of high quality crack-free AlGaN films on GaN templates using plastic relaxation through buried cracks. Journal of Applied Physiology, 94 (10), 6499–6507, 414 Akasaka, T., Nishida, T., Taniyasu, Y., Kasu, M., Makimoto, T. and Kobayashi, N. (2003) Applied Physics Letters, 83 (20), 4140–4142. 415 Shan, W. (1996) Optical studies of GaN and GaN/AlGaN heterostructures on SiC substrate. Applied Physics Letters, 69, 740. 416 Buyanova, A., Bergman, J.P., Monemar, B., Amano, H. and Akasaki, I. (1996) Intrinsic optical properties of GaN epilayers grown on SiC substrates: effect of the built-in strain. Applied Physics Letters, 69, 1255. 417 Davydov, Yu.V., Averkiev, N.S., Goncharuk, I.N., Nelson, D.K., Nikitina, I.P., Polkovnikov, A.S., Smirnov, A.N., Jacobson, M.A. and Semchinova, O.K. (1997) Raman and photoluminescence studies of biaxial strain in GaN epitaxial layers grown on 6H-SiC. Journal of Applied Physiology, 82 (10), 5097. 418 Sun, C.J., Kung, P., Saxler, A., Ohsato, H.H.K. and Razeghi, M. (1994) Journal of Applied Physiology, 75, 3964. 419 Leszczynski, M., Suski, T., Domagala, J. and Prystawko, P. Lattice Parameters of
References
420
421
422
423
424
425
426
427
428
429
the Group III-Nitrides (ed. J.H. Edgar), IEE EMIS Data Review Series, INSPEC, London. Perry, W.G., Zheleva, T., Bremser, M.D., Davis, R.F., Shan, W. and Song, J.J. (1997) Correlation of biaxial strains, bound exciton energies, and defect microstructures in GaN films grown on AlN/6H-SiC(0 0 0 1) substrates. Journal of Electronic Materials, 26 (3), 224. Nikitina, I.P., Sheglov, M.P., Melnik, Yu.V., Irvine, K.G. and Dmitriev, V.A. (1997) Diamond and Related Materials, 6, 1524. Kisielowski, C., Kruger, J., Ruvimov, S., Suski, T., Ager, J.W., III, Jones, E., Liliental-Weber, Z., Rubin, M., Weber, E.R., Bremser, M.D. and Davis, R.F. (1996) Physical Review B: Condensed Matter, 54 (24), 17745–17753. Ahmad, I., Holtz, M., Faleev, N.N. and Temkin, H. (2004) Journal of Applied Physiology, 95, 1692. Edwards, N.V., Bremser, M.D., Davis, R.F., Batchelor, A.D., Yoo, S.D., Karan, C.F. and Aspnes, D.E. (1998) Trends in residual stress for GaN/AlN/6H-SiC heterostructures. Applied Physics Letters, 73 (19), 2808. Waltereit, P. and Brandt, O. (1999) Influence of AlN nucleation on growth mode and strain relief of GaN grown on 6H-SiC(0 0 0 1). Applied Physics Letters, 74, 3660. Krost, A., Dadgar, A., Strassburger, G. and Clos, R. (2003) Physica Status Solidi a: Applied Research, 200 (1), 26–35. Raghavan, S. and Redwing, J.M. (2004) Journal of Crystal Growth, 261 (2–3), 294–300. Han, J., Crawford, M.H., Shul, R.J., Hearne, S.J., Chason, E., Figiel, J.J. and Banas, M. (1999) MRS Internet Journal of Nitride Semiconductor Research, 4S1, G7.7. Kamitani, K., Grimsditch, M., Nipko, J.C., Loong, C.-K., Okada, M. and Kimura, I. (1997) Journal of Applied Physiology, 82 (6), 3152–3154.
430 Amano, H., Iwaya, M., Hayashi, N., Kashima, T., Nitta, S., Wetzel, C. and Akasaki, I. (1999) Control of dislocations and stress in AlGaN on sapphire using a low temperature interlayer. Physica Status Solidi b: Basic Research, 216, 683–690. 431 Twigg, M.E., Henry, R.L., Wickenden, A.E., Koleske, D.D. and Culbertson, J.C. (1999) Applied Physics Letters, 75, 686. 432 Twigg, M.E., Koleske, D.D., Wickenden, A.E., Henry, R.L. and Binari, S.C. (2001) Applied Physics Letters, 79, 4322. 433 George, T., Pike, W.T., Khan, M.A., Kuznia, J.N. and Chang-Chien, P. (1995) A microstructural comparison of the initial growth of AlN and GaN layers on basal plane sapphire and SiC substrates by low pressure metalorganic chemical vapor deposition. Journal of Electronic Materials, 24 (4), 241. 434 Koleske, D.D., Henry, R.L., Twigg, M.E., Culbertson, J.C., Binari, S.C., Wickenden, A.E. and Fatemi, M. (1999) The influence of AlN nucleation layer temperature on GaN electronic properties grown on SiC. Applied Physics Letters, 75 (20), 3141–3143. 435 Ito, T., Ohtsuka, K., Kuwahara, K., Sumiya, M., Takano, Y. and Fuke, S. (1999) Effect of AlN buffer layer deposition conditions on the properties of GaN layer. Journal of Crystal Growth, 205, 22. 436 Fan, Z.Y., Rong, G. and Newman, N. (2000) Defect annihilation in AlN thin films by ultrahigh temperature processing. Applied Physics Letters, 76 (14), 1839. 437 Green, D.S., Gibb, S.R., Hosse, B., Vetury, R., Grider, D.E. and Smart, J.A. (2004) Control of epitaxial defects for optimal AlGaN/GaN HEMT performance and reliability. Journal of Crystal Growth, 272, 285–292. 438 Byun, D., Kim, G., Limc, D., Lee, D., Choi, I.-H., Parkd, D. and Kum, D.-W. (1996) Optimization of the GaN-buffer
j777
j 3 Growth and Growth Methods for Nitride Semiconductors
778
439
440
441
442
443
444
445
446
growth on 6H-SiC (0 0 0 1). Thin Solid Films, 289, 256–260. Lahreche, H., Leroux, M., Laugt, M., Valle, M., Beaumont, B. and Gibart, P. (2000) Buffer free direct growth of GaN on 6H-SC by metalorganic vapor phase epitaxy. Journal of Applied Physiology, 87, 577–583. di Forte Poisson, M.-A., Magis, M., Tordjman, M., Aubry, R., Sarazin, N., Peschang, M., Morvan, E., Delage, S.L., di Persio, J., Quere, R., Grimbert, B., Hoel, V., Delos, E., Ducatteau, D. and Gaquiere, C. (2004) LP-MOCVD growth of GaAlN/GaN heterostructures on silicon carbide: application to HEMT devices. Journal of Crystal Growth, 272, 305–311. Acord, J.D., Raghavan, S., Snyder, D.W. and Redwing, J.M. (2004) in situ stress measurements during MOCVD growth of AlGaN on SiC. Journal of Crystal Growth, 272, 65–71. Doerner, M. and Nix, W. (1988) CRC Critical Reviews in Solid State and Materials Sciences, 14 (3), 225. Floro, J.A., Chason, E., Lee, S.R., Twesten, R.D., Hwang, R.Q. and Freund, L.B. (1997) Journal of Electronic Materials, 26, 969–979. Onojima, N., Suda, J. and Matsunami, H. (2002) Lattice relaxation process of AlN growth on atomically flat 6H-SiC substrate in molecular beam epitaxy. Journal of Crystal Growth, 237–239, 1012–1016. Davis, R.F., Tanaka, S., Rowland, L.B., Kern, R.S., Sitar, Z., Ailey, S.K. and Wang, C. (1996) Growth of SiC and III–V nitride thin films via gas-source molecular beam epitaxy and their characterization. Journal of Crystal Growth, 164, 132. Nakamura, S., Kimoto, T., Matsunami, H., Tanaka, S., Teraguchi, N. and Suzuki, A. (2000) Formation of periodic steps with a unit-cell height on 6H-SiC(0 0 0 1) surface by HCl etching. Applied Physics Letters, 76 (23), 3412.
447 Sverdlov, B., Martin, G.A., Morkoc, H. and Smith, D.J. (1995) Formation of threading defects in GaN and related wurtzite films grown on non-isomorphic substrates. Applied Physics Letters, 67 (14), 2063–2065. 448 Hellman, E.S. (1998) The polarity of GaN: a critical review. MRS Internet Journal of Nitride Semiconductor Research, 3, 11. 449 Sasaki, S. and Matsuoka, T. (1988) Substrate-polarity dependence of metalorganic vapor-phase epitaxy grown GaN on SiC. Journal of Applied Physiology, 64 (9), 4531. 450 Capaz, R.B., Lim, H. and Joannopoulos, J.D. (1995) Ab initio studies of GaN epitaxial growth on SiC. Physical Review B: Condensed Matter, 51 (17), 755. 451 Ren, S. and Dow, J.D. (1996) Latticematching SiC substrate with GaN. Applied Physics Letters, 69 (2), 251. 452 Ren, S. and Dow, J.D. (1997) Role of interfacial-charge in the growth of GaN on a-SiC. Journal of Electronic Materials, 26 (4), 341. 453 Stirman, J.N., Ponce, F.A., Pavlovska, A., Tsong, I.S.T. and Smith, D.J. (2000) Polarity determination and atomic arrangements at a GaN/SiC interface using high-resolution image matching. Applied Physics Letters, 76 (7), 822. 454 Lin, C.F. and Cheng, H.C. (1997) Growth and characterizations of GaN on SiC substrates with buffer layers. Journal of Applied Physiology, 82, 2378. 455 Vermaut, P., Ruterana, P., Nouet, G., Salvador, A. and Morkoc, H. (1997) HREM and CBED studies of polarity of nitride layers with prismatic defects grown over SiC, in III–V Nitrides (eds F.A. Ponce, T.D. Moustakas, I. Akasaki and B.A. Monemar), Materials Research Society, Pittsburgh, PA, Materials Research Society Symposium Proceedings, 449, 317. 456 Torvik, J.T. (1998) Optical characterization of GaN/SiC n–p
References
457
458
459
460 461
462
463
464
465
466
heterojunctions and p-SiC. Applied Physics Letters, 72 (8), 945. Smith, D.J., Chandrasekhar, D., Sverdlov, B., Botchkarev, A., Salvador, A. and Morkoc, H. (1995) Characterization of structural defects in wurtzite GaN grown on 6H SiC using plasmaenhanced molecular beam epitaxy. Applied Physics Letters, 67, 1830. Byun, D., Kim, G. and Lim, D. (1996) Optimization of the GaN-buffer growth on 6H-SiC(0 0 0 1). Thin Solid Films, 289, 256. di Forte-Poissona, M.A., Romanna, A., Tordjmana, M., Magisa, M., Di Persiob, J., Jacquesc, Ch. and Vicentec, P. (2003) LPMOCVD growth of GaN on silicon carbide. Journal of Crystal Growth, 248, 533. Look, D.C. and Sizelove, J.R. (1999) Physical Review Letters, 82, 1237. Byun, D., Kim, G., Lim, D., Choi, I.-H., Park, D. and Kum, D.-W. (1996) Proc of International Symposium on Blue Lasers and Light Emitting Diodes, March 5–7, Chiba University, Japan, p.380. Smith, D.J., Chandrasekhar, D., Sverdlov, B., Bochkarev, A., Salvador, A. and Morkoc, H. (1995) Applied Physics Letters, 67, 1830. Mynbaeva, M., Saddow, S.E., Melnychuk, G., Nikitina, I., Scheglov, M., Sitnikova, A., Kuznetsov, N., Mynbaev, K. and Dmitriev, V. (2001) Applied Physics Letters, 78, 117. Mynbaeva, M., Titkov, A., Kryganovskii, A., Ratnikov, V., Mynbaev, K., Huhtinen, H., Laiho, R. and Dmitriev, V. (2000) Applied Physics Letters, 76, 1113. Zubia, D. and Hersee, S.D. (1999) Nanoheteroepitaxy: the application of nanostructuring and substrate compliance to the heteroepitaxy of mismatched semiconductor materials. Journal of Applied Physics, 85 (9), 6492–42961. PSiC substrates from TDI, Inc. The later varieties have been produced by
467
468
469
470
471
472
473
474
475
Professors W.J. Choyke and R. Devaty at the University of Pittsburgh. Yun, F., Reshchikov, M.A., He, L., Morkoc, H., Inoki, C.K. and Kuan, T.S. (2002) Growth of GaN films on porous SiC substrate by molecular beam epitaxy. Applied Physics Letters, 81, 4142–4143. Sagar, A., Lee, C.D., Feenstra, R.M., Inoki, C.K. and Kuan., T.S. (2002) Morphology and effects of hydrogen etching of porous SiC. Journal of Applied Physiology, 92, 4070. Inoki, C.K., Kuan, T.S., Lee, C.D., Sagar, A. and Feenstra, R.M. (2002) Materials Research Society Symposium Proceedings, 722, K1.3. Yun, F., Reshchikov, M.A., He, L., King, T., Huang, D., Morkoc, H., Inoki, C.K. and Kuan, T.S. (2002) Materials Research Society Symposium Proceedings, 719, F1.3. Okumura, H., Ohta, K., Feuillet, G., Balakrishnan, K., Chichibu, S., Hamaguchi, H., Hacke, P. and Yoshida, S. (1997) Growth and characterization of cubic GaN. Journal of Crystal Growth, 178, 113. Wu, J., Yaguchi, H., Nagasawa, H., Yamaguchi, Y., Onabe, K., Shiraki, Y. and Ito, R. (1997) Crystal structure of GaN grown on 3C-SiC substrates by metalorganic vapor phase epitaxy. Japanese Journal of Applied Physics, 36 (7A), 4241. Headrick, R.L., Kycia, S., Park, Y.K., Woll, A.R. and Brock, Y.K. (1996) Realtime X-ray-scattering measurement of the nucleation kinetics of cubic gallium nitride on b-SiC (0 0 1). Physical Review B: Condensed Matter, 54 (20), 14686. Gerthsen, D., Neubauer, B., Dieker, Ch., Lantier, R., Rizzi, A. and Luth, H. (1999) Molecular beam epitaxy (MBE) growth and structural properties of GaN and AlN on 3C-SiC(0 0 1) substrates. Journal of Crystal Growth, 200, 353. Cervantes-Contreras, M., Lopez-Lopez, M., Melendez-Lira, M., Tamura, M. and Hiroyama, Y. (2001) Molecular beam epitaxial growth of GaN on (1 0 0)- and
j779
j 3 Growth and Growth Methods for Nitride Semiconductors
780
476
477
478
479
480
481
482
483
(1 1 1) Si substrates coated with a thin SiC layer. Journal of Crystal Growth 227–228, 425. Daudin, B., Feuillet, G., Hubner, J., Samson, Y., Widman, F., Philippe, A., Bru-Chevallier, C., Guillot, G., Bustarret, E., Bentoumi, G. and Deneuville, A. (1998) How to grow cubic GaN with low hexagonal phase content on (0 0 1) SiC by molecular beam epitaxy. Journal of Applied Physiology, 84 (4), 2295. Wang, D., Hiroyama, Y., Tamura, M., Ichikawa, M. and Yoshida, S. (2000) Heteroepitaxial growth of cubic GaN on Si(0 0 1) coated with thin flat SiC by plasma-assisted molecular-beam epitaxy. Applied Physics Letters, 76, 1683. As, D.J.,Frey, T., Schikora,D., Lischka,K., Cinalla, V., Pezoldt, J., Goldhahn, R., Kaiser, S. and Gebhardt, W. (2000) Cubic GaN epilayers grown by molecular beam epitaxy on thin b-SiC/Si(0 0 1) substrates. Applied Physics Letters, 76, 1686. Rossner, U., Barski, A., Rouviere, J.L., Bourret, A., Massies, J., Deparis, C. and Grandjean, N. (1995) How to induce the epitaxial growth of gallium nitride on Si (0 0 1). Materials Science & Engineering B: Solid State Materials for Advanced Technology, 29, 74. Zhang, X., Chua, S.J., Feng, Z.C., Chen, J. and Lin, J. (1999) OMVPE growth and characterization of GaN films with composite intermediate layer buffer on Si substrate. Physica Status Solidi a: Applied Research, 176, 605. Kim, S.T., Lee, Y.J., Chung, S.H. and Moon, D.C. (1999) Relationship between crystallographic orientation and 3.42 eV emission bands in GaN grown by HVPE on Si substrate. Semiconductor Science and Technology, 14, 156. Lei, T., Moustakas, T.D., Graham, R.J., He, Y. and Berkowitz, S.J. (1992) Journal of Applied Physiology, 71, 4933. Yokouchi, K., Araki, T., Nagatomo, T. and Omoto, O. (1996) Institute of Physics Conference Series, 142, 867.
484 Ohtani, A., Stevens, K.S. and Beresford, R. (1994) Materials Research Society Symposium Proceedings, 339, 471–476, 485 Basu, S.N., Lei, T. and Moustakas, T.D. (1994) Journal of Materials Research, 9, 2370. 486 Schulze, F, Dadgar, A., Blasing, J. and Krost, A. (2004) GaN heteroepitaxy on Si (0 0 1). Journal of Crystal Growth, 272 (1–4), 496–499. 487 Guha, S. and Bojarczuk, N.A. (1998) Ultraviolet and violet GaN light emitting diodes on silicon. Applied Physics Letters, 72 (4), 415. 488 Tran, C.A., Osinski, A. and Karlicek, R.F., Jr (1999) Applied Physics Letters, 75, 1494. 489 Osinsky, A., Gangopadhyay, S., Wang, J.W., Gaska, R., Kuksenkov, D., Temkin, H., Shmagin, I.K., Chang, Y.C., Muth, J.F. and Kolbas, R.M. (1998) Visibleblind GaN barrier detectors grown on Si (1 1 1). Applied Physics Letters, 72 (5), 551. 490 Johnson, J.W., Piner, E.L., Vescan, A., Therrien, R., Rajagopal, P., Roberts, J.C., Brown, J.D., Singhal, S. and Linthicum, K.J. (2004) 12 W/mm AlGaN-GaN HFETs on silicon substrates. IEEE Electron Device Letters, 25 (7), 459–461. 491 Steves, K.S., Kinniburgh, M., Schwartzman, A.F., Ohtani, A. and Beresford, R. (1995) Demonstration of a silicon field-effect transisitor using AlN as the gate dielectric. Applied Physics Letters, 66 (23), 3179. 492 Martin, G.A., Sverdlov, B.N., Botchkarev, A., Morkoc, H., Smith, D.J., Tsen, S.-C.Y., Thompson, W.H. and Nayfeh, M.H. (1996) Gallium nitride epitaxy on silicon: importance of substrate preparation. Materials Research Society Symposium Proceedings, 395, 67. 493 Zamir, S., Meyler, B., Zolotoyabko, E. and Salzman, J. (2000) The effect of AlN buffer layer on GaN grown on (1 1 1)oriented Si substrates by OMVPE. Journal of Crystal Growth, 218, 181. 494 Lahreche, H., Vennegues, P., Tottereau, O., Laugt, M., Lorenzini, P., Leroux, M.,
References
495
496
497
498
499
500
501
502
503
Beaumont, B. and Gibar, P. (2000) Optimisation of AlN and GaN growth by metalorganic vapour-phase epitaxy (OMVPE) on Si(1 1 1). Journal of Crystal Growth, 217, 13. Morkoc, H. (2003) Growth of III-nitride semiconductors and their characterization, in Wide Energy Bandgap Electronics (eds F. Ren and J. Zolper), World Scientific. Kawaguchi, Y., Honda, Y., Yamaguchi, M., Sawaki, N. and Hiramatsu, K. (1999) Selective area growth of GaN on stripepatterned (1 1 1) Si substrate by metalorganic vapor phase epitaxy. Physica Status Solidi, 176, 533. Seon, M., Prokofyeva, T. and Holtz, M. (2000) Selective growth of high quality GaN on Si(1 1 1) substrates. Applied Physics Letters, 76 (14), 1842. Armitage, R., Yang, Q., Feick, H., Gebauer, J. and Weber, E.R. (2002) Applied Physics Letters, 81, 1450. Vennegues, P., Bougrioua, Z., Bethoux, J.M., Azize, M. and Tottereau, O. (2005) Journal of Applied Physiology, 97, 024912. Krost, A. and Dadgar, A. (2002) GaNbased optoelectronics on silicon substrates. Materials Science & Engineering B: Solid State Materials for Advanced Technology, 93, 77–84. Dadgar, A., Strittmatter, A., Bl€asing, J., Poschenrieder, M., Contreras, O., Veit, P., Riemann, T., Bertram, F., Reiher, A., Krtschil, A., Diez, A., Hempel, T., Finger, T., Kasic, A., Schubert, M., Bimberg, D., Ponce, F.A., Christen, J. and Krost, A. (2003) Metalorganic chemical vapor phase epitaxy of gallium-nitride on silicon. Physica Status Solidi c, 0 (6), 1583–1606. Ishikawa, H., Yamamoto, K., Egawa, T., Soga, T., Jimbo, T. and Umeno, M. (1998) Thermal stability of GaN on (1 1 1) Si substrate. Journal of Crystal Growth, 189–190, 178–182. Almeida, R.M.C. and Baumvol, I.J.R. Physical Review B: Condensed Matter, 62, (2000) R16255.
504 Follstaedt, D.M., Han, J., Provencio, P. and Fleming, J.G. (1999) MRS Internet Journal of Nitride Semiconductor Research, 4S1, G3.72. 505 Lahreche, H., Vennegues, P., Totterau, O., Laust, M., Lorenzini, P., Leroux, M., Beaumont, B. and Gibart, P. (2000) Journal of Crystal Growth, 217, 13. 506 Zamir, S., Meyler, B., Zolotoyabko, E. and Salzman, J. (2000) Journal of Crystal Growth, 218, 181. 507 Amano, H., Iwaya, M., Hayashi, N., Kashima, T., Nitta, S., Wetzel, C. and Akasaki, I. (1999) Control of dislocations and stress in AlGaN on sapphire using a low temperature interlayer. Physica Status Solidi b: Basic Research, 216 (1), 683–689; Amano, H., Iwaya, M., Hayashi, N., Kashima, T., Katsuragawa, M., Takeuchi, T., Wetzel, C. and Akasaki, I. (1999) Improvement of crystalline quality of group III nitrides on sapphire using low temperature interlayers. GaN and Related Alloys Symposium, Materials Research Society, p. G10.1.1/8. 508 Dadgar, A., Blasing, J., Diez, A., Alam, A., Heuken, M. and Krost, A. (2000) Japanese Journal of Applied Physics, 39, L1183. 509 Blasing, J., Reiher, A., Dadgar, A., Diez, A. and Krost, A. (2002) Applied Physics Letters, 81, 2722. 510 Dadgar, A., Christen, J., Richter, S., Bertram, F., Diez, A., Blasing, J., Krost, A., Strittmatter, A., Bimberg, D., Alam, A. and Heuken, M. (2000) InGaN/GaN blue light emitter grown on Si(1 1 1) using an AlAs seed layer. Proceedings of International Workshop on Nitride Semiconductors, Institue of Pure & Applied Physics, Tokyo, Japan, pp.845–848. 511 Krost, A. and Dadgar, A. (2002) Physica Status Solidi a: Applied Research, 194 (2), 361–375. 512 Feltin, E., Beaumont, B., Laugt, M., De Mierry, P., Vennegues, P., Lahreche, H.,
j781
j 3 Growth and Growth Methods for Nitride Semiconductors
782
513
514
515 516
517
518
519
520
521
522
Leroux, M. and Gibart, P. (2001) Stress control in GaN grown on silicon (1 1 1) by metalorganic vapor phase epitaxy. Applied Physics Letters, 79 (20), 3230–3232. Feltin, E., Beaumont, B., Laugt, M., De Mierry, P., Vennegues, P., Leroux, M. and Gibart, P. (2001) Crack-free thick GaN layers on silicon (1 1 1) by metalorganic vapor phase epitaxy. Physica Status Solidi a: Applied Research, 188 (2), 531–535. Marchand, H., Zhao, L., Zhang, N., Moran, B., Coffie, R., Mishra, U.K., Speck, J.S., DenBaars, S.P. and Freitas, J.A. (2001) Metalorganic chemical vapor deposition of GaN on Si(1 1 1): stress control and application to field-effect transistors. Journal of Applied Physiology, 89 (12), 7846–7851. Zamir, S., Meyler, B. and Salzman, J. (2001) Applied Physics Letters, 78, 288. Krost, A and Dadgar, A. (2002) GaNbased devices on Si [MOVPE film growth]. Physica Status Solidi a: Applied Research, 194 (2), 361–375. Dadgar, A., Alam, A., Riemann, T., Blasing, J., Diez, A., Poschenrieder, M., Strassburg, M., Heuken, M., Christen, J. and Krost, A. (2001) Crack-free InGaN/ GaN light emitters on Si(1 1 1). Physica Status Solidi a: Applied Research, 188 (1), 155–158, Feltin, E., Dalmasso, S., de Mierry, P., Beaumont, B., Lahreche, H., Bouille, A., Haas, H., Leroux, M. and Gibart, P. (2001) Japanese Journal of Applied Physics, 40, L738. Dadgar, A., Poschenrieder, M., Blasing, J., Fehse, K., Diez, A. and Krost, A. (2002) Applied Physics Letters, 80, 3670. Sato, H., Tokashi, H., Watanabe, A. and Ota, H. (1996) Applied Physics Letters, 68, 3617. Fujieda, S., Mizuta, M. and Matsumoto, Y. (1987) Japanese Journal of Applied Physics, 26, 2067. Okumura, H., Misawa, S. and Yoshida, S. (1991) Applied Physics Letters, 59, 1058.
523 Andrews, J.E. and Littlejohn, M.A. (1975) Journal of the Electrochemical Society, 122, 1273. 524 Manasevit, H.M., Erdmann, F.M. and Simpson, W.I. (1971) Journal of the Electrochemical Society, 118, 1864; See also Pizzarello, F.A. and Coker, J.E. (1975) Journal of Electronic Materials, 4, 25. 525 OClock, G.D., Jr and Duffy, M.T. (1973) Applied Physics Letters, 23, 55. 526 Liu, J.K., Lakin, K.M. and Wang, K.L. (1975) Journal of Applied Physiology, 46, 3703. 527 Khan, M.A., Skogman, R.A., Schulze, R.G. and Gershenzon, M. (1983) Applied Physics Letters, 42, 430. 528 Hashimoto, M., Amano, H., Sawaki, N. and Akasaki, I. (1984) Journal of Crystal Growth, 68, 163. 529 Matloubian, M. and Gershenzon, M. (1985) Journal of Electronic Materials, 14, 633. 530 Nagatomo, T., Kuboyama, T., Minamino, H. and Omoto, O. (1989) Japanese Journal of Applied Physics, 28, L1334. 531 Sasaki, T. and Matsuoka, T. (1988) Journal of Applied Physiology, 64, 4531. 532 Khan, M.A., Kuznia, J.N., Van Hove, J.M., Olsen, D.T., Krishnankutty, S. and Kolbas, R.M. (1991) Applied Physics Letters, 58, 526. 533 Nakamura, S., Harada, Y. and Seno, M. (1991) Applied Physics Letters, 58, 2021. 534 Amano, H., Akasaki, I., Hiramatsu, K. and Sawaki, N. (1988) Thin Solid films, 163, 415. 535 Akasaki, I. Amano, H., Koide, Y., Hiramatsu, K. and Sawaki, N. (1989) Journal of Crystal Growth, 98. 536 Amano, H., Sawaki, N., Akasaki, I. and Toyoda, Y. (1986) Applied Physics Letters, 48, 353. 537 Akasaki, I., Amano, H., Koide, Y., Hiramatsu, K. and Sawaki, N. (1989) Journal of Crystal Growth, 98, 209, 538 Cho, H.K., Lee, J.Y., Kim, K.S., Yang, G.M., Song, J.H. and Yu, P.W. (2000) Journal of Applied Physiology, 89, 2617.
References 539 Kim, K.S., Oh, C.S., Lee, K.J., Yang, G.M., Hong, C.-H., Lim, K.Y., Lee, Hy.J. and Yoshikawa, A. (1999) Journal of Applied Physiology, 85, 8441. 540 Cho, H.K., Lee, J.Y., Kim, K.S., Yang, G.M., Song, J.H. and Yu, P.W. (2001) Journal of Applied Physiology, 89, 2671. 541 Ito, T., Sumiya, M., Takano, Y., Ohtsuka, K. and Fuke, S. (1999) Japanese Journal of Applied Physics, 38, 649. 542 Wang, T., Shirahama, T., Sun, H.B., Wang, H.X., Bai, J., Sakai, S. and Misawa, H. (2000) Applied Physics Letters, 76, 2220. 543 Hashimoto, T., Yuri, M., Ishida, M., Terakoshi, Y., Imafuji, O., Sugino, T. and Itoh, K. (1999) Japanese Journal of Applied Physics, 38, 6605. 544 Xu, H.Z., Takahashi, K., Wang, C.X., Wang, Z.G., Okada, Y., Kawabe, M., Harrison, I. and Foxon, C.T. (2001) Journal of Crystal Growth, 222, 110. 545 Yuasa, T., Ueta, Y., Tsuda, Y., Ogawa, A., Taneya, M. and Takao, K. (1999) Japanese Journal of Applied Physics, 38, L703–L705. 546 Figge, S., B€ ottcher, T., Einfeldt, S. and Hommel, D. (2000) Journal of Crystal Growth, 221, 262. 547 Yamaguchi, S., Kariya, M., Kosaki, M., Yukawa, Y., Nitta, S., Amano, H. and Akasaki, I. (2001) Journal of Applied Physiology, 89, 7820. 548 Lorenz, K., Gonsalves, M., Kim, W., Narayanan, V. and Mahajan, S. (2000) Applied Physics Letters, 77, 3391. 549 Narayanan, V., Lorenz, K., Kim, W. and Mahajan, S. (2002) Philosophical Magazine A, 82, 5885. 550 Gonsalves, M., Kim, W., Narayanan, V. and Mahajan, S. (2002) Journal of Crystal Growth, 240, 347. 551 Hiramatsu, K., Itoh, S., Amano, H., Akasaki, I., Kuwano, N., Shiraishi, T. and Oki, K. (1991) Journal of Crystal Growth, 115, 628. 552 Hersee, S.D., Ramer, J.C. and Malloy, K.J. (1997) The microstructure of metalorganic-chemical-vapor-
553
554
555
556
557 558
559
560
561
562
deposition gan on sapphire. MRS Bulletin 22, 7. Zhao, D.G., Zhu, J.J., Liu, Z.S., Zhang, S.M., Yang, H. and Jiang, D.S. (2004) Surface morphology of AlN buffer layer and its effect on GaN growth by metalorganic chemical vapor deposition. Applied Physics Letters, 85, 1499. Zhang, J.C., Zhao, D.G., Wang, J.F., Wang, Y.T., Chen, J., Liu, J.P. and Yang, H. (2004) The influence of AlN buffer layer thickness on the properties of GaN epilayer. Journal of Crystal Growth, 268, 24. Ohba, Y. and Iida, S. (2003) Roomtemperature operation of GaN-based blue-violet laser diodes fabricated on sapphire substrates using hightemperature-grown single-crystal AlN buffer layers. Physica Status Solidi a: Applied Research, 200 (1), 126–130. Ohba, Y. and Sato, R. (2000) Growth of AlN on sapphire substrates by using a thin AlN buffer layer grown twodimensionally at a very low V/III ratio. Journal of Crystal Growth, 221, 258–261. Nakamura, S. (1991) Japanese Journal of Applied Physics, 30, L1705. Wickenden, A.E., Wickenden, D.K. and Kistenmacher, T.J. (1994) Journal of Applied Physiology, 75, 5367. Hiramatsu, K., Itoh, S., Amano, H., Akasaki, I., Kuwano, N., Shiraishi, T. and Oh, K. (1991) Journal of Crystal Growth, 115, 628. Fuke, S., Teshigawara, H., Kuwahara, K., Takano, Y., Ito, T., Yanagihara, M. and Ohtsuka, K. (1998) Journal of Applied Physiology, 83 (2), 764–767. Yi, M.S., Lee, H.H., Kim, D.J., Park, S.J., Noh, D.Y., Kim, C.C. and Je, J.H. (1999) Effects of growth temperature on GaN nucleation layers. Applied Physics Letters, 75 (15), 2187–2189. Sugiura, L., Itaya, K., Nishio, J., Fujimoto, H. and Kokubun, Y. (1997) Effects of thermal treatment of lowtemperature GaN buffer layers on the quality of subsequent GaN layers.
j783
j 3 Growth and Growth Methods for Nitride Semiconductors
784
563
564
565
566
567
568
569
570
571
572
573
Journal of Applied Physiology, 82 (10), 4877–4882; Yamamoto, A., Shin-ya, T., Sugiura, T. and Hashimoto, A. (1998) Journal of Crystal Growth, 189–190, 461. Yang, C.-C., Wu, M.-C., Lee, C.-H. and Chi, G.-C. (2000) Microstructural evolution in a multiple composite layer of GaN on sapphire by organometallic vapor phase epitaxy. Journal of Applied Physiology, 87 (9), 4240–4242. Kim, K.S., Oh, C.S., Lee, K.J., Yang, G.M., Hong, C.-H., Lim, K.Y., Lee, H.J. and Yoshikawa, A. (1999) Effects of growth rate of a GaN buffer layer on the properties of GaN on a sapphire substrate. Journal of Applied Physics, 85 (12), 8441–8444. Gibart, P. (2004) Metal organic vapour phase epitaxy of GaN and lateral overgrowth. Reports on Progress in Physics, 67 (5), 667–715. Pozina, G., Bergman, J.P., Monemar, B., Yamaguchi, S., Amano, H. and Akasaki, I. (2000) Applied Physics Letters, 76, 3388. Yamaguchi, S., Kariya, M., Nitta, S., Amano, H. and Akasaki, I. (1999) Applied Physics Letters, 75, 4106. Yamaguchi, S., Kariya, M., Nitta, S., Kashima, T., Kosaki, M., Yukawa, Y., Amano, H. and Akasaki, I. (2000) Journal of Crystal Growth, 221, 327. Chung, H.M., Chuang, W.C., Pan, Y.C., Tsai, C.C., Lee, M.C., Chen, W.H., Chen, W.K., Chiang, C.I., Lin, C.H. and Chang, H. (2000) Applied Physics Letters, 76, 897. Lu, D.-C., Wang, C., Yuan, H., Liu, X. and Wang, X. (2000) Journal of Crystal Growth, 221, 356. Asano, T., Yanashima, K., Asatsuma, T., Hino, T., Yamaguchi, T., Tomiya, S., Funato, K., Kobayashi, T. and Ikeda, M. (1999) Physica Status Solidi a: Applied Research, 176, 23. Liu, B.L., Lachab, M., Jia, A., Yoshikawaa, A. and Takahashi, K. (2002) Journal of Crystal Growth, 234, 637. Huang, D., Reshchikov, M.A., Yun, F., King, T., Baski, A.A. and Morkoc, H. (2002) Applied Physics Letters, 80. 216.
574 Kim, D.-J., Moon, Y.T., Ahn, K.-S. and Park, S.J. (2000) in situ normal incidence reflectance study on the effect of growth rate of nucleation layer on GaN by metalorganic chemical vapor deposition. Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 18, 140–143. 575 Gibart, P., Beaumont, B. and Vennegues, P. (2003) Epitaxial lateral overgrowth of GaN, in Nitride Semiconductors – Handbook on Materials and Devices (eds P. Ruterana, M. Albrecht and J. Neugebauer), WileyVCH Verlag GmbH, Weinheim, Germany. 576 Zheleva, T.S., Nam, O.-H., Ashmawi, W.M., Griffin, J.D. and Davis, R.F. (2001) Journal of Crystal Growth, 222, 706. 577 Zheleva, T.S., Karoui, F., Kirchner, K., Derenge, M., Jones, K.A., Vispute, R.D. and Venkatesan, T. (2001) Materials Research Society Symposium Proceedings, 636, G5.1.1. 578 Davis, R.F., Gehrke, T., Linthicum, K.J., Preble, E., Rajagopal, P., Ronning, C., Zorman, C. and Mehregany, M. (2001) Journal of Crystal Growth, 231, 335. 579 Joyce, B.D. and Baldrey, J.A. (1962) Selective epitaxial deposition of silicon. Nature, 195, 485, 580 For a review, see Goulding, M.R. (1993) Materials Science & Engineering B: Solid State Materials for Advanced Technology, 17, 47–67. 581 Zingg, R.P., Neudeck, G.W., Hoefflinger, B. and Liu, S.T. (1986) Epitaxial lateral overgrowth of silicon over thick steps of SiO2. Journal of the Electrochemical Society, 133 (6), 1274–1275. 582 Tsaur, B-Y, McClelland, R.W., Fan, J.C.C., Gale, R.P., Salerno, J.P., Vojak, B.J. and Bozler, C.O. (1982) Applied Physics Letters, 41, 347. 583 Morrison, A.D. and Daud, T. (1985) Low defect, high purity crystalline layers
References
584
585
586
587
588
589
590
591 592
593
594
595
596
597
grown by selective deposition. US Patent 4,522,661. Kato, K., Kusunoki, T., Takenaka, C., Tanahashi, T. and Nakajima, K. (1991) Journal of Crystal Growth, 115, 174. Parillaud, O., Gil-Lafon, E., Gerard, B., Etienne, P. and Pribat, D. (1996) Applied Physics Letters, 68, 2654. Gibart, P., Tromson-Carli, A., Monteil, Y. and Rudra, A. (1989) Journal of Physics, 50 (C5), 529. Kato, Y., Kitamura, S., Hiramatsu, K. and Sawaki, S. (1994) Journal of Crystal Growth, 144, 133. Nam, O.H., Bremser, M.D., Zheleva, T.S. and Davis, R.F. (1997) Applied Physics Letters, 71, 2638. Zheleva, T.S., Nam, O.-H., Bremser, D.M. and Davis, R.F. (1997) Applied Physics Letters, 71, 2472. Nakamura, S., Senoh, M., Nagahama, S., Iwasa, N., Yamanda, T., Matsushita, T., Kiyoku, H., Sugimoto, Y., Kozaki, T., Umemoto, H., Sano, M. and Chocho, K. (1998) Applied Physics Letters, 72, 211. Sakai, A., Sunakawa, H. and Usui, A. (1997) Applied Physics Letters, 71, 2259. Kurai, S., Nishino, K. and Sakai, S. (1997) Japanese Journal of Applied Physics, 36, L184. Wang, J., Tottori, S., Sato, H., Hao, M.-S., Ishikawa, Y., Sugahara, T., Yamashita, K. and Sakai, S. (1998) Japanese Journal of Applied Physics, 37, 4475. Park, J., Grudowski, P.A., Eiting, C.J. and Dupuis, R.D. (1998) Applied Physics Letters, 73, 333. Nam, O.-H., Zheleva, T.S., Bremser, M.D. and Davis, R.F. (1998) Journal of Electronic Materials, 27, 233. Tadatomo, K., Ohuchi, Y., Okagawa, H., Itoh, H., Miyake, H. and Hiramatsu, K. (1999) MRS Internet Journal of Nitride Semiconductor Research, 4S1, G3.1. Marchand, H., Ibbetson, J.P., Fini, P.T., Wu, X.H., Keller, S., DenBaars, S.P., Speck, J.S. and Mishra, U.K. (1999) MRS Internet Journal of Nitride Semiconductor Research, 4S1, G4.5.
598 Hiramatsu, K., Nishiyama, K., Motogaito, A., Miyake, H., Iyechika, Y. and Maeda, T. (1999) Physica Status Solidi a: Applied Research, 176, 535. 599 Kapolnek, D., Keller, S., Vetury, R., Underwood, R.D., Kozodoy, P., DenBaars, S.P. and Mishra, U.K. (1997) Applied Physics Letters, 71, 1204. 600 Nam, O.-H., Bremser, M.D., Ward, B.L., Nemanich, R.J. and Davis, R.F. (1997) Japanese Journal of Applied Physics, 36 (5A), L532. 601 Yu, Z., Johnson, M.A.L., Brown, J.D., ElMasry, N.A., Muth, J.F., Cook, J.W., Jr, Schetzina, J.F., Haberern, K.W., Kong, H.S. and Edmond, J.A. (1999) MRS Internet Journal of Nitride Semiconductor Research, 4S1, G4.3. 602 Nam, O.-H., Bremser, M.D,. Ward, B.L., Nemanich, R.J. and Davis, R.F. (1997) Japanese Journal of Applied Physics, 36 (5A), L532. 603 Marchand, H., Ibbetson, J.P., Fini, P.T., DenBaars, S.P., Speck, J.S. and Mishra, U.K. (1998) Journal of Crystal Growth, 195, 328. 604 Beaumont, B., Haffouz, S. and Gibart, P. (1998) Applied Physics Letters, 72 (8), 921. 605 Kawaguchi, Y., Nambu, S., Sone, H., Yamaguchi, M., Miyake, H., Hiramatsu, K., Sawaki, N., Iyechika, Y. and Maeda, T. (1999) MRS Internet Journal of Nitride Semiconductor Research, 4S1, G4.1. 606 Kawaguchi, Y., Nambu, S., Yamaguchi, M., Sawaki, N., Miyake, H., Hiramatsu, K., Tsukamoto, K., Kuwano, N. and Oki, K. (1999) Physica Status Solidi a: Applied Research, 176, 561. 607 Kawaguchi, Y., Honda, Y., Yamaguchi, M., Sawaki, N. and Hiramatsu, K. (1999) Physica Status Solidi a: Applied Research, 176, 553. 608 Zhang, X., Dapkus, P.D. and Rich, D.H. (2000) Applied Physics Letters, 77, 1496. 609 Qhalid Fareed, R.S., Yang, J.W., Zhang, J., Adivarahan, V., Chaturvedi, V. and Khan, A.M. (2000) Applied Physics Letters, 77, 2343.
j785
j 3 Growth and Growth Methods for Nitride Semiconductors
786
610 Keller, S., Keller, B.P., Wu, Y.F., Heying, B., Kapolnek, D., Speck, J.S., Mishra, U.K. and DenBaars, S.P. (1996) Applied Physics Letters, 68, 1525. 611 Nitta, S., Kashima, T., Kariya, M., Yukawa, Y., Yamaguchi, S., Amano, H. and Akasaki, I. (2000) MRS Internet Journal of Nitride Semiconductor Research, 5S1, W2.8. 612 Marchand, H., Wu, X.H., Ibbetson, J.P., Fini, P.T., Kozodoy, P., Keller, S., Speck, J.S., DenBaars, S.P. and Mishra, U.K. (1998) Applied Physics Letters, 73, 747. 613 Watanabe, H., Kuroda, N., Sunakawa, H. and Usui, A. (2000) Scanning reflection electron microscopy study of surface defects in GaN films formed by epitaxial lateral overgrowth. Applied Physics Letters, 77 (12), 1786–1788, 614 Zheleva, T.S., Ashmawi, W.M., Nam, O.-H., and Davis, R.F. (1999) Applied Physics Letters, 74, 2492. 615 Kuwano, N., Horichi, K., Kagawa, K., Nishimoto, S. and Sueyoshi, M. (2002) Journal of Crystal Growth, 237–239, 1047. 616 Lilienthal-Weber, Z., Benamara, M., Swider, W., Washburn, J., Park, J., Grudowski, P.A., Eiting, C.J. and Dupuis, R.D. (1999) MRS Internet Journal of Nitride Semiconductor Research, 4S1, G4.6. 617 Liliental-Weber, Z. and Cherns, D. (2001) Journal of Applied Physiology, 89, 7833. 618 Hacke, P., Domen, K., Kuramata, A., Tanahashi, T. and Ueda, O. (2001) Applied Physics Letters, 76, 2547. 619 Sakai, A., Sunakawa, H. and Usui, A. (1998) Applied Physics Letters, 73, 481. 620 Sakai, A., Sunakawa, H., Kimura, A. and Usui, A. (2000) Applied Physics Letters, 76, 442. 621 Fini, P., Zhao, L., Moran, B., Hansen, M., Marchand, H., Ibbetson, J.P., DenBaars, S.P., Mishra, U.K. and Speck, J.S. (1999) Applied Physics Letters, 75, 1706.
622 Kato, Y., Kitamura, S., Hiramatsu, K. and Sawaki, N. (1994) Journal of Crystal Growth, 144, 133. 623 Usui, A., Sunakawa, H., Sakai, A. and Yamaguchi, A.A. (1997) Japanese Journal of Applied Physics, 36, L899. 624 Usui, A., Sunakawa, H., Kobayashi, K., Watanabe, H. and Mizuta, M. (2001) Materials Research Society Symposium Proceedings, 639, G5.6.1. 625 Hiramatsu, K., Matsushima, H., Shibata, T., Sawaki, N., Tadatomo, K., Okagawa, H., Ohuchi, Y., Honda, Y. and Matsue, T. (1998) Materials Research Society Symposium Proceedings, 482, 257. 626 Nataf, G., Beaumont, B., Bouille, A., Haffouz, S., Vaille, M. and Gibart, P. (1998) Journal of Crystal Growth, 192, 73. 627 Parillaud, O., Wagner, V., Buehlmann, H.J. and Ilegems, M. (1998) MRS Internet Journal of Nitride Semiconductor Research, 3, 40. 628 Holst, J., Kaschner, A., Hoffmann, A., Fischer, P., Bertram, F., Riemann, T., Christen, J., Hiramatsu, K., Shibata, T. and Sawaki, N. (1999) Applied Physics Letters, 75, 3647. 629 Bohyama, S., Yoshikawa, K., Naoi, H., Miyake, H., Hiramatsu, K., Iyechika, Y. and Maeda, T. (2002) Japanese Journal of Applied Physics, 41, 75. 630 Sakai, A., Sunakawa, H., Kimura, A. and Usui, A. (2000) Applied Physics Letters, 76, 442. 631 Kuan, T.S., Inoki, C.K., Hsu, Y., Harris, D.L., Zhang, R., Gu, S. and Kuech, T.F. (2000) MRS Internet Journal of Nitride Semiconductor Research, 5S1, W2.6. 632 Hearne, S., Chason, E., Han, J., Floro, J.A., Figiel, J., Hunter, J., Amano, H. and Tsong, I.S.T. (1999) Applied Physics Letters, 74, 356. 633 Zhang, R., Zhang, K., Hansen, D.M., Boleslawski, M.P., Chen, K.L., Lu, D.Q., Shen, B., Zheng, Y.D. and Kuech, T.F. (1999) MRS Internet Journal of Nitride Semiconductor Research, 4S1, G4.7.
References 634 Dunn, K.A., Babcock, S., Stone, D.S., Matyi, R.J., Zhang, L. and Kuech, T.F. (2000) MRS Internet Journal of Nitride Semiconductor Research, 5S1, W2.11. 635 Zhang, L., Gu, S.L., Zhang, R., Hansen, D.M., Boleslawski, M.P. and Kuech, T.F. (2002) Journal of Crystal Growth, 235, 115. 636 Kawaguchi, Y., Honda, Y., Matsushima, H., Yamaguchi, M., Hiramatsu, K. and Sawaki, N. (1998) Japanese Journal of Applied Physics, 37, L966. 637 Kung, P., Walker, D., Hamilton, M., Diaz, J. and Razeghi, M. (1999) Applied Physics Letters, 74, 570. 638 Mao, Z., McKernan, S., Barry Carter, C., Yang, W. and McPherson, S.A. (1999) MRS Internet Journal of Nitride Semiconductor Research, 4S1, G3.13. 639 Yang, W., McPherson, S.A., Mao, Z., McKerman, S. and Carter, B. (1999) Journal of Crystal Growth, 204, 273. 640 Tanaka, S., Kawaguchi, Y., Sawaki, N., Hibino, M. and Hiramatsu, K. (2000) Applied Physics Letters, 76, 2701. 641 Honda, Y., Kawaguchi, Y., Ohtake, Y., Tanaka, S., Yamaguchi, M. and Sawaki, N. (2001) Journal of Crystal Growth, 230, 346. 642 Feltin, E., Beaumont, B., Vennegues, P., Vaille, M., Gibart, P., Riemann, T., Christen, J., Dobos, L. and Pecz, B. (2003) Epitaxial lateral overgrowth of GaN on Si (1 1 1). Journal of Applied Physics, 93 (1), 182–185. 643 Tanaka, S., Honda, Y., Sawaki, N. and Hibino, M. (2001) Applied Physics Letters, 79, 955. 644 Fang, S., Adomi, K., Iyer, S., Morkoc, H., Zabel, H., Choi, C. and Otsuka, N. (1990) Gallium arsenide on silicon: epitaxial growth and characterization. Journal of Applied Physics Review, 68, R31–R38. 645 Kobayashi, N.P., Kobayashi, J.T., Zhang, X., Dapkus, P.D. and Rich, D.H. (1999) Applied Physics Letters, 74, 2836. 646 Linthicum, K.J., Gehrke, T., Thomson, D., Carlson, E., Rajagopal, P., Smith, T.
647
648
649
650
651
652
653
654
655
656
657
658
and Davis, R. (1999) Applied Physics Letters, 75, 196. Davis, R.F., Gehrke, T., Linthicum, K.J., Zheleva, T.S., Preble, E.A., Rajagopal, P., Zorman, C.A. and Mehregany, M. (2001) Journal of Crystal Growth, 225, 134–140. See, for example, Thomson, D.B., Gehrke, T., Linthicum, K.J., Rapagopal, P. and Davis, R.F. (1999) MRS Internet Journal of Nitride Semiconductor Research, 4S1, G3.37. Kim, I.-H., Sone, C., Nam, O.-H., Park, Y.-J. and Kim, T. (1999) Applied Physics Letters, 75, 4109. Gehrke, T., Linthicum, K.J., Rajagopal, P., Preble, E.A. and Davis, R.F. (2000) MRS Internet Journal of Nitride Semiconductor Research, 5S1, W2.4. Gehrke, T., Linthicum, K.J., Preble, E., Rajagopal, P., Ronning, C., Zorman, C., Mehregany, M. and Davis, R.F. (2000) Journal of Electronic Materials, 29, 306. Zheleva, T.S., Ashmawi, W.M. and Jones, K.A. (1999) Physica Status Solidi a: Applied Research, 176, 545. Zheleva, T.S., Smith, S.A., Thomson, D.B., Linthicum, K.J., Rajagopal, P. and Davis, R.F. (1999) Journal of Electronic Materials, 28, L5. Einfeldt, S., Roskowski, A.M., Preble, E.A. and Davis, R.F. (2002) Applied Physics Letters, 80, 953. Roskowski, A.M., Miraglia, P.Q., Preble, E.A., Einfeldt, S., Stiles, T., Davis, R.F., Schuck, J., Grober, R. and Schwarz, U. (2001) Physica Status Solidi a: Applied Research, 188, 729–732. Chen, Y., Schneider, R., Wang, S.Y., Kern, R.S., Chen, C.H. and Kuo, C.P. (1999) Applied Physics Letters, 75, 2062. Kidoguchi, I., Ishibashi, A., Sugahara, G. and Ban, Y. (2000) Air-bridged lateral epitaxial overgrowth of GaN thin films. Applied Physics Letters, 76 (25), 3768–3770. Kidoguchi, I., Ishibashi, A., Sugahara, G., Tsujimura, A. and Ban, Y. (2000) Improvement of crystalline quality of
j787
j 3 Growth and Growth Methods for Nitride Semiconductors
788
659
660
661
662
663
664 665
666
667
GaN films by air-bridged lateral epitaxial overgrowth. Japanese Journal of Applied Physics, 39, L453–L456. Ishibashi, A., Kidoguchi, I., Sugahara, G. and Ban, Y. (2001) High quality GaN films by air-bridged lateral epitaxial overgrowth. Journal of Crystal Growth, 221, 338–344. Davis, R.F., Gehrke, T., Linthicum, K.J., Zheleva, T.S., Rajagopal, P., Zorman, C.A. and Mehregany, M. (2000) Pendeoepitaxial growth and characterization of GaN and related materials on 6H-SiC (0 0 0 1) and Si(1 1 1) substrates. GaN and Related Alloys – 1999 Symposium, Materials Research Society, Materials Research Society Symposium Proceedings, 595, W.2.1.1–W.2.1.10. Fini, P., Zhao, L., Moran, B., Hansen, M., Marchand, H., Ibbetson, J.P., DenBaars, S.P., Mishra, U.K. and Speck, J.S. (1999) High-quality coalescence of laterally overgrown GaN stripes on GaN/ sapphire seed layers. Applied Physics Letters, 75 (12), 1706–1708. Davis, R.F., Nam, O.-H., Zheleva, T.S., Bremser, M.D., Linthicum, K.J., Gehrke, T., Rajagopal, P. and Thomson, D.B. (1999) Lateral- and pendeo-epitaxial growth and characterization of low defect density GaN thin films. Proceedings of the Conference on Microscopy of Semiconducting Materials, Oxford University, pp.335–344. Sugiura, L., Ishikawa, M., Nunoue, S., Onomura, M. and Yamamoto, M. (2000) US Patent 6,015,979. Ikeda, M. (2000) US Patent 6,111,277. Beaumont, B., Gibart, P., Vaille, M., Haffouz, S., Nataf, G. and Bouille, A. (1998) Journal of Crystal Growth, 189–190, 97. Vennegues, P., Beaumont, B., Bousquet, V., Vaille, M. and Gibart, P. (2000) Journal of Applied Physiology, 87, 4175. Beaumont, B., Bousquet, V., Vennegues, P., Vaille, M., Bouille, A., Gibart, P.,
668
669
670
671
672
673
674
Dassonneville, S., Amokrane, A. and Sieber, B. (1999) Physica Status Solidi a: Applied Research, 176, 567. Bousquet, V., Vennegues, P., Beaumont, B., Vaille, M. and Gibart, P. (1999) Physica Status Solidi b: Basic Research, 216, 691. Hiramatsu, K., Nishiyama, K., Onishi, M., Mizutani, H., Narukawa, M., Motogaito, A., Miyake, H., Iyechika, Y. and Maeda, T. (2000) Journal of Crystal Growth, 221, 316. Hirth, J. and Lothe, J (1982) Theory of Dislocations, John Wiley & Sons, Inc., New Tork. Bertram, F., Riemann, T., Christen, J., Kaschner, A., Hoffmann, A., Hiramatsu, K., Shibata, T. and Sawaki, N. (1999) Epitaxial lateral overgrowth of GaN structures: spatially resolved characterization by cathodoluminescence microscopy and micro-Raman spectroscopy. Materials Science & Engineering B: Solid State Materials for Advanced Technology, 59, (1–3), 117–121. Gradecak, S., Wagner, V., Ilegems, M., Riemann, T., Christen, J. and Stadelmann, P. (2002) Microscopic evidence of point defect incorporation in laterally overgrown GaN. Applied Physics Letters, 80 (16), 2866–2628. Wagner, V., Parillaud, O., B€ uhlmann, H.J., Ilegems, M., Gradecak, S., Stadelmann, P., Riemann, T. and Christen, J. (2002) Influence of the carrier gas composition on morphology, dislocations, and microscopic luminescence properties of selectively grown GaN by hydride vapor phase epitaxy. Journal of Applied Physiology, 92 (3), 1307–1316; See also Wagner, V. (2001) Thesis, Ecole Polytechnique F_ed_erale de Lausanne, Lausanne. Nagahama, S.-I., Iwasa, N., Senoh, M., Matsushita, T., Sugimoto, Y., Kiyoku, H., Kozaki, T., Sano, M., Matsumura, H., Umemoto, H., Chocho, K. and Mukai, T.
References
675
676
677
678
679
680
681
682
683
684
685
686
(2000) Japanese Journal of Applied Physics, 39, L647. Miyake, H., Mizutani, H., Hiramatsu, K., Iyechika, Y., Honda, Y. and Maeda, T. (2001) Materials Research Society Symposium Proceedings, 639, G5.3.1. Riemann, T., Christen, J., Beaumont, B., Faurie, J.-P. and Gibart, P. (2004) Selforganized domain formation in low dislocation density GaN. Superlattices and Microstructures, 36, 833–847. Ishida, M., Ogawa, M., Orita, K., Imafuji, O., Yuri, M., Sugino, T. and Itoh, K. (2000) Journal of Crystal Growth, 221, 345. Smart, J.A., Chumbes, E.M., Schremer, A.T., and Shealy, J.R. (1999) Applied Physics Letters, 75, 3820. Ashby, C.I.H., Mitchell, C.C., Han, J., Missert, N.A., Provencio, P.P., Follstaedt, D.M., Peake, G.M. and Griego, L. (2000) Applied Physics Letters, 77, 3233. Katona, T.M., Craven, M.D., Fini, P.T., Speck, J.S. and DenBaars, S.P. (2001) Applied Physics Letters, 79, 2907. Sano, S., Detchprohm, T., Mochizuki, S., Kamiyama, S., Amano, H. and Akasaki, I. (2002) Journal of Crystal Growth, 235, 129. Detchprohm, T., Yano, M., Sano, S., Nakamura, R., Mochiduki, S., Nakamura, T., Amano, H. and Akasaki, I. (2001) Japanese Journal of Applied Physics, 40, L16. Detchprohm, T., Yano, M., Nakamura, R., Sano, S., Mochiduki, S., Nakamura, T., Amano, H. and Akasaki, I. (2001) Materials Research Society Symposium Proceedings, 639, G5.7.1. Strittmatter, A., Rodt, S., ReiÔmann, L., Bimberg, D., Schr€oder, H., Obermeier, E., Riemann, T., Christen, J. and Krost, A. (2001) Applied Physics Letters, 78, 727. Zhang, X., Li, R.R., Dapkus, P.D. and Rich, D.H. (2000) Applied Physics Letters, 77, 2213. Kobayashi, K., Yamaguchi, A.A., Kimura, S., Sunakawa, H., Kimura, A.
687 688
689
690
691
692
693
694
695
696
697
698
and Usui, A. (1999) Japanese Journal of Applied Physics, 38, L611. Li, X., Bishop, S.G. and Coleman, J.J. (1998) Applied Physics Letters, 73, 1179. Rosner, S.J., Girolami, G., Marchand, H., Fini, P.T., Ibbetson, J.P., Zhao, L., Keller, S., Mishra, U.K., DenBaars, S.P. and Speck, J.S. (1999) Applied Physics Letters, 74, 2035. Freitas, J.A., Jr, Nam, O.-H., Davis, R.F., Saparin, G.V. and Obyden, S.K. (1998) Applied Physics Letters, 72, 2990. Li, X., Bohn, P.W., Kim, J., White, J.O. and Coleman, J.J. (2000) Applied Physics Letters, 76, 3031. Kuwano, N., Horichi, K., Kagawa, K., Nishimoto, S. and Sueyoshi, M. (2002) Journal of Crystal Growth, 237–239, 1047. Zheleva, T.S., Ashmawi, W.M., Nam, O.-H. and Davis, R.F. (1999) Applied Physics Letters, 74, 2492. Liu, Q., Hoffmann, A., Kaschner, A., Thomsen, C., Christen, J., Veit, P. and Clos, R. (2000) Japanese Journal of Applied Physics, 39, L958. Liu, Q.K.K., Hoffman, A., Siegle, H., Kaschner, A., Thomsen, C., Christen, J. and Bertram, F. (1999) Applied Physics Letters, 74, 3122. Kaschner, A., Hoffmann, A., Thomsen, C., Bertram, F., Riemann, T., Christen, J., Hiramatsu, K., Shibata, T. and Sawaki, N. (1999) Optical microscopy of electronic and structural properties of epitaxial laterally overgrown GaN. Applied Physics Letters, 74, 3320–3322. Yoshimoto, M., Saraie, J. and Nakamura, S. (2001) Japanese Journal of Applied Physics, 40, L386. Bertram, F., Riemann, T,. Christen, J., Kaschner, A., Hoffmann, A., Thomsen, C., Hiramatsu, K., Shibata, T. and Sawaki, N. (1999) Applied Physics Letters, 74, 359. Pophristic, M., Long, F.H., Schurman, M., Ramer, J. and Ferguson, I.T. (1999) Applied Physics Letters, 74, 3519.
j789
j 3 Growth and Growth Methods for Nitride Semiconductors
790
699 Matthews, M.J., Hsu, J.W.P., Gu, S. and Kuech, T.F. (2001) Applied Physics Letters, 79, 3086. 700 Hsu, J.W.P., Matthews, M.J., AbuschMagder, D., Kleiman, R.N., Lang, D.V., Richter, S., Gu, S.L. and Kuech, T.F. (2001) Applied Physics Letters, 79, 761. 701 Gibart, P., Beaumont, B. and Chua, S.-J. (1999) Journal of Crystal Growth, 201–202, 365. 702 Popovici, G., Kim, W., Botchkarev, A., Tang, H., Solomon, J. and Morkoc, H. (1997) Impurity contamination of GaN epitaxial films from sapphire, SiC, and ZnO substrates. Applied Physics Letters, 71, 3385–3387. 703 Li, X., Bohn, P.W. and Coleman, J.J. (1999) Applied Physics Letters, 75, 4049. 704 Cho, Y.-H., Kim, H.M., Kang, T.W., Song, J.J. and Yang, W. (2002) Applied Physics Letters, 80, 1141. 705 Elsner, J., Kaukonen, M., Heggie, M.I., Haugk, M., Frauenheim, Th. and Jones, R. (1998) Physical Review, 58, 15357. 706 Zubia, D., Zaidi, S.H., Brueck, S.R.J. and Hersee, S.D. (2000) Nanoheteroepitaxial growth of GaN on Si by organometallic vapor phase epitaxy. Applied Physics Letters, 76 (7), 858–860. 707 Goldstein, A. (1997) Handbook of Nanophase Materials, Marcel Dekker, New York, pp. 3–4. 708 Zubia, D. and Hersee, S.D. (1999) Nanoheteroepitaxy: the application of nanostructuring and substrate compliance to the heteroepitaxy of mismatched semiconductor materials. Journal of Applied Physics, 85 (9), 6492–42961. 709 Haffouz, S., Beaumont, B., Vennegues, P. and Gibart, P. (1999) Physica Status Solidi a: Applied Research, 176, 677. 710 Wang, T., Morishima, Y., Naoi, N. and Sakai, S. (2000) A new method for a great reduction of dislocation density in a GaN layer grown on a sapphire substrate. Journal of Crystal Growth, 213, 188–192.
711 Oshima, Y., Eri, T., Shibata, M., Sunakawa, H., Kobayashi, K., Ichihashi, T. and Usui, A. (2003) Preparation of freestanding GaN wafers by hydride vapor phase epitaxy with void-assisted separation. Japanese Journal of Applied Physics, Part 2: Letters, 42 (1A/B), L1–L3. 712 Usui, A., Ichihashi, T., Kobayashi, K., Sunakawa, H., Oshima, Y., Eri, T. and Shibata, M. (2002) Role of TiN film in the fabrication of freestanding GaN wafers using hydride vapor phase epitaxy with void-assisted separation. Physica Status Solidi a: Applied Research, 194 (2), 572–575. 713 Oshima, Y., Eri, T., Shibata, M., Sunakawa, H., Kobayashi, K., Ichihashi, T. and Usui, A. (2003) Preparation of freestanding GaN wafers by hydride vapor phase epitaxy with void-assisted separation. Japanese Journal of Applied Physics, Part 2: Letters, 42 (1A/B), L1–L3. 714 Sagar, A., Feenstra, R.M., Inoki, C.K., Kuan, T.S., Fu, Y., Moon, Y.T., Yun, F. and Morkoc, H. (2005) Physica Status Solidi a: Applied Research, 202, 722. 715 Fu, Yi, Yun, F., Moon, Y.T., Xie, J.Q., € € Dogan, S., Morkoc, H., Inoki, Ozgur, U., C.K., Kuan, T.S., Zhou, L. and Smith, D.J. (2005) Reduction of threading dislocations in GaN overgrowth by OMVPE on TiN micro-porous network templates. Applied Physics Letters, 86 (4), 43108-1-3. 716 Tanaka, S., Iwai, S. and Aoyagi, Y. (1996) Self-assembling GaN quantum dots on AlxGa1xN surfaces using a surfactant. Applied Physics Letters, 69 (26), 4096–4098. 717 Tanaka, S., Hirayama, H., Iwai, S. and Aoyagi, Y. (1997) Materials Research Society Symposium Proceedings, 449, 135. 718 Ramvall, P., Tanaka, S., Nomura, S., Riblet, P. and Aoyagi, Y. (1999) Confinement induced decrease of the exciton-longitudinal optical phonon coupling in GaN quantum dots. Applied Physics Letters, 75 (13), 1935–1937.
References 719 Ramvall, P., Riblet, P., Nomura, S., Aoyagi, Y. and Tanaka, S. (2000) Optical properties of GaN quantum dots. Journal of Applied Physiology, 87 (8), 3883–3890. 720 Wang, T., Liu, Y.H., Lee, Y.B., Izumi, Y., Li, H.D., Bai, J. and Sakai, S. (2002) Journal of Crystal Growth, 235, 177. 721 Lee, Y.-B., Wang, T., Liu, Y.-H., Ao, J.-P., Izumi, Y., Lacroix, Y., Li, H.-D., Bai, J., Nao, Y. and Sakai, S. (2002) Highperformance 348 nm AlGaN/GaN-based ultraviolet-light-emitting diode with a SiN buffer layer. Japanese Journal of Applied Physics, Part 1: Regular Papers, Short Notes & Review Papers, 41 (7A), 4450–4453. 722 Sagar, A., Feenstra, R.M., Inoki, C.K., Kuan, T.S., Fu, Yi, Moon, Y.T. and Morkoc, H. (2005) Dislocation density reduction in GaN using porous SiN interlayers. International Workshop on Nitride Semiconductors, 2004, Pittsburgh, USA, Physica Status Solidi a: Applied Research, 202 (5), 722–726. 723 Yun, F., Moon, Y.-T., Fu, Y., Zhu, K., Morkoc, H., Inoki, C.K., Kuan, T.S., Sagar, A. and Feenstra, R.M. (2005) Efficacy of single and double SiNx interlayers on defect reduction in GaN overlayers grown by organometallic vapor-phase epitaxy. Journal of Applied Physiology, 98 (12), 123502-1-8. 724 Moon, Y.T., Xie, J., Liu, C., Fu, Y., Ni, X., Biyikli, N., Zhu, K., Yun, F., Morkoc, H., Sagar, A. and Feenstra, R.M. (2006) Morphology of GaN seed layers on in-situ deposited porous SiN and its effect on properties of overgrown GaN epilayers. Journal of Crystal Growth, 291 (1), 301–308. € ur, U., € Fu, Y., Moon, Y.T., Yun, F., 725 Ozg€ Morkoc, H. and Everitt, H.O. (2005) Increased carrier lifetimes in GaN epitaxial layers grown using SiN and TiN porous network templates. Journal of Applied Physiology, 97, 103704. € ur, U., € Fu, Y., Ni, X., Morkoc, 726 Xie, J., Ozg€ H., Inoki, C.K., Kuan, T.S., Freeman, J.V.
727
728
729
730
731
732
733
734
735
and Everitt, H.O. (2007) Low dislocation densities and long carrier lifetimes in GaN thin films on SiNx nanonet. Applied Physics Letters, 90 (4), 41107-1–41107-3. Tsen, K.T., Liang, W., Ferry, D.K., Lu, H., € ur, U., € Fu, Y., Moon, William, J.S., Ozg€ Y.T., Yun, F., Morkoc, H. and Everitt, H.O. (2005) Optical studies of carrier dynamics and non-equilibrium optical phonons in nitride-based wide bandgap semiconductors. Superlattices and Microstructures, 38 (2), 77–114. Xie, J., Chevtchenko, S.A. and Morkoc, H. (2007) Defect reduction in GaN epilayers grown by MOCVD with in situ SiNx nano-network. Applied Physics Letters, 90, 262112. Kwon, H.K., Eiting, C.J., Lambert, D.J.H., Wong, M.M., Dupuis, R.D., Liliental-Weber, Z. and Benamara, M. (2000) Applied Physics Letters, 77, 2503. Bunea, G.E., Herzog, W.D., Unlu, M.S., Goldberg, B.B., and Molnar, R.J. (1999) Applied Physics Letters, 75, 838. Jurs?_enas, S., Miasojedovas, 1S, Kurilcik, G., Zukauskas, A. and Hageman, P.R. (2004) Physica Status Solidi a: Applied Research, 201, 199. Sampath, A.V., Garrett, G.A., Collins, C.J., Boyd, P., Choe, J., Newman, P.G., Shen, H., Wraback, M., Molnar, R.J. and Caissie, J. (2004) Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 22, 1487. Jurs?_enas, S., Kurilcik, N., Kurilcik, G., Zukauskas, A., Prystawko, P., Leszcynski, M., Suski, T., Perlin, P., Grzegory, I. and Porowski, S. (2001) Applied Physics Letters, 78, 3776. Chichibu, S.F., Marchand, H., Minsky, M.S., Keller, S., Fini, P.T., Ibbetson, J.P., Fleischer, S.B., Speck, J.S., Bowers, J.E., Hu, E., Mishra, U.K., DenBaars, S.P., Deguchi, T., Sota, T. and Nakamura, S. (1999) Applied Physics Letters, 74, 1460. Izumi, T., Narukawa, Y., Okamoto, K., Kawakami, Y., Fujita, Sg. and Nakamura,
j791
j 3 Growth and Growth Methods for Nitride Semiconductors
792
736
737
738
739
740
741
742
743 744
745
746
S. (2000) Journal of Luminescence, 87–89, 1196. Miyake, H., Bohyama, S., Fukui, M., Hiramatsu, K., Iyechika, Y. and Maeda, T. (2002) Journal of Crystal Growth, 237–239, 1055. Kim, M.H., Choi, Y., Yi, J., Yang, M., Jeon, J., Khym, S. and Leem, S.-J. (2001) Applied Physics Letters, 79, 1619. € ur, U., € Fu, Y., Moon, Y.T., Yun, F., Ozg€ Morkoc, H., Everitt, H.O., Park, S.S. and Lee, K.Y. (2005) Long carrier lifetimes in GaN epitaxial layers grown using TiN porous network templates. Applied Physics Letters, 86, 232106. Hacke, P., Detchprohm, T., Hiramatsu, K., Sawaki, N., Tadatomo, K. and Miyake, K. (1994) Journal of Applied Physiology, 76, 304. Haase, D., Schmid, M., K€ urner, W., D€ornen, A., H€arle, V., Scholz, F., Burkard, M. and Schweizer, H. (1996) Applied Physics Letters, 69, 2525. Fang, Z.-Q., Look, D.C., Kim, W., Fan, Z., Botchkarev, A. and Morkoc, H. (1998) Applied Physics Letters, 72, 2277. Lee, W.I., Huang, T.C., Guo, J.D. and Feng, M.S. (1995) Applied Physics Letters, 67, 1721. Johnstone, D. (2007) Proceedings of SPIE, 6473, 64730L. Kawaguchi, Y., Nambu, S., Sone, H., Yamaguchi, M., Miyake, H., Hiramatsu, K., Sawaki, N., Iyechika, Y. and Maeda, T. (1999) Selective area growth (SAG) and epitaxial lateral overgrowth (ELO) of GaN using tungsten mask. Proc of Materials Research Society, 4S1, G4.1. Hiramatsu, K., Haino, M., Yamaguchi, M., Miyake, H., Motogaito, A., Sawaki, N., Iyechika, Y. and Maeda, T. (2001) GaN layer structures with buried tungsten nitrides (WNx) using epitaxial lateral overgrowth via OMVPE. Materials Science & Engineering B: Solid State Materials for Advanced Technology, 82, 62–64. Amano, H., Iwaya, M., Kashima, T., Katsuragawa, M., Akasaki, I., Han, J.,
747
748
749
750
751
752
753
754 755
756
757
Hearne, S., Floro, J.A., Chason, E. and Figiel, J. (1998) Japanese Journal of Applied Physics, 37, L1540, Iwaya, M., Takeuchi, T., Yamaguchi, S., Wetzel, C., Amano, H. and Akasaki, I (1998) Japanese Journal of Applied Physics, 37, L316. Amano, H., Iwaya, M., Kashima, T., Katsuragawa, M., Akasaki, I., Han, J., Hearne, S., Floro, J.A., Chason, E. and Figiel, J. (1998) Japanese Journal of Applied Physics, 37, L1540. Yang, C.-C., Wu, M.-C., Chang, C.-A. and Chi, G.-C. (1999) Journal of Applied Physiology, 85, 8427–8431. Yang, C.-C., Wu, M.-C. and Chi, G.-C. (1999) Journal of Applied Physiology, 86, 6120–6123. B€ottcher, T., Einfeldt, S., Figge, S., Chierchia, R., Heinke, H., Hommel, D. and Speck, J.S. (2001) Applied Physics Letters, 78, 1976. Bourret-Courchesne, E.D., Kellermann, S., Yu, K.M., Benamara, M., LilientalWeber, Z., Washburn, J., Irvine, S.J.C. and Stafford, A. (2000) Applied Physics Letters, 77, 3562–3564. Benamara, M., Liliental-Weber, Z., Kellermann, S., Swider, W., Washburn, J., Mazur, J. and Bourret-Courchesne, E.D. (2000) Journal of Crystal Growth, 218, 447. Koike, M. (2000) private communication. Koleske, D.D., Twigg, M.E., Wickenden, A.E., Henry, R.L., Gorman, R.J. Freitas, J.A., Jr and Fatemi, M. (1999) Applied Physics Letters, 75, 3141. Huang, D., Visconti, P., Jones, K.M., Reshchikov, M.A., Yun, F., Baski, A.A., King, T. and Morkoc, H. (2001) Dependence of GaN polarity on the parameters of the buffer layer grown by molecular beam epitaxy. Applied Physics Letters, 78, 4145–4147. Hellman, E.S. (1998) MRS Internet Journal of Nitride Semiconductor Research, 3, 11.
References 758 Rouviere, J.L., Weyher, J.L., SeelmannEggebert, M. and Porowski, S. (1998) Applied Physics Letters, 73, 668. 759 Daudin, B., Rouviere, J.L. and Arley, M. (1996) Applied Physics Letters, 69, 2480. 760 Kazimirov, A., Scherb, G., Zegenhagen, J., Lee, T.L., Bedzyk, M.J., Kelly, M.K., Angerer, H. and Ambacher, O. (1998) Journal of Applied Physiology, 84, 1703. 761 Sumiya, M., Tanaka, M., Ohtsuka, K., Fuke, S., Ohnishi, T., Ohkuboi, I., Yoshimoto, M., Koinuma, H. and Kawasaki, M. (1999) Applied Physics Letters, 75, 674. 762 Suyima, M., Yoshimura, K., Ohtsuka, K. and Fuke, S. (2000) Applied Physics Letters, 76, 2098. 763 Smith, A.R., Feenstra, R.M., Greve, D.W., Shin, M.S., Skowronski, M., Neugebauer, J. and Northrup, J.E. (1999) Surface Science, 423, 70. 764 Smith, A.R., Feenstra, R.M., Greve, D.W., Shin, M.-S., Skowronski, M., Neugebauer, J. and Northrup, J.E. (1998) Applied Physics Letters, 72, 2114. 765 Shen, X.-Q., Ide, T., Cho, S.-H., Shimizu, M., Hara, S., Okumura, H., Sonoda, S. and Shimizu, S. (2000) Japanese Journal of Applied Physics, 39, L16–L18. 766 Dmitrov, R., Murphy, M., Smart, J., Schaff, W., Shealy, J.R., Eastman, L.F., Ambacher, O. and Stutzmann, M. (2000) Journal of Applied Physiology, 87, 3375. 767 Ambacher, O., Smart, J., Shealy, J.R., Weimann, N.G., Chu, K., Eastman, L.F., Dimitrov, R., Wittmer, L., Stutzmann, M., Rieger, W. and Hilsenbeck, J. (1999) Journal of Applied Physiology, 85, 3222. 768 Jones, K.M., Visconti, P., Yun, F., Baski, A.A. and Morkoc, H. (2001) Applied Physics Letters, 78 (17), 2497–2499. 769 Ponce, F.A., Bour, D.P., Young, W.T., Saunders, M. and Steeds, J.W. (1996) Applied Physics Letters, 69, 337. 770 Visconti, P., Jones, K.M., Reshchikov, M.A., Cingolani, R., Morkoc, H. and Molnar, R.J. (2000) Applied Physics Letters, 77, 3532; Visconti, P., Jones,
771
772
773
774
775
776 777
778
779 780
781
782
K.M., Reshchikov, M.A., Yun, F., Cingolani, R., Morkoc, H., Park, S.S. and Lee, K.Y. (2000) Characteristics of free standing hydride vapor phase epitaxy grown GaN with very low defect concentration. Applied Physics Letters, 77, 3743. Sonoda, S., Shimizu, S., Shen, X.Q., Hara, S. and Okumura, H. (2000) Japanese Journal of Applied Physics, Part 2: Letters, 39, L202. Shen, X.-Q., Ide, T., Cho, S.-H., Shimizu, M., Hara, S., Okumura, H., Sonoda, S. and Shimizu, S. (2000) Japanese Journal of Applied Physics, 39, L16. Ramachandran, V., Feenstra, R.M., Sarney, W.L., Salamanca-Riba, L., Northrup, J.E., Romano, L.T. and Greve, D.W. (1999) Applied Physics Letters, 75, 808. Liliental-Weber, Z., Benamara, M., Swider, W., Washburn, J., Grzegory, I., Porowski, S., Lambert, D.J.H., Eiting, C.J. and Dupuis, R.D. (1999) Applied Physics Letters, 75, 4195. Widmann, F., Feuillet, G., Daudin, B. and Rouviere, J.L. (1999) Journal of Applied Phycology, 85, 1550. Hellman,E.S.(1998)MRS InternetJournal of Nitride Semiconductor Research, 3, 11. Held, R., Nowak, G., Ishaug, B.E., Seutter, S.M., Parkhomovsky, A., Dabiran, A.M., Cohen, P.I., Grzegory, I. and Porowski, S. (1999) Journal of Applied Physiology, 85, 7697. Bernardini, F., Fiorentini, V. and Vanderbilt, D. (1997) Physical Review B: Condensed Matter, 56, R10024. Morkoc, H., Cingolani, R. and Gil, B. (1999) Solid State Electronics, 43, 1909. Martin, G.A., Botchkarev, A., Rockett, A. and Morkoc, H. (1996) Applied Physics Letters, 68, 2541. Ramachandran, V., Lee, C.D., Feenstra, R.M., Smith, A.R. and Greve, G.W. (2000) Physical Review Letters, 84, 4014. Smith, A.R., Feenstra, R.M., Greve, D.W., Shin, M.S., Skowronski, M.,
j793
j 3 Growth and Growth Methods for Nitride Semiconductors
794
783
784
785
786
787
788
789
790
791
792
793 794
Neugebauer, J. and Northrup, J.E., (1998) Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 16, 2242. Huang, D., Visconti, P., Jones, K.M., Reshchikov, M.A., Yun, F., Baski, A.A., King, T. and Morkoc, H. (2001) Applied Physics Letters, 78, 4145. Widmann, F., Feuillet, G., Daudin, B. and Rouviere, J.L. (1999) Journal of Applied Phycology, 85 (3), 1550–1555. Feenstra, R.M., Chen, H., Ramachandran, V., Smith, A.R. and Greve, D.W. (2000) Applied Surface Science, 166, 165. Smith, A.R., Feenstra, R.M., Greve, D.W., Neugebauer, J. and Northrup, J.E. (1997) Physical Review Letters, 79, 3934. Papcewicz, K., Nardelli, M.B. and Bernholc, J. (1997) Physical Review B: Condensed Matter, 20, R12725. Xue, Q.K., Xue, Q.Z., Bakhtizin, R.Z., Hasegawa, Y., Tsong, I.S.T., Sakurai, T. and Ohno, T. (1999) Physical Review Letters, 82, 3074. Xue, Q.K., Xue, Q.Z., Kuwano, S., Sakurai, T., Ohno, T., Tsong, I.S.T., Qiu, X.G. and Segawa, Y. (2000) Thin Solid Films, 367, 149. Xue, Q.K., Xue, Q.Z., Kuwano, S., Nakayama, K., Sakurai, T., Tsong, I.S.T., Qiu, X.G. and Segawa, Y. (2001) Journal of Crystal Growth, 229, 41. Smith, A.R., Ramachandran, V., Feenstra, R.M., Greve, D.W., Shin, M.-S., Skowronski, M., Neugebauer, J. and Northrup, J.E. (1998) Journal of Vacuum Science & Technology, 16, 1641. Adelmann, C., Brault, J., MartinezGuerrero, E., Mula, G., Mariete, H., Dang, L.S. and Daudin, B. (2001) Physica Status Solidi a: Applied Research, 188, 575. Neugebauer, J. (2001) Physica Status Solidi b: Basic Research, 227, 93. Foxon, C.T., Cheng, T.S., Novikov, S.V., Korakakis, D., Jeffs, N.J., Grzegory, I.
795
796
797
798
799
800
801
802
803
804
805
and Porowski, S. (1999) Journal of Crystal Growth, 207, 1. Hughes, O.H., Cheng, T.S., Novikov, S.V., Foxon, C.T., Korakakis, D. and Jeffs, N.J. (1999) Journal of Crystal Growth, 201–202, 388. Foxon, C.T., Cheng, T.S., Novikov, S.V., Jeffs, N.J., Hughes, O.H., Melnik, Yu.V., Nikolaev, A.E. and Dmitriev, V.A. (1999) Surface Science, 421, 377. Iwata, K., Asahi, H., Yu, S.J., Asami, K., Fujita, H., Fushida, M. and Gonda, S. (1996) Japanese Journal of Applied Physiology, Part 2: Letters, 35, L289. Hughes, W.C., Rowland, W.H., Jr, Johnson, M.A.L., Fujita, S., Cook, J.W., Jr, Schetzina, J.F., Ren, J. and Edmond, J.A. (1995) Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 13, 1571. Ramachandran, V., Lee, C.D., Feenstra, R.M., Smith, A.R., Northrup, J.E. and Greve, D.E. (2000) Journal of Crystal Growth, 209, 355. Chen, H., Feenstra, R.M., Northrup, J.E., Zywietz, T., Neugebauer, J. and Greve, D.W. (2000) Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 18, 2284. Janzen, O., Hahn, Ch., Kampen, T.U. and M€onch, W. (1999) European Physical Journal B, 7, 1. Uchida, K., Watanabe, W., Yano, F., Kougucci, M., Tanaka, T. and Minegava, S. (1996) Proceedings of International Symposium on Blue Lasers and Light Emitting Diodes, March 5–7, Chiba University, Japan, p.48. Keller, S., Keller, B.P., Wu, Y.-F., Heying, B., Kaponek, D., Speck, J.S., Mishra, U.K. and DenBaars, S.P. (1996) Applied Physics Letters, 68, 1525. Grandjean, N., Massies, J. and Leroux, M. (1996) Applied Physics Letters, 69, 2071. Huang, D., Visconti, P., Jones, K.M., Reshchikov, M.A., Yun, F., Baski, A.A., King, T. and Morkoc, H. (2001) Applied Physics Letters, 78, 4145–4147.
References 806 Matsuoka, T., Yoshimoto, N., Sasaki, T. and Katsui, A. (1992) Journal of Electronic Materials, 21, 157. 807 Hamdani, F., Botchkarev, A., Kim, W., Salvador, A., Morkoc, H., Yeadon, M., Gibson, J.M., Tsen, S.C.T., Smith, D.J., Reynolds, D.C., Look, D.C., Evans, K., Litton, C.W., Mitchel, W.C. and Hemenger, P. (1997) Applied Physics Letters, 70, 467. 808 Vispute, R.D., Talyansky, V., Sharma, R.P., Choopun, S., Downes, M., Venkatesan, T., Li, Y.X., Salamanca-Riba, L.G., Iliadis, A.A., Jones, K.A., and McGarrity, J. (1998) Applied Surface Science, 127–129, 431. 809 Wang, R.P., Muto, H., Yamada, Y. and Kusumori, T. (2002) Thin Solid Films, 411, 69. 810 Ueda, T., Huang, T.F., Spruytte, S., Lee, H., Yuri, M., Itoh, K., Baba, T. and Harris, J.S., Jr (1998) Journal of Crystal Growth, 187, 340. 811 Gu, S., Zhang, R., Sun, J., Zhang, L. and Kuech, T.F. (2000) Applied Physics Letters, 76, 3454. 812 Hamdani, F., Yeadon, M., Smith, D.J., Tang, H., Kim, W., Salvaor, A., Botchkarev, A.E., Gibson, J.M., Plolyakov, A.Y., Skowronski, M. and Morkoc, H. (1998) Journal of Applied Physiology, 83, 983. 813 Matsuoka, T., Yoshimoto, N., Sasaki, T. and Katsui, A. Journal of Electronic Materials, (1992) 21, 157. 814 Xia, R.F., Sun, X.W., Li, Z.F., Kwok, H.S., Liu, Q.Z. and Lau, S.S. (1997) Journal of Vacuum Science & Technology A: Vacuum Surfaces and Films, 15, 2207. 815 Doolittle, W.A., Kang, S. and Brown, A. (2000) MBE growth of high quality GaN on LiGaO2 for high frequency, high power electronic applications. SolidState Electronics, 44 (2), 229–38. 816 Kung, P., Saxler, A., Zhang, X., Walker, D., Lavado, R. and Razeghi, M, (1996) Applied Physics Letters, 69, 2116. 817 Yun, F., Reshchikov, M.A., He, L., King, T., Huang, D., Morkoc, H., Nause, J.,
818
819
820
821
822
823
824
825
826
827
Cantwell, G., Maruska, H.P. and Litton, C.W. (2002) Comparative analysis of MBE-grown GaN films on SiC, ZnO and LiGaO2 substrates. Spring MRS Meeting, April, San Francisco, CA. Heying, B., Tarsa, E., Elsass, C., Fini, P., DenBaars, S. and Speck, J. (1999) Journal of Applied Physiology, 85, 6470. Duan, S., Teng, X., Han, P. and Lu, D. (1998) Growth and characterization of GaN on LiGaO2. Journal of Crystal Growth, 195, 304. Fischer, S., Gisbertz, A., Meyer, B.K., Topf, M., Koynov, S., Dirnstorfer, I., Volm, D., Uecker, R., Reiche, P., Ganschow, S. and Liliental-Weber, Z. (1996) Symposium Proceedings EGN-1, Rigi, Switzerland. Leszczynski, M., Beaumont, B., Frayssinet, E., Knap, W., Prystawko, P., Suski, T., Grzegory, I. and Porowski, S. (1999) Applied Physics Letters, 75 (9), 1276–1278. Ponce, F.A., Bour, D.P., Gotz, W., Johnson, N.M., Helava, H.I., Grzegory, I., Jun, J. and Porowski, S. (1996) Applied Physics Letters, 68, 917. Ponce, F.A., Bour, D.P., Young, W.T., Sounders, M. and Steeds, J.W. (1996) Applied Physics Letters, 69, 337. Gassmann, A., Susski, T., Newmann, N., Kiselowski, C., Jones, E., Weber, E.R., Liliental-Weber, Z., Rubin, M.D., Helava, H.I., Grezegory, I., Bockovski, M., Jun, J. and Porowski, S. (1996) Journal of Applied Physiology, 80, 2195. Pelzmann, A., Kirchner, C., Mayer, M., Schwegler, V., Schauler, M., Kamp, M., Ebeling, K.J., Grzegory, I., Leszczynski, M., Nowak, G. and Porowski, S. (1998) Journal of Crystal Growth, 189–190, 167. Manfra, M.J., Pfeiffer, L.N., West, K.W., Stormer, H.L., Baldwin, K.W., Hsu, J.W.P., Lang, D.V. and Molnar, R.J. (2000) Applied Physics Letters, 77, 2888. Miskys, C.R., Kelly, M.K., Ambacher, O., Martınez-Criado, G. and Stutzmann, M. (2000) GaN homoepitaxy by
j795
j 3 Growth and Growth Methods for Nitride Semiconductors
796
828
829
830
831
832
833
834
835
836
metalorganic chemical-vapor deposition on free-standing GaN substrates. Applied Physics Letters, 77, 1868–1870. Smith, D.J., Chandrasekhar, D., Sverdlov, B., Botchkarev, A., Salvador, A. and Morkoc, H. (1995) Applied Physics Letters, 67, 1830. Manfra, M.J., Pfeiffer, L.N., West, K.W., Stormer, H.L., Baldwin, K.W,. Hsu, J.W.P., Lang, D.V. and Molnar, R.J. (2000) Applied Physics Letters, 77, 2888. Prange, R.E. and Girvin, S.M. (1990) The Quantum Hall Effect, 2nd edn, Springer, New York. Macdonald, A.H. (1990) Quantum Hall Effect: A Perspective, Kluwer Academic Publishers. Sarma, S.D. and Pinczuk, A. (1996) Perspectives in Quantum Hall Effects: Novel Quantum Liquids in LowDimensional Structures. WileyInterscience. Reshchikov, M.A., Huang, D., Yun, F., He, L., Reynolds, D.C., Park, S.S., Lee, K.Y. and Morkoc, H. (2001) Photoluminescence of GaN grown by molecular beam epitaxy on freestanding GaN template. Applied Physics Letters, 79 (23), 3779–3378. Morkoc, H. (2001) Comprehensive characterization of hydride VPE grown GaN layers and templates. Materials Science & Engineering R: Reports, 259 (33/ 5–6), 1–73, and references therein. Detchprohm, T., Hiramatsu, K., Sawaki, N. and Akasaki, I. (1994) Journal of Crystal Growth, 137, 170; See also Detchprohm, T., Amano, H., Hiramatsu, K. and Akasaki, I. (1992) Applied Physics Letters, 61, 2688; Detchprohm, T,. Hiramatsu, K., Itoh, K. and Akasaki, I. (1992) Japanese Journal of Applied Physics, 31, L1454. Johnson, M.A.L., Fujita, S., Rowland, W.H., Jr, Hughes, W.C., He, Y.W., ElMasry, N.A., Cook, J.W., Jr, Schetzina, J.F., Ren, J. and Edmond, J.A. (1996) Journal of Electronic Materials, 25, 793.
837 Leszczynski, M., Beaumont, B., Frayssinet, E., Knap, W., Prystawko, P., Suski, T., Grzegory, I. and Porowski, S. (1999) Applied Physics Letters, 75 (9), 1276–1278. 838 Miskys, C.R., Kelly, M.K., Ambacher, O., Martınez-Criado, G. and Stutzmann, M. (2000) GaN homoepitaxy by metalorganic chemical-vapor deposition on free-standing GaN substrates. Applied Physics Letters, 77, 1858–1860. 839 Kirchner, C., Schwegler, V., Eberhard, F., Kamp, M., Ebeling, K.J., Prystawko, P., Leszczynski, M., Grzegory, I. and Porowski, S. (2000) OMVPE homoepitaxy of high-quality GaN: crystal growth and devices. Progress in Crystal Growth & Characterization of Materials, 41 (1–4), 57–83. 840 Komitzer, K., Ebner, T., Thonke, K., Sauer, R., Kirchner, C., Kamp, M., Schwegler, V., Leszczynski, M., Grzegory, I. and Porowski, S. (1999) Photoluminescence and reflectance spectroscopy of excitonic transitions in high-quality homoepitaxial GaN films. Physical Review B: Condensed Matter, 60 (3), 1471–1473. 841 Kirchner, C., Schwegler, V., Eberhard, F., Kamp, M., Ebeling, KJ., Kornitzer, K., Ebner, T., Thonke, K., Sauer, R., Prystawko, P., Leszczynski, M., Grzegory, I. and Porowski, S. (1999) Homoepitaxial growth of GaN by metalorganic vapor phase epitaxy: a benchmark for GaN technology. Applied Physics Letters, 75 (8), 1098–100. 842 Neugebauer, J. and Van de Walle, C.G. (1996) Applied Physics Letters, 69, 503. 843 Reshchikov, M.A., Visconti, P. and Morkoc, H. (2001) Applied Physics Letters, 78, 177. 844 Reshchikov, M.A., Shahedipour, F., Korotkov, R.Y., Wessels, B.W. and Ulmer, M.P. (2000) Journal of Applied Physiology, 87, 3351. 845 Kornitzer, K., Grehl, M., Thonke, K., Sauer, R., Kirchner, C., Schwegler, V.,
References
846 847
848
849
850
851
852
853
854
855
Kamp, M., Leszczynski, M., Grzegory, I. and Porowski, S. (1999) Physica B, 273–274, 66. Meyer, B.K. (1997) Materials Research Society Symposium Proceedings, 449, 497. Dean, P.J., Cuthbert, J.D., Thomas, D.G. and Lynch, R.T. (1967) Physical Review Letters, 18, 122. Scromme, B.J., Jayapalan, J., Vaudo, R.P. and Phanse, V.M. (1999) Applied Physics Letters, 74, 2358. Volm, D., Oettinger, K., Streibl, T., Kovalev, D., Ben-Chorin, M., Diener, J., Meyer, B.K., Majewski, J., Eckey, L., Hoffmann, A., Amano, H., Akasaki, I., Hiramatsu, K. and Detchprohm, T. (1996) Physical Review B: Condensed Matter, 53, 16543. Kuramata, A., Horino, K., Domen, K., Shinohora, K. and Tanahashi, T. (1995) Applied Physics Letters, 67, 2521. Sun, C.I., Yang, J.W., Chen, Q., Khan, M.A., George, T., Chang-Chien, P. and Mahajan, S. (1996) Applied Physics Letters, 68, 1129. Miller, D.A.B. and Chemla, D.S. (1984) Band-edge electroabsoption in quantum well structures: the quantum-confined Stark effect. Physical Review Letters, 53, 2173. Waltereit, P., Brandt, O., Trampert, A., Grahn, H.T., Menniger, J., Ramsteiner, M., Reiche, M. and Ploog, K.H. (2000) Nitride semiconductors free of electrostatic fields for efficient white light-emitting diodes. Nature, 406 (6798), 865–868. See, for example, Langer, R., Simon, J., Ortiz, V., Pelekanos, N.T., Barski, A., Andre, R. and Godlewski, M. (1999) Giant electric fields in unstrained GaN single quantum wells. Applied Physics Letters, 74 (25), 3827–3829. Deguchi, T., Sekiguchi, K., Nakamura, A., Sota, T., Matsuo, R., Chichibu, S. and Nakamura, S. (1999) Quantum-confined Stark effect in an AlGaN/GaN/AlGaN single quantum well structure. Japanese
856
857
858
859
860
861
862
863
864
Journal of Applied Physics, Part 2: Letters, 38 (8B), L914–L916. Tran, C.A. (2006) private communication, SemiLEDs, Chu-Nan, Taiwan. Sharma, R., Pattison, P.M., Masui, H., Farrell, R.M., Baker, T.J., Haskell, B.A., Wu, F., DenBaars, S.P., Speck, J.S. and Nakamura, S. (2005) Demonstration of a semipolar (1013) InGaN/GaN green light emitting diode. Applied Physics Letters, 87, 231110-1. Sasaki, T. and Zembutsu, S. (1987) Substrate-orientation dependence of GaN single-crystal films grown by metalorganic vapor-phase epitaxy. Journal of Applied Physics, 61 (7), 2533–2540. Sun, C.J. and Razeghi, M. (1993) Comparison of the physical properties of GaN thin films deposited on (0 0 0 1) and (0 112) sapphire substrates. Applied Physics Letters, 63 (7), 973–975. Metzger, T., Angerer, H., Ambacher, O., Stutzmann, M. and Born, E. (1996) X-ray diffraction study of gallium nitride grown by OMVPE. Physica Status Solidi b: Basic Research, 193 (2), 391–397. Lei, T., Ludwig, K.F. and Moustakas, T.D. (1993) Heteroepitaxy, polymorphism, and faulting in GaN thin films on silicon and sapphire substrates. Journal of Applied Physiology, 74 (7), 4430–4437. Eddy, C.R., Moustakas, T.D. and Scanlon, J. (1993) Growth of gallium nitride thin films by electron-cyclotron resonance microwave plasma-assisted molecular-beam epitaxy. Journal of Applied Physiology, 73 (1), 448–455. Moustakas, T.D., Lei, T. and Molnar, R.J. (1993) Growth of GaN by ECR-assisted MBE. Physica B, 185 (1–4), 36–49. Iwata, K., Asahi, H., Asami, K., Kuroiwa, R. and Gonda, S. (1997) Gas source molecular beam epitaxy growth of GaN on C-, A-, R- and M-plane sapphire and silica glass substrates. Japanese Journal of
j797
j 3 Growth and Growth Methods for Nitride Semiconductors
798
865
866
867
868
869
870
871
872
Applied Physics, Part 2: Letters, 36 (6A), L661–L664. Tripathy, S., Soni, R.K., Asahi, H., Iwata, K., Kuroiwa, R., Asami, K. and Gonda, S. Optical properties of GaN layers grown on C-, A-, R-, and M-plane sapphire substrates by gas source molecular beam epitaxy. Journal of Applied Physiology, 85 (12), 8386–8399. Craven, M.D., Lim, S.H., Wu, F., Speck, J.S. and DenBaars, S.P. (2002) Structural characterization of nonpolar (1120) a-plane GaN thin films grown on (1102) r-plane sapphire. Applied Physics Letters, 81 (3), 469–471. Ni, X., Fu. Y., Moon, Y.T., Biyikli, N. and Morkoc, H. (2006) Optimization of (1120) a-plane GaN growth by MOCVD on (1102) r-plane sapphire. Journal of Crystal Growth, 290 (1), 166–170. Sun, W.H., Kuokstis, E,. Gaevski, M., Zhang, J.P., Chen, C.Q., Wang, H.M., Yang, J.W., Simin, G., Khan, M.A., Gaska, R. and Shur, M.S. (2003) Strong ultraviolet emission from non-polar AlGaN/GaN quantum wells grown over r-plane sapphire substrates. Physica Status Solidi a: Applied Research, 200 (1), 48–51. Ng, H.M. (2002) Molecular-beam epitaxy of GaN/AlxGa1xN multiple quantum wells on R-plane (1012) sapphire substrates. Applied Physics Letters, 80 (23), 4369–4371. Craven, M.D., Waltereit, P., Speck, J.S. and DenBaars, S.P. (2004) Well-width dependence of photoluminescence emission from a-plane GaN/AlGaN multiple quantum wells. Applied Physics Letters, 84 (4), 496–498. Armitage, R., Yang, Q. and Weber, E.R. (2003) p- and n-type doping of non-polar A-plane GaN grown by molecular-beam epitaxy on R-plane sapphire. MRS Internet Journal of Nitride Semiconductor Research, 8 (6). Chakraborty, A., Xing, H., Craven, M.D., Keller, S., Mates, T., Speck, J.S., DenBaars, S.P. and Mishra, U.K. (2004)
873
874
875
876
877
878
879
Nonpolar a-plane p-type GaN and p–n junction diodes. Journal of Applied Physiology, 96 (8), 4494–4499. Wang, H.M., Chen, C.Q., Gong, Z., Zhang, J.P., Gaevski, M., Su, M., Yang, J.W., and Khan, M.A. (2004) Anisotropic structural characteristics of (1120) GaN templates and coalesced epitaxial lateral overgrown films deposited on (1012) sapphire. Applied Physics Letters, 84 (4), 499–501. Li, D.S., Chen, H., Yu, H.B., Zheng, X.H., Huang, Q. and Zhou, J.M. (2004) Anisotropy of a-plane GaN grown on rplane sapphire by metalorganic chemical vapor deposition. Journal of Crystal Growth, 265 (1–2), 107–110. € ur, U., € Morkoc, H., LilientalNi, X., Ozg€ Weber, Z. and Everitt, H.O. (2007) Epitaxial lateral overgrowth of a-plane GaN by metalorganic chemical vapor deposition. Journal of Applied Physics, 102 053506. Craven, M.D., Lim, S.H., Wu, F., Speck, J.S. and DenBaars, S.P. (2002) Threading dislocation reduction via laterally overgrown nonpolar ð1120Þaplane GaN. Applied Physics Letters, 81 (7), 1201–1203. Imer, B.M., Wu, F., DenBaars, S.P. and Speck, J.S. (2006) Improved quality (1120) a-plane GaN with sidewall lateral epitaxial overgrowth. Applied Physics Letters, 88, 061908. Chen, C.Q., Yang, J.W., Wang, H.M., Zhang, J.P., Adivarahan, V., Gaevski, M., Kuokstis, E., Gong, Z., Su, M. and Khan, M.A. (2003) Lateral epitaxial overgrowth of fully coalesced a-plane GaN on r-plane sapphire. Japanese Journal of Applied Physics, Part 2: Letters, 42 (6B), L640–L642. Chen, C.Q., Zhang, J.P., Yang, J.W., Adivarahan, V., Rai, S., Wu, S., Wang, H.M., Sun, W.H., Su, M., Gong, Z., Kuokstis, E., Gaevski, M. and Khan, M.A. (2003) A new selective area lateral epitaxy approach for depositing a-plane GaN over r-plane sapphire. Japanese
References
880
881
882
883
884
885
886
887
Journal of Applied Physics, Part 2: Letters, 42 (7B), L818–L820. € ur, U., € Fu, Y., Biyikli, N., Xie, Ni, X., Ozg€ J., Morkoc, H. and Liliental-Weber, Z. (2006) Defect reduction in (1120) a-plane GaN by two-stage epitaxial lateral overgrowth. Applied Physics Letters, 89 (26), 262105. Koida, T., Chichibu, S.F., Sota, T., Craven, M.D., Haskell, B.A., Speck, J.S., DenBaars, S.P. and Nakamura, S. (2004) Improved quantum efficiency in nonpolar (1120) AlGaN/GaN quantum wells grown on GaN prepared by lateral epitaxial overgrowth. Applied Physics Letters, 84 (19), 3768–3770. Craven, M.D., Wu, F., Chakraborty, A., Imer, B., Mishra, U.K., DenBaars, S.P. and Speck, J.S. (2004) Microstructural evolution of a-plane GaN grown on aplane SiC by metalorganic chemical vapor deposition. Applied Physics Letters, 84 (8), 1281–1283. McLaurin, M., Mates, T.E. and Speck, J.S. (2005) Molecular-beam epitaxy of ptype m-plane GaN. Applied Physics Letters, 86, 262104. Kawanishi, H., Senuma, M. and Nukui, T. (2007) TM-mode lasing and anisotropic polarization property of AlGaN multiple quantum well lasers in deep-ultraviolet spectral region.SPIEPhotonic West, Opto 07 GaN Material and Devices, January 20–25, San Jose, CA. Matsuoka, T. and Hagiwara, E. (2001) GaN growth on novel lattice-matching substrate: tilted m-plane sapphire. Physica Status Solidi a: Applied Research, 188 (2), 485–489. Baker, T.J., Haskell, B.A., Wu, F., Speck, J.S. and Nakamura, S. (2006) Characterization of planar semipolar gallium nitride films on sapphire substrates. Japanese Journal of Applied Physics, 45 (6), L154–L157. € ur, U., € Baski, A.A., Morkoc, Ni, X., Ozg€ H., Zhou, L., Smith, D.J. and Tran, C.A. Epitaxial lateral overgrowth of ð1122Þ semi-polar GaN on ð1100Þm-plane
888
889
890
891
892
893
894
sapphire by metalorganic chemical vapor deposition. Applied Physics Letters, 90, 182109. Funato, M., Ueda, M., Kawakami, Y., Narukawa, Y., Kosuci, T., Takahashi, M. and Mukai, T. (2006) Blue, green, and amber InGaN/GaN light-emitting diodes on semipolar f1122g GaN bulk substrates. Japanese Journal of Applied Physics, Part 2: Letters, 45, (24–28), L659–L662. Ueda, M., Kojima, K., Funato, M., Kawakami, Y., Narukawa, Y. and Mukai, T. (2006) Epitaxial growth and optical properties of semipolar (1122) GaN and InGaN/GaN quantum wells on GaN bulk substrates. Applied Physics Letters, 89 (21), 211907. Haskell, B.A., Baker, T.J., McLaurin, M.B., Wu, F., Fini, P.T., DenBaars, S.P., Speck, J.S. and Nakamura, S. (2005) Defect reduction in (1100) m-plane gallium nitride via lateral epitaxial overgrowth by hydride vapor phase epitaxy. Applied Physics Letters, 86, 111917. Sun, Y.J., Brandt, O. and Ploog, K.H. (2003) Growth of M-plane GaN films on gamma-LiAlO2(1 0 0) with high phase purity. Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 21 (4), 1350–1356. Ghosh, S., Waltereit, P., Brandt, O., Grahn, H.T. and Ploog, K.H. (2002) Polarization-dependent spectroscopic study of M-plane GaN on gammaLiAlO2. Applied Physics Letters, 80 (3), 413–415. Rau, B., Waltereit, P., Brandt, O., Ramsteiner, M., Ploog, K.H., Puls, J. and Henneberger, F. (2000) In-plane polarization anisotropy of the spontaneous emission of M-plane GaN/ (Al,Ga)N quantum wells. Applied Physics Letters, 77 (21), 3343–3345. Kuokstis, E., Chen, C.Q., Gaevski, M.E., Sun, W.H., Yang, J.W., Simin, G., Khan, M.A., Maruska, H.P., Hill, D.W., Chou, M.C., Gallagher, J.J. and Chai, B. (2002)
j799
j 3 Growth and Growth Methods for Nitride Semiconductors
800
895
896
897
898
899
900
901
902
903
Polarization effects in photoluminescence of C- and M-plane GaN/AlGaN multiple quantum wells. Applied Physics Letters, 81 (22), 4130–4132. Sun, Y.J., Brandt, O., Ramsteiner, M., Grahn, H.T. and Ploog, K.H. (2003) Polarization anisotropy of the photoluminescence of M-plane (In,Ga)N/GaN multiple quantum wells. Applied Physics Letters, 82 (22), 3850–3852. Amano, H., Kito, M., Hiramatsu, K. and Akasaki, I. (1989) Japanese Journal of Applied Physics, 28, L2112. Nakamura, S., Mukai, T. and Senoh, M. (1991) Japanese Journal of Applied Physics, 30, L1998. Amano, H., Kito, M., Hiramatsu, K. and Akasaki, I. (1989) Japanese Journal of Applied Physics, 28, L2112. Bhuiyan, G.A., Hashimoto, A. and Yamamoto, A. (2003) Indium nitride (InN): a review on growth, characterization, and properties. Journal of Applied Physiology, 94 (5), 2779–2808. Xu, K. and Yoshikawa, A. (2003) Effects of film polarities on InN growth by molecular-beam epitaxy. Applied Physics Letters, 83 (2), 251–253, Yoshikowa, A., Xu, K., Hashimoto, N., Terashima, W., Yoshitani, M., Choe, S. and Ishitani, Y. (2003) Effects of film polarities on InN growth by RFmolecular beam epitaxy. MRS 8th WideBandgap III-Nitride Workshop, September 29–October 1, Richmond, VA, USA. Saito, Y., Yamaguchi, T., Kanazawa, H., Kano, K., Araki, T., Nanishi, Y., Teraguchi, N. and Suzuki, A. (2002) Journal of Crystal Growth, 237, 1017. Wu, C.-L., Shen, C.-H., Lin, H.-W., Lee, H.-M. and Gwo, S. (2005) Direct evidence of 8:9 commensurate heterojunction formed between InN and AlN on c plane. Applied Physics Letters, 87, 241916–1.
904 Lu, H., Schaff, W.J. and Hwang, J. (2001) Applied Physics Letters, 79, 1489. 905 Lu, H., Schaff, W.J. and Hwang, J. (2000) Applied Physics Letters, 77, 2548. 906 Xu, K., Terashima, W., Hata, T., Hashimoto, N., Yoshitani, M., Cao, B., Ishitani, Y. and Yoshikawa, A. (2003) Comparative study of InN growth on Gaand N-polarity GaN templates by molecular-beam epitaxy. Physica Status Solidi c, 0 (7), 2814–2817. 907 Xu, K., Terashima, W., Hata, T., Hashimoto, N., Ishitani, Y. and Yoshikawa, A. (2002) Step-flow growth of InN on N-polarity GaN template by molecular beam epitaxy with a growth rate of 1.3 mm/h. Physica Status Solidi c, 0 (1), 377–381. 908 Sato, M. (1997) Japanese Journal of Applied Physics, 36, L658. 909 Yamamoto, A., Shin-ya, T., Sugiura, T. and Hashimoto, A. (1998) Characterization of MOVPE-grown InN layers on a-Al2O3/ and GaAs substrates. Journal of Crystal Growth, 189–190, 461–465. 910 Tsuchiya, T. et al. (1999) Japanese Journal of Applied Physics, 38, 1884. 911 Chen, W.-K., Pan, Y.-C. and Lin, H.-C. (1997) Japanese Journal of Applied Physics, 36, L1625. 912 Matsuoka, T., Okamoto, H. and Nakao, M. (2003) Growth of wurtzite InN using MOVPE and its optical characteristics. Physica Status Solidi c, 0 (7), 2806–2809. 913 Matsuoka, T. (2004) MOVPE growth and Characteristics of nitride semiconductors from GaN to InN, in Advanced Materials in Electronics, Research Signpost (ed. Q. Guo), pp. 46–83. 914 Matsuoka, T., Okamoto, H., Nakao, M., Harima, H. and Kurimoto, E. (2002) Optical bandgap energy of wurtzite InN. Applied Physics Letters, 81 (7), 1246–1248. 915 Aderhold, J., Davydov, V.Yu., Fedler, F., Klausing, H., Mistele, D., Rotter, T., Semchinova, O., Stemmer, J. and Graul, J. (2001) Journal of Crystal Growth, 222, 701.
References 916 Yamaguchi, S., Kariya, M. and Nitta, S. (1999) Journal of Applied Physiology, 85, 7682. 917 Bryden, W.A., Ecelberger, S.A. and Kistenmacher, T.J. (1994) Applied Physics Letters, 64, 2864. 918 Guo, Q. Lin, H.-C. and Pan, Y.-C. (1998) Journal of Crystal Growth, 189–190, 466. 919 Guo, Q., Lin, H.-C. and Pan, Y.-C. (1999) Japanese Journal of Applied Physics, 38, L490. 920 Sato, Y. and Sato, S. (1994) Journal of Crystal Growth, 144, 15. 921 Takahashi, N., Matsumoto, R. and Koukitu, A. (1997) Japanese Journal of Applied Physics, 36, L743. 922 Bu, Y., Ma, L. and Lin, M.C. (1993) Journal of Vacuum Science & Technology, 11, 2931. 923 Vurgaftman, I., Meyer, J.R. and RamMohan, L.R. (2001) Journal of Applied Physiology, 89, 5815. 924 Madelung, O. (ed.) (1996) Semiconductors – Basic Data, Springer, New York, p. 69. 925 Levinshtein, M.E., Ramyantsev, S.L. and Shur, M.S. (eds.) (2001) Properties of Advanced Semiconductor Materials, John Wiley & Sons, Inc., New York, p. 31. 926 Yamada, S., Kato, J.-I., Tanaka, S. and Suemune, I. (2001) Applied Physics Letters, 78, 3612. 927 Schwoebel, R.L. and Shipsey, E.J. (1966) Journal of Applied Physiology, 37, 3682. 928 Heffelfinger, J.R., Medlin, D.L. and McCarty, K.F. (1999) On the initial stages of AlN thin-film growth onto (0 0 0 1) oriented Al2O3 substrates by molecular beam epitaxy. Journal of Applied Physiology, 85, 466–472. 929 King, S.W., Carlson, E.P,. Therrien, R.J., Christman, J.A., Nemanich, R.J. and Divas, R.F. (1999) Journal of Applied Physiology, 86, 5584. 930 Nikishin, S.A., Antipov, V.G., Fransoeur, S., Faleev, N.N., Seryogin, G.A., Elyukhin, V.A., Temkin, H., Prokofyeva. T.I., Holtz, M., Konkar, A.
931
932
933
934
935
936
937
938
939 940
941
942
943
and Zollner, S. (1999) Applied Physics Letters, 75, 484. Assouar, M.B., Elmazria, O., Le Brizoual, L. and Alnot, P. (2002) Diamond and Related Materials, 11 (3–6), 413. Tungasmita, S., Birth, J., Persson, P.O.A., Jarrendahl, K. and Hultman, L. (2000) Applied Physics Letters, 76, 170. Ohta, J., Fujioka, H., Ito, S. and Oshima, M. (2002) Applied Physics Letters, 81, 2373. Liu, L., Liu, B., Edgar, J.H., Rajasingam, S. and Kuball, M. (2002) Journal of Applied Physiology, 92, 5183. Johnson, M.A.L., Fujita, S., Rowland, W.H., Bowers, K.A., Hughes, W.C., He, Y.W., El-Masry, N.A., Cook, J.W., Schetzina, J.F., Ren, J. and Edmond, J.A. (1996) Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 14, 2349. Onojima, N., Suda, J. and Matsunami, H. (2002) Applied Physics Letters, 80, 76; Onojima, N., Suda, J. and Matsunami, H. (2002) Journal of Crystal Growth, 237–239, 1012. Lee, C.D., Dong, Y., Feenstra, R.M., Northrup, J.E. and Neugebauer, J. (2003) Reconstructions of the AlN(0 0 0 1) surface. Physical Review B: Condensed Matter, 68, 205317. Koblmueller, G., Averbeck, R., Geelhaar, L., Riechert, H., H€osler, W. and Pongratz, P. (2003) Journal of Applied Physiology, 93, 9591. Ceperley, D.M. and Alder, B.J. (1980) Physical Review Letters, 45, 566. Perdew, J.P. and Zunger, A. (1981) Physical Review B: Condensed Matter, 23, 5048. Northrup, J.E., Di Felice, R. and Neugebauer, J. (1997) Physical Review B: Condensed Matter, 55, 13878. Northrup, J.E., Neugebauer, J., Feenstra, R.M. and Smith, A.R. (2000) Physical Review B: Condensed Matter, 61, 9932. Qian, G.-X., Martin, R.M. and Chadi, D.J. (1988) Physical Review B: Condensed Matter, 38, 7649.
j801
j 3 Growth and Growth Methods for Nitride Semiconductors
802
944 Coli, G., Bajaj, K.K., Li, J., Lin, J.Y. and Jiang, H.X. (2001) Applied Physics Letters, 78, 1829. 945 Choi, S.C., Kim, J.-H., Choi, J.Y., Lee, K.J., Lim, K.Y. and Yang, G.M. (2000) Journal of Applied Physiology, 87, 172. 946 Jenny, J.R., Van Nostrand, J.E., and Kaspi, R. (1998) Applied Physics Letters, 72, 85. 947 Iliopoulos, E., Ludwig, K.F., Jr, Moustakas, T.D. and Chu, S.N.G. (2001) Applied Physics Letters, 78, 463. 948 Mihopoulos, T.G., Gupta, V. and Jensen, K.F. (1998) Journal of Crystal Growth, 195, 733. 949 Creighton, J.R., Breiland, W.G., Coltrin, M.E. and Pawlowski, R.P. (2002) Applied Physics Letters, 81, 2626. 950 Bauer, J., Biste, L.L and Bolze, D. (1977) Physica Status Solidi a: Applied Research, 39, 173. 951 Baranov, B., Daweritz, L., Gutan, V.B., Jungk, G., Neumann, H. and Raidt, H. (1978) Physica Status Solidi a: Applied Research, 49, 629. 952 He, L., Reshchikov, M.A., Yun, F., Huang, D., King, T. and Morkoc, H. (2002) Properties of AlxGa1xN layers grown by plasma assisted molecular beam epitaxy under Ga-rich conditions. Applied Physics Letters, 81, 2178–2180. 953 Jasinski, J., Liliental-Weber, Z., He, L., Reshchikov, M.A. and Morkoc, H. (2002) Comparison of AlxGa1xN films grown on sapphire by MBE under N-rich and Ga-rich conditions. Presented at 4th Symposium on Non-Stoichiometric III–V Compounds, October 2–4, 2002, Asilomar Conference Grounds, Pacific Grove, CA. 954 Iliopoulos, E. and Moustakas, T.D. (2002) Applied Physics Letters, 81, 295. 955 Koide, Y., Itoh, N., Itoh, K., Sawakiand, N. and Akasaki, I. (1988) Japanese Journal of Applied Physics, 27, 1156. 956 Itoh, K., Kawamoto, T., Amano, H., Hiramatsu, K. and Akasaki, I. (1991) Japanese Journal of Applied Physics, 30, 1924.
957 Einfeldt, S., Kirchner, V., Heinke, H., Diebelberg, M., Figge, S., Vogeler, K. and Hommel, D. (2000) Strain relaxation in AlGaN under tensile stress. Journal of Applied Physiology, 88, 7029–7036. 958 Murray, R.T., Parbrook, P.J., Wood, D.A. and Hill, G. (2001) Microscopy of Semiconducting Materials 2001, Institute of Physics Conference Series 169, p. 289. 959 Jiao, C.G. and Cherns, D. (2001) Microscopy of Semiconducting Materials 2001, Institute of Physics Conference Series 169, p. 327. 960 Blasing, J., Reiher, A., Dadgar, A., Diez, A. and Krost, A. (2002) The origin of stress reduction by low-temperature AlN interlayers. Applied Physics Letters, 81, 2722. 961 Han, J., Waldrip, K.E., Lee, S.R., Figiel, J.J., Hearne, S.J., Petersen, G.A. and Myers, S.M. (2001) Applied Physics Letters, 78, 67. 962 Shiojima, K., Woodall, J.M., Eiting, C.J., Grudowski, P.A. and Dupuis, R.D. (1999) Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 17, 2030. 963 Wang, C.W., Liao, J.Y., Chen, C.L., Lin, W.K., Su, Y.K. and Yokoyama, M. (1999) Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 17, 1545. 964 Miller, E.J., Dang, X.Z. and Yu, E.T. (2000) Journal of Applied Physiology, 88, 5951. 965 Nikishin, S.A., Faleev, N.N., Zubrilov, A.S., Antipov, V.G. and Temkin, H. (2000) Applied Physics Letters, 76, 3028. 966 Amano, H. and Akasaki, I. (2001) Optical Materials, 19, 219. 967 Lee, C.R., Son, S.J., Seol, K.W., Yeo, J.M., Ahn, H.K. and Park, Y.J. (2001) Journal of Crystal Growth, 226, 215. 968 Kashima, T., Nakamura, R., Iwaya, M., Katoh, H., Yamaguchi, S., Amano, H. and Akasaki, I. (1999) Japanese Journal of Applied Physics, Part 2: Letters, 38, L1515. 969 Iwaya, M., Terao, S., Sano, T., Takanami, S., Ukai, T., Nakarmura, R., Kamiyama,
References
970
971
972 973
974
975
976
977
978
979 980
981
S., Amano, H. and Akasaki, I. (2001) Physica Status Solidi a: Applied Research, 188, 117. Wang, H.-M., Zhang, J.-P., Chen, C.-Q., Fareed, Q., Yang, J.-W. and Khan, M.A. (2002) Applied Physics Letters, 81, 604. Amano, H., Iwaya, M., Kashima, T., Katsuragawa, M., Akasaki, I., Han, J., Hearne, S., Floro, J.A., Chasonand, E. and Figiel, J.J. (1998) Japanese Journal of Applied Physics, 37, L1540. Amano, H., Akasaki, I. and Physics, J. (2001) Condensed Matter, 13, 6935–6944. Waldrip, K.E., Han, J., Figiel, J.J., Zhou, H., Makarona, E. and Nurmikko, A.V. (2001) Applied Physics Letters, 78, 3205. Amano, H. (2003) Electroluminescent diodes and laser diodes, in Nitride Semiconductors – Handbook on Materials and Devices (eds P. Ruterana, M. Albrecht and J. Neugebauer), WileyVCH Verlag GmbH, Weinheim, Germany. Yano, M., Detchprohm, T., Nakamura, R., Sano, S., Mochizuki, S., Nakamura, T., Amano, H. and Akasaki, I. (2000) IPAP Conference Series, 1, 292. Ashby, C.I.H., Mitchell, C.C., Han, J., Missert, N.A., Provencio, P.P., Follstaedt, D.M., Peake, G.M. and Griego, L. (2000) Applied Physics Letters, 77, 3233. Detchprohm, T., Sano, S., Mochizuki, S., Kamiyama, S., Amano, H. and Akasaki, I. (2001) Physica Status Solidi a: Applied Research, 188, 799. Kuwano, K., Tsuruda, T., Adachi, Y., Terao, S., Kamiyama, S., Amano, H. and Akasaki, I. (2002) Physica Status Solidi a: Applied Research, 192 (2), 366–370. Chang, L., Lai, S.K., Chen, F.R. and Kai, J.J. (2001) Applied Physics Letters, 79, 928. Pecz, B., Makkai, Zs., di Forte-Poisson, M.A., Huet, F. and Dunin-Borkowski, R.E. (2001) Applied Physics Letters, 78, 1529. Li, H.D., Wang, T., Liu, Y., Ao, I.P. and Sakai, S. (2002) Japanese Journal of
982
983
984
985
986
987
988
989
990
991
992
Applied Physics, Part 2: Letters, 41, L732–L735. Cho, H.K. and Lee, J.Y, (2001) Influence of Mg doping on structural defects in AlGaN layers grown by metalorganic chemical vapor deposition. Applied Physics Letters, 79, 3788–3790. Nakamura, T., Mochizuki, S., Terao, S., Sano, T., Iwaya, M., Kamiyama, S., Amano, H. and Akasaki, I. (2002) Journal of Crystal Growth, 237, 1129. Skierbiszeski, C., Suski, T., Leszczynski, M., Shin, M., Skowronski, M., Bremser, M.D. and Davis, R.F. (1999) Applied Physics Letters, 74, 3833. Adivarahan, V., Simin, G., Tamulaitis, G., Srinivasan, R., Yang, J. and Asif Khan, M. (2001) Applied Physics Letters, 79, 1903. Nam, K.B., Li, J., Nakarmi, M.L., Lin, J.Y. and Jiang, H.X. (2002) Applied Physics Letters, 81, 1038. Stutzmann, M., Ambacher, O., Cros, A., Brandt, M.S., Angerer, H., Dimitrov, R., Reinacher, N., Metzger, T., Hopler, R., Brunner, D., Freudenberg, F., Handschuh, R. and Deger, Ch. (1997) Properties and applications of MBE grown AlGaN. Materials Science & Engineering B: Solid State Materials for Advanced Technology, 50 (1–3), 212–218. Yoshida, S., Misawa, S. and Gonda, S. (1982) Journal of Applied Physiology, 53, 6844. Steude, G., Hofmann, D.M., Meyer, B.K., Amano, H. and Akasaki, I. (1998) Physica Status Solidi a: Applied Research, 165, R3. Tanaka, T., Watanabe, A., Amano, H., Kobayashi, Y., Akasaki, I., Yamazaki, S. and Koike, M. (1994) Applied Physics Letters, 65, 593. Suzuki, M., Nishio, J., Onomura, M. and Hongo, C. (1998) Journal of Crystal Growth, 189–190, 511. Sugiura, L., Suzuki, M., Nishio, J., Itaya, K., Kokubun, Y. and Ishikawa, M. (1998)
j803
j 3 Growth and Growth Methods for Nitride Semiconductors
804
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
Japanese Journal of Applied Physics, 17, 3878. See Tanaka, T., Watanabe, A., Amano, H., Kobayashi, Y., Akasaki, I., Yamazaki, S. and Koike, M. (1994) Applied Physics Letters, 65, 593; Akasaki, I. and Amano, H. (1991) Materials Research Society Symposium Proceedings, 242, 383. Li, J., Oder, T.N., Nakarmi, M.L., Lin, J.Y. and Jiang, H.X. (2002) Applied Physics Letters, 80, 1210. Suzuki, M., Nishio, J., Onomura, M. and Hongo, C. (1998) Journal of Crystal Growth, 189–190, 511. Ruterana, P., De Saint Jores, G., Laugt, M., Omnes, F. and Bellet-Amalric, E. (2001) Evidence for multiple chemical ordering in AlGaN grown by metalorganic chemical vapor deposition. Applied Physics Letters, 78, 344. Korakakis, D., Ludwig, K.F. and Moutakas, T.D. (1997) Applied Physics Letters, 71, 72. Matsuoka, T., Yoshimoto, N., Sasaki, T. and Katsui, A. (1992) Journal of Electronic Materials, 21, 157. Shimizu, M., Hiramatsu, K. and Sawaki, N. (1994) Journal of Crystal Growth, 145, 209. Nagatomo, T., Kuboyama, T., Minamino, H. and Omoto, O. (1989) Japanese Journal of Applied Physics, 28, L1334. Matsuoka, T., Tanaka, H., Sasaki, T. and Katsui, K. (1990) Proceedings of the 16th International Symposium on GaAs and Related Compounds, 1989, Karuizawa, Institute of Physics Conference Series 106, Institute of Physics, Bristol, p. 141. Yoshimoto, N., Matsuoka, T., Sasaki, T. and Katsui, A. (1991) Applied Physics Letters, 59, 2251. Matsuoka, T., Yoshimoto, N., Sasaki, T. and Katsui, A. (1992) Journal of Electronic Materials, 21, 157. Shimizu, M., Hiramatsu, K. and Sawaki, N. (1994) Journal of Crystal Growth, 145, 209. Reed, M.J., El-Masry, N.A., Parker, C.A., Roberts, J.C. and Bedair, S.M. (2000)
1006
1007
1008
1009
1010
1011
Critical layer thickness determination of GaN/InGaN/GaN double heterostructures. Applied Physics Letters, 77, 4121–4123. Piscopiello, E., Catalano, M., Vittori Antisari, M., Passaseo, A., Branca, E., Cingolani, R., Berti, M. and Drigo, A.V. (2001) Influence of the N2/H2 ratio on the structural features of InxGa1xN/ GaN films grown by MOCVD. Materials Science & Engineering B: Solid State Materials for Advanced Technology, 87 (3), 237–24319. Kim, S., Lee, K., Park, K. and Kim, C.-S. (2003) Effects of barrier growth temperature on the properties of InGaN/ GaN multi-quantum wells. Journal of Crystal Growth, 247 (1–2), 62–68. Lefebvre, P., Taliercio, T., Morel, A., Allègre, J., Gallart, M., Gil, B., Mathieu, H., Damilano, B., Grandjean, N. and Massies, J. (2001) Effects of GaAlN barriers and of dimensionality on optical recombination processes in InGaN quantum wells and quantum boxes. Applied Physics Letters, 78 (11), 1538–1540. Ting, S.M., Ramer, J.C., Florescu, D.I., Merai, V.N., Albert, B.E., Parekh, A., Lee, D.S., Lu, D., Christini, D.V., Liu, L. and Armour, E.A. (2003) Morphological evolution of InGaN/GaN quantum-well heterostructures grown by metalorganic chemical vapor deposition. Journal of Applied Physics, 94 (3), 1461–1467. Sharma, N., Tricker, D., Thomas, P., Bougrioua, Z., Jacobs, K., Cheyns, J., Moerman, I., Thrush, T., Considine, L., Boyd, A. and Humphreys, C. (2001) Chemical mapping of InGaN MQWs. Journal of Crystal Growth, 230 (3–4), 438–441. Cho, H.K., Lee, J.Y., Yang, G.M. and Kim, C.S. (2001) Formation mechanism of V defects in the InGaN/GaN multiple quantum wells grown on GaN layers with low threading dislocation density. Applied Physics Letters, 79 (2), 215–217.
References 1012 Kim, I.-H., Park, H.-S., Park, Y.-J. and Kim, T. (1998) Formation of V-shaped pits in InGaN/GaN multiquantum wells and bulk InGaN films. Applied Physics Letters, 73 (12), 1634–1636. 1013 Nakamura, S. and Mukai, T. (1992) Japanese Journal of Applied Physics, 31, L1457. 1014 Piner, E.L., Behbehani, M.K., El-Mastry, N.A., McIntosh, F.G., Roberts, J.C., Boutros, K.S. and Bedair, S.M. (1997) Applied Physics Letters, 70, 461. 1015 Ho, I..-H. and Stringfellow, G.B. (1996) Solid phase immiscibility in GaInN. Applied Physics Letters, 69 (18), 2701–2703. 1016 Osamura, K., Naka, S. and Murakami, Y. (1975) Journal of Applied Physiology, 46, 3432. 1017 Koukitu, A., Takahashi, N., Taki, T. and Seki, H. (1996) Japanese Journal of Applied Physics, 35, L673. 1018 Kawaguchi, Y., Shimizu, M., Hiramatsu, K. and Sawaki, N. (1997) Materials Research Society Symposium Proceedings, 449, 89. 1019 Bedair, S.M., McIntosh, F.G., Roberts, J.C., Piner, E.L., Boutros, K.S. and ElMasry, N.A. (1997) Journal of Crystal Growth, 178, 32–44. 1020 Pozina, G., Bergman, J.P., Monemar, B., Iwaya, M., Nitta, S., Amano, H. and Akasaki, I. (2001) Photoluminescence of InGaN/GaN multiple quantum wells grown by mass transport. Journal of Crystal Growth, 230 (3–4), 473–476. 1021 Tong, Y.Z., Li, F., Zhang, G.Y., Yang, Z.J., Jin, S.X., Ding, X.M. and Gan, Z.Z. (1998) Growth and doping characteristics of InGaN films grown by low pressure MOCVD. Solid State Communications, 109 (3), 173–176. 1022 Wang, T., Saeki, H., Bai, J., Shirahama, T., Lachab, M., Sakai, S. and Eliseev, P. (2000) Effect of silicon doping on the optical and transport properties of InGaN/GaN multiple-quantum-well structures. Applied Physics Letters, 76 (13), 1737–1739.
1023 Yamasaki, S., Asami, S., Shibata, N., Koike, M., Manabe, K., Tanaka, T., Amano, H. and Akasaki, 1. (1995) Applied Physics Letters, 66, 1112. 1024 Ruterana, P., Nouet, G., Van der Stricht, W., Moerman, I. and Considine, L. (1998) Applied Physics Letters, 72, 1742. 1025 Doppalapudi, D., Basu, S.N., Ludwing, K.F. and Moustakas, T.D. (1998) Applied Physics Letters, 84, 1389. 1026 Doppalapudi, D., Basu, S.N. and Moustakas, T.D. (1999) Applied Physics Letters, 85, 883. 1027 Behbehani, M.K., Piner, E.L., Liu, S.X., El-Masry, N.A. and Bedair, S.M. (1999) Applied Physics Letters, 75, 2202. 1028 Wieser, N., Ambacher, O., Felsl, H..-P., G€ orgens, L. and Stutzmann, M. (1999) Compositional fluctuations in GaInN/ GaN double heterostructures investigated by selectively excited photoluminescence and Raman spectroscopy. Applied Physics Letters, 74 (26), 3981–3983. 1029 Narukawa, I.Y., Kawakami, Y., Funato, M., Fujita, Sz., Fujita, Sg. and Nakamura, S. (1997) Applied Physics Letters, 70, 868. 1030 Narukawa, Y., Kawakami, Y., Fujita, S. and Nakamura, S. (1999) Physical Review B: Condensed Matter, 59, 10283. 1031 Wakahara, A., Tokuda, T., Dang, X.-Z., Noda, S. and Sasaki, A. (1997) Compositional inhomogeneity and immiscibility of a GaInN ternary alloy. Applied Physics Letters, 71 (7), 906–908. 1032 Karpov, S.Yu. (1998) MRS Internet Journal of Nitride Semiconductor Research, 3, 16. 1033 Saito, T. and Arakawa, Y. (1999) Atomic structure and phase stability of InxGa1xN random alloys calculated using a valence-force-field method. Physical Review B: Condensed Matter, 60 (3), 1701–1706. 1034 Mattila, T. and Zunger, A. (1999) Predicted bond length variation in wurtzite and zinc-blende InGaN and AlGaN alloys. Journal of Applied Physiology, 85 (1), 160–167.
j805
j 3 Growth and Growth Methods for Nitride Semiconductors
806
1035 Kim, K., Limpijumnong, S., Lambrecht, W.R.L. and Segall, B. (1997) III–V Nitrides (eds F.A. Ponce, T.D. Moustakas, I. Akasaki and B.A. Monemar), Materials Research Society, Pittsburgh, PA, Materials Research Society Symposium Proceedings, 449, 929. 1036 Matsuoka, T., Yoshimoto, N., Sasaki, T. and Katsui, A. (1992) Journal of Electronic Materials, 21, 157. 1037 Shimizu, M., Hiramatsu, K. and Sawaki, N. (1994) Journal of Crystal Growth, 145, 209. 1038 Osamura, K., Naka, S. and Murakami, Y. (1975) Journal of Applied Physiology, 46, 3432. 1039 Yamaguchi, S., Kariya, M., Nitta, S., Kato, H., Takeuchi, T., Wetzel, C., Amano, H. and Akasaki, I. (1998) Journal of Crystal Growth, 195, 309. 1040 Teles, L.K., Furthm€ uller, J., Scolfaro, L.M.R., Leite, J.R. and Bechstedt, F. (2000) First-principles calculations of the thermodynamic and structural properties of strained InxGa1xN and AlxGa1xN alloys. Physical Review B: Condensed Matter, 62 (4), 2475–2485. 1041 Lemos, V., Silveira, E., Leite, J.R., Tabata, A., Trentin, R., Scolfaro, L.M.R., Frey, T., As, D.J., Schikora, D. and Lischka, K. (2000) Evidence for phase-separated quantum dots in cubic InGaN layers from resonant raman scattering. Physical Review Letters, 84 (16), 3666–3669. 1042 Okumura, H., Hamaguchi, H., Koizumi, T., Balanishi, K., Nagatomo, T. and Yoshida, S. (1998) Journal of Crystal Growth, 189–190, 390. 1043 Lima, A.P., Tabata, A., Leite, J.R., Kaiser, S., Schikora, D., Sch€ottker, B., Frey, T., As, D.J. and Lischka, K. (1999) Journal of Crystal Growth, 201–202, 396. 1044 Guo, Q.X., Ogawa, H. and Yoshida, A. (1995) Journal of Crystal Growth, 146, 462. 1045 Pearton, S.J., Lee, J.W., MacKenzie, J.D., Abernathy, C.R. and Shul, R.J. (1995) Dry etch damage in InN, InGaN, and
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
InAlN. Applied Physics Letters, 67 (16), 2329–2331. Abernathy, C.R., MacKenzie, J.D., Bharatan, S.R., Jones, K.S. and Pearton, S.J. (1995) Growth of InxGa1xN and InxAl1xN on GaAs metalorganic molecular beam epitaxy. Journal of Vacuum Science & Technology A: Vacuum Surfaces and Films, 13 (3), 716–718. Pearton, S.J., Abernathy, C.R. and Ren, F. (1994) Dry patterning of InGaN and InAlN. Applied Physics Letters, 64 (26), 3643–3645. Behbehani, M.K., Piner, E.L., Liu, S.X., El-Masry, N.A. and Bedair, S.M. (1999) Applied Physics Letters, 75, 2202. Sing, R., Doppalapudi, D., Moustakas, T.D. and Romano, L.T. (1997) Applied Physics Letters, 70, 1089. El-Masry, N.A., Piner, E.L., Liu, S.X. and Bedair, S.M. (1998) Applied Physics Letters, 72, 40. Moon, Y.T., Kim, D.J., Song, K.M., Kim, D.W., Yi, M.S., Noh, D.Y. and Park, S.J. (2000) Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 18, 2631. Moon, Y.T., Kim, D.J., Song, K.M., Choi, C.J., Han, S.H., Seong, T.Y. and Park, S.J. (2001) Effects of thermal and hydrogen treatment on indium segregation in InGaN/GaN multiple quantum wells. Journal of Applied Physiology, 89, 6514. McCluskey, M.D., Romano, L.T., Krusor, B.S., Bour, D.P., Johnson, N.M. and Brennan, S. (1998) Applied Physics Letters, 72, 1730. Narukawa, Y., Kawakami, Y., Funato, M., Fujita, S., Fujita, S. and Nakamura, S. (1997) Applied Physics Letters, 70, 981. Chichibu, S., Azuhata, T., Sota, T. and Nakamura, S. (1996) Applied Physics Letters, 69, 4188. Cho, H.K., Lee, J.Y., Sharma, N., Humhreys, C.J., Yang, G.M., Kim, C.S., Song, J.H. and Yu, P.W. (2001) Applied Physics Letters, 79, 2594. Ruterana, P., Kret, S. and Poisson, M.A. (2002) Quantitative measurement of In
References
1058
1059
1060
1061
1062
1063
1064
1065
1066
fluctuation inside MOCVD InGaN QWs. Materials Science & Engineering B: Solid State Materials for Advanced Technology, 93 (1–3), 185–188. Chen, H..-J., Feenstra, R.M., Northrup, J.E., Zywietz, T., Neugebauer, J. and Greve, D.W. (2000) Surface structures and growth kinetics of InGaN (0 0 0 1) grown by molecular beam epitaxy. 27th Conference on the Physics and Chemistry of Semiconductor Interfaces, January 16–20, 2000, Salt Lake City, UT, USA, Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 18 (4), 2284–2289. Widmann, F., Daudin, B., Feuillet, G., Rouviere, J.-L. and Pelekanos, N. (1998) Applied Physics Letters, 73, 2642. Feenstra, R.M., Chen, H., Ramachandran, V., Smith, A.R. and Greve, D.W. (2000) 7th International Conference on the Formation of Semiconductor Interfaces, June 21–25, 1999, Goteborg, Sweden, Applied Surface Science, 166, 165–172. B€ ottcher, T., Einfeldt, S., Kirchner, V., Figge, S., Heinke, H., Hommel, D., Selke, H. and Ryder, P.L. (1998) Applied Physics Letters, 73, 3232. Chen, H., Smith, A.R., Feenstra, R.M., Greve, D.W. and Northrup, J.E. (1999) Internet Journal of Nitride Semiconductor Research, 4S1, G9.5. Lin, M.E., Strite, S., Agarwal, A., Salvador, A., Zhou, G.L., Teraguchi, N., Rockett, A. and Morkoc, H. (1993) Applied Physics Letters, 62, 702. Hacke, P., Feuillet, G., Okumura, H. and Yoshida, S. (1996) Applied Physics Letters, 69, 2507. Iwata, .K., Asahi, H., Yu, S.J., Asami, K., Fujita, H., Fushida, M. and Gonda, S. (1996) Japanese Journal of Applied Physics, Part 2: Letters, 35, L289. Ramachandran, V., Lee, C.D., Feenstra, R.M., Smith, A.R., Northrup, J.E. and Greve, D.W. (2000) Journal of Crystal Growth, 209, 355.
1067 Thamm, A., Brandt, O., Takemura, Y., Trampert, A. and Ploog, K.H. (1999) Applied Physics Letters, 75, 944. 1068 Northrup, J.E. and Neugebauer, J. (1999) Physical Review B: Condensed Matter, 60, 8473. 1069 Kubota, K., Kobayashi, Y. and Fujimoto, K. (1989) Preparation and properties of III–V nitride thin films. Journal of Applied Physiology, 66 (7), 2984–2988. 1070 Kubota, K., Kobayashi, Y. and Fujimoto, K. (1989) Preparation and properties of III–V nitride thin films. Journal of Applied Physiology, 66 (7), 2984–2988. 1071 Kim, K.S., Saxler, A., Kung, P., Razeghi, R. and Lim, K.Y. (1997) Determination of the band-gap energy of Al1 xInxN grown by metal–organic chemical-vapor deposition. Applied Physics Letters, 71 (6), 800–802. 1072 Matsuoka, T. (1997) Calculation of unstable mixing region in wurtzite In1 xyGaxAlyN. Applied Physics Letters, 71 (1), 105–106. 1073 Yamaguchi, S., Kariya, M., Nitta, S., Takeuchi, T., Wetzel, C., Amano, H. and Akasaki, I. (2000) Anomalous features in the optical properties of Al1 xInxN on GaN grown by metal organic vapor phase epitaxy. Applied Physics Letters, 76 (7), 876–878. 1074 Lukitsch, M.J., Danylyuk, Y.V., Naik, V.M., Huang, C., Auner, G.W., Rimai, L. and Naik, R. (2001) Optical and electrical properties of Al1 xInxN films grown by plasma source molecular-beam epitaxy. Applied Physics Letters, 79 (5), 632–634. 1075 Yamaguchi, S., Kariya, M., Nitta, S., Takeuchi, T., Wetzel, C., Amano, H. and Akasaki, I. (1998) Observation of photoluminescence from Al1 xInxN heteroepitaxial films grown by metalorganic vapor phase epitaxy. Applied Physics Letters, 73, 830. 1076 Matsuoka, T. (1997) Calculation of unstable mixing region in wurtzite In1 xyGaxAlyN. Applied Physics Letters, 71 (1), 105–106.
j807
j 3 Growth and Growth Methods for Nitride Semiconductors
808
1077 Goano, M., Bellotti, E., Ghillino, E., Garetto, C., Ghione, G. and Brennan, K.F. (2000) Band structure nonlocal pseudopotential calculation of the IIInitride wurtzite phase materials system. Part II. Ternary alloys AlxGa1xN, InxGa1xN, and InxAl1xN. Journal of Applied Physiology, 88 (11), 6476–6482. 1078 Wright, A.F. and Nelson, J.S. (1995) First-principles calculations for zincblende AlInN alloys. Applied Physics Letters, 66 (25), 3465–3467. 1079 Lukitsch, M.J., Danylyuk, Y.V., Naik, V.M., Huang, C., Auner, G.W., Rimai, L. and Naik, R. (2001) Optical and electrical properties of Al1 xInxN films grown by plasma source molecular-beam epitaxy. Applied Physics Letters, 79 (5), 632–634. 1080 Kim, K.S., Saxier, A., Kung, P., Razeghi, M. and Lim, K.Y. (1997), Determination of the band-gap energy of Al1 xInxN grown by metal–organic chemical-vapor deposition. Applied Physics Letters, 71 (6), 800–802. 1081 Peng, T., Piprek, J., Qiu, G., Olowafe, J.O., Unruh, K.M., Swann, C.P. and Schubert, E.F. (1997) Band gap bowing and refractive index spectra of polycrystalline AlxIn1xN films deposited by sputtering. Applied Physics Letters, 71 (17), 2439–2441. 1082 Teles, L.K., Furthm€ uller, J., Scolfaro, L.M.R., Leite, J.R. and Bechstedt, F. (2001) Physical Review B: Condensed Matter, 63, 085204. 1083 Scolfaro, L.M.R. (2002) Physica Status Solidi a: Applied Research, 190, 15. 1084 Tabata, A., Teles, L.K., Scolfaro, L.M.R., Leite, J.R., Kharchenko, A., Frey, T., As, D.J., Schikora, D., Lischka, K., Furthm€ uller, J. and Bechstedt, F. (2002) Applied Physics Letters, 80, 769. 1085 Teles, L.K., Scolfaro, L.M.R., Leite, J.R., Furthm€ uller, J. and Bechstedt, F. (2002) Applied Physics Letters, 80, 1177. 1086 Teles, L.K., Scolfaro, L.M.R., Leite, J.R., Furthm€ uller, J. and Bechstedt, F. (2002) Phase diagram, chemical bonds, and gap
1087
1088
1089
1090
1091
1092
1093
1094
1095
bowing of cubic InxAl1xN alloys: ab initio calculations. Journal of Applied Physiology, 92 (12), 7109–7113. Teles, L.K., Furthm€ uller, J., Scolfaro, L.M.R., Leite, J.R. and Bechstedt, F. (2000) Physical Review B: Condensed Matter, 62, 2475. Chitnis, A., Adivarahan, A., Shatalov, M., Zhang, J., Gaevski, M., Shuai, W., Pachipulusu, R., Sun, J., Simin, K., Simin, G., Yang, J. and Khan, M.A. 2002 Submilliwatt operation of AlInGaN based multifinger-design 315 nm light emitting diode (LED) over sapphire substrate. Japanese Journal of Applied Physics, Part 2: Letters, 41, L320–L322. Nagahama, S., Yanamoto, T., Sano, M. and Mukai, T. (2002) Study of GaN-based laser diodes in near ultraviolet region. Japanese Journal of Applied Physics, Part 1: Regular Papers, Short Notes & Review Papers, 41, 5–10. Bedair, S.M., McIntosh, F.G., Roberts, J.C., Piner, E.L., Boutros, K.S. and ElMasry, N.A. (1997) Journal of Crystal Growth, 178, 32. Yamaguchi, S., Kariya, M., Nitta, S., Kato, H., Takeuchi, T., Wetzel, C., Amano, H. and Akasaki, I. (1998) Journal of Crystal Growth, 195, 309. Han, J., Figiel, J.J., Petersen, G.A., Meyers, S.M., Crawford, M.H. and Banas, M.A. (2000) Japanese Journal of Applied Physics, Part 1: Regular Papers, Short Notes & Review Papers, 39, 2372. Lima, A.P., Miskys, C.R., Karrer, U., Ambacher, O., Wenzel, A., Rauschenbach, B. and Stutzmann, M. (2000) Journal of Crystal Growth, 220, 341. Monroy, E., Gogneau, N., Jalabert, D., Bellet-Amalric, E., Hori, Y., Enjalbert, F., Dang, L.S. and Daudin, B. (2003) Applied Physics Letters, 82, 2242. Dimakis, E., Georgakilas, A., Androulidaki, M., Tsagaraki, K., Kittler, G., Kalaitzakis, F., Cengher, D., BelletAmalric, E., Jalabert, D. and Pelekanos,
References
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
N.T. (2003) Plasma-assisted MBE growth of quaternary InAlGaN quantum well heterostructures with room temperature luminescence. Journal of Crystal Growth, 251, 476–480. Monroy, E., Gogneau, N., Enjalbert, F., Fossard, F., Jalabert, D., Bellet-Amalric, E., Dang, Le Si. and Daudin, B. (2003) Molecular-beam epitaxial growth and characterization of quaternary III-nitride compounds. Journal of Applied Physiology, 94 (5), 3121–3127. Monroy, E., Daudin, B., Bellet-Amalric, E., Gogneau, N., Jalabert, D., Enjalbert, F., Brault, J., Barjon, J. and Dang, L.S. (2003) Journal of Applied Physiology, 93, 1550. B€ ottcher, T., Einfeldt, S., Kirchner, V., Figge, S., Heinke, H., Hommel, D., Selke, H. and Ryder, P.L. (1998) Applied Physics Letters, 73, 3232. Chen, H., Feenstra, R.M., Northrup, J., Neugebauer, J. and Greve, D.W. (2001) Internet Journal of Nitride Semiconductor Research, 6, 11. Huang, J.S., Dong, X., Luo, X.D., Li, D.B., Liu, X.L., Xu, Z.Y. and Ge, W.K. (2003) Journal of Crystal Growth, 247, 84. Hirayama, H., Kinoshita, A., Yamabi, T., Enomoto, Y., Hirata, A., Araki, T., Nanishi, Y. and Aoyagi, Y. (2002) Applied Physics Letters, 80, 207. Moison, J.M., Guille, C., Houzay, F., Barthe, F. and Van Rompay, M. (1989) Physical Review B: Condensed Matter, 40, 6149. Stampfl, C. and Van der Walle, C.G. (1999) Physical Review B: Condensed Matter, 59, 5521. Li, J., Nam, K.B., Kim, K.H., Lin, J.Y. and Jiang, H.X. (2001) Applied Physics Letters, 78, 61. Shatalov, M., Zhang, J., Chitnis, A.S., Adivarahan, V., Yang, J., Simin, G. and Khan, M.A. (2002) IEEE Journal of Selected Topics in Quantum Electronics, 8, 302. Zhang, J.P., Wang, H.M., Gaevski, M.E., Chen, C.Q., Fareed, Q., Yang, J.W.,
1107
1108
1109
1110
1111
1112 1113 1114
1115 1116 1117
1118
Simin, G. and Asif Khan, M. (2002) Applied Physics Letters, 80, 3542. Aumer, M.E., LeBoeuf, S.F., McIntosh, F.G. and Bedair, S.M. (1999) High optical quality AlInGaN by metalorganic chemical vapor phase epitaxy. Applied Physics Letters, 75, 3315. Williams, C.K., Glisson, T.H., Hauser, J.R. and Littlejohn, M.A. (1978) Journal of Electronic Materials, 7, 639. Chen, C.Q., Yang, J.W., Ryu, M.-Y., Zhang, J.P., Kuokstis, E., Simin, G. and Asif Khan, M. (2002) Japanese Journal of Applied Physics, 41, 1924. Lima, A.P., Miskys, C.R., Karrer, U., Ambacher, O., Wenzel, A., Rauschenbach, B. and Stutzmann, M. (2000) Journal of Crystal Growth, 220, 341. Cremades, A., Navarro, V., Piqueras, J., Lima, A.P., Ambacher, O. and Stutzmann, M. (2001) Journal of Applied Physiology, 90, 4868. Meijering, J.L. (1950) Philips Research Reports, 5, 333. Yoffe, A.D. (2001) Advances in Physics, 50, 1. Bimberg, D., Grundmann, M. and Ledentsov, N.N. (1999) Quantum Dot Heterostructures, John Wiley & Sons, Ltd, Chichester, UK. Gerard, J.M., Cabrol, O. and Sermage, B. (1996) Applied Physics Letters, 68, 3123. Arakawa, Y. and Sakaki, H. (1982) Applied Physics Letters, 40, 939. Huang, D., Reshchikov, M.A. and Morkoc, H. (2002) Growth, structure, and optical properties of III-nitride quantum dots, in Quantum Dots (eds E. Borovitskaya and M.S. Shur), World Scientific, International Journal of High Speed Electronics, 25 (1), 79–110. Huang, D., Fu, Y. and Morkoc, H. (2004) Preparation, structural and optical properties of GaN based quantum dots, in Semiconductor Nanostructures for Optoelectronic Applications (ed. T. Steiner), Artech House.
j809
j 3 Growth and Growth Methods for Nitride Semiconductors
810
1119 Das, D. and Melissinos, A.C. (1986) Quantum Mechanics, Gordon and Breach, New York. 1120 Martin, G., Botchkarev, A., Rockett, A. and Morkoc, H. (1996) Applied Physics Letters, 68, 2541. 1121 Jain, S.C., Willander, M., Narayan, J. and Van Overstraeten, R. (2000) Journal of Applied Physiology, 87, 965. 1122 Kuball, M., Gleize, J., Tanaka, S. and Aoyagi, Y. (2001) Applied Physics Letters, 78, 987. 1123 Tanaka, S., Iwai, S. and Aoyagi, Y. (1996) Applied Physics Letters, 69, 4096. 1124 Shen, X.Q., Tanaka, S., Iwai, S. and Aoyagi, Y. (1998) Applied Physics Letters, 72, 344. 1125 Widmann, F., Daudin, B., Feuillet, G., Samson, Y., Rouviere, J.L. and Pelekanos, N. (1998) Growth kinetics and optical properties of self-organized GaN quantum dots. Journal of Applied Physiology, 83, 7618. 1126 Widmann, F., Simon, J., Daudin, B., Feuillet, G., Rouviere, J.L., Pelekanos, N.T. and Fishman, G. (1998) Blue-light emission from GaN self-assembled quantum dots due to giant piezoelectric effect. Physical Review B: Condensed Matter, 58, R15989. 1127 Damilano, B., Grandjean, N., Semond, F., Massies, J. and Leroux, M. (1999) From visible to white light emission by GaN quantum dots on Si(1 1 1) substrate. Applied Physics Letters, 75, 962. 1128 Shchukin, V.A. and Bimberg, D. (1999) Reviews of Modern Physics, 71, 1125. 1129 Eaglesham, D.J. and Cerullo, M. (1990) Physical Review Letters, 64, 1943. 1130 Petroff, P.M. and DenBaars, S.P. (1994) Superlattices and Microstructures, 15, 15. 1131 Li, A., Liu, F., Petrovykh, D.Y., Lin, J.-L., Viernow, J., Himpsel, F.J. and Lagally, M.G. (2000) Creation of quantum platelets via strain-controlled selforganization at steps. Physical Review Letters, 85, 5380.
1132 Daudin, B., Widmann, F., Feuillet, G., Samson, Y., Arlery, M. and Rouviere, J.L. (1997) Stranski–Krastanov growth mode during the molecular beam epitaxy of highly strained GaN. Physical Review B: Condensed Matter, 56, R7069. 1133 Daudin, B., Feuillet, G., Widmann, F., Samson, Y., Rouviere, J.L., Pelekanos, N. and Fishman, G. (1998) Materials Research Society Symposium Proceedings, 482, 205. 1134 Rouviere, J.L., Simon, J., Pelekanos, N., Daudin, B. and Feuillet, G. (1999) Preferential nucleation of GaN quantum dots at the edge of AlN threading dislocations. Applied Physics Letters, 75, 2632. 1135 Daudin, B., Widmann, F., Simon, J., Feuillet, G., Rouviere, J.L., Pelekanos, N. and Fishman, G. (1999) Internet Journal of Nitride Semiconductor Research, 4S1, G9.2. 1136 Shen, X.-Q., Tanaka, S., Iwai, S. and Aoyagi, Y. (1998) Applied Physics Letters, 72, 344; Hirayama, H., Aoyagi, Y. and Tanaka, S. (1999) Internet Journal of Nitride Semiconductor Research, 431, G9.4. 1137 Arlery, M., Rouviere, J.L., Widmann, F., Daudin, B., Feuillet, G. and Mariette, H. (1999) Quantitative characterization of GaN quantum-dot structures in AlN by high-resolution transmission electron microscopy. Applied Physics Letters, 74, 3287. 1138 Feuillet, G., Widmann, F., Daudin, B., Schuler, J., Arlery, M., Rouviere, J.L., Pelekanos, N. and Briot, O. (1997) Materials Science & Engineering B: Solid State Materials for Advanced Technology, 50, 233. 1139 Xie, Q., Madhukar, A., Chen, P. and Kobayashi, N.P. (1995) Physical Review Letters, 75, 2542. 1140 Tersoff, J., Teichert, C. and Lagally, M.G. (1996) Physical Review Letters, 76, 1675. 1141 Gleize, J., Frandon, J., Demangeot, F., Renucci, M., Adelmann, C., Daudin, B.,
References
1142
1143
1144
1145
1146
1147 1148
1149
1150
1151
Feuillet, G., Damilano, B., Grandjean, N. and Massies, J. (2000) Signature of GaN–AlN quantum dots by nonresonant Raman scattering. Applied Physics Letters, 77, 2174. Martinez-Guerrero, E., Adelmann, C., Chabuel, F., Simon, J., Pelekanos, N.T., Mula, G., Daudin, B., Feuillet, G. and Mariette, H. (2000) Self-assembled zinc blende GaN quantum dots grown by molecular-beam epitaxy. Applied Physics Letters, 77, 809. Adelmann, C., Simon, J., Feuillet, G., Pelekanos, N.T., Daudin, B. and Fishman, G. (2000) Self-assembled InGaN quantum dots grown by molecular-beam epitaxy. Applied Physics Letters, 76, 1570. Damilano, B., Grandjean, N., Dalmasso, S. and Massies, J. (1999) Roomtemperature blue-green emission from InGaN/GaN quantum dots made by strain-induced islanding growth. Applied Physics Letters, 75, 3751. Grandjean, N. and Massies, J. (1998) Real time control of InxGa1xN molecular beam epitaxy growth Applied Physics Letters, 72, 1078. Dmitriev, V., Irvine, K., Zubrilov, A., Tsvetkov, D., Nikolaev, V., Jakobson, M., Nelson, D. and Sitnikova, A. (1996) Gallium Nitride and Related Materials (eds R.D. Dupuis, J.A. Edmond, F.A. Ponce and S. Nakamura), Materials Research Society, Pittsburgh, PA. Tanaka, S., Iwai, S. and Aoyagi, Y. (1996) Applied Physics Letters, 69, 4096. Tanaka, S., Hirayama, H., Iwai, S. and Aoyagi, Y. (1997) Materials Research Society Symposium Proceedings, 449, 135. Ramvall, P., Tanaka, S., Nomura, S., Riblet, P. and Aoyagi, Y. (1999) Applied Physics Letters, 75, 1935. Ramvall, P., Riblet, P., Nomura, S., Aoyagi, Y. and Tanaka, S. (2000) Journal of Applied Physiology, 87, 3883. Tanaka, S., Hirayama, H., Aoyagi, Y., Narukawa, Y., Kawakami, Y., Fujita, S.
1152
1153
1154
1155
1156
1157
1158
1159
1160 1161
1162
1163
1164
1165
and Fujita, S. (1997) Applied Physics Letters, 71, 1299. Hirayama, H., Tanaka, S., Ramvall, P. and Aoyagi, Y. (1998) Applied Physics Letters, 72, 1736. Tachibana, K., Someya, T. and Arakawa, Y. (1999) Applied Physics Letters, 74, 383. Nistor, L., Bender, H., Vantomme, A., Wu, M.F., Van Landuyt, J., ODonnell, K.P., Martin, R., Jacobs, K. and Moerman, I. (2000) Applied Physics Letters, 77, 507. ODonnell, K.P., Martin, R. and Middleton, P.G. (1999) Physical Review Letters, 82, 237. Wang, J., Nozaki, M., Lachab, M., Ishikawa, Y., Fareed, R.S.Q., Wang, T., Hao, M. and Sakai, S. (1999) Applied Physics Letters, 75, 950. Lachab, M., Nozaki, M., Wang, J., Ishikawa, Y., Fareed, Q., Nishikawa, T., Nishino, K. and Sakai, S. (2000) Journal of Applied Physiology, 87, 1374. Goodwin, T.J., Leppert, V.L., Risbud, H., Kennedy, I.M. and Lee, H.W.H. (1997) Applied Physics Letters, 70, 3122. Preschilla, N., Major, S., Kumar, N., Samajdar, I. and Srinivasa, R.S. (2000) Applied Physics Letters, 77, 1861. Janic, J.F. and Well, R.L. (1996) Chemistry of Materials, 8, 2708. Coffer, J.L., Johnson, M.A., Zhang, L. and Wells, R.L. (1991) Chemistry of Materials, 9, 2671. Micic, O.I., Ahrenkiel, S.P., Bertram, D. and Nozik, A.J. (1999) Applied Physics Letters, 75, 478. Benaissa, M., Gonsalves, K.E. and Rangarajan, S.P. (1997) Applied Physics Letters, 71, 3685. Micic, O.I., Ahrenkiel, S.P., Bertram, D. and Nozik, A.J. (1999) Synthesis, structure, and optical properties of colloidal GaN quantum dots. Applied Physics Letters, 75 (4), 478–480. Yu, H., Li, J., Loomis, R.A., Wang, L.-W. and Buhro, W.E. (2003) Two- versus three-dimensional quantum
j811
j 3 Growth and Growth Methods for Nitride Semiconductors
812
1166
1167
1168
1169
1170
1171
1172
1173
1174
confinement in indium phosphide wires and dots. Nature Materials, 2, 517–2520. Park, W.I., Yi, G.-C., Kim, M. and Pennycook, S.J. (2003) Quantum confinement observed in ZnO/ZnMgO nanorod heterostructures. Advanced Materials, 15, 526–529. Choi, H.-J., Johnson, J.C., He, R., Lee, S.K., Kim, F., Pauzauskie, P., Goldberger, J., Saykally, R.J. and Yang, P. (2003) Selforganized GaN quantum wire UV lasers. The Journal of Physical Chemistry B, 107, 8721–8725. Chen, L.C., Chen, K.H. and Chen, C.-C. (2003) Group III- and group IV-nitride nanorods and nanowires, in Nanowires and Nanobelts – Materials Properties and Devices: Metal and Semiconductor Nanostructures, vol. 1 (ed. Z.L. Wang), Kluwer Academic Publishers, pp.257–309. Law, M. Goldberger, J. and Yang, P. (2004) Semiconductor nanowires and nanotubes. Annual Review of Materials Research, 34, 83–122. Wagner, R.S. and Ellis, W.C. (1964) Vapor–liquid–solid mechanism of single crystal growth. Applied Physics Letters, 4, 89–90. Morkoc, H., Stamberg, R. and Krikorian, E. (1982) Whisker growth during epitaxy of GaAs by molecular beam epitaxy. Japanese Journal of Applied Physics, 21, L230–L232. Givargizov, E.I. (1975) Fundamental aspects of VLS growth. Journal of Crystal Growth, 31, 20–30. Westwater, J., Gosain, D.P., Tomiya, S., Usui, S. and Ruda, H. (1997) Growth of silicon nanowires via gold/silane vapor–liquid–solid reaction. Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 15, 554–57. Duan, X. and Lieber, C.M. (2000) General synthesis of compound semiconductor nanowires. Advanced Materials, 12, 298–302.
1175 He, M., Minus, I., Zhou, P., Mohammad, S.N., Halpern, J.B., Jacobs, R., Sarney, W.L., Salamanca-Riba, L. and Vispute, R.D. (2000) Growth of largescale GaN nanowires and tubes by direct reaction of Ga with NH3. Applied Physics Letters, 77 (23), 3731–3733. 1176 Wu, Y. and Yang, P. (2001) Direct observation of vapor–liquid–solid nanowire growth. Journal of the American Chemical Society, 123, 3165–3166. 1177 Tang, C.C., Fan, S.S., Dang, H.Y., Li, P. and Liu, Y.M. (2000) Applied Physics Letters, 77, 1961. 1178 Han, W.Q. and Zettl, A. (2002) Applied Physics Letters, 80, 303. 1179 Chang, K.W. and Wu, J.J. (2002) The Journal of Physical Chemistry B, 106, 7796. 1180 Gudiksen, M.S. and Lieber, C.M. (2000) Diameter-selective synthesis of semiconductor nanowires. Journal of the American Chemical Society, 122, 8801–2880. 1181 Wu, Y., Yan, H., Huang, M., Messer, B., Song, J.H. and Yang, P. (2002) Inorganic semiconductor nanowires: rational growth, assembly, and novel properties. Chemistry – A European Journal, 8, 1260–1268. 1182 Liang, C.H., Chen, L.C., Hwang, J.S., Chen, K.H., Hung, Y.T. and Chen, Y.F. (2002) Applied Physics Letters, 81, 22. 1183 Parala, H., Devi, A., Hipler, F., Maile, E., Birkner,A.,Becker,H.W.andFischer,R.A. (2001) Journal of Crystal Growth, 231, 68. 1184 Stach, E.A., Pauzauskie, P.J., Kuykendall, T., Goldberger, J., He, R. and Yang, P. (2003) Watching GaN nanowires grow. Nano Letters, 3 (6), 867–869. 1185 Zhou, S.M., Feng, Y.S. and Zhang, L.D. (2003) A physical evaporation synthetic route to large-scale GaN nanowires and their dielectric properties. Chemical Physics Letters, 369, 610–614. 1186 Dai, Z.R., Pan, Z.W. and Wang, Z.L. (2003) Novel nanostructures of functional oxides synthesized by thermal
References
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
evaporation. Advanced Functional Materials, 13, 9–24. Frank, F.C. (1949) The influence of dislocations on crystal growth. Discussions of the Faraday Society, 23, 48–54. Zhang, R.-Q., Lifshitz, Y. and Lee, S.-T. (2003) Oxide-assisted growth of semiconducting nanowires. Advanced Materials, 15, 635–640. Miehr, A., Ambacher, O., Rieger, W., Metzger, T., Born, E. and Fischer, R.A. (1996) Chemical Vapor Deposition, 2, 51. Jones, A.C., Rushworth, S.A., Houlton, D.J., Roberts, J.S., Roberts, V., Whitehouse, C.R. and Critchlow, G.W. (1996) Chemical Vapor Deposition, 2, 5. Buhro, W.E., Hickman, K.H. and Trentler, T.J. (1996) Advanced Materials, 8, 685. Pan, Z.W., Dai, Z.R. and Wang, Z.L. (2001) Nanobelts of semiconducting oxides. Science, 291, 1947–1449. Yan, H., He, R., Pham, J. and Yang, P (2003) Morphogenesis of onedimensional ZnO nano- and microcrystals. Advanced Materials, 15, 402–405. Yan, H., He, R., Johnson, J., Law, M., Saykally, R.J. and Yang, P. (2003) Dendrite nanowire UV laser array. Journal of the American Chemical Society, 5, 4728–4729. Yoshizawa, M., Kikuchi, A., Fujita, N., Kushi, K., Sasamoto, H. and Kishino, K. (1998) Self-organization of GaN/ Al0.18Ga0.82N multi-layernano-columns on (0 0 0 1) Al2O3 by RF molecular beam epitaxy for fabricating GaN quantum disks. Journal of Crystal Growth, 189–190, 138–141. Sun, Y., Yin, Y., Mayers, B.T., Herricks, T. and Xia, Y. (2002) Uniform silver nanowires synthesis by reducing AgNO3 with ethylene glycol in the presence of seeds and poly(vinyl pyrrolidone). Chemistry of Materials, 14, 4736–4745. Dingman, S.D., Rath, N.P., Markowitz, P.D., Gibbons, P.C. and Buhro, W.E.
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208 1209 1210 1211
(2000) Angewandte Chemie-International Edition, 39, 1470. Vayssieres, L., Keis, K., Lindquist, S.E. and Hagfeldt, A. (2001) Purpose-built anisotropic metal oxide material: three-dimensional highly oriented microrod array of ZnO. The Journal of Physical Chemistry B, 105, 3350–3352. Greene, L.E., Law, M., Goldberger, J., Kim, F., Johnson, J.C., Zhang, Y., Saykally, R.J. and Yang, P. (2003) Lowtemperature/wafer-scale production of ZnO nanowire arrays. Angewandte Chemie-International Edition, 42, 3031–3034. Hashimoto, A., Motiduki, T., Wada, H. and Yamamoto, A. (1998) Materials Science Forum, 264, 1129. Cheng, G.S., Zhang, L.D., Zhu, Y., Fei, G.T., Li, L., Mo, C.M. and Mao, Y.Q. (1999) Applied Physics Letters, 75, 2455. Han, W., Fan, S., Li, Q. and Hu, Y. (1997) Synthesis of gallium nitride nanorods through a carbon nanotube-confined reaction. Science, 277, 1287–1289. Frosch, C.J. and Thurmond, C.D. (1958) The Journal of Physical Chemistry, 62, 611. Frosch, C.J. and Thurmond, C.D. (1962) The Journal of Physical Chemistry, 66, 877. Balkas, C.M. and Davis, R.F. (1996) Journal of the American Ceramic Society, 79, 2309. Zhang, Y., Liu, J., He, R., Zhang, Q., Zhang, X. and Zhu, J. (2001) Chemistry of Materials, 13, 3899. Liu, J., Zhang, X., Zhang, Y., He, R. and Zhu, J. (2001) Journal of Materials Research, 16, 3133. Martin, C.R. (1994) Science, 266, 1961. Nishizawa, M., Menon, V.P. and Martin, C.R. (1995) Science, 268, 700. Jirage, K.B., Hulteen, J.C. and Martin, C.R. (1997) Science, 278, 655. Lee, S.B., Mitchell, D.T., Trofin, L., Nevanen, T.K., Soderlund, H. and Martin, C.R. (2002) Science, 296, 2198.
j813
j 3 Growth and Growth Methods for Nitride Semiconductors
814
1212 Masuda, H. and Fukuda, K. (1995) Science, 268, 1466. 1213 Masuda, H. and Hasegawa, F. (1997) Journal of Electrochemical Society Interface, 144, L127. 1214 Saito, M., Kirihara, M., Taniguchi, T. and Miyagi, M. (1994) Applied Physics Letters, 55, 607. 1215 Routkevich, D., Tager, A.A., Haruyama, J., Almawlawi, D., Moskovits, M. and Xu, J.M. (1996) IEEE Transactions on Electron Devices, 42, 1646. 1216 Al-Mawlawi, D., Liu, C.Z. and Moskovits, M. (1998) Journal of Materials Research, 9, 1014. 1217 Sapp, S.A., Lakshmi, B.B. and Martin, C.R. (1999) Advanced Materials, 11, 402. 1218 Zhang, J., Peng, X.S., Wang, X.F., Wang, Y.W. and Zhang, L.D. (2001) Chemical Physics Letters, 345, 372. 1219 Cheng, G.S., Chen, S.H., Zhu, X.G., Mao, Y.Q. and Zhang, L.D. (2000) Materials Science & Engineering A: Structural Materials: Properties, Microstructure and Processing, 286, 165. 1220 Zhang, J., Zhang, L.D., Wang, X.F., Liang, C.H., Peng, X.S. and Wang, Y.W. (2001) Journal of Chemical Physics, 115, 5714. 1221 Maya, L. (1986) Advanced Ceramic Materials, 1, 150; Maya, L. (1986) Process for preparing transition metal nitrides and transition metal carbonitrides and their reaction intermediates, Report PAT-APPL-6-939-920, 10 pp. 1222 Kimura, I., Ichiya, K., Ishii, M., Hotta, N. and Kitamura, T. (1989) Synthesis of fine AlN powder by a floating nitridation technique using an N2/NH3 gas mixture. Journal of Materials Science Letters, 8, 303–304. 1223 Selvaduray, G. and Sheet, L. (1993) Aluminium nitride: review of synthesis methods. Journal of Materials Science & Technology, 9, 463–473. 1224 Chen, C.-C., Yeh, C.-C., Chen, C.-H., Yu, M.-Y., Liu, H.-L., Wu, J.-J., Chen, K.-H. Chen, L.-C. Peng, J.-Y. and Chen, Y.-F.
1225
1226
1227
1228
1229
1230
1231
1232
1233
(2001) Catalytic growth and characterization of gallium nitride nanowires. Journal of the American Chemical Society, 123, 2791–2798. Wu, Y., Fan, R. and Yang, P. (2002) Blockby block growth of single-crystalline Si/ SiGe superlattice nanowires. Nano Letters, 2, 83–86. Gudiksen, M.S., Lauhon, U.J., Wang, J., Smith, D.C. and Lieber, C.M. (2002) Growth of nanowire superlattice structures for nanoscale photonics and electronics. Nature, 415, 617–620. Bjoerk, M.T., Ohlsson, B.J., Sass, T., Persson, A.I., Thelander, C., Magnusson, M.H., Deppert, K., Wallenberg, L.R. and Samuelson, L. (2002) One-dimensional steeplechase for electrons realized. Nano Letters, 2, 87–89. Solanki, R., Huo, J., Freeouf, J.L. and Miner, B. (2002) Atomic layer deposition of ZnSe/CdSe superlattice nanowires. Applied Physics Letters, 81, 3864–3866. Keating, C.D. and Natan, M.J. (2003) Striped metal nanowires as building blocks and optical tags. Advanced Materials, 15, 451–454. Valizadeh, S. Hultman, L. George, J.-M. and Leisner, P. (2002) Template synthesis of Au/Co multilayered nanowires by electrochemical deposition. Advanced Functional Materials, 12, 766–772. Kim, H.-M., Kang, T.W. and Chung, K.S. (2003) Nanoscale ultraviolet-lightemitting diodes using wide-bandgap gallium nitride nanorods. Advanced Materials, 15, 567–569. Cheng, G., Kolmakov, A., Zhang, Y., Moskovits, M., Munden, R., Reed, M.A., Wang, G., Moses, D. and Zhang, J. (2003) Current rectification in a single GaN nanowire with a well-defined p–n junction. Applied Physics Letters, 83 (8), 1578–1580. Nam, C.Y., Tham, D. and Fischer, J.E. (2004) Effect of the polar surface on GaN nanostructure morphology and growth
References
1234
1235
1236
1237
1238
1239
1240
1241
orientation. Applied Physics Letters, 85 (23), 5878–5880. Cheng, G.S., Zhang, L.D., Zhu, X.G., Chen, S.H., Li, Y., Zhu, Y. and Fei, G.T. (1999) Synthesis of orderly nanostructureof crystalline GaN nanoparticles on anodic porous alumina membrane. Nanostructured Materials, 11 (3), 421–426. Wang, T., Ranalli, F., Parbrook, P.J., Airey, R., Bai, J., Rattlidge, R. and Hill, G. (2005) Fabrication and optical investigation of a high-density GaN nanowire array. Applied Physics Letters, 86, 103103. Kuykendall, T., Pauzauskie, P., Lee, S., Zhang, Y., Goldberger, J. and Yang, P. (2003) Metalorganic chemical vapor deposition route to GaN nanowires with triangular cross sections. Nano Letters, 3, 1063–1066. Khanderi, J., Wohlfart, A., Parala, H., Devi, A., Hambrock, J., Birkner, A. and Fischer, R.A. (2003) MOCVD of gallium nitride nanostructures using (N3)2Ga{(CH2)3NR2}, R ¼ Me, Et, as a single molecule precursor: morphology control and materials characterization. Journal of Materials Chemistry, 13, 1438–1446. Ha, B., Seo, S.H., Cho, J.H., Yoon, C.S., Yoo, J., Yi, G.-C., Park, C.Y. and Lee, C.J. (2005) Optical and field emission properties of thin single-crystalline GaN nanowires. The Journal of Physical Chemistry B, 109 (22), 11095–11099. Johnson, J.C., Choi, H.-J., Knutsen, K.P., Schaller, R.D., Yang, P. and Saykally, R.J. (2002) Single gallium nitride nanowire lasers. Nature Materials, 1, 106–110. Gradecak, S., Qian, F., Li, Y., Park, H.-G. and Lieber, C.M. (2005) GaN nanowire lasers with low lasing thresholds. Applied Physics Letters, 87, 173111. Greytak, A.B., Barrelet, C.J., Li, Y. and Lieber, C.M. (2005) Semiconductor nanowire laser and nanowire waveguide electro-optic modulators. Applied Physics Letters, 87, 151103.
1242 Qian, F., Gradecak, S., Li, Y., Wen, C.-Y. and Lieber, C.M. (2005) Core/multishell nanowire heterostructures as multicolor, high-efficiency light-emitting diodes. Nano Letters, 5, 2287–2291. 1243 Kim, J.-R., Oh, H., So, H.M., Kim, J. and Kim, J.-J. (2003) Rectifying diode made of individual gallium nitride nanowire. Physica E, 18, 25–26. 1244 Huang, Y., Duan, X., Cui, Y. and Lieber, C.M. (2002) Gallium nitride nanowire nanodevices. Nano Letters, 2 (2), 101–104. 1245 D€ urkop, T., Getty, S.A., Cobas, E. and Fuhrer, M.S. (2004) Nano Letters, 4 (1), 35–39. 1246 Chen, R.-S., Chen, H.-Y., Lu, C.-Y., Wang, S.-W., Lan, Z.-H., Tsai, J.T.-H., Chen, L.-C., Chen, K.-H., Liu, Y.-T., Huang, Y.-S., Chen, Y.-F., Yang, Y.-J. and Chen, C.-C. (2007) Ultrahighresponsivity UV photodetector of bridging GaN nanowires. private communication. 1247 Jiang, H.X. and Lin, J.Y. (2003) III-nitride quantum devices microphotonics. Critical Reviews in Solid State and Materials Sciences, 28, 131–183. 1248 Tu, L.W., Hsiao, C.L., Chi, T.W., Lo, I. and Hsieh, K.Y. (2002) Self-assembled vertical GaN nanorods grown by molecular-beam epitaxy. Applied Physics Letters, 82 (10), 1601. 1249 Cerutti, L., Ristic, J., Fernandez-Garrido, S., Calleja, E., Trampert, A., Ploog, K.H., Lazic, S. and Calleja, J.M. (2006) Wurtzite GaN nanocolumns grown on Si (0 0 1) by molecular beam epitaxy. Applied Physics Letters, 88, 213114-1. 1250 Ristic, J., Calleja, E., Trampero, A., Fernández-Garrido, S., Rivera, C., Jahn, U. and Ploog, K.H. (2005) Columnar AlGaN/GaN nanocavities with AlN/GaN bragg reflectors grown by molecular beam epitaxy on Si(1 1 1). Physical Review Letters, 94, 146102-1. 1251 Yamano, K., Kikuchi, A. and Kishino, K. (2005) Self-organized GaN/AlN
j815
j 3 Growth and Growth Methods for Nitride Semiconductors
816
1252
1253
1254
1255
1256
1257
superlattice nanocolumn crystals grown by rf-MBE, in GaN, AlN, InN and Their Alloys, Materials Research Society, Warrendale, PA, USA, Materials Research Society Symposium Proceedings, 831, 483–488. Kishino, K. and Kikuchi, A. (2005) Highly efficient blue to red emissions of InGaN/GaN nano-disks integrated into GaN nanocolumns. 2005 IEEE LEOS Annual Meeting, IEEE, Piscataway, NJ, USA, pp.631–632. Kikuchi, A., Kawai, M., Tada, M. and Kishino, K. (2004) InGaN/GaN multiple quantum disk nanocolumn lightemitting diodes grown on (1 1 1) Si substrate. Japanese Journal of Applied Physics, Part 2: Letters, 43 (12), L1524–L1526. Kikuchi, A., Yamano, K., Tada, M. and Kishino, K. (2004) Stimulated emission from GaN nanocolumns. Physica Status Solidi b: Basic Research, 241 (12), 2754–2758. Goldberger, J., He, R., Zhang, Y., Lee, S., Yan, H., Choi, H.-J. and Yang, P. (2003) Single-crystal gallium nitride nanotubes. Nature, 422, 599–602. Lauhon, L.J., Gudiksen, M.S., Wang, D. and Lieber, C.M. (2002) Epitaxial core–shell and core–multishell nanowire heterostructures. Nature, 420, 57–61. Qian, F., Li, Y., Gradecak, S., Wang, D., Barrelet, C.J. and Lieber, C.M. (2004)
1258
1259
1260
1261
1262
Gallium nitride-based nanowire radial heterostructures for nanophotonics. Nano Letters, 4 (10), 1975–1979. He, M., Minus, I., Zhou, P., Mohammad, S.N., Halpern, J.B., Jacobs, R., Sarney, W.L., SalamancaRiba, L. and Vispute, R.D. (2000) Growth of large-scale GaN nanowires and tubes by direct reaction of Ga with NH3. Applied Physics Letters, 77 (23), 3731–3733. He, R., Law, M., Fan, R., Kim, F. and Yang, P. (2002) Functional bimorph composite nanotapes. Nano Letters, 2, 1109–1112. Wu, Q., Hu, Z., Wang, X., Lu, Y., Chen, X., Xu, H. and Chen, Y. (2003) Synthesis and characterization of faceted hexagonal aluminum nitride nanotubes. Journal of the American Chemical Society, 125, 10176–1077. Fan, R., Wu, Y., Li, D., Yue, M., Majumdar, A. and Yang, P. (2003) Fabrication of silica nanotube arrays from vertical silicon nanowire templates. Journal of the American Chemical Society, 125, 5254–5255. Han, W., Redlich, P., Ernst, F. and R€ uhle, M. (2000) Synthesis of GaN–carbon composite nanotubes and GaN nanorods by arc discharge in nitrogen atmosphere. Applied Physics Letters, 76 (5), 652–654.
j 817
4 Extended and Point Defects, Doping, and Magnetism Introduction
No matter how perfected semiconductors are or would be, defects to varying degrees are still present. This is particularly acute in semiconductors in their early stages of development and particularly so in GaN due to lack of native substrates. In fact, investigations of defects remain a topic for discussion for nearly as long as the semiconductor in question is in use and attracts attention. To highlight the case in a light manner, case tracking of international conferences on defects in semiconductors bear the suggestion that defects issues in semiconductors may never be fully solved [1]. As mentioned, due to a variety of reasons, chiefly due to the nonavailability of native substrates in the commercial sense, GaN epitaxial layers have been mainly grown on substrates such as sapphire, SiC, GaAs, and Si with sapphire being the dominant substrate, particularly for light-emitting diodesLEDs. To improve the quality of layers, much attention has been devoted to the understanding of the particulars of heteroepitaxy. In this vein, various planes of sapphire as well as its vicinal planes have been investigated, as detailed in Chapter 3. The epitaxial relationships between the various substrate orientations and the nitride films grown on them have been investigated, as reported in the same. There is also a thermal mismatch in heteroepitaxy that manifests itself as additional structural defects that are created during cooldown from the growth temperature and during substantial change in growth temperature, or as strain if any defect formed during cooldown is not sufficient to cause complete relaxation. Strain causes a modification of the band structures, and if controlled can be advantageous. GaN is a hard material, and a sufficiently thick film can actually crack the substrate. Transmission electron microscopy (TEM) has been employed to directly observe structural defects in GaN films. To investigate the effect of misorientation on the crystal morphology, Hiramatsu et al. [2] grew GaN on (0 0 0 1) sapphire with a number of different misorientations toward both the ½1 0 1 0 and the ½1 2 1 0 directions, with the exact (0 0 0 1) basal plane leading to the best surface morphology. TEM analysis of wurtzite GaN grown on sapphire substrates [3,4] indicated to no surprise that the
Handbook of Nitride Semiconductors and Devices. Vol. 1. Hadis Morkoc Copyright 2008 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim ISBN: 978-3-527-40837-5
818
j 4 Extended and Point Defects, Doping, and Magnetism major structural defects in GaN are those related to stacking mismatch and dislocations. GaN grown on 6H-SiC would also suffer from the same types of defects, whose density depends among others on the step density. Owing to the stacking order match between GaN and ZnO substrates, stacking mismatch defects are less likely to occur if proper surface preparation procedures could be developed and the chemical reaction be avoided. It should be pointed out that the electrical and optical properties of the material and devices fabricated in it depend on the density and nature of point defects. Point defects are intertwined with extended defects in that complexes may form and/or strain fields induced by extended defect could gather impurities and charges and induce charge because of the piezoelectric (PZE) nature of the material. This is true even if the dislocation cores are fully coordinated with no dangling bonds. Following the discussion of extended defects and their charge nature, point defects are discussed from a theoretical point of view in terms of their formation energies and manifestation of defect in deep-level transient spectroscopic measurements. Doping and point defects are synonymous with each other. Therefore, a considerable effort is expended on this topic. In addition to doping to render the semiconductor either n- or p-type, doping with magnetic impurities that have spin-unpaired electrons is also discussed.
4.1 A Primer on Extended Defects
GaN layers contain large densities of crystallographic defects, among which threading dislocations (TDs) [5,6], the nanopipes [7], the inversion domains [8,9], and the pyramidal planar defects [8,10,11] can cross the whole epitaxial layer and be detrimental to the electrical and optical properties. As pointed out by many workers, a large majority of these defects are made of threading dislocations that originate from the particular growth mode of GaN on top of the (0 0 0 1) sapphire or SiC substrates. This mosaic growth mode leads to islands that are rotated mostly around the c-axis, and therefore, are bounded by mainly a edge dislocations [12,13]. In a conventional semiconductor such defect densities on the order of 1010 cm2 would result in nonusable layers. In GaN-related materials, these high densities of defects, which are bound to lead to the formation of a nonradiative recombination center as well as scattering centers, do not prevent high-performance devices such as LEDs to be fabricated. However, for lasers and FETs, the defects of materials form a formidable obstacle. For preparing lasers techniques such as lateral epitaxial overgrowth [14], discussed in detail in Section 3.5.5.2, are required to lower the dislocation density that could be lower than 106 cm2, with multiple lateral epitaxial growth sequences, in conjunction with thick or freestanding GaN layers grown by hydride vapor phase epitaxy (HVPE). Stacking faults (SFs) represent the planar defects that occur in nitrides, in part due to substrates that do not share the same stacking order as the nitrides. The f1 2 1 0g planar defects have been called translation domain boundaries (TDBs) [15–17], double positioning boundaries (DPBs) [10,18], stacking mismatch boundaries (SMBs) [19], or
4.1 A Primer on Extended Defects
inversiondomainboundaries (IDBs) [8].These faultshavebeeninvestigatedusing highresolution transmission electron microscopy (HREM) and convergent beam electron diffraction (CBED), and it was shown that stacking faults are present on top of both sapphire and SiC [20,21]. In fact, these planar defects have already been studied in the 1960s, and two displacement vectors have been measured by conventional microscopy[22,23].InceramicAlN,Drum [23]investigatedfaults thatintersected onthe basal and prismatic f1 2 1 0g planes. It was shown that the displacement vector was 1/2h1 0 1 0i and that theyfolded tothe basal planesbyleaving a1/6h1 0 1 0i stair rod dislocation atthe intersection. Almost simultaneously, Blank et al. [22] were the first to study the planar defects that folded from the basal to pyramidal f1 2 1 0g planes in wurtzite ZnS, and to interpret them as stacking faults, whereas other authors considered them to be thin lamella of the sphalerite phase in CdS [24]. These pyramidal faults were then shown to be growth domains, and the displacement vector was found to be 1/6h2 0 2 3i, which is the same as that of the I1 basal-stacking fault in the hexagonal compact packed (hcp) structure. A succinct description of the abovementioned defects is given further in the text. The reader can refer to Ref. [25] for details. 4.1.1 Dislocations
As stated in the previous section and throughout Chapters 1 and 3, the large difference in lattice mismatch and in thermal expansion coefficients among the materials involved, result in a large number of dislocations, some of which alleviate the strain. Unless special precautions are taken, as discussed in Section 3.5.5.2, such as epitaxial lateral overgrowth (ELO), the GaN films typically have threading dislocation density of 108–1010 cm2. Threading dislocations in GaN could be the basis for the formation of nonradiative recombination and scattering centers that in turn affect carrier mobility [26–28]. Dislocations can be grouped into misfit dislocations, with fully accommodated strain and interface bound, and threading dislocations. Among the threading dislocations are pure edge, pure screw, and mixed dislocations, which are discussed in Section 4.1.1. In GaN as in any semiconductor, there are basically two types of dislocations: basal plane dislocations and threading dislocations. Dislocations in the pyramidal plane of hexagonal GaN have not seen as much theoretical work as the basal plane dislocations that have line direction along l ¼ [0 0 0!1] and Burgers vectors [0 0 0 1] for the threading screw dislocation and (1/3)½1 2 1 0 for the threading edge dislocation. Basal plane dislocations in GaN are dislocations of the main slip system of the crystal, ! in hexagonal material the {0 0 0 1}h1 1 2 0i system ({1 1 1}i1 1 0h in the cubic system in the cubic symmetry). In these slip systems dislocations inhabit a {0 0 0 1} ({1 1 1} in ! the cubic system) glide plane and possess h1 1 2 0i (h1 1 0i in the cubic system) type line directions and Burgers vectors (as mentioned above). Density functional theory based (DFTB) tight-binding calculations have been applied to treat the basal plane dislocations in GaN (see Ref. [29] for a comprehensive review). In DFTB calculations the electronic wave functions are approximated by a linear combination of atomic orbitals (as in LCAO) involving a minimal basis set of s- and
j 819
820
j 4 Extended and Point Defects, Doping, and Magnetism p-valence orbitals that also include the Ga-3d orbitals in the case of GaN. The two center Hamiltonians and overlap matrix elements result from atom-centered valence electron orbitals and the atomic potentials from single-atom calculations in the frame of the density functional theory (DFT). Exchange and correlation contributions in the total energy (TE) as well as the ionic core–core repulsion are taken into account by a repulsive pair potential. The latter is obtained by comparing DFT calculations. Therefore, the DFTB method can be thought of as an approximate density functional approach in which all DFT integrals (Hamiltonian and overlap matrix) as well as the repulsive potential can be calculated in advance and are tabulated over the interatomic distance of two atoms. Doing so considerably reduces the computational complexity, allowing the treatment of larger models. In materials with strong ionic bonding character and strong charge transfer, the approximations of standard DFT fail the mere superposition of electron densities of neutral free atoms does not describe the electron density of the system. Charge transfer between atoms of different electronegativities necessitates an explicit treatment of long-range Coulomb interactions, which to some extent are included in the self-consistent charge extension of the DFTB method (SCC-DFTB). The charge transfer is typically approximated by spherical atom-centered charge fluctuations that occur in the respective total energy expressions up to the second-order terms. When modeling extended defects, particularly dislocations, medium and long-range stress/strain effects are crucial and must be taken into consideration. A case in point is that the elastic strain energy contained in a cylinder of radius R around a dislocation diverges with ln(R/A ) and forces associated with dislocation–dislocation interaction drop only with R1, where R is the distance between the two dislocations. Consequently, modeling dislocations requires large supercells or clusters for a good representation of the surrounding bulk material or, in case of a supercell approach, for minimizing the dislocation–dislocation interaction with the periodic images in neighboring cells. The simplicity of the SCC-DFT method proves to be a major advantage over the less approximate and computationally more demanding methods: in todays workstations the SCC-DFTB method as described above allows a structural relaxation of systems containing more than 700 atoms. The DFTB method has been applied to model dislocation core structures and energies in various semiconductor materials [30–32] and was found to agree with DFT pseudopotential calculations (AIMPRO [33,34]). Moreover, excellent agreement was found with other DFT calculations [35] carried out on the 90 glide partial dislocation in diamond [31]. In this chapter, the results of DFT-based calculation of perfect edge and screw dislocations are discussed along with experimental observations. For an in-depth discussion on dislocation, a schematic representation of edge and screw of dislocations is shown in Figure 4.1. For illustration, determination of the Burgers vector for an edge dislocation is shown in Figure 4.2. The geometry of all three types of dislocations are shown in Figure 4.3 along with the unit cell of GaN that in the ideal case has the relationship of c ¼ 1.62a [36]. Identification of dislocations (edge, screw, and mixed), can be accomplished with
4.1 A Primer on Extended Defects
z
M
L
N
O
y
x
b
b
(a)
z
r0 R
A M
b
b
L
B
x
O
N r
y
(b) Figure 4.1 Schematic representation of edge and screw dislocations. Courtesy of Liliental-Weber.
bright field TEM images under two beam conditions, that is, for g-vector parallel to the [0 0 0 2] and ½1 1 2 0 directions, respectively. For g-vector parallel to the [0 0 0 2] direction, the edge dislocations are because of the contrast and are not observed. However, for g-vector parallel to the ½1 1 2 0 direction, the same applies for the screw dislocations. On the contrary, mixed-type dislocations appear in both diffraction conditions.
j 821
822
j 4 Extended and Point Defects, Doping, and Magnetism QM
P
O
N
(a) P
M
Q
Burgers vector O
N
(b) Figure 4.2 Schematic description of an edge dislocation (a); determination of Burgers vector for an edge dislocation (b). Courtesy of Z. Liliental-Weber.
4.1.1.1 Misfit Dislocations Dislocations can be divided into two groups: misfit dislocations and threading dislocations. Misfit dislocations are confined to the interface between the GaN epitaxial layer and the substrate interface. To a first extent, misfit dislocations are not supposed to thread up into the layer and are very efficient in relieving strain. However, it has been shown that threading dislocations can be interconnected to the misfit dislocation network [37]. In heteroepitaxy, as has been demonstrated in GaAs epitaxy on Si substrates, these are the best types of dislocations to have. As the misfit in the in-plane lattice constant between the epitaxial film and the substrate increases, the spacing of misfit dislocations decreases, and for high values of misfit the interface is incoherent. If the misfit is not large, the interfacial area of GaN/sapphire can exhibit zones without extended defects, but even then, it is not easy to locate the misfit dislocations at the interface on a high-resolution image. To visualize them, one has to filter the images. A Fourier filtering shows that regularly spaced 60 dislocations are present. If one assumes the core to be located at the interface, the measured average distance of 2 nm shows that they define a stepless relaxed area (Figure 4.4). 4.1.1.2 Threading Dislocations As the nomenclature suggests, threading dislocations propagate through the layer, reach the surface of the film except those that annihilate each other, and are detrimental to the electrical and optical properties of epitaxial layers. The TEM investigations have shown that these dislocations originate at the interface and the large majority propagates to the sample surface [25,38]. There are three types of threading dislocations in GaN: edge, screw, and mixed varieties. The edge
4.1 A Primer on Extended Defects
{1 1 0 0} prismatic planes
3.2 Å
I
6.2Å
B
Edge dislocation
(1 0 0 0) basal plane (a)
(b)
B
I
I
B
Mixed dislocation Screw dislocation (c) Figure 4.3 Schematic representations of (a) GaN unit cell, edge dislocation with a Burgers vector of b ¼ 1/3h1 1 2 0i and [0 0 0 1] line direction, (c) perfect screw dislocation with b ¼ h0 0 0 1i and [0 0 0 1], line direction and (d) mixed dislocation (c þ a type) with
(d) b ¼ 1/3h1 1 2 3i and inclined line direction 10 from [0 0 0 1] toward the Burgers vector. Here a represents edge and c represents screw dislocation. The term (c þ a) represents the mixed dislocations. Patterned after Mathis et al. [36].
dislocations are introduced to accommodate small angular deviations about the [0 0 0 1] axis. The screw dislocations accommodate deviations due to the overgrowth stepped substrate surface. In a-GaN grown along the [0 0 0 1] direction, the edge and screw dislocations, in general, can be characterized with their Burgers vectors bi ¼ ai ¼ 1/3h2 1 1 0i (i ¼ 1, 2, 3) and b ¼ c ¼ [0 0 0 1], respectively [38]. Here, ai represents the three lattice vectors on the basal plane of a hexagonal system, as shown in Figure 4.5. The pure edge and pure screw dislocations are termed as
j 823
824
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.4 Misfit dislocation at the GaN/sapphire interface after Fourier transform where the asterisks mark each misfit dislocation and the missing plane in GaN, every seventh lattice plane, as the in-plane lattice constant is larger than that of sapphire. Courtesy of P. Ruterana.
a3
a2
a1 Figure 4.5 Schematic illustration of the a-GaN structure (space group P63mc) in projection along [0 0 0 1], with symmetry elements superimposed. Open and filled circles denote distinct atom types at heights of 1/4[0 0 0 1]. Solid lines and dashed lines indicate projections of f11 2 0g mirror planes and
f0 1 1 0g glide mirror planes, respectively. The small circles denote inversion centers. The arrows indicate the twofold symmetry axes. The a2 ¼ 1/3 lattice translations a1 ¼ 1/3 ½2110, also indicated. ½1210, a3 ¼ 1/3 ½1120are Courtesy of Dimitrakopulos et al. [38].
a-type and c-type, both with,x as the line direction [0 0 0 1]. Mixed dislocations are termed c þ a type with b ¼ 1/3h1 1 2 3i, and line directions inclined 10 from [0 0 0 1] toward the Burgers vector. 4.1.1.2.1 Edge Dislocations Edge dislocations make a very large percentage of the overall dislocation density, particularly in materials grown by molecular beam epitaxy
4.1 A Primer on Extended Defects
(MBE). Pure screw dislocation density in organometallic vapor phase epitaxy (OMVPE) films is less than 10% of the total. In MBE-initiated films, this fraction is as low as 0.1% for films grown at moderate temperatures under very metal-rich conditions to as high as 20% of the total. The reported fraction of mixed dislocations is up to 2% of the total in all MBE films and nearly 70% for OMVPE films [36]. It has been shown that some of the c and a þ c dislocations tend to bend and annihilate; the density of those that reach the layer surface is only a small percentage of that near the interface [40]. This observation should be treated with caution in that although it may true that the dislocations in the initial part of the layer could be bent or annihilated as reported, this may not necessarily be true as the growth progresses, and also in much thicker films, the dislocation density still remains high. The core structure of a perfect edge dislocation and mixed dislocation has been shown by theory to be of the same eight-atom ring character. Moreover, the lowest energy state is represented by a full core. With first principle calculations for small cell size (small model), the dislocations initially were reported to be electrically inactive except for the effect of the strain fields and impurity gettering [41]. The results of large cell size calculations, however, indicate the case to be otherwise, which will be discussed further in the text. On the experimental side, using high-resolution Z-contrast imaging with a point resolution of 0.13 nm, Xin et al. [42] determined the core structure of a perfect edge dislocation shows that mixed dislocation has eightfold rings in projection along their line direction. The central column in the core of a pure edge dislocation contains an alternating chain of N and Ga atoms in much the same way as their dimers on f1 0 1 0g surfaces, consistent with that reported in Ref. [41]. In addition to the eight-atom ring core atomic configuration, Ruterana et al. [25] noted a contrast that is not consistent with the eight-atom core model. Although this configuration has more or less a frequency consistent with the eight-atom ring core, image simulations show that this core is more compatible with a five to seven atom ring configuration [43]. A discussion of the electronic nature of dislocations can be found in Section 4.1.6, and their impact on point defects can be found in Section 4.3. Threading edge dislocations lie on the f1 0 1 0g planes. They are a dominant species of dislocation in GaN prepared by both MBE and OMVPE. An edge dislocation is formed through an insertion or subtraction of an additional half-atom plane. An example of this is shown in Figure 4.6, which is a plan-view HREM micrograph of a threading edge dislocation in wurtzite GaN. Using the dark spots (atomic columns for the particular imaging condition), a closed right-hand circuit SGHIJKLS is drawn around the line direction x of the defect, which in this points into the page. The circuit is shown to be mapped into the reference space in Figure 4.6, whereby it is seen that closure failure FS arises. The origin, O, has been chosen to be at a location corresponding to a center of symmetry of the holosymmetric parent structure [38]. The core structure of a threading edge dislocation is shown in Figure 4.7 [44]. The relative bond extension and compression at the core are about 9–15% [42,44,45]. The line energy for a threading edge dislocation was found to be 2.19 eV Å1, which is considerably lower than that for a screw dislocation with a narrow opening. The
j 825
826
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.6 (a) Plan-view HREM micrograph of an edge threading dislocation in a-GaN with [0 0 0 1] line direction. A closed right-handed Burgers circuit has been drawn around the line direction of the defect (defocus ¼ 29 nm, thickness ¼ 7.8 nm, atomic columns correspond to dark spots). (b) Mapping of the Burgers circuit to the reference space. Closure
failure FS identifies the defect as an edge dislocation with Burgers vector b ¼ 1/3 < 2 11 0> ([0 0 0 1] projection). Large and small circles denote distinct atomic species. Open and filled circles denote atoms at heights 0 and ca-GaN/8 respectively. The term x depicts the line direction of the defect pointing into the page). Courtesy of Dimitrakopulos et al. [38].
elastic train energy of a dislocation is proportional to kb2. Here, b is the magnitude of the Burgers vector and the constant k is equal to unity for a screw dislocation, and 1/(1 n) for an edge dislocation, where n is Poissons ratio. The ratio of the elastic energies Escrew/Eedge is about 1.66. Therefore, the threading edge dislocation density in GaN films is normally much higher than that of screw dislocation [41,45]. It is instructive to say a few words on the nomenclature used. The terms core atoms,
4.1 A Primer on Extended Defects
5/7 open core
H G
B
3/4 5/6 7/8 9/10
F
Additional half-atom plane
C D
E
y=(1 0 1 0) x=(1 2 1 0) z=[0 0 0 1]
H G
B
1
3/4 5/6
A
2 7/8 9/10
F
C D
E Ga N
(b)
j 827
828
j 4 Extended and Point Defects, Doping, and Magnetism atoms bordering the core, and core region are not all that strictly defined. It could be argued that all atoms whose energy contribution does not follow the elastic limit within some error are to be considered as core atoms. In this case, this would be a classical definition from the elasticity considerations, in the case of which the core is not precisely defined either. Blumenau uses this definition only for the core radius of a dislocation that is defined by energetics. If one refers to core atoms though, this would be mostly in the context of bonding, electronics, and geometry. In this framework, therefore, one defines the core atoms as those that are heavily distorted in their bonding configuration. Although this is not a very precise definition, one could consider the eight-numbered ring of the full-core edge as the core atoms (in the closed-core structure shown in Figure 4.7, and for the open core it would be the 10-numbered ring of atoms, which is often referred to as 5/7 structure shown in the inset of Figure 4.7). 4.1.1.2.2 Screw Dislocations Screw dislocations in hexagonal GaN have a Burgers vector (c) parallel to the dislocation line [0 0 0 1]. Screw dislocations in GaN exist in different forms, that is, full-core screw dislocation, and screw dislocations with a narrow opening or in the form of nanopipes [45]. However, this is controversial in the sense that there exist samples wherein the cores or screw dislocations are full despite the calculations, indicating the open-core ones having lower line energy [46]. Both sides of the issue are discussed below. Figure 4.8 is a model of the core structure of a full-core screw dislocation. It was reported that heavily distorted bonds at the dislocation core could yield deep gap states ranging from 0.9 to 1.6 eV above the valence band maximum (VBM), and shallow gap states at about 0.2 eV below the conduction band minimum. Therefore, the full-core screw dislocation is deemed electrically active and could act as a nonradiative center (see Section 4.1.6 for a detailed discussion). The bond distortion in the open-core screw dislocation is significantly less than that in the full-core screw dislocation [41,46]. Figure 4.9a presents a sketch of a closedcore screw dislocation projected onto the (0 0 0 1) basal plane. The calculated band structure shows both filled and empty states throughout the gap, the deep varieties of which are due to the dangling bonds associated with the threefold coordinated N- and Ga-core atoms (atoms 1/2, e.g., in Figure 4.9b). The lower lying states in the gap are localized on the N-core atoms, whereas states close to the conduction are of mixed 3——————————————————————————— Figure 4.7 (a) High-resolution Z-contrast image of a relaxed edge dislocation. The dislocation core is shown in the boxed region. Courtesy of D. Wang. (b) Supercell containing two full-core pure edge dislocations with their core structure. The broken line in (b) indicates the additional half-atom plane. The threefold coordinated atoms 1 (Ga) and 2 (N) adopt a hybridization similar to the ð1010Þ surface atoms. Numbers 1–8 indicate an eightfold atom-column ring of the dislocation core.
Removal of the most distorted atoms, such as lines of atoms 9 and 10, and atoms 1,2, 3, 4, 5, 6, 7, 8, and their equivalents led to higher energies. The atomic columns in the core of the edge dislocation are indicated with capital letters A, B, C, . . ., H. Removal of column A atoms would lead to open-core edge dislocation, which is also referred to as 5/7 chain atom edge dislocation as shown in the boxed region (upper left). Courtesy of Blumenau et al. (patterned after Refs [29,41]).
4.1 A Primer on Extended Defects
Figure 4.8 Side view (projection onto the ð1 1 2 0Þ plane) of a relaxed and neutral screw dislocation: (a) full-core screw dislocation; (b) Ga-filled screw dislocation. Core of a full-core screw dislocation (discussed in greater detail in Figure 4.10 and the associated text) showing
the double helix of Ga bonds. The supercell is repeated twice in the [0 0 0 1] for clarity. Note that the bonds at the core are heavily distorted. Courtesy of Blumenau et al. (patterned after Ref. [29]). (Please find a color version of this figure on the color tables.)
j 829
830
j 4 Extended and Point Defects, Doping, and Magnetism Core Ga Ga N
[1 0 1 0] [1 2 1 0] z=[0 0 0 1]
(a)
Ga N
1/2 3/4
[1 0 1 0] [1 2 1 0] z = [0 0 0 1]
(b) Figure 4.9 The relaxed atomic structure of a screw dislocation projected in the (0 0 0 1) plane: (a) the full-core structure with the core in the center of the dashed circle; (b) the opencore structure of the same obtained by simply removing the atoms within the dashed circle in (a) before energy relaxation. When energy relaxation is performed, a core opening 7.2 Å in diameter results. See the associated text discussing the limitations of this simple removal of the atoms from the core regions in
constructing the figure. The threefold coordinated atoms 1 (N) and 2 (Ga) atoms adopt sp3 (p2) like hybridization, respectively, similar to the ð1 0 1 0Þ surface atoms, which lowers the surface energy and in so doing the central part of the gap is cleared off the deep states. The remaining states near the band edges are induced by the strain-induced distortion. Courtesy of A.T. Blumenau, J. Elsner, and R. Jones.
4.1 A Primer on Extended Defects
character involving Ga and N atoms, both full-core and Ga-filled-core screw dislocation are electrically active [29]. Although not strictly accurate, when the atoms encircled by the dashed circle in the core are removed, there is an open-core screw dislocation with an opening 7.2 Å in diameter (as shown in Figure 4.9b). For the full-core screw dislocation depicted in Figure 4.9a, one can observe a counterclockwise twist of the ring formed by the core atoms/columns, which also directly affects the region surrounding the core atoms. If one simply removes the central core region (consisting of six atom columns), the final structure of the open-core screw dislocation would not result. This structure still must be minimized in terms of energy. In other words, the structure must be allowed to be energetically relaxed. This relaxation will remove the twist mentioned above, and it will further move the undercoordinated Ga atoms away from the center, giving way to sp2-like hybridization. In Figure 4.9a, this hybridization cannot be seen because the coordinates are taken from the full core. This is not taken into consideration in Figure 4.9b, where all atoms have their relaxed position for the open core. However, the figures have been so constructed to show the general feature of an open-core screw dislocation in the first order. The open-core screw results from the balance of energy required to form a surface at the wall and energy gained by the removal of the atoms in the highly strained core region. The internal surfaces of the dislocation cores are similar to the ð1 0 1 0Þ facets except for the topological singularity required by a Burgers circuit. The Ga (N) atoms adopt sp2 (sp3) hybridization that lowers the surface energy and clears the central portion of the bandgap off the states. However, shallower states remain owing to the strain field by the dislocation. To verify the calculations, unperturbed ð1 0 1 0Þ surfaces are void of these states. It is therefore concluded that screw dislocations with large openings could be free of gap states. The TEM images of screw dislocations reported by Xin et al. [42] point to the filledcore variety being present. If one assumes that the core is filled in the stoichiometric sense, it would conflict with the predictions of Elsner et al. [41] that by a wide margin favor a small open-core dislocation having a diameter of 7.2 Å over the filled-core model. It should also be pointed out that the cell size used in Elsner et al. [41] may have inadvertently missed charge nature of this defect. Nevertheless, Northrup [47,48] attempted to reconcile this apparent discrepancy by arguing that a screw dislocation with a core comprised entirely of Ga atoms is preferred energetically over the hollowcore model in Ga-rich conditions. This so-called Ga-filled-core model is obtained by the removal of all six N atoms per unit cell along the c-axis (5.1 Å) within 2 Å of the center of the dislocation core. A screw dislocation is generated by applying a displacement in the c-direction to each atom in the cell. A schematic projected onto the (0 0 0 1) plane of the atoms within the supercell is shown in Figure 4.10. Because there are six Ga atoms and no N atoms at the core, this mode is also referred to as the 6:0 model. Other configurations besides this particular one were also considered by Northrup, but the 6:0 appears to be most energetically favorable. For this structure the electronic states are distributed throughout the bandgap, which implies that screw dislocations will be a nonradiative recombination center and electrically active if grown under Ga-rich conditions.
j 831
832
j 4 Extended and Point Defects, Doping, and Magnetism y
Core Ga atoms A sites B sites
x
Figure 4.10 The Ga-filled core is obtained by removing the nitrogen atoms from the central core. The Ga atoms in the core form a pair of helices, each with pitch 2c. The filled circles depict Ga and N atoms in the A sites of the wurtzite lattice. The open circles depict Ga and
N atoms in B sites. The H atoms used to saturate the 18 Ga and 18 N dangling bonds in each cell on the six f1 0 1 0g surfaces are not shown. The screw dislocation is generated by applying the displacement (0, 0, Dzi). Courtesy of Northrup [47].
Atomic force microscopy (AFM) view is shown in Figure 4.11 obtained from a screw dislocation in an MBE-grown GaN on HVPE GaN template, after etching in a KOH etch, showing the classical staircase behavior. The screw dislocation emanates from the GaN HVPE grown buffer layer. For comparison, the AFM image of a GaN layer grown by RF MBE on a c-plane low-dislocation high-pressure melt-grown GaN substrate is also shown with atomically stepped surfaces void of any screw dislocation. Whether the screw dislocation core in GaN is filled is not definitive, which means that it depends on the sample preparation method employed. Figure 4.12 shows the screw dislocation core structures as observed in plan-view configuration taken at Scherzer defocus by Liliental-Weber et al. [49,50]. Contrary to expectations, no displacement vector could be observed around screw dislocations because the displacement vector is along the c-axis. Voids, which have a hexagonal shape (lighter contrast in the central part of Figure 4.12), are observed surrounding the dislocation in the HVPE sample. However, the MBE GaN grown on HVPE GaN template (Figure 4.12) does not exhibit voids. This topic has been further discussed in Section 4.1.6. In an effort to interrogate whether the open-core and filled-core screw dislocations ensue in Ga-rich growth conditions, MBE-grown films prepared on HVPE templates under N- and Ga-rich conditions, as well as by OMVPE, were examined. Studying these films, Liliental-Weber et al. [49,50] reported that small voids along a dislocation line were present in HVPE-grown samples for gb ¼ 0. Here, the terms g and b represent scattering and Burgers vectors, respectively. Samples grown by two
4.1 A Primer on Extended Defects
Figure 4.11 (a) AFMviewobtained froma screw dislocation in an MBE-grown GaN on HVPE GaN template showing the classical staircase behavior. The dark spot in the center is the core of screw dislocation. Image size is 500 nm · 500 nm, and
the vertical scale is 1 nm. (b) AFM image of an RF MBE grown GaN epitaxial layer on c-plane highpressure melt grown GaN substrate. Note the atomically stepped surface and lack of screw dislocations. Courtesy of S. Porowski.
j 833
834
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.12 Plan-view images of screw dislocations in (a) MBE and (b) HVPE GaNs. Burgers circuits have no displacement, as expected. Note tilted c-planes around dislocation in (a) and void in (b). Courtesy of Z. Liliental-Weber.
different growth methods (HVPE and MBE) were studied in cross section. Under gb ¼ 0 diffraction conditions for which screw dislocations are out of contrast, small voids along a dislocation line were observed for HVPE-grown samples. These voids were not observed for edge dislocations in the same material and not at all in the MBE or OMVPE overlayer samples grown on top of the HVPE template. The images dealing with voids in HVPE material are shown in Figure 4.13. These voids in HVPE templates have a hexagonal shape. To determine whether the cores are nonstoichiometric, one looks for missing atomic columns expected for the empty core model and for changes in intensity of particular columns to possibly indicate a nonstoichiometric core. However, this is rather complex in that the intensity variations, which are what one looks for, can also be caused by misalignment and tilt. Even a 4 mrad tilt can change the relative intensity of two adjacent atomic columns by up to 4%, but this can be easily detected because pattern symmetry is changing to twofold. Larger tilts change the image pattern so drastically that they are easy to recognize visually without column intensity measurement. Moreover, the primary and diffracted beam intensities forming the image change with sample thickness as well as with stoichiometry. To avoid partially reversed image contrast, sample thickness must be such that all contributing beams have the same sign of phase [49]. This is particularly important where different dislocation core stoichiometries are considered. Calculations of images for the Ga-rich dislocation core model of Northrup [47,48] with no N atoms at the core (6:0 model) show the range of thicknesses within which the distinction between a stoichiometric and a Ga-rich core is clearly possible. This focal-series reconstruction technique has been applied to Ga-rich MBE-grown samples in [0 0 0 1] projection where atomic columns are separated by 1.84 Å. No change in intensity between atomic columns is observed for the Ga-rich or N-rich samples (Figure 4.14), which suggests that dislocation cores in MBE and OMVPE samples grown on HVPE templates are stoichiometric and also have full cores.
4.1 A Primer on Extended Defects
Figure 4.13 Cross-sectional images showing the screw dislocations and vertically stacked voids in HVPE GaN: (a) dislocation in contrast for g ¼ (0 0 0 2) and (b) out of contrast for g ¼ ð1 1 2 0Þ. The inset in (b) shows the pyramidal-shaped voids with their tips slightly shifted from the dislocation line (pyramidal defects caused by Mg doping in GaN and AlGaN and formed in InGaN are discussed in Sections 4.2.2 and 4.2.3). (c and d) Screw
dislocations in the MBE-grown GaN on top of the HVPE GaN template, under Ga-rich conditions, without the voids. The dark spot represents a Ga droplet imbedded in the layer. (e and f) Area near the interface between HVPE template and the OMVPE overlayer. Note voids along screw dislocation in the HVPE material and lack of them in the OMVPE material. Courtesy of Z. Liliental-Weber.
Figure 4.14 (a) Full core of screw dislocation in Ga-rich MBEgrown GaN and (b) in the N-rich GaN. Lack of difference in atomic column intensity between center image and surrounding matrix indicates stoichiometric cores. A model of an open-core screw dislocation by Blumenau et al. [29] and Elsner et al. [41] is shown in Figure 4.9b. Courtesy of Z. Liliental-Weber.
j 835
836
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.15 (a) Image reconstruction of a screw dislocation in HVPE sample. (b) Intensity profiles show reduced (A) and enhanced (B) image intensity, indicating Ga excess and deficiency. Courtesy of Liliental-Weber.
Turning attention to the HVPE buffer layers with specific interest in the cores, focalseries reconstructions similar to the ones discussed above for the epitaxial layer were obtained for [0 0 0 1] projections of screw dislocations [49] accompanied by voids as shown inFigure4.15a.In thedislocation corearea, oneatomic column,which is circled in the upper left corner, appears to be very weakas depicted in Figure 4.15a, and another column, which is circled in the center, appears very bright, as shown in Figure 4.15a. These columns are not juxtaposed to one another, but separated by 8 Å, a distance comparable to that between a dislocation line and the tip of the pyramidal voids observed in the cross-sectional samples. The pyramidal defects caused by Mg doing in GaN and AlGaN and formed in InGaN are discussed in Sections 4.2.2 and 4.2.3. The intensities of the highest intensity and lowest intensity atomic columns lay more than three standard deviations from the mean atomic column peak intensity in the matrix, as indicated in the line scan of the intensity of Figure 4.15b. Therefore, the intensity difference between the highest intensity and lowest intensity atomic columns is about six standard deviations. The observed change in intensity between the highest intensity and lowest intensity atomic columns cannot be obtained simply by sample tilt. Therefore, this difference in their intensities can be assigned only to the stoichiometry of the particular columns [49]. 4.1.1.2.3 Mixed Dislocations Another main type of dislocations in GaN is the mixed dislocation with Burgers vector b ¼ ½1 1 2 3. On the experimental side, very little is available on mixed dislocations owing in part to the lack of spatial resolution in standard electron microscopes, small signal levels, and the local plane bending that makes mixed dislocations difficult to image. From the theoretical point of view, the combination of edge and screw burgers vectors are very demanding in terms of computational time due to the large distortions in the core structure, and many largescale computational systems cannot accommodate them. However, with the advancement of super high-resolution scanning transmission electron microscopes (STEMs) available in a few laboratories, with spatial resolution of approximately 1 Å, this particular dislocation has been imaged with atomic resolution by
4.1 A Primer on Extended Defects
Arslan et al. [51]. Studying an n-type OMVPE-grown GaN, and taking some 50 superSTEM images of mixed dislocations, the core structure of the mixed dislocation was determined to be an eight-atom ring core, as in the case of edge dislocations [52]. A raw Z-contrast image (without image processing) is shown in Figure 4.16a, which is the atomic core structure of a mixed dislocation. The incoherent nature of Z-contrast images makes it possible to obtain the atomic structure directly from the raw image without the need of image simulations as reported by Browning et al. [53] Owing to the resolution of superSTEM, the mixed dislocation can be distinguished from the edge dislocations due to the strain present in the c-direction in mixed dislocations. Figure 4.16b illustrates a ball-and-stick model derived from Figure 4.16a, which helps clarify the atomic structure. As a matter of fact, the core structure of the mixed dislocation has been anticipated to be similar to that for an edge dislocation because the mixed dislocation is a linear combination of the screw and edge components. This means that the edge component should be an eight-atom ring core, but it has not been verified experimentally until the work of Arslan et al. [51]. The large strain as evidenced by a large Burgers vector associated with mixed dislocations causes them not to be favored. Consequently, it is plausible that mixed dislocations would structurally reorder to reduce their energy. Experimental evidence for this reordering or splitting is available in the literature [51]. Displayed in Figure 4.17a is a mixed dislocation that splits into two partial dislocations with a stacking fault structure in between. This referenced stacking fault lies along the ½1 1 0 0 direction and in the ð1 1 2 0Þ a-plane. The reduction in strain due to dissociation reduced the net strain energy in the dissociated core structures and thus has paved the way to image individual atomic columns. The structure of the core appears to be a seven-atom ring core (screw dislocation modified by stacking fault) screw dislocation on the left side and a nine-atom ring core (edge dislocation modified by stacking fault) edge dislocation on the right side with a stacking fault between the two [54]. The screw component induces a distortion in the z-direction, which decreases as the stacking fault widens to accommodate the strain due to the edge component with the insertion of an extra plane, as seen in the artistic stickand-ball diagram of Figure 4.17b. The atomic structure of the dissociated partial dislocation and the associated stacking fault is as follows: starting from the left, the structure commences with a seven-atom ring corresponding to a screw partial dislocation, followed by four- and eight-atom rings and terminating on the right side with a nine-atom edge ring structure. The lines illustrate how the four- and eight-rings get wider to allow an extra atomic plane to terminate the stacking fault at the nine-atom ring. The assignment of these features as edge, screw, and stacking fault can be confirmed with electron energy-loss spectroscopy (EELS) measurements, as discussed in Section 4.1.6. Note that there have been other types of stacking faults observed in GaN to lie on the a-plane, among which are with fRg ¼ 1=2h1 0 1 1i and fRg ¼ 1=6h2 0 2 3i displacement vectors [55]. A detailed discussion of stacking faults is given in Section 4.1.4. However, a point that must be made is that the geometries of these stacking faults that are terminated by partial dislocations are unique.
j 837
838
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.16 (a) Aberration-corrected Z-contrast image of a fullcore mixed dislocation. The box serves to outline the core of the dislocation; (b) ball-and-stick model of the eight-atom ring atomic structure [51].
4.1 A Primer on Extended Defects
Figure 4.17 (a) An aberration-corrected Zcontrast image of a dissociated mixed (partial) dislocation, showing a stacking fault in between the edge and screw dislocation cores of approximately 3 nm length; (b) The ball-andstick model associated with the defect serves to
clarify the atomic structure and illustrates how the stacking fault widens in the plane of the image between the two dislocations to accommodate the strains associated with each dislocation [51].
A consequence of two-dimensional imaging, the Burgers vector of the partial dislocation could be determined only in the direction of the basal plane. The available data are not sufficient to unequivocally determine the slip plane on which the stacking faults form. To do so requires extensive calculation and strain experiments. However, standard models can be used to gain a first-order understanding in hexagonal close packed wurtzitic structures from which the GaN wurtzitic structure is derived. Applying these methods, one suggests that stacking fault generated from the dissociation of a mixed dislocation, ha þ ci ! hai þ hci þ SF, is specifically of the form 1=3½1 1 2 3 ! 1=3½1 1 2 0 þ ½0 0 0 1 þ SFð1100Þ , which is the only of the four different possibilities capable of dissociating into a pure edge dislocation and a pure
j 839
840
j 4 Extended and Point Defects, Doping, and Magnetism screw dislocation. It should be reminded that stacking fault increases the total energy of the configuration, which may be made possible by the segregation of point defects and impurities to the dislocation. The electronic properties of extended defects are treated in Section 4.1.6. 4.1.2 Nanopipes or Hollow Pipes
The nomenclature that describes voids is a term borrowed from SiC where it represents holes of the diameter as large as 10 mm with a small modification, nano instead of micro. In GaN, however, they are much smaller and less frequent and have been observed in both unintentionally doped and donor-doped GaN, albeit their genesis has not been identified unequivocally. In the technical sense, they are empty or filled holes that exhibit a dislocation character. Observations show that void-type defects extending along the c-axis exist in GaN, termed as nanotubes and pinholes [56,57]. In the technical sense, they are empty or filled holes with a dislocation character and generally they have a well-defined standard shape. Moreover, they can extend up to a few tens of nanometers and are usually limited by f1 0 1 0g planes and have been reported to start from V-shaped (pyramidal defects discussed in Section 4.2.2) indentations [56]. Further, the nanopipe density increases with higher concentrations of impurities or dopants, for example, O, Mg, In, and Si. The hollow pipes are often found to be attached to dislocations, and they can be classified into two classes: those with an edge component and those with pure screw dislocations [25,58]. It should also be pointed out that some experimental investigations point to screw dislocations having open cores [59], and others to full cores [47]. Although any role of dislocations in forming hollow defects is controversial, a model is presented here for completeness. Frank [58] proposed that a dislocation with a hollow core can lower the total energy by relieving strain energy at the expense of free surface energy. The equilibrium radius r of the open core is given by r¼
mb2 ; 8p2 g
ð4:1Þ
where m is the shear modulus and g is the surface energy. The term b represents the Burgers vector. Calculations for the hollow pipes in GaN indicate that the equilibrium radius is comparable to the atomic dimensions but is much smaller than the radius observed in experiments. However, there is a consensus that the formation of open-core dislocations is related to impurity atoms as elaborated further in the text. The density of these defects was estimated to be in the range of 105–107 cm2 and their radii in the range 3–1500 nm. Liliental-Weber et al. [56,57] suggested that these two types of defects may have their origin in impurities present in the material, supporting the theoretical work by Elsner et al. [46] who argued that point defects, complexes, carriers, and impurities trapped at the dislocation due to large local strain may be responsible for its electrical activity. For example, O and O-related defect complexes can be formed on the walls of open-core screw dislocations and presumably the
4.1 A Primer on Extended Defects
same arguments can be extended to nanopipes, in GaN. (See Section 4.1.6 for a detailed discussion regarding the electrical activity associated with dislocations and Section 4.3 for point defect complexes.) Furthermore, Liliental-Weber et al. [56,57] also observed that the nanopipes close and open again along their axis as they propagate up through the template much like bamboo canes. As shown in Figure 4.18, as large as 2a edge component may be exhibited in such defects. The nanopipes, which have an edge component, are confined inside the first 200 nm of the epitaxial layer [60]. Furthermore, they contain amorphous material in layers grown on 6H-SiC, whereas they are empty on top of sapphire [61]. It has also been shown that the pure screw nanopipes traverse the entire epitaxial layer [7]. The nanopipes may either keep the same section or close and open up a few times on their way to the surface. The nature of these defects in HVPE templates and MBE overlayers is discussed in detail in Section 4.1.6. The defects stretching along the length of the c-axis with V-shaped termination cannot be opencore screw dislocations, as discussed in the previous section. There is quite a bit of dispersion in terms of the configuration of screw dislocation in that full-core, filled-core, and open-core configurations have been reported and discussed [29,56,62]. The open-core variety has been predicted by theory [41,48]. The open-core dislocation with diameter 7.2 Å appears to have the lowest energy configuration. Other reports indicate that the open-core feature is generally associated with screw dislocations in undoped or n-doped GaN [63]. On the contrary, it has been experimentally [42,52] and theoretically [41] shown that the edge and mixed dislocations are mostly of a full, eight-atom closed-core variety. The theory also indicates this configuration to have the minimum energy. Even then, the four-atom structure and 5/7-atom structure have been observed [64]. However, some samples have been shown to have hollow core edge dislocations [65]. Cherns et al. [66] observed that open core attached to edge (mixed)-dislocation in p-type Al0.03Ga0.97N. They suggested that this might be helpful in understanding the origin of open-core pipe formation in undoped and donor-doped GaN in that the Al0.03Ga0.97N layer investigated was heavily doped with Mg (1020 cm3) grown by OMVPE. Figure 4.19a shows a cross-sectional image of open-core edge and mixed dislocations, with g ¼ (0 0 0 2) reported by Cherns et al. [66]. It was shown the diameter of core!is typically 1–5 nm. The dislocation marked with A is out of contrast with ! g b ¼ 0 and thus is a pure edge dislocation, whereas that marked with B is identified as a mixed dislocation. Due to high Mg concentration in this p-type layer, many pyramidal defects form and appear in the cross-sectional image. A somewhat detailed discussion of pyramidal defects caused by excessive Mg can be found in Section 4.2.2. A close inspection of the data reveals that the density of pyramidal defects decreases significantly within the 50 nm region around the dislocation. However, the dislocations are decorated with pyramidal defect chains. It should be noted that the pyramidal defects on the dislocation have a much smaller size than those in the surrounding region, indicating that the defects on the dislocation are not a result of random distribution of pyramidal defects in the layer. A high-resolution top-view lattice image (Figure 4.19b) on the defect-decorated dislocation reveals that it has a ½1 1 2 0 Burgers vector and an open core. The lack of sharp contrast boundary of
j 841
842
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.18 (a) Nanotubes observed in oxygen-rich or Mg-rich GaN. These nanotubes are extended along the c-axis compared to pinholes; (b) nanotubes stacked on top of each other as pinholes; and (c) termination of a nanotube in a V-shaped surface defect that might be formed by due to O or In present. Courtesy of Z. Liliental-Weber.
4.1 A Primer on Extended Defects
Figure 4.19 Cross-sectional bright-field two-beam image with g ¼ 0 0 0 2, showing pyramid defects on an edge dislocation A and a mixed dislocation B (a); A [0 0 0 1] lattice image of a dislocation with a hollow core. A Burgers circuit shows that this dislocation has an edge component of Burgers vectors (b) [66].
this open core suggests that it has a varying diameter along the [0 0 0 1] direction, which might result from the deposition of impurity atoms on the sidewall. The diameter of these open cores in the sample investigated was found to be 1–3 nm for the edge dislocation and 3–5 nm for the mixed dislocation. With evidence that Mg segregation induced pyramidal defects cannot be found on the screw dislocations, the authors suggested that Mg is attracted into a Cottrel cloud by the dilatational part of the strain field due to the edge component of the Burgers vector. It is likely that Mg,
j 843
844
j 4 Extended and Point Defects, Doping, and Magnetism with a larger atomic size as compared to Ga, segregates to a region under tensile strain. As discussed in Sections 3.5.6 and 4.2.2, segregation of large concentrations of Mg induces a polarity change from Ga-polarity to N-polarity. Because N-polarity GaN grows at a much slower rate than the Ga-polarity GaN, a void will be generated in the N-polarity region. The void may be trapped and closed by the overgrowth of surrounding GaN. However, if there is continual Mg-segregation along the dislocation, a chain of voids or even a continuous hollow pipe could be generated. It should be mentioned that other views, albeit associated with other samples, have been put forth [56,57]. 4.1.3 Planar Defects: Domain Boundaries
Literature is somewhat confusing in terms of definition and also the nomenclature ascribed to planar defects. Even though terms such as double positioning boundaries, stacking mismatch boundaries, translation domain boundaries, and inversion domain boundaries, not in any particular order, are seen in the literature, a picture is emerging. IBD - and Holt-type IBD are well-recognized f1 0 1 0g inversion domain boundaries. Likewise, Drum et al., and Blank (or Amelinckx) models deal with f1 1 2 0g stacking faults that are also called TDBs and DPBs. IDBs in particular are characterized by an inversion operation and should be distinguished from TDBs. Rouviere et al. [40] pointed out that TDBs are considered planar defects where the two crystals on either side of the boundary are related by an operator WTDB that exclusively contains a translation. IDBs are often observed in GaN prepared by various growth methods [68,69]. From a topological point of view, the term TDB is appropriate because it can be clearly contrasted with the IBDs. Many forms of boundaries that have been the topic of theoretical calculations and experimental investigations. The structural and electrical properties of at least some of these boundaries are discussed in Section 4.1.6, which deals with electronic structure of extended defects. Inversion domains come about because of the lack of crystal centrosymmetry in wurtzitic GaN. If one considers the hcp sublattice of anions, its space group (P63/mmc) is centrosymmetric as the 63-axis is normal to the (0 0 0 1) mirror as shown in Figure 2.2. The structure has b1 and b2 tetrahedral sites that are related by the (0 0 0 1) mirror [25]. Moreover, this structure contains two groups of tetrahedral sites that are related by the (0 0 0 1) mirror. The wurtzitic structure is achieved by adding a cation to one tetrahedral site group of the anion network. Doing so removes the mirror from the space group, thus the designation P63/mc. The loss of the mirror normal to the 63-axis paves the way for two crystallographic variants that are related by the broken operator. Consequently, a mirror situated on the (0 0 0 1) anion plane can be used to create an inversion that changes the polarity of the crystal, described by cation–anion bond parallel to the c-axis, the single bond [25]. This is to say that if the bond is directed from the Ga atomic site to the N site, the polarity is considered to be Ga polarity. Presence of inversion-type planar defects leads to the existence of both Ga- and n-polarity regions. Description of an IBD requires a translation R in addition
4.1 A Primer on Extended Defects
to the mirror, m, with the final form of WIBD ¼ m þ R. The operators for Holt and IBD (V-type) inversion domain boundaries are WHolt ¼ m 1/8[0 0 0 1] þ 1/2[0 0 0 1], and WIBD : WV ¼ m 1/8[0 0 0 1], respectively. Several additional models have been proposed, an extensive discussion of which can be found in Refs [25,38]. Substrates other than the GaN templates and ZnO used for wurtzite GaN epitaxy have a different staking order than GaN. As a result of this and nonideal pairing, inversion, and rotation domain boundaries form at each step unless each terrace has the same bonding configuration. Having stated this, it is also possible to have line defect character at the steps even if the two terraces have distinct bonding configurations [70]. IDBs are usually formed near the interface of GaN/substrate at the steps and thread straight to the film surface. They are commonly along the f1 0 1 0g prismatic planes and remain parallel to the [0 0 0 1] direction. They facilitate the coexistence of crystalline regions of inverse polarity, with a resultant influence on physical and chemical properties such as surface morphology, chemical etching behavior, and crystal growth behavior [15,71–73]. They can be electrically active or inactive depending on the atomic coordination along the boundary, which is the topic of Section 4.1.6. Formation of inversion domain boundaries is affected by the surface step structure of the substrate, as pointed out in Sverdlov et al. [19] and examined in detail by Ruterana et al. [25]. In short, unless the stacking order as well as the step heights is such that the surface of each step is consistent with one another, these boundaries would form. For example, on 6H-SiC, these steps need to be 6 bilayer high for this condition to be satisfied, which is possible but requires extreme surface preparation techniques that are nonstandard. Consequently, it is highly desirable to deposit GaN on substrates that are monomorphic, meaning with stacking order common with nitrides. Among the substrates, ZnO is one such substrate with the added advantage that the lattice mismatch between GaN and ZnO is e ¼ 0.017, which leads to a critical thickness between 80 and 120 Å. This implies that coherently strained layers of GaN could be grown with thicknesses up to about 100 Å. Some compositions of the InGaAlN alloy, inclusive of the InGaN alloy, can be made to lattice-match ZnO. Although the opinion is divided, GaN grown on the O face of (0 0 0 1) ZnO exhibits emission characteristics similar to the best ones on (0 0 0 1) sapphire. More work is required to harness the possibilities offered, including easy cleavage planes that align with GaN. For details refer to Section 3.3.4. As mentioned previously, the f1 0 1 0g inversion domain boundaries exhibit two atomic configurations, namely, Holt and IBD , the latter is also referred to by some as V-type IBDs, depending on the growth conditions. It should be mentioned that IBD nomenclature is more commonly accepted by the community and avoids the potential confusion associated with V-type defects formed in Mg-doped GaN (see Section 4.2.2 dealing with pyramidal defects). The samples that contain Holt inversion domains have a flat surface morphology, whereas the IBD (or V-type IDBs) are observed in the center of small pyramids that are nominally 100 nm high, protruding from the sample surface. As compared to the V-type inversion domains, the Holt inversion domains are always smaller, <20 nm, and have relatively high densities (2.5 1010 cm 2). The IBD (or V) variety inversion domains reach 50 nm
j 845
846
j 4 Extended and Point Defects, Doping, and Magnetism in size with a density that is an order of magnitude lower compared to the Holt variety. Generally, the two atomic configurations (IDB s and Holt IDBs) do coexist due to interactions with the basal SFs [74–76]. The inversion domains are generated mostly at surface steps where they minimize the large misfit along the c-axis by about 20%. The IDB -type inversion domain boundary or the V-type inversion domain boundary [68] has four- and eight-atom rings in a ½1 1 2 0 projection. It was suggested that the IDB configuration is stable in ionic materials, because there are no GaGa and/or NN wrong bonds at the inversion domain boundary [77]. The IDB s are formed by shifting one side of an IDB by about c/8 along the [0 0 0 1] direction (WIBD : m 1/8[0 0 0 1]). As can be seen in Figure 4.20a associated with the ð10 10Þ projection, in this case all atoms along the boundary are fourfold coordinated and form GaN bonds across the boundary. In the second configuration, known as Holt-type IDB, the atoms on either side of the boundary are located at interchanged positions in the crystalline structure by shifting a relative displacement of 3c/8 along [0 0 0 1] direction (WHolt : m 1/8 [0 0 0 1]þ1/2[0 0 0 1]). In this fault, referring to Figure 1.2, dealing with sublattice arrangements of b1 and b2 tetrahedrons, two sublattices are exchanged in that atoms A at positions g become atoms A at positions b, and atoms B at positions b become atoms B at positions g. This model has the wrong GaGa and NN bonds at the boundary and have been shown to be electrically active. The other IBD, which we refer to as IBD - or V-type, is characterized by a c/2 translation (WIBD :WV m 1/8 [0 0 0 1]), as compared to the Holt type, to avoid the wrong GaGa and NN bonds. The Holt-type IBDs have higher formation energy than the IDB (V)-type boundaries [77]. Figure 4.20a and b shows the respective stick-and-ball representations for IDB type and Holt-type IDB [74], with the rigid translations p being defined according to the Austerman–Gehman model [78]. Figure 4.21 is a schematic representation of an ideal Holt-type inversion domain boundary formed in growth along the [0 0 0 1] direction in the context of polarization. On the left of the boundary, the growth initiates with N leading to Ga-polarity, and on the right it begins with Ga leading to N-polarity. In N polarity and under tensile strain, the piezoelectric field generated points toward the surface, whereas that for the Ga polarity region points in from the surface. When the strain is compressive, the direction of the field changes. Consequently, the inversion domain boundary in conjunction with the polarization field would introduce an additional scattering mechanism, degrading the materials quality and performance. Refer to Volume 2, Chapter 3 for a discussion of scattering mechanisms. Figure 4.22a illustrates hypothetical epitaxy on a-plane sapphire, in case of which inversion domains can coexist, interrelated by the suppressed (0 0 0 1) mirror that is normal to the substrate surface. Consequently, the epitaxial interfaces on either side of the IDB are crystallographically equivalent and hence energetically degenerate. On the contrary, Figure 4.22b illustrates conventional epitaxy on c-plane sapphire. In this case there is no operation as in (a) for us to interrelate the inverse domains. Hence, if an IDB emanates from the interface, for example, for reducing the misfit along the growth direction as discussed by Ruterana et al. [25] the epitaxial structures on either
4.1 A Primer on Extended Defects IBD * (in [1 0 1 0] plane) l
m
(a)
[0 0 0 1]
P
[1 0 1 0] [1 2 1 0]
c
[0 0 0 1] [1 0 1 0] [1 2 1 0] l
Holt IBD (in [1 0 1 0] plane)
(b) m
P
Figure 4.20 Schematic illustration of f1 0 1 0g inversion domain boundary models. (a) Projection of a f1 0 1 0g IDB (V-type) planar defect in the ð1 1 2 0Þ plane (or the ½1120 projection) with a relative displacement p ¼ c/ 8 of the neighboring domains (WIBD ¼ m 1/8 [0 0 0 1]). (b) Projection of a f1010g Holt-type IDB with a relative displacement p ¼ 3c/8 (WHolt ¼ m 1/8[0 0 0 1] þ 1/2[0 0 0 1]) in the GaN structure (h1 1 2 0i projection; large and
small circles denote distinct atomic species, shading denotes levels 0 and a/2 along the projection direction, and tetrahedral indicate polarity reversal). The p vectors denote displacement of domain m with respect to domain h. The tetrahedra indicate polarity reversal and the rigid translations p are defined according the Austerman–Gehman model. Courtesy in part of P. Ruterana.
j 847
848
j 4 Extended and Point Defects, Doping, and Magnetism Or for (b) IBD *
l
m
[0 0 0 1] [1 0 1 0]
p
[1 2 1 0]
Figure 4.20 (Continued )
side of the IDB are now distinct and hence energetically nondegenerate. In Figure 4.22b, a ½0110 IBD is shown originating from a step at the interface where the IBD depicts the coexistence of nondegenerate interfaces. For simplicity, the misfit between the two crystals is not taken into account. In GaN the large and small circles
E field direction Compressive
[1 0 1 0]
[0 0 0 1]
[0 0 0 1]
Inversion domain boundary Ga polarity
N polarity
E field direction Tensile Compressive
Tensile
Substrate Figure 4.21 Schematic representation of piezoelectric polarization induced field for both tensile and compressive strain on either side of a simple Holt-type inversion domain boundary with wrong bonds intersecting the dashed line. Atom positions are projected onto the ð1 1 2 0Þ plane. The polarity inversion is noted by the dashed line across the boundary. The lightly shaded rods show the wrong bonds between like atoms.
N Ga
4.1 A Primer on Extended Defects
Figure 4.22 (a) Schematic illustration of an a-GaN/sapphire bicrystal caused by a step at the interface in the case of growth on a-plane sapphire. The GaN layer has the epitaxial orientation relationship ð1 0 1 0Þa-GaN// ð11 2 0Þsapph, and [0 0 0 1]a-GaN//½1 1 0 1sapph. The projection direction is ½1 2 1 0a-GaN. (b)
Schematic illustration of an atomic model of an a-GaN/sapphire bicrystal (depicting growth on c-plane sapphire) having the epitaxial orientation relationship (0 0 0 1)a-GaN// (0 0 0 1)sapph, ½2 11 0a-GaN//½1 0 1 0sapph. This case represents epitaxial growth on c-plane sapphire. Courtesy of Dimitrakopulos et al. [38].
indicate distinct atomic species. Open and filled circles denote atoms at height 0 and aa-GaN/2, respectively. In sapphire, crosshatched, and blank circles represent atoms of different species; different shadings represent different levels along the projection direction. Figure 4.23 is an HREM image of an IDB . It was suggested that an IDB nucleates at a step on sapphire substrate, because steps could reduce the energy barrier for formation of an inversion domain [68]. However, it has also been reported that the occurrence of IDBs depends strongly on the growth conditions and Si
j 849
850
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.23 HREM image of an inversion domain and simulation with its corresponding Ga (large circles) and N (small circles) atom overlay in the inset.
doping [74,79,80]. This means that the occurrence of IDBs will be dependent on both a kinetically controlled and a thermodynamically controlled process. As for the formation mechanism of Holt-type IDBs, high-resolution electron microscopy observations showed that the inversion domain boundaries are transformed from the low-energy and electrically inactive IDB -type structure to the high-energy, electronically active Holt-type structure [74]. These transitions are due to the interaction of two distinct planar defects, and can be attributed to the different growth rates of adjacent domains of inverse polarity. A detailed discussion of inversion domain boundaries and their configuration and symmetry operators can be found in Ref. [25]. In epitaxial GaN layers, the IDBs are observed on sapphire substrates but not on SiC when grown under the same conditions on both substrates [61] followed by a detailed study indicating that the layers on SiC are unipolar [21]. This is provided the SiC substrate surface is free of damage and preferably exhibits atomically smooth
4.1 A Primer on Extended Defects
surfaces. In one case IDBs in GaN grown on SiC substrate have been reported when a thin amorphous layer was present at the epitaxial layer substrate interface [81]. The IDBs in GaN layers on sapphire substrates are common to all growth methods, MBE, OMVPE, and HVPE [77]. They can be either bounded by ð1 0 1 0Þ planes and cross the entire epitaxial layer [82] or limited by f1 0 1 0g and f1 0 1 2g planes, and can resemble house-like domains buried near the interface with the substrate [83]. As for the atomic structure of IBDs, Cherns et al. [84] studied displacement fringes and confirmed the model first proposed by Northrup et al. [77], with a c/2 translation, although a translation on the basal plane has also been reported [83], which is not supported by other investigations. In an HREM work on GaN layers grown by MBE, Potin et al. [85] reported the occurrence of Holt-type IDBs. 4.1.4 Stacking Faults
[0 0 0 1]
[1 1 1]
Nitrides with no stacking fault are of 2H type inthat metal, and nitrogen layers alternate along the c-direction, such as AaBb; thus, the term 2H (H standing for hexagonal). Stacking faults are irregularities in the stacking (ordering) of layers of atoms that do not necessitate the breaking of bonds, such as transforming from the hexagonal stacking order to AaBbCc stacking order characterizing cubic (zinc blende) systems, as shown in Figure 4.24. As such, the energies associated with stacking faults are very small, which means that they would be easy to form[86]. The 2Hsystem is the simplest system in which other polytypes, which come about from stacking faults, can be described by
3ML (ZB)
4 ML (WZ)
Figure 4.24 Stacking fault associated with stacking order AaBbAaBbAabCc (between two broken lines) right through wurtzitic and zinc blende stacking orders, causing a stacking fault defect at the dotted line. This particular stacking fault contains 4 ML of the wurtzitic material, and 3 ML of zinc blende material. Patterned after Ref. [88].
j 851
852
j 4 Extended and Point Defects, Doping, and Magnetism the introduction of periodic stacking faults [87]. In other words, instead of maintaining the AaBb stacking along the c-direction throughout the crystal, stacking order such as AaBbAaBbCcBbCcBbAaBb (also called type I and I1) AaBbAaBbCcAaCcAaCcAaBb (also called type II and I2) AaBbAaCcAaBbAaBbAaBb, and AaBbAaBbCcAaBbAaBbAaBb (also called extrinsic) do form. Stampfl and Van de Walle [88] considered the energetics of these four types of stacking faults in GaN, AlN, and InN. Starting with the lowest formation energy, the order is type I, type III, type II, and extrinsic, which means that typeIis the easiest to formand extrinsic is the least probable. The orderremains the same regardless of the binary considered. It should also be mentioned that the aforementioned type III stacking fault has the character of I1 stacking fault related to two superimposed mirror and has not been observed experimentally. It is therefore questionable whether this should be considered a basic stacking fault. Let us now discuss the structural details of stacking faults. The tetrahedral notation is used to represent the structure of the three types of stacking faults encountered in nitride semiconductors as shown in Figure 4.25. Intrinsic SF denoted I1 or type I, I01 : The stacking sequence T1T03 T1T2T01 T2 is 0 0 0 referred to as the I1 stacking fault. Its variant, I 01 : T03 T1T _3T _ 2 T3T2 , which is related to I1 _ _ by a rotation of 180 around the c-axis (the faulted sequence is underlined), is referred to as the I01 intrinsic stacking fault. Intrinsic SF denoted I2 or type II: This results when a 1/3h1 1 2 0i dislocation dissociates into two Shockley partials. Its sequence is T1T03 T1T2T3T02 T3 and for the 0 0 0 twinned variant I0 2 ; it is T03 T1T _ 3_ T _ 2_ T3T _ 2_ . Extrinsic SF denoted E: This is obtained by inserting a complete sequence of the sphalerite (zinc blende) structure such as T1T2T3 or T0 1 T02 T03 in T1T03 ^ T1T03 , leading to 0 0 0 0 0 T1T03 T1T2T3T1T03 or T1T _3T _ T _2T _ T1T _ 3 , which is the twinned variant. This defect is _ 1 _ _ 3 _ _ bounded by a Frank partial dislocation (b ¼ 1/2[0 0 0 1]). Refer to Section 1.1 for the nomenclature used to depict thestacking configuration in semiconductors of interesthere. Early investigations of stacking faults go back to ceramic grade AlN and ZnS, a II–VI material. In ceramic AlN, Drum [23] investigated faults that intersected on the basal and prismatic f1 2 1 0g planes. The displacement vector was shown to be 1/2h1 0 1 1i and faults folded to the basal planes by leaving a 1/6h0 1 1 0i stair rod dislocation at the intersection. In a parallel effort, Blank et al. [22] investigated the planar defects that folded from basal to prismatic f1 2 1 0g planes in wurtzite ZnS, to interpret them as stacking faults, whereas other authors considered them to be thin lamella of the sphalerite phase in CdS [24]. These prismatic faults were then shown to be growth domains and the displacement vector was found to be 1/6h2 0 2 3i that is the same as that of the I1 basal-stacking fault in the hcp structure. These extended faults have been identified in basal and prismatic planes of GaN by Vermaut et al. [20]. In this particular investigation, the fault vectors were found to be R11 ¼ 1/6 ½0 2 2 3, R12 ¼ 1/3 ½0 11 0, RE ¼ 1/2[0 0 0 1] (basal-stacking faults), RA ¼ 1/6½0 2 2 3 and RD ¼ 1/2 ½0 1 1 1, which are in agreement with the models proposed by Drum [23] and Blank et al. [22]. Let us now investigate the three types of stacking faults defined above. On the experimental side, stacking faults are bountiful at the interfacial region between the substrate, low temperature initiation layer, or the buffer layer in the case of MBE, and
4.1 A Primer on Extended Defects
T1
2
3
T3
T'3
T2
T'2
T1
T'1
T3
T3
T2
T2
T2
T1
T1
T1
T'3
T'3
T'3
T1
T1
T1
1
I1 R = 1/6 < 2 2 0 3>
2
3
1
I2 R = 1/3 < 1 1 0 0>
1
2
3
1
E R = 1/2 < 0 0 0 1>
Figure 4.25 Atomic models of the basal-stacking faults projected along ½1 1 2 0. The symbols (s) represent nitrogen, (*) gallium, (—) in-plane, and (¼) out-of-plane bonds. Courtesy of Ruterana et al. [25].
initial portions of the GaN layer grown on top the buffer layer. Somewhat away from the interface, about 20 nm, the defects are in the form of long extended stacking faults that give way to short width faults with an extension less than few tens of f1 0 1 0g lattice spacing. The long extended stacking faults are mainly I1 and I2 faults, even though segments of E fault having a 1/2[0 0 0 1] rigid-body translation are also observed [90]. Early investigations [91] led to the observation of I1 stacking faults in GaN layers grown on sapphire substrates by OMVPE but not I2 and E varieties. Investigations of Ruterana et al. [25], however, led to the conclusion that both I1 and I2 faults are present in the immediate vicinity of the interface with the substrate. Basal plane stacking faults are often observed in epitaxial a-GaN, which occur during growth when atoms occupy either the Bb or the Cc positions above an Aa layer [38]. Experimental observations indicate that most basal-stacking faults are of intrinsic character [90]. Two of the intrinsic staking faults, namely I1 and I2, have been identified in GaN in much the same way as has been done for the hcp structure [87]. In addition to type I1 and I2, type E stacking faults have been observed in thick GaN
j 853
854
j 4 Extended and Point Defects, Doping, and Magnetism both on sapphire and SiC substrates, irrespective of the growth method [92]. Moreover, I1 stacking faults bounded by Frank–Shockley partials have also been observed [93]. The I1 staking faults result from one violation of the stacking rule or from the introduction of one row of zinc blende stacking in the wurtzite-type stacking sequence along the [0 0 0 1] direction. Crystallographically, this fault results from the removal of one basal layer (e.g., Aa) followed by a 1/3h0 1 1 0i shear resulting in AaBbAaBbCcBbCc. . . stacking. The corresponding displacement vector is p ¼ 1/6h0 2 2 3i. The I1 stacking fault cannot be formed by dissociation of crystal dislocations with b ¼ ai (here ai represents the three lattice vectors on the basal plane of a hexagonal system, with i ¼ 1, 2, 3) and, therefore, is formed during the growth [87]. The I2 stacking fault comprises two rows of zinc blende stacking introduced by a 1/3h0 1 1 0i shear, and can be formed by dissociation of b ¼ ai dislocations [38]. Both intrinsic stacking faults are low-energy defects, and they do not affect the nearest neighbor (NN) packing, with I1 being more energetically favorable [94]. Some predictions indicate that these staking faults do not introduce gap states [94]. In terms of optical properties, if the stacking faults induce zinc blende, a material that has a smaller bandgap could lead to quantum well like localization [94]. The resulting band edge transitions such as excitons would be at a lower energy, expected of a wurtzitic material [95]. For illustrative purposes, a schematic representation of an I1 stacking fault is shown in Figure 4.26a. The HREM image of such a fault, which is terminated by partial dislocation, is shown in Figure 4.26b. A closed right-handed circuit, SGHIJKLMS, drawn around the line direction of this partial shows the Burgers vector. The circuit maps to SGHIJKLMF in the reference space (Figure 4.26b and c), and closure failure stacking fault arises [38]. An example of an I2 fault is shown in Figure 4.27 in a high-magnification image taken along the h1 1 2 0i zone axis. These I2 faults may come about from the dissociation of a perfect a ¼ 1/3h1 1 2 0i dislocation into two Shockley partials. But, the Burgers circuit drawn around these faults did not exhibit any closure failure. The two partial dislocations bounding these basal faults have opposite Burgers vectors b ¼ 1/3h0 1 1 0i. Regarding the underlying cause for these defects, the I2 fault could form directly by shear or following a dissociation of a pure edge dislocation into two Shockley partials. The fact is that extended faults of the type are observed in the first few tens of a nanometers in the GaN layer from the sapphire interface point to the nature of the buffer layer. Because the zinc blende phase of GaN is present in the low temperature ~
Figure 4.26 (a) Schematic illustration of an I1 stacking fault in a-GaN (½2 11 0 projection; large and small circles indicate distinct atomic species; open and filled circles denote atoms at height 0 and aa-GaN/2, respectively). (b) HREM micrograph of an I1 Stacking fault terminating at a partial dislocation. An image simulation of the fault is shown in an inset and stacking sequences have been indicated (defocus ¼ 59 nm, thickness ¼ 3.2 nm,
atomic columns correspond to white spots). A closed right-handed circuit has been drawn around the line of the defect using lattice translations and a closing displacement from K to S with Burgers vector b ¼ 1/6½2 0 23. (c) Mapping of the circuit to the reference space. The closure failure FS identifies the line defect as a partial dislocation with Burgers vector b ¼ 1/6½1 2 0 3. Courtesy of Dimitrakopulos et al. [38].
4.1 A Primer on Extended Defects
j 855
856
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.27 I2 stacking fault observed along the h1 1 2 0i zone axis showing no closure failure. Courtesy of Ruterana et al. [25].
initiation layer, it is plausible that these extended faults could be due to imperfect transformation to the hexagonal polytype during heating in hydrogen, underlying the importance of the temperature ramping schedule and the gas environment in hexagonal GaN during heating [96]. Another likely cause for the formation of the extended faults may be the mosaic growth mode of GaN nitridated sapphire layers. The presence of steps, due to the intentional (e.g., 3 ) or unintentional misorientation, leads to the generation of defects because they create an additional shift between the
Figure 4.28 Coexistence of I1 and I2 stacking faults, with I1 defect contributing the closure failure that is equal to c. Courtesy of Ruterana et al. [25].
4.1 A Primer on Extended Defects
epitaxial islands grown on adjacent terraces. The accommodation of the shift along the [0 0 0 1] direction is done by the insertion of stacking faults in the epitaxial layers [97]. The I1- and I2-type stacking faults can also occur in the same vicinity as shown in Figure 4.28, where two neighboring basal-stacking faults are found with a short extension. When a Burgers circuit is drawn around them, the closure failure appears to be equal to c as shown by the white arrow above the scale legend. The I2 fault shown in the left of the figure is bounded by two opposite partial dislocations (b ¼ 1/3 h1 0 1 0i) in much the same manner as those presented in Figure 4.27. This I2 defect (fault) is not due to the dissociation of a dislocation and does not contribute to the closure failure. The I1 fault shown on the right side, however, is bounded by two Frank–Shockley partials and results from a climb dissociation of a perfect c dislocation and causes the Burgers circuit closure error. Plan-view samples of GaN show the f1 1 2 0g defects, which can either terminate by partial dislocations or form closed domains, and are difficult to observe edge along the [0 0 0 1] direction. This is because they can easily fold from the prismatic to the basal plane. It should be noted that the closed domains may limit the inversion domains. These f1 1 2 0g prismatic faults defects can be observed by convergent beam diffraction patterns using the 1011 and 1 0 11 discs recorded inside the domains and in the matrix. The results are identical to those taken along the h1 0 1 2i zone axis [21]. Convergent beam electron diffraction can pave the way for structural identification, thickness, and residual stress. Moreover, the contrast inside the discs, due to dynamical diffraction, reveals the entire symmetry of the structure along the observed direction [98]. To shed some light on the atomic structure of prismatic stacking faults, one can utilize two atomic models that exist in the literature for the f1 1 2 0g prismatic fault in wurtzite materials, as originally characterized by conventional electron microscopy in the 1960s referred to as Blank [22] and Drum models [23]. For the characterization of HREM edge, a defect contained, for example, in the f1 2 1 0g plane should be observed along the ½1 0 1 0 direction. However, the projection of the 1/2½1 0 1 1 fault vector should result in a 1/2[0 0 0 1] translation vector that is not visible in a ½1 0 1 0 HREM image. In ½1 1 2 0 zone axis images, the projection of the fault vector should exhibit one 1/4½1 1 0 0 and one 1/2[0 0 0 1] component. Therefore, observations of the inclined defects should bring information on the fault vector. Unfortunately, the core structure of the defect cannot be directly observed. Direct observation and simulations can be brought to bear for a sound grounding of the atomic nature of the prismatic stacking faults, the results of which are shown in Figure 4.29. Figure 4.30 shows a schematic view of a prismatic SF emanating from a step of partial character on the (0 0 0 1) surface of the 6H-SiC substrate along the ½2 11 0 direction [38]. The stacking fault plane is ð1 2 1 0Þ, which is inclined in this projection, and has been taken here to exhibit a rigid-body translation corresponding to the Amelinckx model [22]. Large and small circles denote distinct atomic species. Open and filled circles denote atoms at heights 0 and ca-GaN/8, respectively. The term x is the line direction of the defect pointing into the page. Stacking faults are very prominent in wurtzitic GaN when it is grown on the polar surfaces of zinc blende substrates such as GaAs and diamond substrates such as Si.
j 857
858
j 4 Extended and Point Defects, Doping, and Magnetism The nature of the two crystal polytypes, namely, wurtzite (a) and zinc blende metastable cubic GaN (b), is pivotal for the discussion of stacking faults on these substrates. Although, wurtzitic GaN grown on SiC and carefully prepared sapphire do not seem to have the zinc blende polytype present, when films are grown on substrates such as (1 1 1) GaAs and (1 1 1) Si, zinc blende phase exists that goes to the heart of stacking faults. TEM studies showed that stacking faults are often observed in the {1 1 1} planes in b-GaN and basal planes {0 0 0 1} in a-GaN. The wurtzite and zinc blende polytypes differ only through the stacking order of the b-GaN {1 1 1} planes,
(1 2 1 0)
Ga, 3/8c out-of-plane N
(a) [1 1 2 0]
(b)
Figure 4.29 Models of the prismatic stacking fault, (a) and (b) the Drum model: 1/2½1 0 1 0, (c) and (d) the Blank model: 1/6½2 0 2 3. Courtesy of Ruterana et al. [25].
Ga N
4.1 A Primer on Extended Defects
[0 0 0 1]
Ga, 3/8c outof-plane N (c)
[0 0 0 1]
Ga N (d) Figure 4.29 (Continued )
with wurtzite having an AaBbAaBb. . . sequence and zinc blende an AaBbCcAaBbCc. . . sequence, see Section 1.1 for details. They have nearly identical nearest neighbor spacing. Therefore, it is not surprising that the stacking fault formation energy in GaN is very low, about 20 meV. Stacking faults are a common form of strain relief in zinc blende fcc crystal structures because their formation energy is fairly low. Figure 4.31 is a high-resolution cross-sectional XTEM micrograph of b-GaN film grown on Si (0 0 1) substrate using a thin SiC as a buffer layer, showing a relatively high density of stacking faults located on the {1 1 1} planes [99]. As shown in Figure 4.31, the stacking faults nucleate near the GaN/substrate interface and their density decreases rapidly within the film away from the interface. This rapid drop in the stacking fault density could be associated with the occurrence of island coalescence [100,101]. For cubic GaN, stacking faults play an important role for stress relief caused by the lattice mismatch and thermal coefficient difference
j 859
860
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.30 Schematic illustration of a prismatic stacking fault emanating from a step of partial character on the (0 0 0 1) surface of the 6H-SiC substrate. The stacking fault plane is ð 1 2 1 0Þ, which is inclined in this projection. The stacking fault has been taken here to exhibit the rigid-body translation corresponding to the
Blank model. (In GaN, large and small circles indicate distinct atomic species, and open and filled circles denote atoms at height 0 and aGaN/2, respectively. In SiC, distinct hatches represent distinct atomic species.) The term x is the line direction of the defect, pointing into the page. Courtesy of Dimitrakopulos et al. [38].
between the GaN and substrate. The formation energy of stacking fault decreases with Si doping compared to undoped GaN [102,103]. This was explained in terms of a decrease in the magnitude of the Mulliken charges on the atoms, as silicon is incorporated into the lattice, and an increase in the overlap populations. Occurrence of stacking faults strongly depends on growth condition. For example, Ga adatoms have a significantly lower diffusion barrier than that of N adatoms on both N- and Ga-polarity surfaces [104]. Under N-rich conditions, the Ga diffusion barrier is greatly increased because the Ga-N bonds are readily formed. The diffusion length of Ga adatoms is significantly reduced under N-rich growth conditions, which
Figure 4.31 High-resolution cross-sectional XTEM micrograph of GaN film grown on Si(0 0 1) Showing stacking faults. Courtesy of D. Wang.
4.1 A Primer on Extended Defects
enhances the formation probability of stacking faults. A large fraction of the dislocations near the interfacial region has partial character because they are associated with stacking faults. Figure 4.32 is a high-resolution image of partial dislocations terminating stacking disorder. Figure 4.32a shows one small stacking fault loop with both ends having a Frank partial. In Figure 4.32b, a wide stacking fault is terminated by a Frank partial. Figure 4.32c shows an area with two Shockley partials on the left-hand side of the image and one Shockley partial on the right-hand side. These Shockley partials were suggested to be one of the sources that are responsible for the formation of a large density of threading edge dislocations. It was suggested that a-type threading dislocations form by the climb of basal plane dislocations present near the interfacial region. These basal plane dislocations are formed primarily by the dissociation of Shockley partials, bounding stacking faults near the GaN/sapphire interface. A Shockley partial (As) can dissociate into another Shockley partial (Bs) and a perfect basal plane dislocation (AB: b ¼ 1/3h1 1 2 0i). AB glides away from the fault and can relax the local stress. The c-type threading dislocations may form as a result of the coalescence of 1/2h0 0 0 1i type Frank faults c þ a threading dislocations may be formed either as a result of the reaction between c and a threading dislocations or by the coalescence of complex 1/6h2 0 2 3i type Frank faults [105].
Figure 4.32 High-resolution image of partial dislocations terminating stacking disorder. (a) Interstitial Frank loop, the position of extra half-plane is indicated by the arrows in the image; (b) isolated Frank partial dislocation; (c) reaction between a dissociated edge dislocation and Shockley partial that is displaced by one basal plane. Courtesy of D. Wang.
j 861
862
j 4 Extended and Point Defects, Doping, and Magnetism 4.1.5 Grain Boundaries
Grain boundary (GB) defects form a compilation of an array of line defects superimposed on some singular configuration [106]. The driving force for forming these array defects is to accommodate small angular deviations from a misorientation. If the angular misorientation is increased, the grain boundaries transform from the low-angle to the high-angle regime, and the array of dislocations becomes very dense. We should note that the GaN crystal rotates about the c-axis when grown on sapphire to reduce misfit. Although there is no simple correlation between geometric parameters and interfacial free energy, considerable evidence is available that supports the notion that the two-dimensionally periodic boundaries are sometimes favorable. There is also some evidence that other one-dimensionally periodic structures are singular, for example, certain boundaries in ZnO [107]. It is illustrative to identify misorientations corresponding to potentially singular values. Because the GaN films studied here grow with their basal plane nearly parallel (tilt) to the Al2O3 substrate, the most common type of misorientation encountered is a relative rotation of one crystal with respect to an adjacent one about their common [0 0 0 1] axis, which is loosely termed as twist. A comprehensive analysis would go along the lines of a method developed by Pond and Vlachavas [108] that identifies the symmetry of dichromatic patterns and complexes. A close examination of grain boundaries indicate that salient ones have rotations close to the special values of 13.17 , 21.79 , and 17.90 with corresponding coincident site densities being S ¼ 7, S ¼ 19 (asymmetric), S ¼ 31 (symmetric), respectively. It is also possible that the interfacial defects are superimposed on a bicrystalline configuration as illustrated in Figure 4.33a, where an HREM micrograph of a disconnection on a ð 23 1 0Þ symmetric tilt grain boundary in a epitaxial a-GaN observed by Potin et al. [109] is shown. A closed circuit SPQTUVS is drawn around the defect to identify its fault character. The circuit maps to SPQTUVF in the reference space (Figure 4.33b), and closure failure FS ¼ b ¼ 1/21½1 5 4 0 appears (expressed in the l coordinate frame). This particular defect is a dislocation arising due to the misorientation of lattice vectors. Using the Volterra approach [110–112] (another method for determining the nature of the defect, which is equivalent to the circuit mapping), it is determined that the defects result from the juxtaposition of two incompatible surface steps. The resulting disconnection separates crystallographically equivalent regions of interfacial structure. The Volterra approach is closely related to the circuit mapping method in that one is the inverse of the other and the two methods are topologically equivalent. In circuit mapping, passive operators are used, that is, operators enacted on an imaginary observer who encircles the defect. In the Volterra approach the operations are active, that is, they are enacted on the medium (i.e., the crystals). The Volterra approach is an a priori technique useful for defect prediction, whereas the circuit mapping is mainly useful for a posteriori topological characterization on HRTEM micrographs. A detailed discussion of these grain boundaries can be found in Ref. [25] and the references therein. There are many sorts of boundaries that have been the topic of theoretical calculations and experimental investigations. The structural and electrical properties
4.1 A Primer on Extended Defects
Figure 4.33 (a) Plan-view HREM image of an interfacial disconnection at a ð2 3 1 0Þ symmetric tilt grain boundary in GaN, which exhibits dislocation as well as step character. A closed circuit SPQTUV is drawn around its line direction to identify its Burgers vector. (b) Mapping of the circuit to the reference space shows that the closure failure FS arises, which
identifies the Burgers vector as b ¼ 1/ 21½1 5 4 0. Large and small circles denote distinct atomic species. Open and filled circles denote atoms at heights 0 and ca-GaN/8, respectively. The term x depicts the line direction of the defect pointing into the page. Courtesy of Dimitrakopulos et al. [38].
of at least some of these boundaries, in the context of double positioning boundaries, are discussed in Section 4.1.6, which deals with electronic structure of extended defects. 4.1.6 Electronic Structure of Extended Defects
Ironically, GaN was not taken seriously early on in part because of large dislocation densities associated with this material. This apprehension was based on experience from other and more established semiconductors for which a large body of cumulative knowledge unequivocally indicated that extended defects are either directly or indirectly responsible for electronic defects that degrade device performance. This assessment in part was based on our understanding of extended defects and their often-obscure relationship to point defects. The demonstration, followed by
j 863
864
j 4 Extended and Point Defects, Doping, and Magnetism marketing, of LEDs led to a premature relaxation of scrutinizing extended defects. This change in approach did not last long as it became clear that to attain higher performance emitters such as lasers and electronic devices, it is necessary to lower the point defect concentration that is exacerbated either directly or indirectly by extended defects. This set the stage for a careful look at the electronic structure of extended defects in wurtzite materials. It should also be pointed out that materials containing large densities of extended defects also suffer from point defects that in one way or another express themselves in the form of degraded device performance. For example, electron mobility and, therefore, FET performance are very much affected by edge dislocations in part because they may trap electrons, providing line scatterers, as discussed in Volume 2, Chapter 3. Therefore, in addition to the structural properties of extended defects, their electronic signature is of great interest, to say the least. Experimentally, it is very difficult to deconvolve the many contributions by many defects, particularly in the early stages of development of a new semiconductor. Therefore, theoretical investigations in which the environment can be specified are extremely useful. Manifestation of point defects in terms of gap states are discussed in Section 4.3. Here, the discussion is limited to atomic structure extended defects, but only pertains to electronic nature. Several calculations dealing with the electronic nature of extended defects have been reported, albeit without complete agreement, as exemplified by the case surrounding screw dislocations. To underscore the point again, pure screw dislocations, pure edge dislocations, and mixed dislocations (having screw and edge components) are among the extended defects that thread through the crystal. The question then is if any of these or other structural defects are electrically active and if so in what form. Let us now discuss the electrical nature of extended defects by first providing an overall glossary of the often conflicting developments. The literature has conflicting reports both in terms of theory and experiments regarding whether extended defects are or are not electrically active. We should hasten to state, however, that a reasonably large body of work leads to the conclusion that extended defects directly or indirectly cause electrically active centers. Even when they are fully coordinated, they can do so by trapping impurities and point defects and by forming complexes due in part to large local strain induced. Pertaining to the structural configuration of screw and edge dislocations, there is considerable divergence whether they have open or full cores. Moreover, there is considerable divergence in the electronic structure of edge and screw dislocations, which is naturally imbricated with the closed- or open-core debate. To start with, experimental evidence on the atomic scale is available in literature indicating that intrinsic dislocations are electrically inactive [52,62]. Countering this again on the experimental side, however, is a study, albeit with lower spatial resolution, which points to the presence of some degree of localized states [113]. The theory is divided on the matter in favor of electrical activity (involving pure edge and screw dislocations) [114,115] and against [41]. The latter point of view may have been arrive at due to the small cell size used. Supporting the former point of view, electron holography [116] and scanning capacitance measurements [117] have shown that edge dislocations are negatively charged. In fact, electron scattering from these line charges is discussed in
4.1 A Primer on Extended Defects
some detail in Volume 2, Chapter 3. Moreover, other electron holography data exist, suggesting that all dislocations are electrically active (negatively charged in n-type GaN) [118]. Furthermore, conductive atomic force microscopy (CAFM) studies have shown that at least a portion of screw dislocations form electrical current conduction paths, the details of which are discussed in Section 4.2.5 dealing with the signature of structural defects. The apparent divergence in the literature might have its genesis in complications emanating from the difficulties associated with delineating the nature and effect of intrinsic dislocations from those of complexes formed between extended defects and point defects and impurities [46,62,119], a topic that is discussed in Section 4.3. 4.1.6.1 Open Core Versus Close Core in Screw Dislocations To restate, Elsner et al. [41] reported that pure screw dislocations in an eightfold ring are atomically fourfold coordinated, see Figure 4.9, and are therefore not electrically active. However, Elsner et al. [41] left the possibility open that the associated strain field may trap point defects and impurities. The open core and implicitly the electrical neutrality predicted by Elsner et al. [41] are consistent with some experimental investigations that point to open-core screw dislocations [59]. Contrary to this, a first principle total energy calculation pointed out that a screw dislocation having a Ga-filled core is energetically more favorable than the open-core structure [47]. This and also the observations that screw dislocations could have nonmetallic closed-core structures are discussed in Sections 4.1.1.2.2 and 4.2.1. In terms of the electrical activity, it has also been predicted that the screw dislocations whose cores contain excess Ga also generate states within the bandgap. Relevant to the topic, Arslan and Browning [121], using multiple scattering simulations along with experimental Z-contrast high-resolution TEM images, reported that changes in the local electronic structure of intrinsic dislocations of all types, inclusive of screw dislocations, are attributable to changes in the local symmetry. Further, no identifiable states result in the bandgap due to changes in local structure. However, they added that any electrical activity appears to be related to segregation of dopants, unintentional impurities, and point defects, consistent with the local strain suggested by Elsner et al. [41]. Using correlated techniques in the scanning transmission electron microscope, Arslan et al. [62] expanded on their earlier work and reported direct atomic scale experimental observations of oxygen segregation to screw dislocations in GaN. The extent of oxygen contamination in each of the three distinct types of screw dislocation core was found to depend on the evolution and structure of the core, giving rise to a varying concentration of localized states in the bandgap. Moreover, substitution of oxygen for nitrogen was observed to extend over as many as 20 monolayers (ML) for the open-core dislocation. The center of the filled nanopipe and the full-core screw dislocation were not observed to have a quantifiable amount of oxygen. The authors stated that presence of a pre-edge in the nitrogen EEL spectrum at positions of appreciable oxygen presence appears to confirm the link between oxygen and the creation of states in the bandgap. The gist of the study is that even when the core is fourfold coordinated with no broken bonds, the strain field and some other
j 865
866
j 4 Extended and Point Defects, Doping, and Magnetism mechanisms cause impurity segregation to the region around the core of the screw dislocation, giving rise to states. Just for reference, O forms shallow donor k levels in bulk GaN. Albrecht et al. [122] invoked the argument of splitting a dislocation into two leads to dangling bonds at the core, therefore rendering it electrically active. The authors investigated radiative and nonradiative recombination at individual dislocations in GaN by cathodoluminescence performed in a transmission electron microscope. In this particular study, the dislocations were produced by indentation of nearly dislocation-free single bulk crystals. The process produced a-type dislocations with Burgers vectors (b ¼ 1/3h1 1 2 0i) that were aligned along the h1 1 2 0i directions in the basal plane. The direct correlation between structural and optical properties on a microscopic scale yielded that 60 basal plane dislocations show radiative recombination at 2.9 eV and screw-type basal plane dislocations act as nonradiative recombination centers. The authors explained the nonradiative recombination by splitting this dislocation into 30 partials that have dangling bonds in the core, which has not been included in the reported calculations. The dissociation width of these dislocations was <2 nm. Somewhat connected to open-core dislocations are the experimental observations that show void-type defects in GaN, termed as nanotubes and pinholes [56,123]. The density of these defects was estimated to be in the range of 105–107 cm2 and their radii in the range 3–1500 nm. Depending on their local configuration and strain around them, they can be eclectically active. If all the bonds are fully (fourfold) coordinated, then any electrical activity would have to be associated with trapped impurities, point defects, and complexes caused by local strain due in part to bond length deviations. Liliental-Weber et al. [56] suggested that nanotubes and pinholes may have their origin in impurities present in the material. The issue of impurity, segregation, point defects, and complexes being attracted to or trapped by fully coordinated open-core screw and edge dislocations is consistent with the suggestion by Elsner et al. [41] (discussed in Section 4.3.1). Further, Liliental-Weber et al. also observed that the nanopipes close and open again along their axis as they propagate up through the sample much like bamboo canes. On the theoretical side, Lee et al. [124] employed density functional calculations to predict that GaN nanotubes are stabile and also provided electronic structures of GaN nanotubes. 4.1.6.2 Edge and Mixed Dislocations As in the case of screw dislocations, whether a given edge dislocation is electrically active or not hinges in part whether its core is open, fully filled, or nonstoichiometric. Pure edge and mixed dislocations have been shown to have the same projected core image, which consists of an eightfold ring [41,42,52] (Figure 4.7). Using a variety of computational methods encompassing local density functional cluster method, as well as density functional theory based tight-binding method, Elsner et al. [41] determined the core structure of pure edge dislocations in their relaxed form. Additionally, they arrived at the conclusion that their lowest energy state is the full-core structure (in contrast to screw dislocations as far as their calculations indicate) and further they are electrically inactive. However, Elsner et al. [41] hastened
4.1 A Primer on Extended Defects
to note that two of the core atoms have very stretched bonds which would give rise to stress fields and could act as traps for impurities and point defects. Electrically inactive scenario has been heavily cited without mentioning the accompanying statement about strain and resultant possible trapping of impurities, carriers, and point defects. Also, not necessarily cited is that Elsner et al. [41] considered a relaxed eight-ring full-core edge dislocation, but there are other edge dislocations such as four core and 5/7 varieties that are discussed below. As a matter of fact, the strain issue is a rather important one in that what may be an electrically inactive defect in the case of fully coordinated atoms (no broken bonds) could be made active deep state by the giant local strain field around the dislocation core in combination with the small lattice constant of GaN. Consequently, an oftencited case, particularly early, that indicated dislocations to be electrically inactive would have to be revisited. In the realm of DFTcalculations performed by Lymperakis et al. [125], pure edge dislocations were found to be electrically active, independent of the core structure. In fact, the authors used high-resolution TEM images of rather thick HVPE-grown GaN along with their DFTcalculations to arrive at that conclusion. This particular investigation is interesting and is detailed at the end of the discussion devoted to electrical nature of edge dislocations. In addition to the effect of strain, ab initio calculations have shown that the edge dislocations may be charged, giving rise to deep gap states [126], which contrast the initial theoretical calculations in which the cell size was too small [41]. Notwithstanding the cell size, the investigation of the effect of the Fermi level position (doping) and the stoichiometry has shown that the dislocation core (pure edge dislocation) in their relaxed state with Ga vacancies is most stable in n-type material. In contrast, a structure without vacancies is predicted in p-type material. Assuming that the yellow emission is associated with Ga vacancies, this picture is consistent with the lack of yellow emission in p-type GaN. Furthermore, in material grown under Ga-rich conditions, a structure having nitrogen vacancies at the dislocation core is predicted to be most stable for p-type GaN, whereas a variety of core structures are possible in n-type GaN, for example, full core, open core, with Ga vacancy, and N vacancy. The underpinning outcome of these density functional theory calculations is that pure edge dislocations would act as electron traps in n-type material and may act as hole traps in p-type material, depending of course on the growth conditions. Leung et al. [127] performed Monte Carlo calculations to determine the charge accumulation on a threading pure edge dislocation in GaN as a function of its density and doping density. Four possible core structures, mentioned above, have been examined. They can all produce defect levels in the gap and may, as a result, act as electron or hole trap. The result further indicates that charge accumulation and the resultant electrostatic interactions can indeed change the relative stability of the four different core structures. According to the simulations, N-rich growth conditions favor the Ga-vacancy structure, whereas Ga-rich growth conditions favor the N-vacancy structure. The charge trapped at the core of pure edge dislocations is of the line charge nature and impacts electron mobility adversely, a topic discussed in detail in Volume 2, Chapter 3.
j 867
868
j 4 Extended and Point Defects, Doping, and Magnetism In the same vein, with an emphasis on the four types of core structure, Lee et al. [115] investigated atomic and electronic structures of threading pure edge dislocations in GaN, using SCC-DFTB. Full-core, open-core (also called 5/7 structure due to juxtaposed five-atom ring and seven-atom ring structure), Ga-vacancy, and N-vacancy edge dislocations (four core structures) have been assumed to be fully relaxed in the computations. The atomic configuration, stick and ball, for all the four core structures are shown in Figure 4.34. Among all the four core structures investigated, the Ga-vacancy dislocation core variety was found to be the most stable in a wide range of Ga chemical potentials. The full-core and open-core dislocation core structures were found to be more stable than the others in the Ga-rich growth regime. It must be noted that during the lattice relaxation near a dislocation in all cases, a partial dehybridization takes place. In terms of the electrical activity, the dangling bonds at Ga atoms are the ones that contribute to the deep gap states the most, whereas those at N atoms contribute to the valence band tails. These finding are somewhat different in that the theories discussed up to this point predict electrical inactivity unless full coordination is not maintained, and large stress field are present. The work of Lee et al. [115] suggest electrical activity in terms of states in the gap or tail states induced by atoms that are heavily deviated from their bulk positions, labeled A–E in Figure 4.34. A somewhat more detailed discussion of the results from this particular approach is given in the following section. Expanding on the above succinct discussion, local density of states (LDOS) were calculated by projecting the density of states (DOS) to specific orbitals of a given atom in an effort to determine any the contribution to the density of states in the gap by atoms defining the dislocation core. Comparing the calculations of LDOS associated with atoms A–E in the full-core configuration, see Figure 4.34, to bulk GaN shows several gap states due to the dislocation. The s-orbitals of Ga atoms contribute to relatively deep unoccupied states, whereas p-orbitals of Ga atoms contribute to both the valence band tail states and gap states. The s-orbitals of N atoms contribute mostly to unoccupied deep states, whereas the p-orbitals mostly contribute to the valence band tail states. It should be pointed out that LDOS calculations overestimate the GaN bandgap at 6.52 eV, the correction for which is typically done by a scissor operation that entails shifting the unoccupied states to the experimental values. This approach is not easily employed when gap states are involved. Consequently, Lee et al. [115] adopted a linear scaling method in which all energy levels in the gap are projected to the valence and conduction bands of bulk GaN. Then the energy levels are shifted downward by the fraction of the bandgap correction. The gap states listed in Table 4.1 have been obtained in this manner. For the full-core configuration (Figure 4.34a) calculating LDOS for Ga and N atoms, A–E, it was found that the s-orbitals of Ga atoms contribute to the relatively deep unoccupied states, whereas the p-orbitals of Ga atoms contribute to both valence band tail states and deep gap states. The d-orbitals do not contribute to gap states. The s-orbitals of N atoms contribute mainly to the unoccupied deep states, whereas the p-orbitals chiefly contribute to the valence band tail states. The unoccupied gap state at 2.1 eV above the VBM originate from the s- and p-orbitals of Ga atoms at A and
4.1 A Primer on Extended Defects
Figure 4.34 Stick-and-ball atomic configuration of fully relaxed geometries of four varieties of perfect edge dislocation cores, namely, (a) fullcore (also called eight-atom ring dislocation), (b) open-core also called pentagon/heptagon 5/7 atom ring dislocation, (c) Ga-vacancy, and
(d) N-vacancy edge dislocations. The larger ball and the smaller ball represent Ga and N atoms, respectively. The heavily distorted atoms are identified as A–E. The boxed region in (a) shows the size of the supercell used in the study of Lee et al. [115].
j 869
870
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.34 (Continued )
4.1 A Primer on Extended Defects Table 4.1 The position of gap states with respect to the valence
band maximum in terms of eV localized at each site of the pure edge dislocation associated with heavily distorted atomic positions A–E. Dislocation type
A site
B site
C site
D site
E site
Full core
2.0 2.7
2.1 3.0
2.0
2.7 2.9
3.0
3.0
1.4
Open core Ga vacancy
0.2 3.2
3.1
N vacancy
1.4 2.2 3.1
0.6 1.9 2.9
1.4 3.2
1.3 2.2 3.0
0.6 2.6
2.8
The values have been obtained by the linear scaling method. Refer to Figure 4.34 for atomic positions and dislocation types [115].
B sites. The level at 2.9 eV above VBM is due chiefly to s orbitals of N atoms at B, D, and E sites. The tails states are mostly due to N atoms at A site. For open-core configuration (Figure 4.34b) that is also called the 5/7 configuration, unoccupied gap states at 1.4 eV above VBM are due to s- and p-orbitals of Ga atoms and p-orbitals of N atoms. Valence band tail states are due to p orbitals of Ga and N atoms, and conduction band tail states are mainly due to s-orbitals of Ga and N atoms. The gap states at 1.4 eVabove the VBM are due to s- and p-orbitals of Ga atoms at site C and to a lesser extent those at site E. The Ga atom at site B contributes to the 3.0 eV state that is of a shoulder character around the conduction band tail states. For the Ga vacancy configuration (Figure 4.34c), no deep gap states were found. The conduction band tail states are due to s-orbitals of Ga and N atoms. The p-orbitals of N atoms mostly contribute to the valence band tail states. The type of dislocation contributing to shallow unoccupied states the most, which exist at 0.2 eV from the conduction band minimum (CBM) or approximately 3.2 eV from the VBM, are due to s-orbitals of the Ga atom at B and D sites. The N-vacancy configuration (Figure 4.34d) induces several gap states over a wide range. The occupied state at 0.6 eV above VBM is due to B and D sites. Note that Ga atoms (small electronegativity) are only at site A, whose excess charge is depleted by nearby N atoms (large electronegativity), which reduces screening. Thus, the repulsive forces among the Ga atoms are increased, causing lattice distortion. The other states shown in Table 4.1 are less prominent. In essence, all edge dislocations can act as deep trap centers, the energies of which are illustrated schematically in Figure 4.35. The lines with filled circles depict the occupied states. As can be seen, several unoccupied gap states exist in all cases even with full-core pure edge dislocations. To reiterate, most of the contribution to the gap
j 871
872
j 4 Extended and Point Defects, Doping, and Magnetism EC 3.2 3.0
3.1 2.9
3.0
2.7
2.6 2.2
2.0
1.9 1.4
1.3
0.6 0.2
EV
Full core
Open core
VGa
VN
Figure 4.35 The schematic diagram of energy levels in the bandgap for each of the four perfect edge dislocation core structures. Filled circles indicate the occupied states [115].
states are due to the s- and p-orbitals of Ga atoms, whereas N atoms contribute to the valence band tail in most cases. Note that all energy levels in the gap have been projected to the valence maximum and conduction band minimum of bulk GaN. The energy levels were then shifted downward by the fraction of the bandgap correction, which is overestimated at 6.52 eV by the LDOS calculations. Charge nature of pure edge dislocations with a line direction of [0 0 0 1] and Burgers vector of 1/3½1 2 1 0 has been interrogated by Blumenau et al. [29] using DFTB method and local density functional pseudopotential (ab initio modeling program – AIMPRO) calculations. Both approaches resulted in localized states in the upper half of the bandgap. The wave function associated with the lower gap states mixes orbitals associated with Ga at the columns labeled A, D, and F, whereas the higher gap states are associated with columns A, E, B, and H, see Figure 4.7b for column designation. The band structure for the negatively charged full-core edge dislocations with two electrons per repeat distance, c0, on the core of the dislocation has been calculated. The simulations indicate that addition of electrons to the dislocation is stabilized by a change in the atomic structure, which acts to lower the localized gap states of the neutral dislocation pushing them deeper into the gap. This charge state has been observed in electron holography experiments [116]. It can therefore be concluded that neutral edge dislocations would be present in p-type GaN, and the negatively charged dislocations give rise to localized states within the bandgap. As in the case of Lee et al. [115], Blumenau et al. [29] as well investigated the charge character of various core structures associated with edge dislocations, namely, the empty core where the central row of atoms are removed shown as A in Figure 4.7b (also see Figure 4.34b), and N- and Ga-vacancy structures. The core structure, however, could fluctuate and
4.1 A Primer on Extended Defects
along the dislocation line there might be both Ga-vacancy and open-core structure present, which is the 5/7 atom ring structure as experimentally observed [16]. From an experimental point of view whether a dislocation is charged or not can be addressed by using charge-sensitive mapping techniques at the microscopic scale with resolution comparable to the core dimensions. Naturally, one would expect differences in the charge state of the dislocation, depending on whether the semiconductor is n-type on p-type as pointed out by Wright and Grossner [126]. In the former case an edge dislocation can act as an electron trap and in the latter as hole trap. An electron holography study reveals the different electric charges associated with n-type and p-type GaN [128]. Figure 4.36a illustrates the experimental and theoretical potential profile across an isolated edge dislocation in n-type GaN, as calculated from the holographic map. The theoretical profile was obtained assuming that the dislocation is a line charge and screening is negligible. The electric field induced by the edge dislocation is E¼
NLe ; 2pee0 r
and the potential is NLe V0 ¼ lnðrÞ þ A; 2pee0
ð4:2Þ
ð4:3Þ
where NLe is the line charge per unit length, r, e0, and e are the radius, the permittivity of both free space and the relative dielectric constant, respectively, and A is a constant. By matching the experimental potential profile, the line charge was identified to be negative with a magnitude of 2e/c, where c is the unit cell spacing of GaN. For comparison, Figure 4.36b shows the map of the inner potential profile of an edge dislocation in p-type (Mg-doped) GaN. From this potential profile it is clearly seen that the dislocations in p-GaN are positively charged, with a magnitude of potential change of about 0.6 V, which is much lower than that of the negatively charged dislocations observed in n-type GaN. A simple model to explain the different charges of edge dislocation in n-type and p-type GaN is shown in Figure 4.37. This model suggests that dislocations produce deep energy states in the energy gap. In the case of dislocations in n-GaN, the deep-level states act as acceptors, and the electrons flow to the dislocations and cause band bending. The existence of these deep-level energy states has been verified by deep-level transient spectroscopy [129]. The situation in terms of dislocation charge in p-GaN is opposite of that in n-GaN, where the electrons flow away from the dislocation and cause band bending in the opposite direction. However, the results for dislocations in p-type GaN show more dispersion than for ntype GaN. The potential profile change at the dislocation core compared to the surrounding matrix can also be close to zero, or even negative. This variability is a result of different electronic structures of dislocations in p-type GaN, meaning either open-core structure or close-core structure would take place. On the experimental side, spatially resolved (on the order of atomic distance) EELS, if combined with scanning TEM, allows probing of the gap states of an extended defect. In the EELS method, the primary high-energy electrons pass through a thin
j 873
874
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.36 Potential profile across the dislocation in n-type GaN deduced from the holographic phase map compared with theoretical profile (a). Courtesy of D. Cherns. False colored map showing phase shifts produced by edge dislocations viewed end-on in nominally undoped GaN. Contour lines emphasize dipole-like phase shifts near
dislocation cores (b). Line profile through indicated dislocation in (b) allows quantification of nominally undoped GaN of electric fields, yielding an estimated bound surface charge of 4 · 1011 e cm2 on either side of the defect (c). Courtesy of M. McCartney. (Please find a color version of this figure on the color tables.)
4.1 A Primer on Extended Defects
2 1.5 1
Phase (rad)
0.5 0 –0.5 –1 –1.5 –2 –150
–100
(c)
–50 0 Distance (nm)
50
100
150
Figure 4.36 (Continued )
∆V Ec EF
Ev (a) Ec ∆V
EF Ev
(b) Figure 4.37 A model to explain the charging of dislocation in n- and p-type GaN observed by electron holography. (a) In n-GaN, electrons charge a dislocation state about 2.5 V below the conduction band, and pin the Fermi level, causing band bending. (b) In p-GaN, pinning of
the Fermi level at a similar, or possibly the same, dislocation state in the bandgap now requires electrons to diffuse away from the dislocation, thus causing band bending in the opposite direction and leaving the dislocation positively charged [128].
j 875
j 4 Extended and Point Defects, Doping, and Magnetism sample, as in the case of TEM, and some are inelastically scattered, losing a fraction of their energy. The energy is lost because of electronic transitions between the valence band and the conduction band, resulting in the low-loss spectra. The interaction can involve single-excitation events or collective modes called plasmons. If the transitions occur between tightly bound atomic core states and the conduction band, the nomenclature of core-excitation EELS or energy-loss near-edge structure (ELNES) is employed. The interaction between the high-energy electrons and the solid can be described as interaction with the polarization field of the solid caused by electrons in the solid. In this description the signal obtained is proportional to the imaginary part of the inverse of the macroscopic dielectric function [130]. K-edge core-excitation EEL spectra in GaN come from electronic transitions between the N-1s core electrons and unoccupied gap or conduction band states. If one neglects the final-state interactions between the excited electron and the core hole to first approximation, and keeping in mind the strong polarization of the initial state, the core EEL spectra are related to the p-like projected density of states on the excited N atoms. Calculated EEL spectra along with experimental data for bulk GaN (in defect-free region) is shown in Figure 4.38 for electron beam orientated normal (z-) and in basal plane (x, y-). The data in the basal plane are isotropic because of the threefold symmetry of the hexagonal crystal. The effect of the Ga-3d states have been
Bulk x,y orientation z orientation Experimental
0.20
EEL spectra signal (au)
876
0.15
0.10
0.05
0.00 0
4
6 2 Energy (eV)
Figure 4.38 Calculated (thin broken and solid lines) and experimental EEL spectra of bulk GaN (bold solid line representing GaN without defects) in arbitrary units. The calculated spectra are for x, y (electron beam is in the basal plane), and z-orientation (electron beam is
8
10
normal to the basal plane) as shown in Figure 4.7 The experimental spectra are taken with electron beam oriented along the ½2 1 1 0 orientation. Courtesy of A. Blumenau, J. Elsner, R. Jones, C.J. Fall, and M. Heggie.
4.1 A Primer on Extended Defects
taken into account by including them as valence states in the pseudopotentials that had little effect on the EEL spectra in the 0–10 eV range. The experimental data are orientation and agree well with the computed spectra, in terms taken along the ½2110 of peaks and shoulders for x- or y-orientation. As mentioned, if the spectra are taken on a dislocation core, they can, at least in principle, provide valuable information regarding the electronic structure of the dislocation, within and above the bandgap. The intent then would be to use that information in determining whether a particular undecorated dislocation is electrically active or not. If so, adverse effects on transport and optical properties would be expected. To be fully useful, experimental EEL spectra must be augmented with calculations to draw specific conclusions, although comparison with defect-free regions can be made to ascertain the charge state of the dislocation. In this vein, nitrogen K-edge spectra collected near a pure GaN edge dislocation have shown an increase in absorption just above the bandgap compared to the bulk regions [42]. Lowloss EELS, where the energy loss is induced by exciting electrons from the valence to the conduction bands, is another method that can be conducted in a scanning TEM that provides detailed information about the gap states. Cross-sectional low-loss EEL experiments in GaN have shown differences in the onset of the EEL spectrum on and off threading dislocations [114]. Fall et al. [131] performed first principles calculations of EEL spectra for edge and screw dislocations in GaN for an energy range of 0–10 eV. In doing so, those authors were able to determine how dislocations lead to changes in EEL spectra compared to bulk regions. Essentially, the gap states associated with undecorated full-core dislocations were found to lead to absorption below the onset energy associated with bulk (defect-free region) in low-loss EELS. Calculations by Fall et al. [131] indicated that a full-core negatively charged edge dislocation is stabilized in n-type material due to a change in core structure. Further, undecorated full-core dislocations were found to invariably lead to supplementary energy absorption signal within the bandgap. This should be observable experimentally with an electron probe precisely positioned on the dislocation core. In performing experiments, however, one must anticipate there is a trade-off in low-loss EEL measurements between deep levels and shallow levels. These create strong changes in the EEL spectra but are very localized and difficult to pinpoint spatially. The latter leads to smaller changes in the absorption spectra but are easier to find spatially. The model developed by Fall et al. of the core-excitation EEL spectra obtained near dislocations, and in particular near a neutral eight-atom pure edge dislocation, points to strong potential variations, almost 1 eV, at the cores of nitrogen atoms near dislocations. These variations would simply broaden any sharp peaks in the core EEL spectra and must be taken into account when interpreting core-excitation experiments. Fall et al. also found that the electrostatic potential at N atoms in the vicinity of a pure edge dislocation varied by an order of a volt, which underscores the complexity, particularly interpretation, of core-loss spectroscopy. Combining the theory by Fall et al. [131] (ab initio calculations within the local density approximation to density functional theory of the GaN and AlN band structure to simulate low electron energy loss spectra) discussed above and experiments utilizing EEL spectra in conjunction with STEM, Gutierrez-Sosa et al. [114]
j 877
878
j 4 Extended and Point Defects, Doping, and Magnetism undertook an investigation. This study probed the electrical nature of dislocations in the energy loss regime at bandgap and at the nitrogen near-edge (at the K-edge that goes by the nomenclature ELNES) structure of not only GaN but also InGaN, and (Al, Ga)N structures, focusing on GaN. The scattering intensity around the onset and the position of the onset energy in locations along the projection lines of isolated dislocations were mapped. Together with low-loss spectra calculations, it was concluded that all dislocated regions reveal bandgap states associated with all dislocation types in GaN. The scattering intensity at 3.3 eV of the simulated spectra, in particular for the full-core screw dislocation, is in qualitative agreement with the experimental findings. An absorption peak at 2.4 eV found in certain regions in the vicinity of dislocations was not present in the calculations and therefore was thought not to be produced by the dislocation and was attributed to impurity and/or point defect trapping by dislocations. This is consistent with that reported by Bangert et al. [132], reconfirming that the 2.4–2.5 eV emission, which appears in all experimental dislocation spectra, is not an intrinsic property of the dislocation as it is not present in calculated spectra. A practical but pertinent issue with EELS investigations is that one cannot extract features below 2 eV with great reliability and furthermore, the prebandgap region is masked by the aforementioned feature at 2.4–2.5 eV, the intensity of which has been noted to increase toward the surface and, of course, toward dislocations. This transition is most likely associated with Ga vacancies complexing with an impurity, such as O, which is discussed in some detail in Volume 2, Chapter 5. While on the point, it may be instructive to note that EELS can also be used to get at the band structure, dielectric function, and reflectivity. However, delocalization effects reduce the spatial resolution of this region of the spectrum quite significantly [133]. The calculated [29] EEL spectra for neutral and negatively charged edge dislocations in GaN using the supercell approach are shown in Figure 4.39a. Both the edge spectrum and corresponding bulk spectra have been modeled in a pure supercell. Consistent with the gap states, absorption lines not expected in bulk GaN (representing the defect-free case) appear below the band edge. When the electron beam is within the basal plane (x and y configuration), absorption energies associated with the edge dislocation are lower than in the case of the z-configuration. Additional absorption lines in the range of 5–7 eV are also seen. The calculated EEL spectra for negatively charged edge dislocations, using the supercell method, in GaN are shown in Figure 4.39b. The absorption within the bandgap for all three orientation of the electron beam is seen at lower energies as compared to the neutral case. The EEL spectra for neutral screw dislocation in the full-core and Ga-filled configurations, described in Figures 4.8 and 4.9, calculated by Blumenau et al. [29] are shown in Figure 4.40. The screw dislocation spectrum has been modeled in a hybrid model as a screw dislocation in the supercell and would be computationally very demanding. For a valid comparison, the bulk spectra also was calculated using such as hybrid model. Strong absorption within the energy gap is seen, as discernable differences in cases where the electron beam oriented in the x, y (in the basal plane), and z-direction along the c-direction is seen. In addition, the Ga-filled screw dislocation appears to differ from the bulk, representing the defect-free case, to a
4.1 A Primer on Extended Defects
Neutral dislocation edge 0.25 Dislocation x orientation y orientation
EEL spectrasignal (au)
0.20
z orientation
0.15
0.10
0.05 Bulk GaN x,y orientation z orientation
0.00 0
4
2
6
10
8
Energy (eV)
(a)
Negatively charged edge dislocation 0.25 Dislocation x-orientation y-orientation
EEL spectra signal (au)
0.20
z-orientation 0.15
0.10
0.05
Bulk GaN x, y-orientation z-orientation
0.00 0 (b)
4
6 2 Energy (eV)
Figure 4.39 The calculated EEL spectra of a region containing a GaN edge dislocation (bold solid stroke for x-orientated electron beam, bold broken stroke for y-orientated electron beam, bold long/short broken line for z-orientated electron beam) and bulk GaN (thin broken line for x, y-orientated electron beam and solid line for z-orientated electron beam represents the defect-free case, termed as bulk GaN). The results are shown for the supercell model
8
10
shown in Figure 4.7b and are given for an electron beam oriented along x, y, and z, using the coordinate system defined in Figure 4.7b; (a) the neutral threading edge dislocation (a): The negatively charged threading edge dislocation (b); the dislocation carries a charge of two electrons per c0. The edge dislocation has been modeled in a pure supercell, as was the corresponding bulk spectrum.
j 879
j 4 Extended and Point Defects, Doping, and Magnetism Full dislocation screw core 0.25
Dislocation x,y orientation z orientation
EEL spectra signal (au)
0.20
0.15
0.10
0.05 Bulk x,y orientation z orientation
0.00 0
4
6
2
(a)
8
10
Energy (eV) Ga-core dislocation screw 0.25
Dislocation x,y orientation z orientation
0.20
EEL spectra signal (au)
880
0.15
0.10
0.05 Bulk GaN x,y orientation z orientation
0.00 0
(b)
4
2
6
8
10
Energy (eV)
Figure 4.40 Comparison between the computed EEL spectra of a region containing a neutral screw dislocation (bold solid stroke for x,y-orientated electron beam, bold long/short broken line for z-orientated electron beam) and bulk GaN (thin broken line for x, y-orientated electron beam and solid line for z-orientated electron beam representing the defect-free case, termed as bulk GaN). Results are given for an electron beam oriented along x, y, and z, using the coordinate system defined in Figure 4.7b. Because of the threefold symmetry
of the screw dislocation, EEL spectra for x and y orientations are identical, which for a full-core screw dislocation is shown in (a) and for Gacore screw dislocation is shown in (b). The screw dislocation has been modeled in a hybrid model because a screw dislocation in a supercell would have been computationally very demanding. For reasons of comparison, also the bulk spectra have been calculated using such a hybrid model. See Figures 4.8 and 4.9 for details regarding the full and Ga screw dislocation.
4.1 A Primer on Extended Defects
larger extent than the full-core structure. The genesis for this can be traced to the metallic-like bonding in the center of the dislocation core. In fact, the EEL spectrum approaches a linear dependence on energy for low energy losses, particularly for the x, y configurations of the electron beam that is consistent with the dielectric function of a free-electron system. The calculations demonstrate that EEL spectra are sensitive to the dislocation, its configuration, and its charge state and would represent a valuable method for the characterization of extended defects. In addition, the coreEEL spectra can also be simulated. The local strain field due to the extended defect would produce an electric field that in turn would cause a shift in the core energies observed. Doing so for a neutral edge dislocation for the N-core levels indicated that the atoms in the compressive region of the dislocation, column A, B, and H, appear in shallower core levels than those in the tensile region of the dislocation indicated by columns D, E, and F in Figure 4.7. The differences calculated are as large as 1 eV, which is not negligible as compared to the bandgap of GaN. These potentials and associated fluctuations tend to broaden the EEL spectra as they directly change the Ncore states. The final empty states are more delocalized and average over the potential fluctuations. Calculated N K-edge EEL spectra for bulk and for a dislocation demonstrate that the one for bulk shows a strong absorption at 5 eV above the valence band maximum that is larger than the bandgap of GaN. States near the conduction band minimum are formed by Ga orbitals and thus have only a small overlap with the initial N-related core EEL spectra. Near the dislocation, the theory predicts substantial absorption. The abovementioned broadening is enhanced for charged dislocations, making the interpretation of EEL spectra difficult, as below the gap absorption does not necessarily point to the presence of empty gap states at that particular energy in the gap. To reiterate, potential shifts near the neutral dislocations do not significantly affect the low-loss EEL spectra because both the initial and final states are delocalized and not very susceptible to short-range potential fluctuations [29]. The electronic structure is naturally sensitive to the type of model used in that the hybrid model is finite in the z-plane and therefore the wave function is confined. Apparently, this mainly affects the bulk peak close to 9.5 eV (Figure 4.39) and shifts it toward 8 eV (Figure 4.40). Effects for the remaining bulk peaks are less substantive and the intensity is really a question of scaling. As for the bulk spectrum that obtained from the supercell spectrum is most precise and thus reliable, as an infinite undisturbed crystal is modeled. Whereas in the hybrid model for bulk, one considers a nanorod of material oriented along [0 0 0 1]. Therefore, the hybrid-bulk spectrum is only useful only for comparing against the respective dislocated model. In what can be characterized as a comprehensive investigation, Lymperakis et al. [125] stressed the importance of considering the role of broken bonds, if any, at the core of the dislocation and just as importantly the local strain induced around the core of the dislocation where atoms are fully coordinated as in the bulk but slightly displaced with respect to the bulk positions. Because of any broken bonds, the core region is expected to be electrically active, giving rise to deep levels, whereas the regions around the core and under strain are expected to produce shallow levels. If the core region is allowed to restructure itself to eliminate broken bonds through
j 881
882
j 4 Extended and Point Defects, Doping, and Magnetism forming new bonds and/or rehybridization, the core can be made electrically inactive. Using a combination of high-resolution TEM experiments and theoretical calculations, Lymperakis et al. performed a comprehensive analysis of edge dislocations. In the process they also discovered a new type of edge dislocation wherein all core atoms are fully coordinated, surrounded by four anions or cations. However, due to local strain and aided by the small lattice constant of GaN, even this fully coordinated dislocation is predicted to be electrically active. Lymperakis et al. [125] were able to obtain high-resolution TEM images of pure edge dislocation with four different core structures with the help of image simulations that are shown in Figure 4.41. Images c and g are for full-core dislocations, images b and f are for the four-core structure, and images d and h are for open-core structure. A pivotal point is that the experimental image in Figure 4.41a with the simulated contrasts in Figure 4.41c and d shows no correspondence. Therefore, the standard core structures we discussed so far in this section fail to reproduce the experimental images and can be ruled out as structural models. To overcome the abovementioned conundrum, a theoretical analysis of the atomic geometry and the energetics was undertaken. In this vein, the atomic geometries calculated with DFT theory were used as input for image simulations, the results of which are shown in Figure 4.42c for the 5/7 edge dislocations, corresponding to Figure 4.41d and h, and in Figure 4.42b for full core (also called eight-core or atom ring structure) corresponding to Figure 4.41c and g and the Ga- and N-vacancy structures, where undercoordinated Ga and N atoms, respectively, have been removed from the full-core structure. The electronic character of the different dislocation cores has been analyzed using the results from the ab initio calculations. An inspection of the total density of states showed that all cores have states in the bandgap and are to be considered electrically active. This is an expected behavior for the open- and full-core structures, because
Figure 4.41 Threading dislocation in GaN at defocus values Df ¼ 23 and Df ¼ 63 nm (unfiltered HR-TEM) (a) and (e). Image simulations for the four-core structure (b)and (f), full-core structure (c) and (g), and the open structure (d) and (h). The upper (lower) panel shows defocus values Df ¼ 23 nm (Df ¼ 63 nm) [125].
4.1 A Primer on Extended Defects
Figure 4.42 Contour plots of dislocation-induced electronic gap states for three edge dislocation configurations, namely, (a) fourcore, (b) full-core, and (c) open-core structures. The plots are obtained by calculating atomic geometries with DFT theory used as input to image simulations. Large (small) balls correspond to Ga (N) atoms [125]. (Please find a color version of this figure on the color tables.)
both have broken bonds in the core region and the expectation is consistent with other DFT calculations [114,131] except for the four-core structure with no broken bonds. To figure out the mechanism that is responsible for the defect state the associated wave function has been analyzed, as shown in Figure 4.42a. The defect state can be characterized by orbitals between the two Ga atoms and is caused by the large local strain field vicinal to the defect made worse by the small lattice constant of GaN. The empty bond states induced by strain appear in all configurations of edge dislocations and can thus be considered a general feature of all a-type dislocations in GaN (Figure 4.42a–c). The general conclusions of this particular investigation are in general agreement with results reported by Bangert et al. [132] who observed a defect state at 3.2 eV above the valence band maximum. The main difference between the fully coordinated four-core structure and the other core structures is that in the latter case the dangling bond orbitals also clearly contribute to this state. In yet another effort to shed some light onto the electronic nature of dislocations, Arslan et al. [51], in the realm of mixed dislocations, undertook an atomic scale investigation that also provided evidence for partial splitting of dislocations in wurtzite structures. The atomic scale nomenclature has taken a new meaning with the demonstration of spatial resolution on the order of 1 Å by Batson et al. [134] with a STEM and calling, referring to the microscope as superSTEM. Arslan et al. [51] utilized a superSTEM equipped with a 0.4 eV energy resolution to perform atomic resolution electron energy loss spectroscopy in an effort to determine the charge state of states around the dislocations. Instruments equipped with aberration-corrected electron optics allow atomic structure images without image simulations, with the resultant delineation of screw component from the edge dislocation in mixed dislocations, taking advantage of the strain present around the screw component. As part of their investigation of the structural aspects of mixed dislocations, Arslan et al. [51] undertook EELS experiments to confirm the dissociated components of mixed dislocation into edge, screw, and stacking fault for confirmation [53] by comparing the spectra obtained near and away from the defect site. The spectrum
j 883
884
j 4 Extended and Point Defects, Doping, and Magnetism taken from the nine-atom ring side on the right, see Figure 4.17b, showed a fine structure similar to that of the bulk. However, the spectrum taken from the sevenatom ring side on the left was drastically different, indicative of increase in the density of unoccupied states at the conduction band minimum in these two areas. Relying on the available data on screw dislocations and analysis of the other two types of dislocation cores [52,121], the fine structure is attributed to the screw component of the mixed dislocation. It should be pointed out that spectrum similar to that observed at the nine-ring region representing the bulk was reported to be similar to that observed for edge dislocations [52,121]. Moreover, the spectrum taken from eight- and four-atom ring regions exhibit fine structure similar to that obtained from screw dislocations, as that region contains distortion in the c-direction. The comparative analysis and deduction suggest that when a mixed dislocation dissociates, one part of the partial maintains perfect edge dislocation character while the other side perfect edge dislocation. Therefore, Arslan et al. [51] concluded that dissociation of a mixed dislocation to pure edge, pure screw, and stacking fault increases the total energy of the configuration. Occurrence of the aforementioned dissociation suggests that impurities and point defects could be segregating to the dislocation. It can be concluded that dislocations are most likely active as they are capable of interacting with point defects and impurities and forming complexes. 4.1.6.3 Simple Stacking Faults: Electrical Nature Whether stacking faults, which are discussed in Section 4.1.4, are electrically active or not has been addressed by Bandic et al. [135]. For the nature of a simple stacking fault along the c-direction of the wurtzitic cell, refer to Figure 4.24 and the text that follows it. Bandic et al. employed the local empirical pseudopotential theory that led them to conclude that the stacking faults introduce electronic levels close to the valence band, which originate from the heterocrystalline wurtzite/zinc blende interface states. The stacking faults in both zinc blende and wurtzite GaN were predicted to introduce electronic levels within the bandgap at the energy of 0.13 0.01 eV above the valence band top. Further, the transitions between conduction band electrons and these levels coincide with the eA-type transitions observed in photoluminescence experiments. The above discussion, however, is not without controversy in that Elsner et al. [136] reported that the prismatic stacking fault composed of fourfold and eightfold juxtaposed rings forming domain boundaries in the f1 1 2 0g planes do have not wrong bonds and do not form deep states and are predicted in the gap, which is controversial [54,88]. Northrup et al. [77] performed total energy calculations for several possible models and concluded that an inversion domain boundary involving a c/2 translation along the [0 0 0 1] direction has a very low domain-wall energy and is thus a suitable candidate for many of the vertical defects observed on f1 0 1 0g planes, see Figure 4.20a for a stick-and-ball illustration. At this shifted inversion domain boundary denoted by IDB , all atoms remain fourfold coordinated with Ga-N bonds across the boundary and therefore do not induce electronic states in the bandgap. Furthermore, Northrup et al. [77] investigated a double-position boundary (DPB-II). Northrup et al. also stated that stacking mismatch boundary has a higher formation
4.1 A Primer on Extended Defects
energy than IBD and thus gives rise to occupied N-derived states in the gap. The DPBII typically forms at the substrate/epitaxial layer interface and could account for the domain boundaries on f1 0 1 0g planes for which no polarity inversion across the boundary is observed [11]. Across the boundary DPB-II would have threefold coordinated Ga and N atoms both in sp2 hybridizations, which gives rise to a deep-acceptor state localized at the lone pair of the sp2 hybridized N atoms. In another theoretical investigation, Elsner et al. [136] utilized local density functional methods to examine a group of domain boundaries on the f11 20g planes, called DB-I, in wurtzite GaN caused by stacking faults in terms of their atomic geometries and energetics. It should be mentioned that relatively larger supercells are required to model domain boundaries terminating in f1 1 2 0g planes. Elsner et al. found that the energetically most favorable configuration is characterized by a displacement of 1/2h1 0 1 1i and has no inversion polarity, which is consistent with high-resolution Z-contrast HR-TEM experiments of Xin et al. [11]. In this model for the aforementioned configuration all atoms at the domain boundary are fourfold coordinated and form strong Ga-N bonds (no wrong bonds), which do not lead to any states deep in the bandgap. However, the same calculations also suggest that electrically active point defects, in particular gallium vacancies, could segregate to the boundary and introduce deep-acceptor states. The prismatic stacking fault, which has a juxtaposed fourfold ring and an eightfold ring with no wrong bonds, is in this category. Unlike the DB-II (also goes by IBD-II) varieties that originate at the substrate/ epitaxial layer interfaces, the DB-I variety could be formed within the epitaxial layer and extend only a short distance. Although a DPB-I has fourfold and eightfold rings with no wrong bonds on the f1 1 2 0g planes, assuming no additional displacement along the [0 0 0 1] direction, it has wrong that is, Ga–Ga (with bond length of approximately 2.7 Å) and N–N (approximately 1.5 Å) bonds in the ð1 0 1 0Þ plane. Due to the very different bond lengths of Ga–Ga and N–N bonds, the wrong bonds give rise to a high energy and thus reduce the stability of the system. This implies that DPB-I should not occur frequently and if it occurs it should exist with different spacing. A spacing of 2.8 Å between the boundary planes minimizes the energy. For the equilibrium distance of 2.8 Å, the structure presents shallow occupied N-derived gap states at approximately 0.2 eV above the VBM and unoccupied gap states at approximately 0.4 eV below the conduction band minimum. For larger distances the influence of the Ga–Ga bonds across the boundary would vanish so that the electrical properties correspond to free f1 1 2 0g surfaces that are electrically inactive. In the foregoing discussion, an assumption was made that there is no vertical displacement along the c-direction involving the DB. Let us now include an additional vertical displacement of 1/2h0 0 0 1i giving a total displacement of 1/2h1 0 1 1i, consistent with that observed by Xin et al. [11]. This additional displacement brings us to the configuration called DPB -I. The calculated domain-wall energy of 99 meV Å2 for DPB -I is significantly lower than the energy of the unshifted DPB-I version, suggesting that DPB -I is a more likely candidate for domain boundaries in f1 1 2 0g planes. No polarity inversion across the boundary for the domain has been observed [11]. DPB -I is thought to be associated with single growth fault in the basal
j 885
886
j 4 Extended and Point Defects, Doping, and Magnetism plane [11,23]. DPB -I starts and ends with a basal plane stacking fault. Although all the bonds are fourfold coordinated with no wrong bonds, some of the bonds are quite distorted. Consequently, DPB -I induces shallow electronic states approximately 0.35 eV above VBM in the bandgap. Reiterating extended defects could trap impurities, point defects, and carriers and also be the source of complexes. These manifestations are either electrically and/or optically active and are discussed in some detail in Section 4.3. Here, a short discussion is given for continuity. The signature of point defects trapped at threading edge dislocations in GaN was considered by Elsner et al. [46] through first principles calculations that indicated that VGa and its complexes with one or more ON have very low formation energies at different positions near the threading edge dislocations. Moreover, the formation energies of VGa, ON, and their complexes at different sites near the edge dislocation are also much lower than those near the defects in bulk material. Energy levels of the defects trapped at dislocations generally shift with respect to the point defects in bulk. The shift, however, is not large [46]. Consequently, the stress field of threading edge dislocations is likely to trap Ga vacancies, oxygen, and their complexes. A variety of the VGa-containing complexes may form acceptor-like defect levels in the lower half of the bandgap and are therefore responsible for some luminescence transitions. The case for p-type GaN may be similar in that the dislocations may trap VN and VN-related complexes. Among the most notorious manifestation of defects as viewed from the optical emission point of view is the yellow luminescence observed in n-type GaN. This is reported to be associated with a deep state [137] that is 0.8–1 eV above the valence band, as discussed in detail in Volume 2, Chapter 5. Several theoretical evaluations have postulated that the deep state responsible for the YL (represents the yellow band emission) could be a complex, gallium-vacancy oxygen, VGa–O [138] (see Section 4.3.2 for details) or nitrogen antisite, NGa [139–141]. Elsner [136] suggested that the stress field in the core of dislocations could act as traps for impurities such as oxygen and intrinsic defects such as VGa, thereby trapping VGa–O complexes. As discussed in Section 4.3.2, evidence appears to suggest that it is most likely that vacancy oxygen (VGa–O) complex in conjunction with extended defects is responsible for the YL emission. It should be kept in mind that deep donors in n-type material and deep acceptors in p-type material usually do not contribute to luminescence [142]. Therefore, dislocations are not likely to show in luminescence experiments, unless they trap point defects owing to large stress fields around dislocations.
4.2 TEM Analysis of High Nitrogen Pressure (HNP) Solution Growth (HNPSG) and HVPE-Grown GaN
Transmission electron microscopy is an indispensable method for not only determining the extended defect density, when it is above 106 cm2, which is often the case in early stages on development, but also shedding much needed light on the nature
4.2 TEM Analysis of High Nitrogen Pressure (HNP) Solution Growth (HNPSG) and HVPE-Grown GaN
and propagation of the extended defects. The HVPE technique is more and more used to produce thick GaN layers followed by the removal of the substrate to produce freestanding GaN for further epitaxy using MBE or OMVPE. In the same vein the GaN templates produced by the HNPSG method represent very high structural quality. It is for these reasons that both the extended and point defect structure in these materials are discussed. A particular attention is also paid to the evolution of defects as the film grows away from the interface in HVPE layers. The two-step HVPE processes employing low-temperature GaN buffer layers have shown good-quality materials with very promising characteristics. Below, extended defects followed by point defect structure of HNPSG templates and HVPE layers (as well as the freestanding templates that have been removed from their substrates by the laser peel off technique) are discussed. Defect discussion is also extended to include epitaxial layers only to the extent that the particular investigations are relevant to issues underlying the extended defects themselves. A case in point is whether the core of a screw dislocation is open or filled, which is discussed in Section 4.1.2. 4.2.1 Extended Defect Characterization
TEM investigations of the extended defect structure in the heteroepitaxial GaN [143] and HNPSG-grown GaN as well as the homoepitaxial layers on them have been undertaken [144–146]. Because the epitaxial layers grown directly on GaN template without methods, such as ELO, tend to reproduce the extended defect nature of the template, investigations by TEM involving epitaxial layers of this kind are lumped together with the templates. Liliental-Weber [144] showed that the N-polar ð0 0 0 1Þ surfaces of the n-type HNPSG GaN crystals, particularly the smaller templates, is often atomically flat with 2–3 ML steps present, and these GaN templates below the surface are practically free of extended defects as far as that can be measured with TEM. We should be reminded that when the extended defect concentration dips much below about 107 cm 2, it is very difficult for the small areas probed by TEM to contain extended defect. If this is the case, other methods, such as defect delineation etches, could be employed, as discussed in Section 4.2.4. On the opposite face of the HNPSG GaN crystals, which is the Ga-terminated face, the surface is rough and the platelet contains a number of extended defects such as stacking faults, dislocation loops, and Ga microprecipitates, extending to 10% of its entire thickness of the template from the surface. It seems that the presence of these defects is related to the growth instabilities often observed on the Ga-polar surfaces of crystals grown without intentional doping, as discussed in Section 3.2.7.1. Detailed TEM studies of Mg-doped crystals [146] have shown that the introduction of Mg induces stacking faults situated on the {0 0 0 1} polar surfaces. Often caused by unstable growth, the defective surface layer must be removed by mechanical chemical polishing and other surface preparation steps prior to epitaxial growth. An issue that is prominent in HNPSG GaN is the observation of void-type defects, that is, nanotubes and pinholes, and whether these have their genesis in dislocations. Refer to Section 4.1.6 for a discussion.
j 887
888
j 4 Extended and Point Defects, Doping, and Magnetism The HNPSG-grown GaN samples doped with Mg have also been investigated in terms of their extended defect structure. Owing to Mg segregation, defects that are unique to these types of samples, such as Mg-rich planar defects under certain growth conditions and the polarity of GaN, have been observed. Liliental-Weber et al. [147,148] and Grzegory et al. [149] reported on the spontaneous ordering in bulk p-GaN grown by high nitrogen pressure solution method, with N-polarity. Selective area diffraction (SAD) patterns indicated that the formation of these regularly spaced planar defects leads to additional diffraction spots dividing the (0 0 0 1) reciprocal distance into 20 equal parts, as shown in Figure 4.43a. These monolayers have a 1=3½1 1 0 0 þ c=2 shift typical for a stacking fault, but the splitting of the (0 0 0 1) and (0 0 0 3) reflections suggest that they also have a characteristic of planar inversion domains. These Mg-rich precipitates are believed to exist in the form of Mg3N2 [150]. The reported thickness of these planar defects is approximately 1 nm and the distance between them is 10.4 nm. This self-formed superlattice structure leads to the intensity of SAD diffraction images along the [0 0 0 1] reciprocal distance being divided into 20 equal parts, as shown in Figure 4.43b. These planar defects have features similar to polytypoids observed in AlN:O or Mg–Si–Al–O ternary or quaternary system [151,152]. Formation of these polytypoids has been attributed to a specific ratio of metal to nonmetal atoms, leading to different lengths of the unit cell along the c-direction. EDX analysis (Figure 4.44) indicates higher Mg and N concentration at the defects as compared to the areas between the defects. This would imply that Mg segregation is responsible for the formation of these defects. It has been reported that Mg together with Al, N, O, and Si can form polytypoids [153]. Therefore, it is possible that defects may also form with Ga instead of Al. It should be noted that the planar defects are not observed in all p-GaN samples grown by high nitrogen pressure solution growth method but only under certain growth conditions. These planar defects are related to Mg segregation in N-polarity surface [154], but more work is clearly warranted to clarify their origin. HVPE GaN layers with varying thicknesses of 1.5, two 5.5, and 55 mm have been used [155] to observe the propagation and annihilation of extended defects in GaN grown by HVPE. HVPE layers investigated were grown on sapphire coated with ZnO in a chloride-transport HVPE vertical reactor [156,157]. For all samples, the crosssectional TEM specimens were prepared in ½2 11 0 and ½1 1 0 0 zone axis orientations. The well-established method CBED was used to determine the polarity of HVPE GaN layers. Because GaN is a noncentrosymmetric crystal the difference in the intensity distribution within (0 0 0 2) and ð0 0 0 2Þ diffraction discs can be attributed to Ga and N distribution within the unit cell. However, this intensity distribution change depends also on sample thickness. Therefore, a comparison of experimental CBED patterns with patterns simulated for the thickness indicated by the pattern in the central disc indicates that these layers grow with Ga polarity (see Figure 4.45) where such comparison is shown for diffraction pattern obtained from an HVPE sample. The density of misfit dislocations at the layer/substrate interface was investigated by HREM. The HREM images were filtered in Fourier space to cause the interfacial dislocations to stand out. An example of an HREM image taken for the HVPE sample
4.2 TEM Analysis of High Nitrogen Pressure (HNP) Solution Growth (HNPSG) and HVPE-Grown GaN
(b)
1
20 19 2
20 Satellite peaks
18
3 17 4 5 7
0
10
247
Figure 4.43 (a) Periodic arrangement of planar defects in the subsurface area from a bulk GaN:Mg crystal grown with N polarity; (b) Diffraction intensity distribution through the pattern of (a) showing division of the [0 0 0 1] reciprocal distance into 20 equal parts. Courtesy of Liliental-Weber et al. [148].
j 889
j 4 Extended and Point Defects, Doping, and Magnetism NK a1
NK a1
Intensity (au)
890
MgK a
OK a1
SiK a1 GaL a1
0
0.5
MgK a
1.0 Energy (keV)
1.5
Figure 4.44 Energy dispersive X-ray (EDX) analysis on the defect area in Mg-doped N-polarity GaN using a 1 nm probe. The inset shows the differences in the intensities for Mg and N peaks when the beam is placed on the defect (shaded) and outside the defect (dotted lines). Courtesy of Liliental-Weber et al. [147].
before and after such filtering is shown in Figure 4.46. This exercise led to a misfit dislocation density at the layer/substrate interface of about 2 · 1013 cm2. Conventional TEM studies indicate that the dominant defects present are the threading dislocations. Bright-field TEM images, recorded under multibeam conditions to image all dislocations with different Burgers vectors, were used to estimate the density of these dislocations with resultant figures to be on the order of about 1010 cm2 near the layer/substrate interface. The density of threading dislocations gradually decreased away from the interface, and for the 55 mm thick layer it reached a value of about 108 cm2 at the surface. Plan-view samples were used to determine the density of threading dislocations at the surface (Figure 4.47). The density of dislocations determined from plan-view samples was in very good agreement with those extrapolated from cross-sectional samples. The threading dislocation densities are plotted as a function of distance from the interface in Figure 4.48. A gradual decrease in density of these dislocations with increase in the distance from the substrate shown on this plot indicates a gradual improvement of layer quality with thickness. The relative numbers of different types of threading dislocations (edge, screw, and mixed) have been investigated with two types of dark-field images – having (0 0 0 2) and ð2 1 1 0Þ type reflections. In the first type of images, only the screw and mixed dislocations are visible, whereas in the second one only the edge and mixed dislocations are observed. The results led to the conclusion that no specific type of
4.2 TEM Analysis of High Nitrogen Pressure (HNP) Solution Growth (HNPSG) and HVPE-Grown GaN
Figure 4.45 Polarity determination for an HVPE layer. Courtesy of J. Jasinski and Z. Liliental-Weber.
threading dislocation dominates and that all three types of threading dislocations (edge, screw, and mixed) are present in HVPE GaN layers in comparable densities. This is somewhat different from MBE in that Ga-polarity layers grown under Ga-rich conditions tend to produce disproportionate concentrations of edge dislocations (109–1010 cm 2) with screw and mixed dislocations in the 107 cm 2 range. However, as the III/V ratio is decreased, the relative number of screw and mixed dislocations increase whereas the edge dislocations decrease. As for N-polarity films, the edge dislocation count decreases even further, but inversion domain boundaries appear, as discussed in Section 3.5.3 in conjunction with growth on SiC substrates and Section 3.5.6 in conjunction with growth on sapphire. A similar treatment was also applied to a freestanding GaN template [158] that was grown by HVPE on sapphire to a thickness of 300 mm and separated from the sapphire substrate by laser induced liftoff [159]. The CBED was applied to determine the polarity of each face.
j 891
892
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.46 High-resolution electron microscopy image of the interface in HVPE LH266l (7,3)G sample (a). An appropriate filtering in the Fourier space enhances visibility of misfit dislocations present at this interface (b). Inset shows clearly one of such dislocation. Courtesy of J. Jasinski and Z. Liliental-Weber.
Figure 4.47 Threading dislocations at the surface of an HVPE layer (D ¼ 1 · 109 cm2). Courtesy of J. Jasinski and Z. LilientalWeber.
TEM study of cross-sectional specimen revealed that the face, which was juxtaposed to the substrate, is of relatively poor quality as expected because it represents the interfacial region between the GaN epitaxial layer and the sapphire substrate. The roughness of this surface is about 0.1 mm. The analysis of CBED patterns obtained on
4.2 TEM Analysis of High Nitrogen Pressure (HNP) Solution Growth (HNPSG) and HVPE-Grown GaN
Density of dislocations (cm–2)
10
LH266lcs LH1089cs LH1106cs LH1089pv LH1106pv
10
10
10
9
8
0
10 20 30 40 50 Distance from the substrate (μm)
60
Figure 4.48 Density of threading dislocations versus distance from the interface in HVPE films. Courtesy of J. Jasinski and Z. Liliental-Weber.
the side previously next to the substrate indicates that it is of ½0 0 0 1 N polarity, which means that a long bond along the c-axis is from N to Ga. The polarity determination by CBED is consistent with chemical etching experiments in which the N-face etched very rapidly in hot phosphoric acid (H3PO4). In addition, Schottky barriers fabricated on this surface exhibited a much reduced Schottky barrier height (0.75 eV versus 1.27 eV on the Ga-face), only after some 30–40 mm of the material was removed by mechanical polishing followed by chemical etching to remove the damage caused by the first mechanical polishing [160]. The density of these dislocations near the N-polarity face as determined from the plan-view sample is estimated to be about 4 1 · 107 cm2. The density of these dislocations determined from cross section was found to be about 3 1 · 107 cm2. For comparison, a density of about 1 · 107 cm2 was obtained by etching the N-face in H3PO4 for 15 s at 160 C followed by counting the etch pits on several AFM images, details of which are discussed in Section 4.2.5. Most of these threading dislocations are of mixed Burgers vector because they are visible in bright-field images with g-vector parallel and perpendicular to the c-axis. However, one needs to be careful with such a conclusion because of the very low statistics (very few dislocations observed within the electron transparent area). TEM studies of a plan-view specimen prepared for the Ga-face side revealed a nearly defect-free surface. The density of dislocations was estimated to be less than 1 · 107 cm2; however, due to the very low statistics, there is a relatively large uncertainty for this estimation. In cross-sectional study one could not find any threading dislocation within the electron transparent area and based on this information one can estimate that density of these dislocations is less than about 0.5 · 106 cm2. The very low defect concentrations on the Ga-face of the sample
j 893
894
j 4 Extended and Point Defects, Doping, and Magnetism necessitate application of another method for a more accurate determination of the dislocation count. To this end, several defect revealing etches, such as hot H3PO4 acid, were employed. Several AFM images, after etching, with large-area scans up to 50 mm · 50 mm, indicated a dislocation count of about 5 · 105 cm2. It is remarkable that the cross-sectional TEM and hot H3PO4 acid methods are in such good agreement. 4.2.2 Pyramidal Defects
Pyramidal defects, or V-shaped defects with f1 0 1 1g polar facets, found in GaN and related heterostructures are associated with some other structural defect such as pinholes (comprising dislocations and inversion domain boundaries) and dislocations. They can also be mitigated by InGaN grown on GaN containing structural defects and be caused by high magnesium doping. This particular section deals with V-shaped defects formed in GaN having pinholes and/or heavy Mg doping. Liliental-Weber et al. [56] investigated V-shaped defects emanating from pinholes in bulk GaN. The facets of the V-shaped defect make 56 angle (depending on the refill, this angle may somewhat change so do the facet planes) with each other as shown in Figure 4.49. As for the V-defects caused by heavy doping with Mg, heavy doping is often tried in an attempt to achieve high hole concentrations. The irony is that doing so or doing so bluntly could lead to V-shaped defects. Attainment of p-type conductivity in GaN catapulted this semiconductor and its alloys to the realm of devices from being a merely a laboratory curiosity. Among all the potential p-type impurities explored, Mg emerged as the only viable candidate despite its high ionization energy (200 meV in GaN and 400 meV in AlN). The conundrum surrounding p-type doping is that hole concentrations in excess of 10 cm3 are needed to reduce the series resistance and the heat dissipation and to increase the efficiency of light emitters. But, increasing the Mg concentration beyond 5 · 1019 cm3 results in self-compensation and drop-off of the hole concentration due to lowering of the Fermi energy that promotes the formation of donor-like defects [161]. The low solubility of Mg is limited by the formation of Mg3N2 precipitates, which usually appear when the Mg doping level is greater than 1020 cm3. Mg segregation [162] and processes associated with it cause many types of defects, the nature of which depends on the growth conditions. Northrup [163] performed first principles pseudopotential density functional calculations for Mg-rich inversion domain boundaries that form on (0 0 0 1) planes in GaN. These boundaries delineate inverted material from the host matrix in the pyramidal inversion domain defects. The model boundary is of the form of GaNMgNGa layers that are stacked in abcab registry, with at least three fourth of the available c sites occupied by Mg atoms. An additional one fourth monolayer of Mg can be incorporated in this layer, provided additional compensating Mg acceptors are located nearby. We discussed this type of defect, some in conjunction with the text associated with Figure 4.13 and others in conjunction with the discussion of V-type IDBs, both in Section 4.1.3 dealing with planar defects. The extended defect nature of these defects
4.2 TEM Analysis of High Nitrogen Pressure (HNP) Solution Growth (HNPSG) and HVPE-Grown GaN
Figure 4.49 A pinhole formed on top of an inversion domain leading to a V-shaped defect. Note that a mixed dislocation approaches the inversion domain and propagates to the sample surface along the pinhole wall (a). The schematic representing the micrograph is shown in (b) for clarity. The height of the inversion domain (I) is about half that of the surrounding layer thickness t3, showing different growth rates for the two polar
directions. Two subgrains [marked in (b) as II and III] grow on the two sides of the inversion domain. Growth proceeds on top of these grains with thickness t2 along the c-axis and t1 perpendicular to the f1011g planes. (c) Atomic model of a GaN crystal in ½1210 projection showing the atomic arrangement along the polar f1011g planes. Note the same angle between these faceted planes and the V edges of the pinhole. Courtesy of Z. Liliental-Weber.
j 895
56°
) 11
01
(1 0
1)
j 4 Extended and Point Defects, Doping, and Magnetism (1
896
(b) Figure 4.49 (Continued )
is investigated by TEM whereas the point defects are generally investigated by methods such as photoluminescence (PL) and deep-level transient spectroscopy (DLTS), which are discussed in this chapter in relation to defects discussed in Sections 4.3 and 4.4. In terms of the structural properties of heavily Mg-doped GaN, some samples have exhibited pyramidal defects [164,165], as shown in Figure 4.50a. With projection along the ½1 0 1 0 direction, a cross-sectional image of a pyramidal defect is of triangle shape, which has an angle of 47 2 between the basal plane and its boundaries (Figure 4.50b). This angle changes to 45 3 when the cross-sectional image is observed along the ½1 1 2 0 direction as displayed in Figure 4.50c. This observation indicates that the pyramidal defect has a set of f1 1 2 3g inclined facets and a hexagonal base with the tip pointing along ½0 0 0 1 of the matrix material that is grown in the [0 0 0 1] direction (Ga polarity). A plan-view image of a pyramidal defect is shown in Figure 4.50d. Although some reports determined these pyramidal defects as being hollow [165], EELS experiments showed that there was no thickness variation between the pyramidal defect and the surrounding matrix [166]. Therefore, it is generally believed that the pyramidal defects are simply Mg-induced inversion domains in the GaN matrix [163,167,168]. Ab initio calculations with regard to the stability of the facets of the pyramidal defects suggest a termination of this surface with Mg and Ga at a ratio of 3 : 1, respectively. Taking into account the 3 : 1 coverage with Mg and Ga, the average Mg concentration at the defect site is around 15–30%, which is some 2 orders of magnitude higher than the Mg concentration in the matrix (0.1%). This enormous accumulation of Mg is a result of either diffusion or segregation that lays the foundation for the formation of this defect. In further investigations, Liliental-Weber et al. [169] determined the atomic structure of the Mg-rich hexagonal pyramids in Mg-doped bulk and also OMVPE thin films of GaN
4.2 TEM Analysis of High Nitrogen Pressure (HNP) Solution Growth (HNPSG) and HVPE-Grown GaN
grown with Ga polarity by direct reconstruction of the scattered electron wave in a transmission electron microscope. Small cavities were also observed to be present inside the defects, confirmed also with positron annihilation (PA) studies. The inside wall surfaces of the cavities were of reverse polarity compared to the matrix. As discussed in Section 3.5.6, Mg segregation is capable of changing the polarity of the GaN epitaxial layers grown by both MBE and MOCVD from Ga polarity to N polarity [168]. While the processes pertinent to growth-related parameter are left to
Figure 4.50 Bright-field cross-sectional TEM image of pyramidal defect induced by Mg (a) [164]; high-resolution TEM cross-sectional image of an Mg-induced pyramidal defect with g ¼ ½10 10. Insert: Projection of the pyramid with a slit tilt along the horizontal direction to show the hexagonal base (b) [164]; high-
resolution TEM cross-sectional image of Mginduce pyramidal defect with g ¼ ½1120 (c). Inset: Projection of the pyramidal defect with a slit tilt along the horizontal direction to show the hexagonal base [164], TEM plan-view of pyramidal defects (d) [165].
j 897
898
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.50 (Continued )
the discussion in Chapter 3, those pertinent to line defect generation and the role of line defects leading to pyramidal defects are covered here. Figure 4.51a shows a crosssectional TEM image of a GaN film grown on SiC by MBE, where the sample surface was exposed to 8 ML Mg during a growth interruption. The polarity change was observed in the subsequent regrown layer with a clear inversion domain boundary (IDB) lying on the (0 0 0 1) plane that divides the sample into two layers with distinct contrast patterns. The detailed features of IBDs in general and their formation are discussed in Section 4.1.3. In the high-resolution TEM image, good crystalline order near the IDB is revealed, as shown in Figure 4.51b. However, in this sample following the change of polarity from Ga polarity to N polarity, a 10-fold increase in the defect density, which is pyramidal stacking faults [54] or screw dislocations, can be observed. Figure 4.52a shows the polarity change (from [0 0 0 1] to ½0 0 0 1) of p-GaN on the undoped-GaN template. The polarity flip was confirmed by convergent beam electron diffraction patterns. The lattice image (shown in Figure 4.52b) of IDB taken along ½1 1 0 0 shows facets that form along the (0 0 0 1) plane at an angle approximately 50
4.2 TEM Analysis of High Nitrogen Pressure (HNP) Solution Growth (HNPSG) and HVPE-Grown GaN
Figure 4.51 Bright-field TEM image in (0 0 0 2) two-beam condition of MBE-grown GaN film on SiC, showing inversion domain boundary labeled with solid arrowheads and growth interruptions labeled with hollow arrowheads (a); high-resolution image of the same sample shown in (a), with brackets indicating the region of the inversion boundary. Courtesy of Feenstra and coworkers [168].
j 899
900
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.52 Convergent beam electron diffraction patterns taken above and below the interface marked with the arrow in a GaN:Mg layer, (a) [167]; Lattice image of the inversion domain boundary taken along the ½1100 axis showing facets that form along the (0 0 0 1) plane with a angle of approximately 50 (b) [167].
to the basic plane. This is consistent with the polarity change by the exposure of the (0 0 0 1) Ga surface to several MLs of Mg. Total energies of different inverted structure were calculated by Ramachandran et al. [168]. These calculations were performed within the local density approximation using first principles pseudopotentials [170]. When the Mg concentration exceeds a certain value, it is energetically favorable to form an Mg-terminated ð0 0 0 1Þ top surface (IDB) than an Mg-terminated (0 0 0 1) surface. The stable phases such as Mg3N2 could also form. A possible converted structure is shown in Figure 4.53a, while the unconverted structure is shown in Figure 4.53b just as a comparison. In the converted structure, a plane of Ga–Ga bonds replaces a normal plane of Ga–N bonds, which results in the outermost Mg layer being terminated by N atoms instead of Ga atoms. This inverted structure is more stable (by 0.5 eV/1 1) than the uninverted and stoichiometrically identical structure. In the subsequent growth, the inverted structure may act as a template for N-polar GaN. The N plane bonded to the upper Ga plane of IDB has two possible stacking orders (zinc blende or wurtzite). A change
4.2 TEM Analysis of High Nitrogen Pressure (HNP) Solution Growth (HNPSG) and HVPE-Grown GaN
(a)
(b)
j 901
Mg
IDB N Ga Figure 4.53 Structural model of a c-plane inversion domain boundary induced by Mg atoms (a); noninverted structure (b). Both models are shown in a ð1 1 2 0Þ projection. Height of the Mg layer above the underlying atomic plane is 1.29 and 2.43 Å for inverted and noninverted structures, respectively. Courtesy of Feenstra and coworker [168].
between these two stacking orders would result in pyramidal stacking faults, as shown in Figure 4.51a. Although most pyramidal inversion domains are identified as N-polarity domains in Ga-polarity matrix, which is supported by the total energy calculations [168], Vennegues et al. [164] observed the opposite that is the inversion domains with Ga polarity being induced by Mg doping in an N polarity matrix. The p-GaN layer under question was grown by OMVPE with Mg doping level around 4 · 1019 cm3. The origin of this Ga-polarity domain still needs to be elucidated. It has also been observed that the shape of pyramidal defects having inclined f1 1 23g facets is very dependent on the polarity of film in Mg-doped GaN layers. Further, TEM plan-view images, depicted in Figure 4.54a and b, confirm that the tips of both the N-polarity inversion domains in the Ga-matrix and Ga-polarity inversion domains in the N-matrix point in the ½0001 direction of the matrix material. In the TEM dark-field images, the Ga-polarity inversion domains and N-polarity inversion domains show distinct contrast. Figge et al. [171] observed the periodicity of magnesium segregation and the formation of pyramidal defects in p-GaN layers along the [0 0 0 1] growth direction. This experimental observation was noted both in p-GaN and p-type AlGaN grown by OMVPE. A defect-free region at the onset of Mg source and following defect density modulation is shown in Figure 4.55a. In the analysis of this defect density modulation, the authors stated that most Mg atoms do not incorporate in the bulk and occupy the Ga site. Instead, they segregate and cover the p-GaN surface at a certain thickness. Because the inversion domains result from Mg accumulation, either by diffusion or by segregation, after the formation of inversion domains Mg is depleted in the vicinity of the domain, and during the subsequent growth a certain layer thickness is required for Mg to reaccumulate. This model then accounts for the origin of the defect density modulations shown in Figure 4.55a. It has also been found that the thickness of the defect-free region and the period of the defect density modulations are strongly dependent on the Mg/group III source molar ratio, as depicted in Figure 4.55b. With lowering of the Mg/group III source ratio, the thickness of the
902
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.54 Dark-field images in multiple beam conditions along the noncentrosymmetric ½1 0 1 0 zone axis. (a) N-polarity defects in Ga-polarity matrix; (b) Ga-polarity defects in N-polarity matrix [164].
4.2 TEM Analysis of High Nitrogen Pressure (HNP) Solution Growth (HNPSG) and HVPE-Grown GaN
Figure 4.54 (Continued)
defect-free region increases. When the aforementioned ratio is less than 0.02, no defect-rich layer could be observed regardless of the p-GaN layer thickness. The effect of Mg doping on the structure has been extended to include Al0.13Ga0.87N layers, specifically the effect of Mg source (Cp2Mg flow) flow rate on the formation of pyramidal or V defects and inversion domain boundaries [172]. Those authors employed high-resolution X-ray diffraction and TEM methods to determine the dislocation density as well as the imaging of the structural defects. The results of HRXRD and TEM were reported to be in good agreement in that the total density of threading dislocations decreased from 4 · 109 cm2 in undoped films to 1.3 · 109 cm2 (in moderately Mg-doped samples with Cp2Mg flow 103 mmol min1). However, it must be pointed out that the reduction is not very significant and may therefore not represent a true trend. The authors also investigated the microstructure of the pyramidal defects with the overall results being that while IBDs were absent in undoped or Si-doped Al0.13Ga0.87N films, they appeared even for Cp2Mg flow rates of 103 mmol min1. The IBDs were observed to originate in the 560 C low temperature GaN nucleation layer and propagate through the entire Mg-doped Al0.13Ga0.87N layer. Surprisingly, IBDs were not observed for Cp2Mg flow rated beyond 0.397 mmol min1. The width of IDBs became broad as the thickness of AlGaN increased, and the faceted step on the surface was on the top region of the inversion domain. In the h1 1 2 0i projection, the facet angle is nearly 52 with respect to the basal plane, which corresponds to the f2 0 23g planes. The facets on the inversion domain were
j 903
904
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.55 The dependence of the density of the pyramidal defects on the sample thickness as obtained from the TEM image shown in the inset. The solid line is a least squares fit, assuming a broadened d-like distribution of the defects (a); Defect-free thickness versus the molar Mg/group III precursor ratio (b) [171].
reported to result from the relatively slow growth rate within the inversion domain region compared to matrix around it, as have been reported for the GaN case [173]. Similarly, as in the case of GaN, the formation of vertical-type IDBs in the low Mg source flow rate cases can be attributed to an inversion in the film polarity in
4.2 TEM Analysis of High Nitrogen Pressure (HNP) Solution Growth (HNPSG) and HVPE-Grown GaN
microscopic Mg-rich regions. For the higher Cp2Mg flow rates in excess of 0.397 mmol min 1, nearly horizontal IDBs were observed in Al0.13Ga0.87N layers. High-magnification TEM images of horizontal-type IDBs, not shown, show multifaceted boundaries and no stacking faults around the IDBs. The facet angle ranged from 45 to 50 with respect to the basal plane, consistent with the GaN layers grown by MBE on Ga-polarity (0 0 0 1) templates. 4.2.3 V-Shaped Defects (Pits) in InGaN Multiple Quantum Wells (MQWs)
The V-shaped defects have also been observed in InGaN in the platform of InGaN/ GaN multiple quantum wells, for example, by Cho et al. [174]. In this particular case, the general belief is that InGaN growth does not initiate at the tip of a threading dislocation that propagated to the surface of the underlying GaN layer. With continual epitaxial growth, inclined facets form. The cross-sectional shape of the void or the region with no growth is of the V shape. This is therefore different from the inverted V defect caused by locally high Mg concentration. However, Cho et al. [174], utilizing cross-sectional transmission electron microscopy, discovered that not all V defects commence on the tip of a threading dislocation, as shown in Figure 3.196. By increasing the indium composition in the InxGa1xN well layer or decreasing the TD density of the thick underlying GaN layer, the authors were able to generate many V-defects from the stacking mismatch boundaries induced by stacking faults that are formed within the MQW due to the strain relaxation. Moreover, TD density in the thick GaN underlying layer was found to affect not only the origin of V defect (pit) formation but also the critical indium composition in the InxGa1xN well on the formation of V defects (pits). With the aid of TEM investigations, Cho et al. [174] observed that the observed V defects have stacking faults on the (0 0 0 1) planes where they nucleate. In specific terms, the stacking order of ABABABAB along the c-axis in normal GaN is transformed to ABABCBCB in the area where the V-shaped pit area commences area, the TEM cross section of which is shown in Figure 4.56a. Consequently, a stacking mismatch boundary is generated at the boundary of the two regions with the ABABCBCB and ABABABAB stacking order. Figure 4.56b depicts an atomic model that illustrates how the SMB can be generated when the faulted area (ABABCBCB) on the left-hand side meets the area on the right with the correct stacking order (ABABABAB) and the V defect can commence where the SMB is. 4.2.4 Structural Defect Analysis by Chemical Etch Delineation
Wet chemical etching is a rapid and inexpensive method for surface and near-surface defect investigation. Hot phosphoric acid (H3PO4), mixed H3PO4/H2SO4 solution and molten potassium hydroxide (KOH) have been shown to etch pits at defect sites on the c-plane of GaN [27,175–179,181]. An AFM image of the as-grown GaN surface revealed few point defects (pits) positioned at surface step terminations (Figure 4.57).
j 905
906
j 4 Extended and Point Defects, Doping, and Magnetism These pits have been previously reported to correspond to the surface termination of pure screw or mixed dislocations [182,183]. Armed with these TEM generated data, AFM can be used to estimate structural defect density on atomically smooth surfaces with a high degree of ordering. Volume 2, Figure 1.61 illustrates AFM image of the etched surface morphology produced by the PEC process after 60 min of etching following the procedure outlined in Volume 2, Chapter 1. The height of the whisker-like features was estimated to be about 700 nm and the lateral size on the order of 100 nm. The density is approximately 2 · 109 cm2 and according to the TEM analysis this value is quite close to the effective density of dislocations. To clarify further the relation between EPD and dislocation density in GaN and look for any consistency among the various chemical etches, H3PO4 and molten KOH were used as defect etchants in GaN, which produce hexagonal-shaped etch pits. By varying the time and temperature, one can optimize the etching process to produce a pitted surface that clearly reveals the size and density of the pits. AFM images of an HVPE GaN sample etched in molten KOH for 2 min at 210 C lead to a pit density of about 1 · 109 cm2. The pits are of hexagonal shape and their size ranges from 40 to 100 nm in diameter and from 10 to 30 nm in depth. Most etch
Figure 4.56 (a) Magnified HR-TEM image of the area where the V-defect commences. The stacking order of ABABABAB on the left-hand side of the stacking mismatch boundary (SMB) in GaN is transformed to ABABCBCB in GaN on the right side of SMB, which is mark of stacking faults. Recall that a stacking fault would
generate a stacking mismatch boundary in the subsequent growth. (b) An atomic model showing how a SMB and V defect could be generated when the faulted area of the left-hand side (ABABCBCB) meets the area on the righthand side with correct stacking order of ABABABAB [174].
4.2 TEM Analysis of High Nitrogen Pressure (HNP) Solution Growth (HNPSG) and HVPE-Grown GaN
[0 0 0 1]
N Ga or In
[1 1 0 0] Stacking Mismatch Boundary
SF
A B C A B C A B C A B C
A B CA B C A B CA B C
(b) Figure 4.56 (Continued )
pits terminated at the surface steps. The pit density is consistent with high concentrations of pure screw or mixed screw edge dislocations found in HVPE GaN samples by TEM study. The surface morphology of another sample from the same GaN wafer etched in H3PO4 for 6 min at 160 C lead to a pit density of 1 · 109 cm2, similar to the KOH etch. The size of the etch pits ranges from 25 to 70 nm in diameter and from 8 to 20 nm in depth. Similar etching experiments have been carried out on the Ga- and N-faces of freestanding GaN H3PO4 at 160 C, with this in mind, until the defects on the surface were clearly revealed. The N face was etched in H3PO4 for 15 s at 160 C, and the
j 907
908
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.57 AFM image (2 mm · 2 mm) of an as-grown GaN. Some defects (pits) positioned at surface step terminations are visible. The average step height is of 0.8 nm and the root mean square roughness 0.4 nm. The vertical scale ranges from 0 to 10 nm.
etched surface is shown in the AFM image of Figure 4.58a. By counting the etch pits on several images, one ascertains that the density on the N face is about 1 · 107 cm2. Also visible on the surface are terraces that have formed, many of which terminate at dislocation sites. The density figure is comparable to 3 1 · 107 and 4 1 · 107 cm2 obtained by cross-sectional and plan-view TEM analysis. Similarly, the Ga-face, which is more impervious to chemical etches, was etched for 50 min in H3PO4 at 160 C and an AFM image of the resultant surface is depicted in was Figure 4.58b. Still visible on the surface are the scratch lines that resulted from the mechanical polishing process, indicating that the nondefective c-plane GaN has not been significantly etched by hot H3PO4 acid. The inset of the figure is a zoom into the region indicated by the white box, and shows that the black dot barely visible in the larger image is in fact a small hexagonal etch pit that has formed on the surface. Three distinct sizes of etch pits were found. These pits, termed (for simplicity) small, medium, and large, can be seen in Figure 4.58b. The total density of etch pits on the Ga-face is 4.3 · 105 cm2, a value that is more than 1 order of magnitude lower than that found on the N face. Because TEM allows observation of only a very small area of the sample, some other method such as defect delineation etches [181], discussed in Section 4.2.2, must be employed to determine the defect density, which is the case with HNPSG templates. Applying molten KOH–NaOH eutectics, Weyher et al. [184] delineated
4.2 TEM Analysis of High Nitrogen Pressure (HNP) Solution Growth (HNPSG) and HVPE-Grown GaN
Figure 4.58 (a) AFM image of the N face of the substrate after etching in H3PO4 for 15 s at 160 C. The surface has been significantly etched at defect sites; so much that valleys, ridges, and terraces have been formed on the surface. The deep etch pits on the surface marked (a), (b), and (c) are 2.1, 1.2, and 1.5 mm wide by 6, 4.7, and 8 nm deep, respectively. (b) AFM image of the Ga face etched for 50 min at 160 C in H3PO4. The lines from the etching are still visible on the surface, indicating that the
nondefective GaN has not been significantly etched. The point defect sites on the surface have been etched by the acid. There are three discreet sizes of defects found on the surface. The large defects are roughly 1.5 mm wide, the medium sized defects (upper left corner of image) are generally 800 nm wide, and the small defects are approximately 200 nm wide. A small defect can be seen in the inset, which is a zoom into the boxed region.
j 909
910
j 4 Extended and Point Defects, Doping, and Magnetism dislocations in both GaN heteroepitaxial layers and GaN templates grown by the HNPSG method. The method revealed a very small etch pit density (10–100 cm2) for templates supporting the notion that the large portions of the template is void of extended defect revealed by delineation etching. 4.2.5 Structural Defect Observations with Surface Probes
Nanoscale scanning probe techniques especially electric force microscopy (EFM), conductive atomic force microscopy, and scanning capacitance microscopy (SCM) has been very useful in characterizing semiconductor materials and devices including GaN materials and devices. EFM, a schematic diagram for which is shown in Figure 4.59, is usually done in two modes: (1) surface potential or Kelvin probe mode and (2) dynamic mode. In surface potential mode of EFM, a DC bias on the tip is varied so that when it equals the potential at the conductive sample surface, the force felt by the tip is zero. Electrostatic force between the tip and sample can be written as F ¼ ((1/2)(dC/dz)V2, where C is the tip-to-sample capacitance and V is the sum of (1) contact potential difference between tip and sample, (2) applied DC and AC bias to the tip, (3) any existing/applied potential on the sample surface, that is, V ¼ Vcp þ VDC þ VAC sin ot þ Vsample [185]. Using this expression for V in the equation of force F we find that force F has components at DC, o, and 2o. Fo ¼ dC/dz(Vcp þ VDC þ Vsample)VAC sin ot. When VDC, the DC potential applied to the tip, equals (Vcp þ Vsample), Fo is zero. This is utilized in determining Vsample if Vcp is known. Dynamic mode EFM senses electric field gradient on the sample by detecting change in the resonant frequency of the cantilever caused by the field gradient. Inversion domains are a major problem in epitaxially grown GaN layers. The surface contact potential electric force microscopy (SCP-EFM) has been used to identify Ga-polar and N-polar regions in MBE-grown GaN films. Because of the
Photodetector
Laser diode Potential feedback
Cantilever Tip
VDC+VAC sin ω t Voltage source
Sample Sample holder Ground Figure 4.59 Schematic of an electric force microscopy setup.
4.2 TEM Analysis of High Nitrogen Pressure (HNP) Solution Growth (HNPSG) and HVPE-Grown GaN
Figure 4.60 Simultaneous tapping mode atomic force microscopy and electric force microscopy image of GaN film. Bright spots in the left image are high points and darker regions in the right image are lower potential regions. By correlating with lower potential in the right image, it is clear that the lower high point in the left image is an inversion domain [186].
different atomic arrangements in these two regions surface potential in them are different. SCP-EFM shows about 55 mV difference in surface potential. Figure 4.60 shows lower potential N-polar regions as dark spots on a Ga-polar substrate [186]. Epitaxially grown GaN contains a large density of threading dislocations. As these dislocations disturb atomic arrangements around them and may contain electronic states in the bandgap, surface potential at and around dislocations are likely to be different than the rest of the crystal surface. Hot H3PO4 cleaned GaN surface shows about 30 mV lower potential at domain boundaries and pits in the domains indicating they are negatively charged [187]. However, pure screw dislocations do not show any variation in potential. This could be because either the electronic states of screw dislocations are not within the bandgap or they are located close to the conduction band. As adsorbed atoms and molecules can change surface electronic properties, surface potential variation can be a good indication of nature of surface contamination. On bare GaN surface electron affinity has been found to be 4.1 eV, about 0.6 eV greater than that of the clean surface. This indicates presence of oxygen at the surface, as intentionally chemisorbed O2 on GaN surface shows similar change in surface potential. Surface potential on AlGaN of AlGaN/GaN heterostructure is found to decrease by 0.6 eV, as the thickness of AlGaN layer increases from 5 to 44 nm. This variation has been attributed to the presence of charged acceptors in the AlGaN
j 911
912
j 4 Extended and Point Defects, Doping, and Magnetism layer [188]. Charging and discharging of surface states can affect device performance significantly. Current collapse phenomena in AlGaN/GaN modulation-doped field effect transistors (MODFETs) have been attributed to such phenomena. Electrons are assumed to be able to reach the surface states by tunneling from the gate. This changes surface potential near the gate, which causes drain current to vary. EFM measurements performed on an operating MODFET confirms that charging of the surface states is related to current collapse, as discussed in Volume 3, Chapter 3. In conductive atomic force microscopy (CAFM) a conductive tip with DC bias is scanned over the sample surface in contact mode. The resulting current flow is amplified to get a simultaneous morphology and current conduction map of the sample surface, as shown in Figure 4.61. Conductive properties of defects as well as other phenomena related to leakage currents and conductivity of surface states have been studied with CAFM [189]. Reverse leakage current of Schottky and p–n diodes causes severe degradation in the performance of GaN-based devices. CAFM measurements of Figure 4.62 indicate that Schottky diodes show reverse leakage current to be localized at some spots that are likely to be associated with defects. As reverse leakage current in films grown under Ga-rich condition is higher, Ga influences electrical activity of the defects. Comparing reverse leakage current in films with different screw dislocation densities but similar overall defect density, it has been deduced that pure screw dislocations are responsible for the leakage current [189–191], as shown in Figure 4.62.
(a)
Tip
Ag paint
n-GaN Sapphire
VDC
Metalplate
(b)
Ag/n-GaN
R
n-GaN/tip
VDC Figure 4.61 Schematic diagrams showing (a) experimental setup and (b) equivalent circuit for n-type GaN at negative bias with respect to tip. The term R represents the bulk resistance of the GaN film.
4.2 TEM Analysis of High Nitrogen Pressure (HNP) Solution Growth (HNPSG) and HVPE-Grown GaN
Figure 4.62 AFM and CAFM images a heterojunction FET sample grown by MBE on an HVPE template. (a) 5 · 5 mm2 image of topography (Dz ¼ 25 nm) and (b and c) simultaneous CAFM current images for reverse and forward biases (Dz ¼ 200 pA), respectively. (d) 2 · 2 mm2 topography and (e, f) simultaneous CAFM for reverse and forward bias (Dz ¼ 10 pA), respectively. White/black circles indicate current leakage centers.
Interestingly, taking advantage of conductive property of the defects, CAFM tip has been used to form a thin oxide layer on the defects. Schottky diodes fabricated on these modified substrates show 2 orders of magnitude lower reverse leakage current [192]. In other reports, both the forward and reverse bias I–V characteristics have been imaged [193]. The AFM topography and CAFM current images for a heterojunction FET sample grown by MBE on an HVPE template are shown in Figure 4.62, where the surface consists of pyramidal hillocks that are 150–500-nm wide and have a density of 5 · 108 cm2. The CAFM images for reverse and forward
j 913
914
j 4 Extended and Point Defects, Doping, and Magnetism bias conditions are shown below the topography images. Enhanced current conduction occurs in both forward and reverse biases at the centers of the pyramidal hillocks, which are likely associated with screw dislocations. Approximately 30% of these features conduct at the maximum reverse bias of 12 V. The number of features demonstrating conduction increases with increasing bias. Note that the larger circles superimposed in Figure 4.62-2b and c indicates hillocks that conduct at both biases, where the extent of such conduction is significantly larger in forward bias. The smaller circles indicate conduction centers in reverse bias that are not prevalent in forward bias. Therefore, only a fraction of the hillocks appear to be amphoteric. Using CAFM, localized I–V spectra were taken from regions located both on and off the hillocks, as seen in Figure 4.63a. Curve 1 on a hillock shows detectable leakage in reverse bias at 9 V, whereas region 2 off a hillock shows no measurable leakage. The forward turn-on voltage is also seen to shift down to 3 V on a hillock. Note that the relatively high turn-on voltages observed for CAFM indicate a significant voltage drop outside the tip–semiconductor junction. The observed shifts in forward and reverse bias for current conduction on the hillock indicate the possibility of charge trapping effects. Subsequent I–V spectra at the same location show a significant decrease in current after three to five scans, also indicating such charging effects. Fitting of the observed forward current data to various current conduction mechanisms, as shown in Figure 4.63b, indicates that current conduction is consistent with Frenkel–Poole (FP) in the defective hillock region and field emission (FE) current away from the defect. I–V data obtained from macroscopic (standard) Schottky contacts indicate that thermionic field emission (TFE) is the primary mechanism at higher current levels in forward bias. In another CAFM investigation, Pomarico et al. [194] performed surface current mapping of HVPE and films before and after defect revealing etch. Two different types of samples, in which the facet planes were present either on the perimeters of as-grown islands or on the edges of etch pits created by postgrowth chemical etching, were studied. The results show that crystallographic planes tilted with respect to the c-plane growth direction show a significantly higher conductivity than surrounding areas. The n-type (or p-type) samples required a negative (or positive) sample bias for current conduction, consistent with the formation of a Schottky barrier between the metallized AFM tip and sample. The time dependence of this enhanced conductivity was also studied, which when detailed could shed valuable light on the dynamics of the particular defect involved. The CAFM images obtained by Pomarico et al. [194] from both the etched and as-grown samples suggest that the off-axis planes situated at such locations are more electrically active than c-plane GaN, as shown in Figure 4.64. It has been reported that the N face of c-plane GaN has a lower Schottky barrier height than Ga face [195,196]. If the same were true for the a, r, and m planes of GaN, then increased current conduction could be expected on those surfaces. The enhanced conduction on the etch pit walls could also be related to their origin. Screw dislocations were found to be the dominant defect in the case of the Si-doped HVPE sample. Investigation of Hsu et al. [189] have shown that such dislocations may provide channels for large and stable current. In contrast, the increased conduction of the island sidewalls on the
4.2 TEM Analysis of High Nitrogen Pressure (HNP) Solution Growth (HNPSG) and HVPE-Grown GaN
Current (pA)
100
Sample A
50 0
2 1
–50 –100
–10
(a)
–5
0 Voltage (V)
5
10
10
–11
10 –12 10 –13 3.6
(b)
1
2
Expt FE
Expt FE FP
4.0
~
Current (A)
10 –10
6.8 7.2 Voltage (V)
7.6
Figure 4.63 (a) Local I–V spectra taken using CAFM on sample A for curve 1 on a hillock and curve 2 off a hillock. The spectra were acquired using a current amplifier that limits the maximum current to 100 pA, and with negative voltages corresponding to reverse bias conditions. (b) Fitting of current conduction mechanisms to the forward bias data of curves 1 and 2, where FE ¼ field emission and FP ¼ Frenkel–Poole.
as-grown sample studied by Pomarico et al. [194] is most likely not associated with dislocations. This could explain the reduced current seen for the pyramidal planes after multiple scans, because bulk charge is not available via dislocations to replenish the surface charge. A substantial spread in current among the etched n-type and Zn-doped samples was also observed. This may be primarily due to higher bulk resistivity of the Zn-doped sample as compared to the n-type one. However, differences in such properties as the surface potential barrier height, adsorbed impurities, and the surface charges may also affect the current. Lateral epitaxial overgrowth is very promising as it can reduce threading dislocation density by several orders of magnitude. In this mode of growth, a thin GaN seed layer is covered with a mask (20 mm wide SiO at 5 mm spacing is an example) that is used in subsequent overgrowth. Scanning probes are especially suitable for characterizing such materials as growth condition varies at sub micrometer scale. Crosssectional CAFM and EFM reveal a high carrier concentration in the region right above the SiO2 mask but a Fermi level deep in the bandgap. This has been attributed to impurities from the SiO2 mask that causes compensation and impurity band
j 915
916
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.64 (a) AFM surface morphology and (b) CAFM current images of an etched, n-type, Si-doped HVPE GaN sample. The sample-to-tip voltage is 1.0 V. The vertical scales are (a) 60 nm and (b) 10 nA, and both image sizes are 1 mm · 1 mm, as indicated. (c) Dual cross section of one of the etch pits with the topography and current signals superimposed. An increased current signal is observed on the edges of the pit.
transport. In the upper regions of growth, stripes with different electrical properties indicate growth dynamics dependent impurity and defect incorporation [197]. Also in samples grown on sapphire at the GaN/sapphire, the interfacial region shows larger carrier concentration but a deeper Fermi level [198]. From these observations it can be deduced that GaN strongly absorbs impurities from the substrate. In scanning capacitance microscopy tip sample dC/dV is determined in contact mode, as shown in Figure 4.65. If the semiconductor does not have a dielectric layer on it, the conductive tip needs to be covered with a dielectric material. The tip sample system forms part of a capacitance detection system. Incremental change in capacitance resulting from incremental change in applied voltage is detected with a lock in technique to minimize noise. SCM performed by varying DC bias can be used to characterize charge potential and mobile carriers in a semiconductor in three dimensions. These charged dislocations are assumed to be the major carrier scattering mechanism in GaN. Using this technique, the locally charged region of doped AlxGa1xN/GaN MODFET epitaxial layer structures and surrounding uncharged regions can be mapped, as shown in Figure 4.66 [199]. SCM measurements on GaN films indicate presence of negative charge around threading dislocations [117].
4.3 Point Defects and Autodoping
Conductive tip and cantilever
Capacitance sensor VAC
Depletion layer
j 917
Lock-in amplifier
Insulation
VDC
C(fixed) dC/dV
Doped semiconductor
VAC
C (VDC)
VDC
Ohmic metal contact Figure 4.65 Schematic of a scanning capacitance microscopy setup.
Capture and emission of carriers by traps associated with threading dislocations affect performance of devices in a major way. A DC bias of 6 V was applied via the SCM tip on the surface of AlGaN/GaN heterostructure. Subsequent SCM scans show some regions, as can be seen in Figure 4.62, to be charged [199]. These regions are thought to contain dislocations. SCM performed on a MODFETunder RF stress can reveal whether the defect states are being charged or not. Bias varied SCM study on AlGaN/GaN heterostructure shows variation of threshold voltage that has been traced to variation of thickness of AlGaN layer and presence of charged dislocations [200].
4.3 Point Defects and Autodoping
Point defects are the most common defects occurring in semiconductors [201,202]. Point defects, also known as native defects or intrinsic defects, unless they are neutral, manifest themselves as background doping or autodoping, compensated dopants, and complicate attempts to dope the semiconductor to control its conductivity. Even though GaN broke the long-standing paradigm of large defect concentrations precluding acceptable device performance, and made it to the market place with LEDs, lasers, and detectors with amplifiers well on their way. However, as it should be, point defects have taken the center stage as they exacerbate efforts to increase efficiency of emitters, increase laser operation lifetime, and eliminate anomalies observed in electronic devices. As in all semiconductors, these defects play an important role in the electrical and optical activity in nitride semiconductors as well. A case in point is the carrier lifetime, which is often dependent on the type and density of these defects. Consequently, they play a pivotal role in the radiative quantum efficiency and the longevity of GaN-based lasers and light-emitting diodes. The point defects also cause charged centers that scatter carriers, degrading the electron mobility and affecting FET performance and characteristics. Associated trapping causes degradation of diffusion lengths and, thus, electrical devices relying on minority carrier transport.
918
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.66 Scanning capacitance microscopy images obtained at 0 V DC sample bias of a locally charged region of doped AlxGa1xN/GaN MODFET epitaxial layer structure and surrounding uncharged regions, obtained (a) immediately (b) 6 min, and (c) 1 h after charging with a 6 V DC bias at the tip. (d) Line scans extracted from the images (a)–(c) at the locations indicated by arrows. Courtesy of Yu and coworkers [199].
4.3 Point Defects and Autodoping
There are three basic types of native point defects: vacancies, self-interstitials, and antisites. Vacancies may be interpreted as the lattice sites missing their atoms. Self-interstitials are the additional atoms in between the lattice sites. Antisites, which are unique to compound semiconductors, are the cations sitting on anion sites and vice versa. Native defects result when bonds in a semiconductor are either broken or distorted; they often give rise to deep levels within the forbidden gap. The Fermi level determines the charge state of a particular point defect. Point defects can be either donor type, acceptor type, or amphoteric. Wide bandgap semiconductors exhibit self-compensation caused by defects. For example, when Si donors are introduced into GaN, the lattice may attempt to create Ga vacancies, VGa, which are acceptors, to reduce the total energy. This is one path through which point defects, that is, vacancies, interstitials, and/or antisites, could be formed. Another way is through defective or incomplete kinetic processes on the growing surface of an epitaxial layer. For example, insufficient N flux at the growing surface could result in N vacancies, VN. Another driving force for the creation of point defects may be the polarization field that are always present in nitrides. Even in the absence of strain (which leads to piezoelectric fields) spontaneous polarization is present, which can give rise to significant fields. Ultimately, these fields are screened in macroscopic samples because of free carriers, but one may wonder whether on microscopic length scales the fields could provide a driving force for defect creation. Not all imaginable point defects are energetically favorable in a given semiconductor. To gain some insight, formation energies of defects are calculated and their likelihood therefore is determined. Point defects in GaN have been tentatively identified, including VN, VGa, and Gai. A wide range of analysis tools have been brought to bear to investigate the point defects and their impact on physical properties of GaN. Among the techniques employed are photo luminescence, deep-level transient spectroscopy, minority carrier lifetime measurements, positron annihilation, electron paramagnetic resonance (EPR), and optically detected magnetic resonance (ODMR). The results obtained using the aforementioned methods as well as their analyses will be discussed in this chapter. 4.3.1 Theoretical Studies of Point Defects in GaN
In early theoretical studies, Jenkins et al. [203,204] calculated the energies of the N and Ga vacancies and antisite defects in GaN by taking advantage of a tight-binding approach. For the calculations, the wave functions of strongly localized defects, like nitrogen vacancies VN, were built up from contributions of the whole Brillouin zone. The calculations indicated small variations with respect to the alloy composition. Those authors suggested that VN would produce an s-like level containing two electrons below the conduction band edge and a p-like level containing one electron above the conduction band edge. Because the p-like level is resonant and its electron is autoionized, it decays to the conduction band edge, and dopes n-type GaN (one electron per vacancy). Gallium vacancy (VGa) was predicted to have a p-like level very
j 919
920
j 4 Extended and Point Defects, Doping, and Magnetism close to the valence band and an s-like level within the valence band. Both types of antisite defects, Ga on a N site, GaN, and N on a Ga site, NGa, lie deep within the energy bandgap. The calculations lend support to the assertion that the nitrogen vacancy is most likely a single donor and responsible for the n-type behavior in undoped GaN. Furthermore, if the nitrogen vacancy were a simple donor with its s-like deep level in the gap just below the conduction band edge, it could explain the 0.2 eV feature in the Tansley–Foley optical absorption data [205]. It should be kept in mind that tightbinding calculations include only the first-neighbor interactions, and likely result in erroneous energy levels. A more accurate description of the energy levels was developed by Neugebauer and Van de Walle [206] and others and will be reviewed below. Using a supercell approach with 32 atoms per cell, a plan-wave basis set with an energy cutoff of 60 Ry, and soft Troullier Martins pseudopotentials, Neugebauer and Van de Walle [207] investigated defect-formation energies and electronic structures for native defects in wurtzite and cubic GaN. It was predicted that 3d electrons are important both for the formation energy and the atomic relaxation. Breaking the additional bonds between the Ga 3d orbitals and the N orbitals to create a vacancy costs more energy. This leads to a significant increase in the formation energy. Taking the Ga 3d electrons as core electrons results in a large relaxation. The introduction of 3d electrons causes the system to be stiffer. As a result, the outward displacement of the surrounding nitrogen atoms is reduced to 0.1 Å and the energy gain to 0.26 eV. The 3d electrons thus prevent the GaN bond length from becoming too short. Essentially, all antisites and self-interstitials are high in energy and hence less likely to occur in reasonable concentrations, and then only as compensating defects [211]. The tight-binding calculations by Jenkins and Dow [203,204] not only had limited accuracy, but unlike the more advanced first principles calculations, gave no information about atomic relaxation and defect formation energies. The latter is very important because some defects are thermodynamically unstable although they have levels in the gap. Low formation energy implies a high equilibrium concentration of the defect; a high formation energy means that defects are unlikely to form. The formation energy can be calculated in the formalism of Zhang and Northrup [208] as X E f ðq; E F Þ ¼ E tot ðqÞ E tot ni mi þ qE F ; ð4:4Þ bulk i
where Etot(q) is the total energy of a supercell containing one defect in the charge state q, E tot bulk is the total energy of the defect-free supercell, ni is the number of atoms of type i in this supercell, mi is the chemical potential of atoms of type i, and EF (the Fermi level) is the electron chemical potential with respect to the valence band maximum. The chemical potentials for Ga and N are usually considered for extreme growth or annealing conditions in that mGa ¼ mGa(bulk) for the Ga-rich case and mN ¼ mN2 for the N-rich case. Note that these values are not independent: mGa þ mN ¼ mGaN. The formation energy of the charged defects depends on the charge and the Fermi level at time of the defect formation (during growth or annealing). As can be seen from
4.3 Point Defects and Autodoping
Equation 4.4, the formation energy of the positively (negatively) charged defects increases (decreases) while the Fermi level moves from the valence band maximum toward the conduction band. The slope of this variation is proportional to the charge state of the defect. The energy at which the levels corresponding to different charge states intersect determines the ionization level of the defect. The total energies in Equation 4.4 are calculated from the first principles, based on self-consistent DFT, usually within the local density approximation (LDA) and the pseudopotential plan-wave method. For a succinct discussion of these ab initio calculations as they pertain to band structure, refer to Chapter 3. Supercells usually contain from 32 up to 96 atoms for the wurtzite GaN cell (earlier calculations used 32 per cell [207]). The Ga 3d electrons are explicitly included in the valence orbitals, or, alternatively, approximately treated using the nonlinear core correction (nlcc). The pseudopotentials are created using different approaches. The appropriateness of the selected pseudopotentials is tested by calculation of the bulk properties of GaN. Plan waves with kinetic energies up to 25–150 Ry are usually included in the expansion of the wave functions. The atomic relaxations associated with the defect are determined by force calculations. The calculations of the total energy may in general give the error of up to a few tenths of electron volts, and different corrections are often used to improve the accuracy. The above approach enables one to calculate not only the formation energies and positions of the defect levels but also binding and dissociation energies of complex defects, local modes of vibrations, migration barriers, or diffusivity of point defects. Finally, it can be applied to surfaces, threading dislocations, and other structural defects. Agreement between different groups, using slightly different approaches, convinces one in the reliability of the results. After one finds the formation energy of a particular defect, the equilibrium concentration of this defect can be estimated as ! Sf Ef c ¼ N sites exp ; ð4:5Þ k kT where Nsites is the concentration of sites in the lattice where the particular defect can be incorporated (Nsites 4.4 · 1022 cm3 for the substitutional defects in GaN) and Sf is the formation entropy, which is about 6k [209,210]. Equation 4.5 assumes thermodynamic equilibrium, which may correspond to the growth temperature or to the temperature at which the concentration of the particular defect becomes fixed. When the defects or their constituents are impurities, Equation 4.5 gives the upper limit of their possible concentrations in an assumption that the impurities are abundantly available during the growth. The growth is nonequilibrium process; however, at high enough growth or annealing temperatures, the conditions may be approximately equilibrium. Equation 4.5 shows that Ef is the key parameter in estimating the defect formation: defects with high formation energy have less probability to be formed. Equation 4.5 shows that defects with high formation energy are formed with low probability and their concentrations are low. The chemical potentials are chosen corresponding to Ga-rich conditions and to maximum
j 921
j 4 Extended and Point Defects, Doping, and Magnetism incorporation of the various impurities, with solubilities determined by equilibrium with Ga2O3, Si3N4, and Mg2N3. Below, the main results of the first principles calculations by Van de Walle and coauthors [211–219] will be presented, although calculations by Boguslawski et al. [220], Mattila et al. [139,140], Gorczyca et al. [221,222], and Elsner et al. [41,46] also give similar results. Figure 4.67 shows the formation energies for all native point defects in GaN as a function of the Fermi level. The slope of each line represents a charge of the defect. For each charge state only the segment giving the overall lowest energy is shown. Thus, the change in slope of the lines represents a change in the charge state of the defect, and the corresponding Fermi energy represents the energy level of the defect that can be measured experimentally. It is clear from Figure 4.67 that self-interstitial and antisite defects have very high formation energies and are thus unlikely to occur in GaN during growth. However, electron irradiation or ion implantation can create such defects in large numbers, and identification of a complex involving Gai has been reported [223]. A divacancy (VGaVN) has relatively high formation energy in GaN and as such is also unlikely to be formed in large concentrations [140]. Nitrogen antisite defect (NGa) is expected to be metastable, with large spontaneous Jahn–Teller (JT) displacement in the [1 1 1] direction, in both cubic and wurtzite GaN [139]. In this treatment, the formation energy of nitrogen vacancies is high in n-type GaN, a topic that will be revisited later on, and low in p-type GaN. Therefore, nitrogen vacancies
Ga i
10
8 Formation energy (eV)
922
N Ga
Ni
6
4
Ga N
2
VN
V Ga
0 0
1
2 Fermi level (eV)
3
Figure 4.67 Formation energies as a function of Fermi level for native point defects in GaN. Ga-rich conditions are assumed. The zero of Fermi level corresponds to the top of the valence band. Only segments corresponding to the lowest-energy charge states are shown. Courtesy of Limpijumnong and Van de Walle [219].
4.3 Point Defects and Autodoping
are unlikely to be formed in n-type material, and can be abundant in p-type GaN. Hence, VN cannot be responsible for the n-type conductivity, as was believed for a long time. Nevertheless, VN can be created by electron irradiation or ion implantation and may increase the electron concentration in such material. The shallow donors with activation energy of 64 10 meV have been introduced by electron irradiation in GaN and attributed to VN [224]. Low formation of the gallium vacancy (VGa) in n-type GaN (when the Fermi level is close to the conduction band) should result in the formation of VGa in large amount. However, complexes of the VGa with the shallow donors should also be taken into account because their formation energy can be even lower as will be shown below. Figure 4.68 shows the formation energy of gallium and nitrogen vacancies in GaN, as well as main shallow donors (SiGa and ON) and acceptor (MgGa), and a complex of ON with VGa. It is clear that SiGa, ON, and VN, appearing with slope þ1, are single donors. Formation of these donors is preferential in p-type GaN, although in n-type
10 VGa
Energy (eV)
8
- 20 -
2-3-
6 0-
V
- 2-
G a –V N
V
Ga –
4
O
N
2
VN
Si Ga
0
ON Ga rich -2 0
0.5
1.0
1.5 2.0 Fermi energy (eV)
2.5
3.0
Figure 4.68 Calculated formation energies and ionization levels for native defects and complexes in GaN in the Ga-rich case. The dashed lines correspond to isolated point defects while the solid lines to defect complexes, respectively [140]. Note the similarity between calculated formation energies for VN and VGa in Refs [140,219].
j 923
924
j 4 Extended and Point Defects, Doping, and Magnetism the formation energies of oxygen and silicon are also low. Gallium vacancies, having quite low formation energies in n-type GaN, may be effective compensating centers for the shallow donors in n-type GaN. The calculated 2/3 level of VGa is about 1.1 eV above the valence band. Its /2 and 0/1 levels are also inside the gap [138,140,222,225]. Evidently, formation of VGa in p-type GaN can be ignored due to very high formation energy. As a mobile and triply negatively charged acceptor, V3Ga can easily form complexes with shallow donors in GaN during the growth or cooling down. Figure 4.68 shows that formation energy for the (VGaON)2 complex is even lower than that for the isolated VGa. Energy level of the (VGaON) /2 is close to 2 =3 the level of VGa [46,139,211]; therefore, both defects are good candidates for a deep acceptor responsible for the yellow luminescence in GaN. By using Equation 4.5, Neugebauer and Van de Walle estimated concentrations of the shallow donors and compensating deep acceptors in n-type GaN grown in Ga-rich conditions (Figure 4.69) [138]. It appears that the Si donor has the highest concentration, whereas the VGaON complex can be the dominant compensating acceptor. The defects are formed much easily at high temperatures (Figure 4.69), although when the system is cooled down after growth, the defect concentrations are expected to remain unchanged. Note the similarity between calculated formation energies for VN and VGa in Refs [140,219]. The case of shallow donors is discussed in Section 4.6.1. In terms of acceptors, Mg acceptor has low formation energy and can easily incorporate into GaN. It prefers Ga site and its shallow acceptor level provides good p-type conductivity. At high Mg flux, Mg2N3 can be formed as the solubility-limiting phase [214]. VN is expected to autocompensate MgGa, but if hydrogen is present in the growth, compensation by hydrogen will dominate (Figure 4.70), and formation of VN will be suppressed [214]. In p-type GaN, the formation energy of VN is significantly lowered, making it a likely compensating center during acceptor doping. The þ/3þ energy level of VGa is estimated at about 0.5 eV above the valence band. The þ/3þ transition of VGa is characterized by a large lattice relaxation [206]. The 2þ charge state is always higher in energy than either the 1þ or the 3þ states, and thus thermodynamically unstable (the so-called negative-U defect). The details regarding acceptor incorporation are provided in Section 4.6.1. 4.3.1.1 Hydrogen and Impurity Trapping at Extended Defects Monatomic interstitial hydrogen may exist in two charge states in GaN: Hþ and H, whereas the H0 state is unstable (Figure 4.71), exhibiting a very large negative-U effect [215]. Hþ prefers the nitrogen antibonding site, whereas for H the Ga antibonding site is the most energetically stable. In a similar vein, Van de Walle calculated the formation energy of interstitial H in GaN as a function of the Fermi level [226]. Hþ is expected to be mobile even at room temperature due to a small migration barrier (about 0.7 eV), while H has a very limited mobility in GaN due to a very large migration barrier (about 3.4 eV) [215]. It follows from Figure 4.70 that the solubility of H is considerably higher under p-type conditions (where it exists as Hþ) than under n-type conditions (H). The value of the negative-U effect (transition of H into Hþ) is extremely high (2.4 eV), larger than in any other semiconductor.
4.3 Point Defects and Autodoping
5 4 VGa –O N
VGa
3 2 1
ON Si Ga
0 –1 1.0 (a)
1.5
2.5
2.0
3.0
Fermi energy (eV)
log 10 concentration
(ccm)
25
VGa –O N VGa
20
ON
Si Ga
15
V Ga
10 0 (b)
500
1000
1500
Temperature (K)
Figure 4.69 Formation energy versus Fermi level for the main native defects and impurities in n-type GaN (a). The corresponding equilibrium concentrations versus temperature are shown in (b). Courtesy of Neugebauer and Van de Walle [138].
Hydrogen can form complexes with other defects in GaN, and often the formation energies of the hydrogenated defects are lower. In p-type GaN, the formation of the (VNH)2þ complex becomes very favorable (Figure 4.70). Note that these complexes may incorporate only during growth: after growth, both H and VN are donors and would repel each other. The energy level of the VNH donor is very close to the conduction band; either it is actually a resonance or slightly below the conduction band minimum [216].
j 925
j 4 Extended and Point Defects, Doping, and Magnetism 2–
VGa
8
(VGa – ON)– 6
Formation energy(eV)
926
(VGa– ON)2– 3–
VGa
+
4
VN 0
Mg Ga
2
H+
3+
VN
VNH2+ + ON
0
+
SiGa
– Mg Ga
–2 0
2 1 Fermi energy (eV)
3
Figure 4.70 Formation energy as a function of the Fermi level for ON, SiGa, MgGa, Hi, VN, VGa, and a complexes of VN and VGa with H and O, respectively. Courtesy of Van de Walle and Neugebauer [214].
In n-type GaN, up to four hydrogen atoms can be bound to gallium vacancy to form the complexes (VGaH)2, (VGaH2), (VGaH3)0, and (VGaH4)þ. The first three complexes are donors, whereas the last one is a single donor. The formation energies of the VGaHn complexes are shown in Figure 4.72. It turns out that the hydrogenated vacancies have lower formation energies than the isolated gallium vacancy, except for the (VGaH4)þ complex. The calculated energy levels of the hydrogenated gallium vacancies (VGaH)2 and (VGaH2) are close to that of the isolated VGa, and thus the (VGaH)2 and (VGaH2) complexes can also contribute to the yellow luminescence in GaN. Local density functional methods were also used to examine the behavior of not only isolated point defects, such as ON and VGa, but also these defects trapped at threading edge dislocations in GaN [41,46]. Elsner et al. [41] explained why the threading screw dislocations (with Burgers vector [0 0 0c]) exist as open-core dislocations, whereas the threading edge dislocations (with Burgers vector a½1120/3) have filled cores. Both threading screw and threading edge dislocations are electrically inactive in wurtzite GaN. However, due to large stress fields near
4.3 Point Defects and Autodoping
e(+/-)
e(0/-)
e(+/0)
Formation energy (eV)
5 4
H0
3 2
H–
1 H+
0 0.0
0.5
1.0
1.5 2.0 2.5 Fermi energy (eV)
3.0
3.5
Figure 4.71 Calculated formation energy of interstitial hydrogen in wurtzite GaN as a function of Fermi level. EF ¼ 0 corresponds to the valence band maximum, and formation energies are referenced to the energy of an H2 molecule. Courtesy of Van de Walle [215,226].
4
Formation energy (eV)
2
H-
(VGaH3)0 +
H
1
) 2-
3) a
3
Ga H 2)
G
Ga H
(V
(V
(V
+ H 4) (V Ga +
Si
0 1.5
2.0
2.5
3.0
Fermi energy (eV) Figure 4.72 Calculated formation energies of hydrogenated Ga vacancies in GaN as a function of Fermi energy. The formation energies of the isolated vacancy ðV3Ga Þ, of interstitial H þ and H , and of the Si donor are also included. Courtesy of Van de Walle [216].
dislocations, impurities and native defects can be trapped at it. In particular, VGa and its complexes with one or more ON have very low formation energies at different positions near the threading edge dislocation (Figure 4.73). Table 4.2 summarizes the calculated formation energies for these defects at different positions.
j 927
928
j 4 Extended and Point Defects, Doping, and Magnetism [1 0 1 0]
[1 2 1 0] 0
1 3
2 4
4*
Figure 4.73 View along the [0 0 0 1] dislocation line of the 312-atom wurtzite supercell (not all area shown) containing a dipole of threading edge dislocations with Burgers vector b ¼ ½1 1 2 0. White (black) circles represent Ga (N) atoms. Position 0 corresponds to a bulk-like
region. Position 1 is at the dislocation core that, similar to the GaN (10–10) surface, contains threefold coordinated Ga (N) atoms in sp2 (p3) hybridizations. Positions 2 and 3 are situated in the stress field near the dislocation core. Courtesy of Elsner and coworkers [46].
Table 4.2 Formation energies of VGa, ON, and VGanON complexes
in bulk and at different sites at the threading edge dislocation (see Figure 4.73). Position
EðV3Ga Þ
EðONþ Þ
E(VGaON)2
E(VGa2ON)
E(VGa3ON)0
Bulk cell Position 0 (bulk like) Position 1 (core) Position 2 Position 3
1.8 1.7 0.2 0.3 0.3
1.7 1.5 0.2 1.3 1.0
1.1 1.0 2.3 0.6 1.0
0.7 0.9 2.5 0.3 1.0
0.8 0.7 3.0 0.3 0.8
Thus, the stress field of threading edge dislocations is likely to trap gallium vacancies, oxygen, and their complexes. The variety of these defects may form defect levels in the lower half of the bandgap and therefore be responsible for the yellow luminescence in GaN [46]. 4.3.1.2 Vacancies, Antisites, Interstitials, and Complexes Defects, in general, and point defects, in particular, have been synonymous with GaN research that began with the initial reports of an epitaxial growth. Characteristically, the early films suffered from large n-type backgrounds. Maruska and Tietjen [227]
4.3 Point Defects and Autodoping
argued that autodoping is due to native defects, probably nitrogen vacancies, because the impurity concentration was, at least, 2 orders of magnitude lower than the electron concentration in their samples. Consequently, unintentional n-type doping in GaN is probably due to nitrogen vacancies became the crutch statement in nitriderelated reports, because there was no unequivocal experiment to prove or disprove it. It should be pointed out that the uncertainty in the experimental determination of impurity concentration was large, and experimental results could be interpreted differently. Perlin et al. [228] and Wetzel et al. [229] observed carrier localization and freeze-out of free electrons at high hydrostatic pressures. In line with existing view of the time, these results were first attributed to a nitrogen vacancy whose neutral localized level at ambient pressure was predicted to be 0.40 0.10 eV above the conduction band edge. Unfortunately, the view still prevails in many experimental reports where increased donor concentration due to various processes are attributed to N-vacancies even though its formation energy in n-type material is high (see Section 4.3.1.2.1). Later, on the basis of their first-principles calculations, Neugebauer and Van de Walle [138,211,212] suggested that in n-type material the thermodynamically stable formation of nitrogen vacancies in appreciable quantities is highly improbable. Instead, they pointed out that contaminants such as silicon or oxygen may be responsible for the large electron concentrations observed in some unintentionally doped samples (see the review by Neugebauer and Van de Walle) [230]. Later on, the characteristic freeze-out of free electrons in GaN above 20 GPa was attributed to oxygen donor that undergoes transition to the strongly localized state at high hydrostatic pressures [231]. To make matters yet more interesting, refined simulations through mainly the use of large supercells indicate that as well [232]. 4.3.1.2.1 Vacancies As in the case of a well-established III–V semiconductor GaAs [233], vacancies in III–V GaN are multiply charged defects causing several defect, levels to appear in the energy gap. Unlike the more established III–Vs, however, the exact nature of defects and their formation details in relation to the Fermi level position are still somewhat controversial. The energy position in the gap of native defects in GaN has been calculated employing employed 96-atom supercells [219], as illustrated in Figure 4.74. In this picture, gallium vacancy (VGa) is the dominant native defect in n-type GaN, whereas the nitrogen vacancy (VN) may abundantly form in p-type GaN. Evidently, N-rich conditions favor formation of VGa, whereas Ga-rich case may facilitate VGa formation. However, as discussed later in this section, other calculations and experimental seem to suggest that even in n-type GaN, N vacancies are viable and should be considered. Gallium Vacancy Gallium vacancy has relatively low formation energy in n-type GaN when the Fermi level is close to the bottom of the conduction band, which is the case for n-type GaN. Therefore, VGa may act as a compensating center due to the sharp increase on the formation energy, deeper in the bandgap (1.5 eV at CBB to 3.2 eV as the Fermi level falls down to 2.5 eV [140]. The 2/3, /2, and 0/1 transition levels of VGa are estimated at 1.10, 0.64, 0.25 eV, respectively (Figure 4.74) [219]. Slightly larger values of the ionization energies (about 1.5, 1.0, and 0.5 eV, respectively)
j 929
930
j 4 Extended and Point Defects, Doping, and Magnetism 1-/0 1+/0 2+/1+ 3+/2+
1-/0 3+/1+ 1-/0 1+/0 1+/0
3-/24+/3+ 2-/1-
2+/1+ 3+/2+
2+/1+
3+/1+
1-/0 VGa
VN
GaN
NGa
Gai
Ni
Figure 4.74 Transition levels for native defects in GaN, determined from formation energies displayed in Figure 4.67.
have also been reported [139,220]. The energies of point defects in zinc blende and wurtzite GaN have been reported, albeit rather old with GaN standards, to be very similar [140], and thus in the view of this particular report gallium vacancies would prevail in both polytypes of as-grown GaN. Continuing on along these lines, the 3 formation energies of VGa were found to be consistently low when the Fermi level is close to the conduction band. In n-type GaN the Ga vacancy is fully filled with electrons, and capture of a photogenerated hole in photoluminescence experiments may lead to radiative transition of an electron from the conduction band or from a shallow donor level to the 3 /2 level of VGa. The calculated migration barrier 3 for VGa is relatively low, 1.9 eV [219]. Therefore, the Ga vacancies are mobile in a wide range of temperatures that overlap temperatures used for growth or thermal annealing. It is therefore likely that Ga vacancies migrate and form pairs and complexes with other mobile, such as N vacancies, and more stable defects. In contrast to N-vacancies, Ga-vacancies do not interact among themselves and thus do not form bound pairs. Nitrogen Vacancy Early calculations predicted the energy levels of the nitrogen vacancy close to or inside the conduction band [204,211,220]. In fact, the highly n-type conductivity consistently observed in undoped GaN was attributed to the positively charged vacancy in the nitrogen sublattice ðVNþ Þ throughout the early investigations of GaN [203]. However, the first principles calculations revealed that the formation energy Ef, of VNþ in n-type material is close to 3 eV, and hence the thermal equilibrium concentration of VNþ is too small to provide an observable number of carriers [139,140,211]. Therefore, based on these particular calculations, the nitrogen vacancies are expected to play a role in the defect microstructure kinetics only in p-doped material [230]. Further, VN might form in minute (barely detectable) concentrations in n-type GaN only under Ga-rich conditions [211,220]. According to these calculations, the electron from the resonant 0/ state would autoionize to the bottom of the conduction band, where it would form an effective-mass state
4.3 Point Defects and Autodoping
bound by the Coulomb tail of the vacancy potential [211,220], allowing VN to act as a donor with a minor role in n-type material owing to its inconsequential concentration. There is only one transition level for VN in the gap: the 3 þ /1 þ state at about 0.5 0.2 eV above the valence band maximum (Figure 4.74) [219,234]. The V2Nþ charge state was reported to be unstable, and transition from VNþ to V3Nþ charge state causes a large lattice relaxation [219,234,235]. The migration barriers for the VNþ and V3Nþ charge states are estimated to be 2.6 and 4.3 eV, respectively [219]. Similar to the case of VGa (migration energy for the V3Ga state is 1.9 eV), the relatively low migration barriers (at least for V3Nþ ) would result in the formation of pairs and larger complexes between VN and VGa as well as other defects during hightemperature growth or annealing, especially in p-type GaN where the V3Nþ state may dominate. Nitrogen vacancies have quite a large binding energy that can pave the way to the formation of small N-vacancy clusters. It should be mentioned that nitrogen vacancies, including those that are negatively charged and their complexes (e.g., mixed Ga-N divacancies (see Section 4.3.1.2.1) have been theoretically shown to be critically important in the defect kinetics of GaN, which in the words of Ganchenkova and Nieminen [232] was unexpected. Their conclusion is based on the improved and extensive ab initio calculations of the electronic structure and energy parameters of vacancies, their clusters, and vacancy-impurity complexes in GaN and its alloys [232]. The effective energy for the vacancy-mediated self-diffusion for V3N and V2N changes from 4.28 to 2.48 eV and 4.85 to 3.61 eV, respectively, as the Fermi level is changed from 2.6 (nearly intrinsic) to 3.2 eV (n-type). These figures are lower than that for the V3Ga defect over the same Fermi level span that leads to the conclusion that the VN accumulation rate in the bulk of the film due to diffusion from the growing surface is higher than that for VGa. This conclusion is arguably supported by N loss from the surface during high-temperature growth experiments, which implies high concentrations of VN at the surface and relatively low formation energy. This assertion is inconsistent with the popular framework that VNþ has the lowest energy over the entire bandgap because the formation energy of VNþ is nearly 3 eV (corresponding to a concentration of approximately 108–109 cm 3 at 900 C) and diffusion activation energy is as high as 7 eV. In contrast, negatively charged VN has a relatively smaller activation energy at 3.53 0.75 eV (corresponding to a concentration of approximately 1019 cm 3 at 900 C) for the Fermi level in the range 2.6–3.1 eV above the valence band maximum. This concentration is sufficiently high to promote thermal decomposition, particularly at growth temperatures employed. This conclusion is globally consistent with improved epitaxial quality obtained at very high temperatures when accompanied with high over pressures of nitrogen. One therefore wonders if the high concentrations of both VN and VGa present in n-type GaN at high temperatures could promote mixed VGa–VN divacancies. The calculations by Ganchenkova and Nieminen [232] indicate that for Fermi levels above 1.5 eV, (VGa–VN)3 is the most stable divacancy, whose formation energy is lower than that for VGa and has substantial binding energy of approximately 2.34 eV. In short the calculations of Ganchenkova and Nieminen [232] lead to the conclusion that the N vacancy can be a dominant defect in GaN for Fermi levels over the entire range of the bandgap, not just for p-type material.
j 931
932
j 4 Extended and Point Defects, Doping, and Magnetism The above conclusion is reached on the basis of the improved computational methodology employed by Ganchenkova and Nieminen [232], over the earlier reports. As typically the case, the calculations were based on the DFT approach within the PAW–LDA approximation implemented in the Vienna ab initio simulation package (VASP) code [236–239] (for additional information on the methods mentioned, see Section 2.1.5) and the use of very large supercells. Defect energy characteristics were calculated using 216- or 300-atomic simulation cells (SCs), though smaller cell sizes (32 and 64 atoms for zinc blende) were used to compare with the earlier published data. The Brillouin zone sampling was done with the 4 · 4 · 4, 2 · 2 · 2 Monkhorst–Pack k-point set and G-point sampling, depending on the considered SC (see Ref. [232] for further details). Any possible error due to the finite supercell size in the formation energy calculations for charged defects was addressed through an analysis of the total energy convergence with the supercell size. In each case only a weak dependence of the nitrogen vacancy configuration energy Econf, that is, the difference between the total energies for supercell with the defect and without it, on the supercell size was noted. This is counter to the case of Ga vacancy, the genesis for which may lie in the differences in the relaxation patterns. Therefore, in the case of the neutral and negatively charged nitrogen vacancies, one notes inward relaxation, whereas for gallium vacancies the relaxation is always outward. This, together with the features of the electron density distribution in a defect, causes the Ga-vacancy configuration energy to be more sensitive to the supercell size. The above mentioned observation regarding N vacancies was reported to be the case for all the charge states ranging from þ1 to 3, with the largest configuration energy change of 0.04 eV for the triply negative state when the supercell size was changed from 64 to 300 atoms. This provided the basis to conclude that both the Madelung-type correction (proportional to q2/L, where q is the charge and L is the supercell size) and the valence band alignment correction for this defect are unimportant (a detailed description of the corrections can be found in Ref. [230]). Although the increased supercell size did not lead to discernable change in the þ prevailing picture of charge states of the Ga vacancy ranging from VGa to V3Ga , this is 3þ not so for the charge states of the N vacancy ranging from VN to V3N . The most notable conclusion is that the nitrogen vacancy is a negative-U defect that exists, depending on the Fermi level position, either in the singly positive or in singly to triply negative charge states. To reiterate, contradicting the earlier conclusions VN (in appropriate charge states) was found to be energetically more favorable than the gallium vacancy over the entire bandgap and not just in p-type material [232]. To shed some light on this obvious contradiction with earlier calculations [230], a set of calculations for VN in the same range of charge states using smaller supercells (from 32 to 216 atoms) in both zinc blende and wurtzite was undertaken. Although the earlier results for positively charged nitrogen vacancies are qualitatively well reproduced even in smaller supercells (no matter how the d-electrons are accounted for and which polytype is considered), the VN was found to be more stable in the upper part of the bandgap than the positive charge states. The calculated ionization levels, based on total energy differences, lie above the DFT single-particle gap calculated for the bulk. Ganchenkova and Nieminen [232]
4.3 Point Defects and Autodoping
have carefully analyzed the distribution of the defect-induced single-particle states and concluded that they are localized in the defect region and not resonant with the conduction band (Ec). In the case of VN the a1 (s-like) state goes below the valence band edge, whereas the t2 (p-like) states lie inside the calculated Kohn–Sham (KS) gap. When VN is negatively charged, extra electrons continually fill the t2 states, which remain below Ec. The filling of the defect levels is accompanied with changes in the relaxation and bonding patterns. Specifically, when the charge state changes from VNþ to V3N , the relaxation changes from outward (for VNþ ) to inward (for all other charge states). With each electron the inward relaxation increases while the distance between the nearest neighbors gets smaller (from 3.17 Å for VNþ to 2.83 Å for V3N as compared to the equilibrium value of 3.15 Å). Moreover, the Td symmetry of VNþ changes to the D2d symmetry for all negatively charged vacancies. This reduction in symmetry is the result of a Jahn–Teller distortion, which points to a bonding trend between the nearest neighbors of the vacancy. In turn, the bonding can be only due to the localization of the added electrons. Indeed, the electron density distribution pattern of the highest occupied electron states clearly shows the contribution of localized electrons to those of the newly formed bonds. Examining the radial distribution of the partial charge densities (with respect to the vacancy center), Ganchenkova and Nieminen [232] observe very pronounced localization of the electrons occupying the highest KS levels within the sphere of the radius of approximately second nearest neighbor distance. If a rigid scissor correction is applied to the bandgap and defect-derived (t2) level states corresponding to VN, the ionization levels (þ/), (/¼) and (¼/3) move upward by 15% to lie between 2.8 and 3.0 eV while remaining below the bandgap, whereas the shift of levels associated with VGa is nearly 30%. This results from the differences in the calculated conduction band bottom positions in the supercells containing VN or VGa. Thus, the scissor corrections change the absolute values of the ionization levels but leave VN to be more favorable than VGa. Experimental evidence for or against the observation of VN is not well settled. For example, extended X-ray absorption fine structure (EXAFS) measurements on both cubic and hexagonal GaN samples grown at 600 C by electron-cyclotron-resonance molecular-beam epitaxy without intentional doping (and hence, most probably n-type) have been explicitly interpreted in terms of the presence of nitrogen vacancies [240]. In contrast, no evidence of negatively charged VN in n-type GaN have been reported in positron annihilation spectroscopy (PAS) investigations that readily detect VGa (see Section 4.6 for details of PAS investigations in GaN) [241]. According to the calculations of Ganchenkova and Nieminen [232] in a heavily n-type doped GaN crystal, where the difference between the formation energies of gallium and nitrogen vacancies is the smallest (0.98 eV), one would expect the ratio of thermal equilibri4 6 th um concentrations of C th VN =C VGa to be approximately 10 –10 in a temperature range of 800–1000 C. The lack of experimental observations in PAS has been attributed to VN being indistinguishable from the bulk due to comparable lifetimes. The calculated positron lifetimes for the nitrogen vacancy are 131 ps for VN , and 132 ps for V2N and V3N , respectively. The measured positron annihilation lifetimes in the bulk lie in the range 160–163 ps compared to the calculated value of 130 ps. In addition,
j 933
934
j 4 Extended and Point Defects, Doping, and Magnetism experimental verification would require the sample to be heated to a temperature where the vacancy concentration would reach its equilibrium value. It should also be noted that the rate of vacancy accumulation is proportional to vacancy currents, which depend not on the vacancy formation energies alone but on the sums of formation and migration energies [232]. The relative ease of N loss from the lattice has been used as indirect evidence for the presence of abundant nitrogen vacancies in n-type GaN. Divacancy A divacancy consisting of Ga and N vacancy (VGaVN) has relatively high formation energy in GaN and is unlikely to form in large numbers (Figure 4.68). The divacancy is expected to lead to at least two deep levels in the energy gap of GaN, and behave as a single acceptor in n-type and as a double donor in p-type GaN [140]. However, taking the calculations of Ganchenkova and Nieminen [232] in heavily ntype GaN at their face value, the high concentrations of both gallium and nitrogen vacancies reached at high temperatures, raising the question whether these mixed Ga-N divacancies (VGa–VN) would appear in n-type crystals. Earlier calculations of Mattila and Nieminen [140] indicated that the binding of two vacancies into a mixed divacancy is energetically favorable over the entire bandgap. On the contrary, the formation energies of mixed divacancies in charge states down to 2 were found to be prohibitively high for their thermal formation. However, Ganchenkova and Nieminen [232] found that for Fermi levels above 1.5 eV from the top of the valence band, the most stable charge state becomes (VGa–VN)3. This defect not only has a substantial binding energy (2.34 eV for dissociation into V0N þ V3Ga ), but its formation energy also turns out to be lower than that of gallium vacancies. 4.3.1.2.2 Interstitials and Antisite Defects Formation of interstitial and antisite defects is often considered to have low probability in GaN due to the small lattice constant of GaN and the large size mismatch between Ga and N atoms [161]; however, under certain circumstances some of these defects may form in small concentrations. Gallium Interstitial Due to the large size of the Ga atom and very large lattice relaxation a Ga interstitial (Gai) would induce, the formation energy of Gai is large. Only the octahedral site is stable for Gai [219]. Although in n-type GaN or in GaN grown under N-rich conditions the formation of Gai is improbable in thermodynamic equilibrium, it may form under electron irradiation in GaN or in p-type growth conditions (Figure 4.67). Formation energies as a function of Fermi level for native point defects are shown in GaN under Ga-rich conditions. The zero of Fermi level corresponds to the top of the valence band. Only segments corresponding to the lowest energy charge states are shown [219]. Similar to VN, Gai is a donor with the resonance 1þ/0 state in the conduction band and a deep 3þ/1þ state (Figure 4.74). Transition levels for native defects in GaN are determined from formation energies displayed in Figure 4.67) [219] and 220. The 3þ/1þ energy level is predicted at about 2.5 eV above the valence band [219]. A metastable behavior is possible in the 3þ state [219]. The migration barrier for Gai is estimated as 0.9 eV [219] in agreement with the recent experimental results of Chow et al. [242] who have discovered mobile
4.3 Point Defects and Autodoping
Gai in irradiated GaN at temperatures below room temperature. Note that the optically detected electron paramagnetic resonance (ODEPR) experiments [242] apparently detected the 2þ charge state of Gai, which was predicted to be unstable [219]. A plausible accounting for this apparent discrepancy is the metastable 2þ state that might be activated by optical excitation [219]. High mobility of Gai even at room temperatures implies that Gai is trapped by some other defects and does not exist in GaN as an isolated defect in equilibrium conditions. Nitrogen Interstitial As in the case of Ga interstitial, the N interstitial has high formation energy, especially in Ga-rich conditions (Figure 4.67). However, once formed, the nitrogen interstitial (Ni) forms an N–N bond [211,220,219,235]. Up to four stable levels corresponding to different charge states of Ni can be formed in the energy gap (Figure 4.74) [219]. The N–N bond distance monotonically decreases with increasing charge of Ni, and for the case of case of N3i þ , it approaches the bond distance in N2 molecule [211,219]. Note that the two N atoms share one N site apparently equally. The highest stable level (1 /0) is expected at about 2 eV above the valence band maximum [219]. Therefore, in n-type GaN, Ni will act as a simple acceptor. However, the formation energy of this defect is too high in GaN grown under equilibrium Ga-rich conditions. The migration barrier for Ni is about 1.5 eV for the 1 and 3 þ charge states, [219] so that diffusion of nitrogen interstitials is likely to occur in GaN at temperatures slightly above room temperature. Gallium Antisite Gallium antisite, meaning Ga on the N site, (GaN) introduces a few deep levels in GaN (Figure 4.74) [219,220]. The 4þ/3þ level of GaN is expected at about 0.9 eV above the valence band [219]. Therefore, in p-type GaN this native defect may essentially cause compensation (each GaN donor can compensate four MgGa acceptors!) if the Ga-rich conditions are employed. A large outward lattice relaxation around GaN has been noted by different theoretical investigations [220,221]. Nitrogen Antisite Nitrogen antisite (NGa) apparently introduces three [219] or even four [139] deep levels in the energy gap of GaN (Figure 4.74). It can also behave as compensating double donor in p-type GaN or acceptor in n-type GaN. The formation energy of NGa is very high for any position of the Fermi level, especially under Ga-rich conditions (Figure 4.67), although Mattila et al. [139,140] predicted a reasonably low formation energy of N3Ga in strongly n-type GaN grown under N-rich conditions. Therefore, this defect may be of interest from an academic viewpoint [139,140,220,235]. The neutral NGa defect may exhibit a metastable behavior, similar to its analogue in GaAs, known as the EL2 defect [243,244]. Mattila et al. [139,140] and Gorczyca et al. [221] predict that the neutral NGa defect can be transferred into VGaNi defect in cubic GaN. 4.3.2 Complexes
The formation energies for simple native point defects in GaN are not energetically as favorable as the complexes, such defects may form with impurities and hydrogen, as
j 935
936
j 4 Extended and Point Defects, Doping, and Magnetism discussed in Section 4.3.1. The same is also in effect with complexes in which impurities with hydrogen would form. It is then fair to consider that these complexes would be the dominant, unintentionally introduced defects. 4.3.2.1 Shallow Donor – Gallium Vacancy Complexes As discussed in Section 4.3.1.2.1 and depicted in Figure 4.67, the gallium vacancy is the dominant native defect in n-type GaN (more so in undoped GaN) because its formation energy is low. As also discussed in the same section, the VGa can easily diffuse even at moderate growth temperatures or thermal annealing and would readily form complexes with other defects. Among the impurities most likely to form stable complexes with VGa are donors. Formation of complexes is driven by electrostatic forces in that a negatively charged acceptor (the charge of VGa is 3 in n-type GaN) and a positively charged donor attract each other. Calculations of Neugebauer and Van de Walle [138] show that VGaON and VGaSnGa complexes act as double acceptors in GaN, and their 1/2 energy levels (at 1.1 and 0.9 eV, respectively) are close to the 2/3 transition level of the isolated VGa (at 1.1 eV). The same calculations also indicate that the electronic structure of Ga vacancy dominates the electronic structure of the VGa shallow donor complexes in a way much similar to the case of n-type GaAs [245–247]. Formation of the VGaON complex is even more favorable than formation of isolated VGa [138]. Similar results were obtained by Mattila and Nieminen [140], who compared the formation energies of the isolated VGa and the VGaON complex at different positions of the Fermi level (Figure 4.68). The VGaON complex has a binding energy of approximately 1.8 eV, as compared to approximately 0.23 eV for the VGaSiGa complex [138]. This supports the notion that the VGaON complex is much more stable. Further, it may be the dominant compensating acceptor in n-type GaN. The VGaCN complex has low formation energy but is unstable in n-type GaN because both constitutes are acceptors and repel each other [138,248]. 4.3.2.2 Shallow Acceptor – Nitrogen Vacancy Complexes The N vacancy and shallow acceptors in p-type GaN essentially are much the same way as the Ga vacancy and shallow donors in n-type GaN. Being a mobile and dominant compensating donor in p-type GaN, VN would be attracted by negatively charged acceptors during growth, cooling down, or thermal annealing. The binding energy for a neutral MgGaVN complex is about 0.5 eV [234,217], although a value of 2.8 eV has been reported by Gorczyca et al. [249]. The formation energy of the MgGaVN complex is notably lower than the sum of the formation energies of the isolated MgGa and VN [217]. The relatively low binding energy for the MgGaVN complex prevents abundant formation of these complexes in p-type GaN in thermal equilibrium [217]. However, the formation of those complexes can be enhanced by kinetically driven processes on the surface during the growth. It is not clear what the energy level of MgGaVN is. MgGaVN has been reported as a deep donor with the energy level at about 0.4 eV below the conduction band [250] and the source of persistent photoconductivity in GaN:Mg along with the associated the bistability with the neutral and 2þ states of this complex [251]. The energy level of (MgGaVN)2þ has
4.3 Point Defects and Autodoping
been estimated at about 0.7 eV above the valence band [234], while an energy level much closer to the valence band has also been reported [249]. Note that the energy level of VN in Ref. [249] is also much lower than that obtained in Refs [217,234]. Similarly to the MgGaVN complexes in GaN:Mg, the BeGaVN complexes are expected to form in Be-doped GaN [252]. Lee and Chang [253] examined the possibility of the formation of MgiVN complexes in p-type GaN. While incorporation of the isolated Mgi is unlikely in GaN due to large atomic radius of Mg, formation of the MgiVN complexes have low enough formation energy when the Fermi level is close to the valence band [253]. The charge state of the MgiVN complex is 3þ in this case, and therefore it can strongly compensate MgGa acceptors [253]. The energy level of the MgiVN complex (or even three close levels) is about 2.8 eV above the valence band maximum. Similar results were obtained by Gorczyca et al. [249] Therefore, apparently not MgGaVN, but MgiVN can be the compensating donor in p-type GaN : Mg, the level of which was detected in a PL study of Kaufman et al. [250] at about 0.4 eV below the conduction band. Note that the MgiVN complex can form only in Ga-rich conditions and when the Fermi level is very close to the valence band [253], that is, passivation with hydrogen or N-rich conditions would prevent formation of this compensating donor. Lee and Chang [253] also assumed that hydrogen passivation can stabilize the formation of the MgiVN complex at higher positions of the Fermi level; however, this possibility is questionable [249]. 4.3.2.3 Hydrogen-Related Complexes Hydrogen readily forms complexes with defects in GaN, and often the formation energies of the hydrogenated defects are lower. In n-type GaN, up to four hydrogen atoms can be bound to Ga vacancy to form the complexes (VGaH)2, (VGaH2), (VGaH3)0, and (VGaH4)þ [216]. The first three complexes are acceptors, whereas the last one is a single donor. The formation energies of the VGaHn complexes are shown in Figure 4.72. As can be seen, the hydrogenated vacancies have lower formation energies than the isolated gallium vacancies, except for the (VGaH4)þ complex. The calculated energy levels of the (VGaH)2 and (VGaH2) complexes are about 1.0 eV above the valence band, which are close to that for an isolated VGa, while the level of (VGaH3)0 is near the valence band maximum [216]. Further, the formation of complexes with several H atoms, such as (VGaH3)0 and (VGaH4)þ, is unlikely because isolated hydrogen exists as H in n-type GaN and it would be repelled from the negatively charged Ga vacancy [216]. Dissociation of the VGaHn complexes is unlikely due to large values of the binding energies [216]. Consequently, once formed during growth in the presence of hydrogen, these complexes cannot dissociate in any postgrowth thermal annealing. In p-type GaN, hydrogen is known to passivate the dominant acceptor (MgGa), as well as the dominant compensating donor (VN) [215,216]. Naturally, H is liberated during the dopant activation process to obtain p-type GaN. In Mg-doped GaN grown by OMVPE, the electrically neutral Mg–H complex has a binding energy of 0.7 eV, with the H atom located in an antibonding site behind the N neighbor of the acceptor [215]. During the postgrowth annealing, the Mg–H complex dissociates, and H diffuses either to the surface or to the extended defects [215]. Similarly, in Be-doped
j 937
938
j 4 Extended and Point Defects, Doping, and Magnetism GaN grown by OMVPE, the Be–H complex may form with a binding energy of 1.81 eV and dissociation energy of 2.51 eV [245]. A postgrowth annealing would also be required to remove hydrogen from the Be acceptors for dopant activation. An N vacancy in p-type GaN can also be passivated by hydrogen during growth by forming a (VNH)2þ complex that has a binding energy of 1.56 eV [216]. The formation energy of the (VNH)2þ complex is lower than the formation of the isolated Hþ and VN when the Fermi level is low in the gap [216]. The VNH complexes can be formed only during growth. Note that diffusion of Hþ toward VN is highly unlikely in p-type GaN because they repel each other. Formation of the (VNH2)þ complex is also likely, while the (VNH3)0 complex is unstable [255]. The VNH complex is expected to have a 2þ/0 transition level at energy of about 2.5 eV above the valence band with the 0/2 level being very close to the conduction band [255]. Earlier calculations indicated the 0/þ transition level of VNH to be near or in resonance with the conduction band [216]. Both V3Nþ and (VNH)2 þ can be formed in abundance in p-type GaN and compensate the dominant acceptor. Note that the hydrogenated vacancies may lose their hydrogen after the sample is grown, either during cooling down and postgrowth annealing, or both. Moreover, the V3Nþ can also migrate during high-temperature annealing [219]. 4.3.2.4 Other Complexes Complexes other than the ones discussed above are discussed succinctly. In Mg-doped GaN, the formation energy of the MgGaON complexes is very low [249]. Therefore, these complexes can readily be formed when oxygen is present in the growth environment. Arguing that both ON and MgGa levels would be pushed out from the energy gap, Gorczyca et al. [249] predicted that these complexes, if formed, would result in semi-insulating GaN. However, relatively low binding energy of the MgGaON complex (0.6 eV) is unfavorable for its formation, unless the kinetically driven processes on the surface pave the way for preferential incorporation of MgGaON [217]. In Be-doped GaN, the BeGaON complex can be formed, which is neutral [254,256]. Moreover, the BeGaONBeGa complex can be formed with the 0/ transition level at 0.14 eV above the valence band maximum [254]. Parallel to the formation of Be2i þ donors, the (BeGaBeint) þ donor complexes may also form in Be-doped GaN, which will compensate the Be acceptor [254]. However, the postgrowth annealing at temperatures above 600 C should result in dissociation of these complexes [254].
4.4 Defect Analysis by Deep-Level Transient Spectroscopy
To gain some insight into the electrical character of point defects, a popular method that is traditionally applied is the deep-level transient spectroscopy [257,258]. In the case of much deeper defects, optical DLTS (ODLTS) and photoconductance with below-gap excitation can be used to discern similar information, albeit the activation energies in this case will be optical ones and can be different from the thermally activated energies. The shortcoming of DLTS is that it is not capable of giving local
4.4 Defect Analysis by Deep-Level Transient Spectroscopy
information or the chemical nature of the defect centers. These can be assuaged to some extent if DLTS is accompanied with systematic studies, such as irradiation damage and measurements in a pressure cell. It can also be used to determine the donor or acceptor nature of a trap, as opposed to isoelectronic traps and vacancies, from the Poole–Frenkel (PF) effect, and it can be used to detect the presence of a capture barrier related to lattice relaxation [259]. The energy, potential profile, and capture barrier can all be used along with theoretical predictions to narrow the possibilities for the composition of a specific trap. 4.4.1 Basics of DLTS
In DLTS, a p–n junction or a Schottky barrier structure is biased into depletion, which causes the defects above the Fermi level (for n-type samples) to release their electrons. A charging pulse (or filling pulse), which reduces the reverse bias across the junction, is applied to fill the previously emptied traps. Upon the removal of the charging pulse, the electrons trapped at deep levels are freed at a rate that depends on the energy level of the trap. This is also related to the rate at which the capacitance returns to its equilibrium value. A schematic representation of this process in terms of the bandgap and depletion region is shown in Figure 4.75. In DLTS, the change in capacitance in various rate windows are plotted as a function of sample temperature. Let us briefly review the fundamentals of deep-level carrier emission and capture in the context of deep-level transient spectroscopy [260,261]. The number of electrons captured by (Nt nt) unoccupied deep trap states during the given time Dt, Dnt, can be written as Dnt ¼ sn hvn inðN t nt ÞDt;
ð4:6Þ
where Nt is the total concentration of deep traps, nt is number of occupied traps, n is electron concentration, sn and hvni are the capture cross section of electrons and root mean square (rms) thermal velocity of electrons. The ratio of change of the number of trapped electrons per unit time to unoccupied trap density, which is the capture rate of electrons cn, can be expressed as cn ¼
Dnt =Dt : N t nt
ð4:7Þ
Using Equation 4.6, the same can be reduced to c n ¼ sn hvn in:
ð4:8Þ
By the same way, the capture rate of holes cp, is given c p ¼ sp hvp ip:
ð4:9Þ
In thermal equilibrium the number of emitted electrons from traps and that of captured ones must be equal as shown below: en nt ¼ c n ðN t nt Þ:
ð4:10Þ
j 939
940
j 4 Extended and Point Defects, Doping, and Magnetism After filling a positive pulse
Steady state bias
on the negative steady bias Empty traps
Depletion region
Newly filled traps
Neutral region
New depletion region
Neutral region
Old depletion region
(a)
V
(b)
0V
t
Vp
VR tp C (b) (a)
(d)
(c)
C0 C(t)
(c) Top and (d) bottom. Figure 4.75 Schematic representation of the depletion (a) followed by the filling pulse (b) in terms of the bandgap and depletion region in the deep-level transient spectroscopy method. The filling pulse is then removed and the relaxation (emptying) of traps is observed and analyzed. (c)
The time evolution of the pulse used in DLTS experiments and resultant evolution of the junction capacitance where regions a and d represent the steady state, (b) represent the filling state, and (d) represents the exponential decay transient following the filling pulse.
4.4 Defect Analysis by Deep-Level Transient Spectroscopy
The same equilibrium condition applies to hole as ep ðN t nt Þ ¼ c p nt :
ð4:11Þ
Therefore, the trap occupancy rate in thermal equilibrium either by electrons or by holes is given by ^t ep cn n ¼ ¼ : N t c n þ e n ep þ c p
ð4:12Þ
It should be noted that the occupancy ratio given by the Fermi–Dirac distribution function for a system with Fermi level at EF is ^t n ¼ Nt
1 g0 Et EF 1þ exp g1 kT
ð4:13Þ
where g0 and g1 are the degeneracy factors when traps at energy level Et are occupied and not occupied, respectively, and k is the Boltzmann constant. Thus, from Equations 4.12 and 4.13, the relationship between emission and capture of electrons and holes can be derived as en g 0 Et EF ; ð4:14Þ ¼ exp cn g 1 kT ep g 1 EF Et : ¼ exp cp g 0 kT
ð4:15Þ
In the following discussion of the equations used in the DLTS analysis, we will assume an electron trap in n-type material, which will behave as a majority carrier trap. When the Fermi level is higher than the trap level, the trap level states are occupied by electrons because cn is larger than en and vice versa. It should be noticed that the capture rate of carriers is dependent on the doping level of materials while the capture cross section and the emission rate of carriers are intrinsic properties of each deep level. When the material is not degenerated, the electron concentration is expressed as in Equation 4.16, where Nc is the effective density of states for conduction band and Ec is the energy level of conduction band edge. Ec EF n ¼ N c exp : ð4:16Þ kT In thermal equilibrium Equations 4.8 and 4.16 can be substituted into Equation 4.14 to obtain an expression about emission rate of electrons from deep traps as a function of temperature in Equation 4.17: g0 Ec Et en ðTÞ ¼ sn hvn i N c exp : ð4:17Þ g1 kT
j 941
942
j 4 Extended and Point Defects, Doping, and Magnetism Each component in Equation 4.17 also can be expressed as follows. 1 3kT 2 hvn i ¼ ; m N c ¼ 2M c
2pm kT h2
ð4:18Þ 32
;
ð4:19Þ
Mc is the number of conduction band minima, m is the effective mass (EM) of an electron, and h is Planks constant. DE s sðTÞ ¼ s¥ exp ; ð4:20Þ kT s1 is the capture cross section at T ¼ 1 and DEs is the activation energy of the capture cross section. Substitution of Equations 4.18–4.20 into Equation 4.17 gives the temperature dependence of electron emission rate, which can be used in obtaining the activation energy of thermal emission of electrons Ena and the apparent capture cross section sna. E na ; ð4:21Þ en ðTÞ ¼ gT 2 sna exp kT where pffiffiffi 3 g ¼ 2 3Mc ð2pÞ2 k2 m h 3 ;
ð4:22Þ
and the apparent cross section for electron sna is given by sna ¼
g0 s¥ : g1
ð4:23Þ
Although it does not appear in Equation 4.21, one must keep in mind that the thermal activation energy Ena and the apparent capture cross section sna do not signify the trap energy level and capture cross section. By looking over the steps taken to derive Equation 4.21, one can easily find that Ena is the summation of (Ec Et) and DEs and sna is the capture cross section extrapolated to T ¼ 1 and modified by the degeneracy ratio. Also, both changes depend on the temperature dependence of (Ec Et). Despite this fact, the plot of en/T2 versus T1, called trap signature, can be used to extract those two values to characterize each deep trap on the basis of Equation 4.21. It is also possible to obtain the trap density by going over several steps of calculation under the condition of large and long enough pulse to completely fill the traps. Assuming a donor trap that is neutral when filled and positive when empty, the total trap concentration is N T ¼ N Tþ þ N 0T ;
ð4:24Þ
4.4 Defect Analysis by Deep-Level Transient Spectroscopy
where N Tþ is the concentration of empty traps and N 0T is the concentration of filled traps. The rate of change is then dN 0T ¼ cf N Tþ eN 0T ; dt
ð4:25Þ
where cf is the capture rate during the filling pulse and e is the emission rate. The emission rate is the same during filling and emission bias conditions, while the capture rate depends on the free carrier concentration, which will be different during the filling pulse and measurement bias. Using the boundary condition that the traps are completely empty at the start of the filling pulse yields the equation for the concentration of filled traps during the filling pulse: N 0T ¼
cf N T ½1 expð ðe þ c f Þt: e þ cf
ð4:26Þ
For large capture rates or long filling pulses, all the traps will be filled. For arbitrary filling pulse widths, and assuming a capture rate much larger than the emission rate during the filling pulse, the initial condition for measurement of trap emission when the bias is changed from filling to measurement conditions is N 0T ¼ N T ½1 ðexpðc f tf ÞÞ:
ð4:27Þ
Then, the concentration of filled traps as they empty during measurement is N 0T ¼ N T ½1 expðc f tf ÞexpðetÞ: The measured capacitance will then be 1=2 qeðns N 0T ðt; TÞÞ : Cðt; TÞ ¼ 2ðV þ fÞ
ð4:28Þ
ð4:29Þ
In the general case where several traps are present with overlapping signals in a given temperature range, the shallow carrier concentration term can be separated and the time-dependent terms can also be expressed as individual terms. This can be rewritten as X t 2 Ai exp Cðt; TÞ ¼ ; t i i where qens qeN T1 ð1 ecf 1 tf Þ and A1 ¼ ; 2ðV þ fÞ 2ðV þ fÞ 0 1 X t 2 Ai exp@ A; Cðt; TÞ ¼ t i i
A0 ¼
A0 ¼ A1 ¼
qens ; 2ðV þ fÞ
qeN T 1 ð1 e cf 1 tf Þ ...; 2ðV þ fÞ
ð4:30Þ
j 943
944
j 4 Extended and Point Defects, Doping, and Magnetism The trap concentration can be obtained from the ratio of Ai to A0. For long filling pulses or fast capture rates, the trap concentration is Ai ns : ð4:31Þ A0 The rate window plot can also be used to determine the trap concentration. One can get the DLTS peak signal S(T) from the rate window plot. The concentration of a trap is given by NT ¼
N t ¼ 2ðDCð0Þ=CÞðN D N A Þ;
ð4:32Þ
where ND NA is the net donor concentration in the n-type material. However, the capacitance immediately after the bias, which is switched from the filling pulse to the measurement bias is difficult to obtain accurately due to the settling time required before the initial capacitance measurement. Alternatively, the trap concentration can be expressed in terms of the rate window and the measured maximum capacitance transient signal [262]. NT ¼
dCmax 2N D expf½r=ð1 rÞln rg dCmax 2N D r r=ðr 1Þ ¼ C0 C0 1r 1r
ð4:33Þ
where C0 is the steady-state capacitance, dCmax is the peak height, and r ¼ t2/t1. This equation assumes that the trap concentration is much less than the shallow carrier concentration, or equivalently, that the capacitance transient is much less than the steady-state capacitance. This equation should be further modified, adjusting for the difference in depletion width where the Fermi level crosses the shallow dopant level and where the Fermi level crosses the deep level. The distance between the two points is 2eðE F E T Þ 1=2 l¼ ; ð4:34Þ q2 N D and depends only on the trap energy [263]. The practical aspects of data collection and analysis according to the previous equations are as follows. The DLTS signal is obtained by measuring the capacitance change of the Schottky diode in a given rate window through thermal scan [264]. The change of the capacitance is caused by the emission of electrons (majority carriers) from the majority carrier traps during the reverse bias period (tm) after a pulse (tf) of near-zero bias during that the traps are filled with electrons. When the pulse bias is zero or less, majority carrier traps in the depletion region will be filled, while, for a p–n junction, minority carrier traps will be filled when the pulse bias injects minority carriers. The sequence of pulse bias and reverse bias alternates with repetition time tr, as shown in Figure 4.75a. Figure 4.75b shows the diode capacitance transient with the bias condition given in Figure 4.75a as a function of time. In this case an exponential transient is assumed, where the capacitance as a function of time can be expressed as CðtÞ ¼ C þ DCð0Þexpðt=tÞ;
ð4:35Þ
where C is the quiescent capacitance under reverse bias, DC(0) is the capacitance change due to the pulse bias at the end of the pulse, and t is the time constant. This form of Equation 4.30 is useful for rate window analysis. The time constant is equal to
4.4 Defect Analysis by Deep-Level Transient Spectroscopy
the inverse of emission rate of electrons. t ¼ en 1 :
ð4:36Þ
Now, the normalized capacitance transient between time t1 and t2 is the DLTS signal and defined as Equations 4.37 and 4.38 SðTÞ ¼ ½Cðt1 Þ Cðt2 Þ=DCð0Þ;
ð4:37Þ
SðTÞ ¼ ½expðt1 =tÞ ½expðt2 =tÞ:
ð4:38Þ
Thus, by differentiating Equation 4.38 and letting the output equal to zero, the value of time constant at the peak of each DLTS signal, tmax, can be obtained as tmax ¼ ðt1 t2 Þ½lnðt1 =t2 Þ 1 :
ð4:39Þ
This means that the DLTS signal is a maximum when the time constant is the same as the value of tmax determined by the selection of t1 and t2. The measured temperature for the peak position of a DLTS signal with a calculated emission rate is one point in the ln(T2/en) versus 1/Tplot. More data points can be obtained by setting different combinations of t1 and t2 and measuring the temperatures for the peak positions of corresponding DLTS signals. The slope of this plot gives the thermal activation energy and the capture cross section appearing in Equation 4.21. Rate window analysis uses only a few pairs of points of the capacitance decay. It is reasonably accurate and relatively immune from noise. However, the energy can be distorted if two traps have overlapping peaks. An alternative to rate window analysis is to record the capacitance transient over several decades of time and plot the signal as SðtÞ ¼ t
d½C2 ðtÞ C20 : dt
ð4:40Þ
This technique is called isothermal capacitance transient spectroscopy (ICTS) [258], which has also been employed for analyzing nitrides. A further alternative, where the data are collected under isothermal conditions, fits the capacitance decay transient for one or more emission rates and generates the Arrhenius plot from the emission rate(s) at each temperature, according to Equation 4.30. The accuracy of several methods, including rate window (or boxcar) analysis, Fourier transform, and fitting using the method of moments, is compared in Ref. [265]. Further information on the nature and origin of the traps can be obtained by varying the electric field and the filling pulse width, or profiling the deep levels. The electric field can be used to identify traps with Coulombic potential profiles, although the presence of a capture barrier can alter the effect [266]. The Poole–Frenkel effect reduces the energy for emission as a result of the electric field tilting the bands in which a Coulombic defect potential resides. The change in energy with electric field is rffiffiffiffiffiffiffiffi hqF ð4:41Þ DU PF ¼ 2q e where Z is the charge state after emission. As an example, for a defect that is singly charged after emission, the shift in energy will be 0.11 eV for a field of 200 kV cm1.
j 945
946
j 4 Extended and Point Defects, Doping, and Magnetism Traps at this peak electric field would be reduced by 0.11 eV, and the trap energy would approach the true value near the depletion zone edge. For a doubly charged defect after emission, the maximum reduction at 200 kV cm1 would be 0.15 eV. One solution to alleviate smearing of the emission transients is to confine the emission to be only from traps in a narrow range of electric field using double-pulse DLTS, where the capacitance signal is from the difference in response from two slightly different filling pulses [267]. Several of the traps are predicted to have significant changes in lattice position, depending on the charge state, such as the VN [219]. In DLTS measurements, this presents itself as a barrier to the capture of carriers. If the capture barrier is large enough, it can be characterized by varying the filling pulse width and measuring the change in capacitance transient amplitude. A capture barrier can also be present as a result of charge buildup along a line defect. In the case of a line defect, the concentration of filled traps increases linearly with logarithm of filling pulse width according to [268] nT ¼ sn hnth itnN T ln½ðtp þ tÞ=t:
ð4:42Þ
The magnitude of a capture barrier associated with a point defect can be determined on the basis of a method by Criado et al. [269] where the peak amplitude is recorded as a function of the filling pulse width for nonsaturating filling pulses. Under partial-filling conditions the capacitance transient amplitude is C0 ðtf ; TÞ ¼ Cinf ½1 expð c n tf Þ
ð4:43Þ
where Cinf is the capacitance transient amplitude for a saturating filling pulse, and cn is the thermally activated capture rate: c n ¼ hnth incap s¥n expð DE b =kTÞ:
ð4:44Þ
Here, DEb is the capture barrier. Plotting the left side of the following equation ln½1 C0 ðtf ; TÞ=Cinf ¼ KT 1=2 expð DE b =kTÞ;
ð4:45Þ
versus the inverse temperature gives the capture barrier as the slope. K is a temperature-independent constant. Capture barriers associated with a line defect should be distinguishable from barriers associated with point defects because the mechanism resulting in the barrier is different in each case. The capture barrier along a line defect depends on the number of carriers already trapped on the line defect, while the barrier associated with a point defect results from the requirement that a defect state assumes an excited vibrational state before a carrier can be accommodated. Other work has shown that the capacitance transient spectral response to various filling pulse widths can be used to identify different charge state arrangements, such as the negative-U configuration [270]. Both the presence of a capture barrier and the correlation energy U can be predicted for certain types of defects and can be used to narrow the field of possibilities for a specific defect. A considerable number of defect studies have been done on GaN and related materials. However, there is also a considerable degree of spread in the reported
4.4 Defect Analysis by Deep-Level Transient Spectroscopy
values. The spread in reported values is not unique to GaN, though. For example, a wide range of energy levels, from 0.49 to 0.58 eV, has been noted for gold deep levels in Si [271]. It was found that gold in silicon can occupy several defect configurations, also that other impurities have signatures that are indistinguishable from gold. Furthermore, the Koster–Slater model has been used in an orthogonalized tightbinding function basis to predict which elements are likely to form substitutional A1-symmetric traps with energy levels deep in the bandgap of covalently bonded semiconductors [272]. The conceptual framework that they developed pointed out that substitutional deep trap energies are asymptotic to the dangling bond energies, so that no traps exist deeper than the dangling bond energy. Because the dangling bond energy is a property of the host lattice, the deep trap wave function is predominantly host-like rather than impurity-like. This explains why impurities whose atomic energies differ by 10 eV may produce trap energies differing by only a fraction of an electron volt. This would suggest that there may only be a few ranges of energies for diverse defects, depending on the configuration of the host lattice around the defect, for example, interstitials or impurities, but there are several other possible reasons that the observed energy levels characterized by DLTS and related methods exhibit a spread. The primary causes of the spread will be introduced next, followed by experimental results, and a subsequent in-depth discussion of the spread in the measured deep-level characteristics. Variations in measured energy and capture cross section can result if there is a large entropy term, or if there is a large Poole–Frenkel effect, or from inappropriate measurement conditions. If a strong Poole–Frenkel effect is present, the trap will have a distribution of energies in the depletion region, as already mentioned above, making fitting of the transients unreliable or shifting the rate window peaks to lower temperature. Conversely, short, nonsaturating filling pulses can truncate the low temperature side of the rate window peaks, making the peak appear at a higher temperature if a capture barrier is present. Another source of spread in measured values comes from the entropy term, because the Gibbs free energy is measured, reported in DLTS measurements as ET, which includes the enthalpy and the entropy as DGn ¼ DHn TDSn ;
ð4:46Þ
so that the emission rate is properly stated as en ¼ sn Xn hnth iN c e DHn =kT ; 0 1 DS n A: X n ¼ exp@ k
ð4:47Þ
In DLTS, the Gibbs free energy is measured, reported as ET, which includes the enthalpy and the entropy expressed as in Equation 4.46 giving rise to the emission rate as in Equation 4.47.The entropy factor in Xn consists of electronic degeneracy and atomic vibrational variations. The degeneracy term in Equation 4.47 has the effect of increasing the apparent capture cross section. Differences in measured capture cross
j 947
948
j 4 Extended and Point Defects, Doping, and Magnetism sections varying by factors of 50 are not uncommon [271]. The vibrational term may be accounted for partially by a capture barrier, and also by the random variety of phonons participating in the emission. Capture barriers are associated with the lattice relaxation predicted by theory for some defects, and thus provide an important factor to discriminate between defects. In the special case of GaN, considering the optical phonon energy of 92 meV, a discrete spread should be seen particularly for lower energy traps depending on the proportions of acoustic and optical phonons by which the lattice accommodates the thermal energy. 4.4.2 Applications of DLTS to GaN
Even though the DLTS method itself is reasonably well developed, the characteristics of defects as analyzed by DLTS for GaN are widely scattered. There is an unacceptably large dispersion in the basic reported parameters of the defects in GaN, let alone their genesis and nature. As many as eight point defects active in DLTS, some of which may be the same due to the aforementioned dispersion, have been reported in n-type GaN, and only one in p-type GaN. To gain a quantitative appreciation of the spread in data, the reported activation energy and capture cross section for the E1 trap span from 0.14 to 0.27 eV and from 3.9 · 1018 to 4.2 · 1016 cm2, respectively. The situation is similar for other trap levels. A good part of this dispersion might be ascribed to the Poole–Frenkel effect wherein the effective barrier is lowered by the electric field and a good deal of the rest to the samples themselves. GaN lends itself to this possibility more than other semiconductors because it is a highly piezoelectric material that can lead to large electric fields, in hundreds of kilovolts per centimeter, as a result of local strain caused by lattice defects and inhomogeneities. For consistency, the results obtained in authors laboratory for HVPE, OMVPE, and RF MBE grown GaN layers are discussed first to avoid any measurement system related variations in the data found in literature. This is followed by a discussion of data found in literature for GaN layers grown by the aforementioned methods. It should be mentioned that the term Ei is used for deep levels in GaN with i ¼ 1, 2, . . .. The higher the subscript, the higher the temperature at which the DLTS spectrum goes through a peak. However, there is noticeable variation in some of the trap energies in GaN prepared by various methods leading to a situation wherein not all Ei traps are consistent among all the growth methods. To avoid confusion, the colleagues in the authors laboratory added two additional letters to the nomenclature that describe the growth method. In the discussion of the data reported in the literature, the nomenclature found in that medium will be repeated, as it is still difficult to correlate defects levels obtained in one laboratory with others fully. To underscore the dilemma mentioned above and also display the distinct differences in GaN trap signature prepared by different methods, DLTS spectra resulting from each growth method are shown in the same scale in Figure 4.76. GaN used in this investigation was grown by MBE, OMVPE [273], and HVPE [274]. The MBE and OMVPE samples were grown on c-plane sapphire. The freestanding HVPE GaN was grown on c-plane sapphire, separated from the substrate, and thinned down
4.4 Defect Analysis by Deep-Level Transient Spectroscopy
Figure 4.76 DLTS spectra for GaN grown by MBE, HVPE, and OMVPE on semilog scale. The rate window is 223 s1.
to a thickness of 200 mm by a series of mechanical polishing and reactive ion etching processes. The carrier concentrations determined from capacitance–voltage (C–V) profiling are 9 · 1015, 1 · 1017, and 2.4 · 1015 cm3, for the MBE, OMVPE, and HVPE GaN, respectively. The trap concentrations measured show a weak proportionality to the shallow carrier concentration for the MBE and OMVPE traps at 340 K, and a rough proportionality for the HVPE sample. The traps at 150 and 500 K have the same concentrations for MBE and OMVPE GaN, and are fewer for the thick GaN grown by HVPE. The dominant trap in HVPE GaN is at 220 K. The same trap may be present in nearly the same concentration in MBE and OMVPE GaN, but the signal is much smaller than that associated with the traps at 150 and 320 K in those samples. The following treats the further characterization of each of the dominant traps in GaN grown by the various methods. A scan of the traps in freestanding HVPE GaN diodes over the temperature range from 80 to 700 K was first done. Six peaks associated with traps, as well as the Arrhenius plot associated with each trap, are evident as shown in Figure 4.77. Table 4.3 summarizes the characteristics for each of the traps in HVPE-, OMVPE-, and MBE-grown GaN. This includes the emission energy, capture cross section, and the presence of a capture barrier as indicated from the variation in capacitance transient amplitude as the filling pulse width is varied, and the Poole–Frenkel effect in the form of PF slope. To determine the capture kinetics, the temperature was fixed near each of the DLTS peaks and the filling pulse width was varied in the freestanding HVPE sample. The traps at 230 K (EHV2) and 350 K (EHV3 and 4) did not show any increase in transient amplitude with increasing filling pulse. The trap at 160 K corresponds to EHV1, the fast trap at 440 K is EHV5, the slow trap at 440 K is EHV6, and the fast trap at 530 K is EHV6. The slow trap at 530 K is a higher energy trap that is not seen in the rate window plot due to soft diode breakdown. At 440 and 530 K, the transients were
j 949
950
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.77 Arrhenius plot for the rate window plot shown in Figure 4.76 for freestanding HVPE GaN.
found to consist of two components. The trap at 160 K shows behavior that would indicate that it has a capture barrier. At 440 K, two components were present, one at 1100 s1, corresponding to EHV5 and one at 80 s1 corresponding either to EHV6 or another trap that was not detected in the rate window plot or the capacitance transient fitting for the Arrhenius plot. The amplitude of one rises as the other falls, suggesting perhaps that the two are related. If the two traps have different charge states, that is, the low energy (1100 s1) trap is an excited state and the higher energy trap is another energy level of the same trap, then the Poole–Frenkel effect should be significantly different for each. The peak at 530 K (there is dispersion from sample to sample as to the temperature at which the peaks occur) also contains two components, 2000 s1 (EHV6) and 300 s1, although only the higher energy trap (300 s1) changes amplitude with pulse width in the range of pulses used. The slower emission rate indicates that there is another trap present with energy higher than 1.08 eV (EHV6). The HVPE diodes consistently exhibited soft breakdown at temperatures above 575 K. To determine whether the charge state of the trap would affect the emission rate therefore the perceived trap energy level, the field was changed for each trap. Traps EHV2, EHV5 (fast), and EHV6 each had negligible dependence of emission rate on electric field. The dependence of the emission rate for trap EHV1, as signified
4.4 Defect Analysis by Deep-Level Transient Spectroscopy Table 4.3 Trap characteristics for HVPE, OMVPE, and MBE GaN.
Label
Trap peak temp (T)
Concentration (cm3)
Energy (eV)
Capture cross section (cm2)
HVPE EHV1 EHV2 EHV3 EHV4 EHV5
160 220 280 350 440
3.8 · 1012 2.5 · 1013 7.5 · 1012 7.0 · 1012 4.8 · 1012
0.212 0.313 0.426 0.612 0.627
2.9 · 1017 1.4 · 1016 1.8 · 1016 3.7 · 1015 2.1 · 1016
EHV6
530
1.6 · 1013
1.084 4.8 · 1014
0.246 0.246 0.536 0.92
3 · 1013 9 · 1016 5 · 1016 1 · 1015
Percentage change from 0.5 to 200 ms filling pulse width
39 0
Poole–Frenkel related slope
0.026 0.0002
0 24 (fast tx) 23 (slow tx)
0.04 Negligible 0.013
0 (fast) 22 (slow)
Negligible 0.016
38
0.003
0 29
0.009 0.002
OMVPE EOM1 EOM2 EOM3 EOM4 EOM5
150
2.4 · 1013
325 520 620
6.6 · 1015 9.6 · 1013 2.1 · 1014
MBE EMB1 EMB2 EMB3 EMB4
120 150 240 320
2.7 · 1012 3.2 · 1013 3.1 · 1013 2.8 · 1014
0.243 0.293
3 · 1013 4 · 1014
11
0.002
0.342
3 · 1018
36 (fast)
550
3.7 · 1014 1.7 · 1014 1.9 · 1014
0.552 1.17
7 · 1016 3 · 1015
Emission rate decreased with increasing field (both)
38 (slow) 0 (fast) 18 (slow)
EMB5 EMB6 EMB7
Negligible
See Section 3.5.5.3 for reduction of defect density by using epitaxial lateral overgrowth methods. Note: EYZW (E ¼ trap designation letter for GaN, YZ ¼ HV for HVPE, OM for OMVPE, MB for MBE), E ¼ 1. . .. The Frenkel–Poole related slope is determined by the slope of the emission rate with respect to the square root of electric field.
by the slope of the emission rate versus square root of electric field plot, has a slope of 0.026, within experimental error of the slope of 0.028 expected for a single charge. Trap EHV4 had a field dependence with a slope of 0.04, which is close to what is 0.039 and expected for a doubly charged trap. The slow trap at 440 K is expected to have the same field dependence as the fast trap at 530 K. However, the slow trap at 440 K exhibits field dependence but not the fast trap at 530 K. This shows that the two are not the same trap even though the emission rates are of the correct magnitude for EHV6.
j 951
952
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.78 DLTS spectra for OMVPE GaN. The rate window is 223 s1.
Turning our attention to OMVPE-grown GaN, which underwent a characterization in the same manner as the HVPE sample, the rate window plot showing the deeplevel spectrum is shown in Figure 4.78. At least five traps are discernable in the rate window plot, including one that shows only partial peaking above 700 K. Fitting of the transients at each temperature resulted in the Arrhenius plot of Figure 4.79. The peak at 150 K is composed of two traps of the same energy but with considerably different capture cross sections (3 · 1013 cm2 for EOM1, and 9 · 1016 cm2 for EOM2). The traps at 150 K (EOM1) and 530 K (EOM4) show a filling pulse dependence for the trap occupation, while the trap at 325 K (EOM3) does not. Attempts at characterizing the capture kinetics and field dependence of the trap at 620 K did not yield reliable results. Field dependence measurements indicate that none of the traps in OMVPE GaN have a slope near 0.028 in plots of emission rate versus square root of electric field. This result contrasts with the field dependence corresponding to a charge state of þ1 for traps EHV1 and EHV4 observed in HVPE GaN. The trap density can be reduced by using the standard epitaxial lateral overgrowth or the nanonetwork in the general context of ELO. The DLTS data for standard ELO and in situ SiNx nano-ELO defect reduction schemes are discussed in Section 3.5.5.3 comparatively. As for the traps in RF MBE grown GaN, the rate window plot is shown in Figure 4.80, and the corresponding Arrhenius plot is shown in Figure 4.81. Similar to EOM1 and EOM2 traps in OMVPE GaN, two low-temperature traps superimposed at 160 K were measured with designations of EMB1 and EMB2. The main peak at 340 K was found to consist of two traps, EMB4 and EMB5. This decomposition has also been seen in several OMVPE samples with high trap concentrations, although the energy for EMB4 is somewhat low. The peak at 240 K did not result in a line on the
4.4 Defect Analysis by Deep-Level Transient Spectroscopy
Figure 4.79 Arrhenius plot for the traps in OMVPE GaN.
Figure 4.80 Deep-level spectra for traps in GaN grown by MBE. The rate window is 223 s1.
j 953
954
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.81 Arrhenius plot for the traps in MBE GaN shown in the rate window plot of Figure 4.80.
Arrhenius plot, but merely appeared as inflection points. Conventional rate window analysis and isothermal capacitance transient analysis also did not result in reliable energy measurements for EMB3, indicating that it does not emit thermionically, or that the Poole–Frenkel barrier lowering is more pronounced in this trap. The broad peak at 600 K might be expected to lead to more than one line on the Arrhenius plot, and there were indications of the presence of additional traps by inflections in the Arrhenius plot. However, the fits of the transients for emission rates were not dependable in this temperature range. As in the case of HVPE and OMVPE samples, the capture behavior was analyzed in terms of any dependence on the filling pulse width, and the charge state was analyzed in terms of the emission rate dependence on electric field. The traps at 120 K (EMB1) and 240 K (EMB3) did not have a high enough capacitance transient signal, clear of other signals, to result in reliable analysis. The traps at 150 K (EMB2), 320 K (EMB4 and EMB5), and 570 K (EMB7) each displayed evidence of a capture barrier. The trap EMB6 does not show capture behavior consistent with the presence of a capture barrier. The field dependence of traps EMB2 and EMB6 showed that neither is a donor trap. An unusual result was obtained for the traps at 320 K. Both EMB4 and EMB5 had emission rates that decreased with increasing electric field strength,
4.4 Defect Analysis by Deep-Level Transient Spectroscopy
opposite to what is expected for the Poole–Frenkel effect. If there is a large electron–phonon coupling, the emission rate can be reduced by the electric field according to 2SkB en en0 exp ; ð4:48Þ hw where S is the Huang–Rhys factor [275]. After presenting the MBE experimental results, a comparison is given of the field dependence of the emission rates for the traps at approximately 325 K for all growth methods. The GaN samples grown by reactive MBE using ammonia for N source behave better in terms of DLTS analysis and will be discussed below as part of data obtained from the literature. To summarize the DLTS data obtained in authors laboratory, the rate window plot of Figure 4.76 appears to indicate that the energies and capture cross sections are similar for a few of the traps appearing at the same temperature in different growth methods. However, the additional characteristics of capture behavior and field dependence of the emission rate indicate that they are either different traps or possibly the same trap but appear different due to GaN polarization properties, dependent on the individual sample properties. Each growth method resulted in traps in the range from 120 to 160 K, EHV1 in HVPE, EOM1 and EOM2 in OMVPE, and EMB1 and EMB2 in MBE GaN. Due to experimental limitations in the range of emission rate, and other reasons such as field effects mentioned already, peaks appearing in the same temperature range in different samples may have slightly different energies and capture cross sections, for example, a lower slope leads to a lower capture cross section calculated from the intercept with the vertical axis. The field dependence and capture dependence are also used to discriminate between traps. The energy, capture cross section, and field dependence would all indicate that the trap appearing in this temperature range in HVPE GaN is a different trap from that appearing in OMVPE and MBE. EMO1 and EMB1 are the same traps, and EMO2 and EMB2 are likely the same traps. When measurable, the capture kinetics indicate that each trap has an associated relaxation. The EHV1 trap in freestanding HVPE sample uniquely showed a field dependence corresponding to a singly charged donor. The traps between 320 and 350 K also showed distinguishing characteristics. EHV2 and EHV3 were discernable only in the HVPE GaN, perhaps due to the higher sensitivity afforded by the lower doping level. EHV4 had not only a higher energy but also field dependence indicating that it is a double donor. EMO3 and EMB5 have very similar energy and capture cross section, but the capture kinetics are different. Traps EMB4 and EMB5 in the MBE GaN have capture barriers associated with them, which do not appear to be related to higher edge dislocation density commonly found in MBE GaN, because the trap tends to saturate as the filling pulse is extended. The third group of traps appearing at high temperatures, from 520 to 620 K and above, was not as well defined as the other traps, but did show some similarities. The measured energy, capture cross section, and field dependence were similar for EHV6, EMO4, and EMB6. However, the capture kinetics for EMO4 would indicate
j 955
956
j 4 Extended and Point Defects, Doping, and Magnetism that it is not the same trap or is affected differently by its environment, for example, higher carrier concentration. The relative trap concentrations can be used as a measure of whether or not equilibrium conditions are approached during growth, which is important for the correlation of theoretical predictions of the defect characteristics to experiment. If it is assumed that the Fermi level is near the same energy during growth for the various growth methods, then the relative concentrations of the same trap present in different growth methods (or at different temperatures in the same growth method) should scale according to the Boltzmann factor, and depend only on the temperature. However, it can be seen from Figure 4.76 that, although the growth temperatures for HVPE and OMVPE are nearly the same, the trap concentrations are nearly equal only for the peak at 200 K. Furthermore, the MBE GaN should have significantly lower defect concentrations than the other two growth methods, but this is only true compared to OMVPE for the peak at 320 K, but not compared to the GaN grown by HVPE. From this, we can conclude that the growth conditions are not near enough to equilibrium to determine the concentration of defects based on the formation energies. Turning our attention to the literature, trap characteristics of GaN analyzed by DLTS prepared by various methods are tabulated in Table 4.4. Despite a reasonably large number of papers on the topic, a closer look reveals the limited scope of the characterization. The literature is almost exclusively on majority carrier Schottky diodes. In spite of more than a few reports on the topic, a more copious use of p–n junctions would allow a thorough characterization of the minority carrier traps also. To underscore the dispersion in the literature, the same published data categorized by each trap is tabulated in Table 4.5 that clearly shows the dispersion in the activation energy and capture cross section reported for DLTS active defects in GaN. Having tabulated the various reported values of trap energy levels and the associated capture cross sections, let us now discuss the particular findings associated with some of those reports. Before delving into a detailed discussion, however, it should be noted that in wide bandgap semiconductors, the very deep states may not efficiently release their electrons, particularly at low temperatures, rendering the traditional DLTS somewhat impractical. To detect traps around mid-gap for GaN, at 1.7 eV, requires unattainably high temperatures and long capacitance transient sampling times. To circumvent this problem, optical excitation in conjunction with capacitance transients can be employed. Deep-level optical spectroscopy (DLOS) is well known as a powerful tool for the characterization of electronic deep levels in the bandgap of semiconductors [296]. This technique measures changes in the depletion region capacitance under optical excitation and can provide detailed mapping of the deep levels that would be undetectable by thermal emission techniques such as DLTS [264] and thermal admittance spectroscopy [297]. Similar to thermal excitation of electrons, when the electrons are optically excited into the conduction band in a reverse biased junction, the charge in the depletion region increases following a resonant optical excitation [298]. The DLOS technique allows one to obtain the spectral distribution of the optical cross sections, son ðhvÞ and sop ðhvÞ, for transitions between the deep level and the conduction band and deep level and the valence band,
Material type
n-type GaN (OMVPE)
n-type GaN
n-GaN(OMVPE) source TMGa
n-type GaN by OMVPE implanted with N
Undoped
n-type GaN by RMBE
Reference
GÅtz et al. [276]
Hacke et al. [277]
Lee et al. [278]
Haase et al. [279]
Hacke et al. [280]
Fang et al. [281]
DLTS
DLTS
DLTS
DLTS
DLTS/ICTS
DLTS
Measurement method
C1 ¼ 0.44 D1 ¼ 0.20 E1 ¼ 0.21
0.26 0.62
E3 ¼ 0.67
1.3e15 2.4e14
E2 ¼ 4.9e17 E3 ¼ 5.8e12 E4 ¼ 5.8e10
E2 ¼ 0.49 eV E3 ¼ 1.63 eV
0.598 0.67
E1 ¼ 3.9e18
E1 ¼ 0.14 eV
E2 ¼ 0.60
1.1–2e15 2.3–3.1e15 0.5–3.8e14
E1 ¼ 0.22–0.27 E2 ¼ 0.56–0.59 E3 ¼ 0.65–0.69
0.27
7e13 cm3 6.3e14 cm3
Emission rate 231 s1
E1 ¼ 0.18 eV E2 ¼ 0.49 eV
1.6e13 2e13
2.48e14 3.34e15
1.72e14
1.3e14 3.6e14 3.8e14
Trap concentration (cm3)
Capture cross section (cm2)
Binding energy (eV)
Table 4.4 Defect characteristics of GaN reported in the literature (the numbers by authors name indicate the reference numbers).
(Continued )
E1: N vacancy
E2: intrinsic defect such as N antisite
E1 an E2 are related to H and C atoms
Comments
4.4 Defect Analysis by Deep-Level Transient Spectroscopy
j 957
Material type
n-GaN (OMVPE)
n-GaN (RMBE)
Mg-doped GaN by OMVPE
n-GaN
n-GaN
Si-doped GaN
n-GaN (MBE)
Reference
Auret et al. [282]
Wang et al. [283]
Nagai et al. [284]
Look et al. [285]
Shmidt et al. [286]
Hacke et al. [287]
Fang et al. [288]
Table 4.4 (Continued)
DLTS
Capacitance transient spectroscopy
DLTS
DLTS TDH
DLTS
DLTS
DLTS
Measurement method
0.89 0.67
E2 ¼ 0.85
0.25 0.59 0.81
3e14 1e15
E2 ¼ 2.3e15 E3 ¼ 1.7e16
E2 ¼ 0.49 E ¼ 0.59 0.06 0.9
E1 ¼ 7.6e17–1.2e12
8e15 1e14
Capture cross section (cm2)
E1 ¼ 0.22–0.475
E1 ¼ 0.234 E2 ¼ 0.578 E3 ¼ 0.657 E4 ¼ 0.961 E5 ¼ 0.240
0.27 0.65
Binding energy (eV)
2.5e13 2.5e13 1.5e14
7.7e14 1.2e15 4.2e15 8.3e15 2.2e14
Trap concentration (cm3)
Dislocations N vacancy
N vacancy
E1: Mg–N–H complex
Comments
958
j 4 Extended and Point Defects, Doping, and Magnetism
n-GaN
Soh et al. [289]
Freestanding GaN
Freestanding GaN by HVPE
Mg-doped GaN (OMVPE)
n-type GaN OMVPE
Reshchikov et al. [290]
Fang et al. [160]
Seghier et al. [291]
Muret et al. [292]
pþn
Material type
Reference
DLTS
DLTS Photocapacitance
DLTS
PL
DLTS
Measurement method
E1 ¼ 4e17 Ec ¼ 1.6e15
E1 ¼ 0.18 Ec ¼ 0.35
0.81eV Hole trap
2e14
E0 B ¼ 1.5e15
E0 B ¼ 0.53
1.1 1.9 (from valence band) hole traps
4e21 yellow band 2e19 green band
2.4 eV
4.2e16 4.2e16 3.9e15 1.6e13 1.3e19 8.1e14
1.2e14 2.8e14 2.6e15 3–6.4e15 4.3e13
0.62 0.41 0.25 0.43–0.48 (MBE) 0.25 (MBE) 0.24 0.56 0.82 1.07 0.23 2.55 (0.85) hole trap
Capture cross section (cm2)
Binding energy (eV)
9e13 2e14 6e14 2.4e15 4.8e13 2e14
Trap concentration (cm3)
(Continued )
E1: N vacancy
Ec: RIE damage
Due to extended defects
Comments
4.4 Defect Analysis by Deep-Level Transient Spectroscopy
j 959
n þ p junction GaN (OMVPE)
n-GaN (OMVPE) hole trap analysis
Undoped GaN
Chen et al. [293]
Auret et al. [294]
Soh et al. [295]
Si-doped GaN
Material type
Reference
Table 4.4 (Continued)
3e14 8.7e18 2.6e18 9e16 2.4e20 7.9e18 4.5e18 5.2e17 5.4e16
0.85 A2 ¼ 0.17 A3 ¼ 0.24 A5 ¼ 0.59 A1 ¼ 0.10 A2 ¼ 0.18 A3 ¼ 0.24 A4 ¼ 0.40 A5 ¼ 0.62
O-DLTS
5e15
Capture cross section (cm2)
0.25
0.59 (e trap in p region)
Binding energy (eV)
TSCAP
DLTS
Measurement method
6.9e13 9e13
5.5e14 8.5e13 1.3e13 5.5e13 8.8e13
2e15 3.5e14
5e14
1e16
Trap concentration (cm3)
A1 – N vacancy A2,3,5 are same as above A4 – Si dopant-induced defects
A2,3 – screw- and mixed-type dislocations A5 – edge dislocations
VN–Mg complex
Comments
960
j 4 Extended and Point Defects, Doping, and Magnetism
4.4 Defect Analysis by Deep-Level Transient Spectroscopy Table 4.5 Dispersion in the reported deep levels in terms of
activation energy and capture cross section (the numbers in parentheses represent reference numbers). E1
E2
Activation energy (eV) 0.18 [276] 0.49 [276] 0.14 [278] 0.49 [278] 0.26 [280] 0.27 [282] 0.234 [283] 0.578 [283] 0.25 [286] 0.59 [286] 0.24 [289] 0.56 [289] 0.25 [294] 0.264 [277] 0.58 [277] 0.27 [279] 0.598 [279] Capture cross section (cm2) 3.9e18 [278] 4.9e17 [278] 1.3e15 [280] 8e15 [282] 4.2e16 [289] 4.2e16 [289] 2.9e15 [277] 5e15 [294] 1.6e15 [277] Trap concentration 7e13 [276] 1.72e14 [278] 1.6e13 [280] 7.7e14 [283] 2.5e13 [286] 9e13 [289] 5e14 [294] 1.3e14 [277]
(cm3) 6.3e14 [276] 2.48e14 [278] 2.0e13 [280] 1.2e15 [283] 2.5e13 [286] 2e14 [289] 3.6e14 [277]
E3
E4
E5
1.44 [278] 0.62 [280] 0.65 [282] 0.657 [283] 0.81 [286] 0.82 [289] 0.85 [294]
1.07 [289]
0.665 [277] 0.67 [279]
2.4e14 [280] 1e14 [282]
1e15 [277]
5.8e12 [278] 3.9e15 [289] 3e14 [294]
1.6e13 [289]
3.34e15 [278] 3.8e14 [277] 1.5e14 [286] 6e14 [289] 2e15 [294]
2.4e14 [289]
respectively. The optical cross sections, son ðhvÞ and sop ðhvÞ can be unambiguously separated and responses from different traps can be resolved from each other. Consequently, son ðhvÞ and sop ðhvÞ can be measured from their threshold up the energy gap of the semiconductor over a large range of temperatures. Coupling of DLOS with standard DLTS allows a clear identification of optical spectra with known levels and simultaneous determination of thermal and optical activities of each trap. Optically excited capacitance method has been applied to GaN. Each time electrons are excited from an otherwise neutral trap, the depletion depth is reduced and thus the capacitance is increased. Each time the photon energy resonates with a deep level, there is a marked increase in the capacitance. The photoemission transient capacitance technique resulted in the observation of much deeper defects as shown in Figure 4.82. Caution must be exercised as optical and thermal activation energies of defects vary substantially due to the Franck–Condon shift [299,300].
j 961
j 4 Extended and Point Defects, Doping, and Magnetism Wavelength (µm)
2.4
3
1.6
1.2
0.8
300
EOT (OL4) = 1.45 eV
250 Steadystatecapacitance (pF)
962
EOT (OL3) = 1.25 eV ∆C SS (OL3)
200
EOT (OL2) = 0.97 eV 150
∆CSS (OL2) 100
EOT (OL1) = 0.87 eV ∆CSS (OL1)
50
n-type GaN T = 150 K
0 0.6
0.8
1.0
1.2
1.4
1.6
Photon energy (eV) Figure 4.82 Photo-capacitance versus optical excitation energy in GaN Schottky diodes [298].
Applying standard DLTS to n-type Schottky and pþ/n GaN diodes and varying the filling pulse width along with the Fourier transform analysis, Soh et al. [295,301] determined the capture kinetics of the deep levels in OMVPE-grown GaN epilayers in the energy range accessible by thermal DLTS. An electron trap level was detected in the range of energies at Ec Et ¼ 0.23–0.27 eV with a wide ranging capture cross section on the order of 1 · 1019–8 · 1016 cm2 for both pþ/n and n-type Schottky GaN diodes. Two traps with lower energy were detected at 0.10–0.11 and 0.17–0.18 eV, with capture cross sections of 3 · 1019 and 8 · 1018 cm2, respectively. Another trap was detected at 0.4 eV, 5 · 1017 cm2. Two other common electron traps were reported at Ec Et ¼ 0.53–0.63 and 0.79–0.82 eV. An electron trap at Ec Et 1.07 eV with a capture cross section of sn ¼ 1.6 · 1013 cm2 was also detected in the Schottky diode samples. Also, a prominent hole trap level was observed at Et Ev 0.85 eV in pþ/n diode samples. The 0.17, 0.23, and 0.56 eV electron trap levels have been attributed to the extended defects based on the observation of logarithmic capture kinetics. However, the trap at 0.53–0.63 eV saturated at 0.02 s in one sample and did not show signs of saturation in another. The capture barrier was estimated as 50–60 meV for all three traps that had an amplitude that depended on filling pulse width. The traps at 0.1 and 0.4 eV could be saturated at shorter filling pulse widths, 10 ms and 1 ms. Saturation at 10 ms would be commensurate with a point defect. The 0.4 eV trap only appeared in Si-doped samples, and the change in amplitude was attributed to segregation of Si along dislocation core sites.
4.4 Defect Analysis by Deep-Level Transient Spectroscopy
A hole trap has been noted in OMVPE-grown n-type hexagonal GaN (Schottky diodes) using minority carrier transient spectroscopy (MCTS) [292]. Minority carrier transient spectroscopy uses an optical filling pulse. The reported ionization energy and capture cross section deduced from Fourier transform transient spectroscopy were 0.81 eV and 2 · 1014 cm2, respectively. Such a capture cross section for holes indicates an attractive potential and hence a negatively charged center before the hole capture. This deep center is a point defect, still negatively charged after a hole capture because it repels electrons, and hence it is a deep acceptor. OMVPE-grown GaN p–n junction diodes on sapphire substrates have investigated [302]. Three electron trap levels with energy levels E1 ¼ 0.59 e V, E2 ¼ 0.76 e V, and E3 ¼ 0.96 e V were observed. The trap concentration NT of the observed levels E1, E2, and E3 were determined to be 2.1 · 1014, 5.3 · 1014, and 17.1 · 1014 cm3, respectively. The corresponding values of the capture cross section determined from the extrapolation of the emission rate signatures data of the aforementioned levels are 2.8 · 1018, 1.5 · 1017, and 7.7 · 1016 cm2. The majority carrier trap E2 completely disappears while levels E1 and E3 undergo systematic reductions of their concentration during repetitive deep-level transient Fourier spectroscopy (DLTFS) scans under electric field. This effect is correlated to the transformation of charged defects into electrically inactive ones, with the help of at least two mechanisms. E2s disappearance leads one to suspect that it may be a NI, which is predicted to be thermally unstable. E3s partial reduction may indicate that it has a NI component, perhaps VGa–NI. Optical DLTS has also been applied for hole trap characterization in n-type ELOgrown GaN [294]. A 380 nm UV-LED with subbandgap energy of 3.27 eV was used for optical excitation. Two hole trap levels at 0.25 and 0.85 eV with respect to the valence band were found in concentrations of 5 · 1014 and 2 · 1015 cm3, respectively. These concentrations are almost 2 orders of magnitude higher than those of the electron traps present in n-GaN. Two electron traps at 0.27 and 0.61 eV were observed with concentrations of 1.6 · 1013 and 3 · 1013 cm3, respectively. After proton irradiation, the concentration of the hole trap at 0.85 eV decreased from to 4 · 1014 cm3. This is a surprising result, considering that the 0.85 eV defect is thought to be due to the VGa, which should not decrease in concentration after irradiation. Deep centers in Si-doped n-GaN layers grown by reactive molecular beam epitaxy (RMBE) have also been studied by DLTS along with their dependence on growth conditions [283,288]. Five deep-level defects [283] have been observed with activation energies E1 ¼ 0.234 0.006 eV, E2 ¼ 0.578 0.006 eV, E3 ¼ 0.657 0.031, E4 0.961 0.026, and E5 ¼ 0.240 0.012 eV. For comparison, defects E1, E2, E4, and possibly E3 have been observed in authors laboratory in RF MBE grown GaN. Among these, the levels labeled E1, E2, and E3 are interpreted as corresponding to deep levels previously reported in n-GaN grown by both HVPE and OMVPE [283]. Si-doped GaN samples grown on Si-doped nþ GaN buffer layers at 800 C showed a dominant trap C, with an activation energy of ET ¼ 0.44 eV and a capture cross section of sT ¼ 1.3 · 1015 cm2, while samples grown at 750 C on undoped semi-insulating GaN buffer layers, which in turn were grown at relatively low temperatures, exhibited the prominent traps D and E, with ET ¼ 0.20 eV and sT ¼ 8.4 · 107 cm2, and ET ¼ 0.21
j 963
964
j 4 Extended and Point Defects, Doping, and Magnetism eV and sT ¼ 1.6 · 1014 cm2, respectively. The trap E1 was assumed related to a Nvacancy defect because the Arrhenius signature of E1 is very similar to the trap E that has been produced by 1-MeV electron irradiation. This irradiation has been shown to induce N-vacancies in GaN grown by both OMVPE and HVPE [281]. Trap C has a strong increase in emission rate with increasing electric field, indicating that it is a donor trap, associated with dislocations, as determined by the linear increase in emission amplitude with logarithm of filling pulse width [268]. However, its concentration increases near the surface, which may be an effect of the chemical potential during growth, surface band bending, and defect formation energy. To uncover the genesis of defects and/or test the resistance of the material to various high energy irradiations, samples are often subjected to such radiation and studied afterward for new defects or modification of defects. g-Irradiation induced defects in slightly and heavily doped GaN grown by low-pressure OMVPE on (0 0 0 1) sapphire substrates has been studied [303]. The g-irradiation decreases the electron concentration for the slightly doped layers, whereas it increases the concentration for the heavily doped epilayers. By means of DLTS measurements, three electron traps were identified before g-irradiation in slightly doped (n 1017 cm3) n-GaN samples: E1 ¼ 0.25 eV, E2 ¼ 0.59 eV, and E3 ¼ 0.81 eV with concentrations of 2.5 · 1013, 2.5 · 1013, and 1.5 · 1014 cm3, respectively. After the g-irradiation, the concentration of the E2 centers increased to 4.3 · 1014 cm3 and also two new centers appeared: E4 ¼ 0.155 eV and E5 ¼ 0.95 eV with concentrations 1016 and 4 · 1016 cm3, respectively. As has been the case in the investigation of evolution of extended defects by TEM as the growth progresses (discussed in Section 4.2), the evolution of deep levels has also been studied. The very thick nature of HVPE layers lends themselves to studies of the evolution of deep levels as a function of distance away from the epitaxial layer substrate interface. The HVPE GaN layers used in the DLTS study were grown on ZnO-pretreated sapphire in a chloride-transport HVPE vertical reactor at thicknesses of 5, 11, 15, 39, and 68 mm. Details of the GaN growth and the sapphire pretreatment techniques for this particular work have been described elsewhere [156], but they are most likely similar to those discussed in Sections 3.4.1.1. Ni/Au Schottky contacts with a diameter of 250 mm, surrounded by large-area Ti/Al/Ti/Au Ohmic contacts were fabricated and tested using a Bio-Rad DL4600 system with a 100-mV test signal at 1 MHz [155]. To determine the activation energy ET of the deep centers, the DLTS spectra were taken at rate windows, ranging from 0.8 to 200 s1, and were analyzed using standard Arrhenius plots. Carrier concentrations in the top region of the five samples, determined by C–V measurements, were found to be in the 1017 cm3 range; these results are similar to those found by Hall effect measurements that have been corrected for the degenerate interfacial layer [304]. As the sample thickness decreased from 68 to 5 mm, the apparent (average) mobility dropped from 740 to 190 cm2 V1 s1. For the three thickest samples (68 13 mm), the apparent carrier concentrations were found to increase from 1.3 · 1017 to 2.6 · 1017 cm3. As for the Schottky barriers, the reverse leakage current is significantly reduced, by almost 3 orders of magnitude, as the layer thickness increased from 5 to 68 mm. The soft breakdown and increased leakage
4.4 Defect Analysis by Deep-Level Transient Spectroscopy
6
DLTS signal (10 14 cm-3 )
5
4
a:68 µm, n = 1.3 × 1017 cm-3 b:39 µm, n = 1.8 × 1017 cm-3 c:15 µm, n = 2.6 × 1017 cm-3 d:11 µm, n = 1.0 × 1017 cm-3 e:5 µm, n = 1.5 × 1017 cm-3
Vb = -3.0 V, Vf = 0 V, Wf = 1 ms, en = 4 /s–1
A1
B A
D
3
C
2
1
e d c
Ax
b
0 50
a
100
150
200
250 300 Temperature (K)
350
400
450
Figure 4.83 Typical DLTS spectra for five HVPE GaN samples with different thicknesses. The carrier concentration of each sample is shown in the legend. Courtesy of Z. Fang and D.C. Look.
current in thin films are most likely caused by carrier tunneling or hopping through defect states possibly induced by threading dislocations. The ideality factor also was indicative of the enhanced quality of the films in that it approached unity for the thickest layer. Typical DLTS spectra for all five samples are presented in Figure 4.83. Note that the spectra in the figure have different base lines, separated by 0.5 · 1014 cm3. Salient features are (i) the dominant centers are Ax (ET 0.72 eV), A (ET ¼ 0.67 eV), and B (ET ¼ 0.61 eV), with concentrations in the 1013–1014 cm3 range for the two thickest samples (see curves a and b); (ii) for the 15-mm sample (see curve c), a center with ET ¼ 0.20 eV (labeled as D) appears at low temperatures; (iii) for the 11-mm sample (see curve d), in addition to a center with ET ¼ 0.17 eV (also labeled as D) at low temperatures, the center A1 (ET ¼ 0.89 eV), with a higher concentration in the 1014–1015 cm3 range, arises; and (iv) for the 5-mm sample (see curve e), B becomes a prominent center and A1 drops significantly, as compared to its value in the 11-mm sample, and the other observed centers are A, C (ET ¼ 0.41 eV), and D (ET ¼ 0.23 eV). Thus, one can state that as the HVPE GaN samples become thinner, both the species and concentrations of the deep centers increase. However, the trend is not monotonic. For example, the concentration of trap D appears to reach a maximum at 11 mm, and traps A and B were lowest for intermediate thicknesses. The effect of thicker layers was to decrease the dislocation density. A reduction in dislocation density by 2 orders of magnitude did not have a proportional change in trap concentration. Interestingly, most of the deep centers found in HVPE GaN have also been observed by us in n-GaN
j 965
j 4 Extended and Point Defects, Doping, and Magnetism C
2.0
DLTS signal (1014 cm-3)
966
Vf = 0V, Wf = 1 ms, en = 80 s –1
SBD on Ga face Vb = -1.0V Vb = -8.0V
1.5 A B
1.0
B' D
0.5
0 100
150
200
250
300
350
400
Temperature (K) Figure 4.84 DLTS spectra for the Ga face of the freestanding GaN template measured as a function of temperature for a series of filling pulse widths (Wf ) at 0 and 8 V reverse bias voltages. Courtesy of Z. Fang and D.C. Look.
grown by other techniques, such as OMVPE and RMBE. Depending on the material parameters such as the layer thickness, electron mobility and dislocation density, the concentrations of these deep centers can be as low as 1013–1014 cm3 for a 4.5 mm thick OMVPE GaN with an electron mobility of 765 cm2 V1 s1 and an estimated dislocation density of low-108 cm2. They can be as high as 1015–1016 cm3 for 0.5 mm thick RMBE GaN layers, with electron mobilities below 250 cm2 V1 s1 and dislocation densities in the 109–1010 cm2 range [318]. In contrast, 200 mm thick freestanding HVPE GaN layers have demonstrated trap concentrations in the 1012 cm3 range, as seen previously in Figure 4.84, and discussed further in the next paragraph. Much thicker GaN in the form of freestanding Samsung GaN templates with both Ga polarity and the N polarity have also been investigated by DLTS [160]. Due to large defect concentration by virtue of growth and mechanical polish, it was necessary to remove some 30 mm of the material from the N face by wet chemical etching to obtain reasonable data. This amount was found to be sufficient based on the Hall measurements. On the contrary, the Ga-face just simply underwent a standard cleaning procedure before fabrication because it had previously been etched with reactive ion etching (RIE). As will be described in detail below, this surface exhibited defects that are nominally attributed to plasma damage caused by the RIE etching. Typical DLTS spectra for a Schottky barrier diode obtained from the Ga-polarity sample measured as a function of temperature for a series of filling pulse widths (Wf)
4.4 Defect Analysis by Deep-Level Transient Spectroscopy
and 0 bias and 8 V reverse bias voltages are shown in Figure 4.84. Compilation of these data and those obtained in samples with thicknesses in the range of 5–68 mm showed six deep traps (labeled as A1, A, B, B0 , C, and D) were observed in the temperature range from 80 to approximately 425 K. It should be mentioned that the A1 trap appears in thinner HVPE films, see Figure 4.83, and RMBE-grown films. Three of these traps, A1, A, and B0 , show significant increases in peak height as Wf increases from 0.2 to 20 ms. The fingerprints of these traps, that is, ET and sT, as determined from the Arrhenius plots of T2/en versus 1/T, are ET ¼ 0.70 eV and sT ¼ 1.7 · 1015 cm2 for A; ET ¼ 0.58 eV and sT ¼ 2.4 · 1015 cm2 for B; ET ¼ 0.53 eV and sT ¼ 1.5 · 1015 cm2 for B0 ; ET ¼ 0.35 eV and sT ¼ 1.6 · 1015 cm2 for C; and ET ¼ 0.25 eV and sT ¼ 1.2 · 1015 cm2 for D. Traps A1, A, B, and D are commonly observed in epitaxial GaN grown by OMVPE, RMBE [281], and HVPE [318]. A significant increase in the DLTS signal as a function of Wf, as found for traps A1, A, and B0 , usually suggests a small capture cross section and/or a trap behaving as a line defect that is typically associated with threading dislocations. Such a behavior in connection with traps C1 and A1 in thin RMBE GaN layers has been reported [281]. The prominence of trap C found in the Ga-face sample might be related to the surface damage caused by RIE, because this center can be detected only in the top 0.4 mm, and is not present at all in the N-face sample that was wet chemical etched to remove any damage present on the surface. Interestingly, the behavior of trap C is very similar to that of a sputter deposition induced trap ES3 in OMVPE-grown n-GaN, reported by Auret et al. [305]. What was considered a new trap B0 was found not only near the surface but also deeper in the sample. This trap has a small capture cross section, and seems to be affected by the strength of the electric field in the depletion region, indicating a donor state. For several of the N-polarity Schottky barriers measured at temperatures below 200 K, trap D, but not trap C, was observed. However, in some of the N polarity Schottky barriers, trap E1 was found, which is shown in Figure 4.85 as a function of filling pulse width, Wf. The trap energy ET and capture cross section sT for trap E1 were determined to be 0.18 eV and 4 · 1017 cm2, respectively. The much lower value of the capture cross section for trap E1, as compared to that of trap D, can explain the strong influence of Wf on the peak height of trap E1. Trap E1 was also observed in thin GaN films grown at 750 C by RMBE, as previously reported [281]. Through comparisons of the DLTS spectra observed in the as-grown RMBE GaN and electron-irradiated OMVPE GaN, trap E1 is very similar to the electron irradiation induced trap E, and is thus believed to be a complex involving the N vacancy. This assignment is supported by the fact that ammonia, the N source in RMBE growth, is rather stable at the low growth temperatures employed, and thus N vacancies might be expected. Moreover, the observation of trap E1 in the N-face Schottky barriers of this study might be an indication of nitrogen deficiency in the early stages of HVPE growth. Reiterating what was mentioned in the TEM section (Section 4.2) dealing with the HVPE films with varying thicknesses, the dislocation density increases from 1 · 108 cm2 in the top region (at t ¼ 55 mm) up to 3 · 108 cm2 in the middle region (at t ¼ 20 mm), then up to 1.6 · 109 and 1 · 1010 cm2 in the region close to the interface
j 967
j 4 Extended and Point Defects, Doping, and Magnetism 6
DLTS signal (1013cm-3)
968
E1
SBD on N face
Wf = 50 ms Wf = 10 ms Wf = 2.0 ms Wf = 0.5 ms
4
Vb= -1.0 V, Vf = 0.01 V, en = 10 s –1
2 D
0 80
100
120
140
160
180
200
Temperature (K) Figure 4.85 DLTS spectra for the N face of the freestanding GaN template measured as a function of temperature for a series of filling pulse widths (Wf ) at 0.01 and 1 V reverse bias voltages. Courtesy of Z. Fang and D.C. Look.
(at t ¼ 3 and 1 mm, respectively). From a cross-sectional TEM image for the nominally 5 mm thick sample (not shown), the dislocation densities at t ¼ 4.5, 3, and 1 mm, are 9 · 108, 1.7 · 109, and 4 · 109 cm2, respectively. The results obtained in the 5 mm thick sample are very similar to those obtained in the interfacial region of the 55 mm thick sample. To summarize this section, the deep-level concentration, the leakage current in Schottky barriers, and threading dislocation density decrease toward the surface, which in turn decreases with increasing layer thickness. By analogy, one can argue that the deep-level density decreasing away from the interface is correlated with dislocations, many of which are either uncoordinated and therefore charged or even when coordinated the strain field generated can cause defect to be attracted to them. Threading dislocations in GaN are electrically active and cause defect states that act as traps and nonradiative recombination centers. This has been demonstrated by studies of scanning capacitance microscopy [117] and plan-view TEM along with cathodoluminescence imaging [306]. Building on earlier formalism developed for charge dislocations in Ge [307,308] and later in GaN [309,310], a theory of charged dislocation line scattering was developed within the framework of the Boltzmann transport equation [311] that describes the temperature-dependent Hall effect data in GaN, the details of which are discussed in Volume 2, Chapter 3. Threading dislocations in GaN have been predicted to be of the VGa or VN nature in their core structures, depending on doping (or type) and growth stoichiometry. The formation
4.4 Defect Analysis by Deep-Level Transient Spectroscopy
of the VGa structure is favored under N-rich conditions and is most stable in n-type material, whereas the formation of the VN structure is favored under Ga-rich conditions and is most stable in p-type material [126]. Because of the uncertainty involved in the formation energy calculations, a mixture of core structures is also possible. Thus, edge dislocations may behave as hole traps in n-type GaN and may act as electron traps in p-type GaN, depending on growth conditions. Regarding the possible point defect nature of the main centers found in n-GaN, recall two important experimental observations, namely, that high concentrations of deep centers in n-GaN are closely associated with high dislocation densities, regardless of the growth technique and that there exists an anticorrelation between centers A1 and B for the two thinnest samples investigated. The anticorrelation relationship was also observed in other n-GaN materials (e.g., for thin OMVPE GaN [312] and for thin RMBE GaN [313], which means that a strong increase in trap B is generally accompanied by a significant reduction of trap A1. The center B (also called E2 in the literature), as one of the highest concentration traps, has been extensively studied, resulting in the suggestion that the center could be due to chemical impurities such as C [314] or Mg [280]. In another investigation, the photoionization of E2 in n-GaN was characterized by using capacitance transient spectroscopy, which yielded an optical activation energy of Eo ¼ 0.85 eV at 90 K and resulted in a Franck–Condon shift [315,316] (the difference between optical and thermal energies needed to excite the electron from the trap level to the conduction band in the coordinate diagram) of 0.3 eV, indicating a possible defect nature from theoretical predictions of relaxations [287]. A study suggested that E2 could be of NGa nature (see Section 4.3.1.2.2 for a theoretical discussion of its likely energy states) and can effectively be suppressed by isoelectronic In doping [317]. The relaxation leading to a Franck–Condon shift should also be revealed as a capture barrier, evident from filling pulse width experiments in DLTS. However, there are at least two traps in this energy range, which complicates the analysis. As presented earlier, measurements from HVPE, MBE, and OMVPE have shown that a capture barrier exists only for the MBE sample. To gain a better insight into the nature of A1, refer to the electron irradiation (EI) induced centers in n-GaN. In OMVPE GaN layer after 1-MeV EI, trap E (at 120 K), which consisted of two components, each of which with a thermal activation energy of 0.06 eV and trap A2 (at 420 K), with ET ¼ 0.85 eV appeared. It is evident that E is related to the nitrogen vacancy (VN), and because A2 shows a production rate close to that of E, it is possible that A2 might be a NI-related defect, created by the reaction nitrogen, leaving the substitutional site and creating an interstitial (NN ! VN þ NI) [318]. This predicts that E should increase and A2 should decrease as the N supersaturation decreases. In addition to the aforementioned observations, some preexisting centers in OMVPE GaN, including B, are not affected by 1-MeV EI at all, ruling out the possibility that defect B is related to the VN defect. Because center A1 is very similar to the EI-induced center A2 and B shows an anticorrelation with A1, A1 is tentatively assigned to a NI-related defect and B to NGa. NGa can be easily formed by the reaction VGa þ NI ! NGa. According to the predictions mentioned above, VGa could be present in n-GaN having a high dislocation density, which has been observed
j 969
j 4 Extended and Point Defects, Doping, and Magnetism 10 6
A2 A
1
As-grown A1
A
EI-induced A2
10 5
As-grown E1 EI-induced E
T 2/en(K2 s)
970
10 4
B
C1
C D
E1
10 3
E
10 2
10 1 2
3
4
5
6
7
8
9
10
1000/T (1/K) Figure 4.86 Arrhenius plots of T2/en for as-grown and electron beam induced deep centers in n-GaN. Courtesy of Z. Fang and D. C. Look.
in undoped OMVPE GaN by positron annihilation spectroscopy [319]. The center D, with ET ¼ 0.17–0.23 eV, also shows a correlation with the thickness reduction or the dislocation increase, which means that D, as an electron trap, could possibly be a defect complex involving VGa (such as VN–VGa). This predicts a decrease in B and an increase in D as the nitrogen supersaturation decreases. As a summary, Arrhenius plots of T2/en for as-grown and electron beam induced deep centers in n-GaN are shown in Figure 4.86. The results of this section are summarized in Table 4.6. A notorious problem with wide bandgap nitrides is that attempts to accomplish p-type doping often results in creation of structural and electrically compensating defects. Minimization of these defects is imperative and determines whether p-type doping can be successfully attained. Although the free hole concentration obtained in p-type GaN has steadily increased over the years, there persists a problem with p-type GaN being more defect prone. Shown in Figure 4.87 are the DLTS spectra obtained in films with and without Mg doping, albeit in the early stages of GaN development. It is clear that more investigations are needed to bring understanding of p-type GaN on par with n-type GaN in terms of deep levels. As can be clearly seen, the E2 line increases remarkably with addition of Mg. Similar effects are seen for the other defects, albeit not to the same extent. 4.4.3 Dispersion in DLTS of GaN
Having simply reported what has been observed in authors laboratory and reported in the literature in regard to DLTS active deep levels in GaN, we now turn our
D
E E1
0.35 in freestanding GaN 0.17–0.23
4 · 1017
1.2 · 1015
1.6 · 1015
1.6 · 1015
1.5 · 1015
2.4 · 1015
1013–1014 cm3
1.7 · 1015
1013–1014 cm3
1014–1015 cm3
Concentration (cm3)
Capture cross section, r (cm2)
EI stands for electron irradiation. Compiled from the data of Z. Fang and D.C. Look.
0.18
0.25 in freestanding GaN
0.35–0.41
0.72 0.61 0.58 in freestanding GaN 0.53
0.67 0.7 in freestanding GaN 0.89
Energy level (eV) below Eg
C
B1
A1 A2 Ax B (E2)
A
Defect
Table 4.6 Summary of the defect states observed in GaN and their nature.
VN Related to VN
Similar to sputtering-induced defect
Surface damage on Ga face of freestanding GaN
NGa not VN
NI NI
Not affected By EI
Nature
Induced by EI As-grown HVPE Induced by EI too In N face of HVPE by Samsung
Not affected by EI Present in all GaN High in thin films
Surface damage in Ga face of Samsung Not affected by EI Present in all GaN except on N face Samsung HVPE
High in thinner films Induced by EI 68 mm HVPE High in thin films Present in all GaN
Not affected by EI Present in all GaN
Comments
4.4 Defect Analysis by Deep-Level Transient Spectroscopy
j 971
j 4 Extended and Point Defects, Doping, and Magnetism 0.02
2×
0.00 E1
DLTS signal (pF)
972
Mg-a level
–0.02
–0.04
–0.06
Undoped Mg-a Mg-b
–0.08
E2
–0.10 50
100
150
200
250
300
350
400
Temperature (K) Figure 4.87 DLTS spectra of undoped and Mg-doped GaN layers [277].
attention to the unusually large dispersion in measured energy and capture cross section of defects. This can result simply from the vibrational and electronic degeneracy entropy term, electric field effect through Poole–Frenkel effect, measurement conditions such as nonsaturating filling pulse, overlapping trap emission signals, whether energy levels are obtained from DLTS data or fitting of the transient, and of course the sample quality and variability. If a strong Poole–Frenkel effect is present due to a local electric field, the trap would have a distribution of energies in the depletion region, as already mentioned above, making fitting of the transients unreliable or shifting the rate window peaks to lower temperature. Conversely, shortfilling pulses will truncate the low temperature side of the rate window peaks, making the apparent peak appear at a higher temperature. Another source of spread in measured values would come from the entropy term. The point of unacceptably large dispersion in the reported trap data can be displayed in imagery also, in addition to that the ones is shown in Tables 4.4 and 4.5. Figure 4.88 shows various trap levels reported in the literature for GaN in the unusual coordinate system of the trap energy and the capture cross section with trap energy along the horizontal axis and capture cross section along the vertical axis. The figure illustrates that there is a distribution in the trap activation energies, as opposed to all the data converging on certain activation energies associated with certain DLTS active traps. Second, for a given trap energy that presumably corresponds to the same trap, there is a distribution in the capture cross sections. Again, Figure 4.88 together with
4.4 Defect Analysis by Deep-Level Transient Spectroscopy 1E-8 1E-9
GaN DLTS trap signatures
1E-10
1E-12 1E-13
0.0035 0.0030
1E-14
0.0025 NT/Ns
s (cm2)
1E-11
1E-15
0.0020 0.0015 0.0010
1E-16
0.0005
1E-17
0.0000 0
1E-18 0.0
0.2
0.4
0.6
0.8
1.0
100
1.2
200
1.4
300 400 500 Temperature (K)
1.6
1.8
600
2.0
700
2.2
2.4
ET (eV)
Figure 4.88 Various trap levels reported in the literature for GaN in the coordinate system of the trap energy and the capture cross section with trap energy along the horizontal axis and capture cross section along the vertical axis. The point is that the dispersion in the reported data is quite substantial as highlighted in Tables 4.4
and 4.5. The inset shows the spectra for one of the samples. (&) Auret et al. [320]; (*) Auret et al. [321]; (~) Johnstone et al. [322]; (!) Polyakov et al. [323]; (^) Lee et al. [314]; (·) Soh et al. [289]; (þ) Hacke et al. [280]; ( ) Cho et al. [324]; (|) Soltanovich et al. [325]; (—) Hacke et al. [277]; (&) Wang et al. [283].
Tables 4.4 and 4.5 points out the dearth of accurate, or complete, measurements that can be used for correlation with theoretical predictions covered in Section 4.3. The bias dependence, electronic degeneracy, and capture barriers can all contribute significant scatter to the measured deep-level characteristics. Each of the traps has reported energies that span a considerable range, for example, 0.5–0.8 eV, for the trap normally observed at 350 K, and capture cross sections that span several orders of magnitude! Naturally, the question arises as to the nature of the aforementioned dispersion/ scatter. There are several possibilities for the sources of variation in reported characteristics. Variations in measured energy and capture cross section can result from vibrational and electronic degeneracy entropy term, Poole–Frenkel effect and measurement conditions such as on-saturating filling pulse, and overlapping trap emission signals. For example, if a strong Poole–Frenkel effect is present due to a local electric field, the trap will have a distribution of energies in the depletion region, as already mentioned above, making fitting of the transients unreliable or shifting the rate window peaks to a lower temperature. Conversely, short-filling pulses truncate the low temperature side of the rate window peaks, making the apparent peak appear at a higher temperature. Another source of spread in measured values comes from the entropy term. A practical limitation on the accuracy is imposed by the limited range of time over which the emission rate can be measured, leading to inaccuracies in
j 973
974
j 4 Extended and Point Defects, Doping, and Magnetism slope and capture cross section. In general, as the energies calculated from the slope decrease, so does the capture cross section, calculated from the intercept with the vertical axis on the Arrhenius plot. This is seen in Figure 4.88 as the linear relationship in the capture cross section versus trap energy. The Poole–Frenkel effect may be the primary cause of the spread in experimental values for donor defects. The Poole–Frenkel effect reduces the energy for emission as a result of the electric field tilting the bands in which a Coulombic defect potential resides. The change in energy with electric field is given by Equation 4.41. As previously stated, with regard to Equation 4.41, for a defect that is singly charged after emission, the shift in energy will be 0.11 eV for a field of 200 kV cm1. The apparent trap energy at this peak electric field would be reduced by 0.11 eV, and the trap energy would approach the true value near the depletion zone edge. For a doubly charged defect after emission, the maximum reduction at 200 kV cm1 would be 0.15 eV. One solution to alleviate smearing of the emission transients is to confine the emission to be only from traps in a narrow range of electric field using double-pulse DLTS, where the capacitance signal is from the difference in response from two slightly different filling pulses [267]. In fact, the electric-field effect on traps A1 and C1 and anomalous capture kinetics of the trap C1 has been pointed out [288] in addition to more anomalous capture kinetics behavior [326]. In the latter report, overlapping energies of point defect related traps in freestanding GaN and line defect related traps in thin epi-GaN were noted. Overlapping D and E trap energies were also noted in RMBE-grown GaN. Similar filling pulse width effects have been observed for trap E1 in both RMBE GaN and N-face freestanding GaN, see Figure 4.85. A comparison of DLTS spectra and Arrhenius plots for three different samples show not only some dispersion in the fingerprints for the traps but also the similarity of traps (by their peak positions) in n-GaN, grown by various techniques [327]. Large trap C found in freestanding GaN measured at low bias, Figure 4.84, is related possibly to surface damage, the extent and nature of which must be considered in trap evaluation. It is expected that much of the dispersion under discussion would dissipate as the sample quality is improved further. Some early evidence of this can be found in the report by Pernot et al. [328] in which OMVPE GaN, ELO GaN, and freestanding GaN samples were investigated, which points to a global agreement with the cumulative work of Fang et al. [326]. In the case where several traps are present with overlapping signals in a given temperature range, the shallow carrier concentration term can be separated and the time-dependent terms can also be expressed as individual terms. This can be rewritten as done in Equation 4.30. A few DLTS systems use this method, rather than the rate window method of picking the temperature from the peak position and using the emission rate as defined by two sampling windows. Collecting the transient over two to three decades allows fitting for multiple exponential components according to Equation 4.30 and accurate energy and capture cross-section measurements. To show the effect of rate window, Figure 4.89 is presented that contains rate window plots for a GaN epitaxial layer at three different rate windows. The trap at 325 K appears as a single peak (for each rate window), but fitting decomposes the peak into two traps. Figure 4.90 shows the fit for the transient, with comparison for one and
4.4 Defect Analysis by Deep-Level Transient Spectroscopy t2 / t1 0.005 s/0.001 s 0.01 s/0.005 s 0.05 s/0.01 s
0.25
C (pF)
0.20 0.15 0.10 0.05 0.00 50
100
150
200 250 300 Temperature (K)
350
400
Figure 4.89 Three rate window spectra for traps in GaN. Each rate window plot shows three distinguishable peaks. Rate window analysis uses the time constant, calculated from t2 and t1, and the peak temperatures to calculate the energies and cross sections. More accurate evaluation is performed by fitting the transients, which decomposes the peak at 325 K to two traps.
two exponential components. The fit for two components has a mean square error nearly an order of magnitude better. A rate window analysis would only detect the single trap, which would be distorted by the second underlying trap. Figure 4.89 exhibits trends related to the individual peaks seen in a DLTS rate window plot. There are typically three dominant peaks found in reported rate window
Figure 4.90 Multiexponential curve fitting for the capacitance transient data. The fit for two exponential components is nearly an order of magnitude better than for one exponential component.
j 975
976
j 4 Extended and Point Defects, Doping, and Magnetism plots of GaN, where a representative rate window plot linking the peaks to the locus of signatures is shown in the inset. The figure suggests that there is a coupling between the deep-level energy and the capture cross section. Note that there are exceptions, where two peaks are observed in the same spectrum that give energies and capture cross sections on the same locus of points. Given the other factors mentioned above that can influence the energy and capture cross section, each trap must be thoroughly characterized to obtain clues to their origin. Additional experimental variables must be taken into account, such as the bias dependence and effect of filling pulse width on transient amplitude. GaN samples prepared by OMVPE on sapphire substrates have been investigated by DLTS by several groups [276–278]. A more inclusive list of typical defect states termed as E1, E2, E3, E4, and E5 are tabulated in Table 4.5 with E1 and E2 appearing in ICTS [276]. Energies of the aforementioned defects, in the same order, have been determined to be in the range of 0.14–0.27, 0.49–0.598, 0.62–0.67, 0.81–0.85, and 1.07–1.44. Roughly these data points fall in the three sets of points when plotted in a field of the capture cross section versus the trap energy as shown in Figure 4.88. The concentrations of the same defects, again in the order listed, are in the ranges of 1.6 · 1013–7.7 · 1014 cm3, 2 · 1013–1.2 · 1015 cm3, 1.6 · 1014–3.8 · 1014 cm3, 1.5 · 1014–2 · 1015 cm3, and 2.4 · 1014–3.34 · 1015 cm3 for E1, E2, E3, E4, and E5 traps, respectively. Finally, for a quick illustration of various defect states detected in DLTS, photoemission transient capacitance methods are schematically shown in Figure 4.91.
Conduction band 0.18 eV DLTS
Photoemission. cap. trans.
0.563 eV 0.66 eV 0.87 eV 0.97 eV
E1 E2
0.269 eV
E3
0.59 eV 0.68 eV
1.25 eV 1.45 eV
2.4 eV 0.85 eV DLTS Hole traps
0.81 eV
0.25 eV
Valence band Figure 4.91 Various deep levels within the gap energy of GaN grown by OMVPE, as observed by DLTS and photoemission transient spectroscopy. The two hole trap energies are with respect to the valence band maximum.
4.5 Minority Carrier Lifetime
4.4.4 Applications of DLTS to AlGaN, In-Doped AlGaN, and InAlN
Although AlGaN forms the barrier layer in both GaN-based optical and electronic devices, the relatively poor quality of AlGaN among others might be the reason why this ternary has not been investigated in much detail in terms of its deep-level characteristics. Because of reduced surface mobility of Al and its predilection for O, among possibly myriad of other reasons, the nonradiative recombination center density increases with Al mole fraction as well as DLTS active peak(s) getting deeper in energy. A dominant DLTS active peak in Al0.28Ga0.72N (grown as part of a multilayer AlGaN/GaN structure) has been reported [329] with an activation energy of 0.9 eV, appearing approximately at 375 K in the spectrum. A bulk Al0.26Ga0.74N measured in authors laboratory showed the dominant peak at 450 K with an activation energy of 1.04 eV at a density of NT ¼ 2.5 · 1015 cm3. What is remarkable, however, is that doping AlGaN with isoelectronic In reduces the deep-level density and improves the layer quality as judged from LED and FETdata. One such In-doped Al0.37Ga0.63N layer measured by DLTS showed a dominant peak at 1.54 eV but with NT ¼ 9 · 1014 cm3, which is below that for the NT ¼ 2.5 · 1015 cm3 layer without In doping, despite the much increased Al mole fraction that otherwise degrades the results. On the contrary, In0.18Al0.72N lattice matched to GaN (with a bandgap of about 4.5 eV that corresponds to approximately that of Al0.5Ga0.5N) showed remarkably different deep-level spectra with two peaks appearing at 170 K and a dominant one at 285 K. The former has an activation energy of 0.17 eV and the latter has 0.67 eV. The deeper and dominant trap with a concentration of NT ¼ 9.45 · 1014 cm3 is much shallower in addition to being lower in concentration as compared to AlGaN that bodes well for both optical and electronic devices. Although the density of the dominant trap in In0.18Al0.72N is similar to that for In-doped Al0.37Ga0.63N layer, its activation energy is much smaller, 0.67 eV, as opposed to 1.54 eV.
4.5 Minority Carrier Lifetime
Minority carriers are rather captured by defect, easily reducing their lifetime and diffusion length in the material. In addition to having important bearing on devices, the minority carrier lifetime can be used to assess the quality of the material. Because the TEM and DLTS results discussed above clearly indicated that the defects are reduced away from the interface, it is only logical to seek the value of the minority carrier lifetime as a function of distance away from the interface. The thick HVPE films provide an excellent medium for this investigation. In this vein, electron beam and optical depth-profiling of thick (5.5–64 mm) quasi-bulk n-type GaN samples, grown by HVPE, were carried out using electron beam induced current (EBIC), micro-PL and correlated to the TEM results discussed above [155]. To cap the properties of the HVPE films used, room temperature Hall measurements (after correction for the highly conductive 200 nm interfacial layer) showed electron
j 977
Direction of line scan e-beam
Au contact
j 4 Extended and Point Defects, Doping, and Magnetism Au contact
978
Sapphire substrate
ToPC
Figure 4.92 Schematic diagram of the experimental setup used for measuring electron beam induced current. Courtesy of L. Chernyak.
concentrations (at the surface) ranging from 8 · 1016 to 1.2 · 1017 cm3, and mobilities from 580 to 750 cm2 V1 s1, respectively, the larger and lower mobilities being associated with the thicker films. The conductive interfacial layer was modeled to have an electron concentration of mid-1019 cm3 and a mobility of about 50 cm2 V1 s1. EBIC measurements were carried out in situ in a scanning electron microscope (SEM) JEOL 6400F [330]. A cartoon of the experimental setup is shown in Figure 4.92. The accelerating voltage, used in these measurements, was varied from 10 to 20 kV. This corresponds to an electron range, R, of 0.36–1.20 mm, respectively [331,332]. After cleavage, vertical gold (Au) stripes (Schottky barriers) of different sizes were formed on one of the edges of the samples by Au evaporation and subsequent liftoff. The minority hole diffusion length, L, was derived from the line-scan EBIC measurements [333,334]. The measurements were carried out at distance, d, from the GaN/sapphire interface [330]. In these measurements the electron beam (positioned perpendicular to the sample edge) was moved from the vertical wall of the Au/ n-GaN Schottky barrier toward another Au-contact. L can be obtained from the EBIC current decay versus distance from the edge of the Schottky contact (for distances > 2L). The details of the diffusion length extraction can be found elsewhere [333,334]. Because the samples used met the condition of thickness 4L, no sample thickness correction in the measurements L was warranted as the sample could be construed as bulk [335]. Likewise, because the ratio R/L 4, no EBIC resolution limitations would be expected [336]. Micro-PL profiling across thick GaN samples was performed in situ in the microphotoluminescence setup [330]. The micro-PL measurements were carried out using a 4 ns pulse 337 nm nitrogen laser with a power of 100 mJ per pulse. The laser beam focused to a 1.5 mm diameter spot allowed probing of a region comparable to that studied by EBIC (electron beam line-scan length is 4.4 mm at ·25 000
4.5 Minority Carrier Lifetime
Figure 4.93 Experimental minority hole diffusion length dependence (open circles) on the distance, d, from the n-GaN/sapphire interface, measured for a 36 mm thick sample. The solid line shows a fit. The open squares show the calculated values of minority hole
pffiffiffiffiffiffi lifetime, t, using L ¼ Dt. The dashed line represents pffiffiffiffiffiffi a second-order polynomial fit using L ¼ Dt. Inset: theoretical minority hole mobility dependence on dislocation density. Courtesy of L. Chernyak.
magnification). A translation stage with a 0.5 mm step resolution was used for laser beam positioning at the predetermined distance from the GaN/sapphire interface. The dependence of the so measured minority carrier diffusion length on distance from the GaN/sapphire interface for one of the samples under investigation (36 mm thick) is shown in Figure 4.93 as open circles. Also shown in the figure are the calculated minority carrier lifetime (open squares) and minority carrier mobility (in the inset), which will be discussed below. L increases linearly from 0.25 mm at the GaN/sapphire interface up to 0.63 mm at the surface. This increase is in agreement with the average space between dislocations, calculated from the experimentally determined dislocation densities using TEM, and shown in Figure 4.94. Comparing the depth dependence of the dislocation density of Figure 4.94 with that of L shown in Figure 4.93, one notes that the decrease in minority carrier diffusion length toward the GaN/sapphire interface is correlated to an increase in the dislocation density. The minority carrier diffusion length at a certain distance from the interface agrees well with the spacing between two adjacent dislocations. For example, L ¼ 0.63 mm at a distance of 32 mm from the GaN/sapphire interface (Figure 4.93). From Figure 4.94, the average spacing between two adjacent dislocations at this distance is 0.69 mm. Similarly, L ¼ 0.25 mm at a distance of 8 mm from the GaN/sapphire interface, while the spacing between two adjacent dislocations at this distance is 0.35 mm. Hence, it is likely that the minority carrier diffusion length
j 979
980
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.94 Experimental dependence of threading dislocation density versus distance from the GaN/sapphire interface (open circles). The measurement error range is illustrated by the horizontal bars. The dashed line represents the fit. Also shown is average dislocation spacing (solid circles), calculated from the experimentally determined dislocation density. The linear fit is shown by the solid line. Courtesy of L. Chernyak.
is determined by carrier recombination on the adjacent threading dislocations. This simply implies that samples used do not allow the measurement of the intrinsic diffusion length in GaN. It is then expected that larger figures would result in the GaN templates with dislocation density some 2 orders of magnitude lower than the films investigated in this particular study. The hole diffusion length in n-GaN may decrease due to a degradation in diffusivity (or minority carrier mobility) and lifetime, t, or both through the expression pffiffiffiffiffiffi L ¼ Dt. The diffusivity, D, and mobility, mp, are related via the Einstein equation. The calculation of minority carrier hole mobility, mp, is a difficult problem in any semiconductor material, because of the degenerate valence bands. However, to get a first-order estimate of mp versus dislocation density, one can simply assume a single valence band with an associated literature hole effective mass of 2.0m0, and then follow previously reported treatments [311]. The results are presented in the inset of Figure 4.93. Note that the 150 cm2 V1 s1 value of mobility at low dislocation densities is much higher than that found in most p-type GaN samples. However, such samples are usually heavily doped (with Mg), and heavily compensated (perhaps by N vacancies), and thus may be expected to produce a much lower mobility than that calculated for a high-quality n-type GaN under investigation. The intensity of PL emission can be related to carrier lifetime [337] that can be used to probe for any trend in t on d. This has been done as a function of distance away from the interface in a 64 mm sample. This is the sample that had been referred to as being 68 mm layer. The discrepancy is due to thickness variation across the sample. To summarize the results, the intensity of the room-temperature band-to-band transition decreases as the probe moves closer to the GaN/sapphire interface. Because the PL intensity is proportional to the carrier lifetime, one can conclude that the latter parameter decreases at lower values of d.
4.5 Minority Carrier Lifetime
With the dependence of diffusion length, L, and lifetime, t, the variation of the diffusion constant with position away from the interface can be discerned. Einsteins relation can then be used to determine the dependence of the minority carrier mobility, hole mobility in this case, on position away from the interface. Because the diffusion length correlates with dislocation density, the basic semiconductor parameters such as the diffusion constant, minority carrier lifetime, and the minority carrier mobility can now be the related to the dislocation density, as are shown in Figure 4.93 and its inset. To find the dislocation density corresponding to a certain distance from the GaN/sapphire interface, at which the L measurements were carried out, one can use the dislocation density versus d data, presented in Figure 4.94. Because L depends linearly on d and on the square root of t, a quadratic fit for t versus d results (see dashed line in Figure 4.93). Note that the values of t so calculated are in agreement with the minority carrier lifetimes (1.4–2.3 ns), probed in InGaN/GaN quantum well, in situ in SEM, using time-resolved cathodoluminescence [338]. Similarly, using the EBIC method indicated a diffusion length of 0.9 mm in the investigations of Miyajima et al. [339]. In yet another investigation somewhat shorter diffusions lengths, determined by electron beam induced current method, were reported. Bandic et al. [340] utilized planar Schottky diodes on unintentionally doped n-type and p-type GaN grown by OMVPE determined the diffusion lengths to be 0.28 0.03 mm for holes and 0.2 0.05 mm for electrons. Minority carrier lifetimes of approximately 7 ns for holes and 0.1 ns for electrons were also estimated using these diffusion lengths and measured mobilities. In the case of HVPE-grown GaN that are typically in the 10 mm or so range, diffusion lengths in the 1–2 mm range were reported. Bandic et al. [340] also attempted to correlate the measured diffusion lengths and lifetimes with the structural properties of GaN. They also sought to correlate the diffusion lengths to linear dislocations as they might act as recombination centers. In yet another investigation but using cathodoluminescence, Godlewski et al. [341] also sought to successfully correlate the microstructure of GaN with the diffusion length measured. Poor quality films led to shorter diffusion lengths. Extending the EBIC investigation to include point defects profiled by DLTS, Soltanovich et al. [342] sought to correlate the two, in particular the 0.54 eV trap from the conduction band. Basically, this particular trap concentration increases notably for dislocation densities ND exceeding 109 cm2. For two different structures having ND of about 108 and 109 cm2, the diffusion lengths of 250 and 130 nm, respectively, were measured. It is clear that thicker epitaxial layers lead to longer diffusion length, which means that the defects play a critical role in its value, which should be obvious when extrapolated from other more established semiconductors in their developmental stages. In addition to the materials properties, measured diffusion length depends on the experimental configuration, making the determination of true diffusion constant difficult. Understandably, we discussed the diffusion length and defects in the same breath as they naturally are interrelated. In GaN and related nitrides, in addition to point defects that could shorten the diffusion length, these materials contain structural
j 981
982
j 4 Extended and Point Defects, Doping, and Magnetism boundaries that are charged and most likely recombination centers for minority carrier. In this environment, the diffusion length to a first extent would be determined by interboundary distance. This may in fact be why the diffusion length measurements by cathodoluminescence did not seem to show discernable difference amongst the nitrides. Because diffusion length is an important parameter in determining the current in p–n junctions, some discussion of this parameter in the context of diffusion current is also treated in Volume 2, Chapter 4.
4.6 Positron Annihilation
The positron annihilation method can be used to interrogate vacancies or vacancyrelated complexes in solids that are primarily negatively charged [343,344]. The technique has been refined for semiconductors to the point that the concentration of the vacancies, bivacancies, or the vacancy-related complexes in the material can be estimated along with determining their charge state. Vacancies can be unambiguously identified along with their open volume (monovacancy/bivacancy/larger cluster) and atomic surrounding (i.e., the sublattice of the vacancy). In many cases, whether a vacancy is surrounded by impurity atoms can also be discerned. Positrons are sensitive to neutral and negative vacancies and their differentiation is possible. Positive vacancies are not detected due to the Coulomb repulsion, unless they are first converted into a neutral or negative charge state (e.g., by excitation with photons). The sensitivity is in the concentration range 1015–1020 cm3. Against a calibrated sample the relative vacancy concentrations in the samples can be determined to an accuracy of better than 10%. The absolute concentration scale is known within a factor of 2. The positron spectroscopy can be applied to both bulk crystals and epitaxial layers. The minimum required thickness of the layer is >100 nm, although thicknesses of about 1 mm (or preferably much thicker) are usually preferred. Positrons can be used to determine the depth profile of vacancies within the range of 0–2 mm from the surface. The depth resolution is typically 20% of the depth. In addition to vacancies, positron spectroscopy is useful for studying other defects containing open volume, such as voids, dislocation lines with open volume, DX centers, among others. There are no limitations concerning the conductivity (n-type, SI, p-type) of the sample except the charged state of the defect being examined. The technique has been applied to vacancies in GaN [319,345]. While the earliest positron experiments in GaN showed negative Ga vacancies in the 1018 cm3 range in unintentionally doped n-type layers and bulk crystals [345], vacancies were observed in semi-insulating or p-type GaN as well [345,346]. The Ga vacancy concentration correlates well with the intensity of the yellow luminescence at 2.2 eV, suggesting that the acceptor states of VGa are involved in this optical transition [345]. In positron annihilation experiments, positrons get trapped at neutral and negative vacancies due to the missing positive charge of the ion cores. The reduced valence and core electron densities at vacancies increase the positron lifetime and narrow the positron–electron momentum distribution. The vacancy defects in thick GaN layers
4.6 Positron Annihilation
and bulk crystals have been investigated using a conventional fast–fast positron lifetime spectrometer (time resolution 200–250 ps) in collinear geometry. For the fast positrons emitted from the source to be stopped in GaN, the layer thickness must be greater than 30 mm. In this setup two identical sample pieces are sandwiched with the positron source (22 Na on 1.5 mm thick Al foil). The lifetime spectrum is then a superposition of exponential decay components, having the form dnðtÞ=dt ¼ P i I i li expðli tÞ, where n(t) is the probability of positron to be alive at time t. The decay constants li ¼ t/ti are called annihilation rates that are simply the inverses of P the positron lifetimes, ti. The average lifetime tave ¼ i Ii ti is the most reliable experimental lifetime parameter, with accuracy better than 1 ps. The average lifetime will increase with increasing positron trapping into vacancies. In more refined studies, positron lifetime experiments have also been performed in thin GaN layers by making use of a pulsed positron beam, where the timing information lost in the beam formation process is recovered from the pulsing electronics. The vacancy defects in thin GaN layers have been investigated by implanting positrons from a monoenergetic beam to depths of up to 1 mm from the surface. The Doppler broadening of the 511 keV annihilation radiation in this case is recorded using one or two Ge detectors (with optional coincidence) and a stabilized multichannel analyzer system. The shape of the 511 keV line is described using the conventional low and high electron momentum shape parameters S (valence annihilation parameter) and W (core annihilation parameter) [344,347]. When positrons annihilate at vacancies, the S parameter increases and the W parameter decreases, because a larger fraction of annihilations takes place because of participation of the valence electrons with lower momentum. Consequently, the lower the S parameter, the less the concentration of Ga vacancies, VGa is. A low electron momentum parameter S represents the ratio of the counts in the central part of the annihilation line to the total number of the counts in the line (see Figure 4 in Ref. [347]). By the same token, the high electron momentum parameter W represents the fraction of the counts in the wing regions of the line over the total number of the counts in the line (see Figure 4 in Ref. [347]). Owing to their low momenta, primarily the valence electrons contribute to the region of the S parameter. On the contrary, only the core electrons have sufficiently high momentum values to contribute to the W parameter. Consequently, S and W are named the valence and core annihilation parameters, respectively. In investigations early on in the development of GaN, both highly Mg-doped bulk crystals and MBE-grown layers were shown to exhibit positron annihilation characteristics of the defect-free GaN lattice [319,345]. Hence, the highly Mg-doped bulk and MBE samples being free of vacancy defects that trap positrons can be used as references for the positron annihilation studies. 4.6.1 Vacancy Defects and Doping in Epitaxial GaN
Positron annihilation experiments have been performed by the positron group under the direction of the late Professor K. Sarinen at Helsinki University of Technology in
j 983
j 4 Extended and Point Defects, Doping, and Magnetism HVPE samples grown at Lincoln Laboratories [353], freestanding GaN samples grown at Samsung Advanced Institute of Technology, and also various OMVPE samples. In terms of the first in the series, the samples ranged in thickness from 1 to 68 mm in an effort to map the Ga vacancy concentration as a function of the sample thickness from the sapphire/GaN interface. Specifically, the experiments were performed in the following samples: HVPE LH1089(3,1) 1 mm, HVPE LH1106 (3,1) 5 mm, HVPE LH1059(3,1) 10–14 mm, HVPE LH706(3,1) 36–39 mm, HVPE LH266(3,1) 49–68 mm, and HVPE LH1232(1,3) 2.6 mm. Recall that the deep-level concentration decreased with increased thickness, the electron mobility increased with increased thickness, and the acceptor concentration decreased with increased thickness (see Section 4.2.1). Because all the acceptor defects are ionized and negatively charged in n-type samples, positron annihilation experiments would add support to the suggestion that the acceptors are caused by Ga vacancies VGa. Also, complexes with substitutional impurities such as ON and SiGa present in the films can be investigated. Conventional positron lifetime spectra taken in a thick (>30 mm) n-type HVPE GaN sample at 550 and 80 K are shown in Figure 4.95. Data taken at 550 K led to a single lifetime ti ¼ 160 ps consistent with that of the Mg-doped MBE GaN samples that are descriptive of films with very little if any vacancies. At 80 K, however, the exponential decay can be described by a two-lifetime process. The lifetime for positron annihilation at vacancies was deduced to be tV ¼ 235 ps. The experimental GaN: (VGa ) ~ 10 17 cm -3
5
10
550 K 80 K
4
10
Counts
984
3
10
2
10
0
500
1000
1500
Time (ps) Figure 4.95 Positron lifetime spectra in thick, >30 mm, HVPE GaN showing an annihilation lifetime of 235 ps indicative of vacancies labeled as Ga vacancies. Courtesy of K. Saarinen.
2000
4.6 Positron Annihilation
Figure 4.96 Schematic of positron densities at Ga and N vacancies, not complexed, in GaN and the lattice diagram showing the local symmetry of a simple VGa and one with a neighboring substitutional O atom on an N site. The experimental lifetime tV ¼ 235 ps can be attributed to the Ga vacancy (relaxed outward) but not to the nitrogen vacancy. The upper part courtesy of K. Saarinen and the lower part Ref. [138].
lifetime tV ¼ 235 ps can be attributed to the Ga vacancy (relaxed outward), but not to the nitrogen vacancy [138], as shown in Figure 4.96. Using the Mg-doped MBE GaN sample as the reference, the density of Ga vacancies VGa was thus estimated to be about 1017 cm3 near the film surface. The vacancies can be positively identified as Ga vacancies and they are responsible for the acceptors in the film, as the concentration of likely acceptor impurities such as Mg (<1015 cm3) and C (<1016 cm3) determined by secondary ion mass spectrometry (SIMS) are much less then the vacancy figure. Although these experiments alone cannot distinguish between the isolated and complexed Ga vacancies in one sample, the body of the evidence presented here and in the literature [347–350] lends support to the premise that the Ga vacancies are associated with defect complexes (such as VGa–ON or VGa–SiGa). To reiterate, the thicknesses of the films used are 1, 5, 10–14, 36–39, 49–68 mm, where 49–68 mm simply represents the thickness variation across the 2-in. wafer. The Doppler broadening experiments indicated that the thinner layers (<30 mm) contain Ga vacancies at even larger densities as would be expected. The increasing S parameter as the thickness of the film is reduced is indicative of larger concentrations of Ga vacancies as shown in Figure 4.97. Again, an Mg-doped sample containing very
j 985
j 4 Extended and Point Defects, Doping, and Magnetism Mean implantation depth (μm)
0
0.09
0.26
0.50
0.79
1.13
HVPE GaN
0.50
10–14 μm
Mg-doped reference
0.49
S parameter
986
1 μm
36–39 μm
5 μm
49–68 μm
VGa
0.48
0.47
0.46 Defect free
0.45 0
5
10
15
20
25
Positron energy (keV) Figure 4.97 Ga vacancies and 1, 5, 10–14, 36–39, 49–68 mm thick HVPE GaN layers indicating increased S parameter, thus increased Ga vacancy concentration toward the GaN/Al2O3 interface in each of the films. A Mg-doped p-type sample with very low, if any, Ga vacancy is shown as the reference. Courtesy of K. Saarinen. (Please find a color version of this figure on the color tables.)
little, if any, Ga vacancies was used as reference. Here, the reference S parameter, which depends slightly on the resolution of the detection system, is S ¼ 0.459. Because the films are assumed reproducible and thinner films represent the earlier stages of growth in thicker films, it is concluded that the Ga vacancy concentration increases toward the interface as is the case with dislocations. In addition, as shown in Figure 4.97, the implantation energy can be used to determine the extent of positron in the films, independently giving a measure of the S parameter as a function of depth in each of the films. Figure 4.98 shows the W (core annihilation parameter) versus S (valence annihilation parameter) parameter plot for all the samples, ranging in thickness from 1 to 68 mm, inclusive of the reference Mg-doped sample. The slope of the curve supports the argument that the Ga vacancy is the same in all the films at the same distance from the interface regardless of the eventual total
4.6 Positron Annihilation 0.058
HVPE GaN 0.056
W parameter
0.054 0.052 0.050 0.048
GaN Mg-doped reference LH 266 (3,-1) 49–68 μm LH 706 (3,1) 36–39 μm LH 1059 10–14 μm LH 1106 (3,1) 5 μm LH 1089 (3,1) 1 μm
0.046 0.044 0.042 0.040 0.45
0.46
0.47
0.48
0.49
S parameter Figure 4.98 The W (core annihilation parameter) versus S (valence annihilation parameter) parameter plot for all the samples. The slope of the curve points to the same Ga vacancy concentration at a given distance from the interface regardless of the layer. Courtesy of K. Saarinen.
thickness of the film. The Ga vacancy concentration was determined in each of the samples and plotted versus the layer thickness that is considered as the distance from the interface and is shown in Figure 4.99. The density drops from about 1020 in 1 mm thick film to below 1017 cm3 in the nominally 68 mm thick films. Because the interfacial region is determined to be n-type and highly conductive, this region must also contain even larger concentrations of O- and/or N-vacancies with lead to n-type material. SIMS results already indicate mid-1019 cm3 O being present in the layer. This has been attributed to O out-diffusion from sapphire as previously reported [351]. Moreover, the preliminary data in sample LH1232 (2.6-mm-thick sample) suggest that this sample contains vacancy clusters instead of Ga vacancies. At lower temperatures, the average positron lifetime increases, indicating that positrons get trapped more efficiently due to attractive Coulomb interaction and lending further support to the proposition that Ga vacancies are negatively charged (i.e., acceptors). If positrons were trapped at negative impurities, such as MgGa ions, as was the case in the measurement conducted in highly n-type bulk GaN such as UNIPRESS bulk GaN crystals, the average lifetime would have decreased at low temperatures not increased. Observation of the opposite effect in the HVPE samples to that in the bulk crystals is a good indication that Ga vacancies are the only negatively charged defects. One can then conclude that the Ga vacancies are the dominant acceptors in thick HVPE layers in the regions of the layer that are more than 5 mm away from the GaN/Al2O3 interface. In short, formation of VGa depends on the thickness of the layer and is strongly enhanced close to the GaN/Al2O3 interface.
j 987
j 4 Extended and Point Defects, Doping, and Magnetism 20
HVPE GaN/Al2O3
10
Ga vacancy concentration (cm -3 )
988
19
10
18
10
17
10
16
10
0
20
40
60
Distance from the interface (μm) Figure 4.99 Ga vacancies in GaN as a function of thickness of the layer. Because the layers are reproducible, the same figure is indicative of the Ga vacancy density at a given distance away from the GaN/Al2O3 interface in HVPE GaN. Courtesy of K. Saarinen.
To further investigate the compensating nature of the Ga vacancies and vacancyimpurity complexes, freestanding HVPE GaN samples (grown at ATMI Inc.) intentionally doped with oxygen were measured with positron lifetime and coincidence Doppler broadening spectroscopy [352]. The free electron and oxygen concentrations were measured with Hall experiments and SIMS, respectively. The thicknesses of the samples varied between 350 and 810 mm and the O concentrations from 2 · 1017 to 2 · 1020 cm3. The measured average positron lifetimes in O-doped HVPE GaN are clearly higher than the lattice lifetime tB ¼ 160 ps for all the samples at low temperatures. By deconvolving, the two pertinent lifetimes are found, where the longer one is t2 ¼ 235 ps, which is typical for a Ga vacancy. The average positron lifetime increases when the sample temperature is decreased. This kind of behavior of the average lifetime is the fingerprint of negatively charged vacancies. Because of the Coulombic nature of the free positron wave function, the trapping at negative vacancies enhances as the thermal velocity of the positron decreases. No other negative centers compete with the Ga vacancy in positron trapping. Negative ions, for example, would localize positrons in a hydrogenic state at low temperatures, strongly decreasing the fraction of annihilation at vacancies and the average lifetime at low temperatures. The same behavior would be expected for possible neutral or negative charge states of the N vacancy where the positron lifetime is very close to that of the GaN lattice. The positron data thus show that the Ga vacancy related defects are
4.6 Positron Annihilation
Figure 4.100 Gallium vacancy concentration as a function of doping. The solid line is guide to the eye. Courtesy of S. Hautakangas.
dominant over the negatively charged acceptors. This is demonstrated above in nominally undoped GaN, but the present data show the universal role of VGa-related defects over almost 4 orders of magnitude of O doping (Figure 4.100). The positron lifetime experiments alone are too insensitive to discern whether the VGa observed in the O-doped HVPE GaN is isolated or complexed with, for example, oxygen. The agreement between the Doppler broadening and ab initio electronic structure calculations together with the lifetime measurements have been utilized to identify the complex. To create a reference sample with isolated Ga vacancies, a nominally undoped HVPE GaN sample [353] was irradiated with 2 MeV electrons at 300 K to a dose of 5 · 1017 e cm2. The average positron lifetime of 205 ps at 380 K in the irradiated sample is significantly higher than that in the GaN lattice, tB ¼ 160 ps. The deconvolution of the lifetime spectrum of electron-irradiated GaN reveals a Ga vacancy related lifetime 235 ps, the same as in O-doped HVPE GaN. The 2-MeV electron irradiation of GaN samples thus generates intrinsic Ga vacancies with an introduction rate of 1 cm1, typical of primary defects formed during electron irradiation [349]. The O concentration of irradiated GaN is more than 1 order of magnitude lower than the Ga vacancy concentration. Therefore, the positrons are trapped at isolated Ga vacancies in the irradiated GaN sample. The deconvolved electron momentum distributions obtained from the Doppler broadening measurements for the vacancy-related defects in electron-irradiated and O-doped HVPE GaN samples are not similar, which paves the way for different complexes to be delineated [354]. In the momentum range between 15 and 30 · 103m0c, the intensity of the electron momentum distribution in irradiated GaN is clearly lower compared to that in the O-doped GaN. This effect can be attributed to oxygen surrounding the Ga vacancy in the defect complex: O ion is smaller than N and thus contributes
j 989
990
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.101 Measured and calculated defectspecific momentum distribution curves for irradiated and O-doped samples and defects. All the curves are shown as ratios to the data obtained in the GaN lattice (note the offset between the theoretical and experimental curves). The defect-specific signal in the experimental curves is obtained by decomposing the original Doppler spectrum by determining the fraction of annihilations from
the positron lifetime measurement. The main contribution to the momentum distribution in the range 15–30 · 103 m0c arises from annihilations with Ga 3d electrons. The decrease in intensity in this momentum region is due to the reduced density of Ga 3d electrons in a Ga vacancy. On the contrary, the difference between VGa–ON and VGa in this region arises from the valence electron states derived from the atomic 2p orbitals. Courtesy of F. Tuomisto.
more for high electron momentum. The same behavior can be inferred from the calculated momentum curves. Moreover, the difference between VGa–ON and isolated VGa arises from the valence electron states derived from the atomic 2p orbitals (Figure 4.101). Hence, to conclude, the results of these particular investigations as well as those preceding them show that the Ga vacancies and Ga vacancy–oxygen complexes are the dominant compensating defects in n-type GaN, with concentrations [VGa] 1016–1018 cm3. Vacancy defects can compensate the p-type conductivity in GaN as well [355]. In this vein, three Mg-doped GaN layers with a thickness of 2 mm grown on a sapphire substrate by MOCVD were investigated. The as-grown sample was of electrical resistivity. The second sample was annealed at 500 C under N2 for 5 min, and had a net space charge concentration NA–ND ¼ 3 · 1016 cm3 as determined by C–V
4.6 Positron Annihilation
measurements. This concentration is generally equal to the net hole concentration p in the valence band. The third sample was annealed at 800 C under N2 for 30 min and had NA – ND ¼ 1 · 1017 cm3. An Mg-doped sample (p 1018 cm3) grown by MBE at 800 C was measured as a reference. A continuous monoenergetic positron beam was then used to study the Doppler broadening of the annihilation radiation and a pulsed positron beam was used to study the positron lifetime in these thin samples. The results obtained at measurement temperatures above 400 K are of importance. In this temperature range, the signal from the vacancy defects is the strongest (no signal of positrons trapping at shallow negative centers with no open volume is present), as shown by the Doppler broadening experiments. In the positron lifetime experiments performed with a pulsed beam, a p-type Mgdoped GaN layer grown by MBE was used as a reference as presented in Figure 4.102. The lifetime spectrum of this sample shows a single exponential component of 154 ps. Within the experimental accuracy, this is nearly the same as that reported for annihilations in the lattice in conventional lifetime experiments in thick bulk crystals (160 ps). No vacancies are thus observed in the MBE-grown GaN:Mg layers. The positron lifetime spectra recorded in the MOCVD-grown Mg-doped samples indicate strong evidence of vacancy defects as displayed in Figure 4.102. From the data three different positron lifetimes can be deduced. The longest one, t3 ¼ 500 ps (intensity 6%), is a typical lifetime for large open volume defects, where at least 10 atoms are
Figure 4.102 The measured positron lifetime spectra in Mgdoped GaN layers, grown by MBE (reference) or by MOCVD (others). The pulsed positron beam energy was 16 keV and measuring temperature 540 K. Courtesy of S. Hautakangas.
j 991
992
j 4 Extended and Point Defects, Doping, and Magnetism missing from their lattice sites. The second lifetime component t2 ¼ 180 ps (intensity 74%) is also larger than the lifetime in the lattice (tB ¼ 154 ps), indicating the presence of monovacancy defects. The shortest lifetime component is typically t1 ¼ 90–140 ps and corresponds to delocalized positrons annihilating in the GaN lattice. Its value is less than tB because it reflects both the annihilation and trapping of positrons. It is clear that annealing affects the lifetime spectra strongly by decreasing the intensities of the vacancy-related lifetime components. The lifetime t2 ¼ 180 ps is clearly smaller than the 235 ps value obtained previously for Ga vacancies, indicating a larger electron density and smaller open volume. The only possible candidates for such defects are N vacancies or complexes involving VN. The lifetime of 180 ps is consistent with the predicted value for VN [345]. On the contrary, the isolated N vacancy is not expected to be a positron trap because of its positive charge. This suggests that neutral or negative complexes involving VN are present in Mg-doped GaN. See Section 4.3.1.2.1 for a discussion of both VGa and VN with their formation energies and charge states. The decomposed electron momentum distributions obtained from the Doppler broadening measurements for the nitrogen vacancy related defects can be compared to ab initio calculations. The calculated momentum distribution for the VN–MgGa complex is in very good agreement with the experimental one, indicating that the nitrogen vacancies are present as complexes with magnesium in Mg-doped GaN. The observation of VN–MgGa complexes is in good agreement with theoretical calculations, which predict a low formation energy (1.4 eV) for such pairs [211,356]. According to theory, the vacancy complexes involving either VN or VGa should be the dominant defects at any position of the Fermi level during growth. The presence of VGa complexes [345] is well established in n-type GaN. Confirming the theoretical prediction, these results show that VN–MgGa pairs exist in semi-insulating and p-type GaN. The positron annihilation experiments in thermally annealed sample in the range of 500–800 C lead to elimination of vacancy complexes. The obvious path is that the VN–MgGa pairs dissociate and VN migrates to the surface. It is interesting to note that the VN–MgGa pairs are observed in as-grown MOCVD GaN, but not in the material grown by MBE as depicted in Figure 4.102. The discrepancy between the two methods is most likely related to the presence of hydrogen in the former and not in the latter. The as-grown Mg-doped MOCVD GaN is semi-insulating because H has been suggested to passivate Mg atoms. A postgrowth heat treatment in N is required to activate the Mg dopants and the p-type conductivity, as discussed in Sections 4.8 and 4.9. Hydrogen is absent in the MBE growth and already the as-grown material exhibits p-type conductivity. The results suggest that the VN–MgGa complexes are stable at the growth temperatures of 800–1000 C only if the Fermi level is close to the midgap; otherwise the pairs dissociate. For neutral VN–MgGa pairs, one obtains an estimate for their concentration of 1017–1018 cm3. Therefore, the dissociation of these pairs during annealing leads to 1017–1018 cm3 electrically active Mg. As the Mg concentration is in the 1019 cm3 range, and one can conclude that the VN–MgGa pairs may contribute significantly to the compensation. It should be noted, however, that hydrogen passivation is treated to be the dominant effect in this estimates.
4.6 Positron Annihilation
In a sense, positron annihilation studies show that vacancy defects act as compensating centers in both n- and p-type GaN. The Ga vacancies and vacancy-impurity complexes have been shown to be the dominant acceptor defects in n-type GaN. Nitrogen vacancies from neutral or negative complexes with magnesium in p-type Mg-doped MOCVD GaN, contribute both to the compensation and activation of the p-type conductivity after postgrowth annealing. 4.6.2 Growth Kinetics and Thermal Behavior of Vacancy Defects in GaN
The positron annihilation spectroscopy has also been applied to study the formation of vacancy defects in GaN grown by RF plasma-assisted MBE for different polar orientations and for different stoichiometry conditions. The N-polarity GaN was grown by nucleating GaN buffer layers directly on sapphire in highly Ga-rich conditions (see Chapter 3 for details). The Ga-polarity GaN was prepared by epitaxy on Ga-polar (0 0 0 1) GaN templates grown by MOCVD on a-plane sapphire. As discussed in Chapter 3, growth on a-plane sapphire still leads to c-plane GaN. The stoichiometry was varied by changing the beam-equivalent pressure (BEP) of gallium during MBE growth experiments. The vacancy defects in the layers can be studied by plotting the layer-specific S parameters as a function of W parameters, as displayed in Figure 4.103. The data points from N-polar samples 1–4 definitely fall in a different group from the Ga-polar samples 6–9. Moreover, the N-polar sample 5, exhibits vacancy defects in these two sample groups that are different. Positron studies have previously identified two types of vacancy defects relevant to this study: Ga vacancies with characteristic parameters SVGa/SB ¼ 1.046 and WVGa/WB ¼ 0.80, and vacancy clusters with an open volume of a divacancy or larger [343] characterized by Sclust/SB ¼ 1.10 and Wclust/WB ¼ 0.75. As shown in Figure 4.103, the S and W parameter values from the Ga-polar samples 6–9 and the N-polar sample 5 agree well with the point characteristic of the Ga vacancy, suggesting a high VGa concentration. On the contrary, the slope of the line between the (S, W) points of the N-polar samples 1–4 and the vacancy-free lattice (SB, WB) agrees well with the characteristic associated with the vacancy cluster. The positron trapping at the vacancy clusters is also suggested by the very high S parameter, S/SB ¼ 1.08. Interestingly, the Ga vacancies are detected in very high concentrations both in Ga-polar and N-polar samples, where the gallium BEP during growth was low (0.5 · 106 Torr). This means the growth approaches N-stable conditions. Note that this also leads to significantly enhanced oxygen incorporation. The S and W values do not significantly differ from SVGa and WVGa, indicating that positron trapping is in saturation. Therefore, it is only possible to give a lower limit estimate, [VGa] 5 · 1018 cm3. It is likely that these vacancies are charged, forming the compensating defects that limit the carrier concentration obtainable from the oxygen. The charged nature is reflected in the extremely low mobility, <20 cm2 V1 s1 at 300 K, in the samples 5–9. The presence of the high VGa concentration in the samples with lower gallium BEP is very reasonable. The incorporation of oxygen is
j 993
994
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.103 The S–W plot of (S,W) values associated with Ga- and N-polar MBE GaN layers. The (S,W) values corresponding to vacancy-free GaN (SB,WB), Ga vacancy (SVGa, WVGa), and vacancy cluster (Sclust, Wclust) are also shown. The parameters of the vacancy-free
GaN are different from those described in the previous section due to the different measurement setup used in this study. The relative parameters S/SB and W/WB can be used for comparison between different experimental setups. Courtesy of M. Rummukainen.
significantly increased both in Ga-polar and N-polar growth direction when the growth is N-stable. The observed high VGa concentration in the N-stable samples thus agrees with the theoretical calculations, predicting that the formation of compensating VGa should be enhanced in the strongly n-type material [138,140]. The observation of the vacancy clusters in the N-polar layers grown in Ga-stable conditions is very interesting. Positron trapping at similar vacancy clusters has been observed earlier in MBE GaN layers (unknown growth polarity) grown on Si (1 1 1) substrate [343] and in Si-doped MBE layers grown on HVPE GaN substrate [357] with Ga-polarity, albeit with inversion boundaries as seen by transmission electron microscopy. It is thus evident that the formation of clusters is associated with the N-polar growth direction. Typically, the quality of the N-polar GaN is inferior to that of Ga-polarity layers although in many cases the morphology and electrical properties when grown on sapphire by MBE might to the first extent look somewhat similar. As discussed in Chapter 3 though, the surface morphology of N-polarity samples is inferior. The higher reactivity of the ð0 0 0 1Þ surface not only enhances the incorporation of oxygen impurities but also enables the formation of vacancy clusters. The high nitrogen pressure method can be used to grow dislocation-free bulk GaN single crystals, and thick overlayers can be grown by HVPW within reasonably short times. This provides an ideal system for investigating the effect of the growth polarity on the impurity incorporation and point defect formation. In addition to the low oxygen content of the substrate bulk GaN (compared to Al2O3), the lack of dislocations minimizes the diffusion of impurities from the substrate. Thus, the properties
4.6 Positron Annihilation
of the HVPE layers should be independent of the layer thickness, in contrast to heteroepitaxial HVPE GaN. The incorporation of impurities and formation of point defects during growth should then be limited by the (polarity dependent) surface kinetics and thermodynamics (independent of the polarity). Positron lifetime spectroscopy has also been applied to investigate HVPE GaN overlayers grown on facets of both Ga and N polarities of bulk HNP GaN substrates [358]. The GaN layers were grown by HVPE to thicknesses in the range of 30–160 mm. Four of the layers were grown on the Ga face and one layer on the N face of the HNP GaN substrates. One of the Ga-polar layers (30 mm) was grown in the same run with the N-polar layer. Apart from the thickness, the properties of the Ga-polar layers were found to be similar to each other. The positron lifetime measured in the HVPE layers grown on the Ga side of the substrate crystal is constant over the entire temperature range and the same for all samples, tave ¼ 160 ps, as seen in Figure 4.104. This value is consistent with the lifetime of positrons (tB) in defect-free GaN. Thus, in the Ga-polar HVPE layers, the concentration of vacancy-type defects is below the sensitivity range of the method, that is, below 1015 cm3. The average positron lifetime measured in the Ga side of HNP GaN is identical to that determined in earlier studies [345,348] and similar in shape to those measured in the N-polar samples shown in Figure 4.104a. In these three samples, tave is longer than tB, which indicates the presence of vacancy defects. A second (higher) lifetime component t2 ¼ 235 10 ps, seen in Figure 4.104b, could be fitted to the experimental lifetime spectra measured on the Ga and N sides of HNP GaN and in the Npolar HVPE layer. This component can be attributed to the Ga vacancy, which is in the negative charge state and most likely complexed with oxygen, as presented above. A third (higher) lifetime component t3 ¼ 470 50 ps, shown in Figure 4.104c, could be fitted to the lifetime spectra measured in the two N-polar samples. This component can be attributed to vacancy clusters that involve at least 20 vacancies. The reduction of the average positron lifetime with decreasing temperature in the samples where tave is longer than tB is a clear indication of the presence of negative ion defects trapping positrons at low temperatures to hydrogenic states where the positron lifetime is equal to tB. Interestingly, the Ga vacancy and O impurity concentrations obtained with SIMS ([VGa] 7 · 1017 cm3, [O] 2 · 1019 cm3) are the same in the N-polar HVPE layer and N-polar side of the substrate bulk GaN, although their growth temperatures were very different (around 1300 K in the HVPE and 1800 K in the HNP growth). Considering the calculated formation energies [138] of VGa and VGa–ON (E fV 1.3 eV and E fVO 1.0 eV in n-type GaN) as well as the available sites of formation that are limited by the oxygen concentration for the latter, the equilibrium concentrations of isolated Ga vacancies at the growth temperatures dominate by a factor of approximately 100 in spite of the lower formation energy of the complex. Thus, the final concentration of VGa–ON in the material is determined by the ability of the isolated vacancies to diffuse and form pairs with the oxygen impurities. However, as will soon be pointed out, the VGa–ON complexes are unstable above 1500 K. Thus, the VGa–ON concentrations are determined by the equilibrium at about 1500 K during cooling down, resulting in [VGa] 1017–1018 cm 3 in both materials.
j 995
996
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.104 Average positron lifetimes (a) and the second and third lifetime components extracted from the lifetime spectra (b), (c) measured in the GaN samples. The solid lines represent the fits of the temperature-dependent trapping model to the data. Courtesy of F. Tuomisto.
By studying both the MBE- and HVPE-grown (hetero- and homoepitaxial) layers, it has been demonstrated that the polarity of the layers plays a pivotal role on the formation of vacancies and vacancy clusters, which are more abundant on the N-polar side. In addition, the vacancy concentrations are similar in both HVPE and HNP GaN in spite of the much higher growth temperature of the latter. This suggests that the stability of the point defect complexes is an important factor determining which defects survive the cooldown process from the growth temperature. To further study the stability of the VGa–ON complex in GaN, annealing under high pressure has been employed as a powerful technique to investigate the thermally induced evolution of the native in-grown vacancy defects in thick freestanding HVPE GaN samples [359]. For this investigation, a 270 mm thick GaN film grown at 1350 K in
4.6 Positron Annihilation
a conventional horizontal HVPE reactor on two-step lateral overgrown MOCVD templates was investigated. Four selected crack-free, self-separated samples from one wafer were annealed at 10 kbar pressure for 1 h at four different temperatures in the range 1423–1723 K. The high pressure was provided by nitrogen to help prevent the dissociation of GaN at temperatures above about 1300 K. The positron lifetime experiments were performed between 10 and 300 K. The fast positrons used in positron lifetime experiments enter the GaN lattice to an average depth of 30 mm with an exponential stopping profile, and thus they probe approximately one third of the sample thickness. Hence, to study the distribution of the defects along the c-axis, the positron lifetime in the samples was measured with both the Ga- and N-polar sides facing the source. In the as-grown HVPE GaN sample, the average positron lifetime and hence the Ga vacancy related defect concentration is clearly higher on the N side than on the Ga side throughout the whole temperature range of the measurement. This is as expected, since the Ga vacancy concentration correlates with the O and dislocation density profiles in HVPE GaN. Annealing of the material at 1523–1623 K decreases the average positron lifetime measured on the N side. This implies that the concentration of the Ga vacancies near the N side decreases with increasing annealing temperature due to the dissociation of the Ga vacancy oxygen complexes. No change in the positron lifetime was observed near the Ga side at annealing temperatures up to 1623 K. However, during annealing at 1723 K the average positron lifetimes near both the N- and Ga-sides change noticeably and become nearly equal, being slightly higher as compared to that near the Ga-side of the asgrown sample. This indicates flattening of the Ga vacancy concentration depth profile, and it is important to notice that the Ga vacancy concentration increases near the Ga-side after the annealing at 1723 K, as seen in Figure 4.105. The O concentration near the N-face of the HVPE GaN samples is relatively high due to the higher dislocation density in the heteroepitaxial interface region, even when a buffer layer is used. The Ga vacancies that are more easily formed in the region of high O content (high n-type conductivity) during growth, survive the cooling down process by forming complexes with the O impurities. Hence, the decrease in the Ga vacancy concentration is due to the dissociation of the VGa–ON pairs. By fitting the concentration data, a binding energy of EB ¼ 1.6 0.4 eV is obtained, which is in good agreement with the value of EB ¼ 1.8–2.1 eV predicted by theory [138,140]. The increase of the Ga vacancy concentration after annealing at 1723 K measured near the Ga-side of the sample indicates that thermally formed vacancies are present in the material. The flattening of the vacancy concentration profile with annealing temperature suggests that the O impurities also are redistributed during annealing, and the sample has reached thermal equilibrium. The results presented above on ingrown vacancy defects in HVPE and HNP GaN indicate that the Ga vacancies are formed as isolated defects and the final concentration is determined by the ability of the Ga vacancies to diffuse and bind to O impurities. Increased O donor ionization shifts the Fermi level up toward the conduction band from Ec0.7 to Ec0.5 eV at 1723 K, at which temperature the Ga vacancies are stabilized by the O impurities.
j 997
998
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.105 The Ga vacancy related defect concentrations determined from the N- and Ga-polar sides of the HVPE GaN layers as a function of annealing temperature. The dashed lines are guides to the eye. Courtesy of F. Tuomisto.
The formation energy of the isolated Ga vacancy can be estimated from the equilibrium concentration at 1723 K to be Ef ¼ 2.5–3.2 eV at Fermi level position Ec EF ¼ 0.5 eV, in very good agreement with the theoretical results of Ef ¼ 2.5–3.5 eV [138,140]. The positron annihilation data presented demonstrate that Ga vacancies are created thermally at the high growth temperature in HVPE and HNP growth, but their ability to form complexes such as VGa–ON determines the fraction of vacancy defects surviving the cooling down. An overall conclusion to this section can be made by stating that positron annihilation is a very powerful technique in analyzing the characteristic and concentration of negatively charged and neutral defects (isolated or complexed) in semiconductors such as GaN.
4.7 Fourier Transform Infrared (FTIR), Electron Paramagnetic Resonance, and Optical Detection of Magnetic Resonance
Fourier transform infrared measurements in the frequency range of 120–320 wave numbers (cm1) in a transmission mode have been undertaken. At frequencies corresponding to vibrational modes of impurities incorporated in the lattice resonant absorption occurs and valuable information about the local nature of impurities can be garnered. Spectral position is unique to a specific impurity and absorption intensity is proportional to concentration. Absorption features in the IR transmission have been obtained in an unintentionally doped GaN layer on sapphire and the results of such an experiment are shown in Figure 4.106. As can be seen, at two distinct energies, 187 and 211.1 cm1 for zero magnetic field, distinct absorption bands were observed. Figure 4.106 also shows the dependence of those absorption bands on magnetic field up to 12 T. The frequency of the
4.7 Fourier Transform Infrared (FTIR), Electron
Figure 4.106 FTIR absorption measurements in the 120–320 cm1 frequency range and up to a magnetic field of 12 T. The frequency of the absorption bands leads to two donors with binding energies of 30.9 and 33.9 meV and the splitting with magnetic field leads to an effective mass of 0.22mo. Courtesy of E. Glaser, W. Carlos, and J. Freitas.
absorption band is indicative of the donor binding energy where the intensity of absorption is related to the concentration of donors. Needless to say that the data point to two donors with binding energies of about 30.9 and 33.9 meV. The former is believed to be due to the unintentionally introduced Si, while the genesis of the latter, which has a much lower concentration, is not clear. The splitting due to the magnetic field, shown with lines emanating from each donor absorption band at zero magnetic field, is consistent with an electron effective mass of 0.22m0. Many defects have electron levels, which split due to different internal fields: exchange interaction, the Jahn–Teller effect, spin–orbit interaction, and so on. Such defects may be paramagnetic or rendered paramagnetic under illumination and, thus, can be detected by magnetic resonance techniques. The EPR method can be used for identification and a thorough study of the predominant defect in a sample. In thin epitaxial layers with defect concentrations at the part per million levels, a more efficient method is often the optical detection of magnetic resonance that is really EPR and PL in one. ODMR provides information on the ground state and
j 999
1000
j 4 Extended and Point Defects, Doping, and Magnetism microscopic origin of defects through obtaining the values and angular dependencies of the Zeeman splitting (g-tensor) and interpretation of any hyperfine structure. It is based on the changes of the PL intensity under varying applied DC magnetic field and resonant microwave absorption. The g-value of the free electron is ge ¼ 2.0023. In the first-order perturbation theory, the deviation of the g-value for a particular center is given by l/DE, where l is the spin–orbit interaction constant and DE is the energy splitting [360]. Usually, donors show g values smaller than free-electrons, and acceptors have a positive g shift. However, there are exceptions to this simple rule [361]. Donors in GaN are typically characterized by g-factors between 1.949 and 1.962, showing a small anisotropy (g g? 0.003–0.006). Acceptors are characterized by comparatively a larger value and anisotropy of the g-factor [362,363]. The magnetic resonance methods give the spin Hamiltonian parameters of the defect (g-factor, hyperfine interaction tensor, etc.) and thus can reveal the symmetry of the defect, the character of the defect wave function, and in principle the chemical identity of the defect. ODMR and EPR could determine whether defects are interstitials, vacancies, impurities, and so on. EPR typically probes the ground state of the system while ODMR reveals information on the excited states. ODMR relies on the spin dependence of recombination processes and combines PL with EPR. The PL provides the high sensitivity needed to study defects at the part per million level or less while the EPR provides the high resolution required to reveal the local symmetry and hyperfine structure of the defect. EPR and ODMR are complementary techniques that can be brought to bear to determine the atomic and electronic structure of residual defects and dopants in GaN. Native defects (vacancies, antisites, and complexes) that are present in semiconductors in moderate concentrations can be produced in large amounts by electron or particle irradiation of the sample. Comparison of the magnetic resonance or luminescence before and after irradiation can augment efforts to identify native defects. Defects may be created or healed by annealing. Annealing in hydrogen or oxygen environment can reveal the related defects in the samples. The ODMR method has already been applied to the study of GaN. The ubiquitous yellow luminescence in undoped GaN and different bands in Mg-doped GaN were studied by Glaser et al. [364]. Identification of transitions suggested in these investigations is inconsistent with the models proposed in conjunction with PL studies [365,366]. The reason may be a two-step process involving spin-dependent capture followed by radiative recombination [366]. Follow-up experiments led to the observation by ODMR of two deep defects in electron-irradiated undoped GaN [367]. One of them, responsible for the broad band at about 0.95 eV, is related to interstitialGa defect. Another band with a sharp zero-phonon line (ZPL) at 0.88 eV is not yet identified. The reported origin or the yellow luminescence is a widely researched field with no shortage of controversy. A detailed discussion of the yellow band often observed in n-type and undoped GaN can be found in Volume 2, Chapter 5. ODMR is useful in determining the nature of the centers causing extrinsic transitions. In the context of this review, ODMR has been applied to the investigation of unintentionally doped GaN grown by HVPE on sapphire substrate at Lincoln
4.7 Fourier Transform Infrared (FTIR), Electron
Figure 4.107 ODMR signal in a 5–10-mm-thick GaN layer from the 2.2 eV band for the magnetic field perpendicular and parallel to the c-axis. The shallow donor (SD) and deep donor (DD) transitions with their associated g-values are given. Courtesy of E. Glaser, W. Carlos, and J. Freitas.
Laboratories (5–10 mm) and a Samsung template, the latter being freestanding. Note that the ODMR was done on the same samples as investigated by EPR. Representative ODMR spectra are shown in Figures 4.107 and 4.108 with the following observation that can be made. The resonance with a g-value of 1.950, assigned to shallow (effective mass like) donors from previously published work, was found for both the 2.2 eV yellow PL band from the HVPE sample (LH1106), and the 2.4 eV green PL band from the Samsung freestanding template (see Figure 4.109). As also observed in the EPR studies that will be discussed below, the notable difference is the larger linewidth of this feature in the Samsung material. The broadening is consistent with the reduced concentration of shallow donors as suggested by Hall effect and EPR measurements (the broadening arises from an increase in the electron–nuclear hyperfine interaction in this regime of nearly isolated donors, i.e., ND 1–3 · 1016 cm3; see, e.g., Glaser et al. [368]). The second feature with g ¼ 1.989 observed on the 2.2 eV band from the HVPE sample (LH1106)
j1001
1002
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.108 ODMR signal in a 5–10-mm-thick GaN layer from the 2.2 (SD) and 3.27 (SD) eV peaks. In both cases, distinct, but broad, deep donor (DD) and Shallow Acceptor (SA) peaks are observable. Courtesy of E. Glaser, W. Carlos, and J. Freitas.
is typically found in this emission from as-grown n-type GaN and is ascribed to a deep defect that may involve Ga vacancies. A different resonance with donor-like character and of deeper nature (based on the g-value of about 1.975 and larger linewidth compared to that found for the effective mass donor resonance) was found for the 2.4-eV PL band. The weak feature labeled by an arrow at about 840 mT with B rotated 30 from the c-axis is more pronounced with B parallel to the c-axis (not shown). The corresponding g-value of about 2.02 suggests an assignment of this feature to deep acceptors. Distinct ODMR spectra were observed for the 1.8 eV PL bands from the HVPE sample (LH1059) and the Samsung template (see Figure 4.109). In particular, different acceptor-like isotropic resonances with g-values of 2.000 and 2.019, respectively, were found for this emission. The same shallow donor resonance (g|| ¼ 1.950 and g? ¼ 1.947) as observed for the 2.2 eV PL band was also observed for the 1.8 eV emission band from the HVPE sample (LH1059). A resonance with a slightly different g-tensor (g|| ¼ 1.958 and g? ¼ 1.954) was found for the 1.8 eV PL from the Samsung template. This line is likely associated with quasi-shallow or perturbed
4.7 Fourier Transform Infrared (FTIR), Electron
Figure 4.109 ODMR signal on 2.2 and 2.4 eV PL bands for GaN grown at both Lincoln Laboratories (2.2 eV) and Samsung Advanced Institute of Technology (2.4 eV), the latter one being a freestanding template. In the case of the Lincoln sample, the B field is parallel to the c-
axis. In the case of the Samsung freestanding template, the B field is rotated 30 off the c-axis. This polarization is shown because the weak feature labeled by an arrow at 840 mT does not appear when B is parallel to the c-axis. Courtesy of E. Glaser, W. Carlos, and J. Freitas.
donors. As suggested by preliminary fits to the spectra, two additional features (labeled by arrows) are also possibly revealed in this emission. In addition to the ODMR, EPR measurements were also undertaken in the same HVPE layer (LH1059) and the freestanding GaN template from Samsung (sample 177). All samples had a resonance due to residual donors as shown in Figure 4.110. The Samsung freestanding template exhibited about 6 · 1015 cm3 uncompensated donors per cubic cm while the HVPE sample (LH1059) exhibited a concentration about a factor of 4 higher. Note that the volume of the Samsung template was 5 mm3 versus 0.1 mm3 for the HVPE sample. The linewidth for the Samsung template is 27 Gauss (G) versus 12 G for the HVPE sample and less than 10 G for typical OMVPE films. The narrower widths observed in HVPE and OMVPE samples are consistent with more isolated donors. In the samples with higher concentration, the hyperfine interaction with the lattice nuclei is averaged out and the lineshape is very Lorentzian. Due to a lower concentration of donors, the Samsung template exhibits a lineshape that has a significant Gaussian character due to unresolved hyperfine interactions. With somewhat purer samples, one can
j1003
1004
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.110 Electron paramagnetic resonance investigation of freestanding and epitaxial HVPE samples. Courtesy of E. Glaser, W. Carlos, and J. Freitas.
expect to perhaps resolve hyperfine interaction with the nearest neighbor lattice nuclei and establish whether the residual donor is on the Ga, N, or interstitial site. Note also that these spectra were taken at 20 K, necessitated by the signals (especially in the Samsung sample) being easily saturated at 4 K. This again is indicative of isolated donors.
4.8 Role of Hydrogen
According to theory [215,216], isolated hydrogen may exist as positively charged (Hþ), being a shallow donor and located near the N atom, or as negatively charged (H), being a deep acceptor and located close to the Ga atom. Hþ prefers the nitrogen antibonding site (1.02–1.04 Å from the nearest N atom), whereas for H the Ga antibonding site is energetically most stable. Hþ is mobile at room temperature due to small migration barrier (0.7 eV), while H is stable (3.4 eV). The solubility of hydrogen is considerably higher in p-type than in n-type GaN. This is consistent with high diffusivity of Hþ, whereas diffusivity and solubility of H is expected to be low in n-type GaN [215]. In p-type GaN, Hþ and H have low and high formation energy, respectively, and the opposite relation is true for n-type GaN. H0 is never stable in GaN, representing a negative-U system. This terminology is due to Anderson [369] who, in an attempt to account for the lack of observation of paramagnetism for localized intrinsic defects in chalcogenide glasses (amorphous material), proposed a model which relies on existence of an effective negative correlation energy U for electrons localized at a defect site. In this model energy gain associated with electron pairing in dangling bonds of defects that are coupled with large lattice relaxation may
4.8 Role of Hydrogen
be able to overcome the Coulombic repulsion of the spin-up and spin-down electrons which would end up supplying a net effective attractive interaction between the electrons at the site. This has come to be known as the negative-U center. It has also been applied successfully to point defects in Si early on [370,371]. Hydrogen molecule H2 is unstable with respect to dissociation into monatomic hydrogen. Besides, it has quite high formation energy. The hydrogen does not form a bond with the Mg atom in p-type GaN, as some investigators argue. Instead, it is bound to N atom in p-type GaN, so that the Mg–H complex has a H–N bond with the calculated stretch mode of 3360 cm1 [215], very close to the stretch mode in NH3 (3444 cm1) [215]. Hydrogen plays an important role in p-type doping. It passivates Mg during growth and prevents formation of native deep donors due to self-compensation process. After the growth, hydrogen can be removed by annealing, leading to good p-type conductivity. Unlike many other semiconductors, formation of H2 molecules is very unlikely in GaN. Thus, most reasonable mechanism of H removal during postgrowth annealing is diffusion of hydrogen atoms to the surface, interface, or structural defects. Hydrogen is likely to form complexes with point defects. In particular, hydrogenated VGa is expected to have lower formation energy than the isolated VGa in n-type GaN, and hydrogenated VN is predicted to have lower formation energy than the isolated VN in p-type GaN [216]. Experimental results on hydrogen in GaN are limited and not well confirmed. High concentrations of hydrogen and carbon have been detected by SIMS and elastic recoil detection analysis (ERDA) in unintentionally doped GaN grown by OMVPE using mixture of triethylgallium (TEG) and ammonia (NH3) in the presence of nitrogen and/or hydrogen or deuterium [372]. It was established that total hydrogen (carbon) content decreases exponentially from 2 · 1021 to 3 · 1019 cm3 (from 1.4 · 1021 to 2 · 1018 cm3) when the substrate temperature was increased from 650 to 1100 C. High concentration of these species is attributed to efficient desorption of hydrocarbons from the GaN surface during growth. Decrease of H and C contamination with increasing temperature is attributed to reduction of effective surface with increasing temperature (increase in crystallite size). In reactive MBE growth, contamination with H and C is expected to be much smaller due to absence of the metalorganic compounds. However, in the MBE growth using ammonia as a source of nitrogen, concentration of hydrogen can be large. Zhang et al. [373] reported high concentration of hydrogen in unintentionally doped GaN grown by MBE with ammonia. A remarkable finding of this work is a linear dependence between the background electron concentration and concentration of hydrogen, as shown in Figure 4.111). Hydrogen also has much higher concentration at the surface due to adsorption. The authors of Ref. [373] suggested that hydrogen sitting near the N site acts as a donor. The activation energy of this donor was estimated as 120 meV from Figure 4.111. It is found that hydrogen is present in relatively high concentrations in as-grown GaN samples, and, moreover, it is easily incorporated during plasma etching, solvent boiling, and wet chemical etching [374].
j1005
j 4 Extended and Point Defects, Doping, and Magnetism 1020
Electron density (cm-3)
1006
1019
1018
1017 1019
10 20
10 21
Hydrogen concentration (cm-3 ) Figure 4.111 Background electron concentration as a function of hydrogen contamination [373].
4.9 Intentional Doping
As is the case in any semiconductor technology, the control of the electrical properties of GaN and related materials remains the foremost obstacle in the progress of device development. Controlled doping in a wide range of concentrations in wide bandgap semiconductors is somewhat difficult for n-type varieties, and very difficult for p-type varieties, as has been pointed out, for example, by Walukiewicz [375]. Unintentionally doped GaN has, in all cases, been observed to be n-type, with the best samples still showing a compensated electron concentration that during the early stages of development approached 1020 cm3 but fortunately got reduced to about 2 · 1016 cm3 in freestanding samples. With the exception of oxygen and silicon, no impurity has been found present in a sufficient quantity to account for the carriers, so researchers have initially, by nature, attributed the background to impurities. However, GaN is very rich in terms of native defects that are more likely or at least likely responsible. All efforts, until early 1990s, obtaining p-type doping has resulted in heavily compensated, highly resistive films. With resilience and clever engineering, development of a p-type doping above the 1018 cm3 level for GaN remained a primary challenge and focus for researchers, which turned out to be successful. Steady progress, even for AlGaN for which the task is even more challenging and also for InGaN, has been made. In this section, various impurities in terms of their formation energies, which go to their likelihood of incorporation and experimental particulars, are discussed. The main discussion of the optical manifestation of intentional dopants is contained in Volume 2, Chapter 5.
4.9 Intentional Doping
Generally speaking, elements such as C, Si, and Ge on the Ga sites and O, S, and Se on the N sites can potentially form shallow donors in GaN. Elements such as Be, Mg, Ca, Zn, and Cd on the Ga sites and C, Si, and Ge on the N sites have the potential of forming relatively shallow acceptors in GaN. A brief review of the formation energies and energy levels calculated from the first principles and by using the effective-mass method is provided herewith. While the former can predict which defect can be easier to form, the latter is much more accurate in terms of determining the ionization energy. 4.9.1.1 Shallow Donors Wang and Chen [376] calculated the energy levels of shallow substitutional donors in GaN using the effective-mass theory accounting for mass anisotropy, central-cell potential correction, and the host conduction band edge wave function. For wurtzite GaN, they deduced donor ionization energies of 34.0, 30.8, and 31.1 meV for C, Si, and Ge, respectively, on the Ga site, and 32.4, 29.5, and 29.5 meV for O, S, and Se, respectively, on the N sites. Boguslawski and Bernholc [209] considered substitutional C, Si, and Ge impurities in wurtzite GaN in the framework of first principles calculations and concluded that formation energies of SiGa and GeGa donors are reasonably low (0.9 and 2.3 eV, respectively, in Ga-rich conditions), while formation of CGa donor has low probability. In other theoretical calculations, the formation energy of neutral CGa has been estimated as 4–4.7 eV in the N-rich case and 5.7–6.5 eV in the Ga-rich case [140,209,248]. Neugebauer and Van de Walle [138] and Mattila and Nieminen [140] have obtained low formation energies for ON and SiGa donors (both being below 2 eV in the Ga-rich case). Calculations show the trend that with decreasing Fermi level, the formation energy of the shallow donors decreases linearly as shown in Figure 4.112. Consequently, one may expect even easier formation of the substitutional shallow donors in high-resistivity or p-type GaN if these impurities are present during the growth. 4.9.1.2 Substitutional Acceptors For p-type doping in GaN, Zn, Cd, Be, Mg, C, Ca, Hg, and Li on the Ga site [377], and C, Si, and Ge on the N site could potentially give rise to relatively shallow acceptors. Ionization energies of main substitutional acceptors in wurtzite and zinc blende GaN have been calculated in the effective-mass approximation by Mireles and Ulloa [378] and Wang and Chen [379] (Volume 2, Table 5.1) the results of which are tabulated in Table 4.7. In the same vein, Neugebauer and Van de Walle [161] reported the results of a comprehensive first principles investigation of several possible acceptors in GaN. As we can see in Volume 2, Table 5.1 the main candidates for the shallow acceptors are BeGa, MgGa, CN, and SiN. From an analysis of the electronegativity differences between the acceptor atoms and the host atoms, P€ od€ or [380] concluded that the ionization energy of BeGa should be only slightly greater than the effective mass value that is estimated to be 85 meV. In increasing order of ionization energies, the other Ga-substitutional acceptors are Mg, Zn, Cd, and Hg, according to increasing electronegativity difference between these impurities and a Ga atom [380]. Park and Chadi [234] examined the atomic and electronic structure of substitutional Be, Mg,
j1007
j 4 Extended and Point Defects, Doping, and Magnetism CGa
CaGa
4 ZnGa Formation energy(eV)
1008
CN
2 Mg Ga BeGa 0 ON
SiGa —2
0
1
2
3
Fermi level (eV) Figure 4.112 Calculated formation energies as a function of Fermi level for shallow donors and acceptors in GaN grown in the most favorable conditions for these dopants (except for SiGa which should have even lower formation energy in Ga-rich conditions). Solid lines, Ga-rich case;
dashed lines, N-rich case. The zero of the Fermi level corresponds to the top of the valence band. The results for Ca, Zn, Mg, and Be are taken from Ref. [161], C – from Ref. [248], and O and Si – from Ref. [217].
and C acceptors in GaN through first principles calculations and obtained that these impurities should result in effective-mass, not AX states in GaN (represent some unknown acceptor-like defect states. The term is coined after the DX centers which represent some unknown deep defect of donor character). The calculated formation energies of some of the substitutional acceptors are shown in Figure 4.70. The formation energies of MgGa and BeGa (the latter is not shown) and their ionization energies are the lowest. However, the Be atom is very small which paves the for its efficient incorporation on the interstitial site where it acts as a double donor [161,254,256,381]. Therefore, from group II impurities Mg and Be are the most promising p-type dopants in GaN, and Be appears the best candidate provided that formation of Bei double donor can be suppressed. Formation energies of acceptors from group IV impurities, such as SiN, and GeN, are relatively high, so that in equilibrium conditions formation of these acceptors is unlikely in GaN [161,209]. Formation energy of CN can be sufficiently low in Ga-rich conditions [209,210,221,248,382]. However, it might be compensated with interstitial C when the Fermi level is close to the conduction band might be compensated with interstitial C when the Fermi level is close to the conduction band [248]. Note also that
4.9 Intentional Doping Table 4.7 Calculated acceptor ionization energies (in meV)
for wurtzite (Wz) and zinc blende (ZB) GaN [383]. Acceptor
EA (Wz) [379]
EA (Wz) [378]
EA (ZB) [379]
EA (ZB) [378]
BeGa MgGa CaGa ZnGa CdGa CN SiN GeN
187 224 302 364 625 152 224 281
204 215 259 331
183 220 297 357 620 143 220 276
133 139 162 178
230 203
147 132
formation energies of the acceptors decrease with increasing Fermi level, so that we may expect efficient formation of CN and SiN acceptors in undoped or n-type doped GaN when these impurities are present during the growth. However, this acceptor might be compensated with interstitial C when the Fermi level is near the conduction band [248]. Also to be noted is that formation energies of the acceptors decrease with increasing Fermi level, so that we may expect efficient formation of CN and SiN acceptors in undoped or n-type doped GaN when these impurities are present in the growth environment, which somewhat raises the bar to success on the experimental front. Expanding the search for a p-type dopant, Neugebauer and Van de Walle [161] examined formation energies of acceptors from group I impurities, such as KGa, NaGa, and LiGa. They predicted that these acceptors have rather high formation energies and large ionization energies. Moreover, the formation energies of the Na and Li interstitials are much lower in p-type GaN, so that these alkali impurities will most likely be interstitials and act as single donors rather than acceptors [161]. 4.9.1.3 Isoelectronic Impurities Arsenic and phosphorus can substitute nitrogen in GaN to form isovalent defects AsN and PN. In silicon, germanium, and most of III–V semiconductors isovalent impurities do not introduce deep energy levels in the bandgap. In contrast, the available theory predicts formation of deep gap states by isovalent impurities in GaN, similar to the situation in II–VI compounds, due to large size mismatch of the isovalent atoms in GaN [384,385]. The first principles calculations have shown that the AsN and PN defects may exist in neutral, 1þ, and 2þ charge states in GaN. Mattila and Zunger [386] predicted the þþ/þ and þ/0 energy levels of AsN at 0.24 and 0.41 eV above the valence band maximum, and those of PN at 0.09 and 0.22 eV, respectively [385]. Van de Walle and Neugebauer [386] obtained values of 0.11 and 0.31 eV for the þþ/þ and þ/0 levels of AsN, respectively. The formation energy of AsN is quite large, and in the most favorable conditions (Ga-rich case, Fermi level is above 2.3 eV), the first principles calculations predict the formation energy of AsN to
j1009
1010
j 4 Extended and Point Defects, Doping, and Magnetism be about 4.5 eV. In p-type GaN, and/or in N-rich conditions, formation of AsGa antisite-type defect is much more likely [386]. This defect is a double donor with the 2þ/0 level being at about 2.5 eV above the valence band maximum and very low formation energy when the Fermi level is close to the valence band [375]. Therefore, the presence of As during growth of p-type GaN can cause significant compensation. 4.9.2 n-Type Doping with Silicon, Germanium, Selenium, and Oxygen
With a successful reduction of the unintentional n-type background, doping GaN with Si has very well been established in both the vapor-phase and vacuum deposition techniques. Silicon is the major dopant for n-type GaN because the electron concentration can be controllably changed from 1017 to 2 · 1019 cm3 by varying the flow rate of SiH4 in OMVPE growth [387]. Although the current technology is sufficient for light emitters, the background impurities and native defects must further be reduced for certain varieties of field effect transistors and detectors. 4.9.2.1 Si Doping Controlled n-type conductivity in binary and ternary alloys of GaN, AlN, and InN is generally achieved by Si doping in both vacuum deposition and metalorganic chemical-vapor deposition techniques [388]. Si substitutes a Ga atom in the lattice and provides a loosely bound electron. In the dilution limit the ionization energy of the Si level in GaN is about 30 meV and decreases with an increasing doping level due to screening, as discussed in Volume 2, Chapter 3. PL measurements in the averagequality GaN yielded a binding energy value of 22 meV [389]. Measurements of highquality GaN layers grown on freestanding GaN wafer indicate the binding energy of Si and O in GaN to be 30.18 and 33.20 meV, respectively [390,391]. The determination is based on magneto-optical studies as well as on detailed analysis of the two-electron 0 n¼1 transitions [392]: ðD01 ; Xn¼1 A Þ2e and ðD2 ; XA Þ2e in which a donor-bound exciton transition is accompanied by excitation of an electron to the n ¼ 2 state of another donor-bound exciton. The difference between the ground state and first excited state transitions of excitons is then equal to three fourth of the donor binding energy in the hydrogenic model that applies, the details for which can be found in Volume 2, Equation 5.53. The solubility of Si in GaN is high and on the order of 1020 cm3, making it suitable for group III-N doping, and is most frequently used. Electron concentrations versus silane flow rates in an OMVPE reactor for GaN and AlGaN are exhibited in Figure 4.113. Although silicon is amphoteric (may appear as a shallow acceptor when substituting the N atom), the first principles calculations (Section 4.6.1) predict that in thermodynamic equilibrium formation of SiN is less probable than SiGa. There are, however, a few experimental observations pointing out the possible existence of SiN acceptors in GaN. This is really not more than attributing the increasing intensity of the UVL band observed to transitions from the shallow donors to the shallow acceptors with increasing Si doping [363,393,394]. In n-type GaN, the
4.9 Intentional Doping 19
10 Electron concentration (cm-3)
Silane doping measured at 300 K
18
10
17
10
GaN Al 0.1 Ga 0.9 N 16 UndopedGaN < 10 cm-3 ΔESi∼ 27 meV 10
0
1
10 SiH4 flow rate (sccm)
10
2
Figure 4.113 The electron concentration versus silane flow rate in an OMVPE reactor for GaN (solid circles) and AlGaN (open circles). Courtesy of I. Akasaki, Meijo University.
intensity of the DAP transitions depends only on the concentration of the acceptors involved; therefore, these results indicate that at least some shallow acceptor incorporates with Si doping. Moreover, the increase in the shallow acceptor concentration with Si doping is confirmed by an increase in the emission intensity from the exciton bound to the shallow acceptor [393]. The SiN acceptor binding energy of 224 meV has been estimated from the position of the zero-phonon peak of the UVL (represents PL emission in the blue region of the optical spectrum) band in GaN: Si [393]. An ODMR study of Si-doped GaN has revealed a highly anisotropic resonance on the UVL band [363], in line with theoretical predictions for the effective-mass shallow acceptor in wurtzitic GaN [395]. Typically, the YL intensity increases with Si doping [396–400]. However, at high doping levels it decreases [387,396]. The enhancement of the YL band with Si doping can be explained by increasing concentration of the VGa-related acceptors, such as VGaON and VGaSiGa, upon the shift of the Fermi level to higher energies during the growth. Unlike ON, demonstrating features of a DX center, SiGa, behaves like a standard hydrogenic donor at high pressures. 4.9.2.2 Ge Doping Ge doping is also well behaved in that the electron concentration increased from 1017 to 1019 cm3 when the flow rate of GeH4 was varied 100 times in OMVPE growth [401]. The observed carrier concentrations of Si-doped GaN were in the range 1017–2 · 1019 cm3, while Ge doping has produced material with electron
j1011
j 4 Extended and Point Defects, Doping, and Magnetism Electron concentration(cm-3)
1012
1019
SiH 4 Si 2H6
18
10
GeH 4
17
10
10–10
10–9
10–8
10–7
Flow rate ( mol/min—1) Figure 4.114 Incorporation rates of Si and Ge n-type dopants as a function of gas flow in OMVPE-grown GaN. Both dopants are well behaved in that the active donor concentrations linearly vary with silane, disilane, and germane flow rates. However, GeH4– requires a factor of 10 higher flow rate to yield the same level doping level expected of silane and silane [401,402].
concentrations of 7·1016–1019 cm3. A linear variation of the electron concentration as a function of both the SiH4 and GeH4 flow rates was observed across the entire experimental range (Figure 4.114). Ge incorporation is roughly an order of magnitude less efficient than Si, as judged by the larger GeH4 flow rates required to obtain similar electron concentrations. Moreover, disilane (S2H6) has also been used in OMVPE growth [402] for Si doping of GaN, which is also shown in Figure 4.114. Other and much safer sources, such as monomethylsilane, have successfully been employed to obtain n-type GaN, but their uses are not yet very common. Goldenberg et al. [403] observed higher n-type GaN conductivities as the NH3 flow was increased during the growth. Although the site selection process, Si substituting for Ga, is very likely enhanced with increased ammonia flow rate, they postulated that H passivation of acceptors in their material leads to the improved electrical characteristics [404–406]. As for the optical manifestation, the intensities of both the excitonic emission and the YL increased with Ge doping. Doping of GaN with Ge above 1020 cm3 led to quenching of the YL as observed by Zhang et al. [407] who attributed this effect to creation of Ga vacancies due to heavy doping. 4.9.2.3 Se Doping As for Se doping electron concentrations of 1.7 · 1018 cm3 [408] and 6 · 1019 cm3 have been achieved in Se-doped GaN OMVPE films [409]. The electron concentration was proportional to the H2Se flow rate employed as a dopant source. The room temperature electron mobilities ranged from 10 to 150 cm2 V1 s1. The high
4.9 Intentional Doping
compensation ratio of 0.4 was nearly constant over the concentration range of 1018–1019 cm3 and became even higher at higher electron concentrations. The Gato-N ratio decreased about 15% with increasing concentration of Se. Although the authors of Ref. [408] attributed the Ga/N ratio change to the formation of NGa defects, formation of VGa due to self-compensation would be a more reasonable explanation in view of theoretical predictions. The YL band intensity increased with Se doping [408,409]. This is consistent with YL being due to the VGa-related defect. Yi and Wessels [409] attributed an increase of the electron concentration in GaN:Se as a cube root of the H2Se partial pressure at n0 > 2 · 1018 cm3 to compensation by a triply charged VGa. Note however that this experiment is not definitive correlating the compensating defect with the isolated 1=n triply charged VGa but not with doubly charged VGaON. Indeed, the N A / N D dependence, where n is the charge of the compensating defect [410], is valid only for a nondegenerate semiconductor, which may not hold for n0 > 2 · 1018 cm3 in GaN. Compensation-related issues in GaN have also been discussed by Yi and Park [411]. Oxygen doping oxygen is a shallow donor in GaN. The activation energy of ON is 33 meV, according to one report [412]. At an electron concentration of n0 > 3 · 1018 cm3, the donor level merges with the conduction band. The Fermi level increases with increasing donor concentration and for an O concentration of 1019 cm3 one may expect optical transition blue shift by 0.1 eV (the Burstein–Moss shift [413]). However, the bandgap renormalization [414] that causes a red shift can amply compensate. As a matter of fact, the combination of the two effects lead to a very small shift in optical transition energies in degenerate GaN : O [415]. Note that the solubility of O in GaN is apparently only 1019 cm3, and if exceeded the excess oxygen formed precipitates in surface microcavities [415]. It should be noted that O is also an unintentional dopant and is present in high concentrations in undoped GaN. For example, in bulk GaN grown at high pressure and high temperature, the concentration of O is on the order of 1020 cm3 [416]. The YL band is very strong in these bulk samples [417]. The correlation between the intensity of the YL and concentration of O has also been noted by Slack et al. [418]. This is consistent with the identification of the YL as being due to the VGaON complex. Note that concentration of both VN and ON increase with O doping. Oxygen in GaN shows the characteristic features of a DX center. Namely, a strongly localized level, associated with distorted O configuration, emerges from the conduction band at hydrostatic pressures above 20 GPa [416]. The pressure dependence of the YL band position experiences change of slope at these pressures [419] thus confirming the attribution of the YL band to transitions from the O donor to a deep acceptor (presumably VGaON). No other PL bands related to O were reported in literature. 4.9.2.4 p-Type Doping There has been much effort aimed at doping GaN and its p-type ternaries by incorporating group II and group IV elements. Many potential p-type dopants have been attempted for incorporation into GaN. Some impurities have been observed to effectively compensate electrons in GaN, leading to highly resistive mate. In addition, much theoretical work has been undertaken in an effort to determine which elements
j1013
1014
j 4 Extended and Point Defects, Doping, and Magnetism are more likely to lead to p-type doping as discussed in Section 4.6.1. Unlike the aforementioned section, here the emphasis is placed on the experimental investigations. Only after postgrowth low-energy electron beam irradiation (LEEBI), reported in 1991, or thermal annealing that converted Mg-doped GaN into p-type material [404]. No other impurity has been as successful as Mg in rendering GaN, AlGaN, and InGaN, with low Al and In mole fractions in the case of p-type alloys. p-Type doping had been a big obstacle for light-emitting diodes based on GaN until Akasaki and Amano made the major breakthrough by using LEEBI on the Mgdoped GaN [420,421]. Later Nakamura et al. [422] also successfully obtained p-type GaN by thermal annealing. From then on, many semiconductor devices based on GaN have been produced and put into commercial use. However, p-type doping in GaN and its ternaries remains as a topic of interest both in terms of technological aspects and also at the fundamental level. 4.9.3 p-Type Doping and Codoping with Donors and Acceptors
Achievement of p-type doping represented the turning point for GaN. This milestone took GaN from a low-level laboratory curiosity and placed it among the most important technology, the applications of which are constantly expanding. As in the case of any wide bandgap material, p-type doping proved to be very elusive and difficult. In what follows, the methods used to achieve p-type doping and inherent defects caused by them are discussed. 4.9.3.1 Magnesium Doping Due to lack of p-type GaN until 1989, attempts at garnering the potential of GaN for emitters relied on metal insulator n-type (m-i-n) GaN blue light emitting diodes first fabricated at what was then the RCA Laboratories, with doped n-GaN and highly resistive i-GaN due to Zn doping. However, a good p–n junction is required for achieving commercial LEDs and any injection laser diode (LD), as detailed in Volume 2, Chapter 5 and Volume 3, Chapter 1. The lack of p-GaN delayed the development of LEDs considerably. Although MgGa is a relatively shallow acceptor [423–425], early attempts of Mg doping were not successful presumably for two reasons: (i) hydrogen is always present in OMVPE and HVPE growth environments, which passivates Mg by forming electrically and optically inactive complexes [426]; (ii) heavily Mg-doped GaN is subject to self-compensation due to donor-like defects that are created [250]. Later, it was established that passivation with H during growth in fact is favorable for obtaining high-conductivity p-type GaN. Indeed, shifting of the Fermi level toward the conduction band facilitates incorporation of Mg (Section 4.9.3) [216]. Postgrowth treatments allow the conversion of the high-resistivity as-grown GaN:Mg into conductive p-type GaN. Hydrogenized Mg can be activated by annealing at temperatures higher than 600 C [372,422,427–429] or by electron beam irradiation [420,430], or even by UV illumination at temperatures above 500 C [431]. In contrast, self-compensation is a permanent effect.
4.9 Intentional Doping
Initial breakthrough came by converting the compensated Mg-doped GaN into conductive p-type material by low-energy electron beam irradiation [432], with hole concentrations as high as 3 · 1018 cm3 and a resistivity of 0.2 O cm [433]. Soon thereafter came the observation that thermal annealing at 700 C under an N2 ambient converts GaN into p-type equally as well. The process was observed to be reversible with the GaN reverting to insulating compensated material when annealed under NH3. Hydrogen was thus identified as the critical compensating agent. Layers of p-type GaN were achieved with no postgrowth anneal required using MBE, employing activated nitrogen or ammonia as the nitrogen source [406]. In the chemical vapor phase epitaxy technique [434], CCp2Mg is utilized for the source of Mg. The Mg chemical concentrations for two different growth temperatures and Mg sources versus the flow rate of Mg sources are plotted in Figure 4.115. The upper curve is for CCp2Mg at a growth temperature of 1020 C. The bottom curve is for MCp2Mg at a growth temperature of 950 C. As the data suggest, CCp2Mg is much more effective and less volatile in that it can be incorporated in GaN at high temperatures where the quality of GaN films is superior to those grown at lower temperatures. To transfer to other reactors, other pertinent parameters such as trimetal Ga (TMG) and ammonia flow rates must be known too. Nevertheless, the point is that Mg concentrations in p-type GaN films are extremely high and approach 1% of the host species. Note that when substitutionally incorporated, only about 1% of the Mg atoms are ionized in GaN. The hole concentration versus flow rate is not shown because the data are not reliable due to large error bars, the likely source of which will be discussed in Volume 2, Chapter 3. In GaN:Mg grown by OMVPE, concentration of free holes at room temperature reaches its maximum value of about 1018 cm3 for the Mg concentration of about 3 · 1019 cm3, and it decreases with farther increase of Mg concentration (Figure 4.116) [435]. Mg acceptors in GaN and AlxGa1xN have activation energy (160–200 meV in GaN and increasing with increasing Al fraction) [425,436,437,439,440] much larger than kT at 300 K, resulting in low activation and therefore low conductivity in p-type GaN and AlGaN, degrading the performance of light-emitting diodes, lasers, and heterojunction bipolar transistors. The large scatter in the measured activation energy of GaN is still not well understood, although concepts such as screening are suggested. In addition, presence of donor-like impurities such as O and Si could cause complex formation that would reduce the activation (ionization energy) as discussed in the text dealing with codoping used to attain increased hole concentrations. Because these materials are of great importance in realizing nitride-based optoelectronic devices, studies that could provide better understanding and realization of highly conductive p-type GaN and AlxGa1xN are urgently needed, particularly for high Al-composition p-type AlGaN-related optoelectronic devices. A few studies have been reported on Mg-doped p-type AlxGa1xN with low Al content 0 < x < 0.27 [441–446]. Figure 4.117 shows activation energies of Mg acceptors in Mg-doped p-type AlxGa1xN as a function of Al content x. For x ¼ 0.27, activation energy as high as 310 meV. The optical processes [447,448] associated with addition of Mg are discussed in detail in Volume 2, Chapter 5.
j1015
j 4 Extended and Point Defects, Doping, and Magnetism Mg concentration (cm—3)
T s = 1020 °C 1020
1019 10—6
10—5
CCp2Mg flow rate (mol min—1)
Ts = 950 °C
Mg concentration (cm—3)
1016
1021
1020 101
10 2 MCp 2 Mg flow rate (mol min—1)
Figure 4.115 Mg chemical concentrations for two different growth temperatures and Mg sources versus the flow rate of the Mg source. The upper curve is for CCp2Mg at a growth temperature of 1020 C. The bottom curve is for MCp2 Mg at a growth temperature of 950 C [434].
Theoretical efforts have been attempted to provide a viable explanation for the relative success of the Mg acceptor in GaN, while other group II metals continue to compensate GaN, even after a LEEBI treatment or annealing. GaN differs considerably from other conventional group III–V semiconductors. For example, due to the strong covalent character, the charge density in GaAs exhibits its maximum near the bond center (BC). There is no such local charge maximum in GaN; rather, there is an
4.9 Intentional Doping
1018 Hole concentration (cm—3)
GaN Mg (RT)
1017
1020
1019 Mg-concentration (cm—3) Figure 4.116 Room temperature concentration of free holes in Mg-doped GaN as determined from the Hall effect versus the concentration of Mg obtained from SIMS measurements [435].
increase in the charge density from Ga toward the more electronegative N in GaN, which causes the charge density around N to be high and nearly spherically symmetric. In view of this, group III–V nitrides are more ionic than other group III–V semiconductors and resemble the band structure of group II–VI semiconductors. Specifically, the presence of a large ionic gap in the valence band density of states causes their lower valence bands (LVBs) to lie deeper beneath the valence band edge. The resulting energy resonance causes the Ga 3d electrons to strongly hybridize with the s- and p-levels of both the upper and lower valence band [449]. Such hybridization is predicted to have a profound influence on the properties of GaN including such quantities as the bandgap, lattice constant, acceptor levels, and valence band heterojunction offsets [449]. It is known in the cases of ZnS and ZnSe that potential acceptors such as Cu, whose d electrons are resonant with the lower valence band, are repelled by the d-hybridized upper valence band in an abnormally deep level, while impurities without resonance form shallow acceptors [450]. Mg has no d electrons, as is the case for Be that will be discussed in Section 4.9.3.2, and turns out to be just sufficiently shallow for RT p-type doping of GaN. On the contrary, Zn, Cd, and Hg, which all have d electrons, arguably form deep levels in GaN as evidenced by the highresistivity films that result when doped with these impurities. The thermal activation energy of acceptors is related to acceptor density and postgrowth annealing temperature [451]. Cheong et al. [452] reported acceptor density and annealing temperature dependent activation energy of Mg in GaN quantitatively. Unlike the OMVPE-grown layers, Mg-doped GaN layers grown by RMBE with ammonia as the nitrogen source exhibit p-type conductivity without any postgrowth treatment [453]. The Mg incorporation during MBE growth is known to depend on the stoichiometry and polarity of the growing surface [454,455] as well as on the substrate temperature and growth rate [456–458]. Several groups investigated
j1017
j 4 Extended and Point Defects, Doping, and Magnetism 0.38 10
19
x = 0.27
0.34 0.32
p (cm -3 )
0.36
Mg activation energy EA (eV)
1018
10
10
0.30 10
E A = 0.31 eV
17
15
13
2
0.28
3
4
5
1000/T (K
-1
6 )
0.26 0.24 0.22 0.20 0.18 0.16
}GaN
0.14 0.0
0.1
0.2
0.3
Al content (x) Figure 4.117 Activation energies of Mg acceptors in Mg-doped p-type AlxGa1xN as a function of Al content x. Solid squares, solid circles, and triangles are data from Refs [443–446], respectively, all obtained by Hall measurements. Open circles indicate data
obtained by PL measurements from Ref. [445]. The inset shows measured temperature dependence of Hall concentration p in the Mgdoped p-type Al0.27Ga0.73N sample from which EA ¼ 0.310 eV was obtained. Courtesy of Lin and coworkers [445].
acceptor-doped GaN, in particular Mg doping, and gave different interpretations for the electronic structure and photoluminescence [459–484]. Much of the abovementioned efforts focused on achieving progressively higher hole concentration and better p-type conductivity. A number of reports have shown that annealing in N2/O2 gas mixtures or in air can aid in the removal of residual hydrogen from GaN [485–488] and potential formation of conducting NiO in Nibased contacts, therefore, significantly improve the contact resistance [489–491]. In this vein, depositing a metal that getters H, such as Pd and Ta, and annealing as part or prior to metallization could help reduce H in GaN and thus increase Mg activation. In a slightly different approach, attention has been paid to the codoping method in GaN and AlGaN in an effort to increase the hole concentration, as discussed in some detail next in Section 4.9.3.1.1. As mentioned above, the difficulty in attaining
4.9 Intentional Doping
low-resistivity wide bandgap semiconductors is in part due to the low solubility and the low activation rate caused by the deep energy level of the acceptors. Let us now discuss codoping as a means of increasing the hole concentration in GaN. 4.9.3.1.1 Codoping for Improving p-Type Conductivity From the discussion we had so far, it is clear that n-type doping in GaN is not very problematic; the same cannot be said about p-type doping. Among the reasons, one can cite the compensation effect due to the low solubility and the low activation rate that is due to the deep energy level of the acceptor or donor. The acceptor energy level is very deep (about a few hundreds of millielectron volts) relative to room temperature (30 meV) because of the small dielectric constant, for example, GaN:Mg (200 meV) and AlN:C (500 meV). The 500 meV acceptor can be activated to yield one hole for every 106 acceptor impurity at room temperature. To attain the low-resistivity wide bandgap semiconductor, the compensation with the increasing solubility of the dopant should be avoided, and the activation rate of the carriers should be increased by reducing the energy level of the acceptor, and also the mobility of the carriers should be increased by reducing the scattering rate of the carriers. The codoping method is interesting and promising [492], in which both nand p-type dopants are provided simultaneously under thermal nonequilibrium crystal growth conditions during MBE or OMVPE. The rational in the form of theory for codoping was provided by Reiss et al. [493,494] during the early stages of Si and Ge development using solution theory. Codoping can lead to enhanced equilibrium solubility (which is a problem for p-type impurities in GaN) lowered ionization energies and might also affect the mobility, all of which happen to be applicable to GaN. The experimental efforts of codoping in GaN and to a lesser degree AlGaN inclusive of the growth methods and salient growth parameters are tabulated in Table 4.8. Essentially, the hole concentration was observed to enhance in p-type GaN: Mg codoped with oxygen donors wherein the hole concentration increased linearly from 8 · 1016 to 2 · 1018 cm3 with increasing oxygen dopant partial pressure. A factor of 3–5 increased resulted in the hole concentration for a fixed oxygen partial pressure during the growth of p-type GaN:Mg. However, unlike the theoretical predictions for reduced formation energy/increased solubility, when Si was codoped with GaN:Mg, the hole concentration remained constant [492]. Let us now give a brief description of the solution theory as well as the modern theories that have been developed [495]. In the equilibrium solution theory, the impurity solubility is dependent on the Fermi level and ion pair formation. In terms of the relation to the Fermi level, the acceptor concentration can be expressed as [493]
NA ¼ exp½ðE Fi E F Þ=kT ; N Ai
ð4:49Þ
where NA and N Ai depict the acceptor concentrations for the doped semiconductor and the intrinsic semiconductor, respectively. The terms EF and E Fi reflect the Fermi level and intrinsic Fermi level, respectively. An exponential dependence of acceptor concentration on the Fermi level is apparent. A similar expression can be written for donors and for a wide range of acceptor concentrations the dependence of NA on
j1019
Growth method
MBE MBE
OMVPE
OMVPE OMVPE
Bulk solution MBE MBE
OMVPE
Codoping pair
GaN : Be-O GaN : Be-O
GaN : Mg-Si
GaN : Mg-Zn GaN : Mg-Si
GaN : Mg-O GaN : Mg-O AlGaN : Mg-O
GaN : Mg-O
DEZn (nmol min1) 0.616 (nmol min1) 0.616 O (cm3) 1018–1019 4 · 1018 4 · 1018
Cp2Mg/Ga 0.7 0.7 Mg (cm3) 1019(0.1–0.15%) 1.6 · 1020 1.6 · 1020 O (ppm) 15 O2/N2 ¼ 1%
1080 1080
SiH4 (nmol min1) 0.11
Cp2Mg/Ga 7.6 · 103
Cp2Mg (mmol min1) 0.36
1080
H2O (Torr) >3 · 1010 H2O (Torr)
Be (cm3) 1020 5 · 1020
1060 1060
1400–1700 780 780
650 650
Donor specie
Acceptor specie
Growth temperature ( C)
Table 4.8 Reported experimental data regarding codoping of GaN and to a lesser extent AlGaN, as tabulated in Ref. [492].
2 · 1018 3 · 1018
— 2 · 1018 2 · 1018
8.5 · 1017 8.5 · 1017
4.3 · 1017
5 · 1018 1 · 1018
p/n (cm3)
17 2.6
— 40 8
10.5 10.5
11.5
70 150
Mobility (cm2 V1 s1)
1020
j 4 Extended and Point Defects, Doping, and Magnetism
4.9 Intentional Doping
donor concentration, ND, is given by [493,494] vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi #ffi u" 2 ND u N D þ N 2A0 ; þt NA ¼ 2 2
ð4:50Þ
where N A0 is the solubility of acceptors for low compensation ratio. The solubility can be increased by forming stable ion pairs, a case in point for which is the Mg–O ion pairs that would form the following Mg þ O: þ ! ½Mg O þ :
ð4:51Þ
The equilibrium constant associated with this reaction can be written following the law of mass action as K0 ¼
aMgO ; aMg :aO þ
ð4:52Þ
where aMg and :aO þ are the activities of ions and aMgO of ion pairs. Assuming a dilute solution, the pair concentration P is expressed as W¼
P ; ðN Mg PÞðN O PÞ
ð4:53Þ
where NMg and NO are the Mg and O concentrations, respectively, and O is a constant that depends on the Coulomb interaction energy as 1 z W¼ exp ; ð4:54Þ 2N kT where N is the total site density that for GaN is 8.8 · 1022 cm3 and z is the Coulomb pair energy which is given by z¼
e2 ; er
ð4:55Þ
where e is the dielectric constant and r is the ion pair distance. For GaN, the Coulomb pair energy z 828 meV using for the nearest neighbor pairs having r ¼ 1.95 Å, neglecting any relaxation and e ¼ 8.9 e0. For NMg ¼ 5 · 1019 cm3 and NO ¼ 5 · 1018 cm3, the computed pair concentration P is 1.5455 · 1018 cm3 at 1300 K, which is a typical growth temperature for OMVPE epitaxial experiments. One would therefore expect increased solubility by a factor of exp(z/kT) that equals about 1.6251 · 103 at 1300 K, which is substantial. In addition to the above discussion centered around increasing solubility of both acceptor and donor pairs that are used, Katayama-Yoshida et al. [495] forwarded a theory that argues that formation of charged nearest neighbor acceptor complex consisting of two acceptors and one donor, that is, acceptor–donor–acceptor (ADA) can also lead to increased hole concentration. Essentially, it is postulated that the codoping method forms the metal-stable acceptor–donor–acceptor complexes for the p-type or donor–acceptor–donor (DAD) complexes for the n-type wide bandgap semiconductors based upon ab initio
j1021
1022
j 4 Extended and Point Defects, Doping, and Magnetism electronic structure calculations [495]. As implied, maximum enhancement in the hole concentration is expected for the ADA complex formation. The increased hole concentration is attributed to reduced ionization energy. In fact, vanishing ionization energies have been predicted. The mechanism for this is that A–D–A or D–A–D complexes decrease the effective formation energy of the dopants under thermal nonequilibrium conditions by reducing the lattice relaxation energy and the Madelung energy because the two pairs of the attractive acceptor–donor (A–D) interaction overcome the one pair of the repulsive acceptor– acceptor (A–A) interaction. Additionally, the formation of the A–D–A or D–A–D complex increases the carrier mobility because the scattering mechanism changes from the long-range Coulomb interaction to the short-range interaction with metallic carrier density upon codoping. The formation of the A–D–A or D–A–D complexes decreases the ionization energy of acceptors or donors, because a donor level is raised and an acceptor level is lowered in the formation of the bonding and antibonding states. When a single acceptor (A) interacts with a single donor (D), the A and D levels will be pushed close to the band edges. Because an additional acceptor upon the codoping interacts with the antibonding donor states with the formation of an A–D–A complex, the acceptor level of an additional A is lowered. In calculations, two types of acceptors are taken the first of which is formed by ADA complex whose concentration is 2ND (ND is the donor concentration) while the second is an isolated acceptor species with a concentration of NA – 2ND (NA being the total acceptor concentration), assuming that every donor forms an ADA complex. The activation energies of the ADA complex and the isolated acceptor are taken to be 0 and 200 meV (representing Mg well). The charge neutrality expression for one donor species (assuming all are ionized, which is the case for a p-type semiconductor) multiple acceptors is given by p þ N Dþ ¼
X N Aj j
1 þ ppj
or simply N A ¼ N Dþ þ p;
ð4:56Þ
where N Aj is the concentration of a given acceptor species j, ND is the donor concentration, and pj ¼ (g/NV)exp (E Aj /kT) with E Aj representing the activation energy of the jth acceptor and g representing the acceptordegeneracy factor that was taken to be 3.6. In the absence of this, a value of 4 is a good default. It should be mentioned that the Fermi level is close to the conduction band and therefore all the donor impurities would be ionized making N Dþ ¼ N D (see Volume 2, Chapter 2 for details). The Fermi level and the hole concentrations can be found by numerically solving N A ¼
NA E ; F 1 þ g exp E AkT
ð4:57Þ
and 2 EV EF p ¼ N V pffiffiffi F 1=2 kT p
ð¥ where F 1=2 ðhÞ ¼ 0
h1=2 dh; 1 þ exp ðh hf Þ ð4:58Þ
4.9 Intentional Doping
specifically substituting Equations 4.57 and 4.58 into Equation 4.56 and performing numerical calculations for the Fermi integral F1/2(Z), which would lead to the Fermi level. The hole concentration then can be calculated using Equation 4.58. Note that in the calculations of Korotkov and Wessels [492], a valence band density of states, NV, of 1.8 · 1019 cm3 reported in Ref. [496] has been used. The calculated hole concentrations for two acceptor doping cases, namely, NA ¼ 1 · 1019 and 1 · 1020 cm3 versus the donor doping concentration, but using a HH valence band effective mass of 2 and corresponding density of states (see Volume 2, Chapter 2 for details) are exhibited in Figure 4.118 with thin (iii) and bold (iv) lines, respectively. The parameters used in calculations are tabulated in Table 4.9. Let us now consider the ADA model in which one donor atom will tend to capture two acceptors and form an ADA complex. The ADA complex behaves like an acceptor with zero activation energy. The calculations can be simplified by considering two cases for NA > 2ND and NA 2ND. For the NA > 2ND, all donors have sufficient number of acceptors with which to combine. Therefore, the concentration of ADA complex will be ND. The remaining acceptor and ADA complex will contribute together acting as ionized centers. The neutrality equation in this case can be expressed as ¼ p; N A þ N ADA
ð4:59Þ
where ¼ N ADA
ND : 1 þ g exp kTE F
ð4:60Þ
For the NA 2ND, case because there are abundance of donors, all acceptors are captured by donors to form the complex and some donors are left as isolated ions, N Dþ . The charge neutrality equation becomes N ADA ¼ p þ N Dþ
ð4:61Þ
where ¼ N ADA
N A =2 ; 1 þ g exp kTE F
ð4:62Þ
and N Dþ ¼ N D N A =2
ð4:63Þ
The simulated hole concentration versus donor concentration within the context of the ADA model are indexed in Figure 4.118 as (i) for NA ¼ 1 · 1019 cm3 and (ii) for 1 · 1020 cm3, which is similar to that reported in Ref. [492] with the exception of a more abrupt change near the inflection point. Shown with diamond symbols is the hole concentration under the condition that NA ¼ 2ND, which represents the upper limit of this method. As argued, an increase in the hole concentration with an
j1023
1024
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.118 Hole concentration versus ND where the acceptor–donor–acceptor complex model of Ref. [495] is shown with thin and bold lines for NA ¼ 1 · 1019 cm3 (i) and 1 · 1020 cm3 (ii), respectively. The optimum hole concentration where NA ¼ 2ND, as expected from the complex formation, is shown with diamonds. For comparative purpose, the simple compensation model which assumes a
single donor and (unpaired) acceptor is depicted with thin and thick lines for NA ¼ 1 · 1019 cm3 (iii) and 1 · 1020 (iv), respectively. The random pair model is also plotted with thin and thick lines for NA ¼ 1 · 1019 cm3 (v) and 1 · 1020 cm3 (vi), respectively. Discussions with Dr R. Korotkov are acknowledged. (Please find a color version of this figure on the color tables.)
increase in the donor concentration, ND, is seen up to a level of NA ¼ 2ND. Because the maximum ADA complex concentration is determined by the acceptor concentration a decrease in the hole concentration for donor concentrations ND > 0.5NA follows. The curve deduced from the simple compensation model [495] is also displayed, as curves iii and iv again for acceptor concentrations of NA ¼ 1 · 1019 and 1 · 1020 cm3.
4.9 Intentional Doping Table 4.9 Parameters used in codoping calculations.
Effective mass for electron Effective mass for hole Activation energy for donor Activation energy for acceptor GaN dielectric constant Temperature Acceptor degeneracy factor
0.22m0 2.0m0 30 meV 200 meV 8.9 300 K 3.6
Let us now turn our attention to the paired ADA model in which the formation of ADA complex is based on the random pairing mechanism. The acceptors can be viewed as being composed of two parts, namely, ionized acceptors and neutral acceptors. The donors (all ionized as the Fermi level is close to the valence band) will first interact with ionized donors to form close pairs. A pair of this kind would capture one isolated neutral acceptor to form the complex. Also activation energy is not zero any more but has the relationship with total impurity density described as E 0A ¼ E A
q2 ; 4per
ð4:64Þ
where the donor–acceptor distance is given by 1=3 3 : r¼ 4pðN A þ N D Þ
ð4:65Þ
To simplify the algorithm used for calculation, as done previously, different regimes can be considered, similar to that performed in the context of the ADA ðnÞ model, but with different terms. First, let us denote N A as the neutral acceptors within the semiconductor given by ðnÞ
N A ¼ N A N A : If
ðnÞ N A >N D ,
N A
ð4:66Þ
all donors will form ADA complexes, resulting in
N D þ N ADA ¼p
where N ADA ¼ ND:
ð4:67Þ
ðnÞ
If, on the contrary N A N D, only a portion of donors can form complexes, resulting in ¼p N A N D þ N ADA ðnÞ
where
ðnÞ
N ADA ¼ NA :
ð4:68Þ
Because N A is not predetermined, iterations are needed to converge on a hole concentration. The hole concentrations for the random distribution model as a function of donor concentration are shown in Figure 4.118 as (v) for Nt ¼ 1 · 1019 cm3 and (vi) for 1 · 1020 cm3. The hole concentration is not seen to decrease as rapidly relative to the simple compensation model (curves iii and iv). However, only a small increase in p for ND 5 · 1019 cm3 is noted.
j1025
j 4 Extended and Point Defects, Doping, and Magnetism To evaluate the effect of donor concentration through the ADA complex formation on the hole concentration, the hole density can be calculated as a function of acceptor concentration using Equation 4.56 with the donor concentration being a parameter (complementary figure to Figure 4.118). These hole concentration versus the acceptor concentration for two donor concentrations, namely, for ND ¼ 1018 and 1019 cm3, are shown with solid lines in Figure 4.119. Hole concentrations obtained using the simple compensation model with one donor, one acceptor and E 0A ¼ 200 meV are also displayed with dashed lines for comparison. An increase describable by a square root dependence of hole concentration is expected from the simple compensation model for high acceptor concentrations [496]. Instead, a very weak dependence of the hole density is observed for the ADA model at high acceptor concentrations. Katayama-Yoshida and coworkers [497–499] investigated the role of n-type dopants, Si and O, in codoping of p-type GaN doped with Be or Mg using ab initio electronic band structure calculations. They found (1) the total energy calculations showed that the formation of Be (Mg)–O–Be (Mg) structures or Be (Mg)–N–Si–N–Be (Mg) ones were energetically favorable; and (2) the variation of the impurity levels caused by the strong interactions between Si (O) and Be (Mg) would enhance the hole concentrations in p-type codoped GaN. They predicted that the p-type codoped GaN using Be or Mg as acceptors and Si or O as donors would exhibit an increased
1019 -3 19 ADA N D =10 cm
ADA ND =1018 cm-3
1018
Holeconcentration(cm-3)
1026
18
1017
= 10
-3
cm
ND
1016
18
= 10
-3
cm
ND 1015 1018
1019
Acceptor concentration (cm-3) Figure 4.119 Calculated hole concentration as a function of the acceptor concentration NA for two donor concentrations of ND ¼ 1 · 1018 and 1 · 1019 cm3 as a parameter for the acceptor–donor–acceptor and simple compensation models that are shown with bold and thin lines, respectively. Courtesy of R. Korotkov and B.W. Wessels.
1020
4.9 Intentional Doping
incorporation of Be or Mg acceptors compared with p-type GaN doped with the acceptors alone. They recently also calculated the codoping method in p-type AlN: [O þ 2C] and n-type diamond [500]. It is demonstrated the codoping method is the efficient and universal doping method by which one can avoid carrier compensation with an increase in the solubility of the dopant, increase in the activation rate by decreasing the ionization energy of acceptors and donors, and an increase in the carrier mobility. Moving onto the technology, predictions regarding codoping are complemented well with experiments in that low-resistivity p-type GaN has been obtained. For example, Korotkov et al. [501] successfully obtained highly conductive p-type GaN by using Mg and O codoping during OMVPE growth. The resistivity of codoped layers decreased from 8 to 0.2 O cm upon oxygen codoping. The activation energy of Mg decreased from 170 to 135 meV (not as much as predicted by the ADA complex theory, a point which is elaborated on below), and the hole concentration was as high as 2 · 1018 cm3. To amplify, to determining the effect of oxygen codoping on carrier concentration for p-type GaN by Korotkov et al. [492] undertook two sets of experiments. The epitaxial films for the first set of experiments were grown at a constant Mg dopant concentration in the gas phase, while the concentration of oxygen was varied from 0 to 80 ppm [501]. The room-temperature carrier concentrations as a function of oxygen doping for this set are displayed in Figure 4.120a. As seen, the hole concentration of the codoped epilayers increased super linearly with oxygen doping. A hole concentration as high as 2 · 1018 cm3 was attained. However, for an oxygen concentration of 30 ppm, the conductivity type changed from hole to electron conduction as indicated by a change in the sign of the Hall effect. For the second set, Korotkov et al. [492] varied the Mg doping while keeping the oxygen partial pressure during growth constant at 4 ppm. The resultant dependence of hole density on Mg flow rate for codoped samples is shown in Figure 4.120b. The hole concentration is seen to increase by more than an order of magnitude up to 1 · 1018 cm3, followed by saturation. For comparison, Mg doping without oxygen grown under similar conditions produced p-type samples with a concentration of 2 · 1017 cm3 and a resistivity of 3.5 O cm. A qualitative agreement between the observed hole dependence on dopant partial pressure and that predicted using the ADA complex model calculations is present. For example, an increase in hole density was observed experimentally with increasing donor dopant as seen in Figure 4.120a, which is predicted by theory. In addition, the data presented in Figure 4.120b indicate that the hole density is nearly fixed at a specific concentration by the donor concentration, ND, which again is in agreement with theory. The solid line in Figure 4.120a (NA is taken as a fitting parameter) and b (ND is taken as a fitting parameter) represents results of calculation using Equation 4.56. Consistent with the work of Katayama-Yoshida and coworkers [495], for these calculations an acceptor binding energy of 0 was assumed for the ADA complex and 200 meV for the isolated acceptor. The value of ND ¼ 1.3 · 1018 cm3 in Figure 4.120b is used to fit the data, and there is again a good agreement between theory and experiment. However, in Figure 4.120a, the increase in the experimental hole concentration is steeper than
j1027
j 4 Extended and Point Defects, Doping, and Magnetism Donor concentration (cm-3) 1018
1019
1020
Carrierconcentration(cm-3)
T = 300 K
1019
1018
p-type
1017 1
n-type
10 Oxygen concentration (ppm)
100
Acceptor concentration, fit (cm-3) 1018
1019
T = 300 K 1018
Hole concentration (cm-3)
1028
1017
1016
0.05
0.1
0.5
Mg flow rate (μmol/min) (ppm) Figure 4.120 (a) Experimental carrier concentration as a function of oxygen concentration during growth. The solid line is obtained by plotting Equation 4.58 with NA ¼ 1 · 1019 cm3. The solid circle symbols are used for p-type while solid square are used to depict n-type conductivity. (b) The hole
concentration is plotted as a function of the Cp2Mg flow rate during growth. The oxygen flow rate was held constant at 4 ppm. The solid line is a fit of Equation 4.58 with ND ¼ 1.3 · 1018 cm3. Courtesy of R. Korotkov and B.W. Wessels.
4.9 Intentional Doping
that predicted by theory. For the data in Figure 4.120a, the fitting parameter NA was taken as 1 · 1019 cm3. The sharp increase observed in the measured hole concentration might to some extent be due to other compensation mechanisms for low hole concentrations. Interestingly, the usual bell-shaped dependence of hole concentration on acceptor concentration typical of GaN:Mg was not observed. In some studies [496] this was attributed to the formation of compensating donor defects. It should be pointed out that the ADA model does not include compensation by donor complexes such as [MgGa–VN]2þ, which might in part account for the discrepancy. Experimental data, presented in Figure 4.120a, are for the heavy doping case (p ¼ 8 · 1016 cm3 and was obtained without oxygen present) with NA on the right-hand side of the bell-shaped maximum [501]. It is likely that oxygen substitutes on the V3Nþ sites, allowing more MgGa acceptors to become active upon codoping with oxygen because the [MgGa–VN]2 þ complex will not form, providing an additional increase of the hole density. While there is a good qualitative agreement between the experiments and the ADA complex model results, several unanswered questions still remain. Among them is the fact that a degenerate acceptor level EA 0 meV was utilized in the calculations, but nonzero activation energies were measured for the samples shown in Figure 4.120 [501] and also in Ref. [504]. In spite of this, degenerate p-type GaN was observed in other investigations, see, for example, Ref. [502], along with the enhancement of hole densities by codoping. The different behavior of Si codoping as compared to O in GaN can be tentatively explained within the framework of the ADA model. The tendency for ADA complex or pair formation depends on the bonding strength, and because the Mg–O bonding is stronger than the Mg–Si bonds, it is natural to expect O codoping to be more effective. Therefore, NA not being dependent on Si donor density is attributed to a donor concentration that is inadequate to produce significant enhancement of acceptors incorporation. There is also an inherent problem with Si in that both Mg and Si compete for substituting for the same (Ga) lattice site. Competitive adsorption also affects the mutual incorporation of Mg and Si. Similarly, the highest hole concentrations were attained in GaN grown with a decreased gallium flow rate with respect to the optimum Ga-flux, as the formation energy of MgGa increases with gallium chemical potential. To reiterate, the ADA model explains some of the observed codoping behavior but not the lack of low ionization energy for acceptors in some codoping studies. This might indicate that there are other important mechanisms in play. One plausible explanation that was forwarded has to do with screening by impurities [493,494]. It is most likely that a combination of several factors, including complex formation, decrease of compensation, increase in acceptor incorporation, and screening effects that lead to reduced ionization energy could all be responsible for the observed increases in hole concentration to varying extents. Interestingly, a potential fluctuation model was also used to describe the observed increase in the hole concentration by codoping, which also predicts very weak temperature dependence of the hole concentration [503].
j1029
1030
j 4 Extended and Point Defects, Doping, and Magnetism Other experiments have been also been conducted. For example, Kipshidze et al. [504] also reported on their successful Mg and O codoping in p-type GaN and AlxGa1xN (x < 0.08) using GSMBE. In MBE growth, it is known that incorporation of Mg takes place on Ga sites of a Ga polarity surface [505]. The stoichiometry of the growing surface is thus important, which can be controlled by adjusting the NH3/Ga and NH3/(Ga,Al) flux ratios. For a fixed Ga and Al flux, the flux of NH3 is increased until the growth rate saturates. The point at which the growth rate becomes independent of the flux of NH3 corresponds to the stoichiometric surface [456]. Because the presence of Al results in more efficient decomposition of NH3, the condition of stoichiometry in AlGaN is obtained at lower fluxes of NH3. In these experiments, the optimum growth rates were in the range 0.5–0.6 mm h1 for both GaN and Al0.08Ga0.92N. Doping with Mg was carried out close to the optimum point, in the region where the growth rate is independent of the NH3 flux (f NH3 ). At lower fluxes, where the growth rate depends on the NH3 flux, p-type doping was not possible and all the samples were n-type or heavily compensated. The decrease in hole concentration for the smallest values of f NH3 is related to the loss of stoichiometric growth condition, that is, when the growing surface becomes Ga rich and the Ga vacancy concentration is greatly reduced. Kim et al. [507] used Mg–Zn codoping in GaN by OMVPE. A low resistivity of 0.72 and hole concentration of 8.5 · 1017 was achieved. Bis(cyclopentadienyl) magnesium (Cp2Mg) is becoming a popular Mg source. Although Cp2Mg is a white crystalline solid with very low vapor pressure, which will lead to transport problems, recently 1 · 1018 hole concentration was obtained in OMVPE growth of GaN [508]. Coimplantation was also used to get high hole concentration. Kent et al. [509] used coimplantation of Be þ O and Mg þ O into GaN to achieve high activation efficiency. Yu et al. [510] used Be implantation into Mg-doped GaN samples. The hole concentration after annealing showed an increase from 5.5 · 1016 to 8.1 · 1019 cm3 determined by Hall measurements. Figure 4.121a shows the change in the hole concentration for GaN (squares) and Al0.08Ga0.92N (circles) as a function of the ammonia flux. These samples were grown consecutively, with the same background pressures of O, as judged by in situ mass spectrometry data and confirmed later by SIMS analysis. The substrate temperature and the growth rate were the same for each sample. Figure 4.121b plots hole concentrations in samples of GaN (squares) and AlGaN (circles) as a function of Mg/O ratio in the layer, as determined by SIMS. The hole concentration strongly depends on the Mg/O ratio. The Mg/O ratio of 40 results in the highest hole concentration of 2 · 1018 cm3 in GaN and in AlGaN. For these samples, SIMS data give Mg and O concentrations of 1.6 · 1020 and 4 · 1018 cm3, respectively. For the best samples the acceptor activation efficiency is thus about 1.3%. The excellent electrical quality of these samples, grown on sapphire, was confirmed by room temperature Hall mobilities of 40 10 cm2 V1 s1 for GaN and 8 2 cm2 V1 s1 for AlGaN. Although it might at first sound counterintuitive, and in fact counter to common perception, codoping with donor impurities along with acceptor impurities has been theoretically and experimentally shown to lead to increased hole concentrations.
Hole concentration(cm-3)
4.9 Intentional Doping
18
10
17
10
15
10
NH 3 flow rate (sccm)
(a)
Hole concentration(cm-3)
25
20
1018
1017
1016
1015 20
(b)
40
60
80
Mg/O ratio
Figure 4.121 (a) Hole concentration versus ammonia flux, solid square: GaN, and solid circle: Al0.08Ga0.92N. (b) Hole concentration versus Mg/O ratio. Solid square: GaN, and solid circle: Al0.08Ga0.92N. Mg concentration in the layer is (1–2) · 1020 cm3. Courtesy of Nikishin and coworkers [504].
However, chemical nature of donor impurities in GaN lattice and their relative tendency for chemical attraction appear to determine which donor acceptor combinations would be best. Because Mg is the most efficient acceptor impurity, this leaves the donor side open for experimentation. Among the candidates, codoping with O appears to be more effective than with Si. Oxygen and silicon donors behave differently as codopants in GaN:Mg, which indicates that the observed codoping effects are not simply due to Fermi level effects. There is experimental evidence that shallow acceptor complexes form upon codoping, as deduced from the measured
j1031
j 4 Extended and Point Defects, Doping, and Magnetism dependence of hole concentration on Mg dopant concentration for oxygen codoped samples. One thing is certain that codoping is a complex phenomenon and more detailed experiments in well-controlled templates are needed. 4.9.3.1.2 Use of Superlattices for Improving p-Type Conductivity Another promising method to get higher acceptor activation and lower resistivity is using AlxGa1xN/ GaN doped superlattices [511–513]. The idea of using AlGaN/GaN heterojunction for enhancement of p-type doping [514–516] has been proposed as a technique to increase the average hole concentration. Increased hole concentration at room temperature through the use of AlGaN/GaN superlattices have been reported [517,518]. The mechanism for hole enhancement is the periodic oscillation of the valence band edge. Acceptors are ionized where the band edge is far below the Fermi energy and the resulting holes accumulate where the band edge is close to the Fermi level, forming a confined sheet of carriers. Although the free carriers are separated into parallel sheets, their spatially averaged density will be much higher than in a simple bulk film. High electric fields due to both spontaneous and piezoelectric polarization within the strained AlGaN layers are expected to strongly impact the band bending within the superlattice. Hole concentration expected from an AlGaN/GaN superlattice both with and without the polarization fields taken into account has been calculated [516]. The fields create a periodic sawtooth variation in the band diagram as shown in Figure 4.122. When the polarization fields are present,
0.2
0.1
A
B
A
B
EMg EF
0.0
Energy (eV)
1032
-0.1
-0.2
EV
-0.3
-0.4
AlGaN
GaN
Figure 4.122 Calculated valence band diagram for the Mg-doped Al0.2Ga0.8N/GaN superlattice with spontaneous and piezoelectric polarization fields taken into account. The thickness of each layer is L ¼ 8 nm. The dashed line indicates the Fermi energy, and the circles represent the energy of the Mg acceptor with
AlGaN solid circles indicating the ionized form. Regions A and B defining each interface are also indicated. The sapphire substrate is on the left and the free surface of the film is on the right side of the figure. Courtesy of DenBaars and coworkers [516].
Spatially average hole concentration (cm -3)
4.9 Intentional Doping
18
10
GaN
AlGaN 17
Holemobility (cm2/Vs)
10
GaN
16
AlGaN 12
8
2
4
6
8
10
12
14
Well (barrier) width (nm) Figure 4.123 Room temperature Hall effect measurements on uniformly Mg-doped OMVPE-grown Al0.2Ga0.8N/GaN superlattices. The superlattice dimension L is varied. The arrows indicate values obtained on bulk samples of Mg-doped GaN and Al0.1Ga0.9N. The dashed lines are eye guides. Courtesy of DenBaars and coworkers [516].
a strong dependence on the superlattice dimensions is obtained: thicker layers yield larger potential changes from the polarization fields, and therefore, higher cracking. The measured hole concentration, spatially averaged, and mobility for the Mg-doped Al0.2Ga0.8N/GaN superlattices are shown in Figure 4.123 as a function of L, the thickness of the GaN and AlGaN layers (the layer thicknesses were kept equal so that the superlattice period is 2L). Further improvements were obtained by minimizing the ionized and neutral impurity scattering mechanisms through modulation doping that maximize the separation of the dopants (ionized scattering centers) from the multiple two-dimensional hole gases (2DHG). In this respect, the effect of modulation doping in Al0.20Ga0.80N/GaN using Hall effect and C–V profiling techniques has been investigated [518]. Mobility, resistivity, and carrier concentration were measured as a function of temperature. The modulation-doped (MD) and shifted-modulationdoped (SMD) samples were shown to have superior electrical properties compared to uniformly doped (UD) samples, especially at low temperatures. Furthermore, C–V profiles were presented that showed the 2DHG of the superlattice structure.
j1033
1034
j 4 Extended and Point Defects, Doping, and Magnetism Figure 4.124 shows self-consistent valence band diagrams of an MD SL (represents superlattice), a SMD SL, and a UD SL. The three ground state hole energies are E0 EF ¼ 5.9 , 21.7, and 21.7 meV, respectively. The term P is the self-consistently solved free hole concentration. The ground state is the only occupied subband at 90 K. The epilayer surface is on the left-hand side. The band diagrams shown in Figure 4.124 are calculated self-consistently using a one-dimensional (1D) Schr€ odinger–Poisson solver. An AlxGa1xN hole mass of (1.76 þ 1.77x) me,z, valence band discontinuity of 0.3 DEg, and energy gap Eg(x) ¼ (3.425 þ 2.71x) eV are used, x being the aluminum concentration. The calculated free hole concentration is shown in Figure 4.124 for each type of SL. At 90 K, only the ground states are occupied. The ionization energy of Mg in AlxGa1xN is not known precisely. However, the available data indicate that it increases from about 170 meV for x ¼ 0 (depending on the hole concentration which affects the screening) to 360 meV for x ¼ 0.27. A simple Vegard-like relationship can be used and put the acceptor level, EA, at (170 þ 704x) meV above the valence band. The calculated average free hole concentrations at 90 K are 2.3 · 1018, 3.3 · 1018, and 3.3 · 1018 cm3 for the MD, SMD, and UD SL, respectively. This agrees favorably with the measured values. A large improvement in mobility for the MD and SMD SLs versus the UD SL was achieved, especially, at low temperatures. At 90 K, the mobility (cm2 V1 s1) was 36, 18, and 2.0 for SMD, MD, and UD SLs, respectively. The free hole concentration versus temperature is presented in Figure 4.125. The effective acceptor activation energies are 16, 30, and 13 meV for the MD, UD, and SMD SLs, respectively, as determined in the temperature range of 250–390 K. These values are much smaller than activation energies of about 200 meV, depending on the doping level found in bulk GaN. Using Mg–Zn codoping in GaN by OMVPE, a low resistivity of 0.72 O cm and hole concentration of 8.5 · 1017 have been achieved [519]. It should be noted that Bis (cyclopentadienyl) magnesium (Cp2Mg) is a very popular Mg source. Although Cp2Mg is a white crystalline solid with very low vapor pressure, which leads to gas transport problems, hole concentrations as high as 1 · 1018 have been reported in OMVPE-grown GaN [520]. Coimplantation was also used to get high hole concentrations. Similarly, coimplantation of Be þ O and Mg þ O into GaN to achieve a high activation efficiency [521]. Again, in terms of codoping Be has been implanted into Mg-doped GaN [522]. The hole concentration after annealing showed an increase from 5.5 · 1016 to 8.1 · 1019 cm3 determined by Hall measurement, a figure that is certainly much higher than other reports, which may necessitate some degree of scrutiny. Despite gallant efforts to obtain p-type doping in GaN and AlGaN with high hole concentration, the problem is still a limitation for optoelectronic devices, particularly for shorter wavelength devices such as real solar-blind (l < 290 nm) detectors, or UV LEDs or LDs, which need p-type AlGaN with high Al composition. For example, a UV LED using AlGaN p–n junction with emission wavelength shorter than 290 nm, can only give submilliwatt or milliwatt power [523]. 4.9.3.1.3 Role of Hydrogen and Defects in Mg-Doped GaN Unlike undoped and n-type GaN, hydrogen has been reported to be very beneficial for p-type growth. It is
4.9 Intentional Doping
Al 0.20Ga0.80 N/GaN superlattices,T = 90 K shaded regions are Mg doped at 1019 cm-3 200
Modulation doped
EF - ---- EA
0 -200 -400
ΨGS
2
-600
EV
Energy E (meV)
GaN
AlGaN
Shifted modulation doped
200 0
- - -200 -
- - -
ΨGS
-400 -600
2
EF - EA
EV
GaN
AlGaN
Uniformly doped
200 0
- - -200 - -
-400
ΨGS
2
-600 600 620 640 660 680 700720 740 Position x (Å)
- - -
EF - EA
EV 760
780
800
Figure 4.124 Self-consistent valence band diagrams of an MD SL, an SMD SL, and a UD SL. The three ground state hole energies are E0 – EF ¼ 5.9 meV, 21.7 meV, and 21.7 meV, respectively. P is the self-consistently solved free hole concentration. The ground state is the only occupied subband at 90 K. The epilayer surface is on the left-hand side. Courtesy of Schubert et al. [518].
well established that growth of Mg-doped GaN in hydrogen ambient improves p-type conductivity after consequent annealing in N2 at temperatures above 700 C [524]. The need for postgrowth annealing to activate Mg in OMVPE-grown samples and reverting them to high-resistivity compensated material after ammonia annealing have both received a good deal of attention. It has been suggested that hydrogen passivates Mg [525], which is also supported by the calculations [526]. Moreover, they demonstrated that hydrogen is beneficial to p-type doping by Mg when compared to the hydrogen-free case, because the hydrogen passivates Mg during growth and thus inhibits the formation of native donors that self-compensate the acceptors. For OMVPE-grown films, it has been argued that H is in its positive charge state, the Hþ proton, and passivates the Mg acceptors that are in their negative charge state during growth; this would prevent the compensating donor-like defects from forming. During post-growth annealing, Hþ is driven out, which results in p-type
j1035
j 4 Extended and Point Defects, Doping, and Magnetism 1.0 0.9
Al 0.20 Ga0.80 N/GaN superlattices 100 Å /100 Å, Mg doped
0.8
Resistivity, ρ (Ω cm)
1036
Uniformly doped
0.7 0.6 0.5
Shifted modulation doped
0.4 0.3 0.2 0.1 0.0
Modulation doped
100
150
200
250
300
350
400
Temperature, T (K) Figure 4.125 Variable temperature three-dimensional carrier concentration data of an MD, an SMD, and a UD Al0.20Ga0.80N/ GaN SL. Courtesy of Schubert et al. [518].
GaN due to the negatively charged Mg acceptors. Although it has been mentioned frequently that Mg and H form a complex in Mg-doped GaN films, the exact mechanism of formation and the release of H upon a postgrowth treatment have not been sufficiently elucidated. In addition, predictions of the position of the H atom from first principles calculation are not consistent. For example, in the Neugebauer et al. [215] calculations, the antibonding site (i.e., a hydrogen sitting next to a N atom, about 1 Å away from the N site, at the antibonding position of the Mg-N bond) is lower in energy than the bond center site. This follows from the argument that the BC position requires an outward motion of the Mg and N atoms, which is very energy consuming in GaN, as GaN is a very hard material. On the contrary, an argument in favor of the BC location for H between substitutional Mg and N nearest neighbors has been provided [527]. In general, H takes the BC position in all other semiconductors. Figure 4.126 displays the passivating H at both the BC and the antibonding sites. Also useful is the determination of the local vibrational modes (LVMs) of the Mg–H complex in GaN, which provides not only the confirmation of hydrogen in GaN but also gives significant information on the structure of the complex. At the present time the stretch frequency is, in fact, the only reliably established physical parameter available from experiment for the Mg–H complex in GaN. To support their spectroscopic identification, Fourier-transform infrared-absorption spectroscopy has been performed on three Mg-doped GaN layers grown by OMVPE [424]. The first sample was as-grown and electrically semi-insulating; the second one was subjected to a thermal anneal and displayed p-type conductivity; and the third sample was exposed to monatomic deuterium at 600 C for 2 h, which increased the resistivity of the material. The as-grown sample displayed an LVM at 3125 cml, which is in very good
4.9 Intentional Doping
Figure 4.126 Proposed bond center and antibonding site incorporation of H in GaN and its passivation of Mg during growth (the Mg atom is directly below the H atom). In part courtesy of C. Van de Walle. (Please find a color version of this figure on the color tables.)
agreement with 3360 cm1 predicted by calculations [215]. After thermal activation of Mg, the intensity of this absorption line is reduced. After deuteration, a new absorption line appeared at 2321 cm1, which disappeared after a thermal activation treatment. The isotopic shift clearly establishes the presence of hydrogen in the complex. The controversy regarding the location of H at Mg–Ga in GaN can be resolved by a Rutherford backscattering experiment with a good-quality crystal and moderately heavy doping [528]. In this method a well-collimated beam of ions is directed down one of the channel axes and observe those that bounce back from a host atom that is not in its perfect lattice site. If H were bond centered then Mg would protrude out into the channel, which was calculated to be energetically unfavorable [215]. If H were in the interstitial space (antibonding site), then the incident ions would strike neither Mg, Ga, or N, but rather H; if there were any detectable recoil ions, which would be doubtful, they would easily be distinguished from those that struck other atoms. Additional support for the interstitial, antibonding site can be discerned from the following phenomenological argument [528]. The Hþ proton should act very much like an idealized positive test charge as it does not have to have its wave function
j1037
1038
j 4 Extended and Point Defects, Doping, and Magnetism orthogonalized to any of the electronic states. The very massive proton is also effectively a point charge, because the energy to confine its wave function to a small volume is less than that to confine the electron to the same volume. Another positive test charge is the positron eþ. A positrons wave function does not have to be orthogonalized to those of the electrons, but the energy to localize it in a small volume is the same as for an electron. In Si, Ge, GaAs, GaP, InSb, and most likely in GaN, and other semiconductors, eþ travels between interstitial spaces in a perfect crystal, whereas e has the least wave function density. This is a natural consequence of maintaining the mass and switching the sign of the charge. However, eþ also needs sufficient volume for its wave function so that the localization energy is not too large, and it finds this volume in the interstitial space. The analogy between eþ and Hþ suggests that Hþ will be in the interstitial site surrounded by N ions. However, because Hþ does not need all of the interstitial space, it might find a lower energy site with a much smaller volume. One might think that the bond site would be a logical candidate. There is a local peak in the electron density at the bond site and it does not result from the Coulomb potential of the ions. If it did result from the Coulomb potential, then obviously Hþ would be repelled rather than attracted. The extra electron density in the bond site results from the constructive interference between the atomic wave function centered on the two atoms participating in the bond; this would be a quantum mechanical effect. Thus, Hþ might reside in a low-energy state. However, if it does, it will then certainly perturb all the bonds in the area rather strongly, which is not energetically favorable over the interstitial antibonding site [215]. Unlike the strongly covalent semiconductors where Hþ is attracted to the bond center, the ionic nature of GaN would tend to expel Hþ into the interstitial site. Thus, it is the interstitial site that will offer the lowest energy [528]. The compensating nature of H is not unique to GaN in that it is known for other group III–V semiconductors [529]. Owing to the incorporation of a relatively large concentration of shallow-acceptor impurities as compared to the concentration of native donor-like defects in GaN, one finds that the Mg acceptor concentration is accomplished primarily through the presence of Mg-H complexes. This compensation is lifted (i) when samples are irradiated with an electron beam of 5–15 keV incident energy for several hours, or (ii) when samples are thermally annealed at constant temperature for half an hour. Also, first principles calculations demonstrate that the same amount of both Mg and H incorporated into the GaN films when they are grown under an H-ambient growth condition, such as that of OMVPE [526]. Furthermore, the calculations predict that more Mg can be incorporated into GaN film when there are more H atoms present. A large body of work exists on optical properties of Mg-doped GaN, which is discussed in Volume 2, Chapter 5. 4.9.3.2 Beryllium Doping Earlier work on Be doping by means of CVD was unsuccessful and led to the incorporation of Be as deep centers [530,531]. Predictions on the basis of ab initio calculations [532] point out that Be on Ga sites would form a shallow acceptor with a thermal ionization energy of 60 meV, and a double donor, if incorporated, as
4.9 Intentional Doping
interstitial. The error in this method of calculation could be 100 meV. Potentially, Be represents the shallowest acceptor level in wurtzite GaN. Until the experimental investigation of Salvador et al. [533], it was believed that Be doping leads to compensated material and yields high-resistivity GaN. Optical measurements in early samples, dating back to the 1970s, yielded a broad emission at energy of about 2.2 eV. It should be noted that this is at about the same energy as the notorious yellow peak found in GaN with native defects (Volume 2, Chapter 5). In an experiment with RMBE-grown films [533], the 380 nm emission peak is the dominant one for low Be doping levels. In samples with higher doping, 420–430 nm emission is observed. Dewsnip et al. [534], Sanchez et al. [535], and Ptak et al. [536] indeed observed the PL band that can be attributed to the shallow acceptor level introduced by BeGa acceptor in Be-doped GaN grown by MBE (Figure 4.127). The band represents a set of peaks separated by the LO phonon energy in GaN (about 91 meV). The peak at 3.38 eV has been attributed to the zero-phonon transition from the shallow donor to the BeGa acceptor [534–536], while that at 3.397 eV to the zero-phonon transition from the conduction band to the BeGa acceptor [536]. The ratio between the intensities of the zero-phonon peak and its phonon replicas is typical of the shallow acceptors with small Huang–Rhys factor (compare with the UVL band caused by the unidentified shallow acceptor in GaN). The assignment of the 3.38 and 3.397 eV peaks to the DAP and e–A transitions has been confirmed through their temperatures and excitation intensity dependencies [534–536]. The 3.38 eV peak decayed nonexponentially and shifted to lower energies with some time delay after pulse excitation, as it is expected
10
7
(e,A):3.397eV A,X:3.472eV 10
DAPand(A,X)-LO 3.38eV
6
PL intensity (au)
DAP(?)and3.27eV 10
3.30eV
AX 3.482eV
5
I 4(?) 3.14eV 10
4
BX LO
10
10
LO
LO
3
2
3.0
3.1
3.2
3.3
3.4
Energy (eV) Figure 4.127 Low-temperature PL from a Be-doped GaN grown by MBE. Courtesy of Myers and coworkers [536].
3.5
3.6
j1039
1040
j 4 Extended and Point Defects, Doping, and Magnetism for the shallow DAP transitions [535]. From the energy position of the DAP and e–A peaks, the ionization energy of the BeGa acceptor can be estimated as 100 10 meV, much lower than that of MgGa [534–536]. In another report [564] the 3.38 eV peak was not observed in Be-doped GaN, and the usual UVL band with the main peak at about 3.27 eV, observed in these works, is apparently related to residual shallow acceptor in GaN (SiN, CN, or MgGa). A small peak observed at 3.35 eV [537,677], which was attributed to BeGa, is questionable. Enhancement of the YL band with the maximum at about 2.2 eV has been noted in GaN samples implanted with Be [377,564,565]. Possibly it is not related directly to Be, but caused by implantation damage [538,564,565]. In the samples doped with Be during growth, an enhancement of emission in the visible part of the spectra has also been reported [534,535]. The emission represented very broad band peaking at 2.4–2.5 eVand related to some deep level or levels apparently formed by Be. In the early work of Salvador et al. [533], the Hall measurements in Be-doped GaN did not indicate p-type conductivity. However, hot-probe measurements indicated p-type conductivity with p–n junctions attained. Later investigations by Yu et al. [539,540] noted an increase in the room temperature hole concentration up to 2.6 · 1018 and even to 8 · 1019 cm3 [540] in Mg-doped GaN after Be implantation. Astonishingly though, Be–O codoping in cubic GaN grown on GaN has been reported to result in a high Be activation in excess of 1020 cm3, with hole mobilities in the 100 cm2 V1 s1 range [541]. One can conclude that extreme care must be exercised in determining the acceptor binding energies by photoluminescence measurements. 4.9.3.3 Mercury Doping Several groups have investigated Hg doping of GaN, but none reported on electrical measurements [406]. Pankove and Hutchby [542] performed optical measurements and observed 2.43 eV emission in Hg-doped samples. This is lower than the 2.9 eV emission reported by Ejder and Grimmeiss [543]. On the basis of measured emission, Ejder and Grimmeiss estimated that the Hg-acceptor level lies about 410 meV above the valence band. We should note that dopant incorporation in nitrides is a rather complex process. Many metals promote defect creation within the gap. Such defects are efficient optical centers and can easily be mistaken for acceptors due to metal incorporation. To this end, available data appear to suggest that unsuccessful attempts end up generating defects. Successful p-type doping, as confirmed by Hall measurements, can be achieved when the unintentional doping level is low, for example, at or below 1017 cm3. 4.9.3.4 Carbon Doping Carbon in both GaN and AlN is amphoteric in nature [210]. Its salient features in both materials are similar: carbon on a Ga or Al site, Ccation, is predicted to be an effective mass donor while carbon on a N site, Canion, is an effective mass acceptor. The ionization energies of the acceptor have been predicted to be equal to 0.2 in GaN and 0.4 eV in AlN. Incorporation of C on a nitrogen site is preferable because the CNformation energy is lower under both Ga- and N-rich conditions of growth. Carbon on a cation site, Ccation, can also assume a metastable DX-like configuration C GaorAl that
4.9 Intentional Doping
requires a broken bond between C and one neighbor. In this configuration, both the host N and carbon atoms are significantly displaced. C GaorAl introduces a singlet at approximately 0.4 and 0.3 eV above the valence bands of GaN and AlN, respectively, and a singlet occupied by one electron at about 0.3 and 1.0 eV below the conduction bands of GaN and AlN, respectively. Investigation of the formation of the þ Ccanion Ccanion nearest ion pairs brought to light that the binding energy is substanþ and C canion . tial and lower than the sum of the formation energies for C canion Therefore, the tendency toward self-compensation is expected to be large. However, for the preferred coordination, CN acceptors compensate residual shallow donors, resulting in semi-insulating GaN [544,545]. As the C concentration is increased the resistivity of GaN decreases, apparently due to formation of deep-level donors [544]. Several deep-level defects have been detected in C-doped GaN by employing methods such as thermally stimulated current spectroscopy [544], photoionization spectroscopy [545], and deep-level optical spectroscopy [546,547]. Most of these defects, however, are apparently not effects directly related to carbon [544]. In one report, hole concentrations up to levels of about 3 · 1017 cm3 with mobilities of about 102 cm2 V1 s1 in CCl4-doped GaN grown by MOMBE on GaAs substrates have been obtained. Unless the partial cubic nature of the resultant films is responsible for this seemingly high mobility, the results are otherwise considered controversial as the mobility reported is characteristic of n-type mobilities [548]. Annealing at 800 C did not increase the hole concentration that construed that the hydrogen passivation of acceptors is not significant. On the contrary, highly resistive CHx-doped GaN films resulted from plasma-assisted OMVPE. SIMS analyses indicated large amounts of C and H ( 1019–1020 cm3) in the samples. After annealing under a nitrogen atmosphere, the films remained highly resistive and suggest that the C–H complex is thermally stable [549]. As et al. [550–552] conducted a series of experiments dealing with C doping in c-GaN. Hole concentrations as high as 6.1 · 1018 cm3 with a mobility of 23.5 cm2 V1 s1 have been obtained. In samples with the hole concentration below 1017 cm3 the hole mobility exceeded 200 cm2 V1 s1. The e–A optical transition in a PL experiment was used to deduce an activation energy for C on the N site of about 215 meV in cubic GaN. Because carbon is present as a contaminant in most growth reactors (although graphite parts have been suggested as the likely source [554–561], the most likely source is hydrocarbons in MBE environment because properly cleaned high purity pyrolytic graphite has been used for crucibles with high quality layers resulting) and in metalorganic sources and possibly graphite parts exposed to high temperatures and hydrogen in OMVPE processes, its unintentional role in the compensation and doping of GaN layers might nevertheless be significant. It should be pointed out that any carbon contamination is most likely contributed by sources other than solid sources such as graphite parts, as the vapor pressure of graphite at growth temperatures is very low, although the effect on H should be considered. Of course, this picture is only applicable when the reaction pathways with those solid sources are rendered inconsequential. It is likely that C doping may be an attractive means of obtaining high-resistivity GaN, as opposed to, for example, Fe-doped GaN with its hysteresis effects.
j1041
1042
j 4 Extended and Point Defects, Doping, and Magnetism In terms of optical transitions, a correlation between the YL intensity and C doping has been attempted [554–561], although the properties of the YL band in undoped and C-doped GaN are apparently different [562,563] from those in undoped GaN. Enhancement of the YL band has also been observed in GaN after implantation of carbon [542,564,565], which may also create VGa resulting in the YL. The intensity of the UVL band apparently also increased with C doping [561]. However, in another report [566], a broad blue band peaking at 3.0 eV, instead of the UVL band, has been observed in highly resistive GaN:C. Seager et al. [566] attributed this blue band to transitions from CGa donor to CN acceptor, even though CN is more favorable. Polyakov et al. [556] also noted an enhancement of the blue band peaking at 3.05 eV in GaN samples heavily contaminated with carbon. Reuter et al. [559] observed a broad red luminescence band with the maximum at 1.64 eV in C-doped GaN. This band could be excited resonantly, and its excitation spectrum involved one or two broad bands in the range from 2.1 to 2.9 eV. The zero-phonon transition for this red band has been estimated at 2.0 0.2 eV [559]. When excited above the bandgap, the red band was buried under the stronger YL peaking at 2.2 eV. 4.9.3.5 Zinc Doping Attempts to achieve p-type conductivity by Zn doping of OMVPE-grown films were unsuccessful. Highly resistive films with resistivities up to 109 O cm were grown [567,568]. It has well been established [567] that among others, Zn causes a dominant level that is about 0.5 eV deep both in GaN and InGaN with a low InN mole fraction. In fact, this deep level was exploited in the earlier versions of InGaN LEDs marketed by Nichia Chemical Ltd in an effort to keep the InN mole fraction in the emission layer small while achieving 450 nm emission that is required by the display society. The nature of Zn centers has received some theoretical attention [569]. As many as four different Zn centers with energies in the forbidden gap, corresponding to emission between 1.8 and 2.9 eV, have been observed experimentally [570] and predicted theoretically [569]. While the 2.9 eV peak is attributed to substitutional Zn on Ga sites, the other three peaks have been assigned by experimentalists to various charged states of Zn on N sites, namely, ZnN ;ZnN2 ;ZnN3 , with Zn atoms binding to as many as three electrons having binding energies of 0.65, 1.02, and 1.43 eV. Because Ga and Zn are both somewhat comparable in size, it is not too difficult to imagine Zn in Ga sites. However, Zn in N sites requires large formation energies and are unlikely to form. The thermal activation energy of Zn in Ga sites has been estimated to be 0.33 eV [570]. Optical properties of GaN doped with Zn are discussed in Volume 2, Chapter 5. 4.9.3.6 Calcium Doping Suggestions have been made that Ca may form a shallow acceptor level in the GaN bandgap [571]. Ca p-type doping of GaN was achieved by ion implantation of Caþ ions or by a coimplantation of Caþ and Pþ followed by a rapid thermal annealing at temperatures 1100 C [572]. The ionization energy of 169 meV was found by temperature-dependent measurements of the hole concentration in a sheet, which should be considered with a good deal of caution. Ca acceptors, like those of Mg, can
4.9 Intentional Doping
be passivated by atomic hydrogen at low temperatures (250 C), and they can be reactivated by thermal annealing at 500 C for 1 h. Pankove and Hutchby [542], Monteiro et al. [573], and Chen and Skromme [574] studied PL from GaN implanted with Ca. A strong green luminescence band with a maximum at about 2.5 eV ensued Ca implantation and subsequent annealing [542,573]. Chen and Skromme [574] also observed another green luminescence band (with a maximum at 2.35 eV) in GaN:Ca; however, they noted that the 2.35 eValso appeared in their semi-insulating GaN samples doped with Mg and Zn and therefore may not be related to Ca. With increasing temperature from 14 to 300 K, the green band shifted from about 2.52 to 2.59 eV [573]. Monteiro et al. [573], proposed that the 2.5 eV emission is due to transitions from a deep compensating donor (0.6 eV below the conduction band) to the Ca-related acceptor (0.3 eV above the valence band). This identification is questioned as the deep DAP transitions are characterized with enormous shift of the band with excitation intensity [366] or with time decay after pulse excitation [479]. Note that no shifts of the green band with excitation intensity nor time delay were detected in Ca-doped GaN [573]. A more likely scenario is that the 2.5 eV band is related to transitions from the conduction band (or shallow donors at low temperature) to the CaGa acceptor. In this case the ionization energy of CaGa can be estimated at about 0.6–0.7 eV. Indeed, the zero-phonon transition for the 2.5 eV band can be expected at 2.8–2.9 eV from the shape of this band at low temperature. 4.9.3.7 Cadmium Doping Cadmium on Ga site (CdGa) is an acceptor. PL features in GaN doped with Cd have been the topic of a few investigations [377,574–577]. Cd-doped GaN exhibits a broad blue band peaking at 2.7–2.9 eV. Lagerstedt and Monemar [576] observed fine structure on the high-energy side of this band with the zero-phonon line at 2.937 eV and a set of LO and what appears to be local phonon replicas. The fine structure, caused by electron–phonon coupling, is similar to that in the Zn-related BL band, although the position of the PL band in Cd-doped GaN is red shifted by about 0.1 eV [228,439,574,576]. Lower photon energy of the Cd-related PL, as compared to Zn-related PL, is consistent with theoretical predictions on ionization energies of these two acceptors. Assuming that the assigning of the 2.937 eV peak in Ref. [576] is reliable, the ionization energy of Cd should be about 0.56–0.57 eV. At 77 K the decay of the Cd-related blue band is nonexponential, in the microseconds range [577]. No obvious variation in the shape and position of this band was observed at different time delays after pulse excitation. Transitions from the conduction band to the CdGa acceptor overlap with transitions from the shallow donors to the same acceptor may be responsible for the nonexponential decay of PL [577]. 4.9.3.8 Other Acceptors in GaN A plethora of group II elements have been used in an attempt to attain acceptors in GaN. In addition to the abovementioned acceptors, Pankove and Hutchby [377] examined the effect of Hg implantation on PL in GaN. Implantation of Hg resulted in a broad green band with a maximum at 2.43 eV. Assumption that this band is due to transition from the conduction band or shallow donor to the HgGa acceptor, the
j1043
1044
j 4 Extended and Point Defects, Doping, and Magnetism activation energy of the HgGa acceptor can be estimated as 0.8 0.2 eV, based on position and shape of this broad band. Pankove and Hutchby [377] also implanted several other elements in GaN that may act as acceptors such as Cu, Ag, Au, Sr, Ba, Li, Na, K, Sc, Zr, Fe, Co, Ni, Dy, and Er. All these elements gave broad PL bands with maxima in the range from 1.5 to 2.2 eV. In all cases deep levels are responsible for the observed PL. However, it is difficult to distinguish emission possibly related to these elements and arising from native defects formed by implantation damage. In case of transitional metals or rare earth elements, the emission in the red part of the spectrum may be also caused by internal transitions in d or f shells. The element Fe has been used for compensating unintentional donor toward achieving semi-insulating (SI) GaN. In this vein Vaudo et al. [578] obtained SI freestanding substrates by hydride vapor phase epitaxy. Variable temperature resistivity measurements were used to determine the resistivity of an iron-doped GaN sample to be approximately 3 · 105 O cm at 250 C that extrapolates to 2 · 109 O cm at room temperature by linear fitting. The near-infrared photoluminescence at 1.6 K exhibited sharp emission at 1.3 eV, which is most likely associated with the 4 T1 ðGÞ ! 6 A1 ð6Þ internal transition of the Fe3 þ charge state. Refer to Section 4.9.6.2 for discussion of optical transitions associated with Fe-doped GaN. 4.9.4 Doping with Isoelectronic Impurities
Isoelectronic impurities, such as AsN and PN, are considered as promising candidates for the mixed anion nitride alloys such as GaAsxN1x and GaPxN1x. The dilute nitride semiconductors in terms of their electronic/optical properties are discussed in Section 1.5.5. In this section, the focus is on GaN doped with small concentrations of isoelectronic impurities. Of particular interest here is the formation of deep levels in these alloys, attributed to AsN and PN, which would induce discontinuous changes in the size of the bandgap [385]. 4.9.4.1 Arsenic Doping Arsenic has been used as a surfactant for the growth on GaN. In addition, As incorporated in GaN introduces a broad blue band peaking at about 2.6 eV [377,574,579–582]. This blue band is usually structureless, although a welldefined phonon-related fine structure associated with it has also been reported [574]. The intensity of the 2.6 eV band increases monotonically with increasing As doping [435,437,580,582], while the concentration of free electrons remains nearly unchanged [583,584]. Several reports noted that intensity of this band is nearly insensitive to temperature in the range from 10 to 300 K [579,580,583]. An activation energy of 50 meV has been estimated associated with quenching the 2.6 eV band between 100 and 300 K [579]. Note that the total decrease in intensity in the work of Li et al. [579] work was only about a factor of two. It can be argued the small decrease might be related to the variation in the capture cross section [142], and the real quenching of this band might take place above room temperature with an unknown activation energy.
4.9 Intentional Doping
The time-resolved PL study revealed that the decay of the 2.6 eV band is nearly exponential even at low temperatures, with a characteristic time of about 0.1 ms [579,581]. The lifetime decreased from 92 to 77 ns with increasing temperature from 8 to 100 K, and subsequently increased up to 148 ns with further increase of temperature up to 300 K [581].Such behavior is typical for internal transitions, rather than those involving the states associated with the conduction or valence bands. Chen and Skromme [574] observed the fine structure on the high-energy side of the blue band related to As implantation. The ZPL was observed at 2.952 eV, followed by two LO phonon replicas. Moreover, the spectrum contained peaks separated from the ZPL by 38 and 75 meV that were attributed to other lattice phonon modes [574]. If the 2.6 eV band were due to recombination of exciton bound to the As-related defect, as it was suggested in Refs [574,579,581], its localization energy would be estimated as 535 meV, the energy difference between the zero-phonon line and free-exciton line [574]. However, if transitions from the conduction band to the As-related defect were responsible for the 2.6 eV emission, the energy level of the As-related defect would be about 560 meV above the valence band. A more detailed analysis, including comparison of the PL lifetime with concentration of free electrons at different temperatures, is warranted for definitive identification of the 2.6 eV band in Asdoped GaN. 4.9.4.2 Phosphorus Doping Similar to the case of As, implantation of P in GaN results in a broad blue–violet luminescence band peaking at about 2.9 eV [377,574,585]. This band has been reported to quench above 100–150 K with an activation energy of about 168–180 meV [150]. Chen and Skromme [574] observed the fine structure of the blue–violet band in GaN:P. The zero-phonon line at 3.200 eV was followed by two LO phonon replicas (91 meV) and a few other phonon replicas (separated by 39, 59, and 77 meV from the zero-phonon line) attributed to various lattice phonon modes [574]. The energy difference between the zero-phonon line and the free-exciton line (287 meV) gives the exciton localization energy if the blue–violet band is due to the isoelectronic PN bound exciton [574]. 4.9.5 Doping with Rare Earths
Strong emission sources in the range of 1.3–1.54 mm are in some demand for optical communications on the basis of silica fibers. For these applications, wide bandgap semiconductors present an advantage in terms of room-temperature stability compared to systems based on narrow-bandgap semiconductors. In addition, wide bandgap nature of GaN and related compounds allow transitions due to rare earths in visiblewavelengths.Initialworkfocusedonrareearth ErinGaNandAlNlayers. Among them is one dealing with AlN doped with Er that produced a strong PL at 1.54 mm. Incorporation of levels in the range of 3 · 1017–2 · 1021 were achieved in MOMBE systems with a solid Er source [586]. A plethora of activity intended particularly for visible applications ensued, which is discussed in Volume 2, Chapter 5.
j1045
1046
j 4 Extended and Point Defects, Doping, and Magnetism 4.9.6 Doping with Transition Metals and Rare Earths
Transition metal impurities are interesting from two points of view. One deals with acceptor-like doping in the context of electronic properties. The other deals with magnetic properties when the transition element concentration is relatively high but still within the dilute limits as to not change the main structural nature of the GaN matrix. On the magnetic side, extending the carrier-mediated magnetic interaction from diluted GaAs:Mn to zinc blende GaN:Mn, Dietl et al. [587] predicted hightemperature ferromagnetism in diluted MnxGa1xN with x 0.05 and potential application of this material to spin transport electronics (spintronics). This started a flurry of activity in GaN doped with Mn and also other transition metals with impetus being provided by the lure of electronics using charge and spin, primarily the latter. 4.9.6.1 Manganese Doping for Electronic Properties In contrast to GaAs where Mn is a relatively shallow acceptor (EA 0.1 eV) providing high concentration of free holes at room temperature, the Mn level in GaN is almost in the middle of the bandgap [588–591], and one cannot expect detectable transport of holes bound to Mn acceptor even at very high concentrations of Mn. When the host material is doped with Mn, emission energies associated with the internal transitions within the Mn atom are modified by the crystal field. Moreover, transitions involving levels induced by Mn in the host material appear, the specifics of which depend on the interaction of the Mn atom with the host material. The nature of the internal transitions in Mn as an isolated atom as well as that affected by the tetrahedral crystal field such as that in GaN is shown in Figure 4.129 for the Mn2þ state. The Mn2þ (d5) levels depicted in Figure 4.129 are expected to hybridize with s–p bands of the host and broaden into d bands, still narrow. An insight can thus be gained by discussing the nature of intratransitions in free Mn atom followed by the same in isolated Mn2þ occupying a cation site in the host material. Transition metal elements have valence electrons corresponding to the 4s orbital, and have partially filled 3d shells, thus the name 3d transition metals (i.e., Mn with the shell structure of 1s2 2s2 2p6 3s2 3p6 3d5 4s2). The partially filled shells of transition metal ions warrant a discussion of protocol involved in labeling the ground and excited states. Consider a free or isolated ion all the electronic shells of which are filled except one which is the 3d shell in transition metal elements and 4f shell in rare earth elements. Suppose the electron levels in the partially empty shell are characterized by an orbital angular momentum l that assumes the value of 2 for 3d states (here n ¼ 3, and l ¼ n 1 ¼ 2). In the particular shell there are 2l þ 1 states [l, (l 1), . . . , 0, . . . , (l 1), l] (assign a letter designation for each of lz) each of which can have two electrons, one with spin up and one with spin down, which would result in 10 states for the 3d shell. If the electrons were not to interact with each other, the ionic ground state would be degenerate. However, this degeneracy, albeit not completely, is lifted by electron–electron Coulomb interaction and electron spin–orbit interaction. The lowest levels after the degeneracy is lifted are governed by a simple set of rules, Russel–Saunders coupling (or LS coupling) and Hunds rules, which come
4.9 Intentional Doping
about as a result of complex calculations [592]. The former rule states that the Hamiltonian is commutative with the total electronic spin angular momentum (S), orbital angular momentum (L), and the total electronic angular momentum (J ¼ L þ S). Because the filled states have zero orbital spin (L ¼ 0), the eigenvalues determine the quantum numbers that in turn describe the configuration of the partially filled shell and the ion. The latter rule has three components, one of which states that the electrons that lie lowest in energy have the largest total spin while adhering to the exclusion principle. This means, for example, that all the spin-up electrons must occupy the partially empty shell while adhering to the exclusion principle. The second Hunds rule prescribes that the total angular momentum L of the lowest lying states has the largest value without violating the first Hunds rule and the exclusion principle. The value is equal to the largest magnitude that lz can have, which means that the first electron will go into the state with the largest |lz|. Because the first Hunds rule indicates that the second electron must have a spin to maximize the spin, it must go into the second state with the same spin as the first. Continuing on with P this rule in mind, the value of L can be calculated using L ¼ j lz j. With the first and second Hunds rules, one can determine the values of S and L, leaving (2L þ 1)(2S þ 1) states that can be further configured according their total angular momentum J. The third rule then helps determine the J values as J ¼ |L S| for n (2l þ 1) and J ¼ |L þ S| for n (2l þ 1) or J ¼ (L S#), and (L þ S"). The 3d band of the Mn2þ ion is exactly half-filled with five electrons among the 10 available states, with a gap between the up-spin (") occupied states and empty down-spin (#) states. For other transition metals, such as Fe, Co, Ni, one of the bands is usually partially filled (up or down), as shown in Figure 4.128. Table 4.10 shows the oxidation and charge states for some of the transition metals in ZnO and GaN [593]. The TM–d bands of the transition metal hybridize with the host valence bands (N–p bands in GaN) to form the tetrahedral bonding. This hybridization gives rise to the exchange interaction between the localized 3d spins and the carriers in the host valence band. In the simplest of pictures, the s-band of the conduction band does not mix with the TM–d bands, but it is still influenced by the magnetic ion. For an element with five 3d electrons (Mn), it means that all five have electrons that would have spin up. The total spin S is calculated at S ¼ (1/2)(n# n") for which Mn (3d) would be (1/2)(5) or 5/2 because all d shell electrons have the same spin. In Fe, however, there are six 3d electrons and one of them would have to have a spin-down configuration. In this case the total spin would be 1/2(5 1) ¼ 2. The methodology for arranging the electrons in the d shell for transition metals including orbital angular momentum, spin angular momentum, total momentum, and the designation for each of the available 10 d shell states is shown in Table 4.11. Because the electronic shell structure of rare earths is not as readily in the mind of the casual reader, it is provided in Table 4.12. The same is tabulated in Table 4.13 for rare earth elements in terms of the 4f shell configuration using the Hunds rules discussed above. As mentioned in the preceding paragraph, the element Mn is a unique case in that it has only five electrons in its 3d shell, half of all the available states, which are all in
j1047
1048
j 4 Extended and Point Defects, Doping, and Magnetism V
Cr
Mn
Fe
Co
Ni
4s23d3
4s13d5
4s23d5
4s23d6
4s23d7
4s23d8
Cu 4s13d10
4s 3d 3 p6 3 s2 2 p6 2 s2 1 s2
Figure 4.128 Electronic configuration of the 3d-states and 4ssates of transition metal elements (from V to Cu).
their ground state, following the Hunds rule, which calls for all electrons to maximize the spin angular momentum ("""""). As tabulated in Table 4.11, the electrons in their ground d shell (in Mn 3d5 state) have their orbital angular momentum L ¼ 0 and spin angular momentum S ¼ (1/2)(5) ¼ 5/2 (the orbital angular momentum L is quenched). The excited states of the d shell electrons would have L ¼ 1, 2, 3, and 4 values and are designated by the letter, in the same order P, D, F, G. For electron to make the transition to one of these excited states, one electron must flip its spin (""""#) that changes the spin quantum number to S ¼ (1/2)(3) ¼ 3/2. In the notation of spectroscopy, the states are labeled with left superscript 2S þ 1 XJ . Because the ground state has S ¼ 5/2 and L ¼ 0, the nomenclature used is 6 S5=25=2 . Similarly, L ¼ 1, 2, 3, and 4 states (with S ¼ 3/2) are labeled as 4 G, 4 P, 4 D, and 4 F. Among these excited states, the 4 G (S ¼ 3/2, L ¼ 4) level has the lowest energy as shown in Figure 4.129. The transition from the 6 S ! 4 G dominates the optical spectra involving a free Mn atom. Because of this reason, the S ¼ 1/2 states ("""##), where an additional electron flips its spin, which lie above the 4 G level, are not discussed. Let us now turn our attention to optical properties of Mn-doped GaN. Some background information is necessary to facilitate the discussion as well as getting
Table 4.10 Expected oxidation and charge state of some candidate transition metals in ZnO and GaN [593].
ZnO Acceptor (negative charge) Neutral Donor (positive charge) Double donor (2þ charge)
3d3
3d4
Cr3þ Mn4þ
Cr2þ Mn3þ Fe4þ
3d5
3d6
Crþ Mn2þ Fe3þ
Mnþ Fe2þ
GaN
Acceptor Neutral Donor
1 2 3 4 5 4 6 7 8 9 10
lz ¼ 1 " " " " " " "# "# "# "#
lz ¼ 2
" " " " " " "# "# "# "# "# " " " " " " "# "# "#
lz ¼ 0
The up and down arrows represent the spin up and spin down. P L ¼ j lz j ¼ jðl 2 Þ þ ðl 1 Þ þ ðl0 Þ þ ðl þ 1 Þ þ ðl þ 2 Þj The label name X is determined as follows: L ¼ 0; 1; 2; 3; 4; 5; 6 X ¼ S; P; D; F; G; H; I
Cu,Zn
Cr,Mn (3d5) Mn (3d4) Fe Co Ni
Sc Ti V
Element
d shell electrons (l ¼ 2) n
" " " " " " "# "#
lz ¼ þ 1
" " " " "#
"
lz ¼ þ 2 1/2 1 3/2 2 5/2 2 2 3/2 1 1/2 0
S ¼ (1/2)(n" n#)
Table 4.11 The electronic configuration for the d shell in transition metals, and the relevant spin, orbital, and total angular momentum along with the nomenclature used.
2 3 3 2 0 2 2 3 3 2 0
L¼j
P lz j 3/2 2 3/2 0 5/2 0 4 9/2 4 5/2 0
J ¼ (L S"), and (L þ S#)
D3=23=2 F2 4 F3=23=2 5 D0 6 S5=25=2 5 D0 5 D4 4 F9=29=2 3 F4 2 D5=25=2 1 S0 3
2
Symbol x in the form 2Sþ1XJ
4.9 Intentional Doping
j1049
1050
j 4 Extended and Point Defects, Doping, and Magnetism Table 4.12 The electronic shell structure of some of the rare earth elements that are pertinent to the content of this text.
Number
57 58 59 60 61 62 63 64 65 66 67 68 69 70
Element
K
L
M
N
O
P
Q
2 sp 26 26 26 26 26 26 26 26 26 26 26 26 26 26
3 spd 2 6 10 2 6 10 2 6 10 2 6 10 2 6 10 2 6 10 2 6 10 2 6 10 2 6 10 2 6 10 2 6 10 2 6 10 2 6 10 2 6 10
4 spdf 2 6 10 2 6 10 2a 2 6 10 3 2 6 10 4 2 6 10 5 2 6 10 6 2 6 10 7 2 6 10 7 2 6 10 9a 2 6 10 10 2 6 10 11 2 6 10 12 2 6 10 13 2 6 10 14
5 spdf 26126-26-26-26-26-26-26126-26-26-26-26-26--
6 spdf 2 2 2 2 2 2 2 2 2 2 2 2 2 2
7 s
La Ce Pr Nd Pm Sm Eu Gd Tb Dy Ho Er Tm Yb
1 s 2 2 2 2 2 2 2 2 2 2 2 2 2 2
Note that Eu and Gd both have seven electrons in their 4f shells. a Note irregularity.
acquainted with the terminology used. When Mn is introduced into GaN, the neutral configuration of Mn is Mn3þ (3d4) or (A0) when viewed as replacing Ga3þ in the lattice. However, in a II–VI material only the 4s2 electrons would be needed for bonding and as such the Mnþ2(3d5) configuration would be the neutral state. Owing to the metal oxide compound terminology, this state is also called the oxidation state. The configuration of Mn as an ionized acceptor is Mn2þ(3d5), with five tightly bound electrons in the Mn d shell. The neutral configuration of Mn3þ(A0) may be realized in two ways, as Mn3þ(3d4) with four electrons localized in the Mn d shell [594] or as an Mn2þ(3d5) þ hole with five electrons in the Mn d shell and a band hole weakly bound in a delocalized orbit [595]. When an Mn atom is substitutionally placed into a tetrahedral crystal of the II–VI type or said to be in its oxidation state (to be more descriptive Mn2þ or Mn2þ(3d5), which also indicates that only two 4s2 electrons participate in bonding), the four neighboring anion atoms exert a crystal field. This field causes the ninefold degenerate 4 G state to split into 4 A1 (non degenerate), 4 E (twofold degenerate), 4 T2 (threefold degenerate), and 4 T2 (threefold degenerate) states. Of these, 4 A1 and 4 E states nearly coincide and are hardly affected by the crystal field as shown in Figure 4.129. In the notation used to describe the intra-Mn2 þ transitions, the ground state label 6 S gives way to 6 A1 that is spherically symmetric and nondegenerate. The calculations show that the crystal field lowers the energies of both the 4 T2 and 4 T1 states. Transitions between the 6 S (S ¼ 5/2, L ¼ 0) and any of the excited states (S ¼ 3/2, L ¼ 1, 2, 3, 4) in free Mn atoms are forbidden because DS ¼ 0 spin parity requirement is
Tb Dy Ho Er Tm Yb
lz ¼ 2 " " " " " " " "# "# "# "# "# #"
lz ¼ 3
" " " " " " " "# "# "# "# "# "# "#
f shell electrons (l ¼ 3) n
1 2 3 4 5 6 7 8 9 10 11 12 13 14 " " " " " " " "# "# "# "# "#
lz ¼ 1
" " " " " " " "# "# "# "#
lz ¼ 0
" " " " " " " "# "# "#
lz ¼ þ 1
" " " " " " " "# "#
lz ¼ þ 2
" " " " " " " "#
lz ¼ þ 3 1/2 1 3/2 2 5/2 3 7/2 3 5/2 2 3/2 1 1/2 0
S ¼ (1/2) (n" n#) 3 5 6 6 5 3 0 3 5 6 6 5 3 0
P L ¼ j lz j
The up and down arrows represent the spin up and spin down. Note that both Eu and Gd have seven 4f shell electrons. P L ¼ j lz j ¼ jðl Þ þ ðl 2 Þ þ ðl 1 Þ þ ðl0 Þ þ ðl þ 1 Þ þ ðl þ 2 Þ þ ðl þ Þj The label name x is determined as follows: L ¼ 0; 1; 2; 3; 4; 5; 6 X ¼ S; P; D; F; G; H; I
Ce Pr Nd Pm Sm Eu,Gd
element
Table 4.13 The electronic configuration for the f shell in transition metals, and the relevant spin, orbital, and total angular momentum along with the nomenclature used.
5/2 4 9/2 4 5/2 0 7/2 6 15/2 8 15/2 6 7/2 0
J ¼ (L S"), and (L þ S#)
F5=25=2 H4 4 I9=29=2 5 I4 6 H5=25=2 7 F0 8 S7=27=2 7 F6 6 H15=215=2 5 I8 4 I15=215=2 3 H6 2 F7=27=2 1 S0 3
2
Symbol x in the form
2S þ 1
XJ
4.9 Intentional Doping
j1051
1052
j 4 Extended and Point Defects, Doping, and Magnetism 4F 4D 4P
CB (S = 3/2,L = 2,3,4)
1, 4s 2
4G
T2 3d 4
1.5 eV E 1.8 eV
6S
VB
(S=5/2,L=0)
Isolated Mn atom
Isolated Mn 3d 4 atom
Figure 4.129 A schematic diagram showing the splitting of the lowest excited states of the 3d5 level ð4 GÞ relative to the ground state ð6 SÞ for an Mn þ þ (or Mn2 þ (d5)) ion in isolated case (left) and in the presence of a tetrahedral crystal field. The arrows indicate possible intra-Mn transitions. The picture is similar for the Mn3 þ
Mn+++ (Mn3+) in a tetrahedral crystal field
shell. The transition with energy of around 1.8 eV in GaN is assigned to the direct emission of holes from Mn3 þ acceptors to the valence band, and the transition observed at around 1.5 eV is assigned to the internal spin-allowed 5 E ! 5 T transition of the deep neutral Mn3 þ state.
not satisfied. For Mn2þ isolated ions in the crystal, the selection rules are relaxed by the lack of inversion symmetry and crystal field. Therefore, during transitions from 6 A1 ground state to the excited states, derivatives of the 4 G states become possible. This is one reasonwhyadifferentnomenclature,fromthatusedforintratransitions infree Mn, is used for isolated Mn2 þ ion. Of the possible transitions, 6 A1 ! 4 T1 has the lowest energy and therefore constitutes the most important transition. The energetics of this transition are discussed in GaN lattice environment later in this section. As mentioned above, the Mn2þ ion is ideally suited in conjunction with discussions dealing with II–VI materials wherein Mn donates two of its electrons (4s2) to bonding which statistically in turn spend more of their time in the orbitals of the O atom making it O2. However, in III–V compounds such as GaAs and GaN, three electrons are needed for forming bonds with anions. This means that either one of the 3d5 electrons would have to be given up to satisfy bonding or an electron would be taken from a donor if the semiconductor contains some, which all compound semiconductors do to varying degree. The former gives rise to an Mn3þ(d4) ion. This further means that both Mn2þ and Mn3þ configuration would be present simultaneously and both must be considered. The Mn3þ configuration has intratransitions that are similarly split due to the crystal field exerted on the ion. In Td symmetry that is applicable to GaN and GaAs, electron spin resonance (ESR) measurements have been undertaken to investigate the splitting of the M3þ levels.
4.9 Intentional Doping
Graf et al. [590,591] have demonstrated that holes bound to Mn in GaN:Mn are located in the d-shell, in contrast to GaAs [596]. The transition associated with Mn3þ and Mn2þ levels, also referred to as the Mn3þ/2þ acceptor level, is located at 1.8 eV above the valence band to an accuracy of about 0.2 eV. The majority of Mn is present in the neutral Mn3þ state in (Ga,Mn)N with 0.2–0.6% of Mn grown by plasma-enhanced MBE. The deep level of Mn states would hinder the presence of free electrons or holes in (Ga,Mn)N. Graf et al. also observed the characteristic absorption spectrum with the ZPL at 1.42 eV, same as that observed by Korotkov et al. [588,589], who inferred their results from internal transition 5 E ! 5 T in Mn3 þ (d4). Transition of holes from Mn3 þ to the valence band can also be seen in the absorption spectrum at photon energies above 1.8 eV [590,591]. Note that when the GaN:Mn sample is codoped with Si, electron is captured by Mn acceptor, converting it from the Mn3 þ (d4) state into Mg2 þ (d5) state. Attempts have been made to determine the nature of intraion levels in the GaN environment. One such investigation is the X-ray absorption spectroscopy (XAS). In Mn L3,2 X-ray absorption, 2p core electrons are excited to unfilled 3d states; therefore, this technique is a direct probe of the magnetically active Mn 2d band. Mn L3,2 X-ray absorption is also highly sensitive to the oxidation state of 3d transition metal ions, and thus can yield valuable information when applied to transition metal doped semiconductors. X-ray absorption in (Ga,Mn)As and (Ga,Mn)N showed that the hybridization between Mn states and charge carriers is reduced in (Ga,Mn)N, which is consistent with the absence of ferromagnetism order across the sample [597]. Similar conclusion was drawn recently from electron spin resonance studies of (Ga, Mn)N. The multiplet structures in XAS is characteristic of a mostly d5 ground state, indicating predominantly Mn2þ impurities [597]. EPR is a technique that can be brought to bear to determine the energy levels associated with magnetic ions in the GaN environment. Wolos et al. [598] carried out EPR magnetization and optical absorption studies in bulk GaN crystals doped with Mn and some samples codoped with Mg acceptors. They concluded that the charge state of the Mn ion in GaN depends on the Fermi level. In n-type samples, Mn is an ionized acceptor A center of Mn2þ(d5). However, in the highly resistive samples, with lowered Fermi level, Mn is most probably in neutral configuration, A0, in the form of either a localized Mn3þ(d4) or a delocalized Mn2þ(d5) þ hole center. Optical absorption spectra of Mn-doped and Mn/Mg codoped samples show typical absorption bands related to Mn. Those authors interpreted these as arising from photoionization of Mn2þ(d5) to GaN conduction band in n-type samples and photoionization of neutral Mn A0 to GaN valence band in highly resistive samples. The location of the Mn acceptor level was derived as 1.8 0.1 eV below the bottom of the GaN conduction band. Note that the accuracy of these measurements is at best about 0.2 eV. Given the fact that GaN has a low temperature bandgap of 3.5 eV, this result agrees with the previous discussion [590]. The observations were interpreted with the help of a coordination diagram shown in Figure 4.130 describing the valence band to Mn2þ level and Mn2þ level to the conduction band. Moreover, Graf et al. [590,591] reported the Mn3þ/2þ level to be 1.8 eV above the valence band maximum. Relevant to the dilute magnetic semiconductor properties, the magnetization data reveal
j1053
j 4 Extended and Point Defects, Doping, and Magnetism 5
A 0+e CB
4
Energy (eV)
1054
Mn 2+(d5)+h ν 3
A -(Mn 2+)
GaN 2
A 0+hν 1
A 0+eCB
E CB opt
Mn 2+(d5)+holeVB
A 0+e VB
E VB opt
0
Q1
Q2
Configuration coordinate Figure 4.130 Configuration coordinate diagram showing Mn-related optical transitions in GaN where parabolas represent the energy levels. The optical transitions occur without changing the configuration coordinate. For example, at point Q1 the direct E VB opt optical transition represents A0 þ hn ! Mn2 þ
ðd5 Þ þ holeVB and at point Q2 the direct E CB opt transition represents Mn2 þ ðd5 Þ þ hn ! 0 A þ eCB . The thermal energies are determined by the minima in the configuration parabola, which are indirect. The relaxation energy is then the difference between the optical transition energy and the thermal energy [598].
Brillouin-type magnetization with Mn2þ(d5) (Mn spin S ¼ 5/2), whereas the highly resistive samples show magnetic anisotropy characteristics in n-type crystals. The highly resistive samples show magnetic anisotropy characteristic for nonspherical transition metal configurations. As mentioned in the previous paragraph, the d5 ground state of Mn was also observed by Edmonds et al. [597] via their Mn L3,2 X-ray absorption study. Wolos et al. [599] also investigated the magneto-optical properties of intracenter absorption band related to neutral Mn acceptor in GaN bulk crystal. The band is built of a ZPL at 1.4166 eV followed by GaN phonon spectrum. No splitting of the ZPL was observed. The only characteristic feature recorded for high magnetic fields was a steplike behavior of the spectral position of ZPL measured in Faraday configuration. The step appeared at a magnetic field of about 7 T, with a shift in energy of about 1.3 meV. The behavior of the band in magnetic field may be reasonably explained in terms of a model of Mn3þ (d4) ion in trigonal crystal field, undergoing Jahn–Teller distortion and spin–orbit coupling, strongly supporting the localized character of Mn neutral acceptor in GaN. Turning our attention to standard photoluminescence measurements, Korotkov et al. [444,600] observed PL transition due to Mn acceptor. It represents a broad structureless band with a maximum at 1.27 eV and FWHM of about 0.26 eV. With increasing temperature from 13 to 300 K, the Mn peak position remained nearly independent of temperature. Transient PL study of this band revealed a very slow
4.9 Intentional Doping
exponential decay of the PL intensity with the characteristic lifetime of about 8 ms [588]. A slow exponential decay of PL is typical of internal transitions. Consequently, it is most likely that the 1.27 eV band is caused by transitions of holes from the excited states of Mg3þ, located close to the valence band, to its ground 5 T level. The Mn-related optical properties in GaN have also been studied as a function of Mn concentration [601] as determined by the so-called elastic recoil detection method. In this experiment, the Mn doping level was varied between 5 · 1019 and 2.3 · 1020 cm3. The optical absorption spectra recorded at 2 K of nominally undoped GaN, (Ga,Mn)N, and (Ga, Mn)N:Si are replotted in Figure 4.131. The absorption peak at 1.414 eV can be attributed to an internal 5 T2 ! 5 E transition of the neutral Mn3 þ state. The intensity of this peak was found to scale with the Mn3 þ concentration in transmission spectroscopy measured at 2 K. In addition, the CL measurements showed that Mn doping concentrations of around 1020 cm 3 reduced the near band
GaN Mn (2.3 x 1020 cm-3)
Absorbance (au)
GaN Mn (1.1 x 10 20 cm-3 )
GaN Mn (5 x 1019 cm-3)
GaN Mn Si
GAN(u) 1.40
1.42
1.44
1.46
1.48
1.50
1.52
Energy(eV) Figure 4.131 Optical absorption spectra at 2 K for MBE-grown undoped, Mn-doped, and Mn þ Si codoped GaN. The Mn concentration varied from 5.4 to 23 · 1019 cm3. The ZPL is at 1.414 0.002 eV. Courtesy of Gelhausen et al. [601].
1.54
j1055
1056
j 4 Extended and Point Defects, Doping, and Magnetism edge emission intensity by about 1 order of magnitude. A complete quenching of the donor–acceptor-pair band at 3.27 eV and strong decrease of the yellow luminescence centered at 2.2 eV were attributed to a reduced concentration of VGa. In the infrared spectral range of 0.8–1.4 eV, three broad Mn doping related CL emission bands centered at 1.01, 1.09, and 1.25 eV were observed, the genesis of which may be from deep donor complexes generated as a result of the heavy Mn doping. The electronic states of (Ga,Mn)N codoped with Mg acceptors have also been investigated utilizing both PL [602] and PLE [603] spectroscopies. It was found that by adding Mg acceptors in (Ga,Mn)N, the weak Mn-related PL band at 1.3 eV was quenched. The change in PL spectra indicates that the Mg addition stabilizes the Mn4þ charge state by decreasing the Fermi level. A series of sharp PL peaks were observed at 1 eV in codoped epilayers and were attributed to the intra-d-shell transition 4 T2 ðFÞ 4 T1 ðFÞ of Mn4 þ ions and their phonon replicas are believed to be involved as well. The relative intensities of the sharp peaks were found to be strongly dependent on the excitation wavelength, indicating that the optically active Mn4 þ centers involved in the separate peaks are different. The temperature dependence of the PL spectrum suggests the presence of at least three distinct Mn4 þ complex centers. On the basis of the optical studies of Han et al. [603], the energy levels of Mn4þ ions in the GaN band diagram can be given as drawn in Figure 4.132. The energy levels in Figure 4.132 can be used to understand the PLE spectrum (shown in Figure 4.133) of the Mn4 þ ½4 T2 ðFÞ 4 T1 ðFÞ transition, which reveals intracenter excitation processes via the excited states of Mn4 þ ions. The contribution involving the contribution by the 1.1 eV transition using the Luckovsky fit [604] is shown in dashed lines in Figure 4.133a. PLE peaks at 1.79 and 2.33 eV shown in Figure 4.133b, after removing the Luckovsky fit, are attributed to the intra-d-shell transitions from the ground 4 T1 (F) state to the 4 T1 (P) and 4 A2 (F) excited states,
Conduction band Mn4+(4A 1(F))
2.3 eV
Mn4+(4T1(P))
1.8 eV Mn4+(4T
2(F))
3.3 eV
1.0 eV Mn4+/3+
Mn4+(4T1(F))
1.11 eV (Lucovsky fit) Mg3+/2+
Valence band Figure 4.132 Energy levels of Mn4þ ions in wurtzite GaN. The energies are given for a crystal temperature of 20 K. Courtesy of Han et al. [603].
4.9 Intentional Doping
Lamp background
PL Intensity (arb. units)
Calibrated PLE response 10
Lucovsky fit
E =1.11 eV
T=20K
th
1 1.0
1.5
2.0
2.5
3.0
3.5
4.0
Energy (eV)
Figure 4.133 Calibrated PLE spectrum of (Ga,Mn)N codoped with Mg at 20 K; (b) the 20 K PLE spectrum after subtracting the Lucovsky fit. Courtesy of Han et al. [603].
respectively. In addition to the intrashell excitation processes, a broad PLE band involving charge-transfer transition of the Mn4 þ /3 þ (or 4 þ /3 þ ) was observed. As determined from the onset of this PLE band, the position of the Mn4 þ /3 þ deep level is 1.11 eV above the valence band maximum, which is consistent with ab initio calculations by Gertsman [605], where the Mn4 þ /3 þ transition level is calculated to be 1.1 eV above the VBM (see Figure 4.135a for GaN transition metal transition levels). PLE results further indicate that 4 þ is the predominant oxidation state of Mn ions in p-type GaN:Mn when the Fermi energy is below 1.11 eV, as measured from the top of the valence band maximum. The optical and electrical properties of GaN films implanted with Mn have been jointly studied [606] by means of optical transmission spectra, microcathodolumines-
j1057
1058
j 4 Extended and Point Defects, Doping, and Magnetism 1A
1S
1
1T 1G
1E
Cr 4+ 3P
1D
2
A1
Cr 4+
V 3+
T2
4.303
3.789
E
4.264
3.743
T1
3.316
3.050
T2
3.174
2.773
3.145
2.743 2.645 2.537
1T
1
3T
1
E
1A
1
T1 A1 A1 T2
V 3+
1T 1A
1
1E 3T
1
1E 3F 3T
2
3A
2
2
2.827
A1 E
2.740
A1 T1
1.958
1.580
T2 E
1.917 1.670
1.537
T1 E
1.208
0.941
T2 A2
1.188
0.925
T2
0 eV
0 eV
1.658
Figure 4.134 Energy level before and after crystal field splitting for the d2-type V3þ and Cr4þ ions on the Ga site in GaN calculated by a numerical diagonalization of the static crystal field energy matrices, neglecting the effects due to the wurtzite trigonal field [608].
cence spectra, capacitance–voltage and capacitance–frequency curves, temperature dependence of resistivity, and DLTS with both electrical and optical injection. Optical transmission on n-type GaN samples implanted with high doses of Mn (3 · 1016 cm2) shows that Mn forms a deep acceptor near Ev þ 1.8 eV. The Mn complexes formed with native defects are deep electron traps with a level near Ec 0.5 eV, which are most likely responsible for a strong blue luminescence band with energy near 2.9 eV.
4.9 Intentional Doping
3+/2+ 3
2+/+
3+/2+ 4+/3+ 3+/2+
2
4+/3+
GaN Eg
Energy (eV)
3+/2+ 3+/2+
4+/3+
3+/2+ 2+
3+/2+
4+/3+ 1 4+/3+
5+/4+ 4+/3+ 0 Ti
V
Cr
Mn
Fe
Co
Ni
Cu
(a)
Figure 4.135 (a) Charge transfer levels of transition metal impurities in GaN. Data in this figure include the results from Gerstman et al. [605] (n), Mahadevan and Zunger [618,619] (?), Heitz et al. [620] (?), Graf et al. [590] (*), Baur et al. [611] (n), Van Schilfgaarde and Myrasov [621] (s), and Han et al. [603] ( ). (b) Calculated charge transfer levels of isolated 3d TM defects in group III
nitrides, AlN, GaN, and InN. The valence and conduction band edges determined via the Langer-Heinrich rule using Cr4þ/3þ as the reference level are also shown to get a glimpse of transition element levels in the nitride host. Note that the 1.9 and 6.2 eV bandgaps for InN and AlN, respectively, have not been updated to 0.8 and 6 eV. Courtesy of A. Blumenau and Ref. [605].
j1059
1060
j 4 Extended and Point Defects, Doping, and Magnetism It is worth noting that, apart from Mn forming deep-acceptor levels in (Ga,Mn)N as reported in the literature, an optical study by Yoon et al. [607] pointed out the presence of an Mn acceptor level at about E ¼ 3.08 eV for a low Mn content (Ga,Mn)N. This suggests that the Mn-bound holes in III nitride exhibit the shallow impurity states. Arrhenius plots of the intensity of the Mn acceptor give an activation energy of E ¼ 310 meV, indicating that the observed thermal quenching of the Mn-related PL peak is due to the dissociation of an acceptor-bound hole. The observation by Yoon et al. [607] is consistent with predictions based on the effective mass theory of the shallow hydrogenic impurity state model. 4.9.6.2 Other TM Doping for Electronic Properties While Mn, for traditional reason, has been investigated in detail in many semiconductors with GaN not providing the exception, other transition and rare earth elements may turn out to be more, if not just as, important. Among the transition elements, V, Cr, Fe have also been investigated and employed for possible ferromagnetism. Baur et al. [608] discussed the optically active transitions associated with these three transition metals in unintentionally doped GaN (Fe3þ and Cr4þ in OMVPE films and V3þin HVPE layers). In terms of the interaction with the host material, hybridization through the second nearest neighbor interaction between the Cr 3d levels and Ga 4s levels leads to gap states, the details of which can be found in Ref. [609]. The focus here is not to entertain the question as to what the source of unintentional impurities is, but rather to what the configuration and energy levels associated with d–d transitions are. Serendipitously, many GaN samples studied by Baur et al. [610] have shown an infrared PL band with a ZPL at 1.30 eV. This line has been previously assigned to the 4 T1 ðGÞ ! 6 A1 ðSÞ crystal field transition of Fe3 þ impurities (internal transition). Moreover, a similar luminescence line has been reported in AlN also and used in conjunction with GaN to determine the band discontinuities [611], as discussed in Volume 2, Chapter 4. The luminescent lifetime of the Fe3 þ PL in GaN is considerably long at 4.5 ms [612], because the radiative transition involved is spin forbidden. Optical properties of p-type GaN epilayers prepared by metalorganic chemical vapor deposition and subsequently implanted with Fe þ ions have also been investigated [613]. The implant activation was accomplished at 850 C for 30 s. Optical transitions observed during photoluminescence measurements at 2.5 eV and around 3.1 eV have been attributed to the presence of Fe. The photoluminescence peak at 2.5 eV has been identified as a donor–Fe-acceptor transition and that at 3.1 eV as a conduction band–Fe acceptor transition. In investigating V in GaN, Baur et al. [608] observed a new PL band with a ZPL at 0.931 eV with a lifetime less than 0.4 ms. This particular band was observed in most GaN HVPE samples but not in the OMVPE samples investigated. At elevated temperatures a hot ZPL appeared 13 cm1 above the main ZPL. Some phonon replicas of the 0.931 eV ZPL were in proximity in energy to intrinsic lattice phonons of GaN [614]. This particular band has been assigned to neutral vanadium impurities with a configuration of V3þ (3d2). Due to the relative weakness of the 0.931 eV PL line,
4.9 Intentional Doping
Baur et al. were not able to measure the excitation spectra. In addition, inherently thick HVPE layers showing the V3þ PL line also exhibited absorption bands at 2.54 and 2.74 eV. Baur et al. [608] also investigated transitions associated with Cr in GaN with particular characteristics of the PL band, with a ZPL at 1.19 eV appearing in all OMVPE and most HVPE GaN layers. Well-resolved TO phonon replicas appeared at about 550 cm1 below the ZPL. At elevated temperatures a hot ZPL appeared at 25 cm1 above the main ZPL. The PL band in question could be very efficiently excited in the red spectral range as well as with various Krþ laser lines ranging between 2.4 and 3.5 eV albeit with the efficiency being at least a factor of 20 less than in the peak of the band. In comparing OMVPE samples with different positions of the Fermi level EF (induced by changing the doping level such as n-type undoped, lightly and heavily Mg acceptor doped), one observes a dramatic increase in excitation efficiency with a drop of EF from near the conduction band to a position deep in the forbidden gap. The genesis of transitions associated with V and Cr warrant further discussion. As the case made for Mn, transition element ground and excited states are split further by the crystal field present. Typically, the transitions to the ground state in crystal field configuration are active. The PL bands associated with V and Cr discussed above can be accounted for by intraelement transitions that are termed as d–d transitions. The PL band observed at 0.931 eV at 2 K can be ascribed to V3þ(3d2) based primarily on the associated ZPL doublet structure and a comparison with the well-understood PL spectra of V3þ in other semiconductors such as GaAs, GaP, and InP [615,616]. The bands observed in those semiconductors result from the 3 T2 ðFÞ ! 3 A2 ðFÞ crystal field transition of V3 þ . Within 3 T2 ðFÞ the first-order spin–orbit splitting is quenched by a moderately strong dynamic Jahn–Teller effect [615]. In this case the four 3 T2 ðFÞ spin–orbit levels, namely, A2, T2, E, and T1 levels, shown in Figure 4.134, merge into a lower lying triply and a higher lying sixfold quasi-degenerate levels separated by a second-order spin–orbit splitting. The cold and the hot ZPL observed correlate well with transitions from these two levels to the T32 A2 ðFÞ ground state (1.88 and 0.925 eV for Cr4 þ and V3 þ , respectively). The splitting between the two lines amounts to 10, 15, and 12 cm 1 for GaAs, GaP, and InP, respectively, appears to be D ¼ 13 cm1 in GaN, and falls into the range observed for V3þ in the conventional III–V semiconductors. Baur et al. [608] also noted that the trigonal field splitting due to the axial field in wurtzite GaN is not observable in PL, indicating that either the trigonal field is small or that such splitting is also quenched within 3 T2 ðFÞ by the Jahn–Teller interaction. The energetic position of the V3 þ 3 T2 ðFÞ ! 3 A2 ðFÞ PL band is to a first extent determined by the cubic crystal field parameter Dq that in GaN is 18% larger than in GaP [617]. For comparison, the 3 T2 ðFÞ ! 3 A2 ðFÞ main ZPL transition in ZnO occurs near 0.85 eV. However, the cation replaced by the transition elements is divalent in ZnO as opposed to trivalent in III–V semiconductors such as GaP:V3 þ and GaN:V3 þ in that the charge state of the host ion replaced (2 þ in ZnO) differs from the impurity charge state (3 þ ).
j1061
1062
j 4 Extended and Point Defects, Doping, and Magnetism While it is natural to ascribe the Cr4þ PL excitation band in GaN to the higher lying 1E(D) and 3Tþ (F) crystal field states of the 3d2 TM element configuration, the origin of the 2.54 and 2.75 eV absorption bands associated with V3þ is less evident. To gain some insight, Baur et al. [608] numerically diagonalized the static crystal field energy matrices, including Coulomb interaction (parameters B and C), the cubic crystal field (parameter Dq), and the spin–orbit coupling (parameter l). The results for V3þ and Cr4þ using the parameter sets listed in Table 4.14 are displayed in Figure 4.134. The results support the assignment of the Cr4þ PL excitation band to the 1 E(D) and 3 T1 (F) states. Further, they suggest that the 2.54 and 2.74 eV absorption bands are related to the 1 T þ (D), and the 3 T þ (P) states of V3 þ , respectively. Once the spin–orbit splitting of 3 T2 (F) predicted by static crystal field theory is known, the dynamic Jahn–Teller model developed earlier for this state can be applied. The dynamic Jahn–Teller model predicts a splitting, D, that matches the observed splitting for V3þ and Cr4þ in GaN. This further supports the Cr4þ assignments discussed above. Before closing the discussion of (Ga,Mn)N, the charge transfer levels of transition metal impurities in GaN reported in the literature reported by Gerstman et al. [605], Mahadevan and Zunger [618,619], Heitz et al. [620], Graf et al. [590], Baur et al. [611], Van Schilfgaarde and Myrasov [621], and Han et al. [603] are summarized in Figure 4.135a. These are of fundamental interest in interpreting and predicting the magnetic properties of GaN-based diluted magnetic semiconductors (DMSs. They are also naturally critical in interpreting the optical transitions in TM-doped GaN. To be complete, the calculated charge transfer levels of transition metal elements Ti, V, Cr, Mn, Fe, Co, and Ni in all the three nitride binaries (cubic AlN, GaN, and InN) obtained by Gerstman et al. [605] are shown in Figure 4.135b. Assuming that the Langer- Heinrich rule is valid, one can also estimate the valence band discontinuities, which leads to 0.91, 1.05, and 2.05 eV for cubic AlN/GaN, GaN/InN, and AlN/InN, respectively. In producing the figure, the Cr4þ/3þ donor level has been taken as the reference level for the alignment of the band edges of all host nitride materials. Figure 4.135b shows that the rule represents a reasonably good approximation, especially for the acceptor levels Co3þ/2þ and Ni3þ/2þ, and also for the donor level V4þ/3þ. The data of Figure 4.135b
Table 4.14 Crystal field parameters in units of cm1 for V3þ and Cr4 þ in GaN.
Cr4þ
k B C Dq
V3þ
Free ion
GaN
Free ion
GaN
11 77 1 040 4 264
115 800 3720 986
105 860 4128
89 810 3660 754
4.9 Intentional Doping
also include the experimentally determined charge transfer levels of transition elements in GaP reported in Refs [622,623] to make the point that the internal reference rule for, for example, determining band discontinuities, to be less reliable when the group III nitrides and other III–V semiconductors are combined, which is not all that atypical to do. The genesis for this inapplicability may have to do with the difference in electronegativity between nitrogen (3.04) and phosphorus (2.19) [624]. In summary, it is believed that a combination of optical, magneto-optical, EPR, ESR, electrical and magnetic measurements, not necessarily in that order, may be required for a full characterization of the state of these impurities in GaN, also applicable to ZnO, and their participation in any magnetization. 4.9.6.3 General Remarks About Dilute Magnetic Semiconductors Transition metal doped semiconductors, dating back to II–VIs such as (Zn–Mn)S, (Zn–Mn)Se, and (Cd–Mn)Te, [625] and later to arsenide-based III–Vs, such as In1xMnxAs [626] and Ga1xMnxAs [627] followed by GaN and ZnO, have been considered for magentoelectronics. If the magnetic moment of transition metal elements can be made aligned, the degree to which that alignment is controlled could pave the way for many device concepts. The II–VI system suffers from a low critical temperature (<30 K) above which magnetic ordering is not maintained. The temperature in GaAs-based system is near 170 K and still improving as modulation doping methods are employed to increase the hole concentration. Heavy doping of GaN with magnetic ions goes beyond the realm of discussion of doped GaN in the hope of obtaining p-type material and extends over into the realm of magnetism. It in this context that a first-order discussion of magnetism in solid is given, which should provide a smoother segue into the issues related to ferromagnetics in GaN. If above room-temperature ferromagnetism were to be accomplished in GaN and/or related nitrides, that might possibly form the base for charge, spin-based, or mixed spin and charge-based devices. The devices utilizing spin in one form or another fall into a newly coined nomenclature, spintronics as in electronics, but utilizing spin in one form or another. As spun, spintronics is a new paradigm in which the spin degree of freedom of the electron, that could also include nuclear spin, which is not discussed here, is harnessed either by exploiting the spin property in conventional charge-based devices or utilizing the spin alone. For successful incorporation of spin into existing semiconductor technology, several technical issues such as efficient injection, transport, control and manipulation, and detection of spin polarization as well as spin-polarized currents must be resolved. Faster and less power-consuming transistors have been reported as being a possibility because flipping the spin takes 10–50 times less power and is 10 times faster than transporting an electron through the channel in traditional FETs. Challenges, however, are formidable in that in addition to coherent spin injection, the device dimensions must be comparable if not less than the spin coherence lengths. For example, to create spin FETs in GaN, the channel length must be much shorter, about 200 nm, although use of wires has been predicted to increase the dephasing length.
j1063
1064
j 4 Extended and Point Defects, Doping, and Magnetism In diamagnetic and paramagnetic materials, small applied magnetic fields lead to an internal magnetic induction that is directly proportional to the applied field through Bint ¼ mH ¼ mr m0 H;
ð4:69Þ
where mr and m0 represent the relative permeability of the medium and permeability of free space, respectively. m is called the permeability of the medium. If the sample is placed in an external magnetic induction B0 or an external magnetic field H, the internal magnetic induction Bint, can be expressed as (with the assumption that demagnetization effects are negligible and the internal magnetic field Hint can be approximated by the external magnetic field H which is justified for diamagnetic and paramagnetic materials) Bint ¼ B0 þ m0 M ¼ m0 ðH þ MÞ:
ð4:70Þ
We should note that many authors call B, not H, the magnetic field, induction already has other meanings in electrodynamics (such as induction of an electric field by changing the magnetic field). To avoid confusion, these two fields will be referred to as the H field and the B field from here on. With the aid of Equations 4.69 and 4.70, we find M ¼ ðmr 1ÞH ¼ cH
or
mr ¼ c þ 1;
ð4:71Þ
where w (dimensionless) is the magnetic volume susceptibility. For ferromagnetic materials Hint cannot be approximated by the external H field and B field can be expressed as B ¼ H þ 4pM
in cgs or
B ¼ m0 ðH þ MÞ in SI units:
ð4:72Þ
Simply, magnetization M is the magnetic moment per unit volume. Magnetic susceptibility w is simply the ratio of magnetization divided by the macroscopic H field: c ¼ M=H:
ð4:73Þ
Similarly, the permeability of the medium is given by the ratio of the B and H fields: m ¼ B/H ¼ m0(1 þ w). The magnetic polarization ( J), also called the intensity of magnetization (I), is defined as J ¼ m0M and its saturation value is depicted with the nomenclature of Js. Because cgs and SI units are used in the literature, a conversion table between the two as well as definition of pertinent magnetism parameters along with their units is tabulated in Table 4.15. Materials with w < 0 and small, which is independent of temperature, are called diamagnetic. Diamagnetism arises from the tendency of electrical charge partially screening the interior of the body from the applied magnetic field. All nonmagnetic materials are diamagnetic, which include all the nonmagnetic semiconductors such as Si, Ge, GaAs, GaN, and so on except when these materials are doped with transition metal elements or rare earths to render them magnetic. Diamagnetism can be understood in the realm that all atoms or ions have spin-paired electrons. In
4.9 Intentional Doping Table 4.15 The relationship between some magnetic parameters in cgs and SI units.a
Quantity B field (aliases magnetic induction, magnetic flux density H field (aliases magnetic field strength/intensity, applied field) Magnetization (M) Magnetization (4pM) Magnetic polarization (J) Specific magnetization (s) Permeability (m) Relative permeability (mr) Susceptibility (w) Maximum energy product (BHmax)
Gaussian (cgs units)
SI units
Conversion factor (cgs to SI)
G
T
104
Oe
A m1
103/4p
emu cm3 G — emu g1 Dimensionless — emu cm3 Oe1 MGOe
A m1 — T J T1 kg1 H m1 Dimensionless Dimensionless kJ m3
103 — — 1 4p · 107 — 4p 102/4p
G, Gauss; Oe, Oersted; T, Tesla. a http://www.aacg.bham.ac.uk/magnetic_materials/type.htm.
the absence of a magnetic field, the electron motion is spherically symmetrical, the angular momentum is zero, circulating current around the nucleus is zero, and the magnetic moment is zero. When a magnetic field is present, a Lorenz force is generated and the Lenzs law dictates that when the magnetic flux changes in a circuit, a current is induced to oppose that change. The centrifugal and centripetal forces are rebalanced by the magnetic force, causing the orbital frequency of an electron with orbital magnetic moment parallel to the field to slow down and with that antiparallel to the field to speed up. Classical and quantum mechanical treatments can be used to determine the magnitude of the magnetic susceptibility that is negative of course and ranges from 5.0 · 109 for Si to 1.6 · 104 for bismuth. When measuring thin films, the diamagnetic contribution from the substrate must be accounted for. To a first extent this contribution is temperature independent unless the temperature affects the ability of charge as it attempts to shield the bulk of the material from the applied magnetic field. One class of materials with w > 0 are called paramagnetic and electronic paramagnetism arises from, for example, atoms, molecules, and lattice defects, and unpaired electrons, causing a nonzero total spin. Free atoms and ions with a partially filled inner shell such as transition elements, ions isoelectronic with transition elements, rare earth, and actinide elements have unpaired electrons in those shells that lead to nonvanishing spin. Examples include V2þ, Cr2þ, Mn2þ, Fe2þ, Co2þ, and Ni2þ among the transition elements and Gd3þ among the rare earths, the latter of which is used very readily in semiconductors and attracted the most attention. The magnetization versus magnetic field H curves in these materials follow a linear relationship, as shown in Figure 4.136. When the magnetic moments of magnetic ions and electron spins are ordered, the material becomes ferromagnetic with nonzero magnetic moment (the susceptibility
j1065
1066
j 4 Extended and Point Defects, Doping, and Magnetism Ferromagnetic
Paramagnetic
Superparamagnetic
G
N
T
Figure 4.136 A cartoon showing ferromagnetic, paramagnetic, and superparamagnetic cases.
is positive and large), even for zero applied field, called the spontaneous magnetic moment. This alignment is broken due to thermal agitations above a certain critical temperature called the Curie temperature. The magnetization versus H curve for a ferromagnet is shown in Figure 4.136 also for temperatures below the Curie temperature. Above the Curie temperature an otherwise ferromagnetic material would behave as a paramagnetic material with linear dependence of magnetization on the H field. For completeness, the magnetic susceptibility in antiferromagnetic (AFM) materials is positive but small, and in ferromagnetic materials it is positive and very large. If the alignment is caused by exchange interaction with carriers, as in the case of GaAs:Mn, the control over the hole concentration would lead to control of the degree to which the material is ferromagnetic. The spin ordering can be simple or helical. If spins are antialigned in an alternating fashion, either simple ordering or canted ordering, the material is called antiferromagnetic. In transition metals having negative exchange energy such as Cr and Mn, electronic spins of adjacent atoms are held in opposite orientations that locally cancel each other. In Cr, which is body centered cubic, the body center atomic spins are directed opposite to those, but in equal number at the cube corners. In this particular case, the saturation magnetization would reduce as the temperature is increased up to a critical temperature called the Neel temperature beyond that the thermal agitations would render the material paramagnetic (in other words the thermal agitations break up the ordering) and thus magnetization would be a linear function of H field. Moreover, if portions of the material are ferromagnetic and others paramagnetic, the material is called superparamagnetic. Artistic views showing paramagnetic, ferromagnetic, and superparamagnetic cases are shown in Figure 4.136. In the paramagnetic case, the magnetic moments
4.9 Intentional Doping
associated with magnetic ions is randomly distributed. In the ferromagnetic case, the magnetic moments are aligned, the mechanisms for which are the topic of this section. In the superparamagnetic case, portions of the material is ferromagnetic and rest is paramagnetic. A schematic representation or M–H curves for diamagnetic, paramagnetic, and ferromagnetic cases is shown in Figure 4.137. In the paramagnetic case, the magnetization would eventually saturates as all the magnetic ions would have their magnetic moments aligned, to the extent possible, by the magnetic field. In the superparamagnetic case, the portions of the material are already ferromagnetic and for temperatures below the Curie temperature and TB (blocking temperature represents the temperature at which the metastable hysteretic response is lost), the M–H curves have the typical hysteretic shape as in the case of ferromagnetic materials. For temperatures above the Curie temperature but below TB, the M–H curves still have their hysteretic behavior but the curves do not saturate. Above both the Curie and TB temperatures, the material behaves as paramagnetic because the thermal agitations destroy any magnetic ordering.
Ferromagnetic M MS Mrem HC
•
•
Diamagnetic M
T
Virgin curve
H
H
M
M
T << T B T < T B < TC T > TB
H
Paramagnetic
H
Superparamagnetic
Figure 4.137 A schematic representation of the magnetization versus magnetic field, or M–H curves, for ferromagnetic, diamagnetic, paramagnetic, and superparamagnetic cases. The term Tc represents the Curie temperature above which ferromagnetism is destroyed by thermal agitations, and TB represents the temperature above which ordering of magnetic moments is destroyed.
j1067
1068
j 4 Extended and Point Defects, Doping, and Magnetism It is instructive to briefly discuss the rudimentary basis for magnetism in an effort to get acquainted with the terminology, basis for various types of magnetism, temperature dependence of magnetism, and values of fundamental nature such as Bohr magneton. In this realm, let us consider N atoms per unit volume, each with a magnetic moment m. Magnetic field, if any present, will align those moments, but thermal disorder would resist the tendency to align. The energy of interaction of the moment m with the applied field B is given by [628] U ¼ m B:
ð4:74Þ
In thermal equilibrium, the magnetization is given by the Langevin equation as M ¼ NmLðxÞ;
ð4:75Þ
where x : mB/kT, the Langevin function L(x) ¼ ctnh x 1/x, and N is the number of atoms per unit volume. For x 1 (or mB kT), L(x) x/3, which leads to the wellknown Curie law, M
Nm2 B C B ¼ cH; ¼ 3kT T m0
ð4:76Þ
where C is the Curie constant and is given by C : Nm2/3k. It should be noted that in some cases the Curie law is expressed in cgs units, with H replaced by B, since the applied field B can be expressed as m0H, and m0 = 1 in cgs units. The Langevin function for this regime is then expressed as LðxÞ ffi
x mB ¼ : 3 3kT
ð4:77Þ
Magnetic moment of an atom or ion in free space is given by m ¼ ghJ ¼ gmB J:
ð4:78Þ
The total angular momentum hJ (h is the Plancks constant) for the electronic system of an atom is the sum of the orbital angular momentum hL and spin angular momentum hS, meaning hJ ¼ hL þ hS. The factor g is called the gyromagnetic or magnetogyric ratio and is represented by the ratio of magnetic moment to h units of the h angular momentum. For electronicsystemsa quantity g is definedto satisfygmB g and is called the g-factor, spectroscopic splitting factor, or Lande splitting factor, as the magnitude of this factor determines how rapidly the energy levels split. Basically, this g-factor is represented by the number of Bohr magnetons divided by h units of angular momentum. For an electron spin g ¼ 2.0023, which is often truncated to 2. If the truncated value of 2 is used, the total magnetic moment of an atom can be expressed as m ¼ mB ðL þ 2SÞ:
ð4:79Þ
This magnetic moment precesses around the direction J and is customarily expressed in Lande terminology as in Equation 4.78. For a free atom with an orbital angular momentum, the g-factor is given by Lande equation as g ¼ 1þ
JðJ þ 1Þ þ SðS þ 1Þ LðL þ 1Þ : 2JðJ þ 1Þ
ð4:80Þ
4.9 Intentional Doping
mj
µs
1/2
-µ
h ω = 2µB -1/2
µ
Figure 4.138 A schematic diagram of spin-induced energy splitting for one electron with only spin angular momentum in a magnetic field normal to the surface. The term g, which represents what is called the g factor, has been substituted by 2 for conduction electrons. The term B represents the magnetic field.
The Bohr magneton is defined as mB ¼
eh 2mc
in cgs units
and
mB ¼
eh 2m
in the SI units:
ð4:81Þ
The energy level of the system in magnetic field is E ¼ mJ gmB B;
ð4:82Þ
where mJ is the azimuthal quantum number and has the values of J, J 1, . . . , 0, . . . , (J 1), J. For a simple spin without orbital moment mJ ¼ 1/2 and g ¼ 2, the total energy splitting of an electron state is given by 2mB with m ¼ gmBS where S is the spin term, as shown in Figure 4.138. In the low-energy state, the magnetic moment is parallel to the B field. The projection of the magnetic moment of the upper branch along the magnetic field direction is given by m and the same quantity associated with the lower branch is m. In the realm of basics of magnetization, let us now consider atoms in a magnetic field. As mentioned above, an atom with angular momentum quantum number J has J, J 1, . . . , 0, (J 1), J as many equally spaced energy levels, 2J þ 1 in total. The magnetization in that case is given by M ¼ NgJmB BJ ðxÞ where
x
gJmB B ; kT
with the Brillouin function BJ(x) given by 2J þ 1 ð2J þ 1Þ 1 x ctnh x ctnh : BJ ðxÞ ¼ 2J 2J 2J 2J 1 y y3 þ . . . : and þ y 3 45 M NJð J þ 1Þg 2 m2B Np2 m2B C ¼ ¼ ¼c 3kT 3kT B=m0 T pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi where p g JðJ þ 1Þ is the effective number of Bohr magnetons.
ð4:83Þ
ð4:84Þ
For x 1, ctnh y ¼
ð4:85Þ
j1069
1070
j 4 Extended and Point Defects, Doping, and Magnetism Electron has associated with it a magnetic moment that is equal to 1 Bohr magneton, mB. The Curie relation for conduction electron magnetization is M¼
Nm2B B; kT
ð4:86Þ
which is temperature dependent. Instead, nonferromagnetic materials have temperature-independent magnetization. Pauli argued that Fermi–Dirac statistic would apply to magnetization also and it would correct the problem and bring the temperature dependence in the picture. Once the Fermi level is taken into account M
Nm2B B T Nm2B ¼ B; kTF kT TF
ð4:87Þ
which is temperature independent and comparable to observations. In the absence of external magnetic field, Pauli magnetism at absolute zero also indicates that the number of electrons in the spin up and spin down states adjusts to make the energies equal at the Fermi level as shown in Figure 4.139. The chemical potential (Fermi level) of the spin-up electrons is equal to that of spin-down electrons as shown in Figure 4.139a. However, when a magnetic field is applied, electrons with one spin, in this case spin down, are moved in energy, albeit very small compared to that observed in the ferromagnetic state due to the large effective molecular field. Consequently, there would be excess of spin-up electrons that are not spin paired, Magnetic field, B
No magnetic field
Excess spin up electrons
E
E
EF
E F-
E F+ 2μ B
N(E)-
N(E)+
(a)
Figure 4.139 Density of states versus energy for the two spin components. In the nonmagnetic states (a), the occupancy by spin-up and spindown electrons is the same. In Pauli magnetism at absolute zero when a magnetic field is applied, the spin-down and spin-up electrons are moved away from each other, which leads to unparity in that the number of electrons with
N(E)-
N(E)+ (b)
one type of spin would dominate over the other (b). The effect of the applied magnetic field is amplified to show the point. In ferromagnetic material and below the Curie temperature, the effective field or the molecular field is so large; thus, this splitting is sizable and no external magnetic field is needed for the shift shown.
4.9 Intentional Doping
as shown in Figure 4.139b. It should be stated that the effect has been magnified to make the point. In ferromagnetic materials be it magnetic semiconductor or ferromagnet below the critical temperature, the picture depicted in Figure 4.139b would hold figuratively without external magnetic field. In ferromagnetic samples and for temperatures above the Curie temperature, the electronic structure is similar to that shown in Figure 4.139a. However, below the Curie temperature, the picture is similar to that shown in Figure 4.139b without any external magnetic field. An isolated nickel atom has the (3d84s2) configuration, and the energy bands in metallic Ni are filled up to the Fermi energy with 9.46 electrons in the 3d states and 0.54 electrons in the overlapping 4s states. When Ni is in a paramagnetic state, for example, above the Curie temperature, the spin-up and spindown bands are equally occupied. On the contrary when Ni is in a ferromagnetic state, meaning below the Curie temperature, the very large effective field produces a very large shift of the spin-up states with respect to the spin-down, states in the case of which the spin-down band is completely filled with five electrons while the spin-up band has only 4.46 electrons, giving rise to spin unparity of 0.54 electrons per atom. In semiconductors samples containing large concentrations of transition metal impurities, the density of states become distorted due to band tail states that cloud the picture somewhat, as shown in Figure 4.140. Another topic to which frequent references are made is the splitting of intramagnetic ion levels due to the crystal environment, termed the crystal field splitting. The same also is applied to elements that typically act as deep acceptors unlike rare earths No magnetic field
Magnetic field, B
E
E
EF
E F-
E F+ 2µB
N(E)-
N(E)+ (a)
N(E)-
N(E)+ (b)
Figure 4.140 Density of states in samples, such as dilute magnetic semiconductors, that are heavily doped with TM elements, causing band tailing. As in the case of Figure 4.139, electrons adjust their numbers to make the energies of spin-up and spin-down electrons to be equal at the Fermi level in the nonmagnetic state (a). The excess spin of the spin-up electrons in the magnetic state or with large magnetic field applied (b).
j1071
1072
j 4 Extended and Point Defects, Doping, and Magnetism where the 4f shell lies deep inside the ions within the 5s and 5p shells in transition metals or the iron group; the 3d shell is the outer most shell and experiences the intense and inhomogeneous electric field, which is called the crystal field. Of the immediate effect, this manifests itself as coupling of the orbital angular momentum (L) and spin angular momentum ( J) gets by and large broken up, and thus the states are no longer specified by their J values. Moreover, the 2L þ 1 sublevels associated with a given L, which are degenerate in the free ion, may be split up by the crystal field, as shown in Figure 4.141. Essentially, this splitting diminishes the contribution of the orbital motion to the magnetic moment. The wave functions giving rise to the charge densities shown in the figure are of the form z f (r), x f (r), and y f (r), and are called the pz, py , and py orbitals, respectively, as shown on the right side of the figure. The triply degenerate energy level present in the free ion is shown with a dashed line. If the electric field does not have axial symmetry, the energy level will all be different. Because it will come up time and again in our discussion of ferromagnetism in dilute magnetic semiconductors, it is timely to provide a rudimentary discussion of the mean field theory (MFT). Given an internal interaction that lines up the magnetic moments and makes the transition from the paramagnetic to the ferromagnetic state, the interaction can be construed as a magnetic field causing it. This field is called the
px, py
Free electron (d) pz
Magnetic ion z
z
z
y
x
y
x
(a)
y
x
(b)
Figure 4.141 The p-states associated with an atom with orbital angular momentum L ¼ 1 in a uniaxial crystal line electric field of the two positive magnetic ions along the z-axis. In the case of free ions, the states mL ¼ 1,0 are degenerate whereas in a uniaxial crystal the
(c) atom has a lower energy when the electron cloud is close to the positive ion (a) than when it is oriented midway between them (b) and (c). In an axially symmetric field, the px and py orbitals are degenerate. The energy levels for a free atom are shown in (d).
4.9 Intentional Doping
exchange field. The effect of exchange field is opposed by thermal agitations, and above certain temperature, which is the Curie temperature, ferromagnetism is destroyed. In the mean field theory the exchange interaction field BE is given by: BE ¼ lM;
ð4:88Þ
where l is a temperature-independent constant. However, for T > TC the system is a disordered paramagnet, and for T < TC the system is a ferromagnet. We should mention that the main idea behind MFT is to replace all interactions to any one body with an average or effective interaction. This reduces any multibody problem, which is generally very difficult to solve exactly, into an effective one-body problem. The ease of solving MFT problems means that some insight into the behavior of the system can be obtained at a relatively low cost even at the expense of loosing some accuracy. Let us find l in terms of TC. If wparamag is the susceptibility, then the magnetization is given by M ¼ cparamag ðHa þ HE Þ in cgs units m0 M ¼ cparamag ðBa þ BE Þ
in SI units
ð4:89Þ
where Ha and Ba are the applied fields. The susceptibility of a paramagnet is given by the Curie law cparamag ¼
C : T
ð4:90Þ
Using Equations 4.88 and 4.89, we obtain MT ¼ CðH a þ lMÞ and c¼
M C ¼ Ha T Cl
in cgs system:
ð4:91Þ
For TC ¼ Cl a singularity occurs in that w ! 1. For C < T < l, we have spontaneous magnetization and c¼
C T TC
or
c¼
C T q
in cgs units;
ð4:92Þ
which represents the Curie–Weiss law. In fact, the Curie law is a special case of the more general Curie–Weiss law. It should be mentioned that y is often used for the Curie temperature as well. We should also mention that in an antiferromagnetic material, the Curie law takes the form c ¼ C=ðT þ T C Þ
or c ¼ C=ðT þ qÞ in cgs units:
ð4:93Þ
j1073
1074
j 4 Extended and Point Defects, Doping, and Magnetism In the above equation y can either be positive, negative, or zero. The case when y ¼ 0 corresponds to the case when the Curie–Weiss law equates to the Curie law depicted in Equations 4.76 and 4.90. A nonzero y implies that there is an interaction between neighboring magnetic moments and the material is only paramagnetic above a certain transition temperature. If, on the contrary, y is positive, the material is ferromagnetic below the transition temperature and the value of y corresponds to the transition temperature (Curie temperature, TC). If y is negative, the material is antiferromagnetic below the transition temperature (Neel temperature, TN); however, the value of y does not relate to TN. We should note that this equation is only valid when the material is in a paramagnetic state. Similarly, it is not valid for many metals as the electrons contributing to the magnetic moment are not localized. However, the law does apply to some metals, such as the rare earths, where the 4f electrons that create the magnetic moment are closely bound. Detailed calculations show that [628] c¼
C ðT T C Þ1:33
;
ð4:94Þ
and
l¼
T 3kT ¼ : T C Ng 2 SðS þ 1Þm2B
ð4:95Þ
Let us apply what we learned so far to some known magnetic materials so as to gain an appreciation for the exchange interaction field BE. For Fe, TC ¼ 1000 K, g is about 2, and S is about 1 and l ! 5000, Ms 1700; thus BE ¼ lM ¼ 1700 · 5000 ¼ 8.5 · 106 Gauss. This exchange field in Fe is huge and much larger than the magnetic field due to ions in the crystal. A magnetic ion produces a field of BE mB/a3 103 G at the neighboring lattice point. From an experimental point of view, measured magnetization data are conveniently plotted following the method reported by Arrott [629] to determine the Curie temperature and susceptibility. The necessary analytical treatment used by Arrott is given here to facilitate the analysis of magnetization data. In the Weiss–Brillouin formalism, the magnetization is given by M ¼ Ms tan h
mðH þ NMÞ ; kT
ð4:96Þ
where Ms is the spontaneous magnetization at absolute zero, m is the magnetic moment per atom, and N is the molecular field constant. This expression can be rewritten as mH mM þN ¼ tanh 1 ðM=Ms Þ: kT kT
ð4:97Þ
4.9 Intentional Doping
And for M Ms, the above expression can be expanded in a power series as mH mM M 1 M 3 1 M 5 þ þ þ ...: þN ¼ kT kT Ms 3 M2 5 M2
ð4:98Þ
Taking the first two of the series on the right-hand side, we can write m H e 1 1 þ M2 ; ¼ kT M Ms 3 M 3s
ð4:99Þ
where e ¼ 1(Tc/T ). At the Curie point (w1 ! 0) m0 H 1 1 M2 : ¼ kT c M 3 M3s
ð4:100Þ
Using M ¼ wH and in the limit of M ! 0, we can deduce from Equation 4.99 that 1 kT N; ¼ c mM s
ð4:101Þ
and at the Curie point the left-hand side tends to zero, yielding Tc ¼
mN Ms : k
ð4:102Þ
Equation 4.99 can be rewritten as M2 ¼ 3eM 2s þ
3mM 3s H : kT M
By plotting M2 versus H/M, one can determine w1 from the intercept with the horizontal (H/M) axis. Extrapolating w1 versus temperature and noting that w1 ! 0 at the Curie point leads to the Curie temperature with the aid of Equation 4.102. At Curie temperature, e ¼ 0, and thus M2 versus H/M curve goes through the origin, as shown in Figure 4.142. The Arrott plots are utilized in the discussion in Section 4.9.6.3. 4.9.6.4 General Remarks About Spintronics Having discussed the optical signature and likely energy levels caused by Mn in the gap of GaN and given a short introduction to the types of pertinent magnetism, let us now turn our attention to spintronics on the assumption that GaN can somehow be made ferromagnetic at high temperature. The use of spin or spin and electrons in a device has attracted much attention, spanning over several decades. While switching and amplification are performed by devices utilizing the charge nature of electrons, magnetic storage media relies on spin. A device combining the charge and spin nature of electrons or just the spin nature would represent a new realm. This realm can be implemented, depending on the application, using magnetic materials suitably integrated with semiconductors and diluted magnetic semiconductors that are alloys between a magnetic elements, such as Mn, Fe, Co, Cr to name a few, and a semiconductor that can inject coherent spin. In other words, DMS depicts a class of
j1075
j 4 Extended and Point Defects, Doping, and Magnetism M 2 (magnetization, abu)
1076
ε< 0 ε= 0 ε>0
H/M
0 FE
AFE
Figure 4.142 The Arrott plot, magnetization squared versus the magnetic field divided by magnetization below the Curie temperature ((e < 0), at the Curie temperature (e ¼ 0) and above the Curie temperature (e > 0). The inverse of magnetic susceptibility, w1, can be determined from the intercept of M2 versus H/ M curve with the horizontal (H/M) axis. Repeating the procedure versus temperature and noting that w1 ! 0 at the Curie temperature, one can obtain the Curie temperature. Practically, this is when the M2
versus H/M curve, which would form a straight line at that point, goes through the origin. If the extrapolations intercept the horizontal axis on the negative side, the sample is ferromagnetic (FE) or the measurement temperature is below the Curie point. If the intercept occurs on the positive side of the horizontal axis, the sample is antiferromagnetic or the measurement temperature is above the Curie point. If the extrapolation goes through the origin, the measurement temperature is identical to the Curie temperature.
semiconductors in which magnetic ions of the same chemical nature but different charge states coexist [630]. Using spin-up and spin-down state in a two-state (binary) system, information can be stored with electron spin. Spin information can be transported, as spin is associated with charge carriers, electrons in this case. At least in metal the spin relaxation length is much longer than the momentum relaxation length that relaxes the eventual device dimensions in a spin-based structure. Spin can be detected optically, through charge–spin coupling, and by polarizers and analyzers made of ferromagnetic materials. Use of spin and charge together could lead to storage and processing capability or use of spin in ferromagnetic semiconductor/ semiconductor heterostructures might pave the way to such applications as quantum computing at the so-called qubit level (quantum bit) [626]. Purported potential advantages are nonvolatility, increased data processing speed, and decreased power consumption, to cite a few [631]. The ultimate hope or fiction for applications is that new spin-based multifunctional devices such as spin-FETs, spin-LEDs, spin resonant tunneling device (RTD), terahertz frequency optical switches, modulators, encoders, decoders, and quantum bits for quantum computation and communication will someday be possible, perhaps even replace the all mighty MOSFET. It should be mentioned that magnetic metals are used for recording and storage. Ferromagnetic semiconductors are researched for expanding the application areas of spin.
4.9 Intentional Doping
Spin transport differs from charge transport in that spin is a nonconserved quantity in solids due to spin–orbit and hyperfine coupling. Spin and charge dynamics can be controlled by external electric and magnetic fields and illumination by light, which can be used to create new functionalities not feasible or ineffective with conventional electronics. The details of these approaches can be found in review articles [632,633]. To take advantage of the spin degree of freedom in semiconductors, spin-polarized carriers must be created, sustained, controlled, and of course detected. The most plausible means for generating spin polarization electrically is by injection of spin-polarized electrons. Doing so with ferromagnetic metal–semiconductor junctions has not been easy, most likely due to scattering at the magnetic metal (Schottky barrier) semiconductor interface. Owing to the fact that efforts in GaN are very sketchy, examples utilizing other semiconductors are provided for the reader to glean an understanding of the physics and applications involved. Tunneling from a ferromagnetic metal through vacuum into a semiconductor has been shown to provide a high degree of polarization [634]. Various combinations of magnetic metal/ semiconductor interfaces have considered for spin-polarized injection with various degrees of success, such as hybrid Au/Co/Cu/NiFe/n-GaAs spin valve Schottky barrier [635], and semiconductor light-emitting diode structure with ferromagnetic metal contact having an injection efficiency of 30% that persists to room temperature [636]. Light emission results from injection of spin-polarized electrons under reverse bias with the aid of Schottky barrier formed at the Fe/AlGaAs interface that provides a natural tunnel barrier. The injected carriers radiatively recombine and emit circularly polarized light. Along similar vein, ballistic spin-polarized transport through a diluted magnetic semiconductor heterostructure with the inclusion of a nonmagnetic barrier has also been investigated wherein, for suitable magnetic fields, the output current exhibited a nearly 100% spin polarization for large forward bias [637]. Spin-polarized tunneling of electrons in a GaInAs/GaAs quantum well lightemitting diode structure from the valence band of GaMnAs into the conduction band of n-type GaAs with Si delta doping at the interface has also been explored [638]. Injected spin-polarized electrons have been detected in the form of circularly polarized light emission from the quantum well that corresponded to magnetooptical Kerr effect (MOKE) loops. Because the angular momentum selection rules are simplified by the strain-induced heavy-hole (HH)/light-hole (LH) splitting, a direct relation between circular polarization and spin polarization can be obtained. On the transistor realm, Si-based spin valve transistors operative at room temperature have been demonstrated using spin-dependent transport over Schottky barriers [639]. That a very good interface between ferromagnet and semiconductor is imperative cannot be overstated. How long the injected spin can exist depends on the spin relaxation time, which can be quite long in lightly doped nonmagnetic semiconductors [640]. For control of spin in a ferromagnetic semiconductor such as (Ga,Mn)As where the effect is carrier spin interaction induced, ferromagnetism might be used. By using gate-induced depletion of the carrier density, ferromagnetism may be turned on and off. For illustrative purposes, photo-generated carriers have been used to induce ferromagnetism in (In,Mn)As [641].
j1077
1078
j 4 Extended and Point Defects, Doping, and Magnetism A succinct glossary of spin polarization is given here for completeness. Spinpolarized transport occurs naturally in any material in which there is an imbalance of the spin populations at the Fermi level, similar to that shown in Figure 4.139b depicting the case without a and with b magnetic field. The disparity between spin-up and spin-down electrons in the ferromagnet or a diluted magnetic semiconductor below Curie temperature versus the normal metal or nonmagnetic materials does not require magnetic field for it to take place. As depicted in the figure, the necessary imbalance occurs in ferromagnetic metals owing to the density of states available for spin-up and spin-down electrons, while often nearly identical; they are shifted in energy with respect to each other [642]. Unequal filling of the bands ensues, providing the source of the net magnetic moment for the ferromagnetic materials. Moreover, this can also cause the spin-up and spin-down carriers at the Fermi level to be unequal in concentration (and mobility), which can lead to a net spin polarization in a transport measurement. However, the sign and magnitude of the resulting polarization is dependent on the specifics. For example, a ferromagnetic metal may be used as a source of spin-polarized carriers for injection into a semiconductor (most likely in conjunction with an insulating barrier to tunnel as in the case of spin FETs discussed below) or a normal metal (as in the case magnetic sensor heads). Obviously, there is a need for 100% spin-polarized conducting materials. These are materials that have only one occupied spin band at the Fermi level. Materials such as Fe, Co, Ni, and their alloys are only partially polarized and have a polarization P of 40–50% [643] but are adequate to develop useful devices. P is defined in terms of the number of carriers n that have spin up (n") or spin down (n#), as Pspin ¼ ðn" n#Þ=ðn" þ n#Þ:
ð4:103Þ
Clearly, the impetus is in place for 100% polarization in the case of which the only states that are available to the carriers are those for which the spin is parallel to the spin direction of those states at the Fermi level; for reference, spin polarization of over 95% has been achieved, albeit at 4 K, in La2/3Sr1/3MnO3/SrTiO3/La2/3Sr1/3MnO3magnetic tunnel junctions (MTJs) [644]. In this particular device, spin polarization can be measured through the tunneling magnetoresistance (TMR) ratio observed when switching the magnetic configuration of one of the FM electrodes from parallel to antiparallel while leaving the other in the parallel configuration. Spin-polarized tunneling (SPT) was reported by Meservey et al. [645], and Tedrow and Meservey [646]. In the associated experimental investigations, determining the spin-polarized tunneling was done by tunneling from a ferromagnetic film at the Fermi level into a Zeeman spin split superconducting Al film that reflects the spin polarization of the tunneling electrons coming from the ferromagnet. These experiments showed that conduction of electrons in ferromagnetic metals are spin polarized and that the spin is conserved in the tunneling process. Julliere [647] formula for ferromagnetic/insulator/ferromagnetic materials tunneling describes the tunnel junction magnetic resistance (JMR). For informational value, the JMR is defined in two manners, pessimistic and optimistic. In the former, JMR ¼ [R(0) R(HS)]/R(0) where R(0) and R(HS) depict the tunnel junction resistance at H ¼ 0 and at saturation magnetic field H ¼ HS, respectively, the latter being much smaller, and represent the case when the magnetic
4.9 Intentional Doping
moments of the two ferromagnetic electrodes are aligned. The former, without the magnetic field, represents the case when the magnetic moments of the two electrodes are antialigned. In the optimistic case, JMR ¼ ½Rð0Þ RðHS Þ=RðH S Þ:
ð4:104Þ
In the so-called pessimistic case, the tunnel junction resistance is then given by JMR ¼
DR 2P1 P2 : ¼ ðR"# R"" Þ=R"# ¼ 1 þ P1 P2 R
ð4:105Þ
Here, R"# and R"" represent the resistances for antiparallel and parallel spin alignment states of the two ferromagnetic layers in the tunnel junction composed of two ferromagnetic materials (magnetic metal or ferromagnetic semiconductor separated by an insulator). It should be mentioned that R"# R"". In addition, P1 and P2 represent the spin polarization values for injector and analyzer ferromagnetic electrodes with the additional assumption that the magnetization in both materials is normal to the interface. They are defined as P¼
DðE F Þ" DðE F Þ# DðE F Þ" þ DðE F Þ#
;
ð4:106Þ
where D" ðE F Þ and D# ðE F Þ represent the density of states of spin-up and spin-down electrons, respectively, at the Fermi level in a given ferromagnetic material (consistent with P ¼ (n" n#)/(n" þ n#) given in Equation 4.103). For equal values of the polarization associated with the two ferromagnetic materials, JMR ¼ 2P2/(1 þ P2). For example, if the ferromagnet has only the spin-up electrons, the spin polarization would be 100%. In nonmagnetic materials there are no spin-unpaired electrons, and therefore D"(EF) and D#(EF) are equal to each other making P ¼ 0. For 100% polarization of ferromagnetic materials on both side of the tunnel junction, the JMR value would be 1. The implicit assumption here is that spin-polarized electrons can be injected into the insulator very efficiently and wherein they do not decohere. In a sense the spin is assumed conserved in tunneling, that is, the tunneling current flows in up- and down-spin channels as if in two parallel but separate wires. If we consider the effect of spin decoherence as well, the JMR expression can be modified as DR 2P1 P2 e d=ls ; ¼ R 1 þ 2P 1 P 2 e d=ls
ð4:107Þ
where d and ls represent the thickness of the medium where the spin-polarized electron transport is considered and the spin coherence length, respectively. In the so-called optimistic approach, the tunnel magnetoresistance (chosen arbitrarily instead of junction magnetic resistance) is defined as [648] TMR ¼ ðR"" R"# Þ=R"" ¼
2P 1 P 2 : 1 P1 P2
ð4:108Þ
For equal polarization of the two electrodes, ferromagnetic materials, TMR ¼ 2P2/(1 P2). The optimistic and pessimistic definitions of TMR differ from each
j1079
1080
j 4 Extended and Point Defects, Doping, and Magnetism other with the sign of the summation in the denominator. The TMR value in the optimistic case explodes for P1P2 ¼ 1. It should be mentioned that the Julliere formula characterizes the magnetoconductance solely in terms of the tunneling spin polarization. This model does not accurately describe the magnetoconductance of free electrons tunneling through a barrier. Another method forwarded by Slonczewski [649] (which also takes into account the orientation of the magnetization vector in relation to the interface) provides a good approximation to the exact expression for free electrons in the limit of thick barriers. In spite of the fact that Slonczewskis model provides a good approximation to the exact expression for free electrons in the limit of thick barriers, MacLaren et al. [650] found that the tunneling of band electrons shows features that are not described well by any free electron picture. This suggests that the details of the band structure of ferromagnetic material at the Fermi energy must be taken into consideration. Relative value of the two models has been debated by others as well [651]. Rigorous theories of tunneling in magnetic junctions have been developed [652], some predicting temperature dependence by taking the thermal smearing into consideration [653]. On the technological side, the seminal work of Meservey and Tedrow [645,646] provided the experimental evidence that the current in a ferromagnetic material and the current that tunneled across a barrier into another material has a net spin polarization. This confirmed the much earlier work of Sir N. Mott, predicting that current in a ferromagnet is spin polarized [654]. Julliere recognized the possibility of a converse effect, of that reported by Meservey and Tedrow [645,646], that is, if the metal on the other side of the tunnel junction (the counterelectrode) is also a ferromagnet, then the density of states available to receive tunnel current would also be different for up- and down-spin electrons. At that time this effect was very small and did not receive much attention which later changed as magnetic tunnel junctions became very popular with room-temperature magnetic tunnel junctions [655], giant magnetoresistance (GMR) taking advantage of the additive nature of the effect in the multiple layers used with GMR values potentially approaching 100% (the group of Fert) [656], the basis for which might go to the work of Gr€ unberg and his colleagues [657]. Getting back to magnetic tunnel junctions, if the magnetization of the materials is reversed, the spin direction of those states also reverses. Consequently, depending on the direction of magnetization of a material relative to the spin polarization of the current, a material can function as either a conductor or an insulator for electrons of a specific spin polarization. An analogy can be made with polarized light passing through an analyzer, which has been made already in conjunction with the proposal for a spin FET, as will be elaborated further in the text. The main difference being that in the optical case crossing the polarization axis at 90 blocks fully the transmission of the light, whereas for spin-polarized electrons, the magnetization must be rotated 180 for the same to occur. Magnetic-recording industry has been rapidly developing devices that rely on spin transport. When the two ferromagnetic metals straddle a normal metal, there can be transport if the polarizations (magnetization or the magnetic moments) of the injecting (polarizer) and collecting (analyzer) ferromagnetic materials are
4.9 Intentional Doping
aligned [633]. When they are antialigned, no transport is possible for the 100% spin polarization case. Technologically, both ferromagnetic and normal materials are deposited in thin films, and unlike the aforementioned scenario, the current is passed in the plane of the films, at the expense of making the spin-polarized transport more complicated. In this case, the effect of the spin exclusion in antialigned films is still observed which relies on interface scattering and channeling of the current into narrowed pathways. When the polarizations of the ferromagnetic films are aligned, both of these effects are removed, and the device would be in the low-resistance state. This two-layer system goes with the nomenclature of spin valve. It is constructed in such as way that the magnetic moment of one of the ferromagnetic layers is very difficult to reverse in an applied magnetic field, while the moment of the other layer is very easy to reverse. This easily reversed or soft layer then acts as the valve control by an external field. The device can be used to measure or monitor those external fields and can have numerous applications among which is the magnetic read head. These concepts can be applied to magnetic recording and nonvolatile memories, which are called magnetic random access memory (MRAM), which gained considerable interest [658,659]. It should be pointed out that because of the announcement of giant magnetoresistance, the magnetic hard disk read technology revolutionized. Ultimately, a new philosophy in computer memory in which the line between storage memory and active memory is blurred, may be spawned. With respect to issues dealing with magnetism, commonly used semiconductors are not magnetic. However, when they are doped appropriately with elements with spin-unpaired shell structures, they exhibit ferromagnetic properties. However, even then the ferromagnetism observed is not retained at room temperature. Remarkable progress has been made in GaAs [626], with transition temperatures in the vicinity of 170 K beyond which the magnetic ordering is destroyed. While the effort in GaAs is predictable and orderly, the same does not necessarily apply to GaN and other nitrides such as AlN. Although the dependability and physical basis are open to debate [660,661], theoretical [587,662,663] and a number of experimental investigations on GaN doped with Mn indicate Curie transition temperatures in excess at room temperature. Although Mn-doped GaN has been explored, as group II elements are potentially dopants that could produce p-type conductivity in GaN. The optical and other manifestations of Mn-doped GaN in this particular context are discussed in Section 4.9.6.1. Here, the focus is on the magnetic properties. The nascent state of GaN doped with transition metals and rare earth elements is such that the definitive observations of standard properties expected of dilute magnetic semiconductors are not yet available with the required clarity. The first group of semiconductor families that exhibited the clearest of all magnetic properties is the II–VI group, which unfortunately suffers from very low Curie temperatures. The semiconductor doped with Mn transition metal GaAs has a relatively higher Curie temperature, near 170 K, and is predictable by carrier exchange interaction induced ferromagnetism. In what follows, a succinct discussion of the theoretical basis for ferromagnetism in dilute magnetic semiconductors is given. When and if II–VI and GaAs examples are able to describe the mechanisms in a better way and/or provide reliable and well-behaved experimental results, they are included in the
j1081
1082
j 4 Extended and Point Defects, Doping, and Magnetism discussion. The results and conclusions of various theoretical approaches applied to GaN are naturally discussed. The discussion also extends to cover possible devices such as spin transistors, generically referred to as spintronics, and polarized lightemitting devices. 4.9.6.5 Theoretical Aspects of Dilute Magnetic Semiconductor The important characteristic of a ferromagnetic material is the spontaneous magnetization below the Curie (TC) temperature, also referred to as the critical temperature. As shown in Figure 4.139 in ferromagnetic materials [642], the d band is divided into spin-up and spin-down subbands, and the up and down states are displaced with respect to one another. The latter is displaced in energy so that the spin-up band is filled first, and the spin-down states contain the remaining, if any, electrons. The difference in the number of spin-up and spin-down electrons gives rise to the observed spontaneous magnetic moment. Above TC, the ferromagnetic material loses its permanent magnetism due to thermal agitations. To have practical applications in functional devices, it would be desirable, to put it mildly, to have a Curie temperature well above room temperature. Further for some device applications, it is also desirable to have the ferromagnetism be of carrier-induced origin, so that the magnetic properties of the DMS can be manipulated by external means such as through manipulation of the hole concentration. A better understanding of the underlying mechanisms will certainly provide the much-needed guidance for material design. To gain insight into the processes involved, a brief tutorial of the recently proposed mechanisms for ferromagnetism in DMS materials is presented in this section. A more detailed treatment of the theoretical results from recent literature about the mechanism of ferromagnetism in TM-doped ZnO and GaN then follows. Essentially, ferromagnetism results when the magnetic moments associated with magnetic ion impurities are polarized to be primarily in the same direction. As will be shown, the nature of the interaction of electrons of magnetic ions with those of conduction and valence bands of the host determine whether the resultant material is ferromagnetic or antiferromagnetic. Intuitively, everything else being equal, the closer the distance between the magnetic ions that occupy the cation sites and anions of the host the better the coupling expected to be. The strength of magnetic ordering in relation to thermal agitation is characterized by the Curie temperature below which ordering is not destroyed by agitations. Empirically, the Curie temperature is then expected to be high for semiconductors with smaller lattice parameters. Normalizing the lattice constant to the cubic form, the measured or the predicted Curie temperatures for various DMS semiconductors are shown in Figure 4.143. Getting back to a brief discussion of the nature of ferromagnetism in DMS, the mechanisms pertinent to magnetism are super direct exchange (antiferromagnetic), free carrier polarization, band polarization, indirect superexchange (could be ferromagnetic), double exchange (ferromagnetic and stronger when mediated by holes), and magnetic polarons, to cite a few. Let us now discuss the various theories developed to address ferromagnetism in magnetic ion doped semiconductors. First, a short glossary for each of the
4.9 Intentional Doping
500
GaN
Curie temperature (K)
400 ZnO
InN
300
AlP
1/a3
200
AlAs
Si 100
CdS
ZnS
GaP
CdSe Ge
CdTe InP
ZnTe
0 4.5
5.0
5.5
6.0
6.5
Lattice constant, a (Å) Figure 4.143 Measured or predicted critical temperature versus the lattice parameter for a group of semiconductors of general interest in terms of DMS. The solid line represents a fit with a3 dependence, where a is the lattice parameter. Courtesy of T. Dietl and J. Furdyna.
mechanisms is presented. What distinguishes DMS materials from more common semiconductors is that in DMS there are two systems, one comprising the semiconductor host and the other comprising the magnetic ion (either transition metal with partially filled d shells or rare earths with partially filled f shells) with their own somewhat preserved properties and their limited interactions. The semiconductor system can be characterized by delocalized band electrons that can be described by extended states. The magnetic ion, however, is characterized by localized 3d or 4f shell. Effective mass theory does well in describing the mobile carriers, that is, electrons in the conduction band and holes in the valence band. In this treatment the crystal Hamiltonian and pertinent perturbations can be described by the Luttinger basis. In addition, tight-binding calculations for wide bandgap semiconductors of interest here indicate that the G6 conduction band Luttinger functions mainly consist of cation s-orbitals whereas the G8 and G7 valence band associated functions to a first extent consist of anion p-orbitals. Electrical and optical properties of the semiconductor are determined primarily by the effective mass carriers. However, the localized magnetic moments associated with the magnetic ions and their interaction with the host semiconductor determine the magnetic properties. The interaction that is responsible for the desired magnetic behavior is sp–d in the case of transition metal magnetic ions and sp–f in the case of rare earth magnetic ions. Spin interactions are
j1083
1084
j 4 Extended and Point Defects, Doping, and Magnetism typically accounted for by the spin-dependent part of the Hamiltonian, which in metals give rise to the Kondo effect. It is for this reason, that the Hamiltonian, describing the spin-dependent coupling between the localized magnetic moments, associated with magnetic ions and band carriers is termed as the Kondo Hamiltonian. Such a Hamiltonian contains two independent mechanisms, namely, the kinetic exchange terms mediated by hybridization and direct Coulomb exchange. A succinct review of this topic is treated by Kacman [630]. 4.9.6.5.1 Carrier – Single Magnetic Ion Interaction Let us discuss the interaction between the host and one ionic magnetic impurity, involving conduction band electrons and the valence band electrons, with conduction band first. Direct Coulomb exchange is a first-order perturbation effect, and for s-like conduction bands, which is the case for the semiconductors under consideration here, and transition metals with their open d-shell, the direct Coulomb exchange interaction leads to ferromagnetic Kondo Hamiltonian. Coupling between the localized magnetic moment and the electron under the influence of strong spin–orbit interactions in the p-like valence band can be described by replacing the spin of the magnetic impurity in the Kondo Hamiltonian with the total eigenvalues of the angular momentum. For rare earths the spin–orbit coupling in the ion should be considered with the outcome that the ionic spin is replaced by the total angular momentum. To account for many magnetic ions in the crystal, a summation over magnetic ions is performed. The magnetic impurities can be described by mole fraction of the magnetic impurity, and in molecular field approximation, the spin operators are replaced with their averages. In the end the carrier-ion direct-exchange Hamiltonian for G6 conduction band electrons (describing s–d interaction in the case of transition metal ions) takes the form
Hex ¼ xN 0 a S== s== ; ð4:109Þ
where xN0 is the concentration of the magnetic ions with x representing the mole
fraction, a denotes the exchange constant for s-like electrons, S== represents the component of the thermodynamical average of magnetic ion spin along the magnetic field, and s//is the component of band electron spin along the magnetic field. Often times the magnetic field is applied along the growth or the z-direction, in the case of which the term hSz i is used to describe the average of magnetic ion spin. The exchange Hamiltonian is a measure of the extent of the exchange interaction and thus the extent of Zeeman splitting of the energy levels in thehost material. The extent of that splitting is generally taken proportional to xN 0 a S== , in the case of which N0a is treated as spin exchange integral for the conduction band. In II–VI DMS materials the parameter a is on the order of about 0.2 eV. Experiments performed mainly in conjunction with II–VI DMS indicated that this parameter for the valence band, going by the nomenclature b, has the opposite sign and is significantly larger than a by a factor of some 5–10 or even more, depending on the host material. As for the valence band, the picture is relatively complicated and there are several approaches that have been employed. For example, within the k-space perturbational approach, the p–d exchange interaction can be visualized as virtual transitions between p-electrons and ionic d-shell. A singly occupied orbital as in the case of
4.9 Intentional Doping
Mn with five electrons in 10 possible states and all electrons having the same spin can participate in virtual transitions that result in both creation and annihilation of p-like band electrons. The spin exchange interaction is governed in part by the Pauli principle that allows only the virtual transitions decreasing the total spin of the ion by removing the electron from d-orbital or by adding one with opposite spin. On the contrary, for an empty or doubly occupied orbital, the Pauli principle does not favor one spin polarity over the other, but the interaction remains spin dependent as the Hunds rule requires that the ion to increase its spin. To take this into consideration, the theory must account for the many electron intershell correlations. In considering a single hole near the top of the G8 valence band and the magnetic ion with five d orbitals occupied by N electrons, the Hamiltonian that is applicable would have ionic, crystal, and hybridization components. The exchange interaction Hamiltonian for p-electrons with a degree of complication in that how the crystal field splits the internal levels associated and which internal levels is strongly coupled to the host p-orbitals must be known. Consider the case of coupling between a G8 valence band p-like electron with three one electron t2g states of the magnetic ion. The hybridization-mediated kinetic exchange depends on the filling of only the t2g orbitals, not all the one-electron d-orbitals of the magnetic ion. The spin-dependent part of the exchange Hamiltonian for interaction between the G8 valence band p-like electrons and all t2g d-orbitals occupied by one electron can be described as Hex ¼
1 xN 0 b S== J == ; 3
ð4:110Þ
where b is the exchange constant, J// is the component of the total angular momentum of the G8 valence band p-like electron parallel to the magnetic field. As in the case of conduction band, the exchange spin Hamiltonian for the valence band electron determines the Zeeman splitting associated with the valence band states of the host. The term b is proportional to what is called the hybridization constant and inversely proportional to the sum of two terms that describe energies needed for transferring a d-shell electron from the magnetic ion into the band and transferring an electron from the band to the d-shell of the magnetic ion. For Mn2þ and Fe2þ, that is, for singly occupied d-orbitals, virtual transition of electrons from the ion to the valence band and also from the valence band to the ion is possible. In the case of Sc2þ or Ti2þ, only the virtual transitions from the valence band to the d-shell of the ion are possible. Band electrons with both spin up and spin down can be transferred. According to the Hunds rule that applies to both types of ions mentioned, the transition that increases the total spin of the ion leads to lower energy. In the case of Fe2þ and Co2þ, the sixth electron of Fe and Co occupy the eg orbitals lying at a lower energy than the t2g orbitals. Consequently, bMn < bFe < bCo. For magnetic ions with empty t2g orbitals, only the transfer of a band electron regardless of its spin into the d-shell of the magnetic ion is possible, and thus the Hamiltonian again has two components in the denominator, but with energies needed for transferring an electron from the band to the d-shell of the magnetic ion with one spin and another spin being subtracted from each other. Hunds rule, however, causes one of these energies to be smaller than the other, making the spin
j1085
1086
j 4 Extended and Point Defects, Doping, and Magnetism exchange interaction, which in this case goes by the nomenclature of g, positive, which means that in this treatment one would expect ferromagnetic interaction for Sc2þ and Ti2þ. When some of the t2g orbitals are singly occupied and some empty or doubly occupied, such as in Cr2þ that has two of its t2g orbitals singly occupied and one empty, strong exchange interactions (due to strong static tetragonal Jahn–Teller effect) result and rather complex splitting occurs. Naturally, exchange interaction has both b and g types of exchange interaction constants. In the opposite limit, when Jahn–Teller effect in none of the three crystallographic directions and the cubic symmetry is conserved, the mean exchange Hamiltonian has the simple Heisenberglike form again composed of b and g types of exchange interaction constants as multipliers with different weighing factors to S== . This represents the case of Cr2 þ in zinc chalcogenides, which has a ferromagnetic character. 4.9.6.5.2 Interaction Between Magnetic Ions Magnetic ion to magnetic ion interaction, or in the case of transition metal ions, d–d interactions could also take place, which is dependent on the density of the magnetic impurity in the host. The exchange coupling tensor in effect can be described with three distinct interactions such as Heisenberg, Dzyaloshinsky–Moriya (DM), and pseudodipolar (PD). Even this partitioning does not lead to any microscopic model of the interaction but only the form of the interaction can be discerned. When spin–orbit interaction is not taken into account, the DM and PD interactions vanish. Further, the DM component of the Hamiltonian vanishes in crystal with the center of inversion. As an example, the DM and PD terms for Mn in II–VI DMS are at least an order of magnitude smaller than the Heisenberg term. The nature of the dominant mechanism determines the sign and strength of the nearest neighbor and the next nearest neighbor (NNN) interaction constants. Let us now consider four main mechanisms leading to spin–spin interaction between magnetic ions. In the Ruderman–Kittel–Kayusa–Yoshida (RKKY) mechanism, the sp–d exchange leads to the polarization of free electrons. In the Blombergen–Rowland (BR) interactions, the sp–d exchange leads to band polarization. In both cases, the spin–spin interaction results from the induced polarization. In the other two remaining mechanisms, namely, the superexchange and double exchange mechanisms, the interaction, within the atomic picture, can be thought of in terms of virtual transitions between the ions (situated at the cation sites) and neighboring anions. The RKKY, BR, and superexchange interaction can be described by an effective Kondo-like Hamiltonian [630]. Superexchange Mechanism The superexchange mechanism is a process wherein the spins of two ions are correlated owing to the spin-dependent kinetic exchange interaction between each of the two ions and the p-like valence band. For superexchange, it is more proper to employ the band picture to describe the spin–spin interaction in semiconductors as opposed to the atomic picture. In II–VI DMS, the superexchange resulting from the sp–d hybridization is by far the dominant spin–spin interaction mechanism for the observed isotropic (Heisenberg) and anisotropic (DM) exchange constants.
4.9 Intentional Doping
Conduction band 2
1 D
C
A
k
B
Valence band
k'
Figure 4.144 A cartoon of the four virtual transitions depicting the superexchange ion–ion interactions. The arrows show the direction of electron transfer for the path ABCD. For the path CADB, all the directions should be reversed. Similar to that in Kacman [630].
Visualizing the virtual transition picture, one can construe the superexchange mechanism as being due to four virtual transitions, namely, from the p-like valence band state to the ions and back to the valence band characterized by arrows A, B, C, and D in Figure 4.144. This can take place in six sequences, namely, ABCD, ABDC, BACD, BADC, ACBD, and CADB. The directions shown are for the transitions ABCD. What is common to all of the sequences is that they all start with the transfer of a band electron k to one of the ions that produces a hole in the band but leads to different intermediate states in the perturbation Hamiltonian matrix depending on the sequence. The unperturbed valence band states, which must be summed up over the entire Brillouin zone, are typically described within the empirical tight-binding model [630]. Blombergen–Rowland Mechanism The Blombergen–Rowland mechanism is a process wherein the spins of two ions are correlated owing to the spin-dependent kinetic exchange interaction between each of the two ions and the p-like valence band and slike conduction band. The BR process differs from the superexchange mechanism only by the specifics of the intermediate states and that it also allows for virtual transitions to the empty conduction band, as illustrated in Figure 4.145. Naturally, this mechanism is less likely than the superexchange mechanism, particularly in large bandgap semiconductors, as the path is more complex. In this method, the intermediate states, jni and jn0 i, in the fourth-order perturbation matrix may correspond to one ion with N electrons and another with N þ 1 electrons and one hole in the p-like valence band. The hole may be considered to be created by a valence band electron being transferred to the ionic d-shell, as in the case of superexchange. In addition, a correspondence to one ion with N electrons and another with N 1 electrons, and one electron in the conduction band is also present. The electron in question in the conduction band is one that is transferred from the d-shell of the ion to the conduction band. In the intermediate jli states, there are N electrons of the two ions involved and a hole in the valence band, and an electron in the conduction band, or there are N þ 1 and N 1 electrons on the ions with neither holes in the valence band nor electrons in the conduction band. This can pave the way to a ferromagnetic d–d interaction even for the Mn2þ ions. However, this interaction
j1087
1088
j 4 Extended and Point Defects, Doping, and Magnetism k'
Conduction band
B
C
1
2 D
A k
Valence band
Figure 4.145 A cartoon of the virtual transitions involved in the Blombergen–Rowland interactions. Unlike the six sequences available in the superexchange mechanism, there are twelve different orders for the four A, B, C and D transitions. These processes commence with
the transfer of an electron either from the valence band to one of the ions (ABCD shown by the arrows, DCBA, ADBC, etc.) or from an ion to the conduction band (e.g., BACD, CBDA, etc.). Patterned after Ref. [630].
was shown to be an order of magnitude less effective than the superexchange [630]. One might expect that Blombergen–Rowland interactions are much enhanced in narrow-gap alloys. Double Exchange Interaction The double exchange interaction, proposed by Zener [664], is based on coupling magnetic ions in different charge states by virtual hopping of the extra electron from one ion to the other, specifically, an Mn2þ–Mn3þ (or d4–d5) pair of ions with one d electron hopping virtually from one ion to the other via the p-orbitals of neighboring anions. Naturally, then the Zener double exchange interaction cannot be the mechanism leading to ferromagnetic correlation between the distant Mn spins, because the magnetic electrons remain localized at the magnetic ion and do not contribute to the charge transport. In this vein Dietl in Matsukura et al. [665] suggested that the holes in the extended or weakly localized states mediate the longrange interactions between the localized spins. The same processes apply to Cr1þ–Cr2þ magnetic ion pairs as well, and also the superexchange interaction may also be in effect. The DMS fitting into this category is not all that uncommon, and the best studied examples are the zero-gap II–VI compounds (HgSe and HgS) with Fe ions, for which the 2þ/3þ donor level is degenerate with the conduction band [630]. The different charge states of a given magnetic ion have also been observed in III–V DMS. The case of GaN doped with Mn where evidence for both charge states is available is discussed in Section 4.9.6.1. The zero-gap II–VI compounds (HgSe and HgS) with Fe ions for which the 2þ/3þ donor level is degenerate with the conduction band have also been investigated [666]. Although it is believed that the Mn ions in GaAs and InAs are in the high spin 2þ charge state [667], the precise nature of the centers associated with Mn ions has not yet been definitively determined. The coexistence of Mn2þ and Mn3þ ions leading to double exchange has been suggested for Mn-doped ZnO [668] and the chalcopyrite semiconductor CdGeP2 [669]. The double exchange mechanism has been successfully used to explain the ferromagnetism observed in (In,Mn)As [664,670,671]. In a DMS material, if
4.9 Intentional Doping
neighboring TM magnetic moments are in the same direction, the TM–d band is widened by the hybridization between the up-spin states. Therefore, in the ferromagnetic configuration the band energy can be lowered by introducing carriers in the d band. In these cases, the 3d electron in the partially occupied 3d-orbitals of the TM is allowed to hop to the 3d-orbitals of the neighboring TM, if neighboring TM ions have parallel magnetic moments. As a result, the d-electron lowers its kinetic energy by hopping in the ferromagnetic state. This is the so-called double exchange mechanism. Theoretical investigations could not progress much because of the fact that both disorder and interactions are strong and must be taken into account nonperturbatively. This model is shown schematically in Figure 4.146. This kind of model can explain the physics of the dilute Mn limit, and can also be easily adopted to include the holes that are localized on ionized antisite defects, such as As in (Ga,Mn) As, rather than Mn acceptors. The double exchange mechanism proposed by Zener couples magnetic ions in different charge states by virtual hopping of the extra electron from one ion to the other. This mechanism was employed by Anderson and Hasegawa [672] to explain the magnetic properties of manganites of perovskite structure. They treated an Mn2þ–Mn3þ pair of ions with one d electron hopping virtually from one ion to the other via the p orbitals of neighboring anions. The relevance of this mechanism in mixed-valence DMS has been discussed in the literature [630]. The DMS materials with the aforementioned property are not that unusual. In addition to the well celebrated II–VI and III–V GaAs [595], GaN will soon be discussed. 4.9.6.5.3 Zener, Mean Field, RKKY, and Ab Initio Treatments Let us now consider these various mechanisms in more detail with pertinence to GaN as needed. In the Zener model, the direct interaction between d shells of the adjacent Mn atoms (superexchange) leads to an antiferromagnetic configuration of the d shell spins because the Mn–d shell is half-filled. On the contrary, the indirect coupling of spins through the conduction electrons tends to align the spins of the incomplete d shells in a ferromagnetic manner. It is only when this dominates over the direct superexchange coupling between adjacent d shells that ferromagnetism is possible. Accordingly, the mean field approach assumes that the ferromagnetism occurs
Mn
Mn As
As
N
N
Mn
Mn
Figure 4.146 Local magnetic moments (Mn2þ(d5)) with spin S ¼ 5/2 are antiferromagnetically coupled to itinerant carriers with spin S ¼ 1/2 in the model semiconductor GaAs:Mn (upper) and GaN:Mn (lower).
j1089
1090
j 4 Extended and Point Defects, Doping, and Magnetism through interactions between the local moments of the Mn atoms mediated by free holes in the material. The spin–spin coupling is also assumed to be a long-range interaction, allowing the use of a mean field approximation. The mean field model calculates the effective spin-density due to the Mn ion distribution. The direct Mn–Mn interactions are antiferromagnetic so that the Curie temperature TC, for a given material with a specific Mn concentration and hole density (derived from Mn acceptors and/or intentional shallow-level acceptor doping), is determined by a competition between the ferromagnetic and antiferromagnetic interactions. It should be mentioned that the Zener theory does not take into consideration the itinerant character of the magnetic electrons and the quantum (Friedel) oscillations of the electron spin polarization around the localized spins. Both of these were later established to be critical concepts for the theory of magnetic metals [587]. Because the mean distance between the carriers is greater than that between the spins, the effect of the Friedel oscillations averages to be zero in semiconductors. Early attempts to understand the magnetic behavior of DMS systems are based on models in which the local magnetic moments are assumed to interact with each other via RKKY-type interactions. The RKKY mechanism was originally introduced to explain the interactions between nuclear spins in metals via the conduction electrons and as such it is efficient only in cases where a high concentration of free carriers is present. Therefore, it is not really suitable to describe ion–ion interactions in DMS. Nevertheless, it formed the basis for the carrier-mediated interionic spin interactions in metals and highly degenerate semiconductors. The basic idea behind the RKKY interaction is based on the exchange coupling between the magnetic ion and the band electrons described by the s–d Kondo Hamiltonian, which is a first-order perturbation effect. The Zener model would become equivalent to RKKY if the presence of the quantum (Friedel) oscillations of the electron spin polarization around the localized spins are taken explicitly into account. It should be mentioned that s and d wave functions are orthogonal and would not lead to any interaction in perfect one electron system in cubic semiconductors. The conduction electron gas is magnetized in the vicinity of the magnetic ion, with the polarization decaying with distance from the magnetic atom in an oscillatory fashion. This oscillation causes an indirect exchange interaction (RKKY) between two magnetic ions on the nearest or next nearest magnetic neighbors. This coupling may result in a parallel (ferromagnetic) or an antiparallel (antiferromagnetic) setting of the moments dependent on the separation of the interacting atoms. The Zener model and RKKY interaction cannot explain well the free carrier based ferromagnetism in such systems. The RKKY interaction between Mn spins via delocalized carriers has been used to explain the ferromagnetism observed in PbSnMnTe [673]. However, if the carriers come from Mn–d states and are localized, which are far from being free electron like, the RKKY interaction may not be realistic. The mean field theory is based on double exchange hole interaction, meaning electronic in nature in p-type GaN and also ZnO. The theory dealing with ferromagnetism driven by the exchange interaction between carriers and localized magnetic ions (localized spin) was first proposed by Zener [664,674,675]. The mean field Zener model proposed by Dietl et al. [587] has been successful in explaining the transition
4.9 Intentional Doping
temperatures observed for p-(Ga,Mn)As, which is roughly 2000· percentile fraction of Mn ions in the matrix (K) and (Zn,Mn)Te. The mean field Zener theory is based on the original model of Zener [675] and the RKKY interaction. The mean field theory basically assumes that the ferromagnetism is a result of interactions between the local moments of the Mn atoms mediated by free holes in the material (double exchange interaction). The spin–spin coupling is also assumed to be a long-range process that allows the use of a mean field approximation. On the experimental side, magnetism even in n-type GaN has been reported. As compared to the RKKY interaction, the mean field Zener model takes into account the anisotropy of the carrier-mediated exchange interaction associated with the spin–orbit coupling in the host material. This process reveals the important effect of the spin–orbit coupling in the valence band in determining the magnitude of the TC and the direction of the easy axis in p-type ferromagnetic semiconductors. On the basis of this model, it was predicted that TM-doped p-type GaN and also ZnO are the most promising candidates for ferromagnetic DMS with high Curie temperature. However, these predictions are predicated on the incorporation of some 5% transition metal element and hole concentrations of above 1020 cm3. Notwithstanding these seemingly yet to be demonstrated high hole concentration (which may in fact never be attainable), this prediction stimulated a plethora of activity to achieve high Curie temperature ferromagnetism by using ZnO- and GaN-based DMSs. Another simple model [676] has been put forth to explain the possible mechanism with specific attention paid to (Ga,Mn)As. In this model, holes are assumed to hop only between Mn acceptor sites, where they interact with the Mn moments via phenomenological exchange interactions. In some other models [587] the ferromagnetic correlation mediated by holes originating from shallow acceptors in the ensemble of localized spins and a concentration of free holes ( 3.5 · 1023 cm3) have been assumed for (Ga,Mn)As. With due reverence and deference to the models mentioned above, it is increasingly becoming clear that a true picture can only be obtained by performing first principles calculations. For example, it is stated that the mean field theory overestimates the critical temperature substantially when the magnetic ion density is small. It has also become clear that in Mn-doped GaN magnetic ion concentration has a profound effect on the way the levels split and whether the ferromagnetic or antiferromagnetic state is stable. Having made the case for ab initio calculations, in the first principles approach the total energy and electronic structures are calculated by using the DFT. At temperature T ¼ 0, the ground state structure of the system corresponds to a minimum of the total energy. DFT [677,678] is a successful approach for the description of ground state properties of metals, semiconductors, and insulators. Implementation of DFT is based on approximations for the exchange correlation potential, which arises from the overlap of the electron wave functions due to chemical bond formation. One effective and common approximation is the local (spin) density approximation that locally allows substitution of the exchange correlation energy density of an inhomogeneous system by that of an electron gas evaluated at the local density, and generalized gradient approximation (GGA) that locally substitutes the exchange correlation energy density by that of an electron gas
j1091
1092
j 4 Extended and Point Defects, Doping, and Magnetism evaluated at the local density and its gradient and higher terms. The magnetic state of the DMSs can be investigated by calculating the electronic structure of a ferromagnetic DMS (all the magnetic moments of TMs are parallel to each other) and that of a spin glass like (magnetic moments of TM point randomly with respect to each other) DMS. The TE is calculated for both states as a function of transition metal density. Then, DE ¼ TE (spin glass state) TE (ferromagnetic state) is calculated to determine the stability of the ferromagnetic state, that is, when DE is positive, the ferromagnetic state is more stable than the spin glass state. In addition to the models mentioned above, an alternative model considers whether ferromagnetic ordering of the Mn moments could originate from carriers (holes) that are present in the material, but localized at the transition metal impurity [679,680]. Furthermore, ferromagnetism in DMS has been accounted for by percolation of bound magnetic polarons (BMPs) [679–686]. This in a sense relies on localized carrier, creating a spin polarization of the magnetic moments within the span of its wave function. The basic idea is schematically illustrated in Figure 4.147. The localized holes of the polarons act on the transition metal impurities
Figure 4.147 Representation of magnetic polarons. A donor electron in its hydrogenic orbit couples with its spin antiparallel to impurities with a 3d shell that is half-full or more than half-full. The figure is drawn for magnetic cation concentration x ¼ 0.1 and when the
orbital radius of the magnetic cation is sufficiently large. Cation sites are represented by small circles. Oxygen is not shown; the unoccupied oxygen sites are represented by squares (after Ref. [684]).
4.9 Intentional Doping
Figure 4.147 (Continued)
surrounding them, thus producing an effective magnetic field and aligning all spins. Transition to the insulating state takes place due to localization of the charge carriers (basically holes) at temperatures higher than the Curie temperature. As temperature decreases the interaction distance (boundary) grows. Below the Curie temperature, the neighboring magnetic polarons overlap and interact via magnetic impurities forming correlated clusters of polarons. A ferromagnetic transition is seen when the size of such clusters is equal to the size of sample. This model is inherently attractive for low carrier density systems such as many of the electronic oxides. The polaron model is applicable to both p- and n-type host materials [681]. Even though the direct exchange interaction of the localized holes is antiferromagnetic, the interaction
j1093
1094
j 4 Extended and Point Defects, Doping, and Magnetism between bound magnetic polarons may be ferromagnetic for sufficiently large concentrations of magnetic impurities. This enables ferromagnetic ordering of the Mn ions in an otherwise insulating or semi-insulating material. Even though the various theoretical approaches discussed above shed light on our understanding, albeit limited, of ferromagnetism in GaN, rigorous calculations must be used to get the true picture. In this vein, the first principles approach has also been used to elucidate magnetism in GaN-based DMS materials. As compared to the shallow acceptor Mn level in (Ga,Mn)As, the addition of Mn to GaN produces a deep impurity band within the GaN bandgap. Litvinov and Dugaev [687] questioned the RKKY interaction employed in the mean field theory and proposed that ferromagnetism in DMS systems is due to localized spins in the magnetic impurity acceptor level of the semiconductor crystal, and these localized spin excite band electrons due to p–d exchange interaction. Their model provided a detailed and quantitative predication of the dependence of the Curie temperature on the Mn concentration for various wurtzitic III–N alloys. Katayama-Yoshida et al. [688] studied GaN with 5% of various transition metals [689] and Ga1xMnxN with x ¼ 0.25 [690]. For a half-filled or less than half-filled d shell such as that in Mn, Cr, and V, the ferromagnetic state in GaN is stable. For a low concentration of Mn, ferromagnetism is favored, whereas for the high concentration the spin glass phase is stable. On the basis of local spin-density approximation functional calculations, Van Schilfgaarde and Myrasov [691] reported that for zinc blende GaN doped with 1–5% concentrations of Mn, Cr, and Fe, the exchange interactions are anomalous and behave quite differently from the picture assumed in simple models such as RKKY [587,692]. Those authors went on to argue that strong attraction between the magnetic elements tends to group them together in small nanoclusters of a few atoms. The magnetic coupling between doped Mn atoms in clusters and crystals of GaN has been predicted to be ferromagnetic by Das et al. [693] who used first principles calculations within GGA. Das et al. suggested that Mn atoms tend to cluster and bind more strongly to N atoms than to Ga atoms, which points out that the Mn concentration in GaN may be increased by using a porous substrate to offer substitutional surface sites. Their calculation also showed that the Fermi level passes right through the fattened impurity band (majority spin), thereby confirming that the impurity level acts as an effective mass acceptor. In a follow-up investigation, Wang et al. [694] examined two different cases where Mn atoms are bonded in bulk GaN as well as thin film forms with ð1 1 2 0Þ surface by allowing full structural relaxation within GGA. The study shows that in the (Ga,Mn)N system, the Mn–Mn separation plays a critical role in magnetic coupling of Mn ions. If the Mn atoms are incorporated into the GaN bulk, they couple ferromagnetically with or without structural relaxation. On the contrary, the coupling in unrelaxed thin film is ferromagnetic, which then converts into antiferromagnetic after relaxation. This may explain some controversial experimental observations for thick and thin film (Ga,Mn)N. Sato et al. [695] calculated the magnetic properties of (Ga,Mn)N from first principles using the Korringa–Kohn–Rostoker coherent potential approximation (KKR–CPR) method. It was found that the range of the exchange interaction in (Ga, Mn)N, being dominated by the double exchange mechanism, is very short ranged
4.9 Intentional Doping
due to the exponential decay of the impurity wave function in the gap. (Ga,Mn)N shows no high-temperature ferromagnetism for low Mn concentrations, so that the experimentally observed very high TC values in GaN should be attributed to small ferromagnetic MnN clusters and segregated MnN phases. The calculated TC for GaMnN is very low as compared to that obtained from the mean field theory that overestimates TC. Kronik et al. [696] reported the ab initio calculations for the electronic structure of (Ga,Mn)N with x ¼ 0.063. The introduction of Mn results in the formation of a 100% spin-polarized 1.5 eV wide impurity band, due to the hybridization of Mn 3d and N 2p orbitals (Figure 4.148). This result is qualitatively different from the case of GaAs due to the different Mn level position in GaAs the shallow acceptor, is 0.1 eV above the VBM in the isolated impurity limit, whereas in GaN deep-acceptor level, it is approximately 1.4 eV above VBM. So, in GaAs Mn hybridizes primarily with the valence band. However, in GaN its interaction with the valence band is therefore much smaller and the introduction of Mn barely polarizes the valence band. For an Mn composition of 6.3%, the impurity band does not hybridize to an extent sufficient for merging with the valence band. The impurity band renders the material half metallic and supports the effective-mass transport within it, which implies that (Ga, Mn)N is a highly suitable material for spin injectors. Kulatov et al. [697] studied the electronic structure and the properties of zinc blende (Ga,Mn)N with Mn concentration ranging from 1.56 to 12.5%. The calculations showed that the ferromagnetic state is lower in energy than those for the paramagnetic and antiferromagnetic states. The magnetic interaction of Mn atoms is short ranged. Their results also showed the important difference between GaAs and GaN in the energy position and localization of the Mn spin-majority states, as in Ref. [696]. 300
Majority spin 200
Density of states (au)
Mn 3d 100
0
−100
−200
Minority spin
N 2p EF
−300 −8
−6
−4
−2
0
2
Energy eV Figure 4.148 Partial density-of-states curves for wurtzite Mn0.063Ga0.937N. Thicker lines: Mn 3d. Thinner lines: N 2p [696]).
4
j1095
j 4 Extended and Point Defects, Doping, and Magnetism Codoping of O for N and Zn for Ga in (Ga,Mn)N shows that the atoms of O and Zn change the occupation of Mn bands and strongly affect both the magnetic moments and the conductivity. In addition, O codoping drastically enhances the ferromagnetic state and also causes a significant increase in the Curie temperature of (Ga,Mn)N. Further, Zn codoping causes strong hybridization of the eg states of Mn with VB and decreases the FM. Sanyal et al. [698] investigated the effect of varying Mn concentration on the electronic and magnetic properties in wurtzite GaMnN by using a first principles plan-wave method. They showed that the d states of Mn form an impurity band completely separated from the valence band states of the host GaN for dilute Mn concentration. Up to x ¼ 0.25, the Fermi level lies only in the spin-up density of states, as there is no state at the Fermi level for the spin-down channel, so that the system is half metallic with a high magnetic moment. For x ¼ 0.5, the system behaves like a ferromagnetic metal with a reduced magnetic moment due to the partially filled spin-down channel. The authors also showed that the DOS for a zinc blende structure is similar to that of a wurtzite structure, as shown in Figure 4.149.
8
(Wurtzite)
4
0
−4 DOS (states/eV)
1096
Mn−d N−p Total
–8
(Zinc blende)
4
0
−4
−8 −10
−5
0 E−E F (eV)
5
10
Figure 4.149 Spin resolved density of states of (Ga,Mn)N for (a) wurtzite and (b) zinc blende structures. Here, the Mn concentration is 6.25%. Courtesy of Sanyal et al. [698].
4.9 Intentional Doping
Mnd projected PDOS (/eV cell)
GaN: Mn t+
e-
t+
e+ t-
GaP: Mn
e-
t+ e+
t+
ee+
t-
t+
tt+
GaSb: Mn
e-
e+ t–3
t-
t+
t-
GaAs: Mn
–4
t-
–2
t-
t+ –1
–F Energy (eV)
1
2
3
Figure 4.150 In Mn d projected partial density of states for a single Mn in GaN, GaP, GaAs, and GaSb, where the symmetry (t2 and e) as well as the spin (þ and ) have been indicated. The shaded region represents the t2þ states (after Ref. [699]). (Please find a color version of this figure on the color tables.)
Mahadevan and Zunger [699] used first principles total-energy calculations to study the trends of Mn in GaN, GaP, GaAs, and GaSb. Figure 4.150 shows the calculated Mn d projected local density of states for neutral substitutional Mn (Mn3þ) in four GaX (X ¼ N, P, As, Sb) compounds. As can be seen, the antibonding t2 level, which is in the neutral state of the impurity, is occupied with two electrons (and therefore one hole) and strongly Mn localized. Moreover, the degree of Mn localization of the hole level decreases along the series GaN ! GaP ! GaAs ! GaSb. The acceptor level of Mn2þ is shown to be very deep in the (Ga,Mn)N, 1.4 eV, and it becomes progressively shallower as the anion x becomes heavier. They predicted a strong ferromagnetic stabilization in (Ga,Mn)N due to the p–d interaction that couples the tþ level of Mn ions to the p-like dangling bond states of the Ga vacancy, despite the fact that the hole orbital is a highly localized deep acceptor. This is in contrast to the model provided by Dietl [662] that assumes a host-like delocalized hole for all materials. In another study, Mahadevan and Zunger [700] explained the electronic structures of Cr and Mn in GaN as follows: introduction of a transition metal impurity in III–V semiconductor introduces a pair of levels with t2 symmetry – one localized primarily on the transition metal atom, referred to as crystal field resonance (CFR), and the other localized primarily on the host-anion atoms next to the transition metal impurity, referred to as the dangling bond hybrid (DBH). In addition, a set of nonbonding states with e symmetry, localized on the transition metal atom, are also introduced. Each of the levels is also spin split. Considering Mn in GaN shown in Figure 4.151, the 3d levels are well above the host cation dangling bonds. The dangling bond states are shown on the right-hand side, and the crystal
j1097
1098
j 4 Extended and Point Defects, Doping, and Magnetism 3d ion d n-1 t-(d )
e-(d )
t+(d)
Mn on Ga site t-CFR
Anion dangling bonds V Ga3-
CFR
e-
t+CFR t+(p)
e+(d) VBM
CFR
e+
DBH
t-
t-(p)
DBH
t+
Figure 4.151 A schematic energy-level diagram for the levels (central panel) generated from the interaction between the crystal field and exchange-split levels on the 3d transition metal ion (left panel) with the anion dangling bond levels (right panel), when the TM d levels are energetically shallower than the dangling bond levels (after Ref. [700]). (Please find a color version of this figure on the color tables.)
field and exchange split Mn d levels are shown on the left-hand side. The t2 (p) levels of the anion dangling bond hybridize with the t2 (d) levels of the transition metal. The levels generated after hybridization are shown in the central panel. The hybridization CFR in the t2-channel creates bonding, transition metal localized CFRs, tCFR þ and t , as DBH DHB well as the host-anion localized antibonding DBHs t þ and t , whereas the e CFR DHB channel creates the nonbonding eCFR þ and e states. The hole resides in the t þ level deep in the bandgap. The symmetry (e versus t2) and the character (DBH versus CFR), as well as the occupancy of the gap level, determine the magnetic ground state favored by the transition metal impurity. The ab initio band structure and total energy calculations [696–699] seem to agree that Mn 3d levels are located in the gap, and that the interaction between substitutional Mn ions is ferromagnetic, at least in the not so high Mn concentration range. On the basis of these understandings and by using a band structure approach and level repulsion model, Dalpian et al. [701,702] proposed a unified picture to account for the magnetic ordering in Mn-doped III–V and II–VI semiconductors. The model of the host p states (VBM) and Mn d levels and level repulsion caused by p–d exchange coupling and d–d coupling between them in ferromagnetic and antiferromagnetic configurations is shown in the schematic diagram of Figure 4.152. The Mn d levels
4.9 Intentional Doping
Antiferromagnetic
Ferromagnetic
(a)
(b )
Spin up (c)
(b ’)
(a’)
(a)
(b )
Spin down (c)
(b ’)
(a’)
t 2d
t 2d
t 2p
t 2d
t 2d
t 2p
t 2p
t 2p
t 2d
t 2d t 2d
t 2d t 2p
t 2p
t 2p
t 2p
or showing the spin up and spin down cases separately
A nti f erromagneti c Ferromagneti c
(a)
(b)
Spin down (c)
(a’ )
t 2d
t 2d
t 2p
t 2p
t 2d t 2d t 2p
t 2p
and Spin-up
Antiferromagnetic
Ferromagnetic
(a)
(b)
(c)
(b’ )
(a’ )
t 2d
t 2d
t 2p
t 2p
t 2d t 2d t 2p
t 2p
Figure 4.152 A schematic model showing the position of the p and d levels and level repulsion between them in FM and AFM configurations for spin-up and spin-down cases. Note that the
Mn d levels are above the VBM. Also note that in (b), (b0 ), and (c), the states have mixed pd characters. Courtesy of Dalpian and coworkers [701].
j1099
1100
j 4 Extended and Point Defects, Doping, and Magnetism are above the VBM, which is the case for (Ga,Mn)N. In the ferromagnetic (FM) configuration, the majority spin state of neighboring Mn atoms couple to each other. Specifically, the spin-up channel of one Mn atom couples to the spin-up channel of neighboring Mn atom. Considering the spin-up channel, step (a) to (b) and step (a0 ) to (b0 ) describe the p–d exchange coupling, which results in raising the t2d state upward by 2D1pd and lowering the t2p state by the same amount. Steps (b) or (b0 ) to (c) illustrate the d–d coupling between the two t2d states of neighboring Mn atoms. Consequently, the holes are placed at a high energy level and electrons are placed at an energy level that is lowered by 2D1dd. The net energy gain in the spin-up channel is therefore 2nhD1pd nhD1dd , where nh is the number of holes in the VBM. In the spin-down channel, the p–d coupling lowers the energy of occupied t2p spin-down state by 2D2pd . The net energy gain in the spin-down channel is 12D2pd for the six electrons in the t2p state. The total net energy gain in the FM configuration is 2nhD1pd nhD1dd –12D2pd . In the antiferromagnetic configuration, the majority spin state of one Mn atom couples only to the minority spin states of the other Mn atom with opposite moment. The situation can be analyzed similarly as in the FM configuration. The total net energy gain in the AFM 1;2 1 2 configuration is (6 nh)D1;2 dd 2nhDpd 12Dpd , where Ddd term represents the level repulsion caused by the coupling between the majority spin d state and the minority spin d state. The energy difference between the FM and AFM state is 1 therefore (6 mh)D1;2 dd nhDdd . This indicates that when the system has holes at the t2d level instead of VBM, the stabilization of the FM or AFM state is not directly related to the p–d exchange splitting, but to the d–d coupling term D1dd (double exchange) and D1;2 dd . In the case of (Zn,Mn)O, there is no hole, so the AFM state is more stable. The FM state is stable for (Ga,Mn)N for moderate Mn concentration (e.g., 6.25% Mn), but it gives way to the AMF state when Mn concentration increases due to the increase of the AFM stabilization energy D1;2 dd . In addition to Mn magnetic ion, theoretical investigations have also been reported for other transition metals such as Cr [690,703,704], Co [705], and vanadium (V) [690,706] in GaN. Das et al. [703] studied the electronic structure, energy bands, and magnetic properties of Cr-doped GaN from first principles with GGA approximation. The coupling between Cr atoms was found to be ferromagnetic in both crystal and small cluster forms of GaN. Kim et al. [704] reported on the electronic structure of Cr-doped GaN observed by hard X-ray photoemission spectroscopy and first principle calculations. They proposed that the ferromagnetic interaction between distinct Cr atoms may be mediated by the Cr 3d–N 2p–Ga 4s hybridization, and more data from magnetic and optical characterization would lend more credence. Switching gears to another transition element, 6.25% Co-doped GaN was calculated for its magnetic properties by Hong and Wu [705] using the fullpotential linearized augmented plan wave (LAPW) method. The authors found that Co atoms prefer to remain close to each other in GaN and form ferromagnetic ordering, with approximately 10 meV lower energy than the ferromagnetic phase. In a similar fashion, vanadium (V) was predicted [690] to be ferromagnetic when doped in GaN according to ab initio calculations within the local spin density approximation, which gives the electronic structure of the 3d metal doped GaN by
4.9 Intentional Doping
the KKR–CPR method. However, at least in one experiment, V-implanted GaN showed paramagnetic behavior up to 320 K [706]. In terms of the theory in regard to Ga1xGdxN, albeit in the cubic form, Dalpian and Wei [707] undertook ab initio band structure calculations with symmetry arguments showing that the magnetic properties of Ga1xGdxN are notably different from that of TM-doped GaN. The coupling between the Gd atoms in the alloy is antiferromagnetic but the 4f orbitals in the rare earth elements are shielded from the host material and more localized as compared to the transition metal atoms; thus, the direct coupling between the 4f ions is expected to be weak. As compared to the 3d transition metal elements, the 4f rare earth elements can have larger magnetic moments and can couple strongly with the host s electrons. Therefore, the ferromagnetic state can be stabilized by introducing shallow donors that are present in unintentionally doped GaN. The large magnetic moments observed can be explained by polarization of donor electrons. Therefore, the electron-mediated ferromagnetism is in effect here. This discussion has a good deal of relevance to the experimental observations discussed in Section 4.9.7. Because many exchange interactions have been invoked, Table 4.16 summarizes salient features of various interactions. It appears that DFT calculations augmented by relevant approximations for exchange correlation energies have the best chance of determining a reasonably true picture, and it is compelling to tabulate the attributes of various approaches within DFT, as done in Table 4.17. Experimental observations of ferromagnetism in both p-type and n-type GaN DMS materials have been amply reported in spite of the preponderance of the theoretical results, mainly centering around non-ab initio calculations, which require p-type GaN for a strong p–d interaction involving holes as compared to s–d interaction involving electrons. The discrepancy between these theories and experiments could perhaps be better understood by invoking the possibility that some observations might be clouded by magnetic contamination, imperfections of material quality specific to growth techniques, such as dislocations in the host materials [708], clusters [709,710] Table 4.16 A comparison of magnetic interactions.
Interaction
Definition
RKKY
Indirect exchange coupling of magnetic moments over relatively large distance via band electrons due to the Coulomb exchange. It becomes efficient when a high concentration of free carriers is present such as in metals for which it was developed. Direct coupling of magnetic ions through overlap of magnetic orbitals Spins of two magnetic ions are correlated due to the exchange interaction between each of the two ions and the valence p-band. Couples magnetic ions in different charge state by virtual hopping of the extra electron from one ion to the other through interaction with p-orbitals.
Direct superexchange Indirect superexchange Double exchange
j1101
Definition Density functional theory
Local density approximation
Generalized gradient approximation
Becke three-parameter Lee–Yang–Parr
LDA
GGA
B3LYP
Method
DFT
Approximation for the exchange correlation energy
Uses hybrid exchange energy functionals and gradientcorrected correlation functionals
Locally substitutes the exchange correlation energy density by that of an electron gas evaluated at the local density and its gradient and higher terms
For regions of a material where the charge density is slowly varying, the local charge density can be considered to be the density of an equivalent uniform homogeneous electron gas
Uses the charge density as the fundamental system variable, and describes the ground state properties by using certain functionals of the charge density
Assumptions
The most popular DFT method. Produces more accurate results
Yields improvement over LDA in the description of finite systems
Simple and produces moderately accurate results in most cases.
Simplifies the many body problem by using the electron charge density as fundamental variable rather than the wave function
Advantages
Table 4.17 A comparison of different methods in DFT calculations to approximate the exchange correlation energy.
Occasional large errors
Overcorrects the lattice constant for semiconductor systems compared to LDA
Underpredict atomic ground state energies and ionization energies, and over predicts binding energies
Applicable only for the ground state, needs approximation for exchange correlation energy functional
Limitations
1102
j 4 Extended and Point Defects, Doping, and Magnetism
4.9 Intentional Doping
precipitates [711,712], antisite defects [713], and nonsubstitutional impurity sites [714]. On the theory side, ab initio calculations, which are comprehensive by their nature in terms of being inclusive of all the mechanisms that may take place, are a must for an accurate determination of whether the material is FM and if so what the Curie temperature is. Already, the mean field theory appears to overestimate the Curie temperature substantially for lower magnetic ion densities. It should be mentioned, however, that the mean field theory is pretty accurate for predicting the Curie temperature in Mn-doped GaAs, at least for concentrations above 5%. Additionally, we must not lose sight of the fact that the results from different theoretical approaches do not agree well to say the least. Presumably, no single model is capable of explaining the properties of a wide class of dilute magnetic semiconductors including ZnO and GaN with their many variants. This being the case, multisourced possible explanations may have to be taken into consideration [715]. In particular, if the solubility limit of the magnetic dopant is exceeded, nonuniform ferromagnetic behavior or precipitates exhibiting ferromagnetism may form. Naturally, the magnetic properties of any such precipitates will depend on the growth conditions. In general, the contribution by precipitates to the overall magnetic properties of the bulk DMS cannot be excluded. A clear need exists for further research in this field if clarity is to be obtained. In this vein, extended X-ray absorption fine structure studies of DMS can shed light on the detailed microscopic structure of the lattice, which is more complicated than assumed in at least some of the theoretical approaches taken. Such careful structural investigations have been undertaken in standard bearer and traditional II–VI compounds [716], and to a lesser extent in GaN, which are discussed in Section 4.9.7. It is believed that with more progress in the synthesis and characterization techniques for DMS materials, improvements in various theories to understand the underlying mechanism will ensue. What is certain is that in due time, the science will correct its course. Before we delve into discussing the properties of DMS materials, particularly those discerned by magnetoelectrical and magnetotransport measurements, it is imperative that a working familiarity with these methods is developed. It is with this motivation that we venture into the world of the aforementioned measurements. 4.9.6.6 A Primer to Magnetotransport Measurements Due to the presence of the anomalous Hall effect, which is also known as the extraordinary or spin Hall effect, the Hall resistance RHall is empirically known to be a sum of the ordinary and anomalous Hall terms. The Hall resistivity in a magnetic semiconductor is given by
rxy ¼ r0xy þ rsxy ¼ Ro B þ Rs ðrxx ÞM ? or RHall
ð4:111Þ R0 RS M? ; ¼ Bþ t t
where the normal contribution r0xy represents the nonmagnetic (normal) component of the off-diagonal resistivity matrix element and is proportional to the external
j1103
1104
j 4 Extended and Point Defects, Doping, and Magnetism magnetic field B, and rsxy represents the magnetic (anomalous) component of the offdiagonal resistivity matrix element and is proportional to the macroscopic magnetization M?, which is normal to the sample surface. The terms R0 ¼ 1/pe and Rs (which is a function of rxx) represent the normal and anomalous Hall coefficients, and t stands for the thickness of the film. Further, Rs arises from the spin–orbit interaction, which induces anisotropy between scattering of spin-up and spin-down electrons. The Rs term is related to the sheet resistance through Rgsheet , where g is a constant and can assume values between 1 and 2, the end points corresponding to skew scattering and side jump mechanism, respectively. The constant g obviously determines the magnitude of the anomalous Hall effect term and scales with the extent of the spin–orbit coupling for the carriers at the Fermi level and the exchange energy that describes the ratio of carrier spin polarization to the magnetization normal to the surface. For a given normal component of magnetization, the anomalous Hall effect is much stronger for holes than for electrons in the tetrahedrally coordinated semiconductors. Moreover, anomalous Hall effect depends on the extent to which the electrons are spin polarized; the effect ceases to be proportional to the magnetization when carrier spin splitting becomes comparable to the Fermi energy. Because magnetotransport measurements are very sensitive to magnetization, they have become a cornerstone in characterizing dilute magnetic semiconductors, and further this method is well applicable to materials where the magnetic ordering or moment is very small. In ferromagnetic samples the anomalous component dominates and the normal component can be neglected. If additionally, the skew scattering is the dominant process, the anomalous Hall coefficient would be proportional to the sheet resistance of the sample [626]. Thus, M can be calculated from the above expression as being proportional to the ratio of the Hall coefficient to the sheet resistance. The Hall measurements should be carried out in the applied magnetic field limit where the magnetization is fully saturated (i.e., at low temperature and high magnetic field). Eventually, if and when GaN-based DMS materials advance to the point where reliable Hall measurements can be made, the anomalous Hall effect would be a reliable means for determining whether the material is ferromagnetic and further what the Curie temperature is. Representative data for the well-established system of GaMnAs are presented in Section 4.9.6.7.2. 4.9.6.6.1 Faraday Rotation, Kerr Effect, and Magnetic Circular Dichroism (MCD) While magnetization measurements with SQUID are imperative in the early stages of development, eventually the extent of magnetic ordering must be such that electrical and optical measurements are feasible and used for further research and transitioning to development. The SQUID measurements are sensitive to any magnetization, inclusive of magnetic impurities, precipitates, clusters, and mixed magnetic phases. The magnetoresistance (MR) measurements and optical measurements, particularly the combination of the two, are therefore the most reliable methods for unequivocally determining whether the material in question has ordered magnetization. The details of the former in nonmagnetic materials are amply discussed in Volume 2, Chapter 3, and a succinct discussion of it is given in Section 4.9.6.6 for magnetic semiconductors. One of the optical measurements relies
4.9 Intentional Doping
on Faraday rotation (FR) that can be simply viewed as the rotation in polarization plane of a linearly polarized light as it propagates through a magnetic medium [717]. Let us take a moment to review linearly and circularly polarized light. Let the in! ! plane coordinates of the material be in x and y and the normal to the surface be in ! ! ! the z direction. The functional dependence of linearly polarized light in x , y , and ! z directions on time can be expressed as C^ x e iwt ; C^y e iwt ;
and
!
C z e iwt ;
ð4:112Þ
where C is a constant (amplitude) and no phase shift is introduced. A right-hand (þ) and left-hand () polarized elliptical polarization would take the form 1 ^ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffi ða^ s x ib^yÞe iwt : a2 þ b2
ð4:113Þ
A special case of which is circular polarization when a ¼ b, in the case of which we can write 1 ^ ¼ pffiffiffi ð^ ð4:114Þ x i^yÞe iwt : s 2 Right-hand and left-hand circular polarizations arise from material exhibiting two different refractive indices, nþ and n, for positive and negative helicity. In semiconductors, application of a magnetic field in the z-direction, parallel to surface normal, causes Zeeman splitting of the electronic energy levels. The transitions associated with some are right-hand and others left-hand circularly polarized in addition to transitions with linear polarization. Faraday Rotation
qF ¼
Faraday rotation is given by the well-known expression as [718]
El ðn n þ Þ; 2hc
ð4:115Þ
where nþ and n are the refractive indices of the medium for right-hand and left-hand circularly polarized light, respectively. Different refractive indices for right-hand and left-hand circularly polarized light forms the basis for circular bifringence. E is the photon energy, l is the length traversed by photons, h is the reduced Plancks constant, and c is the speed of light in vacuum. In the case of H ¼ 0, the dispersion in the refractive index has the form n2 1 /
X 2 i;j E ij
f ij E2
;
ð4:116Þ
where Eij is the transition energy between levels i and j, and fij is the oscillator strength. Interband, intraband, internal d–d transitions associated with the magnetic ion in the host material, and those associated with impurities and defects in principle contribute to the circular bifringence and, through this, to Faraday effect, provided that there is Zeeman splitting. It should be noted that in a DMS, as soon will be discussed, the spin exchange interaction/magnetic field causes large Zeeman splitting of the electronic energy levels.
j1105
1106
j 4 Extended and Point Defects, Doping, and Magnetism For frequencies near Eij, the refractive index given by Equation 4.116 is dominated by the term corresponding to that particular transition. The difference between nþ and n ^þ can then be related to the energy difference DEij between transitions observed in s ^ polarizations. By Taylor expansion, it can be shown that the refractive index and s difference for left-hand and right-hand circularly polarized light can be expressed as ðn n þ Þ ¼
qn DE ij : qE
ð4:117Þ
Then, the Faraday rotation becomes qF
El qn DE ij : 2hc qE
ð4:118Þ
If we focus on a single oscillator of energy E0 in Equation 4.116, representing, for example, the transition at the G point in the Brillouin zone, the refractive index can be written as n2 ¼ n20 þ F 0 ðE 20 E 2 Þ 1 ;
ð4:119Þ
where F0 is a constant associated with the oscillator strength of the excitonic transition, and n represents all contributions to n in Equation 4.117. One can then obtain for Faraday rotation 1 F0l E2 DE 0 ; ð4:120Þ n 2 hc ðE 20 E 2 Þ2 where n represents the refractive index in the absence of magnetic field. For E close to E0 we can assume that n2 ffi F 0 ðE 20 E 2 Þ 1 , which leads to pffiffiffiffiffi El 1 y2 F0 l þ qF ¼ ; ð4:121Þ ðn n Þ ¼ DE 0 2hc E 0 ð1 y2 Þ3=2 2hc qF ¼
where y ¼ E/E0. Shallow impurity and intraband transitions are not taken into consideration in Equation 4.121. By defining pffiffiffiffiffi D y2 F 0 qDE 0 VðEÞ ¼ ; and D ¼ E 0 ð1 y2 Þ3=2 2 hc qH the Faraday rotation expression, expressing the derivative of the rotation angle with respect to the magnetic field, can be simplified to qqF El qn e qn lH ¼ VðEÞl or qF DE or qF or qF ¼ VðEÞHl; qH 2hc qE 2mc 2 ql ð4:122Þ where l is the layer thickness in this case, l is the wavelength, and V(E) is a constant named Verdet constant that depends on the material and the photon energy, E. In nonmagnetic materials, the Verdet constant is very small, for example, 3.25 · 104 ( cm1 Oe1) which prohibits the use of these kinds of materials for magneto-optical effects. However, this constant is large in ferromagnetic materials such as rare earth transition metals. In addition to magnetic properties, the medium should
4.9 Intentional Doping
also be transparent to the wavelength of the light source in use. How the Faraday rotation is additionally related to spin exchange interaction and magnetic circular dichroism is discussed following the discussion of Zeeman splitting. The experimental implementation of the above mentioned magneto-optical effect has quite a few applications, in addition to materials characterization, among which is an optical isolator. A linearly polarized light through a medium is rotated by an angle y. If and when the transmitted light is reflected by a mirror and fed back into the same medium, an additional polarization rotation in the amount of y occurs, leading to a total rotation of the light polarization by 2y, as shown schematically in Figure 4.153. In ferromagnetic materials, the rotation angle is typically plotted versus the magnetic field at temperatures below and above Curie temperature. Ideally any rotation would be minimal above Curie temperature and sizable below Curie temperature. Because the magnetization saturates with magnetic field, the rotation angle would also saturate. Essentially, the rotation angle would be proportional to magnetization. If the magnetization and length of the medium are such that the rotation angle y is equal to 45 , a total rotation of 90 is obtained the second pass through which forms the basis for an optical isolator, as depicted in Figure 4.153. For example, a semiconductor laser could be separated from the medium, such as an optical fiber, which is situated to receive the laser radiation by an optical isolator; there would not be any reflected light coming back into the semiconductor laser to interfere with its operation [719]. Ferromagnetic GaN or AlN for that matter would
l θ
H
I I 2θ
θ
H
I Figure 4.153 A schematic showing the Faraday rotation suffered by light in propagating from left to right (upper figure) and then again from right to left after reflection (lower figure), leading to a total rotation of 2y.
j1107
1108
j 4 Extended and Point Defects, Doping, and Magnetism pave the way for Faraday rotation based isolators operative in a wide range of wavelengths due to the large bandgaps of these semiconductors. Magneto-Optical Kerr Effect The other related phenomenon, called the magnetooptical Kerr effect or Kerr rotation (KR), represents rotation of the polarization plane of the reflected light from the surface of a magnetic material where the magnetization is perpendicular to the surface (polar Kerr rotation), as shown in Figure 4.154. Moreover, the direction of the aforementioned rotation is dependent on the direction of the magnetization M. This is useful in reading the information stored in the form of direction of magnetization and for studying the domain structure of magnetization. From a practical point of view, the value of DE can be inferred from Equation 4.122, using experimentally determined @n/@E. This value is about 35 meV, and is fairly independent of E between 1.25 and 1.6 eV for Ga1xMnxAs. Using DE ¼ N0x(b a)
l
θ
H
I
k
k
M
Figure 4.154 Faraday rotation in reflection is termed the magneto-optical Kerr effect or Kerr rotation wherein the polarization angle of the reflected light changes upon reflection from a magnetized medium such as ferromagnets.
4.9 Intentional Doping
hSzi (a more detailed discussion is given in conjunction with the discussion of magnetic circular dichroism and Equation 4.132), which is valid only at the band edge, N0x(b a)hSzi 1 ev is obtained. The terms N0a and N0b are the spin exchange integrals for the conduction band and valence band, respectively, and hSzi is the thermal average of the transition metal such as Mn spins in the direction of B, typically determined by an independent magnetization measurement. The positive value of N0b N0a reflects the positive sense of the Faraday rotation, which is opposite to that of II–VI DMSs such as undoped (Cd,Mn)Te discussed in conjunction with MCD below. This surprising result has been attributed to a large Burstein–Moss shift due to the high hole concentration specific to Ga1xMnxAs DMS [663,720]. In the context of these pages, it is not the applications that we seek, but rather it is the use of this phenomenon for material characterization that we desire. In experimental conditions concerning ferromagnetic materials, the degree of rotation versus the magnetic field in the ferromagnetic regime, that is, below the Curie temperature, above the critical temperature, that is, the paramagnetic regime, and preferably as a function of temperature can be conducted. From the collected data, one can discern information about the magnetic nature of the material. Magnetic Circular Dichroism (MCD) The other optical effect is the magnetic circular dichroism that is caused by the difference in absorption or transmission of right-hand (sþ) and left-hand (s) circularly polarized light. If the associated transmitted intensities and absorption for right-hand (positive superscript) and left-hand (negative superscript) polarized light are defined as Iþ and I, and aþand a, respectively, the magnitude of the MCD signal in degrees is defined as
Pcirc ¼ ða þ a Þ=ða þ a þ Þ ðT þ T Þ=ðT þ þ T Þ
ð4:123Þ
To gain an understanding of this effect in the particular semiconductor of interest, polarization of each transition involving heavy, light, and spin–orbit split-off bands must be known. In addition, the transition probabilities (oscillator strengths), at least the relative values, for each band are also needed. Continuing on, the knowledge of the selection rules as to the polarization for spin-up and spin-down electron in each of the bands are also required. These conditions reduce what needs to be known to a relationship between Pspin (given by Equation 4.103) and Pcirc (given by Equation 4.121). In this vein, a zinc blende semiconductor such as GaAs, and a generic wurtzitic semiconductor have been chosen for discussion. While the latter choice is obvious, the former is made because of its well-understood properties in the context of optical processes, the nature, and the degree to which those transitions are affected by strain and Zeeman splitting. These processes are intricately tied in with optical properties, the details of which are discussed in Volume 2, Chapter 5. Because the selection rules for GaAs are well established and MCD in ferromagnetic GaAs is, without comparison, more advanced than that in GaN, the MCD data for GaAs are discussed here for model. In many cases, depending on the excitation energy of the incident optical beam in relation to transitions involving the valence band states and conduction band, the right-hand and left-hand circularly polarized light cancel each
j1109
1110
j 4 Extended and Point Defects, Doping, and Magnetism other in the absence of magnetic field. However, with magnetic field (internal, external, or both) one or the other would be enhanced. To reiterate, MCD effect essentially relies on relative strengths of circular polarization and spin polarization. The former depends on the normalized relative intensities of right-hand and left-hand circularly polarized light defined by Equation 4.121. The latter depends on the spin polarization defined by Equation 4.103. Zeeman splitting of valence and conduction band states is caused by external magnetic field or the inherent magnetic ordering. When circularly polarized light is absorbed by a semiconductor, the selection rules are such that the electrons created would be spin polarized. Likewise, if spin selected transitions are dominant in a semiconductor (relative oscillator strengths of the transitions involved), the emitted light would be circularly polarized. By studying the light polarization, one can deduce information about the manner in which and degree to which the band states, particularly the valence band states, are split and can discern Zeeman splitting and therefore magnetization. At the very least the presence of the former can be used to decide if the material under investigation is ferromagnetic or not. In the case of Zeeman splitting by magnetic ordering, the degree to which that splitting occurs depends on the exchange interaction causing the ordered magnetic state. By performing the measurements versus temperature, one can determine the transition temperature. It should be mentioned that spin polarization is lost over time due to several processes such as interaction with magnetic impurities, spin–orbit interaction, to some extent electric field, and so on, which are critical for device applications of spin-polarized materials. In an effort to gain a degree of familiarity, let use first discuss GaAs, as a representative of many III–V zinc blende semiconductors, with no magnetic field and no strain. As will soon be described due to strong spin–orbit interaction in GaAs (DSO ¼ 340 meV), selected circularly polarized light (e.g., by injection of spin-polarized electron) is easily obtained even for bulk samples with degenerate heavy-hole and light-hole states, as the optical transitions involving the HH states are three times stronger than those involving the LH states (therefore, Pspin ¼ 0.5 Pcirc as detailed below). Let us now discuss the details of the process. Various energy levels associated with conduction and valence band states are subject to selection rules as to their participation in optical excitation and recombination. Assuming that spin-polarized carriers (electrons and holes) are generated, they would exist for a time before they recombine. If a fraction of the carriers initial orientation survives longer than the recombination time, if t < ts, where ts is the spin relaxation time, the luminescence (recombination radiation) will be partially polarized [632]. By measuring the circular polarization of the luminescence, it would then be possible to investigate the spin dynamics of the nonequilibrium carriers in semiconductors. Of device relevance, it would be possible to extract pertinent parameters such as the spin orientation, the recombination time, or the spin relaxation time of the carriers. The band structure of GaAs is depicted in Figure 4.155a. The 0 K bandgap of GaAs is Eg ¼ 1.52 eV, the light- and heavy-hole bands are degenerate at the zone center (which can be split by strain and magnetic field) while the spin split-off band is separated from the degenerate heavy- and light-hole bands by the spin–orbit split-off
j
4.9 Intentional Doping 1111 " Figure 4.155 Interband transitions in GaAs selected because of its well-known band structure and also its well-established and wellcharacterized properties in terms of magnetic ion doped diluted magnetic semiconductors: (arrows indicate emission but the concept is just as applicable to transitions from the valence band subband to the conduction band as in absorption). (a) Schematic band structure of GaAs near the G point, the center of the Brillouin zone. As for the terms, Eg is the bandgap and DSO the spin–orbit splitting; CB, conduction band; HH, valence heavy hole; LH, light hole; SO, spin–orbit split-off subbands; G6,7,8 are the corresponding symmetries at the k ¼ 0 point representing conduction, HH, LH, spin–orbit (SO) bands, or, more precisely, the irreducible representations of the tetrahedron group Td (see, e.g., Ref. [721]). The terms s1/2 and p3/2 and p1/2 represent the conduction band (s-like) and valence band (p-like) type of orbitals. (b) Selection rules for interband transitions between the Jz projection of the angular momentum along z-direction, sublevels for circularly polarized light sþ(righthand circular polarization or positive helicity that results from transitions between the Jz ¼ 1/2 conduction band states and Jz ¼ 3/ 2 heavy-hole states, and Jz ¼ þ1/2 conduction band and Jz ¼ 1/2 light-hole states), and s (left-hand circular polarization or negative helicity, which results from transitions between the Jz ¼ þ1/2 conduction band states and Jz ¼ þ3/2 heavy-hole states, and Jz ¼ 1/2 conduction band and Jz ¼ þ1/2 light-hole states). The numbers by each transition indicate the relative transition intensities, with respect to the light-hole subband to the conduction band (absorption or excitation of carriers to higher bands), or the conduction band to the light-hole subband transition (emission), which apply to both excitation and radiative recombination (depicted by the arrows). The circular polarization (s polarization) for light energies that would not excite the spin split-off band is ideally 50%, which becomes 0 if the spin–orbit split-off band is also excited. For completeness, the transitions between the Jz ¼ 1/2 conduction band states and Jz ¼ 1/2 light-hole states, and Jz ¼ þ1/2 conduction band states and Jz ¼ þ1/ 2 light-hole states, which are linearly polarized (p polarization), are also shown as depicted by
two-way arrows in the figure. The transition probability or the emission intensity normalized to the Jz ¼ 1/2 conduction state to the Jz ¼1/ 2 state transitions (indicated with 1) are also indicated in numbers for GaAs. The circular polarization resulting from the conduction band to the heavy-hole states are three times more intense than the circular polarization resulting from the conduction band states to the light-hole valence band states. The linearly polarized transitions are twice as intense as the circular polarization involving light-hole states. (c) Removal of the valence band heavy- and light-hole degeneracy by, for example, strain inducing either by lattice mismatch or by confinement in a quantum well, which increases the electron polarization to nearly 100%. Note that heavy- and light-hole states are no longer degenerate. Both the tensile (left) and compressive (right) in-plane biaxial strain cases are shown. The respective ratios of various transitions (oscillator strengths) have been assumed to be the same as in the relaxed case. Note that spin is indifferent to strain, which means that spin-up and spin-down states are moved in the same direction by strain, but not to magnetic field, as spin-up and spin-down states in a given band are split and moved in opposite directions as shown in (d). In part courtesy of W. Chen, Linko1ping University. (d) Removal of the valence band heavy- and lighthole degeneracy as well as splitting the spin-up and spin-down states by application of magnetic field. The total splitting is enhanced due to sp-d interaction in DMS materials in the form of xN0a<Sz> for the conduction band states, xN0b<Sz> for the HH and LH valence band states, and (1/3)xN0a<Sz> for the spin–orbit split-off band. Here, N0, x, a, b, <Sz> represent the number of cations per unit volume, mole fraction of magnetic ions, the product of Bohr magneton and the g factor for the respective bands, and average spin for each magnetic ion site, respectively. Note that magnetic field/magnetization causes Zeeman splitting, and direction of splitting either up or down in energy is spin dependent. If the semiconductor is ferromagnetic as is the case of GaMnAs, one can either couple polarized light to the symmetry/splitting allowed bands or cause polarized light emission by tuning the wavelength. (Please find a color version of this figure on the color tables.)
1112
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.155 (Continued )
4.9 Intentional Doping
Figure 4.155 (Continued )
j1113
1114
j 4 Extended and Point Defects, Doping, and Magnetism Table 4.18 Angular and spin part of the wave function at the G point in a zinc blende symmetry (GaAs) [632].
J; J z
Symmetry G6
j1=2; 1=2i j1=2; 1=2i
G7
j1=2; 1=2i j1=2; 1=2i j3=2; 3=2i
G8
j3=2; 1=2i j3=2; 1=2i j3=2; 3=2i
Wave function jS"i jS#i E 1=2 ð1=3Þ ½ðX þ iYÞ# Z" E 1=2 ð1=3Þ ½ðX iYÞ" þ Z# E 1=2 ð1=2Þ ðX þ iYÞ" E 1=2 ð1=6Þ ½ðX þ iYÞ# þ 2Z" E 1=2 ð1=6Þ ½ðX iYÞ" 2Z# E 1=2 ð1=2Þ ðX iYÞ#
parameter, DSO ¼ 0.34 eV. Following Ichenko and Pikus [721], Bloch states are denoted according to the total angular momentum J and its projection onto the can be calculated as in atomic positive z-axis Jz as J; J z . The allowed transitions J ¼ 3=2; J ¼ 3=2 (heavy hole) or physics between the ground states with z J ¼ 3=2; J ¼ 1=2 (light hole) and the excited levels with J ¼ 1=2; J ¼ z z 1=2i. Further, expressing the wave functions with the symmetry of s, px, py, and pz orbitals as jSi, jY i, and jZi, respectively, the band wave functions can be written, following Ref. [632] that relied on Ref. [722] with some typos removed, as listed in Table 4.18. To obtain the excitation (or recombination) probability, which is necessary for understanding dichroism, let us consider the photon propagation vector to be along the z-direction. Further, let us assume that sþ and s represent the righthand and left-hand circular polarization (helicity of the exciting light). The dipole operator corresponding to the sþand s optical transitions can be represented as m being proportional to ðX iYÞ / Y 1 1 , where Y l is the spherical harmonic, using Table 4.18. Doing so with the help of h1=2; 1=2jY 1 j3=2; 3=2i 2 1 ¼ 3; h1=2; 1=2jY 1 j3=2; 1=2i 2
ð4:124Þ
1
would allow determination of the relative intensity of the sþ transition between the heavy-hole (Jz ¼ 3/2) subband and conduction band (between the conduction band and the heavy-hole subband in the case of emission) and the light-hole (Jz ¼ 1/2) subband and the conduction band (between the conduction band and light-hole subband in the case of emission) [632]. The relative intensities are indicated in Figure 4.155b with number as 3 for the heavy hole related transition and 1 for the light hole related transition, and 2 for the conduction band to the spin–orbit split-off band transitions. The relative transition probabilities of the other transitions can be found in a similar fashion and are indicated with numbers in Figure 4.155b. If the outgoing
4.9 Intentional Doping
light in the case of emission (incoming light in the case of absorption) is in the zdirection, the helicities are reversed. It should be mentioned that the polarization vector also describes the electron dipole motion in that electric dipole radiates normal to its own motion and does not radiate in the parallel direction. This discussion segues into the selection rules for light absorption and emission. To begin with, only those dipoles in the plane can absorb or radiate. The probability of transitions involving electrons and heavy holes and light holes, even when they are degenerate as in relaxed GaN, are not the same because of different oscillator strengths, the heavyhole transition being three times larger than that that for light holes for GaAs. The transitions involving the spin–orbit split-off bands, when they are excited, are twice as strong as those involving light-hole states. Removal of the valence band heavy and light-hole degeneracy by, for example, strain inducing either by lattice mismatch or by confinement in a quantum well increases the electron polarization to nearly 100%. Note that strain breaks down the heavy- and light-hole state degeneracy regardless of the sign of the strain, that is, tensile of compressive. The sign of the strain naturally has an effect on which valence band states move up in energy and which move down as well as dispersion of the band away from the G point. The relative positions of CB, HH, LH, and SO bands are shown in Figure 4.155c. The respective ratios of various transitions (oscillator strengths) assumed to be the same as in the relaxed case and are indicated below the symbols depicting the nature of polarization. Note that spin is indifferent to strain, which means that spin-up and spin-down states are moved in the same direction by strain. The knowledge of how the bands split with magnetic field in a DMS material, which can be large, can be used in an MCD experiment, which is about to be discussed. In the case of absorption, the spin polarization of the excited electrons is dependent on the photon energy hw. Note that the holes are initially spin polarized also, but they lose spin polarization fast, on the order of momentum relaxation time, and are therefore omitted from this particular discussion. For hw between Eg and Eg þ DSO, only the light- and heavy-hole subbands contribute, leaving the spin–orbit split-off band out of the picture. Denoting the density of electrons polarized parallel (Jz ¼ 1/2) and antiparallel (Jz ¼ 1/2) to the direction of light propagation by nþ and n, one can define the spin polarization (given in Equation 4.103 but repeated here for convenience) as [632] Pn ¼ ðn þ n Þ=ðn þ þ n Þ:
ð4:125Þ
Applying the above equation to the zinc blende symmetry and without magnetic field (nonferromagnetic) and denoting the transitions involving light- and heavyhole transitions as |1/2, 1/2| representing nþ whose normalized intensity is 1 parallel to the direction of light propagation, and |1/2, 3/2| representing n whose normalized intensity is 3 antiparallel to the direction of light propagation, one obtains Pn ¼ ð1 3Þ=ð1 þ 3Þ ¼ 1=2
or 50%;
j1115
1116
j 4 Extended and Point Defects, Doping, and Magnetism for the spin polarization at the moment of photoexcitation (absorption measurements). The spin is oriented against the direction of light propagation, because there are more transitions from the heavy-hole than from the light-hole subbands. The circular polarization of the luminescence in a zinc blende structure is defined as [632] Pcirc ¼ ðI þ I Þ=ðI þ þ I Þ; þ
ð4:126Þ
þ
where I and I represent the radiation intensity for helicity s and s or, righthand and left-hand circular polarizations, respectively. In this particular case, the |Jz ¼ 1/2, 1/2| transition, 1/2 conduction band state to 1/2 valence band light-hole transition, represents right-hand circular polarization parallel to the direction of light with an electron polarization of nþ; the |Jz ¼ 1/2, 3/2| transition, 1/2 conduction band state to 3/2 valence band heavy-hole transition, represents left-hand circular polarization parallel to the direction of light with an electron polarization of 3nþ; the |Jz ¼ 1/2, 3/2| transition, 1/2 conduction band state to the 3/2 heavy-hole state, represents right-hand circular polarization and antiparallel to the light with an electron polarization of 3n; and |Jz ¼ 1/2, 1/2| transition, 1/2 conduction band state to 1/2 light-hole state, represents the left-hand circular polarization antiparallel to the light with an electron polarization of n. Utilizing these delineations and the fact that transitions involving the conduction band states to the heavyhole states are three times more intense that those from the conduction band to the light-hole states, one can calculate the electron polarization for right-hand circularly polarized light as Pcirc ¼
ðn þ þ 3n Þ ð3n þ þ n Þ Pn 1 ¼ ¼ : 2 ðn þ þ 3n Þ þ ð3n þ þ n Þ 4
ð4:127Þ
If the excitation involves transitions from the spin split-off band, that is, if w E g þ DSO , the electrons will not be spin polarized (Pn ¼ Pcirc ¼ 0), underlining h the vital role of spin–orbit coupling for spin orientation. On the contrary, removal of the valence band heavy- and light-hole degeneracy either by strain, confinement or by magnetic field, as shown in Figure 4.155, can substantially increase Pn, perhaps even up to the limit of complete spin polarization [723]. An increase in Pn and Pcirc in strained GaAs owing to a lattice mismatch with the buffer layer/substrate, or due to confinement such as in quantum well heterostructures, has been reported by Vasilev et al. [724] and Oskotskij et al. [725], with Pn values greater than 0.9. A magnetic field, however, causes Zeeman splitting with considerable impact on the band structure, as shown in Figure 4.155d, at the G point in a zinc blende symmetry. In DMS materials, the splitting can be quite large due to the strong sp–d interaction. In a sense, the effect of the magnetic field on the s and p band electrons is amplified by the magnetic moment of the magnetic ion through the sp–d exchange interaction. Let us now discuss the extent and nature of splitting for each of the four bands. For the spin-up and spin-down states in the conduction band: E "c ¼ E 0g ð1=2ÞxN 0 ahSz i and
E #c ¼ E 0g þ ð1=2ÞxN 0 ahSz i;
ð4:128Þ
4.9 Intentional Doping
where E 0g is the bandgap with zero magnetic field, and N0, x, a, hSzi represent the number of cations per unit volume, mole fraction of magnetic ions, a constant that represents the s–d exchange interaction integral, and thermal average spin for each magnetic ion site along the direction of the external magnetic field, respectively. The term a is positive and the term hSzi is normally negative for zinc blende symmetry that means that the spin-up band moves up in energy and spin-down moves down in the presence of magnetic field. The total splitting of conduction band spin-up and spin-down states is xN0ahSzi. Whether a given band moves up or down in energy is also dependent on the sign of a. The G8 degenerate heavy- and light-hole bands split into four bands, 2 HH (Jz ¼ 3/2) and 2 LH (Jz ¼ 1/2) bands with HH energies: DðE HH ; J z ¼ þ 3=2Þ ¼ ð1=2ÞxN 0 bhSz i and DðE HH ; J z ¼ 3=2Þ ¼ ð1=2ÞxN 0 bhSz i;
ð4:129Þ
where b is the p–d exchange interaction integral for the valence band. Recognizing that the spin exchange interaction term iSzh is normally negative and b is negative for zinc blende symmetry, the Jz ¼ þ3/2 would move down in energy and the Jz ¼ 3/2 state would move up in energy. Similarly, for the G8 LH bands split according to DðE LH ; J z ¼ þ 1=2Þ ¼ ð1=4ÞxN 0 bhSz i and DðE LH ; J z ¼ 1=2Þ ¼ ð1=4ÞxN 0 bhSz i:
ð4:130Þ
As in the case of the G8 HH states, recognizing that the spin exchange interaction term hSzi and b are both negative for the zinc blende symmetry, the Jz ¼ þ1/2 move down in energy and the Jz ¼ 1/2 state would move up in energy with magnetic field. Moreover, because the HH states split by an amount twice as that for the LH states, both the Jz ¼ 1/2 and Jz ¼ þ1/2 states are straddled by the Jz ¼ 3/2 and Jz ¼ þ3/2 HH states, as shown in Figure 4.155d. A similar treatment for the G7 SO band leads to DðE SO ; J z ¼ þ1=2Þ ¼ þð1=6ÞxN 0 bhSz i and DðE LH ; J z ¼ 1=2Þ ¼ ð1=6ÞxN 0 bhSz i:
ð4:131Þ
Again, noting that the spin exchange interaction term hSzi and b are both negative for the zinc blende symmetry, the Jz ¼ þ1/2 SO state moves up in energy and the Jz ¼ 1/2 SO state moves down in energy with magnetic field. Note that the direction of splitting is opposite to that of the G8 HH and LH states. In addition, the total splitting is only one third and two third of the HH and LH states, respectively. This too is shown in Figure 4.155d. In the Faraday geometry where both the applied magnetic field and the light propagation are along the crystal growth direction, both the right-hand (sþ) and the left-hand (s) circular polarizations are allowed because of symmetry considerations. As Figure 4.155d indicates, the optical transition energies for right and left circular polarization are different. For example, the energy for transition (s)is larger than that for (sþ) by xN0(b a)hSzi, which represents the Zeeman splitting due to
j1117
1118
j 4 Extended and Point Defects, Doping, and Magnetism sp–d spin exchange interaction. When the effect of the external magnetic field is also included, the energy difference between the abovementioned transitions becomes DE ¼ hSz iN 0 xða bÞ þ g mB H:
ð4:132Þ
The measured energy separation, for example, associated with the two transitions cited above can be used to determine the Zeeman splitting whose magnitude contains information about the magnetic nature of the DMS material under investigation. In addition to the transitions associated with the Brillouin zone center, the G point, the transitions at the L point, between G and L points but near the L point, and also at the L point, both along the [1 1 1] direction, contain quite a bit of information regarding band structure in general and band splitting in particular. The E 1 ¼ Lc6 ! Lv4;5 transition is a case in point. At the L point, the G8 band at the zone center splits giving rise to Lv6 and Lv4;5 levels, the separation of which is typically termed D1. Therefore, analogous to the E0 þ DSO (or E0 þ D0) transition at the zone center, we have E1 þ D1 transition ðE 1 þ D1 ¼ Lc6 ! Lv6 Þ at the L point that can be harnessed in transmission measurements in conjunction with magnetic circular dichroism investigations. The Zeeman splitting due only to sp–d spin exchange interaction for the E1 þ D1 transition is similar to that for the E0 þ DSO and takes into account splitting of the E1 and D1 values. The total splitting is then expressed as DðE 1 þ D1 Þ ¼ r hSz ixN 0 ða bÞ;
ð4:133Þ
where r is a numerical factor and takes values between 0.4 and 0.5 for zinc blende DMS materials depending on the orientation of the magnetic field. To reiterate, Zeeman splitting, inclusive of that induced by external magnetic field and also by the sp–d exchange interaction, can be used to determine the magnitude of the spin exchange interaction. Further, the p–d exchange interaction is stronger than the s–d interaction and |N0b| (associated with the former) is larger than N0a (associated with the latter). Cumulatively, the polarity of N0b can be determined from the polarity of Zeeman splitting of the most intense E0 transition. If Zeeman splitting of two different optical transitions are known along the value of the magnetic moment, both N0b and N0a can be determined independently. The magnitude of the Zeeman splitting can be tens of millielectron volts in typical but real DMS materials, allowing easy measurement. However, if the magnetization is low, and the optical transition spectrum is broadened by doping or by lack of high crystalline quality, the measurements become more difficult to interpret. Regardless, the measurements are made easier if one uses polarization modulation technique in conjunction with MCD. Now that we have gained sufficient knowledge as to the processes in the DMS material, let us now turn our attention to the optical aspects of MCD. As the above discussion laid the ground work, Zeeman splitting causes polarization-dependent optical anisotropy which is the source of MCD. The anisotropy of the refractive index causes the Faraday effect, which simply is the rotation of the polarization plane of the linearly polarized light as it traverses through a magnetic material. Both the Faraday rotation and MCD are related in that they are based on identical electronic structural parameters. Having discussed the Zeeman splitting of the electronic states and polarization associated with each of the transitions between various electronic states, let us relate
4.9 Intentional Doping
the Faraday rotation to magnetization. The transition between the G6 (1/2) and G8 (3/2), with energy Ea, is right circularly polarized whereas that between G6 (þ1/2) and G8 (þ3/2), with energy Eb, is the left circularly polarized (see Figure 4.155d). The energy difference between these two transitions in a DMS is given by ^ þ Þ E b ðs ^Þ ¼ DE 0 ¼ E a ðs
ba M; g eff mB
ð4:134Þ
where M is the magnetization. When Equation 4.134 is substituted into Equation 4.121, the Faraday rotation in a DMS material is expressed as pffiffiffiffiffi 1 y2 F0 b a M l: ð4:135Þ qF ¼ E 0 ð1 y2 Þ3=2 2hc gmB Let us use the Faraday rotation as a segue to extend our discussion to MCD. In the absence of external magnetic field, the transmitted light intensity through a sample is given by I ¼ I 0 e kðEÞl ;
ð4:136Þ
where k(E) represents the energy-dependent absorption coefficient of the material (aspired by the notation used for the imaginary part of the refractive index, which represents the loss term). l is the length of the medium which is the thickness of the sample in this case, and I0 is the intensity of the incident optical signal. The absorption coefficients for right-hand (kþ) and left-hand (k) circularly polarized light can expressed as DE DE k þ ðEÞ ¼ k E þ and k ðEÞ ¼ k E ; ð4:137Þ 2 2 where DE represents the total Zeeman splitting for spin-up and spin-down electrons at energy E given by Equation 4.132. The magnetic circular dichroism in terms of degrees is expressed as (defined in Equation 4.123 but repeated here for convenience) [726] qMCD ¼
180 ðk k þ Þ; also given in terms of the transmitted intensities as 4p
qMCD ¼
90 I þ I ; p Iþ þI ð4:138Þ
þ
where I and I represent the intensities of the right-hand and left-hand circularly polarized light. Utilizing Equation 4.137, yMCD can be rewritten as qMCD ¼
qMCD ¼
45 dkðEÞ DE ; or with spin exchange interaction elaborated p dE
45 dkðEÞ ; hSz iN 0 ða bÞ p dE
ð4:139Þ
j1119
1120
j 4 Extended and Point Defects, Doping, and Magnetism where dk(E)/dE describes the rate of change of the absorption coefficient with respect to energy or the energy derivative of the absorption coefficient. In other words, the MCD signal is proportional to the product of the Zeeman splitting and the energy derivative of the light absorption coefficient, dk/dE. Recall that |N0b| > N0a > 0 because the p–d spin exchange interaction is stronger than the s–d spin exchange interaction. It is deemed instructive to succinctly discuss the experimental aspects of measuring the MCD signal. A typical experimental setup used by Ando et al. [726] for measuring MCD consists of a monochromatic and linearly polarized light impinging on a photoelastic modulator (PEM). PEM then converts the linearly polarized light into alternating right-hand sþ and left-hand s light circularly polarized light that is made to pass through the sample under investigation. The transmitted signal in phase lock with PEM is then detected, as shown in Figure 4.156a. The measurement is repeated for varying light wavelengths, and the absorbance of light for both sþ and s polarization is plotted as the function of energy for a set of magnetic fields and at a set of sample temperatures. Shown in Figure 4.156b is the energy dependence of absorbance for zero (top) and nonzero (bottom) magnetic field. In the case without the magnetic field, the right-hand and left-hand circularly polarized light spectra are identical giving rise to zero difference between the two. However, with Zeeman splitting induced by a nonzero magnetic field and/or spin exchange interaction, the energy levels generating the right- and left-hand circularly polarized light shift, which give rise to a nonzero difference between the transmitted intensities of two circularly polarized lights. This forms the basis of the use of MCD for materials characterization. The photo elastic modulator (PEM) in turn consists of a rectangular bar of a suitable transparent (to the wavelength of light used) material such as fused silica attached to a piezoelectric transducer. The principle of operation of PEM is based on the photoelastic effect, in which mechanically stressed samples exhibit birefringence proportional to the strain generated. By applying an electric field to the PEZ material, which is in mechanical contact with PEM, the strain can be induced. Strain changes the refractive index of the material. The strength of the electric field controls the amount of strain induced, and the polarity of the field determines whether sþ or s polarization is attained, as depicted in Figure 4.157. The PEMs is usually operated in the ultrasound frequencies in the range of 20–100 KHz. Variation in the strain causes a variation in the change of refractive index and phase of light at the exit of PEM, as shown in Figure 4.158a. If the refractive index is made larger and smaller by changing the sign of the strain periodically, the phase shift of the emerging light can be controlled. For l/4 phase shift, righthand and left-hand circularly polarized light is obtained. If the phase shift is smaller, elliptically right-hand and left-hand polarized light would be produced, as depicted in Figure 4.158b. Because GaAs:Mn and transition metal doped II–VI systems are relatively well advanced and the electrical and optical measurements do indeed indicate what the Curie temperature is, a conscious decision was made to discuss the GaAs and II–VI systems to some degree through which it is hoped that the reader would be
4.9 Intentional Doping
Linear
σˆ − σˆ +
Detector
Monochromator Light
Sample Lock-In Amp.
Photoelastic modulator
~
Absorbance of light
Absorbance of light
(a)
LCP No difference between LCP and RCP
RCP
LCP
Difference between LCP and RCP
RCP
(b) Figure 4.156 (a) Experimental setup for measuring magnetic circular dichroism is composed of a monochromatic light being incident on a photoelastic modulator that is capable of generating alternating right-hand sþ and left-hand s light that is made to pass through the sample. The transmitted signal, in phase lock with PEM, is then detected. (b) Energy dependence of absorbance at zero (top) and nonzero (bottom) magnetic field. In the
case without the magnetic field, the right-hand and left-hand circularly polarized light spectra are identical giving rise to no difference. However, with Zeeman splitting induced by nonzero magnetic field and/or spin exchange interaction, the energy levels generating righthand and left-hand circularly polarized light shift, giving rise to a nonzero difference between the transmitted intensities of two circularly polarized lights.
appraised of the type of magnetoelectrical and magneto-optical data that could be expected of GaN when FM state is obtained. In addition, device results associated with GaAs:Mn are also provided for the same reasons. The particulars of DMS and its measurement in GaN are discussed in Section 4.9.7.3, and those of ZnO are discussed in Zinc Oxide: Fundamentals, and Materials and Devices by Morkoc, H. and Özg€ ur, Ü. Wiley 2008.
j1121
1122
j 4 Extended and Point Defects, Doping, and Magnetism Fused silica
PZE
Figure 4.157 Schematic representation of the principle of operation of photoelastic modulator. A piezoelectric crystal that is in mechanical contact with a birefringent transparent material such as fused silica is subjected to strain controlled by a piezoelectric transducer. The direction of the
(a)
field applied to the transducer determines the polarity of light passed through PEM, sþ or s polarization. The figure in the middle indicates compressive strain applied to fused silica while the bottom is for tensile strain. The latter produces right-hand circular polarization. Courtesy of Umit Ozgur.
Ex
Ey
E = xE ˆ y ˆ x+yE
λ /4 Ex = Eocos(ωt–kz) Ex = Eocos(ωt–kz– θ)
t or
z
with θ = π /2,3π/4 (b)
λ /4
λ /4
Figure 4.158 (a) Waveforms of two sinusoidal signals that are phase shifted by l/4, which results in right-hand circular polarization (lefthand circular polarization for l/4 (or 90 ) or left-hand l/4. (b) The polarization of the resultant light with respect to the phase shift
induced. Note that for l/4 peak phase shift, polarization oscillates between right-hand and left-hand circular and linear and elliptical polarization states are observed in between. Courtesy of Umit Ozgur.
4.9 Intentional Doping
4.9.6.7 II–VI and GaAs-Based Dilute Magnetic Semiconductors To provide a clear picture of dilute magnetic semiconductors, because the picture with GaN is not yet clear, magnetic properties of well-established semiconductors doped with transition metals will briefly be discussed. The II–VI- and III–V-based diluted magnetic semiconductors such as Cd1xMnxTe and Ga1xMnx As have attracted considerable attention for the same reason that the spin-dependent magnetic phenomena can be manipulated in these low-dimensional tailored magnetic thin films for various spin-based devices to unprecedented capabilities [727]. Generally, 3d transition metal ions (some species of magnetic ions, i.e., ions bearing a net magnetic moment) are substituted for the cations of the host semiconductors. As a consequence, the electronic structure of the substituted 3d transition metal impurities in semiconductors is influenced by two competing factors: strong 3d-host hybridization and strong Coulomb interactions between 3d–3d electrons. The latter is responsible for the multiplet structures observed in d–d optical absorption spectra. On the contrary, the hybridization between the transition metal 3d and the host valence band gives rise to the magnetic interaction between the localized 3d spins and the carrier in the host valence band [625]. Principally, the majority of DMS studied in an extensive way involved Mn2þ ion as a legitimate member of the transition metal family to be embedded in various AIIBVI as well as III–V hosts. Some of the valid reasons for this choice are as follows: (a) Mn2 þ has a relatively large magnetic moment (spin S ¼ 5/2 and angular momentum L ¼ 0) with a characteristic of a half-filled d-shell; (b) Mn2 þ can be incorporated in sizable amounts into the AIIBVI (up to 80% of Mn) host without affecting much the crystallographic quality of the DMS, whereas about 5 and 35% of Mn are tolerable for III–V-based and ZnO hosts; (c) Mn2 þ is electrically neutral in AIIBVI hosts, acting neither as an acceptor nor as a donor, while it acts as an acceptor in AIIBVI-based DMS. With the significant advance in the materials engineering of thin film and nanoscale heterostructures, the quantized energy levels in semiconductor nanostructures can be coupled either with local magnetic fields created by integrated submicron ferromagnetic structures or with the exchange fields of magnetic ions into the semiconductor lattice itself, hence providing a foundation for quantum spin devices [642] including single-electron spin transistors that rely on spin-dependent tunneling into a magnetic quantum dot (QD) and magnetic field effect transistors (M-FETs) that employ carrier injection into polarized spin transport channels [728]. In this section, an overview of the physics and some generic features of AIIBVIand III–V-based DMS heterostructures are provided. Because several interesting quantum structures derived from the AIIBVI-based DMS (particularly, substituting A-site by Mn) provide a valuable framework for understanding spin transport and dynamics in magnetically active quantum semiconductor structures, some generic features of II–VI-based DMS are described. This is followed by a brief summary of the main experimental properties of (III,Mn)V DMS, and some aspects of semiphenomenological theory have been put forth for the explanation of the ferromagnetism in these compound semiconductors.
j1123
1124
j 4 Extended and Point Defects, Doping, and Magnetism 4.9.6.7.1 II–VI-Based Dilute Magnetic Semiconductors Most of the research works performed so far on II–VI-based DMS materials have been devoted to Mn, which represents rather a simple magnetic system. The family of AII1xMnxBVI alloys [729], along with their crystal structure is presented schematically in Figure 4.159. The phase diagram shows that Cd1xMnxTe forms a ternary alloy of a stable zinc blende structure for x 0.77, whereas Zn1xMnxSe shows wurtzite structure for 0.3 < x < 0.55. The range of the AII1xMnxBVI solid solutions is considerably large. The easy substitution of Mn for the group II elements in both zinc blende and wurtzite structures occurs mainly because the 3d orbitals of Mn2þ are exactly half-filled. As a consequence, all five spins are parallel in this orbital by the Hunds rule, as it will cost CdTe
Cubic
HgTe Cu bi c
HgSe
Cu bi c
0.7 5
CdSe
bic Cu
0.77 6 0.8 MnTe
Telluride group
HgS 7 0.3
0.3 8
MnSe
x He
ZnTe
Selenide group
Sulfide group
bi c Cu
MnS 0.9 4 0.5
0.5
0.6 0.5
Hex
Hex
Cubic
Cubic 0.1
0.35
ZnS
ZnSe A schematic diagram of the AII 1x
VI
Figure 4.159 Mnx B telluride-, selenide-, and sulfide-based alloys and their crystal structures. The bold lines represent ranges of the molar fraction x for which homogeneous crystal phases form. Hex (hexagonal) and cubic structure represent wurtzite and zinc blende structure, respectively.
He x
CdS
4.9 Intentional Doping (a)
(Zn,Mn)Se
(Zn,Mn)Se
ZnSe ZnSe
ZnSe
(b)
(Zn,Cd)Se
ZnSe
(Zn,Cd)Se
ZnSe
MnSe monolayers
MnSe monolayers n–ZnSe
Figure 4.160 Some examples of various spinengineered DMS heterostructures: (a) left, a magnetic barrier QW structure in which the magnetic ions are located in the barrier; right, a magnetically coupled double QW; (b) left, a digital magnetic heterostructure in which the
magnetic ions are incorporated into the QW region in discrete, quasi-2D layers; right, a magnetic 2DEG in which modulation doping is employed to create a 2D Fermi sea that is in contact with magnetic ions.
a considerable amount of energy to add an electron with opposite spin to the atom, resulting in a complete 3d5 orbit. In view of this, Mn atom resembles a group II element, hence attracted many researchers to fabricate these materials using both thin film and crystal growth techniques. On the basis of the above phase diagram, several quantum structures, mainly strained layer superlattices and quantum wells (QWs) of (Cd,Mn)Te/CdTe and (Zn,Mn)Se/ZnSe on GaAs substrates, were grown by MBE [730]. With the advancement of enhanced MBE growth techniques, heterostructures containing a variety of band alignments, strain configuration and DMS alloys were fabricated. Some of the examples are illustrated in Figure 4.160. One may employ bandgap engineering techniques to create new DMS materials by manipulating the various physical parameters involved. Apart from this, the presence of local moments allows the spin engineering of new phenomena through the exploitation of two classes of exchange interactions. The first one is the d–d superexchange between d-electrons of the magnetic ions, and the second one is the sp–d exchange between the d-electrons and the band electrons or holes. This interaction is ferromagnetic for conduction band states and mainly antiferromagnetic for valence bands. This interaction determines the spin splitting of the band states in an external magnetic field. This gives rise to interesting magneto-optical and magneto-transport response in DMS samples. In dilute magnetic semiconductors (when the concentration of M2þ is small, i.e., x < 0.01), the M2þ spins can be regarded as isolated from one another, in the case
j1125
1126
j 4 Extended and Point Defects, Doping, and Magnetism of which the magnetization can be described by Brillouin function and we can write M ¼ xN 0 g Mn mB hSz i;
ð4:140Þ
where x is the Mn ion mole fraction, N0 is the number of cations per unit volume (xN0 therefore representing the number density of Mn ions), and hSzi is the thermal mean of the Mn spin along the z-direction (applied magnetic field direction). In the low-field and high-temperature limit, where gMnmBSH/(kBT) 1 (S ¼ 5/2 for Mn2þ), M is linear in H and static (DC) magnetic susceptibility w, defined by M ¼ wH, is of the Curie form c ¼ xC 0 =T with C0 ¼ N 0 ðg Mn mB Þ2 SðS þ 1Þ=3kB :
ð4:141Þ
For DMS with arbitrary magnetic ion concentration, x, the magnetization M cannot be expressed by the standard Brillouin function because of the M2þ–Mn2þ interactions. At low magnetic fields, M has been found to be linear in H as in the dilute case, which allows for the magnetic susceptibility, w, to be defined as provided below. At high temperatures and low fields, w follows a Curie–Weiss behavior described as c ¼ cd þ CðxÞ=½T qðxÞ;
ð4:142Þ
where y(x) is the Curie–Weiss temperature (the nomenclature Tc is also commonly used as is the case in this text), C(x) is the Curie constant, and wd is the diamagnetic susceptibility of the host. The inverse susceptibility w1 as a function of temperature, with the Mn concentration to be a parameter, for Cd1xMnxSe has been obtained for Mn mole fractions in the range of 5–45% [731], and is shown in Figure 4.161. The inverse susceptibility w1 shows a characteristic departure from the Curie–Weiss law at low temperatures (below about 40 K and shown in solid circles), deemed to be a characteristic of a cluster glass transition [732]. For an external magnetic field B applied along the z-direction, the magnetization Mz of a DMS alloy containing Mn2þ ions is empirically written as Mz ¼ xN 0 hSz i þ xN 0 Ssat B5=2 ð5mB B=kT eff Þ;
ð4:143Þ
where xN0 is the number density of Mn2þ ions and B5/2 (x) is the Brillouin function for S ¼ 5/2, Ssat is the saturation value for the spin of an individual Mn2þ ion (i.e., smaller than 5/2), and Teff ¼ T þ T0 is the rescaled temperature. Along with the distribution of magnetic ions on a DMS lattice, isolated spins, pairs of spins, and triplets are also distributed. Hence, magnetization is dominated by the paramagnetic response of isolated single spins, which are antiferromagnetically coupled. However, if we consider DMS heterostructure, the ferromagnetic s, p–d exchange interaction between conduction electrons and local moments results in an enhanced electronic spin splitting as described below (similar to Equation 4.132 discussed in conjunction with the GaAs case): DE ¼ gmB B xN 0 ½ f ðYÞa gðYÞbhSz i;
ð4:144Þ
4.9 Intentional Doping
Cd1–xMnxSe
120
χ–1 (mol Mn)
100
Mn: x = 0.45 0.35
80
0.25 60 0.05
40
20
0 0
100
200
300
Temperature(K) Figure 4.161 Inverse susceptibility, w1, of Cd1xMnxSe for Mn percentile contents of 5, 25, 35, and 45% as a function of temperature for zero-field-cooled (ZFC) sample at a field 30 G in the low temperature (below about 25 K shown with solid circles) region and 8.5 kg in the high
temperature (above 25 K shown with open circles) region. Note the linear Curie–Weiss behavior above about 40 K. The characteristic downturn of w1 below 40 K and cups observed in all but the data for 5% Mn is indicative of the spin glass transition [731].
where a and b are the s–d and p–d exchange integrals, respectively, and f(C) and g(C) are corresponding factors representing the wave-function overlaps of the conduction and valence band states with the local moments, and a is the s–d exchange integral. The first term represents the simple Zeeman splitting due to the application of a magnetic field. The second term is the splitting caused by the sp–d spin exchange interaction that is dominant in Equation 4.144, in part because the intrinsic g-factor for electrons in II–VI semiconductors is small and in part because of strong sp–d interaction. The exchange integral for the heavy-hole states is typically approximately five times larger than that for the conduction band and light-hole states. Because of this reason, in most optical experiments, which probe heavy-hole excitations, the spin splitting is dominated by that of the valence band states. The effect of magnetic field on the confined electronic states in DMS heterostructures is generally probed using magneto-optical spectroscopy, such as magneto-photoluminescence, magnetic circular dichroism, magneto-absorption, and Faraday/Kerr effect; the latter methods are discussed in Section 4.9.6.6.1. In (II,Mn) VI heterostructures, the spin injection from the DMS layer into the semiconductors happens at very low temperatures of about 5 K and at relatively high magnetic fields of about 1 T [733–735]. In (II,Mn) VI semiconductors, nearly 100% polarization of the electron spins in the conduction band of the semiconductor can be achieved via the giant Zeeman
j1127
1128
j 4 Extended and Point Defects, Doping, and Magnetism splitting of the conduction band states due to the s–d exchange interaction between the spins of the extended band states and the S ¼ 5/2 spins of the localized Mn2þ ions. The (II,Mn) VI layer in the DMS/semiconductor acts as a spin aligner into the adjacent layer. However, the main constraint for applications of this class of paramagnetic semiconductors is that the Zeeman splitting decreases rapidly with increasing temperature and, at room temperature, is orders of magnitude smaller to spin polarize. In addition to Mn, Cr magnetic ions have been exploited particularly in the ZnTe host. This system has been found to be one of the most promising material systems from ab initio calculations. Using gradient-corrected density functional theory and a supercell slab technique, Wang et al. [736] showed that the ferromagnetic phase of Zn1xCrxTe thin film is energetically the most preferable state irrespective of Cr concentration and/or configuration. Therefore, the ZnCrTe system can be unequivocally classified as a DMS system. Specifically, the hybridization between Cr 3d and Te 4p orbitals leads to a ferromagnetic coupling between the Cr spins in Zn1xCrxTe thin films. Furthermore, Wang et al. pointed out that in clusters of CrxN, the coupling between Cr spins has been found to be ferromagnetic, which results from the hybridization between Cr 3d and N 2p electrons. It should be pointed out that Cr2 dimer as well as bulk Cr are antiferromagnetic. Because the Zn1xCrxTe material system has been studied very comprehensively both with magnetization measurements and magneto-optical measurements with excellent agreement, MCD data for this material [737] are discussed first with a follow-up discussion of magneto-optics studies conducted in the MnTe system. Further motivation for the choice of material is provided by the fact that Zn1xCrxTe (with x 0.20) has been reported to show ferromagnetism at room temperature using MCD and magnetization measurements [738,739]. The investigated films by Saito et al. [737] were grown by molecular beam epitaxy on GaAs substrates. The MCD spectra were measured in reflection geometry with the magnetic field being normal to the film surface. Alternating circularly polarized light (right hand and left hand) at a frequency of 50 kHz produced by silica quartz stresser was applied, see Figures 4.156–4.158. The direction of the impinging and reflected light deviated from the film normal by 10 to allow for the measurements to be performed. Magnetization measurements were also performed using a SQUID magnetometer with the H field being normal to the film surface. The diamagnetic contribution from the substrate was accounted for as far as the magnetization data are concerned. Magnetotransport measurements were also performed for a comprehensive evaluation of the films [739]. TEM images and TEM diffraction images were also taken to ascertain that the films did not contain any phases other than the ZB ZnCrTe structure, which could be responsible for the observed magnetization. Displayed in Figure 4.162 is standard SQUID magnetization data for the Zn1xCrxTe (x ¼ 0.20) film investigated. As a standard procedure the diamagnetic contribution from the substrate has been accounted for. From the data obtained at 20 K, which show a large hysteresis, the magnetic moment at m0H ¼ 1 T is about 2.6 mB per Cr ion (requires the knowledge of total number of Cr atom incorporated in the lattice) which is consistent with films containing much lower Cr concentration
4.9 Intentional Doping
Zn1–xCrxTe
T = 20 K
x = 0.20
200 K
2
300 K
0
8
2
20 K 100 K
M
Magnetization M (µB Cr)–1
4
–2
300 K 0 0
–4
200 K
–1.0
–0.5
0.0
μ oH/M
0.5
1
1.0
Magnetic field μ0H ( T ) Figure 4.162 Magnetization curves of Zn1xCrxTe (x ¼ 0 : 20) film at various T. The inset shows the Arrott plots of the magnetization data. Courtesy of Saito et al. [738].
(0.035). Note that the ferromagnetic feature persists at temperatures all the way up to 300 K. For an accurate determination of the spontaneous magnetization and the Curie temperature, Arrott plots, plotting M2 versus m0H/M, were constructed, and intercept of the linear extrapolation from high magnetic fields to m0H/M ¼ 0 gave the quantity M2s, see the inset in Figure 4.162. The Arrott plot analysis is deemed the most reliable method for an accurate determination of Ms and Tc because this method relies on data taken at higher magnetic fields where the effect of magnetic anisotropy and the formation of a magnetic domain are negligible. An intercept at the positive values of M2s is indicative of spontaneous magnetization and at the Curie temperature linear extrapolation goes through the origin and a close inspection of the inset leads to the conclusion that the Curie temperature is Tc ¼ 300 10 K at which the spontaneous magnetization vanishes. To develop further confidence in the conclusions based on the magnetization data only, MCD measurements at m0H ¼ 1 Tand at various temperatures up to 300 K were carried out and analyzed. A few comments on the spectral features of the DMS signal are warranted. For example, a 100 nm thick ZnTe film showed weak MCD intensities at about the G (2.4 eV) and L (3.7 and 4.2 eV) critical points (CP) due to the diamagnetic Zeeman effect. On the contrary, an 80 nm thick Zn1xCrxTe (x ¼ 0.20) film on a thin ZnTe buffer exhibited pronounced MCD spectra at photon energies corresponding to the L-CPs of ZnTe. Moreover, the polarities of the MCD peaks of ZnTe were positive at both L-CPs. However, the Zn1xCrxTe film showed a positive MCD peak at 3.7 eV and a negative signal at 4.2 eV. These particular MCD spectra are indicative of the opposing polarities of the Zeeman splitting for the two L-CPs of Zn1xCrxTe, which is consistent with the general features of the Zeeman splitting caused by the sp–d exchange interaction. Therefore, the Zn1xCrxTe film under discussion was deemed to be a DMS. However, to be certain, the magnetic field
j1129
j 4 Extended and Point Defects, Doping, and Magnetism – MCD ( kdeg cm–1)
20
Zn1-xCrxTe
T = 20 K
x = 0.20 E = 2.2 eV
10
293 K
0 –10
–20 –1.5
–1.0
–0.5
0.0
0.5
1.0
1.5
Magnetic field μ0H ( T ) Figure 4.163 Magnetic field dependence of MCD intensity of Zn1xCrxTe (x ¼ 0.20) film at E ¼ 2 : 2 eV and T ¼ 20 and 293 K. Courtesy of Saito et al. [738].
20 K
300 120 K
200
MCD andM S (a.u.)
dependence of MCD intensity for Zn1xCrxTe (x ¼ 0.20) film on a thick ZnTe buffer near G-CP at T ¼ 20 and 293 K was analyzed to seek consistency with the magnetization data, as shown in Figure 4.163. Clearly, the MCD data agree with the magnetization data of Figure 4.162, confirming that the ferromagnetism persists at room temperature. To establish unequivocal confidence in the conclusion that the sample in question is ferromagnetic, the Arrott plots of the MCD data were plotted, as displayed in Figure 4.164. The inset shows the temperature dependence of MCD
MCD (kdeg2/cm2)
1130
0
MCD MS 0
180 K
100 200 T (K )
300
Zn1–xCrxTe 100
x = 0.20 E = 2.2eV
250 K
293 K
0
0
0.1
0.2
0.3
-μoH/MCD (T/kdeg. cm) Figure 4.164 Arrott plots of MCD intensity obtained in a Zn1xCrxTe (x ¼ 0.20) film at a photon energy of E ¼ 2.2 eV at temperatures up to room temperature. The inset shows the temperature dependence of MCD intensity (open circles) extrapolated to m0H ¼ 0 T, attained Arrott plots, together with the spontaneous polarization, MS (solid circles). Courtesy of Saito et al. [738].
4.9 Intentional Doping
intensity at m0H ¼ 0 T obtained by the Arrott plots, together with that of spontaneous magnetization, Ms. The MCD data plotted in the form of Arrott plots confirm the critical temperature TC to Tc ¼ 300 10 K, which is consistent with that obtained from Arrott plots of the magnetization data. The magnetization and magneto-optical data, particularly their conformity to Arrott plots, and the ensuing analysis clearly indicate that Zn1xCrxTe is ferromagnetic at room temperature. Saito et al. [739] also investigated electrical transport in their Zn1xCrxTe (x ¼ 0.20) samples that were grown on MBE-grown 200 nm thick ZnSe buffer layers that were in turn grown on high-resistivity (0 0 1) GaAs substrates. The film was of insulating nature, and showed a nonlinear behavior in log r versus T1 (here r is the resistivity measured), which means that the conduction is not due to thermally generated free carriers associated with either the conduction band or the valence band. The magnetic field dependence of the Hall resistivity rHall was too small to be measurable below 250 K, which sets an upper limit for the carrier mobility at 0.2 cm2 V1 s1 and points to hopping conductivity. Above room temperature positive sloped rHall versus H curves were noted with a p-type conductivity. With increasing temperature rHall increased, leading to deduction of 37 cm2 V1 s1 for the mobility at 350 K, indicative of band conduction. The magnetoresistance measurements taken at 20 and 150 K under magnetic fields, normal to the sample surface, ranging from 1 T to þ1 T indicated large negative slope with respect to the magnetic field regardless of its polarity. Similar results were also obtained for magnetic fields in the plane of the sample, ruling out any anisotropic magnetoresistance, which occurs in ordinary ferromagnetic metals, being responsible for the observed negative slope. Importantly, the magnetoresistance versus magnetic field curves showed hysteretic behavior at 20 K from which 0.6 T was deduced as the coercive field (HC) which is comparable to that obtained from magnetization measurements performed on the same sample. Defining a percentile change in MR (%) as Dr ¼ [r(H) rmax]/rmax, where rmax is the maximum resistivity in the r(H) versus H curve, MR (%) versus temperature was measured and plotted as a function of temperature up to 350 K under a 1 T magnetic field, as shown in Figure 4.165. At a temperature of 20 K, an MR ratio of 26% (Dr ¼ 4100 O cm) was deduced. With increasing temperature, the MR (%) ratio decreased and its sign changed from being negative to positive between 300 and 350 K. Note that negative MR occurs in the hopping regime of conductance that emanates most likely from localized carriers as well as magnetization. However, the positive MR measured at 350 K is proportional to H2 and also mobility squared, as can be seen in the inset, which is indicative of ordinary MR effect caused by Lorentz force with the associated increase in mobility. The abovementioned results are consistent with other material systems, such as In1xMnxAs [626] and Ga1xMnxAs [740], which also exhibit high resistivity at low temperatures. It has been suggested that the negative MR in these materials systems is due to wave function expansion associated with localized carrier as a result of the application of magnetic field. This picture may also be applicable to the Zn1xCrxTe system. It has been suggested that dissociation of bound magnetic polarons could be the genesis of carrier delocalization upon application of a magnetic
j1131
j 4 Extended and Point Defects, Doping, and Magnetism 5 0 Zn0.8Cr0.2Te H = 10 kOe
–5
TC Magnetoresistancechange(%)
Magnetoresistanceratio(%)
1132
–10 –15 –20 –25
2 350 K
1 0 –1
300 K
–2 –40 –20
0
20
40
H(kOe)
–30
0
100
200
300
400
Temperature, T (K) Figure 4.165 Temperature dependence of magnetoresistance ratio for the Zn0.8Cr0.2Te film investigated at H ¼ 1 T. The Curie temperature TC of the film is indicated by an arrow to be slightly above 300 K, which is consistent with the magnetization measurements and magneto-optical measurements discussed above. The inset shows the magnetoresistance of the film at T ¼ 300 and 350 K. Courtesy of Saito et al. [739].
field [626,740,741]. It is also feasible that giant spin splitting of the Fermi energy induced by strong sp–d interaction might be responsible for wave function expansion [742,743]. Saito et al. [739], however, reasoned that the bound magnetic polaron binding energy, based on the saturation magnetic field being very small, 0.6 kOe, is only 0.05 meV and they would dissociate at room temperature anyhow. Therefore, this model may not be the one responsible for negative MR. More work is warranted to determine the basis for the observations. MCD measurements have been undertaken in MnTe. Addition of nonmagnetic ion into this material and possibility of multiple layers for possible magnetic ordering are of interest. It should be stated that Mn ions in the ZB MnTe lattice interact with their nearest neighbors antiferromagnetically. If one considers the aforementioned antiferromagnetic together with the next nearest neighbor interaction, which is also antiferromagnetic, a variety of magnetic ordering might be achieved. To this end, the CdMnTe system, in the context of CdTe/CdMnTe quantum wells [726,744], has been investigated extensively [745]. MCD signal has been observed and analyzed in MnTe successfully, which results from the Zeeman split G6 conduction band and G8 valence band states. Using thin films to reduce absorption for allowing transmission measurements, MCD measurements [726] have been accomplished with near band edge light and used to deduce the MCD signal, as schematically shown in Figure 4.156. The MCD signal was observed at 15 and 50 K indicating negative and
4.9 Intentional Doping
positive going peaks at 3.38 and 3.6 eV associated with G8 G6 transitions for lefthand and right-hand circularly polarized light. Another positive going peak at 3.9 eV has been attributed to the Zeeman split G7 G6 transition at the L point of the Brillouin zone [726]. The temperature dependence of the negative G8 G6 transition has been used to determine the N_eel temperature to be about 60 K. A word about the N_eel temperature is warranted. In antiferromagnetic solids the total energy of the system at low temperatures is lowest for zero external magnetic field when the dipoles of opposing magnetic moments alternate. This arrangement is stable at low temperatures with susceptibility being small. However, as the temperature is increased, the dipole–dipole interaction begins to falter and the susceptibility increases until the spins become free. The temperature at which this occurs is called the N_eel temperature. Above the N_eel temperature, the spins are free to respond to the magnetic field and the susceptibility follows the modified Curie law of Equation 4.93. In the form of general remarks, the magneto-optical measurements seek to exploit the Zeeman splitting, which is to enhance due to s,p–d interaction over simply that caused by the magnetic field applied. While MCD measurements are sensitive and seek to look at the difference of right-hand and left-hand circularly polarized light, PL measurements can also be made in response to, for example, right-hand and lefthand circularly polarized light to determine the associated transition energies from which one can deduce the Zeeman splitting, the extent of which can be used to discern whether the material is ferromagnetic. In addition, combination of magnetooptical measurements along with carefully designed spin injections can give rise to effects of importance either in terms of basic physics or potential device applications such as optically controlled spintronics. This has been reported in the context of a (CdMn)Te/CdTe (CdMnTe as a spin aligner) DMS [746], and CdTe/(Cd,Mn)Te with magnetically coupled double QW structure wherein the coupling between quantum wells is Zeeman splitting tuned using spin-dependent barrier [747]. In the former case the right-hand and left-hand circularly polarized light emissions clearly show large Zeeman splitting. Moreover, several types of ZnSe/ZnCdSe double QWs coupled by ZnMnSe barrier have been fabricated and studied using different optical spectroscopic techniques [748–750] to understand the magnetically induced changes in the band structure, exciton spin scattering, and coherence in these band structures. Furthermore, CdSe QDs were fabricated by strained layer epitaxy [751,752] to study the coupling of zero-dimensional states with magnetic ions. This of course predicated on the assumption that Mn ions are introduced only in the QDs. The additional motivation may be that the spin coherence time may be lengthened by reduced dimensional structures. 4.9.6.7.2 III–V-Based DMS: (GaMn)As The observation of ferromagnetic transition in (III,Mn) V semiconductors, such as GaMnAs [753–755], at increasingly high temperatures (having a Curie temperature of about 110 K which with modulation doping and annealing has been increased to 170 K) [756,757] has attracted an inordinate amount of attention. Because GaMnAs is a well-developed DMS and is also a III–V semiconductor such as GaN, albeit with different crystalline symmetry when it comes to hexagonal phase of GaN, the GaAs system is covered for the reader to again
j1133
1134
j 4 Extended and Point Defects, Doping, and Magnetism gain an acquaintance with appropriate magnetism, magneto-transport, and magnetooptical data as we have done with II–VI-based DMS materials. The demonstration of reasonably long spin coherence times in GaAs-based semiconductor structures (100 ns for T2 time at 4.2 K in n-GaAs, which represents the spin coherence time [758], 50 ms for the T1 time at 20 mK in GaAs quantum dots, which represents the spin flip time [759], with the predicted value for the T1 time being 1 ms in GaAs quantum dots [760]) and rapid development of information storage technology using ferromagnetic metals have heightened the research activities in this field. The spin coherence time reported for GaN is a few nanoseconds at low temperatures and 35 ps at room temperature [761]. For reference electron spin relaxation effects in semiconductor Si were studied as early as in 1954 by Honig [762] and Honig and Kip [763], which was extended by Feher and Gere [764]. As semiconductors have many potential advantages over metals because of their easy manipulation to form an appropriate heterostructure with impurities, gates, and optical excitation, the actual synthesis of room-temperature ferromagnetic semiconductors is very attractive for device applications. The discovery of ferromagnetism at temperatures much higher than the room temperature in (Ga,Mn)N [587] has fueled hopes that these materials will indeed have profound technological impact. In this section, some important established properties and physics of (Ga Mn)As and (Ga,Mn)N are described. TM-doped GaN is discussed in the following section. Optical and electron paramagnetic resonance experiments [765,766] confirmed that dilute concentration of Mn exhibits S ¼ 5/2 local moments in GaAs (Mn2þ(d5) configuration). Hence, the Mn-induced states near the Fermi energy play a key role in the origin of ferromagnetism and in the magnetotransport properties of (III,Mn)V DMS. Photoemission studies show that Mn states can be associated with the host semiconductor valence bands. Ferromagnetism is not observed [767] for Mn concentrations smaller than about 0.01%, which predicts that ferromagnetism does not occur when all valence band holes are trapped on individual Mn ions or on other defects, such as commonly observed antisite defects due to As in GaAs semiconductors grown at low-temperature by MBE. Although it is generally accepted that Mn acts as an acceptor when substituted for a cation in II–V semiconductor lattice, most holes are trapped not at the Mn acceptors but at other defects for very dilute Mn concentrations. The largest ferromagnetic transition temperature in (Ga,Mn)As occurs for x 5%, having a record Tc 110 K [753]. With a combination of modulation doping and bulk doping and sometimes incorporating InGaAs, this temperature has been raised to about 170 K for annealed layers [756,757]. The ferromagnetic transition temperature drops for higher x values, the exact cause of which is not yet fully understood. However, competition between interstitial and substitutional incorporation of Mn, and possibly clustering in GaAs, which might take place for higher concentrations of Mn might turn out to be the cause, the interstitial one not contributing to ferromagnetism [714,768]. Ideally, the mean field theory predicts the transition temperature to be about 2000 K times the fractional substitutional concentration of Mn. These DMS materials display large anomalous Hall resistivities [767], demonstrating that itinerant valence band carriers are fully participating in
4.9 Intentional Doping
the magnetism due to strong spin–orbit coupling present at the top of the valence band. The valence band in GaAs is degenerate, which can be lifted by Zeeman splitting through the application of an external magnetic field and spontaneous polarization through sp–d exchange interaction if the semiconductor is made ferromagnetic by doping with transition elements such as Mn. If the p layer in a p–n light-emitting diode is made ferromagnetic, the spin-polarized holes and spin-unpolarized electrons that are injected under DC forward bias recombine with electrons in the n-layer only if spins align, giving rise to increased circular polarization. An exchange interaction between localized spins of the Mn atoms and the holes leads to a large g factor and a large Zeeman splitting [769]. All the holes scatter to the lower Zeeman levels, yielding a high percentage of spin-polarized holes. These spin-polarized carriers recombine in the quantum well region to produce circularly polarized photons due to the selection rules. An implementation of this has been done using an InGaAs quantum well sandwiched between a p-type Mn-doped ferromagnetic GaAs and n-GaAs [770]. In this case spin-polarized holes and spin-unpolarized electrons are injected into the InGaAs recombination region. If the spin polarization injected from the p-layer were 100%, the light emission would be as well. The details are discussed in the Section 4.9.12 dealing with spin-based devices. While the GaN-based DMS is in the process of being developed, it is deemed instructive to provide examples of Faraday rotation, along with the follow-up MCD, obtained in established DMS materials such as GaAs [595,771,772]. Faraday rotation below and above the Curie point in a Ga1xMnxAs (x ¼ 0.043) sample obtained at 6 T as a function of photon energy by Kuriowa et al. [773] is shown in Figure 4.166a. The Faraday rotation measured in the same sample as a function of magnetic field at 300 K and a photon energy of 1.49 eV and at about 10 K for a photon energy of 1.55 eV photon energy is displayed in Figure 4.166b. Note that the Faraday rotation versus magnetic field has the general features of the magnetization curves. As discussed in Section 4.9.6.6.1, MCD signal may be obtained by measuring the difference in the reflectance spectra from the sample for right and left circularly polarized light. Typically, alternating right-hand and left-hand circularly polarized light of varying wavelength (chosen to probe the critical points in the band structure, such as E0 and E0 þ DE0 associated with the G minimum and E1 and E1 þ DE1 associated near the L point in the Brillouin zone) is made impingent on the sample at 10 off the sample normal in the experiments of Ando et al. [772] and the spectra of the reflected light is collected whose spectra are recorded. Because semiconductor GaAs is under discussion here, the band structure of GaAs is shown in Figure 4.167 for us to have a clear picture of the critical points. In these experiments 1 degree of MCD rotation corresponds to 7% difference in reflectivity. Shown in Figure 4.168 are the MCD spectra of a reference GaAs sample and two GaxMn1xAs samples with x ¼ 0.05 and 0.074. For the reference sample with x ¼ 0, features in the reflectance spectra appearing at E0 (involving the G6 conduction band states and G8 valence band states at the Brillouin zone center, see Figure 4.167), E0 þ DE0 (involving the G6 conduction band states and G7 valence band states at the Brillouin zone center E1
j1135
j 4 Extended and Point Defects, Doping, and Magnetism 15
(a) (Ga,Mn)As x = 0.043 6T
θ (º)
10
~10 K
5 300 K 0 1.4
1.6
1.8
Energy (eV) 15 10
(b) (Ga,Mn)As x = 0.043
300 K 1.49 eV
5
θ (º)
1136
0 –5
–10 –6
Magnetization (from transport)
~10 K 1.55 eV –3
0 B (T )
3
Figure 4.166 (a) Faraday rotation for a 2 m thick film of Ga1xMnxAs with x ¼ 0.043 measured in a magnetic field of 6 T at 10 and 300 K as a function of the photon energy. These two temperatures represent below and above the Curie points. (b) Faraday rotation as a function of the magnetic field at 10 K (below the Curie point), 1.55 eV, and at 300 K (above the Curie
6 point) 1.49 eV. Note that below the Curie point, the rotation resembles the typical magnetization curves minus the hysteresis. Solid lines show the magnetization determined from magnetotransport measurements at the given temperatures. They have been scaled to match the Faraday rotation data (open symbols) for convenience. Courtesy of H. Ohno.
(involving the L6 conduction band states and L4,5 valence band states at the L point in the Brillouin zone), and E1 þ DE1 critical points (involving the L6 conduction band states and L6 valence band states at the L point in the Brillouin zone) with very small MCD features caused by the external magnetic field induced Zeeman splitting. However, when GaxMn1xAs epitaxial layers doped with Mn (with x ¼ 0.05, middle panel, and x ¼ 0.074, lower panel), the associated MCD signal is significantly increased due to enhanced Zeeman splitting because of strong sp–d spin exchange interaction. Note that the signal from the GaAs reference sample is magnified by a factor of 10. To be certain, MCD data delineate ferromagnetism; those data as a function magnetic field should be attained and analyzed, as shown in Figure 4.169. The notable MCD signal is deemed to originate from the E1 critical point of Ga1xMnxAs and as such it cannot be attributed to other phases such as the MCD spectrum of MnAs that is distinctly different [774]. The MCD spectrum from a sample with x ¼ 0.005 showed optical interference related oscillations at the E0
4.9 Intentional Doping
L 4,5
6
Γ8
Γ8
L6
Γ7
4
Λ6 2
L6 0
L 4,5
Energy (eV)
X6
Γ6
Γ6 Γ8
Γ8
v Λ 4,5
Γ7
L6
–2
Γ7
X7
c
Γ7 X7 X6
–4 –6
L6
X6
L6
X6
–8
–10
Γ6
–12 L
Λ
Γ
Γ6 Δ
X
U,K
Σ
Γ
Reduced wave vector Figure 4.167 Band structure of semiconductor GaAs. A special attention is called for the critical points at the G and L minima with the associated critical points E0 (involving the G6 conduction band states and G8 valence band states at the Brillouin zone center), E0 þ DE0 (involving the G6 conduction band states and
G7 valence band states at the Brillouin zone center), E1 (involving the L6 conduction band states and L4,5 valence band states at the L point in the Brillouin zone), and E1 þ DE1 (involving the L6 conduction band states and L6 valence band states at the L point in the Brillouin zone).
critical point (consistent with the sample thickness) and a sharp negative MCD signal. The negative character of the MCD signal near E0 can be used to infer the character of the p–d exchange involving the upper valence bands, particularly N0b. For reference, in Cd1xMnxTe the p–d spin exchange interaction N0b < 0 and thus the MCD peak is positive, which points to antiferromagnetism. On the contrary, in Zn1xCrxSe the p–d spin exchange interaction N0b > 0 and thus the MCD peak is negative, which points to ferromagnetism. It should be recalled that spin exchange interaction involving the conduction band N0a is much smaller than the spin exchange interaction involving the valence band N0b, and therefore, the sign of N0b would determine the polarity of the MCD signal. A point to note is that the MCD
j1137
1138
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.168 Magnetic circular dichroism spectra of a reference semi-insulating GaAs substrate with no magnetic ions (top), and two epitaxial Ga1xMnxAs films with x ¼ 0.05 (middle) and x ¼ 0.074 (bottom) at T ¼ 5 K and H ¼ 1 T. The spectrum associated with the GaAs reference sample is magnified by a factor of 10 because the signal is weaker than those of Mn-doped Ga1xMnxAs [772].
Figure 4.169 Magnetic field dependence of E1 MCD signal at a phonon energy of 2.83 eV for a Ga1xMnxAs (x ¼ 0.074) sample measured at T ¼ 5 K [744].
4.9 Intentional Doping
data at E1 and E1 þ DE1, associated with the L point, are merged into a broad band structure that is consistent with Mn incorporation into the GaAs lattice and can be attributed to strong sp–d hybridization. The effective g factor of 1.63 reported for GaAs is expected to also hold for Ga1xMnxAs. Additional support for negative N0b at the L point can be gained from the recognition that the sign of the effective g factor is determined from the sign of (N0a N0b/4). Because N0a is positive, the conclusion that N0b is negative is consistent with negative effective g. As in the case of magneto-optical data and previous discussion of the II–VI DMS system, the GaAs:Mn system, because of its well behaved magnetotransport measurements, that is, in terms of anomalous Hall effect data and their predictability, is used here as a model system for displaying transport data expected from DMS. The behavior of rxx and rxy in magnetic fields has been documented in quite a few GaAs: Mn samples, one of which is that reported by Edwards et al. [775] for a sample with Mn content of 0.05. In these measurements the term rxx shows an initial positive magnetoresistance as is typically observed in (Ga,Mn)As with in-plane anisotropy [776], followed by a negative magnetoresistance beyond B 0.5 T. The term rxy shows a rapid rise at low B on the way to magnetic saturation of the ferromagnetic film, followed by a more gradual rise that has contribution from both R0 and Ra. Fitting the measured rxy versus B curves using Equation 4.111 yields the value for p. The hole density corresponds to 90% of the Mn concentration at low Mn concentration and has a maximum value of 1.0 · 1027 m3 when Tc ¼ 125 K. The data configured along the lines of the second form of Equation 4.111 describing the normal and anomalous components of the Hall data have been obtained [777] as well and are shown in Figure 4.170a in the form of the Hall resistance RHall at various temperatures, plotted as a function of magnetic field for 200 nm thick Ga0.947Mn0.053As. The inset shows the temperature dependence of the sheet resistance Rsheet. A close resemblance of the data to the magnetization data (not shown) is indicative of the fact that the contribution of the ordinary Hall term is negligible in the field and temperature range investigated. Consequently, assuming skew scattering, the second term in Equation 4.111, ðRgsheet M ? Þ=t can be reformulated in the form of RHall cRsheetM, where c is a temperature-independent constant. Recognizing that RHall/Rsheet M, Arrotts plots [629] can be employed to determine the temperature dependence of the spontaneous magnetization M. The results of transport measurements are summarized in Figure 4.170b and c. As noted in the prologue to this section, magnetotransport measurements with some caveat that are noted at the end of this paragraph can reliably determine the critical temperature, and in this particular case the critical temperature determined from the magnetotransport measurements is in good agreement with that determined from magnetization measurements. As shown in Figure 4.170c, the value of the Curie temperature follows the empirical relation TC 2000x 10 K, where x represents the mole fraction of Mn in GaAs, which is good up to about 5%, beyond which the critical temperature falls below the value predicted by the empirical expression. Although not known with certainty, this decrease may be due to increased incorporation of Mn on interstitial sites [714]. Unlike the case of GaN, the mean field Brillouin function, see Ref. [628] for an elementary treatment of paramagnetism and the associated Brillouin
j1139
1140
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.170 (a, top) Temperature dependence of the Hall resistance RHall for a 200 nm thick Ga0.947Mn0.053As sample for which direct magnetization measurements have been performed but not shown. The inset shows the temperature dependence of the sheet resistance Rsheet. (b, center) Temperature dependence of the saturation magnetization [RHall/Rsheet]S obtained using Arrott plots (solid circles) and inverse susceptibility 1/wHall (open circles), both deduced from the transport data
shown in (a). Solid lines depict [RHall/Rsheet]S and (c, bottom) 1/wHall (bottom, c) calculated using the mean field Brillouin theory with S ¼ 5/ 2 for the Mn spin and the Curie–Weiss law, respectively. The dependence of magnetic transition temperature TC on Mn composition as determined from the transport data. Courtesy of Ohno and Matsukura [777]. (Please find a color version of this figure on the color tables.)
4.9 Intentional Doping
function treatment, is reasonably close in predicting the saturation magnetization [RHall/Rsheet]S and 1/wHall, as shown with a solid lines in Figure 4.170b and c. The temperature dependence of spontaneous magnetization determined from magnetotransport and magnetization measurements may deviate somewhat. This may have its genesis in the fact that the anomalous Hall effect scales with the spin polarization of the carrier liquid. As such this polarization is proportional to the magnetization only if the spin splitting is much smaller than the Fermi energy and the contribution of the carriers to the total magnetization is negligible. Moreover, the anomalous Hall effect senses only the regions where the carriers venture out. Therefore, it is just in these regions where the carrier-mediated ferromagnetic interaction is strong. Owing to the fact that the carrier distribution is highly nonuniform near the metal–insulator transition, magnetotransport, and direct magnetic measurements may diverge some [587]. 4.9.7 Experimental Results of TM-Doped GaN
Having established the baseline for magnetization data, magneto-optical data and also magnetotransport data for the well-established II–VI and GaAs systems doped with magnetic ions, let us now change gear and bring into focus the GaN system. The demonstration of ferromagnetism in (Ga,Mn)As and the predication that high TC ferromagnetic DMS could be realized by TM-doped GaN have spawned a great deal of experimental interest, in addition to theory, in examining and understanding the magnetic behavior of GaN-based DMSs, albeit still sketchy. The magnetic properties of reported TM-doped GaN are listed in Table 4.19. In this section, the reader is treated to a detailed overview of the progress in experimental studies in Mn, Cr, and other transition metal doped GaN, inclusive of rare earth (Gd) doped variety. 4.9.7.1 Magnetotransport Properties TM-Doped GaN Electrical (magnetotransport) and optical (magneto-optics) measurements contain signature that can shed light into the state of magnetic semiconductors. However, (Ga,Mn)As-based materials being ferromagnetic exhibit anomalous Hall effect, which means that accurate determination of the hole concentration p requires an analysis which takes this fact into consideration [779]. As for GaN, the electrical data are scarce as many of the transition metal doped samples are high resistivity, which precludes reliable Hall measurements from being made. Of the few results that have been reported, the data were of preliminary nature and may not be as reliable. Against this background, attempts have been made to investigate the magnetic behavior of Mn-doped GaN by both vibrating sample magnetometer (VSM) and magnetoresistance measurements, with specific attention paid to the extraordinary Hall effect (EHE) component [780,784]. In this particular investigation, the magnetization of the Mn-doped GaN films was measured by VSM both parallel and perpendicular to the plane of the film. It should be noted that the EHE measures only the perpendicular magnetization component and that the magnetization has an easy axis that is normal
j1141
0.07–0.14
0.03 0.1
0.07 0.03–0.05
0.03–0.05 0.01–0.03
(Ga,Mn)N
(Ga,Cr)N (Ga,Fe)N (Ga, Cr)N (Ga,Mn)N (Ga, Co,V)N
(Ga,Cr)N (Ga,Mn)N
p-GaN p-GaN
4H-SiC (0 0 0 1) Bulk p-GaN c-Sapphire 4H-SiC (0 0 0 1) p-GaN
p-GaN c-Sapphire c-Sapphire p-GaN Bulk
Ion implantation Ion implantation
Sodium flux growtha Implantation ECR MBE MBE Ion implantation
Solid state diffusion MBE MBE Ion implantation Ammonothermal and resublimation MBE
Fabrication method
350 350
700 710 350
750
710
580–720 865 350 1200–1250
Growth temperature ( C)
5 min at 700 C 5 min at 950 C in N2
5 min at 700 C
700–900 C
>300 >300
280 >350 >400
750
220–370 940 10–25 250 >425
250–800 C
700–1000 C
TC (K)
Postannealing
Ferromagnetic, from Mn cluster Ferromagnetic Ferromagnetic Ferromagnetic Spin glass V: paramagnetic, Co: spin glass Ferromagnetic Ferromagnetic
Ferromagnetic Ferromagnetic Ferromagnetic Ferromagnetic Para- and ferromagnetic
Notes
Sodium flux method growth is essentially the vapor liquid solid (VLS) growth scheme for GaN growth in which sodium flux represents the N flux for the N source.
0.01–0.02 0.03 0.07 0.01–0.03 0.02
(Ga,Mn)N (Ga,Mn)N (Ga,Mn)N (Ga,Mn)N (Ga,Mn)N
a
TM content
Material
Substrate (for thin film) or bulk
Table 4.19 List of recently reported magnetic properties of TM-doped GaN, as complied in Ref. [778].
1142
j 4 Extended and Point Defects, Doping, and Magnetism
4.9 Intentional Doping
to the film plane. Briefly, the measured films exhibited saturation magnetization, and the coercive fields varied between 0.01 and 0.05 T (100 and 500 Oe or Gauss). Hysteresis loops could be observed for both parallel and normal field configuration in the VSM measurements with resolution compromised to some extent due to 30 nm thickness, and the resolution of the magnetization curves was limited. The temperature behavior of the extraordinary Hall effect curves was found to undergo a transition from a linear behavior to a typical S shape one with increasing applied magnetic field, the transition point of which was used to deduce the Curie temperature. The ferromagnetic response was observed beyond room temperature with a maximum Curie temperature believed to be 363 K. The ferromagnetic transition temperature was also confirmed from the measurements of the sheet resistance in zero-field at varying temperatures. Additional reports of these kinds of reports preferably in films exhibiting strong magnetization and conductive behavior would go a long way for building confidence that GaN DMS is really at a point where Hall measurements can be employed reliably to determine the critical parameters. 4.9.7.2 Magnetic Properties of Mn-Doped GaN Many of the Mn-doped GaN layers have been prepared by molecular beam epitaxy at relatively low temperatures to increase Mn incorporation. High concentration of magnetic ions, up to a point, is imperative for ferromagnetic behavior. Some researchers [781] have reportedly attained single-phase (Ga,Mn)N layers by gas source MBE, which contained 7.0% Mn, showing anomalous Hall effect, negative magnetoresistance, and magnetic hysteresis at 10 K. The authors [781] believe that Mn-doped GaN is ferromagnetic with a TC between 10 and 25 K because the anomalous Hall term vanished at 25 K. Later on, they prepared a series of Mn-doped GaN samples [782] with Mn composition varying from 3 to 12%, for which the room temperature M–H magnetization loops were observed for the samples with Mn concentrations in the range 3–9%. The M–H measurements indicate a weak but discernable remanent magnetization and hysteresis was reported, indicative of ferromagnetic ordering for Mn concentration up to 9%. No evidence of secondphase formation was observed by either powder X-ray diffraction or high-resolution cross-sectional TEM in the n-type GaN samples with 9% or less Mn. It should be mentioned that any lack of observation by the aforementioned methods does not automatically rule out such phases or small clusters. Of those samples, the material with 3% Mn composition showed the highest degree of magnetic ordering per Mn atom and a TC of 320 K. However, an Mn concentration higher than 9% was found to enhance the antiferromagnetic (AFM) coupling, resulting in a lower magnetic moment per Mn. An extremely high TC of as high as 940 K for Mn-doped GaN has been reported by Sasaki et al. [783] in samples on (0 0 0 1) sapphire prepared by reactive MBE using ammonia for N source. This indeed is astonishingly high as the thermal agitations would be substantial at temperature approaching that value. The authors already acknowledge the coexistence of ferromagnetic and paramagnetic phases in their samples. Reed et al. [784] verified ferromagnetism in (Ga,Mn)N fabricated by solid-
j1143
1144
j 4 Extended and Point Defects, Doping, and Magnetism state diffusion of Mn into OMVPE-grown p-type GaN and subsequent annealing. They used vibrating sample magnetometer and extraordinary Hall effect measurements, in the temperature range of 310–400 K, which is discussed in the preceding section dealing with electrical properties. Dhar et al. [710,712] presented investigations of the magnetic properties of (Ga,Mn)N layers grown on 4H-SiC substrates by reactive MBE. Employing a variety of techniques such as X-ray diffraction and TEM, they confirmed that homogeneous (Ga,Mn)N alloys of high crystallographic quality could be synthesized for Mn concentrations up to 10–12%. Relying on the measurements of temperaturedependent DC magnetization, isothermal remanent magnetization, and frequencyand field-dependent AC susceptibility, the authors concluded that the insulating (Ga, Mn)N alloys behaved as a Heisenberg spin glass with a spin-freezing temperature around 4.5 K. This was attributed to the deep acceptor nature of Mn in GaN, which in turn resulted in the insulating character of the compound. In other reports by the same group, Dhar et al. [712] and Ploog et al. [785] also showed evidence for Mn-rich clusters being embedded in the (Ga,Mn)N alloy matrix to which the ferromagnetic behavior was ascribed. Cubic GaMnN in its metallic phase has been predicted [786] by Monte Carlo simulations to be ferromagnetic with TC in excess of room temperature as well. In this vein, Chitta et al. [787] reported that a p-type cubic GaMnN exhibited roomtemperature ferromagnetism, determined by the M–H hysteresis. The cubic GaN material was grown by MBE on GaAs (0 0 1) substrates, followed by Mn ion implantation corresponding to Mn composition of 0.7–2.8%. After annealing at 950 C, all samples showed sustained M–H hysteresis up to 300 K. Temperaturedependent magnetization measurements for both FC and ZFC conditions confirmed the similarity to hexagonal GaN in terms of magnetic properties. The origin of the observed ferromagnetism was not clearly specified. In most of the reported experimental results, M–H hysteresis data have been used as a measure of ferromagnetism and Curie temperature. However, it is well known that the magnetization curves alone are not sufficient for a conclusive statement on the matter and could in fact be misleading at times. Pearton et al. [660] analyzed three samples, that is, (Ga,Mn)N with 5% Mn (single phase), 50% Mn (contains multiphase of GaxMny), or 5% Mn (contains small amount of multiphase GaxMny), all of which showed magnetization hysteresis loops at room temperature. But in the temperature behavior under FC and ZFC magnetization, which is a more instructive but not so airtight examination of ferromagnetism, three samples showed different behaviors as indicated in Figure 4.171. Only the single-phase (Ga,Mn)N sample with 5% Mn, shown in Figure 4.171 (top), shows ferromagnetism persisting above 300 K as " Figure 4.171 Temperature dependence of field cooled (FC) (top curve in each case) and zero field cooled (ZFC) (bottom curve in each case) magnetic moment for (Ga,Mn)N with 5 at.% Mn (optimized growth) at the top (a), 50 at.% Mn at the center (b), or 5 at.% Mn (termed as unoptimized growth) at bottom (c). Courtesy of Pearton et al. [660].
4.9 Intentional Doping
Magnetic moment (emu)
-4.0 x 10 -6
5% Mn Single Phase
(a)
-8.0 x 10 -6
-1.2 x 10 -5
FC
-1.6 x 10 -5
ZFC 0
100
200
300
Temperature (K) 6.0 x 10 -5
Magnetic moment (emu)
50% Mn multi phase
(b)
FC
4.0 x 10 -5
ZFC 2.0 x 10 -5
0.0
0
100
200
300
Temperature (K) 5.0 x 10 -5
Magnetic moment (emu)
5% Mn multi phase
(c)
4.5 x 10 -5
FC
ZFC 4.0 x 10 -5
3.5 x 10
-5
0
100
200
Temperature (K)
300
j1145
1146
j 4 Extended and Point Defects, Doping, and Magnetism evidenced by the separation between FC and ZFC curves. In contrast, the x ¼ 0.5 multiphase (Ga,Mn)N sample (Figure 4.171 center) showed behavior consistent with spin glass behavior below 100 K. The multiphase x ¼ 0.05 (Ga,Mn)N sample shows behavior consistent with the presence of at least two ferromagnetic phases as shown in Figure 4.171 bottom. These experiments demonstrated that the conclusion of the magnetic properties of DMS materials should not rely only on the magnetization hysteresis. Detailed characterization, especially about the magnetization state of the substitutional transition metal ions, are necessary for a better understanding of the genesis of the magnetism in these materials. Despite many reports of evidence for ferromagnetism in Mn-doped GaN, there are also other reports that are in contrast to these reports [788,789]. In an effort to show that precipitates could be the source of reported ferromagnetism, Zajac et al. [788] synthesized MnxNy precipitates in (Ga,Mn)N crystals by the ammonothermal and chemical transport methods. Both (Ga,Mn)N and the reference MnxNy samples were grown using similar growth parameters such as temperature and pressure. Then the magnetic properties of the samples were investigated in an attempt to determine whether the ferromagnetism observed in Mn-doped GaN could be due to magnetic material inclusions. On the basis of the ferromagnetic behavior observed from MnxNy, it was suggested that the ferromagnetic behavior in Mn-doped GaN could be from MnxNy precipitates that could form under the same growth conditions although they may be undetectable by X-ray diffraction. 4.9.7.3 Magneto-Optical Measurements in TM-Doped GaN As discussed in conjunction with II–VI and III–V GaAs doped with magnetic ions, magneto-optical spectroscopy is a very direct method for evaluating the s,p–d exchange interactions of DMS as it probes the Zeeman splitting of the transitions at critical points associated with electronic band structure of the host material [790]. As such it is not sensitive to other phases that might be present in the magnetic ion doped host material. In addition, the technique can also be used to determine if the DMS is in a paramagnetic or ferromagnetic state. Before delving into the topic of magneto-optics in GaN, it is essential that the band structure of GaN is revisited. While the detailed band structure of GaN without magnetic field is treated in Sections 2.3–2.6, it is imperative that we establish how the conduction and valence band states split in the presence of magnetic field. Specifically, the Zeeman splitting (enhanced by s, p–d, pinteraction) and also the direction in which various bands split must be known (i.e., the sign of the g-factor in addition to its value). The reader should be cautioned that while there is a reasonable degree of work available in freestanding GaN, the magneto-optics data in TM-doped GaN are nowhere near as complete for II–VI and more advanced III–V GaAs. The following discussion would segue into magnetotransport measurements in GaN. The band structure of GaN at the G point is shown in Figure 4.172. The conduction band and valence band HH, LH (spin–orbit split off), and CR (crystal field split off) VB;HH bands are depicted by GCB , G7VB;LH , and G7VB;CR , which in order represent the 7 , G9 conduction band (1/2), valence band heavy-hole G9VB;HH ð3=2Þ, valence band lighthole GVB;LH ð1=2Þ, and valence band spin–orbit split-off G7VB;CR ð1=2Þ states. Unlike 7
4.9 Intentional Doping
+1/2 CB
Jz = −1/2
σ+
σ− π
σ+ π
σ+ σ−
−3/2 −1/2
σ−
+3/2 HH +1/2 LH +1/2 CR
−1/2 VB;HH Figure 4.172 The GCB , G7VB;LH 7 conduction band and G9 VB;CR (spin–orbit split-off band), and G7 (crystal field split off band) valance bands in wurtzitic GaN at the G point along with polarization (sþ right-hand and s left-hand circular polarizations) of various transitions between the conduction and valence band states in the presence of a magnetic field. (Please find a color version of this figure on the color tables.)
the GaAs and II–VI zinc blende cases, due to the lower symmetry, the valence band HH and LH bands are not degenerate. The transitions involving GCB 7 ð 1=2Þ states and VB;LH G9VB;HH ð 3=2Þ states, GCB ð þ 1=2Þ states and valence band G ð 1=2Þ states, and 7 7 VB;CR ð þ 1=2Þ states and valence band G ð 1=2Þ states are right-hand circularly GCB 7 7 polarized, depicted as sþ. On the contrary, the transitions involving GCB ð þ 1=2Þ states 7 VB;LH and G9VB;HH ð þ 3=2Þ, GCB ð 1=2Þ states and valence band G ð þ 1=2Þ states, and 7 7 VB;CR GCB ð 1=2Þ states and valence band G ð þ 1=2Þ states are left-hand circularly 7 7 polarized, depicted as s. The numbers in parenthesis, as in the case of zinc blende symmetry represent Jz. Because (in the absence of Zeeman splitting) the spin-up and spin-down states are not separated in energy, in case of unpolarized light excitation or electron injection, the right-hand and left-hand circularly polarized emissions would cancel each other. This is true irrespective of the excitation wavelength as each allowed VB;HH pair of the GCB ð3=2Þ, G7VB;LH ð1=2Þ(the spin–orbit split-off 7 ð þ 1=2Þ to either G9 VB;CR band) G7 ð1=2Þ (the crystal field split-off band) states is at the same energy and unpolarized light would ensue. Additionally, one must consider the selection rules to ascertain whether a given transition is allowed in the optical geometry used, the details of which are provided in Volume 2, Chapter 5 as well as in Volume 2, Figure 5.42 dealing with free excitons. When a magnetic field is applied that could be parallel or perpendicular to the c-axis, the spin-up and spin-down states split and also change in energy with magnetic field due to Zeeman splitting effect (see Volume 2, Figure 5.41). It should be mentioned that Zeeman splitting could be large if there is a strong spin exchange interaction as in the case of the established diluted magnetic semiconductors. Otherwise, to see a sizable Zeeman splitting a very large magnetic field must be applied in addition to having very sharp optical transitions. Ste2pniewski et al. [791] undertook magnetoreflectivity measurements in a GaN epitaxial layer grown on bulk
j1147
1148
j 4 Extended and Point Defects, Doping, and Magnetism GaN substrate for! a detailed investigation of Zeeman splitting of states in GaN for ! both B ==c and B ? c configurations. They identified the energy positions of the excitonic resonances associated with transitions between the GCB 7 ð1=2Þ conduction band states and G9VB;HH ð3=2Þ, G7VB;LH ð1=2Þ, and GVB;CR ð1=2Þ valence band states 7 by employing the derivatives method, dR/dE, of the reflectivity spectra and assigning the maxima of (dR/dE) to the optically active transverse excitons. The energy positions of the Zeeman split excitons measured for different magnetic fields and for ! ! two experimental configurations ( B ==c and B ? c) are shown in Volume 2, Figure ! 5.41. A schematic diagram for clarity, but for only the B ==c orientation is shown in Volume 2, Figure 5.42. Furthermore, Ste2pniewski et al. [791] used the transitions observed in the magnetoreflectivity data to resolve the symmetry of three excitonic A, B, C resonances, emanating from the band splitting caused by the crystal field and spin–orbit effects on the valence band. The measurements allowed Ste2pniewski et al. [791] to determine the symmetry-breaking diagram of the twofold degenerate conduction band and the sixfold degenerate valence band states. Note that the magnetic field causes Zeeman splitting of each of the bands, the nature of which was also determined and is shown in a simplified form of band-to-band transitions in Volume 2, Figure 5.42. To underscore the point, magneto-optical spectroscopy is a direct method of evaluating s,p-d interaction in diluted magnetic semiconductors as the method seeks to exploit Zeeman splitting in critical point energies in the material, provided that the transition probabilities (oscillator strengths to be discussed briefly shortly) of the relevant transitions are favorable. Any contribution from other phases, if present, do not contribute to the critical point transitions. This marks the difference between magnetization-based measurements made possible by SQUID magnetometers and optical measurements such as magnetic circular dichroism based on magnetooptical transmission/absorption/reflection. For the sake of the argument, let us assume that the Zeeman splitting is very small and indiscernible. If an MCD experiment were conducted in this case, the signals associated with right-hand circularly polarized light (G9VB;HH ð 3=2Þ GCB 7 ð 1=2Þ CB ð þ 3=2Þ G and that for the left-hand circularly polarized light (GVB;HH 9 7 ð þ 1=2Þ occur at the same energy and would cancel each other leading to no MCD signal (the expression for the MCD signal is equations beginning from Equation 4.134). The same is true for the transitions involving LH bands. Because of the fact that the HH and LH states are only separated by 4–6 meV, which is very small, MCD signal would have a mixture of HH and LH hole contributions due to spectral overlap. It should be noted that comparable transition probabilities for these bands have also been mentioned [792–795]. In the context of eventual use of these structures for utilizing the spin properties in devices, lack of spin polarization in electroluminescence (EL) experiment in one particular attempt has been ascribed to fast spin relaxation [796]. For the case where spin-polarized electrons are injected, transitions allowed for that particular spin should be considered. Assuming that injected electrons are spin VB;HH up, then left-hand circularly polarized GCB ð þ 3=2Þ 7 ð1=2Þ conduction band to G9 CB valence band (HH) states, and right-hand circularly polarized G7 ð1=2Þ conduction
4.9 Intentional Doping
band to G7VB; LH ð 1=2Þvalence band (LH) states (see Volume 2, Figure 5.42) would take place. This means that both and right-hand and left-hand circularly polarized light would be emitted, albeit at slightly different wavelengths. However, if these transitions are close in energy and there is spectral overlap due to broadening, the emitted light would contain both polarizations unless cancelled. In GaAs the HHrelated transitions (A excitons) is a factor of 3 stronger than the LH transitions (B excitons), giving rise to a circular light polarization of 50% (see Equations 4.123 and 4.126). In GaN, however, the transition probabilities from the conduction band to the HH and LH states might be comparable in addition to these two valence band being close in energy, in the case of which the light emitted may not be polarized or if polarized the polarization would be small. It is clear that the disparity in transition probabilities is needed for light polarization. Using band-to-band calculations at the G point in wurtzite semiconductors, the transition strengths (or oscillator strengths) for the B (aB) and C (aC) bands relative to that for the A band (aA) can be obtained for light propagation perpendicular to the c-axis as aB/aA ¼ a2 and aC/aA ¼ (1 a2) using [794] qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ðD D Þ þ ðD1 D2 Þ2 þ 8D23 1 2 1 pffiffiffi ; ð4:145Þ a ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ; x ¼ 2 2D3 x 1=x2 þ 1 whereD1 isthe crystalfield energy, and D2 and D3 are the spin–orbitcoupling parameters parallel and perpendicular, respectively, to the c-axis, which leads to a2 ¼ 0.97 and Pcirc ¼ 4% polarization for GaN if we used the parameters reported in Ref. [802] for GaN when both HHand LH bands are involved. If, on the contrary, we use the values given in Volume 2, Chapter 5 by Gil et al. [797], that is, D1 ¼ 10 0.1, D2 ¼ 6.2 0.1, and D3 ¼ 5.5 0.1 meV, we get a2 ¼ 0.62, which gives Pcirc ¼ 23% instead of 1.5% suggested byChoietal.[795]WeshouldpointoutthatChuangandChangarguethattheparameters used by Gil et al. are obtained by linearization while theirs are more precise parameters, that is, D1 ¼ 22 meV and D2 ¼ D3 ¼ 11/3 meV, by using the full expressions to fit the same experimental data. In the band-to-band model, the excitonic effects have been neglected in that the short-range electron-hole spin exchange interaction strongly mixes the excitons built from the G7 and G9 valence bands, but again this effect is small in GaN with a moderate exchange interaction (0.69 meV [798]) compared to the spin–orbit coupling and crystal field splitting parameters, and has almost no effect on the oscillator strengths of the optical transitions [799]. For information, the exchange interaction in ZnO is much larger (4.7 meV) [800,801]. The oscillator strengths and therefore the transition intensities can be modified by strain, induced by either thermal mismatch with the substrate and/or slightly lattice mismatched heterojunctions as in quantum wells. When integrated into the simple band-to-band model, the strain changes the relative oscillator strengths of the A, B, and C excitons drastically [802], as also verified by experiments in GaN [803]. It should be mentioned that the MCD experiments are different in that as long as band coupling to right-hand and left-hand circularly polarized light is sufficiently separate in energy due to Zeeman splitting, an MCD signal should be obtainable.
j1149
j 4 Extended and Point Defects, Doping, and Magnetism GaN:Mn H=1T 0.01 Magnetic circular dichroism (degree)
1150
300K
0 –0.01
0.01
6K
0 –0.01
1
2
3
4
Photon energy (eV) Figure 4.173 MCD spectra of GaN:Mn (6.8%) taken at 300 and 6 K with a magnetic field strength of 1 T. The inset associated with the 6 K data magnifies the MCD structure to show the contribution from diamagnetic GaN. Courtesy of Ando [804].
On the magneto-optics experimental side, Ando [804] employed MCD spectra to investigate possible ferromagnetic behavior in a 300 nm thick (Ga,Mn)N film grown on a 200 nm thick GaN which in was grown on (0 0 0 1) sapphire. The Mn concentration was 6.8% and no other phase, other than GaN, was detected in X-ray data. Further details related to the sample can be found in Ref. [805]. SQUID magnetometer measurement showed the typical hysteresis with an apparent Curie temperature above 300 K and a coercive field of 8.2 mTat the same temperature. The MCD spectra showed structures around the fundamental bandgap, 3.47 eV, as shown in Figure 4.173 wherein a broad MCD peak is visible at around a photon energy slightly below the fundamental band edge. It should be stated that there is no apparent contribution from the sapphire substrate to the MCD signal. Further, the magnetic field dependence of the MCD signal obtained at a photon energy of 3.54 eV is representative of a straight line that is not shown. Because the MCD signal is proportional to magnetization (Equation 4.139), the magnetic field dependence of the MCD signal should be consistent with ferromagnetism, which in this particular sample is not. Figure 4.174 shows the temperature dependencies of the MCD signal corresponding to the positive peak at a photon energy of 3.54 eVat 6 K and a negative peak at 3.20 eV. The dashed line shows the MCD intensity at room temperature, which arises from the Zeeman splitting of the diamagnetic host GaN given by g mBH. We should mention that all semiconductors are diamagnetic at room temperature unless made paramagnetic or
4.9 Intentional Doping
GaN: Mn H = 1T
Magnetic circular dichroism (degree)
0.05
3.54 eV (Γ)
0
3.20 eV
– 0.02 0
100
200
300
Temperature (K) Figure 4.174 Temperature dependences of the MCD intensities of the 3.54 eV peak and the negative signal at 3.20 eV. The dashed line shows the MCD intensity value at room temperature. Courtesy of Ando [804].
ferromagnetic by transition metal or rare earth dopants. Both MCD signal intensities at 3.54 and 3.2 eV show similar Curie–Weiss-type temperature dependencies, which indicate that the sample is paramagnetic and is responsible for the observed MCD spectrum at 6 K. Moreover, a clear MCD peak around the GaN fundamental gap energy (Figure 4.173) clearly suggests that this paramagnetic material has a common band structure with GaN, and its s,p-bands at the G critical point are influenced by the magnetization (d-electrons). Therefore, Ando [804] concluded that the sample is simply a paramagnetic, diluted magnetic semiconductor Ga1xMnxN. Furthermore, the small positive going shoulder on the strong 3.50 eV MCD peak, shown in the circular inset in Figure 4.173, is in reality the MCD signal of the diamagnetic GaN buffer layer, the peak position of which shifts to a higher energy due to the blue shift in the bandgap energy of GaN with decreasing temperature. Let us now analyze and interpret the observations, as was done by Ando [804]. In contrast to the negative MCD peak observed in Cd1xMnxTe [745], note that the MCD peak at 3.54 eV is positive. This particular MCD peak could have its genesis either in a superposition of the A (G9 ! G7) and B (G7 ! G7) transitions (see Figure 4.172 for a descriptive view of the transitions and note that the separation between the A- and B-transitions in GaN is very small [806]) or in the C-transition (G7 ! G7). Ando [804] argued that because GaN comprises light elements as its constituents, its spin–orbit interaction is small [806], in the case of which the Zeeman splitting of the A-, B-, and Ctransitions is expected to have roughly the same magnitude [807,808]. The sign of the
j1151
1152
j 4 Extended and Point Defects, Doping, and Magnetism Zeeman-splitting associated with the B-transition is therefore expected to be different from the ones for the A- and C transitions [807,808]. Consequently, the magneto-optical effect associated with the A- and B-transitions could possibly cancel each other if their transition energies completely fall on each other. This leaves only the magneto-optical effect associated with the C-transition. Ando [804] suggested that while such complete cancellation is generally rare, this possibility must be considered. However, as shown by Ste2pniewskiet al. [791] and inVolume2, Figure 5.41,following Zeeman splitting none of the spin-up and spin-down energies associated with A (G9 ! G7) and B(G7 ! G7) transitions are identical however close they may be. If the sample quality is not sufficiently high and the transitions associated with A and B excitons are broad and/ or both peaks are lumped together, it may not be possible to detect a clear MCD signal. In Mn-doped GaN, broad MCD structures below Eg are also present [804]. The investigations of the optical absorption spectra of lightly Mn-doped GaN have shown that two absorption bands below the bandgap are related to the Mn-acceptor level [588–591,603]. Therefore, the MCD structures observed by Ando [804] below 3.5 eV correspond to these two absorption bands. It should be noted that the same broad MCD structures below the bandgap were also observed in a paramagnetic GaN: Mn [809]. The sample investigated by Ando [804] exhibited possible ferromagnetism even at room temperature by the magnetization curves. However, no MCD signal was detected from that ferromagnetic material by Ando [804]. It is then fair to conclude that any ferromagnetism observed in the M–H curves must most likely have been associated with some unidentified ferromagnetic material that is different from that of Ga1xMnxN. As iterated, this simply is a demonstration that crystallographic techniques such as X-ray diffraction, and TEM would not always be able to detect any different phase responsible for magnetism. Moreover, the SQUID method is so sensitive that even the contribution by minute amounts of ferromagnetic impurities to magnetism could lead to false positive results. In the sample investigated by Ando [804], the volume fraction of the impurities was estimated approximately at about 1% and the measured value of the saturation magnetization of the sample, 7 emu cm3, is about 2 orders of magnitude smaller than the saturation magnetizations of typical ferromagnetic materials. The divergence of the SQUID data and magneto-optical measurements is not unique to GaN in that ZnO:Co [810] and ZnO: Ni [811], deemed to be ferromagnetic by SQUID measurements, did not agree with the paramagnetic or diamagnetic MCD data [812]. To make the point again, other III–V and also II–VI materials doped with transition elements showed ferromagnetism, as determined by SQUID measurements, and also showed magneto-optical response as MCD signal. Experimentally, the conventional II–VI DMS [745] and a paramagnetic III–V DMS Ga1xCrxAs [813] showed strong MCD enhancements at their CPs, as expected. The strong enhancements of the MCD signals at the CPs were also observed in Ga1xMnxAs [772], In1xMnxAs [814], and Zn1xCrxTe [737]. The field dependencies of their MCD intensities clearly proved that these three materials are the intrinsic ferromagnetic DMSs [737,772,814]. On an equally serious note, which does not bode well for Mn-doped GaN in terms of its prospects for ferromagnetism, the holes have been reported to be at the d shell of Mn (Mn2þ (d5)) by several
4.9 Intentional Doping
1.0 0.9
Magnetization (emu/cm3)
Saturationmagnetization(emu cm–3)
researchers [590,591,815], implicating that magnetization induced by hole exchange interaction may not be plausible in (Ga,Mn)N. The scrutiny and concern raised in the previous paragraph are equally applicable to another transition metal doped semiconductor, ZnO:Mn. In fact, the same may even be said with reasonable certainty about Co- and also Cr-doped ZnO. In the case of ZnO:Mn, the argument has been made that observation of hysteresis in magnetization curves alone is not sufficient for determining the Curie temperature. Even within the confines of magnetization measurements alone, the real test actually is more involved in that the diamagnetic component due to the substrate material and paramagnetic component provided by the transition metal alone must all be subtracted from the magnetization curves first. This involves the use of applicable theory in conjunction with parameters determined from experiments. A careful analysis has been undertaken for RF sputtering deposited Mn-doped ZnO, which concluded that the material is paramagnetic not ferromagnetic [816]. A more detailed discussion on ZnO can be found in Chapter 14. In contrast to the picture painted in the foregoing paragraphs, reports of ferromagnetism based on magnetization measurements continued. For example, Reed et al. [817] proposed that in (Ga,Mn)N when the Fermi position is such that there is a partial occupancy of electrons and holes in the Mn impurity band, ferromagnetic behavior occurs. This suggestion needs to be considered in terms of Mn-induced levels/bands within the bandgap to devise a scientific explanation of these observations. The dependence of the ferromagnetic properties on the Fermi level was demonstrated by doping (Ga,Mn)N with n-type or p-type dopants. Shown in Figure 4.175 is the effect of silane flow on the saturation magnetization of the (Ga,Mn)
0.8 0.7 0.6 0.5
1.0 0.5 0.0 –0.5 –1.0 –2000
0.4
Si = 2 nmol min-1 –1000
0
1000
2000
Applied field (Oe)
0.3 0.2 0.1 0.0 0
20
40
60
Silane flow (nmol min–1)
Figure 4.175 The influence of silane flow on the saturation magnetization in (Ga,Mn)N:Si films. Note the reduction in the magnetic strength with increased Si doping. Inset shows the magnetization curve for sample (b) at low magnetic fields, which is typical of the samples
80
used in this study. This sample has a residual magnetization of 0.154 emu cm3, a coercivity of 100 Oe, and saturation magnetization of 0.72 emu cm3 at 2.2 kOe, at room temperature. Courtesy of El-Masry and coworkers [817].
j1153
j 4 Extended and Point Defects, Doping, and Magnetism N:Si films. The data indicate that high silicon concentrations in (Ga,Mn)N:Si films eliminate the FM behavior determined by the magnetization curves, whereas a much stronger FM response was observed for low to moderately Si-doped samples. In the case of high Si doping, it is expected that EF is close to the conduction band; therefore, the argument goes that the deep Mn band is completely filled with electrons, leaving no available holes to mediate the magnetic exchange interaction. When doped with Mg, electrons in the Mn impurity band are absorbed by the normally empty Mg acceptor states that is 1.2 eV below the Mn band, so that these electrons have nearly zero probability of returning to the Mn energy band where they would facilitate the Mn–Mn exchange interaction. Therefore, the addition of Mg to an otherwise FM (Ga,Mn)N film is expected to reduce the FM response, and the introduction of Mg acceptor states at a concentration that is the same as that of Mn nearly eliminates the films FM response, as it has been observed. Arkun et al. [818] reported on the dependence of ferromagnetic properties of MOCVD-grown (Ga,Mn)N films on carrier transfer across adjacent layers. They found that the magnetic properties of (Ga,Mn)N, as part of (Ga,Mn)N/ GaN:Mg heterostructures, depend on the thickness of the GaN:Mg layer. These results were explained based on the occupancy of the Mn energy band and how the occupancy can be altered due to carrier transfer at the (Ga,Mn)N/GaN:Mg interfaces. The authors also found that the magnetic properties of the GaMnN/AlGaN/GaN:Mg depends on the thickness of the AlGaN barrier layer, as shown in Figure 4.176. The presence of this AlGaN barrier layer thus affects the carrier transfer from the GaMnN film to the GaN: Mg layers. These results seem to be consistent with the model proposed by Dalpian et al. [701] in that ferromagnetic state can only be stabilized when Mn levels have both holes and electrons. Mn-doped GaN samples have also been produced by MBE technique [781] and these samples have been reported to exhibit ferromagnetic properties at room 4
Magnetization (emu cm–3)
1154
3 2
AlGaN barrier thickness tb = 50 nm tb = 25 nm
1 0 –1 –2 –3 –4 –5000
–2500
0
2500
5000
Applied filed (Oe) Figure 4.176 Magnetization versus applied field for GaN:Mg/ AlGaN/(Ga,Mn)N/AlGaN/GaN:Mg DHS containing 25 and 50 nm thick AlGaN barriers The thickness of the (Ga,Mn)N and GaN:Mg layers are fixed at 0.38 and 0.75 mm, respectively. Courtesy of El-Masry and coworkers [818].
4.9 Intentional Doping
0.15
Cr: 1%
5K
0.2μ μ B /Cr
Cr: 5%
0.1 Magnetization (emu g–1)
Cr : 3% 0.05
Cr : 0.5% 0
–0.05
–0.1
–0.15 –10 000
–5000
0
5000
10 000
Magnetic field (Oe)
–1 Magnetization, (emu g )
0.05
Cr: 5%
Cr: 1% 0
–0.05 –1000
Hc = 100 Oe
0 Magnetic Field (Oe)
Figure 4.176 (Continued )
1000
j1155
1156
j 4 Extended and Point Defects, Doping, and Magnetism temperature for low concentration of Mn (3% of Mn). Optical properties of (Ga,Mn) N thin films are not available yet. However, optical injection on n-GaN samples [819] implanted with high doses of Mn (3 · 1016 cm2) reveals that Mn forms a deep acceptor near En þ 1.8 eV. The expected wonder of the reader as to whether Mn-doped GaN is ferromagnetic or not is well taken and reflects the murky state of the literature on the topic. In those instances where ferromagnetism has been reported, the Curie temperatures span from 20 to 940 K. Although the growth mechanism seems to play a vital role, the reason for such a wide variation of TC is not understood and it is reasonable to speculate about the likelihood of contributions from materials other than the (Ga, Mn)N proper, if any. There appears to be some hints from theory that ferromagnetism is very much dependent on the doping level and a narrow range of doping levels may lead to ferromagnetic state being stable. From an experimental point of view, it would be fair to conclude that perhaps Mn is therefore not the best choice for obtaining ferromagnetism in GaN. 4.9.8 Magnetic, Structural, Optical, and Electrical Properties of Cr-Doped GaN
Before delving into Cr-doped GaN, it is instructive to revisit briefly the section dealing with the theoretical simulations. Cr-doped GaN was found to have the most stable ferromagnetic state in transition metal doped GaN by Sato et al. [690] Das et al. [703] performed a first principles calculation within the framework of linearized muffin-tin orbital (LMTO) tight-binding method and gradient-corrected (GCA with approximation added toward the end) DFT. They predicted the coupling between Cr atoms to be ferromagnetic, with a magnetic moment of 2.69 mB for the case of bulk GaN and 4 mB for the case of GaN clusters. Prior to the abovementioned theoretical work, Cr-doped GaN single crystals grown by the flux method [820] and thin films grown by ECR MBE [821] had been grown and studied for their magnetic properties. The single-crystal (Ga,Cr)N samples fabricated by adding Cr to GaN single crystal by flux method were reported to show ferromagnetic TC at 280 K, whereas the thin films grown by ECR MBE were reported to display ferromagnetism with a Curie temperature above 400 K. The ferro- to paramagnetic transition at approximately 280 K was also observed in the temperature-dependent resistance measurement at zero-magnetic field. The M–H hysteresis loops yielded coercive fields of 54 and 92 Oe at 250 and 5 K, respectively. The carrier density in this crystal was about 9 · 1018 cm3 (n-type) with a mobility of 150 cm2 V1 s1. It should be pointed out that the electron mobility appears to be inconsistently high for the electron donor concentration not to mention any adverse effect of Cr doping. In another report [821], the (Ga,Cr)N thin films grown by ECR MBE were deemed very encouraging, because they showed ferromagnetism with TC above 400 K and remarkable M–H hysteresis loops with a saturation field of about 2000 Oe and a coercive field of about 55 Oe at 300 K. Again, caution must be exercised here as to the characterization of the observed magnetism.
4.9 Intentional Doping
0.15
Cr: 1%
5K
0.2 μ B /Cr
Cr: 5%
0.05
Cr: 3%
0.05 Cr: 0.5%
0
–0.05
Magnetization (emu g–1)
Magnetization (emu g–1)
0.1
j1157
–0.1 –0.15 –10 000
Cr : 5% Cr :1%
0
H c =100 Oe
–0.05 –1000
0
Magnetic Field (Oe)
–5000
0
5000
10 000
Magnetic Field (Oe) Figure 4.177 Magnetization curves for Crdoped GaN in atomic concentrations of 0.5, 1, 3, and 5% up to a magnetic field normal to the surface of 10 000 Oe (1 T). Note that the film containing 5% Cr does not show any saturation magnetization in the range measured and
appears to be paramagnetic. The blow-up version near the origin indicates hysteresis for 1% Cr sample and a coercive field of 100 Oe. Courtesy of F. Hasegawa. (Please find a color version of this figure on the color tables.)
The GaN films doped with Cr and grown with MBE using NH3 for nitrogen source (either on (1 1 1) Si or (1 1 1) GaAs substrates, which is inconsequential as far as the implications here are concerned) have been thoroughly investigated in terms of direct magnetization measurements and also by magneto-optical measurements such as MCD [822]. Films containing 0.5, 1, 3, and 5 at.% Cr were measured with a SQUID magnetometer with the film with 1% Cr showing the largest hysteresis and a saturation field of 1 T, as show in Figure 4.177. The saturation magnetization of 0.14 emu g1 at 5 K for the 1% sample corresponds to 0.2 mB per Cr atom. The diamagnetic component due to the Si substrate was subtracted, and that with 5% Cr behaved similar to a paramagnetic material and did not show any saturation. The antiferromagnetic behavior of the 5% sample may be due to antiferromagnetic CrN that may form in samples with high Cr concentration. In terms of the electrical conduction, unintentionally doped samples exhibited electron concentrations in the range of 1019 cm3 that decreased to 1015 cm3 by Cr doping initially but climbed back up to 1019 cm3 again for a Cr concentration of 10%. The temperature dependence of magnetization measured at H ¼ 200 Oe for 1, 3, and 5% Cr-containing GaN is shown in Figure 4.178. The temperature dependence in the low temperature region (circled) is consistent with ferromagnetic, combination of ferromagnetic and paramagnetic, and paramagnetic behavior for 1, 3, and 5% Cr (atomic percentile) containing GaN, respectively. Specifically, the dependence for 3 and 5% sample can be described with paramagnetic behavior. The 1% sample shows little temperature dependence, which is consistent with ferromagnetic behavior.
1000
j 4 Extended and Point Defects, Doping, and Magnetism 0.04 H = 200(Oe)
Magnetization (emu g–1)
1158
Ferromagnetic
Cr: 1%
0.02
Paramagnetic + ferromagnetic
Cr: 3 %
Paramagnetic
Cr : 5%
0 0
50
100
150 200 250 Temperature (K)
300
350
Figure 4.178 Temperature dependence of magnetization for 1, 3, and 5% Cr-containing GaN. As indicated, the film with 1% Cr is consistent with ferromagnetic behavior. The films with 3 and 5% Cr exhibit a combination of ferromagnetic and paramagnetic behavior, and paramagnetic behavior, respectively. Courtesy of F. Hasegawa. (Please find a color version of this figure on the color tables.)
Magnetic circular dichroism measurements in Cr-doped GaN (1.5% atomic percentile) samples have also been carried out at a magnetic field of 1 T that was normal to the surface. The MCD spectra displaying 1.65 and 3.49 eV peaks at 6 K are shown in Figure 4.179. The peak appearing at 1.65 eV is most likely associated with Cr4þ1 EðFÞ ! 3 A2 ðFÞ internal transition (d–d ) that occurs at 1.7 eV, see Figure 4.134. In samples grown using di-methyl-hydrazine (DMHy) for N source as opposed to NH3, another peak at 1.9 eV appeared that has been attributed to the 1 T1 ðFÞ ! 3 A2 ðFÞ internal d–d transition of Cr4 þ charge state, see Figure 4.134. The peaks appearing at 3.43 eV in the 295 K spectrum and 3.49 eV in the 6 K spectrum are associated with the band edge of GaN. However, the most striking data are that the MCD intensity of both the 1.9 and near 3.4 eV signals obtained at 6 K varied linearly with magnetic field, with no hysteresis loop [823]. This means that MCD data do not support the presence of any ferromagnetism in spite of the fact that magnetization measurements indicated otherwise. Moreover, another sample containing 1% Cr (which showed substantial hysteresis in SQUID measurements) was also measured in terms of the MCD band edge signal as a function of magnetic field (normal to the sample surface) at 6 K. The MCD and SQUID data measured versus magnetic field are shown together in Figure 4.180. Although the SQUID data show saturation and some hysteresis, the MCD signal varies linearly with magnetic field with no hysteresis [824]. The MCD data are in general in agreement with the Soft X-ray magnetic circular dichroism (XMCD) measurements reported by Makino et al. [825] and performed in Ga0.97Cr0.03N films grown by NH3 MBE in Cr-doped GaN. The temperature
4.9 Intentional Doping
Cr: 1.5% H = 10 kOe
(H ⊥ Sample)
3.43 eV
295 K 0 MCD
3.49 eV 6K 1.65 eV
0
1
2
3 4 5 Photon energy (eV)
6
Figure 4.179 MCD spectra obtained in Cr-doped GaN (1.5%) at 6 and 295 K for a magnetic field of 10 kOe, which was normal to the surface. The peak at 1.65 eV is most likely associated with Cr4þ d–d transitions. The near band edge transition (3.43 eV at 295 K and 3.49 eV at 6 K) is associated with band edge with no clear MCD signal. Courtesy of K. Yamaguchi and F. Hasegawa.
Figure 4.180 The magnetic field dependence of the band edge 3.4 eV MCD peak intensity measured at 6 K for 1 at.% Cr-doped GaN. The SQUID data of the same sample showing saturation and hysteresis are also displayed. The lack of saturation and hysteresis in the MCD data is in contrast to that of the magnetization data and implies that the film is not ferromagnetic. Courtesy of K. Yamaguchi and F. Hasegawa.
j1159
j 4 Extended and Point Defects, Doping, and Magnetism dependence of the XMCD intensity could very well be characterized by the Curie–Weiss law. This sample too showed ferromagnetism up to room temperature while the XMCD data did not show any ferromagnetism. The paramagnetic/diamagnetic nature of the MCD data suggests that homogenously Cr-doped GaN is not responsible for the magnetization data. This simply indicates that the magnetization measurement with a Curie temperature of greater than 350 K is most likely caused by another phase in the material not detected by X-ray measurements. One plausible explanation might be that films contain CrO2 precipitates. The Mn analogue might be MnN4 in Mn-doped GaN, which shows magnetism in magnetic measurements but not in optical measurements. For the sake of completeness, a discussion of optical transitions in Cr-doped GaN is warranted. In this vein, Hashimoto et al. [826] studied the structural and optical properties of (Ga,Cr)N layers grown by ECR MBE, which purportedly exhibited ferromagnetic behavior at room temperature. From XRD and EXAFS measurements, the authors could not detect any second-phase material. Some common compounds of Cr with Ga or N include CrGa4, CrN, Cr2N. Among them, CrGa4 and
3.47 eV
GaCr (Cr : 1.5%) 3.29 eV FWHM: 75 meV
GaN buffer layer
10 K
PL Intensity(au)
1160
30 K 50 K 80 K 120 K 160 K 200 K 300KK 300 2.4
2.8
3.2
Photon energy (eV) Figure 4.181 Temperature dependence of PL spectrum for the (Ga,Cr)N layer with the Cr concentration of 1.5% in the range of 10–300 K. Courtesy of Asahi and coworkers [826].
3.6
4.9 Intentional Doping
CrN have been shown to be nonferromagnetic [827–829], and the magnetic properties of Cr2N are not yet known. From the optical emission point of view, PL emission at 3.29 eV (10 K) was observed in Ga0.985Cr0.015N film, as shown in Figure 4.181, which was attributed to band-to-band transitions in GaCrN. The assignment was deduced from temperature- and excitation-power-density dependent PL measurements. It was also found that the PL emission peak energy in GaCrN decreases with increasing temperature in accordance with the Varshni formula similar to the GaN excitonic transition peak. If truly confirmed as being ferromagnetic, Cr-doped GaN may provide new hope for spintronics applications at room temperature and above. However, for Cr and particularly other transition metal doped GaN, further research is imperative to be certain of the observed results as being due to carrier-mediated ferromagnetism. At the very least, the basis for the observations need to be determined. In another investigation but still within the realm of Cr-doped GaN, Singh et al. [830] studied the effect of substitutional Cr on the magnetic properties of (Ga,Cr)N. For this particular investigation, several (Ga,Cr)N films with identical Cr concentration (3%) were grown at different temperatures by MBE. The fraction of Cr at the substitutional sites (CrGa) was determined from the RBS-channeling angular distribution of Cr and Ga in the h0 0 0 1i axial direction, as shown in Figure 4.182. The results indicate that 78–90% of Cr did indeed occupy the substitutional sites for GaN grown in the temperature range of 700–775 C, and HRTEM showed that Cr distributed uniformly in the lattice. However, only a small fraction of Cr (<20%) was found to be located at the substitutional sites for film grown at 825 C, and a
Figure 4.182 Channeling angular scans in h0 0 0 1i axial direction for (Ga,Cr)N films grown at (a) 700, (b) 740, (c) 775, and (d) 825 C [830].
j1161
1162
j 4 Extended and Point Defects, Doping, and Magnetism significant Cr clustering is revealed from energy-filtered TEM. A higher magnetic moment (0.35 mB/Cr versus 0.13 mB/Cr) and higher percentage of magnetically active Cr ions (12% versus 4%) were found in the films with high substitutional concentration as compared to those with lower substitutional concentration. These results serve to establish that the location of Cr in the (Ga,Cr)N lattice plays a crucial role in determining its magnetic properties. Within the general scope of Cr-doped nitride semiconductors, Cr-doped AlN has also been reported to exhibit high-temperature ferromagnetism. In this vein, Liu et al. [831] have reported TC over 900 K in both (Ga,Cr)N and (Al,Cr)N films prepared by reactive MBE on 6H-SiC substrates. Cr-doped GaN was found to have the highest saturation magnetic moment (Ms) at a Cr content of 3%, which corresponds to 70% of Cr impurities on the substitutional sites for (Ga,Cr)N by channeling Rutherford backscattering. Cr-doped AlN displayed the same trend but with the highest Ms for a Cr content of 7%. Liu et al. [831] have suggested that substitutional Cr is responsible for the observed magnetic behavior. In the same study, electrical transport and magnetotransport properties of Cr-doped GaN have also been measured, as shown in Figure 4.183. The thermally activated process followed R ¼ R0 exp[(T0/T)1/2], which is characteristic of variable range hopping between localized states with a Coulomb gap [832] which in this case is the variable range hopping in the Cr impurity bands. As for the deep-level traps, Cr is known experimentally [833], and predicted theoretically [834], to form a deep level at an energy of 2.0 eV above the GaN valence band. The negative magnetoresistance seen in Figure 4.183b in the range of 2 and 300 K has been attributed to a mechanism, originally proposed by Sivan et al. [835] for nonmagnetic semicon-
Figure 4.183 Temperature-dependent transport measurements of (Ga,Cr)N deposited on sapphire (a) with resistivity (r). Inset shows comparison between experimental data and functional relationship expected for variable range hopping. (b) Magnetoresistance (DR/R), the relative exchange of sheet resistance in a magnetic field. Courtesy of Newman and coworkers [831].
4.9 Intentional Doping
ductors, that takes into account the influence of the magnetic field on the quantum interference between the many paths linking two hopping sites [836]. In the variable-range hopping regime and at small magnetic fields, the MR is consistent with the results of temperature-dependent resistance. The above experimental results strongly suggest that ferromagnetism in (Ga,Cr)N and (Al,Cr)N can be attributed to the double exchange mechanism as a result of hopping between nearmidgap substitutional Cr impurity bands. 4.9.9 Other TM and Rare Earth Doped GaN:(Co, Fe, V, Gd, and so on)
In Co-implanted n-type GaN [606], it was found from the optical transmission data that Co, similar to Mn, also forms a deep acceptor at about Ev þ 1.9 eV. The Co complexes formed with native defects act as deep electron traps with a level near Ec 0.5 eV, which are most likely responsible for a strong blue luminescence band with energy near 2.9 eV. The other deep-level electron traps at Ec 0.25 and 0.7 eV are probably associated with a defective region adjacent to the implanted region, due to out-diffusion during the postimplant annealing treatment. Utilizing another magnetic metal, Shon et al. [613] reported implantation of Feþ ions into p-type GaN. The apparent ferromagnetic hysteresis loops in individual measurements performed at 10 and 300 K were used as the basis to cite ferromagnetism. The ferromagnetism was confirmed by temperature-dependent magnetization measurements that yielded persistent ferromagnetism above 350 K. Optical transitions were observed in PL measurements at 2.5 eV and around 3.1 eV, which were linked to the presence of Fe. Further, the luminescence peak at 2.5 eV has been identified as a donor–Fe acceptor transition, while the 3.1 eV peak as a conduction band–Fe acceptor transition. Vanadium, on the contrary, when implanted into p-type GaN films at doses of 3 · 1016 and 5 · 1016 cm2, has been found experimentally [706] to be paramagnetic up to 300 K, contradicting the theoretical prediction [690] that V-doped GaN has ferromagnetic states even without any additional carrier doping treatment. While on the topic, V-doped ZnO also looks promising. Doping with rare earth metal, such as Gd in GaN, has also been reported to lead to ferromagnetism above room temperature. Teraguchi et al. [837] prepared Ga0.94Gd0.06N ternary alloys on 6H-SiC substrates by RF MBE. Those authors observed hysteresis in M–H curves at all measurement temperatures ranging from 7 to 400 K, with a coercivity, Hc, of approximately 70 Oe at 300 K. Accordingly, an emission peak around 645 nm was observed in the cathodoluminescence spectra believed to originate from Gd3þ, in analogy to Eu-doped GaN [838]. In the GaN environment, there are several phases that could be formed with participation of Gd. Of particular importance to the present topic, Gd element and the rocksalt structure of GdN could indeed form a ferromagnetic material at Curie temperatures of 307.7 K (others reported it being 289 K [839]) and 72 K [840], respectively. The Curie temperature for both materials is lower than that of Ga0.94Gd0.06N observed in the measurement (>400 K) of Teraguchi et al. Therefore, the higher Curie temperature of
j1163
1164
j 4 Extended and Point Defects, Doping, and Magnetism Ga0.94Gd0.06N was deemed to be due to neither the Gd element nor the rocksalt structure of GdN. Rather, it was argued to be plausibly associated with the formation of (Ga,Gd)N ternary alloy. A later investigation performed by Dhar et al. [839] on a series of Gd-doped GaN samples with Gd concentration ranging from 7 · 1015 to 2 · 1019 cm3 demonstrated very high magnetization. Dhar et al. [839] also reported above room temperature ferromagnetism for all (Ga,Gd)N samples even with Gd concentrations less than 1016 cm3, for both MBE-grown (Ga,Gd)N and ion-implanted (Ga, Gd)N films. From the M–H curves as well as the M–T curves measured under both FC and ZFC conditions, the inferred Curie temperature is above 360 K. Further, the Curie temperature continually increased with increasing Gd concentration to well above room temperature. Most strikingly, magnetization measurements yielded an unprecedented magnetic moment of up to 4000 mB per Gd atom observed in the most lightly doped sample, 7 · 1015 cm3, as determined by SIMS analysis. Owing to the large Gd-Gd distance in the very lightly Gd-doped GaN, the microscopic origin of the magnetic coupling leading to ferromagnetism could neither be explained simply in terms of the direct exchange, double exchange, or superexchange interactions amongst the Gd atoms, nor be accounted for by the free carrier mediated RKKY-type long-range coupling, given the electrically highly resistive nature of the alloy after doping with Gd. Dhar et al. [839] believe that the high TC ferromagnetism is closely related to their observation of colossal magnetic moments of Gd in GaN, the latter of which may be explained by the long-range spin polarization of the GaN matrix induced by Gd atoms. This long-range spin polarization of the GaN matrix was reported to be consistent with the circular polarization in magneto-photoluminescence measurements in lightly Gd-doped GaN. The TC dependence on Gd concentration can also be qualitatively explained within the framework of percolation theory [841]. Additional support for ferromagnetism comes from the calculations of Dalpian and Wei [707] who argued that while Gd-doped zinc blende GaN doped with donors is stabilized in the antiferromagnetic state in that intruding electrons and strong exchange interaction with s electrons of the host would stabilize the ferromagnetic state. Magnetization data obtained in a Gd-implanted sample at concentration of 1018 cm3 are shown in Figure 4.184 in the form of magnetization versus magnetic field, magnetization at 1000 Oe versus temperature for FC and ZFC conditions, and Arrott plots determined from magnetization versus magnetic field data. It appears that a magnetization of 1000 mB per Gd atom is achieved, which is indicative of interaction with host material, as this figure is much larger than 8 mB expected from a Gd atom. These findings are very promising, because (Ga,Gd)N may be easily doped with donors (acceptors) with a concentration exceeding that of Gd, to generate spinpolarized electrons (holes) in the conduction band (valence band). Gd-doped GaN with its above room temperature TC might thus be a very attractive candidate for future semiconductor-based spintronics. Because cgs and SI units are both used in the literature, unfortunately sometimes interchangeably, a conversion table between the two as well as definition of pertinent magnetism parameters along with their units is tabulated in Table 4.15.
4.9 Intentional Doping
T=5K T = 300 K nGd = 1018 cm –3
500
0
600
Magnetization (µB /atom)
Magnetization( μ B/ atom–1)
1000
–500
400 200 0
–200 –400
–1000 –3000 –2000 –1000
–600 –500
0
1000
740
2000
3000
ZFC (H = 1000 Oe) FC (H = 1000 Oe)
720
–1 Magnetization(μ Batom )
500
Magnetic field (Oe)
(a)
700 680 660 640 620 Gd–3B
600
n
μ
0
Magnetic field (Oe)
580 0 (b)
18
Gd
= 10 cm
50
–3
100
150
200
250
300
350
Temperature (K)
Figure 4.184 (a) M–H curves showing hysteresis loops at 5 K (circles) and 300 K (squares), for a Gd-implanted sample with Gd concentration of 1018 cm3. The inset is that of (a) but for magnetic fields of 500 Oe. M-T curves at FC and ZFC conditions at H ¼ 1000 Oe are plotted in (b) for a sample with Gd concentration of 1018 cm3, which appear to indicate a Curie temperature above 300 K. The Arrott plot of the same sample measured at 5 and 300 K are shown in (c). If one were to
extrapolate from large magnetic fields on a straight line, those lines would intercept the horizontal axis on the negative side for both 5 and 300 K data indicating that the Curie temperature is above the measurement temperature. However, the Arrott plots do not follow a straight line away from zero magnetic field as they should, see Figure 4.142, and therefore it cannot be stated with requisite certainty what the Curie temperature is.
j1165
j 4 Extended and Point Defects, Doping, and Magnetism 1x10
6
8x10
5
6x10
5
4x10
5
2x10
5
2
5K
2
M (μΒ Gd)–
1166
0 0
300 K
1
2
3
4
5
μ 0H/M Figure 4.184 (Continued)
4.9.10 TM-Doped Nanostructures
Incorporation of Mn into GaN nanowires has also been reported. Deepak et al. [842], Han et al. [843], Choi et al. [844], and Liu et al. [845] produced Mn-doped GaN nanowires with the impetus that small dimensions are needed to take advantage of the spin. These materials could be free of defects, perfectly single crystalline, and have been reported to have a homogeneous distribution of Mn. In addition, these Mndoped GaN nanowires have been reported to be ferromagnetic with a Curie temperature of up to 300 K. It should be pointed out that critical optical and electrical measurements are lacking to corroborate the magnetization measurements. The nanowires are expected to have interesting magnetoelectronic properties because of the confinement of carriers in the radial direction and large magnetic anisotropy energy. In this vein, Wang et al. [846] performed first principles calculations for Cr-doped GaN nanowires with diameters of 0.45 and 1 nm and predicted ferromagnetism regardless of the site occupancy of Cr is a each Cr atom carries a magnetic moment of about 2.5 mB (Bohr magneton). In contrast, in Mn-doped GaN nanowires the magnetic coupling between the Mn atoms is sensitive to the Mn-Mn and Mn-N distances [847]. The magnetic moment at the N site, however, was found to be small and aligned antiferromagnetically to the moments at the Cr atom. The magnetization axis was predicted to be perpendicular to the axis of the wire, but with a small anisotropy in energy. Wang et al. [846,847] also considered Mn-doped nanowires. Using first principles theory, the authors have also shown that Mn-doped GaN nanowires with diameters of 0.45 and 1 nm are ferromagnetic. However, this ferromagnetic coupling between the Mn spins, driven by a double exchange
4.9 Intentional Doping
mechanism, was found to be sensitive to the Mn-Mn and Mn-N distances. Furthermore, calculations of the anisotropic energy showed that the magnetic moment orients preferably along the ½1 0 1 0 direction while the wire axis points along the [0 0 0 1] direction. Deepak et al. [842] reported the synthesis of (Ga,Mn)N nanowires of two different groups with average diameters of 25 and 75 nm, respectively, and with Mn concentration of 1, 3, and 5%. All the samples showed magnetic hysteresis at 300 K. The Mn2þ-related peak was observed in PL measurements, with a significant blue shift of this Mn2þ-related emission for the 25 nm diameter nanowires as compared to the 75 nm ones. Han et al. [843] prepared their Mn-doped GaN nanowires by CVD growth utilizing the reaction of Ga/GaN/MnCl2 with NH3. The diameter of the resultant single-crystalline nanorods was about 50 nm with wurtzite structure and grown along the [0 0 0 1] c-direction. No other phase or clusters was detected by highresolution TEM analysis. The electron energy loss spectroscopy analysis as well as energy dispersive spectroscopy revealed homogenous doping with Mn over the entire nanowire with about 5 at.% of Mn. As repeated on many occasions, these techniques cannot unequivocally rule out the formation of magnetic phases that might be embedded in the diluted magnetic semiconductors as they either do not have the spatial resolution needed or they cannot probe the entire sample. The SQUID magnetometers are very sensitive and a minute amount of magnetic material inclusive of contamination from even tools used to handle the sample could give rise to hysteresis. Continuing on, Han et al. [843] observed hysteresis in magnetization curves at 5 and 300 K with the implication of room-temperature magnetization. In addition, the authors plotted the DM ¼ MFC MZFC curve (using the temperature-dependent FC and ZFC measurements), which represents the difference between magnetization measured while cooling in zero field or cooling under H ¼ 100 Oe and that measured while the temperature is increased; DM dropped to zero at about 300 K. This was used to proclaim ferromagnetism at room temperature, which must yet be scrutinized. Han et al. also were able to perform magnetotransport measurements on individual nanowires in the temperature range 5–300 K, showing the negative MR at temperatures below 150 K.
4.9.11 Applications of Ferromagnetism and Representative Devices
The premise of spin-based electronic devices is predicated on transferring magnetic information from one part of the device to another by using extremely small magnetic elements to encode it onto the itinerant electron spin channels and subsequently reading it off. This coding may be changed by remagnetizing the medium comprising the small magnetic elements and thus enabling the creation of electronic components, the characteristics of which could be modified to respond to applied magnetic fields. The focus of these pages of course is centered
j1167
1168
j 4 Extended and Point Defects, Doping, and Magnetism on semiconductor media made magnetic by doping with magnetic ions such as elements with spin-unpaired electrons found in transition metals and rare earth elements. The use of spin and relevant properties mainly for information storage in media composed of ferromagnetic metals has been the topic of a tutorial review article [642]. The topic has been reviewed amply [631–633], and following a brief mentioning of some of the important device parameters, a few representative device-like concepts are discussed. Briefly, to utilize spin for electronic and optoelectronic applications, one must be able to inject or generate spin-polarized electrons. Second, the spin coherence times must be sufficiently long in relation to device dimensions so that spin coherence would only be perturbed mainly by external controls provided, such as a voltage applied to a gate. Spin-polarized electrons can in principle be injected from a magnetic medium, such as a Schottky barrier made of ferromagnetic metal or diluted magnetic semiconductor or oxide with proper precautions. Spin-polarized electrons can also be produced in a medium by optical excitation with carefully chosen polarized light, as discussed in Section 4.9.6.6.1 dealing with magnetic circular dichroism. If the aforementioned approaches turn out successfully, devices based on spin tunnel junctions, FET-like devices utilizing spin, optical emitters and detectors sensitive to polarization, and elements for quantum computing may be possible. The realization of high TC ferromagnetic DMS might promote the development of spintronic devices, the premise of which is exploitation of electron spin in addition to its charge, for creating a new class of devices with unprecedented functionality. Devices utilizing spin transport rather than charge transport are likely to have the advantages of faster data processing and operation with much less power consumption [631]. To realize spintronic devices for practical applications, spin-polarized carriers need to be first efficiently injected into semiconductors. Spin injection across a semiconductor-ferromagnetic material interface can be adversely affected by the presence of defects at the interface, resulting in low injection efficiencies [848] Diluted magnetic semiconductors with ferromagnetism above room temperature could pave the way for achieving spin polarization and highly efficient spin injection necessary for driving spintronics technologies. The prospects are exhilarating for device applications utilizing the ferromagnetic properties of transition metal doped wide bandgap GaN and ZnO. As discussed earlier in the theoretical studies of TM-doped GaN, Kronik et al. [696] found that the introduction of Mn, in its ordered ferromagnetic phase, results in the formation of a nearly 100% spin-polarized 1.5 eV wide impurity band, primarily due to hybridization of Mn 3d and N 2p orbitals, which renders the material half metallic and supports the effective-mass transport within it. This unique feature, in conjunction with possibly the above room temperature TC and the inherent compatibility with GaN technology, could make GaMnN a potentially significant material for spininjection applications for modern spintronics devices. The caveat is the unequivocal determination of ferromagnetism in this material, which is lacking. ZnO is also promising in achieving high spin-polarized injection efficiencies and carriers due in part to the fact that 3d transition metals have high solubility (up to 30% for Mn and Co). The addition of ferromagnetism to the already unique optical and electric
4.9 Intentional Doping
properties of GaN and ZnO could result in many novel multifunctional spintronics devices. Spin light-emitting diodes (spin LEDs) and spin tunneling junctions based on the current achievements in research of DMSs have already been reported. For the FETlike devices, although proposals have been around for many decades, experimental demonstration has proved to be difficult. Nonetheless, we discuss the principles of operation of a proposed spin FET. Figure 4.185a shows a schematic diagram of the spin FET proposed by Datta and Das [849] and reviewed in Ref. [632]. A metal analogue was also reported by Johnson [850] who performed a spin injection experiment with thin films using a device that became known as the spin transistor. In this device, carriers are injected into the channel from a spin-polarizing electrode (called the polarizer or source), which can be either a ferromagnetic metal or a dilute magnetic semiconductor, through a channel to be collected by another magnetically polarized electrode (called the analyzer or drain). The current through the transistor depends on the relative orientation of the electron spins and the source and drain magnetic moments, the current being maximum when they all align. In a spin FET, application of a relatively low gate voltage causes an interaction between the electric field and the spin precession of the carriers [851], via the so-called Rashba spin–orbit coupling effect [852,853]. It is equivalent to the action of an effective magnetic field Beff that lies within the plane of the 2D electron gas and is perpendicular to the instantaneous wave vector. This results in spatial modulation of the net spin polarization of the current, which can be controlled by an applied gate voltage. If this is sufficient to render the spin orientation of the carriers out of alignment in relation to that of the drain contact (analyzer), then the current is effectively shut off, and this can occur at much lower biases than is needed to shut off the current in a charge-controlled FET. Figure 4.185c represents the schematic of a proposed spin FET based on the GaN material system. Challenges to fabricate a spin FET in the GaN-based system include efficient spin-polarized injection, transport, control and manipulation, and detection of spin polarization as well as spin-polarized current. Devices are required to generate coherent spin injection, with dimensions comparable if not less than the spin coherence length. This, in the case of GaN, means that the channel length must be shorter than some 200 nm. New approaches such as the use of Mn-doped GaN nanotube arrays [854], which have been favorably predicted to enhance the spin polarization or to increase the dephasing length, might have to be brought to bear to realize a spin FET in GaN. Actually, electric field control of spin properties has been demonstrated in GaMnAs. In this system hole interaction is clearly the driving mechanism for ferromagnetism, and therefore if a reverse-biased metal insulator FM semiconductor is employed and made to cause hole depletion, the ferromagnetism can be made to vanish. In contrast, if the metal–insulator–semiconductor junction is forward biased, the depletion depth is reduced and therefore ferromagnetism remains. This is, however, limited to p-type material, which in compound semiconductors is not known for its transport properties and therefore not an attractive material. Ohno et al. [855] reported the demonstration of this concept in an insulated gate (In,Mn)As
j1169
1170
j 4 Extended and Point Defects, Doping, and Magnetism (a)
VG
n Ω
VG,off IDS = 0
k
Drain (spin detector)
Source (spin injector)
(b)
VG
n Ω
VG( = 0),IDS =I ON
k
drain Drain (spin (spin detector) detector)
source Source (spin (spin injector) injector)
(c) GaMnN
Gate
GaMnN
AlN GaN
Figure 4.185 (a) Schematic of the Datta–Daslike spin field effect transistor (SFET). In the case where the channel itself is not ferromagnetic and would not align the electron spin, spin-polarized electrons must be injected and transported all the way to the drain and collected for the drain current to occur. In this case, the source (spin injector) and the drain (spin detector) are ferromagnetic metals or semiconductors, with parallel magnetic moments, and it is assumed that spin-polarized
electrons can be injected with high efficiency. The injected spin-polarized electrons with wave ! vector k move ballistically along a quasi-onedimensional channel formed by, for example, an InGaAs/InAlAs heterojunction in a plane normal to n. Electron spins precess about the ! precession vector W, which arises from spin–orbit coupling and which is defined by the structure and the material properties of the channel. The magnitude of W is tunable by the gate voltage VG at the top of the channel. The
4.9 Intentional Doping
channel structure. Shown in Figure 4.186 are the gated Hall resistance (RHall) data measured at 22.5 K versus the magnetic field with the 125 V (reverse bias for p-type channel), 125 V (forward bias causing hole accumulation), and 0 V bias conditions. The 5-nm-thick (In,Mn)As channel of the insulating-gate field effect transistor (FET) (x ¼ 0.03) was grown on a 10 nm InAs/500 nm (Al,Ga)Sb buffer and a GaAs substrate. A 0.8 mm gate insulator and a metal electrode served to complete the structure. The hole concentration in the channel was estimated to be 5–8 · 1013 cm2 from the change in resistance with the gate voltage and Hall effect at room temperature. In Figure 4.186, the bias of VG ¼ 125 V reduces the sheet hole concentration by 3 · 1012 cm2. The application of a positive gate voltage depletes the channel and reduces the hole-mediated ferromagnetic interaction, which results in a paramagnetic behavior of the magnetization without any hysteresis. A negative gate voltage (125 V), however, causes hole accumulation, which leads to a clear hysteresis as shown in the figure. The magnetization curve resumes its original values as the gate voltage returns to 0 V. At zero gate bias, the channel is weakly ferromagnetic as can be seen from the presence of a small hysteresis. The 125 V swing gives rise to 6% change in the hole concentration and results in a TC change of 4% (1 K). The spin coherence length (dephasing length) is a critical parameter in that the initially spin-polarized electrons are subject to depolarization. The longer the distance between the injecting and collecting electrodes, the more severe is the depolarization. Depolarization strongly depends on material characteristics such as spin–orbit interaction. One benefit to device performance derives from the use of III-Nitride materials because of their small spin–orbit interaction compared to other III–V materials. Selection of the composition of III nitride alloys and of the maximum size of the channel must take into account the appropriate dephasing length, among other factors. To inject (collect) spin-polarized electrons into (from) the transistor channel, ferromagnetic electrodes, either semiconductor or metal varieties whichever could inject spin-polarized electrons efficiently, must be used. GaN-based materials offer certain advantages due to the piezoelectric and pyroelectric phenomena, as do the spin-splitting strained heterojunctions in conventional III–Vs. 3 current is large if the electron spin at the drain points in the initial direction (a); the current is small if the precession period is much larger than the time of flight, leading to a situation where the injected electron spin and detector spin polarization are not aligned or are opposite to one another (b) (patterned after Ref. [632]). In another scenario, the channel is made ferromagnetic by doping the semiconductor with magnetic ions. Under the assumption that ferromagnetism is achieved and device is operating below the Curie temperature,
electrons injected with any common source (spin unpolarized) would be spin polarized due to the ferromagnetic channel and be collected by the spin-polarized drain analyzer. When ferromagnetism is destroyed by some external means, then the spin polarization is not retained and the drain current would reduce. Schematic diagram of a proposed spin FET based on the GaN semiconductor system. InGaN can be substituted for GaN if the Rashba coupling is too small in GaN (c).
j1171
1172
j 4 Extended and Point Defects, Doping, and Magnetism VG < 0 R (kΩ) Hall
0.04
1
1.5 K 5K 10 K
0
20 K
0.02
RHall (kΩ)
–1 –0.5
0.0 B(T)
VG > 0
0.5
0.00
22.5 K
VG = 0V
–0.02
+125 V –125 V 0V –0.04 –1.0
–0.5
0.0
0.5
1.0
B (mT) Figure 4.186 Hall resistance RHall of an insulated gate (In,Mn)As field effect transistor at 22.5 K as a function of the magnetic field for three different gate voltages. RHall is proportional to the magnetization of the (In, Mn)As channel. Upper right inset shows the
temperature dependence of RHall. Left inset shows schematically the gate voltage control of the hole concentration and the corresponding change of the magnetic phase. Courtesy of Ohno et al. [855]. (Please find a color version of this figure on the color tables.)
Because the only attractive GaN FET is that based on two-dimensional electron gas (2DEG), a spin FET based on GaN could also take advantage of transport along the heterointerface. The proximity effects of the exchange splitting in a ferromagnetic metal on a 2DEG in a semiconductor with implications to FET-like devices have been reported [856]. The spin-dependent energy and lifetime in the 2DEG lead to a marked spin splitting in the in-plane current. The planar transport gives way to current leakage into the ferromagnetic layer through the interface, which leads to a competition between drift and diffusion, providing a basis for modulation. The transport theory of a proposed spin valve consisting of a FET with two ferromagnetic gates has been used to demonstrate the effect. The polarization charge at the GaN/AlxGa1xN interface provides an attractive avenue toward optimizing the spin splitting of electrons and holes and thus the performance of the spin-FET. Not much if any is known about the Rashba coupling strength in GaN or InGaN. The zero-field spin splitting in the nitride semiconductor
4.9 Intentional Doping
2DEG can be measured using the beating patterns in Shubnikov–de Haas oscillations [857]. The spin splitting is the combined effect of Rashba [852] and Dresselhaus [858] interactions. The former is due to structural inversion asymmetry and can be varied by an external gate voltage [859–862] while the latter is due to bulk crystalline inversion asymmetry and is insensitive to gate voltage. A spin valve structure can be used to measure the coherence length. At the time of writing this chapter, there was no device demonstration in GaN. However, electric-field control of ferromagnetism has been demonstrated in a (In,Mn)As-based FET at reduced temperatures [855], and optically controlled ferromagnetism has been reported in GaAs with Fe particles. In addition to switching devices, optical devices such as LEDs featuring, that is, ferromagnetic anode (p-type layer which is natural for transition element doping), are of interest because of polarized light emission [863] with applications to optical switching and optical communication with increased bandwidth. What is more is the modulation of the polarization of emitted light by application of an external magnetic field [864,865]. Spin LED allows modulating the polarization of the light emitted by the spin LED by application of an external magnetic field, which is of interest to optical switching and optical communication with increased bandwidth. Spinpolarized light emission has been demonstrated in II–VI semiconductors [866], III–V GaAs [770], and GaN recombination region but with ZnSe(Be:MN) spin aligner [866]. To give the reader a synopsis of the operation of spin LED, the GaAs device is discussed here, particularly because the GaN varieties are not yet available and there are some fundamental issues whether one circular polarization over the other is promoted. The sample structure of a GaAs-based spin LED is depicted in Figure 4.187. The structure contains a p-type ferromagnetic semiconductor (Ga,Mn) As and n-type nonmagnetic semiconductor GaAs, which were epitaxially grown by MBE. The use of ferromagnetic (Ga,Mn)As allows one to inject spins in the absence of magnetic field. Spontaneous magnetization develops below the Curie temperature in the ferromagnetic p-type (Ga,Mn)As semiconductor, indicated by the arrows in the (Ga,Mn)As layer. Under forward bias, spin-polarized holes are injected from the (Ga, Mn)As side into the nonmagnetic region and recombine with spin-unpolarized electrons injected from the n-type GaAs substrate in a nonmagnetic quantum well made of (In,Ga)As (hatched region), as shown in Figure 4.187a, through a spacer layer with thickness d, producing polarized electroluminescence [633]. Injection of spin-polarized carriers through a spacer layer enhances the injection efficiency of spin-polarized carriers. For perfectly spin-polarized carriers (i.e., 100%), the extent of optical polarization is 50%, as Popt ¼ (sþ s)/(sþ þ s) where sþ and s represent the polarization of the HH and LH transitions, respectively. Due to the crystal symmetry of GaAs, injected spin-polarized electrons have been detected in the form of circularly polarized light emission from the quantum well, which corresponds to the magneto-optical Kerr effect loops and can be directly related to spin-polarization. The presence of spin polarization has been confirmed by measuring the polarization of the emitted light. Figure 4.187b shows the observed hysteresis in the degree of polarization of the emitted light at temperatures T ¼ 6–52 K.
j1173
1174
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.187 Injection of spin-polarized holes into a light-emitting p–n diode using a ferromagnetic semiconductor (Ga,Mn)As. (a) Sample structure. Spin-polarized holes hþ travel through the nonmagnetic GaAs and recombine with spin-unpolarized electrons in the (In,Ga)As quantum well. I represents the current, and sþ represents circularly polarized light emitted from the edge of the quantum well. (b) Dependence of the polarization DP of the emitted light on the magnetic field B at temperatures of 6, 31, 52 K, the latter above the Curie value. The solid and hollow symbols represent the degree of polarization when the magnetic field is swept in the positive and
negative directions, respectively. The magnetic field was applied parallel to the surface along the easy axis of magnetization of the (Ga,Mn) As. The temperature dependence of the residual magnetization M in (Ga,Mn)As, where the degree of polarization of the zero magnetic field seen in the emitted light exhibits the same temperature dependence as the magnetization (not shown). (c) Dependence on temperature for B ¼ 0 of the change in the relative remanent polarization, DP, (hollow circles) and magnetic moment measured by a SQUID magnetometer (solid circles). Courtesy of Ohno and coworkers [770]. (Please find a color version of this figure on the color tables.)
4.9 Intentional Doping
∆P 1.00
T=6K
Magnetization
0.75
4 0.50
31 K
2
0.25
52 K 0.00 0
20
40
60
Temperature (K) (c) Figure 4.187 (Continued )
80
100
Magnetization (10-5emu)
Relative polarization,∆P (%)
6 16 K
j1175
1176
j 4 Extended and Point Defects, Doping, and Magnetism The polarized light emission concept has been extended to QD LED heterostructures [867], albeit in the relatively well-established GaAs system, grown by MBE on Sidoped nþ (0 0 1) GaAs substrate. The obvious choice for the QD over QW is due to expected longer spin relaxation times and smaller interband ground-state transition energy in (In,Ga)As/GaAs QD than the bandgap of ferromagnetic (Ga,Mn)As. In this (In,Ga)As/GaAs/(Ga,Mn)As heterostructure, (Ga,Mn)As DMS layer acts as a spin injector. Experimental method employed for producing the structure, however, is such that it is not clear if and why the quantum dot regions are only ones that would be doped with magnetic ions. Assuming this was the case, let us continue. A GaAs spacer layer was grown on an undoped active region, consisting of four coupled (In, Ga)As self-organized QD layers. The collective polarization in (Ga,Mn)As ferromagnetic DMS layer provides a large magnetic moment. Hence, the DMS provides localized magnetic moment and holes. The ferromagnetic behavior of (Ga,Mn)As in the heterostructure is shown in Figure 4.188. The magnetization exhibits the characteristic hysteresis behavior with a saturation field of 150 Oe at 8 K. However, the temperature-dependent magnetization shows a Curie temperature, Tc 44 K. All the holes scatter to the lower Zeeman levels, yielding a high percentage of spinpolarized holes. These spin-polarized carriers recombine in the QD recombination region to produce circularly polarized photons. Among many other measurements, electroluminescence turns out to be one of the popular experiments to reveal the properties of recombination of spin-polarized holes from DMS aligner with electrons in QD or QW. For (In,Ga)As/GaAs/ (Ga,Mn)As heterostructure [867], EL measurements were carried out at 5.1 K in an applied magnetic field of 1000 Oe. The unpolarized EL shows a peak at 1.27 eV with FWHM 67 meV. The light output of the LED was focused onto a zero-order quarterwave plate. The zero-order quarter-wave plate basically converts the circularly polarized output light into horizontal and vertical linearly polarized light, which is detected using an appropriate photomultiplier tube through scanning spectrometer. One such spectra of the circularly polarized light (sþ and s) output is shown in Figure 4.189 from the above DMS heterostructure. The spin degenerate conduction and valence band states in the QDs and the corresponding radiative transitions are shown in the inset according to the selection rule for this structure. The heavy-hole (sþ) and light-hole (s) transitions are polarized in opposite directions. It may also be noted that the transition probabilities of the sþ transitions are three times larger than that of the s transitions. The spin injection efficiency was calculated to be 36% at 5.1 K. Several other groups [746,868] have measured the EL and magnetic properties of (Ga,Mn)As-based heterostructures. Intuitively, GaN has a smaller lattice constant than conventional III–Vs such as GaAs; thus, the magnetic coupling strength is expected to be higher, which also would result in a higher Curie temperature, see Figure 4.143. However, optical experiments performed on GaN/InGaN quantum wells [792] show that the electron spin relaxation time in these structures is shorter than in bulk GaN as well as in GaAs/InGaAs quantum wells. The approach requiring the least semiconductor technology for achieving spin LED in GaN would be to implant Mn into the top contact p-GaN layer of the standard GaN/InGaN LED. Polyakov et al. [869] reported such spin LED
4.9 Intentional Doping
j1177
10 0.4 µm Ga0.974M0.026As Ts = 210 o C TCurie = 44K B = 0.1 T
8
6 M (emu cm—3)
Magnetization (emu cm–3)
15 10 4 5
–300
–200
–100
0
100
200 300 H (Gauss)
–5 2 —10 –15 0 0
100
200
300
Temperature (K) Figure 4.188 Magnetization versus temperature data for the (Ga,Mn)As spin aligner layer. Inset shows the characteristic hysteresis behavior for the (Ga,Mn)As layer at 8 K [867].
structure without polarized light emission. Although, the device produced electroluminescence of low intensity, much higher series resistance and turn-on voltage than those of ordinary LED, due to the difficulty in annealing all radiation damage, has been cited as the bottleneck. The same group studied the inverted spin LED structure grown by MBE with n-type GaMnN on the top also [870]. However, this approach was found not to be successful possibly due to the higher series resistance caused by the lower lateral conductivity of p-GaN inherent to these inverted structures. It should be mentioned that difficulties in obtaining high quality layers on top of a very heavily doped p-GaN are not trivial. The parasitic diode formed with the lower p-GaN layer has been reported as being detrimental to device performance [871]. Furthermore, the reported ferromagnetism in GaMnN was found to be unstable against the type of high temperature anneals used to minimize the contact resistance.
400
j 4 Extended and Point Defects, Doping, and Magnetism Similar spin LED structures [872] have also been reported, albeit without spinpolarized light emission, wherein the nonmagnetic parts of the device was grown by OMVPE and the magnetic component that is limited to the top layer was grown by MBE techniques. It contains four main regions, namely, (1) an intended spin injector of a 120 nm thick n-Ga0.97Mn0.03N layer, (2) a spin detector made up of five periods of nonmagnetic In0.4Ga0.6N quantum wells (3 nm) separated by Si-doped GaN barriers (10 nm each), (3) a nonmagnetic Si-doped GaN spacer (20 nm) (inserted between the aforementioned two regions to avoid direct overlap of electron and hole wave functions from the two regions and thus to ensure a predominant role of the spin injection as the source of spin polarization as monitored in the InGaN MQW spin detector), and (4) a 2-mm-thick layer of Mg-doped p-type GaN for electrical injection of holes into the spin detector. The structure was grown on sapphire substrates starting with a 2 mm thick undoped semi-insulating GaN and also has a top layer of n-type Si-doped GaN (100 nm). The structure shows none or very low optical (spin) polarization at zero field or 5 T, respectively. The weak polarization observed is attributed to the intrinsic optical polarization of the InGaN MQW associated with population distribution between spin sublevels at low temperatures. The device shows loss of efficiency of
B = 0.1 T T = 5.1 K d = 100 nm
σ+
2.0
1.5 EL intensity (au)
1178
σ1.0
0.5
0.0 1.1
1.2
1.3 Energy (eV)
Figure 4.189 Electroluminescence spectra for the right (sþ) and left (s) circularly polarized light measured at 5.1 K with an applied magnetic field of 1000 Oe. The inset shows the spin degenerate conduction and valence band states and the corresponding radiative transitions. See Figure 4.155 for the origin of the right sþ and left s circularly polarized light in GaAs.
1.4
1.5
4.9 Intentional Doping
spin polarization generated by optical spin orientation or electrical spin injection. The I–V characteristics also suffer from a parasitic junction between the (Ga,Mn)N and the n-GaN in the top contact layer due to low conductivity of the p-type layer. Transient and circularly polarized PL analyses of the spin injection dynamics [873] of a GaMnN/ InGaN MQW LED structure suggest that the spin loss is most likely due to fast spin relaxation within the InGaN MQW which destroys any spin polarization generated by either optical or electrical spin injection. In the wurtzite III–nitride system, biaxial strain at the interfaces of heterostructures gives rise to large piezoelectric fields directed along the growth axis that ultimately breaks the reflection symmetry of confining potential, leading to the presence of large Rashba term in the conduction band Hamiltonian. It is this effect that is assumed to adversely contribute to the short spin relaxation times. The adoption of additional stressor layers or even cubic phase of nitrides has been suggested as a cure for larger spin splitting. While the spin-FETs are sought and spin LEDs have been demonstrated, spintronics-based ultrasensitive magnetic sensors would have great impact in magnetic recording, the development of nonvolatile memory, and would greatly increase the magnetic recording density of hard disk drives. Among the devices that have been well developed and that utilize spin is the spin valve, shown in Figure 4.190a in the vertical geometry [642]. The structure utilizes a normal metal straddled by two ferromagnetic metals on either end. When the spins in ferromagnetic metals on either end are aligned parallel to each other, the system is in the low resistance state (top figure). When, on the contrary, the spin of the FM metal on the right is flipped by a magnetic field making the spins of the ferromagnetic metals antiparallel, a highresistance state is attained. A parallel transport configuration of the same concept is shown in Figure 4.190b that requires again two ferromagnetic materials separated from one another by a nonmagnetic conductor whose conductivity is not dominant in the parallel stack. When the magnetic moment of two ferromagnetic layers are antialigned, spin-polarized electrons experience high resistance due to the high interfacial scattering and channeling of the current into narrow pathways. When the spins in the upper and lower films become aligned, the device resistance decreases. To be able to control the spin of one layer only, one material is made hard and the other soft. Because the magnetic moment of one of the ferromagnetic layers is very difficult to be reversed in an applied magnetic field (the hard one), and the moment of the other layer is very easy to be reversed (soft layer), the soft layer then acts as the control valve and is sensitive to manipulation by an external field. The significant change in resistance in response to relatively small magnetic field in layered magnetic thin-film structures is called giant magnetoresistance effect, and is realized by using metallic layers or tunnel junctions in practical applications. In the vertical geometry, using a semiconductor resonant-tunneling diode in conjunction with a ferromagnetic emitter, large tunnel magnetoresistance has been observed in (Ga,Mn)As-AlAs-(Ga,Mn)As tri-layer structures (see Figure 4.191 for the layered structure, and the simplified in-plane band diagram, and the expected I–V characteristics) [874–876]. The I–V characteristics, particularly the derivative of the current with respect to voltage, consistent with resonance states associated with spin split (due to the sp–d exchange interaction) state have been observed. The structure
j1179
1180
j 4 Extended and Point Defects, Doping, and Magnetism
Figure 4.190 (a) Schematic representation of a spin valve, a normal metal straddled by two ferromagnetic metals. When the spins in ferromagnetic metals on either end are aligned parallel to each other, the system is in the lowresistance state top. When, for example, the spin of the FM metal on the right is flipped by a magnetic field, making the spins of the
ferromagnetic metals antiparallel, a high resistance state is attained. (b) Schematic representation of transport that is parallel to the plane of a layered magnetic metal sandwich structure for antialigned (upper figure – high resistance) and aligned (lower figure – low resistance) orientations. (Please find a color version of this figure on the color tables.)
studied by Ohno et al. [874] (see Figure 4.192a) consists of (from the surface that on the right side, down) 150 nm thick (Ga0.965Mn0.035)As, 15 nm undoped GaAs spacer, 5 nm undoped AlAs barrier, 5 nm undoped GaAs quantum well, 5 nm undoped AlAs barrier, 5 nm undoped GaAs spacer, 150 nm Be-doped GaAs (p ¼ 5 · 1017 cm3), 150 nm Be-doped GaAs (p ¼ 5 · 1018 cm3), and p1 GaAs substrate. All of the layers
GaAs: Be
AlAs: (i)
GaAs: (i)
GaAs: (i)
AlAs: (i)
4.9 Intentional Doping
GaAs: (i)
HH1 LH1 HH2
(Ga,Mn)As
HH3 LH2
HH4 (a)
E
E I
k//
V k//
0 (b)
GaAs quantum well
0 (Ga,Mn)As emitter
Figure 4.191 Schematic valence band diagram of a resonanttunneling diode structure (a), simplified diagram of energy versus wave vector parallel to the interface for the GaN quantum well and (Ga,Mn)As emitter, and resulting I–V curve by spin splitting of the valence band of (Ga,Mn)As emitter (b). Courtesy of Ohno and coworkers [874].
were grown at 650 C with the exception of the last (Ga,Mn)As layer, which was grown at 250 C. The derivative I/V characteristics obtained at 6 K indicated peaks associated with HH1, LH1, HH2, HH3, LH2, and HH4 resonant states. Spontaneous magnetization in ferromagnetic semiconductors gives rise to spin splitting of the conduction and valence bands due to the presence of exchange interaction. The magnitude of the splitting between the resonant states in the investigated structure was shown to be proportional to Ms (saturation magnetization) calculated from the Brillouin function, lending credence to the supposition that the origin of peak splitting is indeed the spin
j1181
1182
j 4 Extended and Point Defects, Doping, and Magnetism splitting in the valence band of ferromagnetic (Ga,Mn)As. One can therefore conclude that the splitting observed in the I–V curves is due to the spin splitting of the valence band associated with the development of spontaneous magnetization in (Ga,Mn)As. In the stepped part where the magnetization is antiparallel, the tunneling resistance increases. When the AlAs film that constitutes the tunneling barrier was made thin (1.6 nm), a TMR ratio over 70% was obtained. In this vein, a spin valve transistor with an epitaxial ferromagnetic Fe/Au/Fe(0 0 1) base has been integrated into n-GaAs [877,878]. Parallel transport in trilayer structures, such as the one discussed above, is characterized by the MR ratio (RH R0)/R0, where R0 is the layer resistance in the absence of the external magnetic field, RH (H ¼ 0). For parallel magnetizations M, the structure consists of two (Ga,Mn)As layers and an AlN barrier separating the two Ga: MnAs layer with some thin GaAs spacer layers on either side (Figure 4.192a). The structure was grown with molecular beam epitaxy with the Be-doped buffer layer grown at a typical MBE temperature of 580 C while the Mn-doped layers utilized a low temperature of 250 C. With a moving shutter during the AlAs barrier growth, a wedge shape was achieved allowing the characterization of TMR as a function of barrier thickness [879]. Undoped GaAs spacer layers that were 1 nm thick were inserted on either side of the AlAs tunnel barrier to avoid Mn diffusion/incorporation into the AlAs barrier. Circular 200 mm device structures in which the AlAs thickness ranged between 1.3 and 2.8 nm, owing to the wedge-shaped growth of AlAs with the aid of a moving shutter in front of the substrate. Tunnel magnetoresistance effect in the aforementioned (Ga,Mn)As–AlAs–(Ga,Mn)As trilayer structure with an AlAs barrier thickness of 1.6 nm, measured at a temperature of 8 K, is shown in Figure 4.192b [879], when the magnetic field is parallel to the [0 0 1] axis in the plane. The measurements were performed at a bias of 1 mV, which obviates any possibility of hot carrier effects. The resistance for H ¼ 0 is slightly below 0.015 O cm2. The TMR values are 72% for H ¼ 110–120 Oe (solid curve) and 120–130 Oe when the magnetization of the Ga1xMnxAs becomes antiparallel. The minor loop represented by the thin line is indicative of the fact that the antiparallel as well as the parallel configuration is stable. The tunnel junction resistance measured at 8 K exhibited an exponential dependence on the barrier thickness ranging from the low 103 O cm2 for a barrier thickness of 1.4 nm to about 101 O cm2 for a barrier thickness of 2.8 nm. The barrier thickness dependence of TMR measured at 8 K for magnetic field parallel to [1 0 0] and ½1 1 0 axes is shown in Figure 4.193. A maximum value of 75% was attained for an AlAs barrier width of 1.46 nm when the field is along the [1 0 0] direction. However, the maximum was only about 30% when the magnetic field is parallel to the ½110 axis – again for the same barrier thickness of 1.46 nm. This dispersion is due to the cubic magnetocrystalline anisotropy induced by the zinc blende-type Ga1 xMnxAs crystalline structure, where its easy magnetization axis is h1 0 0i [880]. Above the barrier thickness of 1.46 nm, TMR drops rapidly for both orientations of the magnetic field. The drop in TMR below barrier thickness of 1.46 nm is not clearly understood but could be associated with the ferromagnetic exchange coupling between the two Ga1xMnxAs layers on either side of the barrier because of the thinness of the barrier. Moreover, the TMR decreased with increasing measurement
4.9 Intentional Doping
(Ga0. M 96 n0. )A 04 s(50nm GaAs(1n
j1183
)
m)
AlAs GaAs (1 nm) (Ga0.967Mn0.033)As(50 nm) Be (p) GaAs (100 nm) p+-GaAs substrate (0 0 1)
(a)
60 40
0.020
TMR (%)
Resistance (Ω cm–2)
80
Minor loop
0.025
20 0.015
0 –200
–100
0
100
200
Magneticfield, H//[0 0 1] (Oe) (b) Figure 4.192 (a) Schematic illustration of a wedge-type ferromagnetic semiconductor trilayer heterostructure sample grown by LTMBE. (b) Tunnel magnetoresistance effect curves obtained at 8 K for a Ga1xMnxAs (x ¼ 4.0%, 50 nm)/AlAs (1.6 nm) Ga1xMnxAs (x ¼ 3.3%, 50 nm) trilayer tunnel junction of 200 mm in diameter. The bold solid and dotted curves were obtained while sweeping the magnetic field from positive to negative and from negative to positive directions, respectively (major loop). A minor loop is shown by the relatively thin line. The (Ga,Mn)As
layers are 50 nm thick, and the Mn composition in these layers are 0.04 and 0.033. The AlAs layer is 1.6 nm thick. Because the easy axis of magnetization lies within the plane of the sample, a magnetic field is applied parallel to the sample surface. When a magnetic field is applied along the [1 0 0] direction, a tunneling magnetoresistance effect of over 70% is observed, and when a magnetic field is applied along the [1 1 0] direction, this effect is approximately 30%, as detailed in Figure 4.193 [879].
j 4 Extended and Point Defects, Doping, and Magnetism 100 8K H//[1 0 0]
80
TMR (%)
1184
60
40 H//[1 1 0] 20
0 1.4
1.6
1.8
2.0
2.2
Barrier thickness (nm) Figure 4.193 The TMR values in Ga1xMnxAs (x ¼ 4.0%, 50 nm/ AlAs (dAlAs/Ga1xMnxAs (x ¼ 3.3%, 50 nm) tunnel junctions measured at 8 K versus the AlAs barrier thickness for magnetic field parallel to the [1 0 0] and ½1 1 0 axes [879].
temperature vanishing completely at 50 K, which corresponds to the Curie temperature. It should be pointed out that spin injector layer and medium into which the spinpolarized electrons are injected must match in terms of the ratio of the spin diffusion length and the conductivity to obtain sufficiently high spin-polarized electron injection and thus large TMR ratios. Semiconductor-to-semiconductor, oxide-to-oxide, and metal-to-metal varieties satisfy this requirement. However, this is not so in ferromagnetic metal to semiconductor varieties. To attain a reasonably matched case, a tunneling insulating barrier is typically inserted between the metal and the semiconductor. The spin precession length is a critical parameter for such a spin FET to function as intended. The initially spin-polarized electrons are subject to depolarization with distance, and the depolarization strongly depends on the material characteristics such as the spin–orbit interaction. On the contrary, the use of III nitrides may be beneficial because of possible longer spin coherence times, owing to their weak spin–orbit interaction compared to the more conventional III–V materials. Besides, due to the pronounced piezoelectric and pyroelectric effects in GaN-based materials, the spin splitting of electrons and holes in GaN/AlxGa1xN heterojunctions may be optimized. On the contrary, weak spin–orbit interaction leads to relatively small Rashba coupling. Spin–orbit interaction and the associated spin splitting in zinc blende III–V semiconductor heterostructures have been studied for more than a decade and are relatively well understood. Approaches used for the InGaAs system [859] wherein beating of two close frequencies due to spin splitting Shubnikov–de Haas oscilla-
4.9 Intentional Doping
tions [881], although may not be unambiguous [882,883], can potentially be used for GaN also in determining the spin–orbit interaction parameter. The intrinsic lack of inversion symmetry gives rise to the Dresselhaus [884] interaction. However, the structurally introduced inversion asymmetry such as heterostructures appears [851] to make the Rashba interaction [852,853] more plausible. As mentioned, a spin valve structure, for example, in the form of a spin valve transistor, can also be used to measure the coherence length (A spin valve – a term coined by IBM – is in general a structure that consists of a dedicated GMR trilayer in which the electrical resistance is high or low, depending on the direction rather than the strength of the magnetic field. The name may be somewhat misleading in that there is one fully open or closed valve. Instead, the change in resistance is typical in the range of 5–10%. A spin valve is typically made of only two ferromagnetic layers spaced by a layer of nonmagnetic metal. Contrary to a GMR multiplayer, the two ferromagnetic layers are magnetically decoupled. This is achieved by increasing the thickness of the spacer layer. As a further difference the magnetization of one of the ferromagnetic layers is spatially fixed (pinned) by an antiferromagnetic bottom layer. Thus, it is called the pinned layer, which is magnetically hard, and the other is called the free layer because it should easily follow the external magnetic field and is made of magnetically soft material, meaning it is very sensitive to small fields. The other, on the contrary, is made magnetically hard by various schemes – meaning it is insensitive to fields of moderate size. As the soft layer moves due to applied fields, the resistance of the entire structure changes. The analysis of weak antilocalization (WAL) [885] is much simpler to implement and determine the spin–orbit interaction parameter, as has been done for AlGaAs/GaAs/AlGaAs [886], AlSb(ZnTe)/InAs/AlSb [887], and InGaAs quantum wells with two-dimensional electron gas. As for the wurtzite AlGaN/GaN heterostructures, experiments based on Shubnikov–de Haas (SdH) [888,889], weak antilocalization [890,891], and circular photogalvanic [892] measurements have given conflicting results for the spin splitting. In particular, spin-splitting energies extracted from the beat pattern of SdH measurements are found to be as large as 9 meV, which is about an order of magnitude larger than the theoretical estimates based on the Rashba coupling mechanism for this material system [851]. To account for the discrepancy, Lo et al. [893] have proposed an additional spin-splitting mechanism for wurtzite structures and Tang et al. [894] proposed an alternative interpretation of such data based on magneto-intersubband scattering. In the context of weak antilocalization, Thillosen et al. [890] utilized the weak antilocalization effect to determine the dephasing time tf and the spin–orbit scattering time tSO (1.25 ps) in GaN with the aid of magnetoresistance measurement at very small magnetic fields (in the range of mT) in a two-dimensional electron gas system. The extent of the Rashba spin splitting is typically represented by the spin–orbit coupling (interaction) parameter a. Relying on the notion that the spin–orbit interaction is to a first extent due the asymmetry in the confining potential introduced by the heterostructure [851] and thus by the Rasha effect and with the help pffiffiffiffiffiffiffiffiffiffiffi of [895] aSO ¼ h=kF tSO tq (where tq is the quantum scattering time associated with cyclotron motion), Thillosen et al. [890] obtained a value of aSO 8.54 · 1013 eV m
j1185
1186
j 4 Extended and Point Defects, Doping, and Magnetism that compares with 6 · 1013 eVm obtained by Schumult et al. [891] The corresponding spin precession length lSO ¼ h2 =2am is approximately 200 nm, which is relatively short due in part to the large effective mass. The short value of lSO would indeed provide a tremendous challenge for device fabrication in terms of device lengths that can be used. The spin–orbit interaction parameter a is proportional to the average vertical field in the heterostructure a ¼ b<E>, where the b coefficient is inversely proportional to the effective mass and bandgap of the material [896]. Kurdak et al. [897] have also used weak antilocalization to determine the electron spin-splitting energies in the range of 0.3–0.7 meV (depending on the sheet electron concentration). This compares with DSO 0.3 meV at ne ¼ 1 · 1012 cm2 obtained by Schumult et al. [891] Consistent with such small spin-splitting energies, no beat feature in the SdH oscillations were observed. More importantly, as predicted by the Rashba mechanism, the measured spin-splitting energies were found to scale linearly with the Fermi wave vector. The spin–orbit times varied from 0.74 to 8.24 ps for sheet carrier densities of 5 · 1012 cm2 and slightly under 2 · 1012 cm2, respectively. The spin dephasing time varied between about 40 ps (at 1.8 K) and about 10 ps (at 4 K). Other potential spintronic applications include integration of magneto-optical effect in semiconductor optical isolators by using the Faraday effect to control the spin relaxation time and coherence for optical switches and quantum information processing. It has been shown that electron spin coherence is maintained for periods of at least a few nanoseconds in GaN [898], which raises the possibility of applications of confined electronic states to spin memory and manipulation. 4.9.12 Summarizing Comments on Ferromagnetism
A summary of the state of the possible DMS in GaN doped with magnetic ions is now warranted. With certainty the potential of room-temperature ferromagnetism has rejuvenated the field of dilute magnetic semiconductors. While the devices are the eventual driving force, the importance of understanding and producing ferromagneticdiluted semiconductor materials with an above room temperature Curie temperature, not to mention the junction temperature that is typically well above room temperature, cannot be overstated. The efforts on this topic could also potentially provide an opportunity for mainstream semiconductor device researchers to gain some insight into the world of magnetism, seen as a first and critical step toward realization of devices. Due to small spin–orbit coupling and small interatomic spacing, in addition to more intricate reasons, GaN and ZnO could potentially exhibit ferromagnetism when doped with ions having unpaired partially empty 3d (transition metal) and 4f (rare earth) shells. Intuitively, the smaller lattice constant of GaN and ZnO, compared to the conventional II–VI and III–V semiconductors, would serve to increase the magnetic coupling strength among the magnetic ions through indirect exchange interaction and lead to high Curie temperatures, as shown in Figure 4.143. However, the caveat is that, at least in the modified mean free field theory of Dietl, TM concentrations near 5% or higher and acceptor concentrations above 1020 cm3 are needed to observe above room temperature Curie temperatures. We must hasten to
4.9 Intentional Doping
mention that the mean field theory might overestimate the Curie temperature, particularly for low magnetic ion concentrations. While the above room temperature Curie temperature aspect of that theory is often cited, the conditions at which that temperature hinges are often left out of the critical discussion. As elaborated repeatedly, on the experimental side it is still not sufficiently clear if the uniformly doped host or some other phase is responsible for the observed magnetic hysteresis persisting in some cases well above room temperature. Moreover, the transition energies induced by transition elements, particularly in GaN, are not yet well understood. It is reasonable to expect that only certain states of the TM element (after they are split by crystal field splitting and Zeeman splitting which is to be enhanced by large spin exchange interaction) contribute to interactions supporting the ferromagnetic state. Data in favor and against ferromagnetism in both GaN and ZnO have been reported. Arguments and counterarguments for or against have been made. At least at the time of this writing, the field still appears too controversial and confusing, as the local environment and energy levels of magnetic ions are too sketchy and mechanism(s) leading to ferromagnetism are too many, not to mention the ever-varying sample quality. In the confines of this chapter, what could be described as a gallant attempt was made to discuss those reports with appropriate comments regarding any inconsistencies and difficulties as far as the mechanisms and experimental methods are concerned. In short, not only are the various reported experiments not consistent, the same also holds for the effort on the theoretical side. On the theoretical side, it is not clear which of the many mechanisms put forth is applicable, to what extent, and in what kind of samples. For example, super direct exchange between magnetic ions is antiferromagnetic in the realm of the Zener theory, but indirect exchange interaction between the magnetic ions can lead to ferromagnetism in the realm of Zener and RKKY theories. In the one-electron system, the d wave functions of the transition metals and s wave functions of the host (conduction electrons) are orthogonal and could not hybridize in a cubic system. However, the hybridization of d wave functions of the transition elements and p-states of the host (valence band) is strong and can lead to ferromagnetism. The picture in wide bandgap semiconductors is rather complex in that the level splitting and associated hybridization are magnetic ion and doping level dependent. Therefore, simple arguments, while instructive, may not strictly apply. The mean field theory, expanded by Dietl to include the anisotropy of the valence band spin–orbit interaction, indicates the hole-mediated mechanism to be dominant. In the realm of this theory, the material under investigation must be convincingly p-type and yet ferromagnetism in n-type samples has been reported. In fact, donor-doped samples along with TM doping have been reported to be ferromagnetic and the constructive role of conduction electrons has been touted. In contrast, doping with elements, which would tend to reduce the background electron concentration, has also been touted to be the mechanism for the experimentally observed ferromagnetism. Moreover, magnetization in electron codoped samples has been attributed to bound polarons, particularly in ZnO. A bound polaron is formed when the magnetic moment of the ion is aligned with the spin of an exciton and the magnetic ion
j1187
1188
j 4 Extended and Point Defects, Doping, and Magnetism within its radius. The exciton in turn renders the spin of nearby excitons to align, which in turn is parallel to the spin of a nearby magnetic ion, and so on. Clearly, the field is in its infancy and the picture is rather complex. It is quite possible that strict theories mentioned above are not necessarily applicable to all the samples. It is very likely that ab initio calculations are required. These calculations would also provide a guiding light for experimentalist to focus on exploring those transition or rare earth elements that are more likely to be ferromagnetic in the particular hosts under investigation. A case in point is Mn in these two hosts. The theory indicates that in lightly and heavily doped samples the ferromagnetic phase is not stabilized. The doping concentrations that lead to a particular class of levels splitting are required for GaN:Mn to be stabilized in the ferromagnetic state. This simply means that more detailed and probing investigations are necessary to sort out the various experimental data that have been reported. As has been the characteristic of scientific matters, more clarity will prevail in time as the recent discussions appear to be going in this direction. In closing the discussion of the potential of ferromagnetism and applications to devices, it is not an overstatement that the expected advantages of GaN- and ZnO-based spintronics are truly exciting although the efforts in materials science and devices are still in their embryonic stage. However, there are many challenges to consider including whether high ferromagnetic transition temperature and carrier-mediated ferromagnetism can be realized. In addition to a ferromagnetic semiconductor, successful operation of spintronic devices also requires the support of spin-polarized transport in the device and the amplification and detection of spin polarization (or spin current). With accelerated research efforts toward room-temperature ferromagnetism in GaN and ZnO, and perhaps others such as oxides-based DMS materials, it might be possible to realize semiconductor magnetoelectronics. Perhaps, the real utilization of the spin degree of freedom in spin-based all-semiconductor multifunctional devices such as spin FETs, spin LEDs, spin RTD, sensitive magnetic sensor, high-density nonvolatile memory, and quantum bits for quantum processing will be a reality even at a fraction of the present of expectation. If so, this would represent a truly new chapter in the annals of semiconductor-based devices.
4.10 Ion Implantation and Diffusion for Doping
Ion implantation and diffusion are industrial doping methods for Si. The reports on III-N doping by ion implantation are somewhat contradictory so far. As compared to classical semiconductors, group III-N semiconductors have larger binding energies of the constituent atoms and smaller interatomic distances. Consequently, substitutional impurity incorporation by ion implantation is intrinsically more difficult. On the contrary, the temperatures needed for annealing the implanted lattice defects are limited to the decomposition temperatures of group III-N compounds, which sometimes are not high enough to anneal out the damage created.
4.10 Ion Implantation and Diffusion for Doping
Similarly, it should come as no surprise that experimental reports on diffusion in group III-N semiconductors are limited. Diffusion of Mg at 800 C for 80 h into an unintentionally doped n-type GaN layer in a sealed nitrogen ampoule resulted in p-type material [899]. A hole concentration of 2 · 1016 cm3 and a mobility of 12 cm2 V1 s1 at room temperature were measured. Lower diffusion temperatures were not successful in converting the sample into p-type. However, higher diffusion temperatures caused the film to decompose and evaporate. High-energy implants (80–100 keV) increased strain and defects in GaN [900]. The damage caused by such high-energy implants could not be annealed out after 30 min at 800 C; no conversion into p-type was detected. However, films implanted with Mg ions at lower energies (40–60 keV) recovered after annealing at 800 C and retained their original lattice parameters, as determined by X-ray diffraction. Hot-probe tests showed conversion into p-type. n-Type and p-type conductivities were accomplished by ion implantation of Siþ (200 keV), Mgþ (180 keV) and Mgþ (180 keV) þ Pþ (250 keV) [900]. For these experiments, undoped layers with background electron concentrations of 1–4 · 1016 cm3 as well as a postimplant thermal anneal procedure in the range 700–1100 C for 10 s were employed. To prevent or otherwise limit nitrogen loss, another GaN film was firmly pressed against the surface of the implanted sample during annealing. Mgþ implantation alone did not produce any doping effect. Mgþ/Pþ coimplantation led to a conversion from n- to p-type conductivity after annealing at 1050–1100 C with an activation percentage of about 62%. The effect of coimplantation here is to increase the vacancy concentration. Siþ implantation resulted in a sharp increase in the n-type conductivity after annealing at 1050–1100 C with an activation percentage of 93%. Results somewhat contradictory to the above reports on GaN were obtained by Wilson et al. [901] who observed that no diffusion of implanted Mg and other impurities (Be, C, Zn, Si, Se, and Ge) was discernible even after an anneal at 800 C for 10 min. Only S showed a marked diffusion at temperatures higher than 600 C. Impinging energies of the ions were as follows: H, 40 keV; Li, 100 keV; Be, 100 keV; C, 260 keV; F, 100 keV; Na, 100 keV; Mg, 100 keV; Si, 150 keV; S, 200 keV; Zn, 300 keV; Ge, 500 keV; Se, 500 keV [902]. The lack of diffusion in the implanted samples could be attributed to the fact that the energetic ions are capable of producing defects that, in turn, can trap implanted impurities, causing them to cluster. The annealing temperatures may not be high enough to anneal out such defects. Spurred by the predictions that Ca might be a shallow acceptor [571], implantation of Ca has been attempted. Implantation of Caþ at 180 keV and coimplantation of Caþ with Pþ at 130 keV followed by rapid thermal annealing at T 1100 C produced p-type doping with an ionization energy of 169 meV for a Ca acceptor level, which is similar to that of Mg [903]. In some device fabrication schemes, selective-area conversion of the sample to high resistivity is very desirable. This serves to eliminate the unwanted current paths in devices such as field effect transistors. To this end, proton implantation has been undertaken and high-resistance samples have been obtained by implanting Hþ or Heþ [904], and Nþ or Fþ [905,906]. As endemic to any aspect of nitride semiconductor research and development, many more investigations are needed for improved results and reproducibility. We should, however, be prepared for disappointment if
j1189
1190
j 4 Extended and Point Defects, Doping, and Magnetism ion implantation does not turn out to be universally relevant in nitride development due to a large bond strength and the high N-vapor pressure in this material. On the contrary, one would like to have strong bond strength for stability, and on the contrary, one would like to have a semiconductor with small bond strength to allow the device engineer to modify its characteristics. Ion implantation for the purpose of doping the material with magnetic ions, when applicable, are discussed in conjunction with the applicable topic.
4.11 Summary
A simple conclusion that can be drawn is that defects in GaN-based semiconductors as in any other, are important. Furthermore, they are made more relevant because of relatively strong coupling between polarization, mechanical properties, and thermal properties in that any strain inhomogeneity caused by defects leads to polarization inhomogeneity that in turn could mitigate additional defect generation particularly under high field and/or temperature operation. Unfortunate for GaN, despite the tremendous advances made, the lack of native substrate in the desired form is synonymous with extended and point defects. Although epitaxial lateral overgrowth can be used to reduce the extended defect concentration and naturally the point defect concentration, the window regions as well as the coalescence boundaries still contain inordinate amount of defects. It is possible in some special device structure to use only the wing regions with very low defect concentration, but this is not necessarily applicable to large-area devices. In addition, point defects and p-type doping are forever intertwined, and the nature of active p-type impurity incorporation is not sufficiently understood. Site selection and defect creation to minimize the free energy will require more effort than that expended so far. The question of whether acceptors in GaN are effective-mass-like impurities still lingers despite the predictions that the thermal activation energy for Be is only 60 meV and that for Zn is a large 330 meV [907]. Experiments are lacking in that Hall measurements have not been possible in Be-doped GaN and that Zn-doped GaN often has high resistivity. The predicted thermal activation energy for Mg is 230 meV [907], which is in the range of experimentally reported figures. Acceptor-like impurity incorporation including codoping in nitride semiconductors represents one of the most pivotal issues. Doping with magnetic ions in the hope of achieving above room-temperature ferromagnetism is met with certain degree of stalemate and controversy. Unlike the case of GaN, magnetic ion doped GaN is not p-type conductive. Therefore, even if ferromagnetism is achieved, its control with hole concentration is precluded. In closing, analysis of defects (extended, point, and complexes) in GaN has seen substantial improvement in a relatively short period and helped the practitioners of GaN devices to enhance their craft to the point of being a strong contender for the lucrative and colossal general lighting.
References
References 1 Stavola, M. (1999) To 40 years of defects in semiconductors: may the problem never be solved! Physica B, 273–274, 1–6. 2 Hiramatsu, K., Amano, H., Akasaki, I., Kato, H., Koide, N. and Manabe, K. (1991) Journal of Crystal Growth, 107, 509. 3 Sitar, Z., Paisley, M.J., Yan, B. and Davis, R.F. (1990) Materials Research Society Symposium Proceedings, 162, 537. 4 Ponce, F.A., Major, J.S., Jr, Plano, W.E. and Welch, D.F. (1994) Applied Physics Letters, 65, 2302. 5 Ponce, F.A., Bour, D.P., G€otz, W., Johnson, N.M., Helava, H.I., Grzegory, I., Jun, J. and Porowski, S. (1995) Applied Physics Letters, 68, 917. 6 Vermaut, P., Ruterana, P., Nouet, G., Salvador, A. and Morkoc, H. (1995) Institute of Physics Conference Series, 146, 289. 7 Cherns, D., Young, W.T. and Ponce, F.A. (1997) Materials for Advanced Technology, 50, 76. 8 Rouviere, J.L., Arlery, M., Bourret, A., Niebuhr, R. and Bachem, K. (1995) Institute of Physics Conference Series, 146, 285. 9 Northrup, J.E., Neugebauer, J. and Romano, L.T. (1996) Physical Review Letters, 77, 103. 10 Tanaka, S., Kern, R.S. and Davis, R.F. (1995) Applied Physics Letters, 66, 37. 11 Xin, Y., Brown, P.D., Humphreys, C.J., Cheng, T.S. and Foxon, C.T. (1997) Domain boundaries in epitaxial wurtzite GaN. Applied Physics Letters, 70, 1308–1310. 12 Ning, X.J., Chien, F.R., Pirouz, P., Wang, J.W. and Khan, M.A. (1996) Journal of Materials Research, 3, 580. 13 Ponce, F.A. (1997) MRS Bulletin, 22, 51. 14 Zheleva, T.S., Nam, O.K., Bremser, M.D. and Davis, R.F. (1997) Applied Physics Letters, 71, 2472. 15 Rouviere, J.L., Arlery, M., Niebuhr, R., Bachem, K.H. and Briot, O. (1997)
16
17
18
19
20
21
22
23 24
25
26
Materials Science & Engineering B: Solid State Materials for Advanced Technology, 43, 161. Ruterana, P. and Nouet, G. (2001) Atomic structure of extended defects in wurtzite epitaxial GaN epitaxial layers. Physica Status Solidi b: Basic Research, 227, 177. Dimitrakopulos, G.P., Kehagias, T., Komninou, P., Nouet, G. and Karakostas, T. (2002) Disconnections at translation domain boundaries in epitaxial GaN. Journal of Physics: Condensed Matter, 14 (48), 12709–12715. Smith, D.J., Chandrasekar, D., Sverdlov, B., Botchkarev, A., Salvador, A. and Morkoc, H. (1995) Applied Physics Letters, 67, 1803. Sverdlov, B.N., Martin, G.A., Morkoc, H. and Smith, D.J. (1995) Applied Physics Letters, 67, 3064. Vermaut, P., Ruterana, P., Nouet, G., Salvador, A. and Morkoc, H. (1997) Philosophical Magazine A, 75, 239. Vermaut, P., Ruterana, P. and Nouet, G. (1997) Philosophical Magazine A, 76, 1215. Blank, H., Delavignette, P., Gevers, R. and Amelinckx, S. (1964) Physica Status Solidi, 7, 747. Drum, C.M. (1965) Philosophical Magazine A, 11, 313. Chadderton, L.T. Fitzgerald, A.F. and Yoffe, A.D. (1963) Philosophical Magazine, 8, 167. Ruterana, P., Sanchez, A.M. and Nouet, G. (2003) Extended defects in wurtzite GaN layers: atomic structure, formation and interaction mechanisms, in Nitride Semiconductors – Handbook on Materials and Devices (eds P. Ruterana, M. Albrecht and J. Neugebauer), Wiley-VCH Verlag GmbH, Weinheim, Germany. Speck, J.S. and Rosner, S.J. (1999) The role of threading dislocations in the physical properties of GaN and its alloys. Physica B, 273–274, 24.
j 1191
1192
j 4 Extended and Point Defects, Doping, and Magnetism 27 Hino, T., Tomiya, S., Miyajima, T., Yanashima, K., Hashimoto, S. and Ikeda, M. (2000) Applied Physics Letters, 76, 3421. 28 Weinman, N.G., Eastman, L.F., Doppalapudi, D., Ng, H.M. and Moustakas, T.D. (1998) Journal of Applied Physics, 83, 3656. 29 Blumenau, A.T., Fall, C.J., Elsner, J., Jones, R., Heggie, M.I. and Frauenheim, T. (2003) A theoretical investigation of dislocations in cubic and hexagonal gallium nitride. Physica Status Solidi c, 0 (6), 1684–1709. 30 Blumenau, A.T., Jones, R., Öberg, S., Briddon, P.R. and Frauenheim, T. (2001) Dislocation related photoluminescence in silicon. Physical Review Letters, 87, 187404. 31 Blumenau, A.T., Heggie, M.I., Fall, C.J., Jones, R. and Frauenheim, T. (2002) Dislocations in diamond: core structures and energies. Physical Review B: Condensed Matter, 65, 205205. 32 Fall, C.J., Blumenau, A.T., Jones, R., Briddon, P.R., Frauenheim, T., GutierrezSosa, A., Bangert, U., Mora, A.E., Steeds, J.W. and Butler, J.E. (2002) Dislocations in diamond: electron energy-loss spectroscopy. Physical Review B: Condensed Matter, 65, 205206. 33 Jones, R. and Briddon, P.R. (1998) Identification of Defects in Semiconductors, Semiconductors and Semimetals, vol. 51a (ed. M. Stavola), Academic Press, Boston, MA, Chapter 6. 34 Briddon, P. and Jones, R. (2000) LDA calculations using a basis of Gaussian orbitals. Physica Status Solidi b: Basic Research, 217, 131. 35 Blase, X., Lin, K., Canning, A., Louie, S.G. and Chrzan, D.C. (2000) Structure and energy of the 90 partial dislocation in diamond: a combined ab initio and elasticity theory analysis. Physical Review Letters, 84, 5780. 36 Mathis, S.K., Ramanov, A.E., Chen, L.F., Beltz, G.E., Pompe, W. and Speck, J.S. (2000) Modeling of threading dislocation reduction in growing GaN films. Physica
37
38
39
40
41
42
43
44
45
Status Solidi a: Applied Research, 179, 125–145. Kehagias, Th., Komminou, Ph., Nouet, G., Ruterana, P. and Karakostas, Th. (2001) Misfit relaxation of AlN/Al2O3 interface. Physical Review B: Condensed Matter, 64 (19), 195329. Dimitrakopulos, G.P., Komninou, Ph. and Karakostas, Th. (2003) Topological analysis of defects in nitride semiconductors, in Nitride Semiconductors – Handbook on Materials and Devices (eds P. Ruterana, M. Albrecht and J. Neugebauer), Wiley-VCH Verlag GmbH, Weinheim, Germany. Mathis, S.K., Ramanov, A.E., Chen, L.F., Beltz, G.E., Pompe, W. and Speck, J.S. (2000) Modeling of threading dislocation reduction in growing GaN films. Physica Status Solidi a: Applied Research, 179, 125–145. Rouviere, J.L., Arlery, M. and Bourret, A. (1997) Institute of Physics Conference Series, 157, 173. Elsner, J., Jones, R., Stich, P.K., Porezag, V.D., Frauenheim, Th., Heggie, M.I., Äberg, S. and Briddon, P.R. (1997) Theory of threading edge and screw dislocations. Physical Review Letters, 79 (9), 3672–3675. Xin, Y., Pennycook, S.J., Browning, N.D., Nellist, P.D., Sivanathan, S., Omnes, F., Beaumont, B., Faurie, J.P. and Gibart, P. (1998) Direct observation of the core structures of threading dislocations in GaN. Applied Physics Letters, 72, 2680–2682. Ruterana, P., Potin, V. and Nouet, G. (1998) Materials Research Society Symposium Proceedings, 482, 435. Wang, D., Ichikawa, M. and Yoshida, S. (2002) Dislocation core structures in GaN grown on Si(1 1 1) substrates. Philosophical Magazine Letters, 82, 119–124. Haugk, M., Elsner, J., Frauenheim, T., Staab, T.E.M., Latham, C.D., Jones, R., Leipner, H.S., Heine, T., Seifert, G. and Sternberg, M. (2000) Physica Status Solidi b: Basic Research, 217, 473.
References 46 Elsner, J., Jones, R., Heggie, M.I., Stich, P.K., Haugk, M., Frauenheim, Th., Öberg, S. and Briddon, P.R. (1998) Deep acceptors trapped at threading-edge dislocations in GaN. Physical Review B: Condensed Matter, 58, 12571. 47 Northrup, J.E. (2001) Screw dislocations in GaN: the Ga-filled core model. Applied Physics Letters, 78 (16), 2288–2290. 48 Nortrup, J.E. (2002) Theory of intrinsic and H-passivated screw dislocations in GaN. Physical Review B: Condensed Matter, 66, 045204. 49 Liliental-Weber, Z., Zakharov, D., Jasinski, J., Washburn, J., OKeefe, M.A. and Morkoc, H. (2002) Screw dislocations in MBE GaN layers grown on top of HVPE layers: are they different? Materials Research Society Symposium Proceedings, 743, 243–248. 50 Liliental-Weber, Z., Zakharov, D., Jasinski, J., OKeefe, M.A. and Morkoc, H. (2004) Screw dislocations in GaN grown by different methods. Microscopy and Microanalysis, 10, 47–54. 51 Arslan, I., Bleloch, A., Stach, E.A. and Browning, N.D. (2005) Atomic and electronic structure of mixed and partial dislocations in GaN. Physical Review Letters, 94, 025504-1–025504-4. 52 Xin, Y., James, E.M., Arslan, I., Sivananthan, S., Browning, N.D., Pennycook, S.J., Omnes, F., Beaumont, B., Faurie, J.P. and Gibart, P. (2000) Direct experimental observation of the local electronic structure at threading dislocations in metalorganic vapor phase epitaxy grown wurtzite GaN thin films. Applied Physics Letters, 76, 466–468. 53 Browning, N.D., Chisholm, M.F. and Pennycook, S.J. (1993) Atomic-resolution chemical analysis using scanningtransmission electron microscope. Nature, 366, 143–146. 54 Northrup, J.E. (1998) Applied Physics Letters, 72, 2316. 55 Vermaut, P., Nouet, G. and Ruterana, P. (1999) Applied Physics Letters, 74, 694–696.
56 Liliental-Weber, Z., Chen, Y., Ruvimov, S. and Washburn, J. (1997) Formation mechanism of nanotubes in GaN. Physical Review Letters, 79, 2835. 57 Liliental-Weber, Z., Washburn, J., Pakula, K. and Baranowki, J. (1997) Microscopy and Microanalysis, 3, 436. 58 Frank, F.C. (1951) Acta Crystallographica, 4, 497. 59 Qian, W., Rohrer, G.S., Skowronski, M., Doverspike, K., Rowland, L.B. and Gaskill, D.K. (1995) Applied Physics Letters, 67, 2284. 60 Vermaut, P., Ruterana, P., Nouet, G. and Morkoc, H. (1996) MRS Internet Journal of Nitride Semiconductor Research, 1, 42. 61 Ruterana, P., Vermaut, P., Potin, V., Nouet, G., Botchkarev, A., Salvador, A. and Morkoc, H. (1997) Materials Science & Engineering B: Solid State Materials for Advanced Technology, 50, 72. 62 Arslan, I. and Browning, N.D. (2003) Physical Review Letters, 91, 1655011–165501-4. 63 Cherns, D., Young, W.T., Steeds, J.W., Ponce, F.A. and Nakamura, S. (1997) Journal of Crystal Growth, 178, 201. 64 Bere, A. and Serra, A. (2002) Atomic structure of dislocation cores in GaN. Physical Review B: Condensed Matter, 65, 205323-1–205323-10. 65 Liliental-Weber, Z. (2000) Journal of Electron Microscopy, 49, 339. 66 Cherns, D., Wang, Y.Q., Liu, R. and Ponce, F.A. (2002) Observation of coreless edge and mixed dislocations in Mg-doped Al0.03Ga0.97N. Applied Physics Letters, 81 (24), 4541–4543. 67 Rouviere, J.L., Arlery, M. and Bourret, A. (1997) Institute of Physics Conference Series, 157, 173. 68 Sverdlov, B., Martin, G.A., Morkoc, H. and Smith, D.J. (1995) Formation of threading defects in GaN and related wurtzite films grown on non-isomorphic substrates. Applied Physics Letters, 67 (14), 2063–2065.
j 1193
1194
j 4 Extended and Point Defects, Doping, and Magnetism 69 Romano, L.T., Northrup, J.E. and OKeefe, M.A. (1996) Applied Physics Letters, 69, 2394. 70 Dimitrakopulos, G., Sanchez, A.M., Komminou, Ph., Ruterana, P., Nouet, G., Kehagias, Th. and Karakostas, Th. (2005) Physica Status Solidi c, 2, 2500. 71 Komninou, Ph., Kehagias, Th., Kioseoglou, J., Sarigiannidou, E., Karakostas, Th., Nouet, G., Ruterana, P., Amimer, K., Mikroulis, S. and Georgakilas, A. (2001) Materials Research Society Symposium Proceedings, 639, G347.1. 72 Hellman, E.S. (1998) MRS Internet Journal of Nitride Semiconductor Research, 3, 11. 73 Daudin, B., Rouviere, J.L. and Arlery, M. (1997) Materials Science & Engineering B: Solid State Materials for Advanced Technology, 43, 157. 74 Dimitrakopulos, G.P., Komninou, Ph., Kioseoglou, J., Kehagias, Th., Sarigiannidou, E., Georgakilas, A., Nouet, G. and Karakostas, Th. (2001) Physical Review B: Condensed Matter, 64, 245325. 75 Kioseoglou, J., Komninou, P., Dimitrakopulos, G.P., Kehagias, T., Polatoglou, H.M., Nouet, G. and Karakostas, T. (2003) Microstructure of planar defects and their interactions in wurtzite GaN films. Solid-State Electronics, 47 (3), 553–557. 76 Kioseoglou, J., Polatoglou, H.M., Lymperakis, L., Nouet, G. and Komninou, P.A. (2003) Modified empirical potential for energetic calculations of planar defects in GaN. Computational Materials Science, 27 (1–2), 43–49. 77 Northrup, J.E., Neugebauer, J. and Romano, L.T. (1996) Inversion domain and stacking mismatch boundaries in GaN. Physical Review Letters, 77, 103–106. 78 Austerman, S.B. and Gehman, W.G. (1966) Journal of Materials Science, 1, 249. 79 Wang, D., Hiroyama, Y., Tamura, M., Ichikawa, M. and Yoshida, S. (2000) Applied Physics Letters, 77, 1846.
80 Wang, D., Yoshida, S. and Ichikawa, M. (2002) Journal of Crystal Growth, 242, 20. 81 Liliental-Weber, Z., Sohn, H., Newman, N. and Washburn, J. (1995) Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 13, 1578. 82 Romano, L.T. and Myers, T.H. (1997) Applied Physics Letters, 71, 3486. 83 Wu, X.H., Brown, L.M., Kapolnek, D., Keller, S., Keller, B., DenBaars, S.P. and Speck, J.S. (1996) Journal of Applied Physics, 80, 3228. 84 Cherns, D., Young, W.T., Saunders, M., Steeds, J.W., Ponce, F.A. and Nakamura, S. (1998) Philosophical Magazine A, 77, 273. 85 Potin, V., Ruterana, P. and Nouet, G. (1998) Journal of Applied Physics, 82, 1276. 86 Stacking fault energies in have been calculated by Denteneer, P.J.H. and van Haeringen, W. (1987) Stacking-fault energies in semiconductors from first principles calculations. Journal of Physics C: Solid State Physics, 20, L883–L887. 87 Hirth, J.P. and Lothe, J. (1982) Theory of Dislocations, 2nd edn, Wiley Interscience, New York, p. 354. 88 Stampfl, C. and Van de Walle, C.G. (1998) Energetics and electronic structure of stacking faults in AlN, GaN and InN. Physical Review B: Condensed Matter, 57 (24), R15052–R15055. 89 Vermaut, P., Ruterana, P., Nouet, G., Salvador, A. and Morkoc, H. (1997) Philosophical Magazine A, 75, 239. 90 Potin, V., Ruterana, P. and Nouet, G. (2000) Journal of Physics: Condensed Matter, 12, 10301. 91 Akasaki, I., Amano, H., Koide, Y., Hiramatsu, K. and Sawaki, N. (1989) Journal of Crystal Growth, 98, 209. 92 Potin, V., Vermaut, P., Ruterana, P. and Nouet, G. (1998) Journal of Electronic Materials, 27, 266. 93 Liliental-Weber, Z., Kisielowski, C., Ruvimov, S., Chen, Y. and Washburn, J.
References
94
95
96
97
98
99
100
101
102 103 104
105
106
(1995) Journal of Electronic Materials, 25, 1545. Stampfl, C. and Van de Walle, C.G. (1998) Physical Review B: Condensed Matter, 57, R15052. Salviatti, G., Albrecht, M., ZanottiFregonara, C., Armani, N., Mayer, M., Shreter, Y., Guzzi, M., Melnik, Y.V., Vassilevski, K., Dmitriev, V.A. and Strunk, H.P. (1999) Physica Status Solidi a: Applied Research, 171, 325. Lorenz, K., Gonzalves, M., Kim, W., Narayanan, V. and Mahajan, S. (2000) Applied Physics Letters, 77, 3391. Ruterana, P., Potin, V., Barbaray, B. and Nouet, G. (2000) Philosophical Magazine A, 80, 937. Spence, J.C.H. and Zuo, J.M. (1992) Electron Microdiffraction, Plenum Press, New York. Wang, D., Hiroyama, Y., Tamura, M., Ichikawa, M. and Yoshida, S. (2000) Heteroepitaxial growth of cubic GaN on Si (0 0 1) coated with thin SiC by plasmaassisted molecular-beam-epitaxy. Applied Physics Letters, 76, 1683. Hiramatsu, K., Itoh, S., Amano, H., Akasaki, I., Kuwano, N., Shiraishi, T. and Oki, K. (1991) Journal of Crystal Growth, 115, 628. Wu, X.H., Fini, P., Keller, S., Tarsa, E.J., Heying, B., Mishra, U.K., DenBaars, S.P. and Speck, J.S. (1996) Japanese Journal of Applied Physics, Part 1: Regular Papers, Short Notes & Review Papers, 35, L1638. Wang, D., Yoshida, S. and Ichikawa, M. (2002) Applied Physics Letters, 80, 2472. Chisholm, J.A. and Bristowe, P.D. (2000) Applied Physics Letters, 77, 534. Zywietz, T., Neugebauer, J. and Scheffler, M. (1998) Applied Physics Letters, 73, 487. Narayanan, V., Lorenz, K., Kim, W. and Mahajan, S. (2001) Applied Physics Letters, 78, 1544. Sutton, A.P. and Balluffi, R.W. (1995) Interfaces in Crystalline Materials, Clarendon, Oxford.
107 Sarrazit, F., Pond, R.C. and Kisilev, N.A. (1998) Philosophical Magazine Letters, 77, 191. 108 Pond, R.C. and Vlachavas, D.S. (1983) Proceedings of the Royal Society A, 386, 95. 109 Potin, V., Ruterana, P., Nouet, G., Pond, R.C. and Morkoc, H. (2000) Physical Review B: Condensed Matter, 61, 5587. 110 Pond, R.C. (1989) Dislocation in Solids, vol. 8 (ed. F.R.N. Nabarro), NorthHolland, Amsterdam, p. 5. 111 Pond, R.C. and Hirth, J.P. (1994) Solid State Physics, 47, 288. 112 Dimitrakopulos, G.P., Karakostas, Th., Antonopoulos, J.G. and Pond, R.C. (1997) Interface Science, 5, 35. 113 Gutierrez-Sosa, A., Bangert, U., Harvey, A.J., Fall, C. and Jones, R. (2003) Diamond and Related Materials, 12, 1108–1112. 114 Gutierrez-Sosa, A., Bangert, U., Harvey, A.J., Fall, C.J., Jones, R., Briddon, P.R. and Heggie, M.I. (2002) Band-gap-related energies of threading dislocations and quantum wells in group-III nitride films as derived from electron energy loss spectroscopy. Physical Review B: Condensed Matter, 66, 035302-1–03530210. 115 Lee, S.M., Belkhir, M.A., Zhu, X.Y., Lee, Y.H., Hwang, Y.G. and Frauenheim, T. (2000) Electronic structures of GaN edge dislocations. Physical Review B: Condensed Matter, 61 (23), 16033–16039. 116 Cherns, D. and Jiao, C.G. (2001) Physical Review Letters, 87, 205504-1–205504-4. 117 Hansen, P.J., Strausser, Y.E., Erikson, A.N., Tarsa, E.J., Kozodoy, P., Brazel, E.G., Ibbetson, J.P., Mishra, U., Narayanamurti, V., DenBaars, S.P. and Speck, J.S. (1998) Applied Physics Letters, 72, 2247. 118 Cai, J. and Ponce, F.A. (2002) Physica Status Solidi a: Applied Research, 192, 407–411. 119 Krtschil, A., Dadgar, A. and Krost, A. (2003) Applied Physics Letters, 82, 2263–2265.
j 1195
1196
j 4 Extended and Point Defects, Doping, and Magnetism 120 Qian, W., Rohrer, G.S., Skowronski, M., Doverspike, K., Rowland, L.B. and Gaskill, D.K. (1995) Applied Physics Letters, 67, 2284. 121 Arslan, I. and Browning, N.D. (2002) Intrinsic electronic structure of threading dislocations in GaN. Physical Review B: Condensed Matter, 65, 0753101–075310-10. 122 Albrecht, M., Strunk, H.P., Weyher, J.L., Grzegory, I., Porowski, S. and Wosinski, T. (2002) Carrier recombination at single dislocations in GaN measured by cathodoluminescence in a transmission electron microscope. Applied Physics Letters, 92 (4), 2000–2002. 123 Liliental-Weber, Z., Washburn, J., Pakula, K. and Baranowki, J. (1997) Microscopy and Microanalysis, 3, 436. 124 Lee, S.M., Lee, Y.H., Hwang, Y.G., Elsner, J., Porezag, D. and Frauenheim, T. (1999) Stability and electronic structure of GaN nanotubes from density-functional calculations. Physical Review B: Condensed Matter, 60 (11), 7788–7791. 125 Lymperakis, L., Neugebauer, J., Albrecht, M., Remmele, T. and Strunk, H.P. (2004) Strain induced deep electronic states around threading dislocations in GaN. Physical Review Letters, 93 (19), 196401-1–196401-4. 126 Wright, A.F. and Grossner, U. (1998) The effect of doping and growth stoichiometry on the core structure of a threading edge dislocation. Applied Physics Letters, 73, 2751. 127 Leung, K., Wright, A.F. and Stechel, E.B. (1999) Charge accumulation at a threading edge dislocation in gallium nitride. Applied Physics Letters, 74, 2495–2497. 128 Cherns, D., Jiao, C.G., Mokhtari, H., Cai, J. and Ponce, F.A. (2002) Electron holography studies of the charge on dislocations in GaN. Physica Status Solidi b: Basic Research, 234, 924–930. 129 Polyakov, A.Y., Siminov, N.B., Goverkov, V., Fang, Z.-Q., Look, D.C., Molnar, R.J.
130 131
132
133
134 135
136
137
138
139
and Osinsky, A.V. (2002) Deep hole traps in n-GaN films grown by hydride vapor phase epitaxy. Journal of Applied Physics, 91, 6580. Schattscheider, P. (1986) Fundamentals in Inelastic Scattering, Springer, New York. Fall, C.J., Jones, R., Briddon, P.R., Blumenau, A.T., Frauenheim, T. and Heggie, M.I. (2002) Influence of dislocations on electron energy loss spectra in gallium nitride. Physical Review B: Condensed Matter, 65, 245304–245312. Bangert, U., Gutierrez-Sosa, A., Harvey, A.J., Fall, C.J. and Jones, R. (2003) Electron energy loss studies of dislocations in GaN thin films. Journal of Applied Physics, 93 (5), 2728–2735. Keast, V.J., Scott, A.J., Kappers, M.J., Foxon, C.T. and Humphreys, C.J. (2002) Electronic structure of GaN and InxGa1xN measured with electron energy-loss spectroscopy. Physical Review B: Condensed Matter, 66, 125319-1–125319-7. Batson, P.E., Delby, N. and Krivanek, O.L. (2002) Nature, 418, 617–620. Bandic, Z.Z., McGill, T.C. and Ikonic, Z. (1997) Electronic structure of GaN stacking faults. Physical Review B: Condensed Matter, 56 (7), 3564–3566. Elsner, J., Kaukonen, M., Heggie, M.I., Haugk, M., Frauenheim, Th. and Jones, R. (1998) Domain boundaries on {112 0} planes in GaN: a theoretical study. Physical Review, 58, 15347–15350. Calleja, E., Sanchez, F.J., Calle, F., Tijero, J.M.G., Sanchez-Rojas, J.L., Beaumont, B., Lorenzini, P. and Gibart, P. (1997) Physical Review B: Condensed Matter, 55, 4689. Neugebauer, J. and Van de Walle, C.G. (1996) Gallium vacancies and the yellow luminescence in GaN. Applied Physics Letters, 69 (4), 503–505. Mattila, T., Seitsonen, A.P. and Nieminen, R.M. (1996) Physical Review B: Condensed Matter, 54 (3), 1474.
References 140 Mattila, T. and Nieminen, R.M. (1997) Physical Review B: Condensed Matter, 55, 9571. 141 Chen, H.M., Chen, Y.F., Lee, M.C. and Feng, M.S. (1997) Physical Review B: Condensed Matter, 56, 6942. 142 Reshchikov, M.A. and Korotkov, R.Y. (2001) Physical Review B: Condensed Matter, 64, 115205. 143 Ponce, F.A., Cherns, D., Young, W.T. and Steeds, J.W. (1996) Characterization of dislocations in GaN by transmission electron diffraction and microscopy techniques. Applied Physics Letters, 69 (6), 770–772. 144 Liliental-Weber, Z. (1999) EMIS Data Review Series, No. 23, INSPEC, The Institution of Electrical Engineers, London, p. 230. 145 Christiansen, S.H., Albrecht, M., Strunk, H.P., Foxon, C.T., Korakakis, D., Grzegory, I. and Porowski, S. (1999) Physica Status Solidi a: Applied Research, 176, 285. 146 Liliental-Weber, Z., Benamara, M., Swider, W., Washburn, J., Grzegory, I., Porowski, S., Dupuis, R.D. and Eiting, C.J. (1999) Physica B, 273–274, 124. 147 Liliental-Weber, Z., Benamara, M., Washburn, J., Grzegory, I. and Porowski, S. (1999) Spontaneous ordering in bulk GaN:Mg samples. Physical Review Letters, 83, 2730. 148 Liliental-Weber, Z., Benamara, M., Swider, W., Wasburn, J., Grzegory, I., Porowski, S., Dupuis, R.D. and Eiting, C.J. (2000) Mg segregation: difficulties of p-doping in GaN. MRS Internet Journal of Nitride Semiconductor Research, 5S1, W.9.7 (http:// nsr.mij.mrs.org/5S1/W9.7/article.pdf). 149 Grzegory, I., Jun, J., Bockowski, M., Krukowski, St., Wroblewski, M., Lucznik, B. and Porowski, S. (1995) Journal of Physics and Chemistry of Solids, 56, 639. 150 Vennegues, P., Leroux, M., Dalmasso, S., Benaissa, M., De Mierry, P., Lorenzini, P., Bamilano, B., Beaumont, B., Massies, J. and Gibart, P. (2003) Physical Review B: Condensed Matter, 68, 235214.
151 Yan, Y., Terauchi, M. and Tanaka, M. (1998) Philosophical Magazine A, 77, 1027. 152 Westwood, A.D., Youngman, R.A., McCartney, M.R. and Cormack, A.N. (1995) Journal of Materials Research, 10, 1270. 153 Jack, K.H. (1976) Journal of Materials Science, 11, 1135. 154 Bungaro, C., Rapcewicz, K. and Bernholc, J. (1999) Physical Review B: Condensed Matter, 59, 9771. 155 Fang, Z.-Q., Look, D.C., Jasinski, J., Benamara, M., Liliental-Weber, Z. and Molnar, R.J. (2001) Evolution of deep centers in GaN grown by hydride vapor phase epitaxy. Applied Physics Letters, 78 (3), 332–334. 156 Molnar, R.J., G€otz, W., Romano, L.T. and Johnson, N.M. (1997) Journal of Crystal Growth, 178, 147. 157 Detchprohm, T., Hiramatsu, K., Amano, H. and Akasaki, I. (1992) Applied Physics Letters, 61, 2688. 158 Jasinski, J., Swider, W., Liliental-Weber, Z., Visconti, P., Jones, K.M., Reshchikov, M.A., Yun, F., Morkoc, H., Park, S.S. and Lee, K.Y. (2001) Characterization of freestanding HVPE GaN. Applied Physics Letters, 78, 2297–2299. 159 Kelly, M.K., Vaudo, R.P., Phanse, V.M., Gorgens, L., Ambacher, O. and Stutzmann, M. (1999) Japanese Journal of Applied Physics, 38 (3A), L217–L219. 160 Fang, Z.-Q., Look, D.C., Visconti, P., Wang, D.-F., Lu, C.-Z., Yun, F., Morkoc, H., Park, S.S. and Lee, K.Y. (2001) Applied Physics Letters, 78 (15), 2178–2181. 161 Neugebauer, J. and Van de Walle, C.G. (1999) Journal of Applied Physics, 85, 3003. 162 Liliental-Weber, Z., Benarama, M., Swider, W., Wasburn, J., Grzegory, I., Porowski, S., Dupuis, R.D. and Eating, C.T. (2000) Mg segregation: difficulties of p-doping in GaN. GaN and Related Alloys – 1999 Symposium, Materials Research Society, Materials Research Society Symposium Proceedings, 595, W9.7.1–W9.7.6.
j 1197
1198
j 4 Extended and Point Defects, Doping, and Magnetism 163 Northrup, J.E. (2003) Magnesium incorporation at (0 0 0 1) inversion domain boundaries in GaN. Applied Physics Letters, 82, 2278. 164 Vennegues, P., Benaissa, M., Beaumont, B., Feltin, E., De Mierry, P., Dalmasso, S., Leroux, M. and Gilbart, P. (2000) Pyramidal defects in metalorganic vapor phase epitaxial Mg doped GaN. Applied Physics Letters, 77, 880. 165 Liliental-Weber, Z., Benamara, M., Swider, W., Washburn, J., Grzegory, I., Porowski, S., Lambert, D.J.H., Eiting, C.J. and Dupuis, R.D. (1999) Mg-doped GaN: similar defects in bulk crystals and layers grown on Al2O3 by metal–organic chemical-vapor deposition. Applied Physics Letters, 75, 4159. 166 Egerton, R.F. (1986) EELS in the Electron Microscope, Plenum, New York, p. 291. 167 Romano, L.T., Northrup, J.E., Ptak, A.J. and Myers, T.H. (2000) Faceted inversion domain boundary in GaN films doped with Mg. Applied Physics Letters, 77, 2479. 168 Ramachandran, V., Feenstra, R.M., Sarney, W.L., Salamanca-Riba, L., Northrup, J.E., Romano, L.T. and Greve, W. (1999) Inversion of wurtzite GaN (0 0 0 1) by exposure to magnesium. Applied Physics Letters, 75, 808. 169 Liliental-Weber, Z., Tomaszewicz, T., Zakharov, D., OKeefe, M., Hautakangas, S., Saarinen, K., Freitas, J.A. and Henry, R.L. (2006) Atomic structure of pyramidal defects in GaN:Mg: influence of annealing. Physica Status Solidi a: Applied Research, 203 (7), 1636–1640. 170 Neugebauer, J., Zywietz, T., Scheffler, M., Northrup, J.E. and Van de Walle, C.G. (1998) Physical Review Letters, 80, 3097; Northrup, J.E. and Neugebauer, J. (1996) Physical Review B: Condensed Matter, 53, 10477. 171 Figge, S., Kroger, R., Bottcher, T., Ryder, P.L. and Hommel, D. (2002) Magnesium segregation and the formation of pyramidal defects in p-GaN. Applied Physics Letters, 81, 4748.
172 Cho, H.K. and Lee, J.Y. (2001) Influence of Mg doping on structural defects in AlGaN layers grown by metalorganic chemical vapor deposition. Applied Physics Letters, 79, 3788–3790. 173 Romano, L.T. and Myers, T.H. (1997) The influence of inversion domains on surface morphology in GaN grown by molecular beam epitaxy. Applied Physics Letters, 71, 3486. 174 Cho, H.K., Lee, J.Y., Yang, G.M. and Kim, C.S. (2001) Formation mechanism of V defects in the InGaN/GaN multiple quantum wells grown on GaN layers with low threading dislocation density. Applied Physics Letters, 79 (2), 215–217. 175 Kozawa, T., Kachi, T., Ohwaki, T., Taga, Y., Koide, N. and Koike, M. (1996) Journal of the Electrochemical Society, 143 (1), L17. 176 Hong, S.K., Yao, T., Kim, B.J., Yoon, S.Y. and Kim, T.I. (2000) Applied Physics Letters, 77, 82. 177 Hong, S.K., Kim, B.J., Park, H.S., Park, Y., Yoon, S.Y. and Kim, T.I. (1998) Journal of Crystal Growth, 191, 275–278. 178 Shiojima, K. (2000) Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 18 (1), 37. 179 Hino, T., Tomiya, S., Miyajima, T., Yanashima, K., Hashimoto, S. and Ikeda, M. (2000) Applied Physics Letters, 76, 3421. 180 Stocker, D.A., Schubert, E.F. and Redwing, J.M. (1998) Applied Physics Letters, 73, 2654. 181 Visconti, P., Jones, K.M., Reshchikov, M.A., Cingolani, R., Morkoc, H. and Molnard, R. (2000) Investigation of defects in GaN by photo-electrochemical and hot wet etching. Applied Physics Letters, 77, 3532. 182 Kapolnek, D., Wu, X.H., Heying, B., Keller, S., Keller, B.P., Mishra, U.K., DenBaars, S.P. and Speck, J.S. (1995) Applied Physics Letters, 67, 1541. 183 Tarsa, E.J., Heying, B., Wu, X.H., Fini, P., DenBaars, S.P. and Speck, J.S. (1997) Journal of Applied Physics, 82, 5472.
References 184 Weyher, J.L., Brown, P.D., Rouviere, J.L., Wosinski, T., Zauner, A.R.A. and Grzegory, I. (2000) Journal of Crystal Growth, 210, 151. 185 Girard, P. (2001) Nanotechnology, 12, 485–490. 186 Jones, K.M., Visconti, P., Yun, F., Baski, A.A. and Morkoc, H. (2001) Investigation of inversion domains in GaN by electric force microscopy. Applied Physics Letters, 78 (17), 2497–2499. 187 Hsu, J.W.P., Ng, H.M., Sergent, A.M. and Chu, S.N.G. (2002) Scanning Kelvin force microscopy imaging of surface potential variations near threading dislocations in GaN. Applied Physics Letters, 81 (19), 3579–3581. 188 Koley, G. and Spencer, M.G. (2001) Surface potential measurements on GaN and AlGaN/GaN heterostructures by scanning Kelvin probe microscopy. Journal of Applied Physics, 90 (1), 337–344. 189 Hsu, J.W.P., Manfra, M.J., Lang, D.V., Richter, S., Chu, S.N.G., Sergent, A.M., Kleiman, R., Pfeiffer, L.N. and Molnar, R.J. (2001) Applied Physics Letters, 78, 1685. 190 Hsu, J.W.P., Manfra, M.J., Chu, S.N.G., Chen, C.H., Pfeiffer, L.N. and Molnar, R.J. (2001) Applied Physics Letters, 78, 3980. 191 Hsu, J.W.P., Manfra, M.J., Molnar, R.J., Heying, B. and Speck, J.S. (2002) Applied Physics Letters, 81, 79. 192 Miller, E.J., Schaadt, D.M., Yu, E.T., Poblenz, C., Elsass, C. and Speck, J.S. (2002) Journal of Applied Physics, 91, 9821. 193 Spradlin, J., Dogan, S., Xie, J., Molnar, R.J., Baski, A.A. and Morkoc, H. (2003) Investigation of forward and reverse current conduction in MBE-grown GaN films by conductive atomic force microscopy. Applied Physics Letters, 84 (21), 4150–4152. 194 Pomarico, A.A., Huang, D., Dickinson, J., Baski, A.A., Cingolani, R., Morkoc, H. and Molnar, R. (2003) Current mapping of GaN films by conductive atomic force microscopy. Applied Physics Letters, 82, 1890–1892.
195 Fang, Z.-Q., Look, D.C., Visconti, P., Wang, D.-F., Lu, C.-Z., Yun, F., Morkoc, H., Park, S.S. and Lee, K.Y. (2001) Applied Physics Letters, 78, 2178. 196 Karrer, U., Ambacher, O. and Stutzmann, M. (2000) Applied Physics Letters, 77, 2012. 197 Hsu, J.W.P., Matthews, M.J., AbuschMagder, D., Kleiman, R.N., Lang, D.V., Richter, S., Gu, S.L. and Kuech, T.F. (2001) Applied Physics Letters, 79, 761. 198 Hsu, J.W.P., Lang, D.V., Richter, S., Kleiman, R.N., Sergent, A.M. and Molnar, R.J. (2000) Applied Physics Letters, 77, 2873. 199 Smith, K.V., Dang, X.Z., Yu, E.T. and Redwing, J.M. (2000) Charging effects in AlGaN/GaN heterostructures probed using scanning capacitance microscopy. Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 18 (4), 2304. 200 Schaadt, D.M., Miller, E.J., Yu, E.T. and Redwing, J.M. (2001) Quantitative analysis of nanoscale electronic properties in an AlxGa1xN/GaN heterostructure field-effect transistor structure. Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 19 (4), 1671. 201 Lannoo, M. and Bourgoin, J. (1981) Point Defects in Semiconductors. I. Theoretical Aspects, Springer Series in Solid–State Sciences, vol. 22, Springer, Berlin. 202 Bourgoin, J. and Lannoo, M. (1983) Point Defects in Semiconductors. II. Experimental Aspects, Springer Series in Solid–State Sciences, vol. 35, Springer, Berlin. 203 Jenkins, W. and Dow, J.D. (1989) Physical Review B: Condensed Matter, 39, 3317. 204 Jenkins, D.W., Dow, J.D. and Tsai, M.-H. (1992) Journal of Applied Physics, 72, 4130. 205 Tansley, T.L. and Foley, C.P. (1986) Journal of Applied Physics, 59, 3241. 206 Neugebauer, J. and Van de Walle, C.G., (1996) Festk€orperprobleme, Advances in Solid State Physics, vol. 35, Vieweg, Braunscheweig, p. 25.
j 1199
1200
j 4 Extended and Point Defects, Doping, and Magnetism 207 Neugebauer, J. and Van de Walle, C.G. (1994) Materials Research Society Symposium Proceedings, 339, 687. 208 Zhang, S.B. and Northrup, J.E. (1991) Physical Review Letters, 67, 2339. 209 Boguslawski, P. and Bernholc, J. (1997) Physical Review B: Condensed Matter, 56, 9496. 210 Boguslawski, P., Brigs, E.L. and Bernholc, J. (1996) Applied Physics Letters, 69, 233. 211 Neugebauer, J. and Van de Walle, C.G. (1994) Physical Review B: Condensed Matter, 50, R8067. 212 Neugebauer, J. and Van de Walle, C.G. (1995) Proceedings of the International Conference on the Physics of Semiconductors, ICPS-22, World Scientific, Singapore, p. 2327. 213 Van de Walle, C.G. (2000) Materials Science Forum, 338–342, 1561. 214 Van de Walle, C.G. and Neugebauer, J. (1997) Materials Science Forum, 258–263, 19. 215 Neugebauer, J. and Van de Walle, C.G. (1995) Hydrogen in GaN: novel aspects of a common impurity. Physical Review Letters, 75, 4452. 216 Van de Walle, C.G. (1997) Interaction of hydrogen with native defects in GaN. Physical Review B: Condensed Matter, 56, R10020–R10023. 217 Van de Walle, C.G., Neugebauer, J., Stampel, C., McCluskey, M.D. and Johnson, N.M. (1999) Acta Physica Polonica A, 96, 613. 218 Van de Walle, C.G., Neugebauer, J. and Stampel, C. (1999) Properties, Processing and Applications of Gallium Nitride and Related Semiconductors (eds J.H. Edgar, S. Strite, I. Akasaki, H. Amano and C. Wetzel), INSPEC, The Institution of Electrical Engineering, London. 219 Limpijumnong, S. and Van de Walle, C.G. (2004) Physical Review B: Condensed Matter, 69, 035207. 220 Boguslawski, P., Briggs, E.L. and Bernholc, J. (1995) Physical Review B: Condensed Matter, 51, 17255.
221 Gorczyca, I., Svane, A. and Christensen, N.E. (1999) Physical Review B: Condensed Matter, 60, 8147. 222 Gorczyca, I., Christensen, N.E. and Svane, A. (2002) Physical Review B: Condensed Matter, 66, 075210. 223 Linde, M., Uftring, S.J., Watkins, G.D., Harle, V. and Scholz, F. (1997) Physical Review B: Condensed Matter, 55, 10177. 224 Look, D.C., Reynolds, D.C., Hemsky, J.W., Sizelove, J.R., Jones, R.L. and Molnar, R.J. (1997) Physical Review Letters, 79, 2273. 225 Jenkins, D.W. and Dow, J.D. (1989) Electronic structures and doping of InN, InxGa1xN, and InxAl1xN. Physical Review B: Condensed Matter, 39, 3317–3329. 226 Van de Walle, C.G. (2003) Physica Status Solidi b: Basic Research, 235, 89. 227 Maruska, H.P. and Tietjen, J.J. (1969) Applied Physics Letters, 15, 327. 228 Perlin, P., Suzuki, T., Teisseyre, H., Leszczynski, M., Gregory, 1., Jun, J., Porowski, S., Boguslawski, P., Bernhoic, J., Chervin, J.C., Polian, A. and Moustakas, T.D. (1995) Physical Review Letters, 75, 296. 229 Wetzel, C., Walukiewicz, W., Haller, E.E. and Ager, J., III (1996) Physical Review B: Condensed Matter, 53, 1322. 230 Van de Walle, C.G. and Neugebauer, J. (2004) Journal of Applied Physics, 95, 3851. 231 Wetzel, C., Suchi, T., Ager, J.W., III, Weber, E.R., Haller, E.E., Fischer, S., Meyer, B.K., MoInax, R.J. and Berlin, P. (1997) Physical Review Letters, 78, 3923. 232 Ganchenkova, M.G. and Nieminen, R.M. (2006) Nitrogen vacancies as major point defects in gallium nitride. Physical Review Letters, 96, 196402. 233 Baraff, G.A. and Schl€ uter, M. (1985) Physical Review Letters, 55, 1327. 234 Park, C.H. and Chadi, D.J. (1997) Physical Review B: Basic Research, 55, 12995. 235 Neugebauer, J. and Van de Walle, C.G. (1996) Festk€orperprobleme, Advances in
References
236 237
238 239 240
241
242
243 244
245
246 247
248 249
250
Solid State Physics, vol. 35 (ed. R. Helbig), Vieweg, Braunschweig, p. 25. Kresse, G. and Hafner, J. (1993) Physical Review B: Condensed Matter, 47, R558. Kresse, G. and Furthm€ uller, J. (1996) Physical Review B: Condensed Matter, 54, 11169. Bl€ ochl, P.E. (1994) Physical Review B: Condensed Matter, 50, 17953. Kresse, G. and Joubert, D. (1999) Physical Review B: Condensed Matter, 59, 1758. Katsikini, M., Rossner, H., FieberErdmann, M., Holub-Krappe, E., Moustakas, T.D. and Paloura, E.C. (1999) Journal of Synchrotron Radiation, 6, 561. Oila, J., Ranki, V., Kivioja, J., Saarinen, K., Hautoj€arvi, P., Likonen, J., Baranowski, J.M., Pakula, K., Suski, T., Leszczynski, M. and Grzegory, I. (2001) Physical Review B: Condensed Matter, 63, 045205. Chow, K.H., Watkins, G.D., Usui, A. and Mizuta, M. (2000) Physical Review Letters, 85, 2761. Chadi, D.J. and Chang, K.J. (1988) Physical Review Letters, 60, 2183. Dabrowski, J. and Scheffler, M. (1989) Isolated arsenic-antisite defect in GaAs and the properties of EL2. Physical Review B: Condensed Matter, 40, 15, 10391–10401. Averkiev, N.S., Gutkin, A.A., Osipov, E.B., Reshchikov, M.A. and Sosnovskii, V.R. (1993) Defect and Diffusion Forum, 103–105, 31. Xu, H. and Lindefelt, U. (1990) Journal of Applied Physics, 68, 4077. Myles, C.W. and Sankey, O.F. (1984) Physical Review B: Condensed Matter, 29, 6810. Wright, A.F. (2002) Journal of Applied Physics, 92, 2575. Gorczyca, I., Svane, A. and Christensen, N.E. (2000) Physical Review B: Condensed Matter, 61, 7494. Kaufmann, U., Kunzer, M., Maier, M., Obloh, H., Ramakrishnan, A., Santic, B. and Schlotter, P. (1998) Applied Physics Letters, 72, 1326.
251 Li, J.Z., Lin, J.Y., Jiang, H.X., Salvador, A., Botchkarev, A. and Morkoc, H. (1996) Applied Physics Letters, 69, 1474. 252 Latham, C.D., Jones, R., Öberg, S., Nieminen, R.M. and Briddon, P.R. (2003) Physical Review B: Condensed Matter, 68, 205209. 253 Lee, S.-G. and Chang, K.J. (1999) Semiconductor Science and Technology, 14, 138. 254 Van de Walle, C.G., Limpijumnong, S. and Neugebauer, J. (2001) Physical Review B: Condensed Matter, 63, 245205. 255 Szûcs, B., Gali, A., Hajnal, Z., Deak, P. and Van de Walle, C.G. (2003) Physical Review B: Condensed Matter, 68, 085202. 256 Latham, C.D., Nieminen, R.M., Fall, C.J., Jones, R., Öberg, S. and Briddon, P.R. (2003) Physical Review B: Condensed Matter, 67, 205206. 257 Lang, D.V. (1979) Space–charge spectroscopy in semiconductor, in Thermally Stimulated Relaxation in Solids, Topics in Applied Physics, vol. 37, Springer, Berlin. 258 Kimerling, L.C. (1981) Defects in Semiconductors, vol. 2 (eds S.Narayan and T.Y. Tan), North-Holland, New York. 259 Van Vechten, J.A. and Thurmond, C.D. (1976) Entropy of ionization and temperature variation of ionization levels of defects in semiconductors. Physical Review B: Condensed Matter, 14, 3539. 260 Blood, P. and Orton, J.W. (1992) The Electrical Characterization of Semiconductors: Majority Carriers and Electron States, Academic Press, London. 261 Kim, W. (1998) Growth and characterization of wurtzite GaN epitaxial films on sapphire substrate by reactive molecular beam epitaxy and material characterization. Ph.D. Thesis, University of Illinois, Urbana-Champaign. 262 Schroder, D.K. (1990) Semiconductor Material and Device Characterization, John Wiley & Sons, Inc.
j 1201
1202
j 4 Extended and Point Defects, Doping, and Magnetism 263 Zohta, Y. and Watanabe, M.O. (1982) On the determination of the spatial distribution of deep centers in semiconducting thin films from capacitance transient spectroscopy. Journal of Applied Physics, 53 (3), 1809–1811. 264 Lang, D.V. (1974) Deep-level transient spectroscopy: a new method to characterize traps in semiconductors. Journal of Applied Physics, 45 (7), 3023. 265 Kirchner, P.D., Schaff, W.J., Maracas, G.N. and Eastman, L.F. (1981) The analysis of exponential and nonexponential transients in deep-level transient spectroscopy. Journal of Applied Physics, 52 (11), 6462. 266 Buchwald, W.R. and Johnson, N.M. (1988) Revised role for the Poole–Frenkel effect in deep-level characterization. Journal of Applied Physics, 64 (2), 958. 267 Lefevre, H. and Schulz, M. (1977) Double correlation technique (DDLTS) for the analysis of deep level transient spectroscopy. Applied Physics, 12, 45–53. 268 Wosinski, T. (1989) Evidence for the electron traps at dislocations in GaAs crystals. Journal of Applied Physics, 65 (4), 1566. 269 Criado, J., Gomez, A., Calleja, E. and Munoz, E. (1988) Novel method to determine capture cross-section activation energies by deep-level transient spectroscopy techniques. Applied Physics Letters, 52 (8), 660. 270 Ferrari, E.F., Koehler, M. and Hummelgen, I.A. (1997) Capacitancetransient-spectroscopy model for defects with two charge states. Physical Review B: Condensed Matter, 55, 9590. 271 Lang, D.V., Grimmeiss, H.G., Meijer, E. and Jaros, M. (1980) Complex nature of gold-related deep levels in silicon. Physical Review B: Condensed Matter, 22 (7), 3917. 272 Hjalmarson, H.P., Vogl, P., Wolford, D.J. and Dow, J.D. (1980) Theory of substitutional deep traps in covalent semiconductors. Physical Review Letters, 44 (12), 810–813.
273 Johnstone, D., Dogan, S., Leach, J., Moon, Y.T., Fu, Y., Hu, Y. and Morkoc, H. (2004) Applied Physics Letters, 85, 4058. 274 Reshchikov, M.A., Morkoc, H., Park, S.S. and Lee, K.Y. (2001) Applied Physics Letters, 78, 3041. 275 Pons, D. and Makram-Ebeid, S. (1979) Journal de Physique, 40, 1161. 276 GÅtz, W., Johnson, N.M., Amano, H. and Akasaki, I. (1994) Applied Physics Letters, 65, 463. 277 Hacke, P., Detchprohm, T., Hiramatsu, K., Sawaki, N., Tadatomo, K. and Miyake, K. (1994) Journal of Applied Physics, 76, 304. 278 Lee, W.I., Huang, T.C., Guo, J.D. and Feng, M.S. (1995) Applied Physics Letters, 67, 1721. 279 Haase, D., Schmid, M., K€ urner, W., DÅrnen, A., H€arle, V., Scholz, F., Burkard, M. and Schweizer, H. (1996) Applied Physics Letters, 69, 2525. 280 Hacke, P., Nakayama, H., Detchprohm, T., Hiramatsu, K. and Sawaki, N. (1996) Applied Physics Letters, 68, 1362. 281 Fang, Z.-Q., Look, D.C., Kim, W., Fan, Z., Botchkarev, A.E. and Morkoc, H. (1998) Deep centers in n-GaN grown by reactive molecular beam epitaxy. Applied Physics Letters, 72 (18), 2277–2279. 282 Auret, F.D., Goodman, S.A., Koschnick, F.K., Spaeth, J.-M., Beaumont, B. and Gibart, P. (1998) Applied Physics Letters, 73, 3745. 283 Wang, C.D., Yu, L.S., Lau, S.S., Yu, E.T., Kim, W., Botchkarev, A. and Morkoc, H. (1998) Applied Physics Letters, 72, 1211. 284 Nagai, H., Zhu, Q.S., Kawaguchi, Y., Hiramatsu, K. and Sawaki, N. (1998) Applied Physics Letters, 73, 2024. 285 Look, D.C., Fang, Z. and Polenta, L. (2000) Electrical measurements in GaN: point defects and dislocations. GaN and Related Alloys – 1999 Symposium, Materials Research Society, Materials Research Society Symposium Proceedings, 595, W10.5.1–W10.5.11. 286 Shmidt, N.M., Davydov, D.V., Emtsev, V.V., Krestnikov, I.L., Lebedev, A.A., Lundin, W.V., Poloskin, D.S., Sakharov,
References
287
288
289
290
291
292
293
294
295
296
297
A.V., Usikov, A.S. and Osinsky, A.V. (1999) Physica Status Solidi b: Basic Research, 216, 533. Hacke, P., Ramvall, P., Tanaka, S., Aoyagi, Y., Kuramata, A., Horino, K. and Munekata, H. (1999) Applied Physics Letters, 74, 543. Fang, Z.-Q., Look, D.C., Kim, W. and Morkoc, H. (2000) Characteristics of deep centers observed in n-GaN grown by reactive molecular beam epitaxy. GaN and Related Alloys – 1999 Symposium, Materials Research Society, Materials Research Society Symposium Proceedings, 595, W11.84.1–W11.84.6. Soh, C.B., Chi, D.Z., Ramam, A., Lim, H.F. and Chua, S.J. (2001) Materials Science in Semiconductor Processing, 4, 595. Reshchikov, M.A., Morkoc, H., Park, S.S. and Lee, K.Y. (2001) Applied Physics Letters, 78, 2882. Seghier, D. and Gislason, H.P. (2002) Journal of Physics D: Applied Physics, 35, 291. Muret, P., Philippe, A., Monroy, E., MuÁoz, E., Beaumont, B., Omn_es, F. and Gibart, P. (2002) Journal of Applied Physics, 91, 2998. Chen, X.D., Huang, Y., Fung, S., Beling, C.D., Ling, C.C., Sheu, J.K., Lee, M.L., Chi, G.C. and Chang, S.J. (2003) Applied Physics Letters, 82, 3671. Auret, F.D., Meyer, W.E., Wu, L., Hayes, M., Legodi, M.J., Beaumont, B. and Gibart, P. (2004) Physica Status Solidi a: Applied Research, 201, 2271. Soh, C.B., Chua, S.J., Lim, H.F., Chi, D.Z., Liu, W. and Tripathy, S. (2004) Journal of Physics: Condensed Matter, 16, 6305. Chantre, A., Vincent, G. and Bois, D. (1981) Deep-level optical spectroscopy of GaAs. Physical Review B: Condensed Matter, 23, 5335. Nakano, Y. and Kachi, T. (2001) Current deep-level transient spectroscopy investigation of acceptor levels in Mg-doped GaN. Applied Physics Letters, 79, 1631.
298 Goetz, W., Johnson, N.M., Street, R.A., Amano, H. and Akasaki, I. (1995) Applied Physics Letters, 66, 1340. 299 Franck, J. (1926) Elementary processes of photochemical reactions. Transactions of the Faraday Society, 21, 536–542. 300 Condon, E. (1926) A theory of intensity distribution in band systems. Physical Review, 28, 1182–1201. 301 Soh, C.B., Chi, D.Z., Ramam, A., Lim, H.F. and Chua, S.J. (2001) Materials Science in Semiconductor Processing, 4, 595. 302 Asghar, M., Muret, P., Beaumont, B. and Gibart, P. (2004) Materials Science & Engineering B: Solid State Materials for Advanced Technology, 113, 248. 303 Shmidt, N.M., Davydov, D.V., Emtsev, V.V., Krestnikov, I.L., Lebedev, A.A., Lundin, W.V., Poloskin, D.S., Sakharov, A.V., Usikov, A.S. and Osinsky, A.V. (1999) Physica Status Solidi b: Basic Research, 216, 533. 304 Look, D.C. and Molnar, R.J. (1997) Applied Physics Letters, 70, 3377. 305 Auret, F.D., Goodman, S.A., Koschnick, F.K., Spaeth, J.-M., Beaumont, B. and Gibart, P. (1999) Applied Physics Letters, 74, 2173. 306 Sugahara, T., Sato, H., Hao, M., Naoi, Y., Kurai, S., Tottori, S., Yamashita, K., Nishino, K., Romano, L.T. and Sakai, S. (1998) Japanese Journal of Applied Physics, Part 2: Letters, 37, L398. 307 P€od€or, B. (1966) Physica Status Solidi, 16, K167. 308 Read, W.T. (1954) Philosophical Magazine, 45, 775. 309 Ng, H.M., Doppalapudi, D., Singh, R., Moustakas, T.D., Weimann, N.G. and Eastman, L.F. (1998) The role of dislocation scattering in n-type GaN films. Applied Physics Letters, 73 (6), 821-824. 310 Weimann, N.G., Eastman, L.F., Doppalapudi, D., Ng, H.M. and Moustakas, T.D. (1998) Scattering of electrons at threading dislocations in
j 1203
1204
j 4 Extended and Point Defects, Doping, and Magnetism 311 312
313
314
315
316
317
318
319
320
321
GaN. Journal of Applied Physics, 83, 3656. Look, D.C. and Sizelove, J.R. (1999) Physical Review Letters, 82, 1237. G€ otz, W., Johnson, N.M., Bour, D.P., Chen, C., Liu, H., Kuo, C. and Imler, W. (1996) Materials Research Society Symposium Proceedings, 395, 443. Fang, Z.-Q., Look, D.C., Kim, W. and Morkoc, H. (2000) Materials Research Society Symposium Proceedings, 595, W11.84. Lee, W.I., Huang, T.C., Guo, J.D. and Feng, M.S. (1995) Applied Physics Letters, 67, 1721. Franck, J. (1926) Elementary processes of photochemical reactions. Transactions of the Faraday Society, 21, 536–542. Condon, E. (1926) A theory of intensity distribution in band systems. Physical Review, 28, 1182–1201. Chung, H.M., Chuang, W.C., Pan, Y.C., Tsai, C.C., Lee, M.C., Chen, W.H., Chen, W.K., Chiang, C.I., Lin, C.H. and Chang, H. (2000) Applied Physics Letters, 76, 897. Fang, Z.-Q., Polenta, L., Hemsky, J.W. and Look, D.C. (2000) Proceedings of the 11th International Semiconducting and Insulating Materials Conference (SIMCXI), Canberra, Australia (eds C. Jagadish and N.J. Welham), IEEE, Piscataway, NJ, p. 35. Saarinen, K., Sepp€al€a, P., Oila, J., Hautoj€arvi, P., Corbel, C., Briot, O. and Aulombard, R.L. (1998) Applied Physics Letters, 73, 3253–3255. Auret, F.D., Goodman, S.A., Meyer, W.E., Koschnick, F.K., Spaeth, J.-M., Beaumont, B. and Gibart, P. (2000) Metallisation induced electron traps in epitaxially grown n-type GaN. Materials Science & Engineering B: Solid State Materials for Advanced Technology, 71 (1–3), 77–81. Auret, F.D., Goodman, S.A., Legodi, M.J. and Meyer, W.E. (1999) Emission kinetics of electron traps introduced in n-GaN during He-ion irradiation. Nuclear
322
323
324
325
326
327
328
329
Instruments & Methods in Physics Research Section B: Beam Interactions with Materials and Atoms, 148 (1–4), 474–477. Auret, F.D., Goodman, S.A., Legodi, M.J. and Meyer, W.E. (1999) Emission kinetics of electron traps introduced in n-GaN during He-ion irradiation. Nuclear Instruments & Methods in Physics Research Section B: Beam Interactions with Materials and Atoms, 148 (1–4), 474–477. Polyakov, A.Y., Smirnov, N.B., Govorkov, A.V., Shin, M., Skowronski, M. and Greve, D.W. (1998) Deep centers and their spatial distribution in undoped GaN films grown by organometallic vapor phase epitaxy. Journal of Applied Physics, 84 (2), 870–876. Cho, H.K., Kim, C.S. and Hong, C.-H. (2003) Electron capture behaviors in unintentionally doped and intentionally doped n-type GaN. Journal of Applied Physics, 94 (3), 1485–1489. Soltanovich, O.A., Yakimov, E.B., Shmidt, N.M., Usikov, A.S. and Lundin, W.V. (2003) Correlation of diffusion length and trap concentration with dislocation density in MOCVD-grown GaN. Physica B, 340–342, 479–4833. Fang, Z.-Q., Look, D.C. and Polenta, L. (2002) Dislocation-related electron capture behavior of traps in n-GaN. Journal of Physics: Condensed Matter, 14, 13061–13068. Fang, Z., Farlowr, G., Clafin, B. and Look, D.C. (2004) Deep centers in conductive and semi-insulating GaN. 13th International Conference on Semiconducting and Insulating Materials (IEEE Cat. No. 04CH37599), pp.29–36. Pernot, J., Ulzh€ofer, C., Muret, P., Beaumont, B. and Gibart, P. (2005) Free energy and capture cross section of the E2 trap in n-type GaN. Physica Status Solidi a: Applied Research, 202, 609. Polyakov, A.Y., Smirnov, N.B., Govorkov, A.V., Milvidskii, M.G., Pearton, S.J., Usikov, A.S., Shmidt, N.M., Osinsky, A.V., Lundin, W.V., Zavarin, E.E. and Besulkin, A.I. (2003) Deep levels studies of AlGaN/
References
330
331 332
333
334
335 336
337
338
339
340
GaN superlattices. Solid-State Electronics, 47 (4), 671–676. Chernyak, L., Osinsky, A., Nootz, G., Schulte, A., Jasinski, J., Benamara, M., Liliental-Weber, Z., Look, D.C. and Molnar, R.J. (2000) Electron beam and optical depth profiling of quasibulk GaN. Applied Physics Letters, 77 (17), 2695–2697. Leamy, H.J. (1982) Journal of Applied Physics, 53, R51. Bandi, Z.Z., Bridger, P.M., Piquette, E.C. and McGill, T.C. (2000) Solid-State Electronics, 44, 221. Chernyak, L., Osinsky, A., Temkin, H., Yang, J.W., Chen, Q. and Khan, M.A. (1996) Applied Physics Letters, 69, 2531. Chernyak, L., Osinsky, A., Fuflyigin, V. and Schubert, E.F. (2000) Electron beam induced increase of electron diffusion length in p-type GaN and AlGaN/GaN superlattices. Applied Physics Letters, 77 (6), 875–877. Dimitriadis, C.A. (1981) Journal of Physics D: Applied Physics, 14, 2269. Luke, K.L., von Roos, O. and Cheng, L.-J. (1985) Journal of Applied Physics, 57, (1978). Aumer, M.E., LeBoeuf, S.F., Moody, B.F., Bedair, S.M., Nam, K., Lin, J.Y. and Jiang, H.X. (2002) Effects of tensile, compressive, and zero strain on localized states in AlInGaN/InGaN quantum-well structures. Applied Physics Letters, 80, 3099. Zhang, X., Rich, D.H., Kobayashi, J.T., Kobayashi, N.P. and Dapkus, P.D. (1998) Applied Physics Letters, 73, 1430. Miyajima, T., Ozawa, M., Asatsuma, T., Kawai, H. and Ikeda, M. (1998) Minority carrier diffusion length in GaN and ZnSe. Journal of Crystal Growth, 189–190, 768–772. Bandic, Z.Z., Bridger, P.M., Piquette, E.C. and McGill, T.C. (2000) The values of minority carrier diffusion lengths and lifetimes in GaN and their implications for bipolar devices. Solid-State Electronics, 44, 221–228.
341 Godlewski, M., ºusakowska, E., Goldys, E.M., Phillips, M.R., BÅttcher, T., Figge, S., Hommel, D., Prystawko, P., Leszczynski, M., Grzegory, I. and Porowski, S. (2004) Diffusion length of carriers and excitons in GaN—influence of epilayer microstructure. Applied Surface Science, 223, 294–302. 342 Soltanovich, O.A., Yakimov, E.B., Shmidt, N.M., Usikov, A.S. and Lundin, W.V. (2003) Correlation of diffusion length and trap concentration with dislocation density in MOCVD-grown GaN. Physica B, 340–342, 479–483. 343 Calleja, E., Sanchez-Garcıa, M.A., Basak, D., Sanchez, F.J., Calle, F., Youinou, P., Muñoz, E., Serrano, J.J., Blanco, J.M., Villar, C., Laine, T., Oila, J., Saarinen, K., Hautoj€arvi, P., Molloy, C.H., Somerford, D.J. and Harrison, I. (1998) Effect of Ga/Si interdiffusion on optical and transport properties of GaN layers grown on Si (1 1 1) by molecular-beam epitaxy. Physical Review B: Condensed Matter, 58, 1550–1559. 344 Saarinen, K., Hautoj€arvi, P. and Corbel, C. (1998) Identification of Defects in Semiconductors (ed. M. Stavola), Academic Press, New York. 345 Saarinen, K., Laine, T., Kuisma, S., Nissil€a, J., Hautoj€arvi, P., Dobrzynski, L., Baranowski, J.M., Pakula, K., Stepniewski, R., Wojdak, M., Wysmolek, A., Suski, T., Leszczynski, M., Grzegory, I. and Porowski, S. (1997) Physical Review Letters, 79, 3030. 346 Jorgensen, L.V., Kruseman, A.C., Schut, H., Veen, A.v., Fanciulli, M. and Moustakas, T.D. (1997) Materials Research Society Symposium Proceedings, 449, 853. 347 Saarinen, K. (2000) III–V Nitride Semiconductors: Electrical, Structural and Defects Properties (ed. M.O. Manasreh), Gordon and Breach, Amsterdam, p. 109. 348 Saarinen, K., Nissil€a, J., Hautoj€arvi, P., Likonen, J., Suski, T., Grzegory, I., Lucznik, B. and Porowski, S. (1999) Applied Physics Letters, 75, 2441.
j 1205
1206
j 4 Extended and Point Defects, Doping, and Magnetism 349 Saarinen, K., Suski, T., Grzegory, I. and Look, D.C. (2001) Thermal stability of isolated and complexed Ga vacancies in GaN bulk crystals. Physical Review B: Condensed Matter, 64, 233201. 350 Oila, J., Ranki, V., Kivioja, J., Saarinen, K., Hautoj€arvi, P., Likonen, J., Baranowski, J.M., Pakula, K., Suski, T., Leszczynski, M. and Grzegory, I. (2001) Influence of dopants and substrate material on the formation of Ga vacancies in epitaxial GaN layers. Physical Review B: Condensed Matter, 63, 045205. 351 Popovici, G., Kim, W., Botchkarev, A., Tang, H., Solomon, J. and Morkoc, H. (1997) Impurity contamination of GaN epitaxial films from sapphire, SiC, and ZnO substrates. Applied Physics Letters, 71, 3385–3387. 352 Hautakangas, S., Makkonen, I., Ranki, V., Puska, M.J., Saarinen, K., Xu, X. and Look, D.C. (2006) Direct evidence of impurity decoration of Ga vacancies in GaN from positron annihilation spectroscopy. Physical Review B: Condensed Matter, 73, 193301. 353 Oila, J., Kivioja, J., Ranki, V., Saarinen, K., Look, D.C., Molnar, R.J., Park, S.S., Lee, S.K. and Han, J.Y. (2003) Applied Physics Letters, 82, 3433. 354 Tuomisto, F., Hautakangas, S., Makkonen, I., Ranki, V., Puska, M.J., Saarinen, K., Bockowski, M., Suski, T., Paskova, T., Monemar, B., Xu, X. and Look, D.C. (2006) Physica Status Solidi b: Basic Research, 243, 1436–1440. 355 Hautakangas, S., Oila, J., Alatalo, M., Saarinen, K., Liszkay, L., Seghier, D. and Gislason, H.P. (2003) Physical Review Letters, 90, 137402. 356 Reboredo, F.A. and Pantelides, S.T. (1999) Physical Review Letters, 82, 1887. 357 Laukkanen, P., Lehkonen, S., Uusimaa, P., Pessa, M., Oila, J., Hautakangas, S., Saarinen, K., Likonen, J. and Ker€anen, J. (2002) Journal of Applied Physics, 92, 786. 358 Tuomisto, F., Saarinen, K., Lucznik, B., Grzegory, I., Teisseyre, H., Suski, T., Porowski, S., Hageman, P.R. and
359
360
361
362
363
364
365
366
367
368
Likonen, J. (2005) Applied Physics Letters, 86, 031915. Tuomisto, F., Saarinen, K., Paskova, T., Monemar, B., Bockowski, M. and Suski, T. (2006) Thermal stability of in-grown vacancy defects in GaN grown by hydride vapor phase epitaxy. Journal of Applied Physics, 99, 066105. Pake, G.E. and Estle, T.L. (1973) The Physical Properties of Electron Paramagnetic Resonance, Benjamin, Reading, MA. Weber, E.R., Ennen, H., Kaufmann, U., Windscheif, J., Schneider, J. and Wosinski, T. (1982) Journal of Applied Physics, 53, 6140. Malyshev, A.V., Merkulov, I.A. and Rodina, A.V. (1998) Fizika Tverdogo Tela, 40, 1002; (1998) Physics of the Solid State, 40, 917. Glaser, E.R., Freitas, J.A., Jr, Shanabrook, B.V. and Koleske, D.D. (2003) Physical Review B: Condensed Matter, 68, 195201. Glaser, E.R., Kennedy, T.A., Doverspike, K., Rowland, L.B., Gaskill, D.K., Freitas, J.A., Jr, Asif Khan, M., Olsen, D.T., Kuznia, J.N. and Wickenden, D.K. (1995) Physical Review B: Condensed Matter, 51, 13326. Suski, T., Perlin, P., Teisseyre, H., Leszczynski, M., Grzegory, I., Jun, J., Bockowski, M., Porowski, S. and Moustakas, T.D. (1995) Applied Physics Letters, 67, 2188. Reshchikov, M.A., Yi, G.-C. and Wessels, B.W. (1999) Physical Review B: Condensed Matter, 59, 13176. Bozdog, C., Przybylinska, H., Watkins, G.D., H€arle, V., Scholz, F., Mayer, M., Kamp, M., Molnar, R.J., Wickenden, A.E., Koleske, D.D. and Henry, R.L. (1999) Optical detection of electron paramagnetic resonance in electronirradiated GaN. Physical Review B: Condensed Matter, 59 (19), 12479–12486. Glaser, E.R., Kennedy, T.A., Freitas, J.A., Jr, Shanabrook, B.V., Wickenden, A.E., Koleske, D.D., Henry, R.L. and Obloh, H. (1999) Optically detected magnetic resonance of shallow donor – shallow
References
369
370
371
372
373
374
375
376 377 378 379
380
381
acceptor and deep (2.8–3.2 eV) recombination from Mg-doped GaN. Physica B, 273–274, 58–62. Anderson, P.W. (1975) Model for the electronic structure of amorphous semiconductors. Physical Review Letters, 34 (15), 953–955. Watkins, G.D. and Troxell, J.R. (1980) Negative-U properties of point defects in silicon. Physical Review Letters, 44 (9), 593–596. Troxell, J.R. and Watkins, G.D. (1980) Interstitial boron in silicon: a negative U center. Physical Review B: Condensed Matter, 22 (1), 921–931. Ambacher, O., Angerer, H., Dimitrov, R., Rieger, W., Stutzmann, M., Dollinger, G. and Bergmaier, A. (1997) Physica Status Solidi a: Applied Research, 159, 105. Zhang, J.P., Sun, D.-Z., Wang, X.-L., Kong, M.-Y., Zeng, Y.-P., Li, J.M. and Lin, L.Y. (1999) Semiconductor Science and Technology, 14, 403. Pearton, S.J. (1997) GaN and Related Materials: Optoelectronic Properties of Semiconductors and Superlattices, vol. 2 (ed. M.O. Manasreh), Gordon and Breach, New York. Walukiewicz, W. (2001) Intrinsic limitations to the doping of wide-gap semiconductors. Physica B, 302–303, 123–134. Wang, H. and Chen, A.-B. (2000) Journal of Applied Physics, 87, 7859. Pankove, J.I. and Hutchby, J.A. (1976) Journal of Applied Physics, 47, 5387. Mireles, F. and Ulloa, S.E. (1998) Physical Review B: Condensed Matter, 58, 3879. Wang, H. and Chen, A.-B. (2001) Physical Review B: Condensed Matter, 63, 125212. P€ od€ or, B. (1996) Acceptor ionization energies in gallium nitride: chemical trends and electronegativities. Semiconductor Science and Technology, 11, L827–L829. Limpijumnong, S., Van de Walle, C.G. and Neugebauer, J. (2001) Materials
382
383
384
385
386
387 388
389
390
391
392
393
394
395
Research Society Symposium Proceedings, 639, G4.3.1–G4.3.6. Ramos, L.E., Furthm€ uller, J., Scolfaro, L.M.R., Leite, J.R. and Bechstedt, F. (2002) Physical Review B: Condensed Matter, 66, 075209. Reshchikov, M. and Morkoc, H. (2005) Luminescence properties of GaN. Journal of Applied Physics Review, 97, 061301-1–061301-95. Bellaiche, L., Wei, S.-H. and Zunger, A. (1997) Physical Review B: Condensed Matter, 56, 10233. Mattila, T. and Zunger, A. (1998) Physical Review B: Condensed Matter, 58, 1367. Van de Walle, C.G. and Neugebauer, J. (2000) Applied Physics Letters, 76, 1009. Nakamura, S. and Fosol, G. (1998) The Blue Laser Diode, Springer, Berlin. Koide, N., Kato, H., Sassa, M., Yamasaki, S., Manabe, K., Hashimoto, M., Amano, H., Hiramatsu, K. and Akasaki, I. (1991) Journal of Crystal Growth, 115, 639. Goetz, W., Johnson, N.M., Chen, C., Liu, H., Kuo, C. and Imler, W. (1996) Applied Physics Letters, 68, 3144. Moore, W.J., Freitas, J.A., Jr, Lee, S.K., Park, S.S. and Han, J.Y. (2002) Physical Review B: Condensed Matter, 65, 081201. Freitas, J.A., Jr, Moore, W.J., Shanabrook, B.V., Braga, G.C.B., Lee, S.K., Park, S.S. and Han, J.Y. (2002) Physical Review B: Condensed Matter, 65, 233311. Dean, P.J., Cuthbert, J.D., Thomas, D.G. and Lynch, R.T. (1967) Physical Review Letters, 18, 122. Jayapalan, J., Skromme, B.J., Vaudo, R.P. and Phanse, V.M. (1998) Applied Physics Letters, 73, 1188. Khan, M.R.H., Ohshita, Y., Sawaki, N. and Akasaki, I. (1986) Solid State Communications, 57, 405. Malyshev, A.V., Merkulov, I.A. and Rodina, A.V. (1998) Fizika Tverdogo Tela, 40, 1002 (Physics of the Solid State, 1998, 40, 917).
j 1207
1208
j 4 Extended and Point Defects, Doping, and Magnetism 396 Lee, I.-H., Choi, I.-H., Lee, C.-R., Son, S.-J., Leem, J.-Y. and Noh, S.K. (1997) Journal of Crystal Growth, 182, 314. 397 Kaufmann, U., Kunzer, M., Obloh, H., Maier, M., Manz, Ch., Ramakrishnan, A. and Santic, B. (1999) Physical Review B: Condensed Matter, 59, 5561. 398 Laukkanen, P., Lehkonen, S., Uusimaa, P., Pessa, M., Oila, J., Hautakangas, S., Saarinen, K., Likonen, J. and Ker€anen, J. (2002) Journal of Applied Physics, 92, 786. 399 Lee, I.-H., Choi, I.-H., Lee, C.R. and Noh, S.K. (1997) Applied Physics Letters, 71, 1359. 400 Li, S., Mo, C., Wang, L., Xiong, C., Peng, X., Jiang, F., Deng, Z. and Gong, D. (2001) Journal of Luminescence, 93, 321. 401 Nakamura, S., Mukai, T. and Seno, M. (1992) Japanese Journal of Applied Physics, 31, 195. 402 Sheu, J.K. and Chi, G.C. (2002) The doping process and dopant characteristics of GaN. Journal of Physics: Condensed Matter, 14, R657–R702. 403 Goldenberg, B., Zook, J.D. and Van Vechten, J. (1993) Bulletin of the American Physical Society, 38, 446. 404 Morkoc, H., Strite, S., Gao, G.B., Lin, M.E., Sverdlov, B. and Burns, M. (1994) Journal of Applied Physics Review, 76, 1363. 405 Mohammad, S.N., Salvador, A. and Morkoc, H. (1995) Emerging GaN based devices. Proceedings of IEEE, 83, 1306–1355. 406 Mohammad, S.N. and Morkoc, H. (1996) Progress and Prospects of Group Nitride Semiconductors, Progress in Quantum Electronics, vol. 20, Pergamon, Oxford, pp. 361–525. 407 Zhang, X., Kung, P., Saxler, A., Walker, D., Wang, T.C. and Razeghi, M. (1995) Applied Physics Letters, 67, 1745. 408 Chen, H.M., Chen, Y.F., Lee, M.C. and Feng, M.S. (1997) Physical Review B: Condensed Matter, 56, 6942. 409 Yi, C.-C. and Wessels, B.W. (1996) Applied Physics Letters, 69, 3026. 410 Walukiewicz, W. (1989) Applied Physics Letters, 54, 2094.
411 Yi, G.-C. and Park, W.I. (2001) Compensation model for n-type GaN. Japanese Journal of Applied Physics, Part 1: Regular Papers, Short Notes & Review Papers, 40 (11), 6243–6247; Yi, G.-C. and Park, W.I. (2001) Photoluminescence properties of Se-doped GaN. Japanese Journal of Applied Physics, Part 1: Regular Papers, Short Notes & Review Papers, 40 (7), 4470–4474. 412 Moore, W.J., Freitas, J.A., Jr, Lee, S.K., Park, S.S. and Han, J.Y. (2002) Physical Review B: Condensed Matter, 65, 081201. 413 Burstein, E. (1954) Physical Review, 93, 632. 414 Haug, H. and Koch, S.W. (1989) Physical Review A, 39, 1887. 415 Korotkov, R.Y., Niu, F., Gregie, J.M. and Wessels, B.W. (2001) Physica B, 308–310, 26. 416 Wetzel, C., Suski, T., Ager, J.W., III, Weber, E.R., Haller, E.E., Fischer, S., Meyer, B.K., Molnar, R.J. and Perlin, P. (1997) Physical Review Letters, 78, 3923. 417 Gassmann, A., Suski, T., Newman, N., Kisielowski, C., Jones, E., Weber, E.R., Liliental-Weber, Z., Rubin, M.D., Helava, H.I., Grzerory, I., Bockowski, M., Jun, J. and Porowski, S. (1996) Journal of Applied Physics, 80, 2195. 418 Slack, G.A., Schowalter, L.J., Morelli, D. and Freitas, J.A., Jr (2002) Journal of Crystal Growth, 246, 287. 419 Perlin, P., Suski, T., Teisseyre, H., Leszczynski, M., Grzegory, I., Jun, I., Porowski, S., Boguslawski, P., Bernholc, J., Chervin, J.C., Polian, A. and Moustakas, T.D. (1995) Physical Review Letters, 75, 296. 420 Amano, H., Kito, M., Hiramatsu, K. and Akasaki, I. (1989) Japanese Journal of Applied Physics, 28, L2112. 421 Akasaki, I., Amano, H., Kito, M. and Hiramatsu, K. (1991) Journal of Luminescence, 48–49, 666. 422 Nakamura, S., Mukai, T., Senoh, M. and Isawa, N. (1992) Japanese Journal of Applied Physics, 31, L139.
References 423 Akasaki, I. and Amano, H. (1994) Journal of the Electrochemical Society, 141, 2266. 424 G€ otz, W., Johnson, N.M., Walker, J., Bour, D.P. and Street, R.A. (1996) Applied Physics Letters, 68, 667. 425 Kozodoy, P., Xing, H., DenBaars, S.P., Mishra, U.K., Saxler, A., Perrin, R., Elhamri, S. and Mitchel, W.C. (2000) Journal of Applied Physics, 87, 1832. 426 Nakamura, S., Iwasa, N., Senoh, M. and Mukai, T. (1992) Japanese Journal of Applied Physics, Part 1: Regular Papers, Short Notes & Review Papers, 31, 1258. 427 Harima, H., Inoue, T., Nakashima, S., Ishida, M. and Taneya, M. (1999) Applied Physics Letters, 75, 1383. 428 Myers, S.M., Wright, A.F., Petersen, G.A., Wampler, W.R., Seager, C.H., Crawford, M.H. and Han, J. (2001) Journal of Applied Physics, 89, 3195. 429 Litwin-Staszewska, E., Suski, T., Piotrzkowski, R., Grzegory, I., Bockowski, M., Robert, J.L., Konczewicz, L., Wasik, D., Kaminska, E., Cote, D. and Clerjaud, B. (2001) Journal of Applied Physics, 89, 7960. 430 Myers, S.M., Seager, C.H., Wright, A.F., Vaandrager, B.L. and Nelson, J.S. (2002) Journal of Applied Physics, 92, 6630. 431 Kamiura, Y., Yamashita, Y. and Nakamura, S. (1999) Physica B, 273–274, 54. 432 Amano, H., Kito, M., Hiramatsu, K. and Akasaki, I. (1989) Japanese Journal of Applied Physics, 28, L2112. 433 Nakamura, S., Mukai, T. and Senoh, M. (1991) Japanese Journal of Applied Physics, 30, L1998. 434 Akasaki, I., Amare, H., Kitoh, M., Hiramatsu, K. and Akasaki, Z. (1989) Journal of the Electrochemical Society, 137, 1639. 435 Obloh, H., Bachem, K.H., Kaufmann, U., Kunzer, M., Maier, M., Ramakrishnan, A. and Schlotter, P. (1998) Selfcompensation in Mg doped p-type GaN grown by MOCVD. Journal of Crystal Growth, 195, 270.
436 Goepfert, I.D., Schubert, E.F., Osinsky, A. and Norris, P.E. (1999) Electronics Letters, 35, 1109. 437 G€otz, W., Johnson, N.M., Walker, J., Bour, D.P. and Street, R.A. (1996) Applied Physics Letters, 68, 667. 438 Tanaka, T., Watanabe, A., Amano, H., Kobayashi, Y. and Akasaki, I. (1994) Applied Physics Letters, 65, 593. 439 Kim, W., Salvador, A., Botchkarev, A.E., Aktas, O., Mohammad, S.N. and Morkoc, H. (1996) Applied Physics Letters, 69, 559. 440 Kim, W., Botchkarev, A.E., Salvador, A., Popovici, G., Tang, H. and Morkoc, H. (1997) Applied Physics Letters, 82, 219. 441 Suzuki, M., Nishio, J., Onomura, M. and Hongo, C. (1998) Journal of Crystal Growth, 189–190, 511. 442 Sugiura, L., Suzuki, M., Nishio, J., Itaya, K., Kokubun, Y. and Ishikawa, M. (1998) Characteristics of Mg-doped GaN and AlGaN grown by H2-ambient and N2-ambient metalorganic chemical vapor deposition. Japanese Journal of Applied Physics, Part 1: Regular Papers, Short Notes & Review Papers, 17, 3878–3881. 443 Tanaka, T., Watanabe, A., Amano, H., Kobayashi, Y., Akasaki, I., Yamazaki, S. and Koike, M. (1994) Applied Physics Letters, 65, 593. 444 Akasaki, I. and Amano, H. (1991) Materials Research Society Symposium Proceedings, 242, 383. 445 Li, J., Oder, T.N., Nakarmi, M.L., Lin, J.Y. and Jiang, H.X. (2002) Optical and electrical properties of Mg-doped p-type AlxGa1xN. Applied Physics Letters, 80, 1210. 446 Suzuki, M., Nishio, J., Onomura, M. and Hongo, C. (1998) Journal of Crystal Growth, 189–190, 511. 447 Stepniewski, R., Wysmolek, A., Potemski, M., Lusakowski, J., Korona, K., Pakula, K., Baranowski, J.M., Martinez, G., Wyder, P., Grzegory, I. and Porowski, S. (1998) Impurity-related luminescence of homoepitaxial GaN studied with high magnetic fields. Physica Status Solidi b: Basic Research, 210, 373.
j 1209
1210
j 4 Extended and Point Defects, Doping, and Magnetism 448 Stepniewski, R. and Wysmolek, A. (1996) Acta Physica Polonica A, 90, 681. 449 Fiorentini, V., Metlifessel, M. and Scheffier, M. (1993) Physical Review B: Condensed Matter, 47, 13353. 450 Wei, S.-H. and Zunger, A. (1988) Physical Review B: Condensed Matter, 37, 8958. 451 Kozodoy, P., Xing, H., DenBaars, S.P. and Mishra, U.K. (2000) Journal of Applied Physics, 87, 1832. 452 Cheong, M.G., Kim, K.S., Kim, C.S., Choi, R.J., Yoon, H.S., Namgung, N.W., Suh, E.-K. and Lee, H.J. (2002) Applied Physics Letters, 80, 1001. 453 Kim, W., Salvador, A., Botchkarev, A.E., Aktas, Ö., Mohammad, S.N. and Morkoc, H. (1996) Applied Physics Letters, 69, 559. 454 Li, L.K., Jurkovic, M.J., Wang, W.I., Van Hove, J.M. and Chow, P.P. (2000) Applied Physics Letters, 76, 1740. 455 Ptak, A.J., Myers, T.H., Romano, L.T., Van de Walle, C.G. and Northrup, J.E. (2001) Applied Physics Letters, 78, 285. 456 Guha, S., Bojarczuk, N.A. and Cardone, F. (1997) Applied Physics Letters, 71, 1685. 457 Cheng, T.S., Novikov, S.V., Foxon, C.T. and Orton, J.W. (1999) Journal of Crystal Growth, 109, 439. 458 Smorchkova, I.P., Haus, E., Heying, B., Kozodoy, P., Fini, P., Ibbetson, J.P., Keller, S., DenBaars, S.P., Speck, J.S. and Mishra, U.K. (2000) Applied Physics Letters, 76, 718. 459 Kaufmann, U., Kunzer, M., Maier, M., Obloh, H., Ramakrishnan, A., Santic, B. and Schlotter, P. (1998) Applied Physics Letters, 72, 1326. 460 Sheu, J.K., Su, Y.K., Chi, G.C., Pong, B.J., Chen, C.Y., Huang, C.N. and Chen, W.C. (1998) Journal of Applied Physics, 84, 4590. 461 Zhang, X., Chua, S.J., Li, P., Chong, K.B. and Wang, W. (1998) Applied Physics Letters, 73, 1772. 462 Sugiura, L., Suziki, M. and Nishio, J. (1998) Applied Physics Letters, 72, 1748. 463 Eckey, L., von Gfug, U., Holst, J., Hoffmann, A., Kaschner, A., Siegle, H., Thomsen, c., Schineller, B., Heime, K.,
464
465
466
467 468 469
470
471
472
473
474
475
476
477
Heuken, M., Schon, O. and Beccard, R. (1998) Journal of Applied Physics, 84, 5828. Suski, T., Jun, J., Lesczynski, M., Teisseyre, H., Strite, S. Rockett, A., Pelzmann, A., Kamp, M. and Ebeling, K.J. (1998) Journal of Applied Physics, 84, 1155. Pong, B.J., Pan, J., Teng, Y.C., Chi, G.C., Li, W.H., Lee, K.C. and Lee, C.H. (1998) Journal of Applied Physics, 83, 5992. Ronning, C., Carlson, E.P., Thomson, D.B. and Davis, R.F. (1998) Applied Physics Letters, 73, 1622. Oh, E., Park, H. and Park, Y. (1998) Applied Physics Letters, 72, 70. Zhang, R. and Kuech, T.F. (1998) Applied Physics Letters, 72, 1611. Cheng, T.S., Novikov, S.V., Foxon, C.T. and Orton, J.W. (1999) Solid State Communications, 109, 439. Pankove, J.I., Torvik, J.T., Qiu, C.H., Grzegory, I., Porowski, S., Quigley, P. and Martin, B. (1999) Applied Physics Letters, 74, 416. Youn, D.H., Lachab, M., Hao, M., Sugahara, T., Takenaka, H., Naoi, Y. Sakai, S. (1999) Japanese Journal of Applied Physics, 38, 631. Zaldivar, M.H., Fernandez, P., Piqueras, J. and Solis, J. (1999) Journal of Applied Physics, 85, 1120. Reshchikov, M.A., Yi, G.C. and Wessels, B.W. (1999) Physical Review B: Condensed Matter, 59, 13176. As, D.J., Simonsmeier, T., Schottker, B., Frey, T., Schikora, D., Kriegseis, W., Burkhardt, W. and Meyer, B.K. (1998) Applied Physics Letters, 73, 1835. Korothov, R.Y., Gregie, J.M. and Wessels, B.W. (2001) Materials Research Society Symposium Proceedings, 639, G37.1. Korothov, R.Y., Gregie, J.M. and Wessels, B.W. (2001) Materials Research Society Symposium Proceedings, 639, G639.1. Glaser, E.R., Carlos, W.E., Braga, G.C.B., Freitas, J.A., Jr, Moore, W.J., Shanabrook, B.V., Henry, R.L., Wichen, A.E., Koleske, D.D., Obloh, H., Kozodoy, P., DenBaars, S.P. and Mishra,
References
478 479
480
481
482 483
484
485
486
487
488
489
490
491
U.K. (2002) Physical Review B: Condensed Matter, 65, 085312. Shahedipour, F. and Wessels, B.W. (2000) Applied Physics Letters, 76, 3011. Shahedipour, F. and Wessels, B.W. (2001) MRS Internet Journal of Nitride Semiconductor Research, 6, 12. Scherer, M., Schwegler, V., Seyboth, M., Kirchner, c. and Kamp, M. (2001) Journal of Applied Physics, 89, 8339. Jeong, S.M., Shim, H.W., Yoon, H.S., Cheong, M.G., Choi, R.J., Suh, E.-K. and Lee, H.J. (2002) Journal of Applied Physics, 91, 9711. Kim, K. and Chung, S.J. (2002) Applied Physics Letters, 80, 1767. Koide, Y., Walker, D.E., Jr, White, B.D., Brillson, L.J., Murakami, M., Kamiyama, S., Amano, H. and Akasaki, I. (2002) Journal of Applied Physics, 92, 3657. Krtschyil, A., Kielburg, A., Witte, H., Christen, J., Krost, A., Wenzel, A. and Rauschenbach, B. (2002) Journal of Applied Physics, 91, 178. Ho, J.K., Jong, C.S., Chiu, C.C., Huang, C.N., Shih, K.K., Chen, L.C., Chen, F.R. and Kai, J.J. (1999) Journal of Applied Physics, 86, 4491. Chen, L.C., Chen, F.R., Kai, J.J., Chang, L., Ho, J.K., Jong, C.S., Chiu, C.C., Huang, C.N., Chen, C.Y. and Shih, K.K. (1999) Journal of Applied Physics, 86, 3826. Koide, Y., Maeda, T., Kawakami, T., Fujita, S., Uemura, T., Shibata, N. and Murakami, M. (1999) Journal of Electronic Materials, 28, 341. Hull, B.A., Mokhey, S.E., Venugopalan, H.S. and Rana, J.C. (2000) Applied Physics Letters, 76, 2271. Maeda, T., Koide, Y. and Murakami, M. (1999) Applied Physics Letters, 75, 4145. Chen, L.C., Ho, J.K., Jong, C.S., Chiu, C.C., Shih, K.K., Chen, F.-R., Kai, J.-J. and Chang, L. (2000) Applied Physics Letters, 76, 3703. Zhang, A.P., Luo, B., Johnson, J.W., Ren, F., Han, J. and Pearton, S.J. (2001) Applied Physics Letters, 79, 3636.
492 Korotkov, R.Y., Cregie, J.M. and Wessels, B.W. (2002) Codoping of wide gap epitaxial III-nitride semiconductors. Optoelectronics Review, 10 (4), 243–249. 493 Reiss, H., Fuller, C.S. and Morin, F.J. (1956) Chemical interactions among defects in germanium and silicon. Bell System Technical Journal, 35, 535–636. 494 Reiss, H., Fuller, C.S. and Pietruszkievicz, A.J. (1956) Solubility of lithium in doped and undoped silicon: evidence for compound formation. Journal of Chemical Physics, 25, 650–655. 495 Katayama-Yoshida, H., Kato, R. and Yamamoto, T. (2001) New valence control and spin control method in GaN and AlN by codoping and transition atom doping. Journal of Crystal Growth, 231, 428–436. 496 Kaufmann, U., Schlotter, P., Obloh, H., Kohler, K. and Maier, M. (2000) Hole conductivity and compensation in epitaxial GaN:Mg layers. Physical Review B: Condensed Matter, 62, 10867–10872. 497 Yamamoto, T. and Katayama-Yoshida, H. (1997) Japanese Journal of Applied Physics, 36, L180. 498 Yamamoto, T. and Katayama-Yoshida, H. (1997) Control of valence states by a codoping method in p-type GaN materials, in Gallium Nitride and Related Materials II (eds C.R. Abernathy, H. Amano and J.C. Zolper), Materials Research Society Symposium Proceedings, 468, 105–110. 499 Yamamoto, T. and Katayama-Yoshida, H. (1998) Electronic structures of p-type GaN codoped with Be or Mg as the acceptors and Si or O as the donor codopants. Journal of Crystal Growth, 190, 532. 500 Katayama-Yoshida, H., Nishimatsu, T., Yamamoto, T. and Orita, N. (2001) Codoping method for the fabrication of low-resistivity wide band-gap semiconductors in p-type GaN, p-type AlN and n-type diamond: prediction versus experiment. Journal of Physics C: Solid State Physics, 13, 8901–8914.
j 1211
1212
j 4 Extended and Point Defects, Doping, and Magnetism 501 Korotkov, R.Y., Gregie, J.M. and Wessels, B.W. (2001) Electrical properties of p-type GaN:Mg codoped with oxygen. Applied Physics Letters, 78, 222–224. 502 Ploog, K. and Brandt, O. (1998) Doping of group III nitrides. Journal of Vacuum Science & Technology A: Vacuum Surfaces and Films, 16 (Part 2), 1609–1614. 503 Marfaing, Y. (2002) Phenomenological analysis of codoping role of statistical fluctuations. Physica Status Solidi b: Basic Research, 229, 229–238. 504 Kipshidze, G., Kuryatkov, V., Borisov, B., Kudryavtsev, Yu. Asomoza, R., Nikishin, S. and Temkin, H. (2002) Mg and O co-doping in p-type GaN and AxGa1xN (0 < x < 0.08). Applied Physics Letters, 80, 2910–2912. 505 Ptak, A.J., Myers, T.H., Romano, L.T., Van de Walle, C.G. and Northrup, J.E. (2001) Applied Physics Letters, 78, 285. 506 Grandjean, N., Leroux, M., Massies, J., Mesrine, M. and La€ ugt, M. (1999) Japanese Journal of Applied Physics, Part 1: Regular Papers, Short Notes & Review Papers, 38, 618. 507 Kim, K.S., Han, M.S., Yang, G.M., Youn, C.J., Lee, H.J., Cho, H.K. and Lee, J.Y. (2000) Applied Physics Letters, 77, 1123. 508 Qi, Y.D., Musante, C., Lau, K.M., Smith, L., Odedra, R. and Kanjolia, R. (2001) Journal of Electronic Materials, 30, 1382. 509 Kent, D.G., Overberg, M.E. and Pearton, S.J. (2001) Journal of Applied Physics, 90, 3750. 510 Yu, C.C., Chu, C.F., Tsai, J.Y., Lin, C.F., Lan, W.H., Chiang, C.I. and Wang, S.C. (2001) Japanese Journal of Applied Physics, 40 (5A), L417. 511 Schubert, E.F., Grieshaber, W. and Goepfert, I.D. (1996) Applied Physics Letters, 69, 3737. 512 Kozodoy, P., Smorchkova, Y.P., Hansen, M., Xing, H., DenBaars, S.P., Mishra, U.K., Saxler, A.W., Perrin, R. and Mitchel, W.C. (1999) Applied Physics Letters, 75, 2444.
513 Saxler, A., Mitchel, W.C., Kung, P. and Razeghi, M. (1999) Applied Physics Letters, 74, 2023. 514 Hsu, L. and Walukiewicz, W. (1999) Applied Physics Letters, 74, 2405. 515 Saxler, A., Mitchel, W.C., Kung, P. and Razeghi, M. (1999) Applied Physics Letters, 74, 2023. 516 Kozodoy, P., Hansen, M., DenBaars, S.P. and Mishra, U. (1999) Enhanced Mg doping efficiency in A0.2Ga0.8N/GaN superlattices. Applied Physics Letters, 74, 3681. 517 See also: Bykhovski, A.D., Gelmont, B.L. and Shur, M.S. (1997) Journal of Applied Physics, 81, 6332. Bernardini, F. and Fiorentini, V. (1997) Physical Review B: Condensed Matter, 56, R10024. 518 Waldron, E.L., John, W., Graff, E. and Schubert, F. (2001) Improved mobilities and resistivities in modulation-doped p-type AlGaN/GaN superlattices. Applied Physics Letters, 79, 2737. 519 Kim, K.S., Han, M.S., Yang, G.M., Youn, C.J., Lee, H.J., Cho, H.K. and Lee, J.Y. (2000) Applied Physics Letters, 77, 1123. 520 Qi, Y.D., Musante, C., Lau, K.M., Smith, L., Odedra, R. and Kanjolia, R. (2001) Journal of Electronic Materials, 30, 1382. 521 Kent, D.G., Overberg, M.E. and Pearton, S.J. (2001) Journal of Applied Physics, 90, 3750. 522 Yu, C.C., Chu, C.F., Tsai, J.Y., Lin, C.F., Lan, W.H., Chiang, C.I. and Wang, S.C. (2001) Japanese Journal of Applied Physics, 40 (5A), L417. 523 See, for example, Adivarahan, V., Zhang, J.P., Chitnis, A., Shuai, W., Sun, J., Pachipulusu, R., Shatalov, M. and Khan, M.A. (2002) Japanese Journal of Applied Physics, 41, L435. 524 Nakamura, S., Senoh, M. and Mukai, T. (1991) Japanese Journal of Applied Physics, Part 1: Regular Papers, Short Notes & Review Papers, 30, L1708. 525 Van Vechten, J.A., Zook, J.D., Hornig, R.D. and Goldenberg, B. (1992) Japanese Journal of Applied Physics, 31, 3662.
References 526 Neugebauer, J. and Van de Walle, C.G. (1996) Applied Physics Letters, 68, 1829. 527 Okamoto, Y., Saito, M. and Oshiyama, A. (1996) Japanese Journal of Applied Physics, 35, L807. 528 Van Vechten, J.A. (1997) private communication. 529 Chevallier, J., Clerjaud, B. and Pajot, B. (1991) Hydrogen in Semiconductors, Semiconductors and Semimetals, vol. 34 (eds J.I. Pankove and N.M. Johnson), Academic Press, New York. 530 Pankove, J.I., Duffy, M.T., Miller, E.A. and Berkeyheiser, J.E. (1973) Journal of Luminescence, 8, 89. 531 Ilegems, M. and Dingle, R. (1973) Journal of Applied Physics, 44, 2434. 532 Bernardini, F. and Fiorentini, V. (1997) Applied Physics Letters, 70, 2990. 533 Salvador, A., Kim, W., Aktas, O., Botchkarev, A., Fan, Z. and Morkoc, H. (1996) Applied Physics Letters, 69, 2692. 534 Dewsnip, G.J., Andrianov, A.V., Harrison, I., Orton, J.W., Lacklison, D.E., Ren, G.B., Hooper, S.E., Cheng, T.S. and Foxon, C.T. (1998) Semiconductor Science and Technology, 13, 500. 535 Sanchez, F.J., Calle, F., Sanchez-Garcıa, M.A., Calleja, E., Munoz, E., Molloy, C.H., Somerford, D.J., Serrano, J.J. and Blanco, J.M. (1998) Semiconductor Science and Technology, 13, 1130. 536 Ptak, A.J., Myers, T.H., Wang, L., Giles, N.C., Moldovan, M., Da Cunha, C.R., Hornak, L.A., Tian, C., Hockett, R.A., Mitha, S. and Van Lierde, P. (2001) Materials Research Society Symposium Proceedings, 639, G3.3. 537 Ronning, C., Carlson, E.P., Thomson, D.B. and Davis, R.F. (1998) Applied Physics Letters, 73, 1622. 538 Heinke, H., Kirchner, V., Einfeidt, S. and Hommel, D. (2000) Applied Physics Letters, 77, 2145. 539 Yu, C.-C., Chu, C.F., Tsai, J.Y. and Wang, S.C. (2001) Materials Science & Engineering B: Solid State Materials for Advanced Technology, 82, 82.
540 Yu, C.-C., Chu, C.F., Tsai, J.Y., Lin, C.F. and Wang, S.C. (2002) Journal of Applied Physics, 92, 1881. 541 Brandt, O., Yang, H., Kostial, H. and Ploog, K. (1996) Applied Physics Letters, 69, 2702. 542 Pankove, J.I. and Hutchby, J.A. (1976) Journal of Applied Physics, 47, 5387. 543 Ejder, E. and Grimmeiss, H.G. (1974) Journal of Applied Physics, 5, 275. 544 Fang, Z.-Q., Look, D.C., Armitage, R., Yang, Q. and Weber, E.R. (2004) Materials Research Society Symposium Proceedings, 798, Y5.27. 545 Klein, P.B., Binari, S.C., Ikossi, K., Wickenden, A.E., Koleske, D.D. and Henry, R.L. (2001) Applied Physics Letters, 79, 3527. 546 Armstrong, A., Arehart, A.R., Ringel, S.A., Moran, B., DenBaars, S.P., Mishra, U.K. and Speck, J.S. (2004) Materials Research Society Symposium Proceedings, 798, Y5.38. 547 Armstrong, A., Arehart, A.R., Moran, B., DenBaars, S.P., Mishra, U.K., Speck, J.S. and Ringel, S.A. (2004) Impact of carbon on trap states in n-type GaN grown by metalorganic chemical vapor deposition. Applied Physics Letters, 84 (3), 374–376. 548 Sato, M. (1996) Highly resistive CH-doped GaN grown by plasmaassisted metalorganic chemical vapor deposition. Applied Physics Letters, 68 (7), 935–937. 549 Abernathy, C.R., MacKenzie, J.D., Pearton, S.J. and Hobson, W.S. (1995) CCl4 doping of GaN grown by metalorganic molecular beam epitaxy. Applied Physics Letters, 66 (15), 1969–1971. 550 As, D.J., Schmilgus, F., Wang, C., Sch€ottker, B., Schikora, D. and Lischka, K. (1997) The near band edge photoluminescence of cubic GaN epilayers. Applied Physics Letters, 70, 1311–1313. 551 As, D.J., K€ohler, U. and Lischka, K. (2002) Materials Research Society Symposium Proceedings, 693, I2.3.
j 1213
1214
j 4 Extended and Point Defects, Doping, and Magnetism 552 As, D.J., K€ohler, U., L€ ubbers, M., Mimkes, J. and Lischka, K. (2001) Physica Status Solidi a: Applied Research, 188, 699. 553 Fisher, S., Wetzel, C., Haller, E.E. and Meyer, B.K. (1995) Applied Physics Letters, 67, 1298. 554 Ogino, T. and Aoki, M. (1980) Japanese Journal of Applied Physics, 19, 2395. 555 Niebuhr, R., Bachem, K., Bombrowski, K., Maier, M., Plerschen, W. and Kaufmann, U. (1995) Journal of Electronic Materials, 24, 1531. 556 Polyakov, A.Y., Shin, M., Freitas, J.A., Skowronski, M., Greve, D.W. and Wilson, R.G. (1996) Journal of Applied Physics, 80, 6349. 557 Kucheyev, S.O., Toth, M., Phillips, M.R., Williams, J.S., Jagadish, C. and Li, G. (2002) Journal of Applied Physics, 91, 5867. 558 Silkowski, E., Yeo, Y.K., Hengehold, R.L., Khan, M.A., Lei, T., Evans, K. and Cerny, C. (1996) Materials Research Society Symposium Proceedings, 395, 813. 559 Reuter, E.E., Zhang, R., Kuech, T.F. and Bishop, S.G. (1999) MRS Internet Journal of Nitride Semiconductor Research, 4S1, G3.67. 560 Tang, H., Webb, J.B., Bardwell, J.A., Raymond, S., Salzman, J. and Uzan-Saguy, C. (2001) Applied Physics Letters, 78, 757. 561 Birkle, U., Fehrer, M., Kirchner, V., Einfeldt, S., Hommel, D., Strauf, S., Michler, P. and Gutowski, J. (1999) MRS Internet Journal of Nitride Semiconductor Research, 4S1, G5.6. 562 Zhang, R. and Kuech, T.F. (1998) Applied Physics Letters, 72, 1611. 563 Armitage, R., Hong, W., Yang, Q., Feick, H., Gebauer, J., Weber, E.R., Hautakangas, S. and Saarinen, K. (2003) Applied Physics Letters, 82, 3457. 564 Skromme, B.J. and Martinez, G.L. (2000) MRS Internet Journal of Nitride Semiconductor Research, 5S1, W9.8. 565 Skromme, B.J., Martinez, G.L., Krasnobaev, L. and Poker, D.B. (2001) Materials Research Society Symposium Proceedings, 639, G11.39.
566 Seager, C.H., Wright, A.F., Yu, J. and G€otz, W. (2002) Journal of Applied Physics, 92, 6553. 567 Pankove, J.I., Miller, E.A. and Berkeyheiser, J.E. (1971) RCA Reviews, 32, 383. 568 Amano, H., Sawaki, S., Akasaki, I. and Toyoda, Y. (1988) Journal of Crystal Growth, 93, 79. 569 Bernardini, F., Fiorentini, V. and Nieminen, R.M. (1996) Proceedings of the 23rd International Conference on the Physics of Semiconductors, 1996, Berlin, World Scientific, Singapore, p. 497. 570 Monemar, B., Gislason, H.P. and Lagertedt, O. (1980) Journal of Applied Physics, 51, 625. 571 Strite, S. (1994) Japanese Journal of Applied Physics, 33, L699. 572 Lee, J.W., Pearton, S.J., Zolper, J.C. and Stall, R.A. (1996) Applied Physics Letters, 68, 2102. 573 Monteiro, T., Boemare, C., Soares, M.J., Alves, E. and Liu, C. (2001) Physica B, 42, 308–310. 574 Chen, L. and Skromme, B.J. (2003) Materials Research Society Symposium Proceedings, 743, L11.35. 575 Ilegems, M., Dingle, R. and Logan, R.A. (1972) Journal of Applied Physics, 43, 3797. 576 Lagerstedt, O. and Monemar, B. (1974) Luminescence in epitaxial GaN:Cd. Journal of Applied Physics, 45 (5), 2266–2272. 577 Bergman, P., Ying, G., Monemar, B. and Holtz, P.O. (1987) Journal of Applied Physics, 61, 4589. 578 Vaudo, R.P., Xu, X., Salant, A., Malcarne, J. and Brandes, G.R. (2003) Characteristics of semi-insulating, Fe-doped GaN substrates. Physica Status Solidi a: Applied Research, 200 (1), 18–21. 579 Li, X., Kim, S., Reuter, E.E., Bishop, S.G. and Coleman, J.J. (1998) Applied Physics Letters, 72, 1990. 580 Winser, A.J., Novikov, S.V., Davis, C.S., Cheng, T.S., Foxon, C.T. and Harrison, I. (2000) Applied Physics Letters, 77, 2506.
References 581 Gil, B., Morel, A., Taliercio, T., Lefebvre, P., Foxon, C.T., Harrison, I., Winser, A.J. and Novikov, S.V. (2001) Applied Physics Letters, 79, 69. 582 Novikov, S.V., Li, T., Winser, A.J., Foxon, C.T., Campion, R.P., Staddon, C.R., Davis, S., Harrison, I., Kovarsky, A.P. and Ber, B.J. (2001) Physica Status Solidi b: Basic Research, 228, 227. 583 Guido, L.J., Mitev, P., Gerasimova, M. and Gaffey, B. (1998) Applied Physics Letters, 72, 2005. 584 Jin, S.R., Ramsteiner, M., Grahn, H.T., Ploog, K.H., Zhu, Z.Q., Shen, D.X., Li, A.Z., Metev, P. and Guido, L.J. (2000) Journal of Crystal Growth, 212, 56. 585 Huang, H.Y., Chuang, C.H., Shu, C.K., Pan, Y.C., Lee, W.H., Chen, W.K., Chen, W.H. and Lee, M.C. (2002) Applied Physics Letters, 80, 3349. 586 MacKenzie, J.D., Abernathy, C.R., Pearton, S.J., Hommerich, U., Wu, X., Schwartz, R.N., Wilson, R.G. and Zavada, J.M. (1996) Applied Physics Letters, 69, 2083. 587 Dietl, T., Ohno, H., Matsukura, F., Cibert, J. and Ferrand, D. (2000) Zener model description of ferromagnetism in zinc-blende magnetic semiconductors. Science, 287, 1019. 588 Korotkov, R.Y., Gregie, J.M. and Wessels, B.W. (2001) Physica B, 308–310, 30. 589 Korotkov, R.Y., Gregie, J.M. and Wessels, B.W. (2002) Applied Physics Letters, 80, 1731. 590 Graf, T., Gjukic, M., Brandt, M.S., Stutzmann, M. and Ambacher, O. (2002) Applied Physics Letters, 81, 5159. 591 Graf, T., Gjukic, M., Hermann, M., Brandt, M.S., Stutzmann, M., G€orgens, L., Phillip, J.B. and Ambacher, O. (2003) Journal of Applied Physics, 93, 9697. 592 Ashcroft, N.W. and Mermin, N.D. (1976) Solid State Physics, Holt-Reinhart-Winston. 593 Graf, T., Goennenwein, S.T.B. and Brandt, M.S. (2003) Physica Status Solidi b: Basic Research, 239, 277. 594 Kreissl, J., Ulrici, W., El-Metoui, M., Vasson, A.M., Vasson, A. and Gavaix, A.
595
596
597
598
599
600
601
602
603 604
605
(1996) Physical Review B: Condensed Matter, 54, 10508. Schneider, J., Kaufmann, U., Wilkening, W., Baeumler, M. and Kohl, F. (1987) Electronic structure of the neutral manganese acceptor in gallium arsenide. Physical Review Letters, 59, 240–243. Averkiev, N.S., Gutkin, A.A., Kolchanova, N.M. and Reshchikov, M.A. (1984) Fizika i Tekhnika Poluprovodnikov, 18, 1629 (Soviet Physics: Semiconductors, 1984, 18, 1019). Edmonds, K.W., Farley, N.R.S., Johal, T.K., Campion, R.P., Gallagher, B.L., Foxon, C.T. and van der Laan, G. (2004) Journal of Applied Physics, 95, 7166. Wolos, A., Palczewska, M., Zajac, M., Gosk, J., Kaminska, M., Twardowski, A., Bockowski, M., Grzegory, I. and Porowski, S. (2004) Optical and magnetic properties of Mn in bulk GaN. Physical Review B: Condensed Matter, 69, 115210–115216. Wolos, A., Wysmolek, A., Kaminska, M., Twardowski, A., Bockowski, M., Grzegory, I., Porowski, S. and Potemski, M. (2004) Physical Review B: Condensed Matter, 70, 245202. Korotkov, R.Y., Gregie, J.M. and Wessels, B.W. (2001) Materials Research Society Symposium Proceedings, 639, G3.7. Gelhausen, O., Malguth, E., Phillips, M.R., Goldys, E.M., Strassburg, M., Hoffmann, A., Graf, T., Gjukic, M. and Stutzmann, M. (2004) Doping level dependent optical properties of GaN:Mn. Applied Physics Letters, 84, 4514. Han, B., Korotkov, R.Y., Wessels, B.W. and Ulmer, M.P. (2004) Applied Physics Letters, 84, 5320. Han, B., Wessels, B.W. and Ulmer, M.P. (2005) Applied Physics Letters, 86, 042505. Luckovsky, G. (1965) On the photoionization of deep impurity centers in semiconductors. Solid State Communications, 33, 299–302. Gerstmann, U., Blumenau, A.T. and Overhof, H. (2001) Transition metal defects in group-III nitrides: an ab initio calculation of hyperfine interactions and
j 1215
1216
j 4 Extended and Point Defects, Doping, and Magnetism 606
607
608
609
610
611
612
613
614
optical transitions. Physical Review B: Condensed Matter, 63, 075204-1– 075204-9. Polykov, A.Y., Smimov, N.B., Govorkov, A.V., Pashikov, N.Y., Kim, J., Ren, F., Overberg, M.E., Thaler, G., Abermathy, C.R., Pearton, S.J. and Wilson, R.G. (2002) Journal of Applied Physics, 92, 3130. Yoon, I.Y., Kang, T.W., Jeong, M.C., Ham, M.H. and Myoung, J.M. (2004) Applied Physics Letters, 85, 4878. Baur, J., Kaufmann, U., Kunzer, M., Schneider, J., Amano, H., Akasaki, I., Detchprohm, T. and Hiramatsu, K. (1995) Photoluminescence of residual transition metal impurities in GaN. Applied Physics Letters, 67 (8), 1140–1142. Kim, J.J., Makino, H., Takata, Y., Yamamoto, T., Hanada, T., Cho, M.W., Ikenaga, E., Yabashi, M., Miwa, D., Nishino, Y., Tamasaku, K., Ishikawa, T., Shin, S. and Yao, T. (2004) Hybridization of Cr 3d-N 2p-Ga 4s in the wide band-gap diluted magnetic semiconductor Ga1 xCrxN. Physical Review B: Condensed Matter, 70 (16), 161315-1–161315-4. Baur, J., Maier, K., Kunzer, M., Kaufmann, U., Schneider, J., Amano, H., Akasaki, I., Detchprohm, T. and Hiramatsu, K. (1994) Applied Physics Letters, 64, 857. Baur, J., Maier, K., Kunzer, M., Kaufmann, U. and Schneider, J. (1994) Applied Physics Letters, 65, 2211. Baur, J., Kunzer, M., Maier, K., Kaufmann, U.andSchneider,J.(1994)MaterialsScience & Engineering B: Solid State Materials for Advanced Technology, 29, 61. Shon, Y., Kwon, Y.H., Park, Y.S., Yuldashev, Sh.U., Lee, S.J., Park, C.S., Chung, K.J., Yoon, S.J., Kim, H.J., Lee, W.C., Fu, D.J., Kang, T.W., Fan, X.J., Park, Y.J. and Oh, H.T. (2004) Ferromagnetic behavior of p-type GaN epilayer implanted with Fe þ ions. Journal of Applied Physiology, 95 (2), 761–763. Sobotta, H., Neumann, H., Franzheld, R. and Seifert, W. (1992) Physica Status Solidi, 174, 57.
615 Aszodi, G. and Kaufmann, U. (1985) Zeeman spectroscopy of the vanadium luminescence in GaP and GaAs. Physical Review B: Condensed Matter, 32, 7108–7115. 616 Skolnick, M.S., Dean, P.J., Kane, M.J., Uihlein, Ch. Robbins, D.J., Hayes, W., Cockayne, B. and MacEwan, W.R. (1983) Journal of Physics C: Solid State Physics, 16, 767. 617 Heitz, R., Hoffmann, A., Hausmann, B. and Broser, I. (1991) Journal of Luminescence, 48–49, 689. 618 Mahadevan, P. and Zunger, A. (2004) Applied Physics Letters, 85, 2860. 619 Mahadevan, P. and Zunger, A. (2004) Physical Review B: Condensed Matter, 69, 115211. 620 Heitz, R., Maxim, P., Eckey, L., Thurian, P., Hoffmann, A., Broser, I., Pressel, K. and Meyer, B.K. (1997) Physical Review B: Condensed Matter, 55, 4382. 621 Van Schilfgaarde, M. and Myrasov, O.N. (2001) Physical Review B: Condensed Matter, 63, 233205. 622 Langer, J.M., Delerue, C., Lannoo, M. and Heinrich, H. (1988) Physical Review B: Condensed Matter, 38, 7723. 623 Clerjaud, B. (1986) Semiconductor Physics, Current Issues in Solid State Sciences (ed. A.M. Stoneham), Hilger, Bristol, p. 117. 624 Heitz, R., Thurian, P., Loa, I., Eckey, L., Hoffmann, A., Broser, I., Pressel, K., Meyer, B.K. and Mokhov, E.N. (1995) Applied Physics Letters, 67, 2822. 625 Furdyna, J.K. (1988) Diluted magnetic semiconductors. Journal of Applied Physiology, 64 (4), R29–R64. 626 Ohno, H., Munekata, H., Penney, T., von Molnar, S. and Chang, L.L. (1992) Physical Review Letters, 68, 2664. 627 Ohno, H. (1998) Making nonmagnetic semiconductors ferromagnetic. Science, 281, 951–956. 628 See, for example, Kittel, C. (1996) Introduction to Solid State Physics, 7th edn, John Wiley & Sons, Inc.; Blakemore, J.S. (1985) Solid State Physics, 2nd edn,
References
629
630
631
632
633
634
635
636
637
638
Cambridge University Press; Beam, W. (1965) Electronics of Solids, McGraw-Hill. Arrott, A. (1957) Criterion for ferromagnetism from observations of magnetic isotherms. Physical Review, 108, 1394–1396. Kacman, P. (2001) Spin interactions in diluted magnetic semiconductors and magnetic semiconductor structures. Semiconductor Science and Technology, 16, R25–R39. Wolf, S.A., Awschalom, D.D., Buhrman, R.A., Daughton, J.M., von Molnar, S., Roukes, M.L., Chtchelkanova, A.Y. and Treger, D.M. (2001) Spintronics: a spinbased electronics vision for the future. Science, 294 (5546), 1488–1495. c, I., Fabian, J. and Das Sarma, S. Zuti (2004) Spintronics: fundamentals and applications. Reviews of Modern Physics, 76, 323. Gregg, J.F., Petej, I., Jouguelet, E. and Dennis, C. (2002) Spin electronics—a review. Journal of Physics D: Applied Physics, 35, R121–R155. Alvarado, S.F. (1995) Tunneling potential barrier dependence of electron spin polarization. Physical Review Letters, 75, 513–516. Steinmuller, S.J., Trypiniotis, T., Cho, W.S., Hirohata, A., Lew, W.S., Vaz, C.A.F. and Bland, J.A.C. (2004) Highly efficient spin filtering of ballistic electrons. Physical Review B: Condensed Matter, 69, 153309. Hanbicki, A.T., Jonker, B.T., Itskos, G., Kioseoglou, G. and Petrou, A. (2002) Efficient electrical spin injection from a magnetic metal/tunnel barrier contact into a semiconductor. Applied Physics Letters, 80 (7), 1240–1242. Zhai, F., Guo, Y. and Gu, B.-L. (2003) Current and spin-filtering dual diodes based on diluted magnetic semiconductor heterostructures with a nonmagnetic barrier. Journal of Applied Physics, 94, 5432. Andresen, S.E., Sørensen, B.S., Rasmussen, F.B., Lindelof, P.E., Sadowski, J., Guertler, C.M. and Bland,
639 640
641
642 643
644
645
646 647
648
649
J.A.C. (2003) Spin-polarized electron tunneling across a Si delta-doped GaMnAs/n-GaAs interface. Journal of Applied Physics, 94 (6), 3990– 3994. Monsma, D.J., Vlutters, R. and Lodder, J.C. (1998) Science, 281, 407. Kikkawa, J.M. and Awschalom, D.D. (1998) Resonant spin amplification in n-type GaAs. Physical Review Letters, 80, 4313–4316. Koshihara, S., Oiwa, A., Hirasawa, M., Katsumoto, S., Iye, Y., Urano, C., Takagi, H. and Munekata, H. (1997) Ferromagnetic order induced by photogenerated carriers in magnetic III–V semiconductor heterostructures of (In,Mn)As/GaSb. Physical Review Letters, 78, 4617–4620. Prinz, G.A. (1998) Magnetoelectronics. Science, 282, 1660–1663. Soulen, R.J., Jr, Byers, J.M., Osofsky, M.S., Nadgorny, B., Ambrose, T., Cheng, S.F., Broussard, P.R., Tanaka, C.T., Nowak, J., Moodera, J.S., Barry, A. and Coey, J.M.D. (1998) Measuring the spin polarization of a metal with a superconducting point contact. Science, 282, 85. Bowen, M., Bibes, M., Barthelemy, A., Contour, J.-P., Anane, A., Lemaître, Y. and Fert, A. (2003) Nearly total spin polarization in La2/3Sr1/3MnO3 from tunneling experiments. Applied Physics Letters, 82 (2), 233–235. Meservey, R., Tedrow, P.M. and Fulde, P. (1970) Physical Review Letters, 25 (18), 1270. Tedrow, P.M. and Meservey, R. (1971) Physical Review Letters, 26 (4), 192. Julliere, M. (1975) Tunneling between ferromagnetic films. Physics Letters, 54A, 225–226. Moodera, S. and Mathon, G. (1999) Spin polarized tunneling in ferromagnetic junctions. Journal of Magnetism and Magnetic Materials, 200, 248. Slonczewski, J.C. (1989) Conductance and exchange coupling of two ferromagnets separated by a tunneling
j 1217
1218
j 4 Extended and Point Defects, Doping, and Magnetism 650
651
652
653
654 655
656
657
658
659
barrier. Physical Review B: Condensed Matter, 39, 6995. MacLaren, J.M., Zhang, X.-G. and Butler, W.H. (1997) Validity of the Julliere model of spin-dependent tunnelling. Physical Review B: Condensed Matter, 56 (18), 11827–11832. Qi, Y., Xing, D.Y. and Dong, J. (1998) Relation between Julliere and Slonczewski models of tunneling magnetoresistance. Physical Review B: Condensed Matter, 58, 2783–2787. Mathon, J. (2002) Theory of spindependent tunnelling in magnetic junctions. Journal of Physics D: Applied Physics, 35, 2437–2442. Davis, A.H., MacLaren, J.M. and LeClair, P. (2001) Inherent temperature effects in magnetic tunnel junctions. Journal of Applied Physics, 89 (11), 7567–7569. Mott, N.F. (1936) Proceedings of the Royal Society A, 153, 699. Moodera, J.S., Kinder, L.R., Wong, T.M. and Meservey, R. (1995) Large magnetoresistance at room temperature in ferromagnetic thin film tunnel junctions. Physical Review Letters, 74, 3273–3276. Baibich, M.N., Broto, J.M., Fert, A., Nguyen Van Dau, F., Petroff, F., Eitenne, P., Creuzet, G., Friederich, A. and Chazelas, J. (1988) Giant magnetoresistance of (0 0 1)Fe/(0 0 1)Cr magnetic superlattices. Physical Review Letters, 61, 2472. Binasch, G., Gr€ unberg, P., Saurenbach, F. and Zinn, W. (1989) Enhanced magnetoresistance in layered magnetic structures with antiferromagnetic interlayer exchange. Physical Review B: Condensed Matter, 39, 4828. Tehrani, S., Slaughter, J.M., Chen, E., Durlam, M., Shi, J. and DeHerren, M. (1999) Progress and outlook for MRAM technology. IEEE Transactions on Magnetics, 35 (5 Part 1), 2814–2819. Pohm, A.V., Nanotechnology and magnetic memories, in Recent and Evolving Advanced Semiconductor and
660
661
662 663
664
665
666 667
668 669
670
671 672
Organic Nano-technologies, Part 1 (2003), Academic Press (ed. H. Morkoc). Pearton, S.J., Abernathy, C.R., Thaler, G.T., Frazier, R.M., Norton, D.P., Ren, F., Park, Y.D., Zavada, J.M., Buyanova, I.A., Chen, W.M. and Hebard, A.F. (2004) Wide bandgap GaN-based semiconductors for spintronics. Journal of Physics: Condensed Matter, 16, R209. Liu, C., Yun, F. and Morkoc, H. (2005) ZnO and GaN based ferromagnetic materials: a review. Journal of Materials Science B: Materials in Electronics, 16, 555–597. Dietl, T. (2001) Journal of Applied Physics, 89, 7437. Dietl, T., Ohno, H. and Matsukura, F. (2001) Physical Review B: Condensed Matter, 63, 195205. Zener, C. (1951) Interaction between the d-shells in the transitions metals. II. Ferromagnetic compounds of manganese with perovskite structure. Physical Review, 82, 403–405. Matsukura, F., Ohno, H. and Dietl, T. (2002) III–V ferromagnetic semiconductors, in Handbook of Magnetic Materials, vol. 14 (ed. K.H.J. Buschow), Elsevier, Amsterdam, pp. 1–42. Wilamowski, Z. (1990) Acta Physica Polonica A, 77, 133. Szczytko, J., Twardowski, A., Swiatek, K., Palczewska, M., Tanaka, M., Hayashi, T. and Ando, K. (1999) Physical Review B: Condensed Matter, 60, 8304. Sato, K. and Katayama-Yoshida, H. (2000) Japan Journal of Applied Physics, 39, L555. Medvedkin, G.A., Ishibashi, T., Nishi, T., Hayata, K., Hasegawa, Y. and Sato, K. (2000) Japan Journal of Applied Physics, 39, L949. Akai, H. and Dederichs, P.H. (1993) Physical Review B: Condensed Matter, 47, 8739. Akai, H. (1998) Physical Review Letters, 81, 3002. Anderson, P.W. and Hasegawa, H. (1955) Physical Review, 100, 675.
References 673 Story, T., Galazka, R.R., Frankel, R.B. and Wolff, P.A. (1986) Physical Review Letters, 56, 777. 674 Zener, C. (1951) Interaction between the d-shells in the transitions metals. III. Calculation of the Weiss factors in Fe, Co, and Ni. Physical Review, 83, 299–301. 675 Zener, C. (1951) Interaction between the d-shells in the transition metals. Physical Review, 81, 440–444. 676 Berciu, M. and Bhat, R.N. (2001) Effects of disorder on ferromagnetism in diluted magnetic semiconductors. Physical Review Letters, 87, 107203. 677 Hohenberg, P. and Kohn, W. (1964) Physical Review, 136, 864. 678 Kohn, W. and Sham, L.J. (1965) Physical Review, 140, 1133. 679 Berciu, M. and Bhatt, R.N. (2001) Physical Review Letters, 87, 107203. 680 Kaminski, A. and Das Sarma, S. (2001) Physical Review Letters, 88, 247202. 681 Das Sarma, S., Hwang, E.H. and Kaminski, A. (2003) Physical Review B: Condensed Matter, 67, 155201. 682 Warnock, J. and Wolff, P.A. (1985) Physical Review B: Condensed Matter, 31, 6579. 683 Sawicki, M., Dietl, T., Kossut, J., Igalson, J., Wojtowicz, T. and Plesiewicz, W. (1986) Physical Review Letters, 56, 508. 684 Coey, J.M.D., Venkatesan, M. and Fitzgerald, C.B. (2005) Nature Materials, 4, 73. 685 Dietl, T. and Spalek, J. (1982) Physical Review Letters, 48, 355. 686 Ohno, H., Shen, A., Matsukura, F., Oiwa, A., Endo, A., Katsumoto, S. and Iye, Y. (1996) Applied Physics Letters, 69, 363. 687 Litvinov, V.I. and Dugaev, V.K. (2001) Physical Review Letters, 86, 5593. 688 Sato, K. and Katayama-Yoshida, H. (2002) Semiconductor Science and Technology, 17, 367. 689 Katayama-Yoshida, H., Kato, R. and Yamamoto, T. (2001) Journal of Crystal Growth, 231, 428. 690 Sato, K. and Katayama-Yoshida, H. (2001) Japanese Journal of Applied Physics, 40, L485.
691 Van Schilfgaarde, M. and Myrasov, O.N. (2001) Physical Review B: Condensed Matter, 63, 233205. 692 Jungwirth, T., Atkinson, W.A., Lee, B.H. and MacDonald, A.H. (1999) Physical Review B: Condensed Matter, 59, 981. 693 Das, G.P., Rao, B.K. and Jena, P. (2003) Physical Review B: Condensed Matter, 68, 035207. 694 Wang, Q., Sun, Q. and Jena, P. (2004) Physical Review Letters, 93, 155501. 695 Sato, K., Schweika, W., Dederichs, P.H. and Katayama-Yoshida, H. (2004) Physical Review B: Condensed Matter, 70, 201202. 696 Kronik, L., Jain, M. and Chelikowsky, J.R. (2002) Electronic structure of spin polarization of GaxMn1xN. Physical Review B: Condensed Matter, 66, 041203 (R). 697 Kulatov, E., Nakayama, H., Mariette, H., Ohta, H. and Uspenskii, Yu.A. (2002) Physical Review B: Condensed Matter, 66, 045203. 698 Sanyal, B., Bengone, O. and Mirbt, S. (2003) Electronic structure and magnetism of Mn-doped GaN. Physical Review B: Condensed Matter, 68, 205210. 699 Mahadevan, P. and Zunger, A. (2004) Trends in ferromagnetism, hole localization, and accept level depth for Mn substitution in GaN, GaP, GaAs, and GaSb. Applied Physics Letters, 85, 2860. 700 Mahadevan, P. and Zunger, A. (2004) First principles investigation of the assumptions underlying modelHamiltonian approaches to ferromagnetism of 3d impurities in III–V semiconductors. Physical Review B: Condensed Matter, 69, 115211. 701 Dalpian, G.M., Wei, S.-H., Gong, X.G., da Silva, A.J.R. and Fazzio, A. (2006) Phenomenological band structure model of magnetic coupling in semiconductors. Solid State Communications, 138 (7), 353–358; Wei, S.-H., Gong, X.G., Dalpian, G.M. and Wei, S.-H. (2005) Firstprinciples study of Mn-induced local magnetic moments in host semicon-
j 1219
1220
j 4 Extended and Point Defects, Doping, and Magnetism 702
703
704
705 706
707
708
709
710
711
712
713
714
ductors. Physical Review B: Condensed Matter, 71 (14), 144409-1–144409-61. Dalpian, G.M. and Wei, S.-H. (2005) Transition from ferromagnetism to antiferromagnetism in GaMnN. Journal of Applied Physics, 98, 083905. Das, G.P., Rao, B.K. and Jena, P. (2004) Physical Review B: Condensed Matter, 69, 214422. Kim, J.J., Makino, H., Kobayashi, K., Takata, Y., Yamamoto, T., Hanada, T., Cho, M.W., Ikenaga, E., Yabashi, M., Miwa, D., Nishino, Y., Tamasaku, K., Ishikawa, T., Shin, S. and Yao, T. (2004) Physical Review B: Condensed Matter, 70, 161315. Hong, J. and Wu, R.Q. (2003) Physical Review B: Condensed Matter, 68, 233306. Lee, J.S., Lim, J.D., Khim, Z.G., Park, Y.D., Pearton, S.J. and Chu, S.N.G. (2003) Journal of Applied Physics, 93, 4512. Dalpian, G.M. and Wei, S.-H. (2005) Electron-induced stabilization of ferromagnetism in GaGdN. Physical Review B: Condensed Matter, 72 (11), 115201-1–115201-5. Thaler, G., Frazier, R., Gilar, B., Stapleton, J., Davidson, M., Abernathy, C.R., Pearton, S.J. and Segre, C. (2004) Applied Physics Letters, 84, 2578. Seo, S.S.A., Kim, M.W., Lee, Y.S., Noh, T.W., Park, Y.D., Thaler, G.T., Overberg, M.E., Abernathy, C.R. and Pearton, S.J. (2003) Applied Physics Letters, 82, 4749. Dhar, S., Brandt, O., Trampert, A., Friedland, K.J., Sun, Y.J. and Ploog, K.H. (2003) Physical Review B: Condensed Matter, 67, 165205. Matsukura, F., Abe, E., Ohno, Y. and Ohno, H. (2000) Applied Surface Science, 159–160, 265. Dhar, S., Brandt, O., Trampert, A., D€aweritz, L., Friedland, K.J., Ploog, K.H., Keller, J., Beschoten, B. and G€ untherodt, G. (2003) Applied Physics Letters, 82, 2077. Grandidier, B., Nys, J.P., Delerue, C., Stievenard, D., Higo, Y. and Tanaka, M. (2000) Applied Physics Letters, 77, 4001. Yu, K.M., Walukiewicz, W., Wojtowicz, T., Kuryliszyn, I., Liu, X., Sasaki, Y. and
715 716
717 718
719
720
721
722
723
Furdyna, J.K. (2002) Physical Review B: Condensed Matter, 65, 201303. Dietl, T. (2003) Nature Materials, 2, 646. Bunker, B.A., Pong, W.-F., Debska, V., Yoder-Short, D.R. and Furdyna, J.F. (1987) EXAFS determination of bond lengths in Zn1 xMnxSe, in Diluted Magnetic (Semimagnetic) Semiconductors, vol. 89 (eds R.L. Aggarwal, J.K. Furdyna and S. von Molnar), Materials Research Society, Pittsburgh, PA, pp. 231–235; Pong, W.-F., Mayanovich, R.A., Bunker, B.A., Furdyna, J.F. and Debska, U. (1990) Extended X-ray-absorption fine-structure studies of Zn1xMnxSe alloy structure. Physical Review B: Condensed Matter, 41 (12), 8440–8448. Faraday, M. (1846) Philosophical Transactions of the Royal Society, 136, 1. Bartholomew, D.U., Furdyna, J.K. and Ramdas, A.K. (1986) Interband Faraday rotation in diluted magnetic semiconductor Zn1 xMnxTe and Cd1xMnxTe. Physical Review B: Condensed Matter, 34 (10), 6943–6950. Shinagawa, K. (2000) Faraday and Kerr effects in ferromagnets, in Magneto Optics, Springer Series in Solid State Sciences, vol. 128 (eds S. Sugano and N. Kojima), Springer. Szczytko, J., Mac, W., Twardowski, A., Matsukura, F. and Ohno, H. (1999) Physical Review B: Condensed Matter, 59, 12935. Ivchenko, E.L. and Pikus, G.E. (1997) Superlattices and Other Heterostructures, Symmetry and Optical Phenomena, 2nd edn, Springer, New York. Pierce, D.T. and Meier, F. (1976) Photoemission of spin polarized electrons from GaAs. Physical Review B: Condensed Matter, 13, 5484–5500. Dyakonov, M.I. and Perel, V.I. (1984) Optical Orientation, Modern Problems in Condensed Matter Science, vol. 8 (eds F. Meier and B.P. Zakharchenya), North-Holland, Amsterdam, pp. 11–71.
References 724 Vasilev, A.M., Daiminger, F., Straka, J., Forchel, A., Kochereshko, V.P., Sandler, G.L. and Uraltsev, I.N. (1993) Optical orientation of holes and electrons in strained layer InGaAs/GaAs quantumwells. Superlattices and Microstructures, 13, 97–100. 725 Oskotskij, B.D., Subashiev, A.V. and Mamaev, Y.A. (1997) Polarized photoemission spectra of the strained semiconductor layers. Physics of LowDimensional Semiconductor Structures, 1–2, 77–87. 726 Ando, K., Takahashi, K., Okuda, T. and Umehara, M. (1992) Magnetic circular dichroism of zinc-blende-phase MnTe. Physical Review B: Condensed Matter, 46, 12289–12297. 727 Weisbuch, C. and Vinter, B. (1991) Quantum Semiconductor Structure, Academic Press, San Diego, FL. 728 Samarth, N. and Awschalom, D.D. (1998) Quantum Circuits and Devices (eds K. Ismail, S. Bandyopadhyay and J.P. Leburton), Imperial College Press, London, pp. 136–141. 729 Giriat, W. and Furdyna, J.K. Semiconductors and Semimetals, vol. 25 (eds J.K. Furdyna and J. Kossut), Academic Press, pp.1–32. 730 Gunshor, R.L., Kolodziejski, L.A., Nurmikko, A.V. and Otsuka, N. (1998) Annual Review of Materials Science, 18, 325. 731 Oseroff, S.B. (1982) Magnetic susceptibility and EPR measurements in concentrated spin-glasses: Cd1 xMnxTe and Cd1xMnxSe. Physical Review B: Condensed Matter, 25, 6584–6594. 732 Oseroff, S., Mesa, M., Tovar, M. and Arce, R. (1982) Time and field dependence of magnetization in AuFe, AgMn, and ThGd spin glasses. Journal of Applied Physics, 53, 2208. 733 Oestreich, M., H€ ubner, J., H€agele, D., Klar, P.J., Heimbrodt, W., R€ uhle, W.W., Ashenford, D.E. and Lunn, B. (1999) Applied Physics Letters, 74, 1251. 734 Jonker, B.T., Park, Y.D., Bennet, B.R., Cheong, H.D., Kioseoglou, G. and
735
736
737
738
739
740
741
742
743
744
745
Petroff, A. (2000) Physical Review B: Condensed Matter, 62, 8180. Fiederling, R., Klem, M., Reuscher, G., Ossau, W., Schmidt, G., Waag, A. and Molenkamp, L.W. (1999) Nature, 402, 787. Wang, Q., Sun, Q., Jena, P. and Kawazo, Y. (2005) First principles study of ferromagnetic coupling in Zn1 xCrxTe thin film. Journal of Applied Physics, 97 (4), 43904-1–43904-6. Saito, H., Zayets, V., Yamagata, S. and Ando, K. (2002) Magneto-optical studies of ferromagnetism in the II–VI diluted magnetic semiconductor Zn1 xCrxTe. Physical Review B: Condensed Matter, 66, 081201(R). Saito, H., Zayets, V., Yamagata, S. and Ando, K. (2003) Room-temperature ferromagnetism in a II–VI diluted magnetic semiconductor Zn1 xCrxTe. Physical Review Letters, 90, 207202. Saito, H., Yamagata, S. and Ando, K. (2004) Magnetoresistance in a room temperature ferromagnetic diluted magnetic semiconductor Zn1 xCrxTe. Journal of Applied Physics, 95 (11), 7175–7177. Oiwa, A., Katsumoto, S., Endo, A., Hirasawa, M., Iye, Y., Ohno, H., Matsukura, F., Shen, A. and Sugawara, Y. (1997) Solid State Communications, 103, 209. von Molnar, S., Brigges, A., Flouquet, J. and Remenyi, G. (1983) Physical Review Letters, 51, 706. Fukuyama, H. and Yoshida, K. (1979) Journal of the Physical Society of Japan, 46, 102. Wojtowicz, T., Dietl, T., Sawicki, M., Plesiewicz, W. and JaroszyÁski, J. (1986) Physical Review Letters, 56, 2419. Gaj, J.A., Grieshaber, W., Bodin-Deshayes, C., Cibert, J., Feuillet, G., Merle dAubigne, Y. and Wasiela, A. (1994) Magneto-optical study of interface mixing in the CdTe-(Cd, Mn)Te system. Physical Review B: Condensed Matter, 50, 5512–5527. Ando, K. (2000) Magneto-optics in diluted magnetic semiconductors: new materials
j 1221
1222
j 4 Extended and Point Defects, Doping, and Magnetism 746
747
748
749
750
751
752
753
754
and applications, in Magneto Optics, Solid State Sciences Series, vol. 128 (eds S. Sugano and N. Kojima), Springer. Oestreich, M., Bender, M., H€ ubner, J., H€agele, D., R€ uhle, W.W., Hartmann, Th., Klar, P.J., Heimbrodt, W., Lampalzer, M., Volz, K. and Stolz, W. (2002) Superconductor Science and Technology, 17, 285. Waag, A., Schmeusser, S., BicknellTassius, R.N., Yakovlev, D.R., Ossau, W., Landwehr, G. and Uraltsev, I.N. (1991) Molecular beam epitaxial growth of ultrathin CdTe–CdMnTe quantum wells and their characterization. Applied Physics Letters, 59 (23), 2995–2997. Smyth, J.F., Awschalom, D.D., Samarth, N., Luo, H. and Furdyna, J.K. (1992) Physical Review B: Condensed Matter, 46, 4340. Tulchinsky, D.A., Baumberg, J.J., Awschalom, D.D., Samarth, N., Luo, H. and Furdyna, J.K. (1994) Femtosecond spin spectroscopy in magnetically tunable heterostructures. Physical Review B: Condensed Matter, 50, 10851. Flack, F., Samarth, N., Nikitin, V., Crowell, P.A., Shi, J., Levy, J. and Awschalom, D.D. (1996) Near-field optical spectroscopy of localized excitons in strained CdSe quantum dots. Physical Review B: Condensed Matter, 54, R17312. Nikitin, V., Crowell, P.A., Gupta, J.A., Awschalom, D.D., Flack, F. and Samarth, N. (1997) Zero dimensional excitonic confinement in locally strained ZnxCd1xSe. Applied Physics Letters, 71, 1213. Awschalom, D.D., Freeman, M.R., Samarath, N., Luo, H. and Furdyna, J.K. (1991) Physical Review Letters, 66, 1212. Ohno, H. (1998) Making nonmagnetic semiconductor ferromagnetic. Science, 281, 951. Beschoten, B., Crowell, P.A., Malajovich, I., Awschalom, D.D., Matsukura, F., Shen, A. and Ohno, H. (1999) Physical Review Letters, 83, 3073.
755 Malajovich, I., Berry, J.J., Samarth, N. and Awschalom, D.D. (2001) Nature, 411, 770. 756 Edmonds, K.W., Wang, K.Y., Campion, R.P., Neumann, A.C., Farley, N.R.S., Gallagher, B.L. and Foxon, C.T. (2002) Applied Physics Letters, 81, 4991. 757 Ku, K.C., Potashnik, S.J., Chun, R.F., Schiffer, P., Samarth, N., Seong, M.J., Mascarenhas, A., Jonston-Halperin, E., Mayers, R.C., Gossard, A.C. and Awschalom, D.D. (2003) Applied Physics Letters, 82, 2302. 758 Kikkawa, J.M. and Awschalom, D.D. (1998) Resonant spin amplification in n-type GaAs. Physical Review Letters, 80, 4313. 759 Hanson, R., Witkamp, B., Vandersypen, L.M.K., Willems van Beveren, L.H., Elzerman, J.M. and Kouwenhoven, L.P. (2003) Zeeman energy and spin relaxation in a one-electron quantum dot. Physical Review Letters, 91, 196802. 760 Khaetski, A.V. and Nazarov, Y.V. (2000) Spin relaxation in semiconductor quantum dots. Physical Review B: Condensed Matter, 61, 12639. 761 Beschoten, B., Johnson-Halperin, E., Young, D.K., Poggio, M., Grimaldi, J.E., Keller, S., DenBaars, S.P., Mishra, U.K., Hu, E.L. and Awschalom, D.D. (2001) Spin coherence and dephasing in GaN. Physical Review B: Condensed Matter, 63, 121202. 762 Honig, A. (1954) Polarization of arsenic nuclei in a silicon semiconductor. Physical Review, 96, 234–235. 763 Honig, A. and Kip, A.F. (1954) Electron spin resonance of an impurity level in silicon. Physical Review, 95, 1686– 1687. 764 Feher, G. and Gere, E.A. (1959) Electron spin resonance experiments on donors in silicon. II. Electron spin relaxation effects. Physical Review, 114, 1245–1256. 765 Szczytko, J., Twardowski, A., Swiatek, K., Palczewska, M., Tanaka, M., Hayashi, T. and Ando, K. (1999) Physical Review B: Condensed Matter, 60, 8304.
References 766 Okabayashi, J., Kimura, A., Rader, O., Mizokawa, T., Fujimori, A., Hayashi, T. and Tanaka, M. (2001) Physical Review B: Condensed Matter, 64, 125304. 767 Ohno, H. (1999) Journal of Magnetism and Magnetic Materials, 200, 110. 768 Blinkowski, J. and Kacman, P. (2003) Physical Review B: Condensed Matter, 67, 121204(R). 769 Van Esch, A., Van Bockstal, L., De Bocck, J., Verbanck, G., Van Steenbergen, A.S., Wellman, P.J., Grietens, B., Bogaerts, R., Herlach, F. and Borghs, G. (1997) Physical Review B: Condensed Matter, 56, 13103. 770 Ohno, Y., Young, D.K., Beschotem, B., Matsukura, F., Ohno, H. and Awschalom, D.D. (1999) Nature, 402, 790. 771 Matsukura, F., Ohno, H. and Dietl, T. (2002) III–V ferromagnetic semiconductors, in Handbook of Magnetic Materials, vol. 14 (ed. K.H.J. Buschow), Elsevier, Amsterdam, pp. 1–42. 772 Ando, K., Hayashi, T., Tanaka, M. and Twardowski, A. (1998) Magneto-optic effect of the ferromagnetic diluted magnetic semiconductor Ga1 xMnxAs. Journal of Applied Physics, 83 (11), 6548–6550. 773 Kuroiwa, T., Yasuda, T., Matsukura, F., Shen, A., Ohno, Y., Segawa, Y. and Ohno, H. (1998) Electronics Letters, 34, 190. 774 Ando, K., Hayashi, T. and Tanaka, M., Nanoelectronics Research Institute, National Institute of Advanced Industrial Science and Technology (AIST), Tsukuba Central 2, Umezono 1-1-1, Tsukuba, Ibaraki 305-8568, Japan, unpublished. 775 Edmonds, K.W., Wang, K.Y., Campion, R.P., Newmann, A.C., Foxon, C.T., Gallagher, B.L. and Main, P.C. (2002) Applied Physics Letters, 81, 3010. 776 Matsukura, F., Ohno, H., Shen, A. and Sugawara, Y. (1998) Physical Review B: Condensed Matter, 57, 2037. 777 Ohno, H. and Matsukura, F. (2001) Solid State Communications, 117, 179.
778 Liu, C., Yun, F. and Morkoc, H. (2005) Critical review of ferromagnetism in ZnO and GaN. Journal of Materials Science B: Materials in Electronics, 16, 555–5597. 779 Omiya, T., Matsukura, F., Dietl, T., Ohno, Y., Sakun, T., Motokawa, M. and Ohno, H. (2000) Physica E, 7, 976. 780 Reed, M.L., Ritums, M.K., Stadelmaier, H.H., Reed, M.J., Parker, C.A., Bedair, S.M. and El-Masry, N.A. (2001) Room temperature magnetic (Ga,Mn)N: a new material for spin electronic devices. Materials Letters, 51 (6), 500–503. 781 Overberg, M.E., Abernathy, C.R., Pearton, S.J., Theodoropoulou, N.A., McCarthy, K.T. and Hebard, F. (2001) Applied Physics Letters, 79, 1312. 782 Thaler, G.T., Overberg, M.E., Gila, B., Frazier, R., Abemathy, C.R., Pearton, S.J., Lee, J.S., Lee, S.Y., Park, Y.D. and Khim, Z.G. (2002) Applied Physics Letters, 80, 3964. 783 Sasaki, T., Sonoda, S., Yamamoto, Y., Suga, K., Shimizu, S., Kindo, K. and Hori, H. (2002) Journal of Applied Physics, 91, 7911. 784 Reed, M.L., El-Masry, N.A., Stadelmaier, H.H., Ritums, M.K., Reed, M.J., Parker, C.A., Roberts, J.C. and Bedair, S.M. (2001) Room temperature ferromagnetic properties of (Ga,Mn)N. Applied Physics Letters, 79, 3473–3475. 785 Ploog, K.H., Dhar, S. and Trampert, A. (2003) Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 21, 1756. 786 Boselli, M.A., da Cunha Lima, I.C., Leite, J.R., Troper, A. and Ghazali, A. (2004) Applied Physics Letters, 84, 1138. 787 Chitta, V.A., Coaquira, J.A.H., Fernandez, J.R.L., Duarte, C.A., Leite, J.R., Schikora, D., As, D.J., Lischka, K. and Abram, E. (2004) Applied Physics Letters, 85, 3777. 788 Zajac, M., Gosk, J., Grzanka, E., Kami½ska, M., Twardowski, A., Strojek, B., Szyszko, T. and Podsiadøo, S. (2003) Journal of Applied Physics, 93, 4715. 789 Zajac, M., Doradzinski, R., Gosk, J., Szczytko, J., Lefeld-Sosnowska, M.,
j 1223
1224
j 4 Extended and Point Defects, Doping, and Magnetism 790
791
792
793
794
795
796
797
Kaminska, M., Twardowski, A., Palczewska, M., Grzanka, E. and Gebicki, W. (2001) Applied Physics Letters, 78, 1276. Furdyna, J.K. and Kossut, J. (1988) Diluted Magnetic Semiconductors, Academic Press, New York. Ste2pniewski, R., Potemski, M., Wysmolek, A., Pakula, K., Baranowski, J.M., Lusakowski, J., Grzegory, I., Porowski, S., Martinez, G. and Wyder, P. (1999) Physical Review B: Condensed Matter, 60 (7), 4438–4441. Buyanova, I.A., Izadifard, M., Chen, W.M., Kim, J., Ren, F., Thaler, G., Abernathy, C.R., Pearton, S.J., Pan, C.-C., Chen, G.-T., Chyi, J.-I. and Zavada, J.M. (2004) On the origin of spin loss in GaMnN/InGaN light-emitting diodes. Applied Physics Letters, 84, 2599. Tackeuchi, A., Kuroda, T., Shikanai, A., Sota, T., Kuramata, A. and Domen, K. (2000) No spin polarization of carriers in InGaN. Physica E, 7, 1011. Campo, J., Julier, M., Coquillat, D., Lascaray, J.P., Scalbert, D. and Briot, O. (1997) Zeeman splittings of excitonic transitions at the G point in wurtzite GaN: a magnetoreflectance investigation. Physical Review B: Condensed Matter, 56, R7108. Choi, C.K., Lam, J.B., Gainer, G.H., Shee, S.K., Krasinski, J.S., Song, J.J. and Chang, Y.-C. (2002) Polarization dependence of the excitonic optical Stark effect in GaN. Physical Review B: Condensed Matter, 65, 155206. Chen, W.M., Buyanova, I.A., Nishibayashi, K., Kayanuma, K., Seo, K., Murayama, A., Oka, Y., Thaler, G., Frazier, R., Abernathy, C.R., Ren, F., Pearton, S.J., Pan, C.-C., Chen, G.-T. and Chyi, J.-I. (2005) Efficient spin relaxation in InGaN/ GaN and InGaN/GaMnN quantum wells: an obstacle to spin detection. Applied Physics Letters, 87, 192107. Gil, B., Briot, O. and Aulombard, R.L. (1995) Valence band physics and optical properties of GaN epilayers grown onto sapphire with wurtzite symmetry. Physical
798
799
800
801
802
803
804
805
806
807
808
Review B: Condensed Matter, 52, R17028–R17031. Julier, M., Campo, J., Gil, B. and Lascaray, J.P. (1998) Determination of the spinexchange interaction constant in wurtzite GaN. Physical Review B: Condensed Matter, 57, R6791. Gil, B. and Briot, O. (1997) Internal structure and oscillator strengths of excitons in strained a-GaN. Physical Review B: Condensed Matter, 55, 2530. Langer, D.W., Euwena, R.N., Era, K. and Koda, T. (1970) Spin exchange in excitons, the quasicubic model and deformation potentials in II–VI compounds. Physical Review B: Condensed Matter, 2, 4005. Gil, B., Lusson, A., Sallet, V., Said-Hassani, S.A., Triboulet, R. and Bigenwald, P. (2001) Strain-fields effects and reversal of the nature of the fundamental valence band of ZnO epilayers. Japanese Journal of Applied Physics, Part 2: Letters, 40, L1089. Chuang, S.L. and Chang, C.S. (1996) kp method for strained wurtzite semiconductors. Physical Review B: Condensed Matter, 54, 2491. Skromme, B.J. (1997) Optical and magneto-optical characterization of heteroepitaxial gallium nitride. Materials for Advanced Technology, 50, 117. Ando, K. (2003) Magneto-optical studies of s,p–d exchange interactions in GaN:Mn with room-temperature ferromagnetism. Applied Physics Letters, 82 (1), 100–102. Sonada, S., Hori, H., Yamamoto, Y., Sasaki, T., Sato, M., Shimizu, S., Suga, K. and Kindo, K. (2002) IEEE Transactions on Magnetics, 38, 2859. Dingle, R., Sell, D.D., Stokowski, S.E. and Ilegems, M. (1971) Physical Review B: Condensed Matter, 4, 1211. Arciszewska, M. and Nawrocki, M. (1986) Journal of Physics and Chemistry of Solids, 47, 309. Ando, K., Saito, H., Jin, Z., Fukumura, T., Kawasaki, M., Matsumoto, Y. and
References
809
810 811
812
813
814
815
816
817
818
819
Koinuma, H. (2001) Journal of Applied Physics, 89, 7284. See, for example, Kondo, T., Kuwabara, S., Owa, H. and Munekata, H. (2002) Molecular beam epitaxy of (Ga,Mn)N. Journal of Crystal Growth, 237–239 (2), 1353–1357. Ueda, K., Tabata, H. and Kawai, T. (2001) Applied Physics Letters, 79, 988. Wakano, T., Fujimura, N., Morinaga, Y., Abe, N., Ashida, A. and Ito, T. (2001) Physica E, 10, 260. Ando, K. (2001) Magneto-optical study of ZnO based diluted magnetic semiconductors, 7th Symposium on the Physics and Application of Spin-Related Phenomena in Semiconductors (PASPS), December 17–18, Yokohama, cond-mat/ 0208010 which can be accessed via http://arxiv.org/archive/cond-mat. Saito, H., Zaets, W., Akimoto, R., Ando, K., Mishima, Y. and Tanaka, M. (2001) Journal of Applied Physics, 89, 7392. Ando, K. and Munekata, H. (2004) Magneto-optical study of spin-carrier interactions in (In,Mn)As. Journal of Magnetism and Magnetic Materials, 272–276 (Part 3), 2004–2005. Wolos, A., Palczewska, M., Zajac, M., Gosk, J., Kaminska, M., Twardowski, A., Bockowski, M., Grzegory, I. and Porowski, S. (2004) Physical Review B: Condensed Matter, 69, 115210. Cheng, X.M. and Chien, C.L. (2003) Magnetic properties of epitaxial Mn-doped ZnO thin films. Applied Physics Letters, 93 (10), 7876–7879. Reed, M.J., Arkun, F.E., Berkman, E.A., El-Masry, N.A., Zavada, J., Luen, M.O., Reed, M.L. and Bedair, S.M. (2005) Applied Physics Letters, 86, 102504. Arkun, F.E., Reed, M.J., Berkman, E.A., El-Masry, N.A., Zavada, J.M., Reed, M.L. and Bedair, S.M. (2004) Dependence of ferromagnetic properties on carrier transfer at GaMnN/GaN:Mg interface. Applied Physics Letters, 85 (17), 3809–3811. Polykov, A.Y., Smimov, N.B., Govorkov, A.V., Pashikov, N.Y., Kim, J., Ren, F.,
820
821
822
823
824
825
826
827
828
Overberg, M.E., Thaler, G.T., Abermathy, C.R., Pearton, S.J. and Wilson, R.G. (2002) Journal of Applied Physics, 92, 3130. Park, S.E., Lee, H.-J., Cho, Y.C., Jeong, S.-Y., Cho, C.R. and Cho, S. (2002) Applied Physics Letters, 80, 4187. Hashimoto, M., Zhou, Y.-K., Kanakura, M. and Asahi, H. (2002) Solid State Communications, 122, 37. Hasegawa, F., Suemasu, T., Yamaguchi, K. and Ando, K. (2005) Extinction of dream room temperature ferromagnetic semiconductor with GaN, Presented at the Second Asia-Pacific Workshop on Widegap Semiconductors (APWS-2005), March 7–9, Hsinchu Lakeshore Hotel, Taiwan. Yamaguchi, K., Tomioka, H., Yui, T., Suemasu, T., Ando, K., Yoshizaki, R. and Hasegawa, F. (2004) Disagreement between magnetic and magneto-optical properties in Cr-doped GaN films on Si (1 1 1) substrates grown by metal organic molecular beam epitaxy. Japanese Journal of Applied Physics, 43 (10A), L1312–L1314. Yamaguchi, K., Tomioka, H., Yui, T., Suemasu, T., Ando, K., Yoshizaki, R. and Hasegawa, F. (2005) Cr concentration dependence of magnetic and electrical properties of Cr-doped GaN films on Si (1 1 1) by MOMBE. Physica Status Solidi c, 2 (7), 2488–2491. Makino, H., Kim, J.J., Nakayama, T., Muro, T., Kobayashi, K. and Yao, T. (2004) Soft X-ray magnetic circular dichroism of Cr-doped GaN. Current Applied Physics, 4, 615–617. Hashimoto, M., Tanaka, H., Asano, R., Hasegawa, S. and Asahi, H. (2004) Observation of photoluminescence emission in ferromagnetic semiconductor GaCrN. Applied Physics Letters, 84, 4191. Corliss, L.M., Elliott, N. and Hastings, J.M. (1960) Antiferromagnetic structure of CrN. Physical Review, 117, 929–935. Haussermann, U., Viklund, P., Bostrom, M., Norrestam, R. and Simak, S.I. (2001)
j 1225
1226
j 4 Extended and Point Defects, Doping, and Magnetism
829
830
831
832
833
834
835 836
837
838
839
840
841
Bonding and physical properties of Hume–Rothary compounds with PtHg4 structure. Physical Review B: Condensed Matter, 63, 125118. Inumaru, K., Okamoto, H. and Yamanaka, S.J. (2002) Journal of Crystal Growth, 237–239, 2050. Singh, R.K., Wu, S.Y., Liu, H.X., Gu, L., Smith, D.J. and Newman, N. (2005) Applied Physics Letters, 86, 012504. Liu, H.X., Wu, S.Y., Singh, R.K., Gu, L., Smith, D.J., Newman, N., Dilley, M.R., Montes, L. and Simmonds, M.B. (2004) Applied Physics Letters, 85, 4076. Yang, S.G., Pakhomov, A.B., Hung, S.T. and Wong, C.Y. (2002) Applied Physics Letters, 81, 2418. Polyakov, A.Y., Smirnov, N.B., Govorkov, A.V., Pashkova, N.V., Shlensky, A.A., Pearton, S.J., Overberg, M.E., Abernathy, C.R., Zavada, J.M. and Wilson, R.G. (2003) Journal of Applied Physics, 93, 5388. Oiwa, A., Endo, A., Katsumoto, S., Iye, Y., Ohno, H. and Munekata, H. (1999) Physical Review B: Condensed Matter, 59, 5826. Sivan, U., Entin-Wohlman, O. and Imry, Y. (1988) Physical Review Letters, 60, 1566. Zhang, Y. and Sarachik, M.P. (1991) Physical Review B: Condensed Matter, 43, 7212. Teraguchi, N., Suzuki, A., Nanishi, Y., Zhou, Y.-K., Hashimoto, M. and Asahi, H. (2002) Solid State Communications, 122, 651. Maruyama, T., Morishita, S., Kagatsume, H., Nanishi, Y. and Akimoto, K.D (2000) IPAP Conference Series 1: Proceeding of the International Workshop on Nitride Semiconductors, Institute of Pure and Applied Physics, Tokyo, p.482. Dhar, S., Brandt, O., Ramsteiner, M., Sapega, V.E. and Ploog, K.H. (2005) Physical Review Letters, 94, 037205. Junod, P., Menth, A. and Vogt, O. (1969) Physik der Kondensiterten Materie, 8, 323. Consiglio, R., Baker, D.R., Paul, G. and Stanley, H.E. (2003) Physica A, 319, 49.
842 Deepak, F.L., Vanitha, P.V., Govindaraj, A. and Rao, C.N.R. (2003) Chemical Physics Letters, 374, 314. 843 Han, D.S., Parka, J., Rhie, K.W., Kim, S. and Chang, J. (2005) Ferromagnetic Mn-doped GaN nanowires. Applied Physics Letters, 86, 032506. 844 Choi, H., Seong, H., Chang, J., Lee, K., Park, Y., Kim, J., Lee, S., He, R., Kaykendall, T. and Yang, P. (2005) Advanced Materials, 17, 1351. 845 Liu, B., Bando, Y., Tang, C., Golberg, D., Yamaura, K. and Muromachi, E.T. (2005) Chemical Physics Letters, 405, 127. 846 Wang, Q., Sun, Q., Jena, P. and Kawazoe, Y. (2005) Ferromagnetic GaN–Cr nanowires. Nano Letters, 5 (8), 1586–1590. 847 Wang, Q., Sun, Q. and Jena, P. (2005) Ferromagnetism in Mn-doped GaN nanowires. Physical Review Letters, 95, 167202. 848 Stroud, R.M., Hanbicki, A.T., Park, Y.D., Kioseoglou, G., Petukhov, A.G., Jonker, B.T., Itskos, G. and Petrou, A. (2002) Physical Review Letters, 89, 166602. 849 Datta, S. and Das, B. (1990) Applied Physics Letters, 56, 665. 850 Johnson, Mark (1993) Spin accumulation in gold films. Physical Review Letters, 70, 2142. 851 Litvinov, V.I. (2003) Physical Review B: Condensed Matter, 68, 155314. 852 Bychkov, Yu.A. and Rashba, E.I. (1984) Oscillatory effects and the magnetic susceptibility of carriers in inversion layers. Journal of Physics C: Solid State Physics, 17 (33), 6039. 853 Rashba, E.I. (1960) Fizika Tverdogo Tela, 2, 1224 (Soviet Physics: Solid State, 1960, 2, 1109). 854 Hao, S., Zhou, G., Wu, J., Duan, W. and Gu, B.-L. (2004) Physical Review B: Condensed Matter, 69, 113403. 855 Ohno, H., Chiba, D., Matsukura, F., Omiya, T., Abe, E., Dietl, T., Ohno, Y. and Ohtani, K. (2000) Electric-field control of ferromagnetism. Nature, 408, 944. 856 McGuire, J.P., Ciuti, C. and Sham, L.J. (2004) Theory of spin transport induced
References
857
858 859
860
861
862 863
864
865
866
867
868
869
by ferromagnetic proximity on a twodimensional electron gas. Physical Review B: Condensed Matter, 69, 115339. Das, B., Datta, S. and Reifenberger, R. (1990) Physical Review B: Condensed Matter, 41, 8278. Dresselhaus, G. (1955) Physical Review, 100, 580. Nitta, J., Akazaki, T. and Takayanagi, H. (1997) Gate control of spin–orbit interaction in an inverted In0.53Ga0.47As/ In0.52Al0.48As heterostructure. Physical Review Letters, 78 (7), 1335–1338. Engels, G., Lange, J., Schapers, Th. and Luth, H. (1997) Physical Review B: Condensed Matter, 55, 1958. Heersche, H.B., Schapers, Th., Nitta, J. and Takayanagi, H. (2001) Physical Review B: Condensed Matter, 64, 161307. Sato, Y., Kita, T., Gozu, S. and Yamada, S. (2001) Journal of Applied Physics, 89, 8017. Ohno, Y., Young, D.K., Beschoten, B., Matsukura, F., Ohno, H. and Awschalom, D.D. (1999) Nature, 402, 790. Jonker, B.T., Hanbicki, A.T., Park, Y.D., Itskos, G., Furis, M., Kioseoglou, G. and Petrou, A. (2001) Applied Physics Letters, 79, 3098. Jonker, B.T., Park, Y.D., Bennet, B.R., Cheong, H.D., Kioseoglou, G. and Petrou, A. (2000) Physical Review B: Condensed Matter, 62, 8180. Fiederling, R., Keim, M., Reuscher, G., Ossau, W., Schmidt, G., Waag, A. and Molenkamp, L.W. (1999) Injection and detection of a spin-polarized current in a light-emitting diode. Nature, 403, 787–789. Ghosh, S. and Bhattacharya, P. (2002) Surface-emitting spin-polarized In0.4Ga0.6As/GaAs quantum-dot lightemitting diode. Applied Physics Letters, 80, 658. Hanbicki, A.T., Jonker, B.T., Itskos, G., Kioseoglou, G. and Petrou, A. (2002) Applied Physics Letters, 80, 1240. Polyakov, A.Y., Smirnov, N.B., Govorkov, A.V., Kim, J., Ren, F., Overberg, M.E., Thaler, G.T., Abernathy, C.R., Pearton,
870
871
872
873
874
875
876 877 878 879
S.J., Lee, C.-M., Chyi, J.-I., Wilson, R.G. and Zavada, J.M. (2003) Solid-State Electronics, 47, 963. Polyakov, A.Y., Govorkov, A.V., Smirnov, N.B., Pashkova, N.Y., Thaler, G.T., Overberg, M.E., Frazier, R., Abernathy, C.R., Pearton, S.J., Kim, J. and Ren, F. (2002) Journal of Applied Physics, 92, 4989. Polyakov, A.Y., Smirnov, N.B., Govorkov, A.V., Kim, J., Ren, F., Overberg, M.E., Thaler, G.T., Frazier, R.M., Abernathy, C.R., Pearton, S.J., Buyanova, I.A., Rudko, G.Y., Chen, W.M., Pan, C.C., Chen, G.T., Chyi, J.-I. and Zavada, J.M. (2004) Journal of Electronic Materials, 33, 241. Buyanova, I.A., Izadifard, M., Storasta, L., Chen, W.M., Kim, J., Ren, F., Thaler, G., Abernathy, C.R., Pearton, S.J., Pan, C.-C., Chen, G.-T., Chyi, J.-I. and Zavada, J.M. (2004) Journal of Electronic Materials, 33, 467. Buyanova, I.A., Bergman, J.P., Chen, W.M., Thaler, G., Frazier, R., Abernathy, C.R., Pearton, S.J., Kim, J., Ren, F., Kyrychenko, F.V., Stanton, C.J., Pan, C.-C., Chen, G.-T., Chyi, J.-I. and Zavada, J.M. (2004) Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 22, 2668. Ohno, H., Akiba, N., Matsukura, F., Shen, S., Ohtani, K. and Ohno, Y. (1998) Spontaneous splitting of ferromagnetic (Ga,Mn)As valence band observed by resonant tunneling spectroscopy. Applied Physics Letters, 73, 363–365. Chiba, D., Akiba, N., Matsukura, F., Ohno, Y. and Ohno, H. (2000) Magnetoresistance effect and interlayer coupling of (Ga,Mn)As trilayer structures. Applied Physics Letters, 77, 1873. Tanaka, M. and Higo, Y. (2001) Physical Review Letters, 87, 26602. Monsma, D.J., Vlutters, R. and Lodder, J.C. (1998) Science, 281, 407. Sato, R. and Mizushima, K. (2001) Applied Physics Letters, 79, 1157. Tanaka, M. and Higo, Y. (2001) Large tunneling magnetoresistance in
j 1227
1228
j 4 Extended and Point Defects, Doping, and Magnetism 880
881
882
883
884
885
886
887
888
889
890
GaMnAs/AlAs/GaMnAs ferromagnetic semiconductor tunnel junctions. Physical Review Letters, 87, 026602. Higo, Y., Shimizu, H. and Tanaka, M. (2001) Anisotropic tunneling magnetoresistance in GaMnAs/AlAs/ GaMnAs ferromagnetic semiconductor tunnel junctions. Journal of Applied Physics, 89, 6745. Das, B., Datta, S. and Reifenberger, R. (1990) Physical Review B: Condensed Matter, 41, 8278. Lu, J., Shen, B., Tang, N., Chen, D.J., Zhao, H., Liu, D.W., Zhang, R., Shi, Y., Zheng, Y.D., Qiu, Z.J., Gui, Y.S., Zhu, B., Yao, W., Chu, J.H., Hoshino, K. and Arakawa, Y. (2004) Applied Physics Letters, 85, 3125. Brosig, S., Ensslin, K., Warburton, R.J., Nguyen, C., Brar, B., Thomas, M. and Kroemer, H. (1999) Physical Review B: Condensed Matter, 60, R13989. Dresselhaus, G. (1955) Spin–orbit coupling effects in zincblende structures. Physical Review, 100, 580–586. Hikami, S., Larkin, A.I. and Nagaoka, Y. (1980) Progress of Theoretical Physics, 63, 707. Dresselhaus, P.D., Papavassiliou, C.M.A., Wheeler, R.G. and Sacks, R.N. (1992) Physical Review Letters, 68, 106. Chen, G.L., Han, J., Huang, T.T., Datta, S. and Janes, D.B. (1993) Physical Review B: Condensed Matter, 47, 4084. Tsubaki, K., Maeda, N., Saitoh, T. and Kobayashi, N. (2002) Applied Physics Letters, 80, 3126. Lo, I., Tsai, J.K., Yao, W.J., Ho, P.C., Tu, L.-W., Chang, T.C., Elhamri, S., Mitchel, W.C., Hsieh, K.Y., Huang, J.H., Huang, H.L. and Tsai, W.-C. (2002) Physical Review B: Condensed Matter, 65, 161306. Thillosen, N., Sch€apers, Th., Kaluza, N., Hardtdegen, H. and Guzenko, V.A. (2006) Weak antilocalization in a polarizationdoped AlxGa1xN/GaN heterostructure with single subband occupation. Applied Physics Letters, 88, 022111.
891 Schmult, S., Manfra, M.J., Punnoose, A., Sergent, A.M., Baldwin, K.W. and Molnar, R.J. (2006) Large Bychkov–Rashba spin–orbit coupling in high-mobility GaN/AlxGa1xN heterostructures. Physical Review B: Condensed Matter, 74, 033302. 892 Weber, W., Ganichev, S.D., Danilov, S.N., Weiss, D., Prettl, W., Kvon, Z.D., Belkov, V.V., Golub, L.E., Cho, H.-I. and Lee, J.-H. (2005) Applied Physics Letters, 87, 262101. 893 Lo, I., Wang, W.T., Gau, M.H., Tsay, S.F. and Chiang, J.C. (2005) Physical Review B: Condensed Matter, 72, 245329. 894 Tang, N., Shen, B., Wang, M.J., Yang, Z.J., Xu, K., Zhang, G.Y., Chen, D.J., Xia, Y., Shi, Y., Zhang, R. and Zheng, Y.D. (2006) Physical Review B: Condensed Matter, 73, 037301. 895 Schierholz, C., K€ ursten, R., Meier, G., Matsuyama, T. and Merkt, U. (2002) Physica Status Solidi b: Basic Research, 233, 436. 896 Lommer, G., Malcher, F. and R€ossler, U. (1988) Physical Review Letters, 60, 728. 897 Kurdak, Ç., Biyikli, N., Äzgur, à., Morkoc, H. and Litvinov, V.I. (2006) Physical Review B: Condensed Matter, 74, 113308. 898 Beschoten, B., Johnson-Halperin, E., Young, D.K., Poggio, M., Grimaldi, J.E., Keller, S., DenBaars, S.P., Mishra, U.K., Hu, E.L. and Awschalom, D.D. (2001) Physical Review B: Condensed Matter, 63, 121202. 899 Rubin, M., Newman, N., Chan, J.C., Fu, T.C. and Ross, J.T. (1994) Applied Physics Letters, 64, 64. 900 Pearton, S.J., Vartuli, C.B., Zolper, J.C., Yuan, C. and Stall, R.A. (1995) Applied Physics Letters, 67, 1435. 901 Wilson, R.G., Pearson, S.J., Abernathy, C.R. and Zavada, J.M. (1995) Applied Physics Letters, 66, 2238. 902 Wilson, R.G., Vartuli, C.B., Abernathy, C.R., Pearton, S.J. and Zavada, J.M. (1995) Solid State Electronics, 38, 11329.
References 903 Zolper, J.C., Wilson, R.G., Pearton, S.J. and Stall, R.A. (1996) Applied Physics Letters, 68, 1945. 904 Binari, S.C., Rowland, L.B., Kruppa, W., Kelner, G., Doverspike, K. and Gaskill, D.K. (1994) Electronics Letters, 30, 1248. 905 Pearton, S.J., Abernathy, C.R., Wisk, P.W., Hobson, W.S. and Ren, F. (1993) Applied Physics Letters, 63, 2238.
906 Zolper, J.C., Pearton, S.J., Abernathy, C.R. and Vartuli, C.B. (1995) Applied Physics Letters, 66, 3042. 907 Fiorentini, V., Bernardini, F., Bosin, A. and Vanderbilt, D. (1996) Ab initio shallow acceptor levels in gallium nitride. 23rd International Conference on the Physics of Semiconductors, vol. 4, part 4, World Scientific, Singapore, pp. 2877–28780.
j 1229
j1231
Index a ab initio band structure 1101 ab initio calculations 265, 281, 363, 405, 931, 992, 1038, 1091, 1100, 1103 ab initio tight-binding methods 141, 214, 366 ab initio treatments 1089 absorption, see photoreflectance spectra absorption, see reflectance 626 accelerating energy 420 acceptor 1000 – substitutional 1007 acceptor activation ratios 628 acceptor concentration 1026 acceptor-doped GaN 1018 acceptor-like isotropic resonances 1002 acceptor–nitrogen vacancy 936 acoustic–acoustic interactions, see acoustic– optic interactions acoustic–optic interactions 56 activation energy 581, 583, 657, 1025, 1044 ADA complex formation 1026 ADA model 1023, 1025, 1029 adsorbate–surface interaction 411 adsorption isotherm 390 adsorption process 389, 391, 400, 405 adsorption rate 412 AFM stabilization energy 1100 Al deposition 643 AlInN – growth 695 AlGaN – cracking 658 – growth 653 – low temperature interlayer 659 – ordering 668 – p-type growth 666 AlGaN alloy 90 AlGaN films 657
AlGaN layers 90, 91, 245, 265, 483, 485, 546, 602, 653, 656, 658, 659, 664, 670, 672, 699, 1032, 1033 AlInGaN-based heterostructures 99 AlInGaN multiple quantum wells 100 all-electron approach 136 all-optical method 288 Al mole fraction 654 Al-rich conditions 651 AlN – epitaxial relationship to Si (001) 383 – growth 638 – surface reconstruction 642 AlN buffer layer(s) 381, 437, 446, 505, 631, 733 AlN bulk crystals 176 AlN electron effective mass 180 AlN films 643 – growth 409 – molar fraction 276 aluminum-based compounds 99 aluminum mole fraction 655 aluminum nitride 62 aluminum nitride films 653 ammonia incident kinetic energy 440 ammonia-limited conditions 439 amorphous GaN films 394 amorphous material 1004 anion–cation bond length 7 anionic model potential parameters 141 anisotropic energy 1167 – splitting 159 anisotropic growth 732 anisotropic phenomenon 627 anisotropy 1139, 1188 annealing 643, 1018, 1035, 1178, 1189 annealing, see LEEBI treatment annealing temperatures 1190
Handbook of Nitride Semiconductors and Devices. Vol. 1. Hadis Morkoç Copyright Ó 2008 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim ISBN: 978-3-527-40837-5
j Index
1232
annihilation radiation 991 annular dark field imaging (ADF) 674 anodic alumina membranes 737 anodization parameters 504 antibonding site 1036, 1037 antiferromagnetic behavior 1157 antiferromagnetic configuration 1089 antiferromagnetic coupling 1143 antiferromagnetic solids 1133 antiferromagnetic state 1091 antiferromagnetism 1137 antisite defects 928, 934 antisites–Ga 935 antisites–N 935 antistandard 498 antisurfactant 721 – microscopic mechanism 721 a-plane GAN 373, 376, 484, 604, 612, 616 a-plane sapphire 293, 351, 371, 375 Archimedean displacement measurements 78 Arrhenius form 403 Arrhenius plots 452, 581, 964 Arrott plots 1074, 1129, 1131, 1164 arsenide semiconductors 371, 596 asymmetric reflections 605 atmospheric pressure chemical vapor deposition (APCVD) reactor 541 atmospheric pressure two-flow system 721 atom-centered basis functions 137 atomic arrangement 545, 603 atomic force microscopy (AFM) 550, 740, 832 – image 340, 570 atomic layer epitaxy (ALE) 79, 394, 528, 703 atomic sphere approximation (ASA) 136 Auger electron spectroscopy (AES) 69, 639, 341 augmented plane wave (APW) method 135 Austerman–Gehman model 846 Avogadros number 440 axial lattice parameter 7 axial ratio 7 azimuthal quantum number 1069
b backscattered electrons (BSE) 191 ball-and-stick model 837 band anticrossing (BAC) model 203 band-edge emission 193 band diagrams 1034 band edge emission 604 bandgap 100 – bowing parameters 214, 205 – issues 629
– low-temperature(s) 172, 179 – optical energy 627 – semiconductors 55, 92, 956 – variation 703 band structure 132 band tail model 170 band-to-band model 1149 band-to-band transitions 1161 band wave functions 1114 barrier–buffer interface 283 barrier height 271 basal GaN 373 basal plane lattice parameter 7 basal plane stacking faults 853 basal sapphire plane 373 basal-stacking fault 852 Bastards formalism 213 Bastards method 213 BC position 1036 beam equivalent pressure (BEP) 444, 993 biaxial relaxation coefficients 158 biaxial strain 226 biaxial stress 484 biexponential decay function 580 biexponential fits 578 biexponential PL decay curve 547 binary alloys 1010 binding energies 411 binodal 698 Bloch–Floquent theorem 133 Bloch function 135 Bloch wave 134 Blombergen–Rowland interactions 1088 – mechanism 1087 blue band 1042, 1044 blue luminescence (BL) 609 blue-shifted emission 561 blue–violet luminescence band 1045 Bohr magneton 1167 Bohr radius 213 Boltzmann constant 54, 941, 403 Boltzmann factor 956 Boltzmann transport equation 968 bond center (BC) 1016 bonding energies 412 bond rotation angle 368 boron nitride (BN) 361 Bose–Einstein expression 177 boundary-limited transport 326 bound magnetic polarons (BMPs) 1092 bound sheet density 267 bowing parameter 253, 269, 280 Bragg diffraction 419 Bragg mirrors 752
Index Braggs law 10, 463 Bragg spots 456, 457 Bragg–Williams approximation 390 bright field (BF) TEM images 472, 473, 657 Brillouin function 1126, 1139 Brillouin scattering 37 – measurements 63 Brillouin-type magnetization 1054 Brillouin zone 140, 1133, 1136 – boundaries 135 buffer layer(s) 94, 264 built-in electric field 157 built-in electrostatic fields 616 bulk crystal growth 384 bulk elastic coefficients 42 bulk material properties 214 bulk modulus 37, 38, 63 bulk rods 426 bulk SiC crystals 335 burgers circuit 831 burgers vector 289, 336, 473, 539, 571, 578, 614, 824, 823, 826, 832, 836, 837, 839, 843 Burstein–Moss shift 194, 195
c cantilever epitaxy (CE) 558 capacitance transient amplitude 946 capture process 194 carbon contamination 370, 396, 1041 carbon nanotubes (CNT) 735 carrier-ion direct-exchange 1084 carrier-mediated ferromagnetic interaction 1141 carrier–single magnetic ion interaction 1084 cathodoluminescence (CL) 537, 558 – data 177 – images 471 – spectra 1163 cation–anion bond 844 cation–anion distances 105 cationic model potential parameters 141 C–C bond strength 398 Cd-doped GaN 1043 cell internal parameter 105 central processing unit (CPU) 140 chalcopyrite-like (CH) structure 246, 254 charge-coupled device (CCD) 425 charge–spin coupling 1076 chemical etching behavior 845 chemical inertness 371 chemical mechanisms 653 chemical vapor deposition (CVD) furnace 751 chemical vapor phase epitaxy technique 1015
circular polarization 1110, 1116, 1117 cladding layer thickness 584 c-lattice parameter 6, 7, 36, 494 cleaving process 371 CL emission bands 1056 closed cycle cryostat 607 closed-shell interaction 139 coalescence boundaries 548 coalescence interface 551 coaxial heterostructures 755 codoping method 1018 coefficients of thermal expansion (CTE) 292, 483 collision-free growth mode 431 commercial LEC GaAs 331 common-cation rule 192 common-cation semiconductors 192 compensation model 1025 complexes 928 – Ga-vacancy 936 – H 937 – N-vacancy 936 compliant substrate 327, 328 composition-dependent bowing parameter 95 composition-induced reduction 211 composition pulling effect 94 composition, see bandgap 100 conduction band 216, 1114, 1128 conduction band Hamiltonian 1179 conduction band minimum (CBM) 143, 208, 219, 871 conduction band wave functions 147 conductive AFM (CAFM) 621, 912 confinement energy 220 confocal micro-Raman scattering 637 constant field approximation 288, 289 constant pressure (Cp) 60 continuum elastic theory 560 conventional Czochralsky melt pulling method 359 conventional semiconductor technology 465, 606 convergent beam electron diffraction (CBED) 363, 472, 587, 819 cooldown-induced thermal mismatch strain 260 – effect 260 corrugation maxima 644, 695 – hexagonal array 644 Coulomb exchange interaction 1084 Coulomb gap 1162 Coulombic defect 974 Coulombic potential profiles 945
j1233
j Index
1234
Coulombic repulsion 591 Coulomb interaction(s) 987, 1123 Coulomb repulsion 982 Coulomb tail 931 covalent semiconductors 1038 coverage-dependent function 412 c-plane sapphire 374 c-plane strain 484 crack-assisted pendeo-epitaxy (CAPE) 547 cracked-ammonia flux 444 crack-free AlGaN layers 659 crack-free samples 470 crack-free thick GaN 511 cross-sectional mapping 561 cross-sectional SEM image 530, 542, 626 – GaN layers 556 – SEM micrograph 546 cross-sectional TEM image 519, 551, 586, 620, 535, 668 cross-sectional transmission electron microscopy (XTEM) 486, 489, 631, 744 – micrograph 543 – observation 551, 662 crystal diffraction 418 crystal field 148, 1050 – parameters 1149 – splitting 148, 152, 1058, 1187 crystal-growth behavior 845 crystal-growth direction 1117 crystal-growth experiments 362 crystal-growth technique 323 crystalline GaN particles 723 crystalline grains 522 crystalline SiC 334 crystalline structure 514 crystallographic anisotropy 616 crystallographic directions 517 crystallographic orientation 531 crystallographic planes 529, 531 crystal-plane spacing 519 crystal symmetry 372 cubic crystal field 1061 cubic magnetocrystalline anisotropy 1184 cubic system 161 CuPt (CP) structure 246, 254 Curie temperature 1066, 1107, 1120, 1126, 1129, 1133, 1139, 1143, 1144, 1156, 1160, 1163, 1164, 1167, 1177, 1184, 1187 Curie–Weiss law 1126, 1160, 1073 Curie–Weiss-type temperature dependencies 1126, 1151 current–voltage characteristics 746 current–voltage measurements 575, 749 CVD-based catalytic 733
cyclotron measurements 169 Czochralski method 294, 332, 355, 358
d DAP transitions 1043 dashed curve 56 deBroglie wavelength 212 Debye expression 60 Debye length 264 Debye temperature 54, 55 Debye theory 79 decomposition 416 – processes 400 deep-level carrier emission 939 deep-level defects 194, 1041 deep-level emission 193 deep-level impurities 205 deep-level optical spectroscopy (DLOS) 956, 1041 deep-level transient Fourier spectroscopy (DLTFS) 963 deep-level transient spectroscopic measurements 818 deep-level transient spectroscopy 938, 939, 581 defect(s) – affected by hydrogen 924 – analysis by chemical etching 905 – analysis by deep-level transient spectroscopy 938 – analysis by surface probes 910 – role of hydrogen 1004 – V-type in InGaN quantum well 905 – V-type induced by Mg doping 894, 895, 897 deformation potentials 165 deionized water 329 d-electrons 139 delocalized metallic Ga–Ga bonds 449 density functional calculation 206 density functional theory (DFT) 137, 140, 819, 820 depolarization 1172, 1185 desorption 405, 412 desorption fluxes 406 desorption mass spectroscopy (DMS) 439, 443 desorption processes 389, 391 desorption spectra 51 d-hybridized upper valence band 139 diamagnetic component 1153, 1157 diamagnetic materials 1064 diamagnetism 1064 diamond-lattice structure 332 diatomic 399
Index dichloromethane 470 dielectric constant 18, 20, 21, 22, 27, 29, 32, 34, 74, 89, 90 dielectric function 20,22 dielectric mask 529 dielectric screening function 141 diethylchlorine gallium 539 differential polarization 285 differential temperature 292 diffusion-controlled processs 50 diffusion length 403 dilute GaAsN films 105, 330 dilute magnetic semiconductor(s) 1082, 1063,1123 – II-VI and GaAs based 1123 – II-VI based 1124 – ab initio theories 1088 – applications 1168 – comparison of magnetic interactions 1101 – GaAs based 1133 – RKKY theory 1088 – theoretical aspects 1082, 1141 – nanowires 1167 – Zener theory 1088 dilute nitrides 105, 202, 210 dimethylhydrazine 512 direct magnetization measurements 1157 dislocations – 5/7 chain 828 – charge profile 874 – edge 821, 822, 823 – edge and screw 821, 823 – electronic structure 882 – mixed 823 – open core-5/7 atom ring 869 – open core-eight atom ring 869 dislocation density 394, 526 dislocation density, see electron mobility dislocation–dislocation interaction 820 dislocation-mediated surface morphology 460 dislocation reduction 557, 578 displacement vector D 264 distributed Bragg reflector (DBR) 97, 661 divacancy 934 DLTS – applied to GaN 948 – applied to ternaries of GaN 977 – basics 939 – dispersion in GaN data 970 DLTS method 938, 948 DLTS peaks 949 DLTS signal 944, 945 DLTS spectra 965
DMS heterostructures 1126, 1127, 1177 DMS materials 1116 DMS systems 1090 domain boundaries 495, 844 – annihilation 641 dominant compensating donor 937 donors – shallow 1007 donor–acceptor pair band 604, 1056 donor-binding energies 611 donor-bound excitons (DBE) 514, 610, 624 donor electrons 1101 donor–gallium vacancy complexes 936 doping 1006 – by point defects 917 – co-doping for p-type 1018 – magnesium 1014 – p-type 1013, 1014 doping with As 1044 doping with Be 1038 doping with C 1040 doping with Ca 1042 doping with Cd 1043 doping with Ge 1011 doping with Hg 1040 doping with P 1045 doping with rare earths 1045, 1046 doping with Se 1012 doping with Si 1010 doping with Si, Ge, O 1010 doping with transition metals 1046, 1060 doping with Zn 1042 Doppler broadening experiments 991 Doppler broadening spectroscopy 988 double-pulse DLTS 946, 974 double exchange interaction 1088 double exchange mechanism 1167 double positioning boundaries (DPB) 496, 818, 844 down-spin channels 1079 down-spin electrons 1080 dry processing techniques 329 d shell electrons 1048 dual-flow channel reactor 666 dual-interface heterostructures 132 dynamical theory 423
e Eagle Picher sample 601 e-beam evaporation 577, 722 EBIC measurements 978 EBIC method 981 ECR microwave plasma-assisted MBE (ECR-MBE) 450
j1235
j Index
1236
edge dislocations 656, 824 – electronic structure 866 effective acceptor activation energies 1034 effective mass approximation 270, 272, 281 effective mass parameters 143 eight-atom core model 825, 837 eight-atom ring character 825 Einstein equation 403, 413 elastic constants 37, 132 elastic deformation 328 elastic moduli 23,78 elastic recoil detection analysis (ERDA) 1005 elastic recoil detection method 1055 elastic stiffness coefficients 63, 156, 159 elastic stiffness constants 224 electric field effect 972 electric field strength 954 electric force microscopy 587, 910 electrical spin injection, see spin orientation electro-optic 383 electroluminescence 1176, 1178 electroluminescence (EL) experiment 1148 electroluminescence (EL) images 744 electrolytic etching technique 35 electromagnetic theory 20 electron acoustic phonon scattering 191 electron beam incident 417 electron beam induced current (EBIC) 977 electron beam lithography 541 electron-counting rule (ECR) 649 electron concentration 525 electron cyclotron resonance microwave plasma assisted molecular beam epitaxy (ECR-MBE) 410 electron density 1038 electron diffraction intensities 423 electron dipole motion 1115 electron effective mass 197, 205, 210 electronegativity 66 electron–electron interaction 133 electron energy loss spectroscopy (EELS) 48 electron–hole pairs 288, 673 electron–hole recombination 279 electron-initiated excitation 559 electron–ion interaction 648 electron-irradiated undoped GaN 1000 electron mobility 526, 632, 966 electron mobility, see dislocation density 966 electron momentum 989 electron momentum distributions 992 electron–nuclear hyperne interaction 1001 electron paramagnetic resonance 998 1134 electron–phonon coupling 84, 955, 1043
electron–phonon interaction 39, 173, 193, 194 electron polarization 1115, 1116 electron spin coherence 1186 electron spin relaxation time 1178 electron spin resonance (ESR) 172, 1052 electron transport 84 electronic band structure 131 electronic degeneracy entropy 973 electronic devices 977 electronic energy levels 1105 electronic paramagnetism 1065 electronic Raman measurements 39 electronic states 1038 electronic structures 1091 electronic thermal conductivity 53 electronic wave functions 133 electrooptic modulators 743 ELO 559 – stripes along the <1 1 _2 0> direction in HVPE ELO 561 – three-step 557 – two-step 547 ELO process 622, 625 ELO technique 623 empirical pseudopotential approach 141 empirical pseudopotential method (EPM) 151, 199 empirical tight binding method (ETBM) 151 energy dispersive X-ray (EDX) 688 – analysis 191 – spectroscopy 744 energy-filtered TEM 1162 energy-loss spectroscopy (EELS) measurements 837 energy momentum 218 – dispersion diagram 218 energy transition site 414 enthalpy 50 envelope function approximation 219 environmental transmission electron microscope 728 epitaxial AlN films 639 epitaxial casting strategy 756 epitaxial core–sheath method 756 epitaxial deposition techniques 323 epitaxial film(s) 329, 448, 527, 597 epitaxial GaN 447, 495 epitaxial GaN film(s) 509, 522 epitaxial GaN layer(s) 487, 498 epitaxial growth temperature 525, 526, 603, 641 – effect 525 epitaxial lateral overgrowth (ELO) 6, 59, 324, 483, 511, 528, 1190
Index – HVPE 537 – maskless 557 – Nano-ELO 564 – point defects 558 – selective 537 – SiN and nanonets 569 – stripes along the <1 _1 0 0> direction in HVPE – W masks 583 epitaxial lateral over-growth on Si 539 – a-plane GaN 616 – pendeo-epitaxy 540 – pendeo-epitaxy on Si 544 – pendeo-epitaxy on SiC 542 epitaxial layer(s) 72, 259, 291, 327, 328, 382, 469, 474, 478, 482, 493, 841 epitaxial nitride layers 395 epitaxial process 527 epitaxial relationship 293, 372, 377 epitaxy 353, 369 EPR 998 EPR measurements 1003 equation of state (EOS) 37 equilibrium bond lengths 681 equilibrium growth 406 etch pitch density (EPD) 331, 528 etch pit delineation process 576 etching methods 552 evaporation coefficients 449 Ewald construction 417 Ewald sphere 419 excimer laser deposition technique 469, 599 excitation-induced carriers 283 excitation-power-density 1161 excitation probability 1114 exciton-binding energies 227, 611 excitonic emission 1012 excitonic spectrum 610 excitonic transitions 555 exciton spin scattering 1133 exothermic nature 413 expanded equilibrium vapor pressure data 51 ex-situ preparation 370 ex-situ processes 369 Extended defects – electronic structure 863 extended X-ray absorption fine structure (EXAFS) 933 external quantum efficiency (EQE) 624 extraordinary Hall effect (EHE) 1141
f face-centered cubic sublattices 144, 611 Faraday effect 1105, 1118, 1127, 1186
Faraday rotation 1104, 1105, 1106, 1107, 1108, 1118, 1119 Fermi–Dirac distribution 941 Fermi–Dirac function 196, 197 Fermi energy 1080, 1104, 1132, 1134, 1141 Fermi level position 189, 929, 1153 Fermi wave vector 1186 ferromagnetic behavior 1153 ferromagnetic coupling 1128 ferromagnetic-diluted semiconductor materials 1187 ferromagnetic electrodes 1079, 1172 ferromagnetic exchange coupling 1184 ferromagnetic film 1078, 1139 ferromagnetic (FM) configuration 1100 ferromagnetic GaN 1107, 1109 ferromagnetic interaction 1100, 1171 ferromagnetic layers 1180 ferromagnetic materials 1067, 1080 ferromagnetic metals 1078, 1131, 1168, 1179 ferromagnetic regime 1109 ferromagnetic samples 1071 ferromagnetic semiconductors 1076 ferromagnetic state 1070, 1072, 1101 ferromagnetic transition 1093, 1133 – temperature 1134, 1143 ferromagnetism 1123, 1130, 1134, 1136, 1144, 1150, 1152, 1156, 1158, 1160, 1163, 1164, 1167, 1169, 1171, 1178, 1187, 1188 field-effect transistors 326, 508, 1010 field emission (FE) SEM image 749 filled-core screw dislocations 832 finite difference time domain (FDTD) 742 finite-element method 538 first-order approximation 197 first-order desorption kinetics 402 first-order kinetic model 450 first-order LEED spot 645 first-order phonon Raman scattering 40 first-order Raman modes 738 first-principles calculations 166, 224 first-principles techniques 140 Float Zone (FZ) method 332 FM state 1121 forbidden energy regions 132 four hydrogen atoms 926 Fourier components 424 Fourier filtering 822 Fourier series 424 Fourier transform analysis 962 Fourier-transform infrared-absorption spectroscopy 1036 Fourier transform infrared (FTIR) 998 – measurements 998
j1237
j Index
1238
four probe 54 F parameter 172 fractional ionic character (FIC) 12 Franks screw dislocation 732 – mechanism 732 Frank-van der Merve (FM) 415 free-carrier charge 272 free-carrier density 279 free carrier induced field 279 free-carrier plasma 89 free-carrier screening 270, 278 free-electron approach 132 free-electron approximation 133 free-electron concentration 87 free electron energy 135 free electron model 134 free electron picture 1080 free-electron recombination band (FERB) model 194 free electrons 929 free energy 536 free enthalpy 550 free excitons (FE) 610 free positron wave 988 freestanding bulk layer 580 FTIR 998 full-core screw dislocation 828, 831 full potential linearized augmented plane wave (FP-LAPW) method 138 full potential LMTO (FP-LMTO) calculations 137 full width half maximum (FWHM) 64, 178, 364
g GaAs 329, 1120 – surface preparation 331 GaAs analogy 388 GaAs-based superlattices 264 GaAs model 390 GaAs spacer layers 1182 GaAs substrate 332 GaAs system 174 GaAs technology 477 Ga-bilayer model 414, 596 Ga-containing alloys 697 Ga desorption mechanism 401 Ga-droplet growth regime 455 Ga-droplet regime 458 Ga-face profile 477 Ga-face sense polarity 624 Ga-filled-core model 831 Ga-filled-core screw dislocation 831 GaInAsN 210
GaInPN 212 gallium antisite 935 gallium arsenide (GaAs) 477 gallium interstitial 934 gallium nitride 30, 49 gallium nitride epitaxy 500 Ga melt 363 GaN 35, 47, 56, 61, 361, 362, 365, 372, 392, 398, 416, 438, 466, 506, 528, 608, 719, 1177 – alloy/multiple layer nucleation layers on SiC 491 – doped with Cr 1156 – doped with Gd 1163 – doped with rare earths 1163 – doped with V 1163 – epitaxial relationship to a-plane sapphire 375 – epitaxial relationship to c-plane sapphire 374 – epitaxial relationship to LiGaO2 and LiAlO2, and pervoskites 382 – epitaxial relationship to r-plane sapphire 376 – epitaxial relationship to sapphire 373 – epitaxial relationship to Si 381 – epitaxial relationship to SiC 381 – heteroepitaxial deposition 528 – high temperature nucleation layers on SiC 486, 489 – low temperature interlayers 584 – low temperature nucleation layers on SiC 488 – nucleation layers on SiC 482 – pertinent surfaces, c, a, m planes 365 – seeded growth 363 – seedless growth 362 GaN-AlN – nucleation layers on SiC by MBE 492 GaN and AlN – epitaxial relationship to sapphire 372 GaN-based electronic devices 62 GaN-based optical devices 62 GaN-based semiconductors 1190 GaN buffer layer(s) 446, 598, 636, 638 GaN-carbon composite nanotubes 758 GaN crystalline quality 404 GaN crystals 361, 362, 364 GaN decomposition 527 GaN deposition 382, 386 GaN desorption 399 GaN epitaxial layers 372, 492, 606, 817 GaN epitaxial relationship 372 GaN equivalent flux 440 GaN film orientation 624
Index GaN films 372, 374, 399, 407, 409, 491, 545, 569 GaN formation 417 GaN growth a-plane 613 GaN growth m-plane 623 GaN growth on GaAs 477 GaN growth on porous SiC (PSiC) 503 GaN growth on sapphire 512 – effect of III/V ratio on nucleation layer 523 – effect of pressure on nucleation layer 525 – low temperature nucleation layers 513 GaN growth on GaN templates 605 GaN growth on LiGaO2 and LiAlO2 603 GaN growth on non-c-plane substrates 611 GaN growth on Si 507 GaN growth on SiC 479, 499 GaN growth on spinel (MgAl2O4) 611 GaN growth on ZnO 598 GaN growth p-type 627 GaN growth technology 462 GaN kinetic model 398 GaN lattice parameter 47 GaN layers 384, 435, 456, 586 – structure 586 GaN low-temperature buffer layer 508 GaN on SiC – double positioning boundaries (SMB) 496 – inversion domain boundaries (IDB) 496 – stacking mismatch boundaries (SMB) 496 – substrate misorientation and domain boundaries 495 – polarity 498 GaN optical phonon energies 42 GaN phase diagram 61 GaN phase growth rate 657 GaN platelets 363 GaN-related materials 818 GaN stacking faults and epitaxial relationship to SiC 480 GaN stripes 531 GaN-substrate interface 287 GaN surface 370 – T and H sites 366, 368 GaN surface preparation 369 GaN synthesis reaction 399 GaN synthesis temperature 52 GaN templates 361, 363, 605 Ga-polar films 587, 591 Ga polarity 234, 381 – on sapphire 586, 597 Ga-polarity films 439, 598 Ga-polarity GaN 588 Ga polarity sample(s) 390, 500, 509 Ga-polarity surfaces 273, 365
Ga-polar matrix 587 Ga-polar sample 222 Ga-polar surface 363, 365, 402 Ga-polar wings 620 GaN QD arrays 714 Ga-rich conditions 367, 407, 457, 831, 935 Ga-rich dislocation core model 834 Ga-rich growth regime 455 Ga-rich regime 414, 460 Ga-rich structure 594 Ga solution 384 Ga sticking coefficient 440 Ga-substitutional acceptors 1007 Ga vacancy concentration 997 Ga-vacancy configuration energy 932 Ga-vacancy model 366 Ga vacancy-oxygen complexes 990 GaPN 210 gaseous species 388 gas manifolds 400 gas-phase diffusion rate 400 gas-phase etching 329 gas-phase product 396 gas-phase reactions 389, 488 gas-phase species 390 gas pressure technique 49, 52 gas-source MBE (GSMBE) 442 gas-source molecular beam epitaxy 210 Gaussian-like distribution 170 Gd-doped GaN samples 1164 Ge doping 1011 generalized gradient approximation (GGA) 9, 1091 generalized quasi chemical approximation (GQCA) 698 geometric theory 418 giant magnetoresistance (GMR) 1080 Gibbs free energy 49, 50, 431, 947 grain boundary (GB) 862 green luminescence band 1043 Greens function theory 140 Greens method 421 ground-state energy 283 growth chamber 369 growth kinetics 700, 993 growth of nitrides 462 growth phase 391 growth process 392 growth rate anisotropy 362 growth-related parameters 672
h half-order diffraction 594 halide precursor techniques 385
j1239
j Index
1240
Hall data 1139 Hall effect 87, 197, 1001, 1103, 1104, 1141, 1143, 1144 – measurements 59, 81, 91, 198, 475, 489, 666, 697, 1040, 1191 Hall electron concentration 197 Hall mobility 632 Hall resistance 1171 Hamiltonian matrix 213 Hamiltonian operator 135 Hanscom sample 601 Hartree-Fock method 140 Hartree-Fock (H-F) theory 137 HCl etching techniques 479 heat of evaporation 49 heat of sublimation 17, 49 heavy-hole bands 207, 1110, 1117 heavy hole (hh) effective masses 173 heavy-hole excitations 1127 heavy-hole subbands 1115 heavy-hole transition 1115 He-Cd laser 607 Heisenberg spin glass 1144 Hermann-Mauguin notation 2, 62 H-etched samples 340 heteroepitaxial film 327, 328, 606 heteroepitaxy 822 heterointerface 565 heterojunction bipolar transistors 110 heterojunction field effect transistor (HFET) 35 heterojunctions 384 heterostructure deposition systems 393 heterostructures 264, 425, 1125, 1179, 118, 1186 hexagonal cells 484 hexagonal close packed (hcp) 2, 143, 426 – sublattices 2, 143 – wurtzitic structures 839 hexagonal compact packed (hcp) structure 819 hexagonal GaN 331, 612 hexagonal lattice structure 374 hexagonal morphology 523, 526 hexagonal pits 458 hexagonal SiC substrates 381 hexagonal stacking 498 hexagonal structure 481 hexagonal symmetry 293, 643 hexagonal wurtzite structure 641 Hg-acceptor level 1040 Hg implantation 1043 high-density crack network 658 high electron momentum 990
high energy electron diffraction 631 high-energy O peak 370 high hydrostatic pressures 929 high-mist systems 327 high-mobility GaN 446 high nitrogen flux rate 100 high nitrogen pressure solution growth (HNPSG) 362 high-performance electronic devices 652 high-performance optical emitters 286 high piezoelectric constants 326 high power field-effect transistors 334 high-pressure anvil method 52 high-pressure apparatus 358 high-pressure growth 527 high-pressure method 365 high-pressure techniques 36 high-quality compound semiconductor growth 416 high-quality epilayers 630 high-quality epitaxy 632 high-quality freestanding GaN template 54 high-quality GaN films 468, 654 high-quality OMVPE 635 high-quality semiconductor films 387 high-quality single crystalline material 77 high-quality thin films 366 High-quality ZnO substrates 353 high-resistivity GaN 1039, 1041 high-resistivity material 627, 628 high-resistivity SiC substrates 327 high-resistivity substrates 334 high-resolution electron microscopy 850 high-resolution image 822 high-resolution TEM analysis 1167 high-resolution top-view lattice image 841 high-resolution transmission electron microscopy (HREM) 819 high-resolution X-ray diffraction (HRXRD) 474 – measurements 42 high-resolution X-ray diffractometry 658 high-resolution Z-contrast imaging 825 high-temperature AlN nucleation layers 486 high-temperature annealing technique 346 high-temperature ferromagnetism 1162 high-temperature GaN buffer layers 488, 489, 590 high thermal conductivity 62 hillocklike feature 58 HNPSG Method 363 hole concentration 1015, 1018, 1025, 1026, 1041, 1042
Index hole effective masses 132, 150, 159, 164, 168, 171, 173, 174, 175, 182, 184, 197, 202, 216, 225, 288, 708, 1025 hole exchange interaction 1153 hole wave functions 1179 hollow core screw dislocations 336 hollow pipes 840 Holt inversion domains 845 Holt-type IDB 846, 850 homoepitaxial AlN films 178 homoepitaxial growth 369, 456 homogeneous biaxial stress 158 homogeneous elastic strain 327 homogeneous emission 561 homogeneously mixed 387 homogeneous strain 638 honeycomb silicate adlayer 341 Hookes law 154, 156, 158 hopping conductivity 1131 horizontal dislocations 559 Hunds rule 1047, 1046, 1048, 1124 HVPE-grown films 393 HVPE-grown samples 834 hybridization 1047, 1060, 1128, 1188 – constant 1085 hydrazine 436 hydride vapor phase epitaxy (HVPE) 36, 323, 326, 385, 387, 537, 818 – buffer layers 468, 836, 1061 – freestanding templates 468 – growth mechanism 388 – initial nucleation layer 388 – reactor 625, 997 – technique 599, 654 – template 472, 834 hydrocarbons 1005 – efficient desorption 1005 hydrogen 393, 397, 1004, 1005 – experimental results 1005 – role 1004 hydrogenation process 334 hydrostatic pressure coefficients 42, 64 hyperfine interaction tensor 1000 hypothetical epitaxy 846 hysteresis 1167, 1171, 1172 – loops 1143, 1158, 1163 hysteretic behavior 1131
i image dislocation 550 image-processing system 425 imperial expression 169 impinging species 516 implantation energy 986
InAlGaN – spinoidal decomposition 705 – quaternary 99 InAlN – alloy 97 – miscibility gap 697 – spinoidal decomposition 697, 698 InAsN 208 indirect gap semiconductor 210 indium-silicon codoping 666 indium doping 527 indium nitride 75, 77 inductively coupled plasma (ICP) 428 inferior carrier transport 453 infrared (IR) 432 – measurements 39 – region 695 – spectroscopic ellipsometry 470 InGaAlN – quaternary, growth 699 InGaAs quantum well 211, 1135 InGaN – alloys 92 – composition pulling effect 93 – doping 678 – growth 671 – layers 672 – phase separation 479 – surface reconstruction 689 – V-defects 674 InGaN/GaN quantum wells 286 inhomogeneous strain distribution 519, 682 injection laser experiments 371 InN – growth 629 in-plane components 218 in-plane current 1173 in-plane deformation potentials 160 in-plane heavy-hole mass 225 in-plane homogeneity 470 in-plane lattice mismatch 505 in-plane light-hole mass 225 in-plane rotation 334 in-plane strain 158, 278 in-plane strain anisotropy 155 in-plane stress-strain relationship 158 InPN 209 input fluxes 406 InSbN 209 in situ annealing 577, 597 in situ buffer morphology 482 in situ methods 402, 477 in situ preparation 370 in situ RHEED images 590
j1241
j Index
1242
intentional doping 1006 interaction between magnetic ions 1086 interfacial relationship 480 interlayer-induced strain 659 interlocking spiral ramps 456 internal cell parameter 103 internal magnetic induction 1064 internal parameter 6, 7, 103, 105, 107, 140, 156, 233, 240, 241, 251, 285 interstitial antibonding site 1038 interstitials 928, 934 – Ga 934 – N 935 intracenter absorption band 1054 intramagnetic ion levels 1071 inverse susceptibility 1126 inversion domains 818 inversion domain boundaries (IDBs) 324, 819, 844, 845, 901 – Holt-type 844, 847 – induced by Mg doping 900 – V-type, IBD 845, 847 inversion-type planar defects 844 inverted pyramids 457 ion-assisted reactive 641 ion beam assisted deposition (IBAD) 24 ionic bond 416 ionic core-core repulsion 820 ionic model potentials 141 ion implantation 1189, 1190 ion-ion interactions 1090 ionization energy 1034, 1040 ion scattering spectroscopy 587 IR ellipsometry measurements 190 island growth mode 415 isoelectronic impurities 1009, 1044 isothermal capacitance transient analysis 954 isothermal capacitance transient spectroscopy (ICTS) 945 isothermal remanent magnetization 1144 isotropic parabolic conduction band 160
j Jahn–Teller effect 999, 1061 Jahn–Teller model 1062
k Kane model 197 Kelvin microscopic investigation 563 Kerr effect 1104 Kerr rotation (KR) 1108 kinematical scattering theory 419 kinematic theory 423 kinetic considerations 392
kinetic energy 136 kinetic model 390, 405 kinetic parameters 436 kinetic theory 53 Knoop diamond indenter 63 Knoops hardness 15, 16, 23, 30, 63 Knudsen effusion cells 410 Koster-Slater model 947 k.p calculations 199 k.p model 143 k.p theory 160 Kronig-Penney approach 133
l Langer-Heinrich rule 1062 Langmuir evaporation data 449 large angle convergent beam electron diffraction (LACBED) 619 large bandgap bowing parameters 106 large stress fields 926 large thermal conductivity 334 laser-assisted CVD 394, 638 laser diode (LD) 696, 1014 laser-induced liftoff 472 laser liftoff (LLO) 462, 468, 471, 607 lasers employing films 371 lateral diffusion barrier 414 lateral dislocations 550 lateral epitaxial overgrowth (LEO) 324, 539, 546, 583 lateral growth 394 laterally contracted bilayer (LCB) 596 lateral propagation 543 lattice constants 102, 202, 351 lattice-matched AlInGaN alloy 265 lattice-matched barrier 97 lattice-matched conditions 62 lattice-matched system 710 lattice matching composition 262 lattice misfit 295 – strain 382 lattice-mismatched film 328 lattice-mismatched substrates 324, 326, 377, 752 lattice mismatch (lm) 185, 375 – component 290 – problem 328 lattice nuclei 1003 – hyperfine interaction 1003 lattice parameter 382, 494 lattice parameter mismatch 294 lattice specific heat 53 Lau condition 419, 424 Lau rings 420
Index Lau treatment 419 layer-plus-island growth mode 415 LEEBI treatment 1016 LEED patterns 597 Leibfried-Schloman scaling parameter 79 Lenzs law 1065 Levenberg-Marquardt algorithm 580 LiAlO2 358 LiGaO2 355 light-hole subbands 1115 light-emitting devices 717 light-emitting diodes (LEDs) 616, 696, 817, 1014, 1077 light-hole bands 207, 1117 light-hole dispersion 209 linear combination of atomic orbitals (LCAO) 37, 133, 819 linear interpolation 282 linear muffin-tin orbitals method (LMTO) 78, 136, 139 linear polarization 284 linear system 161 linear thermal expansion coefficients 79 linearized augmented plan wave (LAPW) method 136, 1100, 1156 line-of-sight quadrupole mass spectrometry 401, 402 liquid encapsulated Czochralski (LEC) method 330 lithium gallate (LGO) 355, 603 lithographical step 547 lithographic/etching techniques 736 lithography 469 local density approximation (LDA) 9, 137 local density formalism (LDF) 138 local density functional theory 648 local probe analysis 563 local strain modification 551 local vibrational modes (LVMs) 1036 long haul fiber-based communications systems 211 longitudinal heterostructures 737 longitudinal optical (LO) phonon frequencies 20 long-period corrugation 644 long-period structure 644 long-range Coulomb interactions 820 long-wavelength lasers 110 LO phonon-plasmon coupled modes (LPP) 561, 563 Lorentz force 1065, 1131 Lorentzian peaks 64 low-energy crystal 411
low-energy electron beam irradiation (LEEBI) 628, 1014 low-energy electron diffraction (LEED) 341, 639 low-energy ion-assisted growth 641 low-energy state 1069 lower thermal conductivity materials 55 lower valence bands (LVBs) 628, 1017 lowest intensity atomic columns 836 low free-electron concentration 47 low-lying conduction electron state 219 low pressure (LP) 584 low-pressure Hg lamp 178 low substrate temperature buffer growth 330, 590 low-temperature buffer growth 590 low-temperature CL image 662 low-temperature data 177 low-temperature deposition processes 467 low-temperature free-exciton transition 177 low-temperature GaN buffer layer 478 low-temperature interlayer 584 low-temperature nitridation 590 low-temperature nucleation buffer layers 386, 467, 514 low-temperature photoluminescence mapping 470 low-temperature process 734 LT AlN insertion buffer layers 585 LT insertion layer 658 Luckovsky fit 1056 luminescence 1110, 1116 – band 1163 – efficiency 558 – images 471 – line 1060 luminescent lifetime 1060 Luttinger-Kohn model 215 Luttinger parameters 152, 161, 173, 174, 175, 183, 201 luzonite-like (LZ) structure 246, 254 Lyddane-Sach-Teller relationship 22
m magnesium doping 1014 magnetic anisotropy 1129 magnetic circular dichroism (MCD) 1104, 1109, 1119 – measurements 1158 magnetic coupling strength 1177 magnetic dopant 1103 magnetic electrons 1088, 1090 magnetic field 998, 1064, 1071, 1084, 1116, 1135, 1139, 1147, 1158, 1180, 1182, 1184
j1243
j Index
1244
magnetic field effect transistors (M-FETs) 1123 magnetic ions 1067, 1074, 1076, 1082, 1083, 1084, 1085, 1086 magnetic ion spin 1084 magnetic moment(s) 1063, 1066, 1068, 1069, 1070, 1079, 1080, 1082, 1133 magnetic polarization 1064 magnetic polaron binding energy 1132 magnetic properties – Mn-doped GaN 1143 magnetic quantum dot (MQD) 1123 magnetic random access memory (MRAM) 1081 magnetic-recording industry 1080 magnetic resonance 999 – methods 1000 – optical detection 999 magnetic semiconductor 1075, 1078 magnetic sensor heads 1078 magnetic susceptibility 1066, 1126 magnetic tunnel junctions (MTJs) 1078 magnetization 1108, 1118, 1119, 1126, 1139, 1164, 1167, 1168, 1171, 1173, 1176, 1182, 1188 – curves 1154 – data 1160 – hysteresis 1146 – measurements 1128, 1158, 1163, 1167 magneto-optical data 1139 magneto-optical effect 1107 magneto optical Kerr effect (MOKE) 1104, 1108 magneto-optical measurements 1133, 1157 magneto optical properties – TM doped GaN 1146 magneto-optical spectroscopy 1146, 1148 magneto-optics studies 1128 magneto-photoluminescence 1127 – measurements 1164 magneto electrical measurements 1103 magneto reflectivity data 1148 magneto reflectivity measurements 1147 magnetoresistance 1139, 1131, 1162 – measurements 1104, 1186 – effect 1180 magneto transport 607 – measurements 1103, 1168 magnetron sputter deposition 641 magnitude 420 maskless epitaxial lateral overgrowth 557 mass flow controllers (MFC) 408 mass spectrometer 49, 438, 447, 462 mass spectroscopic techniques 398
mass spectroscopy experiments 432 mass transport 391 material-specific parameters 212 maximum electron mobility 635 maximum growth rate (GR) 330 Maxwell-Garnett approximation 191 MBE 409, 460, 733 – adsorption 411 – buffer layers 597 – decomposition 416 – effect of III/V ratio on growth 455 – experiments 993, 1017 – GAN films 460, 832 – incorporation 415 – MBE growth – N species for growth 451 – PAMBE growth 435, 446 – reactive ion 435 – RF N source 430 – RF N species 431 – RMBE growth 435, 437 – surface diffusion 413 – surface reconstruction 426, 427, 428, 429 – UHV system 369 Magnetic Circular Dichroism (MCD) 1109 – data 1131, 1160 – effect 1110 – intensity 1130 – measurements 1129, 1132, 1133 – signal 1132 – spectra 1128, 1158 mean atomic column peak intensity 836 mean free path 56 mechanical chemical polish (MCP) 338 melting temperature 49, 68 membrane-based method 736 metallic adlayer structures 651 metallic gallium 387 metallic layers 1180 metalorganic chemical vapor deposition 385 metal-organic MBE (MOMBE) 637 metalorganic precursor 397 metalorganic pyrolysis 396 metalorganic vapor phase epitaxy 385 metal-rich growth 647 metal-semiconductor contacts 365 metastable growth process 416, 437 metastable zinc blende GaN epitaxial films 330 Mg-augmented ELO 536 Mg-doped GaN 1000, 1015, 1034, 1035, 1036 Mg-doped p-type GaN layer 753 Mg-doped samples 985, 991 M-H hysteresis loops 1143, 1144, 1156, 1163
Index microcathodoluminescence studies 76, 190 micro-Raman scattering 198 microwave-excited nitrogen 636 Mie resonances 76, 187, 191 Mie resonant absorption 86 Mie theory 191 migration-enhanced epitaxy (MEE) 404, 632 Miller-Bravais indices 343, 372 Miller index planes 531 minority carrier lifetime 977 minority carrier transient spectroscopy (MCTS) 963 miscibility gap – InAlN 697 misfit accommodation (MA) 327 misfit dislocations 822 misfit strain 293, 295 mist-induced piezoelectric polarization 260 mixed anion host materials 212 mixed cation ternary host materials 212 mixed dislocations 836, 841 – electronic structure 866 Mn complexes 1060 Mn-doped ferromagnetic GaAs 1103, 1135 Mn-doped GaN 1048 model dielectric matrix 141 moderate-temperature nucleation layers 514 MODFET structures 268 modulation doped field effect transistor (MODFET) 35 modulation-doped structures 169, 264 molar fraction 259 – function 259 molecular beam epitaxy (MBE) 323, 409, 450, 824 molecular cohesive energy 410 molecular stream epitaxy (MSE) 677 molecular weight 35 MOMBE systems 1045 monatomic interstitial hydrogen 924 monatomic nitrogen 438 monochromatic CL image 559 monocrystalline 605 monolayer units 412 monomethylsilane 1012 monovacancy defects 992 Monte Carlo calculations 83 Monte Carlo simulations 1144 Moss-Burstein effect 86, 186, 188, 189 Moss-Burstein shift 75, 76, 186, 187, 189 Mott critical density 189 Motts transition 197 – concentration 197
multibeam optical stress sensor (MOSS) system 491 multiheterolayer buffer structures 299 multiple interface heterostructures 278 multiple layer nucleation layers 491 multiple quantum disk (MQD) nanocolumn 753 multiple quantum well (MQW) 608 Murnaghans EOS 37
n
nanocrystalline GaN thin films 723 nanoheteroepitaxy (NHE) 503, 564 nanoimprint lithography 569 nanopipes 818, 840, 841 nanostructures 725 – coaxial heterostructures 755 – core multishell structures 744, 746 – FETs 746 – multiple quantum disk LEDs 753 – nanowires and heterostructures 737 – preparation 725 – properties 725 – synthesis 726 – synthesis by confined chemical reactions 735 – synthesis by template based-methods 735 – synthesis by self catalytic vapor-liquid-solid (VLS) process 730 – synthesis by vapor-liquid-solid process 727 – synthesis by vapor phase 726 – vapor-solid process 732 nanotubes 756 nanowire(s) 737 – growth 734 – optical emission 743 – stimulated emission 744 – synthesis in solution 734 nanowire FETs 749 native defects 929, 1000 N-beam case 424 N-containing gases 653 N-deficient layers 409 N-doped GaP 210 near field scanning optical microcopy (NSOM) 617 near-infrared photoluminescence 1044 negative-U system 1004 negative ion defects 995 neighbor bond lengths 105 N-face sense polarity 624 nitridation process 349, 513, 597 nitride-based optoelectronic devices 1015
j1245
j Index
1246
nitride-based semiconductors 75 nitride-based ultraviolet detectors 652 nitride crystal growth 361 nitride epitaxy 324 nitride family 75 nitride film 396 nitride growth techniques 384 nitride heterostructures 270 nitride semiconductor alloys 102, 269 nitride semiconductor family 102 nitride semiconductors 47, 131, 140, 323, 359, 383, 395 nitrides 1, 39, 851 nitrogen antisite 935 nitrogen concentrations 109 nitrogen dissociation pressure 49 nitrogen flux 454 nitrogen incorporation efficiency 440 nitrogen-limited growth 454 nitrogen precursor 585 nitrogen pressure 50 nitrogen-rich conditions 407 nitrogen-rich growth regime 693 nitrogen species 452 nitrogen vacancies 930, 931, 932 N-limited growth conditions 448 nonisomorphic substrates 496 nonlinear core corrections 139 nonlinearity of polarization 245 nonlinear optical properties 641 nonlinear optics properties 383 nonlinear polarization 276 nonmagnetic ion 1132 nonmagnetic materials 1078 nonmagnetic semiconductors 1064 nonplasma-based growth techniques 435 nonpolar surface 223 novel method 638 N-plasma conditions 634 N-polar films 591, 635 N-polarity 234 – films 598 – samples 266, 453 N-polar layer 994 N-polar surface 364, 365 N-polar wings 620 N-rich conditions 367, 450, 594 N-rich growth regime 455 N-rich regime 460 NSOM measurements 620 N-type GaN platelets 364 nucleation buffer layer 479, 480, 482, 492, 523 nuclei formation 527
o octahedral sites 611 ODMR method 998, 1000 Omega rocking curves 614 OMVPE – Ga and N precursor adsorption and desorption 400 – Ga and N precursor adsorption and desorption, activation energy 401 – Ga and N surface diffusion 403 – GaN desorption as it relates to growth 398 – growth mechanism 398 – kinetic model – balance between adsorption and desorption 405 – low-temperature nucleation buffer layers 513 – reactors 393 – samples 581, 628, 948, 952, 984, 1061 – systems 395, 408, 583 – technique 393, 636 OMVPE-grown films 1035 OMVPE-grown layers 1017 OMVPE-sources 467 one-dimensional growth mechanism 738 one-dimensional stress 292 one-dimensional system 269 one-electron model 132 one-electron picture 133 one-electron wave functions 137 one-monolayer adsorption model 390 open-core edge 841 open-core screw dislocation 828, 831, 840 optical absorption coefficient 71 optical absorption measurements 76, 95 optical absorption technique 191 optical beam 1109 optical detection of magnetic resonance 998 optical devices 977 optical emission 704 – spectra 432 optical emitters 629 optical excitation 1110 optically detected electron paramagnetic resonance (ODEPR) 935 optically pumped ultraviolet-blue stimulated emission 741 optical microscopy 459 optical processes 600 optical spectrum 492 optical surface reflectometry 521 optical transitions 1060, 1110, 1163 optical transmission measurements 95 optical transmission spectra 1057 optimal growth temperature 699
Index optimum stripe orientation 621 optoelectronic devices 225, 384, 623, 1034 orbital angular momentum 1072 order–disorder phase transitions 588, 594 organic chemical vapor deposition 684, 733 organometallic precursor gases 395 organometallic source 468, 730 organometallic vapor phase epitaxy (OMVPE) 323, 385, 393, 825 470, 565 orthogonalized plane wave (OPW) method 134 orthorhombic cell 382, 383 orthorhombic structure 382 out-of-plane component 218 out-of-plane deformation potentials 160 out-of-plane lattice constant 463 out-of-plane rocking curves 465 out-of-plane sound velocities 57 out-of-plane strain 276 out-of-plane stress 155 oxidation state 1050 oxygen-bonded silicon 341 oxygen doping 1013 oxygen-terminated surface 356
p parabolic band structure 216 parabolic conduction band 216 paramagnetic materials 1157, 1064 paramagnetic regime 1109 paramagnetism 1004 parametric variations 387 parasitic gas-phase chemical reactions 653 Paulings electronegativity 138 p-doping 667 peak electric field 946 peak electron drift velocity 83 peak transition energy 616 pendeo-epitaxial film 543 pendeo-epitaxial growth 540, 542 pendeo-epitaxial phenomenon 542 pendeo-epitaxy 507, 540, 541, 621 periodic stacking faults 852 perovskite cell 383 perovskite oxide(s) 296, 382 – substrates 295 phase separation 93 phase shift 1120 phonon dispersion curves 41 phonon energies 20 phonon frequency 20 phonon-phonon scattering 57, 79 phonon-phonon Umklapp scattering 53 photoelastic effect 1120
photoelastic modulator (PEM) 1120 photoelectron spectroscopy 208 photoemission experiments 139 photoemission methods 271 Photoemission studies 1134 photoemission transient capacitance methods 961, 976 photoexcitation 1116 photo-generated carriers 1077 photoionization spectroscopy 1041 photolithography step 557 photolithography techniques 530 photoluminescence 288, 407, 462, 635, 1018 – data 578 – efficiency 598 – experiments 185, 930 – intensity 615 – measurements 483, 604, 637, 1040 – spectra 435, 489, 587, 602, 606, 608 photomodulated transmission 95 photon energy resonates 961 photoreflectance (PR) spectra 87, 626 piezo component 276 piezoelectric (PE) coefficients 132, 180, 641 piezoelectric charge 262 piezoelectric constants 240, 259 piezoelectric effects 95, 182, 215, 588, 1185 piezoelectric fields 1179 – scattering carriers 679 piezoelectric material 948 piezoelectric polarization 157, 236, 259, 263, 289 – calculation 259 – effects 132 – non linearities 256 – properties 62 – transducer 1120 piezoelectric tensor 237 piezo-induced polarization 278 pillarlike interfacial layer 327 pin photodiodes 661 planar defects 844 plane wave expansion method 134 plane wave pseudopotential (PWPP) 78 Planks constant 942 plasma-assisted MBE (PAMBE) 428, 657 plasma-assisted OMVPE 636 plasma-enhanced CVD 394 plasma excitation 513 plastic deformation 327 plastic relaxation 483 PL bands 1061 PL emission 1161 PLE spectrum 1056
j1247
j Index
1248
PL intensity 1055 PL spectra 1056 PL transition 1054 point defects theory 919 Poisson effect 155 Poisson equation 270 Poisson ratio 484, 565 Poissons equation 266, 269, 271 Poissons ratio 37, 38, 158, 292, 483, 826 polarization charge 221, 267, 276, 1173 – Ga-polarity single AlGaN/GaN interface 272 – Ga-polarity single AlxIn1-xN/GaN Interface 276 polarization charge density 266 polarization difference 284 polarization effects 131, 230 polarization gradient 269 polarization-induced charge 221, 286, 287, 288, 615 polarization-induced electric field 282, 622, 626 polarization-induced field 283, 753 polarization-induced interface charge 266 polarization-induced interface sheet density 286 polarization-induced red shift 288 polarization-induced surface 266 polarization in heterostructures 264 polarization in quantum wells 278 – nonlinear 280 polarization vector 1115 polarized light 1109, 1110, 1119, 1176, 1177 polarized light emission 1173, 1178, 1179 polycrystalline film 600 polycrystalline growth mode 630 polymeric GaN 731 polymeric product 399 Poole-Frenkel barrier 954 Poole-Frenkel effect 939, 945, 947, 948, 950, 955, 972, 973 porous SiC (PSiC) substrates 503 porous silicon (PS) 509 positive polarization charge 273 positron annihilation experiments 982, 983, 984, 992 positron annihilation spectroscopy (PAS) 933 positron-electron momentum distribution 982 positron lifetime experiments 983, 989 positron lifetime spectroscopy 995 positrons wave function 1038 postgrowth annealing 109, 110, 993 postgrowth evaluation 503
postimplant annealing treatment 1163 powder technique 77 Powells model 441 precursor-mediated model 449 precursor-mediated pathway 438 precursor flux 400 predigital thermometers 296 pre-epitaxy surface preparation 334 preexponential factor 399 pressure coefficient 14 prismatic growth 516 prismatic planes 533, 568 probe-size-related effect 620 process pressure 525 proton implantation 1190 proton irradiation 963 pseudo-hydrogen atoms 649 pseudomorphic temperature 631 pseudopotential calculations 138, 139 pseudopotential method 132, 135 pseudopotential plane wave calculations 42 pseudopotential plane wave method 38 p-type doping 1006, 1007, 1013, 1035 – role of hydrogen 1034 p-type doping with superllatices 1032 p-type GaN 627, 1163 p-type semiconductor 272 p-type impurity atoms 92 pulsed laser ablation/chemical vapor deposition (PLA-CVD) process 739 pulsed positron beam 991 pyramidal planar defects 818 pyramidal structures 559 pyroelectric effects 1185 pyrolysis temperatures 512 pyrolytic decomposition 653 pyrolytic process 653
q QD-like structures 722 quadratic equation 103 quadratic nonlinearity 280 quadrupole mass spectrometer 51 quantitative in situ method 401 quantum-confined Stark effect(QCSE) 264, 286, 611616 quantum dots 706, 707, 711, 723 – colloidal formation 723 – growth 706 – holy grail 707 quantum dots by MBE 712 quantum dots by OMVPE 719 quantum efficiency 612 quantum mechanical treatments 1065
Index quantum mechanics 218 quantum size effect 186 quantum well (QW) structures 90, 132, 213, 221, 215, 279, 280, 673, 706, 1125, 1149, 1179 – thickness 222 quasi-cubic approximation 166, 167, 174 quasi-cubic model 160, 168, 171, 179, 182 quasi-Fermi level 197 quasi-linear temperature gradient 361 quasi-particle approach 141 quasi-particle band structure energies 141 quasi-particle excitations 139 quaternary – InGaAlN 699 quaternary alloys 89, 100, 699, 652 – growth of 652
r radiation-induced defects 190 radiative recombination lifetime 573 radiative transitions 196, 1177 radio frequency plasma excited (RF-MBE) 98, 405, 630 Raman-active optical phonon modes 64 Raman frequencies 559 Raman measurements 483 Raman mode 39, 563 Raman scattering 40, 613, 695 – data 559 – experiments 193, 561 – measurements 61, 64 Raman spectroscopy 42, 79, 563 – measurements 563 Ramsdel notation 2-4 Rare earths 1046 Rashba coupling 1173 – mechanism 1186 Rashba–Sheka–Pikus (RSP) 214 Rashba spin–orbit coupling effect 1169 Rashba spin splitting 1186 RBS-channeling 1161 reactive ion etching (RIE) methods 511, 751, 966 reactive ion molecular beam epitaxy (RIMBE) 72, 435, 441 reactive molecular beam epitaxy (RMBE) 326, 963 rectangular cross section facet 533 reduced recombination efficiency 612 reduced screw dislocations 586 reflectance spectra 177, 1135 reflectance spectroscopy 84
reflection high-energy electron diffraction (RHEED) 6, 341, 417 – intensity 700 – patterns 633 – secular beam 700 refractive index 27, 32, 34, 73, 74, 89, 1120 relative transition probabilities 1114 RF-activated nitrogen 405 RF magnetron sputtering (RF MS) 98, 637, 638 RF plasma-assisted MBE 993 RF plasma source 452 RF-sputtered film 187 RKKY theories 1187 RKKY-type interactions 1090 reactive molecular beam epitaxy (RMBE) 436, 437 RMBE-grown films 1039 rock salt 1, 2, 14, 28, 36, 62 room-temperature absorption coefficient 73 room-temperature electron mobility 87 room-temperature ferromagnetic semiconductors 1134 room-temperature Hall measurements 605 room-temperature thermal conductivity 59 root mean square (rms) 196, 346 r-plane sapphire 293, 326, 373, 377 Russel–Saunders coupling 1046 Rutherford backscattering spectrometry (RBS) 91, 187, 188, 700, 704, 1162 – experiment 1037
s sacrificial buffer layers 469 Samsung Advanced Institute of Technology 468 Samsung freestanding template 1003 Samsung template 477, 1002, 1003 sandwich method 392 Sapphire 342, 343, 344, 346, 388, 452, 512, 709 – stacking order 351 – surface preparation 346 – unit cell 344 sapphire planes 372 sapphire substrates 72, 87, 293, 327, 382, 665, 752 satellite spots 646 saturation magnetization 1066 scanning capacitance microscopy (SCM) 910 scanning electron microscope (SEM) 978 – images 445 scanning thermal microscopy (SThM) 57
j1249
j Index
1250
scanning transmission electron microscope (STEM) 836 scanning tunneling microscopy (STM) 412, 588, 642, 691 – probes 691 scanning tunneling spectroscopy (STS) 367 scattering amplitude 422 Schmidt orthogonalization condition 135 Schoenflies notation 2 Schottky barrier 271, 621 – ultraviolet detectors 508 Schottky characteristics 575 Schrodinger–Poisson equations 265, 282 Schrodinger–Poisson solver 272, 281 Schrodingers equation 136, 266, 215, 219, 270, 420 screw burgers vectors 836 screw dislocation(s) 336, 550, 571, 823, 826, 828 – component 488 – full core 831, 865 – open core 831, 865 screening field 244 s–d exchange interaction 1117, 1126 SdH measurements 1186 secondary ion mass spectroscopy (SIMS) 91, 475, 985 second lithographic process 551 second-order elastic moduli 78 second-order perturbation theory 160, 172 Se doping electron concentrations 1012 segmented silicon lattice planes 567 selected area diffraction (SAD) 668 selective area epitaxy (SAE) 535 selective area growth (SAG) 583 selective epitaxial growth 529, 537 self-catalytic process 731 self catalytic vapor–liquid–solid (VLS) process 730 self-compensation process 1005 self-consistent field method 137 semiconductor-ferromagnetic material interface 1169 semiconductor lasers 216 semiconductor quantum dots (QDs) 706 semiconductors 369 semiconductor substrates 330 semiconductor technology 1006 semiempirical pseudopotential calculations 199 semi-insulating SiC 336 semi-insulating (SI) GaN 1044 semimetallic overlap 205 seven-atom ring core 837
shallow acceptors 1010, 1017 shallow donor acceptor 507 shallow donors 1007 sharper superimposed peak 522 sheet carrier concentration 268, 287 sheet carrier density 278, 653 short-range electron-hole spin exchange interaction 1149 Shubnikov-de Haas data 169 Shubnikov-de Haas measurements 169 Si 332 – surface preparation 333 Si-based electronics 507 SiC 334, 335 – epitaxial layers 503 – hydrogen surface etching 340 – stacking sequence 335 – surface preparation 338 SiC mechanical polishing techniques 479 Si-doped GaN layer 585 Si doping 1010 silica-based fiber dispersion 110 silicon-on-insulator (SOI) 509, 565 silicon substrates 544 SIMS 1005 SIMS analyses 1041 SIMS technique 475 single-crystal AlN films 396 single-crystal diffraction 462 single-crystal epitaxial thin films 73 single-crystalline AlN 638 single crystalline GaN 624 single-crystalline structure 741 single-electron spin transistors 1123 single-interface heterostructures 276 single magnetic ion interaction 1084 single-molecule precursors (SMPs) 740 single quantum well (SQW) 286, 559 – structure 286 single-step ELO technology 535 single-step substrate 496 single-walled carbon nanotubes (SWCNs) 749 Slater-Koster parameters 141 SMB threading defect 498 smooth coalesced films 558 smooth surface morphologies 634 solar-blind detectors 638, 652 solar-blind region 492 solar-blind UV photoconductors 661 solar cells 211 solid-liquid interface 729 space group 143 space grouping 2
Index spatial modulation 1169 sp-d exchange 1125 sp-d exchange interaction 1127, 1129 sp-d spin exchange interaction 1118, 1127 specific heat 17, 25, 29, 31, 53, 60, 61, 67, 68, 79, 80, 81 spectral position 998 spectroscopic ellipsometry 170, 177, 288 specular epitaxial films 523 specular reflection spot 417 sphalerite 2,4 spin angular momentum 1072 spin coherence length 1172 spin coherence times 1134, 1168 spin-dependent kinetic exchange interaction 1087 spin-dependent transport 1077 spin diffusion length 1184 spin-down channel 1100 spin-down electrons 1005, 1070, 1078, 1104, 1109 spin-down states 1115, 1116 spin exchange interaction 1085, 1117, 1137, 1147 spin FETs 1171, 1188 spin-freezing temperature 1144 spin interaction induced 1077 spin LED 1174 spin LED structures 1179 spin light-emitting diodes (spin LEDs) 1169 spinoidal decomposition – InAlGaN 705 spinoidal isotherms – InAlGaN 705 – spin–orbit interactions 148 spin–orbit splitting 152, 171 spin ordering 1066 spin orientation 1169, 1179 spin polarization 1110, 1169, 1176, 1179, 1188 spin-polarized carriers 1078, 1135 spin-polarized current 1171 spin-polarized electrons 1168, 1172, 1173, 1176, 1179 spin-polarized holes 1135, 1176 spin-polarized injection 1171 spin-polarized tunneling (SPT) 1078 spin-polarizing electrode 1169 spin–orbit band 225 spin–orbit coupling 1054, 1116, 1135, 1149, 1187 spin–orbit interaction(s) 148, 151, 174, 1084, 1104, 1110, 1172, 1185 spin–orbit interaction, see Jahn–Teller effect 999
spin–orbit levels 1061 spin-orbit split-off band coupling 215 spin-orbit split-off bands 1110, 1115 spin-orbit split-off band transitions 1114 spin–orbit split-off-hole effective mass 175 spin-orbit split-off mass 183, 201 spin-orbit splitting 148, 151, 152, 160, 161, 172, 183 spin resonant tunneling device 1076 spin splitting 1126 spin-splitting energies 1186 spin transport electronics 1046, 1077 spin tunnel junctions 1168, 1169 spin-unpaired electrons 818, 1168 spin-up channel 1100 spin-up electron 1047, 1070, 1079 spin valve 1180 spin valve structure 1173 spinel-type structure 611 spinodal curves 698 spinodal isotherms 705 spinodal phase separation phenomenon 97 spintronic devices 1168 spintronics 1063 – general remarks 1075 spiral hillock features 459 spiral hillocks 445, 460 split-off band 1110 split-off hole mass 152, 174 spontaneous polarization 231, 241, 265, 277 – charge density 266 – charges 588 – coefficients 132 – nonlinearities 253 sputtering technique 84, 637, 638 squared absorption coefficient 96 SQUID magnetometers 1167 SQUID measurements 1158 stacking-order mismatch 351 stacking fault (SF) 818, 851 – basal 853 – electronic structure 884 – induced by Mg doping 887 – prismatic 860 – type I and I1 852 – type II and I2 852 – type III 852 stacking mismatch boundaries (SMB) 329, 818, 844 stacking sequence 2,4,146 stamping method 569 standard framework 498 standard liter per minute (slm) 467, 486 stark effect 288
j1251
j Index
1252
stark shift 264, 288 steady-state capacitance 944 steady-state four-probe method 54, 59 step-flow growth 403 stick-and-ball representations 4, 846 sticking coefficients 400 STM data 642 stoichiometric boundary 633 stoichiometric growth 632 stoichiometric nitrides 131 stokes shift 170 strain-free films 328 strain-induced piezoelectric 277 strain-induced polarization 264 strain minimization 377 strain-relieving defects 260 strain–stress relationship 154 strained-layer superlattices (SLS) 213, 450 Stranski–Krastanov (SK) mode415 stress – strain-GaN on SiC 483 stress-related phenomena 502 stress–strain relationship 155 stress–temperature coefficient 484 strong exchange interactions 1086 sublimation sandwich method (SSM) 392 submicron seed posts 541 substitutional acceptors 1007 substrate surface preparation 494 substrate temperature 299 superexchange interaction 1086 superexchange mechanism 1086 supersaturation 363 surface acoustic wave measurements 63 surface diffusion 413 – length 404 surface energy 368 surface morphology 406, 455, 460, 845 surface-segregation processes 450 symmetry-conserving stress 42
t tandem shift 204 telecommunication purposes 110 TEM analyses 491 TEM bright-field images 656 TEM data 604 TEM diffraction images 1128 TEM images 1128 temperature controllers 296 temperature-dependent Boltzmann term 412 temperature-dependent CL spectra 178 temperature-dependent emission energy 170 temperature-dependent measurements 1042
temperature-dependent thermal conductivity 55 temperature-independent constant 1073 temperature-independent electron 654 temperature-programmed desorption (TPD) 443 temperature resistivity measurements 1044 template-based methods 735 template-based techniques 737 tensile biaxial strain 262 tensile strain 658 tensile uniaxial strain 160 tentative growth mechanism 505 terahertz frequency optical switches 1076 ternary – alloys 89, 90, 210, 652, 1010, 1124 – growth 652 terrace width 445 tetragonal unit cell 383 tetrahedral bonds 366 tetrahedral notation 852 theoretical frequency shifts 642 thermal activation 406 – energy 1042 thermal annealing 521, 608, 1015 thermal conductivity 53, 59, 79, 499, 638 thermal cracking 436 thermal decomposition rate 453 thermal desorption spectroscopy (TDS) technique 443 thermal emission 942 – techniques 956 thermal energy 412 thermal etching 597 thermal expansion coefficients (TEC) 47, 67, 479, 537, 601, 658, 662 thermal imaging 57 thermally detected optical absorption (TDOA) measurements 190 thermally stimulated current spectroscopy 1041 thermal mismatch 290, 326 – induced strain 290 thermal stability 396 thermal strain 510, 546 thermal stress 291, 297, 660 thermodynamic 436 thermodynamical data 388 thermodynamic equilibrium models 398, 451, 934 thermodynamical (low-stability) barriers 362, 392 thermogravimetric technique 398 thick single-crystalline GaN layers 386
Index thick wurtzite GaN films 329 threading dislocation (TD) 460, 473, 553, 586, 656, 660, 662, 818, 822 – densities 528, 585, 604 three-dimensional (3D) – diagrams 100 – growth 341 – mechanisms 565 – nucleation 544 – strain-minimizing shape 568 – stress relief 565 threefold spot pattern 652 three-step ELO process 557 three-step growth 527 tight binding (TB) approximation 133 tight binding calculation 208 tight-binding model 134, 499 tight binding realm 272 time-resolved luminescence 571 time-resolved PL (TRPL) 279, 573 time-resolved Raman measurements 39 TiN nanoporous blocking method 577 TM-doped GaN 1101, 1141, 1146 total energy (TE) 820 traditional compound semiconductors 205 transient PL study 1054 transient spectroscopy 462 transition elements 1046, 1060, 1063, 1188 transition energies 1187 transition growth regime 455 transition metal doped semiconductor 1153 transition metal impurities 1046 transition point 492 translation domain boundaries (TDBs) 818, 844 transmission cross-sectional TEM 472, 529, 565 transmission electron microscopy (TEM) 565, 817 – images 340, 684, 1128 transmission measurements 635 transverse acoustic (TA) mode 39 transverse optical (TO) phonon frequencies 20 trapezoidal crystals 516, 517 trapezoidal stripes 556 trenched substrates technique 659 trialkyl compounds 409, 468, 513 trialkyl precursors 467 triangular stripes 531, 538, 563 trichloroethane (TCE) 329 triethylgallium (TEG) 393, 395, 530, 1005 trigonal field splitting 1061 trimetal Ga (TMG) 1015
trimethylaluminum (TMA) 395 trimethylgallium (TMG) 395, 467 trimethylindium (TMI) 395 truncated hexagonal features 523 tunneling barrier 1182 tunneling magnetoresistance (TMR) ratio 1078, 1182 tunnel junction magnetic resistance (JMR) 1078 two-boat system 738 two-coupled conduction bands 203 two-dimensional buffer material 328 two-dimensional electron gas (2DEG) 586 two-dimensional growth mode 415, 641 two-dimensional imaging 839 two-dimensional polarization 266 two-dimensional stress 292 two-dimensional version 419 two-electron transitions 610, 1010 two-state (binary) system 1076 two-step ELO 551 two-step epitaxial lateral overgrowth 547
u UHV electron cyclotron resonance (ECR) 638 ultrahigh vacuum (UHV) 330 ultraviolet detectors 652, 667 ultraviolet LEDs 508 ultraviolet photoemission spectroscopy (UPS) 69 uniaxial stress 158, 373 universal Kanes relation 194 unpolarized light 1147 UV-assisted atomic layer epitaxy 89 UV emitters 666 UV wavelengths 432
v vacancies – doping 983 – growth kinetic and thermal behavior 993 vacancy 928, 929 – Ga 929 – N 930 vacancy-impurity complexes 931, 988 vacancy-mediated self-diffusion 931 vacancy-related lifetime components 992 vacuum-deposition techniques 329 valence band 224, 1164 – confinement energies 226 – density 1017 – edge 173 – heterojunction offsets 139 – mass parameters 199
j1253
j Index
1254
valence band maximum (VBM) 207, 828 valence band offset 211, 215 valence band structures 143 valence electrons 983 valence force field (VFF) 680 van der Waals bonding 328 van der Waals epitaxy 328 van der Waals forces 411 van der Waals substrates 328 vapor–liquid–solid (VLS) process 726, 727 vapor–solid (VS) growth 726 vapor–solid method 732 vapor phase composition 387 vapor phase epitaxy (VPE) 385 vapor phase growth 726 vapor phase reactions 520 variable-range hopping regime 1163 Varshni formula 203, 1161 Varshni parameters 170, 177, 192 V-defects – induced by Mg doping 894, 895, 897 – InGaN 674 – InGaN quantum well 905 Vegards law 85, 90, 94, 95 Verdet constant 1106 vertical cavity surface emitting laser (VCSEL) 110 vertical gradient freeze (VGF) methods 330 vibrating sample magnetometer (VSM) 1141 vibrational modes 39 Vinter notation 219 VLS process 729 void-assisted separation (VAS) process 576 Volmer–Weber (VW) mode 415 volume-conserving strain 284 V-shaped defect(s) 664, 666 V-type inversion domains 845
w Wafer bonding 471 wave pseudopotential methods 366 wave vector 420, 647 weak antilocalization (WAL) 1185 weak compressive strain 277 weak photoluminescence peaks 84 Weisbuch notation 219 well-characterized semiconductors 75 well-converged plane wave calculations 138 well-defined angular momentum 137 wet etching techniques 35 Wigner–Seitz cell 135 wurtzite (Wz) 1, 131, 1124 Wurtzite AlN, see bandgap semiconductor wurtzite crystal structure 28, 39
Wurtzite GaN bandgap 169 Wurtzite nitride growth 498 wurtzite phase 508, 612 wurtzite semiconductors 213, 1109, 1149 wurtzite structure(s) 2, 7, 143, 144, 148, 844 wurtzitic crystals 162 wurtzitic InN 185 wurtzitic phase 381, 477 wurtzitic systems 410
x X-ray absorption spectroscopy (XAS) 38, 1053 X-ray analysis methods 521 X-ray beam 474, 619 X-ray data 33, 34, 78, 393, 474, 487, 489, 521, 687, 1150 X-ray diffraction (XRD) 10, 11, 37, 38, 64, 187, 188, 276, 364, 372, 462, 465, 487, 514, 528, 537, 588, 590, 605, 607, 608, 637, 695, 702, 1144, 1146, 1189 – experiments 626 – in and out of plane diffractions 466 – peak 72, 91, 372, 463, 488, 519, 525, 570, 573, 581 – spectra 188 – spectral FWHM 636 – system 475 X-ray linewidth 407, 446 X-ray photoelectron diffraction 587 X-ray photoelectron spectroscopy 208, 301, 349, 639 X-ray photoemission spectroscopy (XPS) 48, 1100 X-ray reciprocal-space mapping 95 X-ray reflections 608 X-ray rocking curve 98, 364, 365, 465, 466, 474, 510, 513, 514, 576, 601, 605, 619, 687, 688 – analysis 636 – peaks 522 X-rays 462, 463 – techniques 67 – wavelength 463 – wave method 587 XRD analysis 77, 364, 489, 600, 687, 688 XRD rocking curves 474, 619, 620, 621
y yellow line (YL) emission 555 yellow luminescence 567, 886, 924, 926, 928, 982, 1000, 1056, 1198 YL band 563, 609, 1011, 1013, 1040, 1042
Index Youngs modulus 15, 16, 23, 24, 30, 33, 38, 63, 156, 157, 158, 291, 331, 333, 337, 345, 358, 483, 565, 566
z Zeeman effect 1129 Zeeman splitting 1110, 1116, 1117, 1118, 1119, 1120, 1127, 1129, 1133, 1135, 1136, 1146,1147,1148,1149,1150,1151,1152,1187 Zener theor(ies) 1090, 1091, 1187 zero-dimensional electronic systems 738 zero-order quarter-wave plate 1176 zero-phonon line (ZPL) 1000 zero-phonon peak 1011, 1039 zero-phonon transition 1039, 1042, 1043 zinc blende 1, 2, 6, 9, 14, 16, 17, 21, 22, 24, 28, 29, 40, 41, 43, 45, 57, 63, 77, 78, 90, 98, 109,
131, 132, 136, 138, 140, 141, 171, 176, 183, 184, 201, 202, 232, 294, 466, 681, 723, 851, 854, 858, 930, 932, 1118 – AlN 24, 28, 29, 183 – growth 507 – phase 9, 14, 16, 17, 22, 23, 24, 36, 477 – polytypes 36, 41, 62, 82, 90 – structure 2, 4, 39, 141, 143, 144, 1096, 1124 – symmetry 1117 zinc blende GaN 172 zinc blende InN 200 Zn-doped samples 628 ZnO 350, 351 – surface preparation 353 ZnO substrates 598 ZPL 1045
j1255
Appendix
j1257
Appendix: Periodic Table of Pertinent Elements
atomic weight based 69.72 on carbon 12 Density 5.91 Ga Symbol solid&liquid:g/cm-3 Ar3d104s2p1 Electronic configuration Solid r r: Radioactive gas:g/liter at 273K&1 atm Name Gallium Atomic number 31
II A
IB
IIIB
IIB
9.0122
IVB
5
10.81 6 12.01 2.34 1.85 B 2.62 Be C 1s22s2p2 1s22s2p1 1s22s2 Solid Solid Solid Carbon Beryllium Boron 13 26.98 14 12 24.305 28.09 1.74 2.70 Al 2.33 Mg Si Solid Ne3s2p1 Ne3s2 Solid Solid Ne3s2p2 Magnesium Aluminum Silicon 20 72.59 40.08 29 65.38 31 69.72 32 63.55 30 Ca 8.96 Cu 7.14 Zn 5.91 Ga 5.32 Ge 1.55 Ar3d104s2p1 Ar3d104s2p2 Ar3d104s2 Ar3d104s1 Ar4s2 Solid Solid Solid Solid Solid Copper Zinc Calcium Germanium Gallium 47 112.41 49 118.69 107.87 48 114.82 50 10.5 Ag 8.65 Cd 7.31 In 7.30 Sn 4
VB 14.007 7 1.251 N 1s22s2p3 Gas Nitrogen 15 30.97
VIIIB 2
1.82 P Ne3s2p3 Solid Phosphorus 33 74.92 5.72 As 4.80 Se Ar3d104s2p3 Ar3d104s2p4 Solid Solid Selenium Arsenic 51 127.60 121.75 52 6.88 Sb 6.24 Te Kr4d105s2 Kr4d105s2p1 Kr4d105s2p2 Kr4d105s2p3 Kr4d105s2p4 Kr4d105s1 Solid Solid Solid Solid Solid Solid Tellerium Cadmium Silver Indium Tin Antimony 79 196.97 80 200.59 81 204.37 82 207.2 83 208.98 19.3 Au 13.53 Hg 11.85 Pb 9.8 Bi Tl 11.4 Xe4f144d106s1
Solid Gold
Xe4f145d106s2 Xe4f145d106s2p1 Xe4f145d106s2p2 Xe4f145d106s2p3
Solid Mercury
Solid Tallium
Solid Lead
Solid Bismuth
63.55
0.1787 He 1s2 Gas VIB Helium 16 10 20.18 8 1.429 O 0.901 Ne 1s22s2p4 1s22s2p6 Gas Gas Oxygen Neon 16 32.06 18 39.95 2.07 S 1.784 Ar Ne3s2p4 Ne3s2p6 Solid Gas Argon Sulfur 34 83.80 78.96 36 3.74 Kr Ar3d104s2p6 Gas Krypton 131.3 54 5.89 Xe Kr4d105s2p6 Gas Xenon