Rf Bulk Acoustic Wave Filters for Communications (Artech House Microwave Library)

RF Bulk Acoustic Wave Filters for Communications

For a listing of recent titles in the Artech House Microwave Library, turn to the back of this book.

RF Bulk Acoustic Wave Filters for Communications Ken-ya Hashimoto Editor

artechhouse.com

Library of Congress Cataloging-in-Publication Data A catalog record for this book is available from the U.S. Library of Congress.

British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library.

ISBN-13: 978-1-59693-321-7

Cover design by Igor Valdman

© 2009 ARTECH HOUSE 685 Canton Street Norwood, MA 02062 All rights reserved. Printed and bound in the United States of America. No part of this book may be reproduced or utilized in any form or by any means, electronic or mechanical, including photocopying, recording, or by any information storage and retrieval system, without permission in writing from the publisher. All terms mentioned in this book that are known to be trademarks or service marks have been appropriately capitalized. Artech House cannot attest to the accuracy of this information. Use of a term in this book should not be regarded as affecting the validity of any trademark or service mark.

10 9 8 7 6 5 4 3 2 1

Contents Preface

ix

CHAPTER 1 Background and History

1

1.1 BAW Technology Background 1.1.1 Basic Definitions 1.1.2 Role of Piezoelectric Materials 1.1.3 Transducers and Resonators 1.1.4 Comparisons with SAW and Plate Wave Resonators 1.1.5 Other Kind of Resonators 1.1.6 Electrical Characteristics of Piezoelectric Resonators 1.1.7 Technology Driving Forces 1.2 Thin Plate Resonators: Towards High Frequencies 1.2.1 Conventional Quartz Crystal Thinning 1.2.2 Bonded Plate Resonators 1.3 Composite Resonators 1.4 Development of Thin Films 1.5 Multidimensional Effects 1.6 Legacy Filter Topologies 1.6.1 Balanced Bridge Filter 1.6.2 Ladder Filters 1.6.3 Lattice Filter 1.6.4 Monolithic Filters 1.7 Some Acoustic Device and Materials Processing Legacy References

1 1 2 3 4 5 7 10 11 11 11 12 13 14 15 15 16 17 18 18 19

CHAPTER 2 Resonator and Filter Topologies

21

2.1 Plate Edge-Supported Resonators 2.1.1 Pothole Membrane 2.1.2 Pocket Membrane 2.1.3 Undercut Air Gap Membrane 2.2 Solidly Mounted Resonators 2.3 Electrode Metallization 2.4 Temperature Compensation 2.5 Electrically Coupled Filters 2.5.1 Ladder Filters

21 21 24 25 27 29 31 34 34

v

vi

Contents

2.6

2.7 2.8 2.9

2.5.2 Balanced Ladder 2.5.3 Conventional Lattice Acoustically Coupled Filters 2.6.1 Stacked Crystal Filter 2.6.2 Coupled Resonator Filter Wide-Bandwidth Tuned Coupled Resonator Filters Hybrid Filters Summary References

CHAPTER 3 BAW Device Basics

37 37 37 38 42 45 47 48 48

51

3.1 Thin Film Bulk Acoustic Wave Resonator 3.1.1 The Prototype Resonator and Piezoelectric Constitutive Relations 3.1.2 The Basic Parameters and Equivalent Circuit 3.2 Basic Physics 3.2.1 Wave Propagation, Transmission, Reflection, and Attenuation of Acoustic Waves 3.2.2 Electroacoustic Conversion 3.2.3 Mason Model 3.2.4 Dispersion Relations and Wave Modes 3.2.5 Resonator Design Based on Dispersion Relations 3.3 Device Design 3.3.1 Effective Coupling Coefficient 3.3.2 Loss Mechanisms and Q-Values 3.3.3 Spurious Modes 3.3.4 The Other Important Parameters 3.4 Summary References

59 62 64 67 70 74 74 78 82 88 89 89

CHAPTER 4 Design and Fabrication of BAW Devices

91

4.1 Design Considerations for BAW Devices 4.1.1 Electromechanical Coupling Coefficient 4.1.2 Quality Factor 4.1.3 Spurious Modes 4.1.4 Power Handling 4.1.5 Temperature Coefficient of Frequency 4.1.6 Area Efficiency 4.1.7 Interconnect Losses and Parasitics 4.1.8 Robustness 4.1.9 Nonlinearities 4.2 Fabrication of BAW Devices 4.2.1 Material Selection 4.2.2 Fabrication of SMR Resonators and Filters 4.2.3 Fabrication Tolerances and Trimming

52 52 57 59

91 91 92 92 93 93 94 94 95 96 97 97 101 102

Contents

vii

4.2.4 Process Controls 4.3 Application Space for BAW-FBAR Technology 4.3.1 RF Filters and Duplexers 4.3.2 Oscillators 4.3.3 Sensors References

108 108 108 112 113 115

CHAPTER 5 FBAR Resonators and Filters

117

5.1 Introduction 5.1.1 Short History of FBAR 5.1.2 The Duplexer 5.1.3 The Package 5.1.4 FBAR in Context with the Rest of the World 5.2 FBAR Technology 5.2.1 Introduction 5.2.2 Modeling of FBARs 5.2.3 Method of Ascertaining Q 5.2.4 The Rayleigh-Lamb Modes 5.2.5 Apodization 5.2.6 Frames 5.2.7 Temperature-Compensated Resonators 5.2.8 Coupled Resonator Filters 5.3 FBAR Filters 5.3.1 Interstage Filters 5.3.2 The Duplexer and Multiplexers 5.4 Conclusions References

117 117 119 122 123 124 124 126 129 133 137 140 145 149 150 150 152 156 158

CHAPTER 6 Comparison with SAW Devices

161

6.1 Introduction 6.2 Structural Comparison and Features 6.3 Resonator Performance and Reliability 6.3.1 Q-Factor 6.3.2 Power Durability 6.4 Filter Design 6.5 Manufacturing Process 6.6 Temperature Compensation Technique 6.7 Application Map References

161 161 162 162 165 166 168 168 169 170

CHAPTER 7 Thin Films Deposition for BAW Devices

173

7.1 Most Commonly Used Piezoelectric Materials 7.1.1 Zinc Oxide 7.1.2 PZT

173 173 173

viii

Contents

7.1.3 Aluminum Nitride 7.2 Methods of Deposition of Piezoelectric Films 7.2.1 Sputtering 7.2.2 Practical Aspects of the Sputter Deposition of the AlN Films 7.2.3 Electron Cyclotron Resonance Deposition 7.2.4 Ion Beam Deposition 7.2.5 Metalorganic Chemical Vapor Deposition 7.2.6 Jet Vapor Deposition 7.2.7 Nonvacuum Deposition 7.3 Metal Deposition for BAW Applications 7.3.1 Aluminum 7.3.2 Molybdenum 7.3.3 Tungsten 7.3.4 Platinum 7.3.5 Ruthenium 7.3.6 Combinations of Metals References

174 175 175 183 187 188 188 189 189 189 191 192 192 193 194 194 194

CHAPTER 8 Characterization of BAW Devices

197

8.1 Introduction 8.2 Single-Layer Material Characterization 8.2.1 Introduction 8.2.2 Dielectric and Piezoelectric Layers 8.2.3 Metallic Layers 8.3 Laser Interferometry 8.3.1 Introduction 8.3.2 Measurement Setup 8.3.3 Evaluation of Dispersion 8.4 Loss Mechanisms 8.4.1 Introduction 8.4.2 Acoustic Leakage 8.4.3 Acoustic Leakage Through the Bragg Reflector 8.4.4 Laterally Leaking Waves 8.4.5 Electrical Losses 8.4.6 Viscoelastic Losses 8.4.7 Scattering Losses 8.5 Electrical Characterization 8.5.1 Introduction 8.5.2 Resonator Measurements 8.5.3 Filter Measurements References

197 198 198 198 200 201 201 201 203 204 204 205 207 211 212 212 214 214 214 214 217 219

CHAPTER 9 Monolithic Integration

221

9.1 Introduction 9.2 Compatibility Issues Between IC and BAW Technologies

221 223

Contents

ix

9.3 Practical Implementation 9.3.1 Technology Description 9.3.2 Filtering LNA 9.3.3 WCDMA RF Front-End 9.3.4 WLAN Oscillator 9.4 Conclusion Acknowledgements References

224 225 227 228 232 232 233 233

CHAPTER 10 System-in-Package Integration

235

10.1 Introduction 10.2 Trends in Front-End Integration for Wireless Applications 10.2.1 Multiband, Multimode Wireless Systems 10.2.2 SiP Versus SoC 10.3 SiP Technologies 10.3.1 Laminate Platform 10.3.2 LTCC Platform 10.3.3 Thin Film Platform 10.4 SiP Design 10.4.1 Electromagnetic Modeling 10.4.2 Design Methodology 10.5 Test and Industrialization, Known-Good Die Concept 10.6 RF-SiP Examples 10.6.1 General Wireless Examples 10.6.2 Examples Including BAW References

235 235 235 239 241 241 242 243 245 246 251 252 253 253 255 257

Glossary

259

About the Author

265

List of Contributors

266

Index

267

Preface Nowadays, electromechanical (EM) resonators are widely used in most sophisticated electronic equipment. For example, bulk acoustic wave (BAW) resonators using crystal quartz are indispensable for frequency or time generation owing to their outstanding performances. The mobile communication market has grown explosively in last two decades. From a technological point of view, this growth is significantly indebted to the rapid evolution of silicon technologies, and most of all, functionalities are now realized by the use of silicon integrated circuits (Si-IC). However, highly precise frequency generation and excellent radio-frequency (RF) filtering are exceptional. They were only realizable by the use of quartz resonators and surface acoustic wave (SAW) devices, respectively. RF-BAW devices employing a piezoelectric thin membrane were proposed in 1980. Although their excellent performance was well recognized, the majority of engineers believed that their applicability was very limited due to extremely tight requirements given to the device fabrication. However, the tremendous efforts of a few believers moved mountains. RF-BAW devices progressed surprisingly in the last decade and are now mass produced. Furthermore, they are attempting to take over the current RF-SAW filter market. The devices also receive much attention from Si-IC industries for their use as a core element in sophisticated RF front-end and/or one-chip radio modules based on the system-on-chip (SoC) or system-in-package (SiP) integration with active circuitry. This book deals with key technologies and hidden know-hows necessary for the realization of high-performance RF-BAW resonators and filters. All the authors are prominent professionals in this field, and they did their best to transfer their knowledge to the younger generation. This book is invaluable not only for young engineers and students who wish to acquire this exotic technology, but also for experts who wish to further extend their knowledge. It is extremely hard for any person to prepare such a monograph solely, and only fruitful collaboration of these authors could make this difficult task possible. By the way, the term film bulk acoustic wave resonator (FBAR) might be more familiar to a majority of readers. However, its use is often limited to the category of a free-standing membrane fabricated by the surface or bulk micromachining technology. Namely, the solidly mounted BAW resonator (SMR) employing the multilayered reflector(s) is excluded from this category. From this reason, we follow this categorization, and the RF-BAW resonator is used as the whole set of these two categories throughout this book.

xi

xii

Preface

In Chapter 1, Dr. Keneth Lakin, a pioneer of the RF-BAW devices and a technical leader in this field, reviews the background and history of the RF-BAW resonators and takes readers on a virtual tour of extensive efforts that brought the technology to its current success. In Chapter 2, Dr. Lakin gives detailed explanations on resonator and filter topologies that frequently appear in current RF-BAW technologies. Electrical characteristics of RF-BAW device are simulated quite well by computer simulation and its use is vital in current device design. In Chapter 3, Dr. Jyrki Kaitila describes the BAW device basics, explaining the one-dimensional modeling, detailing various second effects inherent for the precise simulation, and then discussing numerical techniques and underlying physics. In Chapter 4, Dr. Robert Aigner and Dr. Lueder Elbrecht discuss RF-BAW devices based on the solidly mounted resonator technology. First, they consider their design and then discuss their fabrication for mass production in a semiconductor fabrication environment. In Chapter 5, Dr. Richard Ruby, the father of FBAR, reviews free-standing bulk acoustic resonators (FBARs). Dr. Ruby begins this chapter with a short history about the high obstacles that he and his group encountered, how he struggled, and how he achieved a great triumph at the last minute. In Chapter 6, Dr. Masanori Ueda compares the RF-BAW device with the RFSAW device from various points of view. Dr. Ueda has been involved in the research and development of both of these devices, and can evaluate them without bias. As described before, BAW device performances can be simulated numerically fairly well. However, achievable performances are critically dependent on employed manufacturing process, especially the quality of deposited piezoelectric thin films. In Chapter 7, Dr. Sergey Mishin and Yuri Oshmiansky describe one of the most important technologies for the fabrication of RF-BAW devices, namely, deposition of high-quality thin films mandatory for realization of high-performance BAW devices. In Chapter 8, Dr. Gernot Fattinger and Dr. Stephan Marksteiner discuss one more important factor for the realization of high-performance RF-BAW devices: namely, characterization of RF-BAW materials and devices. They also discuss the major technologies of laser probing and electrical properties. Integration of RF-BAW devices with semiconductor circuitry is one of the most important concerns for the future in this community. In Chapter 9, Dr. Marc-Alexandre Dubois, a principal researcher of the famous MARTINA European Consortium, details monolithic integration of RF-BAW devices on Si. In Chapter 10, Dr. A. Bart Smolders, Dr. Jan-Willem Lobeek, and Dr. Nicolaus J. Pulsford discuss the RF integration from another aspect—system-in-package (SiP) integration. They explain various technologies used in the SiP integration, demonstrate its effectiveness, and then show how the BAW technologies fit well with RF-SiP, which will be the mainstream for further RF integration. Ken-ya Hashimoto Editor Chiba University Chiba-sha, Japan May 2009

CHAPTER 1

Background and History Ken Lakin

1.1

BAW Technology Background The purpose of this chapter is to give a brief history of the development of BAW technology which is covered in technical detail in later chapters of this book. First it is necessary to define what the BAW technology is and then put the history in that context. For the purposes of this book, BAW history is interesting not so much as who did what when (that will be apparent from numerous references) but how other technologies were drawn upon to make the development of the modern thin film BAW technology possible. Microelectronics has played a key role over the years by providing materials-processing techniques previously unavailable. Review papers give an overview of thin film resonator technology [1–5]. 1.1.1

Basic Definitions

The term bulk acoustic wave (BAW) refers to primary acoustic waves that propagation in the bulk of a material whose dimensions are infinite and wherein the wave occupies all of that volume. There are three possible propagation modes called the normal modes of the material. Those modes are well understood for a large number of materials whose elastic properties are known. In more practical terms, a wave in a finite three- dimensional region can only approximate the propagation characteristics of an infinite region. The first approximation required to support a BAW is that the lateral extent of the medium is much larger than the wavelength and cross-section of the wave. The practical definition of BAW is imprecise and depends on what artifacts crop up due to the finiteness of the beam. For example, a beam starting out as being of comparable dimensions to the wavelength would appear as a point source and spread widely, due to diffraction, but could be described as some complex linear combination of the normal modes. The second approximation is that the lateral extent of the wave, and therefore of the medium, is such that the wave is primarily one-dimensional but with some residual effects due to lateral finiteness. In the direction of propagation the material extent may be very finite, such as a half-wavelength thick for a resonator. Yet in such a case, dimensions will appear large in the direction of propagation because the wave bounces within the resonator between parallel surfaces maintaining its characteristics as if propagating over considerable distance. Typical average lateral dimensions might be

1

2

Background and History

approximately 100 times the wavelength for resonators in filters designed for 50-ohm source and load impedances. Whereas finiteness is a distortion imposed on BAW, other modes of propagation are uniquely tied to the finiteness of a structure. For example, waves can propagate along and be guided by a surface or at an interface. The most notable being the solid to air interface that supports surface acoustic waves (SAWs). A feature of waves is that they tend to be guided by regions of slower velocity and lower energy density. If there is a lateral deformation at or very near a surface, the material can expand perpendicular to the force (Poisson effect) out into the air region. That added degree of freedom makes the surface appear mechanically softer and as a result the SAW is confined to the surface. In the case of SAWs the material region must be just a half space with the relevant approximation that the material is sufficiently thick that the wave does not exist at any other surfaces. If the material region is formed as a plate with two parallel surfaces, but large in lateral extent, then another set of waves, plate waves (PW), can propagate along the parallel boundaries of the plate. These waves are most pronounced when the thickness of the plate is comparable to the propagation wavelength. It turns out that such a geometrical constraint is met by a typical BAW resonator. Further, plate waves can be generated in BAW resonators and can plague high-performance BAW resonators with parasitic resonances. Other modes of propagation are possible in the typical BAW structural approximation but PW are the most pronounced. Since a resonator can be though of as a confinement structure for a wave bouncing between reflecting surfaces, it is only a manner of properly generating and confining a wave to make a useful resonator. Two issues then emerge. First, how to generate the wave, and second how to confine the wave so that most of the energy is stored with a minimum amount of energy loss except on a controlled basis. 1.1.2

Role of Piezoelectric Materials

The most straight forward method of generating an acoustic wave is to use a piezoelectric material. The piezoelectric direct and inverse effects are described in general by the equations, T = cS − eE

(1.1)

D = eS + εE

(1.2)

Here (1.1) is Hook’s law of elasticity, T is stress (force per unit area), S is strain, e is the piezoelectric coefficient, c is mechanical stiffness, ε is permittivity, and E is the electric field. The second equation shows the contribution of mechanical strain to electric charge generation and displacement current. Accordingly, mechanical deformations and electric properties are piezoelectrically coupled. As will be shown in subsequent chapters, the strength of the piezoelectric coupling determines the bandwidth of filters and the mechanical losses in the material will determine resonator Q and accordingly filter insertion loss.

1.1 BAW Technology Background

1.1.3

3

Transducers and Resonators

The transduction process requires the application of an electric field to the piezoelectric material. This is done generally in the form of metal electrodes applied to, or in close proximity to, the surfaces of the piezoelectric material. In the technical discussions that follow it will be useful to keep in mind the distinction between transducer and resonator. Figure 1.1 shows cross-sections of three devices. Note the adoption of the microelectronic custom of drawing wherein vertical dimensions are greatly expanded over horizontal dimensions. In Figure 1.1(a) a set of metal electrodes is applied to a piezoelectric plate much as in a simple capacitor. The outer metal surfaces are against air or vacuum so that acoustic waves reflect off these surfaces and effectively stay confined to the material body. The combination of electrodes and piezoelectric plate constitute a transducer and because the energy is confined within the outer surfaces of the electrodes the transducer is also a resonator. In Figure 1.1(c) the electrodes are close to the piezoelectric plate but not in mechanical contact with the plate. Small air gaps between the electrodes and piezoelectric plate insure that most of the applied voltage is applied to the piezoelectric plate. Because of the air boundary on the piezoelectric plate surfaces, sound energy is confined to the piezoelectric plate. Thus the piezoelectric and electrodes are together the transducer but the piezoelectric alone is the resonator. Since the electrodes are not part of the mechanical resonance their thickness can be made large for mechanical strength and electrical conduction purposes. Prior to about 1950, many quartz crystal resonators were made in this format. The areas of those old resonators were made about 2 cm2 in order to accommodate the large currents flowing through resonator circuits of that era. Interest in this configuration has reemerged where thin electrodes have too much electrical loss. In the air gap coupled resonator the added series capacitance of the combined air gaps, Cg, reduces the effective piezoelectric coupling by the factor 1/(1+Ct /Cg) where Ct is the piezoelectric plate capacitance. Figure 1.1(b) shows a transducer atop another plate of material. The resonator in this case is the material region from topside of the top electrode to bottom side of the bottom plate. The resonator may be one or many half-wavelengths thick. Practical resonators have been made where the thickness is around 120 half-wavelengths (mode number equal to 120) at a given frequency. These are called overmoded reso-

PIEZOELECTRIC

d

PIEZOELECTRIC

ELECTRODES (a)

(c)

PIEZOELECTRIC ELECTRODES d

d

ELECTRODES

PIEZOELECTRIC

d

(d)

(b)

Figure 1.1 Cross-sections of BAW resonators. (a) Piezoelectric plate with attached electrodes. (b) Piezoelectric transducer attached to a substrate. (c) Piezoelectric plate with electrodes separated by an air gap from the plate. (d) Piezoelectric plate with lateral electric field excitation.

4

Background and History

nators (OMR). A small change in frequency can cause the resonance to shift up or down one mode number and many resonances can exist over the bandwidth of the transducer. The frequency spacing is the reciprocal of the round-trip time of a propagating wave. In Figure 1.1(d) the lateral field resonator is designed to keep electrodes out of the resonator by exciting the piezoelectric plate with fringing electrical fields that are mostly parallel to the plate surface. Resonance is established between two unelectroded surfaces. Historically, resonators (called crystals then and today) of the type in Figure 1.1(c) were used until the early 1950s. With the need for smaller resonators and the availability of metal plating techniques, crystals of the form in Figure 1.1(a) were produced in ever decreasing sizes. The configuration in Figure 1.1(b) was limited to transducers for delay lines and other applications. Transducer bonding techniques were not advanced enough to support low-loss transduction although some metallurgical techniques showed some promise in special applications. It was the need for BAW delay-line transducers at high frequencies that led to the development of piezoelectric thin film deposition [6]. And, it was not until the introduction of thin film deposition did the composite configuration show some promise as a resonator. The air gap and lateral field resonators have a modern day application in quartz crystals for low-aging applications where metal electrodes would have detrimental effects, or for microwave resonators where acoustic losses in the electrodes are excessive. 1.1.4

Comparisons with SAW and Plate Wave Resonators

It is apparent from the discussion above that an acoustic resonator can be formed by a transduction means and mechanical boundaries that confine the energy. As introduced above, there are other waves that exist at boundaries such as SAWs and PWs. Figure 1.2 shows the cross-section of a SAW transducer wherein the thickness of the substrate is much larger than the wavelength of the SAW. Here spatially periodic

(a)

(b)

Figure 1.2 SAW transducer and resonator. (a) Side view showing driven electrodes and reflectors. (b) Top view of transducer and resonator. In practice there are many more pairs of transducer electrodes and more reflector stripes on both sides of the transducer.

1.1 BAW Technology Background

5

electrodes generate a wave that is synchronous at center frequency with the periodicity of the electrodes. Typical electrodes are a quarter-wavelength long in the propagation direction and much wider that a wavelength in the width (depth in the drawing) direction. Once launched, a SAW propagates in both directions along the surface until intercepted by another transducer. A resonator can be formed by using an array of electrodes on both sides of the transducer that intercept the wave to such an extent as to cause a significant reflection back towards the transducer. The transducer can be designed so that some of the reflection occurs within the transducer itself. Accordingly, a resonator can be formed by the transduction and reflection process. A SAW can also be abruptly reflected by a vertical termination of the material region but this leads to waves that reflect in such a manner as to radiate into the bulk of the material and thus would constitute a loss mechanism. The details of SAW resonators and filters are described in a later chapter. Plate waves propagating in the lateral dimension of a plate are confined to the plate by the top and bottom surfaces. It is assumed that the plate is much larger in width than in thickness. In Figure 1.3(a) is illustrated an electrode pair, one on top and one on the bottom of the plate. The electrodes are approximately a half-wavelength long in the propagation directions. The number of waves that can be excited is quite complex unless means are taken to trap a single mode under the electrodes and have the propagation of other modes cutoff in the external regions. When energy is trapped in the electrode region the device becomes a resonator. Figure 1.3(b) shows the results of a numerical calculation of the mechanical displacement for a trapped resonator. The electroded region slows the wave propagation and allows the energy to stay confined. The standing wave is confined to the electroded region and has an evanescent decay outside the region where other modes are accordingly cutoff. Historically, plate wave resonator and associated monolithic crystal filter (MCF) technologies were overcome by the advancements in SAW devices. The main advantage of the MCF would be comparatively small size at a given frequency but particularly at frequencies below 100 MHz. 1.1.5

Other Kind of Resonators

Figure 1.4 illustrates some other resonators of historical significance that can be implemented in thin film form. They will be discussed in order of apparent stiffness and therefore applicable frequency. In Figure 1.4(a) the top view of a cantilever beam is shown and it is assumed that the lateral dimensions of the beam are much larger than the plate thickness. The flexural beam resonance has the lowest potential operating frequency because of the relatively high degree of compliance of the beam. In Figure 1.4(b) the beam is clamped at both ends and so is substantially stiffer. In Figure 1.4(c) a thin plate is clamped on all sides resulting in a stiffer structure which can support a “drum head” resonance. Finally, in Figure 1.4(d) a beam is supported in the center, which forces a node point, and is allowed to vibrate in the length dimension. This is known as the length extension mode. An extension of this device is a disk supported at the center node point and has motion in the radial direction.

6

Background and History

PIEZOELECTRIC

d

Amplitude, Angstroms

(a)

Electrodes

Horizontal Distance (b) Figure 1.3 Plate wave excitation and trapped energy. (a) Cross-section showing top and bottom electrode stripes used to excite the plate wave. The electrodes are normally comparable to the principal wavelength in the lateral propagation direction and longer in the depth direction of the figure by many times the plate thickness. (b) Calculation of wave amplitude in the vicinity of the electrode showing that energy is actually trapped.

These various forms of resonators may occur by deliberate design or as an artifact in some other more desired resonator. They have also appeared in one form or another in the field of MEM devices. Figure 1.5 suggests a possible implementation in thin film form of a piezoelectric bimorph cantilever beam resonator designed to operate in a flexure mode. A voltage applied to the top piezoelectric plate can excite a length extension strain that causes the beam to bend. The bending of the lower beam is detected by the piezoelectric and generates an output voltage. Although useful in low-frequency resonators and filters, the driven flexure vibration can possibly occur in high-frequency BAW devices as a parasitic effect. Figure 1.6 gives a pictorial summary of frequency ranges for different modes of vibration resonators over a range of plate thicknesses. BAW modes for longitudinal and shear waves are higher in frequency because the material structure is stiffer for those modes and frequency is simply inversely proportional to thickness.

1.1 BAW Technology Background

7

a

a Air Gap

(a)

(c)

a a Air Gap

Air Gap

(b)

(d)

Figure 1.4 Other forms of resonant structures in a plate that is much thinner than the lateral extent of the device. (a) Beam resonator clamped at one end. (b) Beam resonator clamped at both ends. (c) Membrane clamped on all sides. (d) Length-extensional resonator.

(a) Gnd

Substrate (b)

I/O Gnd I/O

(c) Figure 1.5 Beam resonator. (a) Bonded plate low-frequency piezoelectric bimorph. Piezoelectric film implementation as an example of how classical resonator or filter structures can be reinvented in thin film form. (b) Cross-section view. (c) Top view showing electrode run-out.

1.1.6

Electrical Characteristics of Piezoelectric Resonators

This section will briefly describe the electrical properties of a resonator in order to bring to light the reemergence of problems in thin film BAW resonators that have plagued the BAW quartz crystal field.

8

Background and History

FILM THICKNESS, MICROMETERS

100 AlN SHEAR AlN LONGITUDINAL

ZnO LONGITUDINAL ZnO SHEAR CLAMPED MEMBRANE 10

LENGTH EXTENSION BEAMS

1.0

0.1 100 Hz

1 MHz

10 MHz

100 MHz

1 GHz

10 GHz

FREQUENCY

Figure 1.6 Approximate required film thickness as a function of frequency range for a number of resonator technologies. A width or length to plate thickness ratio of 10:1 was assumed for the low-frequency plate wave devices. Highest frequency operation for a given thickness is for longitudinal AlN. Lower frequencies require thicker films and materials of slower velocity, or shear waves would be applicable.

The interaction between the applied voltage and resulting current flow gives rise to a complex impedance that exhibits both series and parallel resonance as will be described in detail in Chapter 3. Without going into the details, the impedance of a simple resonator is given by, ⎛ tan φ⎞ Z = (1 jωC )⎜1 − K 2 ⎟ φ ⎠ ⎝ φ=

kd π ⎛ f ⎞ = ⎜⎜ ⎟⎟ 2 2 ⎝ fp ⎠

(1.3)

(1.4)

f p = V 2d

(1.5)

K 2 = φ tan φ for f at series resonance frequency

(1.6)

where f is frequency, V is the velocity of propagation, d is the thickness of the plate, C is geometric capacitance, and K2 is the piezoelectric coupling coefficient. Figure 1.7 shows a plot of the modeled magnitude and phase of a representative resonator. If the device were not a piezoelectric resonator and just a capacitor then the impedance would have the characteristic 1/f response (nearly a horizontal line on the frequency scale in Figure 1.7). For frequencies outside resonance the resonator has near −90° of phase, characteristic of a capacitor. As series resonance is approached the reactance drops and a large capacitive current flows just below series resonance. At series resonance impedance drops to a minimum value and is resistive. For frequencies slightly above series resonance the current flow is inductive. As parallel resonance is approached, the impedance becomes very large and

1.1 BAW Technology Background

9 90.0

500

75.0 60.0 400

Parallel Resonance

Series Resonance

300

30.0 15.0 0 -15.0

200

Phase, Deg.

Zmag, Ohms

45.0

-30.0 -45.0 100

-60.0 -75.0

0 1450

1500

1550

1600

1650

-90.0 1700

Frequency, MHz

Figure 1.7 Simulated magnitude and phase of impedance over frequency for a one-dimensional finite Q resonator. Series resonance occurs where the phase crosses zero with positive slope and the impedance is a minimum. Parallel resonance is where phase crosses zero with negative slope and impedance is a maximum.

reaches a maximum and resistive value at parallel resonance. For frequencies above parallel resonance the resonator again becomes capacitive. The principal value of a piezoelectric resonator is the realization of a high Q inductance if only over a short range of frequencies. This is the ideal case that can only be approximated in practice. Resonators whose lateral extent is much larger than plate thickness act as many resonators operating in parallel. The various area segments of the resonator tend to operate independently and thickness control can be an issue. For example, suppose a very high-quality resonator would normally exhibit a series resonant resistance of 1 ohm and a parallel resonant resistance of 2,000 ohms. Now assume that the majority of the resonator is at parallel resonance but that 0.05% of the resonator area is at series resonance with a series resistance accordingly scaled by area to be 2,000 ohms. These two resonator portions are electrically in parallel and the combination for the overall resonator is 1,000 ohms. Thus, the apparent parallel resonance resistance of resonator has been significantly reduced. If the area of the resonator is 200 × 200 μm, the area of the parasitic portion would be only 4.5 μm2. Figure 1.8 shows the effect on resonator phase of the parasitic resonance. This is an extreme example only in that the parasitic resonator was assumed to be discrete and well defined. Distributed thickness effects such as roughness can also degrade resonator performance. The parasitic area effect can occur as a result of a lack of parallelism during resonator thinning. As crystal plates were made thinner to reach higher frequencies, the degree of mechanical processing tolerance decreased, making large-area thin resonators very difficult to fabricate. For this and other reasons, quartz crystal reso-

10

Background and History

Phase, Deg.

90 75 60 45 30 15 0 -15 -30 -45 -60 -75 -90 1550

1570

1590

1610

1630

1650

Frequency, MHz Figure 1.8 Modeled phase of a resonator having small-area parasitic resonator having series resonance at the parallel resonant frequency of the principal resonator. In practice such effects could occur over a distribution of frequencies.

nators became smaller in lateral extent compared to thickness to the point of no longer being BAW resonators. Finally, multidimensional wave propagation effects were incorporated for energy trapping and resonance mode control. 1.1.7

Technology Driving Forces

In order to appreciate the contents of this book it is important to understand the driving forces that brought about the development of thin film BAW technology. The discussion above suggests that there is more than one way to build resonators and resonator-based filter, as is certainly the case. The drivers from the technology’s applications are cost, performance, and device size. The performance issue was the need for high-frequency operation well beyond the frequencies reached by quartz crystal technology and to a certain extent beyond SAW technology. This in turn required the development of processes for manufacturing thin piezoelectric films and resonators. Filters are required that exhibit the necessary system bandwidths with low loss. This is because system architectures have largely eliminated IF filtering and require that these functions be carried out at the front end. A front-end filter must have low-insertion loss for receiving and lower loss for transmitting. Filters are a major cost driver in the cell phone market. Not only is the filter cost itself a concern, but an inefficient filter leads to increased battery requirements. The economies brought about by wafer-scale manufacturing have had a significant affect on filter cost in high-volume production. Device size is important as circuit boards become smaller and space on the board a premium. In wafer-scale manufacturing the wafer die count is important in determining end unit cost. In military systems the availability of small filters has a significant affect on system architectures. Systems that might have been deemed undesirable because they required a number of large-area dominating filters become feasible with the existence of small filters. In particular, a thin film BAW filter is

1.2 Thin Plate Resonators: Towards High Frequencies

11

approximately 1,000 times smaller than a ceramic filter for the same frequency and characteristics. From a historical perspective, none of this is really all that new. What has been significantly different is the intense pressure to achieve these goals in a short period of time. The goals then for the development of BAW technology were, and still are today: (1) higher frequencies and better performance, (2) small size, and (3) low cost. Again, from a historical perspective it is interesting to see what other technologies have been brought to bear on this effort. The sections below review the core legacy resonator technologies to see what thin film BAW has been built upon.

1.2

Thin Plate Resonators: Towards High Frequencies 1.2.1

Conventional Quartz Crystal Thinning

The obvious approach to reach higher frequencies with conventional piezoelectric resonator materials is to thin a crystal plate until the desired frequency is obtained. Clearly there are practical limits to thinning large-area crystal plates in mass production with perhaps the most important issue being the need to mechanically support the thin-plate resonator after the fact. For example, AT-cut quartz crystal unsupported plates are commercially available in thickness of less than 25 micrometers (equivalent to approximately 60-MHz fundamental frequency) having areas of approximately 25 mm2. (That these plates survive subsequent processing is probably due to the fact that quartz does not exhibit cleavage planes.) Once thinned to a practical limit, these blanks can then be used as a starting point to further increase the resonator frequency by selective area thinning [7]. One such approach, shown in Figure 1.9, is the inverted mesa configuration wherein a thin resonator region is supported by a much thicker supporting substrate of the same material [8]. Chemical etching techniques have been extensively investigated along with ion milling to produce the thin plates in the mesa [9–17]. Further, considerable effort has been directed towards chemical etching techniques that do not leave a crystal facet roughened surface. The final result can be a large crystal blank having an array of inverted mesas in a wafer-scale manufacturing format. The chemical properties of quartz, that allow it to be relatively easily chemically or plasma etched, have historically not been available for high K2 materials of interest for resonators. 1.2.2

Bonded Plate Resonators

Other fabrication techniques were proposed to obtain thin plates, such as the one suggested in Figure 1.10 [18]. Here a crystal plate is bonded to a substrate having an

Figure 1.9

Inverted mesa quartz plate with the thinned area produced by chemical etching.

12

Background and History

appropriate void region and the mechanical strength required to support the eventual thin crystal plate for the resonator. Once bonded, the crystal plate can be mechanically thinned to the desired amount while the peripheries of the crystal plate are supported by the substrate. Today wafer-bonding techniques developed in microelectronics might be usefully applied in this resonator configuration, and recently more advanced processing techniques and topologies have been proposed for quartz mesa resonators [17]. The advantage of this legacy approach is that materials not producible in thin film form could be processed into resonators. In Chapter 2 an etching process is described that would allow the plate in Figure 1.10 to be bonded to a flat silicon wafer, thinned, and the hole etched in the wafer afterwards. This is a much more practical approach because it avoids the problem of the plate bowing into the hole during polishing when the plate is thin compared to the lateral extent of the hole.

1.3

Composite Resonators Rather than thin down a single crystal plate, it became apparent to researchers early on that growing the resonator material to a desired thickness might be a viable approach [19, 20]. However, these ideas occurred well in advance of the materials science and technology necessary to support actual device fabrication. The lead in to the composite resonator was the microwave bulk wave delay line which required a thin film piezoelectric transducer for high-frequency operation. These delay lines were very thick and amounted to microseconds of time delay. The composite resonator is basically a delay line that does no have an output transducer and is thin.

PIEZOELECTRIC PLATE

SUBSTRATE (a)

(b)

(c)

Figure 1.10 Bonded plate resonator. In (a) a piezoelectric plate with bottom electrodes and substrate with an open area, in (b) the plate has been bonded to the substrate, and in (c) the plate has been thinned and upper electrodes attached. A modern approach would be to form the hole as a final step.

1.4 Development of Thin Films

13

One of the first composite approaches, shown in Figure 1.11, would have resulted in resonators having high mode numbers and low effective coupling coefficients because the likely substrates would have been many half-wavelengths thick. It was not until thin silicon substrates became available that the composite resonator could be demonstrated at high frequency. That device is discussed in detail in Chapter 2. The most important concept forwarded by these and other approaches, irrespective of their relative implementation successes, is that the desirable electrical, mechanical, and processing properties need not reside within a single material but can be realized by a composite of materials and processing techniques. And, that perspective constitutes a major departure from the conventional crystal plate technology. Accordingly, the major developments in resonator technology have been highly dependent on advances in materials processing, primarily those driven by microelectronics integrated circuit technology.

1.4

Development of Thin Films The principal driving force for the development of piezoelectric thin films was the need for higher frequency microwave delay lines, and the lack of adequate fabrication techniques for thinning piezoelectric crystal plates to high frequencies. In addition, methods for bonding piezoelectric plates to delay lines were only moderately successful due in large part to the need for an electrode between the piezoelectric plate and delay line material. In the early 1970s, there was also a desire to have higher velocity substrates for SAW devices to allow operation at higher frequencies. Eventually, high-resolution electron beam lithography, developed by the microelectronics industry, allowed the fabrication of high-frequency SAW devices. The first reported work directed towards thin piezoelectric films was that for zinc oxide (ZnO) and CdS transducers for microwave delay lines [1]. Subsequent work on films quickly moved towards piezoelectric films for SAW devices [21–24]. AlN deposition on sapphire substrates for SAW applications was reported in the early 1970s [22]. A significant amount of work was done on ZnO for SAW devices. The assumption with ZnO was that a process developed for bulk wave delay lines could be easily transferred to SAW applications. However, as with other piezoelectric films, it was discovered that a higher quality of film was required for SAW transduction than for delay lines. In microwave delay lines the transducer is heavily loaded by the delay line and the unloaded Q of the piezoelectric film need not be much higher than the loaded Q of the transducer. When propagation was along the length of the piezoelectric film a higher quality film was required for SAW propagaPiezoelectric Film Substrate

Figure 1.11 Composite resonator composed of a thin film piezoelectric grown on a suitable substrate. The film and substrate will have a 180° phase for a fundamental mode resonator.

14

Background and History

tion. The problem of material Q surfaced again, and in the extreme, when films were applied to thin film BAW resonators where most of the energy is in the film. Films good enough for SAW were not good enough for BAW resonators. AlN films grown by high-temperature organometallic chemical vapor deposition for SAW devices were of the required quality but lacked a viable means of putting an electrode under the resonator. Significant advances have been made in sputter film deposition as will be detailed in a later chapter.

1.5

Multidimensional Effects All structures actually fabricated are of course three dimensional, and it is only a matter of the degree of multidimensionality that affects device performance. The most important issue with BAW resonators is the generation of plate wave modes that can be seen as spurious responses in the normal resonator response. Figure 1.12 illustrates the physics of the problem. In Figure 1.12(a) a simple resonator having electrodes and a lateral dimension comparable to the piezoelectric plate thickness is shown. Assuming a simple longitudinal mode-thickness excitation, vertical deformation causes a lateral deformation through the natural Poisson coupling. This coupling causes lateral vibrations in additional to thickness vibration in time harmonic excitation. In Figure 1.12(b), the plate is assumed to be much larger in lateral extent. Thus when volume element A is subjected to excitation its lateral deformation is canceled by the like lateral deformation of adjacent cells C and B. The result is a one-dimensional deformation locally. End cell D is also driven to a thickness deformation but in this case there is no adjacent cell on the left-hand side to cancel the lateral deformation in that direction. Accordingly, volume element D generates vertical and horizontal deformations capable of exciting lateral wave propagation in the plate. Once excited at the plate edges, plate waves will propagate throughout the plate reflecting off any material or electrical discontinuity. The energy contained in the lateral wave will be dependent on the strength of the excitation. Most important is the ratio of plate wave energy to that in the primary thickness mode. In the small resonator case of Figure 1.12(a), there is no real distinction between the two excitations because the deformations are so tightly coupled and occupy the same volume. In the extended resonator of Figure 1.12(b), the internal volume elements driven in the thickness mode will have an associated energy larger than the plate wave in approximate proportion to the width-to-thickness ratio of the resonator. In Figure 1.12(c) the plate wave problem is cast in a format closer to the thin film BAW case. Here the piezoelectric plate is assumed to be of a lateral extent larger than the electroded region, or more pertinent, the overlap of the electroded region. D (a)

C

A (b)

B

D

C

A

B

(c)

Figure 1.12 Resonator geometry for describing plate wave excitation. (a) Resonator having comparable lateral and thickness dimensions, (b) large lateral-to-thickness dimensions, and (c) part of an extended plate.

1.6 Legacy Filter Topologies

15

The excitation of the plate wave is somewhat softened by the stiff material region in the plate outside the electrodes. Fringing electric fields at the edges can also excite other modes. Total current flow in the resonator is a result of integration of displacement current across the electrodes. Two main factors determine the influence of plate waves on total current flow; first is the strength of the excitation, and second is the lateral extent of the electrode relative to the wavelength of the lateral wave. Figure 1.13 shows a simulation of a BAW resonator done with numerical analysis [25]. It is clear that plate waves cause ripple in the impedance. Figure 1.14 shows the wave distribution across the resonator. Displacement current in the resonator has a similar ripple and as frequency changes the mode number changes for the lateral standing wave. The electrode has the effect of integrating the displacement current flowing into the electrode. For an even number of lateral half-waves the plate wave current component averages to zero but when frequency shifts then there is a plus/minus contribution of current that is left over and that causes the current ripple. Later chapters will discuss how to mitigate this problem.

Legacy Filter Topologies There are a number of legacy filter technologies that have application to high-frequency applications. Most of the quartz crystal technology involved the use of individual resonators integrated into a circuit possibly with the use of transformers and inductors. 1.6.1

Balanced Bridge Filter

The balanced bridge filter is shown in Figure 1.15 is probably the simplest filter short of using a single series resonator between input and output. The bridge is

Phase, Deg.

Impedance

1.6

5.40 5.46 5.52 5.58 5.64 5.70 5.76 5.82 5.88 5.94 6.00 Frequency, GHz x μm

Figure 1.13 Numerical analysis simulation of the phase and amplitude across a piezoelectric plate. Frequency scale is normalized by plate thickness and impedance is normalized by resonator capacitance reactance magnitude.

16

Background and History

Amplitude

Electrodes

Distance (mesh points) Figure 1.14 Numerical analysis simulation of the standing wave distribution across a resonator. The ripple is due to plate waves.

X1

Rload C1

Figure 1.15 Classical bridge filter. In some configurations there is a resonator of shifted frequency in the lower branch as well.

designed so that the capacitor matches the capacitance of the crystal when off the resonant frequency. At series resonance the filter circuit is out of balance and the low impedance of the crystal provides minimum insertion loss. At parallel resonance the bridge is also out of balance but the reactance of the capacitor is much larger than the load resistance and so there is minimal transmission. More complicated bridge filters use resonators in both branches with the resonators slightly offset in frequency for a multipole response. For high-frequency miniature filter applications the transformer presents a problem and other filter topologies are much better. 1.6.2

Ladder Filters

The ladder filter configuration is shown in Figure 1.16 in two formats. In Figure 1.16(a) the shunt branches are capacitors. At low frequencies the Q of the resonators are sufficiently high to overcome the reactance of the capacitors and provide a multipole response. Conventionally these filters have been employed in applications that do not require low insertion loss.

1.6 Legacy Filter Topologies

17

X1

X2

C4

X3

C5 (a)

X1

X2

X4

X3

X5

(b)

Figure 1.16 Ladder-filter circuits. In (a) capacitors are used in the shunt branches as commonly done in low-frequency quartz crystal filters. In (b) resonators are used in both series and shunt branches. This configuration is used in high-frequency BAW resonator filters.

In Figure 1.16(b) the resonators are also used in the shunt branches. As will be described later, the use of shunt resonators gives lower insertion loss and a symmetrical passband. This is the legacy filter technology that is widely used in high-frequency applications today. In practice, more or fewer resonator sections are employed. 1.6.3

Lattice Filter

The lattice filter is another legacy filter configuration that is of increasing interest for high-frequency applications wherein the direct connection with RF ICs is best done with balanced networks. Shown in Figure 1.17 is the schematic of a simple lattice filter. These filters are used to provide a balanced input and output and may be cascaded to give better selectivity. Redrawn, the lattice is a form of a balanced bridge filter. Typically, these filters used transformers for input and output.

X1

X2

X2

X1

Figure 1.17 Lattice-filter configuration that provides a balanced input and output. This configuration can have wider bandwidths than the ladder filters.

18

Background and History

1.6.4

Monolithic Filters

The monolithic filter, shown in cross-section in Figure 1.18, is a relatively miniature filter that was widely used at frequencies up to 100 MHz with some demonstrated at higher frequencies. One R&D filter was demonstrated at around 800 MHz [26]. In the evolution of filter development the monolithic filter was replaced by SAW filters in most applications because the SAW filter is much easier to design and manufacture. Advanced thin film processing techniques show some promise for thin film BAW implementation of the MCF configuration.

1.7

Some Acoustic Device and Materials Processing Legacy It is readily apparent from Figure 1.1 that the BAW resonator is a simple device compared to other microelectronic structures. It appears to be just a piezoelectric plate with two electrodes. The quartz crystal legacy is to take a quartz plate and grind it down to the required thickness to achieve the desired resonance. This has been the most cost-effective manufacturing approach for decades supplemented in more recent times by chemical etching and microelectronics style batch processing. As electronics technology evolved from large vacuum tubes to smaller ones and to solid state the need and desirability of smaller area resonators increased and the size of

(a)

(b)

Figure 1.18 Representation of a three-pole monolithic crystal filter (MCF). (a) Cross-sectional view, and (b) top view. Electrode overlaps form the excitation or I/O regions. The center resonator typically is not connected to outside circuitry. Energy trapping can also be done by thinning the piezoelectric plate between and outside the resonators.

1.7 Some Acoustic Device and Materials Processing Legacy

19

quartz crystal resonators became smaller. The most valuable contribution to thin film BAW from quartz crystal technology was a simple equivalent circuit model and the physical understanding and computational handling of resonator spurious responses. Thin film BAW has most of its roots in areas of microwave acoustics and transduction. Thin piezoelectric films were first developed to support for microwave delay lines in upper-GHz frequencies. The act of attaching a piezoelectric plate to a delay line and grinding down the plate to the desired thickness was a difficult task. It seemed much more desirable to grow the piezoelectric film to the required thickness on an underlying electrode and subsequently fabricate a thin film top electrode. Pioneering work in ZnO and AlN deposition was done initially for microwave delay lines. Perhaps the greatest contribution came from the microelectronics industry in the form of photo lithography, film deposition, wet and dry (plasma) processing, magnetron sputtering of metals and dielectrics, and the availability of precision substrates and wafer planarization techniques. In the area of thin film deposition there are a number of representative contributions [27–38].

References [1] Weigel, R., et al., “Microwave Acoustic Materials, Devices, and Applications,” IEEE Trans. MTT, Vol. 50, No. 3, March 2002, pp 738–749. [2] Ruby, R., “Review and Comparison of Bulk Acoustic Wave FBAR and SMR Technology,” IEEE 2007 Int. Ultrasonics Symp. Proceedings, paper 11E-3. [3] Muralt, P., “Is there A Better Material For Thin Film BAW Applications Than AlN?” 2005 IEEE Int. Ultrasonics Symp. Proceedings, paper 5C1. [4] Lakin, K. M., “Thin Film Resonator Technology,” 2003 Frequency Control Symp. Proceedings, paper WE1A-4 (Invited). [5] Lakin, K. M., “Review of Thin Film Resonator Technology,” IEEE Microwave Mag., Vol. 4, No. 4, December 2003, pp. 61–67. [6] Foster, N. F., et al., “Cadmium Sulphide and Zinc Oxide Thin-Film Transducers,” IEEE Trans. on Sonics and Ultrasonics, Vol. Su-15, No. 1, January 1968, pp. 28–41. [7] XECO, 1651 Bulldog, Cedar City, UT 84720. [8] Guttwein, G. K., A. D. Ballato, and T. J. Lukaszek, “VHF-UHF Piezoelectric Resonators,” U.S. Patent 3,694,677. [9] Hanson, W. P., “Chemically Polished High Frequency Resonators,” Proc. 37th Ann. Freq. Contr. Symp., 1983, pp. 261–264. [10] Hunt, J. R., and R. C. Smythe, “Chemically Milled VHF and UHF AT-Cut Resonators,” Proc. 39th Ann. Freq. Contr. Symp., 1985, pp. 292–300. [11] Lepek, A., and U. Maishar, “A New Design for High Frequency Bulk Resonators,” Proc. 43rd Annual Frequency Control Symposium, Denver, CO, May 31–June 2, 1989, pp. 544–547. [12] Berte, M., “Acoustic-Bulk-Wave Resonators and Filters Operating in the Fundamental Mode at Frequencies Greater Than 100 MHz,” Electronic Letters, Vol. 13, No. 9, April 28, 1977, pp. 248–250. [13] Stern, F. M., et al., “The Fabrication of High Frequency Fundamental Crystals by Plasma Etching,” Proc. 43rd Ann. Freq. Contr. Symp. (AFCS), 1989, pp. 634–639. [14] Wang, J. S., S. K. Watson, and K. F. Lau, “Reactive Ion Beam Etching for VHF Crystal Resonators,” Proc. 34th Ann. Freq. Contr. Symp. (AFCS), 1984, pp. 101–104.

20

Background and History [15] Brauge, J., M. Fragneau, and J. P. Aubry, “Monolithic Crystal Filters Fabricated by Chemical Milling,” Proc. 39th Freq. Cont. Symp., pp. 504–513. [16] Ishii, O., et al., “High Frequency Fundamental Resonators and Filters Fabricated by Batch Process Using Chemical Etching,” Proc. 1995 IEEE Freq. Cont. Symp., pp. 818–826. [17] Lakin, K. M., G. R. Kline, and K. T. McCarron, “Self Limiting Etching of Piezoelectric Crystals,” Proc. 1995 IEEE Int. Freq. Cont. Symp., pp. 827–831. [18] Coussot, G., and E. Dieulesaint, “Method of Manufacturing an Electromechanical System Having a High Frequency Resonance,” U.S. Patent 3,924,312. [19] Curran, D. R., “Composite Resonator,” U.S. Patent 3,401,275. [20] Sliker, T. R., and D. A. Roberts, “A Thin-Film CdS-Quartz Composite Resonator,” J. App. Phys., Vol. 38, 1967, pp. 2350–2358. [21] Manasevit, H. M., F. M. Erdmann, and W. I. Simpson, J. Electrochem. Soc., Vol. 118, No. 1864, 1971. [22] Lakin, K. M., J. Liu, and K. Wang, “Aluminum Nitride on Sapphire,” 1974 IEEE Ultrasonics Symp. Proceedings, Milwaukee, WI, November 11–14, 1974, p. 302. [23] Shiosaki, T., Proc. IEEE 1978 Ultrasonic Symp., Vol. 100, 1978. [24] Hickernell, F. S., Proc. IEEE, Vol. 64, No. 631, 1976. [25] Lakin, K. M., and K. G. Lakin, “Numerical Analysis of Thin Film BAW Resonators,” 2003 IEEE Int. Ultrasonics Symposium, paper 4A-3. [26] Lakin, K. M., G. R. Kline, and R. S. Ketcham, “Low Insertion Loss Filters Synthesized with Thin Film Resonators,” 1987 IEEE Ultrasonics Symposium, Denver, CO, October 14–16, 1987, Vol. 1, p. 375. [27] Hashimoto, K.-Y., et al., “Preparation of Piezoelectric ZnO Films by Target Facing Type of Sputtering Method,” 1998 IEEE Ultrasonics Symp. Proc., Vol. 1, 1998, p. 207. [28] Iriarte, G. F., et al., “Synthesis of C-Axis Oriented AlN Thin Films on Metal Layers: Al, Mo, Ti, TiN and Ni,” 2002 Ultrasonics Symposium Proc., Vol. 1, 2002, pp. 311–315. [29] Emanetoglu, N. W., et al., “MgxZn1-xO: A New Piezoelectric Material,” 2001 IEEE Ultrasonics Symposium Proc., Vol. 1, 2001, pp. 253–256. [30] Mishin, S., et al., “Sputtered AlN Thin Films on Si and Electrodes for MEMS Resonators: Relationship Between Surface Quality, Microstructure and Film Properties,” 2003 IEEE Ultrasonics Symp., 2003, p. 2028. [31] Naik, R. S., et al., “Measurement of the Bulk, C-Axis Electromechanical Coupling Constant as a Function of AlN Film Quality,” 2002 IEEE Ultrasonics Symp. Proc., 2002, p. 292. [32] Lakin, K. M., K. T. McCarron, and J. F. McDonald, “Temperature Compensated Bulk Acoustic Thin Film Resonators,” 2002 IEEE Ultrasonics Symp. Proc., Vol. 1, 2002, pp. 855–858. [33] Bjurstrom, J., et al., “Dependence of The Electromechanical Coupling on the Degree of Orientation of C-textured Thin AlN Films,” IEEE Trans. on Ultrasonics, Ferroelectrics and Frequency Control, Vol. 51, 2004, p. 1347. [34] Guy, I. L., E. M. Goldys, and S. Muensit, “Measurements of Piezoelectric Coefficients of Nitride Semiconductor Films,” Semiconducting and Insulating Materials Conference, 2000. SIMC-XI. International, July 3–7 2000, pp. 55–58. [35] Shiosaki, T., “Growth and Applications of Piezoelectric and Ferroelectric Thin Films,” 1990 IEEE Ultrasonics Symp. Proc., Vol. 1, 1990, pp. 537–546. [36] Driscoll, M. M., et al., “Low Noise, UHF Oscillators Utilizing High Overtone, Lateral-Field Excitation, Lithium Tantalate Resonators,” 1991 IEEE Ultrasonics Symp. Proc., Vol. 1, 1991, pp. 453–457. [37] Lee, C. H., et al., “Silicon Bulk Micromachined RF MEMS Switches with 3.5 Volts Operation by Using Piezoelectric Actuator,” 2004 IEEE MTT-S Digest, p. 585. [38] Krishnaswamy, S. V., B. R. McAvoy, and M. H. Francombe, “Thin Film in Microwave Acoustics,” in Physics of Thin Films, Academic Press, Vol. 16, 1992, pp. 145–224.

CHAPTER 2

Resonator and Filter Topologies Ken Lakin

The principal requirement for a resonator topology is that it must have suitable boundary surfaces that act to confine the sound wave to the resonator region. Of course the structure needs to be practical from a manufacturing standpoint having high yield and low cost. A configuration that might have seemed unpractical in times past can suddenly be of interest through the application of advanced manufacturing techniques. There are two principal thin film resonator topologies, one with two air interfaces and the other having one air interface and one reflector stack as described below. Both topologies are widely used in high-volume manufacturing.

2.1

Plate Edge-Supported Resonators From an acoustic reflector standpoint, an edge-supported membrane having top and bottom air interfaces would seem to be the most ideal topology. The question is how to fabricate such a structure in a practical manner. 2.1.1

Pothole Membrane

The first thin film resonators were made with the so-called pothole process illustrated in Figure 2.1 [1–8]. The process was made possible by some pioneering work in microelectronics [9]. In Figure 2.1(a) a silicon wafer having (100) orientation is given an oxide layer on the backside and a highly doped p layer of Boron on the topside. This is a batch process that can be done in a diffusion furnace processing hundreds of wafers at a time at low cost. Next, in Figure 2.1(b), an oxide window is opened to expose the bottom of the pothole. The pothole is etched from the backside of the wafer using a wet chemical batch process that etches fast along the crystal’s [100] direction but slowly along the crystal’s [111] directions leaving four walls of (111) orientation. Once the surface is deglazed of the highly doped glass layer formed during the diffusion, the wafer is ready for further processing. The patterns formed in silicon are restricted to rectangular shapes because of the anisotropy of the etching process. The p membrane so formed is typically less than 1-micrometer thick and is accordingly translucent, appearing light brown in transmitted light. This degree of transparency allows the location of the pothole to

21

22

Resonator and Filter Topologies

p+ Silicon

(100) Silicon

Oxide

(a)

Pothole (b) Piezoelectric

Electrodes (c)

Removed p+ Layer (d)

Figure 2.1 Pothole membrane process for fabricating thin film resonators: (a) a silicon (100) wafer is given a p layer about 1.5 micrometers thick, (b) a wet etch is used to form the pothole in the backside of the wafer, (c) the piezoelectric device is fabricated on top of the p membrane, and finally, (d) the p membrane is plasma etched to free up the piezoelectric membrane and devices.

be identified from the topside of the wafer during subsequent IC photolithographic processing. Next, the bottom electrodes are formed by a normal IC process. Typically, an Al film is deposited on the surface of the wafer in such a manner as to provide the correct (111) orientation for subsequent AlN deposition. Typically, the piezoelectric film is grown by sputter deposition covering the entire wafer. After the top electrode is processed, a capacitor-type structure is formed. It is possible to conduct resonator measurements at this stage because the resonator is a composite resonator being composed of AlN, electrodes, and the p layer. In early work, the p layer was not removed and was part of the resonator. The last step in Figure 2.1(d) is to plasma etch the p membrane from the bottom side of the wafer to leave an edge-supported piezoelectric plate. Although Figure 2.1(d) suggests a single resonator for illustration purposes, the membrane can actually support complex filter topologies having many resonators. The pothole process was used to fabricate resonators in the 1980s and the first membrane resonators, without the p layer, were reported in 1982 [8]. The freely supported resonator with air on both sides is now often called an FBAR. A disadvantage of this process is that the p layer is still in the wafer field outside of the pothole region. Since p silicon is a relatively good conductor it can give rise to parasitic resonators where top electrode out-traces overlap the p field. In general, the p area can act as an unwanted ground plane unless removed by some other processing.

2.1 Plate Edge-Supported Resonators

23

The step in Figure 2.1(b) could be followed by a patterning of the p layer to remove it from the majority of the topside field and, if necessary, a deposition of silicon dioxide followed by a planarization of the wafer. This process is outlined in Figure 2.2. In Figure 2.2(b) the p layer is selectively removed by an IC process and then covered by an oxide deposition. Planarization of the wafer gives the structure in Figure 2.2(c). Finally, the piezoelectric device is fabricated and the p layer is removed. Topside electrode traces over the oxide are now given a measure of isolation from the silicon wafer (which could of the high-resistivity type). Again, the advantage of the pothole process is that low-cost batch processing can be employed. Critical dimensionality control occurs only with the actual resonator fabrication, but that issue is common to all resonator fabrication techniques. A disadvantage is the large wafer area taken up by the vertical projection of the opening at the backside of the wafer. This could be mitigated to a certain extent by wafer thinning as now routinely practiced in the IC industry. The purpose of the above exercise is to illustrate how a resonator topology first implemented in the 1970s and then mostly abandoned in the early 1990s can be revived through application of advances in IC processing that have taken place since then. These advances include, the ability to use thinner silicon wafers, advanced plasma processing, and the wide spread use of chemical mechanical polishing (CMP) or other planarizing processes.

(a) Oxide

p+ Silicon

(b) Oxide

(c)

Piezoelectric Oxide Silicon

(d)

Figure 2.2 Advanced pothole process. In (a) the wafer is already provided with a p membrane, (b) the p+ membrane is patterned and then overlain with an oxide layer, (c) the wafer is planarized to smooth the wafer surface, and (d) the piezoelectric device is fabricated followed by a plasma etch to remove the p region and in the process smooth the pothole walls.

24

Resonator and Filter Topologies

2.1.2

Pocket Membrane

The pocket process is similar to the pothole process except the manner of carrying out the backside etch is much more sophisticated, offers greater freedom in structural features, but is more expensive. In Figure 2.3(a) an etch stop layer is formed (if needed) followed by complete device fabrication in Figure 2.3(b). The backside of the wafer needs to be oxide patterned and then followed by a deep reactive ion etch (RIE) that leaves vertical side walls, Figure 2.3(c). The deep-etch process involves a complex plasma chemistry that leaves the vertical side walls protected yet allows etching in the direction of the impinging beam. The etching is known to effectively stop on AlN and allow a good degree of process control. The etching process is done one wafer at a time, and that adds to the cost element. The use of thin wafers is important here. One of the features of the deep RIE etching process, that has been exploited by the IC industry, is the very high aspect ratio vertical structures that can be formed. For thin film resonators, the waste of topside area caused by the pothole process is eliminated by the steep vertical walls. The lateral resolution of the process is such that wall-anchored vertical support structures could be fabricated as well as vertical silicon features that are only supported by the membrane. Figure 2.4 suggests that, if needed, some of the piezoelectric membrane can be left with fine-detailed silicon support structures. The deep RIE process can produce very vertical narrow side wall features with high depth-to-width aspect ratios of 100:1 or greater. Wafers are processed individually and sufficient energy is imparted to the wafer as to require wafer cooling to avoid damage.

Etch Stop Layer

(100) Silicon

Oxide

(a) Piezoelectric

Etch Stop Layer

Electrodes

Oxide

(b)

(c)

Figure 2.3 Pocket membrane process. The steps are summarized in (a) an etch stop layer (such as AlN) is applied to the top of the wafer and an oxide layer formed on the back of the wafer to allow definition of the hole, (b) fabrication of the thin film resonator in the normal manner, and (c) a deep RIE to open up the hole and expose the bottom of the resonator.

2.1 Plate Edge-Supported Resonators

25

Support Wall

Piezoelectric Membrane

(a) Piezoelectric Support Wall

Support Wall (b)

Figure 2.4 Pocket process with support structures allowed by the very high aspect ratio provided by the RIE process (a) top view and (b) side view. Height-to-width ratios of over 100:1 are easily achieved by this process.

Subsequent to the pothole membrane development, and about the time of the pocket membrane process, significant advances were made in wafer bonding. Accordingly, wafer bonding would be a way to seal off the backside holes as part of a wafer packaging process. Either pothole or pocket process allows the wafer to be subsequently diced with a wet process, sawing. 2.1.3

Undercut Air Gap Membrane

Perhaps the most elegant of the membrane processes is the one that forms an air gap under the membrane, as a postfabrication dry process implemented from the topside of the wafer [10–13]. For example, in Figure 2.5(a) a patterned sacrificial layer is fabricated on the topside of the substrate. This layer must be compatible with the following metal deposition such that the metal’s crystal orientation is proper for the subsequent piezoelectric deposition. Typically, the sacrificial layer is in a polycrystalline or amorphous state that allows the metal to orient along preferred directions without direct influence from the sacrificial layer. The high-energy state of the layer is also very conducive to the subsequent removal process. On top of the patterned sacrificial layer the lower level metal is deposited and patterned followed by the piezoelectric film deposition. The care with which these steps must be carried out is discussed in a later chapter. Next, in Figure 2.5(b) VIAs are etched to access buried electrodes and the top metal is deposited and fabricated. A VIA hole is etched or ion milled through the overlaying device structure into the sacrificial layer. The wafer is then subjected to

26

Resonator and Filter Topologies Piezoelectric Support Substrate (a)

(b)

(c)

Figure 2.5 Air gap process. In summary, (a) a sacrificial support layer is deposited and patterned on the wafer followed by lower electrodes and then piezoelectric film deposition, (b) top electrodes are formed and then followed by VIAs to access the buried electrodes and the sacrificial layer, and (c) the sacrificial layer is removed by a highly selective etching process.

an etching agent that enters through the hole and etches out the sacrificial layer leaving a membrane as suggested in Figure 2.5(c). The etching hole need not be plugged as long as the wafer is subsequently only dry processed, including trimming to frequency, wafer dicing step, and any packaging operations. Examples of the process are shown in the optical photographs of some test structures in Figures 2.6 and 2.7. The tests were done for two different sized etch holes. The area of the small holes in Figure 2.7 can be compared to the area of the undercut region. The rectangular regions (with tabs in Figure 2.6) mark the boundaries of the

Figure 2.6 Demonstration test patterns. The rectangular regions with tabs are the pattern of the sacrificial layer. There are no metal layers under the AlN film. Holes were etched through the AlN to expose the sacrificial layer and the wafer subsequently etched to remove the sacrificial layer. The optical interference pattern is due to the AlN membrane bowing upward.

2.2 Solidly Mounted Resonators

27

Figure 2.7 Test structure similar to the one in Figure 2.6, except that there are more holes of smaller diameter. Holes could be within the pattern and not just at the edges.

sacrificial layer. The optical interference pattern is caused by mechanical stress in the film, acquired during film growth, relieved by upward bowing the film. The width-to-thickness ratio of the sacrificial layer is over 100:1, yet the activated gas is able to diffuse throughout the area and reaction products escape out the hole. The mechanical pinning of the membrane to the wafer support structure would be as good as that of the pothole or pocket process. The narrow gap, of the order one micrometer, could allow parasitic capacitance from the resonator bottom electrodes down to the silicon, and accordingly couple resonator electrodes. Typical lateral dimensions of electrodes would be 100 times the gap thickness, even if the silicon wafer is of high resistivity, the parasitic capacitance effect might be important at parallel resonance. The smaller etch holes could be effectively plugged by an overlay deposition of sufficient thickness. The problem is that this would allow the film to be stressed by thermally generated pressure in the gas trapped under the film.

2.2

Solidly Mounted Resonators A more mechanically rugged resonator structure, illustrated in Figure 2.8, can be formed by isolating the resonator from the substrate with a reflector array [14–17]. The array is composed of nominally quarter wavelength (acoustic) thick layers sometimes called a Bragg reflector, the optical analog. The number of layers in the reflector depends on the total reflection coefficient required and the reflection occurring at each layer interface. If the substrate has relatively high impedance then the first layer on top of the substrate should be of low impedance, the next layer high impedance, and so forth. A suitable sequence might be SiO2 and AlN or SiO2 and W (tungsten) on a silicon or sapphire wafer. If the total reflector reflection coefficient is large enough, little or no sound enters the substrate. In which case it does not matter if the first layer is of high or low impedance relative to the substrate. Because tungsten has relatively high mechanical impedance, each material boundary has a higher reflection coefficient than if AlN were used and therefore fewer layers are required. Since AlN and most piezoelectric materials have moderately high impedance, the layer under the resonator is of low impedance and is usually silicon dioxide. If conductors are used for the high-impedance layers, then they generally have to be patterned as a means of avoiding parasitic effects in complex arrays of resonators used in filters. Such patterning can leave the wafer nonplanar but a clever

28

Resonator and Filter Topologies Electrodes Piezoelectric

Reflector Layers

Substrate

Figure 2.8 Solidly mounted resonator cross-section. The reflector stack is composed of acoustically quarter-wavelength layers of materials that produce a substantial reflection coefficient at each layer boundary. The bottom boundary of the resonator is 180° of phase down from the top of the resonator.

technique of reflector fabrication and subsequent planarization has been demonstrated [17]. Figure 2.9 shows a plot of mechanical displacement versus depth for an AlN resonator with aluminum electrodes at 1,800 MHz. The resonator plot is for a frequency near series resonance. The standing wave is largest in the piezoelectricelectroded region and decreases gradually through the depth of the nine-layer (not all layers are shown) reflector stack. The entire structure is the resonator and the electroded piezoelectric region is technically just a transducer. Because energy is stored outside of the electrically sampled region the effective piezoelectric-coupling coefficient of the resonator is reduced over that of an isolated thin film resonator.

0.10

Al Electrodes

0.08

0.04 0.02 0.00

AlN

SiO2

AlN

AlN

-0.06

SiO2

-0.04

SiO 2

-0.02

AlN

Displacement, A

0.06

-0.08 -0.10

0

2

4

6

8

10

Distance, μm

Figure 2.9 Simulation of resonator displacement versus depth. The resonator is series resonant at 1,800 MHz. The most wave amplitude, and hence energy, is stored in the AlN–electroded region with wave amplitude decreasing with depth into the reflector. The reflector is composed of nine layers in total on a sapphire substrate.

2.3 Electrode Metallization

29

In the case shown, the transducer region is 180° thick (simply called a half-wavelength even though there are three material regions) and the material boundary to the right of the “bottom” electrode is also the “resonator” boundary. If, for example, the near-in reflector layer were thicker than a quarter-wavelength, the excess portion would appear in the resonator, causing the resonant frequency to shift downward. Because there is more sound energy in the near-in reflector layer than the other layers, that layer has a greater effect on overall resonator performance compared to more distant reflector layers. Resonator properties and some wide bandwidth filter properties are strongly influenced by the reflection spectrum of the reflector. Bandwidth of the reflector is affected by the impedance ratio between layers with the SiO2/W sequence having a much wider bandwidth than an SiO2/AlN sequence as shown in Figure 2.10. In both cases, the number of reflectors was chosen to such that there was no significant increase in resonator performance through the addition of more layers. The various layers in the reflector need not have exactly the same materials in the high/low sequence so long as the sequence alternates between high and low. For example, the first two layers nearest the resonator might be SiO2/W while the remaining layers needed just SiO2/AlN, but fewer than nine in total because of the large reflection at the W interfaces. That way only one W layer need be patterned. Specific details on reflectors will be given in a later chapter.

Electrode Metallization There are two key elements of electrode metallization that strongly affect resonator performance as illustrated in Figure 2.11 [18, 19]. First, the effective resonator-coupling coefficient has a peak in K2 at a particular thickness ratio. Here the ratio is defined as the thickness of one metal layer to the piezoelectric thickness, and the data is for both electrodes of the same thickness. Since filter bandwidth is strongly tied to K2, wide-bandwidth filters may require the use of W or Mo for metallization as a means of increasing effective K2. The second affect on resonators is the reduction in piezoelectric film thickness for a given frequency. The thinner the piezoelec-

Reflection coefficient

2.3

Frequency, MHz

Figure 2.10 Reflection spectrum of a reflector stack. The solid line is for nine layers of the SiO2/AlN sequence and the dashed line is for a SiO2/W sequence. It will be apparent how reflector bandwidth affects the stacked crystal filter as described in a later section.

30

Resonator and Filter Topologies

Au

Al

W

Mo

(a)

Al Mo Au

W

(b) 2

Figure 2.11 Simulation of resonator effective K and frequency constant for AlN thin film resonators having equal thickness metal on both sides shown as a function of the thickness ratio of one of 2 2 the metal thicknesses to the AlN thickness. In the K plot, K has a characteristic peak near the 0.1-thickness ratio for Al. The frequency constant falls rapidly for the higher impedance films of Mo, W, and Au, allowing significantly thinner piezoelectric thickness for a given frequency.

tric the smaller the area required to achieve a desired capacitative reactance. Smaller area translates into more filters per wafer during manufacturing and probably lower per-filter cost. Electrode resistance is a secondary but important factor in metallization that can be controlled to some extent by resonator layout and methods to connect the resonator to other circuit elements or the I/O structures. At 2 GHz the skin depth in Al is approximately 2 micrometers and the optimum metal thickness for electrodes at that frequency is approximately 200 nm (2,000A). This suggests that the current flow in the electrodes is mostly uniform throughout the electrode thickness. As a consequence, there are power losses in the electrodes that must be carefully controlled. Further, the assumed equal potential nature of the electrode may be doubted around series resonance where current flow is the greatest. Composite electrodes composed of a high mechanical impedance electrode such as W might include an Al layer for better conduction. Because electrode metals are more lossy mechanically than the other materials used in the structure, a design tradeoff is required. Thicker metal lowers resistive losses but increases mechanical losses. Somewhere inbetween there is an optimum

2.4 Temperature Compensation

31

metal thickness that depends on the whole resonator design. However, at the upper microwave frequencies it may be necessary to increase metal thickness to reduce resistive losses because thinner metals become somewhat discontinuous and disproportionately more resistive. Figure 2.12 shows a calculation of series resonant Q versus metal thickness for Mo, Au, and Al metal electrodes at 10 GHz. The piezoelectric thickness was reduced to maintain the same resonant frequency as the metal thickness increased. In the model, the Q of AlN was chosen to be 5,000 for an epitaxial film. At near-zero metal thickness the series resistance losses dominate resonator Q. As metal thickness increases a point is reached wherein the electrode resistance and mechanical losses are equal, giving a peak in the Q response. Beyond that point metal mechanical losses begin to dominate. Ironically, one way to reduce metal losses in microwave resonators is to use the air-gap-coupled electrode configuration used in 1940s resonators. Here there is no sound in the electrodes and therefore they can be made as thick as required for mechanical stability while reducing resistive losses. Two problems arise. First, the air gap will form a capacitance in series with the resonator and thereby lower the effective K2. For example, if the combined air gap thickness is equal to one-tenth the piezoelectric thickness, the series air gap capacitance equals the piezoelectric capacitance and K2 is reduced by a factor of two. Second, it is very difficult to maintain an air gap small enough not to significantly impact K2 because of finite strain in the piezoelectric films.

Temperature Compensation Most materials have a negative temperature coefficient for stiffness, meaning quite simply that they get softer as temperature increases. Some ferroelectric materials have a sufficiently large polarization induced piezoelectric effect to affect the stiffness of the material in a manner that is significant for acoustic resonators. Quartz is 900 800 Mo

700

Qseries

2.4

600

Au

500 400 Al

300 200 100 0 0

0.02

0.04

0.06

0.08

0.1

0.12

Metal, μm

Figure 2.12 Simulation of series resonant Q versus metal thickness for each of two electrodes on a piezoelectric thin film resonator. The AlN Q was assumed to be 5,000, Mo Q of 400 with 5-ohms-per-square-sheet resistance at 500A, Au Q of 100 with 2.5 ohms per square at 500A, and Al Q of 200 with 3 ohms per square for 500A.

32

Resonator and Filter Topologies

a unique material that has a positive temperature coefficient of stiffness due to stretching of the Si-O chain upon increased temperature. The effect causes the material to become stiffer with temperature over a useful range of temperatures. In particular, the property is manifest in the polycrystalline form as used in microelectronics as well as in the crystal form. Since there are no temperature-compensated materials readily available in thin film form that are also piezoelectric, it is necessary to use composite structures composed of positive and negative coefficient characteristics. Temperature compensation can therefore be obtained with a composite layering of materials of normal negative coefficient with silicon dioxide which has a positive temperature coefficient [20]. Figure 2.13 shows four material composite configurations that can be used to achieve a degree of temperature compensation. In Figure 2.13(a) the oxide layers are shown below the electrodes. This format is good when Al electrodes are used because it allows the Al layer to escape the high temperature of the oxide deposition (300°C) unless it is already pinned by another layer. So, the oxide is deposited first, followed by the Al film, which has a tendency to orient properly on the polycrystalline surface of the oxide. Next AlN is deposited but not at such high temperature as to cause the Al electrode to hillock. The top layer of oxide is deposited onto the AlN where adherence is not a problem. Finally, the top metal layer is deposited and the electrodes formed. This is the preferred configuration for the SMR because, if there is already an oxide layer next to the resonator, it only needs to be made thicker. In that configuration the resonator is inherently not symmetric and that can give rise to spurious resonances. A high frequency spurious response can be prevented by making the added oxide layer half the thickness of the oxide layer that is on top of the AlN. In Figure 2.13(a, c, d) configurations, the oxide layer inside the resonator acts as a series capacitance that reduces the effective K2 of the resonator. As practiced, the Figure 2.13(a) configuration was used in the design of narrow-bandwidth filters which required temperature compensation. The level of TC is a matter of degree and as the filter bandwidth is narrowed through the use of the series oxide capacitance

SiO 2 Piezoelectric

SiO2 (a)

SiO2 Piezoelectric

SiO2 (b)

SiO 2 Piezoelectric Piezoelectric

SiO2 Piezoelectric

(c)

(d)

Figure 2.13 Possible temperature compensation layouts using silicon dioxide or similar temperature-compensated material. For simplicity, the drawing is for thin film resonators but the same can be done for SMR.

2.4 Temperature Compensation

33

Df/fa, ppm

the required degree of temperature compensation increases in such a manner as to hold the filter stable down to bandwidths of just under 1%. The configuration of Figure 2.13(b) is interesting because the resonator remains symmetrical and for a given frequency smaller in size because of the oxide layers. However, the top oxide must be deposited on top of the top metal electrode, which could be a problem with Al unless the metal is pinned with an initial sputter-deposited oxide layer. There should be no problems using the upper oxide layer with more refractory metals such as Mo or W. The configuration in Figure 2.13(c) has a single oxide layer that is probably not desirable because it leaves the resonator highly nonsymmetric. In Figure 2.13(d) the oxide is in the center of the resonator, at the point of maximum stress, where the effect of the oxide might be most pronounced. However, that configuration would require that the upper piezoelectric layer be grown on the oxide, and that is not likely to result in a satisfactory film. In the SMR, a small fraction of the acoustic energy is stored in the topmost layers of the reflector as previously described. Consequently, the resonator TC is automatically partially compensated if the last reflector layer is a positive TC material such as silicon dioxide (+85 ppm/°C in film form). The normal −25 ppm/°C of an AlN-only resonator is reduced to −15 ppm in this case. The first thin film TC composite resonator was the AlN on silicon p+ resonators using shear wave propagation [21]. Figure 2.14 shows some experimental results for an SMR resonator having a degree of temperature compensation compared to experimental data for AT-cut quartz. The process of compensating is to offset on material with another so as to cancel out the first-order variation. The data in Figure 2.14 suggests that too much oxide was used because the curve is actually turned over in the linear variation. 200 180 160 140 120 100 80 60 40 20 0 −20 −40 −60 −80 −100 −120 −140 −160 −180 −200 −100 −75

Parallel

Series

Quartz

−50

−25

0

25

50

75

100

125

150

Temperature, deg. C

Figure 2.14 Some early experimental results for a SMR compensated with silicon dioxide in the configuration of Figure 2.13(a).

34

Resonator and Filter Topologies

Although first-order temperature-coefficient data is known for many materials, very few higher order elastic coefficient terms are known. Results on resonator aging have been published [20].

2.5

Electrically Coupled Filters With resonators as circuit element building blocks, networks of resonators can be designed to implement various filter characteristics. The sections below will present a general overview of filter topologies of current interest which will include balanced and unbalanced filters, the latter is of increasing interest to cell phone applications and IC integration. 2.5.1

Ladder Filters

Electrically connected resonators can form ladder, lattice, or other similar circuits, introduced briefly in Chapter 1. The interconnect techniques are straightforward and can be implemented to minimize parasitic effects and inordinate conduction losses in the electrode metallization. Figure 2.15 shows the circuit diagram of a simple ladder filter having resonators in series and shunt branches. One or both of the end series resonators can be eliminated or more sections added depending on design constraints. A typical ladder filter response is shown in Figure 2.16, for the purpose of describing how the ladder filter works. In this case all the series resonators have

Figure 2.15

Simple ladder filter having series and shunt resonators.

Figure 2.16 Experimental results for a simple ladder filter having five series and four shunt resonators with AlN and Al electrodes.

2.5 Electrically Coupled Filters

35

the same series and parallel resonant frequencies and likewise the shunt resonators are all identical but different from the series resonators. The filter has five series resonators and four shunt resonators, hence forth called a 5-4. The center frequency of the filter is at the series resonant frequency of the series resonators. There the series circuit branch has the lowest impedance (resistance) and current flow is more or less straight through the filter. To support an unimpeded current flow in the series branches, the shunt resonators are shifted in frequency such that their parallel resonance is at approximately at the series resonant frequency of the series resonators. That way the network has minimal current flow to ground through the shunt elements and minimum insertion loss. For this effect to be of greatest benefit, the shunt resonators must have high parallel resonant resistance, and the series resonators must have high series resonant Q to give lowest resistance. This forces the resonator technology to produce high Q at both resonances. The out-of-band rejection of a ladder filter, at frequencies well outside the acoustically active range of the resonators, is controlled by the capacitive voltage divider nature of the ladder circuit. More ladder sections, or higher capacitance shunt resonators, increase the ultimate rejection but also increase the in-band insertion loss. As frequency is increased from the low frequency side the shunt resonators go through series resonance and that produces the deep notch on the lower frequency side of the passband. Further increasing frequency causes the shunt resonator to become inductive and then parallel resonant at the filter center frequency and becoming more capacitive across the upper half of the passband. The series resonators are series resonant at passband center and become parallel resonant and produce the high-frequency notch. At higher frequencies the resonators become capacitors again. These near in notches are used to advantage in filters designed for cell phones and other high performance applications. Clearly these notches can be spread according to the distribution of resonant frequencies of the building block resonators. The corners of the filter can be sharpened by higher Q resonators and by not having all resonators at the same resonant frequency, as will be described in a later chapter. Figure 2.17 shows a comparison of three types of simple ladder filters. The one having lowest insertion loss is in the 3-2 configuration and also has the lowest out-of-band rejection. The next lowest filter is a 5-4 and has higher insertion loss because of the increased number of resonators in the filter. However, the larger number of resonators gives a higher out-of-band rejection. Typically, one filter would be used in the front-end of a receiver and the second as a post–LNA filter. The narrow-bandwidth filter in Figure 2.17 was made with temperature-compensated resonators in the format of Figure 2.13(a). In general, more complicated ladder filters have been designed for high-performance applications wherein various resonators have slightly different frequencies in order to optimize filter characteristics. The shifted frequencies can be implemented with series inductors (e.g., bond wires), or through incremental thickness adjustments of the metal electrodes. Figure 2.18 shows circuit diagrams of filters having either deliberately introduced inductors or the inherent inductance of the circuit. The advantage of this approach is that the resonator fabrication itself can be simplified by having the set

36

Resonator and Filter Topologies

IL = 1.4 dB BW = 38 MHz

IL = 2.5 dB BW = 33 MHz

IL = 3.7 dB BW = 18 MHz

Figure 2.17 Summary of ladder filters of simple topology. The widest bandwidth filter is a 3-2 type with lower insertion loss but also minimal out-of-band ultimate rejection. The other two filters are of the 5-4 configuration and have the same ultimate rejection (same mask set actually) but one was designed for maximum bandwidth and the other of narrower bandwidth and temperature compensated.

(a)

(b)

Figure 2.18 Ladder filter circuit diagrams showing the presence of intrinsic or introduced inductance used to shift resonator frequencies and thereby provide a more optimized bandwidth. In (a) shunt inductors, possibly wire bonds, are used to shift the shunt resonators’ frequencies, possibly by the use of bond wires. In (b) an inductance is suggested that can be chosen series resonate the shunt resonators, as capacitors, at the filter center frequency.

of resonators set on at most two frequencies. Shifting the frequency of individual resonators generally requires another masking step in an already complex process. The down side of the inductor approach is that there is little room for chip or integrated inductors.

2.6 Acoustically Coupled Filters

37

The circuit of Figure 2.18(b) suggests the use of a common mode inductor that is series resonant with the set of shunt resonators. In the filter passband the shunt resonators are going parallel resonant and therefore the inductor has little effect on the circuit. But, off-frequency where the shunt resonators are capacitors, the series resonance has the effect of putting a low-Q rejection notch over the filter, which increases the near-in isolation of the filter. Because of the LC notch, the design of the filter can use fewer elements and produce lower insertion loss in band. 2.5.2

Balanced Ladder

The balanced ladder filter, in Figure 2.19, is simply a mirror image of the single-ended filter. This type of balanced filter does not give as large a bandwidth as a lattice filter but does have the steep near in skirts associated with the single-ended ladder filter. Balanced filters are of increasing interest in system integration with ICs whose amplifier circuits typically have balanced I/O. 2.5.3

Conventional Lattice

The conventional lattice filter, described in Chapter 1, is readily adopted for thin film BAW implementation where a balanced filter is required. In the lattice filter response the balanced network configuration suppresses the normal pole-zero response of a resonator to give a more conventional multipole response. Figure 2.20 shows a four-pole lattice filter with the top view of a die layout. In this configuration the circuit I/Os do not require contact to the electrodes on the bottom side of the piezoelectric. Electrodes denoted by E and F are buried under the piezoelectric plate and the I/O electrodes are all on the top. If implemented on an IC this would afford a compact layout.

2.6

Acoustically Coupled Filters Resonators may be acoustically coupled to yield more or less classical filter responses. Acoustical coupling takes two general forms, one where propagation is perpendicular to the major plate surfaces, and the other where it is transverse (i.e., parallel to the major plate surfaces). The following discussion will focus on the thickness mode form of coupling.

Figure 2.19

Balanced ladder as a mirror image of a single-ended ladder filter.

38

Resonator and Filter Topologies X1A

C

X2B

X2A

B

X1C

E

A

X2D

X2C

F

X1B

X1D

A

B

C

D

E

X1A

X2B

X1C

X2D

F

X2A

X1B

X2C

X1D

D

Figure 2.20 Lattice filter. The upper-circuit diagram gives the general circuit of a two-section 4-pole ladder filter. The lower drawing is a topside view of the filter as laid out for a thin film resonator fabrication. The resonator and electrode designations can be used to correlate the circuit diagram with the layout.

2.6.1

Stacked Crystal Filter

One of the primary thickness-mode-coupled resonators is the stacked crystal filter (SCF) (Figure 2.21 [22–26]). The SCF is composed of multilayers of piezoelectric and metal layers, as shown in Figure 2.21(a) for a one-pole filter and in Figure 2.21(b) for a two-pole filter. Because one transducer is located directly on top of the other, there is little or no impediment for the sound generated by one resonator propagating between the two transducers. A voltage applied between electrode 2 and the ground drives the top transducer. The wave that is generated in the top 1

GND.

2

VIA

VIA

Piezoelectric

Piezoelectric Reflector Layers or Air (a) GND.

1

2

GND.

Piezoelectric Piezoelectric Reflector Layers or Air (b)

Figure 2.21 Cross-sectional views of a simple SCF. In (a) is shown two piezoelectric layers with intervening electrodes for a single section one-pole filter, and in (b) two sections are connected electrically in series to form a two-pole filter.

2.6 Acoustically Coupled Filters

39

transducer propagates through the structure and reflects off the bottom of the bottom transducer. Thus, the acoustic region established between outer reflecting surfaces of the two transducers forms a resonator. In the SCF then, one transducer drives the resonant structure and the other extracts energy from the resonator. The limited frequency range of the externally loaded resonator is the basis of the filter response. Figure 2.21 also shows the electrical shielding between input and output provided by the ground plane. The lowest order resonance, as shown in Figure 2.22, is for a half-wavelength across the entire structure or an approximate quarter-wavelength across each piezoelectric region, and will be denoted as mode 1. Although transduction is not the most efficient when there is only a quarter-wavelength across the piezoelectric, the structure is nevertheless resonant and a filter response is obtained. The most efficient transduction, and hence the greatest effective electrical coupling to the external circuit occurs at the second overtone, mode 2, where there is a half-wavelength across each transducer. The next major response is at the third overtone, mode 3, and coupling is inefficient because each transducer is operating at three-fourths wavelength. Taken together, these resonances have the effect of placing adjacent spurs at the half- and three-halves frequency around the most efficient transduction frequency. The response of the mode 2 SCF can be improved by fabricating in the SMR format on a limited bandwidth reflector array to effectively attenuate the mode 1 and mode 3 frequency responses, as implied by Figure 2.10. It is useful to compare filters designed for mode 1 and for mode 2 resonant structures. Figure 2.23(b) shows in dashed lines an SCF designed for the mode 2 optimal coupling case. Here the SMR format is used to limit the mode 1 and mode 3 responses that occur relatively near-in and would otherwise severely limit the out-of-band rejection. Also, on the plot is a mode 1 filter centered on the frequency of the mode 2 filter. The mode 2 filter is on a nine-layer reflector stack of AlN and SiO2 whereas the mode 1 filter has the simpler air reflector structure. Clearly, the mode 1 filter has higher ultimate rejection over most of the range shown than does the mode 2 filter, Figure 2.23(b). The overtones for the mode 1 filter occur at higher frequencies far removed and there is no need for the SMR reflector’s rejection characteristics. Table 2.1 gives the data for the two cases. For the chosen example, the electrodes are Al, the piezoelectric is AlN, and two sections are connected in series to

Mode 3

Mode 1

Mode 2

Figure 2.22

Modeled response of an SCF showing the three principal resonance modes.

40

Resonator and Filter Topologies 0 -10

S21, dB

-20 -30 -40 -50

1575.42

-60 -70 -80 1475

1515

1555 1595 Frequency, MHz

1635

1675

(a) 0 -10 -20

S21, dB

-30 -40 -50

1575.42

-60 -70 -80 500

900

1300

1700

2100

2500

Frequency, MHz (b)

Figure 2.23 Comparison of filter responses of two stacked crystal filters, one designed for mode 1 operation, solid line, with air isolation (FBAR) and the other an SMR format mode 2 shown in dashed lines. In (a) the near-in response shows a slightly narrower bandwidth for the mode 1 filter, and in (b) the mode 1 filter has better ultimate rejection.

make a two-pole filter. Note that the mode 1 filter is about 2.7 times smaller than the mode 2 filter. If the mode 1 filter is designed using Mo electrodes then it is about 6.5 times smaller. Clearly, if the mode 2 SMR filter were designed with Mo or W electrodes it would be smaller as well. These size comparisons do not include electrode I/O pads which are probably a fixed area for both cases. More important than size, the mode 1 filter can be built in the simpler membrane structures (i.e., no reflector stack is required). Details for the two filters are given in Table 2.1. Figure 2.24 shows the circuit models of a resonator and an SCF. First note that with acoustical coupling, Co is not in parallel with the series branch of the LRC and is across the source and load impedances instead. The center of the filter passband is at, or very near, the series resonance of the series LRC circuit and there is no parallel resonance of Co giving rise to a near-in notch as for the ladder filter. The absence of

2.6 Acoustically Coupled Filters

41

Table 2.1 Comparison of Fundamental and Second Overtone Modes of an SCF at GPS L1 Parameter

Mode 1 SCF Mode 2 SCF

Metal 1 thickness, μm

0.3

0.3

Piezo 1 thickness, μm

1.15

2.91

Metal 2 thickness, μm

0.2

0.3

Piezo 2 thickness, μm

1.15

3.1

Metal 3 thickness, μm

0.3

0.3

3-dB bandwidth, MHz

18.4

22.8

Insertion Loss

1.5

1.22

Resonator Size, μm × μm

156 × 156

257 × 257

Area ratio

0.37

1

Ca/2

2Ra

2La

Ca Co

Ra Co

Co

La (a)

(b)

Figure 2.24 (a, b) Equivalent circuits of simple resonator and SCF. Note that the SCF does not have Co across the RLC series branch and therefore there are no parallel resonances.

the parallel resonance is important because it allows the resonator to be less exacting than resonators used in ladder filters wherein the parallel resonance of the shunt resonators is of paramount importance. The performance of the SCF is determined by only the series resonance of the resonator since there is no parallel resonance. Analysis shows that the conditions for minimum insertion and maximum bandwidth occur when the magnitude of the reactance of Co is equal to the source and load resistance. Part of this is because Ra is proportional to the reactance of Co and Ra limits the in-band insertion loss of the filter. Increasing Co can be used to reduce Ra but that has the effect of increasing current through the shunt element Co. However, the SCF can be inductor tuned to eliminate the shunt current flow through Co. Then Co can be increased to reduce Ra. The process has diminishing returns when the equivalent parallel resistance of the tuning inductor decreases towards the source and load-resistance values. More on tuning will be discussed in the coupledresonator filter section. It is useful to look at the fabrication layout of a simple SCF to get an idea of what the device looks like and what problems might arise in the layout. In

42

Resonator and Filter Topologies

Figure 2.25 the lightly shaded areas are acoustically active as defined by the overlap of the I/O electrodes with the ground plane. The bottom floater electrode is denoted by d, e, f and is rectangular with no cutouts. Over the floater electrode is a piezoelectric layer and on top of that is the ground plane. Consider the overlap of electrode b with the ground plane the excitation region and note that the ground plane is cutout for out-feeds a and h. The wave generated propagates to the bottom piezoelectric and a voltage is generated at electrode d against ground. Thus, electrode d and b must line up very closely to avoid a parasitic resonator. For example, if I/O electrode a was over ground (no cutout in the ground plane) a resonator would be formed between there and the bottom of the lower piezoelectric. Likewise, when electrode e transfers current to the right-hand side resonator there must not be a parasitic resonator formed with the ground plane in the gap region between electrodes b and g. Therefore, the ground plane must have a rectangular cutout corresponding to the gap between b and g. Note that in Figure 2.25(a) the line to denote the cross-section is irregular shaped to better show the electrode overlaps. The ground contact for the device is shown in Figure 2.25(a) as being on just one side of the device but in practice it should be on both sides. Better I/O isolation is obtained when capacitance between input and output is at a minimum, which is not hard to achieve at the die level. 2.6.2

Coupled Resonator Filter

The SCF discussed above is effectively a single resonator with an arrangement for sampling the energy within the resonator. The resonator either operates in mode 1 Ground VIA Bottom Floating Electrode

c

a

b

g

h

d,e,f Ground Plane Openings (a) a

b

g

c

c d

e

f

(b)

Figure 2.25 Stacked crystal filter layout. (a) Top view of the layout, and (b) the side view. The lightly shaded areas are acoustically active. (Note in part (a) the shift in the cross-section indicator line.)

2.6 Acoustically Coupled Filters

43

without optimal electrical to acoustical coupling or in mode 2 which is an overtone. In both cases the effective K2 is limited resulting in narrower bandwidth than that obtained with a ladder filter. The limited bandwidth of the SCF can be overcome by reducing the coupling between the vertically disposed transducers in such a way that they begin to act as independent resonators rather than as a single resonator. The resulting configuration is called a coupled resonator filter (CRF) (Figure 2.26) to distinguish it from the SCF [27, 28]. In this case it is appropriate to blur the distinction between transducer and resonator because the resonators (transducers) are sufficiently decoupled that they can be properly called resonators. However, keep in mind that the bottom of the top resonator is at a plane that is 180° of phase down from the top reflecting surface, and that plane may or may not represent an actual material boundary. In Figure 2.26(a) the acoustically active region defined by electrode overlaps is indicated between the vertical dashed lines. The bottom electrode is patterned such that it is brought out from under the acoustically active region for eventual contacting. Above the bottom piezoelectric is another electrode that too must be brought out for contact. Next, formed in sequence, are an acoustic isolation region, on top of that another electrode, a top piezoelectric layer, and finally the top electrode. As in the SCF, layout must be such as to avoid unwanted parasitic resonators. In Figure 2.26(b) two sections of CRF are connected in series to produce a 4-pole filter. This arrangement allows the lower two electrodes to float and there-

VIA

1'

2'

2

1

VIA Piezoelectric Piezo.

VIA

Coupling Layers Piezoelectric Reflector Layers or Air (a) 1'

1

2'

2

Piezoelectric Coupling Layers

Piezoelectric Reflector Layers or Air (b)

Figure 2.26 Cross-sectional view of a CRF. In (a) a single section CRF is shown, and in (b) two sections are connected in series electrically and in a form that allows the input and output to be independent.

44

Resonator and Filter Topologies

fore significantly simplifies the fabrication process since only two topside VIAs need be formed. The layout of a CRF is shown in Figure 2.27. Here the electrodes are shown alone without any other layout complications such as VIAs. In Figure 2.27(a) the floater electrodes are arranged so that one has a hole in it to prevent a parasitic resonator from forming at the overlap. The top I/O pads are shown staggered as required to eliminate overlap. In Figure 2.27(b–d) top views of the electrodes are shown. In Figure 2.27(b) the input and output areas are equal, in Figure 2.27(c) the output area is smaller for higher impedance, and in Figure 2.27(d) the output is series connected to raise the impedance by a factor of four. In the CRF acoustical coupling between resonators is used to control filter bandwidth. Figure 2.28 illustrates classic resonator coupling responses in the CRF obtained by altering the strength of the coupling between a pair of resonators. In optimal coupling the group delay is flat or slightly quadratic across most of the passband and the VSWR is also slowly varying. If there is too great a degree of isolation between resonators, insertion loss is high and the bandwidth is narrow just as in classical coupled LRC resonators. With coupling beyond critical, the combined resonance is split because of electrical mismatch with the source and load. This it turns out can be used to greatly increase CRF bandwidth through inductor tuning, which will be described later. Electrical interconnection of filter sections provides a way of increasing the multipole response and, for an even number of poles, allows the I/O electrodes to appear near or at the top of the structure for ease of fabrication as shown in Figure 2.26(b). The crossover electrodes for the bottom resonators are independent of the I/O electrodes, in contrast to the SCF wherein the ground electrode is shared. Having independent electrodes for the top resonators, in the CRF, allows the common I/O electrode to be split into two independent electrodes as shown in Figure 2.27. When the I/O resonators are electrically isolated, except for stray capacitance, the filter can be operated in a full balanced mode or as a balanced-to-unbalanced transition.

I/O

I/O

(b)

(c)

(a)

(d)

Figure 2.27 Simplified layout of a two-section CRF. In (a) only the electrodes are shown in perspective view to show how the I/O pads are isolated. In (b) the CRF sections have equal area, in (c) the areas are not equal, and in (d) the right-hand set are series connected to raise the impedance level.

2.7 Wide-Bandwidth Tuned Coupled Resonator Filters

45

Over Coupled

Critically Coupled Under Coupled

Figure 2.28

Coupled-resonator filter response showing the effects of coupling conditions.

A convenient coupler uses a sequence of nominal quarter-wavelength-thick layers whose transmission response is designed to produce the desired resonator coupling. The coupling layers can take a variety of forms with the goal to partially isolate one resonator from the other. Quarter-wavelength-layer sequences provide one option and may be of the same material types as used in a reflector stack. For precise bandwidth control it may not be possible to use a quarter-wave sequence of known materials because no combination is correct. There is simply not a wide choice of materials available having the desired mechanical impedances. In which case, the effective impedance of a layer can be trimmed. For example if the three-layer sequence SiO2/AlN/SiO2 does not provide enough coupling the AlN layer can be thinned from a quarter-wavelength thick and one or both of the other layers increased in thickness such that there remains 270° of phase across the sequence. The effect is to synthesize a new material of lower impedance than AlN. The reflection coefficient between the SiO2 and the “new material” is reduced and the overall transmission increased. Electrodes also have an effect on coupling in part because they affect the acoustic source and load impedance of the resonators. If the electrode material has high mechanical impedance relative to the piezoelectric then the near-in electrodes can actually be part of the coupling layers. For example, the sequence E/SiO2/E, where E is an electrode of Mo or W, might at first appear to be a single-layer coupler, because the metals are not each a quarter-wavelength thick, when in fact it could be operating as a three-layer coupler. The devil is in the details.

2.7

Wide-Bandwidth Tuned Coupled Resonator Filters The equivalent circuit for the SCF, Figure 2.29(a), can be analyzed to give some guidance on bandwidth limitations. As discussed briefly before, minimum insertion loss would be expected to occur when series Ra is small. However, Ra is proportional to the reactance of Co and making Co larger causes larger current flows to ground which limits bandwidth. The optimum condition is for the reactance of Co to be the same magnitude as the source and load resistances, usually 50 ohms. However, parallel resonating Co with a shunt inductor, as shown in Figure 2.29(b), can

46

Resonator and Filter Topologies 2L

2R

Ca/2

Rg Co

RL

Co

(a) 2L

Rg

Lp

2R

Ca/2

Lp

Rp Co

Co

Rp

RL

(b)

Figure 2.29 Acoustically coupled resonator tuning. In (a) the equivalent circuit of an SCF single section, and (b) tuning Co by shunt inductor Lp and its loss element Rp. R is proportional to the magnitude of the reactance of Co, so increasing Co lowers R and the insertion loss of the filter but only if Co is parallel resonated by Lp to prevent excessive shunt current flow.

be used to eliminate the effect of Co over the bandwidth of the series RLC circuit. With Co resonated out of the circuit its reactance can be dropped (keeping it at parallel resonance with a shunt inductor), effectively decreasing Ra and lowering insertion loss. This approach is limited by the finite Q of the inductor because Rp will drop with the decreased inductance required to resonant the increased Co. The above tuning process is limited by the Q of Lp and the value of Rp relative to the source and load resistances. As Rp approaches Rg and RL increased shunt current through Rp increases insertion loss. The equivalent circuit of the CRF is shown in Figure 2.30 along with a tuning circuit similar to Figure 2.29 for the SCF. The equivalent circuit of a single-section CRF is shown in Figure 2.20(a) and the method of shunt inductor tuning in Figure 2.20(b). For the CRF, there is an added degree of tuning freedom through the controlled acoustic coupling of the two resonators. By deliberately acoustically over coupling the resonators, the split in resonant frequency shows up as an apparent electrical mismatch having two peaks in the transmission response, as shown in Figure 2.28. The passband is flattened by tuning out Co and by adjusting the resonator area for a better match with source and load. The design sequence is to first split the resonance so that the outer corners of the two peaks are near the extremity of the desired filter passband, and then adjust impedance levels and tuning to flatten the passband. The simulation of tuned CRFs is shown in Figure 2.31 for two different designs. More details on the CRF will be discussed in a later chapter.

2.8 Hybrid Filters

47

L

R

Ca

Ca

Rg

R

ACOUSTIC COUPLING

Co

L

RL

Co

(a)

L Rg

Lp

R

Ca

Ca

R

L

Rp

Lp

ACOUSTIC COUPLING

Co

Rp RL

Co

(b)

Figure 2.30 Coupled-resonator equivalent circuits. (a) Without tuning and (b) with tuning. Acoustic coupling is used to split the resonances of the two resonators then Co and Lp are adjusted for proper matching and passband shape.

0 2.8

-10

2.6 -20 2.4 2.2

-40

2.0

VSWR

S21, dB

-30

1.8

-50

1.6 -60

-80 400

1.4

800.0000

-70

560

720

1.2

880

1040

1.0 1200

Frequency, MHz

Figure 2.31 Simulation of inductor-tuned two-pole CRF filters. The inductors are in shunt with the I/O and have a Q of 20. The filter bandwidths are 14% and 22% of center frequency.

2.8

Hybrid Filters It should be fairly obvious that filters of one type can be chained with another type to give an overall improved filter response. For example, the near-in response of a SCF can be improved by a ladder filter increasing the near-in skirt selectivity. An example is shown in Figure 2.32 for a GPS L2 filter.

48

Resonator and Filter Topologies 0 -10 -20

S21, dB

-30 -40

-70 -80 1125

1165

1205

1247.600

1207.600

-60

1227.600

-50

1245

1285

1325

Frequency, MHz

Figure 2.32 Effects of cascading filters. Shown are the individual modeled response of a 4-pole CRF, a simple −20-dB out-of-band ladder filter, and their cascaded response. The overall −50 dB bandwidth is less than 25 MHz. The rejection of the windowing ladder filter can be used to increase the near-in rejection while the CRF response provides the out-of-band rejection.

2.9

Summary This chapter has discussed the thin film bulk acoustic resonator topologies of greatest interest for communications and wireless applications. Thin film resonators form the building blocks for several forms of bandpass filters including, ladder, lattice, stacked crystal, and coupled resonator types. The topologies of resonators and acoustically coupled filters were discussed in some detail as to device layout and the implied impact of modern IC processing and manufacturing. The following chapters will give more specific details on filters, filter applications, device processing, and the all important piezoelectric film growth.

References [1] Grudkowski, T. W., et al., “Fundamental Mode UHF/VHF Miniature Resonators and Filters,” Applied Physics Letters, Vol. 39, No. 11, November 1980, pp. 993–995. [2] Lakin, K. M., and J. S. Wang, “Acoustic Bulk Wave Composite Resonators,” Applied Physics Letters, Vol. 39, No. 3, February 1981, pp. 125–128. [3] Nakamura, K., H. Sasaki, and H. Shimizu, “ZnO/SiO2-Diaphragm Composite Resonator on a Silicon Wafer,” Elect. Letters, Vol. 17, No. 14, July 9, 1981, pp. 507–509. [4] Kitayama, M., et al., “VHF/UHF Composite Resonator on a Silicon Substrate,” J. Appl. Phys., Vol. 22, Suppl. 22–3, 1983, pp. 139–141. [5] Nakamura, K., Y. Ohashi, and H. Shimizu, “UHF Bulk Acoustic Wave Filters Utilizing Thin ZnO/SiO2 Diaphragms on Silicon,” J. Appl. Phys., Vol. 25, No. 3, 1986, pp. 371–375. [6] Vale, C., et al., “FBAR Filters at GHz Frequencies,” 45th Annual Symp. of Freq. Cont. Proc., 1991, pp. 332–336. [7] Su, Q. X., et al., “Edge Supported ZnO Thin Film Bulk Acoustic Wave Resonators and Filter Design,” Proc. 2000 IEEE/EIA Int. Freq. Control Symp. and Exhibition, pp. 434–440.

2.9 Summary

49

[8] Lakin, K. M., et al., “Thin Film Resonators and Filters,” Proc. 1982 Ultrasonics Symp, October 27–29, 1982, Vol. 1, p. 466. [9] Petersen, K .E., “Silicon as a Mechanical Material,” IEEE Proc., Vol. 70, No. 5, May 1982, pp. 420–457. [10] Satoh, H., et al., “An Air Gap Type Piezoelectric Composite Resonator,” 39th Annual Symposium on Frequency Control Proc., 1985, pp. 361–366. [11] Seabury, C. W., et al., “High Performance Microwave Air-Bridge Resonators,” 1995 Ultrasonics Symp. Proc., pp. 909–911. [12] Lanz, R., P. Carazzetti, and P. Muralt, “Surface Micromachined BAW Resonators Based on ALN,” Proc. IEEE Int. Ultrasonics Symp., paper P21-4. [13] Krishnaswamy, S. V., “Piezoelectric/Ferroelectric Films for Microwave/MEMS Applications: Historical Perspective, 2005 IEEE Ultrasonics Symp., September 19–21, 2005, Rotterdam, paper 5B-1. [14] Newell, W. E., “Face-Mounted Piezoelectric Resonators,” Proc. IEEE, Vol. 53, June 1965, pp. 575–581. [15] Lakin, K. M., K. T. McCarron, and R. E. Rose “Solidly Mounted Resonators and Filters,” 1995 Ultrasonics Symp. Proc., 1995, pp. 905–908. [16] Dubois, M., et al., “BAW Resonator Based on Aluminum Nitride Thin Films,” 1999 Ultrasonics Symp. Proc., 1999, pp. 907–910. [17] Aigner, R., et al., “Advancement of MEMS into RF-Filter Applications,” Proc. 2002 IEDM Symp., 2002. [18] Lakin, K. M., et al., “Improved Bulk Wave Resonator Coupling Coefficient for Wide Bandwidth Filters,” 2001 IEEE Ultrasonics Symp., paper 3E-5. [19] Larson, J. D., and Y. Oshmyansky, “Measurement of Effective kt2, Q, Rp, Rs vs. Temperature for Mo/AlN FBAR Resonators,” Proc. 2002 IEEE Ultrasonics Symp., pp. 939–943. [20] Lakin, K. M., et al., “Temperature Coefficient and Ageing of BAW Composite Materials,” 2001 Frequency Control Symp. Proc., pp. 605–608. [21] Lakin, K. M., J. S. Wang, and A. R. Landin, “Low Temperature Coefficient Shear Wave Thin Films for Composite Resonators and Filters,” 1983 IEEE Ultrasonics Symp. Proceedings, Atlanta, GA, October 31–November 2, 1983, Vol. 1, p. 491. [22] Ballato, A., and T. Lukasek, “A Novel Frequency Selective Device: The Stacked Crystal Filter,” Proc. 27th Annual Freq. Control Symp., June 1973, pp. 262–269. [23] Lakin, K. M., “Equivalent Circuit Modeling of Stacked Crystal Filters,” Proc. 35th Annual Freq. Control Symp., 1981, pp. 257–262. [24] Stokes, R. B., and J. D. Crawford, “X-Band Thin Film Acoustic Filters on GaAs,” IEEE Trans. Microwave Theory Tech., Vol. 41, No. 6/7, December 1993, pp. 1075–1080. [25] Lakin, K. M., et al., “High Performance Stacked Crystal Filters for GPS and Wide Bandwidth Applications,” 2001 IEEE Ultrasonics Symp. Proc., pp. 833–838. [26] Lakin, K. M., et al., “Bulk Acoustic Wave Resonators and Filters for Applications Above 2 GHz,” 2002 IEEE MTT-S Digest, Vol. 3, pp. 1487–1490. [27] Lakin, K. M., “Coupled Resonator Filters,” Proc. 2002 IEEE Intl. Ultrasonics Symp., Paper 3D-5, 2002. [28] Fattinger, G., R. Aigner, and W. Nessler, “Coupled Bulk Acoustic Wave Resonator Filter: Key Technology for Single-to-Balanced RF Filters,” Proceedings IEEE 2004 MTS Symp. Digest, 2004.

CHAPTER 3

BAW Device Basics Jyrki Kaitila

In this book we make the definition that the thin film bulk acoustic wave (BAW) resonator is a piezoelectric device. This means that the electromechanical conversion is based on the piezoelectric effect. In the literature some other classes of devices, such as CMUTs (capacitive micromachined ultrasonic transducers), are also sometimes called BAW devices, but here we will reserve the term exclusively for the use given above. Piezoelectric effect is an ability of a material to convert electrical energy into mechanical energy and vice versa. Most materials exhibiting this property are crystalline. However all crystalline materials are not piezoelectric: the criterion is the lack of center of symmetry. This is essential as the mechanism of piezoelectricity is based on spatial separation of positive and negative electrical charges under applied stress. Thin film piezoelectric materials will be discussed in detail in Chapter 7. Properties of crystalline materials are inherently complex and they are even more so when it comes to the piezoelectric phenomena. In this chapter we will make no attempt to explain the detailed workings of piezoelectricity. What will be attempted is to give an overview of the relevant topics associated with design and analysis of thin film BAWs. Some of the models that we will use are very simple; someone understanding the real complexity of the covered issues would probably term them naive. We accept this possible criticism, but take the practical view: even if the models and analysis lack ultimate precision, they nevertheless can explain general behavior of real devices with reasonable accuracy, at least qualitatively. Ultimately designing and manufacturing devices is an engineering art. We are extremely pleased if we can bring any insights into how a practitioner can identify the phenomena described in the following pages and apply the solutions offered to the benefit of his devices. This chapter is about resonators. Resonators form part of many different systems: The thin film BAW technology has started out with filters. However, other applications such as oscillators and various kinds of sensors are being envisioned. The basic three parameters that a designer is interested in are usually sufficient effective coupling coefficient, high Q-values, and operation free of spurious resonances (remember the discussion in Section 1.5). This does not mean that all these are the most important parameters for a given application; neither does it mean that there would not be any other considerations to be taken into account. It all depends on the specifications of the task at hand. The first two sections of this chapter will

51

52

BAW Device Basics

build up a rudimentary base for the analysis done in Section 3.3, concerned with thin film bulk acoustic wave resonator design.

3.1

Thin Film Bulk Acoustic Wave Resonator 3.1.1

The Prototype Resonator and Piezoelectric Constitutive Relations

BAW devices utilize piezoelectric effect to generate a mechanical resonance from an electrical input. Conversely, the mechanical resonance is turned into electrical domain for output. Figure 3.1 shows a prototype resonator consisting of a piezoelectric plate of thickness 2d sandwiched by infinitely thin electrodes. Intuition tells us that, if we consider the material having an acoustic velocity of v then the purely mechanical resonance condition of this system is simply ω n = (n + 1) ⋅

π ν ⋅ , n = 0, 1, 2, K 2 d

(3.1)

which is obtained by setting up a multiple of half-wave lengths in the thickness of the plate. The associated stress fields are plotted in Figure 3.1. We are generally interested in the fundamental mode n = 1. Putting in some representative numbers for a traditional crystals, v = 6000 m/s and 2d = 100 μm, we arrive at resonance frequency f = 30 MHz. On the other hand if we are tasked in building an AlN resonator operating at a fundamental frequency f = 2 GHz we arrive at plate thickness of roughly 3 μm. This is a regime where thin films technology quite obviously enters the picture. Before we enter the world of the thin film devices it is necessary to quickly review the basic equations and theories governing piezoelectric resonators. We will not go into too many details here, neither will we offer any lengthy derivations of equations. There are excellent works already available on these topics, see for example books by Auld [1], Ristic [2], or Rosenbaum [3]. However, for the later discussions in this book it is necessary to include the most important aspects for quick and easy reference. The piezoelectric constitutive relations relate the mechanical and electrical variables. These relations are written as T = c E S − eE

(3.2)

D = eS + ε S E

(3.3)

where T is stress, S is strain, E is electric field, and D is electric displacement. These are the field variables. The terms cE, e, and εS are the material parameters: cE is the

2d n=0

n=1

n=2

Figure 3.1 Mechanical resonances in a plate of thickness 2d. The stress fields associated with the resonances are plotted.

3.1 Thin Film Bulk Acoustic Wave Resonator

53

stiffness constant, and it is the parameter c that appears in the original Hooke’s law for nonpiezoelectric material, relating stress T and strain S through T = cS. In the case of piezoelectric medium, Hooke’s law needs to be modified to (3.2) in order to account for the emergence of stress associated with external electric field (i.e., the direct piezoelectric effect). This is achieved through the piezoelectric (stress) constant e. Similarly (3.3) now has a component describing how internal stress contributes to the electric displacement (i.e., the inverse piezoelectric effect) again through the same material parameter e. The material parameter relating D and E in (3.3) is permittivity of the material and is denoted by εS. We have here written the constants cE and εS with a superscripts to emphasize that the constants need to be evaluated under specific conditions. Therefore what cE S denotes is stiffness under constant (usually zero) electric field. Likewise ε gives permittivity under constant strain. Generally, all material parameters have to be defined this way. It reflects the fact that these constants are true constants only when specific experimental conditions are applied when the parameters are measured. Equations (3.2) and (3.3) give one of the four possible ways of expressing the piezoelectric constitutive relations. Instead of writing stress T and electric displacement D as functions of strain S and electric field E [i.e., T(S, E) and D(S, E)], we could have just as well chosen any one of the remaining three permutations between the four variables. This would in each case invoke a new set of material parameters. Obviously these different sets of material parameters are related through some (fairly simple) transformations. The second important equation is the Newton’s second law, familiar from high school physics, relating force with mass and acceleration, F = ma. In the one-dimensional case we can identify the left-hand side with T/∂z · ΔV and the right-hand side 2 2 with ( ρ ⋅ ΔV ) ⋅ ∂ u/∂t , resulting in ∂T ∂2 u = ρ⋅ 2 ∂z ∂t

(3.4)

Here ρ is the mass density of the material and u is the (particle) displacement. In a nonpiezoelectric medium using the Hooke’s law and the definition of strain S=

∂u ∂z

(3.5)

we end up with the wave equation ∂2 u c ∂2 u = ⋅ 2 ρ ∂z ∂t 2

(3.6)

We assume a time dependence of all the fields as exp(jωt). Therefore the wave equation describes a wave propagating with a phase velocity ν=

c ρ

(3.7)

54

BAW Device Basics

It should be emphasized that this velocity is not the particle velocity associated with the particle displacement u, given by ∂u/∂t. We will refer to v in (3.7) as velocity of the acoustic wave, that is in the pure mode cases either the velocity of the longitudinal or the shear wave, denoted later by vL and vS, respectively. In a piezoelectric medium we get from the constitutive relations (3.2) and (3.3) ⎛ e2 ⎞ e e T = c E ⎜1 + E S ⎟ S − S D = c D S − S D ⎝ c ε ⎠ ε ε

(3.8)

Inserting this into the wave equation, (3.4) and utilizing the fact that D is a constant in the dielectric piezoelectric medium one arrives at an acoustic velocity νD =

cD = ρ

cE ⋅ 1+ K2 = ν ⋅ 1+ K2 ρ

(3.9)

This highlights the first effect of piezoelectricity in our system: the acoustic velocity is higher than would be deduced simply from the material parameter cE. In D E D essence the piezoelectric effect stiffens the material (c > c ). Therefore c is sometimes called the piezoelectrically stiffened elastic constant. We have also in the last forms defined the electromechanical coupling factor K2, given by K2 =

e2 c E εS

(3.10)

It depends only on the material parameters and is a measure of conversion efficiency between electric and acoustical domains in the piezoelectric material. Finally, we can now examine a simple prototype resonator to study some real-life consequences of the previous analysis. We will assume a simple piezoelectric plate of thickness 2d with infinitely thin massless electrodes covering the opposing faces. We will assume that the lateral dimensions of the resonator are much larger than the thickness and this will reduce our system to purely one-dimensional case. Looking at the wave equation, (3.6), we can assume a general Ansatz for the displacement as

[

]

u( z, t ) = a ⋅ sin( kz ) + b ⋅ cos( kz ) ⋅ e jωt

(3.11)

where k is called a (vertical) wave number or propagation constant. The constants a and b are determined by the boundary conditions. Inserting (3.11) into (3.6) we have ω 2 ⋅ u( z, t ) =

k2 c D ⋅ u( z, t ) ρ

(3.12)

and the wave number is therefore k=

ω 2π = D λ ν

(3.13)

3.1 Thin Film Bulk Acoustic Wave Resonator

55

In the last form we have identified the wavelength λ. The stress is now given by (3.8), and is

[

]

T( z ) = c D k ⋅ a ⋅ cos( kz ) − b ⋅ sin( kz ) −

e D εS

(3.14)

We have here dropped off the time dependence of our field variables. Throughout this text the term exp(jωt) will mostly be suppressed to keep the presentation more readable. Assuming the boundary condition of vanishing stress at the upper and lower surfaces, T(±d) = 0, gives T( z ) =

⎤ eD ⎡ cos( kz ) ⋅⎢ − 1⎥ S ε ⎢⎣ cos( kd ) ⎥⎦

(3.15)

The associated displacement is (see (3.8)) u( z ) =

sin( kz ) eD ⋅ S c ε k cos( kd ) D

(3.16)

In order to find the response of the system to the outside electrical stimulus we eliminate stress S from (3.8) and (3.3) and solve for E. The result is E=−

⎛1 e e2 ⎞ T − ⎜ − D S 2 ⎟D S ⎝ε c ε ⎠ c ε D

(3.17)

The voltage over the piezolayer is given by the integral of electric field over the thickness of the body. After some lengthy manipulation this becomes V =

+d

∫ E( z )dz =

−d

tan( kd )⎤ 2 dD ⎡ e2 1 ⋅ − ⋅ ⎢ ⎥ S D S kd ⎥⎦ ε ⎢⎣ c ε

(3.18)

If the piezolayer is dielectric the current is purely a displacement current, J = D/∂t. Therefore the current at the terminals is given by I = jωA · D, where A is the area of the device. Now the impedance is given by Z=

V 1 = I jωC 0

⎡ tan( kd )⎤ ⋅ ⎢1 − Kt2 ⋅ ⎥ kd ⎥⎦ ⎢⎣

(3.19)

where we have introduced yet another electromechanical coupling coefficient Kt2 =

e2 K2 = c D εS K2 + 1

(3.20)

This is called the electromechanical coupling factor for the thickness-longitudinal vibration (also called the piezoelectric-coupling constant for transversely clamped

56

BAW Device Basics

material). For rather weak piezoelectrics, like AlN or ZnO, the two coupling constants are approximately equal, Kt2 ≈ K 2 . The static capacitance C0, given by the familiar expression C0 =

εS A 2d

(3.21)

Note that the factor 2 appearing in the denominator is the consequence of defining the thickness as 2d. In most other works the thickness of the plate is given as d, but in order to keep the definition of plate thickness constant throughout this chapter we have opted to use this one. The resonant frequencies are obtained from (3.19). The antiresonances (or parallel resonances) are obtained when Z → ∞ (or when the admittance Y = 1/Z = 0. This gives kd = (2n + 1) ⋅

π , n = 0, 1, 2, K 2

(3.22)

Using (3.13) this becomes ω a , n = (2n + 1) ⋅

π νD ⋅ , n = 0, 1, 2, K 2 d

(3.23)

The resonant frequencies ωr,n are obtained from solution Z = 0 of (3.19). They are therefore obtained from ⎛π ω ⎞ tan ⎜⎜ ⋅ r ⎟⎟ ⎝ 2 ωa ,0 ⎠ 1 = 2 π ωr Kt ⋅ 2 ωa ,0 D

(3.24)

where we utilized v /d solved from (3.22) for the lowest antiresonance frequency ωa,0. It is interesting to note the apparent similarity, but the subtle difference between (3.1) and (3.23). The first one was obtained by simple reasoning without very much hard physics involved. It describes resonances in a purely mechanical system, which means it is a plate-and-hammer model. That is: what waves would be observed if we simply hit the plate with a hammer (assuming that the hammer really is a wide-frequency band stimulus). On the other hand, (3.23) was derived based on the piezoelectric phenomenon. The difference between the obtained resonances is that the antisymmetric modes present in the purely mechanical treatment are missing from the piezoelectric driven case. This agrees with intuition: the antisymmetric modes are not excited because the constant external electric field cannot drive them. This symmetry argument will be used in the later sections when effective coupling coefficient and spurious modes are analyzed.

3.1 Thin Film Bulk Acoustic Wave Resonator

3.1.2

57

The Basic Parameters and Equivalent Circuit

In order to develop an equivalent circuit it is convenient to write the impedance expression in a slightly modified form. It can be shown [2, 3] that (3.19) can be expressed as Z( ω) =

⎡ ω 2 kn2 ⎤ 1 ⋅ ⎢1 − Kt2 − ∑ 2 ⎥ 2 jωC 0 ⎢⎣ n ωa ,n − ω ⎥ ⎦

(3.25)

where we have introduced the coupling of the nth mode kn2 =

8Kt2

[(2n + 1)π]

(3.26)

2

The choice of the equivalent circuit is not unique; many different topologies that bring about electrical behavior as expressed by (3.25) can be envisioned. However, from (3.25) it is clear that the resonator can be described by a capacitance C0 in parallel with an acoustic arm. Parallel to this we can have further motional arms corresponding to the terms in the sum in (3.25). This circuit is the Butterworth–Van Dyke (BVD) circuit and is shown in Figure 3.2(a). Generally, the higher order harmonics are neglected in the basic analysis concentrating on the main resonance and the circuit takes the form shown in Figure 3.2(b). For the simplified circuit the input impedance takes the form Z( ω) =

j( ωL1 − 1 ωC1 )

(3.27)

1 − ω 2 C 0 L1 + C 0 C1

Again we find the series and parallel resonances by requiring zero and infinite impedances, respectively, and these are ωr =

1

(3.28)

L1 C1

and ωa =

C1 + C 0 C = ωr ⋅ 1 + 1 L1 C1 C 0 C0

Rx

C0

L1

L2

L3

C1

C2

C3

(a)

...

C0

(b)

(3.29)

Lx

L1

C0

C1

R0

L1 C1 R1

(c)

Figure 3.2 (a) Multiresonant BVD circuit. Each motional leg corresponds to a resonance. (b) Single resonance BVD circuit, and (c) the modified BVD (mBVD) circuit taking into account losses.

58

BAW Device Basics

The basic BVD circuit does not have any resistive elements and therefore it cannot take into account any losses in the system. This means that the quality factors of our series and parallel resonances are infinite. A more realistic representation is obtained with the circuit given in Figure 3.2(c) [4]. The resistance Rx in series can be associated with the simple, ever-present resistance of the metal electrodes connecting the device. The inductance Lx can arise because of the measurement configuration (device layout on the wafer). The motional resistance R1 is associated with acoustic losses, of whatever origin, in the system. Now the input impedance takes a rather complicated form as ⎡ ⎤ 1 1 Z( ω) = jωL x + R x + ⎢ + ⎥ ⎢⎣R 0 + 1 jωC 0 R1 + j( ωL1 − 1 ωC1 )⎥⎦

−1

(3.30)

We define the quality factors at series and parallel resonances as Qs = −

1 ∂ϕ ωs 2 ∂ ω ω=ωs

(3.31)

Qp = +

∂ϕ 1 ωp 2 ∂ ω ω=ω p

(3.32)

and

where ϕ is the phase angle of the impedance. Therefore we have from (3.30) approximately Qs ≈

ω s L1 R x + R1

(3.33)

and Qp ≈

ω p L1 R 0 + R1

(3.34)

This shows that at the series resonance ωs the main contributors to the Q-value are Rx and R1. The appearance of Rx is expected because of the high currents associated with the series resonance. The benefit of having the three resistors in our modified BVD circuit is that it allows us to better model the situation where the series and parallel resonance Q-values are different. However, in a simple analysis we can calculate only two Q-values from a measured resonator, the series resonance Qs and the parallel resonance Qp, and in the equivalent circuit we have introduced three resistors. Therefore, the choice of distributing the losses among these three elements is not unique. We will explore this theme in the later chapters of this book. Besides resonator analysis the BVD circuit can readily be used in design of filters.

3.2 Basic Physics

3.2

59

Basic Physics 3.2.1 Wave Propagation, Transmission, Reflection, and Attenuation of Acoustic Waves

We saw previously that phase velocity emerges naturally from the wave equation, 1/2 and is given by the stiffness constant c and mass density ρ as v = (c/ρ) . This view, although entirely correct, is unfortunately just an extreme simplification of the real situation. That is because the stiffness constants come in all kinds of varieties. It is a well-known fact that stress in one direction, for example, z, produces strains also in the perpendicular directions, x and y. What this means is that the three-dimensional stress-strain relationship must contain terms allowing for this spatial cross-coupling. This coupling also obeys Hooke’s law, with certain stiffness constants. Under the isotropic assumption, the simplest case, we have three constants describing the system behavior under stress. These are c11, c12, and c44, where 2c44 = c11−c12, meaning that only two of the components are independent. c11 is the primary term linking, just as an example, the z-direction longitudinal stress to z-direction longitudinal strain. The second constant c12 describes the Poisson interaction: how z-direction stress translates into x- and y-direction (shear) strain. Finally, the constant c44 describes the relationship between perpendicular (shear, xand y-direction) stress and strain. It is also possible to write the stiffness constants using another notation as c44 = μ, c12 = λ, and c11 = λ + 2μ. These λ and μ are the Lame constants. Specifically, μ is called the shear modulus (shear-to-shear interaction). Having multiple stiffness constants means having multiple acoustic velocities. Therefore in the isotropic case we have two acoustic velocities given by νL =

c 11 ρ

(3.35)

νS =

c 44 ρ

(3.36)

and

The first velocity vL is the longitudinal velocity: the particle vibration is in the direction of wave propagation. It is also called the thickness extensional (TE) or compressional wave. The last name suggests the nature of the wave: as the wave propagates there are regions of compression and decompression in the material (i.e., local mass density variations). The second velocity vS is the shear velocity. In this case the vibration is perpendicular to the propagation direction of the wave. There are no local mass density variations as the shear wave propagates. Figure 3.3 shows the particle displacements associated with the longitudinal and shear waves. Naturally having two acoustic velocities also means having two different acoustic impedances. They are now given by ZL = (ρc11)1/2 = ρvL and ZS = (ρc44)1/2 = ρvL. With the aid of vL and vS, we can define the Poisson ratio σ as

BAW Device Basics

Propagation

60

λ

λ

(a)

(b)

Figure 3.3 Particle displacements associated with (a) longitudinal, and (b) shear waves. The propagation direction of both waves is up (or down). In the figure we have noted the wavelength λ(Kλ = 2π), see (3.13).

νL = νS

c 11 = c 44

1− σ 12 − σ

(3.37)

When an acoustic wave meets an interface part of it is reflected and part of it is transmitted. The amplitudes of the reflected and transmitted waves are proportional to the acoustic impedance difference across the interface. This can be shown formally by assuming an normal incidence plane wave in material 1, exp(−jk1z), reflected into reflected-backward propagating wave of a · exp(+jk1z) and a transmitted wave in material 2, b · exp(−jk2z). Applying the continuity of displacement and stress we arrive at the amplitude transmission and reflection coefficients t =

2Z 2 Z 2 + Z1

(3.38)

r=

Z 2 − Z1 Z 2 + Z1

(3.39)

and

These coefficients obey the relation 1 + r = t. If the angle of incidence is nonperpendicular the wave will also experience mode conversion. This means that, for example, a longitudinal wave will convert into four waves: reflected and transmitted longitudinal and shear waves. The angles of the reflected and refracted waves follow the Snell law, but the amplitudes for these four waves cannot be written in any simple manner. We will omit them here and simply refer the reader to previous works by Auld [1], or Dieulesaint and Royer [5]. However, one general point should be mentioned for future reference: for the longitudinal and shear waves this mode conversion happens also at a stress-free boundary.

3.2 Basic Physics

61

Acoustical attenuation is a phenomenon where some of the mechanical energy propagating in a material is converted into heat. This can be taken into account by rewriting Hooke’s law in the form T = cS + η

∂ν ∂S = cS + η ∂z ∂t

(3.40)

where η is called viscosity. The first form illustrates the fact that as the wave travels the stress amplitude decreases because of the acoustic absorption. The second form conveys similar message: strain tends to relax toward its equilibrium state with time. For sinusoidal excitation this becomes T = (c + jωη) ⋅ S

(3.41)

where the material parameter is now called complex elastic stiffness. Using (3.40) the wave equation now reads as ρ

∂2 u ∂2 u ∂3 u c η = + ∂t 2 ∂z2 ∂ t∂ z 2

(3.42)

~ Introducing a complex propagation constant k = k + jα the wave equation becomes ~ ~ − ω 2 ρ ⋅ u = −ck 2 ⋅ u + jηk 2 ω ⋅ u

(3.43)

For the real and imaginary parts we get − ω 2 ρ = −ck 2 + cα 2 − 2 ηkαω

(3.44)

0 = −2ckα + ηk 2 ω − ηα 2 ω

(3.45)

and

respectively. Assuming small absorption, that is both α and η are small compared to k and ω, we get from the real part ω = k

c =ν ρ

(3.46)

Therefore, for small absorption the wave velocity is frequency independent and equal to the lossless case. Dropping the last term of the imaginary part (α<<β) we get α=

ηω 2 ω = 3 2Q ν 2ν ρ

(3.47) 2

From the first form we see that the absorption is directly proportional to ω and inversely proportional to v3. This makes sense: for higher frequency and/or lower velocity the displacement peaks and valleys are closer together, implying higher

62

BAW Device Basics

strain, and consequently higher relative absorption. The second form of the formula introduces the material quality factor Q=

ν2 ρ ω = 2 αν ωη

(3.48)

The absorption coefficients are difficult to measure for materials in thin film form. Therefore the values of α used in calculations are intelligent questimates that will result in resonator Q-values representative of measured devices. 3.2.2

Electroacoustic Conversion

In his classic treatment Piezoelectric Crystals [6], referring to the nature of piezoelectric effect, Warren P. Mason states: “plate cut from a piezoelectric crystal with electrodes attached serves not only as a capacitor for storing electrical energy but also as a motor for turning electrical into mechanical energy and as a generator for turning mechanical energy into electrical energy.” In this section we quickly review the tools for understanding this electroacoustic conversion in resonators. Right from the beginning it is important to note the difference between the piezoelectric material-coupling coefficient K2 and the effective coupling coefficient K2eff. The former is a material property. Certain materials are intrinsically better at making the electroacoustic conversion than others: lead zirconium titanate (PZT) has a much higher K2 than AlN, and amorphous silicon dioxide (SiO2) has none at all. On the other hand, the effective coupling coefficient K2eff is the property of a device (i.e., resonator). The material property, K2, of course, influences what kind of effective coupling coefficient might be expected from a manufactured resonator. Therefore, no matter how much we would be tempted to build a resonator using amorphous SiO2 as the piezoelectric layer there is nowhere to go because this material itself shows no piezoelectricity. Likewise it is actually not at all demanding to manufacture a resonator showing poor K2eff from a material having quite adequate K2. The resonator just needs to be designed wrong. In the BAW literature the effective coupling coefficient, K2eff, is almost invariably discussed. It is the most easily measurable quantity depending only on the measured parallel and series resonant frequencies, fp and fs, of a resonator. The definition recommended by the authors of this book for the experimental effective coupling coefficient is (compare to (3.24)) 2 K eff =

⎛π f ⎞ π fs ⋅ ⋅ cot ⎜⎜ ⋅ s ⎟⎟ 2 fp ⎝ 2 fp ⎠

(3.49)

2

In Chapter 8 the practical issues relating to the measurement and evaluation of K eff are given. In this chapter we are dealing with the theoretical aspects of resonator physics and design. We will here adopt a different notation making certain that the experimental and theoretical effective coupling coefficients are not mixed up. Therefore, throughout this chapter we will denote the theoretical coefficient k2eff, with a lower case k.

3.2 Basic Physics

63

The definition we are going to adapt for the calculation of the effective coupling coefficient is the Berlincourt formula [7]. This formula calculates the k2eff directly from the field variables and therefore allows the investigation of the influence of device geometry on coupling. The general definition for the internal energy of a piezoelectric body, having volume V, is U=

1 (TS + ED)dV 2 V∫

(3.50)

where the field variables are defined as before. Using the constitutive relations, S = s E T + dE

(3.51)

D = dT + ε T E

(3.52)

this can be written as a sum of three terms U = U e + 2U m + U d

(3.53)

We identify these as the elastic energy Ue, mutual energy Um, and electric energy Ud. These are given by 1 Ts E TdV 2 V∫

(3.54)

1 (TdE + EdT )dV 4 V∫

(3.55)

1 Eε T EdV 2 V∫

(3.56)

Ue =

Um =

and Ud =

E

We have here used a different set of material parameters, where s is the elastic compliance at constant electric field, d is the piezoelectric (strain) coefficient and eT the permittivity at constant stress (remember the discussion in Section 3.1.1). The electromechanical coupling factor, (3.10), can be expressed using either of the material parameter sets as K2 =

e2 d2 = c E εS sE ε T

(3.57)

The effective coupling coefficient is now defined as 2 keff =

U m2 U eU d

(3.58)

64

BAW Device Basics

Equation (3.58) is a definition and therefore it should be taken at a face value, as long as it makes some sense. And it does: The numerator U2m clearly describes the interaction of stresses and electric field through piezoelectric coefficient d (or e). It tells us how the electrical and mechanical domains couple. The denominator on the other hand gives us the energies stored in the mechanical Ue and electrical Ud domains of the system. That means that it normalizes the coefficient with a suitable factor. In [8] Chang, Rogacheva, and Chou argue that the Berlincourt formula actually does not give consistent results when compared with other methods. What they propose is a so-called energy formula taking into account the electrical boundary conditions in a more rigorous way. Nevertheless, here we will utilize the formula given by (3.58) simply because it is relatively straightforward to use. In this chapter we are interested in the rather generalized ideas and whether the formula used is the exactly correct way of describing the situation is of secondary importance. 3.2.3

Mason Model

Up to this point we have only analyzed the simple prototype resonator having infinitely thin massless electrodes. These models are somewhat useful in designing traditional crystals where the electrode thickness is small compared to the thickness of the piezolayer. However, this is not the case in thin film resonators, where the electrode thicknesses typically form a substantial part of the stack. Therefore we need a model that can easily incorporate them and other layers, such as the mirror in the case of a SMR. This is accomplished by the Mason model. First, we will quickly show the justification behind the Mason model. For the reader interested in more details previous works by Mason [6], Rosenbaum [3], and Ristic [2] can be recommended. Assume a thin, large piezoelectric plate bound by two planes located at z1 and z2. Starting from (3.3) we have the electric field as E=

1 e 1 e ∂u ⋅D − S ⋅ S = S ⋅D − S ⋅ S ε ε ε ε ∂z

(3.59)

The voltage is now, as before in (3.18), given by z2

V =

∫ E( z )dz =

z1

2 dD e − S ⋅ u( z 2 ) − u( z 1 ) εS ε

[

]

(3.60)

where 2d = z2 − z1 is the thickness of the plate. With the aid of the particle velocity, v =∂ u/∂ t = jω, this becomes V =

2d I e ⋅ + ⋅ ν( z 2 ) − ν( z 1 ) ε S jωA jωε S

[

]

where we again used I = jωA · D. Solving for the current we get

(3.61)

3.2 Basic Physics

65

I = jω C 0 V +

eC 0 εS

[ ν( z ) − ν( z )] 2

(3.62)

1

Here the capacitance C0 is again given by (3.21). Equation (3.61) or (3.62) establishes the connection between current I, external voltage V, and the particle velocities v on the surfaces of the plate. Setting up the equations for displacements u(z1) and u(z2) through (3.11), and solving for the coefficients a and b gives a=

1 ⋅ u( z 2 ) cos( kz 1 ) − u( z 1 ) cos( kz 2 ) sin(2 kd )

[

]

(3.63)

1 b= ⋅ u( z 1 ) sin( kz 2 ) − u( z 2 ) cos( kz 1 ) sin(2 kd )

[

]

The mechanical force at the boundaries is given by e ⎞ ⎛ F = −TA = − ⎜c D S − S D⎟ ⋅ A ⎝ ⎠ ε

(3.64)

where (3.8) was used for the stress T. Inserting strain S defined by coefficients a and b into the expression for F and evaluating at the left boundary z = z1 gives F1 =

kc D A eD ⋅ u( z 2 ) − u( z 1 ) − kc D A ⋅ tan( kd ) ⋅ u( z 1 ) + S ⋅ A sin(2 kd ) ε

[

]

(3.65)

D

Identifying the acoustic impedance kc = ωZ, and using I = jωA·D and the particle velocity v = jωu, this becomes F1 =

ZA e ⋅ ν( z 1 ) − ν( z 2 ) + jZA ⋅ tan( kd ) ⋅ ν( z 1 ) + ⋅I j sin(2 kd ) jωε S

[

]

(3.66)

Similar expression is obtained for the force at the right boundary F2 =

ZA e ⋅ ν( z 1 ) − ν( z 2 ) − jZA ⋅ tan( kd ) ⋅ ν( z 1 ) + ⋅I j sin(2 kd ) jωε S

[

]

(3.67)

These equations relate the forces at the boundaries with particle velocities v and the external current I. Equations (3.62), (3.66), and (3.67) establish a connection between the acoustical and electrical variables: acoustical currents v and forces F at the boundaries, and current I and voltage V. Looking at the form of the equations we can identify an equivalent circuit (transmission line) given by Figure 3.4. There are three ports: on the left and right we have the acoustic ports, and the piezoelectric coupling to electrical variables comes through the electrical port. In the case of a nonpiezoelectric plate, (3.66) and (3.67) are exactly the same less the last term. This is quite expected since this is the only term with the piezoelectricity appearing through piezoelectric constant e. Therefore a

66

BAW Device Basics v1

v2 jZ·tan(kd)

jZ·tan(kd) v1−v2

-jZ / sin(2kd)

F1

−C0

F2

C0

I

V

Figure 3.4

Mason model for a piezolayer.

nonpiezoelectric transmission line is represented by the circuit given in Figure 3.4, with only the acoustical ports present. Combining multiple layers is accomplished by cascading transmission line sections of piezoelectric and nonpiezoelectric layers, corresponding to the physical stack under investigation. Figure 3.5 shows such a configuration. The single top electrode at the left is terminated by a free surface that is represented by an acoustic short. The piezolayer has the now familiar three ports allowing coupling to the electrical domain. The layers to the right are the bottom electrode, and in the case of a SMR device, the mirror layers. The final impedance on the right is the terminating impedance of the substrate (assuming a semiinfinite medium). Effects arising from a finite substrate thickness, such as backside reflections (overtones), can be included by assigning a finite thickness to the substrate and again terminating the model by an acoustic short. Using the Mason model it is also possible to construct devices with multiple electrical ports (i.e., multiple piezolayers), simply by adding additional sections with the configuration given in Figure 3.4. Losses are introduced through the use of complex propagation constants introduced in Section 3.2.1.

Top Piezolayer electrode Air (acoustic short)

Bottom First mirror layer electrode

Other layers

Last layer Substrate terminating impedance

Figure 3.5 Mason model for a multilayer resonator. The circuit is terminated in the left by a stress-free surface (an acoustic short) and on the right by a semiinfinitely thick substrate. Layers are represented by individual sections of transmission lines with the piezoelectric layer having the electrical ports associated with it.

3.2 Basic Physics

3.2.4

67

Dispersion Relations and Wave Modes

The Mason model is a one-dimensional treatment of the resonator. That means in essence that the resonator, from mechanical point of view, is assumed to operate in a single mode. This mode is either the longitudinal or shear wave mode characterized by the acoustic velocities and impedances corresponding to this mode only. In this section we will describe the basic idea behind the so-called dispersion relations. Just as with the Mason model, the dispersion relations constitute a one-dimensional model. They describe the system when we explicitly allow for laterally propagating modes. In Figure 3.6 we have again drawn a simple plate. We assume a wave propagating in a nonperpendicular angle and reflecting from the top and bottom interfaces. We can no longer expect pure mode propagation. This means that the longitudinal and shear waves are coupled through the mode conversion upon reflections at the interfaces. The dispersion relations of a simple plate of Figure 3.6 are shown in Figure 3.7. The question these relations answer is what kind of waves does the plate support at a given frequency? The waves are characterized first by their lateral propagation constant β. Also associated with each branch (or rather point on a branch) is a displacement profile u. The wave numbers β can take either pure real, pure imaginary, or complex values. For the so-called shear horizontal modes the first two options are available, either pure real or imaginary. These waves are depicted in Figure 3.8. The real propagation constant describes a sinusoidal propagating wave, in the figure moving to the right. The imaginary propagation constant on the other hand describes a reflected wave (note that this imaginary constant has nothing to do with absorption), described by an exponentially decreasing amplitude. For the longitudinal modes all three possibilities are present. Therefore the simple two-dimensional representation of Figure 3.7(a) is insufficient to explain the real spectrum of longitudinal modes. That is why we have also drawn these in the Figure 3.7(b) using a three-dimensional plot. The dispersion branches of Figure 3.7(b) display a rather complex behavior. Starting from the high-frequency side of the branch labeled L3, one observes the following: Between points labeled A and B the curve lies on the real axis (pure real β), and the wave is propagating in the x direction, for example. Between B and C the values for β are pure imaginary. The wave is again propagating between points C and D, but now with the opposite direction x. At point D the wave number becomes complex until the zero frequency point E at the lower-left corner of the figure is reached. Branches labeled L1 and L2 show much simpler behavior with the former laying in the real plane (a propagating wave at all frequencies) and the latter having a propagating high-frequency portion and a complex low-frequency behavior. The dotted curves, labeled with negative subscripts, are the mirror images of the

λ

Figure 3.6 Wave propagation in a simple plate. Longitudinal plane waves are shown by solid and shear waves by dashed lines. After certain distance λ, the partial waves reconstruct themselves.

68

BAW Device Basics A

f

f

A

L-3 L-2

L3 L2

B

fc,TE1 fc,TS2

B C

L-1

D

C

L1

-

D

fc,TS1 β Im{β}

Re{β}

Re{β}

0

E Im{β} (a)

(b)

Figure 3.7 Dispersion diagrams: (a) two-dimensional representation showing pure real and imaginary wave numbers β, and (b) three-dimensional representation for longitudinal waves showing also complex wave numbers β [1]. In (a) the cutoff frequencies of TS1, TS2, and TE1 are indicated (dispersion is of type I). In (b) each continuous line represents a particular branch. The dotted lines correspond to a wave propagating in the opposite direction of the solid curves.

Propagation

λ (a)

(b)

(c)

Figure 3.8 Propagation modes in a (multilayer) plate: (a) pure real β, (b) pure imaginary β, and (c) complex β. In (a) we have indicated the wavelength of the propagating wave. Usually the wavelength and wave number are coupled through the expression λβ = 2π.

three branches discussed above. They are simply propagating in the opposite direction compared to their positive subscript counterparts. We have also indicated points A through C in Figure 3.7(a) showing only the pure real and imaginary parts of the curves. Point D- is marked with a superscript to emphasize the opposite propagation direction compared to point D in Figure 3.7(b). In the case of simple plates the dispersion relations can be solved analytically for the shear horizontal modes. However, the longitudinal modes (as well as flexural modes) can not be solved in closed form (these are also called Rayleigh-Lamb or just Lamb modes). This is rather unfortunate since the longitudinal modes are the ones that matter most for us interested in thin film BAW devices. In the end this does not

3.2 Basic Physics

69

make much of a practical difference since we will in any case be analyzing multilayer systems where the problem must always be solved by numerical techniques. In a multilayer case the problem is generally solved as follows [9, 10]: In each of the layers i we set up longitudinal and shear waves traveling upwards and downwards, in Figure 3.9 noted as Li+, Si+, Li−, and Si−. These waves propagate at a certain angles θi with velocities vi (generally θi and vi are different for longitudinal and shear waves). Associated with these are certain propagation constants ki = ω/vi in the direction described by θi. The lateral propagation constant β is therefore the projection of ki in the lateral direction (here chosen as x). At each material interface we must therefore consider eight waves: up and down traveling shear and longitudinal waves from above and below. The boundary conditions of continuous displacement and stress are applied at each interface. These couple the wave amplitudes in adjacent layers. The boundary condition for an air interface is again the familiar stress-free condition and in a case of semi-infinite half-space, such as substrate, we set up waves traveling in one direction only (away from the last interface). These eliminate the remaining amplitudes and the system can be solved for the lateral propagation constant β. The lateral component of this propagation, in all the layers, is given by β. It must be a constant, for a given solution at a certain frequency ω, for all the layers. The resulting matrix equations coupling the different layers are therefore usually solved by searching the β(ω)-space for solutions satisfying the boundary conditions. In practice this means finding the zero of the determinant of the transfer matrix describing the layer stack. An important property of dispersion is called the cutoff frequency fc. This is defined as the frequency for which β = 0 for any particular dispersion branch. It divides, in frequency, the cases between propagating and reflecting waves. Certain branches have their real wave numbers higher in frequency that their cutoff frequency [i.e., positive group velocity, for example TE1 (thickness extensional) and TS1 (thickness shear) in Figure 3.7(a)]. For future use we will define these as having type I behavior. Other branches might behave the opposite way: propagating waves have a lower frequency than their cutoff frequency (i.e., negative group velocity). These will be denoted as type II branches [for example, TS2 in Figure 3.7(a)]. It should be emphasized that this is a property belonging to a dispersion branch; it is not the property of a layer stack or a device. However, later we will also define resonators as being of either type I or II, but this just refers to the branch (resonance) that they are supposed to be operating with. +

S3

L3

+

Layer 3

S2

L2

+

Layer 2

S1

L1

+

Layer 1

+

+

-

S1

-

L1 -

S2

-

L2 -

S3

-

L3 +

Substrate

Figure 3.9

+

LS

SS

Setting up a dispersion analysis for a multilayered plate.

z

70

BAW Device Basics

The cutoff frequencies are identical to the resonance frequencies obtained from a one-dimensional model of a resonator. For the thin electrode case of a single plate these are given by (3.22), and for a general multilayer case they can be found by the Mason model. Since these one-dimensional methods use only one wave type some of the cutoff frequencies are obviously missing. The cutoff frequencies of longitudinal and shear modes are easily calculated by using the appropriate material parameters (longitudinal or shear) in the calculation of the frequency response. Usually devices having type I behavior have their longitudinal cutoff frequency fc,TE1 higher than their second shear wave cutoff frequency fc,TS2. The opposite is true in type II resonators. This provides a quick method for calculating the dispersion type using the Mason model. An example of measured dispersion curves of a 1-GHz ZnO SMR is given in Figure 3.10. The main resonance at approximately 900 MHz has the type I behavior. Multiple other modes are also present. The dispersion in Figure 3.10 is qualitatively similar to the simple plate dispersion of Figure 3.7(a), but the details are dependent on the exact layer stack in question (there are roughly 10 acoustically active layers in the measured device). The measurement of dispersion relations is discussed in detail in Chapter 8. 3.2.5

Resonator Design Based on Dispersion Relations

The dispersion relations can be used to design of a simple prototype resonator. We will here introduce a model first described in the 1960s by Shockley, Curran, and Koneval [11, 12]. This dispersion-based picture will enable us to study lateral effects in the resonator. In the beginning of their 1963 paper, Shockley, Curran, and Koneval stated: “If the portion of the wafer surrounding the resonator has a cutoff frequency higher than the exciting frequency, the resulting vibratory energy is essentially confined to 1.2 1.1

Frequency [GHz]

1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.0

0.5

1.0 1.5 2.0 Wavenumber β [1/μm]

2.5

Figure 3.10 Measured dispersion curves of a 1-GHz ZnO resonator. At fs (at 0.9 GHz) there exist at least four excited higher order modes. Figure courtesy of Kimmo Kokkonen (Helsinki University of Technology, Finland) and Tuomas Pensala (VTT, Finland).

3.2 Basic Physics

71

the resonator with an energy distribution decreasing exponentially with distance away from the resonator. This exponential decay is not associated with energy loss but acts to trap the oscillating energy within a confined region.” In Figure 3.11(a) we have drawn such a structure and an accompanying cutoff frequency diagram. The associated dispersion characteristics near the cutoff are shown in Figure 3.11(b) (only the relevant TE1 branches are drawn). We assume a device structure exhibiting type I dispersion behavior, both in the active and outside regions. Consequently, the wave modes having frequency lower than the cutoff frequency fc have imaginary wave numbers and modes having frequency higher than fc have a real wave number. The gray shading in both the cutoff frequency diagram and the dispersion diagram denote imaginary wave numbers. It might be appropriate here to note an analogy with quantum mechanics and the Schrödinger equation: In quantum mechanics the wave is bound to a quantum well by similar argumentation: it has an exponentially decreasing amplitude in the forbidden band (below the allowed energy for a propagating wave) and a sinusoidal wave in the quantum well itself. The cutoff frequency diagrams proposed here have their origin in the quantum mechanical description of waves (particles) bound in potential wells. Both the Schrödinger equation and our mechanical (acoustical) counterpart here are wave equations and even without a rigorous mathematical proof it is quite easy to believe that similar behavior occurs in both cases. Just as any symmetric one-dimensional quantum well will have at least one bound state so will also our symmetric one-dimensional cutoff frequency well hold at least one trapped acoustic mode (i.e., resonance). To begin our analysis, let us assume displacement amplitude u(x, z) separable in spatial coordinates u( x , z ) = u x ( x ) ⋅ u z ( z )

(3.68)

Now the corresponding stress field in the z-direction is given by Tz ( x , z ) = T( z ) ⋅ u x ( x )

(3.69)

2L f

Outside

fc,o fc,o

βo

βa

Active

fc,a

β

fc,a x = −L

x=0 (a)

x = +L

βo

β=0

βa

(b)

Figure 3.11 (a) Geometry and cutoff frequency diagram of a traditional resonator of width 2L (dispersion type I). (b) Corresponding dispersion diagrams of the TE1 modes for the outside and active regions near the cutoff frequencies. Cutoff frequencies of the outside and active areas are denoted by fc,o and fc,a, respectively. In both diagrams the gray shading denotes pure imaginary wave numbers β. In the dispersion diagram the denoted wave vector values, βo and βa, correspond to the lowest trapped mode drawn in the cutoff frequency diagram.

72

BAW Device Basics

The second term, ux, can be understood as a modulating term arising from the (possible) nonuniformity of the z-direction stress field in the x-direction. The vertical stress T(z) can be handled for example in the simple cases by the calculations shown in Section 3.2.2 or in a general case with the Mason model in Section 3.2.3. The separable displacement assumption is quite questionable: the constituent relations, (3.2) and (3.3), are usually written in tensor form explicitly because the variables are not separable as per (3.68); we touched this point already in Section 3.2.1. However, we will here take the practical view: the equations and analysis in the following will be good enough to explain many observed phenomena in real devices. We do not wish to carry along the extra weight by utilizing the complete description given by the full equations. Whenever we run into serious difficulties explaining the observed effects we will try to clearly point out whether our simplifying assumptions are to be blamed. Turning our attention to coupling coefficient we insert (3.69) into (3.54) to (3.56) and (3.58). Now one finds that the coupling coefficient is composed of two separate terms 2 keff = kz2 ⋅ kx2

(3.70)

where the z-dependent contribution is grouped under kz2. The form of the lateral component kx2 follows from the Berlincourt formula [see (3.58)], and is k = 2 x

(∫ E u dx) x

2

x

2 2 ∫ E x dx ⋅ ∫ u x dx

(3.71)

where Ex is the x-dependence of the z-directional electric field. This is usually a constant 1 as the resonator is a parallel plate capacitor. The electromechanical coupling coefficient K2 is included in the term kz2, and therefore does not appear in kx2. Referring to Figure 3.11, and concentrating our attention at the right half of the device, we write the lateral displacement of bound modes in two parts as ⎧ a ⋅ cos( β a x ) −L ≤ x ≤ +L ux ( x ) = ⎨ b ⋅ − x exp β ( o ) x > +L ⎩

(3.72)

where βa and βo refer to the lateral wave numbers, obtainable from the dispersion curves, in the active and outside areas, respectively. a and b are amplitude normalization coefficients and will be found by application of the boundary conditions. At this point we should note that even if throughout this text we call βo an imaginary wave number, it can readily be seen that in (3.68) it is actually a number with a real value. We could of course write the displacement as exp(−|βo|x), but for simplicity we will keep the formulation as given by (3.72). In this Ansatz, (3.72), we have already dropped the antisymmetric displacement amplitudes of form sin(βax), that would be present in a purely mechanical treatment. It is immediately seen from the equation kx2, (3.71), that these antisymmetric cases are not worth investigating from a coupling point of view: When a constant Ex is

3.2 Basic Physics

73

assumed (the resonator of Figure 3.11 is in essence a parallel plate capacitor) the integrand in the nominator of (3.71) would always be antisymmetric, being a product of a symmetric function Ex and antisymmetric function ux, and consequently the integral vanishes. Therefore, as intuition might already have told us, antisymmetric modes cannot couple. This is identical to the discussion previously given in the thickness direction regarding the purely mechanical modes, (3.1) and (3.22) In order to calculate the physical fields of (3.72) we need the boundary conditions at x = ±L. For simplicity we will assume these to be continuity of lateral displacement ux and its first derivative dux /dx. It is evident that the first boundary condition is a correct one. Just as obviously, in a general case, the latter is incorrect; we will revisit this point a little bit later. The condition dux/dx being continuous is adopted here just to keep the analysis simple and straightforward (it is interesting to note that Shockley et al. use exactly the same condition in their analysis [11]). Using the boundary conditions at x = ±L we arrive at the resonance condition β a tan( β a L) = β o

(3.73)

Once again we note, that both βa and βo are functions of frequency, β β(f). For the dispersion type assumed here moving to higher frequency makes βa a larger real number and βo a smaller imaginary number [see Figure 3.11(b)]. For small changes in frequency one can usually take βo as a constant and just look at the effect on βa. However, what comes out of the analysis, no matter how it is done in a particular case, is a resonant frequency f associated with (3.73) through the frequency dependence of lateral wave numbers, β = β(f). This means that also the lateral mechanical boundary conditions select the operating mode of the device. This is quite a natural and expected result: exactly the same thing happens in the vertical direction discussed in previous sections. The displacements for the five lowest modes according to (3.73) are drawn in Figure 3.12. Here we have still retained the antisymmetric displacement profiles of form sin(βax), n odd. One important consequence of (3.73) is that the resonator actually does not resonate at the cutoff frequency of the active area fc,a. In the previous one-dimensional analyses this was the case. For the resonator discussed here, exhibiting type I dispersion, the resonance frequency is higher than the cutoff frequency, f > fc,a. However, in any real case (i.e., resonator having a realistic size 2L) the frequencies are for all practical purposes identical. This means that for designing the resonators for a particular application the Mason model can be readily applied with the required precision. For a very narrow resonator we need to move up higher along the dispersion curves to find the conditions satisfying (3.73). Therefore the ground state for a very narrow device might be well above the cutoff frequency fc,a. For a resonator exhibiting type II dispersion characteristics the opposite is true: the frequency of a narrow resonator would be lower than the cutoff frequency fc,a. For a quick-and-dirty analysis, one may assume a hard-wall model by letting βo → ∞, and then the resonance condition, (3.73), reads as βaL = π/2. As expected, the resonance is obtained when a half-wavelength fits into the width of the resonator. For a real device with a finite outside region cutoff frequency fc,o the resonant frequency of a narrow device approaches, but never exceeds, this cutoff frequency fc,o.

74

BAW Device Basics

f0

n=0

f1

n=1

f2

f3

n=2

n=3

f4 n=4

Figure 3.12 lateral displacement profiles ux,n of the first five modes of the traditional resonator. Because of symmetry reasons only the symmetric modes with n = even couple with the constant driving force of the electric field.

This is because above fc,o the outside region supports a traveling wave with a real β, and consequently no trapped resonance can occur in a laterally infinite system. 1 Whether there exists more than one trapped mode depends on the depth and width of the cutoff frequency well, and also on the properties of the active and outside regions (i.e., specific dispersion characteristics). The depth is simply a function of the cutoff frequency difference between the outside and active regions fc,o fc,a. The width 2L is simply the physical size of the prototype resonator under investigation. We will return to these other higher order modes later in the discussion of spurious modes in Section 3.3.3.

3.3

Device Design 3.3.1

Effective Coupling Coefficient

In an earlier section we pointed out the difference between effective coupling coefficient k2eff and material coupling coefficient K2. This section will deal with k2eff, the device property. We will analyze a real resonator with finite thickness electrodes and show how most of the material coupling coefficient K2 can be extracted to maximize or optimize k2eff. This effect was already discussed in the previous chapter dealing

1.

In this case, if taken literally, our model really fails when a real-world device is considered. Even if we allow a propagating wave in the outside, a resonance, maybe a poor one but a resonance nevertheless, would be observed in a manufactured device. One of the reasons is that our model does not treat the mechanical and electrical boundary conditions correctly (see [13]). However, for the analysis done here it is good enough to explain even the real-world behavior of the system.

3.3 Device Design

75

with electrode metallization (see Section 2.3). We will again make a simplified Gedanken experiment to highlight the physics behind the effect. Let us assume a resonator that has electrodes made of the same material (or similar in terms of acoustic velocity and impedance) as the piezoelectric itself. This simplifies our calculations as the displacement inside the body can be written as u( z ) = a ⋅ sin( kz )

(3.74)

where k is the vertical (z-direction) wave number given by k=

ω 2 πf = ν ν

(3.75)

Here v is the velocity of the acoustic wave, in both the electrodes and piezolayer. Figure 3.13 shows the geometry under consideration. Using the stress-free boundary condition T(±(d + t)) = 0 we arrive at the resonance condition kn ⋅ ( d + t ) = (2n + 1) ⋅

π , n = 0, 1, 2, K 2

(3.76)

The frequencies are now given by the familiar ω a , n = (2n + 1) ⋅

π ν ⋅ 2 d +t

(3.77)

The difference between this and (3.23) arises from the different definition of the total plate thickness (see Figure 3.14). After some simple manipulation we get from (3.58) the coupling coefficient of nth mode π t ⎤ ⎡ cos 2 ⎢(2n + 1) ⋅ ⋅ 8 e 2 d + t ⎥⎦ ⎣ = S E ⋅ 2 ⋅ , n = 0, 1, 2, K t ⎞ 2 ε c π ⎛ (2n + 1) ⋅ ⎜⎝1 − ⎟ d + t⎠ 2

2 keff ,n

2

(3.78)

2

In Figure 3.14 function cos [(2n + 1) · π/2 · x]/[(2n + 1) · (1 − x)], where x = t/(t + d), is plotted for the three lowest modes. For main mode n = 0 the function +(d+t) +d 0 −d −(d+t)

z Electrode

t

Piezolayer

2d

Electrode

t

Figure 3.13 The geometry of a resonator with electrodes, made of the same material as the piezolayer itself. This also corresponds to the situation where infinitely thin electrodes are placed at a distance of t from the surfaces within the piezoelectric body. The stress field is drawn with solid line and the displacement with a dashed line.

BAW Device Basics

Normalized effective coupling coefficient

76

1

n =0

0.8

0.6

0.4

n =1 n =2

0.2

0 0

0.2

0.4 0.6 Thickness ratio t/(t+d)

0.8

1

Figure 3.14 Normalized effective coupling coefficients for the three lowest resonances in a simple prototype membrane resonator as a function of fractional electrode thickness t/(t+d). The electrodes are made of the same material (i.e., same acoustic velocity and impedance) as the piezolayer. The main mode n = 0 has a maximum at t/(t+d) ≈ 0.26, indicated by the dotted vertical line.

achieves a maximum at x ≈ 0.26 and the value of the function at this point is approximately 1.14. This means that in our simple Gedanken experiment the effective coupling coefficient is maximized for electrode thickness of t/(d + t) ≈ 0.26. The higher 2 harmonics have naturally a lower k eff and certain electrode thicknesses produce a vanishing coupling coefficient (for the first harmonic n = 1 at x = 1/3, and for the second n = 2 at x = 1/5 and x = 3/5). This behavior is easily understood when considering the symmetry of the situation; see numerator of (3.78). Next we will consider a more realistic but still simple symmetric resonator with electrodes made from a material different than the piezolayer. The resonator is assumed symmetric with both electrodes having the same thickness t and material properties (see Figure 3.15). The displacement has now to be considered in two parts ⎧ a ⋅ sin(kp z) −d ≤ z ≤ +d u( z ) = ⎨ ⎩b ⋅ sin( ke z + γ ) z > + d

+(d+t) +d 0 −d −(d+t)

(3.79)

z Electrode

t

Piezolayer

2d

Electrode

t

Figure 3.15 The geometry of a resonator with electrodes. Example corresponds to the case of Zp > Ze. The stress field is drawn with a solid line and the displacement with a dashed line.

3.3 Device Design

77

where the wave numbers kp and ke refer to the piezolayer and electrodes, respectively. The phase term γ preserves the generality of our Ansatz. Using the boundary conditions at the interface z = +d and vanishing stress at z = +(d+t) we arrive at the resonance condition Ze ⋅ tan(kp d ) tan( ke t ) = 1 Zp

(3.80)

1.4 1.2 2:1 1 1:1

0.8 0.6

1:2

0.4 0.2 0 0

0.2

0.4 0.6 0.8 Thickness ratio t/(t+d) (a)

1

Normalized effective coupling coefficient

Normalized effective coupling coefficient

Here Ze and Zp are the acoustic impedance of the electrode and piezolayer, respectively. We could calculate the stress profiles associated with (3.79) and insert these into the Berlincourt formula for a general expression of the effective coupling coefficient (the result, however, looks rather unappetizing). Unfortunately it is not possible to further insert the resonance condition, (3.80), into the obtained equation for a closed form solution of k2eff. Therefore, we will have to be satisfied with a numerical solution for some representative cases. From the figures we see how the effective coupling coefficient depends on the material properties of the electrodes. Most notably it is seen from Figure 3.16(b) that the high acoustic impedance electrodes increase the maximum achievable k2eff. With the available high impedance electrode materials, which are close to the impedance ratio 3:1 given in Figure 3.16 (assuming AlN or ZnO piezolayer), the normalized effective coupling coefficient has a maximum value of approximately 1.19. This has some important real-life consequences when it comes to practical design of resonators. This issue will be discussed in a later chapter. We have here analyzed a FBAR device. The results in this case do not directly apply to a SMR. This is because the stress fields inside the mirror layers lower the achievable k2eff. However, the general trend seen in Figures 3.15 and 3.16 holds just as well to the SMR, even if the exact position and value of the maximum k2eff depend on the specific mirror configuration. Simulations in the case of a SMR device can be found in the original paper describing the effect by Lakin et al. [14].

1.4 1.2 3:1 1 1:1

0.8 0.6

1:3

0.4 0.2 0 0

0.2

0.4 0.6 0.8 Thickness ratio t/(t+d)

1

(b)

Figure 3.16 Normalized effective coupling coefficients as a function of fractional electrode thickness t/(t + d) for resonator with different electrode to piezolayer (a) acoustic velocity ratio and (b) acoustic impedance ratio. The 1:1 case corresponds to (3.78), n = 0.

78

BAW Device Basics

3.3.2

Loss Mechanisms and Q-Values

Quality factors are a measure of losses in the system. The very basic definition follows from the ratio of the total energy in the system to the power lost in a half-cycle: Q=

ω Etot ⋅ 2 ΔE

(3.81)

where Etot is the total energy and ΔE is the power lost per half-cycle. Assuming multiple loss mechanisms, the overall Q-values follow the well-known law 1 = Qtot

1

∑Q i

(3.82)

i

where Qi is the Q-value associated with loss mechanism i. If we assume that the energies associated with loss mechanisms are different we get from the definition of Q-value, (3.81), E 1 1 = ⋅∑ i Qtot Etot i Q i

(3.83)

where Ei is the energy associated with loss mechanism i, and Etot = ∑ E i . These loss mechanisms in a BAW resonator can be grouped under three categories: 1. Electrical losses; 2. Acoustical attenuation; 3. Leaking waves. Electrical losses are associated with finite resistance of the resonator electrodes and leads connecting resonators and bonding/probing pads. In the previous section we saw that the preferred electrode materials for wide-bandwidth filters are high acoustic impedance metals, typically from the group of refractory metals. Unfortunately, these have rather high electrical resistances. It must also be kept in mind that for the high acoustic impedance metals the best obtainable thin film resistivities are generally 1.5 to 2 times higher than the corresponding bulk values. This is not the case with aluminum, copper, and some noble metals (Ag, Au) that usually have resistivities fairly close to bulk values even in thin film form. When electrical resistivity is concerned there are two major effects that should be considered: First the traditional resistivity of the leads and electrodes themselves. Generally there is about one square of electrodes associated with a resonator (i.e., the top and bottom electrodes themselves). The resistive loss mechanism is obviously most pronounced near the operating point where electrical currents are largest, see (3.33). The second resistive part arises from the possible nonuniform stress distribution over the area of the electrodes when the resonator is operated at high frequencies. If the stress distribution is not flat then areas of the resonator vibrating at different amplitudes and/or phases will have redistribution currents associated with them. This means that even if no current would be flowing in and out of the resonator

3.3 Device Design

79

leads connecting it to the outside world these redistribution, or eddy currents, still have a resistance associated with them. FEM simulation of the phenomena can be found in [15]. In the next section dealing with spurious resonance-free resonator design we will encounter a design that tries to accomplish constant displacement (stress) amplitude over the face of the resonator. Even in the cases where this is successfully accomplished and the spurious modes are not generated, one will almost invariably find small ripple in the measured stress distribution near the operating frequency of the device. This ripple arises because some of the energy is coupled into high lateral wave vector dispersion branches (more about this after few lines). At the moment it is unknown whether this effect has any practical Q-limiting value. Previously we described acoustical attenuation as a phenomenon where some of the mechanical energy propagating in a material is converted into heat. We can readily model this effect by assuming a complex propagation constant. However, as was argued before, we have very little knowledge about the actual values of attenuation constants α at the relevant frequencies. Therefore it is difficult to judge whether this effect comes into play in practical devices. However this is the mechanism that is often used in, for example, Mason’s model, to fit measured and predicted Q-values. The third loss mechanism group, leaking waves, can be broken down to three more general subclasses. The first one is SMR specific: waves lost through the mirror into the substrate. In a FBAR device both the upper and lower surface of the resonator are in contact with the air. With the large acoustic mismatch we can be certain that no wave is transmitted into the air and consequently into the substrate. But, in the case of a mirror device, this might not be so. A Mason model calculation gives the longitudinal wave reflectivity of the typical quarter-wavelength mirror as 99.99% (unity for all practical purposes). This would imply Qs easily in excess of 10,000. However, the mirror in the λ/4-configuration is designed for longitudinal waves. Remembering that the shear wave velocity vS is roughly half of the longitudinal velocity vL, it is follows that the mirror is more or less λ/2 for the shear waves. If, for whatever reason, shear waves are generated in the device these can readily pass through the mirror and consequently lower the Q-values. This loss can be significant for even small amounts of energy associated with the shear waves. This can easily be shown by using (3.83): if one assumes a longitudinal Q-value of 10,000 and shear Q-value of 10, and assigns 1% of total energy to the shear waves, the resulting total Q will be roughly 900. In Figure 3.17(a) we have calculated the mirror transmissivity for the traditional λ/4-configuration (transmissivity being T 2 = 1 − R2, where R is the reflectivity). As expected the longitudinal transmissivity shows a minimum at the resonance (denoted by the vertical line). However, the shear wave transmissivity is high validating our earlier expectation. This leakage can be confronted effectively be designing the mirror to reflect both the shear and longitudinal waves. Such a mirror is no longer based on λ/4-thick layers; an example is given in Figure 3.17(b). Currently there are no reliable explanations for the generation of shear waves. However, as always, it is easy to speculate on the origin of these modes. It was previously mentioned that any nonperpendicular longitudinal waves incident upon a material interface will convert into reflected and transmitted longitudinal and shear

BAW Device Basics

Mirror transmissivity [dB]

80 0

0

−10

−10

−20

−20

−30

−30

−40 0.5

1.0

1.5

2.0

2.5

Frequency [GHz] (a)

3.0

3.5

4.0

−40 0.5

1.0

1.5

2.0

2.5

3.0

3.5

4.0

Frequency [GHz] (b)

Figure 3.17 Transmissivity of (a) a λ/4-mirror and (b) a cooptimized mirror for both shear and longitudinal waves [16]. The longitudinal transmissivity is given by the solid lines, and the shear by the dashed lines, respectively. The resonance frequency of the device (approximately 1.8 GHz) is indicated by the vertical line. Both mirrors are constructed as 2.5-pair W/SiO2-layers.

waves. There should be no a priori reason to expect that such waves would not be present in a real device. Remember again that the Mason model is a 1D treatment and there quite obviously exists no mechanism for generating these modes. Possible sources for the nonperpendicular longitudinal waves are for example the electrode edges where boundary conditions might require generation of these waves. This brings us to the second possible source of the shear waves: there is again no a priori reason to believe that these lateral boundaries would not directly create shear wave components. The second leaking-wave loss mechanism is laterally leaking waves. In Section 3.2.4 we proposed a simple dispersion model for explaining the energy-trapping principle. We found that if we consider only the thickness extensional modes near the cutoff frequencies of the active area and outside, perfect energy trapping can be accomplished. It is time to review that assumption. When we make the resonator analysis based on the dispersion relations we explicitly disregard all other dispersion branches except the one under consideration, usually the thickness extensional mode TE1. In essence we assume that there is no coupling from this mode to the others. In an infinite bulk material the coupling between the modes does not occur. However, in any realistic analysis the resonator edges must come into play. The lateral boundaries require the same continuity of displacement and stress as do the vertical boundaries. However, even in the simple model at the boundary between the active and outside regions of our prototype resonator the stress fields are obviously different, even within the piezoelectric layer itself. This arises because the vertical boundary conditions require vanishing stress on the free surfaces. In other words on the side of the outside region, the vertical stress must equal zero, but on the other hand, on the side of the active area it generally does not have to (see Figure 3.11). Therefore, there must be an additional mechanism to facilitate for this mismatch. In the dispersion picture this can only be brought in by adding a contribu-

3.3 Device Design

81

tion from the other branches on the dispersion diagram at the relevant frequency (what else would there be?). Some of these other branches might have a real wave number. Therefore, almost invariably we will set up traveling waves in the outside region and these will propagate and be lost thus lowering our Q-values and possibly the K2eff. It is an experimental fact that laser interferometer analysis of any resonator will invariably show waves propagating in the outside region as well. Figure 3.18(a) shows such a phenomenon in a mirror device. Just as invariably small wavelength ripple will be present in the active area; see Figure 3.10 for a laser interferometer measurement of the active area of a ZnO resonator operating at 1 GHz (the fact that we are able to measure the dispersion curves means that the waves must be present). Also, finite element analysis (FEM) of devices always shows such a behavior; see Figure 3.18(b). FEM is a brute force method that will, when done correctly, reveal all the possible modes. This is actually a problem in setting up a good FEM simulation, as these waves propagating in the outside must be somehow absorbed at the outmost boundaries of the model, in order not to set up additional unphysical standing wave patterns and corresponding resonances. So how big is the lateral leakage effect in real life? The sad answer is that we do not know. It is very difficult to distinguish between the different Q-loss mechanisms in the electrical measurements, except in some special cases. Laser interferometry can not help us much either: it is difficult to judge what the amount of energy carried away from the resonator with the lateral waves really is. It has been argued by some authors, quite convincingly in fact, that in the case of FBAR this loss mechanism is dominant at certain frequencies (see Chapter 5 and [19]). Nevertheless it must be concluded that definite assignment of a certain Q-value to lateral losses is challenging to say the least. Finally surface and/or interface scattering can be important if the surface and/or interface quality of the materials is not good enough. Here the mechanism is in essence the same as in vertically or laterally leaking waves, we just make the separation because in this case the origin of the leaking waves lies with waves generated by the local roughness. These waves can have any propagation direction and therefore,

(a)

(b)

Figure 3.18 (a) A laser interferometer analysis of a SMR showing propagating modes outside the resonator active area [17], and (b) FEM simulation of a FBAR showing propagating modes on the right-hand-side outside area [18].

82

BAW Device Basics

in principle, it is possible to convert into various different lateral and vertical waves that might or might not be lossy. 3.3.3

Spurious Modes

In this section we first develop the hard-wall model mentioned in Section 3.2.5 further to explore the emergence of spurious modes. Let us first elaborate further on the hard-wall assumptions: The lateral wave is not allowed to penetrate the outside region at all (see Figure 3.11). This is accomplished by letting the lateral wave number in the outside region βo → ∞. As previously shown, we now get the resonance condition as β a L = (2n + 1) ⋅

π , n = 0, 1, 2, K 2

(3.84)

corresponding to the poles of the tangent function of (3.73). Inserting this into the general form of the lateral contribution to effective coupling coefficient kx2, obtained by inserting (3.72) into (3.71), we have kx2 =

2 ⋅ sin 2 ( β a L) ⎛ ⎞ 1 β L ⎜1 + ⋅ sin( β a L) cos( β a L)⎟ βa L ⎝ ⎠ 2 a

(3.85)

2

Now we find a series of resonances, labeled with n, with kx2, n =

8 1 ⋅ 2 π (2n + 1) 2

n = 0, 1, 2, K

(3.86)

This equation describes a series of modes having decreasing effective coupling coefficients. We identify the n = 0 case with the largest coupling as our main mode. The modes labeled with n > 0 are the spurious modes. They are excited because the driving force of the electrical field Ex, a constant across the face of the essentially parallel plate capacitor structure under investigation, is able to couple into these laterally symmetric modes as well (see Figure 3.12 for the mode shapes). In other words the convolution integrals in the numerator of the coupling coefficient equation, (3.71), have nonzero values for all symmetric solutions of the lateral wave modes ux. The wave modes ux are sketched in Figure 3.13. It is interesting to note that ∞

∑k

2 x,n

=1

(3.87)

n=0

This means that all the piezoelectricity available is consumed in the series of resonances composed of the main mode and the spurious modes. In Figure 3.19 electrical measurements of two resonators are shown. The first one, Figure 3.19(a), is an AlN SMR exhibiting type I dispersion. The spurious modes are identified as the smaller loops superimposed on the large main resonance. Because the resonator is of type I the spurious modes are located above the series res-

3.3 Device Design

83

(a)

(b)

Figure 3.19 Spurious modes in (a) type I resonator and (b) type II resonator. (From: [21]. © 2001 IEEE. Reprinted with permission.)

onant frequency fs. This is where the active area of the resonator has real wave numbers βa enabling standing wave patterns according (3.73) to be set up. The opposite is true for the resonator of Figure 3.19(b). This example is an AlN-FBAR (i.e., a membrane resonator) having type II dispersion. Since the real wave numbers are now located below the cutoff frequency, the spurious modes appear below fs in frequency. Note that (3.84) to (3.86) hold for both type I and II devices; the only assumption we have made is βa real and βo → ∞, and in principle this can be accomplished with either dispersion type by proper resonator design. In [22] Kokkonen and Pensala studied a ZnO resonator exhibiting strong spurious modes using electrical measurements, laser interferometry, and FEM. Their findings verify the theoretical reasoning given earlier: the acoustical spurious modes are responsible for the spurious modes seen in the electrical response. Figure 3.20 shows the connection between the electrical resonances and the associated wave patterns obtained from laser interferometry. We can also set up a second interesting problem with (3.73). This takes place when we let βo = 0. In this case we get a coupling coefficient ⎧1 n = 0 kx2 = ⎨ ⎩0 n ≠ 0

(3.88)

This describes a resonator with only one single mode excited. That means it is spurious resonance free. The condition β 0 seems ridiculous at first glance. It describes a constant displacement amplitude in the outside region. This is clearly a violation of the problem statement requiring an energy-trapped structure with an exponentially decaying amplitude in the outside. Actually this also corresponds to a case of infinitely large resonator, and this is exactly the configuration that, for example, the Mason model analyzes. In a laterally infinite resonator there are no edges and therefore no possibility of setting up laterally trapped waves and consequently no spurious modes are excited. But nevertheless, the condition o 0 holds an important truth even in the case of a laterally finite resonator: if we are able to create a situation where the boundary

84

BAW Device Basics

-2

Mechanical response [a.u.]

Absorbed power (1-|S11|2) [dB]

0

−4 −6 −8 −10 −12 920 925 930 935 940 945 950 955 960 965 970 Frequency [MHz]

Figure 3.20 Connection between the measured electrical response and wave patterns obtained by laser interferometry [22]. The solid line is the electrical power absorbed in the resonator as a function of frequency. The dotted line gives the mechanical response as summed over measured lateral wave vector β values at each frequency.

condition of dux/dx = 0 at x = ± L is satisfied, we have created a resonator with no spurious modes excited. This spurious resonance-free condition can be achieved by the type I resonator structure given in Figure 3.21. By inserting a narrow border region, having a real wave number b, between the active area and the outside we L is satisfied. In this case we have a can create a situation where dux/dx 0 at x 0, b real and o imaginary, simultaneously fulfilling all the requirements of energy trapping and boundary conditions. We must now assume the displacement profile as being composed of three parts, corresponding to the three regions of the device. The Ansatz for displacement, again immediately dropping the noncoupling antisymmetric solutions, reads as

W

W 2L f Outside Active fc,o

fc,o

βo

fc,a

β b βa

Border

fc,b

fc,a fc,b

β x=0 (a)

βo

βa = 0

βb

(b)

Figure 3.21 (a, b) The structure and cutoff frequency diagram of a spurious resonance-free resonator, type I. The corresponding dispersion characteristics display the operation point with b real, imaginary and a = 0. o

3.3 Device Design

85

⎧ a ⋅ cos( β a x ) −L < x < +L ⎪ u x ( x ) = ⎨b ⋅ cos( β b x + γ ) + L < x < +( L + W ) ⎪c ⋅ exp( − β x ) x > +( L + W ) o ⎩

(3.89)

Coefficients a, b, and c are again for the amplitude normalizing reasons. The phase term γ in the border region, and the sinusoidal form of the displacement in the active area have been introduced to preserve the general nature of the Ansatz. Investigating the special case of a = 0, meaning a constant amplitude a in the active area, we arrive to the resonance condition. This is achieved by requiring both L and x (L W). After some simple manipulaux and dux/dx continuous at x tion we now have the resonance condition β b tan( β bW ) = β o

(3.90)

Comparing (3.90) and (3.73) for the traditional design a striking resemblance is seen: the equations have exactly the same form with L being now replaced by W and ba by bb. This is no coincidence: One can imagine accomplishing the structure by first designing a narrow (width 2W) resonator using the border-region layer stack according to (3.73). The resonant frequency of this narrow device will be well above the cutoff frequency fc,b of the layer stack (for type I device). Now if one splits this resonator in the middle, where quite obviously dux/dx 0 for the symmetric modes, and in between inserts an active region having a cutoff frequency fc,a exactly matching the resonance frequency of the narrow resonator of our Gedanken experiment, the active area will be operated exactly at this cutoff frequency. And, this in other words means a 0. Therefore the resonance condition, (3.90), must read as it does with no dependence on the width of the active area L or the active area lateral wave number a. Again note that (3.89) and (3.90) apply to both type I and II devices, reasoning as before. Concentrating our efforts solely in the active area (i.e., assuming no coupling in the border area, or rather W « L) we can now find the coupling coefficients corresponding to the main mode and the spurious modes. The displacement in the active area reads as ux,a = a · cos(βa x) for the symmetric modes. The solution satisfying L must have βaL = n · π, with n = 0, 1, 2, .... Therefore, we do dux/dx 0 at x indeed get the coupling coefficient of (3.88). The higher order symmetric modes vanish, because for n 0 the active area now supports a multiple of half-wavelengths (see Figure 3.22). We have here not explicitly made the simplifying assumption of holding both βb and βo constant across frequencies, because nothing in the mathematics themselves forces us to do so. However, it should be noted that generally the condition of (3.90) should be fulfilled over the frequency range of interest, in the vicinity of fs and fp. Whether this happens in real devices is an open question: our idealized model does not take into account all the real-world phenomena and this leaves us only with experiment to judge whether or not we can place any faith in it. Luckily, the experimental observations do validate the model. Electrical measurements of resonators show a minimum in the spurious mode content for a certain W. Direct observation of the effect is possible with laser interferometry. In Figure 3.23 electrical measurement of type I ZnO and AlN SMRs with and without appro-

86

BAW Device Basics

n=0

n=1

n=2

n=3

n=4

Figure 3.22 The resonant modes of the spurious resonance-free resonator. Only the main mode, n = 0, couples with the electric field. The odd modes are antisymmetric and cannot couple. The higher order even modes have a multiple of half-wavelengths across the resonator, −L ≤ x ≤ +L, and consequently do not couple (W « L). Note that for clarity we have drawn the border area width W much larger than in typical devices.

(a)

(b)

Figure 3.23 (a, b) Electrical measurements of type I 1-GHz ZnO and 2-GHz AlN resonators with (solid line) and without border rings (also known as the frame or overlap) [20]. The devices with correctly dimensioned borders have superior characteristics with almost no spurious modes. The improvement in Q-values is also evident.

priately dimensioned borders [20]. Figure 3.24 displays laser interferometer analysis of two ZnO resonators. The resonator on the right shows a flat displacement profile corresponding to βa = 0. Furthermore, the values for W given by (3.90) agree fairly well with experiment.

3.3 Device Design

87

(a)

(b)

Figure 3.24 Laser interferometer measurements of the lateral displacement pattern of a 1-GHz ZnO SMR (dispersion type I) [20]. Part (a) shows displacement profile of a traditional design without the overlap. Part (b) shows a flat profile for the resonator with a correctly dimensioned overlap. The measurement frequency used here is slightly above the series resonant frequency fs, of the devices.

In principle the border region can be designed with any width-thickness combination satisfying (3.90). The general trend seen from (3.90) is that for higher value of lateral wave number βb in the border region the width W must be made smaller. Bigger βb means larger cutoff frequency difference fc,a − fc,b (i.e., the cutoff frequency well is deeper). It was previously commented in Section 3.2.4 that any symmetric one-dimensional cutoff frequency profile will hold at least one bound state. The cutoff diagram for the border region in Figure 3.21 is not symmetric. Therefore, whether there exists an isolated mode in this region depends totally on each individual case. When the border region is correctly dimensioned with respect to thickness and width there are no isolated modes, the lowest frequency mode of a type I device has the shape shown in Figure 3.22, n = 0. However, if the border region is too wide and/or too deep an isolated resonance can be formed. What now happens is that there is an exponentially decaying wave in both the outside and active region (i.e., both βa and βo are imaginary), and βb is real. In this case the electrical measurements will show, again in the case of a type I device, an additional resonance below fs, along with a reduced effective coupling coefficient. In Chapter 5 dealing with the FBAR devices we will see examples of the application of the frame concept to a type II membrane device. As expected, in this case the frame (border area) needs to have a cutoff frequency fc,b higher than the active area, in order to achieve a real wave number βb at resonance. Furthermore improved energy trapping in these devices is achieved through the use of an outside area having cutoff frequency fc,o lower than the active area fc,a, again in agreement with the prediction of the dispersion-based model. One could visualize the situation as in the cutoff frequency diagram of Figure 3.25. The reader familiar with semiconductor physics might see this similar to the representation of the valence band of a quantum well device (such as a quantum well laser), as opposed to the conduction bandlike behavior of the type I device in Figure 3.21.

88

BAW Device Basics W

W 2L f Active fc,b

Border

fc,a fc,o fc,b fc,a

Outside βo

β b βa βo

fc,o

βa = 0

β

βb

x=0 (a)

(b)

Figure 3.25 (a, b) The structure and cutoff frequency diagram of a spurious resonance-free resonator, type II. The corresponding dispersion characteristics display the operation point with βb real, βo imaginary, and βa = 0.

3.3.4

The Other Important Parameters

The most important of the other parameters concerning a resonator is its temperature coefficient of frequency (TCF). We can examine this for the prototype resonator by inserting (3.7) into (3.23) and differentiating with respect to temperature. We get 1 dω 1 ⎛ 1 dc 1 dρ ⎞ 1 dd 1 ⎛ 1 dc 1 dV ⎞ 1 dd ⋅ = ⎜ ⋅ − ⋅ = ⎜ ⋅ + ⋅ ⎟− ⋅ ⎟− ⋅ ⎝ ω dT 2 ⎝ c dT ρ dT ⎠ d dT 2 c dT V dT ⎠ d dT

(3.91)

where T is the temperature and V is the volume and the other symbols are as before. In the last form we can identify the linear thermal expansion coefficient αl = 1/d · dd/dT and the volumetric thermal expansion coefficient αV = 1/V · dV/dT. Both of these are usually positive and therefore they tend to cancel each other to some extent. In an isotropic case αV = 3αl holds fairly well, and therefore the net contribution of the last two terms is roughly +1/2 · αl. Typical values for αl range between +1 to +20 ppm/K. The first term describes how the stiffness changes with temperature. Usually materials become softer as temperature rises. This implies that generally 1/c · dc/dT should be negative and that is the way it is for most materials. Temperature coefficients of stiffness constants are typically in the range of a few tens to a few hundreds of ppm/K (negative). It could therefore be argued that this effect usually is the largest component determining the TCF of a resonator. In a thin film BAW the situation is not really well described by (3.91). It might be a good approximation in the case of FBAR, but in the case of SMR the intimate contact with the substrate complicates the situation. As the substrate thickness is typically up to two orders of magnitude larger than the film stack on top of it, it can not be neglected: The thermal expansion of the substrate (area expansion in this case) changes the stress state of all layers deposited on it (these layers might have some intrinsic stresses in them at the beginning). These stresses (strains) can influence the material parameters: remember the discussion in Section 3.1.1.

3.4 Summary

89

The most straightforward method to account for temperature effects in thin film BAWs is to model them directly through the use of the temperature coefficient of velocity 1/v · dv/dT. The travel time through the various layers changes with velocity and to the first-order this describes the TCF of the resonator. In the accurate description one should also take into account the temperature coefficient of acoustic impedance, 1/Z · dZ/dT = 1/v · dv/dT 1/ρ · dρ/dT. Here the two terms do generally have negative temperature behavior resulting in 1/Z · dZ/dT being also negative. However, since the acoustic impedance influences the resonance frequency only through the reflection and transmission coefficients r and t, (3.38) and (3.39), its temperature coefficient can safely be neglected. To elaborate, as the impedances on both sides of the interface have similar temperature behavior it can safely be assumed that the effect of T on r and t, these being functions of the ratio between the two acoustic impedances Z1 and Z2, is for all practical purposes zero. Finally it should be noted that some materials do indeed have a positive temperature coefficient of velocity (or stiffness). Most notably amorphous SiO2 exhibits this behavior (as already pointed out in Chapter 2). Therefore it can be used as TCF compensating material in resonators [23]. This effect is also notable in SMRs, where the low impedance acoustic layers are usually made from SiO2. This is the reason behind generally lower TCF in SMRs as compared to FBARs.

3.4

Summary On the previous pages we have outlined the most important topics regarding modeling of thin film BAW resonators. The basic equations governing the simple prototype resonators were presented and the origin of the lateral effects, most notably spurious modes, were described. Theoretical solutions to these problems were presented. In later chapters we will see practical implementations of the described solutions: How the effective coupling coefficient is optimized with the use of high-impedance electrodes and also in the case of SMR use of suitable materials for the mirror layers. It will also be shown how Q-values are optimized for SMR with the use of a nonquarter-wavelength mirror and how the spurious modes are eliminated with the application of a border region in both FBAR and SMR devices. These solutions have a very concrete practical application in BAW filter production. It is safe to say that the role of BAW in radio frequency filters would not be what it is today without these enhancements.

References [1] [2] [3] [4]

Auld, B. A., Acoustic Waves and Fields in Solids, Vol. I & II, New York: Wiley, 1973. Ristic, V. M., Principles of Acoustic Devices, New York: Wiley, 1983. Rosenbaum, J. F., Bulk Acoustic Wave Theory and Devices, Norwood, MA: Artech House, 1988. Larson, J. D. III, et al., “Modified Butterworth-Van Dyke Circuit for FBAR Resonators and Automated Measurement System,” Proceedings of IEEE Ultrasonics Symp. 2000, San Juan, Puerto Rico, pp. 863–868.

90

BAW Device Basics [5] Dieulesaint, E., and D. Royer, Elastic Waves in Solids, Vols. I & II, New York: Springer-Verlag, 1999. [6] Mason, W. P., Piezoelectric Crystals and Their Application to Ultrasonics, Princeton, NJ: Van Nostrand, 1950. [7] Berlincourt, D. A., D. R. Curran, and H. Jaffe, “Piezolelectric and Piezomagnetic Materials and Their Function in Transducers,” Physical Acoustics, Vol. I-A, ed. W. P. Mason, Academic Press, New York, 1964. [8] Chang, S. H., N. N. Rogacheva, and C. C. Chou, “Analysis of Methods for Determining Electromechanical Coupling Coefficients of Piezoelectric Elements,” IEEE Trans. on Ultrasonics, Ferroelectrics and Frequency Control, Vol. 42, No. 4, July 1995, pp. 630–640. [9] Lowe, M. J. S., “Matrix Techniques for Modeling Ultrasonic Waves in Multilayered Media,” IEEE Trans. on Ultrasonics, Ferroelectrics and Frequency Control, Vol. 42, No. 4, July 1995, pp. 525–541. [10] Adler, E. L., “Matrix Methods Applied to Acoustic Waves in Multilayers,” IEEE Trans. on Ultrasonics, Ferroelectrics and Frequency Control, Vol. 37, No. 6, November 1990, pp. 485–490. [11] Shockley, W., D. R. Curran, and D. A. Koneval, “Energy Trapping and Related Studies of Multiple Electrode Filter Crystals,” IEEE Frequency Control Symp. Proceedings, 1963, pp. 88–126. [12] Shockley, W., D. R. Curran, and D. A. Koneval, “Trapped-Energy Modes in Quartz Filter Crystals,” Journal of the Acoustical Society of America,, No. 41, 1967, pp. 981–993. [13] Milsom, R. F., et al., “Comparison of Mode-Conversion, Energy-Trapping and Lateral Acoustic Coupling in FBAR and SBAR,” 2005 IEEE MTT-S International Microwave Symp. Digest, June 2005, pp. 229–232. [14] Lakin, K. M., et al., “Improved Bulk Wave Resonator Coupling Coefficient for Wide Bandwidth Filters,” Proceedings of IEEE Ultrasonics Symp. 2001, Atlanta, GA, 2001, pp. 827–831. [15] Thalhammer, R., et al., “Ohmic Effects in BAW—Resonators,” Proceedings of MTT-S 2006, San Francisco, CA, pp. 390–393. [16] Marksteiner, S., et al., “Optimization of Acoustic Mirrors for Solidly Mounted BAW Resonators,” Proceedings of IEEE Ultrasonics Symp. 2005, Rotterdam, Netherlands, pp. 329–332. [17] Fattinger, G. G., “Acoustic Wave Phenomena in Multilayered Thin Film Layer Stacks,” Ph.D. Thesis, Johannes Kepler Universität, Linz, 2004. [18] Thalhammer, R., et al., “Spurious Mode Suppression in BAW Resonators,” Proceedings of IEEE Ultrasonics Symp. 2006, Vancouver, Canada, 2006, pp. 456–459. [19] Ruby, R., “Review and Comparison of Bulk Acoustic Wave FBAR, SMR Technology” Proceedings of IEEE Ultrasonics Symp. 2007, New York, 2007, pp. 1029–1040. [20] Kaitila, J., et al., “Spurious Resonance Free Bulk Acoustic Wave Resonators,” Proceedings of IEEE Ultrasonics Symp. 2003, Honolulu, HI, 2003, pp. 84–87. [21] Ruby, R., et al., “Thin Film Bulk Wave Acoustic Resonators (FBAR) for Wireless Applications,” Proceedings of IEEE Ultrasonics Symp. 2001, Atlanta, GA, 2001, pp. 813–821. [22] Kokkonen, K., and T. Pensala, “Laser Interferometric Measurements and Simulations of Waves Transmitted Through the Mirror in Thin Film BAW Resonator,” Proceedings of IEEE Ultrasonics Symp. 2006, Vancouver, Canada, pp. 460–463. [23] Lakin, K. M., K. T. McCarron, and J. F. McDonald, “Temperature Compensated Bulk Acoustic Thin Film Resonators,” Proceedings of IEEE Ultrasonics Symp. 2000, San Juan, Puerto Rico, pp. 855–858.

CHAPTER 4

Design and Fabrication of BAW Devices Robert Aigner and Lueder Elbrecht

4.1

Design Considerations for BAW Devices The design methods for impedance-element-based filters have a long tradition. The theory is covered in Chapter 2. The most significant performance parameters of the resonators constituting filters are described in the section below. The degrees of freedom in designing a BAW filter are significantly less than in SAW because frequency is determined by the layer stack rather than by lithography. Typically only the two discrete frequencies of series and shunt resonators are available in a BAW process, leaving impedance of each individual resonator as the main design instrument. Practical filter design is normally accomplished using behavior-based compact models of resonators which have a size scaling for performance parameters built in. The designer’s main choice is the filter topology and the number of filter stages. A fast circuit simulator with properly set up goal functions will quickly converge to a very satisfactory result for most filters. BAW filter design becomes very simple once a consistent resonator performance has been established. In this section, we will first discuss some fundamental design considerations for BAW devices. 4.1.1

Electromechanical Coupling Coefficient

2 Electromechanical coupling coefficient k eff (as defined in Chapter 3) is a parameter of exceptional importance for the design of BAW components. The width of the fil2 ter passband required for a certain product defines a lower limit for k eff . Insufficient coupling will force the filter designer to use excessive inductance in the ground path of shunt resonators which will harm attenuation in the stopband region tremendously. Unfortunately, ladder filters are unforgiving in terms of coupling. For a PCS transmit filter which requires a bandwidth of 60 MHz at 1,880-MHz center fre2 quency the minimum acceptable value for k eff is 6.3%, but this will leave no manu2 = 6.45% or facturing margins at all. For a high yielding part the goal is to have k eff

higher.

91

92

Design and Fabrication of BAW Devices

4.1.2

Quality Factor

Another important performance parameter is the quality factor Q of the resonator (as defined in Chapters 2 and 5). As superior Q-values are the main selling point for BAW-FBAR over SAW technology, it is extremely important to keep pushing the performance envelope. A first step is to know which type of the loss defines the limit in a certain case. No general answer can be given but the usual suspects are • • • • •

Electrical resistance in electrodes and interconnects; Acoustic leakage in vertical and lateral direction; Substrate losses (parasitic conductivity); Viscous losses (propagation losses) in the layers with high stress and strain; Losses related to surface contamination at the air interfaces.

A lot of progress has been made over the past 5 years to improve the Q-values from less than 1,000 to around 2,500. The dominant loss mechanism in BAW-SMRs used to be leakage of shear wave components through the reflector, which was discovered and fixed in 2002 [1]. After introducing a reflector stack which reflects both longitudinal and shear wavers the Q-values increased from around 700 up to 1,500. A conclusive theory about the true limit of quality factors beyond this point is not available to date. Q-values up to 2,700 have been reported [2]. In order to be able to take full advantage of improved resonator Q-values it is important to reduce losses due to electrical resistance in electrodes and interconnects significantly. 4.1.3

Spurious Modes

Spurious modes refer to narrowband effects causing the impedance and phase of a resonator to deviate from a 1D Mason model or a BVD model around the main resonance. Spurious modes can harm filter performance significantly, and in particular in filters with tight amplitude ripple and group delay ripple specifications. In the case when spurious modes appear as sharp indents (or even an additional loop) in a Smith chart of a resonator most likely lateral standing waves are responsible for the effect. While those cases appear to be the most severe they are often easy to resolve by proper termination of the resonator edge regions (see Chapters 3 and 5). Spurious mode suppression does not only help the smoothness of the filter curves, it will also reduce filter losses as the optimum Q-values coincide with the point of best spurious mode suppression. In other cases the spurious modes are not so obvious, this is when they go along with significant energy losses and are smeared out over a wide-frequency range. In such cases the overall Q-value of the resonator is usually poor. Spurious modes are a dangerous trap for BAW developers. They start showing up as soon as a threshold in Q-value of around 600 has been obtained and typically force the developer to start over with layer stack design.

4.1 Design Considerations for BAW Devices

4.1.4

93

Power Handling

FBAR/BAW devices endure higher power levels better than SAWs mainly because the electrical currents distribute more evenly. There are no narrow IDT fingers like in SAW which are prone to electromigration damage. Even though the minimum feature size of BAW is much larger, the current densities can be enormous. For a BAW at 32-dB transmit power at the upper-passband skirt (worst-case scenario) the following observation has been made: Depending on the electrode materials used, the combined effect of current density and mechanical stress will cause the electrode material to migrate and form rough regions on the resonator surface. The losses of that resonator will increase and so will the temperature of the resonator. As migration effects follow an Arrhenius–type law with temperature, the damage accelerates and the resonator will self-destroy within minutes. The power handling of BAWs is a strong function of the ambient temperature as suggested by the Arrhenius law. It is very important to keep the filter chip as cold as possible; therefore it is necessary to provide a good heat sink. Using a reflector stack which includes metal layers is a significant advantage over dielectric reflector stacks or FBARs. Another key to excellent power handling is to improve the electro- and stress-migration properties of the weakest material involved. Unfortunately, aluminum (which is an excellent electrical conductor) is not as good as denser metals like molybdenum or copper. Significant material research is in progress to find an optimum solution for BAW. The power-handling characteristics of a process should be determined in terms of maximum transmitted and dissipated power per area. The designer of a duplexer needs to keep those numbers in mind and design the filter stages such that those limits are not exceeded in any resonator. If required, the power density of a certain resonator can be lowered by cascading two resonators. Two resonators which are electrically connected in series behave very similar to a single resonator of half the area of one of the two cascaded resonators. As a consequence the power density is reduced by a factor of four. Improvements in power handling can therefore be achieved on cost of chip size. 4.1.5

Temperature Coefficient of Frequency

FBARs are slightly better in terms of temperature drift than conventional SAWs, but not by much. SAWs, based on LiTaO3, have a typical temperature coefficient of frequency (TCF) of −42 ppm/K, while FBARs can achieve around −30 ppm/K on average. BAW-SMRs on the other hand utilize the inverted temperature behavior of (amorphous) SiO2 to obtain partial compensation of the temperature drift of the other materials. The Young’s modulus of SiO2 increases as temperature goes up [3]. In addition to that, the thermal expansion coefficient is very small. A BAW-SMR with a properly designed layer stack will be in a range of −16 to −19 ppm/K. In the most demanding duplexer applications, a low TCF is extremely important as it allows achieving spec-compliance over a wider range of temperatures. Moreover, a low TCF helps to avoid thermal runaway (a situation where self-heating shifts the filter down and losses at the upper-passband edge increase which in turn enhances self-heating, and so on) [4].

94

Design and Fabrication of BAW Devices

It is possible to further improve TCF in SMRs by increasing the SiO2 content and by moving the SiO2 closer to the high-stress regions in the stack. SMRs with essentially zero TCF can be achieved using the methods described in Chapter 3. All of these approaches harm k2eff massively and can only be used for filters and resonators with small fractional bandwidth. 4.1.6

Area Efficiency

At a given frequency and filter topology the size of a BAW-FBAR implementation can vary significantly depending on layer stack and applicable design rules. The most important factors determining area efficiency are: •

•

•

4.1.7

The thickness of the piezolayer at a given frequency. For each electrode material a certain electrode thickness will yield the best performance in k2eff and Q. Depending on the acoustic impedance of the electrode material the piezolayer will vary in thickness. A thinner piezolayer is normally desirable as it reduces the size of the resonators and filter chips. At very high frequencies it can be a disadvantage to reduce resonator size any further. Dead area between resonators. Depending on the design rules there will be a minimum spacing between adjacent resonators. This is defined by processing limitations for the edges of resonators both on FBAR and on BAW-SMR. In BAW-SMR with patterned reflectors it is important to be able to use minimum over-sizing of the reflector relative to the bottom electrode. It is also helpful to be able to process resonators in which bottom and top electrode have essentially the same size. Area consumed by interconnects and packaging. A significant percentage of the actual chip area is consumed by bond-pads, seal-rings, and interconnects between the filter and the output terminals. Often the percentage of active resonator area will be smaller than 50%. The method chosen for packaging/assembly can make a significant difference. While resonators shrink in area according to 1/ f 2 the chip area will not reflect this shrink (because the area wasted to fulfill package design rules does not change). Interconnect Losses and Parasitics

Along the same lines as in area efficiency, the design rules also have an impact on actual filter performance. If design rules require resonators to have a large spacing between each other, the resistance of the leads between those resonators will increase significantly. Moreover the added lead area will increase the parasitic capacity of interconnects and hence the effective coupling of the resonators will decrease. More importantly in a BAW-SMR with patterned reflector the region where the top electrode lead crosses over the extended reflector area creates a significant parasitic capacitance which lowers the k2eff of this particular resonator. The current path in the leads and electrodes of a filter can cause significant current crowding effects. It can be shown that increasing the width of a certain trace will not always have the desired effect of lowering the electrical series resistance because very little current will flow in the added metal area.

4.1 Design Considerations for BAW Devices

95

Another consideration is the current density distribution in the thickness direction. The relative angle of the current density vectors in the top electrode and bottom electrode will determine if the current density is uniform throughout each electrode layer. If currents flow in opposite direction in top and bottom electrodes the current will be pulled towards the piezolayer by magnetic interaction. In a case where a less conductive layer is used facing the piezolayer, the electrode resistance will increase significantly. For this reason the resistive effects of the electrodes can vary significantly as a function of feed-positions for top and bottom electrode. 4.1.8

Robustness

Robustness has a number of aspects relevant to both manufacturing and reliability. It can be categorized into the following areas: •

•

•

Mechanical robustness. Certain processing steps such as spin coating, clean processes, handling, testing, and sawing exert significant stress for structures on the wafer surface. BAW-SMR is inherently more robust than FBAR as there are no structures that can be mechanically damaged. No special precautions have to be taken anywhere in processing or assembly. Environmental robustness. BAW and FBAR devices need to pass the same qualification and test procedures as the semiconductor component they are connected to. Typical consumer products require passing tests in accelerated humidity and temperature conditions. In presence of humidity, the metal in the electrodes will corrode quickly, causing frequency shift and massive degradation of Q-values. Several options exist to prevent the actual device from degrading in these severe conditions. Using the classical SAW-filter package which features a hermetic cavity is one solution. Alternatively, a hermetic seal can be created on wafer-level using a WLP approach with wafer bonding [5]. Finally, for BAW-SMR, the device surface by itself can be passivated in the same way this is done for semiconductor devices. The complication for BAW-SMR is that the passivation layer is part of the acoustic stack and it can therefore not be made as thick as in IC processes. As soon as robustness with regard to humidity is established on a wafer level the creation of an air-cavity can be done in a fairly cheap approach using polymers [6]. Electrostatic discharge (ESD). Both BAW and FBAR devices can achieve excellent robustness with regard to ESD. More than +1-kV ESD HBM (human body model) robustness for a 2-GHz device can be achieved when the process is done well. AlN has a very favorable behavior with regard to electri9 cal breakthrough. High-quality films will show well above 3.0 ⋅ 10 V/m breakthrough strength. ESD damage usually occurs at the edges of a resonator, in particular at spots where AlN growth is imperfect. Any steps that exist directly before the deposition of the piezolayer will potentially create a weak spot. Most critical are the spots where a top electrode lead crosses such a spot. ESD robustness can be enhanced by proper filter design, such as using cascaded series resonators at input and output to reduce the peak voltage across each resonator.

96

Design and Fabrication of BAW Devices

4.1.9

Nonlinearities

The discovery of nonlinear behavior in BAW and FBAR was reported in 2005 [4], after the first problems with BAW-FBAR duplexers had surfaced. While classical cell phone systems had no stringent requirements on linearity this has changed in the third generation of cell phones using WCDMA. Intermodulation of transmit signals with interference signals is a major concern in duplexers for WCDMA. It has long been known that solids exhibit nonlinear stress-strain relationships at high-stress levels [7]. The binding forces of the atoms in a lattice are a strongly nonlinear function of distance. This effect is described by the third-order elastic constants of a material. In addition to that, the piezoelectric constants change as the crystal deforms. It should not come as a total surprise that the elastic constants of a material are modulated by strain generated by large voltage swings at high RF-power levels or by bias voltages. A complete theory about the nonlinear behavior of AlN-based BAW resonators has not yet been published but a few practical tricks exist on how to circumvent the problem. Once more the cascading of two double-sized resonators to replace one resonator in a filter works as it decreases the voltage swing across each resonator by a factor of two. However, this is not possible for all resonators in a filter as it would increase the size of a BAW by a factor of four. Other tricks involve proper termination of harmonics before they can create intermodulation issues. The simplest way to characterize the nonlinearities of a BAW filter is to apply a pure sine wave fo at the input and use a spectrum analyzer at the output to search for harmonics at 2fo and 3fo. A more advanced approach is to apply two tones to the input and analyze the mixing products. For modeling purposes, the characterization of nonlinear behavior should focus on resonator measurements. There are two simple measurements which are well suited to determine nonlinear parameters: 1. Reflected spectrum measurement: One single resonator is connected to an RF source using a directional coupler. The test is performed at frequencies close to the acoustic main resonance with different power levels. The incoming energy provides excitation for the resonators while the reflected energy is fed into a spectrum analyzer. From the harmonics at 2fo and 3fo the IIP2 and IIP3 of a resonator can directly be calculated. 2. Frequency shift by DC bias: The second measurement requires a conventional network analyzer and a (high-voltage) bias T. The S11 of a single resonator is measured in presence of a DC bias voltage across the resonators. The shift of resonance frequency as a function of voltage is caused by stiffening of the piezolayer under bias stress. In equivalence to temperature coefficient TCF we introduce the term “VCF” as a parameter which describes the relative frequency shift in ppm for 1V DC bias. Once measurement data has been collected for resonators of different size it is relatively simple to establish a nonlinear BVD model for a BAW resonator by making capacitance voltage dependent, and the mutual inductance current dependent using a polynomial coefficients. Those components are fairly common in device modeling for active component. Instead of small-signal linear simulation the filter must be simulated using a harmonic balance simulator.

4.2 Fabrication of BAW Devices

4.2

97

Fabrication of BAW Devices It is remarkable that BAW or FBAR used to be a topic of intense R&D, mostly in companies not being main players in surface acoustic wave filters. The schematic cross-section of an FBAR or BAW-SMR device looks very similar to thin-film capacitors or micromachined pressure sensors; things the semiconductor industry has been doing for two decades. So it is no surprise that semiconductor companies have been the first to successfully demonstrate the mass-manufacturing of BAW devices [8, 9]. There are several aspects why BAW manufacturing in a semiconductor fab can make sense: •

•

•

The tools and processes available today for semiconductor device manufacturing have reached a very high level in both process quality (e.g., film thickness uniformity and stability) as well as process cost (e.g., throughput, mean time between failure). The couse of the same equipment for other products besides BAW typically enables lower manufacturing cost due to optimized utilization and shared depreciation. A third motivation could be the option for monolithic integration of microelectronics and bulk acoustic wave devices. This topic is discussed separately in Chapter 9.

4.2.1

Material Selection

The selection of the “right materials” is one of the key elements for fabrication of high-performance BAW devices. “Right” in this case means the optimum selection with regards to electrical properties (e.g., sheet resistance), acoustic properties (e.g., acoustic impedance, material quality factor, temperature behavior, roughness) and other properties (e.g., thermal conductivity, moisture stability). Beside these fundamental material properties, manufacturability (e.g., homogeneity and repeatability of film properties) also is an important selection criterion. We will discuss the materials most commonly used in BAW manufacturing in this section. The acoustic impedance can be calculated from the product of acoustic velocity and mass density (Section 3.2). Figure 4.1 illustrates these two material properties for some typical semiconductor device materials and some other materials for comparison. It should be noted that not all of the shown materials have high acoustic quality factor, especially the “soft” materials with low acoustic velocity typically have also low Q. Metals

For electrode layers the thickness is typically much less than the skin depth at the relevant frequency. As a consequence the electrode resistance is higher than desired. BAW people are facing the dilemma that good acoustic materials are usually lousy conductors while the best conductors have excessive material damping and would create acoustic losses.

98

Design and Fabrication of BAW Devices

Figure 4.1 Acoustic velocity, density, and acoustic impedance (size of the bullets) for various materials. Materials with thick lines are most common materials in integrated circuit fabrication.

•

•

•

•

Aluminum (Al) is typically used in form of aluminum alloys as AlCu, AlSiCu with electrical and mechanical properties very close to pure Al. It has a favorable electrical conductivity (see Chapter 7). However, due to its low acoustic impedance and the subsequently lower achievable bandwidth compared to high acoustic impedance materials such as tungsten and molybdenum, Al is not commonly used as electrode material for BAW devices. Tungsten (W) in contrast to Al has a very high acoustic impedance and is therefore an excellent candidate as part of an acoustic mirror in a SMR-type BAW resonator [1, 10], and as an electrode material (see also Chapter 7). In semiconductor device manufacturing, tungsten is mainly used as via fill in between metal layers. The respective processes (typically CVD deposition) are therefore optimized for good step coverage whereas for a BAW device, optimization for thickness uniformity and residual stress is of higher importance. Copper (Cu) has approximately 1.5 times better conductivity compared to aluminum and is therefore increasingly used for interconnect layers. However, process complexity and process cost are significantly higher than for Al. It requires diffusion barrier layers and equipment dedication to prevent transistor device contamination by copper diffusion into the silicon. The acoustic impedance of Cu is very similar and acoustic Q-value is even worse than aluminum, which makes it also an unfavorable material for BAW electrodes. Titanium (Ti) and Titanium nitride (TiN) are common diffusion barriers and liners for CVD-W layers in IC manufacturing. As Ti has higher resistivity than Al but similar acoustic impedance, it is typically not used in a BAW device for other purposes than these two.

4.2 Fabrication of BAW Devices

•

•

•

99

Tantalum (Ta) is used pure or in form as nitride, silicide, or oxide in semiconductor device manufacturing. Typically, these layers are used as thin diffusion barrier layers (e.g,. for Cu interconnects). Tantalumpentoxide is a high-k dielectric and has been used as part of an acoustic mirror in a SMR-type BAW [11]. Molybdenum (Mo) has an electrical conductivity similar to tungsten, but a slightly lower acoustic impedance. It is therefore also commonly used as electrode material for BAW resonators. It is a reasonable compromise for FBAR 2 but for BAW-SMR it will increases resonator size and will lower the k eff as compared to a tungsten electrode, in particular when used as a bottom electrode. Potential alternatives for electrode materials have been proposed which include iridium [12] and ruthenium [13]. Those materials may be difficult to introduce but they could allow improving electrode resistance somewhat. Little data is available about research on metal alloys showing improved resistance and acoustical parameters but this is a field of great potential.

Semiconductors and Dielectrics •

•

•

•

Silicon (Si) as the most important material of integrated circuits is typically not an essential part of BAW devices. It can be used for micromachining thin membranes either by bulk or surface micromachining [14, 15]; otherwise, it serves just as a carrier substrate for SMR-type BAW devices. In the latter case, high-resistive (undoped) silicon should be used in order to minimize electrical losses. Compared to other substrates it has some proven advantages: Beside cost and availability aspects, silicon is mechanically very robust which eases handling during processing and assembly. Singulation of silicon wafers to individual dies also is a process available at every semiconductor device manufacturer. Silicon oxide (SiO2) actually has multiple interesting properties for BAW devices: It has low acoustic impedance, which makes it a good candidate to be part of an acoustic mirror in SMR-type BAW devices. Also, its temperature behavior is inverted to most other layers [3]: With increasing temperature, SiO2 exhibits an increase of Young’s modulus together with a rather small thermal expansion, which leads to an overall decrease in acoustic delay. It can therefore be used to reduce (or even fully compensate) the temperature dependence of BAW devices. Silicon nitride (SiN) has a similar mass density but higher acoustic velocity than SiO2. However, the difference in acoustic impedance is not large enough to build good acoustic reflectors from a SiN/SiO2 multilayer stack. For membrane-type resonators, SiN layers are sometimes used as supporting membrane [16]. Additionally, like in semiconductor manufacturing, SiN can be used as a passivation layer. Another promising material under evaluation is silicon-oxy-carbide (SiOC, also called silicon carb-oxide). SiOC is a popular material in CMOS where it is used as “low-k” intermetal dielectric to reduce capacitive parasitics. SiOC

100

Design and Fabrication of BAW Devices

•

can be deposited by fairly conventional PE-CVD tools. The tricky thing about SiOC is that its material parameters can vary significantly and are more difficult to control than for SiO2. Acoustic losses in SiOC seem to be significantly higher than in SiO2. A concern is also that SiOC is mechanically less robust and it can introduce roughness which could harm the quality of subsequent layers. Another group of materials with very low acoustic impedance are polymers. Polymers with sufficient temperature stability exist and are under evaluation as acoustic layers. Typically, these polymers are spin coated, baked, and cured at elevated temperatures in order to cross-link all active chains. Polymers exhibit strong material damping and are not well suited to be placed into regions with high mechanical energy density but they can be a good solution for reflector layers deep down in the stack. Another consideration when using polymers is that the acoustic velocity is very low, and as a consequence the layers need to be very thin. Many of the other layers in BAW exhibit large stress which means that the polymer needs to have excellent adhesion to layers below and above or else delamination will be an issue.

Piezoelectric Layers •

From the material perspective, aluminum nitride is the preferred piezo material for integration in a semiconductor fab environment. As discussed in Chapter 7, it can be deposited with high quality on W, Mo, and Al electrodes. It has been so far the only piezoelectric film material that has proven manufacturability in terms of stable and reproducible film quality in high-volume production [5, 17]. There are no contamination concerns. As a positive side effect, AlN is an excellent heat conductor, which helps to improve device reliability at high power levels.

Other Processes Besides Film Deposition •

•

The lithography requirements for BAW devices are less demanding than for most semiconductor devices. The minimum feature size is typically much larger than 1 μm; however, overlay accuracy may be critical in some cases to minimize appearance of unwanted acoustic side-modes. Therefore, the use of a step and repeat system (which costs multiple million dollars each) is preferred over contact lithography systems. Etching processes for BAW need to be developed such that overetch into critical layers is minimized. Regardless if wet or dry etching is used it is important to choose etch processes which are either highly selective or allow use of end-point detection systems in modern dry-etch tools. Very often additional layers have to be introduced to achieve a robust etch stop. It is also important to be cautious with chemical cleaning processes which are very common after any kind of dry-etch to remove residues created at the sidewalls of the etch profiles and photoresist. In many cases the cleaning procedures will show more metal removal than can be tolerated.

4.2 Fabrication of BAW Devices

•

4.2.2

101

As discussed in Chapter 7, chemical mechanical polishing (CMP) to improve surface smoothness and planarity can have positive impact on BAW device performance. CMP processes for polishing tungsten and SiO2 are extensively used in microelectronics fabrication to minimize surface topology. When using these CMP processes for BAW fabrication, particular attention should be given to the prevention of dishing effects in the active resonator stack, as any nonuniformity over the resonator area would degrade the quality of acoustic resonance. Fabrication of SMR Resonators and Filters

In this section we will briefly describe a typical process flow for SMR-type resonators and filters. Starting with a high-resistive substrate, first the mirror is fabricated by deposition of alternating low and high acoustic impedance layers. In the case that metallic layers are used in the acoustic mirror, those have to be patterned in order to minimize parasitic coupling in between adjacent resonators. The construction of such a mirror can either be mesa type [as shown in Figure 4.2(a)] or—for example, by using CMP processes—fully planarized [see Figure 4.2(b)]. Following to the fabrication of the acoustic mirror, the two most critical layers in BAW manufacturing are deposited and patterned: the bottom electrode and the piezolayer. The challenges regarding the bottom electrode are manifold: First, the film microstructure (e.g., roughness) may have significant impact on the quality of the piezolayer deposited on it (see also Chapter 7). Second, the topology before the deposition of the piezolayer should be minimized as piezolayer growth can be disturbed in a way that a fold develops [as indicated in Figure 4.3(a)], which can negatively affect the device performance (e.g., ESD robustness, as discussed in Section 4.1.8). The challenges of depositing high-quality piezolayers are discussed in Chapter 7. In the case that an electrical connection of bottom and top electrode is needed (which is the case for a single resonator, but not mandatory for all filter designs), the piezolayer has to be patterned as well. Subsequently, the top electrode can be deposited and patterned. Figure 4.4 shows the cross-section of a single resonator after completion of these process steps. The figures do not include the “detuning layer,” which is needed in a filter device to put the shunt resonators to a lower frequency than the series resonators. This layer is typically placed either below the bottom electrode or on top of the top electrode. Also not included is the patterning of border rings at the edge of the reso-

(a)

(b)

Figure 4.2 Cross-section of an acoustic mirror having two metallic layers (e.g., W) with high acoustic impedance and three dielectric layers (e.g., SiO2) having low acoustic impedance. (a) Mesa configuration, and (b) planarized configuration.

102

Design and Fabrication of BAW Devices

(a)

(b)

Figure 4.3 Cross-section of a SMR-type resonator after processing of the bottom electrode and deposition of the piezolayer. (a) Mesa configuration without bottom electrode planarization, and (b) planarized configuration.

(a)

(b)

Figure 4.4 Cross-section of a SMR-type resonator after processing of the top electrode. (a) Mesa configuration without bottom electrode planarization, and (b) planarized configuration.

nators in order to suppress unwanted acoustic modes (see Section 3.3.3). Other layers that may be used are passivation layers on top of the upper electrode as well as additional metallization layers for pads or interconnects. 4.2.3

Fabrication Tolerances and Trimming

One of the major yield loss contributors in BAW manufacturing is frequency position of the devices. As an example, for a PCS RX band filter at 1.96 GHz, the frequency position must be met by better than 0.1% (see Figure 4.5). Already 20 MHz below the lowest passband frequency, the filter has to provide high attenuation in the associated TX band. Adding ∼3 MHz of margin to all specifications to accommodate for temperature drift, a filter with a typical roll-off steepness of ∼10 MHz between passband (e.g., 3 dB) and stopband (e.g., 50 dB) has only ± 2MHz of space left for fabrication tolerance. The frequency position of BAW filter is determined by thicknesses of all “acoustically active” layers (piezolayer, electrodes). Assuming constant acoustic velocity and density, the thickness accuracy and uniformity requirements of these layers also have to be in the same order of 0.1%. This is actually two orders of magnitude more demanding of what is typically needed in semiconductor device manufacturing: Integrated circuit processes typically allow for ±10% maximum variation in film thickness for metal or dielectric layers with a typical 1σ variation of 2% (this corresponds to a process capability index cpk of 1.7).

4.2 Fabrication of BAW Devices

103

Transmission

Filter roll-off Temperature margin stopband Temperature margin passband

Manufacturing tolerance

Frequency

Figure 4.5

Illustration of margins for a PCS band RX filter.

As illustrated by Figure 4.6, this would mean that just by using deposition processes with film thickness accuracy as typically used for semiconductor device fabrication, the BAW device yield would be 5% best case, typically much lower. Even if the deposition processes are improved to a level of 1σ = 0.5%, the maximum yield would be limited to less than 20% best case. So in order to achieve a reasonable BAW yield, there is no way around some kind of trimming, which compensates film thickness variations (and acoustic velocity or density variations) by either local etching or deposition. Figure 4.7 shows an example for a local frequency distribution over a single wafer and Figure 4.8 shows a typical histogram for the frequency distribution. Some of the characteristics shown in this plot are typical for a layer stack deposited in a semiconductor fab: •

The frequency pattern has significant rotation symmetric components as the majority of deposition equipment has rotation symmetric reaction chamber geometry.

1850

1900

1950

2000

2050

2100

2150

Figure 4.6 Frequency distribution for a 2-GHz BAW product assuming a 1 = 2% film thickness accuracy. The two vertical lines in the center of the figure indicate the 4 MHz target window.

104

Design and Fabrication of BAW Devices

Figure 4.7 Example plot of frequency variation on a single wafer after deposition of all BAW layers. The small dots indicate the sample positions.

0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 1800

1850

1900

1950

f S /MHz

Figure 4.8

Histogram of frequency on a single untrimmed BAW wafer.

•

Highest gradients in frequency are generated at the wafer edge. The layer thickness of all layers typically gets very thin due to geometrical conditions during deposition (e.g., clamp-rings or field discontinuities at the wafer edge).

•

Compared to the die size of ∼1 mm, the frequency variations are rather long-range effects. It is not necessary to trim with single-die-accuracy (trimming of each individual die would be very expensive).

Ion milling has a long tradition in trimming quartz crystals and enables achieving a tight frequency tolerance. In this process argon is ionized and accelerated in an electric field to an energy of 400 to 1,500 eV. The Ar+ ions hit the surface and knock out material from it, an effect that resembles sandblasting, but on the atomic level. The ion milling process requires components to be in a vacuum chamber. Pure physical etching with ion bombardment is applicable to all kinds of thin films, even mate-

4.2 Fabrication of BAW Devices

105

rials which are hard to etch with dry chemistry (Cu) or chemically inert (Pt, Au). In principle, such a process would work for singulated BAW devices, but it would take an excessive amount of time to handle and trim the multiple thousands of chips that fit individually on a wafer. This approach disqualifies ion milling from consideration as a tool for volume production. A process suitable to compensate frequency nonuniformities, as described above, is local etching by a scanned ion beam [18, 19]. The basic principle of this trimming method is illustrated in Figures 4.9–4.12. A narrow ion beam with a Gaussian intensity profile is scanned along the surface with controlled speed and will remove the desired amount of material at each location. In order to utilize this concept for semiconductor processing the tool must be fully automated with a robotic handling system and a high-performance x-y scanning table carrying a cooled wafer chuck. The ion beam source must include a neutralizer to avoid charge accumulation at the wafer surface. The full width at half maximum (FWHM) of the Gaussian beam is in a range of 10 to 15 mm. The scan pattern is usually a meander with a line-to-line distance of 2 to 6 mm. The ion beam diameter is small enough to correct the thickness gradients which typically occur on a length scale of a few millimeters, but it is much larger then one BAW device. The trimming process works “region by region” rather then “device by device” (as in laser trimming), which is an important advantage for throughput. The local removal rate is controlled by the time the ion beam stays at certain positions on the wafer. The removal profile for each wafer is then described by a velocity profile which is calculated specifically for each wafer for a known etch profile of the ion beam. In a relatively uniform layer with a smooth profile exhibiting small local gradients, the error map and the residence time map look virtually identical. The only limitation for such a case is the maximum velocity at which the scanning system can travel because it determines the minimum removal. The situation changes dramatically for error maps with strong gradients. The deconvolution roughens the velocity map severely. The accelerations required to accommodate the resulting velocity profile can be brutal, in fact a narrow beam is useless if the acceleration limits the ability to make steep gradients. For a given etch rate and beam size the maximum

Ion beam

Scanning path

Figure 4.9

Principle of local trimming by a scanned ion beam.

106

Design and Fabrication of BAW Devices

Figure 4.10 Test profile for localized processing. Minimum removal is 5 nm, maximum removal is 45 nm. Minima and Maxima occur at 20-mm distance from each other. This profile covers beam size, acceleration of scanning system, maximum velocity of scanning system, and errors in beam center position.

scan velocity and acceleration determines whether a certain thickness error can be corrected at all. The velocity limitation can be circumvented by lowering the etch rate or by adding more offset to the deposited thickness and then remove more material. In both cases the throughput of the system will plunge. The other thing to keep in mind is that the method relies on a stable etch rate and beam shape. Any error in etch rate or beam shape will compromise the results, even more so if more material has to be removed. Depending on accuracy requirements and throughput considerations one layer (typically the uppermost passivation layer) or multiple layers can be trimmed. As mentioned before, ion beam etching is able to affect virtually every material. However, some prominent materials used in integrated circuit production (such as aluminum) are more difficult to deal with than others. Like most metals (with the exception of the noble metals), aluminum has a tendency to grow a native oxide layer when exposed to air, the thickness of which depends on exposure time, humidity, and temperature. Unfortunately the ion beam etch rates of Al2O3 are a factor five times lower than those of pure aluminum. As a consequence, when etching Al and AlCu alloy thin films the removal is initially slow but speeds up tremendously when the native oxide is gone. This problem makes it very difficult to obtain accurate removal. In contrast to aluminum, the thickness correction in tungsten and molybdenum works very well despite the existence of native oxide on those metals. This is because the removal rates of the native oxides match those of the bulk metal very well. Dielectric layers do not exhibit this problem at all. It is therefore beneficial to

4.2 Fabrication of BAW Devices

107

Figure 4.11 Deviation of measured removal from planned removal (clusters of dots), lines indicate ±4-nm error limits.

Figure 4.12

Histogram of thickness error after processing for the profile shown in Figure 4.10.

have a thin SiN (silicon nitride SixNy) layer on top of the final resonator just for the purpose of accurate trimming. The sensitivity of a SiN layer at the top of the structure is small. On the positive side this enables very accurate trimming, but on the negative side the trimming range is very limited.

108

Design and Fabrication of BAW Devices

Ion beam etching is compatible with conventional photo-resist on wafers used in acoustic wave device manufacture. This makes it possible to do lithography-defined trimming of certain groups of BAW resonators, which is very important for bandwidth adjustments in these devices. For the generation of the removal profiles, map data of frequency and film thickness has to be available for each wafer. Even though the general concept of feed-forward control of manufacturing processes (“run-to-run control”) is actually implemented in many semiconductor fabs, the amount of data collected during BAW manufacturing to enable accurate trimming is typically much higher. For cost-effective operation, fully automated data collection and trimming for each single wafer is required. 4.2.4

Process Controls

One apparent difference between BAW and semiconductor device processing is the difference in yield loss mechanisms. Whereas in integrated circuit manufacturing constancy in lateral dimensions and defect density are most relevant yield contributors, BAW yield is mainly defined by vertical dimensions (film thickness), and properties related to the acoustic resonance (coupling coefficient of piezolayer, acoustic quality factors). The methods of film thickness characterization are reviewed in Chapter 8. The selection of the right test method may in some cases be different for semiconductor and BAW devices due to the different nature of relevant material parameters (e.g., “electrical thickness” for integrated circuits versus “acoustic thickness” for BAW resonators). While thickness uniformity, as it concerns frequency tolerance is somewhat overemphasized, the severity of the other aspects is often underestimated. At this point it is worth mentioning that many of the methods used to measure and map thickness are indirect and need very careful calibration. For transparent layers the standard tool is an optical spectrometer or ellipsometer. It measures thickness very accurately if refractive index and the optical properties of the layers throughout the stack are well known. However a change in “optical length” is not necessarily related to acoustical length. Other parameters like resulting frequency position of shunt and series resonator, piezoelectric coupling, as well as quality factor of the acoustic resonances are ideally monitored with respective single resonator test structures (see Section 8.5). This data can be handled similar to the process control monitor (PCM) test structures in semiconductor manufacturing.

4.3

Application Space for BAW-FBAR Technology 4.3.1

RF Filters and Duplexers

Wireless broadband communication has gained tremendous popularity. However, allocated frequency spectrum is limited, and the most favorable frequencies are occupied by cell phone bands, by governmental agencies, or by unlicensed bands with restricted transmission range. Whenever new applications are conceived they

4.3 Application Space for BAW-FBAR Technology

109

are forced to deploy relying on less favorable frequency bands that have noisy or “oversensitive” neighbors—sometimes both. The commercial viability of these new bands depends at least in part on the equipment makers’ ability to solve neighborhood problems. RF filters play a key role in minimizing interference between systems operating in different bands. The selectivity of the RF filter determines how big a portion of the total bandwidth will be used—“wasted” in a real sense—for guard bands. The selectivity—respectively, the steepness of the filter skirts—is closely related to inherent losses in the reactance elements of an RF filter. Practical RF filters also show a shift of center frequency as a function of temperature, which complicates the design process. RF filters traditionally used for cell phone applications based on SAW technology. The selectivity of SAW filters is good for a band at 1 GHz but degrades when the band is located closer to the upper limit of 2.5 GHz. Temperature drift is also a concern. Before the advent of BAW-FBAR the only available solutions for broadband communication systems above 2 GHz were dielectric filters, waveguide filters, and LC filters. Dielectric and waveguide filters are based on electromagnetic waves and/or integrated inductor and capacitor combinations. In “pure” electrical LC filters the major losses are related to ohmic resistance, skin effect, and eddy currents in the metal conductors. Another issue related to overall usefulness is the quality of inductors, which is generally below 50 in the frequency range above 2 GHz, making it impossible to provide the required selectivity. Filters based on wave phenomena show significantly less losses than filters based on lumped LC elements. "The Rediscovery of Slowness”

In a fast-paced industry, it is counterintuitive to settle for something less than maximum speed. However, for RF filters based on wave phenomena it is a huge advantage to use waves with a slow velocity. Wavelength is proportional to velocity and inversely proportional to frequency. The only alternative to filters employing electromagnetic waves are those using acoustic waves in solid materials—the foundation of acoustic filter design.

Figure 4.13

Illustration of RF filter performance characteristics.

110

Design and Fabrication of BAW Devices Table 4.1

Comparison of Wave Velocity Wave Velocity

Wavelength at 2 GHz

Electromagnetic wave in air

300 million m/s

150 mm

Electromagnetic wave in dielectric material εrel = 100

30 million m/s

15 mm

Acoustic wave in solid material 4,000 m/s to 12,000 m/s 2 μm to 6 μm

The volume of a material comprising the physical structure of a SAW or BAW filter required to confine an acoustic wave is approximately a factor of 105 smaller than for an electromagnetic wave (Table 4.1). Acoustic waves store and carry energy very efficiently and with extremely low losses. As a result acoustic filters can be tiny in size and exhibit very low losses. Surface Acoustic Wave and Bulk Acoustic Wave Filters

SAW filters are the dominant technology for RF front-end filters and duplexers in almost all wireless phone frequencies. In terms of insertion loss, skirt steepness and relative bandwidth SAW filters exactly match the requirements of traditional cell phone systems. This is not a coincidence; it is the consequence of defining the cell phone standards such that commercially viable filter technologies would be available to fulfill anticipated requirements. It is highly desirable to utilize the available frequency spectrum more efficiently. In order to increase useable bandwidth the part of the spectrum wasted by guard bands has to be minimized (see Figure 4.13). The minimum width of a guard band is determined by the steepness of the filter’s skirts. SAW technology serves classic cell phone applications (all four GSM bands and all CDMA bands except the US-PCS band) very well. The manufactured volume of SAW filters exceeded 5 billion units in 2007. SAW technology is very mature and every aspect of the manufacturing process is optimized to achieve aggressive cost targets. The US-PCS band with its narrow transition range of 20 MHz between transmit (Tx) and receive (Rx) bands provides challenges which are difficult to overcome with conventional SAW technology. The two flavors of BAW (BAW-SMR and FBAR) have successfully filled this void in recent years, granting them a place in the wireless phone market. Deciding which filter technology is right for a certain application is usually a balancing act between performance, size, and cost. In terms of performance, there are several disciplines in which technologies compete: •

•

•

•

•

Maximum achievable filter bandwidth as a percentage of the center frequency (relative bandwidth). Insertion loss in the passband (in particular at the edges of the passband) and steepness of the filter skirts. Temperature dependency of the filter characteristics: temperature coefficient of frequency (TCF). Flexibility in port impedance and port configuration, for example, single-ended input, differential output. Power-handling capability and ESD robustness.

4.3 Application Space for BAW-FBAR Technology

111

In the categories of relative bandwidth and flexibility to accommodate different port configurations, SAW technology is clearly the winner at frequencies up to 2 GHz. Different bandwidth requirements are accommodated in SAW designs by choosing a suitable piezomaterial with a certain crystal-cut angle in the raw wafers. Choices range from very low-bandwidth materials, for example quartz and langasite, to medium-bandwidth materials such as lithium tantalate. A typical high-bandwidth material is lithium niobate. As a general rule, the higher bandwidth materials show larger temperature dependency and higher losses. SAWs have an inherent advantage when it comes to impedance conversion and arbitrary port configuration because these are determined by the transducer mask layout and do not require more complex processing. SAWs also have the ability to include a “balun” function, which can be used to create a differential output signal from a single-ended filter input: a widely used and practical advantage. For SAW it is possible to integrate filters and duplexers for different bands on one chip with little or no additional processing effort. SAW technology approaches practical limits at 2.5 GHz because the requirements for line width and gap dimensions in the transducers call for less than 0.25-micrometer lithography resolution. Manufacturing such a structure requires efforts and investments that are commercially difficult to justify. SAWs with a relative bandwidth of larger than 0.5% show a significant temperature dependency. For example: the most widely used SAW substrate material is lithium tantalate, which will exhibit a TCF on the order of −45 ppm/°C. The resulting frequency shift at −30°C and at +85°C must be accounted for by adding temperature margins to the filter characteristics. Filters in the transmit path are challenged by power handling requirements. The current densities in the tiny metal fingers are significant and coincide with mechanical stress. This gives rise to metal migration effects in the fingers which will destroy the device over time. A carefully designed SAW filter for 2 GHz will have a mean time to failure (MTF) of >10,000 hours for continuous 1W (30 dBm) input power at +55°C ambient temperature. Higher levels of power or higher operating temperatures are very difficult to accommodate. The BAW principle has inherent advantages with regard to losses. Acoustic energy density is very high in BAW designs and the waves are very well trapped. The quality factors (Q-value) that can be achieved with BAW resonators are superior to any other technology suitable for the GHz range. Q-values of 2,000 at 2 GHz represent the state-of-the-art for FBARs and SMR-BAWs. As a result of the high Q-values the filter skirts will be very steep while the insertion loss remains low even at the edges of the passband. This is a key advantage for duplexers in the US-PCS band and the main reason FBAR and BAW were able to conquer a large market share in this particular application. There are no tiny electrode fingers in a BAW resonator and therefore the limit for power handling is defined by exceeding a temperature limit rather than by electromigration effects. The long-term power durability can be pushed up to 4W (36 dBm) at 2 GHz with moderate effort. With regard to ESD robustness a BAW device is by far superior. BAW-SMRs also have significantly less temperature dependency and exhibit particularly favorable TCF compared to SAW, typically −20 ppm/°C. All this having been established and proven in the marketplace, the most important advantage of BAW-SMR is the fact that frequencies up to 6 GHz can be addressed without running into practical manufacturing limits.

112

Design and Fabrication of BAW Devices

The thickness of the layers to be deposited scale with 1/f while the size of a BAW resonator scales with 1/f 2. Both parameters make it favorable to use BAW at high frequencies, but conversely, make it hard to compete at low frequencies with SAW products. BAW-SMR and FBAR require a complex manufacturing process with a factor of 10 more processing steps than SAW. Even though material costs for both filter types are about the same, and even though BAW-SMR/FBAR are manufactured on larger wafer sizes (SAW on 100 mm, BAW on 150 mm or 200 mm), the inherent cost per filter is much higher than for a SAW. As of today the only thin film piezomaterial with proven manufacturability is aluminum nitride (AlN). The piezoelectric effect in AlN is relatively weak and as a consequence the relative bandwidth of FBAR and BAW-SMR is limited to about 4%. This is just enough to handle most of the cell phone applications well, but it is too little for certain broadband wireless applications such as WLAN or WiMAX where the passband can have up to 15% relative bandwidth. The other significant limitation of the current generation of FBAR and BAW-SMR is the lack of flexibility to transform impedance or to provide a built in balun function. Another disadvantage is that, while in theory it is possible to build monolithic BAW devices (which would cover more than one frequency band on a single chip), such an implementation is prohibited by practical and commercial reasons. While the active area of a SAW filter is slightly larger than a BAW between 1 and 2 GHz, a SAW solution for multiband applications typically recovers the lost space by using monolithic integration. Both SAW and BAW have specific strengths and weaknesses. For the most part they complement each other. The number of applications in which they compete against one another is very limited. It appears that any controversy regarding whether SAW or BAW will dominate the filter market has ceased since major SAW players have acquired BAW capabilities. It is relatively simple to map out the application space for SAW and BAW for near-term opportunities (Figure 4.14): BAW will expand the ability to serve high frequency and power applications through its ability to satisfy the requirements of high-performance filters. In summary it can be stated that BAW or FBAR is well positioned to dominate high-performance applications in the frequency range above 2 GHz, in particular applications which are less cost sensitive. Due to inherent cost advantages SAW will unlikely lose market share in applications they currently serve. 4.3.2

Oscillators

An application in which BAW or FBAR has a potential to play a role in the near future is in low-phase-noise oscillators and related clock and timing circuitry for which frequencies above 2 GHz are desired. The key advantage is that high Q-values can be maintained up to 5 GHz and higher. This will eliminate the frequency doublers and phase-locked-loops (PLL) used in many systems requiring frequency references at >2 GHz. Remarkable temperature stability can be obtained by introducing SiO2 layers between one of the electrodes and the piezolayer (as described in Section 2.4). This method reduces the relative frequency spacing between resonance and antiresonance which is acceptable for oscillators. BAW or FBAR resonators can be electrically tuned to a specified frequency using a varactor diode in the same way

4.3 Application Space for BAW-FBAR Technology

Figure 4.14 sated SAW.

113

Application space for RF filters. TC-SAW is an abbreviation for temperature compen-

quartz crystals are “pulled.” The tuning range will typically be less than 1% in frequency or else the oscillator noise performance will degrade massively. An alternative method of tuning can be accomplished by the inherent second-order nonlinearity of the piezo material. In presence of a DC-bias voltage the resonance frequency will change with a coefficient of around 20 ppm/V in a typical 2-GHz resonator with 1.3-μm-thick piezolayer [4]. The tuning effect achievable with this method is small but as it avoids using varactors there is no degradation of phase-noise performance. 4.3.3

Sensors

Chemical/Biological Sensors

BAW or FBAR devices have been studied as mass-sensing elements in applications ranging from chemical sensors to biological/DNA sensors [20]. The devices will have to be coated with a layer that allows the substance to be detected to dock onto the surface. In theory an increase in mass loading on the top surface of a BAW or FBAR will decrease the resonance frequency of the device. The frequency shift (Hz) observed for a certain mass load [kg/m2] increases proportional to the resonance frequency squared. Using a 2-GHz BAW device instead of the classical 10-MHz quartz crystal will increase signal strength tremendously and improve signal-to-noise ratio. In practical applications there are several obstacles which make it hard to gain as much sensitivity as predicted. Most importantly all chemical and biological sensors need a special coating which has a finite thickness. While this coating can be treated as a pure mass load in a 10-MHz quartz, it has to be considered as a delay line in a high-frequency BAW. Hence the coating thickness will modulate the sensitivity for mass loading. Another obstacle is that many biological sensors must be operated in

114

Design and Fabrication of BAW Devices

a liquid environment. As BAW or FBAR is normally operated in longitudinal (thickness extensional) mode significant acoustic energy will be radiated into the fluid covering the resonator. This corresponds to a tremendous drop in Q-values, for example from Q = 2,000 in air to around Q = 50 in deionized water. One approach to circumvent this issue is the attempt to build a BAW which works in a thickness shear mode [21]. The c-axis of the piezolayer is intentionally tilted by as much as 30° from the vertical to enable a significant coupling to the shear wave resonance of the layer stack. Shear waves do not propagate in fluids and thus the losses associated with acoustic radiation should be much lower. It is unclear at this point in time if BAW or FBAR will offer significant advantages over other principles of chemical and biological sensors. Deformation Sensors

Other potential sensor applications of a BAW include the sensing of deformations, for example in the membrane of a pressure sensor. The acoustic velocity of silicon or other structural materials will be slightly modulated by quasi-static stress and strain. By choosing the strained part of a structure to be in the acoustic travel path of a BAW resonator, a frequency change will occur which is proportional to the deformation in the structural material. The largest effect can be obtained if the structural material is a long delay line in an overmoded BAW resonator. Reading out frequency as a sensing signal instead of capacity or resistance for a pressure sensor can be an advantage for wireless remote sensing, for example in tire-pressure monitoring. Inertial Sensors

A more exotic application of BAW is in gyroscopic inertial sensors. The principle is based on wave conversion occurring due to Coriolis force in rotating inertial systems. Imagine a BAW vibrating in a pure thickness extensional (TE) mode. If the whole device now rotates around an axis which is parallel to the chip surface the Coriolis force will induce a polarized shear wave which will propagate through the layer stack. It is relatively simple to isolate this shear wave from the longitudinal wave by using a reflector stack which is highly reflective for the longitudinal wave, and at the same time transparent for the shear wave, and vice versa. It is thus possible to trap the longitudinal and the induced shear wave in a different parts of the layer stack. A second resonator which is optimized to detect the shear wave can be used to extract the rotation rate. To this point there is no experimental proof that sufficient sensitivity can be achieved. In comparison to MEMS gyroscopes which have tiny vibrating masses it is clear that the vibration amplitudes in a BAW is smaller by a factor of 103, but on the other side the vibration frequency can be larger by a factor 104. As a consequence, the velocity of any given mass point is potentially larger in a BAW than it is in a MEMS gyroscope, hence the Coriolis force should be at least as high as in the MEMS case. The advantage of a SMR-type BAW gyroscope over a MEMS gyroscope is that the BAW is potentially much smaller and it is practically unbreakable; it will withstand the worst mechanical shocks without damage.

4.3 Application Space for BAW-FBAR Technology

115

References [1] Marksteiner, S., et al., “Optimization of Acoustic Mirrors for Solidly Mounted BAW Resonators,” Proc. IEEE Ultrasonics Symposium, 2005, pp. 329–332. [2] Aigner, R., “Bringing BAW Technology into Volume Production: The Ten Commandments and the Seven Deadly Sins,” Proceedings of International Chiba Symposium on Acoustic Wave Devices, Chiba, Japan, March 2007. [3] National Institute of Standards (NIST) ceramic database, http://www.ceramics.nist.gov/srd/summary/emodox00.htm. [4] Aigner, R., et al., “Behavior of BAW Devices at High Power Levels,” Proceedings of IEEE IMS-MTT-S 2005, Long Beach, CA, 2005. [5] Ruby, R., et al., “Ultra-Miniature High-Q Filters and Duplexers Using FBAR Technology,” Proc. Solid-State Circuits Conference, 2001, pp. 120–121. [6] Franosch, M., et al., “Wafer-Level-Process Using Photo-Epoxy to Create Air-Cavities for Bulk-Acoustic-Wave RF-Filters,” Proceedings of IMAPS 2004 Conference, Long Beach, CA, November 2004. [7] Mason, W. P., Physical Acoustics, Vol. III, part A, New York: Academic Press, 1966, p. 196. [8] Bradley, P., et al., “A Film Bulk Acoustic Resonator (FBAR) Duplexer for USPCS Handset Applications,” Proc. IEEE MTT Symposium 2001, 2001, pp. 367–370. [9] Aigner, R., et al., “Advancement of MEMS into RF-Filter Applications,” Digest International Electron Devices Meeting IEDM, 2002, pp. 897–900. [10] Lakin, K., G. Kline, and K. McCarron, “High-Q Microwave Acoustic Resonators and Filters,” IEEE Trans. on Microwave Theory and Techniques, Vol. 41, No. 12, 1993, pp. 1517–1520. [11] Smolders, A., et al., “BAW Devices and Integration into System-in-Package (SiP),” Proc. 3rd International Symposium on Acoustic Wave Devices for Future Mobile Communication Systems, Chiba, 2007. [12] Iborra, E., et al., “Aluminum Nitride Bulk Acoustic Wave Devices with Iridium Bottom Electrodes,” Proceedings of IEEE Ultrasonics Symposium 2007, New York, October 28–31, 2007. [13] Ueda, M., et al., “High-Q Resonators using FBAR/SAW Technology and Their Applications,” Proceedings of IEEE IMS-MTT-S 2005, Long Beach, CA, 2005. [14] Madou ,M., Fundamentals of Microfabrication, Boca Raton, FL: CRC Press, 1997. [15] Ruby, R., and P. Merchant, “Micromachined Thin Film Bulk Acoustic Resonators,” Proc. IEEE Symposium on Frequency Control, 1994, pp. 135–138. [16] Dubois, M.-A., et al., “Above-IC Integration of BAW Resonators and Filters for Communication Applications,” Proc. 3rd International Symposium on Acoustic Wave Devices for Future Mobile Communication Systems, Chiba, Japan, 2007. [17] Aigner, R., et al., “Bulk-Acoustic-Wave Filters: Performance Optimization and Volume Manufacturing,” Proc. IEEE MTT-S International Microwave Symposium, 2003, pp. 2001–2004. [18] Zeuner, M., M. Nestler, and D. Roth, “Ultra-Precise Wafer Trimming Technology,” EuroAsia Semiconductor, June 2007, pp. 17–22. [19] European Patent EP1390559B1. [20] Brederlow, R., et al., “Biochemical Sensor Based on Bulk Acostic Resonators,” Electron Devices Meeting, 2003, Technical Digest, December 8–10, 2003, pp. 32.7.1–32.7.3. [21] Bjurstr, J., G. Wingqvist, and I. Katardjiev, “Synthesis of Textured Thin Piezoelectric AlN Films with a Nonzero C-Axis Mean Tilt for the Fabrication of Shear Mode Resonators,” Proceedings of IEEE Ultrasonics Symposium, 2005.

CHAPTER 5

FBAR Resonators and Filters Richard Ruby

5.1

Introduction 5.1.1

Short History of FBAR

Free-standing membrane film bulk acoustic resonator (FBAR) was first demonstrated in 1980 by Grudkowski et al., at United Technologies and independently that same year by Nakamura, et al., at Tohoku University Japan [1, 2]. Throughout the 1980s, research occurred in universities and commercial and government laboratories around the world. In 1982, a visionary paper given by Lakin et al. of TFR [3], described the future potential of BAW (and FBAR) resonators). In particular, the paper emphasized the size. Size and performance (as we learned much later), are crucial for cell phone applications and are the enabling technologies that allowed both the shrinking of the early mobile handsets, while allowing more components and functionality into the slimmer phones. Besides work at small start-ups such as TFR, Inc., major research firms such as Westinghouse [4, 5] made solid contributions to the advancement of BAW filters. Out of that effort an early and classic book on FBAR was written by Rosenbaum in 1988 [6]. The piezoelectric material, zinc oxide, was the material of choice in the early years of BAW research. One could sputter ZnO and have a reasonable chance of having a film with piezoelectric properties. Sputtered aluminum nitride (AlN) was introduced as an alternative piezoelectric by Wang and Lakin [7] in 1981. AlN, in contrast to ZnO, was not easy to sputter and more often than not, sputtered AlN films were only weakly piezoelectric. However, the attractive aspect of AlN was the ease with which it could (if made to work) be integrated into an IC facility. Zinc oxide has issues with volatility and contamination, a serious issue for any IC facility. In 1993, research was started on FBAR at Hewlett Packard Laboratories. Up until that time, there was a strong perception that ordered ZnO (or AlN) could only be accomplished by depositing it on an ordered substrate. Typical choices for electrodes were gold or aluminum. At HP, we opted to focus on molybdenum and tungsten as electrode choices, and AlN for the piezoelectric [8]. Tungsten proved to be too difficult to deposit as a low-stress film (due to tool limitations); however, very good success in depositing low-stress electrodes was obtained using sputtered Mo.

117

118

FBAR Resonators and Filters

One important property about any choice of materials for the acoustic stack is that there needs to be good etch selectivity between piezoelectric and electrode material—an important issue when patterning and etching each layer in the acoustic stack. The combination of Mo with AlN turned out to be a very good pairing of materials. Besides being able to deposit Mo as low-stress films, Mo, itself, has excellent acoustic properties including very high acoustic Q ( much higher than Al or Au). Early on, HP focused on bulk micromachining, etching silicon from the backside using tri-methyl ammonium hydroxide (TMAH) [9]. Figure 5.1(a) is a micrograph taken of the first working FBAR resonator at HP Labs (photo taken from the backside). The meandering trace forms a microheater that can be used to “tune” FBAR frequency. Although results were obtained as early as October 1993, it was clear that this was never going to be a manufacturable process. This realization led to a process based on surface machining. The first “surface machining” approach was to put down a sacrificial layer of phosphorous silica glass (PSG), etch holes (vias) into the film, and then deposit tungsten vias. The vias (also acting as support structures to hold the acoustic membrane above the surface), was accomplished by polishing the W back to the PSG surface. From here, deposition and patterning of the Mo/AlN/Mo acoustic stack with pads was relatively straight forward. The PSG had the very elegant property of etching very quickly in the presence of dilute HF, creating the necessary air/crystal interface on both sides. The first working device using this technology is shown in Figure 5.1(b). In 1995, the process migrated to forming a depression or “swimming pool” into the silicon and back filling with PSG. Then, the excess PSG on the surface was removed by chemical mechanical polishing (CMP) the PSG to the silicon surface. This has since been the preferred method of creating an air/crystal interface on the underside of the acoustic stack. Figure 5.2 shows a cut-away photo of a stacked FBAR—where two resonators were fabricated one on top of the other. This photo clearly shows the swimming pool and a free-standing membrane spanning the swim-

(a)

(b)

Figure 5.1 (a) SEM micrograph of bulk micromachined FBAR (circa 1993). The meandering lines were used to tune FBAR frequency with on-chip heaters. (b) SEM micrograph of an “air bridge” FBAR resonator (circa 1994).

5.1 Introduction

119

Figure 5.2 Membrane spanning an etched silicon pit. The first such devices were SBAR- (or SCF) type BAW devices (stacked bulk acoustic resonators).

ming pool. In the early days, hot KOH was used to etch the swimming pool, thus leaving the tell-tale 54.7o angle in the silicon. It should be noted that early research on FBAR at HP Labs was not always well supported. The reason for this had to do with the simple fact that SAW technology was in high-volume production, was relatively small size, and most important, a “one mask” process. Furthermore, performance of early FBAR devices was not anywhere as good as SAW devices made at that time. The reality was that from 1997 thru early 1999, the FBAR team struggled to exist and find justification for continuing work. Highlights (in between lowlights of having the project being nearly cancelled several times) included demonstration of a 5.2-GHz filter (Figure 5.3) just days before Christmas of 1997. These filter parts actually had a passband, skirts with a reasonable shape factor and good rejection outside the passband. It was also at this time that “apodization” was shown to conclusively improve parasitic lateral modes [10]. Apodization (the use of non-Manhattan geometries for resonator shapes) reduces the ripple in the bandpass of the filter, giving a smoother passband response than filters using Manhattan geometries for the resonator shape. Apodization is discussed in greater detail later in this chapter. Without a strong value proposition, the project was constantly in danger of being shut down. In 1998, a value proposition was articulated—a duplexer for cell phones. At first, making a duplexer seemed impossible, of all the filters in the cell phone, the specifications for the duplexer were by far the most stringent. However, by late 1998, we not only made a duplexer, but had wired one into a working phone and used it to make phone calls to upper management. 5.1.2

The Duplexer

The duplexer was for FBAR the “killer app.” The value proposition is easy to understand. FBAR technology eliminated a large ceramic duplexer in a PCS CDMA cell phone. When samples of early FBAR duplexers were shown to customers, their

120

FBAR Resonators and Filters

Figure 5.3 Micrograph superimposing two 5-GHz filters with and without apodization (vertical axes 5-dB per division; horizontal axes 50-MHz per division; span: 5.0 to 5.5 GHz).

enthusiasm was strong. The dramatic size reduction had clear value to the handset manufacturers. Early duplexers at HP consisted of a printed circuit board (PCB) the size of a stick of gum that allowed for three connections, two places to glue down die, and approximately an inch-long delay line to act as a quarter-waveline between the Rx and the Tx/antenna port. Working die from a yielding wafer (in those days, the frequency variation across the wafer and wafer-to-wafer could be easily 10% to 20%) were picked to be both the Tx and Rx filters. Figure 5.4(a) shows the first FBAR phone and some of the FBAR team calling senior managers [Figure 5.4(b)]. Code division multiple access or CDMA was chosen as the communication protocol by several service providers in the United States (the two largest being Sprint and Verizon). CDMA has the feature of being able to use a given amount of frequency spectrum very efficiently. Part of the efficiency gains come from the fact that CDMA is a full-duplex-based technology (versus half-duplex technology incorporated in GSM phones). GSM worked like a “walkie-talkie” transmitting packets of data and then switching to receive to accept incoming packets of voice data. A full-duplex technology required the receiver to be listening at all times for incoming data whilst simultaneously transmitting data. This gives twice the capacity of GSM; however, the sheer enormity of dynamic range needed is staggering. The power amplifier might be transmitting at 1W while centimeters away, a delicate low-noise amplifier was struggling to resolve signals at 25-femto-Watts. Worse yet, the FCC in the United States had assigned a 1% separation in frequency between the transmit and receive (a 20-MHz, later 15-MHz guard band for the G-block PCS band). To

5.1 Introduction

121

(a)

(b)

Figure 5.4 (a) Picture of the first FBAR duplexer and cell phone. The three-port duplexer (center) consists of a Tx, Rx, and antenna ports, where the Rx die is separated by a one-quarter-wavelength transmission line from the antenna port. A ceramic duplexer is shown in front for size comparison. (b) Members of the original WSD FBAR team in 1998; Rich Ruby, Becky Whittaker (talking on the first FBAR phone), John Larson, Yury Oshmyansky, and Randall Canha.

work within the dictates of a 136-dB dynamic range, the duplexer was a huge piece of carved ceramic giving as much as 50 dB of isolation between input and output (today, phones using FBAR duplexers have a receive sensitivity of −110 dBm, another three times better). Because the FBAR was a bulk device, it could withstand the high powers duplexers were subjected to, meet (and later exceed) the isolation specs, and most important because it was so much smaller than ceramic duplexers, FBAR was quickly adapted into cell phones. Figure 5.5(a) shows the famous Samsung Wrist Watch phone introduced at Comdex in 2000. Figure 5.5(b) shows the phone board “guts” including our first 6 × 11-mm2 duplexer. This “Dick Tracy” phone was hailed as one of the top 5 innovative products of Comdex 2000.

(a)

(b)

Figure 5.5 (a) Picture of the Samsung SPH-S100. The phone was announced at Comdex 2000 as the new “Dick Tracy” phone. (b) Picture of the phone board of the SPH-S100.

122

FBAR Resonators and Filters

With the introduction of FBAR duplexers, a new generation of smaller and sleeker cell phones proliferated the market—cell phones as “bricks” became a thing of the past. One of the first FBAR phone introduced in 2001 by Sanyo was featured in the 2002 sci-fi movie Men in Black II with Tommy Lee Jones and Will Smith. 5.1.3

The Package

Although FBAR resonators are very small and the amount of silicon needed to make the filter is measured in fractions of a square millimeter, the finished duplexer—placed into an LCC chip carrier package—had no hope to shrink much 2 further than the 6 × 11 mm footprint shown in Figure 5.5(b). Furthermore, it quickly became apparent that the manufacturers of ceramic LCC packages were going to enjoy the fruits of FBAR success by keeping their package costs high—unmindful that in commodity applications such as cell phones, average sales price can erode at about 20% per year. The situation was untenable. There was a constant fear that SAW technology would soon find a way to make their technology work at 2 GHz in their sleek LTCC packages. Anticipating this, HP Labs embarked on a program to use wafer-level packaging technology (WLP) that would be smaller and cheaper than the ceramic packages used by SAW manufacturers. Starting in early 1998, and using a manual hydraulic press as the bonder, we demonstrated a feasible version of a working microcap’d filter. This work was presented in 2002 [11]. By late 2003, Agilent (Agilent was spun-off from HP in 1999) introduced the world’s first 5 × 5 mm2 FBAR duplexer using two microcap’d FBAR die. Figure 5.6(a) shows the first duplexer with hermetic packaged die (the overmold was not added so that the microcap’d die with bond wires connecting to the PCB are shown). In many ways, the development of the microcap and wafer-level packaging was as challenging as FBAR. For too long, most folks simply could not conceive that one

(a)

(b)

Figure 5.6 (a) An early version of our 5 × 5 mm duplexer using the first microcap’d FBAR filters for the Tx (left die) and Rx (right die). (b) Picture of an Rx die with the microcap lid removed. Pieces of silicon from the lid have broken off during the removal process and remain attached to the base FBAR wafer. 2

5.1 Introduction

123

could achieve “wholesale” hermetic package of thousands of devices during one simple bonding step. Figure 5.6(b) shows an Rx duplexer die where the microcap lid has been removed. Hermiticity is achieved by using a gasket around the perimeter of the die with smaller gaskets around the perimeter of each contact. A sign of a good bond is to have pieces of silicon break off during the debonding process and stick to the gasket [as seen in Figure 5.6(b)]. In early 2004, we began ramping up our first microcap’d FBAR product and by 2006 (8 years after starting the project), we published data on the reliability of microcap [12]. By late 2008, we had made over a billion microcap’d filters. 5.1.4

FBAR in Context with the Rest of the World

As early as 1996, with publications by Ken Lakin on solidly mounted resonators [13], the possibility of switching from FBAR to solidly mounted resonator or SMR-BAW technology was a viable and tempting option. Free-standing membranes often cracked after release. Stress control was far from perfect. In what was to become a signature of FBAR process, very few deterministic experiments would give anything other than ambiguous results. In a 2-month period in 1998, membranes were cracking and breaking regardless of what deposition or etch condition we tried. Cracking came back to haunt us several times in the early years and solid investigation and luck were needed to figure out all the sources that would cause cracking. In contrast, solidly mounted resonators looked simple and dependable. Furthermore, going to a solidly mounted BAW process kept open the door for future integration with CMOS, with filters and resonators into a single IC. However, the decision was made to stay with FBAR and focus on making FBAR manufacturable. The biggest concern regarding SMR-BAW was the ability to get sufficient Q and coupling between electrical and acoustic fields (the coupling coefficient). With the acoustic energy trapped in between a crystal/air interface, we were more confident of achieving high Q and sufficient coupling between acoustic and electric fields. A big uncertainty for the FBAR team had to do with the competition; that is, SAW technology. Fujitsu had introduced the cell band SAW duplexer in the early 1990s with great success. Other SAW manufacturers were quick to copy and the 800-MHz duplexers proliferated throughout the cell phone market and enabled single-band phones working at 800 MHz. It did not seem that far fetched that SAW technology would soon find itself in the 2-GHz PCS band. With their inherent low cost, small size, and economies of scale, it was widely believed that FBAR duplexers would be quickly replaced. Indeed, as early as December 1999, rumors that Fujitsu (then the leading supplier of SAW duplexers) was going to announce the imminent release of their SAW PCS duplexer [14]. The formal Fujitsu announcement of their SAW PCS duplexer occurred in November 2002, promising high volume, mass production in January 2003 [15]. We would hear from our customers that SAW competitors would offer a 30% lower price than any price we quoted. In July 2002, Robert Aigner of Infineon publicly announced in Taiwan the results and products of a large and mostly secret effort on BAW manufacturing [16].

124

FBAR Resonators and Filters

Unlike Agilent/HP, Infineon had decided to use the SMR-BAW approach to making filters and resonators. Also, as important, Infineon chose to focus their energy on interstage filters—mostly for Nokia. In what may have been a strategic blunder by Infineon’s senior management, their BAW team focused almost exclusively on commercialization of interstage filters (where only cost, not performance, mattered) and not the more lucrative and performance-sensitive duplexer market. (In August 2008, Avago annouced the purchase of the Infineon SMR-BAW group.) By 2005, SAW duplexers were beginning to finally make their presence in PCS CDMA cell phones. But, from a performance and robustness perspective, FBAR had grown out of its adolescent years into a full-blown mature technology with extremely high Q and excellent filter performance. Furthermore, single-digit yields were now a thing of the past and we were able to price our parts competitively. This was crucial; the high prices we charged for FBAR duplexers, early on, ended by late 2003. The estimated decline in average sales price was roughly 15% to 20% a year since 2003. In another irony of geopolitical proportions; many SAW companies felt that they needed to accelerate research and development in BAW, based on the success of HP/Agilent FBAR program. In so doing, precious R&D funds were diverted away from SAW development to begin serious BAW development. The jury is still out as to the wisdom of these investments.

5.2

FBAR Technology 5.2.1

Introduction

The Q-circle for a resonator is a representation of the reflection coefficient versus frequency on the Smith chart, and, can be thought of as a pictorial measure of the figure of merit (FOM) of the resonator. The closer the Q-circle is to the edge of the Smith chart, the better the FOM. Resonators with better FOM values can be made into filters that have better insertion loss of a filter, and a better shape factor (steepness of skirts), than filters using poorer FOM resonators. Insertion loss in the bandpass of a filter varies from the lower frequency edge of the passband to the higher frequency edge. Typically, the worst minimum insertion loss occurs at one of the two band edges. It is the worst-case insertion loss (for the worst temperature case), that defines the specification of the filter (not the minimum insertion loss). For the transmit filter of a PCS duplexer, the worst-case insertion loss usually occurs at the high-frequency band edge, 1,910 MHz, and at the maximum specified temperature, 85°C, while pumping the maximum amount of power through the duplexer from the power amplifier (frequency where maximum self-heating occurs). Since the materials used in the filter typically have negative temperature coefficient of frequency (TCF), the whole filter response moves down in frequency with increasing temperature. It is this case that defines the insertion loss of the filter. Conversely, the Rx filter in a duplexer running at the minimum temperature (−30°C) specified and at low power causes the Rx filter to move up in frequency. This corner case, at the low-frequency edge, defines the insertion loss of the Rx filter of a duplexer.

5.2 FBAR Technology

125

The best filter topology for making filters with steep skirts is a half-ladder design (series and shunt resonators—where the shunt resonators are mass loaded such that their resonant frequency is lower than the series resonators). The only “knob” one can use to make a superior half-ladder filter is to use resonators with a superior figure of merit or FOM. The relevant FOM, for any resonator in a filter, is the product of the coupling coefficient and the Q. Q turns out to be frequency dependent so the FOM is also frequency dependent and is defined as: FOM( f ) = kt2eff * Q( f )

(5.1)

Where the effective coupling coefficient, kt2eff is defined as; kt2eff = ( π 2 ) * (f s f p ) tan

[( π 2) * f

s

fp

]

(5.2)

and Q is the unloaded Q of the resonator. The two points on the Smith chart where the Q-circle crosses the real axes define the frequencies fs and fp, (series and parallel resonances). At these two points the reactance of the resonator is zero and the impedance is real. If the Q-circle is at the edge of the Smith chart, Q would be infinity. A measurement of the resistance at either fs or fp will give measurement of the figure merit (at fs and fp) and from the related FOM and kt2eff one can back out the unloaded Q (at fs or fp). One can relate the resistivity at fs or fp to the FOM as R p ~ FOM(f p ) * X o

(5.3a)

R s ~ X o FOM( f s )

(5.3b)

where Xo is the capacitance reactance of the resonator. One can write Xo in terms of the radian frequency, the plate capacitance, Co, and the effective coupling coefficient, kt2eff.

(

(

X o = ωC o 1 + kt2eff

))

−1

~ ( ωC o )

−1

(5.4)

and where Co = ε*εr*A/d A = Area of the resonator; d = Thickness of the piezoelectric region; εr = relative dielectric constant of the piezoelectric material. 2 The optimum value for kt eff is set by the relative filter bandwidth requirement. 2 The coupling coefficient, kt eff , is a property of the acoustic stack and is a material property that is independent of frequency. In fact, by definition, kt2eff is fixed by the two measured frequencies that define the series and parallel resonance—of the thickness extensional (or main longitudinal) mode, as measured on the Q-circle.

126

FBAR Resonators and Filters

For cell phone applications, the necessary bandwidth typically requires maximum values of kt2eff. There is only a small amount of design flexibility in varying kt2eff. Therefore, the best knob for obtaining large values of Rp (and, conversely, low values for Rs) is to improve Q. The Q, like kt2eff and kt2int, is also a function of the acoustic stack, (i.e., the choice of electrodes and how and where the acoustic stack is supported). Like kt2int, Q is also dependent on frequency. Q losses can have smooth dependencies on frequency as well as spurious losses (sharp degradation of Q at certain frequencies). If there were no parasitic lateral modes, the Q would be smoothly varying and the measured Q-circle would be a nice round circle on the Smith chart. This simple Q-circle could then be precisely modeled in the electrical domain by a few Rs, Cs, and Ls. This will be covered in more detail in the next section. Figure 5.7 shows the evolution of resonator Q from 1993 to 2007. Devices in Figure 5.7(a–e) use molybdenum electrodes in the acoustic stack (frequency ∼2 GHz), whereas in Figure 5.7(f), the Q-circle of an 820-MHz FBAR resonator using tungsten electrodes is shown. 5.2.2

Modeling of FBARs

There are three models that are used to understand and predict resonator performance. The one-dimensional Mason model [6] is a physical model that gives the electrical response as a function of the physical parameters and the thicknesses of these layers in the acoustic stack. One property of the Mason model is that it allows one to determine the effect of layer thickness on frequency. This model will also predict the kt2eff as a function of kt2int, electrode material, and the relative ratio of electrode thickness to piezoelectric material thickness. Thus, one uses the Mason model

Figure 5.7

(a–f) Evolution of the FBAR Q-circle starting from 1992 to 2007.

5.2 FBAR Technology

127

to design the acoustic stack. Figure 5.8(a) is a simplified version of the Mason model used on our ADS simulator (courtesy of Tiberiu Jamneala). The losses due to the top and bottom electrode as well as the AlN piezoelectric material are modeled as lossy transmission lines. However, we have found that over a narrow range of frequencies, one can model the losses by using lossless transmission lines to model the layers and by placing three resistors external to the impedance transformer. This is seen in Figure 5.8(b). The transformer ratio is π Xo/(kt2int*ZAIN), where Xo is given in (5.4), and the acoustic impedance, ZAIN = Sqrt[ρ*c33] where ρ is the mass density and c33 is the stiffness in the longitudinal direction. ZAlN is approximately 35 MRayls for AlN. The next model is really a framework of Newton’s equations along with the constitutive equations to create a finite element model (FEM) of the resonator. The FEM allows one to model the effects of the three-dimensional aspects of the resonator. However, this model takes a tremendous amount of computing power to finish a simulation and is often used in a postdictive capacity. In time, the role of FEM will change from postdictive to predictive. The third model is the Butterworth Van-Dyke model (BVD)—described in earlier chapters. The BVD model consists of a motional inductor, Lm, a motional capacitor, Cm, and a resistive loss term rm to account for the acoustic loss—all in series modeling the series resonance of an FBAR in parallel with a plate capacitor Cp. The BVD model will have two resonances; a series resonance, fs, and an antiresonance (sometimes referred to as the parallel resonance), fp, where

[L * C ] ~ = (1 [L * C ])* (1 + C

f ss = 1 fp

m

m

m

m

m

2 * C p ) = f s * (1 + C m 2 * C p )

(5.5)

We note from (5.5), that (fp − fs) is proportional to the ratio of Cm/Cp. 2 Equation (5.2) for kt eff can be fitted with the linear term (fp − fs)/( fp + fs) in the limit that (fp − fs) << (fp + fs). [Hint: set fs = 1 and plot ζ*( fp − 1)/(fp + 1) alongside the function (π/2)*(1/ fp)/tan(π/2)/fp as a function of fp. One will find that ζ is ∼4.8.] For 1 < fp / fs < 1.03, one can simplify (5.2):

Bot elect. AIN −C o Co

Top elect.

Bot elect. AIN

Top elect.

−C o

r series Co ro

rm (a)

(b)

Figure 5.8 (a) A simplified Mason model implementation in ADS [17]. The loss terms are buried inside the transmission lines that represent each layer in the acoustic stack (bottom electrode, AlN, top electrode). (b) A Mason model implementation in ADS where the transmission lines are made lossless and all the losses unique to FBAR are contained in the three resistors, rseries, ro, and rm.

128

FBAR Resonators and Filters

kt2, eff = 48 . * (f p − f s )

(f

p

+ fs )

(5.6)

This is a simpler expression for kt2eff [in contrast to (5.2)]. If one is working with a restricted set of kt2eff values in their acoustic stack, one can refine the variable ζ. For example, if kt2eff lies between 5% and 6%, ζ is 4.758. One can as easily relate kt2eff to the ratio Cm/ Cp from (5.5) and (5.6) to arrive at kt2eff ≅ 1.2C m C p

(5.7)

In the mBVD model we have three loss terms (versus the BVD model which has only the acoustic—or motional loss, rm). These three terms are rseries, ro, and rm. These are the same three resistance terms used in Figure 5.8(b). Experience with modeling real resonators leads one to realize that the BVD model will not model FBAR resonators very well, and that Q is not constant with respect to frequency (which would be the case for any device that could be fitted to a BVD model). A better fit between measured Q-circle and the BVD model, is to add two more loss terms to the BVD model; a resistor in series with the plate capacitor, ro, and a series resistor, rseries. Hence the term “modified” BVD or mBVD model. Figure 5.9 shows the Q-circle of a modified BVD, (or mBVD) circuit for a nominally 50-ohm reactance (Xo) at 800 MHz which is fitted to the measured Q-circle shown in Figure 5.7(f). The plate capacitance in Figure 5.9 is 3.47 pF. The motional capacitance is 0.176 pF and the ratio Cm / Cp is equal to 0.05072. Using (5.7), we get 2 kt eff = 6.08%. From Figure 5.7(f), one can pick fs and fp and using (5.2), one gets 5.82%. Alternatively using (5.6), one arrives at 5.87% for kt2eff. Even if rm, ro, and rseries of a fitted mBVD model to a measured resonator were independent of frequency, it is now obvious that the Q of the fitted resonator is a

C C2 C = 3.46975 pF

R R1 rm = 0.2525 Ohm

R R3 rseries = 0.3801 Ohm L L2 L = 182.4 nH R= m2

Figure 5.9

C C3 C = 0.1761 pF

m1

A best-fit mBVD to the resonator shown in Figure 5.7(e).

R R2 ro = 0.06 Ohm

5.2 FBAR Technology

129

function of frequency. This can be seen by modeling the series resonance at fs by a LRC circuit where Rs = rseries + rm and by modeling the parallel resonance at fp by a 2 tank circuit where here Rp = Xo /( ro + rm) and X0 is given by (5.4). Since there is no a priori reason why ro and rseries are equal, we know that Q is smoothly varying over frequency. In fact, for a resonator described by second-order equations, one would expect a Lorentzian shape for Q versus frequency. 5.2.3

Method of Ascertaining Q

It is instructional to calculate the unloaded Q at every frequency. This is done by calculating the energy stored in each of the reactance terms in the mBVD model (fitted to a real resonator) and dividing by the losses through each of the resistors.

(

)

Stored Energy = 1 C p * V 2 + C m * V 2 + L m * I 2 2 2 Dissipated Power in a cycle = I12 * rm + I 22 * ro + Itotal * rseries

(

) (2 π f )

Q unloaded = Q u = the ratio of the Stored Energy to the Dissipated Power in a cycle

I1 is the current thru the motional branch, I2 is the current thru the plate capacitor and Itotal = I1 + I2. This method of extracting Qu, uses the following procedure: 1. Fit an mBVD model to the resonator to get values for each reactance and resistor. 2. Calculate the stored energy and the dissipated power in one cycle at each frequency. 3. Taking the ratio to calculate Qu of the resonator. This is referred to as the brute-force method, or BFM, for finding Qu of a resonator [18]. The solid line (Lorentzian curve) in Figure 5.10 is a plot of the unloaded Qu, for the resonator shown in Figure 5.7(f) based on a fit to the measured Q-circle—Figure 5.9. It is worth noting that for this resonator the maximum Qu does not fall at fs or fp, but somewhere in between. It is also important to point out that in this fit, using Agilent’s Advanced Circuit Simulator (ADS), a product of Agilent’s EESOF division, one can find the best fit of a model to data using a least-squares fitting algorithm. The noisy data superimposed on the solid line fit in Figure 5.10 is from the equation Q m ( f ) = 2 πf * τ( f ) * Γ( f )

(1 − Γ( f ) ) 2

(5.8)

where the subscript “m” signifies that this is Q directly taken from measurement, and τ(f) is the group delay and τ(f) = −dφ/dω ≅ −δφ/δω

130

FBAR Resonators and Filters m3 freq = 888.900 MHz Q = 2503.28974 6000

m5 freq = 901.900 MHz Q = 5206.58378

m4 freq = 910.400 MHz Q = 3196.83084

m5

5000

Q3 Q

4000

m4 m3

3000 2000 1000 0 870

880

890

900

910

920

930

freq, MHz Figure 5.10

Q versus frequency using the brute force method and using (5.8).

|Γ(f)| is the magnitude of the reflection coefficient (S11) measured at frequency f It is very important to realize that f, |Γ(f)|, and τ(f) can be read off most network analyzers. Hence, (5.8) is a method of extracting the unloaded Q from the terminals of a resonator (any resonator!). Equation (5.8), when used properly, is relatively quick. There are caveats: 1. δφ/δω is calculated from two adjacent frequency points taken from measurement. If these frequencies are relatively far apart, then errors will accrue. Even using a large number of frequency points in a network analyzer (NWA), there is still the issue of differences of large numbers in the denominator. Hence, one obtains the noisy plot shown in Figure 5.10. 2. Below fs, we see that applying (5.8) gives negative Qs (not physically real). From the Q-circle we see rattles below fs in the south west quadrant of the Smith chart. The reason for the apparent negative Qs at certain frequencies below fs is due to the change in φ with respect to ω as the phase goes counterclockwise with respect to the center of the Smith chart instead of clockwise, hence, the change in sign. These rattles and small circles are due to parasitic lateral modes that coexist with the main piston mode of the resonator. These subresonances are better seen in small, high-impedance resonators with square or Manhattan geometries. 3. Equation (5.8) is only valid in the limit that the Q is high (Q ∼> 20). Equation (5.8) has its roots in a theorem by Bode found in his book Network Analysis and Feedback Amplifier Design [20]. And, indeed, if rm is large (compared to the natural impedance, Zo, of the resonator), then (5.8), as an approximation, fails.

5.2 FBAR Technology

131

4. It would appear that (5.8) fails if |Γ(f)| passes through the center of the Smith chart (|Γ(f)| = 0). However, for reasonable Qs, one can recover the full accuracy of (5.8) by transforming the Q-circle in such a way that it is recentered in the Smith chart [19]. An example of a transformation to a new coordinate system would be to first transform Γ(f) back into an impedance Z(f) where: Z( f ) = (1 + Γ( f )) (1 − Γ( f ))

and the transformed reflection coefficient for (5.8) is Γ ′( f ) = (Z( f ) − Z o′ )

(Z( f ) + Z ′ ) o

where Z o′ would be the impedance point near or at the center of the measured Q-circle. For example, if the resonator impedance, Zo, is 300Ω, then set Z o′ to 150Ω. Care must also be taken to convert the group delay as well, where φ = atan[Im(Γ ′(f))/Re(Γ ′(f))]. A simpler technique (easily accomplished on today’s network analyzer) is to change the source impedance from the default 50Ω to a new impedance such that the Q-circle now centered on the Smith chart. The brute-force method has its limitations. In the case of resonators with significant lateral modes, the simple mBVD does not do a good job fitting the Q-circle. Figure 5.11(a) is a measurement of a 240Ω square resonator resonating around 720 MHz. Superimposed on the Q-circle in Figure 5.11(a) is a simple mBVD fit and a fit

m2

(a)

(b)

Rser SRLC

C plate R plate

SRLC

SRLC

SRLC

SRLC

C mot L mot R mot

mBVD Model fo = 722.562 MHz

R-L Mode 1 fo = 721.12 MHz

R-L Mode 2 R-L Mode 3 R-L Mode 4 fo = 721.563 MHz fo = 721.95 MHz fo = 722.744 MHz

(c)

Figure 5.11 (a, b) A 720-MHz resonator fitted to a mBVD model, and (c) to an mBVD model with additional RLC elements to model some of the parasitic modes.

132

FBAR Resonators and Filters

using a more complicated circuit is shown in Figure 5.11(b). Here, better justice is done to the fit by adding four additional RLC series circuits. This is seen in Figure 5.11(c). Again, a least-squares fit in ADS is used to find the values for each of the elements in the more complicated model. These additional RLC elements represent some of the parasitic lateral modes of the measured Q-circle in Figure 5.11(a) around fs. Figure 5.12(a) shows calculated Q(f) using BFM for the simple mBVD model superimposed with a BFM calculation using the more complex model. Since the additional RLC elements model the parasitic modes due to Rayleigh-Lamb waves generated in the plate, this model is referred to as the RLC-mBVD model. Note that the Q versus frequency curve is no longer Lorentzian. Figure 5.12(b) contains the two calculated curves of Q versus frequency from the two versions of the fitted models (the mBVD and the RLC-mBVD model) superimposed with Q versus frequency measurements using (5.8) applied directly to measured data of the resonator and to a data list generated from the RLC-mBVD fit. One can see that by adding more RLC series elements, both a better fit to the Q-circle is obtained and the Q as calculated by BFM more closely resembles the actual Q extraction from (5.8). More elements added to the circuit shown in Figure 5.11 would give better fits—especially between near fp. 1.0E4 9.0E3 m5

8.0E3

m5 Tx1_freq= 7.22500E8 Q=7826.87892

m6 Tx1_freq=7.39700E8 Q=456.04249

7.0E3 6.0E3

Q

5.0E3 4.0E3 3.0E3 2.0E3 1.0E3

m6

0.0 −1.0E3 7.10E8

7.15E8

7.20E8

7.25E8

7.30E8

7.35E8

7.40E8

7.45E8

7.50E8

Tx1_freq freq, Hz

(a) 1.0E4 9.0E3

m5

8.0E3

m5 Tx1_freq=7.22500E8 Q=7826.87892

m6 Tx1_freq=7.39700E8 Q=456.04249

7.0E3 6.0E3

Q

5.0E3 4.0E3 3.0E3 2.0E3

m6

1.0E3 0.0 −1.0E3 7.10E8

7.15E8

7.20E8

7.25E8

7.30E8

7.35E8

7.40E8

7.45E8

7.50E8

Tx1_freq freq, Hz

(b)

Figure 5.12 (a) A plot of Q versus frequency using the BFM approach on the two versions of the mBVD model described in Figure 5.11. (b) Plot (a), but with additional curves obtained from (5.8) on measured resonator and the RLC-mBVD model superimposed on the two plots.

5.2 FBAR Technology

133

There are two conclusions from this; one is that there is significant energy stored in the resonator in the form of the parasitic modes. These modes actually contribute to making Q(f) larger or smaller, than one would assume using a simple mBVD model. And second, that fitting the resonator with the mBVD model to extract Q has its limitations, in contrast, the Bode equation for Q is quite useful at showing the true unloaded Q for any reasonably high Q resonator. 5.2.4

The Rayleigh-Lamb Modes

The parasitic modes seen on the Q-circles in the previous figures are caused by a variety of modes including the excitation of the Rayleigh-Lamb (RL) modes inherent in a plate. These RL modes are a superposition of a longitudinal mode with a k-vector kl and a shear vertical mode with an associated k-vector, ksv. In an infinite three-dimensional volume, these two waves do not interact. However, in a plate, (infinite in x, y but of finite thickness in z), the scattering of either mode will convert into both shear and longitudinal modes. As shear and longitudinal waves scatter off the top and bottom surfaces of the plate, one can imagine a case where for a certain angle associated with a frequency, the two modes will recombine. In this case, the allowed modes satisfy the “transverse” resonance condition and reconstruct themselves after successive reflection from the lower and upper faces of the plate [21]. These modes are referred to as the Rayleigh-Lamb modes, where the RL modes have a k-vector β, and k12 = ( ω v l ) − β 2 and ksv2 = ( ω v s ) − β 2 2

2

(5.9)

where v1 is the velocity of the longitudinal wave and vsv is the velocity of the shear vertical wave. Figure 5.13 shows the dispersion curve highlighting all of the lateral modes in a simple plate of type II, 2.5 μm thick, piezoelectric AlN. In this figure, all of the various modes have been labeled accordingly. Besides the five Rayleigh-Lamb modes (S0, S1, A0, A1, and A2) there are also the three thickness shear modes (“pure” shear modes versus vertical shear modes). There is no conversion of pure shear into longitudinal or vertical shear modes for a simple plate. The Rayleigh-Lamb modes can be further broken down into two families; the symmetric modes(s) and the antisymmetric(a), or flexure modes. The symmetric RL modes are the ones that create the subresonances seen on the measured Q-circles. It is these subresonances that will cause strong ripple in the passband of any filter using these high Q resonators. Hence, one must take care to analyze the source of these modes and if possible, suppress them or eliminate them. Flexure modes are not easily measured electrically. The reason one cannot electrically “see” flexure modes is that the relative spacing between the electrodes remains fixed and therefore voltage between electrodes due to flexure modes is constant. The reason one can only electrically see the symmetric modes on a measured Q-circle (as opposed to the flexure modes), is that at any given point on the resonator, there is a periodic expansion and contraction of the piezoelectric material, therefore inducing a periodic voltage across the two electrodes at that point. Assuming that a plane defined in the center of the piezoelectric film (parallel to the top and

134

FBAR Resonators and Filters

3.0e+009 A2 mode (flexure) Shear Modes: 1 rst and 2nd Harmonics

2.5e+009

S1(−)

Frequency

2.0e+009

S1(+)

fd

0.5e+009

S1 Sysmmetric Mode: Characteristic “Swoon”

A1 Mode (flexure mode) 1.0e+009 S0 Mode (symmetric mode)

5.0e+008

0th Order shear mode

A0 Mode (flexure mode) 0.0e+000 0.00

Figure 5.13

0.05

0.10

0.15

0.20

0.25

0.30

0.35

0.40

0.45

0.50

Calculated dispersion curve for an AlN plate of thickness 2.5 μm thick.

bottom surfaces) defines 0 voltage, the voltage of the top electrode would measure at V(x,y)/2, while simultaneously, the voltage of the bottom electrode would measure −V(x,y)/2, where V(x,y) is a voltage at any point x,y on the resonator. The mode that has a very strong electrical signal would be the TE-1 or the first thickness extensional mode near fs (also referred to as the S1 mode). Here the top and bottom electrodes move up and down in opposing motion with respect to each other. But, unlike the fundamental longitudinal mode, the vertical displacements vary locally across the face of the resonator. The pure longitudinal mode (occurring only at fs) has periodic opposing motion between the two electrodes and the same phase everywhere on the top plate and the same phase but 180° out of phase on the bottom plate. As energy “sloshes” from side to side in the resonator, below fs, the phase of the voltage as measured at any point on the top electrode varies as a function of the x,y location (and likewise for the bottom electrode). This leads to variations in voltage on the same electrode at any given point in time. If the voltages on an electrode are different in two places, current will flow between the two points. This phenomenon, piezoelectrically induced eddy currents, is a source of energy loss. The hit in Q comes from the I2R losses due to eddy currents flowing in each electrode. For a free-standing, apodized, membrane using AlN for the piezoelectric, the majority of losses due to this phenomenon lie below fs. Figure 5.14 is a measured Q-circle with two separate mBVD fits. In the mBVD model, the only change is the element rseries. Here, the series resistor term was increased five times in order to fit the measured data below fs. All other parameters are unchanged. From Figure 5.14 one

5.2 FBAR Technology

135

mBVD fit abovef s

mBVD fit belowfs f [MHz]

SH-2 of resonator

fs Clear evidence of a cut off frequency

fs

S1- or TE-1 of resonator

Im(ν) L R C Rmotion2 Cmotion2 Lmotion2 L=71 nH R=0.265 Ohm C=91.4 fF R=

R Rseries2 R=0.26 Ohm

C Cplate2 C=1.563 pF

~1.3 Ω

Re(ν) (1/cm)

R C Rmotion2 Cmotion2 R=0.265 Ohm C=91.4 fF

L Lseries2 L=-0.13 nH R=

R Rseries2 R=0.26 Ohm

R Rplate2 R=0.6474 Ohm

C Cplate2 C=1.563 pF

L Lmotion2 L=71 nH R=

L Lseries2 L=-0.13 nH R=

R Rplate2 R=0.6474 Ohm

Figure 5.14 The Q-circle of an FBAR resonator. Two different mBVD models were used to model the device below and above fs, the series resonance.

can see that rseries changes dramatically at fs. This suggests that there is specific RL mode that has a cutoff frequency at fs. Indeed, if one calculates the Rayleigh-Lamb dispersion equations, one will see that just below fs, there are four modes. One of those modes, TE-1 mode, has a cutoff frequency at fs. One can derive the dispersion diagrams for the RL modes based on the “transverse” resonance principle described in volume II of Bert Auld’s book, Acoustic Fields and Waves in Solids [21]. Following the nomenclature of (5.9) and the teachings of Auld, one can write down the dispersion relations of the symmetric case for a simple plate as R = −4 β 2 kl ksv

(k

2 sv

− β2

)

2

(5.10)

and for the asymmetric case R −1 = −4β 2 kl ksv

(k

2 sv

− β2

)

2

(5.11)

where R = tan(ksvd/2)/tan(k1 d/2) and d = plate thickness. Figure 5.15 is measured and calculated dispersion curves (ω−β) for the first two pairs of symmetric and antisymmetric modes, of an 920-MHz W/AlN/W resonator. The measurement and simulation techniques are discussed by Telschaw et al. [22]. The labeled S1 curve (or TE-1 mode) has a minimum at a frequency below fs; we call this the dilational frequency, fd (where B → 0.08 μm and fd = 900 MHz). At this point, S1 bifurcates into two branches, one, with a negative group velocity, S1(−), that terminates at fs where the k-vector is zero. This represents the case where the

136

FBAR Resonators and Filters 1.0

SH

S1

0.9

900

0.8

800 700

r

ea

A1

0.7

Frequency (GHz)

Ω/2p (MHz) ~1/vshear

1000

Sh

0.6

600 500

0.5

0.4

400 300

S0

0.3

A0

0.2

200

0.1

100 0

0.05

0.1

0.15

0.2

K/2π (1/μm)

0.25

0.3

0.35

0.0 0.00

0.05

0.10

0.15

0.20

0.25

0.30

0.35

kx (1/μm)

Figure 5.15 Measured and calculated dispersion curve of a Tungsten electrode resonator (fs ~920 MHz). The thicknesses of the stack were used to generate simulated dispersion curves. One can overlay the two curves and see an excellent match of all frequencies and wave numbers for each mode. (Courtesy of J. Larson and A. Shirakawa.)

lateral wavelength of this TE-1 mode is infinity and the two electrodes uniformly move up and down like a reciprocating pistons. The S1(−) curve is referred to as the thickness extensional mode or TE-1. The TE-1 mode is quite strong in free-standing BAW devices using AlN as the piezoelectric material. However, above fs, the only allowable symmetric RL modes are the S0 mode and the right branch, S1(+), of the S1 mode. These two modes have relatively large wave numbers (β large), corresponding to smaller wavelengths. The device shown in Figure 5.15 has a total thickness of a membrane with 0.8 μm tungsten electrodes and the AlN thickness of 2.24 μm, the resonator will resonate at 920 MHz. The area of the square resonator measured Figure 5.15 would be 25,600 μm2 and an edge would be 160 μm. From Figure 5.15, the wave number for f −1 = fd is 0.08 μm giving a wavelength of 12.5 μm. As one travels along the S1(−) dispersion curve, the wavelength gets longer and longer, becoming infinite at β = 0. There will be fewer wavelengths of the S1(−) mode trapped between electrode edges as frequency approaches fs. In comparison, the wave number (and wavelength) just above fs for the associated S0 mode is 0.16 μm−1 (or 6.25 μm), and the wave number (and wavelength) at −1 just above fs for the associated S1(+) mode is 0.11 μm (or ∼9 μm). For an edge separation of 160 μm, the number of S0 wavelengths in one direction is ∼25.5. And, for the S1(+) mode there are ∼17.5 wavelengths—just above fs. Another observation is the relative group velocities of the S0 mode and right branch or the S1(+) mode, versus the left branch of the S1(−) mode (or TE-1 mode). Ignoring the sign of the velocities, we see that the TE-1 mode group velocities are very small; hence, for lateral boundary conditions given above, we would expect to see lateral subresonances grouped closely together in frequency compared to subresonances of the S1(+) and S0 modes. Simple “eyeballing” the relative slopes from Figure 5.15 would suggest that for the same spacing, TE-1 subresonances will be two to three times closer spaced.

5.2 FBAR Technology

137

Figure 5.16 is a set of interferometer measurements taken just below the dilation frequency (onset of the S1 mode) to fp. For frequencies greater than fs or less than fd, the membrane displacement is relatively small. Furthermore, only fine features (high wave numbers) are seen. But, between fd and fs, the membrane displacement is qualitatively quite different. Here, the local displacements within the membrane become large and as the TE-1 mode S1(−) approaches fs, the wave numbers become larger and the features become more gross. Figure 5.17(a) is a Q-circle of a square resonator and Figure 5.17(b) shows the log magnitude of a measured S11 [23]. One can see that below fs, the TE-1 subresonances are closely spaced. We see that the frequency spacing of these electrically measured subresonances literally map out the dispersion curve of the TE-1 mode. Above fs, we see the electrically measured subresonances spaced further apart, as expected for the higher group velocities associated with the S0 and S1(+) modes. 5.2.5

Apodization

Prior to discussing apodization, it is worth looking at the measured TE-1 dispersion curve for very narrow and long FBAR resonators. Figure 5.18 shows such a layout for two such designs. The long edges are parallel and the two short ends are angled. This topology is chosen to exacerbate the RL TE-1 mode in the narrow direction and “smear” out standing waves in the long direction. As can be seen from the two

Freq = fd – 1 MHz

Freq = fd + 1 MHz

Y (μm)

200

200

200

300

300

300

400

400

400

100

200 300 X (μm)

400

100

200 300 X (μm)

400

100

Freq = fs + 1 MHz

Freq = fs – 1 MHz 100

200

200

300

300

300

400

400

400

100

Figure 5.16

200 300 X (μm)

400

400

100

Y (μm)

200

200 300 X (μm)

Freq = fp

100

Y (μm)

Y (μm)

fd < Freq < fs 100

100

Y (μm)

Y (μm)

100

100

200 300 X (μm)

400

100

200 300 X (μm)

Interferometric measurements of the local displacement in an apodized resonator.

400

138

FBAR Resonators and Filters x = 1 x = 1.5

x = 0.5

x = 2

r = 0

x = 4

x = 0

(a)

r = 1

r = 2

r = 1 2

x = −4 4 x = −2 2 x = −0.5 0.5

x = −1.5 1.5 x = −1 1 0.98 0.96

fd

0.94 0.92

(b)

1.8

1.9

2.1

fs

2.2

0.88

Figure 5.17 Q-circle (a) of a square resonator, and (b) the measured magnitude, |Γ|. The “landmark” frequencies at fd and fs are indicated. Between fd and fs, the quasi-periodic oscillations are 4 to 5 MHz. Above fs, the periodicity changes to 15 MHz. Interference between the S1(+) and the S0 modes can be seen above fs.

sets of dispersion measurements, one sees a discrete set of relatively large-signal intensity (from the acoustic microscope). The wave numbers are equally spaced in wave number (but not in frequency) and are inversely related to the electrode width (in the narrow direction). Electrically, these bright spots are measured at the various subresonances seen on the Q-circle (below fs). If one assumes that the strain is zero at the two edges of the swimming pool, then one can guess a simple solution for the strain over the width to be S( x ) = S o sin(2πβx )

(5.12)

At x = 0 and L, S(x) = 0, then 2πβL = n π (n = 1, 2, 3, 4, ...) If we assume that the forces are opposite in sign at the two edges, then we can replace n with (2 n − 1). Or, for the standing waves in the narrow direction β = (2n − 1) (2 * L) n = 1, 2, 3, 4, K

(5.13)

What can also be seen is that the measured dispersion curve is continuous (or nearly continuous) measured wave numbers. That is, other allowed solutions exist

5.2 FBAR Technology

139

2000 2000

1950

f (MHz)

f (MHz)

1950

1900

1900

. .

1850

1800 0

.

1850

0.05

0.1

+ Electrode

0.15

− Electrode 80 μm wide

1800 0

0.05 0.1 0.15 0.2 0.25 0.3 0.38

+ Electrode

− Electrode 40 μm wide

Figure 5.18 Two 500 μm-long FBAR resonators with 80- and 40-μm widths respectively, apodized at both ends and their measured dispersion curves (measurements courtesy of J. Larson).

between the expected standing waves set up in the narrow dimension. These quasi-continuous states exist due to the angled edges at the far ends. Standing waves here are spaced quite close in frequency and wave number. It is assumed that the acoustic microscope (in this mode of operation) can only see standing waves. Hence, the nearly continuum of measured states on the TE-1 dispersion curve suggests nearly a continuum of standing waves. Since, the spacing of these nearly continuous states are closely spaced, we can assume that the value of n for any point is very large. At nearly every frequency there is another allowed standing wave. If one were to imagine a ray trace emanating from one of the angled edges, and follow that over multiple reflections from the other three edges, there would be a large number of reflections before the ray comes back on itself. This implies many wavelengths, or another way is to say that n in (5.13) is very large. This observation explains how apodization, or changing the shape of the electrodes from Manhattan geometry to a non-Manhattan geometry would help smear those modes above fs. By making edges nonparallel, we recognized that the path lengths for the fundamental subresonance would greatly increase (e.g., n times longer). Therefore, a standing wave can exist at almost every frequency (Δf between standing waves is very small). Thus, apodization smooths out the variation in electrical subresonances. The downside to apodization is obvious; the subresonances are much more closely spaced, giving the appearance of a nearly continuum of states. But, the loss is no longer limited to discrete frequencies but is also smeared out over all frequencies. Figure 5.19 is a Q versus frequency [using (5.8)] of two resonators identical in area, but one shaped as a square and the other as a squashed pentagon. This leads to an interesting observation. The Q is larger for square resonators for those frequencies

140

FBAR Resonators and Filters

1.0E4

m7 m5

5.0E3

freq (700.0 MHz to 760.0 MHz)

0.0 −5.0E3 −1.0E4 −1.5E4 700

Figure 5.19

710

720

730

740

750

760

Q versus frequency plots for two resonators of identical area, but with different shapes.

in-between the subresonances (we are only concerned with those resonances above fs). What this tells us is that energy loss (presumably out of the edges of the resonator) are real and of measurable concern. When one correctly apodizes the membrane, one creates loss at almost every frequency above fs (unlike loses for a square resonator at v/2*L frequency intervals). Apodization smears out the Q losses due to these RL modes S0 and S1(+). However, it neither eliminates nor suppresses these lateral modes. By careful design of the apodization, we have been able to obtain a nearly smooth Q-circle [an example is seen in Figure 5.20(a)]. Here a quadrilateral design was used to apodize the shape. Figure 5.20(b) is the log magnitude of the reflection coefficient. One can see how much smoother the reflection coefficient is above fs. However, the TE-1 modes are still quite strong, and given that the wavelengths are much longer for this mode, the subresonances are still relatively strong. 5.2.6

Frames

Since 1997, work at Avago (then HP) focused on Q enhancements for apodized resonators. As stated earlier, kt2eff is a limited knob when trying to improve Rp. Most high-volume filter applications require bandwidths where only the maximum kt2eff is useful. The kt2eff , in turn, is limited by the intrinsic coupling value (material limited) and the configuration of the electrodes (e.g., W versus Al electrodes). No such limitation is forced on Q. Improve Q by two times, Rp and Rs improve by two times, and with that, the insertion loss and shape factor of any filter using FBAR resonators improves. In 2003, Kaitila et al. presented a paper on the use of frames that would eliminate lateral modes (as opposed to suppressing or smearing of modes via apodization). This is discussed in much more detail in Chapter 2 and in [24]. In all of the published papers on frames since 2003, the focus has been on FBARs using ZnO

5.2 FBAR Technology

141 x = 1 x = 1.5

x = 0.5

x = 2 r = 0

x = 4

x = 0

r = 1

r = 2

r = 1 2 x = −4 x = −2 x = − 0.5 x = −1

x = − 1.5

(a)

0.96 0.94 0.92 1.8

1.9

2.1

2.2

0.88

(b)

Figure 5.20 Q-circle (a) of a quadrilateral (i.e., apodized) resonator, and (b) the measured magnitude, |Γ|. Between fd and fs, the quasi-periodic oscillations are more smeared, but still visible.

as the piezoelectric or SMR-BAW devices that use AlN. In both cases, the TE-1 piston mode exists above fs—not below. In the case of FBARs using AlN as the piezoelectric material, the TE-1 mode lies below fs. This is one fundamental difference between FBAR versus SMR-BAW and has been discussed elsewhere [25]. Thus, the application of a frame around the perimeter of the resonator—as taught by Kaitila—will not work. This is made clear in Figure 5.21(a). In the case of SMR-BAW, the TE-1 mode rises above fs. [Note: Even though ALN is a type II piezoelectric material, if there is sufficient oxide in the acoustic stack (and/or mirror for SMR-BAW) the resonator will behave like a type I acoustic stack where the TE-1 mode is above fs.] The frame region for a type I acoustic stack (a raised frame) will also have a TE-1 mode rising above fs, but due to the thicker acoustic stack, the TE-1 mode for this region is overall at a lower frequency and terminates at f = f s′ where f ′ < f s . If the thickness and width of this frame is chosen correctly, one can create an “eigensolution” consisting of: (1) a uniform displacement throughout the center region (the majority of the resonator area), (2) an exponentially decaying function outside of the resonator, and (3) the frame region where the amplitude and first derivative match at each inter-

142

FBAR Resonators and Filters

f [MHz]

TE-1 mode of resonator

f [MHz]

.

TE-1 mode of raised frame

fs fs ’ SH-2 mode of resonator

Im(β)

Re(β) (a)

(1/cm)

SH-2 mode of resonator

.

TE-1 mode of recessed frame

TE-1 mode of raised-frame region of resonator

Im(β)

Re(β)

TE-1 mode of resonator

(1/cm)

(b)

Figure 5.21 The dispersion curves of a type I resonator such as a SMR-BAW (a) and equivalent curve for a type II resonator such as an FBAR using AlN piezoelectric (b). Also shown are the dispersion curves for the frames (raised or recessed) for the two kinds of resonators. For a thin delta thickness, there will be a solution in the frame such that the center region motion will be uniform.

face. In particular, if the framed region has a real solution such that the eigenfunction and its first derivative are continuous, no other Eigensolution can exist by reason of orthogonality and thus, no subresonances will occur. Without this breakthrough, the SMR-BAW would have never reached the performance it enjoys today. For FBAR using AlN, one would have to use a recessed frame to accomplish the same thing. In this case, the perimeter is at a higher frequency compared to the center or main region of the resonator. If the width and amount of removed material is done correctly, there will be an allowable eigensolution where the center region displacement is uniformly flat across the center region. Figure 5.21(b) is the TE-1 dispersion curve for a type II resonator (an FBAR consisting of a free-standing membrane and used AlN as a piezoelectric material). The solid line represents the center region (the majority of the resonator area) and the dotted line is the TE-1 dispersion curve for the recessed frame around the perimeter. We have done this and we see suppression of the S1(−) mode and an improvement in the Qu below fs. The amount of improvement for an optimized value of the recessed frame width and thickness will vary depending on if the electrode is Mo (a big improvement) or W electrodes (a more modest improvement). However, above fs, at fp we see a degradation of Rp. The Qu, in general, is degraded everywhere above fs. The reason for the degradation near fp, is believed to be the fact that for those RL modes, S0 and S1(+), the recessed frame acts like an acoustic horn, foucusing the energy emitted from the membrane into the silicon anchoring points. It is well-known that one can match two dissimilar transmission lines by the use of a quarter-wave line whose impedance is the geometrical mean of the two transmission

5.2 FBAR Technology

143

lines. In the case of an FBAR, the center region is the most heavily loaded acoustic stack and the region between the electroded FBAR and the anchoring silicon edge (an area missing a top electrode) is much less heavily loaded. The recessed frame—lying between the two regions—is more heavily loaded than the outer region, but less so than the inner region. Thus, the recessed frame acts like an impedance matching element for RL lateral waves. Once energy has left the resonator and is launched into the supporting substrate, one can assume that energy is lost and hence, will limit the Qu. When we observe membrane (with recessed frame) motion at f90 (midway between fs and fp) under the acoustic imaging microscope, we see a large displacement at the perimeter of the resonator. This displacement is larger than the relative displacement of the center region, indicating a large amount of energy escaping the resonator. This is not the case for a resonator with optimized raised frame. So, what happens if one puts a raised frame around the periphery of a type-II acoustic stack? The answer is that if done correctly the Rp is greatly enhanced over a standard resonator without a raised frame [26]. The raised frame for a type II FBAR (i.e., a free-standing resonator using AlN piezoelectric) acts like a large impedance mismatch. In Figure 5.21(b), the dotted line represents the dispersion curve for a raised frame. For RL modes below fs (primarily the TE-1 mode) the perimeter frame acts as a energy barrier. Only evanescent waves are allowed inside the frame FBAR for those frequencies above f s′ of the frame. Thus, the frame acts to be a very high quality reflector for the TE-1 RL generated lateral waves. The net effect is that below fs, the trapped lateral modes have higher Qs and thus create larger rattles in the Q-circle along the southwest quadrant. However, for the S0 and S1(+) modes (below and above fs), the frame will have allowed states. But, above fs, we see the measured Qu to be significantly better than a resonator without raised frames around the periphery. The higher order RL modes (i.e., the S0 and S1(+) modes) are certainly allowed in the frame region FBAR. However, movies made of the motion of the acoustic resonator with a raised frame (using an acoustic microscope), clearly show that the energy is also well trapped inside the resonator (versus no frame). In the latter case, energy is seen leaking out into the silicon with a commensurate amount of bending occurs between the edge of the top electrode and the edge of the swimming pool. It is interesting to look at the fitted mBVD model for two resonators of equal area and stack make-up but one with a raised frame and the other without a raised frame. Indeed, what one finds is that the only term that changes dramatically is the ro term. This is the resistor in series with the plate capacitance. This value changes by 5 to 10 times as one compares a standard FBAR with one having an optimal width raised frame. So, what is happening? We argue that the raised frame belongs to part of a lateral Bragg reflector (BR). Since the physics of frames are quite different, one might wonder at what happens when one combines a recessed frame with a raised frame. A patent on this concept was filed in 2005 [27] and was independently published by Thalhammer et al. of Infineon in 2006 [28]. To see if there is any merit to this concept, we made a series of FBAR resonator layouts with varying frame widths. If there is a dependence due to interference, one would expect to see a periodic dependence on Qu (or Rp—since kt2eff only slowly

144

FBAR Resonators and Filters

changes with frame width) versus width. Also, as part of the experiment, we also laid out resonators with varying widths of recessed frames. The layouts thus constituted a two-dimensional array of resonators where the inner recessed-frame resonator widths varied along the x direction and the outer raised-frame resonator width varied along the y direction. Figure 5.22(a) shows Rp as a two-dimensional function of the recessed-frame width (horizontal axes labeled E) and an outer raised-frame width (vertical axes labeled A). The array is repeated multiple times across a 6” wafer and the median value Rp for each “flavor” of resonator is plotted in Figure 5.22(a). The frame width is varied along the y axes from zero to 11 μm. As one can see, the Rp is periodic with respect to the raised-frame width (1/4, 3/4, and 5/4 periodicity at 2 μm, 6.5 μm, and ∼11 μm). From Figure 5.22(a) and assuming the Bragg relationship, the maximum reflection occurs for that frequency whose wavelength is 8.5 to 9 μm (for a wavenumber of 0.11 μm−1). From Figure 5.22(b), we can infer that the dominant mode that leaks energy from the edges is most likely the S1(+) mode and not the S0 mode. In Figure 5.22(b), we see that at fp, the S1(+) mode had a wavelength of about 8 ∼9 μm. This leads us to the conclusion that the S1(+) mode is the larger contributor to loss (as compared to the S0 mode). Therefore, one can maximize Rp (at fp) by choosing a frame width to be λ (2n − 1) , where λ is the wavelength of the S1(+) mode at fp. However, it should be 4 emphasized that the added benefits of a raised frame around the periphery must be accompanied by appropriate apodization of the resonator. In contrast with the raised frames, varying the widths of recessed frame devices shows no such periodic dependence of Rp with width. Again, this is seen in Figure A E

Rp 11.0 10.0 9.0

980 960

S22_RP

8.0

<= 2500

940

7.0

<= 3000

920

6.0

<= 3500

5.0 4.0 3.0 2.0 1.0 0.0

<= 4000 <= 4500

860

<= 5000

840

<= 5500

820

E (μm) (a)

<= 6500

.

900 880

<= 6500

0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0

f(MHz)

A (μm)

W_FBAR_plain_B8880081_WVFFT

1000

Legend

0

0.05

0.1

.

0.15

0.2

0.25

B (1/μm) (b)

Figure 5.22 (a) A two-dimensional parameter plot of median Rp versus recessed-frame and raised-frame dimensions (courtesy of Alex Shirakawa). The insert gives the detail of the recessed/raised frame design. The measured dispersion curves are given in (b). At fp, the wave number for the S1(+) mode is ~0.11 to 0.12 −1 −1 μm and the wave number for the S0 mode is ~ 0.19 μm .

5.2 FBAR Technology

145

5.22(a); the x-axis is the variation in the recessed frame (inside the raised frame) and varies from 0 to 10 μm. This shows that the two techniques, a raised frame and a recessed frame, are governed by completely different physical phenomena. As one can see from Figure 5.22(a), varying the width of the inner frame only serves to lower Rp. Figure 5.23(a) shows the behavior or kt2eff versus frame width. The decrease in kt2 as the raised-frame width is increased is due to the fact that the raised-frame FBAR acts like capacitor in parallel with the active area. Thus, the ratio of Cm/Cp decreases as width increases and kt2 decreases. In the case of recessed frames, kt2eff increases with recessed-frame width increase. Kaitila theorizes in his paper [24] that the more effective a recessed-frame design is, the less energy will be coupled into the spurious RL modes, and thus, more acoustic energy remains for the desired longitudinal mode. In Figure 5.23(b), we see that only one width (x = 4.5 μm) will work to improve the Q below fs. However, beyond 6 μm, the recessed frame only acts to degrade Qs. Figure 5.24(a) is a plot of a resonator Q-circle with only a raised frame (optimized for width and thickness) superimposed on a Q-circle of a resonator with a recessed frame (optimized for width and thickness). Also shown are the two Q versus frequency plots of the two devices [using (5.8)]. One can see that above fs, the raised-frame device clearly has a better Q than the recessed-frame device and vice versa below fs. Figure 5.24(b) is a plot of two resonators, one with an optimized combination of recessed and raised frames and a resonator with no frames at all. As can be seen from the Q versus frequency plots, the Qu is better everywhere when using an optimized combination of recessed and raised frames. One drawback should be mentioned, each frame requires an additional masking step. Thus making the process that much more complex. The bottom and top electrodes used for these resonators is made of tungsten and average 5,600A and the AlN thickness is about 1.9 μm. 5.2.7

Temperature-Compensated Resonators

Most materials used in BAW devices (AlN, Mo, W) have negative coefficients of frequency terms. This arises from the fact that the stiffness coefficient of these materials is inversely proportional to temperature. As temperature rises, the stiffness decreases. In contrast, oxide films have positive temperature coefficients. Thus, one can compensate for the negative temperature dependence of the other films by placing an oxide layer in the acoustic stack [29]. One can put the oxide outside of the metal electrodes or inside the electrodes. The former case will only marginally improve the temperature drift and inside, the oxide—if designed properly—can cancel out all or nearly all of the linear coefficient terms. However, there will be a cost. The kt2eff will decrease for any amount of oxide inside the stack. For zero-drift resonator applications, the application of an oxide is very useful. Figure 5.25 is the residual temperature drift of an acoustic stack with a thin layer of oxide introduced into the stack. As can be seen, this device has a residual temperature dependence that can be—to the first order—modeled by a parabolic equation. The raw “uncompensated” performance of this resonator is only two to three times

146

FBAR Resonators and Filters k 2t eff

Legend

11.0 10.0

S11_KT2

A (μm)

9.0 8.0

<= 6.000

7.0

<= 6.125

6.0

<= 6.250

5.0

<= 6.375

4.0 3.0

<= 6.500

2.0

<= 6.625

1.0

> 6.625

0.0 0.0

1.0

2.0

3.0

4.0 5.0 E (μm)

6.0

7.0

8.0

9.0

(a)

Qs Legend

11.0 10.0

S22_QS

A (μm)

9.0 8.0

<= 750

7.0

<= 1000

6.0

<= 1250

5.0

<= 1500

4.0 3.0

<= 1750

2.0

<= 2000

1.0

<= 2250

0.0

<= 2500 0.0

1.0

2.0

3.0

4.0 5.0 6.0 E (μm)

7.0

8.0

<= 2750

(b)

Figure 5.23 These two-dimensional parameter plots are of (a) k2t and (b) Q at fs . The data is from the same array of resonators described in Figure 5.20(a) and in the text.

worse than quartz (the two criteria for this comparison is temperature stability and the f*Q product—a normalized figure of merit when comparing phase noise of oscillators at running at different frequencies). For many applications (GPS receivers

5.2 FBAR Technology

147 2500

3500 3000 2500

fs

fp

fs

2000

2000

fp

1500

1500 1000

1000 500

500

0 -500 860 865 870 875 880 885 890 895 900 905 910 915 920

freq, MHz

0 860 865 870 875 880 885 890 895 900 905 910 915 920

freq, MHz

(a)

(b)

Figure 5.24 (a) The Q-circle of a resonator with a recessed frame (superimposed is a resonator with a raised frame). (b) The Q-circle of a resonator with an optimized recessed plus raised frame superimposed on a Q-circle with no frames. Q versus frequency [using (5.8)] is plotted for each pair of devices in (a) and (b). The resonant frequencies fs = 886 MHz and fp = 910 MHz. All resonators are apodized.

XO

Frequency deviation in ppm

30 15 0 −15

f = −α*(T − Txo )

2

−30 −45 −60 −40 −20

0

20

40

60

80

100

Temperature (°C) Figure 5.25 The residual temperature variation after the linear temperature terms of each of the materials (Mo electrodes, AlN piezoelectric, and a positive temperture compensation oxide).

148

FBAR Resonators and Filters

being one notable exception), the temperature drift of this resonator is acceptable. For example, in the case of many oscillator applications, a drift of ±150 ppm is acceptable. Once the linear drift term has been eliminated, one can define the residual parabolic dependence with two variables; the quadratic term, α (typically −22 ppb/oC2) and the temperature cross-over point (TXO) where the derivative of the frequency versus temperature is flat). To get the minimum frequency drift over a temperature range of −20 to 80°C, one would design TXO to be at 30°C. The oxide is relatively soft compared to both AlN and Mo; therefore, one would expect that this would impact Q. Indeed, we do see a consistent trend of lower Q-values for compensated resonators versus uncompensated resonators [29]. Going back to the mBVD model, we would expect that rm, the motional loss term, would be most changed. When modeling two identical resonators (with and without the oxide layer), we see that indeed the rm term increases by ∼8× when comparing resonators without oxide in the stack. Figure 5.26 shows the Q-circles for the two resonators and the best-fit mBVD values for each resonator. Unlike rm, the other loss terms, rseries and ro, remain unchanged. The performance of a 600-MHz differential Colpitts oscillator was presented using a temperature compensated FBAR in 2007 [30]. The device showed ±25 ppm stability over 100°C and a phase noise at 1-kHz offset of −104 dB. This performance compares well with quartz oscillators. Using dividers to bring the frequency down 32 times (by using 5 dividers), one can extrapolate the phase noise performance to be −134 dBc at 1-kHz offset at 20 MHz (where quartz

R R C Rseries1 Rm1 Cm1 R=0.25 Ohm R=0.676 Ohm C=44.7 fF

C Co1 C=0.93 pF

L Lm1 L=1.086 uH R=

L Lseries1 L=0.01 pH R=

R ro1 R=8.76 Ohm

0.7 Ohms 30 Kμm2 Resonator without temperature compensation

m2

30 Kμm2 Resonator with temperature compensation

~ 0.325 Ohms

R Rseries3 R=0.4 Ohm

5.0 Ohms !!

~8.7 Ohms

L R C Rmotion3 Cmotion3 Lmotion3 R=5.0 OhmC=19.53 fF L=3.59 uH R=

C Cplate3 C=0.9 pF

R Rplate3 R=8.82 Ohm

L Lseries3 L=0.01 pH R=

C plate is the same for both cases ~0.9pF

Figure 5.26 The Q-circles for two devices identical except one has an additional oxide layer to create a temperature compensated resonator. This changes the resonant frequency (and hence the Lm, Cm terms in the mBVD model).

5.2 FBAR Technology

149

crystal oscillators work). This is comparable to many of the TCXO quartz oscillators used in phones today [31]. 5.2.8

Coupled Resonator Filters

Ken Lakin describes coupled resonator filters (or CRFs) in Chapter 2 and so it will not be described here. However, in the description given, Ken describes using a quasi-Bragg reflector consisting of three or more layers of differing acoustic velocity as the decoupling medium between the two FBARs. Fattinger et al. [32], used a tungsten layer sandwiched between to layers of oxide. The layers were not precisely quarter-wave, which would be the case for a true Bragg reflector. This perturbation from the quarter-wavelengths, allow for some weak coupling to exist between the bottom and top resonator. What wasn’t described in any detail in Chapter 2 is that there are materials that have sufficiently low acoustic impedance, that the two resonators can also be sufficiently decoupled. This is shown in Figure 5.27. A single coupling (or rather a decoupling) layer is shown. One family of coupling materials would be organic films that can be spun onto the wafer and then processed further. The acoustic impedance of some organic materials is around 2.8 MRayls and is ideal for decoupling the resonators. Certain oxides are also possible choices for the decoupling layer. A simple equation was derived that puts an upper bound on the acoustic impedance of the coupling material [33]. Z c ~< 2 * (%BW ) * Z AIN

(5.14)

For a percent bandwidth of 4%, an acoustic impedance of AlN of 35 MRayls, the coupling acoustic impedance should be equal or less than 2.8 MRayls. The derivation of (5.12) assumes infinitesimally thin electrodes. In the case of high acoustic

Electrode 4

Electrode 3

PAD 4

Electrode 2 Ground ring

Electrode 1 PAD 1

Figure 5.27 A CRF filter consisting of two stacked FBAR resonators separated by a coupling layer. In this configuration, electrodes 2 and 3 are tied together and grounded. The inset is a device with 2 an area of 0.4 × 0.6 mm .

150

FBAR Resonators and Filters

impedance electrodes such as Mo or W, the acoustic impedance of the coupling layer can be increased to something like 5 MRayls. [Note: (5.14) should be taken with a grain of salt. Higher coupling coefficients lead to poorer return loss and swag in the middle of the passband. So, there are other equations that govern the parameter space for a usable CRF device.] One can create a single-ended to single-ended filter with just one resonator (as shown in Figure 5.27) where the two middle electrodes are tied together and grounded. Floating the two top electrodes converts the device into an unbalanced single-ended to differential filter. Using an external inductor will help balance the differential output. The external inductor is designed to resonate out the large parasitic capacitance formed between metal 2 and metal 3 sandwiching the thin dielectric. The drawback to using organics as the coupling material is that they are relatively lossy, somewhere between 1,000 and 2,000 dB/cm. Figure 5.27 inset shows a picture of one incarnation of the CRF where the two middle electrodes are tied to ground. The performance (narrow band and wide band) are given in Figure 5.28. The device has a 4% 2-dB bandwidth at 2.4 GHz and a null near 2.1 GHz—useful for filter applications in the ISM band [34]. Figure 5.29 shows a single-ended to differential-ended filter and its response. Other single-ended to differential topologies are described in [35]. (Note: a filter-balun combination is a property much desired for high-frequency receivers.)

5.3

FBAR Filters 5.3.1

Interstage Filters

Even if the resonator is the heart of the filter or duplexer, by itself, it is not a filter. Multiple resonators must be connected in some fashion as to make a filter response

0 -10 -20 -30 -40

0

-50

m1

-2

m2

-60

m3 freq = 2.484 GHz dB (S(1,1)) = −10.328

-4 -6

-70 -80 -90 1.4

1.6

1.8

2.0

2.2

2.4

2.6

2.8

3.0

3.2

3.4

3.6

freq, GHz

-8 m3

-10 -12 -14 -16 -18 -20 2.34

2.36

2.38

2.40

2.42

2.44

2.46

2.48

2.50

2.52

2.54

2.56

2.58

2.60

freq, GHz

Figure 5.28 A measurement of the single-ended to single-ended device. The 2-dB bandwidth is 4.3%. The wideband performance is shown in the inset. For these devices a null is formed due to a parasitic capacitance between input and output electrodes.

5.3 FBAR Filters

151

50 Ohm 0 −1

Term Term14

m4

m5

−2 −3 −4

External inductor 1

−5

2 3

L L21

Ref

S3P SNP5

-6

0

-7

−10

-8 -9

Term Term13

50 Ohm

−20 2.40

−30

2.42

2.44

2.46

2.48

2.50

2.52

−40 −50 −60 −70 1.8

2.0

2.2

2.4

2.6

2.8

3.0

3.1

freq, GHz

Figure 5.29 A measurement of the single-ended to differential-ended device. The 2-dB bandwidth is 4.1%. The narrowband performance is shown in the inset. An external inductance is needed to balance the two output arms. This is shown in the circuit.

with a passband with some defined level of insertion loss (usually defined by the worst-case corners measured over temperature), an out-of-band rejection (and isolation for duplexers), and wideband specs. The most common topology for filters and duplexers is the half-ladder topology. Here, the series resonators are designed acoustically to have a desired frequency f, whereas the shunt resonators are acoustically designed to have a frequency f − Δf. The value of Δf is targeted to be about 1.5% to 2%. This has the effect of placing a zero above the passband (due to Rp of the series resonators) and a zero below the passband (due to Rs of the shunt resonators). The series resonance of the series resonators provides a pole in the passband, as does the antiresonance of the shunt resonators. To remove all ambiguity, I prefer the nomenclature where a pole “props” up the passband, and, zeros “nail” down the rejection. We use mass loading in the acoustic stack to achieve the Δf in frequency between the series and shunt resonators. This is done by adding a thin film on the order of a few hundred angstroms to just the shunt resonators. One is not restricted to only one mass loading. One can add more resonators with different mass loadings (at the cost of additional processing steps). Adding different frequency series resonators will widen the passband (additional poles). In order to avoid additional mass loadings in the shunts (more processing), inductors of different values in series with each of the shunt resonators are used. The Lext*Cp product for each of the shunt arms in a filter will define the frequency of the zeros in the rejection band. This can be understood as follows: Off resonance, the

152

FBAR Resonators and Filters

shunt resonator looks like a plate capacitor. It will series resonate with the external inductor creating a zero around 1/[Lext* Cp]1/2. Choosing different external inductors (and/or shunt resonator area) for each shunt resonator allows the designer to distribute zeros across the rejection band. If the zeros are too far apart, the rejection floor rises quickly. By spacing the zeros appropriately, the designer nails down the rejection floor across the critical frequency region. There is a price to pay for using external inductors on shunts. These inductors will create “flyback.” This is where the rejection floor will rise quickly and irretrievably as one goes to higher frequencies. This can be explained simply by pointing out that the inductor impedances continue to increase with increasing frequency. Hence, less energy will be diverted to ground and will instead, find itself passing through to the output. Alternatively, one can avoid using shunt external inductors and obtain a respectable filter response by optimizing the areas of the shunt resonators and adding additional stages. At the expense of insertion loss and steep skirts, one can design a filter with a much delayed flyback. Choosing the best design for a half-ladder filter is in the province of a talented filter designer. Depending on specs and, of course, the quality of the resonators, a good designer will choose an appropriate design. Figure 5.30 shows measurements of a single Tx “point” filter used in a large majority of CMDA phones today [36]. Figure 5.30 highlights the importance of having a reproducible process. One can see that there is indeed a spread in performance for these devices as measured across a wafer. The zeros are clearly seen outside the passband and the tightness of the passband insertion loss (worst-case corners) defines the “blood” specs. These are the specs that one MUST apply to each part as it is tested in the factory and—even if the part misses a spec by < 0.1 dB—the part must be rejected. 5.3.2

The Duplexer and Multiplexers

Unlike a filter, the duplexer is a much more complicated device to design. The duplexer is made up of two filters. However, the set of specifications increases more than two-fold. The reason is that not only does each filter have its own set of specifications, but, now there are the isolation specifications of the combined filter pair. Another way to understand this is that a single filter with input/output has three independent S parameters from which specifications are set for (namely, return loss S11, S22, and thru S12). For a duplexer, there are the specifications relating to the return loss for each port; S11, S22, and S33 (where S33 is the antenna port) but there are new specifications relating to S12 (isolation between Tx port and Rx port) as well as S13 and S23. In the filter, duplexer, and higher-order multiplexer, the engine is the resonator. Without controllable and repeatable kt2eff and very high unloaded Q-values, the performance of these types of devices is marginal. The specs that drive acceptance of FBAR into the market place are insertion loss, return loss at each port, and steep skirts resulting in high rejection and isolation close in to the passband. Figure 5.31 shows the progression of duplexer performance from 1999 to 2003. What is telling is the huge improvement in the first four years; from scientific oddity

5.3 FBAR Filters

153

0

−10

−20

−30

−40 B2247088 (S11) B2247080 (S11) B2477080 (S12) B2477088 (S12)

−50

−60 1.75×109

1.80×109

1.85×109

1.90×109

1.95×109

2.00×109

(a) 0 B2477080 B2477088

−10

−20

−30

−40 0

1×10 9

2×10 9

3×10 9

4×10 9

5×10 9

6×10 9

(b)

Figure 5.30 Filter performance from two separate wafers (400 units) of a PCS Tx interstage filter. Narrowband is shown in (a), and the wideband response is shown in (b). The vertical axes are in dB [36].

to commercial realization. Clearly, specs have not eased up at all in the intervening years. The latest device the ACMD-7302 addresses the new G-Block addition to the traditional 60-MHz PCS band. So, instead of only 20-MHz spacing between Tx and Rx (and defining the error budget that includes temperature drift, process variation, and roll-off), the new guard band is only 15 MHz.

Figure 5.31

0

−50

−40

−30

−20

−10

−1.0

Insertion Loss, dB

1.9

1.95×10 9

Frequency, GHz

1.90×10 9

Two Duplexers (mid 2001)

Frequency, GHz

1.8

2.00×10 9

2

Improved specs, but variation from part to part

1.85×10 9

1.7

First Published Duplexer, 1999

A snapshot of duplexer measurements from 1999 to 2003.

−6.0

−5.0

−4.0

−3.0

−2.0

Insertion Loss, dB

2.1

0

0

−50

−40

−30

−20

−10

−20

−10

Insertion Loss, dB −90

−80

−70

−60

−50

−40

−30

Insertion Loss, dB

0

1.88

1.92

1.94

1.96

1.95×10 9

Frequency, GHz

1.90×10 9

Duplexer Builds 1/03/2003

Frequency, GHz

1.9

1.98

2.00×10 9

Improved specs and part to part consistency

1.85×10 9

1.86

Barely meeting specs at the corners

158 Duplexer Builds 7/29/2000

2

154 FBAR Resonators and Filters

5.3 FBAR Filters

155

In a typical CDMA phone today, there are two bands (the cell band at 850 MHz and the PCS band at 1,900 MHz). The two sets of radio frequencies are fed into a diplexer (to minimize the loading of the one duplexer on the other duplexer) whose output is then fed into an antenna. With the requirement of E-911, all CDMA phones need an access port to GPS. The GPS requirement is quite stringent, since the signal from orbiting satellites are so weak, the insertion loss of the GPS filter must be quite low and because any interference will detune the GPS LNA, the rejection out of band must be very good. In fact, until recently, simultaneous GPS (S-GPS) and phone transmit were mutually exclusive. To make S-GPS, the handset manufacturer would have to make a separate transceiver path with separate antenna, filter, and LNA. Alternatively, a switch at the diplexer could be installed to switch from phone use to GPS use. The switch would add insertion loss to the receive channel and simultaneous GPS would not be available. One advantage FBAR resonators have (heretofore not discussed) is the Q-factor—away from resonance. Unlike SAW and SMR-BAW devices where the resonance is created by a Bragg reflector that only works over a narrow range of frequencies, the air/crystal interface of an FBAR resonator will not transmit sound

Figure 5.32 Model layout of Avago AMCD-7102 Quintplexer. This product allows for working GPS during normal phone calls. The two relatively large die on the left are the cell-band duplexer die and the two die on the right are the PCS duplexer die. The lower die in the center is the GPS filter. The three smaller die are chip capacitors.

156

FBAR Resonators and Filters

from the transducers into the substrate. Thus, far from resonance, the FBAR looks like a very high-quality capacitor. One can take advantage of this to design and build a quintplexer [37]. The “quint” eliminates the switch as well as the diplexer and connects all three radios (cell, PCS, and GPS) to one antenna. Thus, overall handset performance is improved and just as important simplified from a phone designer’s perspective. Figure 5.32 is a model of our quintplexer, the ACMD-7102 [37]. Every trace is mutually coupled to each other trace. These mutual capacitances and inductances must be taken into account so that the far-from-band specs are met. For example, the insertion loss of the GPS filter portion of the quintplexer is less than 0.9 dB, and the port isolation between the GPS port and the TX ports of the two transmit filters (cell and PCS) is better than 42. The large isolation is an important selling point for these multiplexers, all of the parasitic mutuals have been cancelled out such that isolation can be met. Figure 5.33 is a side-by-side comparison of our first duplexer product introduced in 2001 with our quintplexer product introduced in 2007. The performance improvements are quite substantial, the functionality and content of the quintplexer is significantly greater than our first duplexer, the size is smaller—and yet, the price is less than half as much as our first product. Such is the speed of progress and effect of competition on our business.

5.4

Conclusions There are five market forces driving the filter needs in the cell phone business. These are: 1. CDMA service providers are demanding that the talk time equal or exceed that of the GSM service provider’s phones, and at the same time, adding power draining features. However, a countertrend has been the consumer’s Then and Now

PCS Duplexer Tx /Rx insertion loss ∼3.8 dB/−4.5 dB Return loss −8 dB Isolation Tx / Rx ∼ −50 dB/ -40dB Volume 144 mm3 two LCC Packages

PCS Duplexer + Cell Duplexer + GPS Tx /Rx insertion loss ∼2.0 dB/−2.5 dB Return loss −11 dB Isolation Tx / Rx ∼ −57 dB/−45 dB Volume 28 mm3 –five Microcaped Filters Price ∼ ½ of the 2001 PCS Duplexer

Figure 5.33

First 6 × 11 mm FBAR duplexer versus the 4 × 7 mm Quintplexer. 2

2

5.4 Conclusions

2.

3.

4.

5.

157

quick adoption of slim phones, which demands a thin battery with limited mA-hour capacity. This places pressure on the duplexer to reduce insertion loss in the Tx band, a major source of power drain for cellphones. Service providers are demanding large cost reductions and more functionality from their handset vendor, while simultaneously tightening specifications—including noise sensitivity and reduction of spurious emissions—both specifications that affect the phone’s ability to meet and improve quality of service (QoS). As the base-band chip and the various transceiver chips migrate from relatively expensive SiGe technology to the cheaper and higher density RF CMOS technology, the quality of the CMOS gates degrades with size and in particular, the CMOS LNAs have poorer noise figure. This puts pressure on the duplexer vendors to improve its Rx insertion loss and rejection in the Tx band, as well as far-from-band rejection. The move to internal antennas has degraded the sensitivity of both Tx and Rx in handsets—again, pushing the duplexer to have better insertion loss specs. The creation of EGSM in Europe and more recently the creation of the G-block spectrum in the United States has started a trend of taking bandwidth away from the guard band between Tx and Rx spectrum and giving this spectrum to the user. This means that the duplexer will have to both increase kt2eff to meet the needs of a wider bandwidth and increase Q to meet the need for steeper skirts—in filter speak, the resonators used must have higher Rps and lower Rss to meet the new requirements. The push to have more radios with coverage in more bands means the phone must have more antennas and/or more switches. Large number of throws on a switch (that already has very tough insertion loss specs and linearity specs) is very difficult and there is a limit to the number of antennas one can put into the phone—both for size reasons and due to increasing cross-talk between antennas.

Furthermore, as the airwaves become more congested and interferers proliferate, the requirements on the filter performance is constantly increasing. Now, specifications include values on the amount of nonlinearity (specifically secondharmonic generation due to the inherent nonlinearity of the piezoelectric film) and intermodulation-distortion (IMD2 and IMD3) as well as sufficient rejection at higher frequencies, specifically, harmonics of the fundamental power going through the PA. This same congestion will, over time, force carriers to adopt higher carrier frequencies. Applications such as WiMAX, LTE (long-term evolution of CDMA), satellite radio, military applications—all operating at high frequencies—will utilize BAW devices. A recent paper by Fujitsu demonstrated the first competent-looking filter at 10 GHz using FBAR resonators [38]. The main advantage of BAW (FBAR or SMR-BAW) devices over their SAW counterparts at higher frequencies is their relative imperviousness to ESD and power handling. At the higher frequencies, the spacing between the fingers of a SAW transducer becomes smaller, control on the actual

158

FBAR Resonators and Filters

widths of the lines and spaces become more difficult and losses due to the thinner fingers begin to take their toll on performance. Between 1993 and 2003, the FBAR project at Avago probably faced project termination on five occasions. There were probably another 10 to 20 times when FBAR hit what seemed to be a fundamental obstacle in its ability to contribute and compete in the commercial arena. Even today, one of the greatest challenges FBAR technology faces is the ability to get costs down, while continuing to improve performance (Q and kt2eff). So, work on die shrink, yield improvements and new markets help amortize the overhead and match the never-ending price erosion.

References [1] Grudkowski, T., et al., “Fundamental-Mode VHF/UHF Miniature Acoustic Resonators and Filters on Silicon,” Applied Physics Lett., Vol. 37, 1980, pp. 993–995. [2] Nakamura, K., H. Sasaki, and H. Shimizu, “A Piezoelectric Composite Resonator Consisting of a ZnO Film on an Anisotropically Etched Silicon Substrate,” Proc. of 1st Symp. on Ultrasonic Electronics, Tokyo, 1980. [3] Lakin, K. M., et al., “Thin Film Resonators and Filters,” Ultrasonics Symp., 1982, pp. 466–475. [4] Driscoll, M., et al., “Recent Advances in Monolithic Film Resonator Technology,” Ultrasonics Symp., 1986, pp. 365–369. [5] Vale. C., et al., “FBAR Filters at GHz Frequencies,” 44th Frequency Control Symp., 1990, pp. 332–336. [6] Rosenbaum, J. F., Bulk Acoustic Wave Theory and Devices, Norwood, MA: Artech House, 1988, p. 180. [7] Wang, J., and K. Lakin, “Sputtered AlN Films for Bulk Acoustic Wave Devices,” Ultrasonics Symp., 1981, pp 502–505. [8] Ruby, R., and P. Merchant, “Method of making Tunable Thin Film Acoustic Resonators,” U.S. Patent # 5,873,153, filed August 1996. [9] Ruby, R., and P. Merchant, “Micromachine Thin Film Bulk Acoustic Resonators,” 48th Frequency Control Symp.,” 1994, pp. 135–138. [10] Larson, J., R. Ruby, P. Bradley, “Bulk Acoustic Wave Resonator with Improved Lateral Mode Suppression,” # 6215375, filed July 1999. [11] Ruby, R., et al., “High-Q FBAR Filters in a Wafer-Level Chip-Scale Package,” International Solid-State Circuits Conference, 2002, pp. 184–185. [12] Ruby, R., and L. Kekoa, “Wafer Level Packaging (WLP) of FBAR Filters,” 3rd International Symp. on Acoustic Wave Devices for Future Mobile Communications Systems, 2007, pp. 101–103. [13] Lakin, K. M., K. T. McCarron, and R. E. Rose, “Solidly Mounted Resonators and Filters,” IEEE Ultrasonics Symp., 1995, pp. 905–908. [14] http://jp.fujitsu.com/group/fmd/en/release/2000/20000106.html. [15] http://jp.fujitsu.com/group/fmd/en/release/2002/20021119.html. [16] ITRI Conference, Taiwan, July 2002. [17] “Advanced Design System” (“ADS”), EESOF, a Division of Agilent Technologies. [18] Feld, D., R. Parker, and R. Ruby, “After 60 Years: A New Way for Computing Q is Warranted,” Ultrasonics Symp., 2008. [19] Ruby, R., R. Parker, and D. Feld, “Method of Extracting Q Applied Across Different Resonator Technologies,” Ultrasonics Symp., 2008.

5.4 Conclusions

159

[20] Bode, H. W., Network Analysis and Feedback Amplifier Design, Chapter 10, New York: Van Nostrand, 1945, pp. 196–225. [21] Auld, B., Acoustic Fields & Waves in Solids, 2nd ed., Vol. II, Chapter 9, Malabar, FL: Kreiger Publishing, 1990. [22] Telschow, K., et al., “UHF Acoustic Microscopic Imaging of Resonator Motion,” Ultrasonics Symp., 2000, pp. 631–634. [23] Ruby, R., et al., “The Effect of Perimeter Geometry on FBAR Resonator Electrical Performance” International Microwave Symp. Digest, 2005. [24] Kaitila, J., “Review of Wave Propagation in BAW Thin Film Devices—Progress and Prospects,” Ultrasonics Symp., 2007, pp. 120–129. [25] Ruby, R., “Review and Comparison of Bulk Acoustic Wave FBAR, SMR Technology,” Ultrasonics Symp. ,2007, pp. 1029–1040. [26] Feng, H., et al., “Thin Film Bulk Acoustic Resonator with a Mass Loaded Perimeter,” #7,280,007, filed November 2004. [27] Ruby, R., “Thin Film Bulk Acoustic Resonator with a Mass Loaded Perimeter,” App. 20060071736, filed June 2005. [28] Thalhammer, R., “Spurious Mode Suppression in BAW Resonators,” Ultrasonics Symp., 2006, pp. 456–459. [29] Kaitila, J., “XXX,” Ultrasonics Symp. Proc., 2007. [30] Lakin, K., K. McCarron, and J. McDonald, “Temperature Compensated Bulk Acoustic Thin Film Resonators,” Ultrasonics Symp., 2000, pp. 855–858. [31] Pang, W., “A Zero Drift Colpitts Oscillator Using Temperature Compensated FBAR,” IEEE Ultrasonics, 2007. [32] Vectron TCXO Oscillator Sheet: http://www.vectron.com/products/tcxo/vtm3.pdf. [33] Fattinger, G., R. Aigner, and W. Nessler, “Coupled Bulk Acoustic Wave Resonator Filters: Key Technology for Single-to-Balanced RF Filters,” 2004 IEEE MTT-S Microwave Symposium Digest, Vol. 2, June 2004, pp. 927–929. [34] Jamneala, T., et al., “Coupled Resonator Filter with Single-Layer Acousic Coupler,” submitted to the IEEE Trans. on Ultrasonics, Ferroelectrics, and Frequency Control, November 2007. [35] Small, M., et al., “A De-Coupled Stacked Bulk Acoustic Resonator (DSBAR) Filter with 2 dB Bandwidth > 4%,” Ultrasonics Symp., 2007, pp. 604–607. [36] Gilbert, S., et al., “An Ultra-Miniature, Low Cost Single Ended to Differential Filter for ISM Band Applications” Submitted and accepted to the International Microwave Symposium in 2008 under Focused Sessions. [37] Feld, D., et al., “A Wafer Level Encapsulated FBAR Chip Molded into a 2.0 mm × 1.6 mm Plastic Package for Use as a PCS Full Band Tx Filter,” Ultrasonics Symp., 2003, pp. 1798–1801. [38] Bradley, P., et al., “A 6-Port Film Bulk Acoustic Resonator (FBAR) Multiplexer for U.S. CDMA Handsets Permitting use of PCS, Cellband, and GPS with a Single Antenna,” Ultrasonics Symp., 2006, pp. 325–328.

CHAPTER 6

Comparison with SAW Devices Masanori Ueda

6.1

Introduction In the wireless communication market, a huge amount of filters and duplexers are utilized as RF components, and SAW and BAW devices are playing a very important role in this market. This chapter describes comparisons between BAW, in particular FBAR, and SAW technologies. As these technologies have merits and demerits in terms of performance, device size, productivity, cost, and so on, it is important to consider which technology is better for the target from the user side or the device development side. Section 4.3, in Chapter 4, also discusses the comparison of these technologies, including SMR technology, and an advantage of SMR which comes from its unique structure are introduced there.

6.2

Structural Comparison and Features Figure 6.1 compares the structures and features of FBAR and SAW [1]. FBAR and SAW have different types of structure and vibration mode. For example, FBAR, which utilizes c-axis orientated AlN or ZnO, vibrates in thickness extension (TE) mode, whereas 42°Y LiTaO3 as a SAW substrate, which is widely used in mobile applications, vibrates in a shear horizontal (SH) mode. Center frequencies of FBAR and SAW are basically determined by film thickness and an electrode pitch of the interdigital transducer (IDT), respectively. These resonators work as impedance elements although they have quite different structures and they can be described as the same equivalent circuit shown in Figure 6.1. A piezoelectric thin film resonator filter on a Si substrate has the following advantages over a SAW filter: 1. Higher Q-factor, particularly in the high-frequency range (above 2 GHz); 2. High-power durability and high resistance to electrostatic discharge (ESD); 3. The possibility of realizing monolithic devices with active RF devices. On the other hand, SAW devices have features which can make them preferable for low-frequency applications. Self evident that they can be used as balun (unbalanced-to-balance transformer) and in addition, the simple layer composition

161

162

Comparison with SAW Devices Piezoelectric thin film Top electrode Bottom electrode Standing wave of BAW

hT Cavity

Si

Si

Piezoelectric thin film resonator (FBAR type)

Co

Standing wave of SAW

L1 C1 R1

One-port resonator

Grating reflectors Comb electrodes λ

Piezoelectric single crystal SAW resonator

Figure 6.1

FBAR and SAW structures.

promotes a cost-effective production process. The features of SAW devices are summarized here: 1. Suitable for low-frequency applications (below 2 GHz); 2. Suitable structure for an unbalance-to-balance or balance-to-unbalance conversion; 3. Few process steps and no need for a special frequency control equipment (low cost).

6.3

Resonator Performance and Reliability 6.3.1

Q-Factor

To obtain a high-resonant frequency of a SAW resonator, a narrow electrode pitch and finger width by using a fine photolithography techniques such as i-line, excimer stepper, or electron beam (EB) exposure system are required. However, this increases the electrical resistance of the electrodes and deteriorates the Q-factor. Figure 6.2 compares the 5-GHz FBAR and 5-GHz SAW layout, which use the same ladder-type design method, where each filter consists of seven resonators in total. The 5-GHz SAW needs a very fine and thin electrode, with a finger width of 0.18 μm and a film thickness of 60 nm. It is very difficult for SAW devices to maintain good performance in the high-frequency range because of the high electrode resistance. Figure 6.3 shows the Smith chart of 5-GHz resonators, where the bold line is the

6.3 Resonator Performance and Reliability

163

SAW filter

FBAR filter 1.4 mm

7 SAW resonators

7 FBARs 0.9 mm

0.9 mm Al-Cu IDT Patterns

mm 0.18 0.18 μm

Figure 6.2

Finger electrode with 180 nm width and 60 nm thickness

Topologies of 5-GHz FBAR and 5-GHz SAW filter.

FBAR SAW

Figure 6.3

Smith chart of 5-GHz FBAR and 5-GHz SAW resonator (S11).

performance of FBAR and the solid line is that of the SAW resonator. As can be seen, the loss factor of the SAW resonator is higher than that of the FBAR, which is mainly the result of the high electrode resistance. The resistance of the electrodes also has an impact on power durability. Calculated passband performances utilizing resonators with various Q-factors are shown in Figure 6.4(a). In this simulation, the quality factors resonate of antiresonant frequency Qa and frequency Qr are assumed to be equal; furthermore C0/Cm, C0: clamped capacitance, Cm: motional capacithe capacitance ratio ( tance) is assumed to be 16. The filter has a ladder configuration composed of four basic sections, in which the capacitance ratio Cp/Cs is 0.35 (Cp: static capacitance of parallel arm resonator, Cs: static capacitance of series arm resonator), and center frequency is 1.9-GHz. Figure 6.4(b) shows the relationship between Qs and insertion losses at passband edges of filters. From these figures, it is obvious that Q gives a strong impact to the filter performance, and in addition the filter loss is reduced about 0.5 dB by improving Q from 500 to 1,000.

164

Comparison with SAW Devices

γ

(a)

(b)

Figure 6.4 Q-factor on a typical ladder-type filter: (a) passband performances, and (b) insertion loss of filter.

Examples for the achievable Q-factors of FBAR, SMR, and SAW resonators derived from a number of studies [1–8] can be seen in Table 6.1. According to these data, it is obvious that FBAR and SMR indicate higher Q. The energy dissipation and therefore the resonator Q factor is dependent upon structure, design, applied materials, and so forth; it is one of the resonator key factors determining not only filter losses, but also the steepness of the passband border region. The electromechanical coupling factor (K2) which is also a key factor determines the filter bandwidth, and the fractional bandwidth is known empirically to be approximately half of the coupling factor. Both Q and the coupling factor should be discussed simultaneously; with resonators having insufficient coupling factors it is hard to realize a low loss and wide passband filter (like it is worldwide commercially applied in mobile phones) even if the resonators provide an excellent Q-factor. Therefore, in some cases the product of coupling and Q-factor is discussed as the resonator figure of merit (FOM) (see Section 5.2.1).

6.3 Resonator Performance and Reliability Table 6.1

165

Examples of Reported Q-Factors Qa

Ref. No.

Technology Frequency

Qr

SAW

0.8 GHz

841

[2]

SAW

1.9 GHz (PCS)

320

1,080 [3]

SAW

5 GHz (WLAN)

192

256

FBAR

0.9 GHz (Band 8) 2,500 3,200 [8]

FBAR

1.9 GHz (PCS)

2,780 2,880 [8]

FBAR

2 GHz (Band 1)

1,500 1,100 [4]

SMR

1.9 GHz

900

SMR

1.8 GHz

1,400

FBAR

9.1 GHz

239

[1]

2,500 [6] [7] 488

[5]

Qr: Q at resonant frequency Qa: Q at antiresonant frequency

6.3.2

Power Durability

A comparison of the power durability between SAW and FBAR 5-GHz filters is shown in Figure 6.5. In this figure, horizontal and vertical axis indicate input power to filters and the lifetime evaluated by the degradation of filter loss. Estimated power durability of the SAW filter is about 15 mW; however, the FBAR indicates 800 mW for an extrapolated lifetime of 100,000 hours. There are mainly two reasons why FBAR devices have a better power durability, namely the already mentioned higher electrode resistance of SAW devices and the different vibration mode. The SAW fabricated on 42oY-X LiTaO3 excites a shear-horizontal mode, whereas FBAR vibration mode is thickness extensional. It is considered that the stress in the finger electrode of the SAW is much higher than that in the plane electrode of the FBAR. In the SH-mode SAW, a large stress is applied to narrow fingers in a high-power conditions. As a result, stress migration occurs, and in addition, the high resistance of the electrode causes Joule heat, which accelerates migration. On the other hand, for the FBAR using TE mode, the stress on the electrode is not as

7

10

6

10

Lifetime (hrs)

5

10

4

10

3

10 10

2

10 1 −1

10

0.01

0.1

1 Input power (W)

Figure 6.5

Power durability of 5-GHz FBAR and SAW filters.

10

166

Comparison with SAW Devices

severe as compared to the SH-mode SAW. Although only 5-GHz filters are discussed, 2-GHz filters, which are widely used commercially, show the same tendency, and long-term power durability of the 2-GHz filter is up to 4W (refer to Section 4.3). From these investigations, FBAR duplexers are confirmed to be highly reliable for RF front-end devices.

6.4

Filter Design Ladder [9] and lattice structures, as shown in Figure 6.6(b, c), are well-known methods for designing single-ended and balance-to-balance filters. Either structure is applicable for both FBAR and SAW filter design, in particular, ladder filters are widely used for duplexers in RF front-end. Unbalance-to-balance or balanceto-unbalance filters are widely utilized as interstage filters, and recently as front-end duplexers for CDMA and WCDMA systems, which require balanced output for the Rx side. The double-mode SAW (DMS) design shown in Figure 6.6(a) is a flexible structure and is applicable not only for single-ended and balance-to-balance, but can also convert an unbalance-to-balance or balance-to-unbalance signal easily, and simultaneously change the input to output impedance ratio. A DMS has a balun transformer function as well as a filter function. DMSs using three ITDs or two IDTs are widely used for unbalance-to-balance and balance-to-unbalance conversion filters. Figure 6.7 shows the drawing of a three IDT-type DMSs for EGM Rx and its fil-

(b)

(a) (c)

Figure 6.6 design.

Filter design methods: (a) DMS design, (b) ladder-type design, and (c) lattice-type

167

Attenuation [dB]

6.4 Filter Design

Frequency [MHz] (b)

Amplitude balance [dB]

Phase balanced [degree]

(a)

Frequency [MHz] (c)

Frequency [MHz] (d)

Figure 6.7 DMS filter configuration and performance: (a) DMS structure, (b) passband performance, (c) amplitude balance, and (d) phase balance.

ter performances. In Figure 6.7(a), two DMSs are connected to achieve better stop-band rejection, and upper and lower DMSs are an input port and an output port, respectively. By connecting one of the terminals of the output or input ports to ground, an unbalanced port is obtained. In this design, an impedance conversion such as 50–100Ω and 50–200Ω is possible by a design modification. Excellent passband and balance performance are achieved as shown in Figure 6.7(b–d) (Fujitsu Media Devices Limited, P/N FAR-F5KB-942M50-B4EB). In BAW technology, a filter with a balun function like DMS is proposed. Figure 6.8 shows a proposed balanced filter from Infineon Technologies, which is based on a SMR structure [10]. Input and output transducers are acoustically connected by the underneath piezoelectric layer inside the acoustic mirror; this balance filter enables balance-to-unbalance and also unbalance-to-balance conversion. EPCOS proposed a WCDMA band two-duplexer which has a single-ended BAW filter for Tx and a balance output SAW for Rx filter [11]. It is a good approach to achieve current requirements.

168

Comparison with SAW Devices

Figure 6.8

6.5

Proposed BAW balance filter by Infineon Technologies.

Manufacturing Process In SAW devices, the center frequency of a filter is basically determined by the electrode pitch of the photomask; however, the center frequency of FBAR is determined by film thickness and highly-accurate tuning is needed (as discussed in Section 4.2). To achieve precise frequency adjustment and better production yield for FBAR devices, it is necessary to strictly control the film thicknesses of electrode, piezoelectric, and mass-loading films within one wafer, and also from wafer to wafer. SAW has a much simpler layer structure than FBAR. For the FBAR, at least three layers for two electrodes and a piezoelectric layer are needed, which implies a minimum of four masks for photolithography to process these layers and to form the cavity below the resonating area. On the contrary, only one layer and one mask can be sufficient for the manufacturing of SAW devices. This directly affects the cost, and generally SAW filters are therefore less expensive compared to FBAR filters. The SMR technology seems to be more complicated than FBAR due to the multilayer acoustic mirror is needed. For assembly, almost the same processes are applicable for both devices and flip-chip technologies are commonly used. Recently, wafer-level package (WLP) technologies have been developed and applied to devices on commercial base.

6.6

Temperature Compensation Technique Temperature compensated SAW technologies have been developed since the 1980s [12]. Recently, WCDMA band 1 and band 2 (US-PCS) SAW duplexers have been developed using temperature compensation techinques [13–15] and therefore are competitors to FBAR devices in the today’s market. The temperature coefficient of frequency (TCF) is determined by the thermal expansion coefficient (TEC) and the temperature coefficient of velocity (TCV) by means of TCF = −TEC + TCV

(6.1)

6.7 Application Map

169

LiTaO3 and LiNbO3 substrates normally have positive TEC and negative TCV, e.g., −30 to −40 ppm/°C for the TCF can be obtained in a 42Y LiTaO3 SAW substrate. There are mainly two countermeasures to compensate the TCF, one is to suppress the TEC of the piezoelectric substrate, the other is to compensate the TCV. Fujitsu developed the US-PCS duplexer [13] with temperature-compensated LiTaO3/ sapphire-bonded substrate shown in Figure 6.9(a). Sapphire has a high Young’s modulus and a low TEC. By bonding sapphire to LiTaO3 by employing surface-activated bonding technology and thinning the LiTaO3 substrate to several tens of a micrometer, the TEC is improved. As a result of this bonded substrate, the SAW substrate TCF is improved from −30 to −15 ppm/°C. There are two big advantages in this technology, no degradation of propagation loss and heat-sink effect of sapphire. Thus, low loss and power durable duplexers can be realized. Murata also developed temperature-compensated US-PCS and WCDMA band 1 duplexers using a SiO2/heavy electrode/LiNbO3 configuration as shown in Figure 6.9(b). The TCV of SiO2 indicates positive value, therefore the filter TCF can be controlled from negative to positive value by adding SiO2. The TCF of WCDMA band 1 duplexer is successfully improved from −80 to −10 ppm/°C [14]. In this technology, rotation angle of substrate, SiO2 thickness, a selection of electrode material, and flattening of the SiO2 surface above IDT are key parameters. Both companies are producing duplexers on a commercial base. Temperature-compensated FBARs are discussed in Section 5.2.7.

6.7

Application Map In mobile communication systems, FBAR duplexers and filters are commercially manufactured for 800-MHz to 5-GHz applications [16–20], such as the cellular IDT IDT (heavy electrode)

LiTaO3

Sapphire

SiO2

<enlargement> LiNbO 3 or LiTaO3

LiTaO3 Amorphous layer

5 nm

Sapphire (a)

(b)

Figure 6.9 Temperature-compensated SAW substrate: (a) bonded-wafer type, and (b) SiO2/IDT/piezoelectric substrate type.

170

Comparison with SAW Devices

band, GPS, WCDMA band 1, band 2 (PCS band), and 5-GHz WLAN system. Practically, FBAR duplexers for US-PCS seems to be in a strong position now. SAW interstage filters successfully cover GSM, CDMA, and WCDMA markets up to the 2-GHz range, and duplexers for WCDMA band 5 (cellular band) and other 800–900 MHz bands are dominated by SAW. In these low-frequency ranges, SAW devices are in good position on performance, cost, productivity, and so on. In the frequency range of WCDMA band 1 and 2, temperature-compensated SAW technologies are applied for this market and competing FBAR technology. In addition, temperature-compensated SAW having a high coupling factor seems to be appropriate for a band 3 duplexer because a system needs a wide passband (relative bandwidth 4.3%), which is hard for AIN based BAW devices, and a very narrow gardband between Tx and Rx (20 MHz). FBAR devices, especially duplexers, are advantageous for high-frequency and high-power applications over the frequency range of WCDMA band 2. In higher frequency ranges, such as WCDMA band 7 (2.5 GHz), WiMAX and four generation system (4G), which is under discussion, BAW technology will play an important role. Section 4.3 also summarizes the application map for SAW and BAW technologies. Both BAW and SAW technologies have been discussed in this chapter. A perfect device which can cover every specification does not exist; therefore, manufactures and users need to consider which technology is better for their target use.

References [1] Ueda, M., and Y. Satoh, “FBAR and SAW Technologies and Their Applications for Mobile Communications,” Asia-Pacific Microwave Conference Workshops & Short Courses Digest, 2006, p. 426. [2] Inoue, S., et al., “Ultra-Steep Cut-Off Filters Using High-Q SAW Resonators with Suppressed Side Radiation,” 32nd EM Symp., 2003, pp. 95–98 (in Japanese). [3] Nakao, T., et al., “Smaller Surface Acoustic Wave Duplexer for US Personal Communication Service Having Good Temperature Characteristics,” Japan. J. Apl. Phs., Vol. 46, 2007, pp. 4760–4763. [4] Taniguchi, S., et al., “An Air-Gap Type FBAR Filter Fabricated Using a Thin Sacrifice Layer on a Substrate,” Proc. IEEE Ultrasonics Symp., 2007, p. 600. [5] Hara, M., et al., “X-Band Filters Utilizing AlN Thin Film Bulk Acoustic Resonators,” Proc. IEEE Ultrasonics Symp, 2007, p. 1152. [6] Fattinger, G., R. Aigner, and S. Marksteiner, “Everything You Always Wanted to Know About BAW,” Asia-Pacific Microwave Conference Workshops & Short Courses Digest, 2006, p. 408. [7] Timme, H. J., and R. Aigner, “Bulk Acoustic Wave Filters for Mobile Cellular Communications,” IEEE Intl. Microwave Symp. Workshop and Tutorial Notes, 2005, WSC. [8] Ruby, R., “Review and Comparison of Bulk Acoustic Wave FBAR, SMR Technology,” Proc. IEEE Ultrasonics Symp, 2007, p. 1029. [9] Ikata, O., et al., “Development of Low-Loss Band-Pass Filters Using SAW Resonators for Portable Telephones,” Proc. IEEE Ultrasonics Symp., 1992, pp. 111–115. [10] Fattinger, G. G., R. Aigner, and W. Nessler, “Coupled Bulk Acoustic Wave Resonator Filters: Key Technology for Single-to-Balanced RF Filters,” IEEE Intl. Microwave Symp. Technical Digest, 2004, p. 927.

6.7 Application Map

171

[11] Marksteiner, S., et al., “Hybrid SAW/BAW System-in-Package Integration for Mode-Converting Duplexers,” 3rd Int. Symp. on Acoustic Wave Devices for Future Mobile Communication System, 2007, p. 97. [12] Yamanouchi, K., and S. Hayama, “SAW Properties of SiO2/128 Y-X LiNbO3 Structure Fabricated by Magnetron Sputtering Technique,” IEEE Trans. Sonics and Ultrasonics Symp., Vol. SU-31, 1984, p. 51. [13] Miura, M., et al., “Temperature Compensated LiTaO3/Sapphire Saw Substrate for High Power Applications,” Proc. IEEE Ultrasonics Symp., 2005, p. 573–576. [14] Kadota, M., et al., “Small Surface Acoustic Wave Duplexer for Wide-Band Code-Division Multiple Access Full-Band System Having Good Temperature Characteristics,” Japan. J. Appl. Phys., Vol. 46, 2007, p. 4714. [15] Nakao, T., et al., “Smaller Acoustic Wave Duplexer for US Personal Communication Service Having Good Temperature Characteristics,” Japan. J. Appl. Phys., Vol. 46, 2007, p. 4760. [16] Ruby, R., “Overview of FBAR Filters, Duplexers, Quintplexers, and Front End Modules (FEM) at Avago,” Asia-Pacific Microwave Conference Workshops & Short Courses Digest, 2006, WS12-1. [17] Elbrecht, L., et al., “Integration of Bulk Acoustic Wave Filters: Concepts and Trends,” IEEE Intl. Microwave Symp. Technical Digest, 2004, p. 395. [18] Handtmann, M., et al., “Bulk Acoustic Filters for GPS with Extreme Stopband Attenuation,” IEEE Intl. Microwave Symp. Technical Digest, 2004, p. 371. [19] Nishihara, T., et al., “High Performance and Miniature Thin Film Bulk Acoustic Wave Filters for 5 GHz,” Proc. IEEE Ultrasonics Symp., 2002, p. 969. [20] Tsutsumi, J., et al., “Extremely Low-Loss SAW Filters and Its Application to Antenna Duplexer for the 1.9 GHz PCS Full-Band,” Proc. IEEE Frequency Control Symp., 2003, p. 861.

CHAPTER 7

Thin Films Deposition for BAW Devices Sergey Mishin and Yury Oshmyansky

7.1

Most Commonly Used Piezoelectric Materials In the development of microelectromechanical systems (MEMS) most widely used materials are zinc oxide (ZnO), lead zirconate titanate (PZT), and aluminum nitride (AlN). Some of the physical properties reported in the literature [1–3] and measured experimentally are reported in the Table 7.1. 7.1.1

Zinc Oxide

Zinc oxide has bean used in MEMS for many years. It has a relatively high coupling coefficient and is easily deposited by RF sputtering. But, as devices moved to higher frequencies it became apparent that high acoustic loses at higher frequencies limited its applications. With Q below 50, it could not be used for many high-frequency filters. Some other limitations have also come up as people started looking at producing high-volume low-cost RF filters. High TCF makes it difficult to use in many transmit filter applications. Many people wanted to manufacture filters either in CMOS facilities or integrate them into CMOS wafers. Since Zn decreases minority carrier lifetimes in silicon, it is not allowed in CMOS facilities. ZnO quickly absorbs moisture so a lot of care has to be taken in making sure that ZnO doesn’t sit around for very long before being encapsulated by another material. As aluminum nitride became more easily manufactured, ZnO became less and less popular. 7.1.2

PZT

PZT structure is shown in Figure 7.1(a) [4]. It has some very desirable characteristics: significantly higher coupling coefficient than either ZnO or AlN, a large zero bias permittivity, both piezoelectric and ferroelectric properties are exhibited, and very thick film thicknesses are viable. Due to its high acoustic loses at high frequencies, it is mostly used for low-frequency BAW devices or applications that don’t require high Q.

173

174

Thin Films Deposition for BAW Devices Table 7.1

Physical Properties of Selected Piezoelectric Films

Material

AlN (0001) ZnO

PZT

Units

Acoustic Velocity

11.05

6.35

4.6

3 (10 m/s)

Young’s Modulus

3.94

2.11

0.49

(10 N/m )

11

-12

2

Piezoelectric Coefficient (d33)

5.6

12.4

200

(10

Electromechanical Coupling (k33)

31

48

72

(%)

Thermal Conductance

280

60

1.80

(W/mK)

4.15

2.92

—

(10 /K)

6.5

8.5

35

(%)

Temperature Coefficient of Frequency (TCF) −25

−60

—

(ppm/ C)

Thermal Expansion (300K) Piezoelectric Coupling Coefficient Kt

2

C/N)

-6

o

(Pb,Zr)TiO 3 (PZT) Pb 2 4+

Ti , Zr

4+

O

2−

(a)

(b)

Figure 7.1

7.1.3

(a) PZT structure, and (b) aluminum nitride wurtzite (B4) structure.

Aluminum Nitride

Aluminum nitride was always considered a highly desirable BAW material. Having high velocity of sound, high Young’s modulus, high thermal conductivity, and relatively low TCF made it highly attractive. It was clear that if it could be produced with relatively high coupling coefficient, it could open markets for high-frequency filters that require high Q that have been dominated by either bulky ceramics or fairly pricy SAW filters. In the late 1960s it was demonstrated that AlN could be

7.2 Methods of Deposition of Piezoelectric Films

175

deposited with reasonable coupling coefficient, but it was not until late 1990s that high-volume production-worthy techniques started being available. At the present time several companies offer methods of AlN deposition that are commercially available, making it a material of choice for high-frequency BAW applications. Stoichiometric AlN is a very stable compound with a strong covalent and ionic bonds. The bonding energy of this structure is about 11.5 eV. It is also a semiconducting material and has a band gap of about 6.2 eV. Aluminum nitride crystal structure is shown in Figure 7.1(b) with positional relation between Al (large balls) and nitrogen (small balls) atoms. As with many other piezoelectric materials, it shows wurtzite crystalline structure with hexagonal symmetry.

7.2

Methods of Deposition of Piezoelectric Films 7.2.1

Sputtering

Sputtering is widely used method for depositing piezoelectric films. There are three main methods that are commonly used in the industry: RF diode sputtering, AC magnetron sputtering, and pulsed DC (bipolar) magnetron sputtering. Basic principles of sputtering are described in many books since sputtering is a mature technology. Thin Film Processes II [5] provides a good description of basic principles involved in most depositions methods described in this chapter. We will focus on the important aspects of sputtering as applied to the piezoelectric deposition. Methods for depositing AlN and ZnO have a lot in common. Although ZnO was developed earlier, much of the current development has been focused on AlN. For this reason, we will concentrate on the AlN next. One important requirement of the sputtered AlN is high coupling coefficient. Most research indicates empirical relation between grain structure of the deposited film and its coupling coefficient. Several studies looked at the cross-section of the film (see Figure 7.2) and correlated crystallinity of the film to the coupling coefficient [6–9]. Such techniques are only good for qualitative studies because it is up to an individual to decide the level of the grain orientation “goodness.” Generally, it is used to demonstrate that extremely poor films have disoriented grain structure and excellent films have fairly uniform columnar grain growth. A more quantitative and nondestructive approach is to look at XRD “rocking curves.” By using this technique, an entire wafer with deposited AlN film can be measured and further processed if necessary. Typically full wave-half maximum (FWHM) values are used (see Figure 7.3). Highly oriented AlN film with (002) orientation shows strong correlation with x-ray diffraction. As aluminum nitride thickness increases, width of the XRD rocking curve doesn’t increase, and maximum increases, resulting in lower number for the thicker aluminum nitride films. Reasonable correlations between rocking curve and coupling coefficients have been shown (see Figure 7.4). This technique can give a quick measure of the film quality without having to make a fully functioning resonator. It is important to notice that the most repeatable results and accurate measurement are obtained on silicon. When rocking curves are used to check AlN on different materials it is important to have a baseline of measurements on the particular film in question in order to be able to make con-

176

Thin Films Deposition for BAW Devices

Figure 7.2

SEM cross-section of AlN film, showing column structure.

AlN XRD Rocking Curve AlN on Si 300000

250000

Thickness (A)

FWHM (degrees)

20,000 15,000 10,000 5,000

1.21 1.32 1.66 2.08

Intensity (cps)

200000

150000

100000

50000

0 15

16

17 2.0 μm

Figure 7.3

18 Omega (degrees) 1.5 μm

1.0 μm

19

20

21

0.5 μm

Rocking curve for different thicknesses of AlN films.

clusions about the quality of the AlN. For example, Figure 7.5 shows how the same quality aluminum nitride film produces significantly different rocking curves on different surfaces. The best rocking curve is reported on epitaxially grown AlN on sap-

7.2 Methods of Deposition of Piezoelectric Films

Figure 7.4

Coupling coefficient correlation with rocking curve FWHM [9].

Figure 7.5

X-ray diffraction of AlN film growth on different substrates.

177

phire substrate and has a value of less than 0.03°, due to good match of film/substrate crystal structures [10]. It is helpful to use basic theory of adatom mobility to understand the impact of different deposition parameters on the film grain orientation. It is postulated that

178

Thin Films Deposition for BAW Devices

before atoms/molecules are incorporated into the surface film, they will diffuse along the surface along the substrate until they are either chemically bonded to the surface or are physically adsorbed to the surface. Mobility of these particles depends on many factors such as temperature of the surface, particle bombardment of the surface, energy of the adatom upon arrival at the surface, surface condition, and the ability to form chemical bonds with the material on the surface. Generally, it is thought that if these particles have more mobility, they are more likely to find low-energy binding sites, resulting in a crystalline growth. One of the approaches to explain different factors that impact the adatom mobility is called “sticking coefficient.” Sticking coefficient tries to quantify the ease of adatom sticking or being trapped on the surface of the substrate. The easier it is for adatom to get trapped as on the surface, the higher the sticking coefficient of the material. This concept is used for explaining why some materials have better step coverage or better grain orientation. The higher the sticking coefficient the more difficult it is to obtain good step coverage or preferred grain orientation with the material in question. Sticking coefficient is influenced by many factors, some are related to the surface condition while others are related to the kinetic and potential energy of the incoming adatom; yet others are attributed to the surface bombardment. One of the important factors related to the surface condition is temperature of the surface. Based on Thornton’s theory [11, 12] there is a strong relation between the structural growth of the film and the material’s melting point. Below 0.1 times melting point, the adatom surface mobility is negligible and film starts to grow in the direction of the material flux. Therefore mobility increases with temperature, but there is an important threshold that must be reached before it plays a significant role. Mobility associated with the temperature typically starts being significant at temperatures above 0.3 times melting point of the material in question. At this point adatom surface diffusion starts to play a significant role and the film reaches a density close to bulk material value. AlN melting point is about 2,200°C. So, below 660°C there is little mobility due to the temperature of the substrate. It actually takes temperature above 0.5 times melting point (or 1,100°C) to get epitaxial growth purely due to the surface temperature [13]. An important point about surface temperature is that most of the correlations are made measuring substrate temperature. Wafer temperature is usually controlled by some type of the heater with closed-loop temperature control. If part of the heat is supplied due to wafer bombardment and part due to the heater, it is possible to have higher mobility of adatom in a system that has most of its energy supplied due to wafer bombardment versus a system that relies on the heater. It is likely that if electrons, ions, neutrals, or condensing metal strikes the wafer the surface will get hot and then transfer heat to the rest of the wafer. On the other hand, if there is little energy that comes from the bombardment and most of the heat comes through the wafer backside, it is likely that at the same wafer temperature, the wafer with the most bombardment will have much higher real surface temperature. Surface roughness is another surface condition that impacts the sticking coefficient and plays a significant role in the growth process. Smoother surface will result in higher mobility and better grain orientation of the films. This works well if AlN is deposited on surfaces such as silicon dioxide or silicon nitride. When metal electrodes are used, it is not always the case that a smoother electrode produces more

7.2 Methods of Deposition of Piezoelectric Films

179

oriented AlN film. There are plenty of extremely “rough” electrodes like tungsten that have an extremely rough surface but produce highly oriented films. It is most likely that the grain orientation of such materials naturally causes AlN molecule to stick to the low-energy binding site even though there is very little mobility on a rough surface. Removing moisture or organic contamination results in higher mobility and better AlN films. Typically sputter-etch or ion mill is used to remove either moisture or resist residue left over from the electrode patterning. It is also consistent with the mobility theory. If the adatom can bond with oxygen or organic contaminant, it is more likely to reduce its surface mobility and cause it to get stuck to an unfavorable site. Increasing mobility can also be accomplished by physical bombardment of the surface. Typically surface is bombarded by either energetic neutrals, electrons or charged ions. It has been observed that increased surface bombardment can lead to better grain orientation. A standard method to increase surface bombardment is to apply negative potential to the substrate. Small substrate bias of less than 100V is usually sufficient; a much higher bias can actually cause damage to the surface and create defects, vacancies, and dislocations in the growing film due to impact of heavy Ar ions with high kinetic energy, resulting in poor crystal orientation. Excessive bias can also cause unacceptably compressive stresses in the film. Another way to increase energy of the surface bombardment is to lower deposition pressure. Lower deposition pressure leads to longer mean free path between collisions. This reduces energy loses (both kinetic and potential) of ions and neutrals before they strike the substrate, resulting in higher adatom surface mobility and lower sticking coefficient. Higher energy adatoms have higher mobility. It has been observed that more energetic plasmas (such as typically observed in a dual magnetron AC deposition) have a lot more energy [14]. For AC frequency less than 100 kHz, a half-cycle period is long enough compared to the time required for the plasma components to come to equilibrium with the electric filed. The behavior of the AC discharge is very similar to the DC discharge, with the only difference that current is reversed every half cycle. AlN film formed on the target surface (when process operates in poisoned mode) is an insulator and works like a capacitor, charging/discharging during half-cycle sequentially. This causes voltage to almost double across negative glow discharge at the beginning of the cycle. A plasma sheath forms near the negative target and ions accelerate across this region by instantaneous electric field. Also, ions are able to cross the plasma sheath in the much shorter time interval compared to the AC period. These energetic ions gain energy approximately equal to the peak of AC voltage. When these high-energy ions arrive at the cathode they produce highly energetic material being sputtered off the target. As an example, AC-deposited films can have either highly tensile or compressive and still be highly oriented and highly piezoelectric. In the DC-powered process, highly tensile films usually have poor crystal orientation and tend to have lower piezoelectric coupling coefficient. The adatom potential energy can be increased by using a positive plasma column. In the positive plasma column electrons reach equilibrium with the electrical field. The electric fields in the glow discharge can put energy directly into the electrons and accelerate them in the positive plasma column. This process represents the

180

Thin Films Deposition for BAW Devices

main source of energy. In this way the positive plasma column is almost equivalent to the RF plasma discharge. One of its important characteristics is a possibility of creating a large plasma volume and increasing the probability of the Penning reaction of ionization. The large volume of the positive plasma column increases ionization of reactive species and probability of the ion reaction between aluminum and nitrogen, resulting in a high potential energy of AlN adatom. For example, ion reaction produces much higher potential energy than reaction between excited atoms Al + N − → AlN + Q1 Al + N* → AlN + Q2

where Q1 >> Q2. Positive plasma column can be achieved by simply increasing the distance between an anode and cathode, or in other words putting an anode far behind Faraday dark space (see Figure 7.6). For dual target configuration with AC power between them, it is just increasing distance between targets. In addition, in the AC process with dual-target configuration, a magnetic field next to the target, which has a positive potential at the moment, creates an anode layer with high plasma density. Electrons created at the cathode are accelerated by the cathode plasma sheath and strike the anode with enough energy to emit additional secondary electrons. This process is especially important at low pressure. Unbalancing of a magnetron is another method of increasing plasma density and plasma volume. This method creates additional electron traps in a volume between target(s) and substrate which can also increase plasma density and ionization/excitation reaction in the path of AlN adatom to the substrate. This allows for more oriented films without having to apply substrate bias. Reactive sputtering can be used with RF, AC, and pulsed DC depositions. Early work on piezoelectric AlN involved RF diode deposition from an aluminum target in argon and nitrogen ambient. RF diode depositions used in these configurations were run at high pressure and no substrate bias and had low energy plasma. In order to obtain high-coupling coefficient films it was necessary to deposit at above

Figure 7.6

Plasma discharge with the positive plasma column.

7.2 Methods of Deposition of Piezoelectric Films

181

1,100°C [15, 16] because the deposition relied mostly on temperature to achieve high adatom mobility. This made it very difficult and slow process, not practical for the production environment. AC deposition of AlN with dual-target magnetron was first demonstrated by Este and Westwood [17] in the late 1980s. In this configuration, two magnetrons both having aluminum targets are alternatively sputtered using argon/nitrogen plasma. Each target acts as anode or cathode alternating at medium frequency, typically 40 kHz. An advantage of this method compared to the RF diode is that there is enough energy transferred to the deposited film that no extra heating of the wafer is required. The speed of the deposition with the magnetron is significantly higher as well. This technique is easily incorporated into a high-volume production system. High-quality piezoelectric films have been demonstrated in production environment using such a configuration [18]. Pulsed DC (bipolar) magnetron has been successfully used to reactively deposit AlN films [19]. In a bipolar deposition a simple DC magnetron is used with a special power supply that removes charge build-up from the target by reversing polarity on the target for a short period during every cycle. Typical frequency for this device is between 10 kHz and 250 kHz. The advantage of this technique is that a fairly standard sputtering system can be modified to deposit AlN films. All it takes is making a special anode to prevent it from coating with AlN too quickly (a so-called disappearing anode problem), using special power supply and making minor modifications to the shields and magnetron. Substrate bias is commonly used to obtain highly oriented piezoelectric films with desired stress. From a more practical standpoint there are issues that must be addressed in order to manufacture highly piezoelectric AlN films. Arguably the most important factor impacting the quality of the deposited AlN film is the ability to have the least amount of oxygen in the deposited film. Most people who studied low kt2 films versus high kt2 films typically observed two things: bad films have poor crystal orientation and higher concentration of oxygen [13, 18]. It is quite likely that these two are mutually inclusive (i.e., films with poor crystal orientation have larger space in the grain boundaries for oxygen to get incorporate or the high concentration of oxygen in the film prevents highly structured crystal growth). Regardless of which is correct, oxygen and crystal orientation are the usual suspects when low kt2 AlN is observed. Oxygen is probably the reason why it has always been easier to get highly piezoelectric-sputtered films of ZnO than AlN. Since ZnO is sputtered in oxygenrich ambient it would be insensitive to the additional oxygen present in the sputtering chamber due to leaks or out-gassing water vapor. In our investigations we have found that regardless of the type of deposition used or the mode of deposition, increase in oxygen level leads to lower coupling coefficient in AlN films. Sources of oxygen in the sputtered AlN films can come from either the surface of the wafer, leaks in a vacuum chamber, or outgassing of the shields in the deposition chamber. In the older vacuum systems it was very difficult to get low oxygen/water vapor levels during the deposition. Most systems had many vacuum seals that could leak and large surface areas that outgassed during elevated temperatures of the deposition. Since the introduction of well-designed cluster tools in the early 1990s, most modern sputtering systems have well-designed vacuum chambers that have low base pressure (10−8 torr range). It is now a common practice to get low level of water

182

Thin Films Deposition for BAW Devices

vapor before attempting AlN deposition. Another precaution commonly used to obtain high-quality AlN films is a small amount of surface cleaning with RF plasma to remove moisture from the substrate surface before AlN deposition. With these fairly easy techniques most systems are capable of depositing good quality aluminum nitride. As with many reactively sputtered films, AlN can be deposited in two modes: “metallic” and “poisoned.” In the metallic mode, partial pressure of nitrogen is kept low enough to prevent formation of a thick insulating layer of AlN on the target. Material coming off the target reacts with nitrogen either on a way to a substrate or on the substrate surface. In the poisoned mode, target surface reacts with nitrogen completely to form a layer of aluminum nitride; see Figure 7.7. This layer is then sputtered off the target as AlN molecule. These molecules land on the surface of the wafer and condense to form AlN film. There is a transition mode between those two regions. This regime is very unstable due to hysteresis function of the transition process parameters from metal to poison and from poison to metal modes. Those instabilities make it unsuitable for high-volume commercial application. In the past metallic mode deposition was a preferred method to deposit AlN film due to the fact that it is about three to five times faster than poisoned mode. In the late 1990s there was a shift to a poisoned mode deposition. Even though it is slower, poisoned mode produces highly piezoelectric films that are repeatable and have controllable stress. In contrast, fast metallic mode depositions have a hard time producing highly piezoelectric films with zero or slightly tensile stress. Authors have been able to produce high-quality films in poisoned mode with stresses ranging from 1,000 MPa tensile to −1,000 MPa compressive without any trouble. In metallic mode, stress had to be above −700 MPa compressive to achieve reasonable coupling coefficient. Another problem with metallic mode is that a sophisticated control of deposition power and partial pressure of nitrogen gas are required to maintain stable process over the entire target life. It is generally reported that higher temperature produces more piezoelectric AlN films [7, 13, 18]. For the early RF diode depositions, temperature in access of

Figure 7.7

Metallic and poison zones of reactive sputtering.

7.2 Methods of Deposition of Piezoelectric Films

183

1,100°C [15, 16] were reported to be required. For DC or AC depositions it was reported that temperature above 400°C were desirable [13, 18]. It is, however, difficult to separate wafer temperature and the energy of the sputtered molecules. It is possible to get good films at low wafer temperature, but with highly energetic plasma and AlN adatoms with high potential energy. It has been observed that even with water-cooled wafer, it is not difficult to deposit good films with either high DC power or medium to high AC power. If one assumes that there is a certain minimum energy level required to get good films, it is easy to see why many combinations of wafer temperature, deposition power, pressure, oxygen level, and surface roughness can produce acceptable results. Most people use low pressure to produce higher quality AlN. It is consistent with adatom mobility theory that at lower pressure AlN molecules will have fewer collisions resulting in more energy remaining for the molecule to find more favorable energy site [13, 18]. Such process regime does have some disadvantages. Columnar growth is perpendicular to the growth surface. Unfortunately, if there are sharp corners the growth at the corner tends to have poor orientation and step coverage that may lead to low-voltage breakdown and leakage problems. Monte Carlo simulations tend to predict fairly well the problems with poor grain formation and voids around such structures (see Figure 7.8 [20] and Figure 7.9 [21]). A typical solution to this problem is to avoid 90° angles. Shallow angles help avoid voids and poor grain structure on sidewall structures. Even though high-quality aluminum nitride films tend to be smoother than poorly oriented films, film surface roughness increases with film thickness (see Figure 7.10). 7.2.2

Practical Aspects of the Sputter Deposition of the AlN Films

Reactive deposition of AlN has been has been successfully performed in a variety of different sputtering machines. Most of the machines on the market can provide adequate AlN films for research applications. It is important, however, to make sure that if the machine is to be used for volume production it meets certain requirements. Metallic films are polycrystalline Microvoids and grain boundaries

Impinging atoms

~0.25 μm

Figure 7.8

Columnar (rough) growth and pores more likely because of oblique incidence and low surface diffusivity

10 nm

Monte Carlo simulation of AlN grain growth as a function of geometry.

184

Thin Films Deposition for BAW Devices Ψm = 0°

Ψm = 0°

Ψm Ψm = 70°

Ψm = 45°

Figure 7.9

Figure 7.10

Comparison of the impact of the sidewall angle on the AlN film growth.

AlN surface roughness as a function of the film thickness.

Typical problems with AlN deposition systems involve flaking of the AlN from the chamber shields after enough material builds up on it. Most machines can run without flaking for a while, but a practical requirement is at least five hundred microns of AlN without flaking. Typically, this problem is addressed by using aluminum depositions between aluminum nitride depositions. This coating not only

7.2 Methods of Deposition of Piezoelectric Films

185

prevents flaking, but in a pulsed-DC system it refreshes anodes that otherwise would get coated by dielectric AlN and electrically disappear. In order to get maximum wafer throughput it is tempting to reduce or eliminate such aluminum coatings. But only running through the entire target will confirm if conditioning is appropriately optimized. Since most BAW applications require tight control of AlN, uniformity of AlN deposition is a critical parameter. Most machines on the market today can demonstrate about 0.3% one sigma uniformity on 6- or 8-inch wafers. Unfortunately, since most systems have either dual targets or multiple erosion rings on a single target, as the target is being consumed uniformity changes. Some systems have ability to compensate for this by adjusting power to each target in a dual target configuration [18]. Single-target systems have ability to change space between substrate and the target. Another compensation technique used by some machines is an external magnetic field that can be manipulated over the target life to compensate for this problem. It is important to check for the uniformity changes throughout the entire target because some systems have better compensation schemes than others. Without actually going through the target, it is impossible to know if the system will provide adequate uniformity or not. Unfortunately, even the best uniformity available on the sputtering machine is not adequate for many applications. It is a typical requirement for many filters to have frequency control of ±0.1 percent total range. There are several approaches used in the industry to obtain such requirements. One can either etch each deposited layer of the device to get perfect film uniformity at each step, or etch the wafer at the end of the process based on the frequency map. The most accurate method is to use a focused ion beam (FIB) to adjust the thickness of each resonator. This approach can get 100% of dice on frequency but requires many hours to tune each wafer in a very expensive machine. Another method relies on using masks to etch only desired areas to be trimmed. An easier way is to use ion mill with a small beam size (from 0.2 to 1 cm) and tune the wafer at the end of the process using frequency map of the finished wafer, or tune based on the thickness map at each deposition step. There are several commercial makers of both FIB and ion mill equipment. Some use chemical-etch enhanced with the ion bombardment. This approach allows removal of very thin layers, but with a side effect of leaving etch chemical implanted into the surface of the wafer. Using an argon-based ion beam eliminates this issue. an ion beam simply removes more or less material based on a preprogrammed uniformity map. There are two modes of operation: one is to use constant power and move the wafer at variable speed; another is to use variable power and constant wafer speed. If the ion source can have stable, quickly controlled power, it is possible to tune the wafer to the thinnest point on a wafer. The constant power approach is limited by the ability to increase or decrease speed of the wafer motion, but it is easy to maintain constant power during the process. Generally, it is hard to go from very slow to very high speed. This leads to a minimum amount of material that can be removed in order to have a smooth wafer motion. Generally, the loss of material is more severe with variable wafer speed than with variable power design. Another way to combat thickness nonuniformity is to deposit extra material in the desired areas. This method is more complicated and requires either very slow FIB deposition die-by-die or masking/depositing/lift-off process. Due to the relative

186

Thin Films Deposition for BAW Devices

difficulty of this method compared to the etch method, it is not widely used in high-volume applications. BAW applications require very good wafer-to-wafer thickness repeatability. It is fairly common to have a “first wafer” effect on most systems. Depending if the system has been recently used or set idle for a while, the first wafer will have a slightly different thickness than the following wafer. Another problem is that as the targets get eroded, the deposition rate tends to fall. Not only does the deposition rate drop due to a self-shadowing effect in the eroded area (also known as “racetrack”) but as Este and Westwood [17] showed in the 1980s, the deposition rate of the reactively sputtered AlN is extremely dependent on the target voltage. As the target erodes, voltage drops. In most modern systems it is very difficult to adjust magnetic field to keep constant voltage over the entire target life, thus, it is important to look at different ways to keep wafer-to-wafer thickness repeatability over the target life. A simple solution to these problems is to use a laser interferometry end point to stop at a required thickness [22]. A less optimum solution is to have a computer adjust deposition time or power based on the historical data from the system. Film stress is an important consideration in production depositions of AlN. In the free-standing membrane applications, excessive stress can cause cracking or peeling of the resonators. In the solidly mounted application, stress can cause wafer bowing that is unacceptable in many photolithographic processes such as steppers, resulting in wafer loss. Stress control can be accomplished by standard means described by Vossen [5]. Using power, pressure and substrate bias works as expected in any sputtering application (i.e., higher power or lower deposition pressure produces more compressive stress). Applying substrate bias produces more compressive stress with diminishing effect as the substrate bias is increased. Some equipment-specific modifications to stress are described by Mishin [18] (see Figure 7.11). Besides such obvious stress control methods, it is possible to modify stress through surface treatment. Making the surface smoother by CMP or etching will

Figure 7.11

AlN film stress as a function of the magnetic field.

7.2 Methods of Deposition of Piezoelectric Films

187

usually result in more compressive AlN films. Breaking films into multiple steps have been reported [23] to modify film stress. PZT deposition by sputtering is a lot more complex than either AlN or ZnO depositions. RF diode deposition of PZT is an extremely slow process when deposited in the commercially available equipment, resulting in deposition rates of 50 to 100 Å/min. Conventional RF magnetron processes can achieve up to 500 Å/min deposition rates by employing an appropriate array of magnets, but it is very difficult to get correct stoicheometry from a given target over the entire target life because of the different yield level of the lead, zirconium, and titanium at different voltages [24]. It is also important to have the right mixture of argon and oxygen to maintain desired ratio of the materials in PZT. Because PZT is a dielectric, it does not lend itself to either DC or AC magnetron deposition techniques commonly used for either AlN or ZnO. An interesting deposition method designed to circumvent these issues is described in [25]. Using a hallow cathode effect with high-rate gas flow, deposition rates of up to 2,500 Å/min were demonstrated. Authors claim films as thick as 16 μm were deposited with excellent film characteristics in only ninety minutes. Typically, hollow cathode consists of a cylindrically shaped target arranged perpendicular to the substrate. Very high plasma density creates high erosion rate of the material from the target. Argon and oxygen flowing through the center of the target carry target material to the wafer. In PZT deposition, target is composed of the individual rings of lead zirconium, and titanium. In order to obtain the desired ratio in the PZT material, the size of the target rings is adjusted to obtain appropriate composition on the surface of the substrate. Temperatures of 550°C to 650°C as well as a small amount of the substrate bias are used to obtain dense, highly oriented PZT. This approach demonstrates a path forward for a PVD deposition of PZT at high deposition rates with good control over the stoichiometry of the PZT. Unfortunately, this system is in the experimental stages and no commercially proven equipment is available on the market. Sputter deposition from ceramic AlN or ZnO targets can be accomplished only in the RF-deposition mode. This is due to a need to couple between backing plate and the insulating target. In order to accomplish this, higher frequency is required. Unfortunately deposition rate from the ceramic are an order of magnitude slower that either AC or pulsed-DC deposition. For this reason, as well as cost and difficulty of making a ceramic target, this technique did not find a wide commercial application. 7.2.3

Electron Cyclotron Resonance Deposition

Electron cyclotron resonance (ECR) has been used to deposit AlN [6] and ZnO films of high quality. Typically, argon/nitrogen mixture is used with aluminum target. A magnetic field of about 875G is driven at 2.45 GHz. The advantage of the technique is that it allows a nearly epitaxial film deposition at temperatures below 300°C. The disadvantage of ECR is that because deposition rate tends to be low there are no high-volume, low-cost, production machines available on the market. Most work reported in the literature is from universities and R&D laboratories.

188

Thin Films Deposition for BAW Devices

7.2.4

Ion Beam Deposition

Ion beam deposition (IBD) is a well-known technology and is described in detail in Thin Film Process II [5]. RF or DC source is used to create argon ions that are accelerated through voltages between 500V and 5,000V towards target material. Material removed from the target is deposited on a substrate. In the last decade ion beam deposition has become a more viable technique, especially for composite dielectric materials such as PZT. The advantage of the IBD over a magnetron deposition is that unlike magnetron, composition of the material coming of the target remains repeatable over the target life. Our experience indicates that by adjusting the angle of the source to the target, deposition rates can be more than doubled for some materials. Figure 7.12 shows dependence of the deposition rate versus source angle for some materials. In the past, slow deposition rates combined with high-maintenance requirements have limited usage of IBD to depositions of about 1,000-Å thickness. In the last two decades, development of large gridless sources, operating at high power increased this range to about 10,000-Å range. This makes IBD more practical for some piezoelectric applications. 7.2.5

Metalorganic Chemical Vapor Deposition

Metalorganic chemical vapor deposition (MOCVD) is one of the methods that have been successfully used for depositing PZT materials. The advantage of this technique is that it allows precise composition control and uniformity over a large area [26]. MOCVD provides conformal step coverage that is highly desired in many applications. MOCVD uses liquid delivery and flash vaporization to deliver chemicals into the deposition chamber. Low-volatility materials are desired for safety reasons. Difficulty of safely handling materials prevented wide commercial use of MOCVD for AlN deposition. For PZT there are several chemicals that can be safely used. Pb(thd)2-pmdeta [Bis(tetramethylheptanedionato)lead with pentamethyldie-

Figure 7.12

Deposition rate as a function of the source to target angle.

7.3 Metal Deposition for BAW Applications

189

thlenetriamine adduct], Zr(thd)4 [Tetra(tetramethylheptaanedionato)zirconium] or Zr(thd)2(O-iPr)2 [Bis(isopropoxy)bis(tetramethylheptanedionato)titanium]. The films are deposited in oxidizing atmosphere at about 1.5-torr pressure. Deposition temperature is typically 550°C to 590°C. Because of the high temperature and oxidizing environment, platinum electrodes are preferred to other commonly used materials. 7.2.6

Jet Vapor Deposition

Jet vapor deposition (JVD) is a low-cost, high-throughput alternative for depositing thin film PZT material [27]. The advantage of the system is that it operates in the low-torr range. Such low-level vacuum allows for low-cost chamber and pumping setup. Vapor source provides condensable material for the deposition. Molecules of this material are carried by inert gas at speed around 1,000 m/s towards the substrate. For larger substrates, multiple jets in conjunction with moving substrate can deposit uniform films on a larger substrate or multiple substrates. Films tend to have very fine grain structure after the deposition. An anneal at 550°C to 650°C can increase the grain size to the desired range. So far this technique was only applied to only PZT depositions. 7.2.7

Nonvacuum Deposition

PZT is the best material for application requiring extremely high coupling coefficient and relatively low frequencies. Unlike AlN or ZnO, there is a simple nonvacuum deposition technique that can be used to get reasonably good piezoelectric films. One of the popular techniques of depositing thin PZT layers is a sol-gel deposition. It is the most widely used method of depositing PZT of intermediate thicknesses. It is highly desirable because it is compatible with standard semiconductor processing [28–30]. Sol-gel, a metal-organic solution that contains oxide components in an organic solvent, is usually spun by the same method as used to deposit photoresist. A small amount of sol-gel is deposited on a spinning wafer. Spin speed is adjusted to obtain the desired thickness and uniformity. Spin step is followed by a low temperature (300°C to 400°C for several minutes) baked, followed by high temperature (600°C to 750°C for about 30 to 60 minutes) anneal. Ideal thickness is between 1,000 to 2,000Å. This process can be repeated multiple times to obtain desired thickness of up to 10 microns. Films deposited by this method show low void density and high coupling coefficient. Even though sol-gel is considered the most viable technique currently in use, there are no proven commercial machines available on the market at the present time. This technique requires careful optimization and a lot of proprietary knowledge to achieve good results, particularly for thicker films.

7.3

Metal Deposition for BAW Applications There are a couple of universal issues that impact most metal electrodes used for BAW applications. Before depositing any electrode it is important to know the con-

190

Thin Films Deposition for BAW Devices

dition and type of surface under the electrode [31]. If the surface has a high degree of crystal orientation and is extremely smooth (such as prime silicon wafer), it is very easy to obtain high-quality electrode material. If the surface is an amorphous silicon oxide or similar material with rough texture, it is important to prepare it before depositing the electrode material. Some of the most common ways to improve surface smoothness are chemical mechanical polishing (CMP) or plasma etching such as ion mill. By using CMP techniques it is possible to obtain surface roughness less than 7Å (see Figure 7.13). There are other techniques of improving surface smoothness, but the common goal is to get close to the same level of smoothness as prime Si wafer (about 5Å RMS). Another common issue with most of the electrodes and particularly with molybdenum, tungsten, and ruthenium is the grain orientation and surface roughness of the deposited film. Films that tend to incorporate oxygen in the grain boundaries can develop poor crystalline orientation and high level of surface roughness. It is best to deposit films with minimum oxygen levels, but sometimes it is not practical or extremely difficult to accomplish. In such cases using CMP or plasma surface etch should be used to improve surface roughness. Mishin [8] shows that surface roughness control of the electrode is critical to achieving good crystallinity of the AlN films. It is also clear that the type of surface can influence the growth of the electrode metal. One example is that if molybdenum is deposited on polished PECVD oxide film, it will have adverse effect on the AlN growth. Using a thin film of AlN on top of the polished PECVD film eliminates this problem. Most BAW applications have a need for metal electrodes that have the following characteristics: acoustic stiffness, low electrical resistance, compatibility with the standard manufacturing techniques, and a favorable surface to orient the piezoelec-

Figure 7.13

CVD-deposited W film surface roughness after CMP process.

7.3 Metal Deposition for BAW Applications

191

tric layer. For solidly mounted bulk acoustic resonator (SMR) devices it is also important to have a large mismatch in acoustic impedance between metal and dielectric in its acoustic mirror stack. Lanz [32] gives a good example of selecting different materials in order to optimize a particular device. Velocity of sound and material density all are important in the choice of the electrode for a particular application. Table 7.2 lists some of the film properties for the films commonly used in the BAW applications. The acoustic impedance and stiffness are both very critical parameters in determining suitability of a given material to be used for a specific BAW application. When acoustic waves travel between two electrodes it is highly desirable for most applications to have most of the energy reflected back by the electrode. The acoustic impedance and stiffness of a material determines how far into the electrode the acoustic wave will penetrate before it is reflected back. If the acoustic stiffness is high, very little energy is lost and penetration into the electrode is minimal. In the SMR applications, acoustic pairs of high and low acoustic impedance are chosen to create quarter-wavelength reflectors. Typically, two pairs with highly mismatched impedance such as tungsten and silicon dioxide, for example, are sufficient to reflect most of the energy back into the stack. On the other hand, when materials with less acoustic mismatch are used, for example, SiO2/AlN, it takes as many as seven pairs to achieve the same result [2]. Many metals have been tried for different piezoelectric materials. Materials such as aluminum, gold, and copper have been tried because of their low electrical resistance and ready availability in many manufacturing facilities. Both gold and copper have been rejected due to their poor acoustic stiffness. Materials like tungsten, molybdenum, platinum, and ruthenium have been used because of their relatively high acoustic stiffness and ability to help orient piezoelectric films. Some additional properties of those metals are shown in Table 7.3 [33]. 7.3.1

Aluminum

Aluminum has been used with reasonable success for AlN and ZnO applications. It is the most readily available material in most fabrication facilities and is easily deposited with a standard DC-sputter deposition. If the device needs low resistance and does not require exceptional coupling coefficient, aluminum is a good choice. Table 7.2

Material

Acoustic Properties of Selected Films Sonic Velocity Acoustic Density V (long.) impedance Stiffness c11 11 (g/cm3) 103 m/s 103 m/s (10 Pa)

AlN(002) 3.26

11.0

3.6

4.2

SiO2

2.2

5.6

1.2

0.7

Pt

21.5

—

—

3.5

Al

2.7

6.3

1.7

1.1

Mo

10.2

6.2

6.4

—

W

19.2

5.2

10

—

Cu

8.9

4.7

4.2

—

192

Thin Films Deposition for BAW Devices Table 7.3

Physical Properties of Selected Metals

Thermal Bulk Thermal Resistivity, Density, Expansion, Conductivity, Material Ohm-cm g/cc ppm W/cm/K Al

2.8

2.7

22.9

2.37

Pt

10.7

21.5

8.9

0.716

5.7

10.2

5.1

1.38

5.39

19.2

4.3

1.73

4.5

8.9

0.46

12.2

9.6

1.17

Mo W Ti Ru

55 7.3

Devices with Kt2 of as high as 5% have been fabricated. Unfortunately, many devices require Kt2 above 6%. For such devices several materials with relatively high stiffness have been used. Molybdenum, tungsten, platinum, and ruthenium have all been used to demonstrate high Kt2. 7.3.2

Molybdenum

Molybdenum (Mo) has been used in high-volume manufacturing of BAW resonators with AlN piezoelectric material mainly because it is a very acoustically stiff material and has moderately low resistance. Mo is refractory metal and has thermal expansion coefficient similar to aluminum nitride which minimizes impact from thermal mismatch and residual stress to Mo/AlN stack. Mo has body-centered cubic crystal structure [33]. It can easily be deposited by either DC or AC sputtering. Care has to be taken when depositing molybdenum. If the sputtering system has high oxygen or water level during deposition, films will have rough surface, poor resistivity, and will also prevent growth of highly oriented AlN. It has been observed that slightly compressive films deposited with either substrate bias or at low pressure usually produce fairly smooth films with relatively low resistivity. Molybdenum XRD is a good indicator of the film quality. If XRD is less than 2° and film has not been severely oxidized before AlN deposition, it will provide a good surface for AlN deposition. Such film can be left for couple of weeks before AlN deposition and will not cause a problem with the aluminum nitride deposition. If XRD is significantly higher than 2° or the molybdenum surface has been severely oxidized in previous processing, AlN film grown on it will have poor orientation and high XRD. 7.3.3

Tungsten

Tungsten (W) has higher acoustic impedance and acoustic stiffness than molybdenum, thus making it easy to deposit films with high Kt2. W is refractory metal and has thermal expansion coefficient similar to aluminum nitride which minimizes impact from thermal mismatch and residual stress to W/AlN stack. Tungsten has body-centered cubic crystal structure [33]. Unfortunately, it has several disadvantages. It is denser than molybdenum, thus, requiring thinner film than molybdenum to produce same frequency loading. When tungsten gets thin, its bulk resistance rises sharply to as high as 20 to 30 micro-ohm-cm, compared to molybdenum that is about 10 for the equivalent electrode when sputtered using a standard deposition

7.3 Metal Deposition for BAW Applications

193

system with argon sputtering gas. Using heavier gases such as krypton or xenon it is possible to control tungsten resistivity in thinner films. Krypton and xenon are not commonly used in high-volume production environments and poses logistical challenges for practical use. Another issue with tungsten is surface roughness. If tungsten surface is not polished with CMP, AlN deposited on it tends to produce highly variable coupling coefficient. With appropriate surface treatment this problem can be eliminated. 7.3.4

Platinum

Platinum (Pt) has face-centered cubic crystal structure. It is one of the best films to orient the piezoelectric films that are deposited on it. It is one of the few materials that can withstand the high temperature, oxidizing environment of the MOCVD process. It is by far the electrode of choice for PZT applications. The most critical parameter in depositing Pt is temperature. It is usual to use 500°C or higher temperature to get desired Pt film’s properties. One of the critical properties of Pt for PZT applications is a very tightly packed grain orientation. It has been showed that for the high-temperature anneals that are frequently used for PZT processing it is critical to have tight grain orientation in platinum to avoid PZT diffusion through the grains of platinum. But, at the same time, higher deposition temperature cause problems with stress and surface roughness (see Figure 7.14). Much less success was obtained when platinum was used for AlN BAW applications. Even though very high Kt2 was obtained with Pt electrodes, typically because of the density of platinum such electrodes tend to be very thin and have high resistance. This has detrimental impact on the Q of the device, making it impractical for most of the applications.

Figure 7.14

Pt film properties versus deposition temperature.

194

Thin Films Deposition for BAW Devices

7.3.5

Ruthenium

Ruthenium (Ru) has been used in BAW applications. Crystal structure of Ru is hexagonal [33], with lattice constant: a = 2.706Å, c = 4.282Å. It has been shown to produce high Kt2 as well as very high Q. In practice, Ru tends to oxidize fairly quickly in the ambient atmosphere, especially if it has poor crystal orientation. Once heavily oxidized, it is critical to remove this oxide before depositing AlN films. If the oxide is not removed, AlN films tend to have poor orientation. Ru XRD is not a good indication of the quality of AlN film that will be deposited on it. For example, XRD of AlN deposited insitu on Ru film that has XRD as high as 12° can be similar to the XRD of AlN film deposited on molybdenum electrode. But unlike molybdenum, if Ru film is left in the ambient atmosphere for couple of weeks, AlN films grown on it will have two to three times higher XRD. Ru is fairly expensive material compared to the other electrode materials. For these reasons, there are very few people who use it in high-volume applications. 7.3.6

Combinations of Metals

Lakin [2] describes a use of a combination electrode to get the best of both worlds: highly stiff electrode contacting the piezoelectric film and a thin layer of highly conducting material to help reduce the resistance of the electrode. It appears that a 1,000Å layer of aluminum can accomplish this requirement. Since aluminum is a relatively light material it doesn’t significantly alter the frequency mass of the electrode. If there are no other compatibility reasons, aluminum is an easy choice for this application.

References [1] MacDonald, N., et al., “AlN Film Bulk Acoustic Resonators,” UCSB MEMS. [2] Lakin, K., et al., “Improved Bulk Wave Resonator Coupling Coefficient For Wide Bandwidth Filters,” IEEE 2001 Ultrasonics Symp., Paper 3E-5, October 9, 2001. [3] Larson, J., S. Gilbert, and B. Xu, “PZT Material Properties at UHF and Microwave Frequencies Derived from FBAR Measurements,” IEEE Ultrasonics Symp., 2004, p. 173. [4] Larson, J ., S. Gilbert, and B. Xu, “PZT Material Properties at UHF and Microwave Frequencies Derived from FBAR Measurements,” IEEE Ultrasonics Symp., 2004, pp. 173–177. [5] Vossen, J., and W. Kern, Thin Film Process II, New York: Academic Press, 1991. [6] Nishihara, T., et al., “High Performance and Miniature Thin Film Bulk Acoustic Wave Filters for 5GHz,” IEEE Ultrasonics Symp., 2002. [7] Loebl, H., et al., IEEE International Ultrasonics Symp., paper 3E-2, 2001. [8] Mishin, S., et al., “Sputtered AlN Thin Films on Si and Electrodes for MEMS Resonator: Relationship Between Surface Quality Microstructure and Film Properties,” IEEE Ultrasonics Symp., Vol. 2, 2003, pp. 2028–2032. [9] Loebl, H. P., et al., “Piezoelectric Materials for BAW Resonator and Filters,” IEEE International Ultrasonics Symp., 2001, pp. 807–811. [10] Uchiyama, A., et al., “Growth of AlN Films by Magnetron Sputtering,” Journal of Crystal Growth, Vol. 189–190, 1998, pp. 448–451.

7.3 Metal Deposition for BAW Applications

195

[11] Thornton, J., “Influence of Apparatus Geometry and Deposition Condition on the Structure and Topography of Thick Sputtered Coatings,” Journal of Vacuum Science and Technology, Vol. 11, 1974, pp. 665–670. [12] Thornton, J., et al., “Internal Stresses in Metallic Films Deposited by Cylindrical Magnetron Sputtering,” Thin Solid Films, Vol. 64, 1979, pp. 110–120. [13] Hsieh, P., M.S. Thesis, MIT, 1999. [14] Glocker, D., “Influence of the Plasma on Substrate Heating During Low-Frequency Reactive Sputtering of AlN,” J. Vac. Sci Tech., A 11, Vol. 6, November/December 1993. [15] Rutz, R., E. Harris, and J. Cuomo, IBM J. Res. Dev., Vol. 17, No. 61, 1973. [16] Shuskus, A., T. Reeder, and E. Paradis, Appl. Phys. Lett., Vol. 24, No. 4, 1974, pp. 155–156. [17] Este, G., and W. Westwood, “A Quasi-Current Sputtering Technique for the Deposition of Dielectrics at Enhanced Rates,” J. Vac. Sci. & Tech., A6, N.3, Part II, May/June 1988. [18] Oshmyansky, Y., et al., “Sputtering Processes for Bulk Acoustic Wave Filters,” Semiconductor International On-line Addition, March 1, 2003. [19] Loebl, H., et al., “Solidly Mounted Bulk Acoustic Wave Filters for the GHz Frequency Range,” IEEE Ultrasonics Symp., 2002, p. 923. [20] www.ima.umn.edu/talks/workshops/6-5-9.2000/osullivan/ima-release.pdf. [21] www.cletus.phys.columbia.edu/~windt/papers/2003_JAP_94_263.pdf. [22] Mishin, S., B. Sylvia, and D. R. Marx, “Improving Manufacturability of AlN Deposition Used in Making Bulk Acoustic Wave Devices,” IEEE Ultrasonics Symp., 2005. [23] Cumbo, F., D. Metacarpa, and M. Serry, “Film Bulk Resonator Process Technology,” Veeco FBAR Appsnotes, 02730. [24] Suu, K., et al., “Lead Content Control of PLZT Thin Films Prepared by RF Magnetron Sputtering,” Integrated Ferroelectrics, Vol. 14, 1997, pp. 59–68. [25] Jacobsen, H., et al., “High-Rate Sputtering of Thick PZT Layers for MEMS Actuators,” MEMS 2006, Instanbul, Turkey, January 22–26, 2006. [26] Chen, I. -S., et al., “Metalorganic Chemical Vapor Deposition Pb(Zr,Ti)O3 and Selected Lower Electrode Structures as a Pathway to Integrated Piezoelectric Microelectromechanical Systems,” J. Vac. Sci. Tech. B, Vol. 19, No. 5, September/October 2001, p. 1833. [27] Golz, J., et al., “Jet Vapor Deposition of Lead Zirconate Titanate (PZT) for Thin Film Pyroelectric Detectors,” Mat. Res. Soc. Symp. Proc., Vol. 284, 1993. [28] Frederick, A., M.S. Thesis, University of Pittsburg, PA, 2006. [29] Kim, S. H., et al., “Dielectric and Electromechanical Properties of PB(Zr,Ti)O3 Thin Films for Piezo-Microelectromechanical System Devices,” Japan. J. Appl. Phys., Vol. 42, Part 1, No. 9B, 2003, p. 5952. [30] Zhou, Q., and K. Shung, “Fabrication of Sol-Gel Modified Piezoelectric Thick Films for High Frequency Ultrasonic Applications,” IEEE International Ultrasonics, Ferroelectrics, and Frequency Control Joint 50th Anniversary Conf., 2004, p. 1958. [31] Iriarte, G., et al., “Synthesis of C-Axis Oriented AlN Thin Films on Metal Layers: Al, Mo, Ti, TiN and Ni,” IEEE Ultrasonics Symp., 2002, p. 315. [32] Lanz, R., M. Dubois, and P. Muralt, “Solidly Mounted BAW Filters for the 6 to 8 GHz Range Based on AlN Thin Films,” IEEE Ultrasonics Symp., 2001, p. 844. [33] Papertech Marketing Group Inc., “Periodic Table of the Elements,” 1994.

CHAPTER 8

Characterization of BAW Devices Gernot Fattinger and Stephan Marksteiner

8.1

Introduction As we have seen in the previous chapters, the response of a thin film BAW resonator can be well predicted (up to first-order effects) by a one-dimensional wave model: the acoustic field in the resonant layer stack is described by waves that travel perpendicular to the various thin film layers. All of these layers contribute somehow to the shape and resonance frequency of the fundamental acoustic mode. This means that in addition to the piezolayer itself, every other layer (whether part of the electrodes, part of the trimming layer on top of the resonator or part of the acoustic mirror under the actual transducer), contributes to the electrical behavior. However, the physical mechanism and the strength of this influence depends strongly on where the corresponding layer is located and what purpose it serves. The electrodes, for example, being located directly beneath and above the piezolayer, have a particularly strong influence on the resonance frequency, especially if they are made from heavy (or high acoustic impedance) materials [1]. In contrast, the lower layers of an acoustic mirror have a basically negligible influence on the resonance frequency. On the other hand, they contribute significantly to the quality factor of the resonance by reflecting residual energy that leaked through the upper mirror layers. So even if a layer has very small frequency sensitivity, it does not necessarily mean that large thickness fluctuations can be tolerated during manufacturing. The bottom line of this discussion is that a good control of the layer thickness—made possible by proper measurement means—as well as reproducible material properties for all constituting layers is of huge importance to optimize the performance and to guarantee the manufacturing stability of the resonator and filter response. In Section 8.2 we will therefore focus on the measurement methods for the characterization of single thin film layers. In Section 8.3 we will discuss methods for directly characterizing the acoustic properties of the multilayer BAW stack. In Section 8.4 we will discuss the dominant loss mechanisms related to various physical effects in SMR/FBAR resonators. In Section 8.5 we will finally turn to the discussion of the electrical characteristics.

197

198

8.2

Characterization of BAW Devices

Single-Layer Material Characterization 8.2.1

Introduction

Any BAW device consists basically of a stack of thin film layers that are deposited and patterned consecutively on a substrate using standard semiconductor/MEMS processing technology. Metallic layers are used for electrodes, interconnects, and potentially also for the high-Z layers in the acoustic mirror. Dielectrics are used (e.g., for passivation purposes), as frequency trimming layers or low-impedance layers in the acoustic mirror. And finally, there is the piezoelectric layer which also belongs to the category of dielectrics. Note that semiconducting layers do not play an active role in BAW devices (with one exotic exception [2]). The silicon wafers commonly used for SMR and FBAR manufacturing serve only as low-cost substrates that can conveniently be handled with standard semiconductor equipment. In contrast, the semiconducting properties of the substrate are unwanted in the sense that they may result in a parasitic capacitive coupling between the various resonators within an SMR/FBAR filter chip. Therefore, proper care needs to be taken to avoid both intrinsic ohmic conduction of the silicon itself, as well as oxide and/or interface charge-induced carriers. From the simplified BAW resonator model (e.g., the 1D Mason model; see Section 3.2.3) it is easy to see that the effect of any layer of thickness t in the BAW stack is uniquely determined by its acoustic delay τ = t/vL of the longitudinal wave (where vL is the longitudinal sound velocity), and by its acoustic impedance ZL. If the material parameters (vL and ZL) are known, then the geometrical thickness t completely determines the contribution of the thin film layer to the total acoustic motion of the stack. In the following subsections we will discuss methods for characterizing the properties of BAW thin film layers. Basically, all these methods are standard methods widely used in semiconductor manufacturing. 8.2.2

Dielectric and Piezoelectric Layers

The by far most important dielectric layers used in BAW manufacturing are silicon oxide and silicon nitride. They can be deposited with good uniformity and thickness control, high quality, and excellent reproducibility. The deposition is typically done by plasma-enhanced chemical vapor deposition (PECVD) at moderate temperatures (<450°C) or by RF sputtering. Other dielectrics sometimes used as high-Z layers in acoustic mirrors are tantalum pentoxide, aluminum nitride, or aluminum oxide. The most commonly used piezoelectric materials are aluminum nitride and zinc oxide. In order to measure the thickness of a dielectric layer several methods can be used. If the thin film is patterned during the process, a direct measurement with a profilometer is possible. With this method the thickness of the patterned layer is characterized by softly setting a probe needle on the surface and scanning it across the etch step. This method is cheap and fast, accurate in the nanometers range (if properly calibrated), is very localized (the needle tip is a few tens of micrometers large) and can also deliver information about short-range thickness nonuniformities (e.g., that occur if chemical mechanical polishing is used in the manufacturing pro-

8.2 Single-Layer Material Characterization

199

cess). However, the accuracy of this measurement is limited by over-etch into the underlying layers during patterning. A more convenient way (that can also be used if the dielectric layer is unpatterned) is to use ellipsometry. This measurement system is contactless, well localized (spot size <10 μm in diameter), and can be performed on process control wafers as well as directly on productive wafers if a proper underlayer provides sufficient reflectivity. The method itself extracts the optical thickness by measuring the change in polarization of a light beam that is reflected off the dielectric layer. Due to the measurement principle, it can simultaneously extract the index of refraction of the dielectric. In modern semiconductor manufacturing, the index of refraction is used as a process control monitor for the thin film quality and can therefore be stabilized within very narrow limits: whenever the measured value is out of the specified bounds, the deposition tool is inspected (and if necessary reconditioned), and the process is adjusted until the index of refraction is back within the spec limits again. A certain drawback of all methods described above, however, is that neither the geometrical nor the optical thickness is really the relevant physical quantity for characterizing the BAW layer stack. As stated earlier, it is in strict sense the ratio of geometrical thickness and longitudinal sound velocity, and the acoustic impedance. A method that directly determines these acoustic properties is picosecond laser pulse probing [3]. With this method, an ultra-short laser pulse is divided by a beam splitter into a first strong excitation pulse, and then a second, much weaker probe pulse. The first pulse induces a thermal stress wave in the material under investigation. The second pulse hits the same spot with a certain time delay and thereby tests if the shock wave from the initial excitation has returned to the surface after being reflected from the bottom of the investigated layer. By sweeping the time delay of the probe beam, a measurement signal is generated that shows a multitude of “echoes” of the initial excitation from interface reflections in the stack underneath. The delays of these echoes are proportional to the thickness of the various layers in the probed stack. Since the initial stress wave is generated via the heating of conduction electrons by the excitation pulse, the laser cannot directly induce a stress wave in the dielectric layer itself. However, the sensitivity of the measurement system is high enough to resolve several reflections from a multilayer stack. Metallic layers that are deposited in a BAW process can therefore be used to probe the dielectric layers above or underneath of them as well. As stated above, the laser pulse method directly measures the acoustic delay τ = t/vL and therefore indirectly also measures a material property, namely the (longitudinal) sound velocity vL. It is interesting to know, that also the second important material property, the acoustic impedance ZL, can be extracted from the acoustic delay with one additional measurement of the weight of the deposited layer [4]. It can be calculated by Z L = ρ ⋅ v L = ( ρ ⋅ t ) ⋅ ( v L t ) = ( w A) τ

where ρ is the mass density of the thin film material and w is the weight of the layer on a wafer with area A. The weight can be directly measured by weighing the wafer before and after the deposition with a precision scale.

200

Characterization of BAW Devices

The just-mentioned weighing is yet another way for characterizing the layer thickness. However, in this case the density of the layer needs to be known in order to extract the thickness from the measured total weight of the thin film. A severe disadvantage of this method, however, is that it is completely delocalized: it only gives the mean value over the whole wafer. In the case of piezoelectrics, the thickness is only one of many important properties that determine the performance of the BAW resonator. Others are the orientation of the piezoelectric axis, the dielectric constant, and the crystal structure. Even though these properties can be characterized by various sophisticated methods like X-ray diffraction (XRD), SEM, or TEM investigation, a direct extraction of the relevant piezoelectric tensor component (which ultimately determines the effective coupling coefficient) is not possible. It is therefore much more practical to extract the above-mentioned material parameter from the electrical measurement of appropriate test structures. At the end of this section it is important to note that the real-world behaviour of a BAW resonator is not just determined by the delays and longitudinal acoustic impedances of the constituting layer stack. In reality, nonvertical and nonlongitudinal waves also play an important role in the physics of the device (see Chapter 3 and Section 8.3), giving rise to energy leakage and spurious responses in the electrical characteristics. The behaviour of the corresponding spurious acoustic modes is determined by the full piezoelectrical, mechanical, and dielectric tensors. A direct measurement of all tensor components is very difficult and therefore not suited for monitoring purpose during manufacturing. However, a more practical way for extracting good estimates for these material parameters will be discussed in Section 8.3.3. 8.2.3

Metallic Layers

As has been mentioned above, the most elegant and accurate way of characterizing metallic films in a BAW layer stack is to use the laser pulse probing system. The spot size is very small (approximately 10-μm diameter) and therefore this method is perfectly suited for both process monitor wafers as well as productive wafers. The thickness resolution is in the nanometer range and it is even able to resolve underlying adhesion layers of only a few tens of nanometers (at least if the change in acoustic impedance from the metal to the adhesion layer is large enough). However, such laser probing systems are very expensive. A more economical way to extract metallic layer thickness is to measure the sheet conductivity, which (up to the first order) is directly proportional to the layer thickness. The thin film conductivity is most accurately measured with a four-point resistance probe. Even though this measurement method is widely used in semiconductor manufacturing (here the sheet resistance of the on-chip wiring is of ultimate importance), it has several disadvantages for BAW devices. Firstly, this method can only be applied to unpatterned, flat surfaces and is therefore restricted to process control wafers. Secondly, the sheet conductivity is a stronger function of the material composition and crystal structure than the mechanical properties. Changes in conductivity due to small variations in grain size can accidentally be misinterpreted as thickness variations.

8.3 Laser Interferometry

201

A somewhat more robust method for determining metal layer thickness is X-ray fluorescence measurements. This method basically “counts” the number of atoms in the thin film by irradiating a predefined spot on the wafer with a relatively monochromatic X-ray beam tuned to a low-energy electronic level of the metal under investigation. The fluorescence signal from the recombination is directly proportional to the number of atoms in the irradiated spot (at least until the thickness of the layer is smaller than the attenuation length of the X-rays). A drawback of this method is—similar to the four-point probe—that it can basically just be applied to unpatterned layers. It is therefore only suited for process monitoring. Again, an economic method that can as well be used for productive wafer monitoring is the above-mentioned weighing of the wafers.

8.3

Laser Interferometry 8.3.1

Introduction

Electrical measurements are an important source to gain hints and clues about the mechanism and the basic behavior of the acoustic waves in thin film layer stacks. Nevertheless, without a possibility to observe the mechanical behavior (at least at the surface), it would be almost impossible to find out why the behavior of a certain resonator deviates from theory. A reason for such deviations is the imperfect layer deposition, with respect to layer quality, structure, and thickness. And even if the stack composition happens to be perfectly known, there are usually a few not so well-known numbers in the material parameters. Now, the most important detail which can be obtained by such measurements might be the dispersion relation of a certain layer stack. To be able to predict the mechanical and electrical response of a resonator it is necessary to know about the dispersion of the three mainly involved regions on top and around the device. A method to calculate the dispersion relation of a certain stack from laser interferometric measurements has been first developed and demonstrated by one of the authors in [5–7], and since been used and adapted by many others [8, 9]. Besides the resolution of the dispersion type, a useful application of optical measurements is the investigation of side resonances (e.g., from the leads or the pad areas). The setup briefly described in the next section, which has been developed by one of the authors during his Master’s thesis, allows us to take a closer look at the vibration of relatively small details. Due to its very high lateral resolution, the vibration of resonator parts like the border ring area (usually 1 to 10 micrometers) can be studied accurately. A more accurate description of the setup can be found in [5–7]. Other setups, allowing for comparable data acquisition capabilities can be found in [8, 9]. 8.3.2

Measurement Setup

The setup is fundamentally based on a modified Mach-Zehnder interferometer; a sketch of this construction can be seen in Figure 8.1. While the device under test (DUT) is driven by a signal source, the whole area of interest is scanned by the laser.

202

Characterization of BAW Devices

Figure 8.1

Sketch of the interferometric measurement setup, realized as part of [10].

A 532-nm laser was used as light source, providing both a higher spatial resolution and a better sensitivity than an ordinary 633-nm HeNe device. The laser beam is split up into a measurement and a reference beam. While the measurement beam hits the sample through a microscope objective, the reference beam is reflected by a mirror. Both beams are combined again by a beam splitter after they were reflected by their respective targets. At this point an interference pattern develops and is subsequently detected by a photodetector. In order to minimize losses due to beam splitters which are passed more than once by a beam, polarized light was used. In combination with polarizing beam splitters and retardation plates, it is possible to adjust the polarization of a beam, so that it passes the beam splitters in the desired directions with almost no losses. For sample translation motorized stages are used, similar to recently published systems [11, 12]. There are two major ideas realized, both improving the performance of such a measurement setup tremendously. First, all earlier published systems had a very limited measurement speed. Due to the fact, that increasing the spatial stepping speed could decrease the overall measurement time only by a factor of 2 or 3, another approach was chosen. A major disadvantage of other systems is that they are limited to one frequency per measurement cycle. Therefore, the possibility to acquire more frequencies within one spatial scan was investigated. Instead of using a spectrum analyzer for photocurrent detection, like in other setups [11, 12], a network analyzer has been used. This configuration has two advantages: First, the possibility of frequency-swept measurements was provided, which might be also achieved with a sweep generator coupled to a spectrum analyzer. The second, major advantage, was that if the network analyzer drives the sample with its own source, it is able to determine the phase of the measured signal. Therefore one

8.3 Laser Interferometry

203

should be able to decide whether the surface displacement is directed up or down at a designated point. And, this automatically leads us to the second novelty. This measurement technique implicates that the optical setup has to be very stable to avoid changes in signal due to varying path lengths causing a change of both sensitivity and phase response. This means that the biasing point of the optical systems has to be fixed exactly at a path difference of π/2 or 3π/2 due to the maximum slope of the system’s response at these points. To get the phase information without any uncertainty, it has to be known to which one of these two the system is locked. To satisfy the above demands, a phase-locking system was developed, based on the fact that a sine-shaped artificial interference signal would produce a second harmonic proportional to the systems deviation from the ideal π/2 to 3π/2-point, respectively. Thus this signal is well-suited for a feedback loop, controlling the length of one interferometer arm. This can be achieved, for example, by a piezo mirror as done in this work. As a source for the artificial modulation, a phase-modulator crystal is used. In addition to the phase control enabled by the second harmonic as an additional benefit, one can use the signal’s first harmonic to correct for the sample’s surface reflectivity, out-of-focus losses, and optical misalignment during a scan. Therefore deviations from ideal conditions can be compensated, providing consistent results even with long scan times. If this signal is used to calibrate the amplitude of the signal produced by the sample vibration, it is possible to calculate the absolute surface deflection, because the signal produced by the phase modulator is known. The vibration amplitudes of the sample surface are small compared to the laser wavelength, thus they cause almost linear response in light intensity and the photocurrent. Of course this is only true for a system biased at the π/2- or the 3π/2-point, as described earlier. With this method, the possible accuracy for vertical surface deflections of a DUT is in the region of a few picometers, depending on the desired measurement speed and the systems signal-to-noise ratio. For resolution tests, we acquired surface details of one sample with different lateral step resolutions. The results revealed a lateral resolution limit of approximately 250 nm. With the measured values for amplitude and phase of the photocurrent compared to the driving source signal, the surface deflection can be reconstructed at every point of the scanned area. The vibration of the surface can then be visualized in a movie, or studied during a frequency sweep. A sample measurement of a resonator at a certain frequency can be seen in Figure 8.2. 8.3.3

Evaluation of Dispersion

One can gain even more information through closer investigation of the spatial amplitude distribution over a device area with a certain layer stack composition. The idea is to perform a two-dimensional Fourier transform of the complex amplitude matrix. If this is done for several different frequencies, one can create a k-vector versus frequency diagram: the dispersion relation for the lateral components of the acoustic waves. A comparison of the lateral mode dispersion with theory [7, 13] provides valuable insights into device physics as well as crucial feedback needed for layer stack optimization. For the calculation of the dispersion curves an FFT algorithm is applied to the dataset, which yields an amplitude distribution over the two-dimensional k-space.

204

Characterization of BAW Devices

(a)

(b)

Figure 8.2 (a) Microscope image of circular-shaped resonator, and (b) interferometer measurement result.

An example can be seen in Figure 8.3, at every point the color in the diagram represents the amplitude of a wave with the corresponding wave number kx, ky. The values are then sorted with respect to the magnitude of their wave number (see Figure 8.4), represented by the distance from the diagram center. Values with the same wave number are then averaged which gives for each frequency a distribution of the wave amplitudes over the propagation constant. Additionally, a special weighting function may be applied to suppress aliasing. Obviously, aliasing effects of the lowest order would first appear as parts of a circle mirrored at the boundaries of part of the k-space consideration in the calculation. The weighting function takes this geometric fact into account in order to avoid predominant first-order aliasing effects. However, aliasing effects remain a relatively small issue since they can be identified clearly in the resulting dispersion diagrams. Their direction is in almost every case perpendicular to the “real” modes, therefore a clear visual distinction between both cases is possible. By calculating the amplitude distribution versus measured wave numbers for each frequency point in the range of interest, a complete dispersion relation diagram can be assembled. The result can be seen in Figure 8.5 and Figure 8.6. One can recognize the different vibration modes, the thickness extensional mode (TE1) is around 1,050 MHz and the thickness shear mode (TS2) is approximately at 900 MHz. The decision which kind of mode a branch represents was done in this case by comparing the results to the simulations. There it is possible to calculate the deflection amplitudes, and thus a classification of each mode is viable.

8.4

Loss Mechanisms 8.4.1

Introduction

In this section the focus will be on theoretically possible and actually verified loss mechanisms in bulk acoustic wave resonators. Both SMR and FBAR will be investigated and a qualitative view of the losses will be given.

8.4 Loss Mechanisms

205

(a)

(b)

Figure 8.3 (a) Active area deflection of a rectangular-shaped resonator, and (b) Fourier transformation of deflection data.

8.4.2

Acoustic Leakage

In general, various loss mechanisms are possible, one of the most obvious and frequently discussed ones being the loss of energy due to acoustic leakage. The SMR BAW resonator is especially prone to this kind of losses compared to the FBAR since the latter lacks an important acoustic loss path: via the Bragg reflector into the substrate.

206

Characterization of BAW Devices

Amplitude/arbitrary units

101

101

101

101

0

0.5

1.0

1.5 k/m−1

2.0

2.5

3.0 × 106

Figure 8.4 Amplitude versus propagation constant k of the surface deflection of a resonator at a certain frequency.

Figure 8.5

Measured dispersion of a 1,050-MHz resonator layer stack [10].

8.4 Loss Mechanisms

207

2.0

f/GHz

1.5

1.0

0.5

0

Figure 8.6

1

2 k/m−1

3

4 ×10

6

Measured wideband dispersion diagram of a 1,050-MHz resonator layer stack.

As discussed in Section 8.1, given a sufficient piezoelectric coupling factor, the resonator quality factor is the main performance parameter deciding about whether a certain application can be addressed with the BAW technology at hand, or not. In the case of the ladder filter topology, which makes up the major part of today’s BAW applications, the available resonator Q has its primary impact on the steepness of the passband roll-off as well as on the minimum insertion loss of a filter. 8.4.3

Acoustic Leakage Through the Bragg Reflector

As discussed in Chapter 3, SMR-BAW devices use an acoustic reflector (sometimes referred to as a Bragg reflector), to isolate the resonating device acoustically from

208

Characterization of BAW Devices

the device substrate. If the reflectivity of the Bragg reflector is not a perfect 100%, energy will be leaking in the form of acoustic waves from the active region of the SMR throughout the reflector layers into the substrate. There it will be transported away from the resonator, scattered or damped on the substrate backside or in the substrate itself due to imperfections, or reflected back into the resonator out-of-phase. In any case, since energy that leaked into the substrate cannot be recovered, this loss mechanism significantly reduces the quality factor of the resonance. [See the earlier definitions (Chapter 5) of resonator Q.] Experiments [14, 15] have shown that even a perfectly matched quarter-wavelength reflector, which should in the (simplest) theory reflect close to 100% of the inbound waves, does show a significantly lower quality factor than expected by theory. In order to get information about the energy which is possibly leaking into the substrate, a resonator with a straight-forward quarter-wavelength reflector which has been measured interferometrically beforehand, was mounted upside-down into the interferometric measurement setup. This enabled the measurement of the surface deflection on the backside of the substrate, caused by the acoustic waves leaking through the Bragg reflector. A few snapshots of the measured surface deflection are shown in Figure 8.7. Although the amplitudes on the substrate backside are approximately a factor of 10 times smaller than those at the front surface, they’re still constituting a significant contribution to the losses and results in a decrease in the resonator Q, effectively limiting the Q to the regime of <700. Now, since obviously a significant part of the wave energy is leaking despite the fact that in theory the quarter-wavelength Bragg reflector should reflect 100% of the inbound longitudinal waves, one is inevitably led to the conclusion that a nonnegligible part of the mechanical energy is stored and traveling in the form of shear waves. At the first glance, given the (naive) image of the strictly vertically moving resonator surface, this seems to be unlikely. However, on closer investigation it turns out that there are a variety of mechanisms promoting the conversion of longitudinal wave energy into shear waves. Tilting of the wave vector due to lateral mode dispersion (see Chapter 3), scattering at layer imperfections and layer roughness, and

Resonator top surface deflection at different frequencies:

(a)

Substrate backside deflection:

(b)

Figure 8.7 Resonator top-surface deflection at different frequencies (a) and corresponding substrate backside deflection patterns (b).

8.4 Loss Mechanisms

209

interface effects adhering to the simple requirements of acoustic boundary conditions on layer interfaces are among the most important ones. The idea of shear waves leaking energy throughout the acoustic Bragg reflector has been put to the test first by Infineon’s BAW group [14, 16], and in the meanwhile by several others as well. Countermeasures have been presented as well in the form of cooptimized Bragg reflectors providing good reflectivity for both, shear and longitudinal waves. Using Mason’s model to simulate the longitudinal and shear wave transitivity of a given Bragg reflector configuration, the resulting limitations to the resonator quality factor can be calculated. Figure 8.8 shows the resulting Q limits for a quarter-wavelength mirror. In contrast to that, Figure 8.9 shows the same parameters, this time calculated for a reflector which has been cooptimized for longitudinal and shear wave reflectivity. Samples with the different reflector types have been prepared, and the resonator Q has been measured for each variant respectively. Representative Smith charts of the measurements are shown in Figure 8.10. Both resonators have been fitted with a modified BVD model (see Chapter 3). The shown Q-value represents the maximum achievable Q at any point of the resonance curve in the absence of spurious modes. Note that the Q-value at the serious resonance frequency is lower than at the parallel resonance frequency, since there’s a strong influence of electrical (ohmic) losses at this point, which will be discussed briefly further down. In general, the parallel resonance Q is, given a spurious-free response, the best parameter to look at while optimizing a BAW device for acoustic losses. The Q-value of the optimized reflector resonator is twice that of the standard quarter-wavelength version. Since the transitivity of the longitudinal waves is calculated to be almost the same in both cases, and only the shear wave reflectivity changes, the result suggests that a major part of the losses in the quarter-wavelength variant has been indeed due to shear waves. Recent reports from various groups among the BAW (SMR) community [4, 16, 17] confirm that careful cooptimization of the reflector for longitudinal and shear

Figure 8.8 Calculated Q limitation due to Bragg reflector, assuming a quarter-wavelength reflector.

210

Characterization of BAW Devices

Figure 8.9 Calculated Q limitation due to Bragg reflector for a longitudinal and shear wave cooptimized reflector.

Figure 8.10 Resonator Q measurements for different reflector types. Left Smith chart shows response of a quarter-wavelength reflector resonator. Right chart shows a resonator incorporation a longitudinal and shear wave cooptimized reflector. Note that the difference is most obvious in at the parallel resonance. The quality factor of the series resonance is limited by electrical conductivity rather than acoustic losses.

waves significantly boosts the usable resonator Q. Without consideration of the shear wave reflectivity, all groups seem to hit a wall at a quality factor of about ∼700. In this case the losses seem to be dominated indeed by the acoustic reflector leakage due to shear waves. Careful reflector design on the other hand can easily boost the resonator Q up to >1,500, some groups reporting quality factors up to >2,500 [16]. It should be mentioned that this cooptimization of the Bragg reflector for shear waves goes hand-in-hand with sacrificing some piezoelectric coupling, since usually layers in the reflector close to the resonator have to be made thicker. This in turn causes a larger part of the stress field to reside outside the piezoelectric material, thus reducing the coupling.

8.4 Loss Mechanisms

211

Also, at this point it seems important to reiterate that the quality factor of a resonator is always dictated by the (at this point) dominant loss mechanism (see Chapter 3). However, most SMR manufacturers seem to be suffering mainly from shear wave losses while they roam in the region of Q < 1,000. 8.4.4

Laterally Leaking Waves

Another acoustic loss path is the energy loss by laterally leaking waves. This mechanism received quite a bit of attention since it is important for SMR as well as FBAR manufacturers. The existence of this type of losses can be verified much easier than the previously discussed losses, since it shows up immediately in interferometer measurements of SMR and FBAR resonators. However, the amplitudes of these waves are considerably smaller than the amplitudes in the active area of the resonator. The difference is more than two orders of magnitude (see Figure 8.11). It has to be pointed out that accurate laser-interferometric measurement of the wave amplitudes outside the metallized active resonator region is not straight-forward since the measurement is in most cases done looking through a dielectric (e.g., the piezo layer) onto the bottom electrode. Thus, the comparison of the mechanical deflection amplitudes is to be devoured with great care in addition to the bottom electrode deflection (e.g., a change of refractive index in the dielectric layers due to stress/strain might contribute to the measured signal as well). Experiments [14, 16] have shown that lateral wave leakage is not a dominant loss mechanism in SMRs operating in the Q-regime up to 2,500. It might very well become significant above that threshold, though. Also, it has been shown experimentally [4] that energy leaking laterally can indeed be confined by appropriate measures. However, a confinement of these waves produced additional unwanted spurious resonances due to standing waves, of course. Some FBAR groups have shown simulations suggesting scattering of wave energy at the suspension points into the substrate [18]. Methods to prevent this scattering have been suggested; however, from the authors’ point of view, FBAR resonators have up to now not been limited by lateral acoustic losses, as long as some fundamental design principles (like keeping the acoustically active region confined to the suspended area) have been followed.

Figure 8.11 laterally.

Interferometric measurement of an SMR resonator device, leaking acoustic waver

212

Characterization of BAW Devices

8.4.5

Electrical Losses

The other, straight-forward and obvious principle limiting the usable resonator Q are the electrical losses due to ohmic resistance. This can be ohmic resistance in the resonator leads from and to the electrodes as well as distribution currents in the electrodes themselves. Furthermore, once the existence of laterally standing waves is accepted, as a consequence redistribution currents within the electrodes, between areas of strong and weak local deflection must exist [19]. These currents result in ohmic losses due to the finite conductivity of the electrode material. The finite conductivity of the electrodes in general mostly impacts the quality factor at the series resonance of a resonator, since its electrical impedance at this point is low and any series resistance has a direct impact on the usable Q. In order to decrease the Q at the parallel resonance by electrical means one would have to have a current path between the two resonator electrodes. Another electric loss path, which is not so obvious at the first glance, is the creation of eddy currents in any metallized region exposed to the varying magnetic field created by the RF current driving the BAW device. The swirling current set up in the conductor is due to electrons experiencing a Lorentz force that is perpendicular to their motion. Hence, they veer to their right, or left, depending on the direction of the applied field and whether the strength of the field is increasing or declining. The resistivity of the conductor acts to damp the amplitude of the eddy currents, thus, eddy currents create losses through heating. 8.4.6

Viscoelastic Losses

The materials used in SMR or FBAR [20] devices expose both viscous and elastic behavior. What this means exactly is best illustrated in Figure 8.12, which shows how various (theoretical) types of materials behave in the time domain. For a slab of

Figure 8.12

Stress and strain versus time for different material types subject to cyclic loading.

8.4 Loss Mechanisms

213

material with a cross-sectional area, A, and a thickness, T, subject to cyclic loading, F(t), the corresponding response is given by the displacement function, x(t). The cyclic stress on the sample material is found by dividing the input load by the cross-sectional area, and the resulting cyclic strain on the material is found by dividing the displacement by the thickness. A (theoretical) purely elastic material is one in which all the energy stored in the sample during loading is returned when the load is removed. As a result, the stress and strain curves for elastic materials move completely in phase. For elastic materials, Hooke’s law applies, where the stress is proportional to the strain, and the modulus is defined at the ratio of stress to strain. A complete opposite of an elastic material would be a purely viscous material, also shown in Figure 8.12. This type of material does not return any of the energy stored during loading. All the energy is lost as “pure damping” once the load is removed. In this case, the stress is proportional to the rate of the strain, and the ratio of stress to strain rate is known as viscosity, μ. These materials have no stiffness component, only damping. Real materials do not fall into either of the above extreme classifications, they all exhibit a more or less pronounced mixture, called a viscoelastic behavior. Some of the energy stored in a viscoelastic system is recovered upon removal of the load, and the remainder is dissipated in the form of heat. The cyclic stress at a loading frequency of g is out-of-phase with the strain by some angle φ, (where 0 < φ < π/2). The angle φ is a measure of the materials damping level; the larger the angle the greater the damping. For a viscoelastic material, the modulus is represented by a complex quantity. The real part of this complex term (storage modulus, E1) relates to the elastic behavior of the material, and defines the stiffness. The imaginary component (loss modulus, E2) relates to the material’s viscous behavior, and defines the energy dissipative ability of the material. Using Hooke’s law to define the modulus for complex values, we can define the complex modulus, E* as: E* = E1 + iE 2 =

σ 0 iφ e ε0

The properties of viscoelastic materials are dependent on many parameters. They can include: frequency, temperature, dynamic strain rate, static preload, time effects such as creep and relaxation, aging, and other irreversible effects. For most parameters important for SMR or FBAR applications, like stiffness or mass density, there are a number of reliable measurement methods available [21]. In contrast, for the viscoelastic damping constants people typically use literature values. Those are mostly based on macroscopic (bulk) measurements instead of thin film values, possibly refined to a certain degree by measurements of the effects of those materials on real devices. To the knowledge of the authors, there are no reliable values for the typical SMRF/BAR materials published yet. Today’s SMR-BAW device losses have not been dominated by viscoelastic losses yet—chances are that this changes once the Q-regime above ∼3,000 is going to be explored.

214

Characterization of BAW Devices

8.4.7

Scattering Losses

Scattering losses occur on layer material imperfections and surface or interface roughness, as well as at the lateral device boundaries. The main loss mechanism is the redirection of vertically moving acoustic energy towards lateral directions. This causes the waves to leave the active resonator region and dissipate either in the device substrate or in the regions surrounding the device laterally. It can be shown that if one goes to extreme roughnesses of hundreds of nm RMS the corresponding losses will of course be visible in today’s devices. However, to the authors’ knowledge, there is no publication investigating the more subtle effects of roughness variations in the common range of several nanometers to tens of nanometers RMS showing a significant impact on the loss.

8.5

Electrical Characterization 8.5.1

Introduction

In Sections 8.2 and 8.3 we discussed several methods for characterizing the properties of single thin film layers as well as the multilayer properties of the whole SMR-FBAR layer stack. Ultimately, however, it is the electrical response of the bulk acoustic wave resonators that counts. The purpose of these measurement methods is to control and stabilize the manufacturing processes with defined limits and to provide a geometrical and acoustical characterization of the BAW device that can be correlated with the electrical response. In the next two sections we will discuss the basic aspects of the electrical response of BAW resonators and filters.

8.5.2

Resonator Measurements

As mentioned in the previous chapters, the simplest BAW device is a resonator consisting of a bottom electrode, a piezo layer, and a top electrode (and in the case of an SMR an acoustic mirror). From an electrical point of view this is a two-pole device and can therefore be characterized using a one-port scattering parameter measurement setup. A typical device geometry for this measurement is shown in Figure 8.13(a). Besides this, a resonator can also be measured in a two-port configuration [see Figure 8.13(b) for a series and Figure 8.13(c) for a shunt setup]. The advantage of this setup is that a network analyzer has typically a much better sensitivity on the transmission channel than on the reflection channel. Nevertheless, in most cases the simple one-port measurement of the resonator structure in Figure 8.13(a) is sufficient. A typical electrical response is depicted in Figure 8.14 on the Smith chart. From the S-parameters the complex resonator impedance can be calculated according to the well-known formula Z = Z ref ⋅ (1 + S 11 ) (1 − S 11 )

8.5 Electrical Characterization

215

Signal

Figure 8.13 (a–c) Typical geometries for electrical characterization of a single BAW resonator. The lighter shade shows the region of the lower (or bottom) electrode, while the darker shade corresponds to the upper (or top) electrode. Note that a via through the piezolayer is needed in order to contact the ground pad.

Figure 8.14

One-port S-parameter trace of the simple test resonator in Figure 8.13(a).

where S11 is the measured reflection coefficient and Zref is the reference impedance of the measurement setup (typically 50Ω). Figure 8.15 shows the phase and the absolute value of the impedance of the resonator measurement depicted in Figure 8.14. The important resonator properties that can be readily extracted from these plots are the resonance frequency (i.e., the minimum of the impedance curve), the separation of series and parallel resonance frequencies (proportional to the effective-coupling coefficient) and the resonator quality factors (most easily extracted from the steepness of the phase curve at the low- and high-frequency zero-crossings). Even though technically easy, a meaningful determination of these quality factors turns out to be a fairly tricky thing: first of all, the quality factor at the series resonance is very sensitive to probe needle and contact lead resistance, making it

216

Characterization of BAW Devices 10 4

Impedance (Ohms)

10 3

10 2

10 1

10 0 1850

1900

Frequency (MHz) (a)

1950

2000

1950

2000

2 1.5

Phase (rad)

1 0.5 0

−0.5 −1 −1.5 −2 1850

Figure 8.15

1900

Frequency (MHz) (b)

Impedance and phase curves of the resonator measurement shown in Figure 8.14.

strongly dependent on test structure layout and measurement quality. Secondly, the frequency points at which the quality factor is most critical (namely at the edges of the filter passband; see next section) is neither the series nor the parallel resonance frequency [22, 23]. Instead, for series resonators the position around 7 and 12 o’clock on the Smith chart are most relevant, whereas for shunt resonators, the positions around 12 and 5 o’clock are important. It is therefore most convenient to fit a lumped element model (the BVD model, or its modifications; see Section 3.1.2) to the measured response and then to use this model in the design of filters intended to be made with these resonators. The difficulty (and well-kept secret of all SMR and FBAR groups); however, lies in the definition of a proper error function for the fit algorithm. The basic strategy is to put high weights to the above-mentioned regions that correspond to the passband edges of

8.5 Electrical Characterization

217

the filter in order to extract the motional components of the BVD model. The static capacitance and other parasitic components (like series resistance and lead inductance) are more conveniently fitted off-resonant. 8.5.3

Filter Measurements

Even though there are many ways for constructing an SMRF or BAR filter (e.g., ladder-, lattice-, CRF-type filters) the common feature of all filtering applications is to transmit electrical signals in a predefined frequency band—called the passband—and to provide sufficient signal blocking outside. Again, the preferred measurement system to characterize RF filters is a network analyzer. However, due to the multiport character of a filter (at least two ports are required—one for the incoming and one for the outgoing signal), the device under test needs to be characterized by the full-scattering matrix with all its reflection and transmission components. The only simplification comes from the passive character of the filter (no signal gain within the filter). Due to this, the transmission matrix is symmetrical, meaning that the transmission from port i to port j is the same as the transmission from port j to port i. In Figure 8.16 an example of a filter attenuation curve is shown. In this case, the curve depicts the transmission from the power amplifier port to the antenna port of a US-PCS duplexer [24]. The specs are only drawn for illustrating the various regions of the plot as discussed next. The filter response consists of three regions: 1. The passband: This region has low insertion attenuation (IA). The shaded region under the curve indicates a maximum allowable limit that is typically

0 −10

Magnitude [dB]

−20 −30 −40 −50 −60 −70 −80 1750

1800

1850

1900

1950

2000

Frequency [MHz]

Figure 8.16 Transmission curve of the TX filter of a PCS duplexer [24]. The gray-shaded regions depict upper and lower limits on the filter attenuation.

218

Characterization of BAW Devices

derived from system level requirements of the radio system in which the filter is used. In general, the passband specifications have to be met over an extended range of operating temperatures (typically from −30°C to +85°C). Due to a nonzero temperature coefficient of frequency (TCF; see Section 3.3.4) some extra-bandwidth (the so-called “temperature margin”) is needed below and above the passband to accommodate for the temperature shift of the BAW resonators. Within the passband the attenuation curve is in general not flat. The difference of the smallest to the largest value is called the amplitude ripple and is typically also limited by system requirements. Another consequence of IA variations is an unwanted dispersion of electrical signals if the IA changes significantly over a single transmission channel. The strength of this effect is typically limited by a maximum group delay specification. Another very important parameter of any acoustic filter is the maximum achievable passband width. This is the frequency separation of those two points on the transmission curve, where the insertion loss drops to the maximum allowed passband attenuation level (to the left and to the right of the passband). The amount that the maximum achievable passband width is larger than the specified (or minimum) passband width can be used to compensate for manufacturing fluctuations, and therefore directly correlates with production yield. 2. The stopband: Somewhat offset from the passband, the filter is required to block incoming signals from other RF applications (otherwise they may overload the low-noise amplifier and thereby block the actual signal) or—if used as a transmit filter—to clean up the frequency spectrum of the amplified RF signal (suppression of higher harmonics and power amplifier noise). 3. The guard band: Between the passband and the stopband exists a transition region in which the filter characteristics changes from transmissive to reflective. This region of the attenuation curve is called “filter skirt” and the frequency range over which the transmission drops from the maximum passband to the minimum stopband attenuation is called the “roll-off.” It is a direct measure for both the performance of the constituting bulk acoustic wave resonators as well as for the quality and sophistication of the filter design. In the guard-bands there are strong limitations on the intended or unintended radiation of any electronic device in order to avoid interference between them. Therefore, the guard-band frequency range is in principal “wasted” transmission spectrum. There is a strong demand from mobile service providers to come up with filters that provide steeper roll-offs and thus offer filtering solution that allow for narrower guard-bands and consequently broader passbands (resulting in more transmission channels). One example for this is the recent extension of the US-PCS band (TX 1,850 to −1,910 MHz and RX 1,930 to 1,990 MHz) by 5 MHz called BC14 (TX 1850 to 1,915 MHz and RX 1,930 to 1,995 MHz). Another important property of any filter is the return loss (i.e., the fraction of the incident electromagnetic energy that is reflected back to the signal source). This parameter is only relevant in the passband. Figure 8.17 shows the reflection curves for the filter of Figure 8.16. In the stopband region, the resonators are basically

8.5 Electrical Characterization

219

Magnitude [dB]

0 −5 −10 −15 −20

1850

1900

1950

2000

Frequency [MHz]

Figure 8.17

Return loss curve of the filter shown in Figure 8.16.

capacitors and therefore the stopband attenuation is established mostly by reflection of the incident wave (return loss close to 0 dB). In the passband a well-designed filter shows only a small reflection (since it is supposed to transmit as much of the incoming signal as possible). The particular shape of the reflection curve with its several notches depends strongly on the filter topology and resonator areas.

References [1] Lakin, K. M., et al., “Improved Bulk Wave Resonator Coupling Coefficient for Wide Bandwidth Filters,” Proc. 2001 IEEE Ultrasonics Symp., Vol. 1, 2001, pp. 827–831. [2] Mutamba, K., et al., “Micromachined GaN-Based FBAR Structures for Microwave Applications,” Proc. 2006 Asia-Pacific Microwave Conference, 2006, pp. 1757–1760. [3] Antonelli, G. A., et al., “Characterization of Mechanical and Thermal Properties Using Ultrafast Optical Metrology,” MRS Bulletin, Vol. 31, 2006, pp. 607–613. [4] Kaitila, J., and G. G. Fattinger, “Measurement of Acoustical Parameters of Thin Films,” Proc. 2006 IEEE Ultrasonics Symp., 2006, pp. 464–467. [5] Fattinger, G. G., and P. T. Tikka, “Laser Measurements and Simulations of FBAR Dispersion Relation,” Proc. IEEE MTT-S Int. Microwave Symp., 2001, pp. 371–374. [6] Fattinger, G. G., and P. T. Tikka, “Modified Mach-Zehnder Laser Interferometer for Probing Bulk Acoustic Waves,” Appl. Phys. Lett., Vol. 79, No. 3, 2001, pp. 290–292. [7] Fattinger, G. G., “Acoustic Wave Phenomena in Multi-Layered Thin-Film Layer Stacks,” Ph.D. Thesis, 2000–2003, Johannes Kepler University, Linz, Austria, Infineon Technologies AG, Munich, Germany, 2001. [8] Makkonen, T., et al., “Estimating Materials Parameters in Thin-Film BAW Resonators Using Measured Dispersion Curves,” IEEE Trans. on Ultrasonics, Ferroelectrics, and Frequency Control, Vol. 51, No. 1, January 2004, p. 42. [9] Telschow, K. L., and J. D. Larson III, “Quantitative Determination of Lateral Mode Dispersion in Film Bulk Acoustic Resonators Through Laser Acoustic Imaging,” Proc. IEEE Ultrasonics Symp., 2006, pp. 448–451. [10] Fattinger, G. G., “GHz-Laservibrometer for Investigation of Vibrating Semiconductor Layers,” Master Thesis, 1999–2000, Johannes Kepler University, Linz, Austria, Infineon Technologies AG, Munich, Germany. [11] Knuuttila, J., et al., “High Resolution Laser-Interferometric Probing of SAW Devices,” Proc. IEEE Ultrasonics Symp., pp. 235–238.

220

Characterization of BAW Devices [12] Tikka, P. T., et al., “Laser Probing and FEM Modeling of Solidly Mounted Resonators,” 1999 IEEE MTT-S International Microwave Symposium Digest, Vol. 3, 1999, pp. 1373–1376. [13] Lowe, M. J., “Matrix Techniques for Modeling Ultrasonic Waves in Multilayered Media,” IEEE Trans. Ultrason., Ferroelect., Freq. Contr., Vol. 42, No. 4, 1995, pp. 525–542. [14] Marksteiner, S., et al., “Optimization of Acoustic Mirrors for Solidly Mounted BAW Resonators,” Proc. IEEE Ultrasonics Symp., 2005, pp. 329–332. [15] Fattinger, G. G., et al., “Optimization of Acoustic Dispersion for High Performance Thin Film BAW Resonators,” Proceedings of IEEE Ultrasonics Symp., 2005, pp. 1175–1178. [16] Fattinger, G., R. Aigner, and S. Marksteiner, “Everything You Always Wanted to Know About BAW,” APMC2006 Workshop Proceedings, Yokohama, Japan, December 2006. [17] Thalhammer, R., R. Aigner, “Energy Loss Mechanisms in SMR–Type BAW Devices,” Proceedings of IEEE IMS-MTT-S, 2005, pp. 225–228. [18] Link, A., et al., “Suppression of Spurious Modes in Mirror-Type Thin Film BAW Resonators Using an Appropriate Shape of the Active Area,” Proceedings of IEEE Ultrasonics Symp., 2005, pp. 1179–1182. [19] Taniguchi, S., et al., “An Air-Gap Type FBAR Filter Fabricated Using a Thin Sacrificed Layer on a Flat Substrate,” Proceedings of IEEE Ultrasonics Symp., 2005, pp. 600–603. [20] Yokoyama, T., et al., “New Electrode Material for Low-Loss and High-Q FBAR Filters,” IEEE Ultrasonics Symp., 2004, Vol. 1, August 23–27, 2004, pp. 429–432. [21] Thalhammer, R., et al., “Ohmic Effects in BAW Resonators,” IEEE MTT-S International Microwave Symposium Digest, 2006, June 11–16, 2006, pp. 390–393. [22] Ruby, R., “Overview of FBAR Filters, Duplexers, Quintplexers, and Front End Modules (FEM) at Avago,” Asia-Pacific Microwave Conference Workshops & Short Courses Digest, 2006, pp. 399–407. [23] Ruby, R., D. Feld, and R. Parker, “A Discussion of Q for FBAR Resonators,” Proc. Intl. Workshop on Piezo-Devices Based on Latest MEMS Technologies, 2007, pp. 31–40. [24] Marksteiner, S., et al., “Hybrid SAW/BAW System-in-Package Integration for Mode-Converting Duplexers,” Proc. 3rd International Symp. on Acoustic Wave Devices for Future Mobile Communication Systems, Chiba University, Japan, 2007, pp. 97–100.

CHAPTER 9

Monolithic Integration Marc-Alexandre Dubois

9.1

Introduction The thin film BAW resonators were born from the nearly simultaneous development, in the early 1980s, of technologies for etching thin silicon membranes, and for processing high-quality piezoelectric thin films at moderate temperature. Hence, instead of the traditional mechanical machining techniques used until then to make crystal resonators, micromachining technologies could be applied for the manufacturing of UHF resonators and filters. In addition to the low thermal budget of these new technologies, the compact size of the thin film BAW devices contributed to underline the possible compatibility between this emerging piezoelectric technology and the fabrication of integrated circuits (IC). So, from the very beginning of the development of the thin film BAW technology, the monolithic integration was not only seen as a possibility by most players in the field, but it was even set as a target by several development teams. The first results were disclosed after a few years by Toshiba [1] and Iowa State University [2]. One of the main driving forces in this race towards a fully integrated RF system-on-chip has been size. Indeed, each new generation of communication system is required to be smaller than its predecessor, or to bring many more functionalities in an equivalent volume. Consequently, the space reserved for the RF-passive components, and among them the resonators and filters, has been shrinking steadily. Monolithic integration of piezoelectric BAW components with active circuitry is one possible way of addressing this challenge. The gain of space can be significant: A single common package can replace two or more individual packages, depending on the number of resonators and filters in the system. And this common package can be of the same size as that of the IC alone, provided that the BAW components are built over the active part of the circuit, as opposed to reserved areas that would be void of transistors or integrated passive devices. But size reduction in itself is not the only advantage of this technology. The pad-ring of the integrated circuit becomes less crowded, since the connections to and from the filter are not located at the edge of the silicon die, but rather close to the corresponding IC blocks. Also, the fact that only one package needs to be soldered on the printed circuit board (PCB), simplifies a lot the design of the latter. Fur-

221

222

Monolithic Integration

thermore, the parasitic losses usually associated with the bonding of each component to its package are significantly reduced, as are those linked to the PCB’s interconnecting strips. The result is a potential increase in electrical performance. As a side effect from the reduction of number of packages is also the fact that the design of the complete RF system can be alleviated from some complex tasks of electromagnetic simulation. As a general rule, the input and output impedance presented by a filter in an RF system is close to 50 ohms, whatever the technology used for the manufacturing. This ensures a wide compatibility between the numerous components used in radio systems. However, this impedance level is not always the best choice in terms of performance. For example, some low noise amplifiers (LNA) exhibit very high input impedance levels, rendering the introduction of an impedance-matching network between the 50-ohm filter and the LNA mandatory. This network being composed of nonideal elements, it contributes to the overall attenuation in the signal path. The monolithic integration approach enables the engineer to choose the impedance of the filter at the design level, so that it fits better to the particular requirements of the circuit with which the filter is meant to operate. There are of course limits to the level of impedance that can be set by the BAW device itself: Impedance is linked to the size of the resonators (through the capacitance of the piezoelectric film), as are the key performances of the latter, such as the coupling coefficient and Q-factors. Nevertheless, through a clever codesign of the circuit and the filter, the impedance-matching network can sometimes be suppressed, reducing further the complexity and the size of the radio system, while increasing its performance. Monolithic integration is, however, not readily applicable to each and every communication system. The main issues are both economical and technical. First of all, many systems need to handle several different communication standards, meaning several different carrier frequencies. Each of these in turn requires BAW components with a specific center frequency. This translates into several different thickness values for the piezoelectric film, and even maybe for the surrounding layers. The complexity of processing increases thus significantly for each additional standard. The larger the number of BAW center frequencies is, the more extreme the challenge of monolithic integration becomes. Even assuming that all the technical hurdles could be solved in a satisfactory manner, the cost of such a technology would be probably high enough to prevent its usage in many multistandard applications. Another limit of this technology is related to the manufacturing yield. Since most of the processing steps, if not all, required for the BAW devices are an addition to the regular manufacturing sequence of the circuit, the fabrication yield of the monolithic system can only be smaller than that of the microelectronic circuit alone. This is all the more true that the yield mechanisms are totally different in the BAW and the microelectronic technologies. For example, the acoustic devices are very sensitive to any thickness variations, or to variations of the coupling coefficient of the piezoelectric layer, whereas the circuit is virtually unaffected by these phenomena. On the other hand, the circuit performance is limited often by electrical charges and defects in the semiconductor, which is not relevant for the resonators. Unfortunately, radio applications most often are borne on high-end IC technology platforms, which are expensive, and any reduction of manufacturing yield has to be fought seriously.

9.2 Compatibility Issues Between IC and BAW Technologies

9.2

223

Compatibility Issues Between IC and BAW Technologies Although BAW resonators and filters are processed, usually, on silicon wafers, the applied technologies are those used in the microelectromechanical systems (MEMS) world, rather than in the IC fabrication facilities. Among the differences are the need for different materials—high-resistivity silicon substrates, a piezoelectric layer and specific electrodes—fabrication tolerances either relaxed or much more stringent depending on the parameter, and possibly the wafers’ size. Consequently, it is generally not possible to share many processing steps between the circuit and the acoustic device. So the monolithic integration is often the concatenation of two manufacturing sequences, one for the electronics, followed by one for the BAW devices. Indeed, the latter has to be done at the end, because the metal electrodes of the resonators would not sustain the very high temperature under which some semiconductor processes, such as the thermal oxidation, the diffusion of doping elements, or the epitaxial growth, are performed. Completed circuits are however also very sensitive to elevated temperatures. Therefore, the overall thermal budget experienced by the IC during the post-process of the BAW devices, must be low enough to avoid any damage to the semiconductor components. This can be achieved by using magnetron sputtering for the deposition of the piezoelectric thin film and the electrodes composing the resonators. Indeed, it is possible to obtain very high-quality films with this method, and at moderate temperature levels, as described in Chapter 7. The impact of postprocessing can be controlled by measuring, before and after manufacturing the BAW devices, the IC parameters that might be affected by the numerous thermal cycles applied to the wafer. The following example of such a study has been obtained in the case of post-processing of FBAR above BiCMOS 0.35-μm SiGe wafers from AMI Semiconductor, in Belgium. The BAW process is described in Section 9.3.1. In that particular case, about 50 different test parameters characterizing globally the performance of the semiconductor technology have been monitored. Table 9.1 shows an excerpt of this measurement campaign, with only a few relevant parameters. Values measured before and after post-processing are given, together with their relative difference. The stable breakdown voltage Vbd of the transistors shows that there is no significant alteration of the gate oxide integrity. The MOS threshold voltage Vt0 is like-

Table 9.1

Impact of BAW Postprocess on BiCMOS Technology Unit

Before

After

Difference

NMOS 10x0.35 Vbd at 10 nA

V

9.4682

9.5689

1.06%

202857 PMOS 10x0.35 Vbd at 10 nA

V

−8.6168

−8.6625

0.53%

NMOS 10x0.35 Vt0

V

0.60358

0.59291

−1.77%

PMOS 10x0.35 Vt0

V

−0.63017

−0.6334

0.51%

M3/M2 chain of vias

Ω/via

0.84329

0.85834

1.78%

M2 resistance

mΩ/sq

48.498

48.805

0.63%

HIPO resistor W100L10

Ω/sq

1,090.4

1,073.52

−1.55%

224

Monolithic Integration

wise not affected. The resistance measurements indicate that the thermal cycles experienced by the circuits do not lead to any intermetallic formation in the interconnections, since the resistive paths are nearly unchanged. In general, no major deviation due to the post-process has been observed in any measured parameter. All differences are below 2% except for two leakage current measurements, which seem actually to have been improved by the BAW processing. This shows that the impact of this particular BAW fabrication sequence upon the BiCMOS circuits is very limited. Of course, the influence of the active circuits on the performance of the acoustic devices should be minimized as well. The different prototypes of monolithic integration published to date show that the resonators and filters manufactured on IC wafers can perform very well. Nevertheless, these studies have all in common the fact that the BAW devices are placed over a die area which is void from any transistor, connecting line, or passive component. So the exact impact of an operating circuit on the acoustical performance of a BAW device that is set directly above has not been reported. Another important issue for the monolithic approach is linked to the substrates: The type and size of wafers are normally set by the IC technology. The BAW technology has been carried from the small size wafers of the beginning, up to 200-mm Si wafers. However, it is unlikely that equipment manufacturers will develop the necessary tools to move to larger substrates. This will prevent the monolithic approach from taking advantage of some of the most advanced IC technologies, available only on 300 mm. Furthermore, stand-alone BAW components perform better on high-resistivity wafers, because of the reduced electrical losses inherent to the capacitive coupling to the substrate. But the usual resistivity of the wafers used in microelectronic manufacturing is in the 10–20 Ω cm range, and it cannot be changed for the monolithic integration with BAW devices, because bulk resistivity is a key parameter for the IC performance.

9.3

Practical Implementation Although the monolithic integration concept had been recognized very early as an asset of the emerging BAW technology, only a limited number of successful developments have been published over the years. The first prototype was announced by Toshiba in 1987 [1]. The chip was a 1-mm2 Colpitts oscillator operating at 400 MHz, based on silicon bipolar transistors and a BAW resonator. The latter was a membrane comprising ZnO as the piezoelectric medium, Au electrodes, and two layers of SiO2 responsible for stabilizing the deformation of the diaphragm by reducing the internal stress. This membrane was fabricated by surface micromachining, with a ZnO sacrificial layer. Owing to the SiO2 layers, the overall TCF was kept below 5 ppm/°C. The main features of the complete oscillator were a consumption of 2.5 mA at a 6V source voltage, for an output power of −19.4 dB, and a carrier-to-noise ratio of 90 dB at 20-kHz offset. The same year, a team at the Microelectronics Research Center of Iowa State University disclosed the successful monolithic fabrication of a 257-MHz Pierce

9.3 Practical Implementation

225

oscillator based on a Si-ZnO resonator and bipolar junction transistors [2]. The membrane of the resonator, a p Si layer topped by aluminum electrodes and a ZnO piezoelectric film, was obtained by electrochemical etching of the Si substrate from the backside. Phase noise was better than −90 dBc/Hz at 1-kHz offset, and temperature stability was −8.5 ppm/°C. This partial compensation of the thermal drift was attributed to the Si p layer. The same team described later a technology, also based on silicon bulk micromachining, to integrate AlN resonators with bipolar junction transistors [3]. An interesting feature of these developments was the use of the same metal layers for the electrodes of the resonator as well as the circuit interconnects. This technology was applied to cointegrate 1.18-GHz resonators with bipolar transistors [4]. A complete filter was successfully embedded with an integrated circuit by TRW Electronic Systems Group [5]. This prototype was a stacked crystal filter made of two AlN layers, connected to a GaAs amplifier based on hetero-junction bipolar transistors. The membrane was obtained through the bulk micromachining of the GaAs substrate. The 1.6 × 3.2 mm2 circuit operated at 1 GHz, with a gain of 14 dB at 12V. During the following decade, no other results about monolithic integration of BAW devices were published, probably because the attention of researchers was mainly focused on the industrialization of the piezoelectric BAW technology. Much more recently, monolithic integration has been given new attention, mainly in Europe. Infineon has described a process flow for manufacturing AlN solidly mounted BAW devices over bipolar RF circuits [6]. A particular feature of this technology is the modification of the IC back-end process flow: a silicon nitride layer is buried in the oxide surrounding the metal lines of the circuit. This silicon nitride acts as an etch stop when cavities are carved in the oxide to embed the acoustic mirror. The fact that the acoustic mirror is partly buried eases the interconnection between the IC and the BAW electrodes, since the latter are nearly at the same level as the topmost IC metal layer. The main limit of this technology is that it requires that the die area covered by the mirror be free from any metal routing lines, and hence probably from any transistors. The potential gain in size offered by monolithic integration is therefore somewhat limited. A consortium of European research centers, universities, and industrial companies have teamed up in 2002, and for three years, in order to evaluate the potential of monolithic integration of FBAR systems, for applications at 2.14 and 5.5 GHz. The IC technology used in this research effort were the BiCMOS 0.25-μm SiGe:C technology from ST Microelectronics, and the BiCMOS 0.35-μm SiGe technology from AMI Semiconductor. The BAW process was developed by CSEM and CEA-LETI. It is described in more details hereafter. 9.3.1

Technology Description

The FBAR process flow developed by CSEM and CEA-LETI is based on AlN grown on platinum (Pt). It has been shown that the Pt electrode promotes efficiently the growth of AlN films with excellent piezoelectric properties. This is due to the hexagonal symmetry of the (111)-plane of Pt that matches the (002)-plane of AlN, and to an extremely smooth surface [7, 8]. The drawback of Pt is its lower electrical con-

226

Monolithic Integration

duction compared to aluminium or molybdenum, which is one of the limiting factors for the series Q-factor of the BAW resonator. The crystalline properties of AlN films deposited with this process have been assessed by X-ray diffraction measurements, yielding a very narrow rocking curve FWHM of 1.07° for the (002) peak. This high crystalline quality has been confirmed by a direct measurement of the piezoelectric d33,f extensional coefficient with a double beam Mach-Zehnder interferometer. A value of 5.3±0.22 pm/V has been obtained, representing a potential coupling coeffsicient keff2 in BAW resonators larger than 6.5%. The fabrication sequence of the BAW resonators and filters above IC is the following: The BiCMOS process is first completed until the last-but-one step. The circuits are covered by the passivation layer, but the latter is not yet patterned, so that the metal pads are not open. Then, a TEOS oxide is deposited and polished by CMP, in order to get a smooth and flat surface for the FBAR fabrication. A 1-μm photoresist sacrificial layer is deposited, patterned, and cured at high temperature for defining the position of each resonator. It is then protected by 0.5-μm of silicon nitride. Next, the active part of the devices is manufactured, with the subsequent deposition and patterning of the Pt bottom electrode (0.2 μm), the piezoelectric AlN layer (1.36 μm), and the Al top electrode (0.2 μm). A protective oxide is deposited and patterned. Via holes are then etched through the different dielectric layers until the last metal level of the integrated circuit, and a thick metal interconnect is deposited and patterned to link the BAW resonator to the IC. This interconnect is made of Ti and Al (0.2 μm and 2 μm, respectively). A thin silicon oxide-loading layer is then deposited and patterned, to shift down the frequency of selected resonators. This is required for building lattice or ladder filters. The thickness of the loading layer depends on the design, but is in the range of 80 nm. Finally, holes are realized around the resonators, which enable the removal of the sacrificial layer by a dry-etch process, thus releasing the membranes. The thickness values given above are valid for a 2.14-GHz operation. At 5.5 GHz, the technology is nearly identical, however with thinner layers. A schematic cross-section of a FBAR connected to the last metal level of an IC wafer is shown in Figure 9.1. The FBAR technology has been kept as simple as possible for these proof-of-concept integrations. Consequently, additional processing steps that would be required for large-scale production, such as passivating the devices for example, have been omitted. Because of the importance of the resistive losses on the Q-factor of the resonators, as well as on the IL of the filters, the quality of the metal interface between the Air gap

Electrodes Loading

Metal interconnect

Metal interconnect

SiO 2

M5 BiCMOS wafer

M5 Passivation

Figure 9.1 Cross-section of an FBAR integrated above IC. The holes used for releasing the membrane are not shown.

9.3 Practical Implementation

227

interconnections and the IC metal layer is extremely important. In particular, any metal oxide naturally present at the surface of the Al contacts of the IC needs to be removed prior to sputtering the thick Ti-Al interconnect layer. Figure 9.2 shows a close-up view of such an interface. The bottom level of the Ti layer appears slightly lower than the TEOS/M5 original interface, which is an indicator of an effective metal oxide removal. To confirm it, a test structure comprising two IC contact pads, connected to each other through a line of FBAR interconnection metal, has been measured. After pad de-embedding, the IL of the test structure has been compared to a simple R-L series model, where the values of the lumped elements have been calculated with an electromagnetic simulator. The close match between model and measurement has assessed that the interfacing of the IC and FBAR technologies does not add any significant resistance in the signal path, and hence should not degrade the Q-factor of the resonators, or the insertion loss of the filters [9]. This technology has enabled the fabrication of 50Ω FBARs with a coupling coefficient keff2 and a Q-factor higher than 6.5% and 900 at 2.14 GHz, respectively. At 5.5 GHz, typical values for 50Ω test resonators are 6.6% and 750. If the coupling coefficients confirm the high-quality of the AlN layer, the Q-factors could be further improved with a better thickness and acoustic impedance ratio between the electrodes and the piezoelectric film and the suppression of the lateral modes propagating in the membrane. The latter can be seen as wavelets on the 5.5-GHz fundamental resonance circle of a test rectangular FBAR in Figure 9.3. The large circle is the fundamental resonance, whereas the two smaller ones are harmonics at 12.5 and 17.8 GHz, respectively. 9.3.2

Filtering LNA

This FBAR process has been applied to realize a differential low noise amplifier (LNA) that comprises two differential broadband current feedback amplifiers connected on each side of a BAW double-lattice filter [9, 10]. The measured perfor-

Figure 9.2

Cross-section (FIB) of metal interface between IC and BAW device.

Monolithic Integration

S 11

228

Frequency [10 MHz–20 GHz]

Figure 9.3

Smith chart of a test FBAR at 5.5 GHz.

mance of the filter by itself is shown in Figure 9.4, with IL level at −3 dB in the 60-MHz passband, very steep skirts, and an out-of-band rejection lower than −50 dB over a large frequency range. This extreme attenuation could be obtained in spite of the 15-Ωcm Si substrate, by introducing a grounded metal shield under the filter. The latter is made from a metal layer from the BiCMOS technology, and reduces the input-output cross-talk by more than 20 dB. The architecture of the filtering LNA and a picture of the fully processed chip are shown in Figure 9.5. The summarized performances of this LNA are a power gain of 21 dB at 2.7-V supply voltage, an input gain compression at 1 dB of −26 dBm (improved to −15 dBm if the power gain is limited to 16 dB), and a noise figure of 3 dB. The BAW filter is responsible for the out-of-band selectivity larger than 50 dB. The active circuit has been implemented in the BiCMOS 0.25-μm SiGe:C technology from ST Microelectronics. 9.3.3

WCDMA RF Front-End

Another and even more complex circuit integrated with this monolithic FBAR technology has been the receiver part of an RF front-end set at 2.14 GHz. An initial version of this front-end circuit contained a low noise amplifier, a single-to-differential converter, a high rejection double-lattice FBAR filter, a matching network, and a mixer. For the characterization of the circuit, the differential mixer was fed by an external signal generator [10, 11]. The inclusion of a differential FBAR-based VCO in this receiver chain has led to the circuit shown in Figure 9.6. The signal coming from the antenna must be fed to the second pad from the left on the top of the chip. The VCO is based on a two-transistor loop structure and exploits two resonators: one is a parallel LC tank and the other is a FBAR in its series-resonant mode

9.3 Practical Implementation

229

0 −10 −20

S21 [dB]

−30 −40 −50 −60 −70 −80 −90 −100 0.0

0.5

1.0

1.5

2.0

2.5

3.0

3.5

4.0

Frequency [GHz] 0

0

−2

−4

−4

−8 −12

−8

−16

−10

S22 [dB]

S21 [dB]

−6

−20

−12 −14

−24

−16

−28

−18

−32

−20 2.00

2.05

2.10

2.15

2.20

2.25

2.30

Frequency [GHz]

Figure 9.4

Measured S-parameters of a double-lattice filter.

[12]. Figure 9.7 shows the measured phase noise of the FBAR VCO compared to the noise of a reference VCO, in which the FBAR is replaced by a second LC tank. The phase noise has been significantly improved by using the FBAR, with the best phase noise being 143.7 dBc/Hz at 1-MHz offset from the carrier. A specific feature of this oscillator architecture is its extended tuning range, compared to usual configurations using piezoelectric resonators with low coupling coefficients. The tuning range of this differential VCO has been measured at 15 MHz. The single-ended version of the same oscillator has even shown a 37-MHz tuning range, corresponding to 1.8%. It is not yet sufficient to cover the whole frequency band as required by the WCDMA standard, but it would be enough for GPS applications.

230

Monolithic Integration

OUT +

100 Ω Load

OUT −

BAW filter 1st LNA IN +

IN −

2nd LNA Linearity control

Figure 9.5 CEA-LETI).

Schematic view and photograph of the filtering LNA (IC design by P. Vincent,

(a)

(b)

(c)

Figure 9.6 Chip micrograph (a) of an integrated WCDMA receiver and SEM inserts of the BAW filter (b), and resonator (c). (IC design by J.-F. Carpentier, ST Microelectronics, and K. Östman, Tampere University.)

The functionality of the monolithic WCDMA receiver is shown by Figure 9.8, which represents the spectrum measured at the output of the mixer with a −40-dBm signal at 2.1138 GHz fed at the LNA input. The supply voltage and VCO control voltage have both been set at 2.4V.

9.3 Practical Implementation

231

−70 −80

Phase noise [dBc/Hz]

−90

LV VCO

−100 −110 −120 −130 −140

FBAR VCO

−150 −160

10

100

1,000

10,000

Offset [kHz]

Figure 9.7 VCO.

Phase-noise measurement of the FBAR-differential VCO compared to a reference LC

0 −10 −20 −30 −40 −50 −60 −70 −80 −90 −100 Center 1.132551683 MHz

179.982 kHz

Span 1.79982 MHz

Figure 9.8 Output spectrum of the monolithic WCDMA receiver, with the demodulated output at 1.13 MHz.

232

Monolithic Integration

9.3.4

WLAN Oscillator

Oscillators have been designed for operating at 5.5 GHz [13, 14]. As an example, the schematic of a Colpitts oscillator is shown in Figure 9.9, together with a micrograph of the chip. The core of the oscillator is a common collector transistor T1, with the feedback capacitors C1 and C2 ensuring the negative resistance necessary to compensate the losses in the resonating FBAR. Transistor T2 is used as a buffer to isolate the resonator from the load impedance. The active circuit has been implemented in the BiCMOS 0.35-μm SiGe technology from AMI Semiconductor. The active circuit and the ground line have intentionally been kept away from the resonator to prevent any possible coupling with the FBAR. However, the silicon area of the oscillator could be reduced significantly by placing the FBAR over the active elements. In that case, a careful shielding of the sensitive part would be needed. The output power of the oscillator is −8.4dBm for a total current consumption of 4.7 mA at 2.7V, out of which 3 mA are drawn by the buffer amplifier. Figure 9.10 shows the output power spectrum of the circuit. A phase noise of −117.7 dBc/Hz has been measured at 100-kHz offset from the carrier.

9.4

Conclusion The few examples described in this chapter show that BAW processing is compatible with advanced IC technologies, and hence enables the cointegration of RF high Q passive and active devices on a single chip. Many circuit blocks such as LNAs or VCOs can take advantage from such a cointegration with high Q BAW devices. The performances of advanced radio systems could hence be further enhanced through the reduction of size, the limitation of interconnection parasitics, the simplification of packaging, or other advantages inherent to that technology. However, the complexity of the monolithic integration will certainly limit the fabrication yield in a production environment. So the validity of this cointegration scheme is probably restricted to niche markets and high-end applications, where the RF performance or extreme miniaturization aspects outweigh the cost issues. Vcc

Rb T1 FBAR

C1

Cc T2

C2

I

Vout

I1

Figure 9.9 Chip micrograph (640 × 650 μm ) and schematic of the oscillator (IC design by M. Aissi, LAAS-CNRS). 2

Acknowledgements

233 0 −10 −20

Power (dBm)

−30 −40 −50 −60 −70 −80 −90 5.4675

5.4676

5.4677

5.4678

5.4679

Frequency (GHz)

Figure 9.10

Output power spectrum of Colpitts oscillator.

Acknowledgements The author is indebted to all actors of the MARTINA European Consortium, and especially to Christophe Billard, Guy Parat, Pierre Vincent, Jean-François Carpentier, Mohammed Aissi, Kim Ostman, and Hocine Ziad.

References [1] [2] [3]

[4]

[5] [6] [7]

[8] [9]

Satoh, H., et al., “A 400 MHz One-Chip Oscillator Using an Air-Gap Thin Film Resonator,” Proc. of IEEE Ultrasonics Symp., 1987, pp. 363–368. Burkland, W. A., et al., “A Thin-Film Bulk-Acoustic-Wave Resonator-Controlled Oscillator on Silicon,” IEEE Electron Device Letters, Vol. 8, No. 11, 1987, pp. 531–533. Weber, R. J., S .G. Burns, and S. D. Braymen, “A Semiconductor Process for Cointegration of BAW Thin-Film Piezoelectrics with Microwave BJTS,” Proc. of IEEE Ultrasonics Symp., Honolulu, HI, December 4–7, 1990, pp. 525–528. Burns. S. G., R. J. Weber, and S. D. Braymen, “High Frequency Oscillators Using Cointegrated BAW Thin-Film Piezoelectrics with Microwave BJTS,” Proc. of IEEE Frequency Control Symp., Los Angeles, CA, May 29–31, 1991, pp. 207–211. Cushman, D., et al., “SBAR Filter Monolithically Integrated with HBT Amplifier,” Proc. of IEEE Ultrasonics Symp., Honolulu, HI, December 4–7, 1990, pp. 519–524. Elbrecht, L., et al., “Integration of Bulk Acoustic Wave Filters: Concepts and Trends,” IEEE MTT-S Digest, Fort Worth, TX, June 7–12, 2004, pp. 395–398. Dubois, M. -A., and P. Muralt, “Stress and Piezoelectric Properties of Aluminium Nitride Thin Films Deposited on Metal Electrodes by Pulsed Direct Current Reactive Sputtering,” J. of Appl. Phys., Vol. 89, 2001, pp. 6389–6395. Loebl, H. P., et al., “Piezoelectric Thin AlN Films for Bulk Acoustic Wave (BAW) Resonators,” Materials Chemistry and Physics, Vol. 79, 2003, pp. 143–146. Dubois, M. -A., et al., “Integration of High-Q BAW Resonators and Filters Above IC,” Proc. of IEEE International Solid-State Circuits Conference, San Francisco, CA, February 6–10, 2005, pp. 392–393.

234

Monolithic Integration [10] Dubois, M.-A., et al., “Monolithic Above-IC Resonator Technology for Integrated Architectures in Mobile and Wireless Communication,” IEEE Journal of Solid State Circuits, Vol. 41, 2006, pp. 7–16. [11] Carpentier, J.-F, et al., “SiGe:C BiCMOS WCDMA Zero-IF RF Front-End Using an Above-IC BAW Filter,” Proc. of IEEE International Solid-State Circuits Conference, San Francisco, CA, February 6–10, 2005, pp. 394–395. [12] Östman, K. B., et al., “Novel VCO Architecture Using Series Above-IC FBAR and Parallel LC Resonance,” IEEE Journal of Solid State Circuits, Vol. 41, 2006, pp. 2248–2256. [13] Aissi, M., et al., “A 5.4GHz 0.35µm BiCMOS FBAR Resonator Oscillator in Above-IC Technology,” Proc. of IEEE International Solid-State Circuits Conference, San Francisco, CA, February 6–9, 2006, pp. 1228–1235. [14] Aissi, M., et al., “A 5 GHz Above-IC FBAR Low Phase Noise Balanced Oscillator,” Proc. of IEEE RFIC Symp., San Francisco, CA, June 11–13, 2006.

CHAPTER 10

System-in-Package Integration A. Bart Smolders, Jan-Willem Lobeek, and Nick J. Pulsford

10.1

Introduction The purpose of this chapter is to show how bulk acoustic wave (BAW) devices can be used as a key component within a wireless system. First, we will discuss the trends in front-end integration in wireless applications. It will become clear that there is a strong need for advanced system-in-package (SiP) technologies. The BAW devices need to be compatible with this technology. System integration can be done in several ways with varying miniaturization levels. The most common way is to provide a reference design for the application. A reference design contains the layout and assembly of the PCB and corresponding software drivers. The idea is that the customer exactly copies the reference design into his application, so that all critical design items have been taken care of. A reference design is normally not intended to obtain smallest size and/or lowest cost. More advanced methods of system integration are SiPs or modules, and system-onchips (SoCs). Both concepts will be discussed in this chapter. However, the emphasis will be on SiP technology for RF applications (RF-SiP), since this fits best with BAW.

10.2

Trends in Front-End Integration for Wireless Applications 10.2.1

Multiband, Multimode Wireless Systems

Advanced system integration has been, and will remain, as one of the key trends in wireless applications in the upcoming decade. This is particularly true in cellular handsets, where the number of features grows very rapidly and where the key-value drivers are size and cost. This means that a high level of integration is required in order to meet the market needs. With the growing number of wireless mobile applications, there is a strong need for a more efficient system partitioning in order to reduce the total silicon area that is used, and to increase flexibility and reuse (multimode) of various ICs. Already in today’s high-end mobile phones we see various wireless applications combined, such as GSM/EDGE (2/2.5 generation cellular), UMTS (3G cellular), Bluetooth, FM radio, GPS, and WLAN. This is illustrated in Figure 10.1, where the penetration

235

236

System-in-Package Integration Applications penetration forecast 2002–2008 (part I) 80%

NFC A-GPS

70%

FM radio

60%

Bluetooth 50%

USB

40%

WLAN TV

30% 20% 10% 0% 2002

Figure 10.1

2003

2004

2005

2006

2007

2008

Penetration rate of wireless functions in a mobile handset. (Source: NXP.)

rate of various popular wireless applications in a mobile handset is shown. For example, the penetration rate of Bluetooth is expected to be more than 65% in 2008. In today’s phones most wireless applications are added to the cellular system as “stand-alone” features, where the integration is only done on a software level. No significant sharing of functions on a hardware level is done yet. In the future mobile systems this will be different as illustrated in Figure 10.2. The future mobile platform will consist of optimized flexible digital signal processors (software-defined radio) and reconfigurable RF system-in-package radios (radio-SiP) that can be used to serve several wireless applications. As shown in the example of Figure 10.2, we will have a dedicated cellular baseband taking care of the entire signal processing that is required for 2G, 2.5G, 3G+ cellular standards. In addition, it includes a dedicated signal-processing unit taking care of all the noncellular and emerging standards (e.g., WLAN, LTE/Wimax). The radio SiPs are partly reconfigurable as well and could potentially also be used to serve several applications operating at different frequency bands (multimode/multiband). In addition, two radio-SiPs could be used for MIMO (multiple-input-multiple-output) and/or diversity purposes as well. The radios are connected to the digital modem via a high-speed serial interface. The key challenge in the route towards reconfigurable or software-defined radios will be in the front-end. Many antennas need to be connected to the multimode digital signal processor. To illustrate this, let us investigate the potential evolution of the front-end from 2007 to 2015+. Figure 10.3 shows a typical front-end of a high-end cellular handset in 2007/2008. Note that only the cellular pipe is shown here. It consists of a transceiver for the 2G/2.5G mode and another transceiver for the 3G mode. Both transceivers are connected to a single baseband processor. In the front-end we have a low-band and a high-band antenna connected to a front-end-module (FEM) for the 2G/2.5G pipe and to a power-amplifier-with-integrated-duplexer (PAiD) module for the 3G pipe. In this example, the 3G pipe only uses the high band. Further, integration of the front-end of Figure 10.3

10.2 Trends in Front-End Integration for Wireless Applications

237

3G RF Cellular BaseBand processor

2/2.5G RF

Application engine

MIPI OFDM radio 1 Connectivity and broadcast modem engine OFDM radio 2

Radio

Digital RF-BB interface

Modem

USB NFC …

Multimedia

Figure 10.2 Embedding of baseband IP in future cellular platforms. The reconfigurable radio and corresponding front-end is highly integrated using SiP technology.

Figure 10.3

Cellular front-end of a high-end multimode/band in 2007.

can be done by expanding the FEM with the PAiD and antenna switch. Advanced SiP technology will be needed to further reduce size and cost. If we look somewhat further in the future we will see additional frequency bands and features coming into the front-end. If we use a similar approach/architecture as used in the 2007 front-ends, we will see the following trends: • • •

Rapid growth of filters/duplexers; Multiple RF interfaces; Growth in number of power amplifiers;

238

System-in-Package Integration

• •

•

Need for higher integration level in the front-end due to size/cost constraints; More and broader band antennas due to new frequency bands on system enhancements like MIMO; Coexistence issues with other standards will increase the requirements on the filters and duplexers.

This would lead to the front-end partitioning in 2010+ as illustrated in Figure 10.4. The front-end would require 15 filters and 5 power amplifiers (PAs). The contribution of the filters to the total cost of a front-end will grow significantly. It is clear that there is a strong need in the market for a high-performance, small-size, low-cost filter technology as a key building block for future FEMs. It is believed that the bulk acoustic wave (BAW) technology will make this happen, since it offers high-performance and is compatible with low-cost standard IC processing. It is clear that the front-end architecture as shown in Figure 10.4 is not the most cost-effective solution, since it uses many components. Therefore, if we look somewhat further in the future alternative, front-end architectures can be expected with a high level of reuse of silicon and other blocks. An example of such a front-end is shown in Figure 10.5. This reconfigurable SiP combines a multimode transceiver with a tunable front-end. The interface to the baseband processor is realized using a low-power high-speed serial link. It is clear that such a partitioning can only be achieved when high Q “tunable” components are available for adaptive PA matching and adaptive filtering. Both BAW as well as RF-MEMS [1] are considered as potential technologies that can be used to build such tunable components. Another function that will be embedded in the future front-end is a low-power high-frequency reference oscillator based on a high Q BAW resonator. In this way, bulky and expensive crystal-based oscillators are not needed anymore. In addition, phase-noise performance will be improved significantly.

Figure 10.4 Cellular front-end of a high-end multimode/band in 2010+. Additional frequency bands and performance-enhancement features (e.g., MIMO) are introduced.

10.2 Trends in Front-End Integration for Wireless Applications

239

Figure 10.5 Reconfigurable radio in 2012+. It consists of a multimode transceiver with digital high-speed interface to the baseband processor. The front-end uses high Q tunable components for adaptive PA matching and tunable selectivity.

10.2.2

SiP Versus SoC

Over the past years SoC has been a buzzword in the semiconductor community. In a SoC, the idea is to integrate both digital and analog blocks into a single monolithic device. In wireless systems it is often not possible to achieve the required system performance with a single SoC, or it is simply too expensive. For example, a high Q BAW resonant cannot be easily integrated into a standard advanced CMOS process. However, examples in literature have been shown where BAW resonators have been integrated with (Bi)-CMOS processes [2]. So in principle it is feasible, but for high-volume production other factors are more relevant. For wireless systems it is often a better approach to put everything in one package using a system-in-package technology. SiPs provide some advantages over SoC solutions. First, a SiP offers the opportunity to integrate a complete system including all the required passive components, such as bandpass filters, baluns, decoupling capacitors, switches, and crystals. Even antennas can be integrated on a SiP [3]. This will result in the smallest possible total solution in the final application. Note that it is estimated that in a mobile telephone passive components account for more than 80% of the total component count. This is illustrated in Figure 10.6, where the large number of components (mostly passives) and functional blocks in a GSM handset can be seen determining its overall size and cost. The challenge is in the passives [4]. Second, the system performance and total cost can be optimized in a SiP by choosing the best technology mixture for each block in the architecture. For example, the digital baseband can be made in an advanced digital CMOS node (like C090/C065/C045), the active RF function can be realized in a (SiGe) BiCMOS process, and the passive components (including BAW devices) can be integrated into a low-cost silicon-based thin-film technology. This clearly has a cost-benefit, since the analog part of the system does not have to use the expensive (USD/mm2) CMOS technology. The passive functionality, such as decoupling and filtering, can be

240

System-in-Package Integration

Antenna switch and SAW filters Antenna

Baseband, memory, and power management functions

PA and power control function

Transceiver function

Figure 10.6

GSM handset circuit board after [4].

located extremely close to the active dies, providing the best possible performance up to and including mm-wave applications. The third advantage of a SiP is related to development time. We will have a relative fast time-to-market, since a SiP supports reuse of subsystems and passive building blocks. In addition, an SiP will bring down the design-in-time and risk at the end customer. To illustrate the benefit of integrating BAW devices in a SiP, we will look at a simple example of a SoC versus SiP for a Bluetooth system that uses a BAW bandpass filter to obtain the required out-of-band blocking. The system consists of a digital baseband processor, an RF transceiver (radio), a BAW bandpass filter and some matching and decoupling components. For simplicity, we will assume that the matching and decoupling components are external. The baseband processor and transceiver are made in an advanced CMOS090 process (90 nm). Typical 2007 prices are used in the cost calculation. Table 10.1 shows the cost calculation for both the SoC and the SiP case. It is assumed that each BAW resonator needs to be tuned during processing and that the final yield of the BAW devices is 80%. Typical prices are used for high-volume production. The yield of the baseband plus radio die is assumed to be 95% and the yield of the BAW filter is 80%. This example clearly shown that putting the BAW device into a SoC is not effective from a cost point of view, a cost adder of around 30% can be expected. Of course there are also technical problems when integrating BAW devices on large (12 inch) silicon wafers.

Table 10.1 Cost Comparison of SoC Versus SiP for a Bluetooth Wireless System Using a BAW Bandpass Filter Cost Type

SiP Cost (USD)

SoC Cost (USD)

Active die CMOS090 2 (baseband + radio ~10 mm )

0.65

0.78 (20% uplift for extra mask steps)

BAW filter in silicon technology 0.03 (incl. bondpads) 0.04 (Embedded) 2 ∼0.5 mm core Assembly, package and test

0.15

0.1

Yield loss

0.05 × 0.83 = 0.04

0.24 × 0.92 = 0.22

Total cost

0.87

1.14

10.3 SiP Technologies

10.3

241

SiP Technologies The technology platform choice for a SiP product development is determined both by the technical requirements of the integrated functionality and the commercial key value drivers being cost, time-to-market, and industrialization. While the technical requirements typically dictate the platform choice, the industrialization aspects have to be addressed early in the product development cycle to ensure the timely introduction of the technology platform into mass production. The technology platforms for SiP integration can be grouped according to the material class of the underlying SiP substrate, as either laminate, LTCC, or thin film as described below. 10.3.1

Laminate Platform

The first platform is based on a laminate substrate with molded overflow to form an encapsulated package. The laminate substrate has typically two, four, or six metal layers commercially available in panel form and is used for the interconnect; and for the integration of passive RF functionality such as striplines, small inductors, and coupled balun structures. The integration in laminate offers more design flexibility and better performance for some functions at high frequencies as compared to surface-mount components (SMDs). In addition, the functionality of the embedded structures is sufficiently well defined with limited RF coupling between structures for the laminate substrate to be designed by a SiP development team and manufactured commercially. The choice of the number of metal layers in the laminate is determined by the integration requirements, cost, and by thermal considerations. Power amplifier applications require a high density of thermal vias that tend to limit the number of metal layers to two or four. Radio transceiver applications require a higher integration level preferring a choice of four or six metal layers. Figure 10.7 shows a cross-section of the laminate stack of a 6-layer laminate which consists of two core layers, a prepreg layer, and two sequential build-up (SBU) layers. Micro-vias are used in both SBU layers providing a dense interconnect, whereas the other layers use more conventional buried and through-hole vias. Multiple active dies (CMOS, BiCMOS, GaAs) can be mounted on the substrate, either as a single die or in a stacked-die configuration. The interconnection from active-dies to the substrate can be either done with a conventional, low-cost 25-μm wire-bonding technology or by using a more advanced flip-chip technology with solder bumps. Most of the passive functions that are required in wireless systems can be either integrated in the laminate substrate itself or in a dedicated thin-film based passive IC. The remaining passive functions (R, L, C, and PIN diodes) can be realized using 0402-size and 0201-size SMD components. SAW filters are typically mounted in a preformed ceramic CSP. BAW filters are mounted as a wirebond bare die with a cap. There is a trend to reduce the size of both SAW and BAW packages by advanced CSP and flip-chip techniques. Molding encapsulation is applied to make the SiP complete. Figure 10.8 shows all the ingredients of a laminate SiP. The typical height of the complete package measures only 1.2 mm.

242

System-in-Package Integration

Core

Prepreg

RCC

Figure 10.7

Cross-section of a six-layer laminate stack.

Figure 10.8 Example of a laminate SiP including wirebond dies and SMD components mounted on a multilayer laminate substrate.

10.3.2

LTCC Platform

The second platform employs low temperature cofired ceramic (LTCC) as a substrate material to increase the functional integration level in the SiP substrate and reduce the package size and component count [5]. LTCC is a multilayer ceramic material for hybrid interconnection (connection of Si/GaAs devices and discrete SMDs) and passive integration (inductors, capacitors, and transmission lines). Ceramic sheets printed with Cu paste conductor patterns are laminated above each

10.3 SiP Technologies

243

other and cofired to form a ceramic plate. Electrical connection between the patterns on different sheets is made with vias. Figure 10.9 shows the construction of a five-layer LTCC substrate. The big advantage of LTCC compared to laminate is the broad choice in dielectric ceramic materials (dielectric constant 5 to 80), metal layer materials (printed resistors included), and the LTCC stack definition (layer number and thickness). This material combination can be tailored to give broader range RF-embedded functions extending the capability of laminate to capacitors, resistors, multilayer inductors, and multilayer filters which would otherwise be assembled as discrete ceramic components. In addition, the flexibility in LTCC design leads to a higher integration density of passive components and the ability to form cavities in the ceramic allows for two-sided assembly of active dies (wirebond or flip-chip) which further improves the level of miniaturization. The high complexity of the LTCC substrate makes the RF design of a LTCC-SiP product more demanding than a larger laminate-based product due to the increased RF coupling between structures and the larger number of metal layers. However, LTCC manufacturers are able to use fast prototype spins to optimize the LTCC-SiP performance and minimize unwanted parasitic effects before proceeding to production. The assembly on LTCC is more demanding than laminate due to potential warpage of the ceramic substrate plates during cofiring, though strong improvements in plate flatness and uniformity have been achieved in recent years. Both the challenges in design and assembly have lead to the main LTCC-SiP applications being driven by companies which have ceramic technology in-house, for example, Murata, Epcos, and TDK. 10.3.3

Thin Film Platform

A thin film passive integration IC technology has been developed on high-ohmic silicon and on glass substrates. The thin film technology features a low number of mask steps and uses standard back-end IC processing with relaxed lithographic resolution in order to minimize the manufacturing cost. The passive dies built from this technology can be either used as a building block for a laminate-based SiP platform as described in Section 10.3.1, or they can be used as host substrate forming a complete SiP product. The advantage of thin film processing compared to laminate or Top side conductor

Thermal via

Via

Dielectric layer

Internal conductor

Figure 10.9

Back side conductor

Cross-section of 5-layer LTCC stack.

Stacked via

244

System-in-Package Integration

LTCC is the higher lithographic resolution for metal lines and the higher capacitance density. The disadvantage is the inability to make use of the vertical direction for improving the RF performance by controlling the coupling between metal lines. In addition, an IC package is required to reliably house the thin film passive integration substrate unlike laminate or LTCC that can be mounted directly in the application. Figure 10.10 shows a cross-section of the NXP semiconductor’s PASSI process that combines three metal layers and two dielectric layers in five mask steps to form integrated inductor and capacitor structures. The high-ohmic silicon substrate (resistivity ρ > 5 kΩcm) combined with a clean interface to the first oxide layer ensures a minimal loss in the silicon substrate and thus a high-quality passive component performance at RF frequencies. The integrated capacitors are formed by the 425-nm thick SiNx dielectric layer between the lower two metal layers with a capacitance density of 145 pF/mm2. For low-frequency decoupling, the capacitance density is increased dramatically by using NXP semiconductors PICS technology which implements deep circular pit capacitors into the high-ohmic silicon substrate as shown in Figure 10.11. This is achieved by etching 20 μm deep vias into the silicon and coating the inner via surface with a layer of thin film silicon oxide and nitride. The via is closed by a layer of doped poly-silicon to complete the capacitor pit structure. Capacitance densities of 100 nF/mm2 are in production with this technique. This is significantly higher than can be achieved on LTCC. Stripline inductors are formed from the thick metal layer, which has a very high lithographic tolerance. For multiturn inductors, feed-throughs are realized with the bottom metal layer, minimizing the parasitic capacitance from the feed-through to the inductor turns. The passive integration technology is optimized to achieve a high level of RF performance for the integrated components. A high stripline inductor quality factor is achieved with values Q > 35 in the frequency range f = 900 MHz to f = 6 GHz, which covers most cellular and wireless standards. The Q-factor is significantly higher than can be achieved in a conventional IC process and similar to inductors integrated in LTCC or laminate. The capacitor RF loss, given by the equivalent series resistance ESR < 100 mΩ is extremely low compared to standard SMD components which have a typical ESR = 300 mΩ. The combination of a low RF loss with a low production tolerance is the key to achieving high-performance integrated passive circuits in applications. Because the passive integration technology leaves little tuning possibility for the designer, accurate predictive simulation tools are vital to

5 μm Al top metal

SiNx thin film capacitor

High ohmic silicon substrate

Figure 10.10

Cross-section of the NXP semiconductor’s PASSI process [6].

10.4 SiP Design

Figure 10.11

245

Cross-section of the NXP semiconductors PICS process.

minimize the number of design cycles in development. The low-loss, thin-film metal and dielectric layer stack used in passive integration technology is ideally suited to commercially available planar electromagnetic simulation software, for example Momentum or Sonnet. These tools are able to accurately and efficiently predict the passive performance of an integrated RF structure and the simulation results can be easily incorporated as S-parameter files into a circuit simulation environment, for example ADS or Spice, to fully simulate the RF-SiP performance. More on this will be explained in the next section. Figure 10.12 illustrates a SiP with double flip-chip technology in which one or more dies (e.g., BAW or an active IC) are mounted on the passive die with solder bumps. This stack of active and passive dies is then mounted into a standard IC package with a height of only 0.85 mm.

10.4

SiP Design Due to the multitechnology nature of SiP, the design is often a rather complicated matter since it requires a variety of design tools. Let us illustrate this by looking at the following example. One of the key applications for BAW filters in SiP is the power amplifier (PA) with integrated duplexer for WCDMA applications operating in the 1.9-GHz band, as illustrated in Figure 10.13. Such a product is also often called a front-end module (FEM). In this case the duplexer is integrated onto the laminate carrier of the PA. Integration of the duplexer with the PA leads to reduced size, and improved performance at the cost of design complications in terms of isolation and matching. When designing this product the following tools are needed:

246

System-in-Package Integration Passive IC

Active or BAW IC

Figure 10.12 SiP build-up using a passive thin-film IC. A double flip-chip technology with solder bumps is used in a standard IC package. Multiple (active or BAW) dies can be mounted on the passive die.

BAW interstage filter

BAW duplexer filter Antenna

PA Tx band

Filter structure with BAW resonators To receiver Rx band

Figure 10.13 RF-SiP that combines a power amplifier (PA) with integrated BAW duplexer and BAW-interstage filter operating in the 1.9-GHz band.

• • • •

Acoustic models to design and model the BAW resonators (e.g., in MATLAB); Circuit-simulators (e.g., ADS, Spectre) for initial design of filter and PA; Harmonic balance simulator (e.g., ADS) to model nonlinear behavior; Electromagnetic models (EM) to model all interconnect and other embedded structures (e.g., Momentum, Sonnet, HFSS).

Models to describe the behavior of BAW devices have already been discussed in previous chapters of this book. We will now give a short introduction into EM-based models and tools that play a very important role in SiP design for wireless applications. 10.4.1

Electromagnetic Modeling

An important element in the design of a SiP for RF applications is to have a proper model for the interconnect between the various components. The interconnect can be considered as parasitics, or in some cases as functional interconnect (e.g., an inductor embedded in the laminate substrate). The induced current flow on such metal structures can be found by solving the Maxwell equations for the electric and magnetic field.

10.4 SiP Design

247

As an example, let us take a closer look at the design of a duplexer for WCDMA applications as part of the SiP of Figure 10.13. In Figure 10.14, the complete simulation setup of a duplexer is shown using a commercial electromagnetic (EM) simulation tool. The duplexer is composed of two BAW filter dies (ICs); one for the transmit (TX) band and one for the receive (RX) band. Both dies are wire bonded and mounted onto the SiP laminate substrate. The multilayer laminate takes care of the interconnect and integrates the high-quality RF inductors (typically 1 to 3 nH) which are required to complete the wideband response of the filter. The wire bonds are approximated by strings of small rectangular boxes. There are several commercial tools available to solve for the electromagnetic fields and corresponding current distribution on the structure of Figure 10.14. As most SiP design problems are three-dimensional planar, it is usually sufficient to use EM tools like Momentum or Sonnet to solve these structures. These tools typically use the method-of-moments (MoM) to solve for Maxwells equations numerically. For true three-dimensional problem one could use for example HFSS, which uses a finite-element-method. For most designers, EM tools are considered as “black-boxes” and even as “black-magic” tools. However, some basic background information about the method that is used by these tools is required for correct interpretation of the results that come out of these tools. We will now give a very brief introduction to the MoM that is used by many commercial EM tools. For more background info, one is referred to literature [7]. The MoM is a general procedure for solving so-called integral equations. In this section we will illustrate this method using a simple example of a perfectly conducting wire (e.g., dipole antenna with radius a and length 2l), located in free-space as shown in Figure 10.15. We will show how the electrical current distribution on the wire is calculated. We will go through the following steps: •

• •

Determine integral equation based on the boundary condition for the tangential electric field for the wire-antenna problem; Expansion of the wire in small segments or so-called subdomains; Calculation of the matrix elements.

Figure 10.14 EM-simulation setup of a BAW duplexer, consisting of two BAW ICs which are connected to a laminate substrate using wirebonds.

248

System-in-Package Integration z

1

I(z)

0

y

2a

x

−1

Figure 10.15 Wire (e.g., dipole antenna) in free-space. The wire is modeled as a cylindrical disk with radius a and length 2λ. Perfect electric conductor is assumed. The excitation is done with a voltage source in the center of the wire (z = 0).

We will assume that the radius of the wire is much smaller as compared to the wavelength, that is a << λ0. Because of this, we can neglect the current on the end-surfaces of the cylinder. The resulting current distribution will now only have a r r component along the z-axis (i.e., J = J z e z ). We can formulate the boundary condition for our problem: The total tangential electric field on the perfectly conducting cylinder with radius a equals zero. In formula form this is written as: r r r r r r r r r e n × Etot ( r ) = e n × E ex ( r ) + E s ( r ) = 0,

(

)

r r ∈ S0

(10.1)

where the surface r s r S0 is the outer surface of the cylinder and where the electric fields r ex r E ( r) and E ( r ) represent the excitation field and the scattered field, respectively. e n is the normal on the metallic cylinder. The scattered field is generated by the induced electrical current on the wire. The excitation field is generated by a voltage source that is connected in the center of the wire. Since formula (10.1) is based only on the electric field, this type of equation is called electric field integral equation (EFIE). The EFIE will form the rbasis of the MoM formulation as we will show next. r The scattered field E s ( r ) can be expressed in terms of a surface integral over the r yet unknown surface current J and the so-called Green’s function of the configuration. The Green’s function is the response due to a point source. In our example of r r Figure 10.15, we can use the free-space Green’s function, represented by G( r , r0 ), r where r0 is the location of the point source. The expression for the free-space Green’s function is relatively simple and can be written in closed form [7, 8]. For more complicated structures like layered media, the Green’s function is represented by an inte-

10.4 SiP Design

249

gral that can be precomputed once for the medium, and can be reused to analyze any metal structure within the medium [9]. More detail on the Green’s functions is outside the scope of this book. Important for our example is the observation [7] rthat the relation between the scattered field and the unknown current distribution J on the wire is a linear relation and can be written in the following form: r r E s ( x , y, z ) = L J( x , y, z )

{

}

(10.2)

where L{}is a linear operator. Equation (10.1) can be solved numerically r by using the MoM. The first step is to expand the unknown current distribution J on the surface of our metal wire into so-called expansion functions according to: r r (10.3) J( x , y, z ) = ∑ I n J n ( x , y, z ) n

in r which In are the mode coefficients that we need to determine. The functions J n ( x , y , z) are called expansion or basis functions. If we want to determine an exact r solution for the current distribution J we would need to have an infinite summation, which is of course not possible in practice. Therefore, we will try to find an approximation by limiting to n = Nmax. We can choose all kinds of basis functions. Typically, we will use local basis functions or subdomain basis functions that are nonzero only over a small part of the total structure under investigation. An example of a subdomain basis function that is often used is shown in Figure 10.16, where piece-wise linear (PWL) basis functions are shown to approximate the behavior of a function. Combining (10.2) and (10.3) gives the following expression for the scattered rs r field E ( r ): r E s ( x , y, z ) =

N max

∑I n =1

n

r L J n ( x , y, z ) =

{

}

N max

∑I n =1

n

r E ns ( x , y, z )

(10.4)

Substitution of expansion (10.4) into the integral equation (10.1) gives: r ⎛ N max r ⎞ r r e n × ⎜ ∑ I n E ns ( x , y, z ) + E ex ( x , y, z )⎟ = 0 ⎝ n =1 ⎠

z0

z1

z2

z3

z4

(10.5)

z5

Figure 10.16 Piece-wise linear approximation of a function. The type of subdomain basis functions are called PWL modes in literature (piece-wise linear). In this example four PWL modes are used to approximate the function.

250

System-in-Package Integration

r on the outer surface S0 of the wire. Let us now introduce the residue R according to: r r ⎛ N max r ⎞ r R( x , y, z ) = e n × ⎜ ∑ I n E ns ( x , y, z ) + E ex ( x , y, z )⎟ ⎝ n =1 ⎠

(10.6)

This residue has to be equal zero on the entire outer surface of the wire of Figure 10.15. This condition will be relaxed somewhat. r The residue will be weighted to zero with respect to some weighting functions J m ( x , y , z) such that r r R; J m =

r

r

∫ ∫ R( x , y, z ) ⋅ J ( x , y, z )dS = 0

(10.7)

m

Sm

r for m = 1, 2, ..., Nmax, where Sm is the surface on which the weighting function J m is nonzero. Note that the set of weighting functions, also called test functions, is the same as the set of expansion functions. This particular choice is known as Galerkin’s method. Inserting (10.6) into (10.7) gives a set of linear equations: rs

N max

r

r ex

r

∑ I ∫ ∫ E ( x , y, z ) ⋅ J ( x , y, z )dS + ∫ ∫ E ( x , y, z ) ⋅ J ( x , y, z )dS = 0 n

n =1

n

m

m

Sm

(10.8)

Sm

for m = 1, 2, ..., Nmax. This set of linear equation can be written in the more compact form: N max

∑I

n

Z mn + Vmex = 0

(10.9)

n =1

for m = 1, 2, ..., Nmax. In matrix notation we get:

[Z][I] + [V

ex

] = [0]

(10.10) ex

in which the elements of the matrix [Z] and of the excitation vector [V ] are given by: r r (10.11a) Z mn = ∫ ∫ E ns ( x , y, z ) ⋅ J m ( x , y, z )dS Sm

Vmex =

r ex

r

∫ ∫ E ( x , y, z ) ⋅ J ( x , y, z )dS m

(10.11b)

Sm

The matrix [Z] contains Nmax × Nmax elements, [I] is a vector with the Nmax ex unknown mode-coefficients and [V ] is the excitation vector with Nmax elements. The matrix equation (10.10) can be solved rather easily using standard numerical routines such as those available in MATLAB. After this matrix equation has been solved we can determine the current distribution on the wire antenna of Figure 10.15 by substituting the mode-coefficient [I] in (10.3). With this current distribution we can then calculate the input impedance or S-parameters and other characteristics.

10.4 SiP Design

251

Now let us go back to the example of a duplexer for WCDMA of Figure 10.14. This structure is of course much more complicated than the simple wire of our example, but we can use exactly the same approach for determining the unknown current distribution on all metallic structures. Of course, commercially available EM solvers can be used for this. The BAW filters can be designed by cascading series and shunt resonators, as illustrated in Figure 10.13. An acoustic-model (see Chapter 3) describing the BAW resonator can be implemented into commercial available software and used to design the filter. Both the EM results as well as the acoustic-based results can be connected on circuit level using an S-parameter representation to simulate the overall performance. Figure 10.17 shows an example of the total response of the BAW duplexer using a full EM simulation as compared to a lumped-element simulation. 10.4.2

Design Methodology

The design methodology of a SiP for RF applications is typically like that indicated by the flow diagram of Figure 10.18. First, an initial design and topology will be made based on a simple lumped-element representation. In the case of BAW components in a SiP the BVD model can be used. Next, an initial layout needs to be generated that will be analyzed using a full-wave EM simulation as described in the previous section. This will require several iterations before the product will be compliant to the specifications. The final stage will be to evaluate hardware and if required to make a next iteration of the design process. 0 −5 −10 −15 −20

Transmission (dB)

−25 −30 −35 −40 −45 −50 −55 −60 −65 −70 −75 −80

2.25E9

Figure 10.17 Predicted response using EM simulation of a BAW duplexer including detailed modeling of the interconnect.

2.24E9

2.22E9

2.23E9

2.21E9

1.99E9

2.20E9

1.98E9

1.96E9

1.97E9

1.94E9

1.95E9

1.93E9

1.91E9

1.92E9

1.90E9

1.88E9

1.89E9

1.87E9

1.85E9

1.86E9

1.83E9

1.84E9

1.82E9

1.80E9

1.81E9

Frequency

252

System-in-Package Integration

Start

Define circuit requirements

Topology study

Lumped-element simulation and optimization

Layout design

EM simulation of layout including package

Measurement of hardware

End

Figure 10.18

10.5

Design flow for RF-SiP.

Test and Industrialization, Known-Good Die Concept When integrating a silicon device like a BAW filter into a SiP that is intended for high-volume production, it is very important to have a good overall control of the yield. Due to the nature of BAW devices, the yield is typically significantly lower as compared to standard digital CMOS IC processes. In standard IC processes the yield is typically well above 95%, whereas in BAW it is in the range between 80% to 90%. This means that if we would integrate without pretesting a relative low-cost BAW device together with an expensive digital baseband processor the overall yield would be determined by the BAW device and would be very low. In this way expensive SiP products that fail the final test would end up as waste. In order to prevent a low final yield of SiPs the concept of “known-good dies” (KGDs) has been introduced by the semiconductor industry. Now all devices on a silicon wafer have been RF pretested completely. The dies that do not comply to the test specification will be dotted with an ink-mark on the wafer. These failed devices will not be used during assembly of the SiP products. In this way, the overall yield of the SiP products can be made very high, comparable to the yield levels of standard digital IC and packaging technologies. Without KGDs the cost advantage of a SiP as illustrated in Table 10.1 cannot be achieved. In the case of BAW devices within the

10.6 RF-SiP Examples

253

SiP, this means that in-line trimming to come to acceptable yield levels is required. Trimming is described in Chapter 4.

10.6

RF-SiP Examples 10.6.1

General Wireless Examples

We will now take a closer look at some examples of SiP products that recently have been introduced into wireless connectivity applications (i.e., Bluetooth and WLAN). Figure 10.19 shows a typical application diagram of a Bluetooth radio. The BiCMOS transceiver IC uses a low-IF architecture in combination with open-loop modulation giving optimal performance with very high receiver sensitivity and extremely clear transmit eye-diagram. The VCO with integrated coils operates around 5 GHz. External functions include a loop filter, de-coupling, baluns, Tx/Rx switch, matching circuits and a bandpass filter. If one would build up a complete radio application with a discrete transceiver IC (packaged in a standard 5 × 5 mm2 HVQFN-package), the required board space 3 in an application will be 200 × 2 mm or more depending on the PCB technology that is used in the application. When the same functionality is integrated into a SiP, 3 the size of the total solution is reduced to 6 × 6 × 1.2 mm with the laminate plat3 form and even to 5 × 5 × 0.85 mm with a thin-film platform. This is illustrated in Figure 10.20 where a discrete solution is compared with both SiP solutions. The three Bluetooth solutions have more or less the same performance using the same transceiver IC and do not require any additional external components, except an antenna. Note that the design of the discrete solution is not trivial and may require several spins of the application board. Figure 10.21 shows the technology mixture that is used in the Bluetooth laminate-based SiP in more detail with all the passive blocks including a bandpass filter, RX/TX switch, baluns, loop filter, and supply decoupling. Note that the photos were taken before the molding process. The package has a standard HVQFN-like footprint. The loop filter, requiring extremely low leakage without any dielectric relaxation effects, is integrated on a passive die with a thin film technology and is

Decoupling network

Loop filter

Tx Balun Switch + match

RF IC

Antenna filter

Rx Balun

Figure 10.19 Typical application diagram of a Bluetooth transceiver IC. External blocks are Tx/Rx baluns, Tx/Rx switch with matching, antenna bandpass filter, loop filter, and decoupling.

254

System-in-Package Integration

(a)

(b)

(c)

Figure 10.20 Total Bluetooth radio solution with sizes of (a) discrete solution 17 × 14 × 2 mm3, 3 (b) laminate-based SiP 6 × 6 × 1.2 mm , and (c) thin-film based SiP with double flip-chip technol3 ogy 5 × 5 × 0.85 mm . All three solutions cover the application diagram as shown in Figure 10.19.

Substrate

Stacked die

RF IC

Antenna filter

Figure 10.21 LAMP-based Bluetooth SiP including all passive components. The HVQFN-package 2 measures only 6 × 6 mm . The mold encapsulation is not shown here. Total height is 1.2 mm.

connected to the transceiver IC using a stacked-die bonding technique. In this way, parasitics are very low and VCO-pulling problems are avoided. The fourth-order LC bandpass filter is also integrated into a passive die. This filter includes an additional notch near 2 GHz to comply with the coexistence specification for mobile phone applications. Another notch is put around 5 GHz to suppress the second-harmonic/VCO leakage. The baluns, Rx/Tx switch, and decoupling are realized with SMD components and by embedding some RF functionality directly (coupling coils and λ/4 strip lines) in the laminate substrate. Next step in integrating more functionality in one package is to include the baseband and clock generation into a SiP. We will describe an example of such a package for WLAN. Figure 10.22 shows a complete WLAN 802.11b system solution including a CMOS baseband, a BiCMOS radio, a BiCMOS power amplifier, crystal, diversity switch, matching circuits, loop filer, and decoupling. A discrete implementation of this total system using a packaged version of the active dies would measure more than 500 mm2, whereas this solution measures only 11 × 16 mm2.

10.6 RF-SiP Examples

255 Crystal

BiCMOS RFIC

CMOS baseband

BiCMOS PA

Antenna filter die

Figure 10.22 WLAN 802.11b-full-system solution including baseband, transceiver, PA, crystal, 2 and all required passive components. The HVQFN-like package measures only 11 × 16 mm .

2

Further size reduction to 10 × 10 mm or smaller is possible for this WLAN system by applying a stacked-die concept or by using the passive thin-film IC technology. Note that the same passive die is used as in the Bluetooth SiP to realize the bandpass filter. This reuse of building blocks is one of the important advantages of using SiPs over SoC solutions. An example of a complete silicon-based SiP is shown in Figure 10.23, where all passive functions like decoupling, PA-matching, baluns, and bandpass filter are integrated into a passive silicon technology (PICS) [10]. In Figure 10.23, a wireless transceiver IC and a PA IC are flipped on a passive IC, containing capacitors (supply decoupling, RF), inductors and resistors needed to complete the radio function. This subassembly can be packaged in a standard HVQFN-type package. The combination (or transceiver function alone) can also be reused as RF subsystem on a module substrate housing the PA, matching, filtering and RF switching functions to provide a one package RF system solution. 10.6.2

Examples Including BAW

A first example of a SiP that includes BAW is the BAW duplexer as shown in Figures 10.13 and 10.14. Details have already been discussed in Section 10.4.

Fully integrated radio function · Transceiver IC · PA IC · Complete application · Matching · Supply decoupling

Figure 10.23

Fully integrated RF system solution [10].

256

System-in-Package Integration

The second example is a BAW-based high-frequency oscillator [11]. In this case the BAW resonator is placed on top of passive silicon die using flip-chip technology. An active BiCMOS die that contains the feedback amplifier and additional control circuits is also placed on the Si-carrier. The BAW oscillator operates at 2 GHz and can replace low-frequency crystal-based oscillators. Main advantages are size, cost, and improved performance (e.g., power consumption and phase noise). Figure 10.24 shows a photograph of the three dies. In the last example a BAW filter is used to provide the required selectivity in a wideband tuner concept. Figure 10.25 shows the basic block diagram. The tuner front-end is a highly linear LNA+mixer that upconverts the input frequency to an IF frequency. The BAW filter is then used to provide the selectivity and to remove interfering signals. The back-end part of the tuner that converts the IF signals to a digital representation has now very relaxed specifications. The active circuits are integrated in a BiCMOS process. The BAW IF filter is put on top of this BiCMOS die by means of flip-chip technology. Finally, the sandwich of BiCMOS die and BAW die are mounted in a standard plastic package using wirebond techniques. Figure 10.26 shows an example of such an IF BAW filter with a relative narrow-bandwidth. This

Figure 10.24 Photo of a BAW-based high-frequency oscillator consisting of a BAW die, a BiCMOS die, and a passive silicon carrier [11]. The BAW and BiCMOS dies are flip-chipped on the passive silicon carrier.

FE IC

BE IC B A W

LO1

Figure 10.25 selectivity.

LO2

A wideband upconverter tuner using a high Q BAW filter to realize the required

0

0

−50

−50

−100 1.16

Figure 10.26 architecture.

S211 [dB]

257

S21 [dB]

10.6 RF-SiP Examples

−100

1.18 Frequency [GHz]

1.20

Narrowband BAW filter characteristic with 8-MHz bandwidth used for upconverter

up-converter concept can be used to realize low-cost high-performance TV tuners for DVB-H/T. In addition, this concept could also be a route to solve the future multimode/band problem as illustrated in Figure 10.4.

References [1] Rijks, T. G. S. M., et al., “MEMS Tunable Capacitors and Switches for RF Applications,” Proceedings of the 24th Int. Conference on Microelectronics, 2004, pp. 49–56. [2] Dubois, M. C., et al., “Above-IC FBAR Technology for WCDMA and WLAN Applications,” Proceedings of the IEEE Ultrasonics Symp., 2005, pp. 85–88. [3] Breur, H., et al., “Bluetooth Radio Module with Embedded Antenna Diversity,” Proc. of the European Microwave Conference, Munich, Germany, October 2003. [4] Smolders, A. B., et al., “RF SiP: The Next Wave for Wireless System Integration,” Proceedings of the IEEE RFIC Symp., Fort Worth, TX, June 2004, pp. 233–236. [5] Sutono, A., et al., “High-Q LTCC-Based Passive Library for Wireless System-on-Package (SOP) Module Development,” IEEE Trans. on Microwave Theory and Techniques, October 2001, pp. 1715–1724. [6] Pulsford, N. J., “Passive Integration Technology: Targeting Small Accurate RF Parts,” RF Design, November 2002, pp. 40–48. [7] Harrington, R. F., Field Computation by Method-of-Moments, New York: IEEE Press, 1993. [8] Harrington, R. F., Time Harmonic Electromagnetic Fields, New York: McGraw-Hill, 1961. [9] Kong, J. A., Electromagnetic Wave Theory, New York: John Wiley and Sons, 1986. [10] van Straten, F., et al., “Multiband Cellular RF Solutions,” IEEE Journal of Solid-State Circuits, Vol. 39, October 2004. [11] van Helmont, F., et al., “A 2 GHz Reference Oscillator Incorporating a Temperature Compensated BAW Resonator,” Proceedings of the IEEE Ultrasonics Symp., Vancouver, 2006, pp. 333–336.

Glossary Acoustic impedance (Z) Ratio of the stress to the particle velocity associating with propagating acoustic waves. In homogeneous media, it is given by ρV, where ρ is the mass density and V is the acoustic wave velocity. Aluminum nitride (AlN) Piezoelectric material with moderate electromechanical coupling for the thickness extensional vibration. High-quality AlN thin films can be deposited by the sputtering method. Antiresonance frequency (fa) Frequency where the admittance of the shunt capacitance is cancelled with that motional or series of the equivalent circuit of the resonator. It is defined as the frequency giving the reactance of the resonator is zero while the resistance is extremely large. From the modified BVD model, it is given by the formula: f a = 1 2 π L1 C1 C 0 (C1 + C 0 ). Antenna duplexer Three-port device where signals in the transmitter (Tx) band can selectively transmit between the Tx and antenna (ANT) ports while signals in the receiver (Rx) band can selectively transmit between the Rx and ANT ports. Balanced input and/or output Signal interface using two dedicated conductors to provide forward and return paths for signal. Signal is transferred as voltage difference between two conductors. Balanced ladder filter Ladder-type filter with balanced input and output interfaces. Balun Functionality or equipment to convert balanced signal to unbalanced signal or vice versa. Bulk acoustic wave (BAW) Acoustic wave propagating inside of a medium. Butterworth-Van Dyke (BVD) equivalent circuit Equivalent circuit consisting of series elements L1, C1, and R1 in parallel with C0, where L1, C1, and R1 represent the motional inductance, capacitance and resistance respectively and C0 the shunt capacitance. Butterworth-Van Dyke (BVD) model Equivalent to Butterworth-Van Dyke (BVD) equivalent circuit. Capacitance ratio ( or r) Ratio of the shunt capacitance to the motional capacitances for BAW resonators. This indicates weakness of the electromechanical coupling. When the modified BVD model is applicable, γ is given by the formula: γ = C 0 C1 =

{(f

p

fs

)

2

}

−1

−1

, where C0 and C1 are the shunt and motional

capacitances, respectively, and fp and fs are the parallel and series resonance frequencies, respectively. Common signal In-phase components appearing in the balanced interface.

259

260

Glossary

Coupled resonator filter (CRF) BAW filter composed of multiple resonators acoustically coupled to each other. It supports multiple resonances synthesizing the passband shape. Differential input and/or output Equivalent to balanced input and/or output. Dissipation factor (D) Inverse of quality factor Q. Double-mode filter Frequency-selective filter using multiport resonator supporting two resonances. Proper allocation of their resonance and antiresonance frequencies offers flat passband and good out-band rejection. Duplexer Equivalent to antenna duplexer. 2 coupling Effective electromechanical coupling factor (Keff ) Electromechanical factor derived from the impedance characteristics of a resonator, where influences of electrodes and/or additional layers are taking into account. Effective electromechanical coupling factor for thickness-extensional vibration 2 (Kteff ) Electromechanical coupling factor for the thickness-longitudinal vibration estimated from measured or calculated admittance characteristics of a resonator by 2 the formula: K teff = πf s 2f p tan πf s 2f p , where fp and fs are the parallel and

(

)

(

)

series resonance frequencies, respectively. 2 Electromechanical coupling factor (K ) Certain combination of elastic, dielectric and piezoelectric constants which appears naturally in the expression of impedance of a resonator. A different factor arises in each particular family of mode of vibration. The factor is closely related to the relative frequency spacing and is a convenient measure of piezoelectric transduction. Alternatively, the coupling factor may be interpreted as the ratio of the electrical or mechanical work which can be accomplished to the total energy stored from a mechanical or electrical power source for a particular set of boundary conditions. Electromechanical coupling factor for thickness-extensional vibration (Kt2) Electromechanical coupling factor for the thickness-longitudinal vibration. When the influences of electrodes and/or additional layers are negligible, it is given by the for2 mula: Kteff = πf s 2f p tan πf s 2f p .

(

)

(

)

where fp and fs are the parallel and series resonance frequencies, respectively. Equivalent circuit Electrical circuit which has the same impedance as a piezoelectric resonator in the immediate neighborhood of resonance. Fast shear wave Refer to shear wave. Figure of merit (FOM or M) Factor indicating performance of the device relative 2 to the other ones. As for RF BAW devices, Kteff × Q is most often used, where Q is 2 the Q-factor and K teff is the effective electromechanical coupling factor. For low-frequency resonators, the value is usually defined by Q/γ, where Q is the Q-factor and γ is the ratio of capacitances. Film bulk acoustic resonator (FBAR) RF BAW resonator comprising a free-standing membrane as a vibrating element. Fractional bandwidth Ratio of the pass bandwidth to the mid-band frequency in the case of bandpass filters. Fundamental resonance The lowest resonance mode in a given family of vibration.

Glossary

261

Group delay Delay time of the signal envelope from input to output. It is given by τ = −∂ {∠H (f )} ∂ (2πf ), where H(f) represents the transfer function such as the scattering coefficient. Guard band Frequency range between the passband and rejection band. Half-wavelength resonance Resonance characterized by the field distribution with the standing wave pattern of circa a half-wavelength. Harmonic resonance Higher-order resonance appearing at frequencies corresponding to integer times the fundamental resonance frequency. Higher-order resonance Resonance mode higher than the fundamental one in a given family of vibration. Impedance ratio Ratio of the absolute value of the impedance at the parallel resonance frequency to that of the impedance at the series resonance frequency. In-band insertion loss Maximum value of the insertion loss in the passband in decibels. In-band ripple Peak-to-peak variation of the insertion loss in the passband in decibels. Inharmonic overtone Equivalent to inharmonic resonance. Inharmonic resonance Higher order resonance excluding harmonic ones. Insertion loss (IL) Logarithmic ratio of the power delivered to the load impedance before and after insertion of the resonator. Isolation Signal leakage between the transmitter (Tx) port and the receiver (Rx) port in antenna duplexer in the Tx or Rx band. Ladder-type filter Frequency-selective filter composed of multiple one-port resonators mutually connected in ladder-form. Lattice-type filter Frequency-selective filter composed of multiple one-port resonators mutually connected in lattice form. Lamb wave Guided wave propagating along a plate and composed of shear-vertical and longitudinal displacement components. Lead zirconate titanate (PZT) Ferroelectric and piezoelectric material with large electromechanical coupling for the thickness extensional vibration. Its high-quality thin film can be deposited by the sputtering method. Longitudinal resonance mode Resonance caused by the wave excitation to the thickness direction. Longitudinal wave Acoustic wave vibrating parallel to the propagation direction. Longitudinally coupled resonator filter (LCF) Resonator filter composed of multiple resonators acoustically coupled in thickness direction. Long-term stability Aging characteristic: variation of a specific frequency or amplitude with time, in order of days to years. Equivalent to dissipation factor D. Loss tangent (tan Mason’s equivalent circuit An equivalent circuit model for piezoelectric resonators employing the thickness-extensional mode. It is rigorous for the one-dimensional analysis. Mason’s model Equivalent to Mason’s equivalent circuit.

262

Glossary

Mid term stability Temperature characteristic: variation of a specific frequency or amplitude with time, in order of minutes to hours. Minimum insertion loss (ILmin) Minimum value of the insertion loss. Modified Butterworth-Van Dyke (BVD) model Modified version of BVD model where a resistance R0 is connected in series with the shunt capacitance C0 for taking account of energy leakage and/or dielectric loss. Sometimes another resistance Rs is added in series with the input terminal for taking account of electrode resistance. Monolithic crystal filter (MCF) Acoustic resonator employing acoustically coupled multiple resonators fabricated on a single chip. Motional capacitance (C1) Capacitance of the motional or series arm of the equivalent circuit. Motional inductance (L1) Inductance of the motional or series arm of the equivalent circuit. Motional resistance (R1) Resistance of the motional or series arm of the equivalent circuit. Out-of-band rejection Difference between minimum value of the insertion loss in the rejection band and maximum value of the insertion loss in the passband. Overtone Equivalent to higher-order resonance. Parallel resonance frequency (fp) Equivalent to antiresonance frequency fa. Parallel resonance Q (Qp) Quality factor estimated at the parallel resonance frequency fp. When the modified BVD model is applicable, Qp is given by the formula: Qp = 2πf p L1 ( R1 + R0 ). Passband Frequency range where the insertion loss of a filter shall be equal to, or less than a specified value. Passband insertion loss Equivalent to in-band insertion loss. Passband ripple Equivalent to in-band ripple. Pass bandwidth Separation of frequencies between which the insertion loss of a filter shall be equal to, or less than a specified value. Quality factor (Q) Factor indicating how long stored energy is preserved in a system. It is closely related to the steepness of the conductance peak. It is defined as stored energy . Q = πf dissipated power in half cycle Quarter wavelength resonance Resonance characterized by the field distribution with the standing wave pattern of circa a quarter-wavelength. Ratio of capacitances ( ) Equivalent to capacitance ratio. Rayleigh wave SAW composed of shear vertical and longitudinal displacement components. Receiver (Rx) band Frequency range used for receiver. Reflection coefficient ( ) Amplitude ratio of the reflected signal to the input signal. When a resonator with the impedance Za is connected to a transmission line with the characteristic impedance Zb, the reflection coefficient at their connection point is given by Γ = ( Z a − Z b ) ( Z a + Z b ). Rejection band Frequency range where the insertion loss of a filter shall be larger than a specified value.

Glossary

263

Relative bandwidth Equivalent to fractional bandwidth. Resonance frequency (fr) Frequency of the motional or series arm of the equivalent circuit of the resonator. It is defined as the frequency where the susceptance of the resonator is zero while the conductance is extremely large. From the modified BVD model, it is given by the formula: f r = 1 2π L1 C1 . Return loss (RL) Value of the reciprocal of modulus of the reflection coefficient Γ, expressed in decibels. Quantitatively, it is equal to: −20log|Γ| [dB]. Series resonance frequency (fs) Equivalent to the resonance frequency fr. Series resonance Q (Qs) Quality factor estimated at the series resonance frequency fs. When the modified BVD model is applicable, Qs is given by the formula: Qs = 2πf s L1 R1 . Shear wave Acoustic wave vibrating normal to the propagation direction. In general, two types of shear BAWs exist in solid body, and are distinguished by their velocities. Namely, fast and slow shear waves. Shear-horizontal wave Shear BAW with the displacement parallel to the boundary surface. Shear-vertical wave Shear BAW with the displacement normal to the boundary surface. Short term stability Noise characteristic: variation of a specific frequency or amplitude with time less than a second. Shunt capacitance (C0) Capacitance in parallel with the motional arm of the equivalent circuit. Shunt resistance (R0) Resistance in parallel with the motional arm of the equivalent circuit. Single-ended input and/or output Signal-interface using a “common” conductor (ground) as a signal return path. Signal is transferred as voltage difference from the “common” conductor. Slow shear wave Refer to shear wave. Solidly mounted resonator (SMR) RF BAW resonator comprising a vibrating element solidly mounted on a supporting substrate. Spurious resonance State of resonance of a resonator other than that associated with the working frequency. Stacked crystal filter (SCF) BAW filter comprising acoustically coupled multiple resonators stacked to the thickness direction. Surface acoustic wave (SAW) Acoustic wave propagating along the surface of an elastic substrate, the amplitude of which decays exponentially into the substrate depth. Temperature coefficient of frequency (TCF) Ratio of fractional change in a characteristic frequency (for example, center frequency) of the device under test with the environmental temperature. Thickness-extensional (TE) mode Resonance mode associating with the longitudinal-type BAW propagating to the thickness direction in a plate. Thickness-shear (TS) mode Resonance mode associating with the shear-type BAW propagating to the thickness direction in a plate.

264

Glossary

Transition band Equivalent to guard band. Transmitter (Tx) band Frequency range used for transmitter. Transversally-coupled resonator filter (TCF) Resonator filter composed of multiple resonators acoustically coupled in the lateral direction. Transverse resonance mode Resonance caused by the wave excitation to the in-plane direction. It appears slightly different frequencies. Transverse wave Equivalent to shear wave. Unbalanced input and/or output Equivalent to single-ended input and/or output. Voltage standing wave ratio (VSWR) Another indicator of the reflection coefficient Γ. It is given by (1 + Γ ) (1 − Γ ). Zinc oxide (ZnO) Piezoelectric material with moderate electromechanical coupling for the thickness extensional vibration. Its high-quality thin film can be deposited by the sputtering method.

About the Author Ken-ya Hashimoto was born in Fukushima, Japan, on March 2, 1956. He received his B.S. and M.S. degrees in electrical engineering in 1978 and 1980, respectively, from Chiba University, Japan, and Dr. Eng. degree from the Tokyo Institute of Technology, Japan, in 1989. In 1980, he joined Chiba University as a research associate, and is now a professor at the University. In 1998, he was a visiting professor at the Helsinki University of Technology, Finland. In the winter of 1998/1999, he was a visiting scientist at the Laboratoire de Physique et Metrologie des Oscillateurs, CNRS, France. In 1999 and 2001, he was a visiting professor at the Johannes Kepler University of Linz, Austria. In 2001, he served as a guest coeditor of the special issue of IEEE Transactons on Microwave Theory and Techniques (MTT), Special Issue on Microwave Acoustic Wave Devices for Wireless Communication and Sensing. He also served as a publicity cochair of the 2002 IEEE International Ultrasonics Symposium. He was appointed as a member of the speaker’s bureau of the IEEE MTT Society. He served as an international distinguished lecturer of the IEEE UFFC Society from July 2005 to December 2006. He also serves as a distinguished lecturer of the IEEE Electron Device Society since 2007. He will serve as a general chair of the 2011 IEEE Ultrasonics Symposium, which will be held in Kobe, Japan. His current research interests include simulation and design of various high performance surface and bulk acoustic wave devices, acoustic wave sensors and actuators, piezoelectric materials, and RF circuit design. Dr. Hashimoto is a fellow of the IEEE and a member of the European Microwave Association, the Institute of Electronics, Information, and Communication Engineers of Japan, the Institute of Electrical Engineers of Japan, and the Acoustical Society of Japan.

265

266

About the Author

List of Contributors Robert Aigner Triquint Semiconductor, Inc. 1818 S. HW 441 Apopka, FL 32703 U.S.A. e-mail: [email protected] Marc-Alexandre Dubois Centre Suisse d’Electronique et de Microtechnique CSEM S.A. Jaquet-Droz 1 CH-2002 Neuchâtel Switzerland e-mail: [email protected] Lueder Elbrecht Avago Technologies GmbH Charles-de-Gaulle-Str. 2 D- 81737 Munich Germany e-mail: [email protected] Gernot G. Fattinger Triquint Semiconductor, Inc. 1818 S. HW 441 Apopka, FL 32703 U.S.A. e-mail: [email protected] Ken-ya Hashimoto Dept. EEE, Chiba University 1-33 Yayoi-cho, Inage-ku Chiba 263-8522 Japan e-mail: [email protected] Jyrki Kaitila Avago Technologies GmbH Charles-de-Gaulle-Str. 2 D- 81737 Munich Germany e-mail: [email protected] Kenneth M. Lakin P.O. Box 310 Redmond, Oregon 97756 U.S.A. e-mail: [email protected]

Jan-Willem Lobeek NXP Semiconductors Nijmegen Gerstweg 2 Nijmegen The Netherlands e-mail: [email protected] Stephan Marksteiner EPCOS AG Anzinger Str. 13 D-81671 Munich Germany e-mail: [email protected] Sergey Mishin Advanced Modular Systems, Inc. 7920 Winchester Circle Goleta, CA 93117 U.S.A. e-mail: [email protected] Yury Oshmyansky Avago Technologies, Inc. 499 Nyes Place Laguna Beach, CA 92651 U.S.A. e-mail: [email protected] Nick Pulsford NXP Semiconductors Nijmegen Gerstweg 2 Nijmegen The Netherlands e-mail: nick.pulsford @nxp.com Richard C. Ruby Avago Technologies, Inc. 350 West Trimble Rd San Jose, CA 95131 U.S.A. e-mail: [email protected] Bart Smolders NXP Semiconductors Nijmegen Gerstweg 2 Nijmegen The Netherlands e-mail: [email protected] Masanori Ueda Fujitsu Laboratories, Ltd. 64, Nishiwaki, Ohkubo-cho Akashi 674-8555 Japan e-mail: [email protected]

Index A Absorption coefficients, 62 Acoustical attenuation, 79 Acoustically coupled filters, 37–45 coupled resonator, 42–45 stacked crystal, 38–42 See also Filters Acoustically coupled resonator tuning, 46 Acoustic Fields and Waves in Solids (Auld), 135 Acoustic impedance, 65 Acoustic leakage, 205–7 BAW resonator, 205 Bragg reflector, 207–11 Acoustic velocities, 59 Acoustic waves attenuation, 59–62 reflection, 60 Adatoms, 179 Advanced Circuit Simulator (ADS), 129 AIN deposition systems, 184 deposition uniformity, 185 film stress, 186 grain growth, 183 sputtered, 186 surface roughness, 184 Air gap process, 25–27 demonstration test patterns, 26 illustrated, 26 test structure, 27 Aluminum (Al), 98, 191–92 Aluminum nitride, 100, 174–75 Apodization, 119, 137–40 downside, 139 Q losses and, 140 Area efficiency, 94

B Balanced bridge filters, 15–16 Balanced ladder filter, 37 BAW devices area efficiency, 94 basics, 51–89 characterization of, 197–219

design, 74–89, 91–96 electromechanical coupling coefficient, 91 frequency distribution, 103–4 high-frequency, 6 integration in SiP, 240 interconnect losses and parasitics, 94–95 material selection, 97–101 nonlinearities, 96 piezoelectric effect, 52 power handling, 93 quality factor, 92 robustness, 95 spurious modes, 82–88, 92 temperature coefficient of frequency (TCF), 88, 89, 93–94 thin films deposition, 173–94 yield, 103 BAW-FBAR application space, 108–14 duplexers, 108–12 manufacturing process, 112 oscillators, 112–13 RF filters, 108–12 sensors, 113–14 BAW filters, 110–12 advantages, 111 balance, 168 frequency position, 102 IF, 256 narrowband, 257 physical structure, 110 thin film, 10–11 types of, 110 wideband upconverter tuner, 256 BAW resonators, xi acoustic leakage, 205 AIN-based, 96 cross-sections, 3 loss mechanisms, 78 plate wave modes, 14 Q-factor, 226 simulation, 15 solidly mounted (SMR), xi thin film, 7, 52–58

267

268

BAW-SMR, 111–12 dispersion curves, 142 manufacturing process, 112 TE-1 mode, 141 Beam resonators, 6, 7 Berlincourt formula, 63 BiCMOS BAW postprocess impact, 223 die, 256 process, 239 Bluetooth LAMP-based, 254 radio solution, 254 transceiver IC, 253 Bonded plate resonators, 11–12 Bragg reflector (BR), 143, 207–11 acoustic leakage through, 207–11 cooptimization, 210 defined, 207 Q limitation due to, 209 quarter-wavelength, 208 reflectivity, 208 for shear waves, 210 Bridge filters, 15–16 Bulk acoustic wave (BAW) defined, 1–2 finiteness, 2 piezoelectric materials, 2 propagation modes, 2 support approximation, 1 technology background, 1–11 technology driving forces, 10–11 thin film, 19 See also BAW resonators; RF-BAW devices Butterworth-Van Dyke (BVD) circuit, 57, 58, 127

C Capacitive micromachined ultrasonic transducers (CMUTs), 51 Cascading filters, 48 Characterization, 197–219 electrical, 214–19 introduction to, 197 laser interferometry, 201–4 loss mechanisms, 204–14 single-layer material, 198–201 Chemical/biological sensors, 113–14 Chemical mechanical polishing (CMP), 23, 101, 118 processing, 101 for surface smoothness, 190

Index

Code division multiple access (CDMA), 120 long-term evolution, 157 service providers, 156–57 wideband (WCDMA), 170, 228–31 Colpitts oscillator, 233 Composite resonators, 12–13 concept, 13 defined, 12 TC, 33 See also Resonators Copper (Cu), 98 Coupled resonator filters (CRF), 42–45 acoustical coupling, 44 bandwidth, 44 cross-sectional view, 43 defined, 43 electrode effect, 45 equivalent circuits, 46, 47 FBAR, 149–50 inductor-tuned two-pole, 47 layout, 44 response, 45 Coupling coefficients, 72 effective, 62, 63, 74–77 electromechanical, 55, 91 equation, 82 material, 74 spurious modes, 85 Cutoff frequencies, 69–70 defined, 69 diagrams, 71 longitudinal, 70 shear modes, 70 spurious resonance-free resonator, 84

D Deformation sensors, 114 Detuning layer, 101 Device design, 74–89 area efficiency, 94 considerations, 91–96 effective coupling coefficient, 74–77 electromechanical coupling coefficient, 91 interconnect losses and parasitics, 94–95 loss mechanisms, 78–82 nonlinearities, 96 power handling, 93 quality factor, 92 Q-values, 78–82 robustness, 95 spurious modes, 82–88, 92

Index

temperature coefficient of frequency (TCF), 88, 89, 93–94 See also BAW devices Device fabrication, 97–108 chemical mechanical polishing (CMP), 101 etching process, 100 ion milling, 104–5 local etching, 105 material selection, 97–101 process controls, 108 SMR resonators and filters, 101–2 tolerances, 102–8 trimming, 102–8 See also BAW devices Device under test (DUT), 201 Dielectric layers, 106–7 Dielectrics, 99–100 Dispersion branches, 67 diagrams, 68 evaluation of, 203–4 Dispersion curves, 70 for AIN plate, 134 BAW-SMR, 142 FFT algorithm, 203 TE-1, 137 Tungsten electrode resonator, 136 Dispersion relations, 67–70 cutoff frequencies, 69–70 multilayered plate, 69 resonator design based on, 70–74 simple plates, 67, 68 Displacement amplitudes, 71, 72 Displacement profiles, 74 Double-lattice filter, 229 Double-mode SAW (DMS), 166 Dual-target magnetron, 181 Duplexers, 108–12 BAW filter dies, 247 ceramic, 121 FBAR, 119–22 illustrated, 121 measurements, 154 PCS, 217 performance, 152–53 SAW, 124 topology, 151 for WCDMA, 170

E Effective coupling coefficient, 62, 63, 74–77 maximized, 76

269

normalized, 76, 77 Elastic energy, 63 Electrical characterization, 214–19 filter measurements, 217–19 geometries, 215 introduction to, 214 resonator measurements, 214–17 See also Characterization Electrical losses, 78, 212 Electrically coupled filters, 34–37 balanced ladder, 37 conventional lattice, 37 ladder, 34–37 See also Filters Electrical resistivity, 78 Electric energy, 63 Electroacoustic conversion, 62–64 Electrodes current path, 94 resistance, 30 resonator with, geometry, 75 Electromagnetic modeling, 246–51 Electromechanical coupling coefficient, 55, 91 Electromechanical coupling factor, 55–56 Electron beam (EB) exposure system, 162 Electron cyclotron resonance (ECR) deposition, 187 Electrostatic discharge (ESD), 95 Environmental robustness, 95 Etching process, 100 Excimer stepper, 162

F FBAR filters, 150–56 coupled resonator, 149–50 duplexer, 152–56 interstage, 150–52 multiplexer, 152–56 performance, 153 topology, 151 FBAR resonators advantage, 155 center region, 143 figure of merit (FOM), 124, 125 modeling, 126–29 package, 122–23 Q-circle, 124, 125, 126, 135 temperature-compensated, 145–49 Figure of merit (FOM), 124, 125 Film bulk acoustic wave resonator (FBAR), xi, 117–58 bulk micromachined, 118

270

Film bulk acoustic wave resonator (continued) devices, 77 duplexer, 119–22 history of, 117–19 process flow, 225 in real world, 123–24 research, 117 structure illustration, 162 TCF, 93 technology, 124–50 See also FBAR filters; FBAR resonators Filtering LNA, 227–28 Filter measurements, 217–19 guard band, 218 passband, 217–18 return loss, 218–19 stopband, 218 Filters acoustically coupled, 37–45 balanced bridge, 15–16 balanced ladder, 37 cascading, 48 in cell phone market, 10 coupled resonator, 42–45 double-lattice, 229 electrically coupled, 34–37 fabrication, 101–2 FBAR, 150–56 hybrid, 47–48 insertion loss, 227 interstage, 150–52 ladder, 16–17, 35–37 lattice, 17, 37 legacy, topologies, 15–18 monolithic, 18 RF, 108–12 SAW, 166–68 skirts, steepness, 109 stacked crystal, 38–42 wide-bandwidth tuned coupled resonator, 45–47 Finite element model (FEM), 81, 127 Flyback, 152 Focused ion beam (FIB), 185 Frames, 140–45 Frequency distribution, 103–4 Frequency shift, by DC bias, 96 Front-end module (FEM), 236, 245 Front-ends architecture, 238 high Q tunable components, 239 integration, 236–37

Index

partitioning, 238 trends, 237–38 Full wave-half maximum (FWHM), 175, 177

G Galerkin’s method, 250 Green’s function, 248, 249 Guard band, 218

H Hook’s law, 2, 53, 59 Hybrid filters, 47–48

I I-line, 162 Inertial sensors, 114 Input impedance, 250 Insertion loss, 227 Interconnects area consumed by, 94 losses, 94–95 Interdigital transducers (IDT), 161 Interstage filters, 150–52 Ion beam deposition (IBD), 188 Ion beam etching, 108 Ion milling, 104–5 Iridium, 99

J Jet vapor deposition (JVD), 189

K Known-good dies (KGDs), 252

L Ladder filters, 16–17, 34–37 balanced, 37 circuit diagrams, 36 circuits, 17 complicated, 35 experimental results, 34 formats, 16 out-of-band rejection, 35 series and shunt resonators, 34 summary, 36 See also Filters Laminate platform, 241–42 Laser interferometry, 201–4 evaluation of dispersion, 203–4 introduction to, 201 measurement setup, 201–3 setup illustration, 202 Lateral displacement profiles, 74

Index

Laterally leaking waves, 211 Lattice filters configuration, 17 conventional, 37 defined, 17 double, 229 See also Filters Legacy filter topologies, 15–18 Lithography, 100 Local etching, 105 Loss mechanisms, 204–14 acoustic leakage, 205–7 acoustic leakage (Bragg reflector), 207–11 electrical losses, 212 introduction to, 204 laterally leaking waves, 211 scattering losses, 214 viscoelastic losses, 212–13 Low noise amplifiers (LNAs), 222 differential, 227 filtering, 227–28 Low temperature cofired ceramic (LTCC) platform, 242–43

M Mason model, 64–66 applying, 73 calculations, 79 defined, 64 implementation in ADS, 127 longitudinal/shear transitivity, 209 mechanical force at boundaries, 65 for multilayer resonator, 66 for piezolayer, 66 Material coupling coefficient, 74 Material selection, 97–101 lithography, 100 metals, 97–99 piezoelectric layers, 100 semiconductors and dielectrics, 99–100 See also Device fabrication MATLAB, 250 Maxwell equations, 246 mBVD model, 128 fitted, 143 fitting resonator with, 133 reactance terms, 129 Mechanical robustness, 95 MEMS gyroscopes, 114 Metal deposition, 189–94 aluminum, 191–92 combination of metals, 194

271

molybdenum (Mo), 192 platinum (Pt), 193 ruthenium (Ru), 194 tungsten (W), 192–93 Metallic layers, 200–201 Metalorganic chemical vapor deposition (MOCVD), 188–89 Metals, 97–99 Method-of-moments (MoM), 247, 248 Microelectromechanical systems (MEMS), 223 Mirror transmissivity, 79, 80 Molybdenum (Mo), 192 Molybedenum (Mo), 99 Monolithic crystal fiber (MCF) BAW implementation, 18 technologies, 5 three-pole, 18 Monolithic filters, 18 Monolithic integration, 221–33 applicability, 222 complexity, 232 in Europe, 225 filtering LNA, 227–28 introduction to, 221–22 of piezoelectric BAW components, 221 practical implementation, 224–32 substrates and, 224 technology description, 225–27 WCDMA RF front-end, 228–31 WLAN oscillator, 232 Monolithic WCDMA receiver, 230, 231 Monte Carlo simulations, 183 Multidimensional effects, 14–15 Multilayer resonators, 66 Multiplexers, 152–56 Mutual energy, 63

N Narrowband BAW filter, 257 Network analyzers, 202 Nonlinearities, 96 Nonvacuum deposition, 189 NXP semiconductors, 244, 245

O Organization, this book, xii Oscillators, 112–13 BAW-based high-frequency, 256 chip micrograph, 232 Colpitts, 233 output power spectrum, 233 schematic, 232

272

Oscillators (continued) TCXO quartz, 149 WLAN, 232 Overmodeled resonators (OMR), 3–4

P Passband, 217–18 Phase-locking system, 203 Piece-wise linear (PWL) approximation, 249 Piezoelectric constitutive relations, 52–56 Piezoelectric Crystals, 62 Piezoelectric materials, 2 aluminum nitride, 174–75 commonly uses, 173–75 physical properties, 174 PZT, 173–74 zinc oxide, 173 Piezoelectric resonators, 7–10 Piezolayers, 55 Mason model for, 66 materials, 100 thickness, 94 Plasma discharge, 180 Plasma-enhanced chemical vapor deposition (PECVD), 198 Plasma etching, 190 Plate edge-supported resonators, 21–27 pocket membrane, 24–25 pothole membrane, 21–23 undercut air gap membrane, 25–27 See also Resonators Plate wave (PW) resonators, 4–5 Plate waves, 14 Platinum (Pt), 193 Pocket membrane, 24–25 process illustration, 24 process with support structures, 25 Poisson ratio, 59 Polymers, 100 Pothole membrane, 21–23 advanced process, 23 disadvantage, 22 process illustration, 22 in resonator fabrication, 22 See also Plate edge-supported resonators Power-amplifier-with-integrated-duplexer (PAiD), 236, 237 Power durability, SAW filters, 165–66 Power handling, 93 Process control monitor (PCM), 108 Propagation constants, 54 Prototype resonator, 52–56

Index

Pulsed DC magnetron, 181 PZT, 173–74 characteristics, 173 deposition, 187 structure illustration, 174

Q Q ascertaining, 129–33 enhancements, 140 frequency versus, 130, 132 reasonable, 131 Q-circles, 124, 125, 126 fitting, 131 illustrated, 135 measured magnitude, 138 parasitic modes, 133 of quadrilateral, 141 of square resonator, 138 symmetric modes on, 133 Q-factors, 162–65 BAW resonator, 226 examples, 164, 165 as figure of merit (FOM), 164 on ladder-type filters, 164 passband performances, 163 Quality factors, 58, 92 Quartz crystal resonators, 9–10 Quartz crystal thinning, 11 Quartz material, 31–32 Quintplexers, 155, 156 Q-values, 92

R Rayleigh-Lamb (RL) modes, 133–37 calculating, 135 defined, 133 families, 133 lateral waves, 143 symmetric, 133 TE-1 mode, 137 Reactive ion etch (RIE) etching process, 24 Reactive sputtering, 180 of AIN, 183 metallic and poison zones, 182 Reflected spectrum measurement, 96 Reflection, 59–62 Reflection coefficients, 140 Removal profiles, 108 Residue, 250 Resonance, 4 Resonant frequencies, 56

Index

Resonator measurements, 214–17 impedance, 214–15, 216 phase curves, 216 Resonators air gap coupled, 3 beam, 6, 7 bonded plate, 11–12 composite, 12–13 dead area between, 94 FBAR, 117–58 geometry for plate wave excitation, 14 lateral extent, 9 modeled phase, 10 piezoelectric, 7–10 plate edge-supported, 21–27 PW, 4–5 Q-factor, 227 quartz crystal, 9–10 SAW, 4–5, 162–66 solidly mounted, 27–29 spurious resonance-free, 84, 86 standing wave distribution across, 16 structures, 7 thin plate, 11–12 types of, 5–7 See also BAW resonators Return loss, 218–19 RF-BAW devices, xi RF filters, xi, 108–12 application space, 113 performance characteristics, 109 RF SiP (RF-SiP), 235 BAW examples, 255–57 design flow, 252 design methodology, 251–52 examples, 253–57 illustrated, 246 wireless examples, 253–55 Robustness, 95 Ruthenium (Ru), 99, 194

S SAW filters, 110–12 active area, 112 advantages, 111 bandwidth, 110 configuration, 167 design, 166–68 design methods, 166 double-mode SAW (DMS) design, 166 performance, 167 physical structure, 110

273

topologies, 163 SAW resonators, 4–5 power durability, 165–66 Q-factors, 162–65 Smith chart, 163 See also Resonators Scattering losses, 214 Semiconductors, 99–100 Sensors, 113–14 chemical/biological, 113–14 deformation, 114 inertial, 114 Shear modulus, 59 Silicon nitride (SiN), 99 Silicon oxide (SiO2), 99 Silicon-oxy-carbide (SiOC), 99–100 Silicon (Si), 99 Simultaneous GPS (S-GPS), 155 Single-layer material characterization, 198–201 dielectric and piezoelectric layers, 198–200 introduction to, 198 metallic layers, 200–201 See also Characterization Snell law, 60 Software-defined ratios, 236 Solidly mounted resonators (SMR), xi, 27–29 acoustic energy, 33 acoustic impedance mismatch, 191 cross-section, 28, 102 displacement versus depth, 28 experimental results, 33 fabrication, 101–2 laser interferometer analysis of, 81 reflection spectrum and, 29 TCF, 94 See also Resonators S-parameters, 214, 215 calculating, 250 double-lattice filter, 229 Spurious modes, 82–88, 92 acoustical, 83 coupling coefficients, 85 defined, 92 emergence of, 82 energy losses and, 92 type I resonator, 82–83 Spurious resonance-free resonator cutoff frequency diagram, 84 resonant modes, 86 Sputtering, 175–83 defined, 175 practical aspects, 183–87

274

Sputtering (continued) PZT deposition by, 187 reactive, 180, 182 Stacked crystal filters (SCF), 38–42 circuit models, 41 cross-sectional views, 38 defined, 38 equivalent circuit, 45 fabrication layout, 41–42 layout illustration, 42 modeled response, 39 performance, 41 Sticking coefficients, 178 Stopband, 218 Stress-free boundary condition, 75 Stress profiles, 77 Surface acoustic wave (SAW), 2 application map, 169–70 center frequency, 168 devices, xi, 161–70 duplexers, 124 manufacturing process, 168 structural comparison and features, 161–62 structure illustration, 162 TCF, 93 technology, 110 temperature compensation, 168–69 See also SAW filters; SAW resonators Surface-mount components (SMDs), 241 Surface roughness, 178 System-in-package (SiP), xi, 235–57 BAW device integration in, 240 complexity, 245 cost calculation, 240 design, 245–52 design methodology, 251–52 electromagnetic modeling, 246–51 integration, 235–37 known-good dies (KGDs), 252 laminate platform, 241–42 LTCC platform, 242–43 multiband, multimode wireless systems, 235–39 RF (RF-SiP), 235 SoC versus, 239–40 technologies, 241–45 test and industrialization, 252–53 thin film platform, 243–45 System-on-chip (SoC), xi, 235 cost calculation, 240 SiP versus, 239–40

Index

T Tantalum (Ta), 99 TE-1 mode, 136 dispersion curve, 137 group velocities, 136 SMR-BAW, 141 subresonances, 137 Temperature coefficient of frequency (TCF), 88, 89, 93–94 determination, 168 FBAR, 93 SAW, 93 SMR, 94 Temperature coefficient of velocity (TCV), 168 Temperature-compensated resonators, 145–49 Temperature compensation, 31–34 early experiments, 33 layouts, 32 Thermal expansion coefficient (TEC), 168 Thickness characterization, 108 effect on frequency, 126 error, histogram, 107 uniformity, 108 Thickness extensional (TE), 59, 161 Thickness shear modes, 133 Thin film BAW, 19, 51 equivalent circuit, 57–58 filters, 10–11 parameters, 57–58 piezoelectric constitutive relations, 52–56 prototype, 52–56 resonators, 7, 52–58 technology, 51 temperature effects, 89 Thin Film Processes, II, 175 Thin films development, 13–14 passive integration IC technology, 243 platform, 243–45 Thin films deposition, 173–94 for BAW applications, 189–94 electron cyclotron resonance (ECR), 187 ion beam (IBD), 188 jet vapor (JVD), 189 metalorganic chemical vapor (MOCVD), 188–89 methods, 175–89 nonvacuum, 189 piezoelectric materials, 173–75 sputtering, 175–83 Thin plate resonators, 11–12

Index

bonded plate, 11–12 quartz crystal thinning, 11 Titanium (Ti), 98 Transduction process, 3 Transmission, 59–62 Transmission coefficients, 89 Transmissivity, mirror, 79, 80 Trimming, 105–6 Tungsten (W), 98, 192–93

U Undercut air gap membrane, 25–27

V Viscoelastic behavior, 213 Viscoelastic losses, 212–13 Viscoelastic materials, 213 Viscosity, 61 Voltage-controlled oscillators (VCO) control voltage, 230 differential FBAR-based, 228–29 phase noise, 229, 231 reference, 229

275

W Wafer-level packaging technology, 122 Wave modes, 67–70 Wave numbers, 54 Wave propagation, 59–62, 67 Wave velocity comparison, 110 WCDMA duplexers for, 170 monolithic receiver, 230, 231 RF front-end, 228–31 Wideband upconverter tuner, 256 Wide-bandwidth tuned coupled resonator filters, 45–47 Wireless systems, 235–39 key challenge, 236 penetration rate, 236 WLAN oscillator, 232

X X-ray diffraction (XRD), 200

Z Zinc oxide, 173

Recent Titles in the Artech House Microwave Library Active Filters for Integrated-Circuit Applications, Fred H. Irons Advanced Techniques in RF Power Amplifier Design, Steve C. Cripps Automated Smith Chart, Version 4.0: Software and User's Manual, Leonard M. Schwab Behavioral Modeling of Nonlinear RF and Microwave Devices, Thomas R. Turlington Broadband Microwave Amplifiers, Bal S. Virdee, Avtar S. Virdee, and Ben Y. Banyamin Computer-Aided Analysis of Nonlinear Microwave Circuits, Paulo J. C. Rodrigues Designing Bipolar Transistor Radio Frequency Integrated Circuits, Allen A. Sweet Design of FET Frequency Multipliers and Harmonic Oscillators, Edmar Camargo Design of Linear RF Outphasing Power Amplifiers, Xuejun Zhang, Lawrence E. Larson, and Peter M. Asbeck Design Methodology for RF CMOS Phase Locked Loops, Carlos Quemada, Guillermo Bistué, and Iñigo Adin Design of RF and Microwave Amplifiers and Oscillators, Second Edition, Pieter L. D. Abrie Digital Filter Design Solutions, Jolyon M. De Freitas Distortion in RF Power Amplifiers, Joel Vuolevi and Timo Rahkonen EMPLAN: Electromagnetic Analysis of Printed Structures in Planarly Layered Media, Software and User’s Manual, Noyan Kinayman and M. I. Aksun Essentials of RF and Microwave Grounding, Eric Holzman FAST: Fast Amplifier Synthesis Tool—Software and User’s Guide, Dale D. Henkes Feedforward Linear Power Amplifiers, Nick Pothecary Foundations of Oscillator Circuit Design, Guillermo Gonzalez Fundamentals of Nonlinear Behavioral Modeling for RF and Microwave Design, John Wood and David E. Root, editors Generalized Filter Design by Computer Optimization, Djuradj Budimir High-Linearity RF Amplifier Design, Peter B. Kenington High-Speed Circuit Board Signal Integrity, Stephen C. Thierauf Intermodulation Distortion in Microwave and Wireless Circuits, José Carlos Pedro and Nuno Borges Carvalho

Introduction to Modeling HBTs, Matthias Rudolph Lumped Elements for RF and Microwave Circuits, Inder Bahl Lumped Element Quadrature Hybrids, David Andrews Microwave Circuit Modeling Using Electromagnetic Field Simulation, Daniel G. Swanson, Jr. and Wolfgang J. R. Hoefer Microwave Component Mechanics, Harri Eskelinen and Pekka Eskelinen Microwave Differential Circuit Design Using Mixed-Mode S-Parameters, William R. Eisenstadt, Robert Stengel, and Bruce M. Thompson Microwave Engineers’ Handbook, Two Volumes, Theodore Saad, editor Microwave Filters, Impedance-Matching Networks, and Coupling Structures, George L. Matthaei, Leo Young, and E.M.T. Jones Microwave Materials and Fabrication Techniques, Second Edition, Thomas S. Laverghetta Microwave Mixers, Second Edition, Stephen A. Maas Microwave Radio Transmission Design Guide, Trevor Manning Microwaves and Wireless Simplified, Third Edition, Thomas S. Laverghetta Modern Microwave Circuits, Noyan Kinayman and M. I. Aksun Modern Microwave Measurements and Techniques, Second Edition, Thomas S. Laverghetta Neural Networks for RF and Microwave Design, Q. J. Zhang and K. C. Gupta Noise in Linear and Nonlinear Circuits, Stephen A. Maas Nonlinear Microwave and RF Circuits, Second Edition, Stephen A. Maas QMATCH: Lumped-Element Impedance Matching, Software and User’s Guide, Pieter L. D. Abrie Practical Analog and Digital Filter Design, Les Thede Practical Microstrip Design and Applications, Günter Kompa Practical RF Circuit Design for Modern Wireless Systems, Volume I: Passive Circuits and Systems, Les Besser and Rowan Gilmore Practical RF Circuit Design for Modern Wireless Systems, Volume II: Active Circuits and Systems, Rowan Gilmore and Les Besser Production Testing of RF and System-on-a-Chip Devices for Wireless Communications, Keith B. Schaub and Joe Kelly Radio Frequency Integrated Circuit Design, John Rogers and Calvin Plett RF Bulk Acoustic Wave Filters for Communications, Ken-ya Hashimoto

RF Design Guide: Systems, Circuits, and Equations, Peter Vizmuller RF Measurements of Die and Packages, Scott A. Wartenberg The RF and Microwave Circuit Design Handbook, Stephen A. Maas RF and Microwave Coupled-Line Circuits, Rajesh Mongia, Inder Bahl, and Prakash Bhartia RF and Microwave Oscillator Design, Michal Odyniec, editor RF Power Amplifiers for Wireless Communications, Second Edition, Steve C. Cripps RF Systems, Components, and Circuits Handbook, Ferril A. Losee Stability Analysis of Nonlinear Microwave Circuits, Almudena Suárez and Raymond Quéré System-in-Package RF Design and Applications, Michael P. Gaynor TRAVIS 2.0: Transmission Line Visualization Software and User's Guide, Version 2.0, Robert G. Kaires and Barton T. Hickman Understanding Microwave Heating Cavities, Tse V. Chow Ting Chan and Howard C. Reader For further information on these and other Artech House titles, including previously considered out-of-print books now available through our In-Print-Forever® (IPF®) program, contact: Artech House Publishers

Artech House Books

685 Canton Street

16 Sussex Street

Norwood, MA 02062

London SW1V 4RW UK

Phone: 781-769-9750

Phone: +44 (0)20 7596 8750

Fax: 781-769-6334

Fax: +44 (0)20 7630 0166

e-mail: [email protected]

e-mail: [email protected]

Find us on the World Wide Web at: www.artechhouse.com