Millimeter Waves in Communication Systems (Innovative Technology Series)

Millimeter Waves in Communication

This page intentionally left blank

INNOVATIVE TECHNOLOGYSERIES INIORMATION SYSTEMS AND NETWORKS

Millimeter Waves in Communication

edited by Michel Ney

HP5

HERMES PENTON SCIENCE

First published in 2001 by Hermes Science Publications, Paris First published in 2002 by Hermes Penton Ltd Derived from Annales des Telecommunications, Vol. 56, no. 1-2, GET, Direction Scientifique, 46 rue Barrault, F 75634, Paris, Cedex 13, France. www.annales-des-telecommunications.com Apart from any fair dealing for the purposes of research or private study, or criticism or review, as permitted under the Copyright, Designs and Patents Act 1988, this publication may only be reproduced, stored or transmitted, in any form or by any means, with the prior permission in writing of the publishers, or in the case of reprographic reproduction in accordance with the terms and licences issued by the CLA. Enquiries concerning reproduction outside these terms should be sent to the publishers at the undermentioned address: Hermes Penton Science 120 Pentonville Road London Nl 9JN © Hermes Science Publications and GET, 2001 © Hermes Penton Ltd, 2002 The right of Michel Ney to be identified as the editor of this work has been asserted by them in accordance with the Copyright, Designs and Patents Act 1988.

British Library Cataloguing in Publication Data A CIP record for this book is available from the British Library. ISBN 1 9039 9617 1

Typeset by Saxon Graphics Ltd, Derby Printed and bound in Great Britain by Biddies Ltd, Guildford and King's Lynn www.biddles.co.uk

Contents

Foreword Michel Ney 1. Heterojunction bipolar transistors for millimeter wave applications: trends and achievements S. L. Delage 2.

3.

4.

5.

6.

7.

vii

1

HEMT's capability for millimeter wave applications S. Bollaert, Y. Cordier, M. Zaknoune, T. Parenty, H. Happy and A. Cappy

20

Progress in millimeter-wave fiber-radio access networks A. Nirmalathas, C. Lim, D. Novak and R. B. Waterhouse

43

Hybrid 3D integrated circuits at millimeter-wave frequencies: advantages and trends Ch. Person, E. Rius and J. Ph. Coupez

68

Printed millimeter-wave reflectarrays D. Pilz and W. Menzel

94

Computer aided design for new microwave filter topologies for spatial applications in Ka band D. Baillargeat, H. Blondeaux, S. Bila, P. Levdque, S. Verdeyme andP. Guillon

113

A simple way to design complex metallic photonic band-gap structures G. Poilasne, P. Pouliguen, C. Tenet, P. Gelin and L. Desclos

135

Index

147

This page intentionally left blank

Foreword

New and future developments for communication systems in the millimeter-wave range With the rapid development of communication systems, the demand for wideband operations has led either to a congestion of the spectrum in the RF and microwave frequency range or an increasing demand for operations in the millimeter-wave range. Opportunities have arisen as frequency allocations have been made available in the millimeter-wave frequencies (above 20 GHz). As a result, applications are now considered or being developed at these frequencies. To name a few, one can mention short-range anti-collision radar (76 GHz), Local Area Networks (LANS), Local Multipoint Distribution Systems (LMDSS). Also, satellite communications (e.g. Ka band for TV channels and services), development of optical fiber networks to provide multi services (fiber to the building FTTB or fiber to the home FTTH) are being developed. Although technological processes and modeling tools have been largely investigated from the low frequency range to few tens of gigahertzs, these are generally not readily transposable to the higher part of the millimeter-wave range. This publication has as its objective to present some new trends and recent developments in millimeter-wave components and systems, focusing on applications to communication systems. The state-of-the-art regarding solid-state devices is discussed in the two first two contributions. First of all, S.L. Delage from the "Laboratoire Central de Recherches" at THALES (former Thomson CSF), Orsay, France, presents the development of Heterojunction Bipolar Transistors (HBT) and circuits for power microwave applications and also of GaN-based field effect transistors. Rapid progress in the field has allowed the use of HBT in the millimeter-wave range and f^ up to 820 GHz has been reported. After a presentation of comparisons between different HBT technological solutions, a short review of millimeter-wave circuits shows that functions such as vco can operate at frequencies above 100 GHz and amplifications up to 85 GHz. In the second paper, the state-of-the-art of High Electron Mobility Transistors (HEMT) is presented by S. Bollaert, Y. Cordier, M. Zaknoune, T. Parenty, A. Cappy and H. Happy of the "Institut d'Electronique et de Microelectronique du Nord (IEMN)", Villeneuve d'Ascq, France. This technology has produced the fastest transistor device, especially with the emergence of nanometer-gate length process. The salient fact is that HEMT cut-off frequency has rapidly increased to currently reach 350 GHz and an LNA amplifier using InP HEMT with 7.2 dB gain has been implemented.

viii

Foreword

The authors first present the quality factors of HEMT to obtain low-noise and high power performances. Then, the most commonly HEMTS used in millimeter wave IC's applications are presented. Finally, the state-of-the-art regarding cutoff frequencies, low-noise and power level is discussed. Through Wavelength Division Multiplexing (WDL), optical fiber networks may be the key to provide quasi-unlimited bandwidth for services to customers. To access the fiber network, broadband millimeter-wave systems are foreseen to redistribute the signal to customers (private home or building) via cables and/or radio links. In their paper "Wavelength Progress in Millimeter-Wave Fiber-Radio Access Networks", A. Nirmalathas, C. Lim, D. Novak and R. B. Waterhouse, (Australian Photonics Cooperative Research Centre, University of Melbourne and the Royal Melbourne Institute of Technology, Australia) discuss the various architectures that have been proposed for fiber mm-wave access systems employing optical fiber distribution networks. Also, some aspects regarding the millimeter-wave and opto-electronic components are presented. Regarding the implementation of passive components such as filters in the millimeter-wave range, the use of Monolithic Microwave Integrated Circuits (MMC) may present some difficulties regarding quality factors. The solution seems to emerge from using hybrid technologies for which functions such as filters couplers and antenna are implemented in various planar structures. In the next contribution, Ch. Person, E. Rius and J. Ph. Coupez from the Laboratory for Electronics and Systems of Telecomunications (LEST), Brest, France, discuss the advantages of three-dimensional hybrid circuits and the future trends. In particular, co-planar technology is presented with original compensation methods to account for the presence of integrated bridges to suppress unwanted dispersive modes. Examples of filters implemented with this technology are presented and results show that the technique yields very good results up to 50 GHz. Then, the authors address the crucial problem regarding the interconnection between planar circuits and active MMIC'S components. They propose various solutions associating coplanar and Thin-Film MicroStrip (TFMS) structures. The objective is to ensure integration reproducibility and reliability at reasonable cost and to provide compatibility between sub-modules. In the next contribution, D. Pilz and W. Menzel (University of Ulm and DaimlerChrysler Aerospace, Ulm, Germany) present an emerging type of antenna that are developed for millimeter-wave applications namely, reflectarrays. They consist of arrays of periodic or quasi-periodic metallic motives printed on planar dielectric substrates. Geometrical shape of elements can be modified to achieve the desired local reflection phase property. Using different configurations, lowloss dual property antenna can be built (dual polarisation or operating frequency). Also, their low-profile reflector improves compactness. Theoretical models are presented and comparison with experiment shows good agreement with applications up to 60 GHz. In the sixth paper, D. Baillargeat, H. Blondeaux, S. Bila, P. Lev6que, S. Verdeyme and P. Guillon from the University of Limoges, France, discuss

Foreword

ix

some aspect regarding the direct implementation of electromagnetic tools to filter response optimization. The electromagnetic model is based on the Finite Element Method (FEM) which is coupled to other mathematical software. The proposed optimization procedure reduces the number of electromagnetic analyses which are costly in terms of CPU time, and the design of a low-loss compact 4-pole elliptic filter is designed using this technique. Finally, the authors propose a new type of circuit that provide both the filtering and radiation functions. This approach has the advantage eliminating losses produced by the connections normally used between the filter section and the radiating elements. Finally, G. Poilasne (UCLA, USA), P. Pouliguen and C. Terret (University de Rennes I, France), P. Gelin (LEST, France) and L. Desclos (NEC C&C Research Laboratory, New Jersey, USA) discuss the design of complex metallic photonic band-gap structures in their paper. The different topics presented in this publication provide a summary of the major activities in the development of components, devices and systems in the millimeter-wave range. It shows that solutions have been found for technological processes and design tools leading to new components that are able to cope with the ever-increasing demand for faster systems in the telecommunication area. Michel NEY Ecole Nationale Superieure des Telecommunications de Bretagne, Brest, France

This page intentionally left blank

Chapter 1

Heterojunction bipolar transistors for millimeter wave applications: trends and achievements S. L. Delage THALES, Laboratoire Central de Recherches, France

I. Introduction IH-V devices are now widely used for microwave applications. Products have finally emerged with a strong market growth due to the emerging wireless market. MESFET'S (metal-semiconductor field-effect transistor) are now the most used devices thanks to their cost and maturity. However HFET'S (heterojunction field effect transistor), the generic name of any field effect transistor including at least one heterojunction with any kind of doping management), first developed in 1980, are also used thanks to their excellent high frequency noise characteristics, power capability and extremely high maximum oscillation frequency f^. Heterojunction bipolar transistors (HBT) [1-2], whose development started in the mid-1980s for microwave applications are now also coming on to the commercial market. Most of the present market, about 80%, is located below 20GHz, but a rapid increase in the frequencies required is foreseen (in telecommunications, automotive avoidance radar etc). For applications above 30 GHz, robust MESFET technology is no longer useable. HEMT (high electron mobility transistor) technology is the present dominant solution to cover this market. Nevertheless, HBT performance improvement might change the current situation and the objective of this paper is to present a rapid overview of the current situation and to estimate the advantage of the different device topologies and their impact. One is not attempting to address the comparison between PET'S and HBT'S here. Maturity of the two families is not equal in the millimetre wave field. On one hand, FET'S have the advantage of a smaller number of process steps and they have therefore reached earlier maturity. The HFET'S currently available on market exhibit higher RF gain than HBT'S by using deep submicron gate length devices. On the other hand, HBT'S benefit from easier lithography and more uniform electrical performances than

Millimeter waves in communication systems

wafers. On-going HBT development gives rise to devices exhibiting equal or even better RF gains than HFET'S. PET'S are clearly very suitable to low noise applications, while HBT'S are very adapted to power applications. However, for millimeter waves the application separation might change the conventional frontier and integration might be a decisive factor, where HBT technology could benefit from their high on-wafer uniformity. Up to 3 years ago, the HBT'S reputation, especially ffl-V ones, was plagued by reliability limitations. However, GaAs- HBT have now invaded the mobile phone market, especially for the power transmitter function. This success story gives now after about 10 years of development high credibility to the HBT approach. It is clear that in near future most of the R&D effort will be carried out on both device families.

II. Material aspect The situation in the near future is quite clear concerning the semiconductor heterojunctions that will be used by microwave manufacturers to produce HBT for millimetre applications, namely SiGe/Si, InGaP/GaAs and InP/InGaAs. That means that three types of substrates will be used: Si, GaAs and InP. These different substrates have intrinsic properties that are quite important for the overall performances of the devices. Table I shows their main features. This provides real world advantages and limitations: • Si is a cost killer due to its wafer size and low defect density and shows good thermal stability associated with good thermal conductivity. Smaller gap and lower resistivity are its two main drawbacks. Table I. Key values of substrates for HBT fabrication. (The values are given at room temperature). Parameter \ Substrate Available size (mm) Robustness 2

Defect density (mm~ ) Energy gap (eV) Intrinsic carrier concentration (cm"3)

Si

GaAs

InP

150-300

75-150

50-75

High

Medium

Brittle

1

-1000

-SxlO 4

1.12

1.424

1.35

1.79 x 106

1.2 xlO 8

1.45xl0

10

4

8

Highest available resistivity (Q cm)

10

10

108

Dielectric constant

11.9

12.61

13.1

1415

1238

1070

1.41x10^

1.87x10-^

4.44 x 10-4

1.5

0.46

0.68

Melting point (°C) 1

Thermal expansion coefficient f) (K" ) Thermal conductivity (W/cm K)

Heterojunction bipolar transistors for millimeter wave applications

3

• GaAs is now available in 150 mm size and the wafers exhibit an acceptable defect density for medium integration. Its thermal conductivity is however the worst of the three. Its main advantage remains in its high resistivity. • InP suffers from low wafer size and from its brittle characteristics. This is a concern to produce low cost chips. However it shows good thermal and resistivity characteristics. The availability of high resistivity substrates constitutes a mere advantage to HI-V components by easing the interconnection losses and by decreasing device parasitic capacitance. This leads to high cut-off frequencies with a limited number of processing steps. However, the higher the operation frequency the lower the loss difference between the three substrates. Moreover semiconductor-on-insulator (soi) such as SMOX is an excellent way to reducing the device parasitic capacitance [3]. The use of thick polyimide as dielectric material associated to thick aluminium interconnection is also a solution to decrease the RF loss contribution of the silicon substrate [4]. Concerning the thermal conductivity, process engineers have tried to reduce its contribution by implementing specific actions such as thermal spreaders close to dissipating junctions [5] or mounting on high conductivity material such as A1N, diamond or metal [6]. This weakens the classical arguments against III-V substrate thermal behaviour.

ni. Heterojunction characteristics The efficiency of electron injection from the emitter to the base is strongly dependent on the bandgap discontinuity at the emitter-base heterojunction [2]. The ideal situation would be to have a heterojunction with a AEc (discontinuity of the conduction band) close to zero and a AEv (discontinuity of the valence band) as high as possible to avoid the injection of holes from the base to the emitter. Table II shows the main heterojunction available today for the three

Table II. Heterojunction band arrangement for different heterostructures. Physical Parameters

Unit

Substrate

As/ GaAs

Gao-siH..* P/GaAs

InP/Gao^ KssA8

Si/SW Ge^

GaAs

GaAs

InP

Si

A1

03Ga0.7

AEc

Electron Volt

0.24

0.17

0.26

0.04

AEv

Electron Volt

0.15

0.31

0.34

0.06

%

38

65

58

22

AEc/AEg

Millimeter waves in communication systems

kinds of substrates. The heterojunction which fulfills the requirement the best is obtained by the GaQ5]In049P/GaAs. Si/SiGe appears on the very low scale of efficiency. A close look at the device shows that the design of such HBT'S are often close to a standard bipolar (low concentration of Ge at the interface), but a germanium concentration grading in the base ensures a carrier transport assisted by an internal field leading to a drastic decrease of the base transit time [7]. The effect is important in the case of devices designed for digital applications where low Bvceo's (common emitter breakdown voltage) are targeted. In this case, the collector thickness is in the 200 nm range and the contribution of base transit time is important compared to the collector one. The transit time through the base is also very important for ultimate operations (Table III). The IQ-V base material show much better performances than SiGe. Table IV shows the figure of merit C^ of various collectors based on Gao and Morkoc/s model [8]. Cfm indicates the most important collector parameters (thickness, doping concentration, maximum electrical field and saturation velocity). This figure of merit is quite similar to Johnson's, but it does include thickness and doping impact. Therefore Cfm gives rise to a closer description of material limitation, since the latter figure of merit supposes the collector thickness infinite and the intrinsic properties of semiconductor are used. Gao and Morkoc/s approach could be refined by taking into account the overspeed effect that is quite important in ffl-V and the effect of temperature over carrier speed. CGaAs represents the figure of merit of GaAs. Si and GaAs appear very similar in terms of performances using this figure of merit while the others are 50% more efficientA refined model including the Kirk effect and its impact on electron overspeed would have favoured GaAs over Si. In conclusion, the intrinsic performances of ffl-V HBT'S are far better than the SiGe/Si ones. Only construction strategy can compensate these SiGe/Si drawbacks. The differences between InP- and GaAs-based HBT'S are smaller and the application will decide the selection.

Table III. Key factors during base transport. Symbol

Unit

Intrinsic electron mobility

M*

cm2 / Vs

Idem in doped base

Uno

Mh

Property

Hole mobility (Nb~5x 1019 cm'3)

GaAs

I%53G%47As

2000

8600

13000

cm2 / Vs

-100

-2000

-1000

cm2 / Vs

52

130

50

0 75 ^*" 0 25

Heterojunction bipolar transistors for millimeter wave applications

5

Table IV Electrical properties of collector materials. Material

Eg

«m

V s (@e = 10s V/cm)

cfm

cfm/ C-GaAs

(eV)

(x 105 V/cm)

(x 107 cm/s)

(xlO25 V cm~3/2 s-5/4)

Si

1.1

4.1

0.86

3.44

1.07

InP

1.35

4.4

1.1

5.02

1.56

GaAs

1.42

4.8

0.72

3.22

1.00

Alo.3 GaQ7As

1.78

6.5

0.86

5.45

1.69

fa

1.89

7.6

0.71

5.01

1.56

Unit

0.49

Ga

0.51P

IV. Structure and device topologies Figure 1 shows a schematic cross section of a ni-V HBT. HBT electronic transport, represented by the big arrow, is perpendicular to the wafer substrate. Four major advantages of these devices are the capability of high breakdown voltage, less demanding lithography, low 1/f noise and high transconductance. Vertical transport is directly involved in the first three advantages since the current is mostly flowing away from free surfaces (no surface breakdown and surface trap influences) and the speed of the device is fixed by the thickness of the different layers while the high transconductance is due to pn junction current control. The vertical transport explains therefore the excellent capabilities of HBT for power amplifier and oscillator applications. HBT'S require the use of heteroepitaxy and

Figure 1. Cross-section of a III-VHBT.

6

Millimeter waves in communication systems

of two types of dopants. III-V processes are mostly using the as-grown wafers produced by an epitaxial reactor including all the different layers used in the final devices; the processing involved the removal of unnecessary material and the deposition of ohmic contacts. In the case of Si technology, the process uses a more symbiotic scheme where doping, passivation, crystalline growth are tightly bound. The Si process engineer builds up the devices from a plain silicon wafer while the ffl-V ones use more an de-embedding approach. The reason stems from the higher constraint on thermal budget for ni-V materials (Table I) and the more complex mechanisms involved during processes like annealing, dopant diffusion or alloy ohmic contact formation. Despite these process approach discrepancies, the basics of the devices remain similar. IV.l. Base and collector optimization A standard HBT uses only one emitter-base heterojunction, that is, base and collector are made of the same material but for doping levels and type. As seen in Table IV, the collector has a strong impact both on the speed (vs) and the breakdown voltage of the component (em). This is one reason that the base-collector junction is also optimized either for speed and/or breakdown voltage by introducing a second heterojunction (double HBT or D-HBT). To emphasize the major difference between HBT and D-HBT, Figure 2 shows the f t (current gain cut-off frequency) and f^^ as a function of emitter-collector bias for InGaP/GaAs HBT and D-HBT with identical doping levels (2 x 1016 cnr3) and thickness (1 jjm) [14]. One can see that the D-HBT measurements have been made up-to 33V while they have been stopped at 18V. Above this bias, the avalanche has destroyed the device. The use of HBT instead of D-HBT can therefore drastically change the optimization and performances of the device.

Figure 2. ft andfmax as a function of emitter-collector bias for InGaP/GaAs HBT and D-HBT with identical collector doping levels (2 x 1016 cm~3) and thickness (1 \m).

Heterojunction bipolar transistors for millimeter wave applications

7

The mastering of D-HBT will be one of the key developments in coming years. The challenge is to obtain easy transport of the electrons outdiffusing from the base into the collector despite the presence of the AEc, which is not favourable for most heterojunctions. The lower the conduction band discontinuity the easier the electron collection. Digital or analog applications do not use the same optimization scheme since the digital ones are more focussed on current gain cut-off frequency f t while analog ones are seeking high maximum power gain, i.e. maximum oscillation frequency fm^. The classical law derived from 50's research is still widely used and is very valuable for designers:

Where Rb is the total RF base resistance and C^ is the collector-base capacitance of the active device. Ft is mostly governed by the time to cross the base and the collector, i.e. the base and collector thickness. To increase ft, at constant doping levels, bias conditions and material semiconductors, one can decrease both base and collector thickness. However, that will definitely lead to increase of base resistance and often of collector-base capacitance. This approach is typically observed for digital applications while the increase of collector thickness will lead to degradation of f t but a C^ lowering will give rise to fmax preservation. These thicker collector layer allow a higher breakdown voltage and this is important for applications requiring output power. This approach is generally chosen for analog components. Nevertheless, the classical boundary between analog and digital is no longer valid for operations above 20 GHz and devices are pushed to their limits. Table V shows a typical base and collector structure and critical device size for InP, GaAs and Si-based HBT'S. SiGe/Si is obviously a D-HBT since the base is in SiGe, which has a smaller gap than the silicon emitter and collector. The use of a SiGe collector is not possible due to lattice strain (SiGe crystalline lattice is not much different from the Si one) and lower breakdown voltage. Low Si/SiGe AEc eases this D-HBT solution (Table II). High performance circuits have been obtained using ultimate lithography and a process scheme for such circuits as the 0.14 pm emitter width indicates. Moreover a Ge grading into the base has been made. III-V D-HBT are also considered, in order to take advantage of higher maximum field both for InP- and GaAs-based HBT'S [12-13].

IV.2. Device geometry Usually the emitter is located above the base layer and this topology is called "emitter-up" (Figure 3). The area of the base-collector junction is at least three times larger than the base-emitter area. Consequently, the total base-collector capacitance CBC does not only include the intrinsic capacitance, but also two

8

Millimeter waves in communication systems

Table V Typical base and collector HBT structures for mmw applications. Device type

Units

Reference Application Heterojunction Base thickness

nm

Base composition grading Base doping level Collector thickness

InP-based

InGaP/GaAs

SiGe/Si

[9]

[10]

[11]

44GHzampl.

High speed device

32.7 GHz bandwidth

HBT

HBT

DHBT

80

30

20

No 3

cm~

3 x 10

nm

700 3

Yes

No 19

IxlO

20

200 16

Collector doping level

cm"

1 x 10

Undoped

Emitter size

|om

1

0.8

Base size

jam

~2

1.6

1 x 1019

Typ. 150

0.14

Figure 3. Base-collector capacitances in standard emitter-up HBT.

extrinsic elements CBC^X. This parasitic capacitance has a large impact on the RF performance since fmax depends directly on CBC and Rb {Equation (1)}. The demands for higher performances drive the developments towards reduction of the extrinsic capacitances. Emitter-up topology optimization Two main solutions have been explored to reduce the extrinsic collector capacitance in the frame of the emitter-up configuration.

Heterojunction bipolar transistors for millimeter wave applications

9

• Figure 4-A shows the reduction by increasing the resistivity of the extrinsic part of the collector by using ion implantation, or by using a dielectric such as silicon oxide. The extrinsic capacitance is therefore reduced and this reduction can be evaluated using plate capacitor equations. The dielectric constant of the underneath material should be also small as possible to reduce this capacitance. In the case of the ion implantation approach the value is equal to that of the semiconductor ones (er ~12), while in case of silicon oxide er ~3.9 or in the 3 range using polymers. • The first scheme is possible with some ffl-V semiconductors by selecting the proper ion species [16]. The process involvedis not easy since the ions cross the extrinsic base layer, which creates electrical and crystalline defects degrading the extrinsic base resistivity. Post-ion implantation P+ layer doping [15] or RTF annealing are sometimes achieved. • The use of dielectric layer is highly compatible with Si-based technology [1011]. In this case, the collector size is defined by selective growth through a dielectric window. • Figure 4-B shows the possible scheme to reduce the base-collector capacitance [9]. During the process, the collector layer is over-etched to remove extrinsic material. This should be done without etching the P+ layer. This approach is possible with D-HBT for instance since base and collector do not have the same electrochemical properties and etching selectivity can be found. The main drawback of this approach is the risk of manufacturing yield degradation. This is a requisite solution in case of InP-based HBT'S where the resistivity increase by ion implantation is more difficult to achieve than for GaAs-based ones. Collector-up topology Collector-up topology is another way of RF performance improvement by decreasing the base-collector extrinsic capacitance C^^. This has been proposed by H. Kroemer [2], winner of 2000 Nobel Prize for his work; more recently S. Luryi pushed further the concept [17]. The concept of collector-up transistor will lead to ultimate performances. With optimum design, of semiconductor material and process, those devices will reach fmax in the THz range. The inversion of the topology of the device using the "collector-up" topology is an appropriate solution to decrease the base-collector capacitance by a factor of three as Figure 5 shows. However a major difficulty of the "collector-up" topology is that the injecting base-emitter diode covers a larger area than the collecting base-collector junction. Electrons injected from the emitter into the extrinsic base regions will recombine in the base precluding the use of such devices if no specific actions are taken (Figure 6). Simple calculations shows that the current gain would be below 0.5 in the case of the schematic shown in Figure 6. Two kinds of solutions can be explored: • Lateral over-etching of the emitter below the base layer down to the width of the collector.

Figure 4. Emitter-up topology improvement: A: By using an insulating layer beneath the extrinsic base layer B: By lateral partial etching the collector layer below ohmic contact base stripes.

Heterojunction bipolar transistors for millimeter wave applications

11

Figure 5. Collector-up HBT -with collector-base capacitance shown.

• Ion implantation of the extrinsic emitter to change this part of this device into high resistivity material. • Transferred substrate HBT'S. It has to be stressed that the emitter size and its alignment below the collector will have a strong impact on the current gain and cut-off frequencies. This is the main difficulty today to achieve these devices. Most of the work has been focussed on IH-V HBT but that can be also realized using Si/SiGe D-HBT [18]. Lateral over-etching The lateral over-etching approach (Figure 7) is similar to the one described in 0. The realization is feasible if etching selectivity is available. Since the emitter and base layers are involved, that device might not be a D-HBT. Pelouard's team has

Figure 6. Parasitic current injection from the emitter to the extrinsic base.

12

Millimeter waves in communication systems

Figure 7. Collector-up HBT with over-etched emitter.

been largely developing this approach for the last 5 years. Demonstration of 160 GHz and above has been achieved [19-20]. Ion implantation This approach is well suited to GaAs-based HBT'S thanks to efficient increase of embedded emitter resistivity by ion implantation. Figure 8 shows the schematic of the device. The difficulty is to define the implantation condition to preserve the extrinsic base sheet resistance and to increase by several orders of magnitude the emitter resistivity [21-22]. Using this technique fj^ over 120 GHz has been successfully achieved with 2 jam wide collector.

Figure 8. Collector-up HBT with emitter size defined by ion implantation.

Heterojunction bipolar transistors for millimeter wave applications

13

Transferred substrate This is the more complex technology to be implemented but that gives rise to the ultimate achievable performance by HBT. M. Rodwell's team is far ahead with this technology [6] [23-24]. They have demonstrated fmax up to 820 GHz for elementary devices which is the dream for any millimeter wave designer. Figure 9 shows the principle of this technology: • A standard HBT process is carried out from the top-side and emitter and base contact and mesas are defined together with some interconnection. A direct contact of the emitter is made using a thermal shunt approach. A polymer with low dielectric constant is also used for signal propagation. • The ground plate is made using copper deposition in the 30-50 |jm range. This copper acts both as an electrical ground and a thermal sink. • The full wafer is then flipped and thinned down to the collector layer. • The collector mesa and contact are then made using high resolution lithography. This technology is quite complex and requires top equipment. For instance it involves both optical stepper and e-beam lithography. Emitter width of 0.6 urn and collector of 1.6 (jm are achieved. To ensure good electron collection the alignment should be as centered as possible. HBT'S are obtained using InP-based compounds and full 50 mm InP diameter wafers are processed. The number of processing steps of brittle InP with very advanced processing is impressive. However, fabrication yields sufficient to produce MMIC have been demonstrated.

Figure 9. Transferred substrate collector-up HBT principle.

14

Millimeter waves in communication systems

V. Electrical properties of the different devices Table VI gives an overview of what is believed to be the worldwide status of HBT technology for millimeter wave applications. In this table, the following data in the type raw are shown: Blank: Emitter-up HBT; D: DHBT; Cup: Collector-up structure; Sc: Schottky contact on the collector; Trans: Transferred substrate. The Schottky contact has not been introduced in this paper but it is a solution to decrease the collector resistance and to gain on thermal resistance [20, 24, 28]. This table shows the following information: • As the fmax and f t increase, the breakdown voltage decreases. This is due to the shrinking of the collector layer thickness. That gives rise to smaller output power per unit cell. • For all HBT'S but SiGe ones, the f^^ is higher than f t by sometimes a factor 4. • Collector-up's show in general low current gain. Reference [6] describes excellent current gain thanks to high quality lithography. SiGe collector-ups do not suffer from this present limitation maybe thanks to the availability of excellent SiO^Si interface and longer minority carrier lifetime in the base. • Millimeter wave HBT'S are still not in general pushing down the limit of lithography like FET'S. They will be less sensitive to process variation and good electrical homogeneity on wafer will be preserved. A classical factor to evaluate the different families of a device is to xplot fmax mo , as a function of the breakdown voltage of the devices, as for analog applications the gain and power are the key parameters. Figure 10 shows this graph extracted from the previous table. M. Rodwell's team has already increased the performance since they have recently achieved 820 GHz f^^ HBT'S with their transferred technique. The extremely high performance exceeds the present record using HEMT devices

Figure 10. f

versus common base breakdown voltage of different HBT technology.

Table VI. Key parameters currently reached for RF applications. Mat.

Unit

AlGaAs

SiGe

InGaP

InP

D

T^pe

InGaP

SiGe

AlGaAs

InGaP

InGaP

D

D-Cup

Cup

D-Cup

Sc-Cup

InP

InP

Sc

Cup

Cup

Trans

Trans

Ref

25

26

10

9

14

18

22

27

28

24

6

Ft

GHz

45

100

105

80

12

28

68

22

31

22

164

Fmax

GHz

100

65

120

200

33

18

128

115

110

61

500

BVcbo

V

16

4

6

13

45

12

12

35

16

4

2

15

100

10

25

15

150

10

3

3

2.7

50

2

1

0.8

1

2

1.6

2

2

2

1

0.8

P Size

|jm

16

Millimeter waves in communication systems

which is in the 600 GHz range. The wide spread of breakdown voltage is mostly due to the different applications which are targeted. The dashed line shows the present limit which is not going to stay that way with device size shrinking. The niV's are always much better for analog millimeter applications than SiGe. InGaP/GaAs and InP-based HBT'S are the best present performing devices. However GaAs foundries will focus on those applications as the market will grow. If one trusts the computations from Table IV, GaAs DHBT'S - most likely InGaP ones thanks to their better reliability - will be widely studied to fulfill future needs.

VI. Circuits A lot of high performances circuits have been demonstrated. As HBT technology is in everyday progress, just a snap shot will be provided. The list demonstrates that HBT'S are capable of worthwhile millimeter wave applications. Functions of up to 108 GHz have been successfully realized. InP HBT'S are now in advanced in this field with a strong contribution by Californian research at Hughes Research Laboratory and at UCLA. VL1. SiGe/Si emitter-up HBT • Design of a 32.7 GHz bandwidth AGC amplifier (19dB, eye pattern up to 25Gb/s)[ll]. VI.2. AlGaAs/GaAs emitter-up HBT • 50GHz bandwidth base-band amplifiers (49.3 GHz bandwidth with a transimpedance gain of 43.7 dBQ) [29]. VI.3. InGaP/GaAs emitter-up HBT • Ka-Band 4-stage low noise amplifier (16 dB gain, 5 dB noise) [30]. • Monolithic 38 GHz balanced reflection-type direct carrier modulation (BPSK) [31]. VI.4. InP-based emitter-up HBT • 2-32 GHz coplanar waveguide InAlAs/InGaAs-InP cascode distributed amplifier (5 dB gain) [32]. • 108 GHz monolithic push-pull vco with low phase noise and wide tuning bandwidth (output power 0.92 dBm, bandwidth 2.73 GHz, phase noise of - 88 d Be/Hz at 1 MHz of carrier) [33]. VI.5. InP-based emitter-up with transferred substrate • 48 GHz digital IC's and 85 GHz baseband amplifiers [34].

Heterojunction bipolar transistors for millimeter wave applications

17

VII. Conclusion A description of the HBT technology has been given and one has tried to indicate what are the current limitations of HBT'S and what has to be made to improve their RF gain. The key interests of HBT remain in their weak lithography requirements, the vertical transport realising high breakdown voltage and high uniformity in terms of electrical performance over all the wafer for short and long distance. All these advantages are extremely important in analogue and digital applications. Fjjj^ as high as 820 GHz have been obtained using highly sophisticated technology while more standard ones are able to give rise to components with 200 GHz fmax' which is quite reasonable for current millimeter wave applications. The best results are today obtained using InP substrates. This might be a temporary situation since GaAs has the capability to work at higher frequency if one sacrifices the present breakdown voltage. These devices would be in this case much more affordable thanks to the GaAs substrate size. Acknowledgements The author would like to thank the HBT team of Thomson-csF/LCR without whom understanding of the physics and technology peculiarities would not have been possible. UMS colleagues in charge of production of InGaP/GaAs HBT'S are deeply acknowledged for they are those who are giving birth to HBT products. The French and European Institutes working with Thomson-csF/LCR will not be forgotten (EMM, IRCOM, LAAS, CNRS-LPSC, Daimler-Chrysler Research Centre Uhn, Technische Universitat Darmstadt, Technische Universitat Chemnitz, NMRC etc) The worldwide contribution of workers over 30 years on HBT'S is the foundation of this paper and that must be acknowledged. The long term support of Thomson-csF, European Union, French DGA, ONES and their representative is also acknowledged. REFERENCES [1] SCHOCKLEY (W.), Circuits element utilizing semiconductor material, U.S. patent. 2569347,(1951). [2] KROEMER (H.), Heterostructure bipolar transistors and integrated circuits, Proc. IEEE. (1982), 70, n° 1, pp. 13-25. [3] FERLET-CAVROIS (V.), MARCANDELLA (C), MUSSEAU (O.), LERAY (J.L.), PELLOIE (J.L.), MARTIN (R), KOLEV (S.), and PASQUET (D.), High-frequency performances of a partially depleted 0.18 urn SOI/CMOS technology at low supply voltage - Influence of parasitic elements, IEEE Elec. Dev. Lett. (1998), 19, n° 7, pp. 265-267. [4] KIM (B.K.), Ko (B.K.), LEE (K.), JEONG (J.W.), LEE (K.S.), and KIM (S.C), Monolithic planar RF inductor and waveguide structures on silicon with performance comparable to those in GaAsM/w/c. Elect. Dev. Meeting., (1995), International.

18

Millimeter waves in communication systems

[5] DHONDT (F.), BARRETTE (J.), HAESE (N.), ROLLAND (P.A.), DELAGE (S.L), Finite-Element electromagnetic characterization of parasitics in multifinger thermally shunted HBT'S, /£££ Microwave Guided Wave Z.ett.(1998), 8, n° 4, pp. 167-169. [6] PULLELA (R.), LEE (Q.), AGARWAL (B.), MENSA (D.), GUTHRIE (J.), SAMOSKA (L.), RODWELL (M.), A > 400 GHz fmax transferred-substrate HBT integrated circuit technology., Device Research Conference. (1997), Fort Collins, Co, USA. [7] HARAME (D.), COMFORT (J.H.), CRESSLER (J.D.), CRABBE (E.F.), SUN (J.Y.-C), MEYERSON (B.S.), TICE (T.), Si/SiGe epitaxial-base transistors - Part I: Material, Physics, and circuits, /£££ Trans. Elec. Dev. (1995), 42, n° 3, pp. 455-468. [8] GAO (G.B), MORKOC (H.), Material based comparison for power heterojunction bipolar transistor, IEEE Trans. Elec. Dev. (1991), 38, n° 11, pp. 2410-2416. [9]

KOBAYASHI (K.W.), NlSHIMOTO (M.), IRAN (L.T.), WANG (H.), COWLES (J.), BLOCK (T.R.),

ELLIOTT (J.), ALLEN (B.), OKI (A.K.), STREIT (D.C.), A 44 GHz InP-based HBT doublebalanced amplifier with novel current Re-use biasing, /£££ Radio Freq. Integ. Circ. Symp. (1998), pp. 267-270. [10] OKA (T), OUCHI (K.), UCHIYAMA (H.), TANIGUCH (T.), MOCHIZUKI (K.), NAKAMURA (T.), High-speed InGaP/GaAs heterojunction bipolar transistors with buried SiO2 using Wsi as the base electrode, /£££ Elec. Dev. Lett. (1997), 18, n° 4, pp. 154-156. [11] OHHATA (K.), MASUDA (T.), OHUE (E.), WASHIO (K.), Design of a 32.7-GHz bandwidth AGC amplifier ic with wide dynamic range implemented in SiGe HBT, IEEE J. Solid-St. Circ. (1999), 34, n° 9, pp. 1290-1297. [12]

NGUYEN (C), Liu (T.), CHEN (M.), SUN (H.C.), RENSCH (D.), Al I n As/Gain As/I nP double heterojunction bipolar transistor with a novel base-collector design for power applications, IEEE Elec. Dev. Lett. (1996), 17, n° 3, pp. 133-135.

[13]

HENKEL (A.), DELAGE (S.L.), diFoRTE-PoissoN (M.-A.), CHARTIER (E.), BLANCK (H.), HARTNAGEL (H.L.), Collector-up GalnP/GaAs double heterojunction bipolar transistor with high fmax. Electronics Letters. (1997), 33, n° 7, pp. 634-635.

[14] Thomson-csF/LCR internal results, courtesy of CHARTIER (E.). [15] SHIMAWAKI (H.), AMAMIYA (Y.), FURUHATA (N.), HONJO (K.), High-fmax AIGaAs/lnGaAs and AlGaAs/GaAs HBT'S with p+/p regrown base contacts, /£££ Trans. Elec. Dev. (1995), 42, n° 10, pp. 1735-1743. [16] WANG (N.L), SHENG (N.H.), CHANG (M.F.), Ho (W.J.), SULLIVAN (G.J.), SOVERO (E.A.), HIGGINS (J.A.), ASBECK (P.M.), Ultrahigh power efficiency operation of commonemitter and common-base HBT'S at 10 GHz, /£££ Trans. Microwave Theory, and Techn. (1990), 38, n° 10, pp. 1381-1390. [17] LURYI (S.), How to make an ideal HBT and sell it too, /£££ Trans, on Elec. Dev. (1994), 41, n° 12, pp. 2241-2247. [18] GRUHLE (A.), KIBBEL (H.), MAHNER (C.), MROCZEK (W.), Collector-up SiGe heterojunction bipolar transistors, /£££ Trans. Elec. Dev. (1999), 46, n° 7, pp. 15101513. [19]

MATINE (N.), PELOUARD (J.-L), PARDO (F.), TEISSIER (R.), PESSA (M.), Novel approach for InP-based ultrafast HBTS, 8th Conf. on Indium Phosphide and re/. MatiRPM '96, pp. 137-140.

Heterojunction bipolar transistors for millimeter wave applications

19

[20] MATINE (N.), Realisation et caracterisation de transistors bipolaires a heterojonction InP/lnGaAs/Metal (structure MHBT). PhD thesis, University Paris XI, U.F.R. Scientifique d'Orsay, (1996). [21] HENKEL (A.), DELAGE (S.L), diFoRTE-PoissoN (M.-A.), BLANCK (H.), HARTNAGEL (H.L.), Boron implantation into GaAs/Ga0 5ln0 5P heterostructures, Jap. J. ofAppl. Phys. (1997), 36, n° 1A, pp. 175-180. [22] YAMAHATA (S.), MATSUOKA (Y.), ISHIBASHI (T.), High fmax collector-up AlGaAs/GaAs heterojunction bipolar transistors with a heavily carbon-doped base fabricated using oxygen-ion implantation, IEEE £/ec Dev. Lett. (1993), 14, n° 4, pp. 173-175. [23] BHATTACHARYA (U.), MENSA (D.), RODWELL (M.J.W.), 170 GHz transferred-substrate heterojunction bipolar transistor, Electron. £ett.(1996), 32, n° 15, pp. 1405-1406. [24] BHATTACHARYA (U.), MONDRY (M.J.), HURTZ (G.), TAN (I.-H.), PULLELA (R.), REDDY (M.), GUTHRIE (J.), RODWELL (M.J.W.), BOWERS (J.E.), Transferred substrate Schottkycollector Heterojunction bipolar transistors: first results and scaling laws for high fmax, IEEE Elect. Dev. Lett. (1995), 16, n° 8, pp. 357-359. [25] WANG (N.-L.), Ho (W.-J.), HIGGINS (J.A.), Millimeter wave AIGaAs-GaAs HBT power operation, IEEE Microwave guided wave Lett. (1992), 2, n° 10, pp. 397-399. [26] KASPER (E.), GRUHLE (A.), AND KIBBEL (H.). High speed SiGe-HBT with very low base sheet resistivity, IEDM 93. pp. 79-81. [27] HENKEL (A.), DELAGE (S.L.), DifoRTE-PoissoN (M.-A.), CHARTIER (E.), BLANCK (H.), HARTNAGEL (H.L), Collector-up GalnP/GaAs double heterojunction bipolar transistor with high fmax, Elec. Lett., 33, (1997), pp. 634-635. [28] GIRARDOT (A.), HENKEL (A.), DELAGE (S.L.), DIFORTE-POISSON (M.-A.), CHARTIER (E.), FLORIOT (D.), CASSETTE (S.), Transistor a heterojonction GalnP/GaAs a hautes performances hyperfrequences en topologie collecteur en haut avec contact Schottky, 11th Jounr4es Nationales Microondes, (1999), pp. 1A4. [29] SUZUKI (Y), SHIMAWAKI (H.), AMAMIYA (Y), NAGANO (N.), NIWA (T.), YANO (H.), HONJO (K.), 50-GHz bandwidth base-band amplifiers using GaAs-based HBTS, IEDM, (97), pp. 143-146. [30] FREUNDORFER (A.P.), JAMANI (Y), FALT (C), A Ka-Band GalnP/GaAs HBT four-stage LNA, IEEE Microwave and Millimeter Wave Monolithic Circ. Symp. (1996), pp.141-144. [31] NAM (S.), SHALA (N.), ANG (K.S.), ASHTIANI (A.E.), GOKDEMI (T.), ROBERTSON (I.D.), MARSH (S.P.), Monolithic millimeter-wave balanced bi-phase amplitude modulator in GaAs/lnGaP HBT technology, IEEE MTT-S Digest, (1999), pp.243-246. [32] KOBAYASHI (K.W.), COWLES (J.), TRAN (L), BLOCK (T.), OKI (A.K.), STREIT (D.C.), A 2-32

GHz coplanar waveguide InAlAs/lnGaAs-lnP HBT cascode distributed amplifier, IEEE Microwave and Millimeter Wave Monolithic Circ. Symp. (1995), pp. 195-197. [33] KOBAYASHI (K.W), OKI (A.K.), TRAN (L.T.), COWLES (J.C.), GUTIERREZ-AITKEN (A.), YAMADA (F.), BLOCK (T.R.), STREIT (D.C.), IEEE J. Sol. State Ore. (1999), 34, n° 9, pp. 1225-1232. [34] MENSA (D.), PULLELA (R.), LEE (Q.), GUTHRIE (J.), MARTIN (S.C.), SMITH (R.P.), JAGANATHAN (S.), MATHEW (T.), AGARWAL (B.), LONG (S.I.), ROOWELL (M.), 48-GHz digital IC's and 85GHz baseband amplifiers using transferred-substrate HBT'S, IEEEJ. of Sol. State Circ. (1999), 34, n° 9, pp. 1196-1202.

Chapter 2

HEMT's capability for millimeter wave applications S. Bollaert, Y. Cordier, M. Zaknoune, T. Parenty, H. Happy and A. Gappy Institut d'Electronique et de Microelectronique du Nord, Departement Hyperfrequences & Semiconducteurs, Cite Scientifique, France

I. Introduction From the first demonstration of mobility enhancement in modulation-doped heterostructures in 1978 [1], we have witnessed an explosion of research and development on the Modulation-Doped Field Effect Transistor (MODFET), also as known as the High Electron Mobility Transistor (HEMT) and Two-dimensional Electron Gas Field Effect Transistor (TEGFET) for millimeter wave device and circuit applications. At the present time, HEMT is the fastest three-terminal semiconductor device in the world. In the last ten years (1990-2000), the speed of state-of-the-art HEMTS, which is commonly characterised by their extrinsic current gain cut-off frequency fT, has increased from 250GHz to 350GHz [2-3]. These significant improvements have been made possible by the realisation of nanometer gate length, as short as 30nm, defined by electron beam lithography. As a result of these improvements, applications at V-and W-bands and beyond, using ultra-fast HEMTS in MMICS, are quite feasible. Applications such as passive imaging systems, Local Area Networks (LANS), Local Multipoint Distribution Systems (LMDSS), car radar has result in an explosive demand for millimeter wave Low noise amplifiers (LNAS) and power amplifiers. In addition, with rapid developments in commercial satellite constellations for wide bandwidth communication, the demand for low-cost millimeter wave MMICS will continue to increase, an LNA working at 190GHz has been demonstrated [4]. There are four types of HEMTS commonly used for millimeter wave MMIC applications: AlGaAs/InGaAs pseudomorphic HEMTS on GaAs substrate (PMHEMTS GaAs), lattice-matched and pseudomorphic InAlAs/ InGaAs HEMTS on InP substrate (LM-and PM-HEMTS InP) and an emerging metamorphic InAlAs/InGaAs HEMTS on GaAs substrate (MM-HEMTS). InP-based HEMTS have demonstrated a highest f T of 350 GHz [3] and the highest fmax of more than 600

HEMT's capability for millimeter wave applications

21

GHz [5]. Moreover, InP-based HEMTS offer the best low-noise performance at Wband [6]. For applications, where cost rather than performance is the driving factor, GaAs based HEMTS such PM-HEMTS are still used. Maximum output power density at 60 GHz of IW/mm has been reported [7-9]. Due to the larger diameter GaAs substrate and more mature technology of PM-HEMTS on GaAs, the cost can be reduced. Moreover InP substrate is more fragile than GaAs one. An alternative is the emerging MM-HEMTS on GaAs, using the InAlAs/InGaAs heterostructure. In this paper, we will first describe the HEMT'S quality factors to obtain high low-noise and power performance. In second part, we will present the most commonly HEMTS used in millimeter wave ic's applications. Finally, we will give status of state-of-the-art on cut-off frequencies, low-noise and power performance of these devices.

H. HEMT's parameters n.l. HEMT's structure: modulation-doped heterostructures A cross section of HEMT is shown in Figure 1. It consists on different layers of semiconductor. Ohmic contacts are deposited on the cap layer, which has generally high doping level to achieve lower ohmic contact resistances. A second layer with high bandgap is necessary to provide good Schottky barrier height with the gate metalization. This layer also serves to form the modulation-doped heterojunction with the channel. A modulation-doped heterojunction consists of an abrupt junction between two semiconductors (Figure 2), which have different electron affinities, %1 > %2, and in which only the semiconductor with the smaller affinity is doped n-type

Figure 1. Typical layer structure of HEMT.

22

Millimeter waves in communication systems

Figure 2. Diagram energy band formation between two semiconductors.

(bulk or delta-doped). The difference in electron affinities translates into a discontinuity of the conduction band AEc at the heterojunction, where:

To simplify this expression, the conduction band discontinuity is also related to the difference of bandgap, which gives:

(2)

= Egl-Eg2

As a consequence, this discontinuity gives a potential well near the heterojunction interface. Electrons initially introduced in the doped material diffuse into the potential well, where they are confined. This creats a twodimensional electron gas (2DEG). Sheet electron densities ns higher than 1012 cm~2 can be obtained at the interface. This value depends partly on the depth of

HEMT's capability for millimeter wave applications

23

the potential well. For higher conduction band discontinuity, a high level of sheet carrier density is achieved in the well. A large bandgap of semiconductor 1 and (or) small bandgap for the channel will lead to large numbers of electrons in the channel. Impurity scattering is greatly reduced simply by separating electrons from the ionised donors. Using this technique, transport in the 2DEG should approach that of undoped bulk material. This behaviour leads to the superiority of HEMTS on MESFET in microwave performance. This phenomenon can be improved by the use of an undoped layer, called spacer, between the doped layer and the active channel (see Figure 1), which leads to reduction of coulomb scattering. H.2. HEMT's quality factors For microwave applications, the main figures of merit are f T the cut-off frequency of extrinsic current gain (deduced from H2l = 1), f^^ the maximum oscillation frequency (obtained for Mason's unilateral gain U = 1 or maximum available gain MAG =1). These parameters are related to the cut-off frequency of intrinsic current gain fc, and to the parasitic elements of the device. n.2.1. High cut-off frequency fc: geometrical and material parameters

The cut-off frequency f c can be expressed as:

This term is inversely proportional to the transit time of electron under the gate. f c can also be written as: (4)

f=

27t(L+2A)

In this expression, , L and A represent respectively the average electron velocity in the channel, the gate length and the distance gate to channel. In this expression, the term 2A represents the contribution of depleted areas around the gate, on the gate-to-source capacitance. This will lead to parasitic capacitance, called fringe capacitance. A lower aspect ratio —- will degrade the cut-off freA quency. Short gate length and good electron transport properties are required for high cut-off frequency. Using optical lithography (uv or deep uv), gate lengths close to 0.25 (jm can be obtained. Down to 0.25 jam size, electron beam lithography is widely used today. Gate lengths of 30 nm are reached today under laboratory conditions. To achieve high electron velocity, a material with good electron transport properties has to be chosen for the conducting channel. Figure 3 shows the energy-band structure of GaAs. In such material, the lowest minimum of the conduction band, corresponding to the F valley, is at the same point as the top of

24

Millimeter waves in communication systems

Figure 3. Energy band structure ofGaAs.

the valence band. The difference between energy of both points is the gap. Lateral bands X, and particularly L, have a strong effect on the transport properties of electrons. Indeed, the effective mass of an electron increases as the electron reaches the lateral band L. This effect is related to the shape of lateral bands. This will drastically reduce the transport property of electron. So a larger intervalley separation will lead to better transport properties. This explains why the InGaAs material is well suited for the conducting channel, particularly for high Indium content. Indeed the mobility and the PL intervalley separation (Ae^) increase (low electron effective mass), when the Indium content increases (Figure 4). Lfi Regarding the gate to channel distance A, a high aspect ratio -f- has to be /\

chosen to obtain good micro wave performance. Moreover, a poor aspect ratio L —- involves shortchannel effect, such as the injection of carrier in the buffer. A Short channel effect is responsible for pinch-off voltage shift, and for the increase of output conductance. This last effect will particularly degrade maximum Lg oscillation frequency. This geometrical parameter —^ should be maintained above five [161.

HEMT's capability for millimeter wave applications

25

Figure 4. Low field mobility and PL intervalley separation in In^a^^s material versus the Indium content x. Energy bandgap is also represented.

n.2.2. Small signal equivalent circuit and cut-off frequencies Figure 6 shows the small signal equivalent circuit of a HEMT. This circuit consists on intrinsic and extrinsic elements. These elements depend on technological process, geometrical dimension of the transistor and on the material properties used. For example, source resistance is an extrinsic element. The value of this element is a combination of geometrical parameters, such a width and as sourceto-gate distance and square resistance of the material and the ohmic contact (Figure 5). This last parameter is a technological parameter. It depends on the metal used to form the contact and the condition of annealing. These elements will determine the cut-off frequencies of the device. Analytical expressions for extrinsic cut-off frequency f T and maximum oscillation frequency fmax are given below [10-11]:

26

Millimeter waves in communication systems

Figure 5. Cross section O/HEMT. where Rj, Rs and R are respectively channel, source and gate resistances. These expressions show that to achieve high performance, HEMT have to exhibit a high intrinsic cut-off frequency fc, low parasitic capacitance and resistance, small output conductance gd and small gate-to-drain capacitance C d In other words, high voltage gain — and high —— ratio are required. The output conductance §d C^ gd and gate-to-drain capacitance C d mainly depend on the device aspect ratio Lg (gate length over gate to channel distance —- ) the recess shape and the buffer A

Figure 6. Small signal equivalent circuit O/FET.

HEMT's capability for millimeter wave applications

27

structure. Degradation of gd and C d are related to the so-called short channel effect, particularly important in a sub-micron device and also in millimeter devices. To ensure low gd and C d a device aspect ratio larger then 5 is required. Parasitic capacitances are mainly pad capacitances CLj anC which depend on the geometry of the bonding pads. Other capacitances are fringing capacitances, which are an extension of the depleted area around the gate and the electrical coupling between the gate and the access region, via the air or the passivation layer (Figure 5). The magnitudes of all these parasitic capacitances can be reduced by proper design of the transistor topology and by the technological process. In the millimeter wave range, the reduction of the parasitic capacitance is a key issue for ultra-short gate length devices. Expressions 5 and 6 have also a large dependence on access resistances Rs and Rd. These elements are related to the ohmic contact Rc, which is determined by the technological process. 0.1 to 0.5 Q.mm are commonly obtained with such devices. In addition, the distance from the source-to-gate L (L^ for drain-togate), device width W and square resistance Rsquare of the material used, increase access resistances Rs (and Rd). The values of resistances can be simply expressed as: /7\

D __2. i D

^ '

W

s

s< uare

l

Sfi

\Y

Gate resistance R is an important limitation factor to reach high microwave and low-noise performance particularly in sub-quarter micron device. Indeed, gate resistance is largely related to the gate length. To ensure low value of gate resistance, a T-shaped gate has to be used (Figure 5). Another way to reduce gate resistance is to increase the number of gate fingers (Equation 8).

where n is the number of gate fingers, Rm the gate metalization resistance which depends on the resistivity of the metals used and the cross section area of the gate. n.2.3. Low-noise parameters of HEMT For low-noise application, a low noise figure NF with a high associated gain Gass is required. The noise figure of a two-port network, in the original Friis definition [12], is defined as the ratio of the available signal-to-noise ratio at its input terminals to that its output:

(9,

NF=

S./N.

28

Millimeter waves in communication systems

where Sj, Nj, S0 and N0 are the signal and noise available powers at the input and the output of the device. At the standard input temperature T0 = 290 K, this expression becomes: (10)

NF=1+^ l

o

where Te is the equivalent noise temperature of the device. For optimal generator admittance Y t, the noise figure will be minimised. The minimum noise figure NF^ is directly related to the minimum noise temperature T^ of the device. Different noise models define this minimum temperature. For example, the Hughes model [13] gives an analytical expression for associated gain and minimum noise temperature as: (11) \ll>

(12) where T and Td are the equivalent gate and drain temperatures respectively. Accurate determination of the noise temperatures is very difficult, and no evident correlation between these temperatures and technological parameters was yet clearly shown. Expressions 11 and 12 demonstrate the large dependence of noise performance on maximum oscillation frequency fmai. To obtain high noise performance devices, a high intrinsic cut-off frequency f c is required (expression 6). This explains the large dependence of noise figure on the gate length of the device. Other elements, such as the parasitic resistances Rs and R , degrade the noise figure. With a small gate length device, a T-shaped gate required to avoid a noisy device. II.2.4. Power parameters of HEMT

The HEMT, with its high current density and cut-off frequency, is an ideal candidate for millimeter-wave power amplifiers. The power performance of a FET is generally characterised in term of maximum output power, associated gain and power added efficiency (PAE). Maximum output power depends mainly on maximum drain current and drain-to-source breakdown voltage. To provide large drain current, HEMTS have to show high 2DEG density, which is related to large conduction band discontinuity AEc. Moreover, for power transistors the structure has been enhanced by the incorporation of a second doping plane. This double heterostructure design allows an increase of 30% in the charge density and consequently the maximum current density. In the same way, large device width using many gate fingers is necessary to obtain large drain current. The breakdown voltage mainly depends on the energy bandgap of the materials used for

HEMT's capability for millimeter wave applications

29

the device. Large bandgap materials are required for high breakdown voltages. The shape (height, width) of the recess is also important. In effect the design of the recess permits relaxation of the maximum of the electric field placed between the gate and the drain to increase the breakdown voltage. Doublerecessed structures are generally used to obtain large drain voltage operation.

III. HEMT's structure for millimeter applications V- and W- band applications required the use of high-speed devices. Conventional AlGaAs/GaAs HEMTS and pseudomorphic AlGaAs/InGaAs/GaAs HEMTS present limited performance for millimeter wave applications. Drawbacks of these devices are the poor sheet carrier density ns and/or the weak low field mobility in the channel. The main transistor families developed today in applications up to W-band and beyond are based on the InAlAs/InGaAs heterostructure. These devices provide low noise and useful power gain up to the millimeter wave frequencies. This is due to the good electron transport properties in the InGaAs material used as the channel (Figure 4), and the larger conduction band discontinuity AEc, which involves large sheet carrier densities. This heterostructure can be realised on InP and GaAs substrates. This will lead to the families being lattice-matched and pseudomorphic on InP, and metamorphic on GaAs. III.l. Pseudomorphic AlGaAs/InGaAs HEMTs on GaAs Pseudomorphic HEMT on GaAs is based on the AlGaAs/InxGa1_xAs heterostructure (Figure 7). In this structure, the AlGaAs material is the Schottky layer, and InGaAs is the channel layer. AlGaAs material has almost same lattice constant than GaAs. The In^Ga^As material has a larger lattice constant than GaAs and AlGaAs (Figure 8). This difference increases with Indium content x. Then it can be grown pseudomorphically on GaAs substrate, as long as its thickness is less than a critical thickness, at which the strain in the InGaAs begin to relax and create misfit dislocations [14], lowering electron transport properties. Mismatched lattice will limit Indium content to 0.3 and leads to a maximum 2DEG density of 2.9xl012cnr2 [15]. m.2. Lattice-matched and pseudomorphic HEMTs on InP The Indium Phosphide (InP)-based HEMTS, in particular the I n y M j As/Ii^Ga^As HEMTS, have demonstrated its high-speed potential for applications up to millimeter wavelengths. A principal cross section of a basic configuration of an InP-based HEMTS is shown in Figure 9. The layer structure is composed of a semi-insulating InP substrate followed by an InAlAs buffer layer, an InGaAs channel layer, an InAlAs spacer, a pulsed-doped donor layer, a Schottky barrier layer and finally an InGaAs cap layer. The standard composition for

30

Millimeter waves in communication systems

Figure 7. Typical layer structure [26] of pseudomorphic HEMT on GaAs substrate (PM-HEMT GaAs). The delta-doped (Sdoped) is the n-type donor (in our case Silicium) density per square meter. Typical value is S.IO16/™2.

Indium is x = 0.53 for the InGaAs channel and y = 0.52 for the InAlAs Schottky layer. Using these Indium compositions, both layers have same lattice parameter as the InP substrate (Figure 8). Those layers are lattice matched to the InP substrate. That provides the so-called lattice-matched HEMTS on the InP substrate. The value of conduction band discontinuity AEc is close to 0.5 eV and leads to a 2DEG density higher than 3.1012 cnr2 [16]. One way to increase performance of HEMTS on InP substrate is to increase the Indium concentration in the channel. This will lead to better transport properties of the carrier. Indeed, material with higher Indium content exhibits larger intervalley separation (Figure 4). For Indium content of 0.8 in the channel, mobility of 15000cm2/V.s has been reported at room temperature. In comparison, the value is approximately 9000cm2/V.s in InGaAs material for x = 0.53 [16]. Moreover, the conduction band discontinuity is involved, and higher 2DEG concentration is obtained compared with the lattice-matched structure. A drawback is that higher Indium content will cause a lattice mismatch between the InAlAs and InGaAs layers and with respect to the InP substrate. This lattice mismatch leads to strain in the InGaAs channel. The concept of using strained layer to to improve the performance of HEMTS has generated a new type of HEMTS, the pseudomorphic HEMTS. However a limitation to the applicability of strained layer exists, the existence of a critical layer thickness up to which the lattice strain can be accommodated without deteriorating carrier transport properties. A strong effect of strain is the formation of dislocations in the crystal, which will affect transport properties. As

HEMT's capability for millimeter wave applications

31

Figure 8. Energy bandgap versus lattice parameter ofInxGa}}lAs and InyAlj As materials for Indium content varying from 0 to 1. AlGaAs is given is plotted for different Aluminium content. Conduction band discontinuity AEc of the In^lj ^As/InyGoj As is also represented.

a consequence, the channel thickness in a pseudomorphic structure should not exceed 100 to 150A. For high Indium content, large lattice mismatch restricts the use of Indium composition higher than 0.8. III.4. Metamorphic HEMTs on GaAs For low noise and power applications in the millimeter-wave range, metamorphic HEMTS (MM-HEMTS) using an InxAl1_xAs/InxGa1_xAs heterostructure grown on GaAs [17-18] constitute a good alternative to pseudomorphic AlGaAs/InGaAs/GaAs (PM-HEMTS) and to lattice matched InAlAs/InGaAs/InP HEMTS (LM-HEMTS). An advantage of MM-HEMTS is the use of a GaAs substrate, which is more suitable for low cost MMIC production. In fact, a GaAs substrate is larger and mechanically more robust than an InP substrate. However, LM-HEMTS on InP substrate exhibit higher cut-off frequencies and better low-noise performance than PM-HEMTS on a GaAs substrate, especially due to the Indium content channel limitation in PM-HEMTS (high electron effective mass). One solution is therefore to realise an InAlAs/InGaAs heterostructure with high Indium content on a GaAs substrate. This can be achieved using a metamorphic buffer to accommodate

32

Millimeter waves in communication systems

Figure 9. Typical layer structure of lattice-matched HEMT on InP substrate (LM-HEMT InP). A higher Indium content in the InGaAs layer channel will give pseudomorphic HEMT on InP (PM-HEMT InP). The delta-doped (Sdoped) is the n-type donor (in our case Silicium) density per square meter. Typical value is

the lattice mismatch between the active layers and the GaAs substrate. In addition, the use of a metamorphic buffer allows us to grow high quality unstrained Ii^Al^As/InjGa^As heterostructures with almost any Indium content x. The Indium content can then be considered as a new structural parameter. The use of a metamorphic buffer means that an heterostructure can be realised with any Indium content. Indeed, the metamorphic buffer serves to accommodate the lattice mismatch between the active layers and the GaAs substrate. Figure 8 shows the energy bandgap versus the lattice parameter of the In Alj As and InxGa1_xAs materials. To achieve the growth of the In Al^^As/Ii^Ga^As heterostructure, the Indium content of both materials must be virtually equal (x ~ y). Using a simple rule A£C/A£' =0.65 [19], the conduction band discontinuity has been plotted versus the mttice parameter (Figure 8). It can be observed that the conduction band discontinuity AEC increases with decreasing of Indium content, with a peak value of 0.7 eV when the Indium content x is close to 0.3. Below this Indium content, the InAlAs bandgap becomes indirect and the AEC decreases. The reduction of AEC involves a lowering of sheet carrier density of the two-dimensional electron gas (2-DEG). Moreover, for x < 0.3, the transport properties (mobility, intervalley separation Afij^J) in the InGaAs channel are rather low. To avoid an indirect gap, Indium contents x, lower than 0.33 were not considered. On the other hand, higher Indium contents will improve the transport properties in the InGaAs channel, so improving the microwave performance of

HEMT's capability for millimeter wave applications

33

Figure 10. Typical layer structure of metamorphic HEMT on GaAs substrate {MM-HEMT GaAs). In this structure, an Indium content between 0.3 and 0.6 can be chosen. The delta-doped (Sdoped) is the n-type donor (in our case Silicium) density per square meter. Typical value is S.IO16/™:2.

the device. However, too high a value of Indium content x will drastically increase the impact ionisation in the InGaAs channel because of its low energy bandgap. This will result in a high output conductance gd, which will not result in a high maximum oscillation frequency f^^ for the device. Another effect of higher Indium content is the reduction of the Schottky barrier quality due to the small energy bandgap of InALAs. In addition, at higher Indium contents, the larger mismatch between the GaAs substrate and the active layer makes it harder to grow metamorphic structure. For x = 0.3, the mismatch between GaAs and the active layer is about 2% and reaches a value higher than 3% for x = 0.5 [20]. Upper value Indium content x at 0.6 can not be used. Using the metamorphic concept, the Indium content of the unstrained Ir^Al^As/hijGa^As heterostructure can be chosen in the range 0.33-0.6. MMHEMTS can be used in a variety of microwave applications. Indeed for lower Indium composition close to 0.3, the MM-HEMT is an attractive device for power applications, because of the larger bandgap (Figure 8) which gives a high HEMT breakdown voltage, larger 2DEG density (higher AEc for x ~ 0.3). Both characteristics will lead to large voltage operation required for power performance. Moreover the high bandgap of InAlAs for low Indium composition should lead to a high Schottky barrier height. MM-HEMTS with higher Indium content (x = 0.6) can be employed in lownoise applications. This value of Indium content will provide better electron transport properties in the InGaAs channel, due to their lower effective mass.

34

Millimeter waves in communication systems

For intermediate values x ~ 0.4, a recent published work by H. Happy et al [21] predicts high microwave and low-noise performance of MM-HEMTS. This characteristic can be seen in Figure 11. Calculated sheet carrier density of the 2DEG using self-consistent resolution of Poisson's and Schrodinger's equations has been plotted against the Indium content. The low field mobility of the IDEG obtained by the Monte Carlo method is reported on the same graph. Two opposite effects can be observed when the Indium composition varies. On one hand, the sheet carrier density in the InGaAs channel decreases when the Indium content increases, due to a reduction of the conduction band discontinuity. On the other hand, the increase of the Indium content gives better transport properties in the InGaAs channel, especially due to a lower electron mass. These opposing trends in the MM-HEMT characteristics lead to an optimum Indium composition of about 0.4.

IV. Present status of HEMTs VI.1. Microwave performance Cut-off frequency of extrinsic current gain f T is largely dependent on gate length of the devices. In Figures 12 and 13, this behaviour is clearly demonstrated. State-of-the-art is reported on these graphs. The 30 nm gate length LM-HEMT on

Figure 11. Calculated low field mobility and sheet carrier density of the 2DEG of InAlAs/InGaAs heterostructure used in metamorphic HEMT. Values are plotted versus the Indium content.

HEMT's capability for millimeter wave applications

35

Figure 12. State-of-the-art of cut-off frequencies fT of extrinsic current gain H21 versus gate length O/HEMTS realised on InP substrate.

Figure 13. State-of-the-art of cut-off frequencies fT of extrinsic current gain H21 versus gate length O/HEMTS realised on GaAs substrate.

36

Millimeter waves in communication systems

InP substrate (Figure 12) reported in [3] exhibits a f T of 350 GHz, which is the highest value yet reported for a three terminal device. This gate length is obtained using modern electron beam lithography system. On the same graph, state-of-the-art of PM-HEMTS on InP is also given versus gate length. Then value of 340 GHz [22] is obtained with a 50nm-gate length device. This similar result is related to the high Indium content of 0.8 in the InGaAs channel of the PMHEMT on InP device. In Figure 13, f T is also plotted versus gate length for devices realised on GaAs substrate. AlGaAs/InGaAs pseudomorphic HEMTS on GaAs (PM-HEMTS GaAs) is also reported on same graph, because of its mature technology. f T of state-of-the-art PM-HEMTS is 150 GHz for a gate length L = 0.1 urn [23]. For same gate length, in 1999 f T of 200 GHz has been reached for MM-HEMT [24-25]. This is the best value ever reported on GaAs substrate with HEMTS. 0.1 [im MMHEMTS f T are also comparable with devices realised on InP substrate with same gate length. This similar result demonstrates the attractive potential of MM-HEMTS in millimeter wave applications. Moreover GaAs substrate is more suitable for low cost MMIC production than InP substrate. Indeed, a GaAs substrate is larger and mechanically more robust than an InP substrate. IV.2. Low-noise performance The HEMT, with its high speed and low parasitics, is an ideal candidate for lownoise amplification at millimeter wave frequencies. Low noise HEMTS are being used in, and developed for a wide range of systems such as satellite communication, radio astronomy and electronic equipment. In Figures 14 and 15, state-of-the-art of minimum noise figure NF^ with associated gain G^ are reported for different HEMT'S structures. Considerations for optimisation of low-noise HEMTS are relatively straightforward. High cut-off frequencies (fT, f^) and low parasitic elements (gd, C d, R , Rs...) are essential for achieving a low noise figure and high associated gain. As shown in Figure 14, which report measurements at 94 GHz, a strong dependence exists between the minimum noise figure and the gate length. Lowering gate length improves the noise figure of HEMTS. Another way to improve noise performance is to use material with good electron transport properties in the channel. This is demonstrated by curves in Figure 14. The use of higher Indium content in the lattice-matched HEMT (x ~ 0.53) involves substantial noise figure and associated gain compared with the pseudomorphic HEMT on GaAs, which presents lower Indium content in the InGaAs channel (x ~ 0.2). Moreover at 94GHz,a noise figure of 2.1dB and associated gain of 6.3 dB are obtained with gate length L = 0.1 jam for the PM-HEMT on GaAs [26]. For the LM-HEMT [27], a gate length of only 0.25 \Lm is required to reach close values. In Figure 15, frequency dependence of the best minimum noise figure on HEMTS is presented. At 60 GHz and 94 GHz (Figure 15), best results of NF^ are respectively 0.8 dB [28] and 1.2dB [6]. These results are obtained with 0.

HEMT's capability for millimeter wave applications

37

Figure 14. State-of-the-art of minimum Noise Figure NFmin versus gate lengths of LM-HEMTS on InP substrate and PM-HEMT on GaAs substrate. Curve on the inset represent associated gain Gass versus minimum noise figure for same devices. Measurements have been achieved at 94 GHz.

Figure 15. State-of-the-art of minimum Noise Figure NFmin versus frequency of different HEMTS.

38

Millimeter waves in communication systems

gate length LM-HEMTS. MM-HEMTS exhibits higher noise figure than devices realised on InP substrate. This is related to the lower Indium content of 0.4 in structure reported in [21]. PM-HEMT on GaAs presents noise figure of 2.4 dB at 94 GHz,, which is 1.2 dB higher than value on LM-HEMT. This means that LMHEMTS on InP is more accurate today for low-noise applications than PM-HEMTS on GaAs. However, primary results obtained with MM-HEMTS on GaAs substrate have demonstrated its potential for low-noise applications.

IV.3. Power performance The HEMT, with its high current density and cut-off frequency, is an ideal candidate for millimeter wave power amplifiers. The AlGaAs/InGaAs pseudomorphic HEMT has been largely studied for power applications and has demonstrated better maximum output power Pout in V- and W-bands over other types of transistors. As can be seen in Figure 16, an output power density of 1 W/mm has been obtained at 60GHz [7]. This value decreases to 392mW/mm at 94GHz for the same device [29]. Status of state-of-the-art of LM-HEMTS on InP is also plotted on Figure 16. At 60 GHz maximum output power density is 530mW/mm [30]. This value is half than that obtained with AlGaAs/InGaAs PM-HEMT on GaAs. The reason is a combination of lower breakdown voltage in the In0 53 GaQ 47As channel and the weak Schottky barrier height of In052Al048As compared with AlGaAs/InGaAs heterostructure. However InAlAs/InGaAs LM-HEMTS present another advantages. This can be explained by plotting power-added efficiency PAE versus associated gain Ga (Figure 17). Indeed, at higher frequencies, in this case 60GHz, LM-HEMTS present better power added efficiency and associated gain than PM-HEMTS. PM-HEMT presents a Pout = IW/mm, PAE = 36% and Ga = 4.4 [7]. Values for the LM-HEMT are p out = 510W/mm, PAE = 39% and Ga = 7.1 [30]. This is due to the better cut-off frequencies of LM-HEMTS. This behaviour demonstrates that LM-HEMTS are more accurate in power amplifiers at 60 GHz and beyond, LM-HEMTS present a best combination of output power density, power-added efficiency and associated gain. ^x^i-x^^K^i-x^8 MM-HEMT'S output power is also given in Figure 16 for Indium content x = 0.33 and 0.43 [25], At 35 GHz, maximum output power density of 830mW/mm is obtained. Power-added efficiency and associated gain are respectively 46.5% and 7.5 dB. Device presents an Indium content of 0.43 in the channel and the Schottky layers. A best value of 910 mW/mm is obtained for

"our

V. Conclusion We have presented a review on the ultra-high speed HEMTS for low-noise and power applications at millimeter wave range frequencies. Lattice-matched and pseudomorphic HEMTS on InP substrate have demonstrated their superiority in cut-off frequency and low-noise performance. A cut-off frequency of extrinsic

HEMT's capability for millimeter wave applications

Figure 16. Maximum output power of state-of-the-art HEMTS versus frequency.

Figure 17. Power added efficiency versus associated gain ofiM-HEMTs on InP and PM-HEMTS on GaAs. Frequency is 60 GHz.

39

40

Millimeter waves in communication systems

current gain f T of 350 GHz has been obtained with a 30 nm gate length LM-HEMT, which is the highest value yet reported for a three terminal device. A best minimum noise figure of 1.2dB with associated gain of 7.2 dB at 94 GHz has been reached with 0.1 jjm gate length LM-HEMT. For power application, PM-HEMT on GaAs presents still best output power density. At 60 GHz, output power density reached is 1 W/mm. THe value obtained with HEMTS on InP is about half than PM-HEMTS on GaAs at 60 GHz. However structures based on InP substrate allow better combination between output power density, associated gain and power added efficiency than PM-HEMT on GaAs. An alternative of these devices is the metamorphic HEMT on GaAs based on the InAlAs/InGaAs heterostructure, in which the Indium content can be chosen in the 0.33-0.6 content range. That device presents several advantages. First, the use of GaAs substrate is is more suitable for low cost MMC production. In fact, a GaAs substrate is larger and mechanically more robust than an InP substrate. Moreover, metamorphic structure allows the use of weak Indium composition (0.3-0.4 range) in power application and higher value (0.4-0.6 range) for low-noise application. Today MM-HEMTS with Indium content of 0.4 present similar cut-off frequency and power density than state-of-the-art of HEMTS. That suggests that MM-HEMTS will become a successful and competitive technology in the market place, in power and low-noise applications at millimeter wave frequencies. REFERENCES [1] DINGLE (R.), STORMER (H.), GOSSARD (A.C.), WIEGRNANN (W.), "Electron mobilities in modulation doped semiconductor heterojunction superlattices," Appl Phys. Lett., 33, n° 7, pp. 665-667, Oct. 1, 1978. [2] CAMNITZ (L.H.), TASKER (P.J.), LEE (H.), VAN DER MERWE (D.), EASTMAN (L.F.), "Microwave characterization of very high transconductance MODFET" in IEDM Tech. Dig., pp. 360-363, Dec. 1984. [3]

SUEMITSU (T.), eta/., "30-nrn gate InAIAs/lnGaAs HEMT'S lattice-matched to InP substrates," IEDM Technical Digest, pp. 223-226, 1998. [4] LAI (R.), BARSKY (M.), HUANG (T.), SHOLLEY (M.), WANg (H.), KOK (Y.L.), STREIT (D.C.), BLOCK (T.), Liu (P.H.), GAIER (T.), SAMOSKA (L), "An InP HEMT MMIC LNA with 7.2dB Gain at 190GHz", IEEE Microwave and Guided Wave Letters, 8, n° 11, Nov. 1998. [5] SMITH (P.M.) eta/, "W-band high efficiency InP-based power HEMT with 600 GHz Fmax" IEEE Microwave Guided Wave Letter, 5, pp. 230-232, luly 1995. [6] DUH (K.H.G.), CHAO (P.C.), Liu (S.M.J.), Ho (P.), KAO (M.Y.), BALLINGALL (J.M.), "A Super Low-Noise 0.1 Lim T-Gate InAlAs-lnGaAs-lnP HEMT", IEEE Microwave and Guided Wave letters, 1, n8 5, May 1991, pp. 114-116. [7] SMITH (P.M.), BALLINGALL (J.M.), SWANSON (A.W.), "Microwave and mm-Wave Power Amplification Using Pseudomorphic HEMTS", Microwave journal, 1st May 1990, 33, n e 5, p. 71.

HEMT's capability for millimeter wave applications

41

[8] KIM (B.), MATYL (R.J.),. WURTELE (M.)( BRADSHAW (K.), KHATIBZADEH (M.A.), TSERING (H.Q.), "Millimeter-Wave Power Operation of an A]GaAs/lnGaAs/GaAs Quantum Well MISFET", IEEE Transactions on Electron Devices, 36, n° 10, October 1989, pp. 2936-2242. [9] SAUNIER (P.), TSERING (H.Q.), "AIGaAs/lnGaAs Heterostructures with Doped Channels for Discrete Devices and Monolithic Amplifiers", IEEE Transactions on Electron Devices, 36, n° 10, October 1989, pp. 95131-9935. [10] LIECHTI (C.A.), "Microwave field-effecttransistors-1976," IEEE Trans. Microwave Theory Tech., 24, 279-300 (1976). [11] DAS (M.B.), "Millimeter-wave performance of ultrasubmicrometer-gate fieldeffect- transistors: a comparison of MODFET, MESFET and PBT structures", IEEE Trans. Electron Devices, 34, n° 7, pp. 1429-1440 (1987). [12] FRIIS (H.T.) "Noise figures of radio receivers," Proc. IKE, 31, pp. 419-422, July 1944. [13] HUGHES (B.), "A Temperature Noise Model for Extrinsic FETS", IEEE Transactions on Microwave Theory and Techniques, 40, n° 9, November 1992, pp. 1821-1832. [14] MATHEWS (J.W.), BLAKESLEE (A.E.),"Defects in Epitaxial Multilayers Misfit Dislocation" J. Cryst. Growth, 27, pp. 118-125, 1974. [15] FlSCHER-COLBRIE (A.), MlLLER (J.N.), LADERMAN (S.J.), ROSNER (S.J.), HULL (R.), "Growth

and characterization of AIGaAs/lnGaAs/GaAs pseudomorphic structures," J. Vac. Sci. Tech., B6, pp. 620-624. 1988. [16] NGUYEN (L.D.), LARSON (I.E.), MISHRA (U.K.), "Ultra-high-Speed Modulation-Doped Field-Effect Transistors: A Tutorial review". Proceeding of the IEEE, 80, n° 4, April 1992. [17] CHERTOUK (M.), HEISS (H.), Xu (D.), KRAUSS (S.), KLEIN (W.), BOHM (G.), TRANKLE (G.), WEIMANN (G.), "Metamorphic InAIAs/lnGaAs lattice HEMT'S on GaAs substrate with a novel composite channels design", IEEE Elect. Dev. Lett., 17, pp. 273-275, 1996. [18] ZAKNOUNE (M.), BONTE (B.), GAQUIERE (C), CORDIER (Y), DRUELLE (Y.), THERON (D.), CROSNIER (Y.), "InAIAs/lnGaAs Metamorphic HEMT with high current density and high breakdown voltage", IEEE Elect. Dev. Lett, 19, n° 9, pp. 345-347, 1998. [19] WIN (P.), DRUELLE (Y), LEGRY (P.), LEPILLET (S.), GAPPY (A.), CORDIER (Y), FABRE (J.), "Microwave performance of 0.4 pm gate metamorphic In029AI0 71 As/ln0 3Ga0 7As HEMT on GaAs substrate" Electron Lett., 29, pp. 169-173, 1993. [20] ADACHI (S.), Physical properties of IH-V semiconductor compounds: InP, In As, GaAs, GaP, InGaAs, InGaAsP, John Wiley & Sons, 1992, ISBN # 0-471-57329-9. [21] HAPPY (H.), BOLLAERT (S.), FOURR£ (H.), CAPPY (A.), "Numerical analysis of device performance of metamorphic lnyAI1_yAs/lnxGa1_xAs HEMTS on GaAs substrate", IEEE Transactions on Electron Devices, ED-45, n° 10, Oct. 1998, pp. 2089-2095. [22] NGUYEN (L.D.), BROWN (A.S.), THOMPSON (M.A.), JELLOIAN (L.M.), "50 nm Self-AlignedGate Pseudomorphic AllnAs/GalnAs High Electron Mobility Transistors", IEEE Transactions on Electron Devices, 39, n° 9, September 1992. [23] NGUYEN (L.D.), TASKER (P.J.). RADULESCU (D.C.), EASTMAN (L.F.), "Characterization of ultra-high-speed pseudomorphic AIGaAs/ InGaAs (on GaAs) MODFET'S," IEEE Trans. Electron Devices, 36, pp. 2243-2248, Oct. 1989.

42

[24]

Millimeter waves in communication systems

BOLLAERT (S.), CORDIER (Y),

HOEL (V.), ZAKNOUNE (M.),

HAPPY (H.),

LEPILLIET (S.), GAPPY

(A.), "Metamorphic ln04AI06As/ln04Ga06As HEMTS on GaAs substrate", IEEE Electron Device Letters, 20, n° 3, mars 1999. [25] WHELAN (C.S.), MARSH (P.P.), HOKE (WE.), MCTAGGART (R.A.), MCCARROL (C.P.), KAZIOR (K.E.), "GaAs metamorphic HEMT (MMHEMT): an attractive alternative to InP HEMTS for hihg performnace low noise and power application". Proceeding of IPKM, May 2000, pp. 337-340. [26] TAN (K.L), DIA (R.M.), STREIT (D.C.), LIN (T.), TRINH (T.Q.), HAN (A.C.), Liu (P.H.), CHOW (P.O.), YEN (C), "94GHz 0.1um T-Gate Low-Noise Pseudomorphic InGaAs HEMTS", IEEE Electron device Letters, 11, n° 12, Dec. 1990, pp. 585-587. [27] DUHI (K.H.G.), CHAO (PC.), Ho (P.), KAO (M.Y), SMITH (P.M.), BALLINGALL (J.M.), JABRA (A.A.), "High-Performance InP-based HEMT Millimeter-wave Low-Noise Amplifier", IEEE MTT-S International Microwave Symposium Digest, 1989, pp. SOSSOS. [28] DUH (H.G.), CHAO (P.C), Liu (S.M.J.), Ho (P.), KAO (M.Y), BALLINGALL (J.M.), "A Super Low-Noise 0.1 urn T-Gate InAlAs-lnGaAs-lnP HEMT", IEEE Microwave and Guided Wave letters, 1, n° 5, May 1991, pp. 114-116. [29] STREIT (D.C.), TAN (K.L), DIA (R.M.), Liu (J.K.), HAN (A.C.), VELEBIR (J.R.), WANG (S.K.), TRINH (T.Q.), CHOW (P.M.D.), Liu (P.H.), YEN (H.C.), "High-Gain W-Band Pseudomorphic InGaAs Power HEMTS" IEEE Electron Device letters, 12, n° 4, April 1991, pp. 149-150. [30] KONG (W.M.T.), WANG (S.W.), CHAO (PC.), Tu (D.W.), HWANG (K.), TANG (O.S.A.), Liu (S.M.), Hu (P.), NICHOLS (K.), HEATON (J.), "Very High Efficiency V-Band Power InP HEMT MMICS", To be published in Electron Device Letters.

Chapter 3

Progress in millimeter-wave fiberradio access networks A. Nirmalathas, C. Lim and D. Novak Australian Photonics Cooperative Research Centre, Department of Electrical and Electronic Engineering, University of Melbourne, Australia

R. B. Waterhouse Australian Photonics Cooperative Research Centre, Department of Communication and Electronic Engineering, Royal Melbourne Institute of Technology, Australia

I. Introduction The design of broadband radio access networks to deliver services such as videoon-demand, interactive multimedia, high speed internet and HDTV has been the subject of intense interest and investigation in recent years. Throughout the last decade the millimeter-wave (mm-wave) frequency band (26-70 GHz) has been considered for commercial radio access communications, primarily to avoid spectral congestion at lower microwave frequencies, with however the significant advantage of offering large transmission bandwidths. Operation at mm-wave frequencies also produces smaller radio cells due to the large atmospheric absorption which enables a large frequency reuse factor for efficient spectrum utilization. Future millimeter-wave broadband radio access systems may employ an architecture in which signals generated at a central location or central office (co) will be transported to remote antenna base-stations (BSS) for wireless distribution to a number of customer units (cus) [1, 2]. The use of optical fiber feed networks in these systems to distribute the radio signals to the base-stations is an attractive approach since the optical fibers provide efficient, low cost, low loss and EMi-free signal transportation. In addition, an architecture such as that shown in Figure 1 enables a large number of base-stations to share the transmitting and processing equipment located remotely from the customer serving area. For full-duplex operation of the radio access system, mm-wave wireless signals from the cus are received at the antenna BS and undergo electrical to optical conversion before they are transported over the fiber links back to the co.

44

Millimeter waves in communication systems

Figure 1. Generic architecture of a mm-\vave fiber-radio network.

One of the key challenges in implementing mm-wave fiber-radio access systems is developing suitable optical network architectures for efficient distribution of the radio signals while maintaining a simple and compact BS configuration. At the antenna BS, the integration of an optical source and photodetector with electronic components such as mixers, amplifiers, and diplexers as well as printed antennas as the radiating elements, will enable the design and development of simple and lightweight BSS for easy installation on building walls and corners, street lights and telephone poles. The choice of the optical network architecture will also determine the hardware required at the co. This paper presents an overview of the progress made in the last decade towards the implementation of mm-wave access systems employing optical fiber distribution networks. The paper first describes the enabling technologies for the implementation of these systems both from the perspective of the signal transport scheme, as radio frequency (RF) over fiber, intermediate frequency (IF) over fiber or baseband-over-fiber and the necessary mm-wave devices and components including optical sources and modulators, photodetectors, and optoelectronic/electronic device integration. We then outline a summary of the system implementations of mm-wave fiber-radio access networks that have been reported to date, including mm-wave radio systems operating at frequencies starting at 28 GHz and extending beyond 60 GHz.

II. Enabling technologies - signal transport schemes H.l. RF-over-fiber The simplest way of interconnecting remote antenna BSS in a mm-wave fiberradio system is via an optical fiber feed network which can transport the mm-wave RF signals directly over fiber without any need for frequency up- or

Progress in millimeter-wave fiber-radio access networks

45

down-conversion at the BSS, as shown in Figure 2(a). The primary challenge for implementing such fiber-radio architectures with RF-over-fiber lies in the search for both suitable high speed optical modulation techniques that have the ability to generate mm-wave modulated optical signals and also high speed photodetection techniques that directly convert the modulated optical signals back into mm-wave signals in the electrical domain. This scheme of "RF-over-fiber" transport has the direct advantage of realizing simple base-station architectures. However, its major drawback is the significant effect of fiber chromatic dispersion (at a wavelength of 1.55 jim) on the detected RF signals [3, 4]. When optical carriers are directly modulated with mm-wave signals in a conventional way, the resulting optical

Figure 2. Optical signal transport schemes for mm-wave fiber-radio systems: (a) RF-over-fiber, (b) iF-over-fiber, and (c) baseband-over-fiber.

46

Millimeter waves in communication systems

spectrum (to a first order approximation) consists of the optical carrier and a modulation side-band on each side of the optical carrier spaced at a frequency from the carrier equal to the center frequency of the modulating nun-wave signal. As the mm-wave modulated optical signal propagates through standard optical fiber, each modulation side-band will experience different amounts of phase-shift with respect to the optical carrier as a result of fiber chromatic dispersion. When the optical signal is detected at the receiver by a square-law photodetector, each side-band will beat with the optical carrier, producing two nun-wave signal components with equal amplitudes and different phase values. The vectorial addition of these two signals will yield the resultant mm-wave signal and the RF power of the resulting signal will vary as a function of phase difference that depends on both the fiber length and the dispersion parameter. These resulting RF power variations will lead to degradations in the carrier-to-noise ratio (CNR) of the link [4]. Considerable research effort has been devoted towards developing RFover-fiber transport schemes which are either dispersion-tolerant or based on dispersion compensation techniques such as chirped fiber Bragg grating filters [5], self-phase modulation in fiber [6] or mid-span phase-conjugation [7]. Much research has been carried out on investigating techniques for the generation of mm-wave modulated optical sources all of which may have real potential for use in mm-wave fiber-radio systems. However, in this paper we focus on the techniques with demonstrated ability to support data modulation and the dispersion-tolerant methods techniques that attempt to remove the effect of fiber chromatic dispersion on the link CNR reductions. These techniques include resonantly enhanced direct modulation of semiconductor lasers, external modulation using optical modulators, and optical heterodyne techniques. Direct modulation of a single-mode distributed feedback (DFB) laser, although simple to implement, is currently limited to less than 30 GHz [8] and commercially available DFB lasers have typical modulation bandwidths of 10 GHz or less. However, by resonantly enhancing the modulation bandwidth of a laser cavity using external cavity lasers [9,10] or monolithic multi-section lasers [11, 12], narrow transmission windows at mm-wave frequencies can be achieved. As the output optical spectrum from such a multi-section laser comprises many optical modes, the effect of chromatic dispersion is an issue and the authors and others have shown that by optimizing the drive conditions of the laser, the effect of fiber dispersion can be minimized [11-13]. External modulation is the simplest technique for achieving a mm-wave modulated optical signal as it involves the modulation of a continuous-wave laser via external optical modulators driven by electrical mm-wave signals. Progress in the development of high-speed optical modulators is discussed in Section III.l. However, as described earlier, the external modulation technique also suffers from the effect of fiber chromatic dispersion. The optical single side-band with carrier (OSSB+C) modulation scheme can overcome the dispersion problem by generating an optical signal with only one modulation side-band. Figure 3 shows

Figure 3. Schematic of optical single side-band with carrier modulation technique using a dual-electrode MZM in conjunction with an electrical hybrid coupler. (Inset shows a measured optical spectrum of an optical carrier with an upper side-band spaced 37 GHz from the carrier and > 20 dB suppression of the lower side-band).

48

Millimeter waves in communication systems

how OSSB+C modulation can be generated directly by using a single dual-drive Mach-Zehnder modulator with each electrode driven by RF signals with equal amplitudes but a phase difference of 90° [14]. Alternative integrated modulator structures capable of generating OSSB+C modulation have been demonstrated [15, 16]. In addition, techniques for generating OSSB+C modulation via optical filters [17] and nonlinear amplification techniques [18] have also been reported. Optical heterodyne techniques are based on the generation of mm-wave signals by beating two optical modes separated at the required frequency that are also locked in phase. A variety of such techniques has also been reported, based on the manner in which the two phase-locked optical carriers are generated and also how the data modulation is applied to one or both of the optical carriers. Generally, the dual mode sources are obtained by injection locked dual-mode sources [19] or mode-locked laser diodes [20-24]. With optical heterodyne techniques, since the output optical spectrum has essentially two modes, the mode beating at the photodetector produces a single RF component resulting in dispersion-tolerant signal transport [3]. Demonstrations of the techniques which can be used to realize RF-over-fiber signal transport schemes for mm-wave fiber-radio systems are summarized in Table I in terms of the particular technique, RF carrier frequency used, and the relevant data modulation scheme and data rate. H2. IF-over-fiber In contrast to the transmission of mm-wave radio frequencies over optical fiber, the effects of fiber chromatic dispersion on the optical distribution of IF signals are reduced significantly. For comparison, a signal at GHz will experience a signal-to-noise reduction of less than 0.1 dB when recovered after transmission through 7 km of standard optical fiber while a signal at 38 GHz will suffer a CNR penalty of over 10 dB after propagation over the same fiber transmission distance. In addition, IF signal transport schemes have the advantage of only requiring optoelectronic devices with reduced bandwidths. IF-over-fiber signal transport for upstream signal transmission has been reported using the direct modulation of a low cost optical source such as a light emitting diode with a bandwidth of less than 2 GHz [40]. Another technique employed direct modulation of an electroabsorption modulator [41]. Downstream IF-over-fiber signal transport has used techniques such as direct modulation [42] and external modulation of semiconductor lasers [43]. The complexity of the BS hardware however increases with IF signal transport schemes for mm-wave radio access systems, as a mm-wave local oscillator (LO) and mixers for the frequency conversion processes are now required (see Figure 2(b)). This may also be a limitation when considering the ability to upgrade or reconfigure the radio network with the provision of additional mm-wave radio channels or changes in RF frequency. The subsequent requirement for a mm-wave LO at the antenna BS in mmwave fiber-radio access networks employing IF-over-fiber signal transport

Table L Summary of reported data transmission experiments that support "RI--over-fiber" signal transport at mm-wave frequencies. MM- Wave Signal Generation Technique

MM- Wave Frequency

Data Modulation

Reference

Resonantly Enhanced External Cavity Lasers

35 GHz

40 Mb/s BPSK

[9,25]

Resonantly Enhanced Monolithic Semiconductor Lasers

45 GHz

50 Mb/s DPSK

[10]

Resonantly Enhanced Monolithic Semiconductor Lasers

41 GHz

2 x 2.5 Mb/s BPSK

[26]

Frequency Doubling

60 GHz

-

[27]

Frequency Doubling

35 GHz

28 Mb/s 16QAM

[28]

External Modulation

38 GHz

5 Mb/s BPSK 16 QAM, 64 QAM

[29]

External Modulation

38 GHz

2.048 Mb/s

[30]

External Modulation

39 GHz

MPEG2

[31]

External Modulation

44 GHz

118 Mb/s QPSK

[32]

Mode-Locked Lasers

37 GHz

500 Mb/s ASK

[21, 22]

Mode-Locked Lasers

37 GHz

255 Mb/s BPSK 3 x analog video

[23]

Mode-Locked Lasers

63 GHz

155 Mb/s BPSK

[24]

Optical Single Side-band Modulation

36-39 GHz

9 x 155 Mb/s BPSK

[33]

Optical Single Side-band Modulation

60 GHz

2x155 Mb/s DPSK

[34]

Optical Single Side-band Modulation

38 GHz

200 Mb/s OOK 625 Mbaud/s 16QAM 3x6.11Mbaud/s64QAM

[35]

Optical Heterodyne

39 GHz

300 Mb/s BPSK

[36]

Optical Heterodyne

63 GHz

140 Mb/s CPFSK

[37]

Optical Heterodyne

60 GHz

120 Mb/s QPSK

[38]

Optical Heterodyne

39 GHz

300 Mb/s BPSK

[39]

50

Millimeter waves in communication systems

schemes, can be overcome by remotely delivering the LO signal from the central office [43, 44]. This also enables centralized control of the LO signals themselves. In [44], an LO signal at approximately 18 GHz was delivered via OSSB+ c modulation for a 38 GHz radio link, while [43] used a lower IF of 4.8 GHz which was frequency multiplied by a factor of 12 at the antenna BS in a 60 GHz radio system. The previous paragraphs described the more conventional technique of frequency conversion when using iF-over-fiber transport schemes, namely detection of the IF signal and then electronic frequency upconversion using an electronic mm-wave mixer. Frequency up conversion however can also be accomplished in the optical domain, by using the nonlinearity of an optoelectronic component, such as an external electro-optic modulator. In this technique an optical source is modulated by an IF signal and harmonic upconversion of the optical signal is realized via further modulation with an external modulator driven by an LO [45-48]. Depending on where the modulator is biased on its transfer function, different frequency harmonics such as f RF = (2k + l)fLO + fIF or f^ = (2fc)fLO + fIF, will be generated where f^,, fLO, and fjp are the mm-wave, LO, and IF signal frequencies, respectively, and k = 0,1, 2,... It has been shown that the effect of fiber chromatic dispersion on the detected RF frequency in such optoelectronic frequency upconversion techniques varies greatly with the bias point of the modulator transfer function [47]. II.3. Baseband-over-fiber So far we have considered two possible methods to transport the informationcarrying radio signals over fiber in mm-wave fiber-wireless systems. As shown in Figure 2(c) the third technique that can be used is to transport the signal from the co to the BS as a baseband signal over fiber, and then upconvert the information to the required mm-wave radio frequency at the antenna BS. Upstream signal transport can also be accomplished by downconverting the received mm-wave frequency at the BS to baseband before transmission to the co. As with iF-overfiber signal transport, the effects of fiber chromatic dispersion are also greatly reduced by distributing the radio signals as baseband-over-fiber [49]. In addition, architectures incorporating baseband-over-fiber enable the utilization of mature electronic circuitry for signal processing at the remote base-station. However, as with iF-over-fiber signal transport schemes, the complexity of the BS increases accordingly, with the requirement of a mm-wave LO and mixer for the frequency conversion processes. With recent advancements in mmic (monolithic microwave integrated circuit) technology, low phase-noise los extending into the mm-wave region have been demonstrated [50, 51]. Such technology has the advantage of enabling optoelectronic and electronic device integration to generate a compact transceiver module at the remote BS. Since the additional LO source and signal processing hardware (frequency conversion, and multiplexing

Progress in millimeter-wave fiber-radio access networks

51

and de-multiplexing of signals from many users) in the base-station may limit the upgradability of the overall system, remote delivery of the LO signal from the central office can also be implemented [52].

III. Enabling technologies - mm-wave optoelectronic devices ULl. High-speed optical modulators Of the many schemes to create an mm-wave modulated optical signal, the simplest and most mature method is to modulate the optical carrier by an external modulator with the mm-wave signal applied to its electrodes. Over the years there has been a variety of external modulators developed capable of operating in the mm-wave frequencies, each with their relative merits and disadvantages. An excellent review article on wide-bandwidth modulators can be found in [53]. For their application in mm-wave fiber-radio systems, it is imperative that these devices have characteristics of: broad bandwidth, low mm-wave drive voltages, good linearity, bias stability, high optical power-handling ability and low optical insertion loss. The subsequent subsections discuss the various alternatives for mm-wave external modulation of an optical signal keeping in mind this set of criteria. Generally, modulators can be classified as either electroabsorption (EA) or electro-optic (EO) devices. III.1.1. Electroabsorption modulators

Electroabsorption modulators (EAMS) are made from ni-V compound semiconductors and utilize quantum well structures to give the quantum confined Stark effect. EAMS based on this principle have displayed bandwidths in excess of 40 GHz [54, 55]. Due to free carrier absorption and band-to-band absorption, the optical propagation loss of an EAM is large, typically 15-20 dB/mm, restricting the size of the device to less than several hundred microns. This is why these devices are considered as lumped elements and their speed of operation is determined by the RC time constant of the circuit, so it is imperative that EA modulators exhibit small device capacitances. Bandwidths in excess of 60 GHz can be achieved by lowering the terminating resistance. However, this is at the expense of modulation efficiency [56]. EAMS typically suffer from relatively poor optical power-handling capabilities, due to carrier pile up; however, techniques have been proposed incorporating tensile strain in the quantum wells that can alleviate this problem to some extent [57]. The characteristics of EAMS are also very sensitive to wavelength and temperature changes and therefore strict bias control is necessary during operation. This dependency can be detrimental when considering fiber-radio systems utilizing WDM techniques for routing signals.

52

Millimeter waves in communication systems

in.1.2. Electro-optic modulators Of the modulators that can be used in mm-wave fiber-radio links, electro-optic (EO) modulators are the most established. These modulators can be divided into categories based on the material used to give the required electro-optic effect: LiNbO3, semiconductors or polymers. To give the broad bandwidth necessary for mm-wave fiber-radio applications, a travelling-wave electrode structure is typically used, where the electrode is designed as a transmission line and therefore its capacitance does not limit the speed of the modulator. Figure 4 shows a schematic diagram of a travelling-wave electrode. Generally Mach-Zehnder (MZ) interferometric structures (as shown in Figure 4) are utilized for these modulators and such devices are known as MZ modulators (MZMS) where the modulating signal on the electrode travels in the same direction as the optical carrier. If both signals travel at the same velocity, the phase change induced by the RF signal is integrated along the length of the electrode. Thus the drive voltages for such devices can be significantly reduced without reducing the bandwidth. Perhaps the most common travelling-wave EO modulator is made from LiNbO3 material, due to its combination of high electro-optic coefficients and high optical transparency in the near infrared wavelengths. Two types of lithium used for modulators are: Z-cut and X-cut, depending on whether the Z-axis of the material is perpendicular to or in the plane of the wafer. In general, devices based on Z-cut LiNbO3 are more efficient than those based on X-cut due to the geometry of the electrode structure [58]. X-cut devices are favored for commercial applications as they are less prone to thermal instability due to the pyroelectric nature of LiNbO3. There have been several reported LiNbO3 MZM devices with bandwidths in excess of 40 GHz with relatively low drive voltages [59, 60]. To push the

Figure 4. Schematic of an electro-optic modulator with electrode and optical waveguide structures.

Progress in millimeter-wave fiber-radio access networks

53

frequency beyond this frequency requires special design of the electrode structure as dielectric losses, radiation losses and leaky wave modes associated within the mm-wave transmission line become important. A technique that has significantly enhanced the performance of LiNbO3 modulators is to incorporate a ridge waveguide [53]. Ridge waveguides further improve the focusing of the electric field under the hot electrode and have been used to achieve drive voltages of 3.5 V and 5.1 V for Z-cut modulators with 3-dB electrical bandwidths of 30 and 70 GHz, respectively [61]. There are several reported examples of broadband EOMS utilizing semiconductor materials [62, 63]. Although these devices have bandwidths in excess of 30 GHz, their drive voltages are somewhat high (> 10 V at 40 GHz) due to the relatively low bulk electro-optic coefficient of semiconductor materials as well as a poor overlap of the applied electric field and the optical mode [53]. Organic polymers have several attractive features for integrated optical applications and can be made electro-optic using high temperature poling methods. Several broadband EO polymer modulators have been developed [64. 65] with bandwidths and drive voltages similar to their semiconductor counterparts, however, there are some issues associated with this technology, namely the power handling capacity and the long-term bias stability. Table II presents a summary of the mm-wave external modulators considered in this subsection, including their 3-dB bandwidths and drive voltages. HL2 High-speed optical detectors Section n described some enabling technologies for the implementation of mmwave fiber-radio systems with regards to the technique for transporting the radio signal over fiber. In addition to mm-wave optical modulators, the optical transport of mm-wave frequencies requires state-of-the-art photodetector (PD) technologies

Table II. Summary of reported mm-wave external modulators and their characteristics.

Type

Bandwidth

Drive Voltage

Comments

Reference

EAM

60 GHz

4V

Reactive impedance matched at 60 GHz

[56]

EOM LiNbO3

40 GHz

7V

Thick film, packaged device

[60]

EOM LiNbO3

75 GHz

5.1V

Ridged waveguide

[61]

EOM GaAs/AlGaAs

50 GHz

13V

Doped epitaxial layers

[62] layers

EOM GaAs/AlGaAs

50 GHz

14V

Undoped epitaxial layers

[63] layers

EOM Polymer

110 GHz

10 V

Fin-line electrodes

[65]

54

Millimeter waves in communication systems

with high efficiencies and bandwidths in excess of 60 GHz. The bandwidth of any PD is limited by both internal transit-time considerations (determined by the electrical field) and extrinsic factors such as electrical parasitics due to RC time constants [66]. High-speed photodetectors operating at longer wavelengths (in the 1.3 - 1.55 Jim range) have been reported based on surface-illuminated (also known as vertically-illuminated) techniques with 3-dB bandwidths greater than 110 GHz [67] being achieved. Surface-illuminated PDS however suffer from an inherent tradeoff between bandwidth and device efficiency, since a larger surface area is needed to increase the absorption of photons which also leads to increased carrier transit times within the device [68]. Edge-illuminated waveguide photodetectors are one device structure that has been considered as a means for enabling both large bandwidth and efficiency [69, 70]. By also designing the waveguide PD to be a 'travelling-wave' type where photon absorption occurs in a distributed manner along the length of the device, the parasitic bandwidth limitations can be reduced due to the electrically distributed structure of the travelling-wave PD [71]. A similar structure to the travelling-wave photodetector is the velocity-matched distributed photodetector, which consists of an array of PDS serially connected by a passive optical waveguide [72]. Such a device can exhibit large bandwidths (49 GHz in [72]) and efficiency, and is also capable of achieving higher optical saturation powers although there is a tradeoff between the saturation current and bandwidth [72]. mm-wave PDS with high saturation powers and large electrical output powers are very useful in the development of antenna base-stations with fewer electronic amplifiers. The PD saturation power is limited by the nonlinearity of its response and it has been shown that this nonlinearity is caused by a decrease in the electric field and reduction in the carrier velocity due to a space-charge effect [73]. A PD structure that can overcome the drawback of a low saturated carrier velocity is the uni-travelUng-carrier (urc) photodetector in which only electrons travelling at a velocity much higher than the saturation velocity, contribute to the space-charge effect [74]. A UTC-PD with a bandwidth of 94 GHz and peak photocurrent of 184 mA (measured via the pulse response) has been reported [74]. EQ.3. Optoelectronic/electronic device integration for compact mm-wave BSs The development of compact, low cost radio BS hardware is essential for the successful deployment of mm-wave fiber-radio systems. While several examples of demonstrations of mm-wave fiber-radio systems (refer to Section IV) have been reported, these generally use a hybrid approach when integrating photonic devices, mm-wave passive and active devices, and the antenna at the BS. Two alternatives to this hybrid scheme are a hybrid integrated circuit (me) approach, where different materials are used for the separate processes and these wafers are integrated via bonding, and an optoelectronic integrated circuit (OEIC) approach that enables all the optical and electrical functions to be performed on a single chip. The OEIC approach ensures device uniformity, lower production costs and reduced parasitic effects for the base-station hardware. There are several

Progress in millimeter-wave fiber-radio access networks

55

reported implementations of monolithically integrated OEIC photoreceivers suitable for mm-wave fiber-radio systems. Recently A P-I-N photodiode operating at 1.55 |im was monolithically integrated on GaAs with 0.15 Jim gate-length pseudomorphic HEMTS using a linear graded metamorphic InGaAlAs buffer, with a measured optical/electrical (O/E) conversion factor of 49.6 V/W at 42 GHz [75]. Fabricating a multi-mode waveguide-photodiode on top of InP-based HEMT layers, a 52 GHz bandwidth photoreceiver with an O/E conversion factor of 105 V/W was developed [76]. Umbach et al [77-79] have developed several integrated mm-wave photoreceivers. Here separate layer stacks for the photodetector (p-i-n or metal-semiconductor-metal) and a HEMT-based amplifier are used to allow separate optimization of each device. Examples of HICS developed for mm-wave fiber-radio systems include a narrow-band 60 GHz fiber-radio transmitter developed in [80], consisting of a photodiode, MMIC preamplifier, and amplifier and a planar antenna. A packaged 60 GHz optical transceiver has also been recently reported in [81] utilizing an electroabsorption transceiver. The main quandary associated with an OEIC basestation module is that the materials used for growing photonic and mm-wave devices have properties unsuited to radiating structures, namely the printed antennas. Recently a simple approach was introduced that may overcome this perceived problem, by electromagnetically coupling low dielectric constant material to the mmic or OEIC wafers to achieve efficient radiators that can be integrated with the active devices [82].

IV. System demonstrations and architectures Earlier link demonstrations of mm-wave fiber-radio systems were mainly proof-of-concept transmission experiments which focused on the development of half-duplex links incorporating various optical transport techniques as described in Section II. A variety of half-duplex mm-wave half-duplex fiberradio experimental demonstrations for RF-over-fiber signal transport have been reported and were summarized in detail in Table III. Various mm-wave optical transport techniques for the transmission of analog and digital signals have been demonstrated with successful signal recovery. For example, Mathoorasing et al [83] have reported the distribution of multi-carrier compressed digital video signals (25 Mb/s 16 QAM at 38 GHz) while Ahmed et al [23] have demonstrated simultaneous transmission of 3 analog video channels and 255 Mb/s BPSK digital data at 37 GHz. While these demonstrations verify the feasibility of mm-wave hybrid fiber-radio systems, development of full-duplex fiber-radio links are critical to assess the commercial viability of such networks and this has been an active area of research over the past decade. In 1993, a full-duplex implementation known as 'modal' was proposed in the European RACE project [84]. Since that time a number of system demonstrations of full-duplex fiberradio systems operating at various mm-wave frequencies have been reported.

Table III. Summary of mm-wave fall-duplex fiber-radio system demonstrations. Mm-wave Radio frequency Band

Radio Downlink Characteristics

• 37 GHz • BPSK 622 Mb/s • 1 channel • 2 m link length

• 35-39 GHz • BPSK 155 Mb/s • 3 channels 26-40 GHz

• SCM

• 5 m link length

• 36-39 GHz • BPSK 155 Mb/s • 3 analog video channels • no radio link • 38 GHz • BPSK 155 Mb/s • 1 channel • no radio link

Radio Uplink Characteristics

Features of Fiber Feed Network

• 39 GHz •BPSK 155 Mb/s • 1 channel • 2 m link length

• Baseband-over-fiber (full duplex fiber with coarse WDM for up/down links)

• 37 GHz • BPSK

51.8 Mb/s • 1 channel • 5 m link length

•51. 8 Mb/s • no radio link

• 34 GHz • BPSK 155 Mb/s • 1 channel • no radio link

• RF-over-fiber • Half-duplex fiber

• RF-over-fiber downlink • IF-over-fiber uplink • Full duplex fiber (coarse WDM for up/down links) • RF-over-fiber • Half-duplex (ring configuration)

WDM

Special Features

• 2 channels down • 1 channel up

• Dispersion tolerant remote LO optical delivery • Novel modulation technique • OSSB + C modulation • 3 channels technique down • Mode-locked • 1 channel up laser upstream transmitter • No LO signal required at the BS • OSSB+C modulation downlink • Remote LO delivery •LED IF modulation uplink • 1 channel down • 1 channel up

• OSSB+C modulation down/up links • Wavelength re-use between up and downlinks

Ref.

[52, 90]

[33]

[44]

[95]

60 GHz

•60 GHz • QPSK 120 Mb/s (1 channel) • 25 Satellite TV channels • 5 m link length

• 59 GHz •QPSK 120 Mb/s • (1 channel) • no radio link

Half-duplex fiber links • RF-over-fiber downlink • IF-over-fiber uplink

• 60 GHz • 512 carrier QPSK/OFDM 50 Mb/s •BPSK 155 Mb/s • 5 m link length

• 57 GHz • 512 carrier QPSK/OFDM 50 Mb/s 5 mlink length

Half duplex fiber links • RF-over-fiber downlink • IF-over-fiber uplink

• 59 GHz • DPSK 155 Mb/s • 1 channel • no radio link

•60 GHz • DPSK 155 Mb/s • 1 channel • no radio link

• RF-over-fiber • Half-duplex (ring configuration)

• 63 GHz • DBPSK 155 • 1 channel • 1 m link length

• (19.2) GHz • DBPSK 155 Mb/s • 1 channel • no radio link

RF-over-fiber with remote upconversion in in downlink • IF-over-fiber in uplink • Half-duplex fiber

• Optical heterodyne • EAM dual function transceiver • No optical source requirement for base-station

• 3 channels down • 1 channel up

• Harmonic upconverison • Remote LO delivery for downconversion for uplink

• 1 channel down • 1 channel up

• EAM dual function transceiver • No optical source

• Harmonic upconversion and downconversion using mode-locked laser diode

[96]

[92]

[93, 971

[24]

58

Millimeter waves in communication systems

The characteristics of these full-duplex mm-wave fiber-radio systems are summarized in detail in Table HI. In addition, recent efforts by groups from the European ACTS FRANS Project [85, 86], Alcatel, Germany [80, 87, 88], and the Communications Research Laboratory, Japan [89] have considered the development of field trials to extend the laboratory demonstrations to real radio propagation environments. In order to extend mm-wave fiber-radio system demonstrations to fully deployable access networks, the generic architecture as shown in Figure 1 must be further developed. The design of such an architecture becomes critical when a network with many BSS are to be deployed. As each BS will require full-duplex connectivity with the co, the number of optical fiber pairs reaching the co scales with the number of base-stations. Recently the application of optical networking concepts to mm-wave fiber-radio networks has been proposed [33, 40, 89, 9092]. The use of wavelength division multiplexing (WDM) can provide a number of benefits by simplifying the network architecture using different wavelengths to feed different antenna BSS, supporting multiple interactive services on one fiber, and greatly simplifying network upgrades by enabling the introduction of new services and the deployment of additional BSS. Figure 5 shows a proposed star-tree architecture for a mm-wave fiber-radio network incorporating WDM [91]. In the system, the fiber links from the co form

Figure 5. MM-wave fiber-radio system star-tree architecture incorporating WDM.

Progress in millimeter-wave fiber-radio access networks

59

the star part of the architecture while the tree part is at the remote node (RN) with each branch feeding a different BS radio cell. The groups of BSS fed by one arm of the star have their own unique WDM wavelength for both the down and upstream directions. On this single arm of the star, the RF channels on any selected WDM wavelength are those transmitted or received within one radio cell while the channels required for adjacent cells are carried by other wavelengths. For the downstream case, the wavelengths are multiplexed together at the co and demultiplexed at the RN, while in the upstream direction the wavelengths are multiplexed at the RN and demultiplexed at the co. An alternative mm-wave WDM fiber-radio architecture is the ring network shown in Figure 6 [93]. The co distributes a number of wavelengths each carrying multiple modulated RF subcarriers. The ring topology allows the allocation of a single wavelength to a particular BS and wavelength routing is enabled via optical add-drop multiplexers (OADMS). Modulating upstream radio signals onto an optical carrier at the same BS wavelength and adding it back into the ring via the OADM, achieves upstream signal transmission. More advanced WDM fiber-radio architectures can be deployed by combining the basic architectures shown in Figures 5 and 6 to suit the physical network layout requirements of any practical implementations of the network. In any optical network incorporating WDM, optical crosstalk occurs due to imperfect optical components and leads to unwanted wavelengths interfering with the desired optical channel [94]. For dimensioning of large-scale WDM fiber-radio networks, accumulation of analog transmission impairments as a result of optical

Figure 6. MM-wave fiber-radio system ring architecture incorporating WDM.

60

Millimeter waves in communication systems

crosstalk will need to be analyzed and measures for controlling such impairments need to be further developed.

V. Conclusion Millimeter-wave fiber-radio systems have been an intense area of research and investigation over the past decade. In this paper we have reviewed the progress in the area of mm-wave fiber-radio systems in terms of their basic enabling technologies, from signal transportation over optical fiber to integration of optical and RF components at the antenna BSS, system demonstrations, and the development of network architectures incorporating WDM. In presenting this overview, we have focused our discussions on techniques and technologies that have been directly linked to implementations of mm-wave fiber-radio systems and networks. However, a significant amount of basic and applied research has gone into the development of other many high-speed lightwave and microwave core technologies which will have a significant impact on the future development of novel techniques, subsystems and system architectures for fiber-radio networks. All these developments will lead to the commercial reality of mm-wave fiber-radio networks. REFERENCES [1] OGAWA (H.)f POLIFKO (D.), BANBA (S.), Millimeter-wave fiber optics systems for personal radio communications, IEEE Trans. Micro, Thy & Tech. (1992), 40, pp. 2285-2293. [2] GRAY (D.), Optimal cell deployment for LMDS systems at 28 GHz, Proc, Wireless Broadband Conf. (1996), Washington DC, USA. [3] SCHMUCK (H.), Comparison of optical millimetre-wave system concepts with regard to chromatic dispersion, Electron. Lett. (1995), 31, n° 21, pp. 1848-1849. [4] GLIESE (U.), NORSKOV (S.), NIELSEN (T.N.), Chromatic dispersion in fiber-optic microwave and millimeter-wave links, IEEE Trans. Micro. Thy. & Tech. (1996), 44, n° 10, pp. 1716-1724. [5] KITAYAMA (K.), Fading-free transport of 60 GHz optical DSB signal in non-dispersion shifted fiber using chirped fiber grating. Proc. Int. Top. Meet. Micro. Photon. (1998), Princeton, NJ, USA, pp. 223-226. [6] RAMOS (F.), MARTI (J.), POLO (V.), FUSTER (J.M.), On the use of fiber induced selfphase modulation to reduce chromatic dispersion effects in microwave/millimetre-wave optical systems, IEEE Photon. Technol. Lett. (1998), 10, pp. 1473-1475. [7] SOTOBAYASHI (H.), KITAYAMA (K.), Cancellation of the signal fading for 60 GHz subcarrier multiplexed optical DSB signal transmission in non-dispersion shifted fiber using midway optical phase conjugation, IEEEJ. Lightwave Technol. (1999), 17. pp. 2488-2497.

Progress in millimeter-wave fiber-radio access networks

61

[8] BHATTACHARYA (P.), Quantum well and quantum dot lasers: From strained-layer and self-organized epitaxy to high-performance devices, Opt. & Quant. Electron, (2000), 32, n°3, pp. 211-225. [9] NAGARAJAN (R.), LEW (S.), BOWERS (J.E.), Millimeter wave narrowband optical fiber links using external cavity semiconductor lasers, IEEEJ. Lightwave Technol, (1994), 12, n° 1, pp. 127-136. [10] GEORGES (J.B.), KIANG (M.), HEPPELL (K.), SAVED (M.), LAU (K.Y.), Optical transmission of narrow-band millimeter-wave signals by resonant modulation of monolithic semiconductor lasers, IEEE Photon. Technol. Lett. (1994), 6, n° 4, pp. 568-570. [11] LIM (C), NOVAK (D.), SMITH (G.H.), Implementation of an upstream path in a millimeter-wave fiber wireless system, Proc. Opt. Fiber Commun. Conf. (1998), San Jose, CA, USA, pp. 16-17. [12] LIM (C.), NOVAK (D.), NIRMALATHAS (A.), SMITH (G.H.), Dispersion-induced power penalties in millimeter-wave signal transmission using multi-section DBR semiconductor lasers, to appear in IEEE Trans. Micro. Thy. & Tech. (2001). [13] OHNO (T.), SATO (K.), FUKUSHIMA (S.), Doi (Y.), MATSUOKA (Y), Application of DBR mode-locked lasers in millimeter-wave fiber-radio system, IEEE J. Lightwave Technol., (2000), 18, pp. 44-49. [14] SMITH (G.H.), NOVAK (D.), AHMED (Z.), Technique for optical SSB generation to overcome dispersion penalties in fibre-radio systems, Electron. Lett. (1997), 33, pp. 74-75. [15] DAVIES (B.), CONRADI (J.), Hybrid modulator structures for subcarrier and harmonic subcarrier optical single side-band, IEEE Photon. Technol. Lett. (1998), 10, pp. 600602. [16] VERGNOL (E.), DEVAUX (F.), TANGUY (D.), PENARD (E.), Integrated lightwave millimetric single side-band source: design and issues, IEEEJ. Lightwave Technol. (1998), 16, pp. 1276-1284. [17] PARK (J.), SORIN (W.V.), LAU (K.Y.), Elimination of fibre chromatic dispersion penalty in 1550 nm millimetre-wave optical transmission, Electron. Lett. (1997), 33, pp. 512-513. [18] YAO (X.S.), Brillouin selective side-band amplification of microwave photonic signals, IEEE Photon. Technol. Lett. (1998), 10, pp.138-140. [19] WAKE (D.), LIMA (C.R.), DAVIES (P.A.), Optical generation of millimeter-wave signals for fiber-radio systems using a dual-mode DFB semiconductor laser, IEEE Trans. Micro. Thy. & Tech, (1995), 43, n° 9, pp. 2270-2276. [20] NOVAK (D.), TUCKER (R.S.), Millimetre-wave signal generation using pulsed semiconductor lasers, Electron. Lett. (1994), 30, n° 17, pp. 1430-1431. [21] AHMED (Z.), NOVAK (D.), WATERHOUSE (R.B.), Liu (H.F.), Optically-fed millimetre-wave (37 GHz) transmission system incorporating a hybrid mode-locked semiconductor laser. Electron. Lett. (1996), 32, n° 19, pp. 1790-1792. [22] AHMED (Z.), NOVAK (D.), WATERHOUSE (R.B.), Liu (H.F.), Millimeter-wave (37 GHz) transmission of data (up to 500 Mb/s) in an optically fed wireless link

62

Millimeter waves in communication systems

incorporating a hybrid mode-locked monolithic DBR laser, Proc Opt. Fiber Commun. Conf. (1997), Dallas, TX, USA, pp. 45-48. [23] AHMED (Z.), NOVAK (D.), WATERHOUSE (R.B.), Liu (H.F.), 37-GHz fiber-wireless system for distribution of broad-band signals, IEEE Trans. Micro. Thy. & Tech, (1997), 45, ne 8, pp. 1431-1435. [24] VON HELMOLT (C.H.), KROGER (U.), KRUGER (K.), GROSSKOPF (G.), A mobile broad-band communication system based on mode-locked lasers, IEEE Trans. Micro. Thy. & Tech. (1997), 45, n° 8, pp. 1424-1430. [25] LEVY (S.), NAGARAJAN (R.), HELKEY (R.J.), HUMPHREY (P.), BOWERS (J.E.), Millimetre wave fiber-optic PSK subcarrier transmission at 35 GHz over 6.3 km using a grating external cavity semiconductor laser, Electron. Lett., (1993), 29, n° 8, pp. 690-691. [26] GEORGES (J.B.), CUTRER (D.M.), KIANG (M.), LAU (K.Y.), Multichannel millimeter wave subcarrier transmission by resonant modulation of monolithic semiconductor lasers, IEEE Photon. Technol. Lett., (1995), 7, n° 4, pp. 431-433. [27] SCHMUCK (H.), HEIDEMANN (R.), HOFSTETTER (R.), Distribution of 60 GHz signals to more than 1000 base stations, Electron. Lett. (1994), 30, n° 1, pp. 59-60. [28] GRIFFIN (R.A.), LANE (P.M.), O'REILLY (J.J.), Radio-over-fiber distribution using an optical millimeter-wave/DWDM overlay, Proc. Opt Fiber Commun. Conf., (1999), San Diego, CA, USA, pp. 70-72. [29] PARK (J.), SHAKOURI (M.S.), LAU (K.Y.), Millimetre-wave electro-optic upconverter for wireless digital communications, Electron. Lett., (1995), 31, n° 13, pp. 1085-1086. [30] CADIOU (J.F.), DEVAUX (R), VEILLARD (J.F.), LE MERDY (B.), GUENA (J.), PENARD (E.), LEGAUD (P.), Electroabsorption modulator for radio over fiber at 38 GHz, Electron. Lett., (1995), 31, n° 15, pp. 1273-1274. [31] PARK (J.), LAU (K.Y.), Millimetre-wave (39 GHz) fibre-wireless transmission of broadband multichannel compressed digital video, Electron. Lett., (1995), 32, n° 5, pp. 474-476. [32] KAWAMURA (H.), IMAI (N.), OGAWA (E.), H. INOMATA (H.), High-speed data transmission using millimeter-wave fiber-optic links. IEICE Trans. Commun., (1996), E79-B, n° 12, pp. 1784-1791. [33] SMITH (G.H.), NOVAK (D.), LIM (C), A millimeter-wave full-duplex WDM/SCM fiberradio access network, Proc Opt. Fiber Commun. Conf., (1998), San Jose, CA, USA, pp. 18-19. [34] KURI (K.), KITAYAMA (K.), STCJHR (A.), OGAWA (Y.), Fiber-optic millimeter-wave downlink system using 60 GHz-band external modulation, IEEE J. Lightwave Technol., (1999), 17, n° 5, pp. 799-806. [35] VERGNOL (E.), TANGUY (D.), CADIOU (J.F.), CARENCO (A.), PENARD (E.), Multi-carrier and m-QAM modulation based on integrated single side band lightwave source, Proc Opt. Fiber Commun. Conf., (1999), San Diego, CA, USA. [36] GEORGES (J.B.), PARK (J.), SOLGAARD (0.), PEPELJUGOSKI (P.), SAYED (M.), LAU (K.Y.), Transmission of 300 Mbps BPSK at 39 GHz using feedforward optical modulation, Electron. Lett. (1994), 30, n° 2, pp. 160-161, 1994.

Progress in millimeter-wave fiber-radio access networks

63

[37] BRAUN (R.P.), GROSSKOPF (G.), RODHE (D.), SCHMIDT (F.), Low-phase-noise millimeterwave generation at 64 GHz and data transmission using optical single sideband injection locking, IEEE Photon. Technol. Lett., (1998), 10, n° 5, pp. 728-730. [38] NOEL (L), MARCENAC (D.), WAKE (D.), 120 MBPS QPSK radio-fiber transmission over 100 km of standard fiber at 60 GHz using a master/slave injection locked DFB laser source. Electron. Lett. (1996), 32, n° 20, pp. 1895-1897. [39] GEORGES (J.B.), PARK (J.), SOLGAARD (O.), CUTRER (D.), PEPEUUGOSKI (P.), LAU (K.Y.), Highdata-rate millimeter-wave subcarrier transmission using feedforward optical modulation, Proc. Opt. Fiber Common. Conf. (1994), Washington, DC, USA, pp. 165-167. [40] SMITH (G.H.), NOVAK (D.), LIM (C), Wu (K.), Full-duplex broadband millimetre-wave optical transport system for fiber wireless access, Electron. Lett., (1997), 33, n° 13, pp. 1159-1160, 1997. [41] NOEL (L), WAKE (D.), MOODIE (D.G.), MARCENAC (D.D.), WestBROOK (L.D.), NESSET (D.), Novel techniques for high-capacity 60-GHz fiber-radio transmission systems, IEEE. Trans. Micro. Thy. 8 Tech. (1997), 45, n° 8, pp. 1416-1423. [42] BRAUN (R.P.), GROSSKOPF (G.), HENTGES (R.), LOCH (S.), ROHDE (D.), SCHMIDT (F.), Fiberoptic microwave generation for bidirectional broadband mobile communications, Proc. IEEE MTTS Int. Micro. Symp. (1997), Denver, co, USA, pp. 225-228. [43] KOJUCHAROW (K.), KALUZNI (H.), SAUER (M.), NOWAK (W.), A wireless LAN at 60 GHz novel system design and transmission experiments, Proc. IEEE MTTS Int. Micro. Symp., (1998), Baltimore, MD, USA, pp. 1513-1516. [44] SMITH (G.H.), NOVAK (D.), Broadband millimeter-wave fiber-radio network incorporating remote up/downconversion, Proc. IEEE MTTS Int. Micro. Symp. (1998), Baltimore, MD, USA, pp. 1509-1512. [45] GOPALAKRISHNAN (G.K.), BURNS (W.K.), BULMER (C.H.), Microwave-optical mixing in LiNbO3 modulators, IEEE Trans. Micro. Thy. & Tech., (1993), 41, pp. 2383-2391. [46] YOUNG (T.), CONRADI (J.), TINGA (W.R.), BER characteristics of ir/4 DQPSK microwave subcarrier signals on optical fiber using Mach-Zehnder modulator nonlinear upconversion, IEEE Photon. Technol. Lett., 0996), 8, pp. 1552-1554. [47] FUSTER (J.M.), MARTI (J.), POLO (V.), CORRAL (J.L), Fiber-optic microwave link employing optically amplified electrooptical upconverting receivers. IEEE Photon. Technol. Lett. (1997), 9, pp. 1161-1163. [48] FUSTER (J.M.), MARTI (J.), CORRAL (J.L), Chromatic dispersion effects in electrooptical upconverted millimeter-wave fiber optic links, Electron. Lett. (1997), 33, n°23, pp. 1969-1970. [49] MESLENER (G.J.), Chromatic dispersion induced distortion of modulated monochromatic light employing direct detection, IEEEJ. Quantum Electron., (1984), 20, n° 10, pp. 1208-1216. [50] RHEINFELDER (C.H.), STROHM (K.M.), METZGER (L), KIBBEL (H.), LUY (J.-F.), HEINRICH (W.), 47 GHz SiGe-MMic oscillator, Proc. IEEE mtts Int. Micro. Symp., (1999), Anaheim, CA, USA, pp. 5-8.

64

Millimeter waves in communication systems

[51] SINNESBICHLER (F.X.), GELTINGER (H.), OLBRICH (G.R.), A 50 GHz HBT push-pull oscillator, Proc. IEEE mtts Int. Micro. Symp. (1999), Anaheim, CA, USA, pp. 9-12. [52] LlM (C.), NlRMALATHAS (A.), NOVAK (D.), WATERHOUSE (R.), GHORBANI (K.), Full-duplex

broadband fiber-wireless system incorporating baseband data transmission and a novel dispersion tolerant modulation scheme, Proc. IEEE mtts Int. Micro. Symp. (1999), Anaheim, CA, USA, pp. 1201-1204. [53] DAGLI (N.), Wide-bandwidth lasers and modulators for RF photonics, IEEE Trans. Micro. Thy. & Tech. (1999), 47, pp. 1151-1171. [54] Lot (K.K.), SAKAMOTO (I.), MEI (X.B.), Tu (C.W.), CHANG (W.S.C), High efficiency 1.3 Lim InAsP-GalnP MQW electroabsorption waveguide modulators for microwave fiber optic links, IEEE Photon. Technol. Lett., (1996), 8, pp. 626-628. [55] IDO (T.), TANAKA (S.), SUZUKI (M.), KOIZUMI (M.), SANO (H.), INOUE (H.), Ultra high-speed multiple quantum well electroabsorption optical modulators with integrated waveguides, J. Lightwave Technol., (1996), 14, pp. 2026-2034. [56] MINEO (N.), YAMADA (K.), NAKAMURA (K.), SAKAI (S.), USHIKOBO (T.), 60 GHz band electroabsorption modulator module, Proc. Opt. Fiber Commun. Conf., (1998), San Jose, CA, USA, pp. 287-288. [57] Moss (D.J.), IDO (T.), SANO (H.), Photogenerated carrier sweep out times in strained \nKGa^ngA\^y quantum well modulators, Electron. Lett., (1994), 30, pp. 405-406. [58] WOOTEN (E.L), KISSA (K.M.), YI-YAN (A.), MURPHY (E.J.), LAFAW (D.A.), HALLEMEIER (P.P.), MAACK (D.), ATTANASIO (D.V.), FRITZ (D.J.), MCBRIEN (G.J.), Bossi (D.E.), A review of Lithium Niobate modulators for fiber-optic communications systems, IEEEJ. Sel. Top. Quantum Electron, (2000), 6, pp. 69-82. [59] DOLFI (D.W.), RANGANATH (T.R.), 50 GHz velocity-matched broad wavelength LiNbO3 modulator with multimode active section, Electron. Lett. (1992), 28, pp. 1197-1199. [60] BURNS (W.)f HOWERTON (M.), MOELLER (R.), KRAHEMBUHL (R.), MCELHANON (R.), GREENBLATT (S.), Low drive voltage, broad-band LiNbO3 modulators with and without etched ridges., J. Lightwave Technol. (1999), 17, pp. 2551-2555. [61] NOGUCHI (K.), MITOMI (O.), MIYAZAWA (H.), Millimeter-wave Ti:LiNbO3 optical modulators, J. Lightwave Technol. (1998), 16, pp. 615-619. [62] WALKER (R.G.), Electro-optic modulation at mm-wave frequencies in GaAs/AIGaAs guided wave devices, Proc. IEEE/LEOS Ann. Meet, (1995), San Francisco, CA, USA, pp. 118-119. [63] SPICKERMANN (R.), SAKAMOTO (S.R.), PETERS (M.G.), DAGLI (N.), GaAs/AIGaAs traveling wave electro-optic modulator with electrical bandwidth greater than 40 GHz. Electron, Lett. (1996), 32, pp. 1095-1096. [64] CHEN (D.), FETTERMAN (H.), CHEN (A.), STEIER (W.H.), DALTON (L.R.), WANG (W.), SHI (Y), Demonstration of 110 GHz electro-optic polymer modulators, Appl. Phys. Lett., (1997), 70, pp. 3335-3337. [65] CHEN (D.), BHATTACHARTA (D.), UDUPA (A.), TSAP (B.), FETTERMAN (H.), CHEN (A.), LEE (S.S.), CHEN (J.), STEIER (W.H.), DALTON (L.R.), High frequency polymer modulators with integrated finline transitions and low Vir, IEEE Photon. Technol. Lett. (1999), 11, pp. 54-56.

Progress in millimeter-wave fiber-radio access networks

65

[66] UMBACH (A.), ENGEL (TH.), G. UNTERBORSCH (G.), Optoelectronic integration of ultrafast photoreceivers, Proc. Int. Top. Meet. Micro. Photon., (1999), Melbourne, Australia, pp. 31-34. [67] WEY (Y.G.), GIBONEY (K.), BOWERS (J.E.), RODWELL (M.), SIEVESTRE (P.), THIAGARAJAN (P.), ROBINSON (G.), 110-GHz GalnAs/lnP double heterostructure p-i-n photodetectors, J. Lightwave Techno/., (1995), 13, n° 7, pp. 1490-1499. [68] KATO (K.), Ultrawide-band/high-frequency photodetectors, IEEE Trans. Micro. Thy. & Tech. (1999), 47, n° 7, pp. 1265-1281. [69] KATO (K.), KOZEN (A.), MURAMOTO (Y), ITAYA (Y.), NAGAT-SUMA (T.), YAITA (M.), 110-GHz, 50%-efficiency mushroom-mesa waveguide p-i-n photodiode for a 1.55-|o.m wavelength, IEEE Photon. Technol. Lett. (1994), 6, n° 6, pp. 719-721. [70] UMBACH (A.), TROMMER (D.), MEKONNEN (G.G.), EBERT (W.), UNTERBORSCH (G.), Waveguide integrated 1.55-|xm photodetector with 45 GHz bandwidth, Electron. Lett., (1996), 32, n° 23, pp. 2143-2145. [71] GIBONEY (K.S.), NAGARAJAN (R.L), REYNOLDS (T.E.), ALLEN (ST.), MIRIN (R.P.), RODWELL (M.J.W.), BOWERS (J.E.), Travelling-wave photodetectors with 172-GHz bandwidth and 76-GHz bandwidth-efficiency product, IEEE Photon. Technol. Lett., (1995), 7, n° 4, pp. 412-414. [72] LIN (L.Y.), Wu (M.C.), ITOH (T.), VANG (T.A.), MULLER (R.E.), Sivco (D.L.), CHO (A.Y.), High-power high-speed photodetectors - design, analysis, and experimental demonstration, IEEE Trans. Micro. Thy. & Tech., (1997), 45, n° 8, pp. 1320-1331. [73] WILLIAMS (K.J.), ESMAN (R.D.), WILSON (R.B.), KULICK (J.D.), Differences in p-side and n-side illuminated p-i-n photodiode nonlinearities, IEEE Photon. Technol. Lett., (1998), 10, n° 1,pp. 132-134. [74] ISHIBASHI (T.), FUSHIMI (H.), Ito (H.), FURUTA (T.), High power uni-travelling-carrier photodiodes, Proc. Int. Top. Meet. Micro. Photon. (1999), Melbourne, Australia, pp. 75-78. [75] LEVEN (A.), BAEYENS (Y.), BENZ (W.), BRONNER (W.), HULS-MANN (A.), HURM (V.), JAKOBUS (T.), KOHLER (K.), LUDWIG (M.), REUTER (R.), ROSENZWEIG (J.), SCHLECHTWEG (M.), GaAS-

based PIN-HEMT photoreceivers for optical microwave and millimeter-wave transmission at 1.55 ^im, Proc. Int. Top. Meet. Micro. Photon. (1998), Princeton, NJ, USA, post deadline paper PD002. [76] TAKAHATA (K.), MURAMOTO (Y.), FUKANO (H.), MATSUOKA (Y.), 52 GHz bandwidth monolithically integrated WGPD/HEMT photoreceiver with large O/E conversion factor of 105 V/W, Electron. Lett. (1999), 35, pp. 1639-1640. [77] UMBACH (A.), VANWAASEN (S.) AUER (U.), BACH (H.G.), BERTENBURG (R.M.), BREUER (V.), EBERT (W.), JANSSEN (G.), MEKONNEN (G.G.), PASSENBERG (W.), SCHLAAK (W.), SCHRAMM (C.), SEEGER (A.), TEGUDE (F.J.), UNTERBORSCH (G.), Monolithic PIN-HEMT 1.55 LUTI photoreceiver on InP with 27 GHz bandwidth, Electron. Lett. (1997), 32, pp. 2142-2143. [78] UMBACH (A.), TROMMER (D.), MEKONNEN (G.), EBERT (W.), UNTERBORSCH (G.), Waveguide integrated 1.55 Lim photodetector with 45 GHz bandwidth, Electron. Lett., (1997), 32, pp. 2143-2145.

66

Millimeter waves in communication systems

[79] ENGEL (T"H.), STRITTMATTER (A.), PASSENBERG (W.), UMBACH (A.), SCHLAAK (W.), DROGE (E.), SEEGER (A.), STEINGRUBER (R.), MEKONNEN (G.G.), UNTERBORSCH (G.), BACH (H.-G.), BOTTCHER (E.H.), BIMBERG (D.), Narrow-band photoreceiver OEIC on InP operating at 38 GHz, IEEE Photon. Technol. Lett. (1998), 10, pp. 1298-1300. [80] HEIDEMANN (R.), VEITH (G.)f Mm-wave photonic technologies for Gbit/s-wirelesslocal-loop, Proc. Optoelectron. Commun. Con/. (1998), Chiba Japan, pp. 310-311. [81] KURI (T.), KITAYAMA (K.), TAKAHASHi (Y.), 60-GHz-band full-duplex radio-on-fiber system using two-RF-port electroabsorption transceiver, IEEE Photon. Technol. Lett. (2000), 12, pp. 419-421. [82] ROWE (W.S.T.), WATERHOUSE (R.B.), NIRMALATHAS (A.), NOVAK (D.), Integrated antenna base station design for hybrid fiber radio networks, Proc. Int. Top. Meet. Micro. Photon. (1999), Melbourne, Australia, pp. 47-50. [83] MATHOORASING (D.), TANGUY (D.), CADIOU (J.F.), LEGAUD (P.), PENARD (E.), Bouchoule (S.), Kazmeirski (C.)f Multicarrier distribution of multiplex digital compressed TV

[84]

[85]

[86] [87]

[88] [89]

channels using harmonic laser source at 38 GHz, Proc. Int. Top. Meet. Micro. Photon, (1998), Princeton, NJ, USA, pp. 13-16. O'REILLY (J.J.), LANE (P.M.), CAPSTICK (M.H.), SALGADO (H.M.), HEIDEMANN (R.), HOFSTETTER (R.), SCHMUCK (H.), RACE R2005: microwave optical duplex antenna link, IEEE Proceedings-J. (1993), 140, pp. 385-391. DEBORGIES (R), MITTRICH (M.), SCHMUCK (H.), JAFFRE (P.) PESCOD (C), Progress in the ACTS FRANS Project, Proc. Int. Top. Meet. Micro. Photon. (1999), Melbourne, Australia, pp. 115-118. AC083 frans Homepages: http://www.infowin.org/ACTS/RUS/ projects/ac803.htm,http://www.infowin.org/ACTS/RUS/trials/ac803.htm. HEIDEMANN (R.), MM-wave photonics - the enabling technology for broadband wireless local loop systems, Proc. Int. Top. Meet. Micro. Photon, (1996), Kyoto, Japan, pp.35-36. HAISCH (H.), T. PFEIFFER (T.), Photonic technologies for hybrid access systems, Proc. Int Top. Workshop Contemp. Photon. Technol. (2000), Tokyo, Japan, pp. 17-20. OGAWA (H.), TSUJI (H.), HIRAKAWA (M.), Novel fiber fed WDM links for millimeter-wave

wireless access systems, (1999), Proc. IEEE MTTS Int. Micro. Symp. Anaheim, CA, USA, pp.1213-1216. [90] LIM (C.), NIRMALATHAS (A.), NOVAK (D.), WATERHOUSE (R.), YOFFE (G.), A WDM architecture for milllimeter-wave fiber radio system incorporating baseband data transmission, Proc. Int. Top. Meet. Micro. Photon., (1999), Melbourne, Australia, pp. 127-130. [91] SMITH (G.H.), NOVAK (D.), LIM (C), A millimetre-wave full-duplex fiber-radio startree architecture incorporating WDM and SCM, IEEE Photon. Technol Lett. (1998), 10, pp. 1650-1652. [92] KALUZNI (H.), KOJUCHAROW (K.), NOWAK (W.), PEUPELMANN (J.), SAUER (M.), SOMMER (D.), FINGER (A.), FERLING (D.), Simultaneous electro-optical upconversion, remote oscillator generation and air transmission of multiple optical WDM channels for a

Progress in millimeter-wave fiber-radio access networks

67

60 GHz high capacity indoor system, Proc IEEE MTTS Int. Micro. Symp., (1999), Anaheim, CA, USA, pp. 881-884. [93] STOHR (A.), KURI (T.), KITAYAMA (K.), HEINZELMANN (R.), JAGER (D.), Full-duplex 60 GHz fiber optic transmission, Electron. Lett. (1999), 35, pp. 1653-1655. [94] GRIFFIN (R.A.), LANE (P.M.), O'REILLY (J.J.), Crosstalk reduction in an optical MMwave/DWDM overlay for radio-over-fibre distribution, Proc. Int. Top. Meet. Micro. Photon., (1999), Melbourne, Australia, pp. 131-134. [95] NIRMALATHAS (A.), LiM (c), NOVAK (D.), WATERHOUSE (R.), Optical interfaces without light sources for base-station designs in fiber-wireless systems incorporating WDM, Proc. Int Top. Meet. Micro. Photon., (1999), Melbourne, Australia, pp. 119-122. [96] NOEL (L), WESTBROOK (L.D.), MOODIE (D.G.), NESSET (D.), 120 Mbit/s QPSK data and multi-channel TV transmission over 13 km fiber to a 60 GHz mobile radio link using an electroabsoprtion modulator as a transceiver, Electron. Lett. (1997), 33, pp. 1285-1286. [97] KURI (T.), KITAYAMA (K.), TAKAHASHI (Y), Simplified BS without light source and RF local oscillator in full-duplex millimeter-wave radio-on-fiber system based upon external modulation technique, Proc. Int. Top. Meet. Micro. Photon., (1999), Melbourne, Australia, pp. 123-126.

Chapter 4

Hybrid 3D integrated circuits at millimeter-wave frequencies: advantages and trends Ch. Person, E. Rius and J. Ph. Coupez Laboratoire d'filectronique et Systemes de Telecommunications, France

I. Introduction Wideband and multimedia services contribute today to the development of millimeter-wave applications regarding required data bit rates and point to multipoint distribution constraints. The maturity of the technology at such frequencies still remains a critical issue for the manufacturers. They need to deliver to private and public users, commercial systems at reduced cost and with multiple possibile utility. In this way, significant efforts are being made to improve and accommodate new integration technologies for both the operators delivering new competitive services with respect to the cable (Local Access Network, Wireless Local Loop etc), and the customers (interactivity, voD.etc). Integration reproducibility and reliability become one of the key parameters, as well as the problem of compatibility between the sub-modules constituting the RF front-ends of such wireless equipment. In thischapterr, we focus on original technological solutions which can be proposed in order to meet the previously described requirements.

II. Emerging integration techniques for mm-wave applications In the past few years, intensive development of mm-wave applications has significantly contributed to the evolution of integration technologies and techniques. The waveguide solution is certainly the best approach in terms of electrical performance (losses), but brings also drastic constraints regarding compatibility (in

Hybrid 3D integrated circuits at millimeter-wave frequencies

69

particular with active MMICS), size, and, as a consequence, cost For these reasons, the traditional machined copper structures have been replaced by plastic injected or molded waveguides [1], This is a significant contribution to the reduction of cost and weight, provided specific metallisation procedures are used (loss considerations), but also for design flexibility (complex geometrical shapes available). For instance, such an approach has been efficiently used for implementing high performance filters, diplexers and antenna arrays [2]. Also, the reliability with active mm-wave components induces the use of restrictive transitions and interconnecting structures for facilitating volume (waveguide) and printed MMIC (Monolithic Microwave Integrated Circuits) technologies. For such reasons, the CPW technology has been considered for a long time as the only solution for developing mm-wave integrated functions, with reasonable tolerances in terms of cost, performances and reliability. Indeed, in comparison with microstrip architectures, that proposed brings interesting properties, especially concerning design flexibilities (according to additional design variables) and losses, the electric and magnetic fields being concentrated at the dielectric-air interface instead of within the dispersive dielectric substrate. The major drawbacks of this emerging technology lie in the control of fundamental parasitic dispersive modes rejection by means of filtering airbridges, as well as microstrip or parallel plate modes potentially generated when a metallic packaging is used. Such constraints have motivated the numerous efforts of MMIC designers to develop integrated air bridges (for reproducibility and modelisation reasons) [3]. In this way, the achievable reduced dimensions (CPW slot widths) are considered for reducing backward radiation in expectation of packaging incidences. On the other hand, the MMIC technology is really dedicated to component mass-production. The actual trend followed by designers is mostly oriented towards hybrid technologies for meeting cost, reconfigurability and other drastic requirements specifically imposed by emerging mm-wave markets. In addition, RF filters and antennas can not reasonably be implemented on GaAs or InP semiconductor substrates for size and cost, as well as for efficiency reasons. The achievement of hybrid technology for mm-wave wireless services appears today as a real opportunity, provided that interconnections with active MMICS can be optimised and electrically well controlled. Furthermore, hybrid technologies are based upon various technological configurations depending on the process and materials involved. This allows one to improve the performances of hybrid systems and sub-systems as original and optimal combination of integrated functions can be manufactured using several methods. Two fundamental questions have to be answered: - Which processes (materials, procedur etc) can be used simultaneously, depending on the basic functions to be associated and integrated? - How can these processes be combined to lead to an universal hybrid integration procedure for mm-wave products?

70

Millimeter waves in communication systems

in. 3D integrationtechniques for improving the uniplanar technology Many studies have shown that the coplanar waveguide can be considered a good alternative to microstrip lines in the millimeter frequency range [4] (high integration level, interconnection facilities, direct ground connections, great flexibility for circuit design). However, there is one major drawback in dealing with coplanar designs: The parasitic and strongly dispersive coupled slotline mode may be excited by non-symmetrical coplanar waveguide-discontinuities, such as bends or T-junctions for example. The suppression of such perturbing modes is achieved by the insertion of bridges over the centre conductor that enforce identical potential values between lateral ground planes. Generally, bridges in microwave-integrated-circuits (MIC'S) are implemented by thin metallic wires ((j> = 17 pm), which are bonded to the ground planes as shown in Figure 1. However, such an operation introduces numerous problems in reproducibility and positioning, which deteriorate the circuit performances. Moreover, the mechanical resistance of such bridges does not guarantee circuit reliability. In this paper, we consider how these problems can be overcome by using a new multi-layer approach for bridge integration. However, dielectric bridges appear as strong discontinuities in coplanar circuits, and must be compensated. Thus, we propose a way to reduce the parasitic influence of the bridges by locally tuning the dimensions of both the bridges and associated compensating lines. m.l. Thick-film multilayer technology for dielectric bridges The basic idea consists of using a protective dielectric layer for interconnecting ground planes by means of a thin conductive printed strip. MMICS solutions have been proposed in the literature, leading to suspended bounding strips as mentioned

Figure 1. Bonding wires on a MIC T-junction.

Hybrid 3D integrated circuits at millimeter-wave frequencies

71

in Figure 2. Unfortunately, this can not be used in MIC technology, because of etching possibilities that are presently available. To overcome such limitations and ensure better mechanical resistance, wellknown multilayer hybrid technologies (MHIC) can be employed for implementing dielectric bridges on MICS. Indeed, in accordance with the significant progress accomplished by dielectric ink manufacturers, it becomes feasible to use standard process, like the thick film technology, for realising integrated mode filtering structures among uniplanar guides. As illustrated in Figure 3, dielectric layers can be locally or globally laid down over a first conductive level via holes are made in this dielectric layer, to ensure electrical connection between the two lower ground planes and the bridge structure. The second conductive level is directly screen-printed and etched on the upper side of the dielectric layer.

Figure 2. MMIC bridges (courtesy ofiEMN-Utte

[3]).

Figure 3. Dielectric bridge in a multilayer technology.

72

Millimeter waves in communication systems

At mm-wave frequencies, photo-printed dielectric materials (HERAEUS ink KQ 110 [5]) are employed to ensure efficient interconnections between the different layers by means of miniature via holes (diameter: 40-50 |im). Notes that laser drilling can also be considered to achieve smaller diameters (#10 jjm). This requires tunable ligthwave equipment (excimer laser) for selection of the materials to be drilled. The proposed thick film process enables us to obtain robust integrated bridges and eliminate potential problems due to positioning and reproducibility. In addition, this process provides an easy control of bridge dimensions and shape. HL2. Parasitic influences of the dielectric bridges Due to the reduced height of the dielectric layers (typically between 10 ^m and 20 pm, for the MIC process used), an additional and significant capacitive parasitic coupling effect is introduced between the inner conductor of the coplanar waveguide and the upper short-circuited strip. Obviously, the characteristic impedance and the effective permittivity are significantly modified in the bridge region. In this study, bridges are modelled by equivalent cascaded transmission lines, taking into account the modified electrical parameters of the local multi-layer structure. Figure 4 shows the influence of the dielectric layer (B) and of the dielectric bridge (C) on the electrical parameters of a 50-Q coplanar waveguide (A). It

Figure 4. Electrical parameters of the integrated bridges.

Hybrid 3D integrated circuits at millimeter-wave frequencies

73

should be noted that this type of dielectric bridge presents a sharp discontinuity for the coplanar waveguide. Fortunately, the overall length of a dielectric bridge is usually small, i.e. about 50 jam. As a result, the influence of one single bridge on the transmitted RF signal remains negligible below 40 GHz. However, practical uniplanar designs require a large number of bridges. This results in an accumulation of parasitic effects in the coplanar circuit and, consequently, deteriorates the electrical responses. Figure 5 represents measured return losses of a 50Q multi-layer coplanar waveguide (strip 80-|jm/slots 55-jim) with (curve 1) and without (curve 2) a dielectric bridge. It should be noted that the performances deteriorate rapidly, with a return loss greater than 15 dB at 40.4GHz. m.3. Compensating dielectric bridges Specific procedures must be found for minimising the distortion due to the low characteristic impedance near to the bridge discontinuity. The basic idea is to associate high characteristic impedance sections on each side of the bridge, to compensate its low characteristic impedance and recover proper matching conditions [6, 7, 8]. This is achieved by reducing, in an appropriate way both the width of the inner conductor under the bridge and the distance (D) it extends each side (Figure 6). The complete structure can now be considered as a low-pass filter (L-C-L equivalent model for A-type compensation) for which the cut-off frequency is related to the parameter D. The dual compensating network can be implemented in a similar way leading to the B-type compensation solution (Figure 6). The main advantage of this type of structures lies in the reduction of technological

Figure 5. Influence of the bridge on a transmission line with/without bridge.

74

Millimeter waves in communication systems

Figure 6. Compensated bridge models.

constraints: it does not require either very narrow strips nor accurate positioning of the bridge. m.3.1. Compensated 50-Q transmission lines: experiment Various dielectric bridge configurations, with and without compensation networks on 50-Q coplanar waveguides can be designed. The length of each dielectric bridge determines the cut-off frequency of the bridge and, therefore, the return loss level within the operating bandwidth. "On wafer" measurements have been carried out on compensated 50-Q transmission lines, over the 1 GHz to 50 GHz frequency band. Both types of compensation yield acceptable results with return loss levels better than 24 dB at 40 GHz (see Figure 7). The compensation technique does not introduce significant additional losses (<0.7 dB up to 40GHz).

Figure 7. B-type compensation.

Hybrid 3D integrated circuits at millimeter-wave frequencies

75

Figure 8. Distributed bend model.

m.3.1.1. Bridges' efficiency: experimental verification Circuits including discontinuities, like T- and cross-junctions or bends, for instance, generate the parasitic coupled slotline-mode on a coplanar structure, especially at high frequencies. Therefore, preliminary measurements were first carried out on Te- and cross-junctions to verify the integrated bridges' efficiency in suppressing the above unwanted mode. All tested circuits, built on an alumina substrate (er = 9.6, thickness = 635 |im), have 60-jim long bridges. Despite lumped electrical L-C models being normally used [6, 8], we have implemented distributed models based upon cascaded transmission lines (bendjunction: see Figure 8). The electrical parameters (Zc, Ereff) per unit length of the lines are calculated with a static finite difference method (2-D analysis). Figure 9 illustrates the behaviour of a cross-junction once all parasitic effects has been properly described and compensated. The measured resonant frequency is equal to 30.4 GHz (30 GHz predicted). According to these results, it thus becomes possible to design accurately a resonator at a fixed frequency. Because the compensation technique can be applied to many cases, e.g. for different bridge lengths and for different numbers of bridges, various topologies of resonators are available for a given response. m.3.1.2 Application to wideband millimeter-wave band-pass filter To validate the above compensated bridges concepts, we designed basic microwave functions such as filters that use integrated bridges. A 5-poles CPW

a

band-pass filter with a 19-dB return loss, 23-GHz operating bandwidth centred at 31 GHz (# 75% relative bandwidth) was synthetised. First, a standard topology is considered. It consists of symmetrical and asymmetrical short circuited stubs and Xg/4 series inverters. The characteristic impedances and dimensions of each section of the structure are presented in the table of Figure 10. Corresponding experimental performances (with classical bonding) are also shown. It should be noted that the same strip width (80 jam) is employed for all resonators. Consequently, only the slot widths are modified to attain the desired characteristic impedances. Differences observed between ideal and experimental values (Figure 10) are mainly due to the fact that junctions were not taken into account in this design model, as well as the air bridges. Figure 11 deals with the same filter theoretical characteristics as above, but implemented this time by multi-layer technology. This facilitates the implicit integration of compensated dielectric bridges. Equivalent distributed scheme of bridges and junctions are included in the design model. Each discontinuity can now be perfectly controlled and matched. The first prototype (A) uses 60-um-long bridges with respect to the central CPW line. The compensation of the prototype (B) was determined in relation to

Hybrid 3D integrated circuits at millimeter-wave frequencies

77

the 100-jjm bridge located on the central conductor. Experimental results are in agreement in both magnitude and phase with the required responses. Moreover, it should be noted that return losses are about 15 dB in the first case and 19 dB in the second case. Insertion losses are around 1 dB at the central frequency (31 GHz) and the rejection level is correct. m.4. CPW technology at mm-wave frequencies: physical and electrical limitations Compared to classical uniplanar technology, the main advantage of multi-layer uniplanar technology is to provide robust and reproducible integrated bridges. Indeed, with this new technology, it becomes possible to control both the shape and the dimensions of bridges. Thus, it is easier to model their parasitic effects in terms of fundamental propagation mode. Nevertheless, coplanar technology can not be universally used, especially for future wireless systems because of their required operating frequencies and bandwidths. For instance, extremely narrow operating bandwidths have to be

78

Millimeter waves in communication systems

Figure 11. Experimental and theoretical results of compensated multilayer coplanar band-pass filters.

Hybrid 3D integrated circuits at millimeter-wave frequencies

79

respected for wireless point to point or point to multi-point nun-wave links. This imposes specific constraints on RF filters for example. One can still consider a CPW implementation by using separating intermediary ground planes. This achieves low coupling levels between adjacent resonant printed structures. Unfortunately, it has some effect on the overall dimension (additional losses) of such passive functions, and complicates the modelisation process (full-wave model required). In this way, additional but complementary guiding structures must be developed and considered to implement both efficient radiating elements and loss-less non-disturbing miniature narrow-band functions. The previously described 3D CPW technology is this time supposed to ensure the interconnections between different sub-modules constituting the RF front-ends. It behaves like a transparent intelligent interconnecting mother-board, including at the same time biasing networks. The 3D integration process can be employed once again, to provide the integration of loss-less interconnecting or filtering structures by means of Thin Film Microstrip Lines (TFMS), or efficient radiating stacked elements by using ceramic/foam composite substrate. Such technological orientation is also wellsuited for improving reliability with active MMICS.

IV. Hybrid integration: benefits of a 3D approach Multi-layer configurations appear today as a good compromise for optimising size and performances in numerous RF functions. As shown before, this can be used for example for improving the reliability of the CPW technology. Previous work was also reported at microwave frequencies. They emphasised the great interest in using overlapped Metal-Dielectric-Metal configurations, in particular with the microstrip approach. Such an orientation brings a complete design flexibility and has been commonly generalized for bypassing intrinsic limitations of basic MIC technologies [9]. At mm-wave frequencies, additional constraints related to the operating wavelength may also encourage designers to employ 3D architectures: - The dimensions of the structures become small, then requiring the use of thin substrates (for preventing substrate surface modes). - Mounting and bonding operations involved with hybrid integrations (to connect active components and passive functions) may introduce nonnegligible perturbations. With the generalisation of chip components and flip-chip techniques, the mother-board substrate dimensions should absolutely become compatible with regard to MMIC pads and bumps. Accessing such low-loss thin substrates has been a major preoccupation in the past few years, with regard to compatibility, measurement, and access conditions. Unfortunately, thin substrates or standard ceramic substrates are not really appropriate for integrating radiating elements regarding radiation efficiency

80

Millimeter waves in communication systems

especially, gain or achievable bandwidth. On the other hand, high coupling levels can be easily achieved through available overlapping possibilities of multileveled architectures. The combination of multi-layer ceramic substrates and low permittivity foam support appears as an optimal composite substrate for such applications. This is discussed in Section IV.2. IV.l. TFMS technology for mm-wave integration: an attractive and complementary transmission line topology Thin substrates are recommended at high frequencies for microstrip technology to prevent parasitic substrate surface modes. Furthermore, in an hybrid integration approach, MMIC active components dedicated to mm-wave applications are characterised by reduced dimensions, especially about external connecting pads. Compatible interconnection techniques must be proposed to minimise electrical associated parasitic effects. The thick-film process can be used for developing Thin Film MicroStrip lines (TFMS) [10] and, as a consequence, 50-Q feeding line widths comparable to MMIC access pad dimensions. Figure 12 describes the basic corresponding topology based upon a motherboard alumina substrate, with a first onductive layer (assumed the TFMS

Figure 12. Transverse geometry of the TFMS topology.

Hybrid 3D integrated circuits at millimeter-wave frequencies

81

ground plane), a thin dielectric sheet (10 to 40 jjin) deposited by a screenprinting procedure, and an upper etched layer (microtrip conductor). In the case of a er = 4.3/40 jjm dielectric sheet, a 50-fi line can be synthesised by means of a very thin strip (W = 75 um) in accordance with the dielectric thickness and, consequently, with common MMIC pad dimensions. In addition, the W/h ratio is compatible with a proper description of the technological structure using standard CAD tools based upon transmission line analysis, despite very high operating frequencies. Avoiding tedious electromagnetic simulations constitutes a significant advantage. Note that this is not the case when dealing with CPW complex architectures or oversized microstrip ones (with respect to wavelengths). IV.1.1. Accessing TFMS structures [11]

The reduced dimensions offered by the TFMS technology are well-suited for designing efficient RF-circuits at mm-wavelengths. However, accessing such technological structures can bring some problems in terms of dimension compatibility with standard feeding technologies (CPW and microstrip), electrical behaviour (parasitic effects, bandwidth & frequency limitations) and mechanical tolerances. Therefore, we have developed high performances CPW to TFMS transitions, keeping in mind then the advantage of the CPW motherboard concept in the millimeter-wave range. Such a transition is based upon Xg/4 overlapped coupled lines separated by a thin dielectric sheet which constitutes simultaneously the TFMS substrate. Note that the TFMS ground plane is directly connected to the CPW ground planes and may also be considered as a potential CPW parasitic mode filtering structure (Figure 13). An optimisation procedure was applied to control the lower and upper cut-off frequencies of this series capacitive coupling structure. Additional inductive and capacitive sections were cascaded, each apart from the MIM (Metal-InsulatingMetal) capacitance to lead to a higher order highpass filter. Quite a wide operating bandwidth can be achieved, as shown in Figure 14 where experimental results are reported for a 40% relative bandwidth transition (for a -17 dB return loss) centred at 41 GHz. Wider bandwidths can be considered, using the numerous tuning variables, and taking advantage of the wide range of coupling achievable with the 3D approach. Insertion losses of about 0.8-dB only have been registered inside the 15-dB return loss bandwidth for a 1 mm long structure. Such transitions appear really as a convenient solution for improving interconnections between active and passive modules in future millimeter-wave hybrid front-ends. IV.1.2. TFMS approach for designing efficient passive functions

With the TFMS technology, the strip width becomes fairly small for a wide set of characteristic impedance values. Thus, it becomes therefore negligible with respect to wavelengths in the 30-100 GHz range. In this case, TEM approximations can be

82

Millimeter waves in communication systems

Figure 13. Design of a back to back CPW to TFMS transition.

considered, and the design of passive function like filters and couplers can be directly implemented using traditional topologies and synthesis methods (parallel coupled lines filters, Kg/4 resonators, etc). Furthermore, since the substrate thickness is kept small, quite low coupling levels are achieved with standard X,g/4 coupled resonators, even when the smallest etchable spacing is reached (the odd and even mode impedances of the coupled lines are quite similar (Zoo # Zoe) because of ground plane proximity). This is especially attractive for synthesising very narrow-band band-pass filters at mm-wave frequencies. Figure 15 describes a TFMS parallel coupled lines filter designed at 41 GHz, and compared with its corresponding microstrip version on 254-(om alumina substrate. Note that the LAV and L/S ratios (W, L: width, length of the resonators, S: spacing between adjacent resonators) for each resonator allows a transmissionline description with the TFMS solution.

Hybrid 3D integrated circuits at millimeter-wave frequencies

83

84

Millimeter waves in communication systems

Figure 15. TFMS/microstrip layouts of a narrow-band filter.

Figure 16 gives some experimental results obtained with the TFMS filter, including CPW to TFMS transitions for probe testing facilities. No tuning operation is required after manufacturing, despite the severe requirements. Some fair agreement is obtained between simulation and measurement given the fact that the transmission losses are not accounted for in the model.

Hybrid 3D integrated circuits at millimeter-wave frequencies

85

Some specific effort must be provided by both manufacturers and designers for decreasing conductive losses with TFMS topologies with the reduced width and thickness of the conductors. Another solution, proposed in [11] consists in artificially lowering the filter order by using an adequate and original synthesis technique. The CPW to TFMS transition previously described can be considered as an overcoupled resonator (overlapped structure). It can be introduced within the filter which contributes to reduce its size, losses and technological sensitivity. FV.1.3. TFMS access for improving compatibility with MMICs

The achievement of commercial mm-wave MMICS and chip is a key factor in the emergence of hybrid technologies in future applications. The critical dimensions of these new products (typically 200*300 pm - thickness 120 um) become also critical from an electrical point of view. The connecting bounding wires usually induce important parasitic effects depending on their length, diameter and contact quality with pads when using microstrip or CPW technologies. Compatible dimensions between the feeding line network and the MMIC pads are not really respected, as well as the common metallic support manufactured for compensating thickness discontinuities. The TFMS configuration is undoubtly an attractive solution for accessing MMICS and chips with lowest perturbation phenomena: - Dimensions active and passive feeding ports are quite similar. - Bounding operations can be made less perturbing by minimizing the number of wires and their length. Since the MMIC is usually mounted on a ground plane, the thin dielectric sheet can be etched, thus ensuring the perfect control of the chip location when reported on the motherboard alumina substrate. In addition, ground plane continuity is readily obtained, and no additional bounding wires are required for interconnecting MMIC and CPW or microstrip ground planes (see Figure 17). IV. 1.4. TFMS process for mm-wave flip-chip components

Finally, metal-dielectric-metal implementation (i.e. TFMS structures) will be found in a number of applications that involve future mm-wave flip-chip components. Photo-imageable dielectric materials are of great interest to realise discrete connections between lower shielded printed structures and upper surface-mounted flip-chip components. Figure 18 depicts a beam-lead schottky diode mounted across a 30 jim width slotline, with a separating 20 (jm dielectric layer. The interconnection between the diode pins and the slotline arms is performed through 40-um via holes. Numerous advantages appear: -Using a 3D technological process, the passive circuit is completely shielded and protected, thus simplifying mounting.

86

Millimeter waves in communication systems

Figure 17. Interconnection between MMIC and TFMS technology.

- Such an implementation also contributes to simplification of the structure analysis. Indeed, the input impedance can be extracted at the diode connecting ports, using either the Agilent-MDS™ multilayer library or 2.5D Agilent-Momentum™ simulator. Consequently, the non-linear modelisation can be done on MDS® once the complete passive structure is modeled as a multiport network. This is illustrated in Figure 19 for a balanced mixer based upon an annular slotline structure [12] (Figure 18). IV.2. Mixing technologies for efficiency and miniaturization Cost reduction and miniaturization efforts will be achieved in the future if global synthesis and integration techniques are considered. Hence, the radiating network should be located as close as possible to the RF-front-end modules. In addition, this will also result in improving the C/N ratio as well as the reconfigurability of such a module. With the selected hybrid integration approach, active and passive functions are integrated on alumina motherboard substrate, i.e. high permittivity substrate, for size and sub-module compatibility requirements. On the other hand,

Hybrid 3D integrated circuits at millimeter-wave frequencies

87

Figure 18. Beam-lead Schottky diode across a slotline ring.

radiating elements must be integrated on low permittivity thick substrates for radiation efficiency and electrical performances (e.g. bandwidth etc). The hybrid process should accommodate the above requirements which enforce opposite trends in terms of permittivity values. Once again, the flexibility brought about by the 3D concept leads to new prospects for mm-wave applications: • Constructing composite ceramic/foam substrates by combining pressmoulding and screen-printing process on alumina motherboard substrate. • Translating and matching the well-known Membrane process to standard 3D hybrid techniques. IV.2.1. Composite ceramic/foam substrate for global optimisation procedure [13] Work on mm-wave antennas is focused today on the use of new well-suited materials, based upon foam (er = 1.07) or honey-cell structures which offer excellent performances up to 60 GHz. The metalisation and etching processes

88

Millimeter waves in communication systems

a - Decomposition of the structure for a local 20 analysis

Figure 19. Non linear analysis of a 3D CPW mixer: procedure.

remain critical procedures. However, new solutions have been recently proposed through press-moulding or chemical deposition techniques. Printed patch arrays have been implemented on such a support, with specific efforts concerning the miniaturization of the basic radiating elements (better control of the antenna array response), by dielectric insert technique. Difficulties arise when considering the microstrip array feeding network, because of the (er,h) values of the substrate: (wide strips "*• parasitic coupling"*• side lobes **, cross-polarisation «*). Therefore, stacked structures and aperture coupled antenna arrays are used rather than planar configurations, emphasising once again on the benefit of 3D approaches. Figure 20-a describes a microstrip patch antenna made on foam substrate fed by a non-resonant slotline located in the ground plane. This slot is

Hybrid 3D integrated circuits at millimeter-wave frequencies

89

Figure 20. Patch antenna on a composite substrate.

excited through a CPW structure made on alumina substrate to obtain efficient coupling. At mm-wave frequencies, the 3D passive and active functions described above can be directly combined with such antenna arrays on foam material, through the common ceramic mother-board substrate and an intermediate slodine feeding network. Such technique has been validated up to 20 GHz only (see Figure 20-b [14]), but the concept can be easily extended to mm-waves. This original association is obviously a good compromise for combining efficiency and miniaturization criteria in a global integration approach. Notes that specific synthesis techniques have been proposed in order to introduce technological constraints during the design task (global synthesis). This leads to increase the bandwidth of the radiating patch.

90

Millimeter waves in communication systems

IV.2.2. Membrane technology as an alternative solution The other opportunity, initially proposed by Rebeiz [15], and considered by numerous authors consists in developing membrane antennas. Thin film membrane having a permittivity close to that of air, radiation efficiency is increased. In addition, optimal coupling between guided and radiating modes is obtained, which yields wideband high efficiency antenna arrays. Interconnecting such a technology with standard ones has been a great challenge during the last few years. Various possibilities are now considered [17], by using electromagnetic coupling (comparable to the method mentioned above) or suited mounting techniques. A 4*4 aperture fed membrane antenna array designed at 40 GHz is shown in Figure 21. Each basic radiating patch is designed on a membrane section (for radiating efficiency), while the microstrip feeding network is implemented on a backside substrate for which the ground plane acts as a shielding plane. The excitation of the different elements is therefore made easier through the slot apertures, while parasitic couplings are properly rejected, each antenna being located on an electromagnetic shielded cavity. Ultra-low cost membrane structures on copper support, processed as flipchip components, and reported on an alumina motherboard substrate were reported a second time [18]. This is illustrated in Figure 22 in the case of a membrane 50-£2 line mounted on an alumina motherboard substrate with CPW access, through miniature bumps (50 jam diameter). In this way, one can easily combine 3D functions which are integrated on alumina substrate with membrane modules. These are used for critical sections like antennas or narrow-band filters for which the membrane characteristics should be efficiently selected (losses, radiation efficiency, etc).

Figure 21. Aperture fed membrane antenna array at 40 GHz.

Hybrid 3D integrated circuits at millimeter-wave frequencies

91

V. Conclusion The main objective of this paper was to demonstrate that hybrid integration techniques appear today as attractive solutions for developing low-cost high efficiency mm-wave modules and sub-systems, especially when 3D integration techniques are considered. On one hand, the reliability of CPW technology, which can be assumed as the reference technology at such frequencies, can be really improved by implementing integrated dielectric bridges for parasitic mode suppression. Original compensation techniques can be used for controlling the corresponding operating bandwidth. On the other hand, the uniplanar approach cannot be considered as an universal response for integrating all mm-wave functions, with any requirements. Using the hybrid orientation, interconnections between basic functions or elements remain sensitive operations, as well as the development of high performances passive devices (narrow-band filters, antennas network etc). The flexibility of 3D integration techniques was therefore investigated, using various available technological architectures, leading to original structures with improved performances with respected to standard approaches. Some efforts remain to be made concerning mounting techniques, especially the flip-chip ones. However, the actual evolution of both the chemical industry (for

92

Millimeter waves in communication systems

delivering high performance dielectric or conductive inks) and electromagnetic simulators should help designers to find optimal integrated hybrid 3D modules. These will remain competitive with regard to MMIC approaches up to 80 GHz [19]. Acknowledgements The results presented in this paper owe much to the contributions of T. Le Nadan and S. Perrot from LEST/ENST Bretagne (France). Thanks are also due to Professor Serge Toutain, from IRCCYN Nantes (France) for discussions and helpful suggestions concerning mm-wave aspects. REFERENCES [1] GOEBEL (U.), REHME (F.), SCHMITT (D.), "New microwave packaging technologies for high-rel. and low-cost applications", Digest of the 28th EuMC, The Netherlands -

1998. [2] MEINEL (H.H.), "Commercial applications of millimeter waves: History, present status, and future trends", IEEE Trans. Microwave Theory Tech., 43, pp. 1639-1653, Jul.1995. [3] BORET [S.] et al., "Modeling of passive coplanar elements for W-band ICs, experimental verification up to 110GHz and parasitic mode coupling study" Digest of the 28th EuMC, Amsterdam, The Netherlands - Oct. 98 - pp. 190-195 [4] HOUDART (M.), "Coplanar lines: application to broadband microwave integrated circuits" - Digest of the 7th EuMC, Prague, 1976, pp.49-53. [5] BARNWELL (P.), SHORTHOUSE (G.), "High performance material technologies for advanced circuit assemblies and MCMs" - Microelectronics International Journal ofiSHM Europe, n° 38, pp. 28-31, Sept 1995. [6] KOSTER (N.H.), KOBLOWSKI (S.), BERTENBURG (R.), HEiNEN (S.), WOLFF (I.), "Investigations on air bridges used for MMICS in coplanar waveguide technique", Digest of the 19th EuMC, London 1989. [7] STEPHAN (L), COUPEZ (J. P.), Rius (E.), PERSON (C), TOUTAIN (S.), "Integration of various types of compensated dielectric bridges for mm coplanar applications" MIT'S symposium, San Francisco 1996. [8] Rius (E.), COUPEZ (J.P.), TOUTAIN (S.), PERSON (C), LEGAUD (P.), "Theoretical and Experimental Study of Various Types of Compensated Dielectric Bridges for Millimeter-Wave Coplanar Applications", IEEE Trans on Microwave Theory and Technique, 48, n° 1, Jan 2000, pp. 152-155. [9] PERSON (C.), SHETA (A.), COUPEZ (J.P.), TOUTAIN (S.), "Design of high performance band-pass filters by using a Multi-layer Thick-film technology", Digest of the 24th European Microwave Conference, Cannes, Sept. 1994, P2.6. [10] ROBERTSON (G.), "Ultra low impedance Transmission Lines for multilayer MMICS" MTT-S Digest, pp 145-148, 1993.

Hybrid 3D integrated circuits at millimeter-wave frequencies

93

[11] LE NADAN (T.), COUPEZ (J), QUENDO (C), PERSON (C.), "3D integrated narrowband filters for millimetre-wave wireless applications", IEEE Mn-Symp. Digest, Boston, June 2000. [12] HETTACK (K.), COUPEZ (J.P.), SHETA (A.), TOUTAIN (S.), "Practical Design of uniplanar broadband sub-systems; application to wideband hybrid magic tee", MTT-S Digest, pp. 915-918, 1994. [13] LATTARD (H.), TOUTAIN (Y.), LE MAGUER (S.), COUPEZ (J.R), PERSON (C.), "High performance miniaturized microstrip antennas on locally non-homogeneous substrates", Digest of the 29th EuMW, Munchen, Germany, Oct. 4-8, 1999. [14] LE NADAN (Th.), TOUTAIN (S.), COUPEZ (J.P.), PERSON (C.), "Optimization and miniaturization of a filter/antenna multi-function module using a composite ceramic-foam substrate", Proc. IEEE Mir-Symp., Anaheim, USA - June 99. [15] REBEIZ (G.M.), KASULIGAM (D.P.), STIMSON (P.A.), Gou (Y.), RUTLEDGE. (B.), "Monolithic Millimeter-Wave Two-Dimensional Horn Imaging Arrays", IEEE Trans.Ant.and Propag., Vol. AP-28, pp. 1473-1482, Sept 1990. [16] PERSON (C.), PERROT (S.), CARRE (L.), TOUTAIN (S.), "Ultra Low-Cost Membrane Technology For Millimeterwave Applications", IEEE MTT-S D/grest,1998. [17] PERROT (S.), PERSON (C.), CARRE (L.), NEY (M.), "Proposed design solutions for lowcost flip-chip membrane devices for mm-wave applications", IEEE MTT-S D/gest,1999, pp. 1905-1908.

Chapter 5

Printed millimeter-wave reflectarrays D. Pilz DaimlerChrysler Aerospace, Ulm, Germany

W. Menzel Microwave Techniques, University of Ulm, Germany

I. Introduction For communication and sensor applications, antennas with low profile, low loss and low production cost are required. While planar antennas are optimal with respect to antenna depth and cost, they suffer from high losses, especially for narrow beamwidth [1, 2]. Arrays of horn antennas with a waveguide feed network [3] or waveguide slotted arrays [4, 5] have lower loss, but partly narrow band and complicated in their design, and they do not readily lend themselves to low cost fabrication. Recently, some work has been reported on the fabrication of waveguide networks and antennas according to [3] using plastic injection molding and electroplating [6]. An alternative solution for printed antennas are quasi-optically fed printed antennas, i.e. printed reflector type antennas consisting of arrays of printed elements acting as fixed reflection phase shifters [8-17]. Different concepts for such antennas have been reported, ranging from Fresnel type antennas [9] to arrays with continuous phase adjustment using microstrip patch antennas with fixed length stubs [8, 12], patches with varying dimensions [10, 11, 13, 14, 16, 17], or even electronically tunable reflection phase angles [15]. This paper describes design and performance of a number of antennas based on the reflectarray principle with emphasis on the exploitation of the dual polarization properties of the planar structures. The following section of this paper gives a short introduction to the basic computation method of periodic and non-periodic antenna structures. In the next section, a first offset fed antenna for a single polarization is presented, together with results of a full wave characterization of the complete antenna. In the fourth section, a dual polarization antenna with equal radiation characteristics in the two polarizations, but different feed points, is presented, and two different approaches for dual frequency reflectarray antennas will be demonstrated, while Section V will

Printed millimeter-wave reflectarrays

95

deal with a combination of focussing reflectarray and polarization twisting, leading - together with a polarizing grid - to low profile folded reflector antennas. Examples of communication antennas in the frequency ranges of 28 GHz and 58/60 GHz will be presented.

II. Some principles of theoretical calculation of antenna structures The basis for the design procedure of reflectarray antennas as described in this paper is a periodic array of dipoles printed on a single dielectric substrate with backside metallization. Typically, the distance between the printed elements is chosen to be about half a free space wavelength. In this way, grating lobes can be avoided for all antenna arrangements. The reflection behavior of such dipole arrays is calculated using a standard spectral domain method and Fioquet's theorem, e.g. [7, 25]. The basic equation describing the fields excited by an incident plane wave on a periodic structure is given by 0)

E^^Z-J+E^E^.

All quantities are Fourier transforms, indicated by the tilde. The electric o fields are taken in the metallizationplane. Z is the dyadic Green's function of a dielectric layer on a ground plane

(2)

a 2 +p 2

and

a and P are the Fourier variables for x and z. J represents the current density on the strip, E^ is the incident plane wave, and E^ is the electric field reflected by the dielectric layer without the dipoles. The indices 1 and 2 in the

96

Millimeter waves in communication systems

preceding equations refer to the free space and the dielectric layer, respectively. The current densities on the metal elements - in this paper, only rectangular elements are considered - are expanded with entire domain functions including edge terms [25]. Galerkin's method is used to transform Equation (1) into an inhomogeneous system of equations. Due to the periodic nature of this arrangement, only discrete Fourier variables have to be taken into account for the numerical procedure. The system of equation can be solved for the amplitudes of the current density functions and the reflection coefficient of the wave. The wave incident on the substrate is completely reflected, the phase angle, however, depends on dipole length and, to a minor degree only, on dipole width. In Figure 1, the reflection phase angle is plotted for a plane wave incident from broadside. Duroid material with a thickness of 0.254 mm and a dielectric constant of 2.22 is used as substrate in this example. Frequency is 58 GHz, and element distance is 2.4 mm in both directions. The resulting phase angle varies over nearly 360 °. In a first approach, these reflection phase angles calculated from a periodic structure can be used for the design of reflectarray antennas consisting of dipoles on a periodic grid, but with varying dipole dimensions. This design procedure mostly gives reasonably good results, although it provides only a rough estimate

Figure 1. Reflection phase angle of a plane wave reflected from a periodic array of printed dipoles, (Frequency 58 GHz, substrate thickness 0.254 mm, dielectric constant 2.22, dipole grid spacing 2.4 mm, normal incidence of the wave).

Printed millimeter-wave reflectarrays

97

of the actual situation. Therefore, Equation (1) was extended to the calculation of a complete reflectarray [18]. The incident field now is the field transformed from the aperture of the feeding antenna into the plane of the dipole elements. The current density now is the superposition of the contributions of all individual dipoles. Although the basic procedure remains the same as with the solution of Equation (1), there are some important differences: • Due to the non-periodic arrangement, the Fourier variables are now continuous. Consequently, the poles occurring in the Green's function {Equations (3), (4)} have to be taken into account by residuum calculations. • Due to the distances between the different dipoles, highly oscillating terms appear in the reaction integrals resulting from use of Galerkin's method. With a modified integration procedure including an analytical integration of an approximated kernel of the integrals, these can be evaluated effectively [19]. From the amplitudes of the current densities, both the radiation diagram of the complete antenna as well as amplitude and phase of the outgoing (plane) wave can be computed. The phase distribution of the outgoing wave calculated immediately in front of the reflectarray allows an association of phase errors to the underlying dipoles, thus enabling an optimization of the antenna.

III. Offset-fed printed reflectarray antenna Based on earlier work [13], an offset-fed printed reflector antenna for a single polarization at 24 GHz was designed, fabricated and tested. The layout of the reflector is shown in Figure 2. Polarization is horizontal, the diameter of the array was chosen to 200 mm, and the feed point is located 150 mm away from the substrate plane and shifted down from the array center by 216 mm. The substrate material has a thickness of 0.76 mm and a dielectric constant of 2.5. The dipoles are 1.5 mm wide. For each dipole, its length was chosen to give a plane wave front of the reflected wave propagating perpendicular to the substrate surface. The spectral domain calculation as described above was applied to the complete antenna in order to check the simple design approach. Theoretical and experimental radiation diagrams of the offset-fed antenna for H- and E-plane are plotted in Figure 3. Up to ±15°, the agreement is rather good; for larger angles, effects like the finite substrate size not included in the calculations or direct feed radiation lead to some deviations of the theoretical results from the measured ones.

IV. Dual polarization antennas As was shown hi Figure 1, the reflection phase angles of the dipoles mainly depend on length. The reflection phase angles in the orthogonal polarization are then determined by the element width. Including some minor corrections of the

98

Millimeter waves in communication systems

Figure 2. Layout of the 24 GHz, single polarization printed offset

reflectarray.

element geometry, reflectarrays thus can be designed for mostly independent performance in the two polarizations. This may include: • • • •

equal radiation characteristics with different feed points, different radiation characteristics with equal or different feed points, operation at different frequencies, generation of circular polarization (to this end, a 90 ° phase difference has to be adjusted between the two beams with equal amplitude performance), • generation of polarization twisting between incident and reflected wave. This will be explained and exploited in detail in Section V. As a first example to evaluate this principle, an antenna is presented having two different offset feed point, but equal radiation characteristics in the two polarizations. Such an arrangement allows an easy separation of orthogonal polarizations without the need for a separate transmission line polarizer. The antenna structure and the layout of the reflectarray are shown in detail in Figure 4, frequency of operation being 20 GHz. The four radiation diagrams are plotted in Figure 5. The antenna diameter of 180 mm results in nearly equal beamwidths of 5°-5.5 ° for both planes and polarizations. Sidelobe level is -17dB or better; some unsymmetry is caused by the offset feeding in both planes. In some applications, a single aperture is required for different applications, e.g. at different frequencies. To this end, two approaches were investigated based on reflectarrays as described before. The first approach simply uses a single polarization antenna as described in Section III or in [13]. As the backside metallization needs to reflect only the respective polarization of

Printed millimeter-wave reflectarrays

99

operation of the antenna, it equally can be replaced by a printed grid (Figure 6). Together with relatively narrow dipoles, this reflectarray is nearly transparent at the orthogonal polarization. In the example presented here, the dipole width is 0.5 mm. To compensate a slight difference between the reflection phase angle of the grid and a solid metallization, the value of the thickness of the dielectric substrate is slightly modified for the design process. The focal length of this (nearly) centrally fed array is 55 mm only. Consequently, the design approach

100

Millimeter waves in communication systems

Figure 4. Layout and structure of the 20 GHz dual polarization reflectarray antenna with two different feed points.

on the basis of a plane wave incident on a periodic structure gives only a first approximation of the necessary dipole geometries. The full-wave computation method therefore was used to optimize the phase distribution in the antenna aperture. Theoretical and experimental radiation diagrams of the antenna as depicted in Figure 6 are plotted in Figure 7. Although some effort was made to approximately include the blocking effect of the feed waveguide in the computation, some deviation in the side lobe level can be recognized between experiment and calculated results. Beamwidth and side lobe level are 4° and about -23 dB in the H-plane and 3.6° and -19 dB in the E-plane, respectively. The transparency of this antenna in the orthogonal polarization was tested in Xband (8-12 GHz), and an insertion loss of 0.2 dB only was found. As a consequence, such an antenna provides a high degree of freedom for a combination of different antenna applications in a common aperture. In the second example, the planar reflectarray is designed for operation at two different frequencies in the two linear polarizations. To evaluate the limit of this procedure, a combination of 24 GHz and 60 GHz in a common aperture was investigated. The full range reflection angles requires a maximum dipole length of X/2 at 24 GHz, including the effect of the dielectric. On the other hand, to avoid grating lobes at 60 GHz, the element spacing must not exceed

Printed millimeter-wave reflectarrays

101

Figure 5. Radiation diagrams of dual a polarization reflectarray for two different feed points. (Substrate thickness 0.76 mm, dielectric constant 2.5, cell size 7.5 mm x 7.5 mm, frequency 20 GHz).

half a free space wavelength at 60 GHz. To fulfill these two requirements, a substrate with a dielectric constant of 6 and a non-quadratic grid were chosen. Substrate thickness is 0.254 mm. The overall reflector with a diameter of 150 mm includes more than 5000 dipoles. The reflector dipole structure is shown in Figure 8. The 24 GHz operation is performed with horizontal, the 60 GHz

102

Millimeter waves in communication systems

Figure 6. Basic principle of a "transparent" antenna using a printed reflectarray with a grid as ground plane.

operation with vertical polarization. A single offset feed position was selected with a distance of 160 mm to the substrate plane and 95 mm below the array axis. Figure 9 displays the azimuth radiation diagrams at the two frequencies. At 24 GHz, a reasonably good radiation characteristic with a beamwidth of 5.2° can be seen. At 60 GHz, shoulders at the sides of the main lobe rise to about -15 dB; the beamwidth is 2.3°. Partly, this is due to the fact that the antenna was tested at a distance of 6 m only, compared to a necessary far field distance of 9 m according to the familiar 2D2/X formula.

V. Folded reflector antennas The focussing array can be modified to include polarization twisting of the electromagnetic field. Together with a printed polarizing grid or a slot array this can be applied to the design of a folded reflector antenna [201. The principal function of this antenna arrangement is indicated in Figure 10. The radiation of the feed is reflected by a printed grid or slot array at the front of the antenna. Then the wave is incident on the printed focussing and twisting reflector. The

Printed millimeter-wave reflectarrays

103

Figure 7. Radiation characteristics of the "transparent" reflectarray. (Substrate thickness 0.76 mm, dielectric constant 2.5, cell size 5 mm x 5 mm, frequency 35 GHz).

twisting principle of the reflector is indicated in Figure 11. The dipole axes are tilted by 45° with respect to the incident electric field. The dimensions of the dipoles are selected in such a way that a phase difference of 180° occurs between the two components of the reflected wave. Thus, the polarization of the total reflected wave is twisted by 90° compared to the incident one. Such a twisting performance is possible for a wide variety of combinations of dipole width and length, differing only in the absolute reflection phase angle. This

104

Millimeter waves in communication systems

Figure 8. Layout of a 24/60 GHz dual frequency reflectarray antenna.

Figure 9. Azimuth radiation diagrams of the dual frequency reflectarray antenna. (Substrate thickness 0.254 mm, dielectric constant 6, cell size 3 mm x 1 mm).

Printed millimeter-wave reflectarrays

105

overall phase shift is adjusted according to the focussing (phase shifting) requirements of the reflector. The outgoing plane wave then can pass the grid or slot array (Figure 10). The original design of this antenna once again is carried out on the basis of periodic structures. For varying dipole dimensions, the reflection phase angles are calculated for both principal polarizations. The optimum combination of phases is then selected from this set of data according to both twisting and focussing requirements. Even circular polarization can be achieved with this type of antenna if the polarizing grid or slot array is replaced by a circular polarizer as described in [21]. Due to the folding of the printed reflector antenna, very compact, i.e. relatively low profile antennas can be realized. From the feed, the power is

Figure 11. Principle of polarization twisting by a printed periodic reflectarray.

106

Millimeter waves in communication systems

distributed by radiation, and most of the dipoles of the reflectarray are not in resonance. Consequently, losses are very low. As the antenna consists of two printed substrates only - together with the feed, it lends itself to low cost design as well. In a final stage, even the feed might be integrated with the reflectarray substrate. After first tests at 20 GHz [20], antennas of this type have been realized at 28.5 GHz, in the 60 GHz range [23, 24], and at 76.5 GHz for automotive applications, including even beam scanning by tilting the reflectarray [22]. Figure 12 shows the layout of the 28.5 GHz antenna designed for a pointto-point communication system. Substrate thickness for the reflector and the polarizing grid are 0.5 mm and 3.28 mm, respectively; the dielectric constant is 2.35. The distance between the two substrates is 55 mm, the size of the antenna 260 mm x 280 mm. A conical horn is used as feed. H- and E-plane radiation diagrams are plotted in Figure 13, top. Beamwidths are 2.4° and 2.3°, and the side lobe level is better than -20 dB. At the bottom of Figure 13, the E-plane radiation diagrams are shown at frequencies of 27.5 GHz and 29.5 GHz compared to 28.5 GHz. Except for a slight widening of the beam at a level below -10 dB, good performance can be stated over a bandwidth of at least 7%. For the 61 GHz ISM band as well as for the 58 GHz communication band, an antenna with a diameter of 100 mm and a depth of only 25 mm was designed;

Figure 12. Layout of the focussing and twisting reflector for the 28 GHz reflectarray antenna.

Printed millimeter-wave reflectarrays

107

the layout of the reflector is plotted in Figure 14. Substrate thickness is 0.254 mm for the reflector and 1.58 for the polarizing grid, the dielectric constants are 2.22 and 2.35, respectively. Figure 15 shows the E- and H-plane radiation diagrams at 58.4 GHz as well as the H-plane diagrams at 59.6 GHz and 61.6 GHz compared to that at 58.4 GHz. Except for some slight shoulders at the highest frequency, side lobe level is below -24 GHz. At 61.2 GHz, the

H-plane at 28.5 GHz. Bottom: E-plane at 27.5 GHz, 28.5 GHz, and 29.5 GHz. Reflector: substrate thickness 0.5 mm, dielectric constant 2.22, cell size 4.8 mm x 4.8 mm. Grid: substrate thickness 3.28 mm, dielectric constant 2.35).

108

Millimeter waves in communication systems

Figure 14. Layout of the 58/60 GHz folded reflectarray antenna.

beamwidths are 4.1° in H-plane and 3.6° in E-plane; this is larger than expected from the aperture, but as the feed horn diameter was too big, an amplitude taper of -15 dB occurred at the reflector edges. Gain was measured to 33 dB, compared to 32.6 dB calculated from the usual formula 270007(0 x O), the beamwidths 0 and O taken in degrees. This demonstrates the low loss performance as mentioned above. First tests even have been initiated to further reduce the depth of this antenna; first results indicate a performance with only slightly increased side lobe levels for an antenna depth of 20 mm. Further activities are directed towards a full wave characterization of this type of antenna to enable an optimization as it was done with the 35 GHz antenna (Section IV, [18,19]). In this case, however, there is a wide variation of dipole widths requiring a higher number of expansion functions for the current densities on the dipoles. This - together with the necessity to take into account two polarizations - leads to a considerably increase of computational effort. First results have been obtained for the 60 GHz antenna, although a good convergence has not really been achieved. Nevertheless, a reasonable agreements between theory and experiments can be stated with this first effort, as can be seen from Figure 16.

Printed millimeter-wave reflectarrays

109

Figure 15. Radiation diagrams of the V-band folded reflector antenna. (Top: E- and Hplane at 58.4 GHz. Bottom: H-plane at 58.4, 59.6, and 61.6 GHz. Reflector: substrate thickness 0.254 mm, dielectric constant 2.22, cell size 2.4 mm x 2.4 mm. Grid: substrate thickness 1.58 mm, dielectric constant 2.35).

VI. Conclusion Design, partly optimization, and experimental results have been demonstrated for a number of reflectarray antennas. Both lengths and widths of the reflecting elements are included into a dual polarization design. This includes antennas with dual frequency operation. If the dual polarization properties are used to pro-

110

Millimeter waves in communication systems

ffl

folded reflector antenna at 61.2 GHz. (Data see Figure 15).

vide a polarization twisting operation, low profile, low loss, and potentially, low cost mm-wave antennas can be realized. Practical examples have been demonstrated at 28.5 GHz and 60 GHz. Some efforts have been briefly described for aking a full wave calculation of these reflectarrays; this ongoing effort will probably lead to optimization of these antennas and to further reduce their depth. REFERENCES [1] RUSSELL (M.E.) et a/., Millimeter-wave radar sensor for automotive intelligent cruise control (ICQ, IEEE Trans. Microw. Theor. Techniques Mrr-45 (1997), pp. 2444-2453. [2] MANSEN (D.), VILLINO (G.), Planar Microstrip Antennas for mmds Application at 40 GHz. 28th European Microw. Conf. 1999, Munich, Germany, Vol. Ill, pp. 9-12. [3] SEHM (T.), LEHTO (A.), A. V. RAISANEN (A. V.), A high-gain 58 GHz box-horn array antenna with suppressed grating lobes, IEEE Transactions on Antennas and Propagation, 47, n° 7, 1999, pp. 1125-1130. [4] ANDERSON (T. N.), MICHALSKI (J.), Hou (YuN-Li), A high power, high performance planar slot array antenna, Microw. Journal, May 1995, pp. 70-77. [5] ANDO (M.) et a/., Novel single-layer waveguides for high-efficiency millimeterwave arrays, IEEE Trans, on Microw. Theory Tech., MTT-46 (June 1998), pp. 792-799.

Printed millimeter-wave reflectarrays

111

[6] DOLP (R.), MAYER (W.), GRABHERR (W.), 58GHz High Gain Flat Panel Antenna for High Volume Production, 28th European Microw. Conf. 1999, Munich, Germany, Vol. Ill, pp. 12-15. [7] MITTRA (R.) et a/., Techniques for analyzing frequency selective surfaces - a review, Proceedings of the IEEE, 76 (12), pp. 1593-1615, Dec. 1988. [8] MUNSON (R.E.), HADDAD (H.A.), HANLEN (J.W.), Microstrip Reflectarray for Satellite Communication and Radar Cross-Section Enhancement or Reduction, US Patent n° 4,684,952, Sept. 24, 1982. [9] HUDER (B.), MENZEL (W.), Flat printed reflector antenna for mm-wave application, Electronics Lett. 1988, p. 318. [10] Quo (Y.J.), BARTON (S.K.), A High-Efficiency Quarter-Wave Zone Plate Reflector, IEEE Microwave and Guided Wave Letters, 2. 1992, 470-471. [11] Quo (Y. J.), BARTON (S. K.), Fresnel Zone Plate Reflector Incorporating Rings, IEEE Microw. and Guided Wave Letters, 3, 1993, 417-419. [12] JAVOR (R. D.), Wu (X.-D.), CHANG (K.), Design and performance of a microstrip flat reflectarray antenna, Microwave And Optical Technology Letters, 7, n°7, May 1994, pp. 322-324. [13] MENZEL (W.), A planar reflector antenna, MIOP 1995, Sindelfingen, Germany, pp. 608-612. [14] POZAR (D.M.), TARGONSKI (S.D.), SYRIGOS (H.D.), Design of millimeter wave microstrip reflectarrays, IEEE Trans, on Antennas and Propagation, Vol. AP-45 (1997), pp. 287-96. [15] PATEL (M.), THRAVES (J.), Design and development of a low cost, electronically steerable, X-band reflectarray using planar dipoles. Military Microwaves, 1994, pp. 174-178. [16] BRADLEY (J.), CUHACI (M.), SHAKER (J.), JETTE (S.), PETOSA (A.), A novel bifocal dualfrequency, dual orthogonal polarization planar reflector for SatCom applications, AP2000, Davos, April 2000, Session 5A2 (paper 0774). [17] ENCINAR (J.A.), ZORNOZA (J.A.), Design and Development of Multilayer Printed Reflectarrays for Dual Polarisation and Bandwidth Enhancement, AP2000, Davos, April 2000, Session 4A9 (paper 0236). [18] PILZ (D.), MENZEL (W.), Full Wave Analysis of a Planar Reflector Antenna, 1997 Asia Pacific Microwave Conf. APMC'97, Dec. 2-5, 1997, Hong Kong, pp. 225-227. [19] PILZ (D.), MENZEL (W.), A mixed integration method for the evaluation of the reaction integrals using the spectral domain method, IEEE Proc. on Microwaves, Antennas and Propagation, 146, n° 3, June 1999, pp. 214-218. [20] PILZ (D.), MENZEL (W.), Folded reflectarray antenna, Electron. Lett., 34, n° 9, April 1998, pp. 832-833. [21] PILZ (D.), MENZEL (W.), A novel linear-circular polarization converter, 28th Europ. Microw. Conf., 1998, Amsterdam, 2, pp. 18-23. [22] Menzel (W.), Pilz (D.), Leberer (R.), A 77 GHz fm/cw radar front-end with a lowprofile, low-loss printed antenna, IEEE Trans, on Microw. Theory Tech., 47, n° 12 (Dec. 1999), pp. 2237-2241.

112

Millimeter waves in communication systems

[23] MENZEL (W.), PILZ (D.), Printed quasi-optical mm-wave antennas. Millenium Conf. on Antennas and Propagation AP2000, Davos, Switzerland, 2000, Session 3A2-1 (paper n° 0023). [24] MENZEL (W.), PILZ (D.), Printed mm-wave folded reflector antennas with high gain, low loss, and low profile, IEEE Antennas and Propagation Conf. 2000, Salt Lake City, USA, July 2000, 2, pp 790-792. [25] ITOH (T.) (editor), Numerical Techniques For Microwave And Millimeter-Wave Passive Structures. John Wiley & Sons, 1989.

Chapter 6

Computer aided design for new microwave filter topologies for spatial applications in Ka band D. Baillargeat, H. Blondeaux, S. Bila, P. Leveque, S. Verdeyme and P. Guillen IRCOM - UMR CNRS 6615, University of Limoges, France

I. Introduction Nowadays, demand for filters used in wireless communication systems is increasing and provides significant challenges to microwave designers. The required characteristics of filters for such communication are small size, low insertion loss, tuning less, high selectivity, high temperature stability and an easy manufacturing [1, 2]. Moreover, in order to be connected to MMICS, the filters must be easily integrated into planar surrounding. Recent advances in microwave computer aided design (CAD), growing computer capabilities and new available tools permit designers to implant electromagnetic (EM) simulators in optimization procedures for powerful design of passive and active microwave circuits. So, by applying direct and specific EM optimization procedures which ke possible reducin the number of global analyses, it is now possible to conceive original microwave devices, as multipole filters. This article is devoted to the design of original microwave filter topology in Ka band and it is divided into two parts. The first describes an original resonant element operating around 20 GHz (Figure 1). Its main characteristics are the following: -

small size: volume < 1 cm3, high integration into a planar circuit, low losses: unloaded quality factor Qu>2000, good frequency isolation: >2 GHz on both side of the resonant frequency.

This element is used to conceive an experimental Tchebyscheff 2 pole filter without tuning. Then, applying a direct electromagnetic optimization method

114

Millimeter waves in communication systems

developed in our laboratory, we design an elliptic 4 pole filter centered at 20 GHz, with a 500 MHz bandwidth at -3dB. We note here, that the em optimization method developed combines three independent softwares [3]: a 3D electromagnetic one based on the finite element method developed in our laboratory, a specific software developed by inria and the third which establishes the objective electrical characteristics. This optimization method will be described in paragraph III.3. The second part of this article presents a new concept of a global function applied to the design of a radiant microwave filter. Indeed, in a telecommunication satellite losses are introduced by the transmission between radiant elements and filter. So the main objective of the global function developed here is to suppress these losses and to design a n pole radiant filter constituting the elementary cell of an emission/reception antenna. In this present paper, the resonant element described previously is used to conceive an opened Tchebyscheff two pole tuning less filter at 20 GHz, with a 500 MHz bandwidth at -3 dB. This device is designed applying a 3D finite element software including perfectly matched layer (PML) conditions, and presents interesting filtering performances and radiation ones.

II. Electromagnetic analysis The theoretical analysis is performed by applying the free or forced oscillations 3D finite element method in the frequency domain between the access ports of the device. The FEM is well known to be useful for solving Maxwell's equations. This method has already been explained in several papers [4] and our purpose is not to describe it here. The mathematical formulation we solve is identical to the commercial simulator HFSS. We note that the device under consideration is meshed with triangles (2D) or tetrahedrons (3D). To solve Maxwell's equations on each node of the mesh, we apply a vectorial E-formulation using mixed finite elements of Nedelec with second order polynomials. Indeed, this formalism permits us to avoid spurious responses [5]. Moreover, using our electromagnetic software, we can analyze complex voluminous or planar devices. The inner materials must be homogeneous and linear, but can be anisotropic and possess dielectric or metallic losses. The structure studied is generally closed by perfect magnetic and electrical conditions or opened considering absorbing conditions (PML, ABC). Two types of access can be considered t be the distributed ones and the localized access which allows introducion of active response of components or MMICS in passive domains. For the problem of forced oscillations imposing the frequency, we obtain the scattering matrix parameters between the access ports. For free oscillations, these ports are short circuited and the analysis gives the resonance frequency, field repartition of each eigenmode, and their unloaded Qu factor.

Computer aided design for new microwave filter topologies

115

III. Direct EM optimization method of microwave filters m.l. Resonator characteristics The resonant structure is composed of a metallized dielectric substrate enclosed in a parallelepipedic cavity (Figure 1) [1]. Thin-film metallization on the upper and lower sides of the dielectric substrate has hollow patches of the same diameter. The portion without metallization is used as a resonator. In order to reduce the device size, this resonator is usedi on its first hybrid mode [6], named HEM!, which presents two orthogonal polarizations at the same frequency. Moreover, to obtain an un-deteriorated response of the filter, we set the frequency isolation at 2 GHz on both side of the resonant frequency. The FEM has been applied to characterize the HEM! mode and to establish the optimal dimensions to obtain a resonant frequency equal to 20 GHz, a high unloaded Qu factor, the desired frequency isolation and a small sensitivity to the metallic cavity dimensions. The results are shown in Figure 1. One of the main advantages of this device is its ease of manufacture and its ability to provide high electrical performances. In fact, in Figure 1, we can observe that the unloaded quality factor (Qu = 2500) obtained by this original resonator is higher that ones obtained using planar technologies at these frequencies [7].

Figure 1. Description of the resonant element.

116

Millimeter waves in communication systems

III.2. Tchebyscheff 2-pole filter A two-pole filter based on this resonator topology has already been studied [8] at 7.5 GHz, so we propose here to validate this technology at 20 GHz. UI.2.1. External Qe factor The input/output excitation system is defined by a coplanar line on the upper surface of the substrate (Figure 2). This excitations type is easily manufactured and integrated on the resonator. The line penetration X on the substrate is longer, more important is the input/output coupling, related to the external Qe factor. Using this solution, the external Qe factor can be obtained with high precision. Figure 2 describes the external Qe factor as a function of the depth penetration X of the excitation lines. This curve is obtained by applying a forced oscillation between the ports 1 and 2. In this case, the losses are not taken into account. ni.2.2. Inter-polarization coupling technique To obtain a two-pole filter, the two polarizations of the HEM! mode must be coupled. In this study, this coupling is obtained by a resonator geometrical modification [8], which needs no tuning. So, the patches geometry becomes elliptic (Figure 3). The elliptic axis is placed at 7C/4 from the probe axis. Then, the interpolarization coupling coefficient K is defined as a function of the difference between the two elliptic radii (Rx-Ry). Using this solution, no strong discontinuities are created around the resonator and the coupling is easily manufactured with high precision.

Figure 2. External Qe factor as a function of the depth penetration line.

Computer aided design for new microwave filter topologies

117

Figure 3. Coupling coefficient as a Junction of the radius Ry.

m.2.3. Tchebyscheff 2-pole filter responses Filtering objectives of the Tchebyscheff two-pole filter to be fabricated are: Center frequency

20 GHz

-3 dB pass bandwidth

500MHz

Return loss

15 dB

Attenuation (at fo ± 1 GHz)

15 dB

The synthesis of the filter gives the following objective couplings: external quality factor Qe = 65 and K = 17.5 10"3. Two previous studies allow definition of the filter dimensions (Figure 4):

X = 1060 Mm, Rx = 3.95 mm and Ry = 3.81 mm The theoretical and experimental filter responses shown in Figure 4 permit validation of the original resonator topology and filter design. In this case, the dielectric and metallic losses of materials aren't taken into account by the EM analysis. In spite of a frequency shift, due to a too fine experimental alumina thickness compared to the theoretical one, the experimentation is in good agreement with theory (2% of error). So this experimental study allows validation of the theoretical EM analysis, the topology of the resonant element, the excitation system and the interpolarization coupling technique at 20 GHz. m.3.4-pole elliptic filter This part is devoted to the theoretical analysis of a four-pole elliptic filter design, applying a direct electromagnetic optimization method described in a previous paper [3].

Figure 4. Theoretical and experimental response of a 2-pole filter.

Computer aided design for new microwave filter topologies

119

EQ.3.1. The direct electromagnetic optimization method

Usually, applying conventional methods [9, 10], the design of a microwave filter starts with the selection of an ideal transfer function which satisfies the electrical objectives defined by the filtering pattern. The synthesis of this ideal transfer function applying filter theory, determines a theoretical coupling matrix which represents all physical couplings that have to be realized to produce the desired filtering response. Classically, each coupling which is related to an element of the filter is analyzed separately applying a segmented method. This analysis provides the initial dimensions of the structure. Generally, the first global electromagnetic analysis using these basic dimensions does not agree with the filtering pattern. An additional optimization which requires several iterations is necessary, fitting the electromagnetic response to the objective one. In our approach, we focus on reducing the number of global electromagnetic analyses, considering accurate means of error estimation between computed and desired results. Exploring this way, we have developed a procedure initialized by a classical filter synthesis which provides an objective coupling matrix. Then, after each global electromagnetic analysis the filter couplings are identified and compared to the objective ones. Couplings identification provides an accurate criterion in order to accelerate the convergence towards the desired behavior. The electromagnetic optimization method presented in this paper, the procedure may be applied to the design of a wide range of microwave filters and multiplexers. This procedure couples three independent softwares. In the first stage of the procedure, the objective coupling matrix is established from the filtering pattern, applying a synthesis software based on filter theory. This software provides objective coupling matrices, taking into account all desired electrical specifications. Considering the objective coupling matrix, the second stage is to define the basic dimensions of the filter performing a segmented electromagnetic synthesis. Coupling coefficients are estimated separately from each other, analyzing successively the structure coupling elements. This electromagnetic synthesis allows definition of tables or analytical functions relating coupling elements dimensions to the coupling coefficients values. The last stage is the optimization loop. Considering the basic dimensions determined previously, a global electromagnetic analysis of the filter is performed. We can note here that all electromagnetic analyses in the different stages of the procedure are performed applying the finite element software described previously. After the initial global electromagnetic analysis, the corresponding coupling matrix is established from the computed scattering matrix, applying hyperion, an identification software developed by INRIA [11]. This software was originally dedicated to solving mathematical problems of data rational approximation. The

120

Millimeter waves in communication systems

transfer matrix, computed in the optimization frequency domain, is approximated to a rational polynomial matrix and an extension of this software based on filters theory allows computation of the coupling matrix from the polynomial matrix poles and zeros. The extracted coupling matrix has a fixed geometry, which guarantees its unicity for a given transfer function. This unicity condition is necessary in order to relate coupling coefficients to their corresponding coupling elements dimensions. Also, if the coupling matrix does not agree with the ideal one, we proceed with the optimization loop presented in Figure 5. m.3.2. Elliptic 4-pole filter design Filtering objectives of the four-pole filter are as follows:

Figure 5. Optimization loop chart.

Computer aided design for new microwave filter topologies

As a first step, the objective coupling matrix filtering pattern:

121

is established from the

Rin = Rout = 1.03 0.000

0.81

0.000

0.353

0.81

0.000

-0.835

0.000

0.000

-0.835

0.000

0.81

0.353

0.000

0.81

0.000

[Mij]objl: ideal coupling matrix

where Rin and Rout are directly allied to the input/output coupling. The studied four pole filter topology is described in Figure 6. As shown on the two pole filter, each resonator is used on its first hybrid mode. The resonator 1 presents the polarizations named 1 and 2, and the resonator 2 the polarizations named 3 and 4. Elliptic resonator dimensions are related to the inter-polarizations coupling coefficients (M12 and M^), and irises dimensions to the inter-cavities coupling coefficients (M14 and M^3). In the second stage applying the 3D FEM, initial studies described previously establish theoretical couplings M12, M34 and Rin, Rout as functions of filter dimensions. So we still have to study inter-cavities coupling coefficient (M14 and N^) through crossed irises, and the compensation of the resonant frequency shift due to the excitation perturbation. - inter-cavities coupling technique As for the resonator, thin-film metallization on the upper and lower sides of the iris dielectric substrate has hollow patches of the same length. The portion without metallization is used to couple elements (Figure 7).

AFigure 6. Description of the 4-pole filter.

122

Millimeter waves in communication systems

Figure 7. Coupling coefficient as a function of the iris dimension. Moreover, several tests show that coupling elements are almost electrically independent. So iris dimensions, related to the inter-cavities coupling coefficients, influence only M14 and M23 [11]. The inter-cavities coupling coefficient K between two dielectric resonators is defined as a function of rectangular iris dimensions. This original coupling system needs no tuning at the operating frequencies and can be manufactured with high precision and allows a good sensitivity to mechanical tolerances. The dielectric iris substrate is TMM3 by Rogers and presents the following characteristics: e = 3.27, TG8 = 20.1
Figure 8. Frequency compensation for X = 1055 \un versus the perturbation length.

Computer aided design for new microwave filter topologies

123

of the excited polarization (polarization 1). If the perturbation is placed at 90° to the excitation axis, the resonant frequency of the un-excited polarization (polarization 2) is shifted; but not the other one. Moreover, the perturbation is placed on both side of the resonator to avoid a direct coupling with excitation lines. This frequency compensation study closes the segmentation approach which permits definition of the following initial dimensions satisfying the objective couplings: -

cavity: Lc = 8.9 mm He = 5.9 mm dielectric resonator: Dr = 7.4 mm hs = 381 jim excitation: X = 1100 |jm elliptic resonator: Rx = Dr/2 Ry = 3.56 mm compensation: perturbation length: lower cavity: Lpj = 850 |jm upper cavity: Lp2 = 850 pm - dielectric iris: width = 0.3 mm iris length coupling 14 = 4.32 mm iris length coupling 23 = 5.21 mm Finally, in the last stage, a global EM analysis is performed to establish the whole structure response taking into account these basic dimensions. Then, the first resulting matrix [Mijjj is extracted using the hyperion software. Rin = 1.56 and Rout = 1.49 -0.308

0.693

-0.003

0.321

0.693

-1.60

-0.882

-0.103

-0.003

-0.882

-0.019

0.71

0.321

-0.103

0.71

-0.189

[Mij]j: first resulting matrix

As we can see all the terms of [Mijjj do not satisfy the objectives ones, and in particular those of [Mii]j which are related to a frequency shift between the resonant frequency f; and the central frequency of the filter response. We can observe too that the [M^j coefficient is not equal to zero. So we proceed to the optimization loop to satisfy [Mij]^. As it was explained in a previous paper [11], we suppose coupling elements are almost electrically independent, so all their dimensions are corrected at the same time taking into account the coupling functions established previously. After three iterations, we define the following dimensions giving the second resulting matrix [Mij]2 and the global EM filter response (Figure 9): - cavity: Lc = 8.9 mm He = 5.9 mm - dielectric resonator: Dr = 7.4 mm hs = 381 um - excitation: X = 1055 im

124

Millimeter waves in communication systems

- elliptic resonator: Rx = Dr/2 Ry = 3.54 mm - compensation: perturbation length: lower cavity: Lpl = 545 urn upper cavity: Lp2 = 476 jam - dielectric iris: width = 0.3 mm iris length coupling 14 = 4.41 mm coupling 23 = 5.15 mm Rin = 1.13 and Rout = 1.17 -0.089

0.805

-0.001

0.385

0.805

-0.061

-0.933

-0.104

-0.001

-0.933

-0.077

0.844

0.385

-0.104

0.844

-0.092

[Mij]2: second resulting matrix

The reflection response is limited around -8 dB in bandpass, this limitation is due to [ML^]; as we can see for [Mij]2, [M^^ is not zero. This term defines a parasitic coupling between the polarizations 2 and 4 through the crossing coupling iris, due to the elliptic geometry of the dielectric resonators. As a consequence, we have to define distinctly the influence of this parasitic coupling on the filter response. Figure 10 described the filter equivalent circuit if the coupling between polarizations 2 and 4 is not taken into account The obtained response is our objective one.

Figure 9. 4-pole filter response after three iterations.

Figure 10. Filter equivalent circuit.

126

Millimeter waves in communication systems

Figure 11 describes the filter equivalent circuit with localized elements taking into account the parasitic couplings M24 and M13. To obtain this circuit, the parasitic effect M^ is symmetrically allocate among M24 and M13 and the corresponding coupling matrix is defined by the matrix [Mij]obj2 after transformation realizing M24 = M13. In this case we introduce all the ideal coupling values given by [Mij]0t,n and the parasitic effects on M^ and M13 established previously. The obtained response shows that if the couplings M24 and M13 are introduced, the filter response is extremely perturbed, and never can satisfy the objective pattern. Our purpose is now to compensate for the parasitic coupling effect to fit the EM response on the objective response. Introducing a shift between the four resonant frequencies of the filter can be a solution to this task. Then, a new objective coupling matrix [Mij]obj2 is established. Figure 11 shows that these new objective coupling matrix respects perfectly the initial filtering pattern. Rin = 1.10 and Rout = 1.13 0.000

0.81

-0.047

0.353

0.81

0.04

-0.835

-0.047

-0.047

-0.835

0.04

0.81

0.353

-0.047

0.81

0.000

[My]obj2: new objective coupling matrix

At present, the optimization procedure applying the new objective coupling matrix is being studied. After three iterations, this study establishes the following dimensions, the third resulting matrix [Mij]3 and the global EM filter response (Figure 12). -

cavity Lc = 8.9 mm He = 5.9 mm dielectric resonator: Dr = 7.4mm hs = 381 urn excitation: X = 1050 jam elliptic resonator: Rx = Dr/2 Ry = 3.54 mm compensation: perturbation length: lower cavity: Lpj = 470 jam upper cavity: Lp2 = 470 urn - dielectric iris: width = 0.3 mm iris length coupling 14 = 4.325 mm coupling 23 = 5.09 mm

Figure 11. Compensation of the parasite effect.

128

Millimeter waves in communication systems

Rin = 1.28 and Rout = 1.20

0.016

0.836

-0.050

0.352

0.836

0.109

-0.856

-0.050

-0.050

-0.856

0.109

0.859

0.352

-0.050

0.859

0.180

[Mij]3: third resulting coupling matrix

As we can see from the final four pole filter response (Figure 12), the reflection response is lower than -16 dB, so the parasitic coupling effect is correctly compensated. On the one hand, these results demonstrate the efficiency of the optimization procedure applied to design of a four-pole filter. And on the other hand, they validate the chosen solution to compensate for the parasitic effect.

IV. Radiant microwave filter Our purpose is here to conceive an opened Tchebyscheff two-pole tuning less filter using two superposed cavities coupled by a metallic iris. Filtering objectives of this filter are the following: center frequency at 20 GHz, 500 MHz 3 dB bandwidth, 15 dB return loss and 15 dB attenuation at fo ± 1 GHz. In a first step, we demonstrate the problem feasibility using two enclosed cavities. The section response obtained constitutes the opened filter objective response. And in a second step, we optimize the cavity-opened surface, taking

Figure 12. Final 4-pole filter response.

Computer aided design for new microwave filter topologies

129

into account several coupling parameters, to obtain this objective response. The cases of a square opening and a circular opening will now be studied. I V.I.a. Enclosed 2-pole filter The two-pole filter topology studied is described in Figure 13. As we can see, we use the resonator described previously. The inter-cavity coupling is obtained by a classic rectangular metallic iris. In this case, the coupling coefficient K between two dielectric resonators is defined as a function of the rectangular iris dimensions. The Tchebyscheff synthesis of the filter by an equivalent localized element circuit gives the different theoretical couplings: external quality factor Qe = 65 and inter-cavity coupling coefficient K = 17.5 10~3. Previous electromagnetic studies allow definition of the filter dimensions satisfying these objective couplings:

Qe = 65 -» X = 1050 jim K = 17.5 IQr3 -> iris length = 5.4 mm Figure 13 shows the corresponding global EM filter response. IV.1.2. Opened 2-pole filter - Filtering characteristics The opened two-pole filter topology studied is described in Figure 14. The filter is composed of two resonant elements superposed and coupled by metallic iris. The upper cavity is constituted of a resonant element without excitation and an opening surface. The lower cavity is constituted of a resonant element excited by a coplanar line. The global EM necessitates surrounding the opening surface with PML conditions.

Figure 13. Enclosed 2-pole filter.

130

Millimeter waves in communication systems

Figure 14. Opened 2-pole filter.

Four parameters are still used to optimize the filter reflection response: the penetration line X on the lower substrate, the iris length, the Hsup dimension and the opening surface Sop. Each parameter is subject to a particular study to quantify its influence on the reflection response. To characterize the radiant element, two studies are combined. The first one, applying the FEM to an opened one-pole filter, determinates the resonant frequency shift for each opening surface, and the second one, using an equivalent circuit of the one-pole filter, determinates the radiation resistance (Rr) for each opening surface (Figure 15). The Hsup parameter is used to compensate the frequency shift of the resonant frequency due to the opening. The synthesis of the opened two-pole filter by an equivalent circuit gives the different theoretical couplings: external quality factor Qe = 57, intercavities coupling coefficient K = 17.5 10~3 and radiation resistance Rr = 80 Q. And the previous studies define the filter dimensions satisfying the objective couplings: Qe = 57 -> X = 1080 jim K = 17.5 10~3 -> iris length = 5.4 mm Rr = 80 Q -* Sop = 22.5 mm2 => Ssquare = 4.75 x 4.75 (mm2) Scircular = JC x (2.68)2 (mm2)

Computer aided design for new microwave filter topologies

131

Figure 15. Radiation resistance variations.

Figure 16 compares the two opened reflection responses with the enclosed one. It shows that filtering objectives are respected. So an opened filter presenting a really good reflection response has been dimensioned. Our purpose is now to study its radiation pattern.

Figure 16. Opened 2-pole filter reflection response.

132

Millimeter waves in communication systems

- Radiation patterns Defining a Huyghens surface surrounding the opening surface, we extract the electromagnetic field tangential to it. And then the radiation integrals allow extraction of the radiation pattern [12]. Figure 17 shows the Ex modulus component of the electrical field on the Huyghens surface above the circular opening and the radiation pattern obtained in j planes. It is in agreement with required ones for such applications. The results are similar for the two openings.

V. Conclusion An original technology of the resonator has been presented in this paper to realize K-band filters, whose volume is < 1 cm3. It allows high integration into planar circuit. Moreover, the frequency compensation, the excitation and coupling techniques are realized in the same manufacturing step as the resonator and need no tuning. A Tchebyscheff two-pole filter around 20 GHz with 2.5% of -3 dB pass bandwidth has been designed and realized. The experimentation is in good agreement with theory, and allows the validation of this technology at these frequencies. An elliptic four-pole filter, using two resonant elements superposed and coupled with a dielectric iris, at 20 GHz with 2.5% of -3 dB pass bandwidth has been simulated. Some original technologies of coupling iris and of frequency compensation, which can be manufactured with high precision, have been presented. A direct electromagnetic optimization method based on three independent softwares developed in our laboratory has been applied to design multipole filters. The procedure is initialized by a classical filter synthesis, based on a segmented electromagnetic synthesis which provides the basic dimensions of the structure. Finally, the optimization loop which combines a global electromagnetic analysis and a coupling identification, improves the structure response compared to an empirical optimization. Using this method, some parasitic effects between two polarizations disturbing the filter's response have been shown and compensated. In order to compensate for the influence of those parasitic effects, a new equivalent circuit of the filter has been defined. Then a new objective coupling matrix satisfying perfectly the initial filtering pattern is established. And, to satisfy correctly this new objective coupling matrix, a new optimization procedure validates the compensation solution and demonstrates the efficiency of our procedure applied to design a four-pole filter. At last, a Tchebyscheff radiant two-pole filter at 20 GHz with 2.5% of -3 dB pass bandwidth has been designed for the first time. A specific study of the radiant element hs been used to determine the radiation resistance and the resonant frequency shift due to each opening surface. This study, combined with a filter synthesis definition of a square and a circular opening giving a satisfactory reflection response. By extraction of the electromagnetic field on a Huyghens surface, we have established the radiation pattern.

Figure 17. Electrical field repartition and radiation pattern.

134

Millimeter waves in communication systems

Acknowledgements The authors would like to acknowledge Alcatel Space Industries for their financial support, the CNES and the INRIA for using Hyperion software. REFERENCES [1] ISHIKAWA (Y), HIRATSUKA fr), YAMASHiTA (S.), ho (K.), "Planar type dielectric resonator filter at millimeter-wave frequency", IEICE Trans. £/ectron.,Vol. E79-C, 5, May 1996, pp. 679-684. [2] HIRATSUKA (T.), SONODA (T.), MIKAMI (S.), SAKAMOTO (K.),TAKIMOTO (Y), "A Ka-Band Diplexer Using Planar TE Mode Dielectric Resonators with Plastic Package", 29th European Microwave. Conference, Munich, Oct 1999. [3] BILA (S.), BAILLARGEAT (D.), AUBOURG (M.), VERDEYME (S.), GUILLON (P.), ZANCHI (C.), SOMBRIN (J.), GRIMM (J.), BARATCHART (L.), "Direct Electronic Optimization Method for Microwave Filter Design", Electronics Letters, 35, n° 5, pp. 400-401, March 1999. [4] AUBOURG (M.), GUILLON (P.), "A miwed finite formulation of microwave devices problems: Application to MIS structure", J. Electromagn. Waves Applications, 5, n°415, pp. 371-386, 1991. [5] NEDELEC (J.C.), "A new family of mixed elements on R3", Numer. Math., 50, pp. 57-87, 1986. [6] ATIA. (A.E), WILLIAMS (A.E), "New types of waveguides bandpass filters satellite transponders", COMSAT Tech. Rev, 1971,1, n° 1, pp. 21-43. [7] ROUCHAUD (R), MADRANGEAS (V.), AUBOURG (M.), GUILLON (P.), THERON (B.), MAIGNAN (M.), "New Classes of Microstrips Resonators for HTS Microwave Filter Applications", IEEE MTT-S, Baltimore, June 1998. [8] MORAUD (S.), VERDEYME (S.), GUILLON (P.), ULIAN (P.), THERON (B.), "A new planar type dielectric resonator for microwave filtering", IEEE MTT-S, Baltimore, June 1998. [9] ALOS (J.T.), GUGLIELMI (M.), "Simple and effective EM-based optimization procedure for microwave filters", IEEE Transactions on MTT, 45, n° 6, June 1997, pp. 856-858. [10] ATIA (A.E.), WILLIAMS (A.E.), "Narrow-bandpass waveguide filters", IEEE Transactions on MTT, 20, n° 4, April 1972, pp. 258-265. [11] BAILLARGEAT (D.), MORAUD (S.), BLONDEAUX (H.), BILA (S.), VERDEYME (S.), GUILLON (P.), "Design of filters for space applications", Workshop MTT "Filter for the masses", Anaheim, CA, June 13-19 1999. [12] HARRINGTON (R.F.), "Time Harmonic Electromagnetic Field", McGraw-Hill, New York

1961, p. 34.

Chapter 7

A simple way to design complex metallic photonic band-gap structures G. Poilasne EE Department, UCLA, California, USA

P. Pouliguen and C. Terret Laboratoire Antennes & Telecommunications, Universite de Rennes I, France

P. Gelin LEST, ENST-Bretagne, France

L Desclos NEC C&C Research Laboratory, Princeton, New Jersey, USA

I. Introduction Photonic band-gap materials (PBG) are periodic structures composed of dielectric or metallic material. The different periodicities and contrasts between the material properties give them some interesting behaviors similar to electron propagation in semi-conductors. They exhibit frequency bands inside which no propagation mode exist [1]. Between these band-gaps are propagation bands with different corresponding modes. These structures are very attractive for antenna applications as they can enhance the gain or shape the antenna radiation pattern [2, 3]. For structure initially composed of dielectric material only, it has been shown that adding metallic wires can enlarge the band-gap [4]. These so called metallo-dielectric PBG are a very interesting evolution of PBG since they offer the possibilities to improve the basic behavior of all dielectric PBG. Unfortunately, their design is still difficult, especially for multiple frequency applications which maybe needed in future multi-mode communications systems [5]. It is then important to clarify the mechanisms involved and create useful tools to design such structures. We have investigated the behavior of complex structures obtained by merging two different PBG. This behavior has been compared to the response of both the structures when they are isolated. It has

136

Millimeter waves in communication systems

lead to establishment of laws in order to design complex and multiple frequency structures in a simple way. In Section II, the design of a mixed MPBG - Bragg mirror structure is presented. This design is performed taking into account the effective medium, which should allow suppression of the first propagation band. It is shown that the control of propagation bands is not so obvious and propagation peaks still remain. It then demonstrates that a simple method can be applied to determine the behavior of mixed structures. In Section ffl, numerical results are presented on the mixing of PEG structures. These results allow us to draw up simple design empirical laws which are helpful for the design of complicated structures. Finally, these results are all the more important because they allow us to take into account realization parameters in order to manufactured PBG structures. Numerical results presented in this communication have been obtained with Ansoft HFSS and HP HFSS, commercial software based on the finite element method.

II. Mixed MPBG and Bragg mirror structure Metallic photonic band-gap materials and Bragg mirrors exhibit dual behaviors, as presented in Figure 1. It has already been demonstrated that cascading different structures allow suppression of propagation modes [6]. However, we can wonder if this behavior is also obtained when two structures are mixed as presented in Figure 2. To design it, basic equations have been used in order to take into account the electrical length when the MPBG and the Bragg mirror are mixed. The authors have demonstrated that the end of the first band-gap of a MPBG composed of parallel wires and excited by a wave with the E field parallel to the wires is approximately defined by: X,c = 3.2x (period-diameter) where Xc is the cutoff wavelength of the band-gap, period and diameter are respectively the period of the MPBG and the diameter of the wires. In the case of the mixed structure, the period along the depth of the MPBG is to be defined as an effective one. If we consider the case of a Bragg mirror composed of air and a dielectric material, the effective period is defined as:

periodeff = thick^ + n^. thick^ where periodeff is the effective period of the MPBG or the electrical length of a layer of the Bragg mirror, thick^ is the thickness of the air layer, and n^ and thick^j are respectively the index and the thickness of the dielectric material layer. The last equation required is well known and can be used to calculate the band-gap position of the Bragg mirror: W4 = thickair = ndiel •thickdiel where X.^^ is the central wavelength of the band-gap and the other parameters are the same as defined previously.

Figure 1. Reflection coefficient of a MPBG composed of metallic wires a) and reflection coefficient of a Bragg Mirror b).

138

Millimeter waves in communication systems

Figure 2. Composition of the mixed structure.

A first approach in designing mixed structures is the following. For application and measurement purpose facilities, the targeted frequency range is centered on 10 GHz. Then, choosing the parameters from the Bragg mirror allow calculation of the effective period of the MPBG. The wire diameter is used to fit the edge of the 1st MPBG band-gap to the beginning of the Bragg mirror gap. As shown in Figure 2, a set of parameters can be: - air thickness:

6.6 mm

- dielectric thickness:

3.3 mm

- dielectric index:

2

- effective period:

13.3 mm

- wire diameter:

4 mm

Figure 3-a) presents simulated results for isolated structures. The design corresponds to the desired behavior which was to superimpose the propagation band of the MPBG and the band-gap of the Bragg mirror. Results obtained with the mixed structure presented in Figure 3-b) shows that the propagation band is not completely suppressed. In order to validate it, a structure has been manufactured and plane wave measurements have been performed using a radar cross section measurement technique. Results are presented in Figure 4. The same propagation peaks appear nearly on both sides of the Bragg mirror alone band-gap. The difference of the propagation peak levels between the simulated and experimental results comes

Figure 3. Numerical results on mixed structure, reflection coefficient of a) MPBG alone and Bragg mirror alone, b) mixed structure.

140

Millimeter waves in communication systems

Figure 4. Experimental results in the mixed structure, comparison with the reflection coefficient of the Bragg mirror alone.

from the fact that in the simulation, it is a real plane wave exciting an infinite structure whereas in the experimentation the structure is finite. Measurements of the MPBG alone are not presented here, as they do not correspond to a 13.3 by 13.3 mm period MPBG but to a 13.3 by 9.9 mm period MPBG. Part of the following studsets out to explain why there are peaks. The observation of the reflection coefficient for the other polarization clarifies the situation. For this polarization, the MPBG do no longer exhibit a band-gap starting from 0 Hz until the cutoff frequency presented previously. The behavior is similar to the Bragg Mirror but the band-gap is frequency shifted. Figure 5 presents the reflection coefficient of the Bragg mirror alone, the MPBG alone and the mixed structures. The metallic structure alone would exhibit a band-gap around 15 GHz even if more layers would be needed to obtain a reflection coefficient equal to 0 dB. It appears that when the Bragg mirror band-gap and the MPBG band-gap are superimposed, the band-gap of the mixed structure disappears. The following section is devoted to the study of this phenomenon.

in. Mixed structure behaviours The measurement of the mixed Bragg mirror - MPBG structure tends to prove that when two structures with the same period exhibit a band-gap with the same

Complex metallic photonic band-gap structures

141

Figure 5. Reflection coefficient of the Bragg mirror alone, the MPBG alone and the mixed structure for an orthogonal polarization.

frequency range, the structure composed of the merger of these two structures does not exhibit a band-gap in accordance. In order check it empirically, some other mixed structures have been tested. They are based on a 2D MPBG as previously and on a 3D MPBG composed of discontinuous metallic wires. The behavior of discontinuous wire MPBG looks like the behavior of dielectric PBG. The authors have also studied the influence of the length of the wires on the reflection coefficient. Figure 6 presents this coefficient for 12 mm period MPBG with different wire length varying from 6mm to 11.5mm. The period along the wires is 12 mm too. Then, different structures have been mixed with the 2D one. The first discontinuous structure used has a 9 mm-wire length. In this case, the first part of the band of analysis corresponds to a propagation band of both isolated structures. In this case, the mixed structure exhibits a propagation band (Figure 7). Then, from 9 to 12.5 GHz, the discontinuous structure has a bandgap and the other still has a propagation band. The mixed ones now exhibit a band-gap. For higher frequencies, it is the other way round for the isolated structures and the mixed one still exhibits a band-gap. The second discontinuous structure we can consider is composed of 10.5 mm long wires. In this case, as presented in Figure 8, for frequencies greater than 14 GHz, both isolated structures exhibit a band-gap. And as obtained experimentally, the mixed structure does not suppress the propagation of waves anymore.

142

Millimeter waves in communication systems

Figure 6. Transmission coefficient of different discontinuous structures function of the length of the wires.

Figure 7. Comparison between the transmission coefficient of the continuous structure alone, the discontinuous one alone (wire length 9 mm) and the mixed one.

Complex metallic photonic band-gap structures

143

Frequency (GHz) Figure 8. Comparison between the transmission coefficient of the continuous structure alone, the discontinuous one alone (wire length 10.5 mm) and the mixed one.

Different other examples have been simulated and they all tend to prove the following empirical laws: - If both isolated structures exhibit a propagation band, the mixed structure exhibits a propagation band, - If one of the isolated structures exhibits a band-gap and the other exhibits a propagation band, the mixed structure exhibits a band-gap, - If both isolated structures exhibit a band-gap, results tend to prove that the mixed structure exhibits a propagation band. These empirical laws are just based on numerical and experimental results. No theoretical work has supported them until now. But they are already enough interesting to point out. One can easily imagine that if one needs a structure within two different frequency ranges, two different structures with the same period and exhibiting a band-gap for both bands can be designed. The mixed structure obtained by merging both MPBG will exhibits a band-gap for each frequency range, as long as the band-gaps of the isolated structures do not overlap. These laws can also be useful for design of MPBG when they are manufactured as described in the following section.

144

Millimeter waves in communication systems

IV. Manufacturing MPBG One solution to manufactured mpbg at low cost consists in printing the desired pattern on different substrate layers and to insert foam layers between these substrates (Figure 9). Glue films are also inserted in order to obtain good rigidity of the structure after having heated and applied pressure to the structure. The thickness is controlled using aluminium stops.

Figure 9. Sandwich structure inside the press in order to obtain the desired thickness.

Different problems must be taken into account when designing structures for such a technology. The first one is due to the fact that when the foam is compressed, the value of the dielectric constant changes. A parametrical study has shown that the evolution of the dielectric constant is a linear function of the compression rate (Figure 10). The second one is due to the substrates on which are printed the patterns. The dielectric constant of such substrate is often larger than the dielectric constant of the foam. To a first approximation, we can consider that the effective dielectric constant is the average of the dielectric constant of the foam and of the substrate.

mousse ' mousse mousse

substrat' substrat substrat

But actually, the succession of foam and substrate is a Bragg mirror. And when the patterns are printed on the substrate, the structure consists of the merge between a Bragg mirror and a MPBG. Therefore, attention must be paid to the design. The Bragg mirror must be analyzed alone and it must be checked that the working band of the MPBG corresponds to a propagation band of the Bragg mirror. Otherwise, propagation band and band-gap of the MPBG may be suppressed.

Figure 10. Dielectric constant versus compression rate for two different initial foam thicknesses.

V. Conclusion A simple design method has been presented in order to realize multiple frequency MPBG. In this method, basic laws allow determination of the behavior of a complex structure composed of two merged structures. When both isolated structures exhibit a propagation band, the mixed structure exhibits a propagation band. When one of the isolated structure exhibits a propagation band and the other exhibits a band-gap, the mixed one exhibits a band-gap. Finally, when both structures exhibit a band-gap, the mixed one exhibits a propagation band. Therefore, one structure can be designed to work within a frequency range and the other one in a next frequency range. The mixed structure will work within both bands so long as band-gaps do not overlap. Acknowledgements This work has been done within the framework of the GdR "Cristaux Photoniques et Microcavites". The authors would like to acknowledge the CRI of the Universitd de Rennes 1, the CELAR and the CNRS, especially the IDRIS for the use of their computers.

146

Millimeter waves in communication systems

REFERENCES [1] YABLANOVITCH (E.), "Inhibited Spontaneous Emission in Solid State Physics and Electronics", Physical Review Letters, 58, n° 20, 1987, pp. 2059-2062. [2] BROWN (E.R.), PARKER (C.D.), YABLONOVITCH (E.), "Radiation properties of a planar antenna on a photonic crystal substrate". Journal of Opt. Am. Soc., 10, n° 2, Feb. 1993, pp. 404-407. [3]

POILASNE (G.), POULIGUEN (P.), MAHDJOUBI (K.)f TERRET (C), GELIN (Ph.), DESCLOS (L.),

"Experimental radiation pattern of dipole inside metallic photonic band-gap materials", Microwave and Optical Technology Letters. 22, issue 1, July 1999. [4] CHONGJUN JIN, BINGYING CHENG, BAOYUAN MAN, DAOZHONG ZHANG, "Two-dimesional metallodielectric photonic crystral with a large band-gap", Applied Physics Letters, 75, n° 9, 30 August 1999, pp. 1201-1203. [5] DESCLOS (L.), MADIHIAN (M.), FLOC'H (J.M.), "Double frequency patch antennas for multimode applications", IEEE-APS, Montreal, Canada, July 1997. [6] AGI (K.), BROWN (E.R.), DILL III (C.), MALLOY (K.J.), "Design of ultrawideband photonic crystrals for broadband antenna applications". Electronics Letters, 30, n°25, Dec. 1994.

Index

m-V HBT 5 base and collector optimization 6 circuits, various 16 et seq collector-up topology 9 electrical properties of devices 14 emitter-up topology 8 geometry 7

antennas 94 dual polarization 97 folded reflector 102 offset-fed printed reflectarray 97 printed folded 105 structure calculation 95 baseband-over-fiber

microwave filters optimization method 115etseq resonator structure 115 CPW technology, limitations 77 dielectric bridges, thick-film multilayer technology 70 compensating 73 efficiency 75 parasitic influences 72 wideband band-pass filter, application 75 electroabsorption modulators 51 electron gas field effect transistor 20 electro-optic modulators 52

50

collector materials, electrical properties 5 common emitter breakdown voltage 4 complex metallic photonic band-gap structure, design of 135 et seq manufacture 144 MPBG and Bragg mirror structure, mixed 136 behaviours 140 composite ceramic/foam substrate, use of 87 computer aided design (CAD) for microwave filter topologies 113 et seq electromagnetic analysis 114

fiber chromatic dispersion 48 fiber-radio access networks 44 et seq optical network architectures 44 signal transport schemes 44 RF-over-fiber 44 transmission data 49 full-duplex fiber-radio systems 56-57 architecture 58 ring 59 heterojunction bipolar transistors etseq heterojunction characteristics 3 heterojunction field effect transistor high electron mobility transistor (HEMT) 1

148

Index

cross-section 26 microwave performance 34 low-noise 36 millimeter wave applications 20 EC sea parameters 21 et seq low-noise 27 power 28 power performance 38 quality factors 23 cut-off frequency 23 structures 21 metamorphic 31 for millimeter application 29 types in use 20 high speed optical detectors 53 high speed optical modulators 51 hybrid 3D integrated circuits 68 et seq benefits 79 uniplanar technology 70 et seq IF-over-fiber 48 integrated air bridges 69 ion implantation 12 Ka band, CAD for new microwave filter topologies 113 et seq 4-pole elliptical filter 117 elliptical 4-pole filter design 120 frequency compensation 122 inter-cavities coupling technique 121 parasitic effect, compensation for 127 radiant microwave filter 128 enclosed 2-pole filter 129 resonator characteristics 115 Tchebyscheff 2-pole filter 116 lateral over-etching 11 local area networks vii local multipoint distribution systems vii

membrane technology 90 metal-semiconductor field effect transistor 1 mm-wave optoelectronic devices 51 et seq modulation, direct and external 46 modulation doped field effect transistor 20 monolithic microwave integrated circuits viii, 69 optical detectors, high-speed 53 optical fiber networks vii optical signal transport schemes 45 optoelectronic/electronic device integration 54 polarization circular 105 twisting 105 reflectarrays viii focussing and twisting reflector 106 printed millimeter wave 94 et seq thick-film multilayer technology 70 thin-film micro-strip viii thin-film micro-strip lines 80 access to 81 efficient passive functions 81 and flip-chip components 85 and miniaturization 86 MMICs compatibility with 85 substrates for HBT fabrication 2 transferred 13 uniplanar technology 30 3D integration technique 70 waveguides, constraints on use 68 wavelength division multiplexing viii