You are on page 1of 705

Handbook of RF and Microwave Power Ampliers

Whether youareanRF transistor designer, anamplier designer, or asystemdesigner,


thisisyour one-stopguidetoRF andmicrowavetransistor power ampliers. A teamof
expert authorsbringsyouup-to-speedoneverytopic, including:
r
devices(Si LDMOSandVDMOS, GaAsFETs, GaNHEMTs);
r
circuit andamplier design(discrete, hybridandmonolithic);
r
CAD;
r
thermal design;
r
reliability;
r
systemapplications/requirementsfor RF andmicrowavetransistor ampliers;
r
amplier measurements.
Covering state-of-the-art developments, and emphasizing practical communications
applications, thisisyour completeprofessional referenceonthesubject.
John Walker is currently European Sales Manager at Integra Technologies, Inc. He
received his Ph.D. fromthe University of Leeds in 1976 and has since held various
industry positions, including MicrowaveHybrids Manager at Thorn-EMI Electronics
andRF DivisionManager at Semelab. HeistheEditor andCoauthor of thebooksHigh
Power GaAs FET Ampliers andClassic Works inRF Engineering. Heis aFellowof
theIEE.
TheCambridgeRFandMicrowaveEngineeringSeries
SeriesEditor
SteveC. Cripps, DistinguishedResearchProfessor, Cardiff University
Peter Aaen, J aimePl aandJ ohnWood, ModelingandCharacterizationof RF and
MicrowavePower FETs
DominiqueSchreurs, M airtnODroma, AnthonyA. Goacher, andMichael Gadringer,
RF Amplier Behavioral Modeling
FanYangandYahyaRahmat-Samii, ElectromagneticBandGapStructuresinAntenna
Engineering
EnricoRubiola, PhaseNoiseandFrequencyStabilityinOscillators
Earl McCune, Practical Digital WirelessSignals
StepanLucyszyn. AdvancedRF MEMS
PatrickRoblin, Nonlinear FRCircuitsandtheLarge-Signal NetworkAnalyzer
MatthiasRudolph, ChristianFager, andDavidE. Root, Nonlinear Transistor Model
Parameter ExtractionTechniques
Forthcoming
SorinVoinigescu, High-FrequencyIntegratedCircuits
DavidE. Root, J asonHorn, andJ anVerspecht, X-Parameters
RichardCarter, TheoryandDesignof MicrowaveTubes
Anh-VuH. Pham, MorganJ. Chen, andKuniaAihara, LCP for MicrowavePackages
andModules
NunoBorgesCarvalhoandDominiqueScheurs, MicrowaveandWireless
Measurement Techniques
Handbook of RF and Microwave
Power Ampliers
Edited by
JOHN WALKER
Integra Technologies, Inc.
CAMBRI DGE UNI VERSI TY PRESS
Cambridge, NewYork, Melbourne, Madrid, CapeTown,
Singapore, S aoPaulo, Delhi, Tokyo, MexicoCity
CambridgeUniversityPress
TheEdinburghBuilding, CambridgeCB28RU, UK
PublishedintheUnitedStatesof AmericabyCambridgeUniversityPress, NewYork
www.cambridge.org
Informationonthistitle: www.cambridge.org/9780521760102
C _
CambridgeUniversityPress2012
Thispublicationisincopyright. Subject tostatutoryexception
andtotheprovisionsof relevant collectivelicensingagreements,
noreproductionof anypart maytakeplacewithout thewritten
permissionof CambridgeUniversityPress.
First published2012
PrintedintheUnitedKingdomat theUniversityPress, Cambridge
Acatalogrecordfor thispublicationisavailablefromtheBritishLibrary
ISBN978-0-521-76010-2Hardback
Thetechnical descriptionsandproceduresinthisbookhavebeendevelopedwiththe
greatest of care; however, theyareprovidedasis, without warrantyof anykind. The
author andpublisher of thebookmakenowarranties, expressedor implied, that the
equations, programs, andproceduresinthisbookarefreeof error, or areconsistent
withanyparticular standardof merchantability, or will meet your requirementsfor
anyparticular application. Theyshouldnot berelieduponfor solvingaproblem
whoseincorrect solutioncouldresult ininjurytoapersonor lossof property.
CambridgeUniversityPresshasnoresponsibilityfor thepersistenceor
accuracyof URLsfor external or third-partyinternet websitesreferredto
inthispublication, anddoesnot guaranteethat anycontent onsuch
websitesis, or will remain, accurateor appropriate.
Contents
List of contributors pagexiv
Preface xv
1 Silicon LDMOS and VDMOS transistors: physics, design, and technology 1
Wayne Burger and Christopher P. Dragon
1.1 Technologyoverview 1
1.1.1 Introduction/history 1
1.2 LDMOSandVDMOSconstruction 2
1.2.1 LDMOS 2
1.2.2 VDMOS 8
1.3 Devicephysics 10
1.3.1 Current transport 10
1.3.2 Behavior of parasiticelements/models 12
1.3.3 BV
DSS
, R
DSon
, HCI boundaries 17
1.3.4 Snapback/ruggedness 22
1.3.5 Operatingvoltageconsiderations 26
1.4 Design/layout 27
1.4.1 Top-downnger layout 27
1.4.2 Bondpadmanifolds 29
1.4.3 Metal design electromigration 30
1.4.4 Thermal 32
1.4.5 Operatingvoltageconsiderations 34
1.4.6 Frequencyconsiderations: gatelength, gatewidth, resistors 36
1.4.7 HVICs 37
References 39
2 GaAs FETs physics, design, and models 42
Rob Davis
2.1 Introduction 42
2.1.1 Propertiesof GaAsandrelatedcompounds 43
2.1.2 TheSchottkybarrier gateandtheMESFET 45
2.1.3 ThePf
2
limit 45
2.1.4 Typesof GaAsFET 46
vi Contents
2.2 Power devicephysics 51
2.2.1 ThedeviceIVcharacteristicandloadline 51
2.2.2 ThedynamicIVcharacteristic 53
2.2.3 Theconsequencesof trappingeffects 54
2.2.4 Devicebreakdown 57
2.2.5 Breakdownmechanismsandoptimisation 58
2.2.6 CommentsonGaAsFET breakdownratings 59
2.2.7 TheFET equivalent circuit 60
2.2.8 Devicegainandguresof merit 61
2.3 Devicedesign 63
2.3.1 Power devicedesign 63
2.3.2 FET channel andrecessdesign 63
2.3.3 Power cell design 67
2.3.4 Power cell combination 71
2.3.5 Thermal design 72
2.4 Devicefabrication 74
2.4.1 Overview 74
2.4.2 Keyprocesssteps 75
2.4.3 Low-cost GaAsdevicefabrication 81
2.4.4 Packaging 81
2.5 Models 84
2.5.1 Devicemodels 84
2.5.2 Small-signal models 84
2.5.3 Largesignal models 85
2.5.4 Load-pull 89
2.6 Concludingremarks 90
References 91
3 Wide band gap transistors SiC and GaN physics, design
and models 103
Robert J. Trew
3.1 Introduction 103
3.2 Background 105
3.2.1 SiC transistors 106
3.2.2 AlGaN/GaN transistors 108
3.3 Material parameters 111
3.4 Transistor amplier operatingprinciples 115
3.5 DevicedesignandRF performance 118
3.5.1 4H-SiC MESFET amplier 120
3.5.2 AlGaN/GaN HFET amplier 123
3.6 Transistor DC andlarge-signal RF models 125
3.6.1 Equivalent circuit transistor models 125
3.6.2 Physics-basedlarge-signal transistor models 128
Contents vii
3.7 Large-signal effects 130
3.7.1 Spacechargelimitedcurrent transport 130
3.7.2 Nonlinear sourceanddrainresistance 133
3.7.3 Gateleakage 144
3.7.4 Reliabilityandtime-dependent performancedegradation 146
3.8 Summary 152
References 153
4 Amplier classes, AS 159
Steve C. Cripps
4.1 Introduction 159
4.2 Activedevicemodels 161
4.3 ClassA 162
4.4 ClassAB andClassB 164
4.5 ClassC 171
4.6 ClassF 173
4.7 ClassJ 176
4.8 Invertedmodes, invertedClassF 179
4.9 ClassE 181
4.10 ClassS 183
4.11 Multimodes 184
4.12 Conclusions 186
References 186
5 Computer-aided design of power ampliers 188
Stephen Maas
5.1 Introduction 188
5.2 Methodsof analysis 188
5.2.1 Linear analysis 188
5.2.2 Harmonic-balanceanalysis 193
5.2.3 Time-domainanalysis 202
5.2.4 Applicationsof analytical methods 205
5.3 Passivecircuit structuresandsimulationaccuracy 205
5.3.1 Scatteringparameter models 206
5.3.2 Closed-formmodels 208
5.3.3 ModelsfromEM simulation 210
5.3.4 Databasemodels 212
5.3.5 Parasiticextraction 212
5.4 Solid-statedevicemodels 213
5.4.1 Power devicemodels 213
5.4.2 Modelingcell interconnectionsinlargedevices 213
5.4.3 Thermal effectsindevicemodels 214
viii Contents
5.5 Special aspectsof power-amplier modeling 216
5.5.1 Lossincircuit metalizations 217
5.5.2 Lossincircuit components 219
5.5.3 Bondwires 219
5.6 Practical aspectsof nonlinear circuit simulation 221
5.6.1 Convergencedifculties 221
5.6.2 SPICE modelsinharmonic-balanceanalysis 226
5.6.3 Problemsizeminimizationandsolutionoptimization 226
5.6.4 Numerical considerations 227
5.6.5 Designow 228
References 230
6 Practical HF/VHF/UHF RF power amplier realization 232
Daniel P. Myer
6.1 Introduction 232
6.2 RF power amplier markets 232
6.3 Therealizationprocess 233
6.3.1 RFPA qualitativespecicationdelineation 234
6.3.2 RFPA specications, genericlist andquanticationguidelines 236
6.3.3 Specication/hardwarerealization 241
6.4 RFPA systemlevel designoverview 242
6.4.1 RF power amplier moduledesignoverview 243
6.4.2 RF power transistor deviceselectionprocessguidelines 246
6.4.3 RF power transistor bias/thermal trackingnetworks 249
6.4.4 RF input/output coupling/decouplingnetworks 250
6.4.5 Power transistor impedancematching 250
6.4.6 Feedbacknetworks 251
6.4.7 Thermal management 251
6.5 Hypothetical amplier designexample 252
6.5.1 Hypothetical applicationexampleoverview 252
6.5.2 Amplier qualitativespecicationdelineation 252
6.5.3 Amplier specicationquantication 253
6.5.4 Amplier hardwaredesign/realization 254
6.6.5 RF transistor selection 255
6.5.6 Gatebias/temperaturetracking/compensationnetwork 257
6.5.7 Input/output RF/DC coupling/decouplingnetworks 259
6.5.8 Input/output impedancematchingnetworks 259
6.5.9 Feedbacknetwork 267
6.5.10 Test setupconguration/analysis 268
6.5.11 Physical RFPA moduleconstruction 271
6.5.12 RFPA moduletest results 273
6.5.13 Beyondthetest data 281
References 283
Contents ix
7 Microwave hybrid amplier realization 284
Dominic FitzPatrick
7.1 Introduction 284
7.2 Printedcircuit boards 285
7.3 Housing 293
7.3.1 Materials 294
7.3.2 Sealingandhermeticity 294
7.3.3 Construction 299
7.3.4 Thermal issuesandheat sinking 305
7.3.5 RF connections 311
7.4 Components 315
7.4.1 Passive lumpedcomponents 315
7.4.2 Passive distributedcomponents 323
7.4.3 Transistors 331
7.5 Amplier design 333
7.5.1 Topologies 333
7.5.2 Matchingandstability 336
7.5.3 Internallymatcheddeviceampliers 343
7.5.4 Combining 344
7.5.5 Modulesize/systemintegration 344
7.6 Biasingandcontrol 345
7.6.1 Control andinterfacing 352
7.7 Tuningtechniques 353
References 355
8 Monolithic power ampliers 357
Inder J. Bahl
8.1 Overviewof MMIC power ampliers 357
8.1.1 Brief historyof MMIC power ampliers 357
8.1.2 Advantagesof monolithicpower ampliers 358
8.2 MonolithicIC technology 359
8.2.1 MMIC fabrication 360
8.2.2 MMIC substrates 361
8.2.3 MMIC activedevices 361
8.2.4 MMIC matchingelements 362
8.3 MMIC designmethodology 370
8.3.1 CADtools 370
8.3.2 Designprocedure 371
8.3.3 EM simulators 372
8.4 MMIC PA summaryandexamples 372
8.4.1 Narrowbandpower amplier 374
8.4.2 Broadbandpower ampliers 376
8.4.3 Ultrabroadbandpower ampliers 377
8.4.4 High-power ampliers 381
x Contents
8.4.5 Millimeter-wave2.4WPA 386
8.4.6 Wireless3Wpower amplier 386
8.4.7 High-voltagemonolithicPAs 387
8.5 Packagingof MMIC PAs 389
8.5.1 Ceramicpackages 390
8.5.2 Plasticpackages 394
8.5.3 Packageassembly 396
8.6 MMIC power amplier characterization 401
References 406
9 RF power amplier thermal design 411
Mali Mahalingam
9.1 Whythermal designdeservescareful attention? 411
9.2 RFPA thermal design basics 413
9.2.1 RFPA thermal designinatypical portableproduct 413
9.2.2 RFPA thermal designinatypical radiobasestation 416
9.2.3 Basicheat transfer processesandtheir roleinanRFPA thermal
performance 419
9.3 Thermo-physical propertiesof materialsinanRFPA 423
9.4 Toolstocharacterizeandpredict thethermal performanceof RFPAs 427
9.5 RFPA thermal designandmanagement advanced 432
9.6 RFPA thermal design trendsandprognostication 440
References 442
10 Reliability 446
Bill Roesch
10.1 Introduction 446
10.2 Vocabularyanddenitions(units, goals, andstrategy) 447
10.2.1 Reliabilitygoals 448
10.2.2 Semiconductor reliabilitystrategy 448
10.3 Failurecriteria 449
10.4 Failuremodes 450
10.5 Failuremechanisms 451
10.5.1 Metalization 451
10.5.2 Dielectric 453
10.5.3 Bulksubstratematerial 454
10.5.4 SchottkygateFET failurecauses 454
10.6 Failuredistributions 455
10.7 Accelerationfactors 458
10.7.1 Thermal acceleration 458
10.7.2 Current acceleration 462
10.7.3 Voltageaccelerationfactors 465
10.7.4 RF biasacceleration 472
Contents xi
10.8 Reliabilitypredictions(MTBF, MTTF, FITs, etc.) 473
10.9 Wear-out versusdefects(accelerationversusreal life) 475
10.9.1 Thermal excursionexampleno. 1. Interconnect vias 475
10.9.2 Thermal excursionexampleno. 2. Copper bump 478
10.9.3 Defect amplicationandK factors 482
10.9.4 Environmental example humidityactivation
energy 488
10.10 Processeffectsandinuence 492
10.11 Designfor reliability 495
10.12 Historical trendsandtechnologycomparisons 501
10.13 Summary 502
References 505
11 Power amplier applications 508
Mustafa Akkul and Wolfgang B osch
11.1 Introduction 508
11.2 Systemdesignparameter tradeoffs 509
11.2.1 Output powerefciencytradeoff 509
11.2.2 Linearity, modulationscheme, andcrest factor 512
11.3 Systemlevel linearizationtechniques 514
11.3.1 Introductiontolinearizationtechniques 514
11.3.2 Digital basebandpredistortion 514
11.3.3 Memoryeffect compensation 517
11.3.4 Impact onpower efciency 517
11.4 Wirelesscommunicationpower ampliers 519
11.4.1 Mobileradiocommunicationtoday 519
11.4.2 Systemlevel andpower amplier requirements 522
11.4.3 Power amplier designoutline 523
11.4.4 Dohertyamplier for efcient basestations 527
11.5 Militarypower ampliers 530
11.5.1 Radar Tx,Rxmodules 530
11.5.2 EWapplications 534
11.5.3 Anti-IEDapplications 538
11.6 In-phasepower combiningtechniques 538
11.6.1 Wilkinsonpower combiners 538
11.6.2 Gysel combiner 542
11.7 Quadrature-phasepower combining balancedampliers 544
11.7.1 Branch-linequadraturehybrid[19] 547
11.7.2 Langecoupler 549
11.8 Anti-phasepower combining pushpull ampliers 552
11.8.1 Coupledcoil transformers 553
11.8.2 Transmissionlinetransformers 554
11.8.3 RF/microwavepushpull amplier 557
xii Contents
11.9 Dohertycombining 559
11.10 Conclusions 567
References 568
12 Amplier measurements 570
Michael G. Hiebel
12.1 Introduction 570
12.2 Power measurements 570
12.2.1 Typical power sensor principles 570
12.2.2 Typical sourcesof measurement uncertainties 574
12.2.3 High-power RF measurementsanddirectional power 576
12.2.4 Power measurementsusingaspectrumanalyzer 579
12.3 S-parameter measurements 580
12.3.1 Theconcept of S-parameters 580
12.3.2 Scalar networkanalyzersandtheir limitations 582
12.3.3 Vector networkanalyzers 586
12.3.4 Introductiontosystemerror correction 588
12.3.5 Calibrationwithdifferent connector types 589
12.3.6 CalibrationwithPCBs, test xtures, andwafer probers 593
12.3.7 Calibrationconsiderationfor high-power setups 596
12.3.8 Residual errorsandmeasurement uncertainties 598
12.4 Further linear measurements 599
12.4.1 Amplier gaindenitions 599
12.4.2 Efciencyfactor 602
12.4.3 Linear distortion, phaseandgroupdelaymeasurement 603
12.4.4 Linear stabilityconsiderations 605
12.4.5 Mixed-modeS-parameters 608
12.5 Nonlinear measurements 611
12.5.1 Inter modulationdistortion(IMD) andharmonic
distortion(HMD) 611
12.5.2 Compressionpoint 615
12.5.3 Large-signal networkanalysis 616
12.5.4 Load- andsource-pull measurements 619
12.5.5 Hot S-parameters 622
12.6 Modulatedmeasurements 623
12.6.1 Crest factor andCCDF 624
12.6.2 Adjacent channel power ratio(ACPR) 625
12.6.3 Noisepower ratio(NPR) 630
12.6.4 Error vector magnitude(EVM) andconstellation
diagram 630
12.6.5 AM/AM andAM/PM measurements 632
12.6.6 Memoryeffects 632
Contents xiii
12.6.7 Pulsedmeasurements 633
12.6.8 Bit error ratio(BER) andsymbol error ratio(SER) 635
12.7 Noisemeasurements 636
12.7.1 Amplier noisefactor andnoisegure 637
12.7.2 Noiseguremeasurement 637
12.7.3 Noiseparameters 640
12.8 Conclusions 641
References 642
About theauthors 644
Index 651
Contributors
Mustafa Akkul
ASELSAN A.S.
Inder Bahl
CobhamSensor Systems
Wolfgang B osch
Graz Universityof Technology
Wayne Burger
FreescaleSemiconductor
Steve Cripps
Cardiff University
Rob Davis
RFMD
Chris Dragon
FreescaleSemiconductor
Dominic FitzPatrick
PoweRFul Microwave
Michael Hiebel
Rohde& Schwarz GmbH& Co. KG.
Stephen Maas
AWR, Inc.
Mali Mahalingam
FreescaleSemiconductor
Daniel P. Myer
CommunicationPower Corporation
(CPC)
Bill Roesch
TriQuint Semiconductor
R.J. Trew
NorthCarolinaStateUniversity
Preface
In 1989, I was responsible for organizing a workshop at the European Microwave
ConferenceonHigh-PowerSolidStateAmpliers. Thisworkshopprovedpopularandso
ArtechHouseaskedmetopersuadethespeakerstoturntheirmaterial intoaformsuitable
for publication, theresult was thebook entitled High-Power GaAs FET Ampliers
of whichI was editor andacoauthor. That book is of coursenot just out of print but
also largely out of date. This book adopts the same philosophy as the previous one
withchaptersondevicetechnology, amplier design, CAD, thermal design, reliability,
measurements, andapplications butwithacompletelydifferentsetof authorsandwith
everychapter completelyre-writtentobringthecontent uptodate.
Thepolitical, economic andtechnical landscapehas changedalmost beyondrecog-
nitionintheinterveningtwodecades. Inthe1980s most RF andmicrowaveengineers
wereworkinginmilitaryelectronics, defensespendingwaslargelyresponsibleforall the
technical advances, andtherewerenomobilephones! Comparethat withthesituation
nowwherethereareprobably just as many RF and microwaveengineers working on
commercial applications as therearein military electronics, commercial applications
oftendrivetechnical advances, andmost householdswill havenot just onebut several
mobilephones andit is themobilephoneindustry that has largely beenresponsible
for this shift toward commercial applications. However, there is one consequence of
this sea-changeintheindustrial andtechnical environment whichhas hadaprofound
knock-on effect when it comes to writing a book such as this. Now the commercial
pressuresof shortest possibletimetomarket andminimumcost, etc. aresointensethat
anyprospectiveauthor workinginthiseldhastobepreparedtocommit endlesshours
of their ownrather thantheir employerstimetothetask. I wanttopubliclyacknowledge
my deepdebt of gratitudeto all theauthors inthis book for makingthat commitment
andhencemakingthisbookpossible.
J ohnWalker
TheHandbook isacomprehensivereferencefor RF andmicrowavepower ampliers.
It includesboththeoryandpracticeaswell asavarietyof different applications. Often
overlookedsupportingtopicssuchasCAD, thermal design, andreliabilityaretreatedin
depth. J ohnWalker has put together anoutstandingteamof authors, eachof whomis
well qualiedto address his topic. Finally, I liketheway it is organizedwithseparate
chapters for three types of RF-power transistors (silicon, GaAs, and GaN/SiC) and
separatechaptersfor ampliersof differentfrequencytypes(HF/VHF/UHF, microwave,
andIC).
Fritz Raab, GreenMountainRadioResearchCompany
J ohnhassuccessfullybrought together, inonebook, thecurrent knowledgefromworld
experts actively involved with thecharacterisation and modelling of devices together
withthosedevelopinganddesigningRF andmicrowavepower ampliers. Thetimely
publicationof thisbookwill serveasauseful referencesourcefor engineersworkingin
boththecommercial andmilitarymarket sectors.
SteveNightingale, CobhamTechnical Services
1 Silicon LDMOS and VDMOS
transistors
Physics, design, and technology
Wayne Burger and Chris Dragon
Freescale Semiconductor
1.1 Technology overview
1.1.1 Introduction/history
Power ampliersareat thecoreof nearly all high-power (i.e., >5W) RF applications.
Theapplication spaceincludes cellular phonebasestation transceiver systems, pulsed
radar, ISM (industrial, scientic, medical), avionics, digital television broadcast, etc.
This diverse and evolving RF power amplier landscape dictates thestrategy for the
design, fabrication, andoptimizationof multiplegenerationsof RF power devices. The
RF power transistor mustsatisfyabroadandoftenconictingsetof applicationrequire-
ments, includingbut not limitedtopower, linearity, efciency, gain, reliability, thermal
management, bandwidth, ruggedness, digital predistortion (DPD) linearizability, and
cost effectiveness. The amplier architecture has also evolved to adapt to the ever-
changingsystemrequirements, most recentlywiththewidespreadadoptionof Doherty
amplierstoboost back-off efciencyinlinear applications. Thesearchitectural evolu-
tions createopportunities for further renements intheRF power transistor to extract
peakperformancefromthearchitecture.
The various major market segments of the RF power market tend to embrace a
dominant devicetechnologythat meetsabroadrangeof theserequirementsuntil anew
technologyemergestooffer amorecompellingsolution. Throughthelate1970s, silicon
bipolar transistorswerethepreferredRF power devicetechnology[12]. Therelatively
low frequencies and amplier requirements of the era were compatible with silicon
bipolar transistor technology, which was capableof providing arobust, cost-effective
solution. Thebipolartransistorshadadequategainandefciency, couldbereadilyscaled
to achievethedesired power levels, and offered linearity that was consistent with the
modest requirements of that era. On the other hand, power gain was relatively poor,
packages with isolated anges were expensive, thermal runaway due to the negative
temperaturecoefcienthadtobecarefullymanaged(usuallyattheexpenseof degraded
performancebecauseof theneedtoincorporateballast resistors), andtheevolvingand
increasinglymorestringentlinearityandefciencyrequirementswerebecomingdifcult
todesignintothetransistors.
The limitations of the silicon bipolar transistor eventually created an opening for
a new generation of transistor technology that offered superior performance without
2 Silicon LDMOS and VDMOS transistors
theselimitations. Theearly 1980s witnessed theemergenceof doublediffused MOS
(DMOS) transistors that were superior to silicon bipolar transistors for many high-
power RF amplier applications[34]. A rangeof factorscontributedtothisimproved
performance, starting with the improved frequency response inherent to a majority
carrierdevicecomparedtotheminoritycarriertransportinthebipolartransistor. Second,
theDMOS transistor structurelends itself to highbreakdownvoltagedesigns without
seriouslycompromisingfrequencyperformance, openingupthepossibilityof increasing
thepower supplyvoltage, loweringthepower supplycost, andsimplifyingthedesignof
ever higher power devices. Another keyadvantageisthat MOSFETsarenot susceptible
tothermal runaway, duetothepositivecoefcient of thermal resistance[5]. Theability
todesignDMOStransistorswithhighlinear efciencyhasalsoemergedasakeyfactor
intheir widespreaddeployment. Thesetopicswill beexploredingreater detail later in
thischapter.
DMOS transistor structureand fabrication technology diverged into two main sub-
groups dependingonthedirectionof current ow, lateral DMOS andvertical DMOS
transistors(LDMOSandVDMOS, respectively) [611]. Eachof thesevariantshastheir
strengthsandweaknesses, andeachhaslargelysucceededinndingappropriatemarket
segmentswithinwhichtoourish. Thedopingproleinthechannel regionof bothtran-
sistorsisformedthroughtheoverlapof lateral diffusionproles, butLDMOSmaintains
thedrainregionandcurrent owlaterallynear thesurfacewhereit canbeeasilymodi-
edandoptimized, makingitmoreattractivewherelinear efciencyandhigh-frequency
operationareimportant. VDMOS, ontheother hand, canachieveexcellent power den-
sity (i.e., extremely lowR
DSon
/area) sincethelargedraindrift regionneededtosustain
highbreakdownvoltagesextendsverticallybelowthesurface. Thissamestructuretends
to limit the scaling of the gate structure, detracting fromthe high-frequency perfor-
mance. This makes it thelogical choicefor applications that requirevery high-power
densityat relativelylowfrequencies. Comparisonsbetweenthesetwotechnologieswill
beexploredthroughout thischapter.
1.2 LDMOS and VDMOS construction
1.2.1 LDMOS
Figure1.1showsapictureof apackagedhigh-power LDMOS transistor, aviewof the
internal construction, andahigher magnicationimageof theLDMOS die. Figure1.2
showsacross-sectionof astandardLDMOSdie.LDMOSdiearen-channel enhancement
modeMOSFETs. TheLDMOS transistor has along, lightly-dopedn-typedrift region
(hereafterreferredtoasthen-driftregion) betweenthedraincontactandthegate/channel
of thedevice. TheLDMOStransistor hasthen-driftregionorientedlaterallyreferenced
to the silicon surface, the origin of the L in LDMOS. The drain supply voltage to
rst order determinesthelengthanddopinglevel inthen-drift region. LDMOSdevices
optimizedfor handsetsmayhaveann-driftlengthof lessthan0.5m, whileanLDMOS
devicedesignedtooperateat50V inanindustrial applicationmayrequireadriftregion
1.2 LDMOS and VDMOS construction 3
(a)
Figure 1.1a 2.1GHz, 170WLDMOSsingle-endedpart inanair cavitypackage.
(b)
5
0
0

m
i
l
Array of
bonding-wires
Gate lead
Flange
MOS capacitors
Transistors
Ceramic
substrate
Embedded
capacitor
Drain lead
Figure 1.1b High-power LDMOSdevicewithlidremovedillustratingtheLDMOSbuilding
blocks, MOSCAPs, andextensivewirebondarraysintheinput andoutput matchingnetworks.
4 Silicon LDMOS and VDMOS transistors
(c)
Gate Lead
Drain Lead
ESD Protection
Figure 1.1c Typical layout of a50WLDMOSbuildingblockdesignedfor 2GHz operation.
Gate
gate oxide
oxide
metal strap
Drain
n
+
drain n-drift region n
+
source
p-type epitaxy
p
+
substrate
Source
Channel
PHV region
p
+
sinker
Figure 1.2 LDMOScross-sectionillustratingkeyfeatures, includingtopsidegateanddrain
connectionsandabacksidesource.
56mlong. Thevastmajorityof cellular infrastructurebasestationsaredesignedwith
asupplyvoltageof 2832V. Whenthetransistoristurnedon, thedriftregionsimplyacts
asavoltagevariableresistorandcreatesavoltagedropsuchthatthepotential inthedrain
regionbelowthegateissignicantlylessthantheappliedDC biasinorder topreserve
the integrity of the gate oxide and ensure that HCI (hot carrier injection) is limited.
Most LDMOSdesignsalsoleverageatechniquetermedRESURF REducedSURface
Fields[12], whichreliesuponarapidtwo-dimensional expansioninthedepletionregion
widthwithincreasingdrainbiasthatkeepsthepeakelectriceldbelowthecritical eld
for impact ionization, without compromisingthelowdrainbiasR
DSon
of thetransistor;
this techniqueenables very highbreakdownvoltages whilemaintainingthelowR
DSon
necessary to achievehigh-power density. Unless statedotherwise, references to power
1.2 LDMOS and VDMOS construction 5
p-type
substrate
n
+
source
n
+
drain
gate
VD = 0 V
5 V
10 V
20 V
50 V
n-region
Figure 1.3 Depletionregionboundariesfor V
DS
voltagesof 0, 5, 10, 20, and50V inanLDMOS
device.
densityrefertoW/mmgateperiphery; withthisdenition, high-powerdensitycorrelates
withimprovedperformancefor most guresof merit. Thenatureof thereactivecircuit
elementsinanRF transistor enablesthepeakdrainvoltagetoreachapproximatelytwice
thedrainsupply voltageV
dd
duringclass AB operation, andevenhigher duringother
modes of operation [13]. The ability to withstand these peak voltages explains why
datasheetsfor transistorsdesignedfor 32V ClassAB operationtypicallyspecify65V
minimumfor drain-to-sourcebreakdownvoltage, BV
DSS
.
Thelightly dopedn-drift regionintheLDMOS device, alongwiththelightly doped
p-epi region, aredesignedtodepleteasthedrainvoltageincreases, inalignmentwiththe
RESURF principle. Theepi depth/dopingaswell asthen-driftsdepth/doping/extension
must beoptimizedsuchthat thepeakelectriceldacrossthisdepletionregiondoesnot
exceedcritical avalanchebreakdownlevelsduringtheapplicationsRF voltageswings.
Figure 1.3 illustrates through simulation how the depletion region edge progresses
throughthen-drift regionasthedrainbiasvoltageisincreasedfrom1V to65V, with
thegatebiasedatatypical voltageforClassABoperation. Sincethisregionisthelargest
parasitic resistancewithin thetransistor, it also determines thesaturation current and
hencepower density. Keeping this resistanceas lowas possiblewhilemaintaining an
appropriatebreakdownvoltageandHCI reliabilityisacritical partof thedesigntradeoff
madeintheLDMOStransistor designprocess. Proprietarytechniquesareemployedto
increasethepowerdensitywithoutcompromisingBV
DSS
orHCI. Thesethreeparameters
(BV
DSS
, HCI, R
DSon
) denetheboundary withinwhichthetransistor drainstructureis
optimized. The lightly doped p-type epitaxial layer is also important to achieve low
drain to source capacitance, C
ds
, which is important to achieve good high-frequency
performance.
Thegateof theLDMOStransistor ismostcommonlycomposedof astackof polysil-
iconandasilicide(e.g., WSi, CoSi) [1415]. WhileaDC current will not owinthe
gateof aMOSFET, displacement current fromtheAC waveformwill owthroughthe
gatecapacitance, resultinginanundesirablevoltagedropacross thewidthof thegate
6 Silicon LDMOS and VDMOS transistors
n
+
drain
n
+
source
n

drift
region
1E16
1E17
1E18
1E19
1E20
1E15
p-type lateral
channel diffusion
N
e
t

D
o
p
a
n
t

C
o
n
c

(
c
m

3
)
Figure 1.4 Lateral dopingprolealongthesurfaceof anLDMOSdevice.
nger. Thesilicidelowersthegateresistancebyat least anorder of magnitudeover that
of highlydopedpolysilicon. Inthecaseof WSi thiscanrangefrom10O/sqtolessthan
1 O/sq, depending on thickness. If thegateresistanceis too high, thepower gain of
thedevicewill suffer. Thegatelengthandgateoxidethicknessarekey indetermining
thefrequency responseof thetransistor (i.e., f
t
, theunity current gainfrequency of the
transistor). Thinner gateoxidesandshorter gatelengthsresultinahigher f
t
. Inaddition,
athinner gateoxideresultsinahigher devicetransconductance(g
m
), butnotnecessarily
higher RF power gain. This is becausethethinner gateoxidealso increases theinput
capacitanceof thedevicewhichcanlower gain. Thisisanother examplewheredesign
tradeoffsmust beconsidered.
Theasymmetrical p-channel regionof thedeviceisoneof thedistinguishingfeatures
that differentiates theDMOS transistor fromthestandardMOSFET. For theLDMOS
transistor, this regionis createdby usingthegateto self-alignamoderatedosep-type
implant (referredtoasthePHV implant) tothesourceedgeof thegateof thetransistor.
A subsequentfurnaceanneal isusedtolaterallydiffuse(theDinDMOS) thisimplant
into the channel. The source-side structure is completed by the self-aligned implant
andsubsequent diffusionof theheavily dopedn-typesource/drainimplant. Figure1.4
presentsthesimulatedprolefromthesourcetothedraincontact alongthesurfaceof
thetransistor, illustratingthefourdistinctregionsof thedevice(n

source, PHV, n-drift,


andn

drain). TheresultisaMOSFET withanonuniformchannel dopingprole, with


thesourcesidemoreheavily doped than thedrain side. Oneadvantageof this is that
thedopant gradient generatesitsownelectric eldwhichprovidesasmall boost tothe
overall currenttransportof thedevice[16]. Moreimportantly, thisdesignallowsthelarge
supplyvoltagesdescribedearlier tobeappliedwithout sufferingpunch-through. Asthe
1.2 LDMOS and VDMOS construction 7
drainvoltageisincreased, thedepletionregionwill spreadaway fromthenpjunction
formedbytheintersectionof then-drift andPHV/p-epi regions. If that depletionregion
weretoreachthesourcesideof thedevice, then

sourcetoPHV junctionbarrier would


beloweredresultinginadramatic increaseinthesupply of electrons injectedinto the
channel andswept tothedrainterminal by theappliedelectriceld. Thisphenomenon
is referredto as punch-through, andresults inaloss of control of thedraincurrent by
thegatevoltage. Sincethedepletionregionwidthisinverselyproportional tothedoping
density, thegrowthof thedepletionregionintothePHV slowsconsiderablyasit moves
towards themoreheavily dopedsourcesideof thechannel inanLDMOS device(see
Figure1.3). Thispreservesthehigh-voltagecapabilityof thetransistor.
Thesourceof thetransistor isuniqueinanRF LDMOSdevicebecauseitgetsshorted
to the body of the transistor. The body cannot be biased separately fromthe source.
This is doneso that theback of thewafer can beused as thegrounded sourcein the
application. Makingelectrical groundconnectiontothebackof thedieobviatestheneed
for sourcewirestobepresenttomakeatop-sideconnection. Byeliminatingthetopside
sourcebond wires, alargeamount of sourceinductanceis eliminated, increasing the
gainof thetransistor. Tomakethis backsidesourcepossible, then

sourceis shorted
toaheavilydopedp-typeregioncalledthep

sinker bymetal 1(typicallyanaluminum


alloy). This metal is not contacted by a bond wire for biasing and simply acts as a
means to short the pn junction between the two regions. The p

sinker is implanted
veryearlyintheprocessandisthermallydiffuseduntil it meetsthep

substratedoping
whichis gradually diffusingupwardduringthis thermal cycle. Thep-epi must not be
entirely consumedby thesubstrateup-diffusionbecauseof thebreakdownvoltageand
capacitanceconstraints describedearlier. A balancebetweenkeepingalow-resistance
paththroughthep

sinker intothep

substrateandretainingamplelightlydopedp-epi
for breakdownandlowC
ds
must bestruck. Thewafer is thenthinnedthroughaback-
grindprocess(tothicknessesinthe26milsrange) andback-metal isdepositedonthe
wafer backsidesothat agood, low-resistancecontact canbemadebetweenthedieand
package.
There are two components of the device design that are located above the silicon
surface: the eld plate and the drain metallization. The eld plate provides an extra
degreeof freedomwithinthen-drift optimizationtradeoff describedearlier. Byplacing
a grounded conductor (i.e., the eld plate) close to the surface of the n-drift region,
the eld plate can perturb the depletion region and electric elds such that a higher
dopingand/or shorter extensioncanbeusedfor then-drift regionfor agivenamount
of breakdown voltage and HCI. In other words, the parasitic drain resistance of the
device can be lowered, the RF power density of the device can be increased, and
the HCI levels in the device can be reduced if the eld plate is designed correctly.
Figure 1.5 is a simulation of the subsurface electric eld for a device both with and
without a grounded eld plate, fromwhich the peak electric eld can be seen to be
dramatically reducedfor thedevicewithaeldplate. Inaddition, sincethiseldplate
isgrounded, itcanactasashieldbetweenthedrainmetalsandthegateof thetransistor,
reducing the feedback capacitance C
gd
. The drain metallization must be designed to
meet theapplications electromigration requirements. RF power devices aretypically
8 Silicon LDMOS and VDMOS transistors
n
+
source
channel
and n

drift
with shield
4.0e+05
2.0e+05
0.0e+00
without shield
L
a
t
e
r
a
l

E
-
F
i
e
l
d

M
a
g
n
i
t
u
d
e

(
V
/
c
m
)
Figure 1.5 Comparisonof thelateral electriceldmagnitudewithandwithout aeldplateshield.
designed to operateat ajunction temperatureup to 200

C at relatively high current


densities. A typical devicedesigntarget might bea100-year electromigrationmedian
timetofailure(MTTF) atratedpower and200

C. Thisrequiresaveryrobustmetalliza-
tion, andistypicallysatisedwithathickaluminumor goldtopmetal withdimensions
(thicknessandlinewidth) that areappropriatetokeepthecurrent densitylowenoughto
meet theMTTF goals.
1.2.2 VDMOS
TheVDMOS transistor (Figure1.6) shares many of thedevicedesignandoperational
considerations describedfor theLDMOS transistor. Themost signicant differenceis
that the body/substrate of the VDMOS transistor is n-type rather than p-type, and it
serves as thedrainof theVDMOS transistor whereas thebody/substrateis thesource
for theLDMOS device. Then-drift regionis alightly dopedn-typeepitaxial layer on
topof aheavily dopedn-typesubstrate; theVDMOS epi thicknessistheequivalent of
then-drift extension in theLDMOS device. This region is also theprimary source
of parasitic resistanceintheVDMOS devicebut it extendsdowntowardsthebackside
of thedierather thanremainingat thesurface. This designallows theepi thickness to
beadjustedtoachievethetarget breakdownvoltage. For veryhighbreakdownvoltages
inthe200V regime, thisvertical designismoreappropriatethanthelateral designof
theLDMOStransistor. VDMOStransistorssuitablefor RF operationatdrainbiaslevels
inexcess of 100V arenowonthemarket [1718], whereas 50V is thehighest drain
voltageoperational ratingonanLDMOStransistor availabletoday[1921]. Increasing
thedrainvoltageisthelogical pathway todevelophigh-power partswithuser-friendly
impedancelevels. This has ledtoadivergenceinthemarket wherethesetechnologies
1.2 LDMOS and VDMOS construction 9
n-type epitaxy
Channel
Source metal
overlay
Gate
Source
oxide
Gate
n
+
substrate
p
+
diffusion
n
+
source n
+
source
Drain
Figure 1.6 VDMOScross-sectionillustratingkeyfeatures. UnliketheLDMOSstructure, thegate
andsourceareonthetopsidewhilethedrainisonthebacksideof thestructure. Adaptedfrom
reference[25].
competeagainst eachother, withLDMOS tendingto havethehighest values of gain,
efciency, andoperatingfrequency, whiletheVDMOScanachievehigher power levels
at higher drainbiasvalues, but at lower frequencies.
Whilethevertical drift regiondesignenableshigher drainvoltageratingsandpower
capability, which are signicant advantages for certain applications, this drift region
design is not amenable to the incorporation of eld plates; the performance gains
achievedbyLDMOSfor thepast half dozenyearswereenabledbytheincorporationof
eldplates to allowfor aggressivereductions inR
DSon
andincreases inpower density
withoutcompromisingreliabilityor breakdownvoltage. Thevertical driftregiondesign
alsoleadstothebacksideof thedevicebeingthedrainratherthansource/groundterminal
(theLDMOS transistor brings thesourceto thedevicebackside). Sincethetransistor
mountingangeis mechanically andelectrically connectedto thePA heat sink andto
ground, this introduces complexity into the packaging environment for the VDMOS
devicecomparedtotheLDMOS transistor. Finally, thetransitionof current owfrom
lateral tovertical inducescurrent crowdingthat tendstolimit performancecomparedto
theLDMOSpurelylateral transport [22].
10 Silicon LDMOS and VDMOS transistors
0.45
0.40
0.35
0.30
0.25
I
D
S

(
A
)
0.20
0.15
0.10
0.05
0.00
0 10 20 30 40
7.0 V
6.0 V
5.4 V
5.0 V
4.4 V
4.0 V
3.4 V
3.0 V
VGS = 2.0 V 0 V
VDS (V)
50 60 70 80
Figure 1.7 I
DS
-V
DS
familyof curvesfor variousV
GS
values.
1.3 Device physics
1.3.1 Current transport
DMOS devices behave largely the same as standard three-terminal n-channel MOS
devices withregardto transistor operation. Thecurrent-voltageresponsecanbechar-
acterizedashavingcutoff, linear, andsaturationregimesof operation(seeFigure1.7).
Currentequationsfor thelinear andsaturationregionsof operationcanbeapproximated
by equations (1.1) and(1.2), respectively [23], whereI
D
is thedraincurrent,
S
is the
electronsurfacemobility, C
ox
isthegateoxidecapacitanceper unit area, Wisthetotal
gatewidth, L is theeffectivegatelength, and V
G
, V
T
, and V
D
arethegate, threshold,
anddrainvoltage, respectively. Duetothegradeddopingprolewithinthechannel of
thedevice, thereis anadditional electric-eldinduceddrift current component which
is not present in standard MOSFETs, providing an additional boost to the apparent
mobility andg
m
. Notethat for small drainvoltages, theV
D
2
termcanbedroppedfrom
equation(1.1), whichthenreducestothefamiliar linear relationshipbetweenI
D
andV
D
.
I
D
=

s
C
ox
W
L
_
(V
G
V
T
)V
D

1
2
V
2
D
_
(1.1)
I
D
=

s
C
ox
W
2L
(V
G
V
T
)
2
(1.2)
ItisworthnotingthatDMOSdevicesascommonlydesignedfor RF operationcannot
beusedas four terminal devices (i.e., gate, drain, source, andbody). InbothLDMOS
1.3 Device physics 11
Gate Drain
Source
Figure 1.8 Illustrationof thecurrent owintheLDMOSstructure. Thecurrent owislateral
acrossthedrainandchannel, andisthenshuntedtothesourceconnectionat thebacksideof the
wafer.
andVDMOSdevices, thebodyof thedeviceisusedasthesourceor drain, respectively.
In both cases this eliminates the need for a top-side contact for all three terminals
of the device (i.e., gate, source, drain). In the case of LDMOS, only the gate and
drain havetop-sidecontacts allowing for thesourceto remain alow-resistance, low-
inductanceconnection(i.e.,wirebondsarereplacedbydiffusionsthatelectricallyconnect
thesourceto thebacksideof thewafer, which is then connected to systemground
see Figure 1.2) which is important for RF applications. VDMOS has only gate and
sourcetop-sidecontacts, whichhaslayout densicationadvantages, especiallyfor very
highvoltageoperation, aswill bediscussedinalater section. Thedrainof theVDMOS
transistorisinternallyshortedtothesubstratewhich, aspreviouslydescribed, requiresan
accommodationduringpackagingsincethewafer backsidecannot bemounteddirectly
tothepackageangeandheat sink.
The current paths for the LDMOS and VDMOS transistors are illustrated in
Figures 1.8 and 1.9, respectively, but remember that current ow is the opposite of
electronow. TheLDMOSdeviceshowscurrentbeginningatthedrainwhereapositive
voltagehas been applied and owing through thelightly doped n-drift region before
crossing the channel. The current then passes through the n

source into the metal


whichshortsthen

sourcetothep

sinker, andthenintothep

sinker. Thecurrent then


moves vertically through thesilicon and out thebacksideof thesubstrateto ground.
TheVDMOS devicehas acurrent pathwhichbegins at theback of wafer andmoves
vertically tothesurface, transitioningthroughthelightly dopeddrift regionformedby
theepitaxial layer. Itthencrossesthechannel andexitsoutof thesourcecontactterminal.
12 Silicon LDMOS and VDMOS transistors
Gate
Gate
Drain
Source
p
+
diffusion
Source metal
overlay
oxide
Figure 1.9 Illustrationof thecurrent owintheVDMOSstructure. Thecurrent owisvertical
throughthedrainregion, turninglateral acrossthechannel andintothesource.
1.3.2 Behavior of parasitic elements/models
InRF power applications, theoperational effectiveness (e.g., gain, power density, ef-
ciency, etc.) of atransistor ismostlylimitedbyitsparasiticelements. Itisinminimizing
theseelementsthat thetruechallengeof devicedesignbecomesapparent. Capacitances
andresistancesposethebiggest problems. Resistancesareaproblembecausethey not
only dissipateenergy but also limit thepeak current andhencepeak power capability,
andcontributetoanincreaseinthekneevoltageandhencedegradethepeakefciencyof
thetransistor. Parasitic resistances, althoughanecessary by-product of certainregions
of thedevice (i.e., then-drift region) to meet breakdown voltageand HCI reliability
goals, tend to degrade the overall performance of the transistor. Many variations of
thebasic DMOS structurehavebeen reported in an attempt to reduceR
DSon
without
compromisingBV
DSS
. Capacitances poseseveral problems. Themost classical impact
issimply todegradethefrequency responseof thetransistor. Equations(1.3) and(1.4)
aresimpliedequationsfor f
T
(unitycurrent gainfrequency) andf
max
(unitypower gain
frequency), respectively[24], whereC
gs
istheinput capacitance, R
out
isthereal part of
theoutput resistance, andR
in
isthereal part of theinput resistance.
f
T
=
g
m
2C
gs
(1.3)
1.3 Device physics 13
C
D
S
/
m
m

(
F
/
m
m
)
VDS (V)
0 10 20 30
1.20E-12
1.00E-12
8.00E-13
6.00E-13
4.00E-13
2.00E-13
0.00E+00
(a)
Figure 1.10a Typical drain-sourcecapacitance(C
DS
) versusvoltagecurvefor anLDMOSdevice.
(b)
6 5 4 3 2 1 0
C
G
S
/
m
m

(
F
/
m
m
)
VGS (V)
1.05E-12
1.10E-12
1.15E-12
1.20E-12
1.25E-12
1.30E-12
1.35E-12
Figure 1.10b Typical drain-sourcecapacitance(C
GS
) versusvoltagecurvefor anLDMOSdevice.
f
max
=
f
T
2

R
out
R
in
(1.4)
The other impact is that many transistor capacitances are nonlinear functions of the
junction voltage and therefore can result in a distortion of the signal being passed
throughthePA. Figure1.10presentsinput capacitanceC
gs
, output capacitanceC
ds
, and
feedback capacitanceC
gd
versus voltagecurves that arerepresentativeof anLDMOS
transistor, illustratingthesensitivity of thecapacitances to terminal voltage. Thevari-
ation of these capacitances degrades the efciency of the input and output matching
networks since the xed value passives in these networks must be designed to oper-
ateinanenvironment wherethecapacitancesbeingmatcheddependonvoltage. What
14 Silicon LDMOS and VDMOS transistors
(c)
C
D
G
/
m
m

(
F
/
m
m
)
VDG (V)
0
5.00E-14
4.00E-14
3.00E-14
2.00E-14
1.00E-14
0.00E+00
5 10 15 20 25 30 35
Figure 1.10c Typical drain-sourcecapacitance(C
GD
) versusvoltagecurvefor anLDMOSdevice.
p
+
substrate
R
S
C
GS
R
G
C
GD
R
D
C
DS
p-epi
Gate Drain
Source
p
+
n
+
n
+
n

p
Figure 1.11 KeyparasiticcapacitancesandresistancessuperimposedontheLDMOSstructure.
ThegateresistanceR
G
isactuallyperpendicular totheplaneof thedrawnstructure(i.e., intothe
page).
follows is amoredetaileddiscussiononeachof thekey parasitic elements of DMOS
transistors.
Figure1.11shows thevarious parasitic resistances andcapacitances inanLDMOS
transistor. Thedrainresistance(R
d
) islargelydominatedbythen-drift regionandmust
bedesignedtosustainappropriatelevelsof breakdownvoltagewhileminimizingHCI.
1.3 Device physics 15
good sinker linkage
poor sinker linkage
1E16
1E17
1E18
1E19
D
o
p
a
n
t

C
o
n
c

(
c
m

3
)
1E20
Figure 1.12 Comparisonof thevertical dopingprolesthroughthesinker regionof anLDMOS
devicewithandwithout goodlinkagetothesubstrate.
Thisisdiscussedinmoredetail inthenext section. Thegateresistance(R
g
) iskept low
throughtheuseof asilicidewhichsits atopthepolysilicongate. Thesilicideprovides
at least anorder of magnitudereductioningateresistanceover just polysilicon. Given
the high-power capability of these devices, total gate widths tend to be measured in
millimeters rather than microns. How this is achieved froma layout perspective is
shown in a later section. The important aspect to consider is that the RF signal is
travelingdownlongstretchesof gateandthereforeit must alsobeconsideredtoact as
atransmissionline. If R
g
getstoohigh, avoltagedropoccursalongthegatewidthand
thegainof thedevicebecomespoor. Finally, R
s
isdrivenprimarilybythesinker region,
thelink to thep

substrate, thep

substrateresistance, andvarious smaller resistances


associatedwiththedieattachandmetal packageange. If onewereto takeavertical
look at thedopant proleseenthroughthesinker tothesubstrateit wouldlook likethe
solidlineinFigure1.12. A failuretoformalow-resistancelink betweenthep

sinker
andthesubstrateisillustratedbythedashedlineinFigure1.12, whichwill degradethe
RF performanceof thetransistor.
The capacitances in the LDMOS device typically have both xed and nonlinear
components. Beginning with the drain-to-source capacitance C
ds
, a typical C
ds
CV
curveisplottedinFigure1.10a. Thenonlinear natureof thecurveisduetothenonlinear
spreadingof thedepletionregionintoboththebodyandn-driftregionasthedrainvoltage
isincreased(seeFigure1.3). It isaffectedby thedopant levelsinthedeviceaswell as
theshielddesignswhichcanperturbthen-driftdepletionsif placedclosetothesurface.
Inaddition, therearexed, voltage-invariantintermetal fringingcapacitanceswithinthe
devicethat shift theentireCVcurveup. Thenonlinear natureof C
ds
canbeaproblem
sincevoltageswings will createarangeof capacitances for eachRF cycle. This leads
to distortionandcanalso becomeproblematic for specic types of PA designsuchas
16 Silicon LDMOS and VDMOS transistors
envelopetrackingthat varythedrainvoltagedynamicallytoadjust output power levels.
Another challengefromnonlinear capacitancesistheimpact of thenonlinearityonthe
matchingnetworkdesign; sincethematchingnetworkcomponentsarevoltageinvariant
(inductorsandMOScapacitors, typically), theinstantaneousimpedancetransformation
will vary across the RF cycle as the device intrinsic capacitance varies, resulting in
compromisedperformanceover most of theRF cycle. Andnally, C
ds
determines, to
rst order, the intrinsic output impedance of the transistor; for silicon transistors in
particular, thisjunctioncapacitancecanleadtoverylowimpedancesthataredifcultto
designbroadbandmatchingnetworksfor.
Thegate-to-sourcecapacitanceC
gs
inthedeviceishighlydominatedbythegateoxide
of thetransistor. Duetothenatureof all MOSFETstheC
gs
CVishighlynonlinear and
showninFigure1.10(b). Priortothedevicereachingthresholdthereisnoinversionlayer
tospanthechannel directly belowtheoxide. Thereforeadepletionregioniscreatedto
uncover chargetobalancetheappliedgatevoltage. Oncethedevicegoesintoinversion,
thereisanamplesupplyof electronsdirectlybeneaththeoxidesurfaceonwhichE-eld
linescanterminate. Thecapacitancebecomesmuchlarger sinceit nowconsistsof only
thegateoxiderather thanthegateoxideinserieswithadepletioncapacitance; theon-
stateC
gs
for anLDMOSdeviceistypicallytwotofour timeslarger thanC
ds
measuredat
28V whereasfor aVDMOSdevicetheratioiscloser tounity. Thisnonlinear behavior
of theinput capacitancewithvoltagealsocreatesproblemswithlinearityintheformof
phasedelaysfromtheinput totheoutput of thedevice.
Thegate-to-drainfeedback capacitance(C
gd
) hasthesameCVshapeasC
ds
but the
magnitudein atypical LDMOS deviceis much lower C
dg
at 28 V is typically less
than5%of C
ds
at 28V. Thenonlinear contributionstemssolely fromwherethen-drift
regionisoverlappedbythegateandisthereforemanipulatedbythen-drift doping, the
extentof thelateral diffusionof thePHV inthechannel, thegateoxide, andthevariation
indepletionregionlocations withbias. Therearealso signicant contributions to C
dg
fromintermetal fringing. Variousshielddesignshavebeenusedtoconceal thegatefrom
thedrainmetal andhencereducethefeedback capacitance. Theshieldisgroundedand
thereforeterminates E-eld lines originating with thedrain. ExcessiveC
gd
can lower
power gaininthedeviceandincreasetheinstability.
ThedescriptionsappliedtotheparasiticresistancesandcapacitancesforLDMOSalso
applytotheVDMOSstructure. Inexchangefor then-driftregionbecomingvertical and
therebyincreasingtheexibilitytodesignforbreakdownvoltagesof 100V orhigher, the
parasiticcapacitancesof theVDMOSstructuretendtobehigher thanfor theequivalent
power RF-LDMOS device. In addition, compared to LDMOS theVDMOS structure
lendsitself towardslower operational frequencies(i.e., lower gainatagivenfrequency).
Thelack of agroundedshieldstructureintheVDMOS device(seeFigure1.6) tends
toincreaseC
gd
, inadditiontonot providingtheadditional devicedesignexibilitythat
agrounded shield layer provides (i.e., thegrounded shield has enabled higher n-drift
dopingconcentrationstoincreasepower densitywithout sacricingHCI performance).
Therearefewbenignparasitic elementswhenconsideringtheperformanceof high-
power RF transistors. A robust design process based upon models that include these
parasitic elements is critical to enableoptimizationof thedesignacross abroadrange
1.3 Device physics 17
of performancemetrics. Anexcellent referencefor thecharacterizationandmodeling
of RF power devicesis[25].
1.3.3 BV
DSS
, R
DSon
, HCI boundaries
Breakdown voltage (BV
DSS
), linear regime on-resistance (R
DSon
), and HCI are three
critical parameters that aretraded off against oneanother in thepursuit of higher RF
performance. Manyaspectsof thetransistor designareconstrainedbytheseparameters
andfor themost part arecontrolledbythedrainregionof thedevice. Manipulatingthe
drainof thedeviceinvariousways(i.e., n-driftdoping, n-driftlength, shieldplacement,
anddesign) is collectively referredto as drainengineering. This sectionis devotedto
thistopic.
Breakdownvoltagebetweenthedrainandsourceof aMOSFET whilethetransistor
channel isOFF (i.e., gatevoltageiszerofor standardLDMOSandVDMOSdevices) is
referredtoasBV
DSS
. Foratypical wirelessbasestationapplicationwiththePA operating
in Class AB bias, thedrain DC supply voltagewill bein the2632 V range, but the
peak RF voltage which occurs on top of the DC bias will essentially be double this
value. ThiswouldimplyaminimumBV
DSS
requirementof 64V. For thisreasonthedata
sheets typically specify 65V minimumBV
DSS
for cellular infrastructureapplications.
Thisisachievedwiththelightly dopedn-drift regionthat isdesignedtooperateinthe
RESURF regime. Discussion of thebreakdown mechanismis required to understand
howthisworks.
The drain-source breakdown in an LDMOS or VDMOS device occurs when the
electric eldacross then-drain/p-sourcejunction (thejunction which is vulnerablein
these devices is actually between the drain and the body of the MOSFET, but recall
that thesourceandbody areshortedso thedrain-sourcevernacular remains accurate)
exceeds thecritical level required for aphenomenon known as avalanchebreakdown
to initiate. Withany p,njunctionthat is reversebiased(as is thecasewhenapositive
voltageisappliedtothen-typedrainwhilethep-typesourceisgrounded), adepletion
region extends into each sideof thejunction creating abalanceof charge. Thereare
nofree-owingelectrons inthen-typedepletionregionor free-owingholes inthep-
typedepletionregion, hencetheyaredepletedof mobilecarriers. Without thesemobile
carriers, thedopant atomswithinthesiliconlatticepresent axedcharge(i.e., positive
chargeinthen-typedepletionregionandnegativechargeinthep-typeregion). These
xedcharges set upanelectric eldacross thedepletionregions. Theintegratedxed
chargeinthedepletionregionsoneither sideof thejunctionisalwaysequal. If thedrain-
source voltage is increased, the depletion regions grow uncovering additional xed
chargewhichin-turnresultsinalarger electric eld. Howlargethedepletionregionis
dependsonthelevel of dopantinthatregion. If theregionishighlydoped, thedepletion
regionisquitesmall sinceavery small depletedareauncoversalargeamount of xed
charge(recall that thexedchargecomesfromthedopant inthelattice). If theregionis
lightly dopedtheoppositeis true: thedepletionregionmust extendalargedistanceto
exposethenecessaryxedcharge. Thisconcept isimportant inthat for agivenapplied
voltage, thepeak valueof theelectric eld that extends over along distanceis lower
18 Silicon LDMOS and VDMOS transistors
thanonewhichoccursover averynarrowregion. Itisthepeakvalueof theelectriceld
whichincitesavalanche[26].
Within thedepletion region electron-holepairs areconstantly being generated that
areswept fromthedepletionregionsbytheelectriceldcreatedbytheappliedvoltage,
resultingintheleakagecurrentinthedevice. Asthevoltageacrossthejunctionincreases,
thepeak electric eldwill eventually reachavaluewherethespontaneously generated
electron-holepairsgainsufcient energy fromtheeldtobreak electronbondsduring
collisions withthelatticeatoms, leadingto thegenerationof newelectron-holepairs.
Thisnewlyformedelectronholepairrepeatsthepattern; itiseasytoseehowtheprocess
can lead to an exponential increasein current for asufciently largeapplied voltage.
Thisprocessistermedavalanchebreakdown. Theresultant electrical curveisshownin
Figure1.7. Inthisexampleit isclear that anexponential growthincurrent isoccurring
at 72V.
Designing for high BV
DSS
is most easily achieved by using alight dopant level on
both sides of thedrainsource(body) junction. In both LDMOS and VDMOS cases,
thebody isalready lightly doped. Thedrainhowever hasmany designelementswhich
canbeadjustedtoachievethedesiredbreakdownvoltage. Themost obviousgiventhe
discussionthusfaristosimplyusealightlydopeddrain. However, if then-drainregionis
short andshallow, thenthedepletionregionwill veryquicklyconsumetheentiren-area
and hit thendrain contact area, pinning thelateral growth of thedepletion region.
This means that length and depth of then-drift region becomeadditional parameters
which must be carefully designed. The result is a two-dimensional depletion region
spread(RESURF) that doesnot occur insimpleone-dimensional junctiontheory [12].
Referring to Figure 1.3, the progression of depletion laterally fromthe channel and
verticallyfromthebodycausesareductionineldstrengthastheoverall electriceldis
nowsplitintovectorswhichareorthogonal tooneanother. A full discussionof RESURF
isbeyondthescopeof thischapter but thetypical patterninlateral electriceldacross
then-driftregionisseeninFigure1.6withtwoelectriceldpeaks: onenear thechannel
andonenear thendraincontact area. To maximizeBV
DSS
then-drift doping, depth,
andlengtharedesignedsothat thesepeaksarenearlyequal.
Another element of drain engineering design is the use of shields or eld plates
abovethen-drift region(seeFigure1.2). Theconcept behindeldplates is to provide
an additional degree of freedomto modify the eld distributions within this critical
region of the device. If a grounded conductive layer is placed close enough to the
surfaceof thedeviceit creates asurfacefor electric eldlines to terminateupon; this
structureiscommonlyreferredtoasaeldplate. Theeldplateservesseveral purposes.
Oneistoreducecapacitivecouplingbetweenthedrainandgatewhichimproveshigh-
frequency performance. It shouldbenotedthat early devices placedagroundedmetal
shieldbetweenthedrainandgatetoreducecapacitance, butfar enoughfromthesilicon
surfaceto haveminimal effect on theelectric eld distribution in thedrain. Over the
past ten years, LDMOS device design has evolved to place the eld plate closer to
the silicon surface to intentionally alter the eld distribution in the drain region. In
this regime, thecouplingbetweenthedrainandtheeldplateenhances theRESURF
behaviorinthedevice, allowingahigherdopantlevel tobeusedtoachieveagivenBV
DSS
.
1.3 Device physics 19
Thehigher dopant level increasesthepower density, improvingdeviceperformance. In
addition, thedeviceengineer canplacetheshieldonly abovetheportionof then-drift
regionthat is neededandcanalsocontrol howcloseit is by choosingthethickness of
thedielectric depositedbelowtheshield, providingadditional exibility inthedevice
design. Itisimportanttonotethattheeldplateintegrateseasilyintothelateral structure
of theLDMOSdevice; theVDMOSstructureisinherentlyincompatiblewitheldplate
structures.
Looking at a typical family of I
D
V
D
curves for various V
G
values there are two
general regionsof MOSFET operationasdiscussedearlier: linear andsaturation. Inthe
linear region of operation theMOSFET current versus voltagecurves exhibit aslope
whosereciprocal is referred to as R
DSon
. Thesteeper this slopeis then thelarger the
RF signal can swing before becoming limited by the capability of the transistor. A
lower R
DSon
valuetypically translates into higher power density and higher efciency
andisconsideredacritical designcomponent inany LDMOS or VDMOS device. The
desireistokeepR
DSon
aslowaspossible. Thelargest contributor toR
DSon
isthen-drift
regionwherethebreakdownvoltagediscussionaboveillustrates theneedfor alightly
doped(moreresistive) design. Thisisoneof thefundamental tradeoffstobemadewhen
designing an RF PA transistor, and it is of littlesurprisethat thevast majority of the
devicedesignactivity isdevotedtodrainengineeringprecisely thisparticular tradeoff.
This drove the need for shields/eld plates in LDMOS and experimentation with a
varietyof dopingtechniquesinthen-drift area. Other contributorstoR
DSon
includethe
sourceresistancecomponentsof theLDMOSandVDMOSdevicesalready coveredas
well asthechannel resistancecontributionwhichisnegligibleif designedproperly.
LDMOSdevicesrelyonthelateral diffusionof ap-typeimplanttocreatethechannel
doping prole. This results in the preferred higher doping at the source end of the
channel andlower dopingat thedrainendof thechannel (seeSection1.3). However,
if thelateral diffusionistoogreat duetoeither athermal cyclewhichistooaggressive
or agatelengthwhichis too short, thep-typedopant will reachthen-drift regionand
overcompensate. Thisresultsinthep-typedopantcounter-dopingthen-typedopantand
that area of overcompensation becomes a p-type region. If there is no n-type region
to link up to thedrain edgeof thegate(seeFigure1.2) then thesmall p-typeregion
becomesalargeparasiticresistance, R
DSon
increasesdramatically, andpower capability
is lost in thedevice. This makes controlling gatelength and lateral diffusion thermal
cyclesacritical manufacturingconcernforLDMOS. TheVDMOStransistorhassimilar
considerationsintermsof controllingthelateral diffusionof thePHV implant.
HCI in MOSFET transistors must beconsidered with respect to theimpact it will
haveinRF PA applications. HCI is thethirdmajor consideration(theother two being
BV
DSS
andR
DSon
). Thereareavarietyof metricsavailabletocharacterizeHCI, including
thresholdvoltageshift, transconductancedegradation, etc. Thetwocritical parameters
impactedby HCI for RF power devicesareshiftsinR
DSon
andbiascurrent (commonly
referred to as I
DQ
). For athorough understanding of theseeffects adiscussion of the
devicephysicsinvolvedisrequired.
Two things must be present for HCI to occur: an electric eld strong enough to
impart signicant energytothecarriersmakingthemhot andthecarriersthemselves
20 Silicon LDMOS and VDMOS transistors
VDS = 0.1 V
VDS = 28 V
0 0.5 1 1.5 2 2.5 3
1.00E+00
1.00E-01
1.00E-02
1.00E-03
1.00E+04
1.00E-05
1.00E-06
1.00E-07
1.00E-08
1.00E-09
1.00E-10
1.00E-11
I
D
S

(
A
)
VGS (V)
Figure 1.13 Sub-thresholdI
D
V
D
curvesfor anLDMOSdevice(V
D
= 0.1V, V
D
= 28V).
(i.e., electrons). IntheBV
DSS
discussiontheconcept of RESURF wasusedtoillustrate
thattherearetwoelectriceldpeakswithinthen-driftregionof anLDMOSdevice. The
electriceldpeak at thedrainedgeof thegateistheonewhichresultsinHCI if it gets
toostrong. Under normal transistor operation, electronsareowingacrossthechannel
withtheaidof alateral electriceld. Aswithavalanchebreakdown, theeldcanbecome
strongenoughthat theelectrons areacceleratedto apoint wherecollisions withother
electron-holepairs or thesiliconlatticeoccur. Whiletheeldis not strongenoughto
begintheavalancheprocess, thecarrierstravelingnear thesurfacecanget misdirected
duringacollisionandendupbeinginjectedintothegateoxide. Howdeepintotheoxide
they areinjecteddepends ontheenergy of theelectronandtheavailableenergy states
intheoxide. Onceinjectedthis electronacts as axednegativechargewhichinduces
apositivechargeinthechannel belowit. Dependingonexactly wheretheelectronis
injectedtherearetwodifferent devicedegradationmechanismswhichcanoccur.
If the electron is injected directly over the channel of the device (see Figure 1.2)
theimpact is onthebias current or I
DQ
under RF operation. Lookingat subthreshold
curvesof atypical LDMOSdevice(Figure1.13) takenwithadrainvoltageof 0.1V and
28V thereisanobservedshift inthecurves. Thethresholdvoltage(V
T
) islower when
28V isappliedtothedrain. Thisisduetoashort-channel effect withintheeld-effect
transistor (FET). At thesurfaceof thechannel alarger depletionregionextendsintothe
channel whenlargerdrainvoltagesareapplied. Thisuncoversxednegativechargeinthe
channel. Whenapositivegatevoltageisapplied, it islookingtogenerateanequivalent
1.3 Device physics 21
negativechargein thechannel. This leads to inversion as electrons arecreated at the
channel surface and the threshold voltage has been exceeded. If the larger depletion
regionhasalready createdsomenegativechargefor thegateelectric eldtoterminate
upon, thenlessinversionelectronsarerequiredtocreateacompletelyturned-onchannel.
Theresult is alower V
T
. PA applications will set theDC bias using the28V (in this
example) drainsupply by increasingthegatevoltageaboveV
T
until thedesiredI
DQ
is
reached. If HCI isoccurringduringnormal deviceoperation, electronsabovethechannel
will induceapositivechargeessentiallyreversingtheincreaseddepletionspreadcaused
by the28V. This increases V
T
andstarts tode-bias thepart (i.e., I
DQ
decreases). Over
timeasmoreelectronsareinjected, thedeviceslowlylosesitsbiasandthepart will no
longer operateas neededinthePA. If theelectronis injectedabovethen-drift region,
theinducedpositivechargesimplyincreasesR
DSon
which, asstatedearlier, will resultin
decreasedpower capability. RobustnesstoHCI must bedesignedintothetransistor and
characterizationperformedtodeneacceptablelevels.
Characterization of HCI affects is performed through stress testing at theDC bias
whichwill beappliedtothedeviceintheapplication. A typical basestationPA could
requireadrain voltageof 32 V and an I
DQ
of 4 mA/mmof total gatewidth. A drain
voltageof 32V isappliedandthenthegatevoltageisincreaseduntil the4mA/mmis
reached. A rapidassessmentof theHCI wouldentail applyingthesteadystateDCstress
tothetransistor for 1648hours sothat anextrapolationcanbemadeout to20years;
theR
DSon
andI
DQ
drift areestablishedbytakingperiodicmeasurementsthroughout the
stress period. Care should be taken to control the temperature of the device under
test (DUT) as well as theambient temperatureas V
T
is temperaturesensitiveandcan
also impact theI
DQ
readings. As described earlier, HCI into thegateoxideabovethe
channel regionreversesthedepletionregionspreadcausedbytheDCbiasdrainvoltage.
Referring to our exampleonceagain, this means that theV
T
curveat 32 V begins to
movetowardtheV
T
curveatadrainvoltageof 0.1V. Thisisaself-limitingphenomenon
whichmeansthattheinitial impacttoI
DQ
isquitelargeandthenadditional injectionhas
lessandlesseffect astimegoeson. HCI degradationcanbeestimatedasalogarithmic
responsebyplottingtheI
DQ
responseagainst thetimeof stressand(Figure1.14). Most
of thedegradationoccurs intherst fewhours andthenlevels off dramatically. Using
this log response, an estimation for the degradation out to 20 years can be made. A
well-designed transistor will keep the20-year degradation in I
DQ
below10%. This is
usually adequatefor ensuring that thePA remains within performancespecications.
R
DSon
increasesarealsotabulatedafter thestresstestingdescribedabove. Again, the20-
year responseshouldbebelow10%butalsoof importanceistheinitial 16hshiftwhich
shouldbelower than5%(preferablylower than3%). ItisimportanttonotethatHCI isa
functionof temperature, voltage, andcurrent, andthat theaboveDC testingisintended
to provideadevicewith acceptableHCI sensitivity under most operating conditions.
Thenal assessment of HCI requires testing in theactual application environment to
properlyaccount for theactual stressconditions.
Many facets of thedevicestructureimpact HCI sensitivity, includingsurfaceoxide
quality, n-drift junctionproles, shielddesign, etc. HCI mitigationstrategies typically
work against another deviceparameter (e.g., reducedn-drift dopingto lower HCI will
22 Silicon LDMOS and VDMOS transistors
0.0024
0.00238
0.00236
0.00234
0.00232
0.0023
I
D
Q

(
A
)
Time (s)
0.00228
0.00226
0.00224
0.00222
0.0022
1E+01 1E+02 1E+03 1E+04 1E+05 1E+06 1E+07
20 years
1E+08 1E+09
Figure 1.14 HCI induceddegradationof thequiescent current (I
DQ
) inanLDMOSdevice.
degradeRdsonandpotentiallyimpact BV
DSS
). ThetradeoffsbetweenBV
DSS
, R
DSon
, and
HCI arefundamental to thedesign of LDMOS and VDMOS transistors. Engineering
variouswaysof improvingthesetradeoffstoallowfor improvementsinRF performance
has driven device development in this application space for more than 10 years and
continuestoday. Extensivedevicesimulationisneededtofullyunderstandthecomplex
interactions whichareinvolvedwithany particular devicedesign. It is also important
toperiodically characterizeHCI under typical applicationconditionstoensurethat the
DC characterizationremainsrelevant initsabilitytopredict applicationHCI behavior.
1.3.4 Snapback/ruggedness
VariousRF applicationsrequirethat thedevicesbeabletowithstanddifferent levelsof
RFstress theyneedtobeconsideredruggedenoughfortheapplication. Usuallywhat
drivestheruggednessrequiredisthelevel of RF voltage/current excursionsexpectedto
beexperiencedby thetransistor. Theseexcursions arefrequently createdby mismatch
conditions that occur at theoutput of thedevice. Radar applications, for instance, use
pulsed signals which may incur transients which stress thedevice, whileapplications
such as aCO
2
laser routinely havethePA operating into what is essentially an open
circuit. Usually various voltagestanding waveratios (VSWRs) areused to stress the
devicestodeterminethelevel of ruggedness. Devicesaretestedat5:1or10:1(orhigher)
VSWRsat different levelsof input overdrivetoassessrobustness. It isalsocommonto
1.3 Device physics 23
p

type epitaxy
p
+
sinker
PHV region
n
+
source n

drift region
gate oxide
Drain
oxide
metal strap
Gate
n
+
drain
p
+
substrate
Source
Figure 1.15 LDMOScross-sectionillustratingtheparasiticnpnbipolar formedbetweenthedrain,
channel, andsourceregions.
characterizeruggednessat elevateddrainvoltageswherethedeviceismoresensitiveto
ruggednessfailures.
Therearetwodevicerelateddesignconcernswhichmustbeconsideredwhenensuring
adequatetransistorruggedness: breakdownvoltageandsnapbackcurrent. Theavalanche
breakdownconcept has already beendiscussedindetail intheprevious section. If RF
voltageswingsareallowedtoexceedthebreakdownvoltagethenthecurrent withinthe
devicerisesrapidlyandthereisariskof acatastrophicthermal failureof thetransistor.
This means that therst measureof defenseagainst ruggedness failures is designing
thepart such that thevoltageswings spend very littletimeexceeding breakdown. Of
courseonecoulddesignthepart withanextremely largeBV
DSS
to ensureahighlevel
of ruggedness but as is madeclear in theprevious section this would result in aloss
inRF performance. Ideally, thetransistor shouldhavethelowest level of BV
DSS
needed
to provide adequate ruggedness for the application. This means that at the extremes
theBV
DSS
will beexceeded, thereforethesecondaspect of ruggednessdesigninvolves
increasingthecurrent level whichcanbewithstoodwhileinbreakdown. This is most
directlylinkedtoaphenomenonknownassnapback.
Referring to Figure 1.15, there is a parasitic bipolar device within the LDMOS
structure(asimilarparasiticbipolardeviceexistswithintheVDMOSstructure indeed,
itisabyproductof typical MOSFET structures). Theemitter isthen

source, thebaseis
thebodyof thedeviceandthecollector isthedrain. Whenimpact ionizationisinitiated
andavalanchebreakdownoccurs, thereis asuddenanddramatic increaseinthelevel
of electronsandholesinthedrainregionof thedevice. Thebuilt-inelectric eldspull
24 Silicon LDMOS and VDMOS transistors
0.45
0.40
0.35
0.30
0.25
0.20
0.15
0.10
0.05
0.00
0 10 20 30 40
VGS = 2.0 V
0 V
3.0 V
3.4 V
on-state
breakdown
off-state
breakdown
4.0 V
4.4 V
5.0 V
5.4 V
6.0 V
7.0 V
50 60 70 80
VDS (V)
I
D
S

(
A
)
Figure 1.16 On-stateversusoff-statebreakdowncurves.
the electrons out of the drain of the transistor, while the holes are injected into the
baseregionof thebipolar transistor. Theholecurrent canforwardbiastheemitterbase
junction, andso moreelectrons areinjectedacross thechannel andinto thehigheld
drainregionwhichcreatesmoreholesandelectronsduetoavalanchingandcreatinga
feedback loop that can result in extremely large, localized current ows that result in
catastrophicthermal failureof thetransistor. Thisisreferredtoassnapback, andcanbe
characterizedbybothasnapbackvoltageandcurrent.
The goal of enhancing ruggedness is to prevent snapback fromoccurring by both
delaying the onset of impact ionization, and to design the transistor to minimize the
injectionof holesintothebaseof theparasiticbipolar onceimpact ionizationhasbeen
initiated. Increasingthesnapback voltagetypically entailsincreasingBV
DSS
. However,
BV
DSS
is the off-state breakdown voltage; it is equally important to increase the on-
state breakdown (see Figure 1.16). The drain region design (doping levels, shields,
etc.) dictates theonandoffstatebreakdownbehavior; designfor ruggedness becomes
another of the tradeoffs of the drain engineering process. Strategies to increase the
snapback current entail both moving thelocation of impact ionization away fromthe
baseof theparasiticbipolar transistor, anddesigningthedevicetoshunttheholecurrent
to ground, bypassing injection into thebaseof thebipolar. Figure1.17is an example
illustratingtheeffectonholecurrentbymodifyingthedrainof thedevicetoaccomplish
bothgoals(movingtheimpactionizationawayfromthebaseof thebipolar, andshunting
theholecurrent toground).
1.3 Device physics 25
Baseline Optimized
Hole current shunted
to substrate
Hole current
injected into base
Gate Drain Gate Drain
Figure 1.17 TCADsimulationof holecurrent densityfor twostructurestakenintoavalanche
breakdown. Theoneof theleft depictsabaselinedevicewiththemajorityof theholecurrent
beinginjectedintothebaseof theparasiticnpn, whiletheoptimizedstructureontheright shunts
theholecurrent tothegroundedsubstrate, preventinglatch-up.
TLP Voltage
10 0 20 30 40 50
snapback
0
0.2
0.4
0.6
0.8
1
1.2
1.4
T
L
P

C
u
r
r
e
n
t

(
A
)
Figure 1.18 Typical snapbackcurveduetoturn-onof theparasiticbipolar transistor inan
LDMOSdevice.
Characterization of snapback voltage and current is typically carried out using a
transmission line pulse generator (TLPG) system, in a similar manner to how ESD
sensitivity is characterized. The systemworks by charging up a transmission line to
successively higher voltages and then throwing a switch allowing the stored energy
on thetransmission lineto enter thetransistor. At each pulsethevoltageand current
arerecorded allowing aplotting of thesnapback curve(seeFigure1.18). SimpleDC
26 Silicon LDMOS and VDMOS transistors
measurementswill endindestructiononcethesnapbackvoltageisexceeded. TheTLPG
systemallowsvariousdesignparameterstobeinvestigatedfor their efcacy inadding
ruggednesstothedevice.Finally,thefullydesignedpartistestedasdescribedearlierwith
variouslevelsof VSWRsandoverdrives. Thisisalsoatesttodestructionwherethenal
level of survivabilityisrecorded. Itisimportanttonotethatruggednessperformanceisa
functionnot simplyof thedevicebut thecompleteoperatingenvironment (surrounding
circuit, inputwaveforms, operatingtemperature, etc.); thenal assessmentof ruggedness
performancemust beconductedinthenal applicationunder realistic operational and
stressconditions.
1.3.5 Operating voltage considerations
Althoughconsiderabletimehasbeenspent inthissectiondiscussingwaysof designing
thebreakdown voltagefor agiven device, thefocus has largely been on basestation
typedesignswhereavoltagesupplyof 2632V isused. LDMOSandVDMOSdevices,
however, can be easily adapted to the voltage supply requirements of a wide range
of applications. Theoptimumvoltagelevel tends to beproportional to theRF power
requirementsof theapplication.
In general, changing then-drift region length (laterally by layout for LDMOS and
vertically by epi thickness for VDMOS) and doping level is theeasiest way to tailor
thebreakdownvoltagetoagivensupply voltagerequirement. For lower voltageappli-
cations such as handset PAs, thevoltagecan drop as low as 3 V, whilefor broadcast
applications 50V is quickly becomingcommonplace. Other applications intheindus-
trial/scientic/medical (ISM) spaceareamenabletoevenhigheroperatingvoltages, with
VDMOSdevicesonthemarketdesignedfor 100V or higher operation(i.e., BV
DSS
over
200V). Typical n-driftregionlengthsrangefrom3mforcellularinfrastructures2832
V requirementsdowntotherangeof 0.5mforthelowvoltage, low-powerapplications,
but canbeashighas69mfor the50V applicationsandveryhighRF powers. Each
endof thisrangehasitsownset of designconcernstoconsider.
At thelow-voltageendof thespectrum, suchashort n-drift regionmakesit difcult
tomakethepartresistanttosnapback. J ustbythenatureof suchasmall driftregion, the
avalancheprocessisgoingtooccur incloseproximitytothebaseof theparasiticbipolar
transistor. Thismakestheuseof ap

regionaroundthen

sourcetolower thegainof the


bipolar transistor thatmuchmoreimportantinthesedesigns. For 50V LDMOSdesigns
thereisthechallengeof achievingbreakdownsinexcessof 100V. Longn-drift regions
coupledwithintelligent shielddesignsareneededtooptimizetheusual set of tradeoffs
(R
DSon,
HCI, ruggedness, andBV
DSS
). Butatsomepoint, thevertical breakdownbeginsto
bethelimitingfactor asopposedtothelateral breakdown. Tocircumventthislimitation,
athicker epitaxial layer must beused to extend theamount that thedepletion region
canextendverticallybeforehittingthehighlydopedp

substrate. Thelink-upbetween
thep

sinker andthesubstratemust beredesignedsincethereisnowathicker epi layer


throughwhichalow-resistancepathmust becreated. TheVDMOS devicestructureis
moreamenabletoincreasingtheoperatingvoltage. InVDMOS theepi layer thickness
anddopinglevel determinethebreakdowncharacteristics. TheLDMOS structurehas
1.4 Design/layout 27
Figure 1.19 LDMOSdiscretetransistor layout for an50Wdevicewith500munit gatewidth
(UGW).
moreexibilitytobedesignedfor veryhigh-power density(W/mmgateperiphery) with
lowparasiticcapacitanceduetothelateral structureandaccesstoshieldlayers, but this
exibilitytendstobelimitedtobreakdownvoltagesinthe100130V range. VDMOS
devices, ontheother hand, canbedesignedwithbreakdownvoltagesinexcessof 200V,
but withrelativelyhigher parasiticcapacitancelevelsthat tendtolimit thefrequencyof
operation.
1.4 Design/layout
1.4.1 Top-down nger layout
LDMOS and VDMOS devices for RF PAs deliver very largeamounts of power. It is
not uncommonfor asingletransistor dietodeliver 50W, andoftentwotofour of these
blocksarearrangedinparallel withinapackagetocreateasingledevicewhichdelivers
inexcessof 200W. Generatingthisamount of power requiresavery largegatewidth.
Singletransistor gatewidths areroutinely over 50mmandhavebeenknowntorunto
over 1m. This is anextremely largeamount of gateperiphery whichmust begivena
layoutdesignwhichisefcientandoptimizedforRF operation. Thissectionwill discuss
thevariouscritical designconcernsregardingtop-downlayoutof LDMOSandVDMOS
devices.
The layout of power transistors with very large gate periphery is designed to sat-
isfy anumber of considerations, includingthermal, aspect ratiofor stressandpackage
compatibility, andfrequencyof operation. Thesolutiontothischallengeistoarrangea
largenumber of shorter gates inparallel suchthat they operateinunisonas onetran-
sistor. This parallel arrangement is referredto as anarray of gatengers. All of these
ngerssit withinonelargeactiveareasurroundedbysometypeof eldoxideisolation.
Figure1.19shows atop-downviewof atypical LDMOS layout designedfor 50W
RF power at 2GHz. Eachgatenger is500mwideandisreferredtoastheunit gate
width(UGW) of thetransistor. Two ngers inparallel yields 1mmof gateperiphery.
Thengersarearrayedsuchthat thereissymmetryaroundthecenter of eachdrainand
eachsource. Thisleadstotwiceasmany gatengersastherearedrainngersaseach
drain(andsource) feedstwogates.
TheRF signal andbiases aregoingtobeappliedtothebondpads at theendof the
ngers. Thismeansthateachngerwill actasatransmissionlineasthesignal progresses
28 Silicon LDMOS and VDMOS transistors
Figure 1.20 Layout showinggatebusesfeedinggatetapsspacedat 100mintervalsalongthe
ngersinanLDMOSdevice.
down its length. To minimizethetransmission linelosses or phasedelays which can
result, theconcept of gatetaps is introduced. NoticeinFigure1.20that thereis agate
contact every100mattachedtoametal lineconnectedtothegatebondpad. Thisgate
metal lineor gatebus is used to carry theinput signal down thelength of thenger
with minimal transmission lineeffects dueto thehigh conductivity of thealuminum
alloy. This bus is then electrically connected to thegateitself such that each 500 m
gateis actually ve100mgates inparallel. Recall that thegateitself also typically
hasasilicideatopthepolysilicontokeeptheintrinsicgateresistancelow. Thissilicide
resistance, however, is two to threeorders of magnitudehigher in resistancethan the
metal gatebus, illustratingthenecessityof thegatebus. Other unitgatewidthsandgate
tapspacings areemployed, typically dictatedby thepower level andfrequency. Larger
andlarger UGWseventuallygeneratetransmissionlinelossesevenwithinthegatebus
whilevery small UGWs makefor very poor aspect ratio devices. Higher frequencies
will causetransmissionlinelossestoappear sooner duetotheshorter wavelengthsand
itisthereforemorecommontoseelargeUGWdevicesoperatinginthe900MHzspace
andbelowinthecellular infrastructurearena.
Finally thedevicepitchmust beconsidered. Thedrain-to-sourcepitchfor LDMOS
or sourcetogatepitchfor VDMOS of agivenlayout isthedistancebetweeneachaxis
of symmetry withinasinglenger (i.e., fromthecenter of thesourceto thecenter of
thedrainfor LDMOS, or center of gatetocenter of sourcefor VDMOS). TheLDMOS
drain region is typically kept to aminimumbecausethen

implant region needed to


1.4 Design/layout 29
makeagood drain contact is alargecontributor to theoutput capacitancewithin the
device. Therefore, theminimumn

drainis determinedby theminimumdraincontact


dimension plus enclosure rules for the n

implant. The rest of the drain contribution


to pitch is set by then-drift region requirements of thetransistor. Thesourcesideof
thedeviceallows for moreexibility andcanusually beexpandedor contractedto t
agivenpackagespaceor meet athermal requirement. Whenshrinkingthesourcearea
for LDMOS caremust betakenthat thep

sinker implant doesnot get tooclosetothe


channel of thedevice. Recall that thep

sinker undergoes anaggressivethermal drive


todiffusethedopant downthroughtheepi tomeet thep

substrate. Lateral diffusionof


thedopant isoccurringat thesametimeandtypically reachesseveral microns. Device
pitch, unit gatewidth, andgatetapspacingareall exibledesignparameters that are
adjustedbasedontheperformancerequirementsof thepart.
1.4.2 Bond pad manifolds
ToprovideanRF signal tothetransistor, wiresmuchbeconnectedfromtheleadsof the
packageto thesilicondie. Inthecaseof LDMOS thereareonly gateanddrainwires
since the source is connected through the package ange to ground. This seemingly
simple electrical connection turns out to be quite complex in the eld of RF device
design, since these elements are not merely electrical conductors but instead these
conductiveelementshavecapacitance, inductance, andtransmissionlinebehaviorsthat
arefundamental totheRF performance. Despitethedesignconsiderationsmentionedin
theprevioussection, thetransistor diestill hasalargeaspectratio. Itisnotuncommonto
havediewhichare13cmwidewithanarrayof ngersspanningmostof thelength(see
Figure1.1). Placingonegatewireinthecenter of this array will causeatransmission
lineaffecttobepresentfromthecenterngerstotheoutsidengers. Eachngerwill not
receivethesameRF stimulusandthiscanintroducenonuniformitiesindeviceoperation
duetophasedifferencesbetweentheindividual ngers. Toremedythissituation, alarge
number of wires in parallel are bonded fromthe package lead to a bond pad which
spanstheentirewidthof thedevice. Thebondpadsincur parasiticcapacitancebut this
isminimizedbyplacingthemontopof theeldoxide. Thegoal istofeedthearrayof
ngers as uniformly as possibleto maximizeperformance. This parallel arrangement
of wires introduces inductanceat theinput and output of thedeviceand this must be
incorporatedintoany matchingintendedfor thetransistor. Moreover, thiswirearray is
typically utilizedandoptimizedby designerstopresent adesirablelevel of impedance
at thepackageleadtoeasethecustomersuseof thepart.
At aner level of detail, thereis designof themetal whichconnects thebondpad
to thenger itself. Theprimary consideration in this region of thedeviceis resistive
lossesduetolargeamountsof RF currentbeingfunneledoutof eachngerintothelarge
expanseof bondpadmetal. However, designingtomeet electromigrationrequirements
typically minimizes this resistive loss (see Section 1.4.3), so this is not typically a
problem. Nevertheless, aressuchasshowninFigure1.21canbeusedtominimizethe
impact.
30 Silicon LDMOS and VDMOS transistors
Figure 1.21 Drainmetal aredesignsfor transitioningfromdrainbustodrainbondpad.
1.4.3 Metal design electromigration
Electromigration is aphenomenon which occurs in metal lines when theDC current
densitywithinthelinesbecomesexcessiveinconjunctionwithelevatedtemperatures
conditionsthatareeasilymetinRF power devices. Momentumtransfer duetocollisions
betweenelectrons andthemetal conductor atoms candisplacethemetal atoms which
leadstoresistanceincreaseandeventuallyanopencircuit under severeconditions. This
is awearout mechanismwhich occurs over thelifetimeof thepart and, as such, is a
reliability consideration. Electromigration is discussed further in Chapter 10, but this
sectiondiscusses howto designadeviceproperly suchthat adequateelectromigration
lifetimesareachieved.
To begin designing for electromigration robustness, the metal being used must be
characterized withvarious current andtemperaturestress tests. Different metal alloys
andmetal typesvarywidelyintheir electromigrationresponses. Gold, for instance, has
much higher electromigration resistance than aluminum. Aluminumalloys, typically
formed by the addition of a small percentage of copper [27] have been developed
and arein widespread usein thesemiconductor industry; AlCu alloys haveexcellent
electromigrationpropertiescomparedtopureAl. Copperisanotherconductorwithgood
electromigrationproperties. It isimportant that theelectromigrationcharacterizationof
themetal conductor usesthesameprocessingandstructures(linewidth, topography) as
theactual devicetoaccurately reect themicrostructureandstresses intheconductor.
Oncethenecessary material constants for thechosenconductor anddesignhavebeen
generated, asimpliedformof Blacks equation[28] canbeusedto beginthedesign
calculations.
MTTF =
A
J
2
e
_
E
A
KT
_
(1.5)
whereMTTF is themedian timeto failure(typically dened as a certain percentage
increasein lineresistance), A is a material constant (includes geometry effects), J is
thecurrent density, E
A
is theactivationenergy, kis Boltzmanns constant, andT is the
temperature.
1.4 Design/layout 31
Figure 1.22 Top-downillustrationof thecurrent owinanLDMOSdevice.
Oneof therst thingstonoticeabout Blacksequationisthat thelifetimeit predicts
isinverselyproportional tothesquareof thecurrentdensity. Assumingthatthematerial
propertiesof theconductor havealreadybeenoptimizedtomaximizetheMTTF perfor-
mance, thecurrentdensityisthenextparameter thatthedeviceengineer will manipulate
to improvetheMTTF. Thecurrent density is typically controlledby usingathick top
metal layer, consistent withfabricationdesignrules andthedevicestructure, stacking
metal layerstoreducethecurrentdensity, andnallybydrawingwiderandwiderlinesto
lower thecurrent densityuntil thetarget MTTF isreached. Thereareof courselimitsto
howwidethemetal layerscanbeduetoparasiticcapacitanceconsiderations, socareful
consideration of electromigration is required in thedesign of thedevice. In LDMOS
devices, thedrainlinescarry thelargest currents. A top-downviewof thecurrent ow
(Figure1.22) showsthatthereisagreaterandgreateramountof currentbeingcarriedby
thelineasitnearsitsconnectionwiththedrainbondpad. Thereforetheelectromigration
critical designpoint istheendof eachdrainnger asit entersthedrainbondpad. One
commonpracticeis to arethemetal makingit wider as it approaches theendof the
nger. This keeps thecurrent density relatively constant along thenger length. The
downsideis that extraparasitic capacitanceis generated between thedrain metal and
theunderlyingstructure(i.e., thegateandsource).
Onetechniquethat canbeusedtoenhanceelectromigrationperformanceistodesign
thehigh-current conductorssothat theyareintheso-calledbambooregime[29]. Each
metal line is comprised of metal grains (see Figure 1.23). If the metal linewidth is
kept belowthemediangrainsizethelinebegins to look likeapieceof bamboo with
thegrainboundariestraversingthelinelaterally. Electromigrationoccurspreferentially
alonggrainboundaries, sokeepingthemetal linewithinthebambooregimeresultsin
32 Silicon LDMOS and VDMOS transistors
Figure 1.23 Thegureontheleft isacross-sectionTEM throughthedrainmetal of anLDMOS
deviceshowingtheintersectionof threealuminumgrains; theimageontheright isa
contrast-enhancedoptical microscopeviewof thealuminumdrainbondpad, showingthe
distributionof grainsinthelm.
greatlyenhancedelectromigrationlifetimes. Typical grainsizescanrangefromlessthan
1mtogreater than5mdependingonthemetal depositionprocess.
Theother critical parameter tonoteinBlacksequationistemperature. Thelifetimeis
exponential withtemperature. Giventhat manyPA power transistorsrunat high-power
densities, temperaturescangetashighas200

C. Hence, thethermal performanceof the


deviceandpackageareimportant considerationsthat affect thepeak temperature. It is
commontondRF power transistorsdesignedtooccupyalarger areathanisnecessary
toreducetheareal power densityandtherebylower thejunctiontemperature. Oncethe
devicedesigniscompletedadesigncurveor calculator istypicallygeneratedandmade
availableto enablethecustomer to calculatetheMTTF for their applicationcondition
(seeFigure1.24).
1.4.4 Thermal
Given the large amount of power dissipated in LDMOS and VDMOS transistors for
PA applications, thermal heating within the device must be accounted for. Excessive
temperatures will degrade both the performance and reliability. This means that the
thermal resistanceof thepart must beengineeredtomanagetheheat generatedduring
operation. Thefactorswhichcontributetothethermal resistancearethematerialswhich
theheat must pass through as well as thecross-sectional areathrough whichtheheat
passes. LDMOSandVDMOStransistorsaremadefromsiliconwhichisgivenametal
backing(typicallygoldorasolderablemetal lm) whichisattachedeithereutecticallyor
solderedtothepackageange, whichinturnismountedtotheheat sink. Factorswhich
must beconsidered in thethermal design includelayout (increasing thesource-drain
pitchreduces thepower density), substratethickness (5075mis atypical thickness
for high-power parts), dieattachtechnology (goldeutectic beingthebest, followedby
1.4 Design/layout 33
10
100
1,000
10,000
P
out
(W)
V
dd
(V)
Drain Eff (%)
25
28
42
110 120 130 140 150 160 170 180 190 200 210
Temperature (C)
Electromigration MTTF vs. Temperature
M
T
T
F

(
Y
e
a
r
s
)
Figure 1.24 A graphof theelectromigrationMTTF vs. temperature. Thevariablesareoutput
power, drainvoltage, anddrainefciency. Theequationfor thecurveisbaseduponBlacks
equation.
solder), voidsformedduringdieattach(paradoxically, accountingfor voidscanactually
causetheoptimumsubstratethicknesstoincreasesinceitactsasaheatspreader around
the void), and ange thermal conductivity and thickness (thermal performance is an
important driver of packagetechnology).
Theheat inaDMOStransistor isgeneratedwithintheprimaryparasiticresistanceof
thedevice: then-drift region. Lookingat thetop-downviewof anLDMOS transistor
(seeFigure1.1) thereis anarray of drainregions whichareall generatingheat during
operation. Fromeachof theseregionstheheat will fanout laterallyasit movestowards
thebacksideof thewafer. Itisthereforeaverygoodapproximationtousethetotal active
tubareaas thecross-sectional areadrivingthermal resistance. Thereareexceptions to
this suchas whenadeviceis quitesmall andtheedgeeffects begintobecomealarge
34 Silicon LDMOS and VDMOS transistors
contributor to theoverall thermal resistancecross-section, but for largetransistors the
approximation is valid. This begins to play a role into how the UGW and pitch are
chosen when designing thetransistor. Choosing thelargest pitch possiblefor agiven
packageareawill yieldthebestthermal resistance. For agivenamountof gateperiphery
and a given package constraint, moving to the largest UGW that is consistent with
electromigration and performance considerations allows the source-drain pitch to be
increased, improvingthermal performance. It isapparent that thereareawiderangeof
considerations that must bemanaged during thedevicedesign process to achievethe
best balanceof performanceandreliability.
Over theyears thesetechniques haveheld increasing importanceas power density
within the device has improved. Customers are always seeking more power out of
a given package footprint, which places continued challenges on maintaining good
thermal resistance. Oneaspectof devicedesignwhichhelpsistheincreaseof efciency.
Asdeviceshavebecomemoreandmoreefcient, either throughintrinsicperformance
or throughhigh-efciencyarchitectureslikeDoherty, theheat dissipationhasimproved
for agivenamountof deliveredpower. Thus, a50Wpartwith45%efciencygenerates
signicantlylessheat thana50Wpart with35%efciency, makingthermal resistance
designamoreimportant factor for the35%efciency part (thesearetypical efciency
levelsincellular basestationsoperatedinDohertyor simpleClassAB, respectively).
It should also be mentioned that thermal properties affect the uniformity of the
transistor. Figure1.25showsaninfraredscanof atypical LDMOStransistor. Thecenter
of atransistorgenerallyrunshotterthantheedges. Thiscreatesnonuniformitywithinthe
deviceastheedgengerswill haveaslightlydifferentthresholdvoltage, etc., duetothe
heatprole. Goodthermal designpracticeswill minimizethesetemperaturevariations.
1.4.5 Operating voltage considerations
ThisbookisfocusedonRFtechnologiesforpowerlevelsexceeding1W, ortogeneralize,
noncellular handset RF power applications. For thesepowerslevelsandfor frequencies
upto3GHz, silicontechnologiesandinparticular LDMOS andVDMOS dominate.
Thereasonsarefairlysimple alow-coststructure, goodperformance(gain, efciency,
linearity), excellent reliability, and relatively straightforward scalability to powers up
to1kW. Themarket hassegmentedby applicationvoltageintothreeprimary bands:
12V, 2832V, and50V or higher. As wouldbeexpected, therangeof devicepower
levelsalsoscalewithoperatingvoltage. The12V market rangesupto70Wdevices,
the2832V market ranges upto300400W, andthe50V market includes devices
ratedat over 1kW.
The12V market applicationis primarily for landmobileapplications (re, police,
taxi, etc.). Thespecied maximumapplication voltageis typically 16 V to allow for
automotivebatterychargersoperatinginworst caseconditions. Excludinglowcost, the
mostimportantrequirementof thismarketgiventheharshenvironmental andapplication
conditions that can be encountered is ruggedness. Ruggedness considerations dictate
BV
DSS
inthe50V or higher range. Thelandmobilemarket is characterizedby slices
of spectrumthat vary by country, but in general operateat frequencies under 1GHz;
1.4 Design/layout 35
7S18125AH
T
e
m
p
e
r
a
t
u
r
e

(

C
)
D
R
A
I
N
G
A
T
E
Figure 1.25 Thermal scanfor a125WLDMOSdevice. ThePA istypicallydesignedtolimit the
maximumjunctiontemperaturebelow150

C, althoughthedevicesaretypicallyratedtooperate
upto200

C to225

C.
efciencyat thesepower levelsandfrequenciescanreachinexcessof 70%at P
1dB
for
classAB operation. Linearityrequirementsarefairlymodest comparedtocellular. The
dominant technologyfor thismarket segment isLDMOS.
The 2832 V market application is dominated by cellular infrastructure, but also
includesbroadcast, avionics, andother noncellular applications. Thecellular infrastruc-
turemarket is also dominatedby LDMOS. This market segment is very performance
competitive, requiring state-of-the-art gain, efciency, and linearity. The application
ruggednessrequirementsarelessdemandingthanfor landmobile. Cost isanimportant
consideration, solongasperformanceiscompetitive. Thecellular infrastructuremarket
hasbeenunder signicant cost pressuresfor thepast decade, whichhasdrivenpackag-
inginnovationsuchashigh-power over-moldedplastictransistorsthathavealower cost
structurethanthehistorical ceramicair-cavitypackages. TheBV
DSS
minimumfor these
voltages is typically 65 V. LDMOS products areon themarket for frequencies up to
3.8GHzforWiMAX applications; VDMOSdevicescompeteinthesubGHz, noncellular
arenawheretheir morelimitedfrequencycapabilityisnot alimitation.
The 50 V and higher markets are concentrated in the relatively broad noncellular
application spacethat includes ISM, avionics, and broadcast markets. Thesemarkets
36 Silicon LDMOS and VDMOS transistors
haveaverydiversesetof performancerequirements, withcertainapplicationsrequiring
high-pulseCWwithnolinearity requirements, whileothersdemandstringent back-off
linearitywithgoodefciency. Thesedevicesrequireexcellentruggednessperformance;
certain applications like CO
2
lasers routinely subject the transistors to open circuit
conditionsathighpowerlevels,representingatestinggroundforatransistorsruggedness
capability. This application spacetypically requires higher power transistors than are
practical with 28 V devices, with products on the market at power levels exceeding
1kW. 50V devicestendtohaveminimumBV
DSS
valuesinthe105V120V range, but
applicationswithextremeruggednessrequirementscouldhaveevenhigher breakdown
values. It isonly withinthepast several yearsthat 50V LDMOSdeviceshavebeenon
themarket [1921]. VDMOS competition is robust, particularly at lower frequencies
and high power levels. Thevertical structureof VDMOS also enables scaling of the
breakdownvoltagetoallowoperatingvoltagesof 100V [1718]. Thelateral LDMOS
structuredoesnot scaleasreadily tosupport BV
DSS
valuesof 200V or higher that are
necessarytooperateat 100V.
1.4.6 Frequency considerations: gate length, gate width, resistors
Beforedelvingintofrequencyconsiderations, it isworthwhiletoconsider theimpact of
transistor gainonefciencyandcost. Itbecomeschallengingtodesignstablepartswith
goodisolationif thegainexceeds about 25dB inasinglestagetransistor (multistage
lower power ICs havebeendesignedwithgainupto35dB [3031]). Thedesirefor
high gain is primarily cost a higher gain, high-power nal stage in the PA lineup
will requirealower power andhencelower cost driver, potentiallyfewer devicesinthe
lineup, and requireless spaceon theprinted circuit board (which also translates into
cost savings). Therearealsolineupefciencybenetswithahigh-gainnal stage. The
power-added efciency (PAE) is ametric that tracks theconversion efciency of DC
fromthepower supplyintoRF power, andisdenedbytheequation:
PAE =
P
out
P
in
P
DC
(1.6)
whereP
out
istheRF output power, P
in
istheRF input power, andP
DC
isthepower from
theDC power supply to thedevice. PAE, as thenameindicates, is ameasureof how
efcientlythedeviceconvertsDCpowertoRFpower, andhencecanbeusedtodetermine
power dissipationinthedevice. A 50%PAE devicemust dissipateanamount of power
equal to its output RF power whilea25%PAE devicedissipates threetimes as much
power asit transmits. Thehighest possiblePAE that meetsother systemrequirementsis
clearlythegoal. Bymanipulationof thevariables, PAE canalsobewrittenas:
PAE =
D
_
1
1
G
_
(1.7)
where
D
is thedrainefciency (denedas theratio of P
out
to P
DC
), andG is theRF
power gain. For again of 20 dB (afactor of 100), thePAE is within 1%of thedrain
efciency. As the gain falls below 15 dB, the PAE begins to fall rapidly, degrading
1.4 Design/layout 37
overall efciencyandincreasingoperatingcostsinadditiontorequiringmoreexpensive
techniquesandmechanical itemstomanagethedissipatedheat. Bythesametoken, there
islittleefciencymotivationforthegaintoexceed20dBfromanefciencyperspective,
althoughtherearestill cost andboardspaceconsiderations.
Simplelter theory predicts for asinglepoletransfer functionthat power gainwill
followa6dB/octaverolloff withfrequency, i.e., a20dB gainpart at 2GHz will have
26dB gainat 1GHz, 32dB gainat 500MHz, etc. A commonapproachduringdevice
design is to develop thetransistor to haveas high again as possibleat its maximum
operating frequency without compromising other parameters (i.e., reliability, rugged-
ness, etc.), and then to limit thegain increaseat thelower frequencies to maintain a
stabledeviceandcircuit. Maximizingperformanceat thehighest frequenciesof opera-
tioncausesthedevicedesigner tomigratetowardstheclassical solutionsof shorter gate
lengths, thinner gateoxides, and aggressively reducing all parasitic capacitances and
resistances. TheLDMOS structureismoreamenabletooptimizingfor highfrequency
comparedtoVDMOS, withLDMOSdominatingthecellular infrastructurefrequencies.
Excessivegain at lower frequencies can becountered by degrading theintrinsic gate
of thetransistor (longer gatelength, thicker gateoxide). Another approachthat offers
ancillary benets is to add series resistanceto thegatefeed network, which not only
decreases the gain to manageable levels but also lowers the Q of the input network,
facilitatingthedesignof broadbandmatchingnetworks.
High-power device design also requires careful optimization to the layout of the
individual ngers. Practical limits of gatewidthat frequencies of 13GHz areof the
orderof 1000m,withthemaximumgatewidthdecreasingasthefrequencyisincreased.
Excessive gate widths exacerbate distributed effects (transmission line delays, phase
shifts, etc.) andlower thegainandcanimpact efciency andlinearity. As frequencies
decreasebelow1GHz, thesedistributedeffects becomeless important andthedevice
layouttendstobedictatedbypackageconstraintsandreliabilityconsiderations, suchas
electromigration.
1.4.7 HVICs
High voltage integrated circuits (HVICs) in the context of high-power RF devices
typically refers to having at least two stages of amplication along with elements of
thematchingnetwork(e.g., inductors, capacitors, resistors) all integratedontothesame
semiconductor substrate. HVICsincellular infrastructurewererstintroducedasdriver
devices that were designed to power the nal stage of the PA lineup. The rst high-
power cellular infrastructureHVIC inproductionwas theMRFIC5001, introducedby
thesemiconductor sector of Motorola(nowFreescale) in1999. TheMRFIC5001is a
10W, 900MHz GSM driver HVIC; thisHVIC isatwo-stagedevicehaving26dB gain
at 26V, andwasbasedonthethird-generationHV3LDMOS platformfromMotorola.
Thedesignof bothdriver stageandnal stagehigh-power HVICshasourishedsince
thistime, withalmost all PA designsnowincludingHVICsinthelineup[3235].
The advantages of integration are well known, and include a dramatic reduction
in component count and board space, lower cost, and reduced overall performance
38 Silicon LDMOS and VDMOS transistors
Output
stage
Input
stage
GND VD1 NC NC NC RF in RF NC VG1 VG2 NC GND in
Figure 1.26 Assemblydrawingof a2GHz, 100WLDMOShigh-power IC. Thisisa2-stageIC,
withafullyintegratedinput andinterstagematch(input impedanceis50O), andanintegrated
shunt-L output match.
variation. ModernHVICs aretwo stagedesigns dueto thehighgainof theindividual
LDMOS stages. A traditional discrete design matching network is constructed using
high-Qinductorsandcapacitors. Theinductor isformedfromwirebondsandcanhavea
Qinthe50100range. ThematchingnetworkpassiveelementsinHVICsincludespiral
metal inductorsandintegratedseriesandshuntcapacitors. Themostchallengingpassive
tointegrateintotheLDMOSowisthespiral inductor which, duetotheheavilydoped
substrate, is limited to Q values in the510 range; integrated capacitor performance
doesnot tendtobethelimitingfactor inHVIC designs. Therelativelylowvalueof the
integratedinductorsisadequatefor thedesignof inputandinterstagematches, butistoo
lowfor theoutput stage. Wirebonds continuetobeemployedat theoutput of thenal
stageof HVICswherethemuchhigher currentsrequirethehighest possibleinductor Q
toachievetarget performancelevels.
Theperformanceof HVICshasadvancedover thepastdecade. Thishasbeenenabled
not only by theimproved performanceof theLDMOS transistor, but also by passive
component optimizationandrenementsinthedesignmethodologytoextract asmuch
performanceas possible. An exampleof thestate-of-the-art in IC design today is the
MW7IC18100N [36], which is atwo-stageIC rated at 100 W at 1.8 2.0 GHz with
30 dB gain, designed as a high-gain, high-power output device for GSM and GSM
Edgeapplications (seeFigure1.26). Another exampleof theprogress madein HVIC
performanceisfoundintheMW7IC3825N[37]. ThisICisdesignedfor 28V operation
References 39
inthe3.43.6GHzband, isratedat25WP
1dB
, andhasbeencharacterizedfor WiMAX
operation. Theseexamples illustratethesignicant progress madeinboththeprocess
technologyaswell asthedesignmethodologytoenablethedesignof high-performance,
high-power HVICs.
Summary
The application space for high-power RF transistors is broad and growing, ranging
fromthe ubiquitous cellular base station to avionics, broadcast, industrial, scientic,
medical, etc. Therequirements placedupontheRF power transistor varies depending
upontheapplicationrequirements, includingpower gain, linearity, efciency, reliability,
thermal management, bandwidth, ruggedness, linearizability and, last but certainly not
least, cost. LDMOS andVDMOS technologies dominatetheseapplications dueto an
excellent combinationof thesefactors. VDMOS is strongest at lower frequencies and
higher power levels where the vertical structure can best be leveraged into a higher
operating voltagecapableof very high power levels. LDMOS is thedominant device
technologyfor cellular infrastructurebasestationPA applications, andhasover thepast
fewyearsbeenintroducedintomarketsthat weretraditionally thedomainof VDMOS
and silicon bipolar transistors. VDMOS and LDMOS together dominate the market
for high-power PAs fromfrequencies in the low MHz range up to 4 GHz, and for
power levels that exceed1kW. Investments continuetobemadeinbothVDMOS and
LDMOStofurther improveperformanceandmeettheevolvingrequirementsof theend
applications.
Acknowledgments
The authors would like to acknowledge the support and assistance provided by our
colleaguesatFreescale,withoutwhichmuchof thischapterwouldnothavebeenpossible.
References
1. J. T. C. Chen and C. P. Snapp, Bipolar microwave linear power transistor design, IEEE
Trans. Microw. Theory. Techn., vol. MTT-27, no. 5, pp. 423430, May1979.
2. C. P. Snapp, Microwavebipolar transistor technology present andprospects, NinthEuro-
peanMicrowaveConference, Sept. 1979, pp. 312.
3. E. Fong, D. C. Pitzer, and R. J. Zeman, Power DMOS for high-frequency and switching
applications, IEEE Trans. ElectronDevices, vol. ED-27, no. 2, pp. 322330, Feb. 1980.
4. O. Ishikawa, H. Yamada, and H. Esaki, A 2.45 GHz Power LD-MOSFET with reduced
sourceinductanceby V-grooveconnections, International ElectronDeviceMeeting, 1985,
pp. 166169.
40 Silicon LDMOS and VDMOS transistors
5. J.-J. Bouny, Advantages of LDMOS in high power linear amplication, MicrowaveEng.
Europe, pp. 3740, 1997.
6. A. Wood, W. Brakensiek, C. Dragon, andW. Burger, 120watt, 2GHz, Si LDMOSRF power
transistor for PCS base station applications, IEEE MTT-S Microwave Symp. Dig., vol. 2,
pp. 707710, 1998.
7. C. Cassan, J. J ones, andO. Lembeye, A 2-stage150W2.2GHzdual pathLDMOSRF power
amplier for highefciencyapplications,IEEEMTT-SMicrowaveSymp. Dig., pp. 655658,
2008.
8. F. vanRijs, Statusandtrendsof siliconLDMOSbasestationPA technologiestogobeyond
2.5GHz applications, IEEE RadioandWirelessSymposium, 2008, pp. 6972.
9. P. H. Wilson, A novel high voltage RF vertical MOSFET for high power applications,
TenthIEEE International SymposiumonElectronDevicesfor MicrowaveandOptoelectronic
Applications, 2002, pp. 95100.
10. M.Trivedi andK.Shenai,Comparisonof RFperformanceof vertical andlateral DMOSFET,
EleventhInternational SymposiumonPower Semiconductor DevicesandICs, 1999, pp. 245
248.
11. J. Zhang, D. Sdrulla, D. Tsang, D. Frey, and G. Krausse, Design of rugged high voltage
highpower p-channel siliconMOSFET for plasmaapplications, 38thEuropeanSolidState
DeviceResearchConference, 2008, pp. 7174.
12. J. A. Appels and H. M. J. Vaes, High voltage thin layer devices (RESURF Devices),
International ElectronDeviceMeeting, vol. 25, pp. 238241, 1979.
13. F. H. Raab, F. H. Raab, P. Asbeck, S. Cripps, P. B. Kenington, Z. B. Popovic, N. Pothecary,
J. F. Sevic, and N. O. Sokal, Power ampliers and transmitters for RF and microwave,
IEEE Trans. Microw. TheoryTechn., vol. 50, no. 3, pp. 814826, 2002.
14. W. R. Burger, RecentadvancesinRF-LDMOShigh-power ICdevelopment,IEEE Interna-
tional ConferenceonIC DesignandTechnology, 2009, pp. 3538.
15. S.J.C.H.TheeuwenandH.Mollee,S-bandradarLDMOStransistors,EuropeanMicrowave
IntegratedCircuitsConference, 2009, pp. 5356.
16. W. XieandB. Li, Ananalytical currentmodel forlateral gradual dopingchannel inLDMOS,
IEEEInternational Conferenceof ElectronDevicesandSolid-StateCircuits, 2009, pp. 1619.
17. STMicroelectronics, RF power transistors HF/VHF/UHF N-channel MOSFETs,
STAC4932B datasheet, Feb. 2010RevisedAug. 2010.
18. Microsemi, RF power MOSFET n-channel enhancement mode, ARF1500datasheet, Rev.
E, Oct. 2008.
19. P. Piel, W. Burger, D. Burdeaux, and W. Brakensiek, 50 V RF LDMOS: An ideal
RF power technology for ISM, broadcast, and radar applications, 2008. [Online] Avail-
able: http://www.mwjournal.com/2008/DownloadablePDFs/FREESCALE50VLDMOS.pdf
[Accessed: 6Aug. 2010].
20. Freescale Semiconductor, RF power eld effect transistor, MRF6VP11KHR6 datasheet,
J an. 2008[RevisedApril 2010].
21. NXP, LDMOS avionics radar power transistor, BLA6H0912500 datasheet, Mar. 2009
[RevisedMay2010].
22. M. Trivedi, P. Khandelwal, andK. Shenai, Performancemodelingof RF power MOSFETs,
IEEE Trans. ElectronDevices, vol. 46, no. 8, pp. 17941802, Aug. 1999.
23. P. H. Aaen, J. A. Pl a, and J. Wood, Modeling and Characterization of RF and Microwave
Power FETs, CambridgeUniversityPress, 2007, pp. 2122.
References 41
24. P. H. Aaen, J. A. Pl a, and J. Wood, Modeling and Characterization of RF and Microwave
Power FETs, CambridgeUniversityPress, 2007, p. 32.
25. P. H. Aaen, J. A. Pl a, and J. Wood, Modeling and Characterization of RF and Microwave
Power FETs, CambridgeUniversityPress, 2007.
26. S. M. Sze, Physicsof Semiconductor Devices, NewYork, NY: J ohnWiley& Sons, 1981.
27. M. C. ShineandF. M. dHeurle, Activationenergy for electromigrationinaluminumlms
alloyedwithcopper, IBMJ. ResearchDevelop., vol. 15, no. 5, pp. 378383, 1971.
28. J. R. Black, Electromigration failuremodes in aluminummetallization for semiconductor
devices, Proc. IEEE, vol. 57, no. 9, pp. 15871594, 1969.
29. S. Vaidya, T. T. Sheng, and A. K. Sinha, Linewidth dependence of electromigration in
evaporatedAl-0.5%Cu, Appl. PhysicsLett., vol. 36, no. 6, pp. 464466, 1980.
30. G. Bouisse, High power silicon MMIC design for wireless basestations, 30th European
MicrowaveConference, 2000, pp. 13.
31. Freescale Semiconductor, RF LDMOS wideband integrated power ampliers,
MW7IC915NT1datasheet, Sept. 2009[RevisedDec. 2009].
32. G. Wang, L. Zhao, andM. Szymanowski, A Dohertyamplier for TD-SCDMA basestation
applicationsbasedonasinglepackageddual-pathintegratedLDMOSpowertransistor,IEEE
MTT-SMicrow. Symp. Dig., pp. 15121515, 2010.
33. L. Zhao, G. Bigny, andJ. J ones, A 120watt, two-stage, LDMOS power amplier IC at 1.8
GHzfor GSM/EDGE applications,IEEE MTT-SMicrow. Symp. Dig., pp. 15091512, 2008.
34. C. CassanandP. Gola, A 3.5GHz25WsiliconLDMOSRFICpower amplier for WiMAX
applications, IEEE RadioFrequencyIntegratedCircuits(RFIC) Symposium, 2007, pp. 87
90.
35. C. D. Shih, J. Sjostrom, R. Bagger, P. Andersson, Y. Yinglei, G. Ma, Q. Chen, T. Aberg, RF
LDMOS power amplier integratedcircuits for cellular wireless basestationapplications,
IEEE MTT-SMicrow. Symp. Dig., pp. 889892, 2006.
36. Freescale Semiconductor, RF LDMOS wideband integrated power ampliers,
MW7IC18100NR1datasheet, May2007[RevisedMar. 2009].
37. Freescale Semiconductor, RF LDMOS wideband integrated power ampliers,
MW7IC3825NR1datasheet, Nov. 2008.
2 GaAs FETs physics, design,
and models
Rob Davis
RFMD
2.1 Introduction
The manufacture of GalliumArsenide FET devices and integrated circuits is now a
matureindustry. TheGaAs FET was rst developedinthe1960s and1970s [1], with
the impetus to establish a manufacturing capability coming in the 1980s driven by
governmental support most notably thecomprehensiveMIMIC programmeinthe
United States. In the intervening time the GaAs FET became the default solid-state
devicefor all manner of RF andmicrowaveapplications. However, thepositionof the
GaAs FET in this arenahas not goneunchallenged. It was soon joined by theGaAs
HBT whichhasdominatedthecellular handset power amplier market. Theupper fre-
quency limit of silicon LDMOS technology has steadily increased over recent years
as its highly mature technology was further rened with the result that this technol-
ogy currently dominates high-power RF applications below 3 GHz. More recently,
galliumnitride devices join the fray. The GaN FET is a device technology of great
promise that is steadily being made available by more vendors as its reliability is
established. Initially, galliumnitride is also targeting the lower frequency bands but
iscapableof beingdevelopedfor applicationsacrossthewholemicrowavebandwidth.
For thehigher millimetre-wavefrequencies indiumphosphidetechnology has aplace.
However, GaAs FET technology is proven, competent, mature, and remains a good
choicefor many applicationsincludinghigh-frequency power andhighlinearity. GaAs
technology also has signicant cost advantages over its nonsilicon competitors. The
economies of scale that the cellular communications market has brought to GaAs
technology has revolutionized the manufacture of GaAs products and has given rise
to dramatic reductions in cost. It is in the area of continued cost reduction that the
most signicant new developments in GaAs device and associated technologies are
focused.
ThischapteraimstointroducecontemporaryGaAs-basedpowerFET technology. Itis
writtenwiththeperspectiveof theuserof thetechnologyinmind. Thematerial properties
andthepertinentdevicephysicsarereviewedandrelevantconceptsarerecappedbriey
asnecessary. Thedevicedesignissuesaredescribedfollowedbyasectiononfabrication
withparticular focusonlow-costmanufacture. Thechapter concludeswithadiscussion
of devicemodelsfor circuit design.
2.1 Introduction 43
Table 2.1 GaAs FET materials properties [2, 3]
Parameter Si GaAs Al
0.22
Ga
0.78
As In
0.2
Ga
0.8
As Units
Bandgap, Eg 1.12 1.424 1.698 1.14 eV
Conductionbandstep,
LEc(wrt AlGaAs)
0.17 0 0.31 eV
Electronmobility, (undoped) 1400 8500 3600 6900 Cm/V.s
Electronmobility, (N
d
= 3E17cm
3
) 4000 Cm/V.s
Latticeconstant 5.653 5.655 5.734

A
Breakdowneld, E
BR
310
5
410
5
(46)10
5
(24)10
5
V/cm
Thermal conductivity, 1.3 0.44 0.2 0.05 W/cm.C
2.1.1 Properties of GaAs and related compounds
GaAs and its related compounds offer inherently good electronic properties for
microwavesemiconductor devices. Key material properties of GaAs andtwocommon
partner materials AlGaAs andInGaAs aregiveninTable2.1withthecorrespond-
ing values for silicon provided for reference. The GaAs-based materials shown have
direct band-gaps and high electron mobilities. High mobility results in lower access
resistanceand rapid acceleration of channel electrons to their saturated velocity over
a short distance. These are important benets for microwave devices. Further per-
formance enhancement may be engineered by combining compatible materials with
differing band-gaps to formheterojunction devices. Suitablecombinations of materi-
als allow very effectivedevicestructures to bemanufactured that can provideahigh
degree of spatial control of the charge thereby allowing device performance to be
optimized. AlGaAs has a wider bandgap than GaAs or InGaAs material. Theresult-
ing step in the conduction band when AlGaAs is used in conjunction with GaAs or
InGaAs channel layers allows the current to be conned in the narrower band-gap
material. The step in the conduction band edge between AlGaAs and InGaAs can
be engineered to be considerably greater than that between AlGaAs and GaAs, and
therefore the former combination provides a signicantly higher degree of electron
connement.
Heterojunction devices are only possible if the desired material combinations are
sufciently compatible to allow defect free growth across the crystal interfaces. The
AlGaAscrystal hasthesameface-centeredcubicstructural formasGaAswithalattice
constant that remainsveryclosetothat of GaAsfor all fractionsof aluminumcomposi-
tion. ConsequentlyAl
x
Ga
1x
Asiscrystallographicallycompatiblefor all valuesof mole
fraction x. Unfortunately, high values of aluminumcomposition xareunattractivefor
other reasons. Therst limitingfactor istheemergenceof ahighdensityof deeplevels
calledDX centers[4] whichareformedfor x25%andtheirdensityrisessharplyfor
valuesof xabovethisvalue. For thecaseof InGaAs, highindiumfractionsaredesirable
as the conduction band offset and mobility improve with increasing indiumcontent.
InGaAsalsohasthesamecrystal formasGaAs, however theindiumatomisrelatively
largecomparedtogalliumwiththeresultthatthelatticeconstantof In
y
Ga
1y
Asincreases
44 GaAs FETs physics, design, and models
withmole-fractiony. Consequently, whenathinchannel layer of InGaAs is grownon
a GaAs or AlGaAs crystal the InGaAs layer structure pseudomorphically adopts the
templateof thehost crystal andthis gives riseto astrained(compressed) layer. Natu-
rally, thereis alimit tothis behavior andfor agiventhickness of theIn
y
Ga
1y
As layer
thereis amaximumvalueof ywhichshouldnot beexceededinorder to maintainan
acceptable degree of strain [5]. For thicknesses or mole fractions beyond the critical
limit thenthecrystal strainwill berelaxedbythegenerationof mist dislocations. For
useful AlGaAs/In
y
Ga
1y
Asdeviceswithachannel thicknessintheregionof 10nm, the
maximumuseablevalueof yisof theorder of 2022%. Theaboveissuesthereforecon-
strainthemolecular compositionsthat arepossiblefor practical AlGaAs/GaAs/InGaAs
devices and the compositions of AlGaAs and InGaAs given in Table 2.1 are chosen
to satisfy the constraints described above and are typical of those used in practical
devicestructures. Although thebulk material datagiven in thetabledoes not strictly
apply tothinor strained-layer structureswithadjacent heterojunctioninterfacesaffect-
ing electron transport, thedatashown is indicativeand useful in conveying thebasic
principles.
A further important attributefor asemiconductor for microwaveapplications is the
abilitytoengineer substrateswithveryhighelectrical resistivityinorder thatRF signals
carried by thetracks on thenished deviceshould experiencelowattenuation. GaAs
is naturally blessedinthis regardduetotherelativeeasewithwhichthematerial may
bemadeintoagoodinsulator. High-purity GaAsnaturally hasahighresistivity andis
deemedtobesemi-insulating(SI). ThehighresistivityarisesbecausetheFermi-level is
pinnedverydeepintheforbiddenbandbyanaturallyoccurringcrystal defectknownas
EL2 (ElectronLevel 2). EL2is anear midband-gapelectrontrapwhicheffectively
clampstheFermi-level sodeepthat very fewfreeelectronsor holesareavailablefor a
meaningful currentow. Thenatural resistivityof pureGaAsistypically10
6
10
7
O-cm
at roomtemperature. Substratemanufacturersfurther renethedegreeof insulationby
incorporatingvery small amountsof carbonduringthecrystal growthprocess. Carbon
is ap-typedopant inGaAs andthis is usedto counter-dopetheslightly n-typenature
of pureGaAs. Bycompensatingthehighdensityof deep-donor EL2defectswithalow
concentrationof shallowacceptorsfromthecarbondoping, theresultingresistivitycan
bene-tuned. A typical commercial SI GaAssubstrateexhibitsaresistivity>10
8
O-cm
at roomtemperature.
Of courseno semiconductor material is ideal and GaAs and its related compounds
comewithsomenatural drawbacksthatmustbeaccommodatedinthedesignof success-
ful products. Notabledisadvantages of GaAs arearelatively lowthermal conductivity
andtheabsenceof anativeoxideor similar passivant. Theformer issuelimitstheability
of GaAsdevicestodissipateheat throughthesubstratethereby makingthermal design
anareaof particular attentioninthedesignof high-power products. Thesecondissue
of imperfect passivationgivesrisetoslow-state phenomena. A number of alternative
terms areusedhereto describethemechanisms andtheir effects. Theterms: traps,
deep-levels, surface-states, anddispersion areall commonly used. They refer to
anundesirablefeatureof semiconductor deviceswherecrystallineimperfectionsresult
inelectronor holestates that areof intermediatedepthintheband-gapsuchthat they
2.1 Introduction 45
interact withthefreecarriersat noticeablelevelsbut at rateswhichareslowcompared
to the intended transistor response. The result is that, in addition to the desired fast
response, therefollows asecondary slowtail that can compromisethedeviceperfor-
manceinavarietyof ways. Effectivecontrol of dispersivephenomenainGaAsdevices
tookmanyyearstoadequatelyresolve. Bothof thesetopicsareaddressedinmoredetail
inthesectionstofollow.
2.1.2 The Schottky barrier gate and the MESFET
A classof transistor whichisverysuitablefor GaAsistheSchottky-gateFET. Thisform
of FET is anatural choicefor GaAs because, unliketheMOSFET, theSchottky-gate
FET can accommodatean imperfectly passivatedsurface. A host of devicevariations
havefollowedsincebut theprimary deviceof thefamily is theMEtal-Semiconductor
FET or MESFET. ThisisessentiallyaSchottkybarrier gatebetweentwoohmiccontacts
onalayer of n-typesemiconductor that formsaconductingchannel. A Schottkybarrier
is formed when ametal is brought into contact with asemiconductor surface. Given
asuitabledifferenceinmaterial work-functions, chargeredistributioninthesemicon-
ductor occurs whichdepletes theadjacent semiconductor regionof its mobilecarriers
(inthemanner of aone-sidedp

-njunction). Theextent of thedepletiondepthiscon-


trollableby theamount of bias across thejunction, and by this means abias applied
to the Schottky gate will modulate the available charge and hence the current in a
FET channel. The rate that the junction can control the current limits the frequency
response of the device. The limiting processes here are the RC time-constant of the
gate junction and the time for the carriers to travel along the channel. The relevant
keymaterial propertiesarethemobilityandthesaturatedvelocity, andinhigh-mobility
materialssuchastheGaAsfamily thenthesaturatedcarrier velocity isthedominating
factor.
2.1.3 The Pf
2
limit
The maximumpower obtainable froma transistor manufactured froma given semi-
conductor material is dependent on the frequency at which the device is required to
operate. Therelationshipof power withfrequency isaninverse-squarelaw, sometimes
referredto as thePf
2
= constant limit [6]. Thefactors that determinethis relation-
ship are the breakdown eld, the saturated carrier velocity, and the physical size of
thedevicefootprint. Therms power density, P

, obtainablefromasinusoidally driven
transistor with the peak current density J
max
and voltage swing of V
min
to V
max
is
givenby:
P
/
=
J
max
(V
max
V
min
)
8
(2.1)
For aFET withitsspeedlimitedbytheelectrontransit-time, travelingat thesaturated
velocity:
sat
, over characteristiclengthL, withapeakvoltagelimitedbythebreakdown
46 GaAs FETs physics, design, and models
eldE
b
, andapproximatingV
min
tozero, then:
P
/
=
J
max
E
br
L
8
(2.2)
=
J
max
E
br
:
sat

8
(2.3)
=
J
max
E
br
:
sat
16 f
T
(2.4)

1
f
T
(2.5)
wheref
T
isthecut-off frequency. ThissameanalysisisusedtodenetheJ ohnsongure
of merit for asemiconductor material J FOM [7]:
J FOM =
E
br
:
sat
2
. (2.6)
Returningtoequation(2.4), J
max
iscrudelyof theorderof 500mA/mmformostavours
of GaAspower FET irrespectiveof operatingfrequency. However, theabilityof aFET
todeliver J
max
acrosstheentiregateperiphery of adevicediminishesasthefrequency
increases. This is primarily due to de-phasing of the input signal along gate ngers
andacrossthemultigatemanifolds. Consequently, theremoteregionsof thedeviceare
drivenprogressively out of phasecomparedto theregionintheimmediatevicinity of
thegateterminal therebyresultinginanetreductionincurrentdeliveredtothedrain. In
orderforthephasingeffectstoremaininvariantthephysical devicewidthmustbescaled
inversely withthefrequency thereby givingafurther 1,f contribution. Incombination
withequation(2.5) theoverall effect onthetotal power, P, isthen:
P
1
f
2
T
. (2.7)
Clearly this is asimplication which omits ahost of other factors such as RC losses,
matchingeffectsandthermal considerations, but it doescapturethedominant limiting
processesfor awell-designedmicrowavepower FET. Validationof equation2.5isgiven
inFigure2.1whichshowstheratedbreakdownvoltagesversusprocessf
T
for avariety
of commercially availablepower FET processes of thevarieties described in thenext
section.
2.1.4 Types of GaAs FET
Four key GaAs FET variants are compared in Figure 2.2. The rst type, shown in
Figure 2.2a is the MESFET consisting of a Schottky gate controlling the current in
a simple uniformly-doped channel. The rst devices were ion-implanted structures
and this approach becamethestandard manufacturing techniquefor GaAs transistors
for a number of years. The MESFET was gradually rened with enhancements that
included optimizing the doping proles, the use of epitaxially grown layers, and the
development of recessedgatestructuresfor enhancedbreakdown. Theadvent of band-
gapengineering introducedAlGaAsasapartner material. A varietyof heterostructure
2.1 Introduction 47
30 25 20 15 10 5
0
20
40
60
80
100
BVgd V
f
T


G
H
z
0.5 m P
0.3 m P
0.25 m P
0.15 m P
0.25 m P
0.15 m P
0.25 m P
0.25 m P
0.15 m P
0.15 m P
0.5 m P
0.5 m M
0.5 m H
0.5 m P
0.5 m M
Figure 2.1 Breakdownvoltage frequencyrelationshipfor commercial power FET processes
(M: MESFET, H: HFET, P: pHEMT).
Figure 2.2 KeyGaAsFET devicetypes; (a) metal-semiconductor FET (MESFET) (b) doped-
channel heterojunctionFET (HFET); (c) high-electronmobilitytransistor (HEMT);
(d) pseudomorphichigh-electronmobilitytransistor (pHEMT).
48 GaAs FETs physics, design, and models
FET (HFET) developmentsthenfollowed. Themost straightforwardHFET, depictedin
Figure2.2b,usesawidebandgapAlGaAsspacerlayerthatspacestheGaAschannel from
thegate[810]. Thisstructureconstitutesaneffectivepowerdevicewithexcellentpower
andlinearity performance[11]. Thetransconductanceachievablefromthisstructureis
relativelymodest, however avaluableattributeisthenear constant valuewithgatebias
[8, 12, 13] that isachievablewhichisimportant for improvinglinearity.
Anumberof developmentsof theHFEThavebeeninvestigated, typicallyinvolvingthe
useof InGaPasanalternativewidebandgaplayerandwithInGaAsasthedopedchannel
layer. Reference[14] summarizesthebenetsanddrawbacksof anInGaP barrier layer
includingitsabsenceof DX centersandthat it islesslikelytosuffer surfaceoxidation.
However, it is also has aless advantageous conduction bandalignment thanAlGaAs,
anditsuseasanHFET barrier layer hasnot foundwidespreaduse.
Thefurther devicedevelopmentsdescribedherefocusonimprovementof thechannel
properties so as to enhance the frequency performance. A key development was the
AlGaAs/GaAs high-electron mobility transistor (HEMT) or modulation doped FET
(MODFET) [15]. TheseareequivalentnamesforadevicestructureshowninFigure2.2c
whichavoidsdopingthechannel directlyandinsteaddopestheadjacent AlGaAslayer.
MobileelectronsthenpopulatetheGaAschannel but, at loweldsat least, suffer much
lessscatteringasthedopingimpuritieshavebeenseparatedfromtheconductionchannel.
HEMTs of this type have been superseded by the pseudomorphic device discussed
below, but theconcept was akey stepping-stoneinthedevelopment of themicrowave
FET.
Thepseudomorphic-HEMT (pHEMT) showninFigure2.2disasignicant enhance-
ment of theHEMT that introduces thebenet of anInGaAs channel [1618]. InGaAs
is a narrow band-gap material with excellent electron transport properties. The rst
incarnation of the pHEMT was a single heterojunction device with an AlGaAs bar-
rier and charge supply-layer above the InGaAs channel. As material growth quality
improved, asecondAlGaAslayer beneaththechannel wasaddedwhichprovidesbetter
chargeconnement and hencehigher current capability. Theprincipleof modulation
doping for thedouble pHEMT is illustrated in Figure2.3 which shows theidealized
banddiagramfor apHEMT channel. However, as describedabove, InGaAs possesses
alarger latticespacingthanGaAs andAlGaAs whichlimits theindiummolefraction
to around20%. This is afairly modest indiumfractionwiththeresult that theexcep-
tionally high mobilities that arethenormin thehigher indiumcomposition channels
of thelattice-matchedIn
0.52
Al
0.48
As/In
0.53
Ga
0.47
As/InPHEMT devicesarenotachieved
in GaAs-based structures. However, it should be realized that the key device benet
achievedinAlGaAs/InGaAs devices comes not fromthefundamental mobility of the
channel material, but rather fromthe separation of the carriers in the channel from
their dopant atoms intheadjacent wideband-gapregion. Inthis respect theincreased
conduction band step introduced by the use of InGaAs for the channel layer is very
benecial. Althoughtheinnatechannel mobility is not improvedabovepureGaAs, it
is substantially improvedabovedopedGaAs andinAlGaAs/InGaAs pHEMT devices
channel mobilitiesexceeding6000cm
2
/Vsareobtained. For aMESFET withadirectly
doped channel then the achieved mobilities would typically be less than half that of
2.1 Introduction 49
+
_ _
+
AlGaAs InGaAs
AlGaAs
E
F
Doping plane Doping plane
High mobility
channel
Figure 2.3 IdealizedpHEMT banddiagramandmodulationdopingprinciple.
thepHEMT value. Thisimprovement iscomfortablysufcient toensurethat thedevice
speedof thepHEMT isnot signicantlylimitedbythechannel mobility.
The semiconductor band-diagrams that correspond to the key device types of
Figure2.2aregiveninFigure2.4. Thediagramswerecalculatedusingapublic-domain
Poisson-Schr odinger equationsolver [19] andthegureshowstheequilibriumconduc-
tionandvalencebandsolutions for thezero-bias conditiontogether withtheresulting
electronconcentration. Thefour banddiagrams highlight thedifferences inthenature
of theconnement of thechannel electrons (electrondensity n) intheregionbetween
theSchottkybarrier of thegateterminal ontheleft-handsideandthemidbandpinning
of the Fermi-level of the insulating substrate that occurs off-scale on the right-hand
side of the plots. For the MESFET of Figure 2.4a the channel charge distribution is
essentially that of thebulk semiconductor layer withaneffectivewidthmodulatedby
theextent of thedepletionof theSchottky gate. Theapplicationof negativegatebias
further increasestheenergydifferencebetweentheFermi-level andtheconductionband
andtherebyextendsthegatedepletionreducingtheavailablechargeinthechannel. For
thesituationwhereapositivebiasisappliedtothegatethenbythereverseprocessthe
depletiondepthreducesandtheMESFET channel widenstowardsthegate. Theuseof
negative, zeroandpositivegatebiasesforaMEFSET structureareshowninFigure2.5a.
TheHFET structurewithabanddiagramshowninFigure2.4bdiffersfromtheprevious
caseduetotheAlGaAs spacer layer beneaththegate, andthemobilechargefromthe
dopinginthis layer is transferredto the(also doped) channel whereit is energetically
favourabletoremain. A changetothegatebiasvoltagemodulatesthedepletionedgein
thesamemanner astheMESFET, butnowthepresenceof theheterojunctionprovidesa
constraint ontheminimumdepthof thedepletionlayer edge. AsshowninFigure2.5b,
under positivegatebiasconditionstheelectronconcentrationremainslargely conned
bytheheterojunctionbarrier andso, unliketheMESFET, theupper extent of depletion
layer edgeisconstrainedanddoesnot movesignicantlytowardsthegate. TheHEMT
50 GaAs FETs physics, design, and models
E
c
E
f
E
v
n
E
c
E
f
E
v
n
E
c
E
f
E
v
n
E
c
E
f
E
v
n
50 0 100 150 200
2.0
1.5
1.0
0.5
0.0
0.5
1.0
Depth (nm)
E
n
e
r
g
y

(
e
V
)
0.2
0.4
0.6
0.8
1.0
x10
18
n

(
c
m

3
)
50 0 100 150 200
2.0
1.5
1.0
0.5
0.0
0.5
1.0
Depth (nm)
E
n
e
r
g
y

(
e
V
)
0.2
0.4
0.6
0.8
1.0
x10
18
n

(
c
m

3
)
50 0 100 150 200
2.0
1.5
1.0
0.5
0.0
0.5
1.0
Depth (nm)
E
n
e
r
g
y

(
e
V
)
0.2
0.4
0.6
0.8
1.0
x10
18
n

(
c
m

3
)
50 0 100 150 200
2.0
1.5
1.0
0.5
0.0
0.5
1.0
Depth (nm)
E
n
e
r
g
y

(
e
V
)
0.2
0.4
0.6
0.8
1.0
x10
18
n

(
c
m

3
)
(a) (b)
(c) (d)
Figure 2.4 Zero-biasbanddiagramsandelectrondensitiesfor keyGaAsFET devicetypes;
(a) metal-semiconductor FET (MESFET); (b) doped-channel heterojunctionFET (HFET);
(c) high-electronmobilitytransistor (HEMT); (d) pseudomorphichigh-electronmobility
transistor (pHEMT).
structureof Figure2.4c has asimilar AlGaAs/GaAs heterojunction to theHFET just
discussedbutwiththedifferencethattheGaAsregionisundoped. Theband-bendingof
thejunctioncreatesasmall well that ispopulatedwithcarriersfromthedopedAlGaAs
region. Themobilityof thechannel isintendedtobethat of theintrinsicmaterial asthe
scatteringfromthedopant ionsiseliminatednowthat theyarespatiallyseparatedfrom
thepath of themobileelectrons. However, theconnement capability of this modest
well isquitelimitedandthisstructureisthereforenot effectiveasapower device. This
issueisresolvedinthepHEMT of Figure2.4dwithanInGaAschannel. Herethedeeper
conductionbandoffsetbetweenAlGaAsandInGaAsprovidesahighdegreeof conne-
ment andtheuseof adoubleheterojunctionwithdopingprovidedfromboththeupper
andlower AlGaAs barrier layers achieves ahighsheet-chargedensity. Also employed
2.2 Power device physics 51
E
c
E
f
E
v
n
E
c
E
f
E
v
n
50 0 100 150 200
2.0
1.5
1.0
0.5
0.0
0.5
1.0
Depth (nm)
E
n
e
r
g
y

(
e
V
)
0.2
0.4
0.6
0.8
1.0
x10
18
n

(
c
m

3
)
50 0 100 150 200
2.0
1.5
1.0
0.5
0.0
0.5
1.0
Depth (nm)
E
n
e
r
g
y

(
e
V
)
0.2
0.4
0.6
0.8
1.0
x10
18
n

(
c
m

3
)
(a) (b)
Figure 2.5 Effect of variationof gatebiasondevicebanddiagramsandelectrondensityfor (a)
MESFET and(b) HFET. Solidlineisthezerogatebiascondition, longdashisfor reversebias
andshort dashisfor forwardbias.
hereis theuseof deltadoping. Inconventional dopingthedopant atoms areincluded
uniformlyinthecrystal at modest concentrations. However, indelta-dopingthesilicon
dopant atoms aredepositedinacontinuous thinlayer just afewatoms deep. This has
benets to deviceoperationinthat thedopant atoms areall very closeto thechannel
ensuring maximumtransfer of electrons into thechannel theso-called modulation
efciency. A further benet is that it is easier to managetheMBE growthreactor to
dopeinthisway.
2.2 Power device physics
2.2.1 The device IV characteristic and loadline
AnidealizedIV characteristic is showninFigure2.6. Thekneevoltage, V
knee
, is the
voltagewherethecurrent saturates. Belowthispoint thedeviceisavoltagecontrolled
resistor andaboveV
knee
theDC current issaturatedandideally isindependent of drain
voltage. Themaximumcurrent I
max
istypicallydenedjust abovethekneevoltageand
isthemaximumcurrentthedevicecansupplybeforethegatejunctionbecomesforward-
biasedandstartstodrawgatecurrent. Another keyparameter for apower deviceisthe
breakdown voltageas this limits thepeak voltagethat thedeviceload-linecan swing
upto. Thepinch-off voltageV
p
isthegatevoltagerequiredtoturn-off thedraincurrent
(typicallytoathresholdvalueof order 1mA/mm).
Thedetailedbehavior of theFET IV characteristic is determinedby thecombina-
tion of theSchottky gatedepletion dependenceon thegate-channel potential and the
velocity-eldcharacteristicof thesource-drainchannel. Inrealitythisisacomplexand
interdependent 2D problem. However, for submicron gateGaAs-based devices where
theelectronvelocity saturatesunder thegateover muchof theIV spacethenauseful
52 GaAs FETs physics, design, and models
V
ds
I
ds
V
knee
V
max
I
max
I
dss
Q
A
Q
B
V
gs
> 0
V
gs
= 0
V
gs
< 0
V
gs
= V
p
Figure 2.6 Ideal FET DC IVcharacteristicwithclassA andB loadlines.
simplied description can be considered. In this model the current is determined by
thesaturatedelectronvelocityandthenumber of availablecarriers. Thefractionof the
maximumchannel current that is available is determined by the extent that the gate
depletionregionextends downinto thechannel. Animportant benet of thesaturated
velocity mode of device operation is that over the main part of IV space the drain
currentisgenerallyamorelinear functionof gatevoltage. Goodlinear behavior maybe
pictorially observedfromthefamily of I
d
V
d
curves that makeupadeviceIV graph
havingapproximateequal spacingasV
g
isvaried. Suchbehavior isunlikethelong-gate
or gradual channel caseof thetraditional J FET wheresaturation is alow-eld and
constant mobilityprocessresultinginasquare-lawdependenceof draincurrent ongate
voltage[3, 20].
It is readily apparent that a good power device has high peak current I
max
, a high
breakdowncapability, alowV
knee
andequally spacedcurves ontheIV characteristic.
Figure2.6illustrates themaximumpower class A and class B load-lines that may be
supported on theidealized IV characteristic. Theclass A loadlineis thesimplest to
understandandthegureshowstheIVlocusswingingfromthepeakcurrentvalueI
max
andminimumvoltagevalueV
knee
toamaximumvoltageat zerocurrent limitedby the
devicebreakdown. TheresultantquiescentbiaspointisQ
A
. Themaximumoutputpower
for theidealized class A situation is given by equation (2.1). Theclass B loadlineis
achievedbyreducingtheoperatingcurrenttobiaspointQ
B
. Thesameloadresistanceis
requiredandasimilar maximumpower isdeliveredbut withhigher efciencyachieved
duetoareducedDC dissipation.
Forareal devicetheIVcharacteristicdepartsfromtheideal inanumberof important
ways. Figure2.7highlightssomekeyfeaturesof amorerealisticDC IVcharacteristic
whichwouldbeobservedusingaslow-sweepcurvetracer. Thegureillustratesthermal
droop where self-heating of the device causes a reduction of electron velocity and
mobility. Thespacingof thelines of constant gatevoltageis also no longer ideal and
exhibitscompressionat thelimitsof thegatevoltagerange.
2.2 Power device physics 53
V
ds
I
ds
V
knee
BV
ds
I
max
I
dss
V
gs
> 0
V
gs
= 0
V
gs
< 0
Figure 2.7 Practical FET DC IVcharacteristicexhibitingbreakdown, thermal droop, and
transconductancecompression.
7 6 5 4 3 2 1 0
0
20
40
60
80
100
V
d
V
I
d

m
A
DC
Q
A
(6V, 0V)
Q
AB
(6V, -0.3V)
Q
B
(6V, -0.6V)
Q
A
Q
AB
Q
B
Figure 2.8 MeasuredpulsedIVcharacteristicfor 0.25mpHEMT processat classA, AB, andB
biaspoints.
2.2.2 The dynamic IV characteristic
A further departurefromtheideal characteristic occurs whenthedynamic responseis
considered. Figure2.8showstypical pulsedIVcharacteristicsthathavebeenmeasured
with a commercial system[21, 22]. A set of dynamic IV plots are overlaid onto a
conventional DC IVmeasurement. Inthepulsedcasethedeviceisbiasedat quiescent
biaspointsindicatedonthegureandshort, low-dutycyclepulsesfromthisbiaspoint
arethen used to exploretheIV planeand map out thecharacteristic. Thebias point
should be chosen to be typical of the intended operating point with the aimthat the
measureddynamic IV will replicatetheRF behavior of thedevice. As is apparent in
Figure2.8thedynamicbehavior differssignicantlyfromthestaticcase. Thedifference
54 GaAs FETs physics, design, and models
is, for themost part, duetothepresenceof slowtraps inthesemiconductor. Traps are
deep-level statesthatcancaptureandretainelectronsor holesfor extendedperiods. The
possiblecauses andlocations arenumerous [2326] andincludetraps at theun-gated
semiconductor surfaceandthesubstrateinterface, DX levelsinAlGaAs, andfreeions
inpassivatingmaterials. A conciseoverviewof thephenomenafor bothGaAsandGaN
devicesisgiveninreference[27].
Thetrapsareenergystatesthathaveenergiessufcientlydeepintothesemiconductor
forbiddenbandthatthelikelihoodof acarrierinteractingwiththestateisrelativelysmall
andwhenacarrierdoesoccupysuchastatethenafurtherlowprobabilityeventisneeded
inorder for ittobereleasedagain. Theresultisthatthelifetimeinadeep-level statecan
bequitelong, and timeconstants of microseconds to milliseconds arecommonplace.
Theimpactof thetrappedchargeinthedeepstatesisthattheassociatedeldaffectsthe
passageof thefreecarriersinthechannel thereby modifyingthedevicecharacteristic.
Becausethecaptureandreleasetimeconstantsof thetrapsareverylongcomparedtothe
periodof amicrowavesignal, theRF (carrier) signal andtrapoccupancydonot directly
interact. However, thetrapoccupancyisaffectedbythemeanbiasconditiongivingriseto
thesituationthatthedynamicIVchangesslightlyasthemeanbiaspositionchanges. A
useful waytothinkof thisisthatthereisntauniqueanddenitiveIVcharacteristicfor
aparticularFET, butratherthereisaslightlymodiedoneforeverymeanbiascondition.
A helpful physical model isthatof theeldassociatedwiththetrappedchargeactingas
aslowlyvaryingvirtual gate that modiestheeffect of thephysical gate. A particular
area of concern for trapping effects is the semiconductor surface. Without a suitable
passivant material thereareinevitably alargenumber of surfacestatespresent. Indeed,
itisworthnotingthatthedensityof availablesurfacestatesistypicallycomparablewith
theintended doping of thedevicechannel. It thereforebecomes an important device
designtasktominimizetheirimpactonthedeviceperformance. Thisisachievedbysuch
measuresaskeepingtheetchedsurfaceareatoaminimum, maximizingthedistanceof
suchareas tothechannel andtheuseof charge-screeninglayers toisolatethechannel
fromchanges to thesurfacepotential. In contemporary optimized FET structures the
trappingeffectshavebeenaddressedtoasubstantial degree. Improvedmaterialsgrowth
quality, improvedwaferprocessingtechniques, andadvancesindevicedesigntechniques
havereduced thedensity of availablestates and mitigated their impact on thedevice
response. Nevertheless, discernable slow-state effects are the normand these effects
haveanimpactthatcanlimitthedeviceperformanceandintroducedifcultiesindevice
characterizationandmodeling.
2.2.3 The consequences of trapping effects
Thephysical processes described abovegiveriseto amultitudeof observabledevice
effectstobeminimizedbythedevicedesignerandaccommodatedbythecircuitdesigner.
Theeffectsaresummarizedinthefollowingparagraphs.
Reducedoutput power: For devicesoperatingasanamplier andbiasedassuch, the
resulting equilibriumtrapping statereduces thepeak devicecurrent dueto increased
surfacedepletion and increases theeffectivekneevoltageby increasing thedynamic
2.2 Power device physics 55
(a) Gate lag (b) Drain lag
Time Time
I
d
Figure 2.9 Typical GaAsFET gateanddrainlagresponses; (a) gatelag; (b) drainlag.
channel access resistances. These modications to the device IV are illustrated in
Figure2.8andreducethemaximumoutput power that adevicecandeliver.
G
m
andg
ds
dispersion: Measurements of devicetransconductance(g
m
= dI
d
,dV
gs
)
and output conductance (g
ds
= dI
d
,dV
ds
) with frequency are observed to undergo a
transitionfromtheir DC values, andthis variationwithfrequency is calleddispersion
[25, 26, 28]. Thetransitionfrequency rangeis typically inthe1kHz to 1MHz range
with thetransconductancedecreasing fromtheDC valueand theoutput conductance
increasing fromtheDC value. Historically, dispersion measurements wereaprimary
assessment tool for devicetrapping effects beforepulsed IV assessment cameto the
fore.ForcontemporaryGaAspowerdevicestransconductancedispersionisusuallysmall
andthedispersionof theoutputconductanceisthedominantphenomena. Thisisclearly
observed in FET DC IV characteristics such as Figure2.8. Thespacing of thelines
of constant gatevoltagehas remained largely invariant indicating minimal difference
in transconductance. However the increase in the slopes for the pulsed characteristic
demonstratestheincreasedoutput conductanceexperiencedbyadynamicstimulus.
It shouldberealized, however, that thedifferencebetweentheDC anddynamic IV
slopesdoesnot indicatethat slow-stateshavecompromisedtheoutput conductancefor
theRF signal. ItismoreaccuratetoviewthedynamicIVmeasurementasrevealingthe
underlyingtrue output conductancethat is obtainablefromtheshort gatestructures
thataretypicallyemployedinGaAs-baseddevices. For theDCcase, theaforementioned
underlyingIVismodiedbychangesintheequilibriumtrapoccupationswhichevolve
withthe(slowly varying) biasvoltageinsuchamanner soastosupplement theaction
of the gate and so reduce the resulting output conductance. The mechanismcan be
visualizedsuchthat asthedrainvoltageisincreasedthentheamount of trappedcharge
in the vicinity of the gate also increases and the eld associated with the trapped
chargeactsinconcertwiththatof theSchottkygatebiastherebyhelpingtosuppressthe
increaseindraincurrent that wouldotherwiseoccur.
Gateanddrain lag: Gateanddrainlagaretermsthat describethedelayedresponse
of thedraincurrent tochangesinthegateanddrainvoltages, respectively[26, 2932].
Typical gatelaganddrainlagresponsesareillustratedinFigure2.9andshowhowthe
dominant fast responseis followed by a slow tail. Thetail can persist for timescales
56 GaAs FETs physics, design, and models
rangingfromafewmicrosecondstomillisecondsdependingonthedetail of thedevice
construction and the fabrication techniques employed. The plots relate to the same
physical test asisperformedinthepulsedIVplot of Figure2.8; however, thex-axison
theplot isnowtimerather thandrainvoltageandthetracesshowthetimeevolutionof
just onesamplepoint onthat IV plot. Withgate-lagthedrainvoltageisheldconstant
andthegatevoltageis steppedfromthequiescent valueto therequiredsamplepoint,
andwithdrainlagthegatevoltageisheldconstant andthedrainvoltageisstepped.
Memoryeffects: Inadditiontothemodicationstothedevicecharacteristicsdescribed
sofar, thedispersionmechanismalsoresultsinhysteresisinthedeviceresponse, or in
otherwordsthedevicecanexhibitamemoryof arecentlyappliedstimulus. Asdiscussed,
the trapped carrier population is a function of the mean bias condition. For a high-
frequency continuous-wavesignal wheretheperiodof theRF is short withrespect to
thetraptime-constant thenanequilibriumtrapoccupationwill beachievedandastable
dynamicdevicecharacteristicisobserved. However, for low-frequencysignalsor, more
likely, an RF carrier modulated with amodulation frequency that is comparablewith
thetraptime-constant, thenthetrapoccupationcanbeinuencedbythislow-frequency
variation. Theresult is that theRF characteristic can besubtly modied by thelower
frequency component of thesignal thereby givingadependenceontherecent history.
Of course this is a familiar problemfor all semiconductor devices even if trapping
mechanisms wereto becompletely controlled as thermal timeconstants havesimilar
consequences. The impact of trapping effects on a modulated signal is simulated in
reference[33].
I naccuracy in large-signal models: Traditional device models are based on IV
characteristicsmeasuredat DC. However, ashasbeenshown, thetrappingeffectsgive
risetodynamicIVswhicharenoticeablydifferent fromthestaticone. Thismeansthat
models whichsimply usethestatic characteristic do not accurately predict thedevice
performance. Invariably, thereal devicewill providelesspowerandexhibitlessgainthan
theDC-derivedmodel. Techniquesfor thegenerationof improvedlarge-signal models
areaddressedlater inthechapter.
I ncreasedbreakdownvoltage: Sofar thelist hasgivenaseriesof detrimental effects
that arise fromthe presence of surface states. However, they do have an important
benecial impact on breakdown voltage. As will be discussed in more detail in the
next section, theassociatedsurfacechargelocatedinparallel withthechannel has the
positive benet of assisting to spread the electric eld in the gate-drain region over
alonger distancethereby reducing thepeak eld developed and henceincreasing the
devicebreakdownvoltage.
Kink effect: The list ends with a phenomenon that has long been observed where
traces on the IV characteristic can exhibit a kink to a higher drain current as if
the gate bias was suddenly adjusted higher as the drain voltage is swept. Similar
effects have been reported in various kinds of semiconductor transistor, such as sil-
icon MOSFETs, GaAs-MESFETs, doped channel HFETs, AlGaAs/InGaAs-HEMTs,
andInAlAs/InGaAs-HEMTs[34].
Various mechanisms havebeen explored and it is clear that thereareanumber of
different kink-effect processes that may be present depending on the detailed device
2.2 Power device physics 57
G
D
S
BV
gs
BV
gd
BV
ds
Figure 2.10 Breakdownvoltagedenitions.
construction. It ispossibletoobservemechanismsthat affect theDC characteristicbut
arenot observed at RF [35], whilefor other structures they may beobserved also or
exclusively inthedynamic characteristic [34, 36]. Thepossiblemechanisms that may
be involved include eld ionization of traps where an increase in the drain voltage
induces releaseof trappedelectrons thereby allowingtheassociatedchannel depletion
to lessen [34, 37]. Other processes involvethepresenceof holechargegenerated by
impact ionization. Hereabuildupof holechargeat thesourceendof thegatecangive
riseto aparasitic bipolar effect that can causecurrent injection thereby reducing the
effectivesourceresistance[38]. It isalsopossiblethat associatedchangeinthechannel
potential reduces theeffectivepinch-off voltage[39]. A further mechanismsuggested
by 2Dsimulationisfor impact ionizationgeneratedholestointeract withandpartially
dischargesurfaceelectrontraps thereby wideningthechannel [40]. Other simulations
indicate a possible contribution fromredistribution of the 2D electric eld when the
lateral extensionof gatedepletionreachestheedgeof therecess[36].
2.2.4 Device breakdown
Thethreebreakdownconditionsgenerally quotedfor FETsaretheGate-Source, Gate-
Drain and Drain-Source breakdown voltages BV
gs
, BV
gd
, and BV
ds
, respectively, as
illustrated in Figure2.10. Thetypical denition employed is thevoltagefor which a
currentof 1mA/mmof gatewidthisobserved. BV
gs
andBV
gd
areso-calledtwo-terminal
tests(i.e., withthethirdterminal oating) and, withthenotationusedhere, arenegative.
Inpowerdevicesthegateisusuallypositionedasymmetricallytooptimizethegate-drain
breakdown value. BV
ds
is athree-terminal test with thegatebias set to asufciently
negativevaluesoastoensurethatthedeviceispinchedoff. BV
ds
ispositive. Itiscommon
for only two-terminal teststobequotedondata-sheets or inwafer acceptancecriteria.
However, the three-terminal drain-source breakdown is also an important parameter,
particularlysofor power devices, asthiscongurationcorrespondstohowthedeviceis
actuallyused. Different physical mechanismsaregenerallyobservedfor thebreakdown
processesexperiencedunder twoandthree-terminal conditionswiththeresultthat BV
ds
can be signicantly less than might be expected froma simple consideration of the
58 GaAs FETs physics, design, and models
combinationof thegate-drainbreakdownandtheappliedgatevoltage(i.e., V
gs
-BV
gd
)
[41].
2.2.5 Breakdown mechanisms and optimization
Signicant attention has been paid to theoptimization of breakdown performancein
order tooptimizeoutputpower, andbreakdownperformancefor power deviceshasbeen
developedtothepointthatpower GaAsFET devicesareusuallythermallylimitedrather
thanbeinglimitedby devicebreakdown. A number of physical processesareinvolved
intheevolutionof thebreakdownprocessanddependontheprecisedeviceconstruction
andontheoperationof thedevice. Thekeyprocessesdescribedhereare[4247]:
r
thermioniceldemission(TFE) over thereverse-biasedgatebarrier;
r
tunneling through the gate barrier which narrows as the device is heavily reverse-
biased;
r
impact ionizationinthechannel;
r
parasiticbipolar effect;
r
electriceldspreadingduetothesurfacepotential.
Theseprocessescanall operateinconcert togiveavariety of interestingbehaviors. A
typical evolutionof abreakdowneventbeginsinahigh-eldlow-currentconditionwith
increasedgate-drainleakagecausedbyeld-emissionof currentover thereverse-biased
gate-drainbarrier. Itisnormal for theeldtobesohighthatthebarrier becomesthinned
whichcausestheeld-emissiontobeenhancedbyquantum-mechanical tunneling. Ener-
geticelectronscanthenndthemselvesinjectedintothehigh-eldchannel withexcess
energy. Theelectrons will relax their energy by various means and onepossibility is
impact ionization whereelectron-holepairs arecreated caused by collisions with the
lattice. Impact ionizationisself-reinforcinginahigh-eldchannel astheelectron-hole
pairsproducedcangoontoseedother ionizationeventsinanavalanchemultiplication
process. Breakdowntendstobeafairlygradual processat rst astheappliedvoltageis
increasedandsteadily over several voltsthetotal breakdowncurrent becomesprogres-
sively dominatedby theimpact avalanchecomponent. Detailedstudies analyzing this
behavior inGaAsFETshaveusedthedifferingtemperaturedependenceof theTFE and
impact ionization processes to identify the relative contributions.
1
Thepictureso far
thenis that of energetic electrons defeatingthegatebarrier, seedingimpact ionization
andthebreakdowncurrentrunningawaywithapositivefeedbackmechanism. However,
thisisnotnecessarilyquitetheendof thestory. Onsomedevicesitispossibletoobserve
a snap-back effect where, as a certain current threshold is crossed, the device can
no-longer support thebreakdown voltageresulting in thebreakdown characteristic of
Figure2.11. Inthiseventthenthevoltagecollapsestoalowvalueandthedevicecurrent
rises markedly [48]. This process may or may not be fatal depending on the device
1
TFE has a positive temperature coefcient which is to say the current over the barrier increases with
temperature[42, 46]. Conversely, inGaAsdevices, thetemperaturedependenceof theionisationcoefcients
actsintheoppositesense.
2.2 Power device physics 59
V
d
I
d
Snap-back
Impact ionization
TFE/Tunnelling
Figure 2.11 Breakdownsnap-backcausedbyaparasiticbipolar effect.
andthecircuit. Themechanismgivingriseto thesnapback is aby-product of impact
ionization wheretheresulting equilibriumholeconcentration can induceaparasitic
bipolar effect (PBE). Oncecreated, theionized holes can followanumber of paths:
Theycanbecollectedbythegateterminal andaddtothegatecurrent, theycanescape
intothesubstrate, or they canowtothesource. A commonunderstandingof thepar-
asitic bipolar effect is that theholes collecting in thesubstrateact as aparasitic back
gateandhavetheeffect of openingthechannel fromtheback [49, 50]. However, in
aheterojunctionFET holestendtobeconnedinthechannel bythevalencebandwell
andinthis casemay not readily owinto thesourcecontact or substrate. Insteadthey
will drift to thesourceregionandanequilibriumholechargeis developedthere. This
localizedpositivechargefavoursinjectionof electronsfromthesourcecontact intothe
channel thereby inducing an increased drain current by another means [49]. Parasitic
bipolar effectsarewell knowninsilicondevices[51, 52] but appear lesssointheGaAs
community.
2.2.6 Comments on GaAs FET breakdown ratings
GaAs FET circuits are often designed to operate quite close to the transistor rated
breakdownlimitswithsafetymarginslessthanaretypicallyemployedwithother tech-
nologies. A number of factors makethis asafethingto do. Thenatureof GaAs FET
breakdownis suchthat it is typically quitegentleinits onset andresults insignicant
circuit performance reduction before device degradation is observed [43]. Also, the
typical dominant agingmechanismof devices operatingunder high-eldconditions is
ahot-electroninducedsurfacedegradation. Thesurfacedamageleads toasubsequent
increaseintrappedsurfacechargecausingincreasedspreadingof theelectric eldand
henceanincreaseinthebreakdownvoltage[53]. Thisso-calledbreakdownwalkout
providesafail-safemechanismwherethefailureprocesseffectivelyhardensthedevice
against further degradation.
60 GaAs FETs physics, design, and models
C
gd
R
d
L
d
R
g
L
g
R
i
C
gs
+
--
R
s
L
s
g
m
R
ds
C
ds
g
m
= g
m0
.e
j
Figure 2.12 CommonGaAsFET equivalent circuit network.
A further relevant phenomenon is theobservation that theRF breakdown of GaAs
FETscanbefrequently higher thantheir DC breakdowndatawouldsuggest [54]. The
literatureisnotcomprehensivebuttheperceptionof devicessafelyoperatingatvaluesnot
commensuratewiththeirDCbreakdownvaluesisacommonone. A popularexplanation
is that of avalanchedelay [55], whichcaninhibit theonset of breakdownbecausethe
periodof amicrowavesignal is typically comparablewiththecharacteristic delays of
theavalancheprocess. Anattempt toquantifytheeffect wasgivenbyShrikov[56] who
measured amodest onevolt enhancement to thedrain-sourcebreakdown. Snap-back
effects(if present) may alsobetooslowtorespondtotheRF signal whichinthat case
couldbeexpectedtogiveanapparent RF breakdownenhancement.
2.2.7 The FET equivalent circuit
Theusual small-signal equivalentcircuitnetworkusedtorepresentaGaAsFETisshown
inFigure2.12. Theprimeelementsarethevoltage-dependentdraincurrentgenerator of
transconductanceg
m0
andthegatecapacitanceC
gs
acrosswhichthecontrollingvoltage
is developed. The remaining elements are unavoidable parasitic components whose
presencedegradesthedeviceperformanceandsowhosevaluesareminimizedasmuch
aspossibleinthedevicedesignandthefabricationapproachesemployed.
Figure2.13illustrates themappingof theequivalent circuit ontothephysical struc-
ture. Theparasiticswhicharecommonlyparticularlysignicant aretheshunt feedback
capacitanceC
gd
andtheaccessresistancesR
g
, R
s
, andR
d
. Thegateandsourceresistances
R
g
andR
s
compromisegainbyreducingthefractionof theinput signal that reachesthe
intrinsicgate. ThesourceanddrainR
s
andR
d
compromiseoutput power andefciency.
Thesourceinductance, L
s
, isoftenacriticallyimportant parasiticwhich, together with
R
s
, givesrisetoseriesfeedback compromisingdevicegain. However, thevalueof L
s
is
dominatedbytheinterconnectionnetworkincludingtheviatogroundandthereforenot
shownonthecross-sectionviewof Figure2.13. Theremainingparasiticsusually have
alesser impact thoughcanstill besignicant.
2.2 Power device physics 61
G S D
R
s
n
+
n
+
Ledge Ledge
Channel
Substrate
R
g
C
gs
C
gd
C
ds
R
d
R
i
R
ds
g
m
Figure 2.13 Correspondenceof GaAsFET physical structureandequivalent circuit network.
Anareaof particular noteinthemappingof theequivalent circuit isthenetwork of
elementsusedtomodel thedepletionregion. Thedepletionregionisasingleentitythat
delineates asingleregionof space-charge. However, it is accessedby all threedevice
terminals and so in the equivalent circuit the depletion region must therefore have a
connection to the three terminals. This is achieved in the lumped model by the use
of two capacitanceelements C
gs
and C
gd
. Thegate-sourcecapacitance, C
gs
, connects
acrossthedepletionregionfromthegatemetal tochannel formingthemaincontributor
totheinput capacitance. Thegate-draincapacitance, C
gd
, connectsfromthegatemetal
acrosstothedrain-sideof thedepletionregionandformsashunt feedbackcapacitance.
Modulation of gatedepletion region edgerequires chargeto beadded to or removed
fromthedepletionregion. Inthelow-eldregionthisprocessisdielectricrelaxation[57]
modeledby theC
gs
-R
i
arrangement wherethegate-chargingresistance, R
i
, represents
thenondepletedlow-eldchannel resistanceof thechannel under thegate. Inthehigh-
eldregionthemodulationof thedepletionregionedgeislimitedbythenitesaturated
velocityof thechannel carrierslimitingtheratethatcarrierscanbesuppliedor beswept
away. Thereisthereforeatimedelaygivenbytheproduct of thelengthof thesaturated
regionandthecarrier velocitythat limitsthespeedof thisprocessandthisgivesriseto
adelayterm,, for thecurrent generator equivalent circuit element.
2.2.8 Device gain and gures of merit
ThekeygainquantitiesforamicrowaveFET areillustratedinFigure2.14whichshowsa
setof typical commonlyusedgaincurvesforapHEMTdevice. Themoststraightforward
microwavegainquantityisthepower gainina50Osystem. InFigure2.14thisisshown
asS
21
. Thiscurveshowsaone-poleresponsedominatedbythe50O sourceimpedance
and the device input capacitance. Once above the 3 dB corner frequency S
21
falls at
6dB/octave. Inorder toachieveauseful gainperformancethedevicemustbepresented
withmoreappropriateterminatingimpedances. Theremainingcurvesontheplotarethe
62 GaAs FETs physics, design, and models
log(Frequency Hz)
G
a
i
n

(
d
B
)
MSG
MAG
h
21
S
21
f
max
f
T
U
Figure 2.14 Microwavegaincurvesandguresof merit.
quantitiesgenerallyemployedtoindicatewhat performanceisachievablefromadevice
for specicterminatingconditions.
h
21
: Thehybridparameter h
21
isthecurrent gainintoashort-circuit load. Theinter-
sectionof this curvewiththeunity gainaxis is akey devicegureof merit calledthe
transitionfrequency, f
T
whichisdiscussedinmoredetail below.
Gmax: TheGmax curveis thecompositeplot of themaximumavailablegain and
maximumstablegaincurves(MAGandMSG, respectively). MAGisthegainobtained
when the input and output are both simultaneously matched for optimumgain. This
quantity may only bedetermined when thedeviceis unconditionally stable. Whereit
is possible that a combination of source and load impedances will cause the device
to oscillatethenMAG is undenedandso insteadtheMSG is plotted. MSG provides
thetheoretical gainobtainedimmediately beforeoscillationoccurs andhas aslopeof
3 dB/octave. The slope of MAG is of the order of 6 dB/octave but varies due to
the variation of the optimal termination conditions which are frequency dependent.
Figure 2.14 shows just one transition frequency between MSG and MAG. In high-
performance FETs it is quite common for there to be a further transition at higher
frequenciesbacktoMSGasthedevicebecomesconditionallystableagain. Inpractice,
Gmaxcanbethought of asthebest gainobtainablebut itsinterpretationiscomplicated
bytheconditional stabilityissue.
U: A somewhat theoretical gain quantity often favoured by device specialists is
theUnilateralized gain U [58], also known as Masons invariant gain [59]. For this
quantity thefeedback hasbeenperfectly neutralizedtogiveagainmeasurethat isfree
fromcomplicationsof theeffectsof conditional stability.
f
T
andf
max
: Twopopular guresof merit for RF devicesaref
T
andf
max
. Theyassist
inthereadyassimilationof adevicesperformanceandtoallowconvenientcomparisons
of different devices. Thetransitionfrequencyf
T
istheunitygainfrequencyof h
21
the
2.3 Device design 63
frequency for which the current gain of the device has fallen to 0 dB. f
T
is a useful
andreasonably unambiguous gureof merit that is convenient to measureandrelates
directlytotheprimaryequivalentcircuitelementsthatdeterminethedeviceRFgain. The
elementsconcernedaretheintrinsictransconductanceof thegateg
m0
andtheassociated
capacitanceC
gs
that limitstherateat whichtheinput voltagemaybevaried. Theusual
approximateexpressionfor f
T
isgiveninequation(2.8).
f
T

g
m0
2(C
gs
C
gd
)
(2.8)
A signicant weaknessof f
T
asanindicator of deviceperformanceisthat it neglects
other important parasitics, andinparticular takesnoaccount of deviceinput resistance.
This is becauseh
21
is thecurrent gainfor thecaseof aninput current generator with
innite output conductance. It is therefore quite possible for a device with a high f
T
ratingto actually havearelatively poor power gain. Clearly however, adevicechosen
for apower amplier shouldhavealowinput resistanceandhighpower gain. A gure
of merit that addresses this requirement is f
max
, theso-calledmaximumfrequency of
oscillation. This parameter is thefrequency for which thepower gains U and Gmax
havefallento0dB, asapower gainof unityistheminimumgainrequiredfor adevice
tobeabletooscillate(seeAppendix2.1andreference[60]). Theexpressionfor f
max
can
bedeterminedfor thenetwork of Figure2.12[61] andisgiveninequation(2.9) which
illustratestherelativesignicanceof thevariousparasiticcomponents.
f
max
=
f
T
2
_
(R
g
R
s
R
i
),R
ds
2 f
T
R
g
C
gd

0.5
(2.9)
A difculty with f
max
is that there is no universally adopted approach to its deter-
minationandit is commonly overestimated. This is discussedinAppendix 2.1where
recommendedmethodsfor thepractical determinationof f
T
andf
max
aredescribed.
2.3 Device design
2.3.1 Power device design
The process of optimizing a power device comprises three main steps: (a) designing
thebasic FET devicestructure, (b) designing thepower cell whereaset of gates are
assembledto formastackableunit, and(c) formingacompositedevicefromaset of
cellstoprovideadevicewiththerequiredpower for agivenrequirement.
2.3.2 FET channel and recess design
Here the task is essentially to select the FET type and gate length appropriate for
theoperatingfrequency andto optimizethecurrent density capability andbreakdown
voltagewithout unduly compromisingtheother competingspecications suchas gain
andlinearity. Keyareasof attentionaretheepi-designandthegaterecess.
64 GaAs FETs physics, design, and models
Epi-layer design: GaAs FET epitaxial layer structures vary incomplexity fromthe
simplest uniformly-doped MESFETs [62], through reasonably straightforward mul-
tilayer designs for HFETs [8], to complex many-layer quantum-well structures for
pHEMTs [63]. In uniformly-doped MESFETs the epi-layer design choices are fairly
limited. Highpower requires highcurrent whichis achievedby ahighdopingdensity
and/or athickchannel. Bothof thesefactorshavelimits. For thecaseof dopingdensity,
then as this parameter is increased the breakdown voltage falls due to the increased
electriceldthat isdeveloped. Inaddition, thesemiconductor mobilityisdegradeddue
to increasedscatteringassociatedwiththedopant atoms. Alternatively, as thechannel
thickness is increasedthen, for agivengatelength, theoutput resistancefalls andthis
cancompromisethedevices ability to deliver current into thedesiredloadresistance.
Toprevent thisproblemthegateaspect ratio(theratioof gatelengthtochannel depth)
shouldbemaintainedtobeof theorder of veor moreinorder toensureasatisfactory
output resistance. Theaspect ratioconstraint presentsnoissuesfor longer gatedevices
butfor higher frequencyapplicationsrequiringsubmicrongatelengths, thentherestric-
tiononchannel depthforceshighlevelsof channel dopingtoachievethedesiredcurrent
densitytherebyresultinginacompromisedbreakdownvoltageandundesirablechannel
mobility.
MESFETs with tailored doping designs are employed with the aimof achieving
improveddevicecharacteristics suchas linearity andnoise. Moreideal deviceperfor-
manceisachievedbyconcentratingthedopingdeeper intothedevicewiththeobjective
of achieving, for example, astep-doped or similar prole. A desirableoutcomefrom
thismeasureistointroducelessvariationindepletiondepthasthechannel ismodulated
therebyresultinginamoreconstantdevicetransconductanceandgatecapacitance. How-
ever, thisbenet comesinevitably at acost of reducedcurrent density for agivenpeak
dopingandgatelength. Suchdevices thereforetendto havelower current capabilities
thanuniformlydopedFETs.
Thelimiting casefor astep-doped proleis theHEMT which seeks to restrict the
current ow to a narrow plane at a xed depth into the semiconductor. The HEMT
achieves this while being substantially free fromthe mobility degradation associated
with increased doping densities as described abovefor theMESFET. This is because
inthecaseof theHEMT thecurrent-carryingchannel isphysically separatedfromthe
donorionsandsomuchhigherdopinglevelscanbeusedwithoutadverselyimpactingthe
mobility. GaAs-basedHEMT structurestypically achievesheet-chargedensitiesabove
1.510
12
cm
2
which, for channel depths of order 100

A, would correspond to an
equivalent bulk-dopingdensityinexcessof 110
20
cm
3
. Evenif suchabulk-doping
densitywasapractical proposition(whichit isnt asthisdensityisconsiderablygreater
than thesolubility limit of then-typesilicondopant inGaAs), themobility wouldbe
enormouslydegradedandnotbemorethanafewhundredV/cm
2
satbest. IntheHEMT
thedopinglimit nowbecomesthat for whichtheassociatedelectronscanbeeffectively
contained in theheterojunction channel. For theAlGaAs/GaAs HEMT thedifference
in band-gaps between the two materials is relatively modest with a commensurately
limited degree of charge connement. Consequently this device has a modest peak
current capability. However, as shown in Table 2.1 theAlGaAs/InGaAs pHEMT has
2.3 Device design 65
Channel
Ledge
Outer recess
Inner recess
n
+
n
+
Gate metal
Cap layer
Figure 2.15 GaAsPower FET recessstructure.
amuch moresubstantial band-gap differencewith theresult that thepHEMT device
variant allows a high current density of order 500 mA/mmto be maintained for all
practical gate-lengths.
Thegate-recessdesign: A mostsignicantadvancethatallowedtheGaAsFET tobe
developedintoauseful powerdevicewasthedevelopmentandoptimizationof thedouble
gate-recess. Theearlier deviceshadasimplesinglerecessthat wastypicallycreatedby
simply etching into the channel until the desired current was achieved. At this point
thegatemetallizationwouldbedepositedwithintherecessedregion. Thebreakdown
voltageis enhancedby increasingthewidthof therecess, thereby givinganincreased
separationbetweenthegatemetal andthedrainn

contactregionandreducingthepeak
eld. However, inpracticetheincreasedexpanseof freesurfacetypicallyresultsinpoor
deviceperformancewithsignicantlyreducedcurrent andslow-stateeffects.
Thesolutionfor this problemis thedoublerecess [6467] depictedinFigure2.15.
Herethegateisdepositedinasmall inner recessthat lieswithinalarger outer one. In
thisconstructiontheouter recessislarger onthedrainsidetoprovidethegatedepletion
with roomto extend towards the drain as the gate-drain voltage is increased thereby
reducingthepeak electricelddeveloped. Incontrast tothesinglerecessstructure, the
variationsinsurfacedepletionontheextendedetchedsurfacecausedbychangesintrap
occupationsnowusefullyattenuatedbyvirtueof increasedphysical separationfromthe
channel. This attenuation is generally further enhanced by the inclusion of moderate
levelsof n-typedopingwhichact asachargescreen.
Considerable attention has been paid to optimizing GaAs FET breakdown in the
recent past. A signicant driver hasbeencellular base-stationPAsrequiringafewhun-
dredwattsof peakpower at0.82.1GHz. For atechnologythat hithertooperatedpower
ampliers with a typical drain bias of the order of 68 V, this application provided
a signicant challenge. Initial attention focused on optimization of the design of the
epi-layers and of gate-recess structure. Figure 2.16 shows the impact of one design
66 GaAs FETs physics, design, and models
3.0 2.5 2.0 1.5 1.0 0.5 0.0
0
10
20
30
40
50
L
gd
(micron)
B
r
e
a
k
d
o
w
n

V
o
l
t
a
g
e

(
V
)
BV
dg
BV
ds
Figure 2.16 Breakdownvoltagedependenceof gate-drainvoltageseparation.
variable, namelythelengthfromthegatetotheedgeof theouter recessonthedrainside
(L
gd
), onbreakdownvoltagefor a0.5mgatepower device. Thetwo-terminal break-
downBV
gd
isseentobereadilycontrolledbythesizeof therecessbutthethree-terminal
breakdown BV
ds
exhibits amorelimited dependence. Thegureprovides areminder
that even though BV
gd
is the normally quoted breakdown voltage measurement, in
reality thedeviceoperates inathree-terminal conditionandfor apower amplier the
drain-sourcebreakdownvoltage, BV
ds
cannotbeneglected. Fortunately, inpractice, com-
mercially availabledevicepower processes areappropriately designedandthequoted
BV
gd
dataisgenerallyagoodindicator of thebreakdowncapabilityinapower amplier
conguration.
Optimizationof therecessgaveasignicantstepforwardinincreasingthebreakdown
voltagebut itsabilitytospreadtheeldislimited. Toimprovefurther agreater degree
of eldspreadingisrequired. Someattemptsweremadetoemployreducedsurfaceeld
p-layersasusedtogreat effect insiliconLDMOS[69, 70], but most focushasbeenon
theuseof eld-plateelectrodeslocatedinthegate-drainrecessarea[68].
Field-plates: As discussed above, thechargetrapped in deep-levels on thesurface
of thegaterecess provides anaturally occurringassistanceintheeldspreading. The
useof eld-plates makes it possibleto engineer amoresubstantial effect andthis has
been an areaof signicant attention for higher voltageGaAs devices. In this context
a eld-plate is an extra electrode positioned on an insulating layer in the high-eld
gate-drain recess region. Its function is to provide a controlled potential that acts to
suppressthepeakeldatthegateedge. Thesimplestconstructionisthegate-connected
2.3 Device design 67
eld-plate investigated by a number of groups [7175]. As the normal shape for a
power FET gate is a T-shape (as discussed in Section 2.3.3), the simplest formof
eld-plateconstruction is merely an extension of thetop of normal T-shaped gatein
thedirection of thedrain to formaso-called I-gate(gamma-gate). Theeld-plate
approach can be very effective in increasing the breakdown eld, and research work
achieved device operation at and beyond a drain bias voltageof 28 V. Unfortunately
however, theimpactonthedevicegainissignicant[76]. Boththegate-sourceandgate-
draincapacitancesarecompromisedmakingthetechniqueunattractivefor frequencies
aboveL band. A morerecent development has been thesource-connected eld-plate
[77]. This conguration requires a more complex manufacturing process but has the
great benet of shielding the gate-drain coupling thereby reducing C
gd
and actually
improvingGmax. Theinputcapacitanceisstill signicantlyaffectedandthedevicef
T
is
consequently compromised, however for applications suchas cellular communications
for which the technology was targeted then this additional input capacitance can be
accounted for in theinput matching circuit. Signicant effort was deployed on eld-
platetechnologyfor GaAsdevicesandwithnotablesuccess. However, theeffortswere
largelyovertakenbywidebandgaptechnologysuchasGaNwhichhasnowbecomethe
preferredtechnologyfor high-voltageRF FET devices.
2.3.3 Power cell design
Gatewidth: Power FET cellsinvariablyrequireasmuchgateperipheryaspossibleand
a key factor which inhibits increasing the width
2
of the gate nger is the impact of
thecumulativeloss and delay of thegatesignal as it travels along thegateelectrode
[7880]. ThegateelectrodeisdepictedinFigure2.17awhichshowsadiscretizedmodel
of a loaded transmission line. The gate metal is modeled by the incremental series
resistanceand inductanceelements dR
g
and dL
g
with theassociated gatecapacitance
elementsdC
g
. Thegatevoltagewaveformappliedtothegatengerpropagatesalongthe
structureandisprogressivelyattenuatedasittravelstherebyreducingtheoverall device
gain. Inadditionthenitetimeto traversethestructurefurther degrades thesignal by
introducing a progressive phase delay that adds to the overall signal degradation. In
order to improvematters attention must bepaid to thegatecross-section as theshort
gatelengthrequiredfor fast transit alongthechannel lengthwouldotherwiseresult in
areducedcross-sectional areathereby providinghighseries resistanceandinductance
alongthenger width. ThesolutionwidelyemployedasshowninFigure2.15istoform
aT-shapedgatetoimprovethepropagationalongthegatemetal whilestill maintaining
ashort gatecontact lengthfor goodgainperformance.
Power cell manifold issues: Having optimized the unit nger the next task is to
designanassemblyof unit ngerssuitablycombinedsoastoachieveanoptimal power
performance that has scaled well with respect to the innate performance of the unit
nger [7981]. Figure2.17(b) illustrates theprinciplefor anexamplenetwork of four
2
By convention, thewidth of agateis thelong dimension perpendicular to thechannel direction and the
lengthof agateisthedimensioninthedirectionof channel current ow.
68 GaAs FETs physics, design, and models
dC
g
dR
g
dL
g
dC
g
dR
g
dL
g
dC
g
dR
g
dL
g
Gate
Terminal
(a)
Gate
Drain
(b)
Figure 2.17 Power FET cell; (a) gatenger equivalent circuit; (b) gatemanifoldwithfour ngers.
ngers combined to makeamodest multinger cell and shows thengers connected
byanassemblyof short transmissionlines. Asngersareaddedthentheoverall device
gainisfurther impairedduetothesuccessivephasedelaycontributionsexperiencedby
theadditional ngers. At rst thought it might beimaginedthat this phasedifference
couldbecorrectedbytheuseof analternativelayout whichcollectedthedraincurrents
withcompensatingdelaysintheoutput circuit. However, thisneglectstheimpact of the
high capacitiveloading on theinput network by thegatecapacitanceresulting in this
network beingrelatively slowcomparedtotheoutput network. Thehighimpedanceof
thedrain sideof thedevicemeans that theoutput current contributions arecollected
withrelativelysmall phasedifferencescomparedtothecumulativephasedifferencesin
theinput that buildupfromtheinternger andalong-nger delays. Another approach
to maintainthegainwouldbeto minimizethengernger phasedelay by designing
thedevicewithngers as closetogether as possible. Unfortunately theheat generated
in power FETs is usually substantial and so thedesign freedomhereis usually quite
limited in order to maintain an acceptable channel temperature required for reliable
deviceoperation.
Common-leadinductance: Asthegateperipheryisincreasedtoachievehigher cur-
rent capabilitythentheimportanceof theinductanceof thesourceconnectionbecomes
2.3 Device design 69
(a) (b) (c)
G G G
D D
D
Figure 2.18 GaAsFET power-cell layout variants; (a) gate-sidesourceviasfor highpacking
density; (b) source-bridgefor reducedinductanceof smaller cells; (c) individuallyviadsource
stripesfor highest frequencyperformance.
increasingly signicant. Thenegativefeedback generatedby thecommon-leadinduc-
tancecanbeasignicant factor reducingthegainof thepower cell andcloseattention
tominimizingits valueis oftenrequired. A very commontopology for power FETs is
showninFigure2.18awhichprovidestwoviasper cell. Theviasarelocatedtotheside
of the device and connection to the source ngers is made by an air-bridge over the
gatemanifold. This layout is very effectivefor power cells as it allows multiplecells
tobeefcientlystackedinarowfor ahighdensityof ngers. Thedisadvantageof this
congurationisthatitresultsinarelativelylongpathfromtheviatoeachsourcenger.
Higher gaincanbeachievedwiththesource-stitched layout of Figure2.18b. Herethe
vias arelocated adjacent to therst and last unit ngers and connected to thesource
stripesbyalowinductancebridgedfeed. Theimprovementisparticularlysignicantfor
amodest numbersof gatestripes, however thevialocationdoesnot makeefcient use
of dieareafor arraysof manyngers. Figure2.18cmakesuseof recentimprovementsin
viatechnology andtheabilitytofabricatenarrowwidthslottedvias. Narrowviawidth
allowssourcengerstohavedirectlyattachedgroundviasandtheslot shapeallowsthe
amount of viawall presentedtothedevicesourceconnectiontobemaximizedkeeping
thesourceinductanceper nger toaminimum. Ingeneral, thestylefor Figure2.18ais
most efcient onsemiconductor areabut styles (b) and(c) havegainbenets that can
beattractivefor higher frequencies.
Many of the factors in the design of the power-cell are amenable to mathematical
analysisandmodeling[82]. However, thepragmaticandmost accurateapproachtothe
determinationof thescalingbehavior ismerely todesignamask set of devicevariants
that coverstherequiredset of layout styles, gatewidths, number of ngersper cell and
gatespacings, andthenfabricatethedevicesandcharacterizethem. Theresultsof such
anexercisefor a0.5mgatepower FET processareillustratedinFigures2.19aandb.
Figure2.19ashowsthef
T
of thepower cell astheunit gatewidthisvariedfor arange
of gatesper cell. Figure2.19bshowsthecorrespondingcurvesfor f
max
. Ingeneral, the
performancereductionwithunit gatewidthis drivenby thecombinationof increased
gatenger resistanceandincreasedviainductance/mmof gateperiphery. Thereduction
withnumber of gatesisdominatedbytheviainductance/mmof gateperiphery.
70 GaAs FETs physics, design, and models
500 400 300 200 100
0
5
10
15
20
25
Unit gate width (micron)
f
T

(
G
H
z
)
8 gates
10 gates
12 gates
14 gates
(a)
500 400 300 200 100
0
10
20
30
40
50
Unit gate width (micron)
f
m
a
x

(
G
H
z
)
8 gates
10 gates
12 gates
14 gates
(b)
Figure 2.19 GaAsFET power-cell performance; (a) f
T
asafunctionof unit gatewidthandnumber
of gates; (b) f
max
asafunctionof unit gatewidthandnumber of gates.
2.3 Device design 71
2.3.4 Power cell combination
Power-cells are required to be combined to forma composite high-power device. Of
course, combining has to bedonein amanner that takes into consideration thesame
gain degradation issues that arose in the design of the power cell. Furthermore, the
approachhastocontendwiththeissuethat theinput impedanceof thecell islikely to
bechallenginglysmall.
Thesimplestapproach, commonlydoneformodestfrequenciesandimpedancelevels,
istocombinetherequirednumberof cells, provideeachonewithappropriatesetsof gate
anddrainbondpads, andleavetheend-user freetocombineinthecircuit asrequired.
Theindividual cells canhavecommongateanddrainbus bars, or bewholly separate
cells depending on thesizeof thecell and theapproach used to achievesatisfactory
stability. Someexamples of thevariety of power devicelayouts used in GaAs device
technology aregiveninFigure2.20. Thedevices shownrangefrom200mmdiscrete
devicesabletoprovideoutput powersof 100Wat 2GHz tomm-wavepower cellswith
apower capabilityof 0.5W.
Thediecanbesuppliedeither packagedor asbaredie. Thelatter approachprovides
thehighest performanceas it avoids theintroductionof signicant packageparasitics
inasensitivepart of thecircuit. However, theassemblycostsarehigher andthecircuit
modulehastoprovideahigher degreeof environmental protection. Thecircuitdesigner
will facethetaskof stabilizing, matchingandcombiningthecellstoachievetherequired
power amplier performance. At thislevel thedevicecombinationcannot betreatedas
a lumped problemand is typically done with a corporate combing approach using
a distributed network [83, 84]. One to three levels of corporate combining can be
consideredwhichcombine2, 4, or8die, respectively. Unfortunatelythelossesassociated
witheachlevel of thecombinationbuildupandsothebenet diminishes. Generallyup
totwolevelsof combinationareeffectivebutgoingbeyondthreelevelsisunlikelytobe
worthwhile.
In order to achieve optimal performance froma packaged power device then it is
commontoperformat least someof theimpedancematchingandcombiningfunctions
inboard of thepackageparasitics. By includingmatchingcircuitry at thedeviceter-
minalstheimpact of thepackageisincurredat alesssensitivepart of thecircuit. This
approach is very common for high-power devices [8588], so much so that thename
internallymatchedFET (orIMFET) hasemergedasalmostadevicetypeinitsownright.
It is normal in IMFET products to combinethefunctions of prematching with power
combiningtechniquesinorder tocombinethepower of multipledevices. Anillustration
of the typical circuit topology used to achieve this is shown in Figure 2.21. Another
aspectthatcanmakelargepower diedifculttodeal withistheir propensitytooscillate.
Their largeperipherygiveshugelow-frequencygainwhichmustbeaccomodated. Even
moreproblematicistheriskof odd-modeoscillationwhichcanariseduetoloopsinthe
combiner networks. Internallymatchingprovidestheopportunitytosubstantiallyallevi-
atethisproblemfor thecustomer byincludingsuitableinternal stabilizationtechniques
withinthepackage[89]. A further optionistoprovideinternal control of theharmonic
terminatingimpedancesasrequiredfor high-efciencyamplier modes[90, 91].
72 GaAs FETs physics, design, and models
(g)
(a) (b) (c) (d)
(e)
(f)
Figure 2.20 Power FET photos: (a) 200mmmulticell L-bandpower FET; (b) 60mmpower
L-bandFET; (c) C-bandpower FET; (d) power combinationinX-bandpower MMIC;
(e) conventional power cell evaluationstructure(BCB coatedfor mechanical handling
protection); (f) sourcebridgepower cell evaluationstructure; (g) electronmicroscopeimage
of power cell (evaporatedsecondmetal).
2.3.5 Thermal design
Thermal designisacritical partof apowerFETdesignprocess. Thejunctiontemperature
mustbemaintainedwithinoperational limitsinordertoensurereliableoperation. Ohmic
contacts to GaAs areobserved to degradeat elevated temperatures, but normally the
dominant degradation mechanismis gate-sinking. Here the gate metal diffuses into
thesemiconductor thereby movingtheeffectivelocationof theSchottky junctionand
so reducingthedevicecurrent [92]. A typical requirement is to maintainthejunction
temperaturebelow 150

C in order to achieveapredicted operating lifeof 1 million


hours. Gatesinkingisdiscussedingreater detail inSection10.7.1.
Adifcultyencounteredinthisendeavoristhelevel of uncertaintyinthedetermination
of thechannel temperature. Theavailablemethodsall havesignicant potential sources
of error. Commonly availablemeasurement methods areinfra-redimaging, useof the
2.3 Device design 73
D
G
Figure 2.21 Corporatecombiningandprematchingcircuit topologyusedinIMFETs.
liquidcrystal transitiontemperature, or useof thegatejunctionasathermometer [93
99]. Infra-red imaging is relatively convenient for surface temperature measurement,
but FETs present difculties as thegate-drain areathat needs to beresolved is small
withrespect totheinfraredwavelength. Theliquidcrystal approachislimitedinthat it
canonlyindicatewhenthetransitiontemperaturethresholdiscrossedrather returninga
valuefor thepeak temperaturefor agivenoperatingcondition. Theelectrical approach
inevitably returns ameasureof theaveragetemperatureacross thedevicerather than
thepeak temperatureand, for thetraditional switchedapproachof reference[98], error
is introduced by the delay involved in switching fromthe active operating mode to
passive sensing mode. Newer measurement techniques being developed are Raman
spectroscopy [100] and scanning thermal microscopy [101]. A recent advance to the
electrical approachhas alsobeenpublishedthat uses thegatejunctionstatedirectly in
anadmirably simplemanner obviatingtheneedtoswitchthegateintoasensingmode
[99]. Inthis latter approachachangeto thebase-platetemperatureis compensatedby
anadjustmenttothedrainvoltageinorder toreturnthegatejunctionthermometer toits
original conditionbeforethebasetemperaturewasadjusted. Fromthismeasurement a
valuefor themeanthermal resistancecanbeobtained.
Analternativeapproach is to usethermal simulation. Thedetaileddevicestructure
andthethermal propertiesof thematerialsusedarewell characterizedand3Dthermal
simulation tools are comfortably able to model structures to the required degree of
problemcomplexity[102104].
Therearealsonumerousapproximatemethodsincommonusage. Typical techniques
are2Danalyticsolutionsorbasicnumerical methodslimitedtolinearthermal conductiv-
ities. Suchapproachesshouldbetreatedwithsomecaution[102] astheapproximations
involvedfrequently donot apply toGaAsFET devices, andinthecaseof power FETs
thethermal operatingwindowavailablecanprovidesignicant designconstraintswith
minimal margin for error. Table 2.2 shows the results of a series of calculations of
thermal resistancefor amicrowavepower FET that illustratesthevariationinpredicted
values for different calculationmethods andfor different levels of physical detail that
areincluded.
Itisalsoimportanttorealizethatthethereisthepotential forinaccuracywitheventhe
mostcomprehensivesimulationtool. For example, thereremainsomeunknownssuchas
thecontributionsof thermal interfaces[100, 104], andusuallysomeuncertaintyover the
preciseconstructionof thethermal problem. It isclear thenthat whatever theapproach
74 GaAs FETs physics, design, and models
Table 2.2 Comparison of peak thermal resistance calculations for a 4 120 m
GaAs FET cell with a junction temperature T
j
of 150

C on 100 m substrates
mounted with 15 m epoxy (the linear analytic cases use an empirical
approximation to include the contribution of the epoxy)
Calculationmethod
RthC/W
(T
j
150

C)
2Dlinear analytic[107] 240
2Dlinear analyticwithend-effect included 206
3Dlinear semianalytic(TXYZ) [107] 184
3Dlinear nitedifference 208
3Dnonlinear nitedifference 237
3Dnonlinear nitedifferencewithsurfacemetallization 221
3Dnonlinear semianalytic[105] 226
3Dnonlinear nitedifferencewithsurfacemetallizationand
platedvias
213
Table 2.3 Simulated thermal resistances (

C/W) for central nger (R


C
) and outer
nger (R
O
) compared against measured values (100

C liquid crystal transition
temperature) for packaged RFMD discrete FET devices [105]
Devicetype R
C
R
O
(R
C
R
O
),2 Measured
FPD1500SOT89 75 54 64.5 60
FPD1500DFN 70 51 60.5 60
FPD2250SOT89 53 37 45 48
FPD2250DFN 50 34 42 40
FPD3000SOT89 41 28 34.5 35
used thereis scopefor signicant error. A good approach then is to support thermal
simulations withexperimental evidenceof cases that canbeaccurately measured. An
exampleof this is giveninTable2.3wherethetemperaturepredictedby simulationis
comparedto100

Cliquidcrystal transitiontemperatureforaseriesof packageddiscrete


FETdevices[105]. A further, particularlydetailedcomparisonisgiveninreference[108]
whereaspecial devicewasconstructedwithonengerof apowercell arrayconnectedas
apassivethermometer. A sensibleandpragmaticwaytomitigateanyresidual systematic
error istoemploy thesametechniqueinthethermal designof product aswasusedby
thefabricationfacilityinthegenerationof thedevicelifedata.
2.4 Device fabrication
2.4.1 Overview
InthissectionGaAsdevicefabricationtechniquesaredescribedfromthestartingpointof
themanufactureof blankwaferswithsuitableactivelayersonthesurfaceof aninsulating
GaAs substrate. For aMESFET asimpledoping schemecomprising achannel layer
2.4 Device fabrication 75
accessedby ahighly dopedupper contact layer is all that is required. MESFET layers
canbecreatedeither by usingion-implantationof dopant ionsintothesubstrate, or by
epitaxial growthof layersontothesubstrate. For morecomplexdevicessuchasHFETs
andHEMTsthenepitaxial growthisrequired. Theselayersareasuitablecombinationof
GaAsandAlGaAs/InGaAspartner layersdopedasnecessary toconstruct therequired
devices. Thegrownwafersarethentransferredintoawaferprocessingfacilitywherethe
semiconductor layersarepatterned, implanted, etched, metallizedandcoatedasrequired
tofashionthedesireddevicesandcircuits.
Inthefollowingdiscussionthefocuswill beonepitaxiallygrownlayers, nowthemore
common approach for GaAs-based FET devicemanufacture. In commercial devices,
epitaxial layers aregrownby molecular beamepitaxy (MBE) or metal organic chem-
ical vapour deposition (MOCVD). In MBE, asubstrateis heated under high vacuum
and beams of theappropriateproportions of theconstituent atoms aredirected at the
substrates, condensingonthesurfaceto formvery high-quality layers of therequired
compounds. Thelayerscanbedepositedwithvery highprecisionwithlayer thickness
control downtoafewatomiclayers. After growththelayersareinspectedfor accuracy,
typically assessingthesheet chargeandlayer thicknesses, themolecular compositions
of theAlGaAs/InGaAs ternary layers and thequality of thesurface. In MOCVD the
constituent atomsaredeliveredtothesurfaceof aheatedsubstratebymeansof suitable
precursor organic gas molecules that chemically decomposeat theheatedsurfaceand
depositthedesiredatomsatthesurfacelayerbylayer. Typical sourcegasesaretrimethyl-
gallium, trimethylaluminum, trimethylindiumand arsinewith ahydrogen carrier gas.
Thecrystal composition is controlled by adjusting therelativeproportions of thegas
owover thesurface.
2.4.2 Key process steps
Uponreceipt of theepi-wafers deviceprocessingcancommence. To formthedevices
a set of process modules are developed that performfunctions such as creation of
ohmiccontacts, recessetching, Schottkygatedeposition, metal interconnectdeposition,
insulatinglmdepositionandetching,substratethinning,andviaetching.Thesemodules
employasetof processstepsthatareoptimizedtoworkinconcertandarecharacterized
andmaintainedto meet thenecessary manufacturingtolerances. Themenuof process
stepsthat aretypicallyemployedaredescribedinthefollowingparagraphs[109, 110].
Lithography: All of thewafer processingoperationsneedtobeselectivelyappliedin
controlledareas. This is achievedby lithography most commonly photolithography.
Hereasuitablephotosensitiveresistlmispatternedwithanimagethathaspreviously
beencreatedonaphotographicglassplatecalledamask. Theresistlmisspunontothe
wafer, exposedwiththerequiredimageandthenchemically developed. For so-called
positive resist theunexposedarearemainsintact therebyshieldingthecoveredregion
fromasubsequent etchingor metal depositionprocessstep. By thismeansthevarious
device features may be patterned as required. Alternatively, negative resist can be
used. This behaves intheoppositesenseso that theexposedarearemains after being
developed.
76 GaAs FETs physics, design, and models
Three types of lithography are in common usage for GaAs wafer processing. The
simplestiscontactprinting. Here, after thephotoresisthasbeenapplied, thetechniqueis
toalignthemasktoexistingfeaturespreviouslyfabricated, clampthemasktothewafer
andthenexposetheassemblytolight. Thisisalow-costapproachcapableof featuresizes
down to 0.5 mand is quitesuitablefor small-volumemanufacture. It does however
sufferfrommaskwearandregistrationaccuracyissuesacrossthewafer. Theresolutionis
fundamentallylimitedbydiffractionof theincidentlightsourceandforhigherresolution
otherapproachesarenecessary. ForGaAsFETsthehighestresolutionrequirementisfor
thegatemetal whichisusuallyof theorderof 0.5morless. Traditionally, electron-beam
lithographyhasbeenthesolutionadoptedfornegeometrygatedenitionbelow0.3m.
Herethegatepattern is created by steering an energetic electron beamto thedesired
areasandtherebyexposingasuitableresistmaterial inthoseregions. Thiseliminatesthe
optical diffractionproblem(theelectrondeBrogliewavelengthforkV electronsisbelow
0.1nm), andgivesaresolutionlimitedbythescatteringintheresist andbackscattering
fromthe wafer. For research devices E-beamgate lengths have been driven down to
0.05mor less. Incommercial devices E-beamgates aretypically availabledownto
0.15m. Thechief disadvantagesof E-beamlithographyarecomplexityandthroughput.
Theserial natureof thewritingprocessmeanstheexposuretimesarelengthy.
Thethird lithographic techniquein common usageis theoptical stepper. Herethe
patternis imagedonaportionof thewafer withrefractiveoptics. Themask is usually
enlarged, typically 5 times greater than the nal image and the pattern, or shot, is
stepped and repeated to cover the wafer. Shot sizes are typically up to the order of
2020mm
2
. Theuseof optical steppers is thedominant approach for high-volume,
high-yield processing. The precision of the stepper optics is extremely stringent and
requiresassociatedcontrol of vibrationandtemperature, compensationfor air pressure
variation coupled with precise alignment tools, stage-stepping control, and complex
focusingsystemscapableof adaptingtolensaberration.
Liftoff: Inthepatterningof metalsonGaAsdevicesmuchuseismadeof aprocedure
called liftoff. This is a different approach to that used in silicon processes where
thealuminumtracks areformedby depositingthemetal lm, applyingandpatterning
thephotoresist andthenetchingback wherethemetal is not required. Thesituationis
different for GaAs devices whichmakeuseof goldtracks andcompositemetal stacks
whicharenot readily etched. Theapproachfor thesemetalsistoapply andpatternthe
resist beforethemetal is depositedandthereforeto usetheresist to control wherethe
metal isdeposited. Whentheresist isdissolved, theunwantedmetal that wasdeposited
ontheresist lmisliftedoff andremoved. Tofacilitatethisprocesstheedgeproleof
theresistaperturesisfashionedbyvariousmeanssoastohaveanoverhangor lipand
themetal isevaporatedwithanear normal incidencetothewafer sothat acleanbreak
inthemetal lmiscreatedbytheshadowingeffect of theoverhang.
Deviceisolation: Waferswithepitaxially grownactivedevicelayersrequirethat the
individual devices onthewafer beisolatedfromeachother. This is either doneusing
mesa etching or ion implantation. With mesa etching, islands of active material are
retainedandtheregionsof interconnectingepi-layersareremovedbyanetchingprocess.
Mesaetchingis aneffectiveapproachthat avoids theneedfor expensiveimplantation
2.4 Device fabrication 77
equipment. However, it introduces undesirablesurfacerelief andintroduces increased
gateleakageat thepoint wherethegatemetal striperises upthesideof themesaand
crossestheactivelayer. Anysurfacerelief isunwantedfromaprocessingperspectiveas
it inhibits uniformphoto-resist coveragefor subsequent process stages, andsteps tend
to compromisetheintegrity of any metal tracks that haveto go over them. With ion
implantationtheactivedeviceregionsareprotectedwiththickphoto-resistandtheareas
tobeisolatedaredeliberatelydamagedbyanenergeticbeamof ionstherebyrendering
theexposedregionstohavehighresistivity.
Forion-implantedMESFETsthendeviceisolationmaynotbenecessaryasthedoping
for thedeviceactiveareascanbedoneselectivelyintheregionswhereit isrequired. In
thiscaseimplantationisservingtheoppositepurposetotheisolationcaseabove. Silicon
donor ionsareimplantedwithenergiestoachievetherequireddepthprole. Thewafer
is then heated so that thedamagedoneto thecrystal structureduring implantation is
annealedoutandthedopantionsincorporatedintothecrystal latticeinorder toactivate
themasdonors. Alternatively, ablanketdopingimplantoverthewholewafercanbeused
andtheactiveareasarethenisolatedeither withanisolationimplant or mesaetchingas
for epi-wafers.
Ohmiccontacts: Thefunctionof anohmiccontactistomakealow-resistanceelectri-
cal connectionfromthemetal trackstothesemiconductor activelayers. Ohmiccontacts
to GaAs aremadeby theuseof anickel, gold and germaniummetal stack deposited
onto aheavily dopedGaAs contact layer andannealedat approximately 400

C. The
essential purposeistoreducethethicknessandeffectiveheight of theSchottkybarrier
thatformsatmetal-semiconductorjunctionstosuchanextentthatthebehaviorisohmic.
Theexact mechanismremainssomewhat elusivebut involvesthegenerationof ahighly
doped surface layer of germaniumsubstituting for galliumin the crystal lattice. The
nickel component rst acts as awetting agent for theGeAu but it is also believed to
enhancethediffusionof germaniumintoGaAs[109].
Gate Etch: Prior to the gate formation a recess is etched into the semiconductor
material to removethehighly-doped contact material abovethechannel. Historically,
thiswasanetch-to-current activity wheretheetchratewouldbecarefully calibrated
andatimedetchwouldbeusedtotargetthedesiredrecessdepth. Subsequentverication
by testing the drain current of the etched structure would result in the wafer being
returned for a top-up etch if the measured current was too high. Such crudity was
eliminatedwiththeadvent of etch-stops wheretheetchchemistriesarechosensoas
to beselectiveto thevarious heterojunctionlayers. Theheterojunctions may therefore
beusedtostoptheetchprocessat precisedepthswithhighaccuracyand, crucially, the
accuracyismaintainedover thewholewafer. Thisadvancewaskeytothedevelopment
of high-yieldmanufactureandtheuseof largeareawafers. For example, reference[111]
demonstrates that theuseof anAlGaAs etch-stoplayer for aGaAs MESFET reduced
theprocessstandarddeviationfor Idssfrom25%to5%. Layersthat arealreadypresent
in theepi-stack for their electrical function may beemployed if appropriate[111] or
specic etch-stop layers may beadded to theepitaxy design that aretherepurely for
control of theetchprocess. Alternativededicatedetch-stoplayers areAlAs [112] and
InGaP [113].
78 GaAs FETs physics, design, and models
Therearetwo mainapproaches to GaAs etching: wet or dry. Wet etchant solu-
tions consist of an oxidizing agent to oxidizethesurfaceand asecond component to
dissolvetheoxide. CommonlyusedetchesabletoprovideetchselectivitywithAlGaAs
aredilutehydrogenperoxide/ammonia, hydrogenperoxide/citricacidandhydrogenper-
oxide/succinic acid [111, 114, 115]. Dry etching uses plasmachemistries involving a
combinationof chlorineanduorineradicalsinconcertwithenergeticionbombardment
[115, 116, 117]. Thechlorineproduces theetchingreactionandtheuorineproduces
an etch-stop reaction with aluminumdueto theformation of anonvolatilealuminum
uorideprotectivelayer onthesurface. Bothapproaches haverelativeadvantages and
disadvantages. A wet chemical etchprovidesalow-damagesurfaceandfor thisreason
isoftenpreferred. However, wetetchingisisotropicandsohaslessdimensional control.
Dryetchinghastheadvantageof gooddirectionalitygivingahighdegreeof dimensional
control, but thiscancomeatacostof somesurfacedamagefromtheionbombardment.
However, it is foundto bepossibleto tailor thedry etchrecipeso as to minimizethe
ionenergy towards theendof theetchprocess andsuitably control thedegreeof etch
damage[116]. ThereforebothwetanddryetchescanbeusedforGaAsFET gateetching
andbothareinuseincommercial processes.
Schottkygateelectrode: ThegatestructuremustmakeagoodSchottkybarriercontact
tothesemiconductor, onethatisstableover thelifeof thedeviceandthatprovidesalow
resistancealongthegatenger. TheSchottkybarrierheightislargelypinnedbysurface-
statestoabout0.7eV ratherthancontrolledbytherelationshipof thesemiconductorand
gatemetal work-functionsasnormallydescribedinintroductorytextbooks. Thereforein
principlemanymetalswill provideadequateSchottkybarriers. Inreality, considerations
suchasmetal adhesionandthermal stabilityprovidethepractical selectioncriteria. The
result isthat thereisachoiceof twoapproachestotheformationof thegateelectrode:
anevaporatedgold-basedgateor asputteredrefractorymetal approach[118, 119]. The
mostcommontechniqueistouseanevaporatedgatetypicallyusingatitanium-platinum-
gold (Ti-Pt-Au) metal stack. Herethetitaniumlayer ensures good adhesion, thegold
provideslowfeedresistanceandtheinterveningplatinumlayeractsasadiffusionbarrier
keepingthegoldsafelyfromdiffusingintothegatejunction. Thesecondapproachisthe
useof awhollyrefractorymetal approach, generallyusingtungsten-silicideor titanium
tungsten. TheTi-Pt-Auapproachis asimpler technology however refractory gates are
morethermally stable. This is advantageous not just for deviceoperationbut also for
devicefabrication. Thethermal resilienceof arefractorygateallowsthegatemetal tobe
depositedbeforetheohmic contact metal thereby makingthecritical gatelithography
mucheasier andallowingtheohmicmetal tobeself-alignedtothegate[120]. Withthe
conventional Ti-Pt-Austack theohmic contact anneal stepmust becompletedprior to
thegatemetal deposition.
For power FETsitisgenerallythecasethatinorder toobtainhighpower itisessential
to beableto operatewith as wideadevicenger as possible. A limiting factor here
is thegatemetal resistanceand so power FETs usually usesomeformof T-shaped
gate where the top of the gate metal is widened to reduce the resistance along the
stripe. Inthecaseof E-beamgatestheT isachievedby amultilevel resist approach,
typicallyemployingPMMA (polymethyl methacrylate) thermoplasticresistmaterialsin
2.4 Device fabrication 79
atypically bi- or tri-layer schemeto produceamushroom-shaped prolein theresist
comprisinganarrowstemandawider T or mushroomtop. This is typically achieved
by usinglayers of resist that differ intheir sensitivity to thedeveloper solutionandto
usethemoresensitivematerial for thedenitionof theT-top. A nal thinlayer maybe
employedto createalipfor improvedliftoff. Followingthecreationof themushroom
cross-sectionintheresist thegatemetal stack isdepositedover thewafer andtheresist
developedaway liftingoff theunwantedmetal andleavingthegatestructures behind.
Anexampleof adevelopedPMMA resist cross-sectionthat providesagoodillustration
of theapproachisgiveninreference[121]. Theresult of themultilayer approachisto
createthedesiredT shapewithminimizedgateresistanceandcapacitance.
Theuseof stepper-based lithography introduces adifferent method of forming the
gate. Heretheapproachis to formtheT-gatewiththeT-topsupportedonadielectric
layer [122124]. The gate-support layer is deposited, a T-stemis dened and etched
in this dielectric layer and theT-top is dened in photo-resist on thetop. Gatemetal
is then evaporated to formagateof therequired shapethereby giving theresistance
benetrequired. ThedrawbackrelativetotheE-beamgateprocessdescribedpreviously
is that the gate capacitance is increased slightly due to a higher degree of dielectric
loading associated with thedielectric layer supporting theT and typically, alarger
T-topoverlaparea.
Dielectriclayers: A number of dielectriclayersarerequiredfor avarietyof purposes
including protective coatings, supporting metal track cross-overs, and for the forma-
tionof integratedcapacitors. A commonly usedmaterial for GaAsprocessesissilicon
nitridedepositedusingplasma-enhancedchemical vapour deposition(PE-CVD). This
technique is compatible with the modest thermal constraints of GaAs device manu-
facture. Inthis approachsiliconnitridelms aredepositedduringaplasma-enhanced
reactionof silane, ammoniaandnitrogengases. Careful processoptimizationisrequired
for the successful deposition of device lms, with particular attention to lmstress
andplasma-induceddamage. Filmstressisasignicant factor for GaAsdevicesasthe
material ispiezo-electric[125]. Theplasmaistypicallygeneratedwitha13.56MHzRF
powersourceandtheresultinglmfromthisarrangementisstressedandtypicallytensile
innature. Control of thestressfromtensiletocompressivecanbeachievedbyanumber
of methodsincludingadjustment of gascompositionor theadditionof acomponent of
lower frequency power, typically 12MHz [125, 126]. Thelatter approachintroduces
ahighenergyion-bombardment of thegrowingsiliconnitridelmandthisresultsina
controllablechangetotheresultantstressstate. Althougheffectiveincontrollingthelm
stress, ionbombardment employedinthevicinity of aGaAs surfaceintroduces unac-
ceptabledegradationof thesurfacetherebyintroducingatradeoff of lm-stressagainst
surfacedegradation. However, devicesareusuallyfabricatedwithanumberof lmlayers
andagooddegreeof stresscontrol canbeachievedbydesigningthestackof composite
layersappropriatelysothat theoverall lmstressisacceptableandthesurfacedamage
arisingfromthenear-surfacelayersisminimal. Inorder topatternasiliconnitridelm
after deposition it must be etched and either dry or wet etching approaches may be
used. Dryetchingispreferredduetoitssuperior dimensional control. Itistypicallyper-
formedusingasulphur hexauoride(SF
6
) plasmadilutedinheliuminorder toachievea
80 GaAs FETs physics, design, and models
controllableetchrate. Wet etchingof siliconnitrideisuncommonbut canbeperformed
usingabufferedHydrouoricacidetch.
I nterconnectmetals: Usually, twolevelsof interconnectmetal arerequiredinorderto
makeconnectionsbetweendevicesandother componentsandexternal bond-pads. The
rst level metal is generally evaporatedgoldandis generally depositedto athickness
of 12m. A second level metal is needed so that onetrack may cross-over another,
for capacitor formation, and to providethicker tracks in order to carry high currents.
Second metal tracks areusually electro-plated gold onto asputtered seed layer. More
recently somemanufacturers arenowemploying evaporated secondmetal rather than
usingelectroplating[127]. Thiscanhavecost advantagesparticularlyfor highvolumes
asthehighlyuniformnishof evaporatedmetal ismoreeasilycompatiblewithautomatic
visual inspectiontools.
Backsideprocessing: GaAsRF technologyisusuallyof themicrostripvarietyrequir-
ing a ground-plane on the backside of the wafer. Connection to the ground plane is
madebythrough-wafer vias. Processingof theundersideof thewafer thereforeconsists
of thinning, viaetchingandmetal deposition. Thecompletedfront surfaceisprotected
and then temporarily adhered to a supporting carrier with a wax or photoresist. The
wafer isthengrounddowntothedesiredthickness. Forpower devicesthenal thickness
requirement is normally governed by thethermal design and is typically in therange
50120m. Through-wafer viaetchingisperformedusingdirectional dryetching. The
backsideisthenmetallizedusingelectroplatedgoldontoasputteredseedlayer.
ProcessMonitoring: Inordertoevaluatetheperformanceof eachwaferandtoprovide
dataforstatistical control of theprocess, anumberof standardizedtestcellsareincluded
oneachwafer. Thecellsarecalledprocesscontrol monitor(PCM) cellsorthecoupon.
Thenumber usedper wafer varies anddepends onthewafer size, thematurity of the
processandlocal policiesandcanvaryfromtentoahundred. ThePCM containsboth
structures to assess individual process steps andstandarddevices whichareevaluated
at variouspointsintheprocessow. Structuresareincludedtoassessthecontacts, the
efcacy of theisolation, thequality of eachof thevarious metal anddielectric layers,
and any GaAs or thin-lmresistors that might also bein theprocess. Theseand the
standarddevicecells aretypically assessedafter thegatehas beendeposited, after the
front-facehas been completed, and nally at theend of theprocess. Thedevicetests
performedwhilethedeviceisintheproductionlineconcentrateoncardinal parameters
including the pinch-off voltage, drain current for V
g
= 0 V (I
dss
), maximumdrain
current, breakdownvoltage, diodebuilt-involtageandideality, gateleakageandtheDC
transconductance. Thesearemonitoredwithaviewtoobtainingconstant feedback on
theprocess so as to keep it in control and for identifying occasional errant wafers so
that theycanbescrappedassoonaspossibletoeliminatethecostof further processing.
Uponcompletionof thewafers, astandardPCM FET structuredesignedtobesuitable
for on-wafer RF testingis usually assessedwithsomelevel of RF test. This typically
consistsof ameasurement of S-parametersat aspecicstandardbiaspoint fromwhich
anindicationof RF performanceis obtainedby extractionof f
T
, f
max
, or Gmax. Some
manufacturersalsoperformequivalent circuit extractioninorder tobeabletomonitor
keyequivalent circuit parameters.
2.4 Device fabrication 81
At theendof theproductionlineaset of themost critical parametersthat havebeen
testedarecollatedandusedinthewafer acceptancetest(WAT). For thistestadened
fractionof thePCM structureshavetobewithinthespecicationlimitsfor thewafer to
beacceptabletogoonfor visual inspectionandreleasetothecustomer.
2.4.3 Low-cost GaAs device fabrication
Therecent increaseinGaAsproductionvolumesandthecost pressuresof thecellular
handset market that havedriven that volumehaverevolutionized GaAs devicemanu-
factureandthecapabilities that canbebrought tobear. Key developments herearean
increaseof GaAswafer diameter from4
//
to6
//
andtheuseof stepper-basedlithography
withitsinherent benetsof highthroughput, uniformity andyield. Asdiscussedinthe
Schottkygateelectrodepartof Section2.4.2, dielectricallydenedgatetechniquesused
instepper-basedprocessesinevitablyhavehigher parasiticgatecapacitancethanunsup-
portedapproachesusedbye-beamprocesses. However, thereductionof performanceis
acceptablefor many applications andthebenets of lower cost andgreater uniformity
aresubstantial. Furthermore, thehigher degreeof dieencapsulation requiredinorder
toprovidethelevel of environmental protectionthat isincreasingly demandedanyway
involvesanincreasedamount of encapsulatingdielectricmaterial. Inthiscase, thefact
that dielectric-dened gate processes start with a higher degree of dielectric loading
becomeslessof anissue. Thesubstantial advantagesof 6
//
stepper-basedwafer fabrica-
tionhas thereforeledto themajor manufacturers adoptingthis approachandoffering
optically dened gatetechnologies [128130]. Theinitial useof optical steppers was
for 0.5mapplicationsandmadeuseof relativelyaffordablei-line steppers(365nm
wavelength). This technology is suitablefor devices with f
T
s of theorder of 25 GHz
andfor applicationstoXband. Morerecentlyprocesseshavebeendevelopedtobeable
to fabricatesmaller gates for higher performancedevices. Theoptions hereareto use
adeep-UV stepper or oneof anumber of gate-shrink approaches. Withadeep-UV
stepper ashorter wavelength of 248 nmis employed to directly imagegates down to
the order of 0.25 m[131]. Alternatively, or in combination, one of the gate-shrink
approaches canbeemployedtopatternthegatedimensionbelowtheresolutionof the
stepper. Techniques that havebeensuccessfully employedincludetheuseof dielectric
sidewall spacers[120], chemical shrink [132], reowedresist [133, 134] andtheuseof
phase-shiftmasktechnology[135]. Withthesetechniquesstepper-basedapproachescan
beemployedinproductionat andbelow0.15m.
2.4.4 Packaging
Packagesareusedfor easeof handingthefragiledieandfor environmental protection.
Therangeof availablepackagetypes is extensivereectingdiverserequirements that
stretchfromDCtomillimetrewave. Thepackagingsolutionsrangefromplasticmolded
structures for high-volumeapplications to hermetically sealed ceramic housings with
high-qualityintegral heatsinksfor thehighestperformanceproducts. Figure2.22shows
aselectionof packagesusedfor GaAspower devices.
82 GaAs FETs physics, design, and models
(c) (d)
(a) (b)
AS
QFN
AF
SOT89
Figure 2.22 Typical power FET packages; (a) SOT89; (b) ceramicangemount; (c) leaded
ceramicsurfacemount; (d) leadlessplasticsurfacemount (QFN).
Traditionally, the packaging route for microwave power transistors was essentially
limitedtothemetal-ceramicair-cavitypackageconsistingof analuminahousingbrazed
ontoametal angesuchastheAF packageinFigure2.22. Thisapproachprovidesgood
heat dissipationproperties, agoodRF ground, awell-controlledparasiticenvironment,
andhermeticsealingoptions. Thisapproachremainsthehighestperformancepackaging
optionbut this is achievedat asignicant cost. Not only is thepackagecost themost
expensivecomponent in apackaged power transistor part, they arealso expensiveto
assembleandtheresultingcomponents arenot amenabletohigh-volumecircuit board
manufacturingprocesses.
The high cost of conventional ceramic packaging encourages the development of
alternativetechnologies andasustainedfocus has beenonthedevelopment of plastic
packages [136]. Theuseof plastic packages introduces anumber of difculties com-
paredtoceramictechnologyincludingincreasedgroundinductance, substantial dielec-
tric loading, and thermal dissipation and expansion issues. Established over-molded
plasticpackagessuchastheSOT89styleequippedwithareasonablyheavylead-frame
areemployedfor modestpower andfrequencyapplications. Moredemandingsituations
have driven appropriate technology developments. Silicon LDMOS devices targeting
2.4 Device fabrication 83
(a)
(b)
Figure 2.23 QFN packagecross-section; (a) over-moldedpackage; (b) air-cavitypackage.
frequencies upto theorder of 2GHz havepursuedover-moldedplastic encapsulation
of devicesmountedonanintegral copper heat spreader. Suchsolutionsaresuitablefor
inexcess of 100W [137]. However, GaAs devices generally address higher frequency
andlower power applications. Surfacemount packages, suchastheASandQFNstyles
inFigure2.22aredesirablefor high-volumeassembly requirements. Leadedandlead-
less surface mount ceramic packages are well established and plastic surface mount
techniques are currently receiving much attention. Of particular note is the example
of theQFN package(QuadFlat No leads, J EDEC standardMO220[138]) of whicha
33mm12-padstyleis showninFigure2.22. This packagestandardwas originally
developedfor low-speedgeneral electronics, however its constructionis amenablefor
development for microwavefrequenciesandQFN packageshavefoundapplicationfor
arangeof microwavedevicesandMMIC circuits. StandardplasticQFNpackagessuch
asisdepictedinFigure2.23aareusedinapplicationsupto18GHz andapproximately
1W dissipation. Developments haveincludedpower variants withaheavier duty lead
frame and employing solder die attach [139, 140]. For higher frequencies the use of
alternativematerialsandconstructionsareattractiveinorder toachievelower dielectric
loading. Signicant attention has been paid to organic polymers which have supe-
rior microwave properties to plastic [141, 142]. Millimetre-wave capable approaches
have been developed using multilayer approaches with air cavities as illustrated in
Figure2.23(b) [143]. Ceramic implementations arealso pursuedfor higher frequency
applications dueto thesuperior mechanical precision of thosematerials and 40 GHz
operationhasbeendemonstrated[144].
Anactivity of signicant interest for low-cost manufacturingis theoptimal routeto
achievingacceptableenvironmental protection. Traditional ceramicpackageswereable
toprovidehighlevelsof hermeticity.Alternatively,theenvironmental protectioncouldbe
providedatthemodulelevel. However, thecontinual drivefor lower manufacturingcost
nowincreasinglyexcludessuchoptionsandrecentworkhasfocusedonthedevelopment
of hermeticlow-costpackages[143] oradequateencapsulationatthedielevel. Thelatter
objectiveis achievedby ensuringthat thenishedsemiconductor diesurvivestandard
84 GaAs FETs physics, design, and models
tests for hermeticity and are sufciently robust that the encapsulation survives the
mechanical handlingencounteredintheplastic packagingprocess. Two standardtests
herearethe85/85THBandHASTtestswhicharediscussedinmoredetail inSection
10.9.3. In the former the packaged parts must survive 1000 h at 85

C temperature,
85%humidity andtheoperatingbias. Thelatter highly acceleratedstress test aims to
replicatethesamestressinamuchshortertimeframe. ThestandardHAST conditionsof
130

C with 85%relativehumidity is achieved at approximately 18 psi overpressure.


Thestandardassumptionisthatthe1000hTHBtestisequivalentto96hof HAST based
onearlier work onsilicondevicetechnology. Thisequivalencehasbeenquestionedfor
thecaseof compoundsemiconductor devicesbyanumber of workers[145, 146] andthe
GaAs deviceindustry has foundtheroutinesatisfactory attainment of the96hHAST
requirement adifcult hurdle[147].
2.5 Models
2.5.1 Device models
Devicemodelsareemployedtosimulatedevicebehavior incircuitdesign. Theaccurate
simulation of GaAs FET power ampliers can present a number of difculties for
thepresently availablemodelingtechniques andthedegreeof success achievedvaries
considerably depending on the precise application. Available models for moderately-
sized devices (up to a few mmof gate periphery) can generally predict rst-order
parameterssatisfactorilysuchasterminal impedances, gain, powerandpowersaturation,
andgiveareasonableindicationof efciency. However, thesituationformoredemanding
requirementssuchasaccuratemodelingof verylargedevices, for thepreciseprediction
of large-signal nonlinearityandfortheimpactof sometransientphenomenasuchasself-
heatingandslow-stateeffectsondigitallymodulatedwaveformsisoftenlesssatisfactory.
2.5.2 Small-signal models
Extraction of the parameters of the equivalent circuit model of Figure 2.12 is well
established. For devices of gateperipheries up to theorder of amillimeter or so, this
model is readily extractablefromasuitableset of bias dependent S-parameters using
thedirect extraction techniqueintroduced by Dambrineet al. [148] and subsequently
further renedbynumerousauthors. Theessential techniquereliesonaset of off-state
or cold-FET biases to extract the embedding parasitic elements froma simplied
equivalent model applicableto this bias condition. Theembeddingparasitics obtained
forthissimplernetworkareassumedtobealsoappropriatefortheon-stateorhot-FET
bias condition. Theseparasitic values may thereforebeusedto de-mbed thehot-FET
data, thereby obtainingthey-parameters of theintrinsic FET equivalent circuit and, at
this point, solving for intrinsic elements is straightforward. Themodel obtained from
this process is usually acceptable for frequencies below around 10 GHz. At higher
frequenciesit isnormal tondthat theaccuracy of thereverseisolationparameter S
12
2.5 Models 85
R
s
R
i
C
gs
C
gd
C
dc
g
m
R
ds
C
ds
R
d
L
d
L
s
R
g
L
g
+
--
g
m
= g
m0
.e
j
Figure 2.24 High-frequencyGaAsFET equivalent circuit network.
becomes unacceptable. Thereasonfor thediscrepancy is duetotheinadequacy of the
equivalent circuittopology. Better tsthanthatobtainedfromdirectextractionmethods
may bereadily obtainedbut it isgenerally foundthat thisinvolvesnon-physical values
for someof theelements. Alternatively, modiedequivalent circuit topologies may be
employedwhichattempt toaddresstheadditional complexityinthefrequencyresponse
at the higher frequencies. Such factors as dipole capacitance and distributed effects
[148152] giveriseto modiedequivalent circuit models suchas that of Figure2.24.
These more complex equivalent circuit topologies dont lend themselves to a wholly
direct extractionalgorithm, however theDambrinemodel canbetakenas thestarting
point and strategies developed to deal with the additional elements in a structured
manner [153].
As mentioned above, direct extraction performs well for devices of modest size.
As thedevicesizeis increasedaboveafewmmof gateperiphery thedevicebecomes
increasinglydistributedinnature, andalsothedevicemeasurementsbecomelessreliable
duetothelowimpedancelevel thatresults. Verylowimpedancesaredifculttomeasure
accurately in a 50 O systemand the obvious solution of scaling up smaller device
measurementstoreplicateaverybigdeviceisnot straightforward. A particular issueis
that thethermal environments canbevery different [154]. However, for themost part,
atleastfor devicesthataredirectlymeasurable, thedevelopmentof small-signal models
isareliableactivity.
2.5.3 Large-signal models
Thesituationfor large-signal modelsislessstraightforward. Herethedesireistomodel
thedeviceresponsetoanarbitrarysignal. Thedifcultyof thischallengeisperhapsnot
always fully appreciated. Thenormal approachistouselarge-signal equivalent circuit
modelswhichhavebeencreatedbytransformingaset of bias-dependent linear models
into a single nonlinear one. At the heart of this approach, at least for commonplace
models, lies thequasi-static assumption whereit is assumed that theinstantaneous
86 GaAs FETs physics, design, and models
values of the equivalent circuit elements are uniquely dependent on their controlling
voltages[155]. Inother wordsthedeviceisassumedtobememory-less. Ashasalready
been discussed this actually isnt the case and both thermal effects and dispersion
effectscausethisassumptiontobeviolated. Consequently, large-signal modelsemploy
various measures in attempt to account for these effects and a range of large-signal
model formulations havebeen developed over theyears. They typically havefocused
onparticular aspects of devicebehavior andaDarwinianprocess of natural selection,
skewedbythechoicesof modelsthat thesimulator vendorshavechosentosupport has
resultedinarangeof modelscomingtothefore.
A FET model for PA applicationsshouldpossessthefollowingattributes:
1. ReplicatetheDC IVwell inorder toaccuratelyreproducethecorrect biaspoint.
2. Account for theeffectsof dispersionsothat theRF swingisaccuratelymodeled.
3. Properlyrepresent thebias-dependenceof thecapacitances.
4. Accurately reproducedifferentialsof thebiasdependent parametersaswell astheir
absolutevalues.
5. Includetheimpact of temperatureondevicecharacteristics.
6. Includetimedelaysfor high-frequencymodels.
Thecommonly availablemodelstendtohavestrengthsthat haveaddressedasubset of
theaboverequirementsandthereisnoobviousbest model. For example, theTriquint
TOMseriesof modelsintroducedaneffectiveapproachtomodel thebiasdependence
of thedrainconductance[156] whereastheAngelov/Chalmersmodel [157] isnotablein
includingthecharacteristicpeakinthetransconductancedependenceongatebiastypical
of HEMT devices. The IV characteristic is the focus of the ParkerSkellern model
[158] which has a exible functional formand well-behaved continuous derivatives.
Many models include dispersion effects with various degrees of sophistication with
the extended Angelov [159, 160] and ParkerSkellern models being comprehensive
examples. Itisnotunusual for modelstoconcentrateheavilyonthedeviceIV, however
for accurate simulation of linearity as frequencies increase then the accuracy of the
reactiveelementsisalsoimportant [161]. TheAngelov [160] andTOM3models[162]
havecomprehensivebias dependent capacitancemodels. Another development is the
(unpublished) Aurigamodel whichisafurther development of theAngelovmodel and
whichclaimsimprovedcapacitancemodelsandamodieddraincurrentequation[163].
The popular EEFET3 and EEHEMT models [164] bring together a number of these
featuresintwowidelyusedforms.
Anappreciation has grownwithtimeof theimportanceof chargeconservation for
the gate capacitance. As discussed in Section 2.2.7, the gate depletion region is a
singleentity but is accessed by all threeterminals. It is seemingly straightforward to
extend thesmall-signal model to employ two bias-dependent capacitors C
gs
(V
gs
, V
ds
)
and C
gd
(V
gs
, V
ds
) to represent the depletion reactances with independent charge or
capacitancefunctions of theremotecontrollingvoltages. However, this approachwill
generally result inanon-conservativesystemso that thetotal gatechargeis (V
gs
, V
ds
)
pathdependent. Theconsequencesof thisarethepossibility of anunintendednet gate
current [165] and, as circuit simulators are required to maintain charge-conservation
2.5 Models 87
at each node, then simulator non-convergence and spurious results can occur. Other
work hasdemonstratedthat chargeconservationisimportant for accuratepredictionof
nonlinear effects [166, 167]. Two approaches to resolvetheproblemarepossible. The
direct andconceptuallystraightforwardapproachistoconstruct themodel intermsof a
singlegatechargeentitythat isafunctionof thelocal variablesV
gs
andV
gd
. Thisisthe
approachusedinthewidelyavailableTOM3model [162]. Thechargefunctioncannotbe
directlymeasuredandmust beinferredfromthesmall-signal C
gs
andC
gd
capacitances.
The resulting model is fundamentally and unequivocally charge-conservative [167].
Alternatively, the charge may be separated into independent functions Q
gs
(V
gs
, V
ds
)
andQ
gd
(V
gs
, V
ds
). To achievechargeconservationtheseelements must beaugmented
by additional chargecontrol elements called trans-capacitances which arerequired in
order toproperly account for thecontributions tothepartitionedreactivecurrents that
arise fromboth controlling voltages [168, 165, 169]. An example of a gate-charge
model employing separategate-sourceand gate-drain functions and employing trans-
capacitancetorestorechargeconservationis theformulationusedintheEEFET3and
EEHEMT models. The functions employed are charge-conservative in the saturation
regionof thedeviceIVandsoarevalidfor power amplier circuits. However, theuse
of smoothingfunctions inorder toforcesymmetrical behavior of thechargefunctions
aroundV
ds
= 0resultsinnon-physical (negative) drain-sourcecapacitanceinthelinear
(i.e., subknee) region[170].
Large-signal models aredevelopedby ttingthemodel equations to measureddata
by numerical optimization. Thenumber of tting parameters can beextensiveand so
toobtaingoodmodels robust methodologies arerequiredtosegment theprobleminto
parameter subsetsandtoselect goodinitial values. Themost straightforwardprocedure
istousemeasuredDC datafor theIVequationandtousebias-dependent S-parameter
data to extract the charge functions and to model the correction terms necessary to
modify thedynamic responseof theIV. Theimpact of dispersioncanbeasignicant
sourceof error for PA designandamoreaccurateapproachcanbetodirectly measure
thedynamic IV withapulsedIV measurement system[21, 22, 171, 172] andtouse
that to represent the model IV [173]. This approach provides a direct model of the
dynamic IV at thequiescent bias point of interest and avoids theneed to develop a
complex empirical correction factor. Themain drawback of this approach is that the
resultingIV model is no longer applicablefor thewholebias planebut is specic to
operationpointsinthevicinityof thequiescent point inthepulsedIVset.
Thetraditional compact devicemodels generally do areasonablejob of describing
rst-order amplier performance and adequately represent the terminal impedances
andpower saturationbehavior. However, they areusually less successful at next-level
parameters suchas linearity measures, andusually do not includesuchrenements as
self-heatingor accuratebias-dependenceof trappingphenomena. Theseshortcomings
have inspired a lot of efforts spanning many years to enhance commonly available
models.
A substantial degree of improvement was obtained by augmenting the quasi-static
modelswithcorrectivetermsfor trappingeffects. Measurestoaccomplishthisinclude
empirical methodstomodifythelarge-signal IVresponsebymeansof correctionstothe
88 GaAs FETs physics, design, and models
dynamicoutputconductance[174]. Laterdevelopmentshavebeentomakeuseof pulsed
IV datato frametheconstructionof morephysics-basedtrappingmodels [175177].
Thermal effectsarealsobeincludedinthiswork bytheuseof self-heatingtermsanda
thermal impedancemodel [178]. Thenextstageof thisapproachistoincludetheimpact
of theself-heatingontothetrappingstatebyincludingatemperaturedependencetothe
trapcorrectionterms[179, 180].
Therenementstothetrappingmodelshavesuccessivelyimprovedthestate-of-artin
thisaspectof modeling. However, anareasomewhatlessdevelopedistheprovisionof a
satisfactoryapproachforthemodelingof largeperipherydevices. A commonexperience
isthatconventional circuitmodel approachesdonotsatisfactorilyscaleaboveafewmm
of gatewidth. However, suchdevices arerequiredfor numerous applications intheL
to C bandrange. GaAs devices with100200mmof periphery arecapabledelivering
50100 W froma single die at 2 GHz but the circuit design approaches are largely
cut and try. Conventional modeling approaches are ill-equipped to cope with this
distributedproblemandanumber of newdimensions needto beaddedif models are
to beadequate. Such devices havemany ngers frequently over ahundred which
seedifferingandcoupledthermal andelectromagnetic(EM) environments. A particular
problemis alsotomodel accurately thestability of suchdevices. Thecompletemodel
for thishighly complex situationisacoupledelectro-thermal andEM model. Thereis
anemergingbodyof academicliteratureonthistopicwithrecentdevelopmentstowards
simplication and manageable computation speed. Reference [181] describes circuit
simulation softwarecoupled with ahighly efcient thermal solver. Individual ngers
of apower cell aremodeled with acompact equivalent circuit model and thengers
arethermally coupledby athermal circuit. Reference[182] specically addresses the
topic of large power devices with a similar approach that also includes coupled EM
simulation. Thelatter approachisnotablefor beingdeployableonstandardcommercial
simulators.
Another directioninmodelingaims to avoidthewholemessy business of tryingto
persuadeanequivalentcircuitmodel withelementsthatfollowprescribedbias-dependent
functionsintorepresentingthemeasureddata. Analternativeapproachinsteademploys
table-basedtechniqueswherethenonlinear dataisaccessedandinterpolatedfromlook-
up tables. The most well-known implementation of this approach is the Root model
[183] availableasaturn-keycommercial modelingsolution. Subsequent developments
of this approachmakeuseof moresophisticatedinterpolationschemes whichprovide
better simulationof nonlinearity[184, 185].
A furtherthemereceivingmuchrecentattentionhasbeentheuseof directlarge-signal
measurement [186]. Techniques explored hereincludetting conventional equivalent
circuit model parametersdirectlytoobservedlarge-signal behavior [187] andthedirect
extractionof extrinsic current andchargefunctions to describethenonlinear behavior
directly at thedeviceterminals [188]. However, theultimatelogical end-point of this
directionistoeliminateanylevel of equivalentcircuitdescriptionaltogether andinstead
to implement a wholly mathematical black-box or behavioral description of the
data. The approach that has been adopted to achieve this is based on poly-harmonic
distortion(PHD) modeling[189] whichdescribeslarge-signal behavior bymeansof an
2.5 Models 89
extensiontolinear S-parameters. Inthisschemeadditional termsareaddedtothelinear
parameters to account for harmonics and intermodulation frequency components. By
this means complex waveforms may be described. The termX-parameters has been
coined to describe the new nonlinear parameters and a commercial nonlinear vector
network analyzer capable of their measurement are available. In order to be useful
for characterizingtransistors theX-parameters must bemeasuredover theappropriate
regionof theSmithchartrequiringtheX-parametercharacterizationtobecombinedwith
aload-pull system[189, 190]. Theresultingdataset is largerequiringthedimensions
of frequency, bias, signal amplitude and impedance state all to be characterized and
recorded. However, theapproach is mathematically rigorous and has been veried to
highlevelsof compression[191].
In many ways conventional empirical models and new behavioral data models are
complementary. The former possess such benets of an innate generality, compact-
nessandscalabilitybyvirtueof theunderlyingphysicsembodiedintheir construction.
Theseareattractivequalitiesfor devicemanufacturerswhoneedtocharacterizeapro-
cess inageneral fashion. They alsomakeuseof relatively straightforwardandwidely
availabletest equipment. Black-box behavioral models offer theprospect of automatic
generationof high-accuracymodelsfor specicdevicesandoperatingconditions. This
latter pictureisattractivefor specicdesignrequirementsfocusedonparticular devices
wherethegenerationof largedatasets andalack of model scalability aremanageable
issues.
2.5.4 Load-pull
A long established, pragmatic, and reliable alternative to the nonlinear device model
is thelong-standing load-pull measurement. Herethedeviceperformanceis explored
with carefully characterized tuners and thecircuit is designed to replicatethedesired
matchingimpedances. Thisapproachhasevolvedtoemploycomputer controlledtuners
thatarenowabletoincludeeffectivecontrol of thesourceandloadharmonicimpedances
[192]. Thekey limitationof thepassivetunersprovidedby commercial vendorsisthat
thelosses arising between thetuner instrument and thedeviceunder test restricts the
maximumreectioncoefcient that canbeattainedandthisrather limitsthesizeof the
devicethat canbecharacterized. Pre-matchingcircuitry canhelpherebut theultimate
solutionis achievedby theuseof anactiveload-pull systemsuchas that describedin
reference[193] wherethereectedsignal issynthesizedasrequiredtoaccount for the
loss so as to achieve theeffective impedance as if an ideal lossless tuner wereused.
Activeload-pull systemshavebeensuccessfullydemonstratedbyanumber of workers
over manyyearsbuttheyhaveyettoachievewidespreaduseoutsideof theR&Dlabdue
toreasonsof cost andcomplexity. Inthislight it isinterestingtonotethat aremarkably
effectivebut simpleand extremely low-cost alternativeto load-pull test equipment is
availableusingnothingbutasimplelinearmodel forthedeviceoutputimpedance[194].
Experiencehasshownthat theestimatefor theoutput power matchconditionobtained
fromthis approachis consistently inexcellent agreement withload-pull measurement
andthetechniqueremainsapopular approachfor rst-cut circuit design.
90 GaAs FETs physics, design, and models
2.6 Concluding remarks
Thischapter hasreviewedGaAspower FET technology. It hascoveredmaterialsprop-
erties, devicetypesandtheir operation, keydevicephysics, andcritical aspectsof power
devicedesign. A summary of GaAs devicefabrication has been provided contrasting
establishedprocesses withnewlow-cost approaches andthechapter concludes witha
brief reviewof devicemodels.
Thewiderangeof subjectscoveredspansseveral decadesof developmentbynumerous
researchgroups andindustrial companies. It is thereforeimpossibleto fully reference
suchabodyof workandabalancehadtobestruckthatweighedrecognitionof historical
signicancewithclarity andbrevity for thecontemporary reader. A further limitation
is this authors limitedcapacity toread, digest andretainthebreadthof literature. It is
consequentlyinevitablethat omissionswill haveoccurredandfor whichit ishopedthe
relevant partieswill understand.
Acknowledgment
Theauthor wouldliketo thank theengineeringteamat theRFMD facility inNewton
Aycliffe(memberspast andpresent) whohaveall contributedtotheunderstandingthat
is containedinthesepages. Particular appreciationis expressedto MikeBrookbanks,
RichardDaviesandRobDrywhogavehelpful support inthewritingof thechapter.
Appendix 2.1
Commentsonthedeterminationof f
T
andf
max
Theh
21
functionisgenerallyverywell behavedandiseasytocalculateinanunambigu-
ousmanner. Thisshouldbeperformedfromalinear partof theh
21
versuslog. frequency
curvewheretherst poleof thefrequencyresponsedominates. For amicrowavedevice
afrequencyof around510GHz istypicallyagoodfrequencytouse. Anextrapolation
to0dB at 6dB/octavewill giveareliablevalueandthismaybesimplycalculatedthus:
f
T
= f h
21
( f ) (A1)
= f 10
h
21
dB( f ),20
(A2)
where f is the frequency of evaluation, h
21
is the magnitude of the forward hybrid
parameter and h
21
dB is its value in dB. The measurements should be properly de-
embeddedtothedevicereferenceplanefor theappropriateterminatingimpedances to
apply. A useful ruleof thumb for awell-designed deviceis that it will likely bean
appropriatechoicefor anapplicationfor frequenciesuptof
T
,2.
Thesituationforf
max
isnotasstraightforward. IntheorythereisnoissueasbothGmax
andU both cross the0dB lineat thesamefrequency andthis point uniquely denes
f
max
[60, 195]. However, in practice for microwave FETs, f
max
is usually somewhat
References 91
higher thantheupper limit of theavailabletest equipment andexaminationof thegain
curvesforfrequencieswell belowf
max
invariablysuggestsGmaxandUwill havedistinct
intercepts. Thesituation typically observed is that of Figure2.14 with U following a
well-behaved6dB/octaveroll off andgivingno hint that it will convergewithGmax.
Vendelin[60] explains howadditional terms inthefrequency responsewill ultimately
restraintheUcurve; however, ajudgmenthastobemadeonthedatathatisavailable. In
practiceoneof theU or Gmaxcurvesischosenandextrapolatedtodeterminethe0dB
intercept. Itisalsoverycommonnottospecifywhichcurvewasusedfor thisprocedure.
Manyworkerschoosea6dB/octaveextrapolationof U. For somethisisduetoabelief
that it is thecorrect oneor becauseof its apparent well-behavedslope. Others choose
Gmax and return a commendably more conservative value; however the complicated
Gmaxcurveprovidesambiguityastohowit shouldbeextrapolated. Giventheseissues
thereis agood argument not to quotef
max
at all but to provideexplicit Gmax curves
or quoteGmaxat particular frequencies. Inany event, adegreeof cautionis required
when comparing devices based on f
max
values that onehas not measured for oneself.
Shouldavaluefor f
max
berequiredanditisbeyondthefrequencyrangeof availabletest
equipmentthen, inthisauthorsopinion, areasonableapproachtoitsdeterminationisto
extrapolateGmaxat 6dB/octavefromafrequency wherethedeviceisunconditionally
stableandwithstabilityfactor, k, comfortablyaboveunitysoastobereliablyfreeof the
gain-peakingnear theMAG/MSGstabilitybreak-point.
References
1. Y. Aoki, Y. Hirano, High-power GaAs FETs, in High-power GaAs FET Ampliers,
J. L. B. Walker. Ed., ArtechHouse, 1993, pp. 43145.
2. IoffePhysico-Technical Institute, Physical propertiesof semiconductors, [Online]. Avail-
able: http://www.ioffe.ru/SVA/NSM/Semicond/.
3. S. M. Sze, Physicsof Semiconductor Devices, 2ndEdn., Wiley, 1981.
4. L. Dobaczewski, A. R. Peaker, and J. M. Langer, DX defect centres in AlGaAs, in
Properties of AluminumGalliumArsenide, S. Adachi, Edn., IET/INSPEC, 1993, pp. 278
288.
5. J.W. Matthews and A.E. Blakeslee, Defects in epitaxial multilayers, J. Crystal Growth,
vol. 27, pp. 118127, 1974.
6. J.V. DiLorenzo, W.R. Wisseman, GaAs power MESFETs: design, fabricationandperfor-
mance, IEEE Trans. Microw. TheoryTech., vol. 27, pp. 367378, 1979.
7. E.O. J ohnson, Physical limitationsonfrequencyandpower parametersof transistors,IRE
Int. Conv. Rec., 13, pp. 2734, 1965.
8. P. Sauiner, W. S. Kopp, H. Q. Tserng, Y. C. Kao, and D. D. Heston, A heterostructure
FET with 75.8% power added efciency at 10 GHz, IEEE MTT-S, Int. Symp. Digest,
pp. 635638, 1992.
9. S. Cooper, K. Anderson, K. Salman, R. Culbertson, J. Mason, D. Bryant, and P. Saunier,
8-watt high efciency X-band power amplier using AIGaAs/GaAs HFET technology,
GaAsIC Symposium, 1992, pp. 183185.
92 GaAs FETs physics, design, and models
10. I. Takenaka, K. Ishikura, H. Takahashi, K. Asano, J. Morikawa, K. Satou, K. Kishi, K.
Hasegawa, K. Tokunaga, F. Emori, andM. Kuzuhara, L/S-band140-W pushpull power
AlGaAs/GaAs HFETs for digital cellular base stations, IEEE J. Solid-State Circuits,
vol. 34, pp. 11811187, 1999.
11. M. Yang, and Y. Chan, Device linearity comparisons between doped-channel and
modulation-doped designs in pseudomorphic AlGaAs/InGaAs heterostructures, IEEE
Trans. ElectronDevices, vol. 43, pp. 1174 1180, 1996.
12. J. Morikawa, K. Asano, K. Ishikura, H. Oikawa, M. Kanamori and M. Kuzuhara, 60
W L-band power AlGaAs/GaAs heterostructure FETs, IEEE MTT-S Int. Symp. Digest,
pp. 14131416, 1997.
13. Y. C. Lin, EdwardYi Chang, H. Yamaguchi, Y. Hirayama, X. Y. Chang, andC. Y. Chang,
Device linearity comparison of uniformly doped and -doped In
0.52
Al
0.48
As/In
0.6
Ga
0.4
As
metamorphicHEMTs, IEEE ElectronDeviceLett., vol. 27, pp. 535537, 2006.
14. D. Geiger, E. Mittermeier, J. Dickmann, C. Geng, R. Winterhof, F. Scholz, and E. Kohn,
InGaP/InGaAsHFET withhighcurrentdensityandhighcut-off frequencies, IEEEElectron
DeviceLetters, vol. 16, pp. 259261, 1995.
15. T. J. Drummond, W. T. Masselink, H. Morkoc, Modulation-dopedGaAs/(Al,Ga)Ashetero-
junctioneld-effect transistors: MODFETs, Proc. IEEE, 74, pp. 773882, 1986.
16. L. D. Nguyen, L. E. Larson, U. K. Mishra, Ultra-high-speedmodulation-dopedeld-effect
transistors: atutorial review, Proc. IEEE, vol. 80, pp. 494518, 1992.
17. J.V. DiLorenzo, B. D. Laterwasser, andM. Zaitlin, IntroductiontopHEMTs,inR. L. Ross
et al. (Eds.), Proceedingsof theNATOAdvancedStudyInstituteonPseudomorphicHEMT
TechnologyandApplications, 1994, pp. 121.
18. N. Moll, M. R. HueschenandA. Fischer-Colbrie, Pulse-dopedAlGaAs/InGaAs pseudo-
morphicMODFETs, IEEE Trans. ElectronDevices, vol. 35, pp. 879886, 1988.
19. G. Snider, 1DPoisson, [Online.] Available: http://www.nd.edu/gsnider.
20. P. H. Ladbrooke, MMIC Design: GaAsFETsandHEMTs, ArtechHouse, 1989.
21. P. H. Ladbrooke, J. P. Bridge, N. J. GoodshipandD. J. Battison, Improvingunderstanding
of theRF circuit behaviour of contemporary semiconductor devicesthroughfast-sampling
I(V) curvetracer measurements, GalliumArsenideApplicationsSymposium(GAAS), 2000.
22. L. Dunleavy, W. Clausen, andT. Weller, PulsedIV for nonlinear modeling, Microw. J.,
vol. 46, pp. 6884, 2003.
23. M. Rocchi, Status of thesurfaceand bulk parasitic effects limiting theperformances of
GaAsICs, PhysicaB., vol. 129, pp. 119138, 1985.
24. S. Lo and C. Lee, Analysis of surface state effect on gate lag phenomena in GaAs
MESFETs, IEEE Trans. ElectronDevices, vol. 41, pp. 15041512, 1994.
25. P. H. Ladbrooke, andS. R. Blight, Low-eldlow-frequencydispersionof transconductance
in GaAs MESFETs with implications for other rate-dependent anomalies, IEEE Trans.
ElectronDevices, vol. 35, pp. 257267, 1988.
26. C. Canali, F. Magistrali, A. Paccagnella, M. Sangalli, C. Tedesco, and E. Zanoni, Trap-
relatedeffectsinAlGaAslGaAsHEMTs, IEE Proc., Part G, vol. 38, pp. 104108, 1991.
27. S. C. Binari, P. B. Klein, and T. E. Kazior, Trapping effects in GaN and SiC microwave
FETs, Proc. IEEE, vol. 90, pp. 10481058, 2002.
28. Y. Hasumi, N. Matsunaga, T. Oshima, andH. Kodera, Characterizationof thefrequency
dispersion of transconductance and drain conductance of GaAs MESFET, IEEE Trans.
ElectronDevices, volume50, pp. 20322038, 2003.
References 93
29. T. Izumi, T. Ohshima, M. Tsunotani andT. Kimura, Newmethodtomonitor thefrequency-
dispersion in InGaAs/AlGaAs PHEMTs, GaAs MANTECH International Conferenceon
CompoundSemiconductor Manufacturing, 2002.
30. A.F. Basile, A. Mazzanti, E. Manzini, G. Verzellesi, C. Canali, R. Pierobon, andC. Lanzieri,
Experimental andnumerical analysisof gate- anddrain-lagphenomenainAlGaAs/inGaAs
pHEMTs, IEEE International SymposiumonElectronDevices for MicrowaveandOpto-
electronicApplications(EDMO), 2002, pp. 6368.
31. G. Verzellesi, A. Mazzanti, A. F. Basile, A. Boni, E. Zanoni, andC. Canali, Experimental
andnumerical assessment of gate-lagphenomenainAlGaAsGaAs heterostructureeld-
effect transistors(FETs), IEEE Trans. ElectronDevices, vol. 50, pp. 17331740, 2003.
32. O. Pajona, C. Aupetit-BerthelemotandJ. M. Dumas, Modellingof thetraprelatedparasitic
effects inmetamorphic HEMT onGaAs substrate, EleventhInternational Symposiumon
ElectronDevicesfor MicrowaveandOptoelectronicDevices, 2003, pp. 151156.
33. F. Wang, W. D. J emison, andJ. C. M. Hwang, A GaAsMESFET transientmodel capableof
predictingtrap-inducedeffectsunder complexdigital modulation, IEEE MTT-SInt. Symp.
Dig., pp. 815818, 2001.
34. J. Haruyama, H. Negishi, Y. Nishimura, andY. Nashimoto, Substrate-relatedkink effects
withastronglight-sensitivityinAlGaAs/InGaAsPHEMT, IEEE Trans. ElectronDevices,
vol. 44, pp. 2533, 1997.
35. J. B. Kuang, P. J. Tasker, G. W. Wang, Y. K. Chen, L. F. Eastman, O. A. Aina, H. Hier, and
A. Fathimulla, Kink effect in submicrometer-gate MBE-Grown InAlAs/InGaAs/InAlAs
heterojunctionMESFETs, IEEE ElectronDeviceLett., vol. 9, pp. 630632, 1988.
36. P. H. LadbrookeandJ. P. Bridge, BenignmechanismgivingrisetokinksinGaAsMESFET
andHEMT I(V) characteristics, Electron. Lett., vol. 31, pp. 19471948, 1995.
37. J.-W. Chen, M. Thurairaj, and M. B. Das, Optimization of gate-to-drain separation in
submicron gate-length modulation doped FETs for maximumpower gain performance,
IEEE Trans. ElectronDevices, vol. 41, pp. 465475, 1994.
38. T. Suemitsu, T. Enoki, N. Sano, M. Tomizawa, and Y. Ishii, An analysis of the kink
phenomenain InAlAs/InGaAs HEMTs using two-dimensional devicesimulation, IEEE
Trans. ElectronDevices, vol. 45, pp. 23902399, 1998.
39. M. H. Somerville, A. Ernst, andJ. A. del Alamo, A physical model for thekink effect in
InAlAs/InGaAsHEMTs, IEEE Trans. ElectronDevices, vol. 47, pp. 922930, 2000.
40. A. Mazzanti, G. Verzellesi, C. Canali, G. Meneghesso, and E. Zanoni, Physics-based
explanation of kink dynamics in AlGaAs/GaAs HFETs, IEEE Electron Device Letters,
vol. 23, pp. 383385, 2002.
41. M. H. Somerville, J. A. del Alamo, andP. Saunier, Off-statebreakdowninpowerpHEMTs:
theimpact of thesource, IEEE Trans. ElectronDevices, vol. 45, pp. 18831889, 1998.
42. H. P. Li, O. L. Hartin, andM. Ray, Anupdatedtemperature-dependentbreakdowncoupling
model including both impact ionization and tunneling mechanisms for AlGaAs/InGaAs
HEMTs, IEEE Trans. ElectronDevices, vol. 49, pp. 16751678, 2002.
43. R. Menozzi, Off-statebreakdownof GaAsPHEMTs: reviewandnewdata, IEEE Trans.
DeviceandMaterialsRel., vol. 4, pp. 54 62, 2004.
44. R. J. TrewandU. K. Mishra, GatebreakdowninMESFETsandHEMTs, IEEE Electron
DeviceLett., vol. 12, pp. 524526, 1991.
45. S. R. Bahl, andJ. A. del Alamo, Physicsof breakdowninInAlAs/n

-InGaAsheterostructure
eld-effect transistors, IEEE Trans. ElectronDevices, vol. 41, pp. 22682275, 1994.
94 GaAs FETs physics, design, and models
46. J. A. del AlamoandM. H. Somerville, Breakdowninmillimeter-wavepower InP HEMTs:
acomparisonwithGaAsPHEMTs, IEEE J. Solid-StateCircuits, vol. 34, pp. 12041211,
1999.
47. M. H. Somerville, R. Blanchard, J. A. del Alamo, K. G. Duh and P. C. Chao, On-state
breakdowninpower HEMTs: measurementsandmodeling,IEEETrans. ElectronDevices,
vol. 46, pp. 10871093, 1999.
48. K. vanderZanden, D. M. M.-P. Schreurs, R. Menozzi, andM. Borgarino, Reliabilitytesting
of InP HEMTs using electrical stress methods, IEEE Trans. Electron Devices, vol. 46,
pp. 15701576, 1999.
49. A. Di Carlo, L. Rossi, P. Lugli, G. Zandler, G. Meneghesso, M. J ackson and E. Zanoni,
MonteCarlo study of thedynamic breakdown effects in HEMTs, IEEE Electron Device
Letters, vol. 21, pp. 149151, 2000.
50. A. Sleiman, A. Di Carlo, P. Lugli, G. Meneghesso, E. Zanoni, andJ. L. Thobel, Channel
thicknessdependenceof breakdowndynamicinInP-basedlattice-matchedHEMTs, IEEE
Trans. ElectronDevices, vol. 50, pp. 20092014, 2003.
51. K. Bock, C. Russ, G. Badenes, G. Groeseneken, and Inuence of well prole and gate
length on the ESD performance of a fully silicided 0.25 mCMOS technology, IEEE
Trans. Components, PackagingandManufacturingTechnol., Part C, vol. 24, pp. 286294,
1998.
52. K. G. Verhaege, M. Mergensb, C. Russb, J. ArmerbandP. J ozwiak, Novel designof driver
andESDtransistorswithsignicantlyreducedsiliconarea, MicroelectronicsRel., vol. 42,
pp. 313, 2002.
53. R Menozzi, P. Cova, C. Canali, and F. Fantini, Breakdown walkout in pseudomorphic
HEMTs, IEEE Trans. ElectronDevices, vol. 43, pp. 543546, 1996.
54. J. Verspecht, and D. Schreurs, Measuring transistor dynamic loadlines and breakdown
currents under large-signal high-frequency operatingconditions, IEEE MTT-SInt. Symp.
Dig., pp. 14951498, 1998.
55. J. P. R. David, J. E. Sitch, andM. S. Stern, GatedrainavalanchebreakdowninGaAspower
MESFETs, IEEE Trans. ElectronDevices, vol. 29, pp. 15481552, 1982.
56. M. S. Shirokov, R. E. Leoni, C. J. Wei, and J. C. M. Hwang, Breakdown effects on the
performanceandreliabilityof power MESFETs, GaAsIC Symposium, 1996, pp. 3437.
57. P. H. LadbrookeandJ. E. Carroll, Dielectric relaxationas alimit ontransistor switching
speed, Electron. Lett., vol. 32, pp. 15111513, 1996.
58. G. D. Vendelin, Design of Ampliers and Oscillators by theS-parameter Method, Wiley-
Blackwell, 1982.
59. S. J. Mason, Power gaininfeedback amplier, Trans. IRE Professional GrouponCircuit
Theory, vol. 1, pp. 2025, 1954.
60. G. D. VendelinandS.-C. Shin, Applyingf
max
, f
t
, andf
mag
for microwavetransistor designs
at microwaveandmillimeter-wavefrequencies, IEEE Microw. Mag., pp. 8490, 2007.
61. S. I. Long, A comparison of the GaAs MESFET and the AlGaAs/GaAs heterojunc-
tionbipolar transistor for power microwaveamplication, IEEE Trans. ElectronDevices,
vol. 37, pp. 12741278, 1989.
62. H. M. Macksey, GaAs power FET design, inJ. V. DiLorenzo, D. D. Khandelwal (Ed.s),
GaAsFET PrinciplesandTechnology, ArtechHouse, 1982, pp. 257276.
63. T. Grave, Pseudomorphic HEMTs: device physics and materials layer design, in Ross
etal. (Eds), PseudomorphicHEMT Technologyandapplications,Proceedingsof theNato
AdvancedStudyInstitute, 1994.
References 95
64. H.M. Macksey, Optimisationof then ledgechannel structurefor GaAs power FETs,
IEEE Trans. ElectronDevices, vol. 33, pp. 18181824, 1986.
65. J. B. Boos, and W. Kruppa, InAlAs/lnGaAs/lnP HEMTs with high breakdown voltages
usingdouble-recessgateprocess, Electron. Lett., vol. 27, pp. 19091910, 1991.
66. J. C. Huang, G. S. J ackson, S. Shaneld, A. Platzker, P. K. Saledas, andC. Weichert, An
AlGaAs/InGaAspseudomorphichighelectronmobilitytransistorwithimprovedbreakdown
voltagefor X andKu-bandpower applications,IEEE Trans. Microw. TheoryTech., vol. 41,
pp. 752759, 1993.
67. Y. Kohno, H. Matsubayashi, M. Komaru, H. Takano, O. Ishihara, andS. Mitsui, Modeling
andsuppressionof thesurfacetrapeffect ondraincurrent frequency dispersions inGaAs
MESFETs, CompoundSemiconductor IC Symposium, pp. 263266, 1994.
68. W. Marsetz, A. Hiilsmann, T. Kleindienst, S. Fischer, M. Demmler, W. Bronner, T. Fink,
K. Kohler, and M. Schlechtweg, High performance double recessed Al
0.2
Ga
0.8
As/
In
0.25
Ga
0.75
As pHEMTs for microwave power applications, European Microwave Con-
ference, 1997, pp. 10301034.
69. C.-H. Chen, and J. Skogen, Improvement of GaAs MESFET performance using sur-
faceP-layer doping (SPD) technique, IEEE Electron DeviceLett., vol. 10, pp. 352354,
1989.
70. M. Hirose, K. Matsuzawa, M. Mihara, T. Nitta, A. Kameyama, andN. Uchitomi, A lightly
dopeddeepdrainGaAs MESFET structurefor linear ampliers of personal handy-phone
systems, IEEE Trans. ElectronDevices, vol. 43, pp. 20622067, 1996.
71. Yasuko Hori, Masaaki Kuzuhara, Yuji Ando, and Masashi Mizuta, Analysis of electric
elddistributioninGaAsmetal-semiconductor eldeffecttransistorwithaeldmodulating
plate, J. Appl. Physics, vol. 87, pp. 34833487, 2000.
72. K. Matsunaga, K. Ishikura, I. Takenaka, W. Contrata, A. Wakejima, and K. Ota, A low-
distortion230W GaAs power FP-HFET operatedat 22V for cellular basestation, IEEE
Int. ElectronDeviceMeeting(IEDM), 2000, pp. 393396.
73. N. Sakura, K. Matsunaga, K. Ishikura, I. Takenaka, K. Asano, N. J wata, M. Kanamori, and
M. Kuzuhara, 100W L-bandGaAs power FP-HFET operatedat 30V, IEEE MTT-SInt.
Symp. Dig., pp. 17151718, 2000.
74. A. Wakejima, K. Ota, K. Matsunaga, W. Contrata, and M. Kuzuhara, Field-modulating
plate(FP) InGaP MESFET withhighbreakdownvoltageandlowdistortion, IEEE Radio
FrequencyIntegratedCircuitsSymp. Dig., pp. 151154, 2001.
75. K. Inoue, M. Nagahara. N. Ui, H. Haematsu, S. Sano andJ. Fukaya, A highgainL-band
GaAs FET technology for 28 V operation, IEEE MTT-S Int. Symp. Dig., pp. 821824,
2004.
76. M. Miller, Design, performanceandapplicationof highvoltageGaAsFETs, Compound
Semiconductor IC Symposium, 2005, pp. 236239.
77. H.-C. Chiu, Y.-C. Chiang, and C.-S. Wu, High breakdown voltage AlGaInP/InGaAs
quasi-enhancement-modepHEMT witheld-platetechnology,IEEEElectronDeviceLett.,
vol. 26, pp. 701703, 2005.
78. R. L. Kuvas, Equivalent circuit model of FET including distributed gate effects, IEEE
Trans. ElectronDevices, vol. 27, pp. 11931195, 1980.
79. A. Higashisaka, Y Takayama, and A high-power GaAs MESFET with experimentally
optimizedpattern, IEEE Trans. ElectronDevices, vol. 27, pp. 10251029, 1980.
80. J. P. Mondal, Distributedscalingapproachof MESFETsanditscomparisonwithlumped-
element approach, IEEE Trans. Microw. TheoryTech. vol. 37, pp. 10851090, 1989.
96 GaAs FETs physics, design, and models
81. D. Teeter, S. Bouthillette, L. Aucoin, A Platzker, C. Alfaro, andD. Bradford, Highpower,
high efciency PHEMTs for use at 8 GHz, IEEE MTT-S Int. Symp. Dig., pp. 323326,
1995.
82. F. Hasegawa, Power GaAsFETs, inJ. V. DiLorenzo, D. D. Khandelwal (Eds.), GaAsFET
PrinciplesandTechnology, ArtechHouse, 1982, pp. 219255.
83. J. L. B. Walker, Combiningtechniques, inJ. L. B. Walker (Ed.), High-Power GaAsFET
Ampliers, ArtechHouse, 1993, pp. 263313.
84. I. J. Bahl, Design of power MMlCs and power combining techniques, International
WorkshoponIntegratedNonlinear MicrowaveandMillimeterwaveCircuits, 1994, pp. 71
91.
85. H. Derewonko, M. Laviron, andJ. Lepage, X- andKu-bandinternally matchedpackaged
GaAsF.E.T., Electron. Lett., vol. 15, pp. 89, 1979.
86. K. Honjo, Y. Takayama, andA. Higashisaka, Broad-bandinternal matchingof microwave
power GaAsMESFETS, IEEE Trans. Microw. TheoryTech. vol. 27, pp. 38, 1979.
87. Z. Shichang, C. Tangsheng, L. Gang, andL. Fuxiao, 8-wattinternallymatchedGaAspower
amplier at 1616.5GHz, International ConferenceonSolid-StateandIntegratedCircuit
Technology(ICSICT), 2006.
88. K. Mori, J. Nishihara, H. Utsumi, A. Inoue, andM. Miyazaki, X-Band14Whighefciency
internally-matchedHFET, IEEE MTT-SInt. Symp. Dig., pp. 315318, 2008.
89. S. T. Fu, J. J. Komiak, L. F. Lester, K. H. G. Duh, P. M. Smith, P. C. Chao, andT. H. Yu,
C-band20watt internallymatchedGaAsbasedpseudomorphicHEMT power ampliers,
GaAsIC Symposium, 1993, pp. 355356.
90. I. Takenaka, H. Takahashi, K. Asano, J. Morikawa, K. Ishikura, M. Kanamori, M. Kuzuhara
andH.i Tsutsui, HighefciencyS-band30Wpower GaAsFETs, IEEE MTT-SInt. Symp.
Dig., pp. 14171420, 1997.
91. A. Wakejima, T. Asano, T. Hirano, M. Funabashi, andK. Matsunaga, C-bandGaAs FET
power amplierswith70-Woutput power and50%PAE for satellitecommunicationuse,
IEEE J. Solid-StateCircuits, vol. 40, pp. 20542060, 2005.
92. C. Canali, F. Castaldo, F. Fantini, D. Ogliari, L. Umena, andE. Zanoni, Gatemetallization
sinking intotheactivechannel inTilWlAumetallizedpower MESFETs, IEEE Electron
DeviceLett., vol. 7, pp. 185187, 1986.
93. T. M. Kole, A comparativestudyof thermal measurementtechniquescurrentlyavailableto
thesemiconductor industry, Proc. GaAsReliabilityWorkshop, 2000, pp. 7998.
94. P. W. Webb, Thermal imagingof electronicdeviceswithlowsurfaceemissivity,IEEProc.,
Part G, vol. 138, pp. 390400, 1991.
95. Mark N. Minot, Thermal characterizationof microwavepower FETsusingnematicliquid
crystals, IEEE MTT-SInt. Symp. Dig., pp. 495498, 1986.
96. H. Fukui, Thermal resistanceof GaAseld-effect transistors, IEEE Int. ElectronDevice
Meeting(IEDM), 1980, pp. 118121.
97. P. W. Webb, Measurement of thermal resistance using electrical methods, IEE Proc.,
Part I, vol. 134, pp. 5156, 1987.
98. AgilentTechnologies, HighFrequencyTransistor Primer PartIII Thermal Properties, Appli-
cationNote59663084E.
99. I. Angelov, andC. K arnfelt, Directextractiontechniquesforthermal resistanceof MESFET
andHEMT devices, IEEE RadioFrequencyIntegratedCircuitsSymp. Dig., pp. 351354,
2007.
References 97
100. A. Sarua, H. J i, M. Kuball, M. J. Uren, T. Martin, K. P. Hilton, andR. S. Balmer, Integrated
micro-Raman/Infraredthermography probefor monitoringof self-heatinginAlGaN/GaN
transistor structures, IEEE Trans. ElectronDevices, vol. 53, pp. 24382447, 1986.
101. J. A. Mittereder, J. A. Roussos, W. T. Anderson, andD. E. Ioannou, Quantitativemeasure-
mentof channel temperatureof GaAsdevicesfor reliablelife-timeprediction,IEEE Trans.
Rel., vol. 51, pp. 482485, 2002.
102. P. W. Webb, and I. A. D. Russell, Thermal resistance of gallium-arsenide eld-effect
transistors, IEE Proc. Part G, vol. 136, pp. 229234, 1989.
103. P. W. Webb, Thermal modeling of power microwave integrated circuits, IEEE Trans.
ElectronDevices, vol. 40, pp. 867877, 1993.
104. J. Wilson, and K. Decker, GaAs MMIC thermal modeling for channel temperatures in
accelerated lifetest xtures and microwavemodules, Proceedings of theSemiconductor
Thermal Measurement andManagement Symposium(SEMI-THERM), 1994, pp. 121128.
105. W. Batty, Analytical solutionincomplicatedvolumesfor detailedcompact thermal model
construction, EuropeanMicrowaveIntegratedCircuitsConference, 2006, pp. 316319.
106. H. F. Cooke, PrecisetechniquendsFET thermal resistance, Microwave, pp. 8587, August
1986.
107. A.G.Kokkas,Thermal analysisof multiple-layerstructures,IEEETrans.ElectronDevices,
vol. 21, pp. 674681, 1974.
108. W. Marsetz, M. Dammann, H. Kawashima, J. Rtidiger, B. Matthes, A. Hiilsmann, and
M. Schlechtweg, Inuence of layout and packaging on the temperature of GaAs Power
PHEMTs, EuropeanMicrowaveConference, 1998, pp. 439442.
109. R. Williams, ModernGaAsProcessingMethods, ArtechHouse, 1990.
110. C. Y. Chang, andF. Kai, GaAsHigh-speedDevices: Physics, TechnologyandCircuit, Wiley,
1994.
111. B. C. Schmukler, P. E. Brunemeier, W. R. Hitchens, B. D. Cantos, W. A. Strier, D. H.
Rosenblatt, and R. D. Remba, Highly selective citric buffer etch-stop process for the
manufactureof very uniformGaAs/AlGaAs FETs, GaAs IC Symposium, 1993, pp. 325
328.
112. K. Alavi, S. Ogut, P. Lyman, W. Hoke, and M. Borkowski, A highly uniform, and high
throughput, double selective pHEMT process using an all wet etch chemistry, GaAs
MANTECHInternational ConferenceonCompoundSemiconductor Manufacturing, 2002.
113. A. W. Hanson, D. Danzilio, K. Bacher, and L. Leung, A selective gate recess process
utilizingMBE-grownInGaP etch-stoplayersfor GaAs-basedFET technologies, GaAsIC
Symposium, 1998, pp. 195197.
114. D. C. Hays, C. R. Abernathy, S. J. Pearton, F. Ren, andW. S. Hobson, Wet anddry etch
selectivity for theGaAs/AlGaAs and GaAs/InGaP systems, Electrochemical Soc. Proc.,
vol. 98, no. 12, pp. 202212, 1998.
115. F. Spooner, W. Quinn, L. Hanes, S. Woolsey, K. Smith, and J. Mason, A reproducible,
highyield, robustwetetchetch-stopprocessusingorganicacid peroxidesolutions,GaAs
MANTECHInternational ConferenceonCompoundSemiconductor Manufacturing, 2004.
116. E. Y. Chang, J. M. Van Hove, andK. P. Pande, A selectivedry-etchtechniquefor GaAs
MESFET gaterecessing, IEEE Trans. ElectronDevices, vol. 35, pp. 15801584, 1988.
117. F. Ren, S. J. Pearton, C. R. Abernathy, C. S. Wu, M. Hu, C.-K. Pao, D. C. Wang, andC. P.
Wen, 0.25-pmPseudomorphic HEMTs processedwithdamage-freedry-etchgate-recess
technology, IEEE Trans. ElectronDevices, vol. 39, pp. 27012706, 1992.
98 GaAs FETs physics, design, and models
118. A. E. Geissberger, I. J. Bahl, E. L. Grifn, andR. A. Sadler, A newrefractoryself-aligned
gate technology for gaas microwave power FETs and MMICs, IEEE Trans. Electron
Devices, vol. 35, pp. 615622, 1988.
119. M. Yanagihara. Y. Ota, K. Nishii, O. Ishikawa, andA. Tamura, HighlyefcientGaAspower
MESFETswithn asymmetrical LDDstructure, ElectronicsLett., vol. 28, pp. 686687,
1992.
120. A. T. Ping, W. Liebl, G. Mahoney, S. Mahon, and O. Berger, A high-performance0.13-
mAlGaAs/InGaAspHEMT processusingsidewall spacer technology,GaAsMANTECH
International ConferenceonCompoundSemiconductor Manufacturing, 2005.
121. M.-J. Hwu, H.-C. Chiu, S.-C. Yang, andY.-J. Chan, A novel double-recessed0.2mT-gate
processfor heterostructureInGaP-InGaAsdopedchannel FET fabrication, IEEE Electron
DeviceLetters, vol. 24, pp. 381383, 2003.
122. G. M. Metze, J. F. Bass, T. T. Lee, D. Porter, H. E. Carlson, andP. E. Laux, A dielectric-
denedprocess for theformationof T-gateeld-effect transistors, IEEE Microw. Guided
WaveLett., vol. 1, pp. 198200, 1991.
123. J.-E. Muller, T. Grave, H. J. Siweris, M. K arner, A. Schafer, H. Tischer, H. Riechert, L.
Schleicher, L. Verweyen, A. Bangert, W. Kellner, and T. Meier, A GaAs HEMT MMIC
chip set for automotiveradar systems fabricated by optical stepper lithography, IEEE J.
Solid-StateCircuits, vol. 32, pp. 13421349, 1997.
124. S. K. J ones, D. J. Bazley, D. R. Brambley, P. A. Claxton, I. R. Cleverley, I. Davies,
R. A. Davies, C. Hill, W. A. Phillips, N. M. Shorrocks, M. Stott, K. Vanner, R. H. Wallis,
andD. J. Warner, Processmodellingandsimulationfor GaAsP-HEMT gateimprovement
and control, GaAs MANTECH International Conference on Compound Semiconductor
Manufacturing, 2001.
125. E. Y. Chang, G. T. Cibuzar, and K. P. Pande, Passivation of GaAs FETs with PECVD
silicon nitride lms of different stress states, IEEE Trans. Electron Devices, vol. 35,
pp. 14121418, 1988.
126. K. D. Mackenzie, B. Reelfs, M. W. DeVre, R.l Westerman, and D. J. J ohnson, Char-
acterization and optimization of low stress PECVD silicon nitride for production GaAs
manufacturing, GaAsMANTECHInternational ConferenceonCompoundSemiconductor
Manufacturing, 2004.
127. M.C. Clausen, and J. McMonagle, Advanced manufacturing techniques for next gener-
ation power FET technology, GaAs MANTECH International Conferenceon Compound
Semiconductor Manufacturing, 2005.
128. D. Fanning, L. Witkowski, J. Stidham, H.Q. Tserng, M. Muir, andP. Saunier, Dielectrically
dened optical T-Gate for high power GaAs pHEMTs, GaAs MANTECH International
ConferenceonCompoundSemiconductor Manufacturing, 2002.
129. M. F. OKeefe, J. S. Atherton, W. B osch, P. Burgess, N. I. Cameron, andC. M. Snowden,
GaAspHEMT-basedtechnologyformicrowaveapplicationsinavolumeMMICproduction
environmenton150-mmWafers,IEEETransSemicond.Manuf.,vol.16,pp.376383,2003.
130. C.-G. Yuan, Y. Y. Hsieh, T. J. Yeh, C.-H. Chen, D. W. Tu, Y.-C. Wang, J. L. S. Murad,
R. Schook, F. Bontekoe, and M. Tomesen, Production ready ultra high breakdown 6
pHEMT technology, GaAsMANTECH International ConferenceonCompoundSemicon-
ductor Manufacturing, 2005.
131. M. F. OKeefe, J. G.E. Mayock, D. M. Brookbanks, J. McMonagle, andJ. S. Atherton, Step-
per basedsub-0.25mprocessfor mm-waveapplications,GaAsMANTECHInternational
ConferenceonCompoundSemiconductor Manufacturing, 2005.
References 99
132. T. Lodhi, J. McMonagle, R. G. Davis, D. M. Brookbanks, S. Combe, M. Clausen, M. F.
OKeefe, A. Collar, and J. S. Atherton, Deep UV stepper based 0.15 mhigh power
150mmGaAspHEMT processfor millimeter waveapplications, CompoundSemiconduc-
tor IC Symposium, 2006, pp. 125128.
133. S. M. J Liu, Y. Cheng-Guan, T. D.-W. R. Wu, J. Huang, Y. Shih-Wei, W. Lai, and P. Yu,
Optical stepper based150mmGaAsPHEMT for microwaveandmillimeter-waveMMIC
applications, IEEE Int. Conf. Microwaves, Communications, AntennasandElectronicSys-
tems(COMCAS), 2008, pp. 16.
134. C.-G. Yuan, S. M. Liu, D.-W. Tu, R. Wu, J. Huang, F. Chen, andY-C. Wang, 0.15micron
optical gate6 power pHEMT technology, GaAsMANTECH International Conferenceon
CompoundSemiconductor Manufacturing, 2009.
135. K. Fujii, J. Stanback, and H. Morkner, 40 to 85 GHz power amplier MMICs using an
optical lithographybasedlowcostGaAsPHEMT,EuropeanMicrowaveIntegratedCircuits
Conference, 2009, pp. 503506.
136. V. Steel, Lowcost packagingtechniquesfor commercial GaAsIC Components, GaAsIC
Symposium, 1996, pp. 1820.
137. T. Ho, F. Santos, R. Uscola, M. Szymanowski, andS. Marshall, A 900MHz, 200Wsilicon
LDMOS power amplier using integrated passive devices in a new over-molded plastic
package, IEEE MTT-SInt. Symp. Dig., 2009, pp. 12691272.
138. FreescaleApplicationNoteAN1902(2008).
139. V. A. Chiriac, T. T. Lee, andV. Hause, Thermal performanceoptimisationof radiofrequency
packagesforwirelesscommunication,J. Electron. Packaging, vol. 126, pp. 429434, 2004.
140. S. Krishnamoorthi, K. Y. Goh, Y. R. Chong, R. Kapoor, andY. S. Sun, Thermal character-
izationof athermally enhancedQFN package, Proceedings of theElectronics Packaging
TechnologyConference, 2003, pp. 485490.
141. K. Aihara, A. C. Chen, A. V. Pham, and J. W. Roman, Development of molded liquid
crystal polymer surfacemountpackagesfor millimeter waveapplications,Topical Meeting
onElectrical Performanceof ElectronicPackaging, 2005, pp. 167170.
142. R. Wormald, S. David, G. Panaghiston, andR. J effries, A lowcost packagingsolutionfor
microwaveapplications, EuropeanMicrowaveIntegratedCircuitsConference, 2006.
143. K. Aihara, M. J. Chen, andA.-V. Pham, Development of thin-lmliquid-crystal-polymer
surface-mount packages for Ka-band applications, IEEE Trans. Microw. Theory Tech.,
vol. 56, pp. 21112117, 2008.
144. Y.-H. Suh, D. Richardson, A. Dadello, S. Mahon, and J. T. Harvey, A low-cost high
performanceGaAs MMIC packageusingair-cavity ceramic quadat non-leadedpackage
upto40GHz, GalliumArsenideApplicationsSymposium(GAAS), 2005, pp. 545548.
145. P. Ersland, H.-R. J en, andX. Yang, Lifetimeaccelerationmodel forHASTtestsof apHEMT
process, Proceedingsof theGaAsReliabilityWorkshop, 2003, pp. 36.
146. W. J. Roesch, Thermal accelerationof compoundsemiconductorsinhumidity, Reliability
of CompoundSemiconductors(ROCS) Workshop, 2005, pp. 111121.
147. A. Extance, PA makers seek module standard, Compound Semiconductor, pp. 2526.
2009.
148. G. Dambrine, A. Cappy, F. Heliodore, andE. Playez, AnewmethodfordeterminingtheFET
small-signal equivalentcircuit,IEEE Trans. Microw. TheoryTech., vol. 36, pp. 11511159,
1988.
149. R. J. Trew, and M. B. Steer, Millimetre-wave performance of state-of-the-art MESFET,
MODFET andPBT transistors, Electron. Lett., vol. 23, pp. 149151, 1987.
100 GaAs FETs physics, design, and models
150. S. Akhtar andS. Tiwari, Non-quasi-statictransientandsmall-signal two-dimensional mod-
eling of GaAs MESFETs with emphasis on distributed effects, IEEE Trans. Electron
Devices, vol. 40, pp. 21542163, 1993.
151. P. J. Tasker and J. Braunstein, New MODFET small signal circuit model required for
millimeter-waveMMIC design: extraction and validation to 120 GHz, IEEE MTT-S Int.
Symp. Dig., pp. 611614, 1995.
152. G. Crupi, D. M. M.-P. Schreurs, A. Raffo, A. Caddemi, andG. Vannini, A newmillimeter-
wavesmall-signal modelingapproachfor pHEMTsaccountingfor theoutput conductance
timedelay, IEEE Trans. Microw. TheoryTech., vol. 56, pp. 741746, 2008.
153. D. M. Brookbanks, Measurement and modelling of high performance MESFET and
HEMT devices, West GermanyIEEE MTT/AP WorkshoponMeasurement Techniquesfor
MicrowaveDeviceCharacterisationandModellling, 1990, pp. 109121.
154. R. Hajji, J. Shumaker, andE. Camargo, 28V high-power GaAsFET large-signal modeling
achievespowerandlinearityprediction,IEEEMTT-SInt. Symp. Dig., pp. 10651068, 2004.
155. C. Rauscher, Simulation of non-linear microwaveFET performanceusing aquasi-static
model, IEEE Trans. Microw. TheoryTech., vol. 27, pp. 834840, 1979.
156. A. J. McCamant, G. D. McCormack, andAnimprovedGaAsMESFET model for SPICE,
IEEE Trans. Microw. TheoryTech., vol. 38, pp. 822824, 1990.
157. I. Angelov, H. Zirath, and N. Rorsman, A new empirical nonlinear model for HEMT
and MESFET devices, IEEE Trans. Microw. Theory Tech, vol. 40, pp. 22582266,
1992.
158. A. E. Parker, andD. J. Skellern, A realisticlarge-signal MESFET model for SPICE,IEEE
Trans. Microw. TheoryTech., vol. 45, pp. 15631571, 1997.
159. I. Angelov, L. Bengtsson, andM. Garcia, Temperatureanddispersioneffect extensionsof
theChalmers nonlinear HEMT and MESFET model, IEEE Radio Frequency Integrated
CircuitsSymp. Dig., pp. 15151518, 1995.
160. I. Angelov, L. Bengtsson, and M. Garcia, Extensions of theChalmers nonlinear HEMT
andMESFET model, IEEE Trans. Microw. TheoryTech., vol. 44, pp. 16641674, 1996.
161. J. Staudinger, M. C de Baca, and R. Vaitkus, An examination of several large signal
capacitance models to predict GaAs HEMT linear power amplier performance, IEEE
RadioandWirelessConference(RAWCON) 1998, pp. 343346.
162. R. B. Hallgren, andP. H. Litzenberg, TOM3capacitancemodel: linkinglargeandsmall-
signal MESFETmodelsinSPICE,IEEETrans. Microw. TheoryTech., vol. 47, pp. 556561,
1999.
163. Y. Tajima, ProgressinRF devicemodeling: fromMESFETstoGaNPHEMTs, Advanced
DeviceModelingSymposium, 2006.
164. Aglient Technologies, ADSDocumentation, Nonlinear Devices, 2007.
165. A. D. Snider, Chargeconservationandthetranscapacitanceelement: anexposition, IEEE
Trans. Educ., vol. 38, pp. 376379, 1995.
166. D. E. Root, Chargemodelingfor FET large-signal simulationandits importancefor IP3
and ACPR in communication circuits, Symposiumon Circuits and Systems (MWSCAS),
2001, pp. 678772.
167. M. Wren and T. J. Brazil, Enhanced prediction of pHEMT nonlinear distortion using a
novel chargeconservativemodel, IEEE MTT-SInt. Symp. Dig., pp. 3134, 2004.
168. D. E. Root, and B. Hughes, Principles of nonlinear active device modeling for circuit
simulation, ARFTGConf. Dig., vol. 14, pp. 124, 1988.
169. S. A. Maas, Nonlinear MicrowaveandRF Circuits, 2ndEdn., ArtechHouse, 2003.
References 101
170. D. M. Brookbanks, Privatecommunication.
171. M. Paggi, P. H. Williams, andJ. M. Borrego, Nonlinear GaAs MESFET modelingusing
pulsed gate measurements, IEEE Trans. Microw. Theory Tech., vol. 36, pp. 15931597,
1988.
172. A. Platzker, A. Palevsky, S. Nash, W. Struble, and Y. Tajima, Characterization of GaAs
devices by a versatile pulsed IV measurement system, IEEE MTT-S Int. Symp. Dig.,
pp. 11371140, 1990.
173. W.Struble,S.L.G.Chu,M.J.Schindler,Y.Tajima,andJ.Huang,Modelingintermodulation
distortionInGaAsMESFETsusingpulsedIVcharacteristics,GaAsICSymposium, 1991,
pp. 179182.
174. J. Staudinger, M. Golio, C. Woodin, andM. C. deBaca, Considerationsfor improvingthe
accuracy of large-signal GaAs MESFET models to predict power amplier circuit perfor-
mance, IEEE J. Solid-StateCircuits, vol. 29, pp. 366374, 1994.
175. Z. Ouarch, J. M. Collantes, J. P. Teyssier, and R. Quere, Measurement based nonlinear
electrothermal modelingof GaAsFET withdynamical trappingeffects, IEEE MTT-SInt.
Symp. Dig., pp. 599602, 1998.
176. O. J ardel, F. DeGroote, T. Reveyrand, J. J acquet, C. Charbonniaud, J. Teyssier, D. Floriot,
andR. Qu er e, Anelectrothermal model for AlGaN/GaNpower HEMTsincludingtrapping
effects to improve large-signal simulation results on high VSWR, IEEE Trans. Microw.
TheoryTech., vol. 55, pp. 26602669, 2007.
177. L. S. Liu, J. G. Ma, and G. I. Ng, Electrothermal large-signal model of IIIV FETs
accountingfor frequencydispersionandchargeconservation,IEEEMTT-SInt. Symp. Dig.,
pp. 749752, 2009.
178. L. Codecasa, D. DAmore, andP. Maffezzoni, Modelingthethermal responseof semicon-
ductor devicesthroughequivalent electrical networks, IEEE Trans. CircuitsandSystems
I: Fundamental TheoryandApplications, vol. 49, pp. 11871197, 2002.
179. A. Raffo, V. Vadal` a, G. Vannini, and A. Santarelli, A newempirical model for thechar-
acterization of low-frequency dispersive effects in FET electron devices accounting for
thermal inuence on the trapping state, IEEE MTT-S Int. Symp. Dig., pp. 14211424,
2008.
180. G. Verzellesi, A. Bade, A. Mazzanti, C. Canali, G. Meneghesso, and E. Zanoni, Impact
of temperature on surface-trap-induced gate-lag effects in GaAs heterostructure FETs,
Electron. Lett., vol. 39, pp. 810811, 2003.
181. S. Luniya, W. Batty, V. Caccamesit, M. GarciaC. Christoffersen, S. Melamed, W. R. Davis,
andM. Steer, Compact electrothermal modelingof anX-bandMMIC, IEEE MTT-SInt.
Symp. Dig., pp. 651654, 2006.
182. C. M. Snowden, Coupled electrothermal and electromagnetic modeling, simulation and
design of RF and microwave power FETs, Asia Pacic Microwave Conference, 2006,
pp. 295304.
183. D. E. Root, S. Fan, andJ. Meyer, Technologyindependentlargesignal nonquasi-staticFET
models by direct construction fromautomatically characterized device data, European
MicrowaveConference, 1991, pp. 927932.
184. I. Angelov, H. Zirath, andN. Rorsman, A newempirical nonlinear model for HEMT and
MESFET devices, IEEE Trans. Microw. TheoryTech., vol. 40, pp. 22582266, 1992.
185. J. Xu, D. Gunyan, M. Iwamoto, A. Cognata, andD. E. Root, Measurement-basednon-quasi-
staticlarge-signal FET model usingarticial neural networks,IEEEMTT-SInt. Symp. Dig.,
pp. 469472, 2006.
102 GaAs FETs physics, design, and models
186. D. E. Root, J. Xu, D. Gunyan, J. Horn, andM. Iwamoto, Thelarge-signal model: theoretical
andpractical considerations, trade-offs, andtrends, IEEE MTT-SInternational Symposium
WorkshopWMB, 2009.
187. D. M. M.-P. Schreurs, J. Verspecht, S. Vandenberghe, andE. Vandamme, Straightforward
andaccuratenonlinear devicemodel parameter estimationmethodbasedonvectorial large-
signal measurements, IEEE Trans. Microw. TheoryTech., vol. 50, pp. 23152319, 2002.
188. M. C. Curr as-Francos, P. J. Tasker, M. Fern andez-Barciela, Y. Campos-Roca, andE. S anchez,
Directextractionof nonlinear FET QV functionsfromtimedomainlargesignal measure-
ments, IEEE Microw. GuidedWaveLett., vol. 10, pp. 531533, 2000.
189. J. Verspecht, and D. E. Root, Polyharmonic distortion modeling, IEEE Microw. Mag.,
vol. 7, pp. 4457, 2006.
190. H. Qi, J. Benedikt, and P. J. Tasker, Novel nonlinear model for rapid waveform-based
extraction enabling accurate high power PA design, IEEE MTT-S Int. Symp. Dig.),
pp. 20192022, 2007.
191. G. Simpson, J. Horn, D. Gunyan, and D. E. Root, Load-pull NVNA = enhanced X-
parameters for PA designs with high mismatch and technology-independent large-signal
devicemodels, ARFTGConferenceDigest, 2008.
192. C. Tsironis, A. J urenas, andW. Liu, Highlyaccurateharmonictunersfor loadpull testing,
AsiaPacicMicrowaveConference, 2001, pp. 13111314.
193. Z. Aboush, C. J ones, G. Knight, A. Sheikh, H. Lee, J. Lees, J. Benedikt, andP. J. Tasker,
Highpower activeharmonic load-pull systemfor characterizationof highpower 100watt
transistors, EuropeanMicrowaveConference, 2005.
194. S. C. Cripps, A theoryfor thepredictionof GaAsload-pull power contours, IEEE MTT-S
Int. Symp. Dig., pp. 221223, 1983.
195. M. S. Gupta, Power gaininfeedback ampliers, aclassicrevisited, IEEE Trans. Microw.
TheoryTech. vol. 40, pp. 864879, 1992.
3 Wide band gap transistors SiC and
GaN physics, design and models
R. J. Trew
ECE Department, North Carolina State University
3.1 Introduction
Althoughsolid-statetransistors havereplacedvacuumelectronics inthevast majority
of microwaveelectronic systemsover thepast 40years therevolutionisnot complete.
In particular, the areas of high RF power for microwave and millimeter-wave radar
andcommunicationstransmitter applications, theabilitytoproduceadequateRF power
levels at frequencies greater than 100 GHz, and the ability of devices to operate at
high temperatures greater than about 250

C remain dominated by microwavetubes.


Further solid-statematerial and transistor developments in theseareas areamong the
last frontiers for semiconductor electronics. In theseareas solid statetransistors have
not beenabletocompetewithvacuumtubedevices, andmostsystemsthatmustdeliver
kWtoMWpower levelsaredesignedusingvarioustypesof microwavetube.
Thecurrentstate-of-the-artformicrowavesolid-statedevicesandformicrowavetubes
is showninFigure3.1. As indicated, solid-statedevices produceRF power levels less
thanabout 100WandoperatewithreasonableRF output power tofrequenciesof about
100GHz. TheRF performancestatusshowninFigure3.1isfor singledeviceoperation,
anddoesnotnecessarilyrepresentatruecomparisonof theRFoutputpowercapabilityof
asystem. Powercombiningandphasedarraytechnologypermittheoutputsof manysolid
statetransistors to becombined, thereby producing signicantly improved RF output
power and solid statesystems can, in practice, competein terms of RF output power
with tube-based systems in some cases. Combining technology can raise microwave
RF output power into thekW range, at least through S band and into Ku band [14],
andtheoreticallytomuchhigher power levels. However, suchmultideviceconceptsare
increasingly difcult to apply as operating frequency increases and cannot extend the
upperfrequencylimitbeyondthepresentstate-of-the-art. Operationatfrequenciesabove
Xbandandupto100GHz withRF output power inthehundredsof wattsor kWrange
will requirenewsemiconductor materialsand/or transistor concepts.
Theupper frequencycapabilityof asolid-statetransistor isfundamentallydependent
uponthechargecarrier velocity inthesemiconductor material fromwhichit is fabri-
cated, andthephysical dimensionsof thedevice. Modernsemiconductormaterial growth
technologyandnelinelithographypermittransistorswithcritical dimensionslessthan
amicron(-10
6
m) tobereadilyfabricated, whichpermitstransistorswithhighcutoff
frequencies to berealized. At highelectric elds most common semiconductor mate-
rials demonstrate a saturated charge carrier velocity on the order of :
s
10
7
cm/s,
104 Wide band gap transistors SiC and GaN physics, design and models
Figure 3.1 AverageRF output power versusfrequencyfor varioussemiconductor andvacuum
electronicdevices[5].
or less. Although GaInAs-based HEMTs demonstrate good RF performance up to
approximately 300 GHz, this performance results primarily fromhigh sheet-charge
density and resulting high device current. These two factors permit transistors with
goodRF performanceat highfrequencytobeachieved. However, theRF output power
fromthesedevicesisverylow, andnotsuitablefor mostpractical applications. Standard
semiconductor devices suchas eld-effect transistors andbipolar transistors designed
for high frequency arefundamentally limited in their RF power generation capability
by lowbreakdownvoltage, whichprohibitstheir operationat thevoltagesnecessary to
generatehigh RF power. Also, relatively poor thermal conductivity makes it difcult
to engineer thedevicefor adequatethermal resistance, anddevices designedfor high
RF output power tend to operateat elevated temperature, which limits deviceperfor-
mance. Also, power devices must be designed for high current and this necessitates
devices with largecross-sectional area. This, in turn, produces lowimpedanceinputs
that aredifcult toimpedancematch, especiallyinpower-combinedschemes. All tradi-
tional semiconductor devices(i.e., thosefabricatedfromSi, GaAs, InP, etc.) arelimited
in operating temperature by relatively low barrier energy, and the electronic barriers
becomeincreasinglyleakyastemperatureincreases.
Advances in semiconductor materials engineering, device design, and fabrication
areprovidingsolutions tomany of theselimitations anddevices for high-power, high-
frequency, and high-temperatureapplications arebeing developed. Thesedevices are
expectedtondwideapplicationduetothehighreliability, small size, andpotential low
costofferedbysolid-stateelectronics. Oneverypromisingapproachisthedevelopment
of microwave transistors fabricated fromwide bandgap semiconductors, particulary
SiC, GaN, and heterostructures of theIII-Nitridesystem. Although research in these
3.2 Background 105
semiconductor materialshasbeenpursuedfor manyyears, thetechnologyfor producing
high-quality bulk SiC material for substrates has been demonstrated only in the last
fewyears, andepitaxial SiC material of sufcient quality for devicefabricationisnow
available. Bulk GaN substrates arenot yet available, althoughdevicequality GaN and
AlGaNepitaxial layerscanbegrownonsapphireandSiC substrates. A varietyof elec-
tronicdevicesfabricatedfromthesematerialshavebeendemonstratedandtheresultsare
promisingfor thedevelopment of transistorsthat canbeusedinhigh-power andhigh-
temperaturemicrowavetransmitters. Dueto highelectronvelocity (:
s
210
7
cm/s)
andhighsheet-chargedensity(n
s
10
13
cm
2
) III-Nitrideheterostructuredevicesshow
promisefor producingheterojunctionFETs(HFETs) withimprovedmm-waveRF per-
formance, potentiallyupto300GHzandabove. Atlower frequencies, ontheorder of X
bandandpossiblyashighasK band, SiCandGaN-baseddevicesshouldbecompetitive
withGaAs-basedandInP-basedtransistors for many applications [5], particularly for
improved RF output power capability. However, the higher current capability of the
GaN-basedheterostructuresover SiC-baseddevicesprovidesafundamental advantage
for higher frequencyoperationandimprovedRF output power aboveX or Kuband.
In this chapter, the physical operation, design, and modeling techniques for wide
bandgaptransistorsarepresented. Themicrowaveperformanceandstatusof transistor
development fromthewidebandgap semiconductor materials arepresented. Problem
areas that arepresently limiting deviceperformanceareindicated. It is demonstrated
that microwavepower ampliers fabricatedfrom4H-SiC MESFETs andAlGaN/GaN
HFETsofferexcellentRFpowerperformance, particularlyatelevatedtemperature. The-
oretical transistor models predict roomtemperatureRF output power on theorder of
530W/mmwithPAEapproachingthetheoretical limitsfor classA andBoperationfor
ampliers fabricatedusing4H-SiC MESFETs andAlGaN/GaN HFETs. Experimental
resultsverifythetheoretical predictions. Also, theoretical transistor modelsindicatethat
practical operationatelevatedtemperature, atleastupto500

C, ispossible. TheRFout-
putpower capabilityof devicesfabricatedfromwidebandgapsemiconductorsisalmost
an order of magnitudehigher compared with transistors fabricated fromSi or GaAs-
based materials. The wide bandgap semiconductor devices are nding application in
RF sourcesandpower ampliersfor basestationtransmittersfor cellular telephonesys-
tems, satellitetransmitters, HDTV transmitters, power modulesfor phased-arrayradars,
surveillanceandair-trafccontrol radars, wide-bandampliers, andother applications.
Thetransistors areparticularly attractivesincethey arereadily combinedfor highRF
power applications. They are also attractive for applications that require operation at
elevatedtemperaturesincetheyrequireminimal heat sinking.
3.2 Background
A variety of electronic devices for high-power andhigh-frequency applications canbe
fabricated fromSiC and GaN, and various heterostructures can be based upon these
materials, particularly GaN-basedheterostructures. Devices for power applications [6]
includePIN diodes, SchottkyBarrier Diodes, MOSFETs, BJ Ts, J FETs, thyristors, and
106 Wide band gap transistors SiC and GaN physics, design and models
various ICs fabricatedby combiningthebasic devices into complex structures. High-
frequency devices includeMESFETs, HFETs, SITs, BJ Ts and HBTs, and IMPATTs.
A variety of electronic devices, includingbothhigh-power RF andmicrowavedevices,
canbefabricatedfromnitride-basedsemiconductors. MicrowavedevicesincludeMES-
FETs, static-inductiontransistors(SITs), Heterojunctionbipolartransistors(HBTs), and
HFETs. Devicesfor high-power applicationsincludediodes, MOSFETs, MOSHFETs,
andbipolartransistors. Excellentperformancehasbeendemonstratedfromresearchand
prototypedevices, although many of thesedevices havenot found widespread useor
insertionintocommercial systems. MicrowaveAlGaN/GaNHFETsarendingapplica-
tionfor communicationsbandbasestationampliersandmicrowaveradar transmitters.
Theperformanceof SiC andGaN-basedtransistorsarereviewedinthissection.
3.2.1 SiC transistors
Bipolar transistors(BJ Ts) havebeenfabricatedinSiC. However, duetohighresistance
associatedwiththelowmobilityof p-typematerial, thebaseresistancefornpntransistors
ishigh, whichlimitsthefrequency responsefor thetransistor [7]. It isshownthat 6H-
SiC BJ Ts arelimitedinfrequency responsetoabout S band(i.e., 24GHz), but good
gain and RF output power arepossible, with RF output power on theorder of 50 W
predicted. Power-added efciency falls rapidly above about 1 GHz. The BJ T device
operationisdominatedbyminoritycarrier (electron) transit-timeacrossthep-typebase
region. Theuseof 4H-SiCresultsinimprovedperformance, anditisshownthatelectron
mobilityinthep-typebaseregioncanbeontheorder of 215cm
2
/V-sfor abaseregion
dopedwithAl toaconcentrationof N
B
= 410
18
cm
3
[8]. For thisdevice, thebase-
collectordepletionregionchargingtime,
c
, andtheparasiticchargingtime,
p
, fromthe
capacitancebetweenmetal padsandtheunderlyingcollectorregiondominatetheoverall
electrontransit timeand, therefore, thef
T
of thedevice. Thetransistor demonstrateda
peak f
T
= 4GHz at acollector voltageof V
CE
= 20V, andanemitter current density
of J
E
= 10kA/cm
2
. Theparasiticchargingtimecanbeminimizedbyimproveddevice
design and removal of theparasitic charging timeproduces apeak f
T
= 15 GHz. An
improveddevicedesignby thesameauthors [9] yieldedanf
T
= 7GHz andanf
max
=
5.2 GHz. Fabrication of the transistor on a semi-insulating substrate with resistivity
greater than 10
5
O-cmpermitted the parasitic charging time to be minimized. The
transistor wasbiasedat V
CE
= 20V, andoperatedat anemitter current densityof J
E
=
10.6kA/cm
2
. Thecalculatedmaximumavailablegain(G
max
) was18.6dB at 500MHz
and12.4dB at 1GHz.
A 4H-SiC BJ T with good gain has been reported by Huang and Cooper [10]. This
transistor utilized athermal oxidation procedure, similar to that employed in 4H-SiC
MOSFETs, to passivatethetransistor surface. Previous SiC BJ Ts suffered fromhigh
surfaceleakagecurrentsduetosurfacerecombinationvelocityintherange10
4
10
5
cm/s
[11, 12]. The passivated BJ T had a current gain = 55 and breakdown voltages of
BV
CEO
= 500V andBV
CBO
= 700V. Thecurrent gain remainedabove50withthe
current density above700 A/cm
2
. A record low on-stateresistancefor 4H-SiC BJ Ts
wasreportedby Zhanget al. [13]. The4H-SiC BJ T useda12mthick drift layer and
3.2 Background 107
producedanon-stateresistanceof 2.9mO-cm
2
, withanopen-basecollector-to-emitter
blocking voltage of V
CEO
= 757 V, and a current gain of = 18.8. The transistor
conducted acurrent of 5.4A (J
c
= 859 A/cm
2
) at aforward voltageof V
CE
= 2.5 V.
Thesameauthorspreviouslyreporteda4H-SiCBJ T thatsupportedavoltageof 9.2KV
[14]. high-power SiC BJ Tshavebeenfabricatedat UHF, and215Wof pulsedclassA
power at 450MHz has beenreported[15]. Thetransistor was biasedat V
CE
= 180V
andpulsedwitha0.1%duty cycletoavoidself-heating. Thegainwas7.5dB, andthe
RF power densitywas4.3W/mmwhennormalizedtoemitter nger length.
SITslookverypromisingfor RF power applications[16]. A SIT isbasicallyavertical
FET. Thedeviceoperates under space-charge-limited(SCL) electrontransport condi-
tions, and is very similar to a vacuumtriode in operation. The device demonstrates
lowcurrent gain(f
T
), but excellent voltagegainandhighpower gainis possible. The
lowgainprevents theSIT fromproducinggoodperformancemuchaboveC band, but
excellent UHF andS-Banddevices andampliers havebeenfabricated. Both6H and
4H-SiC havebeenused. Therst SITswerefabricatedfrom6H-SiC andadevicewith
11cmperiphery producedabout 38W at 175MHz with60%PAE [16]. The6H-SiC
SITs producedvery lowcurrent, andthelowcurrent was foundto beassociatedwith
anisotropic electrontransport behavior. Currentstravelinginadirectionparallel tothe
c-axisof a6H wafer wereabout vetimeslower thancurrentsnormal tothec-axisfor
thesamevoltage. Thisresultedinworkshiftingtotheuseof 4H-SiC, anda4H-SiCSIT
with 38 W RF output power, 9.5 dB of gain, and 45%drain efciency at 3 GHz was
developed [17, 18]. Thedevicewas operated under pulsebias and is useful for radar
applications. Further progressincludesa800WUHF SIT anda900WL-bandSIT [19].
A two-stageamplier with 1 kW RF output power using thesedevices was reported
for theHDTV market [20] and other applications. Themost recent result makes use
of ion-implantation to produceaunit cell devicethat generates 107 W output power
with 8.7 dB gain and 59% PAE under CW operation at 750 MHz [21]. The device
was biased at V
ds
= 81.8 V and I
ds
= 1.87 A. This unit cell deviceis being used in
atencell structureto producea10kW RF solidstatedriver amplier for commercial
applications.
The MESFET is a majority carrier device that can be fabricated using n-type SiC
material so that only electrons are involved in current transport [22]. The MESFET
is very attractive for fabrication of high-performance devices for use at microwave
frequencies. Therst SiC MESFETswerefabricatedfrom6H-SiC andMESFETswith
current gain-bandwidth products of f
T
= 25 GHz were reported [23]. These devices
produced3.5W(1.75W/mm) RF power with45.5%PAE at 6GHz. AlthoughtheRF
output power fromthe6H-SiCdevicesisabout threetimesthatgenerallyobtainedfrom
GaAs MESFETs, 4H-SiC has aloweldmobility about twicethat of the6H-SiC and
mostdevicedevelopment hasfocuseduponthismaterial. Early4H-SiCMESFETswith
RF output power on the order of 2.8 W/mmat 1.8 GHz [24], and 2.27 W/mmwith
65.7%for aclassB amplier werereportedat 850MHz [25]. Inlater work, a4H-SiC
MESFET withanf
max
of 42GHzwasreported[26], indicatingthatthesedevicesshould
be capable of producing excellent RF performance through X-band, and potentially
toK-band. This devicehadagatelengthof L
g
= 0.5mandproduced5.1dB gainat
108 Wide band gap transistors SiC and GaN physics, design and models
20GHz.high-powerampliershavebeenproducedandaSiCUHFtelevisionmodulehas
demonstratedgoodsignal delity at the2000WPEP level. S-band4H-SiC transistors
with over 200 W peak power have been produced for radar applications and X-band
power of over 6 W has been obtained [27]. A 4H-SiC MESFET with 42 mmof gate
periphery onasingledieproduced53W of RF power with37%PAE at 3GHz [28].
X-bandSiC MESFETshaveproduced2.5W/mmof RF power and41%PAE at 8GHz,
and 30 W RF power froma 12 mmgate device at 9.7 GHz [29], and excellent RF
power densityof 5.2W/mmat3.5GHzand4.5W/mmat10GHzwereobtained. Asthe
material quality improvedRF output power has improvedanda4H-SiC MESFET has
produced56W with53%PAE at L-band[30]. Thedevicehadaf
T
= 12GHz andan
f
max
= 17 GHz, and demonstrated minimal current drift up to 1100 h. A 4H-SiC
MESFET hasproduced20Woutput power with60%PAE inS-band[31].
A major limitationtoRF outputpower isbreakdownof thegateelectrodeonthedrain
side. Theelectriceldat thegateedgecanachieveaveryhighmagnitude, particularly
when thedeviceis biased to high drain voltageand operated with largeRF terminal
voltages consistent with high RF power drive. It has been shown that the gate can
leak current, and may demonstrate breakdown. The use of eld-plates [32] has been
demonstratedtoresult insignicantlyreducedelectriceldmagnitudeat thegateedge,
therebyreducinggateleakage. Theeld-platecanbeconnectedelectricallytothegate,
the source, or left oating. The use of the eld-plate permits higher voltages to be
applied, withimprovedRF output power. A eld-plate4H-SiC MESFET withaburied-
gatedesignproducedveryhighRF outputpower withapower densityof 7.8W/mmand
70%PAEat3GHz[33]. ThedevicewasoperatedinclassA/B. Theeld-platepermitted
adrainbias of V
ds
= 65V to beapplied. A two-stagebroadbandintegratedamplier
circuit that produced5W over 10MHz to 2.4GHz was reported[34]. Theintegrated
amplier produced 22dB gain, 37dBmoutput power and 28%PAE. Thethird-order
intercept was47dBm.
A novel FET which used a -SiC nanowire as the conducting channel has been
reported[35]. Thedeviceconsistedof aSiO
2
layer grownontopof aSi substrate. The
-SiCnanowirewaslocatedonthesurfaceof theSiO
2
betweentwometal contacts. The
Si substratewasusedasthegateelectrode, andapplicationof avaryingvoltagepermitted
control of thecurrent owingintheSiC nanowire. Nanowirediameters varyingfrom
1025 nmwith a length of 10 mmwere used. The device demonstrated good high-
temperatureperformanceandit wasconcludedthat thetransistor couldndapplication
asahigh-temperaturegassensor.
3.2.2 AlGaN/GaN transistors
The AlGaN/GaN HFET demonstrates excellent RF performance. High sheet-charge
density resulting fromhigh Al incorporation in the AlGaN layer permits high chan-
nel current to beobtained[36]. Initial HFETs werefabricatedonsapphiresubstrates,
but recent work has focused upon theuseof semi-insulating or p-typeSiC substrates
[3739]. Excellent RF performancehasbeenachievedat S-bandthroughKa-bandfre-
quencies, with the greatest RF power density obtained at S-band and up to X-band.
3.2 Background 109
Most AlGaN/GaN HFETs arefabricatedwithunintentionally dopedAlGaN andGaN
epitaxial layers. However, it isalsopossibletofabricateAlGaN/GaNHFETswithgood
RF performanceusingdopedchannel designs [40], and1.73W/mmRF output power
withgoodgainwasobtainedat8.4GHz. Thesmall signal performanceof thesedevices
demonstratedgainbandwidthproductsof f
T
=39GHzandf
max
=45GHz. Small-signal
performance with intrinsic current gain-bandwidth products up to f
T
= 106 GHz for
adevicewith agatelength of Lg = 0.15 mhas been obtained [41]. Thesedevices
producedabout4W/mmRF power and41%PAEat4GHz. VeryhighRF power density
has alsobeenobtainedand9.8W/mmRF power density with47%PAE at 8GHz has
been reported [42]. Thedevices had gatewidths of W= 2 mmand thedevices were
ip-chipmountedtoAlN substrates for improvedthermal conductance. Other devices
fabricatedusingSiC substrates producedRF power as highas 10.7W/mmat 10GHz
with40%PAE[43], withfurtherimprovementsyieldingslightlyover11W/mm. Devices
fabricatedusingAlNinterfacial layersbetweentheAlGaNandGaNproducedRFoutput
power of 8.4W/mmwithaPAE of 28%at 8GHz [44]. Theintroductionof eld-plate
technology suppresses theelectric eld at thegateedgeand permits larger drain bias
to beappliedresultinginhigher RF output power. A highRF output power density of
>30 W/mmwas reported for a eld-plate device biased at a drain voltage of 120 V
[45]. High PAE has also been reported, and an AlGaN/GaN HFET grown by MBE
ona4H-SiC substrateproduced 8.4W/mmwith67%PAE withadrainbias of 30V
[46]. Silicon has emerged as aviablesubstratematerial for AlGaN/GaN HFETs and
excellent RF performancehas been obtained. J ohnson et al. [47] reported RF output
power of 12W/mmwith52.7%PAE and15.3dB gainfor a0.7mgatelengthdevice.
The HFET was biased at 50 V and operated at 2.14 GHz. The transistor is intended
for communications band applications. Dumkaet al. report 7 W/mmwith 38%PAE
and 9.1 dB gain at 10 GHz froma AlGaN/GaN HFET fabricated on a Si(111) sub-
strate[48]. Thedevicewas biased at adrain voltageof 40 V. Reduction of thedrain
bias to 20 V resulted in a decrease in RF power to 3.9 W/mm, but an improvement
of thePAE to 52%. high-frequency Ka-band performancehas also been reported. At
18GHz Ducatteauet al. report anRF power density of 5.1W/mmwith20%PAE and
9.1 dB gain fromanitrideHFET fabricated on aSi substrate[49]. Thedevicehad a
0.25 mgatelength and acurrent gain bandwidth of f
T
= 50 GHz. An AlGaN/GaN
HFET fabricated on a SiC produced 5 W/mmwith 30.1%PAE and 5.24 dB gain at
26 GHz [50]. Leeet al. [51] report 4.13 W/mmwith 23%PAE and 7.54 dB gain at
35GHz. TheHFET wasbiasedwithadrainvoltageof 30V. TheHFET wasfabricated
onaSiC substrate. At 40GHz anRF power density of 2.8W/mm, 10%PAE, and5.1
dB gainwasobtainedfromadevicewitha0.18mgatelengthdevice[52]. Theperfor-
manceof thedevicewassensitivetofrequency, andRFoutputpowerdensityincreasedto
3.4W/mmbyreductionof theoperatingfrequencyto38GHz. Usingarecessgatedesign
anRF outputpower densityof 5.7W/mmwith45%PAE wasobtainedwithadrainbias
of 20 V [53]. Increasing the drain bias to 28 V resulted in an increase in RF output
power density to 6.9 W/mm. Palacios et al. [54] report excellent RF performance at
40GHz fromanAlGaN/GaN HFET fabricatedona4H-SiC(0001) substrate. Devices
with similar structures were fabricated using both OMCVD and MBE. The device
110 Wide band gap transistors SiC and GaN physics, design and models
produced8.6W/mmwith29%PAE andgainof about5dB. TheOMCVDgrowndevice
hadimprovedperformance, with10.5W/mm, 33%PAE, andabout6dBgain. Attempts
to improvedeviceperformanceincludenovel surfacepassivation and chargeconne-
ment. Lauandher colleagues[55] introducedasurfacepassivationtechniqueinvolving
auoride-basedplasmatreatment. Theuoride-basedplasmatreatment, alongwitha
post-gaterapidthermal annealingstep, wasfoundtoeffectively incorporatenegatively
chargeduorineionsintotheAlGaN barrier andpositively shift thethresholdvoltage.
Thetechniquewasusedtofabricateanenhancement-mode(E-mode), HFET. Shenetal.
[56] usedtheuorineplasmaprocess, alongwithadeeplyrecessedgateHFET design,
tofabricateadevicethatproduced17.8W/mmwith50%PAE and15dBgainat4GHz.
Thepassivationprocess limitedgateleakageandthereby permittedadrainvoltageof
V
ds
= 80V tobeapplied, without theuseof aeld-plate.
Thestrongpolarizationeffectsof theAlGaN/GaNstructuremaybeasourceof some
of thereliability problemsexperiencedwiththesedevices. Attemptstoinvestigatethis
includeutilizationof alternatebarriermaterialsthatarelesspolar. Onesuchstructurecan
befabricatedusingInAlN, rather thanthecommonlyemployedAlGaN. AnInAlN/GaN
HFET with a gate length of 0.7 m produced a gain-bandwidth product of f
T
=
13GHz andandf
max
= 11GHz. The2DEG was very high, withn
ss
= 410
13
cm
2
andanelectronmobility of
n
= 750cm
2
/V-s [57, 58]. AnInGaN layer was usedas
aback-barrier to improveconnement of the2DEG electrons [59]. Theconnement
improvedtheoutput resistance, andadevicewithagatelengthof 100nmproduceda
gain-bandwidthproduct of f
T
= 153GHzandanf
max
= 198GHz. Byadjustingthebias
thesamedeviceproducedanf
max
= 230GHz. A doubleheterojunctiondevicedesign
usinganInGaN notchfabricatedonasapphiresubstrateproducedRF output power of
3.4W/mmand41%PAE at 2GHz [60]. GoodRF performancehasalsobeenobtained
fromaGaN FET, fabricatedusinganovel surfacepassivationconsistingof athinAlN
layer locatedbetweentheGaNchannel andaSiNsurfacepassivation[61]. Theresulting
structureisbasicallyametal-insulator-semiconductor (MIS) FET. A devicewithagate
lengthof 60nmproducedf
T
= 107GHz andf
max
= 171GHz.
Most of theearly resultswerefor deviceswithvery narrowgatewidthstominimize
deviceheatingandthermal effects. Morerecentworkhasfocuseduponproducinghigh-
power devices andampliers suitablefor useinapplications suchas communications
basestationtransmitters. Andoetal. [62] reportedRF outputpower of 10.3Wwith47%
PAE and18dB linear gainat 2GHz. Thisresult wasfor adevicewithagatewidthof
1mm. Linear gate-widthscalingwithdraincurrent andRF output power hasalsobeen
demonstrated. A high-power integrated circuit using 8 mmof gateperiphery yielded
51W RF output power at 6GHz under pulsebias conditions [63]. A communications
bandamplier, usingAlGaN/GaN HFETs fabricatedonSiC substrates, andbiasedat
48V producedCW RF output power of 100W at 2.14GHz [64]. A C-bandamplier
usinga0.4mgatelengthand50.4mmgatewidthAlGaN/GaN HFET fabricatedon
aSiC substrateproducedRF output power of 140Wwith25%PAE. Theamplier was
operatedwithapulsebiasof 40V [65]. A pushpull transmitteramplierfor3Gwireless
basestationapplicationswasconstructedusingAlGaN/GaN HFETsfabricatedonSiC
substrates [66, 67]. At adrain bias of 50 V theamplier produced 250 W RF output
3.3 Material parameters 111
power and, usingdigital predistortionlinearization, anadjacent channel leakagepower
ratio (ACLR) of less than 50 dBc for four-carrier W-CDMA signals was obtained.
Very high RF output power was obtained fromwidegatewidth AlGaN/GaN HFETs
fabricatedonSi(111) substrates[67]. Theindividual HFETshadagatewidthof 36mm
and when operated under CDMA modulation produced 20 W RF power with adrain
efciencyof 27%whenbiasedat aV
ds
= 28V. Theamplier wasfabricatedusingtwo
of thedevices and produced amaximumRF output power of 156 W with 65%drain
efciencyat2.14GHzandnomodulation.Thesameauthorsreportfurtherimprovements
by employing asource-grounded eld plateon theHFET, and when biased at V
ds
=
60V andunder pulsedRF conditions, asaturatedRF output power of 368Wwith70%
PAE wasobtained. Widegatewidthdevicesrequireeffectivemeansfor grounding, and
alaser-assistedprocessing procedurefor fabricating viaholes was reported [68]. The
processpermitswidegatewidthdevicestobeeffectively grounded, anda20mmgate
widthdevicebiasedat V
ds
= 26V produced41.6Wwith55%PAE at 2GHz.
AmpliersfabricatedusingAlGaN/GaNHFETshaveproducedover 400Wof pulsed
power with 600 MHz of bandwidth (2.9to 3.5 GHz) and 50%PAE [69], atwo-stage
amplier has produced 58 W with 38% PAE and 15 dB gain at X-band [70], and a
MMIC amplier has produced 500 mW RF power with 17%PAE and 12 dB gain in
E-Band (7195 GHz) [71]. The nitride devices are being aggressively developed for
applicationinampliersfor S-bandcommunicationsbandbasestationtransmitters, and
asampliersfor X, Ka, andW-bandradar transmitters.
3.3 Material parameters
TheDCandRFperformancecapabilityof electronicdevicesisfundamentallydependent
upontheelectronic, thermal, andmechanical propertiesof thematerialsfromwhichthe
devicesarefabricated.Of particularimportancearethechargetransportcharacteristicsas
afunctionof electriceldfor thematerial. Eachsemiconductor hasadifferentvelocity-
eldcharacteristic, andsemiconductorsof mostinterestfor devicefabricationwill have
high carrier velocity capability. The quality of semiconductor epitaxial material has
continuallyimproved, andtheDCandRFperformanceof semiconductortransistorshave
experiencedsignicantperformanceimprovementsasaresult. A varietyof technologies
forgrowthof semiconductorepitaxial layers, suchasmolecularbeamepitaxy(MBE)and
organo-mellaticchemical vapor deposition(OM-CVD), havebeendevelopedandthese
technologies permit the growth of epitaxial layers of precise thickness and impurity
doping concentration. It is now possible to fabricate solid state devices with layer
thicknessof onlyafewangstromswithprecisedenedimpurityconcentrations, andthis
level of control permitsdeviceswithfrequencyperformancewell over 100300GHzto
befabricated.
The advantages of device fabrication fromwide bandgap semiconductors can be
seenfromacomparisonof fundamental electronictransport andmaterial parameters. A
summaryof thesemiconductor material propertiesmost important toelectronicdevice
performanceislistedinTable3.1for several semiconductors.
112 Wide band gap transistors SiC and GaN physics, design and models
Table 3.1 Material properties for several semiconductors
Material E
g
(eV)
r
(W/

K-cm) E
c
(V/cm)
Si 1.12 11.9 1.5 310
5
GaAs 1.43 12.5 0.54 410
5
InP 1.34 12.4 0.67 4.410
5
3C-SiC 2.3 9.7 4 1.810
6
4H-SiC 3.2 10.0 4 3.510
6
6H-SiC 2.86 10.0 4 3.810
6
GaN 3.4 9.5 1.3 210
6
Diamond 5.6 5.5 2030 510
6
For transistors the most important material properties for fabrication of high-
performance microwave structures include a large energy gap, E
g
(eV), a low value
of dielectric constant,
r
, high thermal conductivity, (W/

K cm), and high critical


electric eldfor breakdownE
c
(V/cm). Wideenergy bandgapgenerally translatesinto
anabilitytosupporthighinternal electriceldsbeforeelectronicbreakdownoccurs, and
alsoprovidesfor improvedradiationresistance. Most transistor fabricationhasbeenin
Si, GaAs, andInP andrelatedcompoundsandthevast majorityof all devicescommer-
cially availablearefabricatedfromthesematerials. TheSiC andGaN-basedmaterials
have energy bandgaps about two to threetimes larger than thosein theconventional
semiconductors, suchas Si, GaAs, andInP. Thedielectric constant is anindicationof
thecapacitiveloading of adeviceand affects theterminal impedance. Generally, for
solidstatedevicesalowvaluefor thesemiconductor dielectricconstant isdesired, and
this permits asolid statedeviceto belarger in areafor aspecied impedancevalue.
Increased areapermits larger RF currents and higher RF power to begenerated. The
widebandgapsemiconductorshavedielectricconstantsabout 20%lower thanthecon-
ventional materials. This, inturn, permits awidebandgapsemiconductor devicetobe
about 20%larger inareacomparedtoacomparabledevicefabricatedfromSi or GaAs
for aspecicimpedancemagnitude, andincreasedareapermitslarger RF currentsand
higher RF power tobegenerated.
Thethermal conductanceof thematerial isextremelyimportant sincethisparameter
indicates theeasewithwhichdissipatedpower canbeextractedfromthedevice. Poor
thermal conductivity resultsindeviceoperationat elevatedtemperaturewithdegraded
performance. Conventional semiconductors are, in general, poor thermal conductors,
particularly theGaAs andInP materials. Conversely, SiC is anexcellent thermal con-
ductor andGaN is about thesameas Si, thebest of theconventional semiconductors.
Diamondhasthehighest thermal conductivityof anyknownmaterial andisoftenused
tofabricateheatsinksfor semiconductor devicesthatmustoperateinhigh-power appli-
cations. Finally, thecritical electriceldfor electronicbreakdownshouldbehigh. This
parameter is an indication of thestrength of theelectric elds that can besupported
internallytothedevicebeforebreakdown. Highelectriceldspermit largeterminal RF
voltagestobesupported, andthisisnecessaryfor thegenerationof highRF power. The
critical elds for thewidebandgapmaterials areexcellent andvery high, typically an
3.3 Material parameters 113
Electric Field (kV/cm)
10 1
0.1
1
10
GaAs
AIGaN/GaN
4H-SiC
6H-SiC
Si
GaN V
e
l
o
c
i
t
y

(
x
1
0
7

c
m
/
s
)
100 1000
Figure 3.2 Electronvelocityversuselectriceldtransport characteristicsfor variousn-type
semiconductors(N
d
=10
17
cm
3
).
orderof magnitudegreaterthanfortheconventional semiconductors. Ingeneral thewide
bandgapsemiconductorshavemoreoptimumvaluesfor all theseparameterscompared
toconventional semiconductors.
Basically, acurrentisdenedasthemovementof chargeandexpressedastheproduct
betweenthechargedensity andtransport velocity. Therefore, theDC andRF currents
thatowthroughadevicearedirectlydependentuponthechargecarrier velocityversus
electriceldtransportcharacteristicsof thesemiconductor material. Generally, for high
currents andhighfrequency, highchargecarrier mobility andhighsaturationvelocity
aredesirable. A comparisonof theelectronvelocity-electric eld(:-E) characteristics
for several semiconductors isshowninFigure3.2. The:-E characteristic isdescribed
in terms of charge carrier mobility
n
, (units of cm
2
/V s) dened fromthe slope of
the:-E characteristicat lowelectriceld, andthesaturatedvelocity:
s
(unitsof cm/s),
dened when the carrier velocity becomes a constant, eld-independent magnitude,
generallyat highelectriceld. Thehighvaluefor electronmobilityof GaAs(typically,

n
5000cm
2
/V s) isthemainreasonthatFETsfabricatedfromthismaterial havesuch
excellenthighfrequencyperformance. A primarydisadvantageof fabricatingtransistors
fromSiCandGaNistherelativelylowvaluesforthechargecarriermobilities(typically,

n
200500cm
2
/V s). Ingeneral, thewidebandgapsemiconductorshaverelativelylow
mobility, butveryhighsaturationvelocity(typically, :
s
1210
7
cm/s). However, the
mobility of SiC andGaN isadequatefor transistorsdesignedfor highpower operation
[72] duetothelargeRF terminal voltagesthesetransistorscansustain. Thelowmobility
producesarelativelyhighkneevoltage(i.e., thetransitionvoltagebetweenthelinearand
saturationregions onthetransistor IV curve), but theability of thedevicetoproduce
goodRF output power andPAE inamplier circuits is not seriously compromisedby
therelativelyhighkneevoltageduetothelargeRF terminal voltages, whichareonthe
order of 1020 times themagnitudeof thekneevoltage. In practice, near ideal PAE
isobtainedfor ampliersfabricatedfromwidebandgapsemiconductor transistors, and
114 Wide band gap transistors SiC and GaN physics, design and models
10
mobility
Sheet Charge Density
100
Temperature (K)
1000
10000
2
D
E
G

M
o
b
i
l
i
t
y

(
c
m

2
/
V
-
s
)
1000
S
h
e
e
t

C
h
a
r
g
e

D
e
n
s
i
t
y

(
c
m

2
)
10
12
10
13
10
14
10
15
10
16
Figure 3.3 Electronmobilityandsheet chargedensityversustemperaturefor a2Delectrongas
fromShubnikovDeHaasmeasurements[5].
theAlGaN/GaN HFET ampliers obtainnear-ideal PAE upto X-band, andpotentially
higher.
For atypical devicedopingdensity of N
d
210
17
cm
3
, theelectronmobility for
6H- and4H-SiCareabout250cm
2
/V sand500cm
2
/V s, respectively. Thefactor of two
increaseinmobility for 4H-SiC comparedto 6H-SiC is oneof themajor reasons that
the4Hpolytypeispreferredfor deviceapplications. Theelectronsaturationvelocityin
both6H- and4H-SiC is :
s
210
7
cm/s, whichis afactor of two higher thanfor Si
(:
s
110
7
cm/s) andafactor of four higher thanfor GaAs(:
s
(0.50.6)10
7
cm/s).
Themobilityandsaturationvelocityfor the2DEGfor theAlGaN/GaN heterointerface
is very suitablefor deviceapplications. Theroomtemperaturemobility of the2DEG
is in the range of 10001500 cm
2
/V s, which is signicantly better than for SiC or
bulkGaN. Thesheet-chargedensityfor thisstructurecanbeveryhighandgreater than
n
ss
10
13
cm
2
duetopiezoelectric andspontaneous polarizationinducedeffects. The
measuredsheet-chargedensityisaboutafactorof vebetterthanisobtainedforthemore
commonly employedAlGaAs/GaAs heterostructure. Thecharacteristics for the2DEG
areshownintheShubnikov-DeHaasandHall mobilitymeasurementsinFigure3.3[5]
for an AlGaN/GaN heterostructure grown on a sapphire substrate. The measurement
over temperature indicates that the 2DEG mobility is very sensitive to temperature,
demonstrating about T
2.3
dependence. This indicates that devices fabricated from
thistypeof structurewill betemperaturesensitiveandperformancewill degraderapidly
withelevatedtemperature.
Themagnitudeof electriceldthat producessaturatedchargecarrier velocityisalso
importantsincethedevicemustbeabletodevelopthesaturationeldtoobtainmaximum
RF performanceandhigh-frequencyoperation. Thesaturationeldsfor 4H-and6H-SiC
areabout E
s
60kV/cmandE
s
200kV/cm, respectively, whicharehighrelativetothe
comparablevaluesof E
s
3kV/cmandE
s
35kV/cmfor GaAsandSi. Thesaturation
eldfor theAlGaN/GaN heterostructure2DEGislessthanfor either 6H- or 4H-SiC.
3.4 Transistor amplier operating principles 115
Network
V
s
Z
s
= R
s
+ jX
S
Z
L
= R
L
+ jX
L
I
L
V
L
P
in
P
out
Z
in Z
out
Figure 3.4 General two-port networkusedfor amplier analysis.
Holemobilities in SiC and GaN-based materials arevery low, and on theorder of
1050cm
2
/V-s, andit is very difcult to observesaturationeffects for holetransport.
Extremelylowmobilityrequiresveryhighsaturationelds, whichapproachthecritical
eldfor avalanchebreakdown. Lowmobility alsoresultsinhighvaluesfor resistance,
whichlimitsdeviceperformance. Thelowholemobility presentsseriousproblemsfor
use of p-type wide bandgap material in devices. For this reason, most devices under
development aremajoritycarrier devices, suchasFETsandstatic-inductiontransistors
that canbefabricatedusingonlyn-typesemiconductor material.
3.4 Transistor amplier operating principles
Thebasic congurationfor anamplier isshowninFigure3.4[73]. Theamplier isa
two-port network that consistsof asourcethat feedstheinput withaloadconnectedto
theoutput. Thenetworkhasgainandtherebyampliesasignal passingthroughit from
thesourcetotheload. RFpowercanonlybegeneratedfromareal source(i.e., resistance)
anddeliveredthroughanetworktoareal load(i.e., resistance). Sinceelectronicdevices
andnetworks, as well as most microwavesources andloads, also includereactanceit
is necessary to employ reactivetuningto obtainoptimumpower transfer. Conjugately
tunedoutput andloadimpedancesdeliver maximumRF output power fromthesource
totheload.
Thepower deliveredtotheload, P
L
, fromthenetworkcanbewrittenas
P
L
=
1
2
Re
_
V
L
I

=
1
2
[I
L
[
2
R
L
(3.1)
whereV
L
andI
L
arethevoltageandcurrent at theloadimpedance, andR
L
is thereal
part of theloadresistance. Thepower deliveredtotheloadcanbewrittenasafunction
of thereectioncoefcient at theload,
P
L
= P
out
_
1[I
L
[
2

(3.2)
whereP
out
istheRF poweravailablefromthenetworkandI
L
isthereectioncoefcient
at theload. MaximumRF power transfer occursfor noreectionfromtheload,
I
L
= 0 (3.3)
116 Wide band gap transistors SiC and GaN physics, design and models
Drain Voltage (V)
Linear
D
r
a
i
n

C
u
r
r
e
n
t

(
A
)
0
0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
1 dB Compression
3 dB Compression
5 10 15 20 25
Figure 3.5 Dynamiccurrent-voltageloadlinessuperimposedupontheDC IVcharacteristicsfor
aGaAsMESFET amplier (thethreedynamicloadlinesindicateoperationfor linear, 1dB,
and 3dB compressionconditions) [5].
This conditionoccurs whentheloadimpedanceis set tothecomplex conjugateof the
networkoutput impedance
Z
L
= Z

out
(3.4)
Theamplier PAE is
P AE =
P
L
P
in
P
dc
x100%= P
in
(G 1)
P
dc
x100% (3.5)
where P
in
is the RF power into the network, P
DC
is the DC power dissipated in the
network, andGisthenetworkgain, expressedas
G =
P
L
P
in
(3.6)
Thedynamiccharacteristicsof anamplier usingaGaAsMESFET astheactivedevice
areillustratedinFigure3.5, whichshowsdynamicloadlines(i.e., IVcharacteristics) for
threeconditions: linear operation; 1dB incompression; and 3dB incompression.
The dynamic load lines are superimposed upon the DC IV characteristics for the
activedevice. For thesituationshowninFigure3.5theGaAs transistor is biasedwith
adrain-sourcevoltageof V
ds
= 8 V, and thenetwork is tuned for maximumPAE for
each dynamic load line. SinceRF power can only begenerated by areal sourceand
deliveredtoareal load, thedynamicloadlinewouldbeastraightlineoscillatingupand
downtheDC loadlinefor thenetwork. However, sincethedevicehascapacitance, the
dynamicloadlinedemonstrateselliptical behavior. Whilethedeviceisoperatingbelow
saturation the load line is conned within the DC IV characteristics. As the device
is driven into saturation thedynamic load lineshifts and extends outsidetheDC IV
characteristics onboththehigh-current andlow-current portions of theRF cycle. The
averagevalueof theRF current also increases, indicating that thedeviceDC current
increases as the device is driven into saturation. The extension of the dynamic load
line outside the DC IV characteristics is possible due to the complex nature of the
network. Thetotal RF currentconsistsof conductionanddisplacementcomponentsand
3.4 Transistor amplier operating principles 117
althoughtheconductioncurrent islimitedby theIV characteristics, thedisplacement
currentmaintainscurrentcontinuityattheterminals. Thatis, asthedeviceisdriveninto
saturationtheconductioncurrentisclippedbytheIVcharacteristicsfor thedevice, but
thetotal RFcurrentcontinuityismaintainedbydisplacementcurrent.Devicecapacitance
increases as it is driven into saturation and inductive tuning is necessary to obtain
optimumRF performance.
Optimizingtheinductiveexternal impedancetomatchthecapacitiveimpedanceof the
transistor resultsinthereversal of dynamic loadlinedirection, asshowninFigure3.5.
Under optimumtuningconditionsthenetwork isessentially aresonant circuit withthe
reactiveenergy shiftingbetweenthecapacitiveandinductiveelds. As thenetwork is
drivenfurther intosaturationthecurrentclippingbehavior increases, withanetincrease
inbothDC current anddevicecapacitance.
The dynamic behavior of the amplier network denes the factors that determine
theRF performancelimits of thedeviceandthematerials fromwhichit is fabricated.
Thepower deliveredtotheloadisaproductof theRF voltageandRF currentthatcanbe
establishedat theload, andthisisdeterminedbytheactivedevice. Semiconductorsare
limitedinthebiasvoltagethatcanbeappliedbythecritical electriceldfor breakdown
of thesemiconductor material. Therefore, semiconductorsthathavehighcritical electric
elds for breakdownaredesirablefor power deviceapplications. Thecritical eldfor
breakdown is a function of bandgap energy and wide bandgap semiconductors are
desirablefor power applications.
TheIV characteristicsshowninFigure3.5canbeusedtoexplainthebasic classes
of amplier operation. For example, if the transistor is biased at a DC voltage and
DC current locatednear themiddleof theIV characteristic plane, andtheinput and
output impedances are tuned so that the dynamic RF IV characteristic is conned
completelywithintheIVcharacteristicplane, theamplier will operateunder classA
conditions. Thetransistor isalwaysinanon stateandthemaximumPAE is50%. By
changingthegate(orbase) biastoreducetheDCdrain(orcollector) currentthedynamic
IV characteristic will begin to clip on thehigh-voltageportion of theRF cycle. The
waveformclippingwill result inno channel conductioncurrent, but theRF waveform
will bemaintainedbycapacitivecurrent. Thereductioninthedraincurrentwill produce
areductionintheDCpower dissipationwithinthetransistor, andthewaveformclipping
will produceareductionintheRF power deliveredtotheload. However, thereduction
in theDC power dissipation occurs morerapidly than thereduction in theRF output
power, with a result that the PAE increases. For a bias condition where one half of
the RF waveformis clipped, the RF output power will decrease by a factor of two
(3dB). Theideal PAE for this modehas atheoretical valueof 78.5%. Althoughthe
PAE isincreased, thehalf-sinusoidcurrent waveformproducesharmonicsat theoutput,
althoughit islinear inthesensethat anxdB increaseintheinput power resultsinanx
dB increaseinoutput power until thedeviceisdrivensufcientlyhardtocauseclipping
of thetopof thecurrent waveform. Linearity isvery important for ampliersdesigned
for communicationssystems. Usingtwotransistorsinapushpull conguration, where
eachtransistor isintheon statefor one-half of theRF cycle, doublestheoutputpower
and hence extends the linear range while maintaining the high PAE. The penalty is
118 Wide band gap transistors SiC and GaN physics, design and models
therequirement for theadditional transistor, alongwiththenecessary RF circuitry to
accomplishthenal circuit. For operationbetweentheclassA modeandtheideal class
B mode, theRF waveformispartiallyclippedandthePAE will besomewherebetween
50%and78.5%. ThismodeisgenerallytermedclassA-B, andmanypractical transistor
ampliersaredesignedtooperateinthismode. Thisisnecessarybecausereal transistors
haveasoft-turn-oncharacteristicandsoif operatedinapureclassBmodethentheyare
nonlinear at small signalsandshowagainexpansionregionbeforesaturatingandthen
eventuallyenteringthetraditional nonlinear gaincompressionregion. A veryhigh-PAE
modecanbeobtainedbybiasingthetransistorwell belowthevoltagethatpermitschannel
conductioncurrent to ow. Inthis mode, termedclass C, thetransistor only conducts
during the peak of the voltage during the RF cycle. The current waveformbecomes
essentiallyapulse. Theideal theoretical efciencyfor classC is100%, althoughthisis
onlyobtainedwithnopower deliveredtotheload. Practical classCampliers, however,
canbedesignedtooperatewithPAE inthe8090%range.
Electronic devices designedfor microwaveandRF applications operateinatransit-
timemodeandarescaledinsizebyfrequencyconsiderations. Under normal operation
theelectriceldswithinthedevicesvaryfromlowmagnitudenear theelectroninjection
locationtoamagnitudesufcient toproduceelectronvelocity saturationinthecharge
control/modulation region. Therefore, largecurrent capability requires semiconductor
materials that have high electron velocity. In general, both high-mobility and high-
saturationvelocity aredesirablefor highRF current. Traditional semiconductors such
as Si and GaAs haveelectron saturation velocities that arelimited to about :
s
= 1
10
7
cm/s, andthislimitsboththepowerthatcanbegeneratedandthefrequencyresponse
of thedevice. Widebandgapsemiconductorshaveelectronsaturationvelocitiesthatcan
beafactor of twohigher. Thecombinationof high-current andhigh-voltagecapability
makewidebandgapsemiconductors very attractivecandidatematerialsfor fabrication
of high-power andhigh-performanceelectronicdevices.
3.5 Device design and RF performance
The most promising devices for high-power, high-frequency RF applications are the
4H-SiC MESFET and the AlGaN/GaN HFET. Since the 4H-SiC MESFET can be
fabricated entirely fromn-type material the losses associated with use of p-type SiC
canbeavoided. Thedeviceisalsorelativelyeasytofabricateduetoasimplestructure.
Thebasic MESFET structureconsists of ahighly doped n-typeepitaxial layer grown
upon ahighly resistivesubstrate, as shown in Figure3.6. Thedrain-to-sourcecurrent
is thereby conned to the highly doped n-type layer. A control electrode (the gate)
is locatedbetweenthesourceanddrainelectrodes, whicharedesignedto haveohmic
current-voltagecharacteristics. ThegateelectrodeisanonlinearSchottkycontact, which
innormal operationisreversebiased, whichcreatesadepletionregionintheconducting
channel, thereby permitting control of thedrain-to-sourcecurrent. Modulation of the
voltage applied to the gate electrode permits the channel current to be modulated,
andsincealargechannel current canbemodulatedwithasmall gatevoltage, alarge
3.5 Device design and RF performance 119
Figure 3.6 SiC MESFET structure.
n
+
cap n
+
cap
Figure 3.7 AlGaN/GaN HFET structure.
transconductance(i.e., gain) is achieved. Themagnitudeof thetransconductanceand
theupper frequencyof operationof thedevicescalewithreductionsingatelength, and
for thisreasonshort gatelengthsaredesirable. Inpractice, gatelengthsontheorder of
L
g
0.11 mareroutinely realized, and this permits operation with good gain to be
realizedaboveXandKu-bandsfor SiC-basedMESFETs.
The AlGaN/GaN HFET is also readily fabricated and demonstrates excellent RF
performance. Thesedevicesaresimilar totheMESFET, butdiffer inthesemiconductor
layer structure. A HFET structureisshowninFigure3.7. Typically, anundopedlayer of
GaN isgrownuponahighly resistivesubstrate, oftenSiC. A GaN buffer layer isoften
usedtoaccountforthelatticemismatchbetweentheSiCandGaNlayers. A thinupdoped
AlGaNlayer isthengrownupontheundopedGaNlayer. Theenergybanddiscontinuity
betweentheAlGaNandGaNlayerscreatesanenergynotch attheheterointerface, and
thisresultsinthecreationof a2Delectrongas(2DEG), whichestablishesaconducting
path between the drain and source electrodes, which are fabricated in an analogous
manner totheMESFET. A Schottkygatecontactislocatedbetweenthedrainandsource
electrodes, as in the MESFET, and the same scaling rules apply. However, since the
120 Wide band gap transistors SiC and GaN physics, design and models
0
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
10 20 30 40 50
V
gs
= 8 V
V
ds
(V)
I
d
s

(
A
)
V
gs
= 5 V
V
gs
= 2 V
+V
gs
= 1 V
Figure 3.8 DC IVcharacteristicsfor aSiC MESFET (N
d
= 10
17
cm
3
, L
g
= 0.5mm,
W= 1mm) [5].
electrontransport characteristicsaremuchsuperior inthenitrideheterointerface2DEG
compared to the bulk SiC, the AlGaN/GaN HFET is capable of much improved RF
frequency performance. In fact, thenitridedevices arecapableof RF operation with
goodgainwell above100GHz.
Inthenext sectiontheDC andmicrowaveperformanceof thesedevicesisdescribed,
and performance projections are presented. The investigation makes use of theoreti-
cal simulationsandtheresultsarecomparedtoexperimental measurements. Excellent
agreementbetweenthesimulatedandmeasureddataisobtained. Oncethedevicesimula-
toriscalibratedandveriedagainstexperimental data, thesimulatorisusedtodetermine
theperformancefor optimized devicestructures. Theoptimized devicestructures are
tunedinClassA andClassA/Bampliernetworkstoinvestigatepredictedperformance.
3.5.1 4H-SiC MESFET amplier
TheMESFET, sinceit isamajority carrier device, isanideal transistor for fabrication
using wide bandgap semiconductors [7]. The DC IV characteristics for a MESFET
fabricatedfrom4H-SiC andwithgatelengthL
g
= 0.5mandgatewidthW= 1mm
areshowninFigure3.8. Thedevicehasbeenoptimizedfor microwaveperformancein
X-bandandhasauniformchannel impuritydopingdensityof N
d
=510
17
cm
3
anda
channel thicknessof a=0.15m. Theconductingchannel isgrownonahigh-resistivity,
semi-insulatingsubstrate. Thetransistor producesamaximumchannel current of I
dss
=
550mA andamaximumtransconductanceof g
m
= 65mS/mm, whichislowbyGaAs
MESFET standards wherethetransconductances aregenerally intherangeof several
hundredmS/mmforanX-bandtransistor. TheIVcharacteristicsindicateakneevoltage
wherethechannel current saturates, withadrainvoltageof about 9V, whichishighby
GaAsMESFET standardswherethedevicestypicallysaturateat lessthanavolt.
3.5 Device design and RF performance 121
0
1 10
Frequency (GHz)
H
2
1

(
d
B
)
,

G
m
a
x

(
d
B
)
100
5
15
10
20
25
30
35
40
G
max
H
21
Figure 3.9 Current gain(H
21
) andpower gain(G
max
) small-signal RF performanceversus
frequencyfor aSiC MESFET amplier [5].
P
in
(dB m)
P
o

(
d
B

m
)
,

P
A
E

(
%
)
,

G

(
d
B
)
0
10
20
30
PAE
P
o
40
50
60
20
G
30
40
35 25 15 10 0 5
Figure 3.10 Large-signal RF performanceversusRF input power for aSiC MESFET amplier
(Freq= 10GHz, V
ds
= 40V, ClassA operation).
Thesmall-signal current (h
21
) andpower gains (Gmax) for thedeviceareshownin
Figure3.9. Although thetransconductancefor the4H-SiC MESFET is low by GaAs
MESFET standards, the device produces a gain-bandwidth product of f
T
= 24 GHz
andamaximumfrequency of oscillationof f
max
= 56GHz. Thef
max
is highdueto a
high-magnitudeoutput impedance, which permits high-voltagegain to bedeveloped.
Thesmall-signal RF parametersshowninFigure3.9indicatethat thedeviceiscapable
of producing good RF output power through X-band, and potentially higher. This is
demonstratedinFigure3.10, whichshowstheoperationof thetransistor whenoperated
in aClass A amplier circuit. Theamplier is biased at V
ds
= 40V and is tuned for
maximumPAE at 10 GHz. The amplier produces a maximumRF output power of
122 Wide band gap transistors SiC and GaN physics, design and models
Frequency (GHz)
P
A
E

(
%
)
,

P
o

(
d
B

m
)
,

G

(
d
B
)
0
10
20
30
PAE
P
o
40
50
60
20
G
30 35 25 15 10 0 5
Figure 3.11 RF performanceversusfrequencyfor aSiC MESFET amplier (V
ds
= 40V, ClassA
operation).
5 W/mmwith a maximumPAE of 50%, the ideal value for Class A operation. The
linear gainof theamplier is14.8dB. Theseresultsareexcellent andsuperior tothose
obtained fromacomparablegatewidth GaAs MESFET, which can only produceRF
outputpower ontheorder of 11.5W/mm. Therelativelylowelectronmobilityof SiC
andhigh-saturationkneevoltageof thetransistor donotlimittheRF performanceof the
devicebecausethe40V drainbiasthat canbeappliedissufcient for theregionunder
thegatetooperateinvelocitysaturationconditionsandefcient gatemodulationof the
channel current is maintained [72]. The gate breakdown voltage for this transistor is
V
gdB
=100V, therebypermittingthe40V drainbiastobeappliedwithoutencountering
RF breakdownphenomena.
Thesmall-signal RF parametersindicatethat theamplier shouldoperateabovethe
X band. To explore the performance of the amplier as a function of frequency it
is operated over a frequency range extending from3 GHz to 30 GHz and tuned for
maximumPAE. TheresultsareshowninFigure3.11. Asindicatedtheamplierproduces
near ideal class A performance through X-band (12 GHz). At 12 GHz the amplier
produces4WRF power with48%PAE and10dB linear gain. AboveX-bandthegain
and PAE decrease due to increased losses that result fromthe low electron mobility.
The PAE decreases from48% at 12 GHz to 26% at 30 GHz. The high-frequency
gain is signicantly reduced and at 30 GHz is only about 3 dB, which is too lowfor
practical use.
Theseresultsstemfromextensiveinvestigationsandhavebeencalibratedandveried
withexperimental results. Thestudyindicatesthat 4H-SiC MESFET amplierswill be
useful through X-band, but will have limited application at higher frequencies. The
lowmobility of SiC producesrelatively highaccessregionandcontact resistancesthat
severelylimit RF performanceat frequenciesaboveX-band.
3.5 Device design and RF performance 123
Frequency (GHz)
P
o

(
d
B

m
)
,

P
A
E

(
%
)
,

G

(
d
B
)
0
10
20
30
PAE
P
o
40
50
60
20
G
30 35 25 15 10 0 5
Figure 3.12 Large-signal RF performanceversusfrequencyfor anAlGaN/GaN HFET amplier
(V
ds
= 25V, ClassA operation).
3.5.2 AlGaN/GaN HFET amplier
FETsfabricatedusingtheAlGaN/GaN heterostructureoffer thepotential toproducea
class of devices withexcellent DC andRF performance. Thechargedensity andelec-
tron transport characteristics of the 2DEG at the heterointerface between the AlGaN
andGaN layersareexcellent, withvery highsheet-chargedensity ontheorder of 10
13
cm
2
routinelyrealized. Thismagnitudeistypicallyafactor of vehigher thanfor the
AlGaAs/GaAs2DEGusedinGaAs-basedHEMTs. Thesheet-chargedensity ishigher
thanwouldbeexpectedfromstandard2DEGtheoryandthishasbeenshowntobedue
topiezoelectricandspontaneouspolarizationeffects. The2DEGattheAlGaN/GaNhet-
erojunctionhasexcellentchargetransportcharacteristicsandthesaturationvelocityhas
amagnitudeof about 1210
7
v/cmandmobility intherangeof 10001500cm
2
/Vs
at roomtemperature. The combination of high sheet-charge density and high carrier
velocityresultinhighcurrentcapabilityfor thetransistor. Inpractice, highchannel cur-
rent isobtainedfromthesestructures, andAlGaN/GaNHFETswithmaximumchannel
currentsover 1A/mmandapproaching2A/mmareroutinelyobtainedexperimentally.
Thesimulatedmicrowaveperformanceas afunctionof frequency for anoptimized
AlGaN/GaNHFET ClassA amplier tunedfor maximumPAE isshowninFigure3.12
[7]. Thetransistor has agatelength of L
g
= 0.5 m, and awidth of W= 1 mmand
is biased at V
ds
= 25 V and V
gs
= 4 V. This gate bias would be expected from
the DC IV characteristics to place the amplier in class A operation. However, due
to recticationeffects under overdrivenlarge-signal operationthetransistor bias point
shiftsasafunctionof frequency andtheamplier shiftsbetweenclassA andclassAB
operationfrom3GHz to18GHz. Thisisevidencedby thePAE, whichwasover 50%
from3GHz to 25GHz. ThePAE peaked at about 58%from12 GHz to 18GHz. At
frequenciesabove18GHz thePAE decreasestoabout 44%at 30GHz duetoincreased
124 Wide band gap transistors SiC and GaN physics, design and models
Frequency (GHz)
P
o

(
d
B

m
)
,

P
A
E

(
%
)
,

G

(
d
B
)
0
10
20
30
PAE
P
o
40
50
60
20
G
30 35 25 15 10 0 5
Figure 3.13 Large-signal RF performanceversusfrequencyfor anAlGaN/GaN HFET amplier
optimizedfor highRF output power (V
ds
= 25V, ClassA operation) [5].
losses. The amplier produces RF output power of about 35 W over the frequency
rangeof 3GHz to30GHz, whichisexcellent for aHFET witha1mmgatewidth. The
linear gainisabove10dB from3GHz to18GHz, andisstill at 9dB at 30GHz.
FETscanbedesignedtomaximizeRF outputpower, gain, orPAE, butitis, ingeneral,
not possible to obtain optimumperformance for all three parameters simultaneously
[74]. For thetransistor described here, themodication of thestructurefor increased
channel current andwithretuningof theinput andoutput impedancesfor anoptimized
combinationof performancemeasures, it is possibletoget anRF output power onthe
order of 1012 W/mmwhilemaintaining high PAE at high operating frequencies, as
showninFigure3.13. Infact, evenhigher RF outputpower couldbeobtained, butatthe
expenseof PAE andgain, whichrapidlydegradeasoperatingfrequencyisincreased. In
fact, aspot RF output power densitygreater than30W/mmat 4GHz hasbeenreported
[45] foratransistorhavingL
g
=0.5mwithadrainbiasof V
ds
=120V. Thebreakdown
voltagefor thetransistor wasreportedtobeV
dB
=170V, whichindicatesthatsignicant
channel breakdownoccurredonthehighvoltageportionof theRF cycle. Nevertheless,
thepeak PAE was 54.8%with an associated gain of 14 dB. For theresults shown in
Figure3.13, thedeviceisbiasedat V
ds
= 40V andthegatebiasisadjustedfor ClassA
operation, whichresults ingoodPAE. ThePAE is about 50%from3GHz to10GHz,
anddeclines monotonically aboveX-band. However, at 30GHz thePAE is still 30%.
Thegain remains above10 dB up to about 25 GHz, and is still 8 dB at 30 GHz. By
retuningfor reducedPAE andgainit ispossibletofurther increasetheRF output power
toslightlygreater than12W/mm.
Theseresultsindicatethat AlGaN/GaN HFETs arecapableof excellent DC andRF
power performance well into the mm-wave frequency spectrum, and potentially well
above100GHz. Excellent WandE-bandRF performancehas beenreported[71, 75],
andanE-bandamplierproduced500mWoutputpowerwith12dBassociatedgainwith
about 10%PAE. Withtransistor power combiningtechnologyamplierswithhundreds
3.6 Transistor DC and large-signal RF models 125
I
out
Two-Port
Network
I
in
V
in
V
out
Figure 3.14 Small-signal two-port network.
to thousands of watts of RF output power shouldbeachievable. Theseampliers may
becapableof competingwithvacuumtubeampliersinmanyapplications.
3.6 Transistor DC and large-signal RF models
Mathematical modelsfor transistorsndwideapplicationfor bothdevicestructureopti-
mizationandcircuit designapplications. Basically, therearetwomajor typesof model
inpractice: (1) equivalentcircuitorientedmodelsthatareusedinDCandRF circuitand
systemdesignapplications; and(2) models that arebaseduponsemiconductor device
physics. Theequivalentcircuitbasedmodelsrequirethatthetransistor befabricatedand
characterizedbeforethemodel canbedened. However, oncedened, themodel canbe
usedinRF circuit andsystemdesignapplicationsandit providesameanstoinvestigate
RF circuit performanceandoptimizationwithout theneedtoactually fabricatethecir-
cuit until anoptimizeddesignis determined. Theequivalent circuit models havebeen
extensively developedandavariety of modelsarenowreadily availableinvirtually all
commerciallyavailablesimulators. Theequivalentcircuitbasedmodelsareonlyaccurate
over therangeof parametersfor whichtheyweredened, andoftenfail whenextended
outsidetheseparameters. For thisreason, newmodelsarecontinuallybeingderivedand
reportedintheliterature. Thephysics-basedmodelsoffer analternateapproach. These
modelsarebaseduponthefundamental semiconductordeviceequationsandcanbeused
to investigatethephysical operationof thetransistor beforefabricationoccurs. Inthis
manner, thephysics-basedmodels canbeusedto investigateanomalous physical phe-
nomenathat areobservedtooccur inthetransistor under variousoperatingconditions,
as well as for devicedesignoptimizationapplications. Thephysics-basedmodels are,
ingeneral, moredifcult to developandrequiremoreintensecomputer resources for
solution. Themodelsaresignicantlymoredifculttointegrateintocircuitandsystems
level simulators, andmost of thesemodels havebeendevelopedas stand-alonedevice
level simulators. A variety of thesesimulators is commercially availableandarevery
useful, particularlyfor devicedesignoptimizationapplications.
3.6.1 Equivalent circuit transistor models
Equivalent circuit models for transistors follow fromlinear two-port circuit analysis.
For example, a linear two-port can be represented by the block diagramshown in
Figure 3.14. Theinput to thecircuit block has input current and voltage, i
in
and :
in
,
126 Wide band gap transistors SiC and GaN physics, design and models
C
gd
R
g
g
d
C
gs
C
ds
R
d
r
ds
g
m
R
s

s
R
i
Figure 3.15 Small-signal tee-equivalent circuit for FETs.
andthecircuit blockoutput hasoutput current andvoltage, i
out
and:
out
. Anytwoof the
four variables may beselected as theindependent variables, and theother two as the
dependent variablesandnetwork equationsestablished. For transistor applicationsit is
commonto select acombinationof theinput voltage, :
in
, andtheoutput current, i
out
,
asthedependent variables, andtheinput current, i
in
, andtheoutput voltage, :
out
, asthe
independent variables. Thenetworkcanthenbedescribedbythematrixequation
_
:
in
i
out
_
=
_
h
11
h
12
h
21
h
22
_ _
i
in
:
out
_
(3.7)
Thisformulationcanbeusedtodenetheequivalent circuit model for thetransistor.
Thevarious h-parameter terms takeonaphysical meaning. For example, theh
11
term
hasunitsof resistanceandrepresentstheinput impedancetothenetworkwhenashort-
circuitis placed at the output terminals. Likewise, the h
22
termhas units of Siemens
(inverse O) and is the output admittance when an open circuit is placed at the input
terminals. Theh
21
parameter hasnounits, andrepresentstheoutput current normalized
to the input current, which is the forward current gain for the network. Likewise,
theh
12
parameter is theinput voltagenormalized to theoutput voltage, which is the
reversevoltagegainfor thenetwork. Other formulations arepossibledependingupon
the parameters selected as the dependent and independent variables. Since the entire
network is linear, oneset of parameters is easily converted to another throughsimple
linear transformations.
Theformulationpermitsanequivalent circuit for thetransistor tobeestablished. All
that needstobedoneistodetermineacongurationof circuit elementsthat replacethe
general network block and that generatetheexact samedynamic responseat thenet-
work terminalsastheoriginal circuit. Variouscircuit congurations canbedeveloped,
suchasthehybrid-pi, or theT (tee) circuit. For microwavetransistorstheteecircuit is
most commonlyused. Thebasiccircuit isshowninFigure3.15, andvariouscircuit ele-
mentsarealsoincludedtoindicatevariousparasiticelementsassociatedwithtransistor
operation. The equivalent circuit is exact in that it accurately reproduces the linear
electrical responseof transistor performanceand themain elements of theequivalent
circuit canbedirectly derivedmathematically fromtheoriginal two-port network. The
3.6 Transistor DC and large-signal RF models 127
equivalent circuit isvery useful for small-signal characterizationof thetransistor since
theequivalent circuit element values can bedirectly determined frommeasured data
by aparameter extractionprocess. Suitableroutines areavailableandnumerous prac-
tical techniques havebeen extensively reported in theliterature. Parameter extraction
softwareiscommerciallyavailablefromavarietyof vendors.
Nonlinear modelsfor large-signal RF performancehavebeendetermined, basedupon
thelinear equivalent circuit model for thetransistor. However, thisprocessisnot exact,
and it is, in general, not possible to theoretically derive an accurate nonlinear large-
signal model directly fromthe linear equivalent circuit. Although many large-signal
equivalent circuit models havebeenreportedintheliteratureandarereadily available
incommercial RF circuit simulators, all of themodelshavebeenderivedbydeveloping
nonlinear expressions for the various equivalent circuit elements and then using the
resulting expressions to dene thelarge-signal equivalent circuit model. Theproce-
durecanbecomecomplex, andgenerallythelinear equivalent circuit model isreduced
to themost important, basic equivalent circuit elements, whicharethencharacterized
by nonlinear functions of various combinations of input and output current and volt-
age. Thereducedequivalent circuit showingonlythebasiccircuit elementsisshownin
Figure3.16. Themostimportantelementsarethecurrentgenerator andtheinputcapac-
itance, andnonlinear expressionsbaseduponpower law, tanh(V
ds
), VolterraSeries, etc.
formulations have been developed. As themodel development progresses, additional
elementsintheequivalentcircuitcanbeformulatedasnonlinear functionsandincluded
in themodel. Theresulting equivalent circuit model can accurately predict thelarge-
signal RF performanceof thetransistor, butrequiresthattheparametersinthenonlinear
expressionsbedetermined. Theonlywaythiscanbeaccomplishedisbyanexperimental
parameter extractionprocesswherebythenonlinear termscanbedeterminedfrommea-
sureddata. Thecompletemodel canbecomplex, withtherequirement todenemany
parameter values fromtheextracted measurement data. Many techniques for accom-
plishingthishavebeenreportedintheliteratureandmostcommerciallyavailablecircuit
andsystemssimulatorsincludesuitablelarge-signal parameter extractionroutines.
A major issue with the large-signal equivalent circuit models is that, since they
arebased upon experimental extraction of thenonlinear elements, they aregenerally
accurateonly for therangeof parameters over whichthey havebeencalibrated. When
themodel is driven to regions outsidetheoriginal characterization space, thereis no
reasontoexpectthecircuitresponsetobeaccurate. Infact, themodelsoftenfail. Forthis
reasonnewmodelsarecontinuallybeingdevelopedandreported. Eachiterationandnew
equivalent circuit model development effort isdirectedtowardssolutionof apreviously
observedor reportedfailure. Thenewmodel, of course, alsorequiresdeterminationof
theelement parameter valuesbyexperimental extractionandcalibrationwithmeasured
data. The process requires that transistors be fabricated and characterized before a
suitablemodel canbedeveloped, andmany transistor manufacturersroutinely produce
equivalent circuit models for their transistors. However, these manufacturers models
generally only consider typical operating range data, and specic applications may
requirethat anewmodel bedened. However, oncetheequivalent circuit models are
determinedtheyhaveprovedveryuseful incircuit designapplications.
128 Wide band gap transistors SiC and GaN physics, design and models
Nonlinear Circuit Elements
Gate
Source
Drain
I
dg
(V
out
, V
in
)
I
gs
(V
in
)
C
gs
(V
in
, V
out
)
I
dg
(V
out
V
in
)
C
ds
(V
out
)
I
ds
(V
in
,V
out
)
R
in
(V
in
,V
out
)
R
ds
(V
in
,V
out
)
C
dg
(V
out
V
in
)
C
gs
(V
in
,V
out
)
I
gs
(V
in
)
R
in
V
in
(t)
I
ds
(V
in
,
V
out
)
V
out
(t)
R
g
R
ds
R
s
C
ds
C
dg
R
d
Drain-gate voltage-controlled current source
due to drain-gate avalanche breakdown
Gate voltage-controlled current source due to
forward biasing of the gate
Drain-source voltage-controlled current source
Drain-gate capacitance
Gate-source capacitance
Drain-source capacitance
Gate-source charging resistance
Drain-source resistance
Figure 3.16 Large-signal tee-equivalent circuit for FETs.
3.6.2 Physics-based large-signal transistor models
Analternateapproachtothedevelopmentof transistor modelsisbaseduponsolutionof
thebasicsemiconductor deviceequations. Thesemiconductor equationsconsist of:
(a) thecurrent densityequationsfor electronsandholes,

J
n
= q
n
n

E qD
n
n (3.8)
and

J
p
= q
p
p

E qD
p
p (3.9)
whereJ isthecurrent density, isthechargecarrier mobility, n, parethefreeelectron
andholedensities, andDisthediffusioncoefcient.
3.6 Transistor DC and large-signal RF models 129
(b) Thecontinuityequationsfor electronsandholes,
n
t
=
nn
0

n

1
q


J
n
(3.10)
and
p
t
=
p p
0

p

1
q


J
p
(3.11)
wheren
o
andp
o
arethethermal equilibriumdensityof electronsandholes, and
n
and

p
aretheelectronandholerecombinationlifetimes.
(c) FaradaysLaw,


E =

B
t
(3.12)
whichcompletesthebasicsetof equations, whereE andBaretheelectricandmagnetic
elds.
Theseequationscanbesolvedsimultaneouslytodevelopamodel forasemiconductor
device. Generally, solutionstotheseequationsappliedtoatransistor structurearecom-
plex anddifcult tosolveanalytically. However, theequationsarereadily solvedusing
numerical techniquesandavariety of simulatorsbaseduponeither nite-differenceor
nite-elementmethodshavebeenreportedandarecommerciallyavailable. Thesedevice
level simulatorspermitdetailedinvestigationof thephysical operationof thedeviceand
canbeusedtobothinvestigatephenomenaobservedinexperimental measurementsor
theycanbeusedfor devicedesignandoptimizationstudies. Ingeneral, thesesimulators
requiresignicant solutiontimeandaredifcult to employ incircuit-level simulators.
Thephysical models, however, canbeextremely accurateas all phenomenaknownto
affect deviceperformancecanbeincluded. Thephysical modelstakeasinput datathe
devicestructure, semiconductor material andtransport parameters, andbiasconditions.
Themodel canbesettotakethevoltageappliedtothedeviceterminalsasinputdataand
returntheresultingcurrentsthat ow, or set uptotakethecurrent appliedtothetermi-
nalsasinputdataandreturnterminal voltages. Thedeviceinputandoutputimpedances
canthenbecalculatedfromtheterminal voltagesandcurrents. Thephysical modelsare
extremely exibleandcanbemodiedto includephenomenathat arefoundto affect
deviceperformance, suchaschargetrapping, breakdownmechanisms, surfaceandinter-
facecharginganddischarging, andleakagecurrents, etc. Additionally, thedevicemodels
canbemodiedtoincludetransient andnonequilibriumphenomena, ballistictransport
effects, and quantumphysics behavior. These effects will increase the complexity of
themodel and generally increasethesimulation time, but theresulting model can be
madeextremelyaccurate. Devicesimulatorsof thistypendwideapplicationindevice
investigationsof operational physics, however, themodelsaregenerallynot suitablefor
inclusionincircuit-level simulators.
Itispossibletogenerateamodiedphysics-basedmodel thatissuitableforintegration
intocircuitandsystems-level simulators[76]. Inorder toaccomplishthis, itisnecessary
tocompromisetheformulationbetweeninclusionof pertinent physical phenomenaand
130 Wide band gap transistors SiC and GaN physics, design and models
solutionefciency. Thegoal is to produceamodel that maintains theimportant phys-
ical phenomenathat dominatedeviceperformance, whileproducingamodel that can
bevery quickly and efciently solved. In this manner theutility of thephysics-based
approachcanbecoupledwiththesimulationefciency of theequivalent circuit-based
approach. The model development proceeds by coupling a two-dimensional Poisson
equationsolutiontechniquewithaone-dimensional currentdensityequation. Byfocus-
ing the Poisson equation solution on the area under the gate electrode in a FET an
analytic solution can beobtained [76]. ThePoisson solution permits theelectric eld
within thetransistor to becalculated as afunction of structure, impurity doping, and
bias conditions. The electric eld is then used to calculate the channel current den-
sity withuseof thecurrent density equation. This approachworks well for structures
where the channel is narrow so that the current ow is essentially one-dimensional.
In fact, this model approach results in a physics-based transistor model that retains
the accuracy of the physics-based approach, but can be solved in an efcient man-
ner. Themodel can beintegrated into circuit and systems-level simulators with great
success.
Thesimulationworkreportedinthischapter makesuseof themodel describedabove,
whichhasbeenmodiedfor usewithwidebandgapsemiconductor devices. Excellent
resultsareobtainedandthemodel veryaccuratelypredictstheDCandRF performance
obtainedexperimentally.
3.7 Large-signal effects
Thewidebandgapsemiconductor FETsarecandidatesfor high-RF power applications
sincetheycanoperateunder high-voltageandhigh-current conditions. However, under
theseoperatingconditionsthedevicesexperienceavarietyof physical phenomenathat
affect their performance, and in some cases, produce deviations fromthe expected
response. Inmost casesthephenomenaarenatural physical responsestothevery high
voltages andcurrents that occur under RF large-signal operation. Themost signicant
of thesephenomenaaredescribedinthissection.
3.7.1 Space charge limited current transport
Under high-currentconditionstheinjectedchargeinasemiconductor canbecomecom-
parable in magnitude to the background impurity density and space-charge limited
transport can occur [77, 78, 79]. This condition can beachieved in practical devices
under large-signal RF operationwhenhigh-magnitudeinput RF power is applied. The
voltage that can be supported by a semiconductor device is limited by the internal
resistance, andwhenhighinput power is applied, theinjectedcurrent will increaseto
satisfy theboundary conditions imposedby theappliedsource. Generally, under these
conditions the input impedance to the device is driven to a reduced magnitude, and
althoughsomeof theinput power isreectedby thereducedimpedanceof thedevice,
3.7 Large-signal effects 131
currentinjectionincreases. Theinjectedchargeandtheinternal electriceldarerelated,
asexpressedbyPoissonsequation,
dE
dx
=
q

(N
d
n) (3.13)
whereE istheone-dimensional electric eldinthedirectionof current ow, N
d
isthe
effectivedonor density that represents thepositivepolarization/piezoelectric chargein
HFETs, n= n
o
n is thefreeelectron density wheren
o
is thethermal equilibrium
density of charge, and n is the density of injected charge. The thermal equilibrium
density of electrons is essentially equal to thedonor density (i.e., n
o

= N
d
) andwhen
theinjectedchargebecomescomparableinmagnitudetothethermal equilibriumdensity
of electronsPoissonsequationiswrittenas
dE
dx
=
q

(N
d
n
o
n)

=
q

n (3.14)
Under high-injectionconditionstheelectriceldisreducedinmagnitudeasafunction
of increasingchargeinjection, andtheresistivity andresistanceof thesemiconductor
material becomeafunctionof current injection. Thiseffect canbecomesignicant for
semiconductor devices operatedunder high-current injectionconditions. For low-level
injectionconditions wheren_ n
o
, theE eldis essentially independent of injection
level.
Inorder to determinetheconditions under whichspace-chargeeffects becomesig-
nicant, it is illustrative to solve equation (3.13) analytically. The current density
is
J = qn: (3.15)
where J (A/cm
2
) is themagnitudeof thecurrent density and: (cm/s) isthenonlinear
velocityeldcurve, whichismodeledas
: =
E
1[E[ ,E
sat
(3.16)
where(cm
2
/V s) istheloweldmobilityandE
sat
(V/cm) isthemagnitudeof electrical
eldthat producesvelocitysaturation, expressedas
E
sat
= :
sat
, = 8.0kV,cm. (3.17)
Equation(3.13) canbewrittenintheform:

E
sat
dE
dx
= 1
J
J
sat
E E
sat
E
(3.18)
wherethe parameter is
=
E
sat
qN
d
(3.19)
and has the dimensions of length. The parameter is typically very small for an
AlGaN/GaN heterojunction2DEG, sincetheeffectivedopingisveryhigh.
132 Wide band gap transistors SiC and GaN physics, design and models
Current Density (MA/cm
2
)
R
e
s
i
s
t
i
v
i
t
y

(

-
c
m
)
D = 0.5 m
D = 1.5 m
D = 2.5 m
D = 3.5 m
41 40.5 40 39.5 39
0
0.2
0.4
0.6
0.8
Figure 3.17 Semiconductor resistivityversuscurrent densityat theonset of spacechargelimited
transport (thevariouscurvesindicateresistivitydeterminedat locationsfromthecurrent
injectionpoint) [77].
The solution of equation (3.18) depends on its magnitude at x = 0 which should
reect acombination of low E and high n. For thelimiting case E(0) = 0, equation
(3.18) hasasolution:
E(x) =
E
sat
J
J
sat
J
_
1 W
_
e
1
(J
sat
J )
2
J
sat
J
x

__
(3.20)
thatcanbesimplywrittenintermsof theprinciplereal branchof theLambertWfunction
for J - J
sat
andtheother real branchof Wfor J > J
sat
. Inequation(3.20), thelength
scale
L
J
= J
sat
J
_
(J
sat
J )
2
(3.21)
islarger thanthelengthscale of equation(3.19) but isstill lessthananAngstromfor
J =
1
2
J
sat
. L
J
divergesnear J

= J
sat
.
Theresistivity, = E
_
J , asafunctionof distancefromthesourcefor semiconductor
fabricated using an AlGaN/GaN heterojunction 2DEG is shown in Figure 3.17. The
resistivity is essentially independent of J until a critical threshold current, J
SC
, is
achieved. In this example J
SC
39.6 MA/cm
2
. For a current density in excess of
J
SC
the resistivity increases rapidly both with current, and with distance from the
source injection point. Therefore, once space-charge limited transport conditions are
established, the resistivity of a semiconductor will rapidly increase, and the effect is
moresignicant withthelengthof thesemiconductor region.
Theanalytic solution of equation (3.18) is continuous at J = J
sat
but its character
changes. For J - J
sat
, E(x)

= E
sat
J
_
(J
sat
J ) isalmostindependentof xexceptinthe
3.7 Large-signal effects 133
region0- x - L
J
whereE(x)

x. For J > J
sat
, E(x)

= x
(J J
sat
)
:
sat
increasesalmost
linearlywithx. Furthermore,
E(x) = E
sat
J
_
(J
sat
J ) (3.22)
is a second exact solution of equation (3.18) for J - J
sat
, in addition to being an
approximationof equation(3.20) intheregion x > L
J
near thegateedge. Incontrast
toequation(3.20), however, equation(3.22) isnot continuousat J
sat
andisnot physical
at J = J
sat
. This peculiar situation raises the possibility of mode-switching between
thecontinuous solutionandtheconstant solution. Themode-switchingtransitionmay
occur for L
J
> L
sg
, when thelength scale L
J
of equation (3.21) exceeds thelength
of theaccess region. This transitionwouldbeabrupt inpractical devices where L
sg
is
on theorder of amicron and is much larger than = .32

A, thelength parameter of
equation(3.19), becauseL
J
> L
sg
onlyfor J
_
J
sat
> 1
_

_
L
sg
asJ approachesJ
sat
.
The magnitude of the J
SC
threshold current is approximately given by the
expression
J
SC

= qN
d
:
sat
(3.23)
and for aheterojunction 2DEG, it is assumed that N
d

= n
ss
,h, whereh is thewidth
of the2DEG quantumwell. For atypical AlGaN/GaN 2DEG space-chargeeffects are
expectedtosetinforathresholdcurrentintherangeof J
SC
40MA/cm
2
. Thisislower
thanthecurrent density measuredinexperimental devices. Practical devices generally
haveamaximumdraincurrent of I
ds
(11.2) A/mm, andfor aquantumwell width
of abouth=25

A, thecurrentdensityisintherangeof J 50MA/cm
2
. Thisindicates
thatthesedevicesmostlikelyoperateunderspace-chargelimitedconditions, particularly
during thehigh-current portion of theRF cycle. In addition, themagnitudeof J
SC
is
expectedtovarywithn
ss
and:
sat
, andtheseparametersvaryinmagnitudewithDCand
large-signal RF operatingconditions. Themagnitudeof J
SC
, therefore, canvary with
HFET biasandRF drive, andthisincreasesthenonlinearityof thegate-sourceresistance
for anHFET under large-signal drive.
3.7.2 Nonlinear source and drain resistance
Theonset of space-chargelimitedcurrent transport inamicrowaveHFET under large-
signal operation will cause the source and drain resistances to signicantly increase
during thehigh-current portion of theRF cycle. Although both thesourceand drain
resistancesareaffected, theincreaseinsourceresistancehasthemost signicant effect
upontransistor performance. Thedrainresistanceisessentially inserieswiththerela-
tively high-magnitudeoutput loadresistanceand, therefore, theincreaseinthedevice
drain resistance has minimal effect upon device performance. The source resistance,
however, iscommontothetransistor inputandoutput, asindicatedinthetee-equivalent
circuit, asshowninFigure3.16, andany increaseinsourceresistancewill degradethe
transconductanceof thetransistor byreducingthevoltagethatdrivesthecurrentgenera-
tor. ForthisreasonitisimperativetoreducethesourceresistanceinaFET tothegreatest
134 Wide band gap transistors SiC and GaN physics, design and models
C
gs
i
g
m
i
R
gs
Figure 3.18 Simpliedequivalent circuit for aFET withanonlinear sourceresistance.
C
gs
ext
g
m
ext
Figure 3.19 Simpliedequivalent circuit for aFET withthenonlinear sourceresistance
transformedtothegatecapacitanceandtransconductance.
possibleextent. Theincreaseinsourceresistanceduetotheonsetof space-chargelimited
transport conditions will degradetransistor performance, both by decreasing thegain
capabilityof thetransistor, andbyintroducingundesirablenonlinearity.
Theeffect of thenonlinear sourceresistancecanbeseenby reducingtheequivalent
circuitinFigure3.16toitsbasiccircuitelements,asshowninFigure3.18.Thisequivalent
circuit canbetransformedto theequivalent circuit showninFigure3.19by rewriting
theelement valuesas
g
m
ext
=
g
m
i
1 R
gs
g
m
i
j R
gs
C
gs
i
(3.24)
and
C
gs
ext
=
C
gs
i
1 R
gs
g
m
i
j R
gs
C
gs
i
. (3.25)
Intheseexpressionsg
m
ext
andC
gs
ext
arethetransconductanceandgate-sourcecapac-
itancethat areobservedat theinput totheequivalent circuit inFigure3.18andg
m
i
and
C
gs
i
arethetransconductanceandgate-sourcecapacitanceintrinsictothetransistor, and
representedby theequivalent circuit showninFigure3.19. For lowfrequencies, these
equationssimplifyto
g
m
ext

=
g
m
i
1 R
gs
g
m
i
(3.26)
and
C
gs
ext

=
C
gs
i
1 R
gs
g
m
i
. (3.27)
According to equations (3.26) and (3.27), both thetransconductanceand gate-source
capacitanceobservedat theterminalsof thetransistor will decreaseasthegate-source
resistanceincreases. ThedecreaseinC
gs
ext
isparticularlyinteresting, sincethisindicates
that themagnitudeof theinput impedanceto theHFET will increaseas thedeviceis
driven into saturation. This is opposite to the normal operation of a FET, where the
input impedanceisdriventoalower magnitudeasthedeviceisdrivenintosaturation.
3.7 Large-signal effects 135
0
1
1.5
C
g
s

(
p
F
)
0.5
4 1 6
P
in
(dB m) (GHz)
Eqn( )
Eqn( )
Cgs_4_0_20
Cgs_4_0_30
11 16 21
2
Figure 3.20 Measuredgatesourcecapacitanceasafunctionof RF input power for an
AlGaN/GaN HFET for V
ds
= 20V, andV
ds
= 30V.
Thebehavior indicatedinequation(3.27) is supportedby measureddata, as shownin
Figure3.20. Inthisgurethemeasuredinputcapacitanceasafunctionof inputRFpower
driveisshownforanAlGaN/GaNHFET. TheincreasedRFpowerdrivecausesincreased
channel current, which has a magnitude sufcient to exceed thethreshold for space-
chargelimitedcurrent transport. TheSCL current conditionsproduceanincreaseinthe
gate-sourceresistance, whichproduces thereductioninC
gs
ex
as expressedinequation
(3.27). The C
gs
ex
magnitude is reduced by almost a factor of two over the measured
rangeof input power, andthisproducesanincreaseinthedeviceinput impedancebya
correspondingfactor. TheSCL currenttransportphenomenonhasthedesirableresultof
increasingtheterminal impedances, whichmakeit easier todesignthetransistor input
amplier, andother circuits.
Theonsetof space-chargelimitedcurrentcantheoreticallyaffectbothdepletionmode
HFETs, as well as enhancement modeMOS typeFETs (e.g., Si LDMOS FETs). The
depletionmodeHFETshaveaninherentadvantageof lowerinputcapacitancecompared
to the enhancement mode FETs for a constant RF output power and supply voltage
dueto geometrical factors (e.g., thicker dielectric layers that result in theconducting
channel beinglocatedfarther fromthegateelectrode). It shouldbenotedthat theonset
of space-charge limited current and the increase in the gate-source resistance under
large-signal operation conditions has not been observed in the normal operation of
Si LDMOS FETs or compound semiconductor MESFETs and HEMTs. Theaffect is
commonly observedandappears to dominateinthenitride-basedHFETs, most likely
due to the higher current densities and internal electric elds under which the wide
bandgapsemiconductor nitride-basedHFETsoperate.
Themeasured and simulated performanceof thesourceand drain resistances as a
functionof currentfor anAlGaN/GaNHFET areshowninFigure3.21. Asshown, once
136 Wide band gap transistors SiC and GaN physics, design and models
Drain Current (A)
R
s
R
d
Measured
Simulated
12
10
8
6
4
2
0
0.1 0 0.1 0.2 0.3 0.4 0.5 0.6
R
s
,

R
d

(
O
h
m
)
Figure 3.21 Measuredandsimulatedsourceanddrainresistancesasafunctionof draincurrent for
anAlGaN/GaN HFET.
space-chargelimitedcurrenttransportconditionsareachievedthesourceanddrainresis-
tancesdemonstratecurrent-dependentcharacteristicsandincreasewithcurrentdrive. In
fact, simulationsindicatethat under high-current driveconditionsthesourceanddrain
resistancefor theHFET canincreasebyuptoanorder of magnitude.
Theeffect of anonlinear sourceresistanceupon on an HFET amplier circuit can
besignicant. Thiscanbeshownbyacircuit simulationinwhichanonlinear, current-
dependent source resistance is included [77, 80]. For this study, a harmonic-balance
simulator that includes a physics-based FET model [76] is used. The HFET device
model hasbeenmodiedbyinclusionof asourceresistanceintheform
R
s
= r
ss
Lr
ss
=
r
ss
1
I
I
SC
. (3.28)
where R
s
is the source resistance, r
ss
is the low-current magnitude of the resistance
in the gate-source region, Lr
ss
is the increase in resistance after the onset of space-
chargeeffects, andI
SC
isthespacechargethresholdcurrent previouslydiscussed. The
nonlinearsourceresistanceisafunctionof thetime-dependentRFcurrentandisincluded
on thetimedomain, nonlinear sideof theharmonic-balanceinterface. Inthis manner
thesourceresistanceisafunctionof theconductioncurrent inthetransistor.
Themodiedsimulator was usedtoinvestigatetheDC andRF operationof acom-
municationsbandAlGaN/GaNHFET amplier. TheHFET devicehadagatelengthand
width of L
g
= 0.8mand W= 0.4mm, respectively. Thedevicewas biased with a
drainvoltageof V
ds
= 28: andwasoperatedclassA-Batafrequencyof F = 2.14GHz.
Thedevicedemonstratedprematuregaincompressionandwas, therefore, selectedasa
3.7 Large-signal effects 137
Drain Voltage (V)
D
r
a
i
n

C
u
r
r
e
n
t

(
A
)
0
0
0.1
0.2
0.3
0.4
5 10 15 20 25
V
gs
= +1 V
V
gs
= 0 V
V
gs
= 1 V
V
gs
= 2 V
V
gs
= 3 V
V
gs
= 4 V
Figure 3.22 MeasuredandsimulatedDC IVcharacteristicsfor anAlGaN/GaN HFET
(L
g
= 0.8mm, W= 0.4mm) [77].
candidatetodetermineif anonlinear sourceresistancecouldexplainthegaincompres-
sionbehavior.
The measured and simulated DC IV characteristics for the HFET are shown in
Figure3.22. AsindicatedinFigure3.22, excellentagreementbetweenthemeasuredand
simulateddataisobtained. Inthesimulationalow-eldmobilityof = 1500cm
2
/V
sec and an electron saturation velocity of :
sat
= 1.2510
7
cm/sec were used. The
mobility wasmeasuredandthesaturationvelocity wasadjustedtoget agreement with
themeasuredIV characteristics. Thesaturationvelocity usedis belowthetheoretical
value for electrons in an AlGaN/GaN 2DEG, but is consistent with measured data.
Without inclusionof thecurrent-dependent nonlinear sourceresistance, thesimulated
current increasingly deviated fromthe measured data as the gate bias voltage was
increasedfrompinch-off andadjustedfor increasingchannel current.
ThemeasuredandsimulatedRF performanceandtheDC drainandgatecurrent as
afunction of input power to theamplier areshown in Figures 3.23, 3.24, and 3.25,
respectively. In the simulation seven harmonics were used in the harmonic-balance
routine. The amplier was tuned for maximumPAE. Excellent agreement between
the measured and simulated device performance and the DC drain and gate current
are obtained. The amplier produced a peak PAE of 53%, with RF output power of
34dBmandagainof 19dB.Thelineargainfortheamplierwas25dB.Theexperimental
amplierdemonstratedprematuregaincompressionandadegradationof gainbeginning
at an input power of slightly below about 0 dBm. Signicantly, thesimulated results
accuratelypredictthechangeinslopeof thegainresponse, asshowninFigure3.23. The
138 Wide band gap transistors SiC and GaN physics, design and models
Input Power (dB m)
P
o

(
d
B

m
)

G
a
i
n

(
d
B
)

P
A
E

(
%
)
10 5 0
0
5
10
15
20
25
30
Nonlinear Source Resistance Onset
Sim. Po
Meas. Po
Meas. G
Sim. G
Meas. PAE
Sim. PAE
35
40
45
50
55
5 10 15 20
Figure 3.23 MeasuredandsimulatedRF output power, gain, andPAE versusRF input power for
anAlGaN/GaN HFET amplier (freq= 2.14GHz, V
ds
= 28V, ClassAB) [77].
Input Power (dB m)
D
r
a
i
n

C
u
r
r
e
n
t

(
A
)
10
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
Simulated
Measured
8 6 4 2 0 2 4 6 8 10 12 14 16 18 20
Figure 3.24 MeasuredandsimulatedDC draincurrent versusRF input power for anAlGaN/GaN
HFET amplier (freq= 2.14GHz, V
ds
= 28V, ClassAB) [77].
3.7 Large-signal effects 139
Input Power (dB m)
G
a
t
e

C
u
r
r
e
n
t

(
A
)
10
0.0002
0
0.0002
0.0004
0.0006
0.0008
0.001
0.0004
8 6 4 2 0 2 4 6 8 10 12 14 16 18 20
Figure 3.25 MeasuredandsimulatedDC gatecurrent versusRF input power for anAlGaN/GaN
HFET amplier (freq= 2.14GHz, V
ds
= 28V, ClassAB).
(a) Time (ps)
G
a
t
e

V
o
l
t
a
g
e

(
V
)
Pin = 14.8 dB m
Pin = 0.0 dB m
0
10
8
6
4
2
0
2
50 100 150 200 250 300 350 400 450 500
Figure 3.26a RF gatevoltageversustimeresponsefor anAlGaN/GaN HFET amplier
(thetwowaveformsindicateoperationat theonset of spacecharge-limitedtransport,
andunder maximumPAE).
simulator predictsboththechangeingainslopeatP
in
=0dBm, andthegainsaturation
that occursafter maximumPAE isachieved. Thechangeingainslopeiscausedbythe
onset of space-chargelimitedcurrent transport conditions.
The time domain voltage and current waveforms at the gate and drain termi-
nals are shown in Figures 3.26 and 3.27. Figures 3.26a and 3.26b show the voltage
140 Wide band gap transistors SiC and GaN physics, design and models
(b) Time (ps)
G
a
t
e

C
u
r
r
e
n
t

(
A
)
Pin = 14.8 dB m
Pin = 0.0 dB m
0
0.2
0.15
0.1
0.05
0
0.05
0.1
0.15
0.2
50 100 150 200 250 300 350 400 450 500
Figure 3.26b RF gatecurrent versustimeresponsefor anAlGaN/GaN HFET amplier (thetwo
waveformsindicateoperationat theonset of space-charge-limitedtransport conditions, and
under maximumPAE.
(a) Time (ps)
D
r
a
i
n

V
o
l
t
a
g
e

(
V
)
Pin = 14.8 dB m
Pin = 0.0 dB m
0
0
10
20
30
40
50
60
50 100 150 200 250 300 350 400 450 500
Figure 3.27a RF drainvoltageversustimeresponsefor anAlGaN/GaN HFET amplier (thetwo
waveformsindicateoperationat theonset of spacecharge-limitedtransport conditions, and
under maximumPAE).
and current waveforms at the gate terminal under low drive (P
in
= 0 dBm) and
large-signal operating conditions wheretheinput power is sufcient to producemax-
imumPAE. The same waveforms at the drain terminal are shown in Figures 3.27a
and3.27b.
3.7 Large-signal effects 141
(b)
D
r
a
i
n

C
u
r
r
e
n
t

(
A
)
P
in
= 14.8 dB m
P
in
= 0.0 dB m
0.25
0.2
0.15
0.1
0.05
0
0.05
0.3
0.35
0.4
Time (ps)
0 50 100 150 200 250 300 350 400 450 500
Figure 3.27b RF draincurrent versustimeresponsefor anAlGaN/GaN HFET amplier (thetwo
waveformsindicateoperationat theonset of spacecharge-limitedtransport conditions, and
under maximumPAE).
ForthelowRFdriveP
in
=0dBmcondition,thegatevoltageandcurrentareessentially
low-amplitudesinusoids, asexpected. SincetheP
in
= 0dB driveconditionissufcient
to producesomesaturation, aslight deviationfrompuresinusoidal behavior is noted,
particularlyintheRF voltage. TheshiftinphaseinthegateRF voltagewaveformatthe
higherdriveconditionshowninFigure3.26aisduetotheshiftingate-sourcecapacitance
atthehigher drivelevels, aspreviouslydiscussed. Asthedeviceisdrivenintosaturation
thegatevoltagegrowsinamplitudeandshowstheeffectsof harmonicgeneration. The
gateRFcurrentwaveformshowninFigure3.26bbecomeshighlynonlinearasthedevice
isdrivenintosaturation,andthesevenharmonicsareclearlyevident.Themechanismthat
causesthenonlinearityisnot evident inthegateterminal waveforms. Thegatevoltage
does not obtain a magnitude sufcient to cause either forward or signicant reverse
conductionof thegateelectrode. Somereverseconductiondoesoccur andthisgenerates
a small, but nite, DC reverse conduction in the gate electrode. The small negative
reversegateconductionwasobservedinboththeexperimental dataandthesimulation.
However, thesmall amount of reverseconductionis not sufcient to clipthegateRF
current waveformand generate the nonlinear behavior observed in the waveformin
Figure3.26b.
TheRF voltageand current waveforms at thedrain terminal areshown in Figures
3.27a and 3.27b, respectively. Again, the waveforms for the low-drive and maxi-
mumPAE conditions areshown. Thelow-driveRF drainvoltagewaveformshownin
Figure3.27aindicatestheonsetof saturation, andslightdeviationfromsinusoidal behav-
ior is observed. Thelarge-signal RF waveformdemonstrates signicant clipping, both
at lowandhighRF voltages, andthewaveformbecomesmoresquared inshape. The
clippingat thelow-drainvoltages is causedby thetotal RF terminal voltagedropping
142 Wide band gap transistors SiC and GaN physics, design and models
N
o
n
l
i
n
e
a
r

S
o
u
r
c
e

R
e
s
i
s
t
a
n
c
e

(
O
h
m
)
P
in
= 14.8 dB m
P
in
= 0.0 dB m
1
1.5
2
2.5
3
3.5
4
5
4.5
Time (ps)
0 50 100 150 200 250 300 350 400 450 500
Figure 3.28 Sourceresistanceversustimeresponsefor anAlGaN/GaN HFET amplier (thetwo
waveformsindicateoperationat theonset of spacecharge-limitedtransport conditions, and
under maximumPAE).
belowtheRF kneeof theIV characteristic, andtheclippingat highdrainvoltages is
causedby theonset of RF breakdownintheconductingchannel. TheRF draincurrent
waveforms areshown in Figure3.27b. The large-signal RF current demonstrates the
squaring behavior caused by thedrain voltagewaveformclipping mechanisms that
occur at lowandhighvoltagemagnitudes. TheRF voltageandcurrent areessentially
out-of-phase, withthecurrentmagnitudebeinghighwhenthevoltagemagnitudeislow,
andviceversa. Thewaveformclippinggeneratesharmonics, whichareclearlyobserved
inthelarge-signal RF current.
TheRFdraincurrentshowninFigure3.27bconsistsessentiallyof conductioncurrent,
which ows through the conducting channel fromthe source to the drain. When the
threshold for space-charge limited ow is reached the resistance of the material will
becomeafunctionof themagnitudeof thecurrent andthegate-sourceresistancewill
become nonlinear. The magnitudes of the source resistance under the low drive and
large-signal conditionsareshowninFigure3.28. Theonset of space-chargedependent
resistanceisobservedfor theP
in
=0dBminputdriveconditionduringthehigh-current
portion of the RF cycle. The source resistance increases froma DC magnitude of
R
s
= 1.85 O to apeak magnitudeof about R
s
= 2.4 O. However, under large-signal
conditions thesourceresistancebecomes highly nonlinear and increases signicantly
during the high-current portion of the cycle. Since the RF drain current shown in
Figure3.27biscomposedessentiallyof conductioncurrent, themagnitudeof thesource
resistanceisdirectlydependentuponthiscurrent. Themagnitudeof thesourceresistance
increasesfromtheDC magnitudeof R
s
= 1.85O toalmost R
s

= 5O duringthepeak
of theRF draincurrent. Increasingthedrainvoltageresultsinanincreasedmagnitudeof
thenonlinearsourceresistanceduringthehigh-currentportionof theRF cycle, asshown
3.7 Large-signal effects 143
Time (ps)
v
ds
= 28 V
v
ds
= 38 V
v
ds
= 48 V
N
o
n
l
i
n
e
a
r

S
o
u
r
c
e

R
e
s
i
s
t
a
n
c
e

(
O
h
m
)
0
1
2
3
4
5
6
7
8
9
10
50 100 150 200 250 300 350 400 450 500
Figure 3.29 Sourceresistanceversustimeresponsefor analgan/ganHFET amplier at maximum
PAE conditionsfor V
ds
= 28V, 38V, and48V.
inFigure3.29. Inthis guretheresults obtainedby increasingthedrainvoltagefrom
V
ds
= 28V, toV
ds
= 48V areshown. For V
ds
= 48V thesourceresistanceincreasesby
almostanorder of magnitudeduringthehigh-current portionof theRF cyclecompared
tothesmall-signal value.
The nonlinear behavior of the source resistance helps explain the behavior of the
RF gate current shown in Figure 3.26b. Although the RF gate current is composed
essentially of displacement current, the gate circuit requires conduction through the
gate-sourceregion, whichisnormallyalow-valueresistance, andthemagnitudeof the
RF gatecurrentisdependentuponthemagnitudeof thesourceresistance. Theharmonic
generationduetotheclippingof theRFdraincurrentistransferredtothegatecircuitand
themagnitudeof thegatecurrent is, therefore, affectedbythemagnitudeof thesource
resistance. Theinput impedancetothetransistor is essentially aseries combinationof
thegate-sourcecapacitanceandthenonlinear sourceresistance.
The nonlinear source resistance has a signicant effect upon the operation of the
device. The source resistance essentially couples the input gate circuit to the output
drain circuit for thedevice, as shown in theequivalent circuit shown in Figure3.18.
Thedrain current generator is driven by thevoltagegenerated across thegate-source
capacitance, accordingtotheexpression,
i
out
= g
mi
:
gs
e
j
(3.29)
wherei
out
is theHFET RF output current, g
mi
is theintrinsic transconductance(mS),
(s) is a delay time, and :
gs
is the RF voltage across the gatesource capacitance.
Thetransconductancethat is developed at thedeviceoutput is reduced by thesource
resistance, accordingtoequation(3.26), aspreviouslydiscussed.
144 Wide band gap transistors SiC and GaN physics, design and models
E

(
V
/
c
m
)
E

(
V
/
c
m
)
@ Mid-Point of Conducting Channel
E = 2 10
6
V/cm
@ Surface of AlGaN Layer
Microns
Microns
Figure 3.30 Electriceldmagnitudeversusdistanceat themidpoint of theconductingchannel
andat thesurfaceof theAlGaN Layer for anAlGaN/GaN HFET (thetopcurveisthetotal E
eldandthebottomcurveisthex-directedE eld. Thedottedlinesindicatethecritical E eld
for breakdowninGaN.) [79].
3.7.3 Gate leakage
WhenahighdrainbiasvoltageisappliedandtheHFET isdrivenwithalargeRF signal
thepeak voltageat thedraincanobtainamagnitudeessentiallytwicethemagnitudeof
thebias voltage. Detailed simulations indicatethat themagnitudeof theelectric eld
at theedgeof thegateelectrodeonthedrainsidecaneasily exceedEc68MV/cm,
as shown in Figure3.30, which is sufcient to producequantummechanical electron
tunneling.
ThecurvesinFigure3.30indicatetheelectriceldatthemid-pointof theconducting
channel andat thesurfaceof theAlGaN layer. Boththemagnitudesof thetotal electric
eldandthex-directed(i.e., inthedirectionof current ow) electric eldareshown.
The dotted line indicates the assumed breakdown voltage, which is in the range of
Ec2 MV/cm. As indicated, both thetotal and x-directed electric eld for thestated
operating conditions signicantly exceed thebreakdown voltage. Theelectric eld at
the gate edge near the surface has a magnitude on the order of E8 MV/cm, which
is sufcient toproducesignicant electrontunneling. Experimental dataindicates that
electrontunneling, infact, occurs.
Theelectronsthattunnel fromthegateelectrodecan(a) accumulateonthesurfaceof
thesemiconductornexttothegate,(b)conductalongthesurfacebyatrap-to-traphopping
mechanism, creatingagate-to-drainleakagecurrent, or (c) possibly travel throughthe
AlGaNlayer tothe2DEGconductingchannel, asshowninFigure3.31. Measureddata
indicatethat thesurfaceleakagepathisdominant under practical operatingconditions,
3.7 Large-signal effects 145
Electrostatic feedback
Surface Leakage
(Primary mechanism)
Gate/Channel Leakage
(Secondary mechanism)
Electrostatic
depletion
Electron
Tunneling
AIGaN
2DEG
GaN
Gate
Figure 3.31 GateelectronleakagepathsinanAlGaN/GaN HFET [79].
and the path through the AlGaN layer only occurs for extreme conditions following
defect creationthat canoccur under highelectricelds, etc.
Inaddition, if theenergyof theelectronsissufcientlyhigh, theycancauseavalanche
ionizationonthesurfacenext to thegate. Whenthis occurs electrons tunnel fromthe
gatemetal tothesemiconductor surfaceareaadjacent tothegatewithsufcient energy
to causeavalancheionization, which is accompanied by light emission fromthegate
edge. Lightemissionfromthegateedgeisoftenobservedinthelarge-signal operationof
GaAsMESFETsandInP-basedHEMTs, andhasbeenobservedinAlGaN/GaNHFETs
under certainoperatingconditions. Thisindicatesthat surfacebreakdownoccursinthe
nitridedevicesandcanbeafactor affectingreliability. Avalancheionizationalsooccurs
in theconducting channel of thesedevices and is afactor in theRF operation of the
devicewhenoperatedunderlarge-signal drive. RFchannel breakdownis, infact, afactor
that affectsgainsaturationintheHFET devices.
When the electrons accumulate on the surface of the semiconductor at the gate, a
virtual gate effect is created, where the gate effectively increases in length as the
electrontunnelingproceeds andthedensity of electrons onthesemiconductor surface
increases. Theelectronsthat accumulateonthesurfaceof thesemiconductor createan
electrostaticchargethatproducesapartial depletionof theconductingchannel electrons,
therebycausingareductioninthechannel current, andacorrespondingdecreaseinRF
output power. Theelectrontunnelingandchargeaccumulationcontinueas afunction
of time. Thismechanismistheprimaryphysical basisfor thenondestructivereliability
problemassociated with these devices. Typical performance degradation is shown in
Figure3.32, whichshowsthemeasuredDC channel current andRF output power asa
functionof time. Theincreasingelectrostatic chargeacts tosuppress further tunneling
of the electrons fromthe gate metal, thereby limiting the effect. In this manner, the
mechanismisself-limiting.
Themeasured DC conduction current degradation shown in Figure3.32 correlates
withadegradationinRF outputpower. Thecurrentconductioncharacteristicsvarywith
timeand with devicedesign, surfaceprocessing, and passivation, and varying power
degradationresultsareobtained. It ispossibletomodifyandreducethetunnel leakage
bytheuseof optimizedeld-platedevicedesigns, andbytheuseof passivation, which
minimizestheRFpowerdegradation. Withproperandoptimizedpassivation, DCcurrent
146 Wide band gap transistors SiC and GaN physics, design and models
C
h
a
n
g
e

i
n

P
o
u
t

(
d
B
)
C
h
a
n
g
e

i
n

I
d
s
s

(
%
)
Stress Time (hr) (a)
(b) Stress Time (hr)
2
0 100 200 300 400 500 600 700 800 900
0 100 200 300 400 500 600 700 800 900
1
0
1
10
20
10
0
Figure 3.32 (a) Measuredchangeinchannel current (I
dss
) and(b) RF output power versustimefor
anAlGaN/GaN HFET (variouslinesindicatedifferent devicesincludedinthemeasurement)
[79].
andRF output power degradationcanbeminimal, at least for limitedrangesof DCbias
voltage.
3.7.4 Reliability and time-dependent performance degradation
Oneof thedominant reliability problems experienced by nitride-basedHFET devices
hasbeenlinkedtogateleakage[79, 80]. Althoughgateleakageisnot theonlyproblem
affectingdevicereliability, itisarst-order problemthatneedstobesolvedbeforethese
devices nd widespread application. The problemis manifested as a time-dependent
decreaseindraincurrent andRF output power, as showninFigure3.32, andhas been
primarilyaddressedthroughsurfacepassivationtechniques.
Thedegradationisobservedtovary signicantly withsurfacepassivationprocesses
andfrommanufacturer tomanufacturer. Also, gateleakageisnot theonly mechanism
that resultsindraincurrent andRF output power degradationbut it wasamongtherst
tobeaddressed. Thephenomenonisreversibleanddoesnotproducepermanentdamage
3.7 Large-signal effects 147
RF channel
breakdown
R
s
I
g
t
t
I
for
I
tun
I
rev
I
con
I
chbd
I
d
R
d
Gate
Drain
Gate tunnel leakage
Source
[C]
Figure 3.33 Large-signal HFET model usedintheperformancesimulations[79].
or degradationto thedevice, andaperiodof inactivity generally results inthedevice
returningtoitsinitial performance. However, thisrecoveryhasalsobeenobservedtobe
afalserecovery aswhenstressisreappliedtothedevice, it quickly degradestoitslast
degradedstate. Additionally, under certainoperatingconditions, asuddenreliability
problemhas beenobserved[81], wherepermanent degradationindeviceperformance
occurs. Devicesthat experiencethisproblemarecharacterizedbyhigh-magnitudegate
leakage.
A model for gate tunnel leakage in GaAs MESFETs has been reported [82]. This
model has been modied for use with AlGaN/GaN HFETs, and the modied model
canbeusedinaharmonic-balancesimulator toinvestigatethegatetunnel mechanism
as afunctionof DC andRF operatingconditions. Themodel is showninFigure3.33.
The gate tunnel leakage is represented as a current generator between the gate and
drainelectrodes. Themodel alsoincludesRF breakdownwithintheconductingchannel,
whichisrepresentedbyacurrent generator betweenthedrainandsource.
Thismodel accuratelysimulatestheDCandRF performanceof AlGaN/GaNHFETs,
andthesimulatedandmeasuredRF performancefor aclassA-B 2.14GHz communi-
cationsbandAlGaN/GaN HFET amplier areshowninFigure3.23, andthemeasured
andsimulatedI
ds
andI
gs
as afunctionof input power wereshowninFigures 3.24and
3.25, respectively. As shown in Figure3.25, thegateconducts asmall, but niteand
negativeleakagecurrent for theentirerangeof input power, until thegatejunctionis
drivenintoforwardconductionataninputRF power of aboutP
in
=17dBm. Themodel
is in excellent quantitativeagreement with themeasured datafor thereverseleakage
conduction characteristics of the gate, and in qualitative agreement for the forward
conduction, but slightly underestimates theinput power requiredto drivethegateinto
forwardconduction.
Aselectronstunnel fromthegatetotheAlGaN surfacethey canaccumulatenext to
thegateelectrode. Thespacechargefromtheelectrons provides anelectrostatic feed-
back to thegate that works to suppress thetunnel leakage, as shown in Figure 3.31.
148 Wide band gap transistors SiC and GaN physics, design and models
n
tun
(t )
NTA
N
d
Electrostatic feedback
Surface charge
Surface conduction
AIGaN
Figure 3.34 Gateelectrontunnel leakageandsurfaceconductionmodel [79, 83].
(a)
6.40E-02
6.30E-02
6.20E-02
6.10E-02
6.00E-02
5.90E-02
0 200 400 600
Time (s)
NTA = 2 10
11
cm
2
I
d
s

(
A
)
NTA = 5 10
11
cm
2
800 1000
Figure 3.35a SimulatedDCchannel current versustimefor twovaluesof theacceptor-likesurface
trap(NTA) density[79, 83].
Thisintroducestimedependencetothegateleakage, withacorrespondingtimedepen-
denceassociatedwiththeRF power degradation. This effect canbemodeledby intro-
ductionof asurfaceconductionlayer thatpermitsavariablesurfacecharge, asshownin
Figure3.34. TheNTA termrepresentstheacceptor-likesurfacetrapdensity, andcanbe
expressedas
NTA = N
d
n
tun
(t). (3.30)
whereN
d
isthesurfaceconductionlayerchargedensity, andn
tun
(t) isthetime-dependent
tunnel chargedensity. Inthismodel, theelectronsthattunnel andaccumulatenexttothe
gateeffectivelyreducethedensityof thesurfaceconductionelectronsinthisregionand
permit avaryingsurfacetrapdensitynext tothegatetobedetermined.
The model shown in Figure 3.34 reproduces the drain and gate currents observed
inmeasureddata. Measuredandsimulatedtime-dependent DC drainandgatecurrents
areshowninFigure3.35aand3.35b, respectively, for two values of theacceptor-like
surfacetrapdensity. Notethat astheelectronsaccumulateonthesurfacenear thegate,
themagnitudeof thetunnel leakagecurrent isaffectedandthegateanddraincurrents
becometime-dependent. AstheNTA densityvaries, thedegreeof electrostaticfeedback
is affected, with corresponding effects upon the gate leakage current, and the drain
current degradation.
3.7 Large-signal effects 149
(b)
2.28E-05
2.32E-05
2.36E-05
2.40E-05
2.44E-05
2.48E-05
2.52E-05
2.56E-05
2.60E-05
0 200 400 600
Time (s)
NTA = 5 10
11
cm
2
I
g

(
A
)
NTA = 2 10
11
cm
2
800 1000
Figure 3.35b SimulatedDC gatecurrent versustimefor twovaluesof theacceptor-likesurface
trap(NTA) density[79, 83].
Themagnitudeof theelectriceldat thegateedgeisafunctionof thedevicedesign
andthemagnitudeof theterminal voltagesexperiencedbythedevicewhileinoperation.
A reductionintheelectriceldwill reducethegateleakagecurrent. AlGaN/GaNHFETs
produceaveryhigh-magnitudeelectriceldatthegateedgeduetothehighsheetcarrier
concentrationinthe2DEG. Very lowchannel resistanceresults, andminimal potential
dropoccursalongthechannel regionfromthedraintothegateuntil thegatedepletion
regionisencountered. Essentiallytheentiredrainpotential issupportedover thenarrow
depletionregionandaveryhigh-peakEeldresults. Techniquestoreducethemagnitude
of theelectric eld at thegateedgeincludetheuseof eld-plates, n-doped GaN cap
layers, controlled polarization-induced surfacecharges [81], and modications of the
2DEGsheet-chargedensity.
Two main current paths for gateleakagecurrents can beidentied. Themain path
is established by electron tunnel leakage fromthegate, with electrons owing along
or near theAlGaN surfaceto thedrain contact. Theelectron conduction occurs by a
trap-to-trap hopping mechanism, where both thermionic emission and tunneling are
likely involved, as illustrated in Figure3.36. Simulations indicatethat it is likely that
the exact conduction mechanismchanges as the electric eld increases due to high
DC and RF terminal voltages. This performance degradation process is essentially
reversibleandnondestructive, andremoval of thebias anddrivesignals, withaperiod
of deviceinactivity, causes thedeviceto return to its initial state. However, as previ-
ouslyindicated, reapplicationof DCandRF voltagesoftenresultinthedevicereturning
to adegraded state, which indicates that somepermanent damagehas occurred. The
second current path consists of electron tunneling fromthe gate, with electron ow
throughtheAlGaN layer to the2DEG conductingchannel. This current pathrequires
ahigher electric eld, andoftenproduces permanent damageto theAlGaN semicon-
ductor lattice, with increased gate leakage. The lattice damage is observed in TEM
images.
150 Wide band gap transistors SiC and GaN physics, design and models
AIGaN Surface Gate Metal
Strained Energy
Band
Electrons can
accumulate creating
virtual gate
Electron tunneling
parameters N
ss
, M
*
tun
V
dg
= V
V
dg
= 0
Thermionic Emission
Tunnel Emission
Surface hopping
parameters G, s
G
E
F
E
s
Figure 3.36 Detailedmodel for gate-tunnel leakageandsurfacetrap-to-traphopping
conduction[79].
(a)
V
gs
= 5 V
V
gs
= 2.5 V
V
gs
= 0 V
Solid Lines: Measured Data
Points: Simulation Data
0.2
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
0 1 2 3 4 5 6 7 8 9 10
V
ds
(V)
I
d
s

(
A
/
m
m
)
Figure 3.37a MeasuredandsimulatedDC IVcharacteristicsfor anAlGaN/GaN HFET usingthe
gate-tunnel leakageandsurfaceconductionmodel [79, 83].
Usingthegatetunnel leakageandsurfaceconductionmodel it ispossibletosimulate
the drain and gate current characteristics with excellent accuracy in comparison to
measureddata[83, 84]. For example, themodel showninFigure3.34producesthedrain
andgateIVcharacteristicsshowninFigure3.37. Figure3.37aandFigure3.37bshow
themeasuredandsimulateddraincurrentandgatecurrentfor aAlGaN/GaNHFET. The
gatetunnel leakageandsurfaceconductionmodel accurately predictsthegateleakage
andsurfaceconductioncurrent andaccuratelysimulatesboththedrainandgatecurrent
3.7 Large-signal effects 151
(b)
V
gs
= 7 V
V
gs
= 5 V
V
gs
= 3 V
V
gs
=7 V (measured)
V
gs
=5 V (measured)
V
gs
=3 V (measured)
V
gs
=1 V (measured)
V
gs
=7 V (simulated)
V
gs
=5 V (simulated)
V
gs
=3 V (simulated)
V
gs
=1 V (simulated)
V
gs
= 1 V
0.00E+00
2.00E-05
4.00E-05
6.00E-05
8.00E-05
1.00E-04
1.20E-04
0.00E+00 2.00E+00 4.00E+00
6.00E+00 8.00E+00 1.00E+00
V
ds
(V)
I
g

(
A
)
Figure 3.37b MeasuredandsimulatedDC gatecurrent characteristicsfor anAlGaN/GaN HFET
usingthegate-tunnel leakageandsurfaceconductionmodel [79, 83].
(a)
0
47.5
47
48.5
48
46.5
46
45.5
44.5
45
100 200 300 400 500
Time (s)
I
d


(
m
/
A
)
Figure 3.38a MeasuredandsimulatedDC draincurrent versustimefor anAlGaN/GaN HFET
includingtheeffectsof gatetunnel leakageandsurfaceconduction(pointsaremeasureddata
andthelineissimulateddata) [79, 83].
characteristics. Themodel canbeextendedtotime-dependent conditions, as shownin
Figure3.38[83, 84]. Figures3.38aandFigure3.38bshowthemeasuredandsimulated
time-dependent DC gateanddraincurrents, respectively.
Thesimulationsareperformedwithamodel that includestheeffectsof electrostatic
feedback fromtheelectrons that tunnel to thesurfaceof theAlGaN layer adjacent to
thegateelectrode. Thesimulationresultsarecomparedtoexperimental dataandexcel-
lent agreement betweenthemeasuredandsimulateddataisobtained. Theelectrostatic
feedbackreducestheelectriceldattheedgeof thegateelectrode, therebyreducingthe
electrontunnel leakage. Aselectronsaccumulateat thegateedgeasafunctionof stress
time, thefeedback producesreducedgateleakagecurrent. Also, theincreasedelectron
152 Wide band gap transistors SiC and GaN physics, design and models
(b)
0
0.001
0.002
0.003
0.004
0.005
0.006
0.007
0 100 200 300 400 500
Time (s)
I
g

(
m
/
A
)
Figure 3.38b MeasuredandsimulatedDC gatecurrent versustimefor anAlGaN/GaN HFET
includingtheeffectsof gatetunnel leakageandsurfaceconduction(pointsaremeasureddata
andthelineissimulateddata) [79, 83].
densityontheAlGaNsurfacepartiallydepletesthe2DEGelectrons, andareductionin
gatecurrent occurs.
3.8 Summary
Widebandgap semiconductors, that is SiC and nitride-based heterostructures, can be
used to fabricate high-frequency transistors with RF power performance superior to
thosefabricatedfromGaAsor Si. Themost promisingRF devicesareFETsfabricated
from4H-SiC andHFETs fabricatedfromtheAlGaN/GaN heterostructure. Optimized
4H-SiC FETs can produce RF output power on the order of 45 W/mm, which is a
factor of four greater than obtainablefromGaAs devices. Ampliers fabricated from
4H-SiC MESFETs will be useful, particularly for RF applications in S and C-band
communications, andpotentially for X-Bandradars. TheAlGaN/GaN HFET canpro-
duceRF power density ontheorder of 1012W/mm, withvery goodPAE. Prototype
nitride-basedHFETshaveproducedaspotRFoutputpowerdensityashighas30W/mm,
althoughthisrequireddrainbiasof V
ds
=120V. Thehighmobilityandsheet-chargeden-
sityof theAlGaN/GaN heterostructurepermit thefabricationof HFETswithexcellent
high-frequencyperformance, anddevicesthat canoperateuptoandpotentiallyexceed
100 GHz havebeen demonstrated. For both 4H-SiC and AlGaN/GaN HFETs power-
added efciencies approach theideal for operation up to X-band in both Class A and
B operation. For X-band and belowtheSiC and AlGaN/GaN transistors arecompeti-
tivewith each other, and both produceRF output power superior to GaAs-based and
InP-based transistors, while providing equivalent gain and PAE. Above X-band the
AlGaN/GaN HFETs will dominate. However, improved thermal design is required to
obtain the theoretically predicted performance, particularly for AlGaN/GaN HFETs,
which are generally fabricated frommaterial grown on SiC substrates. Both 4H-SiC
and AlGaN/GaN devices are likely to nd application in power ampliers for base
References 153
stationtransmitters for wireless communications, HDTV transmitters, power modules
for phased-array radars, andother applications. Thedevices areparticularly attractive
for applications that requirehighRF output power andoperationat elevatedtempera-
ture. Thesesolid statedevices should providean alternativeto theuseof microwave
vacuumtubesinmanytransmitter applications. Thewidebandgapsemiconductor tran-
sistors, duetotheir inherently highinput andoutput impedances, areattractivefor use
inpower-combining, broadbandwidth, andphased-arrayradar applications.
References
1. B. E. Kruger, Efcient widebandhighpower generationfor X- andKu-bandradars, Pro-
ceedingsof the1995IEEE International Radar Conference, 1995, pp. 227232.
2. M. Kumar, M. Hanczor, H. Voigt, G. Cambigians, R. Sachs, and C. Bonilla, 22 kW next
generationlowcost S-bandsolid-statetransmitter for surveillanceandtrafccontrol radars,
IEEE Int. Microw. Symp. Dig., pp. 16011604, 1995.
3. T. Marae, K. Fujii, andT. Matsuno, Highpower S-bandsolid-stateampliersfor survelleil-
lance and trafc control radars, 2001 IEEE International Microwave SymposiumDigest,
pp. 653656.
4. M. Cicolani, Highpower modular S-bandsolidstatetransmittersfamilyfor ATC andnaval
radar applications, IEEE Int. Microw. Symp., pp. 17231726, 2000.
5. R. J. Trew, SiC and GaN transistors: is there one winner for microwave power applica-
tions,Proceedingsof theIEEE, Special IssueonWideBandgapSemiconductors, J une2002,
vol. 90. pp. 10321047.
6. B. J. Baliga, SiliconCarbidePower Devices, WorldScientic, Singapore, 2005.
7. R. J. Trew, J. B. Yan, and P. M. Mock (invited), Thepotential of diamond and SiC elec-
tronicdevicesfor microwaveandmillimeter-wavepower applications, Proc. IEEE, vol. 79,
pp. 598620, May1991.
8. F. Zhao, I. Perez, C-F. Huang, J. Torvik, andB. VanZeghbroeck, Analysis of transit times
and minority carrier mobility in npn 4H-SiC bipolar junction transistors, IEEE. Trans.
ElectronDev., vol. 52, pp. 25412545, Dec. 2005.
9. F. Zhao, I. Perez-Wur, C-F. Huang, J. Torvik, andB. VanZeghbroeck, First demonstration
of 4H-SiCRF bipolarjunctiontransistorsonasemi-insulatingsubstratewithfT/fmaxof 7/5.2
GHz, IEEE Int. Microw. Symp. Dig., 2005.
10. C-F. HuangandJ. A. Cooper, Highcurrent gain4H-SiC NPN bipolar junctiontransistors,
IEEE ElectronDev. Lett., vol. 24, pp. 396398, J une2003.
11. A. Galeckas, J. Linnros, M. Frischholz, K. Rottner, N. Nordell, S. Karlsson, and V. Griv-
ickas,Investigationof surfacerecombinationandcarrier lifetimesin4H/6H-SiC, Mat. Sci.
Eng., vol. B6162, pp. 239243, 1999.
12. T. Kimoto, N. Miyamoto, andH. Matsunami, Performancelimitingsurfacedefects inSiC
epitaxial p-njunctiondiodes, IEEE Trans. ElectronDev., vol. 46, pp. 471477, Mar. 1999.
13. J. Zhang, P. Alexandrov, T. Burke, andJ. H. Zhao, 4h-sic power bipolar junctiontransistor
withavery lowspecic ON-resistanceof 2.9mW-cm2, IEEE Electron. Dev. Lett., vol. 27,
pp. 368370, May2006.
14. J. Zhang, J. H. Zhao, P. Alexandrov, andT. Burke, Demonstrationof rst 9.2KV 4H-SiC
bipolar junctiontransistor, IEE ElectronLett., vol. 40, pp. 13811382, Oct. 2004.
154 Wide band gap transistors SiC and GaN physics, design and models
15. C-F. Huang, I. Perez, F. Zhao, J. Torvik, R. Irwin, K. Torvik, F. Abrhaley, andB. VanZegh-
broeck, 215WpulsedclassA UHF power amplicationbasedonSiC bipolar technology,
DeviceRes. Conf. Dig., pp. 23, J une2123, 2004.
16. C. D. Brandt, R. C. Clarke, R. R. Siergiej, J. B. Casady, S. Sriram, andA. K. Agarwal, SiC
for applications inhigh-power electronics, Chapter 5inY. S. Park, Ed., Sic Materials and
Devices, SemiconductorsandSemimetals, vol. 52, AcademicPress, 1998.
17. R. R. Siergiej, R. C. Clarke, A. K. Agarwal, C. D. Brandt, A. A. Burke, A. Morse, and
P. A. Orphanos, Highpower 4H-SiC staticinductiontransistors, IEDMDig., pp. 353356,
WashingtonDC, Dec. 1995.
18. R. C. Clarke, A. K. Agarwal, R. R. Siergiej, C. D. Brandt, and A. W. Morse, Themixed
mode4H-SiC SIT as anS-bandmicrowavepower transistor, DeviceResearchConf. Dig.,
pp. 6263, SantaBarbara, CA, J une1996.
19. A. W. Morse, P. M. Esker, R. C. Clarke, C. D. Brandt, R. R. Siergiej, and A. K. Agarwal,
Applicationof highpowersiliconcarbidetransistorsatradarfrequencies,1996IEEEMTT-S
Dig., pp. 677680, SanFrancisco, CA.
20. R. C. Clarke, A. W. Morse, P. Esker, andW. R. Curtice, A 16W, 40%efcient, continuous
wave4HSiC L-BandSIT, Int. Microw. Symp. Dig., pp. 141143, 2000.
21. G. C. DeSalvo, P. M. Esker, T. A. Flint, J. A. Ostop, E. J. Stewart, T. J. Knight, K. J. Petrosky,
S. D. Van Campen, R. C. Clarke, and G. M. Bates, Ion implanted SiC static induction
transistor with107Woutput power and59%power-addedefciencyunder CWoperationat
750MHz, Int. J. HighSpeedElectronicsandSyst., vol. 14, no. 3pp. 906908, 2004.
22. R. J. Trew, SiC microwavedevices, Chapter 6inSiC MaterialsandDevices, pp. 272279,
Y. S. Park, Ed., SemiconductorsandSemimetals, vol. 52, AcademicPress, 1998.
23. S. Sriram, R. Barron, A. W, Morse, T. J. Smith, G. Augustine, A.A. BurkJ r, R. C. Clarke, R.
C. Glass, H. M. Hobgood, P. A. Orphanos, R. R. Siergiej, C. D. Brandt, M. C. Driver, andR.
H. Hopkins, Highefciency operationof 6H-SiC MESFETs at 6GHz, DeviceResearch
Conf. Dig., pp. 104105, 1995.
24. C. Weitzel, J. W. Palmour, C. H. Carter, and K. J. Nordquist, 4H-SiC MESFET with 2.8
W/mmpower densityat1.8GHz,IEEEElectronDev. Lett., vol. 15, pp. 406407, Oct. 1994.
25. K. E. Moore, C. E. Weitzel, K. J. Nordquist, L. L. Pond, J. W. Palmour, S. Allen, andC. H.
Carter, 4h-sicmesfet with65.7%power-addedefciencyat 850MHz, IEEE ElectronDev.
Lett., vol. 18, pp. 6970, Feb. 1997.
26. S. Sriram, G. Augustine, A. A. Burk, R. C. Glass, H. M. Hobgood, P. A. Orphanos, L. B.
Rowland, T. J. Smith, C. Brandt, M. C. Driver, andR. H. Hopkins, 4H-SiC MESFETswith
42GHz fmax, IEEE ElectronDev. Lett., vol. 17, pp. 369371, J uly1996.
27. A. W. Morse, P. M. Esker, S. Sriram, J. J. Hawkins, L. S. Chen, J. A. Ostop, T. J. Smith,
C. D. Davis, R. R. Barron, R. C. Clarke, R. R. Siergiej, andC. D. Brandt, Recentapplication
of siliconcarbidetohighpower microwave,IEEE Int. Microw. Symp. Dig., pp. 5356, 1997.
28. R. A. Sadler, S. T. Allen, T. S. Alcorn, W. L. Pribble, J. Sumakeris, and J. W. Palmour,
SiC MESFET withoutput power of 50WattsCWat S-Band, DeviceResearchConf. Dig.,
pp. 9293, 1998.
29. R. A. Sadler, S. T. Allen, W. L. Pribble, T. SAlcorn, J. J. Sumakeris, andJ. W. Palmour, SiC
MESFET hybrid amplier with 30 W output power at 10 GHz, IEEE Int. Microw. Symp.
Dig., pp. 173177, 2000
30. B. Luo, P. Chen. A. Higgins, H. Finlay, K. Boutros, B. Pierce, A.J ones, D. Griffey, and
J. Kolosick, 56 W SiC MESFET transistors with >50% PAE for L-band applications,
Proceedings of the17th International Symosiumon Power Semiconductor Devices & ICs,
SantaBarbara, CA, May2326, 2005, pp. 13.
References 155
31. H. Henry, G. Augustine, G. DeSalvo, R.C. Brooks, J. Oliver, A. Morse, B. Veasel, P. Esker,
and R. Clarke, S-band operation of SiC power MESFET with 20 W (4.4 W/mm) output
power and60%PAE, IEEE Trans. ElectronDev., vol. 51, pp. 839845, J une2004.
32. A. Asano, Y. Miyoshi, K. Ishikura, Y. Nashimoto, M. Kuzuhara, andM. Mizuta, Novel high
power AlGaAs/GaAs HFET with a eld-modulating plate operated at 35v drain voltage,
IEDMDig., pp. 5962, 1998.
33. K. Andersson, M. Sudow, P-A. Nilsson, E. Sveinbjornsson, H. Hjelmgren, J. Nilsson, J. Stahl,
H. Zirath, andN. Rorsman, Fabricationandcharacterizationof eld-plateburied-gateSiC
MESFETs, IEEE ElectronDev. Lett., vol. 27, pp. 573575, J uly2006.
34. A. SayedandG. Boeck, Two-stageultrawide-band5WpoweramplierusingSiCMESFET,
IEEE Trans. Microw. TheoryTech., vol. 53, pp. 24412449, J uly2005.
35. W. M. Zhou, F. Fang, Z. Y. Hou, L. J. Yan, andY. F. Zhang, Field-effect transistor basedon
b-SiC nanowire, IEEE ElectronDev. Lett., vol. 27, pp. 463465, J une2006.
36. Y. F. Wu, B. P. Keller, P. Fini, S. Keller, T. J. J enkins, L. T. Kehias, S. P. Denbaars, and
U. K. Mishra, HighAl-content AlGaN/GaN MODFETsfor ultrahighperformance, IEEE
ElectronDev. Lett., pp. 5053, Feb. 1998.
37. A. T. Ping, Q. Chen, J. W. Yang, M. A. Khan, andI. Adesida, DCandmicrowaveperformance
of high-current AlGaN/GaN heterostructure eld effect transistors grown on p-Type SiC
substrates, IEEE ElectronDev. Lett., pp. 5456, Feb. 1998.
38. G. J. Sullivan, M. Y. Chen, J. A. Higgins, J. W. Yang, Q. Chen, R. L. Pierson, and B. T.
McDermott, High power 10 GHz operation of AlGaN HFETs on insulating SiC, IEEE
ElectronDev. Lett., vol. 19, pp. 198200, J une1998.
39. S. T. Sheppard, K. Doverspike, W. L. Pribble, S. T. Allen, J. W. Palmour, L. T. Kehia, and
T. J. J enkins, High-powermicrowaveGaN/AlGaNHEMTsonsemi-insulatingsiliconcarbide
substrates, IEEE ElectronDev. Lett., vol. 20, pp. 161163, April 1999.
40. Q. Chen, J. W. Yang, R. Gaska, M. A. Khan, M. S. Shur, G. J. Sullivan, A. L. Sailor, J. A.
Higgings, A. T. Ping, andI. Adesida, High-power0.25-mmgatedoped-channel GaN/AlGaN
heterostructure eld effect transistor, IEEE Electron Dev. Lett., vol. 19, no. 2, pp. 4446,
Feb. 1998.
41. L. F. Eastman, V. Tilak, J. Smart, B. M. Green, E. M. Chumbes, R. Dimitrov, K. Hyungtak,
O. S. Ambacher, N. Weimann, T. Prunty, M. Murphy, W. J. Schaff, andJ. R. Shealy, Undoped
AlGaN/GaN HEMTs for microwave power amplication, IEEE Trans. Electron Dev.,
vol. 48, pp. 479485, Mar. 2001.
42. Y. F. Wu, D. Kapolnek, J. P. Ibbetson, P. Parikh, B. Keller, andU. K. Mishra, Very-highpower
densityAlGaN/GaNHEMTs, IEEE Trans. ElectronDev., vol. 48, pp. 586590, Mar. 2001.
43. V. Tilak, B. Green, V. Kaper, H. Kim, T. Prunty, J. Smart, J. Shealy, andL. Eastman, Inuence
of barrier thicknessonthehigh-power performanceof AlGaN/GaNHEMTs,IEEE Electron
Dev. Lett., vol. 22, pp. 504506, Nov. 2001.
44. L. Shen, S. Heikman, B. Moran, R. Cofe, N-Q. Zhang, D. Buttari, I. P. Smorchkova,
S. Keller, S. P. DenBaars, and U. K. Mishra, AlGaN/AlN/GaN high-power microwave
HEMT, IEEE ElectronDev. Lett., vol. 22, pp. 457459, Oct. 2001.
45. Y-F. Yu, A. Saxler, M. Moore, R. P. Smith, S. Sheppard, P. M. Chavarkar, T. Wisleder, U. K.
Mishra, andP. Parikh, 30-W/mmGaN HEMTsbyeldplateoptimization, IEEE Electron
Dev. Lett., vol. 25, pp. 117119, Mar. 2004.
46. A. Corrion, C. Poblenz, P. Waltereit, T. Palacios, S. Rajan, U.K. Mishra, and J. S.
Speck,Review of recent developments in growth of AlGaN/GaN high-electron mobility
transistors on4H-SiC by plasma-assistedmolecular beamepitaxy, IEICE Trans. Electron-
ics, vol. E89-C, no. 7, pp. 906912, 2006.
156 Wide band gap transistors SiC and GaN physics, design and models
47. J. W. J ohnson, E. L. Piner, A. Vescan, R. Therrien, P. Rajagopal, J. C. Roberts, J. D. Brown,
S. Singhal, andK. J. Linthicum, 12W/mmAlGaN-GaNHFETsonsiliconsubstrates,IEEE
ElectronDev. Lett., vol. 25, pp. 459461. J uly2004.
48. D. C. Dumka, C. Lee, H. Q. Tserng, P. Saunier, andM. Kumar, AlGaN/GaN HEMTsonSi
substrates with7W/mmoutput power density at 10GHz, Electron. Lett., vol. 40, no. 16,
Aug. 2004.
49. D. Ducatteau, A. Minko, V. Hoel, E. Morvan, E. Delos, B. Grimbert, H. Lahreche, P. Bove,
C. Gaquiere, J. C. DeJ aeger, andS. Delage, Output power densityof 5.1W/mmat 18GHz
withanAlGaN/GaNHEMT osSi substrate, IEEE ElectronDev. Lett., vol. 27, pp. 79, J an.
2006.
50. C. Lee, H. Wang, J. Yang, L. Witkowski, M. Muir, M. A. Khan, andP. Saunier, State-of-art
CW power density achieved at 26 GHz by AlGaN/GaN HEMTs, Electron. Lett., vol. 38,
pp. 924925, Aug. 2002.
51. C. Lee, P. Saunier, J. Yang, andM. A. Khan, AlGaN-GaN HEMTsonSiC withCWpower
performance>4W/mmand23%PAEat35%,IEEEElectronDev. Lett., vol. 24, pp. 616618,
Oct. 2003.
52. K. Boutros, M. Regan, P. Rowell, D. Gotthold, R. Birkhahn, andB. Brar, Highperformance
GaN HEMTsat 40GHz withpower densityof 2.8W/mm, IEDMTech. Dig., pp. 981982,
2003.
53. J. S. Moon, S. Wu, D. Wong, I. Milosavljevic, A. Conway, P. Hashimoto, M. Hu, M. Antcliffe,
andM. Micovic, Gate-recessedAlGaN-GaNHEMTsfor highperformancemillimeter-wave
applications, IEEE ElectronDev. Lett., vol. 26, pp. 348350, J une2005.
54. T. Palacios, A. Chakroborty, S. Rajan, C. Poblenz, S. Keller, S. P. DenBaars, J. S. Speck, and
U. K. Mishra, High-power AlGaN/GaN HEMTsfor Ka-bandapplications, IEEE Electron
Dev. Lett., vol. 26, pp. 781783, Nov. 2005.
55. Y. Cai, Y. Zhou, K. J. Chen, and K. M. Lau, High-performance enhancement-mode
AlGaN/GaN HEMTs using uoride-based plasma treatment, IEEE Electron Dev. Lett.,
vol. 26, pp. 435437, J uly2005.
56. L. Shen, T. Palacios, C. Poblenz, A. Corrion, A. Chakraborty, N. Fichtenbaum, S. Keller,
S. P. DenBaars, J. S. Speck, and U. K. Mishra, Unpassivated high power deeply recessed
GaN HEMTs with uorine-plasma surface treatment, IEEE Electron Dev. Lett., vol. 27,
pp. 214216, April 2006.
57. O. Katz, D. Mistele, B. Meyler, G. Bahir, and J. Salzman, Polarization engineering of
InAlN/GaN HFET andtheeffect onDC andRF performance, IEDMTech. Dig., pp. 1035
1038, 2004.
58. O. Katz, D. Mistele, B. Meyler, G. Bahir, andJ. Salzman, Characteristicsof InAlN-GaNhigh-
electron mobility eld-effect transistor, IEEE Trans. Electron Dev., vol. 52, pp. 146150,
Feb. 2005.
59. T. Palacios, A. Chakraborty, S. Heikman, S. Keller, S. P. DenBaars, and U. K. Mishra,
AlGaN/GaN highelectronmobility transistors withInGaN back-barriers, IEEE Electron
Dev. Lett., vol. 27, pp. 1315, J an. 2006.
60. J. Liu, Y. Zhou, J. Zhu, K. M. Lau, andK. J. Chen, AlGaN/GaN/InGaN/GaN DH-HEMTs
withanInGaN notchfor enhancedcarrier connement, IEEE ElectronDev. Lett., vol. 27,
pp. 1012, J an. 2006.
61. M. Higashiwaki, T. Mimura, andT. Matsui, AlN/GaNinsulated-gateHFETsusingCat-CVD
SiN, IEEE ElectronDev. Lett., vol. 27, pp. 719721, Sept. 2006.
References 157
62. Y. Ando, Y. Okamoto, H. Miyamoto, T. Nakamura, T. Inoue, andM. Kuzuhara, 10-W/mm
AlGaN-GaN HFET with a eld modulating plate, IEEE Electron Dev. Lett., vol. 24,
pp. 289291, May2003.
63. R. Vetury, Y. Wei, D. S. Green, S. R. Gibb, T. W. Mercier, K. Leverich, P. M. Garber,
M. J. Poulton, J. B. Shealy, Highpower, highefciency, AlGaN/GaNHEMT technologyfor
wirelessbasestationapplications, IMSDig., pp. 487490, 2005.
64. Y. Kamoetal., A C-bandAlGaN/GaNHEMT withCat-CVDSiNpassivationdevelopedfor
anover 100Woperation, IEEE IMSTech. Dig., pp. 495498, 2005.
65. T. Kikkawa, et al., Anover 200Woutput power GaN HEMT push-pull amplier withhigh
reliability, IEEE IMSTech. Dig., pp. 13471350, 2004.
66. W. Nagy, S. Singhal, R. Borges, J. W. J ohnson, J. D. Brown, R. Therrien, A. Chaudhari,
A. W. Hanson, J. Riddle, S. Booth, P. Rajagopal, E. L. Piner, andK. J. Linthicum, 150W
GaN-on-Si RF power transistor, IEEE IMSTech. Dig., pp. 483486, 2005.
67. R. Therrien, S. Singhal, J. W J ohnson, W. Nagy, R. Borges, A. Chaudhari, A. W. Hanson,
A. Edwards, J. Marquart, P. Rajagopal, C. Park, I. C. Kizilyalli, K. J. Linthicum, A 36mm
GaN-on-Si HFET producing368W at 60V with70%drainefciency, IEEE IEDM Tech.
Dig., 2005.
68. O. Kruger, G. Schone, T. Wernicke, R. Lossy, A. Liero, F. Schnieder, J. Wur, andG. Trankle,
Laser-assistedprocessingof VIAsforAlGaN/GaNHEMTsonSiCsubstrates,IEEEElectron
Dev. Lett., vol. 27, pp. 425427, J une2006.
69. K. Krishnamurthy, J. Martin, B. Landbert, R. Vetury, andM. J. Poulton, Wideband400W
pulsedpower GaN HEMT ampliers, IEEE CSIC Symp. Dig., pp. 303306, Monterey, CA,
Oct. 1215, 2008.
70. S. Piotrowicz, E. Morvan, R. Aubry, S. Bansropun, T. Bouvet, E. Chartier, T. Dean, O. Drisse,
C. Dua, D Floriot, M.A. diFortePoisson, Y. Gourdel, A. J. Hydes, J .C. J acquet, O. J ardel,
D. Lancereau, J. O. McLean, G. Lecoustre, A. Martin, Z. Quarch, T. Reveyrand, M. Richard,
N. Sarazin, D. Thenot, andS. L. Delage, Stateof theart58W, 38%PAEX-BandAlGaN/GaN
HEMTsmicrostripMMIC ampliers, IEEE CSIC Symp. Dig., pp. 14, Monterey, CA, Oct.
1215, 2008.
71. M. Micovic, A. Kurdoghian, H. P. Moyer, P. Hasimoto, M. Hu, M. Antcliffe, P. J. Willadsen,
W. S. Wong, R. Bowen, I. Milosavljevic, Y. Yoon, A. Schmitz, M. Wetzel, C. McGruire,
B. Hughes, andD. H. Chow, GaN MMIC PAsfor E-Band(71GHz-95GHz) radio, IEEE
CSIC Symp. Dig., pp. 14, Monterey, CA, Oct. 1215, 2008.
72. R. J. Trew, Widebandgapsemiconductor transistorsfor microwavepower ampliers,IEEE
Microw. Mag. vol. 1, pp. 4654, March2000.
73. R. J. Trew, Highfrequencysolidstateelectronicdevices,IEEETrans. ElectronDev., Special
IssueonVacuumElectronicDevices, pp. 638649, May, 2005.
74. T. A. Winslow, R. J. Trew, P. Gilmore, andC. T. Kelley, Simulatedperformanceoptimization
of GaAsMESFETampliers,ProceedingoftheThirteenthBiennial ConferenceonAdvanced
Concepts in High Speed Semiconductor Devices and Circuits, Ithaca, NY, Aug. 1991,
pp. 393402.
75. M. Micovic, A. Kurdoghlian, H. P. Moyer, P. Hashimoto, M. Hu, M. Antcliffe, P. J.
Willadsen, W. S. Wong, R. Bowen, I. Milosavljevic, Y. Yoon, A. Schmitz, M. Wetzel,
C. McGuire, B. Hughes, andD. H. Chow, GaN MMIC PAs for E-band(71GHz95GHz)
radio, IEEE Compound Semiconductor Integrated Circuits Symposium(CSICS), pp. 14,
Oct. 2008.
158 Wide band gap transistors SiC and GaN physics, design and models
76. M. A. KhatibzadehandR. J. Trew, A large-signal, analyticmodel for theGaAsMESFET,
IEEE Trans. Microw. TheoryTech., vol. 36, pp. 231238, Feb. 1988.
77. R. J. Trew, Y. Liu, G. L. Bilbro, W. W. Kuang, R. Vetury, andJ. B. Shealy, Nonlinear source
resistance in high voltage microwave AlGaN/GaN HFETs, IEEE Trans. Microw. Theory
Tech., vol. 54, pp. 20612067, May2006.
78. G. L. Bilbro and R. J. Trew, RF knee walkout and source access region of unpassivated
HFETs, ElectronicsLett., vol. 42, pp. 14251426, Nov. 2006.
79. R. J. Trew, D. S. GreenandJ. BShealy, AlGaN/GaNHFET reliability,IEEEMicrow. Mag.,
vol. 10, pp. 116127, J une2009.
80. R. J. Trew, Y. Liu, W. W. Kuang, andG. L. Bilbro (invited), Thephysics of reliability for
AlGaN/GaN HFETs, Compound Semiconductor Integrated Circuits Symp. (CSICS) Dig.,
SanAntonio, TX, Nov. 1315, 2006.
81. Y. Inoue, et al., Degradation-modeanalysis for highly reliableGaN-HEMTs, 2007IEEE
IMSDig., pp. 639642. W. Kuang, R. J. Trew, andG. L. Bilbro, Modelingof surfacedefect
relatedgateleakageinAlGaN/GaNHFET,MaterialsResearchSociety(MR)SpringMeeting,
SanFrancisco, CA, April 913, 2007.
82. T. A. WinslowandR. J. Trew, Principlesof large-signal MESFET Operation, IEEE Trans.
MicrowaveTheoryTech., vol. 42, pp. 935942, J une1994.
83. W. Kuang, R. J. Trew, and G L. Bilbro, An analytical model for surfaceleakagecurrents
of AlGaN/GaN HFETs and effects upon device reliability, WOCSDICE, IMEC, Leuven,
Belgium, May1821, 2008.
84. R. J. Trew, Y. Liu, W. Kuang, andG. L. Bilbro(invited), Reliabilitymodelingof high-voltage
AlGaN/GaN andGaAseld-effect transistors, Proc. of SPIE, vol. 6894, 1H17, 2008.
4 Amplier classes, A to S
1
Steve Cripps
Cardiff University
4.1 Introduction
Thealphabetical classicationof electronicampliersappearstodatebacktotheearliest
era of electronics, and as such could well be approaching the centenary mark. Its
survival to thepresent day represents aremarkablecontinuity, given thevast changes
in technology that have taken place in the intervening decades. It can also represent
a distraction for the modern RFPA designer working with solid state active devices
and GHz frequencies, both of which were well below the horizon when the original
classicationcameintogeneral use. Theplaninthischapter istointroduceanddene
the various Classes,
1
and then consider how the original intent is often modied in
typical modernapplications, sometimestothepointwheretheoriginal conceptmigrates
intosomethingpalpablydifferent.
Although thedenitions of Class A, AB, B, and C arewell established and havea
longhistorical precedent, thesubsequentClasses(D, E, F, etc.) areof muchmorerecent
originandinsomecaseshavesufferedfromdifferentinterpretationsbydifferentauthors.
A curious but endemic featureof this subject is theassertiveuseof classications by
authorsanddesignerswhenthenal amplier current andvoltagewaveformshavenot
been(andinmany cases cannot easily be) measureddirectly. This has beenknownto
lead to considerablecontroversy, given that someamplier Classes haveeven been
patented. Another issuewhich comes up when addressing this subject with amodern
perspectiveistheintrusionof digital approachestopower amplication, primarilyinthe
formof so-calledswitchmodes.Asactivedevicetechnologyimproves, thefrequency
rangeat whichit canbemadetobehaveasanear-ideal switchincreases. Until recently,
this range could be reasonably restricted to the HF (MHz to tens of MHz) region,
but newer technologiessuchasgalliumarsenideandgalliumnitridehaveextendedthis
region into theVHF (hundreds of MHz), and morearguably into themicrowave
(GHz) region. Thereisthussomethingof agrayarea, whereaparticular amplier can
beconsideredasasmoothed-outswitchmode, or alternativelyasamoreconventional
analoguePA classwithsomeextraharmoniccomponents. Bothof theseapproacheswill
bedescribedinthelater sectionsonClassS, E, andJ amplier.
1
The word Class has such specic and specialised importance in this chapter that I will capitalize it
throughout.
160 Amplier classes, A to S
Efciency is the central issue in the evolution of amplier classes. The Class A
amplier can beconsidered to bealogical starting point, and will bedescribed rst.
It has several positive attributes, notably simplicity of implementation, linearity, and
highpotential operatingbandwidth. Itdoeshowever haveonlymoderateefciency, even
at maximumsignal drive conditions, and this is what led early amplier developers
to explore Class AB modes, where the device is deliberately shut off for a portion
of each RF signal cycle. This can have a dramatic effect in terms of increased ef-
ciency, but almost alwayscomesat apriceof reducedoverall linearity. Asweprogress
throughthealphabet inthePA idiom, this tradeoff betweenefciency andlinearity in
general continues. A highly efcient Class C amplier, for example, cannot be con-
sidered at all in applications having any formof amplitudemodulation on thesignal
carrier.
Sincethestart of thedigital communicationsera, therehasbeenamarkedemphasis
onlinear RF power amplication, that istosaythesystemRFPA isrequiredtoamplify
the signal in its nal, fully modulated form. This represented something of a sea-
change for the RFPA industry, since in older systems it was more common to use
the RFPA itself as a high-level modulator. Vacuumtubes appear to have been very
amenable to the use of supply voltage variation as a means of imposing amplitude
modulation(AM) ontothecarrier. AssiliconRF power transistorsbegantoappear in
theearly1960s, it seemsthat amajor re-thinktookplace, sincethesedevicesdisplayed
highly nonlinear behavior under conditions of supply modulation. Conveniently (and,
presumably, not serendipitously) there was a shift to angle modulation (frequency
andphasemodulation), especiallyinmobiletransmitters, wherebythetransmitter RFPA
couldberunatconstantamplitudelevel. Thisallowedtheongoinguseof efcientClass
C-type amplier designs and was something of a disincentive for device technology
development towards more linear performance. But the more recent development of
systems usingcomplex, digitally based, modulationschemes has forcedRFPA design,
andtheunderlyingdevicetechnology, tocomplywithstringent linearityspecications.
There is nevertheless a counter-culture evident in the modern RFPA community,
whichseeks to reinstatetheoldregime. Inprinciple, adigitally modulatedsignal can
still be sent using only nonlinear RF amplication. The basic concept is to generate
a constant envelope signal which carries the appropriate phase modulation, and then
use supply modulation (and/or alternatives) to generate the required AM. The non-
linear relationship between thesupply voltageand theoutput RF envelopeamplitude
canbemanagedinamodernsystemthroughtheuseof digital correctiontechniques,
either onthetrackingvoltageitself, thesignal envelope, or both. SuchLINC (linear
amplication using non-linear components) RFPA systems can in principle be much
moreefcient than thelinear approach, although theefciency at which thetracking
voltagesupply can begeneratedis anegativefactor whichmust always betaken into
account. LINC systemdesignhasthusrenewedinterest inthedesignof highlyefcient
amplierswithout theconstraint of linearity. Theclassical solutionof ClassC modeis
less attractivein solid statedesign, and this is theareain which switching, or quasi-
switchingRFPAssuchasClassesDandE maywell haveamajor roletoplayinfuture
systems.
4.2 Active device models 161
Despitethesepossiblefuturedirections, thevast majority of RFPAs in current use
have been designed for Class A (Section 4.3) or Class AB (Section 4.4) operation.
Class A tends to predominate at higher GHz frequencies where applications usually
demandhighlylinear performance, andRF bandwidthscanbe10%or greater. Wireless
communicationssystems,whichtypicallyusemuchnarrowerbandwidths,usuallyfavour
ClassABoperationinordertomaximizeefciency. Linearityrequirementsare, however,
rapidlyapproachingthoseencounteredinsatellitecommunicationsandmicrowavelink
applications, anddigital signal processingusually hastobeemployedinorder tomeet
linearityspecicationsinClassAB operation. WenoteClassB (Section4.4) somewhat
inpassing, asasingular point that dividesClassAB fromClassC (Section4.5).
ClassF (Section4.6) issomethingof thejoker inthepack, inat least onesense, but
hasbeenthesubjectof muchresearchoverthelastdecadeorso. Itcanbeconsideredasa
derivativeof ClassABoperation, andhasbeenusedinbothlinearandLINCapplications.
Inalinear application, it caninprincipleincreasetheefciency at peak power levels,
withoutcompromisinglinearity, butinpracticeposessomedifcultcircuitdesignissues,
especially for higher power devices at GHz frequencies. In LINC applications it can
challengetheefciencyof themorefashionableswitchmodes, andinspeciccasesmay
supplyahigher output power duetobetter control of peakvoltage.
Other PA classes come and go according to the whims of researchers and patent
attorneys, so as weproceed beyond Class F then denitions becomealittleharder to
nd, let alonetosummarize. TherearealsosomediscrepanciesbetweenaudioandRF
electronics in the denition of some amplier classes. The most notable of these is
theClass D audio amplier whichis basically apulsewidthmodulator andis usually
denoted as Class S at RF and Microwavefrequencies (hencethereis no section in
thischapter onClassD). Also, intheaudioworldClassGandClassHarewell dened
but theseterms havenot comeinto general useat RF. Ironically, theaudio denitions
essentiallyutilizeatechniqueknownasenvelopetracking, or ET, whichisusedatRF
but has never beenclassiedalphabetically. Section4.11attempts tosummarizesome
of themiscellaneous categories.
4.2 Active device models
IndeningandanalyzingRFPA Classesit islogical toadopt acommondevicemodel.
Attheoutset, however, ithastobenotedthatRFPA devicesfall intotwodistinctphysical
kingdoms, thebipolar transistor andtheFET. Withinthesetwokingdomslienumerous
genera; for example the most widely used bipolar device at GHz frequencies is the
HeterojunctionBipolar Transistor, or HBT, but at higher power levelsandlower (UHF)
frequencies, the more traditional Si bipolar junction transistor (BJ T) still survives.
FET devices come in a somewhat greater diversity, both in terms of materials and
structures. Galliumarsenide metal semiconductor FETs (GaAs FETs) dominated the
GHz sector for several decades, somewhere between the late 1960s and the 1990s.
Moreadvanced material growing machinery such as molecular beamepitaxy (MBE)
becamecommercially availableintheearly 1990sandledtothedevelopment of more
162 Amplier classes, A to S
0
I
max
V
max
V
max knee
) (<<V
I
ds
V
gs
steps) (linear
I
ds
V
gs
D
S
G
Figure 4.1 Ideal devicemodel usedfor PA analysis.
optimumFET structures such as the high-electron mobility transistor (HEMT) and
the pseudomorphic HEMT (pHEMT), but these were primarily still based on GaAs
substrates.Duringthe2000s,galliumnitridehasemergedasanenfantterrible,sporting
high-voltageoperationandpotentiallybroader bandoperationthanGaAsdevices, albeit
withnospecicinherent frequencyadvantage.
Thesevariousdevicesandtechnologiesaredescribedinearlier chaptersof thisbook,
but for thepresent purposesjust about anyof theFET deviceswill displayaset of IV
characteristicsasshowninFigure4.1.
Thesecharacteristicshavebeenidealizedfor thepurposesof maintainingafocuson
themodeof operation. Theidealityassumptionscanbesummarizedasfollows:
r
constant current-sinkbehavior outsidetheturn-on, or knee region;
r
abrupt cut-off of current whenthegatevoltagedropsbelowathreshold value;
r
saturation of thecurrent aboveadenedvalue, usuallydenotedbyI
max
;
r
linear relationship between output current and input voltagebetween thethreshold
andsaturationpoints(inthischapter thiswill becalledthequasi-linear region);
r
quasi-staticbehavior (sameIV characteristicsregardlessof sweepspeed).
In dealing with PA modes, an additional idealization is often deployed, which is to
assumethat thekneeregionhasanegligibleimpact, andthat theturn-oncharacteristic
of thedevicecanbeignored. This assumptionis almost endemic amongPA theorists,
andcanoftenrepresent themainunderlyingcausefor discrepanciesbetweenmeasured
resultsandtheoretical performancepredictions. For thepurposesof thischapter wewill,
however, complywiththemainstreamviewwhichistoset V
k
= 0.
4.3 Class A
In Class A operation thedeviceis kept entirely within thequasi-linear region. For
maximumpower performance, thedeviceis supplied with astanding bias current of
I
max
,2 and theinput signal voltageis constrained to swing between thelimits of the
quasi-linear range. Figure 4.2 shows the current and voltage waveforms for an ideal
devicewithaninput sinusoidal signal excitation.
4.3 Class A 163
rad.) t, (=
V
ds
V
dc
I
dc
I
ds
I
max
2V
dc
2 3 4
Figure 4.2 ClassA devicewaveforms.
Thesinusoidal waveformsmakethecalculationof output power andefciency very
straightforward. TheRF output power, P
rf
, isgivenby
P
rf
=
I
max
2

2
V
dc

2
=
I
dc
V
dc
2
(4.1)
andtheDC power suppliedis
P
dc
= I
dc
V
dc
. (4.2)
Theoutput efciency isdenedas

o
=
P
rf
P
dc
(4.3)
sothat inthiscaseweobtaintheclassical result that theoutput efciency of aClassA
amplier is
1
2
, or 50%.
Wenote, however, that thismuch-quotedresult will not applyinpracticefor anyreal
device, dueprimarilytothezero-knee assumption. Theeffect of thekneevoltagecan
bemost simply expressed and quantied by assuming that thevoltageswing will be
maintained such that theminimado not encroach into thekneeregion, that is where
V
ds
- V
k
. SotheRF output power under maximumdriveconditionscanberewrittenas
P
rf
=
V
dc
V
k

2
.
I
dc

2
=
(V
dc
V
k
)
2
.I
dc
(4.4)
andthecorrespondingefciencybecomes

o
=
V
dc
V
k
2V
dc
=
1
2
_
1
V
k
V
dc
_
. (4.5)
TheratioV
k
,V
dc
isbothtechnologyandapplicationdependent. If wesurveytherangeof
semiconductortechnologiesincurrentuseatGHzfrequencies, theratioisapproximately
0.1inmostcases, butthisassumesthatthedeviceisbeingoperatedatitsmaximumrated
164 Amplier classes, A to S
DCsupplyvoltage. Soinmost practical casestheClassA efciencycanbeexpectedto
benohigher thanabout 45%.
TheaboveanalysisappliesonlytoaCWinputsignal whichhasthenecessarymagni-
tudetodrivethedeviceintoamaximumcurrentswing, thatisoverthefull quasi-linear
rangefromzero to I
max
. In order to assess theefciency for an amplitudemodulated
signal, it is necessary to obtain an expression for the efciency under conditions of
power back-off (PBO). This is an easy calculation to performin theClass A case,
sincetheDCbiasremainsconstant, thatisindependentof theinputdriveconditions. So
for abacked-off condition, wheretheRF output isP
bo
, theefciencywill be

bo
=
P
bo
P
dc
(4.6)
whichcanbeexpressedintermsof theRF output under full driveconditions, P
max
, as

pbo
=
1
2
.
P
rf
P
max
(4.7)
So the efciency of a Class A amplier backs off in direct proportion to the output
power. For example, at the6dB back-off point, theefciencyisonequarter of thepeak
power efciency. For amodulatedsignal that hasapeaktoaveragepower ratioof 6dB,
anaverageefciencyinthe2025%rangeisthebest that canbeexpectedfromaClass
A amplier. This is an unacceptably low efciency for many applications, and is the
mainreasonClassA isnot muchusedinwirelesscommunicationssystems.
TheClassA modedoeshowever havesomeadvantages. Itslinearityisusuallygood,
due to the fact that the device is kept entirely within the quasi-linear range, where
the nonlinearities are of the weak variety. The power gain is typically several dBs
higher for a given device operating in Class A than in the more popular Class AB
modesconsideredinthenext section. For thisreason, ClassA operationbecomesmore
widespread at higher frequencies, wheretheavailabledevices deliver less than about
10dBof gaininClassA. ClassA ampliersarealsofairlyeasytodesign, inthattheydo
not requirespecic harmonic, aswell as fundamental, matching. For thisreasonClass
A ispreferredfor broadband(greater thanoctave) power ampliersatGHzfrequencies,
althoughat subGHz frequenciesthepushpull classB conguration[1] iswidelyused
for multioctavebandwidths.
4.4 Class AB and Class B
Theuseof areducedconductionangle inthedesignof RF power ampliers is well
known, andalsodates totheearliest eraof electronics. As suchwewill not engagein
lengthypreliminariesbut consider thedevicewaveformsshowninFigure4.3.
ThekeydifferenceinmovingfromClassA toClassABoperationisthatthequiescent
biascurrent ischangedtoalower relativevalue, oftenaslowasabout 10%of theI
max
for thedevice. For aFET typedevicethiscanbeeasilyimplementedbymovingthegate
bias voltagecloser to thethreshold level. An RF signal input can thus still swing the
devicecurrent uptotheI
max
value, asinClassA operation, but duetothesymmetryof
4.4 Class AB and Class B 165
I
dc
I
ds
I
max
2
2
2
3
3
3
4
4
4
(a)
(b)
(c)
V
ds
V
ds
V
dc
V
dc
2V
dc
2V
dc
Figure 4.3 Reducedconductionangle(ClassAB) waveforms; (a) current, (b) output voltagewith
broadbandresistiveoutput termination, (c) voltagewithshort-circuitedharmonictermination.
asinusoidal excitationthenegative-goingpartof thevoltagecyclewill swingthedevice
gatevoltagebelowits thresholdvalue, thus cuttingoff theconductionfor aportion
of theRF cycle. Theresultingcurrent waveformisshowninFigure4.3a, andisusually
describedasatruncated sinewave. It isthemathematical propertiesof suchtruncated
sinewaves that determine the main performance benets of Class AB operation, but
before considering this in more detail we need to consider what now happens to the
deviceoutput voltage.
Dueto thefact that an RF transistor can beconveniently approximated as an ideal
current sink, theoutput voltagecanbeeasily calculated; it isasimplematter of multi-
plyingeachindividual currentharmonicwiththecorrespondingoutputloadimpedance,
or inmoresymboliclanguage,
V
ds
=

n
I
n
.Z
n
(4.8)
where I
n
represents the harmonic components of current and Z
n
the load impedance
valueat thecorrespondingharmonicfrequencies.
Figure4.3bshowsthevoltagewaveformthat wouldresultfromabroadbandresistive
termination, andittakestheformof aninvertedreplicaof thecurrentwaveform. Sucha
result isunlikelyinpracticeat GHz frequencies, wherethecharacteristicsof theoutput
166 Amplier classes, A to S
0
Fundamental
DC
2nd
3rd
4th
5th
0.5
0
Amplitude
(I
max
=1)
2 Conduction
angle
C B A AB (CLASS)

Figure 4.4 Harmoniccomponentsof reducedconductionanglewaveforms.


matchingnetworkasafunctionof frequencywill showlargevariationsinbothresistive
andreactivecomponents. It is thus animportant stipulationinthedesignof Class AB
ampliers that the harmonic impedances are made as close to zero as possible. This
harmonic short isnever ideally achievedandisoftentheunderlyingcauseof RFPAs
performinglesswell thananticipated. However, for thepresentpurposeswewill assume
that theoutput loaddoes present aperfect short circuit to thedevice, so theharmonic
components of thecurrent donot generateany correspondingharmonic content inthe
output voltage, in which casetheoutput voltagewill then besinusoidal, as shown in
Figure4.3c.
Sotheoutput voltageof anideal ClassAB amplier looksthesameasfor theClass
A case, and it is to the current that we look for the differences. Figure 4.4 plots the
DC andfundamental componentsof atruncatedcosinewave. Astheconductionangle
isreduced, theDC component dropsbut thefundamental RF component remainsvery
nearly constant, for conduction angles greater than 180

. This results directly in an


efciency increase, whichis plottedinFigure4.5. But it must beemphasizedthat the
efciencyplot makesseveral assumptions, viz.
r
zerokneevoltage;
r
short circuitedharmonics, resultinginasinusoidal deviceoutput voltage;
r
maximumvoltage(V
DC
amplitude) andcurrent (I
max
,2Amp) swings.
A caseof particular signicance, if maybenotactuallywidelyused, isthezero-biasor
ClassB condition. Inthiscasethedeviceisbiasedpreciselytoitsthresholdpoint and
hencedrawsnocurrent until someinput signal isapplied. Whenthedriveissufcient
tocausethedevicecurrent toswingtoitsmaximumextent, thecurrent waveformwill
4.4 Class AB and Class B 167
0 2
Conduction
angle
0
+5 dB 100%
5 dB 0%
(dB)
Efficiency
(dB)
I
Q
=10%
Figure 4.5 Power andefciencyof fullydrivenreducedconductionanglePA.
(=
I
dc
I
ds
I
max

2
2
3
3
4
4
(a)
(b)
V
ds
V
dc
2V
dc
t, rad.)
Figure 4.6 ClassB devicewaveforms.
becomeahalf-waverectiedsinewave, asshowninFigure4.6. TheDCandfundamental
componentsintheClassB conditionhaveasimpleclosedform,
I
dc
=
I
max

.
I
1
=
I
max
2
.
(4.9)
168 Amplier classes, A to S
sothat theRF andDC power isthesameasfor theClassA condition,
P
rf
=
V
dc
I
max
4
P
dc
=
V
dc
I
max

. (4.10)
Theoutput efciencyisnowgivenby

o
=

4
(4.11)
or about 78.5%.
As withtheClass A analysis, theallowancefor anonzero kneevoltagewill reduce
thisclassical number by anywherebetween5and10%, dependingonthedevicebeing
used.
Thezerobiasconditionisauseful datumpoint, butisnotoftenused. Thisisduetothe
fact that azero-biased devicewill haveveryquirkyperformanceat lowsignal levels,
whereareal devicewill not display anideal cutoff behavior. Inpractice, theefciency
of adeep Class AB amplier, biasedaroundthe10%level, will showanefciency
quiteclosetotheclassical ClassB value, asshowninFigure4.5. Nevertheless, taking
account of kneeeffectsaswell, it isunrealistictoexpect anefciencyhigher than65%
inapractical case, whilemaintainingthedeviceentirelywithinthequasi-linear region.
Thisisnottosaythathigherefcienciescannotbemeasured, andarefrequentlyreported
intheliterature, but suchresults areoftentakenwiththedevicedisplayingsignicant
gaincompression.
Asbefore, it isimportant toconsider thevariationof efciencyasafunctionof input
drivebackoff, and theClass B casewill again beconsidered dueto its mathematical
simplicity. For adeviceoperatinginClassB, andat abacked-off RF output level of P
rf
,
weassumethattheRF loadingisunchangedfromthemaximumpower condition, P
max
.
Sothebacked-off current andvoltageamplitudescanbeexpressedas
I
1
=
I
max
2
.

P
rf
P
max
(4.12)
V
1
= V
dc

P
rf
P
max
(4.13)
I
dc
=
I
max

P
rf
P
max
. (4.14)
Sofor aconstant supplyvoltage, thebacked-off efciencyis

pbo
=
P
rf
P
dc
=
V
dc
I
max
4
.
P
rf
P
max
.

V
dc
I
max

P
max
P
rf
=

4
.

P
rf
P
max
(4.15)
This is aresult of considerablesignicancefor amplitude-modulated signals, sinceit
shows that thePBO efciency is aslower function than in theClass A case, being
inversely proportional to thesquareroot of thePBO ratio. Inthecaseof a6dB PAR,
4.4 Class AB and Class B 169
0.5
0.2
I =0
Q
0.1
Efficiency
0
50%
100%
Output power (2 dB/div)
Figure 4.7 Efciencycharacteristicsfor ideal ClassAB PAsfor backed-off driveconditions.
theefciency at themeanpower level hasonly droppedby afactor of two, asopposed
toafactor of four intheClassA case.
The same analysis can be performed for intermediate Class AB cases, but dees
symbolictreatment; Figure4.7showstheresultingPBOefciencycharacteristicswhich
have been computed for a range of quiescent bias settings. It is clear that the same
conclusionconcerningthesuperior PBOefciency canbemadefor quiescent settings
uptoat least the10%level.
It hasalreadybeenemphasizedthat thedesirableefciencyimprovementsthat Class
ABoperationoffersdocomeataprice. Lowerpowergainandincreasedcircuitcomplex-
ity generally will limit therangeof applications to lower frequencies (whereavailable
devices havemoregainto spare), andnarrowbandwidths (dueto therequirement for
short-circuiting the harmonics). There is, however, another issue which concerns the
linearity of aClass AB amplier. This is something of acontroversial topic, because
devicemanufacturersoftentailor thefabricationprocesstooffer devicesthathavegood
linearityinClassAB, but oftenonlywhenoperatedat atightlyspeciedquiescent bias
setting.
Figure 4.8 does show, nevertheless, that an ideal device will display signicant
nonlinearitywhenoperatedinClassAB. Thisresultsfromthemathematical properties
of truncatedsinewaves; basically, at agivenquiescent biassettingtheconductionangle
isitself afunctionof drivelevel andanonlinear relationshipbetweenthedrivevoltage
andthefundamental currentcomponent results, despitethedeviceitself havinganideal
transconductivecharacteristic. As thequiescent setting approaches theClass B point,
the linearity tends asymptotically towards a linear characteristic, and at around the
10%level it becomes possibleto cancel thegain expansion by tailoring thedevice
transconductancecharacteristic. Whetherthisprocesshaseverbeenimplementedinsuch
ana-priori manner issomethingof amoot point, but sufceit tosay that suchdevices
usually needthequiescent bias point to beset withconsiderableprecisioninorder to
obtainthespeciedlinear performance. Assuch, thedesigner frequentlydoesnot have
170 Amplier classes, A to S
Input Power (2 dB/div)
Output
Power
(2 dB/div)
P
lin
V
q
=0.5
(Class A)
0.25
0.15
0.05
V
q
=0
V
q
=0.1
V
q
=0.25
V
q
=0.5
Figure 4.8 Linearityof ClassAB PA modes.
50
V
dc
bias
network
fundamental
match
input
match
harmonic
termination
Figure 4.9 Topologyof basicClassAB amplier.
afreechoiceontheconductionangle, andsomemanufacturers will evenrecommend
that higher quiescent settingsshouldnot beusedat all.
Figure4.9 shows schematically themain elements in atypical Class AB amplier
circuit at GHz frequencies. The output matching network has to performtwo main
functions, afundamental matchwhichtransformsthedeviceload-lineresistancetothe
systemimpedance level, and a harmonic trap which presents a short circuit at the
harmonic frequencies. Thefundamental matchingnetwork canbevery similar to that
usedfor aClassA amplier, sincetheoptimumfundamental loadwill beverycloseto
thedeviceload-lineresistance. Theharmonic trap can takevarious forms. A popular
textbook solution is to useashort-circuited quarter-wavestub, which thus presents
ashort circuit only at theevenharmonics andalso acts as aconvenient bias insertion
point. But thisisnot oftenusedinpractice, duetothelimitedbandwidthover whichan
acceptably lowimpedancecanbemaintained. A secondoptionistouseashunt series
resonator at the second harmonic. Fourier analysis of the half-wave rectied current
4.5 Class C 171
waveformrevealsthatthesecondharmonicisbyfar thelargestcomponent, andinmany
practical cases harmonics higher than the third can be regarded as trapped within
thedeviceitself, either through theaction of theoutput capacitance, or thelow-pass
characteristicof thedeviceitself.
By far themost common solution for terminating theoutput harmonics, albeit not
alwaysintentionally, istoutilizethedeviceoutputcapacitance. Itcanbeshown[2] thatif
thereactanceof theparasiticoutputcapacitanceisequal to, or lessthanthefundamental
load-lineresistance, theoutputcapacitor isbyitself abletosatisfytherequirementsof a
harmonicshort circuit. Thisapproximationtendstoholdasthefundamental frequency
increases, along with thedeviceperiphery. As aresult, what could bedescribed as a
formof complacencyseemstohavedevelopedamongstPA designersabove1GHz. Itis
found, for example, that thereisawiderangeof applicationswherethedeviceappears
to give satisfactory performance by suitably careful optimization of the fundamental
matching alone. This however does not in any way challenge the basic theory. It is
merely aconsequenceof thefact that availabletransistor technologieshavequitelarge
outputcapacitancevalues(intherangeof about1pF/W), whichalthoughbeingamajor
problemindesigningthefundamental matchover anyuseful bandwidth, justhappensto
solvetheharmoniccircuit problemveryconveniently. Thisconveniencedoes, however,
break down when using a given device at a much lower frequency, and/or when a
new technology comes along (such as galliumnitride) that has a much lower pF/W
characteristic.
The wireless communications industry has stimulated a vast amount of research
and development into the design of Class AB ampliers that give good linearity
and high efciency. These applications have however been focused in the lower fre-
quency strata of themicrowavespectrum, andutilizationof thesebenets becomes
rapidly more difcult above about 8 GHz, due mainly to the lower gain of available
devices.
4.5 Class C
TheClassCmodeisalogical extensionof thereducedconductionangleconcept, where
the conduction angle is reduced to less than half of the RF cycle. This results in a
current waveformthat looks morelikeastringof sharppulses, as indicatedinFigure
4.10. Referring back to Figure 4.6, the fundamental component starts to drop as the
conductionanglecrosses into Class C territory, but theDC component also continues
to drop and themathematics tells us that theefciency climbs ever upwards towards
100%for animpulsivecurrent, as showninFigure4.5. Thereare, however, anumber
of problemsthatneedtobeconsidered, whichtogether haverelegatedtheClassCmode
intoverylimitedpractical usefor certainspecializedapplications. Thiswasnotthecase
inthevacuumtubeera, whentheterminologywasintroduced.
ThereductioninRFpowerinClassCisaseriousproblemforasemiconductordevice,
sinceit meansthat toobtainagivenRF power thesize, or periphery, of thedevicehas
to beincreased. This was less of an issuein thetubeera, sincethehigher efciency
172 Amplier classes, A to S
(= t, rad.)
I
dc
I
ds
I
max

2
2
3
3
4
4
(a)
(b)
V
ds
V
dc
2V
dc
Figure 4.10 ClassC devicewaveforms.
V
T
V
MAX
Input
Voltage
Figure 4.11 Excessinput voltageinveryshort conductionangleClassC.
enabledthedevicetoberunatahigherplatevoltage, thuseffectivelyrestoringthepower
shortfall. Suchfreedominsupplyvoltageselectionisnotavailableinthesemiconductor
world, wheredevices areusually operated at their maximumspecied safeoperating
voltage.
A larger problemwithClassC isillustratedinFigure4.11. Biasingthegatebeyond
itsthresholdpointmeansthataverylargedrivesignal will berequiredinorder toswing
thecurrent up to V
max
, thegatevoltagerequired for thedeviceto drawits maximum
current I
max
. For asinusoidal signal, thismeansthat thenegative-goingpeakswill drop
down to alevel that may causesomebreakdown effects. In particular, given that the
inputvoltagehasaminimumthatcorrespondstothemaximumpeakswingof theoutput
voltage, it becomesall toolikelythat somedrain-gatereversebreakdownwill occur.
4.6 Class F 173
V
ds
2 3 4
V
d
V
max
Figure 4.12 Effect of addinganin-phasethirdharmoniccomponent tothevoltagewaveformof a
ClassB amplier.
A ClassCamplier alsoposesagreater challengeintermsof thenecessaryharmonic
termination. Therelativeharmoniclevelsescalatequiterapidlyastheconductionangle
isreducedbelowtheClassB value, andthesecondharmonicapproximation may no
longer beusefullyvalid.
Despitethesevariousdisadvantages, itisworthmentioningthattheClassCmodehas
foundanimportant nicheinrecent years, asbeingauseful meansof implementingthe
peaking stageof aDohertyPA.
4.6 Class F
TheClassFmodehasbeenthefocusof muchresearch[3,4].Inprincipleitoffersasimple
meansof boostingthepeak efciency of aregular ClassB or deepClassAB amplier
by more than 10%. This is achieved by allowing a third harmonic component in the
voltagewaveform, sothat theoutput voltagelooksmorelikeasquared-up sinewave.
Asapreliminary, it isthereforeimportant tounderstandtheunderlyingmathematicsof
sinewaveshavinganaddedthirdharmoniccomponent.
TheprocessisillustratedqualitativelyinFigure4.12. Theadditionof asmall antiphase
thirdharmoniccomponenttoanysinewaveclearlyreducesthepeak-to-peakswing, since
therelevant thirdharmonic peaks anddips arecoincident intimewithpeaks anddips
of thefundamental. As aresult, theentirewaveformcanbescaledup, givingahigher
fundamental component. This process clearly has a limit, that is to say there is an
optimumlevel of thirdharmonicthat resultsinamaximumincreaseinthefundamental
amplitude; beyondthispoint thetwinpeaks start toincreaseandthebenetsrapidly
fadeaway. Finding this specic condition is something of amathematical puzzlethat
hasintriguedseveral authorsover theyears. Rhodes[5] tackleditbyrecognizingthatthe
optimumconditionwas asingular point. Morerecently [6], thepresent author showed
that theproblemcouldbesolvedby factorizingthevoltageexpression, aformulation
that turnsout tohavesomewider implications. Thesewill bediscussedinalittlemore
detail inSection4.11, but thespecicsolutionfor theClassF caseisnowconsidered.
If thecurrent is assumed to bean ideal truncated cosinusoidal function, thecorre-
spondingClassF voltagewave canbeexpressedintheform
: () = V
dc
V
1
cos V
3
cos3 (4.16)
174 Amplier classes, A to S
whereV
DC
istheDC supply, andV
1
, V
3
arethefundamental andharmonicamplitudes.
If for conveniencewenormalizethevoltagestotheDC level, thisexpressionsimplies
to
: () = 1:
1
cos :
3
cos3 (4.17)
andweseek themaximumvalueof :
1
for which:() remainsgreater thanor equal to
zerofor all valuesof .
Noting that the optimum condition will include a zero-grazing double root of
:() = 0, equation(4.17) canbewrittenintheform
:() = (1 cos)
2
(1 cos) (4.18)
so that relationships between the , parameters can be established with :
1
and :
3
by expanding (4.18) and comparing coefcients of similar terms in (4.17), noting in
particular that thesecond harmonic termmust vanish. Furthermore, (4.18) will force
the :() = 0 condition so long as 0 - - 1. This process results in the following
relationships,
=

2
(4.19)
:
3
=

3
2
(4.20)
:
1
=
_
3
2
8

3
2
_
. (4.21)
The parameter ineffect controlsthelevel of thirdharmonic for aset of zero-grazing
waveformsdenedby(4.18). Wethusseekthevalueof thatgivesthemaximumvalue
of :
1
, throughtherelationshipin(4.21). Simpledifferentiationgivesthisvalueas
=
2

3
(4.22)
correspondingtoamaximumfundamental component of
:
1max
=
2

3
(4.23)
andacorrespondingthirdharmonicvoltage
:
3max
=
1
6
. (4.24)
So for theoptimumClass F case, thevoltageexpression (4.17) can bewritten in the
factorizedform
: () =
_
1

2
3
cos
_
2
_
1

2
6
cos
_
. (4.25)
aremarkableresult of fairlyrecent origin[6].
4.6 Class F 175
(=
I
dc
I
ds
I
max

2
2
3
3
4
4
(a)
(b)
V
ds
V
dc
2V
dc

Figure 4.13 Ideal optimumClassF devicewaveforms.


SincetheDC componentsareunalteredfromtheClassB case, theoptimumClassF
efciencywill be

4
.
2

3
, or about 90.7%.
Theresultingideal ClassFwaveformsareshowninFigure4.13; notethattheoptimum
solutionhasathirdharmonic component that isslightly higher thanthat whichgivesa
maximallyat response.
As always, this result is conspicuously ideal and in practice the knee voltage will
causesignicant degradation. Indeed, ironically, dueto thefact that theideal Class F
voltagespends amuch higher proportion of theRF cyclewithin thekneeregion, the
relativedegradationfromtheideal power andefciencywill besignicantlyhigher than
for asinusoidal voltage. Butcaveatsnotwithstanding, theClassF mathematicscertainly
danglesaveryjuicycarrotwhichmorethanonegenerationof RFPA designershasfound
tobeanirresistiblechallenge. AndcomingupwithsuitableClass F circuit topologies
is indeed quite a challenge. The output matching network now has to performthree
functions,
r
transformthefundamental (load-line) resistancetotheterminationimpedance;
r
present theappropriate resistiveterminationat thethirdharmonicfrequency;
r
short circuit theremainingharmonics, especiallythesecond.
Itisthesecondof theserequirementsthatisnew,andformulatingastrategytodeal with
it hasbeenthesubject of muchdiscussion. Theaboveanalysisgivesaclear quantitative
designgoal for therequiredthirdharmonic voltagecomponent, but transformingthis
intoacorrespondingdesignimpedancepresentssomedifculties. Theproblemcanbe
highlightedbyconsideringtheClassBcase, wheretheideal currentwaveformhasazero
thirdharmoniccomponent: what impedanceisnecessarytodevelopavoltageof V
dc
,6
for azerocurrent ow? Somefurther discussiononthisapparent paradoxcanbefound
inreference[2], but for thepresent purposes it will sufceto say that thedesigngoal
176 Amplier classes, A to S
50
/4,f
o
o/c,3f
o
Figure 4.14 PossibleClassF matchingtopology.
shouldbetopresentashighof aresistanceaspossibletothedeviceatthethirdharmonic,
andinparticular toensurethat all of thereactiveparasiticsareparallel-resonatedat the
device output. This can be a daunting task when dealing with larger devices, whose
output capacitancecanbetensor hundredsof pF, andasaminimumwill bebandwidth
limited.
Numerous circuit topologies have been devised for implementing Class F at GHz
frequencies. A somewhat intuitiveapproachis showninFigure4.14, whereaquarter-
wavestubisusedtoshortthesecond(andinprincipleall of thehigher evenharmonics),
andat thethirdharmonic thedeviceoutput capacitanceis resonatedout withanopen
circuit stub, whose length at the fundamental is low enough such that it only adds a
small extracapacitancethat canbeabsorbedinto thefundamental matchingstructure.
But therearefurther constraintsonmaintainingthehigh-impedanceenvironment at the
third harmonic, in particular thefundamental matching network must haveasuitably
low-passcharacteristic inorder not toload thethirdharmonic impedance. Thismay
requireamoreaggressivenetwork at thefundamental, including ahigh-Q resonator,
not only torealizethehighthirdharmonic impedance, but alsotoblock theextrathird
harmoniccomponentsfromreachingtheoutput.
SothemaindifcultyinimplementingClassF isbandwidth, andtoooftenresultsare
publishedthat represent spot-frequencydesigns.
4.7 Class J
TheClass J modehas afairly recent origin, beingessentially promoted, as opposedto
invented, bythepresent author [2].
Thebasicconcept istoengineer asomewhat similar trick tothat usedintheClassF
mode, but usingsecond, rather thanthethirdharmonic. Onceagain, it paystoexamine
themathematicsof theprocessrst. Figure4.15showswhathappenswhenanantiphased
secondharmoniccomponentisaddedtoacosinewave. Theresultingwaveformbecomes
asymmetrical about theDC level, withahigher peak andaatter minimum, whichis
nowraisedabovethezerolevel. Asaresult, it ispossibletoscaleupthewaveformsuch
that it againbecomeszero-grazingandthusthefundamental component issignicantly
increased. J ust as in theClass F case, it is necessary to determinetheoptimumlevel
4.7 Class J 177
2V
dc
(=t, rad.)
I
max
I
ds
V
ds
V
dc
I
dc
2 3 4
2 3 4
(a)
(b)
Figure 4.15 Effect of secondharmoniconvoltagewaveform.
of second(intheClass J case) harmonic inorder to maximizethefundamental, while
maintainingthenonzerocrossingcondition. Sometrigonometric manipulationscanbe
employedtoshowthat for anormalizedcosinusoidal voltagewave,
: () = 1cos (4.26)
themaximumsecondharmoniccomponentthatcanbeaddedhasanormalizedamplitude
of 1,2, sothat thewaveformbecomes
: () = 1

2cos
1
2
cos2 (4.27)
asshowninFigure4.15.
At rst sight, thefactor of

2onthefundamental wouldimplyanefciencyof
=

4
= 1.11. (4.28)
or 111%, assuming the current waveformwas that of a Class B half-wave rectied
sinewave.
Clearly, this is inadmissible, and the reason for this is that the second harmonic
componentsof voltageandcurrent arein-phase, implyingeither power absorptionor a
negativeresistiveload.
The Class J mode resolves this problemby shifting the entire voltage waveform,
relativeto thecurrent, by 45

at thefundamental. TheClass J voltagewaveformthus


becomes
: () = 1cos sin
1
2
sin2. (4.29)
whichimplies afundamental loadconsistingof theregular load-lineresistivecompo-
nent, but withanequal reactivecomponent. Thesecondharmonic loadis areactance,
178 Amplier classes, A to S
2V
dc
(=t, rad.)
I
max
I
ds
V
ds
V
dc
I
dc
2 3 4
2 3 4
(a)
(b)
Figure 4.16 ClassJ devicewaveforms.
50
+ jR
L
R
L
f
0
jX , 2f
0
,
Figure 4.17 ClassJ output matchingcircuit topology.
of comparable value to the load-line resistance. The efciency is exactly the same
as a Class B amplier, as indicated by the unity normalized amplitude of the fun-
damental cosine voltage component. The resulting Class J waveforms are shown in
Figure4.16.
The Class J mode has an important benet over its regular Class B or Class AB
counterparts, inthat thesecondharmonic does not requireashort-circuit termination.
Indeed, thecapacitivereactancethat is requiredto terminatethesecondharmonic can
insomecases beprovidedby theoutput capacitanceof thedevice. This has probably
causedwidespreaduseof ClassJ inafortuitousmanner. Thisisillustratedinatypical
Class J circuit conguration, shown in Figure 4.17. The output matching topology
consistsessentiallyof acapacitor, whichprovidesthesecondharmonictermination, and
alow-passnetworkfor matchingthefundamental. Dependingonthefrequencyandthe
devicetechnology in use, theparasitic output capacitancemay in itself bewithin the
rangedenedbytheClassJ designequations. Insuchcases, theuninformeddesigner
cantreat thewholedesignproblemas anexerciseinfundamental matching, andsome
judicious a posteriori tuningcanintroducethenecessary reactivecomponent into the
4.8 Inverted modes, inverted Class F 179
V
dc
(=t, rad.)
I
max
I
ds
V
ds
V
dc
I
dc
2 3 4
2 3 4
(a)
(b)
Figure 4.18 InvertedClassF modewaveforms(ideal); (a) current, (b) voltage.
fundamental impedance.Highervoltageharmonicscanusuallyberegardedasnegligible,
dueto thedecliningamplitudeof thecurrent components andtheeffect of theoutput
capacitor. Inpracticetherewill besomeinteractionbetweenthevalueof thecapacitor
at thesecond harmonic and theimpedanceof thefundamental network at thesecond
harmonic.
4.8 Inverted modes, inverted Class F
The modes which have so far been described can all be inverted, which means in
effect that thecurrent andvoltagewaveformsarereversed. So, for example, aninverted
ClassBmodeconsistsof adevicehavingasinusoidal currentwaveformandahalf-wave
rectiedvoltagewaveform. Intheideal case, thepowerandefciencywouldbethesame
at thepeakpower level, but thepower back-off efciencycharacteristicwouldreplicate
the Class A curve in Figure 4.7, and as such this mode is not much used. A more
interestingcaseistheinvertedClassF mode, whichhasreceivedconsiderableattention
intheliteratureover thelastfewyears. Thewaveforms, showninFigure4.18, showone
useful potential advantageof ClassF
1
. Duetothefact that theDC component of the
half-waverectiedvoltagesinewavehasavalueof V
pk
,, thefundamental component
can beincreased by afactor of ,2, assuming that thepeak voltageof V
DC
can be
safelyaccommodated. Thiscorrespondstoapotential power increaseof nearly2dB in
comparisontoaClassF conguration, andabout a2.5dB increaseincomparisontoa
ClassB PA usingthesamedeviceatthesamesupplyvoltage. Inpracticethisextrapeak
voltagemayexceedthebreakdownspecicationof thedevice, althoughtherearecases
wherethismaynot bealimitation.
180 Amplier classes, A to S
(=t, rad.)
I
max
I
ds
V
ds
V
pk
V
dc
I
dc
2 3 4
2 3 4
(a)
(b)
Figure 4.19 PhysicallyrealizableinvertedClassF waveforms; (a) current, (b) voltage.
The waveforms shown in Figure 4.18 are still highly idealized and unlikely to be
realizedinpractice. InvertedClassF isconventionallyengineeredbystartingoff with
thedevicebiased as for Class A operation. Thecurrent clipping can then berealized
by over-driving the device so that it saturates on the peaks and cuts off in the dips.
This, however, will result inamoreof amaximally at current waveform, asshownin
Figure4.19. Thisonlyreducestheefciencyverymarginally, butinpracticeabigger hit
will betakenwhentryingtoengineerthestipulatedvoltage. Theideal half-waverectied
sinusoidcontainsmultipleharmonics, andwill usuallybeapproximatedbyaddingjust
secondharmonic. AsshowninFigure4.19(andasanalyzedinSection4.7), thisallows
thefundamental component tobeincreasedby afactor of

2, somewhat lessthanthe
,2factor that wouldapplyfor theideal half-waverectiedsinewave.
Thepeak power andefciency advantages of Class F
1
aresomewhat temperedby
aClass A-type PBO efciency characteristic, dueto thehighquiescent bias setting
requiredtoengineer thesquared-upcurrentwaveform. Thereis, however, aninteresting
variant, showninFigure4.20. Herethecurrentwaveisahalf-waverectied(co)sinewave
whichhasclippedpeaks. Withjudiciousadjustment, thevoltagecanbeallowedtodip
intothekneeregion, thusclippingthecurrent suchthat it becomesanapproximationto
asquarewave. This clippingcanbeadjustedto null out thesecondharmonic current
component, sothat theopen-circuit impedanceterminationwill allowaviablemodeof
operationthathasanimprovedPBOefciencycharacteristic. Theclippingprocessdoes,
however, signicantly reducethefundamental current component, causingareduction
inRF output power. Theclippedconditionwill alsolikely result inawell-compressed
condition, sothat thisvariant maynot besuitablefor linear applicationsthat useampli-
tudemodulated signals. It appears that this variant has in thepast been observed and
giventhenameof ClassG [7], but thistermhasnot comeintogeneral use.
4.9 Class E 181
(=t, rad.)
I
max
I
ds
V
dsV
pk
V
dc
I
dc
2 3 4
2 3 4
(a)
(b)
Figure 4.20 Clipped variationoninvertedClassF.
V
dc
I
dc
v
c
(t )
C
P I()
Figure 4.21 BasicClassE circuit.
4.9 Class E
TheClass E modeis dened, fundamentally, as aswitching mode, wheretheactive
device characteristics that have been used thus far are replaced by a simple, perfect,
switch. This immediately raises several questions about the validity, and indeed the
relevance, of switchingmodesat GHz frequencies. But beforeconsideringtheseissues
anyfurther, wewill examinethesimplestandmostbasicformof ideal ClassEoperation.
Figure 4.21 shows the simplest possible circuit for Class E operation. The active
devicetakes theformof an ideal switch, which for thepurposes of this analysis will
be assumed to have negligible transition times and can be turned on or off at
discretionarytimeswithineachRF cycle. Theswitchisshuntedbyacapacitor, andthis
in turn is shunted by a series resonant circuit. If we assume that the switch is being
toggledperiodicallyat afrequencythat isclosetotheresonant frequencyof thecircuit,
182 Amplier classes, A to S
I
dc
I ( )
I
max
I
max
I
max
0
0
0
0
I
rf
(a)
(b)
(c)
(d)
Device
Current
V
pk
V
dc
0
0
2 4
Output Cap
Current
Device
voltage
Figure 4.22 Ideal ClassE waveforms.
therewill beasinusoidal current owingaroundtheresonant circuit loop. Figure4.22
showstheresultingwaveforms, includingthecurrentsintheresonant loop, theswitch,
andtheshunt capacitor. Theactionof theswitchistoforcetheresonant current either
into theswitch, when theswitch is closed, or theshunt capacitor, when theswitch is
open. Giventheinertia of theseriesresonator, thecirculatingcurrent cannot change
astheswitchistoggled.Thecapacitor thusendsupwithacurrentwaveformasshown
in Figure 4.22c, which can be integrated to show the voltage across it, as shown in
Figure4.22d.
Looking at the waveforms at the switch terminals, it is clear that there is no time
withintheRF cyclethat current andvoltagearenon-zero simultaneously. Thesystem
thusrepresents100%efcientconversionfromDCtoRF energy, andduetotheresonant
natureof thecircuit whichcontainstheRF load, theenergy will bemainly connedto
thefundamental frequency. However, thehighefciency is as muchaproperty of the
assumedideal natureof theswitch, asthemodeof operation. Thereisalsoanimportant
caveat in that thepeak voltageacross theswitch can beseveral times theDC supply
voltage, causingbreakdownissueswhentheswitchisreplacedbyatransistor.
4.10 Class S 183
Figure 4.23 ClassSamplier concept.
In practice, the process of making a transistor behave like a switch requires some
trickery, which involves the use of the knee region as well as the threshold of the
devicecharacteristics. Thisprocessusually involvessacricingasignicant portionof
thedevicepeak current capability sothat, aswithClassC operation, higher efciency
is obtained at theexpense of obtaining lower power; this can be as much as 23 dB
lower than normal Class AB operation for the same device with the same supply
voltage.
TheClassE PA hasbeenafavouritesubject intheliteraturefor nearlyfour decades,
attracting particular attention fromtheacademic community. Above1 GHz, many of
thesepapersandarticlescanbequestionedinthat they frequently donot showany RF
waveformsatthedeviceplane, andquoteefcienciesthataremuchlower thanwouldbe
expectedfromapureswitchingmode. Efcienciesaround90%havebeenreportedfor
Class E designs inthelowGHz region, but thedeviceis usually operatinginahighly
nonlinear condition. Suchresultshavelimited, albeitstill potentiallyuseful, applications
inmicrowavecommunications.
4.10 Class S
TheClassSmodeistheRF versionof apulsewidthmodulationtechnique, widelyused
at lower frequencies under thenameof Class D. Thebasic process is well known, and
is indicated in Figure4.23. Thesignal is sampled, and atrain of pulses is generated,
whoselengthisproportional totheinstantaneoussampledamplitude. If thispulsetrain
ispassedthroughalow-passlter, it isanelementaryresult of samplingtheorythat the
original signal will bereconstructed, hencethetermreconstructionlter.
RF designershavealwaysdreamed of thedaywhenRF signalscanbegeneratedin
thismanner, andat anygiventimethereisusuallyavociferousfactionwhichproclaims
that the day has come. There are, however, some hazards upon which the unwary
frequentlystumble. Historically, theobviousoutstandingproblemisthatof thenecessary
sampling rate. At audio frequencies, this can be made easily two or three orders of
magnitude higher than the sampled signal bandwidth without posing any particular
challenges onthespeedof readily availableelectronic components. But for asignal at
1GHz thisclearlyposesproblems.
184 Amplier classes, A to S
V
dc
Figure 4.24 ViableClassSamplier conguration.
V
dc
Figure 4.25 Bogus ClassSamplier conguration.
Thereisinfact another problemwhichisfrequentlyoverlooked. Inorder tomaintain
high efciency, thereconstruction process must not generateany signicant power at
any of the spectral frequency components that are caused by the sampling process.
Figure4.24showsonewayof achievingthis. Theactivedeviceswitchestheoutputlter
andloadbetweentwovoltagesources, whichcanbeconvenientlytakenaszero(ground)
andapositiveDC supply. Switchingbetweensuchstiff voltagesources ensures that
theonlycurrentwhichisallowedtoowthroughthelter intotheloadisattherequired
signal frequency band. All other frequency components are presented with the high
reactiveloadof thelter andassuchdonotcreateanypower. Unfortunately, asingleRF
power transistor cannot beusedtoimplement suchascheme, asshowninFigure4.25.
Thetransistor canbeswitchedon or off, but thisineffect meansthat thelter-load
combinationisbeingexcitedbyaswitchedcurrentsource. Itmightbethoughtsufcient
to accommodatethis changeby placing asuitablelter in shunt across thedevice, in
order toprovidealow-impedancepathfor theunwantedspectral components, but then
thevoltageof thedesiredsignal will appear across thedeviceterminals. As such, the
devicewill starttodissipateheatasthelevel of theoutputsignal isreduced, inmuchthe
samewayasaregular ClassA or ClassAB amplier.
Implementation of asuitableswitching conguration thus requires, as aminimum,
acomplementary pair of transistorswhichcanswitchtheloadbetweenthetwosupply
rails. Suchacongurationposesdifcultiesat GHz frequencies, but aslower parasitic
semiconductor technologiesappear thepossibilityof realizingaClassSamplier atlow
GHz frequenciesdoesincrease.
4.11 Multimodes
Theongoingandwidespreaduseof theClass categorizationof RFPAs is somewhat
puzzling. Theseclassicationscitespecicpropertiesof thedevicecurrent andvoltage
4.11 Multimodes 185

I
dc
I
ds
I
max

2
2
3
3
4
4
(a)
(b)
V
ds
V
dc
2V
dc
(=t, rad.)
Figure 4.26 Multimode voltagewaveformcontainingsecondandthirdharmonics.
waveforms, and at GHz frequencies these waveforms are very difcult to measure
directly. They can, of course, besimulated, but it then becomes ajudgment call as to
whether thewaveformscomplywiththeintendedClass.Infact, areal deviceinareal
circuitcanfrequentlydisplaywaveformsthatdonotfall easilyintoaparticular category.
TakeforexamplethecurrentandvoltagewaveformsshowninFigure4.26. Thecurrent
isaregular half-waverectiedsinewave, but thevoltagecontainsbothsecondandthird
harmonic components, andas suchdoes not fall under Class F, D, or J as describedin
thischapter. Thevoltagewavehastheform
V = V
dc
cos V
1Q
sin V
2Q
sin2 V
3Q
sin3. (4.30)
which has the same in-phase fundamental component as a regular Class A or Class
B sinusoidal voltage, but withsomeaddedharmonic components. Sincetheharmonic
components are in quadrature with the cosinusoidal current, they do not contribute
power. Theresult is that suchanamplier will showthesameefciency as aClass B
amplier, but theharmonicvoltagecomponentsimplythat thedeviceisnot terminated
withashort circuit at eachharmonic, andinthiscasetheharmonicterminationswill be
entirelyreactive.
Suchasituationisprobably very commoninpractice. Thetextbook stipulationof a
global harmonic short is often unlikely to befully implemented in apractical circuit,
andthis exampleis just onespecic caseof alargeandcontinuous multidimensional
terminationspace, whichimpliesacontinuumof harmonic matchingconditionsthat
yieldthesamefundamental power andefciency as aclassical Class B amplier. The
ClassJ voltagecondition, denedearlier intheform
V = 1cos sin (1,2)sin2 (4.31)
186 Amplier classes, A to S
is another specic exampleof this space. It was observed by thecurrent author [6]
that thiscanbewritteninafactorizedform,
V = (1cos)(1sin) (4.32)
:() = 1cos :
1q
sin
k=n

k=2
:
kq
sink; (4.33)
it iseasier toconstruct solutionsbymultiplyingnon-zero-crossingfactorssuchas
(1 cosk). (1 sink).
andpowersthereof.
Suchexpressions will havethesameRF power andefciency as theclassical Class
B so long as the in-phase fundamental coefcient remains at unity and the in-phase
harmonics arezero. It is alsopossibletogeneralizetheexpressionfurther andinclude
cosineharmoniccomponents.Althoughthisraisesthepossibilityof generatingunwanted
powerattheharmonicfrequencies, thefundamental termcanbeenhanced. Forexample,
theclassical ClassF modecanbeshowntofall intothenewtheoretical framework,
V =
_
1
2

3
cos
_
2
_
1
1

3
cos
_
(1 sin) (4.34)
with = 1, but themoregeneralized formulation reveals acontinuous set of modes
basedonClassF, but whichcontainadditional quadratureevenharmonics.
4.12 Conclusions
TheRFPA classesdescribedinthischapter formaframeworkaroundwhichmost prac-
tical designswill t. Eachmodehowever representsatleastsomedegreeof idealization,
bothinthecharacteristicsof theactivedevice, andalsothefundamental andharmonic
termination environment. Almost any practical RFPA which operates in thelowGHz
frequencyregionwill likelydisplaysomevariancefromthetraditional waveformClass
denitions. But for themost part, designerswhoareunabletomeasurethedeviceplane
RF waveforms still indulge in a mindset of blind faith that the complex interaction
betweendeviceandcircuit canbefullycharacterizedbyafewlettersof thealphabet.
References
1. J. L. B. Walker, Ed., High Power GaAs FET Ampliers, Norwood: Artech House, 1993,
pp. 1821.
2. S. C. Cripps, RF Power Ampliersfor WirelessCommunications, 2ndEdn., Norwood: Artech
House, 2006.
3. V. J. Tyler, A new high efciency high power amplier, Marconi Rev., vol. 21, 1958,
pp. 96109.
References 187
4. F. H. Raab, ClassF poweramplicationwithmaximallyatwaveforms, IEEE. Trans. Microw.
TheoryTech., vol. 45, no. 11, pp. 20072011, Nov. 1997.
5. J. D. Rhodes, Universalityinmaximumefciencylinear power ampliers,Int. J. Circ. Theor.
Appl., vol. 31, pp. 385405, 2003.
6. S. C. Cripps, P. J . Tasker, A. L. Clarke, J. Lees, andJ. Benedikt, Onthecontinuity of high
efciency modes in linear RF power ampliers, IEEE Microw. Components Lett., vol. 19,
no. 10, pp. 665667.
7. P. Colantonio, F. Giannini, G. Leuzzi, and E. Limiti, High efciency low-voltage power
amplier designbysecondharmonicmanipulation,Int. J. RF Microw. Computer-AidedEng.,
vol. 10, no. 1, pp. 1932, J an. 2000.
5 Computer-aided design of
power ampliers
Stephen Maas
AWR, Corporation
5.1 Introduction
In any book about power ampliers, it seems essential to discuss themost important
tool in their design, circuit-analysis software. The development of such software has
progressedfromimprovementsintheunderstandingof linear and, especially, nonlinear
circuit theory, as well as rapid improvements in computer and software technology
over the past 20 or 30 years. While we all know about these successes, there exists
athird dimension to thematurity of circuit-design software: our ability to createand
especially to maintain largesoftwaresystems to support aversatiledesign owfor a
widevarietyof RF/microwavecomponents. Inspiteof thesesuccesses, however, circuit-
analysis technology has not reached thepoint whereit is perfectly transparent; some
street wisdom onthepart of theuser is still required. By describingtheunderlying
technologyof thesesoftwaresystems, thischapter shouldimpart someof that wisdom.
5.2 Methods of analysis
5.2.1 Linear analysis
Linear analysisisanimportant part of anynonlinear circuit simulator; it isalsointrinsi-
callyuseful, asmanytypesof circuitarequitesatisfactorilytreatedaslinear. Earlylinear
circuit-analysissoftwaretreatedall elementsastwo-portsandconnectedtheminseries,
parallel, or cascade according to the structure of the circuit. Since most microwave
matchingcircuitscanbedescribedeasilythisway, it wasauseful waytocreateaset of
circuit equations.
Unfortunately, many kindsof circuit simply cannot bedescribedasinterconnections
of two-ports. In this case, a more general method is needed, usually resulting in the
creationof anadmittanceorother matrixdescribingthecircuit. Onesuchmethod, which
was used in early general-purpose circuit-analysis programs, was based on so-called
nodal incidence matrices [1]. This method was complicated to implement, so it was
soonsupplantedbynodal analysis.
Nodal analysis
Nodal analysis has a number of attractive features. The nodal matrix, an admittance
matrix of thecircuit at each frequency of interest, can becreated in afully mindless
5.2 Methods of analysis 189
I
1
I
2
V
2
V
1
n
1
n
2
Y
Figure 5.1 WhentheadmittanceY isconnectedbetweennodesn
1
andn
2
, it changesthecurrent in
eachnodeasshown.
manner, thusmakingit perfect for implementationbyamindlessmachine, acomputer.
Nodal analysis creates largesparsematrices (i.e., matrices that consist mostly of zero
entries) and thereforecan takeadvantageof modern numerical methods for handling
such matrices. Although nodal analysis is generally less efcient than analysis based
on cascaded two-ports, its versatility in handling a widevariety of circuit structures,
combinedwiththespeedof moderncomputersandsophisticationof numerical methods
for dealingwithmatrices, makeit thoroughlypractical evenwhenappliedtoverylarge
circuits.
Nodal methodsdohavesomedisadvantages. First, andperhapsmostobviously, many
kinds of circuit element andstructuredo not haveadmittancerepresentations. This is
particularlytroublesomewhencircuitsarepartitioned, anecessarystep, asweshall see,
formanykindsof nonlinearcircuitanalysis. Then, thecircuitmaybecomedisconnected,
causing thenodal admittancematrix to besingular. Similarly, DC analysis of circuits
havinginductors or transmissionlines oftenfails, as nodes becomeinterconnectedby
large, perhapsinnite, conductances. Methodshavebeendevelopedfor circumventing
suchproblems; thesewill bedescribedinduecourse.
Consider theadmittanceelement Y inFigure5.1connectedbetweennodes 1and2.
Whenweconnect it into thecircuit, it changes thetotal nodal current ineachnodeas
follows:
Y(V
1
V
2
) = LI
1
Y(V
2
V
1
) = LI
2
(5.1)
whereV
1
, V
2
arethevoltages at therespectivenodes, measuredbetweenthenodeand
somearbitrarygroundpoint. Thiscanbewritten
_
LI
1
LI
2
_
=
_
Y Y
Y Y
_ _
V
1
V
2
_
(5.2)
implyingthat thematrixin(5.2) issimplyaddedtotheadmittancematrixof thecircuit,
intheimpliedpositions; that is, Y is addedto the(1, 1) and(2, 2) positions andY to
the(2, 1) and(1, 2) positions. equation(5.2) issometimescalledastamp, implyingthat
addingelementstothecircuit matrixinvolvesnothingmorethanstamping thematrix
190 Computer-aided design of power ampliers
V
2
V
3
V
1
+

n
1
n
2
n
3
Figure 5.2 Inanindeniteadmittancematrix, all nodevoltagesarereferencedtoacommon
groundnode.
withapredeterminedpattern. Stampsformorecomplicatedelements, suchascontrolled
sourcesandother admittancematrices, canbegeneratedsimilarly.
The resulting matrix is called an indenite admittance matrix. It is an admittance
matrix inwhichthevoltagesrepresent nodevoltagesrelativetosomearbitrary ground
point; thesituationisillustratedinFigure5.2. Itisclear thatthenodevoltagescannotbe
determineduniquelyinsuchacircuit. For example, wecouldndaset of voltagesthat
satisfy, say, (5.2), addsomeparticular DCquantitytoeachof them(5V might benice),
andthecurrents remainunchanged. It is inevitableinsuchacasethat theadmittance
matrixissingular.
To removethesingularity, at least onenodein thematrix must havesomedened
voltage; inpractice, it isgrounded. If theindenitematrixisasshownbelow,

I
1
I
2
. . .
I
N

Y
11
Y
12
. . . Y
1N
Y
21
Y
22
. . . Y
2N
. . . . . . . . . . . .
Y
N1
Y
N2
. . . Y
NN

V
1
V
2
. . .
V
N

(5.3)
grounding node n simply involves setting V
n
to zero. Then the nth column can be
removed, asitselementsareall multipliedbyzero. Similarly, thecurrentI
n
isthenof no
interest, soitsrowcanbedeletedaswell. Theresultingmatrixisstill squarebut (unless
it hassomeother problem) nolonger singular.
Usually, not all of thenodevoltages areof interest. Only thevoltages at accessible
external nodesareof concern, and, inparticular, weoftenwanttocharacterizetheN-node
network by aP-port or P-nodeadmittancematrix. Thiscanbeaccomplishedasshown
inFigure5.3. Werst select thenodes that will becomeports or accessiblenodes and
sequentially exciteeach port/nodewith acurrent source. Wethen obtain thevoltages
at each port or node. The voltages, divided by the excitation current, are the values
in one column of the impedance matrix. When all the columns have been obtained,
theimpedancematrix canbeconvertedto anadmittancematrix, scatteringmatrix, or
whatever typeisdesired.
5.2 Methods of analysis 191
I
1
V
2
Z
21
= V
2
/I
1
Z
31
= V
3
/I
1
Z
11
= V
1
/I
1
V
3
V
1
+

n
1
n
2
n
3
Figure 5.3 TondaP-nodenodal admittancematrixfromtheN-nodeindenitematrix, theP
externallyaccessiblenodesareexcitedinturnbycurrent sources. Thenodevoltagesresulting
fromeachexcitationprovideasinglecolumnof theZmatrix. TheZmatrixport matricescanbe
foundinasimilar way; thevoltagesof interest arethenthosebetweenthenodesdeningthe
ports. TheZmatrixcannallybeconvertedtoanydesiredform.
Themost commonmethodfor factoringanodal matrixisLU decomposition[2] and
back substitution. Only a single factorization of the large nodal matrix is necessary,
andtheport or nodevoltages resultingfrommultiplecurrent vectors canbefoundby
back-substitution, amuchlesscostlyoperation.
This process is used most frequently for characterizing the linear subcircuit in a
harmonic-balanceanalysis(Section5.2.2). Asweshall see, disconnectingthenonlinear
circuit elements from the linear ones often leaves disconnected nodes, making the
nodal matrix singular. Thisproblemcanbecircumventedfairly easily inthefollowing
manner:
1. Beforetheindenitematrixiscreated, moderate-valueresistorsareconnectedacross
eachport of aP-port matrixor fromthenodetogroundinanodal matrix. Thevalue
of theresistanceshouldbeonthesameorder asresistancesinthecircuit; inmostRF
andmicrowaveapplications, 100O workswell.
2. ThematrixisreducedtoaP-nodeor P-port admittancematrixasdescribedabove.
3. The added resistance now appears along the main diagonal of the P-dimension
admittancematrix. Theresistanceisremovedsimplybysubtractingitsinversefrom
themain-diagonal terms.
Thisprocessalwaysworksforanodal matrix. Inaportmatrix, itistechnicallypossible
for it tofail, but it almost alwaysworksinordinarycircuits.
Modiedanalysis
Many types of element, such as voltage-controlled voltage sources (VCVS), do not
have an admittance representation. Others, such as ideal transformers, do not have a
Y- or Z-matrix representationat all. It is possibleto circumvent someof theselimita-
tions with other elements; for example, aVCVS can berealized fromacascadeof a
192 Computer-aided design of power ampliers
voltage-controlledcurrent sourceandagyrator; atransformer canalsoberealizedbya
cascadeof gyrators. Modiednodal analysisisasomewhat moreelegant alternative.
Suppose we have a VCVS whose control terminals are (j, k) and voltage source
terminalsare(m, n). Thiselement addsaconstraint onthevoltageof the(m, n) branch,
andthismust beincludedinthenodal matrix. Wenotethat
A
:
V
j
A
:
V
K
V
m
V
n
= 0
I
j
= I
k
= 0
I
m
= I
n
= 0
(5.4)
whereA
v
isthevoltagegainandI isthebranchcurrent. Then, wecanaugmentthenodal
matrixwithanextrarowandcolumnrepresenting(5.4):

[
[ 1
[ 1
[
A
:
A
:
1 1

V
j
V
k
V
m
V
n
I

I
m
I
n

(5.5)
Thiscreatesanewstampfortheelement. All elementsthatcannotbedescribedsimply
byanadmittancematrixrequirethisnewstamp.
A full treatment of modiednodal analysis is beyondthescopeof this chapter. The
interestedreader shouldconsult [3] for moreextensiveinformation.
Sparse-matrixmethod
Imaginealargecircuit havingmanythousandsof nodesandconsistingof simple, two-
terminal elements. Eachof theseelementsisconnectedtoonlytwonodes, creatingfour
entries inthematrix. From(5.2) weseethat amatrix position(j, k) has anentry only
if acircuit element isconnectedbetweenthosenodes. Clearly, most nodesdonot have
elementsconnectedbetweenthem, somost entriesinthematrixarezero.
Storingalargematrix consistingmostly of zeroelementswastesmemory, andcom-
putations with such amatrix largely involvemultiplying zero by zero and adding the
result tozero. Thisisespeciallytroublesome, inviewof thefact that LU decomposition
of amatrixisanN
3
process; thatis, theamountof computationincreasesapproximately
asthecubeof thematrixdimension, N.
Thissituationclearlyiswasteful, somethodshavebeendevelopedtoimproveit. An
earlymethod,developedspecicallyforcircuitanalysisandstill inuse(e.g.,inthecircuit-
analysis programSPICE), involves storingthenonzero matrix values indoubly linked
lists, alist for eachrowandeachcolumn. Inthis way, largenumbers of zeroelements
neednot bestored. Tondmatrix elements, it isnecessary totraversethelists, clearly
aslowprocess. Somekinds of access, however, suchas ndinglocations for element
stamps, canbefacilitatedby savingpointers to thoselocations. LU decompositionof
asparsematrix tends to createll-ins; that is, zero locations areoften replaced by
nonzerovalues, andnewlocationsinthelistsmustbecreatedforthem. Variousheuristics
areusedtominimizell-ins.
5.2 Methods of analysis 193
As thecost of computer memory has decreased, recent sparse-matrix methods have
favoredincreasedcomputational speedover minimizedstorage. Furthermore, thematri-
cesinmost kindsof nonlinear circuit analysisarenot nearlyassparseasalinear, nodal
matrix, andinsuchcaseslarger numbersof ll-instendtobegenerated, sothevalueof
minimizingstorageis, inanycase, minimal. Insuchanenvironment, iterativemethods
that operateonthecompletematrixareoftenused.
Thegoal of suchmethods is to reducetheresidual of thematrix. Specically, sup-
posethat onewishes to ndthevector V givenanadmittancematrix Y andexcitation
sourcesI :
YV = I (5.6)
WeestimateV insomewayanddenetheresidual, r(V):
r(V) = [YV I [ (5.7)
Clearlyweneedtominimizer(V), or, equivalently, r
2
(V).
Onecouldviewthiscase, for example, asanoptimizationproblem. Onemight take
thegradient of r
2
(V), r
2
(V), andminimizer
2
(V) intheindicateddirection. Although
intuitively thisprocessmay seemslow, it scaleswithmatrix sizefar better thansimple
LU decomposition. Other methodsaremoresophisticated, takingadvantageof sparsity,
matrixstructure, andtheavailabilityof agoodapproximateinverse. Inharmonic-balance
analysis, which we describe in Section 5.2.2, a method called GMRES (generalized
minimumresidual), oneof aclassof methodscalledKrylovsubspacemethods, hasbeen
favoredfor manyyears.
Suchmethodsareusedprimarilyforhandlinglargesparsesystemsof linearequations,
wherethescalingof computationwithmatrixsizemaybeontheorder of N
1.5
or better,
insteadof LUs N
3
. Their advantagemay beminor or nonexistent for small systems of
equations, andtheir abilitytohandleill-conditionedcases(thoseinwhichthematrixis
nearlysingular) areworsethanclassical LU decomposition. For thisreason, theuser of
nonlinear circuit-analysissoftwaremust beespeciallycareful toavoidsituationswhere
ill conditioning can occur. Useof certain kinds of time-domain model in frequency-
domain simulators; useof models that arepoorly dened, disconnected, or shortedin
certainfrequencyranges; poor choiceof thermal parametersinself-heatingmodels; and
model parametersthat creatediscontinuitiesarecommonproblemsthat areoftenunder
thecontrol of theuser. Weexaminethesematterslater inthischapter.
5.2.2 Harmonic-balance analysis
Harmonic-balanceanalysisseemstohavebeendevelopedsimultaneouslybyanumber
of individuals[46]. Whilesingle-purposeharmonic-balancesoftwarehasexistedsince
the mid 1970s, large-scale, general-purpose harmonic-balance simulators have been
availableonlysincethemid1980s[7]. Sincethen, however, harmonic-balanceanalysis
hasbecomethedominant tool for power-amplier designersat microwavefrequencies,
andit probablyshouldbeusedmorethanit isfor lower frequency(RFIC) applications.
194 Computer-aided design of power ampliers
I
R
V
s
V
+
+

Figure 5.4 A simpleDC circuit includingadiodecannot beanalyzedalgebraically. Thevoltage
andcurrent at thediodecanbefoundonlybyiterativemeans.
Anheuristicintroductiontoharmonic-balanceanalysis
Letsconsider theproblemof ndingtheDC voltageof adiodeinthesimplecircuit of
Figure5.4. Thecircuit isdescribedbytheequation,
I =
V
s
V
R
= I
sat
[exp(V) 1] (5.8)
where
= q,(KT). (5.9)
Thequantities in(5.8) and(5.9) arewhat onemight expect: qis electroncharge, K is
Boltzmannsconstant, T isabsolutetemperature, isthediodeidealityfactor, I
sat
isthe
currentparameter, Visthejunctionvoltage, andI isthejunctioncurrent. Therestof the
terms aredenedby thegure. It shouldbeclear frominspectionthat (5.8) cannot be
solvedalgebraically. Wecould, however, ndVbymeansof thefollowingalgorithm:
1. Denetheerror equation,
f (V) =
V
R
I
sat
[exp(V) 1]
V
s
R
(5.10)
ThisissimplyKirchhoffscurrent law; f(V) = 0whenit issatised.
2. Select somevalueof V as arst estimateof thesolution. If wehavesomeideaof
whatVshouldbe, obviouslythatestimateshouldbeused. Inmostcases, however, we
havenoidea,
1
soperhapssimplychoosingV=0mightbeareasonableplacetostart.
3. Calculatef(V).
4. Bymeansof someappropriatenumerical method, modifyVsothat[ f(V)[ decreases.
5. Repeat theprocessuntil [ f(V)[ issmall enough.
Thisidearaisestwoimmediatequestions; rst, wheredoweobtaintheappropriate
numerical method, and, second, what, precisely is small enough? Therst question
is answeredeasily: themathematicians havebeenhereaheadof us. Any goodtext on
numerical methodswill describemanynumerical techniquesthat might beappropriate.
Onegood, general methodfor ndingthezero of afunctionis Newtons method. This
technique, illustrated in Figure 5.5a, consists of repeatedly estimating the zero by a
linear extrapolation fromaknown point on thecurve. Given somefunction f(x), and
1
Or, moreprecisely, that big stupid machineon which wedepend, called acomputer, has no ideaof the
solution.
5.2 Methods of analysis 195
f(x)
f(x
0
)
x
0
x
0
x
x
0
x
x
0
x
x
df
dx
df
dx
(a) (b)
f(x)
f(x
0
)
Figure 5.5 Estimatingthezeroof anonlinear functionf(x) involvesusingthederivativeto
extrapolatetothexaxis(a). Thisprocessisrepeateduntil thezeroisfoundtoadequateaccuracy.
Themethodcanfail, however, if theinitial point x
0
ispoorlychosen; (b), for example, showsa
casewheretheprocesshasbeentrappedbyarelativeminimum.
aninitial estimatex
0
, wecalculatef(x
0
) anddf(x),dxat x
0
. Theequationfor thelinear
extrapolationis
f (x
0
)
df
dx

Lx = 0
x=x
0
(5.11)
fromwhichweobtainL x. Wethenestimatethezeroas
x = x
0
Lx (5.12)
thusobtainingabetter estimateof thezero. Wenowsimplyrepeat theprocesswiththe
newestimateof thezero as thestartingpoint. If thecurveis smoothandtheoriginal
estimatex
0
was reasonably closetothezero, eventually theprocess will convergetoa
solution.
Theperformanceof thismethoddependsonthestrengthof thenonlinearity andthe
quality of theinitial estimate, x
0
. If it is applied to alinear equation, it will converge
exactly in a single iteration, but if it is applied to a strongly nonlinear function, it
may requiremany iterations. Insomecases, it may fail completely. Figure5.5bshows
an example of a convergence failure, in which the process has been trapped by a
relativeminimum. Other quirks, suchasaninectionpointnear thezero, canalsocause
convergencefailure. Modicationsof themethodcansometimescircumventsomesuch
problems; for example, reducing thestep sizefromthefull Lx to something smaller
easilysolvestheinection-point problem.
Thesecondproblem, determiningwhether thesolutionhasconvergedadequately, is
moresubtle. In theone-dimensional casewehaveexamined here, theanswer simply
dependsonthenecessaryprecision. Inmostpractical cases, however, wehaveavectorof
harmonicvoltages, V=[V
0
, V(
p
), V(2
p
). . . ]
T
where
p
isthefundamental excitation
196 Computer-aided design of power ampliers
frequency. Then, insteadof ascalar f(V), wehaveF(V), avector of current errors. In
this case, the criterion for a solution is much less clear. Should we specify a limit
for the vector magnitude [F(V)[, the magnitude of each component of F(V), or the
fractional error in each component of F(V)? Theanswer largely depends on thetype
of problemweareaddressing. Intherst case, [F(V)[ - , where isthelimit, small
components of F(V) couldbehighly inaccurateeventhough issmall. Thiscouldbe
troublesome for intermodulation analysis, where one expects large differences in the
magnitudesof variousfrequencycomponents. Thesecondcriterion, f
k
(V) - , for all
K components of thevector, is nefor small components but may befar toostringent
for largecomponents. Thethird, fractional error, isdifcult toestimatewhentheerror
is largeand thecorrect valueof each component is unknown. It also tends to betoo
stringentfor small error components. Usually, somecombinationof thesecriteriaworks
best.
Itisimportanttorecognizetheideaunderlyingthislittleexercise: wehaveshownthat
it is not necessary to beableto analyzeanonlinear circuit directly. Infact, except for
trivial cases, it isimpossibletodoso. It isonlynecessarytondsomemethodthat can
reliably improveahypothetical solution. Then, by applyingthat methodrepeatedly, we
canreducetheerror tothepoint at whichit isnegligible.
Amoregeneral case
Now, lets make the problem a little more difcult. Consider the situation in
Figure5.6, wherethediodeis excited by asinusoidal sourceat thefrequency
p
. In
general, thesourceimpedanceiscomplexand, of course, differsateachharmonicof the
excitationfrequency. Wenowhavemadetheproblemmultidimensional, sinceweneed
tosatisfyour equationsat anumber of voltageharmonics, or, equivalently, at anumber
of voltagesamplesinthetimedomain. Wecannolonger writeasimpleequationhaving
the formof (5.8), because the diode junction must be described in the time domain
whilewehavedescribedthelinear part of thecircuit inthefrequencydomain. Howdo
weaccomplishthis?
For themoment, letsassumethatweknowthediodevoltage. Thisisexpressedinthe
frequency domainby V, avector of harmonic components, or equivalently inthetime
domainby thesampledwaveformv(t). Figure5.6(b) showsthat wecanndI
LIN
(k
p
),
thefrequency-domaincurrentinthelinearpartof thecircuitateachharmonicfrequency
k
p
, includingDC, as
I
LIN
(k
p
) =
V(k
p
) V
s
(k
p
)
Z(k
p
)
(5.13)
Notethat V
s
has acomponent only at
p
, theexcitation frequency, and is zero when
k,= 1. Thetime-domaincurrent inthediode, i
NL
(t) inFigure5.6c, is
i
NL
(t) = I
sat
[exp(:(t)) 1] (5.14)
Thetime-domainjunctionvoltagev(t) isperiodicsoit canbefoundbyinverseFourier-
transformingV. Similarly, weFourier transformthetime-domaincurrent to obtainits
5.2 Methods of analysis 197
V
(a)
(b)
(c)
V
s
Z()
V
s
I
LIN
I
LIN
i
NL
(t)
I
NL
(t ) = I
sat
[exp(v(t ) ) 1]
I
NL
+

V
+

v(t )
+

Z()
Figure 5.6 Themorecomplicatedcase, inwhichthediodeisexcitedbyasinusoidal sourceanda
complexsourceimpedance; (a) showsthiscase; (b) and(c) showthedecompositionintolinear
andnonlinear subcircuits, respectively.
harmonic components, I
NL
(k
p
). To satisfy Kirchhoffs current lawat each harmonic,
werequire
I
LIN
(k
p
) I
NL
(k
p
) = 0 (5.15)
Nowwemust facethefact that wedont really knowthevoltagecomponents V(k
p
).
Tondavalidsolution, wemustndthesetof voltagecomponentsV(k
p
), k=0. . . K,
whereK is thegreatest signicant harmonic, that satises (5.15) at each k. In effect,
wehaveK1equations of theform(5.15) andK1variables, thevoltages V(k
p
).
Theproblemis not much different fromtheprevious one; wemust ndthezero of a
nonlinear function. Inthiscase, however, theproblemismultidimensional.
Fortunately, Newtons methodis easily modiedto accommodatemultidimensional
problems. Weformulateourvoltagesandcurrentsinvectorformandemployaniterative
processentirelyanalogousto(5.11) and(5.12). Wedeneour error functionas
F(V) = I
LIN
(V) I
NL
(V) (5.16)
andasolutionisfoundwhen
F(V) = 0 (5.17)
198 Computer-aided design of power ampliers
Thezeroisestimatedas
F(V)
_

V
F(V)
_
LV = 0 (5.18)
whichissolvedfor LV. Thenewestimateof V is

V = V LV. (5.19)
Thederivativeof avector withrespect toavector, whichweseein(5.18), isaJ acobian
matrix. This matrix contains all thederivatives of each component of F with respect
toeachvoltagecomponent V(k
p
). Assuch, it containsinformationabout theeffect of
every voltagecomponent onevery error component. This is all theinformationabout
thelocal error that onecouldpossibly have, andit implies that themethodshouldbe
verypowerful for ndingthezero.
For this reason, as well as considerable successful empirical experience, multidi-
mensional Newtons method has become the favored technique for both time- and
frequency-domainnonlinear-circuitsimulation.CompareNewtonsmethodto,forexam-
ple, anoptimizationapproach, inwhichthegradient, [F(V)[, isusedtodeterminethe
direction in which changes in V should go. That formulation would includeinforma-
tion about theeffect of each V(k
p
) component on [F(V)[, but not on theindividual
components of F(V). It should be expected that such a method would be distinctly
inferior to multidimensional Newton, as implementations of both methods quickly
demonstrate[7].
Fortunately, theJ acobianmatrixissurprisinglyeasytocreate. Thetermsof thematrix
aresimply theFourier components of thediodes conductancewaveform(i.e., its IV
derivativeevaluatedat v(t)) addedtotheadmittancesY(k
p
) = 1/Z(k
p
) inappropriate
locations. Inlargecircuits, however, theJ acobianis invariably large, so solving(5.18)
to obtain LV can be computationally costly. Iterative methods such as GMRES are
extremelyhelpful inminimizingthat cost.
Our nal matter istoshowhowharmonic-balanceanalysisisappliedtolargecircuits.
It should be clear that (5.16) represents Kirchhoffs current law, so it is valid when
V, I
LIN
, andI
NL
represent voltages andcurrents at bothcircuit nodes andfrequencies.
Specically, V couldjust aswell be
V = [V
1
(0). V
1
(
p
). V
1
(2
p
). . . . . V
1
(k
p
). V
2
(0). V
2
(
p
). . .]
T
(5.20)
where V
n
(k
p
) is the voltage at node n and frequency k
p
, and (5.185.19) remain
unchanged. It isnecessaryonlytogeneralize(5.13) intheobviousmanner,
I
LIN
= Y(V V
s
) (5.21)
where Y is the admittance matrix of the linear parts of the circuit, at all harmonic
frequencies, arrangedintheformof (5.20). Althoughpreviously weassumedV
s
tobe
asingle, sinusoidal excitation, (5.21) shows that this restrictionneednot beimposed.
V
s
could be, for example, a nonsinusoidal source or a set of nonsinusoidal sources
connectedtooneor morenodesof thecircuit anddescribedbytheir Fourier series.
5.2 Methods of analysis 199
Harmonicbalancevariants
Asonemightexpect, thestoryisnotassimpleaspresentedabove. A numberof methods
havebeendevelopedtoimprovethespeedandrobustnessof harmonic-balanceanalysis
andtoaccommodatemoretypesof analysis. Theseincludethefollowing:
Normreduction
It isfrequentlyobservedthat takingthefull Newtonstepdenedby(5.18) usuallydoes
not result inrobust convergence. Dynamically adjustingthesizeof thesteptoprovide
anoptimumreductionintheerror functionisinvariablyabetter approach. Thus, (5.19)
becomes

V = V LV (5.22)
where is aconstant that canbevariedas needed. Usually, is initially small andis
increasedby thesimulator until theerror isminimized. SincetheJ acobianneednot be
factoredduringthis process, it is computationally relatively inexpensiveandincreases
thesimulatorsrobustnesssignicantly.
Semanskii iteration
Semanskii iterationissimplyafancynamefor reusingtheJ acobianinsteadof reformu-
lating it. Especially if thecircuit is not too strongly nonlinear, or theprocess is close
to asolution, using asingleJ acobian formulation for several iterations can speed the
solutionprocess.
Semanskii iterationcanbeusedonlywhenLUdecompositionisusedtosolve(5.18);
it is not applicable to Krylov methods. Since literally all modern harmonic-balance
simulatorsuseKrylovmethods, Semanskii iterationisnolonger of great interest.
Krylovsubspacemethods
Ateachiterationof theharmonic-balanceprocess, wemustsolvethematrixequation,
J LV = F(V) (5.23)
where J is theJ acobian. When J is large and at least somewhat sparse, as it usually
is, Krylovsubspacemethods, particularlyGMRES, canbeextremelyhelpful insolving
(5.23) rapidly. This process requires that J and theright sideof (5.23) bemultiplied
by a preconditioner. The preconditioner is an estimate of the inverse of J ; the exact
inverseis, of course, unknown; if it wereknown, theproblemwouldbesolvedby the
preconditionermultiplication. Thesuccessof theprocessdependsstronglyonthequality
of thepreconditioner. Thecloser it is to theinverse, themoreefcient thesolutionof
(5.23).
By now it should beobvious that it is not necessary to solve(5.23) completely. It
is necessary only toreachthepoint whereLV, appliedto(5.22), decreases thecircuit
error. Thus, thesolutionof (5.23) neednot beexact, andit canbeterminatedwhenever
an improvement in theerror is reached. This is especially useful in theearly steps of
a harmonic-balance analysis, when even an exact solution of (5.23) would not result
200 Computer-aided design of power ampliers
in a good estimate of the zero, and thus would represent wasted effort. Because of
thedependenceonsuchpartial solutions, thesemethods aresometimes calledinexact
Newtonmethods.
Weightingof theerror equations
Wehintedat theproblemof determining theaqdequacy of thesolutionearlier inthis
section. We noted that the norm[F(V)[
2
, while the default method for determining
convergence, is by itself apoor criterion, as it discriminates against small components
of F(V). In fact, theproblemis worsethat this. It can beshown that, in general, the
gradient of F(V) does not point in the same direction as LV, and, in some types of
circuit, itisactuallyperpendicular [8]. Theproblemisespeciallyacuteincircuitshaving
controlledsources, whichincludevirtuallyall of solid-stateelectronics. Thismeansthat
agoodNewtonstep, inthesenseof improvingmost of thecomponents of F(V), does
not necessarilyimprovethenorm.
This problemcanbesolvedby weightingF(V) beforedeterminingthenorm. Mul-
tiplyingF(V) by theJ acobianis anappropriateweightingfunction. Usingthis simple
method, beforeevaluating thecircuit error, is asimpleand highly effectivetechnique
for improvingtherobustnessof harmonic-balanceanalysis.
Multitoneexcitations
Sofar, wehaveassumedtheexcitationtobeperiodic, soitcouldbeexpressedasaFourier
series. Thisis, however, anunnecessaryrestriction. Nothinginthepreviousformulation
requiresthat thefrequencycomponentsbeharmonicsor theexcitationsourceshavethe
samefundamental frequencies; theycanbewhatever excitationfrequenciesandmixing
productsareusedandproducedbythecircuit. Ingeneral, thefrequenciesinthecircuit
are
= m
p1
n
p2
(5.24)
wherem, n, . . . , areintegersand
p1
,
p2
, . . . , aretheexcitationfrequencies, assumed
tobenoncommensurate; thatis, notharmonicallyrelated. Equation(5.20) thenbecomes
V = [V
1
(0). V
1
(
1
). V
1
(
2
). . . . . V
1
(
K
). V
2
(0). V
2
(
1
). V
2
(
2
). . . . . V
2
(
K
). . . .]
T
(5.25)
where
k
, k=1, . . . ,K arethesetof nonharmonicmixingfrequenciesdenedby(5.24).
TheonlyproblemistheFouriertransformation; sincethevoltagesandcurrentsingeneral
arenot periodic, wecannot useasimplefast Fourier transform(FFT) to stepbetween
thetimeandfrequencydomains.
A number of methodscanbeusedtoperformthenecessarytime-to-frequencytrans-
formation. Indeed, a cottage industry in developing such methods existed for a time
during the1980s and 90s [916]. Oneobvious choiceis theuseof adiscreteFourier
transform(DFT). Thisworkswell aslongasthefrequencycomponentsarenot closely
spaced, asituationthat istoorestrictivefor manykindsof analysis. Theill-conditioning
of theDFT in such cases can beavoided by using nonuniformly selected timepoints
inthetransform; someof thetransformmethods focus specically onthat time-point
5.2 Methods of analysis 201
selection process [9]. Theuseof an n-dimensional Fourier transformis equivalent to
selectingthetimepointsoptimally[10], but it isuseful onlywhennnoncommensurate
excitationfrequenciesareused. Thislimitsntoapproximatelyn 3. Higher valuesof
naretheoreticallypossiblebut becomecomputationallyexpensiveinpractice.
Envelopeanalysis
It is valuable to be able to use circuit-simulation techniques for excitations that are
modulatedwaveforms. Suchananalysisisstraightforwardinatime-domainsimulator,
but time-domain analysis may not beappropriatefor certain kinds of circuit. For this
reason, harmonic-balancemethodsthatcanhandlesuchwaveformshavebeendeveloped
[1719].
A naive approach to this problemmight be to generate the modulated excitation
waveform, sample it periodically at a rate based on the modulation time scale, and
performaharmonic-balanceanalysisfor eachsample. Thisapproachhastwoproblems:
rst,ityieldsnomoreinformationthananAMAM/AMPMbehavioral model,inwhich
thecircuits amplitudeand phaseresponseto arangeof sinusoidal signal amplitudes
is determinedandstoredinalook-uptable. Second, it does not account for long-term
memory(memoryontheorder of theinversebandwidth) ineither thelinear or nonlinear
parts of thecircuit. In envelopeanalysis, weperformaharmonic-balanceanalysis by
samplingthemodulatedcarrier ataraterelatedtothemodulationbandwidth. Thetricky
part isincluding, inanapproximatemanner, theeffectsof circuit memoryontheorder
of thesampleperiod.
Dealingwithmemory inthenonlinear subcircuit is simple; for example, consider a
capacitor. Thecapacitorschargeis
Q(t) =
1
2
K

k=K
Q
k
(t)exp(j k
p
t) (5.26)
whereQ
k
(t) is themodulationwaveformfor thespectrumat thekthcarrier harmonic,
k
p
. ThecapacitivechargeisQ(V) andthecurrent isdQ(V),dt. Differentiatinggives
i (t) =
dQ(V)
dt

V=:(t)
=
1
2
K

k=K
_
j k
p
Q
k
(t)
dQ
k
(t)
dt
_
exp(j k
p
t) (5.27)
Sincevirtuallyall thecircuit elementsinthenonlinear subcircuit aresmall, thismodi-
cationhasonlyminor effects. Theeffectof thelinear subcircuitismuchmoreimportant.
Dealing with thelinear subcircuit is moreof aproblem. Weneed somemethod to
ndthecircuit currentsinresponsetotheexternal nodevoltages, whicharemodulated
sinusoids. This problemis not much different fromthat of using frequency-domain
datainatime-domainsimulator, andcanbehandledinmuchthesameway. Existing
approachesaretouseaniteimpulseresponse(FIR) model, aninniteimpulseresponse
(IIR) model, or to expand thefrequency responsein aTaylor series [17]. Depending
uponthemethodemployed, theharmonic-balanceprocedurecanthenbeperformedin
thecustomaryfrequencydomainor, alternatively, inthetimedomain(sometimescalled
awaveformbalanceapproach). Theiterativeprocess, in any case, is somewhat more
complexthanfor thesimplesinusoidal analysis.
202 Computer-aided design of power ampliers
Envelopemethodsareclaimedtoreducecomputational cost relativetotime-domain
methods or multitoneharmonic-balancemethods. It is nowunderstood, however, that
thelatter methodsdonot requireuniformsamplingintervalsof afractionof thecarrier
period, andwhenmoreintelligent approachesareused, thecomputational effortsof all
arecomparable. At thesametime, thegrowingsophisticationof behavioral modeling
methodshasmovedmuchof thiskindof analysisfromthecircuittothesystemsimulator;
indeed, most of theinformation provided by envelopeanalysis is used at thesystem-
analysis level, wherethesystems effect onthemodulatedwaveformcanbeincluded.
In the systemsimulator, the computational cost is far lower, as the nonlinear circuit
needs to beanalyzed only as necessary to generateamodel. Systemcalculations can
then be performed indenitely with no further attention to the nonlinear component.
Furthermore, afundamental dependenceonbehavioral modelsformodulated-waveform
analysis allows modeling frommeasurements of real components as well as circuit
analysis. For these reasons, dependence on envelope analysis, which never has been
great, iscurrentlydecreasinginfavor of behavioral modelingapproaches.
5.2.3 Time-domain analysis
Itiswell knownthatalinear circuitcontainingNindependentreactiveelements(i.e., not
includingsuchtrivialitiesastwocapacitorsconnectedinparallel) canbedescribedinthe
timedomainby anNth-order linear differential equation. Furthermore, any Nth-order
linear differential equationcanbeexpressedasaset of Nlinear, rst-order differential
equations. Thesameisgenerallytrueof nonlinear circuits.
A number of methodsexistfor formulatingthetime-domaincircuitequationsdirectly
inmatrix form. Whileit ispossibletouseordinary nodal analysis, inwhichonly node
voltages arethevariables, it is usually moreconvenient to useamodiednodal form,
allowingcurrentstobevariablesaswell. Thecircuit isdescribedbytheequation,
dX
dt
G(X) S= 0 (5.28)
whereX isavector of time-domainnodevoltagesandbranchcurrents, G(X) isavector
of nonlinear functions of those quantities, and S is a vector of source voltages and
currents.
Thekeytotime-domainanalysisistheintegrationof (5.28).Aswithharmonic-balance
analysis, manymethodsof solutionareconceivablypossible, all of whichhavediffering
numerical characteristics. In all cases, it is necessary to represent thederivativeby a
discreteapproximation, whichconverts(5.28) intoaset of nonlinear equationsthat can
besolvedsequentially.
Onesimpleapproachistoestimatethederivativeas
dX
dt

tn
=
X(t
n1
) X(t
n
)
Lt
(5.29)
5.2 Methods of analysis 203
whereLt is thetimeinterval betweenpoints at whichX is evaluated. This expression
converts(5.28) into
X(t
n1
) X(t
n
)
Lt
G(X(t
n
)) S(t
n
) = 0 (5.30)
whichcanbesolvedalgebraicallyforX(t
n1
). Theprocessisthenrepeatedatsubsequent
time points. This method, while simple and fast to evaluate, has unacceptably poor
numerical characteristics. Inparticular, itserror-propagationcharacteristicsarepoor, as
well asitsabilitytohandlestiff systems.
2
A better methodistouse
dX
dt

t
n1
=
X(t
n1
) X(t
n
)
Lt
(5.31)
that is, to treat this as thederivativeat thenext timestep, rather thanthecurrent step.
Then(5.28) becomes
X(t
n1
) X(t
n
)
Lt
G(X(t
n1
)) S(t
n1
) = 0 (5.32)
Wenowhaveasystemof nonlinearequationsthatmustbesolvediterativelyforX(t
n1
);
analgebraicsolutionisnolongerpossible. If thevectorXhasdimensionK, wenowmust
ndthezerosof K nonlinear equations, eachof whichisK-dimensional. Wehaveseen
this problembefore, of course, inharmonic-balanceanalysis, whereNewtons method
wasusedfor thesolution. Thesamemethodisapplicablehere.
Althoughthisapproachrequiresaniterativesolutionat eachtimeinterval, it ismuch
morerobustthantheearlierone. Of course, thecomputational costappearsmuchgreater,
butisactuallynotassevereasonemightexpect. Intime-domainanalysis, thechangesin
X(t) fromsteptosteparegenerallyfairlysmall, soconvergenceisrapid. Thiscontrasts
markedly with harmonic-balance analysis, where Newton steps are often extremely
large. Thegreatest convergencedifcultyintime-domainanalysisusuallyoccursat the
beginning of an analysis, wherethesimulator must determinetheDC bias point and
initial conditionsfor theanalysis, andasolutionmust befoundinamuchlarger space.
The rules for convergence of time-domain methods are largely the same as in
harmonic-balanceanalysis. Therequirementswhichwill bepresentedinSection5.6.1,
for example, applyequallytotime-domainandharmonic-balanceanalysis.
Time-domain variants
Shootingmethods
J ust asharmonic-balanceanalysisinherentlyndsasteady-stateresponsetoaperiodic
excitation, time-domain analysis nds anetworks transient response, withor without
anexcitationthat neednot beperiodic.
To ndsteady-stateconditions withtime-domainanalysis, it may appear necessary
tointegrateuntil thetransient hasdiedout. Thismay beimpossible, inpractice, asthe
circuit may havetimeconstants that areorders of magnitudelonger thantheperiodof
theexcitation, sointegrationthroughalargenumber of cyclesmaybenecessary. After
2
Stiff systemsarethosehavingmultiple, widelyvaryingtimeconstants.
204 Computer-aided design of power ampliers
this longintegration, numerical errors couldbecomeso great that theresults couldbe
meaningless. This situation acombinationof longandshort timeconstants exists
moreoften than not in RF and microwavecircuits. Shooting methods avoid this long
integrationbysearchingdirectlyfor thesteady-stateconditions.
Thesteadystateisreachedwhen
X(t T) = X(t) (5.33)
whereT istheperiodof theexcitation. Theproblem, inessence, istondsomeinitial
conditionX(t) that remainsunchangedafter integrationthroughaperiodT. Again, this
isaprocessof ndingK zerosof K nonlinear equations, andcanbeapproachedinthe
samemanner assimilarlydenedproblems.
Frequency-domainmodels
Many passive-element models, suchasstriptransmissionlinesanddiscontinuities, are
bestdescribedinthefrequency-domainandcanbeanalyzedinastraightforwardmanner
by frequency-domain electromagnetic (EM) simulators. Theinability of time-domain
simulators to accommodatefrequency-domaindatahas beenasignicant impediment
totheir acceptancebydesignersof high-frequencyelectronics.
Over theyears, methodshavebeendevelopedtoallowtime-domainsimulatorstouse
frequency-domaindata. Oneobviousapproachistoderiveanimpulse-responsefunction
by Fourier transformation. For example, givenanimpedancefunctionZ(), aFFT can
beused to createan impulse-responsefunction z(t), which emerges fromtheFFT in
discreteformandcanbeusedin(5.32) inastraightforwardconvolution.
This process has a number of problems. SinceZ() is truncated in frequency, z(t)
extendsover all timeandisthusnoncausal. Furthermore, it happensthat Z() must be
verytightlysampledor artifactsandnonconvergenceinthetime-domainsimulationcan
result. Straightforwardpractical problemscanariseaswell; for example, Z() andz(t)
must havecompatibleintervals, or somekindof interpolationis necessary. This is not
onlyacomplication, it canalsocausenonconvergence.
Although methods for ameliorating these deciencies exist, the process is at best
inefcient. Better methods attempt to determine a Laplace-domain expression of the
form,
Y
i j
= Y
0.i j
sY
1.i j

K

k=1
_
A
k.i j
s p
k

A

k.i j
s p

k
_
(5.34)
whereY
ij
isaY-parameter of amultiport structure. Asmoderntime-domainsimulators
invariablyincludeafacilityfor handlingelementscharacterizedbytheir Laplacetrans-
forms, this methodis very straightforwardto implement. Its numerical characteristics
aregoodandthecharacterizationiscausal.
Multitoneanalysis
WenotedinSection5.2.2that efcient analysis of circuits under multitoneexcitation
required nonuniformsampling intervals. The same is true of time-domain analysis.
Whiletheyarenot asintuitiveasinharmonic-balanceanalysis, techniquesfor handling
5.3 Passive circuit structures and simulation accuracy 205
multitoneexcitations haveextended time-domain methods to such problems as inter-
modulationanalysisof powerampliers.Adescriptionof thesetechniquesiswell beyond
thescopeof thischapter; wenotethat theyexist andareavailableinsuchsoftware.
5.2.4 Applications of analytical methods
Asgeneral-purposecircuit-analysistools, time-domainmethodsareconsiderablyolder
thanharmonic-balanceones, datingfromthelate1960s. Oneof theearliest nonlinear
circuit analysis programs, SPICE, developedat theUniversity of Californiaat Berke-
ley, becameavailableintheearly 1970s. It was releasedas apublic-domainsoftware,
guaranteeingitswideavailability. SPICE still isusedextensivelyfor digital andanalog
integrated-circuit design. SincesiliconRFIC designgrewlargely fromthesiliconana-
logworld, SPICE andlater time-domainprogramshavebeenthedominant softwarefor
thosetechnologiesaswell.
Harmonic-balanceanalysiswasadoptedbythemicrowavecircuit designcommunity
largely because SPICE was not well suited to circuits having distributed structures.
Distributedcircuit elements, suchasnonideal transmissionlinesandmicrostripdiscon-
tinuities, aremorereadily described in thefrequency domain. Furthermore, transient
response, whichiswhattime-domainsoftwareinherentlycalculates, israrelyof interest
to microwavedesigners, whilesteady-stateresponse, which is provided by harmonic-
balanceanalysis, is precisely what they need. RFICs, operatingat frequencies belowa
few GHz, rarely usetransmission-linestructures. Interconnects in such ICs areoften
electrically short andcanbemodeledinother ways; for example, as RC transmission
lines.
Even so, many characteristics of thesemethods aremerging. Modern time-domain
software often includes methods for using frequency-domain models, and harmonic-
balanceanalysisincludesmethodsfor handlingnonperiodicwaveformsandsimulating
transientcharacteristics. Similarly, systemsimulatorsaretakingonsomeof thecapabil-
itiesof circuitsimulators, includingtheabilitytoaccountfor componentmismatches. In
thisway, differenttypesof simulationcanbeintegratedintoasingleprocessfor solving
problemsthat donot t convenientlyintoasinglemethod.
5.3 Passive circuit structures and simulation accuracy
Theproblemof modeling passiveelements in RF and microwavecircuits has been a
dauntingoneaslongashigh-frequencydesignhasexisted. Eventoday, theaccuracyof
circuit simulationismorestronglydependent onmodel accuracythanonthesimulator
itself.
Manyyearsago, most discontinuities(e.g., irisesinwaveguides) hadtobemeasured
andtheresultstabulated. Occasionallytheresultscouldbenormalizedinfrequencyand
dimension, sotheycouldbescaledandappliedtoawiderangeof structures. Eventually,
closed-formexpressions for theelement models werederivedfromthemeasuredand
tabulated data. Today, efcient EM simulators can do much of the heavy lifting
206 Computer-aided design of power ampliers
involvedinmodelingcircuit structures. They areespecially valuablefor themodeling
of power-amplier components, asclosed-formmodelssometimesarenot accuratefor
theconditions of high current and low impedanceoften encountered. Such tools can
becomputationally costly, however, so thedesigner should becareful in using them.
A littleresourcefulness indesign, especially favoringelements that areeasy to model
accurately, candomuchtoensuretheaccuracyof thedesignprocess.
5.3.1 Scattering parameter models
Thesimplest way to model any linear component is by its scattering (S) parameters.
Modern calibration techniques now allow accurate S-parameter measurements to be
madewell intothemillimeter-waveregion. Theuniversality of S-parameters is largely
anhistoricphenomenon, datingfromtimesbeforethewidespreaduseof circuit-analysis
software. Indeed, anyother setof hybridparameterscandescribelinear circuitelements;
admittance(Y) parameters, whichcarry exactly thesameinformationasS-parameters,
havelongbeenthecustomarywaytodescribeRF transistors. Moreover, onecouldclaim
that characterizingdevicesbyY-parametersmakesmoresensethanbyS-parameters, as
all suchdataarerepresentedwithinthesimulator inadmittanceform.
S-parameter models areactually moregeneral than onemight at rst assume. The
output fromEM simulators is invariably in the formof S-parameters (even though,
like circuit simulators, they calculate Y-parameters directly), as is the data from
database models, described later, in Section 5.3.4. Closed-formmodels, discussed in
Section5.3.2, startoutlifeasS-parameter measurementsaswell. Thus, thepointsmade
in this section, which examines several considerations in the use of S-parameters to
model circuit elements, applytoawidevarietyof modelingmethods.
S-parameter models aresomewhat inexible. It is only rarely possibletoscaleS- or
Y-parameterstodescribearangeof element typesor sizes, andtheparametersmust be
interpolatedtoobtainvaluesatfrequenciesbetweenthoseatwhichtheyweremeasured.
The method of interpolation can affect the results of the simulation; simple linear
interpolation between complex values is often unacceptable, as it results in gain and
VSWR curves having aclearly nonphysical, scalloped appearance. Interpolation in a
polar senseismuchbetter, asaresplineandrational-functionmethods[2].
An important consideration in all models (most easily illustrated, however, with
S-parameter models) is that themanner inwhichthecircuit element is measuredmust
beconsistentwiththewayinwhichitisused. Moreprecisely, themodesattheinterface
of themodel, inuse, mustmatchthoseof themeasurement. Thispointisbestillustrated
byanexample.
Consider the measurement of a chip capacitors S-parameters. The capacitor is
mounted on a carrier, with short, precise, 50 O transmission lines, and thecarrier is
placed in acalibrated test xture. Thesituation is illustrated in Figure5.7. Sincethe
test xturesreferenceplanesarelocatedat theedgeof thecarrier, theS-parametersare
thoseof thecapacitor plusitstransmission-lineconnections.
Normally, the transmission lines should not be part of the characterization, so the
referenceplanesaremovedcloser tothecapacitor; oneconventionistoplacetheplanes
5.3 Passive circuit structures and simulation accuracy 207
(a)
(b)
Figure 5.7 A chipcomponent isusuallymodeledfrommeasurementstakeninaparticular
conguration(a). If it isusedinanother conguration(b), themodel maynot bevalidbecause
themodesat theinterconnectionarenot thesame. Intheaboveexample, useof ateejunction
doesnot completelysolvetheproblem.
attheedgeof thecapacitorselectrode.
3
Atthispointweencounterasubtleproblem. The
elds at thecapacitors edgedo not consist solely of thequasi-TEM microstripmode;
theyincludeanumber of higher-order, evanescent modesthat areconcentratednear the
microstrip-to-chipdiscontinuity. Bymovingthereferenceplanetothecapacitorsedge,
wehaveeliminatedthedominantmodebutnottheevanescentones. Whenthecapacitor
isplacedinacircuit, thosemodes must thesameasinthemeasurement, or themodel
may loseits validity. Themodes arethesameonly if thecapacitor is connectedviaa
50O microstripthat islongenoughtoallowthoseevanescent modestodissipate. If the
capacitor isconnectedviaashorter strip, or oneof adifferentimpedance, themodesare
different andthusthemodel is, tosomedegree, invalid.
A second consideration arises when S-parameter models are used in a harmonic-
balanceanalysis. Suppose, forexample, thatwearesimulatinga10GHzpoweramplier
and use 12 harmonics of the fundamental frequency, plus DC, in the analysis. This
appears to suggest that weneedS-parameters fromzero to 120GHz. Not only would
this entail adifcult measurement, but it simply might not bepossibleto model many
elementsat suchahighfrequency. Most microstripdiscontinuities, for example, cannot
bemodeledaccurately at frequencies wherehigh-order modes canpropagateor where
radiation and surface waves may occur. The latter depend on the dimensions of the
3
Another conventionistoplacethereferenceplanesatthecenter of thechip. Inthiscase, thecircuitdesigner
mustaddctitioustransmissionlinesections, eachequal tohalf thecapacitorslength, tomovethereference
planestothemoredesirablelocationat theedgeof thechip.
208 Computer-aided design of power ampliers
circuits housingandthelocationof adjacent structures, whichmight not beknownat
thetimeof thesimulation.
Inpractice, this problemis not as severeas onemight fear. Solid-statedevices have
parasitic capacitances that shunt the nonlinear elements of the device or the device
terminals, sohigh-frequency currents intheexternal circuit areusually negligible: a1
pF gate-to-sourcecapacitanceof alarge, 10GHz power FET doesapretty goodjobof
short-circuitingthegateat120GHz! Thus, itisrarelynecessarythatmodelsbeaccurate
atsuchhighharmonicfrequencies. Itisnecessary, though, thattheybewell behaved; that
is, theS-parametersvarysmoothlywithfrequencyanddonottakeonimpossiblevalues.
S-parameter models often take on bizarre values (e.g., a passive element becoming
active) when asimulator extrapolates low-frequency measurements to obtain missing
high-frequencydata. Toavoidthisproblem, onecansimplyadddummy S-parameters
(e.g., representingashort-circuit) atafrequencywell abovethehighestharmonic. Then,
thesimulator interpolatesthedata, insteadof extrapolatingit, andtheresultsaremore
rmlybounded.
A secondproblemoccurs at DC. Suchelements as strip-transmission-linedisconti-
nuitymodelsmust havelowimpedancesat dc, andextrapolationtoDC fromRF values
rarelyensuresthis. Again, DCS-parametersmustbeprovided; thisisinvariablyasimple
thingtodo.
At DC, most modelsconsist of short circuits(or at least very lowresistances), open
circuits, or simpleresistances. Short circuits can createaproblemin formulating the
admittance matrix of the linear subcircuit, as connecting two nodes by a very low
impedance creates large matrix entries. Conversely, an open circuit can leave a node
oating, makingtheadmittancematrixsingular. Most modernharmonic-balancesimu-
latorshavewaystohandlesuchproblems; simpleonesaretoformulatetheadmittance
matrixdifferentlyat DC thanat RF or toconnect nodesbyanite, but negligiblysmall
resistance, insteadof zero. Other methods, moremathematicallyelegant, operateat the
matrix level. In any case, theuser should beawareof thesepotential difculties and
avoidthemwhenever possible.
This discussionillustrates oneimportant advantageof lumped-element models over
S-parameter models: theformer neednot beinterpolated, andtheyarewell denedand
guaranteedpassiveatall frequencies. Thecontinuousnatureof suchmodelsisanadvan-
tageinsometypesof circuit, especiallyoscillators, wheretheinevitablegraininess of
S-parameterscancausepoor convergence. For thisreason, manykindsof model, while
generatedfrommeasurement dataor EM simulation, areoftenrealizedinthesimulator
inlumped-element form.
5.3.2 Closed-form models
Closed-formmodelsconsistof lumpedanddistributedcircuitelementswhosevaluesare
determinedbyalgebraicexpressions, or, atworst, arelativelysimplenumerical process.
Thoseexpressionscanbederivedinanumber of ways. Frequently, theyarebasedonan
approximateanalysisof thedevice, but sometimesthey arecompletely empirical, with
parameter valuesdeterminedbymeasurements. Most modelsusedincircuit simulation
5.3 Passive circuit structures and simulation accuracy 209
(a)
R L
C
p
C
1
C
2
Air Bridge or
Undercrossing
(b)
Figure 5.8 Single-layer spiral inductor (a) andequivalent circuit (b) inaIVtechnology.
Multilayer inductorsmayrequireamorecomplexequivalent circuit andsiliconimplementations
mayhavetoaccount for additional lossmechanisms.
C
p
C
1
C
3
C
2
R L
Figure 5.9 Equivalent circuit of achipcapacitor (Figure5.7a. C
s
istheratedcapacitanceandC
p
is
theparallel capacitiveparasitic. TheinductanceL arisesfromthecurrentsinthecapacitors
platesandcausesbothparallel andseriesresonances.
areclosed-form, butincreasedcapabilitiesof bothanalytical softwareandEMsimulators
hascreatednewkindsof model withsignicant advantages.
Closed-formmodels consisting of lumped elements often can successfully model
distributedstructures. Anexampleis theuseof suchmodels to describemicrostripor
other strip-transmission-linediscontinuities. J ust as onecanmodel atransmissionline
by a cascade of series inductors and shunt capacitors, a discontinuity usually can be
modeledsuccessfullybylumpedelements.
As anexample, consider theplanar spiral inductor shown, alongwithits model, in
Figure5.8. Aswell asinductance, thespiral hasloss, capacitancebetweenitstheturns
of windings, andcapacitancefromthewindingstoground. If thespiral isnot toolarge
relativetoawavelength(andif itis, theinductor will likelybetoolargefor useanyway),
theseparasitics canbemodeled as lumpedelements. Theinterwindingcapacitanceis
modeled, to a good approximation, by a single capacitor across the terminals of the
spiral, andthecapacitancetogroundbycapacitorsat eachend. Thelossismodeledby
aseriesresistor, and, of course, theinductancebyasimpleinductor.
A second example, a model of the chip capacitor of Figure 5.7a, is shown in
Figure5.9. Thecapacitorhasseriesinductance, simplybyvirtueof itslengthandthefact
210 Computer-aided design of power ampliers
thatitcarriesatime-varyingelectriceld. Italsohasdielectricandmetallizationlosses,
and shunt capacitance between its electrodes and bond pads. The inductance creates
aseries resonanceandtheinterelectrodecapacitances, combinedwiththeinductance,
createaparallel resonance. Thelatterresonantfrequencyismuchhigherthantheformer.
Determiningthevaluesof themodel elementsisusuallystraightforward. Themethod
depends strongly onthetypeof device. For example, many of thecapacitor parasitics
canbefoundfromtheseries andparallel resonant frequencies, andtheloss resistance
fromtransmissionlossatresonance. If nothingelseworks, themodel canbedetermined
fromttingitsparameter valuestomeasuredS-parameters.
Closed-formmodelsarefrequently usedfor striptransmissionlines. Over theyears,
equationsfor suchlineshavebeendevelopedandpolished, inmanycases, toimpressive
accuracy. The critical characteristics of the lines characteristic impedance, phase
velocity, loss, and non-TEM dispersion are expressed in such models by algebraic
equations. Although theexpressions aresometimes fairly long, they can beevaluated
rapidlyandrarelyhaveasignicant effect oncomputationtime.
As one might expect, models for the most frequently used types of line are most
accurate. Microstripmodelsaremorematurethanmodelsforother typesof line; models
of coplanar waveguideandsuspended-substratelinesareprobablynext best. Modelsof
lessusedstructures, suchasslotlines, arenot asgood.
Most closed-formtransmission-linemodels work well incircuit simulators, as they
involvesimply calculating aset of admittanceparameters algebraically fromtheline
dimensions. Whilemostmodelsarerelativelysimple, andarenotcostlytoevaluate, some
can be relatively complex; certain coupled-line models and microstrip-discontinuity
models arean exampleof thelatter. Even so, in comparison to EM simulation, such
modelsinvariablymakequitemodest demandsoncomputational resources.
5.3.3 Models from EM simulation
Ascomputer capabilitieshaveadvanced, sohasthepracticalityof usingEM simulation
for characterizingthepassiveparts of high-frequency circuits. As of this writing, it is
practical toanalyzetheentirepatternof metal interconnectionsinasmall ICasasingle
structure. With time, and predictable improvements in both software and computer
hardware, it should bepossibleto do even more, and it is not unreasonableto expect
that, eventually, virtuallyall but thesimplest passivestructureswill becharacterizedby
EM simulation.
Several kindsof EM simulator, usingvariousmeansfor analysis, areavailabletoday.
A detaileddiscussionof thesesimulatorsisoutsidethescopeof thischapter; Swanson
[20] has given agood discussion of their useand technology. Below, weoutlineonly
their characteristicsasrelevant topower-amplier design.
Two-dimensional simulators
2D simulators analyze the cross-section of a transmission line or similar structure,
determiningitsinductance, capacitance, series-resistanceandshunt-conductancematri-
ces. It is assumed that the structure is innite in length. From these matrices, all
5.3 Passive circuit structures and simulation accuracy 211
characteristics of the line can be determined for its quasi-TEM mode only, although
non-TEM dispersioneffectssometimescanbeincludedfromempirical relations.
Such simulators arevery versatilein thekinds of structurethey can accommodate
andarequitevaluableincaseswhereclosed-formmodelsarepoor. Closed-formmodels
oftenareinaccurateindealingwiththickmetal, especiallyincoupledstriptransmission
lines, cannot analyze multiple, asymmetrical strips at all, and often cannot describe
lines onmultipledielectric layers. 2D EM simulators candeal easily withthesecases.
They areespecially useful for creatingcoupled-microstripmodels inICs, wheremetal
thicknessisoftennot small comparedtothegapwidths.
The lack of any need to specify a strip transmission lines length during the EM
analysis is asignicant advantage. Becauseof this, thestrips lengthcanbevariedin
thecircuitsimulator (e.g., duringnumerical optimization) withoutneedtorepeatthe2D
simulation. For most structures, 2D simulation is very fast, often only slightly slower
thantheevaluationof aclosed-formmodel.
Planar simulators
Thesesimulators, sometimescalled2
1
/
2
-Dsimulators, shouldmorecorrectlybecalled
3Dpredominantlyplanar simulators. Theyarebasedfundamentallyonspectral-domain
momentmethods, whichoriginallywerelimitedtozero-thicknessmetal stripsonlayered
dielectricsandimposedrestrictionsonthegeometry andcurrent distributiononmetal,
suchasviaholes, that wereperpendicular tothesubstrate. Virtuallyall of theseoriginal
limitations havebeencircumventedover theyears, however, andsuchsimulators now
canaccommodateawidevarietyof structures, includingthickmetal, dielectricbricks,
(rectangularareasof dielectricthatdifferfromthesurroundingdielectric), internal ports,
vertical structures, andsoon.
Planar simulators canuseeither anopenor closedformulation. Intheopenformu-
lation, themetal is placed on dielectric layers of inniteextent; on closed structures,
thedielectric is placedinsideametal box. Theopenformulationis perhaps morever-
satile, allowingtheanalysis of patchantennas, for example; theclosedformulationis
somewhat moreaccurate. Similarly, someformulations requirethat metal edges align
withapredeterminedgrid; somedonot. Again, theformer, whilemorerestrictive, are
generally moreaccurate. Thelatter canbemadeequally accurate, althoughsometimes
at thecost of increasedcomputationtime.
Planar simulators are much faster than full 3D simulators. While slower than 2D,
they includeall non-TEM effects. Unlike2D, they can analyzestructures that havea
complex 2D shape and need not be innite in any dimension or have any particular
symmetry. Thismakesthemideal for striptransmissionlinesandtheir discontinuities.
The speed of many simulators is impressive; because of this, they can be used for
determiningtheS-parametersof completecircuit nets, theentirepatternof metal used
for interconnectionsonICs.
Three-dimensional simulators
3D simulators arethemost general but also theslowest of thesimulators considered
here, andtheymakethegreatest demandsoncomputer resources. Full 3D simulators
212 Computer-aided design of power ampliers
can treat a wide variety of problems, including structures having great complexity
in all dimensions. They are not restricted to layered dielectrics. Typical applications
of 3D simulators aretheanalysis of waveguidediscontinuities, waveguide-to-coax or
waveguide-to-microstriptransitions, andcoaxial rotaryjoints.
3D simulatorshaverelatively littleapplicability inplanar circuits; planar simulators
candeal withvirtuallyall problemsthat ariseinsuchcircuits. It isfrequentlyassumed,
quiteincorrectly, thatthepresenceof anyvertical structureinanotherwiseplanar circuit
requirestheuseof full 3Dsimulation. Whilethismayhavebeentrueinthepast, itisno
longer truetoday.
5.3.4 Database models
WhileEM simulationof suchstructuresasmicrostripdiscontinuitiescanbequitefast,
its speedstill canbeprohibitivewhenlargenumbers of suchelements areinvolvedor
manyfrequenciesmust beused. OnesolutiontothisproblemistoprecomputetheS- or
Y-parametersof awidevarietyof structures, savetheminadatabase, andrecall themin
acircuit simulation. Sincethestructures inthedatabasemay not correspondprecisely
tothedimensionsor other characteristicsof thecircuit element, someappropriatetype
of interpolationisnecessary.
Technologies exist today for automatically convertingthedatabaseparameters toan
accuratelumped-elementequivalentcircuit[21]. Doingsoprovidessmoothinterpolation
between frequencies and insures passivity of theresulting network. It also eliminates
small discontinuitiesinthefrequencyresponse,whichcouldcauseconvergenceproblems
insomecalculations.
5.3.5 Parasitic extraction
Earlyanaloganddigital ICsweretraditionallydesignedaslumped-element circuits. As
circuit speedandcomplexity increased, theinterconnections betweentransistors hada
signicant effect onthecircuitsperformance. Thus, it becamenecessary tomodel the
interconnectingconductors. Thedesignow, however, oftenevolvedintoonewherethe
circuits initial designwas basedonideal interconnections. Then, theinterconnections
wereanalyzedandtheir effectincludedasanal stageof thedesign. Thisoftenresulted
inacircuitthatdidnotwork, anditbecamenecessarytoredesignthechipwith, perhaps,
shorter connections. SincesiliconRFIC designevolvedfromanalogdesign, thisrather
disjointedapproachhasbeenadoptedfor RFICsaswell.
Many kinds of softwarehavebeen developed for parasitic extraction, and they use
various methods for modeling the connections. All are, in some sense, types of EM
simulator. ModernRF anddigital ICsuseshort, narrowconductors, whichusually can
be modeled acceptably as RC circuits. Some extractors boast an inductance aware
extraction, whichincludestheconductors inductance.
A preferabledesignowusesconcurrentlayoutandcircuitdesign, inwhichthelayout
iscreatedsimultaneouslywiththecircuitdesign. Somekindsof moderndesignsoftware
5.4 Solid-state device models 213
support this methodology. In those, the design, layout, and parasitic extraction are
integrated, sothedesigner issparedanunpleasant surpriseafter thelayout iscomplete.
Theneedforcharacterizingcircuitmetal aspartof thedesignprocesshasalwaysbeen
obvious in thedevelopment of microwavecircuits. Even in microwaveICs, however,
when layout is delayed to the end of the design process, it often happens that some
structuressimplydonott, andredesignbecomesnecessary. Concurrentdesignprevents
thisfromoccurring.
5.4 Solid-state device models
Whilepower-devicemodelingis covered inChapter 2, andthermal effects indevices
arecoveredinChapter 9, certainaspectsof thedevicemodel affect thewaysimulations
proceed, and, indeed, determinewhether theamplier canbesimulatedsuccessfully at
all. Weconsider someof thosemattersinthissection.
5.4.1 Power device models
Power devices are large, as they must handle large currents and high voltages. To
accommodateahighcurrent, thetotal gatewidthof apower FET oftenislargerelative
to awavelength. To prevent degradationof deviceperformanceby distributedeffects,
the device must be divided into a number of cells, each a smaller FET with its own
gate; sourceanddrainregionsaregenerallysharedwithadjacent cells. Similarly, BJ Ts
andHBTsarerealizedasanumber of individual cellsconnectedinparallel, oftenwith
sharedelectrodesaswell.
InICs, thedesigner may havesomedegreeof freedomindecidinghowmany cells
can beused in a particular device and how they arearranged. Thedesign must then
includeanalysisof theinterconnectionparasiticsandmustdescribethemulticell device
byasingletransistor model or, if necessary, at most afewtransistors.
5.4.2 Modeling cell interconnections in large devices
Thecellsof apower deviceinvariably includealargenumber of interconnections. The
way thoseconnectionsarearrangedandtheamount of metal inthoseinterconnections
canaffect theperformanceof thedevice. Inacellular handset amplier, for example,
theload impedanceat thedevicemay beon theorder of 1 or 2 O; in this case, even
0.1nH of inductancerepresents approximately 1O of reactanceat 1.8GHz, clearly a
nonnegligiblequantity. Attheoppositeendof thedevice, thewaythatthegatesor bases
areconnectedisalsoimportant. Simplestisaparallel connection, althoughinverylarge
devicesthisconnectionmaynot provideuniformdrivetoall thecells. A tree-structured
connectionusually providesmoreuniformdrivebut islarger andmorecomplicatedto
model.
Although it is a frequent practice, modeling interconnect metal by conventional
transmission-lineanddiscontinuity modelsisrarely successful. Suchmodelsareoften
214 Computer-aided design of power ampliers
notveryaccurateinalow-impedanceenvironmentandarecorrectonlywhenwell sepa-
ratedfromeachother. Inapower device, bothdiscontinuitiesandstraight transmission
linesarecloseenoughtogether thattheir eldscouple, violatingafundamental assump-
tion in themodels formulation. EM simulation is a preferabletool for modeling the
interconnectionparasitics.
Since theinterconnects feed individual cells, it may betempting to treat each cell
in thesimulation as a separate transistor. The computational cost of this approach is
rarelyjustiedbythegaininaccuracy. Sinceaproperlydesignedinterconnect structure
providesuniformdrivetothecells, andproper thermal designshouldresult inuniform
cell temperatures, thereis littleto gain by treating thecells as separatedevices. It is
almostalwaysacceptabletoreducethelargedevicetoascaled, singledeviceor atworst
averyfewdevices.
5.4.3 Thermal effects in device models
Inpowerdevices,highpowerdissipationistobeexpected,soitisalmostalwaysnecessary
toincludeself heatingindevicemodels. Oftenself-heatingmodelsarenot available, or
thesimulator does not support them(SPICE does not). Most devicemodels, however,
include at least thermal scaling; that is, at the outset of the simulation, the user can
specify atemperaturefor eachdevice. Oncethesimulationis completeandthepower
dissipationhasbeendetermined, theuser cancorrect thetemperaturesasnecessaryand
rerunthesimulation. Thisisaclumsyprocess, especiallywhenalargenumberof devices
areinvolved.
Self-heatingmodelsdeterminethedevicetemperaturebycalculatingthepower dissi-
pationfromtheterminal voltageandcurrent waveformsandtheuser-speciedthermal
resistance. Thepower dissipatedinthedeviceat anyinstant t is
P
d
(t) =
K

k=1
:
k
(t)i
k
(t) (5.35)
summedover thedevicesK terminals. If thethermal massof thedeviceislargeenough,
andthetimescaleof thevariationsissmall, devicetemperatureis
T
d
= T
BP


j c

_
P
d
(t)dt (5.36)
whereT
d
is thedevicetemperature, T
BP
is themounting-surface(or baseplate) tem-
perature,
jc
isthethermal resistancebetweenthedevicesactiveareaandthebaseplate,
and issomelongperiodof time.
If thethermal timeconstant isnot longcomparedtothetimescaleof theexcitation,
thedevicetemperaturevarieswithtimeandmustbeincludedintheanalysis. Thiscanbe
accomplishedby theelectrothermal equivalent circuit inFigure5.10. Thetemperature
T(t) then becomes a variable quantity within the model, much like any other control
voltageor current in satisfying (5.17). In thegure, thethermal resistanceis treated
asalinear quantity, but inreality thethermal resistanceof all semiconductor materials
is nonlinear, increasing with temperature. This is an important effect in determining
5.4 Solid-state device models 215
C
th
T(t )

jc
P
d
(t )
T
BP
+
+

Figure 5.10 Electrothermal equivalent circuit for determiningthedevice-temperaturewaveform.


Thecurrent isset numericallyequal tothepower dissipationwaveformP
d
(t) from(5.35), isthe
thermal resistance, andC
th
isthethermal capacitance. Thedevicetemperatureisnumerically
equal tothevoltageacrossC
th
.
V
cc
I
c
+

V
ce
+

V
be
+

V
bb
R
bb
+

Figure 5.11 Biasedbipolar transistor subject toself heating. TheresistanceR


bb
providesstability.
devicetemperature, butfewdevicemodelsincludeit. Whenincluded, itcansignicantly
degradethenumerical conditioningof theproblem.
While self-heating models are considerably more satisfactory than thermal scal-
ing, they do tendtobeill-conditioned, sometimes causingconvergencefailureinboth
harmonic-balanceandtime-domainanalysis. Ill conditioningandtheresultingconver-
gencefailurecanbeasignof thermal instabilityinthecircuit, aswell. Suchinstability
canbedifculttopredict, andwhenitoccurs, itcanbedifculttorecognizeasthecause
of convergencefailure. Theproblemcan also beexacerbated by poor behavior of the
model outsideitsnormal operatinglimits.
Theproblemcanbeillustratedby thesimple, dc-biasedbipolar transistor shownin
Figure5.11. Thepower dissipationinthedevice, P
d
, is
P
d
= V
ce
I
c
(5.37)
whereV
ce
andI
c
arethecollector voltageandcurrent, respectively; wehaveassumedthe
basepower to benegligible. Thetemperatureincreasecausedby this dissipation, LT,
is
LT =
j c
P
d
(5.38)
216 Computer-aided design of power ampliers
where
jc
isin

C/W. Finally, wecansay


I
c
= (LT)I
b
(5.39)
and
I
b
=
V
bb
V
be
R
bb
(5.40)
where is thecurrent gain, V
be
is thebase-to-emitter voltage, whichweapproximate
asafunctionof temperatureonly, andtheother termsareasshowninthegure. R
bb
, a
ballast resistor, isincludedspecicallytoimprovethermal stability. Substituting(5.38)
to(5.40) into(5.37) gives
LT =

j c
V
ce
R
bb
(LT)(V
bb
V
be
(LT)) = C
tb
f
b
(LT) (5.41)
whereC
tb
isaconstant andf
b
isthepart of (5.41) dependent onLT.
Theresult isatranscendental equation, whichmust besolvedgraphicallyor numeri-
cally. Thisisdonebyexpressing(5.41) as
LT
/
= C
tb
f
b
(LT) (5.42)
LT
/
= LT (5.43)
andsolvingsimultaneously. ThisisshowninFigure5.12. InsiliconBJ Ts, increases
with temperatureand V
be
decreases. As aresult, (5.42) increases monotonically with
LT. If C
tb
is small enough, the solution is well dened. As C
tb
increases, however,
thesolutionbecomesmultiple, poorly dened, andnally nosolutionexists. Thelatter
casecorrespondstothermal runaway, awell knownpropertyof siliconbipolar devices.
In HBTs, by contrast, decreases with temperature, so the situation is much better.
However, eventhen, ill-conditioningcanoccur if C
tb
istoolarge.
Inasimilarmanner,itispossibletoshowthattheadditionof emitterresistanceremoves
theeffectof , sothethermal stabilityof thesilicondeviceismuchimproved; however, in
HBTs, thenegativethermal feedbackprovidedby islost. Forthisreason, siliconpower
devices areusually emitter-ballasted. Althoughbaseballast provides better stability in
HBTs, it decreases gain signicantly, so most HBT ampliers use a combination of
emitter andbaseballast.
Clearly, if the device itself is not thermally stable, any analysis of the circuit that
includesself heatingislikelytofail. However, evenincaseswherethedeviceisthermally
stable, aself-heatingmodel thatisbadlybehavedoutsideof thenormal rangeof operation
may exhibit ill conditioned behavior. We consider the importance of model behavior
outsidethenormal rangeof operationinSection5.6.
5.5 Special aspects of power-amplier modeling
Somecharacteristicsof circuit-elementmodelshaveanespeciallystrongeffectonpower
ampliers andtheir simulation. Circuit losses, for example, areimportant inall types
5.5 Special aspects of power-amplier modeling 217
T


=

T
T


=

T
T


=

C
tb
f
b
(T)
T


=

C
tb
f
b
(T)
(a)
Very large C
tb
Stable Temperatures
Stable Temperatures
Large C
tb
Large C
tb
Small C
tb
Small C
tb
(b)
T
T
T

Figure 5.12 Stablethermal operatingpointsarefoundbysolving(5.42) and(5.43) simultaneously.


When increaseswithtemperature, asinsiliconBJ Ts, it mayhappenthat nooperatingpoint is
possible. Thissituationcorrespondstothermal runaway. Conversely, inHBTs, decreaseswith
current andthusprovidesinherent stability. Thermal instabilityisstill possible, however, in
HBTs.
of circuit, but in power ampliers they becomecritical. Similarly, whileall chips use
bondwires, largenetworksof bondwiresarecommoninpower devicesandthusmust
betreatedinspecial ways.
5.5.1 Loss in circuit metalizations
Power ampliers have high current not only in their devices but also in their circuit
metal. Asaresult, I
2
Rlossesinthemetal canbesurprisinglyhigh; thisisespeciallythe
caseinICs, wheremetal layersarethin.
TheDC resistanceof arectangular metal sheet isgivenby
R = R
sq
L
W
(5.44)
whereL is thelengthof themetal sheet, Wis its width, andR
sq
is thesheet resistance
in O/square. Thesheet resistanceis theresistanceof asquaresection of themetal, a
quantity that isindependent of thesizeof thesquare. R
sq
issimply ,t, where isthe
metalsresistivityandt isthethicknessincompatibleunits.
218 Computer-aided design of power ampliers
Estimatingmetal lossesisusually straightforward, but afewmatterscancomplicate
it. Thesearelistedbelow.
1. Skineffect: Asfrequencyincreases, thecurrent becomesconcentratedatthesurfaces
of theconductor. Theskindepth, , thedepthat whichthecurrent densitydecreases
toexp(1) of itssurfacevalue, isgivenby
=
1

f
(5.45)
whereisthepermeabilityof thematerial, istheconductivity,andfisthefrequency.
InRFICs, whichoperateat lowfrequencies, skineffect israrelymuchof aconcern,
but inmicrowaveICs it may besignicant. If themetal thickness is morethantwo
or threeskindepths, increasingitsthicknessdoesnot measurablydecreaseitslosses.
Most strip-transmission-linelossmodelsaccount for skindepth.
2. Current distribution: Current in at conductors tends to concentrate at the edges.
Thisistrueof DCaswell asRF currents. Unfortunately, theconductor edgesusually
arefairly rough, especially if they aredenedby chemical etching. This roughness
increasesthelengthof thepaththat thecurrent must follow, thusincreasingresistive
losses.Asonemightexpect,thecurrentdistributionislessuniforminwideconductors
thaninnarrowones.
Microstriplossmodelsaccountfor thisnonuniformcurrentdistribution, asdoloss
estimatesfromEM simulations. They generally donot account for edgeroughness,
althoughtheysometimesincludecorrectionsfor surfaceroughness.
3. Multilayer metallizations: Many types of circuit board and IC use more than one
layer of metal for their metallizations. If thefrequencyislowenoughthat skineffect
canbeignored, thelayers canbetreatedto agoodapproximationas resistances in
parallel; thus,
R
sq
=
R
sq1
R
sq2
R
sq1
R
sq2
(5.46)
whereR
sq1
andR
sq2
arethesheet resistancesof thetwometal layers.
4. Metal imperfections: Especially in ICs, the resistivities of metal layers are invari-
ably greater thantheir handbook values. Thelatter aredeterminedfromlarge, pure
samples of the metal, but deposition technologies rarely provide such perfection.
Changes ingrainstructureandinclusionof impurities canincreasethemetal resis-
tivitysubstantially. Alloys, for example, invariablyhavemuchhigher resistivitythan
puremetals.
5. Metal oxidation and surface roughness: Especially at high frequencies, where the
current isconcentratedat themetal surface, suchimperfectionsasroughnessandan
oxidelayer increaselosses. Inmicrostripcircuits, most of thesurfacecurrent is on
theundersideof theconductor, sothesubstratessmoothnesslargelydeterminesthe
metalssmoothness.
Manyof thesephenomenaaredifcult toquantify. Theycanresult inaneffectivemetal
resistivity that appears to vary with frequency and metal dimensions. Even in simple
5.5 Special aspects of power-amplier modeling 219
cases (e.g., lowfrequencies andsimplemetal structures) many strip-transmission-line
models are not terribly accurate in predicting losses. Perhaps the best simple way to
treat themistousethestandardmodelsincombinationwithaconservativeestimateof
themetal resistivity. Theauthorsgeneral practiceistouseat least doublethehandbook
resistivityvaluesinall transmission-linemodels.
Calculationof losses by means of anEM simulator sometimes is not as rigorous as
one might assume. The usual process is to calculate the surface current distribution
ontheconductors, thento determinethesheet resistance, accountingfor skineffect if
necessary,andtodeterminethelossesbyintegrating.Thismethodisaccurateforlow-loss
conductorsthatareotherwiseideal, butdoesnotdirectlyaccountformetal imperfections
describedabove, anditsusefor multilayer metallizationsisproblematical.
Metal losseshavelittleeffectonthecircuit-simulationprocess. If anything, theytend
to improvetheconditioning of theadmittancematrices, and thus may haveat least a
theoretical effect inpreventingconvergencedifculties. This is likely to beimportant
only in inherently ill-conditioned cases, such as transmission lines that are precisely
one-half wavelengthlong.
5.5.2 Loss in circuit components
Becauseof thehighcurrentsinpower-ampliercomponents, lossesinnominallylossless
components capacitorsandinductors canbehighaswell. It isnot unknownfor chip
capacitors, whichideally dissipateno power, to becomehot enoughto melt thesolder
connecting themto the circuit board! For this reason, it is important in simulations
to monitor thecurrents in such elements and, along with information about their Qs,
calculatetheir power dissipation. Thisisespeciallyimportant for componentsinoutput
circuits, whereresonator currents canbequitehigh, andevensmall lossesmay havea
largeeffect onefciency.
A simulation of a WCDMA handset amplier illustrates the effect of circuit loss.
The matching circuit consists of a simple structure with two shunt capacitors and a
seriestransmissionline, astructurethatistypical for suchampliers. Figure5.13shows
theresults of theamplier simulation. Thesimulationis just of thepower stage; such
ampliers usually includeadriver stageas well. Thegureshows thetotal RF output
power of thecompleteamplier (i.e., includingall harmonics) andthetotal RF power
at theinput of theoutput matchingcircuit. Thedifferencebetweenthesecurves, 0.5dB
at maximumefciency, represents thepower dissipation in thematching circuit. This
analysis shows that 11%of theoutput power is dissipated in thematching circuit. At
rstglance, thislossisdistressing, butincellular handsetPAs, wheresmall, inexpensive
capacitors and inductors must be used in the output matching circuit, it is largely
inevitable.
5.5.3 Bond wires
Whilebondwiresexist inall kindsof circuit, theyareusuallyusedsimplyasintercon-
nections. Inpowerampliers, however, bondwires, whichhaveinductancesof afraction
220 Computer-aided design of power ampliers
0
5
10
15
20
25
30
35
40
0
0 2 4 6 8 10 2 4 6
Power (dB m)
WCDMA Amplifier Power Sweep
Total Output Power (L, dB m)
Power Sweep
Efficiency (R)
Power Sweep
Gain (L)
Power Sweep
20.04 dB m
20.04 dB m
31.59 dB m
31.08 dB m
Total Pwr at MC Inp (La,dB m)
Power Sweep
O
u
t
p
u
t

P
o
w
e
r

(
d
B

m
)
E
f
f
.

(
%
)
8 10 12 14 16 18 20 2223
10
20
30
40
50
60
70
80
Figure 5.13 Simulationof acellular handset power amplier, showinginput andoutput power at
theoutput matchingcircuit. Theoutput losscausedbymatching-circuit elementsinthiscaseis
0.5dB.
of onenanohenry, areoftenusedas matchingelements. This is anespecially common
practiceinoutput circuits, whereverylowinductancesareoftenneeded.
Theuseof bondwires for inductances creates threedifculties: (a) determiningthe
wiresinductanceandresistance, (b) makingsurethat thewirescancarry therequired
current, and (c) making certain that thedesired wirelength and shapeis consistently
produced in a production environment. Even where the bond wire is not used as a
matchingelement, thelowimpedancelevel of thecircuitmaycausethewireinductance
tohaveasignicant effect onmatchingor port VSWR.
TheDC fusingcurrent of a25mdiameter goldbondwireis approximately 0.6A
anditsDCresistanceisapproximately0.05O/mmof length. Thefusingcurrentdepends
somewhat onthebondwires length, andtheresistancemay beaffectedby skineffect
and the kinds of imperfection described earlier. The inductance of the wire is more
difcult to determine. A singlestraight wireover a ground planecan bemodeled as
a transmission line, but all practical bond wires are asymmetrically arched, a more
complicatedsituation. Finally, power devicesrarelyusesinglebondwires; theyusually
usemultiple, closelyspacedwires. Thewiresareinvariablymagneticallycoupled, sothe
impedanceof Nwiresisnotsimply1,Ntimestheimpedanceof asingleone. Suchlarge
bond-wirenetworksmust beanalyzedasawhole, not scaledfromsingle-wireanalyses.
The determination of bond-wire impedance is a straightforward problemfor a 3D
EM simulator (Section5.3.3). Withcare, abondwiresometimes canalso bemodeled
acceptably by aplanar EM simulator, withstraight vertical andhorizontal sections, as
longasthetotal lengthof themodeledwireisthesameasthereal one. Thesesimulations
aregenerally costly, so they should beminimized, perhaps by limiting thenumber of
bond-wirecongurationsinthedesign.
5.6 Practical aspects of nonlinear circuit simulation 221
Theproblemof uniformity may beless than it at rst appears. Automatic bonding
equipmentproduceshighlyuniformbond-wireshapes; aslongastheprocesssstandard
shapeisacceptable, wireuniformityshouldnot bedifcult toachieve.
5.6 Practical aspects of nonlinear circuit simulation
5.6.1 Convergence difculties
Wenotedearlier thatNewtonsmethodisnotguaranteedtoconverge; harmonic-balance
simulators regularly remind us of this fact. Newton-based harmonic-balance analysis
isneverthelessquiterobust, evenwhenappliedtostunningly complex problems. Most
convergenceproblems arenot inherent in thesimulator; they arisefrompoor charac-
teristics of models, which areoften under thecontrol of theuser, and fromtheusers
misunderstandingof thebest waytoset upananalysis.
Model characteristics
Many standardmodelsarenot well conceivedfor useinharmonic-balancesimulators.
Others have undocumented constraints on parameter values, which, if not observed,
causeconvergencefailure. Finally, modelsdesignedfor useintime-domainsimulators
are often transferred unmodied to harmonic-balance simulators, where they do not
workaswell.
A fundamental ruleof Newton-basedcircuitanalysisisthefollowing: all expressions
describingnonlinear circuit elements must becontinuous throughtheir secondderiva-
tives. This is equivalent to saying that aplot of therst derivativemust not haveany
kinksinit. Clearly, sinceNewtonsmethoddependsonderivativestoestimatethezero
of afunction, anysuddenchangeinthederivativemakesconvergencemoredifcult.
Theexistenceof suchproblemscanbeinherent inthemodel or canbecausedbythe
wayparametersaredetermined. Consider, for example, anonlinear passiveconductance
element describedbytheI,Vequations,
I (V) = a
0
a
1
V a
2
V
2
a
3
V
3
V > V
th
I (V) = 0 V V
th
(5.47)
Inthiscase, theparametersa
n
must beselectednot onlytomatchtheI/V characteristic
of thereal device, butalsosuchthatI(V
th
) =0anddI,dV=0atV=. Itcanbeshown
that theseconstraints denetwo of thea
n
values. This leaves only two coefcients to
adjust theshapeandoverall magnitudeof thecharacteristic. A naiveuser of thismodel,
however, might not recognizethis, andselect all thepolynomial coefcients to obtain
thebest overall t to themeasuredI,V curve. Thevirtually certainresult wouldbea
discontinuityat V= V
th
.
Another potential problemis aconsequenceof theway harmonic-balanceanalysis
operates. In the early iterations of an analysis, it is possible, indeed likely, that the
independent variables intheanalysis (usually nodevoltages) becomequitelarge, well
beyond thenormal operating rangeof thedevice. In this case, it is essential that the
222 Computer-aided design of power ampliers
model bewell behaved not only in its normal rangeof operation but at voltages well
outsidethat range. TheSPICE diodemodel serves as agoodillustration. Supposewe
weretouseanordinarytextbook characteristicfor thejunctioncurrent,
I (V) = I
sat
exp(V) (5.48)
where I
sat
is the diodes current parameter and 40. In the initial iterations of a
harmonic-balanceanalysis of adiodecircuit, thevoltagecanreachseveral hundredor
evenseveral thousandvolts, clearlycausinganumerical overowor underow. Tosolve
thisproblem, thefunctionusesaquadratic extensionabovesomelargethresholdvalue
of V; theextensionisdesignedsothederivativesarecontinuousat thethresholdvalue.
Note that a linear extension of the function would introduce a discontinuous second
derivativeat thethresholdvoltage, whichwouldnot beacceptable.
Valuesof independent variablesneednot behugetocausetrouble. Consider adevice
model thatincludesself-heating. Itislikelythatthemodel isnotwell denedattemper-
aturesbelowabsolutezero, but it ispossiblethat thesimulator, whichdoesnt knowthe
differencebetweenavoltandadegree, might, atsomepoint, createathermal valuethat
islessthanzero. Often, naivemodel developerssimplycreateahardlimit of T
0
. Thisis
anexcellent waytoobtainpoor convergencecharacteristics.
This kind of problemis greater in harmonic-balanceanalysis than in time-domain
analysis, asNewtoniterationsintime-domainanalysisbeginwithavalueof voltageor
current, ateachtimeinterval, thatisclosetothecorrectvalue. Theiterativeprocessusu-
allychangesthevalueof theindependent variableonlyslightly. Occasionally, however,
largevaluesof theindependent variablescanoccur intime-domainanalysis, usually at
start-upor whenastepfunctionof theexcitationoccurs.
It shouldgo without sayingthat derivatives of nonlinear model characteristics must
be programmed correctly. This is not as easy to ensure as it may seem, however, as
many nonlinear devicemodels useexpressions that areexceedingly complicated, and
their derivative expressions are even more so. It is quite common for printed model
documentation to contain errors, and for thoseerrors to propagatethrough simulator
implementationsforyears. Onesolution, oftenobservedtoday, isforthemodel developer
to createastandardimplementation, usually inSPICE but sometimes inpseudo-code
or ahigh-level languagesuchas VerilogA. This standardizes theimplementationand
reduces thedanger of multipleimplementations havingtheir ownerrors. Instead, they
all containthesameerrors.
Another solution is the use of automatic differentiation, a technology for creating
exact, analytical derivatives fromafunctional expressioninsourcecode. Inthis case,
the derivatives of the nonlinearity need not be programmed, but derivatives for each
elemental function (such as an exponential, sine function, and so on) must be pro-
grammed, as well as the chain-rule process. The advantage of this technology is the
guaranteeof correct derivatives, aslongasthederivative-generatingprocessiscorrect.
Thedisadvantage, of course, isthatanerrorintheprocessaffectsall models, notjustone.
A nal possibilityistheuseof numerical derivatives; that is, simplyestimating
df (:)
d:

Lf (:)
L:
(5.49)
5.6 Practical aspects of nonlinear circuit simulation 223
usingsomesmall incrementL:. Thisisalmostneverdoneineitherharmonic-balanceor
time-domainanalysis. Itintroducesmanynumerical problems; forexample, determining
thesizeof L: sothat it issmall enoughtoproduceareasonablyaccuratederivativebut
not sosmall that itisaffectedbylossof numerical precision. Empirical experiencewith
numerical derivatives shows that they oftenarenot sufciently accuratetoallowgood
convergence, especiallynear thesolution.
It iscommoninsomesimulatorstoaddlow-valueconductancesacrossall nonlinear
elements, or fromall nodestoground, toimproveconvergencecharacteristics. Theneed
for this canbeunderstoodintuitively by recognizingthat theJ acobianis similar to an
admittancematrix. If anonlinear elementisturnedoff, say, bybiasingaFET well below
its threshold voltage, someof its nodes may beopen-circuited. This results in a row
of theJ acobianhavingzeroor very small entries, renderingit singular or nearly so. In
SPICE, thedefault valuefor theconductanceis10
12
O
1
. Thisvalueof conductance
rarely hasasignicant effect ontheaccuracy of thesolution, but it still canbehelpful
inprovidinggoodconvergence. Thevaluecanbeincreasedby statements inSPICEs
optionsblock. Harmonic-balancesimulatorsusuallyhavesimilar capabilities.
Analysis characteristics
Continuationmethods
Wenoted earlier that Newtons method converges reliably in asingleiteration in any
linear circuit. Thus, onepossibilityfor improvingtheconvergenceissimplytolinearize
thecircuit insomeway.
Thestrengthof any nonlinearity is directly relatedto thelevel of its excitation; any
nonlinear element approaches linearity as its RF voltage or current approaches zero.
This simplefact canbeusedtoobtainconvergenceindifcult conditions: analyzethe
circuit at somelowexcitation level, then increasethelevel using theprevious results
(perhaps scaled according to theexcitation level) as theinitial estimate. Theanalysis
proceedsinthismanner until thedesiredexcitationlevel isreached. Thisprocesstakes
advantageof thenear-linearity of thecircuit at lowlevels, andusessolutionsat higher
levelsasinitial estimates. Theseexpedientsusually areenoughtoprevent convergence
failure.
Thismethodissometimescalledsourcestepping. It isoneof amoregeneral classof
methodscalledcontinuationmethods, inwhichsomeparameter of thecircuit isvaried
stepwiseandasolutionisobtainedat eachstep. Besidessourcestepping, continuation
methods may include parameters that vary the linearity of nonlinear elements from
nearly linear to their speciednonlinearity. Bothhavebeenusedincircuit simulators;
however, becauseof itseaseandgeneralityof implementation, sourcesteppingismost
common.
It isimportant torecognizethat source-stepping, or other continuationmethods, are
effectiveonlywhenstrongnonlinearityorlargeexcitationcausesconvergencedifculty.
Continuationcannot provideasolutionif theproblemisill-conditioned, say, bytheuse
of apoorlyconceivedself-heatingmodel.
224 Computer-aided design of power ampliers
Frequencyset
The selection of a frequency set can affect convergence. By frequency set, we mean
simply the set of frequency components used in the harmonic-balance analysis. In a
simplesingle-toneanalysis, thefrequency set issimply theset of harmonicsfrom0to
K
p
, where
p
istheexcitationfrequencyandK isaharmonicgreat enoughsothat the
result hasadequateaccuracy. Inamultitoneanalysis, thesituationismorecomplicated.
Theset of frequenciesintheanalysisis
= k
p1
l
p2
m
p3
(5.50)
where
pn
are the excitation tones, and the range of the integers k, l, m, . . . must be
limitedinsomeway. A number of limitingschemesarepossible; for example,
k = K . . . K
l = 0. . . L
. . . .
(5.51)
andsoon. Notethat it isnot necessarytoincludel - 0, asthissimplycreatesharmonic
components that already exist in(5.51). This set is sometimes calledarectangular or
boxtruncation[14], asplottingthe(k, l) pairsonarectangular gridcreatesaboxpattern.
Another possibility, calledatriangular truncation, involves theuseof (5.51) withthe
additional constraint that KL - Q, whereQ is somemaximumorder of themixing
productsof interest. A thirdpossibilityisthemixer set, inwhich
= k
p

0
(5.52)
andkislimitedasinthesingle-tonecase. Thisset isusedfor mixer analysis, inwhich

p
isthelocal oscillator and
0
istheIF.
Although thefrequency set is rightly selected according to theproblemat hand, it
nonethelesshasastrongeffectonconvergence, analysistime, andmemoryuse. Itshould
beobviousthat agreater number of frequencycomponentscausesgreater memoryuse
andslower analysis. Theeffect onconvergencemay belessclear, however. Intuitively,
onemight expect theuseof agreater number of frequency components to providea
better estimateof thesolution, andthusconvergencemight bemorerobust. Infact, the
frequencysethasarelativelyweakeffectonconvergence, andminimizingthenumberof
harmonicswhilesignicantly oversamplinginthetimedomainusually providesbetter
convergence.
Thereasonfor thismildlynonintuitivesituationcanbeilluminatedbyanexamination
of thewayharmonic-balanceanalysisoperates. For simplicity, weconsider asingle-tone
problemlimitedtoK harmonicsandasinglenonlinear function. If theindependentvari-
ables arevoltages (whichis always thecaseinnodal analysis), limitingtheharmonics
toK makesV
k
=0, k>K. Thisisequivalenttosettingall theembeddingimpedancesin
thecircuittozeroatthosefrequencies. Sincethefrequencyspectrumisstrictlybandlim-
ited, the:(t) waveformusedtocalculatethecurrent inthenonlinear subcircuit isvery
clean; that is, unaffectedby aliasingor other Fourier-transformerrors. Thecurrent,
i(t), isthenobtainedfromthenonlinear function, i(t) = f
NL
(:(t)). Thefourier transform
5.6 Practical aspects of nonlinear circuit simulation 225
of thisquantityisnot strictlybandlimitedtoK
p
. If thewaveformisoversampled(i.e.,
the sampling interval is much less than the Nyquist limit), the harmonics of interest
arenot affected signicantly by thosehigher harmonics, so theharmonics k > K are
discarded without incurring error. If the sampling interval is minimal, however, the
lower harmonicscanincur aliasingerrors, aformof numerical noisethat changesfrom
iterationtoiteration, makingconvergencedifcult or impossible.
Terminationcriteria
Earlier we made the point that harmonic-balance analysis is a process of iteratively
improving an estimateof thesolution. At somepoint, that solution is good enough
andtheprocessmust terminate. Howdowedecidewhat isgoodenough?
A number of possibilitiesareimmediatelyevident. Therst issimplytorequirethat
themagnitudeof thecurrent-error vector belessthansomethreshold:
[I
LIN
(V) I
NL
(V)[ - (5.53)
Another isthat theindividual current errorsat eachharmonicbebelowsomethreshold;
that is,
[I
LIN
(k
0
) I
NL
(k
0
)[ - (5.54)
for all harmonicsat all nonlinear elements. Finally, wecouldrequirethat thefractional
error ineachharmonicbebelowsomethreshold:
[I
LIN
(k
0
) I
NL
(k
0
)[
([I
LIN
(k
0
)[ [I
NL
(k
0
)[),2
- (5.55)
wherethedenominator of (5.55) isanestimateof theelement current at that particular
harmonic.
All of thesemethods havepitfalls, introducedby thepresenceof harmonics having
very large and very small magnitudes. In (5.53), it is possible for the error to be
relatively small, while errors in individual weak harmonics, which are often of most
interest, arequitelarge. Suppose, for example, weareanalyzingapower amplier under
multitoneexcitation, andweareespecially interestedinits intermodulationdistortion.
An error limit of, say, = 10
6
would be far too small for the fundamental tones,
and might well prevent successful convergence. That error might still betoo great to
resolveintermodulationtones, however, whichcouldbewell belowthislevel. A similar
problemexistswith(5.54); anabsoluteerror limitmaybetoosmall for larger frequency
componentsyet toolargefor smaller ones.
Thelimit shown in (5.55) has theoppositeproblem. A fractional error of, say, 1%
( = 0.01) might be ne for larger components, such as the fundamental-frequency
output, but it ispointlesstodetermineintermodulationtonestosuchasmall error.
Onesolutiontothisdilemmaistheuseof ahybridcriterion. For example, wecould
examineeach current-error component and determinewhether either (5.54) or (5.55)
is satised. If all components satisfy oneor theother, theproblemterminates. When
226 Computer-aided design of power ampliers
this criterion is used, the absolute error of (5.54) can be specied for weak compo-
nents, andthefractional error of (5.55) for larger components. Thisapproachnaturally
accommodatesbothlargeandsmall componentswithappropriateerror criteria.
5.6.2 SPICE models in harmonic-balance analysis
When general-purpose harmonic-balance simulators were originally developed, few
nonlineardevicemodelswereavailable. TheobvioussourceforsuchmodelswasSPICE,
soSPICEmodelsweresimplytransferredtotheharmonic-balancesimulator. Sincethen,
manymoredevicemodelshavebeendeveloped, andtheoriginal implementationof these
models oftenwas inSPICE as well. As aresult, many suchmodels areoptimizedfor
SPICE, andsimilar time-domainprograms, butarenotwell suitedtoharmonicbalance.
The reason for this situation arises in the differences between silicon monolithic
ICs, for which SPICE was created, and microwave hybrid and monolithic ICs, for
whichharmonicbalancesimulatorsaregenerallyused. SiliconICsconsist primarilyof
nonlinearcircuitelements, withrelativelyfewlinearones; microwaveICsconsistmostly
of linear circuitelements, whichareincorporatedintoasingleadmittancematrix. Many
SPICEdevicemodels, however, consistof manynonlinear elementsandfewlinearones.
Separatingthesemodelsintolinearandnonlinearsubcircuitsoftenleavesisolatednodes,
resultinginanill-conditionedharmonic-balanceJ acobianmatrix.
Often, parts of such models that do not affect microwave circuits are included in
SPICE models, introducing unnecessary computational overhead. A good exampleis
the inclusion of a substrate parasitic transistor in the VBIC BJ T model. Microwave
devices, both HBT and BJ T, do not havethis parasitic. Its effects can beremoved by
using device parameters that turn off the elements, but this practice still may create
isolatednodes, and, inany case, adds useless overheadinanalyzingdead nonlinear
elements.
5.6.3 Problem size minimization and solution optimization
Minimizing the size of the problemcan do much to reduce the computational cost
of a nonlinear analysis. At this point, it should be clear that the size of the problem
is essentially the size of the J acobian matrix, which is proportional to the number
of harmonics and to the number of nodes or ports at which nonlinear elements are
connected. Minimizingeither of thesequantitieshelpstospeedtheanalysis.
We have already touched on the matter of selecting a sensible frequency set and
minimizingmodel complexity. Other technologiescandomuchtominimizethesizeof
theproblemthat must besolved.
Inthepast, nonlinear circuits rarely includedmuchDC circuitry. Intodays RF and
microwaveICs, however, it is common to havefairly complex circuitry, such as bias
circuitry, whichcanbetreatedaslinear. Treatingsuchcircuitsaspart of theRF circuit
iswasteful andunnecessary.
Simulator technologies exist that can identify the parts of the circuit that have no
signicant RF voltageand treat themas linear. Thoseparts can then beincorporated
intothelinear subcircuit, reducingtheproblemsize. Insimulatorshavingthiscapability,
5.6 Practical aspects of nonlinear circuit simulation 227
dramatic improvements in simulation speed are observed; reducing the size of the
nonlinear problemalsocanimprovesimulator robustness.
Harmonicbalancecanbeformulatedsuchthattheindependentquantities, thenonlin-
ear element voltages, areat either portsor nodes. Useof aport formulationminimizes
problemsize, whilethenodal formulationismoreversatile.
Becauseof theimportanceof minimizingproblemsize(andthus, presumably, simu-
lationtime), virtually all early microwavesimulators usedaport formulation. Intime,
however, numerical methodsfor handlinglarge, sparsematricesimproved, allowingthe
nodal formulationtohavesimulationefciencyvirtuallyasgoodasaport formulation.
This evolutionparallels theevolutionof linear circuit simulators, which, intheir earli-
est incarnations, usedport concepts. Today, suchmethods areobsolete, andnodal (or
modiednodal) methodsarevirtuallyuniversal.
Eachtimeanewsimulationbegins, thesimulatormust, inessence, nditswaythrough
avariety of possibilities to obtainasolution. It does this by usingtheJ acobianmatrix
topoint tothedirection, at eachiteration, of thefastest decreaseintheerror function.
Unfortunately, someof thestepsinthat processaregoodonesandsomearenot.
Itispossiblefor thesimulator toremember whichstepsweresuccessful indecreasing
the error function and which werepoor. Then, the information can beused to speed
subsequent analyses. The idea is useful as long as the circuit does not change much
betweenanalyses, whichisoftenthecase; after all, intheprocessof tweakingacircuit
to optimizeit, most circuit modications areminor. Suchmethods canbevery useful
inpower-amplier analysis, where, for example, largepumpedcapacitancesseemtobe
quiteeffectiveinsendingthesimulator off towardplaceswhereit shouldnever go.
5.6.4 Numerical considerations
Muchof harmonic-balanceandtime-domainanalysisinvolvessolvinglinear equations.
This is obvious in the case of (5.18), but even the process of performing a Fourier
transformisinherentlyamatrixoperation.
It iswell knownthat ill conditioning(i.e., near singularity) of thematrixcanresult in
largeerrors. GiventheN-dimensional systemof linear equations
Ax = b (5.56)
theerror inthesolutionvector, x, canbeboundedas
|x|
|x|
(A)
|b|
|b|
(5.57)
where (A) is the condition number of the matrix, and is the maximumnormof the
vector v,
|:| = max[:
i
[ 1 i N (5.58)
Theconditionnumber canbefoundinanyof several ways; see[2].
Equation (5.57) says, in essence, that any fractional variation in b is amplied by
theconditionnumber indeterminingthefractional error inx. In(5.18), theright-side
vector isF(V), whichissubjecttoconsiderablenumerical noiseanderror, asitinvolves
228 Computer-aided design of power ampliers
multipleFouriertransformsandlossof precisionfromextensivecomputation.Thus,poor
conditioningof theJ acobiancaneasilycreateerrorslargeenoughtomakeconvergence
impossible.
Wehavediscussedthesourceof ill conditioningperiodicallythroughout thischapter.
Themostcommonsourcesof ill-conditioningprobablyare(a) poorlyconceivedmodels,
and(b) characteristicsof thecircuit itself. Amongtheformer arepoorlyconceivedself-
heatingmodelsandthelatter includepartsof thecircuitthatwouldbedisconnectedsave
for somelargeimpedance.
Modelsfrequentlyareill-conditioned, inthesensethattheyleadtoanill-conditioned
admittancematrix of thelinear subcircuit or toanill-conditionedJ acobian. Many phe-
nomena that affect conditioning havebeen described earlier (e.g., Sections 5.6.1 and
5.6.2), soit sufcessimplytomakethepoint oncemore.
Ingeneral, nonlinear capacitors arecharacterizedby their charge-voltagefunctions,
Q(V), andthecurrentisfoundbydifferentiationinthefrequencydomain, whichinvolves
simply multiplicationby j. Another approach, whichis useful insomemodels, is to
usethecapacitance, denedas dQ(V),dV. Inthat casetheindependent quantity isthe
voltagederivative, not thevoltageitself.
TheJ acobiansentriesnormally havethemagnitudesof circuit admittances, but this
formulationcreatesentrieswhosemagnitudesarethoseof thecapacitances. Thesesmall
entriescreatenear-zerovaluesinthematrix, whichcauseill conditioning. Thesolution
inthis caseis very simple: scalethecapacitances larger (by afactor of, say, 10
9
) and
maketheindependent voltagesaffectingthemsmaller bythesamefactor.
Itisamusingtonotethatthisproblemhasoccasionallybeenpresentedasafundamen-
tal difculty inusingacapacitiveformulation. However, this simpleexpedient solves
theproblemcompletely.
Inaport formulation, loopsof independent voltages(e.g., threenonlinear capacitors
inapi conguration) cancauseill conditioning. Theproblemarisesfromthefactthatthe
loopvoltagesarelinearlydependent; onevoltageisextraneous, asit canbedetermined
fromtheothers. Thisisthereasonfor therequirementinSPICE thatloopsof capacitors
andvoltagesourcescannot beused.
4
Thisproblemisverydifcult toavoid, asthenonlinear elementsinacircuit areoften
hidden insidemodels, and such loops can becreated without any obvious indication.
Thenconvergenceispoor. Useof anodal ormodiednodal formulationusuallyprevents
thisproblem.
5.6.5 Design ow
A fast, robustsimulatorcannotdomuchtospeedthetaskof designinganamplier if the
designersdevelopment processisfraught withbottlenecks. Thedifferencebetweenthe
fastestandslowestcircuit-simulatorenginesmightmake, atmost, afewweeksdifference
4
SPICEalsoproscribescutsetsof inductorsandcurrentsources, forthesamereason. Thisrestrictionisfound
insimulatorsthat usemixedvoltagesandcurrentsasindependent quantities.
5.6 Practical aspects of nonlinear circuit simulation 229
inaproject of several months, but theuseof acumbersomedesignowmight easily
doubleor eventripledevelopment time.
Onegreat impediment toasmoothdesignowisthecommondivisionof thedesign
task into separate, disjointed efforts, such as initial design, parasitic extraction, EM
analysisof critical circuit elements, andlayout. Eachof thesestagescanuncover aaw
requiringsubstantial redesignat atimewhenmucheffort has already beenexpended.
Integratingthesetasksintoasingle, concurrentowcandomuchtoexpeditethedesign
process. Softwarecanbeanimportant part of improvingthedesignprocess, but it must
bedesignednotonlyfor itsanalytical capabilities, buttosupportanefcientdesignow
aswell.
AstheRF/microwaveindustryevolvesfromalmostcompletelymilitaryandaerospace
functions to more commercial ones, with more stringent cost requirements and tight
schedulesdrivenbytheneedforashorttimetomarket, thesoftwareindustryisbecoming
awareof theneed for human as well as analytical functionality. This is aworthwhile
development.
Theintegrationof EM analysis softwarewithcircuit-analysis softwareserves as an
exampletoillustratethisphenomenon. Evenwell intothelate1990s, it wascustomary
for EMsoftwaretoruninbatchmodeononeor morepowerful computers. Thedesigner
oftendidnot knowwhat partsof acircuit hadtobeEM simulateduntil acircuit layout
wascompleted. Then, thecircuitelementstobesimulatedhadtoberedrawnfor theEM
simulator, asthelayoutandEMsoftwaresgraphicsmoduleweregenerallyincompatible,
withconcomitantrisksof error, andthecircuitelementsnallysimulated. Theseresults
werefedback intothecircuit-analysissoftwareandthecircuit resimulated. Theresults
of theresimulation often werenot acceptable, so redesign was necessary, along with
another loopthroughthesimulation-layout-EM process.
Animprovement camewithcosimulation, theabilityof disparatetoolstorunsimul-
taneously and to sharedata. This could bedonein anumber of ways, supported to a
greater or lesser degreebythecomputersoperatingsystem. Examplesof thelatter were
Unix pipes andMicrosoft Windows dynamic data exchange(DDE) capability. These
allowedacertaindegreeof interprocesscommunication, especiallythedirect transmis-
sionof databetweensimulatorswithouttheusersintervention. Whilethesetechnologies
allowedinputandoutputdatasharing, theydidnotconstitutefull integration, asthetypes
of interactionwerelimited.
Today, full integrationof circuitsimulationwithEM, layout, andevensystemsimula-
tionispossible. Thiscapabilityissupportedinpart byoperatingsystemtechnology, in
particular Microsoft Windows component object model (COM) technology, astandard
that allows separatesoftwaremodules to beintegrated at theobject-code level. As a
result, simulatorscanbefully awareof theoperationof other simulatorsandobtainall
theinformationthey needabout theeffect of onesimulator ondatathey deal with. For
example, if theuser changesthedimensionsof amicrostripteejunction, thatchangecan
bereectedinstantlyinthelayout, andtheEM simulator becomesawareof thechange
andcanresimulatethejunction. Inthisway, multipleiterationsthroughthesimulation-
layout-EM loop areavoided, and many sources of error involved in copying dataare
fullyeliminated.
230 Computer-aided design of power ampliers
However valuable these technologies are, they do not let the user off the hook.
Engineeringjudgment intheir useis still essential. For example, it is not yet possible
(althoughit may be, eventually) for softwaretodeterminetheoptimumsetup, interms
of frequency sets and termination criteria, for aharmonic-balanceanalysis. Theuser
must understand the matter and make appropriate decisions. Similarly, some circuit
elements can be modeled more easily and more accurately than others; for example,
straight microstrip transmission-line models are invariably much better than closed-
formmicrostripjunctionmodels. Suchelementsshouldbeusedpreferentially, wherever
possible, inmicrowavedesigns. Finally, theuser must beawareof theeffect of certain
kinds of model on analyses. The interpolation of S-parameters, for example, can be
critical for various types of analysis: linear interpolation can result in passband plots
havingascallopedappearanceandthelackof smoothnesscancauseconvergencefailure.
Earlier, wedescribedtheproblemof S-parametersthat donot spanthefrequencyspace
used in a nonlinear analysis. This also can and usually does cause convergence
difculty.
References
1. N. BalabianandT. A. Bickart, Electrical NetworkTheory, Wiley, NewYork, 1969.
2. G. Dahlquist andA. Bj ork, Numerical Methods, EnglewoodCliffs, NJ : Prentice-Hall, 1974.
3. J . VlachandK. Singhal, Computer Methods for Circuit AnalysisandDesign, SecondEdn.,
Norwood, MA: ArtechHouse, 1994.
4. M. S. NakhlaandJ . Vlach, A piecewiseharmonic balancetechniquefor determinationof
periodicresponseof nonlinear systems, IEEE Trans. Circ. Syst., vol. CAS-23, p. 85, 1976.
5. S. W. Director andK. W. Current, Optimizationof forcednonlinear periodiccurrents,IEEE
Trans. Circ. Syst., vol. CAS-23, p. 329, 1976.
6. F. R. Colon and T. N. Trick, Fast periodic steady-stateanalysis for large-signal electronic
circuits, IEEE J. Solid-StateCirc., vol. SC-8, p. 260, 1973.
7. K. S. Kundert and A. SangiovannIVincentelli, Simulation of nonlinear circuits in the
frequencydomain, IEEE Trans. Computer-AidedDes., vol. CAD-5, , p. 521, 1986.
8. H. YeagerandR. W. Dutton, Improvementinnorm-reducingmethodsforcircuitsimulation,
IEEE Trans. Computer-AidedDes., vol. 8, p. 538, 1989.
9. G. B. Sorkin, K. S. Kundert, and A. Sangiovanni-Vincentelli, An almost-periodic Fourier
transformfor use with harmonic balance, IEEE MTT-S Int. Microw. Symp. Dig., p. 717,
1987.
10. V. Rizzoli, C. Cecchetti, andA. Lipparini, A general-purposeprogramfor theanalysis of
nonlinear microwavecircuits under multitoneexcitationby multidimensional Fourier trans-
form, Proceedingsof the17thEuropeanMicrowaveConference, 1987.
11. E. Ngoya, J. Rousset, M. Gayral, R. Quere, andJ. Obregon, Efcient Algorithmsfor spectra
calculationsinnonlinear microwavecircuitssimulators, IEEE Trans. CircuitsSyst., vol. 37,
p. 1339, 1990.
12. P. Rodrigues, A general mappingtechniquefor fourier transformcomputationinnonlinear
circuit analysis, IEEE Microw. GuidedWaveLett., vol. 7, p. 374, 1997.
13. P. Rodrigues, Anorthogonal almost-periodic fourier transformfor useinnonlinear circuit
simulation, IEEE Microw. GuidedWaveLett., vol. 4, p. 74, 1994.
References 231
14. K. S. Kundert, J. K. White, andA. Sangiovanni-Vincentelli, Steady-StateMethodsfor Simu-
latingAnalogandMicrowaveCircuits, Boston: Kluwer, 1990.
15. J. C. PedroandN. BorgesdeCarvalho, Articial frequencymappingtechniquesformultitone
harmonicbalance, IEEE MTT-SInt. Microw. Symp. Dig. Workshops, 2000.
16. V. Boric, J . East, and G. Haddad, An efcient Fourier transformalgorithmfor multitone
harmonicbalance, IEEE Trans. Microw. TheoryTech., vol. MTT-47, p. 182, 1999.
17. V. Rizzoli, A. Neri, and F. Mastri, A modulation-oriented piecewise harmonic-balance
techniquesuitablefor transient analysisanddigitallymodulatedsignals, Proceedingsof the
26thEuropeanMicrowaveConference, 1996, p. 546.
18. E. Ngoya, J . Sombrin, andJ . Rousset, Simulationdecircuitsetsystemes: methodes, actuelles
et tendances, SeminaireAntennesActives-MMIC, Arles, Arles, France, 1994.
19. E. NgoyaandR. Larcheveque, Envelop[sic]transientanalysis: anewmethodforthetransient
andsteady-stateanalysis of microwavecommunicationcircuits andsystems, IEEE MTT-S
Int. Microw. Symp. Dig., p. 1365, 1996.
20. D. G. Swanson and W. J. R. Hoefer, Microwave Circuit Modeling Using Electromagnetic
FieldSimulation, ArtechHouse, Norwood, MA, 2003.
21. J . Rautio, Synthesisof compactlumpedmodelsfromelectromagneticanalysisresults,IEEE
Trans. Microw. TheoryTech., vol. MTT-55, p. 2548, 2007.
6 Practical HF/VHF/UHF RF power
amplier realization
Daniel P. Myer
Communication Power Corporation (CPC)
6.1 Introduction
Thisdiscussionfocusesonthepractical realizationof radiofrequencypower ampliers
(RFPAs),theprocessthatexistsbetweennothingandsomething,thepathanRFPAdesign
engineer can takefromtheRFPA application conceptual phaseto theconstruction of
actual hardware.
Since theend useapplication/market denes and drives theneed for an RFPA, an
overviewof major applicationareas is coveredinitially. Theapplications will demand
that certainRFPA specicationsaresatised, soanoverviewof genericamplier speci-
cationsrelativetoseveral applicationsislogicallyprovidednext. Thespecicationsare
viewedfromseveral vantagepoints, i.e., they arecoveredqualitatively, inother words,
for aparticular application, whichRFPA specicationsaremost relevant. Thentheyare
dened quantitatively, not so much for a unique application, but with a bias towards
theRFPA itself, withaneffort to provideaguidelineas to what constitutes arealistic
specicationvalue, andwhat doesnot.
Thechapter will end off with adesign examplethat originates with ahypothetical
applicationandusestheconceptspresentedtogenerateaspecicationandRFPA module
designtosatisfytherequirement. Again, thisisachapter onpractical realization, while
itwill cover sometheoretical aspectsof RFPA design, itwill alsocover howtoconstruct
theamplierandemphasizetestconguration/dataanalysis. Thegoal istohelpfacilitate
adesignthatcannotonlybemanufacturedonce, butinvolumewithdesignmargin, and
protability.
6.2 RF power amplier markets
Thereareseveral majormarketsorenduseapplicationareasof RFPAs,themorecommon
onesare:
r
military;
r
medical;
r
scientic;
r
industrial;
r
commercial.
6.3 The realization process 233
Within thesebroad areas thereexist ahost of uniqueapplications wheretheRFPA
ndsahome:
r
Military: communications, communication jamming, improvised explosive device
(IED) J amming, countermeasures, radar, andpsychological warfare;
r
Medical:magneticresonanceimaging/spectroscopy(MRI/MRS),thermotherapy,car-
diactissueablation, benignprostatichyperplasia(BPH) treatment, RF cauterizing;
r
Scientic: nuclear magneticresonance(NMR) spectroscopy, nuclear quadrupoleres-
onance(NQR), electronparamagneticresonance(EPR)
r
Industrial: electromagnetic compatibility testing(EMC), RF heating/drying, adhe-
sivecompoundcuring;
r
Commercial:semiconductorwaferplasmaprocessing, cellularbasestations, AM/FM
radio, HDTV broadcast.
All theseapplications haveonethingincommon; anRFPA is essential andvital to
theprocess. ButjusthavinganyRFPA, will notsufceeither, i.e., anRFPA designedfor
MRI couldnotbeusedeffectivelyfor HDTV broadcastandviceversa. Eachapplication
carries withit auniqueset of requirements or specications that theRFPA, if it were
to beused successfully, must meet. All RFPAs havecertain operating characteristics,
and for each uniqueapplication, somearemoreimportant than others. In asatellite-
based RFPA, for example, efciency is very critical dueto limitedpower availability,
ontheother handanMRI RFPA specicationismorefocusedonlinearity andlesson
efciency.
For the applications listed above, a brief overview of some of the required RFPA
specicationsfor eachdisciplineisprovidedinthefollowingparagraphs.
6.3 The realization process
RFPA realizationis, inabroadsense, athree-stepprocess:
1. RFPA qualitativespecication delineation: A task concernedwithassessingapar-
ticular proposedapplicationthatrequiresanRFPA anddistillingoutof themultitude
of existingamplier performancespecications, only theones necessary andsuf-
cient(inotherwords, requiredspecications) thatoncequantied, will denealistof
amplier characteristicstosatisfytheenduseapplication. For example, consider the
applicationof RF heating. RFPAsareusedinthisrequirementtoheatmaterials(plas-
tics, humantissue, etc.) withitsoutput power. Clearly, thereisnoneed, for instance,
for goodphaselinearity, thematerial tobeheatedwill not respondanydifferentlyto
anRFPA that hasexcellent phaselinearity thanwithonethat doesnot. Thisstepof
qualitativespecicationdelineationisby nomeansatrivial processandisexecuted
best by maintaining a close working relationship between the RFPA engineer and
thesystemengineer intimatewith thegiven application. Simply becausean RFPA
Engineer knows, for example, how to design an amplier with excellent linearity
does not necessarily meanheis awareof what distortiontypes toassess inorder to
234 Practical HF/VHF/UHF RF power amplier realization
enablegoodpicturequalityfromaTV transmitter. TheTV systemengineerwill know
theseparametersbasedonhiseldexperience. Sincetherearemanyapplicationsfor
RFPAs, thisstepisonlycoveredhereinacursorysense, suchthattheRFPA Engineer
isawareandinformedof itsimportance.
2. RFPA specication quantication: Withalist of requiredRFPA specicationsrel-
ativeto agivenapplicationassembled, thenext stepis to assignnumeric values to
each particular one along with a clarication of what the numbers represent; i.e.,
minimum, maximumor typical. Anarrival at aparticular number for acertainspec-
icationusually is theresult of eldtrials wheresystemperformanceis monitored
whileonly theamplier parameter of interest is successively degradedto thepoint
whereitbecomesevidentintheoverall systemperformance. AnRFPA Engineermust
bevery familiar withamplier specications not only for what they imply but also
for whatvaluerepresentsarealisticapplicationdemandandonethatisnotphysically
or economically realizable. Therefore, anoverviewanddenitionof RFPA speci-
cations is providedalongwithhighlightingbothideal andtypical values toprovide
theRFPA Engineer withaframeof referenceor boundaries fromwhichtowork in
duringtherealizationprocess. Thetypical values arewhat onemight expect to get
fromagenericRFPA without anyformof error correctionapplied.
3. RFPA hardwarerealization: Followingthespecicationdelineationandquantica-
tion, adesignexampleis providedto illustratemethods to physically realizeRFPA
hardware. Thisstepwill denethebasicarchitectural componentsof anRFPA mod-
ule: theRF transistor, matchingnetworks, feedback networks, DC bias andsupply
networks. AnRFPA moduledesignthatcanbeappliedtoseveral applicationswill be
shownasadesignexample.
6.3.1 RFPA qualitative specication delineation
Military ampliers: For just about any application in themilitary or defensemarket,
oneamplier characteristic is paramount: reliability. RFPAs for military applications
must endureextreme, harsh environmental conditions (temperature, altitude, salt fog,
exposuretoaviationfuelsandsand, etc.) andperformawlessly. Intheevent that they
doencounter somepartial malfunction, thentheyneedtobebattlesure. Thisimplies
theRFPA must providesomelevel of performanceand beexpected to overridefault
protectionshutdowncircuitryinspiteof thefact that it maybedamaged.
Military communications requireanRFPA tobeextremely broadband, adverseload
VSWR tolerant, andif themodulationformat demands it, haveexcellent linearity. For
applications whereanRFPA is battery powered, highlevels of amplier efciency are
needed.
J amming/countermeasures applications require extreme broadband capability and
antennaloadVSWR tolerance. Linearityisnot overlyimportant asthegoal maysome-
timesbetooverpower enemycommunicationsanddistortioncomponentsmayactually
augment theprocess.
6.3 The realization process 235
Radar RFPAs will need to work well in thetimedomain and maintain good pulse
delity; fast riseandfall transitiondurations (formerly knownas rise/fall times), low-
pulsetilt (amplitudedroop), andhaveandlowlevelsof pulseovershoot andringing.
Ampliersusedforpsychological warfareoperations(Psy-Ops) aregenerallyusedfor
emulationof enemy civilianandmilitary communicationsystems whichcanbeeither
AM/FMradioortelevisionbroadcast. Inthisevent, RFPAswill needextremebandwidth
to cover multiple channels along with high linearity and very low intermodulation
distortion(IMD).
Medical Ampliers: Ampliers for themedical market needto bereliableas well;
however, theenvironment they areexposedto is very benign. Usually RFPAs inmed-
ical applications are located in hospitals or research institutions where the ambient
temperatureremainsat approximately25

C.
MRI andMRS, whichprovidedetailedanatomical andmetabolicallyproledhuman
images, demand the RFPA deliver certain levels of performance in three domains:
time, power andfrequency. Inthetimedomain, theRFPA must deliver highlevels of
pulsedelity (fast risingandfallingtransitiondurations, low-pulsetilt, lowovershoot
andringing). Inthepower domain, theRFPA must exhibit lowlevels of AM/AM and
AM/PM (gainandphaselinearity, respectively) distortion. Inthefrequencydomain, the
MRI/MRS RFPA must deliver uniformperformanceat several key frequencies, while
havinglownoiseoutputatfrequenciesotherthanthecarrier, specicallywhiletheRFPA
istransmittingthecarrier.
For medical heating (thermotherapy, cardiac tissue ablation and benign prostatic
hyperplasia treatments), the focus is on precise power control. To heat human tissue
safely, a feedback control (ALC) loop is the best method for keeping the RF power
output variations extremely low. Linearity is not necessarily critical but does enable
theRFPA to bemoreeasily controlled by feedback loops. Medical heating is usually
narrowband(i.e., 915MHz/ 5MHz), andwhiletheloadVSWR maybeharsh, the
RFPAsoutput canbeprotectedwithCirculators.
Scientic Ampliers: NMR spectroscopy employs all the same principles as MRI
except, instead of analyzing patients, an NMR Spectrometer will evaluate chemical
compoundsor pharmaceuticals. NMRessentiallymakesthesamedemandsonanRFPA
that MRI would, however an amplier for NMR/NQR and EPR will require more
precisepulsedelity(faster rising/fallingtransitiondurations, lower droopandvirtually
nonexistent pulseringing/overshoot).
I ndustrial Ampliers: EMC RFPAs must provideRF power over ultrabroadband-
widthsspanningseveral octavesfromtheaudiofrequencyrangeupintothemicrowave
frequency range (10 kHz to over 1 GHz). EMC RFPAs are used to test the radiated
susceptibility of electronic products. Thisistheproductsability tomaintainnormal
operating functions while being subjected to external RF radiation. The EMC RFPA
output will befedinto widebandantennas whichwill radiateRF energy into products
under evaluation.
Material heatingandCompoundcuringdemandthat theRF power bepreciselycon-
trollableandstableover temperatureandtime.
236 Practical HF/VHF/UHF RF power amplier realization
Commercial ampliers: Ampliers for semiconductor wafer processing must be
able to withstand severe load VSWR. While these RFPAs usually have a matching
network which works to actively match theimpedanceof theplasma, therestill exist
severe transient load VSWR excursions. These ampliers must also have extremely
reliableperformanceasanamplier failurecanleadtoshuttingdownwafer fabrication
productionlines.
Cellular basestationRFPAsmusthandlemultiplecarrierssimultaneously. Duetothis
requirement, theseRFPAs must beextremely linear suchthat IMD distortionis mini-
mized. Typically, theIMDrequirement ismuchlower thancanbeachievedwithastand
alone, un- correctedRFPA. Therefore, error correctionschemes suchas pre-distortion
and feed-forward are employed to reduce IMD components below required limits.
AM radio transmitters require good linearity while FM transmitters have a stronger
emphasis on efciency and low cost. Conventional television transmitters place high
demandsonsignal linearity.
Reviewingtheapplications, itisapparent that certainRFPA operatingparametersare
instrumental indeninganRFPA for oneapplication, but areabsent inanother while
somearecommontoall.
6.3.2 RFPA specications, generic list and quantication guidelines
TheRFPA specicationssimplydene, quantitatively, themanner inwhichaparticular
RFPA must behaveunder agivenset of conditions. Theconditions areinput stimulus
signal characteristics, expectedoutput signal responseperformance(transfer function),
outputloadVSWR, supplyvoltages, environmental (operating/storage) conditions(tem-
perature, shock, vibrationandaltitude).
A listof genericRFPA specicationsisprovidedbelowwithdenitionsaccompanied
by quantications of what anideal amplier woulddeliver, followedby what atypical
onemight provide. Thetypical values represent what is readily accomplished with a
genericbroadbandamplier, themorearequirement demandsaparticular specication
quanticationtopropagatetowardsanideal value, themoredifcult(andcostly) it will
betodesignandmanufacturetheamplier. Thequanticationsarenotindicativeof any
particular enduseapplication, thegoal istoprovidetheRFPA engineer withacommon
sense, afeel for what isreadilyattainable, andwhat isnot.
r
Power output(units: W): theamountof power anRFPA candeliver intoaparticular
load VSWR, over a given frequency and dynamic range. What is ideal: theRFPA
would deliver theprecise amount of power demanded by a particular application,
This power level wouldbeexact, proportionallycontrollable, withzero power level
uctuationsduetotemperature, component variationandfreeof noiseanddistortion
components. What istypical: apower level that variesinaquasi linear fashionand
drifts a fewwatts for every fewdegrees shift in ambient temperature along with a
power spectrumof noiseanddistortioncomponents.
r
Frequency response/range(units: Hz): the range of frequencies that an RFPA is
expectedto uniformly meet all specications. What is ideal: a rangeof frequencies
6.3 The realization process 237
weretheRFPAexhibitsawlessuniform(identical timeandpower domainresponses)
performance(stablepower, no distortionor noise), outsideof this rangetheampli-
er has no responseto any other frequencies. What is typical: a rangeof frequen-
cies where performance is not uniform; i.e.: gain will vary by several dB, distor-
tionandnoisewill bepresent, riseandfallingtransitiondurations varyalongwith
efciency.
r
Gain(units: dB): themeasureof howmuchgreater inpower anRFPA will increase
thelevel of an input signal. What is ideal: theoutput power is an exact, constant,
linear multipleof theinput power that isindependent of frequency, temperatureand
drivelevel. Whatistypical: gainvariationof 15dBacrossagivenbandwidth, ashift
invalueof afewtenthsof adBfor everyfewdegreeschangeinambient temperature.
r
Gainatness(units:/dB): theamount of gainvariationover thespeciedfre-
quency range. What is ideal: absolutely no (/ 0 dB) of gain variation over the
requiredfrequencyrange. What istypical: dependingontherequiredbandwidth, the
gainatnesscanvaryabout anominal valuefrom/ 0.5dBto/ 4dBor more.
Gainatnessisverydifcult tomaintainover widefrequencyranges(>1octave).
r
Gain linearity/AM-AM distortion (units:/dB versus a specied power
domaindynamicrange): theabilityof anamplier toholditsgainconstantthrough-
out theapplicationof anRF input signal withvaryingpower levels. What is ideal:
theRFPAs nominal gainvalue(gain= G dB/ 0dB) remains perfectlyconstant
fromanoutput power of 0Wto themaximumpower demandedbytheapplication.
What is typical: gain variations of/ 1 dB arereadily achieved by Class A and
AB ampliers, over adynamic rangeof 4060dB. Gainlinearityisclassdependent
withClass A beingthemost linear andleast efcient, whileClass D/E arethemost
nonlinear but most efcient.
r
Gaintemperaturestability(units:/dB): theabilityof anRFPA toholditsgain
constant over varying levels of ambient temperature. What is ideal: absolutely no
(/ 0 dB) of gain variation regardless of ambient temperature variations. What
is typical: gain variations of 6 dB over temperatureswings of 10 to80

C are
common. Thevariationsareeasilycorrectedfor withALCcontrol loopsor openloop
gainstabilizationnetworks. Gainstabilityof/ 0.25dBover 4050

Cisachievable
with open loop temperaturecompensation networks. ALC loops can improvethese
valuesfurther.
r
Phaselinearity/AMPMdistortion(units:/

versusaspeciedpowerdomain
dynamicrange): insertionphaselinearity or AM toPM distortion, istheability for
anRFPA toholditsinsertionphaseconstantovervaryingoutputpowerlevels. Whatis
ideal: aninsertionphasevariationof / 0

fromzeropower output tofull rated


power. Whatistypical: aninsertionphasevariationof/ 10to/ 15

arounda
nominal insertionphasevalue() over a4060dBdynamicrangeiseasilyachieved
viaClassAampliers. Predistortedandfeed-forwardamplierscanhavemuchless
phasevariation.
r
Dynamicrange(units: dB):therangeof outputpowerlevelsthatanRFPA mustwork
over. Usuallythemaximumpower outputistheupper limit. Whatisideal: theRFPAs
output wouldbelinearlycontrollablewithnononlinear deviationsfromexactly0W
238 Practical HF/VHF/UHF RF power amplier realization
totherequiredmaximumratedpower. What is typical: thelinear classes of RFPAs;
A, AB, and B offer thebest dynamic range(approx 4060 dB), Class C, D, and E
havelimiteddynamic ranges (-1015dB uncorrected; i.e., without pre-distortion).
Thelowendof thedynamic rangeis limitedbyit noiseoor, thehighendwouldbe
boundedbyunacceptablelevelsof nonlinear gaincompressionor expansion.
r
Efciency(units:%):theamountof DCinputpoweranRFPA will requiretoproduce
agivenamount of RF output power. What is ideal: theoretical maximumefciency
for thespecic amplicationclass. What istypical: ClassC, D, andE offer thebest
methodsfor highefciency, >5070%andrelativelyconstantover alimiteddynamic
range; however, linearity (both gain and phase) will suffer. Broadband Class AB
efciency runs in the 40%range, but drops substantially at reduced power output
levels.
r
Risetime(risingtransitionduration)(units:s): theamountof timeittakesanRFPA
to progress from10 to 90%(in voltage) of any given rectangular RF pulseoutput.
What isideal: theampliersoutput power risingtransitiondurationisexactlyequal
intimeto theRF input signals risingtransitionduration, regardless of howfast or
slow. What istypical: risetimesof 250750nsarereadilyaccomplished.
r
Fall time(fallingtransition duration) (units: s): the amount of time it takes an
RFPA to progress from90%to 10%of any given RF pulseoutput. What is ideal:
theampliersoutput power fallingtransitiondurationisexactlyequal intimetothe
RF input signalsfallingtransitionduration, regardlessof howfast or slow. What is
typical: fall timesof 50500nSarereadilyaccomplished.
r
Pulseovershoot(units: %): theamountanRFPAsoutputdeviatesfromanexpected
100%output power valueduringtheperioddirectlyfollowingtherisetimetransition
duration. What is ideal: 0% overshoot, the ampliers output exactly follows the
input. What is typical: 1015%overshoot is common though controlling overshoot
becomes more problematic with faster rising transition duration times along with
lower operatingfrequencyrangeintothe130MHzrange.
r
Pulsedroop(pulsetilt) (units: %): theamount an RFPAs output either decreases
(negative tilt) or increases (positive tilt) across the duration of a rectangular RF
output pulse. What isideal: 0%, anRFPA withaperfectlyat rectangular RF input
pulse delivers an amplied exact replica on the output. Note: pulse tilt can be an
extremelyelusivepulseparameter todenelet alonedesignfor or evenmeasure. The
most problematicissuewithpulsetilt isthepinpoint assignment of the100%power
amplitudediscretelocationontherectangular RF pulseenvelopewhichwill serveas
thereferencepoint. Pulsewaveformscanmanifestthemselvesinaninniteamountof
subtlydifferent shapes, evenif theyareall classiedasrectangular pulses. What
istypical: apulsetilt valueof 10%iscommon. Lessthan5%becomesverydifcult
tomanageover broadfrequencyanddynamicranges.
r
Ringing/settlingtime: (units: seconds): Thedurationof timethat anRFPAsoutput
overshootsandexponentiallydecayssinusoidalydowntoa100%pulsepower output.
Whatisideal: absolutelynoringingor overshootwhichmightinitiateringing. Whatis
typical: dependingonthefrequencyrangeandrisingtransitiondurationtime, ringing
canoccur andlast for 20500nanosecondsor perhapslonger.
6.3 The realization process 239
r
Distortion,harmonic(units: -dBc): thelevel of unwantedsignal componentswhich
areinteger multiples of theRF input signal frequency that aremeasuredrelativeto
themagnitudeof theRF output signal. What is ideal: absolutelyno harmonics, the
onlyoutput of theamplier isareplicaof theinput RF spectrum, withall frequency
components amplied exactlythesame. What is typical: even-order harmonics are
less than 20 dBc at thesecond-order harmonic and decreasing further at higher
even-order frequencies,Odd-order harmonicsarelessthan 12dBcatthethird-order
harmonicanddecreasingfurther at higher odd-order frequencies.
r
Distortion,inter-modulation(units: dBc): thelevel of unwantedsignal components
that arisefromtheapplicationof twoor moreRF input signalspropagatingthrougha
nonlinear medium. Thedistortionsignalsarecloseinfrequencytotheoriginal input
signals. What isideal: therearenoIMDcomponents, theoutput frequencyspectrum
isanexact, ampliedreplicaof theinput spectrum. What istypical: IMDdistortion
componentswill typicallyfall 20to 30dBcdownfromthetwo-toneoutputsignals.
r
Noiseoor (units: dB relativeto thermal noiseoor): the amount of noise an
amplierputsoutwhenitsinputisterminatedwitha50Oresistor.RFpowerampliers
typicallyarenotconcernedwithnoiseastheirprimarytaskistoprovidelargeamounts
of electrical energy. Inother words, theamplier, per se, canbeasubstantial noise
source. Thereare, however, situationswhereanRFPA mayberequiredtoemitaslittle
transmittednoiseaspossibleat frequenciesother thanthecarrier. What isideal: the
ampliers noiseoutput is zero dB over thenoisepower of a 50 resistor. What is
typical: anoiseoutput level of 1015dBabovethethermal noiseof a50 resistor.
r
VSWR, input (units: dimensionless): a measureof theRFPAs ability to keep its
input impedancecloseto aspecic value(i.e., 50O) over agiven frequency range
so as to achieveamaximumtransfer of power fromasignal sourceto theamplier
input. Whatisideal: a1:1VSWRisaperfectmatch, all thesignal sourcespower will
enter intotheinputportof theRFPAfor all frequencies. Whatistypical: a2:1VSWR
(or less) over abroadfrequencyrangeiscommonlyacceptable. Usuallytheinput of
anRFPA requiresasmall amount of signal power (ontheorder of 0dBm), soa2:1
VSWR corresponds to approximately 90%of thesignal sourcepower entering into
theRFPAsinput port.
r
VSWR,load(units: dimensionless): ameasureof theRFPAsoutputimpedancerel-
ativetoagivenload. AnRFPAsoutputimpedanceisadynamicparameter depending
on avariety of variables; power output, supply voltageand frequency. It is adesir-
ableto matchtheRF transistor to its loadimpedancefor maximumpower transfer.
Unfortunately, for manyapplications, theloadwill varywidelyandpresent aserious
challengetotheRFPA Engineer todesignanRFPA that canwithstandadverseload
VSWRs and maintain specied performance. What is ideal: a perfect match, 1:1
VSWRfor all frequenciesandpower levels. What istypical: thisdependsheavilyon
theenduseapplication, but canvaryanywherefromaclosematch1.2:1toanopen
or shortedload(:1).
r
Stability,spuriousoutput,loadpull dependent(units: -dBc): thisdenesanampli-
ers ability to maintain stable operation (i.e., not generate any unwanted spurious
signals and maintain an output power that remains controlled by the input power
240 Practical HF/VHF/UHF RF power amplier realization
andastabletransfer function) whiletheinput/output loadVSWR is varied. This is
sometimesdenedasloadpull stability. A wordof cautionheretothosewhospecify
or havebeenrequestedto designanamplier as unconditionally stable, implying
the amplier will not oscillate terminated by any input/output load VSWR. While
theredoesexist waystotheoreticallyshowanamplier isunconditionallystableinto
adverseloads, it is strongly argued herethat no such thing as a truly uncondi-
tionallystableamplier hasor ever will exist in thereal world. Thebasisfor this
argument is that any physical amplier whether broadband or narrowband, can be
subjectedto virtually innitenumber of operatingpoints, assembly process control
variations, devicelot/datecodevariationsandchangingenvironments, beitfrequency,
output power levels, operating temperatures, modulation formats, input/output port
isolationvalues andcombinations of complex input/output loadterminations. Tobe
certain, therewill beonecombination of thelistedoperating points that will cause
anoscillationor somelevel of spuriousoutput. Sinceaninniteamount of operating
points exist, it would require an innite amount of time to test and verify a given
amplier is unconditionally stable, unfortunately (or perhaps fortunately, at least
for thepoor soul taskedwithtestinganRFPA for unconditional stability) noonecan
livelongenoughto test andconrmthis. What is ideal: unconditional stability, no
oscillationsfor anyconditionof operation. What istypical: conditional stability, the
amplier will bestableunder adened, discretesetof conditionssuchasLoadVSWR
uptoagivenpoint (i.e., 3:1, fullyrotational about theSmithChart), dynamicrange,
xedfrequencyranges, or atanoutputfor spuriousfrequenciesthatareanacceptable
amountbelowthecarrier. Typical valuesof loadpull spuriousare 40to 60dBc.
r
Operatingtemperaturerange(units: degrees): thetemperaturerangeover which
theamplier will beexpectedtomeetall specications. EverycomponentinanRFPA
will haveelectrical characteristics that aretemperaturedependent. This dependency
can cause an RFPA to be specication compliant at one temperature and out of
specicationat another. Thekey istodesigntheRFPA sothat theeffectsof varying
temperature are minimized. What is ideal: the RFPA will operate uniformly at all
temperatureswithnovariationingain, output power, distortion, etc. What istypical:
theRFPAwill operateover alimitedtemperaturerange(50to125

C, for example)
wherethehighendofthetemperaturerangeisdeterminedbytheRF power transistors
junctiontemperatureandrequiredfailurerate.
r
Altitude(units: feet, metersabovemeansealevel (AMSL)): thealtitudes within
which the amplier is expected to meet full specication. Altitude can impact an
RFPAs performanceprimarily withones that useforcedair coolingas air becomes
thinner arehigher altitudes, renderingtheefcacyof thiscoolingmethodproblemat-
ical. What isideal: anRFPA isoperational fromSubmarinetoOuter Spacewithno
variationinperformance. What istypical: theRFPAislimitedtocertainaltitudesby
themethodbywhichheatisremoved, for lowaltitudesforcedair coolingisadequate,
for highaltitudesandspace, liquidcoolingbecomesmoreeffective.
r
Shock/vibration(units: G, rms): thelevel of six axes (x, y, z and rotational: yaw,
pitch, androll) mechanical energy anRFPA canwithstandandstill bespecication
compliant. What is ideal: theRFPA canwithstandexposureto shock andvibration
6.3 The realization process 241
resultingfromtransportation(shipping) andapplication(militaryapplications: air-
borne, groundtransport, colocatedordinanceetc.) andremainfunctional regardless
of what level of mechanical shock is imparted. What is typical: anRFPA will have
set limits of howmuchshock theunit canendure, ampliers havebeendesignedto
withstandasmuchas5000Gof transient shockandasmuchas60Gof continuous,
random, sixaxisvibration.
r
Conducted/radiatedemissions: thelevel of unwantedRF noisewhether signal, spu-
rious, distortionor other thatgetsoutof theRFPA either onphysical wires(conducted
emissions) or throughanimproperly shieldedenclosure(radiatedemissions). What
is ideal: theRFPA will containall radiofrequencyenergywithintheconnes of the
physical housing of the amplier system. The RF energy will only exit the chassis
through coaxial cables and connectors. What is typical: all RF power ampliers
will emit and conduct somelevel of RF energy unintentionally to other collocated
circuits, subsystems, andequipment, shieldingmeasuresmust bedeployedsuchthat
theemissionsarewithinacceptablelimitsbasedontheparticular applicationEMC
guidelines.
r
Conducted/radiated susceptibility: the ability of an RFPA to maintain its speci-
edperformancewithunwantedsignal/spurious/noiseenergy fromother collocated
circuits, subsystemsandequipment gettingintoitsonphysical wires(conductedsus-
ceptibility) or throughimproperlyshieldedRFPA enclosures(radiatedsusceptibility).
What is ideal: anRFPA canoperatenormallyregardless of beingsubjectedto any
level of electromagnetic(EM) interferenceor impulse. What istypical: All RF power
ampliers will beaffected at somelevel of RF energy unintentionally coupled into
it fromcollocated circuits, subsystems and equipment. Shielding measures must be
deployedsuchthat thesusceptibilitythresholdsarewithinacceptablelimitsbasedon
theparticular applicationEMC guidelines.
r
Mean time to failure (units: hours): the average amount of time an amplier
will function before experiencing a malfunction or failure. What is ideal: a par-
ticular amplier will be operated within its specied ranges and performaw-
lessly indenitely. What is typical: it depends heavily on the temperature of the
RFPAs semiconductor or die temperatures, MTTFs of 20,000 to 100,000 h are
common.
6.3.3 Specication/hardware realization
Regardlessof theapplication, aspecicationdeninganRFPA will drawfromsomeor
all theabovelistedOperatingSpecications. Howtheseparametersarespeciedclosely
inuencehowaparticular RFPA designisrealized.
Therearemultitudesof applicationsforRFPAs, itisnearlyimpossibletoillustrateone
particular methodto realizeanRFPA designfor each. However, it is possibletocover
techniquesthat canaddressSpecicationsthat arecommontoseveral applications. The
stepsexhibitedandtakenwill progressfromspecicationdelineationandquantication
toblock/wiredesign, thentoRFPA moduledesign.
242 Practical HF/VHF/UHF RF power amplier realization
Figure 6.1 Simpliedsystemlevel blockandwiringdiagramfor atypical RFPA system.
RFPA systemdesigncanbebest accomplishedby workingbackwards if youwill.
Thatis, anRFPA systemdesignbeginsbystartingwiththemaximumRF power require-
mentof aparticular application, anddesigningtheRFPA outputsectionrst, asopposed
tolast. Webeginwithoutput sectionrst for thefollowingreasons:
1. Theapplication(or enduse) demandfor RF power istherst andforemost require-
ment for anRFPA tomeet andtheoutput sectionaddressesthisdirectly.
2. Howtheoutput moduledesignof theRFPA evolves, andwhat itstransfer/distortion
characteristicsare, will dictatethedesignof thestagesthat will precedeit (low-level
driver andintermediatepower amplier stages).
3. Dependingonwhat RF power transistors arechosenfor theRFPA sectionandhow
they performfroma DC standpoint (i.e., DC operating voltages and DC current
demands) will alsodeterminewhat typeof DC power suppliesarerequired.
6.4 RFPA system level design overview
ThestepfollowingspecicationdelineationandquanticationforaRFPA systemdesign
istheSystemBlockandWirediagram. A simpliedSystemBlockandWirediagramis
showninFigure6.1for a1kWamplier example. Thearchitectureinthisdiagramcan
beusedtorealizeanyRFPA systemapplicationrequirement. Itshowsthebasicelements
or subsystems:
r
low-level driver stage;
r
intermediatepower amplier stage;
r
RF power divider;
6.4 RFPA system level design overview 243
r
RF power amplier section(usuallyconsistingof multiple, identical RFPAs);
r
RF power combiner;
r
directional coupler;
r
DC supplysection;
r
digital control section.
Thetopof thesystemlevel blockandwirediagramshowstheoutputpower requirement
of eachstageandaboveeachsubsystemblockdiagramaretypical gain/lossvaluesrela-
tivetoeachstage. Workingthepowerrequirementfromrighttoleft, itisapparenthowthe
applicationpower requirementdemandsafocusontheoutputRFPA sectionrst. Based
ontheapplicationpower requirement, weneedtoselectaNRF power transistor thatcan
provideadequateRF power, andmost importantly, providethenecessary power while
concurrentlysatisfyingtheapplicationrequirementsfor linearity, efciency, distortion,
transition duration response times and anticipated load VSWR excursions. Another
reasonfor workingontheRFPA output rst istheinitial vericationof meetingspeci-
cationsonamodular, scaleddownlevel. If, for example, distortionlevelscannotbemet
onamodular level, either thedesigner hastoimprovetheRFPA moduledesignor plan
for waystoprovidesystemlevel error correction.
Thereareavarietyof processesoccurringinanyRFPA system; however, if youwere
tobreakthemdownandclassifythem, thereareprimarilytwo:
r
power amplication;
r
power transfer.
Power amplication, obviously, isaccomplishedwiththeRF power transistors; power
transfer is accomplished with matching networks, dividers and combiners. Therefore,
it isreadily apparent that RF deviceselectionandimpedancematchingwill becritical
steps.
6.4.1 RF power amplier module design overview
After theblock/wiring top level systemdesign, theoutput RFPA section itself can be
broken down further into ageneric lower level block diagramas show in Figure6.2.
RF power moduleshavethesomeor all of thefollowingbasicsections:
r
RF power transistor;
r
devicebias/temperaturecompensationnetwork;
r
input/output RF andDC coupling/decouplingnetworks;
r
input/output matchingnetworks;
r
feedbacknetworks;
r
heat removal.
Thisleadstotherst major taskinRF power amplier stagerealization: selectionof
anRF power transistor. ThisisperhapsthesinglemostimportantdecisiontheRF power
amplier designengineer makes, thereareother decisions, for sure, but thisisthemost
critical. Thisisalsonot adecisionthat ismadebymerelycomparingRF transistor data
244 Practical HF/VHF/UHF RF power amplier realization
Figure 6.2 Blockdiagramof thecomponent level viewof anRFPA module.
sheets of different RF transistor manufacturers; it is adecisionthat is madeafter data
sheetreview, CADsimulation, prototyping, rigoroustestingof individual PA stagesand
thencomparisonof actual, applicationspecicresults. Careful reviewof RF transistor
datasheetsisaveryimportant, albeit initial step.
Figure6.3ispart of atypical RF power transistor datasheet. Althoughthedatasheet
containsalargeamountof data, determiningwhetherornotaparticulardeviceissuitable
for agivenapplicationmaynotbereadilyapparentwhenreviewingit. Thereasonisthat
RF power transistor manufacturerssimplycannotfullyanticipateall theapplicationsfor
which aparticular devicemay bedeployed, they will makegeneric recommendations
for applications, but it isuptotheRF power amplier designengineer tomakethenal
judgment call, andonlyafter several deviceshavebeentestedandevaluated.
Inspiteof itsvaguenature, theRF transistor datasheetisatleastastartingpoint. The
followingisabrief overviewof itsmajor sections:
1. Applications: manufacturers recommendation of potential applications for the
device; i.e.; medical, broadband, VHF communications, etc.
2. Absolutemaximumratings: maximumvaluesfor power dissipation, junctiontem-
perature, supply/breakdownvoltagesanddevicecurrents.
3. Electrical characteristics: quanticationof keyparameters:
r
power output: howmuchpower thedevicecanreliablydeliver whenmatchedinto
a50O load.
r
frequencyrange: what frequenciesthedevicecanbeusedover.
r
gain: typical power gainlevel, usuallyat themaximumoperatingfrequency.
r
efciency: howefcientlythedevicecanconvert DC power intoRF power.
r
thermal resistance: a measure of the devices ability to remove the heat of its
semiconductor diestoanoutsidesurface.
6.4 RFPA system level design overview 245
DIM Millimeters Tol. Inches Tol.
A 19.05 0.50 0.75 0.020
B 10.77 0.13 0.424 0.005
C 45 5 45 5
D 9.78 0.13 0.385 0.005
E 5.71 0.13 0.225 0.005
F 27.94 0.13 1.100 0.005
G 1.52R 0.13 0.060R 0.005
H 10.16 0.13 0.400 0.005
I 22.22 MAX 0.875 MAX
J 0.13 0.02 0.005 0.001
K 2.72 0.13 0.107 0.005
M 1.70 0.13 0.067 0.005
N 5.08 0.50 0.200 0.020
O 34.03 0.13 1.340 0.005
P 1.61R 0.08 0.064R 0.003
D1020UK
Document Number 2599
Issue 5
Semelab plc. Telephone +44(0)1455 556565. Fax +44(0)1455 552612.
E-mail: sales@semelab.co.uk Website: http://www.semelab.co.uk
Semelab Plc reserves the right to change test conditions, parameter limits and package dimensions without notice. Information furnished by Semelab is believed
to be both accurate and reliable at the time of going to press. However Semelab assumes no responsibility for any errors or omissions discovered in its use.
Semelab encourages customers to verify that datasheets are current before placing orders.
P
D
Power Dissipation
BV
DSS
Drain Source Breakdown Voltage *
BV
GSS
Gate Source Breakdown Voltage *
I
D(sat)
Drain Current *
T
stg
Storage Temperature
T
j
Maximum Operating Junction Temperature
389 W
70 V
20 V
25 A
65 to 150 C
200 C
MECHANICAL DATA
M K J
I
O N
A
C
(2 pls)
D
F
E
(4 pls)
B
G
(t yp)
P
(2 pls)
1
5 4
3 2
H
GOLD METALLIZED
MULTI-PURPOSE SILICON
DMOS RF FET
150 W 28 V 400 MHz
PUSHPULL
FEATURES
EXTRA LOW C
rss
SIMPLIFIED AMPLIFIER DESIGN
SUITABLE FOR BROADBAND APPLICATIONS
SIMPLE BIAS CIRCUITS
LOW NOISE
HIGH GAIN 10 dB MINIMUM
DR
PIN 1 SOURCE (COMMON) PIN 2 DRAIN 1
PIN 3 DRAIN 2 PIN 4 GATE 2
PIN 5 GATE 1
ABSOLUTE MAXIMUM RATINGS (T
case
= 25 C unless otherwise stated)
APPLICATIONS
HF/VHF/UHF COMMUNICATIONS
from 1 MHz to 500 MHz
METAL GATE RF SILICON FET
TetraFET
* Per Side
Figure 6.3 Typical datasheet of anRF power transistor (courtesySemelabLtd., UK).
246 Practical HF/VHF/UHF RF power amplier realization
r
breakdownvoltages: voltagelevels(atthedeviceterminals) wherethedevicewill
breakdownandfail.
r
thresholdvoltages: rangeof DCvoltagelevels(for FETs) wherethedevicebegins
toconduct current.
r
loadmismatchtolerance: ameasureof what level of loadVSWR thedevicecan
safelytolerate.
r
largesignal impedances: usually plots of largesignal input/output impedances
plottedonaSmithchart that thedeviceneeds to seeinorder to deliver its rated
power output, gain and efciency at a specic power input, frequency, supply
voltageandbiascurrent.
r
typicaltransferfunctionplots:poweroutput,efciency,gainanddistortionversus
power input andoutput.
r
typical scattering(S)parametersversusfrequencyforbroadbandcomputeraided
simulation.
r
input, reversetransfer and output capacitances versus supply voltage plots
whichvisually showhowthedevicesparasitic capacitancevariesasafunctionof
supplyvoltage.
6.4.2 RF power transistor device selection process guidelines
The following is a discussion in more detail of each major section of a typical RF
transistor datasheet.
Proper RF transistor selection starts by taking into consideration the intended
application, in other words, primarily what RF signal/modulation format the RFPA
will ultimately beexpectedtoamplify, towhat power level withanacceptablelevel of
distortion.
A particular application may not demand apower level that will exceed that of an
individual transistor, however if it does, combiners and dividers must beemployed to
reachadesiredpower output level. Ineither case, whether anRFPA designrequiresone
or moreRFPA stages, thesystemlevel specicationmustbemetwithsubstantial design
margin at theindividual nal output stagelevel (if onetransistor stagehas adequate
power) or scaledup(if onetransistor stagedoesnot haveenoughpower).Thismargin
will beerodedasmoreamplicationstagesarecascaded.
Insidethedeviceselection process exists thedecision not only for what particular
part withinaclass of transistors but also aselectionof thespecic class of transistor
itself. By class of transistor, it is meant bipolar junction transistor (BJ T), vertically
diffusedMOSFET(VDMOS), laterallydiffusedMOSFET (LDMOS), galliumarsenide
FET (GaAsFET), galliumnitride(GaN) and silicon carbide(SiC) transistors. No one
class of transistor is universally better thantheothers, eachclass of devicehas salient
featuresthat makeit moreamenabletoaparticular applicationthananother.
While BJ T transistors have less gain and are more difcult to match across broad
bandwidthsthantheirMOSFETcounterparts,insomepulseapplicationstheycanexhibit
faster risingandfallingtransitiondurations. A disadvantageof BJ Ts currently is that
there are fewer companies manufacturing these devices. MOSFETs (both vertically
6.4 RFPA system level design overview 247
andlaterally diffused) offer higher Gain, easier biascongurations, higher largesignal
impedances and are less prone to thermal runaway. A limitation of MOS FETs is
theavailability of devices operatingbroadbandover 1GHz. GaAsFET andespecially
GaN devices offer excellent choices for ultra-broadband operation from20 MHz to
over 6 GHz. However, GaAS and GaN devices aremoreexpensiveand requiremore
sophisticatedbias schemes (sequencing) tosafely turnthedevices on. SiliconCarbide
transistors offer a signicantly higher maximumjunction temperature (255

C as
opposedto200

C for LDMOS).
Therefore, inthedeviceselectionprocess, theinitial stepisselectionof whichclass
of RF power transistor, thenselectionof aparticular devicewithinthat class amongst
devicepower levelsanddifferent manufacturers.
Althoughit maybedesirabletouseasfewRF power transistorsaspossible, thereare
applicationsandsituationswhereitcanbeprudenttousemultipledevicesasopposedto
fewer. Forexample, inmissioncritical Militaryapplications, wherereliabilityandbattle
sure characteristicsarekey, it ispreferabletousemoreRF power transistorssincethe
heat will bespreadout over moredevices(whichcanyieldlower dietemperatures) and
intheevent thereisasingledevicefailure, theimpact onoverall systemperformanceis
minimized.
With this in mind, RF power transistor data sheets within a preselected class are
rst comparedintermsof enduseapplicationcompatibility. AlthoughRF deviceman-
ufacturers would prefer to make one transistor suitable for all uses, they do at some
pointoptimizetransistorstolendthemselvesbetter tocertainapplications. For example,
AvionicsRFPAsoperateprimarilyinpulsemodeandthereexist RF transistorsthat are
designedtoput out substantial power, but expresslyinpulseformat. Trytoget thesame
power out of this deviceinCW mode(or evenextendedpulsewidths for that matter)
andthedevicewill bedestroyed.
A close review of absolute maximumratings will cover just how far the device
will hold up under extreme conditions such as maximum dissipation and junction
temperature.
Thepower outputof atransistor stateshowmuchRF power adevicecandeliver. Take
careinreviewingthisparameterandnotetheconditionsinwhichthedevicemanufacturer
hasspeciedtheoutput power. Remember, theRFPA hastodeliver power, but ismore
accuratetostatetheRFPA hastodeliver concurrent power, that is, deliver power while
concurrentlymaintainingavarietyof other specicationssuchasdistortionlevels, pulse
delity, efciency, etc. A particular transistor may deliver 300 W of RF power, but if
theapplicationdemands300Wof power withlessthan1dB of GainCompressionand
your deviceis compressing5dB, thedevice, whilecapableof deliveringthepower is
notcapableof concurrentlydeliveringthepower attherequiredGainandGainLinearity
level.
Thefrequencyrangeof aparticular RF deviceshouldnotbethoughtof inanabsolute
sense, that is, if apower transistor has amaximumspecied frequency of 500 MHz,
this does not mean the device will cease to function at 501 MHz. It will function at
501MHz, itmayevenfunctionat600MHz, andyoumaybeabletouseitthere, butbear
inmindif youdo other parameters may not remaininspecicationsuchas minimum
248 Practical HF/VHF/UHF RF power amplier realization
Gain. Thedesigner who does operatetransistors far abovetheir maximumfrequency
shoulddosoonlywithagoodamount of designmargin.
Thegain specicationof RF power transistors depends onmany factors, frequency
of usage, how the device is matched (narrowband versus broadband), output power
level, temperature and load VSWR. Usually, a manufacturer will rate a device at a
minimumgainat aparticular frequency. Thisisaminimumvalue; however thedevices
they supply will usually havegaininexcess of this amount. For theapplicationwhere
large quantities of ampliers are expected to be produced, be cautious to design an
RFPA stagethat anticipates the minimum, and NOT typical gain of thedevice.
Thereason is that over time, thewafer fabrication process may yield transistors with
lower gainthanthetypical value, andif youhavedesignedastageto requireadevice
with typical gain, if a lot code of transistors is delivered with minimumgain, your
amplierwill beoutof specicationandtherewill benorecoursewiththemanufacturer.
A way of insuring your transistors are more uniformin performance is to put their
performanceunder therestrictionsof asourcecontrol drawing, or SCD. TheSCD isa
documentof mutuallyagreeduponRF Transistor performancespecications, wherethe
two parties in theagreement arethetransistor manufacturer and RFPA manufacturer.
This document calls out tighter performancespecications than exist on thestandard
devicedatasheet. It will forcethedevicemanufacturer tocherrypick devicesfroma
lot that meet thevaluesintheSCD. Thiswill invariably leadtohigher transistor costs,
especiallyif thevolumeislow,butthisissuecanbeeradicatedbyhigh-volumeproduction
quantities.
Howefcient atransistor operatesistiedinclosely withwhat classof amplication
thetransistor isbiasedto, if it isoperatedbroadbandor narrowband, what loadVSWR
it seesandwhat typeof power combiner (zerodegreeversusquadrature) isusedtosum
thepower of multiplestages. Usually, theefciencylistedonadatasheet wasmeasured
under narrowband, conjugatematchedconditionssobroadbandefciencywill belower
andfrequency/output power level dependent.
Thethermal resistancevalueis ameasureof howeasily thedevicecanremovethe
heatitsdiegeneratestoanexternal heatsink. Selectingadevicewiththelowestpossible
thermal resistancewill yield lower dietemperatures and failurerates. A low thermal
resistancealsoprovidesfor better pulsetilt performance.
The breakdown voltages quantify what level of voltage the device will fail at, the
larger this value the more a particular device will tolerate load mismatches. Device
manufacturersspecifytransistorstohandlesevereloadVSWRs, andtheymayinfactbe
abletowithstandloadVSWRsof 10:1; however, sometimesasevereloadVSWR may
precipitateoscillationswhichinturncandamagethepart.
Thethresholdvoltages showarangeof wheretheparticular devicebegins to draw
current based on a gate bias voltage. This voltage and its variation as a function of
temperaturewill playakeyroleinthedesignof thebiasthermal trackingnetwork.
Theinput andoutput impedanceof theRF transistor ischaracterizedbylargesignal
impedanceparameters. This is usually presented on thedatasheet as complex series
equivalent impedancethat is plottedonaSmithChart. Thelower theinput andoutput
impedances are the more difcult it becomes to match it to 50 O. The lower the Q
of the impedances the better, as devices with a low Q lend themselves more readily
6.4 RFPA system level design overview 249
Figure 6.4 Genericcircuit architecturefor athermallycompensatedbiasnetwork.
to broadband operation. These impedance levels provide a good starting point when
determininginput andoutput transformationratios.
Transfer function plots provideagoodvisual indicationof howadeviceperforms
over aspecic dynamic rangeof output power levels in terms of gain, efciency and
distortion.
Scatteringor S-parametersandX Parameterswhenprovided, will enablecom-
puteraidedsimulations. CADsimulationsareaninvaluabletool tooptimizeinput/output
matching, broadbandgainandefciency.
I nput, reversetransfer andoutput capacitancesareparasiticcapacitancesthat exist
withinthesemiconductor device. Thesecapacitancevaluesaredynamic(i.e., theyvary
withDC supply voltageandoutput power level). Thelower thecapacitancevalues the
better as they will inuenceahost of parameters includinggain, maximumoperating
frequency, stabilityandphaselinearity.
6.4.3 RF power transistor bias/thermal tracking networks
Asdiscussed, therearethreebroadclassesof RF power transistor:
r
BJ Ts;
r
metal oxide semiconductor FETs (vertically and laterally diffused, VDMOS,
LDMOS);
r
galliumdevices: arsenide(GaAsFET) andnitride(GaN).
All three transistor types require unique bias networks and some formof thermal
trackingtohelpmaintainrelativelyconstantquiescentcurrentswhilebeingsubjectedto
varyingthermal environments. Withoutthethermal trackingnetworks, RF biascurrents
may tend to drift and move into bias points that yield excessive or unstable gain or
undesirabletransfer functions.
Figure6.4showsasimpliedblockdiagramfor athermal trackingnetworkarchitec-
turethatmightbiasBipolar, LDMOS/VDMOSandGaAs/GaNFET transistors. Bipolar
250 Practical HF/VHF/UHF RF power amplier realization
biasingrequires ahigher current capability thanMOSFET devices. Bias networks for
FETs, inthesimplestform, canbeasimplepotentiometer. Inanycase, thebiasnetworks
mustcontainsomesourceof temperaturedependentvoltage. Thermal tracking(or com-
pensation) isaway toreducebiascurrent (or voltage) suchthat thequiescent currents
andbiaspoint of theRF transistor remainsreasonablyconstant over temperature.
Thebiasnetworksfor GaAsFET or GaN devicesaremoreelaboratesequencednet-
works, i.e.; thegateand drain voltages aresequenced or turned on/off in adened
order so as not to damagethedevice. GaAs transistors will draw heavy and perhaps
destructivelevels of draincurrent if adrainvoltageis appliedwithzero gatevoltage.
Tosafely turnthesedeviceson, thegatevoltageneedstobebrought negativerst, the
drainvoltageisthenapplied, andthegatevoltageisincrementallyraised. Theprocessis
reversedtoturnthedeviceoff. Themainroll of thesequencer istocontrol thisprocess.
6.4.4 RF input/output coupling/decoupling networks
TheinputandoutputcouplingcapacitorsarechosentoblockDCandprovideminimum
capacitivereactanceat thelowendof theampliersfrequencyrange. Thesecapacitors
mustmaintainresonancefreeoperationacrosstheampliersoperatingband. Capacitor
manufacturers datasheetsusuallyshowatypical capacitorseriesresonanceperformance
curvesover aparticular frequencyrange. Thesecurvesconveythatevenchipcapacitors
will series self-resonate at some frequency. For narrowband applications, the series
resonant frequency of aparticular chipcapacitor will bethebest frequency to usethe
deviceasacoupling/decouplingcapacitor asit exhibitsthelowest possibleimpedance.
RF chokes areemployedtodecoupletheRF signal andtofeedintheDC operating
bias and supply voltages/currents. As in the case of coupling/decoupling capacitors,
theRF chokes should also exhibit resonance-freeoperation whereaparticular choke
exhibitsonlyinductivereactanceacrosstheentirebandof intendedamplier operation.
6.4.5 Power transistor impedance matching
Thereareavariety of methods to matchtheimpedances of anRF power transistor to
50O. Whatmethodtouseisdeterminedbythefrequencyrangeandrequiredbandwidth.
Highfrequencyandveryhighfrequency(HF, 130MHz, VHF 30300MHz) RFPAs
are best matched with discrete LC networks for narrowband and transmission line
transformersfor broadband(>1octave) applications.
Ultra-high-frequency(UHF, 300MHz-1GHz) RF power transistorsarematchedwith
printedmicro-strip, transmissionlinetransformersor combinationsof thetwo.
For theHF to UHF frequency range, transmission linetransformers areby far the
most versatile matching technique as they are architecturally identical regardless of
whereinthefrequencyspectrumtheyareapplied. Thefrequencyrangemaybedifferent,
however thetransformer coaxial impedances and interconnections arethesame. How
long thecoaxial lineelements are, and whether or not they areferriteloaded arethe
primary differences betweenatransformer operatingat HF or UHF frequency ranges.
Inadditionto extremebandwidthcapability, thetransmissionlinetransformer has the
6.4 RFPA system level design overview 251
Figure 6.5 Schematicof anRLC feedbacknetworkappliedtoanenhancement-mode, N-channel
MOSFET.
ability to convert an unbalanced signal to a balanced drive required for commonly
available Gemini RF Power transistor packages that are prevalent in this frequency
range.
6.4.6 Feedback networks
Feedbackcanbeemployedtoreducelow-frequencygainandhelpimprovetheindividual
amplier modulegainatness. Figure6.5showsthegeneric circuit architecturefor an
resistive-inductive-capacitive(RLC) feedback network. Theinductor (L) andcapacitor
(C) arechosen to resonateat thelowest operating frequency of theRFPA. Theintent
istohavemaximumnegativefeedback wherethegainof thetransistor isgreatest. The
capacitor will alsoblocktheDCsupplyvoltagefromreachingthegates(or bases) of the
RF transistors. Thevalueof (R) adjusts theamount of feedback. This represents only
onemethodof RF Feedback, thereareother morecomplexmethods(transformer based)
that achieveDC isolationbymagneticallycouplingthefeedbacksignal.
6.4.7 Thermal management
Whilegenerallynotconsideredpartof theRFPAscircuitry, themethodbywhichheatis
removedfromanRFPA isequallyvital. Improperheatremoval canleadtodegradationof
anarrayof RF performanceparametersincludinglinearity, efciency, gainandstability,
252 Practical HF/VHF/UHF RF power amplier realization
etc. Inaddition, higher operatingdietemperatureequatestoreducedoperatinglifetime
andincreasedFailureRates.
For most applications, forcedair coolingisadequate, inhigh-altitudeairborneappli-
cations, liquidcoolingis apreferredoptionas thereductioninair density inhibits the
efcacy of forced air. In either case, however, the thermal interface between the RF
power transistor andthemodulebase, heat sinkor chill plateiscritical, sothedetailsof
creatingaproper transistor angethermal interfacearecovered.
6.5 Hypothetical amplier design example: (20400 MHz RFPA subsystem
module for 1 kW amplier application in electronic warfare-
communication jamming)
To helpillustrateandbetter convey atypical RFPA modulerealization, ahypothetical
amplier designexampleispresented. Theprocesswill involve:
r
hypothetical applicationexampleoverview;
r
RFPA qualitativespecicationdelineation;
r
RFPA specicationquantication;
r
RFPA modulehardwaredesign;
r
RFPA modulephysical construction;
r
test setup;
r
test results.
6.5.1 Hypothetical application example overview
RF Power ampliersthatareusedfor electronicwarfare(EW) communicationjamming
applicationsmusthaveverybroadbandwidthcapabilitysuchthattheyhavetheabilityto
jamvariouscommunicationbands. Frequenciesintherangeof 20200MHzarepopular
for land/mobile communications and military aviation bands heavily populate 225
400MHz. It is benecial to haveoneamplier cover bothbands so thedesigngoal is
tospan20400MHz. Theamplier will beusedtoselectivelyinhibit communications;
itsoutput will beconnectedtoabroadbandantenna, sotheloadVSWR will deviatefar
fromaperfect match.
6.5.2 Amplier qualitative specication delineation
Thefollowing is alist of specications that arecritical to broadband jamming appli-
cations. This is acursory initial attempt, only eldtrials andbetatestingwill reveal if
certainparametersaremoreessential thanothers.
1. Highpower: theRFPA must haveenoughoutput capability tooverpower receivers
andblockenemysignal transmissions.
2. Broadbandwidth: theRFPA musthaveadequateoperatingfrequencyrangetocover
avarietyof communicationbands.
6.5 Hypothetical amplier design example 253
3. High gain: theamplier must haveenough gain such that it can bedriven to full
power output byasmall signal level input.
4. Flat frequency response: theamplier shouldperformuniformly at all frequencies
withinthedesiredrange.
5. LoadVSWR tolerant: sincetheRFPA will beusedtojampoint topoint communi-
cations at various frequencies; it will morethanlikely haveto driveantennas that
present less than ideal load VSWRs which may go as high as 5:1. Theamplier
must beabletodriveintotheseloadswithout damage.
6. Stability, spectral loadpull: as theRFPA must not bedamagedby drivingadverse
loadVSWRs, it also shouldnot oscillateat unacceptably highpower levels under
theseconditionsaswell.
7. Linear: althoughhighgainlinearityisnotusuallycritical, certainjammingsituations
will requiretheRFPAsoutput levelstobepreciselycontrolled.
8. Temperaturerange/stability: theamplier will bemost likely usedinadverseeld
environmentswherehightemperaturesarecommon. Theamplier will beexpected
toprovideacceptableperformanceinthesetemperatureextremes.
9. Altitude: theampliermaybeinanavionicsplatformif itisexpectedtojamairborne
communications, inthiscase, forcedaircoolingwill notbeanoption, liquidcooling
andchill platesarepreferable.
10. Shock/vibration: mostmilitaryampliersaredeployedinmobiletransportenviron-
ments. Abilitytowithstandsevereshockandcontinuousvibrationisessential.
11. Radiatedemissionsandsusceptibility: whiletheRFPA isintendedtorender specic
enemy communicationequipment ineffective, it must not interferewithor haveits
own operation impaired by other colocated equipment. The RFPA must then be
adequatelyshieldedfor EMI.
12. Meantimetofailure: theamplier, aboveall, must bereliable, liveswill dependon
it. Toensurethis, thetransistor silicon(or die) temperaturemust bekept aslowas
possible.
6.5.3 Amplier specication quantication
With a generic list of required specications, the next step is to assign quantities to
eachlineitem. SincethefocushereisonHF/VHF/UHF RFPAs, thespecicationsthat
directly address thepower amplier modulewill becovered quantitatively. (Note: for
ease of illustration, quantities will be loosely assigned and may not represent actual
systemrequirements, whichinmanycases, isclassiedinformation).
r
systempower output: 1kW, continuouswave, minimum;
r
bandwidth: 20400MHz, minimum;
r
gain: 60dB, nominal;
r
gainatness: / 3dB, maximum;
r
antennaloadVSWR: -5.0:1, maximum;
r
stability, spectral loadpull-45dBcupto5.0:1loadVSWR, maximum;
254 Practical HF/VHF/UHF RF power amplier realization
Figure 6.6 Systemblockandwiringdiagramfor 20400MHz, 1KWRFPA system.
r
gain linearity (AM to AM distortion): / 1.5 dB over 20 dB dynamic range,
maximum;
r
phase linearity (AM to PM distortion): / 10

C over 20 dB dynamic range,
maximum;
r
temperaturerange: 10to50

C;
r
altitude: 40,000feet AMSL, maximum;
r
shock/vibration: 10Grms, sixaxis, maximum;
r
radiatedemissionsandsusceptibility: Mil-Std-461E;
r
meantimetofailure: >100,000h, minimum, calculated.
6.5.4 Amplier hardware design/realization
Therequirement hasnowbeendenedquantitatively. Theprocesstostart thephysical
realization begins with the SystemBlock and Wire diagramas shown in Figure 6.6.
Thesimplieddiagramshows all thestages intheRFPA system. Whiletheredoexist
RF transistorsthat candeliver 1kWof CWoutput power, thechallengebecomes heat
removal andAC couplingof theRF signal. If anRF deviceputsout 1kWof RF power
at, for example, 50%efciency, then1kW of heat will needto bedissipated. Addin
bad load VSWR and the dissipation will worsen. In addition to heat dissipation, the
RF current that will bepresent at thelow-impedanceoutput of a1kW transistor will
tax even thehighest quality chip capacitor. It will beamorereliableapproach to use
multipleRF power transistors.
One of the benets of a systemlevel block and wiring diagramis it shows the
insertion gain, loss and RF power levels as power propagates through the amplier
6.5 Hypothetical amplier design example 255
stages. Itdisplaysjusthowmuchpower islostthroughdirectional couplers, RF dividers
andcombiners.
Theloss values shownintheblock andwirediagramarefromactual couplers and
hybridcombiners. Notethatfor anamplier toproduce1kWof linear loadpower intoa
5:1VSWR it actuallyneedstobecapableof driving2788Wof power intoa50O load.
After addinglosses for thecoupler andcombiners, thetotal requiredRF transistor die
power is about 4172W. Choosingabinary multipleport combiner with32ports, then
4172Wdividedby32yieldsabout 130W. WenowhaveanapproximatemaximumRF
transistor output power.
Thenext phaseis RF deviceselectionwheretheselectionguidelines arefor anRF
power transistor capableof 20400MHz operational bandwidthat apower level of at
least 130Wof linear power.
Thesystemblock andwirediagramis brokendownfurther inFigure6.7toablock
diagramof thebasicRFPA module.
6.6.5 RF transistor selection
Current popular output power ranges for RF transistors are 100, 150, and 300 W of
output power. Sinceweneedapproximately 130W of CW output power, thefocus is
onthe150W rangeof CW RF power transistors that aredesignedfor operationupto
400MHz.
An overview of classes of transistors was rst considered. Since the frequency of
operationwasonlyto400MHz, GaAsandGaNdeviceswereruledout. Theywill work
farinexcessof 400MHz; theaddedcostforthisunneededcapabilityisnoteconomically
justiable.
BJ Ts were not selected due lower gain and most importantly, a limited amount of
devicemanufacturers. Fewmanufacturersmeanalowprobabilityof obtainingasecond
source (i.e., an alternate manufacturer with an equivalent part that will replace the
primarydeviceintermsof form, tandfunction)of BJ Ttransistors. Itsoundsinnocuous,
butnothavingasecondsourceof areplacementRF transistor isaverydifcultsituation
tobein. Yousimplydonot want toonedayndyourself inthisparticular xandheres
howthismight happen:
1. RF transistor semiconductor wafer fabricationprocessesareckle, althoughtheyare
tightlycontrolled, intheendtheyarerunbyhumans. Atanypointintimeaparticular
device process control can vary yielding devices that may work on a substandard
level andcanrender anRFPA withslimdesignmarginsinviolationof specication.
2. RF transistor devicemanufacturers have, canandsomeday may either discontinue,
de-rateor sell aparticular lineof transistorstoanother manufacturer.
If eitherof thesesituationsoccur, aonceprotableRFPA designcanovernightdevolve
intoalabqueen (anamplier that canonlymeet specicationbycopiousamountsof
tuning, requiringdaysor weeksinthetestlab) or, evenworse, causeastopproduction
modewhereshipmentshaveceasedleavingtheRFPA engineer (yes, thiswouldbeyou)
franticallysearchingfor analternativepart.
256 Practical HF/VHF/UHF RF power amplier realization
Figure 6.7 Multilevel breakdownof anRFPA moduletocomponent level.
6.5 Hypothetical amplier design example 257
4 8
5
5.2
5.4
5.6
2

a
m
p
s
D1020UK V
gs
vs Temp
4
4.2
4.4
4.6
4.8
5
5.2
5.4
5.6
V
g
s
f
o
r

I
d
s
=

2

a
m
p
s
Flange TemperatureC
Figure 6.8 2A I
dq
biaspoint of theD1020UK versustemperature.
This leaves the MOSFET class of devices. For this particular application, either
vertical orlateral MOSFETswill workequallywell. Theselectionprocessnowdescends
tosegregatingpartsthat workacrossthe20400MHZbandat aminimumof 150Wof
linear CWpower withamanufacturer recommendedapplicationfor communicationsin
theHF/VHF andUHF bandandaloadtoleranceat least upto5:1VSWR.
Several devices wereconsidered, their datasheets comparedandRFPA stages were
constructedandactual test datacompared. TheSemelabD1020UK wasselectedbased
onitsabilitytosatisfythedesigncriteria. Thedatasheet for thisdeviceisinFigure6.3.
6.5.6 Gate bias/temperature tracking/compensation network
Thegatevoltageversustemperatureplot of theD1020UK isshowninFigure6.8. This
was obtainedby biasingthedeviceto 2A quiescent andvaryingthegatevoltagefor a
constantI
dq
astheambienttemperaturewasvariedfrom50to150

C. Notethegate
voltagelevel tosustain2A quiescentdropsapproximately4.4mVDC/

C. Inotherwords,
if thegatebias voltageis held constant over increasing temperature, by virtueof the
factthatthe2A biasvoltagepointiscontinuouslyloweringwithincreasingtemperature,
thenthexedgatebiasvoltagemigratesbydefaultintohigher andhigherdraincurrents.
Thismakesthegainandoperatingclasspoint of theFET dependent ontemperature. To
mitigatethisissue, anopenloopthermallytrackedbiasvoltagecanbedeployed. Thisis
accomplishedbymountingatemperaturesensingICmechanicallyadjacenttothedevice
that istobecompensated. Ideally, it isdesiredtohavethebiasvoltagedroppedby the
equivalentamountthatthe2A gatebiasvoltagepointdrops. Whatisniceaboutthebias
voltagevariationof theD1020UK isthatitisapproximatelylinear. Thismakesiteasyto
correct astemperaturesensorsthat havelinear outputs(inmV/

C) arereadilyavailable.
258 Practical HF/VHF/UHF RF power amplier realization
0.1
V
temp
Figure 6.9 Operational amplier basedthermal trackingcircuit for theD1020UK.
2
2.5
3
m
p
s
)
D1020UK Quiescent Bias Current vs Temperature
0
0.5
1
1.5
2
2.5
3
-10 0 10 20 30 40 50
D
1
0
2
0
U
K

I
D
Q
(
A
m
p
s
)
Ambient Temperature (Degrees Celsius)
Uncompensated
Compensated
Figure 6.10 Comparisonof thequiescent currentsof D1020UK without (dashedtrace) andwith
(solidtrace) athermallytrackedbiasvoltage.
All thatneedstobedoneistoadjustthetransferfunctionslopeof thetemperaturesensor
suchthat it hastheinverseslopeof thegatevoltage.
The gate bias/temperature compensation network for the amplier is shown in
Figure 6.9. It consists of a simpleprecision variable voltage divider (5 VDC) net-
work that is summed together with a temperature dependent voltage (Vtemp) that is
mechanically linked to theRF transistor to lower its bias voltageas thedeviceheats
up. Thepart chosenfor thetemperaturesensor istheAnalogDeviceTMP35whichhas
a scale factor of about 10 mVDC/

C. The 5 K O resistor and 10 KO potentiometer


adjustthisscaledownto4.4mVDC/

Candthenitissummedintotheprecisionvariable
5VDC reference.
ThisisanopenloopcompensationnetworkandhelpstheD1020UK maintainamore
temperaturestablebiascurrent. Figure6.10showsthedraincurrent versustemperature
with and without thermal tracking. Without thermal tracking theDrain current varies
6.5 Hypothetical amplier design example 259
1.1A over the 10to50

Ctemperaturerange. Addingthethermal trackingnetwork


reducesthisbyover 80%to0.2A of draincurrent variation.
Whilethis is consideredgoodperformance, it is important to notethat this is open
loopcorrectionandthetemperaturecompensationaccuracy canvary fromlot codeto
lot codeand also on thesamedevicedueto memory effects. Thermal compensation
networkscanbedesignedwithgreater accuracyif theytakeintoaccount thebehavioral
modelingof aparticular classof device.
6.5.7 Input/output RF/DC coupling/decoupling networks
A coupling/decouplingnetwork is merely another namefor capacitors andcoils inan
RFPA. CouplingnetworksareusuallycapacitorsthatpasstheRF power signal fromone
devicetoanother whileblockingDC voltages andcurrents. Decouplingcapacitors are
usedtobypassRF signalstoground. Inductorsincouplingnetworksperformtheduality
function, that istheywill passDC voltagesandcurrentswhilesuppressingRF signals.
InabroadbandRFPA, thevaluesselectedfor thesecomponentsaresolvedfor at the
extremesof theoperatingbandedges. Thesenetworkswill bebasedrelativetothelarge
signal input andoutput impedancesof theD1020UK.
Thecoupling/decouplingcomponentsgenerallyhavetomeet threecriteria:
r
satisfyaminimumreactance(impedance) requirement at thelower frequencylimit;
r
handlehighDC andRF currents, voltages, andpower;
r
sustainresonance-freeoperationbeyondtheRFPA upper frequencylimit.
Byviewingthethreerequirements, oneneedonlysolvefor acomponent valueat the
lower bandedge. Theremainingcriteriaareassessedbymanufacturersdatasheetsand
component vericationtestingonavector networkanalyzer (VNA).
Thevalues for thecouplinganddecouplinginductors andcapacitors aresolvedfor
in thenext section as it will berequired to seewhat impedancematching network is
requiredrst.
6.5.8 Input/output impedance matching networks
Thereisamultitudeof waystomatchtheinput andoutput impedancesof anRF power
transistor. If onelooks at thecharacteristics of theinput andoutput impedanceof the
D1020UK, it becomes apparent (at least for abroadbandamplier application), that it
will bedifcult, if not impossible, toprovideanexact, complexconjugatematchfor the
deviceat all frequenciesandinput drive/output power levelstobeexpectedtobeused.
A balancehastobestruckbetweenwheretochoosetomatchaparticular device. On
theinput, thedeviceshould bematched at thepoint in thefrequency rangewhereits
gain is lowest (thehighest frequency it will operateat). Theoutput is matched at the
highest level of expectedRF output power.
To match impedances over multiple octaves in the HF/VHF/UHF band, the trans-
mission line transformer is the most effective method. In addition, it converts an
260 Practical HF/VHF/UHF RF power amplier realization
unbalancedsignal toabalanceddriverequiredfor Gemini packagedtransistorssuchas
theD1020UK.
The transformer, while able to transformimpedances over wide frequency ranges,
only does axed transformation ratio. RF power transistor terminal impedances will
vary, soatransformerwill transformimpedancesof atransistoreffectivelyoveralimited
frequencyrangeandoutput power level. Outsideof theseareas, input VSWR, gain, and
efciencywill suffer.
Starting with the input matching network, the input terminal impedance of the
D1020UK is extractedfromits Smithchart onthedevicedatasheet. Thegateto gate
seriesequivalentinputimpedanceat400MHzis0.4-j2.3O. Convertingthistoaparallel
equivalent impedanceyieldsareal portionof 13.62O fromgatetogate. Dividing50O
by 13.62equates to 3.67:1. Thenearest realizablebalanceddrivetransformationratio
is4:1.
A rough estimate of output impedance from drain to ground is given by the
equation:
R
o
=
V
2
dd
2P
o
(6.1)
The transistor will be operated at 28 VDC and at approximately 75 W of power
per side. This estimates approximately 5.23 O fromdrain to ground. The pushpull
congurationof theGemini packagedoublesthisto10.45O. Wecanalsoextractaclose
valuealsofromtheSmithchart draintodrainseriesequivalent impedanceaswasdone
for theinput. Theseries equivalent valueis 0.9-j3 O at 400 MHz. Converting to the
parallel equivalent yieldsareal portionof 10.84O. Dividing50O by10.45equatesto
4.78:1. Again, thenearest realizableratiowithabalanceddrivecapabilityis4:1.
Thetermrealizableratiomeansavalueof transformationratiothatcanbephysically
constructedwithanitenumber of transmissionlines. For aparticular transformation
ratio to be physically realized, the square root of the transformation ratio must be a
rational number. If it isnot, aninnitenumber of transmissionlineswouldberequired
torealizetheratio, anobviousimpracticality.
For boththeinput andoutput transformations, anapproximate4:1ratio is required.
Invokingatopological networksynthesisprocedure[1] for transmissionlinetransform-
ers, therst stepis to satisfy thenecessary andsufcient conditions for nitecoaxial
element equal delaytransmissionlinetransformers:

N = Rational Number (6.2)


whereN= requiredtransformationratio. Thesquareroot of 4is2, arational quantity.
Thenecessaryandsufcientrealizabilityconditionissatised. Thesynthesisprocedure
cannowbeginwiththereasonableexpectationof aphysicallyrealizablenetwork.
Therst stepistodeterminethenumber of coaxial linesintherst subgraph:
n
1
=

4= 2. (6.3)
wheren
1
is thetruncationof thesquareroot of thetransformationratio N, andinthis
caseisthenumber 2.
6.5 Hypothetical amplier design example 261
Z
in
Z
o
Z
o
Z
out
Figure 6.11 Subgraphresult of atopological synthesisof a4:1transmissionlinetransformer.
Z
in
Z
o
Z
o
Z
out
Z
o
Figure 6.12 Completedunbalancedtobalanceddrive4:1transformer network.
Thesynthesisprocedurewill terminateif:
N
1
=

N n
1
= 0
N
1
=

42= 0
(6.4)
Theprocess terminates andtwocoaxial lines areinsertedintosubgraph1as shownin
Figure6.11. Thecharacteristicimpedance, Z
0
, of thecoaxial linesissolvedfor by:
Z
0
=
_
R
s
R
l
(6.5)
whereR
s
andR
l
arethesourceandloadresistancesterminatingthetransformer, respec-
tively. These values are 50 and 12.5 O and solve for a characteristic impedance of
25O. Thesynthesisprocedureprovidesatransformer architecturethat isapplicableto
unbalancedtounbalancedloads. TheD1020K inapushpull congurationdemandsa
balanceddrivenetwork. Thereforethetransformer of Figure6.11needstobeconverted
to a true balanced network (one that would present an electromagnetically balanced
distributed network) by interchanging the shield and center conductors of the lower
transmission lineelement. Thetransformer conguration nowprovides abalanced to
balanced drive. A 1:1 balun transformer is added at thehigh impedanceport to pro-
videtheunbalancedtobalanceddriveconversionsuchthat theinput tothetransformer
canbereachedby groundreferencedcoaxial or microstripfeeds. Thenal 4:1unbal-
ancedto balanced transmissionlinetransformer architectureis realizedandshownin
Figure6.12.
Thesynthesized 4:1 architectureis aboiler plate circuit structure, that is, in this
formatitisanengineeringconstructthatcanconceivablyworkinbroadfrequencyspans
262 Practical HF/VHF/UHF RF power amplier realization
Figure 6.13 Simpliedparasiticmodel of 4:1transformer network.
anywherefrom10kHztowell over1GHz. Thatscomforting, buttheRFPA specication
only requires20400MHz. Thetransformer architectureneedssomemassagingtoget
it to have a sweet spot of 20400 MHz. By sweet spot it is desired to have the
transformers absolute values of insertion loss minimized and return loss maximized
from20400MHz.
Intheworldof engineeringconstructs, thephysical transmissionlinetransformer is
anideal transformer nestedinaplethoraof parasitic reactances. Thesereactances will
limittheusablefrequencyrange. Thereactancesthatlimitthelowerandupperfrequency
rangeof thetransmissionlineareprimarilyinductiveinnature. Thelow-frequencyrange
islimitedbyshuntinductance, thehighendislimitedbyparasiticseriesinductance, and
theseareshownschematically inFigure6.13. Parasitic distributedgradient capacitive
reactances will resonate with line inductances and may cause in-band insertion loss
suck-outs, a narrow band of frequencies within the pass band where the insertion
lossspikestovery highvaluesandthenreturnstolowvalues. Additional lossesinthe
transformer arefromferritecores, coaxial linedielectric, copper conductorsaswell as
radiationlosses.
After realizing the4:1 architecture, thenext step is ferriteloading thecoaxial line
elements inorder to suppress evenmode(nontransmissionline) currents andcreatea
netmutuallycoupledinductancethatisinfar excessof theimpedancetobetransformed
sothat itsparallel loadingeffect isimmaterial.
As shown, a physical transmission line transformer is a complex model, an ideal
transformer, mutually coupled inductors, parasitic reactances, and transmission lines.
Anequivalentcircuitof themutuallycoupledinductorsisshowninFigure6.14.Thereare
essentiallyfourinductorsinthe4:1transformersection, twoineachcoaxial transmission
line. For all intentsandpurposes, insidethecoaxial lines, thecoefcient of couplingis
6.5 Hypothetical amplier design example 263
L
4
L
4
L
3
M
12/21
M
34/43
L
2
L
3
L
2
L
1
L
1
i i
Figure 6.14 Mutuallycoupledinductancesof a4:1transmissionlinetransformer.
consideredbi-directionally unity, that is, all theux generatedby thecenter conductor
of thecoaxial lineislinkedtotheoutershieldandviceversa. The4:1isopencircuitedto
helpvisualizethemutuallycoupledinductorswithacommon-modecurrent, i, owing
throughall four inductors. Thenet valueof this inductivereactanceshouldbeat least
510timesthevalueof impedanceof 50O soit doesnot adverselyloaddownthevery
impedancesthetransformer istryingtostepup. Thepathfor ndingthetotal inductance
startswith:
:
i
= L
1
di
dt
M
12
di
dt
L
3
di
dt
M
34
di
dt
L
2
di
dt
M
21
di
dt
L
4
di
dt
M
43
di
dt
(6.6)
where
M
xy
= k
_
L
x
L
y
(6.7)
and
k = 1
L
x
= L
y
(6.8)
So
L
1
= L
2
= L
3
= L
4
= M
12
= M
21
= M
34
= M
43
L
t
= 8L
SW
(6.9)
What theaboveequations stateisthat duetothe4:1transformer congurationof four
mutual, bi directionally unity couplingcoefcient inductors, to solvefor thenet shunt
inductance, simplymultiplytheinductanceof whatwill resultfromwindinganinductor
of asinglewireonagivenferrite(L
sw
) byeight.
Wewouldliketohavetheshunt inductancetobe>510timesgreater than50O so
asnottoloaditdown. Wealsomustkeepthelengthof the25Ocoaxial linesasshortas
possiblesoasnottoincur in-bandresonancesintheresponseof the4:1transformer. We
thereforestartoutwithkeepingthenumber of turnsthroughaferritecoretoaminimum
of twoturns.
264 Practical HF/VHF/UHF RF power amplier realization
Figure 6.15 Measuredresultsof mutuallycoupledinductancesof a4:1transformer.
Solvingfor therequiredinductancefactor at 20MHz:
X
l

= l

= j 500

= 8n
2
A
l
(6.10)
wheren= number of turnsthroughtheferritecore= 2. Notethat theaboveequationis
lacedwithapproximately equal to (

=) signsinsteadof equal to (=). Thereasonis


that theinductancefactor toleranceof ferritetoroidsisabout/ 25%at best, sothere
is littlepoint in trying to solvethis equation precisely as any inductor or transformer
youmight construct usingferriteswill varywildlyinvalue.
It hasbeenfactoredinaheadof timetohavetheshunt inductanceof thetransformer
tobe510times50O, soif youlandinthisregion, thetransformerslow-endresponse
will beacceptable. Theequation was nevertheless solved and an approximateinduc-
tancefactor of 124nh/n
2
was calculated. A ferritecorefromceramic magnetics (Part
6.T503125T-C2050) was chosen as its measured inductance factor is 100 nh/n
2
. The
transformer is constructedwith25O coaxial lines withtwoturns througheachferrite
toroidcore.
Thetransformer wasevaluatedonaVNA withthelow-impedancesideof thetrans-
former loadedby anopencircuit. This will measuretheparallel inductance. Theplot
in Figure6.15 shows theresults with thevalueof inductivereactanceat an adequate
level of 4.3 H which yields a parallel inductive reactance of j540 O, more than
enoughsoasnot toimpair thelow-frequencyreturnlossresponse. Thetransformer can
6.5 Hypothetical amplier design example 265
Figure 6.16 Input returnlossof the4:1input transformer with6.25O chipresistor terminations
onthelow-impedanceport.
then be terminated with 6.25 O chip resistors to evaluate how the device transforms
theresistances to 50 O. Thegraph of thetransformers input return loss is shown in
Figure6.16. Ityieldsanaverageof 21.6dBof returnloss. Theplotof thetransformers
input Insertion Loss is shown in Figure6.17, (notethevalues on thedataneed to be
dividedby twoas theplot is of thetransformers returnloss withanopencircuit load,
hencetheinsertionloss is half). Thereis asmall resonanceat 383.62MHz, thesecan
occur inultra-broadbandtransformers, thebest way to mitigatethemis to shortenthe
lengthof thecoaxial lines. Typically, thetransformerwill alsohaveseriesinductivereac-
tancethatwill impair thehigh-frequencyrange. Thiscanbetunedinwithcompensation
capacitors. Althoughthesecanbecalculated, it isfairlyquicktooptimizeatransformer
bysubstitutingdifferentvaluesof shuntcapacitanceduringtestproceduresandselecting
thevaluethat yieldsthebest broadbandreturnloss.
Since both the input and output ports of the amplier demand a 4:1 transformer,
thesamedevicearchitecturewill beusedbothports. Theoutput transformer useslarger
cross-sectional areaferritesandlargerdiametercoaxial cabletoaccommodatethehigher
power levels.
Withtheimpedancematchingtransformerssolvedfor, it isnowafairly simpletask
to go back to solvefor thecoupling capacitors. Weknowthecapacitancevaluemust
present alowreactanceat 20 MHz. Thecoupling capacitors areto beinserted at the
266 Practical HF/VHF/UHF RF power amplier realization
Figure 6.17 Low-impedanceopenport returnlossof 4:1transformer (toobtaininsertionloss,
divideplottedvaluesby2).
50O point betweenthe4:1and1:1transmissionlinetransformers. It isdesiredtohave
thereactanceat least 1/100
th
of 50O or lessthanj0.5O at 20MHz. Solvingfor C:
C
1
.0.5
= 0.0159F (6.11)
The maximumamount of average current that the series coupling capacitor will see
occursat 150WCW, therefore
I
max
=
:
rms
z
=

P
max
.z
z
=

150.50
50
= 1.73A(rms) (6.12)
Basedonthethreecriteriasetearlierwehavetwoof themsolvedfor, theminimumvalue
of capacitanceandthemaximumaveragecurrent witha50O load. Tobeconservative,
themaximumcurrent will bedoubledtoaccount for drivingadverseloadVSWRs. The
D1020UK isa28VDC device, sotheDC operatingvoltagemust beinexcessof this
value.
Thechipcapacitorselectedisa0.1F, 50WVDC200Bseriescapacitormanufactured
byATC. Thecapacitor canhandleover 9A of averageRF current soit will besuitable
not onlyfor theoutput couplingcapacitorsbut theinput aswell.
The remaining issue is to verify the capacitor, on its own, will maintain a low-
impedance, resonance-free, operationthroughout theentireamplier bandwidth. This
6.5 Hypothetical amplier design example 267
Figure 6.18 Impedanceof 0.1F chipcapacitor from20400MHz.
isveriedbydoingaone-port impedanceanalysisusingaVNA. Figure6.18showsthe
responseof thecouplingcapacitor acrossthe20400MHz bandwidth.
6.5.9 Feedback network
Manual calculationof thefeedbackresistancevaluecanbeveryroughlydeterminedby
thefollowingequation[2]:
R
f
=
(V
2
V
3
)
__
V
1
V
2
R
1
_

_
V
2
R
2
__ R
4
(6.13)
where:
R
f
= feedbackresistanceinO;
V
1
= voltagegatetogateat 400MHz = 9.7V rms;
V
2
= voltagegatetogateat 20MHz = 2.17V rms;
V
3
= voltagedraintodrainonD1020UK at 150Woutput = 43.3V rms;
R
1
= R
2
= impedanceonoutput of input matchingtransformer = 12.5O;
R
4
= output load, draintodrain= 12.5O.
268 Practical HF/VHF/UHF RF power amplier realization
This equation only provides a rough estimation of feedback resistance value. In a
broadband amplier, intentional negativefeedback can swingpositiveif not carefully
modeled. Inthisparticular case, theequationyieldedavalueof 52.5O. At somelevels
of drivetheamplier went intooscillation, sothevaluewasincreasedto100O andthe
amplier becamestable. Withthislevel of feedbackthegainat20MHzwasreducedby
approximately8.3dB. Thesmall signal gainplotsaregiveninSection6.5.12.
6.5.10 Test setup conguration/analysis
Prior to thediscussion of theexampleRFPAs construction and electrical test results,
someeffort will bedevotedto what is requiredto verify theamplier is specication
compliant onamodular level. It isprudent toverify complianceat amodular level for
theobvious reason that if you cant meet specication there, in certain parameters, it
is unlikely performancewill improveat asystemlevel. Knowinghowtotest anRFPA
is every bit as essential as knowinghowtodesignone. Howaccurately thetest results
are acquired will ultimately advise the RFPA engineer of how much design margin
does/doesnt exist. Thetests must beperformedonly ontest equipment that is within
its calibration cycle and has National Institute of Standards and Technology (NIST)
traceability. Thisistoinsurethat whatever test resultsyouobtainwill ultimately agree
withyour customersdata.
Eventhoughthemoduleisfairly small andmay cost only afewthousanddollarsto
fabricate, to fully analyzeand test for this particular requirement will requireseveral
milliondollarsintestequipment. Decisionswill needtobemadealongthewaywhether
or not to purchase, rent, leaseor useatest facilitys equipment. Thekey pieceof test
equipmentistheVNA. Thisunithastheabilitytomeasureinputreturnloss, transformer
insertionloss, small/largesignal gain, gainatness, andinsertiongain/phaselinearity.
Thespectrumanalyzer inconjunction with theparalleled loads and binary stepped
transmission line will monitor the modules spurious response while being subjected
to various loadVSWRs. Thebinary steppedtransmissionline[3] is simply lengths of
coaxial cable cut to specic lengths and switched in line with 2, 3, 4, and 5:1 load
VSWRssuchthat impedancesalongconstant loadVSWR circlescanbeappliedtothe
output of theRFPA module. Thisleadsback toanearlier discussioninthischapter on
thetopicof unconditional stability. Thisistheverytest that isperformedtoassessload
pull stability. It becomesapparent that it isimpossibletopresent acircuit at theoutput
of an RFPA that can emulateall impedancepoints bounded by constant load VSWR
circlesontheSmithChart.
Figure 6.19 is a block diagramthat illustrates the interconnection of a typical test
setupthat will verify asubstantial portionof themodules performance. Theoutput of
theVNA is fedto alow-level test driver to increasethepower to thepoint whereit is
sufcient todrivetheRFPA tofull power. Thepreciseoutput power level of theRFPA
issampledwithacalibrateddirectional coupler andfedtoanRF Power meter. For gain,
gainatness, andinsertiongain/phaselinearity, theVNA hastheabilitytocalibrateout
theresponsevariationsof thelow-level driver. TheVNA cancharacterizeamajorityof
theampliers frequency andpower domainresponses. Inaddition, whentheRFPA is
6.5 Hypothetical amplier design example 269
Figure 6.19 Typical RFPA module/systemtest conguration.
subjectedtotemperature, altitudeandvibrationanalysis, theverysametestsetupcanbe
deployedwiththeinitial responsetransfer functionsstoredinmemoryandcomparedas
temperatureandvibrationlevelsareincreased.
WhilethesetupinFigure6.19cancover amajority of therequiredtests, toexecute
radiatedandconductedemissionstestswill requirethesophisticatedsetupinFigure6.20.
Thesearehighlyelaboratesystems,thecenterof whichare3and10manechoicchambers
whicheffectivelyprovideacontrolledEMC environment inwhichtotest theampliers
susceptibilityto, andemissionof, EM radiation. TheRFPA isplacedina3mchamber
andsubjectedtohigh-power RF energy emittedfromclosely placedbi-conical logand
doubleridgehornantennas. Theamplier is thenmonitoredfor its ability tomaintain
speciedoperationwithout degradationof performancewhilethefrequency andeld
strengthof theradiatedRF energyisvariedover verybroadranges.
For radiatedemissions, theamplier isplacedonaturntablewithinthesemi anechoic
10mchamber;itisthenrotatedwhiletransmittingfull RFpoweroutput.Highlysensitive,
Bi-conical, Logperiodic andActiveLoopantennasarelocatedabout 20feet away and
will beelevatedandloweredbasedontheEMI/EMCspecications. Thesignal received
by the antennas is plotted on a graph with limit lines that show if the amplier is
emittingRF energy beyondacceptableamounts. Radiatedemissions is oneof thefew
requirementsof theRFPA specicationthatif theamplier isnotcompliantatamodule
level, it remains possibleto becompliant at asystemlevel sincetheamplier will be
mounted within ametal chassis that will allowfor further shielding and reduction of
emissions.
270 Practical HF/VHF/UHF RF power amplier realization
Figure 6.20 Simpliedblockdiagramof 3and10mEMC test chambers.
A highlyacceleratedlifetest (HALT) chamber isshowninFigure6.21. Thishasthe
uniqueabilitytoapplyrandomvibrationtotheRFPA moduleinsixaxes(X, Y, Z, yaw,
pitch, and roll). This is moreeffectivethan asingleor dual axis vibration tableas it
caneffectively impart vibrationenergy to themodulethat is closeinlinewithwhat it
will encounter in theeld. In addition, theHALT chamber has theability to change
temperaturefrom100to200

C.
6.5 Hypothetical amplier design example 271
Figure 6.21 HALT chamber.
6.5.11 Physical RFPA module construction
A photoof theindividual breadboardRFPA stageisshowninFigure6.22. Theprinted
circuit board with micro strip interconnecting transmission lines is mounted into an
aluminummodule. TherearetwoPCboards, onefortheinputdivider/matchingnetwork
andonefor theoutput.
The transistor is mounted in a milled-out channel 0.070" deep such that the gate
anddrainleadssit ushwiththePC board. Thischannel isperhapsthemost important
machinedsurfacewithintheentiremodule, carewastakentospecifythatithasasurface
atnessof better than5m/cmalongwithanR
A
(averagesurfaceroughness) of better
than1m. Thedevicemanufacturer hasprovidedthismountinginformation[4].
Withthematingsurfaceready, averythinlayer of thermal compoundwasappliedto
thebottomof theRF transistors ange. Thecompoundis appliedsuchthat thecolor
of the gold plating on the ange is visible through it. The idea here is that the best
thermal interfaceismetal tometal contact, but sincethesurfacesof boththetransistor
andnelymachinedsurfacearenotperfect, verysmall air pocketswill exist. Theroleof
thethermal compoundisthereforenottogetinbetweenthemoduletotransistor (metal
metal) contact, but rather to ll theminuteair pockets. Thenal stepinmountingthe
transistor is to usetheappropriatescrews withtherecommendedmountingtorque. In
thisparticular case, two440screwsweredeployedwithatorqueof 5.0in.lbs.
272 Practical HF/VHF/UHF RF power amplier realization
Figure 6.22 Breadboardof 20400MHz, 150WCWRFPA module; thermal trackingsensor is
locatedinthefront center onthemodulewall.
Theinput1:1transformerconsistsof twoturnsof 0.062
//
inch, sleeved, 50Osemi rigid
coaxial cableona0.5inchCeramic Magnetics ferritetoroidpart 6.T503125T-C2050.
Theinput 4:1transformer consistsof twoturnsof 0.062
//
inch, sleeved, 25O semirigid
coaxial cableontwoof thesamecores.
Theinput couplingcapacitorsisolatetheDCbiasvoltageonthegatesandcouplethe
RF input signal fromthe1:1tothe4:1transformer.
Thetemperaturecompensated bias voltagesensor is mounted to thesidewall and
its output is fed to the gates of the transistor through a 5 KO resistor. The resistor
values can be this high as the gates of a MOSFET present an extremely high DC
impedance.
Theoutput 1:4transformer is constructedwithaheavier gauge(0.085
//
) sleeved25
and50O semirigidcoaxial cable. Threeturnsof 25O coaxial cablearewoundthrough
CeramicMagneticstoroidpart6.T874525T-C2050for the4:1transformer andsimilarly
for the 50 O coaxial cable on the output 1:1 transformer. The selection of toroidal
transformers helps also to meet EMC requirements as transformers woundontoroids
will radiatelessRF energy.
Althoughnot shown, theRFPA modulewill bepopulatedwiththreemoreidentical
amplier stages, theoutputs of whichwill beconnectedtoafour port combiner. With
150 W of output capability per stage, one module will yield about 500 W of output
power after combiner losses.
6.5 Hypothetical amplier design example 273
Figure 6.23 RubberizedEMI gasketseatedinmilledoutchannel inRFPA modulewall topsurface.
Themoduleis amachined out enclosure, thesidewalls haveachannel milled out
toseat ametallizedrubber EMC gasket (Figure6.23) that will seal theinterfacetothe
topcover. This mechanical conguration will provideaneffectivemethodof limiting
unwanted radiated emissions at a modular level. While there exist many numerical
methods to design an RFPA to meet certain electrical specications such as power
output andgain, designinganRFPA tocomply withradiated/conductedemissionsand
susceptibility represents aformidablechallenge. This is dueto thefact that many of
thethings that inuencethemodules shieldingproperties aredifcult to model. This
will tax even themost sophisticated EM simulation programs. Therearepreventative
measuresonecantakeinthedesignphaseof anRFPA:
1. Formall inductive(transformersandchokes)componentsontoroidal coresif possible.
2. Individuallyshieldeachstagewithmilledoutenclosuresandcoversthatseal thegain
stagewithEMI rubberizedgasketing.
3. Useinlinelterconsonall DC/signal feedsintoandout of theRFPA modulewhere
possible.
4. Test eachuniqueRFPA for EMC complianceat amodular level.
6.5.12 RFPA module test results
Thenal RFPA moduleschematic is showninFigure6.24, thetransistor is connected
totheinputoutput transformers, biasadjust andthermal trackingnetworks. Thevalues
274 Practical HF/VHF/UHF RF power amplier realization
Figure 6.24 Complete20400MHz, 150WCWRFPA moduleschematic.
of compensationcapacitancewill changeanddepart fromthosethat providedthebest
returnlosswithchipresistor testterminationsandwill nowbechosentosatisfythebest
input match/gainat 400MHz andbest efciencyat 150O uniformlyacrosstheband.
Thetwoportsmall signal responseof thenetworkisshowninFigure6.25a, b. Figure
6.25ashows theinput matchcharacteristics. TheRFPA modulestagehas highgainat
the20MHz low-frequency bandedgewithpoor returnlossandtheinverseat thehigh
endof theband. TheRLCfeedbacknetworksasshowninFigure6.5wereusedtolower
thegain8.3dB at thelowendof thebandandimprovetheoverall gainatness. Figure
6.25bshowsthegainatnessplot withthisfeedback. Thegainatnessis6.8dB peak
to peak, this can be compensated for on a systemlevel by using a small signal gain
equalizationnetworkor anALC loop.
The feedback capacitor is chosen primarily as a DC blocking component and the
inductancevalueischosensoastoresonateat 20MHz suchthat thefeedbackandgain
reductionismaximizedwherethedevicesgainisgreatest.
The gain and phase linearity (AMAM and AMPM distortion) response of the
amplier is shown in Figure 6.26ae. With a bias current of 2A at a Drain voltage
of 28 V
DC
, thelinearity responseof this deviceis very good for abroadband class
AB biaseddevice. Onaveragethegainlinearity is/ 0.74dB, withapeak deviation
of / 1.05dB. Fromtheaveragevaluevantagepoint, thereis/ 0.76dB of gain
linearitymargin, however thepeakgainlinearitydeviationat 400MHzis/ 1.05dB.
This leaves/ 0.45dB of gainlinearity margin. Inother words, thelow-level driver
gainandintermediatepower amplier stagegainlinearitytransfer functionswill haveto
havesubstantiallylessgainlinearityerror if theaggregateresponseof theentiresystem
(Figure6.6) istomaintainthespecicationcompliancevalueof / 1.5dB.
6.5 Hypothetical amplier design example 275
a)
b)
Figure 6.25 (a, b) Input returnloss, andsmall signal gainof theRFPA module.
276 Practical HF/VHF/UHF RF power amplier realization
a)
b)
Figure 6.26 (ae) Phaseandgainlinearitytest resultsover 20dB dynamicrangeto150W.
6.5 Hypothetical amplier design example 277
c)
d)
Figure 6.26 (cont.)
278 Practical HF/VHF/UHF RF power amplier realization
e)
Figure 6.26 (cont.)
The same is true with phase linearity; overall the average phase linearity error
is/ 4.4

with a peak deviation of / 7.9

at 400 MHz. This leaves/ 2.1

of
allowablephaselinearity error inthefront twostages of theRFPA system. This is not
leavingmuchroomfor thesestagestohaveanynonlinearitypresent intheir owntrans-
fer functions. A possibleremedy is optimizethetransfer functionof theD1020UK at
400MHz, or moveto ahigher frequency part that will haveimprovedgainandphase
linearityat 400MHz.
Figure 6.27 shows the efciency of the amplier from20400 MHz. The average
efciencyis56.8%withaworstcasevalueof 42.9%. Thedevicesdatasheetspeciesa
minimumefciencyof 50%, sobetweenthegain/phaselinearityandefciency, wehave
areasonablygoodindicationthat theoutput impedanceof thedeviceiswell matched.
Usingthetest setupof Figure6.19, theoutput of theRFPA moduleis connectedto
abinarysteppedtransmissionline, theninto50O high-power loadsthat areparalleled.
First two loads areparalleled for a2:1 VSWR. Thebinary stepped transmission line
will switch in50O coaxial cablelengths that will ultimately rotatetheimpedancein
discretesteps about aconstant 2:1VSWR circle. Whilethis is occurring, theRFPAs
frequencyandoutputisincrementedfrom20400MHzand0150W, respectively. The
spectrumanalyzer is monitoring thefrequency spectrumto conrmthat thereareno
6.5 Hypothetical amplier design example 279
50
60
70
80
e
n
c
y
(
%
)
D1020UK Drain Efficiency vs. Frequency @ Pout =150W CW
0
10
20
30
40
50
60
70
80
20 115 210 305 400
D
r
a
i
n

E
f
f
i
c
i
e
n
c
y
(
%
)
Frequency (MHz)
Figure 6.27 Broadbanddrainefciencyof the20400MHz RFPA moduleat 150WCW.
spurious signals that riseabove 45dBc. This process is thenrepeatedfor a3:1load
VSWR, then4:1, andnally5:1.
TheRFPA moduledidexhibit somespurious oscillations that were 60dBc down
belowthecarrierlevel. Onemightbelulledintoafalsesenseof securitythattheSpectral
loadpull stabilityrequirementismetasthespuriousoutputsthatdidappeararelowerthan
45dBc. This may bethecase, but toconrmthis continuously over all frequencies,
powerlevels, loadimpedances, potential modulationformatsandtemperaturesrepresent
countless hours of benchtesting, evenwithautomatedloadpull test setups. Any load
pull stabilitytest, nomatter howcomprehensive, will onlybeacursoryattempt at best.
It maybefair tosaytheamplier modulehasconditional stabilitywithinthelimitsand
scopeof theavailabletest methodsandconditions.
Figure6.28showstheRFPA moduleundergoingsixaxisrandomvibrationandtem-
peraturestress. Aswithother requirements, it isbenecial toevaluaterandomvibration
onamodular level. Randomvibrationperformanceis similar to radiatedemissions in
that anRFPA that meetsspecicationat amodular level will morethanlikelypassat a
systemlevel asthechassisthat themoduleisintegratedintowill absorbagoodportion
of therandomvibrationenergy impartedto it. Themodulewas subjectedto 15G rms
of random, six-axisvibration. Thiswas50%over therequiredspecicationonasystem
level. RFPA circuitry, byvirtueof itsnatureintermsof construction(i.e., chipcapacitors,
resistors) readilylenditself tobeinherentlyimmunetohighlevelsof mechanical shock.
TheweakpointintheRFPA circuitryisanycomponentthatprotrudeswell off of thePC
boardwithsomedegreeof masstoit. Inthisparticular case, theferritesloadedontothe
transmissionlinetransformerscansometimesimpart enoughforceonthecoaxial lines
onwhichtheyarewoundtogenerateenoughtorquethatmayeithercrackthesolderjoint
280 Practical HF/VHF/UHF RF power amplier realization
Figure 6.28 RFPA moduleundergoingvibrationandtemperaturequalicationtesting.
or delaminatethemetallizationof thePCboard. Caremust betakentosecuretheseand
other deviceswithsimilar mechanical properties.
Whilethethermal circuitisusuallyconsideredseparatefromtheelectrical network, it
iseverybitasimportantandhasperhapsmoredirectimpactonreliabilitythananything
else. The section on RFPA module construction covered the details in preparing a
proper thermal interfacebetweentheRF transistor andthemodulebase. Thenext task
is removal of the heat fromthe module of the entire system. While this is outside
the scope of this material, the objective of whatever method of cooling is deployed
is simple; thetransistors dietemperaturemust bekept as low as possible. Lower die
temperature will not only increase MTBF, but lessen the ampliers vulnerability to
failurefromexcessiveoverdriveandloadVSWRs. Bearinmind, a10

Creductionindie
temperaturemayseemtrivial, however, itcanapproximatelyhalvethefailurerateof the
transistor.
The plot in Figure 6.29 shows typical MTBF of RF power transistors versus die
temperature and drain current. Both of these two quantities can be minimized with
optimizedthermal design, systemarchitectureandoutput matchingnetworks. Thermal
designwill cover mechanical interfaces, appropriateselectionof chill plates, heat sinks
and exchangers. If the heat generated by any given architecture yields excessive die
temperatures, then adding modules may bean option to spread theheat out amongst
moredevices. Ineitherof thesecases, theoutputmatchingnetworksof theRFtransistors
must beoptimizedsuchthat theefciencyisthebest it canbe.
6.5 Hypothetical amplier design example 281
Mean Time To Failure
10000000
1000000
100000
10000
1000
120 140 160
5 A
3 A
I
D
= 1 A
T
J
, Junction Temperature (C)
M
T
F
,

M
e
a
n

T
i
m
e

T
o

F
a
i
l
u
r
e

(
H
r
s
)
180 200 220
Figure 6.29 Graphof atypical RF power transistor MTTF versusjunctiontemperatureanddrain
current.
6.5.13 Beyond the test data
Thetestdataontheamplier looksgoodandwithsomeCADbasedcircuitoptimization
will probablyyieldbetter designmargin. It isimportant tonoteat thistimethat thetest
dataitself shouldbeusedfor morethandeterminingwhether or not aparticular module
ismeetingspecication.
Datain discreteformis no doubt useful, it denes theperformanceof aparticular
module, however, RFPA dataincomparativetrendformat is far morepowerful. It not
onlydenestheperformanceof aparticular module, butalsohighlightspotential hidden
processvariationsthat maybeindecline, not onlyinthemanufactureof theRFPA, but
alsointhecomponent suppliers.
The HF/VHF/UHF RF power amplier market has never seen demands for true
commodity commercial volumes of millions of ampliers. As such, attempts to try to
apply statistical control processes suchas Six Sigmacanendinfrustration. Thereare,
however, facetsof SixSigmathat lendthemselvestolower volumeproductionruns.
Short termsigma level (or Z
st
) scorecards look at amplier test data in small lot
quantities(say, for example, 25systems). Inthisenvironment, amplier performanceof
25systemscanbestatisticallycomparedandareaswheretheamplierisrunningcloseto
specicationor experiencingatransientperiodof either marginal or exceptionallygood
performancecan beeasily highlighted and brought to attention. This attention to test
datatrendsalongwithroot causeanalysiswill uncover bothsupplier andmanufacturer
processdeviationsandaws.
282 Practical HF/VHF/UHF RF power amplier realization
Another useful processtoconsider primarilyinthetestingof RFPAsisgaugerepeata-
bilityandreproducibility(gaugeR&R). Theunderlyingconceptbehindthisisdetermin-
ingorquantifyingthevariabilityinameasurementsystembymeasuringtheperformance
of aparticular RFPA several timestodeterminerepeatability. Reproducibility isfound
byhavingseveral differentRF techniciansmeasuringtheRFPA performanceinorder to
quantifythevariationinameasurement systemresultingfromoperators. Thenet result
of agaugeR&Rstudyisthat it will ultimatelyprovideerror tolerancesthat mayrequire
certain amplier parameters to be more tightly specied in order to circumvent the
possibility that evenintheevent of thepresenceof manufacturer measurement errors,
thesystemstill arrivesat thecustomer withinspecication.
Whiletheabovetwotopicsrelatetotestdataandtestequipment, thefollowingtopics
address what the module design should go through next. A manufacturing engineer
shouldassessthemoduleconstructionwithadesignfor manufacturability(DFM) study
to ensure that its construction is amenable to low-cost assembly processes such as
semiautomatedor fullyautomatedassemblyandacceptancetest procedures(ATPs).
Further alonginthedesignprocess, qualityengineersneedtobeinvolvedtoperform
failuremodeandeffectsanalysis(toidentifyRFPA designandprocessproblemsbefore
they occur). Wiebull analysis along with calculated and demonstrated MTBFs will
provideinsight intotheexpectedlifetimeof thenewlymintedRFPA design.
Summary/conclusions: current technology/future trends
in amplier design
It hasbeensaidherethat twothingsprimarilyoccur inanyRF power amplier:
r
efcient amplication;
r
efcient or maximumpower transfer.
Onemust amplifyasignal, transfer it tothenext stage, thendoit all over again. The
devicesresponsiblefor amplicationareobviouslytransistors. Soif wewantedtoknow
what mayhappeninthefuture, wecanextrapolatewhat hasoccurredinthepast.
Twenty-veyearsago,theclassof RFtransistorsmostcommonwereBJ Ts.MOSFETS
werestartingtobecomecommerciallyavailable. Sincethen, MOSFETs(whethervertical
or lateral) havebeentheworkhorses of HF/VHF/UHF RF power ampliers. BJ Ts for
now, appear tobeindecline. Inthe198590timeframe, MOSFETS that couldoperate
upto175MHz at power levelsof 300600Wbecameavailable. Inthelast year, 50V
LDMOSFETshavearrivedthatcandeliver 1KWupto500MHz. Whatwastruethenis
still truenow, namely operatingsuchhigh-power devicespresentsthesamechallenges
of heat removal andtheAC couplingof high-power, high-current RF signals.
GaNandGaAsFET devicescontinuetondthemselvesinampliersthatarebreaking
barriersinultrabroadbandwidths.
ItisapparentthatacontinuingtrendinRFpowertransistortechnologycanbeexpected
with higher and higher power outputs and broader bandwidths found in smaller or
equivalent sized packages. No one, however, has found a way to produce a device
References 283
that isdistortionfreeor hassignicantly improvedefciency performance; all devices
discussedhavevaryingdegreesof nonlinear transfer functionsandinefciencies. This
isnot tosayadeviceof thisnaturewill never arrive, onlythat inover ahalf centuryof
transistor development, it simplyhasnot.
Onthetopicof matchingandmaximumpowertransfer, sufceittosay, if thecomputer
industryprogressedwiththesamespeed, wewouldall still beusingabacuses. A quick
reviewononeof theearliestpapersontransmissionlinetransformersbyGuanella[5]and
Ruthroff [6] showthat theyremainvirtuallyunchangedinconstructionandapplication
sincethe1940s. It isatestimonytowhat elegant andefcient devicestheyare, andalso
tohowdifcult theyaretoimproveupon.
So if we match transistors the same way and if they really are not more linear or
muchlessefcient, thenwherearetheimprovementstocomefrom?Basedonadvances
inthecellular andplasmaprocessingamplier markets, improvement onrawamplier
performancehas, canandwill comeintheformof advancedamplier error correction
techniques such as predistortion, feed-forward and high-speed, digitally based ALC
loops. Theseadvancesevolveprimarilyfromadvancesinhigher speedanalogtodigital
conversion, digital signal processinganderror correction.
Dedication
I wouldliketodedicatethischapter tomywife, CatherineLeigh, sonJ ustinDaniel, and
daughter MikaelaSiennaMyer.
Acknowledgments
Theauthor wouldliketoacknowledgethecontributionsof Robert Schoepfer, Gregory
Muller, andThuyLu.
References
1. D. Myer, Synthesis of equal delay transmission line transformer networks, Microw. J.,
vol. 35, no. 3, pp. 106114, March1992.
2. N. DyeandH. Granberg, RadioFrequencyTransistors-PrinciplesandPractical Applications,
Butterworth-Heinemann, 1993, pp. 193197.
3. R. K. Blocksome, A binary stepped transmission line, R.F. Des., J uly/August 1982,
pp. 2229.
4. N. Padeld, Mountingguidelinesfor SEMELAB RF MOSFETs SemelabPLC Application
Note, pp. 19.
5. G. GuanellaNewmethodof impedancematchinginradio-frequencycircuits, BrownBoveri
Rev., Sept. 1944, pp. 327329.
6. C. L. Ruthroff, Some broad-band transformers, Proc. IRE, vol. 47, pp. 13371342,
Aug. 1959.
7 Microwave hybrid amplier
realization
Dominic FitzPatrick
PoweRFul Microwave
7.1 Introduction
The variety of approaches taken in the design of power ampliers is vast and the
subdividing of the approaches into various categories, such as hybrid is (like the
term microwave itself) a generalization in which the edges are somewhat blurred.
TheCambridgeAdvancedDictionarydenes hybridas somethingthat has beenpro-
duced fromtwo different types. . . especially to get better characteristics, or anything
that is amixtureof two very different things. In this casethemixtureis considered
to be of lumped and distributed components. Until recently a hybrid amplier was
consideredas onewhichusedpackageddevices, however instrivingto achievebetter
performanceat higher frequencies discretedevices and MMICs havebeen integrated
intocircuitscontainingbothdistributedcircuitsanddiscretecomponents, seeFigure7.1.
Hybridampliershavethusbeencreatedasaneffort tocherrypick thebest technical
solutionswithinaneconomicframework.
Theplaceof hybridampliersinthemarket rangesfromtheprototypeandfeasibil-
ity proving stages of product development, to volumemanufacturing, to low-quantity
highest performanceproducts. Microwavehybridpower ampliers (MHPAs) areused
in low-volume applications such high-energy physics particle accelerators to volume
applicationsinmobilephonebasestations. Solidstatepowersampliers(SSPAs) have
becomethetechnology of choicefor themobilecommunications market andasignif-
icant proportionof thesatellitecommunications market. Theadvent of widebandgap
materials has seen hugeimprovements in bandwidth performancebelow 6 GHz, and
thenext generationof commercial products will seetheseadvances cover X bandand
beyond.
Many electrical engineering students undertakean amplier design project as part
of their studies, whilelargeengineeringcompanieshavefoundamplier designgroups
expensiveanddifcult tomanage(oftenleadingtovery successful spin-out compa-
nies!). Somemanagers havestruggledto understandwhy thedesignanddevelopment
of MHPAs is not as predictable a process as the design of other electronic system
components. This chapter seeks to highlight someof thepitfalls that haveso troubled
somedesignprogrammes; itwill alsohopefullyhelpdesignersinselectingtheoptimum
approach to meeting customer requirements. Too often thetermblack art has been
appliedtowhat isineffect aprocesswhichispoorly controlled, andthisstartsright at
theverybasiclevel of thedesign. . . .
7.2 Printed circuit boards 285
Figure 7.1 Mixedtechnologymicrowavehybridpower amplier. Photocourtesyof CreeInc.
12.3GHz SiC 50WAmplier, www.cree.com.
7.2 Printed circuit boards
It might seemstrangetostart achapter onMHPAswithadiscussiononprintedcircuit
boards (PCBs). However, as with housebuilding, this is thefoundation of our struc-
tureandapoor choiceherewill leadto thenal designsolutionbeingcompromised.
MicrowavePCBs aredivided into two categories, hard and soft substrates. Hard sub-
strates are primarily alumina, a ceramic with a tightly controlled dielectric constant
(
r
9.8) althoughother materials suchas glass andsapphireareused. Aluminasub-
strates benet fromhighthermal conductivity andhighoperatingtemperatures which
hasmadethempopular inmilitaryandspaceapplications. Thepatterningof theconduc-
tivecircuitsisachievedeither bydeposition(thicklm) or etching(thinlm). Resistors
canbeintegratedintothecircuitsbyadjustingthemetalization(suchasusinganickel-
chromiumlayer). Groundingisanissueasthematerial isbrittleandholeseither needto
bepunchedwhenthematerial isinitsgreen, unbakedformor laser drilled. Similarly,
themounting or attachment of thesubstratecan causeproblems dueto poor thermal
coefcient of expansionmismatchwithsomeof thecommonmetal housingmaterials,
seeFigure7.2. Newelectricallyconductiveadhesiveshavebeendevelopedwhichallow
asufcientamountof givebetweenthelayers, however thisisneither acheapmaterial
nor isthedispensingandcuringeasy. Metal alloyswithacloser thermal expansionhave
beendevelopedaswill bediscussedlater.
Soft substrateisthetermappliedtoanowvast rangeof productsthat arecomposite
materials, either breor particlebased. Theall-pervadingFR4breglassmaterial of the
conventional electronicsindustryisinappropriateforMHPAsasthedielectricconstantis
poorlycontrolledandthedielectriclosseshigh. Attheother endof thespectrumispure
polytetrauoroethylene(PTFE), asynthetic uoropolymer which has alow dielectric
constant, whichcanbetightly controlled, withlowloss. However, it has poor thermal
performance.BymixingPTFEwithbrebasedboardsthenlowloss,controlleddielectric
constant, andareasonablethermal performancecanbeachieved. Additionally, proling
anddrillingtheboards is cheapandrelatively simple. By addingceramic particles the
286 Microwave hybrid amplier realization
0
0
50
100
150
200
250
300
350
400
450
500
10 20
Thermal Coefficient of Expansion 10
6
/C
T
h
e
r
m
a
l

C
o
n
d
u
c
t
i
v
i
t
y

(
W
/
m

C
)
30 40 50 60 70 80
Solder (PbSn)
Alumina
Kovar
GaAs
Molybdenum
Tungsten
Aluminium
Gold
Copper
Silver
90 100
Teflon
Figure 7.2 Thermal propertiesof commonmicrowavematerials.
dielectric constant can be adjusted, even to the extent that they can be close to that
of alumina, thus offeringcircuits of similar dimensions. A further advantageof these
substratesisthat theyareeasilylaminated, soawiderangeof thicknessesareavailable,
andtheycanbeprovidedwitharangeof metal backings. Processingissimilar tothatof
conventional circuitboardsexceptthatwithPTFE basedmaterialsanadditional stageto
roughenup thesurfacesisnecessary inorder tosuccessfully platetothesurface. As
thedemandfor circuitsfor themobilephoneindustry rosethensothenumber of PCB
processorswhocouldhandlePTFE substratesincreasedandpricesfell.
Thekey parameters of substrates commonly usedinhybridampliers areshownin
Table7.1. Not includedinthetableiscost, thisisbecausewhenconsideringthecost of
acircuit substrateoneshouldnot only consider thepurchasepriceof therawmaterial
but also theprocessing costs. For example, a circuit on a low dielectric maybethree
times larger thanonahighdielectric. Furthermore, thenishedPCB cost needs to be
put intothecontext of thewholeamplier itself. What is thecost/Wof output power?
Knowingthisgureof merit will aidthedecisioninchoosinganoutput PCB; isit more
cost effectivetoopt for anexpensivelow-lossboardor acheaper higher lossmaterial?
Non-PTFEmaterialshavebeendevelopedsothatstandardPCBfabricationtechniques
canbeemployed. A sidebenet of this development has beenthat thestepchangein
dielectric constant,
r
, at around room temperature has been removed. The typical
changein
r
dueto thecrystallinestructurealtering in aPTFE material can beseen
in Figure 7.3. This change causes equally sharp changes in the equivalent electrical
lengthof transmissionlines. Somematerials(suchasceramicloadedPTFE), whilenot
displaying as signicant an inection point, haveamuch greater overall changein
r
7.2 Printed circuit boards 287
Table 7.1 Properties of common substrates used in MHPAs
Soft
RT/duroid
5880
RT/duroid
TMM10i
RT/duroid
R4003C
Substrates
Property Units
Hard
Alumina
1
99.6%
Dielectricconstant 9.9 2.2 9.8 3.55
Thermal coefcient of
dielectricconstant
ppm/

C 125 43 40
Dielectricloss(tanL) 0.0001 0.0009 0.002 0.0027
Dielectricstrength AC-kV/mm 8.7 285 31.2
Volumeresistivity O-cm 10
14
210
13
210
8
1.710
16
Thermal conductivity W/m/K 26.6 0.20 0.76 0.64
Coefcient of thermal
expansion
ppm/

C 8.2 X= 31
Y = 48
Z= 237
X= 16
Y = 16
Z= 20
X= 11
Y = 14
Z= 46
Water absorption
(ASTM-373)
% 0 0.015 0.16 0.06
Type Ceramic PTFE/
microbre
Ceramicloaded
thermoset
plastic
Wovenglass,
ceramiclled
thermoset
1.008
1.006
1.004
1.002
1.000
0.998
0.996
0.994
0.992
0.990
0.988
50 30 10 10 30 50
TempC
Chart 1: RO4000 Series Materials
Dielectric Constant vs. Temperature
70 90
PTFE/Woven Glass RO4350 RO4003
E
r
(
T
)
E
r
(
2
5
)
110 130 150
Figure 7.3 Relativechangeindielectricconstant withtemperature. Courtesyof RogersCorp.
www.rogerscorp.com.
withtemperature. Another advantageof thermoset materials is that they do not soften
whenheatedandthusaremoresuitablefor wirebondingapplications.
Intheideal worldwewouldwant thesubstratetobeatotallyhomogeneousmaterial
wherethedielectric constant is isotropic, i.e., has aconsistent valuethroughout. Due
tothemanufacturingprocessesandmaterial compositionsthisdoesnot happenandthe
1
CoorsTek, ADS-996.
288 Microwave hybrid amplier realization
dielectric constant may even vary with orientation of board, thus if thecircuit layout
andthesubstrateorientationisnot consistent betweenmanufacturingrunstheremaybe
variationsinperformance. It isobviously essential that boardmanufacturersareaware
of thesedifferencesandtakeaccountaccordingly. Theeffectsof thesedielectricconstant
changesaremostdramaticinhighQelementssuchasresonators. Thus, isotropyof the
substratematerial isanimportantconsiderationinthedesignof narrow-bandampliers.
It can also affect wide-band designs where coupling structures are produced on the
PCB, such as Lange couplers [1]. Generally speaking, the ner and more randomly
placed the loading materials in substrate the lower will be the variation in dielectric
constant, thuswovenglassbasedmaterialstendtoexhibit thehighest anisotropy. Very
consistent dielectricmaterialssuchaspurePTFE canbeused; however thesehavetheir
ownproblems.
Thethermal conductivityof thePCB isimportant inmediumpower amplierswhere
surfacemount devices areused, althougheveninthesecases themajority of theheat
generatedisconductedtotheheatsinkthroughthegroundingvias. Inhigh-power ampli-
ers thepower devices areboltedthroughthePCB to thehousingoor or directly to
aheatsink. Theamount of power dissipatedinthecircuit structuresthemselvesshould
not beunderestimated. If apower devicedelivers100Wandtheoutput circuit follow-
ingit has 0.5dB of loss this means that 11W is dissipatedinthePCB (assumingthe
majority of theloss is not radiated). Therearetwo mainmethods of loss inthePCB,
resistiveloss inthemetal conductors anddielectric loss inthesubstrate. Theconduc-
tor loss is determined by the properties and dimensions of the metal used while the
substratelossisdependant not only onthelossof thematerial itself (quantiedby tan
) but by thepercentageof theelectric elds owingthroughthesubstrate. Thesetwo
haveatendency towork against eachother; toget thelowest substrateloss onewould
look to use as thin a substrate as possible and narrower lines, while for minimizing
conductor lossoneneedswider lines. Thethicknessof atransmissionline, particularly
in MHPAs, is affected by two considerations, the DC current and the frequency of
operation. For DC currentsthecross-sectional areaof thetransmissionlineisinversely
proportional totheresistance, i.e., doublethecross-sectional area, halvetheresistance.
At microwavefrequencies however, thecurrents areowingonly inalimitedpropor-
tionof thethickness, theextent of whichis referredto as theskindepth. This term
leads to a common misconception, the current ows in the electrical surface closest
to theground plane(theelectrical elds arebetween thesurfaceof theground plane
andtheundersideof thetransmissionline), thus platingupor changingthemetaliza-
tionof thetopsurfaceconductor may makenodifference. Theformulafor skindepth,
, (inmetres) is:
=

2
2 f
o

r
(7.1)
where is thebulk resistivity of theconductor (O-cm); f is thefrequency (Hz);
o
is
thepermeabilityconstant (H/m) = 4 10
7
, and
r
istherelativepermeability.
7.2 Printed circuit boards 289
Table 7.2 Bulk resistivity of commonly used metals
Material Aluminum Copper Chromium Gold Nickel Palladium Silver
BulkResistivity
(O-cm)
2.65 1.67 18.0 2.30 8.71 10.8 1.59
Table 7.2 lists the bulk resistivities of some common materials. For example, at
5GHztheskindepthinacopper conductor will be0.92m, whilea
1
/
2
oz. copper plated
PCB hasacopper thicknessof 18m. A commonruleof thumbisthat thecopper PCB
traceshouldbeat least 5 tominimizeloss. Alsonotethat thepurposeof goldplating
or ashing on microwave PCBs is to passivatethesurfaceor improve thecontact of
bondingareas, not to reducetheRF resistivity. As theRF current owis primarily on
theundersideof thetrackthesurfaceroughnessof thesubstratewill impact theloss.
Themostcommonconductormaterial iscopper. Therearetwostandardapproachesto
attachingthecopper tothesubstratematerial. Thelowest cost andhencemost common
methodis electrodepositedor ED copper. Thealternativeis rolledcopper, wherethin
sheetsof copper arebondedtothesubstratematerial. Rolledcopper haslower insertion
loss due to the uniformity of the material and the reduced surface roughness; this is
particularly noticeableas frequency increases. However, ED has a better adhesion to
thesubstrate, i.e., higher peel strength. Traditionally, bothtypesarespeciedinounces,
thiscomesfromthenumber of ouncesof copper per squarefoot of boardarea, (1oz.
0.0355mm). Copper cannot beleft bareandsoit iscommonpracticetoeither plateor
coatthecopper. Suchtreatmentsrangefromashingwithanonreactivemetal suchas
gold(typically5m), tohot air solder leveling(HASL), andconductivepolymers.
When selecting the substrate material the most signicant consideration is the
impedancerangethatcanberealized. Thegeneral ruleof microwavedesign, Watchout
whendimensions approachasignicant fractionof awavelength applies here. If the
substratethicknessistoolargetheninsteadof theelectromagnetic(EM)eldsformingin
thequasi-TEMmode, other modespropagate. A goodruleof thumbisthatthethickness
of thesubstrateshouldnot exceed20

phaselengthat thehighest operatingfrequency.


Table7.3summarizestheimpedances, circuit dimensions, andcurrent limitationsfor a
rangeof substratematerials.
Other considerations in theselection of thedielectric and its thickness arethecur-
rent capacity required (usually limited by DC bias currents), and thesizeof discrete
components, suchasdevicetabs. Inlinear designsoftwaremodelstherearelimitations
ontheratio of track widthto substratethickness, whichusually restrict theminimum
impedanceto25O. Thereforeit maybenecessarytouseanEM analysisfor partsof
thecircuit wherewidelinesareunavoidable. Whendecidingwhether ahardsubstrateis
theoptimumsolutionitisimportanttoconsiderthesurfaceareaof thecircuit. Generally,
hardsubstratesarelimitedtoamaximumsizeof 5050mm. Hence, acircuitmayneed
tobemadefromanumber of ceramictiles. Conversely, antennashavebeenmadeon
soft substrates over 1meter long. Typically, however, blank soft substrates sizes range
from250250mmto800600mm.
290 Microwave hybrid amplier realization
Table 7.3 Typical substrate trace dimensions with approximate current rating
Substrate
r
Thickness
(mm)
50O
width
(mm)
,4(90

)
@10GHz
(mm)
Current
for a30

C
temp. rise
2
(A)
Alumina 9.8 0.635 0.61 2.87
(Thin-lm) 12
(Thick-lm) 12
CuClad
217LX
2.17 0.787 2.40 5.46 5
RO4350B 3.66 0.508 1.09 4.41 3.8
TMMi 10 9.8 0.508 0.48 2.90 2.3
Figure 7.4 Impact of processingontrackdimensionsandedgecoupling(slopesexaggerated).
Inthedesignof microwavecircuits it is easy to assumethat thecircuit dimensions
producedbytheCADsoftwarearethosethatwill befabricated; howeveritisessential to
understandtheprocessingthat will beinvolvedinmanufacturingthePCBsandtotake
into account theimpacts. For example, wheretheproductionprocess involves etching
awaytheunwantedcoppertheactual shapeof thecross-sectionof thetrackistrapezoidal,
andthesizeof theetchangleisproportional tothethicknessof thetrack, seeFigure7.4.
TherearetwomainPCBconductorcreationapproaches, (a) subtractive, and(b) additive.
In thesubtractiveprocess theetch angleresults in thetrack width being wider at the
bottom(remember thatthisiswhatdeterminestheRF impedance), whileintheadditive
processthetrackisplatedupfromathinlayer andthusiswider at thetop. Alsoworthy
of note is that the edge coupling between the adjacent tracks (and ground planes) is
2
Approximateasthisdependsonanumber of factorsincludingperipheral circuitfeatures, tracetermination,
backingmaterial andxingmethod(seelater).
7.2 Printed circuit boards 291
assumedinmostsimulator modelstobebetweenvertical walls(or whatever theetching
processwasof thesamplefromwhichthemodelswerederived). A further complication
isthat somePCB manufacturerstakeintoaccount theetchfactor of their process, while
othersdont, hencethedesignermustknowwhetherornottotakethisintoaccountwhen
creatingthemask.
Groundingis animportant part of microwavecircuit designs. Typically, theground
planeisacontinuousconductor onthereversesideof thesubstrate, soit isnecessaryto
connect toit. Thestandardapproachistodrill thesubstratematerial andthenplatethe
holewithaconductor(platedthroughhole PTH)tocreateavia.WithmanyRFsubstrate
materials this has its ownproblems. PTFE materials aredifcult tobondtoandhence
thesurfaceof theholemust beroughenedtopromoteadhesionof thecopper duringthe
platingprocess, andthisisparticularlydifculttoachieveinthrough-holes. Also, dueto
thedifferingthermal coefcientsof expansionbetweentheconductor andthesubstrate,
cracking around the top of the via can result, particularly where the PCB undergoes
signicant temperaturecycling. Thesefactorsthereforepushthedesigner touselarger
holesandthicker plating. Rather thanusingviaholes, slotsmaybecut intheboardand
theiredgesplated, butthismaybeanonstandardapproachforsomePCBmanufacturers,
who would normally do slot cutting after plating and should thereforebehighlighted
in the requirement drawings. Inserting pins through boards is acceptable for simple
prototypecircuits, but theresult is generally not ushwiththesubstratesurfaces and
thusmaynecessitateprolingof theboxoor or restrict theplacement of components.
Substratescanbesuppliedmetalizedonthebacksidewhichmakesitpossibletodirectly
solder thesubstrateto themetal; however this will requireusingbackgroundheating.
Whenusingaluminumbackingit isnecessary toplatethealuminumrst (not atrivial
process), hencecopper or brassbackingarepreferred. Electricallyconductiveadhesives
arealso availableandcanbeusedto bondthemetal andsubstratetogether. For large
bondingareastheadhesivecanbesuppliedinlmswhichmaybemoreconvenient.
Theuseof metal backedsubstratesispopular for anumber of reasons:
r
it providesgoodheat sinking;
r
edgemountedconnectorscanbedirectlyattached;
r
goodmechanical basefor mountinglargecomponents;
r
shrinkageandwarpingof thePCB isreduced;
r
easier attachment tohousings.
However, weight, substrate, andprocessingcostsareincreased. Metal backedmaterials
are particularly popular for test jigs and prototypes where the expense of a custom
housingcanbeavoidedandweight isnot asignicant issue.
MountingPCBs withinhousings or directly to heat sinks canbedonesimply using
screws to clamp theboard in place. However, careshould betaken to ensurethat the
contactbetweentheboardandthebackingmaterial isconsistentbytheuseof appropriate
screwheadsizeandquantitydependentuponthestiffnessof thesubstrate. Areaswhere
particular careshouldbetakenarearoundtheinput andoutput connectionsandwhere
pocketsaremachinedoutfor thedevices. Gapsbetweenthegroundplaneonthebackof
thesubstrateandtheboxoormayproduceresonantcavitieswhichwill altertheresponse
292 Microwave hybrid amplier realization
of theamplier. They canalso providefeedback paths leadingto oscillations. Screws
shouldhavemetal landsunder them, especiallywithPTFE boardsasunder compression
thePTFE hasatendencytocreepawayfromunder thescrew. Itisadvisabletousecap
headstoavoiddamagefromscrewdriversslipping, especiallywheretherearenetracks,
or delicateair-woundinductors. Crinkleor springwashersareusuallyadequatetolock
thescrews. Thestandardsizeissometimeslarger thanthescrewhead, sowherespace
isat apremiumthenext sizedownor animperial sizewill t thebodyof thescrewand
not protrudefar beyondthescrewhead.
Direct screw clamping is not recommended for hard substrate materials such as
aluminaduetothebrittlenessof thematerial. Most commonlythesubstrateissoldered
or epoxied directly onto a metal carrier which in turn is screwed into the housing.
Alternatively, springclips canbeusedto holdthetiles inplace. Modernepoxies have
provedincreasingly popular inmountingbothsubstrates andactivedevices. Thetype
of epoxy usedmust beconsideredcarefully, somecancrack under thermal stress. For
suchapplicationsthermo-plasticadhesiveswhichhavesomegiveareabettersolution.
Conventional PCBs may also besoldered directly into housings. This gives excellent
thermal andelectrical conductivitywiththeaddedadvantagethat noboardreal estate
istakenupby themountingscrews. Inorder toget anevensolder join, solder pasteis
screenprintedonto theundersideof thePCB or apreformused. Pressureneeds to be
appliedacrossthePCBduringreowtopreventareasfromliftingand, if notconstrained
by walls, tokeepthePCB correctly aligned. Attentionmust bepaidtothetemperature
distributionacross theunit as all parts of thesolder joint must reachthereowpoint,
but without going so high as to damagethesubstrate, housing plating or degradethe
solder. A low-ux solder shouldbeused, particularly whenlargeboards areinvolved,
otherwisepocketsof uxmayformunder theboard. Somesolder pastesmayrequirea
dryingperiodafter applicationtothePCB toallowsolventstodissipate.
An alternativeconguration to microstrip is to usecoplanar waveguide(CPW). In
thisconstructionmethodthegroundplaneisbrought tothetopsurface, thisproducesa
circuitwithverytightlycontainedeldswhichisthereforelesssusceptibletoproximity
effectssuchaslidsandtoradiativecoupling. Althoughpopular for low-power devices
it hasdrawbacksfor MHPAs, duetotheneedtoheatsinkdevices. Withangemounted
devicesthereisadiscontinuityintheoddandevenmodesatthedevicepackagejunction,
andfor surfacemountthereneedstobeagroundplaneconnectedtowithvias. Typically
thesubstraterequiresalargenumber of viastoensuregroundcontinuity, hencelosing
oneof theadvantagesof CPW. Anothermajordrawbackisthattuningof lineimpedances
inCPWisverydifcult comparedwithmicrostrip.
ItisoftennecessarytohaveslotsinthePCBmaterial forangemountedcomponents.
Rather than being directly attached to the PCB these are bolted to the housing oor
with leads soldered to the substrate tracks. For microwave devices the alignment of
thePCB andhousingis critical so that gaps areminimal as they cancauseunwanted
impedancechangesandresonances, asshowninFigure7.5. Asthesamerequirements
areimposedontheinput andoutput connector launchesthedimensional toleranceson
thehousingandPCBcanbeextremelytight. Onesolution, particularlysuitabletoPCBs
solderedintohousingsor test xtures, istomachinetheslotsintothePCB andhousing
simultaneously. This ensures precise alignment between the two. Unfortunately, this
7.3 Housing 293
Substrate
Metal Backing
Housing
S
Z = 377
s
w
L = Inductance of RF Link (e.g. bondwire or device tab)
Note: H is in metres
RF Link
W, width of microstrip
H
X
L W
H
Z Z
X = Z tan
2H

Figure 7.5 Groundpatheffects.


optionisnotpossiblewithhardsubstrates, andwithsoftsubstratesheldinplaceamuch
higher degreeof clampingisrequired, andcaremust betakentoensureswarf isnot
forcedintothegapsbetweenboardandgroundplane.
A nal consideration with soft substrates is their moistureabsorption. During pro-
cessing and cleaning, PCBs aresubjected to alargenumber of chemicals. Thesecan
causeavarietyof problems, fromproducingcorrosiveliquidstochangingtheelectrical
propertiesof thesubstrate. Forinstance, thedielectricconstantcanalterandif thecircuit
is tunedto compensatefor this, thenover timeas thesubstratedries out the
r
will
changeandhence, thecircuit performancechanges. Thoroughcleaningfollowedby a
bakingout stageisnecessary.
7.3 Housing
OtherthanforprototypesandtestjigsMHPAsrequirehousing. Althoughthismayatrst
seemtrivial, beforeconsideringtheconstructionof asuitablehousingtherequirements
shouldbeconsidered.
r
interference: tostopthesignalsintheamplier interferingwithandbeinginterfered
byexternal signals, circuitsandmaterials;
r
protection: toprevent thecircuitsbeingharmedbymechanical or chemical action;
r
heat sinking: theheat generatedby thepower components needs toberemovedina
controlledmanner. Thehousingcaneither incorporateor providetheconnectionto
themethodof heat removal;
r
mountingandconnecting: theamplier does not exist inisolation; it requiressignal
connections, supplyconnectionsandamethodof xingtoitssurroundings. Therole
of thehousingistoensurethat thesearereliableandconvenient.
294 Microwave hybrid amplier realization
Therelativeimportanceof eachof thesemust bebalancedagainst theother consider-
ations of cost and weight. Thesameapproach would not betaken in asafety-critical
applicationinaharshenvironmental toalaboratorytest amplier. Asinall elementsof
design, theendsolutionisacompromisebetweenconictingneeds.
7.3.1 Materials
The most common material for MHPA housings is aluminum. It is relatively cheap,
easilymachined, strongandlightweight. Ithasgoodelectrical andthermal conductivity
andsoisexcellent for screeningandheat sinking. Themaindrawback isthat it cannot
bedirectly solderedto. This canbeovercomethroughplating, but this is amultistage
process withtypically rst anickel seedlayer followedby thegoldor tinplating. It is
rarethat thealuminumis left untreated as although theoxidelayer that forms on the
surfaceisfairlyun-reactive, thereisadanger thatsmall residuesof acidicsolutionswill
beleft behindfromtheux usedinmany solders. Thesecanreact withthealuminum
oxidetoformsaltswhichcancausedendritestogrowleadingtoshort circuits. A low-
costaluminumsurfacetreatmentischromateconversion, commonlyknownbythebrand
namesIriditeorAlodine. Someformsof thisprocessarebannedundertheRestrictionof
HazardousSubstances(RoHS) legislationsastheycontaintoxichexavalent chromium.
Careshouldalsobetakenastheelectrical conductivitycanbeinconsistentanddependent
uponthickness. Thecoatingcannot besolderedto. For high-performancehermetically
sealedsystems Kovar is usedas analternativeto aluminumas will bediscussedlater.
Kovarsthermal coefcientof expansion(TCE) iscloser tothatof GaAs, seeFigure7.2;
however, it isthreetimesdenser anditsthermal conductivityisconsiderablylower than
aluminumwhichisanobviousdrawbackfor MHPAs. Aluminumalloy6061ispreferred
duetoitsgoodmechanical propertiesandmachinability. Wherethelidistobewelded
totheunit (rather thanclamped) thenaluminumalloy 4047ispreferred. Thiscontains
siliconwhichimprovestheductilityof theweldandreducescracking.
7.3.2 Sealing and hermeticity
Very few things are as likely to illicit impassioned debate amongst MHPA design
engineersastheissueof theappropriatelevel of sealing. Whenonelooksintotheareain
depth, onecaneasilybeleadtodespair that thereisnothingthat canbedonetoprevent
moistureingress. However, weshouldtakeheart fromthefact that systemscontinueto
operatefor many years, andinfact theruggednessandsurvivability of somerelatively
cheapproductssuchasmobilephones, satellitelow-noisedown-converters(LNBs), and
GPS receivers is impressive. The aimshould be to provide reliability commensurate
withthecost of theunit (includingcost of replacement). Infact manymilitarysystems
providers arenow looking at availability of parts rather than requirements to survive
storageof 20yearsinharshconditions.
Protectionof amplier circuitsisrequirednot onlyfrommechanical damagebut also
fromcorrosion and vapour ingress. In the fabrication of ampliers there are a large
number of chemicalsused, fromthelubricant usedduringmachiningof thehousing, to
7.3 Housing 295
thoseusedinthePCB processingandhousingplating, to theuxes insolders. These,
especiallyinthepresenceof water, canformparticularlycorrosivesolutions. Condensa-
tioncancauseshort circuitsor changestoperformanceinRF applicationsbychanging
theelectric elds aroundtransmissionlines. Avionic systems haveprovedparticularly
vulnerabletocondensation, whereequipmentcanbesittinginwarmmoistair, andwithin
onlyafewminutesbeat highaltitudeandlowtemperatures. Thetemperatureat which
condensationforms iscalledtheDewPoint andvaries withthehumidity of theair; as
theair temperatureincreasessodoesitsabilitytohold water.
The extent of the protection required is dependent upon the expected lifetime of
theproduct, temperaturerange(higher temperaturesincreasechemical activity hence
storagetemperatureisakeyfactor), andsensitivityof thecircuitstocorrosion, whichin
turndependsuponthematerialsusedandhowtheactivecircuitsarepackaged. Standards
wereestablishedfor military systemscoveringmany areasof thedesignandtestingof
electrical systems. MIL-M-38510, thegeneral specicationfor microcircuits, required
that all hybridmicrocircuitsthat containactivesemiconductorsshouldbehermetically
sealed. The MIL standards are no longer supported, mainly because they could not
keeppacewiththespeedof component andprocess developments, but many of their
requirements havebeenassumedinto requirement specications. Recent work [2] has
shown that hermetic sealing to MIL-STD 883 may not beadequateand that modern
plasticpackagingcanprovidebetter reliability. A bigfear intheearlydays, particularly
of GaAscircuits, washydrogenpoisoningandsohydrogengetterswereincorporated
intocircuitpackages.Improvedpassivationof theactivedeviceshasresultedinthevirtual
eliminationof thisissue.
A complete seal against moisture ingress is difcult if not impossible to achieve.
Welding, brazing, or solderingall of thejoints canproduceanadequateseal; however
thereistheneedtoprovideRF andDC interfacesandaccesstotheunit for repair and
maintenance. Moisturecaningressintoahousinginthreeways:
r
diffusion;
r
capillaryaction;
r
breathing.
Diffusion: water vapour will diffuse in if the partial pressure differential is inwards
even if there is an absolute pressure differential in the opposite direction. Hence, a
conditioncanexist wherealler gasinthehousing(suchasnitrogen) canbediffusing
outwardswhilethewater vapour isdiffusinginwards. It isoftennot realizedthat water
vapour moleculesaresmaller thantheother maingasesintheatmosphere oxygenand
nitrogen. Water vapour (H
2
O) has amolecular weight of 18 as compared to nitrogen
(N
2
), 28andoxygen(O
2
), 32. Thus, thewater vapour seal isthehardest toachieve. The
measureof diffusion is moisturevapour transmission rate(MVTR) and is dependent
upontherelativeconcentrationsoneither sideof thebarrier. Table7.4showstheMVTR
values for different composition rubber o-ring seals testedunder identical conditions.
No material provides a100%seal, therelativegas permittivity of various materials is
giveninTable7.5[3], however thereal questionshouldbe, whendoes aleak become
critical, whichcomesbacktotheintendedlifeandoperatingandstorageconditions.
296 Microwave hybrid amplier realization
Table 7.4 Relative rubber o-ring sealing
performance
O-ringmaterial
MVTR
(gm
1
day
1
)
Fluorocarbon 9.510
4
Nitrile 4010
4
Silicone 9010
4
Polyurethane 13010
4
Table 7.5 Relative gas permittivity for various materials
Glasses/
Nylon Silicones Epoxies LCPs ceramics Metals
1 10
2
10
4
10
6
10
8
10
10
Capillaryaction: water may formontheoutsideof theunit fromdirect exposureor
condensation, dependingupontheenvironment. Therateat whichthewater permeates
theunit will dependuponthenatureof thematerial inwhichany holeor crack exists.
Obvious risk areas arealongthelidedgeandany screwholes that break throughinto
thechamber.
Breathing: for unsealedunitsthemajorityof thewater ingresswill bethroughbreath-
ing. Asthepressurechanges betweentheinsideandoutsideof theunit therewill bea
movement of air. Thus, moisturewill becontainedwithintheunit.
A degreeof protectiontocomponentsandtrackscanbeofferedbyconformal coatings,
however thesecanaffecttheRF impedanceof circuits, thereisalsotheproblemof when
to apply them. Application beforetuning may mean that thecoating integrity will be
breachedif itisnecessarytochangecomponentsduringtuningandtest, however adding
thecoatingafterwardsrunstheriskof alteringthecircuitperformanceinanuncontrolled
manner. Also, althoughaconformal coatingprotectsfromshort circuits, condensation
abovetheprotectivelayer canstill detunethematchingcircuits.
The ultimate solution is to hermetically seal the amplier. The argument against
hermetically sealing is that if harmful chemicals build up over timethen by allowing
theunit tobreathe theconcentrationis reduced, whereas inasealedchamber it may
reachharmful levels andcausecorrosion. Themainproblemisduetowater moisture,
thusnot onlyisit necessarytoseal theamplier but alsotoreducethemoisturecontent
within the unit (obviously cleaning so as to remove as much of the contaminants as
possible is essential), therefore air is driven out before sealing and replaced with an
inert gassuchasnitrogen. Thedecisiontohermeticallyseal shouldnot betakenlightly
as this requires special components suchas theRF andDC connectors as well as the
lidseal itself, andwitheachonethereis not only ahigh-integrity seal to producebut
anotherpotential failurepoint. Of courseif itisdecidedthattheunitrequirestobesealed
thenanadditional test stageneedstobeincorporatedtoprovetheintegrity of theseal.
7.3 Housing 297
Studieshaveshown[2] that therequiredlevel of hermeticsealingfor highreliabilityis
510
11
mbar l/s. However, standardleaktestersstruggletomeet theselevelssooften
specicationshaveactuallybeenset lower at about 10
8
mbar l/s.
Hermetic joints must bemetal; epoxies and gaskets do not offer sufcient sealing,
hencesolder or welded joints areused. Thestandard approach to hermetic sealing is
for thelidof thehousingtobelaser weldedor solderedtothewalls. Specialist advice
shouldbesoughtforlasersealingastothedimensionsof thelidandhousingastheseare
critical toproducingagoodseal andrelatedtothelaser power. Of coursethis process
must becarried out after tuning and testing, and if thereis asubsequent failurethen
thelidmust bemachinedoff toallowfor repair/retuning. For RF andDC connections,
hermeticfeed-throughsareusuallymadefromaKovar(anironnickel cobaltalloy) outer
ringandcenter pinwithaglassbody. TheTCE of Kovar isclosetothat of glassandthe
Kovar canbeplatedsothatitcanthenbesolderedinplace. For RF connectionstheratio
of thediameter of thecenter pintotheouter barrel must besuchas topresent a50O
impedance.
It is obvious fromthe reliability of modern electrical components such as mobile
phonesandLNBsthatitisnotnecessarytoprovideahermeticseal toachievereasonable
reliability.Improvementshavebeenmadeinthepassivationlayersonthesurfaceof active
semiconductors and in theconstruction of thedevicepackages themselves. However,
systemspecicationstendtobeconservativeandadegreeof environmental screeningis
oftenrequired. Analternativeapproach, wherefull hermeticsealingisnot required, but
whereprotectionfromharshenvironmentsisnecessary, istouseacompressiongasket
aroundthelid. Sometimesenvironmental andelectrical screeningcannotbeachievedto
thedegreerequiredinasinglegasket sotwoseparateonesused. Thereisalargevariety
of gasketsincludingsolidandhollowtubestoat customforms. Theycanbecomplete
rings, spooledlineormouldedinplace. Thebenetsof thisapproacharethattheunitcan
beopenedat anytimemakingtuningandrepair simpler (especiallyimportant inpower
ampliers!), also no specialist equipment suchas aweldingsystemis required. Some
of thekey considerations areshown in Figure7.6. To ensurethat thecorrect amount
of compressionoccursthedimensionsof thetrougharecritical. Typically, thegasket is
compressed2530%hencethedimensiondinFigure7.6bshouldbesuchthat this is
achieved, similarly wshouldbedeterminedsuchat that at thecorrect compressionthe
gasketisnotrestrictedhorizontally. Wherethegasketisnotrecessed(e.g., atgaskets) it
isadvisabletoincludeprotrusionsinthemachinedfacewhichensurethecorrectamount
of compressionisachieved, but not overdone.
A disadvantageof thegasket sealingapproachis that thewall widthrequiredto t
the gasket and xing screw is greater than without, which can lead to space issues.
Withappropriatecuttingtoolsthematerial under thegasketcanberemovedasshownin
Figure7.6d, howeverthisresultsinmoredifcultassembly, henceitmaynotbeanoption
for volumeapplications. Thelowest cost formof thegasket material is providedona
spool andcanbelaidintherecess, whichcanincludeintricaterouting. Fixingscrews
needtoalwaysbeoutsideof thesealingring, or wherethisisnot possible(for example,
incenter posts) includetheir owngaskets. Whenjoininguptheendsof thegasket it is
better that thetwo ends overlap rather than formabutt joint as shown in Figure7.6e
298 Microwave hybrid amplier realization
Lid
Conductive
Gasket
d
w
Housing
Wall
(a) Before compression.
Fixing hole separation
(c) Housing gasket detail and mounting screws.
(d) Under cutting the gasket
recess.
(e) Joining Gasket ends, butt join (left)
and angled join (right) -preferred.
(b) After compression.
Figure 7.6 Housingsealingusingcompressiongaskets.
andmanufacturers cansupply suitableadhesives to jointheends. Anexampleof this
typeof seal canbeseeninFigure7.7. Althoughnot aMHPA, this is anexampleof a
hybridmicrowaveassemblythat ismountedexternallyandsubject toanorthEuropean
climateandmust behighly reliable. Thehousingis madefromcast aluminumwitha
chromatepassivation. TheRF circuits areelectrically sealed by an internal lid which
includes compartments. Theenvironmental seal is provided by acompression gasket
andalayer of siliconrubber. RF screeningisprovidedbyaninterferencetbetweenthe
internal lidandPCB. Theunit isdesignedtobemountedat ananglesothat if thereis
any condensationwithintheunit it will draintothelower right handcorner away from
theactivecircuits.
7.3 Housing 299
Figure 7.7 Gasket sealedsatellitedownconverter (LNB), manufacturedbyGrundigLtd.
[1] Pinnedthroughboardconnection, [2] waveguidetomicrostripinterface, [3] Diecast main
housingwithgasket recess, [4] FoamRAM for modesuppression, [5] internal lidwithcavity
wallsandconnectingmouse-holes.
OneconsiderationspecictoMHPAsisthat theunwantedby-product whichweseek
to minimize heat actually works to our advantage. Although the danger of short
circuits exists at switch on, especially after storage in cold conditions, the operating
temperatureof most MHPAs will keep theair in theunit abovethedewpoint and so
condensationislesslikely.
7.3.3 Construction
Theconstructionof MHPAs canvary fromthesimpleto thevery intricate. Thebasic
model isacavityintowhichtheRF andDCcircuitsareall tted, asshowninFigure7.8.
ForlargerampliersitisoftenpreferredtoseparatetheDCcircuitsintoanotherchamber.
Thismay besothat biasadjustment canbecarriedout withthelidinplacefor theRF
unit, for electrical isolation, for testingandmonitoring, or becausethesealingmethod
of theDC components is different to that of theRF. It can also reducethelengths of
feedconnectionswhichareareasof danger for interferenceandoscillation. Thevariety
300 Microwave hybrid amplier realization
Figure 7.8 Simpleconstruction, widebandhybridamplier module, beforeandafter lidsealing
andpainting. Courtesyof LabtechMicrowaveLtd. www.labtech.ltd.uk.
of congurations for amplier modules isenormous; Figure7.9describes someof the
morecommonapproaches.
a. Hsection: thisissimpletoconstructandhasthebenetthattheRF andDCcavities
canbesealedindependently. Thefeedlocationsof DC andmonitoringpointstothe
RF PCB can be positioned where needed, with complicated routing being kept in
theDC compartment. Thedrawback withregardto MHPAs is that theheat sinking
for power devicesispoor. Floor thicknesscanbeincreasedunder thepower devices;
however thisstill givesahigher thermal resistanceduetothethermal pathlengththan
other options.
b. Orthogonal Cavities: Similar benets totheH section, but withlonger wirelinks
totheRF PCB. Thedepthof thecavity of theRF sectionisrelatedtotheminimum
height of theDC PCB, whichcouldcauseissues withbox modes. It does offer the
ability to havethepower devices mounted on afacewhich can beattached onto a
heat sinkor coldwall.
c. Wraparound: agoodsolutionwheretherearepower componentsintheDCsection
that also requireheat sinking. Thelong RF section would typically besplit into a
number of chambersusinginternal dividingwallsasshowninFigure7.10.
7.3 Housing 301
RF PCB
DC PCB
Feed through
Coaxial
Connector
(a) (b)
(c)
(d)
Link cable
Lid
(e)
Heatsink/Base
Figure 7.9 Typical amplier housingformats: (a) cross-sectionof H sectionmodule;
(b) cross-sectionof orthogonal cavitymodule; (c) wrap-around module; (d) split section
module; (e) planar module.
d. Split section: this format, althoughrequiringadditional RF connectors andacable,
has thebenet of giving interstageaccess which can beuseful in tuning and fault
nding.
e. Planar: oneof theproblemswiththepreviousmoduleformatsisthat thecircuitsare
within cavities and this can causeproduction issues with assembly and test dueto
302 Microwave hybrid amplier realization
Figure 7.10 (a) X bandand(b) SBandMHPA modules: (a) wrap-aroundconstructionshowing
biascontrol withRF chambersaroundtheperimeter. Notetheweight reductionremoval of
excessmaterial wherepossible; (b) side-by-sideconstruction. Photoscourtesyof SurreySatellite
TechnologyLtd. www.sstl.co.uk.
access. A solutiontothisproblemhasbeentoincorporatethesidewallsintothelid.
The RF and DC circuits may then be incorporated into a single PCB. The circuit
design must have a low susceptibility to ground proximity effects; otherwise the
ttingof thelidwill detunetheperformance. Theconstructionallowspower devices
to be mounted directly to the heatsink thus maximizing thermal transfer. Another
7.3 Housing 303
Figure 7.11 Handlinginternal corners.
issue with this design is that the RF connections need to come vertically through
theboardwhichcangiveriseto difculty linkingto themandlaunchingunwanted
modes(seelater). It ispossibletomount connectionsinlinewiththePCB; however
thejunctionwiththelidrequirescloseattention.
Thehousings themselves may beformedby threemainmethods: direct machining,
casting, and piece parts. Improvements to the performance and speed of computer
numericallycontrolled(CNC) machiningcentershasdramaticallyreducedthecost and
increased the possible intricacy of housings. Designs should be discussed with the
machinistbeforenalizing, asthereareanumber of simpleaspectsthatwill reducecost
andimprovemanufacturability. Theseinclude:
1. Useaslargeacutter aspossiblefor theinternal cavities, wherepossibleavoidtight
corners. If necessaryacorner canbeopenedupusingadrill hole, seeFigure7.11.
2. Holes that requiretappingarebest drilledthroughto stoptaps fromjamming. Use
aslargeaholeandthreadsizeaspossible. Theminimumtapdepthis1.5diameter
of thexing. Blind holes can beastorefor thechemicals and substances used in
processing; it ismucheasier tocompletelycleanthroughholes
3
.
3. Reduce as far as possible the number of cutting axes required, this will speed up
machiningandreducecost.
4. Countersinkingholesisanadditional operation, onlycountersinkwherenecessary.
5. Minimizethenumber of different cuttersanddrillsrequired.
Makingamplier housings fromcast parts is limitedto volumerequirements, andfor
MHPAs theparts will still requiresomemachining to producethenecessary surface
nishesfor mountingpower devices. Complex shapescanbecreatedandthishaslead
tothisbeingapopular optionfor basestationampliers.
3
This caused a problemwith ange mounted transistors failing over time. Removing the blown devices
revealedavery thin layer of sticky goo under thedevices. Assembly technicians wereremindedof the
needforscrupulouscleaningunderthedevicesandaninspectionstagewasintroducedpriortodevicetting.
Still thetransistorsfailed. Eventually, runningacottonbuddownthetransistor mountingholesrevealedthe
sourceof thegoo. Asthedeviceswerebeingclampeddown, dissolvedux wasforceduptheholesand
spreadunder thedeviceange, increasingthethermal resistance. InthiscasetheRF PCBsweresoldered
intotheboxandthencleanedinanultrasonicsurfactant cleaner, leavingresidueintheblindholes.
304 Microwave hybrid amplier realization
Thethirdoptionof usingpiecepartscoversanumberof differentapplicationsranging
fromlow-costcustomassembliestohigh-costlightweightunitsfortheavionicsindustry.
Intherst casesimpleboxextrusionsareusedfor thesidewallswhichcanbeboltedto
heatsinksor lids. Thesecondcaseutilizesaprocesscalledaluminumdipbrazing. This
canproduceassemblies withvery thinwalls (-1mm) thus lightweight, but also with
integratedfeaturessuchasair andwater coolingsectionsthat aresandwichesof corru-
gatedandsheet aluminumthroughwhichwater or air canbeforced, seeFigure7.13e,
andbosses for screwxings. Theparts arecut out andassembledusingtabs andslots
with aluminumsilicate pastes administered along the joints. The whole assembly is
immersedinabathof moltensalt whichsoldersthejoints. Inthisway apressure-tight
seal canbeformed. Suchaprocessisexpensiveandthenumberof manufacturerslimited,
but whereweight isat anabsolutepremiumthiscanbethebest solution.
Wherescrewsareusedtocompressthelid, thenumber andspacingwill dependupon
whether the required screening is primarily electrical or environmental. For environ-
mental screeningthenumber of screwsisdependent uponthestiffnessof thelid( to
thickness) andthegasketmaterial. Whenthereisapressureseal itisnecessaryforthelid
andwallstobethickenough(orstiffenedusingwebbing) sothattheydontbuckleunder
pressure. Theextent of electrical continuityisdependent upontheoperatingfrequency
of theamplier. If thelengthof anyelectrical gapexceeds/8(inair) atanyoperating
frequency the slot so formed can create a radiating element fromwhich signals can
escapeor leak intotheamplier. It shouldalsoberememberedthat it isnot necessarily
themaximumoperatingfrequency of theamplier itself that shouldbeconsideredbut
that of thedevicesinside.
An important sidenoteis screwlocking. In many mechanical applications epoxies
areappliedtothescrewholestoprevent screwscomingloose. Thiscanbeanabsolute
disasterformicrowaveapplicationsasthescrewlockisnonconductiveandcaneffectively
insulate the screw fromthe housing. Appropriate locking methods including epoxy
painting(this canbeauseful additional protectionespecially against chemicals which
couldattack rubber seals) and, whereheight allows, usingpanheadscrewswithspring
washers.
An important consideration in the design of microwave amplier housings is the
resonant frequency(s) of thecavities. A fully enclosedmetal box will havearesonant
frequency dependent on its dimensions [4]. Thepresenceof adielectric material (the
PCB) ononefacewill affect thisfrequencyandthetracksonthePCB will couplewith
thecavity. Thiscannot onlycauseoscillationsbut alsounpredicteddisturbancestothe
amplier gain. Asaroughguide, whenthewidthof thecavityapproaches,2thecavity
will becomeresonant. Thus, cavities should bemadesmall enough that they will not
support anymodeswithintheoperatingband, however thisisnot alwayspractical. The
elds can be broken up by the judicious use of pillars within the cavity to short out
theresonances, however it is important that thesearegroundedat thetopandbottom;
any gap where contact is not made with the lid can turn the pillar into a resonator
itself. This is also true of internal walls. Alternatively, microwave absorber or radio
absorbentmaterial (RAM) canbeusedtoloadthecavity. Thesecomeintwobasictypes,
magnetically loaded and dielectrically loaded. Different sizes aretargeted at different
7.3 Housing 305
frequencyrangessoit isimportant tochoosethecorrect onefor your application. Low-
cost foamversions (seeFigure7.7) areoftenalso electrically conductiveso caremust
betakennot toshort out circuits. Althoughanexcellent solutiontomodingproblems,
oneshouldnot divein anddistributeRAM about thecavity beforeinvestigatingthe
causeof theoscillationor perturbationintheresponse. It couldalsobecausedbypoor
grounding of the circuit board, thus you would be treating the symptomand not the
cause.
7.3.4 Thermal issues and heat sinking
Aboveroomtemperaturethereliabilityof most componentsisinverselyproportional to
temperature, andcomponentswill haveamaximumoperatingtemperatureabovewhich
permanent damagewill bedone. For microwavetransistorsacommongurequotedis
that for every 10

C increasethen themean timeto failure(MTTF) will reduceby a


decade. Power ampliers requiregoodthermal management, not only for maximizing
lifetime but also to obtain the best performance. The maximumtemperature for the
device is specied as thejunction or channel (thecurrent carrying region within the
transistor) temperature. Themanufacturer will alsospecifythethermal resistance(TR)
fromthejunction to thedeviceange. Thedesign task is to ensurethat thejunction
temperatureiskept aslowaspossible, withintheconstraintsof size, weight, andcost.
Under steady-stateconditionstheresistancetoheat owisaproduct of twofactors,
theintrinsic TR of thematerial andtheinterfacewiththenext layer. Initially, thereis
also thermal inertia (TI) (or thermal capacitance) which may be important in pulsed
ampliers, as gainandoutput power areproportional to thechannel temperature. The
thermal components can berepresented by electrical analogues, resistors for TR and
capacitorsfor TI. For example, consider atransistor screwedintoaboxwhichinturnis
boltedtoaheatsink. InthiscasethereistheTR of thedevicechannel toange,
jc
, that
of thelayer betweentheangeandtheboxoor,
fb
, theboxmaterial itself,
bx
, thebox
heatsink junction,
bh
, andnally that of theheatsink (assumedtobeeither instill air
or axedair ow),
h
. All of theseTRsaddtogiveanet
T
, seeFigure7.12a. A noteof
caution, the thermal resistance is not constant, it is proportional to the temperature
difference, thegreater thedifferencethegreater theheat ow. TheTI of theinterface
layers is typically very small and is ignored. The temperature differential between
thedevicechannel andtheheatsinkisthedissipatedpower P
D
, times
T
. TheTI canbe
calculatedby observationof theactual temperatureriseprole[5], whichis described
bytheformula:
T
rise
=
j c
P
d
_
1e
t,
j c
C
j
_
(7.2)
wheret isthetime. A typical responseisshowninFigure7.12b. TheTR of aparticular
junctionis aresult of theintrinsic material thermal conductivity (W/m

C), wherethe
mreferstothethicknessof thematerial inmetres. Whilethiswouldappear tosuggest
that all materialsshouldbeasthinaspossiblethisisnot quitethecase; inorder for the
heatsink to operatemost effectively thetemperaturemust begiventheopportunity to
spread, otherwiseonlyalimitedportionof theheatsink will beeffectiveinremoving
306 Microwave hybrid amplier realization
Channel
Junction to
Flange,
jc
Heat
Source
P
D
160.00
140.00
120.00
100.00
80.00
60.00
40.00
20.00
0.00
0.00 0.20 0.40 0.60 0.80
Time (s)
Device
Device Flange
H
x = H tan 70
70, spread angle
Mounting Holes
T
e
m
p
e
r
a
t
u
r
e

(

C
)
Temperature Rise
1.00
For:

jc
= 2.8 C/W
C
j
= 0.1 10
6
C/s
P
d
= 50 W
1.20 1.40 1.60
(a)
(b)
(c)
Flange
Thermal
Inertia, C
j

T
=
jc
+
fb
+
bx
+
bh
+
h
Box Floor
Thermal
Inertia, C
b
Heatsink
Thermal
Inertia, C
h
Flange to
Box Layer,

fb
Box Floor,

bx
Box to
Heatsink
Layer,
bh
Heatsink to
Ambient
Air,
h
Figure 7.12 Thermal performance: (a) electrical analogueof thermal resistanceandinertia;
(b) examplethermal proleshowingeffect of thermal inertia; (c) thermal spreadingbelow
device.
7.3 Housing 307
theheat. Intermsof heat transfer it isrecommendedthat theheat sourcesbeseparated
suchthat theheat illuminates anareadenedbya70

angleasshowninFigure7.12
(c). Inpractice, other considerationscomeintoplay, suchas:
r
thicknessrequiredfor mountingscrewtapping(1.5diameter minimum);
r
separationof devices, determinedbyangewidthor theRF matchingcircuits. Ideally
Handthedeviceseparationshouldbeset suchthat theilluminated areasjust touch
for eachdevice;
r
boxwidth.
Oneof theadvantagesof usingmultipledevicesover asinglehigh-power deviceisthe
ability to spread thethermal load over alarger area. Theresponsein Figure7.12b is
typical forapowerdeviceange, thishaslimitedthermal capacitybecauseof itsphysical
size. Whenconsideringlargeboxestheeffect canbeamuchgreater timedelay.
Forsurfacemountdevices(resistorsandloadsaswell astransistors) theheatismainly
transferredtothegroundplanethroughviaholes. Theabilityof asingleviatoconduct
heat isgivenbythethermal resistivity,
:
:

:
=
4h
(d
2
o
d
2
i
)
(7.3)
wherehisthesubstratethicknessandd
o
andd
i
aretheouter andinner diametersof the
via. Theconstant, k, isdependent upontheconductor (plating) material andfor copper
is 384 W/m

C. This assumes that theviais unlled, lling theviawill improvethe


thermal conductivity, however the conductivity of solder (Figure 7.2) is signicantly
lessthanthat of copper andhence, it isbest toerr onthesideof cautionandignorethe
llingeffect. A number of vias will berequiredto achievetherequiredTR, it should
beremembered that as theviagets further away fromtheheat sourcethen so its TR
increases.
It can beseen fromFigure7.12athat theeasiest way to reducetheoperating tem-
peratureis to removeaninterface, e.g., directly mountingdevices to theheatsink. For
MHPAsaproblemcomesfrommaintainingagoodRF ground; thegroundplaneof the
matchingcircuitsmustbecontinuouswiththatunder thedevicesasshowninFigure7.5.
Extrudedheatsinkshaveareasonablywidedimensional tolerance, whichmustbecatered
for in approaches such as Figure7.9e. A solution is to manufacturethehousing and
heatsinkasoneunit.
Inorder toreducetheTRof theinterfacelayersathermal compoundisusedtoll the
microscopicair gapsbetweenthematingsurfaces, but applyingit liberallycanactually
increasethethermal resistance; thermal compoundsarenotasgoodthermal conductors
asthemetalsusedintheangeor thehousing. Althoughwidelyusedbelowmicrowave
frequenciesthisapproachisseldomusedabove1GHz, asthebest thermal compounds
areelectricallynonconductive, andattemptstoimprovetheelectrical conductivityhave
degraded the thermal resistance. An alternative is to use soft thin metal shims such
as Indiumunder theange. Thesedeformandcompensatefor any surfaceroughness;
however, theyalsointroduceanother interfacelayer. Thebestsolutionisfor thenishof
themetal surfacestobeasat aspossible. Selectively machiningtherecesswherethe
308 Microwave hybrid amplier realization
devices sit to amirror nishis anacceptablesolutioninlower volumeapplications;
this is smoother than that on the underside of most transistors. Where a heatsink is
boltedtothehousingoor theelectrical conductivityislessof anissueandinthiscase
athermal layer is appropriateand agraphitesheet may beconsidered. This has high
thermal conductivity(20W/m

C) andismanufacturedinsheetsasthinas0.13mm. It
ismucheasier tohandleandlessmessythanthermal grease.
Inadditiontotheatnessof thematingsurfacesitisimportantthatthecorrecttorque
isappliedtothescrewsxingthetransistor. Thisinformationisavailablefromthedevice
manufacturersanddependsonthematerial andthicknessof theange. Over-tightening
cancausetheangetobow, whichwill notonlyincreasethethermal resistancebutcould
damagethebrittlematerialsinsidethepackage(ceramicandactivedevice). Thescrews
should be as large as will t in the ange holes, with both a at and spring washer.
Cap-headsockettypeshavelessriskof thedriver slippinganddamagingsurrounding
circuitry. Whentappingintosoft metalssuchascopper or aluminum, it isimportant to
useaslight countersink. If over tightenedtheedgesof theholecanbecomeraisedand
thus distort themating surface(someanges havethecountersink included which is
equallyacceptable). Thenishandatnessof theundersideof thedeviceangeshould
beinspectedaspartof theassemblyoperation, if notonanindividual basisthencertainly
on asampleof each delivery. Any sign of twisting or scratching of theangeshould
beareasontoreject thedevicebacktothesupplier. Somedevicesupplierssuggest that
theangeatnesscanbeimprovedusingwet/dryabrasivepaper dont accept it! The
manufacturer is responsiblefor providingthedevices inauseablecondition, but such
polishingmaybenecessaryonthehousingmatingsurface. For classA biaseddevicesa
simplecheck istomeasuretheangetemperatureunder knownconditions. Wherethe
budget allows or volumeis sufcient, athermal imagining systemcan quickly spot a
devicethat has not beenmountedcorrectly. Note: issues havebeenseenwithsome
of thehandheldinfra-redthermometerswhenampliersareoperatingwithRF signals,
theexact interferencemechanismisnot currentlyknown.
Detailedthermal calculationsarevery complex. Finiteelement softwareisavailable
thatcanmodel thermal behavior, however thesetendtobeexpensiveandthesimulations
difcult toconstruct. Inaknownenvironment, that is, wherethepower tobedissipated
andthesizeof theamplier housingareknown, it is oftensimplest toconstruct atest
usinganequivalent loadresistor (dissipatingthesameheat asthepower transistor(s) in
theworst casescenario) bolted to theoor of arepresentativehousing. This not only
assistsintheselectionof thebest heatsink andtherequiredair velocity, but alsowhere
sizeandweight arecritical parameters, what customizationof theheatsinkcanbemade
without affectingtheheat transfer totheair. It isimportant toincludeintheexperiment
heat sources for the main power dissipating circuit elements, including those of the
DC bias circuit. Many different types of heatsink areavailable, fromtheconventional
parallel nnedtothepost styleusedwithcomputer processors. Thislatter typeoften
comeswithttingsforfanstobedirectlyattachedandusesaturbulentairow, aswell as
relativelylargesurfaceareatomaximizeheat transfer totheair. Standardheatsinksuse
aparallel arrayof taperedns, whileothersusearoot styleasshowninFigure7.13a.
Thesearemadeusing aluminumextrusions, cut to theappropriatelength. Wherethe
7.3 Housing 309
125
60
1
3
5
1
6
0
0
ABL REF
195AB
196AB
197AB
198AB
199AB
1 2
AIR VELOCITY Meters Second (m/s)
M
U
L
T
I
P
L
I
C
A
T
I
O
N

F
A
C
T
O
R


C
W
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
3 4 5 6
750
600
500
400
300 22
30
38
46
58
0.05
50 100
Length in (mm)
150
195AB
196AB
197AB
198AB
199AB
200
0.10
0.15
0.20
0.25
0.30
No.of fins
50 100
Typical Performance
Length 50 100
0.46 0.35 0.28 0.24
300 200 150
0.64
150 200
Length in (mm)
Black Anodised

C
/
W
C/W
C/W Black

C
/
W
Natural Finish
250 300
0.20
(a)
(b) (c)
0.40
0.60
0.80
1.00
(d)
Figure 7.13 Heatsinktypes: (a) root style, noteimprovement inqfromblackanodizing;
(b) impact of air ow; (c) highpower ridgedfanheatsink; (d) extrudedheatsinkwithcross-cut;
(e) bondednassembly; (f) complexdip-brazedassembly. Photos(a)(c) courtesyof ABL Ltd.,
www.ablcomponents.co.uk, photos(d)(f) courtesyof H.S. MarstonLtd., www.hsmarston.co.uk.
heat is distributedevenly across ahousingoor, sheets of aluminumcanbebrazedor
epoxiedinto analuminumhousingor baseplate, Figure7.13e. Althoughthejunction
between then and theplateis not as thermally efcient as an extrusion, thesurface
areaof theheatsinkisgreater for anequivalent footprint (morens).
Heatsinks can beconstructed out of thehousing itself. An integrated heatsink will
havesuperior performanceasthethermal junctionbetweenthehousingandtheheatsink
310 Microwave hybrid amplier realization
(e)
(f)
Figure 7.13 (cont.)
hasbeeneliminated. Involumeapplicationsthehousingmaybecast; theinternal oor
will requiremachiningtogiveagoodsurfaceatnesswherethepower devicesaretobe
mounted, butwitharougher nishtheheatsinknswill causemoreair turbulenceanda
largersurfacewhichwill improveheattransfer, Figure7.13c. Thecapabilityandspeedof
modernmachiningstationshasmeant that it isoftencost-effectivetoproducehousings
bydirect machining; however, producingtherelativelynarrow, deepgapsrequiredmay
bedifcult usingconventional cutters. Analternativethat hasbeenusedsuccessfullyis
tomakeanarray of diamondwheel cutters onaspindle, withtheappropriategaps set
usingspacers. Thiscanthencut theslotsrequiredinthehousingblock.
The alignment of the heatsink ns relative to the heat producing components is
importantwhenforcedair coolingisemployed. Wherepossiblethearrangementshould
besuchthat thecomponents areperpendicular totheairow, otherwisetheair will be
7.3 Housing 311
preheatedby therst components. Without forcedair theheatsink shouldbemounted
withthens vertical. Theprimary methodof heat loss is convectionbut radiationcan
play asecondary roll. Theability of asurfacetoemit infra-redradiationis referredto
as emissivity (ratio compared to an ideal black body), thus black anodized aluminum
nishis commonfor heatsinks (emissivity of 0.8comparedto 0.05for polished),
seeFigure7.13a. Thetreatment to increaseemissivity must not insulatetheheatsink
fromtheair. As most of thecooling is aresult of convection, it follows that cooling
effectivenessisdependent onair densityandhencealtitude. Analtitudeof 3000mwill
degradeaheat sinksefciencyby20%.
Whenpushedtoreducethejunctiontemperaturethenincreasinglycomplicatedmeth-
odscanbeused. Theeffectivenessof aheatsinkcanbeincreasedbyforcedair cooling.
If this is availablewithinthesystemit is asimplesolution, however if includingafan
withtheamplier, considerationmust begivento failureandmaintenance. Thefanis
probably themost unreliablecomponent (movingparts) andwill havealimitedlife. It
is also moresusceptibleto shock andvibration. Airowimproves theeffectiveness as
showninFigure7.13b.
Aluminumis commonly used in microwave housings for reasons of cost, weight
andmachinability, however its thermal conductivity is almost half that of copper (250
compared with 400 W/m

C). Some ampliers have been constructed with a copper


puck xed into theoor of an aluminumhousing, with thedevices bolted onto the
copper. Water coolingcanbeveryeffective, but manysteer awayfromit for thesimple
reasonthat water andelectricity isnot agoodmixture! Onesolutionistousewater to
cool aplateontowhichtheamplier modulesarebolted. Thiscanbeconstructedusing
dipbrazingtechniquesorhavingcopperpipesepoxiedorsolderedintothecoolingplate.
Another extensionof thisapproachistouseheatpipes. Thesecomponentshavethermal
conductivities upto athousandtimes greater thancopper. Typically, they aremadein
rod formand work by theheat at oneend causing aliquid in thetubeto changeto a
vapour which, dueto thelowpressureinside, quickly moves to theother endwhereit
cools back into aliquid, so transferring energy. Theliquid is absorbed into aporous
liningandisdrawnbycapillaryactionbacktothehot end. Heatpipescanbeembedded
inamplier modules or coldplates withthecoldendclampedto awater-cooledheat
exchanger.
Sometimesthecustomer providesacooled-surfaceontowhichtheamplier module
is clamped. This is also known as acold-wall. Theresponsibility for removing the
heat istheirs, however thedesigner must ensurethat thereissufcient thermal transfer
betweenthetwo surfaces. It is also important that thecustomer is fully awareof how
muchheat will begenerated. Therearebroadbandclass A power ampliers that have
efcienciesof between10and20%whichmaysurprisesome.
7.3.5 RF connections
Microwavesignal connectionsaremost oftencoaxial, althoughdirect waveguideinter-
faces arealso used(Figure7.7). Thetypeof coaxial connector is dependent uponthe
operatingfrequency andpower level. Tomakeconnectorsfunctionat highfrequencies
312 Microwave hybrid amplier realization
Table 7.6 Coaxial connector power handling calculations
Maximumaveragepower (W)
Connector type N SMA 3.5mm 2.4mm
Max. oper. freq. 18GHz 26GHz 40GHz 65GHz
Frequency(GHz) 1 10 1 10 1 10 1 10
Max. power (W) 1900 570 590 180 280 85 130 36
Note: thisassumesaperfect match, 23

C, indryair at sealevel.
Temperature(

C) Altitude(m)
0 40 80 0 1500 15,000
De-rating 1.2 1.0 0.8 1.0 0.95 0.5
thesizeisdecreased; thisreducesboththecurrent handlingandthevoltagebreakdown,
thusreducingthepowerhandlingcapabilities. Besidesmatch, theoperatingenvironment
must also be considered as both temperature and altitude reduce the power handling
ability of coaxial connectors; the quoted gures fromAstrolab Ltd.
4
are included in
Table 7.6. The performance of the connector depends upon the materials used (their
purity), methodof constrainingthecenter conductor, andtheprecisionof themanufac-
turing. Theuseof epoxy center contact captivation (identied by aholein themetal
walls lledwiththeepoxy typically blueor black) shouldbeavoidedinhigh-power
applications; theepoxy has higher loss than thesurrounding PTFE and consequently
experiences localizedheatingwhichcanresult intheepoxy blowingout of thecon-
nector.
Examplede-rating calculation: SMA connector at 10 GHz operation at 80

C and
1500m, then1800.80.95= 136.8W.
The most common method of connecting to the RF PCB is to bolt the connector
through the sidewall of the chassis, Figure 7.14a. The connector center pin can be
supported by an extended dielectric (usually PTFE). In hermetically sealed and eld-
replaceable units the center pin will be a separate part mounted in a glass bead and
solderedor reddirectly into thehousing. Wherethepinis unsupportedthediameter
of theholethroughwhichit passesmust maintain50O. InthePTFE sleevedcase, the
manufacturer sets thePTFE diameter toproducea50O transmissionline. Inthecase
of anair-lledholetherelationshipbetweenthepindiameter andthatof theholecanbe
approximatedbyasimpleformula:
Z
0
= 138log
D
d
(7.4)
which, for 50O impedance, simpliesto
D = 2.303d (7.5)
whereDistheouter diameter anddisthepindiameter.
4
www.minibend.com.
7.3 Housing 313
Figure 7.14 Coaxial launches: (a) panel mount; (b) impedancediscontinuityat transition;
(c) through-boardmodesuppression; (d) stressrelief tapebond; (e) stressrelief slidingcontact.
Photo(e) courtesyof AnritsuLtd.
314 Microwave hybrid amplier realization
Table 7.7 Coaxial transition parameters for
50 O and cut-off frequency
Centrepin Feedhole Cut-off frequency
(mm) (mm) (GHz)
0.5 1.15 115.6
1.0 2.30 57.8
1.5 3.45 38.5
2.0 4.61 28.9
2.5 5.76 23.1
3.0 6.91 19.3
3.5 8.06 16.5
4.0 9.21 14.4
4.5 10.36 12.8
5.0 11.52 11.6
Forpowerapplicationsitisimportanttomaximizethecurrentcarryingcapacityof the
connector, asthediameter increasessothecoaxial transmissionlinecansupport other
modeswhichcanbeexcitedatadiscontinuity, e.g., thetransitionfromcoaxtomicrostrip
andviceversa. Thus, itisnecessarytokeepthedimensionsbelowthefrequencyatwhich
theseother modes canstart to besupported. Table7.7shows therelationshipbetween
pinandholediameter andhigher order modecut-off frequency.
It has been noted that where a discontinuity exists this can launch other modes.
Figure7.14bhighlightstheriskareaaroundtheconnector launch. Thisgureshowsthe
region wherethecoaxial elds of theconnector arechanging to thetransverseelds
of thesubstrate. It can also beseen that thelength of thesignal path and that of the
ground path are different, which causes the even and odd mode phase velocities to
become out of step. Note that this is also an issue where different parts of a circuit
aremounted on carrier tiles. At connector launches this ground path problemcan be
resolved by extending the housing wall to overhang the PCB and by placing ground
padswithviasincludedonthetopsurfaceof thePCB. Anapproachusingthismethod
hasbeendescribed[6] whichshowsexcellentperformanceto>40GHz. A modication
of this is to createataper transition fromcoplanar to microstrip. A drawback of this
approachisthat thePCB needstobemountedunder theconnector.
It is sometimes required to mount the connector through the substrate material as
shown in Figure 7.14c. PCB mount connectors are becoming increasingly popular;
however caremustagainbetakentoavoidmoding. Manyconventional PCB connectors
haveintegral posts at thefour corners of theangewhichareintendedto besoldered
throughthePCB. Aswell asprovidingamechanical connectionthesecanalsohelpto
reducemodingat theright-angleconnection. Astheoperatingfrequencyincreasesfour
posts arenot sufcient and additional vias arerequired. Thesegrounds can beposts,
platedvias, or mountingscrews.
Direct solder attachment of thecenter pintothePCB track isthestandardapproach
withsoftsubstratesasthegiveinthesubstratematerial providesthermal andmechan-
ical stressrelief. Mechanical pressureisputonthepinduringthematingcycle. For hard
7.4 Components 315
substratesthereislittlecompliancesoadirect solder attachment isnot recommended.
Thereareanumber of variationsonthewirebondlink showninFigure7.14d, includ-
inghavingthecenter pinattenedat theendandthewirebondat right angles to the
pin (which is slightly offset fromthe track). A very successful solution is shown in
Figure7.14ewhereaslidingcontact ts over thecenter pin. This contact canitself be
soldereddirectlytothePCB track.
7.4 Components
Thecomponentsusedinhybridmicrowaveamplierscanbedividedintoactiveandpas-
sive. MHPAstakeadvantageof thecompactspacethatalumpedelementcomponentcan
provideor thetailoredfrequencyperformanceof adistributedcomponent. Advancesin
processinghaveproducedlow-temperatureco-redceramic(LTCC) discretepackaged
devices, whichmay containamixtureof discreteanddistributedcomponents inwhat
appearstobeasingleelement, thusthereisevenablurringof thecategorizationbetween
lumpedanddistributed. A conict existsbetweenacceptedpracticeinthegeneral elec-
tronicsengineeringindustryandwhattricksyoucanplaytoimproveRF performance.
Forexample, mountingasurfacemountchipresistoronitssidecanshiftapackagereso-
nanceupinfrequency outof thewantedband butwill giveproductionengineerswith
automated assembly lines high blood pressure! Chip components haverecommended
padsizesfor solder connections. Thesepadswill haveanassociatedcapacitancewhich
thedesigner maywant tominimize. Hence, acompromiseneedstobereachedbetween
theRF performanceandtheproductionrequirements. Intheend, thecomponent must
meet itsRF performancerequirementswhilestill beingmanufacturable.
7.4.1 Passive lumped components
Ampliersrequirecapacitors, resistorsandinductorsfor their operation. Inmicrowave
designsuchelementsdonotbehaveasapureelement. Byknowingthepropertiesof the
element, thesecanbeincorporatedintothecircuitdesignandusedforouradvantage. For
example, afeedback network may compriseof aresistor, acapacitor andaninductor.
The capacitor (C), and resistor (R), will have parasitic inductance. By choosing the
appropriatephysical sizeandshapefor thecapacitor andresistor theycaneither reduce
theinductanceneededor encompassit entirelywithinthetwoelements.
ModelingtheRs, Ls, andCscanbedoneeither usingaphysicallybasedmodel or by
usingexperimentallymeasuredS-parameters. Manycomponent suppliersnowproduce
libraries of dataon their components which can bereadily used in simulators. Users
shouldapproachthesewithcaution, particularlyasfrequencyincreases. Ideally, thesur-
roundingsshouldhavebeenentirelyde-embeddedsothat themodel existsindependent
of itsenvironment. Itisimportanttoverifythemodelsusedintheenvironmentinwhich
the components are intended. It is also useful to understand the construction of the
componentssothat theremaybesomeexpectationof thebehavior withfrequency.
316 Microwave hybrid amplier realization
Resistors
Thesehaveawidevarietyof usesinampliercircuits, e.g., low-frequencybiasing, loads,
attenuators and feedback elements, and signal balancing in combiners. In microwave
applications resistors are mainly made frommetal lms, the metal being chosen for
its resistanceproperties rather thanits conductivity as inPCB tracks. Themajority of
resistorsusedarebaseduponasurfacemountchip, however therearetimes(particularly
below3GHz) whentheinductanceassociatedwithaleadedresistor canbeof benet.
Leaded resistors for such applications should bemetal lmas carbon composition or
wirewound resistors generally contain many parasitic elements and thus havewidely
varyingimpedancecharacteristics. Theinductanceof theleadlengths associatedwith
theresistor iscalculatedfrom:
Inductancein nH = 0.2l
_
2.3log
4l
d
0.75
_
. (7.6)
wherel istheleadlengthanddisthediameter (inmm).
Chipresistorsbasicallyconsistof aceramictileontowhichhasbeendepositedametal
lm(thin lm) or paste(thick lm). Thepasteis red at high temperatures and then
theresistancecanbetrimmedtothenecessaryvalue. Thecontinuedminiaturizationof
electronicshasbeenbenecial tothemicrowaveindustryasthesmaller thecomponent
thenthelower theparasiticsandhencethecloser theperformanceistothatof thepure
element. With sizereduction comes lower power handling. Theexact power handling
ability isdependent onalargenumber of factorsandhasbeenexcellently documented
in other work [7]. As discussed in Section 7.2, at high frequencies thecurrent is not
evenly distributedthroughtheconductingmaterial, thus thequality of theconducting
surfaceisof paramountimportance, andthistendsnottobewell controlledinthicklm
productsduetothenatureof thepaste. Thekeyaspect of thicklmisthat it islowcost.
Thin-lmresistorsaremoreexpensivetomanufacture. A thinresistivelayer, typically
of nickel/chromium(nichrome) or tantalumnitrideisdepositedontheceramicandthen
conductiveterminations aredeposited at either end. Sheet resistances vary between 5
and 250O/sq which arevery suitablefor thetypical resistances required in amplier
circuits. Itisimportanttocheckhowtheresistancevaluesaretrimmed, seeFigure7.15a,
as sometimes a path is cut in the conductor which can have intrinsic capacitive and
inductiveeffects. Theendterminationsonstandardresistorstendtobewrap-around
asthetopsideisalsooftenprintedwiththeresistivematerial. Better RF behavior maybe
achievedby mountingtheresistors upsidedownas this reduces theeffectiveelectrical
length, but it will alsoincreasethecapacitancetoground. High-performancethin-lm
resistors areoftenofferedwiththeoptiontohavetheendterminations onthetopside
only, andaremountedipchip. For high-power applicationathirdmetallizationcan
beaddedtotheundersideof theresistor sothatitmaybesoldereddirectlytotheground
planefor optimumthermal transfer. Specic values for resistors, such as 50 O, have
beenproducedwithoneof theendterminationswrappedaroundtocover thecomplete
undersideof theresistor. Thisisfor useinterminationsor loads.
The majority of surface mount resistors are made using alumina substrates. For
high-power applications berylliumoxide(BeO), thermal conductivity 290W/m

C, or
7.4 Components 317
Helical Trim
Pulsed Trim
PORT
P=1
Z=50 Ohm
PORT
P=2
Z=50 Ohm
IND
ID=L1
L=LPS nH
CAP
ID=Cf
C=0.021 pF
RES
ID=R1
R=100 Ohm
CAP
ID=Ci
C=0.02 pF
CAP
ID=Ci
C=0.02 pF
Meander Trim
(a)
6
8
10
12
14
0 5
DB(|S(2,1)|)
100R_Modelled_RTD
DB(|S(2,1)|)
100R_Modelled_AI
DB(|S(2,1)|)
100R_Sparams
DB(|S(2,1)|)
Lumped Model AI
DB(|S(2,1)|)
Lumped Model RTD
10
Frequency (GHz)
Thru Loss
I
n
s
e
r
t
i
o
n

L
o
s
s

(
d
B
)
15 20
(b)
(c)
Figure 7.15 Surface-mount resistors: (a) resistor trimmingapproaches; (b) simpleequivalent
circuit for 0603resistor on0.8mmthick
r
= 2.2withrecommendedmountingpads; (c)
performanceof 0603100O resistor, manufacturersrawdata(100R_Sparams), resistor model
andEM on0.635mmaluminaandRT duroid5880substratesandlumpedequivalent circuit.
aluminumnitride (AlN), thermal conductivity 170W/m

C, are used. Despite having


lower thermal conductivity, AlN isincreasinglypopular asberylliaishighlytoxicinits
powderform. Itisbannedinanumberof applicationsandcountries, productscontaining
it must beappropriately labelled, and it is difcult to disposeof. Increasing volumes
and the number of suppliers have seen the price of AlN products fall to acceptable
318 Microwave hybrid amplier realization
Table 7.8 Conventional thick-lm resistors mounted on
0.8 mm FR4 substrate
Power rating
Reference Size(mils) Size(mm) (mW)
0201 2010 0.510.25 50
0402 4020 1.020.51 63
0603 6030 1.520.76 100
0805 8050 2.031.27 125
1206 12060 3.051.52 250
2010 200100 5.082.54 500
2512 250120 6.353.05 1000
levels. Wheresingleterminationscannot handlethepower level required, resistorscan
beusedinparallel. Thishastheaddedadvantageof reducingtheseriesinductance, but
capacitancecanincrease.
Resistorsinmicrowavecircuitsareoftenapproximatedbyasimpleequivalentcircuit,
asshowninFigure7.15b. However, it shouldbenotedthat thevalueof C
i
isdependent
uponthepadsizeandthesubstrateused, henceageneric model for theresistor should
not beused. Thevalueof C
f
depends upontheresistor size, paddimensions, andend
terminations. Anexampleof howdifferent modelsandsubstrateswill affect themodel
performance is shown in Figure 7.15c. A 100 O resistor has measured S-parameters
from45MHz to2GHz; theseareextrapolatedto20GHz asareference. Theresistor is
modeledusingbothalumpedelementequivalentcircuitasinFigure7.15bandausinga
modiedequivalent circuit replacingC
i
withanEM simulationfor themountingpads.
Theparasiticinductanceandtheresistanceareconstant for all simulations. Thisshows
that theperformanceof theresistor itself cannot betakeninisolation. If S-parameters
areusedtheyneedtocover thefull frequencyrangeandbeproperlyde-embeddedfrom
themeasurement test xture. Themountingpadsneedtobeincludedinthesimulation,
either directly or as tracks feedingthecomponent. Table7.8lists thepower ratingand
sizeof commonsurface-mount resistors.
Capacitors
These are essential in the operation of microwave ampliers, they have functions at
bothDC andRF andtherequirements areoftenat odds. They areusedfor interstage
biasde-coupling, matching, by-passingandlocalizedchargestorage. Aswithresistors,
not only must their primary characteristic capacitance beconsidered but also the
parasiticelements. Initsmost basicformacapacitor consistsof twoparallel platessep-
aratedbyaninsulatingmaterial. Suchcapacitors, calledsingle-layer capacitors(SLCs),
arecommonly usedinmicrowaveapplications. Thecapacitances that canbeachieved
are determined by the relative dielectric constant of the insulating material,
r
. The
capacitance, C inpF isgivenby
C =
0.00885
r
A
d
. (7.7)
7.4 Components 319
0
1
2
3
4
5
6
0 1 2 3 4 5
Frequency (GHz)
Horizontal Orientation
C17AH101K-7UN-X0T 100.0 pF Temp = 25
Vertical Orientation
C17AH101K-7UN-X0T 100.0 pF Temp = 25 C
S
2
1

(
d
B
)
0
1
2
3
4
5
6
S
2
1

(
d
B
)
6 7 8 9 10
0 1 2 3 4 5
Frequency (GHz)
End termination
Capacitive Plates,
end termination
removed.
6 7 8 9 10
Figure 7.16 Effect of plateorientation. Courtesyof DielectricLaboratories, Inc.,
www.dilabs.com.
whereAistheareaof theplatesinmm
2
anddistheseparationinmm. Workingagainst
theapplicationof thecapacitor inMHPAsisthat thecloser theplatesarethenthelower
thebreakdown voltage, and thehigher the
r
theworsethetemperaturestability and
oftentheloss. Inorder to achievehighcapacitancevalues layers of capacitors canbe
madewithalternateplates joinedtogether as showninFigure7.16, hencetheir name
multilayer capacitors (MLCs). Inthiswayhigher capacitancevaluescanbeachieved
for the same foot print. The parasitic elements of a capacitor are largely due to the
dielectricmaterialsandthephysical size. Theinductanceisproportional tothelengthof
theplates. Thelossinthedielectric, (theenergy that isdissipatedasheat) isexpressed
asaresistanceinparallel, R
S
. Modelingcapacitorsinthemicrowaveregioncanbecome
verytricky, not least becausetheperformancechangesdramaticallywithorientation, as
showninFigure7.16. Withtheplatesparallel tothegroundplane(horizontal) aseries
of resonancesexist, rotatingthecapacitor 90

sothat theplatesarenowperpendicular
to theground plane(vertical) removes half of theresonances. Thereareanumber of
320 Microwave hybrid amplier realization
Table 7.9 Characteristics of various dielectric materials and approximate changes with
frequency
Material
r
Tan (DF) Q
100MHz
Q
1GHz
Q
10GHz
BariumTitanate 12008000 0.030.1 3310 310 -1
Ceramic 30 0.002 500 50 5
Alumina 10 0.0005 2000 200 20
Porcelain 15 0.00007 14000 1400 140
theoriesregardingthisbehavior, oneisthat it isduetothephysicallylonger path-length
between the signals of the top and bottomplates in horizontal mounting, another is
that thecapacitor actslikeafoldedtransmissionlineconnectedtothefeedtrack at one
endandopencircuit at theother, thusbecomingaresonator [8]. It isarguably simpler
and morepractical to measuretheS-parameters of acapacitor than develop amodel,
particularly if onewouldneedto makeameasurement to verify themodel! As inthe
discussion on resistors, theeffect of thesubstratecannot beignored. For MHPAs the
most important characteristic is the insertion loss. Assuming that a suitable value of
capacitanceischosentokeeptheimpedancelow(1/C) thenthelossisaresult of the
dielectricloss. Thishasseveral measures, qualityfactor (Q), dissipationfactor (DF) or
tan.
Q =
1
DF
=
1
tan
=
X
C
R
S
(7.8)
An ideal capacitor would have the RF current lead the voltage by 90

. In the real
worldthisisnot thecaseandthephasedifferenceis. Thisismorecommonly quoted
as tan (wherethemeasureof substrateloss comes from). Tan is dependent upon
frequency and temperature, but unfortunately fewmanufacturers provideinformation
onthisrelationship. For example, havingmeasuredtheoutputpower of adeviceandthe
insertionlossof theoutputmatchingcircuitincludingadecouplingcapacitor, itisfound
that theoutput power islower thanexpected. Themeasurement of thematchingcircuit
wasundersmall signal conditions, underpowerconditionsthelosscausesheating, which
increasesthetan, increasingtheloss. It must berememberedthat aninsertionlossof
0.1dB equates to2%of thepower beinglost inthedevice. In100W that is 2W ina
verysmall space. Table7.9givesasummaryof thecharacteristicsof someof themost
commonmaterialsusedincapacitor dielectrics.
The manufacturers measurements are generally made at low frequencies and the
valuesfor Qinthetableareapproximatebaseduponobservedperformance. Although
exact datais rarely providedindatasheets at thefrequencies, temperatures andpower
levels that adesigner might want, what is clear is that thematerials usedshouldhave
thelowest possibleloss. Theinsulationresistance, R
P
, (theDCcurrent paththroughthe
dielectric material), is typically of theorder of 100,000MO andso is usually ignored
for RF purposes.
For frequenciesupto8GHz whereMLCsareused, thecapacitancevaluesdecrease
withfrequencytoavoidself-resonancewiththeinternal inductance. SLCsarepreferred
7.4 Components 321
above8GHzalthoughtheir planar constructionmaynecessitatewirebondingor careful
handsoldering. Dielectric Laboratories offer aGapCap, wherethebottomplateof
the SLC is split in half effectively making two series capacitors. This may be more
suitable for some production environments than a SLC, but note that it does have a
slightlylower self-resonantfrequency(SRF). For narrowbandapplicationsitispossible
touseMLCsabovetheSRF forDCblocking, however itshouldbenotedthataboveSRF
theimpedanceof thecapacitor increaseslikeaninductor. Indeed, insomeapplications
chipcapacitorshavebeenusedabovetheir SRF asinductiveelementsastheyaremore
repeatable, lowercost, andeasiertohandlethanveryhigh-frequencyinductors. Itisrisky
tooperatenear theSRF astheimpedancevariesrapidlyinthisvicinityandissensitive
to temperature. It shouldalso benotedthat althoughcapacitors fromasinglesupplier
are highly repeatable, performance will vary for similar types of capacitor between
suppliers, particularlytheSRF.
Theelectronic industries association (EIA) divides capacitors into threeclassica-
tions, whicharebasically determinedby thestability of thecapacitance. For matching
andcouplingEIA class1areused, forchargestorageandlow-frequencybiasdecoupling
class3isused. Agingisgenerallynotasignicantissueinclass1capacitors; however in
class3thevariationcanbesignicant. Thisiscausedbytemperatureandhigh-voltage
eldstrengthsaffectingthecrystallinestructureof thedielectric.
As mentioned, for the parts of the circuit where the RF signal ows it is required
that theinsertionlossbeaslowaspossible, but therearepartsof thecircuit wherethis
characteristic is not thecase. Wherebroadbanddecouplingis required, for exampleat
theends of bias lines, theintentionis to ensurethat unwantedsignals areterminated.
Damping resistors in thebias lines causeproblems with changing voltagedrops with
frequency, but alossy decouplingcapacitor will not affect theDC performance. Here
amixtureof capacitor typesareused, ahigh-quality MLC andoneor moreincreasing
capacitanceceramics.
Inductors
The properties of inductors include the fact that their impedance increases with fre-
quency. To increase the inductance the wire can be wound in a coil. This increases
themagnetic ux linkage, whichopposes theowof current throughthewire, hence
increasingtheinductance. Thecoil alsohascapacitancebetweentheturns, whichcauses
theinductor to haveaself-resonant frequency. Conversely to thecasewiththecapac-
itor, aboveSRF theinductor behaves as acapacitor, i.e., its impedancedecreases with
frequency. For thelower microwaveregionswoundinductorscanstill playapart, how-
ever they areusually air coredandtheir impedancecanvary dramatically. Their main
applicationis inbias circuits whereafewturns of spacedcoils canproduceauseable
inductance; at higher frequencies a single loop of wire can be used. A danger with
thesetypesof component isthat theeldsarenot well containedandthusthey canbe
sourcesof transmissionandreceptionof radiation. Wherethey areusedthey areoften
positionedsuchthat they areat right angles, andphysically separatedas far as possi-
ble. Somecompanieshavesucceededinproducinglow-inductancesurfacemount chip
322 Microwave hybrid amplier realization
inductors, howevertheircurrenthandlingandrangeisrelativelylimited. Theinductance,
L, of anair coredcoil isgivenby[9]:
L =
0.394r
2
N
2
9r 10l
. (7.9)
wherer is thecoil radius in cm, N is thenumber of turns, and l is thelength of the
coil. Also, for optimumQ, l = 2r. Very widebandinductances havebeenachievedby
windingnegaugewireeither onaremovableconical former (hence, air cored) or ona
conical ironformer toincreasetheinductance.
An advantage of using air cored inductors or even singleloops of wireis that the
effective impedance can be tuned, either by altering the height of the wire above a
groundplane, or by alteringthespacing(andtheparasitic capacitance) of theturnson
thecoil. For manymicrowaveapplicationsit ispreferredtousedistributedcomponents
asdiscussedlater.
Integratedcomponents
ThedevelopmentsinLTCCtechnologyhavegivendesignerstheoptionof incorporating
anumber of elements inasinglepackageddevice. Thesearestill volumeapplications
unless the size/weight benets can justify the setup costs. The approach consists of
incorporating conductive, resistive and dielectric pastes onto green ceramic tapes.
Different layers arelaminated together and thewholeassembly is then red. Internal
viascanbecreated, andeachlayer isinspectedprior tolaminating. Inthiswayavariety
of components fromsimplestructures suchas lumpedelement lters or bias tees to
completefunctional blockscanbecreated.
Isolatorsandcirculators
Another benet of the mobile communications revolution was the packaging devel-
opment of thesemagnetic/ferritecomponents. Inthe1980s themanufacturingcontrol
of theferriteand magnetic material was not as rened as currently and much of the
assemblywasbyhandandintegral withtesting. Thehighvolumesrequiredfor thebase
stationindustry put emphasis onpackagingfor automatic placement oncircuits. This
hasresultedinpartswhicharenowreadily availableincommonfrequency bands, and
whicharesmall andcost effective. Isolatorsandcirculatorshavethreemainfunctions,
(a) to provideisolationbetweenstages, (b) to protect circuits fromreversepower, and
(c) togivetheamplier acceptableoutput impedance(Figure7.10). Thecirculator isa
threeportdeviceconsistingof asnowakecopperfoil sandwichedbetweentwoferrite
layerswithamagnetadjustedforthecorrectbiasingof theferrite. Themagnetensures
that theelds areorientated correctly. Theoperation of thecirculator is described in
Figure7.17. Thebehavior isduetothenonreciprocal natureof theferritematerial. The
signal enteringport 1splitsandtravelsaroundtotheother portsinoppositedirections.
Asthephasevelocitydependsuponthedirectionof travel, thenadditionor cancellation
can bearranged at theappropriateport by correctly applying themagnetic eld. The
center conductor ispatternedtoensuresymmetryof thestructureandimpedancematch-
ing (theimpedancein theferritesection is less than 50 O), thewider thebandwidth
themoreintricatethis structure. An isolator is acirculator with port 3 terminated in
7.4 Components 323
Signal Flow: Port 1-2 Port 3 Isolated
Port 2-3 Port 1 Isolated
Port 3-1 Port 2 Isolated
Physically the circulator is symmetrical internally.
Port 1 Port 2
Port 3
Figure 7.17 Functional diagramof acirculator.
aload. Circulators canbemadeto handlepower levels upto kWs; power handlingis
proportional tosize, thehigher thepower thestronger themagneticeldandthelarger
thecopper center conductor. Inisolatorstheloadoftendeterminesthemaximumpower
rating. Circulators/isolatorshaveabandpasscharacteristic, but careshouldbetakenin
high-power applications, hysteresis inthemagnetic elds results incirculators having
nonlinear characteristics andhencethey cangenerateharmonics at highpower levels,
whichcancauseintermodulationproblems.
Insystems wherethereis substantial electrical gaintherealso tends to bearisk of
oscillation if the reverse isolation between stages is inadequate. Attenuators between
thestagescanhelp, however throwingawaypower isanathematoPA designengineers.
Isolators, whicharelowlossintheforwarddirection(typically-0.5dB) andhighloss
in thereverse(20 dB), can beabetter solution. In class C pulsed applications, the
input impedanceof deviceschangeswithappliedpower. Therefore, theloadimpedance
seenby thedrivingdevicechanges duringtherisingandfallingedges of thepulse. It
can bedifcult to ensureacompletely stablenetwork especially when frequency and
temperaturearealsothrownintothemix. Theproblemisexacerbatedasonemovesdown
theamplier chain as each deviceis not only producing achanging input impedance
duetoitsowndrivelevel, buttheloaditseesisalsochanging. Anisolator inthemiddle
of theline-upcanreducethisproblemdramatically. Wheretheisolator isbeingusedto
protect theamplier thensmart loadsarebecomingmorepopular. Theseincorporate
atemperaturesensor ontotheloadwhichcanbeusedtotrigger ashut-downsequenceif
thereectedsignal istoohigh. Caremustbetakenwhenhandlingandstoringisolatorsas
theyaremagnetizedcomponentsandtheir behavior dependsonthestrengthof thiseld.
7.4.2 Passive distributed components
Distributed elements are those structures whose physical dimensions fundamentally
determinetheir electrical characteristics. Hence, thetolerancesandrepeatability of the
manufacturingprocesses haveadirect effect ontheperformance. Onaluminacircuits
capacitorsandresistorscanbeincorporateddirectly duringthecircuit fabricationpro-
cess. For resistors, aswiththeir lumpedelementequivalents, resistivepastescanbeused
inthick lmcircuitsor aNiCr layer inthinlm. Theseresistivematerialsarespecied
intermsof O/squareandtheresistance, R, isproportional totheratioof thelength, l, to
thewidth, n.
R = resistivity
l
n
(7.10)
324 Microwave hybrid amplier realization
The width can be adjusted to match that of connecting transmission lines. Common
resistivitiesare50and100O/sq.
Capacitorscanbeincorporatedbyputtingdownadielectricontopof conductorsand
thenoverlappingwithaconductivepaste/plating. Althoughthesehavebeenproposedfor
useinsoftsubstrates, particularlyusingconductiveinks, therehasnotbeenalargescale
takeupasyet.Inductorsonmicrostripcanbedifculttodistinguishfromhighimpedance
lines. Theexception is wherethelineis wound in aspiral (round or squaresided) as
is popular in MMIC applications. This is used less frequently in hybrid applications
duetotherelativelylargesizeandlower Qcomparedtowoundcomponents. Similarly,
interdigital capacitors are used less often in hybrid circuits. An exception is when
edgecoupledlines usedinsomebandpass lter applications full two functions, that
of ltering and DC blocking. For narrow band applications where the volumes are
extremelyhigh, edgecoupledlinesmaybejustiedontheir own.
Themost commondistributedcomponentsaretransmissionlines. Asmentionedear-
lier, the impedance ranges that can be created are limited due to moding and etch
tolerances. In practice, circuit structures are also limited by the ability to simulate
themwithin design tools. Distributed circuits have re-entrant properties, that is, the
impedances repeat (approximately) at multiples of 90 and 180

. Thebehavior of dis-
tributed components with frequency differs to that of pureelements. For example, at
a specic frequency a shunt inductor of impedance X
L
can be replaced by a shorted
transmissionlineof impedance, Z
0
tan whereZ
0
isthecharacteristicimpedanceand
istheelectrical length. However, whereasX
L
increaseslinearlywithfrequency, theline
impedanceincreases with tan , which is periodic. Thereareavariety of equivalents
betweenlumpedanddistributedcomponents, someof whicharegiveninFigure7.18.
Therealizationof distributedseriescapacitors, asmentionedabove, isvery difcult in
hybridapplications. Anexceptioniswhereverysmall valuesarerequiredwhichcanbe
achievedbyusingnarrowgaps.
A short circuited transmission lineor short-circuit stub, is basically atransmission
line with one end terminated in zero impedance, but this becomes more difcult to
achieve as frequency increases due to parasitics. Common methods of producing a
short circuit includeusing aviahole, edgewrapping and asolid ground plane(with
or without viaholes). Oneof thebenets of adistributedcircuit is that ashort circuit
at theend of a ,4 linelooks likean open circuit at theother end of thelineat the
designfrequency. Thus, another way of creatingashort circuit istoattacha,4open-
circuitstubatthepointwhereashortcircuitisdesired. Thisisinherentlyanarrowband
structure; to broaden thebandwidth aradial stub can beused. Even moreeffectiveis
adoubleradial or buttery stub [10]. Thesesolutions areoften adopted as frequency
increases and the inductance/phase length of via holes has more impact, or when an
RF short is required but not one at DC. In bias feeds it is required that DC current
be injected into the circuit but that the feed arrangement not load the RF matching
network. Although microstrip impedances aretypically limited to between 25 and 90
O, it is possible to create effectively lower impedances by adding two o/c stubs in
parallel. A comparison of the performance of various distributed stubs is shown in
Figure7.19.
7.4 Components 325
Inductor
X
L
= L
Z
0
= L/4
Z
0
= 4L/
Z
0
= 2L/
X
L
= 1/(C)
Z = Z
0
sin
Z = Z
0
sin
Z = Z
0
cot
Z = Z
0
tan
Capacitor
Transmission
Line (Series)
Transmission
Line (Series)
Where <45
For <<90
In the special cases where = 90(quarter wavelength) at the resonant frequency:
And when = 180(half wavelength) at the resonant frequency:
Open Circuit
Stub
Transmission
Line (Shunt)
Figure 7.18 Equivalent lumpedanddistributedcircuits.
Whendesigningwithmicrostripelementsitisimportanttoremember thatthemodels
usedweredevelopedandoptimizedfor specicsubstratethicknesstolinewidthratios.
Thereareoftenavarietyof modelsfor thesamestructuresandit isimportant tochoose
themostappropriateonefor thematerialsandfrequencyrangeused. Whereappropriate
modelsdonotexist, theuseof EMsimulationisnecessary. Thiscanbeusedforaspecic
sectionof thedesign, assimulatingawholecircuit inthisway canbetimeconsuming
anddifcult tooptimize.
Transistorshaveanatural gainslopein[S
21
[ of 6dB/octave. A methodfor compen-
sating for this is to uselossy stubs or equalizers. Thesedo not provideaDC path to
groundandhencedonot upset devicebiasing. Thebasic versionof thisapproachcon-
sistsof aresistor connectedtoano/c stub, asisshowninFigure7.20a. By alteringthe
326 Microwave hybrid amplier realization
Table 7.10 Equalizer behavior as a function of resistor value
Nominal resistor Approximateslope Worst-casereturn
value(O) (dB/octave) loss(dB)
25 5.9 6.1
50 3.4 9.6
75 2.4 12.2
100 1.8 14.1
Swp Max
8 GHz
S(1,1)
Radial Stub
S(1,1)
OC Stub
S(1,1)
Via
S(1,1)
Butterfly Stub
S(1,1)
Double Stub
Swp Min
2 GHz

5
.
0
1 0 . 0

4
.
0

3
.
0

2
.
0

1
.
0

0
.
8

0
.
6

0
.
4

0
.2
0 0
.
2
0
. 2
0
.
4
0
.
6
0
.
81
.
0
2
.
0
3
.
0
4
.0
5
.0
10.0
0
.
4
0
.
6
1
0
.
0
5
.
0
4
.
0
Figure 7.19 Relativeperformanceof short-circuit elements.
resistancedifferentslopescanbeachievedasshowninFigure7.20bandsummarizedin
Table7.10. In narrowband applications theequalizer can beused for stopping high-
frequencyoscillationbyintroducinglossat theproblemfrequency.
Couplers
Thereareanumber of casesinamplier designwhereit isuseful tohavetheabilityto
samplethesignal. Runningatrack closetothetransmissionlinewill intercept someof
theelectrical eldsfromthemainlineandasaresult power will becoupled; thecloser
theline, thehigher thecoupling. Also, themoreeldthemainlinedistributesintothe
surrounding, thegreater thecoupling. A 3dB coupler will transfer half of theenergyto
thecoupledline, however thiswill requireverytightcoupling. Asaruleof thumb, when
theseparationisof theorderof thesubstratethicknessthecouplingwill beabout 20dB.
7.4 Components 327
PORT
P= 1
Z=50 Ohm
PORT
P=2
Z=50 Ohm
WS=0.6
LS=9.2
ID=TL1
W=0.6 mm
L=5.8 mm
ID=TL3
W=0.6 mm
L=5.8 mm
TFR
ID=TL4
W=WR mm
L=LR mm
RS=50
F=5000 MHz
MLEF
ID=TL5
W=WS mm
L=LS mm
Lossy Equalizer
Frequency (GHz)
I
n
s
e
r
t
i
o
n

L
o
s
s

(
d
B
)
R
e
t
u
r
n

L
o
s
s

(
d
B
)
6 GHz
0.08673 dB
DB (|S(2.1)|)(L)
Equalizers
DB (|S(1.1)|)(R)
Equalizers
3 GHz
6.143 dB
3 GHz
5.967 dB
MSUB
Er=9.8
H=0.635 mm
T=0.02 mm
Rho=1
T and=0.0005
ErNom=9.8
Name= SUB 1
(a)
(b)
0
1
2
3
4
5
6
0 2 4 6 8 10
30
25
20
15
10
5
0
MTEES
ID=TL2
1 2
3
Figure 7.20 Lossystubequalizer: (a) schematicof lossystub; (b) performance, S21andS11, as
resistanceisvariedbetween50O (faint traces) and25O (boldtraces).
Couplers are frequency dependent, (,4 long). To increase the bandwidth multiple
sections can be used. In order to achieve tight coupling designers have moved away
fromplanar microstrip structures to multilayer stripline. This allows very small gaps
between thetracks on different layers separated by athin dielectric layer. Thesehave
becomestandarddiscretecomponentsthat canbeeither surfacemountedor boltedina
similar manner toangemounteddevices. Inthiswaycompact quadrature(90

) hybrid
328 Microwave hybrid amplier realization
Figure 7.21 Discretequadrature(90

) hybridcouplers, narrowandoctavebandwidthsfrom2to
8GHz. Courtesyof SJ Technologie, www.sjtechnologie.com.
couplers havebeenproducedwhicharevery important increatingbalancedamplier
designswhicharedescribedinSection11.7. Widebandwidthdesignscanbeboughtoff
theshelf, Figure7.21, andstandardproductsareavailableuptoabout 8GHz, however
for reasonablequantities(>1000p.a.) anumber of companieswill bepreparedtocreate
acustomdesign.
5
Figure7.22showsanMHPA utilizingthesecomponents.
Wherevolumesarenot sufcient andthestandardproductsdont meet thenecessary
performancethereisanalternativeproduct whichcanbeeasily customizedtoproduce
octave band, 3 dB couplers over the desired frequency range. Commonly known as
Wirelines
R _
, these consist of two wires with a tightly controlled separation within a
50Oenvironmentmaintainedbyanouter jacketsuchthattheylooksimilar toapieceof
semirigidcable, asshowninFigure7.23. Theyhavebeensuccessfullyusedinampliers
upto at least 6GHz, althoughcareful attentionto thewireconnections is requiredto
avoidtheinductanceof thewiresreducingperformance.
QuadraturecouplerscanbecreateddirectlyonthesubstrateasshowninFigure7.24.
Themost basicapproachistouseaWilkinsonsplitter witha,4line, Figure7.24a, to
producea90

phasedifferencebetweentheoutputports. Thebalancebetweentheports
isexcellentoverawidebandwidth; howeverthephasedifferenceisfrequencydependent.
Theresistorbetweentheoutputarmsisusedtodissipateanyimbalanceinthevoltages, it
maybeomitted(andoftenisinhigh-powerapplications)atthecostof degradingisolation
5
Thisapproachhasbeensuccessfullyusedina48GHz amplier application.
7.4 Components 329
Figure 7.22 MHPA incorporatingquadraturecouplers: (1) Surface-mount quadraturehybrid
coupler; (2) bolt-inquadraturecoupler; (3) edgecoupler; (4) output circulator. Courtesyof
MicrowaveAmpliersLtd., www.maltd.com.
Figure 7.23 Wirelinecouplers. Photocourtesyof SageLaboratoriesInc., www.sagelabs.com.
betweentheoutputports.Partof theproblemwiththisresistoristheassociatedparasitics,
andas thepower increases andtheresistor gets larger, so theproblemgets worse. For
narrowbanddesignstheresistorcanbeoffsetby,4linesreducingtheparasiticeffectsat
thecenter frequency. TheBranchlineCoupler, Figure7.24bhasanumber of advantages;
330 Microwave hybrid amplier realization
-0
-0.1
-0.2
-0.3
-0.4
-0.5
0
-2
-4
-6
-8
-10
0
-2
-4
-6
-8
-10
160
140
120
100
80
60
40
20
-10
-14
-18
-22
-26
-30
2 4
Frequency (GHz)
DB(|S(2,1) |) (L)
Wilkinson
DB(|S(1,1) |) (R)
Wilkinson
DB(|S(3,1) |) (L)
Wilkinson
Wilkinson
6 8
-2
-4
-6
-8
-10
0
-5
-10
-15
-20
-25
0
-5
-10
-15
-20
-25
0
-10
-20
-30
-40
2 4
Frequency (GHz)
Large
Frequency (Ghz)
DB(|S(2,1) |) (L)
Branchline
DB(|S(2,1) |) (L)
Large
DB(|S(2,1) |) (L)
Lange Back2Back
DB(|S(1,1)|) (R)
Lange Back2Back
DB(|S(3,1)|) (L)
Lange Back2Back
DB(|S(1,1)|) (R)
Large
LargeSdB Back2Back
Frequency (GHz)
Frequency (GHz)
PhaseDifference
DB(|S(3,1)|) (L)
Large
DB(|S(1,1) |) (R)
Branchline
DB(|S(3,1) |) (L)
Branchline
Branchline
6 8
2 4 6 8
2 4 6 8
(a)
(b)
(d)
(e)
(c)
2 4 6 8
Figure 7.24 3dB 90

Couplersandtheir characteristics: (a) Wilkinsonsplitter with90

extension:
(b) branchlinecoupler: (c) 3dB Langecoupler: (d) twoback-to-back8dB Langecouplers:
(e) comparisonof phaseperformance.
7.4 Components 331
theloadresistorisoffsetfromthesignal path, thephaseisrelativelyatoveruptoa20%
bandwidth(thiscanbeextendedbyusingmultiplesections, however thelossincreases
proportionally), andthedesigncanbeadjustedtonon50O output impedances which
can easetheproblemof matching to atransistor. For octaveand greater bandwidths,
Langecouplers [1], Figure7.24c, areextremely effective, thedrawbacks arethesmall
geometries that are required and the wire links between non adjacent ngers. One
solutionhasbeentoproducetheLangeonaseparatesubstratetotherest of thecircuit,
insertingit intopockets. WherethePCBsaresolderedtothehousingoor or acarrier
thenthiscanbedonequitesuccessfully. Itispossibletowire-bondthelinksevenonsoft
substrates, however solderingnewires(thestrandsof 0.2,7equipment wireareideal)
usingsolder pasteandamicroscopeisalsopossible. A methodusedat 3GHz involved
windingseveral turns of wirearoundaneedleandadjustingthespacingto matchthe
pitchof thengers. Applyingsolder pastetotheappropriateareaof thengers, while
holdingthecoil inplace, thejointsweremadewithahotairsolderingpencil. Onalumina
substrateswirebondingisthestandardapproach, althoughwithprocessesthat include
air-bridges thelinks canbefabricatedinsitu. Onesolution[11] tothenegeometries
istoproducetwo8dB Langecouplersbacktoback, Figure7.24d. Wider bandwidths
can be achieved in this way, at the expense of size, complexity and higher insertion
loss.
7.4.3 Transistors
Beware of headline power Figures! A certain amount of gamesmanship is played
amongpower transistor manufacturers. Statementslike200WCWPower Achieved,
needtobeexaminedcarefully,thekeypieceof data,achievedinclassA/BwithaCDMA
signal, under averagepower levelsof 40W,issometimeshiddenaway. Thereisnoneed
togoover theexactspecicationargumentshere, butinbroadtermsif anyoneclaimsto
beabletohandlemorethat4045WCWinapackageddevice, alarmbellsshouldstartto
ring. VeryhighPAEsandpoweroutputshavebeenachieved[12], buttheseareinclassB
andover verynarrowbandwidths. InbroadbandclassA applicationsaGaAsFETsef-
ciency andhenceoutput power isgreatly reducedduetothermal limitations. Although
newwidebandgapdevicesoffer higher power densities, it isactuallycurrent packaging
technologiesthat limit theheat dissipationandhencethemaximumoutput power. Wide
bandgaptransistors havemaximumjunctiontemperatures typically 75

C higher than
GaAs, but the thermal resistance fromthe junction to the case tends to be higher as
the active region of the GaN devices is smaller for the same RF power. Table 7.11
compares some of the main characteristics between 45 W GaAs and GaN pack-
aged devices. It should be noted that the GaAs device has band-specic internal
matching whereas the GaN device is intended for wideband operation and hence
unmatched.
Datasheetsshouldalsobeexaminedwithcareasmanyparametersarequoteddiffer-
ently by manufacturers. For example, somequotelinear gain, whileothers quote1dB
compressedgain.
332 Microwave hybrid amplier realization
Table 7.11 Comparison of commercial GaN and GaAs 45 W C band transistors
Characteristic GaN CGH40045F GaAs TIM374245SL-341 Units
Gain 12.1@3.6GHz 11@3.6GHz dB
P1dB 44(typ.) 40(min.) W
Drainsourcevoltage(max.) 84 15 V
Gatesourcevoltage(max.) 10to 2 5 V
Operatingvoltage 28 10 V
Operatingcurrent 3.5 9.5 A
Thermal resistance 2.8 1.2

C/W
Operatingjunctiontemperature 225 175

C
Packagesize 20.56 24.517.4 mm
Manufacturer Cree Toshiba
Thechoiceof transistor technology essentially boilsdowntoabucks/W decision.
Thisismorecomplicatedthanasimplecomparisonof transistor prices. Theproblemis
simplestinthestandardcommunicationsbandsaround0.9, 1.9, 2.1GHz, wheredevices
arepitchedat specic applications. LDMOS andGaN devices tendtorunoff 28V
whilemost GaAsFETsare 1012V. Dependingonthesupplyrail availabletheprice
of DCDC converters may needto beincludedinthecalculationandwithGaAs and
GaN anegativerail is alsorequired. Theapplicationmay not only specify aparticular
output power, but also thelinearity at that level, inwhich caseclass of operation and
technologywill playafactor.
Thedecision becomes harder away fromthestandard bandwidths. In this casethe
designer needs to look at theinput andoutput impedances andtheir frequency depen-
dence. Althoughimproving, theability of circuit modelstopredict largesignal perfor-
manceisnot exact, andanumber of manufacturersstill onlyprovideS-parameter data.
Nonlinear datathat doesexist isonly accurateover thefrequency rangeandat thebias
and power levels at which they werecharacterized, thus even if known they limit the
designengineers ability to optimally usethedevice. Recent advances inactiveload
pull measurement systems[13] haveenableduserstointeractivelycharacterizeandtest
devicesinsynthesizedimpedanceenvironments. Commercially, thesehavebeenlimited
tonarrowbandsystemstargetedat thecommunicationsfrequencies.
Thereareverylittlebenetsintermsof performanceforoperatingGaAsFETspulsed.
Theprimary limitations arethecurrent capacity of feed networks and packageleads,
andthemaximumbreakdownvoltage. Higher voltagedevicessuchasLDMOS, silicon
bipolar and GaN allow signicantly higher power to be achieved within the average
power packageconstraints. Pulsedradar andL bandavionicsarepracticallytheonly
area that silicon bipolar still holds it own in themicrowave region, although thereis
much work being done producing GaN devices and other new technologies such as
highvoltagevertical FETs [14]. Therelatively lowcost of silicon, thematurity of the
processing, andthelegacy systemsstill operatingwill ensurearequirement for silicon
bipolar devicesfor sometime.
7.5 Amplier design 333
7.5 Amplier design
Theamplier isbiasedtoprovidethenecessarypower, gainandlinearityperformance.
Theoutput match is determined such that thenecessary compression point, saturated
power or PAE is achieved. This is doneeither fromthedevicemodel by simulationor
by direct measurement usingload-pull systems. Theinput matchmay besuchthat the
maximumamount of gain is achieved, aspecic gain or set so that it provides aat
gainresponsewithfrequency. It may benecessary todeliberately mismatchthedevice
at certainpoints to prevent that or subsequent stages frombeingoverdriven. First, we
will look at thetopologies that arecommonly employedandthenwewill takeamore
detailedlookat howimpedancematchingcanberealizedfor MHPAs.
7.5.1 Topologies
Thenumberof stagesrequiredinaline-upisafunctionof theoverall gain, thetechnology
used, thefrequency range, and thetopology. Thegain of theoutput block determines
theoutput power level fromtheprecedingblockandsoon. Inmost casesit isimportant
that thecompressioncharacteristics aredeterminedby thenal stage, thus theearlier
stagesdonot limit theoutput performanceof thewholeamplier. Inthecaseof single
endedline-ups, thatis, whenonedevicedirectlydrivesanother, thedeterminationof the
requiredpowerfromeachstageisfairlysimple. Theoutputpowerof thedriverequalsthe
output power of thesucceedingstagelessthesucceedingstagesgainplustherequired
margin. Although at facevalueit may seemthat theoutput devicewill bethelargest
power transistor availableinthefrequency range, infact therearemany benets from
choosing lower power devices and combining themin parallel to achievethedesired
outputpower. Suchreasonswouldincluderedundancy, bandwidth(higher power usually
meansloweroutputimpedancehencemoredifcultmatching), economics(outputpower
isproportional toprice, it may bemorecost effectivetostandardizeononedeviceand
combinethis is parallel), andspreadingthermal loads. For example, a1kW 12GHz
CWamplier hasbeenconstructed[15] from12810WGaAsdevicesinparallel; one
of thestatedadvantagesisthat afailureof anyonedevicehasnosignicant impact on
theoutput power of thewholeamplier.
If binary combiningisusedasshowninFigure7.25athenthecoupler lossismulti-
plied by thenumber of stages. This loss multiplication factor, n, is proportion to the
number of output stages, s:
n=
logs
log2
(7.11)
AnalternativeistouseamultiwaycombinerontheoutputasshowninFigure7.25b. This
will havelesslossthantheequivalent binary combiner andusing, for example, N-way
Wilkinsoncombiners without resistors it is possibleto achievewidebandwidths with
simpleif largelayouts. Thedisadvantageisisolation; however thiscanbecompensated
forbyusingquadraturecombiningaroundthedevicesthemselves.Singlestage,multiway
334 Microwave hybrid amplier realization
Figure 7.25 Parallel combining: (a) quadraturebinarycombination, output combiner losses
increaseswithnumber of parallel stages; (b) quadraturestagescombinedusingmultiway
combiner lossesproportionallylower asthenumber of parallel stagesincreases.
combinerscanbecreated[16] butthesearenonplanarandthusmoresuitableformodule
combining.
A disadvantageof usingabalanceddesignisthatthetotal outputpower isreducedby
theinsertionlossof thecombiner. For narrowbandsthiscanbeaslowas0.2dB, while
for broader bandwidths theloss will increase. Someof thebenets suchas spreading
thermal loads have already been mentioned, for quadrature combiners there is a key
7.5 Amplier design 335
additional benet: match. Inquadraturecombiners (anddividers) thereis a90

phase
relationship between the combining ports. This means that any reected signals get
dissipatedintheloadresistor, providedthephaseandmagnitudeof thereectedsignals
arethesame. Thebenet of agood match of thecombined channels, whatever their
individual match (provided they are the same), is that deliberate mismatches can be
introduced to achieve gain attening across theamplier bandwidth. This may beto
ensurethatthegainshapeisat, or prevent thedevicefrombeingoverdrivenatoneend
of theband. Devicegainreduceswithincreasingfrequency; hencemoredrivepower is
requiredat higher frequenciesfor thesameoutput power.
Inordertosimultaneouslyachieveimprovedmatchandatgainoverwidebandwidths
feedback can be employed. The theory has been well documented [17], for power
amplier applications series feedback is rarely used. WithFETs shunt feedback must,
bynecessity, incorporateseriescapacitanceaswell asresistancetoseparatethegateand
drainDCvoltages.Boththecapacitorandresistorwill haveparasiticinductance.Thiscan
actuallybeof benet asbyincorporatinginductanceinthefeedback model, it canhelp
toincreasetheRF impedanceandincreasethegainat theupper end. Leadedresistors,
not normally used at microwave frequencies, can be specically chosen to introduce
therequiredinductance. Usingshunt feedback, bandwidths of multipleoctaves canbe
achieved. Thefeedback elements must becapableof handlingthepower levels of the
signals travelling through them, but as they get larger to handle higher powers their
parasiticcomponentsincrease.
Oneof theproblemswithintroducingfeedbacktoMHPAsisthatof incorporatingthe
feedbackelementswithinthedesiredspace. Notonlyarethegateanddrainsseparatedby
several mm, butthereisalsousuallyalargeange. Sometimesitispossibletoconstruct
thefeedback network inthreedimensions, goingover thetopof thedevicerather than
remainingplanar. This approachis labour intensiveandnot suitablefor automation. It
alsotendstobemoresusceptibletoperformancevariationsduetothelackof consistency
incomponent formingandplacement. Heat sinkingof thefeedback elementsintheair
or on top of the package is not easy. An alternative approach, suitable for narrower
band applications has been outlined [18]; originally this was intended for low-noise
applications to avoid the feedback introducing noise back to the input. However, it
has theadvantagefor power applications of separating thefeedback components and
amplifyingdevice. Thecircuit incorporates two ,4lines betweenthedeviceandthe
feedback resistor. Thus, on a low-dielectric material (2.2) at 5 GHz, the feedback
resistor cannowbeabout20mmoffsetfromthemaintrack. Thefeedbackarrangement
isshowninFigure7.26a. Thelow-passlter isarrangedsuchthat thereisa180

phase
shift at the operating frequency (or towards the top of the band in wider bandwidth
applications), such that the feedback has little effect on the performance. At lower
frequencies, theresistor isinbandandadjuststheamountof feedback, thusreducingthe
bottomendgain. Thedevicewithfeedback ismatchedwithatransformer oninput and
output. Thebiascanbeincorporatedwithinthefeedback loopFigure7.26b. Although
thisapproachimprovesstability over partsof theband, caremust betakentocarry out
athoroughstability analysis as at specic frequencies thefeedback canactually cause
oscillation. This approach also reduces theimpact of different devices. Thecircuit of
336 Microwave hybrid amplier realization
TLIN
ID=TL2
Z0=25 Ohm
EL=90 Deg
F0=FF GHz
TLIN
ID=TL3
Z0=25 Ohm
EL=90 Deg
F0=FF GHz
CAP
ID=C2
C=1000 pF
CAP
ID=C1
C=10 pF
CAP
ID=C3
C=10 pF
RES
ID=R1
R=50 Ohm
SRLC
ID=RC2
R=60 Ohm
L=2 nH
C=1000 pF
TLIN
ID=TL2
Z0=ZB Ohm
EL=LB Deg
F0=3.5 GHz
TLIN
ID=TL4
Z0=Zip Ohm
EL=Lip Deg
F0=3.5 GHz
Device Data_S21 & Stability
Gain and Stability
TLIN
ID=TL3
Z0=Zop Ohm
EL=Lop Deg
F0=3.5 GHz
SUBCXT
ID=63
NET=mg0951p
TLIN
ID=TL1
Z0=ZB Ohm
EL=LB Deg
F0=3.5 GHz
SUBCKT
ID=62
NET=Feedback
TLIN
ID=TL4
Z0=100 Ohm
EL=90 Deg
F0=FF GHz
PORT
P=2
Z=50 Ohm
PORT
P=1
Z=50 Ohm
30
20
10
0
10
20
10
0
10
20
4
3
2
1
0
0.5 2.5
Frequency (GHz)
4.5 6
0.5 2.5
Frequency (GHz)
4.5 6
0
0.5
1
1.5
2
PORT
P=2
Z=50 Ohm
PORT
P=1
Z=50 Ohm
(a)
(b)
(c) (d)
TLIN
ID=TL1
Z0=100 Ohm
EL=90 Deg
F0=FF GHz
SRC
ID=RC1
R=RD Ohm
C=C8 pF
RF=800
C8=47
FF=6
DB(|S(2.1)|) (L)
FLC 107WG
K( ) (R)
FLC 107WG
K( ) (R)
MGF0951p
DB(|S(2.1)|) (L)
MGF0951p DB(|S(2.1)|) (L)
FLC 107WFBG
DB(|S(2.1)|) (L)
MGF0951pFB
K( ) (R)
FLC 107WGFB
K( ) (R)
MGF0951pFB
Figure 7.26 Feedbackamplier: (a) feedbackcircuit; (b) devicewithfeedback, biasfeedsand
simpleinput matching; (c) comparisonof [S
21
[ andstabilityfactor kbetweentwodevices;
(d) comparisonof gainandstabilityof thecircuit withbothdevices.
Figure 7.26b was optimized for a Mitsubishi MGF0951 but also simulated with the
EudynaFLC107, Figures7.26candd.
7.5.2 Matching and stability
Beforestartingtodescribespecicmatchingstructuresthereareanumber of common
relationshipsthat needtobeestablished. Fromanamplier point of viewmatchisoften
described by return loss (RL), the ratio of incident to reected signal in dB. When
matching, reection coefcient (I) and the actual impedances are more useful. The
7.5 Amplier design 337
Table 7.12 I, Return loss, transmission loss, and VSWR
I 0.1 0.18 0.2 0.25 0.35 0.4 0.5 0.71 0.8
RL (dB) 20 15 14 12 9 8 6 3 1.9
L
T
(dB) 0.04 0.14 0.18 0.28 0.58 0.76 1.25 3.0 4.44
VSWR 1.22 1.43 1.50 1.67 2.1 2.33 3.00 5.85 9.00
relationshipsbetweenthesetermsaregivenbelowandasummaryof real valuesisgiven
inTable7.12.
I =
z z
0
z z
0
(7.12)
RL (dB) = 20log[I[ (7.13)
When a signal is reected from a mismatch there is an associated loss in power
transferred to the output; this loss, L
T
, in dBs can be related to the reection
coefcient:
L
T
= 10log
_
1[I[
2
_
(7.14)
Itcanbeseenfromthetablethatinordertomaximizethepoweroutputwemustminimize
theloss dueto mismatch. Althoughthis may seemtrivial it is important to realizethe
implicationsof aparticular matchspecication. Thebetter matchedatransistor theless
power isrequiredtomeet aspecication. Oftenanisolator will beaddedtotheoutput
of anamplier inorder tomeetanoutputreturnlossspecication, butthiswill typically
have 0.5 dB of insertion loss and will do nothing in terms of translating the output
impedanceof thedeviceto 50O. Whiletheremay begoodsystemconsiderations for
adoptinganisolator (suchas gainrippleonlongcables), inMHPAs it loses hardwon
power. A solutionbaseduponthebestpower matchthatcanbeachievedwouldbemore
efcient.
Any real impedancecanbematchedtothesystemimpedanceat asinglefrequency,
thedifcultyisdoingitover abandwidthandall amplier circuitsneedtohaveatleasta
limitedbandwidthtoaccountfor changesinbehavior withtemperature. Therearemany
combinationsof matchingelementsabletomovefromanimpedanceononepart of the
SmithChart toanother [19]. However, becausethey arelargely treatedaspurelumped
elementstheir useinMHPAsisrestrictedtoanunderstandingof thetheory, inpractice
the matching elements used are complex. The most common matching elements are
openandshort-circuit stubs. Combinedwithaseriestransmissionlinethesecanmatch
animpedanceover adenedarea, this is best explainedgraphically as inFigure7.27.
Microwavepower deviceoutput impedances will typically liewithin theshaded area,
andsooftenanopencircuit stubistherst matchingelement.
Whenconsideringtheeffects of frequency oneof themost useful considerations is
thequality factor, or Q of theload. Therearemany uses of thetermQ withintheRF
areaandit isimportant not toget themconfused. Inthiscase, werefer toQ
T
, theratio
338 Microwave hybrid amplier realization
PORT
P=1
Z=50 Ohm
PORT
P=2
Z=Zr Ohm
PORT
P=1
Z=50 Ohm
PORT
P=2
Z=Zr Ohm
Figure 7.27 Stubmatchingapproaches. Left-handside: short-circuit canmatchanyimpedance
outsidetheshadedarea; right-handside: open-circuit canmatchanyimpedancewithinthe
shadedarea.
of thereactancetotheresistanceof aseriesimpedance, (for parallel admittanceit isthe
ratioof theconductancetothesusceptance). ThesecanbeplottedontotheSmithchart
asQcurves. Thecloser matchingnetworksstay tothereal axisontheSmithchart the
broader thebandwidththat canbeobtained. Theconverseof thisisthat for anyreactive
matchingthereisanitelimit totheachievablebandwidth. Work intothisrelationship
wasconductedby Fano[20] whodevelopedthetheoremthat wherethesourceor load
includesareactiveelementthematchcannotbeperfectover awidebandwidthnomatter
howmany elements areused. For example, inthecaseof aparallel resistor-capacitor
load(typical of most power devices) theformulais:
_

0
ln
1
[I[
d

RC
(7.15)
As I decreases, the value of the integral increases. For a nite frequency range the
best achievablereectioncoefcient, I
min
that canbeachievedfor agivenloadcanbe
dened. Thisrequirestheuseof another Q term, Q
l
denedas:
Q
l
=
F
0
F
upper
F
lower
(7.16)
I
min
= e
Q
l
Q
T
(7.17)
Or, alternatively, thebandwidthwecanachievefor agivenI is
F
upper
F
lower
F
0
=
1
Q
l
=

Q
T
lnI
min
(7.18)
7.5 Amplier design 339
20
2.708 GHz
15 dB
4.435 GHz
15 dB
5.565 GHz
15 dB
7.292 GHz
15 dB
0
20
40
60
2 4
Frequency (GHz)
R
e
t
u
r
n

L
o
s
s

(
d
B
)
6 8
DB(|S(1,1)|)
Single Section
DB(|S(1,1)|)
Three Section
Figure 7.28 Comparisonof bandwidthachievedfromoneandthree-sectionquarter-wave
transformersbetweenthesameresistiveloads.
Dueto increasing loss in thematching elements themselves, in MHPAs it is rarely
worthgoingbeyondthreematchingsections.Thisisnotthecaseinlterswhereveryhigh
Q low-loss elements canbeusedandhencemulti element structures areencountered.
As arough guideasinglequarter-wavematching structurebetween impedances of a
ratio of 6:1 can achieve a bandwidth of 22% (15 dB return loss). Using multiple
sections this can be increased, e.g., matching the same load with three quarter-wave
transformers, the bandwidth that can be achieved increases to 90%. Clearly, wider
bandwidths can be achieved by reducing the impedance ratio; hence the benet of
using higher voltageand widebandgap materials with higher output impedanceand,
conversely, why with technologies such as LDMOS with its high output capacitance,
onlynarrower bandwidthscanbeachievedinthemicrowaveregion. Figure7.28shows
theimprovement fromoneto threesections, usingideal transmissionlines andpurely
resistive loads. The impedance of the quarter wave matching line, Z
T
, is determined
fromtheformula:
Z
T
=
_
(Z
0
R
L
) (7.19)
For multiplesections onerst needs to calculatetheintermediateimpedancebetween
thesections, Z
i(n)
, [21] fro