roubik gregorian-introduction to cmos op-amps and comparators-wiley (1999)

Upload: maanas-khurana

Post on 08-Mar-2016

198 views

Category:

Documents


20 download

DESCRIPTION

Roubik Gregorian-Introduction to CMOS OP-AMPs and Comparators

TRANSCRIPT

  • Pageiii

    IntroductiontoCMOSOPAMPsandComparators

    RoubikGregorian

  • Pageiv

    Thisbookisprintedonacidfreepaper. Copyright1999byJohnWiley&Sons.Allrightsreserved.PublishedsimultaneouslyinCanada.Nopartofthispublicationmaybereproduced,storedinaretrievalsystemortransmittedinanyformorbyanymeans,electronic,mechanical,photocopying,recording,scanningorotherwise,exceptaspermittedunderSections107or108ofthe1976UnitedStatesCopyrightAct,withouteitherthepriorwrittenpermissionofthePublisher,orauthorizationthroughpaymentoftheappropriatepercopyfeetotheCopyrightClearanceCenter,222RosewoodDrive,Danvers,MA01923,(978)7508400,fax(978)7504744.RequeststothePublisherforpermissionshouldbeaddressedtothePermissionsDepartment,JohnWiley&Sons,Inc.,605ThirdAvenue,NewYork,NY101580012(212)8506011,fax(212)8506008,EMail:[email protected],1800CALLWILEY.

    LibraryofCongressCataloginginPublicationData

    Gregorian,Roubik.IntroductiontoCMOSOPAMPsandcomparators/RoubikGregorian.p.cm."AWileyIntersciencepublication."IncludesIndex.ISBN0471317780(hardcover:alk.paper)1.OperationalamplifiersDesignandconstruction.2.ComparatorcircuitsDesignandconstruction.3.Metaloxidesemiconductors,Complementary.I.Title.TK7871.58.06G741999621.39'5dc219823233CIP

    PrintedintheUnitedStatesofAmerica.1098765432

  • Pagev

    Tomywife,AgnesAndourchildren,ArisandTalin

  • Pagevii

    Contents

    Preface ix

    1Introduction

    1

    1.1ClassificationofSignalProcessingTechniques 1

    1.2ExamplesofApplicationsofOpAmpsandComparatorsinAnalogMOSCircuits

    6

    Problems 16

    References 16

    2MOSDevicesasCircuitElements

    17

    2.1Semiconductors 17

    2.2MOSTransistors 21

    2.3MOSTransistorTypes:BodyEffect 27

    2.4SmallSignalOperationandEquivalentCircuitofMOSFETTransistors 30

    2.5WeakInversion 39

    2.6ImpactIonization 40

    2.7NoiseinMOSFETS 41

    2.8CMOSProcess 44

    Problems 45

    References 47

  • Pageviii

    3BasicAnalogCMOSSubcircuits

    48

    3.1BiasCircuitsinMOSTechnology 48

    3.2MOSCurrentMirrorsandCurrentSources 55

    3.3MOSGainStages 63

    3.4MOSSourceFollowers 74

    3.5MOSDifferentialAmplifiers 77

    3.6FrequencyResponseofMOSAmplifierStages 84

    Problems 92

    References 94

    4CMOSOperationalAmplifiers

    95

    4.1OperationalAmplifiers 95

    4.2SingleStageOperationalAmplifiers 100

    4.3TwoStageOperationalAmplifiers 106

    4.4StabilityandCompensationofCMOSAmplifiers 112

    4.5DynamicRangeofCMOSOpAmps 126

    4.6FrequencyResponse,TransientResponse,andSlewRateofCompensatedCMOSOpAmps

    132

    4.7NoisePerformanceofCMOSOpAmps 137

    4.8FullyDifferentialOpAmps 140

    4.9CMOSOutputStages 149

    4.10OpAmpswithRailtoRailInputCommonModeRange 164

    Problems 170

    References 173

    5Comparators

    175

    5.1CircuitModelingofaComparator 175

    5.2SingleEndedAutoZeroingComparators 177

    5.3DifferentialComparators 182

    5.4RegenerativeComparators(SchmittTriggers) 192

    5.5FullyDifferentialComparators 198

    5.6Latches 205

    Problems 212

    References 213

    6DigitaltoAnalogConverters

    214

    6.1DigitaltoAnalogConversion:BasicPrinciples 214

    6.2VoltageModeD/AConverterStages 218

  • Pageix

    6.3ChargeModeD/AConverterStages 231

    6.4HybridD/AConverterStages 234

    6.5CurrentModeD/AConverterStages 238

    6.6SegmentedCurrentModeD/AConverterStages 244

    Problems 252

    References 254

    7AnalogtoDigitalConverters

    255

    7.1AnalogtoDigitalConversion:BasicPrinciples 255

    7.2FlashA/DConverters 263

    7.3InterpolatingFlashA/DConverters 270

    7.4TwoStepA/DConverters 273

    7.5SuccessiveApproximationA/DConverters 282

    7.6CountingandTrackingA/DConverters 294

    7.7IntegratingA/DConverters 295

    Problems 300

    References 301

    8PracticalConsiderationsandDesignExamples

    303

    8.1PracticalConsiderationsinCMOSOpAmpDesign 303

    8.2OpAmpDesignTechniquesandExamples 316

    8.3ComparatorDesignTechniquesandExamples 349

    Problems 355

    References 355

    Index 357

  • Pagexi

    Preface

    Operationalamplifiers(opamps)andcomparatorsaretwoofthemostintricate,andinmanywaysthemostimportant,buildingblocksofananalogcircuit.Thesecomponentsareusedinsuchdevicesasswitchedcapacitorfilters,analogtodigital(A/D)anddigitaltoanalog(D/A)converters,amplifiers,modulators,rectifiers,peakdetectors,andsoon.Theperformanceofopampsandcomparatorsusuallylimitsthehighfrequencyapplicationanddynamicrangeoftheoverallcircuit.Withoutathoroughunderstandingoftheoperationandbasiclimitationofthesecomponents,thecircuitdesignercannotdetermineorevenpredicttheactualresponseoftheoverallsystem.HencethisbookgivesafairlydetailedexplanationoftheoverallconfigurationsandperformancelimitationsofopampsandcomparatorsexclusivelyinCMOStechnology.Whilethescalingpropertiesoftheverylargescaleintegration(VLSI)processeshaveresultedindenserandhigherperformancedigitalcircuits,theyhavealsochangedthedesigntechniquesusedforCMOSanalogcircuits.Therefore,themainpurposeofthesediscussionsistoillustratethemostimportantprinciplesunderlyingthespecificcircuitsanddesignprocedures.Nevertheless,thetreatmentisdetailedenoughtoenablethereadertodesignhighperformanceCMOSopampsandcomparatorssuitableformostanalogcircuitapplications.

    Themainemphasisofthisbookisonphysicaloperationanddesignprocess.IthasbeenwrittenasaunifiedtextdealingwiththeanalysisanddesignofCMOSopampsandcomparators.Itisintendedforclassroomadoptiontobeusedasaseniororgraduateleveltextintheelectricalengineeringcurriculumofuniversitiesandalsoastrainingandreferencematerialforindustrialcircuitdesigners.Toincreasetheusefulnessofthebookasatextforclassroomteaching,numerousproblemsareincludedattheendofeachchaptertheseproblemsmaybeuseforhomeworkassignments.Toenhanceitsvalueasadesignreference,tablesandnumericaldesignexamplesareincludedtoclarifythestepbystepprocessesinvolved.Thefirsttwo

  • Pagexii

    chaptersprovideaconcise,basiclevel,and(Ihope)cleardescriptionofanalogMOSintegratedcircuitsandthenecessarybackgroundinsemiconductordevicephysics.TheremainderofthebookisdevotedtothedesignofCMOSopampsandcomparatorsandtothepracticalproblemsencounteredandtheirsolutions.ThebookalsoincludestwointroductorychaptersontheapplicationsofopampsandcomparatorsinA/DandD/Aconverters.Foramoredetaileddiscussionontheimportantsubjectofdataconverters,readersarereferredtothePrinciplesofDataConversionSystemDesignbyBehzadRezavi,andDeltaSigmaDataConverters:Theory,DesignandSimulationbyStevenR.Norsworthy,RichardSchreier,andGaborC.Temes.

    ThisbookisbasedinpartonapreviousbookIcoauthoredwithGaborC.Temes,titledAnalogMOSIntegratedCircuitsforSignalProcessing.TheoriginalmaterialhasbeenaugmentedbythelatestdevelopmentsintheareaofanalogMOSintegratedcircuits,inparticularopampsandcomparators.Mostofthematerialandconceptsoriginatedfromthepublicationscitedattheendofeachchapteraswellasfrommanypracticingengineerswhoworkedwithmeovertheyears.

    Sincetheoriginalbookevolvedfromasetoflecturenoteswrittenforshortcourses,theorganizationofthematerialwasthereforeinfluencedbytheneedtomakethepresentationsuitableforaudiencesofwidelyvaryingbackgrounds.HenceItriedtomakethebookreasonablyselfcontained,andthepresentationisatthesimplestlevelaffordedbythetopicsdiscussed.Onlyalimitedamountofpreparationwasassumedonthepartofthereader:mathematicsonthejuniorlevel,andoneortwointroductorylevelcoursesinelectronicsandsemiconductorphysicsaretheminimumrequirements.

    Thebookcontainseightchapters.Chapter1providesabasicintroductiontodigitalandanalogsignalprocessing,followedbyseveralrepresentativeexamplesofcircuitsandsystemsutilizingCMOSopampsandcomparators.Thismaterialcanbecoveredinonelecture(twohourlecturesareassumedhereandthroughoutthepreface).

    InChapter2thephysicsofMOSdevicesisdescribedbrieflyandlinearizedmodelsofMOSFETs,aswellasMOScapacitorsandswitchesarediscussed.ThetechnologyusedtofabricateCMOSdevicesisalsodiscussedbriefly.Onceagain,dependingonthebackgroundoftheaudience,twoorthreelecturesshouldsufficetocoverthecontentofthischapter.

    Chapter3coverssomeofthebasicsubcircuitscommonlyutilizedinanalogMOSintegratedcircuits.Thesesubcircuitsaretypicallycombinedtosynthesizeamorecomplexcircuitfunction.Completecoverageofalltopicsofthischapterrequiresaboutthreelectures.

    InChapter4circuitdesigntechniquesforrealizingCMOSoperationalamplifiersarediscussed.Themostcommoncircuitconfigurations,aswellastheirdesignandlimitations,areincluded.Fullcoverageofalltopicsinthischapterrequiresaboutfourlectures.

    InChapter5theprinciplesofCMOScomparatordesignarediscussed.Firstthesingleendedautozeroingcomparatorisexamined,followedbysimpleandmul

  • Pagexiii

    tistagedifferentialcomparators,regenerativecomparators,andfullydifferentialcomparators.Twolecturesshouldbesufficientforcompletecoverageofthischapter.

    Chapters6and7,whichcoverCMOSdigitaltoanalogandanalogtodigitalconverters,serveaspracticalapplicationexamplesofopampsandcomparators.Thefundamentalsandperformancemetricsofthedataconvertersarepresentedfirst,followedbyadiscussionofpopulararchitecturesofNyquistrateconverters.Digitaltoanalogconvertersaredividedintovoltage,charge,andcurrentscalingtypes.Analogtodigitalconvertersincludehighspeedflash,mediumspeedsuccessiveapproximation,andlowspeedserialconverters.Completecoverageofalltopicsmayrequirethreetofourlectures.

    InChapter8thedesignprinciplespresentedinChapter4and5areemployedtoworkoutseveraldesignexamplestoacquaintthereaderwiththeproblemsandtradeoffsinvolvedinopampandcomparatordesigns.Practicalconsiderationssuchasdcbiasing,systematicoffsetvoltage,andpowersupplynoisearediscussedinsomedetail.AlltopicsinthischaptercanbecoveredinthreelecturesifthedetaileddiscussioninSections8.2and8.3iscondensed,thematerialcanbepresentedintwolectures.

    Thus,dependingonthedepthofthepresentation,fullcoverageofallmaterialinthebookmayrequireasmanyas20twohourlecturesorasfewas16.

    Iamgratefultomanypeoplewhohavehelpedmedirectlyorindirectlyintheelaborateandsometimesoverwhelmingtaskofpublishingthisbook.Inparticular,IwouldliketothankmycolleaguesDrs.S.C.Fan,B.Fotouhi,B.Ghaderi,andG.C.Temes,whoreadandcriticizedversionsofthemanuscript.Theircommentshavebeenmosthelpfulandaregreatlyappreciated.MostofthedifficulttypingtaskwasdonebyMs.W.IrwinandD.Baker.Iamgratefulfortheirexcellentandpainstakinghelp.Last,butnotleast,Iwouldliketoexpressmygratitudetomyfamilyforgraciouslysufferingneglectduringthewritingofthisbook.Withouttheirunderstandingandsupportthisworkwouldnothavebeenpossible.

    ROUBIKGREGORIANSARATOGA,CALIFORNIAJANUARY1999

  • Page1

    Chapter1Introduction

    Operationalamplifiers(opamps)andcomparatorsaretwoofthemostimportantbuildingblocksforanalogsignalprocessing.Opampsandafewpassivecomponentscanbeusedtorealizesuchimportantfunctionsassummingandinvertingamplifiers,integrators,andbuffers.Thecombinationofthesefunctionsandcomparatorscanresultinmanycomplexfunctions,suchashighorderfilters,signalamplifiers,analogtodigital(A/D)anddigitaltoanalog(D/A)converters,inputandoutputsignalbuffers,andmanymore.Makingtheopampandcomparatorfasterhasalwaysbeenoneofthegoalsofanalogdesigners.Inthischapterthebasicconceptofdigitalandanalogsignalprocessingisintroduced.Thenathirdcategoryofsignalprocessing,thesampleddataanalogtechnique,whichisinbetweenthetwomainclassifications,isdescribed.Finally,afewrepresentativeexamplesaregivenofcircuitsandsystemsutilizingCMOSopampsandcomparators,toillustratethegreatpotentialofthesecomponentsaspartofanMOSLSIchip.

    1.1ClassificationofSignalProcessingTechniques[14]

    Electricalsignalprocessorsareusuallydividedintotwocategories:analoganddigitalsystems.Ananalogsystemcarriessignalsintheformofvoltages,currents,charges,andsoon,whicharecontinuousfunctionsofthecontinuoustimevariable.Sometypicalexamplesofanalogsignalprocessorsareaudioamplifiers,passiveoractiveRCfilters,andsoon.Bycontrast,inadigitalsystemeachsignalisrepresentedbyasequenceofnumbers.Sincethesenumberscancontainonlyafinitenumberofdigits(typically,codedintheformofbinarydigits,orbits)theycanonlytakeondiscretevalues.Also,thesenumbersarethesampledvaluesofthesignal,takenatdiscretetimeinstances.Thusboththedependentandindependentvariablesofa

  • Page2

    digitalsignalarediscrete.Sincetheprocessingofthedigitalbitsisusuallyperformedsynchronously,atimingorclockcircuitisanimportantpartofthedigitalsystem.Thetimingprovidesoneormoreclocksignals,eachcontainingaccuratelytimedpulsesthatoperateorsynchronizetheoperationofthecomponentsofthesystem.Typicalexamplesofdigitalsystemsareageneralpurposedigitalcomputeroraspecialpurposedigitalsignalprocessordedicatedto(say)calculatingtheFouriertransformofasignalviathefastFouriertransform(FFT),oradigitalfilterusedinspeechanalysis,andsoon.

    Bycontrast,analogsignalprocessingcircuitsutilizeopamps,comparators,resistors,capacitors,andswitchestoperformsuchfunctionsasfilters,amplifiers,rectifiers,andmanymore.Tounderstandthebasicconceptsofthemostcommonlyusedconfigurationsofananalogcircuit,considerthesimpleanalogtransferfunction

    ItiseasytoverifythattheRLCcircuitshowninFig.1.1acanrealizethisfunction(Problem1.1).Althoughthiscircuitiseasytodesign,build,andtest,thepresenceoftheinductorinthecircuitmakesitsfabricationinintegratedformimpractical.Infact,forlowfrequencyapplications,thiscircuitmaywellrequireaverylargevalued,andhencebulky,inductorandcapacitor.Toovercomethisproblem,thedesignermaydecidetorealizethedesiredtransferfunctionusinganactiveRCcircuit.ItcanreadilybeshownthatthecircuitinFig.1.1b,whichutilizesthreeoperationalamplifiers,iscapableofprovidingthetransferfunctionspecifiedin

    Figure1.1Secondorderfilterrealization:(a)passivecircuit(b)activeRCcircuit.

  • Page3

    Figure1.2Switchedcapacitorrealizationofaresistivebranch.

    Eq.(1.1).Thiscircuitneedsnoinductorsandmayberealizedwithsmalldiscretecomponentsforawidevarietyofspecifications(Problem1.2).Itturnsout,however,thatwhileintegrationofthiscircuitonabipolarchipis,inprinciple,feasible(sincetheamplifiers,resistors,andcapacitorsneededcanallbeintegrated),therearesomemajorpracticalobstaclestointegration.TheseincludetheverylargechipareaneededbytheRCcomponents,aswellasthestringentaccuracyandstabilityrequirementsfortheseelements.Theserequirementscannotreadilybesatisfiedbyintegratedcomponents,sinceneitherthefabricatedvaluesnorthetemperatureinducedvariationsoftheresistiveandcapacitiveelementstrackeachother.Theresultingpolezerovariationsaretoolargeformostapplications.

    Priortomid1970s,analogcircuitssuchastheoneshowninFig.1.1wereimplementedusingintegratedbipolaropampsanddiscretepassivecomponents.Inthe1970stwodevelopmentsmadeitpossibletofullyintegrateanalogcircuitsinmetaloxidesemiconductor(MOS)technology.Thefirstdevelopmentwastheemergenceofatechniquecalledswitchedcapacitor(SC)circuits[6],whichisaneffectivestrategyforsolvingboththeareaandthematchingproblemsbyreplacingeachresistorinthecircuitbythecombinationofacapacitorandafewswitches.ConsiderthebranchesshowninFig.1.2.Here,thefourswitchesS1,S2,S3,andS4openandcloseperiodically,ataratewhichismuchfasterthanthatofthevariationsoftheterminalvoltagevAandvB.SwitchesS1andS4operatesynchronouslywitheachotherbutinoppositephasewithS2andS3.ThuswhenS2andS3areclosed,S1andS4areopen,andviceversa.NowwhenS2andS3close,Cisdischarged.WhenS2andS3open,S1andS4close,andCisrechargedtothevoltagevC=vAvB.Thiscausesachargeq=C(vAvB)toflowthroughthebranchofFig.1.2.Next,CisagaindischargedbyS2andS3,andsoon.IfthiscycleisrepeatedeveryTseconds(whereTistheswitchingperiodorclockperiod),theaveragecurrentthroughthebranchisthen

    ThusiavisproportionaltothebranchvoltagevAvB.Similarly,forabranchcontainingaresistorR,thebranchcurrentisi=(1/R)(vAvB).ThustheaveragecurrentflowinginthesetwobranchesarethesameiftherelationR=T/Cholds.

  • Page4

    Figure1.3Secondorderswitchedcapacitorfiltersection.

    Physically,theswitchestransformthecapacitorC,anondissipativememoriedelement,intoadissipativememoryless(i.e.,resistive)one.

    ItisplausiblethereforethatthebranchofFig.1.2canbeusedtoreplaceallresistorsinthecircuitofFig.1.1b.Theresultingstage[3]isshowninFig.1.3.Inthiscircuit,switchesthatbelongtodifferent''resistors"butperformidenticaltaskshavebeencombined.Furthermore,thesecondoperationalamplifier(opamp)inFig.1.1b,whichactedmerelyasaphaseinverter,hasbeeneliminated.ThiswaspossiblesincebysimplychangingthephasingoftwooftheswitchesassociatedwithcapacitorC3,therequiredphaseinversioncouldbeaccomplishedwithoutanopamp.

    AsFig.1.3illustrates,thetransformedcircuitcontainsonlycapacitors,switches,andopamps.Amajoradvantageofthisnewarrangementisthatnowalltimeconstants,previouslydeterminedbythepoorlycontrolledRCproducts,willbegivenbyexpressionsoftheform(T/C1)C2=T(C2/C1).HeretheclockperiodTisusuallydeterminedbyaquartzcrystalcontrolledclockcircuitandhenceisveryaccurateandstable.TheotherfactorofthetimeconstantisC2/C1,thatis,theratiooftwoonchipMOScapacitances.Usingsomesimplerulesinthelayoutoftheseelements,itispossibletoobtainanaccuracyandstabilityontheorderof0.1%forthisratio.Theresultingoverallaccuracyisatleast100timesbetterthanwhatcanbeachievedwithanonchipresistorandcapacitorfortheRCtimeconstant.

    Adramaticimprovementisalsoachievableforthearearequiredbythepassiveelements.Toachieveatimeconstantintheaudiofrequencyrange(say10krad/s),evenwithalarge(10pF)capacitor,aresistanceof10M isrequired.Sucharesistorwilloccupyanareaofabout106m2,whichisprohibitivelylargeitisnearly10%oftheareaofanaveragechip.Bycontrast,foratypicalclockperiodof10s,thecapacitanceoftheswitchedcapacitorrealizinga10M resistorisC=T/R=105/107=1012F=1pF.Thearearequiredrealizingthiscapacitanceisabout2500m2,only0.25%ofthatneededbytheresistorthatitreplaces.

  • Page5

    TheseconddevelopmentthatmadetherealizationofthefullyintegratedanalogMOScircuitspossiblewasthedesignoftheMOSopamp.Perhapsthemostgenerallyusefulanalogcircuitfunctionisthatoftheoperationalamplifier.Priortoabout1977,thereexistedaclearseparationofthebipolarandMOStechnologies,accordingtothefunctionrequired[1,5].MOStechnology,withitssuperiordevicedensity,wasusedmostlyfordigitallogicandmemoryapplications,whileallrequiredanalogfunctions(suchasamplification,filtering,anddataconversion)wereperformedusingbipolarintegratedcircuits,suchasbipolaropamps.Sincethattime,however,rapidprogressmadeinMOSfabricationtechniquesmadeitpossibletomanufacturemuchmorecomplexandflexiblechips.Inaddition,newdevelopmentsoccurredincommunicationtechnology(suchasdigitaltelephony,datatransmissionviatelephonelines,adaptivecommunicationchannels,etc.)whichrequiredanaloganddigitalsignalprocessingcircuitryinthesamefunctionalblocks.Theanalogfunctionsmostoftenneededarefiltering(forantialiasing,smoothing,bandseparation,etc.),amplification,sampleandholdoperations,voltagecomparison,andthegenerationaswellasprecisescalingofvoltagesandcurrentsfordataconversion.Theseparationoftheseanalogfunctionsfromthedigitalonesmerelybecauseofthedifferentfabricationtechnologiesusedisundesirable,sinceitincreasesboththepackagingcostsandthespacerequirementsandalso,duetotheadditionalinterconnectionsrequired,degradestheperformance.HencetherewasstrongmotivationtodevelopnovelMOScircuits,whichcanperformtheseanalogfunctionsandwhichcanalsosharetheareaonthesamechipwiththedigitalcircuitry.

    Comparedwithbipolartechnology,MOStechnologyhasbothadvantagesanddisadvantages.MOSdevicehasextremelyhighimpedanceatitsinput(gate)terminal,whichenablesittosensethevoltageacrossacapacitorwithoutdischargingit.Also,thereisnoinherentoffsetvoltageacrosstheMOSdevicewhenitisusedasachargeswitch.Furthermore,highqualitycapacitorscanbefabricatedreliablyonanMOSchip.ThesefeaturesmaketherealizationofsuchcircuitsasprecisionsampleandholdstagesfeasibleonanMOSchip[1].Thisisusuallynotpossibleinbipolartechnology.

    Onthenegativeside,thetransconductanceofMOStransistorsisinherentlylowerthanthatofbipolartransistors.AtypicaltransconductancevalueforamoderatesizedMOSdeviceisaround2.5mA/Vforabipolartransistor,itmaybeabout50timeslarger.ThisleadstoahigheroffsetvoltageforanMOSamplifierthanforabipolaramplifier.(Atthesametime,however,theinputcapacitanceoftheMOStransistoristypicallymuchsmallerthanthatofabipolartransistor.)Also,thenoisegeneratedinanMOSdeviceismuchhigher,especiallyatlowfrequencies,thaninabipolartransistor.TheconclusionisthatthebehaviorofanamplifierrealizedonanMOSchiptendstobeinferiortoanequivalentbipolarrealizationintermsofoffsetvoltage,noise,anddynamicrange.However,itcanhavemuchhigherinputimpedancethanthatofitsbipolarcounterpart.

    Asaresultoftheseproperties,thelargestuseoftheMOSopampisexpectedtobeaspartofanMOSLSI(largescaleintegration)chip.Herethedesignoftheopampcantakeadvantageoftheimportantperformancespecificationsthatareneeded.Theloadingoftheopampisoftenverylightandusuallyonlyasmall

  • Page6

    valuedcapacitorhastobedrivenbytheseopamps.Switchedcapacitorcircuitsfallespeciallyintothiscategory,whereelementvalueaccuracyisimportantbutthesignalfrequencyisnottoohighandthedynamicrangerequiredisnotexcessive.Voiceandaudiofrequencyfilteringanddataconversionareinthiscategoryandrepresentthebulkofthepastapplications.

    InadditiontofrequencyselectiveswitchedcapacitorfilteringintroducedinFig.1.3,whichhasbeenthemostcommonapplicationofMOSopamps,therearemanyotherfunctionsforwhichopampsandcomparatorscanbeused.Theseincludeanalogtodigital(A/D)anddigitaltoanalog(D/A)dataconversion,programmablegainamplificationforAGCandotherapplications,peakdetection,rectification,zerocrossingdetection,andsoon.Theyhavealsobeenusedextensivelyinlargemixedsignalanalog/digitalsystemssuchasvoicecodecs,highspeeddatacommunicationmoderns,audiocodecs,andspeechprocessors.Thisrangewillexpandcontinuouslyasthequality(bandwidth,dynamicrange,powerconsumption,etc.)ofthecomponents,especiallyopampsandcomparators,improves.

    1.2ExamplesofApplicationsofOPAMPsandComparatorsinAnalogMOSCircuits

    Inthissection,afewselectedexamplesofpracticalanalogMOScircuitsaregivenwhereCMOSopampsandcomparatorsareusedextensively.Ofcourse,thereadershouldnotexpecttounderstandthedetailsofthesesystemsatthisstage.However,thediagramsmaygiveanideaofthepotentialsofthesecomponentsinanalogsignalprocessing.

    Asmentionedearlier,oneofthemostimportantapplicationsofCMOSopampsisinswitchedcapacitorfilters.Figure1.4ashowsthecircuitdiagramofaseventhorderswitchedcapacitorfilter.ItsmeasuredfrequencyresponseisshowninFig.1.4b.Themeasuredpassbandvariationforthedeviceislessthan0.06dB.Thisrepresentsasuperiorperformance,whichcouldnothavebeenachievedwithoutextensivetrimmingusinganyotherfiltertechnology.

    AnobviousapplicationofaCMOSopampistherealizationofchargemodedigitaltoanalogconverters(DAC).Itcanbeobtainedbycombiningaprogrammablecapacitorarrayandanoffsetfreeswitchedcapacitorgainstage.AnexampleofanNbitchargemodeDACisshowninFig.1.5,whereVrefisatemperaturestabilizedconstantreferencevoltage.TheoutputoftheDACistheproductofthereferencevoltageandthebinarycodeddigitalsignal(b1,b2,b3,...,bN).InChapter6thedesignofsuchcircuitsisdiscussedinsomedetail.

    Modulators,rectifiers,andpeakdetectors[6]belongtoanimportantclassofnonlinearcircuits,whichcanbeimplementedwithacombinationofopampsandcomparators.Inanamplitudemodulatortheamplitudeofasignalx(t)(usuallycalledthecarrier)isvaried(modulated)bym(t),themodulatingsignal.Hencetheoutputsignaly(t)istheproductofx(t)andm(t),ory(t)=x(t)m(t).Aperiodiccarriersignal,whichisreadilygeneratedfromastableclocksource,isasquarewavealternatingbetweentwoequalvaluesV.Aneasywaytoperformmodulationwith

  • Page7

    Figure1.4(a)Circuitdiagram(b)measuredfrequencyresponseofaseventhorderswitchedcapacitorlowpassfilter.

  • Page8

    Figure1.5Multiplyingdigitaltoanalogconverter.

  • Page9

    Figure1.6Switchedcapacitormodulatorwithtwoclocksignals.

    asquarewavecarrieristoswitchthepolarityoftheinputsignalm(t)periodically.AstrayinsensitiveswitchedcapacitormodulatorcircuitwhichperformsaccordingtothisprincipleisshowninFig.1.6.Theclockphases 1and 2areoperatedatthefastclockrate c,whilethephase achangesattheslowcarrierfrequencyrate

    ca.Normally, cismuchlarger(byafactorof30ormore)than ca.

    Anothernonlinearcircuitisafullwaverectifierthatconvertsaninputsignalv in(t)toitsabsolutevalue|v in(t)|.Asimplewayofimplementingaswitchedcapacitorfullwaverectifieristoaddacomparatortoanamplitudemodulator.ThecircuitofaswitchedcapacitorfullwaverectifierbasedonthemodulatorofFig.1.6isshowninFig.1.7a.HereAissetto"1"ifv in>0andto"0"ifv inVmax,theopampoutputgoeshighandM1conducts,chargingCuntilvout v inisreached.Ifv in

  • Page10

    Figure1.7Switchedcapacitorfullwaverectifier:

    (a)completecircuit(b)offsetcompensatedcomparator.

    A/DconverterisshowninFig.1.10.AnalogtodigitalconvertersarediscussedindetailinChapter7.

    WiththerecentrapidprogressmadeinMOSfabricationtechniquesandtheemergenceofthesubmicronCMOStechnology,manyintricatesystemscontaininganaloganddigitalfunctionshavebeencombinedinafullyintegratedform.OnedrawbackofthesubmicronCMOStechnologyisthereductioninthepowersupplyvoltage,whichresultsinareducedsignalswingandhencealowerdynamicrange.Toimprovetheperformanceofthesystemandreducetheeffectsofnoiseinjectionfromthepower,ground,andclocklines,mostmodernhighperformancemixed

  • Page11

    Figure1.8Continuoustimepeakdetector.

    Figure1.9FivebitsuccessiveapproximationA/Dconverter.

  • Page12

    Figure1.10ConceptualdiagramofanNbitA/Dconverter.

    signalintegratedcircuitsmakeuseoffullydifferentialsignalpaths.Withopampsandcomparators,thefullydifferentialsignalpathsrequirefullydifferentialoutputsaswellasinputs,andtheyareknownasfullydifferentialopampsandcomparators.Sincethistechniqueusessymmetricallayout,manyofthenoisevoltages(powersupplynoise,clockfeedthroughnoise,offsetvoltages)appearascommonmodesignals.Theyaretoaconsiderableextentcanceledinthedifferentialoutputvoltagevoutatallfrequencies.AhighfrequencyhighQswitchedcapacitorbandpassfilterthatusesafullydifferentialsignalpathisshowninFig.1.11.Thisfilteristypicallyusedinaradiofrequency(RF)receiversystem,whichrequireshighselectivityathighfrequencies[7].Thetwocomplementaryswitchblocks(X1andX2)areshowninFig.1.12.ThefilterusesfullydifferentialsinglepoletransconductancefoldedcascodeopampswithsourcefollowercommonmodefeedbackasillustratedinFig.1.13[8].Thisopampachieves100MHzunitygainbandwidthand60dBofgainwith1mAoftotalcurrentconsumption.FullydifferentialopampsarediscussedindetailinChapter4.

  • Page13

    Figure1.11Schematicdiagramofasixthorderswitchedcapacitorallpolebandpassfilter.

  • Page14

    Figure1.12Twoswitchblocksforadoublesampling.

    Anotherapplicationofthefullydifferentialopampsisinoversampling,ordeltasigmaA/Dconverters.Theoversamplingconvertersoperateatsamplingratesof16to512timestheNyquistrateandincreasethesignaltonoiseratiobysubsequentfiltering.Theoversamplingtechniqueslendthemselvesmostfavorablytoapplicationsthatrequirearelativelylowfrequency(12bits).Themostobviousapplicationofdeltasigmaconvertersisindigitaltelephony

    Figure1.13Widebandopampforthefilter.

  • Page15

    Figure1.14(a)FullydifferentialCMOSimplementationofasecondorder

    deltasigmamodulator(b)twophaseclockscheme.

    anddigitalaudio.Figure1.14showsafullydifferential,switchedcapacitorCMOSimplementationofasecondorderdeltasigmamodulator[9].Itconsistsoftwoparasiticinsensitiveintegrators,acomparatorthatservesasa1bitA/Dconverter,andatwolevel(1bit)D/Aconverter.Useofafullydifferentialconfigurationattenuatespowersupplynoise,clockfeedthrough,andevenorderharmonicdistortion.Themodulatoroperatesontwophasenonoverlappingclocksconsistingofasamplingphaseandanintegrationphase.Itachieves16bitdynamicrangewithanoversamplingratioof256andasignalbandwidthof20kHz.

  • Page16

    Astheexamplesaboveillustrate,presentdayCMOSopampsandcomparatorsandtheiruseinanalogMOScircuitshavereachedacertainlevelofmaturity.Already,almostanyanalogsignalprocessingtaskinthevoiceoraudiofrequencyrangehasapossiblesolutionusingsuchcircuits.Asfabricationtechnologyandcircuitsdesigntechniquescontinuetoadvance,thespeedanddynamicrangeofthesecircuitswillincrease,allowingtheiruseinsuchlargevolumeapplicationsasvideoandradiosystems,imageprocessing,highspeedtransmissioncircuits,andsoon.

    Problems

    1.1.ShowthatthecircuitofFig.1.1acanrealizethetransferfunctionofEq.(1.1).WhatshouldbetheelementvaluesR,L,andC?

    1.2.CalculatethetransferfunctionoftheactiveRCcircuitofFig.1.1b.AssumethatthecircuitistorealizethetransferfunctionofEq.(1.1).Writetheavailableequationsfortheelementvalues.Howmanyelementvaluescanbechosenarbitrarily?

    References

    1.R.W.Brodersen,P.R.Gray,andD.A.Hodges,Proc.IEEE,67,6175(1979).

    2.Y.Tsividis,Proc.IEEE,71,926940(1983).

    3.R.Gregorian,K.W.Martin,andG.C.Temes,Proc.IEEE,71,941966(1983).

    4.D.J.AllstotandW.C.Black,Jr.,Proc.IEEE,71,967986(1983).

    5.P.R.GrayandR.G.Meyer,AnalysisandDesignofAnalogIntegratedCircuits,2nded.Wiley,NewYork,1984.

    6.R.GregorianandG.C.Temes,AnalogMOSIntegratedCircuitsforSignalProcessing,Wiley,NewYork,1986.

    7.BangSupSongandP.R.Gray,IEEE,J.SolidStateCircuits,SC21(6),924933(1986).

    8.T.C.Choi,R.T.Kaneshira,R.W.Broderson,P.R.Gray,W.B.Jett,andM.Wilcox,IEEEJ.SolidStateCircuits,SC18(6),652664(1983).

    9.B.P.Brandt,D.E.Wingard,andB.A.Wooley,IEEEJ.SolidStateCircuits,SC26(4),618627(1991).

  • Page17

    Chapter2MOSDevicesasCircuitElements

    InthischapterthephysicsofMOS(metaloxidesemiconductor)devicesisdiscussedbriefly.Themostimportantandsimplestcurrentvoltagerelationsaregiven,andsimplemodelsintroducedforMOStransistorsinlinearoperation.Thediscussionhereisinthesimplestpossibleterms,aimedatprovidingsomephysicalunderstandingofthehighlycomplexdeviceoperationforthecircuitdesigner.Precisionanddepthhaveregretfullybeensacrificedintheprocess.Theambitiousreaderisreferredtotheexcellentspecializedworkslistedasreferencesattheendofthechapter.

    2.1Semiconductors

    Inmetals(e.g.,aluminum,copper,silver)thataregoodelectricalconductors,theatomsarearrangedinaregularcrystalarray.Theelectronsfromtheouter(valence)shelloftheatomsarefreetomovewithinthematerial.Sincethenumberofatoms,andthusthenumberoffreeelectrons,isverylarge(ontheorderof1023cm3),evenasmallelectricfieldresultsinalargeelectroncurrenthencethehighconductivityobservedforthesemetals.

    Thepictureisquitedifferentforaninsulatorsuchassilicondioxide(SiO2).Herethevalenceelectronsformthebondsbetweenadjacentatomsandhencearethemselvestiedtotheseatoms.Thusnofreeelectronsareavailableforconductionandtheconductivityisverylow.

    Semiconductors(suchassiliconorgermanium)areinbetweenconductorsandinsulatorsintheirelectricalproperties.Atverylowtemperatures,thevalenceelectronsareboundtotheiratoms,whichagainformaregularlattice.However,asthetemperatureisraised,duetothethermalvibrationsoftheatoms,somebondswillbebroken,andanelectronescapesfromeachofthesebonds.Suchelectronsare

  • Page18

    capableofconductingelectricity.Furthermore,eachfugitiveelectronleavesachargedeficit(calledahole)behindinthebond.Avalenceelectroninabondclosetoaholecaneasilymoveover,fillingtheholeandleavinganewholeinitsownbond.Theeffectisthesameasiftheholehadmovedfromonebondtothenext.Sincethehole''moves"inadirectionoppositethatofthemovingvalenceelectron,inanelectronicfielditbehaveslikeapositivelychargedparticle.

    Electricalconductionisthuspossibleforasemiconductoratroomtemperature.Thedensityofthermallygeneratedelectronsandholesis,however,muchsmallerthanthatofthefreeelectronsinmetal.Typicalnumbersare1010chargecarrierspercubiccentimeterforsiliconand1013ingermanium.Inwhatfollows,thecurrentlydominantmaterial,silicon,isdiscussedexclusively.

    Addingforeignelements(dopants)tothepuresiliconcanraisethenumberoffreechargecarriersinasemiconductor.Silicon(andgermanium)hasfourvalenceelectrons.Ifanatomofanelementwithfivevalenceelectrons(suchasarsenic,phosphorus,orantimony)isaddedtothesemiconductor,itmaytaketheplaceofasiliconatominthecrystallattice.Thusfourofitsvalenceelectronswillparticipateinthefourbondstyingtheatomtoadjacentsemiconductoratomsinthelattice.Thefifthvalenceelectronoftheforeignatom,however,willnothaveaplaceinanybondandwillthusbefreetomoveawayfromitsparentatom.Hencesuchadopantelement(calledadonor,sinceitcontributesfreeelectronstothesemiconductor)enhancestheconductivityofthematerial.

    Addingatomsofanelementwiththreevalenceelectronswillalsocontributetotheconductivity.Nowtherewillbeabondlackingavalenceelectronforeachdopantatom.Thuseachsuchatomcreatesonehole.Thesedopants(e.g.,boron,aluminum,andgallium)arecalledacceptors,sincetheholeswillpropagatebyacceptingboundvalenceelectronsfromadjacentsemiconductoratoms.

    Indopedsemiconductorstherewillbecarriersduetothermaleffectsaswellastothedonor(oracceptor)atoms.Materialscontainingdonorswillthushavebothfreeelectronsandholes,buttherewillbemoreelectronsthanholes.Suchsemiconductorswillbecalledntype,wherenstandsfornegative.Materialscontainingacceptorswillhaveamajorityofholestheyarecalledptypesemiconductors,wherepstandsforpositive.

    Asemiconductorstructurecanalsobefabricatedthatcontainstwoadjacentregionsofdifferenttypes(Fig.2.1).Thesurfacejoiningthetworegionsiscalledapnjunction.Whenthejunctionisfabricated,therandomthermalmotionofthe

    Figure2.1Apnjunctiondiode.

  • Page19

    Figure2.2Ionlayersinapnjunction.

    majoritycarriers(electronsinthentyperegion,holesintheptyperegion)willcauseelectronstospilloverfromthentyperegiontotheptyperegion.Viceversa,holeswillmovefromtheptyperegiontothentypesemiconductor.Thusthisrandommotion(calleddiffusion)resultsintheptypesemiconductorbeingchargednegativelywhilethentyperegionischargedpositively.Theeffectwillbestrongestnearthejunction:There,intheptyperegion,thenegativelychargedacceptoratomswillnolongerbeneutralizedbyholes,and(inthentyperegion)freeelectronswillnolongersurroundthepositivelychargeddonorions.Henceinthisareaadipolelayeroffixedionswillbeformed(Fig.2.2).Theelectricfield compensateforthelargernumberofavailablemajoritycarriers.

    Theequilibriumwillbeupsetifavoltagesourceisconnectedtothewiressolderedtothesemiconductor(Fig.2.3).Assumefirstthatthepolarityofthesourceissuchthatitmakesthepregionmorepositivewithrespecttothenregionthatis,v>0inFig.2.3.Thenvwillreduce causedby,say,abatteryofv=0.8Vcanresultinalargemajoritycarriercurrent(say,

    Figure2.3Circuitfortestingapndiode.

  • Page20

    Figure2.4Currentversusvoltagecharacteristics

    ofapnjunctiondiode.

    i=1A)inthecircuit.Hencevwiththepolarityindicatedwillbecalledforwardvoltageandiforwardcurrent.

    Letusnowreversethepolarityofthevoltagesourcesothatv

  • Page21

    saturationcurrentISflowsforadcvoltagev,andsinceadjacentpositive(+Q)andnegative(Q)chargesarestoredinthedepletionregion(Fig.2.3).Sincethechargestoredisanonlinearfunctionofv,thecapacitanceisnonlinear.WeshalldefinethecapacitanceCbytheincrementalrelationC=dQ/dv.Itcanthenbeshown[1,Sec.3.32,Sec.6.5]thatforthedeviceillustratedinFig.2.3,

    holds.Here S 1.04pF/cmisthepermittivityofthesilicon S= 0KS,where 0isthepermittivityoffreespace( 0 8.861014F/cm)andKS 11.7isthedielectricconstant(alsocalledtherelativepermittivity)ofsilicon.AistheareaofthejunctioninsquarecentimetersandNa(Nd)isthenumberofacceptor(donor)atomspercubiccentimeter.NotethatCdecreaseswith|v|.Itcanbeshown[1,Sec.3.3]thatthequantityunderthesquarerootsignis( S/xd)2,wherexd(incentimeters)isthewidthofthedepletionregionhenceC= SA/xdholds.

    2.2MOSTransistors[1,2,Sec.9.2]

    ConsidernextthestructureshowninFig.2.5.Itisasandwichofseverallayers:Fromtoptobottom,itcontainslayersofmetal,silicondioxide(SiO2,anexcellentinsulator),ptypesilicon,andasecondmetallayerconnectedtoground.Itiscalledametaloxidesemiconductor(MOS)structure.Letvbenegativethenanelectricfieldwillbecreatedacrossthedioxidelayer,whichwillattractpositivecharges(holes)totheregionRunderthetopmetalelectrode(Fig.2.5).ThusnegativechargeswillbestoredinthetopmetalelectrodeandpositivechargesinR.ThedevicewillthusbehaveasacapacitorCofmagnitude

    where oxisthepermittivityoftheSiO2,and ox= 0Kox 0.35pF/cm,whereKoxisthedielectricconstantofSiO2(Kox 3.9).Aistheareaofthetopelectrode,

    Figure2.5Metaloxidesemiconductor

    (MOS)structure.

  • Page22

    Figure2.6Capacitanceversusvoltagecharacteristics

    ofanMOSstructure.

    andlisthethicknessoftheSiO2layer.*TheptypeSilayerbetweenRandthebottommetallayerbehavesasaresistorhencetheoverallstructuresimulatesalossycapacitor.

    Next,letvbeasmallpositivevoltageinFig.2.5.Theelectricfieldwillnowrepelholes.Asaresult,thefixednegativelychargedionsinRwillbeabandonedbythemobileholes,andanetnegativespacechargewillappearinR,whichisnowadepletionlayer.Thuschargeisagainstoredinthetopelectrodeandacapacitoriscreated.Forverysmallvaluesof Eq.(2.3)willremainvalidforthemagnitudeofthecapacitance.Asthevalueofvisincreased,however,thechargeinRbecomesgreatersincethedepletionregionwidens.Sincetheaverageionisnowfartherawayfromthesurface,theeffectivevalueoflinEq.(2.3)increasesandCdecreases.

    Ifvisincreasedevenfurther,aneweffectappears.Sincethethermalgenerationofholesandelectronsoccurscontinuouslyinthesemiconductor,ifthefieldcreatedbyapositivevisstrongenough,itcanattractthermalelectronstoRthesewillthenmovetothesurface.Whenthisoccurs,thecapacitorstorespositivechargesinthetopelectrode,whilenegativeones(electrons)arestoredinthesurfacelayer.Thus,inEq.(2.3),lagainbecomestheSiO2thickness,andhenceChasthesamevalueasithadfornegativevoltagev.TheoverallbehaviorofCasafunctionofvisillustratedschematicallyinFig.2.6,whichalsogivesthenamesofthethreeoperatingregions.Thenamesofthefirsttwoareevidentthethirdoneiscalledtheinversionregion,since(duetothehighvoltagev)mobileelectronsareattractedintoR,whichthusbehavesasann(ratherthanap)typematerial.Itshouldbenotedthatsincethermalelectronsaregeneratedataslowrate,thevoltagevmustbepresentforsometimebeforetheinversionlayerisformedhenceitwillnotappearifvisahighfrequency(say,f>1kHz)signalratherthanaconstantvoltage.

    ConsidernextthestructureshowninFig.2.7.Anewfeatureisthepresenceoftwon+(i.e.,heavilydopedntype)regionsintheptypematerial.TheoneontheleftwillbecalledthesourceavoltagevSisconnectedtoit.Then+regiononthe

    *Often,theoxidethicknesslismeasuredinangstroms(1=108cm).Usualvaluesoflarebetween50and200.

  • Page23

    Figure2.7MOStransistor.

    rightwillbecalledthedrainitsvoltageisdenotedbyvD.ThetopmetalelectrodewillbecalledthegateitsvoltageisvG.Thebodyofthesemiconductorisusuallycalledthesubstrateorbulk.TheoveralldeviceistheMOStransistor.Itsoperationisdiscussedbrieflynext.

    Letthesourcebegrounded,sothatvS=0.Also,letvDhaveasmallpositivevalue,say0.5V.WewillconsiderthebehaviorofthedraincurrentiDasvGisraisedfromzerotohigherpositivevalues.Sincethegateisinsulatedfromtherestofthedevicebytheoxidelayer,itwillnotconductanycurrent.Then+drainregionandthesurroundingptypesubstrateformapnjunction.Sincethesubstrateisgrounded,whilevD>0,thisjunctionisreversebiased.HenceforvG=0,iD 0.

    AsvGisincreased,theregionRunderthegatewillfirstbedepleted,theninverted,asdiscussedearlierinconnectionwithFigs.2.5and2.6.WhenRisdepleted,iDremainszero,sincetheareaaroundthedrainisstillreversebiased.However,thesituationchangeswhenvGissolargethatinversionoccurs,sothatRisfilledwithelectrons.Now,alayercontainingmobileelectrons,calledaninversionlayerorchannel,connectsthedraintothesource.Sincethedrainispositivewithrespecttothesource,electronswillflowfromthesourcetothedrainandapositivecurrentiD>0willbeobserved.ThesmallestvoltagevGnecessarytoproduceachanneliscalledthethresholdvoltageandisdenotedbyVT.Usually,VTisgivenasthevGvalueneededforiD=1Aitmayrangefromafractionofavolttoseveralvolts.

    ItshouldbenotedthatforthestructureofFig.2.7,mostoftheelectronsinthechanneldonotoriginatefromthermaleffectsinthebulkinstead,theyaredrawnbytheelectricfieldduetovGoutofthesource.Someelectronsarealsodrawnfromthedrainhowever,sincevD>0,thedrainsubstratejunctionismorereversebiased,andhenceitisharderforelectronstoescapefromthedrain.

    SinceapotentialdifferencevDexistsbetweenthetwoendsofthechannel,theelectronsinthechannelwillbeattractedtothedrain.Therefore,inadditiontotherandomthermalmotionoftheelectrons,asteadymotion(calleddrift)willoccur,whichcausesthecurrentflow.ForsmallvD,thechannelwillthereforebehaveasaresistor,andhenceiD vD/R,wherethechannelresistanceRisgivenby

  • Page24

    Figure2.8PinchoffinanMOStransistor.

    HereListhelengthandWthewidthofthechannel,andnisthemobilityoftheelectronsinthechannel,*definedbytherelation(electrondriftvelocity)=(mobility)(electricfield).Finally,Qnisthechargedensity(inC/cm2)oftheelectronsinthechannel.SincevGcanbeconsideredasthesumoftwoterms,VT(necessarytomaintainthedepletionregionunderthechannel)andvGVT(necessarytomaintainthechannel),wehave

    whereCox= ox/listhecapacitance(perunitarea)oftheoxidelayerseparatingthegatefromthechannel.Hence,forsmallvD( ),therelation

    holds.Thusthetransistoractsasaresistor,withresistanceR=[nCoxW/L(vGVT)]1controlledbyvG.

    IfvDisincreasedsothatitisnolongernegligiblecomparedtovG,Eq.(2.6)willbecomeinaccurate.SincethepotentialofthechannelatthegroundedsourceiszerowhileatthedrainitisvD,wecanassumethatitsaveragepotentialisvD/2.Hencetheaveragevoltagebetweenthegateandchannelis(vGvD/2).ReplacingvGby(vGvD/2)inEq.(2.6)gives

    Equation(2.7)remainsagoodapproximationforvD

  • Page25

    fied.Asthefigureindicates,duetothevariationofthepotentialalongthechannel,thechargedensityQndecreasesnearthedrain.IfvD=vGVT,atthedrainthegatetochannelvoltageisnolongersufficienttomaintainthechannel.Thusthedepletionregionsurroundingthesource,thechannel,andthedrainextendsallthewaytothesurface.Thisphenomenonissometimescalledpinchoff,andtheregionwhereitoccursisthepinchoffpoint(Fig.2.8).IfvDisincreasedfurther,thepinchoffpointwillmovetowardthesource,sincetheareawherevGvD VTwillincrease.Hencethechannelwillnowextendonlyfromthesourcetothepinchoffpoint,thelatterbeingsomewhereunderthegate.Theregionbetweenthepinchoffpointandthedrainisdepleted.Electronsfromthechannelareinjectedintothisdepletionregionatthepinchoffpointandareswepttothedrainbythefieldcreatedbythepotentialdifferencebetweenthedrainandthepinchoffpoint.Thevoltage isthusdividedbetweenthetwoseriesconnectedregions:thechannelbetweenthesourceandthepinchoffpoint,andthedepletionregionbetweenthepinchoffpointandthedrain.Clearly,thelatterhasahigherresistance,andhencemostofvDSinfactappearsacrossit.AnyincreaseofvDwill,toagoodapproximation,resultinanequalvoltageincreaseacrossthedepletionregionandwillhardlychangeiD.Thus,forvDvGVT,fromEq.(2.7),

    ThisphenomenoniscalledsaturationvDsat=vGVTisthedrainsaturationvoltageandiDsatasgivenbyEq.(2.8),isthedrainsaturationcurrent.

    Thedraincurrentdoes,inreality,increasesomewhatwithincreasingvD.ThiscanbeattributedtothemoveofthepinchoffpointtowardthesourceforincreasingvDandhencetotheshortenedchannelasEq.(2.8)indicates,iDincreasesasLisreduced.Asanapproximation,thiseffect(oftencalledchannellengthmodulation)canbeincludedintheformulaforiD(vD)intheformofanaddedfactor(1+lvD).HerelisadeviceconstantthatdependsonL,onthedopingconcentrationofthesubstrate,andonthesubstratebias(discussedinthenextsection).ForL 10mm,typicallyl 0.03V21generally,l~1/L.

    Itisusualtointroducetheabbreviations ThenthesaturationcurrentgivenbyEq.(2.8)becomes

    whichincorporateschannellengthmodulation.Figure2.9showsthevariationofiDwithvGforconstantvD.Figure2.10illustratesitsdependenceonvDforvariousvGvalues,wherevG1,vG2,vG3.

    AllderivationsofthissectionwereperformedforthestructureshowninFig.2.7,whosesource,drain,andchannelwereallntype.ThisdeviceiscalledannchannelMOS,orNMOStransistor.Asimilararrangementcanbeconstructedbycreatingp+drainandsourcediffusionsinanntypesubstrate.NowanegativevGisneededtocreateaptypechannelunderthegate,andanegativevDisusedto

  • Page26

    Figure2.9Draincurrentversusgatevoltage

    characteristicsofanMOStransistor.

    attracttheholesinthechanneltothedrain.Also,iDwillbenegativeifthereferencedirectionofFig.2.7isused.TheresultingdeviceiscalledapchannelMOSorPMOStransistor.Formulas(2.3)to(2.9)remainvalidifsomesmallchangesaremade.Themobilitynofelectronsmustbereplacedbyp,theholemobilityinthechannel.Aswouldbeexpectedfromthemoreelaboratemechanismofholeconduction,p

  • Page27

    Figure2.11Transistorsymbols.

    ThecircuitsymbolsusedforNMOSandPMOStransistorsareshowninFig.2.11aandb,respectively.Ifthetypeisunimportant,thesimplifiedsymbolofFig.2.11cmaybeusedforbothNMOSandPMOSdevices.

    Sincetheoperationofthedevicesdescribedinthissectionisdependentontheelectricfieldinducedbythegatevoltage,theyarecalledfieldeffectdevices(FETs),orMOSFETs.*

    SincePMOStransistorsaremoreeasilyfabricatedthanNMOStransistors,theywereinitiallypredominant.However,later,whenthetechniquesforthereliableproductionofNMOSdevicesweredeveloped,thelatterbecamestandard.Themainreasonforthisisthehighermobilityofelectrons,whichmakestheNMOStransistorsfasterthanPMOStransistors.

    2.3MOSTransistorTypes:BodyEffect

    TheMOStransistorsdescribedinSection2.2,bothNMOSandPMOStypes,shareseveralfeatures.Inthestructure,thegateisinsulatedelectricallyfromtherestof

    *Sincethechargecarriershereareeitherelectronsorholes(notboth),FETsarealsosometimescalledunipolardevices,tocontrastthemwithbipolartransistors,inwhichbothelectronandholecurrentsexist.

  • Page28

    Figure2.12Symbolsfordepletionmodetransistors.

    thedevicebytheSiO2layerunderit.Henceitisoftencalledaninsulatedgatefieldeffecttransistor(IGFET).Also,thevoltagevGinducesandenhancesthedraincurrent.Thusthedevicesdescribedoperateintheenhancementmode.

    ItisalsopossibletofabricateanMOStransistorthatconductsdraincurrentwhenvG=0.Forexample,anntypelayercanbeintroducedbydoping,whichconnectsthesourceanddrainofanNMOSdevice.Withsuchadopedchannel,thefieldofthegateisnotneededtoproduceaninversionlayertheregionR(Fig.2.7)nowhasa''builtin"conductingntypechannel.

    However,ifanegativegatevoltageisapplied,thefieldthuscreatedwillrepelelectronsandcreateadepletionlayerinthechanneladjacenttotheSiO2surface,therebyreducingtheconductivityandthusthedraincurrent.Ifthemagnitudeofthenegativegatevoltageissufficientlylarge,thechannelbecomescompletelydepletedandiD 0results.ThevaluevGatwhichthisoccursisagaincalledthresholdvoltageandisdenotedbyVT.Now,however,VT0,byestablishingaptypedopedchannel.

    Therelations(2.6)to(2.11)remainvalidfordepletionmodedevicesifthevalueandsignofVTischosenappropriately,asdescribedabove.TwosymbolsoftenusedtodenotedepletionmodeMOSFETsareshowninFig.2.12.

    Atotallydifferentstructurecanalsobeusedtoproduceadepletionmodefieldeffecttransistor(Fig.2.13).Here,alightlydopedntypelayer(channel)connectsthen+sourceanddrainregionsandthegateisap+regionimplantedinthislayer.Hence,forvS=vG=0andvGD>0,adraincurrentwillflow.IfvGismadenegative,thep+implantactingasthegatewillbesurroundedbyadepletionlayerthegreater|vG|,thedeeperthelayer.Themobileelectronsinthechannelcannot

    Figure2.13Junctionfieldeffecttransistor.

  • Page29

    enterthisdepletionlayer,ortheonealongthepnjunctionbetweenthechannelandthesubstrate.Hencetheeffectivecrosssectionofthechannelisreducedas|vG|isincreased.AtsomevaluevG=VP(VP

  • Page30

    TABLE2.1.KeyUnitsandConstantsforMOSTransistors

    Thisphenomenon,thebodyeffect,isamajorlimitationofMOSdevicesoperatedwithvS vBitsevilinfluencewillbelamentedrepeatedlylaterinthebook.AsEqs.(2.12)and(2.13)show,toreducethebodyeffect,Nimpshouldbemadesmall.However,forverysmallNAvalues(say,NA

  • Page31

    TABLE2.2.DrainCurrentRelationsforMOSFETsinLargeSignalLowFrequencyOperation

  • Page32

    Inthefollowingdiscussion,weconcentrateonthelinearizedapproximationandmodelingprocessforMOSFETsoperatingintheirsaturationregions,whichistheusualconditionforlinear(analog)operation.Afterward,wegiveabriefoverviewofthelinearizationandmodelingfordevicesthatoperateintheirtriode(nonsaturated)orcutoffregion.AssuminganNMOStransistor,andcombiningEqs.(2.9)and(2.12),therelation

    results.Hereweused wecanwrite

    Here( iD/ vGS)0andsoondenotethepartialderivativesevaluatedatthebiaspoint. iDisthedeviation(increment)ofiDfromitsbiasvalue vGS, vDS,and vSBaretheincrementsofvGS,vDS,andvSB.AlldeviationsmustbesmallforEq.(2.15)tohold.Ifonlytheincremental(smallsignalac)componentsareofinterest,Eq.(2.15)canbewrittenas

    Figure2.14LowfrequencyequivalentcircuitofaMOSFET.

  • Page33

    where

    Heregdisthe(incremental)drainconductancegmandgmbaretransconductancesthatcanberepresentedbyvoltagecontrolledcurrentsources(VCCSs).Henceanequivalentcircuitmodel,showninFig.2.14,canbeconstructed.Thevaluesofgm,gmb,andgdcanbefoundfromEq.(2.14):

    Hence,toagoodapproximation,gmandgmbareproportionalto

    TheotherimportantcomponentsofthecompletesmallsignalmodeloftheMOSFETarethecapacitorsrepresentingtheincrementalvariationsofstoredchargeswithchangingelectrodevoltages.Theseplayanimportantroleinthehighfrequencyoperationofthedevice.TheintrinsiccomponentsoftheterminalcapacitancesoftheMOSFETdevices(associatedwithreversebiasedpnjunctionsandwithchannelanddepletionregions)arestronglydependentontheregionofoperation,whiletheextrinsiccomponents(duetolayoutparasitics,overlappingregions,etc.)arerelativelyconstant.Assumingagainthatthetransistoroperatesinthesaturationregion,itcanbeassumedthatthechannelbeginsatthesourceandextendsovertwo

  • Page34

    thirdsofthedistancetothedrain.Inthisregionofoperation,themostimportantcapacitancesarethefollowing:

    1.Cgd:GatetoDrainCapacitance.Thisisduetotheoverlapofthegateandthedraindiffusion.Itisathinoxidecapacitance,andhencetoagoodapproximationcanberegardedasbeingvoltageindependent.

    2.Cgs:GatetoSourceCapacitance.Thiscapacitancehastwocomponents:Cgsov,thegatetosourcethinoxideoverlapcapacitance,and isnearlyvoltageindependentinthesaturationregion.

    3.Csb:SourcetoSubstrateCapacitance.Thiscapacitancealsohastwocomponents:Csbpn,thepnjunctioncapacitancebetweenthesourcediffusionandthesubstrate,and whichcanbeestimatedastwothirdsofthecapacitanceofthedepletionregionunderthechannel.TheoverallcapacitanceCsbhasavoltagedependencewhichissimilartothatofanabruptpnjunction.

    4.Cdb:DraintoSubstrateCapacitance.Thisisapnjunctioncapacitanceandisthusvoltagedependent.

    5.Cgb:GatetoSubstrateCapacitance.Thiscapacitanceisusuallysmallinthesaturationregionitsvalueisaround0.1COX.

    Figure2.15illustratesthephysicalstructureofanNMOStransistorandthelocationsofthecapacitancesinthecutoff(Fig.2.15a),saturation(Fig.2.15b),andnonsaturationortriode(Fig.2.15c)regions.Table2.3liststheterminalcapacitorsoftheNMOSdeviceandtheirestimatedvaluesinthethreeregionsofoperation.ThenotationsusedarethoseshowninFig.2.15atoc.Figure2.16depictsthecompletehighfrequency(ac)smallsignalmodeloftheMOSFET.InanalyzingthesmallsignalbehaviorofMOSFETs,themodelofFig.2.14canbeusedifonlylowfrequencysignalsarepresentifthecapacitivecurrentsarealsoofinterest,thecircuitofFig.2.16mustbeapplied.

    FromthemodelsofFigs.2.14and2.16andaccompanyingdiscussions,anumberofgeneralstatementscanbemadeaboutthedesirableconstructionofaMOSFET:

    1.Forhighacgain,gmshouldbelarge.Thiswillbethecase,byEq.(2.18),if shouldbeaslargeastheallowabledcpowerdissipationpermits.

    2.AsthenegativesigninEq.(2.19)indicates,thebodyeffectreducesthegain.Tominimizegmb,byEqs.(2.19)and(2.13),weneedlargeCOX,smallNimp(i.e.,lightlydopedsubstrate),andalargebiasvoltage forthesource.

  • Page35

    Figure2.15ParasiticcapacitancesinMOSFETinthe(a)cutoffregion,(b)saturationregion,and(c)trioderegion.

    (Ofcourse,ifvSBisconstant,noincrementalbodyeffectoccursandtheserequirementsareirrelevant.)

    3.Ideally,theMOSFETinsaturationshouldbehaveasapurecurrentsource.Hence,asFig.2.14illustrates,gdshouldbesmall.ByEq.(2.20)thisrequiresasmallbiascurrent andasmall .Since isintroducedbychannellengthmodulation,itcanbereducedbyincreasingLandalso[1,Sec.8.4]byincreasingNimp.AsummaryoftheformulasderivedinthissectionisgiveninTable2.4.

  • Page36

    TABLE2.3.TerminalCapacitancesofaMOSFETintheThreeMainRegionsofOperationa

  • Page37

    Figure2.16HighfrequencyequivalentcircuitofaMOSFET.

    Next,wediscussbrieflythelinearmodelofaMOSFETthatisbiasedinitsnonsaturated(triode)region.Usually,suchadeviceisusedasaswitchthatisopenedorclosed(turnedonoroff)byalargegatevoltageorasafairlylinearlargevaluedvariableresistor.Hence,herewederiveitsmodelonlytoanalyzeitsbehaviorinsuchapplications.WeassumethatvGSisconstantandthat andhencenegligible.UndertheseconditionsitcanbeseenfromEq.(2.7)thatwhenthedeviceconductsdraincurrent,itbehaveslikearesistorconnectedbetweenthedrainandsourceterminals.TheequivalentsmallsignaldraintosourceresistanceforthecaseofvDSnearzeroisgivenby

    andisthuscontrolledbythegatevoltageoverdrive

    Inhighfrequencyapplication,thedevicecapacitancesmustalsobeincludedinthemodel.AsimpleequivalentcircuitisshowninFig.2.17.Sincenowacontinuouschannelextendsfromthesourcetothedrain,thegatetochannelcapacitanceC'gsisconnectedtoboththedrainandthesource.Anaccuratehighfrequencyrepresentationofthechannelshouldincludeadistributedresistivelineextendingfromthesourcetothedrainandcapacitivelycoupledtoboththegateandthesubstrate.However,asafirstapproximation,C'gscanbetreatedasalumpedcapacitancepartitionedequallybetweenCgsandCgd,asindicatedinthelastrowofTable2.3.

    Finally,ifthedeviceiscutoff,nochannelexistsandthemodelcontainsonly

    Figure2.17HighfrequencymodelofaMOSFETinitstrioderegion.

  • Page38

    TABLE2.4.SmallSignalParametersofMOSFETSinSaturationa

    aSeeFig.2.16.| vDS| 1isassumedinallformulas.

  • Page39

    Figure2.18HighfrequencymodelofaMOSFETinitscutoffregion.

    thecapacitances,withthevalueslistedinthefirstrowofTable2.3.AsimplifiedmodelofaMOSFETinthecutoffregionisshowninFig.2.18,wherethedrainsourceresistanceisinfinityandCgsandCdscapacitorsareduetogateoverlapandfringingcapacitances.ThegatetosubstratecapacitanceCgdinthecutoffregionis,however,highlynonlinear,andforthegatetosourcevoltagearoundzeroitsvalueisapproximatelyequaltoWL'Cox,whereL =L2LovandLovisthelengthoftheoverlapbetweenthegateandthesourcedraindiffusionregions.

    2.5WeakInversion

    ThetriodeandsaturationregionsofoperationdiscussedearlierinthischapterassumethatthedeviceisoperatedinstronginversionandVGSVT 100mV(foranNMOStransistor).IfVGSVT

  • Page40

    2.6ImpactIonization[4]

    OneofthesevereproblemsinsubmicronMOStechnologiesoperatingatsupplyvoltagesaround5Visimpactionization.Figure2.19illustratesannchannelMOSdevicecrosssectionshowingtheimpactionizationcurrentflowandtheIVcharacteristicasaresultofimpactionization.Asdepictedinthefigure,whenthedraintosourcevoltageisincreased,thestrengthoftheelectricfieldatthedrainendofthechanneleventuallybecomeshighenoughtoinducesignificantimpactionizationcurrentwhichoriginatesfromthedraindepletionregionandflowsintothesubstrate.Oncethishappensthecurrentthatflowsintothedrainterminalhastwocomponents.OnecomponentistheMOStransistorchannelcurrentthatflowsfromthedraintothesource,andtheotheristheimpactionizationcurrentthatflowsfromthedraintothesubstrate.Theimpactionizationcurrentisnotafunctionofthetransistorchannellength,andthemagnitudeofthecurrentisnotreduceddramaticallysimplybymakingthelengthlonger.Thecurrentislargelydeterminedbythepeakelectricfield,whichinturnisafunctionofthegateoxidethickness,drainjunctiondepth,dopingconcentrationinthesubstrate,thevoltagebetweenthedrainterminalandthedrainendofthechannelregion,andthegatetodrainvoltage.Intechnologieswithfeaturesizesintherangeof2m,forannchannelMOSdevice,theimpactionizationcurrentequals1%ofthedraincurrentwhenthevoltagebetweenthedrainandthedrainendofthechannelisintherange4to9V,andthedeviceisbiased

    Figure2.19(a)AnnchannelMOSdevicecrosssectionshowingimpactionizationcurrentflow(b)IVcharacteristic

    observedasaresultofimpactionization.

  • Page41

    inthesaturationregion.InpchannelMOSdevicestheeffectoccursatsubstantiallyhigherfieldstrengths.

    Theimpactionizationhasseveralpotentiallydamagingsideeffects.OneseriousnegativeconsequenceshowninFig.2.19bisdegradationofthetransistoroutputimpedance,whichresultsinreducedgaininamplifierstagesthatusetransistorsasactiveloads.Onewaytodealwiththisproblemisthroughcircuittechniques,whereashieldingnchanneldeviceisplacedinserieswiththetransistor,preventingitfromhavingaVdSgreaterthanhalfthesupplyvoltage[4].Theseconddamagingeffectisthepossibilityoftriggeringlatchupduetotheohmicdropinducedbytheionizationcurrentthatflowsintothesubstrate.Latchupisaphenomenoncausedbytheparasiticlateralpnpandnpnbipolartransistorscreatedonthechip.Thecollectorsofeachtransistorfeedtheother'sbase,andthiscreatesanunstabledevicesimilartoapnpnthyristor[5].Thiscausesasustaineddccurrentthatmaycausethechiptostopfunctioningandmayevendestroyit.Latchupmaybepreventedbypropersubstratestrappingandusingguardringstosurroundsomecriticaltransistorsonthechip.Anotherstrategyistoreducethesubstrateresistance.Inthismethodthepandnchanneltransistorsareformedinalightlydopedepitaxiallayerthatisgrownonalowresistivitysubstrate.Finally,thethirdserioussideeffectofimpactionizationisthethresholdshiftoftheMOSdeviceduetothecontinuousoperationintheimpactionizationmode.Thisphenomenonisduetothehighelectricfieldwhichcreateshighenergycarriesthatcanbetrappedinthegateoxide,resultinginlongtermthresholdshift.Severalprocessmodificationshavebeenproposedthatareeffectiveatraisingthevoltageatwhichimpactionizationbecomesaproblem.Onetechniqueistolowertheimpuritygradientinthedrainjunctionusinglightlydopeddrain(LDD)structures.

    2.7NoiseinMOSFETs

    Therearethreedistinctsourcesofnoiseinsolidstatedevices:shotnoise,thermalnoise,andflickernoise.

    ShotNoise

    Sinceelectriccurrentsarecarriedbyrandomlypropagatingindividualchargecarriers(electronsorholes),superimposedonthenominal(average)currentI,thereisalwaysarandomvariationinS.Thisisduetofluctuationinthenumberofcarrierscrossingagivensurfaceintheconductorinanytimeinterval.ItcanbeshownthatthemeansquareofinSisgivenby

    whereq=1.61019Cisthemagnitudeoftheelectronchargeand fisthebandwidth.Thisformulaonlyholds,however,ifthedensityofthechargecarriersissolowandtheexternalelectricfieldissohighthattheinteractionbetweenthe

  • Page42

    Figure2.20Thermalnoiseinaresistor:(a)noisyresistor

    (b)and(c)equivalentcircuits.

    carriersisnegligible.Otherwise,therandomnessoftheirdensityandvelocityisreducedduetothecorrelationintroducedbytherepulsionoftheircharges.ThenoisecurrentisthenmuchsmallerthanpredictedbyEq.(2.24).

    InaconductingMOSFETchannel,thechargedensityisusuallyhighandtheelectricfieldislow.Therefore,Eq.(2.24)doesnothold.Thenoisecurrentduetorandomcarriermotionsishencebetterdescribedasthermalnoise,whichisdiscussednext.

    ThermalNoise

    InarealresistorR,theelectronsareinrandomthermalmotion.Asaresult,afluctuatingvoltagevnTappearsacrosstheresistorevenintheabsenceofacurrentfromanexternalcircuit(Fig.2.20a).ThustheThveninmodelofthereal(noisy)resistoristhatshowninFig.2.20b.Clearly,thehighertheabsolutetemperatureToftheresistor,thelargervnTwillbe.Infact,itcanbeshownthatthemeansquareofvnTisgivenby

    HerekistheubiquitousBoltzmann'sconstant,and fisthebandwidthinwhichthenoiseismeasured,inhertz.(Thevalueof4kTatroomtemperatureisabout1.661020VC.)

    IfEq.(2.25)wastrueforanybandwidth,theenergyofthenoisewouldbeinfinite.Infact,however,forveryhighfrequencies( 1013Hz)otherphysicalphenomenaenter,whichcause todecreasewithincreasingfrequencysothattheoverallnoiseenergyisfinite.

    Theaveragevalue(dccomponent)ofthethermalnoiseiszero.Sinceitsspectraldensity isindependentoffrequency(atleastforlowerfrequencies),itisawhitenoise.Clearly,Fig.2.20bmayberedrawnintheformofaNortonequivalentthatisasa(noiseless)resistorRinparallelwithanoisecurrentsourceinT(Fig.2.20c).Thevalueofthelatterisgivenby

    whereG=1/R.

  • Page43

    Figure2.21Equivalentmodelsofthe

    thermalnoiseinaMOSFET.

    SincethechannelofaMOSFETinconductioncontainsfreecarriers,itissubjecttothermalnoise.Therefore.Eqs.(2.25)and(2.26)willhold,withRgivenbytheincrementalchannelresistance.Thenoisecanthenbemodeledbyacurrentsource,asshowninFig.2.21a.Ifthedeviceisinsaturation,itschanneltapersoff(Fig.2.8)andtheapproximationR 3/2gmcanbeusedinEq.(2.26).

    InmostcircuitsitisconvenienttomodeltheeffectofinTcausedbyavoltagesourceconnectedtothegateofan(otherwisenoiseless)MOSFET(Fig.2.21b).This''gatereferred"noisevoltagesourceisthengivenby

    BothinTandvnTdependthusonthedimensions,biasconditions,andtemperatureofthedevice.Asanexampleoftheirordersofmagnitude,foratransistorwithW=200m,L=10m,andCox=4.34108F/cm2(correspondingtoanoxidethicknessof800)whichisoperatedinsaturationatadraincurrentiD=200A,thegatereferrednoisevoltageatroomtemperatureisabout

    Ifthedeviceisswitchedoff,Rbecomesveryhigh,andtheequivalentnoisecircuitwillbeacurrentsourcewithavaluegivenbyEq.(2.26).Clearly, isverysmallhenceforusual(lowormoderate)externalimpedancelevels,theMOSFETcanberegardedasanoiselessopencircuitifitisturnedoff.

  • Page44

    Flicker(1/f)Noise

    InanMOStransistor,extraelectronenergystatesexistattheboundarybetweentheSiandSiO2.Thesecantrapandreleaseelectronsfromthechannel,andhenceintroducenoise[6,7].Sincetheprocessisrelativelyslow,mostofthenoiseenergywillbeatlowfrequencies.Asbefore,apossiblemodelofthisnoisephenomenonisacurrentsourceinparallelwiththechannelresistance.Thedcvalueofnoisecurrentisagainzero.ItsmeansquarevalueincreaseswithtemperatureandthedensityofthesurfacestatesitdecreaseswiththegateareaWLandthegateoxidecapacitanceperunitareaCox.Fordevicesfabricatedwitha"clean"process,thegatereferrednoisevoltageisnearlyindependentofthebiasconditionsandisgivenbytheapproximatingformula

    HereKdependsonthetemperatureandthefabricationprocessatypicalvalue[8,p.31]is31024V2F.Forthetransistordescribedintheprecedingexample,theformulagivesanoisevoltageof

    Thenoiseprocessdescribedisusuallycalledflickernoiseor(inreferencetothe1/ffactorin 1/fnoise.Astheexamplegivenillustrates,atlowfrequencies(say,below1kHz)itisusuallythedominantnoisemechanisminaMOSFET.

    Inconclusion,thechannelnoiseinaMOSFETcanbemodeledbyanequivalentnoisecurrentgenerator,asinFig.2.21a.InthesmallsignalmodelthisgeneratorwillbeinparallelwiththecurrentsourcesgmvGSandgmbvBS(Fig.2.16).Itsvaluecanbechosenastherootmeansquare(RMS)noisecurrent,whichfromEqs.(2.26)(2.28)is

    Notethatthemeansquaresofthenoisecurrentsareadded,sincethedifferentnoisemechanismsarestatisticallyindependent.Alternatively,thenoisecanberepresentedbyitsgatereferredvoltagesource(Fig.2.21b),inserieswiththegateterminal.Thevalueofthesourceisin/gm,withingivenbyEq.(2.29).

    2.8CMOSProcess

    TheCMOSprocessprovidesthemostflexibilitytothecircuitdesigner,duetotheavailabilityofcomplementaryMOSdevicesonthesamechip.TheoriginalmotivationfordevelopingtheCMOStechnologywastheneedforlowpowerandhighspeedlogicgatesfordigitalcircuits.Therequiredisolationbetweenthetwodifferentdevicetypesisaccomplishedbytheuseof"wells,"thatis,large,lowdopingleveldeepdiffusions,whichserveasthesubstratesforoneofthetwodevicetypes.Asanexample,Fig.2.22showspartofannwellCMOSchip,wherehighresistivityptypesubstrateisusedforthenchanneldevices,anddiffusednwellsforthepchanneldevices.

  • Page45

    Figure2.22Devicestructurefabricatedinahighperformance

    nwellCMOSprocess.

    Aswillbeshowninlaterchapters,aCMOScircuitcanbeoperatedwithasinglepowersupply,anditcanbeusedtorealizehighspeed,highgain,lowpoweranalogamplifierstages.Anadditionaladvantageisthatforthedevicesinthewell(inFig.2.22,thePMOStransistor),thesourcecanbeconnectedtothewell,therebyeliminatingthebodyeffect,andifthedeviceisusedinanamplifier,increasingthegainofthecircuit.This,however,resultsinalargestraycapacitancebetweenthesourceandthesubstrate,duetothelargesizeofthewelltobodyinterface.AnotherimportantadvantageoftheCMOSprocessistheavailabilityoftransmissiongatesmadeofaparallelconnectionofcomplementarytransistorsthatcanbeusedasswitches.Whensuchtransmissiongatesareused,thesignalisnolongerlimitedtoalevel,whichisathresholdvoltagebelowthatofthehighclocksignal,asisthecasewhensinglechannelswitchesareused.Inaddition,inCMOSchipsabipolartransistorcanbefashionedfromasourcediffusion,thewell,andthesubstrate.Thiscanbeusedinanemitterfollowerbufferstage(describedlater),inabandgapvoltagereferencecircuit,andsoon.

    Inadditiontotransistors,analogMOScircuitsusuallyrequireonchipcapacitors,andsometimesalsoonchipresistors.Inasilicongate"doublepoly"process,asecondlayeroflowresistivitypolysiliconisavailableforuseasaninterconnectorfortheformationofafloatinggateformemoryapplications.Thesetwolayersofpolysiliconcanalsobeusedasthetopandbottomelectrodesofamonolithiccapacitor.Figure2.22showstheconstructionofacapacitorwithtwopolysiliconelectrodes.ResistorscanbecreatedonanMOSchipusingadiffusedorimplantedlayeronthesurfaceofthesubstrate.Sincethesheetresistanceoftheseresistivelayersisrelativelylow(typically25to70 forasquarelayer),thesizeoftheresistorsobtainableonareasonablysmallareaislimitedtoabout100k .Thehigherresistivitywelldiffusionisalsoavailableasaresistor.Thisresistor,however,hasmuchhighervoltageandtemperaturecoefficientscomparedtodiffusedorimplantedones.

    Problems

    2.1.Apnjunctiondiodeisconnectedtoanexternalvoltagevintheforwarddirection(Fig.2.3).Reversingthepolarityofthevoltagereducesthecurrentbyafactor

  • Page46

    Figure2.23CircuitforProblem2.8.

    106.AssumethatthediodesatisfiesEq.(2.1)andisatroomtemperature.Whatisv?

    2.2.Forapnjunction(Fig.2.3),NA=ND=1016ions/cm3,|v|=5V,A=0.34mm2,andthemeasuredvalueofCis27pF.Howmuchisxd,thewidthofthedepletionlayer?Howmuchis i?

    2.3.UsingthedefinitionR=1/( iD/ VDS),calculatethechannelresistanceofanNMOStransistorfrom(a)Eq.(2.6),(b)Eq.(2.7),and(c)Eq.(2.9).

    2.4.ForanNMOStransistor,n=103cm2/Vs,thethicknessofthegateoxideis103(1=108cm),W=25m,andL=5m.Thethresholdvoltageis4V.CalculateiDforvS=vB=0VandvG=6V,and(a)vD=0.1V,(b)vD=2V,and(c)vD=4V.

    2.5.RepeatthecalculationsofProblem2.4ifvS=0,vB=3V,and p=0.3V.Whatconclusionscanyoudrawfromyourresultsregardingthebodyeffect?

    2.6.ForanNMOStransistor,k =2A/V2,W=30m,L=10m, p=0.3V, =1.5V,and =0.03V1.Findtheincrementalconductancesgm,gd,andgmbforvSB=0V,v0DS=5V,andi0D=10A.RepeatyourcalculationsforvSB=2V!

    2.7.AnNMOSswitchtransistorhasagatetosourcevoltagevGS>VT.Itsdrainisopencircuited.HowmuchisvDS?Why?

    2.8.InthecircuitofFig.2.23,theswitchSisopenedatt=0.(a)Isthetransistoroperatinginitslinearorsaturationregion?(b)Neglectingbodyeffectandchannellengthmodulation,findv(t)bysolvingtheappropriatedifferentialequationforthecircuit.

    2.9.InthecircuitofFig.2.24anoisevoltagevnisgeneratedduetothermaland

    Figure2.24CircuitforProblem2.9.

  • Page47

    Figure2.25CircuitforProblem2.10.

    shotnoiseeffects.ForwhatvalueofRwillthetwonoisevoltagesvnTandvnSbeequal?

    2.10.Calculatetheincrementalimpedance v/ iseenatnodeAofthecircuitsshowninFig.2.25.

    2.11.ShowthatthetransconductancegminthesaturationregionisequaltothedrainconductanceinthetrioderegionforagivendeviceandafixedVG.

    References

    1.R.S.MullerandT.I.Kamins,DeviceElectronicsforIntegratedCircuits,Wiley,NewYork,1977.

    2.A.S.Grove,PhysicsandTechnologyofSemiconductorDevices,Wiley,NewYork,1967.

    3.Y.P.TsividisandR.W.Ulmer,ACMOSvoltagereference,IEEEJ.SolidStateCircuits,SC13(6),774778(1978).

    4.C.A.Laber,C.F.Rahim,S.F.Dryer,G.T.Uehara,P.T.Kwoh,andP.R.Gray,IEEEJ.SolidStateCircuits,SC22(2),181189(1987).

    5.S.M.Sze(Ed.),VLSITechnology,McGrawHill,NewYork,1983.

    6.M.B.DasandJ.M.Moore,IEEETrans.Electron.Devices,ED21(2),247257(1974).

    7.P.R.GrayandR.G.Meyer,AnalysisandDesignofAnalogIntegratedCircuits,Wiley,NewYork,1977.

    8.P.R.GrayandD.A.Hodges,andR.W.Brodersen(Eds.),AnalogMOSIntegratedCircuits,IEEEPress,NewYork,1980.

  • Page48

    Chapter3BasicAnalogCMOSSubcircuits

    InthischaptersomeofthebasicsubcircuitscommonlyutilizedinanalogMOSintegratedcircuitsareexamined.Theseblocksincludeavarietyofbiascircuits,currentmirrors,singlestageamplifiers,sourcefollowers,anddifferentialstages.Thesesubcircuitsaretypicallycombinedtosynthesizeamorecomplexcircuitfunction.Theoperationalamplifierandcomparator,coveredinlaterchapters,areexamplesofhowsimplesubcircuitsarecombinedtoformmorecomplexfunctions.

    ThefirstpartofthischaptercoversthesubjectofthebiascircuitsinCMOStechnologyandthecurrentmirrors.Next,theCMOSgainstageisintroduced,withparticularemphasisontheuseofactivedevicesasactiveloads.Thecurrentmirrorsubcircuitcoveredasabiasingelementisutilizedasadynamicloadtoobtainveryhighvoltagegainsfromasinglestageamplifier.Thedifferentialamplifier,whichrepresentsabroadclassofcircuits,isdiscussednext.Thedifferentialamplifierisoneofthemostwidelyusedgainstages,whosebasicfunctionistoamplifythedifferencebetweentwoinputsignals.Finally,thelastpartofthechapterdealswiththesmallsignalanalysisandfrequencyresponseofCMOSamplifierstages.AgoodunderstandingofthetopicspresentedinthischapterisessentialfortheanalogCMOSdesigner,asmostdesignsstartatthesubcircuitlevelandprogressupwardtorealizeamorecomplexfunction.

    3.1BiasCircuitsinMOSTechnology

    Opampandamplifierstages,describedindetaillater,needvariousdcbiasvoltagesandcurrentsfortheiroperation.Anidealvoltageorcurrentbiasisindependentofthedcpowersupplyvoltages(VDD>0andVSS 0)andoftemperature.

    ToobtainthedcbiasvoltagesVo1,Vo2,...,Von,whereVSS

  • Page49

    Figure3.1(a)DiodeconnectedNMOStransistor(b)currentvoltage

    characteristicofdiodeconnectedtransistor.

    Von

  • Page50

    andVSS=0ischosen.SinceVGS=VDShereforbothdevices,theconditionforsaturation,

    issatisfied.Hencethecommonvalueofthedraincurrentsisgivenapproximatelyby

    HereIbiasisusuallyspecifiedthenVo1andVo2canbeselectedandEq.(3.3)usedtofindtheW/LratiosofthedevicesandthevalueRoftheresistor.

    AnundesirablefeatureofthisconfigurationisthatthebiasvoltagesandcurrentdependonthesupplyvoltagesVDDandVSS.Infact,thebiascurrentincreasesrapidlywithincreasingpowersupplyvoltage.Sincesuchbiasstringsareusedtoprovidebiasforotherdevicesinthecircuit,thedcpowerconsumptionoftheoverallcircuitbecomesheavilydependentonthesupplyvoltages.

    ACMOScircuitwith(theoretically)perfectsupplyindependenceisshowninFig.3.3.IfQ3andQ4arematchedtransistorssothat(W/L)3=(W/L)4,theyideallycarryequalcurrents.Choosing(W/L)1=(W/L)2willthenresultinVGS1=VGS2,

    Figure3.3VTreferencedsupplyindependentCMOS

    biassource.

  • Page51

    andthevoltageacrosstheresistor,IbiasR,willbeequaltoVGS0.Thisequilibriumconditionleadstotheequation

    Equation(3.4),whichisindependentofVDD,canbesolvedtoobtainIbias.Notethatinthisanalysisweneglectedtheeffectsofchannellengthmodulation(i.e.,weassumedthatIDisindependentofVDS).

    AnalternativeversionofthebiascircuitshowninFig.3.3thatusesthebaseemittervoltage(VBE)ofabipolartransistorasthereferencevoltageisshowninFig.3.4a.InaCMOSprocess,thesubstrate,well,andthesourcedrainjunctioninsidethewellcanbeusedtoformaverticalbipolartransistor.Forexample,Fig.3.4b

    Figure3.4(a)VBEbasedsupplyindependentbiascircuit(b)verticalpnpbipolartransistorinannwell

    CMOSprocess.

  • Page52

    showsaverticalpnpdevicethatisformedinannwellprocess.Thecollectorofthispnpdeviceisthepsubstratethatispermanentlyconnectedtothemostnegativevoltageonthechip.Forthebipolartransistor,thecollectorcurrentisgivenby

    whereVBEisthebaseemittervoltage,VT=kT/q,andIsisaconstantcurrent,whichisproportionaltothecrosssectionalareaoftheemitter,whichisusedtodescribethetransfercharacteristicofthetransistorintheforwardactiveregion.

    InFig.3.4a,asinFig.3.3,if(W/L)1=(W/L)2and(W/L)3=(W/L)4,equalcurrentsareforcedthroughthetwobranchesofthebiascircuitandthevoltagedropacrossresistorRequalsVBE.Thusthebiascurrentisgivenby

    CombiningEqs.(3.5)and(3.6),wehave

    ThisequationcanbesolvediterativelyforIbias.

    Bothsupplyindependentbiascircuitshaveasecondtrivialsteadystatecondition,cutoffwhenIbias=0.Topreventthebiascircuitfromsettlingtothewrongsteadystatecondition,astartupcircuitisnecessaryinallpracticalapplications.ThecircuittotherightofthedashedlineinFig.3.5functionsasastartupcircuit.IfIbias=0,Q5isoffandthevoltageatnodeAishigh,causingQ6toturnonanddrawacurrentthroughQ3,forcingthecircuittomovetoitsotherequilibriumstate.Oncethecircuitsettlesinthedesiredstate,Q5turnsonandnodeAgoeslow,turningoffQ6.Atthisstatethestartupcircuitisessentiallyoutofthepicture.

    Figure3.5Supplyindependentbiascircuitwithstartup.

  • Page53

    Anotherimportantperformanceaspectofthebiascircuitsistheirtemperaturedependence.Unfortunately,supplyindependentbiascircuitsarenotnecessarilytemperatureindependent,becausethebaseemittervoltage(VBE),andgatetosourcevoltage(VGS)arebothtemperaturedependent.IfthetemperaturecoefficientTCFisdefinedastherelativechangeofthebiascurrentperdegreeCelsiustemperaturevariation,wehave[1]

    UsingthedefinitionaboveandEq.(3.6),therelativetemperaturecoefficientoftheVBEbasedbiasgeneratorcanbecalculated:

    Sincethetemperaturecoefficientofthebaseemitterjunctionvoltageisnegative(2mV/C)whileresistorstypicallyhaveapositivetemperaturecoefficient,thetwotermsinEq.(3.9)add,resultinginanetTCFthatisquitehigh.ThetemperaturebehaviorofthethresholdbasedbiasgeneratorofFig.3.3issimilartotheVBEbasedcircuit.

    Analternativesupplyindependentbiasgeneratoristhe VBEbasedcircuitshowninFig.3.6a.Theoperationofthiscircuitisbasedonthedifferencebetweenthebaseemittervoltagesoftwotransistorsoperatedatdifferentcurrentdensities.InFig.3.6a,asinFigs.3.3and3.4,(W/L)1=(W/L)2and(W/L)3=(W/L)4.Therefore,equalcurrentsflowthroughthetwobranchesofthecircuitandVGS1=VGS2.Also,thepnptransistorM1,hasanemitterareathatismtimestheemitterareaof,M0.ThevoltageacrosstheresistorRis VBE=VBE0VBE1.FromEq.(3.5),

    VBEappearsacrossRandproducesacurrentofvalue

    Obviously,theresultingbiascurrentisindependentofthepowersupplyVDD.Thiscircuitalsohastwooperatingstates:oneatthedesiredoperatingcurrentgivenbyEq.(3.13)andtheotheratzero.Topreventthecircuitfromoperatinginthecutoffstate,astartupcircuitsimilartotheoneshowninFig.3.5isrequired.

  • Page54

    Figure3.6(a) VBEbasedsupply

    independentbiasgenerator(b)highperformance VBEbased

    supplyindependentbiasgenerator.

    ThetemperaturecoefficientofthebiascurrentcanbecalculatedfromEq.(3.13):

  • Page55

    Since VT/ Tand R/ Tarebothpositive,thetwotermsinthetemperaturecoefficientstendtocanceleachother.ThuscomparedtoVBEorthresholdbasedbiascircuits,the VBEbasedbiascircuitcanproduceamuchsmallertemperaturecoefficient.

    Onedrawbackofthe VBEbasedbiasgeneratoristhestrongdependenceofIbiasonthemismatchesbetweenQ3Q4andQ1Q2devicepairs.ThemismatchbetweenQ3Q4willresultindifferentcurrentstoflowinthetwobranchesofthecircuit.If1+ representstheratioofthetwocurrents, VBEwillbecome

    whichisequivalenttomodifyingm,theratiooftheemitterareas,by1+ .ThemismatchbetweenQ1andQ2andthecurrentdifferenceduetothemismatchofQ3andQ4willmaketheVGSvaluesofQ1andQ2different.ThisisequivalenttoadcoffsetvoltageVos= VGS,whichmodifiesEq.(3.13)to

    Assumingthatm=8, =0.01,andVT=26mVatroomtemperature, VBE=261n[8(1+0.01)]=54.3mVresults.Fora VGS=5mVoffsetvoltage,fromEq.(3.17),Ibiaswillchangeby10%.ToreducethisvariationspecialcareshouldbetakeninthelayoutofQ1Q4.Forbettergeometricalmatching,thesedevicesshoulduseacommoncentroidlayoutstrategy[2].

    ThecurrentmatchingaccuracyofthebiasgeneratorofFig.3.6aisfurtherdegradedduetothemismatchbetweenthedraintosourcevoltagesofQ3Q4andQ1Q2transistorpairs.Thecircuitcanbemadesymmetrical,andthedraintosourcevoltagedropsequalized,byaddingtransistorsQ5toQ8tothetwobranchesofthecircuit(Fig.3.6b).Theimprovedconfigurationalsousesthecascodecurrentmirrorprinciple,describedinSection3.2,toimprovethepowersupplyrejection.Ontheotherhand,theminimumpowersupplyvoltageisincreasedcomparedtothecircuitofFig.3.6a,duetotheextravoltagedropsrequiredbythetwocascodedevices.Thisbecomesamajorshortcominginadvancedsubmicronprocesstechnologies,orinlowpower/lowvoltageapplicationswherethepowersupplyvoltageislimitedto3.3V.

    3.2MOSCurrentMirrorsandCurrentSources

    Aswillbeseeninlatersections,constantcurrentsourcesandcurrentmirrorsareimportantcomponentsinMOSamplifiers.TheMOScurrentsourcesarequitesimilartothebipolarsources[1,3],wherethecurrentmirrorsworkontheprinciplethatidenticaldeviceswithequalgatetosourceanddraintosourcevoltagescarryequal

  • Page56

    Figure3.7(a)nchannelcurrentmirror(b)outputvoltagecurrentrelationship

    andusefuloutputrangeofthecurrentsource.

    draincurrents.AnNMOSrealizationofacurrentmirrorisshowninFig.3.7a.InthiscircuitQ1isforcedtocarryacurrentI1,sinceitsinputresistanceatitsshortedgatedrainterminalsislow(muchlowerthanr0),anditsgatepotentialVD1adjustsaccordingly.IfQ1andQ2areinsaturation,theirdraincurrentsI1andI2aredeterminedtoalargeextentbytheirVGSvalues.SinceVGS1=VGS2=VG,thecondition

    willthereforehold.Moreaccurately,sincedrainsaturationcurrentisgivenby

    ifthetransistorshavethesame ,k andVT,then

    ThecurrentI1isthus''mirrored"inI2.

    UsingEq.(3.19)andignoringtheeffectof ,thegatevoltageVGisgivenby

    Forthetransistorstooperateinthesaturationregion,VD VGVTmusthold.UsingEq.(3.21),therefore,thedrainsaturationvoltageis

    whereVDsatistheminimumdrainvoltagethatkeepsthetransistorsinsaturation.

  • Page57

    Figure3.8SmallsignalequivalentcircuitoftheMOScurrentmirror.

    FortheMOScurrentsourceofFig.3.7a,theoutputvoltagevouthastobegreaterthanVDsattokeepQ2inthesaturationregion.Figure3.7bshowstheoutputvoltagecurrentrelationshipofthecurrentsourceofFig.3.7a.

    Forsmallsignalanalysis,theequivalentcircuitofFig.2.14canbeusedtomodelQ1andQ2.TheresultingcircuitisshowninFig.3.8.Herer0istheincrementalresistanceofthecurrentsourceI1andv inisthetestvoltageconnectedtothedrainofQ2formeasuringtheoutputimpedanceofthecircuit.Thesmallsignaloutputimpedanceissimply

    whereEq.(3.19)wasused.

    Clearly,thecurrentsourceofFig.3.7ahasanoutputimpedancethatisnotbetter(i.e.,higher)thantheoutputimpedanceofasimpleMOStransistor,alsoinaccuratecurrentmatchingduetothevariationofthedrainvoltageofQ2withtheoutputvoltage,andfinallyareasonablywideoutputvoltagerange,whichislimitedatthelowerendbyVDsat.

    Theoutputimpedanceroutcanbeincreased,andthusthecircuitmadetoperformmorelikeanidealcurrentsource,byaddingonemoredeviceandmodifyingtheconnectionsslightly.Theresultingcircuit(Fig.3.9)istheMOSequivalentofWilson'scurrentsource[1,3].Inthiscircuit,ifI2increases,Q2causesv1tobecomelarger.Thisresultsinadropofv3whichthencounteractstheincreaseofI2.Thusanegativefeedbackloopexists,whichtriestoholdI2constant.Thesmallsignal

    Figure3.9MOSversionofWilson's

    currentsource.

  • Page58

    Figure3.10SmallsignalequivalentcircuitofWilson'scurrentsource.

    equivalentcircuitisshowninFig.3.10asimplifiedcircuitisshowninFig.3.11.Thelatterwasobtainedbycombiningr0andrd1into byreplacingtheselfcontrolledcurrentsourcegm2v2byaresistor1/gm2,andbyneglectingrd2,whichisnowparallelwiththe(usuallymuchsmaller)resistor1/gm2.SolvingforiininFig.3.11gives

    Typicalvaluesforgmarearound1mA/V,whilerdisontheorderofhundredsofkiloohms.Hence Then,fromEq.(3.24),

    Here,ontherighthandsidethevalueofthefirstfactorisaround100,whilethatofthesecondisaround1.Hencetheoutputimpedanceroutistwoordersofmagnitudelargerthanrd3.TheoutputimpedanceofthecurrentsourcedropsassoonastransistorQ3entersthelinear(triode)region.TheminimumleveloftheoutputvoltageswingoftheWilson'scurrentsourceisthuslimitedtoVmin=VGS2+VDsat3.Insummary,thecharacteristicsofWilson'scurrentsourcearethefollowing:highoutputimpedance,restrictedoutputvoltageswing,andpoorcurrentmatchingaccuracy,duetothemismatchbetweenthedraintosourcevoltagesoftransistorsQ1andQ2.

    Figure3.11Simplifiedsmallsignalequivalentcircuitof

    Wilson'scurrentsource.

  • Page59

    Figure3.12ImprovedMOSWilson's

    currentsource.

    Thecircuitcanbemadesymmetrical,andthedrainsourcevoltagedropsofQ1andQ2equalized,byaddinganothertransistor,Q4(Fig.3.12).ItcaneasilybeshownthattheoutputimpedanceoftheresultingimprovedMOSWilson'scurrentsourceisagaingivenbyEq.(3.25).Thedetailedanalysisofthecircuitislefttothereader(Problem3.6a).

    AslightlybetterversionofthecircuitofFig.3.12(oftencalledcascodecurrentsource)isshowninFig.3.13a.ItssmallsignalequivalentcircuitisshowninFig.3.13bandinasimplifiedforminFig.3.13c.ThiscircuitalsousesfeedbacktomaintainI2constant,anditalsoequalizesthedrainpotentialsofQ1andQ2,thusimprovingthecurrentmatchingpropertiesofthecurrentmirror.Itcaneasilybeshown(Problem3.6b)thatnow

    Hencethereisagaina100foldincreaseoverthesingleMOSFEToutputresistance.Inaddition,nowtheinternalimpedanceroofthecurrentsourceI1isinparallelwith1/gm1+1/gm4,alowinputimpedance,ratherthanwithrd1.Henceitsloadingeffectismuchreduced.

    ThecascodecurrentsourceissimilartotheimprovedWilson'scurrentsource:Itischaracterizedbyhighoutputimpedanceandaccuratecurrentmirroringcapability.However,acommondisadvantageofbothcircuitsisthattheminimumleveloftheoutputvoltageswingishigherthanthatofthesimplecurrentmirrorofFig.3.7.Thisreducestheavailablevoltageswingofthestage(s)drivenbythemirror.ForthecircuitofFig.3.13,theminimumvoltageswingbeforeQ3makesthetransitionfromthesaturationregiontothelinearregionis

  • Page60

    Figure3.13(a)Cascodecurrentsource(b)equivalentsmallsignalcircuitofthecascodecurrentsource(c)simplifiedsmallsignalcircuit

    ofthecascodecurrentsource.

    TheoutputvoltagecurrentplotforthecascodecurrentsourceisshowninFig.3.14.Theplotshowsthreeoperatingregions.IntheregionwherebothQ2andQ3areinsaturation(VD VT+2VDsat)theoutputimpedancehasthehighestvalue.IntheregionwhereQ2isinsaturationandQ3isthelinearregion(2VDsat VD

  • Page61

    Figure3.14Outputvoltagecurrentrelationshipforcascode

    currentsource.

    Asmentionedearlier,amajordrawbackofthecascodecurrentsourceisitslimitedoutputvoltageswing.AsEq.(3.27)shows,theminimumvoltageattheoutputofthecurrentsourcethatkeepsbothQ2andQ3insaturationisnowVT+2VDsat.Thislossofvoltageswingisespeciallyimportantwhenthecurrentsourceisusedastheloadofagainstage.Toreducethelossandincreasetheoutputswing,wecanbiasthedrainofthelowertransistorQ2attheedgeofthesaturationregion.TheresultinghighswingcascodecurrentsourceisshowninFig.3.15[4].

    Inthiscircuitasourcefollower(Q5andQ6)hasbeeninsertedbetweenthegatesofQ3andQ4inFig.3.13a,andtheW/LratioofQ3hasbeenreducedbyafactorof4.UsingthesimplifiedMOSIVequationID=k W/L(VGSVT)2forthesaturationregion,andtheW/LratiosgiveninFig.3.15,wehave

    Figure3.15Highswingcascodecurrentsource.

  • Page62

    Figure3.16Highswingimprovedcurrentsource.

    where FromFig.3.15,thenodevoltagesVA,VB,andVDS2canbefound:

    Q2isthereforebiasedattheedgeofsaturation.Thelowestleveloftheoutputvoltageswingisnowlimitedto2VDsat,whichisamajorimprovementcomparedtothatofthecircuitofFig.3.13a.Thehighswingcascodecurrentsourceexhibitsahighoutputimpedance,similartothatofthecascodecurrentsource,andanimprovedoutputvoltageswing.Thecurrentmatching,however,suffersduetothemismatchinthedraintosourcevoltagesofthemirrortransistorsQ1andQ2.ForQ1,VDS1=VT+VDsat,whileforQ2,VDS2=VDsat.

    Animprovedhighswingcascodecurrentsource,alongwithitsbiasingcircuit,isshowninFig.3.16.Herenisapositiveintegernumber[5,p.560].Onceagain,usingthesimplifiedMOSIVequationresultsin

    where(VDsat)W/ListheminimumdraintosourcevoltagerequiredtokeepdevicesQ1andQ3withacurrentIandaspectratioW/Linsaturation.FordevicesQ3andQ5wehave

  • Page63

    UsingEq.(3.31)yields

    Clearly,sinceVDS1=VDS2=(VDsat)W/L,bothQ1andQ2arebiasedattheedgeoftheirsaturationregions.Also,assumingthatVDS3=VT VDsat3=n(VDsat)W/L,Q3willbebiasedinthesaturationregion.Asaresult,thecircuitwillhaveahighoutputimpedance.Noticethattheoutputnodevouthashighswingcapability.Actually,aslongasvoutisgreaterthan(n+1)(VDsat)W/L,theoutputwillmaintainitslargeoutputimpedance.FortheimprovedcurrentmirrorthedevicesQ1andQ2haveequalVGSaswellasequalVDSvaluesandthereforegoodcurrentmatchingcapability.

    Theformulasandnumericalestimatesgivenforthecurrentsourcesaresomewhatoptimisticsincetheyneglectthebodyeffectofthefloatingdevices(transistorsQ3andQ4inFigs.3.13to3.16).Also,realMOStransistorsdonotdisplayanabrupttransitionfromthesaturationtolinearregion.Therefore,itisnecessarytobiasthedrainvoltagesofthemirrordevicesQ1andQ2slightlyabovetheidealsaturationvoltageproducedby toachievethehighoutputimpedancederivedearlier.

    3.3MOSGainStages[68]

    AsimpleNMOSgainstagewithresistiveloadisshowninFig.3.17.Q1isbiasedsothatitoperatesinitssaturationregion.ThelowfrequencysmallsignalequivalentcircuitisshowninFig.3.18.Thevoltagegainofthestageisclearly

    Inintegratedcircuitrealization,theresistorRLisundesirablesinceitoccupiesalargeareaandintroducesalargevoltagedropandhenceisusuallyreplacedbya

    Figure3.17ResistiveloadMOS

    gainstage.

  • Page64

    Figure3.18Smallsignallowfrequencyequivalentcircuitoftheresistiveloadgainstage.

    secondMOSFET.IfanNMOSenhancementmodedeviceisusedasaload,thecircuitofFig.3.19results.ThedrainandgateofQ2areshortedtoensurethatvds>vgsVT,andhencethedeviceisinsaturation.ThesmallsignalequivalentcircuitoftheloaddeviceQ2aloneisshowninFig.3.20.Herethevoltagecontrolledcurrentsourcegmvdsisacrossthevoltagevdshenceitbehavessimplyasaresistorofvalue1/gm.Similarly,sincevsb=vout,thesourcegmbvsbcorrespondstoaresistor1/|gmb|[recallthatbyEq.(2.19),gmb

  • Page65

    Figure3.20Smallsignalequivalentcircuitofthe

    enhancementloaddeviceQ2.

    Includingbodyeffect(butstillneglectingchannellengthmodulation),usingEqs.(2.12),(2.19),and(3.34),

    results.For| p|=0.3V, ,and =1,thedenominatoris1.21hencethegainisreducedfrom10to8.26.

    Inconclusion,theNMOSenhancementloadgainstageprovidesalowgain.Thisstageisnonethelessoftenusefulinwidebandamplifiers,wherealowbutpredictablegaincanbetoleratedandthegainstageexhibitsawidebandwidthduetothelowresistanceoftheload.Forhighgainapplications,however,thestageneedsalargesiliconareaandsincetheloaddevicehasahighresistance(smallW/Lratio),ithasalargedcvoltagedropacrossitwhichreducesthesignalhandlingcapabilityandhencethedynamicrangeofthestage.

    ToimprovetheperformanceandincreasethegainoftheMOSamplifiers,acurrentsourceloadcanbeused.Anyofthecurrentsourcesdescribedintheearliersectionscanserveasaload.AcommonsourceMOSgainstagethatusesanNMOSinputdeviceandthepchannelversionofthesimplecurrentsourceofFig.3.7aisshowninFig.3.21.TheoperationofthiscircuitissimilartotheresistiveloadgainstageofFig.3.17,buttheresistiveloadisreplacedbythesmallsignaloutputimpedancerd2ofthePMOScurrentsource.UsingEq.(3.33),thegainoftheCMOSamplifiesstageis

    Forgm1=0.2mA/Vandrd1=rd2=1M ,thesmallsignallowfrequencygainisAv=100.Obviously,thegainisproportionaltothetransconductanceofthe

    Figure3.21CommonsourcegainstagewithNMOSinputandpchannelcurrentsourceasactiveload.

  • Page66

    Figure3.22Enhancementloadgainstage

    withcapacitiveload.

    inputdeviceandthesmallsignaloutputresistanceofthestager0=rd1||rd2.Since foragivensizeoftheloaddevice,largevaluesofgaincanbeachievedinamoderatelysmallsiliconarea.Usingacascodecurrentsourcewithsignificantlyhigheroutputimpedancewillnotincreasethegainoftheamplifierdirectly,becausetheoutputresistanceofthestageislimitedbytheoutputimpedanceoftheinputdevice.

    ForthegainstageofFig.3.21tooperateproperly,bothinputandloaddevicesshouldoperateintheirsaturationregions.TheoutputsignalswingisthuslimitedtoVDD|VDsat2|andVDsat1onthepositiveandnegativeside,respectively.Theoutputsignalmustthereforeremainintherange

    andthetotaloutputswingis

    Forhighfrequencyapplications,allgainstagesdiscussedsofarhaveacommonshortcoming.ConsiderthecircuitofFig.3.22,whichincludesthesourceresistanceRSandthecapacitiveloadCLofthegainstage.Includingtheparasiticcapacitancesinthesmallsignalequivalentcircuit,thediagramshowninFig.3.23aisobtained.Combiningparallelconnectedelements,thecircuitofFig.3.23bresults,where

    ThenodeequationsfornodesAandBare

    whereallvoltagesareLaplacetransformedfunctions.

  • Page67

    Figure3.23(a)EquivalentcircuitoftheMOSgainstage

    (b)simplifiedequivalentcircuitoftheMOSgainstage.

    SolvingEq.(3.41)gives

    Toobtainthefrequencyresponse,smustbereplacedbyj .Formoderatefrequencies, hold.Thenagoodapproximationis

  • Page68

    Figure3.24Approximateequivalentcircuitofthe

    MOSgainstage.

    HereA0v=gm1/GLeqisthedcvalueofAv(j ),and

    Av(j )inEq.(3.43)canberecognizedasthetransferfunctionofthecircuitshowninFig.3.24.ThusthecapacitorCgd1whichisconnectedbetweentheinputandoutputterminalsofthegainstage(Fig.3.23a)behaveslikeacapacitance thehighfrequencygainwillbeseriouslyaffectedandthebandwidthconsiderablyreducedbythisphenomenon.

    TopreventtheMillereffect,thecascodegainstageofFig.3.25canbeused.HereQ2isusedtoisolatetheinputandoutputnodes.Itprovidesalowinputresistance1/gm2atitssourceandahighoneatitsdraintodriveQ3.Ignoringthebodyeffect,thelowfrequencysmallsignalequivalentcircuitisintheformshowninFig.3.26.Neglectingthesmallgdadmittances,clearly

    Hence,forlowfrequencies,

    Figure3.25Cascodegainstagewith

    enhancementload.

  • Page69

    Figure3.26Lowfrequencysmallsignalcircuitof

    cascodegainstage.

    ThegatetodraingainofQ1isgm1/gm2,andthereforetheCgd1ofthedrivertransistorQ1isnowmultipliedby(1+gm1/gm2).Choosinggm1=gm2,thisfactorisonly2.Theoverallvoltagegaingm1/gm3,however,canstillbelarge,withoutintroducingsignificantMillereffect,sincethereisnoappreciablecapacitancebetweentheinputandoutputterminals.

    Asbe