蒙特卡洛分析ppt课件.ppt

MonteCarlosimulation forbetteryieldandperformance Atutorial 1 start Systemrequirement Statisticalanalysisincludeprocess mismatcheffects Initialdesign Designmeetsthegoal end NO YES MonteCarlosimulation forbetteryieldandperformance SomedesignmaydegradeinperformanceOveralldesignyieldcouldbeunexpectedlylow Iffabricationprocessparameteranddevicemismatcheffectonsamediearenottakenintoaccountthen Hencestatisticalanalysismustfindahighplaceindesigncycle 2 WewillperformMonteCarloanalysisonanRF frontendLNAandcomparetheresultifnostatisticalanalysisisdone WewillalsoseehowtoanalyzeyieldandscalardatainMonteCarlowiththehelpofLowpassfilterexample MonteCarlosimulation 3 MonteCarlosimulation example Linearity Inputmatching BiasN W Outputmatching Cascodearch toreducefeedbackcapacitance RF frontend LNA KnowingSystemrequirementInitialdesignbasedonrequirementlikenoise gain narroworwideband 4 MonteCarlosimulation Cadencesimulationsetup Normal Choosingaffirmaanalogartist 2 ChoosingSpectresimulator 5 ChoosingmodelcontainsallMOS reg capmodelparameters MonteCarlosimulation 1 Choosesetup modellibraries 2 Browseandchoosemodelthedirectory Cadencesimulationsetup Normal 6 Setupanalysis dc ac spetc createnetlistandrunsimulator MonteCarlosimulation 1 Chooseanalysistorun 2 Chooseoutputtoplot 3 Createnetlistandrun Cadencesimulationsetup Normal 7 Plottingresults MonteCarlosimulation 1 Choosedirectplotforanalysis 2 Clicktoviewthedesiredresult 3 Analyzewaveform Cadencesimulationsetup Normal 8 MonteCarlomodelinginCadencespectresimulator ProcessSection describesmanufacturingparameter theirstatisticalvariationandamodelfordevicethatcalculatesits width length cap res Etc accordingtoprocessparameter Design SpecificSection designeraccordingtohisneedcanspecifyMonteCarloanalysis Forexampleinacurrentmirrorcircuit matchedtransistorsareusedanddesignercangivesomecorrelationfactorbetweenthesematchedtransistor MonteCarlosimulation Cadencesimulationsetup MonteCarlo 9 TypicalModelFile ProcessSection Allparametersetstotheirnominalvalue nostatisticalvariationdefinedModel NMOS sRg iscalculatedusingnominalparametervalue 2 1 MonteCarlosimulation Cadencesimulationsetup MonteCarlo 10 Definingprocess mismatchparameterasstatisticallyassignedvalue Assessesthedevicemismatchondifferentdie whichcouldhavegonethroughsomedifferentprocessparametersduringfabrication Assessesthedevicemismatchonsamedie whichcouldhavegonethroughsomedifferentprocessparameter Variationdefinedasadistributedfunction MonteCarlosimulation ProcessSection Cadencesimulationsetup MonteCarlo 11 DesignSpecificSection Thisincludesthecircuitconnectivity tworesistors andcorrespondingcurrentsourcesthatfeedthem Definingcorrelationbetweentwodevices R1 R2 Note AlternativelythisinformationcanalsobeinsertedthroughArtistMonteCarloTool MonteCarlosimulation Cadencesimulationsetup MonteCarlo 12 MonteCarlosimulation ModelforLNAexample Note Thisisnotbasedonfoundrydatabutmodeledforillustrativepurposes Cadencesimulationsetup MonteCarlo 13 MonteCarlosimulation AfterInitialdesignthatmeetsthesystemrequirement statisticalanalysismusthavetobecarriedout Makesuretheadditionofprocessandmismatchparametersectioninmodelfile Makecertaintoincludetheparticularsection forexa Statsinspectre insimulationmodellibraryGototool MonteCarloinaffirmaanalogartist Cadencesimulationsetup MonteCarlo 14 MonteCarlosimulation Choosenoofiteration default 100 1 ChoosewhichvariationtoincludeProcess devicemismatcheffectontwodiff dieMismatch devicemismatcheffectonsamedie 2 Clickifyouwanttoseethefamilyofcurvei e curvefromeachiteration 3 Definetheexpressions signalsonwhichMonteCarloanalysiswillbeperformed Note calculatorcanalsobeusedtogettheseexpression Finallyruntheanalysis Cadencesimulationsetup MonteCarlo 15 MonteCarlosimulation Analyzingwaveform Matching Normalsimulation withoutstatisticalvariation MonteCarloSimulation withstatisticalvariation Processparameterandmismatcheffect Input OutputmatchingN W DEGRADES Overalldesignperformance noise gainetc DEGRADES S11 S22 16 MonteCarlosimulation Analyzingwaveform Matching VSWR IttellshowwellinputandoutputN Warematched MonteCarlosimulation Normalsimulation VSWR1 VSWR2 VariationsinVSWR 17 MonteCarlosimulation Analyzingwaveform Matching forwardandreversetransmissiongain Ithasdeterioratedtheperformancesignificantly asaminimumS12andmaximumS21valueisdesirable Normalsimulation MonteCarlosimulation S12 S21 18 MonteCarlosimulation Analyzingwaveform Normalsimulation MonteCarlosimulation Stability AKfvalue 1 isdesiredforanstableamplifier Kfvaluehasbecome 1 andconsequentlycreatingapotentialunstability hencealargemarginisrequiredatinitialdesignphase Stabilityfactor 19 MonteCarlosimulation Analyzingwaveform Normalsimulation MonteCarlosimulation NoisePerformance Asvisible designhasarobustnoiseperformanceatdesiredband 2 4 2 5GHz BUT Noisefigure PeriodicOutputnoise 20 MonteCarlosimulation Analyzingwaveform But LNAasanRF frontendhastoprovideenoughgainwithmaximumnoisesuppressiontomaintainanallowableSNRatdemodulator sinput Itfailstomeetthegainrequirement Gain 21 MonteCarlosimulation WewillquicklygooveranotherexampleoflowpassfilterandseehowtoanalyzescalardataandyieldthroughMonteCarlosimulation 22 MonteCarlosimulation InitialDesign Circuitdesigningaccordingtosystemrequirement Low PassFilter 23 MonteCarlosimulation 1 Runningnormalanalysis 2 Specifyingstatisticalvariationinmodelfile 3 RunningMonteCarloanalysis 1 2 3 Cadencesimulationsetup MonteCarlo 24 MonteCarlosimulation Simulationshowsdb20andphasevaluesaregreatlyaffectedbystatisticalvariationsintroducedintransistor Hencetheneedforredesigningthecircuit Analyzingresults 25 MonteCarlosimulation AnalyzingScalardata 1 Chooseresults plot Histogram 2 Chooseparameterstoplot 3 Analyzethehistogramappearedinwaveformwindow 26 MonteCarlosimulation 1 Chooseresults specificationlimits 2 Setboundsandlimits 3 ChooseResults yield simpleinanalysiswindow 4 Setsuppressionvalueforyield 5 Analyzeyield Only64 iterationspassesthespecifiedlimitsforbandwidthandymax AnalyzingYield 27 MonteCarlosimulation PLLComponents Overview Referenceclock Detectoroutput Loopfilterresponse Oscillatoroutput Clockdivideroutput Phase frequencydetectordeterminesthedifferencebetweenthephaseorfrequencyoftwosignals Theloopfilterremovesthehigh frequenciesfromthevoltage controlledoscillator VCO controllingvoltage TheVCOproducesandoutputfrequencycontrolledbyavoltage 28 MonteCarlosimulation PLLComponents NoiseSources Detectornoise VCOnoise Quantizationnoise InPLLdesignitishighlydesirabletobeabletoseetheimpactofallnoisesources whichinturnaffectstheoverallPLLperformance Duetoreferencejitter Duetovariationincontrolvoltage Duetouncertaintyinvolvedindiscretization 29 MonteCarlosimulation VCO Anoscillatorisacircuitcapableofmaintainingelectricoscillations Frequencyofoscillation 1 LC 1 2Controlledbyvoltagedependentcapacitance varactor Powerefficientsincebiascurrentissharedbetweenthetwotransconductors ComplimentaryCross CoupledLCVCO equivalent Foroperationincurrent limitedregime VO 4 Ibias Req Idealswitching VO apx Ibias Req Highfrequency 30 Causesofspectralpuritydegradation phasenoise 1 Randomnoiseinthereferenceinput thePFD loopfilterandVCO alsodividersifthePLLisafrequencysynthesizer 2 Spurioussidebands highenergysidebandswithnoharmonicrelationshiptothegeneratedoutputsignal Itissystematicinorigin Whyisspectralpurityimportant MonteCarlosimulation VCO PhaseNoise Phasenoiseproducesadjacentchannelinterference Phasenoisecandegradethesensitivityofareceiverduetoreciprocalmixing 31 MonteCarlosimulation VCO PhaseNoise Howdotheprocessandmismatchvariationaffectphasenoise wewillperformmontecarloanalysistoassessthis Step1 VaryingtheprocessparameteronlyStep2 investigatingthedevicemismatch indiffVCOonesidemismatchedtotheother inpresenceofprocessvariation Thestatisticsblockcontainsthedistributionsforparameters DistributionsspecifiedintheprocessblockaresampledonceperMonteCarlorun areappliedatglobalscope andareusedtypicallytorepresentbatch to batch process variations Distributionsspecifiedinthemismatchblockareappliedonaper subcircuitinstancebasis aresampledoncepersubcircuitinstance andareusedtypicallytorepresentdevice to device onchip mismatchfordevicesonthesamechip CadenceSpectremodeling 32 MonteCarlosimulation VCO PhaseNoise modelfile Definestatisticalblocksinthemodelfile ideallyitshouldbeprovidedfromthefoundry Processsection Mismatchsection 33 MonteCarlosimulation VCO PhaseNoise STEP 1 RunningMonteCarloforprocessvariationonly Normalsimulation MonteCarlosimulation Withappliedstatisticalvariation inmodelfile anincreaseinnoisecanbeobserved andatthisrunresultednoiseisworstandunacceptable 34 MonteCarlosimulation VCO PhaseNoise STEP 2 RunningMonteCarloformismatchin2sidesofDiff VCO Normalsimulation MonteCarlosimulation Againsimilarlookingbutnotthesameresultsappearsandnoiseatthisrunisunacceptable Note Whenthesameparameterissubjecttobothprocessandmismatchvariations thesampledprocessvaluebecomesthemeanforthemismatchrandomnumbergeneratorforthatparticularparameter 35 MonteCarlosimulation VCO PhaseNoise moreinsight Togetmoreinsightwewillvaryonlyfewparameterandcheckhowvaluesareassignedfordifferentrunaswellasthesimulationresult Definingvariationforonlytwoparametersinthemodelfile Vth tox 36 MonteCarlosimulation VCO PhaseNoise moreinsight Herebothnmos pmos transistorshavebeenassignedsameprocessvariation Ineachruntheytakeondifferentparameteraccordingtodistributiondefined Processvariationonly NM0 NM1 PM1 PM0 37 MonteCarlosimulation VCO PhaseNoise moreinsight ProcessandMismatchbothvariationtogether withcorrelationof0 2betweenthetwonmos pmos transistor Asconspicuouseachnmos pmos transistorisgettingdifferentparametervalueineachrun 38 MonteCarlosimulation VCO PhaseNoise moreinsight Processvariationonly Process mismatchvariation Asvisibleinthecaseofprocessvariationwithdevicemismatchnoisehasbeenincreased 39 MonteCarlosimulation PLLataglance InaPLLalltheseprocessvariationcandegradeit soverallperformancesignificantly Toseetheimpactofprocessvariationweprobetheoutputaftertheloopfilter MonteCarlosimulation Asclearinonecasecontrolvoltage i e loopfilteroutput isrampingrapidlycomparetootherandthuswillresultindifferentperformance 40 MonteCarlosimulation InourdesignPLLhasasettlingtimeof65us Tosimplyruntheanalysis transistorlevel forthismuchperiodmaytake2 3daysonasinglemachine Todomontecarlosimulationevenfor10runwillmakethesituationworse TospeedupMonteCarloanalyses tomakethemruninminutesasopposedtodays WeneedtoreducetheruntimeandcanutilizeParallelsimulation Suchasvariancereductiontechniquecanbeemployed 41 MonteCarlosimulation Seedno parallelsimulation Note 1 Inputhave scs extension forexa input scs 2 InspectreonecannotspecifydifferentseedfromGUI bydefaultitalwaystakesseed 1 IfMonteCarlosimulationfordifferentseedisrequiredthen Step1 Createnetlist inputfile a Eitherfromanalogartistorb Tools montecarlo simulation create input files Seed 42 MonteCarlosimulation MonteCarlosimulation Seedno parallelsimulation Step2 Editinput scs editSEED line numberyouwant Seed 43 MonteCarlosimulation MonteCarlosimulation Seedno parallelsimulation Step3 Runspectrefromcommandlinewithoptionforexample spectre envartist4 4 6 log psf spectre out formatpsfbin raw psfinput scs Hereoneshouldexecutespectrecommand orexecutablefile fromthenetlistdirectory Forexampleonewantstosimulate PLL designfromcommandlineThengotoyoursimulationdirectorycd simulation pll spectre schematic netlistandhereexecutespectrecommand Seed 44 MonteCarlosimulation MonteCarlosimulation Seedno parallelsimulation Step4 ResultscanbeplottedwitheitherfromcalculatororfromMonteCarlotool Seed 1 Seed 11 Seed 3 Fig Plotsfordifferentseedvaluesimulation Seed 45 MonteCarlosimulation Seedno parallelsimulation Anotherwayofdoingsimilarthing givingdifferentseedvalue fromGUIwouldbetostartsimulationfromdifferentrun orsaytoskipsomeinitialrunasshowninthefig Butbewareskippingtheserunscouldtakemuchlongertimeforacomplexdesign Hereitwillskipfirst10runsandsimulatefrom11 thto110 thrunfor100iterationThisisquitesimilartoassigningdifferentseedvalue AwayaroundfromGUI 46 MonteCarlosimulation Seedno parallelsimulation Runningmultipleanalysisfromonefile Thiscanbedonebydefiningmultiplemontecarloanalysisstatementintheinputshownbelow Note Foreachanalysisadifferentnametochildanalysis forexampleac dc tran andtooutputtobeassigned