IEEEJOURNALOFSOLID-STATECIRCUITS,VOL.40,NO.12,DECEMBER2005ACMOSSmartTe - PDF document

IEEEJOURNALOFSOLID-STATECIRCUITS,VOL.40,NO.12,DECEMBER2005ACMOSSmartTemperatureSensorWitha3Inaccuracyof 0.1 CFrom 55 Cto125 MichielA.P.Pertijs,Member,IEEE,KoÞA.A.Makinwa,SeniorMember,IEEE,andJohanH.Huijsing,Fellow,IEEEAsmarttemperaturesensorin0.7 mCMOSisac-curatetowithin 0.1 C(3 55 Cto125 C.ThesensorusessubstratePNPtransis-torstomeasuretemperature.Errorsresultingfromnonidealitiesinthereadoutcircuitryarereducedtothe0.01 Clevel.Thisisachievedbyusingdynamicelementmatching,achoppedcurrent-gainindependentPTATbiascircuit,andalow-offsetsecond-ordersigma-deltaADCthatcombineschoppingandcorrelateddoublesampling.Spreadofthebase-emittervoltagecharacteristicsofthesubstratePNPtransistorsiscompensatedbytrimming,basedonacalibrationatonetemperature.Ahightrimmingresolutionisob-tainedbyusingasigma-deltacurrentDACtoÞne-tunethebiascurrentofthebipolartransistors.IndexTerms— 55 to125 C).Moreimportantly,theyarerelativelyinaccuratecomparedtoconventionalsensors.Atypicalinaccuracyinthementionedtemperaturerangeis 2 C,whiletheinaccuracythatcanbeobtainedwith,forinstance,aclass-Aplatinumresistorinthatrangeis 0.5 C[6].Higheraccuracycanbeobtainedbycalibrationatmultipletemperatures,butthiswouldundomuchofthecostadvantageofCMOSsmarttemperatureMostCMOSsmarttemperaturesensorsarebasedonthetem-peraturecharacteristicsofparasiticbipolartransistors[1],[2].ManuscriptreceivedApril11,2005;revisedJuly25,2005.Thisworkwas 0.5 C(3 )[5].Inthispaper,aCMOSsmarttemperaturesensorispresentedthatachievesaninaccuracyofonly 0.1 C(3 )overthemili-taryrange[10].Allerrorscausedbythereadoutcircuitryarere-ducedtothe0.01 Clevel.Inadditiontooffsetcancellationandcurvaturecorrection,precisionbiasingtechniquesareused,anddynamicelementmatchingisusedtoreducemismatch-relatederrors.Asaresult,thespreadofthecharacteristicsofthebipolartransistorsistheonlysigniÞcanterrorsourceinthesensor.AcalibrationatasingletemperatureisthensufÞcienttodetermine and .ThesevoltagesarecombinedtoproducethePTATandreferencevoltagesmentionedabove,whichareconvertedtoadigitaltemperaturereading ananalog-to-digitalconverter(ADC).0018-9200/$20.00©2005IEEE IEEEJOURNALOFSOLID-STATECIRCUITS,VOL.40,NO.12,DECEMBER2005 Fig.1.Operatingprincipleofthetemperaturesensor. Fig.2.Temperaturedependencyofthekeyvoltagesinthesensor.Thebase-emittervoltage ofabipolartransistorinitsforward-activeregioncanbedescribedbythefollowingwell-knownlogarithmicequation: (1)where isBoltzmannsconstant, istheelectroncharge, theabsolutetemperature, thetransistorssaturationcurrent, isitscollectorcurrent,determinedbyabiascircuit[11].InthecaseofasubstratePNPtransistor,whichisbiasedviaitsemitter,theresultingcollectorcurrentisaffectedbythescurrentgain.Thiseffectwillbeignoredfornow,anddiscussedinSectionIV-D.Asaresultofthestrongtemperaturedependencyofthesat-urationcurrent ,thebase-emittervoltagehasanegativetem-peraturecoefcientofabout 2mV C(Fig.2).Theexactvaluedependsontheabsolutevalueof and .Theextrap-olatedvalueof to0Kisroughly1.2Vandisoftendenoted ,asitisrelatedtothesiliconbandgapenergy.Thisvoltageisindependentoftheabsolutevaluesof and Using(1),itcanbeeasilyshownthatthedifferenceinbase-emittervoltage betweentwobipolartransistorsoperatedata collector-currentratioisPTAT: Thisdifferenceonlydependsontheratio ,making accuratemeasureoftemperature[2].Alargerratioresultsinalarger ,butcarehastobetakentoensurethatthetwotransistorsremaininthesameoperatingregion.Inourdesign,acurrentratioof isused,resultinginatemperaturecoefcientof0.14mV AbandgapreferencecanbeusedasthereferencevoltageofanADCdesignedtodigitize .AsillustratedinFig.2,thisreferenceisgeneratedbyaddinganampliedversionof to soastoobtainatemperature-independentvoltage : Forthementionedtemperaturecoefcients,againofabout isrequired.AnADCthatconvertstheratioof and beusedtoobtainadigitaltemperaturereading wherethecoef and arechosensoastoobtainadigitaloutputindegreesCelsius[5].B.CorrectionforProcessSpread isinsensitivetoprocessspread, ontheabsolutevaluesofboththesaturationcurrentandthebiascurrent,andwillthereforespreadwithvariationsintheICprocess.Assumingthatthespreadofthesecurrentsdoesnotcantlychangetheirtemperaturedependency,theresultingspreadof canbewrittenas (5)where and aretherelativeerrorsinthecollectorandsaturationcurrent,respectively.Thisshowsthattheerrorin withrespecttoitsnominalvalueisPTAT.Providedallothererrorsinthesensorarenegligible,thisPTATerrorin canbedeterminedfromacalibrationatonetemperature.Theerrorcanthenbecorrectedbymeansoftrim-ming.Thisshouldbedoneinsuchawaythattheresultingcor-rectionisPTAT,andwillthusnullthePTATerroroverthecom-pleteoperatingtemperaturerange[2].APTATcorrectionof canbeestablishedbytrimmingthetransistorsemitterarea,itsbiascurrent,orbyaddingapro-grammablePTATvoltagetoit[5].Wechoosetotrimthebiascurrentusingasigma-deltadigital-to-analogconverter(DAC)(seeSectionIV-B)toobtainahightrimmingresolution[12].C.AccuracyRequirementsTomakethementionedtrimmingschemework,allerrorsourcesotherthanPTATspreadof havetobemadenegli-giblebydesign.Systematicerrors,whichresultinnonlinearity,canbecorrectedforbyintroducingacompensatingnonlinearityinthereadoutcircuitry,aswillbeshownbelow.Randomerrorsneedtobereducedbydesigntothe0.01 Cleveltoobtainanoverallinaccuracyof 0.1 ndwhatrandomerrorsin canbetolerated,thesen-sitivityofthesensorsoutput tochangesin hastobe PERTIJSetal.:ACMOSSMARTTEMPERATURESENSORWITHA3 INACCURACY Fig.3.(a)Simulatedtemperaturedependencyofthereferencevoltage asafunctionofthebiascurrentatroomtemperature.(b)Associatednonlinearityofthesensorsoutput (i.e.,errorwithrespecttoalinearcalculated.Thissensitivityislargestatthelowerendofthetem-peraturerange,whereitisapproximately(1/3) C/mV.Amax-imumtemperatureerror0.01 Cthereforeimpliesamaximumrandomvoltageerrorin of30 V.Similarly,themax-imumrandomvoltageerrorin canbefoundtobeabout V.Thismeansthatallcircuiterrors,suchastheinput-re-ferredoffsetofthereadoutelectronics,havetobereducedtotheselevels.D.CurvatureCorrectionSofar,ithasbeenassumedthat isalinearfunctionoftemperature.Inpractice,however, isslightlynonlinear.Themagnitudeofthisnonlinearity,whichwillbereferredtoascur-vature,dependsonthetemperaturedependencyofthesaturationcurrent,andonthatofthecollectorcurrent.ForaPTATcollectorcurrent,itcanbewrittenas[11] (6)where isthetemperatureexponentintheanalyticalexpressionforthesaturationcurrent(aparametercalled inSPICE), isareferencetemperature.Forourprocess .AswillbediscussedinSectionIV-C,weuseabiascurrentderivedfromaPTATvoltageusingapolyresistor.Ifthetemperaturedependencyofthisresistorisalsotakenintoaccount,apredom-inantlyquadraticcurvatureofabout4mVisfoundoverthemil-itaryrange.Asthiscurvatureispresentinthereferencevoltage ,itresultsinaninversequadraticnonlinearityattheoutputofthesensor.ThisisillustratedinFig.3,whichshowsthetemperaturedependencyof andtheresultingnonlinearityoftheoutputgivenby(4),bothwiththebiascurrentatroomtemperatureasaparameter.Forabiascurrentof1 isaconventionalbandgapreferencevoltage,withazerotemperaturecoefatroomtemperatureandasmallquadraticcurvature.Thecur-vaturethenleadstoaquadraticnonlinearityofalmost1 Cattheoutput. Fig.4.Blockdiagramofthetemperaturesensor.Ifthebiascurrentisincreased, becomesanovercom-bandgapreferencevoltage,withapositivetemper-aturecoefcient.Fig.3showsthattheassociatednonlinearityattheoutputthendecreases.Aminimumnonlinearityofabout Cisreachedforabiascurrentbetween4and5 Thiscanbeexplainedasfollows.Thelineartemperaturede-pendencyof givesrisetoanonlinearityattheoutput,sinceitappearsinthedenominatorin(4).Withaproperchoiceofthetemperaturecoefcientofthereference,thisnonlinearitywillcompensateforthenonlinearityduetothecurvatureof Thus,thecurvaturecanlargelybeeliminatedbyusingarefer-encevoltagewithasmallpositivetemperaturecoefcient[7],[7],E.BlockDiagramFig.4showsablockdiagramofthesensor.Abipolarcoregeneratesthevoltages and .Theseareinputtoa modulator,whichproducesabitstream ofwhichtheaveragevalueisequaltotheratioof and .Thecharge-balancingconversionappliedinthisswitched-capacitormodulatoristhetopicofSectionIII,whiledetailsofthemodulatorwillbediscussedinSectionV.Adecimationlterisusedtolterthequantizationnoisefromthebitstreamandperformtherequiredscalingtoobtaintheoutput asgivenby(4).AswillbeexplainedinSec-tionV-E,thislteralsointroducesasmallnonlinearitytocom-pensatefortheabove-mentionedresidualnonlinearityofabout Aspecialbiascircuitisusedtoprovideaccuratebiascurrentstothebipolarcore.ThiswillbediscussedinSectionIV. IEEEJOURNALOFSOLID-STATECIRCUITS,VOL.40,NO.12,DECEMBER2005 Fig.5.Blockdiagramofthemodulator.III.CA.Charge-BalancingPrincipleFig.5showsablockdiagramofthe modulator.Themod-ulatorconsistsofalooplterandaclockedcomparator.Forsimplicity,onlyarst-orderlooplterisshown.Intheactualmodulator,asecond-orderlterisused(seeSectionV).Everyclockcycle,thecomparatorproducesabitofthebitstream basedonthepolarityoftheoutput ofthelooplter.Thefeedbackisarrangedsoastodrivetheoutputoftheintegratortozero.Ifthebitstreaminagivenclockcycleiszero, integrated,while isintegratedifthebitstreamisone.Asaresultofthefeedbackinthemodulator,theaverageinputtotheintegratoriszero.Inotherwords,thechargeaddedby isbalancedbythechargeremovedby .Iftheaveragevalueofthebitstreamisdenotedas ,thischargebalancingcanbeexpressedas Solvingfor gives Thus,withasimplecharge-balancingscheme,adigitalrepre-sentationoftheratioin(4)isobtained[1].B.Switched-CapacitorImplementationIn[1]and[5],continuous-timeimplementationsoftheabove-mentionedcharge-balancingschemewerepresented.Inthiswork,wechooseforaswitched-capacitorimplementa-tion,becausethisfacilitatestheimplementationofthevariousoffsetcancellationanddynamicelementmatchingtechniquesrequiredtomeetthe 0.1 Cinaccuracytarget.Asimpliedcircuitdiagramofthefront-endofthe ulatorisshowninFig.6.Theintegratorusescorrelateddouble-samplingtoeliminatetheoperationalamplisoffsetand noise[13]andissimilartothedesignpresentedin[8].Theintegratoroperatesintwophases.Suppose,fornow,that ,andhencethe7 capacitorsarenotused.During ,whentheopampisswitchedinunity-gain,thevoltage issampledoncapacitors withrespecttotheopampsvirtualground .Inthesecondphase ,theintegratorcapacitor isswitchedinthefeedbackpath,and Fig.6.Simpliedcircuitdiagramofthefront-endofthemodulator.theinputvoltagechangesto .Asaresult,theoutputoftheintegratorchangesby Thisshowsthatslowlyvaryingsignalsattheopampsvirtualground,suchasitsoffsetand noise,areeliminated.Thechangein isproportionaltothechangein Themodulatorsinput isgeneratedbytwosubstratebipolartransistors and .Thebiascurrentsofthesetran-sistorsaredeterminedbyaswitchnetworkbasedontheinte-gratorphaseandthevalueofthebitstream .If , hastobeintegrated.Toachievethis,the isdirectedto duringphase ,whilea5timeslargercurrentisdirectedto .Asaresult, Duringphase ,thecurrentsareswapped,sothat .Therefore,thechangeintheintegratorsoutputispro-portionalto .Notethatmismatchbetween and doesnotaffecttheaccuracyofthischange. , hastobeintegrated.Duringphase , isshortedtoground,while isbiasedbythecurrent .Asaresult, .Duringphase , isshortedandthecurrent isswitchedto ,sothat .Thechangeintheintegratorsoutputisthenproportional ,sothateffectivelytheaveragebase-emittervoltageofthetwotransistorsisused.Thegain canbeimplementedintwowaysinthisinte-grator.Onewayistousealargersamplingcapacitancewhenintegrating .Alternatively, canalsobeintegratedmultipletimeseverytime .Weuseacombi-nationofthesetechniques:an8timeslargersamplingcapaci-tanceisusedfor ;moreover, isintegratedtwiceif ,while isintegratedonlyonceif .Thus,again of isrealized.IV.BIRCUITRYA.DynamicallyMatchedCurrentRatioMismatchwilllimittheaccuracyofthe1:5currentratioinFig.6,andhencethatof .Itcanbeshownthatthecur-rentratiohastobeaccurateto 0.011%tolimitthetempera-tureerrorresultingfrommismatchto 0.01 C.Suchaccurate PERTIJSetal.:ACMOSSMARTTEMPERATURESENSORWITHA3 INACCURACY Fig.7.Current-sourcecongurationusedtoobtainadynamicallymatched1:5biascurrentratioforgenerating matchingcannotbeexpectedfrompreciselayoutalone.There-fore,dynamicelementmatchingisusedtoaverageoutmis-Fig.7showshowthishasbeenimplemented[2].SixPMOScurrentsourcesprovidecopiesofthecurrentgeneratedinthebiascircuit.Thesecurrentsourcesarenominally1 Aeach.Usingasetofswitches,eachcurrentcaneitherbedirectedto orto .Oneofthemisswitchedtoonetransistor,providingtheunitcurrentinthe1:5ratio,whiletheremainingcurrentsareswitchedtotheothertransistor.Theerrorintheresultingcurrentratiodependsonthemismatchbetweentheunitcurrentsourceandtheaverageoftheothercurrentsources.Byalter-natingtheunitcurrentsourceinsuccessivecyclesofthe modulator,mismatcherrorsareaveragedout.Therequiredav-eragingisperformedbytheintegratorof modulator[14].B.CurrentTrimmingAsdiscussedinSectionII-B,thebiascurrent usedfor hastobetrimmedinordertocompensateforthespread inthenominalvalueofthetransistorssaturationcurrentandthespread ofthebiascurrentitself.Whiletheequivalenttrimmingresolutionatthesensorsoutputhastobeintheorderof0.01 C,thetemperatureerrorduetospreadcanbeseveraldegrees.Thisimpliesatrimmingrangeofabout10Conventionaltrimmingtechniquesincludeadjustmentoftheemitterareaorthebiascurrentusingswitchablebinary-scaledtransistorsorbiascurrentsources[2].Giventhelargerequiredrange,suchtechniqueswouldbecomecomplexandrequirealargechiparea.Instead,weuseacompactsigma-deltaDACtotrimthebiascurrent[12].Fig.8showshowthishasbeenimplemented.ThesamesixPMOScurrentsourcesusedforgeneratingthe1:5cur-rentratioareusedtogenerate .Thisispossiblebecause and areneverneededatthesametime.Fiveofthesesourcesareusedforcoarsetrimming,andareswitchedonoroffbasedonthedigitalinput .Thesixthsourceisusedfornetrimming,andismodulatedbythebitstream ofa modulator.Theresultingtotalcurrent isthusswitchingbackandforthbetween and timestheunitcurrentof1 A.Theinput ofthedigitalmodulatorcanbe Fig.8.Current-sourcecongurationusedtoobtainatrimmablebiascurrent forgenerating usedtoprogramtheaveragevalueofthecurrent.Therequiredaveragingtakesplaceintheintegratoroftheanalog lator.An8-bitrst-orderdigital modulatorisusedtoobtainatrimmingresolutionof4nA,whichcorrespondsto0.01 attheoutputofthesensor.Acompactimplementationofsuchamodulatorisan8-bitaccumulatorofwhichthecarrybitisusedtogeneratethebitstream[15].Thetotaltrimmingrangeof0 Aissufcienttocompensateforpracticalspreadof and Intheimplementation,currentsourceswhoseoutputsarenotusedareconnectedtoanextradiode-connectedtransistor(notshowninFig.8),soastolimittheswitchingtransientsattheirC.ChoiceofBiasCurrentTypeWhilespreadintheabsolutevalueofthebiascurrentcanbetolerated,asitcanbetrimmedout,othererrorsinthebiascurrent,suchasvariationwiththesupplyvoltageandspreadofitstemperaturedependency,shouldbeminimized.Abiascurrentisgenerallyderivedfromabiasvoltageusingaresistor.Adifferenceinbase-emittervoltageisagoodcandi-dateforthebiasvoltage,because,asdiscussedbefore,itonlydependsonacurrentratio.AnadditionaladvantageofusingsuchaPTATbiasvoltageisthattheresultingtemperaturede-pendencyofthebiascurrentreducesthecurvatureof paredto,forinstance,aconstantbiasvoltage[11].Asbiasresistorweuseahigh-resistivitypolysiliconresistor.Thetemperaturedependencyofthisresistoraffectsthecurva-tureof ,andhasbeentakenintoaccountinthedesignofthecurvaturecorrection(seeSectionII-D).Spreadofthistemper-aturedependencyresultsinanon-PTATspreadof ,whichcannotbetrimmedout.Itisthereforeimportanttochooseabiasresistorwithareproducibletemperaturedependency.Sincedataonthisreproducibilitywerenotavailablefortheresistorsofourprocess,wechosetouseahigh-resistivitypolysiliconresistor, IEEEJOURNALOFSOLID-STATECIRCUITS,VOL.40,NO.12,DECEMBER2005whichisavailableasaprocessoptionforanalogapplications,andwasthereforeexpectedtobereasonablywellcontrolled.D.Current-GainIndependentBiasCircuitAsasubstratePNPtransistorhastobebiasedviaitsemitter,thetransistorsforwardcurrent-gain affectsitscollectorcur-rent,andhencethegeneratedbase-emittervoltage: Itcanbeshownthatspreadof doesnotresultinapurelyPTATspreadof ,asaresultofthestrongtemperature-de-pendencyof .Theresultingtemperatureerrorsthereforecannotbecompletelytrimmedout.Foranominalcurrent-gain of22,asfoundinour0.7 mCMOSprocess,theresultingerroraftertrimmingisabout0.05 Cattheextremesoftheoperatingtemperaturerangefora 10%spreadof .ThiserrorwillbelargerformoremodernCMOSprocessesinwhichsmallervaluesof arefound.Fig.9showsaPTATbiascircuitthatcanbeusedtoeliminatethecurrent-gaindependencyof .TwoPNPtransistors and arebiasedata1: currentratio.Asaresult,thedif-ferencebetweentheirbase-emittervoltagesis Thefeedbackloopensuresthattheinputvoltageoftheopampiszero,sothatwecanwrite Solvingfor gives thatis,thegeneratedbiascurrentdependson .Substitutingthiscurrentin(10)showsthatthe -dependenttermscancel(assumingidealmatching),sothatthegeneratedbase-emittervoltage isindependentofthecurrentgain.Thepresentedbiascircuithasaninherentlyhighpower-supplyrejection,asallPMOStransistorsconnected havematcheddrain-sourcevoltages,includingtheoutputtransistorthatprovidesbiascurrenttothebipolarcore.Intheimplementedcircuit,cascodinghasbeenusedtofurtherimprovethepowersupplyrejection.E.OffsetCancellationOffsetsinthebiascircuitofFig.9introducespreadin thatcannotbecompletelytrimmedout.Thisappliestoboththeoffsetoftheopampandthatduetomismatchbetween and .Theseoffsetsmustbesmallcomparedto .Itcanbeshownthatforacurrent-ratio ,thecombinedoffset Fig.9.PTATbiascircuitthatmakes independentoftheforwardcurrent-gain. Fig.10.Implementationofthebiascircuitthatuseschoppingtocanceltheoffset . hastobesmallerthan 70 Vtoresultinatemperatureerrorsmallerthan 0.01 C.Thisimpliesthatsomeformofoffsetcancellationisrequired.Fig.10showstheimplementationofthebiascircuit,inwhichchoppingisusedtocanceltheoffset .Acurrentratio andabiasresistor areusedtoobtainnominalcurrentsof0.5and5 A.Thenominalbiascurrentprovidedtothebipolarcoreis1 Thebiascircuitischoppedsynchronouslywiththeintegratorofthe modulator(Fig.6).ThepositionoftheswitchesshowninFig.10correspondstophase oftheintegrator.ThecircuitcongurationisthenthesameasinFig.9,andtheoffset addsto ,resultinginabiascurrentwhichisabittoolarge.ViathetrimmingcircuitofFig.8,thiscurrentisap-pliedtotransistor .Withtheswitchesintheotherposition(whichcorrespondstophase ),thepolarityoftheoffsetisef-fectivelyreversed,resultinginabiascurrentwhichisabittoosmall.Thiscurrentisappliedtotransistor .Sincetheinte-gratoreffectivelyintegratesthesumofthebase-emittervoltages and ,theerrorduetotheoffsetislargelyeliminated.Duetothelogarithmiccharacteristicoftransistors,thecancella-tionisnotperfect,buttheresidualquadraticerrorisnegligible.Thechoppingschemerequiresanextrabiasresistorof120k comparedtotheunchoppedcircuit.Itcanbeshownthatthereisnoseverematchingrequirementbetweenthebias PERTIJSetal.:ACMOSSMARTTEMPERATURESENSORWITHA3 INACCURACY Fig.11.Circuitdiagramofthesecond-ordermodulator.resistors.Theopampisasimplefolded-cascodedesign.Itsoutputcurrentmirrorischoppedtomaintainnegativefeedback.V.SELTAA.ModulatorTopology ADC,consistingofa modulatorandadeci-lter,converts and toadigitaltemperature .Aquantizationerrorintheorderof0.01 Cisde-sired,whichcorrespondstoabout16bits(sincethefullscaleisequivalenttoabout600K;seeFig.2).Thedesiredconversiontimeis100ms. ADCisoperateddifferentlythanconventional susedforaudioorcommunicationapplications.Insuchapplications,the isrunningcontinuously,producingabit-streamwhichislteredanddecimatedtoNyquist-rateoutputdatabyadecimationlter.Thesensor,incontrast,requiresatypeofoperation.Afterpower-up,themodulatorandthedecimationlterareresetinordertobringtheminawell-de-nedstate.Themodulatorthenproducesagivennumberofbits,whichareturnedintoasingletemperaturereadingbythedeci-lter.Finally,thesensorispowereddownagaintosavepower.Suchoperationisalsoreferredtoaseration,andimpliesthatarelativelysimpledecimationlterofthesameorderasthemodulatorcanbeused[16].Whenusinga modulator,asimplecountercanbeusedasdecimationlter[1],[7].However,2 cyclesarere-quiredtoachievearesolutionof16bits.Toobtainaconversiontimeof100ms,aclockfrequencyofabout650kHzwouldberequired,whichwouldresultinanundesirablyhighpowercon-Toreducetherequiredclockfrequency,asecond-ordermod-ulatorisused.Suchamodulator,operatedasanincrementalADC,canobtainaresolutionof16bitsinonlyafewhun-dredclockcycles(dependingontheexactimplementation)[16].Thus,theclockfrequencycanbereducedtoafewkilohertz.Asingle-loopsecond-ordertopologyisusedthatissimilartothatpresentedin[5].Afullydifferentialswitched-capacitorimple-mentationisused,ratherthanthesingle-endedmixedcontin-uous-timeswitched-capacitorimplementationdescribedthere.B.Switched-CapacitorImplementationFig.11showsacircuitdiagramofthe modulator.Ignoringthechopperswitchesfornow,therstintegratorworksasdescribedinSectionIII-B:ina cycleinwhichthe , isintegratedintwointegrationcycles; , isintegratedinoneintegrationcycle(seethetimingdiagraminFig.11).Thesamplingcapacitoroftheintegratorissplitineightunitcapacitorsof5pFeach.Allofthemareusedwhenintegrating ,whileonlyoneisusedwhenintegrating .Thus,again isrealized.Mismatchbetweentheunitcapacitorslimitstheaccuracyof .Itcanbeshownthat hastobeaccurateto 0.0067%tolimitthetemperatureerrorresultingfrommismatchto 0.01 C.Assuchaccuratematchingcannotbeexpectedfrompreciselayoutalone,dynamicelementmatchingisused.Byalternatingtheunitcapacitorusedinsuccessivecyclesofthe inwhich ,mismatcherrorsareaveragedout[14].rstintegratoressentiallydeterminestheaccuracyofthemodulator.Toguaranteenegligibleerrorsinthisintegratorduenitegain,again-boostedfolded-cascodeimplementationwithaDCgainof100dBwasused[17]. IEEEJOURNALOFSOLID-STATECIRCUITS,VOL.40,NO.12,DECEMBER2005Attheendofa cycle,theoutputoftherstintegratorissampledoncapacitors (of2pF),whicharedischargedintothesecondintegratoratthebeginningofthenext cycle.Toensurestabilityofthemodulator,afeedforwardbranchfromtheinputtothesecondintegratorisused.Thesamplingcapacitors (of1pF)and ofthisbranchareswitchedwiththesametimingastheinputsamplingcapacitors.Sinceerrorsintroducedbythesecondintegratorareattenuatedbythegainoftheintegrator,nooffsetcancellation,dynamicelementmatching,orgainboostingisneededhere.Asimplefolded-cascodeopampisused.Thepolarityofoutputofthesecondintegratorisevaluatedattheendofevery cyclebyaclockedcomparator.Theresultdeterminesthevalueofthebitstreamforthenextcycle.Thecomparatorisimplementedasadynamiclatchprecededbyapreamp,whichpreventskickbacktotheoutputofthesecondintegrator.Themodulatorusesnonoverlappingclocks.Switchinginthefront-endcircuitryandupdatingofthebitstreamoutputtakeplaceinthetimegapbetweentheclockphases and sothatanyresultingchargeinjectiondoesnotresultinerrors.Clockswithdelayedfallingedges(e.g., )areusedtopre-ventsignal-dependentchargeinjection[18].C.System-LevelChoppingWhileoffsetand noiseoftherstintegratorarereducedbytheappliedcorrelateddouble-sampling,charge-injectionmismatchintheswitchesdrivenby and resultsinresidualoffset.Minimum-sizeNMOSswitchesareusedtominimizethisoffset.Nevertheless,anoffsetofafewtensof Vremains.Thisistoolarge,giventhattheerrorin hastobeintheorderof2 Tofurtherreducetheoffset,themodulatorischoppedatsystem-level.Achopperswitchattheinputandaswitchattheoutputperiodicallyreversethepolarityoftheinputsignalandthebitstream.Toavoiddisturbingtheoperationofthemod-ulatorwhenchopping,itsstateisalsoinvertedbyswappingtheintegrationcapacitorsofbothintegrators.Thechoppingisdoneataslowspeedtomakeerrorsduetocharge-injectioninthechopperswitchesnegligible.Twochoppingperiodsperconversionareused,soastomodulatetheoffsettothezeroofthedecimationlter(seeFig.12).D.TimingofaTemperatureConversionThetimingofacompletetemperatureconversionisillus-tratedinFig.12.Afterthechiphaspoweredup,theintegra-torsofthemodulatorarereset.Themodulatorthenrunsfor400cycles,producingabitstreamwhichisfedintothedecimationlter.Theresultingconversionresulthasaresolutionof0.01 AnarbitraryfragmentofthebitstreamisshowninFig.12toillustratethetimingofthedynamicelementmatchingofthe1:5currentratioandthe1:8samplingcapacitorratio,andthetimingofthe -modulatedtrimmingofthebiascurrent.Thecurrentsourceusedforgeneratingtheunitcurrentinthe1:5ratioisselectedbyacycliccounterthatcountsfrom1to Fig.12.Timingdiagramofacompletetemperatureconversion.6.Thiscounterisenabledonlyifthebitstreamis0.Similarly,theunitcapacitorusedforsampling isselectedbyacyclic1to8counter,whichisenabledonlyifthebitstreamis1[14].Thedigital modulatorthatproducesthe (Fig.8),isonlyclockedattheendof cyclesinwhich .If (indicatedbytheshadedareasinFig.12),themodulatorisfrozen.Fig.12showshowthisworksoutif hasa50%dutycycle:arepetitive0101patternappearsinsuccessive cyclesinwhich operationofthedigital ulatorpreventsquantizationnoisefrombeingmodulatedintothesignalbandduetointermodulationbetweenthetwobit-streams[12].Suchintermodulationcanoccur,becausethetwobitstreamsareeffectivelymultiplied;thisisaresultofthefact ,whichismodulatedby ,isonlyintegrated E.DecimationFilterThedecimationlterproducesatemperaturereading fromthe400bitsproducedbythemodulator.A isused,ofwhichthesymmetricaltriangularimpulseresponseisshowninFig.12.Anonsymmetricaltriangularimpulsere-sponsecouldbeusedtoobtaintherequiredresolutioninfewerclockcycles[16].Suchalter,however,wouldnotproperlyav-erageoutthemodulatedoffsetandDEMresiduals.Moreover,theuseofalterwithasymmetricalimpulseresponseensuresthattheconversionresultisarepresentationoftheaveragetem-peratureduringaconversion.AsmentionedinSectionII-D,thedecimationlterisusedtointroduceasmallnonlinearitythatcorrectsforthenonlinearitythatremainsaftercurvaturecorrection.Thisnonlinearityisonlyintheorderof0.1 C.Inthepresentimplementation,thecorrec-tionwasimplementedbyanoff-chiplookuptable.Withonlyasmallcircuitoverhead,itcanbeimplementedon-chipusingthetechniquesdescribedin[19].VI.EXPERIMENTALESULTSThesensorwasrealizedina0.7 mCMOSprocesswithlinearcapacitorsandhigh-resistivitypolyresistors.AchipmicrophotographisshowninFig.13.Thechipareais4.5mm , PERTIJSetal.:ACMOSSMARTTEMPERATURESENSORWITHA3 INACCURACY Fig.13.Chipmicrophotographofthetemperaturesensor. Fig.14.Measuredpowerspectrumofthebitstream,withthedigitaltrimmingmodulatoroperatedinfree-runningandinbitstream-controlledmode(4096bits,16averaged,Hanningwindowed).whichincludesbondpadsandsometestcircuitry.Thedeci-lteranddigitalcontrolcircuitrywereimplementedoff-chipfortestingexibility.Fig.14showsmeasuredpowerspectraofthebitstreamofthe modulator.Thesecond-ordernoiseshapingisclearlyvisible.Thegureshowstheeffectivenessofthebit-stream-controlledoperationofthedigitaltrimmingmodulator.Withafree-runningtrimmingmodulator,thenoiseoorin-creasesbyabout30dBasaresultofintermodulatedquantiza-tionnoisethatendsupinthesignalband.Todeterminethetemperatureerrorofthesensor,24samplesfromonebatchweremountedinceramicpackages,placedinanoven,andcomparedwithaplatinumthermometer.Thisther-mometerwascalibratedto20mKattheDutchMetrologyInsti-tute.Thetemperatureerrorofthe24samplesbeforetrimmingisshowninFig.15.The3 spreadovertherangeof 55 to125 Cis C.Thefactthatthespreadincreasestowardthehighendofthetemperaturerangeisconsistentwiththeas-sumptionthatitismainlyduetothePTATspreadof Fig.16showsthemeasuredtemperatureerrorofthesamesamplesaftertrimming.Thetrimmingconsistedofadjustingthecoarseandnetrimmingparameters and (seeFig.8)soastonullthetemperatureerrorat30 C.Thisreducesthe spreadatthetrimmingtemperatureto 0.03 C,whilethespreadoverthefulloperatingtemperaturerangeisreducedto 0.1 C.Thisperformancemeetstheoriginaltarget,andshowstheeffectivenessoftheappliedreadouttechniques. Fig.15.Measuredtemperatureerrorof24devicesbeforetrimming;boldlinesindicatetheaverageerrorandvalues. Fig.16.Measuredtemperatureerrorof24devicesaftertrimming;boldlinesindicatetheaverageerrorandvalues.TABLEIUMMARY Theabove-mentionedmeasurementswereperformedatasupplyvoltageof3.3V.Thesensorisfunctionalforsupplyvoltagesfrom2.5to5.5V.Overthisrange,thepower-supply IEEEJOURNALOFSOLID-STATECIRCUITS,VOL.40,NO.12,DECEMBER2005TABLEIIOMPARISONOFNACCURACY sensitivityis0.03 C/V,whichisatenfoldimprovementoverpreviouswork[5].Thiscanbeattributedtothesupply-in-sensitivebiascircuitandthefullydifferentialcircuitry.AperformancesummaryisgiveninTableI.TableIIcomparestheachievedperformancewithpreviouswork[3][5].Sincemostworkintheeldofsmarttemperaturesensorsisdoneinindustry,thespecicationsoffourleadingcommercialsensorshavealsobeenincluded[20][23].Ourmeasuredinaccuracyintherange 55 Cto125 Cisavefoldimprovementoverthehighestperformancepublishedtodate.VII.CACMOSsmarttemperaturesensorwithintegratedsecond- ADChasbeenpresented.Thesensorwasdesignedtoachieveanoverallinaccuracyof 0.1 Thedesignphilosophywastoreducealltemperatureerrorsresultingfromcircuitnonidealitiestothe 0.01 Clevel.Er-rorsduetooffsetwerereducedbyacombinationofcorrelateddouble-samplingandsystem-levelchoppinginthe lator,andchoppinginthebiascircuit.Aspecialbiascircuitwasusedtomakethesensorinsensitivetovariationsinthecurrentgainofthesubstratebipolartransistorused,andtovariationsinthesupplyvoltage.Mismatch-relatederrorswereeliminatedbymeansofdynamicelementmatching.Curvature,nally,wascorrectedforbyusingatemperature-dependentreferenceandaslightlynonlineardecimationlter.Spreadofthebase-emittervoltageofthebipolartransistorsisthentheonlyremainingsig-canterrorsource.Thiserroristrimmedoutbasedonacali-brationatonetemperature,bymeansofahigh-resolutiontrim-mingcircuitbasedona currentDAC.Therealizationin0.7 mCMOSmeetsthetargetof 0.1 overthefulltemperaturerangeof 55 Cto125 C.Thisis,todate,thehighestreportedaccuracyforthisclassoftemperature[1]A.BakkerandJ.H.Huijsing,High-AccuracyCMOSSmartTemperatureSensors.Boston,MA:KluwerAcademic,2000.[2]G.C.M.Meijer,G.Wang,andF.Fruett,TemperaturesensorsandvoltagereferencesimplementedinCMOStechnology,IEEESensors,vol.1,no.3,pp.225234,Oct.2001.[3]A.BakkerandJ.H.Huijsing,MicropowerCMOStemperaturesensorwithdigitaloutput,IEEEJ.Solid-StateCircuits,vol.31,no.7,pp.937,Jul.1996.[4]M.Tuthill,Aswitched-current,switched-capacitortemperaturesensorin0.6-mCMOS,IEEEJ.Solid-StateCircuits,vol.33,no.7,pp.1122,Jul.1998.[5]M.A.P.Pertijs,A.Niederkorn,X.Ma,B.McKillop,A.Bakker,andJ.H.Huijsing,ACMOSsmarttemperaturesensorwitha3inaccuracy Cfrom CCto120 IEEEJ.Solid-StateCircuits,vol.40,no.2,pp.454461,Feb.2005.[6]G.C.M.MeijerandA.W.vanHerwaarden,ThermalSensors.Bristol,U.K.:IOPPublishing,1994.[7]M.A.P.Pertijs,A.Bakker,andJ.H.Huijsing,Ahigh-accuracytem-peraturesensorwithsecond-ordercurvaturecorrectionanddigitalbusinterface,Proc.ISCAS,May2001,pp.368[8]C.Hagleitneretal.AgasdetectionsystemonasingleCMOSchipcomprisingcapacitive,calorimetric,andmass-sensitivemicrosensors,IEEEISSCCDig.Tech.Papers,Feb.2002,pp.430[9]G.C.M.Meijeretal.Athree-terminalintegratedtemperaturetrans-ducerwithmicrocomputerinterfacing,SensorsActuators,vol.18,pp.206,Jun.1989.[10]M.A.P.Pertijs,K.A.A.Makinwa,andJ.H.Huijsing,ACMOStemperaturesensorwithainaccuracyof Cfrom Cto IEEEISSCCDig.Tech.Papers,Feb.2005,pp.238[11]G.C.M.Meijer,Thermalsensorsbasedontransistors,SensorsActu-ators,vol.10,pp.103125,Sep.1986.[12]M.A.P.PertijsandJ.H.Huijsing,Bitstreamtrimmingofasmarttem-peraturesensor,Proc.IEEESensors,pp.904907,Oct.2004. PERTIJSetal.:ACMOSSMARTTEMPERATURESENSORWITHA3 INACCURACY[13]C.C.EnzandG.C.Temes,Circuittechniquesforreducingtheeffectsofop-ampimperfections:autozeroing,correlateddoublesampling,andchopperstabilization,Proc.IEEE,vol.84,no.11,pp.15841614,Nov.[14]M.A.P.PertijsandJ.H.Huijsing,Asigma-deltamodulatorwithbit-stream-controlleddynamicelementmatching,Proc.ESSCIRC,Sep.2004,pp.187[15]G.v.d.HornandJ.H.Huijsing,IntegratedSmartSensors:DesignandCalibration.Boston,MA:KluwerAcademic,1998.[16]J.RobertandP.Deval,Asecond-orderhigh-resolutionincrementalA/Dconverterwithoffsetandchargeinjectioncompensation,IEEEJ.Solid-StateCircuits,vol.23,no.3,pp.736741,Jun.1988.[17]K.BultandG.J.G.M.Geelen,Afast-settlingCMOSopampforSCcircuitswith90-dBDCgain,IEEEJ.Solid-StateCircuits,vol.25,no.12,pp.13791384,Dec.1990.[18]S.R.Norsworthy,R.Schreier,andG.C.Temes,Eds.,Delta-SigmaDataConverters:Theory,DesignandSimulation.NewYork:IEEEPress,[19]P.Malcovati,C.A.Leme,P.OLeary,F.Maloberti,andH.Baltes,SmartsensorinterfacewithA/Dconversionandprogrammablecali-IEEEJ.Solid-StateCircuits,vol.29,no.8,pp.963966,Aug.[20]LM92DataSheet,NationalSemiconductorCorp.(2005,Mar.).[On-line].Available:http://www.national.com[21]DS1626DataSheet,MaximIntegratedProducts.(2005,May).[Online].Available:http://www.maxim-ic.com[22]SMT160-30DataSheet,SmartecB.V.(2003,May).[Online].Available:http://www.smartec.nl[23]ADT7301DataSheet,AnalogDevicesInc.(2004,Aug.).[Online].Available:http://www.analog.com MichielA.P.Pertijs05)wasbornonMay31,1977.HereceivedtheM.Sc.degreeinelectricalengineering(cumlaude)fromDelftUniversityofTechnology,Delft,TheNetherlands,in2000.InNovember2005,heexpectstoreceivethePh.D.degreefromthesameuniversityforhisworkonhigh-accuracyCMOSsmarttemperaturesensors.SinceAugust2005,hehasbeenworkingforNa-tionalSemiconductorinDelft.From2000to2005,heworkedasaresearchassistantattheElectronicInstrumentationLaboratoryofDelftUniversityofTechnology.In2000,hewasaninternwithPhilipsSemiconductors,Sunnyvale,CA,workingonanalogcircuitdesign.From1997to1999,heworkedpart-timeforEARSB.V.,Delft,ontheproductionanddevelopmentofahandheldphotosynthesismeter.Hisresearchinterestsincludeanalogandmixed-signalinterfaceelectronicsandsmartsensors. A.A.Makinwa05)receivedtheB.Sc.andM.Sc.degreesfromObafemiAwolowoUniversity,Ile-Ife,Nigeria,theM.E.E.degreefromPhilipsInternationalInstitute,Eindhoven,TheNetherlands,andthePh.D.degreefromDelftUniversityofTechnology,Delft,TheNetherlands.From1989to1999,hewasaresearchscientistatPhilipsResearchLaboratories,wherehedesignedsensorsystemsforinteractivedisplays,andanalogfront-endsforopticalandmagneticrecordingsystems.In1999,hejoinedDelftUniversityofTechnology,whereheiscurrentlyanAssistantProfessorattheElectronicInstrumentationLaboratory.HeholdsnineU.S.patents,hasauthoredorco-authoredover20technicalpapers,andhasgiventutorialsattheEurosensorsandtheIEEESensorsconferences.Hismainresearchinterestsareinthedesignofprecisionanalogcircuitry,sigma-deltamodulatorsandsensorinterfaces.Dr.MakinwaisontheprogramcommitteesoftheIEEEInternationalSolid-StateCircuitsConference(ISSCC)andtheIEEESensorsconference.In2005,hereceivedtheVeniawardfromtheDutchTechnologyFoundation(STW). JohanH.Huijsing97)wasbornonMay21,1938.HereceivedtheM.Sc.degreeinelectricalengineeringfromtheDelftUniversityofTechnology,Delft,TheNetherlands,in1969,andthePh.D.degreefromthesameuniversityin1981forhisthesisonop-erationalampliHehasbeenanAssistantandAssociateProfessorinelectronicinstrumentationattheFacultyofElec-tricalEngineeringofDelftUniversityofTechnologysince1969,wherehebecameafullProfessorinthechairofElectronicInstrumentationin1990,andhasbeenProfessorEmeritussince2003.From1982to1983,hewasaSeniorscien-tistatPhilipsResearchLaboratories,Sunnyvale,CA.Since1983,hehasbeenaconsultantforPhilipsSemiconductors,Sunnyvale,andsince1998alsoacon-sultantforMaxim,Sunnyvale.Hisresearchworkisfocusedonthesystematicanalysisanddesignofoperationalampliers,analog-to-digitalconverters,andintegratedsmartsensors.Heisauthororco-authorofsome200scienticpa-pers,40patentsandninebooks,andco-editorof11books.Dr.HuijsingisaFellowofIEEEforcontributionstothedesignandanalysisofanalogintegratedcircuits.HewasawardedthetitleofSimonStevinMeesterforAppliedResearchbytheDutchTechnologyFoundation.Heisinitiatorandco-chairmanoftheInternationalWorkshoponAdvancesinAnalogCircuitDe-sign,whichhasbeenheldannuallysince1992inEurope.HewasamemberoftheprogramcommitteeoftheEuropeanSolid-StateCircuitsConferencefrom1992to2002.HehasbeenchairmanoftheDutchSTWPlatformonSensorTechnologyandchairmanofthebiennialNationalWorkshoponSensorTech-nologyfrom1991until2002.