jackyxiong：
Reliability,C&DTechnologies Usersofpowersupplyproductsdemandincreasinglyhigherlevelsofreliabilityandperformance.Althoughthesuppliersofindividualcomponentscanconfidentlyprovideimpressivelifeandreliabilitydata,thecompoundeffectonoverallreliabilitycanbesignificantwhenalargenumberofindividualcomponentsarecombinedinamodulesuchasapowersupply.Perhapsmoreimportantintermsofproductreliabilityisthequalityandrepeatabilityoftheassemblyprocess.Solderjoints,connectorsandmechanicalfixingsareallpotentialoriginsforproductfailure.Inuse,operatingtemperatureandotherenvironmentalfactorsalsoaffectthelongevityandreliabilityofapowersupply.Burn-inandvariousotherformsoflifeandstresstestinghelpprovidethedatatoenablepowersupplymanufacturerstocontinuallyimprovethereliabilityoftheirproducts.Indeed,whenanalyzedcorrectlyandfedbackintothedesignandassemblyprocess,theaccumulateddatacanbeusedtooptimizethetestandburn-inprocess.TheBurn-InProcessThepurposeoftheburn-inprocessforpowersuppliesistoweedout“infantmortalities,”asseeninthefirstportionofthewell-known“bathtubcurve”offailurerateversusoperationaltime(Fig.1).Theselatent,earlylifefailuresmaybeduetointrinsicgrossfaultswithinthebought-incomponents,assemblyerrorsorfaultsinducedincomponentsbyinappropriatehandling,e.g.ESDdamage.Itshouldbenotedthattherearenoabsolutesintheworldofreliabilitytesting,onlyprobabilityandconfidencelevelsforlargepopulations.Hence,thereisneveraguaranteethatallinfantmortalitiesarecaughtbytheburn-inprocess.Theconventionalapproachtopowersupplyburn-inovermanyyearshasinvolvedrunningthesuppliesatanelevatedtemperature,oftenthemaximum-ratedoperatingtemperaturelistedintheproductdataspecifications,wheretherateofappearanceoflatentdefectsisaccelerated.Thesuppliesarerununderfullloadwithpowercycling,andtheinputvoltageisrunateitherthemaximumorminimumvoltagetoprovideeithermaximumvoltagestressormaximumcurrentstress,dependingonthedesigntopology.Careinthechoiceofconditionsisnecessarybecausesomecomponentsinsometopologiescanseemorestressatlightloads,suchassnubbernetworksinvariable-frequencyconverters.Someingenuitycanalsobeapplied.Forexample,ifaproductisintendedtooperatenormallywithforcedair,itcouldberuninstillairatlightloadandstillachievecomparabletemperaturestresslevelsofthehottestcomponents.However,withoutthe“temperaturespreading”effectoftheforcedair,othercomponentsmightseelittlestressundertheseconditions.AtechniquesometimesusedbyC&DTechnologies,dependentontheproducttopology,istoburn-inproductsintooutputscycledbetweenshortandopencircuit.Thiscanapplyanappropriatecurrent-stresslevelwhileexercisingtheinbuiltprotectioncircuitryonshortcircuitandimposingahigh-voltagestressleveltomanycomponentsonopencircuit.Thereisamajorbenefitinthefactthatthepowerintheshort-oropen-circuitloadistheoreticallyzero,althoughpracticallytheshortmightbeaMOSFET,turnedon,dissipatingafewwatts.Thismethodalleviatestherealproblemofenergywasteinburn-inloads.However,sometypesofcomponentstressesarenotappliedwiththismethodbecausetheoverallpowersuppliedbytheunitislow,andthereforeself-heatingmaybelow.Anelevatedambienttemperaturewillcompensateforthisinpart,perhapsusingthewasteheatfromtheburn-inloads.Asmentioned,someproducttopologiesarenotsuitableforthisburn-inmethod,suchasthosethathaveapoorlydefinedorstronglyre-entrantshort-circuitcurrentcharacteristic.Thatis,ifona“hard”shortcircuittheoutputcurrentreducestomuchlessthantheratedmaximumoutputcurrent,thelevelofburn-instressmaybetoolowtobeeffective.Thedecisiononburn-inconfigurationismadejointlybetweenthedesignandreliability/qualityengineerstoensureoptimizedscreening.Dataloggingandanalysisoftheunitsundertestisimportantfordeterminingwhetherafailurehasoccurred,andifso,when.Ifallfailuresoccurinthefirstfewminutesofa48-hrburn-insequence,therewouldbegoodreasontoshortenthetimeandincreasethroughputwhilesavingenergy.C&DTechnologiestestsproductscomprehensivelybeforeandafterburn-intoensurethatanychangesinperformanceareidentified.Thisalsocanshowwhetherthereareanyintermittentproblems.Understandingandusingburn-indatatomodifyproductdesignandmanufacturingprocessescanresultinimprovedreliabilityandyield.C&DTechnologiesusesitsburn-indatatodrivethecontinuousimprovementqualityprocess.Experienceinburn-intestinghasshownthatthermalcyclingprecipitatesmoreinfantmortalitiesthanaconstantelevatedambient,althoughthesetsoffailuresdon'tcompletelyoverlap.Thermalcyclingwithadwelltimeateachthermalextremeisthereforethepreferredprocess.IncreasingthethermalrateofchangeprecipitatesmorefailuresinfewercyclesasillustratedinFig.2.Notethatwithincreasedthermalrate,differentpopulationsoffailurescanappearthataremoreorlessaffectedbythistypeofstressandtheoccurrenceofsomeresidualfailuretypesisunaffected.Eventhoughthereisequipmentavailabletoachievethermalratesofchangeofupto60°Cperminute,somemanufacturersdon'texceed40°Cperminutetopreventexcessivethermalstressthatmaycausecrackingofmultilayerceramiccapacitors(MLCCs).Intheabsenceofthermalcyclingchambers,powercyclingatanelevatedambientwithjudiciouslyselectedcycletimesapproachestheeffectivenessofthethermalcycling/dwellprocess.Caremustbetakentoensurethattheproductsarenotstressedoutsideoftheirratingsintheoften-atypicalenvironmentofburn-in.Ifoverstressed,someusefullifeofagoodproductcouldbeusedup,andatworst,hardorlatentfailurescouldactuallybeinducedinotherwisegoodproduct.AtC&DTechnologies,theburn-inprocessnormallystartswithadurationof48hr,withadecisionprocesstoreducethetimeofburn-inwhennofailuresoccurafterasetnumberofhours.Dependingontheproduct'scomplexityandtopology,adecisionismadetoreducethefutureburn-inhoursbyhalfafter200to500unitshavegonethroughtheprocesswithnofailuresoccurringinaquarterofthecurrentburn-intime.Thisprocessiscontinueduntiltheburn-intimeisreducedto2hr,whereitisheldfortheremainderofproduction.Somecontendthatburn-incanbeeliminatedwhennofailuresoccuraftermultipleproductionbuilds.However,itcouldbearguedthatthisremovestheinsuranceagainstagroupofdefectivecomponentsbeingusedand/oraprocessanomalyoccurring.Involumeproductionofpartsthatareknowntohaveasignificantinfantmortalityrate—perhapsbecauseofthedegreeofmanualassembly—aregimeofvariableburn-incanbeusedwherebyfailuresareexpected.However,whenaprecalculatedperiodoffailure-freeoperationofabatchhaselapsed,burn-inisterminated.Thisperiodisfoundfromstatisticaltables,giventheexpectedpercentageofinfantmortalities,theirknownfailurerateanddistributiontype,batchsizeandpercentageconfidencelevelrequiredthatonlyagivennumberoflatentfailuresremain.Forexample,considerabatchof10,000unitsthathistoricallyhashad10infantmortalitiesperbatchofatypefoundtohaveameantimetofailure(MTTF)of10hrattheburn-intemperature.Inthiscase,tablesinthebookElectronicComponentReliability:Fundamentals,Modelling,Evaluation,andAssurance[1]byFinnJensenshowthatafailure-freeperiodof13hrmustpasstogivea90%confidencelevelthatonlyonelatentproductfailureremains.Theperiodextendsto24hrtohavethesameconfidencelevelthatnolatentinfantmortality-typefailuresremain.Somemanufacturershavetakentheburn-inprocessfurtherafterfindingthatthetypesofburn-indescribeddonoteliminate,withinareasonabletime,allofthefailuresseentooccurintheearlylifeofapowersupply.Also,conventionalburn-indoesnotprovokeearlyfailuresthatcouldbearesultoftheshockandvibrationofshippingandhandling.Tocombatthis,amoreaggressivehighlyacceleratedstressscreen(HASS)canbeusedthatappliesmechanical,thermalandelectricalstresstypicallybeyondproductratingsbutwithindesignmargins.Accelerationfactorsofmorethan40overconventionalburn-inhavebeenclaimedforthismethod,givingcorrespondinglyshortertesttimes.Aproblemhoweveristhatthestresslevelsaresoextremethereisariskofdamaginggoodproductwithhardorlatentfailures.Inanswertothis,thehighlyacceleratedlifetest(HALT)processwasdesignedtoidentifytherealdamagelimitsinaproductbystressingtheproducttofailurewithtemperatureextremes,thermalcycling,progressivelyhigherlevelsofvibration,andthenacombinationofthermalcyclingandvibration.Duringthistesting,thedestructionlimitsofthepowersupplyareidentified.Theseoperatinglimitsarethenusedtosettheless-severeHASStestlevels.HALTalsoisusedextensivelyduringproductdevelopmenttoidentifypotentialweaknessesinthedesign.ThetestequipmentrequiredtodoHALTmusttypicallyramptemperaturebetween-55°Cto125°Cwhileapplyingsix-axislinearandrotationalrandomvibration.Thisrequiresamajorcapitalinvestmentandisoftensubcontractedtospecialisttesthouses.SomevendorssuchasC&DTechnologiesalreadyhaveinternalHALTfacilities.TheNoBurn-InModelAsdescribedearlierinthearticle,onceburn-infailureshavereducedtoacertainlevel,somemanufacturersfeelthattheprocesscanbedroppedcompletely.Thiscanbeconsideredonlyifthemanufacturingprocessisentirelypredictableandthequalityofbought-inmaterialissuchthatithasnogrosslatentintrinsicdefects.Inotherwords,thebought-incomponentsthemselvesdon'texhibitsignificantinfantmortalitiesandonlyhavetheirintrinsiclow-levellatentdefectrate.Althoughcommoditycomponentsapproachthisqualitylevelandmodernmanufacturingqualitycontrolcanminimizeprocessvariations,thereisstillarealriskthatacustomermayseesomeearlylifefailures.Thecostofthisintermsofgoodwillhastobeweighedagainstthecostofburn-in.Rememberthatcustomerswillstillseetheintrinsicfailurerateoftheproductinitsservicelife.Asmallextranumberoffailuresattributabletoinfantmortalitiesmaynotbesignificant.Forexample,oneproductfromC&DTechnologiesthatusesqualitycomponentsisbuiltusingastable,matureprocesswithoutburn-inandhasanobservedfieldmeantimebetweenfailure(MTBF)ofmorethan25millionhr.Thisfigureisderivedfrom130failuresinthetotalsalesof4.37millionpartsshippedevenlyoversixyears.Inthiscase,itisassumedthatthepartsarepoweredfor25%ofanygivenperiodandthatonly10%offailuresareactuallyreported.Whileextendedburn-intestsmaybeemployedonsmallnumbersofunitstogagewhetherallinfantmortalityfailureshavebeenidentified,atC&DTechnologies,ongoinglifetestsarerunforuptosixmonthson25to50unitsatamoderatelyelevatedtemperature.Thesetestsarenormallyonlyusedwhentherearelargequantitiesofunitsbuiltonacontinuingbasisandcangiveanestimateoftheintrinsicreliabilityofaproductinservice,thatis,MTBF.Theaccuracyofthisfiguredependsontherelativelymildfailure-rateaccelerationduringthetesthavingaknownrelationshiptothereal-lifefailurerate.TheArrheniusequationcangiveavaluefortheaccelerationfactorgivenaconstantfailurerateafterinfantmortalities.TheArrheniusequationhasitsoriginsinchemistry.Sointheory,itrequiresaknowledgeofeffective“activationenergies”forallfailuremodes.Butinpractice,theruleofthumbistodoubletheaccelerationfactorforeach10°Criseabovethereal-lifeoperatingtemperature.Asanexample,50unitsrunningforsixmonthsat70°Cwithnofailuresgives219,000operationalhr.Fromstatisticaltables,thisrepresentsafailurerateof4110failuresin109hrofoperation(FITs)witha60%confidencelevelor10,502FITswith90%confidence.Atalowertemperatureofsay40°C,ourruleofthumbforanaccelerationfactorto70°Ciseight,sothefiguresreduceto514FITsand1313FITs.FITisλ×109,andMTBFis1/λ,sothesefiguresrepresent1.95millionhror760,000hrMTBFat60%and90%confidencelevels,respectively.Itmayseemoddthatatestwithnofailuresgivesafinitefailurerate.Thisisbecauseitisassumedthatthefirstfailureisjustabouttohappen.Itshouldbeemphasizedthatrealfieldfailurerateisthemostaccuratemeasureofthereliabilityofaproduct.AcalculatedMTBFcanbecomparedwiththedemonstratedfigureobtainedthroughlifetestingtocheckforconsistency.However,thecalculationscanbemisleadingdependingonthebasefailureratesusedforcomponentsandthemethodofcalculation.ArecentsurveybyC&DTechnologiesfoundavariationofafactorofmorethan100betweenMTBFfiguresforthesamecircuitcalculatedbyseveraldifferentpowersupplymanufacturers.DifferentstandardssuchasMIL-HDBK-217FandTelcordiaSR332willgivedifferentanswers.Inaddition,theMILstandardalsogivestwodifferentmethods.Onemethodisthepartscount,whichgivesaquickbutconservativemeasure,andtheotheristhepartstressmethod,whichrequiresdetailedknowledgeoftheelectricaloperatingconditions.Thelattermethodismorerealistic.AsanexampleofapartstresscalculationaccordingtoMIL-HDBK-217F,ageneral-purposediodehasafailureratepermillionhoursgivenby:λP=λBΠTΠSΠCΠQΠEwhereλBisabasefailureratefordifferenttypesofdiodesandthePfactorsarefortemperature,electricalstress,internalconstruction,manufacturingqualityandenvironmentofuse,respectively.ForaSchottkypowerdiodeoperatingatajunctiontemperatureof80°C,withavoltagestressof75%ofitsrating,metallurgicallybondedconstruction,plasticcommercialpackagingandoperatedina“groundbenign”environment,thecalculationischangedbysubstitutionsfromthetablesinthestandardtobecome:λP=0.003×5×0.58×1×8×1=0.0696failurespermillionhours,or69.6FITs.OptimizingProcessControlTheimportantpointtonoteisthatqualityandreliabilitycannotbe“testedin”or“inspectedin.”Burn-intestingisultimatelyanotherinspectionprocess,butservesasamechanismforprocesscontrolandfeedback.Failuresinburn-inalongwithfieldfailurespromptfailureanalysisandcorrectiveactiontoensurethattheproductdesignandprocesshavebeencenteredandoptimizedtoprovidethebestproductpossibletothefield.Studieshaveshownthathigherfactoryyieldsgivehigherproductreliability,happiercustomersandlowerwarranty-returncosts.