數(shù)字系統(tǒng)設計 - 驗證與測試

1講義第九章驗證與測試29.1引言系統(tǒng)可測性設計（designfortestability，DFT）DFT是整個設計過程一個非常重要的組成部分在設計流程中需要盡早考慮如果你不測試它，系統(tǒng)很可能不工作DFT策略提供必要的電路以使測試過程快速且全面提供測試激勵矢量測試過程需要的降低成本希望測試序列盡可能短但仍能覆蓋大部分可能存在的缺陷。39.2生產測試過程診斷測試

芯片和板級調試期間識別和指出失效的部位對于給定的失效部件

功能測試確定一個制造出的元件是否能工作。每一個制造出來的芯片都要經過這一測試直接的影響芯片成本測試盡可能簡單快速參數(shù)測試在各種工作條件（如溫度和電源電壓）下檢查許多非離散參數(shù)，如噪聲容限、傳播延時和最大時鐘頻率參數(shù)測試一般分為靜態(tài)和動態(tài)測試。 49.2.生產測試過程IntroductionThefabricationprocessisoneofthemostprecisemanufacturingmethodsweknowof,butitisstillnotperfect.Eachtimeachipismade,thereisafinitechancethatatleastoneofthemillionsoftransistorsorwireswillhaveanerrorinit.ProductiontestSinceusersofthechipsassumethatthechipisfunction,weneedtohavesomemethodtosortoutthefunctionchipsfromthebadones.acriticalpartofmanufacturinDesignfortest.

Giventhecomplexityoftoday’schips,designersneedtoaddfeaturestothechiptomakeproductiontestingpossible.56DesignMethodologyinDetailDesignSpecificationDesignPartitionDesignEntryBehavioralModelingSimulation/FunctionalVerificationPre-SynthesisSign-OffSynthesizeandMapGate-levelNetListPostsynthesisDesignValidationPostsynthesisTimingVerificationTestGenerationandFaultSimulationCellPlacement/ScanInsertation/RoutingVerifyPhysicalandElectricalRulesSynthesizeandMapGate-levelNetListDesignIntegrationAndVerificationDesignSign-Off7一個典型的生產測試8自動測試儀9109.3.可測性設計TestingofLogicCircuitsFaultModels故障模型

TestGenerationandCoverageFaultDetectionDesignforTest在設計過程的早期考慮測試可能簡化整個驗證過程所有可能的輸入矢量并觀察相應的響應實現(xiàn)驗證該電路的正確性11FaultModel故障模型Stuck-At固定故障ModelAssumeselectedwires(gateinputoroutput)are“stuckat”固定logicvalue0or1故障Modelscurtainkindsoffabricationflawsthatshortcircuitwirestogroundorpower,orbrokenwiresthatarefloatingWirewstuck-at-0:w/0Wirewstuck-at-1:w/1Oftenassumethereisonlyonefaultatatime—eventhoughinrealcircuitsmultiplesimultaneousfaultsarepossibleandcanmaskeachotherObviouslyaverysimplisticmodel!12FaultModelSimpleexample:Generateatestcasetodetermineifaisstuckat1Try000Ifastuckat1,expecttoseef=0,butsee1insteadw1w2w3a/1bcdf0000see1butshouldbe013FaultModelSimpleexamplew1w2w3abcdfTestw1w2w3000001010011100101110111a/0X

X

Xa/1X

X

Xb/0X

b/1X

c/0X

c/1X

d/0X

d/1XX

X

f/0X

X

X

X

Xf/1X

X

XFaultDetectedTestSet14ProblemswithFaultModel故障模型Ingeneral,n-inputcircuitsrequiremuchlessthan2ntestinputstocoverallpossiblestuck-at-faultsinthecircuitHowever,thisnumberisusuallystilltoolargeinrealcircuitsforpracticalpurposesFindingminimumtestcoverisanNP-hardproblemtoo（NP難題，non-deterministicpolynomial縮寫）15PathSensitizationWire-at-timetestingtoolaboriousBettertofocusonwiringpaths,enablingmulti-wiretestingatthesametime“Activate”apathsothatchangesinsignalpropagatingalongthepathaffectstheoutput16PathSensitizationSimpleExample:Toactivatethepath,setinputssothatw1caninfluencefE.g.,w2=1,w3=0,w4=1ANDgates:oneinputat1passestheotherinputNORgates:oneinputat0invertstheotherinputTotest:w1setto1shouldgeneratef=0ifpathokfaultsa/0,b/0,c/1causef=1

w1setto0shouldgeneratef=1ifpathokfaultsa/1,b/1,c/0causef=0Onetestcancaptureseveralfaultsatonce!w1w2bfcaw3w410117PathSensitizationGoodnews:onetestchecksforseveralfaultsNumberofpathsmuchsmallerthannumberofwiresStillanimpracticallylargenumberofpathsforlarge-scalecircuitsPathideacanbeusedto“propagate”afaulttotheoutputtoobservethefaultSetinputsandintermediatevaluessoastopassaninternalwiretotheoutputwhilesettinginputstodrivethatinternalwiretoaknownvalueIfpropagatedvalueisn’tasexpected,thenwehavefoundafaultontheisolatedwire18FaultPropagationw1w2bfcgw3w4hkw1w2fw3w4b/001111DD00D19FaultPropagationw1w2bfcgw3w4hkw1w2fDw3w4g/1110000DDD20TreeStructuredCircuitsTotestinputs

stuck-at-0atgivenANDgateSetinputsatothergatestogenerateANDoutputofzeroForceinputsatselectedgatetogenerateaoneIffis1thencircuitok,elsefaultTotestinputs

stuck-at-1atgivenANDgateDriveinputtotestto0,restofinputsdrivento1Othergatesdrivenwithinputsthatforcegatesto0Iffis0thenOK,elsefaultw1w3w4w2w3w4w1w2w3f21TreeStructuredCircuitsw1w3w4w2w3w4w1w2w3f12345678w110001110w311010100w410011000w201111010w311010100w401100111w101110001w201111010w300101011w110001110w201111010w300101011w401100111Stuck-at-0Stuck-at-1ProductTermTest111010000000Stuck-at-0022TreeStructuredCircuitsw1w3w4w2w3w4w1w2w3f12345678w110001110w311010100w410011000w201111010w311010100w401100111w101110001w201111010w300101011w110001110w201111010w300101011w401100111Stuck-at-0Stuck-at-1ProductTermTest010111110000Stuck-at-0023TreeStructuredCircuitsw1w3w4w2w3w4w1w2w3f12345678w110001110w311010100w410011000w201111010w311010100w401100111w101110001w201111010w300101011w110001110w201111010w300101011w401100111Stuck-at-0Stuck-at-1ProductTermTest000101111000Stuck-at-0024TreeStructuredCircuitsw1w3w4w2w3w4w1w2w3f12345678w110001110w311010100w410011000w201111010w311010100w401100111w101110001w201111010w300101011w110001110w201111010w300101011w401100111Stuck-at-0Stuck-at-1ProductTermTest011110110100Stuck-at-1125TreeStructuredCircuitsw1w3w4w2w3w4w1w2w3f12345678w110001110w311010100w410011000w201111010w311010100w401100111w101110001w201111010w300101011w110001110w201111010w300101011w401100111Stuck-at-0Stuck-at-1ProductTermTest011110110010Stuck-at-1126TreeStructuredCircuitsw1w3w4w2w3w4w1w2w3f12345678w110001110w311010100w410011000w201111010w311010100w401100111w101110001w201111010w300101011w110001110w201111010w300101011w401100111Stuck-at-0Stuck-at-1ProductTermTest011110110001Stuck-at-1127TreeStructuredCircuitsw1w3w4w2w3w4w1w2w3f12345678w110001110w311010100w410011000w201111010w311010100w401100111w101110001w201111010w300101011w110001110w201111010w300101011w401100111Stuck-at-0Stuck-at-1ProductTermTest101100011100Stuck-at-1128TreeStructuredCircuitsw1w3w4w2w3w4w1w2w3f12345678w110001110w311010100w410011000w201111010w311010100w401100111w101110001w201111010w300101011w110001110w201111010w300101011w401100111Stuck-at-0Stuck-at-1ProductTermTest101100011001Stuck-at-1129TreeStructuredCircuitsw1w3w4w2w3w4w1w2w3f12345678w110001110w311010100w410011000w201111010w311010100w401100111w101110001w201111010w300101011w110001110w201111010w300101011w401100111Stuck-at-0Stuck-at-1ProductTermTest110011000010Stuck-at-11Anyotherstuck-at-1casescovered?30TreeStructuredCircuitsw1w3w4w2w3w4w1w2w3f12345678w110001110w311010100w410011000w201111010w311010100w401100111w101110001w201111010w300101011w110001110w201111010w300101011w401100111Stuck-at-0Stuck-at-1ProductTermTest100101011010Stuck-at-11Anyotherstuck-at-1casescovered?Wasthatcasealreadycovered?31TreeStructuredCircuitsw1w3w4w2w3w4w1w2w3f12345678w110001110w311010100w410011000w201111010w311010100w401100111w101110001w201111010w300101011w110001110w201111010w300101011w401100111Stuck-at-0Stuck-at-1ProductTermTest000001101001Stuck-at-11Allinputsstuck-at-1’scoverednow32RandomTestingSofar:deterministictesting(確定性)電路中的同一個缺陷為許多輸入圖形所覆蓋，檢測出這樣一個缺陷只需要這些矢量中的一個，而其他矢量是冗余的。Alternative:randomtestingGeneraterandominputpatternstodistinguishbetweenthecorrectfunctionandthefaultyfunction放寬必須檢測出所有缺陷這一要求可以大大減少矢量的數(shù)目。為此一般的測試過程只要求95-99%的故障覆蓋率。NumberofTestsProbabilityFaultDetectedSmallnumberoftests

hasreasonable

probabilityoffinding

thefault33SequentialTestingDuetoembeddedstateinsideflip-flops,itisdifficulttoemploythesamemethodsaswithcombinationallogic為了測試一個狀態(tài)機中一定的故障僅僅應用正確的輸入激勵是不夠的。因為首先必須使這個被測部件處于所希望的狀態(tài)。這需要應用一系列的輸入。同時把電路響應傳送到其中的輸出上。測試一個FSM中的單個缺陷需要一系列的測試向量。Alternativeapproach:designfortestScanPathtechnique:FFinputspassthroughmultiplexerstagestoallowthemtobeusedinnormalmodeaswellasaspecialtestshiftregistermode在測試過程中把反饋回路斷開，從而把時序電路變成組合電路。自測試（self-test）34設計可測試性的重要特性設計可測試性的重要特性高的可控性。只利用輸入引線就可以使一個電路節(jié)點進入某一指定狀態(tài)如果只用一個輸入向量就可以把一個節(jié)點帶到任何狀態(tài)，那么容易控制一個具有低可控性的節(jié)點或電路需要一個很長的向量序列才能到預期的狀態(tài)。高可觀察性在輸出引線上觀察一個節(jié)點的值對于一個具有高可觀察性的節(jié)點，可以在輸出引線上直接監(jiān)測到它的值一個低可觀察性的節(jié)點則需要多個周期才能使的狀態(tài)出現(xiàn)在輸出口上。當電路的復雜性和引線數(shù)目一定時，一個可測電路應當具有較高的可觀察性。專門測試（adhoctest）同應用類型相關集合了一些可用來提高設計的可觀察性和可控性的技術額外的I/O引線為了減少可能需要的額外壓焊塊的數(shù)目，可以采用在同一個壓焊塊上分路選擇測試信號和功能信號的方法。I/O總線在正常工作期間作為數(shù)據(jù)總線，而在測試期間則用來提供測試圖形和收集響應。35DesignforTestabilityBasicideaConvertasequentialcircuitintoacombinationalcircuitControleveryflip-flopcontentObserveeveryflip-flopoutputHow?Connectflip-flopsintooneormoreshiftregistersSCANDesignforTestability(ScanDFT)OtherbenefitsDiagnosis&debugLowcosttestequipment36ScanPathTechnique掃描測試ConfigureFFsintoshiftregistermode(redpath)Scanintestpatternof0sand1sNon-stateinputscanalsobeonthescanpath(thinksynchronousMealyMachine)激勵向量通過引線ScanIn輸入邏輯模塊Runsystemforoneclockcyclein“normal”mode(blackpath)—nextstatecapturedinscanpath結果鎖存到寄存器中Returntoshiftregistermodeandshiftoutthecapturedstateandoutputs寄存器中的結果通過引線ScanOut送出電路并與期望的數(shù)據(jù)進行比較。CombinationalLogic串聯(lián)掃描方式激勵向量37ScanPathExamplew,y1,y2testvector001Scan01intoy1,y2FFszY1Y2DQQDQQ

01

01y1y2wScan-inScan-outG/S038ScanPathExamplew,y1,y2testvector001Scan01intoy1,y2FFszY1Y2DQQDQQ

01

01y1y2wScan-inScan-outG/S10039ScanPathExamplew,y1,y2testvector001Scan01intoy1,y2FFszY1Y2DQQDQQ

01

01y1y2wScan-inScan-outG/S101040ScanPathExamplew,y1,y2testvector001Scan01intoy1,y2FFsNormalw=0zY1Y2DQQDQQ

01

01y1y2wScan-inScan-outG/S1010041ScanPathExamplew,y1,y2testvector001Scan01intoy1,y2FFsNormalw=0Outputz=0,Y1=0,Y2=0zY1Y2DQQDQQ

01

01y1y2wScan-inScan-outG/S10000042ScanPathExamplew,y1,y2testvector001Scan01intoy1,y2FFsNormalw=0Outputz=0,Y1=0,Y2=0ObservezdirectlyzY1Y2DQQDQQ

01

01y1y2wScan-inScan-outG/S00000043ScanPathExamplew,y1,y2testvector001Scan01intoy1,y2FFsNormalw=0Outputz=0,Y1=0,Y2=0ObservezdirectlyScanoutY1,Y2zY1Y2DQQDQQ

01

01y1y2wScan-inScan-outG/S000044ScanPathExamplew,y1,y2testvector001Scan01intoy1,y2FFsNormalw=0Outputz=0,Y1=0,Y2=0ObservezdirectlyScanoutY1,Y2zY1Y2DQQDQQ

01

01y1y2wScan-inScan-outG/S00045ScanOperationsScan-in:testpatternshiftedinApplyTestMode=1Numberofclockcyclesappliedinthismode=numberofflip-flopsinscanchain(shiftregister)?Applyprimaryinputs?Capture:combinationallogicresponsetothetestpatterncapturedinflip-flopsApplyTestMode=0,and1ormoreclockcycles?Observeprimaryoutputs?Scan-out:capturedresponseshiftedoutApplyTestMode=1Numberofclockcyclesappliedinthismode=numberofflip-flopsIn·scanchain(shiftregister)46部分掃描并不是設計中所有的寄存器都需要掃描

流水線寄存器只是為了提高性能，并不增加電路的新狀態(tài)只需使輸入和輸出寄存器可掃描就可以了47邊界掃描設計（boundaryscan）邊界掃描

把一個板上部件的輸入-輸出引線連接成一條串聯(lián)的掃描鏈在正常工作時，邊界掃描壓焊塊pads作為正常的輸入-輸出器件在測試模式，向量可以從這些壓焊塊處掃入掃出，從而在部件的邊界上提供可控性和可觀察性?？梢岳酶鞣N控制模塊來測試各個部件以及板上的互連線。這方法的開銷是要求稍微復雜一些的輸入-輸出壓焊塊以及一個附加的片上測試控制器。目前大多數(shù)產品部件都提供邊界掃描。板級測試的邊界掃描方法48邊界掃描鏈邊界掃描鏈由邊界掃描單元串行組成邊界掃描單元能夠完成對電路節(jié)點的控制和觀察功能邊界掃描單元的結構49為了實現(xiàn)PCB測試，芯片的邊界掃描鏈要求必須在所有的管腳添加邊界掃描單元。對于輸入管腳、兩態(tài)輸出管腳、三態(tài)輸出管腳和雙向管腳，邊界掃描單元添加方式有所不同當不實現(xiàn)INTEST指令時，輸入管腳的邊界掃描單元只用一個觀察級的掃描單元即可，當需要實現(xiàn)INTEST指令時，還必須添加一個帶控制級的掃描單元。50兩態(tài)的輸出管腳必須添加帶輸出鎖存的邊界掃描單元51三態(tài)輸出管腳除了添加帶輸出鎖存的邊界掃描單元外，還必須在三態(tài)控制信號上也添加一個帶輸出鎖存的邊界掃描單元雙向邊界掃描單元由一個三態(tài)輸出掃描單元和一個只帶觀察級的輸入掃描單元組成52邊界掃描除了在管腳上添加邊界掃描單元外，為了實現(xiàn)芯片內部的電路級可測試性，還必須在內部的電路節(jié)點添加掃描單元。掃描單元一般添加在寄存器的位置，用帶掃描功能的寄存器替換設計中的所有寄存器，即可實現(xiàn)全掃描設計。全掃描資源消耗很大，同時掃描鏈過長也使得串行數(shù)據(jù)掃描的速度急劇下降，使得測試速度變慢。需要根據(jù)電路邏輯結構采用部分掃描設計。53Built-inSelf-Test(BIST)內建測試內建自測試讓電路自己生成測試圖形電路自己能夠決定它所得到的測試結果是否正確TestVectorGeneratorPseudorandom

testswithafeedbackshiftregister窮盡測試，所有可以得到的輸入信號空間，所有可測的故障都會被檢測到。對于N值較大的電路，通過整個輸入空間的操作所需要的時間是無法接受的。

SeedgeneratesasequenceoftestpatternsOutputscombinedusingthesametechniqueGeneratesauniquesignature簽名

thatcanbecheckedtodetermineifthecircuitiscorrectTestVectorGeneratorCircuitUnderTestTestResponseCompressorx0...xn-1P0...Pm-1Signature原則是應當能得到合理的故障覆蓋率54LinearFeedbackShiftRegisterDQQDQQDQQDQQDQQDQQDQQDQQPSignatureRandomTestPatternInputfrom

circuitundertest由多個一位的寄存器串聯(lián)構成有些輸出被異或（XOR）并反饋回移位寄存器的輸入

把這個寄存器初始化為一定的種子值就會產生不同的偽隨機序列。55LinearFeedbackShiftRegisterStartingwiththepattern1000種子,generates15differentpatternsinsequenceandthenrepeatsPattern0000isano-noDQQDQQDQQDQQx3x2x1x0x3x2x1x0ff10001110011110111110011111011001011101011101001100001111001001000001000001110001……InitialConfiguration56LinearFeedbackShiftRegister響應分析器將所生成的響應與存放在片上存儲器中的預期響應進行比較更經濟的技術是在對它們進行比較之前把這些響應進行壓縮

Multi-inputCompressor每一個進來的數(shù)據(jù)字被一個接一個地與LFSR的內容相比較(XOR)。在測試序列結束時，LFSR中含有了這個數(shù)據(jù)序列的簽名或特征字，可以用來與正確電路的特征字進行比較。

DQQDQQP3P2DQQP1DQQP0SignatureCircuitUnderTestOutputs改進的線性反饋移位寄存器用同一硬件來完成圖形生成和信號分析57TestCompression?BISTbenefitsretained,BISTissuesavoided?MAJORproliferationintheindustry?Exampleconfigurations:n=200,p=k=15;n=500,p=k=20;n=1,000,p=k=5058一位信號流的壓縮簽名分析器響應分析器有一個動態(tài)壓縮被測電路輸出的電路以及一個比較器構成。被壓縮的輸出也常常稱為該電路的簽名，而整個方法稱為簽名分析。這一電路只是計數(shù)輸入流中0->1和1->0的翻轉數(shù)目。這一壓縮并不能保證接收到的序列是正確的，也就是說許多不同的序列可以有相同數(shù)目的翻轉。但是翻轉數(shù)目不對，電路肯定有錯。59TestCompressionAdvantages?Exponentialreduction(50–80X,sometimesevenmore)Testdatavolume,time,pins,testerchannels?Hightestquality?XtoleranceNewX-Compactortechnique–anexampleofcombiningVLSIwithinformationtheory?CommercialATPGtoolcompatibility?Diagnosissupport?Faultmodel&testsetindependence?Verylittleareaoverhead,noextradelay?Easyautomation60CompleteSelf-TestSystemCombinationalCircuitFFsandMuxesMICSICScanoutPRBSGScaninMUXPRBSGNormalInputsRandomTest

SequencesMulti-inputCompressorRandomTest

SequencesSingle-inputCompressor61Built-inLogicBlockObserver(Bilbo)內建邏輯塊監(jiān)測器Testgenerationandcompressioninasinglecircuit!M1,M2=11:RegularmodeM1