1.INTRODUCTION
With the scalingof semiconductor process technology,
theperformanceofmodern VLSIchipsimproves significantly.Wehaveseen
operatingfrequenciesof integratedcircuits
reachmulti-gigahertz,resultinginmorerigorous timingrequirements.
Timing related defects originated from manufacturing
process-related problems, such as resistive opens and shorts, metal
mousebites, viavoids, etc.,willbecomemorecommon.
Consequently,delayfaultscausedby thesephysical defects,which
preventthe circuit from meeting the timing requirements, are of
growing concern in nanometer technologies.Moreover,itshouldbenoted
thatthemanufacturingprocessisbecoming
moredifficulttobecontrolledwith theincreasingcomplexity
ofmodernVLSIchips. Therefore,electrical parameters,such
assaturation current,gatecapacitance,threshold
voltage,etc.,mayvaryfromonedevicetoanother.Asaresult,thedelayof
gatesand timing-critical
pathswillhavelargevariationsandcanhardlybepredictedduring the
designstagedueto the imprecisionofverification models. Furthermore,
thecircuit timingwouldalsobeby
theapplicationenvironmentconditionssuchastemperature,
impactedsupplyvoltagenoise,etc.Hence thereisan urgentneed
toconducteffective
delaytestingforascertainingthecorrectoperationofchips atthe
ratedfrequency.Delay faults are a category of faults which cause
functional chip to fail at
specifiedclockspeedandtheobjectiveofdelayfaulttestinganddebugis
todetecttiming defects andensurethatthedesignmeets
thedesiredperformancespecifications.Although timingverification
andfunctionalsimulation during thedesign cycle ensurethatachip
meetsitsperformancespecifications,itshouldbenotedthatthesetechniquesareapplied
toamodelof
anIntegratedCircuit(IC)andnottoactualsiliconandhencecannotdetect
delayfaultswhich
arecausedbyfactorslikecrosstalkinduceddelay,excessivevoltage drop
andswing on
thesupplynets,etc.Additionally,processvariationscanhavea
significantinfluenceonachipsfailuretomeetspecifiedperformance.Process
parameter variations canresultindistributeddelayfaultsinthechip,
whichcauseminordelayfaults onmultiplegatesin agiven path
toaccumulateandresultin thepathfailing tomeet performance
specifications. Adding moredetails tomodels ofanICtoincorporate
thesefactors willcausecomputational
costsoftimingverificationmethods tobecome
prohibitive.Diminishingfeaturesizeslimittheobservabilityofchips,makingtestand
debugmoredifficultespeciallyfortimingviolations.Theuseof
on-chiptestingcircuitry
allowsforat-speedtestingessentialforaccuratedetectionoftimingviolations.
CMOS Digital Intergrated Circuits-TermworkHighPrecisionOCDM
ArchitectureThe on-chip path delay measurement techniques have been
gained many
attentionsforresearchersinrecentyears,foritcanprovideacost-effectivealternative
waytoperformdelaydefectdetectionandsilicondebuginmodernVLSI
chips.Valuable information,which
pointstheperformancelimiterandcircuitfailure,canbe obtainedby
theon-chippathdelaymeasurementtechniquewithamuchhigherconfidence.
Amrita School of Engineering, CoimbatoreDept.ofECE4
2.COMPARISON WITHRELATED WORKS
Severalarchitectureshavealreadybeenproposedfordelaytestingand
silicon debug.Oneofthosewhicharecloselyrelatedtothisworkis
themodifiedVernierDelay Line(VDL)techniqueforpath
delaymeasurement.However;theVDLbased techniques
havesomedisadvantages [1].AtypicalVDLis showninFig2.1.
Fig.2.1TypicalVernier Delay Line
Herethemaximumdelaythatcanbemeasuredislimitedton (td1-td2).
Fig.2.2ModifiedVDL
Theauthorsaremotivatedby thefactthatalthough
VDL-basedtechniquescan
providehighdelaymeasurementresolution,it,however,needslots ofstages
toachievea largemeasurementrange
[2].Asaresult,thehardwareoverheadsof thesetechniquesare
considerable.Also,due
tothemanystages,readingoutthemeasurementresultsstoredin
theVDLneedsalotof
scanclockcycles,thusthewholedelaymeasurementprocessis
time-consuming.Moreover,thedelaysoftheimportlines,whichareusedforfeedingthe
PUMintothepathdelaymeasurementunit,shouldbetakenintoaccounttoavoidtheloss
of precisionfordelaymeasurementof
circuitpaths.Inthispaper,itispresentedanon-
chippathdelaymeasurement(OCDM)circuit,inwhichthedelayrangesof
delaystages areincreasedbyafactor oftwograduallyfrom
thelasttothefirstdelaystage.Thetime differenceof
twoinputsignalsinthenextdelaystagereliesonthestoredvalueof the
currentdelay stage.Further,itisproposedadelay calibration
techniquetocalibratethe
delaydifferenceoftheimportlinesforfeedingthePUMintotheOCDMcircuit.The
maincontributions ofthis workcanbelistedinthefollowing:1)Without
decreasingthedelaymeasurement resolution,theproposedapproachcan
expanddelay measurementrangemuch easierwith
significantlylesshardwareoverhead comparedtoprevious
VDL-baseddelaymeasurementapproaches.Bykeepingasmall quantity
ofthedelaystages,theproposedapproachcanalsoachieveshortdelay
measurementtime.2) A calibration circuit and a signal transition
converter are incorporated into the proposed on-chippath
delaymeasurementarchitecture,thereby providingamoreprecise
andeffectivewayforpathdelaymeasurement.3)Foreachoftheflip-flopsexistedattheendpoints
ofthePUMs,a2-to-1multiplexeris added.Thedatainputs
ofthe2-to-1multiplexerareconnectedtothedata-inputanddata-
outputoftheflip-flop,respectively.Bythisway,thedelaydifferenceoftheimportlines,
whichareusedforconnecting thePUMintothepath
delaymeasurementunit,canbe
calibrated.Henceahighprecisionofpathdelaymeasurementis
guaranteed.
3.OCDMCIRCUIT
ThepurposeoftheproposedOCDMcircuitistoreducethenumberofdelay
stagesintheVDL,thustoachieveasignificantlylesshardwareoverheadaswellaslessdelaymeasurementtime
[3].TheproposedOCDM circuitis shownbelow.
Fig.3.1.ProposedOCDM circuit
3.1 Structureandoperation
ThebasicstructureoftheproposedOCDM
circuitisshowninFig.3.1,whichcan convertthepath delay
ofthePUMintoaseriesofdigitalvaluesthatcanbestoredin the
flip-flopsof theVDL chain.EachdelaystageconsistedintheVDL
chainisconstructed
byapositiveedgetriggeredD-typeflip-flop,fourmultiplexers,andseveralbuffers.In
theproposedOCDMcircuit,itisassumedthattheinputxisfedbytheoutputofthe
PUM,whiletheinputyisfedbytheinputofthePUM.Soalwaysyswitchesearlierthan
xdoesduringthedelaymeasurementperiod.Inordertoexplaintheoperationof
the OCDM circuit,lets
considerthecasethatboththeinputandoutputsignalsofthePUM
are risingtransitions.
Theupperdelayunit(UDU)refers tothebuffer chainthatstarts
attheinputofthe
delaystageatwhichthetransitionsignalispropagatedfromnodey,andendsatthe
inputofthemultiplexerwhoseoutputisconnectedtothedatainputoftheflip-flopin
eachdelaystageof
theOCDMcircuit.Thelowerdelayunit(LDU)issimilartoUDU, exceptthat it
startsatthe input ofthedelaystageat whichthetransitionsignalis
propagatedfromnodex andendsattheinputof
anothermultiplexerwhoseoutputis
connectedtotheclockinputoftheflip-flopineachdelaystage.Thedelayrangeis
definedasthedelaydifferencebetweenthetwodelayunitsineachdelaystageof
the OCDM circuit.FromthelaststagetothefirststageoftheOCDM
circuit,thedelayrange
ofeachstageisincreasedbyafactoroftwo.Supposetheinputandoutputsignalsof
theupperdelaychaininthefirstdelay stage arey1
andy2,respectively,accordingly,x1andx2areassumedforthelowerdelay
chain. All flip-flops ofthe OCDM circuit are initialized to logic
ZERO values by
assertingtheresetsignal.Thedelaymeasurementmodeisactivatedbyassertingthe
mode signal.Asx1andy1 signals
propagatethroughtheirrespectivedelayunits,thetime
differencebetweenthetwo signals willbereduced.
Fig.3.2.Relationshipbetweentimedifferenceoftwoinputsignals
andthatoftwo outputsignals inthefirstdelaystage.(a)LogicONE
intheflip-710flop. (b) Logic ZEROinthe flip-flop.
As showninFig.3.2(a),assumingthatx1
signallagsthey1signalbyenoughtime
(i.e.,y1switchesmuchearlier),andhencealogic-highvalue
willbeholdintheflip-flop.
Asaresult,y2willbethesignalthatpassesthroughtheUDUandthebufferBUFBin
theupperdelay chainfromy1,whilex2willbethesignal thatpassesthrough
theLDU andthebufferBUFBinthelowerdelay chainfromx1.Thedelay
ofBUFBislarge
enoughtoensurethatastablelogichighvaluecanbestoredintheflip-flopbeforethe
twotransitionsignalsarriveattheinputs ofthemultiplexers
whoseoutputs areconnected
totheinputsofthenextdelaystage.Clearly,thetimedifferencebetweenandisreduced
byanamountwhichequals thedelayrangeofthis
delaystage.ThebuffernamedBUFAineachdelay
stagehasadelayvaluethatislargerthan thecumulativedelayof
thepaththatcontainsLDUandBUFBof thesamedelaystage.
Likewise,ifthetimedifferenceofy1andx1issmallerthan
thedelayrange,theflip-flop
willholdlogicZEROvalue.Therefore,thesignalspropagatingthroughBUFA
are then selectedby
themultiplexers.Asaresult,thetimedifferencebetweeny2andx2willbe
equaltothatofy1andx1,as
showninFig.3.2(b).Consequently,theprincipleof theOCDMcircuitis
thatifthetimedifference
betweenthetwoinputsofeachdelaystageislargerthanthedelayrangeofthesame
stage,alogicONEvaluewillbestoredintheflip-flopof
thedelaystage.Thetime differencebetweenthetwooutputsignals
willbeupdatedbysimply subtractingthedelay
rangefromthatbetweentheinputsignalsofthedelaystage.Otherwise,theflip-flopof
thedelaystage willholdalogicZEROvalue,andthetimedifferencebetween
theoutput signals willkeepthesameas thatbetweentheinputsignals
ofthedelaystage.Notethatthereexistsasetuptimeinthestoreblockof
eachdelaystageasshown inFig.3.1,whichconsistsof
aDflip-flopandtwomultiplexers.If thetimedifference
betweenthetwoinputsof
thestoreblockissmallerthanthesetuptime,anerrorlogic value may be
hold in the flip-flop. Therefore, inorder to provide better delay
measurementprecision,the DCUnitcellconstructedby
twobufferlinesisproposedfor
delaycompensation,whichmeansthattheupperandlowerdelayunitsof
DCUnitare designedsuch thatthedelay differencebetween
themisapproximatelyequalto thesetup time.
Thevaluesstoredinthedelaylinecanbeshiftedoutseriallyusingtheclock
signalshiftclkintheshiftmodebyde-assertingthemodesignal.Thedelay
ofthePUM canthenbeobtained.
3.2DelayRangeCalibration
Thedelayrangeof eachdelaystageintheOCDMcircuitwouldbevarieddueto
theprominentprocessvariations.Itisthusnecessarytocalibratethedelayrangesto
assuretheprecisionofpathdelaymeasurementresultbeforeusingtheOCDMcircuit.
Fig.3.3shows thebasicstructureof
thecalibrationcircuit,whichcanbeembeddedinto theOCDMcircuit
fordelayrangecalibration.Theoutputsofthecalibrationcircuit,
denotedasy
andx,aredirectlyconnectedtotheinputswiththesamenotationsofthe
OCDMcircuit,respectively.Twoinputs
ofthecalibrationcircuit,denotedaspinandpout,
arefedbytheinputandoutputof
thePUM,respectively.Clearly,whentheCSsignalis
setto1,thegeneratedtransitionsatthey
andxcanbesentintotheOCDMcircuitfor delaymeasurement.When
theCSsignalissetto0,thedelay range calibration processis
conducted.
Fig.3.3. Calibrationcircuit
Fig.3.4.Timingwaveformfor calibrationcircuit
Thesimplifiedtimingwaveformforthedelay rangecalibrationisshownin
the Fig.3.4.TheFF1andFF2arerisingandfallingedgetriggeredflip-flops
respectively. First,theflip-flopsof
FF1andFF2areinitiatedwithlogicZERObytheresetsignal.
Then,logicONEwillbeloadedintotheFF1andFF2bytherisingandfallingedgesof
theclocksignal respectively.Clearly,thetimedifferencebetween
thegeneratedrising
transitionsatyandxisequaltothewidthofthepositivehalfcycleoftheclocksignal.
Thetimeperiodofthe clocksignalcanbe programmeddeterministicallywith
high resolutionusingtheon-chipclockgenerator.Assumingthenumberof
delaystagesintheOCDMcircuitism.Thefollowing
presentsthecalculationmethodforobtainingthedelayrangeofeachdelaystage,which
is describedby Eq.3.1:a11x1+a12x2++a1mxm =b1
a21x1+a22x2++a2mxm =b2
..
am1x1+am2x2+..+ammxm =bm ..(3.1)
Let
Wehave
AX=B(3.2)
wherexj(1