Material Removal Mechanism in Chemical Mechanical

112IEEETRANSACTIONSONSEMICONDUCTORMANUFACTURING,VOL.14,NO.2,MAY2001MaterialRemovalMechanisminChemicalMechanicalPolishing:TheoryandModelingJianfengLuoandDavidA.DornfeldAbstract—Theabrasionmechanisminsolid-solidcontactmodeofthechemicalmechanicalpolishing(CMP)processisinvestigatedindetail.Basedonassumptionsofplasticcontactoverwafer-abra-siveandpad-abrasiveinterfaces,thenormaldistributionofabra-sivesizeandanassumedperiodicroughnessofpadsurface,anovelmodelisdevelopedformaterialremovalinCMP.Thebasicmodelis=removed,whereisthedensityofwafer,thenumberofactiveabrasives,andremovedthevolumeofmaterialremovedbyasingleabrasive.Themodelproposedin-tegratesprocessparametersincludingpressureandvelocityandotherimportantinputparametersincludingthewaferhardness,padhardness,padroughness,abrasivesize,andabrasivegeom-etryintothesameformulationtopredictthematerialremovalrate().Aninterfacebetweenthechemicaleffectandmechan-icaleffecthasbeenconstructedthroughafittingparameter,a“dynamical”hardnessvalueofthewafersurface,inthemodel.Itreflectstheinfluencesofchemicalsonthemechanicalmaterialremoval.Thefluideffectinthecurrentmodelisattributedtothenumberofactiveabrasives.Itisfoundthatthenonlineardownpressuredependenceofmaterialremovalrateisrelatedtoaprob-abilitydensityfunctionoftheabrasivesizeandtheelasticdefor-mationofthepad.Comparedwithexperimentalresults,themodelaccuratelypredicts.Withfurtherverificationofthemodel,abetterunderstandingofthefundamentalmechanisminvolvedinmaterialremovalintheCMPprocess,particularlydifferentrolesplayedbytheconsumablesandtheirinteractions,canbeobtained.IndexTerms—CMP,materialremovalmechanism,modeling.I.INTRODUCTIONTHEchemicalmechanicalpolishing(CMP)processhasbeenwidelyacceptedinthesemiconductorindustryforoxidedielectricandmetallayerplanarization[1].Itisusedtoinsurethattheinterconnectsbetweenmultilayerchipsareachievedreliablyandthatthethicknessofdielectricmaterialsisuniformandsufficient.DuringCMPprocesses,awaferisrotatedaboutitsaxiswhilebeingpressedface-downbyacarrierandacarrierfilmagainstarotatingpolishingpadcoveredwithcolloidalsilicaslurrywithspecificchemicalproperties(Fig.1).Theabrasiveslurry,composedof,forexample,50–70-nmfusedsilicainanaqueoussolutionwithbetween8.5–11,playsanimportantroleinthematerialremovalmechanisminCMP.MicroscopicobservationsofpolishedsurfaceshaveshownthatmaterialremovalinCMPoccursasaconsequenceManuscriptreceivedDecember15,1999;revisedNovember17,2000.ThisworkwassupportedbytheNationalScienceFoundationthroughawardNSFDMI-9813039andUniversityofCaliforniaSMARTprogramunderContract97-01.TheauthorsarewiththeLaboratoryforManufacturingAutomation,DepartmentofMechanicalEngineering,UniversityofCaliforniaatBerkeley,Berkeley,CA94720-1740USA(e-mail:[e-mail];[e-mail]).PublisherItemIdentifierS0894-6507(01)03522-9.Fig.1.Schematicofthechemicalmechanicalplanarizer.ofacombinationofchemicalreactionoftheslurrychemicalswiththewafersurfacematerialsandtherepeatedsliding,rolling,orindentationoftheabrasiveparticlesagainstthewafersurface[2].MostanalysesofthemechanismofCMPtreatthechemicaleffectsandmechanicaleffectsoftheslurryseparately.Inthispaper,wefocusonstudyingthemechanicaleffect,particularlytheabrasionduetotheabrasive–waferandabrasive–padcontact.Thedetailedmechanicalinteractionbetweenthewafer,pad,andslurryhasbeenthesubjectofresearchforsometime[3]–[9].Mostoftheearliermodelswereatthemacro-scaleandconcentratedonsignificantinputvariablestotheCMPprocesssuchasslurryflowrate,pressure,velocity,frictionforce,lubri-cation,padandwafergeometry,andoutputvariablesincludingpolishrate,planarizationrate,polishrateuniformity,andsurfacequality.TheabrasiveandabrasionmechanisminCMPneedsfurtherinvestigation.Theabrasion/wear-basedPreston’sequation[3],whereisthematerialremovalrate,thedownpressure,therelativevelocityofwafer,andaconstantrepresentingtheeffectofotherremainingparameters,hasbeenwidelyusedinCMPprocesscontrolandconsumabledevelopmentforintegratedcircuit(IC)fabricationandmanufacturing.However,Preston’sequationonlyreflectstheinfluenceofprocessparametersincludingdownpressureandrelativevelocity.ExperimentalresultshaveshownthatinCMPmorefactors,especiallytheconsumablessuchasapolishingpadmadeofpolymermaterialsandslurrycomposedofnano-scaleabrasives,havelargeinfluencesonthematerialremovalrate[10]–[13].Forexample,itisfoundthatasofterandrougherpolishingpadwillyieldalargermaterialremoval.ThiscannotbeexplainedbyPreston’sequation.Furthermore,experimentalresultsshowthatthepressuredependenceofremovalrateforCMPwithsoftpadssatisfiesanonlinearrelationship[14],[15].Althougharoughlylinearcurvewithanonzeroinitialvalueatzerodownpressurecanbefittedforsomenonlinearexperimentaldata,leadingtoalinearrevisedPreston’sequation,no0894–6507/01$10.00©2001IEEELUOANDDORNFELD:MATERIALREMOVALMECHANISMINCMP:THEORYANDMODELING113physicalorexperimentalevidencesupportstheinitialmaterialremovalrate.Actually,itisfoundthattwoseparatelinearcurvesmustbeusedtofittheexperimentaldataforunderhigherpressuresandunderlowerpressures,ifthenon