PLANARIZATION BY CHEMICAL MECHANICAL POLISHING A R

PLANARIZATIONBYCHEMICALMECHANICALPOLISHING:ARATEANDUNIFORMITYSTUDYA.E.FreenyW.Y-C.LaiAT&TBellLaboratoriesMurrayHill,NewJersey07974ThismemorandumwillbeincludedinthevolumeStatisticsintheSemiconductorIndustry:CaseStudiesofProcess/EquipmentCharacterizationtobepublishedbySEMATECHin1993.Planarizationbychemicalmechanicalpolishing(CMP)isgainingimportanceasVeryLargeScaleIntegration(VLSI)pushesbelowthe0.5μmregime.Theglobalplanaritywhichcanbeachievedbythistechnologypermitsmaintainingfinefeatureswithincreasingmetallevels.Inthisarticlewedescribeadesignedexperimentthatweusedtohelpidentifyanoptimumprocessingregimeandquantifyprocessreproducibility.Lookingattheeffectsofplatenandwaferrotationonrateanduniformityofoxideremoval,wediscoveredatradeoffbetweenrate(throughput)anduniformity.Aprocessdriftovertimewasquantified.Theseresultscontributetoajudgementofcostandtiminginintroducinganewprocessandequipmentintomanufacturing.1.EXECUTIVESUMMARY1.1ProblemAdvancedphotolithographyinVeryLargeScaleIntegration(VLSI)increasinglydemandsglobalplanarityacrossachip-sizedprintingfieldforfineresolution.Planarizationbychemicalmechanicalpolishing(CMP)isasimpletechniquethatcanachievethis.InadesignedexperimentweevaluateapolisherwhichdoesoxideplanarizationbyCMPforpossibleuseinthewaferfabricationmanufacturingprocess.Ourgoalistofindthemaximumrateofoxideremovalwhichcanbeused(forthroughputreasons)withoutdegradingtheuniformityoftheremovaloverthesurfaceofthewafer.1.2SolutionStrategyInCMPawaferisheldbyarotatingcarrierandispolishedbypressingthewaferfacedownontoapolishingpadonarotatingplaten.Theimportantparametersforthepolishingprocessareplatenandwaferrotationfrequenciesandpolishingpressure.Designofexperiments(DOE)isaneffectivewaytoinvestigatetherelationshipoftheseparameterstooxideremovalrateanduniformityacrossthewafer,whilequantifyingthereproducibilityoftheremovalprocessfromwafertowafer.WedesignedanexperimenttostudytheeffectsofplatenandwaferrotationsforanautomaticwaferpolishingmachinesuitableforCMP.Allcombinationsoffivewaferandthreeplatenrotationfrequencieswereused.Threewaferswerepolishedateachcombinationofwaferandplatenrotationfrequencies.Atendencyofthepolisherremovalratetodriftlowerthroughtimehadpreviouslybeennoticed.Thusanimportantdesignissuewastochoosetheorderofthefifteencombinationssoastoallowestimationoftheeffectsofthedesignparametersindependentofalineardriftaswellastoquantifythemagnitudeofthedrift.Graphicalmethodsandanalysesofvarianceallowdetectionoftheeffectsofthe-2-parametersandestimationoftherelationshipofuniformityandremovalratetoplatenandwaferrotationfrequencies.1.3ConclusionsBoththepolishingrateandacross-the-wafernon-uniformityshowedlargeincreaseswithincreasingplatenrotationandslightincreaseswithincreasingwaferrotation.Asimplelinearmodelisappropriateforquantifyingtheseeffects.Apreferredoperatingregimeforplatenandwaferrotationfrequencieswasderived.Thepresenceofthesuspecteddriftwasconfirmedasadecreaseinpolishingrateovertimeregardlessoftherotationfrequencies.Thisdrift,combinedwithlargewafer-to-wafervariability,resultsinimperfectprocessreproducibilityevenwithautomation.ThislackofreproducibilityisanimportantfactorinjudgingthecostandtimingofintroducingCMPintomanufacturing.2.PROCESSChemical-mechanicalpolishing(CMP)[1]isanovelapproachtoplanarizingoxidetopography.Theprocedurepressesarotatingwaferface-downontothesurfaceofarotatingpaddedplatenimpregnatedwithaslurryofextremelyfineabrasive.Thekeyparametersaffectingtheremovalrateanduniformityacrossthewaferarethewaferpressureagainstthepolishingpad,theplatenandwaferrotationfrequencies.Thepolisherusedintheexperimentwasanautomaticcassette-to-cassetteloadingsingle-waferpolisherwithfivestations:input,primarypolish,secondarypolish,rinse,andoutput.Wafersareremovedsinglyfromtheinputcassetteandtransportedviathecarriertotheprimarypolishingstation.Thepolishingactionconsistsofthreerelativemotions:thewaferrotationaboutitscenter,theplatenrotation,andasmalloscillationofthearmsupportingthewafercarrier.Afterprimarypolishingiscompleted,thecarriertransportsthewafertothesecondarypolishingstation,whichperformsaclean-upstep.Thepolishedwaferisunloadedviaawatertrackintotheoutputcassette,andkeptsubmergeduntilcassetteremoval.3.DATACOLLECTIONPLAN3.1ThicknessMeasurementsandDerivedResponsesFigure1showsthe37sitesfromthe49possiblesiteswhereoxidefilmthicknesswasmeasuredonexperimentalwafers.Eachwaferwasmeasuredbeforeandafterpolishingsothatacomparisoncouldbemadeonasite-by-sitebasis.Inthisstudy,measurementsfromsites27-33and36-40wereeliminated,leaving37sites:theformerbecauseofproximitytothewaferscribemark;thelatterbecauseofproximitytothewaferflat.Both-3-thescribemarkandtheflatinducenonuniformoxideremovalbecauseoflocaltopographydisruption.Theamountofoxideremovedwascalculatedasthedifferencebetweenthethicknessesbeforeandafterpolishingateachofthe37sites.Theresponsesofinterestweretheoxideremovalrateandtheuniformityoftheremovedoxidefilm.Therate,R,wasdeterminedbydividingthemeanofthethicknessesoftheremovedoxidelayermeasuredatthe37sitesbythepolishingtime(threeminutesin