半導體界統(tǒng)計和R的應(yīng)用

ContentBackgroundRUsageatSMICIntegratedCircuit(IC)ManufacturingCDSEMOverlayMacroCDSEMFilmsStepper/TrackEtchEtcher5XXX8XXXASET-F52401HitachiFilmsDep2.ICManufacturing1.ICDesignN-WellP-WellP+P+N+N+Metal13.FabricatedWafersSemiconductorDataFlowWATFTWS/CPWIP(MES,iEMS)WaferstartFaboutContinuallyDefectinspectionandreview:inspectiontoolKLA,compass)reviewtool:SEMLeica(OM)ContinuallyADI,AEI,CD,etcmeasurementcalledmetrology(MES)METDefect(KLARF)

WIPEQlogdataTestdataProcessdataAlsoreliability,memorybit,QCandotherdataStatisticiansatSMICReliabilityAnalysisQualityControlPriceTimetoMarketOriginalYieldRampOurMission:

ImprovedYieldRampYieldAnalysisRUsageatSMIC1、線性模型/非線性模型/廣義線性模型LinearmodelforinlinemonitorBasicmixedeffectmodelappliedtosemiconductordataVariouscorrelationanalyses2、統(tǒng)計圖形：闡述清楚統(tǒng)計原理與圖形元素的對應(yīng)關(guān)系Usefulandinterestingplots:histogramw/splits,plotw/table(Excellike),wafermap(3-D)3、非參數(shù)統(tǒng)計：各種基于秩的檢驗以及光滑方法Smoothsplineusage,EWMAControlChart5、多元統(tǒng)計：一方面展示已有方法的應(yīng)用，如主成分、聚類等，另一方面體現(xiàn)出R在矩陣運算方面的簡便Waferpatternclassification:clusteringanalysis7、數(shù)據(jù)挖掘和機器學習:Logistic回歸、kNN、以及神經(jīng)網(wǎng)絡(luò)、SVM等Logisticregressionapplication9、程序接口：C/Fortran/C++/Java等或者R(D)COM、RservePython/VB/Delphi/mySQLlinkingwithRR(D)COMunderstudyClimbtheMountainPeakofR,Fighting!!!Thankyou謝謝Q&A請您提問LinearModelforInlineMonitorInlineanalysisobjectivesAutomaticallyexamineALLmetrologyitemsAlarmonlysignificanttrendsandrelatedequipmentDrillDownABWATUniformityAnalysis(MixedEffectsModelExample)ProblemDescription:BreakWATparametersvariationintolot,wafer&diecomponents;Highlightlargeorincreasingvariancecontributorstobettercontrolthemanufacturingprocess.i:lotnumber1,2,…a;j:wafernumber1,2,…b;k:sitenumber1,2,…n.ReferenceBook:MixedEffectsModelsinSandS-Plus.AnalysisMethod:

MixedEffectsModel:three-levelnestedlinearmodel;Appliedareas:agriculture,biology,economics,manufacturing&geophysicsUsingR(nlme),lot,wafer&dievariancecomponentsareestimated.WATUniformityAnalysisExampleResultObsComponentVarComponent%ofTotalSqrt(VarComp)1Lot0.22531.60.4752Wafer0.41958.80.6483Die0.0689.60.2624Total0.713100.00.844OtherCorrelationAnalysesEquipmentComparison9days12daysQueueTimeInterestingPlot1HistogramplotInterestingPlot2WaferMapSmoothSplineUsageSplitReducenoiseSplitcontrollimitSmoothsplinesimulatesthedataforreducingnoise

EWMAControlChartThePurposeofEWMAControlChart:Detectdatatrendshift;ReduceSPCX-barchart'sexcursionfalsealarmrate.ClusterAnalysisBlockissueSeriousedgeissueUsed“neuralnetworklike”methodsPeriodicfailpatternCompositewafersanalysisClusterfailpatternContinuousfailpatternPeriodicpatternContinuousfail

IllustrationsforSingleWaferAnalysisContinuousfailZonepatternSinglewaferanalysisClustersingle&multibinReticle/DUTpatternZonepatternContinuousfail

Clusterfail

Reticlefail

ProcedureInterface/DataPreparationProcedureInterface:

Python(Perl)/VB/Delphi/mySQLlinkingwithR.(Studying)DataConvert:Commonuse：reshape,merge,sort/order,apply,as.P