Analytic Solutions ForSemiconductorMay

ni.comThrough the acquisition of OptimalPlus,NI accelerates companies’ digitaltransformation initiatives by couplingNI leadership in automated test withnew advanced product analytics forenterprises.“We’re confident NI’s enterprisesoftware strategy unlocks the valueof test data by embracing digitaltransformation and bringing it to theanalog world.”Eric StarkloffNI CEO AND PRESIDENT

It’s AChanged WorldTechnological innovation hastransformed our lives.Products and devices aremore intelligent and connected.These products rely onthousands of electroniccomponents that must bemore reliable than ever

anty costsrelated to electronicsand semiconductors23xCar recall increasefrom 2014-2016 dueto electronics390McarsIgnition switch failureFailure to parkTakata airbag recall22.515x Drive per day4:1.5hr traditional car vs.22.5hr autonomous car 25Bhrsni.comCar innovations andnew features aredriven by electronics11 Automotive change drivers for the next Decade, EY, 20162 BMW - AEC Automotive electronics reliability workshop, 20173 NHTSA Recall Data4 Audi, DVCon Munich, 2017AUDI SAYS1 CAR FAILUREEVERY HOUR4Reliableelectronicsis a must

ni.com2005FOUNDED yzing hugevolumes of dataOpen innovationplatform for edgedeployment of realtime analytics andAI/MLapproach takingI4.0 and IIoT to thenext levelReady-madeLifecycle visibilityacross supply chainsand industriesCloud Or On-PremAWS (partner),Azure, GCP100bnDEVICES PER YEARROI track record andloyal trusted partnerProduct-CentricBig data analyticswith expertise inmanufacturingsolutions forAutomotive andSemiconductorindustries

Trusted By Leading BrandsCustomersSupply

Challenges We Address In The Semi Industry12ni.comGrowing chipcomplexity, includingadvanced materials,processes, andpackagesQuality requirementsget more demanding34Efficiently managefragmented supplychains – suppliers,sites, equipment,systemsNeed to improve timeto market of newproducts56Continuous pressureon profitabilityWhere and how touse AI/ML to maintaincompetitiveness

Our VisionLifecycle Analytics Through Product-Centric ApproachMANUFACTURINGDESIGN AND ENG.DESIGNCHIPBOARDMODULECUSTOMERPRODUCTIN USERETURNSDesign Spec Machine Process Metrology Test Rework Genealogy Performance Reliability Usage

System ArchitectureEdge (Factory Floors)MESDataControlRoom Factory AAction(internal or outsourced)Factory BDataActionDataActionDataActionDataActionPLM, ERP, CRM Portal www Rules O Edge Analytics24x7 rule execution and orchestrationO Data PlatformEdge repositoryDataAction/RulesDataActionRulesOther EquipmentAndData SourcesDataTesterCO ProxyDataTesterBO ProxyActionAssy.DataTesterAO ProxyDataFabDataCentralO Central Analytics24x7 rule executionand orchestrationO Data PlatformCentral repositoryFactory Cni.comActionable Insights Across All Manufacturing And Test ProcessesCloudOrOn-Prem.

Semiconductor SolutionsQuality,Reliability andBrand ProtectionYield Analysisand ReclamationEfficiencyTime To MarketAI/ML Deployment For ManufacturingData SecurityData Lifecycleni.comSupplierTransparency

Providing Innovative Solutions Collect lots of dataTypicalMethods Use it primarily when there is a problem: Bad Yield, RMA (Returns), Etc. Find the problem but frequently not the root cause Process is often manual and reactive, not proactive Use of many tools, but not an integrated solutionCollect Lifecycle dataharmonization of any typeO Solution Product, machineand process data Data securityni.comDetectAct Prescriptive analytics Automatic AI / Machine Learning Distributed 24x7 analytics engine Controlled Real-timeA unique, automated and proactive integrated solution

The Value We BringQuality, Reliability andBrand ProtectionMinimizeRMAsAnalyzeroot causeProtect yourbrandComply withautomotivestandardsYield Analysisand ReclamationImproveoverall yieldMinimize site-tosite variationsOptimizere-test policyIdentify equipmentperformance issuesEfficiencyEnable consistenttester availabilityand utilizationAvoid excessiveindex and pausetimesIdentifytest time variationsper testerEnsure efficientretest policies andexecutionShorten NPI timeOptimize balancebetween time,cost, and qualityFacilitate multiteamcollaborationShare learningsfrom NPI to HVMand backTime To MarketSupplier iersEnsure supplier compliancewith flows for every chip

Proxy 1Real-TimeData Collection2Real-TimeControl Runs on all major semiconductortest platforms Identifies issues as soonas they occur Ensures consistent data qualityand high-speed delivery Alerts operators Includes a wealth of informationnot provided in regular data logsfor accurate OEE analysis andsoftware/hardware validation Pauses the tester3Platform ForReal-Time Action Automated re-binning Adaptive test time reduction Drift detection Data-feed-forward and much more Agnostic to, and supports all testprogramsni.comProxy is an agent running on the tester, enabling real-time data collection, control and action

ni.comProxy – Optimal Ambassador On The ATEFor:Data collectionAdaptive testingand tester controlProxy ntersTester Executive SoftwareTestData

RulesTargeting Challenges 24x7Library of standard rulesaccommodate most thechallenges faced by ourindustryCustom rules available forunique monitors and actionsincluding support for R andPython scriptsDeployed at any level of yoursupply chain (central vs edge)Rules engine running

Rules Turning Challenges Into ActionsAction Categoriesni.comEquipment ActionsPauseEngineeringtool alertProcess ActionsPut materialson holdRe-binningRecipe AdjustmentsRe-testskip/addAdaptivetestingData ionAlertsQualityoutlier alertsYield alertsPredictive/Anomaly alerts

Data Security Solution1End-to-end securedata lifecycle Authenticate with testerOS and test program (TP) Secure channel betweenTester and Local Server Secure data at rest ontester and local server Secure transfer of data logs toHQni.com234Key exchangemechanismOffline mode foroffline data logs andrecoverySensitive data filteringto securely share datalogs with suppliers56Allowing Real-timeand offline ruleswithout exposingsensitive dataCompatible with olderTPs not implementingencryption

Our MarketectureGO ActionEdgeML Edge Real-timeEnablersGO WECOSPLRe-BinningEWMAScratchDetectionGOCore AnalyticsSolution SuiteEdgeni.comLifecycleData SourcesAuto-HoldData FeedForward andTP APIShop FloorControlEscapePreventionAdaptive TestSimulationNPIPortal BIProxy Data Lifecycle PlatformData LakeLevel 1-2Level 3Level 4Mfg Machines (Fab, Sort, Assembly, BI, FT, AOI, SLT)MES, QMS, YMSERP, CRM, PLM,.GO SecurityCentralAi / MlAdvanced AnalyticsSolution SuiteOutlierDetectionAdaptive Test(incl. SmartRamp)

ni.comSolution Examples

ni.comQuality, Reliability andBrand ProtectionSolutions

Quality and ReliabilityMinimize excursionsMinimize RMAsEscape PreventionAnalyze root causeSpecial quality algorithms –WECO, EWMA, SPL, Scratch DetectionProtect your brandAuto-hold (via MES)Comply with automotive standardsni.comOutlier DetectionRe-bin (via MES)Data Feed Forward and Test Program API

EXAMPLEOutlier DetectionNNR (Near Neighbor Residual)Wafer Map – Original Data TableStatistical WidgetStatistical WidgetNNR

ni.comBivariate Outlier DetectionBivariate outliersmay be relatedto pairs of testsfrom the sameor differentoperations

ni.comMultivariate Outlier DetectionSeveral MLtechniques canbe used to screenmultivariateoutliersPC’s as virtualtests selectionThe methodologyhere is to usePCA (PrincipalComponentAnalysis) todefine the mainvirtual tests, andthen performDPAT on suchtests

Specialty Algorithm ExampleScratch Detectionni.comPatent pending

ni.comEscape Prevention ExampleNot Enough Tests Performed On PartsStandard PRR(53 tests) Chart showsthat thenumber oftests for agood device is53 tests 5% of the unitson one lothave 39tests Automatedrule detectsthis inproduction andprevents theparts fromshippingLow PRR(39 tests)MISSINGTESTS

PAT For Packaged Units (FT PAT)Skip next testing for units marked as ‘outlier’ bins* Requires ECIDFinal Test“n”O PATVirtually Bin OutliersFinal Test“n 1”Physically Bin OutSpec LimitsSpec LimitsPAT Limitsni.comPat Limits

Defectivity Index (I-PAT) Correlation To SortBetter screening using both test and defectivity dataApplying I-PAT defect outlier recognitionUsing G-PAT to detect clusters using combination of test and I-PAT dataSmart I-PAT Mapni.comBin MapI-PAT can identifyindividual statisticaloutlier die, and drilldown to root causeBin Map PostStandard Outlier DetectionBin Map PostEnhanced Outlier DetectionHB99 G-PAT(test only)HB998 I-PAT Static PATHB997 G-PAT outliers(test and defectivity)

Image Processing Flowni.comSAM inspectiongenerates board image(jpg)Image algorithmanalyzes each weldinglocation (pin) on theboardParameters are loadedinto O to allowanalytics, rules andcorrelations tomachine/product data18 parameters aregenerated for each pinon the board (instead ofjust one parameterpreviously collected– the welding area)

ni.comYield Analysis andReclamation Solutions

YieldOverall yieldBaseline yield and SBL monitoringSite-to-site yieldTest equipment performanceRe-test policyTest and retest policies and executionEquipment and hardware performanceissuesTests limits validationCross-operation correlationTargets against any measure/

ni.comCustomer Use Case: Operational YieldSite IssueIteration 0Device:NetworkProblem:Yield lossIteration 1Standard O Rules FoundWith no monitoring – Site-Site issue not detected – This case is 16 lotsIssue:Yield by testervaries

ni.comYield Improvement ExampleTight Spec LimitsExisting Spec LimitsCutting Distribution TailProposed limits aresensitive togross outliers Current testlimits are tootight, causing0.4% yield loss Proposed testlimits willreduce yieldloss withoutimpactingproduct quality


EfficiencyInconsistent tester availability andutilizationExcessive index and pause timesTest time variationsper testerInefficient retest policies and executionAdaptive Testing using Machine LearningTest equipment performanceTest and retest policies and executionTesters availability and utilization(OEE analysis)Classical Test Time Reduction(TTR analysis, ROA)Adaptive Test Time Reduction (ATTR)Cross-operation correlationsni.comShop Floor Control

ni.comTest Efficiency OpportunitiesBetter resolution of time during testActual test time maximization (vs. index time)Retest optimizationTest time consistencyTester utilization – owned, consigned or paid for

ni.comCustomer Use Case: Efficiency ProblemIncreasing Test TimeDevice:Microcontrollerwith flashProblem:Capital avoidanceIssue: Needed 10more test stationsProblemDiscovered:Issue with testprogramFix:Standard O rules foundTesters had different throughputsTest Time Increasing from 120 Sec to 300 SecResult: Saved 8 test stations 12M in CapEx and OpEx SavingsImproved O rulefor monitoring forall futuretesters/devices

ni.comOverall Equipment EfficiencyTester Usage Breakdown By CategoryO collectsdetailed data ontester operationTester usagestatistics allowto performin-depthproductivityanalyses(e.g. OEE) whichhelp eliminatewasted time

Customer Use CaseTest Time ReductionFacility AHQPublishSimulateMESProxy/ServerCreate RuleTester/ProxyFacility BPublishMESTester/Proxyni.comIdentify tests that can be skipped, create rules and publish to the testers, wherever they areProxy/Server

ni.comAdaptive Test Time ReductionExample Run Showing TTR ElementSamplingZero failsvalidation(beforeskipping tests)in each runSKIPPEDTOUCHDOWNSVALIDATIONTTR SAMPLETOUCHDOWNSINGLE RUN


Time-To-MarketShorten NPI timeOptimize balance between time, cost,and qualityFacilitate multi-team collaborationShare learnings from NPI to HVMand backAdaptive test (reduction or augmentation)and smart rampData loading rulesLoad and create conditionsSandbox to edit metadataDatasets Virtual “workbench” Shared analyses and dataaugmentation Full chain of custodyLimits, Correlation andGR&R ApplicationsReport

ni.comNPI Areas Of Focus and FlowO DataCollectionSandboxO DatasetManagementMinimize timeto marketAnalysisReportingAnalyze split lotsDetermineproduction limits Proxy OTDF Data Cleansing AssociationPVT Analysis Static/Scheduled DropBox SAF Mapping AugmentationCorrelation App Flexible Validation Attributes STDF TemplatesLimits AppGRR App Intuitive CustomizedIdentify designsensitivities

ni.comCustomer Use Case: Time-To-MarketLimit Simulation AppDevice:Cell phoneProblem:Limits notoptimizedIssue:Would not failquestionablemeasurementsFix:Run analysis usinglimits applicationO standard tools found: Limits too wideResult: Immediate feedback Faster product launch

ni.comSupplier TransparencySolutions

Supplier Transparencyni.comBenchmark suppliersSite to site comparisonEnsure supplier compliance with flowsfor every chipSupplier to supplier comparison

ni.comCustomer Use Case Providesconsolidatedviews ofoperationsacross allsuppliers mfg.sites Enablesobjectivebenchmarkingof suppliers Highlights KPIsthat requireattention Enablesdrilldown forroot-causeanalysisSuppliers Benchmark Dashboard For Key KPIsSupplier #1 has lowerGPH than the othersSupplier #1#2#3#4Supplier #1 Avg.touchdown test timelonger than others#5Supplier #1#2#3#4#5

ni.comCustomer Use CaseSupplier Transparency Into Consigned Test-Fleet Performance Providesconsolidatedview of100s-1000sfleet tools Enables realtime andconsistentequipment setsbenchmarking Highlights KPIsrequireattention Enablesdrilldown forroot-causeanalysis 20% of the fleet underperforming – FPY issuesTest tools wasting 60-80%of test time in Pause mode

ni.comPartner with us to enhanceyour big data strategy with ouropen platformSynergetic With Any Data Lake Cloud and On-Premise Accessible Optimized Schema AI and Machine Learning Collect and Act Anywhere Enhance Data Scientist Productivity

Data Platform NeedsVoice of the Market“How can I combine, and do more withmy siloed data systems?”“Our data retention is at least 10 yearsfor our automotive products.”“I know we need to do ML, we justdon’t know how to get started.”“How can we store old data so it doesn’ttake so long to reload and use?”“My teams are proficient inPython or R and I want to leverage this.”“Can we have programmatic access to O data?”“We already have a corporate license ofTableau, can we use this to visualize O ?”“I want to leverage fab/assembly data (i.e. defectand inspection) to improve my quality.”

Consolidated ChallengesCTO/CIOsand IT ProfessionalsProduct, Quality andYield Engineering TeamsData Scientists andEngineering Teamsni.comConcerned about enterprise TCO (Total Cost of Ownership)Need a solution providing analytics that scaleNeed a collaborative ecosystem

Platform Goals12ni.comSupport bi-directionaldata integration with anycustomer data lakeEnable easyconsumption ofOptimalPlus data by 3rdparties and BI tools34Integrating withmachine learning datascience frameworks,leveraging OptimalPlusdeployed infrastructureBoost developer'sinnovation by leveragingOptimalPlus rich API’s,algorithms andinfrastructure5Data security andencryption

Industry Focused Open PlatformOptimal PlatformSynergetic with anybig data strategyConnected to existinginfrastructureOpen for all kinds of dataAccelerates innovationExtensible through bothdata and algorithmsOptimal data pipelinePortal Desktop and Web UIAnalytics Engine and APIData PipelineIBM StreamsColumn Store DatabaseMetadata index “Hot” cacheSQL over HadoopOptimized“Manufacturing” DataCustomer Data Lakeni.comCustomer PlatformCloud(public,private,hybrid)Customer Data

The Full MachineLearning LifecycleLearn from data and evaluatebusiness valueLEARNACTVALIDATEADAPTDeploy and act upon the modelMonitor data and model performanceto identify changesUnderstand changes and updatemodel/

AI/ML Deployment ChallengesLearnGetting dataData scientists wastetime getting andorganizing dataFeature extractionIt is difficult to extractcomplex features fromthe data setFreedom of choiceData scientists want touse their favorite toolsand the latest-andgreatest algorithmsni.comActComplex “plumbing”Data scientists waste timedealing with the “plumbing”associated with getting amodel into productionActionabilityTaking action requiresintegration with equipmentand systemsDistributed mfg.Issues compounded indistributed, outsourcedmfg.ValidateAdaptOngoing validationProduction models needto be validated all thetimeStale modelsProduction changesinevitably cause modelsto go staleOngoing data collectionData collection becomesan ongoing concernRelearningModel relearning is oftenmanualTechnical debtData scientists end upspending time monitoring“old” projects instead ofinvesting in new ones

ni.comHidden ComplexityThe Google ViewDataVerificationConfigurationData nagementAnalysis ToolsProcessManagementToolsSource: Google article from 2014: Hidden Technical Debt in Machine Learning ringServingInfrastructureIt’s allabout theinfrastructure

Optimal Covers The Entire aDataPLM, ERP, CRM Portal www Rules ConfigurationDataon/RulesO Central Analytics24x7 rule executionand orchestrationData CollectionMLCodeCloudOrOn-Prem.O Data PlatformCentral ntAnalysis entToolsOptimal covers the full scope all the way through ML deploymentni.comSource: Google article from 2014: Hidden Technical Debt in Machine Learning chnical-debt-in-machine-learning-systems.pdf

SummaryLifecycle AnalyticsSolutionsProduct-CentricApproachAI/MLOpen PlatformEnd-To-End SupplierTransparencyDomain Expertiseni.comturning data into actions or immediate ROIfor improved quality and reliability andoperational efficiencysupport digital transformation inmanufacturingindustry focused for seamless integrationwith any big data strategyacross operations and industriesapplying data science to solve industrychallenges

Significant Business Impactni.comQuality, Reliability andBrand Protection50% case avoidanceYield Analysisand Reclamationincrease up to 10% NPI 2% HVMEfficiencyup to 25% test cost savingsTime To Marketfrom weeks to days NPI, TTM, RCASupplier TransparencyConsistency and compliance

Ask Our Customers“Escape Prevention enables us to identifyspecific manufacturing and test issues thatdrive advanced quality screening andcomprehensive product management.”Michael CampbellKeith KatcherSENIOR VP OF ENGINEERINGVP OF OPERATIONS ENGINEERING“Global Ops for Electronics enables us torapidly identify and respond to the source ofany PCB and systems manufacturing issue,down to an operation, facility, line or station.”Vincent TongSENIOR VP OF GLOBAL OPERATIONS AND“Optimal gives us real-time visibility of our testoperations, enabling us to monitor every criticalparameter to ensure that every product is of thehighest quality and performs as expected.”“We see Optimal as a strategic partner. Theiropen architecture enables us to create synergyacross different tools and systems across theglobe and accelerate innovation”David ReedEXECUTIVE VP OF TECHNOLOGIES AND OPERATIONS

ni.comThank You

Prescriptive analytics AI / Machine Learning 24x7analytics engine Real-time Automatic Distributed Controlled Collect lots of data Use it primarily when there is a problem: Bad Yield, RMA (Returns), Etc. Find the problem but frequently not the root ca