20250121_PreConF_Hutner.PDF

1、Achieving On-Time Delivery of Production-Ready DesignsMarc HutnerDirector of Product Management for Silicon Learning SiemensMotivation2On-Time DeliveryProduction-ReadyIngredients for hitting marketing window and profitable(Chiplet differentiation is of primary importance,but economics matter)Where d

2、oes yield learning happen3Wafer TestPackage TestSystem Level TestYield Learning happens at every stageWhy does yield matter?4Device yieldTimeEarly process adopterMajority process adopterLate process adopterHistorical Process driven yield curveDesign driven yield curveYield learning identifies layout

3、 issues enabling defect driven yield The outcome is to enable device yield to the process yield curveNew Tech Process driven yield curve(more to learn/stabilize)Chiplet costPer Wafer Cost#good chiplets per waferEquation to minimizeChiplet example of impact of yield5Basic DefinitionExample300mm wafer

4、10mm*10mm parts594 manufactured parts per waferThe goal is to maximize number of good parts for highest possible yieldWhy does this matter?If each wafer costs$10,000*Where each device sells for$25Yield=#good parts#parts manufacturedYield#Good partsCost per partProfit per million parts produced*70%41

5、5$24.10$903,61480%475$21.05$3,947,36890%535$18.73$6,273,408*For latest process nodes cost per wafer$25,000*Highly simplified model,other saving terms in detailed model1%yield improvement=$300kConsider Economics of Multi-Die packages6Consider a 3D/2.5D package4 chiplets per packageEach chiplet cost$2

6、5Package test yield loss 10%Package cost of$50Shifting yield impacts materialImproving yield to 7%at package testCustomer produces 30,000 less parts(less waste)Higher profitability due to less materialProfitability improvement is 3-5x scrap costWhy does this matter?*Highly simplified models,other sa