A yield aware sampling strategy for inspection tools capacity optimization
Résumé
The product quality in semiconductor manufacturing is ensured with 100% inspection at each process step; hence inspection tools quickly run out of capacities resulting in the production cycle delays. To best utilize the production and inspection capacities, existing sampling (static, dynamic and smart) strategies are based on the risk and delays. These strategies, however do not guarantee a reliable lot sample that represents a likely yield loss and there is a high risk of moving a bad production lot to next production steps. We present a 3-step yield aware sampling strategy to optimize inspection capacities based on the likely yield loss with the predictive state (PSM) and alarm (PAM) models as: (i) classify potentially suspected lots, (ii) cluster and/or populate suspected lots in the priority queues and (iii) apply last in first out (LIFO) to optimize capacities. This strategy is implemented with two heuristics. We also present a data model with ASCM (Alarm and State Control Management) tool for the multi source data extraction, alignment and preprocessing to support the validation of [PSM, PAM] predictive models.