Stochastic Gradient Boosting Methodology for Virtual Metrology on a PVD Tool

摘要

Equipment qualification and periodical equipment checks are often related to processing of test wafers. We run bi-daily test wafer on our PVD tool (Physical Vapour Deposition) for monitoring the film thickness, the film reflectivity and the film resistivity, respectively. By means of FDC data we want to use classification and regression trees as well as methods based on regression tree models like stochastic gradient boosting for modelling these metrology results. Reduction of test wafers, higher throughput and therefore cost reduction is the goal. Description Tool qualification and tool monitoring is a cost and time consuming procedure for semiconductor manufacturing and especially metal film deposition with a PVD tool. Therefore we decided to choose a PVD tool qualification as a fist case study of the Stochastic Gradient Boosting methodology. Every day we run one test wafer for chamber 3 and chamber 4 of the PVD tool with the bi daily test recipe P-AlCu-800. After film deposition the wafers were moved to several metrology tools: XRF (film thickness), RES01 (resistivity) and INT01 (reflectivity).