A Novel Control Strategy to Improve On-Product Overlay with Context-Based Wafer Grouping for an Advanced Node Logic Layer in High-Volume Manufacturing Environment

Context-based wafer grouping is a technique to enhance run-to-run control performance by using historical wafer process context information. In this work, we show how machine learning with overlay metrology data and process context data can be used to detect which process steps dominate the correctable overlay and generate a group-based model for inline run-to-run control. The resulting model is used in shadow-mode run-to-run control for one key MOL layer of an advanced node Logic product. The predicted improvements are a drop in the rework rate by 9% as well as a reduction in the mean-plus-three-sigma stacked overlay by 0.29 nm.