A Workflow of Hotspot Prediction based on Semi-Supervised Machine Learning Methodology

Fast lithography hotspot (HS) detection flow becomes critical with continuously improving manufacturing conditions. A feature-vector-based machine learning tool is introduced to predict hotspots on different process stages and cross products based on known hotspots, which is called State of Nature Reduction (SONR) semi-supervised technology from Mentor's Calibre SONR. The workflow can be summarized into three steps, feature vector (FV) extraction, SONR model generation, and HS prediction. In this paper, two rounds HS prediction are done, one is within same products, and the other one is prediction on product B from known hotspots from product A, which is cross products. Verification data are collected post Chemical Mechanical polishing (CMP) stage.