Wafer pattern classification and auto disposition by machine learning — James Lin

摘要

In order to increase bad wafer detection rate and reduce engineer trouble-shooting efforts, this work reports a combination of machine learning techniques to decode and extract the wafer map information from tool log via text mining, transform to useful virtual wafer images by image processing, auto wafer pattern classification by deep learning, and perform auto disposition (scrap, rework, add inspection, or waive) by flow chart based decision rules. 70% hold time can be reduced for certain hold codes, so that the engineers can focus on more valuable jobs rather than trouble-shooting.