 |
International Society of Science and Applied Technologies |
| Self-Training for Accuracy Improvement in Wafer Map Defect Pattern Classification Using AI | ||||
| Author |
Masaru Kurokawa
|
|||
| Co-Author(s) |
Yoshikazu Nagamura; Masayuki Arai; Satoshi Fukumoto
|
|||
| Abstract | Classification of wafer map defect patterns is important to monitor occurrence and further to assist root cause analysis of manufacturing-process-induced systematic defects. Recently, AI-based methods have attracted particular attention, but they require a large amount of labeled data for highly accurate learning. In this study, we propose a self-training-based labeling method to increase the number of training data by labeling unlabeled data. In the proposed method, a small amount of labeled data is first used to train the AI. The trained AI then makes predictions on unlabeled data. We adjust confidence score of prediction for the data that is likely to be misrecognized, and then label a part of unlabeled data based on the score. We report the results of a simulated application of the proposed method to labeled data and show that the labeling accuracy was 96.7%.
|
|||
| Keywords | wafer map, defect pattern classification systematic defect, self-training, adjusted confidence score, ResNet | |||
| Article #: RQD2025-91 | ||||
Proceedings of 30th ISSAT International Conference on Reliability & Quality in Design |