• 목록

Detecting and classifying defect patterns on wafers in semiconductor manufacturing processes is crucial for wafer quality management and prompt analysis of defect causes. However, due to the continuous technological innovation and advancement in semiconductor industry processes, the likelihood of unknown defect patterns emerging is increasing. These unknown defect patterns present a challenge that traditional classification and clustering methods, relying solely on existing training data, struggle to effectively address. To overcome this challenge, this study proposes a novel methodology based on Open-Set Recognition (OSR) using EEOC-SVM (Entropy Estimation One-Class SVM) to accurately detect unknown defect patterns. This methodology introduces unique preprocessing steps, including C-Mean filtering and Radon transformation, to remove noise and efficiently extract features from wafer bin maps (WBM). Through this approach, the study demonstrates the successful detection of unknown defect patterns in test data using only known defect pattern data. Additionally, this study evaluates the practicality and efficiency of the proposed method by utilizing new, undisclosed defect patterns occurring in an actual semiconductor manufacturing environment for assessment, achieving a detection performance of over 98% for various potential defect classes in semiconductor manufacturing data. These results validate the proposed method as a robust tool for effectively detecting and addressing unexpected defect patterns, thereby expecting to significantly impact quality management and maintenance in future semiconductor manufacturing processes.