76‐1: Using Machine Learning Solutions to Accurately Classify Imbalanced LCM Aging Data to Reduce Defect Rates

In semiconductor display manufacturing, liquid crystal display module (LCM) aging is a critical process for assessing the longevity, reliability, and performance degradation of devices over time. However, LCM aging data from real production lines are severely imbalanced, which reduces the accuracy of classifying minority classes, causing significant issues at product quality control. This paper proposes a hybrid method that combines machine learning algorithms, cost‐sensitive learning, and SMOTE‐ENN techniques to address the imbalanced classification problem in the LCM aging process. The method demonstrates high predictive accuracy, achieving an average skip rate of 74.79%, yield rate of 99.95%, and recall of 80.20%. This indicates that when implemented in a real production line, the AI model can potentially reduce the need for up to two‐thirds of aging facilities while maintaining a defect rate of less than 0.05%. Additionally, the resources saved can be focused on aging samples that require longer aging times, thereby reducing the rate of defective product release and enhancing product quality.