Memory-augmented convolutional neural networks with triplet loss for imbalanced wafer defect pattern classification

Abstract

A wafer bin map (WBM) represents wafer testing results for individual dies on a wafer using a binary value representing pass or fail. WBMs often have specific defect patterns, which occur because of assignable causes. Therefore, identifying defect patterns in WBMs helps to understand root causes of process failure. Previous studies on the classification of WBM defect patterns have shown effective performances. However, in previous studies, class imbalance over WBM defect patterns has not been considered, although in practice it is more reasonable to assume that there is a significantly large number of WBMs without any defect patterns because defect patterns occur when there are process faults. In this paper, we propose memory-augmented convolutional neural networks with triplet loss for classifying defect patterns in highly imbalanced WBM data. We use a triplet loss-based convolutional neural networks as an embedding function to make a well-separated low-dimensional space according to defect patterns. Then, we use a memory module to balance the number of WBMs between classes of defect patterns. We train the proposed model end-to-end to learn the embedding function and update the memory simultaneously. We validate the proposed model using simulated WBM data. The proposed model shows high classification performance and effective embedding results for imbalanced WBM data.