A Deep Learning Analysis Framework for Complex Wafer Bin Map Classification

Abstract

Wafer maps are extremely critical data that need to be carefully analyzed for quality control and yield improvement. A wafer bin map (WBM) presents the chip probing test results for a wafer, based on the passing or failing categories (bins) of each die. Likewise, classification of the spatial failure pattern of WBMs corresponds to a key task, which currently relies heavily on manual methods. While automated classification methods have been proposed in recent years, most of them deal with simplified scenarios where only binary WBMs are studied with each WBM containing only one spatial failure pattern, which results in significant limitations in their applications. Accordingly, in this study, a deep learning analysis framework is proposed for complex WBMs that contain multiple bin numbers and multiple spatial failure patterns. The proposed framework is validated with a real-world dataset and a synthesized dataset based on WM-811K. The obtained results show that the proposed framework exhibits more effective classification performance compared with other competitive approaches.