A Learning-Based Approach for Wafer Defect Detection in Production Quality Control

Wafer map defect detection has been a popular research topic due to its strong competition in the semiconductor industry. As a result, various methodologies have been investigated and used to address the wafer map defect detection and recognition issue. The aim of this study is to propose a model for a single-defect wafer classification problem based on an imbalanced dataset. In this model, a convolutional autoencoder (CAE) is proposed to deal with the imbalance and a Detection–Classification (D–C) strategy based on convolutional neural networks (CNN) model to classify defect patterns. The model is trained, validated, and tested on the WM-811K real wafer map dataset. The experimental results have proven to be applicable to actual wafer map inspection system. Some preliminary attempts can be made to improve the current results and future work.