Stacked convolutional sparse denoising auto-encoder for identification of defect patterns in semiconductor wafer map

Abstract

In semiconductor manufacturing systems, those defects on wafer maps tend to cluster and then these spatial patterns provide important process information for helping operators in finding out root-causes of abnormal processes. Deep learning has achieved many successes in image and visual analysis. This study concentrates on developing a hybrid deep learning model to learn effective discriminative features from wafer maps through a deep network structure. This paper proposes a novel feature learning method, stacked convolutional sparse denoising auto-encoder (SCSDAE) for wafer map pattern recognition (WMPR) in semiconductor manufacturing processes, in which the features will be extracted from images directly. Different from the regular stacked denoising auto-encoder (SDAE) and convolutional neural network (CNN), SCSDAE integrates CNN and SDAE to learn effective features and accumulate the robustness layer by layer, which adopts SDAE as the feature extractor and stacks well-designed fully connected SDAE in a convolutional way to obtain much robust feature representations. The effectiveness of the proposed method has been demonstrated by experimental results from a simulation dataset and real-world wafer map dataset (WM-811K). This study provides the guidance to applications of hybrid deep learning in semiconductor manufacturing processes to improve product quality and yields.