A Defect Detection Model for Imbalanced Wafer Image Data Using CAE and Xception

The development of technology in modern society causes consumers to create new demands. And consumers' demands lead to improved product quality. In particular, as the mobile era enters, the development of semiconductor technology is essential for electronic products. In electronics, semiconductors are used as various precision parts and control the performance of products. Therefore, improving the yield of semiconductors is the most laborious task for semiconductor companies. In the semiconductor manufacturing industry, semiconductor wafer defects are a major problem causing large losses. In semiconductor manufacturing, which includes many processes, wafer defects cause various variations, resulting in great losses. Accurately identifying and classifying defects would bring great benefits to the semiconductor manufacturing industry. Wafer defect inspection is being conducted passively by experts. Wasting such passive and human resources can be prevented through machine learning. In this paper, a deep learning-based model using Xception is proposed to proceed wafer defect detection and classification. Xception has a total of 36 convolution layers and consists largely of three flows. In addition, to solve the imbalance problem of the dataset, data augmentation was performed using the convolutional autoencoder. Through the proposed method, it was possible to improve the detection and classification of wafer defects while solving the problem of data imbalance.