Deep Learning for Wafer Map Defect Detection: A Review

Wafer defect recognition is a critical task in semiconductor manufacturing, as the detection and early identification of defects can significantly impact yield and product quality. Various inspection methods are employed to detect defects on semiconductor wafers, and traditional approaches often involve manual inspection by domain experts. However, these methods are time-consuming, labor-intensive, and prone to errors due to the complexity of wafer images. As a result, computer-aided methods have gained traction to alleviate human burden and reduce errors caused by fatigue and subjective differences. In recent years, deep learning has emerged as a powerful approach for wafer defect recognition. Its hierarchical architecture enables the extraction of high-level features, and its strong feature extraction capabilities contribute to accurate defect classification. Recent studies have shown that deep learning algorithms can achieve higher accuracy and efficiency than manual classification by experts. This review article systematically examines the application of deep learning in wafer defect recognition, focusing on four typical algorithms: autoencoders, convolutional neural networks (CNNs), generative adversarial networks (GANs), and recurrent neural networks (RNNs). The mechanisms, developments, and applications of these algorithms are introduced. Subsequently, their applications in wafer defect recognition are comprehensively reviewed, highlighting their strengths and limitations. Finally, looking ahead, the potential for further development of deep learning in wafer defect recognition is promising. Future research could explore more complex architectures, opening new avenues for real-time and efficient defect detection applications.