A Novel Deep-Broad Learning Network for Mixed Wafer Map Defect Recognition

In the integrated circuit industry, automated wafer defect recognition has become a crucial step in ensuring product quality. However, traditional deep learning methods, especially models based on convolutional neural networks (CNN), rely heavily on large-scale annotated datasets for model training. To address this limitation, we propose an innovative hybrid wafer defect recognition network that integrates the advantages of broad learning. Specifically, we introduce a deep-broad feature extraction module with an image pyramid input structure to capture rich contextual information from different scales, thereby extracting more comprehensive feature representations. These optimized features are processed using broad learning algorithms, achieving more accurate identification and classification of wafer defects. This method not only improves detection accuracy but also reduces the dependence on large amounts of training data, demonstrating its great potential in industrial applications. Compared with the current state-of-the-art methods, it further proves the superior classification performance of the proposed network. Ablation experiments performed on the basis of different backbone networks validate the enhancement brought by the broad learning approach, especially when the training data accounts for a small percentage.