Combining Residual Network and Bidirectional Long Short-Term Memory with Additive Attention for Wafer Defect Classification

Abstract Accurate classification of wafer-map defect patterns is critical for boosting yield and reducing cost in semiconductor fabrication. To solve this image classification problem, we propose a new methodology that combines the advantage of the Residual Network (ResNet)—its ability to minimize information loss in deep networks—with Long Short-Term Memory (LSTM) to capture both spatial and sequential features.We propose the Shortcut3-ResNet with Five Sets and Bidirectional LSTM with Additive Attention Network (SCRBLAA-Net), a lightweight composite model. In this model, a Shortcut3-ResNet with Five Sets (SCR5) block first distills spatial features from standardized inputs. This vector is then reshaped and analyzed by an attention-augmented Bidirectional-LSTM (Bi-LSTM), and the refined sequence is fused back with the original spatial representation to minimize information loss. On the WM-811K dataset from which the “none” class was removed, SCRBLAA-Net achieves an F1-score of 93.99%, outperforming the baseline SCR5 network by 1.6% points and surpassing SCR5-LSTM and SCR5-Bi-LSTM variants by 0.78 and 0.21% points, respectively.