Efficient Wafer Defect Patterns Recognition Using Deep Convolutional Neural Network

Defect recognition in semiconductor wafers is crucial in manufacturing process to identify the root cause. This paper presents an efficient wafer defects pattern recognition methodology based on deep convolutional neural networks (DCNN) to automate the classification. In DCNN design, we revamped LeNet-5 deep learning model. The proposed DCNN comprises seven layers including three convolutional and two pooling layers. We implemented an adaptive moment (Adam) optimizer to fine-tune model parameters in DCNN. We tested the proposed model on publicly available semiconductor wafers datasets to verify the effectiveness of the model. The experimental study suggests that the proposed model is highly efficient in classifying wafer defects with an accuracy of 99.22% in testing.