Enhancing Defect Recognition: Convolutional Neural Networks for Silicon Wafer Map Analysis

Convolution Neural Network (CNN) has been utilized to automate time-consuming and manually detecting wafer map failure root causes to improve productivity and yield enhancement in the semiconductor industry. This research provides a reliable method for detecting defects from multiclass images of wafer mapping. The overall research content can be stated as, we used random sampling to extract a total of 902 images through the combination of the original dataset for each class. Then a Deep Convolution Neural Network model was proposed to generate a multiclass classification model along with Gradient-weighted Class Activation Mapping(grad-cam) to highlight the region of an image that contributes to neural network prediction and f1score to evaluate its performance. Then a comparison was made between the proposed model and three other popular deep learning models-VGGNet, ResNet, and EfficientNet and showed that our proposed model outperformed those. Thus, the experimental results show a high accuracy of 91.7% and confirms CNN effectiveness in wafer map failure recognition offering a reliable efficient solution for optimizing semiconductor manufacturing.