Opposition Crow Search-based Optimal Feature Selection for Defect Classification by SVM in Semiconductor Wafer

In semiconductor manufacturing, the classification and detection of wafer defects and taking immediate action to find the root cause of defects are documented in quality management information. The problem statement of existing methods cannot handle camera and light source locations in the segmented process properly, which can lead to a reduction in wafer detection systems. The WM-811K dataset used in the proposed approach contains more data that has been trained and tested. The proposed method is optimal-based feature selection using the Opposition Crow Search (OCS) algorithm and Support Vector Machine (SVM)-based classification to identify wafer defect images. The numerical validation of this approach has an accuracy of 99.23%, precision of 99.39%, recall of 99.54%, and F1-score of 99.68%. By comparing existing methods like Attention-Augment Convolutional Neural Network (A2CNN), Ensemble-CNN (ECNN), Recurrent-CNN (RCNN), and Deep Convolutional Generative Adversarial Network (DCGAN), the proposed method overcomes all the limitations of the existing method.