A Novel Convolution Neural Network Model for Wafer Map Defect Patterns Classification

Abstract

Classification of wafer map defect patterns (WMDPs) plays an important role nowadays to improve the production quality in semiconductor industries. In the fabrication process, due to the complexity of semiconductor slices pattern, manual detection of wafer map defects is quite challenging and increases production costs. Thus, early detection of wafer map defects is necessary to improve product quality and hence reduce production costs. Correct classification of wafer map defects improves the inspection of fabricated wafers. In recent years, computer-aided automatic detection of wafer map defects classification is widely used by the semiconductor manufacturers. This paper presents a novel convolution neural network (CNN) model for the classification of images of WMDPs. The work presented considers an open dataset, namely WM-811K, consisting of 9 defects in wafer maps. The proposed CNN model reduces the parametric calculation and efficiently classifies different types of semiconductor wafer images from high volume production datasets. The experimental results using the proposed CNN model demonstrated that it can be classify the defects in wafer maps with higher accuracy of 99.70%. Further, the performance of the proposed model is also evaluated by comparing the results with the other state-of-the-art models in terms of F1-Score.