Wafer Map Defect Classification using Deep Learning Framework with Data Augmentation on Imbalance Datasets

Abstract Wafer map defect classification is a key task for the semiconductor industry to improve the yield rate. Most wafer map defect classifications suffer from the problem of data imbalance and insufficient data. This paper proposes a global-to-local generative adversarial network (G2LGAN) method using the deep learning framework. It extracts global features and local features separately to generate effective data even in the imbalanced dataset. We use random under-sampling to suppress the majority class of data. We use MobilenetV2 as the classifier, and use two datasets for validation. One is open dataset 1WM-811K and the other is called 21-Defect built from the industry. Based on the serious dataset imbalance problem, this paper integrates data enhancement and random undersampling methods to optimize the dataset and uses the proposed classification network for classification tasks. The results of WM-811K dataset show that the proposed method has a classification accuracy of 98.39 and an F1-Score of 93.01. We also conduct cross-validation on the 21-Defect dataset and the results show that the proposed method has good robustness.