Convolutional Neural Network (CNN) Based Automated Defect Classification (ADC) with Imbalanced Data

Recently, deep learning (DL) convolutional neural network (CNN) has been employed for automated defect classification (ADC), with its diverse modeling approaches and network configurations, aiming to provide the best performance classifiers for wafer defect inspection. However, in semiconductor wafer inspection, critical killer defects data samples are usually very few although it is critical to classify these defects correctly in early stage of the wafer inspection process. Without specifically handling the imbalanced data problem, a classifier induced from the imbalanced data set is more likely to be biased towards the majority class and results in very poor classification result on the minority class (critical killer defects). This paper proposes a CNN for wafer ADC while addressing class imbalance issue via generative adversarial network (GAN) generated images. The experimental imbalanced dataset, consisting of scanning electron microscopy (SEM) images, is collected with ASML-HMI inspection tools.