Minimization of CNN Training Data by using Data Augmentation for Inline Defect Classification

Detecting the defects in silicon wafers generated by semiconductor manufacturing is essential for quality assurance, and requires the acquisition and accurate classification of high-resolution images by scanning electron microscopy. However, owing to the difficulty of automation, the classification process is costly and its efficiency must be improved. To improve the classification accuracy and the cost of creating a classifier, which are the main bottlenecks of conventional technology, we propose a deep convolutional neural network (CNN) based on the VGG16 architecture, and perform appropriate data augmentations on training images. The CNN was successfully trained on a very small number of images, and achieved high defect classification accuracy.