Deep Learning for Automatic IC Image Analysis

Abstract

We propose a systematic training and validation approach for obtaining a deep learning model for automatic IC image analysis, i.e. IC image semantic segmentation. Our approach divides IC images into different regions of interest and provides for noise rejection training. We discuss steps for obtaining such a model. By experiment and by comparison with competing image processing techniques, deep learning models obtained by our approach demonstrate good generalization capability, high prediction accuracy and low circuit annotation error.