Clustering for Semiconductor Defect Using a Small Number of Training Data and Qualitative Knowledge

Abstract

In semiconductor wafer manufacturing processes, defect candidates are usually extracted by an inspection system. The defect candidates are composed of true defects such as open circuits, contaminants, and bridge as well as non-defect patterns, called nuisances, which dominate true defects. The goal of this study is to classify the defects candidates into each true defect and nuisance using a small number of training data given by SEM inspection. It is shown that the accuracy of the clustering is considerably improved by introducing qualitative knowledge, which is a priori given by an inspector, in the clustering processes.