Reclaim Wafer Defect Classification Using Backpropagation Neural Networks

Abstract

Silicon wafer is a part of main cost of material in semiconductor manufacturing. Reclaim wafers have been extensively used in the semiconductor industry for many years. Reducing the usage of prime wafers and using reclaim wafers is one of the most cost effective solutions for a manufacturing plant. If any defects such as void, scratches, particles, and contamination found on the surface of reclaim wafers, these wafers will be classified as defective ones (NG). Due to previous study, we found reclaim wafers can be re-polished if its defect is not fatal and its thickness is enough for a repolish. Currently, those NG reclaim wafers must be screened by experienced human inspectors to determine re-usability through its defective mapping. This screening task is tedious, costly, and unreliable. This study proposed an artificial intelligent system to construct an automated reclaim wafer defect classification system, in which back-propagation (BPN) were used to analyze the pattern of defective mapping, and determined whether reclaim wafers can be re-polished or not. Experiments showed the BPN gave good performance in identifying NG and Good wafers. The proposed method was successfully increase the reusability of RM, and currently serves as important module in the MIS system of case-study company.