Paper Title
Defect Classification of Reclaim Wafer By Deep Learning

Abstract
In semiconductor manufacturing,silicon wafer is the most critical material. Due to the limited resource, the reclamation of monitoring wafers and dummy wafersby refurbishing process can be dramatically lower the running cost for a semiconductor factory, which can become the competitive edge in this industry. It is found from previous study thatmonitoring wafers and dummy wafers can be re-polished if the defects such as void, scratches, particles, and contamination found on the wafers’ surfaceare not fatal and the wafer has enough thicknessfor re-polish.Currently, those No-Good (NG)wafers which are not suitable for reclamation must be screened out by experienced human inspectors to determine re-usability through its defective mapping. The screening task is tedious, costly, and unreliable. In this paper, an automated defect classification system for the reclaim wafer is constructed, which applies deep learning methodology to analyze the pattern of defective mapping and determines whether reclaim wafers can be re-polished or not. Experiments showed deep learning method can give excellent performance in identifying NG and Good wafers. It is also showed that the experimental results are better than using back-propagation neural network (BPN) method. Keywords - Reclaim wafer, Wafer Defect Classification, Deep Learning, Convolutional Neural Networks.