Semi-supervised learning for simultaneous location detection and classification of mixed-type defect patterns in wafer bin maps

Abstract

In semiconductor manufacturing processes, it is crucial to identify the patterns of defective chips in wafer bin maps (WBMs) because different defect patterns are related to different root causes of process failures. Recently, as the semiconductor manufacturing processes have become more complicated, mixed-type defect patterns (i.e., multiple defect patterns in a single wafer) have occurred more frequently. Previous methods for classifying mixed-type defect patterns in WBMs have focused on outputting the class labels of the defect patterns only, not their locations, although the location information of the defect patterns can be useful to track the failure root causes and improve the processes. Moreover, the previous methods have mainly used labeled WBM data only, although a lager quantity of unlabeled WBM data is more accessible because of the costly process of label annotation in practice. In this work, we propose a semi-supervised learning method to classify the mixed-type defect patterns and detect their locations simultaneously using both labeled and unlabeled WBM data. Specifically, the proposed method extends the previous Attend-Infer-Repeat, which is a recent unsupervised object detection method, in a semisupervised fashion to perform both object detection and classification simultaneously. We verify the performance of the proposed method using WBM datasets of various sizes. The results demonstrate effective classification and location detection performance of the proposed method.