Anomaly Detection and Segmentation for Wafer Defect Patterns Using Deep Convolutional Encoder–Decoder Neural Network Architectures in Semiconductor Manufacturing

Abstract

Abnormal defect pattern detection plays a key role in preventing yield loss excursion events for the semiconductor manufacturing. We present a method for detecting and segmenting abnormal wafer map defect patterns using deep convolutional encoder-decoder neural network architectures. Using a defect pattern generation model, we create synthetic wafer maps for 8 basis defect patterns, which are used as training, validation, and test datasets. One of the key capabilities for any anomaly detection system is to detect unseen patterns. We demonstrate that by using only synthetic wafer maps with the basis patterns for network training, the models can detect unseen defect patterns from real wafer maps.