Design and Development of an Image Classification Model for Wafer Defects using Deep Learning

From production standpoint, the semiconductor industry benefits from AI adoption at various process points to improve production efficiency and yield management. Implementation of defects auto-classification allows the offload of previously required manual operations. This research proposes an Automatic Defect Classification (ADC) system based on deep learning for the automation in wafer surface defects classification which focuses on defects identified at the metal layers. It adopts a deep convolutional neural network (CNN) object detection architecture as the model for defect detection and classification using only SEM images as inputs. This research produces a model which achieves industrially pragmatic defect classification performance.