The Classification of Semiconductor Wafer Defects by Means of Feature-based Transfer Learning Approach
摘要
Wafer defect detection is a critical part of the semiconductor wafer manufacturing industry. Human-based manual defect detection is tedious and often prone to misjudgement. Hence, there is a push towards automatic wafer defect detection in the industry. This paper shall investigate the efficacy of a transfer learning pipeline that utilizes different pre-trained VGG models, i.e., VGG16 and VGG19, that act as a feature extractor coupled with a Random Forest (RF) model in classifying wafer defect images. The hyperparameter of the RF classifier is further fine-tuned through the grid-search technique via a five-fold cross-validation method. It was shown in the study that the optimised VGG16+RF pipeline provided a better classification of the wafer defects.