AI-based enhanced detection algorithm for optical inspection of critical defects on 2nm and 1.4nm technology nodes

Defect inspection remains a pivotal process in semiconductor manufacturing; however, the continuous miniaturization of devices poses significant challenges to accurate detection. Addressing this challenge requires techniques that preserve high detection sensitivity while ensuring reliable discrimination of false positives. This study, performed on a DUV wafer inspection system and evaluated with Samsung’s 2nm and 1.4nm nodes, proposes an AI-based detection algorithm incorporating image enhancement and classification to tackle this challenge. The deep learning-based image enhancer improves optically scanned images to increase defect visibility, while the classifier utilizes the extracted features to predict the probability of a defect being yield-killing. Together, these components facilitate the separation of yield-killing defects from nuisance signals. Results demonstrate up to a 20% improvement in defect capture rate or an 89% reduction in SEM review budget, highlighting improved sensitivity for enhanced defect control.