Spatial Attention Enhanced Wafer Defect Classification Algorithm for Tiny Defects

• Main Authors: Ma, Can, Jie, Luyang, Yao, Yiming, Xu, Tong, Hu, Lilei
• Format: Conference Proceeding
• Language: English
• Subjects: Classification algorithms; Costs; Deep learning; defect classification; Inspection; Optical imaging; Semiconductor device manufacture; Semiconductor device modeling; semiconductor inspection; spatial attention
• ISSN: 2770-1603
• DOI: 10.1109/ICAIT59485.2023.10367417

Wafer inspection is a crucial step in the semiconductor manufacturing process. Accurate inspection of wafer defects is of great significance in ensuring chip quality, improving production efficiency, and reducing cost losses. Surface defects, such as pits and scratches, have a significant impact on chip yield during the intricate wafer manufacturing process. For further higher inspection capability, conventional semiconductor inspection methods are limited by optical imaging resolution and signal capture capabilities, resulting in low efficiency. Therefore, there is a need for a more efficient and high-performance automatic wafer defect inspection method in the semiconductor manufacturing industry. To address this challenge, a deep learning network for wafer defect classification that uses spatial attention block and residual structure is proposed. Moreover, as defects in wafer images only occupy a very small portion of the entire image scale, therefore, easily overlooked. Therefore, hot-spot mapping and denoising were carried out to effectively enhance the defect features in the wafer images. The proposed algorithm is evaluated by classifying defects in real wafer data collected by a home-developed optical system. The algorithm has demonstrated a comprehensive classification performance, even with a limited number of defect samples, surpassing other models.