Article: Wafer surface defect detection with enhanced YOLOv7 Journal: International Journal of Information and Communication Technology (IJICT) 2024 Vol.25 No.6 pp.1 - 17 Abstract: Silicon wafers are crucial materials for semiconductor chip manufacturing. Detecting surface defects on wafers is essential for enhancing yield rates and identifying manufacturing issues. Traditional defect detection methods, relying on manual monitoring, are inefficient and inaccurate. Thus, there is a growing interest in leveraging deep learning for defect detection. However, existing algorithms still suffer from missed detections and slow processing speeds. To address these challenges, our study proposes a refined algorithm based on YOLOv7 for detecting wafer defects. We integrate SPD-Conv into the YOLOv7 MP module to enhance feature extraction accuracy and reduce computational complexity. Additionally, we incorporate the CBAM attention mechanism module into the backbone network to adapt to complex scenes. Moreover, we employ the SIoU loss function to improve bounding box regression accuracy. The WM-811k dataset is utilised for testing and evaluating the enhanced algorithm, achieving a recognition accuracy of 92.23%, a recall rate of 94.1%, and a mAP of 92.5%. Additionally, the frame rate remains stable at 136 frames per second, outperforming existing algorithms. Inderscience Publishers - linking academia, business and industry through research

Title: Wafer surface defect detection with enhanced YOLOv7

Authors: Chen Tang; Lijie Yin; Yongchao Xie

Addresses: Hunan Railway Professional Technology College, No. 18 Tianxin Avenue, Zhuzhou, Hunan Province, China ' Hunan Railway Professional Technology College, No. 18 Tianxin Avenue, Zhuzhou, Hunan Province, China ' Hunan Railway Professional Technology College, No. 18 Tianxin Avenue, Zhuzhou, Hunan Province, China

Abstract: Silicon wafers are crucial materials for semiconductor chip manufacturing. Detecting surface defects on wafers is essential for enhancing yield rates and identifying manufacturing issues. Traditional defect detection methods, relying on manual monitoring, are inefficient and inaccurate. Thus, there is a growing interest in leveraging deep learning for defect detection. However, existing algorithms still suffer from missed detections and slow processing speeds. To address these challenges, our study proposes a refined algorithm based on YOLOv7 for detecting wafer defects. We integrate SPD-Conv into the YOLOv7 MP module to enhance feature extraction accuracy and reduce computational complexity. Additionally, we incorporate the CBAM attention mechanism module into the backbone network to adapt to complex scenes. Moreover, we employ the SIoU loss function to improve bounding box regression accuracy. The WM-811k dataset is utilised for testing and evaluating the enhanced algorithm, achieving a recognition accuracy of 92.23%, a recall rate of 94.1%, and a mAP of 92.5%. Additionally, the frame rate remains stable at 136 frames per second, outperforming existing algorithms.

Keywords: surface defects on wafers; improved YOLOv7; SPD-Conv; CBAM attention mechanism; SIoU.

DOI: 10.1504/IJICT.2024.141433

International Journal of Information and Communication Technology, 2024 Vol.25 No.6, pp.1 - 17

Received: 28 Mar 2024
Accepted: 02 Jul 2024
Published online: 12 Sep 2024 *

Title: Wafer surface defect detection with enhanced YOLOv7

Keep up-to-date