Title: Deep learning-based wafer fabrication quality assessment in semiconductor manufacturing

Authors: Jinxing Zhao; Quan Meng; Haolan Zheng; Yuhao Fan; Zinuo Zeng

Addresses: School of Mechanical Engineering, University of Shanghai for Science and Technology, Shanghai, 200093, China ' School of Mechanical Engineering, University of Shanghai for Science and Technology, Shanghai, 200093, China ' School of Mechanical Engineering, University of Shanghai for Science and Technology, Shanghai, 200093, China ' School of Mechanical Engineering, University of Shanghai for Science and Technology, Shanghai, 200093, China ' School of Mechanical Engineering, University of Shanghai for Science and Technology, Shanghai, 200093, China

Abstract: Semiconductor manufacturing relies heavily on wafer fabrication quality, as it directly determines the performance and reliability of downstream electronic and optoelectronic devices. To address the challenges of reliability and efficiency in wafer quality assessment, we propose a deep learning-based multi-label defect detection and classification method for wafer fabrication. The method employs the ShuffleNetV2 for feature extraction, and the Sigmoid activation function for multi-label outputs. The CB-Focal Loss was introduced to tackle class imbalance while the coordinate attention mechanism was integrated to enhance the model's focus on defect regions. Experiments on our own AFM-Wafer dataset demonstrate that, compared with the baseline model, new model achieves 0.51%, 1.39%, 1.31%, and 1.24% increases in Accuracy, L-Precision, L-Recall and L-F1 score, respectively while maintaining nearly the same parameter count and inference speed. This research demonstrates how advanced computer vision and deep learning complement traditional manufacturing and economic merits.

Keywords: wafer fabrication; deep learning; defect detection; quality assessment; semiconductor manufacturing.

DOI: 10.1504/IJMMS.2025.155740

International Journal of Mechatronics and Manufacturing Systems, 2025 Vol.18 No.4, pp.340 - 355

Received: 01 Aug 2025
Accepted: 04 Jan 2026
Published online: 12 Aug 2026 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article