Tsao, Yu-Ci (National Taiwan University of Science and Technology), Lauwardy, Jimmy (National Taiwan University of Science and Technology), Yen, Jia-Yush (National Taiwan University of Science and Technology), Chen, Liang-Chia (National Taiwan University), Tsao, Tsu-Chin (University of California Los Angeles)

Sharp Image Recovery by Combining High-Speed Rectilinear Motion Deblurring with Deep Learning Network for Whole Wafer Inspection

Scheduled for presentation during the Regular Session "Estimation and filtering" (WeCT1), Wednesday, July 16, 2025, 18:10−18:30, Room 105

Joint 10th IFAC Symposium on Mechatronic Systems and 14th Symposium on Robotics, July 15-18, 2025, Paris, France

This information is tentative and subject to change. Compiled on July 16, 2025

Abstract

This research proposes a concatenation of the rectilinear motion deblurring with the deep learning image deblurring network for a wafer inspection system. The wafer inspection system employs high-speed scanning to fulfill the full wafer inspection requirement by the current high-end semiconductor manufacturing process. The high-speed scanning process bears the unique property of a constant scanning speed. The proposed method makes use of this information to achieve a speed-insensitive deblurring algorithm. The deep learning algorithm is an integration of DeepDeblur, U-Net, and ResBlock to take advantage of the strengths of the individual networks. The underlining wafer inspection system achieves a scanning speed of 200 𝑚𝑚/𝑠𝑒𝑐 . The test image is taken from the line scan camera mounted on the inspection optics. The blurred image is obtained from the blurring algorithm described along with the deblurring formula. The test results show that the conventional deblurring algorithm alone does not achieve satisfactory image recovery. The deep learning algorithms alone can achieve image recovery, but the concatenated algorithm achieved speed-insensitive deblurring results.