Robust Inspection of Integrated Circuit Substrates Based on Twin Network With Image Transform and Suppression Modules
- 주제(키워드) Deep learning , defect detection , packaging , integrated circuit substrate , transform module , attention module , printed circuit board , reference comparison , siamese network , twin network
- 주제(기타) Computer Science, Information Systems; Engineering, Electrical & Electronic; Telecommunications
- 설명문(일반) [Choi, Eunjeong; Kim, Jeongtae] Ewha Womans Univ, Dept Elect & Elect Engn, Seoul 03760, South Korea; [Kim, Jeongtae] Ewha Womans Univ, Grad Program Smart Factory, Seoul 03760, South Korea
- 등재 SCIE, SCOPUS
- OA유형 gold
- 발행기관 IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
- 발행년도 2023
- 총서유형 Journal
- URI http://www.dcollection.net/handler/ewha/000000211326
- 본문언어 영어
- Published As https://doi.org/10.1109/ACCESS.2023.3290914
초록/요약
Because existing IC substrate inspection methods do not utilize information in the design file, those are prone to failing detection of critical defects such as missing patterns. To remedy the problem, we propose a novel twin network-based inspection system for integrated circuit (IC) substrates that compares the design file (i.e., a Gerber image) with a test image to be inspected. The proposed method is composed of an image transform module and an image comparison block. The image transform module transforms a Gerber image into an image that has similar characteristics to the test image. Without the transform module, many false positives may occur because the characteristics of the Gerber and test images such as noise, color, and pattern thickness are different. To compare the transformed Gerber image with the test image, we propose a twin network-based image comparison block with a feature suppression module that suppresses features from regions where defects do not exist while emphasizing features from defective regions. We confirmed the performance of the proposed method in comparison with existing methods using a real-world IC substrate dataset. Within the experiments, the proposed method achieved significantly improved performance from the existing inspection methods.
more