Aung, A. P. P., Lum, L. Y. X., Zee, B., Chin, J. M., Lim, Y. K., & Senthilnath, J. (2025). AI-driven pixel-level defect localization using magnetic current images. In Proceedings of EPTC 2025.
Abstract:
In semiconductor manufacturing, electrical defect detection is a tedious multi-step process. Usually, the electrical samples go through electrical testing such as (1) functional testing and curve tracing; (2) fault isolation using thermal or magnetic images; followed by (3) physical testing using X-ray; (4) material analysis and root cause analysis for a complete pipeline of failure analysis (FA) procedure. We motivate this work to do fault isolation using magnetic current images (MCI). We propose our AI-driven Pixel-level Defect Localization method (PixelDL) to localize the power short defects during the fault isolation step. The experiments with real defect samples obtained from semiconductor industry show that our PixelDL method can achieve IOU of more than 0.85 and RMSD of 2.5 pixels. Moreover, we achieved a lateral error of only 35.4um at z=100um in a focused case study.
License type:
Publisher Copyright
Funding Info:
This research / project is supported by the A*STAR - Industry Alignment Fund - Pre-Positioning: Machine Learning Guided Failure Analysis & Diagnostic Capability Development for Next Generation 3D-IC Packaging
Grant Reference no. : M23K8a0050