Chang, R., Wang, J., Chong, S. C., Yang, X., & Pahwa, R. S. (2025). Efficient Visual Inspection Framework of High-Bandwidth Memory Bumps with Generative and Deep Learning AI. 2025 IEEE 75th Electronic Components and Technology Conference (ECTC), 920–926. https://doi.org/10.1109/ectc51687.2025.00161
Abstract:
The process of visual inspection has been significantly improved with addition of AI with human in the loop. In this paper, we introduce new features to enhance our end–to‒end visual inspection framework for HBMs. We detect individual HBMs consisting of multiple components along with providing a full report for the scanned chip providing 3D metrology information, details on defects and possible root cause analysis for the defects. Our modular approach can integrate any object detection and segmentation deep learning model based on the inspection requirements. We introduce a multi-view approach to increase the robustness of the detection and identification framework for defective bumps. Our approach can directly process high-resolution 3D scans and generate automatically an inspection report from our generative AI model. On our HBM dataset, the defective bump identification increased by 46%. We also conduct a study on learning with less labels. With varied percentages of labeled data, we achieved segmentation accuracy of 87% on memory bumps respectively with only utilizing 25% labelled bumps for training our AI models. The Inspection reports are customized using our 3D Metrology toolbox and Large Language Models (LLMs) providing possible causes and suggestions. We then propose a fully automated system, that with one click is able to provide a comprehensive report of the entire dense 3D scan consisting of more than 400 bumps in under 5 minutes. This system shows immense promise to expedite the inspection process and ultimately be used for inline inspection with 3D scanners.
License type:
Publisher Copyright
Funding Info:
This research / project is supported by the Agency for Science, Technology and Research - Manufacturing, Trade, and Connectivity programmatic fund
Grant Reference no. : M23L7b0021