Related Experiment Video
Updated: Jul 24, 2025
10:14
3D Scanning Technology Bridging Microcircuits and Macroscale Brain Images in 3D Novel Embedding Overlapping Protocol
Published on: May 12, 2019
7.3K
Robust Detection, Segmentation, and Metrology of High Bandwidth Memory 3D Scans Using an Improved Semi-Supervised
Jie Wang1, Richard Chang1, Ziyuan Zhao1
1Institute for Infocomm Research (I2R), Agency for Science, Technology and Research (A*STAR), 1 Fusionopolis Way, #21-01, Connexis South Tower, Singapore 138632, Singapore.
Sensors (Basel, Switzerland)
|July 8, 2023
Summary
This study introduces advanced 3D semi-supervised learning for detecting and segmenting buried structures in X-ray scans. The new models significantly improve accuracy in object detection and semantic segmentation for semiconductor analysis.
Area of Science:
- Artificial Intelligence
- Computer Vision
- Materials Science
Background:
- 3D deep learning accelerates accuracy and processing in fields like medical imaging and robotics.
- Semi-supervised learning leverages unlabeled data to improve model performance.
Purpose of the Study:
- Develop advanced 3D models for object detection and segmentation of buried structures in X-ray semiconductor scans.
- Enhance detection and segmentation by utilizing unlabeled data through semi-supervised learning.
Main Methods:
- Employed 3D semi-supervised learning for object detection and segmentation.
- Utilized contrastive learning for data pre-selection in detection models.
- Applied a multi-scale Mean Teacher paradigm for 3D semantic segmentation.
Main Results:
- Achieved up to 16% improvement in object detection and 7.8% in semantic segmentation.
- Demonstrated accurate localization of structures, components, and void defects.
- Automated metrology package reported a mean error below 2 μm for critical features.
Conclusions:
- The proposed 3D semi-supervised learning approach offers state-of-the-art performance in semiconductor metrology.
- The method effectively enhances the analysis of buried structures and defect identification in high-resolution scans.
