Optimizing Scanning Acoustic Tomography Image Segmentation With Segment Anything Model for Semiconductor Devices

This paper enhances the segmentation process of Scanning Acoustic Tomography (SAT) images using SemiSA, a deep learning model fine-tuned from the Segment Anything Model (SAM), for non-destructive evaluation of semiconductor devices.

Paper

Abstract

Scanning Acoustic Tomography (SAT) has become a vital technique for characterizing semiconductor devices in non-destructive evaluation. Precise and efficient segmentation of SAT images is crucial for detecting defects and assessing material properties in the semiconductor industry. This study enhances the segmentation process of SAT images using SemiSA, fine-tuned from the Segment Anything model, trained and evaluated on a large-scale dataset from the Ohlabs TSAM-400 system encompassing various semiconductor devices. Results demonstrate that SemiSA achieves a 17.89% enhancement in Dice Similarity Coefficient scores and a 24.26% improvement in Intersection over Union scores on average across all tasks.

Key Contributions

• Fine-tuned Segment Anything Model (SemiSA) for SAT image segmentation
• Evaluated on diverse semiconductor devices: Flip Chip, Power Semiconductor, Wafer (6/12-inch), Transistor, and MLCC
• Efficient segmentation framework tailored specifically for SAT images

Recommended citation: T. T. H. Vu, T. H. Vo, T. N. Nguyen, J. Choi, S. Mondal and J. Oh, "Optimizing Scanning Acoustic Tomography Image Segmentation With Segment Anything Model for Semiconductor Devices," in *IEEE Transactions on Semiconductor Manufacturing*, vol. 37, no. 4, pp. 591-601, November 2024, doi: 10.1109/TSM.2024.3444850.