Authors:

Javier Guio Gomez

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

X-ray computed tomography (XCT) is a powerful tool that can provide 3D information of the internal microstructures non-destructively. Due to this, it is widely employed in the semiconductor industry to detect internal defects. XCT works on the principle of taking thousands of X-ray projections at incremental degrees of rotation and reconstructing them into a 3D dataset, which can be very time consuming. Recent advances in machine learning provide new ways to address this challenge. This project explores the use of ZEISS DeepRecon Pro, a deep neural network-based reconstruction tool, to accelerate tomographic imaging by taking fewer projections without compromising image quality. Our study focuses on internal defects in solder ball grid array samples subjected to thermal cycling. The models were trained in full-projection scans and used to reconstruct down-sampled datasets which simulate faster scans to evaluate the model's ability to provide the same level of detail. The ground-truth and model-generated 3D images were compared to assess the model's performance through both visual inspection and quantitative metrics. This study provides insights for improving the robustness and reliability of the DeepRecon Pro model which will establish a foundation for integrating AI-based reconstruction into XCT workflows, making time- efficient and high-quality imaging more accessible for dynamic studies. Keywords: [no keywords provided]

Source:

Purdue University / 2025

Topics:

No topics listed

Co-authors:

Javier Guio Gomez