Putting AI to Work: A Practical and Simple Application to Improve 3D X-ray FA

This paper presents the demonstration of a deep learning-based reconstruction approach for working with 3D X-ray tomography/microscopy data, focusing on improving workflows in microelectronics failure analysis and reliability applications. Whereas the industry-standard filtered back projection (known as FDK) method of X-ray tomography reconstruction has been used for many years due to its simplicity and robustness, it has constrained the results of 3D X-ray scanning in terms of both image quality and scan speed. Powered by artificial intelligence technologies, the new deep learning high resolution reconstruction (DLHRR) approach discussed here offers broad improvements across diverse sample types including microelectronics, increases scan speed by a factor of 4X or more, is as easy to use as FDK without requirement for a machine learning expert, and is implemented on a desktop workstation PC. Results will be shown on IC packages and commercial battery devices.