AI-Based Image Enhancement Improves Detection of Interconnect Defects in Microelectronic Packaging

Abstract Traditional failure-analysis (FA) methods face growing challenges due to the increasing miniaturization and complexity of modern semiconductor devices. In particular, the detection of cracks in the extreme low K layers of flip-chips, utilized in advanced packaging architectures at the back end of line (BEOL), is challenging. To improve detectability of these so-called “white bumps”, we propose an AI-based image-enhancement workflow, specifically tailored to improve scanning acoustic microscopy (SAM) images. By applying a deep CNN with skip connections and network in network (DCSCN) model to low-resolution C-scan SAM images, we are able to improve the perceptual quality of images. Furthermore, this approach boosts resolution-dependent analysis methods like AI-based object detection of white-bump defects, improving the recall from 5.2% (95% CI: 0.1–26.0%) to 100.0% (95% CI: 82.2–100.0%). Our work demonstrates that AI-based image enhancement can significantly improve FA workflows, offering a robust and efficient solution for modern semiconductor analysis.