X-ADAPT: AI-driven design-based strategy to address x-ray compatibility challenges in advanced packaging metrology

Advanced packaging technologies present metrology challenges due to their intricate, high-density structures. While x-ray inspection offers high-resolution, non-destructive capabilities, it faces compatibility issues due to noise that obscures defects. We propose the X-ADAPT framework that leverages IC design data and an AI-based probabilistic classifier to address these challenges. By evaluating the likelihood of noise susceptibility for specific design regions, X-ADAPT identifies and prioritizes critical areas for targeted inspection. To validate this approach, we present an experimental setup demonstrating the application of X-ADAPT, along with a case study where synthetically generated data is used to predict x-ray noise with different AI models. This evaluation highlights the framework’s effectiveness, its predictive capabilities, and how the resulting noise estimation guides inspection strategies. Furthermore, we discuss the advantages, challenges, and future research directions for integrating AI-driven noise prediction in x-ray metrology for advanced packaging. Our findings demonstrate that X-ADAPT enhances defect detection, reduces inspection inefficiencies, and improves yield, offering a robust solution to x-ray imaging limitations in semiconductor manufacturing.