AI-Enhanced Multimodal Approaches for Electronics Metrology and Failure Analysis

The rapid growth of 3D advanced packaging introduces new challenges in inspection and failure analysis, where complex structures such as microbumps, redistribution layers (RDLs), and through-silicon vias (TSVs) demand reliable non-destructive testing (NDT). Conventional approaches, including Scanning Acoustic Microscopy (SAM) and X-ray imaging, are limited by noise, resolution, and defect visibility, creating barriers for reproducible and scalable analysis. To address these challenges, our work advances an AI-powered multimodal inspection framework that couples physics-informed machine learning with structured data infrastructure. A Physics-Informed Neural Network (PINN) approach enhances SAM imaging by embedding acoustic wave physics into reconstructions, producing higher-fidelity images validated through structural similarity and physical accuracy metrics. Complementing this, multimodal data fusion across SAM, X-ray laminography, optical microscopy, and CT establishes richer defect detection and cross-validation. Central to this effort is the creation of multimodality benchmark datasets built on standardized acquisition protocols, structured metadata schemas, and annotation pipelines. These datasets provide not only a foundation for AI model training but also enable reproducibility, traceability, and interoperability across future programs.
Speaker(s): Navid Asadi,
Virtual: https://events.vtools.ieee.org/m/498529