Reliability issues in microelectronic systems are often closely linked to the underlying microstructural features of materials, such as grain boundaries, porosity, and phase distributions. Accurate characterization of these microstructural attributes is essential for identifying the root causes of defects like cracks, voids, and degradation of connected regions, which directly impact device performance and lifespan. Image-based methods, particularly those enhanced by machine learning (ML), offer a powerful, automated means of extracting both microstructural and defect-related information from high-resolution microscopy images. By integrating microstructure characterization into failure analysis (FA) workflows, it becomes possible to not only detect failures with higher accuracy but also gain deeper insights into the material behaviours and conditions that give rise to these reliability concerns.

This study presents an integrated workflow to analyse key microstructural features, such as phase distributions in aged intermetallic compounds and cracks in solder joints valid across multiple samples, while simultaneously identifying defects responsible for sample failure or degradation. The ML models, trained on diverse microscopy datasets, achieved detection accuracies between 88% and 92% compared to existing approaches that includes conventional image processing and manual methods in both precision and speed, with a 30x to 40x improvement in processing time. These models were integrated into an external software platform to evaluate interoperability and usability within existing digital microscopy workflows. This dual-capability system enables not only rapid failure localization but also quantitative microstructural assessment, significantly improving the efficiency and depth of analysis in microelectronic diagnostics through a unified, ML-driven pipeline.