The rapid advancement in high-performance electronic components has significantly increased current density in ball grid array solder joints, often surpassing critical thresholds and intensifying the effects of electromigration (EM). EM, driven by atomic diffusion under high electric current, causes structural defects such as voids, hillocks, and cracks, ultimately compromising the reliability and longevity of electronic devices. Geometrical variations in solder joint size, shape, and thickness exacerbate these effects by creating uneven current density, which accelerates defect formation, further intensifies EM, and adversely impacts reliability.This paper introduces a novel framework that integrates a machine learning-based surrogate model with conformal inference to predict solder joint lifespan under electromigration. Physics-based computational modeling generates training samples, while conformal inference addresses geometrical uncertainties and quantifies prediction uncertainty. The framework’s flexibility enables the integration of various predictive models, including polynomial regression, K-Nearest Neighbors, and random forests, providing a uniform basis for performance comparison. This approach facilitates the identification of the most accurate and robust model for predicting solder joint lifespan. The results demonstrate that this innovative framework delivers accurate and reliable predictions, effectively addressing challenges posed by electromigration. Additionally, conformal inference’s ability to quantify uncertainty provides valuable insights for reliability assessment and decision-making in electronic device design and operation.

• 2 papers to be published