AI-driven variability-aware physics-based EM simulation framework for jmax estimation

Abstract

A physics-based simulation framework of electromigration in nano-interconnects is presented. By extending our earlier model, that accounted for diffusion heterogeneity among different diffusion paths, the new framework considers metal barrier shunting and accounts for microstructure. The model is strengthened by combining it with a machine learning-based microstructure generator, enabling the incorporation of microstructural variability and its impact on the statistical distribution of EM lifetimes. The framework allows for the estimation of the maximum allowable current density (jmax) for a given technology. As case studies, it is estimated that jmax decreases by ~30% when going from 45 nm to 20 nm linewidth, while reducing the barrier resistivity by half for a fixed thickness causes a 50% increase in void growth time.