Adaptive Machine Learning-Enabled Evolutionary Optimization for Reliability-Based Design of Through Silicon Via (TSV) Structures Under Uncertainty

Abstract

Through silicon via (TSV) technology has been widely employed as a promising 3-D packaging technology to achieve significant reduction in device dimensions. Due to the existence of uncertainty in device dimension and material properties, significant thermal stress can be generated in TSV to detartrate the performance of TSV-based 3-D chips. This article presents an adaptive machine-learning-enabled evolutionary optimization approach for the reliability-based design of TSV structures under uncertainty. In detail, a finite element model is developed for TSV structures under thermal cycling loads to determine its thermomechanical performance. A Kriging model is then utilized to establish as a surrogate to predict the maximum thermal stress. With the surrogate model, an adaptive machine-learning-enabled efficient evolutionary optimization (aMLEO) approach is proposed to reduce the volume of TSV structures while enhancing their reliability.