Reliability-constrained die stacking order in 3DICs under manufacturing variability

Abstract

Abstract-3D integrated circuits (3DICs) with through-silicon vias (TSVs) are an important direction for semiconductor-based products and "More than Moore" scaling. However, 3DICs bring simultaneous challenges of reliability (power and temperature in stacks of thinned die) as well as variability (performance and power) in advanced technology nodes. In this paper, we study variability-reliability interactions and optimizations in 3DICs. Initial motivating studies show that in the presence of manufacturing variability, different die stacking orders can lead to as much as 2 years (∼44%) difference in MTTF of a 3DIC stack. We study MTTF-driven die-stacking optimization with consideration of variability, and propose a "rule-of-thumb" guideline for stacking optimization to improve peak temperature as well as reliability in 3DICs. We also propose integer-linear programming (ILP) methods for reliability-driven die-stacking optimization. Our methods can achieve ∼7% and ∼28% improvement in average and minimum MTTF, respectively, of 3DICs; we also achieve ∼3% improvement in performance under fixed reliability constraints. Our stacking optimizations can help improve 3DIC product yields under reliability requirements. Our research also yields the notable observation that a limited amount of manufacturing variation can "help" improve 3DIC product reliability when die-stacking optimization is applied.