Drift reliability optimization in VLSI ciruit design

Abstract

Due to aggressive scaling of transistor feature size in contemporary Very Large Scale Integrated (VLSI) circuits, some wearout failure mechanisms have become much more serious than before. Design-for-Reliability is thus of significant importance. The catastrophic failure and the drift failure are the two kinds of failures caused by those failure mechanisms. To date, most reliability studies have been limited to the catastrophic failures only. However, the drift failures have started to have a significant influence on the overall circuit (or system) reliability. In this dissertation, research is presented on the methodologies and the computer tools to analyze and improve drift reliability of VLSI circuits. First, a novel generalized formulation and its algorithmic solutions for drift reliability optimization are proposed. This generalized formulation can be used in various reliability applications. The algorithmic solutions include a novel lifetime approximation algorithm, the Stochastic Approximation optimization approach, and a new random perturbation based gradient estimator. These methods have been implemented in the existing circuit optimization environment and applied to the VLSI circuit design. Then, an alternative drift reliability optimization strategy is presented. This approach is very efficient in suppressing the hot-carrier induced degradation if certain conditions are fulfilled. Its basic idea is to adjust the important designable parameters in order to reduce stress imposed on critical transistors only, which leads to the reduction of an overall performance degradation. Design of Experiments approach is used for screening important circuit parameters and identifying critical transistors. The effectiveness and applicability of the above two optimization approaches is discussed. Based on the proposed approaches, a generic software package to perform drift reliability analysis and optimization for VLSI circuits design has been implemented and added to the existing statistical IC design system called GOSSIP. As an example of practical applications for real life VLSI circuit design, performance degradation resulting from hot electron effects is studied using the package. The results show that the reliability of VLSI circuits can be improved significantly by the proposed approaches. Therefore, both the methodologies and the computer tools developed in this research can be widely used in the engineering design practice.