Variation-aware multimetric optimization during gate sizing

Abstract

The aggressive scaling of technology has not only accentuated the effects of intradie parametric variations in devices, but it has also impacted the effects of optimizing a certain performance metric on the optimality of other metrics. Thus, there is a need for optimization methods that can perform the simultaneous optimization of multiple metrics considering the effects of process variations. In this article, a novel variation-aware gate sizing framework has been developed that can perform simultaneous optimization of multiple performance metrics. In this framework, the relationships between the optimization metrics (like dynamic power, leakage power, and crosstalk noise) are modeled as a function of the gate sizes in the objective function. The delay values obtained from unconstrained delay optimization and the noise margins derived from coupling capacitance information form the constraints for the multimetric optimization problem. As an abstract framework, it is independent of the type of mathematical programming approach as well as the metrics chosen to be optimized. The framework has been implemented using a mathematical programming approach and has been tested on ITC'99 benchmarks for different combinations of multimetric and single-metric optimizations of delay, dynamic power, leakage power, and crosstalk noise. The results indicate that the framework identifies good solution points, and is efficient for postlayout optimization via gate sizing.