-
Design, Automation and Test in Europe, DATE 2011, Grenoble, France, March 14-18, 2011; 01/2011
-
[show abstract]
[hide abstract]
ABSTRACT: Monte Carlo methods and simulation are often used to estimate the mean, variance, and higher order statistical moments of circuit properties like delay and power. The main issues with Monte Carlo methods are the required long run time and the need for prior detailed knowledge of the distribution of the variations. Additionally, most of available circuit simulation tools can run Monte Carlo analysis for Gaussian, lognormal and uniform distribution only. In this paper, in order to estimate these statistical moments, we propose a new method based on a uniform sampling technique and a weighted sample estimator. The proposed method needs significantly fewer simulation runs, and does not need detailed prior knowledge of the variation distributions. Furthermore, it can be used for any type of probability distribution irrespective of the circuit simulation tool used for the analysis. The results obtained show that the proposed method needs 100× fewer simulations iterations than Monte Carlo runs for accurate moments estimation of delay and power for standard cells in 45 nm and 32 nm technologies.
Journal of Low Power Electronics 11/2010; 6(4):578-587.
-
J. Low Power Electronics. 01/2010; 6:578-587.
-
[show abstract]
[hide abstract]
ABSTRACT: Monte Carlo methods and simulation are often used to estimate the mean, variance, and higher order statistical moments of signal properties like delay and slew. The main issues with Monte Carlo methods are the required long run time and the need for prior detailed knowledge of the distribution of the variations. Additionally, most of available circuit simulation tools can run Monte Carlo analysis for Gaussian, lognormal and uniform distribution only. In this paper, in order to estimate these statistical moments, we propose a new method based on the uniform sampling technique and weighted sample estimator. The proposed method needs significantly less simulation runs, and does not need detailed prior knowledge of the variation distributions. Furthermore, it can be used for any type of probability distribution irrespective of the circuit simulation tool used for the analysis. The results obtained shows that the proposed method needs 100× fewer simulations iterations than Monte Carlo runs for the moments estimation of the delay for standard cells in 45nm and 32nm technologies.
