SCPlace: A Statistical Slack-Assignment Based Constructive Placer

Evriklis Kounalakis and Christos Sotiriou
FORTH-ICS, Heraklion, Crete, Greece; and University of Crete, Heraklion


Abstract

Statistical optimization algorithms must be able to optimize both mean delay and standard deviation, sigma, as they need to either improve a circuit's yield or to operate under a yield constraint. Wire delays pose a significant challenge for statistical placement, as they skew signal arrival times, and thus easily affect standard deviation. This paper presents a statistical, constructive placement algorithm, SCPlace, which is based on two novel, statistical slack assignment strategies, namely MSSA (Minimum Sigma Slack Assignment) and TSZSA (Target Sigma Zero Slack Assignment). MSSA assigns wire delays on nets so as to derive a lower sigma bound for a technology-mapped, standard-cell circuit, while TSZSA assigns wire delay bounds on nets for meeting a combined (mean, sigma) constraints. Experimental results, illustrate that SCPlacer's constructive algorithm achieves legal, routable placements, and the statistical slack assigment strategy achieves a 4.68% yield improvement average for the IWLS 2005 benchmarks over an existing commercial placement tool.