Sub-Threshold Standard Cell Sizing Methodology and Library Comparison

摘要

Scaling the voltage to the sub-threshold region is a convincing technique to achieve low power in digital circuits. The problem is that process variability severely impacts the performance of circuits operating in the sub-threshold domain. In this paper, we evaluate the sub-threshold sizing methodology of [1,2] on 40 nm and 90 nm standard cell libraries. The concept of the proposed sizing methodology consists of balancing the mean of the sub-threshold current of the equivalent N and P networks. In this paper, the equivalent N and P networks are derived based on the best and worst case transition times. The slack available in the best-case timing arc is reduced by using smaller transistors on that path, while the timing of the worst-case timing arc is improved by using bigger transistors. The optimization is done such that the overall area remains constant with regard to the area before optimization. Two sizing styles are applied, one is based on both transistor width and length tuning, and the other one is based on width tuning only. Compared to super-threshold libraries, at 0.3 V, the proposed libraries achieve 49% and 89% average cell timing improvement and 55% and 31% power delay product improvement at 40 nm and 90 nm respectively. From ITC (International Test Conference 99) benchmark circuit synthesis results, at 0.3 V the proposed library achieves up to 52% timing improvement and 53% power savings in the 40 nm technology node.