Geostatistical-Inspired Metamodeling and Optimization of Nano-CMOS Circuits

With the continuous progression of semiconductor technology, nanoscale effects have become a persistent issue in the design of analog/mixed-signal (AMS) circuits. The cost of exploration and optimization of the design space increases to infeasible levels with conventional design methodologies. Different modeling techniques to reduce the cost of design exploration, while ensuring the accuracy of such models, have been introduced and continue to be a research problem. In this paper, a geostatistical inspired metamodeling and optimization technique is presented for fast and accurate design optimization of nano-CMOS circuits. The proposed design methodology incorporates a simple Kriging based metamodel which efficiently and accurately predicts design performance. The metamodel (instead of the circuit netlist) is subjected to a Gravitational Search Algorithm for optimization. This design methodology is applicable to AMS circuits and is illustrated with the optimization of power consumption of a 45nm CMOS thermal sensor. The method improves the power performance of the thermal sensor by 36.9% while reducing the design optimization time by 90% even with 6 design parameters.