Sensitivity analysis in the optimal sizing of analog ICs by evolutionary algorithms

Abstract

A multi-parameter sensitivity approach based on Richardson extrapolation, and applied to the optimal sizing of analog integrated circuits (ICs), is presented. First, the multiobjective evolutionary algorithm (EA) called non-dominated sorting genetic algorithm (NSGA-II), is applied to compute the feasible sizes of analog ICs, i.e. the optimal width and length (W/L) of every metal-oxide-semiconductor field-effect-transistor (MOSFET) is found. At this stage, the simulation program with integrated circuits emphasis (SPICE) is used to evaluate the electrical characteristics of analog ICs. Second, the multiparameter sensitivity analysis based on Richardson extrapolation, is applied to approximate the partial derivatives associated to the sensitivities on the performances of the ICs with respect to W/L of every MOSFET. The cases of study are three analog ICs, namely: voltage follower (VF), positive-type second generation current conveyor (CCII+), and current-feedback operational amplifier (CFOA). The proposed approach selects W/L feasible sizes presenting the lower sensitivities that are computed from the corresponding Pareto sets. Finally, 18 feasible (W/L sizes) solutions accomplishing 18 performance objectives and guaranteeing low W/L sensitivities for a complementary metal-oxide-semiconductor (CMOS) CFOA, are listed in Table I.