A hybrid genetic algorithm method for optimizing analog circuits

Abstract

An approach is presented for the automated sizing of analog circuits based upon a combination of a genetic algorithm (GA) with a least squares (Gauss-Newton) gradient search. The method combines the global-search properties of the GA with the fast local convergence properties of the least squares method to produce a circuit design from random initial component values in a reduced time compared to the application of a direct GA method, or a restart least squares algorithm. Results are presented to demonstrate the application of the method in the design of both passive and active circuits.