Metamodeling Techniques in Analog Circuit Design

Abstract

Computational time in analog circuit design is mostly expended by simulators such as PSPICE to determine circuit performance specifications. In many classical as well as contemporary methodologies, metamodels are used as surrogates of circuit simulators to increase computational efficiency of performance specifications, and with accuracy comparable to simulation results. The accuracy of a metamodel depends on its type and the experimental design used to generate it. Two prevalent metamodel types include the classical Response Surface Models (RSMs), and the more recent and increasingly more popular Kriging metamodels. This work compares RSM and Kriging metamodels in analog circuit design, whereby we conclude that the increased popularity of the Kriging techniques may be unjustifiable‐‐provided that minimum bias experimental designs are used in RSMs generation.