Articles written in Sadhana
Volume 46 All articles Published: 16 April 2021 Article ID 0080
Metamodels are widely used in design optimization of analog ICs. Optimization of these ICs to meet required performance specifications is computationally intensive when circuit simulators are involved.Metamodels can be used to replace simulators to determine IC performance specifications and to increase optimization efficiency while maintaining accuracy. Metamodel accuracy is largely dependent on the type of metamodel, and the computer experimental design used to generate it. In this work, two popular metamodeling types are investigated: the classical Response Surface Models (RSMs), and Kriging metamodels which have recently gained popularity. Kriging metamodels are generated from Latin hypercube (LHC) designs in many cases. The techniques of this work are compared using analog ICs examples. The objective of this work is to renew interest in the classical RSMs and demonstrate that better metamodel accuracy is achievable at the same cost as Kriging metamodels, provided that minimum bias designs (MBDs) are used in RSMs derivation.