ROMJIST Volume 27, No. 2, 2024, pp. 183-195, DOI: 10.59277/ROMJIST.2024.2.05
Vasile GROSU, Emilian DAVID, Liviu GORAS, Georg PELZ On the Modelling Possibilities of Integrated Circuits Behavior Using Active Learning Principles
ABSTRACT: There are many situations in applications like circuit design, optimization or verification where the simulation time and licensing costs of simulator can become very prohibitive. Therefore, developing metamodels that would mimic circuit behavior for these applications might be highly desired since they can be used at least as a fast preliminary design tool by the engineers to speed up the development process. Efficient sampling strategies can be further employed for further reducing the simulation related costs for designing such metamodels. In this paper we propose two Active Learning sampling schemes that can be used to minimize the number of samples needed for creating reliable metamodels. We validate and compare the approaches with classical fixed sampling schemes on a set of synthetic functions, a simulated circuit and a power device.KEYWORDS: Active learning; adaptive sampling; integrated circuit design, Gaussian process regression, metamodels; neural networks regressionRead full text (pdf)