Automated abstraction of nonlinear analog circuits to reliable set-valued models with reduced overapproximation

Abstract This paper tackles the analog and mixed-signal modeling for verification challenges. It proposes an adjustable automated modeling approach, which provides set-valued models with reduced overapproximation. The models reliably enclose parameter variations and modeling errors. The reduced overapproximation is obtained by computing the intersecting set of models with intervals and affine forms. The nonlinear circuit examples show a reduced overapproximation up to 86%.