Modeling circuit performance functions by a combination of physical models and black-box approximation

Two novel modeling architectures are proposed for combining physically based analytical circuit performance function models with 'black-box' approximations. The black-box model part provides only the error correction to the 'physical part,' while the physical model part provides the physical insight normally lacking in black-box models. Model Structure I is an extension to the Composite MFQI/GMDH architecture. It generates complex but very accurate models. Model Structure II generates simpler ('parsimonious') models with more physical 'explaining power.' Circuit designers can choose between the two structures. Physical models of insufficient accuracy can be automatically eliminated, if so desired.< >