On-the-Fly Fidelity Assessment for Trajectory-Based Circuit Macromodels

Abstract

Trajectory methods offer an attractive methodology for automated extraction of macromodels from a set of training simulations. A pervasive concern with models based on regression is the lack of certainty about where they fit correctly. The authors show how the unique structure of a scalable trajectory model allows it to monitor the "fidelity" of the fit automatically, and flag where additional model training is warranted. Experimental results demonstrate this self-monitoring ability in practical circuit examples