25.3 Scalable Trajectory Methods for On-Demand Analog Macromodel Extraction Saurabh K Tiwary Rob A Rutenbar Carnegie Mellon University Pittsburgh,PA USA Carnegie Mellon University Pittsburgh,PA USA stiwary@ece.cmu.edu rutenbar@ece.cmu.edu ABSTRACT of some target circuit, and are fast enough to support full system simulation. Today, the tools we have to create such macromodels are extremely ad hoc. The most common strategy parameterizes a simple circuit template via curve-fitting to match relevant behaviors of the target circuit. Unfortunately, just as analog circuits themselves are most often created by experts, so too are their macromodels. Indeed, the larger problem is that for any given circuit–and especially custom circuits–we often lack a suitable template, fitting recipe, or modeling expert. In an ideal world, we should be able to extract macromodels on demand, as needed. The increasing number of mixed-signal designs only magnifies this problem. The essential difficulty is that we seek reduced models of nonlinear behaviors. We have today a rigorous foundation for reduced order linear modeling [4] [5]. However, we lack any unified theory for the general nonlinear case, although there is promising work for important sub-problems, e.g., Volterra models of weakly nonlinear behavior [6] [7] [8]. In this paper, we focus on a class of macromodels called trajectory methods [9] [10] [11] [12] [13]. Trajectory methods sample the state trajectory of a circuit as it is ...
Purchase document to see full attachment