Knowledge Extraction in Sizing: Caffeine

Abstract

This chapter presents a tool to accelerate designer insight in sizing, by extracting whitebox performance models. The tool is called CAFFEINE [Mcc2005a, Mcc2005b, Mcc2005c, Mcc2005d, Mcc2006a, Mcc2006b, Mcc2006c] [Mcc2008b, Mcc2009a, Mcc2009b, Mcc2010]. CAFFEINE implements a method to automatically generate compact, interpretable symbolic performance models of analog circuits with no prior specification of an equation template. The symbolic models capture mappings of the design variables to individual performances or to Cpk (a robustness measure). CAFFEINE takes SPICE simulation data as input. This enables modeling of whatever SPICE handles: arbitrary nonlinear circuits, arbitrary circuit characteristics (including transient and noise performance measures), modern technology processes, environmental effects, and manufacturing variations. The possible expressions for the model are defined as a set of canonical form functions, structured as layers of product-of-sum terms that alternate with layers of sumof-product terms. These canonical form functions are modeled as a grammar, which is subsequently searched via grammatically-constrained genetic programming. Novel evolutionary search operators are designed to exploit the structure of the grammar. By employing multi-objective search, CAFFEINE generates a set of symbolic models which collectively provide a tradeoff between error and model complexity.

All models are wrong, but some are useful.

– George E.P. Box