An analytical-knowledge-based statistical method is developed to derive macromodels for the highly nonlinear A.C. response functions of CMOS Op-amp circuits. Simple circuit analysis is used to motivate the use of logarithmic transformations. By this transformation, linear models are fitted over a large domain of the independent variables (typically, transistor widths) with very few circuit simulations. From circuit analysis it is deduced that poles and zeroes are appropriate for empirical modeling instead of directly modeling the response functions like Unity Gain Frequency or Phase Margin. We have demonstrated the usefulness of the proposed method by using it to model the transferfunction of a two-stage compensated CMOS Op-amp circuit. The model is compared with SPICE3 simulations and shown to have excellent predictive capability, with a speed-up of more than 500.
Pradip Mandal, V. Visvanathan