| Automated nonlinear macromodeling refers to numerical techniques for extracting small macromodels of analog circuit blocks automatically from their SPICE-level descriptions. The generated macromodel can be used to replace the original circuit block for fast system-level simulations. It is expected that computer-aided design (CAD) tools for automated macromodelling will be crucial for effective design of next generation radio frequency (RF), analog and mixed-signal circuit applications. This dissertation presents a generally-applicable automated macromodelling approach based on piecewise polynomial (PWP) representations. Numerical experiments show that the PWP-generated macromodels can capture weak nonlinearities in small-signal analysis, as well as strongly nonlinear phenomena excited by large signal inputs. Significant speedups can be achieved by deploying the macromodel in system-level simulations. A novel weakly nonlinear macromodeling approach, termed the Multiple Pseudo-Inputs (MPI) method, is also developed for use as a low-level engine for PWP Important implementation details---such as smoothing the transitions between different regions for better convergence, merging multiple trajectories to improve state-space coverage and generating one unique projection basis---are addressed in this dissertation. Finally, preliminary investigations of geometric methods for nonlinear macromodeling are presented. |
