Parameterized Model Order Reduction for Nonlinear Dynamical Systems

The presence of several nonlinear analog circuits and Micro-Electro-Mechanical (MEM) components in modern mixed signal System-on-Chips (SoC) makes the fully automatic synthesis and optimization of such systems an extremely challenging task. The research presented in this thesis concerns the development of techniques for generating Parameterized Reduced Order Models (PROMs) of nonlinear dynamical systems. Such reduced order models could serve as a first step towards the automatic and accurate characterization of geometrically complex components and subcircuits, eventually enabling their synthesis and optimization. This work combines elements from a non-parameterized trajectory piecewise linear method for nonlinear systems with a moment matching parameterized technique for linear systems. Exploiting these two methods one can create four different algorithms for generating PROMs of nonlinear systems. The algorithms were tested on three different systems: a MEM switch and two nonlinear analog circuits. All three examples contain distributed strong nonlinearities and possess dependence on several geometric parameters. Using the proposed algorithms, the local and global parameter-space accuracy of the reduced order models can be adjusted as desired. Models can be created which are extremely accurate over a narrow range of parameter values, as well as models which are less accurate locally but still provide adequate accuracy over a much wider range of parameter values. Thesis Supervisor: Luca Daniel Title: Assistant Professor of Electrical Engineering and Computer Science