Model-Order Reduction of Massively Coupled Parameterized Systems via Clustering

Parametrized Model Order Reduction(PMOR) methods reduce the full equations for large parametric networks into a reduced number of parametrized equations. The reduced model is obtained to match the variations in the response of the original full circuit due to variations in its key design parameters. Applying existing PMOR techniques to parametrized systems with many inputs often results in extremely large and dense reduced-order models. This thesis presents a new approach to construct parametrized reduced-order models for multi-parameter networks with many input/output terminals. The new method leverages the efficient application of the multi-parameter moment-matching-based approach by exploiting the superposition paradigm. Solving the resulting overall reduced system to obtain variations in the original circuit response yields significant computational savings. The presented algorithm is an effective means to manage the Monte Carlo simulations of large multiport parametrized linear networks. It is highly suited for multithreading implementation and thus facilitates parallel time-domain variability analysis.