Second-Order Arnoldi Reduction Using Weighted Gaussian Kernel

Abstract

Modeling and design of on-chip interconnect continue to be a fundamental roadblock for high-speed electronics. The continuous scaling of devices and on-chip interconnects generates self and mutual inductances, resulting in generating second-order dynamical systems. The model order reduction is an essential part of any modern computer-aided design tool for prefabrication verification in the design of on-chip components and interconnects. The existing second-order reduction methods use expensive matrix inversion to generate orthogonal projection matrices and often do not preserve the stability and passivity of the original system. In this work, a second-order Arnoldi reduction method is proposed, which selectively picks the interpolation points weighted with a Gaussian kernel in the given range of frequencies of interest to generate the projection matrix. The proposed method ensures stability and passivity of the reduced-order model over the desired frequency range. The simulation results show that the combination of multi-shift points weighted with Gaussian kernel and frequency selective projection dynamically generates optimal results with better accuracy and numerical stability compared to existing reduction techniques.