Comparison of techniques for model order reduction of frequency-dependent networks

Abstract

The constant increase in complexity and functionality of integrated circuit (IC) systems has resulted in large-scale systems that require advanced techniques for simulation at the verification stage. Signal integrity analysis has become more crucial in helping reduce time and production manufacturing costs. This is particularly critical for high-speed networks with transmission-line where crosstalk, skin effect, dispersion and jitter, often lead to malfunction and faulty products. From a circuit analysis perspective, the dynamical order of such systems is in principle infinity. Moreover, the number of ports of such systems can be in the hundreds which makes the analysis of such networks prohibitive. Model order reduction (MOR) has been used not only to reduce the size of the problem but to extract an optimum, relatively smaller system or macro-model for time-domain simulations. This paper explores two of the currently most popular techniques for performing MOR, namely, the Vector Fitting and Loewner Matrix methods. A comparison between the two methods is drawn and some examples are illustrated.