Reduced-order modeling of high-speed channels using machine learning techniques: Partitional and hierarchical clusterings

Abstract

Two well-known machine learning techniques, partitional and hierarchical clustering techniques, are applied to simplify the order of complex rational function models of high-speed interconnect channels. In order to retain the physical features of the system, the cluster centers are biased toward the dominant poles using a different kind of distance measure, called inverse distance measure (IDM) during clustering. The proposed procedure is computationally inexpensive and numerically stabile. To illustrate the validity of the methods, examples of frequency-domain simulations of a high-speed channels are provided.