As process technology continues to scale into the nanoscale regime and overall system complexity increases, the reduced order modeling of on-chip interconnect plays a crucial role in characterizing VLSI system performance. In this paper, we develop a dynamic multi-point rational interpolation method based on Krylov subspace techniques to generate reduced order interconnect models that are accurate across a wide-range of frequencies. We dynamically select interpolation point by applying a cubic spline-based algorithm to detect complex regions in the system's frequency response. The results indicate that our method provides greater accuracy than techniques that apply multi-shift Krylov subspace methods with uniform interpolation points.
I. INTRODUCTIONAggressive feature scaling and increasing operating frequencies greatly impact the performance of high speed integrated circuits [1]. As the complexity of on-chip systems increases, the accurate and efficient modeling of interconnect is vital for estimating and controlling delay, cross-talk noise, and power consumption [2]. Motivated by the expanding complexity of nanoscale integrated circuits, the model order reduction (MOR) of RLC interconnect models has been the focal point of substantial research efforts over the last decade [3]- [6]. MOR provides an invaluable tool for the modeling and design of interconnect in high performance integrated circuits.The methods used for the model order reduction of interconnect fall into two main categories: singular value decomposition (SVD) methods and Krylov subspace projection methods. In SVD methods for linear systems, Hankel-norm approximation, balanced truncation, and singular perturbation methods are well known. The most popular technique is balanced approximation where the primary aim is to generate a balanced representation of the system with same degree of reachability and observability. Balanced approximation based algorithms such as PMTBR [7], [8], PR-TBR [9] and FABT [10], [11] have been proposed. However, SVD methods have high computational complexity with dense computations