Low-rank balanced truncation of RLCk models via frequency-aware rational Krylov-based projection
Christos Giamouzis, Dimitrios Garyfallou, George I. Stamoulis, Nestor Evmorfopoulos · 2024
Model order reduction (MOR) techniques are crucial for addressing computational challenges in the simulation process of large-scale RLCk models extracted from modern integrated circuits. Specifically, projection-based balanced truncation (BT) is typically employed for producing compact reduced-order models (ROMs), even when dealing with multi-port circuits. The rational Krylov subspace (RKS) method proves powerful in constructing low-rank approximate solutions of the computationally expensive Lyapunov equations involved in BT. However, a limitation of the RKS method lies in the optimal selection of expansion points. In this paper, we propose an efficient lowrank BT method based on RKS projection with an adaptive methodology for choosing the expansion points. The proposed method targets a frequency range of interest and achieves exceptional ROM order by accelerating convergence through two effective criteria. Experimental results indicate that our approach can produce more compact and accurate ROMs faster than a conventional multi-point moment matching technique.