MORCIC: Model Order Reduction Techniques for Electromagnetic Models of Integrated Circuits
Dimitrios Garyfallou, Athanasios Stefanou, Christos Giamouzis, Moschos Antoniadis, Georgios Chararas, Konstantinos Chatzis, Dimitris Samaras, Rafaela Themeli, Anastasios Michailidis, Vasiliki Gogolou, Nikos Zachos, Nestor Evmorfopoulos, Thomas Noulis, Vasilis F. Pavlidis, Alkiviadis Hatzopoulos, Elpida Chatzineofytou, Yiannis Moisiadis
TL;DR
This work tackles the scalability challenge of model order reduction for large-scale electromagnetic $RLC_k$ networks in IC design. It introduces MORCIC, which combines balanced truncation with a low-rank extension based on the Extended Krylov Subspace to efficiently solve Lyapunov equations and produce compact multi-port ROMs. The approach achieves substantial ROM size reductions (average $×3.1$, up to $×5.5$) while preserving accuracy in S-parameter predictions, facilitating fast post-layout simulations for million-element models. The results indicate strong potential for industrial adoption, enabling faster design cycles and reduced memory requirements without sacrificing essential circuit behavior.
Abstract
Model order reduction (MOR) is crucial for the design process of integrated circuits. Specifically, the vast amount of passive RLCk elements in electromagnetic models extracted from physical layouts exacerbates the extraction time, the storage requirements, and, most critically, the post-layout simulation time of the analyzed circuits. The MORCIC project aims to overcome this problem by proposing new MOR techniques that perform better than commercial tools. Experimental evaluation on several analog and mixed-signal circuits with millions of elements indicates that the proposed methods lead to x5.5 smaller ROMs while maintaining similar accuracy compared to golden ROMs provided by ANSYS RaptorX.