Abstract:Model order reduction (MOR) is essential in integrated circuit design, particularly when dealing with large-scale electromagnetic models extracted from complex designs. The numerous passive elements introduced in these models pose significant challenges in the simulation process. MOR methods based on balanced truncation (BT) help address these challenges by producing compact reduced-order models (ROMs) that preserve the original model's input-output port behavior. In this work, we present an extended Krylov subspace-based BT approach with a frequency-aware convergence criterion and efficient implementation techniques for reducing large-scale models. Experimental results indicate that our method generates accurate and compact ROMs while achieving up to x22 smaller ROMs with similar accuracy compared to ANSYS RaptorX ROMs for large-scale benchmarks.

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper aims to solve the model order reduction (MOR) problem of large - scale RLCk models in integrated circuit design. Specifically, the authors propose a low - rank balanced truncation method based on the extended Krylov subspace and frequency - aware convergence criteria to address the challenges of high computational complexity and excessive memory consumption faced by traditional balanced truncation methods when dealing with large - scale models. #### Background and Problem Description 1. **Simulation Challenges of Large - Scale Electromagnetic Models** - In integrated circuit design, large - scale electromagnetic models extracted from complex designs contain a large number of passive components (such as resistors, inductors, capacitors, etc.), which makes the computational cost of simulation extremely high. - For RLCk models with millions of components and multiple ports, direct simulation is almost impossible. 2. **Limitations of Existing MOR Methods** - **Moment Matching (MM) Method**: Although efficient, it requires manual selection of the number of moments, and the size of the final reduced - order model is related to the number of moments and the number of ports, limiting its scalability. - **Balanced Truncation (BT) Method**: It provides a clear theoretical error bound and does not depend on the number of ports. However, due to the high computational complexity (O(N^3)) of solving the Lyapunov equation, the BT method is only applicable to small - scale models. #### Proposed Solutions The authors propose the following improvement measures: 1. **Low - Rank Balanced Truncation Method** - Use the Extended Krylov Subspace (EKS) method to solve the Lyapunov equation, thereby significantly reducing the computational load. - Introduce frequency - aware convergence criteria to ensure accuracy within the frequency range of interest. 2. **Efficient Implementation Techniques** - Take advantage of the different characteristics of sparse matrices and dense matrices, and use specialized data structures and numerical techniques for optimization. - Use parallel CPU and GPU acceleration techniques to improve computational efficiency. - Adopt the Bartels - Stewart algorithm to solve small - scale Lyapunov equations. #### Experimental Results The experimental results show that this method can generate a reduced - order model that is approximately 22 times smaller than ANSYS RaptorX ™ ROMs while ensuring accuracy. In addition, this method exhibits good performance on different types of circuit models, with a maximum error of less than 0.14%. ### Summary By introducing the extended Krylov subspace and frequency - aware convergence criteria, the authors have successfully solved the computational complexity and memory consumption problems of traditional balanced truncation methods when dealing with large - scale RLCk models, providing a more efficient and more compact reduced - order model generation method.

A low-rank balanced truncation approach for large-scale RLCk model order reduction based on extended Krylov subspace and a frequency-aware convergence criterion

Reduction of large-scale RLCk models via low-rank balanced truncation

A Broadband Enhanced Structure-Preserving Reduced-Order Interconnect Macromodeling Method for Large-Scale Equation Sets of Transient Interconnect Circuit Problems

An efficient finite element solution using a congruent transformation based model order reduction for interconnect simulation

An Efficient Time-Domain Trapezoidal Difference Based Model Order Reduction Method for Interconnect Circuits

MORCIC: Model Order Reduction Techniques for Electromagnetic Models of Integrated Circuits

Model Order Reduction for Large Scale RC Networks Based on Optimal Elimination

TurboMOR: an Efficient Model Order Reduction Technique for RC Networks with Many Ports

A Block Arnoldi Algorithm Based Reduced-Order Model Applied to Large-Scale Algebraic Equations of a 3-D Field Problem

An Efficient Time-domain Multi-step Integration Based Model Order Reduction Algorithm for Interconnect Circuits

Nonlinear Balanced Truncation: Part 2 -- Model Reduction on Manifolds

Review and analysis of model order reduction techniques for high-dimensional complex systems

Mixed Riccati–Lyapunov Balanced Truncation for Order Reduction of Electrical Circuit Systems

Proper Orthogonal Decomposition In The Generation Of Reduced Order Models For Interconnects

Power System Dynamic Model Reduction Based on Extended Krylov Subspace Method

Balanced truncation on empirical gramians for model-order-reduction of Non-Quasi-Static effects in MOSFETs

High Order Distribute Circuit Model of Laminated Busbars and Its Order Reduction Techniques for Simulating Fast Turn-on and Turn-off Transients

An Efficient Spectral Graph Sparsification Approach to Scalable Reduction of Large Flip-Chip Power Grids

Structure-Preserving Model Order Reduction for Thermal Analysis

Reduced-order models for array structures mounted on platforms with parameters variations

Balanced Truncation of Linear Systems with Quadratic Outputs in Limited Time and Frequency Intervals