Parallel Model Order Reduction for Sparse Electromagnetic/Circuit Models

This paper describes a parallel Model Order Reduction (MOR) technique for Linear Time Invariant (LTI) electromagnetic/circuit systems with sparse structure. The multi-point Krylov-subspace projection method is adopted as framework for the model order reduction and a parallelization strategy is proposed. More specifically, a multi-point version of the well-known PRIMA algorithm is proposed, which is parallelized with respect to the computation of the error between the original model and the reduced one. The number of moments to be matched for any expansion point is chosen adaptively as well. The numerical results show that the proposed parallelized MOR algorithm is able to preserve the accuracy of the reduced models while providing a significant compression and a satisfactory speedup with respect to the sequential one.