Benchmarking Parallel Computing Solutions for Rough Surface Scattering Canonical Problems

Accurate modeling of electromagnetic wave scattering from large-scale ground profiles is essential in remote sensing applications. However, the computational burden associated with the underlying numerical methods — particularly those based on full-wave solvers — can become prohibitive. To address the simulation of scattering from one-dimensional random rough surfaces, we have developed specialized parallel implementations of a Method of Moments (MoM)-based solver, optimized for various heterogeneous computing platforms, including shared-memory multicore CPUs, manycore GPUs, and distributed-memory multinode systems. For solving the resulting dense linear systems in parallel via LU factorization across different architectures, we leverage cutting-edge numerical libraries developed in the HPC research community, such as PLASMA, MAGMA, and ScaLAPACK. These libraries serve as parallelized counterparts to the well-known LAPACK suite, each tailored to exploit distinct levels of hardware parallelism. Representative case studies are presented to validate the numerical behavior of the implemented solvers across diverse computational architectures. Parallel performance is thoroughly assessed through empirical benchmarks, demonstrating significant speedup, scalability, and computational efficiency. The developed parallel prototypes effectively harness the different degrees of parallelism available, enabling faster EM simulations, evaluation of scattering phenomena at higher resolutions, and higher-fidelity modeling of electrically large surfaces. This work paves the way for the efficient computation of electromagnetic scattering from complex and/or multilayered structures, facilitating large-scale simulations on current super-computing infrastructures.