Techniques for implementation of the FDTD method on a CM-5 parallel computer

In this paper, techniques for efficiently parallelizing the FDTD method for radiation problems are presented. Methods of parallelizing the core FDTD algorithm, PML ABC, Mur´s ABCs, and the near-zone-to-far-zone transformation are discussed. A technique which makes the parallel FDTD more efficient can be applied to Berenger´s PML in a straightforward manner, and apparently makes the code for PML ABC faster. Performance of the code is shown, and the computer-time usage for various parts of the code is given. On a 32 processor CM-5, the core FDTD algorithm is 100 times faster than an existing serial code, run on a SUN SPARC-2 workstation, and the calculation of the radiation patterns for two orthogonal planes is approximately 27 times faster. The techniques can easily be extended to other cases, not included in this paper, such as FDTD codes for magnetic materials, and non-uniform-mesh FDTD codes