Parallel Research And Application Of Coherent Time Domain Finite Difference Method

FDTD has been rapidly developed and widely used in electromagnetic engineering field. But the dependence on the computing resources is more serious. Especially to simulate the large and complex structures, the memory and computation time can’t be competent for the single machine. The parallel technique can be used to analyze the same problem with multiple processors. The memory of existing servers is relatively large, which can meet the computing demand. The FDTD is naturally parallel because of its properties.The conformal FDTD has higher accuracy than conventional FDTD, and retains the FDTD’s advantages such as simple and easy to program. Hence, it has a wide range of applications. Firstly, Numerical results have verified the advantage of CFDTD with metal conformal and dielectric conformal technique. The Runge-Kutta index differencing technique has been joined in the program to analyze the non-magnetized plasma. On the algorithm platform, the MPI technology was used to achieve parallel CFDTD program. The divisions of computing domain, the pre-treatment with octree and the parallel iteration have been introduced detailedly. Numerical results have verified the correctness and validity of parallel programs, and some complex engineering analysis models in different fields have shown the significance of MPI-based parallel algorithm.With the obvious advantage of Graphic Processing Unit (GPU), its application has obtained more and more attention. The FDTD algorithm acceleration with GPU is a hot research direction currently. This paper describes two different GPU-based acceleration techniques used on CFDTD algorithm. The first is based on OpenACC GPU acceleration technology. Its principles and application have been detailed. Its main feature is user-friendly to accelerate the process with a large number of pre-compiled commands, but the accelerated effect should be further improved. Another platform is based on GPU CUDA acceleration technology. It has been applied to CFDTD program successfully. According to the characteristics of each memory on GPU, the optimization of data access has been proposed. The full advantage of CUDA platform has been taken to improve the acceleration ratio. The numerical examples have been given to reflect the advantages of GPU acceleration and its application prospects.Finally, the conclusions and the future efforts have been summarized.