ACCELERATION TECHNIQUE OF FDTD MODEL WITH HIGH ACCURACY FOR NANOSTRUCTURE PHOTONICS

Abstract

To accurately model nanophotonic structures, a conformal dispersive flnite difierence time domain (FDTD) method based on an efiective permittivity technique is presented, which can describe exactly the behaviors of evanescent waves in the vicinity of curved interface. A mismatch between the numerical permittivity and the analytical value introduced by the discretization in FDTD is demonstrated, thus, very flne time-step size is always necessary for nanostructures modelling, which greatly increases the required overheads of CPU time as compared to usual FDTD simulations. To resolve this problem, the performance of parallel FDTD code is investigated on a Gigabit Ethernet, and the acceleration technique for parallel FDTD algorithm is presented, which is developed by means of the replicating computation based on overlapping grids, the OpenMP multithreading technique and the vectorization based on SSE instruction. The comparison of relevant numerical results shows that these methods are able to reduce the expense of the system communications and enhance the utilization ratio of the CPU efiectively, which improves greatly the performance of parallel FDTD with high time-consuming.