Parallel FDTD method for EM scattering from a rough surface with a target

The finite-difference time domain (FDTD), a numerical method, is an accurate approach to calculate the electromagnetic scattering coefficients of target and rough surface. However, it is very time-consuming to solve the problem for FDTD executing on a single CPU. We develop the CUDA-based parallel FDTD to accelerate the computation of electromagnetic scattering from a three-dimensional target and a two-dimensional rough surface. The FDTD lattices are truncated by a uniaxial perfectly matched layer (UPML), and finite difference equations are employed in the whole computation domain. Our optimization efforts on GPU-accelerated FDTD for a target and a rough surface model can be achieved at a speedup of 40.16× on NVIDIA Tesla K40c GPU.