Acceleration of finite-difference time-domain electromagnetic simulations using graphics processor units

Abstract

The Finite-Difference Time-Domain (FDTD) method is used extensively in microwave engineering and optics for Electromagnetic (EM) simulations. However, FDTD runs too slowly for some simulations to be practical, especially when run on standard desktop computers, but even when run on clusters of computers or supercomputers. The suitability of Graphics Processor Units (GPUs) for the acceleration of FDTD has been investigated. It is demonstrated that consumer GPUs can be used to accelerate two-dimensional FDTD simulations by a factor of roughly seven, relative to compiler-optimized code running on an Intel CPU of similar technology generation. In order to demonstrate this acceleration, an off-the-shelf GPU has been programmed to solve three 2-dimensional electromagnetic problems: (i) a cavity resonator; (ii) a band­gap structure with periodic boundaries; and (iii) a waveguide with ceramic fingers and Perfectly Matched Layer (PML) boundaries. OpenGL is the Application Programming Interface (API) used to program the GPU.