Hardware Accelerated Design of Millimeter Wave Antireflective Surfaces: A Comparison of Field-Programmable Gate Array (FPGA) and Graphics Processing Unit (GPU) Implementations

Engineered materials that demonstrate a specific response to electromagnetic energy incident on them in antenna and radio frequency component design applications are in high demand due to both military and commercial needs. The design of such engineered materials typically requires numerically intensive computations to simulate their behavior as they may have electrically small features on a large area or often the overall system performance is required, which means modeling the entire integrated system. Furthermore, to achieve an optimal performance these simulations need to be run many times until a desired solution is achieved, presenting a major hindrance in arriving at a feasible solution in a reasonable amount of time. One example of such applications is the design of antireflective (AR) surfaces at millimeter wave frequencies, which often involves sub-wavelength gratings in an electrically large multilayer structure. This paper investigates the use of field-programmable gate arrays (FPGAs) and graphics processing units (GPUs) as coprocessors to the CPU in order to expedite the computation time. Preliminary results show that the hardware implementation (100 MHz) on Xilinx Virtex4LX200 FPGA is able to outperform a single-thread software implementation on Intel Itanium 2 processor (1.66 GHz) by 20 folds. However, the performance of the FPGA implementation lags behind the single-thread implementation on a modern Xeon (2.26 GHz) by 3.6x. On the other hand, modern GPUs demonstrate an evident advantage over both CPU and FPGA by achieving 20x speedup than the Xeon processor.