基于GPU加速及改进谱FDTD方法的周期性结构分析
Analysis of Periodic Structures Using Improved Spectral FDTD and GPU Computing
-
摘要: 为解决作为分析周期性结构的时域数值算法本身在计算单角度入射时依然存在效率偏低的这一问题,提出了一种改进的谱FDTD方法,并运用图形处理器(GPU)对算法进行硬件加速. 改进的算法在保证单频点运算结果精确的前提下,通过降低单次运算对运算结果频谱分辨率的要求以降低总体的运算时间. 算例验证表明,在保证同等精度的前提下,改进后的算法将单角度斜入射问题的计算效率提高了1倍以上,并在此基础上通过GPU硬件加速成功实现了20倍以上的加速比,这证明了GPU加速的改进谱FDTD法的可行性与高效性.Abstract: Due to the features of easy implementation and high-efficiency of spectral FDTD, it has been successfully used in periodic structure simulation. To deal with the low efficiency issue in the case of off-normal single angle incident, an improved spectral FDTD is proposed, together with the implementation of GPU acceleration. The improved method reduces overall runtime by modifying spectral resolution requirement of each simulation run while ensuring the accuracy of each frequency response. As is verified in the simulation example, the improved spectral FDTD is at least twice as fast as the original scheme dealing with off-normal single angle incident case, while GPU acceleration offers another 20-fold speed up, that proves the high efficiency of the improved FDTD method with the use of GPU acceleration.
下载: