Multi-GPU-accelerated FDTD method for PhC fibers characterization

In the work, we have studied the PhC fiber dispersion characteristics by means of the FDTD technique. The method has been accelerated using CUDA-compatible GPU cards and its stability was compared with classical CPU computing. In contrast to other numerical methods, the FDTD allows studying the beam propagation along the fiber taking into account material dispersion and the nonlinearity. To improve the method performance, we have concentrated attention on the method details that allowed reduce dramatically the computation time and achieve the accuracy close to the one provided by the finite elements method.