Speed efficiency optimization for GPU accelerated rigorous coupled-wave analysis program

Rigorous Coupled Wave Analysis (RCWA) method is highly efficient for the simulation of diffraction efficiency and field distribution patterns in periodic structures and textured optoelectronic devices such as VCSELs, LEDs, and DOEs. RCWA provides exact solutions provided the Fourier expansion has infinite order. In practice, the Fourier expansion must be truncated due to computer memory limitations. Researchers are trying to utilize fast convergence algorithms such as the 'normal vector method' and 'Li's rule' which could obtain accurate TM mode results with fewer Fourier orders. However, to thoroughly investigating the behavior of a structure usually requires thousand and even millions of RCWA simulations which may last hours and days. GPU is highly suitable for solutions of complex systems allowing large-scale multi-threaded parallel programming (< 1000 / low-end GPU, <5k / high-end GPU) to speed up matrix computations significantly. In this paper, we present a high-speed RCWA program utilizing optimized CUDA-GPU code and MAGMA libraries. It achieves 2-6 X speedup compared to conventional multithreaded CPU-based code utilizing the Intel MKL library running on IRIDIS 5 super-computer (1 NVIDIA v100 GPU, 40 Intel Xeon Gold 6138 2.0GHz cores CPU)