Accelerating diffractive optics design with GPU-based parallel technique

In this paper GPU based data parallel technique is used to reduce calculating time in diffractive optics elements (DOE) design. By using compute unified device architecture (CUDA) framework, comparison of acceleration radio of some major methods using in DOE design, including Gerchberg Saxton algorithm for global optimization in conjunction with Rayleigh-Sommerfeld diffractive integrating transformation, genetic algorithm with a finite-difference time-domain method and simulated annealing algorithm, is made with demonstrational DOE-lens design, whose focal spot size is 0.5 μm, incident light wavelength is 650 nm, focal length is 3mm, DOE aperture size is 2mm.