The Gpu-based Optical Waveguide Devices Fdtd Parallel Algorithm

The Finite-Difference Time-Domain (FDTD) method is a most popular numerical method for the simulation of optical waveguide devices. It was first proposed in 1966 by Yee to solve problems in electromagnetic scattering. With the increase in computer's computational power and decrease in the computer's price, it has been widely used in almost every area of computational electromagnetics, and larger and more complex problems have been solved by using the FDTD method.The FDTD method was originally proposed for electromagnetic waves with long wavelength such as RF and microwave. It can also be used for the analysis of optical devices. Due to the numerical dispersion limit in the FDTD method, the spatial discretization width is usually less than λ/10, which results in a large grid number in an electrically large device simulation. The spatial discretization width is extremely short when wavelength is of micrometer order. For large-scale optical waveguide simulations, such as simulations of three dimensional or large-scale two dimensional optical waveguide problems, extremely large computer memory space and a long computational time is required.The parallel computing techniques can be used to reduce the computation time significantly and have been widely applied in various complex FDTD applications. The current large parallel computer system no longer satisfies the demands of large optical devices simulation in the industry because the price of the system is too high. A research on a low-cost parallel FDTD computation system is needed for the FDTD application.But CPU based parallel computer system doesn't have enough power to calculate large-scale problems. Graphic Processor Units (GPU) can be used to accelerate FDTD simulation. It is demonstrated that comsumer GPU can be used to accelerate two-dimensional FDTD simulation by a factor of roughly 20, relative to compiler-optimized code running on an Intel CPU of similar technology generation. In order to demonstrate this acceleration, a GPU has been programmed to solve electromagnetic problems.A new parallel FDTD system based on GPU and the low-cost and high performance PC cluster is proposed and implemented. The boundary condition processing and the FDTD applications in this computing system are studied. Each parallel processing unit uses two dedicated Gigabyte Ethernet interface port to exchange FDTD boundary data with its neighbors. There methods are proposed to optimize the parallel algorithm of the FDTD method on Linux operation system and a PC cluster system. The methods are optimizing boundary values exchange method, making algorithm fit the characteristic of CPU cache and pipelines, and optimizing network device driver and socket parameters. A software package is developed for the parallel FDTD computation system, including numerical algorithm engine and friendly graphical user interface.Finally, a 2-dimension optical waveguide problem is simulated and analyzed using this parallel system.The simulation results show that the optimized parallel FDTD algorithm gets a very high speed up rate. The results also show that this complete FDTD computation system has good parallel performance and can be used for time-domain analysis of optical waveguide devices instead of the expensive huge parallel computer system.