A Study Of GPU-based FDTD Method And Its Application In LED Performance Optimization

Nowadays, LEDs have been widely used for the advantages such as small volume, energy conservation and long life, the luminous efficiency of which has been greatly improved with the technology improvement. In present situation, while blue and red LED's internal quantum efficiency can reach up to 80%, extraction efficiency is still very low and becomes a bottleneck to get higher efficiency. Therefore, it is essential to improve extraction efficiency in order to enhance performance and expand application fields of LED.In current researches, many approaches have been proposed and attempted to overcome the poor extraction efficiency, such as photonic crystals, roughened semiconductor surfaces, patterned substrates and flip-chip packaging. However, most researches mainly focus on preparation and experiment, analyses of influence mechanism on extraction are relatively few. As LED's complicated model and relatively large size compared with wavelength, there are many difficulties for traditional electromagnetic simulation software to calculate and analyze LED. Therefore, a high-speed platform for LED simulation is needed.Powerful in data processing and high-speed in calculating, the General Purpose Graphics Processing Units (GPGPU) based on Computer Unified Device Architecture (CUDA) is an ideal platform to meet our demand for highspeed calculating in LED analyzing, which can apply GPU's ability of parallel computing in general purpose calculating.This paper aimed to realize parallel FDTD algorithm based-on CUDA and use it to analyze the extraction efficiency in LED simulation. On the high-speed platform, photonic crystal structures were optimized, and the influence of parameters was analyzed. First, the luminescence principle of LEDs and photoelectric properties were briefly introduced. The reason for poor extraction efficiency was analyzed, the optimization methods and achievements were presented. The numerical method FDTD and the Convolutionary Perfectly Matched Layer(CPML) absorbing boundary was discussed in detail. Then General Purpose Graphics Processing Units (GPGPU) technique and CUDA model was introduced. On the basis of LED theory, FDTD algorithm and CUDA architecture, the main work and achievements of the researches are as following:1. Researches on the implementation and optimization of parallel FDTD algorithm based-on CUDA. In the heterogeneous computing system, task of GPU and CPU was designed in consideration of FDTD's algorithm feature. Then parallel FDTD algorithms of one-dimensional, two-dimensional and three-dimensional were implemented under the mapping between calculation task in different dimension and the structure model of CUDA thread. Based on the property of software and hardware in CUDA, the procedure was further optimized, which improves algorithm's parallelism and utilization of computing resources. As a result, the accelerating ratio can reach more than 25 compared with the performance of CPU, which provides a high-speed calculating platform for the study of LED extraction efficiency.To imitate electromagnetic computing in infinite space, CPML absorbing boundary was adopted and simplified in the implemented parallel FDTD of two-dimensional and three-dimensional. In the field of time-harmonic point source and incidence of plane wave, the boundary can absorb outgoing wave well, which can be used to assist application of FDTD in practical electromagnetic problems like LED.2. Modeled and simulated LED on the implemented algorithm platform, investigated the influence of photonic crystal on the light extraction efficiency of GaN-based LED. A method of calculating extraction efficiency was proposed on the basis of Poynting's theorem. In the FDTD method, overall emitted energy and the energy extracted from the structure was calculated by integrating energy-flux density in time and space, which was verified by compare the analytical solution with numerical solution of the time-harmonic electric dipole's average radiation power.By calculating the extraction efficiency of LED containing surface photonic crystal and embedded photonic crystals of different parameters, the influence of several parameters of triangular lattice photonic crystals with cylinder holes was scanned and analyzed, including filling medium, lattice period, filling ratio, height and position. With the optimized parameters, the surface photonic crystal and embedded photonic crystal improve the extraction efficiency by 184% and 179% respectively.The work and conclusions above provides a high-speed platform for LED electromagnetic analyzing, which is of practical significance for application of FDTD. In the simulation of LED,the calculation and improvement of extraction efficiency is instructive in optimization of LED performance.