The Application Of GPU Computation In STM Simulation

In rencent years, GPU (Graphics Processing Unit) has developed very rapidly. GPU has a large advantage to CPU in computation capacity and bandwidth. Using GPU, together with CPU, in large-scale data-intensive tasks has better performance and is more cost-effective than clusters and supercomputers.STM has been widely used in surface science research, with the help of STM image simulation. However, intensive STM simulation can become very time consuming. In this thesis, we focuse on accelerating STM simulation with GPU using CUDA (Computer Unifie Device Architecture), which is a GPU computation platform.We also study the NDR effect in MPc moleculars. We simulate the STS of MPc with our STM simulation method. Mechanism and application of the NDR effect are discussed.In the first chapter, we introduce the GPU hardware firstly, including the GPU basic architecture and its recent development. Then, we describe how GPU is used as a general computation tool from GPGPU to CUDA. Finally, we introduced in details about CUDA, including the basic design principles, software architecture, thread mode, memory structure, and debug principles.In the second chapter, we first talk about the working principles of STM, and then introduce several basic STM simulation methods briefly. They includes MBA (Modified Bardeen Approximation) method and the Tesoff-Hammann method which can be considered as a special case of the MBA method. In the end of the chapter, we introduce a new STM simulation method which developed from Bardeen method.In the third chapter, we first verify the correctness of our program through two testing systems. Then, we introduce the algorithm of our program, and analyze some important lines of the codes. Finally, we test the speed-up ratio of each part, and also the total acceleration of our program. In this part, we compare two implementations, one of which use FFTW, and the other use CUFFT. The final chapter is on the NDR effect. We first introduce several mechanisms of NDR, focusing on the Zhenpeng Hu's work in 2007. And then, we made some simulations based on Hu's work. We explain the simulation results, and forecast the applications of NDR. We also consider the polarization situation.