Font Size: a A A

The Research Of Fast TEM/STEM Image Simulation Based On GPU

Posted on:2022-10-20Degree:MasterType:Thesis
Country:ChinaCandidate:P J YuanFull Text:PDF
GTID:2518306737454244Subject:Integrated circuits
Abstract/Summary:PDF Full Text Request
With the rapid development of big data and artificial intelligence(AI),the central processing unit(CPU)is not good enough in parallel computing of big data.More and more software developers begin to use the multi thread parallel computing of graphics processor(GPU)in computer hardware to obtain higher performance and efficiency,Compared with CPU,it can save several times or even dozens of times for some largescale and complex computing.In order to analyze the atomic size structure of crystal,transmission electron microscope(TEM)and image simulation technology are important and indispensable tools.In practical application,TEM images are often simulated by commercial or non-commercial software,and then the atomic structure of crystal is determined by comparing with experimental images.Scanning transmission electron microscope(stem),high resolution transmission electron microscope(HRTEM)and convergent beam electron diffraction(CBED)are the three basic imaging modes of TEM.In image simulation,the first thing to consider is the accuracy of simulation.There are many factors that affect the accuracy of image simulation,such as: when calculating the potential field,consider whether the calculated atom is neutral or ionic.In which form does thermal diffuse scattering(TDS)effect affect atomic potential field,and whether to use frozen phonon(FL)model or complex potential field.The whole image simulation process includes three parts: the simulation of atomic potential field,the simulation of outgoing wave function,and the imaging in TEM electron optical system.In addition,because the imaging mode of scanning the sample point by point is adopted in stem simulation,the calculation accuracy and efficiency in stem mode are particularly important for stem image simulation.In this paper,based on GPU multithreaded parallel computing,we developed a code to quickly generate the electric image,and developed a simple and easy-to-use client using MATLAB.The software,named totem,supports windows and Linux systems,and can quickly and easily simulate the image of transmission electron microscope.Totem includes Peng's five Gaussian functions and lobato's most accurate parameters,which can be used to calculate atomic potential field.In addition,five Gaussian potential fields are modified.More importantly,the software adopts the improved multilayer method proposed by the research group to simulate the outgoing wave.This method takes into account the three-dimensional shape of each atom and the influence of each atom's potential field on the upper and lower multilayer.The potential field is quickly distributed to the multilayer,which can accurately reflect the height information of different atoms in the same layer.In addition,the general parallel computing architecture CUDA of NVIDIA is used to develop the code of electron potential field calculation and wave propagation calculation.The GPU's built-in powerful batch(inverse)fast Fourier transform computing capability greatly speeds up the simulation speed of scanning transmission electron microscope and integral differential phase contrast(IDPC)through GPU's multi thread parallel computing mode.The super cell of Sr Ti O3 is simulated on a personal computer equipped with NVIDIA GPU,and the time of multithreading calculation by Matlab is compared.These tests verify the calculation speed and accuracy of totem,and prove its advantages in simulating large crystals.
Keywords/Search Tags:CUDA acceleration, GPU parallel, STEM, MATLAB
PDF Full Text Request
Related items