A Study On New Algorithms Of Electron Exit-wave-function Reconstruction Technique

Through focus exit wave reconstruction techniques are usually implemented by recording a series of images under different defocus and using linear or nonlinear methods to reconstruct the exit wave.Through focus exit wave reconstruction techniques can effectively increase the point resolution of the electron microscope to the resolution limit.However,there are still problems in the through focus exit wave reconstruction techniques.First of all,compared with the traditional imaging mode in TEM,whether the reconstructed phase has advantages in monotonic imaging of sample thickness is worth discussing,which is of great significance for the quantitative analysis of the reconstructed results.In addition,residual aberrations have always affected the perfo rmance of through focus exit wave reconstruction techniques in practical applications,and a method that can be applied to crystals and can automatically correct low-order and highorder aberrations is needed.Finally,as a powerful optimization algorithm,whether the neural network algorithm can help the through focus exit wave reconstruction techniques to obtain more accurate results has also attracted our attention.In view of above problems,we will discuss from three aspects in this thesis: the monotonicity of different atomic imaging signals in TEM,us ing crystals to correct the residual aberrations in the wave function reconstruction,and the implement of through focus exit wave reconstruction using neural network.Based on the scientific research on the above issues through theoretical analysis,numerical simulation and experimental design,this thesis draws the following innovative conclusions:(1)After comparing the high-resolution image and the phase image of the exit wave through simulation and experimental data,it is found that for light atoms(Al),the monotonic interval width of the high-resolution image is only14.58 nm,but the monotonic interval width of the phase image can reach 40.50 nm.For heavy atoms(Au),the monotonic interval width of the high-resolution image is 2.46 nm,but the monotonic interval width of the phase image is 3.67 nm.Therefore,for either heavy atoms or light atoms,the phase of the exit wave can change monotonously within a larger samp le thickness and keep showing true position and shape of atoms within a larger thickness range(30 nm).Nonlinear information will introduce fake spots to the high-resolution image and affect the image contrast when the thickness reaches 5 nm,which will make it difficult to explain the contrast of the high-resolution image.(2)Based on the minimum amplitude contrast criterion,a simple and robust method for measuring up to second-order residual aberrations in the reconstructed exit waves using the crystalline samples was developed."Error accumulation" is a problem often encountered in the process of multi-parameter optimization.Considering that low-order aberrations and high-order aberrations have different effects on image contrast,this method effectively avoids the adverse effects of "error accumulation" by correcting the low-order(defocus,second-order astigmatism)and high-order aberrations(third-order astigmatism,coma)step by step.In addition,this method can correct residual aberrations more effectively when the crystal structure is relatively complex(orthogonal,monoclinic,triclinic crystal system),the image resolution is high and the sample thickness is small.(3)Combining the idea of iterative wave function reconstruction with neural network algorithm,a neural network iterative wave function reconstruction algorithm was developed(IWFR-NN).Unlike other neural networks,the network built in this study converges very quickly,reaching converg ence after about 100 cycles.From the simulation data,it is found that the method can reconstruct the exit wave with a small number of images(less than 10 images).And it can "resist" the influence of image noise,image drift,and offset of the defocus on the reconstruction result.For the experimental data,the phase image obtained by IWFR-NN can still reflect the atomic structure information and is consistent with the information reflected by RIWFR.But the results of the two have a certain difference in image contrast and value.