Synthesis Of Two-Dimensional Transition Metal Dichalcogenides And Machine Learning Based Fast Characterization

Two-dimensional materials refer to a class of nanomaterials that are composed of single or a few atomic layers,and the size is reduced to the limit in one dimension.Accompanied by the successful preparation of graphene by mechanical exfoliation in2004,the exploration of the two-dimensional material family has continued to deepen.The preparation and characterization of two-dimensional materials are two important topics in this field.This thesis is developed along these two research lines.In terms of preparation,chemical vapor deposition(CVD)is a reliable method that is expected to be applied to large-scale production of two-dimensional materials.It has the advantages of good controllability,high repeatability,and a relatively low cost.The crystal morphology changes during the growth process,and the understanding of its growth and etching mechanism is of great significance to the preparation of high-quality and large-scale two-dimensional materials.Meanwhile,the corresponding optical and electrical performance characterization is a necessary step to expand the future industrial application of two-dimensional materials.In this thesis,three transition metal chalcogenide(TMDs)materials were prepared by CVD method:molybdenum disulphide(Mo S₂),molybdenum diselenide(Mo Se₂),and tungsten disulfide(WS₂).The samples were characterized by Raman spectroscopy,photoluminescence spectroscopy,scanning electron microscope,and transmission electron microscope respectively.The mechanism of the etching during CVD growth is explained,and then the hexagonal shape TMDs material obtained during the etching process is used to fabricate the device,and the catalytic hydrogen evolution performance of hexagonal Mo Se₂is tested because a large number of active sites exposed at the edge is expected to favor the hydrogen evolution reaction.The results are summarized as follows:The growth parameters of Mo S₂,Mo Se₂and WS₂ by CVD method were determined through a large number of experiments,and the crystal quality of the three TMDs materials were confirmed by Raman spectroscopy and photoluminescence spectroscopy.The shapes of the TMDs during etching process were photographed by high resolution scanning electron microscope,thus confirm its morphology.According to previous report,the difference between the chemical potentials of chalcogen atoms and transition metal atoms is used to calculate the growth rate of three possible edges in TMDs,based on the feature that the fastest growing edge will disappear during growth and will retain during etching process,the shape that may appear during the etching process is obtained.Moreover,the final hexagonal shape with specific angle because of the edge is shown,by measuring the angle of this hexagonal Mo Se₂ in low-magnification TEM image,we got a consistent result that matches the calculation result,thus proved the etching process.The high-angle annular dark field image of the edge of this special-shaped TMDs material shows that the edge has a large number of defect sites.Based on this,we further tested the electrocatalytic performance of the hexagonal Mo Se₂ with the specific angle and obtained a Tafel slope of 58 m V/dec at the edge and84 m V/dec in the center,showing excellent catalytic hydrogen evolution performance.In addition,we also introduced the experimental process of the one-step growth of Mo S₂/WS₂ heterostructure,and characterized its structure by Raman characteristic peaks mapping,and proved the epitaxial growth of WS₂ outside Mo S₂ due to the different evaporation temperature of the precursor,the clean and sharp contact interface is demonstrated at the same time.In terms of characterization,the methods used to characterize two-dimensional materials in the past,such as atomic force microscopy,Raman spectroscopy,X-ray photoelectron spectroscopy,transmission electron microscopy,etc.,usually take up much characterization time.In addition,shortcomings of these traditional characterization methods lie in limited characterization range,damage to samples,and expensive equipment.In order to improve the characterization efficiency of research on two dimensional materials and accelerate the industrial application of two-dimensional materials,researchers urgently need a method that is able to quickly identify and characterize two-dimensional materials.Recently,as a major branch of artificial intelligence,machine learning has been widely used in the field of materials science due to its high automation and high throughput.In this thesis,combined with the artificial neural network,the chromaticity values RGB and HSV in the optical microscope images of the two-dimensional materials are utilized as input data to quickly classify the two-dimensional materials.The results are summarized as follows:We applied optical microscope photos of 8 monolayer and bilayer two-dimensional materials taken under different light intensities as the data set.The artificial neural network is able to quickly and accurately determine the material type and number of layers,the pixel level average prediction accuracy rate reached 91%.In addition,we utilized the RGB chromaticity values of CVD-grown Mo S₂ with different mono-sulfur defect concentrations(0.40 nm^-2,0.90 nm^-2,1.50 nm^-2)to establish a ternary linear model to predict concentrations of this dominant vacancy in TMDs.The determination coefficient is 0.84,showing a good fitness of this model.The prediction result of sample with a defect concentration of 1.20 nm^-2 is 1.24 nm^-2,thus proves the feasibility of this method.It reveals a new way to estimate the quality of two-dimensional materials merely from the optical images,making it easier and faster for researchers to understand the crystal quality of the sample,and adjust the corresponding growth parameters according to the prediction result.Finally,we characterize the chemical composition and interface information of the Mo S₂/WS₂ heterostructure according to the well-trained ANN,and compare the performance with the result of Raman characterization.The prediction of the chemical composition of the heterostructure matches well with the Raman characterization.In addition,this method can identify the roughness of the interface information of heterostructure,which offers reference for selecting two-dimensional material heterostructures with clean interfaces.This highly automated characterization of two dimensional materials based on machine learning method can greatly save the time for sample preparation and actual characterization,thus improve the research efficiency.