Analysis Of Single-molecule Charge Transport Data Based On Deep Clustering

Molecular electronics is a subject to explore the electrical properties of the single-molecule or single-atom at nanoscale.The study of molecular electronics aims to reveal chemical and physical properties that occur at the scale of single-molecule or single-atom to provide theoretical understanding and technical support for the development of future molecular materials and devices.Nowadays,the single-molecule break junction techniques are widely used to investigate the charge transport properties at the single-molecular or atomic scale and has attracted much attention in the field of molecular electronics.However,due to highly stochastic nature of the molecular junction formation in break junction measurement,statistical analysis methods are still widely used to find and characterize the charge transport characteristics of single-molecule.To overcome the statistic mean feature and improve the performance in characterizing the charge transport of single-molecule,it is urgent to develop more effective data analysis methods for break junction techniques.Recently,deep learning has achieved much process in many fields,such as data mining,pattern recognition,natural language processing and computer vision,which demonstrates the advantages of these methods in data analysis.To overcome the shortcoming of conventional statistical analysis methods and help researcher to find more accurate information from charge transport data,according to the features of charge transport data,a new data analysis method,called deep auto-encoder K-means(DAK)algorithm,is developed in this thesis by combining the deep learning algorithm in features extraction and machine learning algorithm in clustering.The main contributions and results of this thesis are as follows:(1)According to the features of single-molecular charge trcansport data,DAK algorithm is developed for data analysis.This algorithm employs the stacked auto-encoder algorithm to extract significant features from charge transport data and then applies K-means++ algorithm to the features,resulting in the charge transport events in break junction automatically recognized and distinguished by trace.This method greatly improves the ability of identifying and characterizing charge transport properties of single-molecule.(2)A simplified model of single-molecular charge transport is built,and then a dataset including four kinds of charge transport events is generated.Based on the simulation dataset,the effectiveness and performance of DAK algorithm are evaluated and demonstrated.Compared with multi-parameters vector-based classification process algorithm(MPVC)developed in other work,the proposed DAK algorithm has better clustering accuracy.(3)DAK algorithm is applied to the analysis of charge transport data that are obtained from three different molecular break junction systems,including a simple molecular system with two different charge transport events,a multi-component molecular system with three different charge transport events and a molecular system with chemical reaction.Through the analysis results,we demonstrated that DAK algorithm not only can identify the different molecular configurations formed during the break junction process,but also can used to investigate and characterize the dynamic of reaction at single-molecular scale by the dependent analysis.