Study On Thermal Properties Of Germanium Based On Deep Learning And Molecular Dynamics

In recent years,the semiconductor industry has ushered in a rapid development,and the size of microelectronic devices has shrunk to less than 10 nanometers.Microelectronic devices are usually compact and have high power,which requires better thermal management techniques to solve their heat dissipation problems.By analyzing the properties of heat transport in nanoscale systems,the thermal conductivity of microelectronic devices can be better controlled.Molecular dynamics methods are widely used to predict the thermal properties,and the accuracy of the results is related to the empirical potrntial.Deep learning algorithms can be used as a powerful tool to develop the precise empirical potrntial used in molecular dynamics simulations.In this paper,the thermal transport properties of germanium crystals and germanene nanoribbons are studied by molecular dynamics method.Firstly this paper takes crystal germanium as the research object and compares the consistency of the equilibrium(EMD)and non-equilibrium state simulation(NEMD)Molecular dynamics simulation method to study the thermal conductivity properties of materials.The research shows that the NEMD simulation results are better.On this basis,NEMD method was used to study the influence of different factors on the thermal conductivity of two-dimensional and three-dimensional materials: For germanium crystals,it is found that the temperature,the length of the system and the vacancy defects have a greater impact on the thermal conductivity of the material: For germanene nanoribbons,it isfound that the different chirality,the length of the nanoribbons and the potential function selected in the simulation have a greater impact on its thermal conductivity.In this paper,related phenomena are explained from the perspective of lattice dynamics and phonon propagation,which provides corresponding simulation data reference for the thermal management of nano-scale materials.First,the energy and force errors predicted by the neural network potential are compared with the results of AIMD.In addition,based on this potential function,the phonon dispersion curve of crystal germanium and the lattice thermal conductivity at200 to 800 K are obtained.The study shows that the NNP can retain the second and third order force constants of the material well.Finally we apply this potential to the molecular dynamics simulation to calculate the thermal conductivity of crystal germanium.In the simulation,this potential not only preserves the structure of the material well,but also the thermal conductivity obtained by the simulation at 300 K is in good agreement with the experimental measurement value.This study provides a high-precision solution for predicting the thermal properties of micro-nano systems,showing that deep neural networks are a powerful tool for simulating thermal transport at the micro-nano scale.