Construction And Application Of Machine Learning Potentials For Interfacial Heat Transport Properties

In recent years,5G communication has also been gradually applied to smart cars.In5G communication,the reliability of smart chips is higher,and the working conditions of cars are very harsh,including high temperature and strong vibration.A very big challenge.Therefore,it is urgent to carry out research on advanced functional materials with negative Poisson’s ratio and high thermal conductivity（κ）to achieve advanced shock absorption protection and high-performance thermal management and improve the reliability and stability of chip operation.This paper focuses on the development and application of machine learning potential functions in the study of interfacial heat transport properties.First,it explores and compares two commonly used methods of machine learning（ML）fitting potential functions.The moment tensor potential function（Moment Tensor Potential,MTP）and neural-network potential（NNP）methods fit the potential functions of graphene and T-carbon,respectively.The results show that the negative Poisson’s ratio phenomenon of graphene can be well reproduced using the trained MTP potential function.By analyzing the evolution of key geometrical parameters,it is found that the increase in bond angle is responsible for the negative Poisson’s ratio in graphene.Subsequently,the neural-network potential（NNP）method was used to fit the potential function of T-carbon,and molecular dynamics simulations were used to calculate the thermal conductivity of T-carbon,which was in good agreement with the first-principles calculations.In the study of the interfacial thermal transport properties of BAs-Ga N,ML,DFT（Density functional theory）,MD and FEM（Finite Element Method）were used to develop multiscale simulations of thermal transport properties systems and in-depth physical mechanism studies.Using molecular dynamics simulations,the ultra-high thermal conductivity of BAs was directly verified from the perspective of atomic motion,and an ultra-high interface thermal conductance（ITC）of 265 MW m^-2K^-1 was obtained in the BAs-Ga N heterostructure,which are consistent with the experimental measurements.In-depth analysis shows that the underlying physical mechanism for the high interfacial thermal conductance lies in the well-matched lattice vibrations of BAs and Ga N.Based on the understanding of the interfacial thermal transport process from micro-and nanoscale DFT and MD simulations,the physical mechanism between grain size and effective thermal conductivity is investigated.The research results provide a basic theoretical basis for the interfacial heat transfer of BAs-Ga N heterostructures.At the same time,the research shows that the multi-scale simulation method driven by machine learning potential function can be applied to practical engineering and design of complex thermal management systems,thereby further Promote the miniaturization of electronic equipment.