Construction And Application Of Neural Network-Based Interatomic Potential For N-Ga-Al Semiconductors

N-Ga-Al semiconductors has enabled its many applications in light-emitting diodes,high-electron mobility transistors(HEMTs),etc.due to its outstanding mechanical and thermodynamic properties,but the knowledge of the microstructure determining the superb properties remains yet insufficient.In molecular dynamics(MD)simulations,there are many interatomic potentials(force fields)used to study the properties of GaN and AlN,which have achieved some success in different aspects,but have many deficiencies in the simulation of the GaN/AlN heterogenous interface.To accurately study the properties of GaN/AlN heterostructure,in this work,an interatomic potential(force field)based on a deep neural network technique and firstprinciples calculation is developed for N-Ga-Al semiconductors.This potential can accurately reproduce the energy and force of crystal and amorphous structures,with the root-mean-square errors(RMSEs)less than 4 me V/atom and 0.18 me V/?,respectively,and compared to the Stillinger-Weber(SW)potential on amorphous,the accuracy of the trained neural network potential(NNP)is improved by an order of magnitude.The radial distribution functions(RDFs)of GaN and AlN were calculated by MD simulations based on NNP,which showed that GaN and AlN have relatively similar lattice models.and the elastic constants and elastic moduli of GaN and AlN of crystal and amorphous structure were also calculated,which showed that the Young’s modulus,bulk modulus and shear modulus of the disordered structure have decreased(about 30%)compared to the crystalline structure.The RDFs and elastic constants obtained from the NNP-based MD simulations are consistent with the experimental and DFT results.The lattice thermal conductivity of GaN and AlN at different temperatures was calculated using equilibrium MD based on NNP,which shows that the decrease in thermal conductivity value with increasing in temperature,and the simulated results are in agreement with the experimental and first principles data and much better than the calculated values of empirical potentials.The interfacial thermal conductance of GaN/AlN heterostructures with different interfacial morphologies are further studied using non-equilibrium MD simulations with NNP,it is found the atomic interdiffusion and disorder at the interfaces dramatically reduces the interfacial thermal conductance.This work shows the validity of trained NNP for MD simulation of N-Ga-Al semiconductors.