基于第一性原理的金刚石薄膜热导率神经网络预测研究  

Prediction of thermal conductivity of diamond film by neural network based on first principles

作　　者：王彪[1] 任海杰 曹文鑫[1] 胡彦伟[3] 何玉荣[3] 韩杰才[1] 朱嘉琦[1] WANG Biao;REN Hai-Jie;CAO Wen-Xin;HU Yan-Wei;HE Yu-Rong;HAN Jie-Cai;ZHU Jia-Qi(Center for Composite Materials and Structures,Harbin Institute of Technology,Harbin 150080,China;State Key Laboratory of Aerodynamics,Mianyang 621000,China;School of Energy Science and Engineering,Harbin Institute of Technology,Harbin 150080,China)

机构地区：[1]哈尔滨工业大学复合材料与结构研究所,哈尔滨150080 [2]空气动力学国家重点实验室,绵阳621000 [3]哈尔滨工业大学能源科学与工程学院,哈尔滨150080

出　　处：《中国科学：物理学、力学、天文学》2022年第9期58-70,共13页Scientia Sinica Physica,Mechanica & Astronomica

基　　金：国家重点基础研究发展计划(编号:2020YFA0709700);中央高校基本科研业务费专项资金(编号:HIT.OCEF.2021016);国家自然科学基金(编号:52102039);先进焊接与连接国家重点实验室开放基金(编号:AWJ-22Z04)资助项目。

摘　　要：金刚石作为高导热材料对微电子冷却等应用至关重要,对其热导率进行跨尺度分析有利于微纳电子元器件的设计和热管理.玻尔兹曼输运方程是描述多尺度声子输运现象的理想工具,然而因其高维性,利用其对不同尺度金刚石的热导率进行数值求解预测极具挑战性.本文基于第一性原理,结合以玻尔兹曼输运方程为物理信息的神经网络,有效地预测了金刚石内声子的多尺度热输运问题,并与线性化玻尔兹曼输运方程的求解结果对比,两者具有较高一致性.经分析,金刚石在毫米尺度热导率便会出现尺寸效应,同一温度下,金刚石尺度越小其归一化热导率越低,而温度越低尺寸效应越明显,且金刚石每个模态的声子热导率的变化规律是相同的.As a high thermal conductivity material, diamond is significant for microelectronic cooling and other applications.Multiscale analysis of diamond’s thermal conductivity is beneficial to designing micro-nano electronic components and thermal management. The Boltzmann transport equation is an ideal tool to describe the phenomenon of multiscale phonon transport. However, due to its high dimensionality, it is incredibly challenging to use it to calculate and predict the thermal conductivity of diamonds at different scales. Based on the first principle, physical information neural network and Boltzmann transport equation, this paper effectively predicts the multiscale thermal transport of phonons in diamond and makes related analysis. After the phonon dispersion information, the diamond’s group velocity and scattering rate were calculated by the first principles. Relevant results were input into the physical information neural network as training samples. To introduce the Boltzmann transport equation as physical information in physical information neural network, we embedded related control equations and boundary conditions in the loss function of the neural network.When the loss function of the physical information neural network drops to a set standard, the trained model can accurately predict the phonon energy distribution. In this paper, physical information neural network was used to predict the transport law of phonons in a one-dimensional diamond film at different temperatures and scales. The solution results of the physical information neural network and linearized Boltzmann transport equation have high consistency. It could be deduced that the thermal conductivity of diamond had a size effect on the millimeter scale. At the same temperature,the smaller the scale, the lower the normalized thermal conductivity of the diamond, and the lower the temperature, the more pronounced the size effect is on the thermal conductivity. Under the same scale, the normalized thermal conductivity was lower, and the change l

关 键 词：物理信息神经网络 声子-玻尔兹曼方程 金刚石热导率 第一性原理 

分 类 号：O469[理学—凝聚态物理] TP183[理学—电子物理学]