二元碲化物多晶薄膜高压熔化与结晶动力学的大规模分子动力学研究

It is crucial to unravel the high-pressure melting and crystallization kinetics as well as the underline mechanism of polycrystalline phase-change materials in order to fabricate the desirable phase-change random access memory devices with low-power consumption, high-speed and long data-retention. Due to the extremely tiny spatial-temporal scales in nanoseconds and nanometers of the reversible phase-change areas, which is also taken place under confined circumstances, the present-day experimental techniques cannot present the phase-change material behavior under service conditions. Although molecular dynamics modeling can make it, ab initio molecular dynamics (AIMD) simulations are not able to handle such large spatial-temporal scale modeling for polycrystalline phase-change materials, while classical molecular dynamics suffer from precise potentials. To this end, this project will firstly use artificial neural network methods combined with high throughput ab initio total energy calculations to generate accurate high-dimensional neural network potentials (NNPs) for polycrystalline GeTe and Sb2Te3 phase-change materials, and then integrate the NNPs into classical molecular dynamics code like LAMMPS and perform large-scale molecular dynamics to simulate the high-pressure melting, quenching the liquids to room temperature for amorphous states and then recrystallizing the amorphous states at high temperatures. Finally, with a few experiments, this project will unravel the behavior of melting points with pressures, the crystallization kinetics of amorphous states as well as the underline mechanism for polycrystalline GeTe and Sb2Te3 phase-change materials under service conditions. Meanwhile, the role of grain boundary, material volume size and initial amorphous will also be revealed. The results of this present will provide fundamental guides for designing better novel phase-change materials for practical applications.

阐明多晶相变存储材料的高压熔化特征及其非晶的晶化动力学行为是获得功耗低、存储速度快和数据保持寿命长的相变存储器的前提和关键。由于存储操作发生在纳秒、纳米级的时空尺度且在受限的环境下，现有实验技术难以研究；分子动力学有望揭示多晶相变材料在真实服役条件下的材料行为。但是，第一原理分子动力学具有时空尺度极小的局限性，而经典分子动力学缺乏精准的势函数。针对上述问题，本项目将利用人工神经网络法结合第一原理高通量计算，构建相变材料GeTe和Sb2Te3的精准高维神经网络势，并将其集成到经典分子动力学模拟软件，实现相变材料服役条件下的大规模分子动力学模拟；结合实验验证，阐明多晶GeTe和Sb2Te3薄膜的熔化规律及熔点与压力的关系，揭示非晶态的晶化动力学特征及其微观机理，并阐明非晶相初始态的影响。本项目研究结果将为设计性能更好的相变存储材料提供理论基础。

以硫族化合物为存储介质的相变存储器是下一代主流非易失存储器。相变材料的非晶晶化时间、非晶态热稳定性、电/热输运性质是获得低功耗、高速度、长寿命相变存储器的关键。针对这些关键问题，项目开展了以下研究: 跨尺度机器学习碲化物神经网络势函数及其与分子动力学软件的集成；元素掺杂对碲化物晶体/非晶结构及电/热输运性质的影响；C/N掺杂对多晶GeSb2Te4结构与电子性能的影响；W掺杂对立方态GeTe结构与性质的影响；空位及掺杂对碲化物的结构和相变机理的影响；新型硫族化合物半导体的高通量筛选与性能预测；Y掺杂Sb2Te3及其相变存储器件的制备与性能研究。重要结果如下：自主开发出势函数拟合软件PotentialMind，获得了用于大规模分子动力学模拟的Sb2Te3神经网络势函数，为服役条件下相变材料的模拟提供了基础；高通量计算筛选出Sb2Te3的最佳掺杂元素Y和Sc；阐明了Y、Sc对碲化物结构和电/热输运性质的影响，发现Y、Sc均能提高非晶态的热稳定性且降低晶态的电导率和热导率，有利于降低器件功耗并提高其寿命；揭示了不同掺杂浓度的C/N在多晶GeSb2Te4中的原子组态及对电导率的影响趋势；发现了W掺杂导致GeTe的反常晶格收缩；发现了立方态和菱方态Y-Sb2Te3之间的可逆位移型相变，并阐明了其相变机理；基于GeTe-Sb2Te3伪二元碲化物设计出144个硫化物，预测出56个具有极低本征晶格热导率的新型稳定半导体，揭示了其热电输运性质；实现了对YxSb2-xTe3成分的精确实验控制，基于Y0.25Sb1.75Te3制备出的T型相变存储器件展现出极具竞争力的操作速度（6ns）和超低功耗（1.3pJ）的特点；设计出以Y-Sb2Te3为存储介质的多级相变存储器，具有超低器件功耗（0.6 pJ~4.3 pJ）和超低电阻漂移系数（<0.007）、超快擦写速度（3.2 ns）和优异循环性能（>10^6）的优势，开辟了一条无需复杂制造工艺和迭代编程操作即可实现先进多级相变存储的新途径。