A machine-learned interatomic potential for Ge-rich GexTe alloys has been developed aiming at uncovering the kinetics of phase separation and crystallization in these materials. The results are of interest for the operation of embedded phase change memories which exploits Ge-enrichment of GeSbTe alloys to raise the crystallization temperature. The potential is generated by fitting a large database of energies and forces computed within Density Functional Theory with the neural network scheme implemented in the DeePMD-kit package. The potential is highly accurate and suitable to describe the structural and dynamical properties of the liquid, amorphous and crystalline phases of the wide range of compositions from pure Ge and stoichiometric GeTe to the Ge-rich GexTe alloy. Large scale molecular dynamics simulations have suggested a crystallization mechanism which depends on temperature. At 600 K, segregation of most of Ge in excess was observed to occur on the ns time scale followed by crystallization of nearly stoichiometric GeTe regions. At 500 K, nucleation of crystalline GeTe was observed to occur before phase separation, followed by a slow crystal growth due to the concurrent expulsion of Ge in excess.

中文翻译：

---

来自具有机器学习原子间电位的大规模模拟的富锗 Ge[公式省略]Te 合金的结晶动力学


已经开发了一种用于富锗 GexTe 合金的机器学习原子间势，旨在揭示这些材料中相分离和结晶的动力学。该结果对嵌入式相变存储器的操作具有重要意义，该存储器利用 GeSbTe 合金的 Ge 富集来提高结晶温度。该电势是通过将密度泛函理论中计算的能量和力的大型数据库与 DeePMD-kit 包中实现的神经网络方案拟合而产生的。该电位非常准确，适合描述从纯锗和化学计量 GeTe 到富锗 GexTe 合金的各种成分的液相、非晶相和结晶相的结构和动力学特性。大规模分子动力学模拟表明了一种取决于温度的结晶机制。在 600 K 时，观察到大部分过量的 Ge 的分离发生在 ns 时间尺度上，然后是近化学计量的 GeTe 区域的结晶。在 500 K 时，观察到结晶 GeTe 的成核发生在相分离之前，随后由于同时过量排出 Ge，晶体生长缓慢。