Atomistic study of the supercooled liquid and amorphous state of the phase-change material GeTe

Phase-change materials (PCMs) are an important family of alloys employed in non-volatile memories and neuromorphic devices. Phase-change memories are expected to fill the gap between DRAM and flash memories, showing high operation speed comparable with DRAM and non-volatility. Intel Optane is the first "stand-alone" memory based on PCMs. PCMs also show great potential in neural-computing devices, such as in-memory computing, since multi-level resistance states can be easily achieved. All of these devices exploit the unique properties of PCMs: large resistance contrast combined with rapid and reversible transitions between amorphous and crystalline phases. In memory devices, since the two phases are used to represent the logical states "0" and "1", they need to be stable at ambient condition. However, the crystallization must be fast at elevated temperatures to ensure fast switching. These two "conflicting" behaviours imply a very strong temperature dependence of the crystallization kinetics, which is attributed to the high fragility of the supercooled liquid phase. Hence, the temperature dependence of the dynamic properties is one of the key aspects considered in device design and optimization. However, the fragility of supercooled PCMs is hard to determine due to the fast crystallization in the deeply supercooled regime, which is thus called "no-man's land". Recently, it was also found that the two PMCs $\mathrm{Ag_4In_3Sb_{67}Te_{26}}$ and $\mathrm{Ge_{15}Sb_{85}}$, besides showing vitrification at the glass transition temperature, display a liquid-liquid phase transition point at higher temperature. The two liquid phases are characterized by different structural properties, which were detected by femtosecond X-ray scattering experiments. With the help of simulations, the liquid-liquid phase transition was linked to the suppression of Peierls-like distortions. This discovery adds further complexity to the thermodynamic properties of the supercooled liquid phase of PCMs and also stimulates research into the polyamorphism of these materials. Most of PCMs are antimonides or chalcogenides. In particular, the family of Ge-Sb-Te PCM compounds with stoichiometry $\mathrm{(GeTe)_x(Sb_2Te_3)_{1-x}}$ has been intensively studied and is used in commercial products. In this thesis, we focus on the PCM GeTe, not only because it is the parent compound of the Ge-Sb-Te alloys but also due to its own remarkable combination of properties. We investigate both the polyamorphism of GeTe and the dynamical properties in the supercooled range. We first conduct neural-network based molecular dynamics simulations to generate models of amorphous GeTe at different pressures and study pressure-induced polyamorphism. For comparison, another binary compound, GeSe, is also considered. The structural and electronic properties at different pressures are thoroughly analyzed and compared with the corresponding experimental results, conducted with high-energy synchrotron X-rays. For both amorphous GeTe and GeSe, our analysis reveals a high-pressure state characterized by diminished Peierls-like distortion, narrower energy gap and reduced compressibility with respect to the low-pressure state. Nevertheless, the transition occurs at higher pressure in GeSe. Our findings underscore the crucial role of Peierls-like distortions in amorphous PCMs with local octahedral order. These distortions can be controlled through pressure and composition, which opens up the possibility to design PCMs with optimal properties for PCM-based devices. We then move our attention to the supercooled liquid phase. Exploration of the potential energy landscape (PEL) is a powerful approach for understanding the thermodynamics and dynamics of glass forming liquids. Such exploration can be carried out using molecular dynamics simulations. However, a large set of configurations in a wide range of temperatures is needed for a good sampling of the PEL, requiring long simulation times beyond the reach of ab initio methods. Here we use neural-network based molecular dynamics to investigate the PEL of liquid and supercooled liquid GeTe. Combining the exploration of the PEL with thermodynamic integration, we determine the configurational entropy of deeply supercooled GeTe. Then we investigate the dynamical properties: the relaxation time and the shear viscosity are calculated using linear response theory in the same temperature range. Finally, the Adam-Gibbs relation is used to extrapolate the data to low temperatures and estimate the glass transition temperature and the fragility index. Our results indicate that GeTe is a highly fragile system with fragility index of order 135-140.