High-dimensional neural-network potentials for phase change materials for data storage

In this seminar I will review our recent study of the properties of phase change materials by means of molecular dynamics simulations based on Neural Network potentials [1]. The materials we are dealing with are chalcogenide alloys (typically Ge2Sb2Te5 or GeTe) which are the subject of extensive experimental and theoretical research because of their use in optical (digital versatile disc, DVD) and electronic (phase change memories, PCM) storage devices [2]. Both applications rely on the fast and reversible transformation between the crystalline and amorphous phases induced by heating either via laser irradiation (DVD) or Joule effect (PCM). The two states of the memory can be discriminated because of the large difference in optical reflectivity and electronic conductivity of the two phases. In the last few years atomistic simulations based on density functional theory have provided useful insights on the properties of materials in this class [3]. However, several key issues such as the crystallization dynamics occurring at the ns time scale, the properties of the crystalline/amorphous interface and the thermal conductivity at the nanoscale, just to name a few, are presently beyond the reach of fully ab-initio simulations. A route to overcome the limitations in system size and time scale of ab-initio molecular dynamics is the development of empirical interatomic potentials. However, traditional approaches based on the fitting of simple functional forms are very challenges due to the complexity and variability of the chemical bonding in the crystal and amorphous phases of these materials as revealed by ab-initio simulations. A possible solution was demonstrated few years ago by Behler and Parrinello [2] who developed classical interatomic potentials with close to ab-initio accuracy by fitting large ab-initio databases by means of a NN scheme. By means of this technique, we have developed a interatomic potential for GeTe [4] which is one of the compounds under scrutiny for applications in phase change memories. The NN potential with a close to ab-initio accuracy has allowed us to address several key issues for applications in data storage such as the homogeneous and heterogeneous crystallization of the amorphous, the glass transition of the supercooled liquid and the thermal conductivity of the amorphous phase. [1] J. Behler and M. Parrinello, Phys. Rev. Lett. 14, 146401 (2007); J. Behler. Chemical Modelling, 7, 1 (2010). [2] M. Wuttig and N. Yamada, Nature Mater. 6, 824 (2007) [3] S. Caravati, M. Bernasconi, T.D. Kuehne, M. Krack, and M. Parrinello, Appl. Phys. Lett. 91, 171906 (2007); ibidem, Phys. Rev. Lett. 102, 205502 (2009); J. Phys. Cond. Matt. 23, 265801 (2011); J. Phys. Cond. Matt. 21, 255501 (2009); D. Lencer, M. Salinga and M. Wuttig, Adv. Mater. 23, 2030 (2011). [4] G. C. Sosso, G. Miceli, S. Caravati, J. Behler, M. Bernasconi, arXiv:1201.2026v1.