Chalcogenide phase change alloys (GeSbTe alloys, GeTe and related materials) have been extensively investigated in recent years because of their use in non-volatile phase change memories (PCM), which are an emerging data storage technology. This application relies on the fast (10-100 ns) and reversible transformation between the crystalline and amorphous phases of these alloys induced by Joule heating. During the cell programming, the amorphous phase is obtained by rapidly quenching the melt, while the crystal is obtained via recrystallization of the amorphous. The two states of the memory can be discriminated thanks to the large difference in electronic conductivity of the two phases.
Important issues for the development of PCMs are the overall reduction of cell size and of power consumption for the cell programming operations. A very attractive option in this respect involves the change of the PCM geometry by using nanowires (NWs) owing to their higher crystallinity and controllable sizes down to very low scale. NWs show a decrease of programming currents with the NW size due to the reduction of cross-sectional area and the reduction of the melting temperature. Moreover, structural relaxations in the amorphous phase, are reduced in NWs with respect to the bulk. In fact, while the crystal is thermodynamically stable, the amorphous is metastable and subject to aging resulting in an increase of the electrical resistivity of the amorphous with time. This phenomenon called drift is detrimental in PCMs devices and is weaker in NWs with respect to the bulk.
In this work we focus on the crystallization kinetics and aging of the GeTe compound for which an interatomic potential for large scale simulations is actually available. The potential is built by fitting a huge DFT database with a Neural-Network method [1, 2]. By means of molecular dynamics techniques, we have estimated the melting temperature of the NW and simulated the crystallization process at the crystal-liquid interface in a NW with a diameter of about 9.0 nm (~16000 atoms) and in bulk models of about 10000 atoms. It turns out that the melting temperature for the NW is about 100 K lower than that of the bulk. A lower but still high crystal growth velocity is found for the NW with respect to the bulk due to the lower melting temperature.
As regards the aging process, an amorphous NW generated by rapidly quenching a melted portion of the NW, shows shorter chains of homopolar Ge-Ge bonds with respect to bulk models, which we propose to be at the origin of the reduction of structural relaxations and hence of the weaker resistance drift in NWs with respect to the bulk.
 G. C. Sosso et al., Phys. Rev. B, 85, 174103 (2012).
 J. Behler and M. Parrinello, Phys. Rev. Lett., 98, 146401 (2007).