Impact of solid-liquid interfacial thermodynamics on the phase change memory RESET process

Abstract

A model of the RESET melting process in conventional phase-change memory (PCM) devices is constructed in which theGibbs-Thomson effect, representing local equilibrium at the solid-liquid interface, is included as an interfacial condition for the electro-thermal model of the PCM device. A comparison is made between the Gibbs-Thomson model and a commonly used model in which the interfacial temperature is fixed at the bulk melting temperature of the PCM material. The model is applied to conventional PCM designs in which a dome-shaped liquid/amorphous region is formed. Two families of solutions are computed representing steady state liquid regions, distinguished by their thermodynamic aspects. There is a family of solutions representing a liquid nucleation process, and a family of larger steady-state liquid solutions representing the limit of the melting process. A linear stability analysis is performed on the steady states, showing that the nucleus state is the threshold for further growth of the liquid phase which proceeds towards the melting limit state, which is the final stable state in the system. A comparison with a spherical symmetric model shows that in the isothermal limit the system is identical with the case of classical nucleation theory. The melting limits enable calculation of minima in voltage and corresponding current required for the RESET process. In this PCM configuration, the Gibbs-Thomson effect constrains the equilibrium solid-liquid interface temperature to remain above the bulk melting temperature during melting. The magnitude of this temperature difference increases with decreasing device size scale, thus requiring an increase in the required voltage and current needed for RESET compared to the case in which the interface temperature is approximated by the bulk melting temperature. This increase becomes substantial for active device dimensions in the $<$20nm range. The impact of this phenomena on PCM device design is discussed, emphasizing the increased motivation to explore alternative designs that avoid or reverse the cost penalty due to solid-liquid interfacial thermodynamics. By reducing the required RESET power, such design decisions have the potential to improve the performance of PCM for a multitude of applications, including storage class memory, neuromorphic computing, and in-memory computing for machine learning applications.