Multiphysics modeling of Ge2Sb2Te5 based synaptic devices for brain inspired computing
Modeling nanoscale devices that emulate the functionality of synapses of the biological brain is a fundamental operation for developing brain-inspired computational systems. Phase-change material based synaptic devices offer promising performance in speed, spatial and power efficiency metrics, up to human brain level, when connected in a massively parallel crossbar array architecture. In this work, we modeled electrothermal characteristics of a single synaptic device consisting of phase-change material based memory and its selector. First, we proposed a finite element method based simulation framework for modeling electrical, thermal and probabilistic crystallization dynamics of the memory unit. Gradual phase transitions that form device memory between amorphous and crystalline states are studied under nanosecond voltage pulses. Second, we implemented time and temperature dependent resistance drift saturation model for phase-change material based selector device. Our model is in close agreement with the ultrafast saturation phenomena which is observed for the first time in fabricated devices with 8 nm node technology.