Computational Investigation of Nanoscale Memristor Devices for Neuromorphic Computing

Field effect transistors (FETs) with their Boolean operation have been the fundamental building block of digital computers. While this digital framework is excellent at performing complex arithmetic and logic calculations, it lags far behind the human brain in key areas such as adaptivity, generalization, and pattern recognition. Neuromorphic computation using nanoscale adaptive oxide devices or memristors is a very promising alternative to this framework. Oxides of transition metals such as hafnium (HfO₂) are proven to be excellent candidate materials for these devices, which show non-volatile memory and analog switching characteristics, while employing only two voltage terminals. This work focuses on the computational investigation of local and temporal variation of voltage, current and temperature, so that the coupled nature of heat transfer and current flow through these devices is analyzed and the effects of substrate materials, set/reset voltages on relative on-state/off-state resistances, current magnitude and filament temperature are determined. These analyses are expected to provide insightful information on switching mechanisms of these adaptive oxides and warrant their scaled-up implementation for neuromorphic computing.

Nomenclature

a

Vacancy Hopping Distance, m

BE

Bottom Electrode

c

Specific Heat J/kg-K

D

Diffusion Coefficient, m²/s

E

Activation Energy, J

F

Filament

f

Hopping Frequency, Hz

k

Thermal Conductivity, W/m-K

Boltzmann Constant, J/K

n

Concentration, /m³

S

Thermophoresis Coefficient, /K

T

Temperature, K

t

Time, s

TE

Top Electrode

v

Oxygen Vacancy Drift Velocity, m/s