An InGaZnO Synaptic Transistor Using Titanium-Oxide Traps at Back Channel for Neuromorphic Computing

Abstract

Synaptic transistors have attracted growing interest due to their potential in bio-inspired computing. Conventional synaptic transistors typically rely on charge traps, dipoles, or mobile ions/vacancies in the gate dielectric for channel conductance modulation, which causes performance deterioration owing to Coulomb scattering. Herein, a new dual-gate InGaZnO (IGZO) synaptic transistor is proposed. An unisolated top gate with post metal annealing (PMA) treatment is designed to block moisture and form titanium-oxide (TiOx)-associated defects at the IGZO back channel. By using the TiOx defects rather than the gate dielectric to regulate the channel conductance, Coulomb scattering is avoided, and so the device shows relatively high carrier mobility [~11.5 cm2/(V $\cdot $ s)] and small subthreshold swing (~231 mV/dec). Additionally, typical biological synaptic functions are successfully mimicked, and a relatively low device-to-device variation (~7.2%) for conductance modulation is obtained. Furthermore, a convolutional neural network (CNN) based on this device achieves high accuracy in the image classification tasks, demonstrating the great potential of the proposed device in neuromorphic computing.