Multitasking Memristor for High Performance and Ultralow Power Artificial Synaptic Device Application
Ni Yang, Ji Zhang, Jing‐Kai Huang, Yang Liu, Junjie Shi, Qianli Si, Jack Yang, Sean Li
Abstract
The emergence of in-memory computing has shed light on solving high-power consumption and low computation efficiency of the traditional computers built with von Neumann architecture. Memristor, which exhibits history-dependent conductivity modulation, can simulate the synaptic behaviors in the biological brain. However, it remains as a key challenge to fabricate devices that can demonstrate a wide range of synaptic plasticity and maintain stable switching responses over repetitive operating cycles. Hereby, the memristor made of Au/Ti/TiO2/Nb:SrTiO3 (Nb:STO) heterojunction shows a partial nonvolatile bipolar resistive switching behavior with an initial high on/off switching ratio of ∼104, and “writing” and “erasing” with long endurance across 1.5 × 104 cycles. Furthermore, we experimentally developed a single device that possesses a 5-bits (32-states) reservoir computing system to recognize the binary patterns. We also demonstrated the multidata storage for neuromorphic computing in a 10 × 10 neuromorphic array to recognize the patterns of multilevel resistance states with an ultralow operation power of 4.1 pJ. In addition, various synaptic dependent plasticity performances, including spike-duration, -interval, and -number dependent plasticity, have been realized. Such an on-demand neuromorphic device exhibits a multitask shifting potential for analog bipolar memory and bistate and multistate neuromorphic networks and paves a way to develop the highly efficient memristor devices.