Bi₂Se₃-Based Memristive Devices for Neuromorphic Processing of Analogue Video Signals

Abstract

Bismuth selenide (Bi2Se3), a layered semiconductor, has attracted a great deal of attention as a thermoelectric material as well as a potential topological insulator. Here, we present a work showing that Bi2Se3 can also be used for making memristive devices capable of directly processing analog video signals. In this work, Bi2Se3 memristors are produced by multiplexing rubbing-induced site-selective growth, which potentially enables scalable implementation of such memristor arrays for constructing large-scale neuromorphic systems. The fabricated Bi2Se3 memristors exhibit prominent memristive switching characteristics under the application of time-sequential voltage pulses. Especially, such a Bi2Se3 memristor exhibits a reliable dependence of memristive responses on the duty cycle of programming pulses, fast recovery behavior from a dynamically modulated state, and a large drive current. These properties could be employed for extracting spatiotemporal information from analogue signals and realizing practical neuromorphic sensory functions. Our additional tests strongly imply that the memristive output of a Bi2Se3 memristor in response to analogue video scanline signals could be implemented to construct future hardware-based computer vision systems capable of rapidly acquiring graphic information and directly actuating robotic systems with minimal data transmission and energy consumption. Finally, we attribute the observed memristive characteristics to field-mediated drift and diffusion of the selenium vacancies in the Bi2Se3 layers. The simulated memristive response based on this hypothesis model is consistent with the experimental result. This work provides a potentially upscalable device solution to realize memristor-based neuromorphic sensory or edge computing systems.