Two-terminal photovoltaic neuromorphic device with temporal dynamics for reservoir computing

Photovoltaic (PV) neuromorphic devices with photocurrents under illumination as readouts have gained increasing attention due to their ultralow latency and excellent energy efficiency during reading process. However, they face significant challenges in processing temporal data because of the lack of inherent temporal dynamics, limiting their application in reservoir computing (RC) systems. Here, we have developed a simple two-terminal PV neuromorphic device based on an indium tin oxide/Nb-SrTiO3 oxide Schottky heterojunction, which features multi-level PV responses by adjusting the built-in electric field. The spontaneous recapture of electrons by charged defects leads to relaxation of the built-in electric field over time, providing inherent temporal dynamics for the PV device. Using this device, we designed a RC system that achieved high-accurate recognition of image letters and spoken-digits. This work offers an efficacious approach to design neuromorphic devices that combine temporal dynamics with low-energy consumption.