A self-powered artificial retina perception system for image preprocessing based on photovoltaic devices and memristive arrays

• Published in: Nano energy 2020-12, Vol.78, p.105246, Article 105246
• Main Authors: Yang, Xiaoyang, Xiong, Ziyu, Chen, Yongji, Ren, Yi, Zhou, Li, Li, Huilin, Zhou, Ye, Pan, Feng, Han, Su-Ting
• Format: Article
• Language: English
• Subjects: Artificial retina; Memristor; Neuromorphic computing; Perovskite; Self-powered
• ISSN: 2211-2855
• DOI: 10.1016/j.nanoen.2020.105246

Artificial retina perception system is significant to pattern recognition and visual function emulation. However, the recent artificial retina system is mainly reported on the integration of three-terminal transistors, whose structural limitations may result in low processing speeds and high energy consumption due to a low array density and complex line design. Furthermore, the external power source is required to drive devices so that the power consumption of the system would increase. Here we present a self-powered artificial retina perception system by utilizing two-terminal solar cells as artificial neurons and perovskite-based memristors as artificial synapses, ensuring the bio-inspired retina system with extendable crossbar array structure for high-density and low power consumption neural networks. By a light stimulation with various wavelengths and intensities, the electrical pre-synaptic signal is generated in the solar cell and subsequently transferred to the perovskite-based memristor to perform further information preprocessing. Especially, the applicability of the artificial retina system to neuromorphic computing is demonstrated for contrast enhancement and noise reduction. The retina perception system is capable of feature extraction by to implement partial functions of convolutional neural networks (CNNs) on the hardware level with improved recognition rate, boosted recognition speed, and reduced energy consumption. A self-powered artificial retina perception system is proposed by integrating two-terminal solar cells and perovskite-based memristors. The STP-LTP functionality of perovskite-based memristors is achieved via the passivation. This self-driven artificial optic-neural system can execute both the emulation of biological synapses and partial functions (eg. the feature extraction for external signals) of convolutional neural networks. [Display omitted] •We design a self-powered artificial retina perception system composed of solar cells and perovskite-based memristors.•Synaptic plasticity can be implemented by solar cells due to the low operating voltage of ~ 0.4 V.•The system has capabilities of image preprocessing, such as contrast enhancement and noise reduction.•High recognition rate of ~ 86% can be implemented within 180 training epochs.