神经形态工程学
计算机科学
记忆电阻器
多光谱图像
突触
卷积神经网络
终端(电信)
极性(国际关系)
人工智能
横杆开关
计算机体系结构
计算机硬件
人工神经网络
电子工程
神经科学
化学
电信
工程类
生物
生物化学
细胞
作者
Jong‐Min Lee,Bum Ho Jeong,Eswaran Kamaraj,Dohyung Kim,Hakjun Kim,Sanghyuk Park,Hui Joon Park
标识
DOI:10.1038/s41467-023-41419-y
摘要
An optoelectronic synapse having a multispectral color-discriminating ability is an essential prerequisite to emulate the human retina for realizing a neuromorphic visual system. Several studies based on the three-terminal transistor architecture have shown its feasibility; however, its implementation with a two-terminal memristor architecture, advantageous to achieving high integration density as a simple crossbar array for an ultra-high-resolution vision chip, remains a challenge. Furthermore, regardless of the architecture, it requires specific material combinations to exhibit the photo-synaptic functionalities, and thus its integration into various systems is limited. Here, we suggest an approach that can universally introduce a color-discriminating synaptic functionality into a two-terminal memristor irrespective of the kinds of switching medium. This is possible by simply introducing the molecular interlayer with long-lasting photo-enhanced dipoles that can adjust the resistance of the memristor at the light-irradiation. We also propose the molecular design principle that can afford this feature. The optoelectronic synapse array having a color-discriminating functionality is confirmed to improve the inference accuracy of the convolutional neural network for the colorful image recognition tasks through a visual pre-processing. Additionally, the wavelength-dependent optoelectronic synapse can also be leveraged in the design of a light-programmable reservoir computing system.
科研通智能强力驱动
Strongly Powered by AbleSci AI