神经形态工程学
材料科学
异质结
光电子学
突触
纳米技术
计算机科学
记忆电阻器
突触可塑性
人工神经网络
蒸发
功率(物理)
电压
半导体
量子点
峰值时间相关塑性
光开关
逻辑门
机器视觉
功率消耗
作者
Jian-Yu Jiang,Yunjie Liu,Jiang Li,Ankai Sun,Shuangshuang Li,Fuhai Guo,Ming-cong Zhang,Wenjing Jie,Lanzhong Hao,Jian-Yu Jiang,Yunjie Liu,Jiang Li,Ankai Sun,Shuangshuang Li,Fuhai Guo,Ming-cong Zhang,Wenjing Jie,Lanzhong Hao
标识
DOI:10.1021/acsami.5c16987
摘要
Due to the high energy and processing efficiencies, brain-inspired optoelectronic synaptic systems provide a promising solution for next-generation artificial vision computing. However, synapses based on single oxides face the challenge of high-power consumption, which seriously limits their practical application. This study presents multifunctional heterojunction optoelectronic synapses with low power consumption. Layered MoO3 and photochromic WO3 films are deposited in turn onto the ITO-covered quartz substrates by using the electron beam evaporation technique, and metal-oxide heterojunction synapses of MoO3/WO3 are fabricated. The synaptic devices exhibit versatile neuromorphic functionalities under both electrical and optical modulation. The heterojunction enables long- and short-term plasticity and achieves an accuracy of up to 92.4% in handwritten digit recognition. Under light stimulation, the device successfully demonstrated basic and advanced synaptic functions. More importantly, the power consumption of the synaptic event is only 67.6 fJ, which is far below those of other similar devices and close to biological synapses. The optoelectronic synapse arrays of 4 × 4 are developed to realize real-time visual perception and memory behaviors. This work provides effective strategies and a scientific foundation for developing next-generation ultralow-power artificial intelligence vision chips.
科研通智能强力驱动
Strongly Powered by AbleSci AI