ZnO-ITO/WO_(3-x)heterojunction structured memristor for optoelectronic co-modulation neuromorphic computation

用于光电共调制神经形态计算的氧化锌-氧化铟锡/氧化钨异质结忆阻器

Traditional transistors confront severe challenges of insufficient computing capability and excessive power consumption in large-scale neuromorphic systems.To address these critical bottlenecks,we propose an optoelectronic memristor based on zinc oxide-indium tin oxide/tungsten oxide(ZnO-ITO/WO_(3-x))heterojunctions as a promising solution.Through applying different types of electrical and optical signals,the device successfully emulates diverse synaptic functions including short-term/long-term synaptic plasticity,alongside short-term and long-term memory.Introducing the ZnO-ITO functional layer enhances the photoresponse of the WO_(3-x)-based memristor and demonstrates“learning-forgetting-relearning”behavior under optical modulation.Furthermore,based on the photoelectric cooperative memristor array,a convolutional neural network for vehicle type recognition is constructed,which solves the problem of zero weight and negative weight complexity.In regard to energy efficiency,the neural network built with this device operates at a power level of only 10^(-3)W,representing a reduction of more than 4 orders of magnitude compared with a standard central processor.Hence,the photoelectric memristor proposed in this work provides a new idea for neuromorphic computing and is expected to promote the development of energy-efficient brain-like computing.

传统晶体管在大规模神经形态系统中面临计算能力不足和功耗过高的严峻挑战.为了解决这些关键问题,我们提出了一种基于氧化锌-氧化铟锡/氧化钨异质结的光电忆阻器,作为解决上述问题的有效方案.通过应用不同类型的电和光信号,该器件成功模拟了多种突触功能,包括短期/长期突触可塑性以及短期/长期记忆.引入氧化锌-氧化铟锡功能层增强了基于氧化钨忆阻器的光响应,并展示了光调制下的“学习-遗忘-再学习”行为.此外,在光电协同忆阻器阵列的基础上,我们构建了用于车型识别的卷积神经网络,解决了零权重和负权重实现复杂的问题.在功耗方面,使用该器件构建的神经网络仿真功率仅为10^(-3)W,与标准中央处理器相比至少降低了4个数量级.因此,这项工作所提出的光电忆阻器为神经形态计算提供了新的思路,有望推动高能效类脑计算的发展.