摘要
随着数据信息的迅速膨胀,现代基于von Neumann架构的计算机正面临着严峻的挑战,像人脑一样能够对信息进行学习、记忆和灵活处理的智能计算机是未来计算机发展的方向和目标.人脑几乎控制着人类所有复杂的生命活动,大脑神经元间的信息传递依赖于名为"突触"的结构,其突出特点——突触可塑性是学习与记忆的重要分子基础,因此突触仿生和突触可塑性模拟被认为是实现高效类脑人工神经网络的第一步.忆阻器作为第4种基本电路元件,拥有独特的类神经突触非线性电学传输特性,它的出现和发展为实现这一目标提供了可能.本文全面总结了近年来各科研小组使用不同的忆阻器件和各种实验方法成功模拟的突触可塑性,包括按记忆时间长短分类的短时程可塑性(双脉冲抑制、双脉冲易化、强直后增强)和长时程可塑性,以及放电时间依赖可塑性、放电速率依赖可塑性、经验学习、非联想性学习功能,还有用多个忆阻器及其他基本元器件(电阻、电感、电容、晶体管等)复合模拟的联想性学习、突触缩放等复杂功能,对比了不同器件的优缺点.最后,简述了当前忆阻器研究存在的问题和挑战,并对忆阻器在突触仿生中的研究前景进行了展望.
With the rapid expansion of data information, modern computers based on the von Neumann archi- tecture are facing severe challenges. Intelligent computers that can learn, store, and process information flexibly like a human brain will be the direction and goal of computers＇ development. Brain controls almost all the com- plex life activities of human beings, and information transmission between cerebral neurons relies on the structure called ＂synapse＂, whose outstanding property -- synaptic plasticity-- is thought to be an important molecular basis of learning and memory. Therefore, it is widely believed that emulation of synapse and synaptic plasticity is the first step to realize effective artificial neural networks. Owing to the birth and development of the fourth fundamental passive circuit elements, memristors, which have unique nonlinear synaptic electrical transmission characteristics, it is possible to achieve this goal. Thus, over the past few years, a great deal of efforts have been made in mimicking synapse functions though memristors. In this review, recent simulations of synaptic plasticity using different memristor devices and various methods are comprehensively summarized, including short-term plasticity （paired-pulse depression, paired-pulse facilitation, and post-tetanic potentiation）, long-term plasticity, spiking-rate-dependent plasticity, spiking-timing-dependent plasticity, learning experience, associative memory, and synaptic scaling. Finally, the current problems faced in the research and the development prospects in this area are briefly discussed.
出处
《中国科学：信息科学》
CSCD
北大核心
2018年第2期115-142,共28页
Scientia Sinica(Informationis)
基金
国家重点基础研究发展计划(批准号:2014CB648300
2015CB932200)
国家自然科学基金(批准号:61475074
61204095)
江苏省自然科学基金(批准号:BK20160088)
省级大学生创新训练计划(批准号:SYB2016009)
国家自然科学优秀青年基金(批准号:21322402)
长江学者和创新团队(批准号:IRT 15R37)
中国江苏省教育委员会自然科学基金(批准号:14KJB510027)
江苏省高校优势学科建设工程(PAPD)资助项目
关键词
忆阻器
突触
突触可塑性
人工神经网络
memristor, synapse, synaptic plasticity, artificial neural networks
