基于忆容器件的神经形态计算研究进展

Research progress of neuromorphic computation based on memcapacitors

人工智能的快速发展需要人工智能专用硬件的快速发展,受人脑存算一体、并行处理启发而构建的包含突触与神经元的神经形态计算架构,可以有效地降低人工智能中计算工作的能耗.记忆元件在神经形态计算的硬件实现中展现出巨大的应用价值;相比传统器件,用忆阻器构建突触、神经元能极大地降低计算能耗,然而在基于忆阻器构建的神经网络中,更新、读取等操作存在由忆阻电压电流造成的系统性能量损失.忆容器作为忆阻器衍生器件,被认为是实现低耗能神经网络的潜在器件,引起国内外研究者关注.本文综述了实物/仿真忆容器件及其在神经形态计算中的最新进展,主要包括目:前实物/仿真忆容器原理与特性,代表性的忆容突触、神经元及神经形态计算架构,并通过总结近年来忆容器研究所取得的成果,对当前该领域面临的挑战及未来忆容神经网络发展的重点进行总结与展望.

The rapid development of artificial intelligence(AI)requires one to speed up the development of the domain-specific hardware specifically designed for AI applications.The neuromorphic computing architecture consisting of synapses and neurons,which is inspired by the integrated storage and parallel processing of human brain,can effectively reduce the energy consumption of artificial intelligence in computing work.Memory components have shown great application value in the hardware implementation of neuromorphic computing.Compared with traditional devices,the memristors used to construct synapses and neurons can greatly reduce computing energy consumption.However,in neural networks based on memristors,updating and reading operations have system energy loss caused by voltage and current of memristors.As a derivative of memristor,memcapacitor is considered as a potential device to realize a low energy consumption neural network,which has attracted wide attention from academia and industry.Here,we review the latest advances in physical/simulated memcapacitors and their applications in neuromorphic computation,including the current principle and characteristics of physical/simulated memcapacitor,representative synapses,neurons and neuromorphic computing architecture based on memcapacitors.We also provide a forward-looking perspective on the opportunities and challenges of neuromorphic computation based on memcapacitors.