Multilevel resistive switching and synaptic plasticity of nanoparticulated cobaltite oxide memristive device

Multilevel resistive switching(RS)is a key property to embrace the full potential of memristive devices for non-volatile memory and neuromorphic computing applications.In this study,we employed nanopar-ticulated cobaltite oxide(Co3O4)as a model material to demonstrate the multilevel RS and synaptic learning capabilities because of its multiple and stable redox state properties.The Pt/Co3O4/Pt memris-tive device exhibited tunable RS properties with respect to different voltages and compliance currents(CC)without the electroforming process.That is,the device showed voltage-dependent RS at a higher CC whereas CC-dependent RS was observed at lower CC.The device showed four different resistance states during endurance and retention measurements and non-volatile memory results indicated that the CC-based measurement had less variation.Besides,we investigated the basic and complex synap-tic plasticity properties using the analog current-voltage characteristics of the Pt/Co3O4/Pt device.In particular,we mimicked the potentiation-depression and four-spike time-dependent plasticity(STDP)rules such as asymmetric Hebbian,asymmetric anti-Hebbian,symmetric Hebbian,and symmetric anti-Hebbian learning rules.The results of the present work indicate that the cobaltite oxide is an excellent nanomaterial for both multilevel RS and neuromorphic computing applications.     

A review:Photonics devices,architectures,and algorithms for optical neural computing

The explosive growth of data and information has motivated various emerging non-von Neumann computational approaches in the More-than-Moore era.Photonics neuromorphic computing has attracted lots of attention due to the fascinating advantages such as high speed,wide bandwidth,and massive parallelism.Here,we offer a review on the optical neural computing in our research groups at the device and system levels.The photonics neuron and photonics synapse plasticity are presented.In addition,we introduce several optical neural computing architectures and algorithms including photonic spiking neural network,photonic convolutional neural network,photonic matrix computation,photonic reservoir computing,and photonic reinforcement learning.Finally,we summarize the major challenges faced by photonic neuromorphic computing,and propose promising solutions and perspectives.     

视觉模型中的抑制突触可塑性研究

论文研究了神经元抑制性突触的可塑性和在视觉模型中的作用。在一个三层前馈视觉模型中,除了考虑兴奋性突触的可塑性外,将抑制性突触的可塑性加入到模型中。比较了赫布学习和反赫布学习窗口,发现抑制性突触在反赫布学习窗口下,符合视觉系统的生物特性。并进一步研究了抑制性突触可塑性对眼优势可塑性的影响,发现抑制性突触权重在可塑性学习下增大,是导致眼优势关键期关闭重要的原因。研究结果一方面有助于更好地理解脑视觉神经网络的可塑性,另一方面对于治疗斜视、弱视等视觉疾病提供理论依据。     

仿生型脉冲神经网络学习算法和网络模型

为解决脉冲神经网络训练困难的问题,基于仿生学思路,提出脉冲神经网络的权值学习算法和结构学习算法,设计一种含有卷积结构的脉冲神经网络模型,搭建适合脉冲神经网络的软件仿真平台。实验结果表明,权值学习算法训练的网络对MNIST数据集识别准确率能够达到84.12%,具备良好的快速收敛能力和低功耗特点;结构学习算法能够自动生成网络结构,具有高度生物相似性。     

Synaptic Suppression Triplet‐STDP Learning Rule Realized in Second‐Order Memristors

The synaptic weight modification depends not only on interval of the pre‐/postspike pairs according to spike‐timing dependent plasticity (classical pair‐STDP), but also on the timing of the preceding spike (triplet‐STDP). Triplet‐STDP reflects the unavoidable interaction of spike pairs in natural spike trains through the short‐term suppression effect of preceding spikes. Second‐order memristors with one state variable possessing short‐term dynamics work in a way similar to the biological system. In this work, the suppression triplet‐STDP learning rule is faithfully demonstrated by experiments and simulations using second‐order memristors. Furthermore, a leaky‐integrate‐and‐fire (LIF) neuron is simulated using a circuit constructed with second‐order memristors. Taking the advantage of the LIF neuron, various neuromimetic dynamic processes, including local graded potential leaking out, postsynaptic impulse generation and backpropagation, and synaptic weight modification according to the suppression triplet‐STDP rule, are realized. The realized weight‐dependent pair‐ and triplet‐STDP rules are clearly in line with findings in biology. The physically realized triplet‐STDP rule is powerful in developing direction and speed selectivity for complex pattern recognition and tracking tasks. These scalable artificial synapses and neurons realized in second‐order memristors can intrinsically capture the neuromimetic dynamic processes; they are the promising building blocks for constructing brain‐inspired computation systems.     

基于ZYNQ集群的神经形态计算加速研究与实现

基于脉冲神经网络（SNN）的神经形态计算由于工作机理更接近于生物大脑,被认为有望克服深度学习的不足而成为解决人工智能问题的更佳途径。但是如何满足高性能、低功耗和适应规模伸缩需求是神经形态计算系统需要解决的挑战性问题。基于FPGA异构计算平台ZYNQ集群,在NEST类脑仿真器上,重点解决了具有脉冲时间依赖可塑性（STDP）突触计算复杂度高、并行度低、硬件资源占用大的问题。实验结果表明,设计的方法在8节点ZYNQ 7030集群上,性能是Xeon E5-2620 CPU的14.7倍。能效比方面,是Xeon E5-2620 CPU的51.6倍,是8节点ARM Cortex-A9的20.6倍。     

Delay learning and polychronization for reservoir computing

We propose a multi-timescale learning rule for spiking neuron networks, in the line of the recently emerging field of reservoir computing. The reservoir is a network model of spiking neurons, with random topology and driven by STDP (spike-time-dependent plasticity), a temporal Hebbian unsupervised learning mode, biologically observed. The model is further driven by a supervised learning algorithm, based on a margin criterion, that affects the synaptic delays linking the network to the readout neurons, with classification as a goal task. The network processing and the resulting performance can be explained by the concept of polychronization, proposed by Izhikevich [Polychronization: computation with spikes, Neural Comput. 18(2) (2006) 245–282], on physiological grounds. The model emphasizes that polychronization can be used as a tool for exploiting the computational power of synaptic delays and for monitoring the topology and activity of a spiking neuron network.     

低功耗非对称性可调STDP突触电路设计

基于65 nm CMOS工艺设计了一种可用于脉冲神经网络系统的低功耗、高能效、结构紧凑的突触电路.突触电路采用开关电容电路结构,直接接收来自神经元电路的脉冲信号,根据脉冲时间依赖可塑性(STDP)学习规则调节突触权重,并实现了权重学习窗口的非对称性调节,使突触电路可以适应不同情况.仿真结果表明,突触电路耗能约为0.4 ...     

基于多层忆阻脉冲神经网络的强化学习及应用

人工神经网络（Artificial neural networks,ANNs）与强化学习算法的结合显著增强了智能体的学习能力和效率.然而,这些算法需要消耗大量的计算资源,且难以硬件实现.而脉冲神经网络（Spiking neural networks,SNNs）使用脉冲信号来传递信息,具有能量效率高、仿生特性强等特点,且有利于进一步实现强化学习的硬件加速,增强嵌入式智能体的自主学习能力.不过,目前脉冲神经网络的学习和训练过程较为复杂,网络设计和实现方面存在较大挑战.本文通过引入人工突触的理想实现元件——忆阻器,提出了一种硬件友好的基于多层忆阻脉冲神经网络的强化学习算法.特别地,设计了用于数据——脉冲转换的脉冲神经元;通过改进脉冲时间依赖可塑性（Spiking-timing dependent plasticity,STDP）规则,使脉冲神经网络与强化学习算法有机结合,并设计了对应的忆阻神经突触;构建了可动态调整的网络结构,以提高网络的学习效率;最后,以Open AI Gym中的CartPole-v0（倒立摆）和MountainCar-v0（小车爬坡）为例,通过实验仿真和对比分析,验证了方案的有效性和相对于传统强化学习方法的优势.     

CMOL implementation of spiking neurons and spike‐timing dependent plasticity

Successful implementation of spiking neural networks onto CMOS‐Molecular (CMOL) architecture has already been proposed, but the ability of dynamic learning has not yet been addressed. Here, we propose a spiking neural topology with spike‐timing‐dependent learning ability and provide its basic building blocks that are easily mapped onto CMOL architecture. The learning method modifies state of synaptic switches, using spatially and temporally local information which is available at the synapse when state modification is performed. The performance of the proposed topology is analyzed with regards to pre‐ and post‐synaptic spike timing, and simulation results are provided for a synapse with spike‐timing‐dependent plasticity properties. Furthermore, its performance as spike‐timing correlation learning and synchrony detection in a small feed‐forward network is demonstrated as a case example. Copyright © 2010 John Wiley & Sons, Ltd.