用于神经形态计算的自旋突触的全电控制:将多态记忆与量子化联系起来,以实现高效的神经网络
Tzu-Chuan Hsin1, Chun-Yi Lin1, Po-Chuan Wang1
1Department of Materials Science and Engineering, National Taiwan University, Taipei, 10617, Taiwan.
Advanced science (Weinheim, Baden-Wurttemberg, Germany)
|April 26, 2025
概括
这项研究引入了新的spintronic自旋突触装置,用于节能的神经形态计算. 倾斜的异质性装置显示出神经网络中具有高精度的复杂突触仿真的前景.
科学领域:
- 这就是Spintronics.
- 神经形态计算是一种神经形态计算.
- 材料科学 材料科学 材料科学
背景情况:
- 神经形态计算需要能效的记忆设备来模仿突触行为.
- 当前的设备在准确性和适应性方面面临着大脑启发系统的挑战.
研究的目的:
- 开发和评估用于神经形态应用的全电控制,无场自旋突触装置.
- 为了对设备性能进行基准测试,专注于循环对循环的变化和多状态内存功能.
主要方法:
- 设计了三种旋转器件结构:Néel色皮效应,间层Dzyaloshinskii-Moriya相互作用 (i-DMI) 和倾斜的异构性.
- 使用基准测试框架来评估周期间 (CTC) 的变化.
- 设备在卷积神经网络 (CNN) 中实现了训练后定量化.
主要成果:
- 倾斜的异性变异装置显示出11个状态的记忆,CTC变化最小 (2%).
- 在 ResNet-18 中在 CIFAR-10 数据集上的每通道量子化实现了高分类准确性 (在 min-max 中高达 81.51% ,在 MSE 观察者中高达 81.12%).
- 性能接近基线准确度,验证了设备的潜力.
结论:
- 无场自旋突触突触为先进的神经形态架构提供了一个有希望的,面积高效的解决方案.
- 这些设备整合了多状态功能和强大的切换,推进了以神经过程为灵感的节能,高性能计算.
相关概念视频
01:25Neural Circuits
1.1K
Neural circuits and neuronal pools are two of the main structures found in the nervous system. Neural circuits are networks of neurons that work together to carry out a specific task or process. They consist of interconnected neurons and glial cells, which provide structural and metabolic support.
Neuronal pools are collections of nerve cells with similar functions and interact through chemical and electrical signals. These pools include both interneurons (the central neural circuit nodes that...
Neuronal pools are collections of nerve cells with similar functions and interact through chemical and electrical signals. These pools include both interneurons (the central neural circuit nodes that...
1.1K
01:28Electrical Synapses
8.2K
Electrical synapses found in all nervous systems play important and unique roles. In these synapses, the presynaptic and postsynaptic membranes are very close together (3.5 nm) and are actually physically connected by channel proteins forming gap junctions.
Gap junctions allow the current to pass directly from one cell to the next. In contrast, in the chemical synapse, the neurotransmitters carry the information through the synaptic cleft from one neuron to the next. They consist of two...
Gap junctions allow the current to pass directly from one cell to the next. In contrast, in the chemical synapse, the neurotransmitters carry the information through the synaptic cleft from one neuron to the next. They consist of two...
8.2K
01:28Neuronal Communication
816
Neurons, the fundamental units of the brain and nervous system, communicate through complex electrochemical signals that underpin all cognitive and bodily functions. This communication is primarily facilitated by a process involving the generation and propagation of an action potential along the axon of the neuron. When the internal electrical charge of a neuron surpasses a certain threshold, an action potential is triggered. This rapid change in voltage travels swiftly along the axon to the...
816
01:28Integration of Synaptic Events
1.5K
Synaptic integration mainly includes the summation of graded potentials. Graded potentials, regardless of their type, cause subtle alterations in membrane voltage, resulting in either depolarization or hyperpolarization. These incremental changes, when combined or summed, can propel the neuron toward its threshold. Consider, for example, a membrane experiencing a +15 mV shift, causing it to depolarize from -70 mV to -55 mV. In this scenario, graded potentials govern the membrane's ability to...
1.5K
01:19The Role of Ion Channels in Neuronal Computation
3.2K
A postsynaptic neuron usually receives numerous impulses from several other presynaptic neurons. The axon hillock of the postsynaptic neuron integrates all these signals and determines the likelihood of firing an action potential.
Sometimes a single EPSP is strong enough to induce an action potential in the postsynaptic neuron. However, multiple presynaptic inputs must often create EPSPs around the same time for the postsynaptic neuron to be sufficiently depolarized to fire an action potential....
Sometimes a single EPSP is strong enough to induce an action potential in the postsynaptic neuron. However, multiple presynaptic inputs must often create EPSPs around the same time for the postsynaptic neuron to be sufficiently depolarized to fire an action potential....
3.2K
01:03Atomic Nuclei: Nuclear Spin State Overview
903
NMR-active nuclei have energy levels called 'spin states' that are associated with the orientations of their nuclear magnetic moments. In the absence of a magnetic field, the nuclear magnetic moments are randomly oriented, and the spin states are degenerate. When an external magnetic field is applied, the spin states have only 2 + 1 orientations available to them. A proton with = ½ has two available orientations. Similarly, for a quadrupolar nucleus with a nuclear spin value of...
903