摘要
类脑计算仿照人脑神经网络通过人工神经元和突触器件实现信息处理,有望为加速人工智能发展、解决复杂认知问题提供更为高效节能的方案.由于传统半导体器件与类脑计算要求(如非易失性等)不匹配,其进一步发展需要从硬件层面着手,研发新型器件.自旋电子学器件天然具备非易失性,是类脑计算硬件的潜在选择,特别是磁斯格明子,因其独特的拓扑性质、稳定性、低功耗及类孤子特性,有望成为类脑器件的高效信息载体.本文简要回顾了近年来磁斯格明子类脑器件的发展,从斯格明子器件相关的基本物性展开,进一步阐述了基于斯格明子的人工神经元和突触的基本工作原理及相关科研进展,最后对磁斯格明子类脑器件未来发展中面临的机遇与挑战进行了讨论.
Neuromorphic computing mimics the brain’s neural networks, using artificial neurons and synapses to process information.This approach holds promise for more efficient and energy-saving solutions in artificial intelligence (AI) applications.However, traditional semiconductor devices struggle to meet the demands of neuromorphic computing, particularly interms of non-volatility and integrated memory-processing functions. To advance this field, new hardware solutions arerequired. Spintronic devices, which utilize the spin of electrons in addition to their charge, offer a promising alternative forneuromorphic computing due to their inherent non-volatility. Among these, magnetic skyrmions—nanoscale, vortex-likemagnetic structures—are especially attractive. Their unique topological properties, stability, and low-power manipulationmake them ideal for use as information carriers in neuromorphic systems. Skyrmions are stabilized in materials through theDzyaloshinskii-Moriya interaction (DMI), which breaks spatial inversion symmetry. This stability, combined with theability to move skyrmions using low-energy electrical currents, makes them highly efficient for information storage andprocessing. Unlike conventional magnetic domain walls, skyrmions are more stable against external perturbations,providing an advantage in neuromorphic applications, where robustness and energy efficiency are critical.This review examines recent developments in skyrmion-based neuromorphic devices. It begins by outlining thefundamental physical properties of skyrmions, including their stabilization mechanism and electrical current-drivenmotion. The review then discusses how these properties can be exploited to replicate the functions of artificial neurons andsynapses—the core components of neuromorphic systems. In skyrmion-based artificial neurons, the motion of a skyrmionthrough a nanowire can simulate the integration of input signals. When the skyrmion reaches a specific point, it triggers theneuron to “fire”, mimicking the behavior of biological neurons. Similarly, skyrmion-based synapses regulate the strength ofconnections between neurons by adjusting the number of skyrmions in a region, thereby modifying the synaptic weight—akey feature for learning and memory. We then explore different device architectures proposed for implementing theseskyrmion-based neurons and synapses. These include structures based on magnetic tunnel junctions and nanostructuredspintronic materials, offering benefits in scalability, energy efficiency, and non-volatility. Furthermore, some researchshows the potential for these devices to integrate with conventional CMOS technology. Despite significant progress,challenges remain before skyrmion-based neuromorphic devices can be implemented into neuromorphic systems. Theseinclude improving skyrmion stability at room temperatures, optimizing material systems for better performance, andachieving integration with existing electronics. Additionally, most current research is either theoretical or in earlyexperimental stages, with large-scale demonstrations yet to be realized. Magnetic skyrmions offer a novel path forneuromorphic computing, providing a potential solution to the limitations of traditional semiconductor devices. Withfurther research, skyrmion-based systems could lead to highly efficient, low-power computing architectures capable ofemulating the brain’s complex functions.
作者
刘艺舟
杜海峰
Yizhou Liu;Haifeng Du(High Magnetic Field Laboratory,Hefei Institutes of Physical Science,Chinese Academy of Sciences,Hefei 230031,China)
出处
《科学通报》
北大核心
2025年第13期1884-1892,共9页
Chinese Science Bulletin
基金
中国科学院稳定支持基础研究领域青年团队计划(YSBR-084)资助。
关键词
磁斯格明子
类脑计算
神经拟态计算
拓扑磁结构
自旋电子学
magnetic skyrmion
brain-inspired computing
neuromorphic computing
topological magnetic texture
spintronics
