CAREER: Capturing Biological Behavior in Three-Terminal Magnetic Tunnel Junction Synapses and Neurons for Fully Spintronic Neuromorphic Computing

职业：捕捉三端磁隧道连接突触和神经元的生物行为，以实现全自旋电子神经形态计算

• 批准号: 1940788
• 负责人: Jean Anne Incorvia
• 金额: $ 50万
• 依托单位: University of Texas at Austin
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2020
• 资助国家: 美国
• 起止时间: 2020-03-01 至 2025-02-28
• 项目状态: 未结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1940788&HistoricalAwards=false
• 关键词: CAREER; Capturing; Biological; Behavior; Three

Brain-inspired computing is a revolution in computing that is already seeing applications in a myriad of areas, from image recognition to developing learning rules that allow computers to intelligently process big data sets. This field is inspired by the human brain, which is very efficient at certain tasks. For example, the brain can recognize a face or voice using a million times less power than a modern supercomputer. But, so far machine learning has largely focused on restructuring how the computer is put together, but where the building blocks themselves are silicon transistors. Neuroscientists have recognized certain behaviors of neurons and synapses that are central to processing and learning. This is an opportunity to use new materials and new physics to capture the biological behavior of the brain in artificial neurons and synapses. In the proposed work, physics present in nanoscale magnetic devices will be used to model, build, and measure artificial magnetic neurons and synapses that capture biological behavior, such as neurons that build up energy over time, relax that energy when they are not stimulated, and have inter-neuron interactions. Synapses, the connection between neurons, will be designed, built, and measured where the strength of the connection will be adjusted based on the timing of signals from the neurons. The neurons and synapses will then be connected into circuits that make use of the more advanced biological behavior. The performance will be compared to silicon and other emerging materials. This research will further the understanding of one solution for next-generation computing. It will lead to broad impacts in all areas of big data, from biology (e.g. genome sequencing) to defense (e.g. tracking a flying object in real time) to Internet of Things (e.g. sensors in smart cities). In addition to training the next generation of scientists and engineers, the proposed education program emphasizes exposing young scientists to the creative process of research. This will be accomplished by creating nanotechnology hands-on activity events for K-12 students, hosting high school summer research students, and creating graduate course curricula on the creative design of new magnetic devices.The goal of this CAREER proposal is to model, build, and measure three-terminal magnetic tunnel junction (3T-MTJ) devices that can, as closely as possible, capture the biological behavior of the brain. Like the brain, compared to traditional computers, magnetic materials have relatively slow switching but with low voltage, nonvolatility, and with digital, analog, stochastic, and oscillatory behavior. Little research has moved beyond simple multi-weight synapses and stochastic neurons to capture more robust biology that allows the brain to perform data-intensive tasks with low power consumption and in real time. The proposed research method to address this problem is to understand the biological behavior by working with neuroscientists; map the biological behavior onto magnetic properties; develop and simulate the device; fabricate and test the device; and then measure the switching energy and probability in comparison to biology and other artificial neurons and synapses (e.g. silicon and other resistive memories). Neuromorphic computing is a promising approach to the ever-increasing present and future demands for real-time processing of massive amounts of data. The proposed work will test the hypothesis that the similar properties of magnetic materials to the brain make magnetic devices suitable for neuromorphic computing. It will also show how the magnetic behaviors differ from the brain, which will enable new circuit and architecture design. The proposed work will provide a monolithic platform, where the same magnetic thin film stack will be used for both synapses and neurons, and which can be extended beyond the scope of this work to new and creative devices.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

大脑启发计算是计算领域的一场革命，已经在无数领域得到应用，从图像识别到开发允许计算机智能处理大数据集的学习规则。这一领域受到人类大脑的启发，人类大脑在某些任务上非常有效。例如，大脑可以识别人脸或声音，使用的功率比现代超级计算机少一百万倍。但是，到目前为止，机器学习主要集中在重新构建计算机的组装方式，但构建模块本身是硅晶体管。神经科学家已经认识到神经元和突触的某些行为对处理和学习至关重要。这是一个使用新材料和新物理学来捕捉人工神经元和突触中大脑生物行为的机会。在拟议的工作中，纳米磁性器件中存在的物理学将用于建模，构建和测量捕获生物行为的人工磁性神经元和突触，例如随着时间的推移积累能量的神经元，当它们没有受到刺激时放松能量，并具有神经元间的相互作用。突触，神经元之间的连接，将被设计，构建和测量，其中连接的强度将根据神经元信号的时序进行调整。然后，神经元和突触将被连接成电路，利用更先进的生物行为。性能将与硅和其他新兴材料进行比较。这项研究将进一步了解下一代计算的一个解决方案。它将对大数据的所有领域产生广泛影响，从生物学（例如基因组测序）到国防（例如真实的跟踪飞行物体）再到物联网（例如智能城市中的传感器）。除了培养下一代科学家和工程师外，拟议的教育计划还强调让年轻科学家接触研究的创造性过程。这将通过为K-12学生创建纳米技术实践活动活动，举办高中暑期研究生，以及创建新磁性器件创意设计的研究生课程来实现。该CAREER提案的目标是建模，构建和测量三端磁性隧道结（3 T-MTJ）器件，该器件可以尽可能接近地捕获大脑的生物行为。与大脑一样，与传统计算机相比，磁性材料的开关速度相对较慢，但具有低电压，非易失性，并且具有数字，模拟，随机和振荡行为。除了简单的多权重突触和随机神经元之外，很少有研究能够捕捉到更强大的生物学特性，使大脑能够以低功耗和真实的时间执行数据密集型任务。解决这个问题的拟议研究方法是通过与神经科学家合作来理解生物行为;将生物行为映射到磁特性;开发和模拟设备;制造和测试设备;然后测量与生物学和其他人工神经元和突触（例如硅和其他电阻存储器）相比的开关能量和概率。神经形态计算是一种很有前途的方法，以不断增长的现在和未来的需求，实时处理大量的数据。拟议的工作将测试磁性材料与大脑相似的特性使磁性设备适合神经形态计算的假设。它还将展示磁行为与大脑的不同之处，这将使新的电路和架构设计成为可能。这项提议的工作将提供一个整体平台，其中相同的磁性薄膜堆栈将用于突触和神经元，并且可以扩展到新的和创造性的设备。该奖项反映了NSF的法定使命，并通过使用基金会的智力价值和更广泛的影响审查标准进行评估，被认为值得支持。

项目成果

Controllable Reset Behavior in Domain Wall–Magnetic Tunnel Junction Artificial Neurons for Task-Adaptable Computation

用于任务适应性计算的磁畴壁磁隧道结人工神经元中的可控重置行为

• DOI: 10.1109/lmag.2021.3069666
• 发表时间: 2021
• 期刊: IEEE Magnetics Letters
• 影响因子: 1.2
• 作者: Liu, Samuel;Bennett, Christopher;Friedman, Joseph;Marinella, Matthew;Paydarfar, David;Incorvia, Jean Anne
• 通讯作者: Incorvia, Jean Anne

Lateral inhibition in magnetic domain wall racetrack arrays for neuromorphic computing

用于神经形态计算的磁畴壁跑道阵列的横向抑制

• DOI: 10.1117/12.2568870
• 发表时间: 2020
• 期刊: Spintronics XIII
• 作者: Cui, Can;Akinola, Otitoaleke G.;Hassan, Naimul;Bennett, Christopher H.;Marinella, Matthew J.;Friedman, Joseph S.;Incorvia, Jean Anne
• 通讯作者: Incorvia, Jean Anne

Magnetic skyrmions and domain walls for logical and neuromorphic computing

用于逻辑和神经形态计算的磁性斯格明子和畴壁

• DOI: 10.1088/2634-4386/acc6e8
• 发表时间: 2023
• 期刊: Neuromorphic Computing and Engineering
• 作者: Hu, Xuan;Cui, Can;Liu, Samuel;Garcia-Sanchez, Felipe;Brigner, Wesley H;Walker, Benjamin W;Edwards, Alexander J;Xiao, T Patrick;Bennett, Christopher H;Hassan, Naimul
• 通讯作者: Hassan, Naimul

Intrinsic Lateral Inhibition Facilitates Winner-Take-All in Domain Wall Racetrack Arrays for Neuromorphic Computing

内在横向抑制促进神经形态计算领域壁跑道阵列中的赢家通吃

• DOI: 10.1109/iscas48785.2022.9937784
• 发表时间: 2022
• 期刊: 2022 IEEE International Symposium on Circuits and Systems (ISCAS
• 作者: Cui, Can;Akinola, Otitoaleke G.;Hassan, Naimul;Bennett, Christopher H.;Marinella, Matthew J.;Friedman, Joseph S.;Incorvia, Jean Anne
• 通讯作者: Incorvia, Jean Anne

Fabrication of Domain Wall based Magnetic Tunnel Junction Devices with Intrinsic Neuromorphic Functionality

具有内在神经形态功能的基于磁畴壁的磁隧道结器件的制造

• DOI: 10.1109/tmrc59626.2023.10264022
• 发表时间: 2023
• 期刊: 2023 IEEE 34th Magnetic Recording Conference (TMRC
• 作者: Leonard, Thomas;Liu, Samuel;Jin, Harrison;Friedman, Joseph S.;Bennett, Christopher;Incorvia, Jean Anne
• 通讯作者: Incorvia, Jean Anne