EAGER: An Experimental Exploration for Spin-Based Neuromorphic Computing

最近的自旋电子学实验揭示了在单个器件结构中模仿大脑功能的机会，该结构可以在非常低的终端电压下操作。这些非易失性存储器技术可以以交叉阵列的方式布置，以通过应用基尔霍夫定律来实现神经形态系统中的内存计算。自旋电子学使能的神经形态计算具有显著提高机器学习硬件效率的潜力。然而，实验研究仍处于起步阶段。该项目的目标是通过在底层硬件基板中模仿生物大脑，利用自旋轨道耦合的功能特性，实验性地演示计算原语的仿真。EAGER计划的重点是通过实验证明基于自旋的神经形态器件的可行性，与互补金属氧化物半导体技术相比，这种器件有可能在性能和能耗方面取得显着改善。如果成功的话，这项研究可能会导致计算范式的发展，这种范式本质上利用纳米磁性设备作为构建类脑硬件的机制。来自宾夕法尼亚州立大学Schreyer荣誉学院的研究生和本科生将参与该项目的研究活动。PI计划将该项目的结果整合到电气工程系K-12夏令营中。之前的实验工作主要集中在单个器件表征上，而没有解决可扩展性和系统级演示的问题。该提案旨在通过以下研究重点弥合这一差距：（i）在室温下制造和表征基于重金属的霍尔棒结构，以实验证明各种神经和突触功能的可行性，包括泄漏尖峰神经元、随机尖峰神经元、尖峰定时依赖性可塑性和短期可塑性。器件结构将由铁磁体（CoFe/CoFeB）-重金属（Ta/Pt）霍尔条组成，其中流过器件的电流将用于切换磁体，并将使用异常霍尔效应检测最终状态。(ii)该器件将表征性能和非理想性，如磁化钉扎、随机编程、热噪声诱导开关和编程分辨率。(iii)将制造这种设备的阵列，以研究设备到设备和周期到周期变化的影响，并演示原型计算系统。还将努力通过纳入可用于预测大规模系统性能的设备约束和非理想性来开发逼真的设备模型。成功完成该项目为机器学习平台的效率的变革性改进奠定了基础，该平台具有在自主系统中执行实时决策的能力。该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Recent experiments in spintronics have revealed the opportunity of mimicking brain-like functionalities in a single device structure that can be operated at very low terminal voltages. These non-volatile memory technologies can be arranged in crossbar array fashion to realize In-Memory Computing in neuromorphic systems by application of Kirchoff's law. Spintronics enabled neuromorphic computing has the potential of enabling a significant increase in efficiency of machine learning hardware. However, experimental investigations are still in the infancy stage. The goal of this project is to experimentally demonstrate the emulation of the computational primitives by mimicking biological brain in the underlying hardware substrate exploiting the functional properties of Spin-Orbit Coupling. The EAGER program is focused to prove the feasibility of experimentally demonstrating spin-based neuromorphic devices that has the potential to enable significant improvements in performance and energy consumption as compared to Complementary Metal Oxide Semiconductor technologies. If successful, the research can lead to the development of computing paradigms that inherently exploits nanomagnetic devices as a mechanism to build brain-like hardware. Graduate students and undergraduates from Penn State's Schreyer Honors College will be engaged in the research activities of this project. The PI plans to integrate the results from this project into the Electrical Engineering departmental K-12 summer camp.Prior experimental efforts have mainly focused on single device characterizations without addressing the issues of scalability and system-level demonstrations. This proposal aims to bridge that gap through the following research thrusts: (i) Fabrication and characterization of heavy-metal based Hall-bar structures at room temperatures to experimentally demonstrate the feasibility of various neural and synaptic functionalities including leaky-spiking neuron, stochastic-spiking neuron, spike-timing dependent plasticity, and short-term plasticity. The device structure will consist of a ferromagnet (CoFe/CoFeB) – heavy metal (Ta/Pt) Hall-bar where current flowing through the device will be used to switch the magnet and the final state will be detected using anomalous Hall effect. (ii) The device will be characterized for performance and non-idealities like magnetization pinning, stochastic programming, thermal noise induced switching, and programming resolution. (iii) Arrays of such devices will be fabricated to investigate the impact of device-to-device and cycle-to-cycle variations and demonstrate a prototype computing system. Efforts will also be placed to develop realistic device models by incorporating device constraints and non-idealities that can be used to predict performance of large-scale systems. Successful completion offers the basis for transformative improvements in the efficiency of machine learning platforms having the capability of performing real-time decision-making in autonomous systems.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.