CAREER: Toward energy-efficient bio-inspired magnonic processing with nanomagnetic arrays

该项目由材料研究部的凝聚态物理计划和建立的激励竞争研究计划(EPSCoR)联合资助。非技术描述：人工智能(AI)的迅猛发展使强大的工具和应用程序的创造成为可能，这些工具和应用程序在几年前是无法想象的。然而，随着人工智能和机器学习的快速发展，相关的能源成本和温室气体排放正在爆炸式增长。这一巨大的不可持续的趋势威胁着社会实现净零的未来。因此，迫切需要低功率计算和人工智能处理的范式转变。该项目通过提供与磁性纳米结构中的基本激发有关的基础知识和技术来帮助应对这一历史性挑战，以创建一种从大脑中获得灵感的变革性计算方案。当前的计算体系结构依赖于在独立的内存和处理器之间不断地穿梭数据，这是非常低效的。此外，目前的计算平台是基于电子电荷的流动，导致以焦耳加热的形式耗散。为了绕过这些问题，研究小组的目标是利用纳米磁铁相互作用的网络中的动态，通过A)缓解处理器-内存信息传输瓶颈和B)实现基于波而不是移动电荷的数据传输和处理。该项目的教育宣传部分促进公众更多地参与科学研究。教育目标旨在让多层次的波物理学习：1)为终身学习者开发了一门新课程；2)通过创建可访问的波演示，为与不同学生群体合作的学校教师开发了培训计划。技术描述：自旋波及其量子--磁子--是磁系统的基本集体激发。磁子可以在不移动电荷的情况下传输和处理信息，因此，磁子设备比电子设备的耗散更小。纳米磁性阵列类似于神经网络，在同一个单元中提供记忆和计算能力：它们可以保留存储在磁化方向上的信息，并通过磁子激发来处理这些信息。该项目通过确定相互作用的纳米磁铁光刻阵列中的磁振子性质，探索了纳米磁学中的几条路径，在这些阵列中，信息通过磁非磁振子耦合作为“突触”在纳米磁“神经元”之间传递。因此，理解纳米磁体网络中的动态模式耦合是有必要的。这个项目解决了对强相互作用磁网络的基本理解中的关键知识空白。这四个具体目标是：1)控制二维纳米磁铁阵列中的磁子，2)操纵磁振子-磁振子相互作用，3)了解磁性纳米结构中的非线性动力学，4)实验实现下一代神经形态磁子计算概念。采用电子束光刻、电子束蒸发、剥离等方法制备了纳米磁性网络，并用光学、电学和微波等方法对其进行了研究。实验研究得到了微磁模型的支持。这一奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

This project is jointly funded by the Condensed Matter Physics program of the Division of Materials Research and Established Program to Stimulate Competitive Research (EPSCoR).Nontechnical description:The surging development of artificial intelligence (AI) enables the creation of powerful tools and applications that were unimaginable just a few years ago. However, as AI and machine learning rapidly grow, the associated energy costs and greenhouse emissions are exploding. This massively unsustainable trend threatens to prevent society from achieving a net-zero future. Hence, a paradigm shift for low-power computing and AI processing is urgently needed. This project contributes to tackling this historic challenge by delivering foundational knowledge and technology concerning the fundamental excitations in magnetic nanostructures to create a transformative computing scheme taking inspiration from the brain. Current computing architectures rely on a constant shuttling of data between separate memory and processor, which is highly inefficient. Furthermore, current computing platforms are based on the flow of electronic charges, leading to dissipation in the form of Joule heating. To circumvent these problems, the research team aims to harness the dynamics in networks of interacting nanomagnets for bio-inspired processing by A) alleviating the processor-memory information transfer bottleneck and B) enabling the transport and processing of data based on waves rather than moving charges. The educational outreach component of this project fosters increased public participation in scientific research. The educational goals are designed to engage multiple levels of learning in wave physics: 1) a new course is developed for lifelong learners and 2) training programs are developed for schoolteachers working with a diverse student population by creating an accessible wave demonstration. Technical description:Spin waves, and their quanta - magnons - are the fundamental collective excitations of a magnetic system. Magnons can transport and process information without moving charges, and hence, magnonic devices can be less dissipative than their electronic counterparts. Nanomagnetic arrays are similar to neural networks, providing memory and computing abilities in the same unit: they can retain information stored in their magnetization orientation and process that information by magnonic excitations. This project explores several paths in nanomagnonics by determining the magnon properties in lithographically defined arrays of interacting nanomagnets, where information is passed between nanomagnetic ‘neurons’ via magnon-magnon coupling acting as ‘synapses’. Therefore, advances are needed to understand dynamic mode coupling in networks of nanomagnets. This project addresses critical knowledge gaps in the fundamental understanding of strongly interacting magnetic networks. The four specific aims are 1) controlling magnons in two-dimensional arrays of nanomagnets, 2) manipulating magnon-magnon interactions, and 3) understand nonlinear dynamics in magnetic nanostructures to 4) experimentally realize the next-generation of neuromorphic magnonic computing concepts. The nanomagnetic networks are fabricated by electron-beam lithography, electron-beam evaporation, and lift-off and studied by optical, electrical, and microwave methods. The experimental investigations are supported by micromagnetic modeling.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.