用于部署边缘计算神经形态硬件的磁隧道结设计

Davi Rodrigues, Eleonora Raimondo, Riccardo Tomasello, Mario Carpentieri, Giovanni Finocchio
{"title":"用于部署边缘计算神经形态硬件的磁隧道结设计","authors":"Davi Rodrigues, Eleonora Raimondo, Riccardo Tomasello, Mario Carpentieri, Giovanni Finocchio","doi":"arxiv-2409.02528","DOIUrl":null,"url":null,"abstract":"The electrically readable complex dynamics of robust and scalable magnetic\ntunnel junctions (MTJs) offer promising opportunities for advancing\nneuromorphic computing. In this work, we present an MTJ design with a free\nlayer and two polarizers capable of computing the sigmoidal activation function\nand its gradient at the device level. This design enables both feedforward and\nbackpropagation computations within a single device, extending neuromorphic\ncomputing frameworks previously explored in the literature by introducing the\nability to perform backpropagation directly in hardware. Our algorithm\nimplementation reveals two key findings: (i) the small discrepancies between\nthe MTJ-generated curves and the exact software-generated curves have a\nnegligible impact on the performance of the backpropagation algorithm, (ii) the\ndevice implementation is highly robust to inter-device variation and noise, and\n(iii) the proposed method effectively supports transfer learning and knowledge\ndistillation. To demonstrate this, we evaluated the performance of an edge\ncomputing network using weights from a software-trained model implemented with\nour MTJ design. The results show a minimal loss of accuracy of only 0.1% for\nthe Fashion MNIST dataset and 2% for the CIFAR-100 dataset compared to the\noriginal software implementation. These results highlight the potential of our\nMTJ design for compact, hardware-based neural networks in edge computing\napplications, particularly for transfer learning.","PeriodicalId":501083,"journal":{"name":"arXiv - PHYS - Applied Physics","volume":null,"pages":null},"PeriodicalIF":0.0000,"publicationDate":"2024-09-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"A design of magnetic tunnel junctions for the deployment of neuromorphic hardware for edge computing\",\"authors\":\"Davi Rodrigues, Eleonora Raimondo, Riccardo Tomasello, Mario Carpentieri, Giovanni Finocchio\",\"doi\":\"arxiv-2409.02528\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"The electrically readable complex dynamics of robust and scalable magnetic\\ntunnel junctions (MTJs) offer promising opportunities for advancing\\nneuromorphic computing. In this work, we present an MTJ design with a free\\nlayer and two polarizers capable of computing the sigmoidal activation function\\nand its gradient at the device level. This design enables both feedforward and\\nbackpropagation computations within a single device, extending neuromorphic\\ncomputing frameworks previously explored in the literature by introducing the\\nability to perform backpropagation directly in hardware. Our algorithm\\nimplementation reveals two key findings: (i) the small discrepancies between\\nthe MTJ-generated curves and the exact software-generated curves have a\\nnegligible impact on the performance of the backpropagation algorithm, (ii) the\\ndevice implementation is highly robust to inter-device variation and noise, and\\n(iii) the proposed method effectively supports transfer learning and knowledge\\ndistillation. To demonstrate this, we evaluated the performance of an edge\\ncomputing network using weights from a software-trained model implemented with\\nour MTJ design. The results show a minimal loss of accuracy of only 0.1% for\\nthe Fashion MNIST dataset and 2% for the CIFAR-100 dataset compared to the\\noriginal software implementation. These results highlight the potential of our\\nMTJ design for compact, hardware-based neural networks in edge computing\\napplications, particularly for transfer learning.\",\"PeriodicalId\":501083,\"journal\":{\"name\":\"arXiv - PHYS - Applied Physics\",\"volume\":null,\"pages\":null},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2024-09-04\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"arXiv - PHYS - Applied Physics\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/arxiv-2409.02528\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"arXiv - PHYS - Applied Physics","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/arxiv-2409.02528","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0

摘要

稳健、可扩展的磁隧道结(MTJ)的电可读性复杂动力学为推进超形态计算提供了大好机会。在这项工作中,我们提出了一种 MTJ 设计,它具有一个自由层和两个极化器,能够在器件级计算西格码激活函数及其梯度。这种设计能够在单个器件内同时进行前馈和反向传播计算,通过引入直接在硬件中执行反向传播的能力,扩展了之前在文献中探索的神经形态计算框架。我们的算法实现揭示了两个关键发现:(i) MTJ 生成的曲线与软件生成的精确曲线之间的微小差异对反向传播算法的性能影响微乎其微;(ii) 设备实现对设备间的变化和噪声具有高度鲁棒性;(iii) 提议的方法有效支持迁移学习和知识积累。为了证明这一点,我们评估了边缘计算网络的性能,使用的权重来自软件训练的模型,该模型由我们的 MTJ 设计实现。结果表明,与最初的软件实现相比,在时尚 MNIST 数据集和 CIFAR-100 数据集上的准确率损失分别仅为 0.1% 和 2%。这些结果凸显了我们的 MTJ 设计在边缘计算应用中基于硬件的紧凑型神经网络的潜力,特别是在迁移学习方面。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
A design of magnetic tunnel junctions for the deployment of neuromorphic hardware for edge computing
The electrically readable complex dynamics of robust and scalable magnetic tunnel junctions (MTJs) offer promising opportunities for advancing neuromorphic computing. In this work, we present an MTJ design with a free layer and two polarizers capable of computing the sigmoidal activation function and its gradient at the device level. This design enables both feedforward and backpropagation computations within a single device, extending neuromorphic computing frameworks previously explored in the literature by introducing the ability to perform backpropagation directly in hardware. Our algorithm implementation reveals two key findings: (i) the small discrepancies between the MTJ-generated curves and the exact software-generated curves have a negligible impact on the performance of the backpropagation algorithm, (ii) the device implementation is highly robust to inter-device variation and noise, and (iii) the proposed method effectively supports transfer learning and knowledge distillation. To demonstrate this, we evaluated the performance of an edge computing network using weights from a software-trained model implemented with our MTJ design. The results show a minimal loss of accuracy of only 0.1% for the Fashion MNIST dataset and 2% for the CIFAR-100 dataset compared to the original software implementation. These results highlight the potential of our MTJ design for compact, hardware-based neural networks in edge computing applications, particularly for transfer learning.
求助全文
通过发布文献求助,成功后即可免费获取论文全文。 去求助
来源期刊
自引率
0.00%
发文量
0
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
确定
请完成安全验证×
copy
已复制链接
快去分享给好友吧!
我知道了
右上角分享
点击右上角分享
0
联系我们:info@booksci.cn Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。 Copyright © 2023 布克学术 All rights reserved.
京ICP备2023020795号-1
ghs 京公网安备 11010802042870号
Book学术文献互助
Book学术文献互助群
群 号:481959085
Book学术官方微信