SHF: Small: Ultra Low Power Neuromorphic Computing with Spin-devices

SHF：小型：使用自旋设备的超低功耗神经形态计算

• 批准号: 1320808
• 负责人: Kaushik Roy
• 金额: $ 50万
• 依托单位: Purdue University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2013
• 资助国家: 美国
• 起止时间: 2013-08-01 至 2018-07-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1320808&HistoricalAwards=false
• 关键词: SHF; Small; Ultra; Low; Power

MOS (metal-oxide-semiconductor) transistors, being on/off switches, have served as an ideal match to the abstractions of switching functions and Boolean logic, which (together with von Neumann architecture) form the underpinnings of modern computing. While current computing platforms are well-suited to applications that involve arithmetic computations and storing and retrieving large amounts of data, they are known to be highly inefficient - requiring orders of magnitude more energy consumption - for performing tasks that humans routinely perform, such as visual recognition, semantic analysis, and reasoning. The key insight behind the proposed research is that emerging spin torque based devices make it possible to realize the neuron functionality in a highly energy efficient manner, well beyond the capabilities of CMOS (complementary MOS). This is due to two factors - the inherent match between the characteristics of these devices and the functionality of a neuron (leading to a drastic decrease in the number of devices required), and the possibility of ultra-low voltage operation (~20mV). Inspired by this vision, the PIs (principal investigators) propose a research program that spans from devices to architectures to investigate spin-based neuromorphic computing. The goals of the proposed research are to (i) establish the benefits of spin-based device technologies for neuromorphic computing (where CMOS implementations are energy-inefficient), (ii) synergistically explore spin devices, circuits and architecture in a regime where these devices are integrated with CMOS to augment its capabilities and (iii) incorporate the findings from the proposed research in education and outreach programs. The project will leverage outreach programs at Purdue, including the outreach program at NCN (Network for Computational Nanotechnology), and the Women and Minority in Engineering programs in the College of Engineering, to involve undergraduates and minority students in the proposed research. The PIs will develop a summer REU (Research Experiences for Undergraduates) program that builds on their prior experience in integrating undergraduate students into their research groups. The broad nature of the proposed project provides an attractive opportunity for undergraduate students to not only explore novel research, but also get exposed to the emerging fields of spintronics and neuromorphic computing. The project will also establish research, mentoring and education partnerships with primarily undergraduate and minority serving institutions.

作为开/关开关的MOS(金属氧化物半导体)晶体管是开关功能和布尔逻辑抽象的理想匹配，它们(与冯·诺伊曼体系结构一起)构成了现代计算的基础。虽然目前的计算平台非常适合涉及算术计算以及存储和检索大量数据的应用程序，但众所周知，它们在执行人类经常执行的任务(如视觉识别、语义分析和推理)方面效率极低-需要多消耗数量级的能量。这项拟议研究背后的关键见解是，新兴的基于自旋力矩的设备使以高能效方式实现神经元功能成为可能，远远超过了互补MOS(互补型MOS)的能力。这归因于两个因素--这些器件的特性与神经元功能之间的内在匹配(导致所需器件数量的急剧减少)，以及超低电压操作(~20 mV)的可能性。在这一愿景的启发下，PI(首席调查人员)提出了一个从设备到架构的研究计划，以研究基于自旋的神经形态计算。这项拟议研究的目标是：(I)确定基于自旋的器件技术用于神经形态计算的益处(其中，CMOS实现是低能效的)；(Ii)协同探索自旋器件、电路和体系结构，其中这些器件与CMOS集成以增强其能力；以及(Iii)将拟议研究的结果纳入教育和推广计划中。该项目将利用普渡大学的外联计划，包括NCN(计算纳米技术网络)的外联计划，以及工程学院的妇女和少数族裔工程计划，让本科生和少数族裔学生参与拟议的研究。PIS将开发一个暑期REU(本科生研究经验)计划，以他们之前将本科生纳入他们的研究小组的经验为基础。拟议项目的广泛性质为本科生提供了一个有吸引力的机会，不仅可以探索新的研究，还可以接触到自旋电子学和神经形态计算等新兴领域。该项目还将与主要为本科生和少数族裔服务的机构建立研究、指导和教育伙伴关系。