Exploiting Metal-Insulator-Transition in Strongly Correlated Oxides as Neuron Device for Neuro-Inspired Computing

利用强相关氧化物中的金属-绝缘体转变作为神经元设备进行神经启发计算

• 批准号: 1903577
• 负责人: Shimeng Yu
• 金额: $ 30.22万
• 依托单位: Georgia Tech Research Corporation
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2018
• 资助国家: 美国
• 起止时间: 2018-07-01 至 2021-07-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1903577&HistoricalAwards=false
• 关键词: Exploiting; Metal; Insulator; Transition; Strongly

A radical shift in computing paradigm towards the neuro-inspired computing is attractive for performing data-intensive applications such as image/speech recognitions. The neuro-inspired architecture leverages the distributed computation in the neuron nodes and the localized storage in the synaptic elements. The neuron node today is generally implemented by tens of silicon transistors. Compared to the crossbar array of synaptic elements, the silicon neuron is power-hungry and area-inefficient, thereby reducing the parallelism of computing system. In such context, how to design a single device that can efficiently emulate the neuronal behavior (e.g. integrate-and-fire) is critical to the neuromorphic hardware design. This project aims to exploit the metal-insulator-transition phenomenon in strongly correlated oxides as a compact neuron node that can self-oscillate, namely oxide neuron, to overcome the aforementioned limitations of silicon neuron. The proposed research will have a profound impact on the society that is embracing the artificial intelligence. For instance, a compact design of neuromorphic hardware may enable intelligent information processing on power-efficient mobile platforms, e.g. autonomous vehicle, personalized healthcare, wearable devices, and smart sensors. The objective of the research and education integration is to train undergraduate/graduate students and next-generation workforce with interdisciplinary skills. The cross-layer nature of this project ranging from materials engineering, semiconductor device, circuit-device interaction and artificial neural network provides an ideal platform for this educational goal. The project also plans to engage minority and unrepresentative students in research. Technology transfer will be performed through video or on-site seminars and student internships with industrial collaborators.The goal of this research is to advance the artificial neuron device design by exploiting the volatile and threshold switching behavior in strongly correlated oxides, with the purpose of significantly reducing the area and energy of the neuron node, and making it compatible for the integration with crossbar array of resistive synaptic elements. The scope of the project is to explore various material systems of the strongly correlated oxides, in particular, NbO2 and SmNiO3 to demonstrate the self-oscillation behavior in the artificial neuron node. When such oxide device is connected with a series synaptic element whose resistance is within the on/off dynamic range of the oxide device, the node voltage between the oxide device and the synaptic element will start self-oscillation, and the oscillation frequency represents the synaptic conductance. This project aims to explore such self-oscillation to emulate the integrate-and-fire neuronal behavior. To achieve the aforementioned research goal, device fabrication, physical and electrical characterization, device modeling, and circuit-device co-design will be performed to demonstrate the feasibility of the concept and further optimize the device performance. The intellectual significance of this project is two folded. From the fundamental science perspective, the physical switching mechanism of metal-insulator-transition in strongly correlated oxides will be investigated. From the applied engineering perspective, the oxide neuron device will be integrated with the resistive crossbar array for demonstration of a neural network for solving a practical problem, i.e. the image pattern classification.

计算范式向神经启发计算的根本转变对于执行数据密集型应用具有吸引力，例如图像/语音识别。这种受神经启发的体系结构利用了神经元节点中的分布式计算和突触元素中的局部存储。如今的神经元节点通常由数十个硅晶体管实现。与交叉式突触单元阵列相比，硅神经元具有耗电和面积低的特点，从而降低了计算系统的并行性。在这样的背景下，如何设计一个能够有效地模拟神经元行为的单一设备(例如集成并激发)是神经形态硬件设计的关键。该项目旨在利用强关联氧化物中的金属-绝缘体-跃迁现象作为一种能够自振荡的致密神经元节点，即氧化物神经元，以克服硅神经元的上述局限性。这项拟议的研究将对拥抱人工智能的社会产生深远影响。例如，神经形态硬件的紧凑型设计可以在高能效移动平台上实现智能信息处理，例如自动驾驶汽车、个性化医疗保健、可穿戴设备和智能传感器。研究和教育一体化的目标是培养具有跨学科技能的本科生/研究生和下一代劳动力。该项目的跨层性质涵盖了材料工程、半导体器件、电路-器件交互和人工神经网络，为实现这一教育目标提供了理想的平台。该项目还计划让少数族裔和没有代表性的学生参与研究。技术转让将通过视频或现场研讨会和与行业合作者的学生实习进行。本研究的目标是通过利用强关联氧化物的挥发性和阈值开关行为来推进人工神经元设备的设计，目的是显著减少神经元节点的面积和能量，并使其与阻性突触元件的交叉杆阵列相兼容。该项目的范围是探索强关联氧化物的各种材料体系，特别是NbO2和SmNiO_3，以展示人工神经元节点中的自振荡行为。当这种氧化物器件与串联突触元件连接时，其电阻在氧化物器件的通断动态范围内，氧化物器件和突触元件之间的节点电压将启动自振荡，振荡频率代表突触电导。该项目旨在探索这种自我振荡，以模拟整合和放电神经元的行为。为了实现上述研究目标，将进行器件制造、物理和电气特性、器件建模和电路-器件协同设计，以论证该概念的可行性，并进一步优化器件性能。这个项目的智力意义是双重的。从基础科学的角度，我们将研究强关联氧化物中金属-绝缘体-转变的物理开关机制。从工程应用的角度出发，将氧化物神经元器件与阻性交叉杆阵列相结合，展示了解决实际问题的神经网络，即图像模式分类。

项目成果

Integrated Crossbar Array With Resistive Synapses and Oscillation Neurons

• DOI: 10.1109/led.2019.2921656
• 发表时间: 2019-08-01
• 期刊: IEEE ELECTRON DEVICE LETTERS
• 影响因子: 4.9
• 作者: Woo, Jiyong;Wang, Panni;Yu, Shimeng
• 通讯作者: Yu, Shimeng