CAREER: Scalable monolithic integration of Graphene/MoS2/Graphene artificial neurons and synapses for accelerated machine learning

非技术：集成电路的发展是电子黄金时代的一个里程碑。该项目将通过在相同材料平台上开发人工神经元和突触来复制这一壮举。这将使我们能够创造出能够模仿人脑进行模式匹配的能力和速度的系统，击败目前体积更大、速度更慢、耗电量更大的系统。这将给机器学习领域带来革命性的变化，并创造出机器学习硬件接近传感器的设备，从而产生目前无法创建的系统。基于这些设备的神经形态电路包括智能可穿戴设备，可以监测健康生物识别，并为在家独居的老年人发出一级分诊。它们可以改善陆地车辆的自动驾驶。这些电路将对宇宙飞船和太空漫游者特别有用，因为这些超轻型电路不会成为火箭有效载荷的瓶颈。通过这些深度神经网络的快速模式识别能力将增强便携式电子设备的语音识别能力，并改进交通分析和控制系统。在与佛罗里达州中部STEM联盟(CFSA)计划的合作下，来自当地社区大学的代表不足的少数族裔将在PI的实验室获得实践研究机会。每年对人数不足的K-12学生进行定期视频实验室参观，并为两名精选的K-12学生提供为期一周的实验室体验，将培养他们对STEM教育的兴趣。奥兰多科学中心将展示突出这项研究的主要方面的横幅，以提高公众的认识。这些努力将引导电子行业通过佛罗里达高科技走廊在佛罗里达州投资，为该州的工程师创造机会。技术：该提案的目标是利用全二维(All-2D)石墨烯/MoS2/石墨烯记忆异质结构开发可扩展的单片集成人工神经元和突触，用于神经形态计算。在这些异质结构中，石墨烯作为电极，MoS_2作为开关介质。在垂直放置的MoS_2薄片中观察到的挥发性阻性开关将被用来实现积分与放电(IF)神经元，并将研究其放电的随机性质。利用水平MoS2薄片中的多能级非挥发开关，将开发出每个开关事件所需能量为亚皮焦耳的人工突触。我们将利用电学和材料表征技术研究垂直MoS_2薄膜中的挥发电阻开关与水平MoS_2薄膜中的非挥发电阻开关这一有趣现象背后的机制。石墨烯电极在异质结构中的必要性将得到证明。通过异质结构工程设计，人工神经元和突触的性能将被优化，以产生随机阻性开关IF神经元和低功率突触，并将其单片集成。利用石墨烯和二硫化钼的机械柔性优势，将在柔性平台上制造和测试这些神经形态器件。拟议的研究计划旨在将两个新兴研究领域凝聚在一起，形成一项尖端技术。虽然2D材料在冯·诺伊曼范式中接替硅有巨大的前景，但非冯·诺伊曼方法一直在处理传统材料。这项变革性的研究将把这两个世界的最好的东西结合在一起。大面积2D材料的使用将增强这些奇异装置的实际变现性。石墨烯/MoS2/石墨烯人工突触的超低功率操作与稀疏烧成的石墨烯/MoS2/石墨烯人工神经元将提供节能的神经形态计算。该奖项反映了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。

Nontechnical:Development of integrated circuits was a milestone in the golden age of electronics. That feat will be replicated by this project through the development of both artificial neurons and synapses on the same materials platform. This will allow us to create systems that can emulate the capacity and speed of the human brain for pattern matching, beating current systems that are bulkier, slower and more power hungry. This will revolutionize the field of machine learning and create devices where the machine-learning hardware is close to the sensor, giving rise to systems that are currently impossible to create. Neuromorphic circuits based on these devices include smart wearables that can monitor health biometrics and issue first level triage for elderly people alone at home. They can improve autonomous driving for terrestrial vehicles. These circuits will be especially useful for spacecrafts and space rovers because these ultra-light circuits will not be a bottleneck to the rocket's payload. Fast pattern recognition abilities through these deep neural networks will enhance speech recognition in portable electronics and improve traffic analysis and control systems. In collaboration with Central Florida STEM Alliance (CFSA) program, underrepresented minorities from local community colleges will be provided hands-on research opportunities in the lab of the PI. Regular video lab tours to underrepresented K-12 students and week-long lab experience for two selected K-12 students annually will foster their interest in STEM education. Banners highlighting key aspects of this research will be displayed at the Orlando Science Center for general public awareness. These efforts will lead the electronics industry to invest in Florida through the Florida High Technology Corridor, creating opportunities for engineers in the state.Technical:The objective of this proposal is to develop scalable monolithically integrated artificial neurons and synapses using all-two-dimensional (all-2D) graphene/MoS2/graphene memristive heterostructures for neuromorphic computing. In these heterostructures graphene acts as the electrodes and MoS2 as the switching medium. Volatile resistive switching observed in vertically-standing sheets of MoS2 will be harnessed to realize integrate-and-fire (IF) neurons, and the stochastic nature of their firing will be investigated. Artificial synapses with sub-picojoules of energy requirement per switching event will be developed using the multi-level non-volatile switching in horizontal MoS2 sheets. The mechanisms behind this intriguing phenomenon of volatile resistive switching in vertical MoS2 sheets versus non-volatile switching in horizontal MoS2 sheets will be investigated using electrical and materials characterization techniques. The necessity of graphene electrodes in the heterostructure will be justified. Through engineering of the heterostructure design, the performance of the artificial neurons and synapses will be optimized to create stochastic resistive-switching IF neurons and low-power synapses which will be integrated monolithically. The advantage of mechanical flexibility of graphene and MoS2 will be exploited to fabricate and test these neuromorphic devices on a flexible platform. The proposed research program aims at bringing together two emerging research areas cohesively into a cutting edge technology. While 2D materials have immense prospects in succeeding silicon within the von Neumann paradigm, non-von Neumann approaches have always dealt with conventional materials. This transformative research would bring together the best of both these worlds. The use of large-area 2D materials will enhance the practical realization of these exotic devices. Ultra-low power operation of graphene/MoS2/graphene artificial synapses with sparse firing graphene/MoS2/graphene artificial neurons will provide for energy-efficient neuromorphic computing.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.