FET: Small: 2D Material Compound Synapse Arrays for Robust In-Memory and Neuromorphic Computing (2DNeuro)

模拟大脑的新型计算系统受到了科学和工程界的极大关注，因为它们可以在只消耗少量能量的情况下帮助分析大量信息。这些受神经启发的系统已经取得了重大进展，但在便携式设备上的实际应用仍然存在局限性。这项工作将研究可以减轻这些限制的新材料和设计，从而更好地执行关键的数学函数。这项工作带来的科学突破将为数据密集型机器学习应用提供强大、高效和大规模并行计算的支持。通过为支持未来认知系统和人工智能的新型计算硬件奠定基础，将产生深远的社会影响。该项目的另一个目标是通过培训学生（包括研究生和本科生研究人员），开发基于网络的交互式平台形式的教育工具，以及支持关于新型集成记忆和神经启发计算架构主题的研究生水平课程，将研究和教育结合起来。该项目将为基于二维（2-D）原子薄材料的新存储技术奠定基础，扩展神经启发系统的实际工程极限。这些存储设备被称为忆阻器，可以根据外部施加的刺激被编程为不同的状态。当以特定的配置组织时，忆阻器阵列可用于实现系统在决策过程中使用的复杂功能。然而，由于忆阻器工作的随机性，变化是不可避免的，影响了计算的准确性和基于系统的决策的可信度。这项工作旨在将新型二维材料与新的阵列配置相结合，以减轻可变性的影响，从而提高计算的准确性和性能，同时保持低能耗的好处。该项目的智力意义从与二维材料的原子级存储相关的基础科学延伸到与神经启发计算应用的鲁棒记忆电阻阵列演示相关的应用工程贡献。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Novel computing systems that emulate the brain have received significant attention from the scientific and engineering community, as they can help analyze large amounts of information while consuming only small amounts of energy. Significant advances have been made on these neuro-inspired systems, but there are still limitations that prevent their practical application on portable devices. This work will investigate novel materials and designs that can alleviate these limitations, enabling a better execution of critical mathematical functions. The scientific breakthroughs resulting from this work will support robust, efficient, and massively parallel computing for data-intensive machine learning applications. A profound societal impact will be achieved by laying the foundations for novel computing hardware that will support cognitive systems and artificial intelligence of tomorrow. Another objective of this project is to integrate research and education by training students (including graduate and undergraduate researchers), by developing educational tools in form of web-based interactive platforms, and by supporting a graduate-level course in the topic of novel integrated memory and neuro-inspired computing architectures.This project will establish the foundations of new storage technology based on two-dimensional (2-D) atomically-thin materials, extending the practical engineering limits of neuro-inspired systems. These storage devices, known as memristors, can be programmed to different states, based on an externally applied stimulus. When organized in specific configurations, memristor arrays can be used to implement complex functions that are used by the system in a decision-making process. However, because of randomness in the operation of memristors, variation is inevitable, affecting the accuracy of the computation and the trustworthiness of the system-based decisions. This work aims to combine novel 2-D materials with new array configurations to mitigate the impact of variability in order to improve accuracy and performance of computation, while maintaining the benefits of low-energy consumption. The intellectual significance of this project extends from fundamental science relating to atomic-scale storage in 2-D materials, to applied engineering contributions relating to the demonstration of robust memristor arrays for neuro-inspired computing applications.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.