SBIR Phase I: Oscillator Processing Unit - Physical Reservoir Computing on the Edge

这个小型企业创新研究（SBIR）第一阶段项目的更广泛影响将来自于创建振荡器处理单元（OPU），这是一种计算处理器，将是在网络边缘运行的模拟计算设备的颠覆性技术。到2032年，边缘计算设备市场预计将达到1570亿美元。在此基础上，机器学习，特别是深度神经网络（DNN），依赖于云基础设施来进行模型训练和推理的大规模计算，因此OPU将具有有益的安全和环境影响，因为它们将减少对云的依赖。所提出的OPU可以克服冯·诺依曼瓶颈，同时还可以实现更小的外形尺寸，提高能效和更快的速度。随着美国寻求减少对外国微芯片制造商的依赖，OPU也可以提供一个强大的，可行的替代品，可以在美国制造。这个项目的技术影响将来自于对振荡器的更基本的理解，振荡器是宇宙中最多产的动力系统之一，也可以被重新考虑为物理计算机。这个小企业创新研究（SBIR）第一阶段项目旨在利用两种类型的基于神经元的计算机。通过探索振荡器计算机的动力学，将发展出对非线性动力学如何转化为计算能力的更好理解。此外，预计这将为如何为振荡器处理单元（OPU）构建最佳振荡器核心提供见解。这些增强的OPU将融合两种独立的模拟计算方法：物理水库计算机和自适应振荡器。最终，由于振荡器核心的存储器和处理不是独立的，OPU可以提供冯诺依曼瓶颈的解决方案。这项工作将为物理学和信息之间的联系建立一个基本的科学理解。利用这两种不同形式的神经形态智能也将是一个强大的振荡器处理单元的基础，能够作为人工智能推理处理器和通用计算处理器。这个奖项反映了NSF的法定使命，并被认为值得通过使用基金会的智力价值和更广泛的影响审查标准进行评估来支持。

The broader impact of this Small Business Innovation Research (SBIR) Phase I project will result from creating the Oscillator Processing Unit (OPU), a computational processor that would be a disruptive technology for analog computing devices operating at the network edge. The edge computing device market is projected to reach $157B by 2032. Expanding on this, machine learning, especially deep neural networks (DNNs), relies on cloud infrastructure to conduct massive computation for both model training and inference, so OPUs would have beneficial security and environmental impacts since they would reduce reliance on the cloud. The proposed OPUs could overcome the von Neumann bottleneck while also enabling a smaller form factor, increased energy efficiency, and faster speeds. As the US seeks to reduce reliance on foreign microchip manufacturers, OPUs could also provide a powerful, viable alternative that could be manufactured in the US. The technological impacts of this project would result from a more fundamental understanding of how oscillators, which are one of the most prolific dynamic systems in the universe, can also be reconsidered as physical computers. This Small Business Innovation Research (SBIR) Phase I project seeks to leverage two types of oscillator-based neuromorphic computers. By exploring the dynamics of oscillator computers, an improved understanding of how nonlinear dynamics are translated into computational ability will be developed. Further, this is expected to provide insights into how optimal oscillator cores could be constructed for Oscillator Processing Units (OPUs). These enhanced OPUs will converge two separate methods of analog computing: physical reservoir computers and adaptive oscillators. Ultimately, since an oscillator core’s memory and processing are not independent, OPUs could provide a solution to the von Neumann bottleneck. This work would establish a fundamental scientific understanding of the link between physics and information. Leveraging these two disparate forms of neuromorphic intelligence will also be the basis of a powerful Oscillator Processing Unit capable of acting as both an AI inference processor and a generalized computing processor.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.