ASCENT: Using Optical Frequency Comb for Ultrafast Nature-Based Computing for Machine Learning Algorithms

ASCENT：使用光学频率梳进行机器学习算法的超快基于自然的计算

• 批准号: 2231036
• 负责人: Michael Huang
• 金额: $ 149.99万
• 依托单位: University of Rochester
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2022
• 资助国家: 美国
• 起止时间: 2022-10-01 至 2026-09-30
• 项目状态: 未结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2231036&HistoricalAwards=false
• 关键词: ASCENT; Using; Optical; Frequency; Comb

Expanding the boundaries of current computing system performance calls for disruptive innovations to enable next-generation architectures beyond the traditional, so-called von Neumann paradigm. In this project, a novel photonic non-von Neumann system will be developed that pushes the envelope of nature-based computing in efficiency, capability, and applicability. Products and insights from this ASCENT collaboration have strong transformative potentials to bring nature-based computing to the state of compelling infrastructure and directly impact the gamut of application domains of machine learning (ML) in scientific discovery, industry, assistive technologies, robotics-aided healthcare, economic development, and consequent improvements in quality of life. Envisioned broader impacts will permeate to the integrated photonics community, with new functions being realized at the chip level for microcombs that can serve as key enablers in new sensing and communication platforms. Project outcomes will generate new knowledge and disruptive innovation for hybrid photonic and electronic interfaces; enable systems and architectures beyond the von Neumann paradigm, and thus impact Future Semiconductor Technology (FST) platforms -- a strategic national priority; and train next generation engineers for continued innovation in this area.While nature-based, non-von Neumann computing machines such as D-Wave’s quantum annealers are showing promise, these current machines are far from compelling due to their demanding (e.g., cryogenic) operating conditions, significant bulk, their relatively high energy consumption, and their limited applicability to combinatorial optimization problems. They will only be truly viable when they are significantly more capable and efficient than state-of-the-art von Neumann platforms in solving a non-trivial section of real-world problems. This project’s ambitious and broad vision is to bring such a machine to fruition, which can only be realized via convergent research in devices, circuits, algorithms, and ML. Nanophotonics is a promising direction to catalyze the required transformative advances, through an optical frequency microcomb that can be controlled to function as a large-scale computing system. A microcomb-based non-von Neumann system will be developed, which can accelerate a variety of ML algorithms. Building medium- to large-scale system prototypes calls for developments in physical hardware for learning systems, integration of silicon photonic circuits exploiting microcombs, and co-design of novel ML algorithms that leverage the unique features of this machine. This project will educate and train the next generation of researchers to think outside the box of their niche discipline, instill in them the excitement of crossing disciplinary boundaries, and give them first-hand appreciation of the need for convergent efforts towards making progress in engineering system applications with high societal impact.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.

扩展当前计算系统性能的边界需要颠覆性创新，以实现超越传统的所谓冯诺依曼范式的下一代架构。在这个项目中，将开发一种新型的光子非冯·诺依曼系统，推动基于自然的计算在效率，能力和适用性方面的发展。ASCENT合作的产品和见解具有强大的变革潜力，可以将基于自然的计算带入引人注目的基础设施状态，并直接影响机器学习（ML）在科学发现，工业，辅助技术，机器人辅助医疗保健，经济发展以及随之而来的生活质量改善等领域的应用领域。设想的更广泛的影响将渗透到集成光子学社区，在芯片级实现微梳的新功能，可以作为新的传感和通信平台的关键推动者。项目成果将为混合光子和电子接口产生新的知识和颠覆性创新;使系统和架构超越冯诺依曼范式，从而影响未来半导体技术（FST）平台-国家战略优先事项;并培养下一代工程师在这一领域的持续创新。虽然以自然为基础，诸如D-Wave的量子退火机的非冯·诺依曼计算机器显示出希望，但是这些当前的机器由于它们的要求（例如，低温）操作条件、显著的体积、它们相对高的能量消耗以及它们对组合优化问题的有限适用性。只有当它们在解决现实世界问题的重要部分时比最先进的冯诺依曼平台更有能力和效率时，它们才真正可行。该项目雄心勃勃的愿景是实现这样一台机器，这只能通过设备，电路，算法和ML的融合研究来实现。纳米光子学是一个很有前途的方向，通过一个可以控制的光频微梳作为一个大规模的计算系统来催化所需的变革性进展。将开发一种基于微梳的非冯·诺依曼系统，它可以加速各种ML算法。构建中到大规模的系统原型需要开发用于学习系统的物理硬件，集成利用微梳的硅光子电路，以及利用该机器独特功能的新型ML算法的共同设计。该项目将教育和培训下一代研究人员跳出其利基学科的框框进行思考，向他们灌输跨越学科界限的兴奋，先给他们--该奖项反映了NSF的法定使命，并被认为值得通过使用基金会的学术价值和更广泛的影响审查标准。

项目成果

Supporting Energy-Based Learning with an Ising Machine Substrate: A Case Study on RBM

• DOI: 10.1145/3613424.3614315
• 发表时间: 2023-04
• 期刊: 2023 56th IEEE/ACM International Symposium on Microarchitecture (MICRO)
• 作者: Uday Kumar Reddy Vengalam;Yongchao Liu;Tong Geng;Hui Wu;Michael Huang