CAREER: Multi-Dimensional Photonic Accelerators for Scalable and Efficient Computing

职业：用于可扩展和高效计算的多维光子加速器

• 批准号: 2337674
• 负责人: Nathan Youngblood
• 金额: $ 55.22万
• 依托单位: University of Pittsburgh
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2024
• 资助国家: 美国
• 起止时间: 2024-04-01 至 2029-03-31
• 项目状态: 未结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2337674&HistoricalAwards=false
• 关键词: CAREER; Multi; Dimensional; Photonic; Accelerators; CAREER; Multi; Dimensional; Photonic; Accelerators

Despite advances in parallel computing platforms such as graphics processing units (GPUs), the growing demand for high computing power in emerging artificial intelligence (AI) applications far exceeds hardware efficiency improvements in current electronic systems. Optical computing promises improved efficiency and speed over conventional computing hardware, but is currently limited by power consumption, precision, and scalability issues. This project aims to revolutionize the field of AI by addressing these outstanding issues in current photonic computing architectures, thus unleashing the advantages of photonic computing for AI. In this CAREER proposal, the PI plans to advance optical computing through: (1) integrating photonic circuits and image sensors, (2) demonstrating multi-dimensional photonic computing, and (3) developing a simulation framework for large-scale photonic neural networks. Beyond technical advancements, the project’s educational goals include cultivating a diverse high-tech workforce in Pittsburgh through affordable educational tools, annual STEM workshops, and mentoring undergraduate researchers. Voluntary assessments in collaboration with Pitt's Engineering Education Research Center will measure educational outcomes, providing quantifiable metrics for long-term impact.This project aims to address three major limitations of current photonic computing platforms—power hungry electrical readout, limited analog precision, and poor scalability—to enable a fast and efficient computing architecture which could transform the field of artificial intelligence (AI). Despite notable advances in parallel computing, gains in hardware efficiency are unable to keep pace with the growing demand for extremely high computing power required by emerging AI applications and services. This is primarily due to the fundamental trade-off between clock speed and computational efficiency in the electronic domain stemming from the capacitance and Joule heating of metal interconnects. The PI will address this fundamental issue by performing computation in the optical domain using multiple photonic degrees of freedom for improved compute efficiency and speed. In this CAREER proposal, the PI will create new knowledge and extend the boundaries of optical computing through three unified tasks which: (1) integrate photonics and image sensors for robust and scalable matrix operations; (2) demonstrate multi-dimensional photonic computing for complex-valued matrix operations; and (3) develop a simulation framework to model the compute efficiency and latency of the proposed hardware for large-scale deep neural networks. Beyond technical advancements, the PI aims to cultivate a diverse high-tech workforce in the greater Pittsburgh area. Initiatives include creating affordable educational tools exposing students to nanotechnology applications in AI, conducting annual STEM workshops in collaboration with Pitt's outreach program (LEAD), and mentoring undergraduate researchers through Pitt's EXCEL summer research program. Voluntary assessments will measure educational outcomes, providing quantifiable metrics for the project's broader impact on workforce diversity and innovation in AI.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.

尽管在平行计算平台（例如图形处理单元（GPU））中取得了进步，但在新兴人工智能（AI）应用中对高计算能力的需求不断增长，远远超过了当前电子系统的硬件效率。光学计算承诺在常规计算硬件上提高效率和速度，但目前受到功耗，精度和可伸缩性问题的限制。该项目旨在通过解决当前光子计算体系结构中的这些杰出问题，从而释放AI的光子计算的优势来彻底改变AI领域。在此职业建议中，PI计划通过：（1）整合光子电路和图像传感器，（2）展示多维光子计算，（3）为大型光子神经元网络开发模拟框架。除了技术进步之外，该项目的教育目标包括通过负担得起的教育工具，年度STEM研讨会和对本科研究人员进行心理培养在匹兹堡培养潜水员的高科技劳动力。与皮特的工程教育研究中心合作的自愿评估将衡量教育成果，为长期影响提供可量化的指标。该项目旨在解决当前光子计算平台的三个主要局限性 - 能力饥饿的电气读数，有限的模拟精确性和不良的可扩展性，以实现快速有效的计算体系结构，可以实现人类的智能（人工智能）（一个人的人工智能）。尽管在并行计算方面取得了显着进步，但硬件效率的提高仍无法跟上对新兴AI应用程序和服务所需的极高计算能力需求的增长。这主要是由于金属互连的电容和焦耳加热所引起的电子域的时钟速度和计算效率之间的基本权衡。 PI将使用多个光子自由度在光学结构域中进​​行计算来解决这个基本问题，以提高计算效率和速度。在此职业建议中，PI将通过三个统一任务创建新知识，并扩展光学计算的边界：（1）集成的光子学和图像传感器，以实现可靠和可扩展的矩阵操作； （2）演示了用于复杂值矩阵操作的多维光子计算； （3）开发一个模拟框架，以建模大规模深神经网络所提出的硬件的计算效率和延迟。除技术进步外，PI还旨在在大匹兹堡地区培养潜水员高科技劳动力。倡议包括创建负担得起的教育工具，使学生在AI中暴露于纳米技术应用程序，与Pitt的外展计划（LEAD）合作进行年度STEM研讨会，并通过Pitt的Excel Summer Research计划对本科研究人员进行心理化。自愿评估将衡量教育成果，为该项目在AI中对劳动力多样性和创新的更广泛影响提供可量化的指标。该奖项反映了NSF的法定任务，并被认为是通过基金会的智力优点和更广泛的影响标准通过评估来评估的。