CAREER: Optimizing Scalability and Reconfigurability in Silicon Photonic Switch Fabrics

职业：优化硅光子交换结构的可扩展性和可重构性

• 批准号: 2046226
• 负责人: Mahdi Nikdast
• 金额: $ 57.51万
• 依托单位: Colorado State University
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2021
• 资助国家: 美国
• 起止时间: 2021-05-01 至 2026-04-30
• 项目状态: 未结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2046226&HistoricalAwards=false
• 关键词: CAREER; Optimizing; Scalability; Reconfigurability; Silicon

Our daily lives depend heavily on efficiently moving and processing data generated by different applications, from social networks and online shopping to healthcare and educational applications to emerging applications such as autonomous driving. The rapid growth of such data is putting increased pressure on datacenter networks, where enormous amounts of data must be processed. To aid in this task, the underlying datacenter network switches, which are responsible for moving data from one server to another, should provide the fastest possible switching with high bandwidths, while minimizing the energy required. To address these requirements, this project involves transformative research to design and optimize a new class of network switches based on silicon photonics technology that will be capable of transferring large amounts of data with ultra-low latency. The project includes research on this new technology and its system application in large-scale datacenter networks. It will incorporate opportunities for students (including underrepresented groups) at K-12, undergraduate, and graduate levels to work on real-world problems. It also includes a new outreach program to K-12 schools in Northern Colorado and Southern Wyoming with low college-attendance rates.This project proposes a long-term, integrated program of research to optimize the scalability and reconfigurability in silicon photonic switches for future high-performance computing and datacenter systems. In particular, the researchers propose comprehensive optimizations across the entire design space of silicon photonic switches, including device, network, and control layers, to realize ultra-fast reconfigurable, large-scale, and energy-efficient silicon photonic switches. At the device layer, a new class of interferometric silicon photonic switching devices will be designed and optimized to realize fast switching elements with ultra-low optical losses and crosstalk noise. At the network layer, topology-aware optical power loss and crosstalk noise models will be developed to be integrated into a new network-optimization platform, creating and optimizing various types of high-radix networks specifically for large-scale silicon photonic switches. The new research at the control layer will analyze traffic flows in the switch fabric to develop an ultra-fast and energy-aware switch reconfiguration solution. Last, a new open-source silicon photonic switch cross-layer design and co-optimization tool will be developed to incorporate the new models and optimization solutions at each design layer.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.

我们的日常生活在很大程度上依赖于有效地移动和处理不同应用程序生成的数据，从社交网络和在线购物到医疗保健和教育应用，再到自动驾驶等新兴应用。这些数据的快速增长给数据中心网络带来了越来越大的压力，在数据中心网络中必须处理大量的数据。 为了帮助完成这项任务，负责将数据从一台服务器移动到另一台服务器的底层数据中心网络交换机应该提供最快的高带宽交换，同时最大限度地减少所需的能源。为了满足这些要求，该项目涉及变革性研究，以设计和优化基于硅光子技术的新型网络交换机，该交换机将能够以超低延迟传输大量数据。 该项目包括对这种新技术及其在大规模数据中心网络中的系统应用的研究。 它将为K-12，本科和研究生级别的学生（包括代表性不足的群体）提供解决现实问题的机会。它还包括一个面向北方科罗拉多和南方怀俄明州的K-12学校的新的推广计划，这些学校的大学入学率较低。该项目提出了一个长期的综合研究计划，以优化硅光子交换机的可扩展性和可重构性，用于未来的高性能计算和数据中心系统。特别是，研究人员提出了对硅光子交换机整个设计空间的全面优化，包括设备、网络和控制层，以实现超快速可重构、大规模和节能的硅光子交换机。在器件层，将设计和优化一类新的干涉型硅光子开关器件，以实现具有超低光损耗和串扰噪声的快速开关元件。在网络层，将开发拓扑感知的光功率损耗和串扰噪声模型，以集成到新的网络优化平台中，创建和优化专门用于大规模硅光子交换机的各种类型的高基数网络。控制层的新研究将分析交换结构中的流量，以开发超快速和节能的交换机重新配置解决方案。最后，将开发一个新的开源硅光子交换机跨层设计和协同优化工具，以在每个设计层整合新的模型和优化解决方案。该奖项反映了NSF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。

项目成果

Pruning Coherent Integrated Photonic Neural Networks Using the Lottery Ticket Hypothesis

使用彩票假设修剪相干集成光子神经网络

• DOI: 10.1109/isvlsi54635.2022.00035
• 发表时间: 2022
• 期刊: Proc. IEEE Computer Society Annual Symposium on VLSI (ISVLSI
• 作者: Banerjee, Sanmitra;Nikdast, Mahdi;Pasricha, Sudeep;Chakrabarty, Krishnendu
• 通讯作者: Chakrabarty, Krishnendu

Machine Learning Accelerators in 2.5D Chiplet Platforms with Silicon Photonics

• DOI: 10.23919/date56975.2023.10137317
• 发表时间: 2023-01
• 期刊: 2023 Design, Automation & Test in Europe Conference & Exhibition (DATE)
• 作者: Febin P. Sunny;Ebadollah Taheri;M. Nikdast;S. Pasricha

On the Impact of Uncertainties in Silicon-Photonic Neural Networks

硅光子神经网络中不确定性的影响

• 发表时间: 2022
• 期刊: IEEE design test of computers
• 作者: Banerjee, Sanmitra;Nikdast, Mahdi;Chakrabarty, Krishnendu
• 通讯作者: Chakrabarty, Krishnendu

Characterizing Coherent Integrated Photonic Neural Networks Under Imperfections

• DOI: 10.1109/jlt.2022.3193658
• 发表时间: 2022-07
• 期刊: Journal of Lightwave Technology
• 影响因子: 4.7
• 作者: Sanmitra Banerjee;M. Nikdast;K. Chakrabarty

ReSiPI: A Reconfigurable Silicon-Photonic 2.5D Chiplet Network with PCMs for Energy-Efficient Interposer Communication

• DOI: 10.1145/3508352.3549432
• 发表时间: 2022-08
• 期刊: 2022 IEEE/ACM International Conference On Computer Aided Design (ICCAD)
• 作者: Ebadollah Taheri;S. Pasricha;M. Nikdast