NeTS: Small: Can Disaggregation and Silicon Optical Interconnect Technology Co-exist?

每秒钟大约有一千亿个数据包（相当于几千部《大英百科全书》）从一栋典型的办公楼传输出去。现代通信网络的瓶颈或故障会立即影响航空旅行、医疗保健、金融市场、教育和娱乐以及国家安全——简而言之，影响现代生活的方方面面。为了在不相应地增加能源消耗的情况下增加未来十年的带宽，工程师们现在使用光子（光）而不是电子（电压）在计算机之间传输数据。在不久的将来，光子学还可以用于计算机内部微芯片之间的通信，或者微处理器和存储器之间的通信。该项目解决了一个新兴的研究挑战，即确保大规模光连接网络（如仓库大小的数据中心）的可扩展性、控制和有效管理。分解用Infiniband和PCIe等通信网络取代了主板总线架构，从而实现了可扩展性和可控性，但要求网络处理在带宽和延迟要求上跨越几个数量级的新型数据流。这些要求与目前设计的硅光子芯片类型不兼容。该项目将使用硅光子学更有效地实现数据中心内的分解；利用光通信与传统电子线路相比的关键优势：不同颜色的光能够相互穿过而不受干扰，从而在相同的物理网络拓扑结构中实现网络中的不同逻辑互连。为了测试和发展这一概念，硅光子微芯片将与拥有硅光子代工资源的工业和/或政府组织合作设计和制造。这些创新通信芯片组的性能将在光网络测试平台上进行研究。

About one hundred billion data packets (equivalent to a few thousand Encyclopedia Brittanicas) are transmitted every second from a typical office building. Bottlenecks or breakdowns in modern communication networks instantly impact air travel, healthcare, financial markets, education and entertainment, and national security - in short, every aspect of modern life. To increase bandwidth for the coming decade without commensurately increasing energy consumption, engineers are now using photons (light) rather than electrons (voltage) to carry data between computers. In the near future, photonics may also be used to communicate between microchips inside computers, or between microprocessors and memory. This project addresses a emerging research challenge in ensuring the scalability, control and effective management of massive optically-connected networks such as warehouse-sized data centers. Disaggregation replaces motherboard bus architecures with communication networks such as Infiniband and PCIe, thus enabling scalability and controllability, but requires the network to deal with new types of data flows that span several orders of magnitude in bandwidth and latency requirements. These requirements are not compatible with the type of silicon photonic chips that have been designed so far. This project will use silicon photonics to more effectively enable disaggregation within data centers; leveraging key strengths of optical communications compared to traditional electronic wires: the ability of light of different colors to pass through each other without interference, thus enabling different logical interconnections in a network across the same physical network topology. To test and develop this concept, silicon photonic microchips will be designed and fabricated in collaboration with industry and/or government organizations that host silicon photonic foundry resources. The performance of these innovative communication chipsets will be studied in a optical networking testbed.