Guangcan Mi – Huawei

Bio: Guangcan Mi received her PhD in Silicon Photonic and Plasmonics in 2016. She has since focused her career on optical interconnect technologies and standards. Her expertise spans from optical transceiver to optical switching, for applications of data center networks and telecommunications/carrier systems. Currently her work is concentrated on the development of 200G/lane optical interconnect technologies and standards, with a particular focus on IEEE 802.3df and 802.3dj.

Presentation Abstract: The exponential growth of Large Language Model has accelerated the arms race of AI and drives forward the evolution of connectivity technologies. The options are clear, go wider, go faster or boost the symbol efficiency. The choice however is difficult, especially for scale-out and front-end network, where both electrical and optical connectivity are used, and where interoperability holds a stake. As the AI cluster further expands its scale, the connectivity solutions needs to at least maintain the same reaches as today for the next hop, i.e. going beyond 200G/lane. Such request impinges both physical challenges and  system-level challenges, blurring the application boundary between electrical and optical technologies. This presentation aims to roadmap the interconnect path for scale-out and front-end network given the system-level constraints and explore how co-designing optical and electrical interfaces at beyond 200G/lane could provide the next generation of connectivity for AI