Abstract—This paper reports the performances of a silicon pho-tonics optical switch matrix fabricated by using large-scale three-dimensional (3-D) integration. In AI training clusters, thousands or even tens of thousands of GPUs perform All-Reduce operations, generating massive “east-west” traffic. This traffic exhibits high burstiness, extremely high bandwidth demands, and extreme sensitivity to latency. The network is no longer merely a pipeline. Silicon photonics has developed into a mainstream technology driven by advances in optical communications. More precisely, silicon photonics. Broadband nonvolatile electrically programmable silicon photonic switches Broadband nonvolatile electrically programmable silicon photonic switches Rui Chen,11Zhuoran Fang, Johannes E. Fröch, Peipeng Xu,2Jiajiu Zheng,1* Arka Majumdar1,3* 1Department of Electrical and Computer Engineering.
[PDF Version]