AI大模型正引领下一个十年的信息与通信技术(information and communications technology,ICT)产业发展热点。智算中心网络是支撑AI大模型分布式训练的通信底座,是决定AI集群效能的关键要素之一。AI大模型的数据量和参数量不断扩张,给...AI大模型正引领下一个十年的信息与通信技术(information and communications technology,ICT)产业发展热点。智算中心网络是支撑AI大模型分布式训练的通信底座,是决定AI集群效能的关键要素之一。AI大模型的数据量和参数量不断扩张,给智算中心网络带来了严峻的挑战,同时给关键网络技术进行代际性创新带来了机遇。在AI大模型训练和推理过程中,提供数据的高性能和高安全传输是AI业务对智算中心网络的两大核心需求。高效的负载均衡、拥塞控制技术和网络安全协议是其中的关键网络技术。为应对大规模AI业务带来的严峻挑战,提出全调度以太网(global scheduled Ethernet,GSE)作为对应的解决方案,并搭建真实的测试环境对GSE和RoCE(remote direct memory access over converged Ethernet)网络进行性能对比测试。测试结果证明,GSE相较RoCE网络显著改善了任务完成时间(job completion time,JCT)。展开更多
A new dual directional silicon-controlled rectifier based electrostatic discharge (ESD) protection device suitable for 12-V applications is proposed in this paper. The proposed device (NPEMDDSCR) is based on the embed...A new dual directional silicon-controlled rectifier based electrostatic discharge (ESD) protection device suitable for 12-V applications is proposed in this paper. The proposed device (NPEMDDSCR) is based on the embedded DDSCR (EMDDSCR) structure, in which the P+ electrode and P+ injection are removed from the inner finger. Compared with the conventional modified DDSCR (MDDSCR), its high holding voltage meets the requirements for applications. Compared with the embedded DDSCR (EMDDSCR), it has good conduction uniformity. The MDDSCR, EMDDSCR, and NPEMDDSCR are fabricated with an identical width in a 0.5-μm CDMOS process. In order to verify and predict the characteristics of the proposed ESD protection device, a transmission line pulse (TLP) testing system and a two-dimensional device simulation platform are used in this work. The measurements demonstrate that the NPEMDDSCR provides improved reliability and higher area efficiency for 12 V or similar applications. The measurement results also show that the NPEMDDSCR provides higher robustness and better latch-up immunity capability.展开更多
文摘AI大模型正引领下一个十年的信息与通信技术(information and communications technology,ICT)产业发展热点。智算中心网络是支撑AI大模型分布式训练的通信底座,是决定AI集群效能的关键要素之一。AI大模型的数据量和参数量不断扩张,给智算中心网络带来了严峻的挑战,同时给关键网络技术进行代际性创新带来了机遇。在AI大模型训练和推理过程中,提供数据的高性能和高安全传输是AI业务对智算中心网络的两大核心需求。高效的负载均衡、拥塞控制技术和网络安全协议是其中的关键网络技术。为应对大规模AI业务带来的严峻挑战,提出全调度以太网(global scheduled Ethernet,GSE)作为对应的解决方案,并搭建真实的测试环境对GSE和RoCE(remote direct memory access over converged Ethernet)网络进行性能对比测试。测试结果证明,GSE相较RoCE网络显著改善了任务完成时间(job completion time,JCT)。
基金Project supported by the National Natural Science Foundation of China(Grant Nos.61704145,61774129,and 61827812)the Natural Science Foundation of Hunan Province,China(Grant No.2019JJ50609)the Key Technology Program of Changsha City,China(Grant No.kq1902042).
文摘A new dual directional silicon-controlled rectifier based electrostatic discharge (ESD) protection device suitable for 12-V applications is proposed in this paper. The proposed device (NPEMDDSCR) is based on the embedded DDSCR (EMDDSCR) structure, in which the P+ electrode and P+ injection are removed from the inner finger. Compared with the conventional modified DDSCR (MDDSCR), its high holding voltage meets the requirements for applications. Compared with the embedded DDSCR (EMDDSCR), it has good conduction uniformity. The MDDSCR, EMDDSCR, and NPEMDDSCR are fabricated with an identical width in a 0.5-μm CDMOS process. In order to verify and predict the characteristics of the proposed ESD protection device, a transmission line pulse (TLP) testing system and a two-dimensional device simulation platform are used in this work. The measurements demonstrate that the NPEMDDSCR provides improved reliability and higher area efficiency for 12 V or similar applications. The measurement results also show that the NPEMDDSCR provides higher robustness and better latch-up immunity capability.
