当前位置:首页 > 报告详情

RUNoC:重新注入地下网络缓解大规模NoC拥堵.pdf

上传人: 芦苇 编号:651815 2025-05-01 23页 1.88MB

1、RUNoC:Re-inject into the Underground Network to Alleviate Congestion in Large-Scale NoCXinghao Zhua,Jiyuan Baia,Zifeng Zhaoa,Qirong Yua,Xiaofang Zhoua,Gengsheng ChenabaSchool of Microelectronics,Fudan University,Shanghai,China bJiashan Fudan Institute,Jiaxing,Zhejiang Province,China2025.1.22Outline1

2、Introduction2Motivation3RUNoC Architecture4Experiments5Conclusion1Strong Demand for NoCNetwork-on-Chip(NoC)gains considerable attention for its remarkable advantages in scalability and high bandwidth.Typical Network-on-Chip(NoC)RNRNRNRNPPPPPNode(Core,Mem,Acc)RNRouterPortPFig2.Typical NoC architectur

3、e.Fig1.Celerity block diagram1.Introduction21S.Davidson et al.,The Celerity Open-Source 511-Core RISC-V Tiered Accelerator Fabric:Fast Architectures and Design Methodologies for Fast Chips,in IEEE Micro,vol.38,no.2,pp.30-41.Typical Node Architecturea)RouterRoutingthepacketstotheirdestination.b)Netwo

4、rk InterfaceRX:Receiving and analyzing thepackets from network.TX:Transferring the informationinto flits and send them to network.c)ProcessorCPU/Mem/Computing UnitTypical Network-on-Chip(NoC)Fig.Typical node architecture.Introduction3Long Path CongestionExisted Solutiona)Carefully designed routing a

5、lgorithm Self-relief mechanismOften require extra paths and relay on global information,which results in route latency and high resource usageLong Path Congestion in NoCIntroductionFig2.Prolonged path congestion.Fig1.Node congestion.4b)Two-Level Network(TLN)Pseudo-3D topologySymmetric TLN(S-TLN):two

6、 layers have equalsizes,whereeachnodeinonelayerisassociated with a corresponding node in theother layer.AsymmetricTLN(A-TLN):consistsofacomplete network and a sparse network withfewer nodes.Existing architectures have their own lack including power and area consumption,load balancing and deadlock.Ex

word格式文档无特别注明外均可编辑修改,预览文件经过压缩,下载原文更清晰!
三个皮匠报告文库所有资源均是客户上传分享,仅供网友学习交流,未经上传用户书面授权,请勿作商用。
本文提出了一种名为RUNoC的新型网络架构,旨在缓解大规模网络中长路径拥塞问题。RUNoC包含主网络(M-NET)和地下网络(U-NET),通过将拥塞数据重新注入U-NET来平衡两层的工作负载。文章采用共享行缓冲(SRB)和修改的网络接口(NI),确保没有死锁或活锁发生。CDRU决定何时根据M-NET的拥塞信息和包目的地距离重新注入数据,以减少延迟和实现负载平衡。实验表明,RUNoC在性能上比XY路由提高了60%,在面积效率上比现有技术提高了34%。未来工作包括进行真实世界的交通实验和将RUNoC嵌入多核系统中。
"RUNoC如何缓解大规模NoC中的拥堵问题?" "RUNoC架构的创新之处在哪里?" "RUNoC在性能和面积效率上相比现有技术有哪些优势?"
客服
商务合作
小程序
服务号
折叠