片上网络高性能互连技术的研究

项目名称： 片上网络高性能互连技术的研究

项目编号： No.60803038

项目类型： 青年科学基金项目

立项/批准年度： 2009

项目学科： 无线电电子学、电信技术

项目作者： 顾华玺

作者单位： 西安电子科技大学

项目金额： 20万元

中文摘要： 片上网络解决了总线结构芯片所面临的通信效率低、功耗大以及不易扩展等问题，本项目重点研究了片上网络的两项关键技术拓扑结构和路由算法。我们研究了树形拓扑结构，提出Butterfly Fat-Tree光片上网络结构，该结构具有更少的节点和链路。针对现有网络对局部流量支持力度不够的问题，设计了基于分层簇的拓扑结构并将簇引入传统的Clos网络。提出基于蜂群智能、以低能耗和负载均衡为优化目标的拓扑映射算法，完成IP核到网络节点的匹配。结合小世界网络特性，通过在已有拓扑结构中增加长链路来构建新的拓扑结构，缩小网络直径来降低时延。路由算法方面，本项目针对流量均衡、容错、服务质量保证、死锁等各个单项要求进行研究。基于人工势场理论提出新的容错路由算法，放宽对故障区域形状的要求，同时具有一定的流量均衡能力。基于新的片上网络系统级热量模型提出热量均衡路由算法，能够在保证数据传输性能的同时有效的解决热效应问题。借鉴蜜蜂觅食过程中表现的智能提出一种QoS路由算法。在此基础上提出一种分布式多功能路由算法，实现流量均衡、容错、服务质量保证等功能。此外，我们还研究了拥塞控制机制、路由器结构和调度算法等。

中文关键词： 片上网络；拓扑结构；路由算法

英文摘要： Network-on-Chip (NoC) has overcome the problems of bus based chips, including low communication efficiency, high power consumption and weak scalability, etc. Two key technologies in respect of topology and routing algorithm in NoC have been researched in this project. We study the tree topology and propose a Butterfly Fat-Tree optical NoC which has fewer nodes and links. To support local traffic by existing topologies, a hierarchical cluster based topology is proposed. Moreover, the concept of cluster is also used by Clos network. With the optimization goals of low power consumption and load balance, a mapping algorithm, on the basis of bee swarm intelligence, is proposed to match IP cores to the network nodes. Furthermore, new topology has been achieved by inserting long links into the existing regular topology with the consideration of small world phenomenon, and therefore reducing the network diameter. In respect of routing algorithm, we have carried out research in terms of load balance, fault tolerance, QoS guarantee, deadlock and so on. New fault tolerant routing algorithm has been proposed based on the artificial potential field (APF) model. The APF based fault tolerant routing algorithm can gain load balance and also relax the requirement of fault regions shape. With a new system level thermal model for NoC, a thermal balance routing algorithm is designed to address the problem of thermal effect while the network performance can be maintained. A bee swarm intelligence based QoS routing algorithm is proposed and it can exhibit the foraging process of honeybees. Eventually, a distributed multifunctional routing algorithm is designed with the ability of load balance, fault tolerance and QoS guarantee. Other critical subjects are also studied in this project such as congestion control mechanism, router architecture and scheduling algorithm, etc.

英文关键词： Network-on-Chip; Topology; Routing Algorithm