三维片上网络功耗评估及映射算法-微电子学与固体电子学专业论文.docxVIP

摘要 摘要 三维 三维片上网络功耗评估及映射算法 摘要 采用 TSV 作为垂直互联的三维片上网络结构结合了三维电路和片上网络两者 的优势，能获得更高的带宽、更低的延时和更高的吞吐量。但三维片上网络的高 集成度也给芯片温度管理问题提出了更高的要求，而该问题的基础是低功耗问题 的研究。所以本文以片上网络的互联功耗为研究对象。 本文首先建立了三维片上网络的互联功耗模型。互联包括路由和链路两部分， 而链路又分为水平链路和垂直链路 TSV 两类。每一部分的功耗都分为动态功耗和 静态功耗两部分，分别进行讨论。接着，本文阐述了片上网络的映射问题，并介 绍了三种映射优化算法。文章最后评估了二维片上网络结构、一层 Cache 和两层 Cache 的三维片上网络结构在不同注入率下的互联功耗，讨论了不同端口的路由和 针对不同包格式的片上网络结构在功耗方面的优劣，并比较了遗传算法、分支界 限算法和模拟退火算法三种映射算法的优化结果。 仿真结果明确了三维片上网络结构相较于二维结构在功耗方面的优势，在包 格式为 16bits*16flits 的情况下，一层 Cache 和两层 Cache 的三维结构与二维结构 相比，功耗分别降低了 12%和 22%。采用包交换策略时，包格式和路由结构的选 择都会对片上网络的功耗有显著影响。映射算法的仿真结果表明，不同的映射算 法对功耗有不同程度的优化。模拟退火算法与遗传算法和分支界限算法相比，优 化效果更明显，功耗降低的比例高 12%。 关键词：三维片上网络 TSV 功耗建模 映射算法 Abst Abstract 三维 三维片上网络功耗评估及映射算法 Abstract 3D NoCs can achieve high bandwidth, low latency and large throughput，which combine the benefit of through silicon vias and 3D ICs. Due to high packaging density of 3D NoCs, temperature management becomes a crucial problem, therefore, the research objects of this paper are interconnect power and cache power. This thesis focuses on establishing detailed power consumption models of interconnection and cache in 3D NoCs. The interconnected components include router and links which can further be separated into horizontal links and vertical links. The power of each component is divided into dynamic power and static power. Then, Genetic Algorithm, Branch-bounds Algorithm and Simulated-annealing Algorithm are introduced to solve mapping problem of NoCs. Finally, the energy performances of 2D NoC, 3D NoC with one cache layer and two cache layers under different injections are compared. Additionally, the energy performance of 3D NoC structures with different kinds of routers and different packet formats are explored. And the results of three optimization algorithms are also analyzed. The simulation results indicate that 12% and 22% power consumption can be respectively reduced by adopting two kinds of 3D Mesh NoC, and other factors, such as router parameters and packet formats, have a great