基于量子粒群的三维片上网络低功耗映射算法研究

发布时间：2018-01-18 01:29

本文关键词：基于量子粒群的三维片上网络低功耗映射算法研究　出处：《天津工业大学》2016年硕士论文　论文类型：学位论文

【摘要】：二维片上网络(2D NoC)是为了克服基于总线系统的芯片(SoC)体系结构在功耗、通信带宽以及物理设计等方面的局限而诞生的。但随着芯片集成度的进一步提高,2D NoC在布局布线、面积、封装密度以及功耗等方面都已经到达了瓶颈,因而,三维片上网络(3D NoC)应运而生。3D NoC拥有更低的互连损耗、更短的全局互连、更小的体积、更高的封装密度以及更高的性能等诸多优势。在3D NoC的研究中,如何将计算任务映射到3D NoC节点上是关键问题之一,3DNoC映射问题对系统的功耗、延迟等性能均有很大影响,映射优化已成为解决3D NoC降低功耗、改善散热等问题的重要手段,从多个角度研究更好的3D NoC映射算法非常必要。本文对3D NoC映射算法进行了研究,主要完成了以下工作。首先,利用量子粒子群算法的全局收敛性和收敛速度更快的特点,首次将量子粒子群算法应用到3D NoC低功耗映射问题中,并与基于粒子群的3D NoC映射算法进行了对比,仿真结果表明,基于量子粒子群的映射算法比基于粒子群的映射算法收敛速度更快,最大提高90.48%；同时有效地降低了3D NoC的映射功耗,尤其是对于核数为20-80的应用特征图优化效果更为明显,功耗最大降低了20.99%；其次,为了解决大规模3D NoC的低功耗映射问题,提出一种基于多样性控制量子粒子群的低功耗映射算法,并与基于量子粒子群的映射算法进行对比,仿真实验结果表明,当应用特征图规模较大(120核以上)时,该算法仍能保持较稳定的功耗优化效率(4.08%-8.04%),并且收敛速度更快,最大提高了66.7%。
[Abstract]:Two dimensional network on chip (2D NoC) is to overcome the bus system based on chip (SoC) architecture in power, communication bandwidth and physical design of the limitations of the birth. But with the further improvement of chip integration, 2D NoC in the layout, size, package density and power consumption etc have been reached the bottleneck, therefore, three-dimensional network on chip (3D NoC).3D NoC has a lower interconnection loss arises, global interconnect shorter, smaller volume, many advantages of higher packaging density and higher performance. In the research of 3D NoC, how will the computing task is mapped to the 3D node is NoC one of the key issues, power 3DNoC mapping problem of the system, such as delay performance are mapping optimization has become an important means to reduce the power consumption of 3D NoC, improve heat dissipation, from the 3D NoC map of a point of the study is better Method is very necessary. This paper studied the 3D NoC mapping algorithm, mainly completed the following work. Firstly, the characteristics of global convergence and convergence speed of the quantum particle swarm algorithm is faster, the quantum particle swarm algorithm is applied to the 3D NoC low-power mapping problem, and compare with the algorithm of 3D NoC mapping particle based on the population, the simulation results show that the convergence speed based mapping algorithm based on mapping algorithm, quantum particle swarm particle swarm faster, the biggest increase of 90.48%; while effectively reducing the power consumption of the 3D mapping of NoC, especially for the audit for the application of 20-80 feature map optimization effect is more obvious, the maximum power consumption is reduced by 20.99%; secondly, in order to solve the problem of low power mapping large-scale 3D NoC, proposes a control algorithm based on the diversity of low-power mapping quantum particle swarm, and the mapping algorithm based on quantum particle swarm The simulation results show that the algorithm can maintain a more stable power optimization efficiency (4.08%-8.04%) when the application characteristic map is large (120 cores or more), and the convergence speed is faster, and the maximum 66.7%. is increased.

【学位授予单位】：天津工业大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TN47

【相似文献】