Research On Multi Objective Network On Chip Mapping Optimization Algorithm

With the increasing demand for computer performance and the continuous development of integrated circuit technology,more and more processing cores and memory components are integrated on one chip.The traditional bus-based System on Chip(SoC)communication is inefficient and has poor scalability,which cannot handle the communication tasks between cores well.Network on Chip(NoC)has become an important development direction in this field by virtue of its efficient transmission of data between multiple cores.The mapping problem is a hot issue in the field of NoC research,and its mapping results will directly affect the power consumption,latency and other properties of the chip.Among them,power consumption and delay are the most important performance indicators of NoC,and their optimization requirements are very urgent.The NoC mapping problem is an NP-hard problem.How to effectively obtain a low-power and low-latency mapping scheme has become a research difficulty.In view of the above problems,this paper first analyzes the power consumption and delay indicators of NoC.In order to accurately evaluate and calculate the power consumption and delay data of NoC system,the power consumption and delay of NoC are scientifically evaluated and mathematically modeled,and a multi-objective mapping optimization model for low power consumption and delay is established.Then,this paper studies the mapping problem of NoC.After in-depth analysis of the task characteristics and mapping process,the mapping is divided into two phases: mapping from task to IP core,mapping from IP core to NoC platform.Finally,this paper improves on the basic genetic algorithm,and proposes a low-power and low-latency NoC multi-objective mapping optimization algorithm TI_GA to solve the mapping problem in the heterogeneous multi-core network system environment.The algorithm improves the initial population quality and accelerates the convergence speed of the algorithm by constructing a good population individual in the initial population.Then,a two-population genetic mechanism is adopted in the iterative process of the algorithm.The two sub-populations evolve independently of each other,and in the evolution process,according to the requirements of evolutionary algebra,individuals exchange between the two groups.The improvement of the above algorithm not only ensures the diversity of the population,but also avoids the algorithm falling into the local optimal solution,and finally finds a more effective low-power and low-latency mapping scheme.The above schemes are proposed to minimize the power consumption and delay of NoC system as much as possible.This paper realizes the simulation of NoC on BookSim simulator,and combines three test examples of MPEG-4,MWD and VOPD.The horizontal comparison experiments of convergence,single-objective optimization and multi-objective optimization are carried out by using basic genetic algorithm,ant colony algorithm and TI_GA algorithm.The experimental results show that the convergence rate of TI_GA,a heterogeneous multi-core Network on Chip mapping algorithm based on improved genetic algorithm,is much faster than that of the other two algorithms.In the experiment of single objective optimization,compared with the basic genetic algorithm and ant colony algorithm,the power consumption of TI_GA is reduced by 13.8% and 14.2% on average,and the delay is reduced by 18% and 15.3% on average.In the experiment of multi-objective optimization,the overall performance of NoC is improved although the power consumption and delay are slightly lower than that of single objective optimization.TI_GA can weigh the weights of optimization indices according to the actual requirements of application,so as to obtain the maximum optimization effect of power consumption and delay.So the mapping algorithm TI_GA for multi-objective optimization proposed in this paper is very suitable for NoC and has a good development prospect.