Research On NoC Mapping Algorithms For Low-Power

Network on Chip(NoC) is an important direction of System-on-Chip development.Development of low-power NoC is one of the Key Technologies which promotedevelopment of NoC. At present, the mapping algorithm is designed to reduce power,which has received good effect. But as a newly developed algorithm, it also has manydefects and needs to be developed and improved furtherly.Aiming at the low-power mapping problems in NoC design, two different mappingalgorithms are proposed. Firstly, NoC low-power mapping algorithm based on geneticalgorithm has poor convergence. In order to increase convergence speed, the mappingalgorithm is improved from two aspects: fitness function and fitness for appropriateparent population, this population is composed of some excellent individual andStochastic Population. Secondly, standard genetic-simulated annealing algorithm has thedisadvantages of premature and slow convergence. The improved genetic-simulatedannealing algorithm is designed from two aspects of adding memory and adjusting crossand mutation probabilities adaptively using premature judgment sign. It is applied toNoC mapping problems. And NoC low-power mapping algorithm based on improvedsimulated annealing genetic algorithm is constructed.The experimental results show that two mapping methods mentioned above havegood effect on the amount of communication traffic, average communication distanceand total amount of power consumption. The comparison indicates that the two methodsin this paper have different superiority for different maping problems. Because mappingalgorithm is designed just according to power consumption priority model, in thefollow-up work, NoC mapping algorithm for multiple target optimization will bestudied furtherly.