Research On Multi-core SoC Mapping Method Based On Discrete Firefly

Study on NoC mapping is one of the hottest topics in NoC research area. Performance, such as latency or powerof NoC, is directly influenced by the way of mapping. Most of existing research can solve the problem of NoC mapping by using heuristic optimization algorithm, but cause hugetimeconsumption. Especially when the size of NoC is increased, more time and resources are needed. Improved algorithm can promote the speed of optimizationat the cost of the accuracy of solution. Firefly algorithm is a new heuristic optimization algorithm. This dissertation can solve the problem of NoC mapping based on Discrete Fireflyalgorithm. The main work is as follows:1. We propose 2D NoC DFA mapping algorithm. In order to solve the problem that Discrete Firefly Algorithm can’t apply to 2D NoC, we redefine firefly class. In this dissertation, initialization rules of 2D NoC mapping algorithm is redefined,in order to randomly generatea mapping scheme. Method of calculating the distance between fireflies is redefined and it is used as an impact factor to judge the fireflies moving. In order to find better mapping schemes, firefly moving rule is defined as α moving step and β moving step. The experiment results of 2D NoC mapping show that the total communication costs of the best mapping scheme is optimized by 41.46%, compared with the optimal mapping scheme of initialization.Compared with DPSO mapping algorithm, the explore ability of DFA mapping algorithm is better.The micro flit delay is optimized by 58.005%, and the packet delay is optimized by 42.874%.2. We propose 3D NoC DFA mapping algorithm.2D NoC DFA mapping algorithm cannot solve the mapping problem in 3D NoC, so we need to redefine the relevant definitions, initialization rules, and the calculation of distance between the fireflies in 2D NoC DFA mapping algorithm. The experiment results of 3D NoC DFAmapping show that it can find the optimal mapping scheme, compared with the DPSO mapping algorithm.3. We design Network interface support for multiple access and traffic generator which isbased on self-similar. For one to many, many to one communication relationship in the sub task of applications, a network interface RTL model support for multiple accesses is designed, and combined with the existing NoC simulation platform to complete the mapping simulation experiments. In order to simulate the flow characteristics in actual NoC better, we design a traffic generator based on self-similarity. The sending packet cycle and idle cycle length are produced by the ON/OFF model. Simulation experiments of these two modules show that their function is correct.