Research Of Hardware And Software Partitioning Methodology Based On Genetic Algorithm And Ant Annealing

With the rapid development of the micro-electronics technology, largeamounts of embedded products enter into people’s daily life. The efficientdeveloped products have the characteristics of low cost, small volume and low-power. Software-hardware codesign is software/hardware partition’s mainstreammethods that avoid traditional product development’s defects of mistakelocalization, long cycle times and high cost. In order to meet embedded systemdevelopment, some factors taken into consideration include system modeling,system description, software/hardware partition, selection of partitioningalgorithm.The optimal solution in software/hardware partition is a major researchcontent of this subject and also a starting point of whole embedded system design.In the whole process of software-hardware codesign, selection of partitioningalgorithm is a key to solve software/hardware partition. System performance andcost achieving optimal combination is a factor to be considered for algorithm andis also a major problem that need to be solved for software and hardware.This paper begins with recent research at home and abroad and applicationarea of software-hardware codesign to analyze the defects of traditionalsoftware/hardware partition and then to describe the problem of software-hardware codesign, it sufficiently studies the characteristics of several systemmodeling methods that was suitable for software/hardware partition. GeneticAlgorithm and Ant Algorithm are major research partitioning algorithms for thispaper. Through analysis and comparison of two algorithms’merits and demerits,and synthesizing the advantage of two algorithms, a hybrid algorithms isinnovatively put forward and is applies to SoC(system-on-a-chip). System modeluses TGFF to generate DAG. This algorithm is the first to take advantage of genetic algorithm to obtain several optimal solutions that become initial valueparameters of the ant algorithm. The distributions of pheromone are produced byant algorithm initialization by means of genetic algorithm and ant algorithmcrossover and mutation of positive feedback characteristic to search for theoptimal solution.Finally, through comparative analysis of the data generated by tgff and gotfrom programming and experiment of GA、AA and GAAA, it suggests thatGAAA can overcome the defects of GA’s poor capable of local search and AA’sdeficiency of initial pheromone and thereby can search for a higher precision anda more adaptable solution.