Research Of Hardware And Software Partitioning Based On The Large-scale Embedded System

The more rapid development of embedded system and microelectronic technologies is, the higher integration level of hardware which makes integrating CPU,memory and I/O devices into a single chip possible is. SoC become a mainstream design approach of embedded systems with its high integration, good reliability and short time-to-market. With the increase in system complexity, the traditional methods can't meet the particularities of SoC. Therefore, Hardware/Software co-design emerges as needed. , while Hardware/Software partitioning is one key of Hardware/Software co-design.This dissertation generalizes the domestic and abroad developing situation in the research field of hardware and software co-design and proposes Clustering Genetic Algorithm (CGA) based on trends of embedded system development. The new algorithm adopts a classical Genetic Algorithm (GA) as the basic partitioning algorithm, connecting Clustering algorithm and Genetic Algorithm. The clustering algorithm, by which system is pre-partitioned based on the reference task nodes which are visual, is improvement of K-means algorithm. That is dynamic process. The number of reference task nodes is denoted with K value which is dynamic. That makes the result of preliminary partition more precise. After the classification finished, classic GA is applied to the new tasks which are reference task nodes. With the scale of system expanding, CGA could solve the shortcoming of GA, which value is local optimum, especially have better effect on algorithm running time.At the end of this dissertation, GA and CGA are respectively programmed and verified on algorithm running time, cost and fitness using the real data generated by TGFF tool. The verification results show that CGA is able to reduce the complexity of the problem, which not only give a better and more accurate partitioning result, but also have shorter algorithm running time.