The Study Of Large-Scale Hardware-Software Partitioning Method For SoC Design

With the development of embedded systems and microelectronics technologies, SoC emerges as times require and becomes a mainstream design approach of embedded systems. The traditional methods cannot meet the particularities of SoC designs, and have become primary limits for SoC design. Hardware-software co-design technology is high-effective means to design SoC products, and hardware-software partitioning is one of the key technologies of software and hardware co-design, but its solution is a NP-complete problem. If its function is complex in SoC design, the traditional algorithms will be terrified by the sight of its huge solution space, and the result is with difficulty satisfying. The hardware-software partitioning methodology for SoC designs is systematically researched in this dissertation, some main works and contributiveness including:Traditional design and hardware-software co-design have been compared, and flow and characteristic also have given. Summarized the existing hardware-software partitioning methodology, and came to the conclusion that large-scale question is the new question to be faced. Also, key and step have been given for large-scale hardware-software partitioning in SoC design. Above these foundations, architecture model and specification model for hardware-software partitioning have been described, and then the goal and system parameters have been given.Proposed one kind of real-coded frame for hardware-software partitioning, and gave a non-uniform mapping method. With constrain of system performance, we have studied the real-coded mechanism for hardware-software partitioning, and both the uniform and non-uniform mapping method has been analyzed by comparisons.We have given the definition of constrained optimization, as well as its research status. We have studied Runarsson and Yao's SR and ISR, as well as Zhang Min's DSS-MDE. Based Runarsson and the Yao's work, through studying selection strategy and difference evolution, proposed a new algorithm SRDE. Finally, SRDE has been tested on 13 benchmark problems and hardware-software partitioning problems.Through studying the three key issues of cooperative optimization algorithm, as well as the difference evolution algorithm, has given the cooperative difference evolution algorithm. We focused on how to reduce the dimension to decompose the question, and the stochastic ranking was put forward within the framework of differential evolution, as well as combined method was proposed in fitness evaluation. We have studied hardware-software partitioning for multi-processor structure SoC. First of all, the proposed cooperative difference evolution algorithm was used to assign processing units, and then got system execution time using task scheduling algorithm. By analyzing the existing task scheduling algorithm, system algorithm framework has been given.