Research Of SoC Hardware/Software Partitioning Methodology Based On Multi-target Performance Evaluation

The birth of embedded system marked the beginning of a new era. Currently, embedded systems are widely used in every area including aerospace, weapon equipments, communication, industrial control, personal electronic products, which form a huge industry.As the rapid development of embedded system and micro electronic technology, the integration of the hardware becomes more and more intensive, which makes it possible to integrate CPU, storing devices and I/O on one chip, SoC emerges on demand, and soon becomes the mainstream technology of modern SoC design for its features of high integration, good reliability, short time-to-market and so on. The abuse of the traditional method of embedded system design is not able to meet the need of modern SoC design any more, this brings a huge challenge and chance to our system designer. For this reason, the research on SoC design technology turns into a hot topic of embedded system design, therefore, hardware/software co-design methodology emerge. Moreover, hardware/software partitioning methodology is the crucial part in co-design technology, hence, researching of SoC hardware/software partitioning methodology, setting up a reasonable description of system model, proposing a partitioning method and performing the optimization will do have significant academic and application value.As there are several factors affect the performances of the embedded systems, the paper particularly analyzes these factors that may affect the performances of the embedded systems, and introduces a point that hardware/software partitioning base on multi-target performance evaluation. Bring IP and software framework reuse into hardware/software partitioning method. Map the complex embedded system configuration on DAG(Direct Acyclic Graph). Introduce the concept of performance priority, perform optimization on condition traversal algorithm by adding the concept of pre-processing the given data and branch-and-bound in operational research into partitioning algorithm, and verify that this algorithm is superior to previous one on solution convergence speed.