Research In SoC HW/SW Partitioning Technology Based On Multi-Objective Optimization

With the fast development of IC's integrative degree, one electronic system or sub-system can be integrated into one IC chip, which brings the design of IC into an epoch called SoC. During the process of the SoC design, the developments of the design technologies make it possible to tie more and more function into one IC chip. To turn it into realization, a novel design method based on cores is emerged. In this method the system's function is divided into many different cores that can use IP (Intellectual Property) cores to realize their special functions, because special IP cores not only have preferable performance, but also provide a new approach to reduce the system's power. In order to contain the complex IP modules and embedded software into the SoC design, the traditional method used to design SoC system is no longer suitable. Hence, the HW/SW (hardware/software) co-design technology, which takes the HW/SW partitioning into account to adjust the domain of HW and SW at the very beginning of the design, and set the optimal design as its objective, fires up researchers' interest.During the whole process of HW/SW co-design, the HW/SW partitioning algorithm is a key technology. How to keep the balance between the system's speed and its cost, and obtain the maximum of performance cost ratio, is the purpose of HW/SW partitioning. But SoC HW/SW partitioning is regarded as an NP-hard problem, and now can only be approximately settled through some kinds of optimal algorithm. The main work of this dissertation focuses on how to get the optimal solution of HW/SW partitioning during SoC design.Firstly, this thesis systematically and thoroughly discussed the HW/SW partitioning problems during SoC design, and indicates that it is essentially a multi-objective optimal problem.In addition, after discussing the difference among the traditional optimal algorithms, and considering the normal weight-added method that has great subjectivity, a novel HW/SW partitioning method based on multi-objective is proposed in this thesis, which is built on NSGA (non-dominated sort genetic algorithm) and possesses the ability to search multi-objective synchronously. To keep the population's diversity and enhance the algorithm's speed of convergence, the niche technology andelitism-preserving strategy separately are applied to the novel method.Furthermore, a new performance evaluation method based on list scheduling is provided during the process of the system performance evaluation, considering that the traditional algorithms, such as ASAP (as soon as possible) and ALAP (as late as possible), don't take the system's global information into account, but rather schedule the whole system through only its local information that leads the inefficient evaluation. Two key problems, scheduling priority determination and performance evaluation parameter determination, are discussed and the whole pseud-code is provided.Finally the new method's validity is verified by two experiments from the aspects of individual and population.After thoroughly considering the particularity of SoC design, a SoC HW/SW partitioning system is suggested. The system consists of five modules, including database of IP core's characteristic, system model, characteristic analyzing, multi-objective optimal and solution analyzing.As a conclusion, the main contributions of this dissertation are reviewed in the end, and the future work of SoC HW/SW partitioning researches is briefly addressed.