【作者】 任长宁；

【导师】 马光胜；

【作者基本信息】 哈尔滨工程大学 ， 计算机应用技术， 2004， 硕士

【摘要】 随着硅片上集成的晶体管数量迅猛增加，基于IP核的SOC已成为VLSI实现技术发展的趋势。目前SOC的系统设计主要采用基于IP核的配置并执行方法。这种方法是从一个预设计的参数化SOC体系结构出发，以参数化的IP核为组件，通过编写代码、设置参数的方式，对可编程的IP核进行配置以实现设计，通过模拟器仿真执行以完成功能验证，最后产生物理芯片。由于IP核的多样性及其可优化参数的矛盾性，使得SOC的设计空间极其复杂。能否找到一组最优的参数，直接关系到设计的成败。SOC系统综合的主要任务之一就是针对具体的应用在可能的设计空间中找到一组满足设计约束的IP可行配置集，其本质是求多目标优化问题的最优解。 多目标优化问题的经典求解方法是使用目标函数线性聚合或者基于Pareto方法。这类方法通常是将若干个子目标聚合成向量函数，转换成单目标优化问题，其最大的缺点是优化结果为单个解而非Pareto最优集合。使用演化算法求解多目标优化问题的优点在于该方法将解集作为群体，并行搜索多个Pareto最优解对应的目标空间节点。在对SOC系统综合本质和多目标演化算法分析的基础上，针对SOC设计空间的复杂性，提出了一个新的搜索策略，使得SOC系统应用在消耗较低功率的同时，执行的更快。该搜索策略以多目标演化算法为核心，依据参数依赖性概念对设计空间进行大幅度的缩减并使用空间阈值技巧增加了策略的适应性。通过与敏感度分析策略和穷举搜索策略的对比，证明了该搜索策略具有运行时间快、求解质量高的优点。更多还原

【Abstract】 The availability of large numbers of transistors on a chip has lead to the growth of IP (Intellectual Property) based design SOC (System-on-a-Chip) architectures. Many IP based SOC design approaches focus on mapping an application onto a previously designed complex SOC architecture built from an existing IP by configuring and extending the architecture, representing a configure and execute methodology. Such a parameterized SOC architecture is supported by a programming simulation and emulation environment, and may be provided as HDL source code, as an actual chip, or both. Due to the varieties of IP and the conflict of IP parameter, the SOC design space is very complex. An important SOC design work is the configuring of all cores’ parameters, such that the architecture is tuned for the application, i.e., the software running on the SOC architecture, and for the power, size and performance constraints of SOC. One main task of the SOC system-level synthesis is design space exploration. The essential of the task is finding the optimal set of solutions to a multiobjective optimization problem.The classic solution to multiobjective optimization problem is using objective function linearity aggregation or Pareto based approach. These kinds of methods usually aggregate some subobjective function into vector function so as to convert multiobjective problem into single objective. The most defect of these approaches is that the optimized result is a single solution, not Pareto-optimal set of solutions. The advantage of the Evolutionary Algorithm is that it explores parallel numerous Pareto-optimal solutions corresponding to the objective space node by regarding solution set as population.Based on the analysis of SOC system-level synthesis and multiobjective evolutionary algorithm theory, we proposed a new design space exploration strategy to find a tradeoff of power and execute time for an application running onthe parameterized SOC architecture. The strategy uses IP parameter interdependency to reduce the design space and chooses exploration algorithm to explore design space according to the threshold of space size. The strategy is proved to be able to find better Pareto-optimal configuration effectively, and at the same time accelerate the speed of the design space exploration compared with sensitivity analysis strategy and exhaustive search algorithm.更多还原