基于遗传算法的NoC处理单元映射研究

NoC Process Elements Mapping Using Genetic Algorithm

传统的基于总线的SoC体系结构及设计方法在解决多处理器的复杂系统中将遇到瓶颈,有效解决方案NoC(片上网络)成为新的发展趋势。本文研究了广泛使用的二维规则型网络(2D-mesh)对影响系统性能的重要因素——功耗建立模型,以及形成处理单元位置映射等问题。最后运用遗传算法来寻找已建立的功耗模型最优解或近最优解。试验结果表明,遗传算法能够使得目标函数很快收敛,起到了很好的全局寻优效果。

Traditional bus-based SoC architecture and methodology become data exchange bottleneck in complex multi- processor system. New trend of Network on Chip provides a viable solution. 2D-mesh is a widely used NoC topology. This paper studies 2D-mesh network optimization by way of minimizing communication power consumption, which is very import to the system performance. The authors first establish a 2- mesh communication power consumption model, formulate the problem of energy-aware mapping, and then apply genetic algorithm on the model. The result is a fast converged and optimized solution.