基于遗传算法的电压岛感知的多电压分配

摘要

功耗是当前SoC设计所面临的最大挑战之一,多电压设计是一种降低SoC芯片功耗的有效方法.在后布局阶段应用多电压设计,首先对现有电源网络复杂性度量方法进行改进,然后提出了一个同时考虑功耗、电源网络复杂性及电平转换器的新目标函数,并采用遗传算法进行最优电压分配.对GSRC测试电路的实验结果表明,所提出的算法不仅能有效降低芯片功耗,同时可以将多电压设计的额外开销控制在一个较低的水平.此外,改进的电源网络复杂性度量方法在功耗节省和电平转换器数量方面较已有的有一定的优势.%Power consumption has become one of the most crucial challenges in modern SoC design.Multiple Supply Voltages (MSV) is an effective way to reduce the SoC's power consumption.In this paper,we apply the MSV design method in the post-floorplanning stage.At first,we propose an improved method to model the power supply network complexity,then an novel objective function simultaneously considering power consumption,power network complexity and level shifters' number is proposed and the genetic algorithm is exploited to assign the most appropriate voltage to each module of SoC.Experimental results on GSRC benchmarks show that the proposed algorithm can effectively reduce the chip power consumption,while controlling the overhead of MSV design at a lower level.In addition,the improved model of power supply network complexity has some advantages in power savings and the reduction of the level shifters' number.