龙芯1号处理器结构级功耗评估有效性分析

Validation of Architecture Level Power Estimation for Godson-1 Microprocessor Design

结合龙芯1号处理器实际设计过程,介绍了处理器功耗评估的方法和功耗模型,分别对结构级、电路级功耗评估和实际芯片3种情况进行测试程序仿真.经过量化分析和比较表明:结构级功耗评估具有仿真速度快、评估结果误差和测试程序相关,并且同一测试程序误差能够追随电路级功耗评估等特点,说明了龙芯1号处理器进行结构级功耗评估的有效性.采用该方法可显著提高低功耗处理器结构的设计效率.

Through the development of godson-1 microprocessor design, this paper introduces the methodology of power estimation at architecture level and circuit level, and analyses their power models. We run a lot of benchmarks at architecture level, circuit level and real chip respectively, and the experimental results show that the simulation speed at architectural level is nearly 100 times or more faster than that at circuit level, and power estimation at architectural level has more error than that at circuit level which depend on benchmark, and the power estimation at architectural level can track with power consumption of real chip with different benchmarks. That proves it is valid of godson-1 power model at architectural level, and the architecture level power estimation will be efficient for low-power microprocessor design.