摘要
在功耗与信号统计分析的基础上,采用贝叶斯推理技术建立周期精确的功耗宏模型.通过分析信号特征对电路功耗的影响,选择输入信号密度Pin、输入跳变密度Din和输出跳变密度Dout作为贝叶斯推理的三维特征参数,证明了上述特征参数对信号时间和空间相关性信息的覆盖.实验结果表明,该方法较目前的门级功耗分析速度提高400余倍,周期功耗平均误差可以控制在10%以内.
Based on power and signal statistics analysis, this paper chooses Bayesian inference method to build cycle-accurate power macro-model. Through analyzing the influence of signal features on power, we choose input signal probability, input transition density and output transition density as 3D coefficients used in Bayesian inference, and prove those coefficients can denote the temporal and spatial correlation information of signals. The experimental results indicate that the use of the proposed model results in significant （400 ＋ times） speed-ups in power estimation time, while average error in each cycle can be limited to 10 %.
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2007年第10期1241-1246,1251,共7页
Journal of Computer-Aided Design & Computer Graphics
基金
国家"八六三"高技术研究发展计划(2004AA1Z1010)
关键词
功耗模型
贝叶斯推理
三维特征参数
power model
Bayesian inference
three-dimensional coefficient
