基于三维特征参数的贝叶斯推理电路功耗模型  

Three-Dimensional Coefficient Based Bayesian Inference Power Model

作　　者：陈杰[1] 赵晓莺[1] 李险峰[1] 佟冬[1] 程旭[1] 

机构地区：[1]北京大学微处理器研发中心,北京100871

出　　处：《计算机辅助设计与图形学学报》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)

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 %.

关 键 词：功耗模型 贝叶斯推理 三维特征参数 

分 类 号：TP391.77[自动化与计算机技术—计算机应用技术]