论文部分内容阅读
为改善周期精确级功耗分析的准确度和速度问题,使用多维特征参数建立贝叶斯推理的动态功耗模型.基于功耗分布与电路内部节点状态的分析,发现仅使用端口信息作为参数的不足.定义了门单元级数的计算和对应切片的概念,提出使用切片分析的技术提取电路内部关键层的翻转密度作为参数,与端口信息共同参与贝叶斯推理.基于ISCAS85基准电路的实验结果表明,该方法使原始模型的误差降低21.9%,均方差降低25.0%,同时保持了相对现有门级功耗分析700倍的加速比.
In order to improve the accuracy and speed of accurate periodic power analysis, a multi-dimensional feature parameter is used to establish Bayesian inference dynamic power model. Based on the analysis of power consumption distribution and the state of nodes in the circuit, we find that only the port information is used as a parameter Define the concept of gate unit level calculation and corresponding slice, and propose the use of slice analysis technology to extract the flip density of the key layer in the circuit as a parameter, and participate in Bayesian reasoning with the port information.Based on the experimental results of ISCAS85 reference circuit It shows that this method reduces the error of the original model by 21.9% and the mean square error by 25.0% while maintaining the 700 times speedup compared with the existing gate-level power analysis.