摘要
由于多驱动及回路特征使得Mesh结构时钟网络分析较为复杂,现有的定性或定量分析方法都难以针对具体Mesh结构设计寻求到准确的时钟不确定性约束,为此提出基于Mesh结构在片波动简化模型的时钟不确定性的遗传算法求解方法.首先将众多片上偏差源转化为单级延迟概率密度分布,然后进行多级传播叠加为Mesh结构末级驱动点延迟分布,进而缩减变量数目,合理分离时钟网络中树形结构和Mesh结构.在此基础上,借助遗传算法的全局趋优搜索能力来求解Mesh结构性能不确定性问题,以得到更为合理的时序裕量估算.与传统的蒙特卡洛分析方法及定性解析分析方法相比,基于65nm工艺的仿真实验结果证明了该方法的有效性.
Multi-driving and circuit loop characteristics complicate timing analysis on clock mesh networks,and so far as we known,accurate timing uncertainty on specific mesh structures can hardly be pursued by any existing qualitative or quantitative characterization in literature.A simplified on-chip variation model on mesh structure and a genetic approach to timing uncertainty problem in clock mesh network were proposed.The impacts of tremendous variation sources could be converted to delay distributions in the single level,and then added into delay distributions for input drivers in mesh structure by multi-level propagation,which helps to decouple tree and mesh structure in clock network.Based on the simplified on-chip variation model,genetic algorithm was applied to find a more reasonable timing uncertainty in clock networks by the ability of global optimization.In comparison with Monte Carlo analysis and qualitative analysis methods,experimental on mesh structure at 65nm technology node validated our proposal method.
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2010年第11期2045-2052,共8页
Journal of Computer-Aided Design & Computer Graphics
基金
国家"九七三"重点基础研究发展计划项目(2005CB321600)
国家"八六三"高技术研究发展计划(2009AA01Z125)
国家自然科学基金(60803029
60801045)
