基于广义极值分布的产量驱动时钟偏差调度.pdf

1、1Yield-driven Clock Skew Scheduling Based on Generalized Extreme Value DistributionKaixiang Zhu,Wai-Shing Luk*and Lingli WangState Key Laboratory of Integrated Chips and SystemsFudan University,C,luk,Jan.21st,202530th Asia and South Pacific Design Automation Conference21 Useful Skew Design:Why and W

2、hy not?Good:If you do it right,spend less time struggling about timing,orget better chip performance or timing yield.Bad:Need more engineer training.Balanced clock-trees are harder to build.Dont know how to handle process variation,multi-corner multi-mode,.,etc.Adjust clock skew to enhance IC perfor

3、mance or robustness.31 Useful Skew Design:Why and Why not?65nm and below,non-Gaussian distributionpath delay.Skewed and long-tailed Traditional Statistical Clock Skew Scheduling(CSS)relies on Gaussian distribution.Real-world path delays:asymmetric,making Gaussian-based methods insufficient.Gaussian

4、distribution.Real path delay distribution.42 Proposed Solution:Generalized Extreme Value(GEV)Distribution:Flexible,asymmetric distribution from extreme value theory.Capture skewness and asymmetry in path delay distributions effectively.Advantages Over Gaussian:High accuracy in approximating delay di

5、stributions.Better modeling for timing yield-driven optimization.GEV distribution.Better timing yield.2 Contributions of Our Work:Introduce GEV distributionReplace Gaussian distribution with GEV for clock skew scheduling.Propose a framework for timing yield-driven CSSEvaluate Parameter Estimation Me

6、thodsCompare MLE,MoM,and L-Moments for fitting accuracy.Achieves superior timing yield Improve timing yield by 8%on benchmark circuits.563 PreliminariesStatistical Static Timing Analysis(SSTA)SSTA:an extension of traditional STA,accounts for process variations.Modern STA tools based on Gaussian dist