基于机器学习的快速时序校准方法

Fast Time Calibration Method Based on Machine Learning

针对布局布线工具和时序签核工具的时序分析差异,导致的迭代次数多、时序收敛困难的问题,提出一种基于机器学习的快速时序校准方法。首先,基于55 nm工艺,利用开源设计收集数据样本;然后,分别采用Lasso线性回归、BP神经网络、随机森林算法完成寄生参数预测模型的训练、测试及对比;最后,通过实验验证该方法的时序校准效果。实验结果表明,该方法可减少布局布线工具和时序签核工具间的时序分析差异。

A fast timing calibration method based on machine learning is proposed to solve the problems of multiple iterations and difficult timing convergence caused by the difference of timing analysis between the layout and routing tool and the timing signature tool.First,based on the 55 nm process,data samples were collected by open source design;Then,Lasso linear regression,BP neural network and random forest algorithm are respectively used to complete the training,testing and comparison of parasitic parameter prediction models;Finally,the timing calibration effect of this method is verified by experiments.The experimental results show that this method can reduce the time sequence analysis difference between the layout and routing tool and the time sequence signature tool.