基于LS-SVM模型和扰动估计的光刻过程智能R2R预测控制

摘要

针对光刻过程非线性、时变和产品质量不能在线测量的特性,提出了一种基于最小二乘支持向量机(LS-SVM)预测模型、灰色模型估计扰动和克隆选择滚动优化的智能run-to-run (R2R)预测控制策略；由历史批次样本数据构建光刻过程的离线预测模型和由前驱批次的预测误差通过灰色GM (1,1)模型估计扰动实现反馈校正,提高了预测模型的精度；通过基于克隆选择滚动优化算法求解最优控制律,提高了控制精度;性能分析结果表明,提出的控制策略显著降低了关键尺寸(CD)的均方根误差,与EWMA及NMPC方法相比较,RMSE分别平均降低了66％,63％和51％、48％.%For lithography process characteristics of nonlinear, time-varying and not being in-situ measured, this paper proposes a lithography process intelligent run-to-run (R2R) predictive control strategy based on LS-SVM predictive model, the disturbance estimated by grey model and clonal selection receding optimization. LS-SVM off-line predictive model constructed through sample data of historical batches and feedback correction by way of the disturbance estimation obtained from grey GM (1, 1) model constructed by previous predictive errors improves the prediction accuracy. Optimal control law achieved from clonal selection receding optimization improves the control precision. Performance analysis results illustrate that root mean squared error for critical dimension (CD) is brought down substantially and RMSE is reduced about 66% , 63% and 51% , 48% compared with the EWMA and the NMPC method.