摘要
化学机械抛光(chemical mechanical polishing,CMP)过程中由于柔性弹性隔膜的存在使得各腔室之间压力相互耦合,从而使得多区腔室的压力控制变得复杂。针对这一耦合现象,提出了一种将基于动态回归神经网络(dynamic recurrent neural network,DRNN)在线辨识与神经元解耦和分段变参数复合控制相结合的方案。利用DRNN的非线性映射能力以及神经元的在线实时动态解耦特性,获得对象的逆模型,消除了各区之间的耦合;采用分段变参数控制策略减少了由于初始时刻逆控制模型辨识不准而带来的不利影响和系统动荡,使得整个控制系统趋于稳定。实验结果表明:该方案不仅具有很好的在线辨识和解耦能力,同时较常规定参数比例积分微分(proportional integral derivative,PID)控制还具有自适应能力强、响应速度快、超调量小以及鲁棒性好等特点。
During chemical mechanical polishing (CMP) process, the flexible elastic diaphragm couples the processes, which makes pressure control more complicated. The coupling is handled by multi-zone decoupled adaptive control using dynamic recurrent neural network (DRNN) online identification + neural decoupling -t- several section variable parameter compound control. The nonlinear mapping ability of the neural network gives real-time dynamic decoupling for an inverse model of the object to reduce the coupling effects. The several-section variable parameter compound control not only eliminates the adverse effects of the initial inverse model identification errors and system turbulence, but also stahilizes the entire control system. Tests show that the method can provide online identification and decoupling with strong adaptive ability, fast response, small overshoot, and robustness compared with traditional proportional integral derivative (PID) control.
出处
《清华大学学报(自然科学版)》
EI
CAS
CSCD
北大核心
2012年第11期1622-1630,共9页
Journal of Tsinghua University(Science and Technology)
基金
国家科技重大专项(2008ZX02104)
清华大学摩擦学国家重点实验室基金项目(SKLT08B08)
关键词
化学机械抛光
动态回归神经网络
神经元解耦
自适应逆控制
chemical mechanical polishing (CMP)
dynamicrecurrent neural network (DRNN)
neuron decoupling
adaptive inverse control