中国科学院沈阳自动化研究所机构知识库
Advanced  
SIA OpenIR  > 信息服务与智能控制技术研究室  > 学位论文
题名: CMP过程Run-to-Run预测控制方法研究
其他题名: Research on Run-to-Run Predictive Control Methods for CMP Process
作者: 王亮
导师: 胡静涛
分类号: TP273
关键词: 半导体制造 ; 化学机械研磨 ; Run-to-Run控制 ; 预测控制 ; 过程控制
索取号: TP273/W34/2012
学位专业: 机械电子工程
学位类别: 博士
答辩日期: 2012-05-29
授予单位: 中国科学院沈阳自动化研究所
学位授予地点: 中国科学院沈阳自动化研究所
作者部门: 信息服务与智能控制技术研究室
中文摘要: 半导体制造产业正在全球范围内高速发展,先进的半导体制造厂商已经开始大规模采用芯片关键尺寸为65nm的半导体制造技术,300mm晶圆已成为主流产品。随着半导体器件关键尺寸的不断减小、集成度的不断提高和晶圆尺寸的不断增大,半导体制造过程变得越来越复杂,对半导体制备及其自动化水平要求越来越高。半导体制造过程控制性能直接决定半导体产品的良率、再工次数和半导体制造企业的产能及经济效益,半导体制造过程控制的研究具有重要的意义。受过程扰动和漂移的影响,半导体制造不同批次间需采用不同的制程方案(Recipe)。由于大部分半导体制造过程缺乏在线传感器,产品的质量不能在线测量,只能在当前批次结束后离线测量,因此Run-to-Run(R2R)控制成为半导体制造过程控制的主要方式。R2R控制根据对历史批次数据的分析,通过某种R2R控制方法调节不同批次间的制程方案,达到抑制扰动和漂移的影响,降低不同批次间产品的质量差异,进而提高产品良率的目的。    化学机械研磨(CMP)是半导体制造中重要的过程之一,能实现器件多层金属互连和晶圆表面全局平坦化。CMP过程性能对后续工艺如光刻、蚀刻非常重要,是实现超大规模集成电路(ULSI)制造的关键技术。CMP过程包括复杂的物理和化学过程,受环境噪声、研磨残料和研磨垫老化等因素引起的过程扰动和漂移的影响,随着批次的增加,产品质量不可避免的存在差异,R2R控制成为CMP过程控制的主要手段。CMP过程R2R控制成为半导体制造过程控制中重要的研究内容,CMP过程R2R控制方法的研究具有重要的理论意义和实用价值。    本文依托国家科技重大专项课题“集束式控制平台与软件”(2009ZX02001-005)、“设备模块控制器开发与系统参数分析及模块国产化” (2009ZX02008-003)和沈阳市科技计划项目“符合SEMI标准的IC装备自动化控制系统开发平台”(108155-2-00),针对非线性、时变和扰动不可测量的CMP过程R2R控制问题,重点开展了以下五个方面的研究工作:    首先,对CMP过程机理和预测控制的原理进行了充分的分析,给出了CMP过程的模型,提出了一种基于非线性预测模型和智能优化算法的CMP过程R2R预测控制方法的框架。由于半导体制造过程均具有非线性和时变等特征,此框架对半导体制造其它过程R2R控制器的设计具有重要的参考价值。    第二,提出了基于T-S模糊预测模型和广义预测控制算法的CMP过程单输入单输出R2R预测控制方法。使用历史批次样本数据由G-K聚类算法、最小二乘法辨识CMP过程的T-S模糊预测模型并采用递推最小二乘法在线更新此预测模型,解决了线性预测模型的失配问题,提高了预测模型的精度。将T-S模糊预测模型转化为CARIMA模型,采用广义预测控制算法求取最优控制律,提高了控制精度。仿真实验验证了方法的有效性。    第三,提出了基于RBF神经网络预测模型和粒子群(PSO)滚动优化算法的CMP过程多输入单输出R2R预测控制方法。使用历史批次样本数据由减聚类算法、最小二乘法建立CMP过程的RBF神经网络预测模型,采用GM(1,1)模型由前驱批次的预测误差对后续批次的扰动进行估计实现反馈校正,解决了难以建立CMP过程非线性预测模型的难题。采用基于PSO的滚动优化算法求取最优控制律,解决了基于梯度的滚动优化算法易于陷入局部最优的问题,提高了控制精度。仿真实验验证了方法的有效性。    第四,提出了基于在线灰色GM(1,N)预测模型和克隆选择(CSA)滚动优化算法的CMP过程多输入单输出R2R预测控制方法。使用历史批次样本数据建立CMP过程的灰色GM(1,N)预测模型,采用加权平均法由前驱批次预测误差对后续批次的扰动进行估计实现反馈校正,解决了难以建立CMP过程在线非线性预测模型的难题,提高了在线预测模型的精度。采用基于克隆选择的滚动优化算法求取最优控制律,提高了控制精度。仿真实验验证了方法的有效性。    第五,提出了基于最小二乘支持向量机预测模型和克隆选择多目标滚动优化算法的CMP过程多输入多输出R2R预测控制方法。使用历史批次样本数据分别建立CMP过程的MRR、WIWNU的最小二乘支持向量机预测模型,采用贝叶斯证据框架方法对模型中的参数进行优化,采用双指数加权移动平均方法由前驱批次的预测误差对后续批次的扰动进行估计实现反馈校正,解决了难以建立CMP过程非线性多输入多输出预测模型的难题。采用基于克隆选择的多目标滚动优化算法求取最优控制律,解决了CMP过程多变量控制的难题和提高了控制精度。仿真实验验证了方法的有效性。    综上所述,本文提出了基于非线性预测模型和智能优化算法的CMP过程R2R预测控制方法的框架,在此框架的基础上分别提出了单输入单输出、多输入单输出和多输入多输出的R2R预测控制方法,为CMP过程和其它半导体制造过程的R2R控制器的研究提供了一种新的研究思路和方法
英文摘要: Semiconductor manufacturing industry is growing rapidly worldwide. Advanced semiconductor manufacturers have begun a large-scale use of the semiconductor manufacturing technology that the chip critical dimension is 65nm and the wafer diameter is 300mm. With the decreases of critical dimension v.s the raises of the integration of chips and the increases of wafer size, the semiconductor manufacturing processes have begun more complex and required more higher equipment automation level. As a result of the yield,rework and capacity of the semiconductor manufacturing decided directly by process control performance,the research of semiconductor manufacturing process control has important significance. Due to influences of disturbances and drifts, different recipes needs in different runs. Owing to the lack of in-situ sensors and impossible of online measurement of wafer in most caces,run-to-run(R2R) control has begun an key process control means.R2R control can improve yield, suppress disturbances and drifts,and reduce the quality differences between different runs based on the analysis of datas of historical runs and adjustment of recipes of different runs.    Chemical Mechanical Polishing(CMP)process that can achieve multi-layer interconnection and global planarization have begun an important process in semiconductor manufacturing.CMP process performance is getting more critical to the next processes such as lithography or etch and is a key process in production of ULSI.Run-to-Run control is an key means in CMP process comprised of complicated physical and chemical processes.Due to influences of disturbances and drifts caused by environmental noise,pad aging,grinding residue and other factors,with the increase of the runs,product quality variates inevitably.CMP process R2R control research has important theoretical significance and practical value.    With the support of the major national science and technology special issues named "The cluster control platforms and software" (2009ZX02001-005),and "The equipment module controller development ,the system parameters analysis and modules localization"(2009ZX02008-003)and the Shenyang Science and Technology Program named "The  development platform for IC equipment automation and control system meeting SEMI standard "(108155-2-00), aiming at the R2R control problem of CMP process with characteristics of nonlinear ,time-varying and product quality no being measured,the paper focused the research and development work on the following five aspects:    First, the mechanism of the CMP process and predictive control theory are analyzed and CMP process model is given. Then,a framework of CMP process R2R predictive controller based on nonlinear prediction model and intelligent optimization algorithm is proposed and this framework has an important reference value to the other semiconductor manufacturing process controller designing.    Second, a CMP process SISO R2R predictive control method based on T-S fuzzy prediction model and general predictive control(GPC) is proposed. T-S fuzzy prediction model that is off-line identified by G-K clustering algorithms and least squares method and conclusion parameters is on-line updated by recursive least squares method solves the dismatch problem of linear model and improves prediction control .T-S fuzzy prediction model changed into CARIMA model and optical control law obtained from GPC method improves the control precision.The simulation results show the effectiveness of this R2R predictive control method.    Third, a CMP process MISO R2R predictive control method based on RBF neural network prediction model and particle swarm optimization(PSO) algorithm is proposed.RBF neural network prediction model constructed by the datas of historical runs,subtractive clustering algorithm and least squares method and feedback correction achieved from disturbances estimation by grey GM(1,1) model solves the problem of modeling of CMP process nonlinear prediction model. Optimal control law derived from PSO receding horizon optimization algorithm solves the problem of local optimum about gradient-based optimization and improves the control precision. Finally,the simulation results illustrate the effectiveness of this R2R predictive control method.    Fourth, a CMP process R2R predictive control method based on online grey GM(1,N) prediction model and clonal selection algorithm(CSA) is proposed. CMP process GM(1,N) prediction model constructed by the datas of historical runs and grey theory and feedback correction got from disturbances estimation by the method of weighted average of prediction errors of prior runs solves the problem of modeling of CMP process nonlinear online prediction model.Optimal control law received from CSA receding horizon optimization algorithm improves the control precision. The simulation results illustrate the effectiveness of this R2R predictive control method.    Finally, a CMP process MIMO R2R predictive control method based on least squares support vector machine (LS-SVM)prediction model and clonal selection multi-objective optimization algorithm. Prediction models of material removal rate(MRR) and within-wafer nonuniformity(WIWNU )constructed by datas of historical runs and LS-SVM and optimized by Bayes evidence framework and feedback correction implemented by disturbances estimation from prediction errors of prior runs and dEWMA method solves the difficult problem of modeling of CMP process MIMO nonlinear prediction model.Optimal control law received from CSA multi-objective receding horizon optimization algorithm solves the difficult problem of CMP process MIMO R2R control and improves the control precision.The simulation results illustrate the effectiveness of this R2R predictive control method.    In summary, the framework of CMP process R2R predictive methods based on nonlinear prediction model and intelligent optimization algorithm is put forward. Based on this framework,four methods involving SISO, MISO and MIMO CMP process R2R predictive control are proposed.The proposed methods can provide a few new methods for R2R controller research of CMP process and other semiconductor manufacturing processes.。
语种: 中文
产权排序: 1
内容类型: 学位论文
URI标识: http://ir.sia.cn/handle/173321/9413
Appears in Collections:信息服务与智能控制技术研究室_学位论文

Files in This Item:
File Name/ File Size Content Type Version Access License
CMP过程Run-to-Run预测控制方法研究.pdf(1410KB)----限制开放 联系获取全文

Recommended Citation:
王亮.CMP过程Run-to-Run预测控制方法研究.[博士学位论文].中国科学院沈阳自动化研究所.2012
Service
Recommend this item
Sava as my favorate item
Show this item's statistics
Export Endnote File
Google Scholar
Similar articles in Google Scholar
[王亮]'s Articles
CSDL cross search
Similar articles in CSDL Cross Search
[王亮]‘s Articles
Related Copyright Policies
Null
Social Bookmarking
Add to CiteULike Add to Connotea Add to Del.icio.us Add to Digg Add to Reddit
所有评论 (0)
暂无评论
 
评注功能仅针对注册用户开放,请您登录
您对该条目有什么异议,请填写以下表单,管理员会尽快联系您。
内 容:
Email:  *
单位:
验证码:   刷新
您在IR的使用过程中有什么好的想法或者建议可以反馈给我们。
标 题:
 *
内 容:
Email:  *
验证码:   刷新

Items in IR are protected by copyright, with all rights reserved, unless otherwise indicated.

 

 

Valid XHTML 1.0!
Copyright © 2007-2016  中国科学院沈阳自动化研究所 - Feedback
Powered by CSpace