GOALI: Adjustment and Monitoring Methods for Multiple-Stream and Process-Oriented Quality Control

多元统计过程控制研究已经产生了一些工具，可以用来识别生产中的不规则现象，并描述这种变化的组成部分。在工艺调整的意义上，诊断和控制行动没有建模，而是由工艺工程师来解释和纠正变化的原因。随着时间的推移而漂移的质量特征（自相关）和在多个特征之间以类似的方式变化的质量特征（相互关联）的存在使得多变量统计过程控制成为一项困难的任务。由于这些原因，人们对将过程调整技术与统计过程监控工具集成在一起很感兴趣。质量数据变化的主要成分可以通过根据主成分分析分解数据来发现，但这是一种面向数据的方法，而不是基于任何过程知识，这使得解释变得困难。面向过程的基表示（POBREP）分析使用过程知识将质量数据分解为与原因相关的组件。在POBREP中，每个潜在的生产问题都与一个基本元素相关联。本研究探讨了POBREP可以为流程调整提供一个有效的工具。前面已经展示了如何将POBREP用于过程监视目的。对于过程调整，将调查以下问题：(1)纳入POBREP知识的调整的适当统计模型是什么？(2) POBREP何时可能提供性能优势？(3) POBREP能否有效应用于晶圆制造工艺？这项研究有几个好处。基于预期问题和干扰的监测和调整策略可以改变综合控制策略的无效性能。这项工作包括亚利桑那州立大学、英特尔公司和宾夕法尼亚州立大学的研究人员之间的合作。这项合作包括宾夕法尼亚州立大学和亚利桑那州立大学的研究生研究助理在英特尔实习8个月，宾夕法尼亚州立大学的教职员工访问英特尔，亚利桑那州立大学的教职员工定期访问英特尔。这种合作方式具有广泛的基础设施效益：(1)四位先前分别资助的NSF研究人员协调研究的协同效益；(2) GOALI福利，包括课堂上的工程师，教师参观行业等；(3)利用现有实验室设备利用结果的机会；(4)加强宾夕法尼亚州立大学和亚利桑那州立大学现有的应用统计学课程。

Multivariate statistical process control research has produced tools that can be used to identify when irregularities in production occur and to characterize the components of this variation. The diagnosis and control actions, in the sense of process adjustment, are not modeled and it is up to the process engineer to interpret and correct causes of variation. The presence of quality characteristics that drift with time (auto-correlation) and that vary in similar ways across several characteristics (cross-correlation) makes multivariate statistical process control a difficult task. For these reasons, interest exists on integrating process adjustment techniques with statistical process monitoring tools. The major components of variation in quality data can be found by decomposing the data according to principal component analysis, but this is a data-oriented approach and not based on any process knowledge, which makes interpretation difficult. The process-oriented basis representation (POBREP) analysis uses process knowledge to decompose quality data into cause-associated components. In POBREP, each potential production problem is associated with one basis element. This research investigates the thesis that POBREP can provide an effective tool for process adjustment. It has been shown previously how POBREP can be used for process monitoring purposes. For process adjustment, the following questions, among others, will be investigated: (1) What are the appropriate statistical models for adjustment that incorporate POBREP knowledge? (2) When is POBREP likely to provide a performance advantage? and (3) Can POBREP be applied effectively to a wafer fabrication process? There are several benefits associated with this research. A monitoring and adjustment strategy based on anticipated problems and disturbances can transform the ineffective performance of an omnibus control strategy. The work includes collaboration between researchers at Arizona State, Intel, and Penn State. The collaboration includes eight-month internships at Intel for Penn State and Arizona State graduate research assistants, visits to Intel by Penn State faculty, and regular visits to Intel by Arizona State faculty. There are extensive infrastructure benefits related to this collaborative approach: (1) synergistic benefits of coordinated research from four previously separately sponsored NSF researchers; (2) GOALI benefits, including engineers in the classroom, faculty visits to industry, etc.; (3) the opportunity to leverage results using existing laboratory equipment; and (4) to enhance existing courses in Applied Statistics at Penn State and Arizona State.