Analysis of Correlated Functional Variables for Manufacturing Process Diagnosis

功能过程变量(FPV)在决定各种制造过程的性能方面起着重要作用。FPV的一个例子是在化学机械抛光(CMP)过程中晶片上的机械压力分布。例如，分析FPV之间的相关性将有助于更好地理解化学机械抛光过程中复杂的晶片-焊盘-浆料相互作用。这种更好的理解可能会影响20%的晶片产量，并影响单个晶片制造厂28亿美元的收入来源。因此，该项目的目标是开发一种新的方法来分析相关的FPV，以便实现对复杂制造过程的有效监测和诊断。这项研究将首先对相关FPV的复杂时间和空间变化进行建模。对于时间变化，将每个fpv分解为幅度和相位分量，以区分时间相关性和幅度相关性。将执行FPV的全局-局部分解，以区分全局和局部变化。对于空间变化，将用一个非线性动力学模型来描述FPV之间的时间相关性。将发展一种非线性主成分方法来模拟FPV之间的幅度(幅度/全局/局部)相关性。在FPV模型的基础上，将开发统计程序来检测和诊断相关FPV的变化。通过研究非线性动力学模型中的系数，可以诊断FPV中时间相关性的变化。应用主曲线回归模型对震级相关性的变化进行诊断。建议的方法将使用美国旧金山的化学机械抛光测试仪进行验证。与行业合作伙伴的合作将促进半导体制造领域的科学和技术发现的广泛传播。它还将帮助广泛的行业实现更好的过程控制和持续的差异减少。拟议项目的成功将通过开发跨学科的课程材料、建立一个化学和通信技术实验台和基于网络的虚拟实验室以及密切的产学研合作来促进教与学。通过向研究生/本科生/K-12学生提供跨学科培训和分析微/纳米级材料去除过程中的FPV的新方法，这提供了培训过程控制和质量改进方面新的高技能劳动力的机会。女性/少数族裔学生将通过Bridges被招募到NSF和USF工程学院的博士学位、斯隆奖学金和REU(本科生研究经验)补充课程。

Functional process variables (FPVs) play a significant role in determining the performance of various manufacturing processes. An example of FPVs is the distribution of mechanical pressure on a wafer during chemical-mechanical polishing (CMP). The analysis of the correlation among FPVs will provide, e.g., a better understanding of the complex wafer-pad-slurry interactions in the CMP process. The improved understanding could affect 20% of wafer yield and impact a revenue stream of $2.8 billion for a single wafer fab. Therefore, the objective of the project is to develop a novel methodology for the analysis of correlated FPVs in order to achieve effective monitoring and diagnosis of complex manufacturing processes. The research will first model the complex temporal and spatial variations in correlated FPVs. For temporal variations, each FPV will be decomposed into amplitude and phase components for distinguishing the timing correlation and magnitude correlation. Global-local decomposition of FPVs will be performed to discriminate global and local variations. As to the spatial variations, a nonlinear dynamics model will be used to depict the timing correlation among FPVs. A nonlinear principal component method will be developed to model the magnitude (amplitude/global/local) correlation among FPVs. Based on the FPVs modeling, statistical procedures will be developed to detect and diagnose variations in correlated FPVs. The change of timing correlation in FPVs can be diagnosed through investigating the coefficients in the nonlinear dynamics model. The change of magnitude correlation in FPVs is to be diagnosed using the principal curve regression model. The proposed methodology will be validated using the CMP tester at USF.The collaboration with industry partners will facilitate a broad dissemination of scientific and technological discoveries in semiconductor manufacturing. It will also assist a broad array of industry to achieve better process control and continuous variation reduction. The success of the proposed project will promote teaching and learning through the development of interdisciplinary curricular materials, establishment of a CMP testbed and web-based virtual lab, and a close Industry-University collaboration. This provides opportunities to train a new highly skilled workforce in process control and quality improvement by exposing graduate/undergraduate/K-12 students to interdisciplinary training and novel methods of analyzing FPVs in micro/nano scale material removal processes. Women/minority students will be recruited through Bridges to Doctorate, Sloan fellowships, and REU (Research Experience for Undergraduate) supplements from NSF and USF College of Engineering.