A Recurrent Nested Bayesian Non-parametric Model for Real Time Monitoring of Pattern Dependent Surface Topography in Chemical Mechanical Planarization (CMP) Operations

该奖项的研究目标是开发一种新方法，以有效捕获化学机械平面化（CMP）过程中多维过程状态的潜在非线性和非平稳演变，从而实现CMP中依赖于图案的表面形貌的早期缺陷检测，即碟形/侵蚀。提出的研究将：(1)建立CMP过程异常与在线传感器信号提取的特征之间的基本关系，即传感器特征与不断变化的盘蚀和侵蚀之间的映射，从而能够早期发现表面形貌相关缺陷；(2)利用具有非参数特性和数据驱动特性的递归嵌套狄利克雷过程（RNDP）模型建立了新的在线预测模型，该模型能够准确地捕捉CMP过程的非线性/非平稳性，避免了模型可能出现的过拟合和欠拟合。如果成功，这项研究将带来技术突破，可以充分利用/集成CMP工艺数据，从而实现早期缺陷检测/缓解，从而提高晶圆良率。预计该提案将对促进半导体工业过程监控和控制的技术进步产生重大贡献，从而提高CMP工艺的产品质量和工艺生产率。新的课程，REU，结合本科生和研究生指导计划，将吸引潜在的学生，特别是来自代表性不足的群体，通过让学生接触基础研究和行业实践，从事工程相关的研究和教育。研究成果的传播包括专业报告和出版物、网站开发、媒体推广和学生出版物，以及与行业伙伴的合作。

The research objective of this award is to develop a new approach to effectively capture the underlying nonlinear and nonstationary evolution of the multi-dimensional process states in Chemical Mechanical Planarization (CMP) process, to enable early defect detection of pattern dependent surface topography in CMP, i.e., dishing/erosion. The proposed research will: (1) establish the fundamental relationships that connect process abnormalities in CMP with extracted features from online sensor signals, i.e., a mapping between the sensor features with the evolving dishing and erosion, thus, to enable early detection of surface topography related defects; and (2) create a new online predictive model with a novel recurrent nested Dirichlet process (RNDP) model which has a non-parametric property and data-driven nature, and can accurately capture CMP process nonlinearity/nonstationarity and avoid the possible model over- and under-fitting.If successful, this research will result in a technological breakthrough that can fully utilize/integrate the CMP process data and thus enable early defect detection/alleviation for wafer yield improvement. It is anticipated that this proposal would generate significant contributions toward promoting the technological advances in process monitoring and control for semiconductor industry, leading to a better product (IC) quality and higher process productivity for the CMP process. The new curricula, REU, combined with undergraduate and graduate student mentoring programs, will attract potential students, especially from underrepresented groups, to engineering related research and education by exposing students to both fundamental research and industry practices. Dissemination of research outcomes includes professional presentations and publications, website development, media outreach and student publications, as well as collaboration with industry partners.