EAGER Real-D: Real-time Data-Based Modeling and Control of Plasma-Enhanced Atomic Layer Deposition

CBET-1836518PI：Christofides，Panagiotis D.下一代电子设备需要使用改进的材料和非常精确的材料处理技术。为了缩小特征尺寸和提高能效，这些设备采用了极薄的层、高纵横比、原子锐化的接口或其任意组合。由于应用实时在线监测和控制薄膜性能的固有困难，工厂操作员通常依赖批量薄膜沉积和蚀刻循环，然后对沉积的薄膜进行扫描电子显微镜(SEM)或X射线光电子能谱(XPS)表征，以确定可控反应器参数对结果产品的影响。这种经验性的方法降低了生产率，并且不能提供薄膜加工中常见的室式反应器的行为和操作的完整数据。多尺度计算流体动力学(CFD)建模通过减少经验主义并允许开发可用于实时优化和控制反应堆运行条件的完整数据集，为解决这些问题提供了一种手段。这项研究是探索性的，致力于开发一个多尺度的CFD建模和控制框架，该框架可以通过等离子体增强原子层沉积(PE-ALD)来控制薄膜的制造，以实时优化所沉积的薄膜的微结构。CFD模型已经被证明能够捕捉到等离子体带电反应器中存在的复杂反应和传输现象，而微观模型，通常基于动力学蒙特卡罗(KMC)算法，已经成功地再现了沉积薄膜的表面特征。一个涵盖这两个领域的多尺度CFD模型将代表着在理解通过PE-ALD进行薄膜加工方面向前迈出的重要一步，并可以改进对室式反应堆操作的实时在线监测和控制。然而，这样的模型不适合开发实时优化器和基于模型的控制器，因为CFD模拟通常需要计算，并且不能与在线模型预测控制方案相联系。尽管如此，所提出的多尺度CFD模型可以作为一种无风险和有效的工具来研究PE-ALD反应器以前未探索的运行条件，并创建一个数据库，该数据库可用于推导出用于PE-ALD实时控制的计算高效的数据驱动模型。即将开发的多尺度CFD模型将允许应用一种新的、计算效率高的基于数据的贝叶斯人工神经网络(ANN)。此外，使用反应堆模型开发的数据驱动模型将形成实时过程优化和控制的基础。基于数据的模型将用于开发PE-ALD的实时操作决策策略，以减少薄膜沉积时间，这是采用该技术的必要步骤。所提出的方法可以为下一代沉积系统的实时优化和控制奠定基础，并可适用于广泛的工业过程。研究成果的传播将包括通过网络访问数据库和成果储存库。除了培养博士生，PI还计划将研究成果整合到课程中，将CFD建模及其与控制的集成纳入PI为加州大学洛杉矶分校研究生和本科生提供的高级过程控制课程，并将CFD概念和工具整合到本科数值方法、过程设计和过程控制核心课程中。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

CBET-1836518PI: Christofides, Panagiotis D. Next generation electronic devices require the use of improved materials and very precise material processing techniques. To reduce feature sizes and improve energy efficiency, the devices employ extremely thin layers, high aspect ratios, atomically-sharp interfaces, or any combination thereof. Due to inherent difficulties in applying real-time in situ monitoring and control of film properties, factory operators typically rely on batch thin film deposition and etching cycles, followed by scanning electron microscopy (SEM) or x-ray photoelectron spectroscopy (XPS) characterization of the deposited thin films to determine the effect of controllable reactor parameters on the resulting product. This empirical approach reduces productivity and fails to provide complete data on the behavior and operation of chambered reactors common to thin-film processing. Multiscale computational fluid dynamics (CFD) modeling provides a means for addressing these concerns by reducing empiricism and allowing the development of complete data sets that can be used to optimize and control reactor operating conditions in real time. The proposed research is exploratory in nature and focuses on developing a multiscale CFD modeling and control framework that can enable control of thin film manufacturing via plasma-enhanced atomic layer deposition (PE-ALD) to optimize in real time the microstructure of the deposited thin films. CFD models have been shown to capture the complex reaction and transport phenomena present within plasma charged reactors, while microscopic models, typically based on kinetic Monte Carlo (kMC) algorithms, have successfully reproduced the surface features of deposited films. A multiscale CFD model encompassing both domains would represent a significant step forward in understanding of thin-film processing via PE-ALD and could allow for improved real-time online monitoring and control of chambered reactor operations. However, such a model will be unsuitable for the development of real-time optimizers and model-based controllers because CFD simulations are generally computationally demanding and cannot be linked to online model predictive control schemes. Nonetheless, the proposed multiscale CFD model can be used as a risk-free and effective tool to investigate previously unexplored operating conditions of the PE-ALD reactor and create a database which can be utilized to derive a computationally efficient data-driven model for PE-ALD real-time control. The multiscale CFD model which will be developed will allow for the application of a novel, computationally efficient data-based Bayesian artificial neural network (ANN). Furthermore, data-driven models developed using the reactor model will form the basis for real-time process optimization and control. The data-based model will be used to develop real-time operational decision strategies for PE-ALD that reduce thin film layer deposition times, which constitutes a necessary step for adoption of this technology. The proposed methodology can form a basis for real-time optimization and control of next generation deposition systems and may be adapted to a wide range of industrial processes. Dissemination of research results will include web-based access to a database and results repository. In addition to training a PhD student, the PI plans to integrate research results into the curriculum through the inclusion of CFD modeling and its integration with control in the Advanced Process Control course that the PI offers to both graduate and undergraduate UCLA students, as well as through the integration of CFD concepts and tools into the undergraduate numerical methods, process design and process control core courses.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.