EAGER: Transforming Additive Nanomanufacturing with Machine Learning

光电衬底在各种功能器件中构成关键组件，其中衬底的功能是允许光通过，同时保护器件免受周围环境的影响。应用包括显示器、太阳能电池、智能手机、平板电脑、发光二极管（led），以及这些光电设备的新兴柔性版本，如射频识别（RFID）标签、人造皮肤和电子纸。高性能光电器件的可用性极大地影响了可穿戴设备、物联网等技术，从而为国家的经济和安全做出了贡献。目前，刚性玻璃基板通常与抗反射层一起使用。然而，这些涂层不能提供大范围波长或角度的抗反射，并且缺乏其他所需的多功能。这项探索性研究（EAGER）项目的早期概念资助，支持为将机器学习方法应用于纳米制造过程创建框架的研究。具体目标是利用机器学习和优化方法在表面上设计和构建纳米光子结构，以实现不同的光学特性，如防雾和抗菌。增材纳米制造是一种创建具有纳米级特征的复杂三维结构的通用方法。由于许多表面工程设计和功能是可能的，因此需要机器学习方法。该项目研究纳米制造方法，包括半导体器件制造行业中常用的无掩膜和可扩展的蚀刻和沉积工艺。这项研究活动是高度多学科的，体现了工业工程和材料工程在未来纳米制造研究和培训未来劳动力方面的独特作用。该项目创建了一个框架，将纳米制造方法（如反应离子蚀刻）与机器学习和优化工具、物理模拟和多功能表征相结合，以展示具有高性能光子管理特性（如抗反射和雾霾管理）的耐用和灵活的纳米结构光电衬底。最近在各种刚性和新兴柔性光电器件的多功能纳米结构涂层表面工程方面的大部分工作都涉及传统的试错方法，这些方法最多只能提供对参数空间小区域的碎片化和有限的系统研究，而没有任何嵌入式历史知识。主要的挑战存在于证明制造过程的可扩展性。本研究旨在验证与当前增量方法相比，机器学习和优化框架可用于更快地设计和工程光电基板的假设。集成了机器学习方法来拟合实验数据，预测新结构的性能，并为其他实验提供启发式。目前的限制是通过创建新的机器学习方法来克服的，这种方法可以简洁地学习纳米结构-添加剂纳米制造-属性关系，并具有跨领域推广的能力。开发了机器学习模型，以确定如何使用添加剂纳米制造在玻璃和塑料上制造3D纳米结构表面。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Optoelectronic substrates form a critical component in a variety of functional devices where the substrate functions to allow light to pass through and, at the same time, protect the device from the ambient environment. Applications include displays, solar cells, smart phones, tablets, light emitting diodes (LEDs), as well as emerging flexible versions of these optoelectronic devices such as radio frequency identification (RFID) tags, artificial skin, and e-paper. The availability of high performance, optoelectronic devices greatly impacts technologies such as wearables, the Internet of Things, and more, thus contributing to the nation's economy and security. Currently, rigid glass substrates are typically used with an antireflection layer. However, these coatings do not provide for antireflection across a wide range of wavelengths or angles and lack other desired multi-functionality. This EArly-concept Grants for Exploratory Research (EAGER) program award supports research to create a framework for applying machine learning methods to nanomanufacturing processes. A specific goal is to utilize the machine learning and optimization approach to design and construct nanophotonic structures on surfaces to achieve different optical properties such as anti-fogging and anti-bacterial. Additive nanomanufacturing is a versatile method to create complex 3D structures with nano-scale features. Since many surface engineering designs and functions are possible, a machine learning approach is needed. The project studies nanomanufacturing approaches involving maskless and scalable etching and deposition processes that are commonly used in the semiconductor device fabrication industry. This research activity is highly multidisciplinary and exemplifies the unique role that industrial engineering and materials engineering play in the future of nanomanufacturing research and in training the future workforce.The project creates a framework that integrates nanomanufacturing methods, such as reactive ion etching, with machine learning and optimization tools, physical simulations, and multi-functional characterizations to demonstrate durable and flexible nanostructured optoelectronic substrates with high performance photon management properties, such as antireflection and haze management. Most of the recent work in surface engineering of multi-functional nanostructure coatings for a wide variety of rigid and emerging flexible optoelectronic devices has involved traditional trial-and-error approaches that offer, at best, fragmented and limited systematic studies of small regions of the parameter space absent any embedded historical knowledge. Major challenges exist in demonstrating the scalability of manufacturing processes. This research seeks to test the hypothesis that a machine learning and optimization framework can be utilized to more rapidly design and engineer optoelectronic substrates compared to current incremental approaches. Machine learning methods are integrated to fit experimental data, predict the performance of new structures, and provide heuristics for additional experiments. Current limitations are overcome through the creation of new machine learning methods that succinctly learn nanostructure-additive NanoManufacturing-property relationships with the ability to generalize across domains. Machine learning models are developed to determine how to manufacture 3D nanostructured surfaces on glass and plastics using additive nanomanufacturing.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.