FMSG: Integrating Artificial Intelligence in Chemical Vapor Deposition for In-situ Predictive Crystal Growth Manufacturing.

在电子学中，大晶体硅被用作半导体计算机芯片和电网应用开关设备的基础。电子器件的效率取决于晶体的完美，因为它提供了更好的控制电子流而不损失。不同类型的半导体晶体，如钻石，可以胜过硅，但基本上无法使用。目前的项目建议使用人工智能对晶体生长过程中产生和收集的数据进行预测参数，而不是对无缺陷晶体的生长进行反复试验。使用人工智能将评估生长过程中产生的数据本身，晶体生长的当前状态，并预测生长结果。化学气相沉积过程中深度学习人工智能架构的开发和集成将使生长预测更加准确，并将缺陷评估添加到金刚石材料系统制造的预测中。该项目的成果将加快开发周期，降低制造工艺的成本，这将适用于广泛的电子晶体生长工艺。项目中开发的概念将整合到现有课程中，顶点项目将为学生设计，教育模块将为培训操作员开发。将开设一门关于数据收集、处理和解释的课程，使职业工人能够在工作环境中理解、适应并与人工智能增强制造机器合作。该课程将通过与工业制造联盟“自动化协会”（Automation Alley）合作，向制造业社区传播。拟建项目将设计和开发一个整体的人工智能平台，以解决传统方法在大规模晶体金刚石材料系统生长中的问题。该方法将侧重于提高图像收集的分辨率，并训练程序来解决时空数据的问题，包括：(1)棋盘伪影，(2)缺乏照片真实感，(3)在保持大帧分辨率的同时无法防止特征丢失。此外，为该项目开发的人工智能架构将合并到基于输入时间序列参数（如温度和缺陷）的帧预测解决方案中，以实现生长状态预测中最先进的精度指标。大规模无缺陷金刚石材料系统是最具挑战性的技术之一，有望彻底改变电力器件技术。该项目提出的基于更高分辨率图像和结合显微镜缺陷数据的增强预测能力，将使金刚石生长过程的过程控制成为可能，并将为制造晶体生长过程的全自动过程控制开辟道路。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。

In electronics, large crystal of silicon is used as the basis for semiconductor computer chips and switching devices for electric grid applications. The efficiency of electronic devices is dependent on the perfection of the crystals as it offers better control of electron flow without loss. Different types of semiconductor crystals, like diamond, can outperform silicon but are essentially unavailable for use. The current project proposes to use artificial intelligence on the data generated and collected during crystal growth to predict parameters instead of trial and error for growth of defect free crystals. The use artificial intelligence will assess the data generated during the growth process itself, the current state of crystal growth, and predict the growth results. Development and integration of deep learning artificial intelligence architectures in the Chemical Vapor Deposition process will make growth predictions more accurate and add defect assessment to the prediction for manufacturing of diamond material System. Outcome of the project will accelerate the development cycles and reduce costs for manufacturing processes which will be adaptable to a broad range of crystal growth processes for electronics. Concepts developed in the project will be integrated into existing courses, capstone projects will be designed for students, and education modules will be developed for training operators. A course in data collection, handling, and interpretation will be developed for vocational workers to understand, adapt, and team with artificial intelligence augmented manufacturing machines in the work environment. The course will be disseminated to manufacturing community by partnering with the Automation Alley, an industry manufacturing consortium.The proposed project will design and develop a holistic artificial intelligence platform to solve the problems of traditional approaches for growth of large-scale crystalline diamond material system. The approach will focus on increasing the resolution of image collection and training the program to resolve problems with spatio-temporal data, including: (1) checkerboard artifacts, (2) lack of photo-realism, and (3) inability to prevent feature loss, while maintaining a large frame resolution. Further, the artificial intelligence architectures developed for this project will be merged into solutions for frame prediction based on input time series parameters like temperature and defects to achieve state-of-the-art accuracy metrics in growth state prediction. The large scale and defect free diamond material system is one of the most challenging and holds the promise of revolutionizing power device technology. The enhanced predictive capabilities proposed in this project derived from higher resolution images and incorporation of microscope defect data will enable in-process control of the evolving growth process for diamond and will lead the way for fully automated process control of crystal growth processes for manufacturing.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.