Collaborative Research: Machine Learning-assisted Ultrafast Physical Vapor Deposition of High Quality, Large-area Functional Thin Films

点击翻译按钮获取中文摘要

This grant supports research to produce high-quality, large-area functional thin films using a machine learning-assisted ultrafast thin film manufacturing approach. Functional thin films, such as oxides, chalcogenides and nitrides, have a wide range of applications in semiconductor, communication, and energy industries. However, conventional methods for scalable manufacturing of functional thin films are time-consuming and wasteful, relying on solvents and trial-and-error approaches. The goal of this project is to apply machine learning to overcome challenges posed by structural and chemical defects associated with conventional thin film deposition, thereby improve film quality and manufacturing efficiency. Machine learning accelerates optimization of thin film growth conditions via training the experimental and computational data and speeds up the development of thin films with desired functionality. This award supports fundamental research to enable faster and cost-effective manufacturing of high-quality and large-area functional thin films for a broad range of applications in electronics, photonics, and energy conversion. Results from this project benefit the US economy and society by addressing semiconductor manufacturing and clean energy challenges facing the nation. This research involves multiple disciplines including materials science and engineering, machine learning, and advanced manufacturing. This interdisciplinary approach increases the participation of underrepresented groups in engineering research and education. The limitations of conventional thin film deposition are lack of defect control and composition manipulation, long development time, and material waste. This project applies machine learning to ultrafast physical vapor deposition to overcome these limitations and manufacture high quality, large-area functional thin films. In physical vapor deposition, film thickness, microstructure, chemical composition, and property can be engineered by tailoring the processing parameters. Closely integrating machine learning, physical property calculations, and thin film growth conditions improves film quality, shortens development cycle and reduces material waste. This research uses machine learning algorithms, such as, linear and nonlinear regression and Bayesian optimization, to train film growth and property data generated by experiment and collected from literature. Machine learning models, in conjunction with in-situ monitoring, are used to optimize growth conditions such as substrate temperature, deposition time, partial pressure, and ramping and cooling rates and achieve the targeted electronic and optical properties at a lower cost and faster development cycle. The machine learning-assisted scalable manufacturing of functional chalcogenide thin films not only enriches the materials portfolio for solar energy conversion, but also advances their applications in electronics and photonics, such as photodetectors, phototransistors, thermoelectrics, and light emission diodes. This approach can also be applied to accelerate the development of other renewable energy materials.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.