CVD Technology Innovation Behind the Nobel Prize

Recently, the announcement of the 2024 Nobel Prize in Physics has brought unprecedented attention to the field of artificial intelligence. The research of American scientist John J. Hopfield and Canadian scientist Geoffrey E. Hinton uses machine learning tools to provide new insights into today's complex physics. This achievement not only marks an important milestone in artificial intelligence technology, but also heralds the deep integration of physics and artificial intelligence.

Ⅰ. The Significance and Challenges of Chemical Vapor Deposition (CVD) Technology in Physics

The significance of chemical vapor deposition (CVD) technology in physics is multifaceted. It is not only an important material preparation technology, but also plays a key role in promoting the development of physics research and application. CVD technology can precisely control the growth of materials at the atomic and molecular levels. As shown in Figure 1, this technology produces a variety of high-performance thin films and nanostructured materials by chemically reacting gaseous or vaporous substances on the solid surface to generate solid deposits1. This is crucial in physics for understanding and exploring the relationship between the microstructure and macroscopic properties of materials, because it allows scientists to study materials with specific structures and compositions, and then deeply understand their physical properties.

Secondly, CVD technology is a key technology for preparing various functional thin films in semiconductor devices. For example, CVD can be used to grow silicon single crystal epitaxial layers, III-V semiconductors such as gallium arsenide and II-VI semiconductor single crystal epitaxy, and deposit various doped semiconductor single crystal epitaxial films, polycrystalline silicon films, etc. These materials and structures are the basis of modern electronic devices and optoelectronic devices. In addition, CVD technology also plays an important role in physics research fields such as optical materials, superconducting materials, and magnetic materials. Through CVD technology, thin films with specific optical properties can be synthesized for use in optoelectronic devices and optical sensors.

Figure 1 CVD reaction transfer steps

At the same time, CVD technology faces some challenges in practical applications², such as:

✔ High temperature and high pressure conditions: CVD usually needs to be carried out at high temperature or high pressure, which limits the types of materials that can be used and increases energy consumption and cost.

✔ Parameter sensitivity: The CVD process is extremely sensitive to reaction conditions, and even small changes may affect the quality of the final product.

✔ CVD system is complex: The CVD process is sensitive to boundary conditions, has large uncertainties, and is difficult to control and repeat, which may lead to difficulties in material research and development.

Ⅱ. Chemical Vapor Deposition (CVD) Technology and Machine Learning

Faced with these difficulties, machine learning, as a powerful data analysis tool, has shown the potential to solve some problems in the CVD field. The following are examples of the application of machine learning in CVD technology:

(1) Predicting CVD growth

Using machine learning algorithms, we can learn from a large amount of experimental data and predict the results of CVD growth under different conditions, thereby guiding the adjustment of experimental parameters. As shown in Figure 2, the research team of Nanyang Technological University in Singapore used the classification algorithm in machine learning to guide the CVD synthesis of two-dimensional materials. By analyzing early experimental data, they successfully predicted the growth conditions of molybdenum disulfide (MoS2), significantly improving the experimental success rate and reducing the number of experiments.

Figure 2 Machine learning guides material synthesis

(a) An indispensable part of material research and development: material synthesis.

(b) Classification model helps chemical vapor deposition to synthesize two-dimensional materials (top); regression model guides hydrothermal synthesis of sulfur-nitrogen doped fluorescent quantum dots (bottom).

In another study (Figure 3), machine learning was used to analyze the growth pattern of graphene in the CVD system. The size, coverage, domain density, and aspect ratio of graphene were automatically measured and analyzed by developing a region proposal convolutional neural network (R-CNN), and then surrogate models were developed using artificial neural networks (ANN) and support vector machines (SVM) to infer the correlation between CVD process variables and the measured specifications. This approach can simulate graphene synthesis and determine the experimental conditions for synthesizing graphene with a desired morphology with large grain size and low domain density, saving a lot of time and cost² ³

Figure 3 Machine learning predicts graphene growth patterns in CVD systems

(2) Automated CVD process

Machine learning can be used to develop automated systems to monitor and adjust parameters in the CVD process in real time to achieve more precise control and higher production efficiency. As shown in Figure 4, a research team from Xidian University used deep learning to overcome the difficulty of identifying the rotation angle of CVD double-layer two-dimensional materials. They collected the color space of MoS2 prepared by CVD and applied a semantic segmentation convolutional neural network (CNN) to accurately and quickly identify the thickness of MoS2, and then trained a second CNN model to achieve accurate prediction of the rotation angle of CVD-grown double-layer TMD materials. This method not only improves the efficiency of sample identification, but also provides a new paradigm for the application of deep learning in the field of materials science⁴.

Figure 4 Deep learning methods identify the corners of double-layer two-dimensional materials

References:

(1) Guo, Q.-M.; Qin, Z.-H. Development and application of vapor deposition technology in atomic manufacturing. Acta Physica Sinica 2021, 70 (2), 028101-028101-028101-028115. DOI: 10.7498/aps.70.20201436.

(2) Yi, K.; Liu, D.; Chen, X.; Yang, J.; Wei, D.; Liu, Y.; Wei, D. Plasma-Enhanced Chemical Vapor Deposition of Two-Dimensional Materials for Applications. Accounts of Chemical Research 2021, 54 (4), 1011-1022. DOI: 10.1021/acs.accounts.0c00757.

(3) Hwang, G.; Kim, T.; Shin, J.; Shin, N.; Hwang, S. Machine learnings for CVD graphene analysis: From measurement to simulation of SEM images. Journal of Industrial and Engineering Chemistry 2021, 101, 430-444. DOI: https://doi.org/10.1016/j.jiec.2021.05.031.

(4) Hou, B.; Wu, J.; Qiu, D. Y. Unsupervised Learning of Individual Kohn-Sham States: Interpretable Representations and Consequences for Downstream Predictions of Many-Body Effects. 2024; p arXiv:2404.14601.