硅酸盐学报

机器学习在晶体生长中的应用研究进展

PDF下载

杨明亮 ¹^,²^,³ , 王瑞仙 ¹^,²^,³ , 孙贵花 ¹^,³ , 王小飞 ¹^,³ , 窦仁勤 ¹^,³ , 何異 ¹^,²^,³ , 张庆礼 ¹^,³

硅酸盐学报 | 综合评述 2024,52(7): 2412-2424

收起

硅酸盐学报 |综合评述 2024 , 52 (7) : 2412 -2424

机器学习在晶体生长中的应用研究进展

全屏

杨明亮¹^,²^,³, 王瑞仙¹^,²^,³, 孙贵花¹^,³, 王小飞¹^,³, 窦仁勤¹^,³, 何異¹^,²^,³, 张庆礼¹^,³

作者信息

^1．中国科学院合肥物质科学研究院，安徽光学精密机械研究所光子器件与材料安徽省重点实验室，合肥 230031

^2．中国科学技术大学，合肥 230026

^3．先进激光技术安徽省实验室，合肥 230037

通讯作者:

张庆礼（1973—），男，博士，研究员。

作者简介:

杨明亮（1996—），男，博士研究生。

Research Progress on the Application of Machine Learning in Crystal Growth

Mingliang YANG¹^,²^,³, Ruixian WANG¹^,²^,³, Guihua SUN¹^,³, Xiaofei WANG¹^,³, Renqin DOU¹^,³, Yi HE¹^,²^,³, Qingli ZHANG¹^,³

Affiliations

^1.The Key Laboratory of Photonic Devices and Materials, Anhui Province, Anhui Institute of Optics and Fine Mechanics, Hefei institutes of Physical Science, Chinese Academy of Sciences, Hefei 230031, China

^2.University of Science and Technology of China, Hefei 230026, China

^3.Advanced Laser Technology Laboratory of Anhui Province, Hefei 230037, China

Corresponding Author:

ZHANG Qingli (1973-), male, Ph.D., Professor. E-mail: zql@aiofm.ac.cn

YANG Mingliang (1996-), female, Doctoral candidate. E-mail: yangmingliang@mail.ustc.edu.cn

出版时间: 2024-07-20 doi: 10.14062/j.issn.0454-5648.20230621

文章导航

摘要

收起

制备大尺寸优质晶体是材料领域的重要课题，当前晶体生长仍十分依赖于技术人员的生长经验，耗时长，成本高，导致研制进度十分缓慢。随着计算机科学的发展，机器学习有望加快大尺寸晶体生长工艺的探索速度。本文概述了机器学习在晶体生长中的应用，重点介绍了进化算法和神经网络算法，并概括了国内外相关领域的研究成果。在晶体结构预测方面，已成功开发出晶体结构预测软件，如Crystal Structure Prediction Network（CRYSPNet）、Universal Structure Predictor: Evolutionary Xtallography（USPEX）和Crystal structure AnaLYsis by Particle Swarm Optimization（CALYPSO）。在晶体生长条件优化方面，机器学习已成功用于铸锭炉热场优化、SiC、YAG、Si单晶温场以及生长工艺参数优化，取得了显著的效果。在晶体生长控制方法方面，机器学习已成功改进了激光晶体、Si单晶的直径控制，并在Si单晶晶体质量控制以及模型的自适应控制方法方面也取得了一定的进展。机器学习具有高精度、高效率的特点，可避开复杂的物理化学过程，通过从大量数据中挖掘信息和规律，实现晶体结构预测、晶体生长条件优化以及晶体生长精准控制。但机器学习在晶体生长中的应用也面临诸多挑战，包括大尺寸多元复杂体系的晶体结构预测、模型物理解释和数据处理、晶体生长过程的物理机制和控制规律以及多因素的耦合和复杂系统的模拟等。

关键词

计算机科学 / 机器学习 / 晶体生长 / 结构预测

Abstract

收起

With recent development on various functional crystal materials and large-size crystal preparation technologies, the reliance on experience and technological bottlenecks in traditional crystal growth becomes increasingly prominent. To meet the growing theoretical and technical demands, some methods and technologies for crystal growth are developed. Also, the development of computer science and machine learning technologies provides some opportunities in this field. Machine learning, through the analysis of vast amounts of data, can automatically extract the underlying knowledge and patterns, thereby enabling a prediction of crystal structures and an optimization/control of the crystal growth process.

Crystal structure prediction involves determining the microstructure of a crystal under given chemical compositions. Traditionally, the structure of crystals can be just determined through the related experiments. However, the experimental methods are time-consuming and costly processes. In contrast, machine learning methods can learn from the crystal structural data and predict the structure of crystals without the experiments. A series of crystal structure prediction softwares named CRYSPNet, USPEX, and CALYPSO are developed.

Conventional methods of optimizing growth conditions usually require extensive experience and experimental trial-and-error. Machine learning, via analyzing the existing data on crystal growth, can predict the optimal parameter combinations, guide the selection of parameters in actual production and accelerate the optimization process. Compared to conventional experiments and CFD simulations, machine learning offers faster and more accurate predictions. For instance, model construction based on neural networks is approximately 10⁷ times faster than the CFD simulation. The application of such novel technologies in crystal growth can promote the research and development in the related fields.

The crystal growth process is complex and variable, involving the interaction of multiple factors. Conventional control methods largely rely on experience and actual operations. However, with the application and development of machine learning and automatic control methods, crystal growth control is no longer limited to subjective and empirical judgments, but can leverage the capabilities of computer algorithms and data analysis to achieve more accurate and precise control of crystal growth. Machine learning via utilizing large datasets to train and optimize machine learning models can predict and determine the dynamic changes in crystal growth, and timely adjust and control, thereby achieving an effective control over the crystal growth process.

Summary and prospects

This review provided a brief summary of the research progress in the application of machine learning to crystal growth, and discussed mainly three aspects, i.e., crystal structure prediction, optimization of crystal growth conditions, and methods of controlling crystal growth. In terms of crystal structure prediction, machine learning achieves significant results, but the accuracy and computational efficiency still need to be improved, especially for large-size, multi-component complex systems. In the future, it is necessary to further improve and develop algorithms to increase the accuracy and computational efficiency of prediction models. Regarding the optimization of crystal growth conditions, machine learning methods can predict the optimum parameter combinations, speeding up the optimization process. However, there are still some issues with data accuracy, process complexity, and model interpretability that need to be addressed. Future work should involve the integration of more machine learning algorithms, combined with theoretical and practical research, to develop reliable and interpretable models. In terms of methods for crystal growth control, machine learning algorithms can precisely control the crystal growth process, improving the stability and quality of crystal growth. However, there are still some challenges such as in-depth studies on the physical mechanisms of crystal growth, the laws of control, and the simulation of complex systems with multiple coupled factors. In the future, it is necessary to combine advanced machine learning algorithms and optimization methods to enhance the simulation of multi-factor coupling and complex systems, further improving the control capability of crystal growth.

Key words

computer science / machine learning / crystal growth / structure prediction

引用本文

杨明亮, 王瑞仙, 孙贵花, 王小飞, 窦仁勤, 何異, 张庆礼. 机器学习在晶体生长中的应用研究进展. 硅酸盐学报, 2024 , 52 (7) : 2412 -2424 . DOI: 10.14062/j.issn.0454-5648.20230621

Mingliang YANG, Ruixian WANG, Guihua SUN, Xiaofei WANG, Renqin DOU, Yi HE, Qingli ZHANG. Research Progress on the Application of Machine Learning in Crystal Growth[J]. Journal of the Chinese Ceramic Society, 2024 , 52 (7) : 2412 -2424 . DOI: 10.14062/j.issn.0454-5648.20230621

基金

收起

国家重点研发计划(2022YFB3605700)
国家自然科学基金(52272011; 11875248)
中国科学院青年创新促进会(2023463)
安徽省实验室重点基金(AHL20220ZR04)

参考文献

收起

文献

收起

参考文献引证文献

排序方式：

2024年第52卷第7期

PDF下载

引用本文

BibTeX

文章信息

doi: 10.14062/j.issn.0454-5648.20230621

接收时间：2023-08-21
首发时间：2026-10-22
出版时间：2024-07-20

补充材料

相关文章

文章信息

作者

出版历史

收稿日期：2023-08-21
修回日期：2023-09-13

基金

国家重点研发计划(2022YFB3605700)

国家自然科学基金(52272011; 11875248)

中国科学院青年创新促进会(2023463)

安徽省实验室重点基金(AHL20220ZR04)

作者信息

^1．中国科学院合肥物质科学研究院，安徽光学精密机械研究所光子器件与材料安徽省重点实验室，合肥 230031

^2．中国科学技术大学，合肥 230026

^3．先进激光技术安徽省实验室，合肥 230037

通讯作者:

张庆礼（1973—），男，博士，研究员。

参考文献

分享链接

https://castjournals.cast.org.cn/joweb/gsyxb/CN/10.14062/j.issn.0454-5648.20230621

分享至

全文二维码

扫描看全文

引用本文

BibTeX

本文的引用情况

2种不同金属材料的力学参数

科 Family	属数 Number of genus	种数 Number of species	占总种数比例 Percentage of total species (%)	属 Genus	种数 Number of species	占总种数比例 Percentage of total species (%)
鹅膏菌科Amanitaceae	2	11	5.26	鹅膏菌属 Amanita	10	4.78
小菇科 Mycenaceae	2	12	5.74	丝盖伞属 Inocybe	5	2.39
多孔菌科 Polyporaceae	8	14	6.70	蜡蘑属 Laccaria	5	2.39
红菇科 Russulaceae	3	23	11.00	小皮伞属 Marasmius	6	2.87
				小菇属 Mycena	11	5.26
				光柄菇属 Pluteus	5	2.39
				红菇属 Russula	17	8.13
				栓菌属 Trametes	5	2.39

关闭全屏

BibTeX
EndNote
RefWorks
TxT