TST, Vol. 17, No. 4, PP. 47-60
(Invited paper) Design and optimization of terahertz filter devices based on deep learning
Zhang Lin-Feng 1, Ju Xue-Wei 1*, 2, 3, Zhu Guo-Feng 1, Chen Yan-Qing 1, Chen Ying 1*, 4, and Wang Xiang-Feng 1*
1 Fujian Provincial Key Laboratory of Terahertz Functional Devices and Intelligent Sensing, School of Mechanical Engineering and Automation, Fuzhou University, Fuzhou 350108, China
2 The Engineering Research Center for CAD/CAM of Fujian Universities, Putian University, Putian 351100, China
3 Engineering Research Center of Smart Distribution Grid Equipment, Fuzhou 350108, China
4 College of Physics and Information Engineering, Fuzhou University, Fuzhou 350108, China
*1 Email: juxw@fzu.edu.cn
*1 Email: chenying26@fzu.edu.cn
*1 Email: xfwang@fzu.edu.cn
(Received December 2024)
Abstract: In order to meet the demand for efficient design of terahertz functional devices, this paper starts from the target function of the device and quickly and accurately designs the desired filter structure. The design of conventional metasurface electromagnetic filters is relatively cumbersome, and it is difficult for manpower alone to complete a large amount of data analysis. The whole process wastes time and consumes computing resources. How to quickly and accurately design and optimize metasurface electromagnetic filters has become a major problem in the current field of metasurface research. Although machine learning is currently widely studied in the field of metasurfaces, there are few studies on metasurface electromagnetic filters using machine learning. Since electromagnetic filters have strong practical value and in order to avoid the shortcomings of conventional design methods, this paper uses deep learning to study metasurface electromagnetic filters. In addition, a forward spectrum prediction network and a reverse structure prediction network are designed using convolutional neural networks. The prediction results show that deep learning can well learn the physical relationship between the spectrum and structure of terahertz filters, which will greatly reduce the design time of researchers.
Keywords: Reverse design, Terahertz metasurface, Deep learning, Neural network
doi:
View Full Text: PDF
