Deep Learning‐Assisted Design of Bilayer Nanowire Gratings for High‐Performance MWIR Polarizers

• Published in: Advanced materials technologies Vol. 9; no. 19
• Main Authors: Lee, Junghyun, Oh, Junhyuk, Chi, Hyung‐gun, Lee, Minseok, Hwang, Jehwan, Jeong, Seungjin, Kang, Sang‐Woo, Jee, Haeseong, Bae, Hagyoul, Hyun, Jae‐Sang, Kim, Jun Oh, Kim, Bongjoong
• Format: Journal Article
• Language: English
• Published: 07-10-2024
• Subjects: bilayer nanowire gratings; deep learning‐assisted design; MWIR polarizers
• ISSN: 2365-709X; 2365-709X
• DOI: 10.1002/admt.202302176

Optical metamaterials have revolutionized imaging capabilities by manipulating light‐matter interactions at the nanoscale beyond the diffraction limit. Bilayer nanowire grating configurations exhibit significant potential as exceptional elements for high‐performance polarimetric imaging systems. However, conventional computational approaches for predicting electromagnetic responses are time‐consuming and labor‐intensive, and thereby, the practical implementation remains challenging through an iterative design, analysis, and fabrication process. Here, a deep learning‐based design process is presented utilizing an artificial neural network (ANN) trained on finite element method (FEM) simulations that enables the prediction of bilayer nanowire gratings‐based electromagnetic responses. The study validates predictions through nanoimprinted bilayer nanowire gratings, demonstrating the reliability of the ANN's predictions. Furthermore, the research identifies critical geometric parameters significantly influencing transverse magnetic (TM) and transverse electric (TE) transmission. The ANN model effectively tailors design for specific mid‐wavelength infrared (MWIR) wavelengths, which may provide a practical tool for rapidly designing and optimizing metamaterial for high‐performance polarizers.
Herein, a deep learning‐based design process, which presents the efficiency in predicting transmission is demonstrated by systematically varying geometric parameters. Comprehensive validations confirm the accuracy of predictions through comparisons among measurements, simulated results, and predictions from the ANN model. Notably, the study explores the ANN's efficacy in designing high‐performance MWIR polarizers, demonstrating its capability to identify resonances at target wavelengths.