Data analysis on the three defect wavelengths of a MoS2-based defective photonic crystal using machine learning

• Published in: Scientific reports Vol. 13; no. 1; pp. 21699 - 13
• Main Authors: Ansari, Narges, Sohrabi, Atieh, Mirbaghestan, Kimia, Hashemi, Mahdieh
• Format: Journal Article
• Language: English
• Published: London Nature Publishing Group UK 07.12.2023; Nature Publishing Group; Nature Portfolio
• Subjects: 639/624/399; 639/925/357; 639/925/357/1018; Absorption; Asymmetry; Bandwidths; Data analysis; Defects; Design; Graphene; Humanities and Social Sciences; Learning algorithms; Light; Machine learning; Molybdenum; Molybdenum disulfide; multidisciplinary; Optical properties; Physics; Regression analysis; Science; Science (multidisciplinary); Symmetry; Wavelength; Wavelengths
• ISSN: 2045-2322; 2045-2322
• DOI: 10.1038/s41598-023-49013-4

To reduce the dimension of optoelectronic devices, recently, Molybdenum disulfide (MoS 2 ) monolayers with direct bandgap in the visible range are widely used in designing a variety of photonic devices. In these applications, adjustability of the working wavelength and bandwidth with optimum absorption value plays an important role. This work proposes a symmetric defective photonic crystal with three defects containing MoS 2 monolayer to achieve triple narrowband defect modes with wavelength adjustability throughout the Photonic Band Gap (PBG) region, 560 to 680 nm. Within one of our designs remarkable FWHM approximately equal to 5 nm with absorption values higher than 90% for the first and third defect modes are achieved. The impacts of varying structural parameters on absorption value and wavelength of defect modes are investigated. Due to the multiplicity of structural parameters which results in data plurality, the optical properties of the structure are also predicted by machine learning techniques to assort the achieved data. Multiple Linear Regression (MLR) modeling is used to predict the absorption and wavelength of defect modes for four datasets based on various permutations of structural variables. The machine learning modeling results are highly accurate due to the obtained R 2 -score and cross-validation score values higher than 90%.