Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

HL: Fachverband Halbleiterphysik

HL 30: Poster 2

HL 30.8: Poster

Donnerstag, 8. September 2022, 11:00–13:00, P3

Machine Learning-Based Optimization of Chiral Photonic Nanostructures: Evolution- and Neural Network-Based DesignsOliver Mey1, •Manan Shah2, and Arash Rahimi-Iman21Fraunhofer IIS/EAS, Fraunhofer Institute for Integrated Circuits, Division Engineering of Adaptive Systems, Dresden — 2I. Physikalisches Institut und Zentrum für Materialwissenschaften, Justus-Liebig Universität Gießen, D-35392, Germany

Machine learning (ML) techniques such as deep learning (DL) and evolutionary algorithms (EA) exhibit unprecedented capabilities in the scientific ML realm. DL uses artificial neural networks to infer unintuitive solutions for complicated design requirements and specific functionalities. Likewise, the EA attempts to find the optimized solution by utilizing principles such as mutation of parameters and extinction of less promising solutions. These approaches are faster and more effective in the inference of nanostructure design parameters for desired properties, such as wavelength coverage and peculiar response functions, compared to conventional numeric simulations.

We present a nano-patterned GaP dielectric substrate that favors single-handed circularly polarized light (CPL) in reflection or transmission [1]. The optimization in chiral dichroism (CD) by neural networks is compared with the evolutionary algorithm. The increased CD in simulated spectra for designs with stronger reflectivity of right CPL and lower transmissivity of left CPL makes the ML techniques effective to optimize a myriad of properties for metamaterials and photonic nanostructures. [1] Phys. Status Solidi RRL 2022, 16, 2100571.

100% | Bildschirmansicht | English Version | Kontakt/Impressum/Datenschutz
DPG-Physik > DPG-Verhandlungen > 2022 > Regensburg