SKM 2023 – wissenschaftliches Programm

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

O: Fachverband Oberflächenphysik

O 43: Poster: Plasmonics and Nanooptics I

O 43.3: Poster

Dienstag, 28. März 2023, 18:00–20:00, P2/EG

Deep learning for the extraction of optical parameters of multilayer samples in scanning near-field optical microscopy — •Dario Siebenkotten, Lara Harren, Clemens Elster, and Bernd Kästner — Physikalisch-Technische Bundesanstalt, Abbestr. 2-12, 10587 Berlin

Scattering-type scanning near-field optical microscopy (s-SNOM) is a powerful method for optical material characterization at the nanoscale. However, owing to the complex interaction between tip and sample, extensive modelling is needed for the extraction of optical parameters, particularly for layered samples. The extraction of optical parameters typically requires fitting, which quickly becomes unstable [1] and can be very time intensive due to the complexity of the models. Deep Learning algorithms offer a fast alternative for the optical parameter extraction but have only been applied to bulk materials. Here, we show the extension of these approaches to systems consisting of one and two layers of polar crystals exhibiting surface phonon-polariton resonances on top of a substrate by training neural networks with model data. We present the trained neural networks and discuss the extraction accuracy for the cases of one and two layered samples. While this study is limited to polar crystals, the application to other systems, defined for example by free charge carriers or band transitions, is straight forward. [1] McArdle et al. Phys. Rev. Research 2, 023272 (2020)