Overcoming Limits in Nano-Optical Simulations, Design and Experiments Using Deep Learning

Peter Wiecha (1) , Guilhem Larrieu (1) , Aurélie Lecestre (2) , Otto Muskens (3)

1 LAAS-MPN - Équipe Matériaux et Procédés pour la Nanoélectronique
2 LAAS-TEAM - Service Techniques et Équipements Appliqués à la Microélectronique
3 University of Southampton

Subwavelength small particles can be tailored to fulfill manifold functionalities when interacting with light. During the past twenty years, tremendous research efforts have therefore been put into the field of nano-optics, leading to astonishing results and applications like flat optics, optical cloaking or negative index meta-materials. However, there are physical and/or methodological constraints which have proven hard to overcome. For instance, the optical diffraction limit is a difficult obstacle in many applications ranging from microscopy to optical information storage. In nanooptics modeling, inverse problems like the rational design of nano-structures are another example for a difficult tasks. We show how problems that were until recently considered very hard to solve, can be tackled efficiently using methods of artificial intelligence (AI) and specifically deep learning.

Domaines: Physique Numérique [physics.comp-ph]; Optique [physics.optics]; Matière Condensée [cond-mat]; Systèmes mésoscopiques et effet Hall quantique [cond-mat.mes-hall]

Peter Wiecha, Guilhem Larrieu, Aurélie Lecestre, Otto Muskens. Overcoming Limits in Nano-Optical Simulations, Design and Experiments Using Deep Learning. META, Jul 2021, Warsaw, Poland. ⟨hal-03611990⟩