Deep Neural Networks for the design of photonic devices

• Grant number: 24/04835-4
• Support Opportunities: Regular Research Grants
• Start date: October 01, 2024
• End date: September 30, 2027
• Field of knowledge: Engineering - Electrical Engineering - Telecommunications
• Principal Investigator: Hugo Enrique Hernández Figueroa
• Grantee: Hugo Enrique Hernández Figueroa
• Principal researcher abroad: Stéphane Lanteri
• Institution abroad: Institut National de Recherche en Informatique et en Automatique (INRIA Sophia Antipolis), France
• Host Institution: Faculdade de Engenharia Elétrica e de Computação (FEEC). Universidade Estadual de Campinas (UNICAMP). Campinas , SP, Brazil
• Associated researchers: Silvia Vaz Guerra Nista ; Silvio Domingos Silva Santos

Abstract

The precision enabled by nanostructured materials in controlling light-matter interactions at the nanoscale has prompted a transformative shift in nanophotonics. This interdisciplinary field, investigating the interaction of light waves with nanostructured matter, is witnessing a surge in scientific and technological significance. Particularly, metasurface research has transcended academic boundaries, gaining momentum in mainstream applications and offering substantial contributions to pioneering product development. Metasurfaces, exemplifying the forefront of nanophotonics, face challenges, especially with large-scale photonic devices like metalenses. Designing such devices based on physics-based models demands advanced computational methods due to complexities amplified by large-scale structures and intricate material subwavelength structuring. Addressing these challenges is critical for the design of large-scale nanophotonic devices essential in various applications like laser optics, imaging, microscopy, and active beam steering.To expedite the design process and overcome computational limitations, recent studies explore the potential of Artificial Intelligence (AI), particularly Deep Learning (DL), for developing rapid surrogate models. This project, called DNN4Photonics, will be developed by the teams from INRIA (France) and UNICAMP (Brazil), let by Stéphane Lanteri and Hugo Figueroa, respectively, with the support of the French company SOLNIL. The main objective is to develop innovative methodologies based on Deep Neural Networks (DNN) to model and optimize various configurations of metasurfaces, emphasizing large-scale structures. Our research thrust focuses on exploring data-driven DL techniques to provide rapid and reliable surrogates for simulating complex three-dimensional spatial domains involving scattering objects. Moreover, we aim to devise intelligent and efficient formulations for DNN-based inverse design strategies tailored for large-scale photonic devices, enhancing modeling efficiency and reliability while tailoring designs for precision-demanding applications.Three classes of devices will be modeled and experimentally implemented during DNN4Photonics, with potential rapid commercialization facilitated by our non-academic partner, SOLNIL. These include small-scale metalenses optimized for micro-LED displays, a large-scale metalens for achromatic microscopy imaging, and a large-scale metasurface for high-power laser applications. Our focus at visible frequencies, relevant to human vision and high-power lasers for industrial processes, underscores the versatility and applicability of our proposed method-ologies in addressing real-world challenges in nanophotonics. The novel designed metasurfaces will also be fabricated and characterized. (AU)