Deep Neural Networks for the design of photonic devices – DNN4Photonics

The urgent need for innovative solutions underscores the transformative impact of enhanced modeling in advancing large-scale nanophotonics. To address these challenges and expedite the design process, recent studies have explored the potential of Artificial Intelligence (AI), specifically Deep Learning (DL), for developing rapid surrogate models in the numerical characterization of light-matter interactions within micro-/nanostructured matter. DL, is an attractive approach for big data-driven problems. It has seen substantial success in computer vision and natural language processing. Notably, Artificial Neural Networks (ANN) are now being investigated and applied in scientific computing across diverse physical domains. In DNN4Photonics, our primary objective is to develop innovative methodologies relying on Deep Neural Networks (DNN) for the modeling and optimization of diverse metasurface configurations, with a particular emphasis on large-scale structures. Within the framework of DNN4Photonics, our research thrust revolves around exploring data-driven DL techniques to furnish rapid and dependable surrogates capable of emulating the simulation of three-dimensional spatial domains involving complex scattering objects. Moreover, in DNN4Photonics a key focus is to find an intelligent and efficient formulation for DNN-based inverse design (e.g. retrieving the structural morphology of a metasurface provided a target function) strategies tailored for large-scale photonic devices. This encompasses not only enhancing the efficiency and reliability of large-scale metasurface modeling but also strategically tailoring the designs for applications demanding high precision, such as the optimization of RGB metalens at visible frequency for achromatic, micrometer- and millimeter-sized metalenses. We identify three classes of devices that will be modeled and implemented experimentally during DNN4Photonics. Each of them entails a potential rapid exploitation of the technology by patent filing on the design and a rapid commercial roll-out thanks to the presence of the non-academic partner.