We report an unprecedented lithography-free paradigm for an integrated photonic processor, which establishes remarkable field programmability and functionality from a global perspective, enabling accurate multiply-accumulate operations and in-situ training of optical neural networks.

We discuss recent improvements to photonics simulation tools that significantly improve the efficiency of the design process and the performance of large-scale photonic integrated circuits.

AI-based Inverse design, beyond high level calculations, has emerged as an efficient tool in material informatics to accelerate the material structure design for a given property. In this talk, we will present our latest AI-based invers design results on the field of quantum simulation or quantum optical system and the optimal optical design toward realizing Fabry–Pérot interference effect in quantum dot-embedded perovskite solar cell

This talk will introduce two of our recent works: Inverse design in quantum nanophotonics, combining local-density-of-states and deep learning [1], and Inverse design of diffusion–absorption hybrid metasurfaces by highlighting multi-objective optimization [2].

We develop a scalable photonic neural network utilizing the discrete frequency degree of freedom of light with the ability to train itself based on an in-situ backpropagation method with minimal reliance on external computers.

For photonic systems operating in a frequency band, on-the-fly predictions of near-field responses are instrumental to accelerating the dispersion engineering. We propose a neural operator architecture, dubbed HyperNO. The key is to augment an implicit neural operator with a hypernetwork so that the frequency dependency of fields can be captured through the single network. The architecture is validated for optical metasurfaces made of germanium (Ge) exhibiting significant dispersion. The numerical experiment demonstrates that the HyperNO improves both interpolation and extrapolation with respect to frequency.