26th Australian Institute of Physics Congress and the Australian and New Zealand Conference on Optics and Photonics

Inverse Design of Spontaneous Parametric Down Conversion Metasurfaces Using Adjoint Optimisation and Machine Learning Surrogates

Not scheduled

20m

Belinda Hutchinson Building (The University of Sydney )

Contributed Oral ANZOS | Photonics and Optics (ANZCOP)

Description

Metasurfaces offer unprecedented control over light-matter interactions at the nanoscale and have emerged as a promising platform for nonlinear and quantum photonic applications [1]. Particularly, metasurfaces optimised for spontaneous parametric down-conversion (SPDC) serve as compact and highly customisable sources of entangled photon pairs [2]. However, it remains challenging to identify metasurfaces that achieve desired quantum properties due to the large design space and high computational cost of numerical simulations required by numerical optimisation.

In this work, we investigate two complementary inverse design approaches for optimising SPDC metasurfaces: topology optimisation and shape optimisation. First, we build on an existing topology-optimisation gradient-based framework using adjoint methods [3] to explore broad design spaces and identify suitable metasurfaces. By efficiently computing gradients for a large number of design variables, adjoint optimisation provides high-performance structures beyond conventional intuition-driven design. Second, we develop a machine-learning-assisted shape-optimisation framework for the rapid design of parameterised metasurfaces. The differentiable surrogate framework provides a foundation for gradient-based optimisation and accelerates design-space exploration, offering a scalable approach for future inverse-design studies. So far, we have created patterns for unidirectional SPDC outperforming the thin film of the same thickness and material by 20 times in brightness while ensuring near-maximal polarisation entanglement between the signal and idler states.

These two approaches show how topology optimisation and shape optimisation can complement each other in the inverse design of quantum metasurfaces: topology optimisation to discover novel device architectures and machine-learning-assisted shape optimisation to rapidly refine designs while reducing computational cost. This work contributes to advanced computational design methods for next-generation quantum photonic devices and integrated entangled photon sources.

References

[1] A. H. Dorrah and F. Capasso, Science 376, 367 (2022).

[2] J. Ma et al., Science Advances 11, eadu4133 (2025).

[3] N. Li, J. Zhang, D. N. Neshev, and A. A. Sukhorukov, Nanophotonics, 137 (2024).