Description
Photonic neural networks promise high-bandwidth, low-latency analogue computation, but most implementations rely on spatially multiplexed mesh architectures where an N-mode operation demands $O(N^2)$ individually controlled elements, placing significant demands on chip area and electrical interconnects. Electro-optically modulated ring resonators offer a compelling alternative by encoding information across discrete frequency modes within a single ring, reducing this to only $O(N)$ drive signals and dramatically shrinking the hardware footprint. Cascaded dual-ring resonators extend this capability further, enabling programmable unitary transformations across a bounded set of frequency-bin modes on a compact integrated chip.
Building on the resolvent-based scattering matrix framework of Buddhiraju et al. [1], we show that cascading several modulated rings enables the synthesis of arbitrary unitary matrices, and via singular-value embedding, arbitrary non-unitary matrices, with high fidelity. These capabilities directly realise the linear matrix-multiplication layer at the heart of a frequency-domain photonic neural network, and equally underpin frequency-bin qubit gates for linear-optical quantum information processing [2].
As a proof of principle, we present numerical simulations on a single bounded dual-ring lithium niobate on insulator (LNOI) platform [3], targeting all four single-qubit Pauli operators as post-selected frequency-bin gates. Target unitaries are found through multi-restart inverse design of the per-tone modulation amplitudes and phases. Simulations with parameters matched to the fabricated chip show target-subspace fidelities exceeding 0.99 across the full Pauli set.
We test these simulations against experiment with a fabricated LNOI chip, demonstrating a clear path towards arbitrary unitary transformations via cascaded rings established in [1]. Together, these results position modulated dual-ring LNOI resonators as a practical and scalable platform for both frequency-encoded photonic quantum logic and frequency-domain photonic neural networks.
References
[1] S. Buddhiraju et al., Nat. Commun. 12, 2401 (2021).
[2] J. M. Lukens and P. Lougovski, Optica 4, 8 (2017).
[3] H. Dinh et al., arXiv:2603.01422 (2026).
| I am the presenting author | Yes |
|---|