Deep-learning-based design strategy for cladding-free and crosstalk-free photonic crystal waveguide systems
Phys. Rev. Applied 25, 054051 – Published 20 May, 2026
DOI: https://doi.org/10.1103/7fxd-ntk8
Abstract
Recently, cladding-free waveguide systems without nearest-neighbor crosstalk have opened a path toward the miniaturization of photonic integrated circuits. However, there is still a prominent chance of crosstalk between next-nearest-neighbor channels, which severely degrades signal integrity and limits device performance. To overcome this challenge, we propose a deep-learning-based design strategy that further suppresses next-nearest-neighbor crosstalk below −20 dB across the operational bandwidth, thereby maintaining good signal isolation across almost all channels in the system. This approach offers a scalable solution for high-density photonic circuits and phased arrays, facilitating potential applications including optical interconnects, quantum simulations, and radar antennas.
