Breaking the Q-limit: wide-range and high-precision metasensing empowered by deep learning

Area of Science:

• Photonics and Artificial Intelligence
• Metasurface optical sensing

Background:

• Traditional refractive index sensing relies on high-Q resonances in precisely fabricated metastructures.
• This approach is vulnerable to fabrication imperfections, limited spectral resolution, and environmental instability.

Purpose of the Study:

• To introduce a new sensing paradigm based on computationally learned latent representations.
• To demonstrate a computational-photonic hybrid approach for refractive index sensing that bypasses the need for high-Q features.

Main Methods:

• Experimental demonstration of an end-to-end variational autoencoder.
• Direct retrieval of refractive index from transmission spectra of a generic silicon metasurface.
• Autonomous learning of a compact latent manifold encoding refractive-index-dependent spectral information.

Main Results:

• Accurate refractive index retrieval without requiring high-Q features.
• Robust sensing performance under strong noise and fabrication variability.
• Successful operation under conditions beyond the training distribution.

Conclusions:

• The developed computational-photonic hybrid approach removes the traditional dependence on resonance finesse.
• This redefines metasurfaces for optical sensing, offering enhanced robustness and accuracy.
• The method enables reliable refractive index sensing even with imperfect fabrication and noisy data.