Researchers develop wave-based fingerprint identification system using metasurface neural networks

Jul 31, 2026

Researchers developed an all-wave computational ultrasonic fingerprint identification system using a metasurface-driven loop-diffractive neural network. While experimental data and metasurface design files are publicly accessible via Zenodo, and benchmark evaluation codes are hosted on GitHub, raw fingerprint images remain restricted. These biometric images are withheld to protect potentially identifiable personal data under an approved Institutional Review Board protocol.

Wave-based fingerprint identification

▪Researchers developed an all-wave computational ultrasonic fingerprint identification system utilizing a metasurface-driven loop-diffractive neural network.
▪The fingerprint identification system incorporates an optimized diffractive metasurface designed using a trained loop-diffractive neural network model.

Metasurface neural networks

▪The loop-diffractive neural network model was benchmarked against a one-layer universal-diffractive neural network model using MNIST and Fashion-MNIST datasets.
▪The loop-diffractive neural network model was used to design an optimized diffractive metasurface for the fingerprint identification system.

Research data availability

▪Experimental and figure data for the loop-diffractive neural network study are hosted publicly on Zenodo under the digital object identifier 10.5281/zenodo.18859686.
▪The Zenodo repository for the study contains a spreadsheet file with processed and binarized fingerprint data used in the quantitative analyses.

Code repository access

▪The GitHub repository includes training scripts, evaluation routines, and model configurations to reproduce benchmark results on MNIST and Fashion-MNIST datasets.
▪The benchmark implementation code for evaluating the one-layer universal-diffractive neural network and loop-diffractive neural network models is hosted on GitHub.

Biometric privacy restrictions

▪Raw fingerprint images from the study are withheld from public availability because they contain potentially identifiable biometric information.
▪The raw fingerprint data are subject to institutional ethical restrictions under an approved Institutional Review Board protocol.

1 source

Nature

All-wave computational ultrasonic fingerprint identification with metasurface-driven loop-diffractive neural network