All-optical object classification using an edge-detecting spin-differential diffractive network.

Area of Science:

• Photonics and Optical Computing
• Artificial Intelligence and Machine Learning
• Metasurface Nanophotonics

Background:

• All-optical computing promises high speed and low power consumption, essential for exceeding Moore's Law limits.
• Conventional single-wavelength diffractive deep neural networks (D2NNs) struggle with simultaneous edge-feature extraction and classification.
• Optimizing optical edge-feature extraction and classification synergistically is a key challenge in all-optical computing.

Purpose of the Study:

• To propose an edge-detecting spin-differential diffractive neural network (ESD-DNN) for single-wavelength all-optical object classification.
• To achieve synergistic co-optimization of edge-feature extraction and classification in a diffractive neural network.
• To enhance classification accuracy and computational efficiency compared to existing D2NN architectures.

Main Methods:

• Implemented a Pancharatnam-Berry phase gradient metasurface for rapid edge-feature extraction.
• Utilized a spin-differential mechanism with left-/right-handed circularly polarized (LCP/RCP) components for classification inference.
• Performed end-to-end optimization of diffractive layers for co-optimization of network functions.

Main Results:

• The single-layer ESD-DNN achieved 97.5% (MNIST) and 87.5% (Fashion-MNIST) classification accuracy.
• Demonstrated a 5-fold increase in computational efficiency and 80% reduction in time complexity compared to a four-layer D2NN.
• Maintained >90% classification accuracy under environmental challenges like turbulence and thermal lensing, showing robustness.

Conclusions:

• The proposed ESD-DNN effectively integrates edge-feature extraction and classification for all-optical computing.
• This approach significantly improves performance metrics and robustness over traditional single-wavelength D2NNs.
• Paves the way for advanced applications in AI, remote sensing, industrial inspection, and space optical communications.