Fully Photon-Modulated Synaptic Devices with Bidirectional Plasticity
This 2026 paper by Y. Yan et al., published in Advanced Photonics, presents an artificial synapse that operates entirely with light, requiring no electrical signals at any stage.
Photon-driven synapse advances low-power neuromorphic systems Image: spie.org - Photon-driven synapse advances low-power neuromorphic systems
The device uses a rare-earth-doped long-afterglow crystal that stores optical information as trapped charge carriers. According to SPIE, the system achieves bidirectional plasticity through two mechanisms:
- UV light produces paired-pulse facilitation (excitatory response), where a second pulse yields stronger output because earlier excitation partially fills trap states.
- Near-infrared light produces paired-pulse depression (inhibitory response), as the first pulse empties trapped carriers, weakening subsequent responses.
The researchers developed a physical model tracking how carriers are generated, trapped, and released over time, explaining the history-dependent behavior without electrical control.
For practical demonstration, the team paired the crystal with a silicon imaging sensor to build a prototype neuromorphic camera. This in-sensor processing enhanced contrast and suppressed noise directly at capture. A neural network modeled on the device’s behavior achieved 95.99% accuracy on handwritten digit classification after denoising, compared to roughly 78% without it, SPIE reports.
The device currently operates on millisecond-to-second timescales, comparable to biological visual processing. The authors suggest scaling down and modifying material properties could improve both speed and energy efficiency.
The full Gold Open Access paper is available at Advanced Photonics 8(4) 046001 (doi: 10.1117/1.AP.8.4.046001). For a readable summary, the SPIE news article provides the most complete overview.
Sources: SPIE News, Advanced Photonics (SPIE Digital Library)
Fully photon-modulated synaptic devices with bidirectional plasticity for neuromorphic vision and recognition
Submitted 4 hours ago by cm0002@europe.pub to science@mander.xyz
https://www.researching.cn/articles/OJ1c8af7e2b84132f7