• • The device achieves a recognition accuracy of 95.03% on the MNIST dataset, demonstrating high fidelity in in-sensor optical processing and classification, which is critical for deploying neuromorphic systems in real-world pattern recognition tasks.
• • Hierarchical synaptic plasticity is realized through distinct weight modulation timescales: electrical gating induces rapid, short-term changes, while UV-excited long-afterglow optical stimuli produce persistent, long-term potentiation, enabling multi-scale learning dynamics.
• • The integration of a long-afterglow material converts transient UV excitation into sustained visible emission, extending the optical stimulus duration beyond milliseconds and facilitating cascade interactions between optical and electrical pathways for enhanced plasticity tunability.
• • The dual-output architecture concurrently provides visible-light emission for direct display and electrical signal processing, achieving seamless integration of sensing, display, and computation within a single device framework, which reduces system complexity and energy overhead in neuromorphic hardware.