Synaptic plasticity based on photo-activated materials for mobile neuromorphic computing

In the era of artificial intelligence, neuromorphic computing systems that emulate the efficiency of the human brain have emerged as promising alternatives to conventional architectures. The exploration of novel materials capable of low-power, high-speed optical operation is becoming increasingly critical. Among these, stimulus-responsive photoluminescent materials have attracted interest due to optical behaviors that resemble synaptic plasticity. We present a photonic computing layer based on carbon nanomaterials exhibiting dual fluorescence-phosphorescence emission and photoactivation dynamics, mimicking short- and long-term memory as well as synaptic potentiation. The layer was engineered to convert optical inputs into characteristic visible emission time decays. To evaluate its computational capability, the MNIST digit classification task was employed. Due to the material’s brightness, the optical responses were recorded using a mobile device by sampling discrete time points from the emission dynamics. These readouts were used as inputs to an artificial neural network for offline training and inference. The system achieved a maximum training accuracy of 95.3% and a testing accuracy of 93.5%. These results demonstrate that the proposed photonic layer can perform optical computation suitable for neuromorphic information processing. Furthermore, mobile-based implementation highlights the system’s accessibility and portability, enabling low-cost deployment with consumer-grade devices.