Ultra-sensitive 2D Dion−Jacobson perovskite phototransistors for optoelectronic neuromorphic computing

Metal halide perovskite semiconductors demonstrate significant advantages over traditional semiconductor materials in next-generation weak-light-responsive devices, owing to their exceptional photoelectric conversion capabilities and outstanding charge transport properties. Under illumination with a power density as low as 0.1 μW/cm2, these devices can efficiently respond to faint optical signals, achieving an ultrahigh detectivity of 1017 Jones—highlighting their exceptional sensitivity in low weak-light conditions. Furthermore, driven by intrinsic charge trap/de-trap dynamics, the devices successfully emulate fundamental biological synaptic behaviors, including optical pulse-modulated transitions from short-term to long-term memory. Leveraging these tunable synaptic characteristics, we further demonstrate complex visual learning simulations. Specifically, the system successfully modeled a facial recognition training process, where increasing optical stimulation progressively enhanced facial feature extraction, closely mimicking human visual cognition. By exploiting the adjustable conductance of the improved synaptic device, we construct convolutional neural networks to execute synaptic weight modifications. After 100 training epochs, the proposed system achieves a remarkable recognition accuracy of 98.91% on handwritten digit classification tasks. Collectively, these results systematically underscore the immense potential of BDASnI4 FETs as highly stable, multifunctional hardware platforms for next-generation photoelectrically integrated artificial visual systems.