Linearity Engineering of Perovskite Optoelectronic Synapses via Stimulus Modulation for Neuromorphic Photonic Computing

Neuromorphic computing requires synaptic devices capable of ultrafast, low‐power, and linear weight updates—yet existing electronic memristors and optoelectronic devices cannot simultaneously provide weight retention, rapid response, and low‐energy operation within a single uniform. Here, we introduce the photo‐induced memristive perovskite diode, which merges photodetection and resistive switching into an optoelectronic element for zero‐bias optical readout. Unlike prior phototransistors that rely on separate writing and reading mechanisms, our vertically stacked architecture intrinsically forms and collapses conductive filaments under hybrid electrical/optical modulation, enabling linear photo‐induced weight updates with energy consumption down to 34 pJ and ultrafast optical response times of <29 µs. Furthermore, we uncover a previously unexplored mechanism in which long‐chain organic spacers confine halide migration pathways and regulate filament density, yielding stable long‐term potentiation/depression (LTP/LTD) over 3000 cycles with improved linearity. Incorporating the extracted synaptic metrics into artificial neural network simulations, the optimized device reaches 92.89% accuracy using Artificial Neural Network (ANN) simulations and enhanced Convolutional Neural Network (CNN) simulation performance. Our findings establish a new device paradigm—a zero‐bias optical readout, photo‐induced memristive perovskite diode with mechanism‐driven filament control—and highlight a generalizable strategy for developing fast, energy‐efficient weight updates with improved linearity, particularly in the LTP regime.