Ultrafast Light‐Modulated Sliding Ferroelectric Tunnel Junctions for Synaptic in In‐Memory Computing

Abstract Developing neuromorphic synaptic devices that simultaneously offer polymorphic conductance modulation, ultrafast switching, and low power consumption remains a critical challenge for efficient brain‐inspired computing. Here, an innovative optically controlled synaptic memristor is proposed, in which the memristive layer is based on a bilayer sliding ferroelectric semiconductor—boron arsenide (BAs). Interlayer sliding, triggered by femtosecond laser pulses, enables rapid and reversible polarization switching. First‐principles and time‐dependent simulations reveal polarization reversal completed within 417.4 femtoseconds (fs), highlighting an ultrafast response that exceeds conventional gate‐controlled switching speeds. Tuning the optical pulse parameters allows precise modulation of the polarization‐induced interface barrier, thereby enabling reversible switching. The device achieves two stable conductance states with high/low conductance (ON/OFF) ratio up to 10 6 and exhibits robust long‐term non‐volatile retention. Moreover, continuously programmable multi‐conductance states can be achieved during the switching process, supporting synaptic weight modulation and nonlinear response modeling. Integration into a Residual Neural Network‐18 (ResNet‐18) neural network yields 94.7% online learning accuracy on the Fashion‐MNIST (FMNIST) dataset, closely matching the performance of full‐precision models while maintaining robustness against noise and conductance drift. These results establish a material‐to‐device framework for high‐speed, low‐power optically modulated synaptic elements, paving the way toward scalable neuromorphic computing systems with ultrafast learning capabilities.