A multifunctional optoelectronic memristor based on Sb₂Se₃ nanorod arrays for high-dimensional reservoir computing

The emerging paradigm of in-sensor computing demands the seamless integration of optical sensing, memory, and localized data processing within a single monolithic architecture. However, existing integrated devices face a fundamental physical trade-off between rapid photogenerated carrier extraction for sensing and prolonged carrier retention for synaptic memory, severely limiting their overall processing efficiency. To break this bottleneck, a multifunctional optoelectronic memristor based on 1D Pd/Sb₂Se₃/TiN nanorod arrays is demonstrated. The unique 1D nanorod architecture helps to mitigate grain boundary recombination, granting the device exceptional multimodal capabilities. Operating as a pure photodetector, it exhibits an ultrafast response time of 35 μs and a high ON/OFF ratio of ~ 2×10⁴. Under optoelectronic co-modulation, the intrinsic carrier decay dynamics precisely emulate robust synaptic plasticity with an ultralow energy consumption of ~3.6 pJ per spike. Leveraging this reliable nonlinear responsiveness to mixed optoelectronic stimuli, a dual-feature extraction strategy is implemented to construct a high-dimensional reservoir computing (RC) system. This approach efficiently enhances computational dimensionality without redundant network complexity, achieving outstanding recognition accuracies of 90.4% on the MNIST dataset and 92.3% in dynamic hand posture recognition tasks. This work provides a fundamental material-to-system paradigm for developing energy-efficient artificial visual systems and advanced in-sensor computing architectures.