Wafer-Scale All-Silicon Self-Rectifying Memristor for Synaptic Response and Reservoir Computing.

Abstract

Silicon p-n junctions have remained an indispensable building block of electronics since their invention in the Shockley days. Likewise, an abrupt p-n junction has served as a foundational model in semiconductor textbooks. In this work, we report on an p-n junction in silicon with an oxide interfacial layer, enabling memristive functions with highly rectifying resistive switching and reproducible synaptic behaviors for reservoir computing (RC). The device exhibited a rectification ratio of ∼5000 with stable endurance of 4.5 × 10⁶ cycles, without filament formation. Charge-trapping dynamics enable key synaptic behaviors including paired-pulse facilitation, spike-timing-dependent plasticity, and transitions between short- and long-term memory. Leveraging these behaviors, the device performs RC via 4-bit pulse stimulation, achieving 86.9% accuracy in handwritten digit classification. This interface-engineered all-silicon device bridges classical diode physics with modern neuromorphic computation, providing a prospect for wafer-scale platforms for neuromorphic applications.