# Dual Bipolar Resistive Switching in Wafer‐Scalable 2D Perovskite Oxide Nanosheets‐Based Memristor

## TL;DR

A new memristor based on 2D nanosheets enables efficient neuromorphic computing by mimicking brain-like learning rules with high accuracy.

## Contribution

A wafer-scalable 2D memristor with dual bipolar switching enables both STDP and anti-STDP in a single device for supervised spiking neural networks.

## Key findings

- The memristor exhibits stable dual bipolar resistive switching with >10³ s retention and low set variation.
- It replicates STDP/anti-STDP rules using bidirectional plasticity under dual-polarity pulses.
- A spiking neural network model using this memristor achieves 86.4% accuracy on the MNIST dataset.

## Abstract

Memristors based on 2D materials are promising for compact and energy‐efficient neuromorphic hardware. However, conventional devices require paired elements to implement bidirectional weight updates, such as spike‐timing‐dependent plasticity (STDP) and anti‐STDP for supervised spiking neural networks (SNN) such as the remote supervised method. Here, an Au/Ti/2D Sr2Nb3O10 perovskite‐oxide nanosheet (SNO PON)/Pt memristor is demonstrated that exhibits dual bipolar resistive switching, supporting clockwise (interface) and counter‐clockwise (filament) switching. Ultrathin (≈5 nm) SNO PONs, fabricated over wafer‐scale areas by Langmuir–Blodgett deposition, serve as dynamic reservoirs for oxygen ions and vacancies. Voltage‐induced redox reactions at the Ti electrode are accompanied by the formation of oxygen vacancies in the SNO, as confirmed through cross‐sectional transmission electron microscopy and electron energy‐loss spectroscopy. The memristor exhibits stable resistance states with >103 s retention and <0.2 V set variation across 30 cells. Bidirectional plasticity under dual‐polarity pulse trains replicates STDP/anti‐STDP rules, enabling a 3 × 3 array to encode pixel patterns with opposite‐polarity pulses. A leaky integrate‐and‐fire SNN model achieves 86.4 % accuracy on the MNIST dataset using identical pre‐ and post‐synaptic spike waveforms. These findings establish dual bipolar 2D memristors as scalable and efficient components for high‐density, simplified supervised SNN hardware.

A wafer‐scalable memristor based on 2D Sr2Nb3O₁₀ perovskite oxide nanosheets exhibits dual bipolar resistive switching through controllable oxygen ion migration and redox reactions. This single device enables both STDP and anti‐STDP synaptic functions, achieving 86.4% MNIST accuracy in supervised spiking neural networks, offering a compact, energy‐efficient platform for neuromorphic computing.