Sodium-Doped Titania Self-Rectifying Memristors for Crossbar Array Neuromorphic Architectures

Area of Science:

• Materials Science
• Nanotechnology
• Solid-State Electronics

Background:

• Memristors in crossbar-array architectures (CAAs) are key for artificial neural networks but suffer from low reliability and parasitic currents.
• Crosstalk and sneak currents in CAAs hinder the performance and scalability of memristor-based neuromorphic computing.

Purpose of the Study:

• To develop high-reliability memristors for selectorless CAAs.
• To investigate sodium-doped titanium dioxide (Na-doped TiO2) memristors for improved performance in neuromorphic applications.
• To demonstrate the feasibility of Na-doped TiO2 memristors in practical image-recognition tasks.

Main Methods:

• In situ growth of Na-doped TiO2 thin films using atomic layer deposition (ALD) with an aqueous NaOH reactant.
• Fabrication of memristors utilizing the Na-doped TiO2 films, characterized by electroforming-free and self-rectifying behavior.
• Integration and testing of memristors in a 6x6 crossbar array to evaluate performance, including sneak current suppression and synaptic behavior.

Main Results:

• Demonstrated high-reliability Na-doped TiO2 memristors with reversible sodium migration as the switching mechanism.
• Achieved electroforming-free and self-rectifying resistive switching, suitable for selectorless CAAs.
• Measured negligible sneak currents (<100 pA) in a 6x6 crossbar array, enabling effective node addressing with low read currents (<1 µA).

Conclusions:

• Na-doped TiO2 memristors grown by ALD offer a promising solution for reliable nonvolatile memory in artificial neural networks.
• The developed memristors exhibit excellent synaptic characteristics, achieving >99.1% accuracy in image recognition tasks using convolutional neural networks.
• These findings pave the way for practical, high-performance, selectorless crossbar arrays for advanced neuromorphic computing applications.