A Self-Rectifying Synaptic Memristor Array with Ultrahigh Weight Potentiation Linearity for a Self-Organizing-Map...

Area of Science:

• Materials Science
• Computer Science
• Electrical Engineering

Background:

• Self-rectifying (SR)-synaptic memristors are key for high-density neuromorphic computing and 3D integrated systems.
• SR-synaptic memristors can suppress sneak path currents in crossbar arrays.
• Existing SR-synaptic memristors face challenges with nonlinear weight potentiation and steep depression, limiting their use in artificial neural networks (ANNs).

Purpose of the Study:

• To develop a SR-synaptic memristor and cross-point array that overcomes limitations in linearity and synaptic depression.
• To demonstrate the device's capability in image processing tasks.
• To implement an unsupervised self-organizing map (SOM) neural network for efficient and accurate orientation recognition.

Main Methods:

• Fabrication of a Pt/NiO/WO3-:Ti/W based SR-synaptic memristor and cross-point array.
• Characterization of device linearity and sneak path current suppression.
• Demonstration of image contrast enhancement and background filtering.
• Development and testing of an unsupervised SOM neural network for orientation recognition.

Main Results:

• Achieved ultrahigh-weight potentiation linearity up to 0.9997.
• Successfully demonstrated image contrast enhancement and background filtering.
• Developed an unsupervised SOM neural network with 0.98 recognition accuracy and high training efficiency.
• Showcased high resilience to noise and steep synaptic depression.

Conclusions:

• The developed SR-synaptic memristor and array effectively address the linearity and depression challenges in conventional ANNs.
• The device enables high-density, efficient, and accurate neuromorphic computing applications.
• This work expands the potential of large-scale oxide SR-synaptic arrays for advanced computing.