An Artificial Neural Network Based on Oxide Synaptic Transistor for Accurate and Robust Image Recognition

Area of Science:

• Materials Science
• Neuroscience
• Computer Engineering

Background:

• Synaptic transistors are key components for hardware artificial neural networks (ANNs).
• Programmable conductance in synaptic transistors enables emulation of neural plasticity.
• Solution-processed dielectric films offer a low-temperature fabrication route for synaptic devices.

Purpose of the Study:

• To engineer AlO_x/InO_x synaptic transistors using a solution process.
• To investigate the long-term potentiation characteristics of these transistors.
• To construct and evaluate a hardware ANN for noisy image recognition.

Main Methods:

• Fabrication of AlO_x/InO_x synaptic transistors via a solution process.
• Characterization of transistor behavior, including long-term potentiation under gate voltage pulses.
• Implementation of a 9x3 synaptic transistor-based ANN for image recognition tasks.

Main Results:

• The engineered transistors exhibited stable long-term potentiation, mimicking synaptic plasticity.
• The hardware ANN successfully recognized 3x3 pixel images of characters 'z', 'v', and 'n'.
• The network achieved 85% accuracy in recognizing characters even with up to 40% noise.

Conclusions:

• Solution-processed AlO_x/InO_x synaptic transistors are viable for building hardware ANNs.
• These metal-oxide transistors demonstrate significant long-term potentiation for accurate noisy image recognition.
• The developed hardware ANN shows potential for robust pattern recognition in challenging conditions.