α-Fe₂O₃-based artificial synaptic RRAM device for pattern recognition using artificial neural networks

Area of Science:

• Materials Science
• Neuroscience
• Computer Science

Background:

• Artificial neural networks (ANNs) require efficient hardware for image recognition.
• Artificial synaptic devices mimic biological synapses to improve ANN performance.
• Resistive random access memory (RRAM) devices offer non-volatile memory with analog switching.

Purpose of the Study:

• To develop and characterize an alpha-Fe2O3-based artificial synaptic device.
• To evaluate the device's performance in simulating synaptic learning rules.
• To assess the device's potential for practical artificial neural network implementation in image recognition.

Main Methods:

• Fabrication of a Silver/alpha-Fe2O3/Fluorine-doped Tin Oxide (Ag/α-Fe2O3/FTO) device.
• Demonstration of synaptic plasticity: long-term potentiation, long-term depression, and spike-time-dependent plasticity.
• Implementation of off-chip training using a backpropagation algorithm with device-derived synaptic weights.

Main Results:

• The Ag/α-Fe2O3/FTO device exhibited non-volatile analog resistive switching characteristics.
• Successful emulation of key synaptic learning rules was achieved.
• High pattern recognition accuracy: 88.06% on Fashion-MNIST and 97.6% on MNIST.
• Effective integration with a backpropagation algorithm for enhanced accuracy.

Conclusions:

• The fabricated alpha-Fe2O3-based artificial synaptic device shows promising characteristics for ANNs.
• The device's performance, combined with a novel weight mapping strategy, enables high accuracy in image recognition.
• This research highlights the potential of the device for practical applications in artificial neural networks.