Self-Organizing Neural Networks Based on OxRAM Devices under a Fully Unsupervised Training Scheme

Area of Science:

• Neuromorphic Engineering
• Materials Science
• Artificial Intelligence

Background:

• Neuromorphic architectures aim to mimic the brain's structure and function.
• Self-organization and unsupervised learning are key challenges in developing intelligent systems.
• Oxide-based Resistive Random Access Memory (OxRAM) offers potential for efficient synaptic emulation.

Purpose of the Study:

• To develop a fully unsupervised learning algorithm for neuromorphic architectures.
• To experimentally demonstrate spike-timing dependent plasticity (STDP) in OxRAM devices.
• To enable autonomous hierarchical computing through concatenated neuromorphic layers.

Main Methods:

• Experimental demonstration of STDP in OxRAM devices.
• Proposal of specific waveforms to induce symmetric conductivity changes.
• Development of an empirical model to characterize OxRAM plasticity.
• Simulation of a neuromorphic system employing the STDP learning rule.

Main Results:

• Successful experimental demonstration of STDP in OxRAM.
• Validation of an unsupervised learning algorithm in a simulated neuromorphic system.
• Empirical model accurately describes observed plasticity.
• Demonstration of concatenated neuromorphic layers for hierarchical processing.

Conclusions:

• The proposed unsupervised learning algorithm and STDP in OxRAM enable self-organization in neuromorphic systems.
• The developed approach facilitates autonomous hierarchical computing.
• OxRAM devices are viable for implementing efficient, brain-inspired learning.