Brain-like associative learning using a nanoscale non-volatile phase change synaptic device array.

Area of Science:

• Neuroscience
• Nanotechnology
• Computer Engineering

Background:

• Neuroscience and nanoscale electronics are driving interest in brain-like computing hardware.
• Emerging nanoscale memory devices are being explored as synaptic elements for neuromorphic systems.
• Previous network-level studies were limited to simulations, lacking experimental validation.

Purpose of the Study:

• To experimentally demonstrate array-level associative learning using phase change synaptic devices.
• To investigate the brain-like pattern storage and recall capabilities of a synaptic grid.
• To analyze the robustness and energy efficiency of the proposed hardware.

Main Methods:

• Utilized phase change memory (PCM) cells as synaptic elements in a grid configuration.
• Implemented Hebbian learning rules for training the synaptic grid.
• Experimentally tested pattern storage and associative recall capabilities.

Main Results:

• The experimental synaptic grid successfully stored presented patterns and recalled missing ones associatively.
• The system demonstrated robustness to device variations, which can be managed by adjusting training epochs.
• A trade-off between network variation tolerance and energy consumption was identified, with lower tolerance decreasing energy use.

Conclusions:

• Experimental validation of array-level associative learning using phase change synaptic devices is achieved.
• The developed hardware mimics biological brain organization for pattern association tasks.
• The findings highlight the potential for robust and energy-efficient neuromorphic computing hardware.