A Split-Gate Positive Feedback Device With an Integrate-and-Fire Capability for a High-Density Low-Power Neuron...

Area of Science:

• Neuromorphic Engineering
• Solid-State Electronics
• Artificial Intelligence Hardware

Background:

• Conventional hardware-based spiking neural networks (SNNs) face challenges with large circuit area and high power consumption.
• Existing neuron circuits struggle to efficiently mimic the integrate-and-fire function of biological neurons.

Purpose of the Study:

• To propose a novel split-gate floating-body positive feedback (PF) device as an efficient neuron for SNNs.
• To demonstrate the device's capability in mimicking biological neuron functions with reduced area and power.

Main Methods:

• Fabrication and characterization of a split-gate floating-body positive feedback (PF) device.
• Implementation of PF device-based neuron circuits to analyze energy consumption and area efficiency.
• Simulation of a dense multiple PF neuron system for demonstrating neural network functionalities like reset and lateral inhibition.

Main Results:

• The PF device exhibits a super-steep subthreshold swing (SS) below 0.04 mV/dec due to positive feedback.
• PF neuron circuits achieve ultra-low energy consumption of approximately 0.25 pJ/spike, ~100x lower than conventional circuits.
• The PF neuron area is reduced by ~17 times compared to conventional designs, owing to charge trapping for integration.
• Successful simulation of a dense multiple PF neuron system performing on-line unsupervised pattern learning and recognition.

Conclusions:

• The proposed PF device offers a highly efficient solution for hardware-based SNNs, addressing area and power limitations.
• The device's characteristics enable efficient mimicry of biological neuron functions, paving the way for advanced neuromorphic computing.
• Demonstrated feasibility in a dense neuron system highlights the potential of PF devices for on-line learning and pattern recognition in neural networks.