An ultra-compact leaky-integrate-and-fire model for building spiking neural networks

Area of Science:

• Neuroscience
• Electronic Engineering
• Materials Science

Background:

• The leaky-integrate-and-fire model is fundamental for artificial neurons.
• Implementing complex neuronal dynamics in compact, low-power circuits remains a challenge.

Purpose of the Study:

• To introduce an ultra-compact electronic circuit for artificial neurons.
• To demonstrate biologically realistic neuronal features and modularity for deep neural networks.

Main Methods:

• Designed a circuit with two transistors and a silicon controlled rectifier (SCR).
• Incorporated spike-frequency adaptation, refractory period, and voltage modulation.
• Utilized out-of-the-shelf components for circuit construction.

Main Results:

• Achieved an ultra-compact neuron circuit with only three active devices.
• Demonstrated control over characteristic times via resistive parameters.
• Showcased modularity for constructing multi-layer deep neural networks.
• Highlighted low power consumption due to its normally-off state.

Conclusions:

• The developed circuit offers a highly compact and efficient realization of the leaky-integrate-and-fire neuron model.
• Its modularity and low power consumption make it suitable for large-scale neural network applications.
• Future work could explore Mott materials for even greater miniaturization.