A Neuromorphic Quadratic, Integrate, and Fire Silicon Neuron with Adaptive Gain.

Area of Science:

• Neuroscience
• Electrical Engineering
• Computer Science

Background:

• The Quadratic, Integrate, and Fire (QIF) neuron model is a fundamental component in computational neuroscience.
• Implementing complex neuron models in silicon is crucial for developing energy-efficient neuromorphic hardware.
• Previous implementations may lack optimization for low power or specific functionalities.

Purpose of the Study:

• To design, manufacture, and test the first integrated circuit (IC) implementation of the QIF neuron model.
• To optimize the QIF neuron IC for low current consumption and high functionality.
• To validate the circuit's ability to replicate key QIF neuron behaviors.

Main Methods:

• Designed a QIF neuron circuit using $0.5 \mu \mathrm {m}$ silicon technology.
• Incorporated hysteretic reset, non-inverting integrator, and voltage-squarer circuits.
• Optimized for low power, achieving 1.56 mA current consumption per circuit.

Main Results:

• Successfully manufactured and tested the QIF neuron IC, measuring $268 \mu \mathrm {m}$ height $\times 400 \mu \mathrm {m}$ width.
• The IC demonstrated true spiking behavior, including bistability, monotonic, and excitability.
• Achieved adjustable time constants via an external capacitor and adaptive gain via an external resistor.

Conclusions:

• This novel QIF neuron IC represents a significant advancement in silicon neuron technology.
• The design offers low power consumption and flexible parameter control for neuromorphic systems.
• The successful implementation paves the way for more complex and efficient brain-inspired computing architectures.