CMOS LIF Neurons With Local Membrane Dynamic Biasing Based on Reciprocal Inhibition for Self-Oscillatory Neural...

Area of Science:

• Neuroscience
• Electrical Engineering
• Computer Science

Background:

• Biological neural oscillators exhibit complex self-oscillatory biasing behaviors.
• Existing artificial neuron models often lack the ability to replicate these dynamic biasing mechanisms.

Purpose of the Study:

• To design and implement a CMOS-based neuron network capable of emulating biological self-oscillatory biasing.
• To investigate the generation of self-patterned output spikes using dynamic thresholds.

Main Methods:

• Utilized leaky integrate-and-fire (LIF) neuron models with reciprocal inhibition and synaptic fatigue.
• Incorporated an excitation integrator and a bias controller for membrane potential biasing.
• Implemented the network using a 250-nm CMOS process.

Main Results:

• The proposed network successfully generated oscillating membrane potential biases.
• Self-patterned output spikes, including switching and dynamic firing rate patterns, were observed.
• The circuit demonstrated stable operation with low power consumption (99.31μW per neuron) and acceptable variance under process variations (18% std dev in gain, 12% in oscillation period).

Conclusions:

• The developed CMOS neuron network effectively replicates the self-oscillatory behaviors of biological neuron models.
• The design provides a foundation for more biologically plausible artificial neural networks.
• The self-oscillating mechanism offers a novel approach for dynamic thresholding in neuromorphic computing.