CMOS Leaky Integrate-and-Fire Neurons with Local Dynamic Biasing and Inhibitory Self-Oscillation

This paper presents a CMOS-based neuromorphic circuit that can simulate self-oscillatory firing behaviors. Based on the leaky integrate-and-fire neuron model, a dynamic bias controller and an excitation integrator are added for controlling inhibitory connection between neurons as well as switching the operation mode with an integrated excitatory control signal. The proposed neuron circuit replicates extracellular fluidic biasing of membrane potential that oscillates at low frequency on top of the temporal integration of the action potential firing at high frequency. The proposed neuron can also replicate the synaptic fatigue by implementing inhibition declining over time, which leads the bias to oscillate. The proposed neuron model was implemented with 250nm CMOS process at 2.5V supply voltage. Each neuron consumes 46.16pJ energy/spike from leaky integration. Simulation results verified that a pair of proposed neuron cells can generate an oscillatory rhythm, where the bias waveform oscillates with the output spikes generated based on the bias level change. The proposed neuron circuit model can be utilized to reproduce more complicated oscillatory network behaviors of biological neurons.