Biologically inspired tonic and bursting LIF neuron model for spiking neural network: a CMOS implementation

Abstract

The human brain has been encompassed by neurons and synapses, where the signal is propagated as a spike. This perception seeks to create hardware systems called neural cores that are comprised of artificial bio-neurons and synapses; these systems resemble the brain’s functions and emulate the different spiking patterns to operate the neuromorphic processors. Through this motivation, the paper presents a low-power Schmitt trigger-based Leaky Integrate and Fire (LIF) neuron model that offers significantly less energy per spike and showcases the dynamic spike frequency and refractory periods that are regulated by the membrane and reset capacitance in addition to refractory circuitry. The incorporation of a hysteresis comparator, e.g., the Schmitt trigger, enhances noise immunity and facilitates dynamic threshold adjustments, enabling faster switching and reducing energy consumption. The neuron models are simulated in Cadence Virtuoso GPDK 45 nm Technology to obtain dynamic tonic and burst spiking patterns; subsequently, the significantly smaller spike pulse width of the proposed model is measured from the dynamic pattern as 1.867 ns, and the refractory period is measured as 0.2 ns respectively. This proposed model consumes less energy, 524.415aJ per spike, under the 1 V power supply and 1 ms step pulse. Typically, the spike pulses have different shapes and magnitudes based on the functions of membrane potential, which are applicable to realize the spiking behavior of SNNs.