Leaky Integrate-and-Fire Neuron with a Refractory Period Mechanism for Invariant Spikes

Spiking neural networks (SNNs) are a widespread research topic, as they are a promising solution for efficient and low-energy signal processing. The advantage in energy consumption of SNN algorithms pre-developed in software is obtained by their transfer to neural application-specific integrated circuits (ASICs). The neurons and synapses that are used on algorithm level have to be mapped by special circuits on the hardware level. One of the most widely used neuron models due to its low computational complexity and high biological inspiration is the leaky integrate-and-fire (LIF) neuron. In this paper, a modification of an energy-efficient LIF neuron is presented with a novel method to use a refractory period (RP) to generate invariant output spikes. By modifying the RP mechanism, the controllability and stability of entire SNN systems on the circuit level can be significantly improved. Due to the low energy of 1.4 pJ / spike and the invariance of these, it becomes easier to predict the total energy consumption of a large-scale SNN. The concept is verified in measurement by fabricating the circuit in a 130 nm BiCMOS process.