A compact phenomenological digital neuron implementing the 20 Izhikevich behaviors

Abstract

The possible end of Moore's law drives a growing interest for a paradigm shift from classical von Neumann to bio-inspired computing architectures. Toward this objective, the field of neuromorphic engineering studies spike-based information processing with neuron and synapse elements. While analog circuit design allows to emulate the neuron biophysics with only a few transistors, digital circuit design shows high potential to leverage technology scaling with short design times at the expense of limited area efficiency. As recent developments in synapse implementation allow to reach densities as low as 1μm2 per synapse, it is now crucial to optimize the neuron area. In order to push digital design to compact implementations and take advantage of nanoscale CMOS technologies, we developed a phenomenological digital neuron that implements the full repertoire of the 20 Izhikevich behaviors with time constants configurable from biological-to accelerated-time. The proposed implementation occupies only 574μm2 in 28nm FDSOI CMOS.