Sub-nJ per Decision Schmitt Trigger Comparator for Neuromorphic Spike Detection in 55 nm Technology

Abstract

Neuromorphic circuits are known for their promising ultra-low power AI applications in IoT field. However, sub-100 mV supply voltages hamper digital-enable devices due to their non-discrete and highly non-linear response. In this paper, a low-power Smith trigger comparator is proposed to interface spiking analog eNeurons and digital circuits. To this end, a subthreshold bias and BiCMOS 55 nm node technology are chosen. The proposed comparator is post-layout validated, having a maximum decision frequency of 400 kHz and an energy efficiency of 747 aJ/spike. This performance is compatible with existing artificial neurons.