Low-Power Neuromorphic Axon-Hillock Neuron with Memristor-based Synapses for High-Efficiency Spiking Neural Networks

Neuromorphic computing is a technology that imitates the brain’s incredible efficiency in processing information, and it is seen as a potential source of solutions for computing tasks that require low power and high speed. We have developed a low power Axon- Hillock neuron design coupled with a triple memristor synapse architecture processed in 45 nm Cadence Virtuoso CMOS technology. The approach in the paper comprises two Axon-Hillock neurons linked via three memristors, which facilitates the fine adjustment of synaptic weights with a drastically lowered energy usage. The simulated results indicate that the design attains an average power of 11.74 nanowatts(nW), an energy per spike of 244.3 femtojoules(fJ), and a frequency of 47.6 kilohertz(kHz), thus being able to outperform the existing CMOS neuron realizations. Such an architecture manages to strike a compromise between very fast spiking and extremely low energy consumption, thereby becoming a perfect candidate for large-scale neuromorphic systems. This study is a step towards realizing energy-efficient artificial intelligence hardware of the future that leverages the potential of memristor-based neuron circuits.