Wavelength-Multiplexed Operation of Photonic-Electronic Resonant Tunnelling Diode Neurons for Scalable Optical Neuromorphic Computing

Neuromorphic computing is a fast-moving field that aims to mimic the high parallelism, and low energy operation, of neurons and neural networks to create crucial pathways for fast and efficient, next generation computing technologies. One potential candidate, proposed to serve as a fast and efficient photonic-electronic artificial spiking neuron is the Resonant Tunnelling Diode (RTD). RTDs are semiconductor devices that have a highly non-linear current-voltage relationship, exhibiting regions of negative differential resistance (NDR) due to their ultra-fast quantum tunnelling effects [1]. When biased in close proximity to their NDR region, RTDs have demonstrated multiple neuron-like behaviours such as excitable spiking, refractoriness and temporal integration. These neuronal behaviours have all been achieved, both electrically and optically, in response to fast (sub-nanosecond) inputs [2]. However, in this work, we demonstrate experimentally (for the first time) the ability of photonic-electronic RTD spiking neurons to successfully capitalise on a (optical) degree of freedom, namely the wavelength of the optical input channel.