An analogue sum and threshold neuron based on the quantum tunnelling amplification of electrical pulses

Abstract

We demonstrate a spatial neuron that sums and regenerates electrical pulses in real time. The neuron uses a monolithic web of micro-transmission lines to propagate electrical pulses to a 'soma' where they are regenerated via quantum tunnelling amplification. The gain of the neuron follows a sigmoid curve similar to the one that controls the firing of real neurons. We report on the dependence of the regeneration threshold on bias parameters and obtain a good fit of the measured threshold by computing the stability diagram of the soma. The neuron is shown to regenerate coincident pulses with a timing sensitivity of 10 μs compared to milliseconds for real neurons. The present design demonstrates that the physics underpinning analogue computation in biological neurons has an equivalent in modern semiconductor structures. © IOP Publishing Ltd and Deutsche Physikalische Gesellschaft.