Stochastic resonance in an ensemble of single-electron neuromorphic devices and its application to competitive neural networks

Neuromorphic computing based on single-electron circuit technology is gaining prominence because of recent claims about its massively increased computational efficiency and its increasing relevance between computer technology and nanotechnology [1], [2]. Its impact will be strongly felt maximum when single-electron circuits can operate at room temperature, based on fault-and noise-tolerant neural structures. In this paper, inspired by stochastic resonance (SR) in an ensemble of spiking neurons [3], we designed a basic single-electron neural component and examined its statistical results on the network. A single-electron box, in which a quantum dot lies between a nanoscale tunnel junction with capacitance C and a gate capacitance CG whose order of magnitude is close to that of C, is the simplest tunnel junction circuit [4]. We see a significant similarity between single-electron boxes and conductance-based neuron models, that is the tunnel junction and CG in a single-electron box correspond to a voltage-controlled gate and membrane capacitance in the neuron model, respectively. Furthermore, electron tunneling at the junction results in a sudden voltage change on the quantum dot (therefore it corresponds to a spike generation in neurons) and is easily perturbed by thermal fluctuations, as in real neurons