Boltzmann machine neural networkdevices using single-electron tunnelling

Abstract

We proposed a method of implementing theBoltzmann machine neural network on electronic circuits bymaking use of the single-electron tunnelling phenomenon. Thesingle-electron circuit shows stochastic behaviour in itsoperation because of the probabilistic nature of theelectron tunnelling phenomenon. It can therefore be successfullyused for implementing the stochastic neuron operation of theBoltzmann machine. The authors developed a single-electronneuron circuit that can produce the function required for theBoltzmann machine neuron. A method for constructing Boltzmannmachine networks by combining the neuron circuits was alsodeveloped. The simulated-annealing operation can be performedeasily by regulating an external control voltage for thenetwork circuits. A sample network was designed that solves aninstance of a combinatorial optimization problem. Computersimulation demonstrated that, through the simulated-annealingprocess, the sample network can converge to the global minimumenergy state that represents the correct solution to theproblem.