Hardware implementation of self-organizing maps using memristors, a simulation study

Memristors are being widely proposed for hardware implementation of perceptron multilayer neural networks. In this work, we explore the use of memristors as the elements that store the actual state of neurons in self-organizing maps, other class of neural networks used for unsupervised classification tasks. We address the problem of the representation of the neuron elements into a discrete set of distinguishable memristor conduction states, taking the device variability into account.