Memristor devices are considered to have the potential to implement unsupervised learning, especially spike timing-dependent plasticity (STDP), in the field of neuromorphic hardware research. In this study, a neuromorphic hardware system for multilayer unsupervised learning was designed, and unsupervised learning was performed with a memristor neural network. We showed that the nonlinear characteristic memristor neural network can be trained by unsupervised learning only with the correlation between inputs and outputs. Moreover, a method to train nonlinear memristor devices in a supervised manner, named guide training, was devised. Memristor devices have a nonlinear characteristic, which makes implementing machine learning algorithms, such as backpropagation, difficult. The guide-training algorithm devised in this paper updates the synaptic weights by only using the correlations between inputs and outputs, and therefore, neither complex mathematical formulas nor computations are required during the training. Thus, it is considered appropriate to train a nonlinear memristor neural network. All training and inference simulations were performed using the designed neuromorphic hardware system. With the system and memristor neural network, the image classification was successfully done using both the Hebbian unsupervised training and guide supervised training methods.

Memristor devices are generally suitable for incorporation in neuromorphic systems as synapses because they can be integrated into crossbar array circuits with high area efficiency. In the case of a two-dimensional (2D) crossbar array, however, the size of the array is proportional to the neural network’s depth and the number of its input and output nodes. This means that a 2D crossbar array is not suitable for a deep neural network. On the other hand, synapses that use a memristor with a 3D structure are suitable for implementing a neuromorphic chip for a multi-layered neural network. In this study, we propose a new optimization method for machine learning weight changes that considers the structural characteristics of a 3D vertical resistive random-access memory (VRRAM) structure for the first time. The newly proposed synapse operating principle of the 3D VRRAM structure can simplify the complexity of a neuron circuit. This study investigates the operating principle of 3D VRRAM synapses with comb-shaped word lines and demonstrates that the proposed 3D VRRAM structure will be a promising solution for a high-density neural network hardware system.