Modelling of nanostructured memristor device characteristics using Artificial Neural Network (ANN)

The present paper reports modelling of nanostructured memristor device characteristics using Artificial Neural Network (ANN). The memristor is simulated using linear drift model and data generated thereof is applied for learning, testing and validation of ANN architecture. In the present investigation we demonstrate optimum ANN architecture for the said modelling by varying the number of hidden neurons and percentage of testing data. The percentage of validation data is varied in order to accomplish tuning of the experiment. Performance of ANN architecture thus derived has been measured in terms of Mean Squared Error (MSE) and Pearson correlation coefficient (r). The hidden units consist of nonlinear sigmoid activation functions and training algorithm is based on a Levenberg-Marquardt Backpropogation method. The reported ANN architecture reveals best performance at lower numbers of hidden neurons and further lower percentage of testing and validation data. Additionally, optimized ANN structure is selected for modelling of other characteristics of memristor such as, flux-charge relation, time domain memristance and width of doped region. The results support, ANN as the preeminent tool for modelling of nonlinear devices such as memristor and the suite of other emerging nanoelectronics devices.