Prediction of IV curves for a superconducting thin film using artificial neural networks

• We have proposed to tune the current-voltage (IV) characteristics on a superconducting thin with a square array of antidots.
• Artificial neural network based approach has been proposed for extrapolating the IV curves for a wide range of temperature and magnetic field values.
• We have trained the artificial neural network on experimental data with varying number of parameters.
• Bayesian backpropagation framework has been used with Levenberg-Marquardt algorithm for cost function minimization.
• We have used mean squared error for Performance measure between experimental and the predicted results.

We propose a framework using artificial neural networks that predicts the IV characteristics of a superconducting thin film with square array of nano-engineered periodic antidots, called holes. We adopt the conventionally used commercial physical properties measurement system to obtain a dataset comprising transport measurements, and use this dataset to train our artificial neural network. Once trained, the model is capable of predicting the curve for varying temperature and magnetic flux values, which are cross validated by the physical properties measurement system. Consistent with the works in literature, our framework suggests Josephson Junctions like behavior near transition temperature and at stronger magnetic fields. Our study is important since repeated measurements using the conventional method are time consuming and costly; we demonstrate that the proposed method may be effectively used to classify the IV characteristics over a wide range of temperature and magnetic field values.