Development of Neural Network Models for Current-Voltage Characteristics Prediction

As an undergraduate researcher under Prof. Narayana Aluru's research group, I worked with Pikee Priya, a postdoctoral research associate, to develop a neural network for prediction of Current-Voltage characteristics when given a set of material properties. The normal way of finding those characteristics relies on a numerical model where transport and Poisson’s equations need to be solved and the whole process is extremely computationally expensive. Hence, by inputting enough data, a neural network could be trained, and the prediction of characteristics will be much more effective compared with the regular method. My main tasks were to gather all the data, determine the architecture of the network, train the network using supercomputer (BlueWater), and optimize the performance of the network. The final model was used to make numerous predictions with an average percentage of error around 2%.