Predicting the transport properties of silicene nanoribbons using a neural network

Atomistic quantum transport simulations are used to generate the electronic and transport properties of 10,000 realistic silicene nanoribbons (SiNRs) with edge defects. This ensemble of 20 nm-long and 2.1 nm-wide SiNRs is divided into the training and inference set for the artificial neural network (ANN) employed for the prediction of edge-defect-limited carrier mobility from the known values of bandgap and nanoribbon conductance. We find that an optimized ANN with 3 hidden layers can predict SiNR mobility values and variability histograms with acceptable accuracy, thus providing a useful supplement to atomistic quantum transport simulations that take several hours or days for large device ensemble sizes.