Applying recurrent neural networks to predict the work function of metal adatom adsorption on graphene

Abstract

The barrier height has a large impact on the performance of transistors, lasers, and solar cells. To tune such a barrier height, a bidirectional long short-term memory network (LSTM) model is proposed to predict the work function of adsorbed metal atoms on graphene. The largest relative errors predicted by the proposed model compared to those obtained from first-principles calculations are less than 10 %. However, more than one order of magnitude difference in field-emission currents can be caused by such errors according to calculations. It implies that the proposed model can be used to quickly screen surface structures with its work function being around a target work function before performing first-principles calculations. Thus, it can save a lot of computing costs and time costs. The selected adsorbed metal atoms at different sites on graphene can be implemented to improve the performance of graphene-based devices by using the proposed method.PACS numbers85.30.De, 85.30.Tv, 73.40.Kp.