The MXenes family has already demonstrated considerable potential in energy storage, electromagnetic, and electrochemical applications. It is usually created by selectively etching the A layer from the bulk MAX phase. The goal of this work is to broaden the family of MAX phases and examine the possibility of etching them to create a 2D system. Here, we provide a machine learning (ML) approach that is able to accurately predict the relative formation energy (ΔH) on small dataset. From the calculated results, our model shows high prediction accuracy as proved by the RMSE of 0.052 for ΔH. From a dataset of 1320 potential MAX candidates, we predicted 734 MAX phases that could be synthesized by experimental. Moreover, we observed that exfoliating 2D MXenes is more difficult when picking an A atom of S element in the MAX phase, however it is simpler when selecting a group III-A and Ⅳ-A element with a high atomic number. Finally, we identified 75 MXenes candidates with the most potential to be exfoliated from their layered bulk phase. Our machine learning technique can speed up the prediction of possible 2D MXenes while reducing calculation time by more than an order of magnitude.
Read full abstract