Deep Bandgap and Band Structure Engineering by Machine Learning

The ability to deform and keep silicon at large strains harbingers a new age of deep elastic strain engineering (ESE) of electronic materials. Current strained-Si technology thus represents only “tip of the iceberg” of what silicon can do as the most versatile and processable electronic material. Deep ESE explores the full six-dimensional space of admissible elastic strain and its effect on physical properties, beyond linear elasticity and perturbation theory. Here we present a general method that combines machine learning and ab initio calculations to guide rational ESE whereby unprecedented material properties and performance could be designed. This method invokes recent advances in artificial intelligence by utilizing a limited amount of ab initio data for the training of a surrogate model. In particular, an artificial neural network predicts the electronic bandstructure within the accuracy of 19 meV. Our model is utilized to discover the indirect-to-direct bandgap transition and semiconductor-to-metal transition in the entire strain space. By finding out the most energy efficient deformation manner to achieve a desirable bandgap, we demonstrate for the first time how to identify novel pathways to tailor any material figure of merit, which is of central importance for ESE.