Machine-learning-assisted optimization of Ga-free type-II superlattices for enhanced vertical hole mobility

Gaussian process regression is used to develop a model for predicting carrier transport in superlattice (SL) structures grown on GaSb and 6.2 Å substrates. This model is used to search SL structures optimized for enhanced hole transport in the vertical (growth) direction. Nonequilibrium Green’s functions calculations are used to determine the vertical hole mobility of several chosen structures in both ideal and disordered cases. It is demonstrated that the conductivity effective mass can be used in some cases as a qualitative predictor for the relative hole mobility between different structures. However, in the case of disordered SLs, the effective mass must be calculated from quasi-random disordered structures as the results may differ significantly from the ideal case. Ultimately, a methodology for predicting SL structures optimized for high hole transport efficiency in the case of ideal and disordered SLs is demonstrated.