Machine Learning and Electronic Phases and Band Structures of Thin Film Narrow-Gap and Semimetallic Electronic Materials for New Chips and New Energy

The current semiconducting and computing materials system is mostly based on thin films. When exploring the next generation systems for electronics, researchers have found that narrow-band-gap materials are of great interest, mainly because of their ultra-high mobility and small power consumption. When narrow-band-gap materials are made into thin films, the quantum confinement effect is obviously higher than in wide-band-gap semiconductors. 

Since the confinement is correlated with the non-linear band shape, the traditional effective-mass model or k?p perturbation is not adequately accurate in describing the band structure in two-dimension. Though first principle calculations can be used for accurate band structure prediction, the computational cost is still high for such nanoscale alloys. 

In this present work, we provide a new methodology to predict the band structure and electronic phases of narrow-band-gap thin films, using modern techniques of machine learning, including supporting the vector machine, the decision tree, and the artificial neural networks.  The accuracy of different machines, as well as the impact of training pool size will be discussed.