Machine Learning to Predict the L-Point Direct Bandgap of Bi1-xSbx Nanomaterials

TL;DR

This paper employs machine learning techniques to accurately predict the L-point direct bandgap in Bi1-xSbx nanomaterials, overcoming computational challenges of traditional methods.

Contribution

It introduces machine learning models as efficient tools for predicting bandgaps in complex nanostructured materials, improving accuracy and reducing computational costs.

Findings

Support vector regression achieved high accuracy (~0.99 fit)

Effective prediction for thin films and nanowires

Machine learning outperforms traditional ab initio calculations

Abstract

With the development of modern nanoscience and nanotechnology, Bi1-xSbx can be synthesized into different nanoscale and nanostructured forms, including thin films, nanowires, nanotubes, nanoribbons, and many others. However, due to the strong correlation between electrons and holes at the L-point in the Brillouin zone, the direct band evolves in an anomalous manner under the quantum confinement when nanostructured. Due to the alloying and the low symmetry, predicting the L-point direct bandgap in a nanomaterial using either ab initio calculations or kp perturbations can be computationally costive or inaccurate. We here try to solve this problem using the machine learning methods, including the support vector regression, the regression tree, the Gaussian process regression, and the artificial neural network. A goodness-of-fit of ~0.99 can be achieved for Bi1-xSbx thin films and nanowires.