Insight into the Electronic Properties of Semiconductor Heterostructure Based on Machine Learning and...

Area of Science:

• Materials Science
• Computational Chemistry
• Condensed Matter Physics

Background:

• First-principles calculations offer insights into material properties but are computationally intensive for discovering new materials.
• Material genetic engineering necessitates efficient methods for predicting material performance.
• Machine learning presents a promising alternative for data mining and performance prediction in materials discovery.

Purpose of the Study:

• To develop and validate machine learning frameworks for predicting the band gap opening value of graphene in heterostructures.
• To screen semiconductor materials for forming effective heterostructures with graphene.
• To compare machine learning predictions with density functional theory calculations.

Main Methods:

• Utilized a dataset of 2076 binary compounds from the Materials Project library.
• Applied t-distributed stochastic neighbor embedding (t-SNE) for unsupervised dimensionality reduction.
• Developed prediction frameworks using Extreme Gradient Boost (XGBoost) and Support Vector Machine (SVM) algorithms.
• Validated predictions against density functional theory (DFT) calculations of energy bands and density of states.

Main Results:

• Achieved an 88.3% accuracy rate for the prediction model.
• Demonstrated minimal discrepancies between predicted and calculated band gap opening values.
• Successfully screened semiconductor components for graphene heterostructures.

Conclusions:

• The developed machine learning framework effectively predicts the band gap opening value in graphene heterostructures.
• This approach shows significant potential for accelerating the discovery of multiphase materials with tailored electronic properties.
• The study highlights the utility of integrating machine learning with first-principles calculations for materials design.