DeFecT-FF: a machine learning force field framework for high throughput defect modeling in CdTe-based solar cells

Abstract

We developed a framework for predicting the energies and ground state configurations of native point defects, extrinsic dopants and impurities, and defect complexes across zinc blende-phase Cd/Zn-Te/Se/S compounds, important for CdTe-based solar cells. This framework, named DeFecT-FF, is powered by high-throughput density functional theory (DFT) computations and crystal graph-based machine learning force field (MLFF) models trained on the DFT data. The Cd/Zn-Te/Se/S chemical space is chosen because alloying at Cd or Te sites is a promising avenue to tailor the electronic and defect properties of the CdTe absorber layer to potentially improve solar cell performance. The sheer number of defect configurations achievable when considering all possible singular defects and their combinations, symmetry-breaking operations, and defect charge states, as well as the expense of running large supercell calculations, makes this an ideal problem for developing accurate and widely-applicable force field models. Here, we introduce our dataset of structures and energies from HSE06 geometry optimization, including bulk and alloyed supercells with and without defects. Data were gradually expanded using active learning and accurate MLFF models were trained to predict energies and atomic forces across different charge states. Via accelerated prediction and screening, we identified many new low energy defect configurations and obtained high-fidelity defect formation energy diagrams using HSE06 calculations with spin-orbit coupling. The DeFecT-FF framework has been released publicly as an online tool on the nanoHUB platform, allowing users to upload any crystallographic information file, generate defects of interest, and compute defect formation energies as a function of Fermi level and chemical potential conditions, thus bypassing expensive DFT calculations.

References in corpus (18)

• Crystal Graph Convolutional Neural Networks for an Accurate and Interpretable Prediction of Material Properties
• Graph Networks as a Universal Machine Learning Framework for Molecules and Crystals
• A Universal Graph Deep Learning Interatomic Potential for the Periodic Table
• DScribe: Library of Descriptors for Machine Learning in Materials Science
• Achieving DFT accuracy with a machine-learning interatomic potential: thermomechanics and defects in bcc ferromagnetic iron
• Nonrad: Computing Nonradiative Capture Coefficients from First Principles
• Identifying the ground state structures of point defects in solids
• Unified Graph Neural Network Force-field for the Periodic Table
• Gate-Tunable Semiconductor Heterojunctions from 2D/3D van der Waals Interfaces
• Transfer learning for materials informatics using crystal graph convolutional neural network
• Updates to the DScribe Library: New Descriptors and Derivatives
• Rapid Recombination by Cadmium Vacancies in CdTe
• Lone-pair effect on carrier capture in Cu${}_2$ZnSnS${}_4$ solar cells
• Defect identification based on first-principles calculations for deep level transient spectroscopy
• doped: Python toolkit for robust and repeatable charged defect supercell calculations
• Machine-learning structural reconstructions for accelerated point defect calculations
• A Deep-learning Model for Fast Prediction of Vacancy Formation in Diverse Materials
• Machine learning for impurity charge-state transition levels in semiconductors from elemental properties using multi-fidelity datasets