Defect Properties · foundry-ml
semiconductor_defectlevels_v1-1
Computed defect charge state transition levels for 2,910 semiconductor-impurity pairs to predict doping and trap states.
What semiconductor_defectlevels_v1-1 is
semiconductor_defectlevels_v1-1 is a tabular materials science dataset compiled by Maciej P. Polak, Ryan Jacobs, Arun Mannodi-Kanakkithodi, Maria K. Y. Chan, and Dane Morgan. It compiles density functional theory (DFT) calculations of defect charge-state transition levels across 2,910 semiconductor-impurity pairs.
Charge-state transition levels determine how impurity atoms behave within a host semiconductor, dictating whether they provide carrier doping or act as detrimental carrier traps. This dataset provides descriptors and multi-fidelity DFT benchmarks to support machine learning models that predict transition levels directly from elemental and host properties.
How it was made
The dataset records 2,910 semiconductor-impurity configurations alongside tabular input features and target values. Input features include host material composition, sublattice identities, experimental host bandgap in electronvolts, lattice constants in angstroms, host dielectric constants, initial and final defect charge states, and one-hot encodings indicating whether impurities reside on lattice sublattices or various interstitial positions.
Target values measure defect charge-state transition levels in electronvolts relative to the valence band maximum (VBM). Values are supplied at two levels of theory: DFT using the Perdew–Burke–Ernzerhof (PBE) functional and hybrid DFT using the Heyd–Scuseria–Ernzerhof (HSE) functional, accompanied by band alignment correction shifts.
Key facts
- Dataset size
- 2,910 semiconductor-impurity pairs
- Data format
- Parquet
- Modality
- Tabular
- Release year
- 2022
- Publisher
- Materials Data Facility / Foundry-ML
- Task categories
- Tabular regression, tabular classification
- Primary targets
- PBE defect level relative to VBM (eV), HSE defect level relative to VBM (eV)
- Provided splits
- train
What you can use it for
- Impurity transition level prediction
- Train regression models to predict PBE and HSE defect transition energy levels relative to the host valence band maximum using elemental and lattice descriptors.
- Dopant and trap state screening
- Screen host-impurity combinations to identify which defects introduce shallow carrier states suitable for doping versus deep carrier-trapping levels.
- Multi-fidelity machine learning
- Develop multi-fidelity transfer learning algorithms that link lower-accuracy PBE defect calculations with higher-accuracy HSE defect calculations.
Explore the data
The schema, real rows, full-text search and a SQL filter — read in your browser, nothing downloaded.
Explore
Read the real rows without downloading anything
Reading rows…
LiveRead from foundry-ml/semiconductor_defectlevels_v1-1 at the moment you asked. Nothing is cached or stored — every row above came from that request.
Strengths and limits
Good at
- Provides paired transition level targets from both standard PBE and hybrid HSE density functional theory calculations.
- Contains structured geometric indicators, including specific lattice and interstitial site classifications alongside host dielectric and bandgap measurements.
- Formats tabular inputs and targets directly for tabular regression and classification tasks.
Watch out for
- Delivered with a single designated train split, leaving validation and test partitioning to the researcher.
- Confined to the host material compositions, impurity species, and site types included in the 2,910 calculated records.
Questions
Can I look inside semiconductor_defectlevels_v1-1 without downloading it?
Yes. This page reads its schema, real rows and per-column statistics in the browser, with full-text search and a SQL filter. Nothing is downloaded to your machine.
What licence is semiconductor_defectlevels_v1-1 under?
Its authors declare cc-by-4.0. Read the terms on the original page before reusing the data.
What reference point is used for the defect transition energy levels?
Defect levels are reported in electronvolts relative to the host material's valence band maximum (VBM).
What host material properties are included as inputs?
Features include the host composition, constituent elements on specific sublattices, experimental bandgap in electronvolts, lattice constant in angstroms, and dielectric constant.
How are defect locations identified in the dataset?
Defects are described by site type, removed host element, impurity identity, and boolean or one-hot indicators marking whether the atom occupies an A-site, B-site, or one of several interstitial site configurations.
How to cite it
Polak, Maciej P. and Jacobs, Ryan and Mannodi-Kanakkithodi, Arun and Chan, Maria K. Y. and Morgan, Dane. Machine learning for impurity charge-state transition levels in semiconductors from elemental properties using multi-fidelity datasets. Materials Data Facility, 2022. https://doi.org/10.18126/ite0-3iah
Written from semiconductor_defectlevels_v1-1's own dataset card. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.