AI-Guided Semiconductor Materials Screening
Semiconductor discovery requires simultaneous control of crystal stability, band gap, band-edge alignment, carrier transport, dielectric response, elemental availability, toxicity, and process compatibility. Our service connects curated materials data, application-specific machine learning, and targeted first-principles calculations to reduce a broad chemical space to a traceable shortlist for experimental evaluation.
Projects begin with a target product profile rather than a generic property search. We translate device conditions—such as absorber architecture, operating temperature, substrate, deposition route, dimensionality, and restricted elements—into explicit screening objectives and rejection rules. Candidate rankings are predictions, not experimental validation. We therefore report provenance, model applicability, uncertainty, and the physics-based checks applied to each recommended material. This service is part of our AI for Semiconductor and Electronic Materials portfolio.
Core Screening Services
Search-Space and Data Engineering
We aggregate customer data and public crystal/property records; standardize formulas, structures, units, calculation settings, polymorphs, and experimental conditions; and document deduplication and leakage-aware grouping.
AI Property and Class Screening
Composition- and structure-aware models can address stability class, band gap, direct/indirect character, dielectric response, effective-mass proxies, or other agreed endpoints using validation suited to the data regime.
Multi-Objective Candidate Ranking
Property windows are combined with elemental restrictions, abundance, toxicity flags, novelty, synthesis evidence, and process constraints. Pareto fronts retain transparent trade-offs when objectives conflict.
Physics-Based Verification
High-value candidates are recalculated with harmonized DFT settings for structural relaxation, formation energy, energy above hull, and electronic structure, with optional phonon, defect, dielectric, or higher-fidelity band-gap analysis.
Applicability and Uncertainty Analysis
Ensemble or conformal uncertainty is quantified where supported, out-of-domain chemistries are identified, and interpolation is distinguished from extrapolation before recommendations are made.
Active-Learning Study Design
We recommend the next calculations or measurements that are most informative for the target space, focusing resources on candidates that improve both performance and model knowledge.
Integrated Project Workflow
| Stage | Key Activities | Decision Gate |
|---|---|---|
| 1. Project scoping | Define use case, property windows, operating conditions, prohibited elements, cost and synthesis constraints. | Approved target profile and ranking logic. |
| 2. Data curation | Normalize structures and metadata; assess missingness, duplicates, calculation fidelity, and class imbalance. | Auditable dataset and data-quality report. |
| 3. AI screening | Establish baselines, train validated surrogate models, screen candidate space, and map applicability. | Uncertainty-aware preliminary shortlist. |
| 4. Physics checks | Perform harmonized DFT or other agreed calculations on priority candidates and near-threshold controls. | Remove unstable or electronically unsuitable candidates. |
| 5. Decision package | Rank candidates, document trade-offs, and propose synthesis, characterization, and feedback experiments. | Customer-ready shortlist and validation plan. |
Decision-Ready Deliverables
Data and Model Package
Curated dataset, provenance, units, quality flags, model card, validation design, performance metrics, applicability domain, and documented limitations.
Candidate Scorecard
Ranked candidates with predicted and recalculated properties, confidence indicators, constraint status, trade-offs, and rejection rationale.
Validation Plan
Structured CSV/JSON outputs, technical report, reproducible calculation settings, synthesis hypotheses, characterization priorities, and go/no-go criteria.
Representative Applications
Photovoltaic Absorbers
Screen bulk, thin-film, perovskite, and 2D candidates for target band gaps, band-edge alignment, stability, and restricted-element criteria.
Power and RF Semiconductors
Prioritize wide- and ultrawide-band-gap candidates using dielectric, effective-mass, thermal, stability, and dopability-related evidence.
Light Emitters and Detectors
Identify direct-gap and compositionally tunable materials for UV, visible, or infrared windows, followed by higher-fidelity electronic checks.
Quantum and 2D Platforms
Filter host materials using band gap, symmetry, dimensionality, stability, defect-related requirements, and fabrication constraints.
Scientific Basis and Open-Access Evidence
Open research demonstrates why screening should combine candidate generation, sequential filtering, property prediction, and explicit attention to extrapolation.1 Complementary work shows that interpretable search can convert property models into experimentally useful chemical rules.2
Official Materials Project guidance also notes that semilocal DFT band gaps are systematically underestimated and require method-aware interpretation.3 Our screening plans therefore match model fidelity to the decision and reserve higher-cost calculations or experiments for critical candidates.
1 Sorkun MC, Astruc S, Koelman JMV, Er S. An artificial intelligence-aided virtual screening recipe for two-dimensional materials discovery. npj Computational Materials, 2020, 6: 106. https://doi.org/10.1038/s41524-020-00375-7. Open Access, CC BY 4.0.
2 Choubisa H, Todorović P, Pina JM, et al. Interpretable discovery of semiconductors with machine learning. npj Computational Materials, 2023, 9: 117. https://doi.org/10.1038/s41524-023-01066-9. Open Access, CC BY 4.0.
3 Materials Project. Electronic Structure Methodology. https://docs.materialsproject.org/methodology/materials-methodology/electronic-structure. Accessed July 23, 2026. Publicly available official documentation.
Why Work with CD ComputaBio?
Our workflows are modular, traceable, and structured around the customer's experimental decision. Each candidate is linked to its source data, model scope, uncertainty, verification status, and next validation step. We do not present computational ranking as proof of synthesizability or device performance; we identify the evidence needed to test those hypotheses.
To discuss your semiconductor target profile, available data, or process constraints, please contact us or submit the inquiry form below.
