the platform
The Semiconductor Defect Modelling Platform
A GPU-accelerated defect-modelling engine for any semiconductor material. Machine-learning potentials paired with first-principles DFT, feeding a structured, growing defect database. The material changes as an input. The engine doesn't.
Proven across five lattice types: Nitride, oxide, arsenide, antimonide, phosphide. See the full method on the Services page.
InP
GaSb
GaN
GaAs
IGZO
5. Database entry & report
Results are added to the structured defect database and delivered as evidence for your team.
4. Formation energies & CTLs
Charge-transition levels and thermodynamic stability across the relevant chemical potential range.
3. Charged-defect DFT
Hybrid-functional validation with finite-size correction, potential alignment and chemical potential accounting.
2. Defect enumeration & ML screening
Vacancies, antisites, interstitials and complexes, triaged with GPU-accelerated ML potentials (MACE) on Isambard-AI.
1. Structure validation
Establish and validate the pristine crystal structure for your material system.
Technology
One engine, any semiconductor material
The core platform pairs GPU-accelerated machine-learning potentials with hybrid-functional DFT validation. Engineered and maintained in-house and not outsourced.
The method
Proven, and transferable
Validated across five materials spanning nitride, oxide, arsenide, antimonide and phosphide chemistries: InP, GaSb, GaN, GaAs, IGZO
Ga2O3 - next
SiC - next
SiO2 - next
AlN - next
Not calculations for hire
Every engagement strengthens the same reusable engine and the dataset behind it, not individual DFT calculations sold one at a time.
Technology
In development
Semiconductor Defect Data
Every calculation feeds a growing, structured defect database, queryable directly by a device team, a fab, or an investor.
Material & structure
Material, structure and phase: crystalline or amorphous
Defect properties
Defect type, site, charge state, formation energies, charge-transition levels, activation energies, charge carrier mobilities etc.
Provenance
Correction methodology, functional, workflow version and source. Proprietary, published, or experimental kept distinct.
Technology
Automated Simulation
The engine that runs structure validation, defect enumeration with ML screening, and charged-defect DFT. Standardised across materials rather than scripted per project.
What it handles
Pristine structure validation
Defect generation & enumeration
Hybrid-functional charged-defect DFT
Chemical-potential accounting
Finite-size correction
Potential alignment
SLURM/HPC portability
Technology
R&D
AI & Machine Learning
Machine-learning interatomic potentials (MACE) run on Isambard-AI to triage defect candidates before expensive DFT is ever committed.
The goal isn't to replace hybrid-functional DFT. It's to make sure DFT time is spent only on the candidates worth validating.
Where this is headed
As the defect database grows, the same models become the basis for active learning. Uncertainty-guided selection of which calculations to run next, closing the loop between simulation, data and prediction.
Technology
Frontier-level compute, matched to the workload
Tier 1 - CPU
ARCHER2 (and successor)
The CPU-architecture workhorse for hybrid functional, charged defect DFT - the calculations behind our highest-stakes client claims.
Tier 1 - GPU
Isambard-AI
GPU-native compute for ML-accelerated defect screening and high-throughput triage.
Tier 2
MMM Hub
Day-to-day project compute and methods development, reducing dependency on Tier 1 allocation timing.