An open-source, multi-physics simulator for post-silicon and 3D-stacked semiconductor devices. It predicts which 2D channel materials, device geometries (planar, GAA nanosheet, CFET, monolithic 3D) and stacking choices are worth fabricating, and ranks every candidate on performance, thermal limits and manufacturability together, before anything goes near a fab.
The core rule of the project: never present a number a solver didn't
actually produce. Every result carries provenance (is_stub, backend,
citation), and every material property that isn't computed cites a named
paper or database.
Project write-up: https://shobhittripathi.com/quantum-foundry/
The June-2026 imec/ASML/TSMC 300mm 2D-material CMOS demonstration established MoS₂ as the strongest available n-type channel and WSe₂/WS₂ as the strongest p-type, with no single material doing both well. Quantum Foundry's sweep and Pareto engine searches material, dielectric and stacking combinations for that n/p-asymmetry problem, while staying general enough for other materials and geometries.
-
Materials database: a typed schema (bandgap, effective masses,
mobility, contact resistance
$R_c$ , thermal conductivity, work function, defect density) seeded from JARVIS-DFT and C2DB, cross-checked against the Materials Project for gate dielectrics ($HfO_2$ ,$ZrO_2$ ). Every value carries a per-field citation. -
Multi-physics pipeline: each layer sits behind a Protocol with a fast
stub backend and a real-solver backend:
- Ab-initio: Quantum ESPRESSO + Wannier90
- Device transport: DEVSIM (drift-diffusion) + Kwant (atomistic NEGF)
- Thermal/mechanical: FEniCSx (3D FEA) + Elmer (coupled thermo-mechanical) + HotSpot (fast compact-model pre-filter)
- Manufacturability/yield: custom Monte Carlo plus closed-form Poisson/Murphy yield models. No mature open-source package exists for this layer.
- Sweeps and Pareto optimization: parameter sweeps across materials, stacking and dielectrics, with PPA + thermal + yield trade-off reports.
See docs/ARCHITECTURE.md for the design and
docs/DECISIONS.md for why each choice was made.
Requires Python 3.11+. The stub pipeline needs no heavy solvers:
uv venv --python 3.11 .venv && source .venv/bin/activate # or python3.11 -m venv .venv
pip install -e ".[dev]"
qf --help
qf run MoS2_monolayer --backend stub
pytest -m "not requires_real_solver"See examples/ for runnable scripts.
Real backends (Quantum ESPRESSO, Wannier90, DEVSIM, Kwant, FEniCSx,
HotSpot) need conda/mamba: run scripts/setup_env.sh, which builds the
quantum-foundry environment from environment.yml, clones HotSpot at a
pinned commit and installs gmsh's headless build for mesh generation. Elmer
runs in a container (docker/elmer.Dockerfile). Before any
--backend real run, qf preflight reports what is actually available and
refuses to proceed silently if something required is missing.
| Layer | State |
|---|---|
| Materials database | 10 cited 2D materials and dielectrics |
| Validation/benchmarks | 6 literature benchmarks, each anchored to a figure/table |
| Transport | Kwant NEGF real; DEVSIM backend written, needs a BLAS-equipped runner |
| Thermal | HotSpot compact model real (incl. multi-tier stacks with Kapitza interfaces); 3D mesh generation real; FEniCSx/Elmer FEA designed, not built |
| Stacked devices (CFET/GAA/M3D) | per-tier orchestration and multi-tier thermal real; coupled 3D electrostatics not built |
| Yield | Monte Carlo + Poisson/Murphy real |
| Ab-initio | designed, not built |
Anything not marked real runs through a stub and is labelled as such in every
output. See docs/ROADMAP.md for what's next.
Contributions are welcome. Please read CONTRIBUTING.md
(every commit needs a DCO sign-off) and the
Code of Conduct. Report security issues privately as
described in SECURITY.md.
If you use Quantum Foundry in research, please cite it using
CITATION.cff.
Apache-2.0. See LICENSE, NOTICE (including CC BY 4.0
attribution for bundled materials data) and
THIRD_PARTY_LICENSES.md.