Installation¶

Requirements¶

Python 3.11+
Palace EM simulation software

Optional:

COBRA installed/importable if you use ORCA surrogate outputs in circuit optimization

Note

ORCA supports Python 3.11 to 3.14.

Clone Repository¶

git clone https://github.com/DI-PASSIONATE/ORCA
cd ORCA

Install ORCA¶

Option A: uv (recommended)Option B: venv + pip

curl -LsSf https://astral.sh/uv/install.sh | sh
uv sync

or explicitly:

uv venv --python 3.13
source .venv/bin/activate
uv pip install -e ".[train]"

To pick a specific PyTorch build, use uv sync with one of the build selectors from pyproject.toml instead of uv pip install:

uv sync --extra train --extra cpu     # CPU-only PyTorch wheels
uv sync --extra train --extra cu130   # CUDA 13.0 wheels (driver >= 580)
uv sync --extra train --extra cu126   # CUDA 12.6 wheels for older drivers

python3 -m venv .venv
source .venv/bin/activate
pip install -U pip
pip install -e ".[train]"

pip installs the PyPI build of PyTorch (CUDA-bundled on Linux). For a CPU-only or a specific CUDA build, install torch first with the command from PyTorch.org; pip then keeps that version.

Simulation-only install

PyTorch and the other model-training packages are an optional extra. If you only want to generate layouts and run Palace simulations — on an HPC cluster, say — drop the [train] extra: pip install -e .. GDSGenerator, GDSConverter and PalaceSimulator work without it; ModelTrainer, OnnxExporter and ModelTester report the missing packages when used.

Verify Setup¶

Run the following check in your activated environment:

orca

Expected behavior: ORCA GUI starts.

To verify the script workflow:

python examples/main.py

Warning

examples/main.py requires a working Palace installation and a valid geometry configuration.

External Tool Notes¶

Palace¶

ORCA uses Palace as the EM simulation backend.
Install Palace by following the official Palace installation instructions.
Palace can be run directly or via an Apptainer/Singularity container.

COBRA (Optional)¶

COBRA consumes ONNX surrogate models exported by ORCA for circuit-level optimization.
See COBRA documentation for details.

Next Steps¶

Continue with Getting Started -> Quickstart for a first run.
Use User Guide -> Custom Classes to bring your own geometry.