!! Still under research & development !!
Modeling the optical response of diffractive structures using Rigorous Coupled-Wave Analysis (RCWA) integrated with Machine Learning. The RCWA engine is TORCWA (GPU-accelerated and differentiable).
v6-inverse-design/— physics-informed inverse design of a 1D sine-corrugated TiN grating. A neural surrogate predicts the RCWA response, then inverse design recovers the grating geometry(amp, per)that matches a target response spectrum, with optional refinement through the differentiable RCWA solver. See its README.
archives/— earlier research versions (v0–v5): hyper-network weight prediction, order-convergence extrapolation, etc. Kept for reference; their data pipelines contained bugs (e.g. all 32 S-parameter columns identical) that are fixed inv6-inverse-design.
- Python 3.10 or higher
- PyTorch (CUDA optional)
- Required libraries (listed in
requirements.txt)
Clone the repository (including submodules):
git clone --recurse-submodules https://github.com/Alexin-CH/DiffractML.git
cd DiffractMLInstall the required dependencies:
makeSee v6-inverse-design/README.md for the current project workflow
(generate → train → design). Regression tests:
cd v6-inverse-design && python -m pytest tests/