phoxla 0.1.0a1

Differentiable photonic design and simulation on XLA accelerators

PHOXLA

PHOXLA ("FOX-luh") - Differentiable photonic design and simulation on XLA accelerators.

PHOXLA is a framework blueprint for building modern photonic inverse-design systems with JAX/Flax. The goal is to treat differentiable physics as a first-class deep learning component, so photonic simulation modules can be plugged into end-to-end trainable workflows.

Vision

• XLA-native photonic simulation and optimization (GPU/TPU first, CPU compatible)
• Differentiable physics modules usable as model layers/loss terms
• Inverse design pipelines integrated with standard deep learning tooling
• Extensible interoperability with JAX-based solvers (including FDTDX-style workflows)

Current Status

This repository is an initial blueprint and governance scaffold. Core simulation and training modules will be added incrementally.

Planned Package Layout

• src/phoxla/sim: differentiable photonic simulation core
• src/phoxla/nn: NN-facing modules and differentiable operators
• src/phoxla/inv: inverse design workflows and optimizers
• src/phoxla/integrations/fdtdx: interoperability layer

Repository Bootstrap

See:

• docs/REPO_LAUNCH_GUIDE.md
• docs/REPO_ABOUT_AND_TOPICS.md
• docs/PACKAGE_NAMESPACE_PLAN.md
• docs/PYPI_CLAIM_PLAYBOOK.md
• docs/NAMESPACE_PRIORITY.md
• docs/ARCHITECTURE_BLUEPRINT.md
• docs/COMMIT_MESSAGE_GUIDE.md
• ROADMAP.md

Quick Release Commands

make venv
make release-check

# then with PyPI token in env:
make publish-pypi

License

Apache-2.0 (LICENSE)