Changelog

All notable changes to the COMPASS project are documented here. This project follows Semantic Versioning.

v0.1.0 -- Initial Release

Release date: 2025-02-10

The first public release of COMPASS, establishing the core simulation framework with multi-solver support, material database, and visualization tools.

Core framework

• Pydantic-based configuration schema (CompassConfig) with full validation
• Hydra/OmegaConf integration for YAML config loading and CLI overrides
• Internal coordinate system: micrometers for all lengths, degrees for external angles, radians for internal angles
• Light propagation convention: -z direction (air at z_max, silicon at z_min)

Geometry engine

• PixelStack and GeometryBuilder for solver-agnostic pixel structure definition
• Superellipse microlens profile with configurable height, radius, squareness, and CRA shift
• Multi-layer BARL (Bottom Anti-Reflective Layer) support
• Color filter with Bayer pattern (RGGB) and configurable metal grid
• Silicon layer with embedded photodiode and DTI (Deep Trench Isolation)
• Planarization (over-coat) layer
• Layer slice generation for RCWA (2D permittivity grids at each z-height)

Solvers

• SolverBase abstract interface for all EM solvers
• SolverFactory with lazy import and plug-in registration
• torcwa RCWA solver adapter (PyTorch, CUDA GPU)
• grcwa RCWA solver adapter (NumPy/JAX)
• meent RCWA solver adapter (NumPy/JAX/PyTorch multi-backend)
• fdtd_flaport FDTD solver adapter (PyTorch, CUDA GPU)
• Unified SimulationResult output format across all solvers

Numerical stability (5-layer defense)

• PrecisionManager: TF32 disable, mixed precision eigendecomposition
• StableSMatrixAlgorithm: Redheffer star product (no T-matrix divergence)
• LiFactorization: Li's inverse rule for high-contrast boundaries
• EigenvalueStabilizer: Degenerate eigenvalue handling, branch selection
• AdaptivePrecisionRunner: Automatic float32 -> float64 -> CPU fallback

Sources

• PlanewaveSource: Single wavelength, sweep, and list modes
• TE, TM, and unpolarized (TE+TM average) polarization support
• ConeIllumination: Exit pupil illumination model with CRA, F-number, angular sampling (Fibonacci, grid), and weighting functions (uniform, cosine, cos4, Gaussian)

Analysis

• QECalculator: Per-pixel quantum efficiency extraction
• EnergyBalance: R + T + A = 1 verification
• SolverComparison: Pairwise QE difference, relative error, runtime comparison

Runners

• SingleRunner: Single simulation execution from config
• SweepRunner: Wavelength sweep orchestration
• ComparisonRunner: Multi-solver comparison workflow
• ROISweepRunner: Sensor-position sweep with CRA interpolation and microlens shift

Visualization

• plot_pixel_cross_section: 2D cross-section views (XZ, YZ, XY planes)
• plot_qe_spectrum: QE vs wavelength per color channel
• plot_qe_comparison: Multi-result QE overlay with difference plot
• plot_crosstalk_heatmap: Cross-pixel energy distribution
• field_plot_2d: EM field distribution visualization
• 3D pixel stack viewer (PyVista/Plotly)

I/O

• HDF5 result storage with embedded metadata and config
• CSV/JSON export for post-processing

Materials database

• Built-in materials: air, silicon (Green 2008), SiO2, Si3N4, HfO2, TiO2, tungsten, polymer (n=1.56), color filter dyes (R/G/B)
• User-defined materials via CSV (wavelength, n, k)
• Analytical dispersion models: Sellmeier, Cauchy

Infrastructure

• Python 3.10+ with type hints throughout
• pytest test suite with unit and integration tests
• ruff linting (E, F, W, I rules)
• mypy static type checking
• Google-style docstrings