A Python validation pipeline for checking whether simulated diffractive optical designs are physically buildable and measurable.
Developed a Python physical-validation tool to assess whether trained diffractive phase masks and phase-surface stacks are physically buildable and measurable under propagation constraints, noise and fabrication non-idealities.
Implemented the Angular Spectrum Method propagation engine.
Modelled field evolution through stacked periodic diffractive layers and periodic diffractive arrays, with a geometry structurally analogous to cascaded metasurface sheets.
Implemented five sequential checks: optical aliasing and ASM bandwidth/band limits; energy conservation; photon budget; per-pixel/per-element SNR; and Monte Carlo fabrication tolerance.
Validated field distributions, energy conservation and array tolerance.
Implemented Monte Carlo analysis with 200 trials and 7 perturbation types.
Implemented gradient-based phase-profile optimisation through a differentiable wave-propagation model.
Integrated Monte Carlo fabrication-tolerance analysis directly into the gradient-based optimisation loop, treating robustness to layer-thickness and etch-depth variance as initial design constraints rather than post-hoc checks.
Repository: github.com/muhammadzareii/d2nn-feasibility.