FabGAN-ID Benchmark: Fabrication-Aware Inverse Design of Variable-Index, Multilayer Photonic Sensor Front-Ends

# FabGAN-ID Benchmark: Fabrication-Aware Inverse Design of Variable-Index# Multilayer Photonic Sensor Front-Ends Companion dataset for:T. M. Mahim et al. "FabGAN-ID: Learned Generative Process Twins in a Fully Differentiable Fabrication Loop forYield-Aware Inverse Design of Multilayer Photonic Sensor Front-Ends,"submitted to IEEE Sensors Journal. ## Contents ### benchmark_designs.npz (design benchmark)| array | shape | description ||---------|-------------|----------------------------------------------------|| d_um | (300, 20) | layer thicknesses [um], range [0.020, 0.120] || n0 | (300, 20) | layer refractive indices at lambda0 = 550 nm, || | | range [1.6, 2.4] (HDPCVD variable-index SiNx) || T | (300, 161) | exact transmittance spectra (validated transfer- || | | matrix solver, normal incidence) || lam_um | (161,) | wavelength grid [um], 0.40-0.80 | `benchmark_designs.csv` contains the same 300 recipes (columns d0..d19,n0..n19) for spreadsheet access. ### process_traces.npz (historical process traces)| array | shape | description ||----------|------------|----------------------------------------------------|| recipe_d | (400, 20) | commanded layer thicknesses [um] || recipe_n | (400, 20) | commanded layer indices at lambda0 || fab_d | (400, 20) | fabricated thicknesses under the held-out || | | ground-truth process || fab_n | (400, 20) | fabricated indices under the same process | 200 recipes x 2 runs each. The generating process (systematic rate bias andindex drift, inter-layer intermixing, AR(1)-correlated design-conditionalthickness noise, skewed log-normal index noise, rare particulate events) isspecified in the article's Supplementary Sec. S-II and is intended to betreated as HELD OUT: methods should learn only from these traces and bescored against fresh draws. ## ProvenanceMaterial dispersion derives from the public-domain (CC0) refractiveindex.infodatabase (Si3N4: Luke et al., Opt. Lett. 40, 4823, 2015; SiO2: Malitson,JOSA 55, 1205, 1965). Design ranges follow the variable-index SiNx platformof Yesilyurt et al., Nanophotonics 12, 993 (2023). All numerical data weregenerated by the authors; no experimental wafer data are included. ## LicenseCreative Commons Attribution 4.0 International (CC BY 4.0). ## CitationPlease cite the article above and this dataset (DOI assigned by Zenodo).