Machine learning-assisted spectroscopic ellipsometry of chromium thin films for microalgae biosensing.

Rapid and label-free detection of microalgae is increasingly required for environmental surveillance and bio-industrial process control, where decisions must be made from subtle interfacial changes rather than from bulk concentration alone. In this work, chromium (Cr) thin films (20-75 nm) were deposited by RF magnetron sputtering and evaluated by spectroscopic ellipsometry (SE) to develop a thickness-optimized optical transducer for microalgae biosensing. Thickness-dependent optical constants (n, k) were extracted using a Drude-Lorentz dispersion model incorporating an effective-medium roughness layer. The 75 nm film exhibited the most bulk-like and spectrally stable response with comparatively low effective loss across the visible range, while 20-30 nm films showed larger deviations attributable to microstructure- and interface-mediated scattering contributions in the ultra-thin regime. Biosensing was implemented by forming Cr/PVA and Cr/PVA + microalgae stacks and quantifying the differential phase response. The 75 nm Cr/PVA platform delivered the strongest microalgae-induced modulation, exhibiting a Δ phase shift of 40.6° within 2.65-3.20 eV, thereby identifying a high-contrast spectral window for detection. Machine learning was required because Ψ-Δ spectra are high-dimensional, nonlinear, and strongly correlated, and multilayer spectral blending (Cr/PVA/biological loading) limits reliable thresholding and linear separation. A multitask deep neural network was trained to learn the coupled Ψ-Δ response for rapid prediction, and support vector machines were used for supervised discrimination of film stacks. By converting dense SE signatures into decision-ready labels on a thickness-optimized substrate, the proposed SE-ML framework advances an intelligent, non-destructive route for rapid microalgae screening and environmental diagnostics.