Spectroscopic Reflectometry

Spectroscopic reflectometry, metal-film thickness measurement, transparent-film thickness measurement, and multilayer-film metrology all extend thin-film thickness measurement to specific film types that standard single-layer ellipsometry handles poorly — a mirror-like metal surface, a film thick or complex enough that simple reflectometry models struggle, or a stack of several films where thickness has to be separated layer by layer.

Metal films need their own measurement approach because they're opaque, which breaks the transmitted-light assumption most transparent-film techniques rely on. Standard ellipsometry and reflectometry models for dielectric films often account for light transmitting partway through the film and reflecting off the substrate below; a metal film is opaque enough that essentially no light reaches the substrate at all, so metal-film thickness measurement has to work entirely from the metal's own reflected signal and known optical constants rather than any substrate-interaction term — a fundamentally different measurement model, not just a parameter change.

Multilayer-film metrology's real challenge is separating correlated thicknesses when several layers all interact with the same measurement signal at once. Measuring one film's thickness in isolation is a well-posed problem; measuring three or four stacked films' individual thicknesses from one reflectance or ellipsometry spectrum means fitting a model with multiple unknowns to one measured signal, which can leave some parameter combinations ambiguous — thinner-layer-A-plus-thicker-layer-B looking nearly identical in the measured signal to thicker-A-plus-thinner-B. Careful model constraints, known-reference layers, and sometimes complementary techniques are needed to resolve that ambiguity reliably.

Spectroscopic reflectometry, metal-film methods, and multilayer-film metrology all extend the same basic optical thickness-measurement idea to situations the simplest single-layer transparent-film model can't handle on its own — opacity in one direction, multiple interacting unknowns in the other — and getting either one wrong produces a thickness number that looks plausible but is quietly incorrect.