From nominal thickness to manufacturing variation

A nominal design gives one answer for one set of thicknesses. Manufacturing analysis asks how that answer changes when the deposited layers differ from the nominal values.

Start with a sensitivity check

This example keeps Si at 20 nm and perturbs Zr around 100 nm. It uses a freestanding Zr/Si stack, normal incidence, unpolarized illumination, smooth interfaces, and densities of 6.52 and 2.33 g/cm³. Only Zr thickness changes.

Define errors before estimating a pass rate

1. Choose the performance specification: the quantity, wavelength or band, and acceptance threshold.
2. Specify what the error represents. An absolute error in nanometers and a percentage error are different models.
3. Distinguish independent layer scatter from a shared calibration offset or a process drift.
4. Choose the sample count and reproducible seed where supported; check whether conclusions stabilize with more samples.
5. Record constraints and any invalid samples rather than silently discarding them.

The three deterministic rows above have no probability distribution and cannot establish a manufacturing yield. A Monte Carlo pass rate is conditional on the error model, optical data and specification supplied.

Compare nominal and robust candidates

A design with the best nominal response may not be the most reliable under an assumed thickness distribution. In Pro, compare nominal performance with the manufacturing analysis, and use robust optimization when that tradeoff is part of your design objective.

Retain both candidates and the analysis settings. Process qualification still requires measurements; simulation can identify sensitivity and guide that work.

Explore the robust-optimization workflow →