model-based ocd

**Model-Based OCD** is the **computational engine behind optical scatterometry** — using electromagnetic simulation (RCWA, FEM, or FDTD) to compute the expected optical response for a parameterized geometric model, then fitting the model parameters to match the measured spectrum. **Model-Based OCD Workflow** - **Geometric Model**: Define a parameterized profile (trapezoid, multi-layer stack) with parameters: CD, height, sidewall angle, corner rounding. - **Simulation**: Use RCWA (Rigorous Coupled-Wave Analysis) to compute the theoretical spectrum for each parameter combination. - **Library**: Build a library of pre-computed spectra spanning the parameter space — or use real-time regression. - **Fitting**: Match measured spectrum to library using least-squares or machine learning — extract best-fit parameters. **Why It Matters** - **Accuracy**: Model accuracy directly determines measurement accuracy — the model must faithfully represent the physical structure. - **Correlations**: Parameter correlations limit the number of independently extractable parameters — model complexity must be balanced. - **Floating Parameters**: Only a few parameters can "float" (be extracted) — others must be fixed or constrained. **Model-Based OCD** is **solving the inverse problem** — computing what the structure looks like by matching measured optical signatures to electromagnetic simulations.