regression-based ocd

**Regression-Based OCD** is a **scatterometry approach that iteratively adjusts profile parameters to minimize the difference between measured and simulated spectra** — using real-time RCWA simulation and nonlinear least-squares fitting instead of a pre-computed library. **How Does Regression OCD Work?** - **Initial Guess**: Start with estimated profile parameters (from library match or nominal design). - **Simulate**: Compute the optical spectrum for current parameters using RCWA. - **Compare**: Calculate the residual between measured and simulated spectra. - **Optimize**: Use Levenberg-Marquardt or other nonlinear optimizer to adjust parameters. - **Iterate**: Repeat until convergence (typically 5-20 iterations). **Why It Matters** - **Flexibility**: No pre-computed library needed — handles arbitrary parameter ranges and new structures. - **Accuracy**: Can explore parameter space more finely than discrete library grids. - **Combination**: Often used after library matching for refinement ("library-start, regression-finish"). **Regression-Based OCD** is **real-time fitting for profile metrology** — iteratively adjusting simulations to match measurements for precise dimensional extraction.