ilt convergence

**ILT Convergence** is the **convergence behavior of Inverse Lithography Technology optimization** — ILT solves for the optimal mask pattern using gradient-based optimization, requiring many iterations to converge to a mask shape that maximizes the patterning process window. **ILT Convergence Details** - **Objective**: Minimize $sum_{(x,y)} |I(x,y) - I_{target}(x,y)|^2$ summed over process window conditions. - **Gradient Descent**: Compute the gradient of the cost function with respect to mask transmission at every pixel. - **Iterations**: ILT typically requires 50-200+ iterations — far more than rule-based OPC. - **Constraints**: Mask manufacturability rules (MRC) are enforced during or after optimization — adds complexity. **Why It Matters** - **Computation**: ILT is vastly more compute-intensive than OPC — GPU acceleration is essential for full-chip ILT. - **Quality**: ILT often produces superior process windows compared to rule/model-based OPC — worth the computational cost. - **Local Minima**: Non-convex optimization can get trapped in local minima — initialization and regularization matter. **ILT Convergence** is **the optimization journey to the ideal mask** — iteratively refining mask pixel values until the patterning objective function converges.