Inverse Lithography: manufacturable mask optimization

Problem

For a TU/e modeling week with ASML, I worked on an inverse lithography problem: optimize a mask pattern to match a target under a forward model. Pixel-by-pixel optimization is flexible, but it is memory-intensive and can produce non-manufacturable solutions.

My role

• Proposed a geometric parameterization of the manufacturable mask space

• Helped formulate the optimization problem and interpret the results.

• Mentored MSc students and translated between domain and optimization language.

Project Setting

This work was part of a TU/e modelling week with ASML.

Approach

I proposed replacing pixel variables with rectangle positions and sizes, turning the problem into a lower-dimensional optimization problem with direct geometric meaning. The parameters were optimized in Python against the forward-model loss.

Takeaway

This approach reduced computational complexity and enforced manufacturable structure, at the cost of some fidelity relative to a pixel-level baseline. The project highlighted a practical trade-off between accuracy and computational / manufacturing feasibility.

Tools

Python; gradient-based optimization; mathematical modeling.

Context

This work was part of a TU/e modeling week with ASML. My contribution was primarily in problem formulation, optimization design, and technical mentoring, rather than ownership of the full downstream implementation.