Questions about pixelILT and levelsetILT

[mmx-Lv]

I have noticed that the dataset contains both pixelILT and levelsetILT. I would like to ask which one is closer to the actual fabricated mask, and what is the definition of levelsetILT?
Why does pixelILT seem to have many assist features around the main patterns? Is this due to the reasons related to optical imaging? Moreover, through my actual comparison, the lithography simulation results of pixelILT are indeed better.
Thank you very much for your time and valuable insights. I would greatly appreciate it if you could spare a moment to address my questions.

[shelljane]

Thank you for your interest in our work!

The extra assist features you see with PixelILT are expected and, in fact, one of ILT's strengths. They emerge naturally because the lithography simulator's gradient drives the optimization; suppressing them (e.g., by lowering the learning rate or reducing iterations) would also degrade the quality.

Level-set ILT produces fewer assist-features by design. We initialize far-from-boundary pixels to large values, so the optimizer modifies only the boundary regions first. This conservative update keeps the shapes regular but sacrifices the extra degrees of freedom.

In my opinion, for actual fabrication, PixelILT's results may require a more advanced mask writer, while LevelILT's results may be more cost-friendly. Both of them have their own advantages.