Answers are grounded in this paper and cite their sources.
Constructing a dense focus–exposure matrix (FEM) for lithographic process-window analysis is often limited by the cost of acquiring (real or synthetic) scanning electron microscopy (SEM) images at every focus and dose condition. We propose a layout-to-SEM generative surrogate, PADE (process-adaptive denormalization), that injects the lithographic process condition (exposure dose, focus level, or both) into every denormalization layer of an image-to-image translation backbone, paired with a variational latent that supplies the cell-specific SEM texture. It can synthesize SEM images corresponding to unseen intermediate conditions in a single forward pass by interpolating the per-cell encoded latent and feeding the target condition directly to the generator after training on a sparse grid of focus–dose anchors only. On a 3×3 focus–dose anchor grid over 40 line–space cells rendered by a physics-based simulation pipeline, the generated SEM images successfully reconstruct the simulated FEM with accurate metrology metrics. For example, the critical dimension measured from the generated SEM images precisely tracks the simulated ground-truth values with a Pearson correlation of r=0.999 and a root-mean-square error of 1.72 nm. The exposure latitude extracted from the generated Bossung surfaces also agrees well with the ground truth at r=0.97. Once trained, PADE can produce SEM images almost in real time (7108 times faster than conventional physics-based simulation), enabling continuous Bossung-curve extraction and comprehensive process-window analysis in process development.
No takes yet. Share an insight, caveat, or question.
Lee et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: