SCALE: Self-Supervised Layout Generation for Sub-2nm DRV Fixing
July 27, 2026
SCALE uses a self-supervised framework to fix design-rule violations in advanced semiconductor nodes by serializing multi-layer layout geometry into structured text. The fine-tuned model learns to reconstruct masked polygons from BEOL context without requiring explicit violation labels during training.
HOW THIS AFFECTS YOU
●
builderThis introduces a method for automating local place-and-route tasks using language models.
●
researcherThe serialization of geometric layouts into text for LLM training offers a new approach to EDA.