Generic Lithography Modeling with Dual-band Optics-Informed Neural Networks
TimeThursday, July 14th11:37am - 12pm PDT
Location3007, Level 3
Event Type
Research Manuscript
Physical Design and Verification, Lithography and DFM
DescriptionLithography simulation is a critical step in VLSI design for manufacturability.
Legacy solutions with rigorous models are time consuming when seeking high simulation accuracy even equipped with various approximation techniques. Although machine learning solutions are investigated recently, these learning-based methods are either having limited usage scenario or lacking in modeling accuracy. To tackle these concerns, we present a dual-band neural network design that is compliant with optical physics beneath lithography, allows via/metal layer contour simulation at 1nmX1nm pixel resolution with any tile size. The model is 10X smaller than existing solutions when achieving even much better contour generation results.