Reproduce “Physics-Constrained Adaptive Neural Networks Enable Real-Time Semiconductor Manufacturing Optimization with Minimal Training Data” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Rubén Darío Guerrero
Rubén Darío Guerrero developed a physics-constrained adaptive neural network framework that optimizes EUV lithography photomasks, achieving an average Edge Placement Error of 1.369 nm across 19 patterns. This approach operates with 90% less training data and offers a 15-20x inference speedup over rigorous electromagnetic solvers.
Key insights
- Ablation studies identified optical blur (controlled by a learnable parameter) as the singular dominant physics effect, contributing 50-80% of the accuracy improvements across all evaluated geometries.
- The framework's learnable physics parameters enable automatic calibration of electromagnetic approximations, facilitating robust cross-geometry generalization with substantially less training data.
- The methodology's effectiveness is ultimately bounded by the physical resolution limits of EUV lithography, demonstrating diminishing returns for features below 8-12 nm.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.