LithoBench-PDE

High-fidelity 3D reference simulations of photolithography for photomasks from the LithoBench dataset, with ground-truth input-output pairs for three PDE learning tasks.

Tasks

Task	PDE	Input	Output	Shape
Mask diffraction	Maxwell's equations	M photomask	E diffracted near field	[H, W] → [2, 2, H, W] (complex64)
Post-exposure bake	Reaction-diffusion equation	h photoacid concentration	m deprotection image	[25, H, W] → [25, H, W]
Development	Eikonal equation	R development rate	T development time	[25, H, W] → [25, H, W]

Each sample is a .pt dictionary with the keys M, E, h, m, R, T.

Photomask categories

Category	Samples	Size	Description
Metal_I	1,600	512×512	Curvilinear metal photomasks
Metal_T	1,600	512×512	Rectilinear metal target layouts
Metal_P	61	512×512	Packed Metal
Standard	216	512×512	Standard line-space patterns and variants
Contact_I	163	512×512	Curvilinear contact photomasks
Contact_T	163	512×512	Rectilinear contact target layouts
Metal_L	5	512×512 to 1024×1024	Large-area metal masks

Total: 3,808 samples (~436 GB).

Files

Path	Content
LithoBench_PDE/<category>/*.pt	Samples
LithoBench_PDE/seed_manifest.json	Exposure seed of every sample
checkpoints/SONO/	Trained NeuroLitho neural operators
fig_data/	Raw data of the figures
vis/	Visualization code and notebooks
LithoBench_PDE.py	PyTorch dataset class

Usage

import torch
from huggingface_hub import hf_hub_download, snapshot_download

# one sample
path = hf_hub_download("AISDL-SNU/LithoBench-PDE", "LithoBench_PDE/Contact_I/INV_X8__0_0.pt", repo_type="dataset")
sample = torch.load(path, map_location="cpu")

# one category
snapshot_download("AISDL-SNU/LithoBench-PDE", repo_type="dataset",
                  allow_patterns="LithoBench_PDE/Contact_I/*", local_dir="./data")

Code

The NeuroLitho code is provided as a Code Ocean capsule. The DOI will be added upon publication.

License

• Dataset (LithoBench_PDE/, checkpoints/, fig_data/): CC-BY-4.0
• Software (LithoBench_PDE.py, vis/): MIT License
• Layouts from LithoBench: Copyright (c) 2023 Su Zheng, MIT License (LICENSE-LithoBench)

Citation

TBA