---
configs:
- config_name: metal
  data_files:
  - split: train
    path: lithosim_path/metal_train_val.csv
  - split: test
    path: lithosim_path/metal_test.csv
- config_name: opc_metal
  data_files:
  - split: train
    path: lithosim_path/opc_metal_train_val.csv
  - split: test
    path: lithosim_path/opc_metal_test.csv
- config_name: via
  data_files:
  - split: train
    path: lithosim_path/via_train_val.csv
  - split: test
    path: lithosim_path/via_test.csv
- config_name: opc_via
  data_files:
  - split: train
    path: lithosim_path/opc_via_train_val.csv
  - split: test
    path: lithosim_path/opc_via_test.csv
- config_name: ood
  data_files:
  - split: test
    path: lithosim_path/ood.csv
license: afl-3.0
language:
- en
tags:
- code
pretty_name: litho
size_categories:
- 1B<n<10B
task_categories:
- feature-extraction
---

The benchmark of "LithoSim: A Large, Holistic Lithography Simulation Benchmark for AI-Driven Semiconductor Manufacturing"

The corresponding GitHub repo can be found at https://dw-hongquan.github.io/LithoSim/

## Data Construction

- 4 in-distributed dataset (OPC_Metal/Metal/OPC_Via/Via).
- 1 out-of-distribution (OOD) dataset.
- Each main dataset has a train_val and a test folder with compressed data file.
- Each set of data contains a source_simple.src description of the source, a layout.png, a mask.png, and a number of RI under different process variations.