Skip to content

OpenLithoHub

Open-source computational lithography benchmarking and workflow toolkit for advanced EUV/curvilinear mask processes.

Looking for the project home, the live leaderboard, or the hackathon sign-up? Those live on openlithohub.com. This site is the technical reference — install, CLI, Python API, contributor guides, and RFCs.

Design vs no-OPC aerial vs ILT-optimized aerial. Generated by scripts/generate_hero_figure.py.

Overview

OpenLithoHub provides a unified evaluation and workflow framework for computational lithography research. It bridges the gap between academic tensor-based optimization and industrial mask manufacturing.

┌─────────────────────────────────────────────────────────────────┐
│                       OpenLithoHub                              │
├─────────────┬──────────────┬──────────────┬───────────┬─────────┤
│  Data Layer │  Benchmark   │   Workflow   │ Vis & UX  │   CLI   │
│ LithoBench  │  EPE/PVBand  │ Tiling/Stitch│ Paper figs│ eval    │
│ LithoSim    │  MRC/DRC     │ Contour Ext. │ Jupyter   │ optimize│
│ Transforms  │  Stochastic  │ OASIS Export │ EDA bridge│leaderbd │
│ Dummy gen.  │  Shot Count  │ B-spline Fit │           │         │
└─────────────┴──────────────┴──────────────┴───────────┴─────────┘

Key Features

  • Unified dataset access — single interface to LithoBench, LithoSim, GAN-OPC, ICCAD'16 hotspot, ASAP7, FreePDK45 + NanGate OCL, and ORFS-routed RISC-V layouts
  • Hermetic dummy layouts — generate_dummy_layout for CI / Colab without network or KLayout
  • Standardized metrics — EPE, PV Band, shot count, EUV stochastic robustness, hotspot detection (recall / precision / F1)
  • Manufacturing compliance — MRC/DRC rule checking as hard-fail gating
  • OASIS workflow — end-to-end pipeline from tensor to fab-ready mask (manhattan & curvilinear)
  • EDA bridge — minimal Calibre nmDRC / IC Validator templates emitted alongside OASIS exports
  • Paper-ready visualization — IEEE / SPIE column-width contour figures via openlithohub.vis
  • Model-agnostic evaluation — plug any OPC/ILT model into the benchmark via a minimal interface
  • JIT-accelerated forward model — Hopkins/SOCS forward is wrapped with torch.compile by default for kernel-fusion speedups on PyTorch 2.x; use --no-compile to disable
  • PDK-aware synthetic layouts — openlithohub.synth / openlithohub synth run produce ASAP7 / FreePDK45 layouts that pass MRC by construction, for hermetic CI and quick iteration
  • Public leaderboard — track SOTA results across models, datasets, and process nodes