How openopc/openilt works
Project Type and Purpose
• OpenILT is an open-source research platform for inverse lithography technology (ILT).
• It decouples the ILT workflow into four orthogonal stages—simulation, initialization, optimization, evaluation—so researchers can swap in/out new methods easily.
High-Level File/Directory Structure
• benchmark/ : Input designs (ICCAD2013 .glp, .gds)
• config/ : Text configuration files for different simulators & ILT methods
• kernel/ : Precomputed optical kernels & scale factors for litho simulators
• opc/ : Classical OPC implementations (ICCAD13, simple OPC)
• pycommon/ : Shared utilities (GLP/GDS reader, settings, logging)
• pylitho/ : Lithography simulators (exact, simple) as PyTorch Modules
• pyilt/ : ILT solvers & support code (initializers, level-set, multilevel, SimpleILT, evaluation)
• utils/ : GDS‐specific helpers (layout conversion, polygon ops)
• tmp/ : Output directory for masks, figures, intermediate results
• requirements_*.txt, LICENSE, README.md
Main Components and Their Interactions
A. Input/Preprocessing
• Design files (.glp or .gds) loaded via pycommon (readGDS, glp)
• Configuration loader (pycommon/settings) parses config/*.txt
B. Initialization
• Initializer classes in pyilt/initializer.py produce:
– target image (binary)
– initial mask parameters
C. Lithography Simulation
• pylitho/LithoSim (exact.py, simple.py)
– Loads optical kernels (kernel/) and config
– Simulates aerial & printed images at nominal/max/min process corners
D. ILT Optimization
• pyilt/SimpleILT (and other pyilt solvers)
– Accepts target & params
– Feeds params (mask) into LithoSim → printed images
– Computes losses (L2, PVBand, optional EPE)
– Backpropagates gradients via PyTorch optimizers
– Updates mask parameters
– Iterates until convergence → best mask
E. Classical OPC
• opc/*.py provide alternative mask correction flows (iccad13.py, simpleopc.py)
F. Evaluation
• pyilt/evaluation.py: Basic, EPEChecker, ShotCounter
• Outputs metrics: L2, PVBand, EPE, Shots
G. Output
• Final mask images saved in tmp/
• Figures in figs/
Architectural Patterns & Design Principles
• Modular, plug-and-play stages (Simulation, Initialization, Optimization, Evaluation)
• Clear separation of concerns: file I/O (pycommon), simulation (pylitho), optimization (pyilt), evaluation (pyilt + thirdparty)
• Use of framework (PyTorch) to unify differentiable components
• Configuration-driven: all numeric parameters or kernel selections live in config/ and kernel/
Technologies, Frameworks, Libraries
• Python 3.8–3.10, PyTorch (≥1.8) for GPU-accelerated tensor ops and auto-diff
• NumPy, SciPy for numeric routines
• Custom binary kernel files for litho simulation
• Standard Python packaging and requirements files