GDSTK

Gdstk (GDSII Tool Kit) is a C++ library for creation and manipulation of GDSII and OASIS files. It is also available as a Python module meant to be a successor to Gdspy.

Key features for the creation of complex CAD layouts are included:

• Boolean operations on polygons (AND, OR, NOT, XOR) based on clipping algorithm
• Polygon offset (inward and outward rescaling of polygons)
• Efficient point-in-polygon solutions for large array sets

Typical applications of Gdstk are in the fields of electronic chip design, planar lightwave circuit design, and mechanical engineering.

Documentation

The complete documentation is available here.

The source files can be found in the docs directory.

Installation

C++ library only

The C++ library is meant to be used by including it in your own source code.

If you prefer to install a static library, the included CMakeLists.txt should be a good starting option (use -DCMAKE_INSTALL_PREFIX=path to control the installation path):

cmake -S . -B build
cmake --build build --target install

The library depends on zlib and qhull

Python wrapper

The Python module can be installed via pip, Conda or compiled directly from source. It depends on:

• zlib
• qhull
• CMake
• Python
• Numpy
• Sphinx, Read the Docs Theme, and Sphinx Inline Tabs (to build the documentation)

From PyPI

Simply run the following to install the package for the current user:

pip install --user gdstk

Or download and install the available wheels manually.

From source

Installation from source requires the build module (plus CMake and Ninja, for faster compilation):

pip install --user build

With that, simply build the wheel package using:

python -m build -w

This will create a dist directory containing the compiled .whl package that can be installed with pip.

Support

Help support Gdstk development by donating via PayPal or sponsoring me on GitHub.

Benchmarks

The benchmarks directory contains a few tests to compare the performance gain of the Python interface versus Gdspy. They are only for reference; the real improvement is heavily dependent on the type of layout and features used. If maximal performance is important, the library should be used directly from C++, without the Python interface.

Timing results were obtained with Python 3.11 on an Intel Core i7-9750H @ 2.60 GHz They represent the best average time to run each function out of 16 sets of 8 runs each.

Memory usage per object for 100000 objects: