LibrePDK

The LibrePDK is the library driving Danube River and the Pad Cell Generator

ALibrePDK screen shot

It is responsible for generating discrete parts with specific parameters for a specific process.

The properties of the parts can be optimized by utilizing the calibration values extracted from the measurements of taped out Danube River test wafers.

Adding a new technology

Technologies currently supported can be found in the technologies subfolder.

https://gitlab.libresilicon.com/generator-tools/librepdk/-/tree/master/LibrePDK/technologies?ref_type=heads

New technologies can be added by modifying scripts/update_technologies.sh and adding a tech.python script to the technologies folder.

After that, LibrePDK should be capable of auto discovering the new process after running the update script.

Dependencies

Ubuntu

We are using Ubuntu 25.04 in our Docker containers

Simply run

apt-get update && apt-get install -y --no-install-recommends \
    build-essential cmake git gcc g++ bison flex libfl-dev libeigen3-dev autoconf automake libtool \
    libglpk-dev zlib1g-dev libcairo2-dev liblemon-dev libsparsehash-dev pkg-config curl ca-certificates cargo \
    rustc sqlite3 bash imagemagick libcairo2-dev tcllib tklib make g++ libreadline-dev libblas-dev \
    z3 texlive-latex-recommended unzip glpk-utils libglpk-dev git wget mmv libglu1-mesa-dev libcurl4-gnutls-dev rsync \
    pdf2svg libsqlite3-dev git-lfs clang llvm libxaw7-dev time tcl8.6-dev tk8.6-dev tcl software-properties-common vim \
    gcc libz-dev ruby ruby-dev libgit2-dev libgit2-glib-1.0-dev curl sudo python3-full python3-dev python3-pip checkinstall \
    gosu

Boost

In order for Limbo to successfully compile we need to install Boost.

We are currently using boost 1.91.0, of which the tar ball can be obtained here https://archives.boost.io/release/1.91.0/source/boost_1_91_0.tar.gz

With the following command Boost can be installed quickly

./bootstrap.sh \
    --prefix=/usr \
    --with-python-version=3.13

./b2 install

Note that the Python version we compile boost for is 3.13, because LibrePDK hasn't been tested with any later version.

ngspice

For the simulation and characterization of standard cells you will need the ngspice frontend and backend library

Currently we are using version 44.2 of ngspice, the tar ball can be obtained from here: https://download.industrysoftware.automation.siemens.com/open-source/ngspice-44.2.tar.gz

Backend configuration

The backend library is needed for Charlib in order to run characterization of our cells

./autogen.sh
./configure --with-ngshared --enable-xspice --enable-cider --enable-openmp --disable-debug
make -j$(nproc)
make install

After compilation and installation you should have a libngspice.so in your ldpath which charlib can pick up on and load it.

Frontend configuration

This will compile and install the ngspice CLI binary which is needed when you wanna simulate your circuits.

./autogen.sh
../configure --with-x --with-readline=yes --enable-xspice
make -j$(nproc)
make install

ngspice is also used when loading and importing new technologies into LibrePDK

OpenVAF models

IHP's SG13G2 technology node uses OpenVAF models for the ngspice simulation tool.

The following script will make sure that rust and the OpenVAF tool are present and then compiles the models into the osdi format and places them into the technology directory ready to be used by LibrePDK.

Simply run the following script and confirm the installation by checking for the LibrePDK/technologies/spice/SG13G2/devices/*/*.osdi files.

This requires the LibrePDK repo to already have been cloned and updated as shown in.

./scripts/update_ngspice_extensions.sh

LP solver

Google now officially runs the project and you can get the most recent version from GitHub

Install is by cloning and building it

git clone https://github.com/lp-solve/lp_solve
pushd lp_solve/lpsolve55
rm -rf bin/ux64
sh ccc
popd

Then you can copy the shared object file in solve/lpsolve55/bin/ux64 into your /usr/lib64 and copy the headers with

mkdir /usr/include/lpsolve
cp lp_solve/*.h /usr/include/lpsolve/

OR, you can install the system package and devel package with your package manager

Lemon

That library has been developed by a Hungarian university which doesn't maintain their Mercurial setup. Best approach is to use the version you find in your distribution

Limbo

The official version of Limbo has been a total mess when it comes to building libs and linking them. I had to make some severe modifications which makes CMake properly build shared object files and detects the system wide installation of the dependencies using proper CMake detection functions

Just run

git clone https://gitlab.libresilicon.com/leviathan/limbo.git
mkdir Limbo/build
pushd Limbo/build
cmake ..
make
make install
popd

Installation

Before you start installing LibrePDK, you first need to install its dependencies so that its C++ extensions and other C++ tools compile successfully.

After that, please clone the repo.

pip install https+git://gitlab.libresilicon.com/generator-tools/librepdk.git

Don't forget to make sure that all the submodules and their submodules are cloned

git submodule update --init --recursive

For placement of discrete componentes used in more complex components like Driver Circuits, OpAmps, etc. IdeaPlaceExPy is being used.

IdeaPlaceExPy requires the Python system headers to be installed and the virtual env has to match the Python version with which it was compiled.

Using LibrePDK in a Virtual Environment (FINAL SETUP)

It is recommended to use LibrePDK in a Python virtual environment to avoid dependency conflicts with system-wide Python packages.

After you've installed all the below dependencies the recommended way of installing the remaining dependencies is to run

uv sync --no-cache

Building your own Docker images

Inside of the LibrePDK source folder you will find a script where you can uncomment and comment specific build stages of the multistage Docker build covering all the dependencies

Check out the script scripts/build_docker_images.sh

Components

LibrePDK provides generators for the basic components usually found within a VLSI/ULSI design, such as resistors, capacitors, diodes and transistors.

Capacitors

LibrePDK can calculate the specific geometry based on the device rules and available parameters for generating any desired target capacitance value. Below a 50pF capacitor can be seen. You will notice the enormous dimensions of the structure.

Usually we deal with femto Farad in VLSI design so you should never be in a situation where you have large capacitors on your chip.

LibrePDK still can generate you a device, you just won't be happy about it.

Resistors

There's two types of resistor structures available: Meander and strip resistors

LibrePDK automatically adds a guard ring around any resistor which should be on a well

The meander here is 200 Ohms for GF180A@3.3V

The meander here is 500 Ohms for GF180A@3.3V

Diodes

Example of a diode

While normal fingered diodes now have been implemented Schottky diodes still are work in progress.

Schottky diodes

Those are not yet implemented

Transistors

In order to make sure that our transistors don't go up in flame, we have to take the hot carrier migration and thermal budget into consideration when we decide what transistor to use and whether it should have only one gate or should be fingered.

LibrePDK takes care of this and chooses the right transistor with the right amount of fingers for you based on the target operating voltage and current you plan to pump through it, you provide.

Additionally, you can also overwrite the thermal budget which usually is assumed to be for an internal circuit which isn't bonded directly to the outside.

When LibrePDK calculates that electron migration and thermal budget constraints don't allow for a single gate transistor it will dynamically create a fingered structure, either with bulk and source connected or not with the proper guard ring.

Example of a fingered transistor

Libre PDK may also decide to just generate a single gate transistor in cases where there's very little power involved

Example of a single gate transistor

Pad Cells

Last but not least: It contains the Pad Cell Generator which produces beauties like this