This repository contains simulation and analysis code for the LP-PUF. The current version of LP-PUF is 1 to allow for future improvements.
All code in this repository uses pypuf, a PUF cryptanalysis tool. It can be installed using
python3 -m pip install pypufFurthermore, pandas and seaborn are used for data analysis and visualization. Some code is organized in traditional Python modules, most analyses are run in Juypter notebooks.
The simulation of the LP-PUF is based on the
Arbiter PUF simulation of pypuf.
An LP-PUF simulation instance can be created by passing the relevant security parameters and a seed which is used
to initialize the PRNG to obtain "physical" intrinsic parameters of the involved Arbiter PUFs:
import lppuf
my_lp_puf = lppuf.LPPUFv1(n=64, m=8, seed=1)where n specifies the challenge length, and m defines the number of Arbiter PUFs in the first layer.
Additionally, the parameters noisiness_1 and noisiness_2 may be given by non-negative floats to control the
reliability of the first and third layer, respectively.
The defined PUF instances can be evaluated on challenges like so:
import pypuf.io
my_lp_puf.eval(pypuf.io.random_inputs(n=64, N=3, seed=1))The LP-PUF is analyzed for its bias, reliability, uniqueness, and bit sensitivity in the corresponding notebook files.
The security analysis with respect to the Logistic Regression / Splitting Attack and the MLP attack split into two parts. In the first part, the attacks are run on LP-PUF simulations. In the second part, the results are analyzed using a Jupyter notebook. This split was done as the attacks require a relatively long time to complete and are preferably run on a SLURM computing cluster.
To run the attacks on a single computer, use
python3 -m lppufv1_lr_full results/v1/lr 0 1 # for the Logistic Regression / Splitting Attack
python3 -m lppufv1_mlp_full results/v1/mlp 0 1 # for the MLP attackThe results will be stored at results/v1/lr and results/v1/mlp, respectively.
Afterwards, the analysis can be done by running the LR Attack and
MLP Attack notebooks.
The results are not contained in this repository due to their large file size. However, the notebooks contain the analysis results.
As the analysis for reliability-based attacks is done on reliability values rather than by running the actual attack, the run time is relatively short and the whole analysis can be done within a Jupyter notebook. There are analyses for Layer 1 and Layer 3
All source code in this repository is licensed under the GPLv3. Everything else is licenced as CC BY 4.0 (full license).