This artifact accompanies the "pyecsca: Reverse engineering black-box elliptic curve cryptography via side-channel analysis" paper and contains:

• The pyecsca toolkit itself, as a git submodule, implementing the methods presented in the paper. This is under pyecsca/ and also in the repository https://github.com/J08nY/pyecsca.
• The notebooks used to evaluate the methods in the paper. This is under pyecsca/notebook/ and also in the repository https://github.com/J08nY/pyecsca-notebook.
• The documentation for the toolkit. This is under docs/ and also on the site https://pyecsca.org.
• The expanded EFD formula dataset as described in the paper. This is under expanded_efd/ and also on https://zenodo.org/doi/10.5281/zenodo.10908697.

To reproduce the results in the paper follow these instructions. All of the tables, figures and datasets are reproduced here. If you wish to explore the toolkit more, use the notebooks linked next to sections in the paper or at https://pyecsca.org/notebooks.html.

We performed the evaluation on a machine with Ubuntu 23.04, 64GB RAM, and AMD Ryzen 9 7950X 16-Core CPU. However, a machine with much less resources should be perfectly usable, yet slower. You may run into out-of-memory errors, which manifest as seeing A process in the process pool was terminated abruptly while the future was running or pending. error messages in the notebook and OOM killer messages in the system log (sudo dmesg...). You may mitigate this by using less worker which decreases the amount of memory that is used in parallel as well as by freeing more memory for the computation.

Follow the steps below to install the core package as well as the requirements for the notebooks (they can be found in requirements.txt in the notebook repository). For more information, see the installation docs.

git submodule update --init --recursive
python3 -m venv virt
. virt/bin/activate
pip install pyecsca/[pari,gmp]
pip install -r pyecsca/notebook/requirements.txt

Then run either Jupyter Lab or Notebook:

jupyter lab

Note that the pari and gmp extras are required for performant computation of some of the results, they require libpari and libgmp, respectively.

Assuming the pyecsca package is installed, open the formulas.ipynb Jupyter notebook in the pyecsca/notebook/re/ directory. Run its cells sequentially and observe the expanded formula dataset generated in the final cell. Change the num_workers value to something that suits your setup (i.e. number of your CPU cores minus two). This may take roughly an hour with 10 cores (one job takes significantly longer than others).

Assuming the pyecsca package is installed, open the configuration_space.ipynb Jupyter notebook in the pyecsca/notebook/ directory. Run its cells sequentially and observe the tables being printed (with additional columns that are not present in the paper). Note that the cell computing all of the configurations takes some time to finish (~ 10 minutes), as it enumerates them internally.

Assuming the pyecsca package is installed, open the rpa.ipynb Jupyter notebook in the pyecsca/notebook/re/ directory. Run its cells sequentially. Change the num_workers value to something that suits your setup (i.e. number of your CPU cores minus two). Reproducing Figure 5 may take an hour with 10 cores, while computing the asymmetric case (no figure in paper) may take ~12 hours with 10 cores. Reproducing Figure 6 may take ~14 hours with 10 cores.

You can observe the tree metrics for RPA-RE in Table 3 from the following cell:

re = RPA(set(multipliers))
with silent():
    re.build_tree(p256, tries=10)
print(re.tree.describe())

You can observe Figures 5 and 6 as direct output of some cells, as well as saved into files:

• rpa_re_success_rate_symmetric.pdf and rpa_re_query_rate_symmetric.pdf for Figure 5
• rpa_re_errors.pdf for Figure 6.

Assuming the pyecsca package is installed, open the zvp.ipynb Jupyter notebook in the pyecsca/notebook/re/ directory. Run its cells sequentially, upto and including the evaluation part. Optionally, you can skip some cells that are there to store and subsequently load various intermediate datasets. These can, however, help you save your progress if something happens to the computation later on. Change the num_workers value to something that suits your setup (i.e. number of your CPU cores minus two).

If your ZVP point computation seems very slow (estimated >24 hours), one of a few things may be going on:

• You do not have the pari extra and libpari installed, thus pyecsca falls back to slow division polynomial calculation. See the installation instructions, you may verify this by running from pyecsca.ec.divpoly import has_pari; print(has_pari).
• You are computing ZVP points for the expanded set of formulas (much larger than the EFD). See the load_expanded boolean in the notebook.

You can observe the tree metrics for ZVP-RE in Table 3 from the following cell:

print("Zero hit")
print(tree_remapped.describe())
print("\nZero count")
print(tree_count.describe())
print("\nZero position")
print(tree_position.describe())

You can observe Figures 7 and 8 as direct output of some cells, as well as saved into files:

• zvp_re_success_rate_symmetric.pdf, zvp_re_query_rate_symmetric.pdf, zvp_re_precise_rate_symmetric.pdf, and zvp_re_amount_rate_symmetric.pdf for Figure 7
• zvp_re_success_rate_binomial.pdf, zvp_re_query_rate_binomial.pdf, zvp_re_precise_rate_binomial.pdf, and zvp_re_amount_rate_binomial.pdf for Figure 8