This repo contains the code and a link to the dataset of traces used to evaluate the robustness of masked neural networks (NNs) against side-channel attacks. The results of this study are presented in the paper ``Bake It Till You Make It: Heat-induced Power Leakage from Masked Neural Networks'' (https://eprint.iacr.org/2023/076.pdf). ModuloNET, a masked NN offering first-order security, is implemented on Artix-7 FPGA on Chipwhisperer CW305 target, and the Chipwhisperer Lite and CW Husky boards control the capturing process. A set of power traces captured for this study is gievn in the dataset folder.
The dataset is available here.
The dataset could be partially downloaded. To accomplish that safely, you should still maintain the folder structure for the individual experiments (BITUMI_scape->Experiments/metadata.json->(downloaded the need experiment)->(partially download the datasets too)).
Experiments list:
- intermediate_values: datasets with correct intermediate values for all the experiments.
Below is the table explaining the characteristics of the experiments with traces:
| Experiments | Segmented | Unsegmented | Zipped | TVLA | DPA |
|---|---|---|---|---|---|
| exp_hg_off_prng_off | ❌ | ✔️ | ✔️ | ❌ | ✔️ |
| exp_hg_on_prng_on | ❌ | ✔️ | ✔️ | ✔️ | ✔️ |
| exp_hg_off_prng_on | ❌ | ✔️ | ❌ | ✔️ | ✔️ |
| exp_hg_off_prng_on_thermal_chamber | ✔️ | ❌ | ❌ | ✔️ | ✔️ |
| exp_hg_off_prng_on_thermal_chamber_room | ✔️ | ❌ | ❌ | ✔️ | ✔️ |
In the Python folder, you will find Jupyter python notebooks, which can be used to replicate the results using the dataset provided.
@article{mehta2023bake, title={Bake It Till You Make It: Heat-induced Leakage from Masked Neural Networks}, author={Mehta, Dev M and Hashemi, Mohammad and Koblah, David S and Forte, Domenic and Ganji, Fatemeh}, journal={Cryptology ePrint Archive}, year={2023} }