Deep Learning Side-Channel Collision Attack

This repository contains source code related to the article Deep Learning Side-Channel Collision Attack and provides some easy examples. Please note that the provided scripts have been intentionally condensed to their fundamental functionality and deliberately kept brief and straightforward.

Short Description

We present a new methodology, that is able to exploit side-channel collisions in a black-box setting. In particular, this attack can be performed in a non-profiled setting and requires neither a hypothetical power model nor details about the underlying implementation. While the existing non-profiled DL attacks utilize training metrics to distinguish the correct key (e.g., here), this attack is more efficient by training a model that can be applied to recover multiple key portions, e.g., bytes. In order to perform this attack on raw traces instead of pre-selected samples, we further introduce a DL-based technique that can localize input-dependent leakages in masked implementations, e.g., the leakages associated to one byte of the cipher state in case of AES.

Requirements

To run the Python scripts, a working installation of TensorFlow, keras, and scikit-learn is required. GPU acceleration can be beneficial when working with large datasets.

Getting Started

• Step 1: Execute LeakageIdentification.py to obtain the leaking position for a specific key portion. You should execute it with a wide range of sample points first and select a smaller range around the peak afterwards (see Section 3.2 in paper). It might take several tries until a suitable range is found. The final range is determined by visual inspection.
• Step 2: Execute TrainingPhase.py with the previously determined leaking range. You will train a corresponding model that can be used to find collisions in the other key portions.
• Step 3: When the offsets between the processing of different portions are known, you can either calculate the ranges for the other bytes yourself or use CalcRanges.py. If the offsets are unknown, you have to repeat Step 1 and receive the output weight-files. CalcRanges.py will help you to calculate the range for the target byte based on the weight-files.
• Step 4: Execute AttackPhase.py with the previously calculated ranges for the target byte.

Examples

All python scripts should be executable without any prior changes on the parameters to target the seventh byte of the ASCADv1 dataset. The dataset can be found here, all other required files are contained in the Examples folder. When you aim to target any other bytes, make sure to follow the order of execution as explained in the paper.

Contact and Support

Please contact Marvin Staib (marvin.staib@rub.de) if you have any questions, comments or if you found a bug that should be corrected.

Licensing

Please see LICENSE for licensing instructions.

Publication

Deep Learning Side-Channel Collision Attack.