This is a highly advanced and versatile tool for side-channel analysis in the realm of Deep Learning (DL) and other AI methods. SideDL builds upon the foundation laid by the SCA-toolbox that I also worked on and incorporates numerous additional features to enhance its functionality and usability.
One of the key strengths of SideDL lies in its ability to directly read from a wide range of (custom) datasets and file types. Whether it’s .h5, .yaml, or even SQL databases, SideDL effortlessly handles various data formats, enabling researchers and practitioners to seamlessly integrate their data into the analysis pipeline.
Automation is a paramount aspect of SideDL, as it streamlines the process of optimizing neural networks. Leveraging sca-specialized techniques, SideDL automates network optimization, allowing users to maximize the performance of their models without the need for extensive manual tuning. This feature significantly accelerates the research and development cycle, empowering users to focus on the core aspects of their projects.
The SideDL toolbox is also equipped with an array of powerful analysis tools. These tools encompass a wide range of functions, such as statistical analysis, profiling, and non-profiling methodologies. By employing these tools, users can gain deep insights into their models’ vulnerabilities, evaluate the effectiveness of different countermeasures, and ultimately enhance the security and robustness of their encryption systems.
While SideDL brings a wealth of benefits to the field of side-channel analysis, it is regrettable that the project did not receive the necessary funding to gain widespread publicity. Nonetheless, SideDL remains an exceptional tool, offering significant value to researchers and practitioners working in the domain of Deep Learning and AI-based side-channel analysis.
To achieve its comprehensive functionality, SideDL is organized into several packages. The “crypto,” “database,” “datasets,” “deeplearning,” “graphic,” and “sidechannel” packages provide a well-structured framework that encompasses various aspects of side-channel analysis. This seperation allows for easy extendability and custom improvements, which I myself utilized in all further research work on Side-Channel Analysis.
SideDL represents a significant advancement in the field of side-channel analysis for Deep Learning and AI methods. With its extensive capabilities, seamless dataset integration, automated neural network optimization, and diverse analysis tools, SideDL empowers researchers and practitioners to conduct in-depth analyses, strengthen the security of their AI systems, and drive innovation in the field. Despite its underexposure, SideDL stands as a remarkable achievement and a testament to the possibilities of leveraging AI for side-channel analysis.
As always, you can find the source code of this project, and the applcation itself here