AISY Framework: Deep Learning for Side-Channel Analysis

Side-Channel Analysis

AISY – Deep Learning-based Framework for Side-Channel Analysis

AISY framework is a python-based framework that allows efficient and scalable applications of deep learning to profiling side-channel analysis (SCA). This project was implemented as a result of several years of research on deep learning and side-channel analysis by AisyLab at TU Delft (The Netherlands).

Integrated SQLite Database

AISY Framework comes with the option to store all analysis results in an SQLite database. Standard libraries are implemented in the framework and users can easily add custom tables to the project. The creation of custom tables does not require any specific background knowledge on databases.

Web Application

AISY Framework is built on top of the Flask python-based web framework. A web application is integrated with a web-based user interface.

The web application provides a user-friendly way to visualize analysis, plots, results, and tables. Note, however, that the interface is only intended to provide an easy way to visualize results and keep them organized on databases. From the web interface, the user cannot run scripts or manipulate analysis settings (this will appear in future versions of the framework).

Reproducible results with one-click script generation

In the Web Application, a user can generate the full script that was used to produce results stored in the database. This is particularly important when the user wishes to reproduce results after he or she changed the script and don’t keep track of the changes. Another advantage of this feature can be seen when a user shares the database with a second user. The latter generate all the scripts from the original database.

State-of-the-art deep learning-based SCA

AISY Framework brings state-of-the-art deep learning-based side-channel attacks. We follow recent publications and keep the framework updated accordingly. The current version is 0.2 and it already provides state-of-the-art features such as custom loss functions, hyperparameter search, visualization, custom metrics, and data augmentation.

Main Features

  • SCA Metrics
  • Gradient Visualization
  • Data Augmentation
  • Grid Search
  • Random Search
  • Early Stopping
  • Ensemble
  • Profiling Analyzer
  • Custom Metrics
  • Custom Callbacks
  • Custom Loss Functions
  • Confusion Matrix
  • Easy Neural Network Definitions
  • Data Augmentation
  • GUI – plots, tables
  • Automatically generate scripts
  • Fully reproducible script

Install & Use

Copyright (c) 2021 AISyLab @ TU Delft