Paper 2022/890

One Network to rule them all. An autoencoder approach to encode datasets

Abstract

Side-channel attacks are powerful non-invasive attacks on cryptographic algorithms. Among such attacks, profiling attacks have a prominent place as they assume an attacker with access to a copy of the device under attack. The attacker uses the device's copy to learn as much as possible about the device and then mount the attack on the target device. In the last few years, Machine Learning has been successfully used in profiling attacks, as such techniques proved to be capable of breaking implementations protected with countermeasures. In the deep learning-based profiling attack, a core problem is finding efficient neural network architectures to evaluate an implementation's security correctly. Unfortunately, this process is time-consuming, and a different neural network configuration usually needs to be defined for every target. Hence, we propose the following process: train a separate autoencoder for each dataset obtained from different cryptographic implementations and devices to receive an encoded version for each one. After that, define a universal model that can break multiple (encoded) datasets. Thus, instead of finding dataset-specific neural network architectures, we reduce the effort to find autoencoders to encode the datasets and a single neural network to break~them.