Toward practical autoencoder-based side-channel analysis evaluations

In the field of side-channel analysis, profiled attacks are one most powerful types attacks. Nevertheless, major issues with in sensitivity to noise and high-dimensional nature signals used for training, generating a less efficient classifier conduct attack phase. Consequently, evaluating security cryptographic implementation hardware devices like IoT becomes more complex as analysis evaluation easily falls into false-positive results. This paper assesses efficacy applying feature reduction process deal signals. We propose practical procedure using autoencoders leakage evaluations. Two autoencoder architectures compared while performing showing that our proposed architecture keeps relevant information. Our proposal is tested on ASCAD random key database high desynchronization value produced results outperform other state-of-the-art techniques. The guessing entropy converges 1 after around 500 traces.