Time Sequence based AES S-box Implementation Cryptoanalysis using Deep Learning Approaches

So far profiling attack on block ciphers is the most effective side channel analysis method. Conventional profiling attack approaches are always dependent on the evaluation models coordinating with target implementation devices. In this paper, we concentrate on exploring the routine of analysis the profile of generalized instruction execution on generalized computing platforms. We analyse and compare the basic routine of profiling attack using template attack, machine learning based attack and deep learning approaches, and take full advantage of deep learning approaches to make a time sequential AES S-box implementation cryptoanalysis. We use an accurate register operation timer to record the byte substitution of a multiple rounds, and use the deep neural network to recover the target key. According to our experiments, our approaches are more reliable, generalized, and can achieve at least the same performance as the conventional attack routines.