Deep Learning Enhanced Side Channel Analysis on CRYSTALS-Kyber

The combination of Deep-learning (DL) and Side-channel analysis (SCA) has been proven by several attacks targeting symmetric key cryptography implementations such as AES. This paper aims to demonstrate the effectiveness of DL in attacking a Post Quantum CRYSTALS-Kyber implementation to recover the private key. We propose a CNN model with additional ciphertext knowledge to attack each 12-bit coefficient of the polynomial vector representing the private key. The model assigns a label to each trace by combining the values of each coefficient from the private key and so the attacker does not require any knowledge about the implementation and little or no knowledge about the Kyber algorithm. The model needs only 50 traces to reveal the coefficients of the polynomial vector which represents the entire private key.