As we mentioned in a previous article, rounds that make up the AES algorithm can be divided into three different categories: initial rounds, standard round and final round. When encryption, the information is subjected to a different number of rounds (10, 12 and 14 respectively) in each case depending on the length of the key (128, 192 and 256 respectively). Each such round is made up of combinations of transformations, the possible transformations include: ByteSub, ShiftRow, MixColumns and AddRoundKey. Figure # 1 shows the relationship of the transformation and corresponding rounds.
Figure # 1: Distribution of the transformations in each round.
As we can see in Figure # 1, in the original algorithm, the final round differs from standard rounds by omitting the MixColumns transformation, further, the initial round is only composed by the AddRoundKey transformation. In relation to these two variants in the initial and final rounds, relative to standard rounds, the authors of AES Joan Daemen and Vincent Rijmen have mentioned that they have proposed them in order to keep the symmetry between encryption and decryption, as they mention :
“In order to make the cipher and its inverse more similar in structure, the linear mixing layer of the last round is different from the mixing layer in the other rounds. It can be shown that this does not improve or reduce the security of the cipher in any way. This is similar to the absence of the swap operation in the last round of the DES.”
It is commonly known, and the same authors say in the quote above, that the omission of MixColumns do not affect or reduce the security of the encryption, however, a recent study by Orr Dunkelman and Nathan Keller  shows that security can be compromised by the omission of such transformation.
This study takes a plaintext with the corresponding encryption for analysis (for the simplest case of a single round) omitting MixColumns and compares it with other attack in which this transformation has not been omitted. The results show that in the case of the analysis by omitting MixColumns, the algorithm complexity is reduced by a factor of 216. This same method was applied to the proposed attack by Raphael C.-W. Phan  (discussed above) the same results for the attack on 7 rounds with a key length of 192 bits was obtained.
Furthermore, the observations obtained in , appear not to apply to any type of attack, some of them being immune, in the complexity reduction, to the omission of MixColumns transformation in the final round of AES. Despite, the authors claimed that AES security is affected by the omission of MixColumns for attacks where a reduction in the number of standards rounds are used, and therefore, the complete algorithm may also be affected.
It is important to note that the presented attack is an attack on reduced rounds (7 rounds). This means that despite the successful attack proposed by the authors, AES maintains a fairly wide margin of safety with respect to this research.
- Joan Daemen VR. AES Proposal: Rijndael. NIST AES Proposal (1998).
- Orr Dunkelman NK. The effects of the omission of last round’s MixColumns on AES. Information Processing Letters (2010) 110: pp. 304-308.
- Raphael CP. Impossible differential cryptanalysis of 7-round Advanced Encryption Standard (AES). Information Processing Letters (2004) 91: pp. 33-38.
Luis Sequeira is an IT professional with experience in cloud environments, quality of service and network traffic analysis, who loves looking for solutions to engineering challenges, share knowledge. At work, the main challenge is to integrate different network and software technologies to provide solution in a wide range of areas, e.g., virtual network functions, machine learning, autonomous driving, robotics and augmented reality.Website: https://www.luissequeira.com