Experimental Verification of Super-Sbox Analysis — Confirmation of Detailed Attack Complexity

This paper implements the super-sbox analysis on 8-round AES proposed by Gilbert and Peyrin in order to verify its correctness and the attack cost. The attack consists of three parts; the first outbound phase, inbound phase with a super-sbox technique, and the second outbound phase. Gilbert and Peyrin estimated that the attack would require 2

48

computational cost and 2

32

memory, which could be feasible but not easy to practically implement. In this research, we first analyze the relationship among memory, computational cost, and the number of solutions in the inbound phase, and then show that the tradeoff exists for the super-sbox analysis. With this tradeoff, we implement the attack for each of the outbound phase independently so that the cost for the entire attack can be estimated by the experiments. As a result of our experiment, we show that the computational cost to obtain a pair of values satisfying the inbound phase is approximately 4 times higher and the freedom degrees are 4 times smaller than the previous estimation, which indicates that applying the super-sbox analysis is harder than expected.