We compare two key recovery methods for single trace attacks
on the AES key schedule. The 2018 CHES capture-the-flag (CTF)
challenge which includes an unprotected key schedule raises the question,
which method performs best during key recovery: Soft Analytical
Side-Channel Attacks (SASCAs) or Algebraic Side-Channel Attacks (ASCAs).
SASCAs as well as ASCAs exploit knowledge about the attacked
algorithm by leakage recombination and allow for a computationally efficient
key recovery based on e.g. Hamming Weight (HW) leakage. We use
Belief Propagation (BP), which is the most popular choice for SASCA
and a SAT solver as an ASCA algorithm. In this work we attack real
traces of the CTF challenge to demonstrate the limitations of SASCAs
while handling the XOR operation. We exemplify that SASCAs may not
always be the most favorable solution. The comparison is solidified by
evaluating the success rate of SASCAs and ASCAs with simulated HW
leakage on varying noise levels. During attacks on the AES key schedule
the convergence of BP is not only graph dependent but data dependent.
Further, we discuss possible graph clusters and adaptations of the input
distributions to mitigate the influence of the XOR operations and
increase the success rate of BP. All experiments are compared against
equivalent SAT solver approaches. Based on our results we propose a
combination of brute-force and BP to level the performance of the SAT
solver and BP. Apart from this, we address unsolved questions regarding
the benefit of an early break of BP and point out implementation details
which lead to a better success rate.
«
We compare two key recovery methods for single trace attacks
on the AES key schedule. The 2018 CHES capture-the-flag (CTF)
challenge which includes an unprotected key schedule raises the question,
which method performs best during key recovery: Soft Analytical
Side-Channel Attacks (SASCAs) or Algebraic Side-Channel Attacks (ASCAs).
SASCAs as well as ASCAs exploit knowledge about the attacked
algorithm by leakage recombination and allow for a computationally efficient
key recovery based on...
»