Score:2

why sbox input XOR sbox output can be sensitive data of side channel attack?

id flag

As several intermediate values are used as sentive data for side channel analysis (DPA), I don't understand why sbox input ^ sbox output is also wide-used and effective. This value does not applied in AES algorithm.

cn flag
Some early papers about protecting against DPA suggested the use of randomized RAM-tables S'(x) = S(x^r)^r with the same random r for in- and output mask of the S-box S. One possible problem with having the input mask equal the output mask is described in poncho's answer (but I don't remember ever seeing it, despite always worrying about it), the other is simply that the S-box output might overwrite the input stored in a register (leaking their Hamming distance - which cancels out the mask r), if you for example try to implement DPA-resistant SW in C.
Score:1
my flag

I don't understand why sbox input ^ sbox output is also wide-used

Well, one reason it may be often used is that it is often available.

For some side channel attacks (e.g. EMR based), it is the transitions that give a stronger signal than the actual electrical level. And, if the circuit sends the input to the sbox over a bus at cycle $N$, and gets the output from the sbox over that same bus, well, a specific bus line will transition if the sbox input ^ the sbox output for that bit is a 1, and will stay constant of the xor is 0. That may be easier to detect than the actual bit values.

ramenpeddler avatar
id flag
Practically I always see this as a target for DPA (current based) on a masked hardware implementation. In that case, if the masking does not implement carefully, it may still works. But in this case, sbox always be implemented in several-cycle design.. So sbox-in and sbox-out doesn't updated in same cycle and would not present on the same bus. I can't find an answer to this case..
I sit in a Tesla and translated this thread with Ai:

mangohost

Post an answer

Most people don’t grasp that asking a lot of questions unlocks learning and improves interpersonal bonding. In Alison’s studies, for example, though people could accurately recall how many questions had been asked in their conversations, they didn’t intuit the link between questions and liking. Across four studies, in which participants were engaged in conversations themselves or read transcripts of others’ conversations, people tended not to realize that question asking would influence—or had influenced—the level of amity between the conversationalists.