Cryptography Reference
In-Depth Information
4.2 FPCA on Masked DES
Basically, masking technique is considered to be a powerful countermeasure
against SCA. Indeed, it aims at masking the intermediate values that occur
during encryption and decryption process. Many masking schemes have been
proposed to the cryptographic community for symmetric encryption algorithms
(DES, AES, ...) [32,14,22]. Basically, they differ in term of hardware design
complexity. But, they all aim at fulfilling the same goal by ensuring the resis-
tance against first-order SCA like DPA and CPA. Statistically, an ideal masking
implementation is one for which all references, for all made partitions, are the
same when using the mean as CS . However, it has been proved that masking
technique is still susceptible to first-order SCA as long as glitches problem re-
mains not completely resolved [32]. For instance, authors in [18], have shown
that one masked structure so-called “Universal Substitution boxes with Mask-
ing” (USM) is vulnerable against DPA. Moreover, masked implementations are
not resistant against new variants of SCA like VPA which is mainly based on
the variance analysis. It is also shown that a full-fledged masked DES imple-
mentation using a ROM (Masked-ROM) is breakable by VPA attack, in spite of
its high resistance against first-order attacks. In what follows, we use the same
power consumption model as described in [18] to perform the FPCA on USM
and Masked-ROM DES implementations.
Fig. 3. Unprotected DES guessing en-
tropy metric
Fig. 4. Unprotected DES 1st-order suc-
cess rate metric
First, in order to make a fair evaluation for our attack on USM DES structure
we kept the “mean” as CS and we classified traces into five partitions for each
key hypothesis k j . For reasons of clarity, comparison is made between FPCA
and DPA for which we realised the best performance with regards to DoM and
CPA. Results are deduced from Fig. 5 and Fig. 6. Obviously, according to the
first-order success rate metric shown in Fig. 6, FPCA is more ecient than DPA.
Indeed, 15000 traces are needed for DPA to achieve a rate of 0.8. Whereas, for
the same rate, FPCA attack requires only 10000 traces. The guessing entropy
metric, is quite equivalent for both attacks. Second, we targeted a Masked-ROM
DES implementation. For this purpose, we chose the variance as CS ,asithas
 
Search WWH ::




Custom Search