Pushing the Limits of Generic Side-Channel Attacks on LWE-based KEMs - Parallel PC Oracle Attacks on Kyber KEM and Beyond

Abstract

In this work, we propose generic and novel adaptations to the binary Plaintext-Checking (PC) oracle based side-channel attacks for Kyber KEM. These attacks operate in a chosen-ciphertext setting, and are fairly generic and easy to mount on a given target, as the attacker requires very minimal information about the target device. However, these attacks have an inherent disadvantage of requiring a few thousand traces to perform full key recovery. This is due to the fact that these attacks typically work by recovering a single bit of information about the secret key per query/trace. In this respect, we propose novel parallel PC oracle based side-channel attacks, which are capable of recovering a generic P number of bits of information about the secret key in a single query/trace. We propose novel techniques to build chosen-ciphertexts so as to efficiently realize a parallel PC oracle for Kyber KEM. We also build a multi-class classifier, which is capable of realizing a practical side-channel based parallel PC oracle with very high success rate. We experimentally validated the proposed attacks (upto P = 10) on the fastest implementation of unprotected Kyber KEM in the pqm4 library. Our experiments yielded improvements in the range of 2.89× and 7.65× in the number of queries, compared to state-of-the-art binary PC oracle attacks, while arbitrarily higher improvements are possible for a motivated attacker, given the generic nature of the proposed attacks. We further conduct a thorough study on applicability to different scenarios, based on the presence/absence of a clone device, and also partial key recovery. Finally, we also show that the proposed attacks are able to achieve the lowest number of queries for key recovery, even for implementations protected with low-cost countermeasures such as shuffling. Our work therefore, concretely demonstrates the power of PC oracle attacks on Kyber KEM, thereby stressing the need for concrete countermeasures such as masking for Kyber and other lattice-based KEMs.

Analyzing CRYSTALS-Kyber's Susceptibility to Side Channel Attacks: An Empirical Exploration

“…Pushing the Limits of Generic Side-ChannelAttacks on LWE-based KEMs -Parallel PCOracle Attacks on Kyber KEM and Beyond The research conducted by Rajendran and colleagues [49] presents an oracle-based framework where it is assumed that the attacker has physical access to a target device executing the Kyber Key Encapsulation Mechanism (KEM) decapsulation process using the pqm4 implementation [28]. The attacker is capable of interacting with the target device by submitting selected ciphertexts according to their preferences.…”

Implementation attacks on post-quantum lattice-based cryptography

“…Very recently, Rajendran et al [130] and Tanaka et al [342] demonstrated improved PC oracle based side-channel attacks, which are capable of more than one bit of information per query. They demonstrated the ability to recover a generic P number of bits of information about the secret key in a single query (P oe Z + ) through construction of modified ciphertexts ct = (u, v) oe (R k q ◊ R q ) as follows:…”

“…The realization of such a P -way parallel PC oracle, reduces the number of attack traces/queries for key recovery, by a factor of P , compared to the Binary_PC_Oracle_CCA attack [94,276]. In this respect, Rajendran et al [130] and Tanaka et al [342] experimentally demonstrate that there is enough information present in power/EM side-channel leakage from the re-encryption procedure to distinguish between 2 P possible values of the message m Õ in a single trace for P < 10.…”

“…Evaluation: We subsequently tested several existing side-channel attacks [130,268,276] and found that for all attack ciphertexts used in these attacks there was a significant probability of triggering the countermeasure and thus discarding the secret key. More specifically, these attacks focus on one coe cient of the secret, and for all attack ciphertexts at least one possible value of this coe cient of the secret leads to detection of the attack.…”

“…More specifically, these attacks focus on one coe cient of the secret, and for all attack ciphertexts at least one possible value of this coe cient of the secret leads to detection of the attack. The attack of Rajendran et al [130] also includes an attack that targets multiple coe cients at once, but this improvement only increases the probability of triggering the countermeasure. We did not find parameter sets that reliably avoided our countermeasure.…”