International Association
for Cryptologic Research

Secure multi-party computation and homomorphic encryption are two primary security primitives in privacy-preserving machine learning, whose wide adoption is, nevertheless, constrained by the computation and network communication overheads. This paper proposes a hybrid Secret-sharing and Homomorphic encryption Architecture for Privacy-pERsevering machine learning (SHAPER). SHAPER protects sensitive data in encrypted or randomly shared domains instead of relying on a trusted third party. The proposed algorithm-protocol-hardware co-design methodology explores techniques such as plaintext Single Instruction Multiple Data (SIMD) and fine-grained scheduling, to minimize end-to-end latency in various network settings. SHAPER also supports secure domain computing acceleration and the conversion between mainstream privacy-preserving primitives, making it ready for general and distinctive data characteristics. SHAPER is evaluated by FPGA prototyping with a comprehensive hyper-parameter exploration, demonstrating a 94x speed-up over CPU clusters on large-scale logistic regression training tasks.

In this work, we present the first fault injection analysis of the Number Theoretic Transform (NTT). The NTT is an integral computation unit, widely used for polynomial multiplication in several structured lattice-based key encapsulation mechanisms (KEMs) and digital signature schemes. We identify a critical single fault vulnerability in the NTT, which severely reduces the entropy of its output. This in turn enables us to perform a wide-range of attacks applicable to lattice-based KEMs as well as signature schemes. In particular, we demonstrate novel key recovery and message recovery attacks targeting the key generation and encryption procedure of Kyber KEM. We also propose novel existential forgery attacks targeting deterministic and probabilistic signing procedure of Dilithium, followed by a novel verification bypass attack targeting its verification procedure. All proposed exploits are demonstrated with high success rate using electromagnetic fault injection on optimized implementations of Kyber and Dilithium, from the open-source pqm4 library on the ARM Cortex-M4 microcontroller. We also demonstrate that our proposed attacks are capable of bypassing concrete countermeasures against existing fault attacks on lattice-based KEMs and signature schemes. We believe our work motivates the need for more research towards development of countermeasures for the NTT against fault injection attacks.

In 2018, Zhang et al. introduced the Persistent Fault Analysis (PFA) for the first time, which uses statistical features of ciphertexts caused by faulty Sbox to recover the key of block ciphers. However, for most of the variants of PFA, the prior knowledge of the fault (location and value) is required, where the corresponding analysis will get more difficult under the scenario of multiple faults. To bypass such perquisite and improve the analysis efficiency for multiple faults, we propose Chosen-Plaintext based Persistent Fault Analysis (CPPFA). CPPFA introduces chosen-plaintext to facilitate PFA and can reduce the key search space of AES-128 to extremely small. Our proposal requires 256 ciphertexts, while previous state-of-the-art work still requires 1509 and 1448 ciphertexts under 8 and 16 faults, respectively, at the only cost of requiring 256 chosen plaintexts. In particular, CPPFA can be applied to the multiple faults scenarios where all fault locations, values and quantity are unknown, and the worst time complexity of CPPFA is O(28+nf ) for AES-128, where nf represents the number of faults. The experimental results show that when nf > 4, 256 pairs of plaintext-ciphertext can recover the master key of AES-128. As for LED-64, only 16 pairs of plaintext-ciphertext reduce the remaining key search space to 210.

Algebraic Fault Analysis (AFA) is a cryptanalysis for block ciphers proposed by Courtois et al., which incorporates algebraic cryptanalysis to overcome the complexity of manual analysis within the context of Differential Fault Analysis (DFA). The effectiveness of AFA on lightweight block ciphers has been demonstrated. However, the complexity of the algebraic systems prevents it from attacking heavyweight block ciphers efficiently. In this paper, we propose a novel cryptanalysis called Redundancies-assisted Algebraic Fault Analysis (RAFA) to facilitate the solution of algebraic systems in the setting of heavyweight block ciphers. The core idea of RAFA is to expedite SAT solvers by modifying the algebraic systems, which is accomplished via two methods. The first method introduces redundant constraints, which is proposed for the first time in the context of algebraic cryptanalysis. The second one is a sophisticated linearization of the nonlinear Algebraic Normal Form (ANF). It takes RAFA for about 9.68 hours to attack AES-128. To the best of our knowledge, this is the first work that uses a general SAT solver to attack AES with only a single injection of byte-fault. Moreover, RAFA can attack AES-128 in 50.92 and 27.54 minutes for nibble- and bit-based fault model, respectively. In comparison, the traditional DFA algorithm implemented by pure C takes 4 ~ 5 hours under all three fault models investigated in this work. Moreover, in order to show the generality of RAFA, we also apply it to other heavyweight block ciphers. The best results show that RAFA could recover the key of Serpent-256 and SPEEDY-r-192 in 20.7 and 1.5 hours using only three faults, respectively. In comparison, AFA could not break these two ciphers even when 30 bits and 50 bits of their keys are known, respectively. Furthermore, no DFA work on Serpent or SPEEDY is known using comparable fault models.

Persistent Fault Analysis (PFA) is a new fault analysis method for block ciphers proposed in CHES 2018, which utilizes those faults that persist in encryptions. However, one fact that has not been raised enough attention is that: while the fault itself does persist in the entire encryption, the corresponding statistical analysis merely leverages fault leakages in the last one or two rounds, which ignores the valuable leakages in deeper rounds. In this paper, we propose Algebraic Persistent Fault Analysis (APFA), which introduces algebraic analysis to facilitate PFA. APFA tries to make full usage of the free fault leakages in the deeper rounds without incurring additional fault injections. The core idea of APFA is to build similar algebraic constraints for the output of substitution layers and apply the constraints to as many rounds as possible. APFA has many advantages compared to PFA. First, APFA can bypass the manual deductions of round key dependencies along the fault propagation path and transfer the burdens to the computing power of machine solvers such as Crypto-MiniSAT. Second, thanks to the free leakages in the deeper round, APFA requires a much less number of ciphertexts than previous PFA methods, especially for those lightweight block ciphers such as PRESENT, LED, SKINNY, etc. Only 10 faulty ciphertexts are required to recover the master key of SKINNY-64-64, which is about 155 times of reduction as compared to the state-of-the-art result. Third, APFA can be applied to the block ciphers that cannot be analyzed by PFA due to the key size, such as PRESENT-128. Most importantly, APFA replaces statistical analysis with algebraic analysis, which opens a new direction for persistent-fault related researches.

Code-based masking is a highly generalized type of masking schemes, which can be instantiated into specific cases by assigning different encoders. It captivates by its side-channel resistance against higher-order attacks and the potential to withstand fault injection attacks. However, similar to other algebraically-involved masking schemes, code-based masking is also burdened with expensive computational overhead. To mitigate such cost and make it efficient, we contribute to several improvements to the original scheme proposed by Wang et al. in TCHES 2020. Specifically, we devise a computationally friendly encoder and accordingly accelerate masked gadgets to leverage efficient implementations. In addition, we highlight that the amortization technique introduced by Wang et al. does not always lead to efficient implementations as expected, but actually decreases the efficiency in some cases.From the perspective of practical security, we carry out an extensive evaluation of the concrete security of code-based masking in the real world. On one hand, we select three representative variations of code-based masking as targets for an extensive evaluation. On the other hand, we aim at security assessment of both encoding and computations to investigate whether the state-of-the-art computational framework for code-based masking reaches the security of the corresponding encoding. By leveraging both leakage assessment tool and side-channel attacks, we verify the existence of “security order amplification” in practice and validate the reliability of the leakage quantification method proposed by Cheng et al. in TCHES 2021. In addition, we also study the security decrease caused by the “cost amortization” technique and redundancy of code-based masking. We identify a security bottleneck in the gadgets computations which limits the whole masked implementation. To the best of our knowledge, this is the first time that allows us to narrow down the gap between the theoretical security order under the probing model (sometimes with simulation experiments) and the concrete side-channel security level of protected implementations by code-based masking in practice.

Persistence fault analysis (PFA) is a novel fault analysis technique proposed in CHES 2018 and demonstrated with rowhammer-based fault injections. However, whether such analysis can be applied to traditional fault attack scenario, together with its difficulty in practice, has not been carefully investigated. For the first time, a persistent fault attack is conducted on an unprotected AES implemented on ATmega163L microcontroller in this paper. Several critical challenges are solved with our new improvements, including (1) how to decide whether the fault is injected in SBox; (2) how to use the maximum likelihood estimation to pursue the minimum number of ciphertexts; (3) how to utilize the unknown fault in SBox to extract the key. Our experiments show that: to break AES with physical laser injections despite all these challenges, the minimum and average number of required ciphertexts are 926 and 1641, respectively. It is about 38% and 28% reductions of the ciphertexts required in comparison to 1493 and 2273 in previous work where both fault value and location have to be known. Furthermore, our analysis is extended to the PRESENT cipher. By applying the persistent fault analysis to the penultimate round, the full PRESENT key of 80 bits can be recovered. Eventually, an experimental validation is performed to confirm the accuracy of our attack with more insights. This paper solves the challenges in most aspects of practice and also demonstrates the feasibility and universality of PFA on SPN block ciphers.

Decades of research in both cryptography and distributed systems has extensively studied the problem of state machine replication, also known as Byzantine consensus. A consensus protocol must usually satisfy two properties: {\em consistency} and {\em liveness}. These properties ensure that honest participating nodes agree on the same log and dictate when fresh transactions get added. They fail, however, to ensure against adversarial manipulation of the actual {\em ordering} of transactions in the log. Indeed, in leader-based protocols (almost all protocols used today), malicious leaders can directly choose the final transaction ordering. To rectify this problem, we propose a third consensus property: {\em transaction order-fairness}. We initiate the first formal investigation of order-fairness and explain its fundamental importance. We also provide several natural definitions for order-fairness and analyze the assumptions necessary to realize them. We also propose a new class of consensus protocols called Aequitas. Aequitas protocols are the first to achieve order-fairness in addition to consistency and liveness. They can be realized in a black-box way using existing broadcast and agreement primitives (or indeed using any consensus protocol), and work in both synchronous and asynchronous network models.

Persistence is an intrinsic nature for many errors yet has not been caught enough attractions for years. In this paper, the feature of persistence is applied to fault attacks, and the persistent fault attack is proposed. Different from traditional fault attacks, adversaries can prepare the fault injection stage before the encryption stage, which relaxes the constraint of the tight-coupled time synchronization. The persistent fault analysis (PFA) is elaborated on different implementations of AES-128, specially fault hardened implementations based on Dual Modular Redundancy (DMR). Our experimental results show that PFA is quite simple and efficient in breaking these typical implementations. To show the feasibility and practicability of our attack, a case study is illustrated on the shared library Libgcrypt with rowhammer technique. Approximately 8200 ciphertexts are enough to extract the master key of AES-128 when PFA is applied to Libgcrypt1.6.3 with redundant encryption based DMR. This work puts forward a new direction of fault attacks and can be extended to attack other implementations under more interesting scenarios.