Abstract
This paper introduces DCSCA (Differential Ciphertext Side-Channel Analysis), a novel "no-fault" attack that breaks lightweight ciphers like GIFT-128 without physical interference. By combining the principles of Differential Fault Analysis (DFA) with passive side-channel leakages, the authors demonstrate how to recover a master key in approximately $2^{18.39}$ operations. This methodology poses a significant threat to several NIST Lightweight Cryptography (LWC) competition candidates, including GIFT-COFB, ESTATE, HYENA, and SUNDAE-GIFT.
Introduction
In the Internet of Things (IoT) ecosystem, cryptographic ciphers must be "lightweight," optimizing for minimal power consumption and hardware area. The GIFT cipher family is a prominent example, utilizing bit permutations—rearranging bits rather than complex transformations—which are computationally "free" in hardware.
However, these designs often face a trade-off in physical security. Traditional Differential Fault Analysis (DFA) is highly effective but requires active, invasive interference, such as laser injection or voltage glitching. Conversely, Side-Channel Analysis (SCA) is passive but often limited when the attacker lacks control over the input plaintext. This research bridges the gap by proposing a DFA-style attack that requires no physical fault injection, targeting the bit permutation layers of GIFT-128.
The DCSCA Methodology
The proposed DCSCA is a hybrid methodology that leverages statistical occurrences of "natural" differentials. The attack follows four primary steps:
- Passive Observation: The attacker observes a large number of encryptions of random, unknown plaintexts.
- Differential Logic: The attacker identifies pairs of ciphertexts whose differences suggest that only a single S-box (Substitution Box) in the final round received a different input.
- Side-Channel Assistance: To verify this "natural" differential, the attacker monitors side-channel leakage (e.g., power or EM) from the penultimate round. A change in only one specific intermediate value confirms the desired differential.
- Key Recovery: By analyzing these specific ciphertext pairs and their associated leakages, the attacker mathematically prunes the last-round key candidates until the master key is revealed.
Exploiting Bit Permutations
The vulnerability lies in the transparency of bit permutations. While they provide excellent diffusion by spreading influence across the state, they do not alter the bit values themselves. This allows an attacker with side-channel access to "trace" the output of an S-box to its destination in the subsequent round.
Because DCSCA operates backwards from the ciphertext—which is typically unmasked and accessible—it effectively bypasses many common SCA defenses that focus on masking the plaintext.
Key Results and Performance
The attack is remarkably efficient. Simulations indicate that the master key of GIFT-128 can be recovered using roughly $2^{13.39}$ AEAD sessions (assuming 32 encryptions per session), a task well within the capabilities of modern attackers.
A critical factor in the attack's success is the Signal-to-Noise Ratio (SNR) of the side-channel measurements. The authors demonstrated that the attack remains highly practical on real hardware (simulated as an ARM Cortex-M4) provided the side-channel SNR is $\ge 16$.
The broad impact of this vulnerability is significant, as it successfully targets four NIST LWC candidates: GIFT-COFB, ESTATE, HYENA, and SUNDAE-GIFT.
Limitations and Countermeasures
While the attack is powerful, it has specific requirements:
- SNR Sensitivity: In extremely noisy or shielded environments, distinguishing the "natural" differentials becomes significantly more difficult.
- Countermeasures: Standard SCA defenses such as masking (splitting variables into random shares) and shuffling (randomizing operation order) can mitigate DCSCA. However, these protections often impose performance overheads that conflict with the "lightweight" goals of these ciphers.
Future research will likely explore how different implementation styles, such as bitslicing versus nibble-based approaches, influence leakage patterns and the overall feasibility of DCSCA in diverse hardware environments.
Summary updated