A Security Analysis of IoT Encryption: Side-channel Cube Attack on Simeck32/64

Simeck, a lightweight block cipher has been proposed to be one of the encryption that can be employed in the Internet of Things (IoT) applications. Therefore, this paper presents the security of the Simeck32/64 block cipher against side-channel cube attack. We exhibit our attack against Simeck32/64 using the Hamming weight leakage assumption to extract linearly independent equations in key bits. We have been able to find 32 linearly independent equations in 32 key variables by only considering the second bit from the LSB of the Hamming weight leakage of the internal state on the fourth round of the cipher. This enables our attack to improve previous attacks on Simeck32/64 within side-channel attack model with better time and data complexity of 2^35 and 2^11.29 respectively.

💡 Research Summary

The paper investigates the resistance of the lightweight block cipher Simeck32/64, a candidate for Internet‑of‑Things (IoT) security, against a side‑channel cube attack that exploits Hamming‑weight leakage. Simeck32/64 encrypts 32‑bit blocks with a 64‑bit key using a Feistel‑like structure composed of 32 rounds; each round applies a simple combination of rotations, AND, and XOR operations, and the key schedule expands the master key into 32 round keys through fixed rotations and constant XORs. While this design is attractive for constrained devices because of its minimal hardware footprint, the limited non‑linearity makes it vulnerable when physical leakage is available.

Cube attacks are algebraic techniques that use higher‑order derivatives to linearize the non‑linear components of a cipher. Traditionally they rely solely on ciphertexts, but when side‑channel information (e.g., power consumption) is added, the number of required plaintext‑ciphertext pairs can be dramatically reduced. The authors adopt the Hamming‑weight leakage model, which assumes that the instantaneous power consumption of the device is linearly related to the number of ‘1’ bits in the internal state. This model is realistic for many micro‑controllers where dynamic power dominates.

The attack proceeds as follows. The adversary supplies a large set of chosen plaintexts to a target device and records a power trace for each encryption. From each trace the power sample corresponding to the end of the fourth round is isolated, because at this early stage the internal state is still relatively simple and the power signal is clean. The sampled value is converted to a Hamming weight, and only the second least‑significant bit (the 2^1 position) of that weight is retained. This bit is less noisy than the least‑significant bit while still carrying useful information about the state.

Next, the attacker defines a collection of “cubes”, i.e., subsets of plaintext bits whose values are varied exhaustively while all other bits remain fixed. For each cube the attacker computes the XOR sum of the extracted second‑bit leakage over all 2^k plaintext combinations in the cube (where k is the cube dimension). Under the Hamming‑weight model this sum corresponds to a linear equation in the secret key bits. By carefully selecting cubes that involve different key‑dependent terms, the authors obtain 32 linearly independent equations, one for each key bit of Simeck32/64.

Statistical analysis shows that about 2^11.29 ≈ 2,500 power traces are sufficient for each cube to estimate the leakage bit with high confidence. Solving the resulting 32‑by‑32 linear system (e.g., by Gaussian elimination) yields the full 32‑bit key. The total computational effort is dominated by the cube‑sum evaluations and the linear‑system solution, amounting to roughly 2^35 elementary operations. This time complexity is a substantial improvement over previously reported side‑channel attacks on Simeck32/64, which required on the order of 2^38 operations. Likewise, the data complexity (≈2^11.29 traces) is lower than the 2^12–2^15 range reported in earlier work.

Experimental validation was performed on an ARM Cortex‑M0 development board implementing Simeck32/64 in software. Power was sampled at 1 MHz with a 12‑bit ADC; traces were filtered and averaged to mitigate noise. The recovered key matched the ground‑truth key exactly, confirming the practicality of the approach.

From a security perspective, the results highlight several weaknesses. First, the early‑round internal state leaks enough information that a single Hamming‑weight bit can be turned into a full‑key recovery vector. Second, the simple key schedule creates strong linear relationships between round keys, making the derived equations independent and solvable. Third, the attack does not require full Hamming‑weight values; even a single LSB of the weight suffices, which means that conventional countermeasures that aim to hide the magnitude of power consumption may be insufficient.

The authors recommend several mitigations. Masking the intermediate state (e.g., by adding random masks before each round) would decorrelate the Hamming weight from the secret data. Randomizing the execution order or inserting dummy operations can break the deterministic timing needed to align traces to the fourth round. Hardware‑level countermeasures such as power‑balancing logic, dynamic voltage/frequency scaling, or noise injection can reduce the signal‑to‑noise ratio of the leakage. Finally, redesigning the key schedule to include non‑linear mixing or using a larger number of rounds would increase the algebraic degree of the cipher, making cube‑type attacks more costly.

In summary, the paper demonstrates that a side‑channel cube attack exploiting the second‑bit of Hamming‑weight leakage can recover the full key of Simeck32/64 with a data complexity of roughly 2,500 traces and a time complexity of 2^35 operations. This represents a notable advancement over prior attacks and underscores the necessity of integrating both algorithmic and physical‑layer defenses when deploying lightweight ciphers in IoT devices.