Block ciphers like AES derive their security from a single nonlinear primitive: the Substitution-Box (S-Box). Everything else — key mixing, permutation, diffusion — is linear and therefore invertible by an attacker with enough equations. The S-Box is the barrier that breaks linearity. If it is weak, the cipher is weak.
What Makes a Good S-Box?
Cryptographers measure S-Box quality using several formal criteria:
- Nonlinearity (NL) — Hamming distance from the nearest affine function. Higher is better; theoretical maximum for 8→8 bit is 120.
- Strict Avalanche Criterion (SAC) — Flipping one input bit should flip each output bit with probability ½.
- Differential Uniformity (DU) — Lower values resist differential cryptanalysis. AES achieves DU = 4.
- Bit Independence Criterion (BIC) — Output bit pairs should behave independently.
The Chaos Connection
Chaotic dynamical systems — systems governed by differential equations with extreme sensitivity to initial conditions — produce pseudo-random trajectories that are deterministic but practically unpredictable. When we map these trajectories to the input-output space of an S-Box, we get bijections with naturally high nonlinearity.
In our PSO-based work (IEEE Access, 2018), we combined 2D Lorenz-derived chaotic maps with Particle Swarm Optimization to search the high-NL region of the S-Box landscape. The PSO objective function maximised NL and penalised DU above 4. The result: S-Boxes with NL ≥ 108, outperforming many AES alternatives.
Extending to Medical Image Encryption
Our more recent work (Fuzzy Logic S-Box, IEEE Access 2023) addressed a different threat: securing DICOM medical images in telemedicine pipelines. We extended the hyperchaotic S-Box construction with a fuzzy membership function that adapts the substitution mapping based on pixel intensity distributions, producing encryption schemes that are both statistically robust and computationally lightweight for hospital edge devices.
Current Research Direction
We are now exploring 4D memristive hyperchaotic maps — a class of systems with memory-dependent dynamics — that offer significantly larger Lyapunov exponent spectra and therefore better S-Box candidates. Early results show differential uniformity ≤ 4 with nonlinearity approaching 112.