optimizing Diffusion Layer with AVX Shuffling: A Study on SKINNY

The development of the Internet of Things (IoT) requires efficient information transmission among resource-limited devices. Designing lightweight block ciphers with high software and hardware efficiency becomes a research hotspot. The design of block ciphers needs to follow the principles of diffusion and confusion. Nonlinear sboxes are broadly used in confusion layers. In contrast, linear bit/byte permutations or diffusion matrices are generally used in the diffusion layer. This paper designs a lightweight diffusion layer for iterative lightweight 8-bit sboxes. The fast shuffling operations of AVX instructions are used in block cipher designs. In addition, the hardware and software performances are related to the number of XORs for the diffusion matrix operation. For different sboxes, the selection of the optimal diffusion layer is diverse. The main idea is to replace the ShiftRows of the SKINNY with a byte permutation and design a diffusion matrix. We use the MILP method to search the differential and linear paths of the modified SKINNY. Our scheme can improve the lower bounds on the number in both differential and linear active sboxes. Moreover, software performance is also improved by using AVX512 instructions.