Recursive Diffusion Layers for (Lightweight) Block Ciphers and Hash Functions

Diffusion layers with maximum branch numbers are widely used in block ciphers and hash functions. In this paper, we construct recursive diffusion layers using Linear Feedback Shift Registers (LFSRs). Unlike the MDS matrix used in AES, whose elements are limited in a finite field, a diffusion layer in this paper is a square matrix composed of linear transformations over a vector space. Perfect diffusion layers with branch numbers from 5 to 9 are constructed. On the one hand, we revisit the design strategy of PHOTON lightweight hash family and the work of FSE 2012, in which perfect diffusion layers are constructed by one bundle-based LFSR. We get better results and they can be used to replace those of PHOTON to gain smaller hardware implementations. On the other hand, we investigate new strategies to construct perfect diffusion layers using more than one bundle-based LFSRs. Finally, we construct perfect diffusion layers by increasing the number of iterations and using bit-level LFSRs. Since most of our proposals have lightweight examples corresponding to 4-bit and 8-bit Sboxes, we expect that they will be useful in designing (lightweight) block ciphers and (lightweight) hash functions.