An Optimised Constant-Time Implementation of KASUMI FI Function

Substitution boxes (S-boxes) are a key component of many modern ciphers and their optimization can contribute to significant overall performance improvements. Although often implemented as lookup tables, mathematically, many common S-boxes can be represented as non-linear, Boolean mappings. This property can be leveraged to efficiently compute S-box values in constant time, thereby securing implementations against timing attacks. We propose an acceleration method for one such S-box, that of the KASUMI cipher, and integrate it into the Intel(R) IPSec Multi-Buffer Library [9] implementation. The Kasumi algorithm is a block cipher, widely used in 3G mobile communication networks. It was specified for the 3rd Generation Partnership Project (3GPP) to ensure the confidentiality and integrity of wireless data communications, such as voice and data transmissions. Although the prevalence of 3G is declining, Kasumi remains in operation in legacy systems and is consequently still provided as part of the Intel(R) IPSec Multi-Buffer Library. The optimization techniques proposed demonstrate the acceleration capabilities of modern CPU architectures for parallel S-box computation (and the enclosing FI function in Kasumi) and may hold the potential for wider application to cryptographic algorithms which incorporate Boolean S-boxes. The proposed implementation is an optimised version of the existing constant-time implementation of KASUMI in the Intel(R) IPSec Multi-Buffer Library, improving performance by a factor of 9.47.