Survey on the evolution and key technologies of differential distinguisher design

Abstract

The differential distinguisher is pivotal in differential cryptanalysis, distinguishing authentic from random differential distributions. This paper surveys its evolutionary trajectory and key technologies, covering the paradigm shift from traditional manual-driven to neural network-aided methods. Traditional distinguishers rely on Differential Distribution Tables and manual high-probability paths, offering strong interpretability but poor efficiency and adaptability to complex ciphers like Speck and Simon. Neural network-aided designs innovate in input feature engineering and network architectures. The paper identifies challenges including poor cross-cipher generalisation and insufficient interpretability, proposing future directions like adaptive design and lightweight optimisation. It serves as a comprehensive reference for lightweight cipher security analysis.