S-NET

Abstract

Side channel attacks are recognized as one of the most powerful attacks due to their ability to extract secret key information by analyzing the unintended leakage generated during operation. This makes them highly attractive for attackers. The current countermeasures focus on either randomizing the leakage by obfuscating the power consumption of all operations or blinding the leakage by maintaining a similar power consumption for all operations. Although these techniques help hiding the power-leakage correlation, they do not remove the correlation completely. This paper proposes a new countermeasure type, referred to as confusion, that aims to break the linear correlation between the leakage model and the power consumption and hence confuses attackers. It realizes this by replacing the traditional SBOX implementation with a neural network referred to as S-NET. As a case study, the security of Advanced Encryption Standard (AES) software implementations with both conventional SBOX and S-NET are evaluated. Based on our experimental results, S-NET leaks no information and is resilient against popular attacks such as differential and correlation power analysis.