S-box Pooling: Towards More Efficient Side-Channel Security Evaluations

Abstract

Nowadays, profiled attacks are the standard penetration tests for security evaluations. Often the security evaluators have to perform profiled attacks on each S-box to quantify the security strength of the target symmetric cryptographic algorithm implementations more accurately. The required time to conduct such profiled attacks is very long due to the number of profiling traces (for many certification bodies, at least 1,000,000 are mandated). It is getting even more time-consuming after introducing deep learning profiled attacks. Furthermore, some certification bodies instruct up to 5,000,000 or 10,000,000 profiling traces because modern embedded secure IC products have more and more countermeasures against side-channel attacks. It is a challenge to simultaneously decrease the number of required profiling traces and the required profiling time while retaining the attack performance for profiled attacks. In this work, we propose a simple yet remarkably effective pooling approach to address this problem for security evaluations. That is, pooling over the S-boxes to build a large profiling set and perform the profiling on this large set once. Intensive experiments are conducted with this pooling approach using different profiling tools (template attack and its pooled variant, stochastic model and deep learning) on three different AES implementations (a sequential S-box software AES implementation without masking, a sequential S-box software AES implementation with first-order masking and a parallel S-box hardware AES implementation with first-order masking). The experimental results have shown that the proposed pooling approach can lead to similar attack performance while decreasing both the required number of profiling traces and the required profiling time by a factor of 8 or even 16.