Profiled Attacks Against the Elliptic Curve Scalar Point Multiplication Using Neural Networks

In recent years, machine learning techniques have been successfully applied to improve side-channel attacks against different cryptographic algorithms. In this work, we deal with the use of neural networks to attack elliptic curve-based cryptosystems. In particular, we propose a deep learning based strategy to retrieve the scalar from a double-and-add scalar-point multiplication. As a proof of concept, we conduct an effective attack against the scalar-point multiplication on NIST standard curve P-256 implemented in BearSSL, a timing side-channel hardened public library. The experimental results show that our attack strategy allows to recover the secret scalar value with a single trace from the attacked device and an exhaustive search over a set containing a few hundreds of the sought secret.