Neural Network Architectures for Second Order Side Channel Attacks

Graduate internship (Master 2), Summer 2026

💼 About the role

PQShield is a cybersecurity scaleup that specialises in post-quantum cryptography, protecting information from today's attacks while preparing organisations for the threat landscape of tomorrow. It demonstrates quantum-safe cryptography on chips, in applications and in the cloud. We are headquartered in Oxford, with additional teams in the Netherlands, Germany and France.

PQShield SAS, based in Paris (France), concentrates the research activities of PQShield. Our mission is to come up with innovative algorithmic and/or protocol-level solutions to real-world cryptographic problems. Besides post-quantum cryptographic primitives, our research interests include implementation security and advanced cryptosystems and protocols such as secure messaging, threshold schemes, and multiparty computation.

In order to compute key exchanges or digital signatures, embedded cryptographic devices require the use of a secret key that must remain confidential. Side channel attacks (SCA) exploit statistical dependencies between the power consumption of a cryptographic device and the value of data on which it computes in order to recover the secret key. Countermeasures against SCA, such as masking, rely on the randomization of secret data during the computation of the cryptographic algorithm.

The use of machine learning in SCA on masked implementations has become increasingly popular over the last decade (https://eprint.iacr.org/2025/471). Neural networks seem particularly well suited for modeling how the data inside a processor impacts its power consumption. In particular, recent works have tried to design specific neural network architectures to attack masked implementations.

Objectives

The intern will use PQShield’s tools to design neural networks and test their effectiveness on existing datasets that consist of power measurements of PQShield’s post quantum cryptographic hardware accelerator and/or open source implementations.

This project comprises the following phases:

1. Literature review
• Understand the background of the datasets: SCA
• Review the supervised machine learning methods used in the state of the art
2. Practical Evaluations
• Explore the hyperparameter space for neural networks and design neural network architectures by taking into account the specifics of the power measurements within the datasets
• Compare the effectiveness of supervised learning methods other than neural networks
• Analyse the effectiveness of the various neural network architectures on datasets generated from implementations protected by various SCA countermeasures

🎯 Required Skills And Qualifications

Ideal candidates should possess:

• Prior experience in supervised machine learning methods for classification problems, including neural networks
• Proficiency in Python and experience with Keras/Tensorflow
• Some familiarity with cryptography
• Excellent communication skills
• Optional but big plus: familiarity with SCA

Supervision and Mentorship: The intern will be mentored by members of PQShield's Product Security Team in our office in Paris, with frequent progress meetings to facilitate knowledge exchange and to track progress.

Interested? Apply today to be part of the future of secure cryptography! 🚀