SCA strikes back : reverse engineering neural network architectures using side channels

Our previous work selected for Top Picks in Hardware and Embedded Security 2020 demonstrates that it is possible to reverse engineer neural networks by using side-channel attacks. We developed a framework that considers each part of the neural network separately and then, by combining the information, manages to reverse engineer all relevant hyper-parameters and parameters. Our work is a proof of concept (but also a realistic demonstration) that such attacks are possible and warns that more effort should be given to developing countermeasures. While we have used microcontrollers for our experiments, the attack applies to other targets like FPGAs and GPUs.National Research Foundation (NRF)Accepted versionThis research is partly supported by the National Research Foundation, Singapore, under its National Cybersecurity Research & Development Programme / Cyber-Hardware Forensic & Assurance Evaluation R&D Programme (Award: NRF2018NCR-NCR009-0001