The workshop takes place on October 11, 2026.
The venue is in the same building as CHES, but a separate
registration to the CHES-affiliated event (PROOFS 2026)
is mandatory.
Meeting Room 4
|
Sunday, October 11, 2026
|
| 09:00 – 10:05 |
Opening + Invited Talk (TBD) |
| 10:05 – 10:30 |
Coffee Break |
| 10:30 – 11:00 |
Contributed Paper #1
VeriSRAM: Placement-Constrained In-SRAM Cryptography
for Microcontrollers
|
| 11:00 – 11:30 |
Contributed Paper #2
Behavioral Security Invariants for Proactive Ransomware
Detection in Embedded and Cyber-Physical Systems
|
| 11:30 – 12:00 |
Contributed Paper #3
Gem5-ML: Machine Learning-Enhanced Cycle-Accurate
Simulation for IoT Microcontroller Design Space
Exploration and Anomaly Detection
|
| 12:00 – 12:30 |
Contributed Paper #4
RISCGuard: RISC-V Static Analysis Framework for
Detecting Microarchitectural Vulnerabilities (online)
|
| 12:30 – 13:30 |
Lunch |
Bio:
Vincent Grosso is a CNRS researcher in Saint-Étienne, France. His research
focuses on the security of cryptographic implementations, with a particular
interest in physical attacks, including side-channel and fault attacks, and
their countermeasures. His work covers both symmetric and post-quantum
cryptography, as well as the use of machine learning and algebraic techniques
for side-channel analysis. More recently, he has been exploring the interplay
between artificial intelligence and physical security, both using AI to improve
physical attacks and investigating physical attacks against neural network
implementations.
Abstract:
Artificial intelligence is increasingly changing the way we analyze the physical
security of embedded systems. Machine learning, and neural networks in particular,
can provide powerful tools for exploiting side-channel leakages, sometimes
challenging the traditional assumptions and methodologies used in side-channel
analysis.
In this talk, we will explore different interactions between AI and physical
attacks. We will first revisit the use of neural networks for classical
side-channel attacks, before considering collision-based approaches with different
levels of supervision. We will then discuss how machine learning can assist other
stages of a side-channel attack, such as preprocessing and leakage detection.
Finally, we will turn the problem around: as neural networks are increasingly
deployed on embedded devices, they themselves become valuable targets for physical
attacks. We will discuss ongoing work on extracting information about neural
network models and their parameters from their physical implementations.
Overall, this talk aims to illustrate the evolving relationship between AI and
physical security: AI can be a powerful tool for the attacker, but AI implementations
can also become the target. And, appropriately enough, part of this abstract was
generated with the help of an AI.
