Ph.D. researcher in hardware security
Sorbonne Université · LIP6 · CNRS
Ioannis
Kaskampas
Security and trust for neuromorphic AI: attacks on spiking neural networks and the defenses that stop them, from software to silicon and across federated learning.
reading clean input
- Doctorate, since 2024 Sorbonne Université LIP6 · CNRS, expected May 2027
- Focus Security & trust for neuromorphic AI hardware
- Background Computer engineering Integrated master's, University of Patras
- Latest paper IEEE AICAS 2026 SNN2Bitstream: from trained SNN to FPGA bitstream
Research
What I work on at LIP6, and the hardware it is grounded inI am a Ph.D. researcher in the CIAN team at LIP6 (Sorbonne Université, CNRS), supervised by Dr. Haralampos-G. Stratigopoulos. My research asks a simple question of modern AI hardware: can you trust the chip? I study security attacks on spiking neural networks (SNNs), both in software and in neuromorphic hardware, and the defenses that detect and stop them.
A second line of work brings spiking networks to federated learning, where many devices train one shared model without pooling their data. I explore SNNs across the different forms of federated learning and under client heterogeneity, and study how to keep these systems secure when some participants or their hardware cannot be trusted. Finally, I use neural architecture search (NAS), with federated learning for training and evaluation, to find architectures suited to FPGA deployment.
-
Hardware security and trust
Can the chip that runs the model be trusted? Detecting hidden malicious logic such as hardware Trojans, checking the integrity of the silicon, and protecting spiking networks and neuromorphic chips from attacks that software cannot see.
-
Spiking neural networks
Event-driven models that integrate input spikes until a threshold is crossed and then fire, computing with sparse binary events instead of dense numbers.
-
Federated learning
Many devices train one model without pooling their data; each round sends updates in and a global model back out, on hardware that has to be trusted.
-
Neuromorphic computing
Chips built like the networks they run: many small cores that pass spikes to one another and work only when an event arrives, prototyped on FPGA from a trained model to a bitstream.
Recent publications
Most recent first-
SNN2Bitstream: Automated Model-to-FPGA HLS-Based Framework for Spiking Neural Network Hardware Accelerators
2026 IEEE 8th International Conference on Artificial Intelligence Circuits and Systems (AICAS), Ha Long, Vietnam, September 2026.
-
Input-Triggered Hardware Trojan Attack on Spiking Neural Networks
IEEE International Symposium on Hardware Oriented Security and Trust (HOST), San Jose, CA, USA, May 2025.
Biography
Patras, the Hellenic Marines, then ParisIoannis Kaskampas obtained his Integrated Master's degree in Computer Engineering and Informatics from the University of Patras (Polytechnic School) in December 2022, ranking in the top 15% of his cohort (7.28/10). His thesis, graded 10/10, implemented a low-cost thermal-imaging system with real-time hotspot detection on FPGA, built on a MicroBlaze-based SoC with custom hardware IP for thermal image processing.
From March to November 2023 he served in the Hellenic Army Special Forces (575 Marine Battalion) as a Reservist Marine, Signal Specialist and Cryptologist. He operated and maintained cryptologic devices, radios and communication systems, handled the main data server for maintenance and operational tasks, and was promoted from Private to Lance Corporal in the final month of his service.
Since May 2024 he has been a Ph.D. Researcher at LIP6 (Sorbonne Université, CNRS) in Paris, working toward a doctorate in Electrical and Computer Hardware Engineering, expected in May 2027. He is fluent in Greek (native) and proficient in English (C2).