Research & development

Seven lines of work, one conviction.

Intelligence, frugality and security cannot be bolted on afterwards. Each R&D line below feeds the others — and several have already become products.

CELTIC-NEXT RAI-6Green

Energy · Edge AI · European R&D

Edge Energy Arbitration

CELTIC-NEXT RAI-6Green

Real-time arbitration of residential energy assets on the nYOI Energy Arbiter, federated learning between appliances, data sovereignty by design.

ANEMOD’s energy arbitration line takes decisions where the data lives. Each nYOI appliance forecasts and arbitrates PV, battery, EV charging and HVAC locally, then joins a federated learning loop so policies improve across homes without any raw load curve leaving the household.

The work is developed inside the CELTIC-NEXT RAI-6Green project — three edge nodes have been in continuous service since Q1 2026. It is structured as three complementary layers on the same appliance: a bi-directional data connector for residential assets, a safe federated reinforcement-learning arbitration engine, and an edge-coordinated marketplace that pools household flexibility for local energy communities, aggregators and DSOs.

Sub-10 W ARM64 appliance, offline-capable
SAREF4ENER / IEC CIM harmonization, NGSI-LD streams
Federated RL with a mathematical safety shield
mTLS, signed firmware, NIS2 / IEC 62443 alignment
Connector and libraries open source (Apache 2.0)

Artificial Intelligence

SLMs / Edge AI

Small Language Models optimized for embedded systems — AI that runs on the device, without a cloud round-trip, and keeps its accuracy through knowledge distillation.

ANEMOD develops compact language and decision models sized for mobile NPUs and TPUs. Distillation from larger teachers, quantization and pruning bring inference on-device with ultra-low latency and a fraction of the energy budget.

Compact models, fewer parameters
Low energy consumption
Deployment on mobile NPU / TPU
Ultra-low latency inference

Hardware cybersecurity

H2S — Hardware Security Server

A secure element combining HSM functions with a TLS 1.3 server, for the secure distribution of cryptographic keys — the first TLS 1.3 server embedded in a secure element.

H2S puts the TLS 1.3 endpoint inside certified silicon. Keys are generated, stored and used within an EAL6+ secure element; the host only ever sees ciphertext. The concept scales from a pocket device (LeMonolith) to a 16-element grid (GRID001) and underpins ANEMOD’s zero-trust architectures for edge and 6G infrastructures.

EAL6+ / EAL7 certification path
TLS 1.3 inside the secure element
Cryptographic key protection
Zero-trust architecture

Energy optimization

Energy Frugality

Distillation, quantization, pruning and low-rank factorization to cut the energy cost of AI models drastically — without sacrificing performance.

Frugality is a design constraint, not an afterthought. ANEMOD applies model compression end to end — from FP32 to INT8/INT4 quantization, structured and unstructured pruning, low-rank factorization — so that intelligence fits the power envelope of embedded and edge hardware.

Knowledge distillation
Quantization (FP32 → INT8 / INT4)
Structured and unstructured pruning
Low-rank factorization

Intelligent cybersecurity

Cyber-vaccine & IoB

Behavioral AI and the Internet of Behaviors to predict and neutralize attacks before they happen — proactive defense instead of reactive response.

By continuously tracking behaviors over time and feeding them to behavioral AI, ANEMOD builds a Cyber Threat Intelligence foundation that anticipates attacker moves and “vaccinates” systems ahead of the attack.

Advanced behavioral AI
Threat prediction
Cyber Threat Intelligence (CTI)
Continuous temporal analysis

Networks & embedded systems

6G / Embedded Systems

Embedded systems ready for 6G networks: interconnected femto-data centers with low-power GPU/NPUs, operating autonomously and distributed.

ANEMOD designs the control software and hardware for mobile embedded digital infrastructures — femto-data centers that compute, secure and decide at the edge of tomorrow’s 6G networks.

6G-ready architecture
Distributed femto-data centers
Low-power GPU / NPU
Autonomous, interconnected systems

Security assurance

EAL6+ / EAL7 Certification

Reaching the highest assurance levels for end-to-end protection of AI models, data and communications in critical applications.

ANEMOD’s security work targets Common Criteria EAL6+ / EAL7 — the levels required for the most sensitive systems — with end-to-end protection, advanced resistance testing and international certification.

Common Criteria compliance
End-to-end protection
Advanced resistance testing
International certification

Publications

Published work behind these lines.

Peer-reviewed articles and reference works by ANEMOD’s team — Guy Pujolle (President of ANEMOD, Emeritus Professor at Sorbonne University) and co-authors.

Published 2025

Intelligent Energy Arbitration in Edge Networks with Distributed Digital Twins

K. Al Agha, G. Pujolle — IEEE Wireless Days 2025

Introduces an AI-driven arbitrator coordinating distributed digital twins to decide where, when and in which form edge functions run, minimizing a cost that combines total power, variance, peaks and carbon intensity — the theoretical basis of ANEMOD’s energy arbitration engine.

energyedgefrugality R&D: Edge Energy Arbitration
Reference book 2024

Les Réseaux

G. Pujolle — Eyrolles, 10th edition

The French-language reference on networking, continuously updated since 1985 — from protocols and architectures to edge, cloud and 6G.

edge
Published 2023

Internet of Edges architecture for 6G

G. Pujolle — 2nd International Conference on 6G Networking (6GNet), IEEE

Defines the Internet of Edges: 6G trust zones built from nodes embedding very small data centers, processing most requests locally and communicating in direct mode — the architecture in which ANEMOD’s edge appliances and femto-data centers operate.

edge doi:10.1109/6GNET58894.2023.10317768 R&D: 6G / Embedded Systems