Research Engineer (M/F) Energy-Proportional On-Chip Communication Networks for Multi-Chiplet Systems

Job Description

Missions

The rise of large AI models, particularly LLMs and deep neural networks, has significantly increased computational, memory, and communication demands. This evolution is accompanied by a shift toward multi-chiplet architectures, which improve modularity, integration density, and design flexibility. In these systems, Networks-on-Chip (NoCs) play a central role, interconnecting compute cores, memory units, specialized accelerators, and inter-chiplet interfaces. However, traditional NoCs remain poorly suited to AI workloads: their power consumption is not always proportional to actual activity, and their behavior poorly adapts to traffic variability. It is therefore necessary to design energy-proportional NoCs capable of adapting their power consumption and communication mechanisms to actual usage levels while maintaining the performance required by AI applications.
This project aims to explore, design, and evaluate energy-proportional NoC architectures for multi-chiplet AI-dedicated systems. The work will focus on four main axes.
First, it will involve characterizing the traffic generated by different model families, particularly LLMs and deep neural networks. The goal is to identify critical phases, dominant communication patterns, and temporal variability in data exchanges. Representative benchmarks will be developed to assess the suitability of NoCs for real-world applications.
The project will also investigate reconfigurable topologies, selective power-down mechanisms, adaptive routing strategies, and dynamic management techniques for links, buffers, and data flow control (synchronous or asynchronous). The challenge is to reduce power consumption without excessively penalizing performance.
The proposed solutions will be evaluated using NoC simulation tools, including BookSim, combined with appropriate power models. The analysis will rely on metrics such as power consumption, latency, throughput, temporal variability, and hardware cost. A comparison with the state of the art will be expected.
The project may include interaction with software and algorithmic layers to adapt certain communication schemes to the constraints of energy-proportional NoCs. Techniques such as compression, quantization, or traffic reduction may also be explored.

Activity

The planned activities include proposing new NoC designs, implementing them, and conducting associated validation experiments. Active participation in the various tasks of the collaborative project funding this work is also expected.

Your Profil

Skills

The candidate must hold at least a Master's degree or an engineering degree in system architecture, microelectronics, embedded systems, or NoC design.
Experience in modeling, simulation, or prototyping of Networks-on-Chip will be highly valued. Proficiency with tools such as BookSim or OMNeT++, as well as programming languages like C, SystemC, or VHDL, will be an asset.
Skills in power evaluation using tools like McPAT, CACTI, or other power estimation solutions will also be sought. Experience with FPGA or ASIC design flows would be a plus.
Finally, the candidate must demonstrate autonomy, rigor, a strong collaborative spirit, and excellent proficiency in scientific English.

Your Work Environment

Teamwork and collaboration with other team members working on related topics, as well as with the other partners in the collaborative project.

Constraints and risks

None

Compensation and benefits

Compensation

Starting from €3175 gross per month, with adjustments based on professional experience in equivalent-level positions.

Annual leave and RTT

44 jours

Remote Working practice and compensation

Pratique et indemnisation du TT

Transport

Prise en charge à 75% du coût et forfait mobilité durable jusqu’à 300€

About the offer

About the CNRS

The CNRS is a major player in fundamental research on a global scale. The CNRS is the only French organization active in all scientific fields. Its unique position as a multi-specialist allows it to bring together different disciplines to address the most important challenges of the contemporary world, in connection with the actors of change.

CNRS

The research professions