General information
Organisation
The French Alternative Energies and Atomic Energy Commission (CEA) is a key player in research, development and innovation in four main areas :
• defence and security,
• nuclear energy (fission and fusion),
• technological research for industry,
• fundamental research in the physical sciences and life sciences.
Drawing on its widely acknowledged expertise, and thanks to its 16000 technicians, engineers, researchers and staff, the CEA actively participates in collaborative projects with a large number of academic and industrial partners.
The CEA is established in ten centers spread throughout France
Reference
2026-41996
Position description
Category
Miscellaneous
Contract
Internship
Job title
Adaptive Learning for Transformers with ReRAM Analog In-Memory Computing
Subject
This internship investigates how Transformer models can adapt after deployment using ReRAM analog in-memory computing. Only a subset of the network parameters remains trainable, while the rest stays fixed. The challenge is to maintain effective adaptation with strongly quantized weights and a limited number of programming events imposed by ReRAM endurance. The project will explore approaches such as Low-Rank Adaptation (LoRA) and retraining selected MLP layers, with the aim of defining architectures and learning strategies for energy-efficient adaptation on analog AI hardware.
Contract duration (months)
4 to 6
Job description
Within CEA-Leti’s Laboratory of Memory and Computing Devices in Grenoble, you will work on the algorithm-hardware co-design of adaptive Transformer architectures using ReRAM analog in-memory computing.
Using PyTorch and in-house ReRAM characterization data, you will develop reference models and compare different choices for the trainable part of the network. You will evaluate how weight quantization, device variability and restricted programming budgets affect fine-tuning performance, and investigate learning strategies compatible with these constraints.
The expected outcome is to identify viable trade-offs between adaptation performance, the number of trainable parameters and memory programming requirements, and derive design guidelines for future analog AI accelerators.
The internship lasts 4 to 6 months. For outstanding candidates, a PhD continuation is possible in partnership with Weebit Nano.
Applicant Profile
M2 student in Electrical Engineering, Computer Science, Applied Mathematics, Data Science, Computational Neuroscience or related fields, with strong quantitative skills and experience in Python/PyTorch.
Interest in machine learning, neural networks and hardware-aware AI is expected. Prior experience with emerging memory devices or analog computing is a plus but not required.
Position location
Site
Grenoble
Job location
France, Auvergne-Rhône-Alpes, Isère (38)
Location
Grenoble
Candidate criteria
Languages
English (Intermediate)
Prepared diploma
Bac+5 - Master 2
PhD opportunity
Oui
Requester
Position start date
01/02/2026
Create candidate area
