Automatic modelling language variations for socially responsive chatbots
Conversational agents are increasingly present in our daily lives thanks to advances in natural language processing and artificial intelligence and are attracting growing interest. However, their ability to understand human communication in all its complexity remains a major challenge. This PhD project aims to model linguistic variation to develop agents capable of socially adaptive interactions, taking into account the socio-demographic profile and emotional state of their interlocutors. It also focuses on evaluating linguistic cues at different levels, leveraging both spoken and written language varieties, and assessing the generalization capacity of models trained on multilingual and multi-situational data, with the goal of improving interaction modeling with conversational agents.
Compositional Generalization of Visual Language Models
The advent of the foundation models led to increase the state-of-the art performance on a large number of tasks in several fields of AI, in particular computer vision and natural language processing. However, despite the huge amount of data used to train them, these models are still limited in their ability to generalize, in particular for a use case of interest that is in a specific domain, not well represented on the Web. A way to formalize this issue is compositional generalization, i.e. generalising to a new, unseen concept from concepts learned during training. This "generalization" is the ability to learn disentangle concepts and to be able to recombine
them into unseen composition when the model is in production. The proposed thesis will address this issue, aiming at proposing visual representations that enable generic visual language models to generalize compositionally within specific domains. It will investigate strategies to reduce shortcut learning, promoting deeper understanding of compositional structures in multimodal data. It will also address the problem of compositional generalization beyond simple attribute–object pairs, capturing more subtle and complex semantics. The proposed thesis aims at proposing preogress at a quite theoretical level but has many potential practical interest, in the fields of health, administration and services sectors, security and defense, manufacturing and agriculture.
Development of new strategies for robotic computed tomography with variable magnification
The Department of Numerical Instrumentation joins expertise on different research fields through experimental and software platforms. In the Monitoring, Control and Diagnostic Unit, one of the important research areas is the industrial inspection with X-ray methods. Within this framework, a robotic inspection facility is being employed for innovative research and validation of new algorithms and on instrumentation aspects.
One of the most important features of robotic inspection is the possibility to scan large samples. In most application cases, region-of-interest areas are defined, for which a higher spatial resolution is targeted. In this context, a research program covering several topics is proposed, with the main objective of facilitating the setup and the use of the robotic inspection configuration for industrial application cases. A focus can be set on one or two research topics, depending on the background and on personal R&D interests and initiatives of the candidate.
A first topic will consist of developing CT reconstruction algorithms for a scan configuration using a variable magnification ratio, in a first phase with analytical algorithms such as the one proposed by Dennerlein [1] and then to adapt iterative reconstruction algorithms of SART type.
A second topic will consist on a work related to adapt the algorithms towards a multi-resolution representation of the reconstructed volumes, through octree or wavelet decomposition. An approach involving the correlation of experimental data to the CAD model of the sample will allow a better implementation in order to improve the VOI (volume of interest) tomography.
A third topic focused on instrumentation will deal with the experimental validation and additionally it will aim to develop a system capable to verify the accurate positioning of the scene elements with the help of precision distance sensors. A simultaneous measure of the distance source - to part surface together with the radiographic image will allow implementing corrections for positioning errors for every scan point and in a second phase to use this additional information directly in the reconstruction process.
Merging Optomechanics and Photonics: A New Frontier in Multi-Physics Sensing
Optomechanical sensors are a groundbreaking class of MEMS devices, offering ultra-high sensitivity, wide bandwidth, and seamless integration with silicon photonics. These sensors enable diverse applications, including accelerometry, mass spectrometry, and gas detection. Optical sensors, leveraging photonic integrated circuits (PICs), have also shown great potential for gas sensing.
This PhD focuses on developing a hybrid multi-physics sensor, integrating optomechanical and optical components to enhance sensing capabilities. By combining these technologies, the sensor will provide unprecedented multi-dimensional insights, pushing MEMS-enabled silicon photonic devices to new limits.
At CEA-Leti, you will access world-class facilities and expertise in MEMS fabrication, photonics, and sensor integration. Your work will involve:
-Sensor Design – Using analytical Tools and simulation software for numerical analysis to optimize device architecture.
-Cleanroom Fabrication – Collaborating with CEA’s expert teams to develop the sensor.
-Experimental Characterization – Conducting optomechanical and optical evaluations.
-Benchmarking & Integration – Assessing performance with optics, electronics, and fluidics.
This PhD offers a unique chance to merge MEMS and silicon photonics in a cutting-edge research environment. Work at CEA-Leti to pioneer next-generation sensor technology with applications in healthcare, environmental monitoring, and beyond. Passionate about MEMS, photonics, and sensors? Join us and help shape the future of optomechanical sensing!
Towards a low-resistive base contact for the InP-HBT transistor
Join CEA LETI for an exciting technological journey! Immerse yourself in the world of III V
based transistors integrated on compatible CMOS circuits for 6 G future communications
This thesis offers the chance to work on a ambitious project, with potential to continue into
a thesis If you're curious, innovative, and eager for a challenge, this opportunity is perfect
for you!
As the consumption of digital content continues to grow, we can foresee that 6 G
communication systems will have to find more capacity to support the increase in traffic
New Sub THz frequencies based systems are a huge opportunity to increase data rate but
are very challenging to build and maturate the power amplifier required to transmit a
signal will have to offer sufficient power and energy efficiency which is not obtained with
actual silicon technology InP based HBTs (Heterojunction Bipolar Transistors) developed
on large Silicon substrates have the potential to meet the requirements and be integrated
as close as possible to the CMOS circuits to enable minimal system/interconnect losses
Sb based semiconductors for GaAsSb HBT are emerging as highly promising materials,
especially for its electrical properties to integrate the Base layer of the Transistor It is
therefore necessary to produce high performance electrical contacts on this type of
semiconductor while remaining compatible with the manufacturing processes of the Si Fab
technology platforms
Throughout
this thesis, you will gain a broad spectrum of knowledge, beneficiate from the
rich technical environment of the 300 200 mm clean room and the nano characterization
platform You will collaborate with multidisciplinary teams to develop a deep understanding
of the ohmic contacts and analyse existing measurements Several apsects of the metal
(Ni or Ti p GaAs 1 x Sb x contact will be investigated
•Identify wet and plasma solutions allowing the GaAsSb native oxide removing without
damaging the surface with XPS and AFM
•Characterize GaAs 1 x Sb x epitaxy doping level (Hall effect, SIMS, TEM)
•Understand the phase sequence during annealing between the semiconductor and the
metal with XRD and Tof SIMS Manage this intermetallic alloys formation to not
deteriorate the contact interface (TEM image associated)
•Evaluate electrical contact properties using TLM structures Measurement of the
specific contact resistivity, sheet resistance of the semiconductor ant transfer length
associated The student will be a motive force to perform electrical tests on an automatic prober
architecture for embedded system of Automated and Reliable Mapping of indoor installations
The research focuses on the 3D localization of data from measurements inside buildings, where satellite location systems, such as GPS, are not operational. Different solutions exist in the literature, they rely in particular on the use of SLAM (Simultaneous Localization And Mapping) algorithms, but the 3D reconstruction is generally carried out a posteriori. In order to be able to propose this type of approach for embedded systems, a first thesis was carried out and led to a choice of algorithms to embed and a draft of the electronic architecture. A first proof of concept was also realized. Continuing this work, the thesis will have to propose a method allowing the localization device to be easily embedded on a wide range of nuclear measuring equipment (diameter, contamination meter, portable spectrometry, etc.). The work is not limited to a simple integration phase; it requires an architectural exploration, which will be based on adequacy between algorithm and architecture. These approaches will make it possible to respect different criteria, such as weight and small size so as not to compromise ergonomics for the operators carrying out the maps and quality of the reconstruction to ensure the reliability of the input data for the Digital Twin models.
Cryogenic characterization of emerging memories for space and/or quantum computing applications
Low temperature computing is a new proposal to boost the technological performances beyond the frontiers in the aerospace, high performance servers, quantum computing and data center domain.
Different emerging technologies have been showing promising features at single device level during the ongoing Ph.D. work: the programmability of the OxRAMs was proved down to 4K and efforts were focused on the understanding of the interactions between the selector and the resistor composing the memory cell. FeRAMs show a better programming efficiency and stability at low temperature probably due to a crystallographic change driven by the program operation; hypothesis that lies unproved. PCM also showed a programmability down to 12K and may be included in the analysis.
Statistical behavior of R&D chip at low temperature will be the key theme of this proposition knowing that very few publications appeared in the scientific literature leaving much room for analysis and comprehension.
Throughout this Ph. D., you will gain a broad spectrum of knowledge, spanning cryogenics, microelectronics reliability and device physics. Different technologies developed in LETI will be statistically screened in this innovative scenario. Modeling of the conduction phenomena might be also considered. You will be part of a team of 7-8 people between permanent, researchers and students and you will be managed to share your work with them.
Bayesian Neural Networks with Ferroelectric Memory Field-Effect Transistors (FeMFETs)
Artificial Intelligence (AI) increasingly powers safety-critical systems that demand robust, energy-efficient computation, often in environments marked by data scarcity and uncertainty. However, conventional AI approaches struggle to quantify confidence in their predictions, making them prone to unreliable or unsafe decisions.
This thesis contributes to the emerging field of Bayesian electronics, which exploits the intrinsic randomness of novel nanodevices to perform on-device Bayesian computation. By directly encoding probability distributions at the hardware level, these devices naturally enable uncertainty estimation while reducing computational overhead compared to traditional deterministic architectures.
Previous studies have demonstrated the promise of memristors for Bayesian inference. However, their limited endurance and high programming energy pose significant obstacles for on-chip learning applications.
This thesis proposes the use of ferroelectric memory field-effect transistors (FeMFETs)—which offer nondestructive readout and high endurance—as a promising alternative for implementing Bayesian neural networks.
Field Effect Transistor with Oxide Semiconductor Channel: Multi-Level Synaptic Functions and Analog Neurons
This thrilling PhD position invites you to dive into the groundbreaking field of 2T0C (Two-Transistor, Zero-Capacitor) BEOL FET (Back-End-of-Line Field-Effect Transistor) based neurons and synapses, a revolutionary approach poised to transform neuromorphic computing. As a PhD student, you will be at the forefront of research that bridges advanced semiconductor technology with brain-inspired architectures, exploring how these innovative neuron circuits can emulate synaptic functions and enhance data processing efficiency.
Throughout this project, you will engage in hands-on design and characterization of cutting-edge 2T0C neuron circuits, utilizing state-of-the-art tools and techniques. You’ll collaborate with a dynamic, multidisciplinary team of engineers and researchers, tackling exciting challenges related to device performance and energy optimization.
Your work will involve extensive characterization of BEOL FET devices and circuits. You will have the opportunity to propose, specify and design new memory read architectures, that enables the exploration of multi-level synaptic behaviors toward the implementation of more energy and area efficient next-generation neuromorphic systems.
Join us for this unique opportunity to push the boundaries of technology and be part of a transformative journey that could redefine the future of computing! Your contributions could pave the way for breakthroughs in brain-inspired systems, making a lasting impact on the field.
DCDC converter based on chiplet for HPC applications
High-performance computing (HPC) is becoming more and more critical for the AI
advancement. HPC requires significant amount of power for the needed processing
and struggles with thermal dissipation issues. Recent research highlights the need for
innovative solutions to improve power management for AI processors.
In our research team, you will be responsible for developing an enhanced power
management unit to provide a stable power supply for high-performance processors.
By exploring cutting-edge DC-DC converter topologies and emerging silicon passive
components (inductors and capacitors), the main objective is to design a highly efficient
DC-DC converter that optimizes both power efficiency and density. This project also
involves analyzing power distribution networks and integrated circuit design to optimize
the overall power delivery efficiency with relatively small form factor.
As a PhD student, you will be engaged in various technical tasks, from system-level
analysis to IC design. Working in an IC-design lab, you will collaborate with digital
design and component teams to address both device- and system-level challenges.
Your tasks will be distributed across system architecture (40%), passive component
analysis (20%), and converter design (40%).