Attention-based Binarized Visual Encoder for LLM-driven Visual Question Answering
In the context of smart image sensors, there is an increasing demand to go beyond simple inferences such as classification or object detection, to add more complex applications enabling a semantic understanding of the scene. Among these applications, Visual Question Answering (VQA) enables AI systems to answer questions by analyzing images. This project aims to develop an efficient VQA system combining a visual encoder based on Binary Neural Networks (BNN) with a compact language model (tiny LLM). Although LLMs are still far from a complete hardware implementation, this project represents a significant step in this direction by using a BNN to analyze the context and relationship between objects of the scene. This encoder processes images with low resource consumption, allowing real-time deployment on edge devices. Attention mechanisms can be taken into consideration to extract the semantic information necessary for scene understanding. The language model used can be stored locally and adjusted jointly with the BNN to generate precise and contextually relevant answers.
This project offers an opportunity for candidates interested in Tiny Deep Learning and LLMs. It proposes a broad field of research for significant contributions and interesting results for concrete applications. The work will consist of developing a robust BNN topology for semantic scene analysis under certain hardware constraints (memory and computation) and integrating and jointly optimizing the BNN encoder with the LLM, while ensuring a coherent and performant VQA system across different types of inquiries.
Turbulence synthesization methods for hybrid URANS/LES CFD approaches in multi-scale simulation of nuclear cores
Problem description: Fluid-structure interactions in nuclear reactor cores are a result of mechanisms occurring at different space scales. The component scale represents the global flow inside the core and is generally simulated though porous media methods. The local scale represents the fuel assembly: it requires CFD scale-resolving methods to calculate consistent fluid forces on the structures, and it features a certain degree of fluid-structure coupling. With the goal of performing multi-scale simulations of a core, the local scale requires the generation of boundary conditions from the component scale. This can be achieved only by a synthetic generation of turbulence, based on the flow results at the component scale. However, the porous media approach used at the component scale does not contain details on the turbulent quantities: the development of new numerical methods is required for generating consistent synthetic turbulence in this configuration.
Objectives:
1. Identify proper hybrid URANS/LES approaches for fuel assembly vibration related issues
2. Identify available turbulence parameters in porous media methods and explore bottom-up scaling approaches
3. Develop a turbulence synthesization method applicable to any fuel array inside a core
Expected results:
1. A novel approach for fluid-induced vibration analysis based on a multi-scale method
2. Clarify the key parameters to generate proper turbulence-resolved boundary conditions in the specific configuration studied
3. Validate the new methods on available experimental configurations
Designing a fast reactor burnup credit validation experiment in the JHR reactor
The primary mission of the Jules Horowitz experimental nuclear Reactor (JHR) is to meet the irradiation needs of materials and fuels for the current nuclear industry and future generations. It is expected to start around 2032. The design of the first wave of experimental devices for RJH already includes specifications for GEN2 and 3 industrial constraints. On the other hand, the field of experiments essential to GEN4 Fast Breeder Reactor remains quite open in the longer term, while no fast-spectrum irradiation facility is currently available.
The objective of this thesis is to study the feasibility of integral experiments in the JHR or another light water reactor, for validation of the reactivity loss with innovative FBR fuels.
In the first part of this thesis, fission products (FPs) that contribute to the loss of reactivity in a typical FBR will be identified and ranked by importance. The second part is the activation measurement and evaluation of the capture cross section of stable FPs in a fast spectrum. It involves the design, specification, implementation and achievement of a “stable” FBR-FP target in the ILL reactor or in the CABRI reactor fuel recovery station (potentially with thermal neutron shields). The third and final part is the design of an experiment in the JHR to generate and characterize FBR FPs. This experiment should be sufficiently representative of fuel irradiation conditions in a FBR. The goal is to access the FP inventory by underwater spectrometry in the JHR and integral reactivity weighing before/after irradiation in CABRI or another available facility.
The thesis will be carried out in a team experienced in the physics and thermal-hydraulics characterization of the JHR. The candidate will be advised by several experts based in the department. The candidate will have the opportunity to promote his/her results before the nuclear industry partners (CEA, EDF, Framatome, Orano, Technicatome etc.).
Modeling and upscaling of sodium boiling flow within a 4th generation nuclear reactor core
The stabilized boiling in sodium is a subject that has been studied for many years at CEA in order to improve the validation of scientific calculation tools such as CATHARE3. Being able to reproduce properly this phenomena is a key safety related question for liquid metal liquid 4th generation reactors. When an unprotected loss of flow (ULOF) happens in the reactor and the safety measures are not deployed, the coolant can reach saturation, which can ultimately lead to a degradation of the subassembly. In order to avoid this situation, new fuel assembly designs provide negative neutronic feedback as the void fraction is generated. To understand how this void fraction evolves in the sub-assembly (within the rod bundle or the top plenum), the code requires a state of the art sodium modeling in terms of momentum, heat and mass transfer.
To improve the qualification of the CATHARE3 code for such situations, the doctoral student will implement CFD models allowing a better understanding of the boiling mechanisms in sodium-cooled subassemblies. New CFD models, such as large interface modelling, wall boiling, heat and mass exchange at the interface will be applied, yielding detailed information on local variables. Subsequently, this detailed information will be transferred to the 1D system code during an upscaling operation. Once this information is properly gathered and transferred, new models will be developed and implemented into the system code. Finally, these new models will be confronted to experimental data in a validation exercise over the CATHARE code validation database. Ultimately, the aim is to increase the confidence in the CATHARE3 1-D simulation tool for predicting the specific physics of sodium boiling during an unprotected loss of flow transient.
The doctoral student will be based in a research unit on innovative nuclear systems at CEA/IRESNE Cadarache, in a dynamic and international environment. Travel to CEA-Saclay and EDF-Chatou is planned during the thesis, as well as participation in international conferences.
Scalability of the Network Digital Twin in Complex Communication Networks
Communication networks are experiencing an exponential growth both in terms of deployment of network infrastructures (particularly observed in the gradual and sustained evolution towards 6G networks), but also in terms of machines, covering a wide range of devices ranging from Cloud servers to lightweight embedded IoT components (e.g. System on Chip: SoC), and including mobile terminals such as smartphones.
This ecosystem also encompasses a variety of software components ranging from applications (e.g. A/V streaming) to the protocols from different communication network layers. Furthermore, such an ecosystem is intrinsically dynamic because of the following features:
- Change in network topology: due, for example, to hardware/software failures, user mobility, operator network resource management policies, etc.
- Change in the usage/consumption ratio of network resources (bandwidth, memory, CPU, battery, etc.). This is due to user needs and operator network resource management policies, etc.
To ensure effective supervision or management, whether fine-grained or with an abstract view, of communication networks, various network management services/platforms, such as SNMP, CMIP, LWM2M, CoMI, SDN, have been proposed and documented in the networking literature and standard bodies. Furthermore, the adoption of such management platforms has seen broad acceptance and utilization within the network operators, service providers, and the industry, where the said management platforms often incorporate advanced features, including automated control loops (e.g. rule-based, expert-system-based, ML-based), further enhancing their capability to optimize the performance of the network management operations.
Despite the extensive exploration and exploitation of these network management platforms, they do not guarantee an effective (re)configuration without intrinsic risks/errors, which can cause serious outage to network applications and services. This is particularly true when the objective of the network (re)configuration is to ensure real-time optimization of the network, analysis/ tests in operational mode (what- if analysis), planning updates/modernizations/extensions of the communication network, etc. For such (re)configuration objectives, a new network management paradigm has to be designed.
In the recent years, the communication network research community started exploring the adoption of the digital twin concept for the networking context (Network Digital Twin: NDT). The objective behind this adoption is to help for the management of the communication network for various purposes, including those mentioned in the previous paragraph.
The NDT is a digital twin of the real/physical communication network (Physical Twin Network: PTN), making it possible to manipulate a digital copy of the real communication network, without risk. This allow in particular for visualizing/predicting the evolution (or the behavior, the state) of the real network, if this or that network configuration is to be applied. Beyond this aspect, the NDT and the PTN network exchange information via one or more communication interfaces with the aim of maintaining synchronized states between the NDT and the PTN.
Nonetheless, setting up a network digital twin (NDT) is not a simple task. Indeed, frequent and real-time PTN-NDT synchronization poses a scalability problem when dealing with complex networks, where each network information is likely to be reported at the NDT level (e.g. a very large number of network entities, very dynamic topologies, large volume of information per node/per network link).
Various scientific contributions have attempted to address the question of the network digital twin (NDT). The state-of-the-art contributions focus on establishing scenarios, requirements, and architecture for the NDT. Nevertheless, the literature does not tackle the scalability problem of the NDT.
The objective of this PhD thesis is to address the scalability problem of network digital twins by exploring new machine learning models for network information selection and prediction.
Automatization of quantum computing kernel writing for quantum applications
The framework of Hamiltonian simulation opens up a new range of computational approaches for quantum computing. These approaches can be developed across all relevant fields of quantum computing applications, including, among others, partial differential equations (electromagnetism, fluid mechanics, etc.), quantum machine learning, finance, and various methods for solving optimization problems (both heuristic and exact).
The goal of this thesis is to identify a framework where these approaches—based on Hamiltonian simulation or block-encoding techniques—are feasible and can be written in an automated way.
This work could extend to the prototyping of a code generator, which would be tested on practical cases in collaboration with European partners (including a few months of internship within their teams).
Increasing the electrothermal robustness of new SiC devices
Silicon Carbide (SiC) is a semiconductor with superior intrinsic properties than Silicon for high temperature and high power electronics applications. SiC devices are expected to be extensively used in the electrification transition and novel energy management applications. To fully exploit the SiC superior properties, the future semiconductor devices will be used under extreme biasing and temperature conditions. These devices must operate safely at higher current densities, higher dV/dt and higher junction temperatures than Si devices does.
The objective of this thesis is to study the SiC devices fabricated at LETI under these extreme operating conditions, and to optimize their design to fully use the theoretical potential of SiC. The thesis work will include several phases that will be strongly coupled:
- Advanced electro-thermal characterisation (50%), by proposing new approaches to testing components in a box or on a suitable support, using artificial intelligence (AI) tools for data extraction and processing. The work will include adapting standard measurement methodologies to the specific switching characteristics of SiC.
- An assessment (15%) of the design and technological parameters responsible for the operating limits of the components.
- A physico-chemical characterisation component (15%) to analyse failures under these extreme conditions.
- The inclusion of predictive models (20%) for the sensitivity of architectures to extreme conditions and faults, based on modelling.
Design and optimization of color routers for image sensors
Color routers represent a promising technology that could revolutionize the field of image sensors. Composed of nanometricstructures called metasurfaces, these devices allow the modification of light propagation to improve the quantum efficiency of pixels. Thanks to recent technical advances, it is now possible to design and manufacture these structures, paving the way for more efficient image sensors.
The thesis topic focuses on the design and optimization of color routers for image sensors. Several research avenues will be explored, such as the implementation of new metasurfacegeometries (`freeform`) or innovative configurations to reduce pixel pitch (0.5µm or 0.6µm). Various optimization methods can be used, such as the adjointmethod, machine learning, or the use of auto-differentiable solvers. The designs must be resilient to the angle of light incidence and expected variations during manufacturing. After this simulation phase, the proposed structures will be manufactured, and the student will have the mission to characterize the chips and analyze the obtained results (quantum efficiency, modulation transfer function...).
This thesis will be co-supervised by STMicroelectronics and CEA LETI in Grenoble. The student will be integrated into the teams of engineer-researchers working on this project. He/she will be led to collaborate with various specialists in various fields such as lithography and optical characterization.
The student's main activities:
- Optical simulation using numerical methods (FDTD, RCWA)
- Development of optimization methodologies for metasurfacedesign (adjointmethod, topological optimization...)
- Electro-optical characterization and analysis of experimental data
CTC electrolyte pour LiS system
Lithium-Sulfur (Li-S) Batteries are among the most promising energy storage technologies for the fifth generation of batteries, often referred to as post-Li-ion. With a theoretical energy density five times higher than that of conventional Li-ion batteries and an abundant availability of sulfur, the Li-S system offers a unique potential to meet the growing demand for sustainable energy storage. However, current technology is limited by major challenges related to the dissolution of polysulfides in the electrolyte, leading to active sulfur loss, poor cycle life, and insufficient electrochemical performance. These limitations currently hinder the market deployment of this technology.
This thesis aims to explore an alternative approach based on an all-solid electrochemical sulfur conversion mechanism. To achieve this, a next-generation organic solid electrolyte developed in the laboratory will be implemented. This electrolyte features a unique lithium-ion conduction mechanism within a crystalline lattice, preventing polysulfide solubilization. The main objectives are:
1. To understand and control the ionic conduction mechanisms in these electrolytes.
2. To integrate this solid electrolyte into an innovative Li-S system.
3. To optimize the cathode structure for the solid-state mechanism and evaluate the electrochemical performance on a representative prototype scale.
The PhD candidate will use a wide range of characterization and analysis techniques to carry out this project:
• Formulation and characterization of the organic solid electrolyte: Techniques such as FT-IR and NMR to analyze chemical structures and identify the properties of synthesized materials (DSC, TGA, XRD, etc.).
• Electrochemical characterization: Analyses using electrochemical impedance spectroscopy (EIS), cyclic voltammetry (CV), and symmetric cell cycling tests to study ionic conduction properties and electrolyte stability.
• Formulation and performance study of the cathode: Formulation of carbon/sulfur composites and sulfur cathodes integrating the solid electrolyte; galvanostatic cycling tests and advanced interface analyses to understand and optimize solid-state sulfur conversion.
The research will progress in three main phases:
1. Development and characterization of the solid electrolyte: Material development, analysis of conduction mechanisms, and optimization of ionic and mechanical properties.
2. Design and optimization of the cathode structure: Improving electrolyte/cathode interfaces for solid-state sulfur conversion.
3. Electrochemical performance evaluation: Experimental validation of prototypes through in-depth tests, including cyclability and power performance.
Sub-THz programmable electromagnetic surfaces based on phase change material switches
Spatiotemporal manipulation of the near- and far-electromagnetic (EM)-field distribution and its interaction with matter in the THz spectrum (0.1-0.6 THz) is of prime importance in the development of future communication, spectroscopy, imaging, holography, and sensing systems. Reconfigurable Intelligent (Meta)Surface (RIS) is a cutting-edge hybrid analogue/digital architecture capable of shaping and controlling the THz waves at the subwavelength scale. To democratize the RIS technology, it will be crucial to reduce its energy consumption by two orders of magnitude. However, the state-of-the-art does not address the integration, scalability, wideband and high-efficiency requirements.
Based on our recent research results, the main objective of this project will be to demonstrate novel silicon-based RIS architectures s at 140 GHz and 300 GHz. The enhancement of the THz RIS performance will derive from a careful choice of the silicon technology and, from novel wideband meta-atom designs (also called unit cell or element) with integrated switches based on PCM (phase change material). The possibility of dynamically controlling the amplitude of the transmission coefficients of the meta-atoms, besides their phase, will be also investigated. Near-field illumination will be introduced to obtain an ultra-low profile. To the best of our knowledge, this constitutes a new approach for the design of high-gain antennas in the sub-THz range.