Advancing Semantic Representation, Alignment, and Reasoning in Multi-Agent 6G Communication Systems
Semantic communications is an emerging and transformative research area, where the focus shifts from transmitting raw data to conveying meaningful information. While initial models and design solutions have laid foundational principles, they often rest on strong assumptions regarding the extraction, representation, and interpretation of semantic content. The advent of 6G networks introduces new challenges, particularly with the growing need for multi-agent systems where multiple AI-driven agents interact seamlessly.
In this context, the challenge of semantic alignment becomes critical. Existing literature on multi-agent semantic communications frequently assumes that all agents share a common understanding and interpretation framework, a condition rarely met in practical scenarios. Misaligned representations can lead to communication inefficiencies, loss of critical information, and misinterpretations.
This PhD research aims to advance the state-of-the-art by investigating the principles of semantic representation, alignment, and reasoning in multi-AI agent environments within 6G communication networks. The study will explore how agents can dynamically align their semantic models, ensuring consistent interpretation of messages while accounting for differences in context, objectives, and prior knowledge. By leveraging techniques from artificial intelligence, such as machine learning, ontology alignment, and multi-agent reasoning, the goal is to propose novel frameworks that enhance communication efficiency and effectiveness in multi-agent settings. This work will contribute to more adaptive, intelligent, and context-aware communication systems that are key to the evolution of 6G networks.
Combined Software and Hardware Approaches for Large Scale Sparse Matrix Acceleration
Computational physics, artificial intelligence and graph analytics are important compute problems which depend on processing sparse matrices of huge dimensions. This PhD thesis focuses on the challenges related to efficiently processing such sparse matrices, by applying a systematic software are hardware approach.
Although the processing of sparse matrices has been studied from a purely software perspective for decades, in recent years many dedicated, and very specific hardware, accelerators for sparse data have been proposed. What is missing is a vision of how to properly exploit these accelerators, as well as standard hardware such as GPUs, to efficiently solve a full problem. Prior to solving a matrix problem, it is common to perform pre-processing of the matrix. This can include techniques to improve the numerical stability, to adjust the form of the matrix, and techniques to divide it into smaller sub-matrices (tiling) which can be distributed to processing cores. In the past, this pre-processing has assumed homogenous compute cores. New approaches are needed, to take advantage of heterogeneous cores which can include dedicated accelerators and GPUs. For example, it may make sense to dispatch the sparsest regions to specialized accelerators and to use GPUs for the denser regions, although this has yet to be shown. The purpose of this PhD thesis is to take a broad overview of the processing of sparse matrices and to analyze what software techniques are required to exploit existing and future accelerators. The candidate will build on an existing multi-core platform based on RISC-V cores and an open-source GPU to develop a full framework and will study which strategies are able to best exploit the available hardware.
Enhancing Communication Security Through Faster-than-Nyquist Transceiver Design
In light of the growing demand for transmission capacity in communication networks, it is essential to explore innovative techniques that enhance spectral efficiency while maintaining the reliability and security of transmission links. This project proposes a comprehensive theoretical modeling of Faster-Than-Nyquist (FTN) systems, accompanied by simulations and numerical analyses to evaluate their performance in various communication scenarios. The study will aim to identify the necessary trade-offs to maximize transmission rates while considering the constraints related to implementation complexity and transmission security, a crucial issue in an increasingly vulnerable environment to cyber threats. This work will help identify opportunities for capacity enhancement while highlighting the technological challenges and adjustments necessary for the widespread adoption of these systems for critical and secure links.
Low temperature selective epitaxial growth of SiGe(:B) for pMOS FD-SOI transistors
As silicon technologies for microelectronics continue to evolve, processes involved in device manufacturing need to be optimized. More specifically, epitaxy, a crystal growth technique, is being used to fabricate 10 nm technological node FD-SOI (Fully Depleted-Silicon On Insulator) transistors as part of CEA-Leti's NextGen project. Doped and undoped Si and SiGe semiconductor epitaxy is being developed to improve the devices' electrical performances. The thesis will focus on selective SiGe(:B) epitaxy for channels and source/drains of pMOS transistors. A comparison of SiGe and SiGe:B growth kinetics will be made between growth under H2, the commonly used carrier gas, and N2. Innovative cyclic deposition/etching (CDE) strategies will also be evaluated, with the aim of lowering process temperatures.
Advanced Surface Analysis of Ferroelectrics for memory applications
CEA-Leti has a robust track record in memory technology. This PhD project aims to contribute to the development of HfO2-based ferroelectric devices. One of the major challenges in this field is to stabilize the orthorhombic phase while reducing film thickness and thermal budget. To gain a deeper understanding of the underlying mechanisms, a novel sample preparation method will be adapted from a previous PhD project and further developed for application to ferroelectric memories. This method involves creating a beveled crater that exposes the entire thickness of the film, allowing for access by multiple characterization techniques (XPS, TOF-SIMS, SPM) on the same area. This approach will enable the correlation of compositional and chemical measurements with electrical properties. Furthermore, heating and biasing within advanced surface characterization instruments (TOF-SIMS, XPS) will provide insights into how device performance is influenced by compositional and chemical changes.
You possess strong experimental skills and a keen interest in state-of-the-art surface analysis instruments. You excel in team environments and will have the opportunity to collaborate with experts across a wide range of techniques on the nanocharacterization platform, including advanced numerical data treatment. Proficiency in Python or similar programming languages is highly desirable.
ALD materials for FE and AFE capacitances
Ultrathin HfO2-based materials are regarded as promising candidates for embedded non-volatile memory (eNVM) and logic devices. The CEA-LETI has a leadership position in the field of BEOL-FeRAM memories ultra-low consumption (<100fj/bit) at low voltage (<1V). In this context, the developments expected in this thesis aim to evaluate the impact of HfO2-based ferroelectric FE and antiferroelectric AFE layers (10 to 4 nm fabricated by Atomic Layer Deposition ALD) on the FeRAM properties and performances.
In particular, the subject will permit a deep understanding of the crystallographic phases governing the FE/AFE properties using advanced measurements techniques offered by the CEA-LETI nano-characterization platform (physico-chemical, structural and microscopy analysis, electrical measurements). Several integration solutions for ferroelectric capacitances FeCAPs using ALD FE/AFE layers will be studied including doping, interface layers, sequential fabrication w/wo air break…
Thus, the developments based on FeCAPs stack fabricated using 300mm ALD deposition tool aspires to explore the following items:
1-Doping incorporation in FE/AFE layers (La, Y…)
2-Engineering of the interface between FE/AFE layers and top/bottom electrode
3-Plasma in-situ treatment of bottom electrode surface
4-Sequential deposition with and without air break
[1] S. Martin et al. – IEDM 2024
[2] Appl. Phys. Lett. 124, 243508 (2024)
Digital reconstruction of an industrial tank for the improvement of real-time monitoring instrumentation
In the context of industrial digitalization and real-time monitoring, accessing 3D fields (velocity, viscosity, turbulence, concentration, etc.) in real time can be crucial, as local sensor networks are sometimes insufficient to provide a comprehensive view of the system's dynamics. This PhD project aims to investigate a methodology for the real-time reconstruction of fields within an instrumented industrial tank equipped with a mixing system. The proposed approach relies on finite element modeling of the relevant physics within the tank (e.g., fluid dynamics, thermal processes) and model reduction techniques such as physics-based Machine Learning (virtual sensor approach). A key focus of this thesis will also be the development of the tank instrumentation and its associated acquisition chain, both to validate the models and to generate a database for applying the proposed methodology.
Enhanced Quantum-Radiofrequency Sensor
Through the Carnot SpectroRF exploratory project, CEA Leti is involved in radio-frequency sensor systems based on atomic optical spectroscopy. The idea behind the development is that these systems offer exceptional detection performance. These include high sensitivity´ (~nV.cm-1.Hz-0.5), very wide bandwidths (MHz- THz), wavelength-independent size (~cm) and no coupling with the environment. These advantages surpass the capabilities of conventional antenna-based receivers for RF signal detection.
The aim of this thesis is to investigate a hybrid approach to the reception of radio-frequency signals, combining atomic spectroscopy measurement based on Rydberg atoms with the design of a close environment based on metal and/or charged material for shaping and local amplification of the field, whether through the use of resonant or non-resonant structures, or focusing structures.
In this work, the main scientific question is to determine the opportunities and limits of this type of approach, by analytically formulating the field limits that can be imposed on Rydberg atoms, whether in absolute value, frequency or space, for a given structure. The analytical approach will be complemented by EM simulations to design and model the structure associated with the optical atomic spectroscopy bench. Final characterization will be based on measurements in a controlled electromagnetic environment (anechoic chamber).
The results obtained will enable a model-measurement comparison to be made. Analytical modelling and the resulting theoretical limits will give rise to publications on subjects that have not yet been investigated in the state of the art. The structures developed as part of this thesis may be the subject of patents directly exploitable by CEA.
3D ultrasound imaging using orthogonal row and column addressing of the matrix array for ultrasonic NDT
This thesis is part of the activities of the Digital Instrumentation Department (DIN) in Non-Destructive Testing (NDT), and aims to design a new, fast and advanced 3D ultrasound imaging method using matrix arrays. The aim will be to produce three-dimensional ultrasound images of the internal volume of a structure that may contain defects (e.g. cracks), as realistically as possible, with improved performance in terms of data acquisition and 3D image computation time. The proposed method will be based on an approach developed in medical imaging based on Row and Column Addressed (RCA) arrays. The first part will focus on the development of new data acquisition strategies for matrix arrays and associated ultrafast 3D imaging using RCA approach in order to deal with conventional NDT inspection configurations. In the second part, developed methods will be validated on simulated data and evaluated on experimental data acquired with a conventional matrix array of 16x16 elements operating in RCA mode. Finally, a real-time proof of concept will be demonstrated by implementing the new 3D imaging methods in a laboratory acquisition system.
Embedded systems for natural acoustic signals analysis while preserving privacy
The PhD topic aims at developping Embedded systems to record and analyze natural acoustic signals. When targeting city deployement, the privacy issue is raised: how can we keep a satisfactory analysis level while never record or transmit human voices?