Archive ouverte HAL-LAAS
Not a member yet
12189 research outputs found
Sort by
Comparison of different feedback controllers on an airfoil benchmark
No full textInternational audienceThe present paper proposes a comparison of three well-established controllers: a robust proportionalintegral-derivative (PID) controller (Conord and Peaucelle, 2021), a model-free control (Fliess and Join, 2013, 2022) and an adaptive sliding-mode control based on the super-twisting algorithm (Shtessel et al., 2023). The benchmark considered is an airfoil section equipped with trailing edge jets, load sensors and a perturbation system. The objective is to track the lift command under external wind perturbations. The outcome of this work is the comparison of performances for three control laws that are suitable when little knowledge is known from the physics. This study quantifies performance not only in terms of load control, but also in the needed implementation effort
Evolution of the liquid/solid interface roughness in SiGe layers processed by nanosecond laser annealing
No full textInternational audiencePulsed laser annealing is a relevant alternative to conventional thermal processes for future technology nodes as it enables the application of a fast and local thermal budget. Such high-energy process can lead to the formation of a liquid phase that recrystallizes upon heat dissipation, through a high velocity liquid/solid interface moving towards the surface. Here, we report on the evolution of the liquid/solid interface roughness and its influence on the crystallinity of Si 1-x Ge x layers depending on multiple parameters (strain state, doping level, Ge content, and pulse duration). This has been conducted with a roughness quantification method based on cross-section STEM-HAADF micrographs. It has been established that the liquid/solid roughness can be decreased by: (i) a compressive strain decrease, (ii) the use of short duration laser pulses or (iii) a reduction of the initial Ge content. The Ge content and strain must correspond to suitable values for optimized MOSFET performances. Consequently, strain and pulse duration were found to be pertinent levers for liquid/solid interface roughness reduction. Increasing the amount of boron atoms in s-Si 1-x Ge x :B/Si systems is another relevant strategy, as compressive strain decrease would then be associated with a beneficial contact resistance lowering in the sourcedrain regions of p-type MOSFET devices
Secondary ion mass spectrometry (SIMS) analysis of (113) PIN & NIP diamond structures
No full textInternational audienceWith tremendous physical properties, diamond is considered to be the ultimate semiconductor for high power electronics. Indeed, diamond is awaited to endure high electric fields with low leakage current. It is then a candidate of choice for high voltage and high temperature power electronics. Even if some incursions were made on the (110) orientation, the conventional crystalline orientations used for diamond are (111) and (100). Lying in between, (113) is a stable growth orientation during chemical vapor deposition and could allow obtaining enlarged crystals. For p-type doping with boron, the LSPM lab has proven the interest of the (113) orientation by growing p-type free-standing plates with equivalent quality to (100) orientation and, very recently, enlarged (113) p+-substrates [1]. For n-type doping with phosphorus, the GEMaC lab has shown that the (113) orientation give access to lower compensation ratio than (100) with higher electron mobility [2] at temperature above 450°C. All those results pave the way for the realization (113) diamond bipolar devices. In the framework of the ANR-LAPIN113 project [3], the LAAS lab, specialized in power devices simulation and fabrication, has defined the “ideal” PIN and NIP stacks that might ensure the target breakdown voltage. LSPM and GEMaC labs were then in charge of the synthesis to get requested PIN and NIP structures. Thanks to secondary ion mass spectrometry (SIMS), we analysed the depth-distributions of dopants over each structure. The depth profiles reveal the ability of the grower labs to achieve requested PIN and NIP structures on (113) orientation. The knowledge of the exact doping profiles will allow to simulate diode characteristics and help to understand the future experimental measurements of the diodes that will be performed on the PIN and NIP structures. References1.R. Mesples-Carrère, R. Issaoui, A. Valentin, L. Banaigs, O. Brinza, F. Bénédic, J. achard. Diamond Relat. Mater. 149 (2024), 111659. https://doi.org/10.1016/j.diamond.2024.111659 2.M.-A. Pinault-Thaury, I. Stenger, R. Gillet, S. Temgoua, E. Chikoidze, Y. Dumont, F. Jomard, T. Kociniewski, J. Barjon. Carbon 175 (2021) 254. https://doi.org/10.1016/j.carbon.2021.01.011 3.For information about the ANR-LAPIN113 project see https://anr.fr/Projet-ANR-20-CE05-003
De la Simulation au Terrain : Une Approche Sans Vérité Terrain pour le Suivi 3D des Vergers
No full textInternational audienceLack of annotated data as well as model transfer challenges limit accurate structural analysis of 3D orchards in field conditions by LiDAR and machine learning techniques. This study explores a scalable framework that trains deep learning models on synthetic labelled data. It then applies them to unlabelled real point clouds acquired with an unmanned ground vehicle in an apple orchard. Integrating classification, skeletonisation, and contrastive learning the framework segments trees, generates structural maps, and detects anomalies without ground-truth annotations. These results suggest potential contributions to flexible orchard monitoring by supporting variability analysis based on descriptors derived from the structural representations of trees.La falta de datos anotados, así como los desafíos de transferencia de modelos, limitan el análisis estructural preciso de huertos 3D en condiciones de campo mediante LiDAR y técnicas de aprendizaje automático. Este estudio explora un marco escalable que entrena modelos de aprendizaje profundo en datos sintéticos etiquetados. A continuación, los aplica a nubes de puntos reales sin etiquetar adquiridas con un vehículo terrestre no tripulado en un huerto de manzanos. Integrando clasificación, esqueletización y aprendizaje contrastivo, el marco segmenta árboles, genera mapas estructurales y detecta anomalías sin anotaciones reales. Estos resultados sugieren posibles contribuciones a la supervisión flexible de huertos mediante el apoyo al análisis de variabilidad basado en descriptores derivados de las representaciones estructurales de los árboles.Le manque de données annotées ainsi que les défis liés au transfert de modèles limitent l'analyse structurelle précise des vergers 3D sur le terrain par LiDAR et les techniques d'apprentissage automatique. Cette étude explore un cadre évolutif qui entraîne des modèles d'apprentissage profond sur des données synthétiques étiquetées. Elle les applique ensuite à des nuages de points réels non étiquetés acquis à l'aide d'un véhicule terrestre sans pilote dans un verger de pommiers. Intégrant la classification, la squelettisation et l'apprentissage contrastif, le cadre segmente les arbres, génère des cartes structurelles et détecte les anomalies sans annotations de vérité au sol. Ces résultats suggèrent des contributions potentielles à la surveillance flexible des vergers en soutenant l'analyse de la variabilité basée sur des descripteurs dérivés des représentations structurelles des arbres
Thermophotovoltaic conversion: from principles to applications
No full textInvited seminar at IUSTI (Institut de Thermique, Mécanique, Matériaux), Reims, France, February 12th 2025. Speaker: Rodolphe Vaillon
Impulsive switching signals with functional inequalities: Stability analysis using hybrid systems framework
No full textInternational audienceIn this work, we introduce a class of impulsive switching signals described via functional inequalities which govern the switching among different modes with state resets. By choosing the parameters of the inequalities appropriately, we can recover several known classes of switching signals and also allow for signals that depend on time, mode or state of the system. Signals from this class can also be generated online via the use of an auxiliary timer while the dynamical system is running. Via a multiple Lyapunov functions approach, we provide sufficient conditions on the functional parameters of the switching signal which ensure that the equilibrium is globally asymptotically stable (GAS) for autonomous impulsive switched system. In case of inputs, similar methodology is used to provide sufficient conditions for input-to-state stability (ISS) and integral-input-tostate stability (iISS) uniformly over the proposed class of impulsive switching signals. As case studies, we consider switched systems which do not satisfy ISS (respectively, iISS) property for switching signals with arbitrarily large dwell-times but they are shown to be ISS (resp. iISS) for our proposed class of impulsive switchings signals described via functional inequalities.</div
Exact asymptotic characterisation of running time for approximate gradient descent on random graphs
No full textInternational audienceIn this work we study the time complexity for the search of local minima in random graphs whose vertices have i.i.d. cost values. We show that, for Erd\"os-R\'enyi graphs with connection probability given by (with and ), a family of local algorithms that approximate a gradient descent find local minima faster than the full gradient descent. Furthermore, we find a probabilistic representation for the running time of these algorithms leading to asymptotic estimates of the mean running times
Mécanismes physiques de HEMT GaN révelés par l'instabilité de la tension de seuil
No full textNational audienceDans les études de fiabilité des composants, différents paramètres électriques comme la tension de seuil (VTH) sont caractérisés afin de suivre la dégradation du composant. Cependant, pour les HEMT GaN, les mesures de VTH sont souvent instables en raison de mécanismes comme le piégeage des charges induits par l'historique des polarisations. Cette instabilité peut être considérée comme caractéristique de la structure du transistor et n'est pas liée au viellissement. Ce travail se concentre sur la compréhension de l'origine de l'instabilité de VTH des transistors GaN normally-off, en utilisant des mesures répétées de VTH. Les mesures successives de VTH, entrecoupées de polarisations de drain ou de grille respectant les limites de la datasheet, génèrent des dérives reproductibles de VTH, formant ainsi une signature unique du composant. À travers cette signature, l'instabilité initiale de VTH sera illustrée, où les principaux acteurs de cette instabilité sont les zones de field plates et la grille p-GaN. Deux références sont testées, et les signatures uniques obtenues révèlent les différences de structure entre les composants
Very High Frequency Interpolation for Direct Torque Control
No full textInternational audienceTorque control enables agile and robust robot motion, but deployment is often hindered by instability and hardware limits. Here, we present a novel solution to execute whole-body linear feedback at up to 40 kHz on open-source hardware. We use this to interpolate non-linear schemes during real-world execution, such as inverse dynamics and learned torque policies. Our results show that by stabilizing torque controllers, high-frequency linear feedback could be an effective route towards unlocking the potential of torque-controlled robotics
AFEDA : Enhancing Network Slices Acceptance Ratio with Transformer-based Feature Extraction
No full textInternational audienceThe advent of 5G network slicing technology makes it possible to divide a shared infrastructure into logical networks called network slices, each providing a customized quality of service (QoS). Consequently, the QoS satisfaction has shifted from "how to provide tailored QoS?" to "how to improve the acceptance ratio of deployed network slices?". Existing works have successfully employed Deep Reinforcement Learning (DRL) agents to address this challenge, proposing various placement strategies guided by reward functions. In this paper, we introduce AFEDA, an improved DRL agent by coupling it with a transformer-based active features extractor and demonstrate that beyond the reward function, extracting active features from observations significantly helps a DRL agent to enhance the acceptance ratio of slices. Through extensive simulations on two infrastructures, we show that AFEDA is capable of placing 14% to 31% more slices compared against a DRL agent with same configurations while using the same observations as raw features