Scientific direction Development of key enabling technologies
Transfer of knowledge to industry

PhD : selection by topics

Tensor PCA as a Framework for One-Class Classification

Département Intelligence Ambiante et Systèmes Interactifs (LIST)

Vision & Ingénierie des Contenus (SAC)

01-09-2019

SL-DRT-19-0634

mohamed.tamaazousti@cea.fr

The objective of OCC (One-Class Classification) is to distinguish in the observed data an instance of one target object, from all other nontarget objects. The OCC is used in various sectors such as biomedical, biometry, healthcare, video surveillance and cyber-security. The key element of OCC is to find a data formalization that can allow representing in a compatible way the raw data and its intrinsic salient information. In this thesis, we plan to develop an advanced framework, based on Tensor-PCA to deal with OCC for real time embedded data analysis. This framework relies on the use of tools dedicated to tensor analysis developed by the theoretical physics community in the discrete random geometry and gravity quantification. The developed tensorial tools, such as melonic graphs and combinatorial enumeration of triangulations, are not yet applied in Tensor PCA. We will explore these promising tools in order to develop a completely breakthrough OCC framework.

Epitaxial growth and nanosecond laser annealing of GeSn/SiGeSn heterostructures

Département Technologies Silicium (LETI)

Laboratoire

01-10-2019

SL-DRT-19-0635

Pablo.ACOSTAALBA@cea.fr

Since 2015, CEA LETI has the capacity of depositing GeSn/SiGeSn heterostructures on 200 mm substrates. We are currently at the state-of-the-art in several of their application domains. In ordre to fabricate electically-pumped lasers able to operate at room temperature and performant Infra-Red photodetectors, we will explore during this PhD thesis the n-type and p-type doping of such layers, be it by ion implantation or in-situ during the epitaxial growth itself. In order to take full advantage of those doped layers, we will perform recristallisation and electrical activation anneals. With standard annealing techniques, we would be faced with the significant instability of GeSn / SiGeSn stacks (tin precipitation / surface segregation). This is why we will evaluate, during this PhD thesis, the interest of using nanosecond laser anneals and their impact on the structural and electrical properties of those stacks. Those studies, which will be conducted in our brand new SCREEN-LASSE LT3100 tool, will be among the first ever conducted on this type of semiconductors. We will notably focus on the evolution of cristalline quality, doping level, surface roughness, tin agglomeration / segregation and chemical content with the various process parameters (epitaxy and laser anneals). Such a know-how will be put to good use for the fabrication of innovative optoelectronics devices.

Study and mitigation of ionomer degradation in PEMFC electrodes by combining electrochemistry and Operando Neutron/X-Ray characterizations

Département de l'Electricité et de l'Hydrogène pour les Transports (LITEN)

Laboratoire Analyse électrochimique et Post mortem

01-10-2019

SL-DRT-19-0638

sylvie.escribano@cea.fr

Despite huge improvements in the past decade, advances in the performance and durability of PEMFC are still necessary for them to compete with existing technologies. The proton conducting ionomer, contained in the electrodes of PEMFC, is conjectured to be one of the major causes in performance losses. Few studies have been carried out on this ionomer within the electrodes because of the difficulty in characterizing its distribution and properties. Thus, the degradation mechanisms of the ionomer during operation, highly dependent on water content, are still largely hypothetical but believed to lead to modifications of its distribution, chemical and physical structure, transport properties and its contamination by cations. In this PhD, we wish to elucidate the mechanisms by coupling electrochemical and microstructural characterizations with in-situ and operando experiments using Neutrons (ILL) and X-Ray probes (ESRF, SOLEIL), furthermore including accelerated aging tests to simultaneously correlate performance degradation and local modifications of the materials. Specifically, the structure of the ionomer will be investigated by SANS and the water content in the electrodes using neutron radiography. As a result of these investigations we also aim at improving durability of PEMFC by tuning the composition of the electrodes or proposing more appropriate operating strategies as two kinds of mitigation pathways which will be validated towards selected ageing protocols. Achieving these goals is essential for the widespread adoption of PEMFC in clean transportation systems.

Atomic resolution imagery composition measure applied to superarrays

Département Technologies Silicium (LETI)

Autre laboratoire

01-10-2019

SL-DRT-19-0639

nicolas.bernier@cea.fr

For crystalline materials sensitive to electron beam radiation damage, it is necessary to quantify the chemical composition at the atomic scale while minimizing the electron dose. The usual analytical techniques in the transmission electron microscope (TEM) can not be used because of the high probe current and the relatively long acquisition time. On the other hand, atomic imaging, more precisely using the high-angle annular dark field (STEM-HAADF), is performed at a reduced dose and exhibits contrasts proportional to the atomic number of the elements. In addition, the TEMs Titan on the PFNC are equipped with an aberration corrector to acquire state-of-the-art HAADF images in terms of atomic resolution. However, for the contrast in these images to be quantitatively related to the chemical composition of the material, controlled TEM acquisition conditions and electronic scattering simulations must be developed. In parallel, another imaging technique in the TEM is attracting growing interest: ptychography, or "4D data STEM". This technique, consisting in acquiring a diffraction pattern for each position of the incident electron beam, provide the projected potential in the sample. The development of the quantitative aspect of these imaging techniques has many applications: the one targeted in this thesis is the understanding of the atomic order of GeTe / Sb2Te3 superlattices, materials considered as the most promising for phase change memories (PCRAM).

Measurement of nuclear decay data for beta decay and electron capture using metallic magnetic calorimeters

DM2I (LIST)

Laboratoire de Métrologie de l'Activité

01-09-2019

SL-DRT-19-0643

matias.rodrigues@cea.fr

In the framework of ionizing radiation metrology, one of the tasks of the Laboratoire National Henri Becquerel (LNHB), the French national laboratory for ionizing radiation metrology, is the precise determination of nuclear decay data. During this PhD thesis, cryogenic detectors will be developed for the precise measurement of the shapes of beta spectra, photon emission probabilities and capture probabilities of radionuclides decaying via electron capture. These data are required in various fields of research and application, including nuclear medicine, nuclear energy and waste management, or neutrino physics research. The PhD student will conduct experiments comprising the conception and fabrication of cryogenic detectors, their operation in a complex cryogenic setup, work with highly specific electronics, Monte Carlo simulations, and data analysis using sophisticated methods. The measured data will be compared with theoretical calculations and help to improve nuclear data tables.

Explaining predicitive-model decisions: towards automatic interpretation of tree-ensemble models

DM2I (LIST)

Laboratoire d'Analyse des Données et d'Intelligence des Systèmes

01-09-2019

SL-DRT-19-0644

pierre.blanchart@cea.fr

Until recently, the focus in predictive modelling has mainly been set on improving model prediction accuracy. Many successful models scaling to big amounts of heterogeneous data have been proposed in the literature, and widely used implementations of these models are available. Unfortunately, these models generally do not intrinsically come with an easy way to explain their predictions, and are often presented as black-box tools performing complex and non-intuitive operations on their inputs. This can be an issue in many applications where the interpretation of the model decision may have a greater added-value than the decision itself. Examples include medical diagnosis where the interpretation would consist in identfying which combination(s) of characteristics presented by an individual contributes most to the diagnosis. In this thesis, we propose to add interpretability to a specific class of machine learning models known as tree-ensemble models, without impacting the performance of the model we want to interpret. In the continuation of the work already initiated in the laboratory, the objective is to analyze the combinations of input features along with their respective numerical values, so that each instance-level decision taken by the model can be explained by a set of input features having particular numerical values. Fault detection in connected manufacturing provides an interesting application for such approaches, and data as well as the fault detection models will be provided as a starting point for this thesis work.

73 Results found (Page 10 of 13)
first   previous  8 - 9 - 10 - 11 - 12  next   last

Voir toutes nos offres