Reliable In Memory Computing Implementation of stochastic ultra-low-power bio- inspired neural networks

The automated resolution of cognitive tasks primarily relies on learning algorithms applied to neural networks which, when executed on standard CMOS based digital architectures, lead to a power consumption several orders of magnitude larger than what the brain would require. Moreover, Conventional Edge neural network solutions can only provide output predictions, lacking the ability to accurately convey prediction uncertainty due to their deterministic parameters and neuron activations, resulting in overconfident predictions. Being able to model and compute the uncertainty of a given prediction allows the user to make better decisions (e.g. in classification or decision making processes) that can be therefore explained, which is crucial in a variety of applications, such as the safety-critical tasks (e.g., autonomous vehicles, medical diagnosis and treatment, industrial robotics, and financial systems). Probabilistic neural network is a possible solution to deal with uncertainty prediction. In addition, the power consumption can be drastically reduced by using hardware computing systems with architectures inspired by biological or physical models. They are mainly based on nanodevices mimicking the properties of neurons such as the emission of stochastic or synchronous spikes. Numerous theoretical proposals have shown that nanoscale spintronic devices (MTJ) are particularly well adapted. They can be used as stochastic components or as deterministic components.

Multi-level functionality in ferroelectric, hafnia-based thin films for edge logic and memory

The numerical transition to a more attractive, agile and sustainable economy relies on research on future digital technologies.

Thanks to its non-volatility, CMOS compatibility, scaling and 3D integration potential, emerging memory and logic technology based on ferroelectric hafnia represents a revolution in terms of possible applications. For example, with respect to Flash, resistive or phase change memories, ferroelectric memories are intrinsically low power by several orders of magnitude.

The device at the heart of the project is the FeFET-2. It consists of a ferroelectric capacitor (FeCAP) wired to the gate of a standard CMOS transistor. These devices have excellent endurance, retention and power rating together with the plasticity required for neuromorphic applications in artificial intelligence.

The thesis will use advanced characterization techniques, in particular photoemission spectroscopy and microscopy to establish the links between material properties and the electrical performance of the FeCAPs.

Operando experiments as a function of number of cycles, pulse amplitude and duration will allow exploring correlations between the kinetics of the material properties and the electrical response of the devices.

The thesis work will be carried out in close collaboration with NaMLab (Dresden) and the CEA LETI (Grenoble).

Studying inflation with quasars and galaxies in DESI

Measurements of the statistical properties of the large-scale structure (LSS) of the universe provide information on the physics that generated the primordial density fluctuations. In particular, they enable us to distinguish between different models of cosmic inflation by measuring primordial non-Gaussianity (PNG), the deviation from the initial conditions of the Gaussian random field.

Our strategy for studying LLS is to use a spectroscopic survey, DESI, whose instrument was commissioned at the end of 2019. DESI will observe 40 million galaxies and quasars. Observations take place at the 4-m Mayall telescope in Arizona. In the spring of 2021, the project began a five-year period of uninterrupted observations, covering a quarter of the sky.

For this thesis project, LSS are measured with two tracers of matter: very luminous red galaxies (LRG) and quasars, very distant and very luminous objects. These two tracers enable us to cover a wide redshift range from 0.4 to 4.0.

During the first year of his/her thesis, the student will contribute to the final analysis of the first year of DESI observations. In particular, he/she will study LSS with quasars and galaxies (LRG). His/her work will also involve assessing all possible sources of bias in the selection of quasars and LRGs that could contaminate a cosmological signal. In a second phase, the student will develop a more sophisticated analysis using three-point statistics such as the bispectrum with an extended sample to the first three years of DESI observations.

Innovative concepts for particles plasma acceleration and radiation emission in laser – overdense plasma interaction at ultra-high intensity

The present PHD work aims at exploring theoretically and numerically the generation of fast particle beams in ultra-relativistic (above 10^21 W/cm2) laser-overdense solid interaction by using properly-structured or shaped targets. Surface characteristics inducing local electromagnetic modes more intense than the laser field and where nonlinear and relativistic effects play a major role will be investigated.

On the basis of the work already carried out, the new scheme for particle acceleration will be extended in the ultra-relativistic regime of laser plasma interaction. It may lead to groundbreaking ultra-short synchronized light and electron sources with applications in probing ultrafast electronic processes. In this context, this theoretical and numerical study will allow to suggest new experimental schemes feasible on the Apollon facility and multi-PW lasers.

Hybrid solid electrolytes for "all-solid" batteries: Formulation and multi-scale characterization of ionic transport

Lithium-ion batteries, widely present in our daily lives, have revolutionized portable applications and are now used in electric vehicles. The development of new generations of batteries for future applications in transport and storing electricity from renewable sources is therefore vital to mitigating climate change. Lithium-ion technology is generally considered as the preferred solution for applications requiring high energy density, while sodium-ion technology is particularly attractive for applications requiring power.

However, the intrinsic instability of liquid electrolytes results in safety issues. Faced with the requirements concerning the environment and safety, solid-state batteries based on solid electrolytes can provide an effective solution while meeting battery energy storage needs. The barriers to overcome allowing the development of all-solid-state battery technology consist mainly in the research of new chemically stable solid electrolytes with good electrical, electrochemical and mechanical performance. For this goal, this thesis project aims to develop “polymer/polymer” and “ceramic/polymer” composite solid electrolytes with high performance and enhanced safety. Characterizations by electrochemical impedance spectroscopy (EIS) will be carried out in order to understand the cation dynamics (by Li+ or Na+) at the macroscopic scale in composite electrolytes, while the local dynamics will be probed using advanced techniques of Solid-state NMR (23Na / 7Li relaxation, 2D NMR, in-situ NMR & operando). Other characterization techniques such as X-ray and neutron diffraction, XPS, chronoamperometry, GITT ... will be implemented for a perfect understanding of the structure of electrolytes as well as aging mechanisms at the electrolyte / electrolyte and electrolyte/electrode interfaces of the all-solid battery.

Key words: composite solid electrolyte, all-solid-state battery, interfaces, multiscale characterization, dynamics of Li + and Na + ions, electrochemical performance, solid-state NMR, X-ray / neutron diffraction.

Involvement of Rad51 paralogs in Rad51 filament formation in DNA repair

Homologous recombination (HR) is a major repair mechanism for DNA double-strand breaks induced by ionizing radiation. A key step in HR is the formation of Rad51 nucleoprotein filaments on the single-stranded DNA generated from these breaks. We have shown that strict control of these filaments is essential, so that HR does not itself induce chromosomal rearrangements (eLife 2018, Cells 2021). In humans, functional homologs of control proteins are tumor suppressors. Thus, the control of HR appears to be as important as the HR mechanism itself. Our project involves the use of new molecular tools enabling a real breakthrough in the study of these controls. We will be using a functional fluorescent version of the Rad51 protein developed for the first time by our collaborators A. Taddei (Institut Curie), R. Guérois and F. Ochsenbein (I2BC, Joliot, CEA). This major advance will enable us to observe the influence of control proteins on DNA repair by microscopy in living cells. We have also developed highly accurate structural models of control protein megacomplexes in association with Rad51 filaments. This study also led to the identification of specific domains for each paralog protein, outside the structurally conserved Rad51-like core, that might define the specificity of each paralog proteins. We will use a multidisciplinary approach based on genetic, molecular biology, biochemistry, protein structure and live microscopy methods and yeast as model organism to study the consequences of the ablation of these specific domains. We will also search for proteins specifically binding these domains. Their identification would be crucial to understand the function of Rad51 paralog complexes and help to develop new therapeutic approaches.

Search for Higgs boson production with a single top and study of the CP properties of the top-Higgs coupling in the diphoton channel with the CMS experiment at the LHC.

Ten years ago, the ATLAS and CMS experiments at LHC at CERN discovered a new boson, with a dataset of proton-proton collisions of about 10 fb-1 at the centre of mass energy of 7 to 8 TeV [1,2]. Since then, the properties of this particle have been tested by both experiments and are compatible with the Higgs boson properties predicted by the Standard Model of particle physics (SM) within the uncertainties. In absence of direct probes of New Physics, increasing the accuracy of the measurements of the properties of the Higgs boson (its spin, its parity and its couplings to other particles) remains one of the most promising path to pursue.
The measurement of the ttH production allows the direct access to the top quark Yukawa coupling, fundamental parameter of the SM. ttH production is a rare process, two orders of magnitude smaller than the dominant Higgs boson production by gluon fusion. This production mode has been observed for the first time in 2018 [3, 4] separately by the CMS and ATLAS experiments, by combining several decay channels. More recently, with the full Run 2 dataset (data recorded between 2016 and 2018, with a total of 138 fb-1 at 13 TeV), this production mode was observed also using solely the diphoton decay channel, and a first measurement of its CP properties was provided again by both experiments, with the exclusion of a pure CP odd state at 3s [5, 6]. The associated production with a single top quark is about 5 times smaller than the ttH production and has never been observed. Thanks to the searches in the diphoton and multilepton channel, very loose constraints on this production modes were set for the first time recently (see Ref. [7]). This production mode is very sensitive to the H-tt coupling CP properties, since in case of CP-odd coupling, its production rate is largely increased. We propose in this thesis to study jointly the two production modes (ttH and tH) and the H-tt coupling CP properties with Run 3 data (data being recorded now and until 2026, with potentially about 250 fb-1 at 13.6 TeV) in the diphoton decay channel. If there was some CP violation in the Higgs sector, excluding small pseudo-scalar contributions will require more data. Pursuing these studies with Run 3 and beyond may allow to pinpoint small deviations not yet at reach. We propose to bring several improvements to the Run 2 analysis strategy and to use novel reconstruction and analysis techniques based on deep-learning, developped in the CEA-Saclay group by our current PhD students but not yet used in physics analyses, in order to make the most of the available dataset.
[1] ATLAS Collaboration, “Observation of a new particle in the search for the Standard Model Higgs boson with the ATLAS detector at the LHC,” Phys. Lett. B 716 (2012) 1.
[2] CMS Collaboration, “Observation of a new boson at a mass of 125 GeV with the CMS experiment at the LHC,” Phys. Lett. B 716 (2012) 30.
[3] ATLAS Collaboration, “Observation of Higgs boson production in association with a top quark pair at the LHC with the ATLAS detector”, Phys. Lett. B 784 (2018) 173.
[4] CMS Collaboration, “Observation of ttH Production”, Phys. Rev. Lett. 120 (2018) 231801.
[5] CMS Collaboration, “Measurements of ttH Production and the CP Structure of the Yukawa Inter- action between the Higgs Boson and Top Quark in the Diphoton Decay Channel”, Phys. Rev. Lett. 125, 061801.
[6] ATLAS Collaboration, “CP Properties of Higgs Boson Interactions with Top Quarks in the ttH and tH Processes Using H ? ?? with the ATLAS Detector” , Phys. Rev. Lett. 125 (2020) 061802.
[7] CMS Collaboration, “A portrait of the Higgs boson by the CMS experiment ten years after the discovery”, Nature 607 (2022) 60.

Detecting the first clusters of galaxies in the Universe in the maps of the cosmic microwave background

Galaxy clusters, located at the node of the cosmic web, are the largest gravitationally bound structures in the Universe. Their abundance and spatial distribution are very sensitive to cosmological parameters, such the matter density in the Universe. Galaxy clusters thus constitute a powerful cosmological probe. They have proven to be an efficient probe in the last years (Planck, South Pole Telescope, XXL, etc.) and they are expected to make great progress in the coming years (Euclid, Vera Rubin Observatory, Simons Observatory, CMB- S4, etc.).
The cosmological power of galaxy clusters increases with the size of the redshift range covered by the catalogue. The attached figure shows the redshift ranges covered by the catalogues of galaxy clusters extracted from experiments observing the cosmic microwave background (first light emitted in the Universe 380,000 years after the Big Bag). One can see that Planck detected the most massive clusters in the Universe in the redshift range 0<z<1. SPT and ACT are more sensitive but covered less sky: they detected tens of clusters between z=1 and z=1.5, and a few clusters between z=1.5 and z=2. The next generation of instruments (Simons Observatory starting in 2024 and CMB- S4 starting in 2032) will routinely detect clusters in 1<z<2 and will observe the first clusters formed in the Universe in 2<z<3.
Only the experiments studying the cosmic microwave background will be able to observe the hot gas in these first clusters at 2<z<3, thanks to the SZ effect, named after its discoverers Sunyaev and Zel’dovich. This effect is due to high energetic electrons of the gas, which distorts the frequency spectrum of the cosmic microwave background, and is detectable in current experiments. But the gas is not the only component emitting in galaxy clusters: galaxies inside the clusters can also emit in radio or in infrared, contaminating the SZ signal. This contamination is weak at z<1 but increases drastically with redshift. One expects that the emission from radio and infrared galaxies in clusters are of the same order of magnitude as the SZ signal in 2<z<3.
One thus needs to understand and model the emission of the gas as a function of redshift, but also the emission of radio and infrared galaxies inside the clusters to be ready to detect the first clusters in the Universe. Irfu/DPhP developed the first tools for detecting clusters of galaxies in cosmic microwave background data in the 2000s. These tools have been used successfully on Planck data and on ground-based data, such as the data from the SPT experiment. They are efficient at detecting clusters of galaxies whose emission is dominated by the gas, but their performance is unknown when the emission from radio and infrared galaxies is significant.
This thesis will first study and model the radio and infrared emission from galaxies in the clusters detected in the cosmic microwave background data (Planck, SPT and ACT) as a function of redshift.
Secondly, one will quantify the impact of these emissions on existing cluster detection tools, in the redshift range currently being probed (0<z<2) and then in the future redshift range (2<z<3).
Finally, based on our knowledge of these radio and infrared emissions from galaxies in clusters, we will develop a new cluster extraction tool for high redshift clusters (2<z<3) to maximize the detection efficiency and control selection effects, that is the number of detected clusters compared to the total number of clusters.

Modelling point defects for quantum application including electron-lattice interaction and surface effect

The rise of room-temperature applications - nanoscale magnetometry, thermometry, single photon emission, solid-state implementation of qubits - of the negatively charged nitrogen-vacancy NV- center in diamond has motivated a renewed interest in the search, with theoretical methods, of other point defects - in diamond in another material- with a desired property for quantum application, e.g. a bright photoluminescence adna long coherence time of the spin ground state.

However, the fact that the local atomic structure of the defect ground-state or of the excited states is hardly accessible with direct experimental techniques prevents a direct understanding of the thermodynamics stability of defect charge states in the bulk, and of the expected quantum property. This makes the on-demand control of the defect charge state challenging, a problem even more complex near to the surface, because band bending induces a surface modification of the charge state and surface states of ubiquitous defects may be present.

In this Ph.D. work, theoretical methods will be used to predict the defect charge states and explore the effect of the proximity of the surface on the thermodynamic stability and on the spin structure. The objective is threefold: To apply the theoretical framework developed at LSI and predict the defect charge states in bulk; To study changes in the charge state brought by the proximity of the surface; To extend the Hubbard model used to compute the excited states and to account for the electron-lattice interaction in order to compute the zero-phonon line also for the excited states that cannot be predicted by the DFT only. Materials under considerations are carbides -diamond and silicon carbide- and borides - elemental boron and boron compounds. The theoretical method will rely on the Hubbard model developed at LSI in collaboration with IMPMC, and density functional theory (DFT) calculations.

Testing the Standard Model in the Higgs-top sector in the multilepton final using the ATLAS detector at the LHC

The thesis proposes to measure in a coherent way the different rare processes of production of top quarks in association with bosons in the final state with multiple leptons at the Large Hadron Collider (LHC). The thesis will be based on the analysis of the large dataset collected and being collected by the ATLAS experiment at a record energy. The joint analysis of the ttW, ttZ, ttH and 4top processes, where one signal process becomes background when studying the other ones, will allow to get complete and unbiased measurements of the final state with multiple leptons.
These rare processes, which became accessible only recently at the LHC, are powerful probes to search for new physics beyond the Standard Model of particle physics, for which the top quark is a promising tool, in particular using effective field theory. Discovering signs of new physics that go beyond the limitations of the Standard Model is a fundamental question in particle physics today.