The von Neumann electronic architecture is currently one of the main obstacles to low-energy artificial intelligence (AI). Indeed, the separation between processing and memory cores induces major energy overconsumption. Faced with this energy crisis leading to an unsustainable situation, neuromorphic electronics based on spiking neural networks (SNN) and the computing-in-memory (CIM) paradigm, pave the way for low-energy AI through more efficient software-hardware synergy [1].
Thanks to the recent emergence of ferroelectric capacitive memories (i.e., memcapacitors), capacitive CIM via charge or voltage summation is establishing itself as an ultra-low-energy approach [2], free from the inherent limitations of resistive CIM based on memristors [3] which have been explored for the past decade (leakage currents, voltage drops, high currents). Most studies on memcapacitors are based on HfO2 and its variants [4] (Gd:HfO2, HZO, La:HZO), ferroelectric materials initially developed for FeRAM applications. Although this approach has enabled proof-of-concept demonstrations of capacitive CIM for capacitive neural networks [5], the understanding and performance of memcapacitors remain limited, with capacitive memory windows (CMW) on the order of µF/cm² and rare demonstrations of multi-level capacitance programming (a key characteristic for capacitive CIM). Moreover, a truly non-destructive read out operation compatible with how the device will be exploited in neuromorphic circuits remains to be demonstrated.
Meanwhile, the next-generation ferroelectric material AlScN is attracting significant attention [6] due to its exceptional remanent polarization (>100 µC/cm²) and its low-temperature deposition process by PVD (<300°C) that requires no crystallization annealing. These characteristics promise to significantly increase the CMW, thereby multiplying intermediate capacitance levels and facilitating their reading, while simplifying fabrication processes. Nevertheless, to date, no study has focused on the memcapacitive and computing performance of AlScN-based memcapacitors. Furthermore, efforts are required to make AlScN memories CMOS-compatible, as their current low maturity is characterized by academic fabrication processes (e.g., lift-off), round Pt electrodes, AlScN thicknesses of several tens of nanometers, and coercive voltages beyond 5 V. In that scope, advanced PVD process is considered as the most promising method enabling high-quality CMOS-compatible thin films with high control on the thickness and stoichiometry.
References
(1) Mannocci, P.; Farronato, M.; Lepri, N.; Cattaneo, L.; Glukhov, A.; Sun, Z.; Ielmini, D. In-Memory Computing with Emerging Memory Devices: Status and Outlook. APL Mach. Learn. 2023, 1 (1), 10902. https://doi.org/10.1063/5.0136403.
(2) Chen, K.-T.; Shih, L.-C.; Chen, Y.-C.; Lin, K.-H.; Chen, J.-S. Memcapacitors in Neuromorphic Architectures: The Mechanisms, Challenges, and Applications. J. Mater. Chem. C 2025, 13 (26), 13098–13120. https://doi.org/10.1039/d5tc00993f.
(3) Chen, W.; Song, L.; Wang, S.; Zhang, Z.; Wang, G.; Hu, G.; Gao, S. Essential Characteristics of Memristors for Neuromorphic Computing. Advanced Electronic Materials 2023, 9 (2), 2200833. https://doi.org/10.1002/aelm.202200833.
(4) Lee, M.; Narayan, D. M.; Kim, J.-H.; Le, D. N.; Shirodkar, S.; Park, S.; Kang, J.; Lee, S.; Ahn, Y.; Ryu, S. W.; Kim, S. J.; Kim, J. Hafnium Oxide-Based Ferroelectric Devices for In-Memory Computing: Resistive and Capacitive Approaches. ACS Appl. Electron. Mater. 2024, 6 (8), 5391–5401. https://doi.org/10.1021/acsaelm.4c01071.
(5) Zhou, Z.; Jiao, L.; Zheng, Z.; Chen, Y.; Han, K.; Kang, Y.; Zhang, D.; Wang, X.; Kong, Q.; Sun, C.; Xie, J.; Gong, X. Ferroelectric Capacitive Memories: Devices, Arrays, and Applications. Nano Convergence 2025, 12 (1), 3. https://doi.org/10.1186/s40580-024-00463-0.
(6) Zhang, Y.; Zhu, Q.; Tian, B.; Duan, C. New-Generation Ferroelectric AlScN Materials. Nano Micro Lett. 2024, 16 (1), 227. https://doi.org/10.1007/s40820-024-01441-1.
Techniques / methods
1. Thin film deposition: Development of a confocal PVD process for AlxSc1-xN utilizing dual-target magnetron co-sputtering (AlN and Sc targets). This work will leverage the state-of-the-art PLASSYS MP500S system acquired by LTM that will be installed at LMGP by late 2026. This tool will allow for precise control of the stoichiometry and achieve sub-10 nm thicknesses required for low-voltage ferroelectric switching.
2. Structural & chemical characterization: Comprehensive investigation of crystalline structure, phase purity, stoichiometry, and interface quality using a multi-technique approach:
X-ray Diffraction (XRD) & Reflectivity (XRR) to determine crystal phase (ferroelectric wurtzite vs. cubic), film density, and precise thickness;
X-ray Photoelectron Spectroscopy (XPS) to analyze chemical bonding states and establish stoichiometry;
Electron Microscopy (SEM & TEM) to assess grain morphology, cross-sectional interface and bulk quality, and defects at the nanoscale. Energy Dispersive X-ray (EDX) spectroscopy will enable quantitative stoichiometry verification (Al/Sc ratio), local elemental mapping to assess homogeneity, and line-scan profiling to detect diffusion at the AlScN/electrode interface;
Raman Spectroscopy to identify local structural distortions and stress states within the thin films.
3. Device electrical assessment: Using the Plateforme Technologique Amont (PTA) and CIME Nanotech technological platforms, the postdoc fellow and the PhD student at LTM will fabricate CMOS-compatible Metal-Ferroelectric-Metal (MFM) capacitors based on the AlScN films developed by the Master’s student. The latter will participate in the electrical characterizations including:
Ferroelectric switching with PUND (Positive-Up-Negative-Down) pulse protocol to quantify remanent polarization (2Pr) and coercive fields (Ec);
Capacitance-Voltage (C-V) measurements to evaluate the CMW;
Pulsed I-V sequences to demonstrate and optimize multi-level capacitance programming and non-destructive capacitive sensing via transient currents readout;
Cycling measurements to assess endurance and variability.
Applicant skills
The ideal candidate should possess a solid foundation in solid-state physics, microfabrication techniques and materials science, with a particular interest in ferroelectric phenomena and thin-film technologies. Practical experience or strong theoretical knowledge of sputtering deposition processes, Python coding and electrical characterization methods (C-V, I-V, DC and pulsed) is a big plus. Furthermore, the applicant must demonstrate strong analytical skills to interpret complex structural and electronic data from physical-chemical investigations, alongside the ability to work effectively within a collaborative research environment involving multiple laboratories and advanced instrumentation.
The development of new metallic materials and protective coatings increasingly relies on combinatorial synthesis approaches, in which compositionally graded thin films can be produced by magnetron co-sputtering. Such material libraries enable the rapid exploration of broad composition ranges and the identification of promising formulations. However, corrosion resistance is often assessed only at a later stage, although it can ultimately determine whether a material is suitable for realistic operating conditions. Rapid and spatially resolved corrosion screening is therefore essential to integrate environmental stability into early materials selection.
At LRS, a microdroplet electrochemical methodology has been developed for local investigation of metallic surfaces. It provides access to corrosion potential, polarization measurements and, when appropriate, electrochemical impedance spectroscopy. The approach has also been coupled with operando Raman microspectroscopy, enabling electrochemical behaviour and interfacial chemical transformations to be monitored simultaneously.
The main objective of the internship will be to finalize and validate a Raman microdroplet cell for high-temperature spectroelectrochemistry, with the aim to extend measurements from room temperature toward approximately 500 °C in selected molten-salt environments. Initial experiments will be performed on model materials and sputtered thin films to establish robust and reproducible protocols. Raman spectroscopy will provide complementary information on surface films and soluble corrosion products, while electrochemical measurements will quantify corrosion and passivation behaviour.
Once validated, the methodology will be applied to compositionally graded and multicomponent or architectured coatings produced on the CEA-INSTN PVD platform within the parallel internship. Spatially resolved measurements will allow corrosion behaviour to be mapped as a function of composition and microstructure, providing the experimental basis for establishing composition–property relationships.
A further objective will be to test first machine-learning models linking deposition parameters, composition and materials descriptors to corrosion performance. These initial models will provide a proof of concept for data-driven prediction, help identify the most relevant descriptors and indicate which additional experiments would be the most valuable. The objective of the internship will not be to develop a complete ML framework, but to establish the foundations for the subsequent development of such approaches.
The two internships will therefore form two complementary components of a single materials-discovery strategy:
PVD synthesis → composition/microstructure → spatially resolved corrosion characterization → common dataset → machine-learning models → accelerated screening.
The resulting dataset and first ML models will constitute a major outcome of the overall project. Depending on the results obtained during the two internships, the project is expected to continue as a PhD thesis, with funding to be sought through the CEA Cross-Cutting Programme on Materials Science and Engineering. The thesis would build on the materials libraries, experimental methodologies and dataset generated during the internships and develop more advanced machine-learning approaches to establish a complete processing–composition–microstructure–corrosion performance framework, ultimately enabling the prediction and optimization of protective coatings for molten-salt and other extreme environments.
Techniques / methods
Microdroplet electrochemistry
Operando Raman microspectroscopy
High-temperature spectroelectrochemistry
Polarization and impedance measurements
Molten-salt corrosion
DC magnetron and HIPIMS sputtering
Compositionally graded thin films
Spatially resolved screening
Materials databases
Machine learning and data-driven materials discovery
Applicant skills
Master 2 student or third-year engineering student with a background in physical chemistry, electrochemistry and/or materials science. Autonomy, rigor and experimental skills are expected. An interest in data analysis and machine learning would be appreciated. A good English level is required.
This project will establish a common Python-based benchmarking framework for comparing bio-inspired and conventional global optimization algorithms on the same nanoalloy. The overall aim is to determine how efficiently different bio-inspired algorithms navigate the classical or Tight-Binding potential energy surface of metallic nanoparticles containing between one and five chemical components, and how their performance compares with traditional structure-search approaches.
Techniques / methods
Global optimization
Classical molecular dynamics
Semiempirical Tight Binding
Python
Benchmarking
Applicant skills
Expected skills:
Good Python programming skills
Basic knowledge of atomistic simulations
Basic understanding of materials science, chemistry, or computational physics
Ability to work independently with scientific literature and computational tools
Beneficial skills:
Experience with ASE, LAMMPS, xTB, or similar atomistic simulation software
Knowledge of global optimization, evolutionary algorithms, or machine learning
Experience with HPC/Linux environments and parallel computing
Familiarity with structural descriptors such as SOAP or MBTR
Previous experience with metallic nanoparticles or computational materials modelling
The von Neumann electronic architecture is currently one of the main obstacles to low-energy artificial intelligence (AI). Indeed, the separation between processing and memory cores induces major energy overconsumption. Faced with this energy crisis leading to an unsustainable situation, neuromorphic electronics based on spiking neural networks (SNN) and the computing-in-memory (CIM) paradigm, pave the way for low-energy AI through more efficient software-hardware synergy [1].
Thanks to the recent emergence of ferroelectric capacitive memories (i.e., memcapacitors), capacitive CIM via charge or voltage summation is establishing itself as an ultra-low-energy approach [2], free from the inherent limitations of resistive CIM based on memristors [3] which have been explored for the past decade (leakage currents, voltage drops, high currents). Most studies on memcapacitors are based on HfO2 and its variants [4] (Gd:HfO2, HZO, La:HZO), ferroelectric materials initially developed for FeRAM applications. Although this approach has enabled proof-of-concept demonstrations of capacitive CIM for capacitive neural networks [5], the understanding and performance of memcapacitors remain limited, with capacitive memory windows (CMW) on the order of µF/cm² and rare demonstrations of multi-level capacitance programming (a key characteristic for capacitive CIM). Moreover, a truly non-destructive read out operation compatible with how the device will be exploited in neuromorphic circuits remains to be demonstrated.
Meanwhile, the next-generation ferroelectric material AlScN is attracting significant attention [6] due to its exceptional remanent polarization (>100 µC/cm²) and its low-temperature deposition process by PVD (<300°C) that requires no crystallization annealing. These characteristics promise to significantly increase the CMW, thereby multiplying intermediate capacitance levels and facilitating their reading, while simplifying fabrication processes. Nevertheless, to date, no study has focused on the memcapacitive and computing performance of AlScN-based memcapacitors. Furthermore, efforts are required to make AlScN memories CMOS-compatible, as their current low maturity is characterized by academic fabrication processes (e.g., lift-off), round Pt electrodes, AlScN thicknesses of several tens of nanometers, and coercive voltages beyond 5 V. In that scope, advanced PVD process is considered as the most promising method enabling high-quality CMOS-compatible thin films with high control on the thickness and stoichiometry.
The development of new metallic materials and protective coatings increasingly relies on combinatorial synthesis approaches, in which compositionally graded thin films can be produced by magnetron co-sputtering. Such material libraries enable the rapid exploration of broad composition ranges and the identification of promising formulations. However, corrosion resistance is often assessed only at a later stage, although it can ultimately determine whether a material is suitable for realistic operating conditions. Rapid and spatially resolved corrosion screening is therefore essential to integrate environmental stability into early materials selection.
At LRS, a microdroplet electrochemical methodology has been developed for local investigation of metallic surfaces. It provides access to corrosion potential, polarization measurements and, when appropriate, electrochemical impedance spectroscopy. The approach has also been coupled with operando Raman microspectroscopy, enabling electrochemical behaviour and interfacial chemical transformations to be monitored simultaneously.
The main objective of the internship will be to finalize and validate a Raman microdroplet cell for high-temperature spectroelectrochemistry, with the aim to extend measurements from room temperature toward approximately 500 °C in selected molten-salt environments. Initial experiments will be performed on model materials and sputtered thin films to establish robust and reproducible protocols. Raman spectroscopy will provide complementary information on surface films and soluble corrosion products, while electrochemical measurements will quantify corrosion and passivation behaviour.
Once validated, the methodology will be applied to compositionally graded and multicomponent or architectured coatings produced on the CEA-INSTN PVD platform within the parallel internship. Spatially resolved measurements will allow corrosion behaviour to be mapped as a function of composition and microstructure, providing the experimental basis for establishing composition–property relationships.
A further objective will be to test first machine-learning models linking deposition parameters, composition and materials descriptors to corrosion performance. These initial models will provide a proof of concept for data-driven prediction, help identify the most relevant descriptors and indicate which additional experiments would be the most valuable. The objective of the internship will not be to develop a complete ML framework, but to establish the foundations for the subsequent development of such approaches.
The two internships will therefore form two complementary components of a single materials-discovery strategy:
PVD synthesis → composition/microstructure → spatially resolved corrosion characterization → common dataset → machine-learning models → accelerated screening.
The resulting dataset and first ML models will constitute a major outcome of the overall project. Depending on the results obtained during the two internships, the project is expected to continue as a PhD thesis, with funding to be sought through the CEA Cross-Cutting Programme on Materials Science and Engineering. The thesis would build on the materials libraries, experimental methodologies and dataset generated during the internships and develop more advanced machine-learning approaches to establish a complete processing–composition–microstructure–corrosion performance framework, ultimately enabling the prediction and optimization of protective coatings for molten-salt and other extreme environments.
CNRS, SIMaP
Grenoble
This project will establish a common Python-based benchmarking framework for comparing bio-inspired and conventional global optimization algorithms on the same nanoalloy. The overall aim is to determine how efficiently different bio-inspired algorithms navigate the classical or Tight-Binding potential energy surface of metallic nanoparticles containing between one and five chemical components, and how their performance compares with traditional structure-search approaches.
Institut Jean Lamour
Nancy
The global steel industry stands at a critical juncture: the decarbonization of primary production routes (e.g., blast furnace/BOF) necessitates a shift toward electric arc furnace (EAF) pathways, which rely heavily on scrap metal recycling. This transition introduces unprecedented challenges in steel purity, as scrap-derived residual elements (e.g., Cu, Sn, As, Sb) accumulate in liquid steel and cannot be efficiently removed using current refining technologies. These impurities profoundly alter austenite grain boundary mobility at high temperatures, leading to:
• Uncontrolled austenite grain growth during annealing,
• Disrupted phase transformations upon cooling, and
• Deteriorated mechanical properties (e.g., toughness, ductility, weldability) in low-carbon, low-alloy steels.
Recent breakthroughs—such as those from the OPTISCRAPS project (https://www.pepr-diadem.fr/projet/optiscraps/) and international teams such as Leoben or Warwick Universities (e.g., [1, 2])—have revealed the nonlinear, often synergistic effects of residual elements on austenite grain boundary mobility. These findings were enabled by in situ confocal laser scanning microscopy (CLSM) experiments. This method uniquely allows real-time observation of austenite microstructure evolution during thermal treatments, bypassing the need for time consuming post-mortem reconstructions (which are infeasible in low-hardenable steels due to the absence of martensitic transformation).
Two critical bottlenecks currently hinder the scalability of CLSM:
• Low experimental throughput. Each experiment requires painstaking setup (e.g., oxidation control), limiting investigations to ~1 test/day—a prohibitive constraint for mapping the vast compositional spaces of residual elements and processing conditions.
• Image analysis complexity. Austenite grain boundaries are visualized via thermal grooving (surface depression at boundary-surface intersections). However, these grooves persist temporarily even after boundary migration, creating "ghost" artifacts in images. Manual post-processing to distinguish real vs. ghost boundaries is time-consuming, operator-dependent, and unscalable.
This master thesis aims to radically enhance the productivity of in situ CLSM by integrating machine learning (ML) and combinatorial metallurgy, inspired by the DIADEM program’s high-throughput philosophy. The candidate will tackle two interlinked challenges:
• The development of an automated machine learning-based image analysis tool for detecting grain boundaries, differentiating between real and ghost boundaries. Preliminary results from a student project at Mines Nancy have shown that a U-Net CNN with temporal recurrence over a stack of images can be effectively trained and yield promising results. However, significant work remains to be done on loss functions and the initial fragmentation of images in the training database. The code also needs to be trained on a large number of images and made usable for operators. This work will be based on observations conducted as part of M. Noun's PhD thesis at IJL/GPM, particularly on model grades with various residual element contents, observations to which the master student will also strongly contribute.
Innovation: Enable fully automated, high-accuracy boundary tracking, reducing analysis time from hours to minutes.
• The development of a combinatorial metallurgy approach for in situ confocal microscopy. Confocal microscopy can be performed in situ on relatively large observation fields, typically 1 mm, compatible with compositional gradients obtained in diffusion couples, similar to those required for combinatorial experiments in synchrotron XRD. These gradients are sufficiently spread out relative to the microstructure size, with compositional variations being minor at the scale of an austenitic grain. Thus, it would be possible to study all compositions between the two bounds of the diffusion couple in a single experiment, examining the effect of one or more residual elements. The objective is to explore wide compositional spaces in a limited number of experiments, thereby increasing scientific productivity. The goal of this internship will be to demonstrate the feasibility of this method and study its limits by conducting and analyzing a few exploratory experiments (samples from A. Decroocq’s Phd thesis at SIMAP/IJL).
Innovation: Reduce the number of tests needed to map multi-element compositional spaces by an order of magnitude.
[1] Duan, J., Ju, Y., Dastur, P., Slater, C., Strangwood, M., & Davis, C. (2025). Assessing the solute drag of scrap residual elements of Sn and Cu in austenite. Journal of Materials Research and Technology.
[2] Hatzenbichler, L., Gocnik, M., Haslberger, P., Galler, M., Stark, A., Glushko, O., ... & Schnitzer, R. (2025). Alteration of phase transformation behavior and prior austenite grain structure due to elevated tramp element concentrations in a hypoeutectoid steel. steel research international, 2500172.
CEA
Saint-Paul-Lez-Durance
Uranium dioxide is the primary fuel used in nuclear reactors, making its accurate modeling crucial for industrial applications.
Recent advances in machine-learning interatomic potentials (MLIPs) have revolutionized the simulation of UO2, enabling large-scale simulation with near ab-initio accuracy [1].
However, most existing MLIPs rely on short-range interactions, assuming negligible atomic interactions beyond a certain cutoff radius.
For polar materials like UO2, this approach overlooks the significant long-range dielectric interactions, which are essential for an accurate description of the material's behavior.
Recently, a new class of MLIPs have emerged that learns directly these long-range dielectric interactions through the energy [2,3].
These models not only promise improved accuracy in energy predictions but also unlock access to previously inacessible dielectric properties, such as infrared spectra and charge redistribution.
In this internship, the candidate will contribute to the first-principles modeling of UO2 by developing and training MLIPs that explicitly incorporate long-range dielectric interactions.
The project will involve :
• Model development: Training state-of-the-art MLIPs for UO2, with a focus on capturing long-range dielectric effects.
• Benchmarking: Comparing the performance of these new models against existing MLIPs for UO2 to quantify the improvements brought by the inclusion of dielectric interactions
• Property prediction: Using the trained models to compute dielectric properties of UO2 and validating the results against experimental measurements [4]
This work will provide deeper insights into the behavior of UO2, with potential implications for nuclear fuel performance and safety.
[1] E. T. Dubois et al, Phys. Rev. Mater. 8, 025402 (2024)
[2] D. S. King et al, Nat. Commun. 16 (1), 8763 (2025)
[3] P. Zhong et al, npj Comput. Mater. 11 (1), 384 (2025)
[4] J. D. Axe and G. D. Pettit, Phys. Rev. 151, 676 (1966)
Poste pourvu
IRCER, CNRS
Limoges
Scientific Description
Transition-metal carbides have attracted renewed interest owing to their exceptional properties with broad potential applications, spanning refractory and hard materials, superconductivity and HER/OER electrocatalysis [1]. These materials uniquely combine ceramic hardness, thermal and chemical stability with metallic conductivity. The origin of their multifunctional properties can be traced down to the intricate interplay between metallic and covalent bonding, which governs their structural, electronic and mechanical behaviour [2, 3].
Among different metal-to-carbon stoichiometries, metal mono-carbides (MC) and metal hemicarbides (M2C) have been extensively studied in 1970s and 1980s with particular emphasis on understanding of their crystal structure, phase diagram and optimization of synthesis routes [4, 5]. Structurally, these systems can be described as the combination of a metal sublattice forming an underlying core framework with occupation of the interstitial voids by the relatively small C atoms. Experimental reports establish that the metal sublattice in MC compounds primarily adapt FCC-derived arrangement with C atoms filling all the octahedral interstitial sites. The ordered occupation of the octahedral sites gives rise to the characteristic rosk-salt structure, in which the metal and carbon atoms form two interpenetrating FCC sublattices. In case of larger metal ions, the MC compound adapts hexagonal (AA stacking) arrangement [6]. Recently, high entropy carbides stabilizing in rock-salt phase have been reported [7]. Thus, the structural chemistry of MC systems is comparatively well established.
On the contrary, the M2C systems are characterized by HCP-derived metal sublattice framework with partial occupation of the interstitial voids by the C atoms. The low carbon content with respect to the total number of available interstitial voids naturally leads to additional degrees of freedom associated with different degree of C/vacancy ordering [4, 5]. Furthermore, variations in the size and electronic configuration of the metal atoms, along with the differences in M-C bonding interactions promotes distortions of the metal framework and heterogeneity in local-atomic arrangements. Consequently, this combination of structural features gives rise to multiple competing polymorphs with distinct crystal symmetries and local structural motifs. Despite these interesting aspects, the experimental exploration and characterization of M2C compounds have been challenging due to the requirement of high tempera- ture and precise control of processing conditions for their synthesis; thereby leading to difficulties in stabilization and characterization of individual polymorphs [8].
A natural pathway towards the exploration of the complex phase-space of M2C systems and parallely addressing the knowledge gap is the adaptation of suitable ab-initio methods (primarily based on density functional theory) such as crystal structure prediction (CSP). However, their application to M2C systems quickly becomes prohitive due to the significant increase in computational costs associated with the requirement of large supercells for an explicit representation of different disorders. In principle, machine learned interatomic potentials (MLIPs) can be utilised with CSP to alleviate these bottlenecks. But, they have seldom been used or tested for these specific complex disordered systems [9]. For instance for Mo2C, only four experimental phases have been reported, i.e. ↵-Mo2C (Pbcn), -Mo2C (P63/mmc), C6-CdI2 antisite (P ̄3m1) and the "-Fe2N (P ̄31m) [4, 5]. Additionally, few other polymorphs have been proposed theoretically [10, 11, 12]. All these reported structures predominantly comprise of HCP-derived metal frameworks [13]. But, this conventional picture has been recently challenged by our work in Ref. [14] which reports multiple non-HCP derived phases. With the structural flexibility to accommodate different disorders, the M2C systems are strong contenders for the experimental realization of high entropy carbides considering the recent success of MC systems [7].
Within this master’s thesis project, we aim to explore the phase space of 2-3 representative M2C compositions using the Minima-Hopping (MH) crystal structure prediction method [15, 16], which has the capability to efficiently explore complex chemical systems [17, 18, 19]. The MH runs will be carried out with universal MACE foundation models [20, 21], to describe the atomic interactions. The primary goal is to understand and map structural diversity across different chemistries. In order to capture different metal-carbon interactions across the transition metal series, representative M2C compositions will be selected from the 3d, 4d, and 5d transition-metal chemistries. The generated structures will be filtered, clustered, and analyzed to identify structural motifs and assess their relative thermodynamic stability. This will be subsequently followed by DFT validation primarily through open sourced Quantum-Espresso (QE) package [22]. Where feasible, the resulting DFT dataset will be used to fine-tune the foundation model to improve accuracy of the MLIP and thereby enable more efficient exploration of the selected M2C compositions.
This work will lead to a better understanding of the diversity of structural motifs across different M2C chemistries, the role of entropy in phase stabilization and free-energy landscapes, and other thermodynamic properties. In addition, it will provide further insight into the structure-property relationship, particularly with respect to electronic and mechanical properties. The success of this work, both in terms of physico-chemical understanding and computational methodology will form a strong basis for further exploration of multinary M2C systems.
Scientific Team
The proposed master project will be carried out under the supervision of Dr. Santanu Saha (Chair Junior Professor CPJ, PI) and Dr. Sylvian Cadars (CNRS CR, co-PI) based at the Institute of Research for Ceramics, IRCER CNRS UMR 7315 Laboratory and Universit �e de Limoges, 87000 Limoges. Dr. Saha and Dr. Cadars have extensive experience in application of CSP methods on investigations of different chemical systems [17, 18, 23]. They are co-PIs of the PEPR-DIADEM AMADEUS project (ANR-23-PEXD-0003) on “Machine learning-assisted automated design of high- entropy perovskite-type oxide ceramics for energy applications�. In addition, Dr. Saha is also PI of the project on “Ab-initio design of high entropy ceramics� (CPJ fund ANR-22-CPJ1-0039-01). The proposed work will be build upon the computational methodologies and tools developed within these project and tools available in the modelling group based at IRCER, Limoges.
This work will be carried out in collaboration with the team of Prof. Dr. Stefan Goedecker based at the Dept. of Physics, University of Basel, Switzerland. The minima hopping method was developed by the Goedecker group and has been successfully applied on different class of complex materials using DFT, force-field and recently MLIPs [19, 24, 25]. The Goedecker group will share their expertise in adapting the workflow associated with MH based data generation and their use in fine-tuning of MACE potentials. If required, they will assist in introduction of relevant additional features or developments within MH algorithm for accelerated searches of M2C systems.
Currently, the scientific team does not have any funding on this scientific topic. However, based on the acceptance and progress of the current project, the involved PIs are planning to submit funding proposals in future regional and national funding (ANR) calls (2027-2028).
References
[1] Y. Xiao et al. J. Mater. Chem. A, 4(27):10379–10393, 2016
[2] VI Ivashchenko et al. J. Appl. Phys., 125(23), 2019
[3] W. S. Williams. Mater. Sci. Eng. A., 105:1–10, 1988
[4] J. Dubois et al. Acta Metall., 36(8):1891–1901, 1988
[5] T. Epicier et al. Acta Metall., 36(8):1903–1921, 1988
[6] A.S. Kurlov and A.I. Gusev. Inorganic Materials, 42(2):121–127, 2006
[7] P. Sarker et al. Nat. Commun., 9(1):4980, 2018
[8] B. Deng et al. Nat. Commun., 13(1):262, 2022
[9] E. A. Bejarano et al. Phys. Chem. Chem. Phys., 27(41):22240–22250, 2025
[10] W. Sun et al. Ceram. Inter., 50(11):20796–20801, 2024
[11] J. Lei et al. J. Mater. Chem. C, 5(14):3438–3444, 2017
[12] Y. Guo et al. Struc. Sci., 78(6):848–856, 2022
[13] C. Deng et al. Mater. Res. Exp., 11(5):056512, 2024
[14] Asif Iqbal et al. Polymorphism in di-molybdenum carbides mo2c. (to be submitted), 2026
[15] S. Goedecker. J. Chem. Phys., 120(21):9911–9917, 2004
[16] M. Krummenacher et al. SoftwareX, 25:101632, 2024
[17] S. Saha et al. Phys. Rev. B, 102(2):024519, 2020
[18] S. Saha et al. Phys. Rev. Materials, 7(5):054806, 2023
[19] M. Krummenacher et al. arXiv preprint arXiv:2501.18372, 2025
[20] I. Batatia et al. Adv. Neural Inf. Process. Syst., 36:55260–55284, 2023.
[21] I. Batatia et al. arXiv preprint arXiv:2510.25380, 2025
[22] P. Giannozzi et al. J. Phys. Cond. Matt., 29(46):465901, 2017
[23] S. Cadars et al. Phys. Chem. Chem. Phys., 27(4):2158–2176, 2025
[24] M. I. Safa et al. Nanoscale Adv., 2026
[25] O. Tayfuroglu et al. Commun. Phys., 2026
CO₂ mineralization in Mg- and Ca-rich minerals is a promising route for long-term carbon storage, but its microscopic mechanisms remain poorly understood. This limits our ability to design and optimize materials for CO₂ capture. Recent work at CPCV has shown, using a fine-tuned MACE machine-learning potential, that at the forsterite–H₂O–CO₂ interface, surface protonation strongly modifies CO₂ speciation and promotes Mg²� dissolution, a key precursor to carbonate formation. This identifies surface acidity and hydroxylation as potential handles for tuning mineralization.
The proposed Master project aims to extend this approach toward an AI-guided exploration of reactive mineral interfaces. Starting from the existing forsterite model, the student will combine foundation machine-learning potentials, targeted ab initio calculations, and active learning to investigate different surface terminations, with varying hydroxylation and acidity, and, where feasible, related Mg- or Ca-bearing minerals. Reactive molecular dynamics and free-energy calculations will be used to compare key descriptors of mineralization activity: CO₂ speciation, surface protonation, cation dissolution, and formation of ion pairs. The aim is to establish structure–reactivity relationships and identify mineral structures that are particularly favorable to carbon mineralization.
The project is directly aligned with DIADEM through the use of AI and machine-learning interatomic potentials to accelerate the exploration and selection of materials for a targeted chemical functionality. The knowledge gained from the AI-driven exploration will help to identify relevant surface parameters for controlling the different stages of mineralization, providing guidelines for experimental design of mineral surfaces.
Platinum-based nanoalloys (PtNi, PtCu, CuPtNi) are reference catalysts, in particular for the oxygen reduction reaction. Their properties depend finely on particle size, composition and chemical order (solid solution, core–shell, ordered phases, surface segregation). Grown out of equilibrium by sequential deposition or co-deposition in molecular beam epitaxy (MBE), these particles follow metastable growth pathways that remain poorly understood. Synchrotron X-ray scattering (GIWAXS/GISAXS) can follow them in situ, but quantitative extraction of population descriptors from intensity profiles is an ill-posed inverse problem that classical methods treat only approximately.
The internship proposes to solve this problem through a complete simulation → scattering → machine learning approach: (i) build, by molecular dynamics with a TBSMA potential, a database of equilibrium and out-of-equilibrium CuPtNi nanoparticles, including growth simulations; (ii) compute their diffraction patterns with the Debye equation; (iii) describe realistic populations by joint distributions of size, composition and chemical order, then sample aggregated intensity profiles to form a training set; (iv) train a 1D convolutional neural network to predict the parameters of these distributions; (v) apply the model to already acquired synchrotron data (Cu on PtNi, Pt on CuNi, co-deposition) to reconstruct the evolution of the population during growth. The molecular-dynamics and Debye codes as well as the experimental data are available in the laboratory.
Lipid nanoparticles (LNPs) are the leading non-viral carriers for RNA and other polynucleotides. However, their development remains largely empirical, and their internal structure cannot be predicted from lipid composition alone. The manufacturing pathway—including mixing conditions, solvent transfer, pH and dilution history—may play a decisive role in determining particle size, homogeneity, encapsulation and stability.
This Master’s thesis will investigate how pH and dilution pathways control LNP assembly using the automated 2FAST microfluidic platform. A reference lipid composition and a model polynucleotide will be used to compare single-stage and two-stage assembly routes. A second microfluidic mixing step will enable controlled post-assembly operations such as a pH shift, dilution or ionic-strength adjustment.
Each condition will be characterized using online dynamic and static light scattering and transmission measurements, complemented by pH and encapsulation assays on collected aliquots. Experimental design, Bayesian optimization and active learning will be used to identify informative conditions while limiting material consumption. Selected samples may be analysed by SAXS to connect rapid measurements with internal nanostructure.
The expected outcomes are a reproducible experimental workflow, a process map and mechanistic rules linking assembly pathway to LNP properties.
CEA, LITEN
Grenoble
Context
Proton-exchange membrane fuel cells (PEMFCs) and proton-exchange membrane water electrolysers (PEMWEs) play an important role in the green energy transition. Their electrodes (or specifically the cathode in PEMWEs) are porous composite materials consisting of carbon particles—the electronic conductor—supporting platinum nanoparticles as the catalyst, all impregnated with a proton-conducting polymer (the ionomer) acting as the electrolyte. Currently, state-of-the-art electrodes rely on perfluorosulfonic acid (PFSA)-based ionomers, such as Nafion.
To limit the environmental impact of fluorinated materials, alternative PFAS (Per- and polyfluoroalkyl substances) -free ionomers are being developed. While membranes made from these new ionomers exhibit promising intrinsic properties, electrodes incorporating them still often show significantly lower electrochemical performance compared to Nafion-based designs. Because spatial distribution within the catalyst layer is a key parameter governing overall electrode efficiency, it is crucial to characterize the 3D organization of PFAS-free ionomers, which is expected to differ substantially from Nafion due to distinct molecular architectures and chemical interactions.
State of the Art and Our Expertise
Our research group pioneered the use of electron tomography to demonstrate that Nafion forms an ultra-thin layer—only a few nanometers thick—surrounding carbon particles [1]. More recently, we extended this analysis to larger volumes of real electrodes, revealing that the ionomer preferentially localizes in concave cavities on carbon aggregates. Building on these experimental observations, we showed that this distribution can be predicted using a numerical model proposed by Inoue et al., which accounts for capillary forces during ink drying [2,3]. Applying this model enables quantitative reconstruction of the ionomer network across larger volumes imaged by Focused Ion Beam Scanning Electron Microscopy (FIB-SEM), providing modelers with precise 3D microstructural inputs essential for predictive performance modeling. Extending this methodology to PFAS-free ionomers is the logical next frontier.
[1] M. Lopez-Haro et al., Nat Commun, 5, 5229 (2014)
[2] L. Guetaz et al., Electrochemical Society Meeting Abstracts 248, 2063-2063 (2025)
[3] G. Inoue et al., Journal of Power Sources, 439, 227060 (2019).
Internship Objectives
The candidate will apply advanced electron tomography to obtain 3D images and quantify the spatial distribution of emerging PFAS-free ionomers within the electrodes. This work sits at the cutting edge of materials characterization, involving complex experimental steps from specialized sample preparation to 3D reconstruction and image segmentation. Furthermore, a key challenge will be to optimize and automate the reconstruction workflow—which includes, in particular, an image inpainting approach—to enable high-throughput data processing.
These analyses will then be conducted on electrodes with various formulations (such as different ionomer-to-carbon ratios) and fabrication process parameters, with a view to developing an optimized PFAS-free electrode.
π-Conjugated polymers, which exhibit excellent stability and tunable optical and electrochemical properties, are vital in organic electronic devices such as organic field-effect transistors (OFETs). While these polymers are invaluable for transistors and can be found in radio-frequency identification tags (RFID) or biosensors, traditional methods rely heavily on toxic reagents and complex purifications, limiting their environmental sustainability and recyclability. Although these polymers have a number of advantages, they have one major drawback: their life cycle does not include a recycling stage after use. Most conjugated polymers have limited biodegradability and poor long-term biocompatibility mainly due to their chemical structure. To overcome the resistance of double and triple bonds to hydrolytic and enzymatic degradation, one solution relies on introducing cleavable chemical units along the polymer chain, such as imine, while retaining conjugation. Such bonds are obtained via greener synthetic routes (e.g. metal-free synthesis) and offer a promising strategy to induce biodegradability in π-conjugated polymers while maintaining their great optoelectronic properties. The rising demand for flexible, light-weight and cost-efficient devices has also highlighted the need for more accessible synthetic method for π-conjugated polymers. It is thus necessary to address these challenges to scale up polymerization processes (e.g. polycondensation) using the advantages of flow chemistry. While flow chemistry has been effective for chain polymerizations, its application in step-growth polymerizations is relatively new. Continuous flow technology offers benefits like increased efficiency, reproducibility, and enhanced safety in highly exothermic reactions due to the higher surface-to-volume ratios in tubular reactors. Currently, the field relies on cross-coupling and oxidative coupling reactions, which often involve costly and fossil metal catalysts and hazardous reagents and suffer from secondary reactions leading to batch-to-batch variations. Purification of these polymers to remove impurities increases costs and waste, conflicting with green chemistry principles. Polyazomethines are a promising alternative for organic electronics and could offer the unique ability to degrade in a programmed manner or on demand. Unpublished results from a PhD student, co-supervised by the scientific coordinator, gave an unprecedented PCE of 8% after incorporation of a biosourced polyazomethine in the active layer of an organic solar cell, demonstrating their stability in devices. While polyazomethines are already known for their interesting optoelectronic properties, most of the examples available in the literature have shown degradation studies of such π-conjugated polymers in harsh environments or in environments that reproduce biological conditions with difficulty. Assessing the biodegradability of polymers under conditions that closely mimic actual device use and disposal is crucial for realistic evaluations. Preliminary degradation results via respirometry have demonstrated 30% of degradation in waste water after 28 days showing a promising potential biodegradation of polyazomethines.
Thus, the BOATFLOW project aims to produce metal-free conjugated polymers, like polyazomethines, to transform what are currently non-degradable polymers into biodegradable materials while maintaining high perfomance in field-effect transistors. The use of continuous flow techniques will overcome the limitations of traditional synthetic methods.
The main hypothesis of this project is that the introduction of cleavable units along the polymer backbone could induce hydrolytic and/or enzymatic degradation of the material leading to a potential biodegradation. This gives rise to other hypotheses such as (i) dynamic covalent chemistry offers a more sustainable route to conjugated polymers containing cleavable linkers with no metal catalysts, less hazardous reagents and less batch-to-batch variation; (ii) continuous flow synthesis is a promising tool for rapid evoluation of metal-free synthetic approaches with potential for increased efficiency and reproducibility as well as possibilities of fast-screening for the discovery of new polymers and scaling-up semiconducting polymer production;
(iii) conjugated polymers with cleavable units can keep interesting optoelectronic properties and be integrated in organic electronic devices. In this context, the objectives of the project are to (i) produce conjugated polymers with cleavable units along their backbone, like polyazomethines, thanks to continuous flow synthesis in a reproducible manner; (ii) obtain conjugated polymers that perform as well as, if not better than, the usual semiconducting polymers used in the literature that do not contain cleavable bonds; (iii) fabricate fully biodegradable organic field-effect transistors and study their biodegradation in real environmental conditions.
Permanent magnets are key components of motors and generators, and as such they are essential for the transition to clean energy. While the demand for high performance magnets is ever increasing, today’s high-performance magnets are based on critical rare earth elements. There is thus an urgent need to develop new magnets which contain less or ideally no critical elements. The traditional trial-and-error approach to materials development, in which a sample of a given composition is treated to a given set of preparation conditions before being individually characterized, is both time and resource consuming. The so called “combinatorial� approach, in which both sample fabrication and characterization are carried out in a high-throughput fashion, can greatly accelerate while at the same time significantly reduce the cost of materials exploration. When combined with Machine Learning (ML) analysis, combinatorial exploration holds enormous potential to guide data driven design of materials. Our group has developed an entire workflow for a thin-film based combinatorial search for new sustainable hard magnetic materials [1].
High speed mapping of hysteresis loops, using an in-house developed scanning polar Magneto Optic Kerr effect (MOKE) magnetometer, is a critical capability within our high-throughput characterisation suite. MOKE magnetometry is a surface sensitive technique, and we have recently demonstrated the possibility to probe magnetic properties through the depth of a film, by cyclically removing a controlled amount of material from the top surface and remeasuring it. The subject of the M2 project is to contribute to the development of high throughput tomographic MOKE magnetometry. Initial focus will be on the study of wafer-level-etching of different families of hard magnetic materials. The student will then develop protocols for data analysis to allow the construction of 3D maps of magnetic properties.
The student will work closely with post-docs and PhD students within the research group using the thin-film combinatorial approach. Interactions with a project parter specialising in Machine Learning analysis and Micromagnetic modelling is also envisaged.
[1] W. Rigaut et al., Advanced Engineering Materials 28 (2026) e70884
https://doi.org/10.1002/adem.70884
Muconate esters are emerging bio-based monomers that combine renewable sourcing with a conjugated structure offering opportunities for post-polymerization modification and chemical recycling. Group transfer polymerization (GTP) is particularly attractive for their polymerization, providing well-defined polymers under mild, metal-free conditions. However, the very high polymerization rates involved—typically reaching complete conversion within less than 30 seconds—make conventional kinetic studies challenging.
This project will exploit the AMETHYST automated flow-chemistry platform to investigate the GTP copolymerization kinetics of a series of dialkyl muconates. Continuous-flow experiments will be used to rapidly map monomer conversion and copolymer composition as a function of reaction conditions and feed composition, allowing determination of monomer reactivity ratios and establishment of quantitative structure–reactivity relationships. The resulting kinetic understanding will be used to control copolymer composition and thereby tune the mechanical properties of the resulting materials.
In parallel, Bayesian optimization will be coupled with automated experimentation to identify polymerization conditions giving targeted molecular and compositional characteristics, building on our recent work on data-driven optimization of organocatalytic polymerization in flow. If time permits, the thermal depolymerization of selected copolymers will also be investigated to assess the relationship between copolymer structure, properties and chemical recyclability.
Ce site peut utiliser des cookies et des scripts externes. Plus d'informations