BiMAn: Bimetallic magnetic nanoparticles with controlled anisotropy: from the chemical synthesis to the fabrication of optimized materials. Machine learning assistance
Coordinator: Lise-Marie LACROIX
LCPNO – Laboratoire de Physique et Chimie des Nano Objets
(UMR 5215 INSA Toulouse/Univ Toulouse 3/CNRS/INSERM)
Keywords: Nanostructured materials, Rare-earth free magnets, multiscale analysis, on-line SAXS/WAXS, orientational order parameter, Seed-mediated growth, Directed assembly, microfluidic platforms, (un)supervised automatic learning, experimental and simulated datase
Magnetic materials play a major role in the current energy and societal transitions. However, the development of sustainable yet performant materials, without critical elements such as rare-earths (RE), remains a challenge. Therefore, the aim of the BiMAn project is to develop a generic bottom- up approach for the rational design of efficient RE free magnetic materials, by the fine tuning of the intrinsic properties of individual particles, through their anisotropy, and their controlled assembly resulting in optimized collective properties.
In this perspective, the BiMAn project will combine high-throughput, on-line characterization and machine learning (ML) to :
- benefit from an ultra-efficient screening to synthesize Fe-based anisotropic NPs
- control the mesoscale ordering within assemblies to yield performant nanostructured magnetic materials.
To make the BiMAn project a viable reality, a multidisciplinary consortium is essential to gather expertise in chemistry to synthesize bimetallic nanoparticles with controlled anisotropy; in physics, for the multiscale study of structural and magnetic properties, and the design of in silico (micro)structural models; in chemical engineering, for the design of ad-hoc microfluidic and millifluidic platforms combining high temperature, reducing atmosphere and large dilution factors; and in machine learning and data science for the elaboration of algorithms for the prediction of (micro)structural properties and identification of correlations within the project.
The expected achievements of the BIMan project fulfills with the requests of the PEPR framework. The different partners (LPCNO, LGC, LPS, LIONS, ISEC) will join their complementary expertise to achieve the development of innovative tools for magnetic NPs production from data base construction and AI tools. The experimental microfluidic and millifluidic tools coupled to multiscale structural characterizations will be opened to the community. The design and fabrication of a versatile high temperature microfluidic platform working under reductive atmosphere and allowing for large dilution of seeds will open new perspectives for the study of nucleation and growth of complex NPs. The systematic study of heterogenous nucleation by seeding the reaction medium and of the role of the seed structure and mixture of ligands on the anisotropic growth will be possible. This should eventually increase the impact of the 2FAST and FastNano plateforms by extending their activities to new classes of compounds.
Machine Learning (ML) will be an essential tool to analyze the scattering profiles and predict the reaction conditions to control the size, the structure and the anisotropy of the magnetic NPs, from the seeds to the final NPs. The second challenge concerns the efficient, larger-scale synthesis of assemblies of optimized anisotropic NPs, a key process to the development of small magnetic devices. In the absence of multi-scale magnetic simulation tools, the ML-driven analysis of large datasets of SAXS images will enable us to reliably link the magnetic properties of assemblies, measured by magnetometry, with the nature of the NP building blocks (diameter; aspect ratio, polydispersity) and the structure of their mesoscale assembly in terms of orientation distribution and packing fraction, probed locally in relation to the conditions of elaboration. ML and data science algorithms will also contribute to rationalize experimental outcomes by efficiently extracting meaningful features.
Such an innovative data science strategy in an open access spirit with the creation of in silico databases and python libraries and codes for these two challenges is a strong point of the BiMan project.