NACRE: bimetallic NAnoparticles, Chemical ordeR and machinE learning

Bimetallic catalysts, for example, allow improving significantly the activity, selectivity and stability with respect to monometallic nanoparticles. A prediction of the BNP physico-chemical properties is, however, extremely difficult as these depend on a multitude of parameters: chemical composition, size, environment, chemical order, etc. BNPs in their complexity are thus ideally suited for an approach to their characterization and interpretation that uses artificial intelligence.

Based on our recent work on the spatially resolved chemical quantification within individual AgAu BNPs using different machine learning approaches, we are now able to propose the complete characterization of such particles, including not only size, shape and environment but also of the chemical order of sub-10 nm BNPs. This last property, i.e. the spatial distribution of the two metals within the BNP, has so far been extremely difficult to measure quantitatively despite its importance notably in catalysis where BNPs often undergo chemical rearrangement during redox cycles.

We propose to develop standardized procedures for the characterization of BNPs particularly relevant for heterogeneous catalysis and other applications such as hydrogen storage. We will work with pure (surfactant-free) BNPs fabricated in a gas phase cluster source that allows for the independent control of all important parameters: chemical composition, size and environment. Different transmission electron microscopy methods (imaging, element specific spectroscopy EDX and EELS, in situ and under environmental conditions, two- and three-dimensional) will then be used to determine the stable ground state structures of BNPs as well as their chemical transformations during redox cycles. Complementary machine learning techniques will be used (1) for advanced
data treatment, notably concerning spatial and spectral correlations; (2) for the optimization of atomistic potentials in accompanying simulations; and (3) the obtained reference data on selected systems will be added to databases unifying experimental and theoretical inputs for the accelerated development of new materials via other artificial intelligence approaches within the DIADEM network.

Well-established benchmark BNP systems such as PtPd or AuPd will be used initially in order to validate the experimental and data treatment procedures. In the second half of the project we will investigate the partial replacement of expensive and polluting elements such as Pt or Pd by more abundant and cheaper materials such as Fe, Cu, or Ti.

The NACRE project combines expertise in BNP fabrication and the determination of their physicochemical properties, various electron microscopy techniques for characterization, the development and interpretation of machine learning routines and simulations of structures and reactivity. It relies on complementary experimental setups within the “METSA-SET-DIA targeted project” (CEA Grenoble) and in the Lyon region and thus constitutes a clear extension of the available techniques and approaches of the DIADEM network at the nanoscale.