Materials selection and process optimisation through quantum chemistry and artificial intelligence

The NICKEFFECT project focuses on the search for new efficient electrocatalysts for hydrogen evolution reactions, catalysts for fuel-to-power conversion, and coatings for low-power consumption digital storage devices. More precisely, the goal of NICKEFFECT is to replace the widely used platinum-group metals (PGMs) with nickel (Ni), a cheaper, earth-abundant and ferromagnetic element. In order to do so while keeping high performances, innovative Nickel alloys with peculiar compositions and/or deposition techniques need to be developed.   

 

Materials selection can be greatly accelerated through quantum chemistry and artificial intelligence. New chemistries and atomic arrangements can be explored at the atomic scale using state-of-the-art techniques such as molecular dynamics and density-functional theory. These can be used to predict materials’ properties without relying on experimental parameters, making them choice tools for characterizing new Ni-based (electro)catalysts, or the magnetic properties of Ni-based alloys for digital storage. This is particularly interesting when many compounds need to be tested (e.g., with various atomic compositions) which is a very long and costly task to handle experimentally. Indeed, a virtual screening of the possible compounds can therefore be performed first, so that only the best materials need to be synthesized and tested experimentally.  

 

These so-called first-principles computations are very powerful, but they can sometimes be very demanding in terms of computational time due to the complexity of the systems to simulate. In other cases, such as finding the best experimental setup to grow a material (e.g., electrodeposition temperature or current density), they are simply not adequate for the problem at hand. It may also happen that experimental data is already available. In these cases, artificial intelligence can be used as a guide towards the best material candidates or the best experimental setup at a fraction of the computational cost of other methods.  

 

At Matgenix, we are able to leverage these techniques in the most effective way. For the NICKEFFECT project, we help our experimental collaborators by guiding them in the search for the best Ni-based (electro)catalysts and digital storage devices. Combining the experimental data with machine learning, we can suggest the next experimental setup to be tested targeting the maximization of the catalytic activity and the durability of the films. Once these experiments are performed, the new results can be added to the data used to train the machine-learning model, therefore increasing its predictive power. This active learning approach can drastically decrease the time needed to find the best Ni-based electrocatalyst.