A fast hybrid methodology based on machine learning, quantum methods, and experimental measurements for evaluating material properties

Abstract

We present a hybrid approach based on both machine learning and targeted ab-initio calculations to determine adhesion energies between dissimilar materials. The goals of this approach are to complement experimental and/or all ab-initio computational efforts, to identify promising materials rapidly and identify in a quantitative manner the relative contributions of the different material attributes affecting adhesion. Applications of the methodology to predict bulk modulus, yield strength, adhesion and wetting properties of copper (Cu) with other materials including metals, nitrides and oxides is discussed in this paper. In the machine learning component of this methodology, the parameters that were chosen can be roughly divided into four types: atomic and crystalline parameters (which are related to specific elements such as electronegativities, electron densities in Wigner-Seitz cells); bulk material properties (e.g. melting point), mechanical properties (e.g. modulus) and those representing atomic characteristics in ab-initio formalisms (e.g. pseudopotentials). The atomic parameters are defined over one dataset to determine property correlation with published experimental data. We then develop a semi-empirical model across multiple datasets to predict adhesion in material interfaces outside the original datasets. Since adhesion is between two materials, we appropriately use parameters which indicate differences between the elements that comprise the materials. These semi-empirical predictions agree reasonably with the trend in chemical work of adhesion predicted using ab-initio techniques and are used for fast materials screening. For the screened candidates, the ab-initio modeling component provides fundamental understanding of the chemical interactions at the interface, and explains the wetting thermodynamics of thin Cu layers on various substrates. Comparison against ultra-high vacuum (UHV) experiments for well-characterized Cu/Ta and Cu/α-Al2O3 interfaces is also discussed. The new hybrid methodology can be applied to effectively and rapidly screen material options for further investigation, and provide a tractable list for experiments to accelerate optimization and integration of new materials in microelectronics. Our unique hybrid methodology lays a framework for using a combination of experimental and computational simulations to estimate properties for faster screening of materials.

Application of computational approach in plastic pyrolysis kinetic modelling: a review

“…In plastic pyrolysis, the ML and QM could be useful to predict kinetic parameters and degenerated-state or transition-state energy levels. Although this "strange couple" may sound like science fiction, this partnership is growing and should lead to important breakthroughs in the coming years [103]. Recently, a multidisciplinary team involving the Technical University of Berlin, the University of Warwick, and the University of Luxemburg have developed an ML method to predict molecular wave function and electronic properties of molecules.…”

Data-centric science for materials innovation

“…The integration of information will help to create a new paradigm for the next generation of databases-transforming them from repositories of data to "laboratories" where information and data are fused to help unravel the complexity of materials engineering problems. [52][53][54][55]…”