Complex amorphous oxides: property prediction from high throughput DFT and AI for new material search. Issue 23 (26th October 2022)

• Record Type: Journal Article
• Authors: van Setten, Michiel J.; Dekkers, Hendrik F. W.; Pashartis, Christopher; Chasin, Adrian; Belmonte, Attilio; Delhougne, Romain; Kar, Gouri S.; Pourtois, Geoffrey
• Is Part Of: Materials advances. Volume 3:Issue 23(2022)
• Page Start: 8413
• Page End: 8427
• Publication Date: 2022-10-26
• Subjects: 620.11
• DOI: 10.1039/d2ma00759b
• Languages: English
• ISSNs: 2633-5409

Abstract

Abstract : To search for the next semiconductor breakthrough, we calculate properties for binary amorphous systems and model to predict them for complex compositions. Abstract : With decreasing dimensions and increasing complexity, semiconductor devices are getting more difficult to fabricate. In particular the allowed deposition temperature becomes lower. Amorphous materials, which do not require annealing steps, are therefore becoming more interesting. First principles modelling of amorphous materials is, however, way more complex than modelling crystalline ones. Especially to screen for new materials, a fully ab initio approach is hence too expensive. We take on this challenge by employing a combination of high throughput first principles calculations and artificial intelligence (AI). We construct 4500 atomistic models, each containing 200-atoms, to capture the properties of amorphous phases of primary, X–O, and binary, X–Y–O, metal oxides. For these models, we calculate the relevant properties for a transistor channel. Expanding this exercise to more complex metal oxides would lead to a prohibitively large number of options. We solve this problem by training support vector regression models based on the data generated for the primary and binary oxides to predict the properties of ternary and more complex oxides. By combining the trained models, we construct an objective function that, at its minimum, points to the optimal composition in terms of electronic performances and material stability. After screening a series of objective functions, we identify the Zn–Mg–Al–O metal oxide (Zn and Mg around 40–50 at%, Al below 10 at%) as being the most interesting improvement to the current industry standard a-IGZO for the development of a high charge carrier mobility layer of a thin film transistor compatible with low deposition temperature requirements. It is predicted to combine an improvement in terms of electron mobility and chemical stability with respect to a-IGZO. This method, combining first principles calculated data with AI, is however not restricted to finding new materials for the active layer in a thin film transistor. From a general perspective, this approach can be used for any alloy or compound discovery problem, in which the pivotal material properties can be calculated for, in the order of a thousand, relevant one- and two-element materials.