Highly Reliable Implementation of Optimized Multicomponent Oxide Systems Enabled by Machine Learning-Based Synthetic Protocol.

Abstract

Multicomponent oxide systems are one of the essential building blocks in a broad range of electronic devices. However, due to the complex physical correlation between the cation components and their relations with the system, finding an optimal combination for desired physical and/or chemical properties requires an exhaustive experimental procedure. Here, a machine learning (ML)-based synthetic approach is proposed to explore the optimal combination conditions in a ternary cationic compound indium-zinc-tin oxide (IZTO) semiconductor exhibiting high carrier mobility. In particular, by using support vector regression algorithm with radial basis function kernel, highly accurate mobility prediction can be achieved for multicomponent IZTO semiconductor with a sufficiently small number of train datasets (15-20 data points). With a synergetic combination of solution-based synthetic route for IZTO fabrication enabling a facile control of the composition ratio and tailored ML process for multicomponent system, the prediction of high-performance IZTO thin-film transistors is possible with expected field-effect mobility as high as 13.06 cm2 V-1 s-1 at the In:Zn:Sn ratio of 63:27:10. The ML prediction is successfully translated into the empirical analysis with high accuracy, validating the protocol is reliable and a promising approach to accelerate the optimization process for multicomponent oxide systems.