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

Area of Science:

• Materials Science
• Semiconductor Physics
• Computational Chemistry

Background:

• Multicomponent oxide systems are crucial for electronic devices.
• Optimizing these systems experimentally is challenging due to complex cation correlations.
• Indium-zinc-tin oxide (IZTO) is a ternary semiconductor with potential for high carrier mobility.

Purpose of the Study:

• To develop a machine learning (ML)-based synthetic approach for optimizing multicomponent oxide systems.
• To accurately predict the carrier mobility of indium-zinc-tin oxide (IZTO) semiconductors.
• To accelerate the discovery of high-performance IZTO thin-film transistors.

Main Methods:

• Utilized a machine learning approach, specifically support vector regression with a radial basis function kernel.
• Employed a solution-based synthetic route for facile control of IZTO composition.
• Trained the ML model with a small dataset (15-20 data points).

Main Results:

• Achieved highly accurate mobility predictions for multicomponent IZTO semiconductors.
• Predicted a field-effect mobility of 13.06 cm² V⁻¹ s⁻¹ at an In:Zn:Sn ratio of 63:27:10.
• Validated ML predictions through empirical analysis, confirming high accuracy.

Conclusions:

• The ML-based synthetic approach is a reliable and promising method for optimizing multicomponent oxide systems.
• This protocol significantly accelerates the optimization process compared to traditional experimental methods.
• High-performance IZTO thin-film transistors can be realized through this synergistic ML and synthesis approach.