Photoacoustic Characterization of TiO₂ Thin-Films Deposited on Silicon Substrate Using Neural Networks

Area of Science:

• Materials Science
• Nanotechnology
• Computational Physics

Background:

• Accurate characterization of thin films is crucial for advanced material applications.
• Traditional methods for determining thermoelastic and geometric properties can be complex and time-consuming.
• Titanium dioxide (TiO2) thin films possess unique properties relevant to various technological fields.

Purpose of the Study:

• To investigate the use of neural networks for determining the thermal, elastic, and geometric characteristics of TiO2 thin films.
• To compare the accuracy of individual neural networks versus a combined network for parameter prediction.
• To assess the feasibility of using machine learning for precise thin film property estimation.

Main Methods:

• A two-layer model comprising a TiO2 thin film on a silicon substrate was employed.
• Neural networks were utilized to predict thin film parameters including thickness, thermal diffusivity, and coefficient of linear expansion.
• Multiple neural networks were trained, with one analyzing all parameters simultaneously and others analyzing parameters individually.

Main Results:

• A single neural network analyzing thickness, expansion, and thermal diffusivity concurrently demonstrated the highest prediction accuracy.
• Neural networks predicting only one parameter at a time were found to be less reliable.
• The study confirmed that neural networks can accurately estimate the thermoelastic properties of thin films.

Conclusions:

• Neural networks offer a highly accurate and reliable method for characterizing thin TiO2 films.
• A multi-parameter analysis using a single neural network is superior to single-parameter prediction models.
• This approach provides a powerful tool for advancing the understanding and application of thin film materials.