Machine Learning Analysis of Zinc Titanate Nanocomposite Using Spectroscopic Ellipsometry

By Muhammad OsamaReviewed by Susha Cheriyedath, M.Sc.Jan 25 2024

In an article published in the journal Scientific Reports, researchers from Iran and Turkey demonstrated the use of machine learning (ML) algorithms to analyze the optical properties of zinc titanate (ZnTiO₃) nanocomposite using spectroscopic ellipsometry (SE) data. They compared the performance of the artificial neural network (ANN) and support vector regression (SVR) techniques with the conventional nonlinear regression (NLR) method. Moreover, they discussed the synthesis and characterization of nanocomposites with potential applications in various fields such as optoelectronics, catalysis, and sensors.

Background

SE is a non-destructive optical technique that measures the change in the polarization state of light reflected from a surface and can be used to determine the optical constants, thickness, and composition of thin films and nanostructures. It can also be used to observe the optical properties of materials. By analyzing the ellipsometric parameters, such as the amplitude ratio (ψ) and the phase difference (Δ) of the reflected light, one can obtain information about the optical constants, dielectric function, band gap energy, and critical point energies of the sample material.

However, the analysis of SE data is often challenging and requires expert intervention, as it involves solving a nonlinear inverse problem with multiple models and parameters. ML can be applied to analyze SE data, as they can learn the complex relationship between the ellipsometry parameters and the optical properties of the sample, without requiring prior knowledge or initial guesses.

ZnTiO₃ is a composite of zinc oxide (ZnO) and titanium dioxide (TiO₂). ZnO and TiO₂ are two wide-band semiconductors with unique features and applications in various fields such as optoelectronics, catalysis, sensors, and photocatalysis. ZnTiO₃ can be synthesized in different forms such as fibers, films, ceramics, and powders by various methods. It has advantages over ZnO and TiO₂ separately, such as reducing the charge recombination, altering the band gap energy, and shifting the optical response from UV to visible regions. This composite can be used as a transparent conductor, a light-emitting diode, a solar cell, a gas sensor, a catalyst, and a photocatalyst.

About the Research

In the present paper, the authors synthesized a ZnTiO₃ nanocomposite by using a microwave-assisted method and prepared a thin film of the nanocomposite on a silicon substrate by using a spin coating technique. They then characterized the nanocomposite by using field emission scanning electron microscopy (FESEM), UV-Vis spectroscopy, Raman spectroscopy, atomic force microscopy (AFM), and attenuated total reflectance (ATR)-Fourier transforms infrared (FTIR) spectroscopy. The study measured the SE data of the nanocomposite in the photon energy range of 0.59-4.59 eV and analyzed the data by using NLR, ANN, and SVR.

The NLR data fitting method minimizes the error between the experimental and theoretical values of the ellipsometric parameters by adjusting the model parameters iteratively. The ANN ML algorithm mimics the structure and function of the human brain using a network of interconnected nodes that process and transmit information. The SVR method is another ML algorithm that constructs an optimal decision boundary or a hyperplane, which separates the data into different categories using support vectors and a kernel function.

The study employed ANN and SVR to analyze the SE data of a ZnTiO₃ nanocomposite, comparing their performance with the NLR method. The SE data were then fed into the ML algorithms, which were trained and tested to reconstruct ψ and Δ. These parameters are related to the complex reflectance ratio of the sample.

Research Findings

The outcomes showed that the ZnTiO₃ nanocomposite had a band gap energy of 3.83 eV, which is close to the reported values. The authors identified six critical point energies in the 3.39-4.44 eV range, which correspond to the interband transitions or excitonic peaks in the dielectric function and optical conductivity spectra. They observed that the refractive index and dielectric constant of the nanocomposite at zero frequency were 3.24 and 8.47, respectively, and the high-frequency dielectric constant was 19.7. Additionally, they found that the minimum level of the optical loss function was 1.15 eV, and the maximum level of optical conductivity was 3.4 eV.

The researchers compared the performance of the three methods in terms of accuracy, execution time, and error. They found that the ML algorithms were able to analyze the SE data with high accuracy and low error and were faster than the NLR method. The SVR algorithm achieved the best performance among the three methods, with the highest R2 score and the lowest mean absolute error for both ψ and Δ. It also took the least time to execute, while the NLR method took the longest time.

The ANN algorithm showed a slightly lower accuracy and higher error than the SVR algorithm but was still better than the NLR method. The study also used the SE data to calculate the spectral-dependent optical constants, such as dielectric function, refractive index, extinction coefficient, absorption coefficient, optical band gap, optical loss function, and optical conductivity of the ZnTiO₃ nanocomposite and discussed their physical meanings and implications.

Conclusion

In summary, the ML algorithms, especially the SVR algorithm, were effective and efficient tools for analyzing the SE data of the ZnTiO₃ nanocomposite. The study reported the synthesis and characterization of the nanocomposite, highlighting the promising optical, electrical, and photocatalytic properties of the ZnTiO₃ nanocomposite. The results suggested its potential application in data analysis for various device materials and photocatalytic properties.

The researchers acknowledged challenges and limitations and suggested that further studies could include the effects of different parameters and models on the ML algorithms and the use of ML algorithms for other optical characterization techniques.