Artificial neural networks for material parameter extraction in terahertz time-domain spectroscopy

Area of Science:

• Spectroscopy
• Materials Science
• Artificial Intelligence

Background:

• Terahertz time-domain spectroscopy (THz-TDS) enables complex refractive index determination without Kramers-Kronig relations.
• Current THz-TDS analysis involves manual pre-processing, iterative fitting, and potential accuracy loss.
• Analytical approximations offer speed but significantly compromise accuracy.

Purpose of the Study:

• To investigate machine learning for interpreting spectroscopic THz-TDS data.
• To develop a computationally efficient artificial neural network for material parameter extraction.
• To improve accuracy and ease of implementation in THz-TDS data analysis.

Main Methods:

• Training artificial neural networks on large datasets of simulated light-matter interactions.
• Utilizing machine learning for spectroscopic data interpretation.
• Developing a model for efficient material parameter extraction.

Main Results:

• A computationally efficient artificial neural network for material parameter extraction was developed.
• The trained model surpasses the accuracy of approximate analytical methods.
• The neural network approach is easier to implement and faster than iterative methods.

Conclusions:

• Machine learning, specifically neural networks, can significantly improve THz-TDS data analysis.
• Neural networks can overcome common challenges like phase unwrapping and low-frequency accuracy.
• This approach offers a more accurate, efficient, and accessible method for THz-TDS data interpretation.