Related Experiment Video
Updated: Aug 15, 2026
13:44
Simulation, Fabrication and Characterization of THz Metamaterial Absorbers
Published on: December 27, 2012
Attention-driven simultaneous extraction of semiconductor multilayer parameters via terahertz spectroscopy
Optics Express
|August 14, 2026
Summary
This study introduces a prior-free transformer model for analyzing terahertz spectra. It accurately determines thin film thickness and refractive index without initial guesses, enabling rapid, non-destructive characterization.
Area of Science:
- Materials Science
- Optoelectronics
- Spectroscopy
Background:
- Accurate thin film characterization is vital for semiconductor and optoelectronic device fabrication.
- Conventional spectral fitting methods require initial estimations, risking local minima and limiting automation.
- Terahertz (THz) transmission spectroscopy offers a non-destructive method for thin film analysis.
Purpose of the Study:
- To develop a prior-free regression model for extracting thin film thickness and refractive index from THz transmission spectra.
- To leverage a transformer encoder architecture for autonomous spectral inversion without initial parameter guesses.
- To enable rapid and reliable thin film characterization for quality control in device fabrication.
Main Methods:
- A transformer encoder model was developed for direct mapping of THz spectra to thin film parameters.
- A scattering-matrix algorithm was used to generate a comprehensive training dataset.
- The model utilizes a multi-head-attention mechanism to analyze Fabry-Pérot interference fringes.
Main Results:
- The prior-free transformer model achieved accurate extraction of thickness and refractive index from THz spectra.
- The model demonstrated autonomous inversion capabilities, eliminating the need for initial estimations.
- Parameter extraction was achieved on the millisecond scale, significantly faster than conventional methods.
- Validation on a two-layer film structure confirmed the model's rapid and reliable performance.
Conclusions:
- The proposed transformer encoder model provides a rapid, accurate, and autonomous method for thin film characterization using THz spectroscopy.
- This approach overcomes limitations of traditional spectral fitting, facilitating non-destructive quality control in semiconductor and optoelectronic manufacturing.
- The methodology holds potential for advancing non-contact characterization techniques in advanced material fabrication.
