Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Physics-based AI methodology for Material Parameter Extraction from Optical Data

M. Koumans, J. L. M. van Mechelen

TL;DR

Problem: extracting optical material parameters from transmission spectra is challenging due to underdetermination and complex sample geometry. Method: a hybrid physics-based AI that unfolds the transfer-matrix determination of $\epsilon(\omega)$ in a three-layer stack into a forward network, updating $p^{[\ell]}$ to converge toward $\hat p$, with $\epsilon(\omega)$ governed by a Drude–Lorentz model. Contributions: end-to-end integration of physics with deep learning, enabling autonomous retrieval within a trained parameter domain, and demonstration of improved generalization and robustness over a purely data-driven baseline on THz–IR simulated data. Significance: provides a scalable, physics-consistent, time-efficient framework for optical material parameter extraction across configurable spectral ranges and numbers of oscillators.

Abstract

We report on a novel methodology for extracting material parameters from spectroscopic optical data using a physics-based neural network. The proposed model integrates classical optimization frameworks with a multi-scale object detection framework, specifically exploring the effect of incorporating physics into the neural network. We validate and analyze its performance on simulated transmission spectra at terahertz and infrared frequencies. Compared to traditional model-based approaches, our method is designed to be autonomous, robust, and time-efficient, making it particularly relevant for industrial and societal applications.

Physics-based AI methodology for Material Parameter Extraction from Optical Data

TL;DR

Problem: extracting optical material parameters from transmission spectra is challenging due to underdetermination and complex sample geometry. Method: a hybrid physics-based AI that unfolds the transfer-matrix determination of in a three-layer stack into a forward network, updating to converge toward , with governed by a Drude–Lorentz model. Contributions: end-to-end integration of physics with deep learning, enabling autonomous retrieval within a trained parameter domain, and demonstration of improved generalization and robustness over a purely data-driven baseline on THz–IR simulated data. Significance: provides a scalable, physics-consistent, time-efficient framework for optical material parameter extraction across configurable spectral ranges and numbers of oscillators.

Abstract

We report on a novel methodology for extracting material parameters from spectroscopic optical data using a physics-based neural network. The proposed model integrates classical optimization frameworks with a multi-scale object detection framework, specifically exploring the effect of incorporating physics into the neural network. We validate and analyze its performance on simulated transmission spectra at terahertz and infrared frequencies. Compared to traditional model-based approaches, our method is designed to be autonomous, robust, and time-efficient, making it particularly relevant for industrial and societal applications.

Paper Structure

This paper contains 3 sections, 2 figures.

Table of Contents

  1. Introduction
  2. Results
  3. Conclusion

Figures (2)

  • Figure 1: Ground truth signal $y_{\text{sim}}$ and fitted signal $f(\omega, p^{[\ell]})$ using our physics-based AI approach, for $p$ containing (a) the initial values, and the updated values after (b) one unfold, (c) two unfolds, and (d) three unfolds.
  • Figure 2: Loss related to (a) training and (b) validation of a hybrid approach (orange) and purely data-driven approach (blue). (c) Performance variability evaluated by $|\mathcal{L}_{\text{val}} - \mathcal{L}_{\text{train}}|$ for each epoch.