AAU Student Projects - visit Aalborg University's student projects portal
A master's thesis from Aalborg University
Book cover


Metamaterial- and Machine Learning-Assisted THz Sensing

Authors

; ;

Term

4. term

Publication year

2022

Submitted on

Pages

95

Abstract

Dette speciale undersøger, om metamaterialer kan skabe feltforstærkning, der forbedrer følsomheden af terahertz tidsdomænespektroskopi (THz-TDS) ved analyse af organiske forbindelser. En tredimensionel finite-difference time-domain (FDTD) model blev udviklet til at designe og optimere metamaterialer, primært som periodiske arrays af guldantenner på siliciumwafere samt split-ring resonatorer, med variationer i antennelængder og gitterdimensioner. På baggrund af modellens dimensioner blev prøver fremstillet ved hjælp af UV direct write-litografi og magnetron-sputtering. Glukose og sukrose blev deponeret på metamaterialeprøverne og målt i et THz-TDS setup. Maskinlæringsmetoder, herunder feature-udvælgelse, feature-transformation og klassifikation, blev tilpasset og anvendt til at skelne målinger fra forskellige prøver, sammenligne forskellige støjparametre og evaluere metamaterialernes feltforstærkende egenskaber. Arbejdet samler numerisk modellering, fremstilling, spektroskopi og dataanalyse i en samlet ramme for at vurdere anvendeligheden af metamateriale- og maskinlæringsassisteret THz-sensing; detaljerede præstationsresultater er ikke angivet i denne del af rapporten.

This thesis examines whether metamaterials can provide field enhancement to improve the sensitivity of terahertz time-domain spectroscopy (THz-TDS) for organic compounds. A three-dimensional finite-difference time-domain (FDTD) model was developed to design and optimize metamaterials, primarily periodic arrays of gold antennas on silicon wafers and split-ring resonators, with variations in antenna lengths and lattice dimensions. Using dimensions obtained from the model, samples were fabricated via UV direct write lithography and magnetron sputtering. Glucose and sucrose were deposited on the metamaterial samples and measured in a THz-TDS setup. Machine learning methods—including feature selection, feature transformation, and classification—were adapted to distinguish measurements from different samples, compare noise parameterizations, and evaluate the field-enhancement properties of the metamaterials. The work integrates numerical modeling, fabrication, spectroscopy, and data analysis into a unified workflow to assess the viability of metamaterial- and machine-learning-assisted THz sensing; detailed performance outcomes are not specified in this excerpt.

[This summary has been generated with the help of AI directly from the project (PDF)]