Sculptured Thin Films as a Platform for Machine-Learning Based Optical Sensing of Analytes
Open Access
- Author:
- Mcatee, Patrick D
- Graduate Program:
- Engineering Science and Mechanics
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- October 22, 2021
- Committee Members:
- Jeffrey Catchmark, Outside Unit & Field Member
Osama Awadelkarim, Major Field Member
Mark Horn, Major Field Member
Akhlesh Lakhtakia, Chair & Dissertation Advisor
Albert Segall, Program Head/Chair - Keywords:
- Sensing
Plasmonics
Machine Learning
Sculptured thin films - Abstract:
- A surface plasmon-polariton (SPP) wave can propagate along the planar interface of a homogeneous metal and a homogeneous dielectric material. Furthermore, the metal layer in a metal/dielectric combination can provide binding sites for analytes of interest. Binding changes the electromagnetic properties of the dielectric material within the region where the electric field of a SPP wave is maximum in magnitude. Therefore, SPP waves blue can be used for sensing. If the dielectric material is also periodically nonhomogeneous in the direction perpendicular to the metal/dielectric interface, then several SPP waves may be excited. An example of such a periodically nonhomogenous material is a type of sculptured thin film (STF) called a chiral sculptured thin film (CSTF). A CSTF comprises closely nested nano-helixes which are grown on a substrate by a process called physical vapor deposition. STFs are porous and can be infiltrated by fluids containing analytes. Many problems arise in the discipline of sensing. The obvious goal is to sense ever diminishing concentrations of the analyte. This results in ever smaller changes in reflectance of light in a prism-coupled apparatus. When analyzing reflectance data, a human will have a bias towards identifying only those features which arise due to known phenomena. It is wise to assume that there are phenomena which have not yet been identified. Also, the theory describing known phenomena may be incomplete. The solution to these problems is to use machine-learning algorithms, such as artificial neural networks (ANNs) and support vector machines (SVMs). Another problem is sensing multiple analytes simultaneously. STFs can help in this regard by utilizing vertical multiplexing of binding sites. This work addressed the use of STFs with machine learning to sense analytes. First, the correct type of STF, a CSTF, was chosen based on uniform porosity and electromagnetic periodicity. Next, using ANNs and varying incidence conditions, it was verified that the efficacy of this CSTF sensor is comparable to that of basic plasmonic sensors. This was done theoretically and experimentally. Also, a multi-analyte sensing scheme using vertical multiplexing was investigated. Finally, the capabilities of sensing an analyte in solution were compared between two types of CSTF sensors and a basic plasmonic sensor. It was found that (1) a CSTF is better suited for sensing purposes than other structurally periodic STFs because of uniform porosity; (2) p-polarized light is best for sensing using CSTFs; (3) experimentally, the statistical measures of the ANN performance from CSTF sensors are comparable to that of basic plasmonic sensors for predicting the refractive index of a homogeneous liquid; (4) calculated reflectance data from a CSTF in a prism-coupled configuration can be used to train an ANN to identify solutions with different concentrations of two analytes with high accuracy, in which the non-SPP features are important for ensuring high accuracy of classification; (5) measured reflectance data from CSTF sensors with an embedded metallic nanoparticle layer can be used to train an SVM to identify solutions with different concentrations of a single analyte with high accuracy, the CSTF sensors performing better than basic plasmonic sensors.