Surface-Enhanced Raman Scattering‒Based Plasmonic Sensors

Starting from the basics of plasmonics and Raman scattering, the present chapter will introduce surface-enhanced Raman spectroscopy (SERS), where the role of electromagnetic enhancement due to plasmonic coupling will be discussed. Further, the role of plasmonic nanostructure optimization in maximizing the SERS signal will be presented through an example of plasmonic nanosculptured thin films (nSTFs). Further, various modalities of SERS sensing with examples will be provided. At last, an insight into the introduction and utility of machine learning (ML) techniques for Chapter 8 Surface-Enhanced Raman Scattering‒Based Plasmonic Sensors Plasmonics-Based Optical Sensors and Detectors Edited by Banshi D. Gupta, Anuj K. Sharma, and Jin Li Copyright © 2023 Jenny Stanford Publishing Pte. Ltd. ISBN 000-000-0000-00-0 (Hardcover), 000-000-0000-00-0 (eBook) www.jennystanford.com Sachin Kumar Srivastavaa and Ibrahim Abdulhalimb aDepartment of Physics, Indian Institute of Technology Roorkee, Roorkee-247667, Uttarakhand, India bElectrooptics and Photonics Engineering Unit, School of Electrical and Computer Engineering and Ilse Katz Institute of Nanoscale Science and Technology, Ben Gurion University of the Negev, Beer Sheva84105, Israel sachin.srivasta[email protected].ac.in 240 Surface-Enhanced Raman Scattering‒Based Plasmonic Sensors SERS analysis and sensing of species in complex analyte matrix will be presented. The discussions will be substantiated with examples from relevant literature. 8.1 Surface-Enhanced Raman Spectroscopy 8.1.1 Plasmons Plasmons are the quanta of collective longitudinal oscillations of free electrons in (generally) bulk metals. Let us consider a noble metal as a sea of free electrons. When an external stimulation leads to local displacement in the position of a collection of electrons, the mutual electrostatic repulsion among the displaced electrons sets up a restoring force, which leads to collective longitudinal oscillations. This type of wave can be thought analogous to the waves emanating by dropping a pebble in calm water. The waves in calm water traveling away from the center of excitation are longitudinal in nature while the ones on the surface are complex in character. They possess both longitudinal and transverse characters. Such kinds of waves on surfaces can be termed surface plasmon waves and are partially transverse in nature, and can be excited using electromagnetic waves, such as light. Figure 8.1 (a) Schematic of LSPR excitation. (b) Electromagnetic field enhancement in a 40 nm Ag sphere, simulated using COMSOL Multiphysics v5.6. When the dimensions of noble metals are reduced to a few or tens of nanometers, even if there are plasmonic excitations, the structure cannot withstand a wave. Such a kind of plasmonic excitation is termed localized surface plasmon resonance (LSPR) (Fig. 8.1a) [1]. In such a case, the electromagnetic field gets confined in structures 241 Surface-Enhanced Raman Spectroscopy much smaller than the wavelength of incident light (Fig. 8.1b). When a molecule is brought closer to such a plasmonic nanostructure, it experiences a high electromagnetic field [2], which, in turn, leads to enhanced spectroscopic signals from the molecule. 8.1.2 Enhancement of Raman Signal by Plasmons Molecules can be subdivided, in general, into two categories; the first possesses permanent dielectric moment, while the second does not possess any dipole moment. When the molecules of the first category are subject to any incident electromagnetic field, there is an interaction of dipoles of the molecule, which leads to a class of optical phenomena such as fluorescence, etc. In the case of molecules with no dielectric moment, when subjected to the electromagnetic field, such molecules get polarized, thereby creating induced dipoles. The induced dipole moment of such molecules depends on the polarizability of the molecule. Such a class of molecules possesses scattering processes, which are predominantly elastic in nature. Elastic scattering means that these molecules predominantly scatter the radiation of a frequency similar to that of the incident one. This is termed as Rayleigh scattering. A very small fraction of the incident photons on such molecules, say one in 10 million gets scattered inelastically. To say, the scattered radiation has a frequency either larger or smaller than that of the incident radiation. This phenomenon is called Raman scattering. The scattered radiation with a smaller frequency is called Stokes, while the one with a larger frequency is called anti-Stokes radiation. The small shift in the frequency of the incident radiation is a characteristic of the molecular bonds and the Raman bands of every molecule being unique, this is called molecular fingerprinting. As stated earlier, the probability of Raman scattering is very low, 1: 10,000,000; hence the spectroscopic throughput of the phenomenon is very low. Various techniques for the enhancement of Raman signal from the molecules of interest have been invented. One of the most popular techniques is surface-enhanced Raman scattering/spectroscopy (SERS). SERS is the enhancement of Raman signals in the vicinity of plasmonic nanostructured surfaces [3–5]. The induced dipole moment,  p of the molecule, is given as:  pE =a(8.1) 242 Surface-Enhanced Raman Scattering‒Based Plasmonic Sensors where a is the polarizability of the molecule and  E is the applied electric field. Out of the two factors on which  p depends, the contribution of a, known as chemical contribution to SERS [6], is of the order of 102. The contribution of SERS enhancement due to enhancement in the electric field  E is termed electromagnetic contribution and is of the order of magnitude of 106–1010 [7, 8]. Such enhancement in Raman intensities enables to achieve single molecule limit of detection [9]. 8.1.3 Estimation of Electromagnetic Enhancement of Stokes Lines Since most of the molecules remain in the ground state prior to excitation, the probability of Stokes transitions is about 104 times higher than that for anti-Stokes. Let us try to estimate the extent of SERS enhancement due to plasmonic nanoparticles. For this purpose, let us assume a small molecule is kept at a distance d from the surface of a plasmonic nanosphere of radius R, as shown in Fig. 8.2. Figure 8.2 Schematic of a molecule placed at a small distance near a plasmonic nanosphere. Let us denote the dielectric function of the nanoparticle as em and that of the medium surrounding it as e0. Let E0 be the magnitude of an incident in the electric field and Eplasmon that of the plasmons. If the molecule plus the nanosphere is illuminated with a laser of electric field E0, the total electric field experienced by the molecules will be EMolecule = E0 + EPlasmon. (8.2) 243 Surface-Enhanced Raman Spectroscopy EPlasmon due to the metallic nanosphere is given by ER RdE m m Plasmon=- ++ ee ee 0 0 3 30 2 () .(8.3) The field enhancement factor, A can be defined as A=Fieldatthe position of molecule Incident field. (8.4) Hence, if ν is the frequency of the incident laser, AE E R Rd molm m () () () () () () n n n en e en e =~ - ++ 0 0 0 3 3 2. (8.5) Therefore, the total electromagnetic enhancement factor for Stokes SERS power, GAA em SLS () () () nnn=22 ~- + - + È Î Í˘ ˚ ˙+ Ê Ë Áˆ ¯ ˜ en e en e en e en e mL mL mS mS R Rd () () () () 0 0 0 0 212 22 , (8.6) where νS is the Stokes frequency. From Eq. (8.6), it can be concluded that the SERS enhancement in power is roughly of the order of E4. However, this enhancement is inversely proportional to ~d12, which means that it is a very small-range phenomenon and is effective only when the molecule is very close to the plasmonic surface. That is why it is termed SERS. Further, if the size of the analyte is big, say ~100 nm, the plasmonic nanostructures may not be able to enhance the Raman signal from the whole molecule. Bringing a molecule near a nanostructured plasmonic surface leads to about 106–108 enhancement in the SERS power. Since the Raman signal of each molecule is a unique signature and it gets highly enhanced in SERS, SERS can be used for highly sensitive and selective detection of various molecules of interest. A number of bio and chemical sensor studies based on SERS, ranging from clinical diagnostics to water sensing, etc. have been reported in the literature [10–15]. It can also be observed from Eq. (8.6) that for optimum enhancement, the plasmon resonance must coincide with the laser wavelength and the Stokes line [16]. The nanospheres of about 60–70 nm size, which can be prepared with good control on size distribution possess plasmon resonances around 500–600 nm wavelength, a 244 Surface-Enhanced Raman Scattering‒Based Plasmonic Sensors 532 nm laser can be a good choice for SERS spectroscopy. However, the fluorescence signals possessed by certain moieties dominate over the Raman signal at such wavelengths. Therefore, it is preferred to work on near infra-red (NIR) wavelengths to avoid any fluorescence signals. The cost paid for that is low throughput power. It is also difficult to make nanospheres plasmonically resonant at NIR wavelengths. Moreover, the dipole approximation-based analysis is only approximate and does not accurately predict the actual SERS enhancement. In general, for 785 nm excitation, a nanorodlike structure is employed, which works as a nanoantenna and is one of the high-performance tools for SERS. This nanorod-shaped nanoantenna can be fabricated using lithographic techniques, but it is very costly to fabricate them. A relatively cost-effective technique for the fabrication of high-performance nanorod-like structures called nSTFs is the glancing angle deposition (GLAD) technique under physical vapor deposition (PVD) [17]. Figure 8.3 SEM image of various nSTFs (Adapted with permission from Ref. [18]). 245 Surface-Enhanced Raman Spectroscopy 8.1.4 Nanosculptured Thin Plasmonic Films and GLAD Plasmonic nSTFs are basically porous films of plasmonic metals, which possess generally nanorod-like structures. A scanning electron microscope (SEM) image of various nSTFs has been shown in Fig. 8.3. These films are usually grown by the GLAD technique, as mentioned in the earlier paragraph. Pre-templating gives a rather defined topography of the grown nanosculptures, as evident from the SEM images. A schematic of the GLAD setup inside a vacuum chamber has been shown in Fig. 8.4a. The substrate to be coated with the nSTF is kept at glancing angles to the collimated metal plume directed toward it. It has provisions for the control of tilt, online rotation and heating of the substrate, an influx of desired gases during deposition, etc., which enable one to fabricate desired, rod-like, zigzag, and helical porous nanosculptured films. Figure 8.4b illustrates the basic mechanism of the formation of Figure 8.4 (a) Schematic of GLAD setup and (b) mechanism of nSTF formation (Adapted with permission from Ref. [19]). columnar films. During the initial stage of deposition, certain nucleation sites of the incoming vapor flux are grown on the substrate. These nucleation sites form a shadowing region for the later incoming vapor flux, thereby making it a columnar, porous film [20]. Pre-templated substrates grow columnar structures in a regular 246 Surface-Enhanced Raman Scattering‒Based Plasmonic Sensors fashion, as the templates play the same role as that of the nucleation sites, which were randomly formed. At later stages, the high-speed adatoms get deposited near the tip and diffuse toward the bottom to provide relatively smoothers and homogeneous films. The tilt of the nanorods is a function of the tilt angle. However, vertical nanorods can be achieved by adequate rotation of the substrate under suitable conditions [21]. By controlling the temperature of the substrate, deposition rate, and gas influx rate, one may control the porosity of deposited films. 8.1.4.1 Performance optimization Optimization with respect to nSTF height Though Raman scattering is a wavelength-independent phenomenon, SERS enhancement depends on the resonant excitation of the plasmonic nanostructures along with the Stokes frequencies. Hence, the choice of both the plasmonic films and the laser wavelengths becomes crucial. For a wavelength of 785 nm of the excitation laser, hence, it is important to optimize the nSTFs of silver such that they are resonant with the laser wavelength. For this purpose, the nSTFs with rod lengths ranging from 100 to 400 nm having fixed (optimum) porosity (to be discussed later in this section) were fabricated on Si using GLAD. The incubation of these nSTFs in 1% (W/W) 4-Aminothiophenol (4-ATP) in ethanol solution for about an hour resulted in the formation of a self-assembled monolayer (SAM) of 4-ATP over the Ag nSTF. The SAM forms due to the bonding of the thiol (-SH) group in 4-ATP with Ag, while the NH2 group of 4-ATP offers avenues for further bonding over it. After the SAM formation, the nSTFs are taken out and cleaned in running ethanol to remove any unbound 4-ATP molecules from the nSTF surface. The SERS spectra from the 4-ATP functionalized Ag nSTFs were recorded and studied using a custom-made fiber optic Raman spectrometer system. The schematic of the fiber optic Raman system has been shown in Fig. 8.5. Light of wavelength of 785 nm launched into the excitation fiber of a fiber optic Raman probe, equipped with suitable laser line and long-pass filters, lenses, and dichroic and folding mirrors, was focused on the SERS chip and the Raman signal scattered off, and it was collected from the same spot by using the same lens. The SERS signal accessed from the sample was fed to 247 the fiber optic Raman spectrometer via collection fiber. The Raman spectrometer was further interfaced with a computer to assess and analyze the SERS spectra. SERS spectra from 4-ATP functionalized Ag nSTFs of varying heights using the aforementioned setup were studied by Shalabney and Abdulhalim [22] and have been plotted in Fig. 8.6. It can be observed that for a closed film of about 50 nm thickness (with 0% porosity), no SERS enhancement was observed. However, for porous nSTFs, SERS peaks characteristic of 4-ATP were observed. This is due to about one order of magnitude larger electromagnetic field enhancements in localized surface plasmons, as compared to extended or propagating surface plasmons. Figure 8.5 Schematic of a custom-made fiber optic Raman spectroscopy setup (Adapted with permission from Ref. [23]). It can be observed that with an increase in the height of the nSTF from 100 to 400 nm, the SERS enhancement increased, became maximum around 300 nm height, and then decreased again at 400 nm height. It was concluded that the optimum height of nSTF for maximum SERS enhancement is about 300 nm. It can be noted that Surface-Enhanced Raman Spectroscopy 254 Surface-Enhanced Raman Scattering‒Based Plasmonic Sensors the behavior of the first type of substrate. The concentrations of the reagents were chosen such that all the substrates possessed a comparable nanoparticle coverage density of ~30 particles/µm2. 8.2.1.2 Anion sensing experiment Aqueous solutions of ClO4–, CN–, SCN–, and SO42– ranging from micromolar to nanomolar concentrations were prepared and the SERS spectra for each sample solution over the substrate of type I were recorded. The SERS spectra for the abovementioned four anions with varying concentrations have been shown in Fig. 8.11. It can be observed that all the different anions possess qualitative signatures of Raman bands specific to them. Furthermore, with an increase in concentration, in each spectral window, a rise in SERS intensity is observed, which leads to quantitative prediction as well. The limits of detection (LoDs) of 1 ppb, 8 ppb, 7 ppb, and 4 ppb were achieved for ClO4–, CN–, SCN–, and SO42–, respectively. Figure 8.11 SERS spectra of aqueous solutions of ClO4–, CN–, SCN–, and SO42– using the first type substrate (Adapted with permission from Ref. [28]). A comparative study of the performance of the three substrates for a fixed concentration of SCN– was made to assess the optimal sensing capabilities of the three sensors. In Fig. 8.12a, the SERS spectra assessed from 5 ppb SCN– in an aqueous medium over I, II, and III types of sensor substrates have been plotted. It can be observed that among all the types of sensors, type I has the best performance. The substrate with Ag– nanoparticles has the 255 worst performance toward anions sensing among the three types of sensors. In type II, though positively charged BPEI maintains a similar charged environment as that of type I sensor, the distance between the anions and the plasmonic surface becomes relatively larger, thereby leading to a relatively weaker SERS signal. In Fig. 8.12b, the absorption isotherm shows a similar behavior, which is explained as previously. Figure 8.12 (a) SERS spectra for all the three types (I, II, and III) of substrates at 5 ppb concentration of SCN–. (b) Adsorption isotherm of SCN– for all the three substrates (Adapted with permission from Ref. [28]). 8.2.2 Indirect Detection: Escherichia coli Detection Using nSTFs Escherichia coli (E. coli) is a rod-shaped, gram-negative, facultative anaerobic bacterium, which is considered to be an indicator of fecal contamination (fecal coliforms). Some strains of E. coli are highly pathogenic and lead to food or waterborne gastrointestinal diseases. Consumption of E. coli contaminated water and/or food may lead to hemolytic-uremic syndrome (HUS), especially in elderly children. Prolonged HUS may cause hemolysis and kidney failure, which may result in seizures, strokes, or even death [29]. According to the World Health Organization (WHO) survey reports, approximately two billion people annually suffer from gastrointestinal diseases caused by E. coli [30]. SERS-Based Sensing Mechanisms 256 Surface-Enhanced Raman Scattering‒Based Plasmonic Sensors 8.2.2.1 Sensor chip development A stepwise illustration of sensor chip development has been presented in Fig. 8.13. Firstly, an Ag nSTF chip was incubated in 4-ATP solution in ethanol (1% w/w) for 1 hour. This led to the formation of a self-assembled monolayer (SAM) of 4-ATP over the Ag nSTF surface. The nSTF chip was rigorously washed with ethanol and then a copious amount of Milli-Q water to remove any remnants and dried using a blow of nitrogen (N2) gas after the SAM formation. The chip was then incubated in an aqueous solution of glutaraldehyde (5% v/v), a cross-linker, for 1 hour to substantiate receptor binding. After binding of cross-linker on the sensor chip, it was again washed with Milli-Q water to remove any unbound molecules and blow-dried with N2 gas. The chip was further incubated in T4 bacteriophage solution for 4 hours to form an E. coli specific receptor layer. After this, the chip was incubated in an aqueous solution of BSA, an antifouling agent, of 1 mg/ml concentration in 50 mM PBS buffer for 1 hour to enable the prevention of putative non-specific bindings on the sensor surface. This step leads to increased specificity, thereby improving the overall performance of the sensor while reducing the foul signal from other bacteria present in the sample. After BSA immobilization, the sensor chip was taken out, washed rigorously in a copious amount of Milli-Q water and PBS, and blow-dried with N2. The chip was stored in a refrigerator at 4 oC and was taken out only to perform SEM, atomic force microscope (AFM), and SERS studies. Figure 8.13 Illustration of sensor chip development for detection of E. coli (Adapted with permission from Ref. [23]). 257 Figure 8.14 AFM and SEM images of the nSTF sensor: (a), (b) before and (c), (d) after functionalization, respectively; (e), (f) SEM images of E. coli B attached to the sensor surface (Adapted with permission from Ref. [23]). The AFM and SEM images of the E. coli sensor chip before and after the nSTF functionalization have been shown in Fig. 8.14a–d, respectively. In both the AFM (Fig. 8.14c) and the SEM (Fig. 8.14d) images, the leg-like parts of the attached T4-bacteriophages are visible. Figure 8.14e,f shows the AFM and SEM images of a number of E. coli (rod-like microstructures) captured at the sensor surface, respectively. The sample solutions of different concentrations ranging from 1.5×102 to 1.0×105 cfu/mL were prepared in PBS buffer from the stock solutions of E. coli B, E. coli μX, P. aeruginosa, CV026, and P. dentrificans. Bacterial strains other than E. coli B were chosen for negative control experiments to be detailed later in Fig. 8.15. The sensor chips interacted with the sample solutions for 10 minutes. SERS spectra from the chip before and after the interaction were assessed for each sample solution to ensure intensity referencing. After the assessment of SERS spectra, the sensor chip was washed with 20 mM aqueous solution of NaOH to detach bacteria from the surface, thereby regenerating the sensor surface. The sensor surface was then washed twice with PBS to remove the remnants of the NaOH solution, blow-dried with N2 gas, and then used for other solutions. The SERS spectra for different concentrations of E. coli B, E. coli μX, P. dentrificans, P. aeruginosa, and CV026 bacteria were recorded. SERS-Based Sensing Mechanisms 258 Surface-Enhanced Raman Scattering‒Based Plasmonic Sensors The peak SERS intensity of the most prominent peak at 1077 cm–1 Raman shift was considered for quantitative analysis. The variation of peak SERS intensities @1077 cm–1 with bacterial concentrations has been plotted in Fig. 8.15. The X-axis representing concentration has been chosen logarithmic as it spans a large dynamic range. The background SERS signal from the bare sensor chip was subtracted from the SERS signal of samples over the chip to correct the baseline for all the measurements. The Y-axis, therefore, represents the A/D counts for the chip with sample minus that from the bare sensor chip (ISample – ISensor) @1077 cm–1. Figure 8.15 SERS-based E. coli sensor response curves for various bacteria: Negative control (Adapted with permission from Ref. [23]). The processed signal was termed differential Raman enhancement and has been used so throughout the chapter. The symbols represent the differential SERS Intensities at varying concentrations of studied bacteria, while the lines through symbols are the best curve fits. With an increase in the concentration of bacterial concentrations except for the two kinds of E. coli, the differential SERS intensity does not show any change. For both the strains of E. coli, E. coli B, and E. coli µX, differential SERS intensity first increases at small concentrations and then becomes nearly constant, as all the receptor sites get consumed in binding. This kind of response confirms that the sensor possesses high specificity for E. coli. Approximately 10 µL volumes 259 of the bacterial samples were used for SERS sensor studies. It can be estimated that taking 10 µL from a well-mixed sample of 1.5×102 cfu/mL will nearly have only one bacterial cell; thereby reaching the ultimate limit of detection of a single bacterium. 8.2.2.2 Indirect detection: DNA hybridization-based sensors Gold nanostructures connected to single-stranded DNA (ss-DNA) can be useful as micro-SERS probes for reaching ultra-low limits of detection. Here, we briefly discuss a couple of schemes for the detection of heavy metal ion Hg2+, which leads to the hybridization of ss-DNA into double-stranded DNA (ds-DNA), thereby changing the SERS signal. Two such schemes have been illustrated in Fig. 8.16a,b. These schemes have been taken from Refs. [31] and [32] and a detailed discussion of these could be found there, respectively. As shown in Fig. 8.16a, the micro-SERS probe contains two interconnected Thymine (T) rich ss-DNA strands, one of which acts as a probe and the other, complementary. One end of the probe DNA has the thiol (-SH) group to facilitate binding with Au nanoshell, while the other end has tetramethylrhodamine (TAMRA) to provide an enhanced SERS signal. When Hg2+ is introduced to this kind of micro-SERS sensor, the Hg2+ leads to hybridization of the T-bases, thereby bringing TAMRA, close to the Au-nanoshell surface. This leads to the enhancement of the SERS signal, as SERS is a highly distance-dependent phenomenon. Figure 8.16b illustrates the hybridization of two gold nanostar immobilized ss-DNA strands when Hg2+ is introduced. The hybridization leads to enhanced SERS signal, thereby providing nondirect information about the presence of Hg2+. These sensing mechanisms provide highly sensitive detection of analytes of interest. However, when an analyte is present in a complex matrix of molecules/species having almost similar or overlapping Raman bands, it becomes extremely difficult to accurately predict the presence of a particular analyte. As an analogy, one may find it as difficult as finding a needle in hay. To overcome this difficulty, in recent times, artificial neural networks (ANNs) equipped with machine learning/deep learning (ML/DL) algorithms have been employed. In the next sub-section, starting with the preface of the methods and various steps involved in these algorithms, the results of a recent article as a case study using one kind of DL algorithm for SERS-based detection will be discussed. SERS-Based Sensing Mechanisms 260 Surface-Enhanced Raman Scattering‒Based Plasmonic Sensors Figure 8.16 DNA hybridization-based SERS sensors for detection of heavy metal ion (Hg2+) using (a) gold nanoshells (Adapted with permission from Ref. [31]), (b) gold nanostars (Adapted with permission from Ref. [32]). 8.2.3 Machine Learning‒Enabled SERS Sensors Raman spectroscopy is a non-invasive method to characterize molecules by retrieving vibrational information from different chemical bonds and using this information to identify components in the given samples. However, it comes with certain challenges associated with it, due to the complex nature of the technique, such as it is still a very hard problem to find different hidden molecular components in Raman spectra of mixtures. To overcome this, researchers are always searching for new advanced data processing 261 techniques to extract certain meaningful features and analyze complex Raman spectra. For problems with relatively simple Raman signals of pure components, a plethora of techniques are readily available such as Euclidean distance, linear regression, and multivariate data analysis algorithms to recognize and categorize simpler molecular vibrational spectra of pure components with relatively high accuracy. But in practical applications, usually, multiple components are present in Raman spectra of complex samples. And for complex Raman spectra of mixtures of entities of similar structure/functional groups, it still is a challenging task to qualitatively analyze and identify components in Raman spectra of mixtures with high accuracy and sensitivity. For quantitative analysis of Raman spectra of mixtures, many methods have been suggested and used in the past, such as statistics, chemometrics, and various search-based methods. Most of the proposed methods are based on the characterization of peaks, distance, and similarities between spectra and dataset-searching algorithms [33–36]. Moreover, these methods are required to have a spectral data preprocessing step, which can cause some information loss at the cost of increasing variance and thus can introduce errors. Also, one has to choose the searching algorithm and similarity criteria for these methods; hence these methods are not versatile and thus are fit for only certain types of Raman spectra analysis. Due to the advancement in computer hardware technology, there has been a renewed interest in artificial intelligence (AI), where the algorithms are developed over ANNs and trained via ML/DL to extract accurate information from complex or big datasets. In SERSbased sensors, it could be helpful in the assessment of a large dataset of vibrational spectra of complex chemical mixtures. Moreover, it has a large number of applications in various fields such as computer vision, natural language processing, chemistry, biology, sensors, and even in spectroscopy. It is similar to the training of a brain via repeated exposure of the sensory organs to various features of different systems, as illustrated in Fig. 8.17. Once trained, the brain gets acquainted with the systems and can easily identify them. Similarly, when a DL algorithm gets trained to a dataset via exposure through a number of epochs/iterations, it can then test and validate the same dataset through certain features. This can both be used for regression and classification. DL basically extracts features from the SERS-Based Sensing Mechanisms 262 Surface-Enhanced Raman Scattering‒Based Plasmonic Sensors Figure 8.17 Illustration of training and classification using AI through ML (Adapted with permission from Ref. [39]). 263 data, then learns and models complex relationships. Convolutional neural networks (CNNs) are a class of deep neural networks that are mostly used in computer vision but have much wider applications in different fields and also can be used with different classes of neural network architectures. A few notable models with CNN architectures are ResNet developed by Microsoft, VGGNet [37] developed by Visual Geometry Group, and Inception Net [38] developed by Google, etc. Figure 8.18 General scheme presenting the generation pipeline of an ANN applied to SERS spectroscopy (Adapted with permission from Ref. [39]). At first, the experimentally acquired Raman/SERS spectral dataset is divided into three parts, 80%, 15%, and 5%, respectively, named training, validation, and test datasets and preprocessed identically. Each spectrum composing the abovementioned datasets is then labeled to a certain corresponding identity, defined as a feature or label. For example, a SERS spectrum can be converted into a 1D 1000:1 vector, having 1000 features, which need to be fed for the training of an untrained ANN algorithm, as shown in Fig. 8.18a. An ANN is composed of a number of artificial neurons/ nodes, arranged and layered to create a complex network, which is capable of extracting information from the input. The nodes are the basic computational units of the ANN and have been shown as grey spheres in this figure. After the spectrum is converted into features, each feature is assigned certain weights depending on the importance of the feature. For example, a SERS peak is given high SERS-Based Sensing Mechanisms 270 Surface-Enhanced Raman Scattering‒Based Plasmonic Sensors It was observed that SERS enhancement factors of about 107–108 per molecule could be achieved using optimal plasmonic nSTFs. Further, the optimized nSTFs were shown to work as highly sensitive and selective sensors. Pros and cons of plasmonic nanostructure optimization were discussed and optimal structure was used to demonstrate non-direct sensing of E. coli bacteria. Direct detection of anions and DNA hybridization-based detection of Hg2+ were discussed. ML techniques for the assessment of relatively complex matrices of analytes were introduced. Their functionality along with recent development for the classification of 30 bacterial strains along with 08 antibiotics was presented. This chapter presents a blend of both the basics and applications of SERS for bio and chemical sensing applications. Acknowledgments Sachin K. Srivastava gratefully acknowledged financial support from the MoE-STARS grant, SERB Start-up research grant, FIG IIT Roorkee, DST: Indo-Korea JNC, DST-INSPIRE Faculty Award, and DST-BDTD grants. This work was partially supported by the EU MSCA-RISE H2020 project IPANEMA grant agreement N° 872662. References 1. Hutter, E. and Fendler, J. H. (2004). Exploitation of localized surface plasmon resonance. Adv. Mater., 16, pp. 1685–1706. 2. Hao, E. and Schatz, G. C. (2004). Electromagnetic fields around silver nanoparticles and dimers. J. Chem. Phys., 120, pp. 357–366. 3. Fleischmann, M., Hendra, P. J., and McQuillan, A. J. (1974). Raman spectra of pyridine adsorbed at a silver electrode. Chem. Phys. Lett., 26, pp. 163–166. 4. Jeanmaire, D. L. and van Duyne, R. P. (1977). Surface Raman spectroelectrochemistry: Part I. Heterocyclic, aromatic, and aliphatic amines adsorbed on the anodized silver electrode. J. Electroanal. Chem. and Interfac. Electrochem., 84, pp. 1–20. 5. Moskovits, M. (2005). Surface-enhanced Raman spectroscopy: A brief retrospective. J. Raman Spec., 36, pp. 485–496. 6. Valley, N., Greeneltch, N., van Duyne, R. P., and Schatz, G. C. (2013). A look at the origin and magnitude of the chemical contribution to the 271 enhancement mechanism of surface-enhanced Raman spectroscopy (SERS): Theory and experiment. J. Phys. Chem. Lett., 4, pp. 2599–2604. 7. le Ru, E. C., Blackie, E., Meyer, M., and Etchegoint, P. G. (2007). Surface enhanced Raman scattering enhancement factors: A comprehensive study. J. Phys. Chem. C, 111, pp. 13794–13803. 8. Ding, S. Y., You, E. M., Tian, Z. Q., and Moskovits, M. (2017). Electromagnetic theories of surface-enhanced Raman spectroscopy. Chem. Soc. Rev., 46, pp. 4042–4076. 9. Nie, S. and Emory, S. R. (1997). Probing single molecules and single nanoparticles by surface-enhanced Raman scattering. Science, 275, pp. 1102–1106. 10. Pang, S., Yang, T., and He, L. (2016). Review of surface enhanced Raman spectroscopic (SERS) detection of synthetic chemical pesticides. Tr. Anal. Chem., 85, pp. 73–82. 11. Pilot, R., Signorini, R., Durante, C., Orian, L., Bhamidipati, M., and Fabris, L. (2019). A review on surface-enhanced Raman scattering. Biosensors, 9, pp. 9020057. 12. Sharma, B., Frontiera, R. R., Henry, A. I., Ringe, E., and van Duyne, R. P. (2012). SERS: Materials, applications, and the future. Mater. Today, 15, pp. 16–25. 13. Shvalya, V., Filipič, G., Zavašnik, J., Abdulhalim, I., and Cvelbar, U. (2020). Surface-enhanced Raman spectroscopy for chemical and biological sensing using nanoplasmonics: The relevance of interparticle spacing and surface morphology. Appl. Phys. Rev., 7, pp. 031307 14. Cialla, D., März, A., Böhme, R., Theil, F., Weber, K., Schmitt, M., and Popp, J. (2012). Surface-enhanced Raman spectroscopy (SERS): Progress and trends. Anal. Bioanal. Chem., 403, pp. 27–54. 15. Fateixa, S., Nogueira, H. I. S., and Trindade, T. (2015). Hybrid nanostructures for SERS: Materials development and chemical detection. Phys. Chem. Chem. Phys., 17, pp. 21046–21071. 16. McNay, G., Eustace, D., Smith, W. E., Faulds, K., and Graham, D. (2011). Surface-enhanced Raman scattering (SERS) and surface-enhanced resonance Raman scattering (SERRS): A review of applications. Appl. Spec. 65, pp. 825–837. 17. Liedtke, S., Grüner, C., Lotnyk, A., and Rauschenbach, B. (2017). Glancing angle deposition of sculptured thin metal films at room temperature. Nanotech., 28, pp. 385604. 18. Abdulhalim, I. (2014). Plasmonic sensing using metallic nanosculptured thin films. Small, 10, pp. 3499–3514. References 272 Surface-Enhanced Raman Scattering‒Based Plasmonic Sensors 19. Srivastava, S. K., Shalabney, A., Khalaila, I., Gru¨ner, C., Rauschenbach, B., and Abdulhalim, I. (2014). SERS biosensor using metallic nanosculptured thin films for the detection of endocrine disrupting compound biomarker vitellogenin. Small, 10, pp. 3579–3587. 20. Robbie, K. and Brett, M. J. (1998a). Sculptured thin films and glancing angle deposition: Growth mechanics and applications. J. Vac. Sci. Tech. A: Vac., Surf. Films, 15, pp. 1460. 21. Gall, D. D. and Gall, D. (2011). Sculptured thin films: Nanorods, nanopipes, nanosmiles. Proc. SPIE, 8104, pp. 158–164. 22. Shalabney, A. H., Khare, C., Bauer, J., Rauschenbach, B., and Abdulhalim II, I. (2012). Detailed study of surface-enhanced Raman scattering from metallic nanosculptured thin films and their potential for biosensing. J. Nanophoton, 6, pp. 061605. 23. Srivastava, S., Hamo, H., Kushmaro, A., Marks, R., Gru¨ner, C., Rauschenbach, B., and Abdulhalim, I. (2015). Highly sensitive and specific detection of E. coli by a SERS nanobiosensor chip utilizing metallic nanosculptured thin films. Analyst, 140, pp. 3201–3209. 24. Novotny, L. (2007). Effective wavelength scaling for optical antennas. Phys. Rev. Lett., 98, 26, pp. 266802. 25. Schneider, C. A., Rasband, W. S., and Eliceiri, K. W. (2012). NIH image to ImageJ: 25 years of image analysis. Nat. Meth., 9, pp. 671–675. 26. Chung, T., Lee, S.-Y., Song, E. Y., Chun, H., and Lee, B. (2011). Plasmonic nanostructures for nano-scale bio-sensing. Sensors, 11, pp. 10907– 10929. 27. Tang, H., Zhu, C., Meng, G., and Wu, N. (2018). Review—surface enhanced Raman scattering sensors for food safety and environmental monitoring. J. Electrochem. Soc., 165, pp. B3098–B3118. 28. Tan, S., Erol, M., Sukhishvili, S., and Du, H. (2008). Substrates with discretely immobilized silver nanoparticles for ultrasensitive detection of anions in water using surface-enhanced Raman scattering. Langmuir, 24, pp. 4765–4771. 29. MicroSEQ® E.coli O157:H7 Detection Kit (2014), http://tools. lifetechnologies.com/content/sfs/brochures/CO13764.pdf, Accessed January 21, 2022. 30. Kalele, S. A., Kundu, A. A., Gosavi, S. W., Deobagkar, D. N., Deobagkar, D. D., and Kulkarni, S. K. (2006). Rapid detection of Escherichia coli by using antibody-conjugated silver nanoshells. Small, 2, pp. 335–338. 273 31. Han, D., Lim, S. Y., Kim, B. J., Piao, L., and Chung, T. D. (2010). Mercury (ii) detection by SERS based on a single gold microshell. Chem. Comm., 46, 30, pp. 5587–5589. 32. Ma, W., Sun, M., Xu, L., Wang, L., Kuang, H., and Xu, C. (2013). A SERS active gold nanostar dimer for mercury ion detection. Chem. Comm., 49, pp. 4989–4991. 33. Lee, S., Lee, H., and Chung, H. (2013). New discrimination method combining hit quality index based spectral matching and voting. Anal. Chim. Acta, 758, pp. 58–65. 34. Khan, S. S. and Madden, M. G. (2012). New similarity metrics for Raman spectroscopy. Chemomet. Intel. Lab. Sys., 114, pp. 99–108. 35. Rodriguez, J. D., Westenberger, B. J., Buhse, L. F., and Kauffman, J. F. (2011). Quantitative evaluation of the sensitivity of library-based Raman spectral correlation methods. Anal. Chem., 83, pp. 4061–4067. 36. Vandenabeele, P., Hardy, A., Edwards, H. G., and Moens, L. (2001). Evaluation of a principal components-based searching algorithm for Raman spectroscopic identification of organic pigments in 20th century artwork. Appl. Spec., 55, pp. 525–533. 37. Simonyan, K. and Zisserman, A. (2014). Very deep convolutional networks for large-scale image recognition. 3rd International Conference on Learning Representations, ICLR 2015 - Conference Track Proceedings. 38. Szegedy, C., Liu, W., Jia, Y., Sermanet, P., Reed, S., Anguelov, D., Erhan, D., Vanhoucke, V., and Rabinovich, A. (2015). Going deeper with convolutions. In Proc. IEEE Conf. Comp. Vis. Patt. Recog., pp. 1–9. 39. Lussier, F., Thibault, V., Charron, B., Wallace, G. Q., and Masson, J.-F. (2020). Deep learning and artificial intelligence methods for Raman and surface-enhanced Raman scattering. Tr. Anal. Chem., 124, pp. 115796. 40. Ho, C.-S., Jean, N., Hogan, C. A., Blackmon, L., Jeffrey, S. S., Holodniy, M., Banaei, N., Saleh, A. A., Ermon, S., and Dionne, J. (2019). Rapid identification of pathogenic bacteria using Raman spectroscopy and deep learning. Nat. Commun., 10, pp. 1–8. 41. Hossin, M. and Sulaiman, M. N. (2015). A review on evaluation metrics for data classification evaluations. Int. J. Data Min. and Knowled. Managem. Proc., 5, pp. 1–11. References