UNIVERZA V MARIBORU FAKULTETA ZA NARAVOSLOVJE IN MATEMATIKO Oddelek za fiziko MAGISTRSKO DELO. Simon Hamler

UNIVERZA V MARIBORU FAKULTETA ZA NARAVOSLOVJE IN MATEMATIKO Oddelek za fiziko MAGISTRSKO DELO Simon Hamler Maribor, 2015

UNIVERZA V MARIBORU FAKULTETA ZA NARAVOSLOVJE IN MATEMATIKO Oddelek za fiziko Magistrsko delo VPLIV TEMPERATURE NA POVRŠINSKO OJAČAN RAMANSKI SPEKTER 2,4,6-TRINITROTOLUENA Master Thesis INFLUENCE OF TEMPERATURE ON THE SURFACE ENHANCED RAMAN SCATTERING SPECTRA OF 2,4,6-TRINITROTOLUENE Mentor: doc. dr. Marko Jagodič Kandidat: Simon Hamler Somentor: dr. Hainer Wackerbarth Maribor, 2015

ACKNOWLEDGEMENT I would like to thank my mentor dr. Marko Jagodič and co-mentor dr. Hainer Wackerbarth for help and guidance when writing this thesis. I would also like to thank my family for always being there for me.

UNIVERZA V MARIBORU FAKULTETA ZA NARAVOSLOVJE IN MATEMATIKO IZJAVA Podpisani Simon Hamler, rojen 4. 6. 1987, študent Fakultete za naravoslovje in matematiko Univerze v Mariboru, študijskega programa Fizika, izjavljam, da je magistrsko delo z naslovom Vpliv temperature na površinsko ojačan ramanski spekter 2,4,6-trinitrotoluena pri mentorju dr. Marku Jagodiču in somentorju dr. Hainerju Wackerbarthu avtorsko delo. V magistrskem delu so uporabljeni viri in literatura konkretno navedeni; teksti in druge oblike zapisov niso uporabljeni brez navedb avtorjev. Maribor, Podpis:

UNIVERZA V MARIBORU FAKULTETA ZA NARAVOSLOVJE IN MATEMATIKO Hamler S.: Vpliv temperature na površinsko ojačan ramanski spekter 2,4,6- trinitrotoluena. Magistrsko delo, Univerza v Mariboru, Fakulteta za naravoslovje in matematiko, Oddelek za fiziko, 2015. IZVLEČEK Zaznavanje sledi eksplozivov, kot je trinitrotoluen (TNT), je pomembno področje pri preprečevanju terorističnih napadov. Površinsko ojačana ramanska spektroskopija (SERS) je postala močna detekcijska tehnika za identifikacijo majhnih količin analitov. V magistrskem delu so predstavljeni podatki o TNT raztopini, naneseni na nanostrukturirano zlato površino, ki je ogreta do 60 C. Zaznane spremembe, ki jih opazimo na podlagi mikroskopskih slik in SERS spektrov, razložimo s pomočjo izhlapevanja, faznega prehoda in razgradnje TNT molekul. Vpliv temperaturne odvisnosti na SERS učinek je bil raziskan na kemisorbiranem monosloju 4- nitrothiophenol molekul. Da bi zmanjšali izhlapevanje TNT molekul, je bil med plazmonsko površino in TNT vstavljen samosestavljiv monosloj mercaptoheksanola (MCH). KLJUČNE BESEDE: površinsko ojačana ramanska spektroskopija, eksplozivi, temperaturna odvisnost, izhlapevanje, fazni prehod, molekulski razpad.

UNIVERZA V MARIBORU FAKULTETA ZA NARAVOSLOVJE IN MATEMATIKO Hamler S.: Influence of the Temperature on the Surface Enhanced Raman Scattering Spectra of 2, 4, 6-Trinitrotoluene. Master Thesis, University of Maribor, Faculty of Natural Sciences and Mathematics, Department of Physics, 2015. ABSTRACT The detection of trace amounts of explosive like trinitrotoluene (TNT) is an important issue in the prevention of terrorist attacks. Surface enhanced Raman scattering (SERS) spectroscopy has become a powerful detection technique for identification of minute amounts of analytes. This thesis presents data of TNT in solution, deposited on a nanostructured gold surface, which is heated up to 60 C. The observed changes in the microscopy images and in the SERS spectra are explained by evaporation, phase transition and decomposition of the TNT molecules. The impact of temperature dependence of SERS effect is studied on a chemisorbed 4-Nitrothiophenol monolayer. To minimize the evaporation of TNT molecules, a self-assembled monolayer of mercaptohexanol (MCH) was inserted between plasmonic surface and TNT. KEYWORDS: surface enhanced Raman spectroscopy, explosives, temperature dependence, microscopy, evaporation, phase transition, decomposition.

CONTENTS 1 INTRODUCTION... 1 2 RAMAN SPECTROSCOPY... 4 2.1 Energy units... 4 2.2 Degrees of freedom and molecular vibrations... 5 2.3 Basic theory... 7 2.3.1 Comparison of Raman and Fluorescence processes... 12 2.4 Polarizability tensor... 13 3 SURFACE ENHANCED RAMAN SCATTERING (SERS)... 14 3.1 Electromagnetic Enhancement (EM)... 15 3.1.1 Hot spots... 21 3.2 Chemical mechanism... 21 4 EXPERIMENTAL SECTION... 23 4.1 Raman setup... 23 4.2 Laser... 25 4.3 Spectrometer and CCD camera... 26 4.4 Plasmonic substrate... 28 4.5 Microscope... 28 4.6 Sample preparation... 29 4.7 Data analysis... 30 5 RESULTS AND DISCUSSION... 31 5.1 Microscopic observations... 31 5.2 SERS measurements of TNT... 32 5.2.1 Evaporation... 35 5.2.2 Phase transition... 35 5.2.3 Decomposition... 36 5.2.4 Temperature dependence of the SERS effect... 37 5.3 TNT solution deposited on the substrate covered with mercaptohexanol (MHC) monolayer... 39 6 CONCLUSIONS... 41 7 RAZŠIRJENI POVZETEK V SLOVENSKEM JEZIKU... 43 REFERENCES... 46

1 INTRODUCTION Owing to numerous attacks and attack attempts during the last years, the protection of our society against terrorism has gained meaning. In this context, the detection of explosives and their associated compounds is an important issue. This has led to development of new detection technologies, especially in the field of homeland security, to face the problems of hidden explosives at public places, such as airports, bus and train stations. Many techniques have been investigated for this purpose. However, the majority is not ideal for explosive detection, since they have disadvantages such as invasiveness, detect only certain explosive, but fail to detect others, or require complicated sample preparation. Vibrational spectroscopy has shown to be an excellent technique for rapid, accurate quantitation and can be used for studying very wide range of sample types and can be carried out from a simple identification test to an in-depth, full spectrum, qualitative, and quantitative analysis [1]. Vibrational spectroscopy includes several different techniques. However, the most important of them are infrared (IR) spectroscopy and Raman spectroscopy. Both of these techniques can provide a complementary information about molecule vibrations in many instances. Although both study the interaction of radiation with the molecule, they differ in a manner in which photon is transferred to the molecule by changing its vibrational state. IR spectroscopy measures transitions between molecular vibrational energy levels, based on the direct absorption of light quanta. Absorption of photons occurs, when the frequency of radiation by the polychromatic light matches that of a vibration. Therefore, the molecule is prompted to a vibrational excited state. The loss of this frequency of radiation from the beam after it passes through the sample is then detected [1, 2]. On the other hand, Raman spectroscopy is based on a scattering mechanism and requires monochromatic light for detection of molecular vibrations. A portion of the incident photons will be scattered inelastically. Therefore, the energy of scattered photons will differ from that of the incident photons. The energy difference corresponds to the difference between the vibrational levels of the molecule. The IR and Raman vibrational bands are characterized by their frequency (energy), intensity (polar character or polarizability), and band shape (environment of bonds). Since the vibrational energy levels are unique to each molecule, the IR and Raman spectrum provide a fingerprint of a particular molecule. The frequencies of these molecular vibrations depend on the masses of the atoms, their geometric arrangement, 1

and the strength of their chemical bonds. Their spectrum provides information on a molecular structure, dynamics, and environment [1]. In this thesis, the Raman spectroscopy technique was used. There are two main advantages of Raman over IR spectroscopy. The first one is that the samples can be confined or sealed in optical transparent materials (glass, quartz), since they do not absorb the light. Because of this, Raman spectroscopy is suited for the analysis of reactive or environmentally sensitive compounds. The second advantage is that it is suitable for examining samples in aqueous solution. Vibrations of water are weakly Raman active and hence does not interfere the Raman signal, while in IR spectroscopy, the absorption of water is very strong and thus often superimpose the signals of interest. Raman spectroscopy is already an established technique in analytical and forensic science, collecting a unique chemical signature of molecules. Almost all explosives can be identified by their Raman spectra. Therefore, neat explosives have been extensively studied by Raman spectroscopy [3 5]. Further advantages are that the detection is rapid and non-invasive. There are hardly any limitations to the sample, which can be present in different physical states or as a composition of different compounds. Moreover, there is hardly sample preparation. However, Raman spectroscopy is not suitable for the detection of trace amounts, as it is needed for the prevention of bomb attacks. Nevertheless, the inherently weak Raman process can be greatly improved using surface enhanced Raman scattering (SERS). SERS combines low detection limits with high information content about molecular identity making it highly suitable for trace analysis [6, 7]. Based on SERS single molecule, detection was achieved. The enhancement factors can be as high as ~10 14-10 15, if SERS is combined with other effects like resonance Raman. However, the substantial contribution to the enhancement comes from SERS [8-11]. The surface enhanced Raman effect is a product of two mechanisms, the electromagnetic and chemical enhancement. The electromagnetic mechanism is believed to be responsible for the bulk of the enhancement (~10 4-10 8 ) and is based on the increase of the electromagnetic field strength in the vicinity of nanostructured metal surface. The enhancement is greatest when the surface plasmon frequency and the incident light are in resonance. In comparison to electromagnetic effect, the chemical enhancement is quite modest (~10-10 2 ) and arises from interaction between adsorbed molecule and the metal surface. The charge transfer between adsorbate and metal increases the 2

polarizability of adsorbed molecules. Another advantage of SERS is the quenching of fluorescence, which is a known obstacle in Raman spectroscopy of explosives [3]. The aim of this work was to study how increasing temperature affects the SERS spectra of TNT, deposited on a nanostructured gold surface and also if at the elevated temperature, SERS measurements are still possible. Chapter 2 and 3 include necessary theoretical background of normal Raman and surface enhanced Raman scattering. The experimental setup is introduced in Chapter 4, with detailed description of its major components. In the following chapter, we start with the microscopic observations of TNT deposited on nanostructured gold substrate and continue with the investigation of the temperature dependence of the intensity of TNT SERS spectra. Similar study is also performed for the 4-Nitrothiophenol adsorbed on a substrate and mercaptohexanol (MCH), which is placed between the surface and the TNT molecules. In conclusions, we summarize our results. 3

2 RAMAN SPECTROSCOPY The phenomena of inelastic scattering of light was first predicted by Smekal in 1923 and first experimentally observed in 1928 by C.V. Raman and K.S. Krishnan in India and, independently, by L. Mandelstam and G. Landsberg in the former Soviet Union. Since then, this phenomenon is referred as Raman effect. In the original experiment, sunlight was focused onto the sample by the telescope and the second lens, which was placed by the sample, collected the scattered light. Using the system of optical filters, they showed the existence of scattered light with a frequency different from that of the incident light the basic characteristic of Raman spectroscopy [2]. 2.1 Energy units Light is an electromagnetic radiation that can act like waves or like a stream of small packets of energy called photons. This is also known as wave-particle duality. Figure 1 illustrates a wave of linearly polarized electromagnetic radiation propagating in the x- direction. It consist of electric and magnetic field components, which always oscillate in phase with each other and are perpendicular to one another to the direction of wave propagation. Since the Raman effect does not involve the magnetic component, only the former will be further discussed. The oscillation of electric field of strength (E) at a given time (t) is expressed as [12]: E E cos 2t, (1) 0 where E 0 is the amplitude of the incident electric field and is the frequency of the radiation. In EM radiation, the frequency and the wavelength proportional to each other [12]:, are inversely c, (2) where c is speed of light. In Raman spectroscopy, instead of the wavelength the wavenumber is used and given by: 7-1 10 cm c nm. (3) 4

Figure 1: Linearly polarized electromagnetic radiation [13]. A transfer of energy from electromagnetic radiation to the molecule occurs when following condition is satisfied: where E c E h h hc, (4) is the energy difference between two quantized states and h is Planck 34 constant ( 6.626 10 Js ). Thus, wavenumber is directly proportional to the energy 14 of the transition. For example, 1 ev corresponds to 1240 nm, 2.4 10 Hz, and 8065 cm -1 [12]. 2.2 Degrees of freedom and molecular vibrations Degrees of freedom describes the motion of the atoms in x, y, or z direction. An N-atom molecule will therefore have a total of 3N degrees of freedom of motion. Three of these degrees of freedom describe the translational motion of the molecule and three of them describe the rotational motion of the nonlinear molecule about the three principal axis of rotation, which go through the centre of gravity. A linear molecule only has two rotational degrees of freedom, since rotation around its own axis is not considered a degree of freedom of motion (no nuclei displacements are involved). After subtracting translational and rotational degrees of freedom from total 3N degrees of freedom, the net vibrational degrees of freedom (number of normal vibrations) is 3N 6 for nonlinear and 3N 5 for linear molecule. This means that for diatomic molecule, we will have only one vibration. In the case of H2O molecule, we have 33 6 3 normal vibrations as shown in Figure 2. These modes of vibration are symmetrical stretch, bending, and asymmetric stretch. The linear CO2 molecule has 4 modes of vibration. 5

However, the model discussed here has only three. The fourth mode is also bending vibration but in a different plane as shown in Figure 2. Such pair of vibrations with the same frequency, different only in their direction, are called doubly degenerate vibrations [2, 12, 14]. Figure 2: Normal modes of vibration for CO2 (+ and denote vibrations going upward and downward, in direction perpendicular to the paper) and H2O. Based on [12]. For better understanding of molecular vibrations, which are responsible for characteristic bands in Raman spectra, we consider a simple model of diatomic molecule, as shown on Figure 3. Atoms, with masses m 1 and m 2 are connected with chemical bond, which in this case can be regarded as massless spring, with force constant, k. Displacement of atoms from their equilibrium position is x 1 and x 2. Displacement of each of the two masses varies periodically over the period of time as a sine (or cosine) function. Atoms oscillate with different amplitudes, but with the same frequency, thus both masses go through equilibrium position simultaneously. The classical vibrational frequency for diatomic molecule is: 1 1 1 k. (5) 2 m 1 m 2 It can be seen from the above equation that the frequency of diatomic oscillator is a function of atomic masses of the two atoms, involved in the vibration and the force constant k, which is a measure of bond strength between the two atoms. If the atoms are connected with double or triple bond, the force constant will be bigger and consequently the frequency will also be higher [1, 12]. 6

Figure 3: Motion of a simple diatomic molecule [1]. 2.3 Basic theory The Raman effect is a light scattering phenomenon. When a monochromatic light of energy h 0 interacts with the molecule in a material, it can be scattered. The oscillating electric field of light distorts (polarize) the electron cloud around nuclei and form a virtual state, which is not necessarily a true quantum state of the molecule. Because the state is not stable, the photon is quickly re-radiated [15]. In vibrational spectroscopy, the detected energy changes are those that require changing the vibration of nuclei. The dominant scattering, also called the Rayleigh scattering, is a process where only electron cloud distortion is involved. This scattering is referred as elastic scattering, since the energy/frequency of the photon is the same as before interacting with the molecule. On the other hand, Raman scattering is a weak process, where only one in 10 6-10 8 of scatter photons are Raman scattered. This occurs when incident light induce a change in the nuclear motion and energy will be transferred either from molecule to scattered photon or from incident photon to molecule. This process is referred as inelastic scattering, since the energy of scattered photons differs from that of the incident photons. If a molecule at the ground vibrational state is excited by the incident photon to the virtual state and relaxes to a higher vibrational excited state, the molecule gains energy and the energy of scattered photon is smaller than that of incident photon h 0 m. This process is called Stokes scattering. If the molecule is already in an excited vibrational state due to the thermal energy, the scattered photon may gain energy from the molecule h 0 m, leading to anti-stokes scattering. Since most of 7

the molecules at room temperature are in the ground vibrational state, the majority of Raman scattering is Stokes scattering [2, 15]. Figure 4: Diagram for Rayleigh and Raman scattering [16]. The classical description of Raman scattering can be explained by Eq. (6). As already mentioned earlier, the oscillating electric field of light E interacts with a molecule and distorts electron cloud, thereby inducing an electric dipole moment P in the molecule. The magnitude of induced dipole moment P depends on the polarizability of the molecule and the strength of the electric field E of the incident radiation. This can be expressed as: 0 0 P E E cos 2 t. (6) Polarizability is proportionality constant and can be described as the ease with which molecular orbitals are deformed, by the presence of the external field. The more easily the electron cloud of molecule is distorted, the bigger the polarizability and thus greater the induced dipole moment of the given field. If the molecule vibrates with a frequency m, the nuclear displacements q can be written as q q cos 2 0 mt (7) where q 0 is the vibrational amplitude. Using a small amplitude approximation, polarizability can be expressed as a linear function of displacement in the form of Taylor series: 8

0 q q q0 (8) Here 0 is the polarizability at the equilibrium position, and qq0 represents the rate of change in polarizability with respect to the change in displacement from the equilibrium position. If the derivative is equal to zero (no change in polarizability), the vibration does not yield Raman scattering. Oscillations of polarizability cause the induced dipole moment to oscillate at frequencies other than the incident frequency Combining Eq. (6) with Eq. (7) and Eq. (8), we obtain [12,15]: 0. P E cos 2 t 0 0 0E0 cos 2 0t qe0 cos 2 0t q q0 0E0 cos 2 0t q0e0 cos 2 0t cos 2 mt q q0 1 0E0 cos2 0t q0e 0 cos2 0 mt cos2 0 mt 2 q q0 (9) 1 was used in the 2 The trigonometric identity cos cos cos cos final step of Eq. (9). According to classical theory, Eq. (9) demonstrates that the light will be scattered at three different frequencies. The first term is the Rayleigh scattering and represents an oscillating dipole which radiates light at the same frequency as the incident light 0. The second term corresponds to Raman scattering where oscillating dipole radiates light at frequencies, which are different from the frequency of incident beam that is 0 (anti-stokes) and 0 m (Stokes). The magnitude of these shifts m reflects the characteristic vibration of the molecule [12]. Some conclusions can be made from Eq. (9): 1) As already stated before, if q 0 0, the second term vanishes. The vibration q is not Raman active, since the molecular polarizability does not change during the vibration. 2) If the vibration does not greatly change the polarizability, then the polarizability derivative will be near zero and the signal from Raman scattering will be low. 9

Scattering intensity is proportional to the square of the induced dipole moment P, which is proportional to the square of the polarizability derivative q 3) The equation also shows two possibilities to increase the Raman intensity. The one is from the molecules with the larger polarizability and the other one is the stronger electric field experienced by the molecules [17]. 2. Thus, in Raman spectroscopy we measure the shift of the vibrational frequency 0 from the incident beam frequency. A Raman spectrum consist of scattered intensity plotted vs. frequency shift between incident and scattered photons. The frequency shift, also called Raman shift, is defined as [15]: m E 1 1 hc incident scattered (10) where E is the energy difference between initial and final vibrational state of the molecule. Raman shift is independent of wavelength of the incident beam incident if we change incident a way, that the, since, the wavelength of the scattered photons scattered changes in such remains the same. As already mentioned in the beginning of this chapter, there are always more molecules on the ground vibrational state than in the excited vibrational state at the room temperature. This is why Stokes lines are much stronger than anti-stokes lines. The ratio of Stokes and anti-stokes intensities depends on the population in ground and excited vibrational states and can be obtained from the Boltzmann distribution [15]: I I AS S 4 0 m h m exp 4 k 0 m BT (11) 23 where k B is the Boltzmann constant 1.38 10 J K. Measurement of this ratio can also be used for temperature measurements. Since peaks from both lines are positioned symmetrically with respect to the Rayleigh peak. Usual Raman spectrometers only acquire Stokes spectra. A typical Raman spectrum, in this case CCl4, is illustrated in Figure 5. 10

Figure 5: Raman spectrum of CCl4 [12]. The intensity of Raman scattering IR is given by [1]: 2 4 IR I0N t q (12) where is the frequency of the incident radiation, I0 is the intensity of the incident radiation, N is the number of scattered molecules, is the polarizability of molecules, q is the vibrational amplitude, and t is the acquisition time. The above expression shows that increasing laser flux power or using shorter wavelength excitation gives us a higher Raman intensity. However, since the molecules usually have a bigger absorption cross section at lower wavelengths towards UV, the fluorescence, which is a competing process and millions of times more efficient than the Raman effect, is also higher and can thus overwhelm the Raman signal. Because of such an inequality in signal strength, even the trace quantities of fluorescent materials can mask the Raman signal of high-concentration analyte. This is why we usually use excitation at longer wavelengths. Therefore, the wavelength selection is a balance between minimizing the fluorescence and maximizing the signal strength [18]. 11

2.3.1 Comparison of Raman and Fluorescence processes Both Raman scattering and fluorescence produce photons with the frequencies different from that of the incident photon, however, they are fundamentally different from each other: Raman scattering in a molecule is an instantaneous event in which an incoming photon from the laser at 0 excites a molecular vibration m while emitting a scattered photon at s o m. The incident photon does not need to be absorbed and induce electronic transitions in the molecule, since Raman process can be considered as an interaction with the virtual state, as depicted in Figure 4. Because of this, Raman effects can take place at any frequency of the incident light, whereas fluorescence is anchored at a specific excitation frequency. Fluorescence, on the other hand is not an instantaneous, but a stepwise process. The initial step involves the absorption of the incident light, where the system is transferred from the ground single state S0 to a state in the vibrational substructure of the first singlet state S1. The absorption process is shown in Figure 6a. Unlike in Raman, the photon must have enough energy to reach S1 and start fluorescence event. Once in the excited state, the molecule undergoes a series of vibrational relaxations process, reaching the vibrational ground state of S1 (Figure 6b). After certain amount of time (typically around few nanoseconds), the molecule relaxes back to the vibrational levels of the ground state, thus emitting the photon (Figure 6(c)). In fluorescence, the emission process is completely independent of the initial absorption, since both photons are not linked to each other in coherent and instantaneous way like in Raman, where for each photon taken from the laser, there will be a scattered photon (one cannot exist without the other). However, in fluorescence, there are situations where some potentially emitted photons from the ground state of S1 go missing (e.g. in non-radiative combination). Once the molecule is excited to the S1 state, the best-case scenario is to recover all the photons that have been excited in the initial absorption process. However, small fraction will usually be missing through a process that allows the molecule to relax back to the ground state of S0 without emitting a photon. Therefore, the two processes are effectively disconnected in fluorescence (unlike in Raman) [19]. 12

a) b) c) Figure 6: Schematic presentation of fluorescence process as a sequence of events over time. a) Fluorescence starts with the absorption of a photon. b) In the first electronically excited state S1, the molecule undergoes vibrational relaxation and c) after few nanoseconds, it relaxes back to the vibrational levels of the ground state S0, and thereby emits a photon [19]. 2.4 Polarizability tensor To discuss Raman activity, we have to look more carefully at the polarizability. In actual molecules, a nice linear relationship P E does not hold, since molecular response to the applied electric field is not the same in every direction. Both P and E are vectors, consisting of three components in x, y and z direction. Thus, Eq. (6) can be written in the matrix form [12]: Px xx xy xz Ex P E. (13) y yx yy yz y P z zx zy zz E z The first matrix on the right side is the polarizability tensor of second order. The tensor is symmetric. The Raman scattering occurs when one of the components in polarizability tensor changes during the vibration. For small molecules, it is easy to see whether polarizability changes during the vibration. If we consider diatomic molecules (e.g. H2) or linear molecules (e.g. CO2), the electrons are more polarizable (larger ) along the chemical bond than in direction perpendicular to it. Figure 7a shows changes in polarizability from the vibrations of the CO2 molecule. Polarizability tensor is graphically represented as the polarizability ellipsoid. This is a three-dimensional body, whose distance from the electrical centre of the molecule is proportional to 1, where i is the polarizability in i-direction from the centre of gravity in all directions. When i 13

xx yy zz polarizability ellipsoid will be a sphere and molecule is said to be isotropic. For a completely anisotropic molecule, xx yy zz applies. The vibration is Raman-active if the polarizability ellipsoid changes in its size, shape or orientation and the intensity will depend on extent of this change. The 1 vibration is Raman-active since polarizability changes in all directions. On the other hand vibrations 2 and 3 are Raman-inactive. Although in both cases the polarizability changes during the vibration, the size and shape of the ellipsoid at +q and q are identical by symmetry. Note that the Raman activity is determined by the slope near the equilibrium position, qq0 (Figure 7b) [12]. a) b) Figure 7: a) Changes in polarizability ellipsoid during three normal vibrations of CO2 molecule. b) The polarizability of CO2 as a function of displacement coordinate q for 1 and 3 vibrations (the function of the displacement is the same for 2 and 3 ) [12]. 3 SURFACE ENHANCED RAMAN SCATTERING (SERS) Surface enhanced Raman scattering (SERS) was discovered by Fleischmann and coworkers in 1974 when they obtained an unusually strong Raman signal from pyridine adsorbed on electrochemical roughened silver electrode. The reason to roughen the electrode was to increase the surface area and thus the number of adsorbed molecules. In 1977, Jeanmarie et al. and Creighton et al. confirmed the results and pointed out that the Raman signal of molecules adsorbed on metal was enhanced by a factor of ~10 6, 14

compared to a signal from molecules in absence of metal. They reported that such enhancement cannot be explained just by an increase in surface alone, but is also related to an intrinsic surface enhancement effect. 40 years since discovery, improvements in instrumental capabilities, better understanding of the enhancement effect and advances in nanotechnology, made SERS spectroscopy a powerful analytical tool, used in various fields, including physics, chemistry, and biology. It exploits the interaction of light, molecules and metal nanostructured surface to enhance the Raman signal, in some cases even to 14 orders of magnitude, thus allowing detection of single molecules. The total enhancement is a product of two mechanisms, electromagnetic and chemical or electronic enhancement. The dominant effect is electromagnetic enhancement (~10 4-10 8, depending on the nanostructured surface) and is associated with magnification of both incident and Raman-scattered fields, while chemical enhancement ( 10 2 ) arises from electronic interaction between metal and adsorbed molecules. Both mechanisms will be discussed in detail in the next chapter [20]. 3.1 Electromagnetic Enhancement (EM) The electromagnetic enhancement effect occurs at the metal-air interface. When electromagnetic wave interacts with the metal surface, it causes collective oscillations of the conduction electrons in metal - surface plasmons (SP). A plasmon is a quantum of plasma oscillation. The plasmon can be consider a quasiparticle since it arises from the quantization of plasma oscillations, just as phonons are quantizations of mechanical vibrations. Thus, plasmons are collective oscillations of the free electron gas density, for example, at optical frequencies. Surface plasmons are those plasmons that are confined to surfaces and that interact strongly with light resulting in a polariton. If the incident light is in resonance with plasmon frequency, the electromagnetic field at the surface is enhanced. The electric field of SP can be expressed as 0 15 i kxxkzzt E E e (14) with for z 0, for z 0 and where x is direction of propagation parallel to the surface, z direction perpendicular to the surface, k x and k z are wave vectors components along the x- and z-axis and is the frequency of the longitudinal oscillation. The electromagnetic field disappears at z and is the strongest when z 0, which is typical for surface waves.

The solution of the Maxwell equations for the electric field from Eq. (15) at the metaldielectric interference with dielectric constants M and D yields the dispersion relation SP of SP [21]: SP M D cksp, (15) M D where k SP is the wavevector of SP. The dielectric constant of metals is expressed as a ' '' complex value i. The real part M M M ' M is associated with the polarizability of incident light and imaginary part '' i M with the absorption. The complex function is usually frequency dependent. In order to achieve the enhancing effect of the plasmons, the electric field of the incident photon must oscillate parallel to the plane of the incidence, so that its components lies in the direction of SP propagation. For a resonant coupling of SP and photons, the energy E and the momentum p k has to be conserved. The wavevector component of incident light k Ph, x depends on the incident angle and can be described with the help of the dispersion relation of the incident photon c Ph kph, where following equation [21] D k Ph is a wavevector of an incident photon with the Ph, sin (16). c k Ph x D For resonance, the conservation of energy is given when Ph SP and the following equation applies k k M D Ph, x D sin SP c c M D (17). The Eq. (17) relationship is shown graphically in Figure 8 [21]. 16

Figure 8: a) Wavevectors components of incident photon, k Ph and surface plasmons, k SP, along a smooth metal surface. b) Dispersion relation of incident photon and surface plasmons [based on 21]. Figure 8b shows that SP dispersion relation curve never intersects with the dispersion relation line of a light in air with a dielectric constant, 1. Consequently, on a smooth metal surface, SP cannot be excited directly by just free-space light. The SP inplane wavevector is greater than that of incident light and since the momentum must be conserved, the SP cannot radiate to the surrounding media. To turn a light line to the point, where it intersects with dispersion relation curve of SP to get a resonant effect, and thus enhancement, the dielectric constant of the surrounding medium has to be bigger than 1. This is usually achieved using a coupling medium such as prism. Another possibility for resonant excitation of SP is roughening (usually nanostructuring) the metallic surface to get a grating. In this way, k SP D k Ph, x is matched with, by increasing the parallel wavevector component of the incident light with the wavevector of the grating k ph, x. Dispersion relation in this case is fulfilled by the sum 2 kph, x sin k ph, x sin n ksp (18) c c a where n is the integer and a is the grating constant. k ph, x 0 gives no solution to dispersion relation. This is shown schematically on a Figure 9 [21]. 17

Figure 9: a) Wavevectors components of incident photon k Ph and surface plasmons k SP along a nanostructured metal surface. b) Dispersion relation of incident photon and surface plasmons [based on 21]. Due to the direction of SP, modes involving changes in molecular polarizability with a component along the surface are the most enhanced. The most often used metals in SERS are Ag, Au and Cu, since those materials have a negative real and small positive imaginary dielectric constant (proportional to the damping of surface plasmons). Also these materials fulfill the resonance condition in the visible or NIR frequency range. SP can either be propagating in the x- and y-direction (~10-100 µm) along the metaldielectric interface and decay evanescently in the z-direction (~200 nm), or can be localized on spherical particle for example. In the latter case, we talk about localized surface plasmons (LSP). Since both, SP and LSP, are sensitive to the surrounding dielectric environment, both are also used for SERS sensing experiments [8, 20, 22]. Figure 10: Illustration of the a) SP and the b) LSP [8]. A simplified schematic diagram for understanding the concept of electromagnetic SERS enhancement is shown in Figure 11. The metallic nanostructure is a small sphere with the dielectric constant in a surrounding medium with a dielectric constant 0. Since 18

the radius of the sphere is much smaller than the wavelength of light r 0.05 electric field is uniform across the particle and the Rayleigh approximation can be used [23]., the Figure 11: Simple schematic diagram for understanding the concept of EM SERS enhancement. A molecule in the vicinity of the sphere (distance d) is exposed to a field EM, which is the superposition of the incident field E0 and the field of a dipole Esp induced in the metal sphere, therefore EM E0 E. The magnitude of the dipolar field Esp is given sp as [23]: E sp 3 0 r 2 0 rd E 0 (19). The field enhancement factor molecule and the incident field A is the ratio of the field at the position of the A 3 EM 0 r E0 2 0 r d (20). A is particularly strong when the real part of the denominator is zero (i.e., real part of is equal to 2 0 ). Additionally, the imaginary part of the dielectric constant needs to be small for a strong electromagnetic enhancement. This condition describes the resonant excitation of surface plasmons of the metal [23]. In an analogous way to the incident light, the scattered Stokes/anti-Stokes is enhanced. Taking into account enhancement of the incident and the Stokes light, the electromagnetic enhancement factor G S for Raman signals can be written as: 19

2 2 12 r G A A r d 2 2 L 0 S 0 em S L S L 2 0 S 2 0 (21). The above equation shows that the enhancement scales as the fourth power of the local field of the metallic nanostructure and that is particularly strong when both incident light is in resonance with the surface plasmons and the inelastically scattered light is close to this resonance [23]. The Eq. (21) also indicates that electromagnetic SERS enhancement is a long range effect, which means that the adsorbate is not required to be in direct contact with the surface of the metal. Enhanced EM fields generated by the surface plasmon resonance (SPR) enables the detection of molecules even few nanometers from the surface of the substrate. Long range effect of EM mechanism differs from the chemical enhancement mechanism, where molecules have to be in direct contact with the surface. The detection of the molecules nearby to the surface, which are not necessarily bound to it, can be very useful in SERS applications, since many analytes have low or no affinity to Ag or Au. In such cases, the surface can be modified with adlayers to improve specificity of the analytes. The field enhancement around metal sphere decays with the growing distance, described by the decay of the field of a dipole over the distance 3 1 d, as shown in Eq. (20), to the fourth power, resulting to the 12 1 d (see Eq. (21)) [8]. The power of SERS signal is proportional to the following parameters: The power of the Raman signal 2 2 R SERS L L S ads P N I A A (22). P SERS depends on the number of molecules N, laser intensity IL, enhancement factors of excitation A L, the scattered field A S R the Raman cross section of the adsorbed molecule ads [20]., and on The increase in Raman intensity when the molecule is adsorbed on a SERS active substrate is described by the enhancement factor (EF). The average EF for a SERS system, which is evaluated at the single excitation frequency and the same acquisition, is given as ISERS N surf EF (23) I N NRS vol 20

where I SERS is a surface-enhanced Raman intensity, N surf is the number of molecules adsorbed to the metallic surface that contribute to the SERS signal, I NRS is the intensity of normal Raman scattering and the N vol is the number of molecules in the excitation volume [24]. The SERS enhancement factor also strongly depends on the orientation of the adsorbed molecules with the respect to the metal surface. Enhancements are stronger if the vibrations of the adsorbate are parallel to metal surface. 3.1.1 Hot spots Hot spots are highly localized regions of intense local field enhancement, which are caused by local surface plasmon resonances (LSPR). Hot spots occur when intense electromagnetic field of two nanostructures superimpose. Nanostructures can be nanoparticles or structures with a gap in-between. This phenomenon is strongly dependent on the excitation wavelength, particle size, shape, and separation as well as arrangement with respect to the polarization direction of the incident light. Many articles in literature describe hot spots created between two nearby nanoparticles (gap junctions) that line up their induced field with the external field. In other words, the incident field induces two in-phase dipoles along the direction of the incoming field. If d a, where d is the distance between particles and a is a particle diameter, the nearfield interactions will dominate and fall according to d 3. Therefore, in order to create extremely high field confinements (hot spots), it is important to produce very small (few nanometers) gap junctions, i.e. d a. When optical excitation is localized in such small area, extremely large electromagnetic SERS enhancement up to generated, thus allowing observations of single molecules [23, 25]. 12 10 can be 3.2 Chemical mechanism Chemical or electronic SERS effect is a common name for different mechanisms, which require direct contact between molecule and metal surface. Early researchers found out that the SERS intensities between molecules of CO and N2 differ by a factor of 200 under the same experimental conditions. This result was very hard to explain just with electromagnetic mechanism, since the polarizabilities of the molecules are nearly 21

identical and even the most radical variation in orientation upon adsorption could not produce such large differences [26]. Chemical enhancement mechanism can be explained by the electronic coupling between metal and a molecule, adsorbed on the metal surface. This adsorbate-surface formation produces an increase of Raman cross section of the adsorbed molecule, compared to the cross section of molecules in normal Raman experiment. Other possible explanation involve resonance Raman effect, which can occur due to shifted and broadened electronic levels in the adsorbed molecule (compared to the free one) or due to the new electronic transition in metal-molecule system. The later occur through photon driven charge transfer process (PDCT) between metal and adsorbates, which is shown in Figure 12, where HOMO and LUMO denote for the highest occupied molecular orbital and the lowest unoccupied molecular orbital of the adsorbate, respectively. The energies of HOMO and LUMO are approximately symmetric relative to the Fermi level of the metal. The whole process can be described by the following four steps [23, 26]: Step 1: An electron-hole pair of the metal is created by the incident photon with energy h 0 and the electron is excited to the hot-electron state. Step 2: So-called hot electron tunnels into the LUMO of the adsorbed molecule, generating a charge transfer excited state. Step 3: Hot electron tunnels from LUMO (with changed normal coordinates of some internal molecular vibrations) back to the metal. Step 4: The electron recombines with the hole created in the step 1, which leads to a vibrationally excited neutral molecule and to a emission of a Raman shifted photon, with the energy h. 22

Figure 12: Schematic diagram of the photon-driven charge transfer model for a molecule adsorbed on a metal [27]. Chemical enhancement mechanism is a short-range effect (0.1 0.5 nm), limited only to first layer of adsorbed molecules. It depends on the geometry of bonding, the adsorption site and the energy levels of the adsorbate molecule. The contribution of chemical enhancement to the SERS intensity is estimated to be approximately 10-10 2. However, is generally agreed that the electromagnetic enhancement is significantly larger in magnitude. Although the chemical enhancement is not the general mechanism and is restricted only to specific adsorbate-metal systems, it can still provide us useful information on chemisorption and hence interactions between adsorbate and metal [20]. SERS spectrum can show some deviations in relative intensities compared to a normal Raman spectrum of the same molecule. Interactions between molecule and metal may cause, that the Raman lines are slightly shifted in frequency and changed in line width compared with a free molecule. Despite small changes, which can occur in SERS spectrum, it still provides us a very clear fingerprint of the molecule [23]. 4 EXPERIMENTAL SECTION 4.1 Raman setup The Raman setup used in our experiments consists of four major components: excitation source (laser), light focusing and collecting system (in our case a fiber optic probe head), spectrometer and CCD detector. A schematic of the apparatus is shown in Figure 13. 23

Figure 13: Schematic presentation of major components in our Raman setup [28]. The SERS spectra were collected with a standard system (Kaiser Optical System Inc., Ann Arbor, MI, USA); the 785 nm (linewidth 0.06 nm) GaAlAs diode laser (Invictus, Kaiser Optical Systems, Inc.) beam was focused onto the sample. The excitation and scattered light were guided onto/from the sample by a multimode optical fiber equipped with the probe head. The incident power of the laser emission was about 100 mw at a probe head for 5 s recording with 1 accumulation on the detector. The scattered light was coupled into the optical fiber by a confocal aperture and guided to the spectrograph (P h AT System TM, Kaiser Optical Systems Inc.), which uses Volume Phase Holographic (VPH) transmission gratings to perform filtering and dispersion functions. Prior to entering the spectrograph, the scattered light goes through a holographic notch filter, which cuts off photons at the laser frequency (i.e. Rayleigh scattering). The diffracted light was recorded with a CCD camera (idus, Andor Tecnology plc.) with a spectral resolution of 5 cm -1. The most important prerequisite when comparing Raman shifts is the reproducibility (or repeatability) of the experiment. Kaiser Optical Systems provide a Raman shift tolerance between ± 0.5 and ± 1.0 cm -1, the individual system performance will not vary to this extent. Upon calibration, a system should yield Raman shift values reproducible to ± 0.1 cm -1 [7]. 24

Figure 14: Front view of the Raman spectrometer. 4.2 Laser Lasers are ideal excitation sources for Raman spectroscopy, since they provide a monochromatic light with narrow bandwidth and high intensities necessary for generation of a sufficient amount of Raman scattered photons. In addition, laser beams have a small spot diameters that can be further reduced using optical lenses (smallest possible diameter is approximately equal to the laser wavelength) for higher photon flux at the measurement zone. Since the scattering intensity scales with the fourth power of laser frequency (Eq. (12)), the most logical thing for improving Raman sensitivity would be to use highest possible frequency. However, the problem that arises with the use of high frequencies (or short wavelengths) is the emergence of fluorescence, which can cover the Raman signal. The presence of the fluorescence can be reduced by using longer wavelengths. Nowadays, the most common light sources for Raman spectroscopy are diode lasers operating at near-infrared (NIR) wavelength. Fluorescence at such wavelengths is not completely absent, but it is significantly suppressed. The Raman intensity is however weaker, since the energy of radiation is lower and the fourth power law applies. We used a continuous wave (cw), Invictus 785 nm NIR diode laser, with a maximum output power of 450mW. It is rated as class IIIb laser, meaning, that eye damage can occur upon direct exposure to the laser beam. This class applies for laser with no more than 500 mw of radiant power. It uses external cavity design to provide a narrow 25

linewidth and excellent wavelength stability. The Invictus laser also has integrated holographic bandpass filter, which rejects any spontaneous emission from the diode that is not at the lasing wavelength [29]. The probe head uses non-contact optics, which are optimized for incident NIR radiation. The working distance is 1 cm, while the aperture ratio is f/2.0 [30]. 4.3 Spectrometer and CCD camera The spectral separation of Raman scattered light was performed using P h AT System TM spectrograph from Kaiser Optical Systems Inc. Instead of a classical surface relief reflection grating (usually in Czerny-Turner design) (Figure 15a) as dispersive element, the P h AT System TM spectrograph uses Volume Phase Holographic (VPH) transmission grating to perform filtering and dispersion functions (Figure 15b). VPH grating is made from a layer of transparent material, usually dichromated gelatin, which is sandwiched between two layers of clear glass or fused silica. When the light passes through the optical thin film that has a periodic differential hardness or refractive index, its phase is modulated. Hence the term Volume Phase. This is the biggest difference in comparison to conventional reflection gratings, where the phase of the incident light is modulated by the depth of a surface relief patterns. As in the conventional reflection gratings, the spectral dispersion or angular separation of wavelength components in diffracted light is determined by the spatial frequency of the periodic structure [31]. a) b) Figure 15: a) Classical surface relief reflection grating. b) VPH transmission grating [31]. Spectrally separated light is collected with the idus DU420-BRDD charge-coupled device (CCD) camera, which consists of rectangular two-dimensional arrays of 1024 26

256 photosensitive elements (pixels). The pixel site is 26 26 µm, while the image area is 26.6 6.6 mm. For readout, the detector uses so-called Full Vertical Binning (FVB) method. Collected signal is converted into a 16-bit grayscale image. The operation temperature of the camera was at 66 C. The silica based CCD sensors of the DU420-BRDD CCD camera have the highest quantum efficiency in the NIR region (Figure 16). Since we used lasers with an excitation wavelength of 785 nm, our spectral range of interest ranges from 785-915 nm, which corresponds to approximately 1800 cm -1 (Eq. (10)). In this region, quantum efficiencies from 50 to 90% can be achieved. Figure 16 also shows a decrease in quantum efficiency for wavelengths above 750 nm. The reason for the decrease lies in the wavelength dependant absorption of photons in silica, which is why excitation wavelengths longer than 785 nm can be unfavourable for detection with a CCD camera, since weak Raman signals cannot be detected [21]. The so called deep depletion technologies enables high quantum efficiencies in the NIR. Devices manufactured with this technology have thicker photosensitive silicon layers, which offer longer absorption path to photons with longer wavelength and thus increase the probability for creation of excited electron-hole pairs [21]. Figure 16: Quantum efficiency of idus DU420-BRDD CCD camera [adapted from 21]. 27

4.4 Plasmonic substrate All SERS spectra were recorded on a commercially available nanostructured gold substrate (Klarite, Renishaw Diagnostics). The size of the substrate is 6 mm 10 mm, with an active area of 4 mm 4 mm. The active area of the substrate consists of goldcoated periodic square lattice of inverted pyramid pits (~1.4 μm wide and ~1 μm deep), shown on Figure 17. The pyramid pits were produced using conventional optical lithography on a (100) oriented silicon wafer followed by an anisotropic chemical etching [6]. The substrate was opened from a vacuum-sealed package just prior to experiment, to prevent any possible surface contamination. Figure 17: SERS substrate with visible active area (left) and scanning electron microscope images of the nanostructured gold surface (middle), and inverted pyramid pits (right) [6]. The substrate was heated with Peltier element, with the size of 25 25 mm and maximum working temperature of 138 C. 4.5 Microscope TNT solution on the SERS substrate was examined by Carl Zeiss El-Einsatz Axioskop microscope We used a 100 microscope objective, with the numerical aperture of 0.90 and dark field illumination technique. Images were taken by the Lumenera s INFINITY2-2 digital CCD camera. 28

Figure 18: Carl Zeiss El-Einsatz Axioskop microscope. 4.6 Sample preparation In SERS measurements, it is essential that the molecules are delivered to close proximity of a metal surface. The molecules were analysed as thin films by dropping a small volume of solution on the substrate. We used TNT solved in methanol/acetonitrile (1 mg/ml), which was purchased from AccuStandard, Inc. (New Haven, USA). TNT is a nitroaromatic compound (Figure 19a), which is mostly used for military and industrial explosives applications. The melting point of TNT is at 80 C and is thus far below the temperature at which it will spontaneously detonate. It is also relatively insensitive to shock and friction and the explosive cannot be initiated without a detonator. Self-assembled monolayers of 4-Nitrothiophenol (technical grade from Sigma-Aldrich) were generated by soaking the substrates in a 13 mm nitrothiophenol / ethanol (p.a. from Sigma-Aldrich) solution for 24 h (Figure 19b). Before SERS measurements, the substrate was rinsed with ethanol and left to dry in air for 10 min. In the same way, by soaking the substrate in a 1 mm mercaptohexanol (MCH) / ethanol solution for 24 h, was also made a self-assembled monolayer of MCH (technical grade from Sigma-Aldrich) (Figure 19c). 29

a) b) c) Figure 19: a) Chemical structure of TNT, b) 4-Nitrothophenol and c) MCH adsorbed on a gold surface [6]. 4.7 Data analysis When recording a Raman or SERS spectra, in addition to the signal from the sample we also detect some background noise, which can alter the profile of Raman bands. Background noise can emerge because of different reasons; from not-fully suppressed Rayleigh light from the laser, fluorescence or stray light. Because of this, background subtraction or so-called baseline correction (Figure 20) is performed for each TNT SERS spectrum. To perform a baseline correction it is necessary to create a baseline based on a recorded spectrum, which is then subtracted from the spectrum. The baseline was calculated with the help of local and global minima. Here, spectrum is divided into intervals and in each of these intervals local, and global minima are searched. Per interval, we have one support point or a node. These nodes are connected together with the cubic spline, resulting in a baseline, which can be seen as a red curve in the Figure 20. The baseline is then subtracted from the recorded spectrum and we get the correct TNT spectrum without a background noise (blue line in Figure 20). Baseline subtraction on all spectra was performed with Origin 9.0. 30

Figure 20: Principle of the baseline correction. From a recorder TNT spectrum (black) we subtract baseline (red), calculated on the basis of a local on global minima. Result is a TNT spectrum without a background noise (blue). 5 RESULTS AND DISCUSSION 5.1 Microscopic observations TNT diluted in methanol/acetonitrile was dropped on a nanostructured gold surface. Even at small volumes, the solvents spread across the surface and evaporate, leaving molecules adsorbed on the surface. Before and after the SERS measurements, in which the substrate was heated to 60 C, we examined the nanostructured surface with the microscope. One corner of the substrate was chosen to make sure we would observe the same area during the experiments. Figure 21a shows the nanostructured surface with inversed pyramid pits after depositing a drop of solution on it. The border to the unstructured area is also clearly seen. At first, it appears that observed spheres on the surface could be liquid bubbles of the solution, but after being stable for 30 minutes, we can conclude that all volatile solvents already evaporated. We assume that the flattened spheres are pure TNT crystals. This agrees quite well with the SERS measurements recorded immediately after microscope observations, which did not show any signs of 31

methanol or acetonitrile in the spectra. The distribution of TNT crystals on this substrate is heterogenic and the crystal sizes are in the range from ~2 µm to ~12 µm. Figure 21b shows the microscope image after heating it up to 60 C and then cooling it back to 20 C. Almost all of the bright spheres on a substrate have disappeared and dark patches are seen on the gold surface, where the large bright spheres were located before. The smaller bright spheres disappeared completely, indicating evaporation of the TNT microcrystals. Figure 21: Microscope images of a solution on the nanostructured gold substrate a) before and b) after heating to 60 C. 5.2 SERS measurements of TNT After the microscope examination, we immediately started the SERS measurements. The temperature dependence of the intensity of the TNT Raman bands was studied. We started at 20 C and continued to 60 C in 5 C intervals. The acquisition time was 5 s. A typical SERS spectrum of TNT is shown on Figure 22. The characteristic bands of TNT are in the range between 200-1800 cm -1. Moreover, we observed peaks in the region around 3000 cm -1, which can be assigned to different C-H vibrations. At each temperature, the SERS spectrum was obtained for three random spots on the substrate surface (Figure 23). Different intensities of the TNT bands for different spots can be explained by the heterogenic distribution of TNT crystals in the excitation focus on the nanostructured surface. Each spectrum for a given temperature is the average of three spectra at different locations on the surface. 32

Figure 22: SERS TNT spectrum at 20 C. Figure 23: Three SERS spectra of TNT at 20 C, obtained at three random spots on the substrate surface. 33

TNT can be easily identified by the vibrational modes at the following frequencies: 323 cm -1 (2,4,6 C-N in plane torsion, ring in-plane bend), 792 cm -1 (ring in-plane bend, C- CH3 stretch), 824 cm -1 (nitro group scissoring mode), 1207 cm -1 (ring breathing), 1356 cm -1 (4-NO2 symmetric stretching, C-N stretch), 1542 cm -1 (NO2 asymmetric stretching) and 1616 cm -1 (phenyl modes) [32]. TNT SERS spectra from 20 C to 60 C are shown in Figure 24. Due to the temperature change of the SERS substrate, a red-shift of the position of the frequencies up to 12 cm - 1 arises between SERS spectra of 20 C and 60 C. In Figure 25, the behaviour of the most dominant band of the TNT spectrum at 1356-1 cm (NO2 symmetric stretching vibration) is shown. The intensity decreases by a factor of 5. In the following, contributions of evaporation, phase transition, decomposition, and temperature dependence of the SERS effect are discussed on the base of the microscopy and SERS results. Figure 24: SERS spectra of TNT from 20 C to 60 C in 5 C interval. 34

Figure 25: Intensity of the dominant TNT band (1356 cm -1 ) at different temperatures. 5.2.1 Evaporation Evaporation of TNT molecules from the surface is a major contribution for the decrease of the signals, as indicated by the microscopic images, in which some of the shiny spheroids are disappeared after the heating. However, evaporation could not explain dark patches and the changes in the spectra. 5.2.2 Phase transition The change of the shiny TNT crystals to dark patches (Fig. 21) can be explained by a phase transition. The intensity of the band at 1356 cm -1 decreases between 20 and 60 C (Figure 25). Above 35 C the decrease in intensity is steeper than at lower temperatures, whereas above 55 C the intensity appears to be constant. The melting point of bulk TNT is at 80 C. However, the melting point is size-dependent. Therefore, the drop of the melting point can be explained by the small size of the TNT crystals on the surface [33]. Moreover, volatile solvents can also have some effect on crystallization of TNT, since the crystals could be formed differently, which may affect their quality, resulting also in a melting point depression. Considering the heterogenic size distribution of TNT crystals and their size dependent melting points, the temperature dependent behaviour of the SERS intensities resembles conceivably a sigmoidal 35

melting curve, supporting the phase transition. The change of the shiny crystals and the sigmoidal curve shape of the temperature dependence indicate melting of the TNT crystals upon the heating. What exactly are these dark patches and in which phase are they, is unknown. 5.2.3 Decomposition Decomposition of the molecules is an issue in SERS spectroscopy, in particular as the laser beam is focused on a small area at the surface. However, we have not observed decomposition of TNT on such a substrate at room temperature under these conditions (laser power, acquisition time) before. After heating the substrate to 60 C, we cooled it back to 20 C and recorded a spectrum. The comparison of SERS spectra at 20 C before and after heating to 60 C is shown in Figure 26. The heating and recording of SERS spectra result in a non-reversible chemical process. Figure 26: Comparison of TNT spectra at 20 C at the beginning of SERS measurements (black) with spectra at 20 C, after cooling it down from 60 C (red). Characteristic TNT bands at 323 cm -1, 792 cm -1, 824 cm -1, 1207 cm -1 and 1542 cm -1 have totally vanished. The intensity of the dominant band at 1356 cm -1 has dropped to approximately 24% of the original value and has also shifted by 12 cm -1 to a lower 36

frequency. Moreover, the band at 1617 cm -1 is shifted to 1607 cm -1. In contrast, the band at 1128 cm -1, which had a low intensity at the beginning of the measurements, increases for about four times. This clearly indicates that we can observe newly generated unknown chemical species. A simple elimination of a nitro group which would decompose TNT to DNT could not be explanation, as DNT can be identified by two characteristic bands at 834 cm - 1 and 1327 cm -1 [34]. Consequently, it looks like that melting of the crystals is followed by the decomposition of the TNT molecules by heating the plasmonic surface. 5.2.4 Temperature dependence of the SERS effect Temperature dependence of the SERS effect could be further explanation for the decreasing intensities in the SERS spectra. Pang et al. have published a study about the temperature dependence of the SERS effect [35]. These authors found a decrease of the SERS effect with rising temperatures. From 15 K to 300 K the enhancement drops by a factor of approximately 3. To verify the impact of the temperature dependence of the SERS effect, we adsorbed 4-Nitrothiophenol on the surface. These molecules form a covalent gold-sulfur bond with the nanostructured substrate leading to a defined and stable monolayer. Thus, evaporation and phase transition effects should not be relevant. Spectra were recorded from 20 C to 80 C in 5 C intervals; spectra were taken at three different positions on the surface. Figure 27 shows SERS spectra of 4-Nitrothiophenol between 20 C and 80 C. The characteristic bands of 4-Nitrothiophenol are at 1081 cm -1 (C S stretching vibration), 1343 cm -1 (N O symmetric stretching mode) and 1570 cm -1 (C C stretching mode of benzyl ring) [36,37]. The position of the most dominant peak shifts slightly from 1344 cm -1 to 1342 cm -1 (Figure 27 inset). The temperature dependence of the intensity of this band is shown in Figure 28. The behaviour can be described by a linear decrease (Figure 28 inset). The overall decrease in intensity between 20 C and 80 C is approximately 11 %. Besides this small drop in intensity, we cannot find any significant changes in the spectra. The small decrease could nevertheless be a consequence of desorption of the molecules caused by the heating. On the other side, the drop can also be caused by the temperature dependence of the SERS effect. However, the effect is small compared to the drop of the TNT intensities by heating. It may contribute to rather small extent to the decrease. Moreover, 37

we can conclude that the used SERS substrate is suited for applications up to 80 C. That means that detection by SERS at elevated temperature is possible. Figure 27: SERS spectra of 4-Nitrothiophenol recorded at 20 C (black), 50 C (red) and 80 C (blue). Figure 28: Intensity of the dominant 4-Nitrothiophenol band at 1343 cm -1. 38

5.3 TNT solution deposited on the substrate covered with mercaptohexanol (MHC) monolayer MCH was adsorbed on a gold surface, where the molecules formed a defined and stable self-assembled monolayer, similar as with 4-Nitrothiophenol. Then we dropped the TNT solution on top of the formed monolayer. The main idea of inserting a layer of MCH between the gold surface and the TNT was to diminish the evaporation of the TNT molecules by the formation of hydrogen bonds between hydroxyl group of MCH and the nitro group of the TNT. Before the SERS experiment, we have taken microscope images with just MCH monolayer adsorbed on the surface. As seen on Figure 29a, the MCH monolayer is not visible and we can see just a nanostructured gold surface with some impurities on it (bright yellow dots). After measuring the spectra of the MCH at 20 C, we dropped the TNT solution on top of the monolayer (Figure 29b). In comparison to Figure 21a, there are not visible any TNT crystals in the shape of the flattened spheres. Noticeable are just dark patches on the surface. Since on the planar gold surface area, TNT crystals are clearly seen, we can assume that the added monolayer of MCH had an effect on the adsorption of the TNT molecules on the surface. The formation of dark patches was also observed after heating the uncoated substrate. Obviously, the heating and the MCH layer cause the formation of a non-crystal phase of TNT molecules on the surface. The substrate was then heated to 60 C. Figure 29c shows a nanostructured surface after cooling it back to 20 C. Dark patches that were visible before heating are now disappeared, which indicates that the TNT molecules evaporate from the surface. a) b) c) Figure 29: Microscope images of a) MCH monolayer and b) MCH monolayer with added TNT on it, before heating to 60 C. c) MCH monolayer with added TNT after cooling it back to 20 C. 39