Tailoring photonic response for enhanced Raman spectroscopy

(1)

Tailoring photonic response for

enhanced Raman spectroscopy

Master Thesis

Author: Tomas Kaandorp Student number: 10356169 Supervisor: Prof. Dr. A. F. Koenderink Second assessor:Prof. Dr. E. C. Garnett

Daily supervisor: Msc. K. G. Cogn´ee

Resonant nanophotonics group, Amolf

and

Institute of physics, University of Amsterdam

(2)

Abstract

We attempted to establish if one can do Raman scattering in hybrid plasmonic-dielectric resonators. These resonators had been developed in the Resonant Nanophotonics group to provide plasmonic hot spots at wavelengths of around 780 nm, and at quality factors of 500 to 5000. First, we build a setup capable of interrogating these resonators to perform Raman spectroscopy. A unique prop-erty of this setup is the fact that it is outfitted with a tuneable laser allowing us to relate the pump- and Raman enhancements to the features of photonic structures in a SERS scenario. We applied this unique feature to measure the Raman scattering enhanced by the hot spot of a gap mode plasmonic nanocube on a metal film. We observed Raman signals from ligands attached to individ-ual cubes, and found that their intensity depends on how well the pump and Raman line are tuned to the cube resonance. The spectra further show a broad Stokes and anti-Stokes background. This might either be due to a broad Ra-man response of the long carbon chains of the ligand, or alternatively might be due to fluorescence by the metal nanoparticle itself, as reported by Carattino. We recommend further research to determine if luminescence or indeed Raman scattering of the carbon chain lead to these spectra.

(3)

Introduction

In the beginning of 1928 Sir Raman and his student K. S. Krishnan led a se-ries of experiments on the scattering of light. During these experiments, they discovered a form of light scattering which is unique per material species [21]. This scattering process of light by a molecule not only endorsed the quantum nature of light, it also led to a strong identification tool for materials.

Nowadays, Raman spectroscopy is used in a number of fields. For example in bio-sensing, where it is used for the detection of Anthrax or glucose[22]. Portable Raman spectroscopy devices are now also used for the detection of explosives[17], for example on airports. These examples are only a small insight into the use of Raman spectroscopy and one can imagine that Raman spectroscopy is applied in much more disciplines than only the medical or safety industries as mentioned before.

Further advances in technology have led to the detection of single molecules [14]. These advances are remarkable since the Raman scattered light is in general very feeble and of low intensity, meaning that it is sometimes difficult to identify single molecules. One method of increasing the Raman signal strength, involves Surface Enhanced Raman Spectroscopy(SERS). A rough metal surface creates hotspots in the electric field which lead to an increase in the Raman signal strength. This allows for higher Raman signal strength at reduced molecule numbers under study.

These surfaces have a drawback, they are not spatially selective. Firstly, the width of the laser beam, even the most tightly focused, is not small enough to only excite one hotspot and secondly, any particle present at a hotspot will be enhanced. This means it is never clear if the Raman spectra of the desired molecule is measured or if the measured Raman signal is polluted by other molecules[21].

The long term goal of the research project envisioned at the start of my project, was to establish if one can do Raman scattering in hybrid plasmonic-dielectric resonators. These resonators had been developed in the Resonant Nanophotonics group to provide plasmonic hot spots at wavelengths of around 780 nm, and at quality factors of 500 to 5000. In this thesis work I set out

(6)

to adapt a set up for interrogating these resonators to perform Raman spec-troscopy. Compared to other Raman scattering set ups a unique property is the ability to tune the excitation wavelength. In this thesis this capability is applied to Raman scattering in the hot spot of gap modes of plasmonic nanocubes on a metal film.

Lastly, at the end of this thesis, some advice is given for future research and possible options for further improvements on the system and how to move towards measuring Raman scattering in hybrid plasmonic-dielectric resonators as envisioned at the start of the project.

(7)

Chapter 2

Principles of Raman

spectroscopy

In this chapter, we will give the reader a feeling for the principles of Raman

scattering. We start by a simple masses on springs description of chemical

bonds to understand the origin of molecular vibration. Then we lay out the two main mechanism by which light can couple to these vibration: direct infrared (IR) absorption of photons and Raman scattering, and explain how this physics allow for powerful spectroscopic tools to identify chemical species. Then, we briefly explain the influence of quantified vibrational energy levels on Raman scattering, in particular the efficiency of the process. And finally, we present techniques used so far to enhance Raman scattering cross-sections.

2.1 Classical description of Raman scattering

In first order approximation, a molecule can be approximated by an ensemble of masses (atoms) bound together via springs (chemical bonds), see figure 5.6. Quite similarly, a crystal can be described an infinite array of masses bond via springs, the vibration is however delocalized over the full crystal. When in contact with a thermal environment (away from 0 K), these chemical bonds can be excited and start vibrating.

In the simplest and didactical case of a diatomic molecule (Fig. 5.6), the

coupled equations of motion of the two masses under arbitrary driving Fdr,1/2

and negligible damping are

m1x¨1= −k(x1− x2) + Fdr,1

m2x¨2= −k(x2− x1) + Fdr,2

, (2.1)

where x1/2 are the centers of mass position of the 2 atoms. we place ourselves

in the barycentric referential system, and introduce the distance between the 2

(8)

Figure 2.1: Graphical representation of a chemical bond for a di-atomic

molecule. Two atoms of masses m1 and m2 are linked via a spring(chemical

bond of stiffness k

motion to describe the system. Indeed, we get from Eq. 2.1 ¨ δx = −k ₁ m1 + 1 m2 δx +Fdr,2 m2 Fdr,1 m1 . (2.2)

We recognize the equation of a driven harmonic oscillator ¨q = −Ω2νq + Gdr, and

identify from Eq. 2.1 the resonance frequency of our harmonic oscillator

Ων = s k ₁ m1 + 1 m2 = r km1+ m2 m1m2 (2.3) .

This resonance frequency depends on the atomic species (masses) and the type of chemical bond (spring stiffness), and therefore can be seen as a finger-print of a certain molecule (or at least a specific chemical bond).

2.1.1 Coupling to electromagnetic fields

Without entering into details about the difficult topic of electronic orbitals, as a molecule is composed of charged particles (atoms), even-though neutral, a molecule can carry electric dipole moments p. We are particularly interested in the dynamic dipole moment that a monochromatic electromagnetic wave, we

assume a laser from now on, at a frequency ωLcan induce. This is expressed as

p = α(ωL)ELcos(ωLt) (2.4)

where α(ωL) is the polarizability of the molecule at the frequency ωL. We

identify two case scenarios of interest here:

• The laser frequency approaches the resonance Ων of the molecule

vibra-tion, usually in the infrared region of electromagnetic spectrum, and the

dipole moment of the molecule is then oscillating at ωL with an exalted

(9)

one measure the extinction of scanned laser beam has the laser frequency is resonant with a molecular vibration.

• The laser is oscillating much faster than Ω, usually in the visible range (NIR-soft UV). The polarizability is weak and non resonant, but since the molecule is vibrating (this always happens, to a certain extend, away from

0K temperature), the polarizability is weakly modulated at Ων; α(t) =

α0+ δα cos(Ωνt) and therfore

p = (α0+ δα cos(Ωνt))ELcos(ωLt)

= α0ELcos(ωLt) + δαEL[cos((ωL− Ων)t) + cos((ωL+ Ων)t)]

. (2.5)

We see that the dipole moment, and therefore the light scattered by the molecule is oscillating at a superposition of three frequencies;

• At ωL, the light is scattered at the same frequency as the incoming laser;

this is Rayleigh scattering.

At ωL± Ων, we have two-sidebands shifted by the same amount from the

driving frequency; these are Raman scattering lines. The red-shifted line

(ωL− Ων) is called ”Stokes” peak, whereas the blue-shifted one (ωL− Ων)

is referred to as ”anti-Stokes” peak.

Note that the distinction between IR and Raman scattering is more pro-found than what we described. Indeed, the coupling of light field to the charge distribution of the molecule is different for both processes, leading to different selection rules depending of the symmetries of chemical bonds and polarization of incoming light. therfore in practice, IR and Raman spectroscopy are often complementary methods, where some chemical bonds can only be interrogated via one technique and not the other.

2.1.2 Molecular composition

It was quickly realized that this peculiar scattering process could have applica-tion in identifying chemical species based on the collecapplica-tion of peaks that appear in scattering spectra upon light illumination. And with the appearance of strong and monochromatic light sources(lasers), Raman spectroscopy became a tool of paramount importance in all disciplines involving chemistry.

To allow for easy identification of different chemical species, the community uses a certain set of unique units. Indeed, as shown earlier, Raman peaks are

shifted from the pump frequency with a certain energy Ων, and to match the IR

spectroscopy community which uses wavenumber in cm−1 instead of frequency

(or energies): ν(cm−1) = _λ(nm)107 , Raman spectra are plotted with x-axis in cm−1

but shifted to have the pump wavelength centered at 0.

∆ν(cm−1) = 10 7 λ(nm)− 107 λL(nm) (2.6)

(10)

where λ is the wavelength of scattered light, and λL is the wavelength of

inci-dent pump light. Therefore for Raman spectra plotted in shifted, cm−1, using

a different laser frequency will only change the relative strength of measured peaks, but not there positions. Note that in practice, most users flip the x-axis to have Stokes peaks on the positive side of the axis. This is the case in Figure 2.2, where we present five different materials and their tipical Raman spectra and explain what type of bond leads to which destinctive raman peak.

Figure 2.2: Five different raman spectra with corresponding uniek Raman shifts

per chemical bond type.

http://bwtek.com/raman-introduction-to-raman-spectroscopy/ retrieved on 31-07-2019

2.2 Implication of quantum nature of molecular

vibrations

As mentioned earlier, the chemical bonds between species in a molecule can be approximated to an harmonic oscillator. This approximation still remains valid

for the first few energy level EN = ¯hΩν N +1₂. However, since at room

tem-perature, kBT ≈ 26 meV , and a Raman shift of ∆ν = 1000 cm−1 corresponds

to an energy of ¯hΩν = 126 meV , it is usually a good approximation to assume

(11)

the ground state N = 0 and the first excited vibrational state N = 1. In these conditions,

• IR absorption means that an IR photon of energy ¯hΩν is absorbed and

thus puts the molecule into the excited vibrational state. see Fig2.3

• Rayleigh scattering (with light of energy ωL >> Ων) can be described

as absorption of the high energy photon that brings the molecule from the ground state into a extremely short lived virtual state (which is often describe as a superposition of other real energy states of the molecule, vibrational, electronic ...), and immediately re-emitted with the same en-ergy, see Fig.2.3.

• For Stokes scattering, the incoming photon also brings the molecule from the ground state to a virtual state, however the emitted redder photon leaves some energy in the molecule which therefore ends up in the vibra-tional excited state. The molecule then goes back to the ground state via the same mechanisms as in IR spectroscopy.See window c in Fig.2.3. • Finally, for anti-Stokes scattering, the molecule is already in the

vibra-tional excited state when it absorbs the incoming photon to a virtual energy state. Then the molecule goes back directly to the ground state by emitting a more energetic photon, see window d in Fig.2.3.

Regardless of selection rules and transitions probability (Fermi’s golden rule), one easily understand that the thermal occupancy of the ground and ex-cited vibrational state play a decisive role in the ratio between Stokes and anti-Stokes scattering cross-section. Indeed, the probability density of the molecule to be in the excited state can be approximated via a Maxwell-Boltzman distri-bution and therefore the ratio between Stokes and anti-Stokes emission is

IAS

IS

≈ e−kB T¯hΩν _(2.7)

where I(A)S is the (anti-)Stokes intensity, KB the Boltzmann constant and T

the temperature. Therefore, for a Raman shift of ∆ν = 1000 cm−1 (¯hΩν =

126 meV ) at room temperature (kBT ≈ 26 meV ), the ratio I_IAS

S is 7.86 ∗ 10

−₃

or in more common notation: IAS

IS =

1 127.23.

(12)

ħω ħω+ħωas ħω-ħωs Rayleigh scattering ħω ħω A B C D ħω

IR Absorption Stokes scattering Anti stokes scattering

Figure 2.3: Four relevant scattering processes. Figure a shows IR absorption, Figure b shows Rayleigh scattering, (the scattered wavelength is equal to the excitation wavelength. Figure C shows Stokes scattering, notice that the scat-tered energy is smaller than the excitation energy. Figure d shows Anti Stokes scattering, notice that the energy of the scattered light is now bigger than the energy of the excitation beam.

2.3 Enhancement of Raman scattering

As we mentioned earlier, Raman scattering cross-sections are extremely small (at least 3 order of magnitude compared to Rayleigh scattering). This becomes a problem when attempting to measure very dilute solutions, or thin-layers, especially when the incident pump power cannot be increased due to damage threshold of the samples.

Luckily, interactions with an interface can increase greatly the signals. We are particularly interested in photonic origins of these enhancement, agnostic of the species under investigation. The most common way to enhance Raman cross-section is to use hotspot occurring upon coupling of light to plasmon in a metallic structure to enhance light matter interaction. Such techniques allow for a boost of up to 11 order of magnitude increase in the Raman cross-section, enough to overcome the related reduction of the number of analysed molecules (only the molecules within a certain volume around the hotspot see their Raman scattering enhanced).

Surface enhanced Raman spectroscopy was the first example of these tech-niques introduced in 1973[11]: a rough metallic surface exhibits plasmonic hotspots upon illumination. The roughness of the surface can be engineered to have a better response for certain illumination wavelength and Raman shifts, it is however practically impossible to precisely address a single molecule. There-fore, a variant, tip-enhanced Raman spectroscopy (TERS)[20] was introduced: enhancement of Raman scattering occurs only at the point of a near atomically sharp pin. This allows to overcome the optical diffraction limit and interrogate not only single molecules, but also specific chemical bonds.

(13)

Finally, if one could target a specific species by some chemical adhesion techniques, it is possible to use single photonic nanostructure on a substrate without the requirement for sophisticated setup as for TERS, but still being able to reach single chemical bond study ([8], [3]. Another added benefit is being able to tailor the resonance to optimize the Raman enhancement. Indeed, a commonly accepted rule in the Raman spectroscopy community is the fact

that the enhancement of Raman scattering by a photonic structure scales asE_E44

0

,

where E and E0 is the electric field at the position of the molecule, respectively

in presence or not of the photonic structure. This rule has however evolved to

E(ωL)2

E0(ωL)2

E(ωS/AS)2

E0(ωS/AS)2, where the first fraction corresponds to an enhancement of the

intensity of the focused pump, the second term quantifies the ease for a Raman photon to be emitted and is related to the local density of electromagnetic states (LDOS). For plasmonic resonances, which can usually be assumed to be broad compared the Raman shift, these two terms are almost the same.

But there also exist Raman enhancement techniques based on dielectric cav-ities[1], where high-Q photonic resonances are much narrower than the Raman shift. These do not allow for spatial selectivity and are not particularly suited to single molecule studies, but the allow for spectral selectivity, which can be a power full tool for discrimination of specific species in a complex chemical solution. Finally, we mention hybrid resonator solutions which would in theory allow for both a spatial and spectral selectivity by playing of the two terms of

the enhancement ”E_E44

0

” rule[1], while also potentially benefiting from superior LDOS properties[10].

(14)

Chapter 3

Experimental setup

With a basic understanding of the key concepts of Raman spectroscopy in mind,

we can start building a setup that can measure Raman scattering. Firstly,

since the Rayleigh scattering is quite strong compared to the Raman signal, the Rayleigh scattered light needs to be filtered out to prevent it from overpowering the Raman scattered light. Next, in order to have a well defined Raman shift, the linewidth of the excitation wavelength needs to be very narrow. Lastly, since the Raman signal is in general very feeble, the collection efficiency needs to be optimized. Taking care of these steps will allow us to reliably determine any possible enhancement of the signal due to nano-resonators.

As an extra feature, a tuneable laser is installed. This will allow the scan-ning of the excitation wavelength. Since the Raman shift is independent of the excitation wavelength, scanning the excitation wavelength aides the identifica-tion and validaidentifica-tion of measured peaks. A measured Raman shift shows an equal absolute shift in wavelength as the shift in pump wavelength, therefore, allowing for the differentiation between Raman peaks or other measured peaks.

Since we are aiming for a spectral sensitive system, it is also vital to scan over the resonance frequency of the optical antenna used to enhance the Ra-man signal. In theory, scanning over the resonance frequency of this optical antenna should lead to change in signal intensity when optimal parameters are reached/scanned over.

In this chapter we will explain how all these steps are accounted for and

implemented in a pre-existing setup. In the end this allows us to measure

Raman spectroscopy.

3.1 Description of used spectroscopy setup

As mentioned, an existing, home build cavity resonance setup is modified. This setup is used and adjusted because it is already fitted with a narrow linewidth spectrometer(Andor Sr-303I-A-SIL) that allows for a resolution of 0.2nm. It is also outfitted with a tunable diode laser(Newport new focus diode laser

(15)

TLB-6712) which can scan the excitation wavelength over a range of 15 nm ranging from 766 nm up to 781 nm and which has a linewidth of 0.4 pm. So two of the requirements are already checked of by using this existing setup. A simplified schematic overview of the setup can be seen in figure: 3.1(for a more detailed schematic overview, see the appendix). Laser light exits from the output facet of a single mode fiber, passes through a (filter performance discussed below). Subsequently the beam is adjusted in diameter, and sent to an objective via a galvo-scanning mirror and a set of scan lenses( f is respectively: 25.4, -25, -100, 100) setup as a petzval pair. This allows for the confocal scanning microscopy. Where the beam focus on the sample can be scanned in x and y directions over the sample. The light forms a collimated beam on the back focal plane of a microscope objective (Nikon, 0.95 NA 100x), leading to a spot size of 0.4 micron on the sample plane. The scattered light follows the same path, via the galvo-scanning mirror up to a notch filter(filter performance discussed below). After the notch filter the beam is coupled into a multimode fiber that leads to the spectrometer(Andor Sr-303I-A-SIL). From now on we will refer to this setup as the ”Victorinox” setup. For a more detailed schematic overview of the setup, see figure 7.2 in the appendix.

3.2 Cleanup-filter

Since the linewidth of the excitation laser determines the accuracy of the mea-sured Raman shift, it is vital to filter out all unwanted laser frequencies.

In figure 3.2 a, the spectrum of the laser can be seen. The intensity is plotted on a log scale. It can be seen that the laser has a sharp peak at 779 nm and a broad shoulder in lower intensity ranging from 765 nm up to 795nm. The laser in our setup is limited to a range of 765 nm up to 781 nm. To probe the molecule with a well defined excitation wavelength in this range, the broad shoulder needs to be suppressed, atleast between 765 and 781 nm, so only the sharp peak at 779 nm remains to excite the sample. Therefore a cleanup filter is needed. We used a Shemrock TBP01-790/12 filter. This filter is transparent, i.e. has an optical density (OD) of 0, for a range of 28 nm which fits exactly in the scanning range of our laser. Outside this range the OD quickly increases to OD 3-5 which is needed to suppress the shoulder. The filter is designed for 790 nm, but by installing it under an angle, 24 degrees according to specifications made by Shemrock compared to the optical axis, it could still be used in the setup within the mentioned range of 765 nm up to 781 nm. Figure 3.2 b shows

the characteristics of the filter installed under an angle of 24◦. Without this

filter, we could see that the sample was excited over a broad range resulting in broad not clearly defined or detectable Raman peaks.

This filter is chosen such that the bandwidth of the filter allows for scanning the laser in the range of 766 to 781 nm without having to adjust the angle of the filter, thus preventing any changes in between measurements, therefore increasing repeatability of measurements. To fully filter out the broad shoulder of the laser, a second filter was installed in order to increase the OD value of

(16)

Figure 3.1: Simplified graphical overview of the experimental setup as described above. Laser light excites from an output facet, passes through a cleanupfilter, subsequently the beam diameter is adjusted and send to an objective via a galvo scanning mirror(not represented in the overview). The collimated beam follows the same path, but now passes through a notch filter and is coupled to a multimode fiber connected to the spectrometer. For a detailed layout of the setup see appendix.

(17)

(a) Laser output power as a function of wavelength plotted in Log scale. On the x axis wavelength is plotted in nm and on the y axis the laser inten-sity is plotted in dB, retrieved from https://www.newport.com/p/TLB-6712 on 17-9-2019

(b) Filter range of the used TBp01-790/12 cleanupfilter installed in an angle of 24 degrees(based on

spec-ifications). On the y axis the OD

value is plotted over wavelength in

nm on the x axis. Retrieved from

https://www.semrock.com/filterdetails.aspx?id=tbp01-790/12-25x36 on 17-9-2019.

Figure 3.2: characteristics of the laser and cleanupfilter used in the experimental setup. The broad shoulder in figure 3.2A is suppressed by the filter character-istics in figure 3.2b leaving us with the clear peak visible in the top of 3.2a

(18)

3.3 Notch filters

As mentioned, the majority of the scattered light is Rayleigh scattering. In order to be able to measure the Stokes or Anti stokes scattering, the Rayleigh scattered light needs to be filtered out. Since we want to build a setup that is capable of measuring both the Stokes as the Anti stokes scattered light, a notch filter is needed. The bandwidth of the filter, needs to be equal to, or wider than the scanning range of the laser in order to allow for the scanning of the laser. In our case this means the notch filter needs a bandwidth of at least 15 nm from 766 up to 781 nm. To allow for extra tuneability, we used the NF785-33 produced by Thorlabs. By varying the angle of the filter to the optical axis, the filter range can be shifted. The exact placement and filter range can be checked using a broadband light source (LED in our case). The characteristics of this filter can be seen in 3.3

Figure 3.3: Filtering power and range for the used notch filter. This filter allows the scanning of the excitation laser from 766 up to 781 nm

3.4 Illumination and collection efficiency

Since the Raman signal is directly proportional to the excitation laser power[4, 7, 12], it is vital that there is sufficient laser power on the sample. Therefore we need to determine how much energy is lost by the setup (mostly reflection by optics). In order to monitor the position of our laser focus on the sam-ple, part of the signal beam needs to reach an imaging camera. To do this a 50/50 beamsplitter is used. However for the measurement of Raman spectra this beamsplitter results in at least a factor 2 signal loss. By simply removing the beamsplitter after focusing, this problem is solved. Therefore this beam-splitter is shown in figure 3.1 but it must be noted that it is not installed during measurements.

(19)

These adjustments lead to a maximum power on the sample of 6.6 mW.

Estimated from the NA of the objective used, this corresponds 1.5∗105M W/m2.

For reference purposes(see next chapter), we also used another home built setup called ”Puntbron” in this experiment. Puntbron is fitted with a 532 nm laser and is capable of measuring Raman spectroscopy with excitation power of 1 up to 10 mW. Thus, the home built setup has sufficient power on the sample to excite the molecule for Raman spectroscopy. To prevent us from damaging samples, the power on the sample can be lowered using a optical density (OD) filters which were already installed in the pre-exsisting setup.

With the power on the sample in a reasonable range, the collection efficiency needs to be optimized in order to correctly determine a possible enhancement caused by a resonator. As mentioned before, the imaging beamsplitter needs to be removed before measurement. Dispite using a pelicule beamsplitter, this results a shift of the focus of our signal on the incoupling channel of our spec-trometer which therefore needs to be re-optimized systematically. This also means that when focusing on a sample, we need to make sure the sample is placed as close to the optical axis as possible, and that there will always be some misalignment between the spectrometer and the camera.

(20)

Chapter 4

Validation of the

experimental setup

At the beginning of this project, in the research group no Raman measurements had been performed. One goal for the project was to validate our set up for mea-suring Raman spectra with a tunable excitation laser. To obtain this validation we first needed to assess what the typical performance for a Raman microscope is. Since in literature, Raman reports do not frequently report quantitative count rates, power densities, exposure times and soforth, we used a confocal fluorescence microscope ”Puntbron” with pump laser at 532 nm wavelength as a benchmark. This wavelength is one of the most used wavelengths for Raman scattering, while the use of Puntbron as single molecule fluorescence microscope means that its detection is state of the art in sensitivity. We test the validity of the setup for two observables, namely, the spectral position of the Raman peaks detected and secondly on the sensitivity (number of counts). To quantify the performance or our setup, first a reliable reference sample is needed. Due to it’s simple chemical structure and corresponding Raman spectrum, a silicon

wafer (12x12mm) is chosen. Silicon has a sharp Raman peak at 520cm−1and a

weaker but wider peak around 1040 cm−1. The Raman spectrum of silicon can

be seen in figure 4.1.

With the Raman spectra of the silicon reference sample measured in the reference setup , we can compare and validate the Victorinox setup(the setup with the tuneable laser of 766nm to 781 nm ). This is done in first place by com-paring the measured peaks of a silicon reference sample to values documented in literature (see 4.1, secondly Raman peak position is compared to a Raman spectrum measured on a pre-existing and already calibrated setup called Punt-bron. Lastly, enhancement factors from SERS substrates are tested to see if the system also behaves linearly as a function of concentration like documented in the literature accompanying a SERS substrate.

In short, testing these three variables allows us to, independently, determine peak position and counts rate to see if the new setup performs in a reliable way.

(21)

Figure 4.1: Raman spectrum measured by Richter, Wang and Lay[18]. It can be

seen that Silicon has a clear peak at roughly 520 cm−1. The x axis is displayed

in cm−1 and the Y axis is presented in arbitrairy units. Richter et al also did a

simulation that matches the results measured. We can use this result to see if our setup measures the Raman signal at the correct shift.

(22)

4.1 Puntbron benchmark

As mentioned, first the Raman spectra of Silicon will be checked against the Raman spectrum measured with an existing setup called Puntbron. In this section, we will first describe the setup and subsequently we will compare the measurement results for both Victorinox and Puntbron setups.

4.1.1 Puntbron setup

The Puntbron setup is fitted with a 532 nm laser(Time Bandwidth Nd YAG+). Multiple filters are installed in the Puntbron setup, of which we used the LPD01-532Rs and the BLP01-532R-25 filters. The LPD01-LPD01-532Rs is a laser flat beam-splitter designed for a 532 nm laser and the BLP01-532R-25 is a longpass filter from 500 nm. This means that Raman spectra between 500 and 800 nm can be detected. Finally, the signal is measured using a Spectrometer(Action 2300i, equiped with a TE-cooled back-illuminated PIXIS100B ccd camera)

od filters

Figure 4.2: Graphical overview of the Puntbron setup. A 532 nm pulsed diode laser? is directed at the sample using the LPD01-532Rs beamsplitter. The reflected signal is filtered by the BLP01-532R-25. Next the optical setup allows to select either a camera or a spectrometer for detection.

4.1.2 Silicon reference measurement

First, a white light and a camera(Hamamatsu Orca 4) are used to focus the laser on the camera. Finally, the correct power setting out of a range from 1 mW up to 10 mW is selected by adjusting the OD filters in the filter wheel(see figure 4.2).

(23)

Next we measured the Raman spectrum using a 1s integration time and a power of 1 mW. An example of a measured spectrum can be seen in 4.7b. After this measurement the laser is blocked and a background measurement is done, again with 1s integration time. This background measurement allows for the subtraction of dark counts.

-1500 -1000 -500 0 -500 1000 1500 2000 2500 Raman shift (cm^-1) -1000 10000 8000 6000 4000 2000 12000 14000 0

Figure 4.3: Background corrected Raman spectra of Puntbron. Due to the long pass filter no signal is measured for wavelengths smaller than 530 nm. At 0

cm−1 a small peak for the excitation wavelength can be seen. At 520 cm−1 a

sharp peak can be seen which is a characteristic Raman peak for silicon. Finally

also a smaller peak around 1040 cm−1 can be seen which is also characteristic

for the Raman spectrum of silicon During this measurement, a pump power of 1 mW and a integration time of 1 second was used.

One can easily see that the location of the peaks measured with Puntbron setup match to the peak location found in literature, see 4.1 and figure 4.3. The number of counts between literature and the measurement in Puntbron cannot be compared since we do not know what power of setup where used

in literature. The steady increase from 500 cm−1 and higher might be due to

photo-luminescence of silicon [15].

For the spectrum measured with the Victorinox setup we can see that the

Stokes peak position of the 520 cm−1and the peak at 1040 cm−1again match the

spectra found with Puntbron and literature. As an added bonus, we can now

also see that the anti-Stokes peak, found at −520 cm−1 is also at the expected

(mirror image of the Stokes peak) position.

4.1.3 Conclusion comparing Victorinox to Puntbron

With these data set we can now compare the spectrum measured on both the Puntbron and Victorinox setup. As mentioned, the Raman spectrum of silicon measured with the Puntbron setup matches the Raman spectra of silicon found

(24)

Figure 4.4: Background corrected Raman spectra measured on the Victorinox

setup. Again a small peak can be seen at 0 cm−1 _{which again is the pump.}

At 520 cm−1 _{a sharp peak can be seen, which is, just like the peak around}

1040 cm−1, characteristic for the Raman spectra of silicon. Lastly, also notice

the anti Stokes peak at −520 cm−1 which can be measured due to the use of a

notch filter. During measurement a power of 1 mW and an integration time of 1 second was used.

compare these two spectra to the spectra measured with Victorinox. The Raman spectra measured with Victorinox is shown in figure 4.4. We can see that we

have measured a Stokes peak at 520cm−1and an anti-Stokes peak at −520cm−1.

These values match with the values we found using the Puntbron setup and values found in literature. Therefore, we conclude the setup is measuring the correct Raman shift.

4.2 SERS Ocean Optics benchmark

Next a comparison with a commercially available SERS substrate is made to see if the measured Raman signal scales like the producer of the SERS sub-strates documents. A SERS substrate (RAM-SERS-AG) from Ocean Optics is

chosen. This SERS substrate has an active area of 23.75mm2 _{consisting of}

sil-ver nanoparticles. Furthermore, its documentation comes with example Raman spectra for some common species. Out of these documented spectra, Rhodamine 6g(R6G) is chosen since it has nice Raman peaks and was already available at Amolf.

A dilution series was created by dissolving the R6G in acetone. The series

contains the following concentrations: 10−2M , 10−3M , 10−4M , 10−5M , 3.4 ×

10−6_{M , 1.1×10}−6_{M , and 3.8×10}−7_{M . These concentrations were chosen since}

they match literature provided by Ocean Optics.

In 4.6 the dilution series with a step size of a factor 10 in concentration is plotted. First note that for different concentrations, the position(shift) of the

(25)

1000 500 0 500 1000 1500 2000 cm^-1 0 250 500 750 1000 1250 1500 1750 Counts

ten times diluted

0.01 M 1 mM 0.1 mM 0.01 mM

Figure 4.6: Raman spectra for different concentration of Rhodamine 6G

peaks on the x-axis is constant and comparable with the graphs from Ocean Optics, again indicating that the setup is measuring the Raman spectra cor-rectly. Secondly, note the number of counts on the y axis. Documentation from Ocean optics tells us that the signal intensity should increase with a factor 3 as concentration increases with a factor 1000 for an integration time of 3s. We can also see this in the measured Raman spectra. These enhancement factors are in line with the literature provided by ocean optics. Both spectra report a linear response over concentration and a factor 3 increase in signal. However, since we do not know the absolute intensity or the count rate measured by Ocean optics, we cannot fully claim that we measure the same factor 3 increase in signal. Since we want to determine a possible enhancement it is important to know that the change in measured Raman intensity is reasonably reliable.

4.3 Collection efficiency

Based on literature, Rayleigh scattered light power should scale with 1

λ4. If the

collection efficiency of the Victorinox setup is calibrated correctly, we should be able to see this scaling factor in the measured Raman spectra for silicon as well.

Based on _ω14 and the excition wavelengths used (775 nm and 532 nm) we would

expect approx 4.7 times less signal. The sharp peak at 520 cm−1 for silicon is

used to compare the number of counts of both setup. For ease of comparison both spectra are again plotted in figure 5.4.

Looking at these spectra, We can see that the peak measured for a pump wavelength of 532 nm produces 13800 counts per second and the Raman peak excited with 775 nm produces 340 counts per second. However ideally we would need to integrate over the peaks to compare the area beneath the curve since

(26)

(a) Counts measured of a silicon

sam-ple with the Victorinox setup.

Mea-sured at 775 nm with 1 mW of laser power and 1 second of integration time. On the x axis the Raman shift is

pre-sented in cm−1, the number of counts

is presented on the y axis. For a larger figure, see figure 4.4

-1500 -1000 -500 0 -500 1000 1500 2000 2500 Raman shift (cm^-1) -1000 10000 8000 6000 4000 2000 12000 14000 0

(b) Counts measured of a silicon

sam-ple with the Puntbron setup.

Mea-sured at 532 nm with 1 mW of laser power and 1 second of integration time. On the x axis the Raman shift is

pre-sented in cm−1, the number of counts

is presented on the y axis. For a bigger scale, see figure 4.3

Figure 4.7: characteristics of the laser and cleanupfilter used in the experimental setup

spectrometer resolution and efficiency also have an effect. The stated count

rates indicate that the two setups do not scale with _λ14 since there is 40 times

less signal and not 4.7 times as expected. However, this was to be expected since it is difficult to achieve comparable collection efficiency for a setup which is coupled to the spectrometer via free space and a setup which is coupled to the spectrometer via a fiber. Furthermore, the setups both have different spectrometers which most likely also have a different efficiency and resolution. The fact that we measure higher signal intensity for a lower pump wavelength, can thus be explained by the scaling law and the collection efficiency which is optimal in the Puntbron setup and the Victorinox setup probably has room for improvement in efficiency.

(27)

Chapter 5

Silver nanocubes

As mentioned in the Introduction and Theory section, enhancement of the mea-sured Raman signal from even single molecules is already possible with the use of SERS. However SERS is not spatially selective and the measured signal is susceptible to being overwhelmed by background from sample contaminations. By using an antenna instead of a material/substrate we can be more spatially selective. The increase in selectivity is due to the fact that the antenna hotspot

mag be smaller than the wavelength2_{. Thus, the antenna can only detect a}

material that surrounds the antenna.

In this chapter we will explain what sort of antennas we used. We will quantify the resonance frequency of the antennas, determine their shape and dimensions and finally, we will try to measure the Raman spectra of a PVP layer attached to the antennas. Lastly, we will also determine how much laser power can be used on these antenna’s before we will start to deform them.

5.1 Characterization of silver nanocubes

For the antennas we have chosen to use silver nanocubes with dimensions of 75x75x75 nm, which where readily available at Amolf. They where kindly pro-vided to us by Eithan Oksenberg in the group of Erik Garnett. The dipolar res-onance frequency is determined by the dimensions of the cube which is around 570 nm. However, the geometry of the sample allows for the existence of a nanogap mode, of interest for our experiment as it enables strong hotspots[16, 19]. Indeed, the sample consists of a 330 micron thick Sapphire substrate, fol-lowed by a 50 nm thick gold layer and covered with a 3 nm thick Aluminum oxide spacer, onto which the nanocubes are dropcast. Simulations showed that a 3 nm thick spacer should allow for a resonance frequency in the range of the tunable laser that is installed in our setup (see chapter 3). A graphical representation of the sample can be seen in figure 5.1.

Finally, we need to place a Raman active species in the proximity of the cubes to obtaine a Raman signal. Fortunately, due to the fabrication process

(28)

Figure 5.1: Graphical representation of the sample. In blue the base layer of sapphire is represented, followed by aa 50 nm gold layer and topped of by a 3nm spacer layer made off aluminum oxide. On top the silver cube with sides of 75 nm can be seen in gray, covered by the PVP layer which is depicted in orange.

of the cubes, and to prevent them from oxidizing, our cubes are covered with a PVP ligand layer. The PVP layer is depicted as the orange layer in figure 5.1. Besides a broad Raman background (assigned to vibrations of the long carbon

chain of PVP), PVP has some well defined, sharp peaks around 1600 cm−1. As

a reference, the Raman spectra of PVP can be seen in figure 5.2.

Figure 5.2: Raman spectra of PVP. At 1600 cm−1a clear peak can be seen that

is easily detectable in the spectrum. Therefor we will focus on this peak during further measurements [9].

5.2 Experimental procedure

In order to determine a possible enhancement of the measured Raman signal, first a number of parameters such as laser intensity, integration time and res-onance frequency of the system needs to be defined. Secondly a reliable and repeatable measurement procedure needs to be determined. To start, the di-mensions and shape of the cubes need to be checked, to be sure that we are measuring on a cube with the correct properties. Next, the resonance wave-length of the gap mode needs to be determined. There is no point in measuring a cube if the resonance frequency is to far from the pump wavelengths(766-781 nm). These steps resulted in a set of cubes with the correct specifications. This

(29)

means that in theory, we should be able to measure the Raman spectra of the PVP layer attached to these cubes.

5.2.1 Determining the resonance frequency

There are two options available for us to determine the resonance of the nanocubes. A first option to determine the resonance frequency would be spectrally resolved darkfield microscope. In darkfield microscopy, the sample is illuminated by light that is not collected by the objective. The objective only captures scattered light. Therefore it creates a contrast between a dark background and bright spots caused by scattering objects on the surface of the sample. To prevent bright background scattering, the sample should be as flat as possible. Un-fortunately,in our case, the structure of the sample gave too much background signal.

This background signal originates from structures on the sample that also scatter light. Since these structures act as landmarks to navigate on, these structures cannot be removed. Furthermore, we suspect that in the darkfield, the light does not couple well to the gap mode, which makes it unable for us to accurately differentiate the gap mode from the cube resonance. The scatter-ing from the landmarks and the decrease in couplscatter-ing efficiency create to much ambiguity for us to use darkfield to define the gap mode resonance frequency.

By using bright field illumination on the sample one can also determine the gap mode resonance frequency. The bright field illumination is created with a supercontinuum laser. The supercontinuum laser has a spectral range of 500 nm up to 900 nm and is filtered by an acousto-optic monochromator, which has bandwidth of 1 nm. By measuring the reflected signal, the resonance frequency can be determined. When the gap mode or the cube mode is on resonance, the light is scattered in all directions (and partially absorbed as well) instead of being directly reflected. The scattering of light in all directions leads to a local minimum (a darkspot) in the reflected intensity. Since the cube dipole resonance has a different resonance frequency than the gap mode, scanning over frequency would show two dips in reflected intensity, corresponding to both resonances

Figure 5.3 shows the setup used for measuring the frequency of the reflected light whilst scanning the laser frequency. A lens collimates the beam, next the beamsplitter directs the beam toward the sample. A Nikon 0.95NA 100x objective focuses the light on the sample. The reflected light passes through the objective, beamsplitter, through a telescope and finally a camera takes a 2048x2048 pixels image. An example of the measured images can be seen in figure 5.4. In figure 5.4a we can see an image of the sample where neither the gap mode nor the cube resonance mode is excited. In figure 5.4b we can see a dark spot in the red circle. This darkspot was not present in 5.4a. This darkspot indicates that either the cube mode or the gap mode is on resonance. Finally, figure 5.4c does no longer show a darkspot at the position marked in figure 5.4b. Indicating that neither the cube mode nor the gap mode is resonating. Theory and simulations tell us that the cube resonance mode should be around 540 nm and that the gap mode resonance frequency should be around 780 nm. This

(30)

Figure 5.3: The Duimelijn setup used to determine the resonance frequency of the cubes

allows us to conclude that the darkspot found must be the gap mode resonance frequency. Scanning to smaller wavelengths indeed shows a second darkspot aprearing at the same location.

(31)

(a) Reflection measured at 600 nm. Neither gap or cube are resonating. Dark spots are probably dirt

(b) Reflection measured at 750nm. Dark spot in the red circle is a cube on it’s gap resonance. This dark spot is not visible in figure a and c.

(c) Reflection measured at 850 nm. Neither cube nor gap are resonating so the dark spot in figure b has disappeared.

Figure 5.4: Three main scattering processes involved in determining both res-onance frequencies. The Images show one specific cube, excited with different wavelengths ( 600 nm, 750 nm, 850nm). In figure a and c, no cube(darkspot) is visible in the area indicated with a red circle. In figure b a black dot can be seen, indicating that the cube is on resonance.

During measurement, one image of the sample is made per excitation wave-length. After the measurement, the obtained hyperspectral data of the cube is processed using Matlab(2016b): we define a region of interest containing the cube for each image. For this region of interest, the signal is integrated and

(32)

compared to the background (signal outside the ROI). The integrated value is then plotted as a function of wavelength. Finally all plots are compared to a lorentzian fit to get the resonance frequency and the linewidth.

During processing we marked all dark-spots identified to be a cube in a image for easy identification in a later process of measurement. Landmarks on the sample are used for identification. Per cube, the resonances and linewidth are presented in 5.1. Normal-ized in-tensity Wavelength (nm) 0 1 0.8 0.6 0.2 -0.2 0.4 700 750 800 850 900 650

(a) Resonance spectra of multiple cubes. Each color line represents a dif-ferent cube. There are two outliers out of the 9 graphs. These outliers might be caused by different thickness in the spacer layer. 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 650 700 750 800 850 900 Wavelength(nm) 1 - r eflec tion

(b) Resonance spectra of a sin-gle cube. The blueline shows the measured resonance spectra of the cube, the red line is a lorentzian fit that allow us to determine the resonance frequency and the linewidth.

Figure 5.5: Resonance frequencies of multiple cubes

In table 5.1 the resonance frequency and the linewidth for all investigated cubes is presented. We can see that some cubes have a resonance frequency that deviates more than the standard deviation of 35.6 nm from the average of 767 nm. These differences might be caused by differences in the thickness of the spacer layer (see figure 5.1). This was investigated in SEM and it could be seen that these cubes where indeed located on a surface that looks different from the normal surface of the sample. Further investigation of these thickness might be useful.

(33)

Cube Resonance frequency(nm) Linewidth(nm) Measurement date A51.1 782 64 2019-05-27 A51.2 768 67 2019-05-27 A51.3 785 82 2019-05-27 A51.4 735 82 2019-05-27 A81.5 773 72 2019-05-28 A81.6 815 46 2019-05-28 A81.7 768 79 2019-05-28 A71.8 788 49 2019-05-28 A71.9 778 104 2019-05-28 A42.2 813 167 2019-07-03 A42.3 725 151 2019-07-03 A42.4 785 169 2019-07-03 A63.1 699 314 2019-07-03 A63.5 677 98 2019-07-03 A63.6 762 162 2019-07-03 A72.3 780 91 2019-07-03 A72.4 799 91 2019-07-03 A73.3 780 949 2019-07-03 A73.4 762 263 2019-07-03

Table 5.1: Resonance frequency,linewidth and date of measurement for all cubes tested. Differences in frequency might be caused by differences in spacer layer thickness (see 5.1) These variations in spacer layer thickness are caused by the evaporation of gold, and are than also imprinted in the Sapphire spacer layer. The increase in linewidth might be due to degeneration of the cubes over time. The average resonance frequency is 767 nm with a standard deviation of 35.6 nm

5.2.2 SEM

To determine the shape and dimension of the cubes, we used a FEI Helios

Scanning Electron Microscope (SEM). By angling the camera to an angle of 50◦

(34)

Figure 5.6: SEM image of a cube, roughly 500.000 times magnified and tilted

in an angle of 45◦ to see all sides of the cube. The rough background is the

sapphire spacer layer as shown in figure 5.1

It is well known, that SEM also deposits a small layer of carbon on the sample [5]. This Carbon layer negatively effects the resonance(and possibly even the quality factor and field enhancement) of the cubes. The deposition of this carbon layer increases when the image is zoomed in since the amount of deposition is constant but the area of deposition decreases. Therefore, after trial and error, we determined a maximum scale of 1:100.000 as a baseline to prevent too much carbon deposition.

By combining SEM and Duimelijn measurements, we can reliably distinguish dirt or misshaped cubes. Duimelijn and SEM measurement are complementary in the process of identifying cubes. First using Duimelijn allow us to identify potential cubes without spoiling them with carbon deposition. After Raman measurements, the dimensions of the cubes used can still be checked in SEM.

5.2.3 Signature of Raman spectrum measured on the cubes

After determining the resonance frequencies of the cubes, we need to check if we can measure the Raman signal of the cubes. As a reference, the Raman spectrum of PVP can be seen in figure 6.1. Since the Raman signal is feeble, and linear with pump power we will use the maximal laser power of the setup, or at least the maximum power given the thermal damage threshold in section 5.2.4, to see if we can measure the Raman signal of the PVP layer surrounding the cubes. To be sure we did focus the laser on the cube we scanned the laser over an area of 1166 x 550 nm in 11 by 11 steps. The measured results can be seen in figure 5.7

(35)

Figure 5.7: Raman spectra of PVP measured on a cube. We see 5 different spectra measured directly after eachother with an integration time of 2 seconds. The small differences in the spectra is caused probalby caused by an increase in temperature of the cube. The red dotted lines indicates the integration range used for the 2D plot of the Stokes peak. The blue dotted lines indicate an integration range of equal size but then for the symmetric anti-Stokes peak. The Stokes peak has around 1700 counts and the anti Stokes peak 700.

Next, we integrate the signal for Stokes and anti-Stokes peaks in an area indicated with the dotted lines as shown in figure 5.7. We also integrated the

signal of the Rayleigh scattering around 0 cm−1. As mentioned before, only

when the laser is focused on a cube with PVP attached, we expect to see signal in these 2D plots. Rayleigh scattering, i.e. the integrated signal for the Rayleigh scattering should not increase in signal at the position of the cubes (actually we even expected a decrease of the reflected signal).

(36)

S -1 -0.5 0 0.5 1 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 1 2 3 4 5 6 104

(a) 5.5 mW of excitation power is used to plot the intensity of the Stokes Ra-man line. L -1 -0.5 0 0.5 1 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 1000 1050 1100 1150 1200

(b) 1.6 mW of excitation power is used to plot the intensity of the Rayleigh scattering. AS -1 -0.5 0 0.5 1 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 11000

(c) 1.6 mW of excitation power is used to plot the intensity of the anti Stokes Raman line.

Figure 5.8: 2D plots for respectively the Stokes peak, the laser and the Anti Stokes peak. When the laser is scanned over a cube, an increase in signal is measured.

The 2D plots indicate that the Stokes and anti Stokes line are intense only when the excitation and detection spot overlap with the cube, suggestive of SERS. Meaning we can indeed measure the Raman spectra of the PVP layer

that is attached to the cubes. The fact that the measured signal intensity

surrounding the cubes is almost zero, indicates that the background has no Raman signal. The lack of Raman signal indicates there is no cube with PVP attached to it present at that position. We indeed observe not only Stokes

signal for the PVP at 1600 cm−1 and 1450 cm−1 but also the corresponding

anti-Stokes at respectively -1600 cm−1 and -1450cm−1

As we mentioned in the theory section [13], the ratio between Stokes and anti-Stokes depends on the temperature. This was observed experimentally in measurements, as can be seen in Figure5.7. we used the maximum power of our

setup(1.5 ∗ 102GW/m2) focused on the cubes. We can see a clear Stokes peak

at -1600cm−1. At room temperature, anti-Stokes peaks should be at least 100

(37)

5.7 we can also determine the temperature. We inverted the anti Stokes side of the spectra to make it overlap with the Stokes side and determined the ratio in counts of Stokes to anti Stokes. Next we plotted this ratio against equation 2.7(see appendix), in order to get the linear part of the slope. Since the slot is

equal to _T1 we can now determine the temperature. This puts the temperature

of the cubes in the order of 1400K, which is higher than the melting point of gold and silver.

5.2.4 Heating of the cubes by laser

The response of a resonator is not always linear due to material properties. In particular, an increase in temperature can lead to a phase change. This phase change influences the material response and even the shape of the cube or the gap. Thus, directly influencing the resonance frequency. As we mentioned, a quick estimation of the Raman spectra presented in 5.7 indicates a temperature in the order of 1400K, which is higher than the melting temperature of gold an silver.

This temperature estimate, even if it should be taken with a grain of salt, suggest that during this measurements, cubes are very likely to have changed shape. Additionally, we repeated the measurements 5 times in a row, and we can see in Figure 5.9 that the signal is unstable, presumably because of shape fluctuation of the cubes or PVP layer. Figure 5.10d confirmed this hypothe-sis after measurements.After measurement, this cubes was imaged using SEM. Figure 5.10 shows the SEM images of the cube used for these measurements.

Therefore, we needed to figure out how much laser power would damage the cubes, i.e. irreversibly change the response. To check this, laser power

was varied from low power 1.4 ∗ 101 GW/m2 to high power 1.5 ∗ 102 GW/m2.

A different cube was selected per step in laser power, and we measured the Raman spectra for every power setting. During the measurements, the signal from the cubes was measured over time, stepping the beam on and off of the cube. This allows us to determine any possible differences in the signal intensity of the Raman spectra over time caused by the heating or even destruction of the silver cubes. Since we want to see if these cubes enhance the Raman signal of the attached PVP layer, measurements need to be reliable and repeatable over time. This means the cubes can not change shape or dimensions during measurements since this would effect the resonance frequencies.

In 5.9 we show two intensity plots of Raman spectra. In figure 5.9a 1.3 ∗

102 _GW/m2 _{of power was used, and it can be seen that the signal intensity}

decreases over time. This is most likely due to the heating of the cube. Once the cube reaches it’s melting point it will start to deform, thus changing the resonance frequency. The change in resonance frequency will cause a decrease in signal. In figure b, the intensity plot of the Raman spectra excited with

3.8 ∗ 101 GW/m2can be seen. here the signal intensity stays constant overtime,

suggesting that the cube retains it’s shape and dimensions. This means the resonance frequency is not affected and the signal intensity remains equal.

(38)

650 700 750 800 850 900 Wavelength (nm) Time (ID ) Time (ID ) 5 15 20 10 OD 0.13 (a) 1.3 ∗ 105 M W/cm2 of excitation

power is used to plot the intensity of the Raman spectra over time.

Time (ID ) 5 15 20 10 650 700 750 800 850 900 Wavelength (nm) OD 1.57 (b) 3.8 ∗ 101 GW/m2 excitation power

is used to plot the intensity of the Ra-man spectra over time.

Figure 5.9: Signal intensity of the Raman spectra measured for an excitation wavelength of 775 nm. On the y axis the time is plotted and the the color bar shows the measured intensity. In figure a, the PVP Raman spectrum obtained

with a cube excited with 1.3 ∗ 102 _GW/m2 _{is plotted and it can be seen that}

after some time there is no more signal detected. For lower excitation power

(3.8 ∗ 101 _GW/m2 _{) the signal remains constant over time. The dark band in}

the range 766 nm to 790 nm in the middle is the notch filter

determine possible shape changes. Below a power of 2.4 ∗ 101_GW/m2_{, changes}

in shape appear to be negligible. In Fig. 5.10 the SEM images are shown as a function of laser power. We see that for low power (figures a and b), the shape and dimensions of the cube stay unchanged. For higher excitation power(figures c and d) the shape of the cubes is changed dramatically, explaining the decrease of measured Raman signal. Finally for reference, in figure e, a cube is shown that is not used for any measurement, meaning it has it’s original shape.

Therefore, we can conclude that during experiments, laser power should be

smaller or equal to 2.4 ∗ 101 _GW/m2_{. For consistency, from now on we will}

use 2.4 ∗ 101 _GW/m2 _{or 1 mW of power during experiments. Furthermore,}

we are now sure that we can indeed measure the Raman spectra of PVP layer attached to the cubes. This allows us to scan the pump wavelength over the resonance frequency of the gap mode, which might lead to which might lead to variations in Raman signal in dependence of the matching of both pump and Raman-shifted frequency to the plasmon resonance.

(39)

(a) SEM image of the cube shape after exposure with od 0.6

(b) SEM image of the cube shape after exposure with od 0.3

(c) SEM image of the cube shape after exposure with od 0.1

(d) SEM image of the cube shape

after exposure with od 0 (e) SEM image of the cube without_{exposure to the laser}

Figure 5.10: Shape change ea melting of the cubes as a function of exposure power. The lower the OD filter value the more the cubes are deformed. Figure e shows an unused cube as reference.

(40)

Chapter 6

Enhanced Raman

spectroscopy

Now we have the key properties of the silver nano cubes determined, we can measure if the Raman spectrum is indeed enhanced. As we mentioned in the theory section, ideally we would need the centre wavelength(λ) of the resonance

to be 1₂∗ ∆ν. In this case, both the pump wavelength and the Scattered light

are maximally enhanced.

The scanning range of our setup of 766 mm up to 780 nm limits the cubes we can use for the measurements. Again a Raman spectra of PVP is plotted in 6.1.

We can see that the Raman spectra of PVP has a clear peak near 1600cm−1.

This peak can easily be detected and thus, during the experiment we will focus

on the peak at 1600cm−1.

.

Figure 6.1: As a reminder, again the Raman spectra of PVP from literature, also plotted in figure 5.2 [9]

(41)

As we mentioned, PVP has a strong Raman peak around 1600 cm−1. We can convert this Raman shift to nm scale, using a pump wavelength of 775nm (average of the range used) and formula 2.5. This means the shift of the Raman

peak at 1600 cm−1 or 882 nm, corresponds to a shift of roughly 100 nm. In

an ideal situation, we want to use a cube with a gap mode resonance frequency of 825 nm. Since we don’t have full control over the exact resonance frequency of the cubes, we use the cubes with which are one resonance with the laser pump. This should still allow us to see a change in measured signal intensity as a function of pump wavelength.

For the measurement of the Raman spectra, all the cubes that have a res-onance frequency in the right domain were tested. It proved difficult to focus the laser beam correctly onto the cubes with the procedure now followed, which

meant that not all cubes showed a clear PVP Raman peak at 1600 cm−1.

Figure 6.2 shows the Raman spectra measured for two cubes named A and B. These cubes have a resonance frequency at respectively 768 nm and 780 nm. The resonance curve is also plotted in the figure. During measurement, they where excited by a laser from 770 nm to 780 nm in steps of 2 nm. Figure 6.2 shows the measured Raman signal for both cubes.

(a) cube A with a gap resonance at 768nm. excitation wavelength are plot-ted from 770 nm up to 778 nm. A num-ber of Raman peaks can clearly be seen for all wavelengths at the same posi-tion.

(b) Raman spectra of PVP on cube B. It can be seen that the resonance of the cube is at 780 nm.

Figure 6.2: Raman spectra of PVP on silver nano cubes showing enhancement as the pump wavelength is scanned towards the resonating frequency

Figure 6.2a shows the Raman spectra of PVP surrounding cube A. Notice that the signal intensity increases as the pump wavelength approaches the reso-nance frequency of the cube. Figure 6.2b shows the Raman spectra measured on cube B. Again we should notice that the measured signal intensity increases as the pump wavelength approaches the gap resonance frequency of the cube(780 nm). The general broad shoulder is most likely due to the vibration of the long carbon chain of PVP with added sharp Raman peaks of the PVP.

However, the general shape of the Raman spectra have also been reported before by A.Carattino[6]. He investigated gold nanorods without a metalfilm

(42)

and claims the general shape and decay in both the Stokes and anti-Stokes scattering is due to luminescence of gold. He claims the luminescence is caused by the recombination of excited electrons. These excited electrons leave holes which can be excited by the pump. The interaction of the excited electrons

with the phonon bath can lead to anti-Stokes scattering. The difference in

counts between the Stokes and anti Stokes peaks is therefore a measure of the temperature of the material.

We can also see this characteristic tail in the measured spectra and also the difference in counts between the Stokes and anti Stokes scattering. This means that the broad signal measured could be due to luminescence. With added small peaks due to the Raman spectra of the PVP layer. However, the long carbon chain of PVP might also have an effect on the general shape of the Raman spectra recorded. One can argue that Carattino might also have a broad background due to a ligand attached to his nanorods. Therefore, more research is needed to confirm if it is either the long carbon chain of the ligand or indeed luminescence.

As we explained, the Raman scattered light is maximally enhanced if the

pump wavelength is ∆ν −1₂ × ∆λl. In figure 6.2 we see that the blue curve

(770nm) is closest to half of the Raman shift and therefore maximally enhanced. Increasing the pump wavelength towards 780 nm leads to a decrease in signal since the Raman shift is enhanced less efficiently.

(a) Raman spectra of PVP combined with the resonance spectra for cube A. The gap mode has a resonance fre-quency of 768 nm

(b) Raman spectra and resonance spec-tra of PVP combined with the res-onance spectra for cube B. The gap mode has a resonance frequency of 780 nm

Figure 6.3: Raman spectra of PVP on silver nano cubes showing enhancement as the pump wavelength is scanned towards the resonating frequency

In figure 6.3 the resonance curves for both cubes are plotted along with the Raman spectra. This makes it easier to see the increase in signal intensity as the pump wavelength approaches the resonance frequency of the respective cube.

For cube A we can see that the resonance peak has a smaller bandwidth defined than cube B. This leads to a better defined Raman peak that can be measured. We can see this at 800 nm. Comparing this to cube B we can see that the broad resonance frequency leads to a broader enhancement of the

(43)

measured Raman spectrum ea peaks are less defined and also the spectra in between the peaks in enhanced, thus lowering the contrast. This means that a small bandwidth or q factor of the cubes is indeed important.

For both cubes we can see a decrease in signal once the pump wavelength has past the resonance frequency of the gap mode. Meaning that their is indeed a relationship between the intensity of the measured Raman spectra and the pump wavelength. Thus the cubes do indeed increase the signal intensity of the Raman spectra.

(44)

Chapter 7

Conclusion and Outlook

With the promising creation of photonic resonators that can be tailored in their spectral response, new possibilities will arise for sensing and detection of molecules. By creating resonators with narrow linewidth, a system can be designed that can selectively enhance a certain range of the Raman spectrum.

In this report we have characterized a setup for the detection of Raman scattering using resonators that operate near 780 nm bandwidth. A unique feature of this setup is the fact that it is outfitted with a tuneable laser allowing us to relate the pump- and Raman enhancements to the features of photonic structures in a SERS scenario. Using this setup we observed Raman signals from ligands attached to individual cubes, and find that their intensity depends on how well the pump and Raman line are tuned to the cube resonance. The spectra further show a broad Stokes and anti-Stokes background. This might either be due to a broad Raman response of the long carbon chains of the ligand, or alternatively might be due to fluorescence by the metal nanoparticle itself, as reported by Carattino.

Finally we identify further improvements for the measurement procedure and choice of resonator. Since these possibility’s might allow for improved detection of molecules with higher q factors. However further improvements are possible in the enhancement of Raman spectroscopy. For example, The measured signal in Raman spectroscopy might benefit from a better procedure, a different polymer attached to the silver cubes and possibly even the aid of a hybrid structure combined with the silver cubes.

7.1 Placing a different ligand or polymer

In this study we investigated the enhancement of Raman scattering by the nano gap mode of silver nanocubes deposited on a gold layer. During fabrication, these cubes are covered in a PVP layer to prevent oxidation. Since PVP is Raman active and already attached to the cubes, we could use it’s signature as a testbed for our measurements. However, the Raman spectrum of PVP

Tailoring photonic response for enhanced Raman spectroscopy