Probing hotspots of plasmon-enhanced Raman scattering by nanomanipulation of carbon nanotubes

We present a two-step procedure to probe hotspots of plasmon-enhanced Raman scattering with carbon nanotubes. Dielectrophoretic deposition places a small carbon nanotube bundle on top of plasmonic Au nanodimer. After 'pre-characterising' both the nanotubes and dimer structure, we subsequently use the tip of an AFM to push the bundle into the plasmonic hotspot located in the 25nm wide dimer gap, characterize its location inside the gap, and observe the onset of plasmon-enhanced Raman scattering. Evidence for the activation of the carbon nanotube's double-resonant D-mode by the near-field of the plasmonic hotspot is discussed.


Introduction
Metallic nanostructures enable the localization of electromagnetic waves into nanoscale volumes far smaller than the wavelength of light at optical frequencies [1,2]. The electromagnetic wave couples to the collective excitation of electrons in the metal nanostructure called localized surface plasmon resonance (LSPR). This leads to a strong enhancement of the local near-field at the metal surface. When two nanostructures are closely spaced, coupling of the confined LSPRs (at certain polarizations) leads to a collective excitation across both, generating particularly strong field enhancement in the gap between them. Such a region is termed plasmonic hotspot. The field enhancement varies considerably inside and around such a plasmonic hotspot and may differ by several order of magnitude over tens of nanometers [3,1,2,4].
One of the most spectacular applications of LSPRs is plasmon-enhanced Raman scattering (PERS), where the incident light and that inelastically scattered from an object inside a plasmonic hotspot is greatly enhanced [5,6].
PERS has enabled the detection of single molecules and offers great potential for studying the fundamental aspects of light-matter interaction as well as biological and chemical sensing applications. [7,8,9,10,11,12,13,14,15,16,17] A recurring problem in quantifying PERS are small geometric variations between plasmonic structures of the same design, which may strongly alter the enhancement experienced by a Raman scatterer that is nominally located at the same position inside a corresponding plasmonic hotspot [18,19]. It is therefore desirable to probe the PERS enhancement at different locations in and around one plasmonic hotspot using the same Raman scatterer. Kusch et al. recently suggested to read out the PERS enhancement at a plasmonic hotspot via the Raman signal of a sharp Si-tip of a scanning nearfield optical microscope [20]. The spatial extension of the tip, however, masks subtle variations of the enhancement on the nanometer scale. Similar restrictions apply to molecules, the traditional probe for PERS, because their exact location and orientation inside a plasmonic hotspot are impossible to control and, more importantly, cannot be altered. A nanoscale Raman scatterer that probes the enhancement in and around a plasmonic hotspot is therefore highly desirable.
Carbon nanotubes (CNTs) overcome some of these limitations and have recently emerged as a promising alternative to probe and quantify plasmon-enhanced Raman scattering on the nanoscale as their one-dimensional nature allows us to experimentally obtain the exact location and orientation within a plasmonic hotspot [21,22,23,24,25,26,27,28]. Interfacing CNTs with rationally designed hotspots, on the other hand, remains a challenge. We recently demonstrated the benefit of using directed dielectrophoretic assembly (DEP) to place carbon nanotubes precisely into the plasmonic hotspot at the ∼ 25 nm gap of gold nanodimers, Fig. 1(a), and observed a 10 3 − 10 4 enhancement of the Raman intensity [24,25]. More importantly, CNTs are perhaps the only nanoscale object that is able to probe PERS enhancement at different locations in a plasmonic hotspot, as the location and orientation of a nanotube can be purposefully alterted by nanomanipulation. [29,30] Here we demonstrate a two-step scheme that combines dielectrophoretic deposition of carbon nanotubes with tip-based nanomanipulation to probe plasmon-enhanced Raman scattering outside and inside the plasmonic hotspot of a gold nanodisc dimer. AC voltage drives the CNTs in solution to deposit between the electrodes, where the plasmonic dimers are located. Based on Ref. [24], DEP has a < 5% yield for nanotubes passing through the gap, as in (c), and ∼ 25% for nanotubes crossing one of the discs forming the dimer in (d).
spatial localization of the Raman signal after nano-manipulation. Beyond quantifying plasmon-enhanced Raman scattering, the proposed scheme will enable the realization of nanotube-nanoplasmonic experimental systems that were previously not accessable.

Experimental
The sample fabrication and dielectrophoretic deposition process follows the procedures described in Refs. [24] and [25]. In short, arrays of dimers and electrodes for DEP were fabricated on a SiO 2 (90 nm) on Si substrate by electron beam lithgraphy using a LEO Witec alpha300 (λ = 633 nm) single grating spectrometers using 100× objectives (NA 0.9), and piezo stages for spatial mapping. To measure the polarization, we rotated the sample by 90 • . We used integration times up to 10 s and laser powers between 1 mW (532 nm) and 50 µW (633 nm). The luminescence background from the Au nanostructures was subtracted for all Raman spectra.

Results and Discussion
We fabricated arrays of dimer structures consisting of two closely spaced nanodiscs by electron beam lithography on top of a SiO 2 /Si substrate (see Methods). The dimer's dimensions (cylinders with diameter ∼ 100 nm, height 45 nm, separated by a gap of ∼ 25 nm) were chosen to provide resonant enhancement for λ = 633 nm excitation polarized along the dimer axis [33,24,34]. The strongest coupling to the high-intensity near-field occurs for a nanotube placed inside the plasmonic hotspot at the nanoscale gap of the dimer [1,2]. We realized such an interface for plasmon-enhanced Raman scattering of CNTs using directed dielectrophoretic deposition from solution as described in detail in Refs. [24] and [25]. Dielectrophoretic forces drive the nanotube to deposit between electrodes where we have placed the plasmonic dimers, c.f. Fig. 1(b). Ideally the CNT connects the electrode tips in a straight line and passes through the gap as shown in procedure proposed in this work, because it allows us to characterize the CNT and its Raman signatures before it is interfaced with the plasmonic hotspot at the dimer gap.
We show an AFM phase image of such an initial scenario in Fig. 2(a). A small CNTbundle (height ∼ 7 nm, labelled CNT-B) crosses the left disc of a plasmonic dimer. The phase image is overlaid with a spatial map of the nanotube G-band just below 1600 cm −1 , the most prominent Raman feature in CNTs [35]. The corresponding Raman spectrum, taken with the laser centered on the dimer, is shown in Fig. 2(b). The D-mode at ∼ 1350 cm −1 indicates the presence of defects. It also appears in Raman spectra of CNTs dropcasted from the as purchased solution on a bare SiO 2 /Si substrate and is therefore not caused by DEP [24,25]. Both the energy of the excitation (λ = 532 nm) and its polarization P Y (perpendicular to the dimer) prevent plasmonic enhancement from the dominant dipolar LSPR of the dimer.
The spatial distribution of the integrated G-mode intensity I G (P Y ) in Fig. 2(a) matches that of CNT-B. We attribute the increased intensity around the dimer disc to a maximized overlap between CNT-B and the laser spot in combination with minor plasmonic enhancement from the disc's LSPR or due to electric field line crowding at the disc edges [1]. We did not observe a considerable signal for P X , as the strong antenna effect in CNTs suppresses Raman scattering for polarizations perpendicular to the tube axis [35,36]. For an excitation wavelength that overlaps with the dimer resonance (λ = 633 nm), no Raman signal within our detection threshold was observed for both P X and P Y . Before we discuss the nano-manipulation of CNT-B, we investigate more closely the topography of CNT-B and its interface with the dimer structure. A high-resolution AFM topography image is shown in Fig. 3(a). It confirms that CNT-B crosses the left nanodisc. Height profiles parallel to the dimer axis are shown in Fig. 3(b). They connect the dots of corresponding colour in Fig. 3 Fig. 3(d). We achieved the same resolution as in Fig. 3(a) prior to the nanomanipulation -the observed profile of CNTB is the real profile convoluted with the tip apex profile -which shows that the tip did not measureably blunt while in contact with the surface.
The topography in Fig. 3(d) reveals that CNT-B was moved to the right such that it passes through the dimer gap. We will now deduce the exact configuration of the nanotube in the gap by comparing the corresponding height profiles in Fig. 3(e) to those take before moving CNT-B in Fig. 3 The observation of plasmon-enhanced Raman scattering from CNT-B after nanomanipulation confirms that we have successfully interfaced CNT-B with the plasmonic hotspot in the dimer gap. Figure 4(a) shows Raman spectra of CNT-B measured on the dimer with λ = 633 nm for both P X (red) and P Y (blue). The Raman intensity for P X dominates the spectrum, and is a clear sign of enhancement due to near-field from the dipolar LSPR of the dimer structure. The signal is much stronger than for the non-resonant case (λ = 532 nm), where we did not observe any signal for P X . It is also stronger than for P Y , along the CNT-axis, compare Fig. 2(b). We have previously identified this inverted polarization behaviour as a clear sign of plasmonic enhancement from CNTs in dimer hotspots [24,25].
The spatial distribution of the G-mode signal for P X is shown in Fig. 4(b). It is strongly localized in space and shows the point-like character of the Raman enhancement arising from hotspot the dimer gap. Compared to P X , the relative magnitude of the enhanced signal for P Y and its localization, Fig. 4(c), may seem surprising. It originates from minor plasmonic enhancement that couples favourably to CNT-B because the near field is polarized parallel to the axes of the tubes forming CNT-B. This is in stark contrast to the Raman intensity for P X that scales only with the projection of the near field polarization on the tube axis [33,24], explaining the relative intensities in Fig. 4(c). From the inverted polarization behaviour and signal localization presented in Figs. 4(b) and (c), we estimate 2.7 × 10 2 as the lower limit of the enhancement factor of the G-mode, details see Ref. [24].
The D-mode is the dominant feature in our PERS spectra, Fig. 4(b). The ratio I D /I G is typically regarded as a measure for the defect concentration in nanoscale graphitic material [35]. For CNT-B, it increased from 0.43 (532 nm, P Y ) before the nanomanipulation to 0.61 (633 nm, P Y ) and 1.22 (633 nm, P X ) in the presence of enhancement. For bulk quantities of our CNT starting material and for small CNTs bundles deposited directly into the dimer gap by DEP, the I D /I G ratio increases upon changing the excitation wavelength from 532 nm to 633 nm, see Ref. [24]. Therefore, the increase in I D /I G for P Y is intrinsic to the CNTs used in this work and not caused by nanomanipulation. Did moving CNT-B, however, induce structural defects at tube segments now located at the plasmonic hotspot, which could explain the increase in I D /I G from 0.61 (633 nm, P Y ) to 1.22 (633 nm, P X )? Previous studies showed that moving CNTs by AFM does not damage the nanotubes [37,38]. Yano and co-workers moved carbon nanotubes with an AFM tip and subsequently characterized them by tip-enhanced Raman spectroscopy [39]. They did not observe any defects at the manipulated tube segments while probing with a spatial resolution of 20 nm, comparable to the localized nature of the enhancement in this work. As we were moving a bundle of CNTs, any stretching of the bundle due to nanomanipulation can be relaxed by interfacial sliding between tubes in the bundle which occurs at a much lower force than the introduction of structural defects within the nanotube [40]. We further exclude radiation-induced damage because we used low powers (50 µW) when acquiring the PERS spectra at λ = 633 nm.
In the light of these studies, we argue that the increase in the relative D-mode intensity in our PERS spectra is not caused by structural defects. Instead, it is a consequence of the strongly localized nature of the near-field in the plasmonic hotspot.
A recent study observed a strong D-mode for defect-free graphene interfaced with a single plasmonic hotspot [41,34]. It was suggested that the confinement of the light fields in space provides the necessary momentum to excite non-vertical optical transitions. This activates the double-resonant Raman process that gives rise to the D-mode in graphene without requiring a real defect for momentum conservation [42,43]. As the D-mode in carbon nanotubes has the same origin as in graphene [35,36], we argue that the strong D-mode in the PERS spectra of CNT-B also arises from the localized near-field at the hotspot. This interpretation explains the factor two difference in the experimentally observed I D /I G ratios for P Y and P X . The near-field is less localized for P Y than it is for P X , resulting in a smaller -or more likely negligible -plasmon-induced contribution to the D-mode.
The two-step approach combining DEP and nanomanipulation presented here will find use beyond probing PERS from plasmonic dimers. It will allow to place carbon nanotubes into truly nanoscale gaps (≤ 5 nm) of bowtie antennas or plasmonic nanoclusters, which provide the electric field strengths necessary to observe phenomena such as the field gradient effect in PERS [44,45,46]. The low yield of placing CNT directly in such narrow gaps by DEP makes this technique impractical. Using DEP to carry the nanotubes close to the hotspots followed by nanomanipulation as described here, however, will achieve the desired interface.

Conclusions
In conclusion, we suggest directed dielectrophoretic deposition of carbon nanotubes followed by tip-based nanomanipulation to probe plasmon-enhanced Raman scattering from nanoscale plasmonic hotspots. With the tip of an AFM, we pushed a small carbon nanotube bundle 25 nm into the gap of an Au nanodimer. Its location at the hotspot was confirmed directly by AFM and indirectly by plasmon-enhanced Raman scattering.
We observed a 10 2 enhancement of the G-mode that was strongly localized in space in combination with an inverted polarization behaviour. Given the ability to characterize the nanotube beforehand, the strong D-mode after nanomanipulation was interpreted as a signature of plasmonic enhancement rather than being caused by structural defects while moving the nanotubes. Our two-fold scheme will allow the reliable placement of CNTs into nanoscale gaps of plasmonic structures and thereby enable experiments previsouly not accessable.