1. Introduction

Current research into surface enhanced Raman scattering (SERS) is increasingly focused on establishing intricate relationships between field enhancement (FE) effects at excitation and emission frequencies of radiation, its polarization properties and effective SERS enhancement factors [1]. Realization of strong FE effects driving efficient SERS is extremely important for highly sensitive detection of low molecular concentrations [2], ultimately reaching the limit of single molecule detection [3]. In this connection, metal nanostructures supporting localized and/or propagating surface plasmon (SP) resonances have been intensively investigated with respect to achieving strong FE effects [4]. An important research direction is thereby the design of metal nanostructures ensuring well-pronounced SP resonances within a given wavelength range along with substantial and robust FE effects. Several extensive FE-related studies have been performed dealing with various shapes and configurations of metal nanostructures ranging from individual pointed particles [5,6] to their pairs [7–9] and periodic [10] and random [11,12] ensembles.

We have previously used signal levels of two-photon excited photoluminescence in scanning optical microscopy for the evaluation of electromagnetic FE factors achieved with various types of samples, such as individual metal nanostrips [13], periodic metal nanoparticles [10,14], and fractal shaped metal nanostructures [15,16]. The main goal in these experiments was to realize the SP resonances in a predesigned wavelength range and characterize local FE factors aiming at obtaining strong and robust FE effects that are of great importance for surface-enhanced spectroscopies, especially for SERS. Very recently, we have performed the first (to our knowledge) point-by-point comparison of SERS spectra from locations separated by submicron distances within periodic gold nanoparticle arrays [17].

In this paper we undertake the spatially-resolved SERS characterization of periodic arrays of and individual rectangular gold nanoparticles. We employ high-spatial resolution Raman imaging using the well-known Raman spectrum from Rhodamine 6G (R6G) dye molecules adsorbed at the nanostructures and acting as a probe for the local electric FE at each nanoparticle. We compare reflection spectra and SERS enhancements observed in the Raman images for five different particle sizes and two perpendicular excitation polarizations, elucidating the SERS spectral dependence on the locally-enhanced electric field E at each nanoparticle. The description of this relationship is nontrivial and should be extended beyond the usual so-called |E|⁴ approximation [18], as pointed recently out on the basis of conventional SERS spectroscopy dealing only with collective Raman spectra averaged by simultaneous illumination of larger areas with periodic nanoparticle arrays [1].

Another important issue targeted in this work is related to the interplay between collective and individual (localized) SP resonances under the conditions of strongly focused illumination (to a submicron-sized spot) and confocal detection that are essential for obtaining high-resolution Raman images. This issue is extremely important since periodic SERS substrates are usually designed for the plane-wave excitation, so that collective (resonant) SP modes can be very efficiently excited, whereas the usage of strongly focused radiation result in a broad spatial-frequency spectrum, reducing thereby significantly the excitation efficiency and, consequently, lowering FE effects [17]. Here we use scanning confocal Raman microscopy with a high-numerical aperture objective (so that the spatial resolution of ~0.35 µm is achieved at the illumination wavelength of 0.53 µm) for direct comparison of local SERS factors obtained with individual and periodic arrays of identical nanoparticles. The overall results presented in this paper emphasize the importance and quantify the influence of particle dimensions, polarized excitation, collective resonances and SERS locations.

2. Fabrication and linear reflection spectroscopy

The sample consist of both individual and two-dimensional 740-nm-period arrays of rectangular 50-nm-high gold nanoparticles of 5 different lateral dimensions (fixed width d_x=250 nm, lengths d_y=250, 340, 440, 540, and 630 nm) fabricated by electron-beam lithography (EBL), sputter coating and liftoff, on top of a 200-nm-thick gold film sputter-coated onto a silicon substrate and was imaged by scanning electron microscopy (SEM) (Fig. 1). Previous investigations of FE effects showed the importance of both particle dimensions and array periods, and samples with similar particle height, width and period as described above exhibited resonances around the desired 530-nm wavelength [10].

 

Fig. 1. SEM images of arrays with 50-nm-high nanoparticles having lateral dimensions (a) 250×250 nm, (b) 250×440 nm, and (c) 250×630 nm at a fixed array period (Λ_x=Λ_y≈740 nm) on top of a 200-nm-thick gold film.

Download Full Size | PDF

We use reflection spectroscopy [10] to map the SP resonances before covering the sample with R6G and exploiting selected arrays for SERS microscopy. Our experimental setup for reflection spectroscopy has been described previously for experiments with gold nanostrip antennas directly on glass substrates [13, 20]. Here we use the similar method but with a ×100 objective (numerical aperture NA=0.90) and take the reference spectra R_gold from the smooth gold film (instead of from the glass substrate) in order to directly compare with the reflection R from the particle arrays.

 

Fig. 2. (a)–(b) Reflection spectra from the periodic array of square gold nanoparticles on top of a 200-nm-thick gold substrate for the indicated nanoparticle dimensions along with (c)–(d) Raman spectra obtained from the nanoparticles and a reference spectra obtained at the smooth gold film outside the array for (a),(c) x- and (b),(d) y-polarized excitation. The shaded columns in (c)–(d) indicate the spectral range used for Raman imaging in Fig. 3, and the vertical lines in (a)–(b) mark the wavelength range 530 nm–583 nm (shifts from 600 to 1650 cm^-1) visible for the calculated Raman enhancement in Fig. 6.

Download Full Size | PDF

The reflection spectra measured from the particle arrays reveal fundamental differences depending on whether the excitation and detection polarization of the electric field E is along the short- (x) or long- (y) axis of the nanoparticles, corresponding to Figs. 2(a) and 2(b), respectively. For x-polarization [Fig. 2(a)] all five particle lengths exhibit the same resonance position at ~550 nm, although there seem to be a slight red shift for longer particles, which can be ascribed to small increments of the particle width due to proximity effects in the EBL fabrication. The increasing depths in the reflection spectra for longer particle lengths originate from the relatively larger area covered by nanoparticles within each array, but all of them being probed for the resonance along the fixed (250 nm) width. Note, however, that adsorption of R6G to the sample by exposure to even µM concentrations can actually cause the nanoparticle resonances designed around 550 nm to be slightly red-shifted (5–15 nm) and broadened [19]. For y-polarization [Fig. 2(b)], the reflection spectra clearly map the influence of different nanoparticle lengths. Only the spectra obtained for square (250 nm×250 nm) particles is largely independent on polarization, as should be expected, although we do observe a minor increase in the reflection depth with y-polarization for the square particles.

3. Raman imaging and spectroscopy

After obtaining the reflection spectra and directly before the Raman measurement the sample was covered by an aqueous 10^-7 M solution of R6G for approximately 2 hours and subsequently letting the solution run off the substrate by gently blowing with compressed air. However, for these experiments the concentration of R6G adsorbed at the sample surface is not really important, since we primarily want to combine our understanding of electromagnetic FE with Raman microscopy, using R6G and the SERS spectral shape for probing of the SP resonances.

The experimental setup used for Raman microscopy is the commercially available confocal scanning Raman microscope (Alpha300R) from Witec and measurements were obtained using x- or y-polarized excitation of wavelength 532 nm, 600 lines/mm diffraction grating, and ×100 objective (N.A.=0.90), whereas we use unpolarized detection in order to have significant signal to noise ratio and clear details in the Raman images. Typical Raman spectra obtained from the brightest positions at the nanoparticles show significant dependence on polarization as well as particle dimensions [Figs. 2(c) and 2(d)], and being in good qualitative agreement with the mutual strength of nanoparticle resonances observed by their relative depths in the reflection spectra [Figs. 2(a) and 2(b)]. The reference Raman spectra were obtained using the same parameters, but from the smooth gold ~50 µm outside the periodic array and is practically identical for the two perpendicular excitation polarizations, as should be expected. The slightly increased intensity of the Raman spectra obtained with square nanoparticles for y-polarized (compared to x-polarized) excitation corresponds to the slightly increased depth measured in the reflection spectra for y-polarization. Since the tendency actually repeated itself in SERS from individual nanoparticles shown in Figs. 4 and 5, this small discrepancy should mainly be explained by minor variations in the square particle dimension (rather than in the period). Judging from the SEM image [Fig. 1(a)] the bottom right corner of each nanoparticle is actually slightly rounded indicating that both our reflection and SERS spectroscopy setups are highly sensitive to small variations in particle parameters.

 

Fig. 3. Polarized Raman images (3×3 µm²) obtained by mapping the Raman intensity of peaks integrated over 1468–1614 cm^-1 from R6G adsorbed on the 740-nm-periodic array of gold nanoparticles having lateral dimensions (a), (d) 250×250 nm, (b), (e) 250×440 nm, and (c), (f) 250×630 nm and with the excitation polarization indicated by a white arrow.

Download Full Size | PDF

Detailed Raman images were formed by mapping the spatial dependence of SERS intensity integrated around the main Raman peaks within the shift range 1468–1614 cm^-1 for each of the 30×30 points in the scan and an integration time of 400 ms at each point (Fig. 3). These scan-parameters were selected as a compromise between minimum damage/bleaching of the R6G dye molecules, significant signal to noise ratios, and sufficient step size (100 nm) compared to the spatial resolution of ~0.35 µm. The brighter image regions represent the highest Raman intensity and indicate a rather homogenous substrate giving reproducible Raman intensity from each nanoparticle. The Raman images obtained from the periodic array of square (250 nm×250 nm) particles reveal well separated and periodic increments of the bright Raman intensity corresponding to the center position of each nanoparticle and being similar for both polarizations as expected [Figs. 3(a) and 3(d)]. However, for relatively longer particles, e.g. 250 nm×440 nm [Figs. 3(b) and 3(e)], there is a clear polarization dependence which becomes even more remarkable for the longest nanoparticles [Figs. 3(c) and 3(f)], where bright spots are still clearly connected for x-polarized excitation, while images obtained with y-polarized excitation again appear like well separated particles although the interparticle spacing along the y-axis is actually only around 120 nm (much smaller than the resolution limit of ~0.35 µm). The reason for this behavior must be that for y-polarization the enhancement at the nanoparticle terminations and within the narrow ~120-nm-gap between the longest nanoparticles is significantly higher compared to the enhancement from the particle center as well as from the smooth gold film. This can be further verified by investigating Raman images from individual nanoparticles having the same dimensions and still located on smooth gold (Fig. 4).

 

Fig. 4. Raman images (1.6×2 µm²) obtained by mapping the Raman intensity integrated over 1468–1614 cm^-1 from R6G adsorbed on individual gold nanoparticles having lateral dimensions (a), (d) 250 nm×250 nm, (b), (e) 250 nm×440 nm and (c), (f) 250 nm×630 nm and with the excitation polarization indicated by a white arrow.

Download Full Size | PDF

Imaging individual nanoparticles eliminates the influence of the periodic array and isolates the shape dependence of SERS from the rectangular nanoparticles. Unfortunately, the relatively low spatial resolution of our reflection spectroscopy setup (focus spot diameter ~50 µm) is not suitable for measurements from individual nanoparticles. However, using the high-spatial resolution (~0.35 µm) scanning Raman microscopy even the distribution of SERS intensity across the individual nanoparticle can be mapped in detail (Fig. 4). The SERS intensity from the adsorbed R6G molecules actually probes the local FE around each individual nanoparticle revealing the most pronounced differences for x- and y-polarized excitation of the longest nanoparticles [Figs. 4(c) and 4(f)]. For x-polarization the SERS intensity is practically constant and relatively high along the entire nanoparticle length, whereas for y-polarization the SERS intensity from the center of the nanoparticle is significantly smaller compared to that from the ends. This increased electric field at the ends is mainly resolved for the longest particles, and can be ascribed to edge effects and breaking of symmetry, similar to what is observed for two-photon excited photoluminescence microscopy [13].

We consider the spatial distribution of SERS intensity across each nanoparticle to be very interesting and highly relevant when designing sensors, often utilizing collective SERS from larger areas (e.g., of periodic nanoparticle arrays), since for such SERS substrates only a ratio of the nanostructure (at the nanoparticle or even only at an interparticle gap) will actually support the highest enhancement. Hence, the density and design of nanostructures on the SERS substrate should be largely dependent on the application, e.g. spatially high resolved point by point investigations (SERS only at each nanoparticle) versus averaged overall SERS from closely packed nanoparticles such as self-organized nanostructures.

 

Fig. 5. Raman spectra obtained from individual nanoparticles at positions having the brightest SERS intensity and reference spectra obtained at the smooth gold film for (a) x- and (b) y-polarized excitation. The shaded columns indicate the range used for Raman imaging in Fig. 4.

Download Full Size | PDF

SERS spectra of R6G obtained with the individual nanoparticles at the spatial locations exhibiting the brightest SERS intensity reveal significant dependence on polarization as well as particle dimensions (Fig. 5). It is interesting, that for x-polarized excitation the dimension 250 nm×340 nm actually supports much higher SERS compare to both the square (250 nm×250 nm) and the more rectangular shaped particles (lengths 440 nm, 540 nm, and 640 nm). For y-polarization it is also the dimension supporting the mutually strongest SERS among the rectangular nanoparticles (i.e., beside the square shaped nanoparticle).

4. Raman enhancement

Following careful subtraction of the average fluorescence background, and showing only the spectral range having detectable Raman peaks, we evaluate the enhancements as the ratio between levels of reference and six main Raman peaks from R6G obtained at the brightest SERS positions on the nanoparticles (Fig. 6). The Raman enhancements are approximately two times lower for individual nanoparticles compared to the same nanoparticles situated in periodic arrays, indicating the important influence of collective resonances mediated by the presence of an array with period specifically designed to match SP resonances around 550 nm. Based on the relatively broad resonances mapped by the reflection spectra [Figs. 2(a) and 2(b)] one should not expect sharp wavelength dependent details for enhancement within the narrow spectral range 550–583 nm. However, the average relative enhancement strengths for the periodic arrays follow the depths and shape observed in the reflection spectra. The most pronounced polarization dependence is observed for nanoparticles within periodic arrays, but minor differences are also observed for individual nanoparticles. This is consistent with our previous experience and we expected large enhancements for x-polarization, i.e. across the short nanoparticle axis of fixed 250 nm width [10]. However, contrary to the reflection spectra, the Raman spectra are actually always affected by the SP resonances for both polarizations, i.e. depending on all four polarization combinations of excitation and detection [18]. This can explain the rather high SERS obtained for (off-resonant) y-polarized excitation of the longest nanoparticles, where the response will be a combination of resonances for both x- and y-polarization. Since we here use unpolarized detection of the Raman emission one should not expect to see the isolated effect of only one of the resonances.

Compared to Raman enhancements estimated using the similar method [1], where the values stated are usually only ~3–5, these enhancements of up to ~32 are significantly higher indicating properly selected design parameters. However, part of the increase can be related to the high spatial resolution allowing us to specifically characterize the Raman intensity only at the nanoparticles instead of averaging over larger array areas, i.e. also from between the nanoparticles, which would of course decrease the overall enhancement estimation. On the other hand, the reference Raman signal might already have been significantly enhanced due to chemical interactions between R6G and the gold film itself [21,22]. Note that, using Figs. 2 and 5, we can evaluate the fluorescence enhancement from periodic arrays and individual nanoparticles as ~4–6 and ~2–3, respectively, corresponding approximately to the square root of the Raman enhancements. This can be explained by the fact that for an enhanced electric field E, the fluorescence enhancement scales only as ~|E|² compared to the usual ~|E|⁴ dependence of the Raman enhancement, in very good agreement with the analysis in [23].

 

Fig. 6. The relative Raman enhancement from the nanoparticles arranged either in the periodic array (a)–(b) or as individual particles (c)–(d) and estimated by comparing levels at each Raman peak with levels obtained in the reference spectra. The excitation polarization in (a), (c) and (b), (d) were selected along x and y, respectively. For each curve the uncertainty marks are based upon 3–4 measurements from different places with the same configuration.

Download Full Size | PDF

5. Summary

In conclusion, we have investigated SERS from Rhodamine 6G homogenously adsorbed on both periodic arrays of and individual gold nanoparticles using high-resolution Raman imaging with polarized excitation. Rectangular 50-nm-high nanoparticles of different sizes chosen to ensure the presence of localized SP resonances close to the 532-nm excitation wavelength were fabricated with electron-beam lithography on the surface of a smooth gold film and arranged both individually (i.e., placed sufficiently far apart) and in 740-nm-period arrays. We compared linear reflection spectra and high-resolution Raman images obtained from arrays of nanoparticles, revealing good correspondence in the spectral dependences of reflection and local SERS enhancements, which was measured at the top of nanoparticles. The latter were related to those observed with individual nanoparticles and we furthermore noticed good agreement with the related only ~|E|² dependence of the fluorescence enhancement (vs. ~|E|⁴ for SERS). The results obtained emphasize the importance and quantify the influence of particle dimensions, polarized excitation, collective resonances and SERS locations.