Red shifted spectral dependence of the SERS enhancement in a random array of gold nanoparticles covered with a silica shell: extinction versus scattering

Randomly distributed gold NPs are prepared by electron beam evaporation (EBE) on a SiO₂ substrate. A 2 nm thin layer of gold, corresponding to a gold amount of 1 · 10¹⁶ atoms/cm², is evaporated at room temperature using high purity (99.99%) gold pellets as a source on the substrate heated at 480 °C. This technique permits a perfect control of the thickness of the gold film deposited on the substrate by using a quartz microbalance, confirmed by Rutherford backscattering (RBS) analysis. Gold atoms arrive on the heated substrate, diffuse over the substrate and immediately start a ripening process that leads to cluster formation. In order to promote the adhesion of the gold NPs to the substrate, the clusters were covered by a thin silicon oxide layer (2–3 nm) produced by RF magnetron sputtering. Structural characterization of the Au NPs was performed by Gemini Field Emission scanning electron microscopy (SEM) Carl Zeiss SUPRA 25. Notably, the presence of SiO₂ fosters the robust sticking of gold to glass without the need to deposit chromium or titanium layers, known to damp the LSPR of the NPs [32]. A thick flat gold film (surface roughness 1.1 nm_RMS) grown by evaporation is used as a reference for the evaluation of the SERS enhancement.

MB is used as probe molecule. MB has Raman active vibrations in the 400–1650 cm⁻¹ region with the most intense peaks at 450 cm⁻¹ and 1620 cm⁻¹ [33]. MB features a strong electronic absorption in the red part of the spectrum (monomers and dimers strongly absorb light at 670 and 620 nm, respectively) leading to a resonant Raman enhancement of ca. two orders of magnitude with respect to excitation in the NIR (see [7, 9] for more details). MB is absorbed on the gold NPs and the flat gold film by immersion into the aqueous solution of dye for 1 h. The MB solution is prepared mixing deionized water with the powder (Aldrich) at the concentration of 10⁻⁴ M. The samples are then rinsed in water and dried in a vertical position under continuous airflow to avoid the formation of macroscopic aggregates (typical coffee ring effect) of MB. This method guarantees that one or few layers of MB dye adhere onto the surfaces [see 7, 15, 34 and references therein]. The silica layer prevents the detachment of the NPs from the glass substrate during immersion, a phenomenon observed otherwise.

Measurements are carried out with a HR800 Horiba Jobin Yvon micro-spectrometer combined with an optical microscope (Olympus BX51). For the extinction measurements (figure 1(a)), we exploit the white light xenon lamp embedded in the microscope of the HR800 spectrometer. A 10X objective (NA = 0.25) is used to collect the light transmitted through the sample and the spectrometer is used to acquire the optical signal. The LSPR profile is obtained from the ratio between the light transmitted in absence (I₀) or in presence of NPs (I_NPs) according to the formula E(λ) = −log₁₀(I_NPs/I₀). In order to investigate the scattering contribution, we use a setup depicted in figure 1(b). White light is launched through an optical fiber and a collimator with an incident angle of 45° with respect to the sample plane. The scattered light is then collected through a 10X objective (NA = 0.25, working distance 1 cm, collection angle 2α = 30°) normal to the sample plane in order to avoid collection of the reflected light, and analyzed by the spectrometer. The scattering profile is obtained normalizing the scattering signal to the spectral profile of the light source. Multi-wavelength SERS measurements are carried out at 515, 532, 633, 660, 691, and 785 nm in a backscattering configuration (figure 1(c)). Measurements are done focusing a laser beam on a submicron spot with a 100X microscope objective (NA = 0.90). All the spectra were acquired with integration times from 10 to 120 s and powers from 4 to 400 μW, while the accumulation number and the slit aperture were kept fixed at 1 and 100 μm, respectively.

Zoom In Zoom Out Reset image size

Numerical calculations have been carried out using the discrete dipole approximation (DDA) [34, 35] to solve numerically Maxwell's equations. DDA is based on discretizing the target shape into a set of polarizable and interacting elements, and solving the induced system of unknowns. Here we consider a set of coated spheres randomly distributed with gold core (dimension ranging between 8 and 24 nm) and a shell of 2 nm of SiO₂. The substrate is taken into account by using an effective medium approximation with effective refractive index [36]. Then, an additional shell is added to each sphere to take into account the absorption of the probe molecule. To bench the DDA and choose the size of used dipoles, we compare the results for a unique coated sphere with the Mie theory [37].

SEM of the gold NPs sample (figure 2(a)) shows spherical- and ellipsoidal-like particles with average diameter d ∼ 18 nm and a dispersion ($2{\sigma })$ of 11 nm around d (figure 2(b)). Sparse larger NPs (d ∼ 50 nm) are also present. The gap sizes range from few nm up to few tens nm. We cannot exclude, due to the limited resolution of the SEM, that in some points the NPs could be in contact.

Zoom In Zoom Out Reset image size

Extinction measurements (figure 2(c)) show different LSPR between the bare NPs (black line) and the NPs after adsorption of probe molecules. Trans-1,2-bis4-pyridylethene (BPE), featuring no electronic transitions in the visible–NIR, causes a shift of only few nm in the LSPR profile (figure 2(c), green line). MB, instead, (figure 2(c), red line) has strong absorption between 600 and 700 nm [7, 15] and yields a red-shift of the LSPR of about 35 nm (from 575 to 610 nm) and a marked broadening toward the near-infrared part of the spectrum. Shifts are likely due to the change of effective dielectric constant of the medium surrounding the NPs upon absorption of the probe molecules [17]. Numerical calculations carried out by DDA on small NPs clusters (eight particles with random arrangement and dimension ranging between 8 and 24 nm) confirm how the deposition of thicker and thicker layers of probe molecules (we assume a molecule with constant refractive index, for simplicity) yields an increasing red-shift of the LSPR (figure 2(d)). The backscattering (figure 2(c), blue symbols) peaks around 750 nm, remarkably red-shifted with respect to the extinction maxima.

Some qualitative analysis of the SERS enhancement expected from the gold NPs clusters can be drawn based on the E⁴ model, assuming that the local- and the re-radiation field enhancement factors scale with the extinction intensity, i.e., EF ∼ Q_e(λ_L) × Q_e(λ_R). Excitations at 515, 633, and 785 nm (indicated by the colored vertical lines in figure 2(c)) cover both the ascending and descending regions of the LSPR profiles and are, therefore, well suited to gain a first insight on the spectral dependence of the EF. In particular, if the extinction of the bare NPs is considered for the calculation (equation (2)), SERS is expected to be maximum at 515 nm (where both the excitation and the Raman photon energies are close to the peak LSPR) and negligibly small at 785 nm (where both laser and Raman photons are off-resonance). If, instead, we consider the MB-induced red-shifted extinction profile, maximum SERS is expected at 633 nm (laser is resonant with LSPR). Excitation at 785 nm is in the tail of the LSPR, so some SERS is expected although the enhancement should be much smaller than at 633 or 515 nm. Excitation at 633 nm represents the situation closest to Wokaun's rule. Some considerations can also be drawn on the expected relative amplification of the different vibrational modes at a given excitation wavelength. MB is well suited to this aim, featuring several vibrations in the 400–1650 cm⁻¹ region (indicated by the dashed boxes in figure 2(c)). According to equation (2), for a given λ_L, we expect EF ∝ Q_e(λ_R). Based on the measured extinction profiles, at 515 nm, we expect a progressive increase of the intensity of the SERS peaks passing from the low frequency MB modes to the higher ones. An opposite behavior is foreseen for excitations at 633, 660, 691, and 785 nm, in which the condition of maximum re-radiation is better verified for the 450 cm⁻¹ peak than for the 1620 cm⁻¹ one.

In order to study the spectral dependence of the SERS enhancement and check the validity of Wokaun's rule on gold NPs clusters, SERS experiments have been carried out using six lasers tuned from the visible to the near-infrared (515, 532, 633, 660, 691, 785 nm). For each excitation, the SERS intensity of the different MB vibrational bands (peaked at 450, 500, 897, 1152, 1301, 1390, 1425, and 1620 cm⁻¹ [7, 38]) are compared with the corresponding Raman intensities acquired (with the same setup) on a reference sample consisting of flat gold film known to provide negligible (if any) amplification [7, 9].

Normalization of the SERS signal measured on the gold NPs to the Raman signal from a flat gold surface allows one to get rid of any modification of the Raman cross-section due to chemical interactions of the molecule with gold (e.g., chemical enhancement) and resonant Raman enhancement effects related to the resonant excitation of the molecule (as occurs for MB when using red lasers), highlighting the pure electromagnetic effects in SERS [7]. The normalized signal is, in addition, independent from the wavelength-dependent response of the spectrometer. In figure 3 we plot the SERS gain G, calculated as the SERS to Raman intensity ratio, normalized to the power and integration times [9], as a function of the excitation wavelength (different colors) and, for each excitation, as a function of the wavelength of the Raman photons (colored symbols). It is evident the incremental behavior of G; at 515 nm, we have a modest enhancement ranging from ∼7, for the low frequency modes, to ~60 for the band at 1620 cm⁻¹. At 532 nm, G increases from 30 to 100 and at 633 nm from 220 to 350. At 660 and 691 nm, the gain reaches three orders of magnitude, and seems to be constant whatever the considered mode under these irradiation wavelengths. Maximum gain is observed for excitation at 785 nm, for which we calculated an enhancement of 3 × 10³ for the band at 450 cm⁻¹, that decreases to 1 × 10³ for the 1620 cm⁻¹ mode. Notably, the SERS gain G is smaller than the actual SERS enhancement factor, defined as $EF=\left({I}_{{\rm{SERS}}}/{N}_{{\rm{SERS}}}\right)/\left({I}_{{\rm{REF}}}/{N}_{{\rm{REF}}}\right)$ [39], by a factor N_REF/N_SERS related to the different number of molecules probed in the SERS (N_SERS) and the reference (N_REF) measurements. The factor N_REF/N_SERS is typically much higher than one since the molecules experiencing SERS are those located in the nanometric hot spots excited by the laser spot, i.e., much less than the number of total molecules illuminated by the laser spot itself.

Zoom In Zoom Out Reset image size

More insight on the mode-selective SERS amplification, and therefore on the re-radiation enhancement, is obtained comparing the ratios $\eta =\left({I}_{1620}/{I}_{450}\right)$ between the high and the low energy vibrational modes of MB measured in SERS (spectra in figure 4, colored lines) and in Raman (black lines) as a function of the main excitation wavelengths (515, 633, and 785 nm). The results are summarized in table 1.

Zoom In Zoom Out Reset image size

Table 1.  Comparison of the ${{\boldsymbol{I}}}_{1620}/{{\boldsymbol{I}}}_{450}$ ratios as a function of the excitation wavelength in Raman and SERS.

${\eta }$	Raman	SERS	ηSERS/ηRAMAN
515 nm	3.8	9	2.37
633 nm	1.2	1.8	1.5
785 nm	0.5	0.27	0.54

The intensity ratios $\eta ,$ intrinsic to the MB molecules and measured in the Raman reference spectra, are modified in the SERS spectra. At 515 nm, the 1620 cm⁻¹ mode is more amplified than the band at 450 cm⁻¹ of a factor ${\eta }_{{\rm{SERS}}}/{\eta }_{{\rm{Raman}}}=2.37.$ At 633 nm, this selective amplification is reduced to 1.5. At 785 nm, the opposite is observed, with a selective amplification of the band at 450 cm⁻¹ with respect to the 1620 cm⁻¹ peak by a factor 1.85. This phenomenon can be assumed as a signature of the different re-radiation experienced by the Raman modes as the excitation is changed.

Indeed, Wokaun's rule is not verified on our randomly distributed gold NP arrays, no matter if we consider the bare NPs LSPR profile or the MB-induced red-shifted one. Maximum SERS enhancement is observed for excitation at 785 nm, i.e., at a wavelength red-shifted by more than 160 nm with respect to the peak of the LSPR profiles. For what concerns the different (re-radiation) enhancement of the SERS modes, the experimental results at 515 nm (higher amplification of the 1620 cm⁻¹ mode) and 785 nm (higher amplification of the 450 cm⁻¹ mode) are in agreement with what is expected from the LSPR profiles. Conversely, the trends at 633 and 660 nm (higher amplification of the 1620 cm⁻¹ mode) and 691 nm (equal amplification) are opposite to what is expected from Wokaun's rule. Surprisingly the trend of the SERS gain shown in figure 3 and of the selective mode enhancement (figures 3 and 4) qualitatively agrees with the wavelength dependence of the scattering spectrum reported in figure 2(c, blue line), peaked at 750 nm. Indeed, both the laser wavelengths tuned at 515 and 633 nm excite the ascending part of the scattering profile, while the NIR excitation matches its maximum. Furthermore, the SERS intensity re-radiated by the NPs in the visible range increases for the highest frequency Raman modes of MB and for the highest excitation wavelengths (from 515 to 691 nm), while an opposite trend is verified in the near-infrared for the excitation at 785 nm. Our findings therefore suggest that the scattering could provide useful information for the optimization of the SERS efficiency in randomly distributed metal NPs, with the advantage of being a quantity easily measurable in the far-field.