1. Introduction

Localized surface plasmon resonances (LSPR) are electromagnetic modes bound to nanometer-size metallic structures, where collective oscillations of surface electrons are excited upon illumination with light of appropriate wavelengths. Due to the appealing properties like giant field enhancement in an extremely small volume, LSPR exhibits significant advantages in enhancing the interaction of light with matter in close proximity to nano-structured metal surface, and boosts a broad range of applications in optical sensing, biological imaging, and nonlinear optics. In particular, the contribution of LSPR to surface-enhanced Raman scattering (SERS) is twofold: not only the local field experienced by a molecule but also the Raman scattered field could be enhanced, leading to more total photon-molecule interactions and also more detectable scattered photons [1]. Similar statements hold for other nonlinear optical processes too, i.e., both excitation and emission could be enhanced.

It is now well understood that simple single nanoparticles such as isolated spheres or rods of gold or silver could offer modest field enhancement. On the other hand, assembled or lightly aggregated nanostructures [2–7] and also single nanoparticles with particular morphologies like spiky shells [8–12] offer large field enhancement in small gaps or around sharp corners where surface plasmons are localized into “hot spots”. Strong enhancement of field can also be achieved in metal-insulator-metal (MIM) geometries where a thin insulator layer is in between two metal layers [13,14]. Immobilized nanoparticles on appropriate substrates have been proven as interesting and effective in SERS application, but the fabrication of such substrates based on controlled assemblies of nanostructures involves complicated chemical processing steps. Thus, in recent years, many techniques have been developed to produce SERS substrates that have deliberately designed periodic surface structures [15–22] (“top-down” process). Despite impressive results in these works, the working band is always narrow, and different substrates need to be fabricated to fit different excitation sources or different analytes. Furthermore, concurrent multiple or broadband resonances would be beneficial to provide enhancement at the frequencies of not only the excitation laser but also the Stokes shifted Raman vibration modes [15, 18, 23].

In this paper we fabricate a special kind of plasmonic nanostructures with intrinsic electromagnetic hot spots by oblique angle deposition (OAD) of noble metals, on either planar or three-dimensionally (3D) structured substrates. OAD is a convenient and fast method to prepare plasmonic substrates for sensing applications [24–27]. Through OAD method, the deposited metal layer is not a continuous film but rather consisting of nanoparticles [28]. In addition to the OAD of metals, we also include a design of three-layer MIM nanostructures, that is, a dielectric spacer layer is inserted between the top metal nanoparticles and the bottom metal layer. Our study also suggests that this method can form sub-10 nm air gaps between metal particles on the surface. Inherent electromagnetic hot spots are generated due to the coupling between two stacked metal layers (the MIM nanostructures) and also the coupling between those adjacent metal particles with sub-10 nm air gaps. Creating random distribution and shape (or size) of metal nanoparticles rather than periodic and uniform structures will broaden the overall LSPR response of the ensemble, which is promising for broadband SERS applications in the visible and near infrared region.

2. On planar substrates

We first look at the case of planar silicon or glass as supporting substrates. To realize the three-layer MIM nanostructures, a reflective mirror (Ti/Ag, 5 nm/120 nm at 0.1 nm/sec) and a dielectric spacer (Al₂O₃, 20 nm at 0.02 nm/sec) were deposited firstly with normally incident vapor flux by e-beam evaporation (DENTON VACUUM, base pressure 1 × 10⁻⁶ Torr, with a continuously water cooled rotating substrate holder). The thickness of the dielectric spacer was chosen carefully to get an optimized field enhancement and the details could be found in Refs. [29–31]. After that, the chamber was opened to tilt the substrates (the surface normal is at an angle of 70° with respect to the direction of the incident vapor flux). The oblique deposition of silver was performed at a slow rate (0.02 nm/sec). The nominal thickness of the top silver layer by OAD is varied from 20 nm to 60 nm (measured by the film thickness monitor positioned normal to the vapor flux). Control experiments were done to compare the performance of the three-layer MIM structures with single-layer silver nanoparticles. The single-layer substrates were fabricated on glass substrates by OAD simultaneously (the glass substrates were cleaned by acetone/IPA/DI water in an ultrasonic bath, and then further treated with oxygen plasma). The insets of Fig. 1(a) show the digital photographs of the fabricated three-layer (left; on silicon) and single-layer (right; on glass) substrates with nominally 20 nm thick Ag on the top. These two substrates appear different colors, indicating quite different optical properties.

 

Fig. 1 Top-view SEM images of the three-layer substrates with nominally (a) 20 nm, (b) 30 nm, and (c) 60 nm thick silver on the top. The insets in (a) are photographs of the three-(left) and single-layer (right) substrates with nominally 20 nm thick silver on the top. The inset in (b) is the corresponding cross-sectional SEM image. (d) Histogram of the effective radius of silver nanoparticles when the nominal thickness of the top silver layer is 30 nm.

Download Full Size | PDF

The microstructure and optical property of silver films depend on many deposition parameters [32]. Figs. 1(a)–1(c) are the SEM images of three-layer substrates with nominally 20, 30, and 60 nm thick OAD silver on the top, respectively. To estimate the size distribution of Ag nanoparticles on the top, we calculated the effective radius of silver nanoparticles from the SEM image statistically. The effective radius is the radius of a circle that would have the same area as the area of a nanoparticle occupies. We tabulate the effective radius distribution of the nanoparticles in a histogram as shown in Fig. 1(d) when the nominal thickness of the top Ag layer is 30 nm. The effective radius distribution can be fitted to a Gaussian function (red curve), and the median is about 15.5 nm (the median for other nominal thicknesses is indicated in each SEM image in Fig. 1). The size of Ag nanoparticles increases with the nominal thickness, and a continuous film will not form even with a nominal thickness of 100 nm.

To show the difference in tuning optical properties by the three- and single-layer substrates, we measured the reflection (R) and transmission (T) spectra by a spectrophotometer (Perkin Elmer Lambda 950 with a 150 mm integrating sphere) and show the absorption spectra (A = 1 − R − T) in Fig. 2 (an uncalibrated Labsphere Spectralon Reflectance Standard was used as the reference for reflection measurement, and this may cause 1–2% error). Apparently, for both kinds of substrates, the optical properties can be tuned by varying the nominal thickness of the top silver layer, but in Fig. 2(a) we can see that, the absorption peak shifts to longer wavelengths with increasing thickness more dramatically. Figure 2 also indicates that more energy of incident light is absorbed by the three-layer substrate which means stronger LSPR (thus stronger field enhancement) occurs on its surface. In contrast, as seen in Fig. 2(b), the single-layer substrates do not show apparent absorption peaks for wavelengths longer than 500 nm, and the overall absorption is low, indicating that they may not exhibit high field enhancement.

 

Fig. 2 Absorption spectra of (a) three-layer and (b) single-layer substrates with different nominal thickness of the top silver layer.

Download Full Size | PDF

From Fig. 2(a) we see that the absorption peak is located near the wavelength of 633 nm when the nominal thickness of the top silver layer is 30 nm. We deduce that with this nominal thickness strong enhancement can be achieved by using 633 nm laser as the excitation source for Raman measurement. To verify this, we employed 4-methylbenzenethiol (4-MBT) as the probe molecules because it can form a well-defined monolayer on silver surfaces with characteristic molecular footprints [4]. The substrates were immersed into 1 mM ethanol solution of 4-MBT for 2 hours. After that, the substrates were taken out, washed with copious amounts of ethanol, and finally dried under a stream of nitrogen [33]. The Raman spectra were acquired by using a Renishaw in Via Raman microscope system with a 633 nm laser source (the laser power after the objective was about 78.7 μW). The collection time was set as 10 sec, and the Raman signals were collected via a 5× magnification objective. All data were baseline corrected.

In Fig. 3(a), we show a series of SERS spectra from three-layer substrates with different nominal thickness of the top silver layer. Two Raman peaks can be clearly identified at 1077 and 1592 cm⁻¹ from all substrates, with the 1077 cm⁻¹ peak due to a combination of phenyl ring breathing, C–H in-plane bending and C–S stretching and the 1592 cm⁻¹ peak arising from phenyl ring stretching of 4-MBT [34]. The coefficient of variation (the ratio of the standard deviation to the mean intensity) of the characteristic peaks of 1077 cm⁻¹ and 1592 cm⁻¹ (collected at 10 different positions) on each substrate are less than 10%, which indicates a good uniformity and repeatability for SERS applications. From the variation trend shown in Fig. 3(c), we find that the signal from the three-layer substrates is strongest when the nominal thickness of top silver layer is 30 nm, which is in agreement with the absorption spectra measurement in Fig. 2(a).

 

Fig. 3 SERS spectra of 4-MBT on (a) three-layer and (b) single-layer substrates with different nominal thickness of the top silver layer. (c) Intensity variation trend of the characteristic peaks of 1077 cm⁻¹ (red) and 1592 cm⁻¹ (black) as a function of the nominal thickness of the top silver layer.

Download Full Size | PDF

The Raman signal from the three-layer substrate can be 5 times as strong as that from the single-layer substrate. To quantify the Raman signal enhancement ability, we further calculated the enhancement factor (EF) values of the three-layer and single-layer substrates, which is defined as EF = (I_SERS/N_Surf)/(I_RS/N_Vol) [35], where I_SERS and I_RS are the intensities of the selected Raman peaks in the SERS and non-SERS spectra, and N_Surf and N_Vol are the amounts of adsorbed molecules in the scattering volume for the SERS and non-SERS experiments, respectively. When estimating N_Surf, the silver particles were assumed to be semi-spherical with the statistically obtained radius and covered with a dense monolayer of molecules. The non-SERS experiments is measured from 1 mM 4-MBT ethanol solution in a quartz cuvette. We chose the peak 1077 cm⁻¹ to calculate the enhancement factor. As a result, the enhancement factor for the best three-layer (single-layer) is 4.08 × 10⁷ (8.21 × 10⁶).

To get some insight about the enhancement, we adopted the method described in Refs. [36, 37], i.e., to cut the SEM images of random structures into several parts, calculate the spectra of each part and then average them, and finally obtain the spectra that agree well with the measurement. We picked the SEM image of 30 nm top silver layer and cut it into 16 equal parts with area 1 μm × 1 μm, converted into binary images and then imported into Lumerical FDTD (assuming rod-like particles with thickness estimated from cross-sectional SEM images) for simulations. The optical constants of silver were taken from Ref. [38]. The incident wave propagates along the z direction with the electric field polarized along the x direction. The simulation results are shown in Fig. 4. As shown in Figs. 4(a) and 4(b), the light cyan lines are the spectra of individual parts and the blue line is their average. For reference the measured spectra from the macroscopic sample are also shown. In order to calculate the representative electric field distributions, as shown in Figs. 4(c) and 4(d), we picked the instance with its spectra closest to the averaged spectra, i.e., the red lines in Figs. 4(a) and 4(b). By comparing Fig. 4(c) with Fig. 4(d), we can see that the enhancement of the three-layer substrate is an order of magnitude higher than the single-layer substrate. In Fig. 2 it was shown that the three-layer substrates have larger absorption-peak shifting than the single-layer ones. This phenomenon can be reproduced by FDTD simulation, as shown in Figs. 4(e) and 4(f). We carefully checked the cross-sectional SEM images and found that the average size of Ag nanoparticles and the gap size between the nanoparticles were similar for both kinds of substrates. In three-layer substrates, the plasmonic coupling between metal nanoparticles and the bottom metal film, can boost not only the electric field intensity by at least an order of magnitude, but also the shifting of absorption peak with increasing nominal thickness of the top silver layer.

 

Fig. 4 FDTD simulation of structures with random metal nanoparticles. (a, b) Absorption spectra of the three- and single-layer substrates with nominally 30 nm Ag on top. The blue line is the average of 16 instances (light cyan lines) and the red line is the instance which is closest to the average. For reference, the measured spectra from the macroscopic sample are also shown (black line). (c, d) Electric field intensity distribution (at the wavelength of 633 nm; normalized to the intensity of incident light) on the top surface of the structures with nominally 30 nm Ag. (e, f) Absorption spectra of three- and single-layer substrates with different nominal thickness of the top silver layer, indicating larger absorption-peak shifting for the three-layer substrates than the single-layer ones.

Download Full Size | PDF

3. On structured substrates

For SERS measurement, substrates with more hot spots can produce higher Raman signal intensity. Compared with 2D structures, 3D structures can have more hot spots and more analyte molecules in the laser confocal volume during the measurement [39–41]. Meanwhile, well patterned structures can make measurement more reproducible. In this section we present results with 3D structured substrates and therefore there is no need to tilt the substrates during the vacuum evaporation. Here the 3D structured substrates were prepared by wet chemical etching of silicon to form tilted walls. Two kinds of silicon structures were used, i.e., macroporous black silicon [42–44] and inverted-pyramidal pits. For substrates with surface structures of high aspect ratios, nanostructures can be formed by evaporating certain metals directly on it, just like the way of oblique angle deposition [45] as described in the previous section. By inserting a dielectric spacer between two successive metal evaporations, we can thus produce MIM nanoparticles, as highlighted by a red ellipse in Fig. 5(a) (with 50/10/50 nm Ag/SiO₂/Ag deposited by e-beam evaporation in a single run). Similar MIM nanoparticles were previously fabricated by vacuum deposition through nanohole templates created by e.g. e-beam lithography [18,46,47]. The morphology of metal nanostructures we obtained is distinct from those where a single metal layer was deposited on nanopillars or nanocones [48–50], and those where the interparticle spacing was controlled by shrinking the air gaps with initial width of tens of nanometers [20, 51]. Here the bottom silver layer was set to 50 nm thick and the dielectric spacer layer 10 nm, so that there are more air gaps to generate more active hot spots. As shown in Fig. 5(b), two kinds of plasmonic coupling between metal particles can be identified: one is through the air gap between particles within the same metal layer (highlighted by red arrows), and the other is through the dielectric spacer between two metal layers (highlighted by green arrows), as schematically illustrated in the cartoon in Fig. 5(e). We emphasize that templates like nanoholes or nanopillars are not necessary to form these coupled MIM nanoparticles. By comparing Figs. 5(d) with 5(b), we can see that the spacer layer is more conformal to the local substrate surface if using Al₂O₃ as the dielectric spacer (highlighted by the green arrows). We would like to mention the difference between our structures and the commercial “Klarite” substrate and its derivatives [52–56] where a thick and continuous metal film (e.g., 300 nm thick gold) is deposited on inverted or upright pyramidal structures, for which the hot spots arise from the coupling of surface waves on adjacent facets and therefore the shape and size of pyramids need to be engineered.

 

Fig. 5 (a) Cross-sectional SEM image showing 50/10/50 nm (the thickness of bottom/spacer/top layer, quoted from the film thickness monitor) Ag/SiO₂/Ag deposited on macroporous black silicon with blade-like walls of about 70° slope. The top silver layer has a brighter contrast than that of the bottom layer. One MIM nanoparticle is highlighted by a red ellipse, with two bright particles separated by a dark spacer. (b) An enlarged view of the rectangular window in (a). The red arrows highlight the air gaps between the silver particles within the same layer, and the green arrows highlight the dielectric gaps between two silver layers. (c) Cross-sectional SEM image showing 50/10/20 nm Ag/Al₂O₃/Ag deposited on silicon walls of an inverted pyramidal pit. The top silver layer has a brighter contrast than that of the bottom layer. (d) Cross-sectional SEM image showing the titled MIM nanoparticles obtained by depositing 50/10/60 nm Ag/Al₂O₃/Ag on silicon walls with slope of about 70°. The green arrows highlight the dielectric gaps between two silver layers. (e) Schematic illustration of electromagnetic “hot spots” between metal nanoparticles on a slope.

Download Full Size | PDF

The near-field enhancement of the MIM nanoparticles was probed by two-photon induced luminescence and SERS. Two-photon luminescence is an optical process where two near infrared photons are simultaneous absorbed and a single photon in the visible region is emitted. The visible luminescence from noble metal nanostructures is originated from radiative recombination of interband transitions, which could be enhanced by LSPR [57–59]. Since the excitation is proportional to the square of the near infrared laser power, a spatial resolution beyond the limits of conventional far-field optics can be achieved. The substrates had an additional 50 nm Al₂O₃ coating to protect silver against oxygen and water in the air. An ultrafast laser capable of 100 fs pulses at a repetition rate of 80 MHz was brought into the rear port of an inverted microscope (Olympus IX81) and focused with a 60× oil-immersed objective (NA=1.35) to a focal spot size of 0.35 μm as the excitation source. The fluorescence was collected by the same objective and passed through a dichroic mirror (FF670-SDi01, Semrock) and a shortpass filter (FF01-665/SP, Semrock). For SERS measurement, again we employed 4-MBT as the probe molecules.

The luminescence spectra are quite broad without sharp features, and the images from the green channel (495–540 nm, Olympus) are shown in Figs. 6(a)–6(c) for different silver nanostructures deposited on two kinds of structured silicon substrates as in Fig. 5. The images were taken at the same excitation and detection levels. The image in Fig. 6(a) (with 50/10/30 nm Ag/Al₂O₃/Ag deposited on the macroporous black silicon) is much brighter than Fig. 6(b) (with 50 nm Ag deposited on the same substrate), partially due to higher electric field enhancement from the MIM nanoparticles. Comparison between Fig. 6(a) (on the macroporous black silicon) and Fig. 6(c) (on the array of inverted pyramidal pits) indicates that the former produces higher enhancement. A more pronounced comparison is that no luminescence spots were observed for the case with 50 nm Ag deposited on the inverted pyramidal pits, in contrast to Fig. 6(b). Therefore the slope of the local substrate does matter. For example, for the wall of an inverted pyramidal pit shown in Fig. 5(c) with a slope of 54.74°, the evaporated 50 nm Ag at the bottom forms a continuous network with air gaps (the air gap gets smaller with increasing film thickness and finally disappears if the film is too thick); while for the blade-like wall shown in Fig. 5(d) with a slope of about 70°, the 50 nm Ag at the bottom forms only nanoparticles. Therefore, the number density of particles (and also air gaps) gets larger if the slope gets steeper. In contrary, the film thickness, measured along the normal to the local surface, gets smaller for steeper slopes.

 

Fig. 6 Two-photon induced luminescence images of silver deposited on silicon substrates with pores/pits arranged in a hexagonal array of 6 μm period: (a) 50/10/30 nm Ag/Al₂O₃/Ag deposited on macroporous black silicon, (b) 50 nm Ag deposited on macro-porous black silicon, (c) 50/10/30 nm Ag/Al₂O₃/Ag deposited on array of inverted pyramidal pits. The images (only show the green channel) were taken at the same excitation and detection levels. The substrates have an additional 50 nm Al₂O₃ coating to protect against oxygen and water in the air. (d) Raman spectra of 4-MBT molecules measured from inverted pyramidal pits and planar silicon decorated with different Ag nanostructures under 532 nm laser excitation. (e) Repeatability measurement of Raman spectra of 4-MBT molecules from 10 inverted pyramidal pits decorated with 50/10/30 nm Ag/Al₂O₃/Ag, showing the uniformity of SERS signal. For each measurement, the cross hair was placed on the apex of individual pit, as shown by the inset in (d). The coefficient of variation (the standard deviation divided by the mean intensity) is 6.2% for the 1077 cm⁻¹ peak, and 7.8% for 1592 cm⁻¹ peak.

Download Full Size | PDF

Although much higher enhancement can be achieved on the blade-like walls of macroporous black silicon, it is difficult to find the right focus during the Raman spectra measurements. From Fig. 6(c) we can see that the luminescence intensity around the apex of individual inverted-pryramidal pits is uniform across different pits, therefore we only present Raman spectra measured from array of these inverted-pyramidal pits, without changing the focus during the measurement process. As can be seen from Fig. 6(d), all the inverted pyramidal samples show obvious Raman peaks of 4-MBT, including the prominent peaks at 625, 796, 1015, 1077, 1185, 1381, 1487 and 1592 cm⁻¹. However, under the same experimental condition, as indicated by the pink curve in Fig. 6(d), not all the spectral peaks of 4-MBT on the plain sample are observable. Obviously, the 50/10/30 nm Ag/Al₂O₃/Ag deposited on inverted pyramidal pits indicates the maximal Raman enhancement, which benefits from the MIM nanoparticles. As the reproducibility of Raman signal from SERS substrates has been a challenging issue for practical applications, we also show in Fig. 6(e) the Raman spectra acquired from 10 apexes of individual pits. The coefficient of variation for the 1077 cm⁻¹ peak is approximately 6.2%, which is excellent.

4. Conclusion

To summarize, we have described a cost-effective way to fabricate large-scale SERS substrates with broadband and tunable absorption peak around a given excitation laser wavelength. Oblique angle deposition is used to form silver nanoparticles, without any additional nanofabrication steps. The proposed substrate has two typical advantages: with broadband optical response and tunable absorption peak. The broadband property is beneficial to the practical SERS detection of matters with different Raman shifts, and the tunable absorption peak can be used to fit a given excitation laser wavelength. Intrinsic electromagnetic hot spots are generated due to the field coupling between metal nanoparticles and the bottom metal film, and also the coupling between adjacent metal particles. These hot spots are easily accessible for the analytes. By two-photon induced luminescence imaging and Raman measurements, we demonstrate that strong and uniform field enhancement can be achieved by the proposed substrates containing MIM nanoparticles over a large area. The fabrication method can be extended to produce bi-metal particles which combine two different metals within a single nanostructure in a controllable fashion and may result in new optical and chemical properties [60, 61].