1. Introduction

Metal nanoparticles have attracted considerable interest in a variety of fields, due to their intriguing optical properties attributed to their surface plasmon resonance [1]. Surface-enhanced Raman spectroscopy (SERS) is one of the most important applications utilizing noble metal particles, which improves the feasibility of many detection techniques, particularly in the research fields of biomedicine, materials science and electrochemistry [2]. Enabled by the advancements of nanofabrication technologies and developments of simulation capabilities in the recent years, a wide range of metal nanostructures have been demonstrated as SERS-active substrates, which can provide dramatic enhancement of the detection sensitivity up to 15 orders of magnitude [3]. For example, individual metal colloidal nanoparticles have been used to amplify a spectroscopic signal of adsorbed molecules [3]. Nanohole and nanodisk arrays have been proved to exhibit strong SERS effects dependent on the periodicity of these arrays [4]. It has been demonstrated that, compared to an individual metal nanoparticle, a close proximity between particles, introducing regions of strengthened electromagnetic field intensity and therefore producing large enhancement factors, makes them excellent supports for SERS [5]. For practical SERS applications of metal particles, the compatibility with other planar devices such as micro-optical waveguides and microfluidics is also important and has become an attractive research trend in recent years [6–8].

Metal nanoparticles are synthesized in glass matrix by various methods, such as direct ion implantation [9], sol-gel methods [10] and ion exchange [11]. Conventionally, ion exchange is used to introduce silver ions in glass by immersing the samples into a molten salt, and additional processing such as light irradiation or thermal annealing then follows to generate particles. In iron-floated glasses, such as soda-lime glasses and commercial microscope slides, iron ions in glass act as reducing agents to achieve internal oxidation-reduction process and make silver ions reduced to metallic clusters [12]. For high quality optical glasses without thermoreducing impurities, we have demonstrated that masked ion exchange technique enables the fabrication of silver nanoparticle patterns by making silver ions capture the electrons from the Al mask and reduce into silver particles, and these silver nanoparticle patterns can be utilized for SERS applications [13,14]. However, the particles are only formed underneath the mask edge, and the densest particle distribution and the highest SERS signal are obtained just on a thin line with the width of a few hundred nm close to the mask edge. When used as a SERS-active substrate, very high magnification microscope objective (100 × in [14]) is needed to carefully locate this pattern position, and submicron accuracy is needed in the alignment to focus the excitation laser on these points. This increases the difficulties and complexity of the measurement, limits the usability in practical applications, and may also produce variations in the measured SERS response. When combining the active areas with other components, such as optical waveguides and microfluidic channels, very careful alignment is then required. Besides, a photolithographic mask is compulsory in formation of the particles, which are thus fabricated by relatively more expensive equipment in a more complex photolithography process. Therefore, for the practical applications of SERS, a more versatile method for the preparation of nanoparticles is needed, to produce nanoparticle aggregates uniformly in regions which have desired dimensions and shapes.

In this paper, we report on the formation of silver nanoparticles (Ag NPs) and uniform nanoparticle clusters (over large areas) embedded in glass prepared by a two-step ion exchange method. Silver ions are introduced into glass in silver ion exchange, and reduced into metallic silver in subsequent potassium ion exchange. The galvanic replacement reaction takes place in potassium ion exchange to reduce silver ions into metallic silver, enhanced by the electrolytic deposition, which further promotes the formation of Ag NPs. The formed particles are embedded between the depths of 50 nm and 220 nm from the glass surface. The random distribution of particles forms silver nanocluster structures. After etching the glass surface to partially expose the clusters, their SERS activity is demonstrated using Rhodamine 6G as an analyte. The SERS performance of these nanoparticle clusters is better than that of discrete isolated particles [14].

The paper is organized as follows. Section 2 explains the processing steps to synthesize the Ag NPs and describes the characterization methods and equipment. Section 3 presents the experimental results. Section 4 discusses the mechanism of the formation of Ag NPs and the relationship between the surface features and the SERS performance. Finally, some conclusions are drawn.

2. Experimental

2.1 Chemicals and materials

Silver nitrate (≥99% AgNO₃), sodium nitrate (≥99% NaNO₃), potassium nitrate (≥99% KNO₃) and Rhodamine 6G (~95% R6G) were obtained from Sigma Aldrich. All the nitrates were used without further purification. R6G was mixed with deionized water to make a solution. The Corning 0211 glass substrates were cleaned by acetone and IPA in an ultrasonic bath, and then by piranha solution before the ion exchange.

2.2 Synthesis of Ag NPs embedded in glass

Ag NPs embedded in glass were synthesized by a two-step ion exchange process. A pre-cleaned Corning 0211 glass was immersed into a molten mixture of 5% AgNO₃ in a 50/50 mixture of NaNO₃ and KNO₃ at 300 °C for 6 hours to exchange silver ions into the glass. A 100-nm-thick Al film was then evaporated on the ion-exchanged glass by an electron-beam evaporator. The sample (#1) was dipped into a KNO₃ salt at 400 °C for 2 hours to promote the formation of particles. In order to understand the formation of particles, two additional samples were fabricated: i) annealed in the air at 400 °C for 2 hours with the Al layer (#2) and ii) K⁺ ion-exchanged at 400 °C for 2 hours without the Al layer (#3). Table 1 summarizes the processing details of these three samples.

Table 1. The processing details of three different samples.

View Table

2.3 Material’s characterizations

UV-Vis spectra of the fabricated glass samples were recorded by a UV spectrometer (Perkin Elmer Lambda 950). In order to characterize the shape and size of the Ag NPs, a specimen prepared by a focused ion beam (FIB, FEI Helios Nanolab 600) milling was observed by a transmission electron microscope (TEM, JEOL JEM-2200FS). The lateral distribution of particles was scanned by an atomic force microscope (AFM, NTEGRA Prima) with a semi-contact mode. To demonstrate the SERS activity of the particles, samples were etched to different depths and incubated in a 1 µM R6G solution for 10 min. Raman measurements were performed using a confocal Raman microscope (Witec alpha300 RA) with a 20 × magnification. The microscope uses a backscattering configuration and the excitation laser is a 532 nm frequency doubled Nd:YAG. The collection time used was 1 s. The Raman measurements were conducted in ambient conditions.

3. Results

Figure 1 shows optical absorption spectra for the samples prepared under different experimental conditions. All of the three samples were Ag⁺ ion-exchanged for 6 hours, but subsequently processed in different ways (see Table 1). Both samples #1 and #2, which, during the 2^nd process step, had an evaporated Al layer on the ion-exchanged glass, show a strong resonance peak. The peak in absorbance for the sample #1 is much higher, indicating that more Ag particles are formed by immersing the sample into a KNO₃ salt than by just annealing it in the air at an elevated temperature. For the sample #3, which had no Al film during the 2^nd step, there is no clear resonance peak seen in the spectrum. This result reveals the essential role of both the Al layer and the KNO₃ salt in the subsequent process. Both of them together enhance the formation of particles.

 

Fig. 1 Optical absorption spectra of different glass samples with embedded Ag NPs. The samples were prepared under different experimental conditions.

Download Full Size | PDF

Figure 2(a) shows the cross-section distribution of Ag NPs under the glass surface for the sample #1. The particles appear with a high density at the depth of some 50 nm from the glass surface, and extend to the depth of about 220 nm. Figure 2(b) gives the bright field TEM image of Ag NPs with high magnification. It can be seen that the particles are nearly spherical, with a diameter of about 5-10 nm. Some of the particles are separated by a narrow gap, which can give a strong interparticle coupling effects in SERS [5]. The gap could be a perfect location for the adsorbed molecules to show the enhanced Raman signal.

 

Fig. 2 (a) The overview image showing the distribution of Ag NPs under the surface of the sample #1. (b) Bright field TEM image of the particles observed with high magnification.

Download Full Size | PDF

To study the distribution of the Ag NPs and the corresponding SERS performance, the surface of the sample #1 was etched to three different depths, to partially expose the Ag NPs at 55 nm, 110 nm and 220 nm. Figure 3 shows AFM images scanned from the etched surfaces. The scan area is 6 μm × 6 μm. When the sample is etched to 55 nm, the height scale is much smaller than that of the sample etched to 110 nm and 220 nm. The root-mean-square (RMS) roughness of each image has been calculated as 24.2 nm, 38.6 nm, and 46.1 nm respectively, when etched to 55 nm, 110 nm and 220 nm.Two-dimensional power spectra (2D-PS) images obtained from the Fourier transform (FT) of the AFM images are shown in Fig. 4 . Brightness in the 2D-PS images represents the intensity of frequency components in the surface image. The spatial frequency domain ranges from –(2Δ)⁻¹ to + (2Δ)⁻¹ [15], where Δ is the sampling interval. In our case, the sampling interval (the AFM scan step) has been calculated as

 

Fig. 3 AFM images (6μm × 6μm) scanned from the sample #1 etched to different depths (a) 55 nm, (b) 110 nm, and (c) 220 nm. The height scale is shown at the bottom of each image.

Download Full Size | PDF

 

Fig. 4 2D-PS images of AFM images scanned from the sample #1 etched to different depths (a) 55 nm, (b) 110 nm, and (c) 220 nm, and these three images have the same brightness scale. (d) the logarithmic spectra of 1D-PS integrated from the 2D-PS images, and the curve showing the function dependent on the spatial frequency to the power of −3.3.

Download Full Size | PDF

The Raman spectra of 1 µM R6G obtained on the glass substrate (the sample #1 etched to different depths) are shown in Fig. 5(a) . The fluorescence backgrounds have been subtracted in each spectrum. Figure 5(b) gives the average Raman intensity and the relative standard deviation (RSD) of the Raman intensity from a set of Raman results on the substrate of the sample #1 etched to different depths. These results match well with the TEM image. At the depth of 55 nm, few particles are exposed and only weak Raman peaks appear. The RSD of the Raman intensity is as high as 30% due to a poor signal-to-noise ratio. At the depths of 110 nm and 220 nm, dense particle clusters are distributed at the glass surface and clear Raman signals are obtained. The corresponding RSD is about 10%, showing good reproducibility as a SERS-active substrate.

 

Fig. 5 (a) Raman spectra, (b) average Raman intensity and RSD of Raman intensity of 1µM R6G observed from the sample #1 etched to different depths 55 nm, 110 nm, and 220 nm.

Download Full Size | PDF

4. Discussion

4.1 The formation of Ag NPs

Conventionally, galvanic replacement reaction happens when placing one sacrificial metal in an aqueous solution containing another less active metal ion. The driving force of the reaction is the electrical potential difference between these two metals. For example, pre-synthesized Ag nanostructures can be oxidized by HAuCl₄ since the standard reduction potential of AuCl₄ ^-/Au (0.99 V versus the standard hydrogen electrode, SHE) is higher than that of Ag⁺/Ag (0.8 V versus SHE) [16]. In a commercial sacrificial metal-based replacement reaction, Al powder has also been used to reduce the desired metal salt precursors, because it is commercially available, cost-effective and very reactive [17].

In our experiments, as shown in Fig. 6 , the deposited Al layer is used as the sacrificial template because the reduction potential of Al³⁺/Al (−1.676 V versus SHE) [17] is much lower than that of Ag⁺/Ag. The difference between the traditional galvanic replacement reaction and our process is that, Ag⁺ ions are not dissolved in a salt precursor but distributed in the glass matrix. However, at an elevated temperature, the glass can be considered as a solid electrolyte. The structure of glass is not moving significantly, and only monovalent ions such as Ag⁺ ions are relatively free to move in the glass structure. This enables the galvanic replacement reaction including the oxidization of Al³⁺ ions and the reduction of Ag⁺ ions.

 

Fig. 6 Proposed mechanism for the formation of Ag NPs during K⁺ ion exchange with an Al layer.

Download Full Size | PDF

In the optical absorption spectra in Fig. 1, when there is no Al layer during the K⁺ ion exchange (sample #3), there is no plasmon peak in the spectrum, implying no formation of Ag NPs. In our process, the Al layer is responsible for providing the electrons to reduce Ag⁺ ions into metallic Ag. Therefore, under the experimental conditions of K⁺ ion exchange without an Al layer, there is no source of electrons, and thus, no formation of Ag NPs. For the other two samples (samples #1 and #2) in Fig. 1, Ag NPs are formed because for both samples the last process is implemented with an Al layer at an elevated temperature. It satisfies the conditions of galvanic replacement reaction: i) using the Al layer as the sacrificial template, ii) the glass can be considered as a solid electrolyte at an elevated temperature.

In addition, the optical absorption results show that more particles are obtained by immersing the sample into a KNO₃ salt (sample #1) than by just annealing it in the air at an elevated temperature (sample #2). This is due to an ionic current of Ag⁺ ions arisen by the electrical potential differences among the salt melt, the glass and the Al layer. During the K⁺ ion exchange process, the Al layer has the trend to release the electrons to Ag⁺ ions in glass in the galvanic replacement reaction, and the left Al³⁺ ions intend to go into KNO₃ salt. Hence, the Al layer is at a negative potential (V_am) with respect to the melt. Meanwhile, another electrical potential (V_gm) arises between the glass and the melt, as Ag⁺ and Na⁺ ions are more active and mobile than K⁺ ions; more Na⁺ or Ag⁺ ions are diffusing out of the glass than K⁺ ions are diffusing in. This electrical potential difference will lead to an ionic current flowing in glass and drive more Ag⁺ ions moving closer to the Al layer. For the sample annealed in the air at an elevated temperature with the Al layer (sample #2), the reduction of Ag NPs happens only because of the galvanic replacement reaction. This process will slow down or even stop when the system attains equilibrium. For the sample K⁺ ion-exchanged with the Al layer (sample #1), there is an additional Ag⁺ ion flow from the deeper depth of glass to the interface between the glass and the Al layer. Also, since Ag⁺ ions lost from glass can be compensated by new ions such as K⁺ ions from the salt melt, this current can keep flowing. This provides a continuous source of Ag⁺ ions and promotes the formation of Ag NPs.

4.2 The analysis of AFM images and Raman spectra

AFM images present 3D-data of the surface topography of the scanned sample. Previously, some methods have been utilized to quantitatively evaluate AFM images, such as standard deconvolution techniques [18], a statistical analysis of surface roughness [19], a discrete wavelet transform [20], and the Fourier transform [21]. Some of them are used to evaluate the AFM probe by comparing AFM images of the same sample scanned with different tips [21]. On the other hand, under the same experimental conditions in AFM measurements, they can also be used to study the surface features and to distinguish similar surface structures [20].

In our experiments, the AFM images shown in Fig. 3 are scanned by the same AFM system with the same AFM tip, and all the experimental parameters are maintained the same. In order to compare the differences of these three samples, the RMS roughness and the FT analysis are discussed. The RMS is a statistical measure of varying height quantities in surface morphology. It is a simple calculation to get information of vertical variations in AFM images. Higher value of RMS roughness obtained when etched to a deeper depth indicates that there are more rough features on the surface. Moreover, the FT is a powerful technique for analysis of surface images, and it can characterize the features identified by a frequency analysis of the AFM images. Firstly, as shown in Fig. 4(d), the power spectrum signal is higher almost throughout the range of spatial frequency when etched to a deeper depth, indicating that the height variation is larger. This is in a good agreement with the height scale and the RMS roughness given by the AFM images. Secondly, because of the limitation of sampling, the maximum spatial frequency is 21.33 μm⁻¹ and the corresponding smallest lateral size (SLS) that can be resolved in AFM images is 23.43 nm. Even with a better sampling interval, the SLS would still be limited by the highest resolution of AFM images determined by the size of AFM tips (usually around 10 nm). It means that it is impossible to resolve the size of the individual particles in the AFM images in our experiment (5-10 nm observed by the TEM image). However, Fig. 4(d) shows that the power spectra follow quite well proportionality to f ^-3.3. Such a power-law function is scale-invariant, indicating that the power spectrum might keep this trend when going to higher frequencies, up to the limit given by the particle sizes. Actually, this suggests a self-similar fractal structure with fractal dimension around 1.85, and fractal aggregates of metal nanoparticles are known to be efficient for SERS [22]. Finally, it can be found in Fig. 4(d) that, in the frequency range of 0 to 5 μm⁻¹, the intensity of the 1D-PS is much larger when etched to 110 nm and 220 nm than that of the sample etched to 55 nm. The spatial frequency is relatively low, and the corresponding lateral size is as large as 100 nm. According to the size of these features, this frequency might be related to the clusters of particles. This is also indicated by the AFM images, as larger features of the clusters are observed when etched to 110 nm and 220 nm, and the surface features are closer to the average particle size when etched to 55 nm.

From the analysis, we find that when etched to a deeper depth, more roughness is obtained, the particles are separated with deeper and narrower gaps, and more clusters of particles are exposed. These features might have already been embedded in the glass when forming the particles during the K⁺ ion exchange, as is actually suggested by the TEM images, or they could also be to some extent produced during the etching. All these surface features revealed after the etching play an important role in SERS performance. Comparing the AFM analyses to the Raman spectra, when etched to 55 nm, the surface shows less roughness and clusters of particles thus giving weaker enhancement; and when etched to 110 nm and 220 nm, better roughness and more clusters are obtained resulting in stronger enhancement.

5. Conclusion

In conclusion, Ag nanoparticles embedded in glass are prepared by a two-step ion exchange process. Optical absorption spectra of the samples prove the important role of the evaporated Al layer and the KNO₃ salt. TEM images clearly reveal the distribution of the particles, give their size and shape, and suggest the narrow gap between the particles which is an important feature for SERS performance. AFM images are analyzed and their relationship with Raman spectra is discussed. The formation of the particles is initialized by galvanic replacement reaction and further enhanced by the electrolytic deposition. The mechanism is quite different from the metal particle formation in our previous process [13,14], which was driven by potential differences between the mask, glass and the salt melt within distances in the micron range. In the two-step process presented here, the two glass surfaces in contact with the salt melt and with the mask, respectively, are separated by a 0.5 mm thick glass substrate acting as a solid electrolyte. Also, the Ag nanoparticle aggregates are formed uniformly under the metal mask which has dimensions in centimeters. The results demonstrate the formation of the particles and their SERS application.

The proposed method is simple and low-cost for large scale fabrication of Ag NPs in high-quality glasses with no need for the complicated or expensive equipment. By separating ion exchange into two steps, it significantly improves the control of the particle formation, increases the versatility of the SERS applications of the particles, and provides a significant improvement in the potential developments based on this technique. The particles are uniformly formed over broad areas under the Al film, and the SERS signal can be measured over the entire sample without the need to focus on any particular position or the need to use high magnification. This is a very important feature in practical applications of SERS-active substrates. Moreover, the method enables the patterning of the particle patterns when applying a photolithographic mask in either the Ag⁺ ion exchange or the K⁺ ion exchange process. The geometrical characteristics of the sensor areas, such as the shape and the dimensions, are totally determined by the mask pattern. In principle, any shape and dimensions are possible, which makes the development of integrated sensor chips much easier with increased flexibility in the integration of planar sensor devices.