Role of Graphene in Constructing Multilayer Plasmonic SERS Substrate with Graphene/AgNPs as Chemical Mechanism—Electromagnetic Mechanism Unit

Graphene–metal substrates have received widespread attention due to their superior surface-enhanced Raman scattering (SERS) performance. The strong coupling between graphene and metal particles can greatly improve the SERS performance and thus broaden the application fields. The way in which to make full use of the synergistic effect of the hybrid is still a key issue to improve SERS activity and stability. Here, we used graphene as a chemical mechanism (CM) layer and Ag nanoparticles (AgNPs) as an electromagnetic mechanism (EM) layer, forming a CM–EM unit and constructing a multi-layer hybrid structure as a SERS substrate. The improved SERS performance of the multilayer nanostructure was investigated experimentally and in theory. We demonstrated that the Raman enhancement effect increased as the number of CM–EM units increased, remaining nearly unchanged when the CM–EM unit was more than four. The limit of detection was down to 10−14 M for rhodamine 6G (R6G) and 10−12 M for crystal violet (CV), which confirmed the ultrahigh sensitivity of the multilayer SERS substrate. Furthermore, we investigated the reproducibility and thermal stability of the proposed multilayer SERS substrate. On the basis of these promising results, the development of new materials and novel methods for high performance sensing and biosensing applications will be promoted.


Introduction
Surface-enhanced Raman scattering (SERS) is a powerful detection and analysis tool that can detect low-concentration molecules with an enhancement factor of 10 14 or higher [1][2][3][4][5], exhibiting wide application in biomedicine [6], catalytic monitoring [7], environmental analysis [8], food safety [9], and other fields. Although the complete mechanism of SERS is still under debate, two enhancement mechanisms are widely accepted to explain the SERS effect: electromagnetic mechanism (EM) and chemical mechanism (CM) [10,11]. It is generally believed that hot spots are the main contribution for the EM [12,13]. As the main component of the hot spot, Au and Ag nanostructures with strong

Preparation of the SERS Substrate with Different CM-EM Units
The quartz substrates were initially cleaned with ultrasonic processing in acetone, alcohol, and deionized water for half an hour separately to remove the surface contamination. Ag colloids were synthesized according to the method described by our previous work [28]. The graphene was synthesized on copper foil by chemical vapor deposition (CVD), as reported in detail in our previously study [29]. Next, we dissolved FeCl3 (81 g) in 300 mL of deionized water under stirring. Then, the copper foil after the deposition of graphene was placed in the FeCl3 solution for 8 h to etch the Cu foil away. After completely removing Cu, we used a clean quartz substrate to gently drag the graphene film out to deionized water to keep it floating and to soak it for 10 min. After the move to clean ionized water and soaking for 10 min, we repeated the process three times to ensure that the residual ferric chloride was removed. The AgNP plasmonic films were prepared by self-assembly at the liquid-liquid interface using the solution of AgNPs with hexane and ethyl alcohol with a volume ratio of 2:1:1. A total of 10 mL Ag colloids and 5 mL of hexane were mixed. Then, 5 mL ethanol was gradually dropped into the above solution. With the increase of the added ethanol, a layer of AgNP film could be clearly seen covering the hexane-Ag colloid interface. After hexane was evaporated, the AgNP film was obtained and removed with a graphene-based quartz substrate. Repeating the step as shown in Figure 1, the multilayer substrate with different CM-EM units can be fabricated.

Characterization
The morphologies of the prepared samples were characterized by scanning electron microscope (SEM, ZEISS Sigma500) with energy dispersive spectrometer (EDS). All the SEM images in the paper were measured at 3 kV voltages. The more detailed morphology and composition were characterized by transmission electron microscope (TEM, JEM-2100F) and high-magnification transmission electron microscopy (HRTEM).
The absorbance spectra of the AgNPs were obtained with a spectrophotometer (PERSEE, TU-1900) using the absorbance mode. The spectrum range was typically from 300 to 800 nm. SERS spectra were detected by Raman spectrometer (Horiba HR Evolution 800) with a laser wavelength of 532 nm. The laser excitation energy and spot were 0.48 mW and 1 μm, respectively.

Characterization
The morphologies of the prepared samples were characterized by scanning electron microscope (SEM, ZEISS Sigma500) with energy dispersive spectrometer (EDS). All the SEM images in the paper were measured at 3 kV voltages. The more detailed morphology and composition were characterized by transmission electron microscope (TEM, JEM-2100F) and high-magnification transmission electron microscopy (HRTEM). The absorbance spectra of the AgNPs were obtained with a spectrophotometer (PERSEE, TU-1900) using the absorbance mode. The spectrum range was typically from 300 to 800 nm. SERS spectra were detected by Raman spectrometer (Horiba HR Evolution 800) with a laser wavelength of 532 nm. The laser excitation energy and spot were 0.48 mW and 1 µm, respectively. Throughout the experiment, the diffraction grid was set as 600 gr/mm and the integration time was set as 8 s. The laser light was coupled through an objective lens of 50×.

FDTD Simulations
The electromagnetic field distributions were simulated with finite-difference time domain (FDTD) simulation. In theoretical simulations, the absorption boundary condition is the perfect matching layer (PML). Cross-stack spherical AgNPs that were obtained from SEM with diameters of 57 nm and a 3 nm gap were stacked along the x-, y-, and z-directions, and the 532 nm incident wave polarized along the x-direction was set. Considering that graphene does not remain flat when transferred to AgNPs, the graphene in simulation model was set as a curved surface. The numerical data of the refractive index of Ag and graphene were obtained from the reported works [30][31][32].

Results and Discussion
Graphene can provide an excellent nano-platform for the manufacture of SERS active substrates due to its two-dimensional (2D) planar structure. Figure 2a presents the Raman spectrum of the single layer graphene, where the D, G, and 2D peaks of graphene located near 1350, 1580, and 2700 cm −1 can be observed, respectively. The intensity ratio of the G peak to the 2D peak was about 0.5; the full width at half maximum of the 2D peak was less than 35 cm −1 ; and the intensity of the D peak was weak, implying that high-quality and monolayer graphene was successfully transferred to the substrate [33,34]. To investigate the role of graphene for AgNP large-area depositing, we performed a contrast experiment on the substrate with and without the support of graphene in Figure 2b. It can be seen clearly the AgNPs distributed more uniformly on the region with graphene support compared with the region without graphene. Moreover, the AgNPs supported by graphene were closely arranged, forming a large-area stable and uniform film structure. Different from the bare substrate, the AgNP film can be facially transferred to the flexible graphene film substrate, and it tightly adheres to the graphene film surface via van der Waals force [35]. Figure 2c shows the TEM image of the AgNPs with a diameter of about 57 nm. Moreover, the HRTEM inset in Figure 2d was measured to further analyze the details of AgNPs, where the two distinct inter-layer spacings with values of 0.24 nm were agreement with the (111) plane of AgNPs. The absorption peak of AgNPs was approximately 437 nm, as shown in Figure 2d, which matched well with the incident light and was beneficial for generating a stronger local electric field.
SEM was carried out to investigate the surface morphology of the prepared multilayer substrate with different CM-EM units. The image shown in Figure 3a exhibits the self-assembled AgNPs' firmly ordered arrangement on the surface of the graphene, which is beneficial for the uniform hot spots. The SEM images shown in Figure 3b-f carefully chosen at the boundary of the multilayer substrate clearly present the distribution of AgNPs above and below the graphene. It can be clearly seen that each layer self-assembled AgNP film was compact and uniform, with no visible stacking, and the AgNPs covered with graphene were relatively dark. It is worth mentioning that when the number of units was more than three, the difference in the distribution of AgNPs was almost invisible. The EDS spectrum shown in Figure 3g provides further credible evidence for the successfully fabrication of the multilayer substrate.     To investigate the effect of the CM-EM unit for the SERS performance of the multilayer SERS substrate, we used R6G with the concentration of 10 −6 M as the probe molecule. Figure 4a shows the SERS signal of R6G collected from the CM-EM unit substrate with different numbers of units. All the typical Raman peaks of the R6G at 613, 774, 1365, 1510, and 1652 cm −1 were observed, which was consistent with previous reports [36]. Figure 4c plots the intensity of the SERS signal at 613, 774, and 1365 cm −1 for R6G. It can be observed that the Raman signal of R6G firstly increased and then decreased as the number of units increased, reaching a maximum intensity in the four units. As the CM-EM units were successively stacked, the strong plasmnoic coupling between the AgNPs and the coupling between the graphene and the AgNPs were dually excited, and were further synergistically enhanced in the unit. In addition, the introduced graphene also benefitted from the amplification of the SERS signal with the assistance of CM. However, the reason for the intensity of R6G SERS signal remaining almost invariable for more than four units was potentially due to the laser penetration depth, wherein an upper limit for the coupling between different units exists.
To better understand this, we built a simulation to explore electric field distributions for multilayer substrate with 1-6 units. As shown in Figure S1, we observed that the intensity of the maximum electric field increased continuously as the number of units increased, and the maximum electric field intensity tended to be stable for more than four units, which indicates that the multilayer structure existed in an upper limit for coupling. However, the Raman signal of R6G reached its maximum value on four units in the experiment, which was ascribed to the inconsistent penetration depth of the incident light in the simulation and experiment. It is worth noting that due to the influence of the curved surface of the graphene, the particles may not be round. On the other hand, the interparticle distances determined the coupling strength. The distance between the lower unit and the higher unit increased as the units were superimposed. Therefore, each of the EM units were plasmonically decoupled between higher and lower units. It should be emphasized that the molecules were adsorbed on the top unit of the multilayer SERS substrate due to the presence of the graphene. However, it should be recognized that the electromagnetic enhancement effect of the AgNPs would not be shielded by the interlayer graphene, and the charge transfer between graphene and probe molecule could produce CM enhancement. In order to prove the universality and reliability of this conclusion, we repeated the experiment with CV and found that the same conclusion can be drawn to prove our point (the relevant results are shown in Figure 4b,d). Interestingly, we also observed the same phenomenon in Figure 4b-the intensity of the SERS signal also increased up to four units stacked. Then, the intensity decreased and finally remained almost unchanged when there were more than four units. The variation of average intensity of CV bands at 914, 1175, and 1619 cm −1 relative to the number of the units was also exhibited, as shown in Figure 4d. The transmittance of the multilayer substrate with different units was observed, as shown in Figure 4e. We can observe clearly that the transmittance of multilayer substrate remarkably decreased as the number of units increased, which indicates the attenuation of the incident light from layer to layer and was the reason for the different SERS activity of the substrate with different CM-EM units. On the other hand, the effective penetration depth of incident light is also another reason for the phenomenon of the Raman signal of R6G not continuing to rise when there were more than four units. The laser penetration depth was proportional to the wavelength, and different wavelengths of light penetrated the different depths of substrate. The penetration depth was estimated by the following equation , where λ 1 is the wavelength of the light beam in medium 1; i is the angle of incidence; and n 1 and n 2 are the refractive indices of medium 1 and medium 2, respectively. Because the refractive index of the sample was not only related to the incident wavelength but also to the degree of light absorption of the sample, the refractive index of the sample changed drastically where strong absorption occurred. Therefore, the maximum penetration depth of the 532 nm laser was about 266 nm. Moreover, there was also a maximum collection depth for the Raman detector. Combined with the transmittance of the multilayer structure, the loss of gain for more than four units could result from the lower efficiency of light reaching the lower unit. Hence, the maximum enhancement effect was achieved on the four units.
Above all, the multilayer substrate with four unit possessed the optimal SERS performance, which was maintained for the further research throughout the following experiments.
transmittance of the multilayer structure, the loss of gain for more than four units could result from the lower efficiency of light reaching the lower unit. Hence, the maximum enhancement effect was achieved on the four units. Above all, the multilayer substrate with four unit possessed the optimal SERS performance, which was maintained for the further research throughout the following experiments. To further investigate the Raman properties of multilayer nanostructure, we directly fabricated multilayer substrate with four CM-EM units. R6G and CV molecules were successively diluted with water solution from concentration of 10 −6 M to 10 −14 M and from 10 −5 M to 10 −12 M, respectively. Then, 2 μL diluted solution was dropped on the surface of the substrate and dried up naturally before SERS detection. The substrate exhibited superior detection capacity for analytes with ultra-low concentration. Raman spectra of R6G and CV molecules collected from the substrate are shown in Figure 5a  To further investigate the Raman properties of multilayer nanostructure, we directly fabricated multilayer substrate with four CM-EM units. R6G and CV molecules were successively diluted with water solution from concentration of 10 −6 M to 10 −14 M and from 10 −5 M to 10 −12 M, respectively. Then, 2 µL diluted solution was dropped on the surface of the substrate and dried up naturally before SERS detection. The substrate exhibited superior detection capacity for analytes with ultra-low concentration. Raman spectra of R6G and CV molecules collected from the substrate are shown in Figure 5a  The enhancement factor (EF) is an effective method to evaluate the contribution of the proposed multilayer SERS substrate from the enhanced Raman spectra of R6G molecules. The EF is estimated according to the comparison of the limit Raman signal collected from multilayer substrate with four CM-EM units that were obtained from ordinary substrate of probe molecules. The EF was estimated by the following equation [37]: EF = (I SERS × N SiO 2 )/(I SiO 2 × N SERS ), where I SERS , I SiO 2 , N SiO 2 , and N SERS represent the intensity of the SERS signal, the Raman signal intensity obtained from SiO 2 , the number of analyte molecules within the laser spot on the SiO 2 substrate, and the number of molecules within the laser spot on the SERS substrate, respectively [38]. To scientifically guarantee the results, we chose the 10 −13 M R6G solution as the limit concentration for EF calculation. Similarly, the 10 −10 M CV was also chosen as the limit concentration for the calculation of the enhancement factor. The reason for this choice is because the concentration belongs to the same level of precision (around 6%) in relative standard deviation (RSD), as shown in Figure S2. The Raman signals of R6G and CV with concentration 10 −2 M obtained from SiO 2 substrate for reference are shown in Figure 6a. The EF of the multilayer SERS substrate was calculated to be 5.86 × 10 10 for R6G and 9.62 × 10 8 for CV, which is better than that of the previously reported SERS substrates with similar multilayer structure [3,[39][40][41][42][43][44]. The homogeneity and stability are also significantly essential for SERS substrates. Figure 6b shows the SERS mapping at 613 cm −1 peak of 10 −6 M R6G on multilayer substrate with four CM-EM units in an area of 20 × 20 µm 2 . The step-size of the SERS spectra collection is 2 µm. Clearly, it can be concluded that there was good homogeneity from the relatively smooth and uniform color distribution, with only a small dark region. Reproducibility is a significant parameter for practical application. Figure 6c presents the SERS spectra of the R6G with a concentration of 10 −6 M collected from 10 different batches of multilayer substrate with four CM-EM units, with each spectrum being the average of the 20 random spots collected from each substrate. The spectra of R6G were greatly well consistent with each other and the intensities for various peaks only fluctuated quite mildly. As demonstrated in Figure 6d, the intensities of the three main R6G characteristic peaks at 613, 774, and 1365 cm −1 were collected and the RSDs were 3.52%, 6.22%, and 5.92%, respectively. A similar phenomenon was observed for CV with a concentration of 10 −5 M in Figure 6e. Moreover, the RSDs of CV at 914, 1175, and 1619 cm −1 peaks were 3.26%, 3.01%, and 4.46%, respectively, as shown in Figure 6f. In addition, the RSD on low concentrations of R6G and CV is shown in Figure S2. The RSDs of R6G at 10 −12 M, 10 −13 M, and 10 −14 M were calculated as the values of 5.83%, 6.23%, and 6.87%, respectively, and the RSDs of CV at 10 −10 M, 10 −11 M, and 10 −12 M were 5.83%, 6.23%, and 6.87%, respectively. The RSDs of R6G and CV were less than 10% at low concentrations, which indicates the excellent capability of substrate in SERS detection. These results demonstrate the excellent reproducibility of the multilayer SERS substrate. The expected uniformity and reproducibility could be ascribed to the well-distributed AgNPs and the existence of graphene films in CM-EM units. The well-arranged AgNPs can provide dense electromagnetic hot spot distribution from different units, and graphene can make molecules be well distributed around hot spots, allowing molecules to contact well with the substrate in consequence. Furthermore, the plasmonic coupling between in-plane AgNPs and graphene in each unit can be generated under the motivation of the laser. Additionally, the colloidal self-assembly method was low-cost, simple, and used repeatable techniques that could contribute to excellent reproducibility and make the multilayer SERS substrate with CM-EM unit possess great potential for real applications.
In order to further understand the SERS activity of the multilayer SERS substrate with CM-EM units, we carried out a control experiment to measure Raman spectra on different types of substrate, as depicted in Figure 7a,b. All typical Raman peaks of R6G are shown in Figure 7a, and it is evident that intensities of Raman spectra on the four CM-EM unit substrates were much stronger than others. Figure 7b indicates the intensity of the SERS signal at 613, 774, and 1365 cm −1 on the different types of substrate, and the intensities on the four CM-EM unit substrates were about 1.4-3 times stronger than others. As shown in the SEM image in Figure 7c, the arrangement of AgNPs on four EM units was all in a muddle due to the absence of graphene, which can lead to the molecules keeping away from hot spots, whereas the reason for the enhancement being relatively weak was due to the poor binding force between the AgNPs and the detection molecule. Moreover, the molecule was difficult to adsorb on the metal surface, resulting in worse SERS signal stability. As shown in Figure 7d, the distribution of AgNPs on the four EM-CM unit substrates was similar to that on the four CM-EM unit substrates, which proves that there was no particularly large difference in hot spots among AgNPs on different units. The difference for the SERS activity for the four EM-CM units and the four CM-EM units may have been introduced by the top graphene layer. To further identify the perfect SERS behavior of the multilayer substrate with four CM-EM units, we analyzed the electric field properties of these substrates on the basis of the finite difference time domain method (FDTD). The simulation set-up is shown in Figure 7e. The x-z views of the electric field enhancement variation (E/E 0 ) of the hot spots versus three types of multilayer substrate are shown in Figure 7f. Note that the intensity of the electric field varied with different structures. The intensity of the maximum electric field continuously increased with the addition of graphene. In order to ensure the reliability of the model, we show in Figure S3 the electric field of the particles between layers horizontally offset with respect to the next layer. We note that the field enhancement showed almost no change if the particles in the layer were horizontally offset along the xand y-directions with respect to the next layer. The simulation result consisted of our experimental analysis that graphene can effectively enhance the electromagnetic coupling effect. It can be observed clearly that the prominent electric fields were all generated at the nanogaps region produced between the AgNPs, and the electric field strength was further enhanced after adding graphene. We also note that the results in the maximum electric field were not strictly consistent with the experimental results. As a matter of fact, the SERS effect is synergic effect of EM enhancement and CM enhancement and it is incorrectly directly related to the maximum electric field.  Figure 8b was dropped to below about 35% of the original intensity. It is evident in Figure 8c that the intensities of R6G spectra at 200 °C on the four-unit CM-EM substrate were relatively stable up to 200 °C, with little change in the intensity of peak. The measured spectra signal on four EM-CM unit substrates decreased more than that of four CM-EM  Figure 8b was dropped to below about 35% of the original intensity. It is evident in Figure 8c that the intensities of R6G spectra at 200 • C on the four-unit CM-EM substrate were relatively stable up to 200 • C, with little change in the intensity of peak. The measured spectra signal on four EM-CM unit substrates decreased more than that of four CM-EM unit substrates because of the lack of protection of the top graphene. Compared with Figures 8d and 7c, we found that the substrate with four EM units after heating underwent tremendous changes. The agglomeration phenomenon occurred in the AgNPs, which resulted in a large loss of hot spots. Therefore, the destruction of the hot spots between the units when the agglomeration occurred was the main reason for the substantially reduced Raman signal. As shown in Figure 8e for the case of the four EM-CM unit substrates, the existence of the graphene layer improved the deformation resistance of the multilayer substrate, and the AgNPs will not agglomerate at a large scale. Figure 8f reveals that the introduction of graphene as the top shielding layer can further alleviate the degree of particle aggregation. The measured spectra signal on four EM-CM unit substrates decreased more than that of four CM-EM unit substrates because of the lack of protection of the top graphene. The phenomenon can be explained by the fact that graphene can provide structural support in the longitudinal direction, making the structure tougher and not easy to deform, as shown in Figure 8f. Moreover, graphene can also isolate the contact between AgNPs and air and achieve an anti-oxidation effect, which can broaden the practical application under harsh conditions. Moreover, graphene on the CM-EM unit substrate is a flexible material that can also expand as the AgNPs move after heating, making the CM-EM structure unbroken, which is conducive to maintaining the electric field coupling effect. Nanomaterials 2020, 10, 2371 12 of 15 unit substrates because of the lack of protection of the top graphene. Compared with Figure 8d and Figure 7c, we found that the substrate with four EM units after heating underwent tremendous changes. The agglomeration phenomenon occurred in the AgNPs, which resulted in a large loss of hot spots. Therefore, the destruction of the hot spots between the units when the agglomeration occurred was the main reason for the substantially reduced Raman signal. As shown in Figure 8e for the case of the four EM-CM unit substrates, the existence of the graphene layer improved the deformation resistance of the multilayer substrate, and the AgNPs will not agglomerate at a large scale. Figure 8f reveals that the introduction of graphene as the top shielding layer can further alleviate the degree of particle aggregation. The measured spectra signal on four EM-CM unit substrates decreased more than that of four CM-EM unit substrates because of the lack of protection of the top graphene. The phenomenon can be explained by the fact that graphene can provide structural support in the longitudinal direction, making the structure tougher and not easy to deform, as shown in Figure 8f. Moreover, graphene can also isolate the contact between AgNPs and air and achieve an anti-oxidation effect, which can broaden the practical application under harsh conditions. Moreover, graphene on the CM-EM unit substrate is a flexible material that can also expand as the AgNPs move after heating, making the CM-EM structure unbroken, which is conducive to maintaining the electric field coupling effect.

Conclusions
In conclusion, we proposed a multi-layer hybrid SERS structure with graphene as a chemical mechanism (CM) layer and Ag nanoparticles (AgNPs) as an electromagnetic mechanism (EM) layer, forming a CM-EM unit. We investigated the SERS activity of the multilayer with different types of units experimentally and in theory. We demonstrated that the multi-layer structure with a CM-EM unit can serve as a highly sensitive, uniform, and stable SERS substrate. Meanwhile, the reproducibility and thermal stability were also measured from the multilayer SERS substrate. The proposed strategy for self-assembly of multilayer substrate can pave the way towards SERS, and it will possess great potential in the field of application for biosensing.

Supplementary Materials:
The following are available online at www.mdpi.com/2079-4991/10/12/2371/s1. Figure S1: (a) The x-z views of electric field distribution of different number of CM-EM unit substrates at 532 nm wavelength. (b) Electric field enhancement (E/E0) for substrate with different CM-EM units. Figure S2: The

Conclusions
In conclusion, we proposed a multi-layer hybrid SERS structure with graphene as a chemical mechanism (CM) layer and Ag nanoparticles (AgNPs) as an electromagnetic mechanism (EM) layer, forming a CM-EM unit. We investigated the SERS activity of the multilayer with different types of units experimentally and in theory. We demonstrated that the multi-layer structure with a CM-EM unit can serve as a highly sensitive, uniform, and stable SERS substrate. Meanwhile, the reproducibility and thermal stability were also measured from the multilayer SERS substrate. The proposed strategy for self-assembly of multilayer substrate can pave the way towards SERS, and it will possess great potential in the field of application for biosensing.

Supplementary Materials:
The following are available online at http://www.mdpi.com/2079-4991/10/12/2371/s1. Figure S1: (a) The x-z views of electric field distribution of different number of CM-EM unit substrates at 532 nm wavelength. (b) Electric field enhancement (E/E 0 ) for substrate with different CM-EM units. Figure S2: The Raman signal intensity distribution of (a) R6G at 613 cm −1 and (b) CV at 914 cm −1 on multilayer substrate with four CM-EM units from 10 −12 M to 10 −14 M and from 10 −10 M to 10 −12 M, respectively. Figure S3: The x-z views of electric field distribution of substrate shifted horizontally.