High-resolution SEM imaging characterized the physical characteristics of the substrates as presented in figure 1. Although SEM offers only two-dimensional information about the morphology of the substrates, it is a fast and accessible approach that can satisfy the resolution requirements of features as small as tens of nanometers. Furthermore, the third dimension can be studied by measurements at an angle. Atomic force microscopy, an alternative to SEM, can provide three-dimensional images. However, this technique suffers from artifacts like enveloping due to narrow but deep crevices and abrupt profile changes [30].

Figure 1. Top-view SEM images of (a1) PiCO Au, (b1) ATO ID Ag, (c1) Hamamatsu Au, (d1) SERSitive Ag, and (e1) Silmeco Au SERS substrates' nanostructures. The dashed red squares (b2)–(e2) highlight a magnified section of the top-view images. Beneath the squares, the polished cross-sections are presented (a3)–(e3). (f) summarizes the characteristics of the substrate types.

Download figure:

Standard image High-resolution image

Each substrate is randomly structured to an extent. For instance, figure 1(a) has a grid-like structure with random feature sizes, making it less random than the randomly deposited nanoparticles in figure 1(d). Thus, the qualitative extent of order or randomness is one of the factors investigated for the SERS substrates here. Moreover, every nanostructure has specific building blocks, here called features, that are primarily responsible for the enhancement. Consequently, the additional factors investigated for the substrates here are the features' lateral size, height, and the size of the gaps between them. Table 1 provides an overview of the mentioned characteristic factors for the investigated SERS substrates. The feature-related factors are visualized in figure 1(f).

Table 1. The summary of the commercial SERS substrates' surface structure characteristics as measured by SEM.

SERS substrateFabrication technologyFeature size (nm)Gap size (nm)Feature height (nm)Structural orderReferencesPiCONano-transfer printing40 ± 1031 ± 933 ± 5Stripe pattern[31]ATO IDFemtosecond laser128 ± 52a14 ± 5101 ± 47bRandom—HamamatsuNanoimprinting123 ± 9225 ± 16133 ± 14Dot array—SERSitiveElectrodeposition189 ± 6787 ± 46370 ± 187Random—SilmecoReactive ion-etching119 ± 1779 ± 36646 ± 199Random[32]

a 1504 ± 622 nm and b 2371 ± 514 nm for the landscape.

Each substrate was fabricated using a different nanofabrication technology with a different structure and detection strategy. PiCO uses nano-transfer printing technology to create 3D structures using nanowires in silicon; the structure is then coated with gold and nanoparticles for increased sensitivity [31]. PiCO's strategy is to make the structures as small and ordered as possible aspiring to achieve high and stable enhancement. The nanoparticle decoration augments the substrate's enhancement capability. As presented in figure 1(a), this substrate has randomly shaped features and gaps arranged in a grid-like structure. The features have a mean lateral size of 40 ± 10 nm, gap size of 31 ± 9 nm, and height of 33 ± 5 nm. The smallest feature size is around 10 nm. Furthermore, ATO ID fabricates its SERS substrates in soda lime glass using a femtosecond laser nanofabrication technique resulting in stochastic structures that can be seen in figure 1(b). The substrate's structure can be viewed as a random distribution of small features in a roughened landscape. ATO ID's detection strategy is that the variety of feature sizes makes the substrate suitable for a wide range of excitation wavelengths. The small features can be as small as tens of nanometers and have a mean lateral size of 128 ± 52 nm, gap size of 14 ± 5 nm, and height of 101 ± 47 nm. The landscape's lateral features can be as large as several micrometers with a mean lateral size of 1504 ± 622 nm and an average height of 2371 ± 514 nm. It is important to note that the reported sizes are merely average values and that this structure is highly random. Moreover, Hamamatsu uses nanoimprint technology to create a highly ordered nanopillar array in silicon as viewed in figure 1(c). These highly ordered and repeatable structures can stimulate signal stability and repeatability. The mean lateral feature size is 123 ± 9 nm, the mean feature height is 133 ± 14 nm, and the mean gap size is 225 ± 16 nm. In comparison to the other substrates, the structure of this substrate is more uniformly ordered. Furthermore, SERSitive creates its SERS substrates through the electrodeposition of metallic nanoparticles on ITO glass. Figure 1(d) shows the random shapes and sizes of the deposited nanoparticles. The largest particles are micrometer-sized agglomerates, while the smallest ones are tens of nanometers. SERSitive's average lateral feature size is 189 ± 67 nm with a mean gap size of 87 ± 46 nm and feature height of 370 ± 187 nm. Beware that this substrate's structure is random and that these numbers are only an average quantification of its feature's dimensions. Finally, the 'SERStrates' of Silmeco are fabricated in silicon using maskless reactive ion etching featuring leaning pillars [32]. Silmeco's detection strategy is to increase the features' height in addition to minimizing the lateral feature sizes, to enable more hotspots. Silmeco promotes molecule trapping via its pillar-leaning feature. Figure 1(e) shows this substrate's randomly positioned pillars, with some conjoined by the coating. This substrate's pillars have a mean size of 119 ± 17 nm, gap size of 79 ± 36 nm, and height of 646 ± 199 nm.

BPEs four characteristic Raman peaks, 1020, 1198, 1606, and 1636 cm−1 are used to investigate the SERS performance metrics [33]. The measured SERS spectra and the peaks of interest of BPE are shown in figure 2. The peaks' slight deviation from the mentioned values, in the order of a few data points, is an artifact of averaging the map's spectra; point to point variations can add up to a peak maximum slightly shifted. Furthermore, the substrates that contained silver, e.g. ATO ID Ag, showed extra peaks due to silver's reactive nature, e.g. oxidation. However, these peaks are usually smaller than 100 counts and will be subdued with a strong analyte signal as in Silmeco Ag's case. Furthermore, the signal-to-noise ratio can be seen from the noise that is superimposed on the spectra; Silmeco Ag's spectrum is much smoother compared to ATO ID Ag as the signal is much stronger than the noise.

Figure 2. Raw SERS spectra of 50 µM BPE for the 5 substrate types averaged over a 15 µm × 15 µm map i.e. 81 spectra. Vertical dashed blue lines represent the peaks of interest.

Download figure:

Standard image High-resolution image

The first performance metric is signal stability which is the most important characteristic of a detector used for quantification [34]. Different surface morphologies can lead to different spatiotemporal signal instabilities depending on the number of hotspots covered with the exciting laser. Expanding the illuminated area on a structure increases the number of hotspots producing the signal. Thus, the signal's variations are averaged over a larger number of hotspots, and as such the signal becomes more reproducible. The same reasoning applies when comparing different structures: if a structure's number of hotspots is more than another's under the same illumination, its signal is relatively more stable. For instance, a 1 µm2 area sees a lot more variation in the case of ATO ID (figure 1(b)) compared to PiCO (figure 1(a)). The signal stability over time and area is imperative for quantitative analyses or SERS imaging, while a strong signal suffices for identification alone. To study the spatial signal stability, several maps were measured across the SERS active area. An instability metric was introduced as the STD of a characteristic peak scaled by its intensity over the map, i.e. a modified definition of the CV. To calculate this metric for a map, find the STD over the map and divide it by the map's average intensity. Figure 3(a) demonstrates the instability of the studied substrates for the characteristic peaks of BPE. The substrate with the highest relative structural order, Hamamatsu, has the lowest spatial instability, equal to 0.09% with a 0.01% map-to-map variation for the 1636 cm−1 peak. Silmeco Au comes second in stability but has a higher map-to-map variation than Hamamatsu because of the random position and height of its pillars. PiCO, Silmeco Ag, and SERSitive substrates have the same stability, although PiCO's structural order and features' dimensions give its map-to-map variation an edge over the more randomly ordered structures. Finally, ATO ID has the highest instability and map-to-map variation among the studied substrates. This can be associated with this substrate's random structure and relatively lower hotspot density. Moreover, temporal instability is demonstrated by the signal variations of a point over time. Figure 3(b) visualizes this instability as the CV for the different substrates. The CV values were averaged over the center points of the 3 measured windows. Although this is a temporal study, it inherently has traces of spatial instability due to the averaging. Nonetheless, the temporal instability was found to have an inverse relationship with the signal's strength. Moreover, BPE's peaks of interest have the same trend except for 1020 cm−1, as seen in figure 3. This peak is the weakest of the four and its signal-to-noise ratio is close to unity. This can be seen better when considering figure 2 and comparing ATO ID Ag to the other substrates. As this substrate's signal-to-noise ratio is relatively smaller and closer to the noise of the measurements, it is less stable over space and time.

Figure 3. The instability of the SERS signal over (a) the measured maps and (b) single point measurements for the 7 SERS substrates studied for a BPE concentration of 20 μM. The error bars demonstrate the coefficients of map-to-map variation.

Download figure:

Standard image High-resolution image

Understanding the spatiotemporal limitations of a SERS substrate helps with its further development or applications. To elaborate, PiCO has a better structural order compared to SERSitive, but by increasing the beam spot to spatially average more structural variations, SERSitive can have similar signal stability. However, that limits this SERS substrate's minimum resolution for SERS imaging applications. The temporal instabilities can be averaged out by longer acquisition times. Despite that, an advantage of using SERS substrates is faster acquisitions but large temporal variations hinder the minimization of the measurement time.

The enhancement capability of the substrates can be discussed through their EF. A higher EF translates to a stronger signal and higher sensitivity. Thus, the substrates' enhancement capability follows the same logic as the signal's spatiotemporal stability; different structural features lead to different enhancements. Furthermore, the higher the number and strength of the hotspots in the investigated area, the higher the enhancement. Figure 4 highlights the lateral distribution of EF values that are maximum normalized over the SEM-measured windows for the different substrate types. Hamamatsu which had the most ordered structure (figure 1(c)) has the highest lateral EF homogeneity i.e. lowest CV over the surface. The maximum EF for this substrate in the measured window is 1.2 × 106 ± 3.5 × 104, as seen in figure 4(f). Silmeco and SERSitive come second and third in EF's spatial homogeneity while they are first and second in maximum EF, respectively. PiCO and ATO ID have the highest spatial variation and the lowest EF within the substrates studied. Table 2 summarizes the statistics of the maps' EF values for two analyte concentrations.

Figure 4. Maximum normalized EF maps of the 3 SEM measured windows of (a1)–(a3) PiCO Au, (b1)–(b3) ATO ID Ag, (c1)–(c3) Hamamatsu Au, (d1)–(d3) SERSitive Ag, (e1)–(e3) Silmeco Au for the 1636 cm−1 BPE peak and 20 μM concentration. The bar plot, (f), shows the absolute maximum EF averaged over the 3 maps. The error bars show the map-to-map variation of the maximum EF. The vertical axis has a factor of 1000.

Download figure:

Standard image High-resolution image

Table 2. An overview of the maps' EF values' descriptive statistics calculated for the 1636 cm−1 peak of BPE. The errors represent the map-to-map variation. The reported number in this table have a factor of 1000.

 1 μM20 μMSERS substrateMean (a.u.)Maximum (a.u.)STDa (a.u.)Mean (a.u.)Maximum (a.u.)STD (a.u.)PiCO Au266 ± 20619 ± 9140 ± 455 ± 779 ± 810 ± 1ATO ID Ag165 ± 391029 ± 183358 ± 4218 ± 1074 ± 3220 ± 2Hamamatsu Au2878 ± 1114111 ± 192541 ± 22963 ± 71173 ± 3587 ± 9SERSitive Ag1145 ± 1062911 ± 520710 ± 1331450 ± 4422185 ± 645259 ± 110SERSitive AgAu2449 ± 3424917 ± 770953 ± 184190 ± 3393 ± 4267 ± 14Silmeco Au7261 ± 135811 737 ± 13451677 ± 1572669 ± 2163466 ± 314362 ± 83Silmeco Ag17 143 ± 481630 498 ± 95674046 ± 8212455 ± 6763740 ± 859463 ± 85

a STD: standard deviation over a 15 μm × 15 μm map.

The features' aspect ratio is found to be correlated to the SERS structures' hotspot density. The maximum EF (figure 4(f)) increases with the aspect ratio of the features. Furthermore, the CV over the surface was found to be correlated to the error bars representing the map-to-map variation of the maximum EF. While PiCO and ATO ID have a low EF in all maps and consequently a small error bar, Hamamatsu, SERSitive, and Silmeco's CV follow their structural order. In addition, silver's superior plasmonic activity with respect to gold can be seen by comparing the EF of Silmeco Ag vs Au. Another interesting behavior regarding SERS material is the concentration range in which the signal is enhanced. Comparing the SERSitive materials, one can see that SERSitive AgAu has a better EF than SERSitive Ag for 1 μM concentration and vice versa for 20 μM concentration. This behvaior's origin can be the additional Au nanoparticles in SERSitive AgAu, consequently the denser structure. A denser structure performs better in a lower concentration due to its higher number of hotspots. Because of SERSitive Ag's lower structural density and stronger hotspots, the intensity might be weaker for a low analyte concentration but increases more rapidly as the concentration increases. The relatively low intensity count for SERSitive AgAu in figure 2 despite the 50 μM analyte concentration arises from the same material-related behavior. Moreover, by changing the analyte from BPE to crystal violet with a 1 μM concentration, the maximum EF for Hamamatsu Au and Silmeco Ag are 6.7 × 103 ± 500 and 3 × 106 ± 6.3 × 105 respectively. It is also important to note that baseline correction strongly affects the results presented here. Without baseline correction, the maximum EF becomes 6.6 × 105 ± 2 × 104 and 1.7 × 107 ± 1.4 × 106, respectively for Hamamatsu Au and Silmeco Ag. To summarize, EF values can be increased by the choice of the performance analyte, baseline correction, analyte concentration, etc. However, this does not mean that the substrate's SERS performance has improved or is better in comparison. The SERS results presented here are not comparable to the commercial substrates' respective publications as the same methodology has not been used here. Finally, the reported performance is specific to the settings mentioned in subsection 3.2. Deviating from these settings could change the reported numbers, as illustrated in table 2.

An advantage of signal enhancement is the resulting improvement in detection sensitivity calculated through calibration plots of the analyte's characteristic peaks. To make a calibration plot, a wide concentration range is favorable as it encourages the inclusion of saturation and below-sensitivity concentrations. When the availability of analyte molecules is below the sensor's sensitivity, no peak shows up in the spectrum. The corresponding point in the calibration plot will be close to zero until the concentration is raised above the sensor's sensitivity and peaks start showing up. Then, the peak intensity will increase linearly with increasing concentration until the number of analyte molecules starts hindering the signal. From that point on, the changes in peak intensity become nonlinear. The rate of the changes becomes smaller as well i.e. the signal saturates. To summarize, the calibration plot's trend starts from almost zero, raises nonlinearly, becomes linear for a range of concentrations, and finally nonlinearly raises to saturation. This is a sigmoidal behavior. As demonstrated earlier, the spatial instability of a SERS substrate can be high, leading to a different concentration range for which the peak's intensity changes linearly. Hence, fitting a sigmoid function to the calibration plot can help to seek the linear range automatically. Figure 5(a) shows how fitting a sigmoid function to the calibration plot with a wide concentration range could help with the selection of the data points for LOD calculations. The center of the fitted sigmoid function indicates the center of concentrations for which the peak intensity changes linearly. Therefore, seven concentrations were selected in the vicinity of the sigmoid fit's center. Although two points are sufficient to fit a line, seven points are taken in the interest of accuracy. The concentrations were selected symmetrically around the sigmoid fit's center. If the center was closer than three points to the starting concentration, then the first seven concentrations would be selected for the calibration plot. The line fitted to the selected concentrations was then used to calculate the LOD, as shown in figure 5(b).

Figure 5. LOD was found using (a) an extended-range calibration plot with a sigmoid fit, then, (b) a calibration plot with the selected points, resulting in LOD values for the 4 characteristic peaks of BPE. (c) The minimum LOD value averaged over the 3 maps for each substrate type. The error bars show the map-to-map variation of the LOD. The example calibration plots brought here are for 1636 cm−1 BPE peak using a Hamamatsu substrate.

Download figure:

Standard image High-resolution image

The minimum LOD values for the investigated substrates are presented in figure 5(c). Silmeco Ag has the lowest LOD of 0.18 ± 0.19 μM and a LOQ of 0.59 ± 0.64 μM. Moreover, Hamamatsu has the most stable results, as seen in figure 5(c)'s error bars. The detection sensitivity is inversely related to the EF results (figure 5(f)), as expected. Furthermore, the 1020 cm−1 minimum LOD is noticeably different compared to that of the other 3 peaks. This can be explained by the fact that 1020 cm−1 marks the weakest of the four characteristic peaks. Therefore, it might be more unstable compared to the other peaks. Furthermore, a statistical overview of the maps' LOD values is provided in table 3. LOQ differs from LOD by only a factor, thus, everything discussed for LOD applies to LOQ as well.

Table 3. LOD and LOQ of the 1636 cm−1 characteristic peak of BPE for the investigated SERS substrates. The errors represent the map-to-map variation.

 LOD (nM)LOQ (nM)SERS substrateMeanMinSTDaMeanMinSTDPiCO Au4323 ± 2068343 ± 493675 ± 124713 712 ± 61041145 ± 16211 044 ± 2785ATO ID Ag12 927 ± 866610 ± 4629715 ± 39126 258 ± 8182033 ± 154012 047 ± 393Hamamatsu Au900 ± 8210 ± 37567 ± 863000 ± 26700 ± 1231890 ± 286SERSitive Ag927 ± 52690 ± 51749 ± 4983090 ± 1754298 ± 1702496 ± 1661SERSitive AgAu1693 ± 69540 ± 92674 ± 14044770 ± 1771135 ± 316989 ± 3329Silmeco Au307 ± 14218 ± 5396 ± 2701025 ± 47259 ± 171321 ± 899Silmeco Ag178 ± 1926 ± 5224 ± 225593 ± 64119 ± 18748 ± 749