1.

Introduction

1,3,5-triazine-2,4,6-triamine commercially known as melamine, ${\mathrm{C}}_3{\mathrm{H}}_6{\mathrm{N}}_6$, is generally used in laminates, coatings, or glass wares as melamine formaldehyde resins. However, the high nitrogen content (66% by mass), lured unreliable manufacturers to use it to fake the protein content, indicated by nitrogen, in infant milk formula and petfood. Melamine and cyanuric acid, or any other melamine related compounds such as ammeline, ammelide, and 1-(2-mercaptoethyl)-1,3,5-triazinane-2,4,6 trione (MTT),^1–3 may produce complexes in the human body^4–6 when they combine together via hydrogen bonds. At higher concentrations (more than $1\mathrm{ppm}/\text{day}$ for child and $\text{3\hspace{0.17em}\hspace{0.17em}}\mathrm{ppm}/\text{day}$ for adult) those complexes will aggregate in the kidneys and form kidney stones^4,5,7 causing urinary system inflammation^4,5 or urolithiasis. The World Health Organization (WHO) estimated the “tolerable daily intake,” to be 0.2°mg per kg of body mass.⁸ Such toxic additives have to be detected quickly and mass production of food items has to be controlled by authorities.

This calls for reliable and accurate sensing techniques and materials with reasonable ease-of-use. Conventionally, melamine is detected by high performance liquid chromatography (HPLC),^9,10 liquid chromatography coupled mass spectroscopy (LC-MS),¹¹ LC-MS/MS, gas chromatography,^9,12,13 matrix assisted laser desorption/ionization mass spectrometry,¹⁴ variety of ionization mass spectrometry,^15,16 commercial enzyme linked immunosorbent assay technology,¹⁷ and infrared spectroscopy.¹⁸ These laboratory-based techniques, although reliable and accurate, are time consuming, and challenging for on-site common use. Relatively easier methods include optical absorption^19–23 and Raman spectroscopy (RS), including surface-enhanced Raman spectroscopy (SERS).^24–27 Recently, optical absorption studies are being used to detect melamine using MTT functionalized gold nanoparticles (AuNP).²² MTT attaches at one end to the AuNPs, by the thiol bond, and upon exposure to melamine bind to it through hydrogen bonding, thus agglomerating the AuNPs into clusters. The agglomeration will alter the surface plasmon absorption wavelength from 520 nm, for MTT stabilized AuNPs, to $\sim 670\mathrm{nm}$ for the melamine exposed AuNPs. The shift, a function of melamine concentration, in this absorption frequency, observable by naked eye, is used for colorimetric sensing of melamine. However, such studies always used functionalized or modified AuNPs. Similar colorimetric determination of the presence of melamine can be done using ${\mathrm{Fe}}_3{\mathrm{O}}_4$ magnetic NPs with 2.5 ppm detection levels.²³ Recently, SERS is also being used for the sensing purpose.²⁴ For SERS noble metal (Au, Ag) NP decorated substrates were used to enhance the scattering cross-section of melamine and hence lowering the detection limit. SERS is a technique that can be used on-site with relatively smaller degree of difficulty.

In the present work, we demonstrate how the target analyte, melamine, itself induces agglomeration in unmodified gold nanoparticles, resulting in fractals on solid support in air, enabling target detection by SERS. Using finite difference time domain (FDTD) optical simulation we provide a preferred condition of plasmon distribution on the Au-fractal substrate that would be necessary for the sensing.

2.

Experimental

3.

Results and Discussion

For colorimetric^22,23 or Raman spectroscopic detection^24,30,31 of melamine, noble metal or magnetic NPs are generally used. In this case also we have used AuNPs, but without any cumbersome functionalization, for the detection using SERS. Figure 1 contains the details of the AuNP properties. The size control of AuNPs, prepared by the Turkevich method,²⁹ was achieved by varying the concentration of the reactants to obtain distinctive AuNP colloids with purple, red, and orange colors. An optical image of the AuNP colloid and its size distribution, measured by Zetasizer, is shown in Fig. 1(a) and 1(b), respectively. The particle size was predominantly in the 20 to 30 nm range with a distribution as expected of the Turkevich reaction. Representative TEM images of the AuNPs, on the TEM grid, as dispersed and after exposure to 50 ppm melamine is shown in Fig. 1(c), and 1(d), respectively. The AuNPs showed triangular and hexagonal features, and readily agglomerated upon exposure to melamine. Representative SEM image of the AuNPs after melamine exposure is shown in Fig. 1(e). Absorption spectra of the as-grown AuNP colloid, and those after addition of 50 and 100 ppm of melamine is shown in Fig. 1(f). The results showed a surface plasmon absorption peak at $\sim 530\mathrm{nm}$. When melamine with different $M_c$ was added to the AuNP solution, the absorption peak intensity, at 530 nm, decreased [Fig. 1(f)] but no new peak, at higher wavelengths, found in the colorimetric detection methods, was observed that would imply significant agglomeration of the AuNPs in the solution. This is expected since the electrostatic repulsion of the Cl- surface coverage resisted the agglomeration. This is why MTT²² or riboflavin²⁰ functionalization of the AuNPs was done. However, if sufficient amount of melamine is added (50 to 100 ppm) to the AuNP solution such an agglomeration could be actually observed even in unmodified AuNPs but they are too weak to trigger the second peak at a higher wavelength. This proves that melamine have this inherent tendency to weakly agglomerate nonfunctionalized AuNPs even in solution.

Fig. 1

Characterization of the colloidal gold nanoparticles (AuNPs). (a) A conventional optical photograph of the colloidal AuNPs. (b) The size distribution of the AuNPs. TEM image of the AuNPs (c) before, and (d) after 50 ppm melamine exposure. (e) SEM image of agglomerated AuNPs on Si after 50 ppm melamine exposure. (f) Optical absorption spectra, in colloidal solution, of the as-prepared AuNPs (0 ppm), and AuNPs exposed to 50 and 100 ppm of melamine.

The use of MTT²² for AuNP agglomeration is prohibitive for three reasons. First, the production yield of MTT is low, secondly, its production cost is high, and lastly, it introduces an additional functionalization step in the whole detection process. Hence, our method aimed at discarding this step for the detection purpose. We tried to enhance the agglomeration effect on dry substrates that are devoid of the polar solvent, in this case, water. We dispersed 30 μL of the as-grown AuNPs on hydrofluoric acid (HF) cleaned commercial Si wafers and dried them in vacuum to get a clean two-dimensional (2-D) dispersion. Large population density of AuNPs is avoided (1) to restrict three-dimensional (3-D) clustering that would obstruct the agglomeration phenomenon, and (2) to simplify the optical simulations done to construct the plasmon distribution design (discussed later). There is no chemical bond between the Si substrate and the AuNPs covering it. Figure 2 contains SEM images of AuNPs before and after the exposure to different amounts of melamine. The as-dispersed AuNPs (dry), on Si in air, showed a density of $300\pm 30\mathrm{NPs}/{\mu \text{m}}^2$ [Fig. 2(a)]. SEM was repeated [Fig. 2(a), all panels] at different magnifications and areas to verify the distribution that predominantly depended on the volume of the AuNPs solution and the wetting properties of the HF cleaned Si surface that showed a water contact angle of 43 deg at room temperature. Figure 2(b) to 2(d) showed the AuNPs distribution after exposure to 10, 1, and 0.1 ppm of melamine, respectively. As melamine is introduced, these AuNPs readily agglomerates forming an interconnected fractal-like network³² leaving polygonal islands in between with less than $80\pm 20\mathrm{NPs}/{\mu \text{m}}^2$ [Fig. 2(b) and 2(c)]. Such kind of fractal structures were previously obtained where a chain aggregate was constructed by sharing faces, corners, and edges of a polygon.^33,34 These edges, corners, or faces are the areas expected to have high SERS enhancement and ideal for melamine sensing that we would soon demonstrate. The peripheries of these islands are 400 to 600 nm thick containing chained and agglomerated AuNPs with a high 2-D spatial density approaching an ideal SERS enhancement condition³⁵ of $d=2r$ where $d$ is the interparticle separation and $r$ is the radii of the AuNPs. In other words, the SERS efficiency is pronounced when AuNPs are nearly touching each other. The AuNP agglomeration was pronounced for $M_c=10\mathrm{ppm}$, but the clustering weakens and the continuity of the fractal-network is broken as $M_c$ is reduced to 0.1 ppm with the island boundaries thinning to $\sim 100\mathrm{nm}$ when present at all [Fig. 2(d)]. In comparison, the AuNPs chains on the island boundaries were $\sim 600$ [Fig. 2(b)] and 350 nm [Fig. 2(c)] thick when 10 and 1 ppm of melamine was introduced, respectively.

Fig. 2

Representative SEM images of (a) as-grown AuNPs, and the fractal-like AuNP clustering, on silicon, post melamine exposure of (b) 10, (c) 1, (d) 0.1 ppm. The left, middle, and right panel in each row contains images with 5000, 33000, and 50000 times magnification, respectively. The width of the AuNP chains or the fractal interconnects are shown in the (b) and (c) panels.

The density of the AuNPs, counted manually by making $1{\mu \text{m}}^2$ meshes on 10 large area SEMs and neglecting meshes with fewer than $300\pm 30\mathrm{NPs}/{\mu \text{m}}^2$ (as-dispersed density), were $2100\pm 100$, $1450\pm 80$, and $900\pm 50/{\mu \text{m}}^2$ for exposures of $M_c=10$, 1, and 0.1 ppm, respectively. In the large number of SEM images studied, we found very few 3-D clustering, which is not ruled out, but most of the melamine-induced agglomeration resulted in 2-D clustering under the given experimental conditions.

Figure 3(a) to 3(e) contain detailed SERS data of melamine (with $M_c=10$, 5, 1, 0.5, and 0.1 ppm) dispersed on the AuNPs coated Si. For comparison, the RS data (not SERS) of melamine is shown for bare Si substrates (without the AuNPs) [Fig. 3(f) to 3(j)]. Spectra were obtained on different points on each substrate with a focus on the signature melamine (bulk) peak at ${676\mathrm{cm}}^{-1}$ (Fig. 4). Repeated scans and their relative standard deviations (rsd) are shown for each melamine concentration. To have a qualitative estimate of the efficacy of the AuNP-fractals in enhancing the signal, corresponding Raman data of melamine is provided on bare Si substrates [Fig. 3(f) to 3(j)] on which the ${676\mathrm{cm}}^{-1}$ signal was only visible for $M_c=10$ and 5 ppm. The region of interest being around the strongest signal from melamine, 650 to ${700\mathrm{cm}}^{-1}$, is marked in each plot in Fig. 3 scanned from 550 to ${900\mathrm{cm}}^{-1}$. The ${676\mathrm{cm}}^{-1}$ signal (integrated area ${\mathrm{I}}_{676}$) could be identified unambiguously down to the limit of 0.1 ppm [Fig. 3(e)] [with a signal: noise ratio ($S/N$) above 3:1] although its intensity decreased with decreasing $M_c$ [Fig. 3(a) to 3(e)]. The rsd is used to estimate the repeatability (precision) of the measurements. Along different regions of the spectrum (550 to ${900\mathrm{cm}}^{-1}$) the rsd varied between 10% to 30%, however, in the region of interest, 650 to ${700\mathrm{cm}}^{-1}$, the rsd was $<15\%$ and flat indicating a small deviation of the signal intensity compared to other parts of the spectrum. The weak non-SERS spectra of melamine on bare Si substrates reveal qualitatively the efficacy of the AuNP-fractals in enhancing the signal.

Fig. 3

Surface-enhanced Raman spectra (SERS), collected at 5 different spots of the same AuNP fractals, of melamine with (a) 10, (b) 5, (c) 1, (d) 0.5, and (e) 0.1 ppm concentration. Corresponding Raman spectra (RS) for melamine, on bare Si (non-SERS), with the same concentrations are shown in (f) to (j). The relative standard deviation (rsd) of the statistical data is shown at the bottom of each plot (before the scale break on the vertical axes). The spectra were shifted vertically for clarity. The spectral region of interest for melamine detection is marked by a horizontal bar in each figure.

Fig. 4

Normalized Raman scattering data of melamine (bulk) powder, dispersed on Si, showing the signature peak at ${676\mathrm{cm}}^{-1}$.

The variation of the average $I_{676}$ as a function of $M_c$ is shown in Fig. 5, for both the SERS (on AuNP fractals) and non-SERS (reference plot, on bare silicon) cases. The slope of this plot (Fig. 5) is indicative of the sensitivity of detection in the studied concentration range. It is to be noted that the integrated signal intensity $I_{676}$ is not only a function of $M_c$, but also a function of the number of SERS active sites ($N_s$) within the illuminated region ($\sim 12.5{\mu \text{m}}^2$) and with some simplifying assumptions we can write $I_{676}\propto N_s\times M_c$. One has to note that all AuNPs are not SERS active.³⁶ Given the increasing coverage (area) of AuNP clustering with $M_c$ (Fig. 2), we expect an increase in $N_s$ too, and hence a stronger $I_{676}$ was observed [Fig. 3(a)]. In fact, Doering and Nie,³⁷ while comparing metallic single particle and aggregates for single molecule SERS, observed that the active sites or hot spots are most likely to be at the junction of two single particles in metallic aggregates. This supports our finding why the AuNP fractals helped in the detection.

Fig. 5

Variation of the average intensity of the ${676\mathrm{cm}}^{-1}$ line of melamine Raman spectrum as a function of melamine concentration (in water, Fig. 3) when dispersed on (⊙) Silicon (non-SERS) and () gold fractals (SERS). A linear fit to the data is also shown. The $R^2$ values for the fitting are 0.9 and 0.92 for the SERS and non-SERS data set, respectively.

It is difficult to estimate the SERS enhancement factor ($\mathrm{EF}={\sigma }_{\mathrm{SERS}}/{\sigma }_{\mathrm{RS}}$, $\sigma$ being the Raman scattering cross-section) since the number of melamine molecules and $N_s$ within the excited area, and the value of ${\sigma }_{\mathrm{RS}}$, are all unknown.^38,39 However, with some approximation, a rough estimate of the EF can still be made using the following equation:^38,40

In order to determine why the melamine detection was possible, even without functionalization of the AuNPs, the FDTD simulations (Figs. 6 and 7) were carried out. The presentation of the FDTD results begins with a representative SEM image of a given AuNP distribution. A replica, in size and density, of a section of this AuNP distribution is constructed theoretically. This AuNP distribution model when imported into the FDTD software, the optical extinction spectrum could be calculated. This spectrum will show the surface plasmon absorption peaks of the AuNP distribution at specific wavelengths. A single peak for individual unagglomerated AuNP and two or more peaks for agglomerated AuNPs will be obtained under orthogonal polarization directions ($x$, $y$) of the electric field ($E$) of the incident light. Next, the software scans the replica of the AuNP distribution, constructed previously, with incident light wavelength in resonance with that of the surface plasmon absorption peaks obtained in the respective extinction spectrum. This would result in the surface plasmon $E$-field mapping of the entire AuNP distribution. This procedure is repeated for the two polarization directions ($x$, $y$) and with either one or two wavelengths depending on the respective surface plasmon absorption peaks calculated. In addition, the $E$-field mapping of the AuNP distribution was also performed with incident wavelength of 633 nm that was used for the Raman measurement.

Fig. 6

Numerical simulations using FDTD of the AuNP distribution post 10 ppm melamine exposure assuming surrounding medium index of 1.0. (a) Representative SEM image of the AuNP agglomeration. (b) A model replica of the AuNP distribution, from a selected area of the SEM, for use in the FDTD software. Calculated extinction spectra of the model AuNP distribution [as shown in (b)] with (c) horizontal and (d) vertical incident electric field ($\overline{E}$) polarization. Plasmon electric field mapping on the AuNP distribution with horizontal $\overline{E}$ polarization of (e) 525 and (f) 633 nm excitation, and with vertical $\overline{E}$ polarization of (g) 525 and (h) 633 nm excitation. The wave propagation vector $K$ is perpendicular to the plane of the paper. Same scale bars for (e) to (h).

Fig. 7

Numerical simulations using FDTD of the AuNP distribution post 0.1 ppm melamine exposure assuming surrounding medium index of 1.0. (a) Representative SEM image of the AuNP distribution. (b) A model replica of the AuNP distribution, from a selected area of the SEM, for use in the FDTD software. Calculated extinction spectra of the model AuNP distribution [as shown in (b)] with (c) horizontal and (d) vertical incident electric field ($\overline{E}$) polarization. Plasmon electric field mapping on the AuNP distribution with horizontal $\overline{E}$ polarization of (e) 525 and (f) 633 nm excitation, and with vertical $\overline{E}$ polarization of (g) 525 and (h) 633 nm excitation. The wave propagation vector $K$ is perpendicular to the plane of the paper. Same scale bars for (e) to (h).

Figures 6 and 7 contain comparative FDTD optical simulations of the AuNP distribution that were used to detect melamine. A representative SEM image ($x,y$) of highly agglomerated AuNPs ($M_c=10\mathrm{ppm}$) [Fig. 6(a)] was used to create exact theoretical replica [Fig. 6(b)] that yielded the extinction spectra (scattering cross-section) using two perpendicular polarizations ($Ex$ and $Ey$) of the electric field of the incident light [Fig. 6(c) and 6(d)] propagating along the $z$-direction. The extinction spectra showed double peaks at 525 and $\sim 770\mathrm{nm}$ (agglomerated AuNPs surface plasmon) [Fig. 6(c) and 6(d)]. Such double peaks in the spectra are present even for the 0.1 ppm melamine exposed AuNPs showing weak agglomeration. Surface plasmon maps of the AuNP distribution shows stronger E-field distribution under 525 nm excitation than under 633 nm, under both vertical [$Ex$, Fig. 6(e) and 6(f)] and horizontal polarization [$Ey$, Fig. 6(g) and 6(h)]. This is due to the lower surface plasmon absorption at 633 nm [Fig. 6(c) and 6(d)].

The same presentation scheme was maintained for the case of 0.1 ppm exposed AuNPs (Fig. 7). Since the FDTD calculations depend on the density and the size of the AuNPs, the chosen area of the SEM [Fig. 7(a)] and its theoretical replica [Fig. 7(b)] roughly had similar AuNP distribution ($\text{particles}/{\mu \text{m}}^2$) for 0 [Fig. 2(a)] and 0.1 ppm [Fig. 2(d)] melamine exposure. This is assumed for simplicity and no loss of technicality. In this case, only one scattering peak (as expected for mostly individual AuNPs at a low spatial density) is observed at $\sim 525\mathrm{nm}$ [Fig. 7(c) and 7(d)] for both $Ex$ and $Ey$. The plasmonic E-field map, at 525 and 633 nm, shows weak local plasmons centered around the individual AuNPs under both polarization conditions [Fig. 7(e) to 7(h)].

SERS-aided detection is a plasmonic phenomenon where the local electric field at the analyte site dictates ${\sigma }_{\mathrm{SERS}}$.⁴¹ In other words, a strong plasmon [red color in Figs. 6(e) to 6(h), and 7(e) to 7(h)] field coverage of the detector surface (per unit area) would increase the scattered intensity ($I_{676}$) to detectable limits. The simulated plasmon mapping shown in Figs. 6(e) to 6(h), and 7(e) to 7(h) was obtained assuming spherical AuNPs. However, the TEM results [Fig. 1(c)] clearly showed that the AuNPs distribution contained triangular and hexagonal AuNPs with sharp corners and edges. The shape dependence of SERS, reported earlier, indicate that though spheres are excellent model systems for simulations, the maximum enhancement comes from aggregates or fractal like structures of ellipsoids.⁴² However, even the spherical model approximation for the FDTD calculations gives us a lower bound for the plasmon field distributions. In reality, the electric fields would be higher considering the triangular or hexagonal AuNPs. The other usefulness of the FDTD calculations is that it can predict the absorption frequencies for the clusters/fractals under consideration. While a sphere can resonate at one particular frequency (or wavelength), a spheroid can resonate at three different frequencies and a random fractal may resonate at any given frequency over a broad spectral region.⁴³ These localized resonances in fractals increase in number from visible to near-IR increasing the probability of obtaining higher enhancement factors in SERS. Under the given experimental conditions, higher $M_c$ (10 to 1 ppm) could agglomerate the AuNPs so that a strong plasmon field with wide coverage [Fig. 6(e) to 6(h)] could be obtained over the detector surface. Such area of strong plasmon field decreases with decreasing $M_c$. This is why $I_{676}$ decreases with decreasing $M_c$ [Fig. 3(a) to 3(e)]. At $M_c=0.1\mathrm{ppm}$, such AuNP clusters still remained so that we could detect the melamine signal, though weak, quite consistently [Fig. 3(e)]. However, for AuNPs that did not agglomerate when $M_c=0.1$ [Fig. 7(b)] or 0 ppm [Fig. 2(a)], the plasmon field, with an absorption peak at $\sim 525\mathrm{nm}$ [Fig. 7(c) and 7(d)], is weak with low surface coverage [Fig. 7(e) to 7(h)] and hence detection was either difficult or not possible at all. The scarce presence of red color in Fig. 7(e) to 7(h) only indicates a lower probability of scattering and greater degree of uncertainty in measurement under both 525 or 633 nm excitation.

However, the calculated surface plasmon absorption of the AuNP distribution, though strong around 525 nm, was weak at the Raman measurement wavelength of 633 nm [Fig. 6(c) to 6(d); Fig. 7(c) to 7(d)]. Within our experimental rage of $M_c$ (10 to 0.1 ppm), the index of the surrounding medium was assumed to be 1.0 in our FDTD calculations shown in Figs. 6 and 7. If we take into account the presence of melamine within the fractals, then the effective index would be higher than 1.0, depending on its volume fraction, scaling with $M_c$. Although the exact index cannot be calculated, due to the uncertain coverage, we performed the FDTD calculations with higher representative index of 1.33 under 525 and 633 nm excitation. The results, irrespective of the AuNP distribution and incident $\overline{E}$ field polarization, show an increasing plasmon absorption band around $\sim 600\mathrm{nm}$ (Figs. 8 and 9) where the actual Raman measurements were performed. Usual red shift, towards higher wavelength, of the absorption bands are expected and observed with increasing index as reported before.^44,45 Although the 525 nm band was still the strongest [Figs. 8(a), 8(b), 9(a), and 9(b)], the plasmon field distribution at 525 [Fig. 8(c) and 8(d)] and 633 nm [Fig. 8(e) to 8(f)] excitation was markedly improved at background index of 1.33 compared to calculations done with background index of 1.0 for each excitation wavelength [Fig. 6(e) and 6(h)]. These results can now explain why we could still detect melamine with high sensitivity at 633 nm. The limit of detection would have been lower if we could measure at 525 nm because of the strongest surface plasmon absorption at that wavelength in the given AuNP fractal distribution. In short, we demonstrate how 100 ppb of melamine could be reproducibly detected on AuNP-fractals that creates a given plasmon field distribution as shown in Fig. 6(e) to 6(h) and Fig. 8(c) and 8(f).

Fig. 8

FDTD simulations of the AuNP distribution post 10 ppm melamine exposure [as shown in Fig. 6(b)] with surrounding medium index of 1.33.Calculated extinction spectra with (a) horizontal and (b) vertical incident electric field ($\overline{E}$) polarization. Plasmon electric field mapping on the AuNP distribution with (c) horizontal and (d) vertical $\overline{E}$ polarization of 525 nm, and (e) horizontal and (f) vertical $\overline{E}$ polarization of 633 nm excitation. The wave propagation vector $K$ is perpendicular to the plane of the paper.

Fig. 9

FDTD simulations of the AuNP distribution post 0.1 ppm melamine exposure [as shown in Fig. 7(b)] with surrounding medium index of 1.33. Calculated extinction spectra with (a) horizontal and (b) vertical incident electric field ($\overline{E}$) polarization. Plasmon electric field mapping on the AuNP distribution with (c) horizontal and (d) vertical $\overline{E}$ polarization of 525 nm, and (e) horizontal and (f) vertical $\overline{E}$ polarization of 633 nm excitation. The wave propagation vector $K$ is perpendicular to the plane of the paper.

To confirm, whether the strategy will still work when we extract melamine from a milk colloid, we performed similar SERS measurements with samples extracted from real contaminated milk. The extract contained melamine suspended in the liquid of milk with its proteins removed. Figure 10 shows SERS data of milk extracts containing different $M_c$. Experimental designs and conditions remaining the same, similar results were obtained with the $I_{676}$ diminishing in intensity with a lowering of $M_c$. Some background signals from the Si substrate and remnant milk components could be seen as well. The variation of the normalized $I_{676}$ as a function of $M_c$ (in milk extract) is shown in Fig. 11 whose slope is indicative of the sensitivity of detection in the studied concentration range. Here also, we could detect up to 100 ppb of melamine in milk from the feature at ${676\mathrm{cm}}^{-1}$ in the SERS spectra (Fig. 10). Lowering $M_c$ to 10 ppb resulted in virtually no signal ($I_{676}$) (Fig. 10). However, we could say, with confidence, that at least 100 ppb of melamine could be detected with a simple AuNPs distribution, with particle size $\sim 20$ to 30 nm, which demonstrated a plasmon field as shown in Figs. 6(e) to 6(h) and 8(c) and 8(f).

Fig. 10

Surface-enhanced Raman scattering data of melamine containing milk extract on Au fractals. Normalized (to $I_{676}$) SERS spectra of melamine containing milk extract with $M_c=10$, 5, 1, 0.1, and 0.01 ppm, arranged in order.

Fig. 11

Variation of the normalized intensity of the ${676\mathrm{cm}}^{-1}$ line of melamine Raman spectrum as a function of melamine concentration (in milk extract, Fig. 10) when dispersed on gold fractals (SERS). A linear fit to the data is also shown. The $R^2$ value of the fit is 0.93. The slope of the line is an indication of the sensitivity of the measurements.

4.

Conclusions

Presence of melamine was detected in raw milk by surface-enhanced Raman scattering using silicon substrates dispersed with unmodified 30 nm AuNPs. Melamine was found to agglomerate the AuNPs with as-dispersed densities of $300\mathrm{AuNPs}/{\mu \text{m}}^2$, to predominantly 2-D clusters with densities of $\sim 2100\mathrm{AuNPs}/{\mu \text{m}}^2$ (at 10 ppm concentration), that helped to generate a strong plasmonic field under illumination. The absorption and plasmon field strength of the agglomerated AuNPs were simulated by FDTD method. The optical simulations helped to qualitatively estimate the plasmon distribution to enable melamine detection to a limit of 100 ppb. Melamine containing milk extracts when applied to these unmodified AuNPs coated surfaces yielded strong Raman signals at ${676\mathrm{cm}}^{-1}$ indicating the presence of melamine. The absence of the AuNPs on the same substrates yielded no Raman signals under similar experimental conditions. A limit of detection of 100 ppb could be reproducibly obtained in this technique.