Large-area sensors using Cd(Zn)O plasmonic nanoparticles for surface-enhanced infrared absorption

2.1 SEIRA sensing

The analysis of SEIRA with NPs was initially approached by a far field study using the techniques described in the methods section. Sample B1, which has the largest surface coverage, was chosen to exemplify the detailed analysis of the SEIRA performance carried out in all of the samples of the series since it is the one with the best performance. The results for such sample are presented below, and in a latter section, those of the whole series.

To better understand the mechanisms underlying the SEIRA enhancement, the transmittance spectra of sample B1 were obtained for two cases, with a 15 nm or a 50 nm-thick layer of PMMA deposited on top of it. Both spectra were analyzed as described in the methods sections, and the results are shown and compared with a reference sample of PMMA-coated GaAs substrate in Figure 5. In it, the asymmetric s-like shape of the Fano resonance can be seen, indicative of the resonant coupling of the absorption from the molecule and the LSP mode from the nanoparticles [40]. For the 15 nm-thick PMMA case, and using the coupling between the low energy LSP mode and the molecule bond, we report an absorption enhancement factor of 3.1 for the vibrational signal from the stretching mode of the C=O bond (1,732 cm⁻¹) of PMMA. This magnitude of enhancement is comparable to previously reported values in both metals and semiconductors [13], [14], [15], [16]. Also note that by depositing PMMA on the surface of the sample, the permittivity of the environment that surrounds the nanoparticles is changed, producing a slight shift of the plasmonic resonances when compared to the spectra shown in Figure 1. Similarly, the molecules’ interaction with the substrate and the NPs directly influences their binding energy. As a result, the experimentally observed vibrational lines exhibit a slight infrared shift from those found in the literature.

Figure 5:

Transmittance spectra (purple) of Cd_0.9Zn_0.1O NPs with a 15 nm (a) and 50 nm-thick (b) layer of PMMA spinned on top of the sample. In both (a) and (b), the purple line corresponds to the measurement, the dotted black line corresponds to the baseline obtained by using Eilers method (see Methods), and the blue line to the vibrational signals obtained upon subtraction of the baseline from the measurement. The green curve shows the signal from PMMA deposited on a bare GaAs substrate and serves as a reference. Hence, the enhancement in the absorption lines is obtained by normalizing the amplified signal by that from the reference.

Figure 5 also shows the spectra for the case of a 50 nm-thick PMMA coating. In this case, the vibrational signal enhancement is a factor of ×2.1, considerably smaller than that of the 15 nm-thick film of PMMA. There are two possible explanations for this matter. The first one can be extracted from observing how SEIRA is originated in the near field, and the second one from studying the frequency detuning. To begin with, Figure 6 presents the near field enhancement (|E_NP|²/|E₀|²) for a canonical NP with both 15 and 50 nm PMMA coverages. As can be seen, the field is highly amplified around the vertices where the NP meets the substrate. It is around these areas, named hot spots, where PMMA experiences the largest absorption enhancement. Hence, it becomes clear from the results in Figure 6 that the ratio of total PMMA volume where there is high near field amplification is smaller in the thickest film. This explains well the smaller amplification experimentally observed in Figure 5 for the 50 nm-thick PMMA film.

Figure 6:

Cross-sectional normalized field enhancement at 1732 cm⁻¹ for the low energy LSP mode of the NPs covered with a 15 nm (a) and 50 nm-thick (b) film of PMMA at an incidence angle of 45°. The white color represents the areas where there is no enhancement when compared to a model without the NP.

In terms of frequency detuning, the low energy mode is at a higher energy than the molecular absorption of the C=O bond for the 15 nm case but resonant for the 50 nm case. As reported by Vogt et al. [50], the slight detuning of the molecule absorption line (ω _C=O) from the plasmonic resonance frequency (ω _LE) results in a higher vibrational signal. This results in the vibrational signal for the 15 nm PMMA layer being higher than that of the 50 nm one, as the frequency tuning ratio (ω _C=O/ω _LE) is around 0.87 compared to 1, respectively, as shown in Figure 5. Ultimately, both the influence of the field distribution with the PMMA thickness and the frequency detuning contribute to the vibrational signal being larger for the 15 nm PMMA film case.

The stretching of PMMA’s C–H bond has absorption frequencies at 2,956 and 3,001 cm⁻¹ (2,947 and 2,986 cm⁻¹ in the literature due to the aforementioned shift), which aligns perfectly with the frequency of the high energy LSP mode of the NPs. Thus, a coupling between the bond and the LSP mode is expected to occur, and indeed it is visible in Figure 5, yielding an enhancement factor of 1.5. However, the enhancement factor for the C–H bond coupled to the high energy LSP mode is clearly smaller than that of the C=O stretching bond coupled to the low energy LSP mode. As discussed earlier, this is likely a result of the much lower amplification of the electric field provided by the high energy LSP mode. Further, this amplification appears on the apex of the NP (see Figure 4), which could explain why with the thickest PMMA layer, covering entirely the NP, the resonance from the bond shows a slightly higher enhancement. Given that both the enhanced and unenhanced material in the 50 nm layer is bigger than in the 15 nm layer, it adds up to similar vibrational enhancement overall. In terms of detuning, the trend is the same as for the low energy mode, which could also explain the slight difference in vibrational signal; however, the change is too small (0.1) to reach further conclusions as both mechanisms contribute to the effect.

The geometry of our samples offers a radical departure from most recent studies regarding semiconductor nanoantennas for sensing applications. As compared to evenly spaced devices that use electron beam lithography [13], [14], [15], [16], Cd(Zn)O self-assembled NPs provide a stochastic approach that requires an alternative analysis. Nevertheless, we first need to understand how SEIRA is usually quantified. The figure of merit typically used to quantify SEIRA sensing is the Enhancement Factor (EF), which is defined as [2]:

where S _NP represents the enhanced signal strength, while S ₀ denotes the unenhanced signal strength. A _NP and A ₀ both represent areas covered with molecules, the former accounting for those in contact with the hotspots, and the latter for the entire surface of the unit cell. Notably, the active area A _NP is not precisely defined and requires careful consideration, being commonly assessed indirectly via electromagnetic simulations [2]. Since we have a thin analyte film covering the whole surface of the sample, it is fair to assume all the hotspots are in contact with analyte molecules. The ambiguity would be in the condition to consider what is a hotspot (i.e., the threshold field amplification), which is arbitrary. This arbitrariness together with the possibility to modify the ratio A ₀/A _NP by changing the density of hotspots is what compels us to avoid using the EF.

Therefore, although the EF is established as the norm, we consider it is not the fairest metric for our samples due to the high density of hotspots. We believe that the vibrational signal is a better guiding parameter for these types of studies, as the widely accepted EF fails to provide a consistent approach. However, in order to benchmark our results with the previously reported bibliography, we must provide an estimate of such metrics.

In our case, A _NP can be extracted from the plan view image of the NP present in Figure 7. The field is highly enhanced and concentrated in the areas that appear black in the colormap. Such areas correspond to a cutoff of |E_NP|²/|E₀|² > 150. With the help of ImageJ, the total area in white (A _NP) can be quantified, yielding a result of 94.5 nm². On the other hand, the total area of the unit cell, A ₀, is computed just as the square of its side, resulting in a value of 1.44 × 10⁴ nm². Thus, the ratio A _NP/A ₀ yields a value of 153, which multiplied by the vibrational enhancement results in the EF. From our previous discussion, we have stablished that the maximum enhancement that we have observed is for the C=O stretching vibrational signal for the 15 nm-thick PMMA film. In this case, the S _NP/S ₀ ratio equals 3.1, yielding an EF of 4.73 × 10², a value comparable to the values reported by Huck et al. using metallic nanorods [2]. However, such EF value is 2 orders of magnitude below most current studies [40]. As an example, the study by V. Vogt, where nanoantennas arrays are processed through e-beam lithography, reports EFs up to 2.7 × 10⁴, while the maximum enhancement in the vibrational signal is 2.11 [50]. This means that even though their reported signal enhancement is 33 % smaller than the one that we report, their geometry allows an EF 20 times larger than ours. Other reports on semiconductor nanostructures [52], [55], [56], [57] show EFs in a broad range, but when it comes to vibrational enhancements, the reported values are found within the order of magnitude of the results from this article.

Figure 7:

Plan view of the normalized field enhancement for the low energy LSP mode of the NPs at an angle of 45°. The black regions correspond to an enhancement of |E_NP|²/|E₀|² > 150.

From the application standpoint, we believe that our approach offers realistic applicability. We argue that in the realm of commercial sensing devices, instead of devices where the enhancement per nanostructure is extremely high but the overall acquired signal is very low and costly to probe, it would be more beneficial to obtain a much larger signal by greatly increasing the number of nanostructures per unit area that provide SEIRA (i.e., increasing the ratio of SEIRA-active area to total area of the sample), as we propose. This large number of NPs not only yields a larger signal but also greatly increases the probability of the probed analyte to be found around one of them. Furthermore, as the Cd(Zn)O NPs are formed by a self-assembly process, the device fabrication would be faster and more cost-efficient than reported approaches that involve electron beam lithography. In fact, our proposed approach uses readily available substrates and allows to coat them with NPs yielding an area as large as 2 cm². Such large coating area also facilitates the implementation of our transducers into commercial detectors, as the spot size of the illumination sources available is in the realm of the millimeters.

As pointed out above, it is clear that the density of nanoparticles strongly influences the total enhancement of the vibrational signal. Thus, a study of the four samples with different surface coverages (micrographs in Figure 1) is due. The samples were processed, measured and analyzed as explained earlier, yielding the SEIRA signals that are shown in Figure 8, for both the thicker and the thinner PMMA coatings. The results, derived from the vibrational coupling between the low energy LSP mode and the C=O stretching bond of PMMA, show a clear linear increase of the vibrational enhancement with the surface coverage, thus confirming our hypothesis. This implies that, as long as the NPs do not coalesce, the increase in surface coverage through maximization of the number of NPs carries an increase of the effective number of hotspots where the absorption of the PMMA film is enhanced.

Figure 8:

Dependence of the vibrational signal enhancement of the C=O stretching bond coupled to the low energy LSP mode, on the NP surface coverage for the 15 nm (red) and 50 nm-thick (blue) PMMA films. The error bars included in the graph account for the standard deviation from the mean, which is due to heterogeneities found in the samples as the experiments were performed on eight different spots on the surface of each sample.

Lastly, a study of SEIRA was also conducted to test the efficiency of our device on sensing other molecular bonds. A useful compound to do so in the mid-IR is vanillin as it features several strong IR active vibrational lines of diverse functional groups. Within a range of less than 750 cm⁻¹, we can find the bending of the CH₄ and C–H bonds as well as the stretching of the C=O, C–C, C–O, and CH₃ bonds, all of them spectrally located within the wide low energy LSP mode of the Cd_0.9Zn_0.1O NPs.

As specified in the Methods section, approximately 10 nmol of vanillin (molecular mass of 152.1 g/mol) are deposited uniformly on the entire 0.25 cm² surface of the transducer. If we assume that the adsorption of vanillin onto GaAs follows a Langmuir isotherm model [58], then a monolayer coverage is achieved with noninteracting adsorbed molecules. This results in approximately 2 ng of vanillin within a 200 × 200 μm region, which is the area under illumination. An area equal to the unit cell described in the model, A ₀, contains 9 × 10⁻⁵ pg per NP, which means there are roughly 3 × 10⁵ molecules per NP. However, the vibrational enhancement is only given by the molecules that are primarily located in the intense near-field hotspots of the NP. As the effective area accessible to the analyte (A _NP,) represents a 0.65 % of A ₀, we can conclude that approximately 2 × 10⁴ molecules are triggered per NP. This number is two orders of magnitude larger than the state of the art in SEIRA sensing [13].

The results of the study are presented in Figure 9. The black spectrum on the bottom graph serves as a guide to observe the resonance frequencies of the molecular bonds. A very large enhancement of the vibrational signal of most molecule resonances is found again, being the CH₃ bending the largest one, with an enhancement factor of ×4.3. Other peaks, such as the C–H bending and the C–C stretching bonds, also show high levels of enhancements comparable to the ones discussed in the previous study for PMMA. Given such vibrational enhancement, the achieved EF is 1.5 × 10³, again a value in the order of magnitude of the current state of the art.

Figure 9:

Transmittance and SEIRA vibrational enhancement for Cd_0.9Zn_0.1O NPs covered with vanillin. The purple line corresponds to the measurement, the dotted gray line corresponds to the baseline obtained by using Eilers method (see Methods), and the blue line to the vibrational signals obtained upon subtraction of the baseline from the measurement. The green curve shows the signal from vanillin deposited on a bare GaAs substrate and serves as a reference. Hence, the enhancement in the absorption lines is obtained by normalizing the amplified signal by that from the reference. In black, vanillin’s main absorption lines to serve as a guide.

A question now prevails: why is the EF of vanillin larger than that of PMMA for the same surface architecture? Even though we do not have the means to realize an in-depth study on this issue, we propose the following hypothesis. When dissolved, vanillin is a small molecule, while PMMA is a polymer chain formed by the addition of sequential monomers, resulting in particles sizes that can range from hundreds of nanometers to hundreds of micrometers. This outstanding size difference ultimately means that the access of the molecules to the hot spots in the NPs is radically different between both compounds. On one hand, the small size of the vanillin molecules may allow them to adsorb easily anywhere on the surface, including the vertices where the NP meets the substrate and where the electric field is highly enhanced. On the other hand, the size of the PMMA chains could prevent them from fully covering the mentioned hotspot, thus showing a smaller coupling and ultimately a lower vibrational enhancement.

4.1 Mid-IR spectroscopy

Infrared spectra were taken using a Fourier Transform Infrared (FTIR) spectrometer equipped with a deuterated triglycine sulfate (DTGS) detector, a KBr beamsplitter, and a SiC glowbar. The samples were measured in a transmittance configuration with an incident angle of 45° and p-polarized light (magnetic field perpendicular to the light incidence plane). All spectra were recorded with a resolution of 4 cm⁻¹ and 1,024 iterations.

4.2 Nanoparticle growth and structure

The Cd(Zn)O NPs samples featured in this work were grown by MOCVD on doubled-side polished semi-insulating GaAs wafers, using tertiary butanol, dimethylcadmium, and diethylzinc as precursors, with N₂ as the carrier gas. The nominal Zn molar concentration was varied between 0 and 20 %, and the temperature was maintained within the mass-transport-limited regime (300–350 °C) for the growth of NPs of the ternary compound Cd(Zn)O [54]. In contrast, for the binary oxide CdO NPs, the so-called “thermodynamically dominated” regime was used to prevent nanoparticle coalescence. The surface density and morphology of the samples were controlled by adjusting the growth conditions through several key parameters, including temperature, precursor flow rates, deposition time, and the sample position within the reactor, which proved to be particularly relevant due to the horizontal reactor geometry. These studies enabled the reproducible fabrication of samples with a wide range of particle densities and sizes while maintaining the hemispherical shape of the NPs. The samples’ characteristics and growth parameters are summarized in Table S1.

The structural characterization of the samples was done using a FEI Inspect F50 Scanning Electron Microscope (SEM) and a Dimension Icon Atomic Force Microscope (AFM) from Bruker. To evaluate the surface coverage of the samples, the SEM micrographs were analyzed using the particle analyzer tool from ImageJ free software. AFM scans were visualized and studied with the Gwyddion free software.

The samples are not intentionally doped. The high electron concentration in Cd(Zn)O is believed to originate from intrinsic defects that act as shallow donors, most likely hydrogen interstitials [60], [61], [62] whose concentration may be increased with increasing Zn content as a result of the gas precursors used for Zn [29].

4.3 SEIRA sensing

After the preparation and characterization of the nanoparticles, the samples were cleaned in two subsequent ultrasonic baths of acetone and isopropyl alcohol (IPA) for 10 min each. Then two different protocols were followed to deposit the PMMA layers depending on the desired thickness. For the 50 nm-thick layers, PMMA was directly spin-coated on the samples at 4,000 rpm for 60 s. For the 15 nm-thick layers, a solution of 10 % in volume of PMMA in acetone was spin-coated with the same parameters as the former case, followed by a thermal treatment on a hot plate at 160 °C for 2 min to evaporate any traces of the solvent. For both cases, an Eppendorf Research plus mechanical micropipette was used to drop cast 20 µL of the solution on the surface of the samples. The aforementioned thickness of the PMMA layers was calibrated by AFM on a bare GaAs substrate.

Once the samples with NPs are coated, they are measured in the spectrometer along with a GaAs substrate with a similar PMMA layer to act as a reference. The measured spectra are fitted using the smoothing algorithm proposed by Eilers [63]. The fit serves as a reference signal to standardize the measured data. Consequently, a predominantly flat line, except for the vibrational signals, is generated, enabling us to derive the vibrational signal intensity from this baseline-corrected spectrum by measuring the peak-to-peak amplitude of the vibrational signals [42].

For the measurement of SEIRA in vanillin, an alternate approach was carried out when depositing the compound. First, a 2 mg/mL solution of vanillin/IPA was prepared to then drop cast a 1 μL droplet of it onto the sample surface. The sample is left to dry for a few seconds and then measured in a Hyperion Vertex 70 FTIR microscope fitted with a × 36 Schwarzschild objective.

4.4 Finite elements method model

A 3D finite element model was created in COMSOL Multiphysics to analyze the electric field distribution in and around the NPs and PMMA layer. The model is comprised from top to bottom of the subsequent thin layers: Air, PMMA, and GaAs. The nanoparticles are sitting on the GaAs substrate either embedded on the PMMA layer or half-covered by it depending on the thickness of the layer. Conformal layers of PMMA were not considered for the sake of simplicity of the study. The chosen geometrical parameters were adapted from the AFM and SEM results obtained on the NPs samples. The following assumptions were made. The NPs are perfectly hemispherical and evenly distributed on the surface in an infinitely periodic lattice. Their radius is 30 nm and they are separated 60 nm from each other, thus complying to a mean surface coverage of 30 %. The light intensity enhancement maps shown on this work are calculated in the following manner. The values of |E_NP|² are obtained from the model with the NP and the values of |E₀|² from a simulation without it. When both field distributions are compared, the enhancement is obtained. The model uses linearly polarized light to emulate the incident beam of the FTIR spectrometer. The electric field is polarized in TM configuration and has an incidence angle of 45°.