Transmission electron microscopy reveals clusters of Au–Ag nanoparticles formed in TiO2 thin film, with enhanced plasmonic response

This work reports on the influence of nanoparticle (NP) size distribution and the chemical nature of gold (Au) and/or silver (Ag) NPs in the localized surface plasmon resonance (LSPR) responses. The NPs were produced embedded in a titanium dioxide (TiO2) thin film, deposited by reactive magnetron sputtering technique followed by in-vacuum thermal treatment at 400 °C. High-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) gave quantitative key information in terms of both the size and distribution of the noble metal NPs. The average Feret diameter was 17 nm (σ = 8) and 55 nm (σ = 28) for Au/TiO2 and Ag/TiO2 films, respectively, while the Au–Ag/TiO2 film showed intermediate values, with an average size of 22 nm (σ = 9). HAAD-STEM, complemented by EDX chemical mapping, revealed an unusual formation of cluster structures containing local distributions of bimetallic (alloyed) Au–Ag NPs. The synergetic characteristics and properties of such bimetallic Au–Ag NPs resulted in an outstanding LSPR sensitivity compared to the monometallic counterparts. Furthermore, the analysis of the average nearest neighbor distances (about one order of magnitude lower than counterparts) suggests the existence of plasmonic hotspots relevant to be explored in sensing and surface-enhanced spectroscopies.


Introduction
Localized surface plasmon resonance (LSPR) based optical transducers have gained significant importance in various fields due to their unique optical properties and versatile applications [1,2].LSPR is an optical phenomenon in which localized surface plasmons are excited on the surface of a metallic nanostructure (e.g. a nanoparticle, NP) by an incident electromagnetic (EM) field [3,4].Such excitation results in a collective and coherent oscillation of the conduction band electrons when the NP size is one order of magnitude lower than the incident radiation wavelength [5,6].The LSPR band characteristics are highly dependent on the chemical and physical properties of the NPs.They can be tuned by changing the NPs' size, interparticle distance, shape, chemical composition, and surrounding dielectric [7,8].
Gold (Au) and silver (Ag) NPs are the most extensively studied materials in plasmonics since their resonance conditions are met in the visible range of the EM spectrum [9].As a result, Au and Ag are the main focus of plasmonic research for the numerous applications that can originate from LSPR, such as bio [10][11][12][13] and chemical sensing [14,15], optical imaging [16], phototherapies [17,18], catalysis [19,20], and surfaceenhanced spectroscopies [8,21].Au NPs are chemically inert and highly biocompatible [4], while Ag NPs display sharper LSPR bands than other metals [7].Thus, research in alloyed Au-Ag NPs gains momentum for their improved chemical stability and optical properties [22].As both Ag and Au display face-centered cubic (fcc) features while crystallized, it is possible, theoretically, to produce Au-Ag bimetallic NPs in a wide compositional range [23].
One method to prepare NPs is to synthesize them from metals dispersed in a dielectric matrix, and this topic has gained attention in recent years [24,25].Physical Vapor Deposition techniques, such as magnetron sputtering, allow the production of thin films with randomly dispersed noble metal atoms in a dielectric matrix.By adjusting the deposition parameters, such as current, deposition rate [26], target material [27], and noble metal content, it is possible to optimize the thin films to produce reliable LSPR-based optical transducers.However, just using magnetron sputtering does not ensure the formation of NPs with LSPR behavior in the thin film.As the deposition temperature normally reaches below 200 • C, only small (less than 10 nm) and non-crystallized NPs are formed due to the nucleation of dispersed atoms [28].
The growth of NPs follows well-established mechanisms, and to obtain such nanostructural domains from dispersed noble metal atoms and a few clusters, it is necessary to apply energy to favor atom diffusion.In this case, a thermal post-treatment ensures NPs growth and crystallization, resulting in nanocomposite thin film [29][30][31].
In the present work, thin films of Au, Ag, and Au-Ag NPs hosted by a TiO 2 dielectric matrix were prepared to study their nanostructure and correlate it with the optical and plasmonic responses.The thin film deposition parameters were adjusted to obtain thin films with similar total noble metal content.The resulting thin films were evaluated using scanning transmission electron microscopy (STEM) and energy-dispersive xray spectroscopy (EDX), and the resulting NP size and interparticle distance distributions were correlated with the optical Transmittance measurements and refractive index sensitivity (RIS), using T-LSPR spectroscopy.

Experimental details
Nanoplasmonic thin films composed of Au, Ag, or Au-Ag NPs dispersed in a TiO 2 dielectric matrix were prepared using a custom-made reactive DC magnetron sputtering system.The sputtering cathode was a rectangular pure titanium (99.99%) target with noble metal pellets placed in its erosion track, as summarized in table 1.The deposition parameters were optimized in previous work, suggesting that the use of a total noble metal content of approximately 20 at.%, and a thickness below 50 nm [32] are optimal conditions to enhance the LSPR sensitivity.In-depth chemical composition was obtained by Rutherford backscattering spectrometry [33], and the thickness of the thin films was obtained by cross-section scanning electron microscopy (NanoSEM-FEI Nova 200 (FEG/SEM) scanning electron microscope).Both the chemical composition and thickness of the films are displayed in table 1.
The target was sputtered with the current densities described in table 1, in a plasma composed of Ar and O 2 (3.8 × 10 −1 Pa and 3 × 10 −2 Pa, respectively), and a base pressure below 6.0 × 10 −4 Pa, resulting in TiO 2 thin films with embedded noble metal atoms aggregates, as illustrated in figure 1.
The thin films were deposited onto fused silica (SiO 2 ) and NaCl for optical measurements and TEM analysis, respectively.All substrates were plasma cleaned and activated.This procedure was performed using a Zepto Plasma System (Diener Electronic) using a 13.56 MHz generator at a power of 50 W and 80 Pa of working pressure, firstly for 5 min in O 2 , and then 15 min in Ar, for SiO 2 .To avoid damaging the NaCl substrates, a mixture of Ar and O 2 at the same working pressure for 1 min.The objective of this plasma treatment is to remove contaminants from the substrate and increase the thin film adhesion.To induce NP growth in the TiO 2 matrix and achieve LSPR responses, thin films were annealed at 400 • C.This temperature was selected because higher temperatures trigger the diffusion of noble metals to the thin film surface, especially Ag, forming larger particles and degrading the LSPR signal, in accordance with previous optimization studies [32].Additionally, with the objective of producing low-cost LSPR sensors, the annealing temperature should be kept as low as possible to allow the use of inexpensive substrates, such as glass).This thermal treatment was performed in a vacuum furnace with a base pressure of approximately 8 × 10 −6 Pa, with a heating ramp of 5 • C min −1 until it reaches 400 • C, an isothermal period of 5 h, followed by free cooling until it reaches room temperature.The annealing treatment promotes the diffusion of the dispersed noble metal atoms inside the TiO 2 matrix, crystallizing both the matrix in the anatase phase and the formed noble metal NPs.
For TEM analysis, the NaCl substrates containing annealed samples of Au, Ag, and Au-Ag/TiO 2 were dissolved in deionized water, and the floating layer was transferred to a copper grid.
The analysis was performed under high vacuum and at 300 kV, using an FEI-TITAN ETEM in the high-angle annular dark-field (HAADF)-STEM mode and obtaining images suitable for NP's analysis.These images were processed, resulting in thresholded black-and-white images.The resulting measurements were plotted in histograms and fitted to obtain the values of the average Feret diameter (MATLAB environment) and nearest neighbor (ImageJ environment).Furthermore, the noble metals present on the thin films were mapped using STEM-EDX in the same microscope with an X-MAX EDX detector (Oxford Instruments).
The bulk RIS was determined by monitoring the LSPR band shifts in the presence of media with different refractive indexes (figure S1).The measurement cycles were performed with deionized water (η = 1.3325RIU) and a 20% (w/w) sucrose solution (η = 1.3639RIU), with the transmittance spectra monitored for 2 min for each half-cycle.A custommade optical system consisting of an LED light source (LS-LED, SARSPEC, Lda), an enclosed thin film holder, and a modular spectrometer (SPEC RES + UV/Vis, SARSPEC.Lda) composed of a diffraction grating adjusted to the wavelength range of 420-720 nm and a CCD detector.Spectra were acquired with a 3 ms integration time and an average of 200 scans.NANOPTICS software allowed the processing of the acquired spectra and the changes in the LSPR band due to the presence of different surrounding media [34].

Optical response of the thin films
The thin film's optical response and LSPR bands were evaluated by spectrophotometry in transmittance mode, and the resulting spectra are presented in figure 2.
Considering the as-deposited thin films (figure 2(a)), as expected, no LSPR bands were found since the noncrystallized NPs formed by nucleation during the deposition process have first to reach sizes above the quantum limit (>10 nm) to contribute to the appearance of LSPR bands [28].After thermal annealing at 400 • C, all three nanoplasmonic systems display LSPR bands, each one with its minimum positioned at different wavelengths of the visible range (figure 2(b)).For the analysis of the LSPR bands, a similar approach to previously published studies was used [35].Analyzing the LSPR band using NANOPTICS software, the peak position for both wavelengths of transmittance minimum (λ min ) and transmittance minimum (T min ) was obtained.It also allowed the characterization of the band's full width at half height (FWHH) and the LSPR band's full height (BFH), which results from the difference between the T min in and the transmittance maximum at the band's left tail (table 2).
For Au/TiO 2 thin films annealed at 400 • C, the LSPR band has its minimum positioned at λ min = 638.0nm and T min = 26.0%.As seen in previous work, the LSPR band for Au/TiO 2 thin films has a flatter right tail, causing a high FWHH of 339.8 nm and a shorter BFH of 13.3 pp (percentage points).When considering the thin film of Ag/TiO 2 annealed at 400 • C, changing the noble metal to Ag causes the LSPR excitation to appear at lower wavelengths.The LSPR band that derives from Ag NPs usually has sharper LSPR bands due to higher extinction efficiency and at lower wavelengths, when compared to Au NPs (or other metals) [7].The LSPR peak is found at λ min = 560.6 nm and T min = 29.3%.As expected, the FWHH for the Ag/TiO 2 nanoplasmonic system is substantially lower, about 267.0 nm, while the BFH is slightly higher (17.4 pp), showing a narrower LSPR response.Finally, depositing both Au and Ag in the TiO 2 matrix with a 1:1 ratio led to an LSPR band positioned between the bands observed for monometallic Au and Ag systems.For Au-Ag/TiO 2 annealed at 400 • C, the LSPR band minimum is positioned at λ min = 612.0nm and T min = 22.1%.The LSPR band is broader than the LSPR band from Au/TiO 2 , with an FWHH of 358.1 nm but with a higher BFH (22.3 pp), showing a less flat right tail.

Nanoparticles' distribution.
A transmission electron microscopy investigation was performed to correlate the LSPR response of the different films with the size distribution of the NPs after annealing.For this, HAADF-STEM mode was mostly used, and the obtained images were processed in MATLAB software environment to produce thresholded black and white regions concerning matrix and NPs, respectively.The resulting Feret's diameters were plotted in histograms.
Figure 3 presents micrographs of Au-TiO 2 thin film annealed at 400 • C (figures 3(a)-(c)) at different magnifications.In the first analysis, most of the Au NPs (nearly 90%) showed sizes below 10 nm, which have a negligible contribution to the LSPR band, as they are below the quantum limit.
To statistically analyze the NPs, the threshold parameters were adjusted to ignore sizes below 10 nm.The resulting Feret's diameter was plotted in a histogram (figure 3(d)).
Considering only the NPs that directly impact the LSPR response (approximately 10% of the total NPs), around 70% have sizes between 10 and 20 nm, while the remaining 30% reveal sizes between 20 and 50 nm.The average Feret's diameter is 17 nm, with a broad distribution of sizes (σ = 8).The aspect ratio (AR) was also analyzed using the same method in the MATLAB environment.For Au/TiO 2 thin films annealed at 400 • C, the NPs were found with an average AR of 1.7 (σ = 0.7) (figure S2(a)-supplementary material), showing spheroid-like NPs.However, this analysis is limited due to the 2D nature of the HAADF-STEM images.
Changing the noble metal to Ag (Ag/TiO 2 thin films) caused a drastic change in the NPs sizes' distribution and overall aspect of the thin film (figures 4(a)-(c)).For Ag/TiO 2 thin films annealed at 400 • C, the same threshold conditions were used in the MATLAB environment.In comparison with Au/TiO 2 thin films, more than 95% have sizes above 10 nm and, hence, are contributing to the LSPR response (figure 4(d)).As such, there was no need to disregard the NPs with sizes below 10 nm, as they only constitute less than 5% of the total analyzed NPs.
Unlike Au/TiO 2 thin films annealed at 400 • C, the differences between the NPs and the matrix can clearly be distinguished.Ag NPs have an average Feret's diameter of 55 nm, with a much broader size distribution (σ = 28).The higher recorded Feret's diameter for Ag/TiO 2 thin films is 141 nm,   Finally, for the bimetallic Au-Ag nanoplasmonic system, the STEM analysis revealed some distinct nanostructural features compared to metallic counterparts, as can be observed in the micrographs of figures 5(a)-(c).Like Au/TiO 2 , a few bigger NPs were formed during the thermal annealing step, being surrounded by many small NPs.Size distribution analysis also revealed that most of them (80%) have sizes below 10 nm.So, a similar threshold was applied when analyzing the Au-Ag NPs in MATLAB environment, disregarding sizes below 10 nm.The resulting Feret's diameters were plotted in the histogram (figure 5(d)) and represent 20% of the NPs.These NPs have sizes typically ranging from 10 to 40 nm, with an average Feret's diameter of 22 nm, in a broad distribution of sizes (σ = 9).The average AR is about 1.4 (σ = 0.4), also showing spheroid-shaped NPs (figure S2(c)).
Besides the reported size distribution values, the most relevant characteristic is related to its unusual morphology.The Au-Ag NPs are organized into isolated clusters, each one containing a reasonable number of NPs.This type of nanostructural arrangement is certainly different from what would be  expected, and it promises different optical properties and plasmonic responses.Therefore, the micrographs revealing the clustering nanostructures obtained in the Au-Ag/TiO 2 thin films, annealed at 400 • C (figure 6(a)), were deeply analyzed in the ImageJ software environment, with thresholded black and white, similar to the analysis done in MATLAB, yet, this time, the nearest neighbor (N.N.) distance was evaluated, firstly between adjacent NP clusters (figure 6(b)) and then within each NP cluster (figure 6(c)).
When evaluating the N.N.distance between the NP clusters, the cluster's center was considered, and the N.N. was found at an average distance of 115 nm, with a broad distribution of distances (σ = 24), meaning that the clusters are relatively distant from each other.
Considering the N.N.distances within a cluster, each one was evaluated independently in the same software environment.The N.N. was found at an average distance of 32 nm (σ = 24).Since the N.N.measurement is done considering the centroid for each NP and the average Feret's diameter (22 nm), the average distance between the surface of two adjacent NPs can be as low as 10 nm.This means that Au-Ag NP hotspots could have been formed during the annealing at 400 • C. Plasmonic hotspots can occur when the distance between two adjacent NP surfaces is lower than the quantum limit (10 nm) and results in a significant enhancement of the EM field near the NPs [36][37][38].This strong enhancement of the EM field could prove useful for producing LSPR platforms for surface-enhanced Raman spectroscopy (SERS) [39,40], plasmon-enhanced photocatalysis [41], etc.

Elemental composition of NPs for Au-Ag/TiO 2 thin films.
To determine the chemical nature of the NPs in the bimetallic system, the Au-Ag/TiO 2 thin film was also probed using STEM-EDX (figure 7) by mapping the noble metal present in the thin film.
The chemical nature of the NPs was established by mapping both Au (figure 7(b)) and Ag (figure 7(c)) separately using EDX mapping.In previous work, the 1:2 ratio of Au to Ag content showed the possibility of the formation of Ag-enriched Au-Ag alloyed NPs, clearly visible by the higher accumulation of both noble metals in the bigger NPs, while Ag was also found throughout the matrix in smaller NPs [35].With the 1:1 ratio of Au to Ag, prepared in this study, both Au and Ag seem to concentrate on every single NP, with an even distribution, shown by the overlapping of both noble metal maps on the STEM image (figure 7(d)).This could indicate that with these deposition conditions, the resulting NPs are composed of an Au-Ag alloy, composed of both the smaller Au-Ag NPs that have a negligible contribution to LSPR and larger Au-Ag NPs (between 10-40 nm), contributing to the LSPR band.

Refractive index sensitivity for LSPR thin films
Intending to apply these nanoplasmonic systems as LSPRbased optical transducers for biosensing, with enhanced performance, RIS is the first parameter to evaluate, and it goes beyond the simple evaluation of the LSPR band spectra.For this reason, RIS was evaluated by measuring the LSPR bands of the thin films over time while immersed in media with different refractive indexes.Cycles of alternating deionized water (η = 1.3325RIU) and a 20% (w/w) sucrose solution (η = 1.3639RIU) allowed to obtain several spectra that were processed using the NANOPTICS software.
Firstly, in the case of Ag/TiO 2 thin films annealed at 400 • C, an increased instability of the LSPR band signal was observed when the thin film was immersed in deionized water, with the transmittance spectrum revealing a higher noise than expected.
When this happens, the NANOPTICS software is unable to produce reliable results from the cycles' measurements, and hence, Ag/TiO 2 film was disregarded for liquid environments.
As for Au/TiO 2 and Au-Ag/TiO 2 annealed at 400 • C, the resulting cycles from the RIS experiments are presented in figure 8.
The cycles concerning the RIS evaluation for Au/TiO 2 thin films annealed at 400 • C are presented in figure 8(a).At first glance, some noise during the cycles' measurements is visible.After analyzing each cycle, the average LSPR band shift was 2.5 ± 0.1 nm, with a signal-to-noise ratio (SNR) of 8.4.Using the refractive indexes for each media, a RIS of 80 ± 4 nm RIU −1 was calculated.
For the RIS evaluation of Au-Ag/TiO 2 thin films annealed at 400 • C, the measured cycles are displayed in figure 8(b).In contrast with the analysis made for Au/TiO 2 , the measurements have much less visible noise.Analysis of each cycle revealed an average LSPR shift of 4.51 ± 0.03 nm and an SNR of 57.9.The calculated RIS for Au-Ag/TiO 2 thin films was 147.7 ± 0.9 nm RIU −1 .
As evidenced by the results displayed in figure 8, adding Ag to the Au/TiO 2 nanoplasmonic system in equal parts allowed almost double the LSPR response in the same experimental conditions.Firstly, this improvement could be due to the presence of alloyed Au-Ag NPs.From Mie's theory, it is known that Ag NPs have higher absorption and scattering efficiencies when compared to Au NPs [42].As a result of being more efficient in scattering light at the resonance conditions for LSPR, Ag NPs have a higher responsiveness to environmental changes [43].As such, the stable mixing of Au and Ag into bimetallic NPs benefits from the Ag properties, achieving higher extinction efficiencies [44].On the other hand, the increase in NPs size from Au/TiO 2 to Au-Ag/TiO 2 can also contribute to the higher sensitivity of Au-Ag NPs since a higher average NP diameter is also associated with higher extinction efficiencies, and thus improved sensitivity to the surrounding media [45].Therefore, these optimized properties seem to provide a better signal, with a much higher SNR and increased sensitivity for different refractive indexes, which could allow the production of a more reliable and stable system to develop LSPR-based optical transducers for (bio)sensing.
Finally, based on the results obtained by STEM, it can also be claimed that the formation of clusters of NPs, the latter close to each other (originating hotspots), and consequent near-field enhancement might also have contributed to an enhanced sensitivity of Au-Ag NPs.In the literature, the effect of plasmonic hotspots on LSPR sensitivity is a controversial subject.For instance, Feuz et al published a study comparing the sensing capabilities of nanodiscs and hot spots between two adjacent structures, showing that, while the hotspot presented an approximately 20% lower RIS, the SNR increased approximately 20 times [46].Another work published by Yockell-Lelièvre et al dwells on the formation of self-assembled Au NPs-based sensors.It shows that when the interparticle distance is small (3 nm gap) and plasmonic hotspots are formed, no RIS is measurable.Still, the SERS enhancement was reported to increase up to 3 orders of magnitude [47].Lastly, Lee et al described the influence of nanogaps in highly advanced plasmonic structures, with the highest RIS found where the highest EM confinement is expected, suggesting that hotspots improve the sensibility of plasmonic nanostructures [48].As such, continuous study on this topic is still needed to determine the effects of hotspots on the sensibility of LSPR-based optical transducers.

Conclusions
This work reports on the comparison between nanoplasmonic systems composed of Au, Ag, or Au-Ag alloyed NPs dispersed in a TiO 2 dielectric matrix in terms of NPs distribution and optical response of the LSPR band.
The nanostructural analysis using transmission electron microscopy showed the formation of NPs after annealing at 400 • C for all nanoplasmonic systems.For Au/TiO 2 thin films, while most NPs have sizes below the quantum limit and did not contribute to the LSPR phenomenon, approximately 10% of the NPs were found with an average size of 17 nm, contributing to a RIS of 80 nm RIU −1 .Using silver instead of gold in the production of nanoplasmonic thin films, led to the production of Ag/TiO 2 thin films with embedded Ag NPs.The average size of the NPs was 55 nm (increasing three times when compared to Au/TiO 2 thin films), yet it was not possible to calculate the RIS for these plasmonic films, due to instability of the LSPR band signal when the thin film was immersed in liquid environment.Finally, the formation of Au-Ag NPs in the bimetallic nanoplasmonic system can be highlighted.The NPs can be divided into two groups: the large majority dispersed in the TiO 2 matrix with sizes below 10 nm, and the remaining organized into clusters of larger NPs (sizes above 10 nm), contributing to the LSPR phenomenon, with an improved sensitivity for changes in the refractive index, giving rise to an outstanding RIS of 147.7 nm RIU −1 .
In conclusion, electron microscopy techniques allowed to discern nanoscale differences between the nanoplasmonic systems in the study and correlate the improved capabilities of Au-Ag/TiO 2 thin films with their morphological features for LSPR optical sensing.The nanoscale analysis also showed the possibility of applying these thin films as SERS platforms or in photocatalysis due to the formation of plasmonic hotspots in the NP clusters.

Data availability statement
The data that support the findings of this study are available upon reasonable request from the authors.

Figure 6 .
Figure 6.(a) Au-Ag NP clusters of nanoparticles obtained by STEM; (b) histograms for nearest neighbor distance distribution, considering the distance between clusters; (c) histograms for nearest neighbor distance distribution, considering the distance between NPs within a cluster.

Figure 7 .
Figure 7. HAADF-STEM and STEM-EDX analysis of nanoparticles present in Au-Ag/TiO 2 thin film annealed at 400 • C: (a) TEM image of the selected area; (b) Au and (c) Ag EDX maps in the selected area; and (d) Au and Ag maps overlapping with the HAADF-STEM image.

Figure 8 .
Figure 8. LSPR band minimum wavelength shift, processed by NANOPTICS software, for thin films, annealed at 400 • C, showing the cycles due to the change of refractive index of the surrounding media (deionized water vs. a model-solution of 20% (w/w) of sucrose).The last cycle is the reference cycle (water vs water).

Table 1 .
Deposition parameters for nanoplasmonic thin films concerning noble metal content and plasma current density.

Figure 1 .
Schematic representation of the two-step procedure to achieve TiO 2 thin films with embedded noble metal NPs, starting from (a) transparent substrate; (b) the Ti-noble metal composite target is sputtered onto a substrate, resulting in a TiO 2 thin film with dispersed noble metal; and (c) the annealing procedure is conducted to promote noble metal crystallization and NPs growth inside the TiO 2 matrix.

Table 2 .
Summary of LSPR band minimum position (λ min and T min ), FWHH, and BFH for nanoplasmonic thin films determined by NANOPTICS.