1. Introduction

Over the last decades, plasmonics has been almost exclusively studied by using gold or silver nanostructures, exhibiting plasmonic properties from the infrared to the visible spectral region [1]. Due to inherent limitations related to their electronic structures, these metals do not sustain plasmonic resonances in the ultraviolet (UV). Nevertheless, emerging applications will require the extension of plasmonics toward higher energies, particularly in the UV-range. Aluminum (Al) sustains plasmonic resonances in the UV and visible range while keeping relatively low losses, and is now widely regarded as one of the most promising metal for pushing the spectral limits of plasmonics towards high energies [2–4]. Al plasmonics is efficient for numerous potential applications including UV-SERS (Surface Enhanced Raman Scattering) [5,6], optoelectronics [7], structural color [8], solar desalination [9], non-linear optics [10], and photocatalysis [11]. Additionally, Al is cheap and abundant, since it is the third most abundant element on the earth. Last but not the least, Al is compatible with metal-oxide-semiconductor (CMOS) technology [12]. That being said, oxidation of Al nanostructures is an unavoidable issue, which has to be taken into account. It is well known that Al experiences an unavoidable natural oxidation process. When exposed to air, Al surface oxidizes quickly, resulting in a 3-5 nm amorphous layer [13,14]. This dense layer is believed to slow down the subsequent oxidation by preventing oxygen diffusion, acting as a passivation layer [15]. Because of their wide use as solid propellants and explosives, oxidation of Al particles has been investigated since the last decades [16]. To prevent further oxidation of Al nanostructures and improve their long-term stability, effective surface passivation agents like transition metals or organic coating were studied using wet-chemical method [17,18]. Nevertheless, the above mentioned studies about Al nanoparticles oxidation were mostly performed for combustion-based issues and did not focus on their optical properties. Due to their polycrystalline nature, evaporated Al nanostructures contain defects such as grain boundaries and exhibit a certain roughness. Consequently, the long-term stability of their optical properties should be investigated, as Al nanostructures are widely produced and used in research facilities and proposed for future technological applications. This is an important detail which has not been studied thoroughly although it is mentioned in literature [19,20]. In this article, we investigate the plasmonic properties of thermally evaporated Al nanodisk arrays fabricated by electron beam lithography (EBL) over a long period of time, typically 250 days. Their optical properties are preserved for 90 days when stored in ambient air, whatever their diameter, due to a very slight oxidation during this period. Over the following months, the nanodisks loose their optical properties more or less rapidly depending, this time on their diameter. We explain these effects by a competition between oxidation and self-annealing. Particularly, nanodisks with a diameter of 100 nm are surprisingly stable, sustaining plasmonic resonances lasting over 250 days. Furthermore, when Al nanodisks are exposed to water environment, their lifetime and optical properties are shortened to 5 days due to corrosion. The related results are of importance for further use of conventionally evaporated Al nanostructures for research or optical applications, as they should remain stable for a long period of time.

2. Experiment and methods

Finite-difference time-domain (FDTD) simulations were performed using FDTD Solutions Lumerical to compute the extinction spectra of Al nanodisk arrays. To simulate the periodic array, periodic boundary conditions in the x and y directions, and perfectly matched layer in the z direction were used with a minimum mesh size of 0.5 nm. The dielectric functions of the different materials were all taken from the database of the software (Palik data for Al and Al₂O₃). Normal incident plane wave (200-700 nm) located in the quartz substrate was chosen as the excitation source. The transmission spectra (Q_tra) were obtained directly using a power monitor located over the NP in air. Extinction spectra (Q_ext) were calculated by Q_ext = 1 − Q_tra. The periodic nanodisk arrays of 40 × 40 µm² (diameter D = 70, 100, 130, and 160 nm, height 50 nm, pitch 2.5D) were fabricated using EBL process (E-Line Raith, area dose 175 μC/cm²) on quartz substrate. The experimental results shown in this article come from the sample on the same substrate under the same thermal evaporation cycle (Plassys MEB 400, in a vacuum 5 × 10⁻⁶ Torr at a rate of 0.2 nm/s). Scanning Electron Microscopy (SEM) was performed using a Field Emission Gun (FEG Hitachi SU8030) under 1 kV. UV-Visible extinction spectra were acquired using a homemade microscope with spatial confocal filtering through an optical fiber.

3. Results and discussions

3.1. Models to simulate the oxidation

FDTD simulations were used to calculate the extinction spectra of Al nanodisk arrays with an increasing oxide layer of alumina (Al₂O₃) around each nanodisk. In order to obtain the extinction spectrum for different values of thickness, a simplified model is adopted, as depicted to the left in Fig. 1(a). In this model, the volume of the nanodisk is kept constant during the oxidation, corresponding to the initial volume of the unoxidized structure. Then, the uniform Al₂O₃ layer of thickness L is continuously increased from zero until the disappearance of the metallic core. Nevertheless, an extension of the volume of Al nanostructures occurs during the oxidation (V_Al: V_Al2O3 ≈0.8), due to their different densities. If we consider this effect, an additional thickness ∆ of oxide has to be added, as pictured to the right in Fig. 1(a). This improved model presumes that a uniform oxide layer is formed during an isotropic linear expansion process without shape deforming (the height and radius have the same oxidation rate). The increased total height and radius due to the expansion as a function of the metallic core height or radius can be calculated as shown in Fig. 1(b-c). The thickness of oxide layer is plotted in function of the Al core radius in Fig. 1(d) for the two models. At the initial stage of oxidation (metallic core radius decreased by 10 nm), the difference of oxide layer thickness between the two models is about 1.5 nm. Even if considering a full oxidation (the metallic core is completely oxidized), the difference between the two models is only about 4 nm. Similar results have been observed for different sizes of nanodisks (not shown here). For smaller disks (diameter 50 nm, height 50 nm), the maximum difference is less than 2.5 nm. For larger disks (diameter 160 nm, height 50 nm), the maximum value of this difference is around 5 nm.

 

Fig. 1 (a) Schematic representation of the simplified and expanded models. Diameter D, metallic core radius r, oxide layer thickness L for the simplified model. Additional thickness ∆ of oxide layer in the expanded model. (b-d) Calculations obtained for an Al nanodisk (radius 50 nm and height 50 nm) with the expanded model: total radius (b) and height (c) respectively plotted as a function of the Al core radius and height; (d) comparison of oxide layer thickness between the simplified model (no volume expansion) and the expanded model (taking the volume expansion during oxidation into account).

Download Full Size | PDF

The additional thickness of Al₂O₃ in the expanded model will induce a stronger redshift of the localized surface plasmon resonances (LSPR) sustained by the Al nanodisks than the simplified model [21]. However, the discrepancies between the two models are relatively small. In practice, taking into account the expanded volume during oxidation demands a much thinner mesh in the FDTD simulations, thus dramatically increasing the computing time. Moreover, the simulated extinction spectra obtained from the simplified model are in good agreement with the experimental results, as it will be shown below. Consequently, the simplified model is used in the following to perform the FDTD calculations.

3.2. Calculated extinction spectra

The extinction spectra were calculated for Al nanodisk arrays (diameter D = 70, 100, 130, and 160 nm, height 50 nm, pitch 2.5D, named D70, D100, D130, and D160 respectively in the following) with an increasing thickness of oxide layer. The extinction spectra obtained from the simulations are displayed in Fig. 2(a). The dipolar resonances are obvious, ranging from 280 to 600 nm, depending on the diameter of the considered nanodisk. One can also observe the evolution of the resonances according to the oxide layer thickness. Note that higher modes, attributed to quadrupolar resonance, appear at shorter wavelengths (200-300 nm) in the extinction spectra, particularly noticeable for D70 and D100. These modes tend to redshift and vanish with the increasing thickness of alumina. We attribute their signature to hybridized modes, a mix of pure dipolar and quadrupolar eigenmodes sustained by the nanostructure in vacuum, which emerge from the image response of the substrate. They have been already studied in Ag nanocubes, for instance [22]. As they are very confined near the metallic core, they experience a more pronounced redshift and intensity decrease than the dipolar resonances with oxidation. Finally, as the studied system consists in arrays, it may also experience some electromagnetic coupling between each nanodisks (acting as scatterers). This effect is due to interference of the plasmonic resonances and the grating Rayleigh anomalies resulting in the so-called lattice modes. These hybrid modes exhibit also a Fano-like lineshape, effect well visible on the calculated spectra, especially for D130 and D160. The dashed lines plotted in Fig. 2 correspond to the calculated Rayleigh anomalies for both the substrate and air refractive indexes. See reference [23] for an exact analysis of these modes in Al nanodisks arrays.

 

Fig. 2 (a) Calculated extinction spectra of Al nanodisk arrays (diameter D = 70, 100, 130, 160 nm, height 50 nm, and pitch 2.5D) with increasing oxide layer. The scale bar stands for the extinction intensity. Spectral shift (b) and maximum intensity (c) of the dipolar resonance peak normalized with the spectrum of nanodisk arrays without oxide layer, plotted in function of the oxide layer thickness. The color of each box in (a) corresponds the line color indicating the size of nanodisk in (b) and (c).

Download Full Size | PDF

Now, we focus on the dipolar mode and its evolution with oxidation. The spectral shift of the dipolar resonance and its intensity, which are normalized with their value taken from the spectra corresponding to the nanodisks with no oxidation (L = 0 nm), are plotted in Fig. 2(b) and (c). First, the resonances sustained by smaller nanodisks (D70, and D100) experience a redshift until a certain oxide layer thickness (up to about 12 nm). For thicker oxide layer, this effect is counterbalanced by a blueshift until the disappearance of LSPR (meaning that the Al core is almost completely transformed into oxide). Second, the maximum value of the spectral redshift attainable with oxidation decreases when the size of nanodisk increases. The smallest nanodisks (D70) sustain dipolar resonances with a maximum redshift of 8 nm due to oxidation, while larger nanodisks (D100) present a smaller value of maximum redshift, equals to 5 nm. For nanodisks with larger diameters (D130 and D160), the behavior is different: their dipolar resonances keep an almost constant maximum value up to an oxide layer thickness of 30 nm, with no significant spectral shift. For thicker oxide layer, they experience a pronounced blueshift and their intensity is continuously reduced, until they vanish. The decreasing size of Al core induces a blueshift of the resonance, while the increase of the Al₂O₃ layer induces a redshift [24]. For smaller nanodisks (D70 and D100), the increase in the shell thickness dominates until it reaches a value of 12 nm, and then the shrinkage of Al becomes dominant until the disappearance of the resonance due to full oxidation. In the case of larger structures (D130 and D160), these two mechanisms offset each other until the metallic core volume is greatly reduced (L ≈ 25 nm). This results in a constant value of the resonance wavelength during oxidation. Another group has previously reported this phenomenon [25].

The maximum intensity of the dipolar resonance decreases continuously as the thickness of Al₂O₃ layer increases regardless of the diameters as shown in Fig. 2(c). However, the intensity of the resonances sustained by the smaller nanodisks experiences a much faster decreasing rate than the larger nanodisks. This is attributed to the ratio between the metallic core and the increasing oxide layer, which decreases much faster for smaller nanodisks during oxidation. The maximum intensity of the resonance is therefore a relevant and complementary parameter to the spectral shift of the resonance, in order to determine the degree of oxidation of the Al nanostructure. This is particularly true for the resonances sustained by larger nanodisks, as they sustain LSPR which do not exhibit significant spectral shift even for a relatively strong oxidation. The monitoring of the plasmonic resonance intensity and spectral position could act as a calibration process giving the oxidation degree of a particular Al nanostructure. Nevertheless, the calculations presented here do not take into account effects like self-annealing or recrystallization, which will be encountered in experiments as discussed in the following.

3.3. Long-time experimental measurements

Al nanodisk arrays (diameter D = 70, 100, 130, and 160 nm, height 50 nm, pitch 2.5D) were fabricated using EBL and thermal evaporation. After fabrication, the samples were exposed to ambient air in cleanroom (T = 21 ^◦C, relative humidity around 40%) and extinction spectra were measured continuously during a long-term period, over 250 days. Experimental extinction spectra are plotted as a function of time in Fig. 3(a), for the four values of diameter. Figure 3(b) and (c) show the spectral shift and the maximum intensity of the dipolar resonance as a function of time, which are normalized with their values taken from the spectra corresponding to the measurements on the first day. The first important observation is the stability of the plasmonic resonances sustained by Al nanodisks over a period of roughly 90 days, whatever their dimensions. This is an important result as thermally evaporated Al nanostructures with significant roughness (5 nm RMS measured) retain their plasmonic properties for a relatively long time. This period is long enough to study Al nanostructures in research facilities, but it is of course detrimental for their long-term use. Let us also note that high order modes do not appear on the experimental spectra. We attribute that to the roughness of the nanostructures, as also evidenced by numerical simulations (not shown here). We will report this effect elsewhere. Asymmetrical profiles visible on larger nanodisks are attributed to lattice coupling as mentioned above, and will not be discussed here.

 

Fig. 3 (a) Experimental extinction spectra of Al nanodisk arrays (diameter D = 70, 100, 130, and 160 nm, height 50 nm, pitch 2.5D) as a function of time. The scale bar stands for the extinction intensity. (b) Peak shift and (c) peak intensity of the resonance normalized with the spectrum measured on the first day. The color of each box in (a) corresponds to the line color in (b) and (c) indicating the size of nanodisk.

Download Full Size | PDF

Within about 90 days from the day of fabrication of the sample, the dipolar resonances do not shift significantly for larger nanodisks (D130 and D160) while keeping a constant maximum value. For smaller nanodisks (D70 and D100), they are slightly shifted towards longer wavelengths.

This indicates that oxidation governs the structural evolution of the Al nanodisks over the first 90 days, at least partly. Nevertheless, another phenomenon, which has not been predicted by the calculations related above, occurs in experiments. Indeed, one can observe a slight increase of the extinction intensity over time in Fig. 3(c), except for smaller nanodisks (D70), which could be ascribed to self-annealing of Al nanodisks as explained below. This effect lasts for a surprisingly long time on the arrays containing the nanodisks of 100 nm in diameter (D100). The latter sustains a very stable dipolar resonance, still visible on days 250. Even if after a year, it still preserves a strong plasmonic resonance. For the two larger diameters, this effect is weaker and temporary, occurring roughly between day 15 and day 70. Self-annealing could also give rise to a blueshift of the resonance, actually noticeable for D100 in Fig. 3(b). Self-annealing effect is actually a common phenomenon widely observed for various materials with different fabrication methods during their long-term storages at room temperature [26–29]. The driving force of this natural recrystallization is believed to be related with the internal strains and defects linked to atomic dislocations. Factors like deposition conditions and the concentration of impurities play an important role in the natural grain growth rate. Evaporated metallic nanostructures contain more defects (such as grain boundaries) than those obtained by chemical methods [30,31]. Additionally, thermal evaporation could induce a high strain inside the Al nanostructure, resulting in a higher enthalpy state [32,33]. Therefore, due to thermodynamic effects, Al nanostructures could release energy by increasing grain sizes naturally, leading to a slow spontaneous annealing at room temperature over a long period of time.

As a consequence, the experimental results presented in Fig. 3 could be interpreted as a balance between self-annealing and oxidation. After a relative stability of 90 days, the dipolar resonances sustained by the small nanodisks (D70) experience a continuous decrease in intensity and redshift due to oxidation. Indeed, the larger surface to volume ratio of smaller nanodisks could increase the oxidation rate, and self-annealing effect is not observed for these ones. For larger nanodisks (D130 and D160), the predicted resonance maximum wavelength is nearly constant during the initial oxidation, corresponding to the first 10-15 nm of oxide shell growth as shown previously by calculations in Fig. 2(b). This behavior lasts for 90 days in experiments as shown in Fig. 3(b). Then, D130 and D160 show increased oxidation after about 90 days, confirmed by slightly blueshifted peaks and continuously decreased intensities, as also predicted by calculations. Self-annealing is barely visible for these two diameters, as only a slight increase of the maximum intensity of the dipolar resonance is visible in Fig. 3(c), with no measurable blue shift. The diffusion of oxygen during the oxidation process is likely to occur along the grain boundaries and defects inside the nanostructure. Therefore, more defects inside larger nanodisks will facilitate the oxidation. The balance happens for D100 as noticed above: the nanodisks sustain a surprisingly stable resonance during 250 days. These results are likely due to self-annealing, which compensates oxidation during the observing period. The maximum value of the resonance peak measured from the D100 sample is therefore increased by 20% from day 1 to day 250, accompanied with a blue shift of 6 nm, measured on individual spectra taken from Fig. 3(a) (available on request). The reproducibility of the results is confirmed by several samples fabricated at different times.

In our experiments, evaporated nanodisks sustained stable plasmonic properties within 90 days after their exposure to ambient air. Samples made by different methods might show different performances due to changes inside the nanostructure such as grain boundaries. Actually, the natural oxidation of Al nanostructures during the period of storage is driven by a complicate process, involving the environment (such as humidity and temperature), interface diffusion of aluminum cations and oxygen anions, possible morphology change due to the different mechanical properties of metallic core and oxide shell, and phase transformation [6,14,34–36]. Although the natural oxide layer formed around Al nanostructures is believed to stabilize their optical property, a more complex behavior including self-annealing is unveiled for evaporated Al nanodisks in our experiments. These experimental results are of importance as they prove that long-time stability of evaporated Al nanostructures does not last indefinitely. If longer stability is further required, strategies like encapsulation exist to prolong and stabilize the plasmonic properties of evaporated Al nanostructures [17,18,37]. Another solution is the epitaxial growth of aluminum thin films or nanostructures, which sustain very stable optical properties [38], but this method is much more expensive and time-consuming.

3.4. Corrosion in water

Al nanodisk arrays were immersed in distilled water (40 ml in a beaker, pH = 7, temperature 21 ^◦C) in order to test their resistance to water corrosion. Measured extinction spectra are plotted in Fig. 4 before and after immersion. The extinction spectra measured on the samples immersed for 16 hours were strongly modified compared to pristine samples. A clear diminution of the extinction peak intensity is observed as well as a strong redshift. After 5 days, there is almost no plasmonic resonances signature, unveiling strongly oxidized nanostructures. Due to the reaction between water and Al, the immersion of the nanostructures in water accelerates their oxidation with complex aluminate hydrates. This results in a morphological and structural alteration of the nanostructures as evidenced by SEM images presented in Fig. 4. Al nanodisks became highly inhomogeneous with a rough surface compared with pristine ones. This evidences the strong corrosive effect of water to Al nanodisks. Markus Schwind et al. made a full investigation of corrosion kinetics on Al nanodisk arrays in a well-controlled chamber filled with degassed water [39]. An initial slow growth of a homogeneous oxide layer within 2 days was observed followed by a rapid oxidation. Our results are in good agreement with their experiments. The role of the environment to store or use Al nanostructures is of importance, especially if it is a liquid. For instance, in SERS measurements, where it is common to drop cast liquids on the sample, the use of water with uncontrolled pH and impurities should be avoided. Moreover, LSPR chemical sensing, based on the functionalization of nanostructures with key molecules in water-based solutions is not trivial with Al nanostructures due to corrosion.

 

Fig. 4 Experimental extinction spectra measured on Al nanodisk arrays (diameter D = 70, 100, 130, and 160 nm, height 50 nm, pitch 2.5D) before (solid line) and after (dashed line) being stored in water for 16 hours. Corresponding SEM images show the nanodisk arrays before (left side, solid line box) and after 5 days (right side, dashed line box) of immersion in distilled water. Scale bar 500 nm.

Download Full Size | PDF

4. Conclusions

The long-term stability of the optical properties of aluminum nanostructures over time is crucial for further applications like structural color and biosensing, and should be considered for large-scale commercial production in the future. In this article, the long-term stability of aluminum nanodisk arrays with varying sizes is investigated. By using FDTD means, the evolution of the plasmonic resonances sustained by the Al nanodisks during the oxidation is unveiled using a simplified model for the oxidation growth. More precisely, the evolution of the plasmonic properties is linked to the geometrical parameters of the nanodisks and the thickness of their oxide layer, allowing a simple calibration process to determine the degree of their oxidation. In the long-term experiments, Al nanodisk arrays were exposed to ambient air. A 90-day period of stability was observed whatever their diameter. The nanodisks with a diameter of 100 nm were found to be much more stable than others, likely due to counterbalanced effects between self-annealing of the nanostructure and its natural oxidation. By immersing samples into water, the strong corrosion shortens the lifetimes of Al plasmonic resonances, lasting no more than 5 days. The results reported here are of importance as they conclude that appropriate treatments are needed to extend the lifetime of evaporated Al nanostructures in order to stabilize their plasmonic properties for more than 90 days.