1. Introduction

The optical properties of metal nanoparticles (MNPs) have been extensively studied for decades [1,2]. Typically, the change of the electromagnetic (EM) field around MNPs is dominated by the collection of induced oscillation of conduction electrons in MNPs, which is also known as localized surface plasmon resonance (LSPR). This resonance causes a redistribution of charges in the particle, generating multipole moments in response to the EM field. In 1908, Gustav Mie presented a solution to Maxwell’s equations in the form of infinite series of multipole wave functions [3], which can be used to describe the scattering and absorption spectra of spherical particles. LSPR can generate strong scattering and absorption, resulting in an optical cross-section that is tens of times larger than the geometric cross-section of the particles. If the MNPs are placed on the interface between two dielectric materials, most of the scattered light will be prone to traveling toward the material with higher refractive index [4,5], offering a promising light trapping strategy for many dielectric materials. The LSPR of such a system is highly sensitive to the morphology and metallic properties of the MNPs and the dielectric properties of the surrounding materials. Therefore, in addition to theoretical work, there has been growing interest in tuning the spectral properties of LSPR by adjusting the particle size, density, and dielectric environment [6–11]. Such unique and tunable properties of LSPR can be applied to various optoelectronic technologies including surface-enhanced Raman scattering (SERS) [12–15], biological and chemical sensing [16–20], metamaterials [21–23], and photovoltaic cells [24–28].

Gold (Au) nanoparticles (NPs) are one of the noble metal nanoparticles (NMNPs) that have been widely used due to its stability in both liquid and atmosphere [29]. Another kind of NMNPs, silver (Ag) NPs, have a better overall scattering efficiency (Q_sca) than Au NPs in the ultraviolet-visible (UV-vis) region, but their application is limited because Ag tends to react with the atmosphere and degrades rather rapidly [30]. While Pinkhasova et al. [31] shows that Ag₂O could easily decompose in atmosphere at 400 ˚C, the potential of forming silver sulfide (Ag₂S­) and silver chloride (AgCl) remains an issue for the stability of Ag NPs even at room temperature [32]. In order to avoid the atmospheric degradation and maintain long-term optical performance, the Ag NPs have to be protected by dielectric coatings such as titanium dioxide (TiO₂) [33], aluminum oxide (Al₂O₃) [30,34] and silicon nitride (Si₃N₄) [35].

The most common process to form Ag NPs on substrates is the vacuum-based thermal evaporation of a metal film followed by annealing at a moderate temperature (200-300 ˚C) for hours [27], while some other methods include photocatalytic growth [36], the direct use of colloidal MNPs [37,38], and chemical synthesis [31,39]. Recently, a low-cost vacuum-free electroless plating method has been used for the deposition of Ag NPs on silicon (Si) substrates [15]. A typical non-vacuum electroless plating solution contains silver nitrate (AgNO₃) and hydrofluoric acid (HF) and is commonly used in metal-catalyzed chemical etching (MACE) where Si atoms are etched by Ag via the galvanic displacement reaction (GDR) [40–42]:

In this work, a thin layer of dense Ag NPs is deposited on the surface of Si substrate by electroless plating, followed by a water bath at 90 ˚C to form Ag NPs of different sizes. The fabricated Ag NPs are exposed in laboratory environment for 2 weeks to study the effect of Ag degradation in atmosphere. To isolate the NPs from the atmosphere, a conformal layer of Al₂O₃ is coated on the samples using atomic layer deposition (ALD). A scanning electron microscope (SEM) is used to obtain nanoscale images of Ag NPs on the Si surface. An atomic force microscope (AFM) is used for imaging samples coated with Al₂O₃. The size distribution of NPs on the surface is obtained by analyzing the SEM images. The reflection spectrum of samples in the UV-VIS region is measured in a spectrometer with an integrating sphere.

2. Experimental methods

2.1 MATLAB simulation for spherical Ag NPs

A MATLAB analytical model is utilized to simulate the LSPR of spherical Ag NPs for comparison with the experimental data. The model is based on the work of Mätzler which is able to obtain the quantitative solutions to the Mie theory of scattering [43]. In the model, the LSPR of a single spherical NP of a given size and optical constants in a uniform environment is derived by computing a number of multipole wave functions up to a finite truncation [44]. LSPR due to a single multipole moment can be extracted by computing the corresponding term in the series. The dielectric environment is assumed to be lossless and non-dispersive so that only a universal refractive index N is used, while the Ag NP itself has dispersive refractive index n(λ) and extinction coefficient k(λ). The dispersive dielectric constants n and k are obtained from values reported in Wu et al. [45] for atomically smooth epitaxial Ag films. N is chosen to be 1 and 1.6 to simulate the environment of air and Al₂O₃, respectively. The calculated absorption efficiency Q_abs (or scattering efficiency Q_sca) is defined as the ratio of the absorption (or scattering) cross-section to the geometrical cross-section of the NP on the plane perpendicular to the direction of the incident light. The extinction efficiency Q_ext is the sum of Q_abs and Q_sca that represents the total amount of light that deviates from the direction of incidence due to the interaction with a NP.

2.2 Fabrication of Ag NPs

Commercially available single-side polished two-inch mono-crystalline Si (100) Czochralski (CZ) wafers (p-type, 1-10 Ω-cm, University Wafer) are first immersed in a solution containing 90% sulfuric acid (H₂SO₄) and 1% hydrogen peroxide (H₂O₂) (Nano-strip, Cyantek) for 5 min. They are then cleaned using the standard Radio Corporation of America (RCA) protocol. Prior to electroless plating, the cleaned Si substrate is pretreated in a 2 min 10% HF solution dip to make the surface highly hydrophobic. The substrate is then immediately submerged in the Ag plating solution consisting of DI water and various concentrations of HF and AgNO₃. HF is varied at two different concentrations of 2.4 M and 4.8 M. The concentration of AgNO₃ is varied at 1 mM, 2 mM, and 4 mM. To explore the different topographies that can be formed with different plating times, the substrate is exposed to the solution for 5 s, 10 s, or 20 s. After the substrate is plated, the sample is subjected to heat treatment to help the formation of NPs. The sample is immersed in a 90˚C water bath for 2 min and finally dried using a dry nitrogen (N₂) gun. Surface characterization is performed after the completion of the sample.

2.3 ALD of Al₂O₃ on samples

Two Ag NP plated samples and a bare Si substrate is coated simultaneously with a conformal dielectric layer of Al₂O₃ deposited using an oxygen (O₂) plasma enhanced ALD (PE-ALD) (OpAL, Oxford Instruments) with a trimethylaluminum (TMA) metal precursor. Using the spectroscopic ellipsometer (M-2000X, J.A. Woollam), the deposited Al₂O₃ layer is measured to be 27 nm thick with a refractive index of 1.62 at 632.8 nm.

2.4 Reflection measurements and imaging

Reflection measurements are taken on a Cary spectrophotometer (Cary 5000 UV-Vis-NIR, Agilent). The spectrum weighted average reflectance (R_ave) between 300 to 800 nm is computed using the standard radiance of AM1.5G [46]. The percent reflection difference (ΔR%) is calculated by subtracting the reflection of the plated sample (R_Ag) from the reflection of a bare Si wafer (R_Si):

3. Results and discussion

The dispersive Mie efficiencies Q_sca and Q_ext of spherical Ag NPs in a uniform medium are shown in Fig. 1. It can be seen from Fig. 1(a) that a small Ag sphere 20 nm in diameter resembles an electric dipole with a single LSPR peak. When the medium material changes from air (N = 1) to Al₂O₃ (N = 1.6), the peak Q_ext value shifts from 8.6 at 356 nm to 28.7 at 416 nm, implicating a redshift of the peak and a stronger interaction with the incident light. For a larger particle with a diameter of 60 nm in air (Fig. 1(b)), the Q_ext peak is 14.3 at 367 nm, which is stronger than and slightly redshifted from the Q_ext of a 20 nm Ag NP. When N increases to 1.6, not only does the dipole Q_ext peak shift to 400 nm, but a smaller resonance due to the quadrupole also occurs at 466 nm. For the case of a 100 nm Ag NP (Fig. 1(c)), due to the growing interaction between the incident radiance and higher multipole moments, more than one peaks appear in the visible spectrum region (at 353 nm and 391 nm for N = 1, and at 393 nm, 426 nm and 554 nm for N = 1.6), and the total LSPR becomes broadband.

 

Fig. 1. The calculated Mie extinction (blue) and scattering (red) efficiencies for a spherical Ag NP with a diameter of (a) 20 nm, (b) 60 nm, or (c) 100 nm in a uniform medium with a refractive index of N = 1.0 (solid) or N = 1.6 (dashed).

Download Full Size | PDF

Meanwhile, as the particle size increases, the ratio of Q_sca to Q_ext grows rapidly because the scattering intensity grows much faster than the absorption intensity, which is consistent with the fact that the Rayleigh scattering has a size dependence of D⁶, where D is the diameter of the Ag nanoparticle, while the absorption is only proportional to the volume (∼D³) [44]. When N = 1, the nearly zero Q_ext shoulder at about 325 nm does not shift with the varying particle sizes. Such a size-independent transparency is strongly related to the dispersive dielectric constant of Ag, where the refractive index is close to 1 [48].

In this paper, the Ag NPs are roughly separated into three groups in terms of their sizes (Table 1) according to their simulated LSPR behavior in air and Al₂O₃ in the visible spectrum. A small Ag NP (D ≤ 30 nm) has a single narrow resonance peak due to the interaction of between the dipole and the field, and the scattering is weaker than the absorption. The LSPR of a medium Ag NP (30 nm < D ≤ 70 nm) is still dominated by the dipole resonance, but unlike for small NPs, the scattering is greater than the absorption and the resonance due to the quadrupole is not negligible, especially when N increases to 1.6. The large Ag NP (D > 70 nm) has a broadband LSPR because the higher multipole resonances are comparable to or stronger than the one due to dipole.

Table 1. The categorization of different sized Ag NPs studied in this work.

View Table

The numeric model is different from our experiments in some aspects. In the wet chemical plating, usually the distribution and morphology of Ag NPs are not highly uniform, and clusters of NPs may form. Meanwhile, the NPs sit at the Si surface. The overall reflection measured from the Si surface is the sum of the backscattered incident light from the NPs plus the reflection from the Si surface. The goal of using such a simple model is to provide insight on the LSPR of Ag NPs as well as qualitative requirements on the Ag NPs for the preferable optical performance, since it is costly to simulate irregular randomly distributed Ag NPs on the Si surface.

During the electroless plating process, the formation of Ag NPs is complex since after the displacement of Si with Ag, different effects can govern the process. In general, the spherical Ag NPs are usually formed by nucleation and Ostwald ripening [36], while some elongated or random-shaped NPs can be formed through the coalescence among nearby NPs. Figure 2(a) shows that immediately after the electroless plating, the coalescence governs the formation process because dense Ag islands are formed with random shapes with a surface coverage of 46.6%. This Ag island morphology on the Si surface results in the reflection and ΔR% spectra shown as blue curves in Figs. 2(c) and 2(d), despite a weak reflection reduction around 368 nm, the as-plated Si substrate has a higher reflection than the bare Si wafer from 400 nm to 700 nm, rendering an overall reflection increase. The density and shape of the Ag NPs can be significantly modified after a water bath at 90 ˚C for 2 min, as shown in Fig. 2(b), which reduces the CSFF_tot to only 15.5% and changes the shape of most NPs into a sphere. For reference, this bath is named sample A. Its CSFF_i distribution is shown in Fig. 4(a) as the black curve, which indicates that most of the Ag NPs on sample A have diameter between 30 to 70 nm, or medium sized diameters. The red curves in Figs. 2(c) and 2(d) show that the modified sample surface has a lower reflection than the bare Si in the entire visible spectrum, with a strong decrease (>20%) around 371 nm and a weaker one (∼2%) in the longer wavelength region. The hot water bath method for generating spherical Ag NPs has also been adopted by other researchers and has proven effective [49,50].

 

Fig. 2. Surface SEM images of an Ag NP decorated Si substrate (a) before and (b) after a 2 min 90 ˚C water bath. Spectral reflection (c) and (d) ΔR% of bare Si (dashed black line), Si with Ag NPs pre-water bath (solid blue line), and Si with Ag NPs post-water bath (solid red line).

Download Full Size | PDF

The ΔR% curves in Fig. 2(d) have a strong correlation with the simulated LSPR of single NPs. The two ΔR% shoulders at 325 nm agree with the calculated LSPR minima around 325 nm in Fig. 1. The ΔR% peak for the post-bath sample can be explained as the sum of multiple LSPR of medium-sized NPs with peaks around 370 nm, while the weak long wavelength response may be induced by the coupling between NPs [51,52]. As for the as-plated sample, a strong backscattering is anticipated because the total coverage of Ag NPs is too high. When the CSFF_tot approaches 100%, a highly reflective Ag film is presented. As a result, to take advantage of the LSPR of NPs and obtain a significant reduction in reflection, careful tuning of the shape and inter-particle distance of the Ag NPs is required.

Figure 3 shows the surface SEM images of Ag NPs plated using different recipes followed by a 2 min 90 °C water bath. The CSFF_tot values and the CSFF_i distributions presented in Fig. 4(a) are calculated based on three 2.56 µm $\times $ 1.92 µm SEM images taken on different locations of each sample. We can understand the LSPR spectra of the sample by studying the contribution from an individual NP. The error bars on the curves in Fig. 4(a) indicate differences in CSFF_i values among images taken on the same sample. The ΔR% of the samples in Fig. 3 are shown in Fig. 4(b). Prior to the water bath, sample A is plated in a solution containing 4.8 M HF and 4 mM AgNO₃ for 10 seconds, and for comparison, all the other samples use different recipes with only one parameter different from the one used for Sample A. Sample B (Fig. 3(b)) is plated using only half (2.4 M) of the standard sample A HF molarity (4.8 M). The AgNO₃ molarity for samples C and D (Figs. 3(c) and 3(d)) is 25% (1 mM) and 50% (2 mM) of the standard sample AgNO₃ molarity (4 mM), respectively. Samples E and F (Figs. 3(e) and 3(f)) are plated for 5 s and 20 s, respectively, using the same molarity for HF and AgNO₃ as that in sample A’s recipe.

 

Fig. 3. Surface SEM images of Ag NP decorated Si substrates after a 2 min 90 °C water bath. The substrates are dipped in a plating solution of (a) 4.8 M HF and 4 mM AgNO₃ for 10 s, (b) 2.4 M HF and 4 mM AgNO₃ for 10 s, (c) 4.8 M HF and 1 mM AgNO₃ for 10 s, (d) 4.8 M HF and 2 mM AgNO₃ for 10 s, (e) 4.8 M HF and 4 mM AgNO₃ for 5 s, and (f) 4.8 M HF and 4 mM AgNO₃ for 20 s. The samples are labelled in the same order.

Download Full Size | PDF

 

Fig. 4. The CSFF_tot and CSFF_i vs diameter data (a) and the ΔR% spectra (b) for samples plated using different recipes: A: 4.8 M HF and 4 mM AgNO₃ for 10 s; B: 2.4 M HF and 4 mM AgNO₃ for 10 s; C: 4.8 M HF and 1 mM AgNO₃ for 10 s; D: 4.8 M HF and 2 mM AgNO₃ for 10 s; E: 4.8 M HF and 4 mM AgNO₃ for 5 s; F: 4.8 M HF and 4 mM AgNO₃ for 20 s. The vertical black dashed line in (b) indicates the wavelength of 371 nm.

Download Full Size | PDF

By comparing CSFF’s for samples A, C, D, E, and F in Fig. 4(a), different stages of the formation of Ag NPs can be demonstrated. When the solution has 4 mM AgNO₃ and 4.8 M HF, the 10 s plated sample A and 20 s plated sample F have a Gaussian shaped CSFF_i distribution centered at medium sizes from 30 to 70 nm, and the longer the plating time, the larger the CSFF_tot. When the plating solution has a much lower AgNO₃ molarity of only 1 mM, such as for sample C, the generated NPs cover only 4.3% of the entire surface and the largest diameter is only 110 nm. Its CSFF_i is rather low with a small peak ranging from 30 to 90 nm. When the AgNO₃ concentration or plating time is changed, samples D and E can be assumed to have more plated Ag than sample C but less than sample A. Although they both contain high ratios of small NPs, they have a totally different CSFF_i distribution in larger diameter regions. On sample E, small Ag NPs start to aggregate into medium ones, mostly with diameters from 30 to 50 nm, and not many large-sized NPs are formed. However, on sample D, another process prevails in which most of the medium NPs continue to aggregate into larger ones while the rest of the small NPs are not dense enough to group together, which results in two distribution peaks in the small (10 - 30 nm) and large (90 - 110 nm) size regions.

Although the six samples do not have the same Ag NP distributions, their optical ΔR% spectra are similar, as shown in Fig. 4(b). Sample A has the highest reflection reduction of 24.8% at 371 nm. The ΔR% peaks of samples A, C, E, and F are all located at 371 nm, while the ones of samples B and D are at 375 nm and 384 nm, respectively. The measured ΔR% peaks at 371 nm are very close to the calculated extinction (and scattering and absorption) peak at 367 nm of the 60 nm Ag NP in air (Fig. 1(b)). It is an important evidence of the correspondence between the calculated LSPR values and the ΔR% since the Ag NPs on those samples are dominated by the medium Ag NPs of diameters from 30 to 70 nm. Furthermore, the peak ΔR% ratio of sample A to C is 4.3, while their CSFF_i ratio for the medium sizes, CSFF_med,A/CSFF_med,C, is 4.0, indicating that the medium Ag NPs may contribute to the majority of the forward-scattering. The effect of small NPs is not significant since sample E has the highest number of small NPs among all the six, but its peak ΔR% is still at 371 nm and only half in magnitude of sample A’s, indicating that small NPs do not shift the ΔR% location nor contribute to its magnitude much. It is also an implication of a small N, because from the simulation in Fig. 1(b), the Q_ext for a small Ag NP is much weaker in air than in Al₂O₃. If the surface is covered by too much Ag, the coupling between NPs will increase backscattering, hence the overall reflection. Therefore, sample F with a high CSFF_tot of more than 38% generates a lower ΔR% peak compared to sample A. In general, there is always a trade-off when the backscattering increases while the desired forward-scattering is increasing.

To further justify that the fabricated NPs can be simulated using the proposed model, the side view cross section SEM of sample A is shown in Fig. 5. Most of the NPs on sample A are spherical or slightly ellipsoidal, with only a small surface area touching the Si substrate. Therefore, it is reasonable to assume that the average refractive index of the medium surrounding each Ag NP is nearly 1. Meanwhile, the Si surface is smooth on the scale of 10 nm, so the effect of surface roughness can be negligible.

 

Fig. 5. The side view cross section SEM image of the surface of sample A.

Download Full Size | PDF

We conducted an aging experiment to study the time-dependent atmospheric degradation of Ag NPs. The ΔR% of sample A after sitting in ambient laboratory environment for 0, 1, and 2 weeks are plotted in Fig. 6(a). After 2 weeks, the ΔR% peak of sample A is still at 371 nm, but the magnitude has changed. After the first week, the peak ΔR% value of sample A reduces from 24.8% to 23.7%, while a small increase of around 1.5% is also observed at wavelength after 510 nm. However, after 2 weeks, the degradation is more significant since the ΔR% peak is only 20.2%. The relative change of 18.5% in ΔR% in 2 weeks indicates that it is important for the Ag NPs to be protected from atmospheric degradation for use in optoelectronic devices. A 27 nm thick ALD Al₂O₃ is chosen as the protective dielectric material for Ag NPs and is deposited on two samples that are plated with Ag NPs using the same recipes for samples A and B. The degradation study for the former one is plotted in Fig. 6(b). It is shown that the optical properties of an Al₂O₃ coated sample do not change markedly even after 90 days, indicating an outstanding protection on Ag NPs.

 

Fig. 6. (a) The ΔR% value of sample A right after water bath (solid blue) and after sitting in the laboratory environment for 1 (dashed red) and 2 (dotted green) weeks. (b) The reflection data of a sample plated using the same recipe as sample A and coated with Al₂O₃ right after deposition (solid blue) and after sitting in the laboratory ambient for 90 days (dashed red).

Download Full Size | PDF

The AFM image of the Al₂O₃ coated medium sized Ag NP dominated sample is shown in Fig. 7(a), and the R_ave and reflection data of a bare Si wafer and the two Al₂O₃ protected samples before and after the Al₂O₃ deposition are shown in Fig. 7(b). Without Al₂O₃, the two plated samples are proven to have similar reflection spectra shape to the ones of samples A and B despite a small difference in magnitude, which confirms that the Ag NPs on the Si surface of the two samples are mainly medium sized and large sized, respectively, but the numbers vary slightly due to the randomness in the wet chemical deposition process. According to the AFM image, the NPs are coated with a smooth and dense Al₂O₃ film, which preserves the spherical shape of the NPs. But the diameter of the NPs increases by 54 nm, causing the coated Ag NPs to look clustered instead of spaced out individual NPs.

 

Fig. 7. (a) AFM image of a medium diameter Ag NPs dominated surface coated with ALD Al₂O₃ film. (b) Reflection and R_ave values of a bare Si surface without Ag NPs (blue dashed), a medium diameter Ag NPs dominated Si surface (red), and a large diameter Ag NPs dominated Si surface (green) before (dashed) and after (solid) the ALD Al₂O₃ deposition.

Download Full Size | PDF

Since Al₂O₃ does not contribute electrons to LSPR, it only introduces an increase of medium refractive index which would cause redshift, peak widening, and a slight magnitude increase according to the simulation in Fig. 1. As a result, in Fig. 7(b), the planar Si surface only has a slight R_ave decrease of about 6% after deposition, but the optical performance of the two samples with Ag NPs is effectively improved. The large Ag NP dominated sample has a R_ave of 37.6% without Al₂O₃ and has a nearly flat reflection curve of about 18% in the visible spectrum after Al₂O₃ deposition. The one dominated by medium Ag NPs has a R_ave of 12.2% compared to the previous value of 35.2% and reaches a reflection minimum of only 7.6% at 662 nm. Different from the localized behavior of the samples without Al₂O₃, a broadband reflection reduction that is much stronger than any of the affects from only Ag NPs or Al₂O₃ is achieved, which coincides with the multi-peak LSPR of the spherical Ag NP in the simulation study.

4. Conclusion

We report a time- and cost- effective electroless plating method to fabricate randomly distributed Ag NPs on the surface of Si which can reduce the reflection of Si in the visible spectrum. A reflection decrease of 24.8% at 371 nm relative to the bare Si wafer is achieved with Ag NP decoration on the Si surface. We also demonstrate a strong broadband LSPR response due to Al₂O₃ protected Ag NPs on Si, that reduces R_ave from 35.2% to 12.2%, and the reflection at 662 nm from 32.8% to 7.6% by protecting the NPs with a thin layer Al₂O₃, pointing out a potential strategy for antireflection coatings. Compared to the calculation from Mie scattering theory, we are able to determine the effects of different sized Ag NPs on the measured reflection data. The LSPR properties markedly depend on the size and distribution of Ag NPs, leading to a way to tune the surface optical properties of Si. An ALD Al₂O₃ dielectric protection reduces atmospheric degradation of the Ag NPs along with broadening the plasmonic response into reduced surface reflection across the visible spectrum. The proposed method can be utilized in fabricating surface plasmonic materials with selective light absorption and in realizing light confinement on Si surfaces.