1. Introduction

Zinc oxide (ZnO) has a band gap of 3.37 eV and exciton binding energy of 60 meV, the latter providing excellent thermal stability compared, for example, with GaN (binding energy 25 meV). As with all wide-bandgap n-type semiconductors, it is difficult to achieve stable p-type semiconductivity, but initial reports of p-type ZnO have generated a flurry of interest in ZnO optoelectronic properties and applications [1–3]. ZnO exhibits two peaks in its photolu-minescence (PL) spectrum: a near-UV band edge peak at 3.3 eV attributed to exciton recombination and a broad donor-acceptor-pair (DAP) recombination peak centered near 2.3 eV, conventionally understood to be a superposition of emission lines due to the recombination of electrons near the conduction-band edge with deep holes on singly ionized oxygen interstitials, Zn anti-site vacancies and oxygen vacancies [2,4].

The coupling of plasmons in Ag, Au, and Al with ZnO excitons and DAPs for spectral control and amplification has attracted particular attention [5–12]. Plasmon-exciton interactions in ZnO-metal composites amplify the band edge photoluminescence (PL) by a factor 10 in Ag-ZnO and Al-ZnO bilayer structures [5,11,12] and as much as 20 in Au-ZnO nanostructured composites [6]. Band edge PL enhancement in bilayer structures has been attributed both to coupling of band-edge excitons to surface-plasmon polaritons (SPPs), reflected in an increased Purcell factor [5], and to localized surface plasmons (LSPs) that absorb and scatter the ZnO band-edge emission [7,12]. The band-edge PL enhancement in Au-ZnO nanostructures has been attributed to excitation of LSPs by ZnO impurity emission, followed by transfer of hot electrons to the ZnO conduction band [6,8]. Moreover, both the band-edge enhancement and the Purcell factor have a similar dependence on temperature, lending further credence to the idea that enhanced band-edge emission results from coupling to SPPs [11].

In this paper, we demonstrate that the enhancement of the ZnO PL as a function of distance from rough metal films is a result of three distinct mechanisms. First, we show that the band-edge enhancement in nanoparticulate metal-ZnO film bilayers disappears upon introduction of an insulating MgO spacer film. This implies that the enhancement is due to charge transfer, as proposed in earlier studies of Au and Ag nanoparticles in contact with ZnO nano-particles and nanorods [6,8]. Second, while the band-edge enhancement for these systems disappears upon introduction of MgO spacer layers, the DAP PL enhancement remains, indicating a local-field enhancement of the DAP emission by both Ag and Au LSPs, as argued previously for Ag [7]. By varying the MgO spacer thickness, we demonstrate that the enhancement of the DAP PL for both Ag-ZnO and Au-ZnO bilayers decays as 1/z, suggesting that this enhancement results from LSP-DAP dipole-dipole scattering. Third, we exhibit local-field enhancement of the band-edge emission in bilayers of ZnO with rough metal films of Ag and Au. We argue that this is a result of Purcell enhancement of the ZnO band edge emission by Ag and Au SPPs. Although other authors have demonstrated this enhancement for Ag-ZnO bilayers, [5,11] this is the first demonstration of ZnO band-edge PL enhancement by Au SPPs. Varying the insulating MgO spacer thickness provides the linchpin for our argument, as we thereby can demonstrate that the band-edge enhancement decays exponentially with decay constants consistent with the evanescent field fringing field depth for Au and Ag SPPs. The measurement of this distance dependence goes beyond previous studies and thus provides the most substantial support to date for the argument that the Purcell enhancement is indeed initiated by both Ag and Au SPPs.

2. Experimental procedures

Figure 1 shows the samples used in these experiments. Ag and Au films were deposited on 2 cm × 1 cm Si(100) substrates by electron beam evaporation (Thermionics, 100-0030) at a pressure of 2.0 μT. The deposition rate was maintained at 0.1 Å/s, as monitored by an Inficon 750-211-G1 quartz-crystal microbalance. Before deposition, half of the substrate was masked so that the metal occupied 1 cm × 1 cm squares with thicknesses of 10 nm–40 nm in 5 nm increments. The morphology of the metal films was observed by scanning electron microscopy (SEM, Raith e-line), and the thickness was confirmed by Rutherford backscattering (RBS).

 

Fig. 1. SEM images for (a) 10nm Ag film, (b) 10nm Au film, (c) 30nm Ag film, (d) 30nm Au film, and (e) a schematic of the deposited films.

Download Full Size | PDF

After metal deposition, the masks were removed, and the full substrate was coated by electron beam evaporation with 70 nm ZnO. An equal number of samples were prepared with 10 nm MgO spacers deposited by electron beam evaporation after masks were removed but before depositing ZnO. In order to analyze the dependence of the spacing between the plasmons and excitons in these interactions, MgO films were also deposited 10 nm–60 nm thick in 10nm increments on 30nm Au and 30nm Ag films and then coated with 70 nm ZnO as before.

Photoluminescence (PL) measurements were made using a vertically polarized HeCd laser, (wavelength 325 nm, photon energy 3.82 eV) which provided an excitation with a power density of 600 W/cm² at a 45° incidence angle. The ZnO emission was then measured using a 0.5s exposure on an Oriel monochromator MS257TM with a thermoelectrically cooled CCD detector. The spectral range of the monochromator is 1.86–3.78 eV, so both the band-edge emission and visible emission could be characterized. Other authors using a frequency tripled Nd-YAG laser pump showed that the DAP emission begins to saturate for powers greater than 1500 W/cm², indicating — in agreement with our own observations — that saturation of the DAP emission occurs at powers higher than those used in our experiment [13].

3. Results

SEM images of the metal films shown in Fig. 1 demonstrate that, for nominal 10 nm thickness, the films are composed of nanosize islands that merge to form nearly continuous films at deposition thicknesses of 30 nm. The quasicontinuous films will support SPPs, being still sufficiently rough to overcome the photon-plasmon momentum mismatch [14]. LSPs, on the other hand, should be supported on all of the films—either on individual nanoparticles in the thin films, or on holes or asperities in the thicker films.

 

Fig. 2. ZnO photoluminescence for 70 nm ZnO thin films in contact (a) with 10 nm metal particulate films, and separated (b) by a 10 nm MgO spacer layer. (c) Ratio of Ag/ZnO visible emission to ZnO visible emission for metal thicknesses of 10 nm to 40 nm.

Download Full Size | PDF

The PL spectra for 70nm ZnO films with and without 10 nm metal-island underlayers are shown in Figs. 2(a) and 2(b). The peak band-edge emission is doubled in the presence of the Au and Ag underlayers, but the enhancement disappears altogether when the 10nm insulating MgO spacer is introduced. In fact, the metal-ZnO band-edge emission is quenched to half of the ZnO band-edge emission upon introduction of the spacer. In contrast, the metal-ZnO visible emission is enhanced by a factor of roughly 1.5 both with and without the MgO spacer for both Ag and Au. Fig. 2(c) demonstrates that the visible emission enhancement in the Ag-ZnO films depends on metal thickness, and also demonstrates that the general form of the dependence is the same with an MgO spacer as without. The enhancement is a maximum for 30 nm Ag films with and without a 10nm MgO film.

To shed further light on the PL enhancement mechanisms for the UV and visible emission, we compared the ratio of ZnO PL emission with a 30nm rough metal undercoating to the ZnO PL emission without any metal undercoating for MgO spacer thicknesses of 10–60nm in Fig. 3 for both Ag and Au. As expected, the enhancement is greater for Ag-ZnO bilayers than for Au-ZnO bilayers over the full spectral range, but the same general behavior is observed for both metals. Interestingly, while the band edge enhancement factor for both Ag and Au decays to well below 1 as the MgO thickness increases, the enhancement for Ag-ZnO emission at 2.85 eV increases to a factor 2, and the enhancement for Au-ZnO at 2.5 eV increases to a factor 1.5.

Further insight can be gained by examining the band edge enhancement and visible emission enhancement as a function of MgO thickness╌illustrated in Fig. 4. Fitting constants for these curves are shown in table 1. The band-edge luminescence shows a maximum enhancement near 20 nm thickness, while the visible-emission PL enhancement is monotonically decreasing. These features provide additional insight into the exciton-plasmon and defect-plasmon interactions as described in the following section.

 

Fig. 3. Average ratio of ZnO emission with metal to ZnO emission without metal for gold (a,b) and silver (c,d) and silver (c,d) as a function of MgO spacer thickness. Error bars are the standard deviation from three separate samples.

Download Full Size | PDF

 

Fig. 4. Band edge (a) and visible emission (b) enhancement as a function of insulating spacer thickness.

Download Full Size | PDF

Table 1. Fit parameters for data of Fig. 4(a) and 4(b)

View Table

4. Analysis

The experimental data present two different sets of observations that bear on the mechanism of exciton-plasmon and DAP-plasmon interactions. Both contact-related effects and local-field effects must be considered to understand the whole picture.

First, there is the band-edge PL enhancement that arises from direct contact with a thin metallic film—demonstrated in Figs. 2(a) and 2(b) for 10 nm metal films. The disappearance of PL enhancement upon introduction of an insulating MgO spacer layer indicates that direct charge transfer is responsible; these data thus corroborate the model proposed by Lin [6] and Lee [8] to account for PL enhancement in a different experimental configuration.

The origin of the hot electrons can be understood in terms of excitation of LSPs in the metal nanoparticles. The particles in Figs. 1(a) and 1(b) are roughly 50–80 nm in lateral dimension, so that for both 10 nm Ag and 10 nm Au films, LSP resonances should be supported between 1.8 eV and 2.5 eV, depending on the particular geometry of the individual nanoparticles (NPs) and on the dielectric environment [15]. Mie scattering calculations for spherical particles with a normal size distribution approximately the same as Fig. 1 demonstrate that the Au LSP resonance energy for these systems is roughly 1.8 eV and the Ag LSP resonance energy is roughly 2.05 eV. The fact that the band-edge emission is actually quenched when the MgO is introduced suggests that the plasmon energy may be preferentially coupling into waveguide modes in the Si substrate when charge transfer to the ZnO is not possible [12].

The absence in Fig. 2 of the visible emission damping observed by Lin and Lee cannot be solely attributed to different LSP resonance energies due to the different geometry and dielectric environment. Instead, because similar visible emission enhancement is observed with and without MgO spacers in Fig. 2(c), the visible enhancement must be understood to be a result of DAP-LSP dipole-dipole scattering.

If the local field enhancement of the band-edge emission is due to Purcell enhancement, then the exciton recombination rate can be described by the Fermi golden rule, and the Purcell enhancement will increase linearly with the recombination rate [11]:

Here 〈f|Ĥ_int|g〉 is the electron-hole dipole matrix element with a Hamiltonian describing the perturbing plasmon field, and ρ(ħω) is the plasmon density of states (DOS).

For SPPs, the DOS is proportional to k(dk/dω), and so is maximized at energies near the horizontal asymptotes of the dispersion curve, which is plotted in Fig. 5 for Ag and Au at ZnO and MgO interfaces [16–19]. Figure 5 demonstrates that the SPP resonance energy — and thus the maximum plasmon density of states — occurs near 2.9 eV at Ag-ZnO interfaces, 3.2 eV at Ag-MgO interfaces, and 2.4 eV-2.5 eV at Au-ZnO and Au-MgO interfaces. Because the resonance energy for Ag is much closer to the ZnO band edge than that of Au, and because the imaginary component of the silver permittivity is much smaller than the real component, silver SPPs should induce significantly greater band-edge enhancement than gold SPPs.

 

Fig. 5. Dispersion relationship for SPPs in metal dielectric bilayers.

Download Full Size | PDF

The SPP exists as an evanescent wave in the dielectric with an exponential decay constant given by $\hat{z}=(\lambda /2\pi )\mathrm{Im}\sqrt{\left({\epsilon }_1+{\epsilon }_2\right)/{\epsilon }_2^2}$. Evaluating this fringing field depth at the Ag-ZnO and Au-ZnO SPP resonance energies found in Fig. 5, we obtain 23.4nm for Au-ZnO, 39.0nm for Au-MgO, 21.3nm for Ag-ZnO, and 31.7nm for Ag-MgO.

These values provide a significant context for Figs. 3 and 4. In Figs. 3(b) and 3(d), for thicknesses of 50nm and 60nm, the ZnO band edge emission is quenched by roughly 40%, while the emission at the metal-ZnO SPP resonance energy is amplified by a factor of 1.45 for Au and by a factor of 1.8 for Ag. We therefore argue that for MgO thicknesses smaller than the metal-MgO fringing field depth, there exist two mechanisms for SPP energy loss. The SPP evanescent field can cause Purcell enhancement of the ZnO band-edge PL, or the SPPs can scatter directly to photons. Because the rate at which the SPPs scatter to photons is a function of the asperity of the metal surfaces, we see a larger standard deviation in the enhancement factors near the ZnO bandgap and near the SPP resonance energies for small MgO spacers. As the MgO thickness increases to values greater than the metal-MgO fringing field depth, SPPs are exponentially less able to couple with excitons, resulting in enhanced emission at the SPP resonance energy. This is significant evidence for the role of SPP initiated Purcell enhancement of the ZnO bandgap PL for both Au and Ag SPPs.

Analysis of Fig. 4 provides further evidence of SPP-exciton interactions and LSP-DAP interactions. It is clear from equation 1 that the Purcell enhancement of the band edge emission should decay as exp(-2z/ẑ), but Fig. 4(a) demonstrates that there is a competing mechanism which slows the decay of the enhancement factor, which we postulate to be a result of plasmon scattering. Numerical calculations modeling the decay rate of a dipole near an infinite metal surface have demonstrated a function similar to that of Fig. 4(a), so these results are not unexpected [20]. The best fit for the band edge enhancement in Fig. 4(a) is of the form(1 + z ²/A ²)exp(-2z/B). This fitting function results in experimental decay constants consistent with the SPP fringing field depths calculated above, providing significant confirmation of the existence of the Purcell enhancement of ZnO band-edge emission by both Ag and Au SPPs.

If the visible emission enhancement were a result of Purcell enhancement of the DAP PL, the enhancement factor would be expected to decay as the cube of the LSP dipole field, or as 1/z⁶. If the visible enhancement were instead a result of LSP-DAP dipole-dipole scattering, then the enhancement factor would be expected to decay as 1/z³ [21]. However, the best fit to the observed visible PL enhancement decay is of the form A + B/z. It therefore seems that the same—currently unidentified—mechanism responsible for the slowed decay of the band-edge enhancement is responsible for slowing the decay of the visible enhancement, and the DAP PL enhancement can be understood to be the result of LSP-DAP dipole-dipole scattering.

5. Conclusion

Introducing insulating MgO spacer films of varying thickness to ZnO/Au and ZnO/Ag bilayers has enabled us to distinguish PL enhancement due to hot-electron transfer from PL enhancement due to local-field interactions.

The enhancement of ZnO PL in ZnO/metal bilayers has frequently been attributed to excitonic interactions with evanescent fields [5, 11, 12]. By depositing variable thicknesses of MgO between metal films and ZnO films, we have demonstrated the scattering of Au and Ag SPPs to photons when the MgO thickness is much greater than the calculated SPP fringing field thickness. After accounting for this scattering, we have shown that the exponential decay of the band-edge enhancement with increasing MgO thickness has a decay constant that is consistent with the calculated value from the SPP dispersion relations. This confirms the role of local-field interactions between the ZnO exciton and the SPPs propagating on both Ag and Au films. Accounting for plasmon scattering in the same manner, the 1/z decay of visible enhancement provides independent, additional support for LSP-DAP dipole-dipole scattering.

Previous authors studying band-edge PL enhancement in ZnO/Au and ZnO/Ag nanocomposites have attributed the enhancement to hot electron transfer to the conduction band edge [6, 8]. We have confirmed this mechanism by demonstrating that band-edge enhancement disappears when a 10nm MgO spacer film separates the metal and ZnO films — providing critical evidence for hot electron transfer across the metal-ZnO interface as the PL enhancement mechanism.

While our measurements help us to characterize the nature of the coupling between PL and localized or propagating surface plasmons, they do not take into account non-radiative processes that affect the overall quantum efficiency of the process. Our measurements also do not provide any information on the coupling between the band-edge exciton and the DAPs, which is believed to occur by tunneling [22]. By measuring PL enhancement as a function of MgO spacer thickness, we are normalizing out the non-radiative effects except possibly for those kinetic processes that might be determined by absolute intensities or excitation rates.

The growing interest in plasmon-exciton interactions in ZnO/metal composites has resulted in reports of photoluminescent enhancement attributed to both local surface plasmons and propagating surface plasmon polaritons. The existence of ZnO band-edge enhancement due to both charge-transfer mechanisms and local field mechanisms demonstrates a fundamental problem with studying exciton-plasmon interactions in composites utilizing rough metal films, namely, that because both LSPs and SPPs are supported in such systems, the enhancement cannot be well characterized without the introduction of insulating spacers. Similarly, while Fig. 2(c) shows unequivocal evidence of both local field interactions and of DAP emission enhancement varying with the LSP resonance peak, we must consider other interactions, most notably the possibility of ZnO emission being reduced by non-radiative coupling to the Si substrate. Future experiments would benefit from the introduction of lithographically patterned metal films designed to support either LSPs or SPPs but not both, and from the introduction of time-resolved spectroscopy that will allow us to better distinguish between the various interactions.