1. Introduction

Light emission from silicon has attracted a lot of attentions in the scientific community because it has promising potential to be integrated with Si-based electronic device and to form Si-based optoelectronics. However, bulk silicon is indirect bandgap semiconductor, and the efficiency of light emission is weak. To solve this intrinsic problem, porous silicon has been demonstrated to emit visible light in the 1990s [1–7], but light emission was not stable due to the problems of aging, oxidation, and surface passivation in porous silicon [4, 8–11]. Later, silicon nanoparticles (Si NPs) embedded in SiO₂ or sub-stoichiometric oxide have emerged as alternates to porous silicon owing to their stable light emission, good repeatability, and well-controlled particle size [12]. Both red and near-infrared (NIR) light emissions have been demonstrated in Si/SiO₂ superlattice [13–17], Si NPs in SiO_x [18–21], and Si NPs in ZnO [22]. Blue and green light emissions have also been achieved from Si NPs [19, 23, 24]. However, to our knowledge, there is no report to demonstrate such a broad tunability of photoluminescence (PL) from ultraviolet (UV) to NIR regime.

In this work, by controlling the condition of plasma-enhanced chemical vapor deposition (PECVD), different ratios of O to Si were achieved, resulting in tunable wavelength of photoluminescence after thermal annealing. We characterized the samples with emission from UV to NIR regime, but particularly focused on the UV-blue emission since most previous studies had already investigated the visible-NIR regime [18–21]. The experimental results indicated that the origin of light in UV-blue emitting sample is from defects at the interface of Si and SiO_x matrix although nanoparticles with size smaller than 2 nm were found. This observation is not trivial because the size-dependence of silicon nanoparticles relating to PL can be easily misattributed the origin of light emission to quantum confinement effects [3]. It should also be noted that the UV-blue emitting samples with short lifetime in our work is different from the widely studied porous silicon by wet-etching techniques [3, 4, 8–10] in 1990s. Because stable and tunable light can be obtained from annealed SiO_x in contrast to unstable porous-silicon, it can be utilized for application of optoelectronic devices [25, 26]. Our investigation has shown the great potential to extend annealed SiO_x materials for UV-related optoelectronic devices.

2. Experimental methods

The Si-rich SiO_x film was grown on p-type (100)-oriented Si substrate by using the PECVD system. By controlling the N₂O/SiH₄ fluence ratio, different O/Si compositions were achieved. The samples were annealed in a quartz furnace with flowing N₂ gas at 1100°C. Growth conditions and the physical properties of the samples are listed in Table 1. O/Si composition ratios were characterized by XPS (X-ray photoelectron spectroscopy).

Table 1. The growth condition of the studied samples and their optical properties.

View Table

Cross-sectional transmission electron microscope (TEM) specimens were prepared with tripod polishing and ion milling using the Gatan PIPS system operated at 3 kV. The TEM experiments were carried out using a field-emission-gun TEM (FEI, Tecnai F20) operated at 200 kV equipped with Gatan image filter (GIF, Tridiem 865). The real-space spectral-imaging (SI) in energy-filtered TEM with a tunable energy-selection slit, was conducted on the same microscope [27]. For time-integrated PL and time-resolved PL (TRPL) in the UV-blue region, the excitation light source was from the fourth harmonic generation (266 nm or 4.67 eV) of a 40 MHz, 1064 nm femtosecond fiber laser (Fianium, Ltd.). For blue-visible PL lifetime measurement on the time scale of 10 ns to 1 μs, a 0.5 MHz, 405 nm picosecond diode laser (PicoQuant GmbH) was used as the excitation light source. The TRPL were measured using time-correlated single photon technique. The PL excitation (PLE) spectroscopy was performed with a commercial spectrofluorometer (FluoLog, Horiba, Ltd.).

3. Results and discussion

Real-space TEM SI was first performed to confirm the presence of Si NPs embedded in SiO_x with different O/Si composition ratios. Figure 1(a) shows a typical SI with the energy-selection slit positioned at the energy loss region of 16 ~18 eV, which monitors the plasmon peak of bulk Si at ~16.7 eV [28, 29]. Figure 1(b) shows the corresponding electron-energy loss spectra extracted respectively from position 1 (Si NPs) and position 2 (SiO_x) marked in Fig. 1(a). The result reveals that the bright spots shown in Fig. 1(a) have the same spectral feature as Si bulk plasmon, indicating those areas contain Si NPs. In contrast, the gray areas correspond to amorphous SiO_x. Furthermore, the TEM image shows a wide size distribution, 1.0~2.5 nm in diameter, of Si NPs in Sample A. With increasing annealing time (Sample B), the particle size distribution of Si NPs became narrower (~1.8 nm ± 0.2 nm in diameter). Compared with Samples A and B, the Si NP size distribution in Sample C is larger (2.2~3.0 nm in diameter) due to the aggregation of more excess Si. The histograms of Si NP size distributions are shown in Fig. 1(c)-1(e) for Samples A-C. Overall, the mean diameter of Si NPs decreases when the O/Si composition ratio increases [19].

 

Fig. 1 (a) EFTEM-SI image of Sample C with the energy-selection slit tuned to the energy loss region of 16 ~18 eV. Circle 1 indicates a region containing Si NP and circle 2 indicates a region of the surrounding SiO_x. Their corresponding EELS spectra are shown in (b). Histogram of Si NP size of (c) Sample A (d) Sample B and (e) Sample C. The particle number is obtained from a 20 nm x 100 nm TEM image.

Download Full Size | PDF

Figure 2 shows the PL spectra of all the samples. The two peaks at ~340 nm and ~410 nm were observed both in Samples A and B. For Samples C and D, the PL shows a broad feature with peaks at ~600 nm and ~770 nm, respectively. When the excess Si decreases (or x increases in SiO_x), the peak of PL spectra shifts from red (Sample D) and yellow (Sample C), to UV-blue (Samples A and B). The light emission of the SiO_x samples with various annealing time was also investigated. It was found that the PL spectra of Samples C and D showed red-shift when the sample annealing time increased. These results agree with previous reports that the main origin for visible light from annealed SiO_x is from quantum confinement (QC) effects [19]. The blue-shift phenomenon in PL spectra was attributed to the shrinking of Si NP size which leads to a larger bandgap. Longer time of annealing will also result in larger Si NP due to aggregation of excess Si. For UV-blue emitting samples in our work, the Si nanoparticles were found even smaller from our TEM images, which were generally thought as evidence of QC effects. However, in contrast to visible-light-emitting samples, the PL spectra showed no significant peak shift but decreasing intensity when the sample annealing time was increased for Samples A and B.

 

Fig. 2 The PL spectra of all studied samples, where curves A-D denote Samples A-D, respectively.

Download Full Size | PDF

The high-resolution TEM (HRTEM) images of individual Si NPs in Samples A and C are presented in Figs. 3(a) and 3(b), respectively. Figures 3(c) and 3(d) show their corresponding fast Fourier transform (FFT) patterns, which are equivalent to the experimental electron diffraction patterns of the selected local regions indicated by the dash circles in Figs. 3(a) and 3(b). For Sample C the lattice fringe can be resolved in Fig. 3(b), and the corresponding FFT pattern in Fig. 3(d) reveals the spot-like pattern, indicating the Si NPs are crystalline. In contrast, Sample A HRTEM investigation [Figs. 3(a) and 3(c)] reveal a featureless image and a diffuse ring pattern, indicating the Si NPs are amorphous. This evidence clearly exhibits the tendency of poor Si NP crystallinity when the O/Si composition ratio is intentionally tuned to higher value for embedding SiO_x with smaller Si NPs.

 

Fig. 3 HRTEM images of (a) Sample A and (b) Sample C. Their corresponding FFT patterns are shown in (c) and (d). The location of Si NPs is indicated by the dash circles in (a) and (b).

Download Full Size | PDF

We further investigated samples A and C with PLE spectroscopy. Strong resonant features located at ~275 nm are resolved when one monitored the light emission at 340 nm [Fig. 4(a)] and 410 nm [Fig. 4(b)]. This resonant feature is related to defect luminescence and was also observed for Sample B. It is worthy to note that UV-blue emission was observed in porous silicon [11] and amorphous Si:H:O films [30], while similar emission bands were also found in SiO₂ [11, 30]. Specifically, it was reported recently that –SiO₃ group has a resonant feature in the range from 260 nm to 280 nm, with the emission at ~370 nm in porous silicon [31]. Our PLE results strongly suggest that the PL of UV-blue emitting samples (A and B) should be associated with molecule-like luminescent defects in Si-SiO_x complexes.

 

Fig. 4 The PLE spectra of Sample A monitored at the emission wavelength of (a) 340nm and (b) 410 nm. (c) The PLE spectrum of Sample C monitored at the emission wavelength of 620 nm. All the spectra were normalized to the maximum at ~275 nm for comparison.

Download Full Size | PDF

Figure 4(c) shows the PLE spectrum of Sample C by monitoring the main emission wavelength at 625nm. The spectrum is similar to the absorption spectrum of bulk silicon, but a weak resonant feature at 275 nm can be identified. This weak resonant feature implies the existence of fewer UV-blue luminescence defects in Sample C, but this should not be the dominant mechanism of light emission. Previous reports [19] showed that shorter wavelength light emission corresponded to smaller Si NP size in visible-NIR regime following the prediction of QC effects [3]. In addition, we also observed red-shift effects in light emission when the annealing time of Sample C is longer. This evidence also supports the QC effects because longer annealing time results in aggregation of Si and larger Si NPs. In contrast, we did not see shift of spectra for Sample A and B with different annealing time. Apparently the dominant mechanism of visible-NIR emitting samples is different from that of UV-blue emitting samples. It is thus more likely that the visible-NIR light emission of Sample C is dominantly originating from the QC effect of the crystalline Si NPs. But we do not rule out another possibility such as other types of luminescent defects in Si-SiO_x complexes [15, 16, 20].

The lifetimes of UV emission from Sample A were measured as shown in Fig. 5. Similar result was also obtained for Sample B (not shown here). In order to quantitatively analyze the lifetimes, we chose multiple-exponential function ${\text{A}}_0+{\displaystyle {\sum }_{i=1}^3{A}_i\exp (-t/{\tau }_i)}$to fit each TRPL curve in a commercial TRPL fitting software (Fluofit, PicoQuant). In this fitting function, τ_i and A_i denote the lifetime and amplitude of each exponential decay component, and A₀ is the background level. The intensity-weighted averaged PL lifetime was then determined by ${\displaystyle {\sum }_{i=1}^3{A}_i{\tau }_i^2/{\displaystyle {\sum }_{i=1}^3{A}_i{\tau }_i}}$. For emission wavelengths in the range from 275 nm to 345 nm, we noticed that some of the TRPL curves showed a higher apparent background, resulting from the long-lived PL contribution of earlier laser pulse cycles. To determine the PL lifetime accurately, we had to carefully correct this uncompleted PL decay effect [32] by using the background level of the 275 nm TRPL curve as a fixed parameter A₀ in the above curve-fitting process. Figure 5(b) shows the obtained averaged PL lifetime of Sample A and B as a function of the emission wavelength, including other published results [4, 9, 12, 33] for comparison. The dashed line for eye-guiding which was obtained by power-law fitting [34]: $\text{2}\times {\text{10}}^{\text{5}}{\text{(h}\upsilon \text{-0}\text{.8)}}^{\text{5}}\mu {\text{s}}^{\text{-1}}$, where $\text{h}\upsilon$is the photon energy in eV. Note that most previous studies focused on the light emission in the NIR-visible regime (1.4 ~2.1 eV), with reported PL lifetime on the order of microseconds [4, 9, 12, 33]. Here, we have systematically measured the PL lifetime in the UV-blue regime for Sample A and B.

 

Fig. 5 (a) TRPL traces of Sample A. The emission wavelengths measured in these traces are 275, 285, 295, 305, 315, 325, 335, 345 nm, respectively, from bottom to top. (b) The averaged PL lifetimes as a function of the emission photon energy. Note the solid line is for eye-guiding purpose only.

Download Full Size | PDF

The lifetimes of PL from Samples A in visible regime were obtained to be on the order of several tens of nanoseconds, which are at least one order of magnitude lower than that of Sample C and previous reports [4, 9, 12, 33]. Theoretically, the QC effects only relate the light emission wavelength to the particle sizes. The broad spectral profile of light emission is attributed to the size distribution of Si NPs, and the PL lifetimes of different Si NP samples at the same emission wavelength are expected to be the same or similar. The huge difference in PL lifetimes in visible regime for Sample A and Sample C also supports our argument that light emission of Sample A is from defects instead of QC effects.

The lifetimes of PL from Samples A and B in UV regime were also obtained to be on the order of several nanoseconds. Overall, the lifetimes of UV-blue emitting samples decrease monotonically with increasing photon energy of PL. The absorption of light for photoexcitation of carriers could be either through defect absorption band or band transition in the Si NPs. After the carriers are excited with high photon energy, they relax to defect-related states on the order of picosecond timescale [35] and PL transition occurs through defect luminescent centers. Shorter lifetime is thus expected for higher PL photon energy [15]. It should be noted that in Fig. 5(b), Sample C and previous reports [4, 9, 12, 33], which attribute mechanism of light emission to QC effects also have the similar trend of lifetime with photon energy. The lifetime also decreases monotonically with increasing photon energy of PL. According to the QC model, smaller Si NPs have faster PL lifetimes because the spatial confinement enhances the electron-hole wave function overlap and the momentum scattering probability (so-called the breakdown of momentum-conservation rule) to assist the radiative transition [12].

Note that the UV emission lifetime is on the order of few nanoseconds and its UV emission efficiency is in fact comparable to those found in many direct band gap semiconductors such as GaN, GaAs and CdSe with nanosecond lifetimes. This makes Si-based UV photonic devices possible based on annealed SiO_x. It is worthy to mention that PL of porous silicon, also with the nanosecond UV-blue emission from native oxide, is not stable and may not be suitable for application [10]. It should also be emphasized that defect-related UV-blue emission from annealed SiO_x is not trivial. Historically, QC effects were possible origin of UV-blue emission from oxide-free porous silicon containing smaller Si nanoparticles [4, 9]. In our UV-blue annealed SiO_x samples, Si NPs with size < 2 nm were indeed observed and have stable light emission because the embedded Si NPs are not exposed to air. The blue-shift in PL spectra with smaller Si NPs may also mislead to the conclusion of QC effects. Here, we have clearly demonstrated that the origin of UV-blue emission in this work is from defects although smaller Si NPs were formed in comparison with visible-NIR emitting samples (such as C and D).

4. Summary

We have reported that tunable photoluminescence between UV and NIR regime can be achieved by annealed SiO_x by controlling the ratios of O to Si in the condition of plasma-enhanced chemical vapor deposition (PECVD). When the O/Si composition ratio was increased to 1.5 or higher, smaller Si NPs (<2 nm) could form in the thermally annealed SiO_x. But we have concluded that the origin of UV-blue emitting samples is related to defects. These smaller Si NPs were found to be amorphous and turned into Si-SiO_x complexes instead. We have obtained the lifetimes of different PL wavelengths of UV-blue emitting samples. The lifetimes of visible light are on the order of several tens of nanoseconds and those of UV light are on the order of several nanoseconds. The nanosecond lifetimes of UV-blue emitting samples make it comparable with light efficiency of direct bandgap materials. This reveals great potential of annealed SiO_x for Si-based photonic devices from NIR-visible to UV regime.