1. Introduction

There has recently been extensive research [1–9] on the sensitization of rare-earth ions (especially erbium) by silicon nanoclusters (SiNC). SiNC typically exhibit a strong absorption cross-section for pump light in the visible region, and can efficiently transfer this energy to nearby Er ions [3–5]. This has raised hopes for Er doped waveguide amplifiers and lasers in a silicon-compatible material, transversely pumped by broadband sources such as LED arrays [10]. Both crystalline and amorphous SiNC can act as sensitizers, but there is some evidence that small [2, 10] and amorphous [11] SiNC are best in terms of maximizing Er luminescence efficiency. Motivated by the latter, we have investigated the properties of relatively low temperature annealed (~500 C) silicon monoxide (SiO) films doped with Er (Er:SiO) [12]. Bulk SiO is believed to be an inherently phase-separated material [13–15], with nanometer-scale amorphous domains of Si, SiO, and SiO₂. Some of us have recently reported [16] that annealed SiO films contain these same 3 phases, with their relative ratio determined mainly by annealing temperature. The photoluminescence (PL) of Er:SiO films is optimized at an annealing temperature of ~ 500 C [12], and the films contain a high density (~10¹⁹ cm^-3) of amorphous SiNC with a mean diameter ~ 2.8 nm. As a host material, these films exhibit efficient SiNC-mediated excitation of Er in agreement with similar work by Roberts et al. [17].

There have been numerous studies concerning PL optimization of Er-SiNC systems (as cited above), but considerable uncertainty remains with respect to the optimization of waveguide devices [18–22] based on these materials. Moreover, the underlying material properties that enabled early reports [10, 18] of gain in transversely pumped Er:SiNC amplifiers have yet to be fully explained or independently reproduced. Some of the key research questions are summarized as follows:

While silicon monoxide has been proposed as a low-loss guiding medium for the near infrared [17], to our knowledge this is the first experimental study of SiO waveguides. Using waveguides fabricated in the Er:SiO system (described in the introductory paragraph), we have conducted experiments aimed at assessing the basic properties of SiO as a waveguide core material, as well as addressing the Er:SiNC research questions outlined above.

2. Waveguide fabrication and characterization

The synthesis of Er:SiO films has been described in detail elsewhere [12]. Briefly, thin films were deposited by thermal evaporation of bulk SiO from a baffled box source. Er was incorporated via concurrent electron-beam evaporation of erbium oxide (Er₂O₃). Deposition of the separate materials was monitored via separate quartz-crystal thickness monitors. The as-deposited films are nearly stoichiometric SiO [16], and for the present work some films were doped with 1–2 at. % Er as estimated from relative deposition rates.

To facilitate optimization and interpretation of the transverse pumping experiments described below, the band edge absorption was estimated directly from normal incidence transmittance data [Fig. 1(a)] for SiO films on glass substrates. For an amorphous medium, the optical (Tauc) gap typically corresponds to the point at which the absorption coefficient attains 1000 cm^-1 [28]. For the present material, this occurs at a wavelength of ~575 nm; ie. the Tauc gap is ~2.2 eV. Using the technique developed by Swanepoel [28], the index of refraction was extracted in the near infrared region [Fig. 1(b)]. Note that the results were almost identical for undoped and Er-doped films. The solid curve is a fit based on the single oscillator Wemple-DiDomenico model [28], using the values E₀=5.4 eV and E_d=14.3 eV for the model parameters. As is typical [28], E₀ is approximately 2.5x the optical gap. The refractive index at 1550 nm (n~1.94) was confirmed using a prism-coupling technique on slab waveguides and is similar to values reported for SiO in the literature [29]. High index contrast waveguides and cavities can thus be realized using SiO as a core material in combination with standard SiO₂ and polymer cladding materials.

 

Fig. 1. (a). Transmittance versus wavelength for an SiO film on glass annealed at 500 C. Inset: Absorption coefficient versus wavelength near the band edge. Periodic oscillation above 500 nm is due to Fabry-Perot interference effects. (b). Refractive index versus wavelength for the same SiO film. The solid curve is a Wemple-DiDomenico fit as described in the text.

Download Full Size | PDF

Two types of SiO-based waveguides were fabricated. For the first set, the fabrication process was similar to that described in Ref. [30] and the sequence of steps is shown in Fig. 2. Silicon wafers were cleaned in a piranha solution and then placed in a wet thermal oxidation furnace to grow ~ 300 nm of SiO₂. CHF₃-based reactive ion etching was then used to pattern the SiO₂ layer. Using the patterned SiO₂ as an etch mask, high aspect ratio ribs were etched into the silicon in an inductively coupled plasma reactive ion etch (ICPRIE) chamber (Oxford Plasmalab 100). A cryogenic etch recipe based on sulfur hexafluoride (SF₆) and oxygen (O₂) process gases was employed [31]. Without removing the SiO₂ etch mask, the wafers were again placed in a thermal oxidation furnace to grow a thicker SiO₂ layer (~1.4 µm) for undercladding purposes. As shown in Figs. 2(b) and (c), and of key importance for realization of low loss waveguides, the ribs exhibit little roughness after this oxidation step. This was followed by deposition of either SiO or Er:SiO. Owing to the verticality of the rib structures, the evaporated film grows preferentially on the horizontal surfaces atop and between the ribs. A relatively thin and porous film forms on the sides of the ribs. This resulted in isolated regions of dense SiO or Er:SiO (~0.65 µm thick) atop the ribs; each of these regions becomes the core of a rectangular strip waveguide. Cleaved pieces (~2 cm side length) of the wafers were subjected to an optimized annealing process (500°C for 1 hour in a mixed atmosphere of hydrogen and nitrogen [12]), and then a benzocyclobetene (BCB) polymer (Cyclotene, Dow Chemical) upper cladding was spun-cast over the waveguides to improve mechanical durability and to reduce scattering losses. These pieces were further cleaved to produce waveguides of different length (~3–10 mm). The end facet of a completed waveguide is shown in Fig. 2(d). Due to their high core-cladding index contrast, the waveguides are predicted to support multiple propagation modes. However, even the widest guides (~6 µm) exhibited predominately single-mode behavior as described below. We speculate that the higher order modes are greatly attenuated by sidewall scattering and leakage into the silicon rib.

The second set of waveguides was fabricated using a more conventional approach. Briefly, silicon wafers were thermally oxidized to form a SiO2 undercladding (~2 µm thick) and then deposited with ~ 1 µm of Er:SiO and annealed as above. A thin (~20 nm) SU-8 negative photoresist layer was spun-cast, patterned, and developed to form strip-loaded waveguides. Although the SU-8 loading strips are very thin, these guides exhibited twodimensional confinement and single mode behavior (not shown). For both sets of guides, pieces free of upper cladding were kept aside to enable elemental analysis (by electron microprobe analysis) of the Er:SiO layer. Most of the results below pertain to the buried strip waveguides of 6 µm nominal width, which consistently exhibited the lowest overall insertion loss amongst both sets.

 

Fig. 2. (a)-(d) SEM micrographs showing the sequence of steps in the fabrication of buried strip waveguides (a) End facet view of a tall rib etched in silicon. (b) End facet view of a rib after thermal oxidation. The protrusions near the upper edges of the SiO₂ evolve as a result of leaving the SiO2 mask from the silicon etch step. (c) SEM image of the structure in (b), but from a different angle. (d) End facet view of the final waveguide structure, after Er:SiO core and BCB upper cladding deposition.

Download Full Size | PDF

Waveguide propagation was initially studied at a wavelength of 1300 nm, to avoid Errelated absorption. Light from a fiber pigtailed laser source was passed through a fiber-based polarization controller and then directly end coupled (using index matching fluid) to waveguides under test via a high numerical aperture (NA), small mode field diameter (MFD) fiber (Nufern, NA~0.26, MFD ~4.8 µm). To assess the experimental near-field mode profiles, light was collected from the output facets using an objective lens. The magnified image was mapped by scanning a pinhole-covered photodetector in the transverse plane. The system magnification was calibrated by performing the measurement on optical fibers with known mode-field diameters. Both the effective index method and a commercial software package (OptiBPM, Optiwave Corp.) were used to predict the modal properties of the buried strip guides. The fundamental TE mode solution at a wavelength of 1300 nm and for a core size of 0.65 µm by 6 µm is shown in Fig. 3(a). The experimental mode profile for a guide with approximately these same dimensions is shown in Fig. 3(b). There is good qualitative agreement between the results, especially in the horizontal (in-plane) direction. Since the theoretical MFD in the vertical dimension is ~ 0.8 µm, the discrepancy in that direction is likely due to the limited resolving power of the objective lens. To facilitate the loss study below, careful consideration was given to the coupling efficiency between the input fiber and the strip waveguides. Considering only power launched into the fundamental TE mode, the theoretical coupling loss for the 6 µm wide strip waveguide is ~4.4 [dB]. To experimentally assess the input coupling loss, light from the output facets was first collected using the objective lens and an iris-covered photodetector. The iris (along with an infrared CCD camera) was used to ensure that stray light propagating in cladding modes was eliminated. Without changing the input coupling conditions, the objective lens was replaced by an endcoupled fiber (of the same type used at the input). Assuming equal coupling loss at the input and output facets, the difference in insertion loss for the two measurements is approximately equal to the coupling loss between the fiber and the waveguide. From data collected for several 6 µm wide waveguides, the estimated coupling loss is 4.8+/-0.5 dB, in reasonable agreement with the theoretically predicted value.

 

Fig. 3. (a). Near field profile (as simulated using OptiBPM) of the fundamental TE guided mode for a buried strip waveguide. The geometry of the simulated waveguide is overlaid. (b). The experimental near field mode profile obtained from a buried strip waveguide with nominally the same dimensions as in part (a). A wavelength of 1300 nm was used in both cases.

Download Full Size | PDF

Several conventional methods for measuring propagation loss were considered. However, it was difficult to consistently realize high quality facets by cleaving because of the small chip sizes available (limited by the dimensions of the annealing furnace). As a result, Fabry-Perot and cutback techniques were deemed impractical. Analysis of light scattered from the top of the waveguides was also attempted, but results were not reliable because of the random distribution and size of scattering defects. Thus, an insertion loss method [32] was used to approximate propagation losses. Similar to above, output light was collected using an objective lens and delivered to an iris-covered detector or a CCD camera. By measuring the transmitted power with and without the waveguide sample in place, the overall insertion loss of a waveguide was estimated. Index matching fluid was used to minimize reflections, so that most of the loss is attributable to input coupling loss and propagation loss. By simply subtracting the previously determined coupling loss from the overall loss, a first-order estimate of the propagation loss is obtained. To be conservative, we used the theoretically predicted coupling loss (4.4 dB for the 6 µm waveguides) in all calculations. This measurement was performed on dozens of buried strip waveguides, yielding fairly consistent results. Table 1 lists results for 6 of the lowest loss waveguides measured. Because the samples are short and have relatively low loss, there is considerable error (approximately +/- 1 dB/cm [32]) inherent to these estimates.

Table 1. Propagation loss at a wavelength of 1300 nm, estimated by subtracting theoretical coupling loss from experimental insertion losses. Results shown are representative of the lowest loss buried strip waveguides measured, with core width 6 µm, and for TE polarized input light. Coupling loss of 4.4 dB was assumed in all cases.

View Table | View all tables in this article

In general, waveguides that exhibited higher loss also exhibited obvious scattering defects, reflecting the experimental nature of the process. Loss was higher for narrower waveguides, likely due to sidewall scattering. In all cases loss was higher for TM polarized light. This can be attributed to higher leakage into the silicon rib through the relatively thin lower cladding. In fact, the leakage loss contribution was estimated theoretically to be >0.1 dB/cm and >0.3 dB/cm for the TE and TM fundamental modes, respectively. It is expected that further refinement of fabrication processes could result in SiO-based waveguides with even lower losses. These results suggest that SiO is a promising material for high index contrast, low-loss waveguides operating in the near infrared.

3. Spectroscopic analysis

Absorption and PL spectroscopy were used to estimate the magnitude of the ⁴ I _15/2 to ⁴ I _13/2 cross-sections. Broadband transmission scans were obtained using an experimental setup similar to that described in Section 2, but with the 1300 nm laser source replaced by a tunable laser (Santec) operating in the 1520–1620 nm range. Light from the tunable laser was passed through a polarization controller and then a 90/10 coupler, the latter used to tap a portion of the input light for calibration purposes. The 90 percent port of the coupler was delivered to waveguides under test via the Nufern high NA fiber, and light at the output facets was collected by an objective lens and delivered to an iris-apertured photodetector. The measurement system was calibrated to remove any wavelength dependence by performing reference scans without a waveguide sample in place. The correctness of this procedure was verified by subsequently performing scans on undoped SiO waveguides. After calibration their response was flat (within ~0.5 dB) across the entire 1520–1620 nm band, suggesting that the wavelength dependence of scattering losses is minimal over that range. Calibrated scans for Er-doped waveguides are shown in Fig. 4(a), with the curves referenced by assuming zero Er-related absorption at 1620 nm. The latter assumption is only approximately true, and might lead to a slight underestimate of the cross-section.

 

Fig. 4. (a). Transmission scans for strip loaded and buried strip waveguides, corrected for system response and referenced to 1620 nm. (b). The estimated absorption cross-section spectrum for a buried strip guide (6 µm core width), extracted from the data in (a). Also shown is the estimated emission cross-section spectrum, based on experimental photoluminescence data and scaled using the Fuchtbauer-Ladenburg expression.

Download Full Size | PDF

Using the data in Fig. 4(a), the ⁴ I _15/2 to ⁴ I _13/2 absorption cross-section (σ ₁₂) can be estimated via the expression α_erb ~ ΓN_erbσ ₁₂, where α_erb is the erbium related absorption coefficient, Γ is the core confinement factor, and N_erb is the active Er concentration. Electron Microprobe Analysis (EMPA) was used to assess the total Er concentration in each case. The EMPA results were calibrated using reference samples previously characterized by Rutherford Backscattering Spectroscopy (RBS). Note that the concentration of optically active Er (in the Er³⁺ state) can be less than the total Er concentration, if some portion of the Er is in metallic clusters or in the form ErSi₂ [5–6]. Given the strong Er-related absorption observed, and in the absence of evidence to the contrary, we’ve assumed that all of the Er is active. The extracted cross-section spectrum for a typical buried strip waveguide (core width 6 µm) is shown in Fig. 4(b). The analysis of the peak absorption cross-section is summarized in Table 2, for both sets of waveguides. The peak cross-section in both cases is approximately 2 to 3 times the value for Er in SiO₂, and is in good agreement with recent estimates for similar Er:SiNC systems [22–23]. The enhanced cross-section can be partially attributed to a higher local field correction factor in our high index films [23].

Table 2. Peak ⁴I_15/2 to ⁴I_13/2 absorption cross-sections estimated for the two different types of waveguides studied. For the buried strip guides, different core widths (4-6 µm) produced slightly different estimates within the range indicated. This might be due in part to the presence of higher order modes in wider guides.

View Table | View all tables in this article

Waveguide PL spectra were collected for both transverse and co-propagating pump arrangements (see below). The emission cross-section spectrum can be extracted from the PL spectrum using the Fuchtbauer-Ladenberg expression [33]:

where λ_P is the wavelength of peak emission, I_PL(λ) is the PL spectrum normalized to a peak value of 1, n is refractive index, τ_rad is the radiative lifetime for the transition, and Δλ_eff=∫I_PL(λ)dλ is the effective linewidth. From lifetime measurements in the low Er concentration limit [12], we estimate τ_rad~3 ms for our Er:SiO films. Using this value, a typical emission cross-section spectrum was plotted alongside the absorption cross-section spectrum in Fig. 4(b). Encouragingly, the estimated peak emission and absorption crosssections are of similar magnitude. The metastable lifetime (τ ₂₁) was measured using the setup described in our previous work [12], and was approximately 0.5 ms and 1 ms for the strip loaded and buried strip waveguide samples, respectively. The reduction in lifetime with increasing Er concentration is attributable to concentration quenching effects [4, 12]. Finally, we estimate Δλ_eff~57 nm, which is somewhat larger than the value (47 nm) reported by Roberts et al. [17] and considerably larger than in most silicate glasses [33].

4. Optical pumping

As mentioned above, we have previously described [12] non-resonant pumping of Er ions in SiO. To assess the feasibility of Er:SiO for waveguide amplifiers, experiments were conducted using either a transverse or a co-propagating pump beam. The transverse pump source was a frequency-doubled Nd-YAG laser operating at 532 nm, where the SiO films have an absorption length ~ 4 µm. This ensures fairly uniform pumping of the entire depth of the SiO core material. Also, the polymer upper claddings are transparent at 532 nm. The pump laser was focused to a line with 1/e ² beam widths ~250 µm and ~4 mm, as measured using a beam profiler.

The focused green light was used to pump 4 mm long buried strip waveguides, resulting in reasonably uniform pump intensity along the entire length of the guide. Neutral density filters were placed in the beam line to control the pump intensity. For simplicity, the beam area was defined as the product of the 1/e ² widths and this area was used in estimating the effective pump intensity. Using the tunable laser described above as a signal probe, transmission scans in the 1520 to 1620 nm range were taken at each of several pump intensities. A schematic of the experimental setup is shown in Fig. 5(a), and a photograph of a waveguide under test is shown in Fig. 5(b). As above, the intrinsic wavelength dependence of the measurement system was taken into account. By collecting luminescence data without the input probe signal, we also verified that amplified spontaneous emission could be neglected in the analysis.

 

Fig. 5. (a). Schematic of the transverse pumping experiment. PD1 and PD2 are photodetectors, PC is a polarization controller, and WG is the waveguide under test. (b). Photograph of a waveguide under transverse pumping with green light.

Download Full Size | PDF

A typical set of calibrated transmission scans is shown in Fig. 6(a), revealing two significant trends. The dominant trend is an induced loss that increases monotonically with pump intensity and is relatively wavelength independent. This is the signature of FCA/CCA [19–21]. A more subtle trend is the reduction of pump-induced loss within the Er emission band peaked near 1535 nm. This second trend is clarified by plotting the Er related signal enhancement (SE_erb) [20],

with T(I_P,λ) the transmitted power at wavelength λ for pump intensity I_P. This assumes negligible Er-induced signal change at 1620 nm and also that the FCA/CCA is wavelength independent. The result is plotted in Fig. 6(b), clearly evoking the shape of the Er emission band shown in Fig. 4(b).

 

Fig. 6. (a). Transmission scans for a buried strip waveguide under transverse pumping by 532 nm light of varying intensity. (b). The erbium related signal enhancement extracted from the data in (a), as described in the text.

Download Full Size | PDF

The fraction of inverted Er ions can be estimated as [18]

where N₂ is the volume density of Er ions in the metastable (⁴ I _13/2) state. Based on this equation, ~16 % of the active Er ions were inverted for the highest pump intensity tested. Given the estimated active Er concentration (~4×10²⁰ cm^-3) and the SiNC density (~1.3×10¹⁹ cm^-3) in the present films [16], this corresponds to ~ 5 excited Er ions per nanocluster. There is clearly an onset of saturation in the erbium-related signal enhancement data (and thus in the inverted fraction of Er ions) for the data in Fig. 6(b). As mentioned in the introduction, this type of behavior has often been reported in Er:SiNC systems and attributed to various physical mechanisms. To illuminate the origin of this saturation in the present waveguides, we collected pump-probe data [similar to that in Fig. 6(a)] using a 980 nm wavelength copropagating pump. In that case, the pump and probe were combined using a standard fiber WDM coupler. To separate pump and probe light at the output of the waveguide, the input probe light was chopped and lock-in detection was employed at the output. The wavelength dependence of the system was taken into account as above. The coupled pump power was estimated by taking into account the theoretical coupling loss at 980 nm. Furthermore, the pump intensity inside the waveguide was estimated by dividing the coupled power by the effective area of the fundamental waveguide mode.

Figure 7(a) shows the relative transmission at the peak and outside of the Er emission band, as a function of pump intensity. While FCA/CCA is reduced and seems to saturate at a lower level for 980 nm pump, it is still the dominant pump-induced mechanism. The photon energy for 980 nm pump is well below the optical bandgap of the host films. However, amorphous materials are characterized by weak absorption extending well below their nominal bandgap [28]. Charge carriers can thus be generated by pump photons with wavelength extending into the near infrared [17]. Long transients (seconds to tens of seconds) associated with the FCA/CCA, as reported by others [19, 21], were also observed here. Another interesting observation was the absence of green emission [6] under 980 nm pumping, in spite of the fact that significant homogeneous and inhomogeneous upconversion is expected for the Er concentrations studied here (see below). We speculate that this is due to an efficient and non-radiative energy back-transfer from the upper Er levels to the SiNC [34].

 

Fig. 7. (a). Relative transmission (relative to the unpumped case) inside and outside the Er emission band, for transverse and co-propagating pump. (b). Erbium inversion versus pump intensity, extracted from the raw data as described in the text.

Download Full Size | PDF

The inverted fraction of Er is plotted for both pump configurations in Fig. 7(b). There is strong saturation evident in both curves, but especially for the resonant pump. Given the range of 980 nm pump intensity studied, it is evident that even under resonant pumping only ~15–20 % of the Er ions are invertible. This is in contrast to other results (for films with similar Er concentration as studied here) [20, 22, 26–27], where ~1–12% of Er ions were excitable via SiNC absorption but ~100% by resonant means. In the present case, the excitable fraction limit (for SiNC-mediated excitation) cannot be entirely attributed to a distance limitation for the SiNC to Er energy transfer [26]. Nonlinear de-population mechanisms can also limit the Er inversion level; these include Auger de-excitation of Er by energy transfer to excitons in SiNC [5], homogenous upconversion [4–5], and excited state absorption (ESA) [27]. For low to moderate pump intensity, a limit of one excited Er ion per SiNC has been associated with both Auger de-excitation and ESA [5, 27]. Again, this is not entirely consistent with the data presented above. However, given the high Er concentrations in our waveguide samples, it is very likely that homogeneous upconversion [4] and concentration quenching [4, 12] (which increases the non-radiative decay rate of the metastable Er level) contribute to the saturation in Fig. 7(b).

Since the curves for resonant and non-resonant pump seem to converge to a similar saturation level, it appears that ~80% of the Er ions are non-invertible (unbleachable). Wojdak et al. [25] reported similar behavior for films with similar Er concentration; 70% of their Er ions were unbleachable by any means. Perhaps the simplest possible explanation for this behavior is the well-known pair-induced quenching (PIQ) (also called inhomogeneous upconversion) mechanism [35]. In PIQ, paired (or clustered in general) Er ions are active in the sense that they contribute to the usual absorption features. However, they are noninvertible due to rapid ion-ion interactions within the clusters [36]. The onset of significant clustering in pure SiO2 occurs for N_erb~10¹⁸ cm^-3 [35], well below the Er concentrations studied here. Given that Er is generally thought to reside preferentially within the SiO₂ regions of Er:SiNC systems [4, 10, 26], it is somewhat surprising that PIQ is rarely cited [5] as an explanation for the excitable fraction limit. Further, the solubility of Er in Si is even lower than that in SiO₂ [8]. However, it should be noted that Pellegrino et al. [26] saw no evidence of Er clustering in Er:SiNC films annealed at temperatures less than 850 C. Within the framework of a simplified two-level model for the Er ion [3], the inversion level in the presence of clustering can be approximated as [35]

where k is the fraction of clustered (unbleachable) Er ions, σ_eff is the effective excitation crosssection, ϕ_P is the pump flux, and τ₂₁ is the lifetime of the metastable Er level. Thus, we can define a characteristic saturation flux: ϕ_ps=1/(σ_effτ₂₁). Assuming direct resonant excitation of Er ions in the 980 nm pump case, and using σ_eff~10^-20 cm² (ie. similar to the values in Table 2) and τ₂₁~1 ms (see the discussion in Section 3 and in Ref. [12]), we estimate a saturation pump intensity I_PS~2×10⁴ W/cm². This is in reasonable agreement with the curve in Fig. 7(b). However, there is likely some indirect excitation of Er via SiNC (given the CCA/FCA observed for 980 nm pump), which will increase σ_eff. For the 532 nm transverse pump case, using σ_eff~10^-15 cm² [5, 12] and the same value for τ₂₁ produces I_PS~0.4 W/cm², also in reasonable agreement with the data. Note that the 532 nm pump is quasi-resonant with Er absorption lines, causing some direct excitation. However, given typical cross-sections for Er ions in insulating hosts (<10^-20 cm²) and the maximum photon flux in our transverse pumping experiment (~10¹⁹ cm^-2), the fraction of directly excited Er ions can be neglected.

The data in Fig. 7 clearly confirm the 3–5 orders of magnitude increase in excitation crosssection widely reported for Er:SiNC systems [1–9]. Compared to Er:SiNC systems with lower silicon content [20, 25], the Er:SiO films enable a higher excitable fraction of Er and the Er inversion saturates for considerably lower pump intensity. In fact, significant Er inversion is exhibited at pump intensities within the range of high power LEDs [10]. However, the onset of CCA/FCA in the present films occurs at pump intensity similar to that in reference [21] and much lower than that in Ref. [20].

5. Summary and conclusions

Our results show that SiO is a promising waveguide medium for the near infrared, characterized by high index (n~1.94) and low scattering and absorption losses. Highly confining waveguides with modal effective area ~4 µm² and propagation loss ~ 1 dB/cm were realized. As a host medium, appropriately annealed SiO enables efficient broadband excitation of Er ions. Saturation of the Er inversion was demonstrated for transverse pump intensity <10 W/cm² at 532 nm. However, at least 80% of the Er was not invertible, probably due primarily to inhomogeneous upconversion in our heavily doped films. Optimization of the Er concentration might address this limitation. A greater barrier to the use of Er:SiO as an amplifying medium is the significant carrier-induced absorption we observed at modest pump intensities. For both a transverse 532 nm pump and a co-propagating 980 nm pump, carrierinduced absorption exceeded Er-induced gain at all pump intensities investigated. The results provide insight into the impairments that must be addressed if Er:SiNC systems are to achieve practical implementation.