1. Introduction

Chalcogenide glasses (ChGs) are of great interest for planar photonics on account of the many interesting optical properties they possess. Their transparency deep into the infrared region (which includes most of the important range for vibrational spectroscopy) makes them very attractive for applications in chemical and biosensing [1]. Their high linear refractive index (n≈2-3) and large nonlinear refractive index n₂, (100-1000x silica) with low linear and nonlinear absorptions at telecommunication wavelengths, make them serious candidates for applications in on-chip all-optical signal processing [2]. The low phonon energy of chalcogenide glasses is also advantageous for rare-earth doped devices to enable transitions unavailable in common high phonon energy hosts like silica (both MIR transitions and others such as the 1300 nm transition in Pr³⁺) [3]. The large rare-earth solubility in some ChGs [4] also promises high gain in a short device which is essential for integrated optics applications. The possibility of having both high gain and outstanding nonlinear optical performance in a single device could also lead to practical low pump power all-optical signal processing devices. However, despite intensive study of bulk rare-earth doped ChGs, so far, the authors are only aware of four reports of gain or lasing in ChG [5–8] prior to 2015, these having a maximum wavelength of 1330 nm. There are no reports of high inversion in Erbium based devices.

Recently, limited internal gain between 1570 and 1630 nm was achieved for the first time in a chalcogenide based on Erbium doped As₂S₃ planar waveguides [9]. In that report, the total Erbium ion population inversion was estimated at 30% limited by clustering of the Erbium even with concentrations as low as 0.15 mol% (~4.5 × 10¹⁹ ion/cm³). This is essentially the minimum usable doping level for a planar waveguide device, as at this concentration, the peak absorption (at 1540 nm) is 1.7 dB/cm. With 1490 nm pumping, the maximum inversion is 65% which limits the maximum possible internal gain to ~0.5 dB/cm, even before waveguide losses, ion-ion effects, etc are considered which further reduce the net gain. Long waveguides with lower doping are not a solution here as the passive losses at the pump wavelength then become a problem (passive losses of ~3-5 dB are tolerable for the pump which limits amplifier length to perhaps 15 cm for the ~0.3 dB/cm propagation losses common for high index contrast waveguides of this type [10]). Obviously, significant improvement is necessary to realize a useful high gain amplifier. Different approaches to doping Erbium into As₂S₃ films for a better Erbium homogeneity without thermal annealing or the use of hosts with better intrinsic Erbium solubility are considered promising solutions to this issue.

Amongst the family of ChGs, Gallium containing materials are known to have better rare-earth solubility. Rare-earth doped Ga-containing ChGs have been extensively investigated, with for example promising outcomes in terms of photoluminescence properties e.g [4, 11, 12]. Of particular note amongst the published findings, Ikuta et al.’s report [4] underlines the utility of Gallium in increasing Erbium solubility, the measured photoluminescence (PL) intensity from Er³⁺ doped (GeSe₂)_1–x(Ga₂Se₃)_x glasses being found proportional to the Er³⁺ concentration up to 2 at % indicating no clustering. The importance of Gallium for Er³⁺ activation was also studied and the PL results indicated the critical Gallium to Erbium concentration ratio was about 5, with excess Gallium likely causing issues with homogeneity [4].

However, only a few reports on rare-earth doped Ga-containing chalcogenide films exist and none on the use of such materials in waveguide amplifiers. Takahiko et al. reported the properties of Ga-Ge-Se films on fused silica substrates deposited by sputtering, and the lifetime was 1.8-2.6 ms for ⁴I_13/2 energy level when excited by a 973 nm laser [13]. In Nazabal’s report [14], Ga-Ge-Sb-S(Se) films were fabricated by pulsed laser deposition or RF magnetron sputtering, both physical and optical properties of the films were investigated. The ⁴I_13/2 level fluorescence decays decreased from 1.6 ms to 1.1 ms in sputtered films as the Erbium concentration increased from 0.3 to 1.5 at.% (3.4 × 10¹⁹ to 1.65 × 10²⁰ ions/cm³). So, whilst much promise has been revealed, actual investigation into amplification in waveguides has not been undertaken.

A large part of the difficulty in film/waveguide amplifier experiments, particularly in Gallium containing ChGs, relates to the difficulties in fabricating high quality Erbium doped films. Thermal evaporation is perhaps the simplest and therefore most widely used method of preparing ChG films. However many ternary and quaternary ChGs display the undesirable property of forming phase separated molecular liquids on melting, with the different phases boiling off at different temperatures and rates. This can result in films with often quite different composition to the starting materials, e.g [15]. The weight percentage of Gallium was found to be significantly reduced from the value in the starting material in films deposited by standard thermal evaporation [16]. Fortunately, elemental co-thermal evaporation provides a solution to this issue as it offers the possibility of controlling the evaporation rate of each element in the film independently. Thus, the final composition of the film can be precisely controlled.

This co-thermal evaporation approach was applied in previous work on Erbium doped Ge-Ga-Se films [17]. Accurate film compositional control was achieved. An intrinsic lifetime of 0.87 ms in a 0.7 mol% doped film showed the potential solubility of Erbium in this host material. However, a high density of particulates was observed on the film surface, which caused a high propagation loss in the finished waveguides. After careful investigation, these particles were found to be metallic in nature, likely ejected from the Gallium source.

In this work, particles from Gallium were diminished with a refinement of the evaporation setup. High quality Erbium doped Ge-Ga-Se films were characterized optically, and then patterned into waveguides via photolithography and ICP dry etching. The performance of the waveguides under resonant pumping was then investigated and high inversion levels demonstrated.

2. Experiments

Erbium doped Ge-Ga-Se films were deposited by co-thermal evaporation where separate Erbium, Gallium, Germanium and Selenium sources (each equipped with their own dedicated quartz microbalance rate monitor plus one for the wafers), were employed with high purity (5N) materials. The Gallium furnace also had a baffle fitted to it to prevent line of sight evaporant particle ejection during deposition. Films were deposited on 100 mm diameter thermally oxidised Silicon wafers (2 µm of oxide) at room temperature in a vacuum of ~10⁻⁷ Torr. Due to known issues in plasma etching Erbium containing films arising from Erbium involatiles, a 450 nm As₂S₃ layer was deposited without breaking vacuum after the 650 nm Erbium doped Ge-Ga-Se layer to enable a strip loaded waveguide geometry where etching of the Erbium containing layer is not necessary. After deposition, the bilayer film was thermally annealed at 130 °C for 24 hours in a vacuum oven to bring the As₂S₃ film back closer to the bulk state. The thicknesses and refractive indices of the two layers were measured using a dual angle spectroscopic reflectometer (SCI FilmTek 4000) using a Tauc-Lorentz model for each film to fit the reflection data. The 650 nm Erbium doped Ge-Ga-Se layer had a refractive index of 2.43 at wavelength of 1550 nm, while the 450 nm As₂S₃ layer had a refractive index of 2.41 at 1550 nm after annealing. This minor difference in refractive index enables a well confined rib waveguide.

The obtained Erbium doped Ge-Ga-Se film was carefully checked under an optical microscope using a × 100 objective in dark field mode, and no particles were observed, indicating the baffled furnace effectively stopped the Gallium spitting that was previously problematic [17].

The composition of the Ge-Ga-Se layer was determined as Ge_2.6Ga_27.7Se_69.7 by Energy Dispersive X-Ray Analysis (EDXA). This was a considerable difference to the films obtained previously [17] and the desired composition (Ga₁₁Ge₂₅Se₆₄). It was not possible to tune back to the desired composition as the Ge furnace was operating very close to its temperature limit and so the Ge flux could not be increased, and slowing the rate of the other evaporants was not possible as the Er rate became unstable at the low rate required for the desired doping after reducing the rates of the other furnaces. The reasons behind the lowered Ge deposition rate compared to previous experiments were uncertain and are under investigation. Also, because of the limited compositional resolution of EDXA, the low Erbium concentration cannot be accurately read from the EDX results directly. However, comparing the optical Erbium absorption peak in the final waveguides (discussed below) with our previous Erbium doped films in a similar host, the Erbium concentration in this film was estimated at 0.5 mol% (~1.5 × 10²⁰ ion/cm³).

An all-fiber confocal setup as previously described in [18] was employed to measure the PL decay of the ⁴I_13/2 excited state with laser excitation at 1490 nm. Here an intrinsic lifetime which approximates the true radiative lifetime in the material is defined as the decay constant from a fit to the later pure single exponential part of the PL vs time curve, typically at a time when the PL intensity has dropped >10-100x from the initial value. Under these circumstances all population related quenching effects should be small and the intrinsic lifetime should approximate the radiative lifetime in the absence of impurity related or multiphonon based nonradiative decay. With this definition, an intrinsic lifetime of 0.99 ms was obtained from this film under an excitation intensity of 350 kW/cm². The dependence of lifetime and PL intensity versus pump power is shown in Fig. 1. Intrinsic lifetime remains at 0.99 ms showing no dependence with pump intensity ranging from 15 kW/cm² to 1200 kW/cm² as would be expected. The 1/e lifetime drops from 0.88 ms at 15 kW/cm² to 0.80 ms at 1200 kW/cm² pump intensity, indicating interactions between pump photons and excited ions or ion-ion effects, but the 1/e lifetime is only slightly less than the intrinsic lifetime and exhibits only a weak dependence on pump intensity. A PL decay curve under 350 kW/cm² excitation intensity is shown as an insert in Fig. 1(a), and was typical of the decay curve shape seen at all pump intensity levels. Whilst a difference between the 1/e and intrinsic lifetimes was observed even in bulk glasses at this concentration [17], in the film sample the difference is greater than in the bulk glass. This likely indicates that either other parasitic effects are occurring or that the Erbium is not homogeneously distributed in the film. The approximately quadratic trend of PL intensity versus pump intensity shown in Fig. 1(b), also implies either or both of the occurrence of energy exchange effects and/or PL saturation under high pumping intensity.

 

Fig. 1 (a) 1/e lifetime, intrinsic lifetime and a lifetime decay curve (inserted), and (b) PL intensity versus pump intensity at 1490 nm.

Download Full Size | PDF

A longer intrinsic lifetime of 1.34 ms could be achieved by annealing Er doped Ge-Ga-Se films at 280 °C for 24 hours indicating the Erbium was going into solution better as expected from the solubility of Erbium in Ga containing hosts [4, 11]. The T_g of the obtained film could not be directly measured, but from the literature, glasses with similar composition (Ge₅Ga₂₀Se₇₅) had a T_g of 270 °C [19], and T_g increased with increasing Ga content due to rise in mean bond energy of the system. Thus it is expected that film used in this experiment had a T_g higher than 270 °C. The large T_g difference between the layers (As₂S₃ layer has T_g of 180 °C), limited annealing to only 130 °C to protect the As₂S₃ layer and consequently potentially lowered Er performance in the waveguides due to polyatomic Er evaporation [9].

3. Waveguide Results and Discussion

3.1 Waveguide fabrication and propagation loss measurement

The rib waveguides were structured using contact lithography with standard positive Photoresist and Reactive Ion Etching (RIE) with CHF₃ gas [20]. 400 nm of the 450 nm As₂S₃ was etched away to leave the Erbium doped layer pristine and this process resulted in a smooth surface. The cross section of the designed structure for a 2 μm waveguide, cross section of obtained 2 μm waveguide by SEM and simulated TE fundamental mode are shown in Fig. 2. With this structure, the overlaps for both TE and TM fundamental modes and the Er doped area were around 71% for both pump and signal wavelengths as calculated via a public domain full vector finite difference based code [21], thereby promising effective use of both the pump energy and the excited Erbium ions.

 

Fig. 2 (a) Details of the designed structure, (b) cross section of obtained 2 μm waveguide by SEM and (c) simulated TE fundamental mode for the 2-μm waveguide.

Download Full Size | PDF

Cut-back measurements were performed in TE mode at 1610 nm as this wavelength is beyond the end of the Erbium absorption tail and so provides a true measure of the propagation losses, and the results are shown in Fig. 3(a). The propagation loss of the 2 µm waveguides was 1.7 ± 0.1 dB/cm, measured starting with a waveguide length of 62 mm cut to 40 and 22 mm lengths. Compared with previous data [17] and the thinner total waveguide thickness with greater etch depth, there is an improvement in propagation loss, but the losses are still much higher than the ~0.3 dB/cm obtained with pure As₂S₃ waveguides of slightly smaller dimensions [10]. The wavelength dependence of the propagation loss in a 2 µm waveguide was measured using a supercontinuum source and optical spectrum analyzer and is shown in Fig. 3(b). It is well known that propagation loss following a 1/λ² dependence is expected for sidewall scattering, whilst a 1/λ⁴ dependence is indicative of scattering off nanoscale inhomogeneities as Rayleigh scattering. Very good propagation loss fitting was obtained for the measured waveguides with the 1/λ² formula, indicating the sidewall roughness in the waveguide is the main contributor to propagation loss. This waveguide has an air top cladding resulting in larger scattering losses from the larger index difference compared the polysiloxane top clad (n = 1.535 at 1550 nm) all As₂S₃ devices. Sidewall scattering is often approximated as scaling with the square of the core-clad index difference [22], which would imply a sidewall roughness loss of ~0.8 dB/cm might be expected in this device if the etch quality was up to the standards of the all As₂S₃ waveguides and scaling from the 0.3 dB/cm figure there.

 

Fig. 3 (a) Raw cutback insertion loss data with linear fit line, (b) measured optical propagation loss spectrum of 2 μm waveguides with fitted data.

Download Full Size | PDF

Nanoscale phase separation is one process that is known to occur in chalcogenide glasses [23] that can produce the nanoscale inhomogeneities that manifest as Rayleigh scattering. Erbium clustering or the growth of different structures around the Erbium itself can also produce similar effects as was observed in the case of previous work on Er:As₂S₃ [9]. In waveguides measured in this work, the Rayleigh scattering component is so small that is almost absent from our fitting results, indicating there were far less film growth issues. Also as will be shown shortly, based upon the pumping results obtained from this waveguide, very little of the Erbium is clustered in this film and this is consistent with the result of the above analysis that scattering caused by doped Erbium is unlikely in this waveguide. Therefore it appears that the etching in this device was not as good, and in fact given the square dependence on roughness and the acute sensitivity to the correlation length in this region of operation [24], then only a small change is required to degrade the loss to the levels observed.

3.2 Optical enhancement and photoinduced absorption

Signal enhancement in this paper is defined as the output power of a test source with the rare earth pump source on minus the amplified spontaneous emission power (test source off) all divided by the output power with the rare earth pump off. Thus an enhancement factor of unity corresponds to transparency and values above unity to optical gain. Given the low Er³⁺ absorption at 1490 nm evident in Fig. 3(b) above, a diode laser operating at 1505 nm was employed as the pump source. A heavily attenuated supercontinuum source (<-35 dBm input power across the Erbium band to avoid exciting the Erbium ions) with range from 600 to 2000 nm was used as the signal source. The pump source and signal were combined together through a 90/10 coupler (pump on 90% port) and then delivered to a 26 mm long waveguide via a lensed fiber with a 2.5 µm 1/e² diameter. The emerging light was collected by another lensed fiber and then recorded on an optical spectrum analyzer (Ando AQ 6317).

The resulting optical enhancement curve and its dependence on excitation power intensity are shown in Fig. 4. From the enhancement curve, about 9 dB enhancement was achieved with the available maximum 4100 kW/cm² pump intensity launched into the waveguide. However, it is clear from looking at the data at 1620 nm, where no enhancement would be expected, that the whole level of the curve drops by up to 4 dB at maximum pump power. Increased loss was also seen across the whole transmission spectrum of the device and so does not just represent excited state absorption on the long wavelength side of the Erbium absorption. This loss was completely and relatively rapidly reversible (~1 minute) by turning off the pump, and was also verified to be associated with the Erbium as injecting high power at 1430 nm (where the Erbium absorption is negligible) had no effect.

 

Fig. 4 Erbium absorption curve and optical enhancement spectrum under 1505 nm excitation.

Download Full Size | PDF

The correlation with pumping the Erbium strongly suggests an effect connected with upconversion to shorter wavelengths, especially as such processes will be enhanced in chalcogenide glasses compared to oxide hosts due to the much longer upper state lifetimes resulting for much lower multiphonon recombination rates [25]. Illumination with light at and above the band gap has previously been established to result in photoinduced wideband absorption in some materials [26]. Whist most of the study was on As₂S₃ and As-S-Se based fiber and showed relatively weak absorption at 1550 nm (<0.2 dB/m at 10 mW/cm² of transverse bandgap light [26]), effects were also noted in Ge-As-Se glasses but little data presented. With this in mind 1550 nm light was propagated through waveguides made from undoped Ge₃Ga₂₈Se₆₉, As₂S₃, Ge₃₃As₁₂Se₅₅ (unannealed) and Ge_11.5As_23.5Se₆₅ (unannealed) whilst exposing them to ~10 mW/cm² red and green LED based light from above. The As₂S₃ waveguide exhibited a change of <0.01 dB/cm propagation loss, whereas all the Ge containing waveguides showed much stronger effects. The effect was most dramatic in the Ge₃₃As₁₂Se₅₅ waveguide which exhibited an increase in propagation losses approaching 10 dB/cm, compared to ~0.5 dB/cm in the Ge_11.5As_23.5Se₆₅ and ~1 dB/cm in the Ge₃Ga₂₈Se₆₉ for red illumination and ~2 dB/cm for green. The bandgap of this Erbium doped Ge-Ga-Se film is about 1.75 eV (708 nm) based upon the results from FilmTek measurements. The time response of the photoinduced absorption in the materials was best fitted with a triple exponential decay. For the Ge-As-Se glasses, recovery was complete within 2 seconds, whereas in the Ge-Ga-Se glass complete recovery required ~90s with fitted exponential time constants of ~0.1s, ~1.5s, and ~30s. In all cases we also looked for a fast decay component indicating a free carrier component, but no response faster than ~50 ms could be found even when looking with a 1MHz bandwidth detection system. The much increased decay time may indicate a higher density of defect states in the sub bandgap region where the carriers move between them as a result of thermalisation thereby surviving longer. Annealing studies are needed to resolve this.

Up-conversion related emissions at 980, 800, 670, 540 and 520 nm, associated with radiative decay from the ⁴I_9/2, ⁴F_9/2, ⁴S_3/2 and ²H_11/2 states respectively, are all located in the bandgap or Urbach tail of this host material. Due to the small mode area (~2 µm²) of the waveguides, the intensity of up-conversion related emissions in the waveguides could certainly be high enough to induce photoinduced absorption, and this coupled with the insensitivity to high power 1430 nm light strongly suggests this is the cause of the increased loss. In addition, Er³⁺ ions excited to sates with energy equal to or higher than the bandgap energy of the glass matrix may directly transfer energy to the electronic states of the glass matrix [27, 28]. This may also be a contributor for the observed photo-induced loss, but data is not reported on such optical loss.

To get an accurate Erbium population inversion ratio, the effects of photoinduced loss should be eliminated from the calculation of enhancement. Andriesh [29] found an exponential fit to the photoinduced absorption in the above bandgap and Urbach tail region, but the dependence then became non exponential beyond ~1100 nm. Given that the data being used is in the >1100 nm region and that a different material was being considered, a phenomenological approach was adopted. The photoinduced losses were extracted from the wideband difference between the zero-excitation and the 4100 kW/cm² excitation curves and empirically found to be well fitted by the formula: $loss=k/{\lambda }^4+b$, where the k and b are fitting coefficients and λ is the wavelength. At this stage no interpretation can be placed on this dependence. Figure 5 illustrates the excellent fit afforded by this relationship.

 

Fig. 5 Loss increment curve due to photoinduced absorption effect and the fitted curve.

Download Full Size | PDF

With the fitted photoinduced loss curve, the real signal enhancement curve under 4100 kW/cm² pump intensity was obtained and is shown in Fig. 6(a) as the solid line. Optical enhancements corresponding to different population inversion ratios from 10% up to 65% were calculated and also shown in Fig. 6(a) as dashed lines [30]. From these curves, the experimental enhancement curve fits with the calculated line having 50% population inversion. This is far higher than obtained in the best previous measurement [9] and produces internal gain across most of the spectrum, as shown in Fig. 6(b). The sharp peak and big power fluctuation around 1505 nm in Fig. 6(b) is caused by the residual pump power centered at 1505 nm. The maximum pump efficiency curve is plotted in Fig. 7(a), which represents the maximum inversion possible versus pump wavelength taking into account the effect of pump initiated stimulated emission under the resonant pumping in this region. The curve was calculated according to the method in [31]. From the curve, the maximum possible inversion calculated for 1505 nm pumping is 65%. To attain reasonable gain requires an inversion of 60-65% as is evident from Fig. 6(a) where more than another 4 dB of enhancement is available at the peak, and of course more actual gain with higher doping.

 

Fig. 6 (a) Calculated optical enhancement as a function of the population inversion and experimental optical enhancement; (b) experimental optical enhancement and the Erbium absorption from a 26 mm waveguide.

Download Full Size | PDF

 

Fig. 7 (a) Calculated maximum pump efficiency versus pump wavelength; (b) measured peak enhancement at 1538 nm as a function of pump power (measured at fibre connector).

Download Full Size | PDF

Peak enhancement at 1538 nm was extracted from each measured enhancement curve in Fig. 4 and its dependence on pump intensity is plotted in Fig. 7(b). It is clear that the saturation region has almost been reached and further increases in the enhancement will be small (~1dB) corresponding to a further small increase in the inversion to perhaps 55%. Note that with combined fiber lens, overlap, and reflection losses of ~4 dB per facet, the expected intensity coupled into the waveguide was ~4100 kW/cm² for 130mW of incident pump power but no further pump power was available to push to saturation as would be obtained with optimized pump coupling. This is to the author’s knowledge the first time inversion in Erbium has been attained at these levels in a chalcogenide host, and indicates that sufficient inversion for high gain amplification can be achieved with Erbium. The saturation at levels below the theoretical maximum indicates that a small proportion of Erbium ions are still clustered, in energy contact, or are optically inactive in this host material as grown, consistent with the low power lifetime shortening compared to the intrinsic lifetime noted above. Further improvement in both deposition and/or more effective annealing are required to match the theoretical performance. To attain this however requires a change of the waveguide design to use a higher T_g strip glass, or the use of waveguide fabrication methods such as lift off [32] or hot embossing [33, 34]. A practical device also requires much lower passive waveguide losses (should be achievable with improved processing and film growth with appropriate compositions) and also the resolution of the photoinduced losses. It remains to be seen whether the photoinduced loss also occurs in films with the desired composition of Gallium doped Germanium Selenide, or whether a different composition with a higher bandgap will be more advantageous.

4. Conclusions

In this work, high quality Erbium doped Ge-Ga-Se films were deposited and rib waveguides based on these films were patterned. Significant signal enhancement at 1.5 µm was observed and 50% Erbium population inversion was obtained, with a saturation maximum of ~55% being possible, in waveguides with 1.5 × 10²⁰ Er ion/cm³. To the author’s knowledge this is the highest level of inversion ever demonstrated for Erbium in a chalcogenide host and is an important step towards future devices operating at 1550 nm and on the MIR transitions in Erbium. Photoinduced absorption loss caused by upconversion products in the waveguides is the remaining hurdle to achieving net gain. Further research is needed to find suitable compositions that possess high rare-earth solubility whilst avoiding the detrimental photoinduced losses.