Ce3+/Yb3+/Er3+ triply doped bismuth borosilicate glass: a potential fiber material for broadband near-infrared fiber amplifiers

Erbium doped bismuth borosilicate (BBS) glasses, possessing the broadest 1.55 μm near infrared (NIR) emission band among oxide glasses, stand out as excellent fiber material for optical fiber amplifiers. In this work, we demonstrate that both broadened and enhanced NIR emission of Er3+ can be obtained by sensibly combining the effects such as mixed glass former effect, phonon-assisted energy transfer (PAET) and de-excitation effect induced by codopant. Specially, by codoping CeO2 in a controlled manner, it leads to not only much improved optical quality of the glasses, enhanced NIR emission, but also significantly suppressed energy transfer up-conversion (ETU) luminescence which is detrimental to the NIR emission. Cerium incorporated in the glasses exists overwhelmingly as the trivalent oxidation state Ce3+ and its effects on the luminescence properties of Er3+ are discussed. Judd-Ofelt analysis is used to evaluate gain amplification of the glasses. The result indicates that Ce3+/Yb3+/Er3+ triply doped BBS glasses are promising candidate for erbium doped fiber amplifiers. The strategy described here can be readily extended to other rare-earth ions (REs) to improve the performance of REs doped fiber lasers and amplifiers.

Since Payne et al. first demonstrated the feasibility of erbium doped fiber amplifiers (EDFAs) working at 1.55 μ m near-infrared (NIR) wavelength 1 , a wide range of applications in telecommunications, medical surgery, eye-safe measurement and spectroscopy have been realized 2,3 . Nowadays, research is mainly oriented in two directions: either broaden the gain spectrum or enhance the output power of Er 3+ doped fibers, to meet the insatiable demand for optical data transmission capacity and high power lasers. However, the gain width of conventional SiO 2 -based EDFAs is restricted to 40 nm 4 , limiting the wavelengths that can be simultaneously amplified 5 . In contrast, Er 3+ doped fluoride, tellurite and especially bismuth glasses exhibit gain spectra wider than 50, 70 and 80 nm, respectively [6][7][8] , thus are more promising candidates for broadband EDFAs.
When it comes to lasing efficiency, although codoping with Yb 3+ has been routinely applied 9 , the quasi-threelevel scheme of Er 3+ still suffers from the problems limiting the NIR lasing efficiency, such as the population inversion bottle neck between the 1.55 μ m lasing level 4 I 13/2 and the ground level 4 I 15/2 , the undesirable energy transfer up-conversion (ETU) emission and excited state absorption (ESA). Recent studies have focused on methods speeding up the non-radiative transition (multi-phonon relaxation, MPL) between the 980 nm pumping level 4 I 11/2 and the lasing level 4 I 13/2 to increase the population of the latter. Since the multi-phonon relaxation rate is exponentially proportional to the maximum phonon energy of host materials 10 , utilizing comparatively high phonon energy (HPE) materials is a sensible choice, for example, conventional SiO 2 -based EDFAs have representative vibrational phonon energy 1100 cm −1 , higher than that of tellurite (700 cm −1 ) and fluoride (500 cm −1 ) fibers, it can provide gain amplification up to more than 1000 times easily. Borate crystals of even higher phonon energy (~1350 cm −1 ), such as Na 2 B 4 O 7 11 , Ca 4 YO(BO 3 ) 3 (YCOB) and YAl 3 (BO 3 ) 4 (YAB), have been demonstrated as excellent laser materials too 12,13 . The HPE also brings in an added merit that it effectively restrains the deleterious ETU emission and/or ESA.

Experiments
Glass samples of (40-x)Bi 2 O 3 -50B 2 O 3 -10SiO 2 -xCeO 2 -1Yb 2 O 3 -0.5Er 2 O 3 (x = 0, 0.1, 1, 3, 5, 10, 15, 20 in mol.%) were prepared by conventional melt-quenching method. The raw materials were weighed 20 g, mixed in an agate mortar for at least 15 min., stored in an alumina crucible and then put in an electric furnace. To prevent evaporation of Bi 2 O 3 , the materials were pre-sintered at 450 °C for 30 min., samples with varying content of CeO 2 were melted at different temperatures in a range from 950 to 1550 °C. The reason to increase the melting temperature is to dissolve more concentration of CeO 2 , otherwise the samples suffer from serious crystallization (see Results part below). All the obtained glasses were annealed around T g for 180 min. Finally, the glasses were cut and polished carefully to meet the requirements for optical measurements.
Absorption spectra were measured by a Perkin-Elmer Lambda 950 UV-VIS spectrophotometer over the spectral range of 300-2000 nm. Emission spectra were measured by an Edinburgh FS920 fluorescence spectrometer equipped with a liquid-nitrogen cooled steady state InSb detector and luminescence decay was recorded by a TDS3000C digital phosphor oscilloscope. Sample density was obtained by Archimedes method with alcohol as the immersion liquid. Er 3+ concentration was calculated from the measured density and initial composition. The X-ray photoelectron spectra spectroscopic (XPS) was measured by ESCALAB250Xi (Thermo Fisher U.S.), the glass samples were broken in the XPS chamber and only those newly created cross-sections were measured. Raman spectra were measured by RenishawInvia Raman microscope (Renishaw, Gloucestershire, UK) with the excitation wavelength of 515 nm. All the measurements were carried out at the room temperature. Figure 1 shows coloration of the glasses with varying content of CeO 2 and melted at different temperatures. Depending on the content of CeO 2 and melting temperature (MT), the figure can be divided into four parts. At high MT (> 1200 °C) and low CeO 2 content (< 5 mol.%) region (b), the glasses suffer from serious darkening/ devitrification and cannot be obtained due to the evaporation of Bi 2 O 3 ; at high MT and CeO 2 content region (c), the addition of CeO 2 markedly improves the transparency and avoids the devitrification of the glasses. Obviously, the solubility of CeO 2 increases with MT such that up to 20 mol.% CeO 2 can be incorporated into the glasses. At low MT and CeO 2 content region (a), on the other hand, the addition of CeO 2 increases the absorption of the glasses in the visible region. Finally, glasses cannot be prepared at low MT and with high CeO 2 content (d) due to severe crystallization resulted from the undissolved CeO 2 .

Results
The XRD patterns of the samples sitting on the border separating those that can form glass and those cannot were measured (cf. Fig. S1). The amorphous nature of the glasses was confirmed. DSC curves of some representative samples (showing the best NIR emission properties) were also measured (cf. Fig. S2 and Table S1). The Δ T = T x − T g (T x , onset temperature of crystallization and T g , glass transition temperature) values were considerably larger than 100 °C, indicating excellent thermal stability and fiber drawing potential. On the other hand, as is known, traditional bismuth glasses have relatively low melting (< 1000 °C) and glass transition temperatures (< 450 °C), the glasses volatilize and the optical properties worsen terribly when melted or handled at high temperatures, all of which pose a serious barrier to fiber drawing and fusion with conventional silicate optical fibers 8 . As shown in Fig. 1, addition of a moderate amount of CeO 2 (e.g., 3 mol.%) makes excellent glasses even when melted at temperature as high as 1350 °C. And the glass transition temperature can be significantly increased up to 638 °C for the glass containing CeO 2 up to 20 mol.% (Table S1). Therefore, the advantage to increase the melting and glass transition temperatures is to mitigate the difficult in the fiber drawing and fusion with conventional silicate fibers.
Absorption spectra of all the studied glasses were measured (cf. Fig. S3). Here, we show only spectra of the samples enclosed by two rectangles in Fig. 1. At low MT (e.g., 1050 °C), the addition of CeO 2 increases the absorption in the visible region such that the cut-off edge of the glasses undergoes a red-shift to longer wavelength ( Fig. 2(a)). At high MT (e.g., 1250 °C), the transparency of the glasses improves markedly upon the addition of CeO 2 , while the cut-off edge again shifts to longer wavelength with CeO 2 ( Fig. 2(b)).
The NIR emission spectra of all the studied glasses were measured (cf. Fig. S4), in all cases a very broad emission band with the full width at half maximum (FWHM) about 80 nm (Table 1), covering S (1460~1530 nm), C (1530~1565 nm) and L (1565~1625 nm) silica fiber communication bands, can be clearly observed. Codoping of CeO 2 resulted in the desired variations of the emission spectra. On one hand, both the intensity and bandwidth of the emission increase upon the finite codoping of CeO 2 , for example, no more than 1 mol.% for the glasses melted at MT lower than 1050 °C. Further increase in CeO 2 , however, exerts an adverse effect on the NIR emission intensity ( Fig. 3(a)). We would like to stress that even though the enhancement of the NIR emission shown in Fig. 3(a) is not that breathtaking, the result is quite reliable and reproducible, because we have tested different batches of the glasses with the same designed composition and melted under approximately the same condition, the similar enhancement was observed in every run. The CeO 2 induced enhancement is more evident at high MT, for example, a two-fold increase in the NIR emission intensity was observed for the glasses melted at 1150 °C (cf. Fig. S4(c)). A thorough study also show that the optimal doping concentration of CeO 2 leading to the largest enhancement is MT dependent, and increases with MT ( Fig. 3(d)). The variation of the NIR emission lifetime with the content of CeO 2 is shown in Fig. 3(c), it decreases with CeO 2 very mildly.
On the other hand, the ETU emission almost disappears completely upon even slight co-doping of CeO 2 , and decreases gradually with increasing content of CeO 2 (insert in Fig. 3(b)). It is noted that the red 660 nm ETU emission is unusually stronger than that at the green 532 nm wavelength which will be discussed later on.  We also studied the MT dependence of the NIR and ETU emission spectra of the glasses. Fixing the doping concentration of CeO 2 at 1 mol.%, both the NIR and ETU emissions decrease with increasing of MT ( Fig. 4(a,b)), similar to our previous work 22 . However, the full width at half maximum (FWHM) of the NIR emission band remains to be 80 nm (Table 1). On the other hand, the lifetime of the NIR emission varies arbitrarily with MT ( Fig. 4(c)).
Since the NIR emission is subject to phonon energy of glasses too, the distribution of phonon energies (lattice vibrations) of the studied glasses was examined by Raman scattering experiment, and the results are shown in

Discussion
In the present work, the effects of CeO 2 are two-fold. Firstly, it works as an effective oxidant to against the darkening/crystallization of the BBS glasses due to the precipitation of Bi metal or clusters during glass melting 22 . The above processes can be expressed by the following equations: high melting temperature 2 high melting temperature It is the release of oxygen gas (Eq. 2) at high MT that oxidizes the Bi metal or clusters (Eq. 3) and improves the transparency of the BBS glasses ( Fig. 1) 25 . According to Eq. 2, CeO 2 itself goes into Ce 2 O 3 during the melting, the consumption of oxygen gas by the Bi metal or clusters favors the formation of Ce 3+ ions, especially at high MT. The XPS measurement clearly backs up this assumption (Fig. 6). Although both Ce 3+ and Ce 4+ can co-exist in glasses, the XPS spectra show that trivalent Ce 3+ dominates in the studied BBS glasses, as confirmed by the resemblance of the XPS spectra to that of the standard reference CeF 3 crystal and the absence of the characteristic doublet related to Ce 4+ at ~916 eV 26,27 . However, either the presence of Ce 3+ or Ce 4+ can increase the absorption in the visible region up to 500 nm (Fig. 2), which is induced by either the 4f-5d transitions of Ce 3+ or the charge transfer between O 2− and Ce 4+ . Secondly, the role of codoping CeO 2 is to improve the NIR luminescent properties of Er 3+ . The variation of the NIR emission with the content of Ce 3+ is complex; however it can be analyzed based on the phonon assisted energy transfer (PAET) between the relevant electronic energy levels of Er 3+ and Ce 3+ ions. The PAET may involve one, two or more than two phonons depending on the phonon energy of the host, concentration of and critical radii between rare earth ions (REs) in solids 28 . Generally, one phonon assisted process accounts for 20~30%, while two and more than two phonons assisted processes only 10 −6~1 0 −3 of the total transition  probability of REs. Referring to the energy level diagrams of Er 3+ and Ce 3+ (Fig. 7), it is seen that the following two PAET processes which may influence the NIR emission: (a) Er 3+ : 4 I 11/2 + Ce 3+ : 2 F 5/2 → Er 3+ : 4 I 13/2 + Ce 3+ : 2 F 7/2 , and (b) Er 3+ : 4 I 13/2 + Ce 3+ : 2 F 5/2 → Er 3+ : 4 I 15/2 + Ce 3+ : 2 F 7/2 have energy mismatches approximately 1400 cm −1 and 4300 cm −1 , respectively. The maximum phonon energy of the glasses is 1350 cm −1 (Fig. 5), which means that the former process (a) leading to the increased population of the NIR lasing level 4 I 13/2 of Er 3+ and enhanced NIR emission requires one phonon, while the latter (b) more than two phonons. Although the latter may quench the NIR emission, it hardly plays a role for low doping concentration, because the PAET involving more than two phonons generally has extremely low efficiency. In fact, the luminescence decay measurement (cf. Table S2) shows that the lifetime of the NIR luminescence varies by only 5% when the content of Ce 3+ is low (< 1 mol.%).
However, at too high concentration of Ce 3+ (e.g., > 5 mol.%), one phonon assisted process may also occurs between one Er 3+ ion and a pair of Ce 3+ ions (via Er 3+ : 4 I 13/2 + Ce 3+ : 2 F 5/2 + Ce 3+ : 2 F 5/2 → Er 3+ : 4 I 15/2 + Ce 3+ : 2 F 7/2 + Ce 3+ : 2 F 7/2 ). The energy transfer efficiency (ƞ) between Er 3+ and Ce 3+ can be calculated according to the equation: ƞ = 1 − τ/τ 0 , where τ and τ 0 are lifetimes of the donor (Er 3+ ) with and without acceptor (Ce 3+ ). A rough estimate of ƞ shows that it equals to 30~60% for high doping concentration of Ce 3+ , suggesting that one phonon assisted process indeed occurs. A more rigorous calculation of the PAET efficiency requires knowledge of the absorption cross-section of the Ce 3+ 2 F 7/2 → 2 F 7/2 transition (~2200 cm −1 ), which is impossible to obtain because the intrinsic phonon cut-off edge of the samples lies at shorter wavelength (~3000 cm −1 ). Incidentally, the PAET has also been  found a very successful means to improve the lasing efficiency of Yb 3+ /Ho 3+ and Yb 3+ /Tm 3+ pairs in solids 29 . We have failed to measure the lifetimes of 4 I 11/2 level of Er 3+ before and after codoping of CeO 2 because the 1000 nm and 2.7 μ m emissions from 4 I 11/2 level were too weak to be recorded, thus we are not able to deduce the PAET efficiency between Er 3+ and Ce 3+ .
One may have noticed that the lifetime of the 1.55 μ m emission of Er 3+ was rather short (~1 ms) in the studied glasses ( Fig. 3(c)). This is because the glasses contain a large amount of B 2 O 3 as high as 50 mol.%, giving rise to much faster multi-phonon relaxation between the lasing level 4 I 13/2 and the ground level 4 I 15/2 as compared to glasses containing a less amount of B 2 O 3 (~3 ms) 30 . In this respect, it seems preferable to use glasses containing a limited amount of B 2 O 3 , however, when referring to our previous work 22 and other published papers 30 , it appears that the broadband NIR emission with a FWHM over 80 nm can only be obtained in glasses containing relatively high content of B 2 O 3 8 . Since the emission bandwidth of rare-earth (RE) ions is largely determined by the variations of ligand fields around RE ions from site to site (inhomogeneous broadening), and the ligand fields around RE ions are mainly influenced by glass formers, consequently, the broadband NIR emission observed in the BBS glasses can be partially accounted for by the so-called "mixed glass former effect" due to the presence of a multiplicity of glass former motifs (Fig. 5) 31 . However, a compromise indeed exists between the emission lifetime and the bandwidth regarding the optimized concentration of B 2 O 3 . Fortunately, because Bi 2 O 3 -B 2 O 3 -SiO 2 ternary system has a very broad glass forming region, providing a rich tuning range of glass compositions, the lasing and other relevant physical/ optical properties are amenable to being tailored for specific requirements. Work is now under way to further refine the lasing properties of the glasses and to understand the line broadening mechanism induced by B 2 O 3 .
Previous report showed that when the content of B 2 O 3 increased to 8 mol.% in Er 3+ /Yb 3+ codoped tungsten-tellurite glasses, the red ETU emission was already getting stronger than the green one 32 . This phenomenon is also owing to the presence of HPE compound B 2 O 3 , which tends to decrease the lifetime of 4 I 11/2 level and as such increase the population of 4 I 13/2 level, consequently, the ETU emission involving the intermediate 4 I 11/2 level, viz., the green one, was reduced, whereas the red one involving the intermediate 4 I 13/2 level enhanced. As mentioned above, the studied glasses contain a high concentration of B 2 O 3 , it is no wonder that the red ETU is stronger than that of the green one ( Fig. 3(b)).
Judd Ofelt (JO) analysis is the most useful tool in estimating the forced electric dipole transitions of RE ions. JO parameters can be derived from the absorption spectra and the results are shown in Table 1. According to the theory, three JO intensity parameters (Ω t , t = 2, 4, 6), which depend on the local chemical environments, determine the radiative transition rate of RE ions. Of the three parameters, Ω 2 is the most sensitive to local structure, and previous studies have suggested that it increases with the degree of asymmetry around RE ions 33,34 . The other two parameters, Ω 4 and Ω 6 are not so much susceptible to the environmental changes as Ω 2 . However, Ω 6 has been found to increase with the extent of the overlap between the 4f and 5d orbitals of Er 3+ ions, or instead, it decreases as the 5d electron density diminishes. The substitution of CeO 2 for Bi 2 O 3 reduces the number of highly polarizable Bi-O bonds (Fig. 5), and according to our previous study 22 , this may in turn decrease to some extent the microsymmetries around Er 3+ ions, which partly accounts for the increased Ω 2 .
Gain coefficient of the NIR emission, G(λ), was calculated according to λ where P is the population inversion given by the ratio between the population of Er 3+ ions at the lasing level ( 4 I 13/2 ) and total concentration ( , where c is the speed of light, n is the refractive index and A rad is the radiative transition rate obtained from JO analysis, Δ λ eff was calculated by Positive gain already appears when only twenty percent of Er 3+ ions are populated to the lasing level (insert in Fig. 8(a)). The maximum gain increases with the concentration of CeO 2 . It should be mentioned that the gain may also depend on the ESA of Er 3+ as in the case of Er 3+ -doped silica fibers 35,36 , so in reality, the calculated G(λ)-values can suffer from overestimating. On the other hand, with the fixed doping concentration of CeO 2 , the gain coefficient barely changes with MT ( Fig. 8(b)).The sudden drop in the gain value is due to the crystallization of the samples melted at high MT.

Conclusion
Depending on the melting temperature, a large amount of CeO 2 (up to 20 mol.%) can be doped into the bismuth borosilicate (BBS) glasses, and cerium ions present as trivalent oxidation state Ce 3+ overwhelmingly. The benefits of the doping of CeO 2 include, on one hand, improving significantly the transparency of the glasses especially for those melted at high temperatures; on the other hand, enhancing the 1.55 μ m near-infrared (NIR) emission without sacrificing the lifetime due to the phonon-assisted energy transfer (PAET) between Er 3+ and Ce 3+ . However, the doping concentration must be limited because otherwise both the intensity and lifetime of the NIR emission degrade. In addition, the undesirable energy transfer up-conversion (ETU) emission and excited state absorption of Er 3+ are severely quenched by doping of CeO 2 stemmed from the accelerated non-radiative transition 4 I 11/2 → 4 I 13/2 of Er 3+ . The unique combination of enhanced NIR emission intensity, broadened NIR emission bandwidth and suppressed ETU emission makes Ce 3+ /Yb 3+ /Er 3+ triply doped BBS glass a potential fiber material for broadband EDFAs. And because Bi 2 O 3 -B 2 O 3 -SiO 2 system has a very broad glass forming region, providing a rich compositional tuning range, the lasing and other related physical/optical properties can be tailored for specific requirements. The strategy described here, namely, de-excitation by PAET between rare-earth (RE) ions, increasing non-radiative transition by using high phonon energy compounds and line broadening through mixed glass former effect, can be readily extended to other RE pairs such as Yb 3+ -Tm 3+ and Yb 3+ -Ho 3+ to enhance the 2.0 μ m emissions of Tm 3+ and Ho 3+ doped solids. More work has to be done on optimizing the concentration of B 2 O 3 and exploiting the physical basis for the line broadening induced by B 2 O 3 .