Prospects and challenges for all-optical thermal management of fiber lasers

It is hard to overstate the utility of lasers in modern technology. Optical-fiber-based lasers are of particular value thanks to their combination of small form factors, afforded by the coilability of the thin strands of fiber, and high beam-quality output. The optical fiber geometry also possesses a relatively high surface-area-to-volume ratio, rendering thermal management somewhat more straightforward than in other bulk laser types. Regardless, the generation of heat during the lasing process can still be problematic for a myriad of reasons, and conventional methods of thermal management do not comport with the potential compactness and elegance of fiber lasers as technological solutions. This Perspective summarizes recent advances in glass science and optical fiber engineering to support the provocative premise that heat generation in future laser systems can be entirely managed by a combination of fiber materials and novel laser physics. Letting the fiber manage heat itself would have significant impacts on enhancing system performance while greatly reducing size, weight, power-consumption, and cost.


Introduction
The commercial ubiquity of lasers may lead one to believe that lasers are fully mature technologies with little new physics to unveil.However, the practical need to reduce the size, weight, and power consumption (SWaP) of laser systems has led to a renaissance in the exploration and development of new materials and exciting laser physics.
Optical-fiber-based lasers have arguably led the way in this reawakening.Fibers offer long gain lengths over which the propagating mode is strongly confined, which promotes high output powers and excellent beam quality.Further, fibers can be coiled, thus affording more compact packaging than bulk solid-state laser systems.Like all current commercial lasers though, fiber lasers also generate heat, which is especially problematic at high powers.As discussed in greater detail below, this heat must be managed for any system to be practical, which often confounds efforts to reduce SWaP.

Sources of heating in fiber amplifiers and lasers
There are numerous sources of heat in an active fiber laser.It is informative to organize these into two general categories: intrinsic and extrinsic.There are two principal intrinsic sources of heating in active systems: (i) the quantum defect (QD) and (ii) non-radiative relaxations.Both are omnipresent and may or may not be wholly negated by choice of fiber material composition or processing, nor by fiber or laser design.
The QD is defined as the energy difference between the excitation source (i.e. the pump) and the laser emission (i.e. the signal), energy that is generally released in the form of heat during lasing or amplification.Recast in terms of wavelengths, QD = 1−(λ p /λ s ), where λ p and λ s are the pump and signal wavelengths, respectively.QD heating is an unavoidable price for optical gain.For the most common fiber laser dopants and transitions, the QD can range from a few percent (e.g.5%-8% for Yb 3+ ) to nearly 60% (e.g.Er 3+ ) [1].
Non-radiative relaxation arises when electrons in an excited state relax to a lower lying energy level through the production of phonons, or heat.When the energy spacing between two energy levels is less than around a few phonon energies in the host material, non-radiative relaxation has a higher probability than spontaneous emission.Such rapid non-radiative transitions can be desirable, for example in erbium-doped fiber amplifiers pumped at 980 nm where they quickly populate the metastable level ( 4 I 13/2 in the case of erbium).However, like in the case of QD the price to pay for non-radiative relaxation is undesirable heating of the gain medium.For completeness, intrinsic optical scattering (e.g.Rayleigh) is not, per se, a source of heat generation unless the scattered light is subsequently absorbed by, for example, the fiber coatings, and thus contributes indirectly to heating the fiber.
More deleterious forms of non-radiative relaxation are those that prevent the production of photons from what would otherwise be excited metastable states.There are several sources for non-radiative relaxation, including, most importantly, multiphonon relaxation and concentration quenching.The former arises from coupling between the excited-state electrons and host phonons, which stimulates their non-radiative relaxation along with the creation of phonons.Concentration quenching is a mechanism by which the energy of an excited-state electron is transferred to either an impurity, another dopant ion in the ground state, or another dopant ion in the excited state, depending on the ion and the transition, with the subsequent creation of phonons via non-radiative relaxation [2].The probability of this process increases rapidly when the dopant ions are clustered together, which facilitates energy transfer between adjacent ions.Clustering is a thermodynamic phenomenon strongly dependent upon the glass fabrication process [3], hence it is an intrinsic property for a given dopant/host combination.
Extrinsic sources of heat generation may be reduced through careful experimental controls.As relates to the active optical fiber itself, impurities that either absorb pump or signal light, or that facilitate quenching, are of primary concern.For silica-based fibers fabricated using commercial chemical vapor deposition (CVD) methods, absorbing impurity concentrations are essentially non-existent given the self-purifying nature of the underlying chemistry [4].However, hydroxyl (OH) can remain if the glass is not suitably dried.Such species, in the form of Si-OH linkages, exhibit large vibrational energies, larger than those of silica [5], and can promote non-radiative relaxation of dopant excited states by multiphonon processes [6].They might also indirectly, via mechanisms not yet elucidated, promote or enhance quenching [7].In glasses not fabricated using CVD processes, residual impurities from the glass precursors, particularly transition-metal impurities, can increase background losses and therefore exacerbate heat generation.
More generally, however, the quantum efficiency (QE) usually lumps both intrinsic and extrinsic sources of non-radiative relaxation into a single parameter.The QE is defined here to be the proportion of absorbed pump photons that contribute to radiative relaxation.To illustrate the importance of thermal management in high power fiber laser systems, a simple example is presented.Consider a fiber laser at 1061 nm cladding-pumped with 10 kW at 976 nm, or a QD of 8%.The fiber is assumed to be a generalized commercial fiber with a core/cladding/coating diameters of 25/400/600 µm, and a quantum (radiative) efficiency of 98%.While these values were selected primarily for illustrative purposes, it should be noted that the coating diameter can vary among manufacturers and be somewhat smaller than 600 µm.The QE is arbitrarily selected but represents a very good value in that it is sufficiently large to enable cooling by anti-Stokes fluorescence (ASF).More details are provided later in the paper.The total thermal power due to the QD is 800 W, while the QE-induced heating is 200 W.If the thermal load is taken to be uniform over a 12 m fiber length, it is equal to 83.3 W m −1 .In the absence of any active cooling (i.e. the fiber is in still air at a room temperature of 293 K), the steady-state fiber temperature is governed by comparable convective and radiative heat transfer coefficients (roughly 45 W m −2 K −1 each) [8], leading to an unacceptably large core temperature of around 835 K (562 • C).Let us now assume that the fiber undergoes forced convective cooling with water (total convective cooling coefficient of 1000 W m −2 K −1 ). Figure 1 plots the simulated radial temperature profile of the fiber under this new cooling condition (solid curve).The change in core temperature drops by a factor of around 5.5, to around 100 K.This is still a significant temperature change that requires further thermal management.Figure 1 also shows (dashed curves) the reduction in fiber temperature profile when halving the non-radiative contribution only, halving the QD only, and halving both the QD and non-radiative components together.In this case, the greatest improvement comes from the reduction in QD.Note that changing the coating thickness by a few tens of micrometers has a largely negligible impact on these results.
There are other sources of extrinsic heat generation.The first of these are optical losses relating to how pump light is coupled into the doped fiber.For example, poor fusion splices or beam combining can lead to the coupling of pump light into the buffer (or coating) where it is at least partially absorbed.Similar issues arise for the signal power within the fiber-laser system.In addition, the use of a cladding mode stripper [9] near the laser output can lead to excess local heating.Finally, both spontaneous emission and amplified spontaneous emission (ASE) can be sources of heating.Since the wavelengths of the mean spontaneous emission and ASE are usually longer than that of the pump, they both usually yield a positive QD.The temperature dependent mean fluorescence wavelength is defined to be λ F = ´∞ −∞ λI (λ, T) dλ/ ´∞ −∞ I (λ, T) dλ where I (λ, T) is the spontaneous emission spectrum.Additionally, spontaneous emission may be absorbed by the coating material.It should be noted, however, that under normal operating conditions, fiber lasers produce little spontaneous emission.A generalized power budget for a Yb-doped fiber laser is shown in figure 2. As stated earlier, the QD sets the highest theoretical slope efficiency of a Yb-doped laser or amplifier to λ p /λ s .A nominal QD of 7% is shown in the figure, but it will vary, depending on the choice of pump and signal wavelengths.The second most significant source of power loss might be found in the coupling of pump light from the semiconductor(s) to the pigtail fiber in the module, and the subsequent delivery of this power through a pump combiner to the active fiber.Such combiners are typically quoted as having around 95% combining efficiencies (which can vary, depending on the configuration).Similarly, there may be sources of insertion or other losses to the signal, and the QE of the active fiber will be less than unity (it is assumed to be 97% in figure 2).Finally, the laser may produce some nominal spontaneous emission.The values selected here are mainly for illustrative purposes and can vary upward or downward depending on the quality of the individual elements comprising the system.

Detriments of heating in fiber amplifiers and lasers
The impacts of heat generation on the properties and performance of a fiber amplifier or laser are many fold and range from parasitic to catastrophic.The more consequential detriments include (i) adverse effects on the active ion spectroscopy (absorption and emission cross-sections as well as non-radiative relaxation rates) [10], (ii) degradation or failure of the protective plastic coatings on the fiber [11], (iii) thermal lensing [12] and/or transverse mode instability (TMI) [13], and (iv) excess frequency (or phase) and intensity noise [14].
Proper thermal management or, ideally, minimizing the heat generated during the lasing process (the topic of this Perspective), potentially permits reduced thermal noise and thermal nonlinearities (e.g.thermal lensing and TMI), and it opens the door to ultra-stable lasers, spot-cooling for micro-electronics, and power-scalable high-energy laser systems with significantly reduced SWaP demands.

Conventional methods of heat mitigation
It is important to point out that changes in operating temperature are not only important in high-power systems, but also in precision, low-power laser sources for sensing applications.For example, the fundamental lower limit to the thermodynamic temperature variance in a system (due to entropy, or movement of heat) is given by ⟨∆T 2 ⟩ = k B T 2 /ρVC i [15], where k B is Boltzmann's constant, T is the temperature, ρ is the mass density, V is the volume, and C i is the specific heat.Through thermal expansion and the thermo-optic effect, this temperature variance leads to both phase and amplitude fluctuations in the laser output [16].To avoid increases in temperature when in operation, the laser must therefore be actively cooled.The most common forced convective methods include heat exchangers that rely on flowing either air or liquid (such as water) around the gain medium.Each active cooling method requires excess mechanical movement in the laser system, which can fail over time.Be this from flowing water or air or small rotating fans, this introduces mechanical vibrations that can impact the refractive index, and therefore laser phase, through the photoelastic effect.Therefore, vibration-free cooling would be preferred for many applications [17].As a point of reference, the convective heat transfer coefficients are typically around h = 100 W (m 2 K) −1 [18] for flowing air and over 1000 W (m 2 K) −1 for the flowing water [19,20].
Other temperature control methods include the use of thermo-electric coolers [21].They work well and do not induce vibrations, but they can only extract so much power per unit volume, their energy efficiency is relatively low, and unless used in special geometries not always compatible with the form factor of optical fibers or chip-scale waveguides, they can produce asymmetric cooling.Any asymmetry induces temperature gradients in the gain medium that in turn can result in a range of deleterious effects, including mode-shape distortion through the thermo-optic effect at low power, as has been observed in bulk lasers [22,23], and catastrophic failure at high power.Therefore, careful attention should be paid to the thermal design when incorporating these cooling elements.

Routes to in-fiber thermal management
As noted, this Perspective aims to provoke new thinking regarding all-glass thermal management in future laser systems.This philosophy stems from recent efforts by the authors to mitigate internal heating of fiber lasers in general, and parasitic optical nonlinearities stemming from internal heating in particular, by judicious selection of the glass composition of the active fiber [1,24,25].

Material considerations
The first design principle relates to the properties of the glass comprising the active fiber.As mentioned, heat generation during lasing in commodity silica fibers primarily arises from the QD and non-radiative relaxations.While the spectroscopic properties of active dopants do not change as much from host to host in glasses compared to crystals, they do change enough to be meaningful to this discussion.For example, although the QD for Yb 3+ lasing at a wavelength of about 1 µm is between 5% and 8%, Yu et al reported a Yb-doped laser with a QD less than 1% [26].The active fiber was one in which the concentration of fluorine in the core glass was large enough to result in a significantly narrowed emission spectrum and a shorter average emission wavelength relative to other common laser glasses, such as aluminosilicates.This phenomenon is often attributed to the nephelauxetic effect, in this case to the decreased degree of covalency in the vicinity of the rare-earth ion [27].Since gain is favored at shorter wavelengths, the pump and signal wavelengths could be brought close enough together that a roughly five-fold reduction in QD is realized.
The significance of non-radiative relaxations to heat generation in a fiber laser depends on the phonon spectrum of the host glass as well as the energy level structure of the active dopant.As described above, when the energetic difference between dopant excited states is small in comparison to the maximum phonon energy of the host, electron-phonon coupling can yield non-radiative de-excitation of the excited-state electrons, creating host phonons, which manifests itself as heat.Since the choice of active dopant is generally dictated by the application, heat generation by multiphonon relaxation is primarily defined by the properties of the host.Indeed, there are glasses with very low phonon energies, such that radiative transitions between dopant excited states have an innately higher efficiency.For the purposes of this Perspective, silica fibers are assumed, given their remarkably high purity, resulting in near-intrinsically low attenuation, high mechanical strength, and industrial acceptance.Accordingly, if both the active dopant and the glass (i.e.silica) are defined, one might presume no substantive means to tailor spectroscopic or phononic properties are available.Fortunately, this is not entirely the case.There has been a significant interest in nanoparticle (NP)-doping of silica glass to sequester the active dopant inside a dielectric environment different from that of the principal glass host [28,29].The NPs are then used to tailor the spectroscopic properties of the dopant without degrading the mechanical and other useful properties of the host (silica).Such NP-doped silica fibers have been reported to modify absorption and emission spectra, radiative lifetimes, and quantum efficiencies, with negligible changes in background attenuation when carefully fabricated.
Put another way, silica-based fibers have been fabricated using conventional methods that exhibit markedly reduced QDs and enhanced quantum yields, while also exhibiting reduced optical nonlinearities.With this existing material foundation, opportunities for internal cooling based on laser physics and design can further complement the mitigation of heat and loosen the demands for thermal management.

Laser cooling via ASF
First proposed in 1929 by Pringsheim [30], anti-Stokes fluorescence pumping (ASP) refers to optically pumping a laser material at a wavelength longer than its mean fluorescence wavelength-as opposed to shorter than the mean fluorescence wavelength in conventional Stokes pumping.As illustrated in figure 3(a), lower energy pump photons excite ground-state electrons into the lower levels (less energetic) of the laser material's upper manifold.To satisfy Boltzmann's distribution, these electrons must redistribute themselves among the upper (more energetic) levels of the upper manifold.They do this by acquiring energy from the phonon bath of the material (phonon annihilation).When these more energetic electrons relax back to the ground state, they emit so-called ASF photons that have a higher mean energy than the pump photons.These ASF photons escape the gain medium, taking away the added energy they acquired from the phonon bath.The net energetic budget of this process is negative for the laser host, and it cools.
An experimental demonstration of ASP was first reported in 1995 by Epstein and co-workers in a free-space sample of Yb-doped fluoride glass (ZBLANP) that was pumped at 1015 nm with 700 mW and cooled by 0.7 K from room temperature [31].Four years later, Bowman applied the same principle to produce the first laser with no net internal heat generation [32], a mode of operation now broadly referred to as radiation balanced (figure 3(b)), or less commonly as athermal.Bowman's radiation-balanced laser was a solid-state single-crystal Yb:YAG; at an output of 81 W the heat load on the laser rod was still slightly negative.Since then, many hosts and trivalent rare-earth ions have been successfully cooled via ASP.Most demonstrations were performed in free-space samples [31][32][33][34] to increase the volume of the pumped region, which increases the total number of laser ions that participate in the cooling process and enables greater cooling.Most samples were placed in a vacuum chamber to minimize convective heat transfer from the warmer surrounding air into the sample, and hence achieve greater negative temperature changes.Samples have also been mounted in clamshells to reduce radiative heating [33,34].The lowest absolute temperature reached with ASP cooling to date is 91 K [34].It was achieved in a highly purified 4 × 4 × 12 mm YLiF 4 crystal doped with 10% Yb 3+ (probably atomic percents) and pumped in 11 passes with 54 W at 1020 nm.
Applying this process to cool an optical fiber, or a fiber laser or amplifier, leads to smaller temperature changes for several reasons.First, the density of active ions that can be doped into the glass comprising the fiber is generally lower than in crystalline laser hosts.In other words, for the same volume of doped material, the absolute number of laser ions performing the cooling is smaller.Second, in a fiber only the core is typically doped with a laser ion, and consequently the core has to cool not only itself but also the inert cladding.Third, for practical reasons it is much preferable to cool a fiber placed in air rather than in a vacuum, which means that the fiber gets a continual input of heat from the warmer surrounding air.These limitations are particularly noticeable in silica fibers, which cannot be as heavily doped with rare-earth ions as fluoride fibers for example, and in single-mode fibers, which have a smaller core than multimode fibers.
The material requirements for cooling a fiber can be understood and quantified simply by returning to the basic physics of ASF cooling and modeling the heat that can be extracted from a fiber by this process.Intuitively, every time a laser ion excited by an anti-Stokes photon relaxes radiatively to the ground state, it emits a photon with a mean energy equal h < ν f >, where h is Planck's constant and <ν f > is the mean fluorescence frequency of the laser transition.This emission happens at a rate of 1/τ rad , where τ rad is the radiative lifetime of the transition, assumed purely radiative.The number of ions per unit volume that emit is the excited-state population density N 2 in the pumped volume of laser material.It then follows that the amount of heat extracted power per unit time and volume, dQ(z)/dt, from the material is given by the very simple expression: The net extracted heat is equal to this (negative) quantity plus the (positive) power per unit volume that is injected in the same volume by the pump that creates this excited-state population.When applied to the most common cooling ion, which is Yb 3+ , and to an optical fiber, this analysis leads to the following expression for the maximum heat extracted per unit volume assuming no absorptive impurities (or absorptive loss) in the fiber core [35]: ( dQ dt This result is intuitive.The extracted heat is proportional to (1) the doped area A f = πa 2 where a is the doped core radius (the larger the doped core, the more ions are involved in the cooling process), (2) the concentration of Yb ions N 0 (same explanation), (3) the reciprocal of the radiative lifetime (the faster the ions relax radiatively the more heat they extract per unit time), and (4) the ratio of emission cross section σ p e to absorption cross section σ p a of the Yb ions at the pump wavelength (the higher the probability of emission, the less excited-state population will be created by the pump, and the lower the cooling).Importantly, the extracted heat is proportional to the difference in the front term, which quantifies the balance between the pump energy (which injects energy into the fiber) and the ASF mean energy (which extracts energy from it).In this term, τ is the total upper-manifold lifetime given by τ (N 0 ) = 1/(τ rad −1 + τ nrad −1 + τ q (N 0 ) −1 ), where τ nrad is the nonradiative lifetime and τ q the concentration-dependent quenching lifetime [36] N c is the critical quenching concentration.In the absence of quenching (N 0 ≪ N c ), assuming negligible nonradiative contribution as is the case for Yb 3+ in silica and fluoride, from equation (3) τ q is infinite, τ = τ rad , and the front term in equation (2) simplifies to the difference between the pump photon energy and the mean fluorescence photon energy, which is negative: the material cools.If the Yb concentration is increased, τ becomes faster (smaller) than τ rad , the first term (pump) in the bracket increases, and the bracket term is less negative: the material cools less.If the first term increases too much (too much quenching), it exceeds the second term, and the bracket becomes positive: the material heats up instead of cooling.So, on one hand it is desirable to increase N 0 to increase the last term in equation ( 2) and increase cooling; on the other hand, N 0 cannot be too large, or the first bracket becomes positive, and cooling is no longer possible.Because of these two opposing trends, there is an optimum Yb concentration that maximizes cooling [35].This optimum is typically approximately around 10% of N c , depending on the ratio of pump energy to mean fluorescence photon energy.This analysis points to two of the most critical requirements for efficient cooling, which is that the level of quenching must be minimized (higher possible N c ), and that the Yb concentration be optimized for that level of quenching.The superior cooling performance of rare-earth-doped crystals so far [34] can be attributed almost entirely to their greater resistance to quenching (higher N c ) and therefore higher optimized rare-earth concentration.
The second key requirement is that the absorptive loss due to core impurities, in particular OH and transition metals, must be very low.A good estimate of the additional heat load caused by residual impurity absorption can be obtained with a simple model.If the power absorbed per unit length in a fiber is P abs /L, and the  [39,40,42], where they were conducted in bulk glass preforms.The asterisk ( * ) denotes data reported here for the first time.
fiber is passively cooled by convection of still air, then its temperature will rise by ∼31.3P abs /L, in Kelvin [37].An absorption coefficient α ba will then increase the temperature of a fiber by ∼31.3α ba P, where P is the pump power in the fiber.At the power that gives maximum cooling in a few-mode fiber, which is around 200-500 mW, the temperature rise is then 1.4-3.6 mK per dB km −1 of residual absorption.In a fiber with 10 dB km −1 of background absorption, broadly speaking between a quarter to half of the heat removed by ASF is negated by this spurious source of internal heat 4 .A Yb-doped silica fiber with state-of-the-art level of quenching will not cool well, or not cool at all, if its residual absorption exceeds approximately 30 dB km −1 .Interestingly, the solid that was cooled to the lowest temperature to date, the single crystal of Yb:YLiF 4 mentioned earlier, had an absorptive loss of 43 dB km −1 [34], which is actually much larger than the lowest value reported in a silica fiber (less than 5 dB km −1 [7]).This superior performance of silica fibers is due to the extremely low levels of impurity that can be achieved with the modified vapor deposition used to fabricate silica fibers.
Figure 4 plots the critical quenching concentration values that have been reported to date for silica and silicate fibers and preforms (blue circles) [7,[38][39][40][41], and for ZBLAN fibers (red diamonds) [42][43][44][45] that have been successfully cooled by ASP.This data is presented chronologically to illustrate recent progress made in the realm of silica.The solubility of ytterbium fluoride in ZBLAN being much higher than that of ytterbium oxide in silica, at the start of silica cooling in 2019 ZBLAN fibers had critical quenching concentrations around 3-4 × 10 27 m −3 , which is quite high, although not as high as the best Yb-doped crystals.The best 4 To be accurate, a calculation of the penalty of absorptive loss on cooling requires a full-blown model of ASF cooling like the one reported in [35].A more intuitive range of penalty can be obtained with a simple calculation using the fibers of [38] as an example.In three of these fibers, the most cooling was induced for a pump power absorbed per unit fiber length around 100 mW m −1 .Of this, about 1.5%-2% was converted into extracted heat by ASF, or 1.5-2 mW m −1 .The pump power at the site of the cooling was typically 150-200 mW.In such fibers, if the background absorption due to impurities were 10 dB km −1 , the pump power absorbed by impurities would be 0.35-0.46mW m −1 , of which 100% is converted into heat.Therefore, the heat generated by 10 dB km −1 of background absorption would be 20%-30% of the heat extracted by ASF.The range is actually broader, because it depends on many spectroscopic parameters, in particular the fiber length and the pump power launched into it.silica fibers and preforms then stood at critical quenching concentrations of 1.2 × 10 27 m −3 .The 2008 data point in figure 4 was measured on a silica preform (a bulk sample) [39].There was no attempt in that work to cool the sample; this data point is presented for comparison.Since 2019, while quenching in reported ZBLAN fibers has remained essentially unchanged, important refinements in the fabrication of silica fibers [1] has resulted in a marked increase in the value of N c for silica fibers.In aluminosilicate fibers doped with CaF 2 NPs, the authors reported last year values 30% higher than this previous record (1.6 × 10 27 m −3 ).Here, we add to this list a new aluminosilicate fiber (labeled ' * ' in figure 4) with an N c as high as 2.85 × 10 27 m −3 , which is more than twice as high as it was three years ago and a new record for a silica fiber.Earlier this year, the Kashyap group at Polytechnique Montreal reported a value about twice as large again (6.16 × 10 27 m −3 ) in a phase-separated aluminosilicate preform co-doped with yttrium [40].We note that although significant progress has been made by the University of New Mexico/Fraunhofer Institute team in cooling silica preforms to record levels, unfortunately the quenching performance of their glasses could not be added to this chart because their quenching levels were not quantified and cannot be inferred from the data provided in their publications.The bottom line of figure 4 is that significant reductions in quenching have been achieved in recent years in silica fibers and preforms, to the point where silica is now exceeding the performance of fluoride fibers on this critical metric.
This general improvement in quenching has also been paralleled by a commensurate increase in the Yb concentration that has been achieved in silica fibers and preforms, as illustrated in figure 5.This graph plots values of the Yb concentrations of silica fibers [7,38,40,41] and silica/silicate preforms [40,46,47] that were successfully cooled, as well as (fluoride glass) ZBLAN fibers [43][44][45] that cooled.The data were divided to highlight the relative performance of four different kinds of cooled objects, namely ZBLAN fibers, NP-doped  [44]).The gap between the materials is closing.We note that the new silica fiber labeled * , also featured in figure 4, has a ratio of Yb concentration to N c of only 4.5%, versus ∼10% for the optimum ratio.It means that for optimum cooling, this host composition should be doped with 2.2 times more Yb, which would bring this composition to the best level for a silica fiber.The aluminosilicate preforms from the New Mexico team have the lowest concentrations of all silicates that cooled, even though they cooled more than all other silica samples in figure 4. The reasons for the greater cooling are again their much larger doped area (typically 900 µm versus 21 µm or less for the silica fibers), and the fact that they were cooled in a vacuum [46,47].
The Yb concentration cited in [40], reported in an yttrium-doped aluminosilicate glass, is the highest value reported in a silicate glass that was successfully cooled.We note for completeness an inconsistency between this high concentration (4 × 10 26 m −3 at the center of the glass), measured by electron-probe microanalysis, and the relatively low measured absorption of the glass (2.6 cm −1 at 976 nm) [40], which would suggest that adding yttria to the glass greatly reduced the absorption cross-section, or that the Yb 3+ concentration may be lower than stated.The Yb concentration shown in figure 5 is 33% lower than stated in [40] (2.64 × 10 26 m −3 instead of 4.0 × 10 26 m −3 ) to correct for this error, in accordance with [48].The same correction was applied to the data point for [40] in figure 4, since this value is proportional to the Yb concentration [40].In spite of this disparity, which may result from multiple reasons, this work demonstrates that more highly modified glasses show definite promise for ASF cooling.
Many observations of ASF cooling have now been reported for fibers and preforms, made of either silica or ZBLAN.The cooling performance is summarized in figure 6 in the form of a plot of the largest negative temperature change from room temperature as a function of the pump power injected into the sample.The data are parsed into four groups to distinguish between silica and ZBLAN, and between measurements carried out at atmospheric pressure and in vacuum.The samples reported in figure 6 had doped diameters of 7.8-21 µm for the silica fibers, 52-200 µm for the multimode ZBLAN fibers, 3 µm for the single-mode ZBLAN fiber, and 900 µm and 2.6 mm for the silicate preforms, respectively.We leave it to the interested reader to pull this and other relevant data from the referenced papers.Most powers are the power launched into the core of the sample (either fiber or preform).Because of expected differences in how the powers are reported in the literature, a few values shown in the figure were measured at the site of the cooling, a few centimeters down the fiber, as in [7,41,45]; in a couple of cases the values are pump power as the launched power, but the unabsorbed pump power emerging from the sample was recycled through the sample with a reflector, as in [46].In spite of these small disparities, the trend is clear: proportionally more power is required to induce greater cooling (the dashed line in the figure is a guide to the eye with a slope of unity).All data points in figure 6 above the −100 mK level were measured in either single-mode or few-mode fibers, either in ZBLAN or silica [7,38,41,45].All points below this line (greater cooling) were reported for preforms in the case of silica/silicate samples and preforms or highly multimoded fibers in the case of ZBLAN samples [40,[42][43][44][45][46][47]49].This last group, especially preforms, have produced so far the largest temperature changes, again because of their larger volumes of doped materials, and also because in most studies, though not all, they were placed in a vacuum.The current record, credited to the University of New Mexico team, is −18.4K in an aluminofluorosilicate preform with a 900 µm core diameter pumped with 10 W at 1035 nm [47].In comparison, aluminosilicate fibers have cooled significantly less because their core areas are between three and four orders of magnitude smaller, and because they were cooled at atmospheric pressure.The record cooling for a silica fiber at atmospheric pressure with a small core (11 µm in diameter) is 62 mK.It was achieved in the aluminosilicate fiber labeled (1) in figures 4 and 5.It had a V number of 6.3 at 1064 nm.The power that was removed from the fiber to produce this level of cooling, easily calculated from [37], is 2 mW m −1 The pump power absorbed per unit length at the site of cooling was measured to be 85 mW m −1 .The cooling efficiency was then the ratio of these two quantities, or 2.4%.In all silica tests, the efficiency was in the broad range of 1%-2.5%.
These material advances have led in recent years to several 'first' demonstrations, including ASF cooling of an aluminosilicate rod fiber in vacuum by 18.4 K [47] and of a small-core aluminosilicate fiber in air by −70 mK (both from room temperature) [7].The heat extracted by ASF in a doped fiber is high enough that stimulated emission (which always generates heat) can be used to now make lasers and amplifiers that run cold, or at least at room temperature.This exciting possibility was demonstrated in 2021 with the report of the first radiation-balanced fiber amplifier [50] and laser [51], both at 1065 nm in Yb-doped silica fiber.The laser emitted 114 mW and had an optical efficiency of 8.4% with respect to launched pump power, and the amplifier had a gain of 17 dB.A radiation-balanced Yb:YAG disk laser with 1.05 W of output power and an optical efficiency of 15% was also recently reported [52].

Low-quantum-defect pumping
Already discussed above was the reduction of the QD by setting the pump and signal wavelengths to be closer together.In this case, advantages are gained from materials whose spectroscopy favors shorter wavelength operation (such as the fluorosilicates [26] and phosphosilicates [53]).Another promising approach to reducing the laser QD can be found in multi-wavelength pumping [54], referred to here as excitation balancing.This method has been introduced to solid-state lasers quite recently (2021), and therefore it has not been studied as extensively.
In this configuration, the laser makes use of a pump wavelength that is longer than the intended lasing wavelength.This anti-Stokes pump (P 1 in figure 3) effectuates the extraction of heat from the active fiber.More specifically, since one long-wavelength pump photon plus a phonon gives rise to a stimulated emission photon, thermal energy is carried away from the host by the laser signal.However, P 1 cannot alone provide net gain to the signal wavelength.A second pump (P 2 ), one whose photon energy is higher than that of the signal wavelength, is therefore required to realize gain.A caveat, however, is that the two pumps should not significantly overlap in space and time, otherwise the excitation will favor the longest wavelength, since it will experience the largest gain of all three wavelengths.As a result, this configuration is best suited for pulsed lasers and amplifiers.
A basic energy level diagram for the simplest embodiment (not to scale) is shown in figure 3(c).P 1 first excites some proportion of the Yb 3+ ions into the upper state.After a very short period following the absorption of pump photons, the population distribution in the upper-state manifold reaches thermal equilibrium (i.e.Boltzmann distribution).P 2 then follows P 1 and brings the system to a point of positive gain for the signal wavelength, suitable for either a laser or an amplifier.For a laser, the gain must be equal to the loss (e.g.cavity losses, background losses in the fiber, etc) to reach threshold.In the case of a pulse amplifier, the seed pulse follows the P 1 to P 2 pumping cycle.
The two-color pumping scheme described above is similar to that presented in [55], wherein a QE greater than one (1.008) was reported in a Rb D 2 line laser.Although the fiber demonstration reported in [26] did not exhibit a QE greater than one, it was shown that absorbed anti-Stokes power can contribute to stimulated emission.The laser configuration was a gain-switched Fabry-Perot laser employing a pair of fiber Bragg gratings.The best-case slope efficiency with respect to the launched anti-Stokes pump power was reported to be 38.3%[55].This value was limited in part by the large background loss in the newly developed fluorosilicate fiber.The lower absorption coefficient at the longer wavelength also led to significant pump leakage in an optimized length of fiber.In response, pump recycling methods were recommended.A QD as low as 0.7% was demonstrated for the two-color pumping configuration, reduced by about 30% from the single-pump case.However, net QD values near zero are plausible [1].More recent theoretical work has shown that pulse energy scaling to the multi-mJ level is possible in this configuration [56].As may be readily contemplated, laser performance, such as efficiency and thermal generation, is a strong function of the pump and signal wavelengths.It was shown that, generally, the highest pulse energies and lowest thermal generation occur when the anti-Stokes pump wavelength is close to that of the signal.Laser efficiency was relatively higher when pumping near the zero-phonon line.Finally, it was also shown that if the pump wavelengths are carefully selected relative to the mean spontaneous emission wavelength, ASF cooling, as described above, can cooperate to further reduce heat generation in pulsed lasers pumped with multiple wavelengths.

Grand challenges
Regarding internal cooling via ASF, critical research directions include higher Yb-doping levels in the core glass prior to the onset of clustering and quenching.However, while certain glass families are known to possess high quenching concentrations for rare-earth dopants, e.g.fluoride and phosphate glasses [57][58][59], an eternal verity in optical fiber is 'if you can use silica, use silica' [60].This is due to the very low losses, intrinsically high purity, and economy of scale afforded by the chemical vapor deposition process and the commoditization of silica fibers courtesy of global data communication networks, respectively.Silica is unfortunately naturally limited in its capacity to be doped, so creative approaches, such as the NP doping, are needed to increase the rare-earth doping levels achievable in silica.Further, as-fabricated preforms and drawn fibers can have high OH concentrations if drying processes are not thoughtfully employed.OH has a generally negative impact on cooling efficiency because it absorbs broadly around 1 and 1.3 µm, and it may also be a precursor to quenching [35,[61][62][63][64]. Reducing OH concentrations is therefore critical to maintaining high quantum yields and more efficient cooling.Methods for drying, such as the use of Cl 2 , may promote the reduction of desired Yb 3+ active ion to undesired Yb 2+ .Divalent Yb can promote photodarkening [65,66] while also reducing the concentration of Yb 3+ , thus directly lessening the potential cooling.Accordingly, understanding and balancing the drying and other fabrication processes to yield very low OH and Yb 2+ concentrations are paramount.Beyond OH, other impurities, particularly rare-earths and transition metals can also contribute to quenching and to increased absorptive losses [67,68].

Future opportunities
A key application of optical cooling is power scaling of fiber lasers and amplifiers.The main challenge is then to scale the relatively low level of cooling achieved so far in Yb-doped fibers to much larger extracted heat per unit time.This goal can be achieved using two broad avenues.The first one is using large-core fibers that support a single mode.In such fibers, the Yb-doped core diameter is typically up to tens of microns, which is about one order of magnitude larger than the core diameter of a typical single-mode fiber at a wavelength around 1 µm.This 10-fold increase in diameter translates into a 100-fold increase in core area, and a commensurate increase in the heat extracted per unit time [35].A large number of options are available to produce single-mode fibers with very large cores, including chirally-coupled-core fibers [69], multi-trench photonic-crystal fibers [70], and many others [71].
The second promising avenue, which has been studied theoretically [72][73][74] and demonstrated experimentally [43], is to use a double-clad fiber in which both the core and the cladding are doped.Both the core and the cladding are optically pumped, for example at the same wavelength if they are both doped with the same rare-earth ion.The core then serves the same purpose as in a conventional fiber laser or amplifier, namely it provides the required gain to amplify an external signal (amplifier) or spontaneous emission (laser), and it also provides a little cooling.The majority of the cooling, however, is performed by the cladding, which has a much larger diameter, and is pumped with a much higher power than the core, and therefore extracts much more heat from the fiber.Many variants of this scheme can be used, including using a different pump wavelength for the core and the cladding, or a different dopant in the core and the cladding [73].Modeling showed than in a Yb-doped silica fiber with a 30 µm core diameter, a 350 µm cladding diameter, and a critical quenching concentration of 1.64 × 10 27 Yb m −3 , up to 115 W of radiation-balanced power can be obtained [72].This value can be increased by using more recent fibers, which have a critical quenching concentration several times higher, a larger core, and a larger cladding, and/or operating at a laser wavelength closer to the pump wavelength.With further material progress and optimization, it is quite conceivable that a radiation-balanced fiber laser approaching the kW level may be attainable.A few kW may be within reach if the fiber temperature can be allowed to raise a moderate amount above room temperature, at which point a smaller external cooler can handle the small residual heat load.In principle, the outputs of multiple such fiber lasers can then be combined, coherently or incoherently, to produce higher output powers perhaps to the 100 kW level.Attention will then need to be paid to the relative spatial arrangement of the fibers, which must be such that the ASF is allowed to escape.On the other hand, with current silica science, radiation balancing a fiber laser or amplifier comprised of a single fiber at the 100 kW level appears to be far more questionable.That said, it is possible that anti-Stokes pumping, while not capable of maintaining such a high-power device at room temperature, may reduce its heat load sufficiently to lessen the burden on the external coolers.It will then be interesting in the future to perform simulations of the temperature distribution in large-core fiber amplifiers operating at the 100 kW level to quantify how low the temperature of the fiber can be kept and assess the potential role of ASF cooling in these significantly higher power sources.
One of the challenges of this application is that the cladding cannot be too thick, or it will reabsorb some of the ASF that is escaping the fiber radially, resulting in extraneous heating.Care must therefore be taken to design fibers that alleviate this general problem, which appears to be a necessary step to scale radiation-balanced fiber lasers or amplifiers to the 10 kW level and above.It is important to note, however, that the fiber does not need to be cooled exactly to room temperature.Reducing the core temperature down to a few Kelvin or even 20 K above ambient is generally sufficient, a relaxed requirement that will help power-scale this cooling technique to higher powers.
Another application that can potentially benefit from the ability to cool internally with light is fiber and free-space gratings.When used in conjunction with high peak or continuous wave (CW) powers, for example for pulse compression in pulsed amplifiers, these components heat up internally due to residual absorption of the light by the grating.This heat results in distortion of and instability in the optical pulses, sometimes severe, or damage of the optical grating.The cooling methods described in this Perspective can be used to offset this undesirable effect.One possible implementation is to dope the grating with an appropriate rare-earth ion and to optically pump the grating to refrigerate it internally.The pump can be either the signal itself (if it has a wavelength larger than the mean fluorescence wavelength of the laser ion), or a separate pump multiplexed with the signal (when both the signal and the pump are CW or quasi-CW).Here too, the grating temperature can be cooled close enough to room temperature to achieve the desired pulse stability-perfect radiation balancing is not necessary.
Because ASF cooling is performed internally, heat is removed at the core rather than at the surface of the fiber, which may have benefits in some laser/amplifier systems depending on their dynamics.As an example, when cooling a fiber, the heat is carried away from the host in the time it takes (1) the excited-state population to reach steady-state, and (2) phononic energy to be transferred from the host glass to the excited Yb 3+ electrons.The first step is by far the slowest, with a time constant of the order of 1 ms for Yb 3+ in silica, unless the pump power far exceeds the saturation power, in which case it is somewhat faster.If a single short pulse is sent through this fiber to be amplified by the same excited-state population, the heat generated by the pulse amplification will heat the core as fast as the population is depleted, in a time constant that depends on the pulse width and intensity.Since it takes ∼1 ms for ASF to re-cool the core, the next pulse will still see a 'warmer' core if it arrives less than 1 ms after the first pulse.For comparison, if the same fiber was cooled externally, with forced air for example, because of the low thermal conductivity of glass it would take several ms for this heat to diffuse into the (colder) cladding, depending on the dimension of the core (the larger, the longer).One can then expect more rapid cooling with ASF than with other forms of external cooling, enabling pulse rates up to 1 kHz for ASF cooling with Yb-doped silica, versus a few 100 Hz or less for external cooling.This general issue of the dynamics of cooling with ASF has important implications, in particular in the effectiveness of the cooled-FBG application mentioned in the previous paragraph.It is probably best answered by performing advanced simulations of the dynamics of ASF and external cooling.
The same general benefits are expected for other optical systems, for example ring resonators, especially those that exhibit internal gain, like ring-laser gyroscopes that rely on stimulated Brillouin scattering [75] or electronic transitions in Er 3+ [76] as the gain mechanism.In these active devices, the gain medium generates internal heat that causes fluctuations in the index and length of the ring waveguide.These fluctuations translate into instabilities in the resonance frequencies of the ring resonator, and therefore in the output of the ring laser.In turn, these instabilities induce increased noise and/or drift in the sensor output, which degrades its performance.Here too, the implementation of internal cooling via ASF can reduce these sources of error in the gyro output and improve its performance.
Other areas of application for self-cooled high-power lasers are in space.For example, there is renewed interest in lunar exploration; with an increasing number of countries sending unmanned probes to the lunar surface, situational awareness in the cislunar range is becoming increasingly important.Lunar exploration can be supported through the transmission of power via laser beams [77].Furthermore, since the moon is 384 400 km from the earth, actively probing the cislunar region of space will require lasers in the kW range.Such lasers can serve as illuminators or pulsed lidar transmitter sources for large-area telescopes looking in that region of space from the ground or from space.The former would give rise to a bistatic sensing configuration.

Conclusions
The goal of this Perspective was to shine light on the importance of thermal management in the next generation of high-power and low-noise fiber systems, and to underscore the crucial role that anti-Stokes pumping methods will play.Even if not completely displacing flowing water systems, the current state-of-the-art in anti-Stokes cooling can offset at least part of that burden.Looking forward, state-of-the-art reports of ASF cooling were tabulated and presented graphically, highlighting the rapid evolution of silicate fibers from infancy over just a few years.Should this trend continue, ideal Yb-doped fibers with near 100% QE and significantly scaled quenching concentrations are just over the horizon.Coupled with judicious fiber and laser design, athermal fiber lasers, amplifiers, and sensors will follow soon thereafter for a wide range of applications.

Figure 1 .
Figure 1.Thermal profile of an idealized cladding-pumped laser fiber (25 µm core diameter, 400 µm cladding diameter, 600 µm coating diameter, 0.06 numerical aperture) with an 83.3 W m −1 thermal load, corresponding to a 10 kW-class laser absorbing 833 W m −1 of pump with 10% conversion (quantum defect and non-radiative relaxation) to heat (solid curve).The dashed curves show the impact of decreasing either the non-radiative component (green), the quantum defect (purple), or both (blue).Cladding background absorption was neglected in the calculation.

Figure 2 .
Figure 2. Representative power budget for a Yb-doped fiber amplifier.

Figure 3 .
Figure 3. Energy-level diagram for cooling processes based on anti-Stokes pumping.(a) anti-Stokes fluorescence (see text for details).(b) For a radiation-balanced laser, a pump, P, electrons are excited into a lower sub-level of the upper manifold.These electrons acquire energy from the phonon bath (thermalization) to satisfy Boltzmann distribution.Upon relaxation to the ground state, anti-Stokes fluorescence photons are emitted that carry the acquired energy out of the system, leading to cooling.(c) For an excitation-balanced laser, a first, lower energy pump, P 1 , brings the upper state population to a starting level.The second pump, P 2 , then brings the system to a point of positive gain to the signal wavelength.The blue squiggly lines represent cooling due to phonon annihilation while the red squiggly lines represent heating.

Figure 4 .
Figure 4. Improvement over time in published values of the critical quenching concentration Nc of aluminosilicate (oxide) and ZBLAN (fluoride) glass fibers.All values were inferred from measurements conducted in optical fibers, except for[39,40,42], where they were conducted in bulk glass preforms.The asterisk ( * ) denotes data reported here for the first time.

Figure 5 .
Figure 5. Improvement over time in published values of the Yb concentration in aluminosilicate and ZBLAN (fluoride) glass fibers and preforms that were successfully cooled by anti-Stokes pumping (the pure-silica sample was not tested for cooling; it is shown here as a reference to a typical highly-doped fiber that predates the recent work cooling).The asterisk ( * ) denotes data reported here for the first time.

Figure 6 .
Figure 6.Compilation of the measured temperature changes reported in the literature for ZBLAN fibers and silica fibers and preforms, plotted as a function of pump power.