Implementation of Laser-Induced Anti-Stokes Fluorescence Power Cooling of Ytterbium-Doped Silica Glass.

Laser cooling of a solid is achieved when a coherent laser illuminates the material, and the heat is extracted by annihilation of phonons resulting in anti-Stokes fluorescence. Over the past year, net solid-state laser cooling was successfully demonstrated for the first time in Yb-doped silica glass in both bulk samples and fibers. Here, we report more than 6 K of cooling below the ambient temperature, which is the lowest temperature achieved in solid-state laser cooling of silica glass to date to the best of our knowledge. We present details on the experiment performed using a 20 W laser operating at a 1035 nm wavelength and temperature measurements using both a thermal camera and the differential luminescence thermometry technique.

Introduction

The possibility of heat extraction from sodium metal vapor via anti-Stokes fluorescence (ASF) was first suggested by Pringsheim in 1929.[1] Nearly 7 decades later, in 1995, Epstein et al. reported the first experimental observation of laser-induced ASF cooling of a solid in Yb-doped ZBLANP (ZrF4–BaF2–LaF3–AlF3–NaF–PbF2).[2] Several attempts have since confirmed laser cooling in various solid-state materials, primarily in rare-earth-doped (RE-doped) crystals and glasses. Laser cooling of RE-doped crystals has been the most successful so far;[3−6] the record cooling of a Yb-doped YLiF4 (Yb:YLF) crystal was reported at the University of New Mexico in 2016.[7] Several RE-doped glasses have been successfully cooled over the years;[8−15] Yb-doped silica glasses are the most recent additions to the list of successfully cooled RE-doped materials.[16,17,24] Although laser cooling of RE-doped silica was thought to be elusive over the years, recent investigations pointed out its possibility[18−20] and eventually led to its experimental observation.[16,17,21−24] In all these reports, the temperature drop of the laser-cooled Yb-doped silica was less than 1 K. Here, we present, to the best of our knowledge, the largest temperature drop to 6 K below room temperature.

There are many potential applications for optical cooling through ASF. In principle, it can be used for compact, vibration-free refrigeration systems,[5,8] for example, when precision cooling is demanded in low-thermal-noise detectors and reference cavities of ultrastable lasers, or even in physiological applications.[24] One can even envision laser-cooled silica’s potential usage as the substrate in silicon photonics devices.[25−28] Another important potential application for ASF cooling is in radiation-balanced fiber lasers (RBFLs), where ASF cooling balances the waste heat generated in the laser.[29−34] Historically, RE-doped ZBLAN glasses have been more amenable to the stringent requirements needed for laser cooling. Unfortunately, ZBLAN fibers are low in mechanical durability and chemical stability and hard to cleave and splice, so they are generally less desirable than silica fibers. However, the recent advances in laser cooling of Yb-doped silica glass open a potential pathway for future applications in RBFLs. Of course, much more substantial cooling is required to make a viable impact on fiber laser designs; this paper is a step in this direction.

Review of the Recent Results

The cooling efficiency, ηc, characterizes the potential of a material to cool via laser-induced ASF. It is defined as the net power density (per unit volume) extracted from the material (pnet) per unit total absorbed power density (pabs). The cooling efficiency is a function of the pump laser wavelength λp and can be expressed as[16]

The mean fluorescence wavelength is represented as λf. The external quantum efficiency, ηext, and the absorption efficiency, ηabs, are defined aswhere Wr and Wnr are the radiative and non-radiative decay rates of the excited state in the RE dopant, respectively, and ηe is the fluorescence extraction efficiency. αb is the background absorption coefficient, and αr(λp) is the resonant absorption coefficient due to the RE dopants. Note that the attenuation due to scattering, including Rayleigh scattering, does not contribute to the material’s heating; therefore, αb represents only the background absorption and not the total parasitic attenuation.

For net solid-state optical refrigeration, the cooling efficiency must be positive; therefore, we must show that ηc > 0 is reachable over a range of λp. The laser pump wavelength λp cannot be much longer than λf; otherwise, the pump absorption cross-section would become too small. This would result in a small αr and hence a small ηabs and a negative cooling efficiency. In practice, to observe net cooling, λp can only be slightly longer than λf, and both ηext and ηabs must be near unity. To realize the ηabs ∼ 1 limit for λp ≫ λf, one must increase the RE dopant density to achieve αr(λp) ≫ αb. However, increasing the RE dopant density results in an increase in the non-radiative decay rate, Wnr, primarily because of the RE clustering and quenching, hence decreasing the external quantum efficiency, ηext. This unfortunate circle of undesirable influences was recently overcome in Yb-doped silica. It was shown that by adding certain modifiers such as Al, P, F, and Ce, the quenching concentration of silica glass could be increased significantly.[18,35,36] The result was the successful cooling of high-Yb-concentration silica as a fiber preform by Mobini et al.[16] up to 0.7 K and as an optical fiber by Knall et al.[17] up to 50 mK.

To investigate the cooling efficiency as a function of the pump wavelength and obtain the optimum value of λp for maximum cooling, we performed a laser-induced thermal modulation spectroscopy (LITMoS) test[5,37] on our Yb-doped silica samples.[16] In Figure , we show −ηcαr as a function of λp for the sample used in this paper (the same as sample A studied by Mobini et al.,[16] but some of the cladding is removed; see Table ). This quantity is proportional to the change in the sample temperature for a fixed pump laser power. Figure shows that the maximum temperature drop can be obtained at around 1035 nm. At the time when we carried out our experiments for ref (15), the only viable high-power source in our laboratory was a 1053 nm laser. In this paper, as will be explained later, we use a λp = 1035 nm source to achieve a higher temperature drop.

Figure 1

Value of −ηc αr(λ), which is proportional to ΔT at a fixed input laser power (in the low-absorption regime), versus the pump laser wavelength for our sample. The solid line presents the best fit to the experimental measurements reported by Mobini et al.[16] This figure is adapted from Figure S3 of Mobini et al.[16]

Table 1

Properties of the Yb-Doped Silica Glass Sample

parameter	value	error
codopants	Al, P
Yb2O3 [mol %]	0.12	±0.01
Yb density [1025 m–3]	5.3	±0.4
OH– concentration [ppm]	3.0	±0.5
core diameter [mm]	1.7	±0.1
cladding diameter [mm]	2.9	±0.1
length [mm]	15.1	±0.1
αb [dB km–1]	10	±2

Results

Power Cooling Experiment

The samples that we laser-cooled in our experiments reported in ref (15) were surrounded by undoped (no Yb doping) silica glass cladding regions, which provide significant thermal load. The cooling in our experiments was achieved in spite of this large thermal load. For this work, we chose sample A studied by Mobini et al.[16] and removed most of its undoped cladding region to reduce the thermal load and enhance the cooling effect. Moreover, we built a high-power source at the optimum cooling wavelength of 1035 nm as described below.

The characteristics of the (fiber preform) sample are listed in Table . The Yb2O3 concentration is measured by electron probe micro-analysis. The Yb density is calculated from the measured Yb2O3 concentration. The error for the Yb2O3 concentration is related to the applied method’s uncertainty in this concentration range. The OH– concentration and parasitic background absorption (αb) are measured by the cut-back method in the fiber form, for which the errors express the repeatability of the measurement setup.

To make a high-power source at the nearly optimum λp = 1035 nm wavelength, we have designed and built a fiber amplifier to amplify the output of our continuous-wave tunable Ti:Sapphire laser. The fiber amplifier’s gain medium is a 1.2 m piece of Yb-doped double-cladding fiber pumped using a high-power diode laser at the wavelength of 976 nm. The amplifier’s input is approximately 300 mW, and the amplified output of the fiber amplifier is on the order of ∼20 W at the 1035 nm wavelength. We note that any residual diode pump power at the 976 nm wavelength can be a significant source of heating in the material because of the Yb-silica sample’s absorption peaks at 976 nm. Therefore, to observe laser cooling, a spectrally pure laser is essential. To reduce the fiber amplifier’s 976 nm pump leakage in the output as much as possible, we implement a cladding mode stripper scheme at the fiber amplifier’s end. We also use a stack of two 1000 nm wavelength long-pass dichroic mirrors to filter out the rest of the 976 nm pump leakage.

The experimental setup for the power cooling experiment is shown in Figure . The Ti:Sapphire laser is tuned to a wavelength of 1035 nm, which is then amplified by the fiber amplifier. The output laser light is then collimated and filtered. The collimated light is coupled to the sample through a long-focal-length lens from the outside of the vacuum chamber. The vacuum chamber pressure is maintained at 10–6 torr during the power cooling experiment to minimize convective heat transfer. A spectrometer captures the sample’s fluorescence through a KCl salt window mounted in the chamber. Similarly, the thermal images are recorded via a thermal camera through the thermally transparent KCl salt window, and the images are post-processed to measure the changes in the sample’s temperature.[16] The mean fluorescence wavelength is calculated from the fluorescence emission measured using an optical spectrum analyzer and is found to be λf = 1010 nm. In order to decrease the thermal contact on the sample, the sample is mounted on very thin glass fibers. To minimize the back reflection into the fiber amplifier, the sample and chamber windows are tilted slightly, and a beam block is used to capture the laser after exiting the sample.

Figure 2

Wavelength-tunable continuous-wave Ti:Sapphire laser is coupled to a homemade fiber amplifier’s input through a 20× microscope objective. The amplified laser light is collimated again using a lens with the focal length of f = 5 cm. The collimated light is then filtered using a stack of two one-micron long-pass dichroic mirrors. The filtered and collimated light is coupled to the Yb-doped silica glass sample using a lens with the focal length of f = 12 cm. The sample is held inside a vacuum chamber. The upper-left inset shows a sketch of the Yb-doped silica glass sample supported by a set of thin silica fibers to minimize the heat load.

The red dots in Figure show the sample’s temperature evolution, measured using the thermal camera, as a function of exposure time to the 20 W laser light at 1035 nm. The temperature drop is ΔT = Ts – T0, where Ts is the sample temperature and T0 ≈ 23 °C is the ambient temperature. The thermal camera saturates at ΔT ≈ – 6 K, so the temperature may have dropped below the saturation value [see subsection (3.2); Differential Luminescence Thermometry]. In this power cooling experiment, the sample’s temperature evolution as a function of time follows the following exponential form (see Mobini et al.[16] for a derivation)where we use the following definitionsHere, V is the sample volume, ε = 0.85 is the emissivity of the implemented Yb-doped silica glass fiber preform, σ = 5.67 × 10–8 W·m–2·K–4 is the Stefan–Boltzmann constant, T0 is the ambient temperature, A is the surface area of the sample, ρ = 2.2 × 103 kg·m–3 is the silica glass mass density, and cv = 741 J·kg–1·K–1 is the specific heat of the silica glass.[38−40]Pabs is the absorbed laser power that can be estimated from the Beer–Lambert law in a single pass[41,42]Here, Pin is the input power coupled into the fiber preform at z = 0, l is the sample length, and αr(1035 nm) ≈ 1.93 × 10–2 cm–1.[16] In fact, by combining eqs and 5, we can see that ΔTmax ∝ ηcαr, which is the vertical axis in Figure used to estimate the optimum pump laser wavelength. By fitting the exponential form in eq to the experimental data (red dots) in Figure , we obtain ΔTmax = 6.02 ± 0.01 K and τc ≈ 166 ± 1 S—the error bars are estimated by the fitting procedure. The dashed blue line is the theoretical fit and agrees with the experiment quite well. Using the measured value of ηc ≈ 0.016 at λp ≈ 1035 nm reported by Mobini et al.[16] for sample A, we use eqs and 5 to estimate ΔTmax ≈ 9 K. This theoretical estimate is consistent with the measured fitted value of ΔTmax = 6.02 ± 0.01 K because the heat conduction from the fiber-holder contact and also the parasitic heating from fiber facet imperfections are not included in the theoretically ideal form of eq . Moreover, the fitted value for τc agrees quite well with the measurement reported by Mobini et al.[16] once the difference in geometry is taken into account (τc ≈ 175 S vs τc ≈ 166 S). The goodness of the fitting in Figure indicates that despite the saturation of the camera, the actual value of ΔTmax cannot be much larger than 6 K.

Figure 3

sample’s temperature change is plotted as a function of time when exposed to the high-power 1035 nm laser light. The red dots correspond to the experimental results, and the blue dashed line represents the fitting of the exponential function in eq to the experimental data. The insets show two thermal images corresponding to before laser exposure and after the final temperature stabilization.

Differential Luminescence Thermometry

In this technique, the variation in luminescence intensity distribution with temperature is used to determine the sample’s temperature. This variation is due to the temperature dependence of the Boltzmann population of the crystal field levels of the emitting state and the homogeneous line width of the individual crystal field transitions.[43] Differential luminescence thermometry (DLT) has been successfully used to measure temperature variations on the order of tens of Kelvin;[7] however, it can be quite noisy and less accurate for smaller temperature variations such as those reported here. The reason is that unlike semiconductors where substantial spectral shifts are observed as a function of the temperature,[44] the 4f electrons in REs are shielded from the environment in a solid. The noise in our DLT measurement is dominated by the standard spectrometer noise, mainly due to thermal and mechanical effects.

For DLT, the temperature-dependent emission spectral density S(λ,T) is obtained in real time and is referenced to a spectrum at the starting temperature T0. The normalized differential spectrum is defined as

Normalization to the spectral peak Smax is performed to eliminate the effect of input power fluctuations. The scalar DLT signal is given bywhere the limits of integration bracket the sample’s spectral emission, eliminating possible contributions from the spurious laser line scattering; we choose λ1 = 895 nm and λ2 = 955 nm. The temperature drop from the ambient, ΔT, is linearly proportional to SDLT: ΔT = γ·SDLT, where γ is the proportionality constant. To use DLT for temperature measurements, we first perform a calibration measurement by mounting the sample on a variable-temperature cold plate, while pumping the sample with the Ti:sapphire laser and collecting the spectrum. We find that for our sample, γ = −34 ± 2 K.

We use the DLT calibration result to measure the sample’s temperature evolution over time, while being exposed to the 20 W laser light at 1035 nm by collecting the emission spectral density every 10 s. The results are shown in Figure . The DLT data points are in blue dots, where the error bars are due to the error in γ as estimated from the calibration. The results are compared with the thermal camera measurements in red dots. The DLT results are quite noisy as expected; however, the trend agrees with the temperature values from the thermal camera and also hints that the sample is cooled slightly more than 6 K below the ambient temperature, consistent with the results presented in subsection (3.1), Power Cooling Experiment.

Figure 4

sample’s temperature change is plotted as a function of time when exposed to the high-power 1035 nm laser light. The blue dots are based on the DLT method, and the red dots represent the temperature measurements using the thermal camera.

Discussion

We have demonstrated the laser power cooling of Yb:silica glass to 6 K below room temperature. This result constitutes almost an order of magnitude improvement compared with the previous result of 0.7 K reported by Mobini et al.[16] Our work points to the feasibility of an all-fiber-based cryocooler using Yb:silica after a moderate improvement in preform synthesis. For comparison, Yb:ZBLAN (2% doped) having a background absorption of 2 × 10–4 cm–1 was cooled to about 200 K with P ≈ 10 W (at 1026 nm).[11,45] Our current Yb:silica glass having a lower background absorption by an order of magnitude has the potential to outperform ZBLAN. With a moderate improvement in doping concentration (i.e., 1%), one can envision all-fiber refrigerators reaching 150 K with no moving parts. It should be noted that the emission spectrum and, consequently, the absorption spectrum of Yb-doped silica and ZBLAN are reasonably close. Future improvements are possible by increasing the pump power and implementing a multipass scheme, improving material specifications, and optimizing the sample geometry. A video clip of the cooling evolution of the sample is presented in the Supporting Information section. The video shows the temporal evolution of the sample’s temperature as captured by the thermal camera in the high-power laser cooling experiment. The thermal image of the sample gets darker as the sample cools due to the exposure to the high-power laser.