Professional Documents
Culture Documents
A DISERTATION
SUBMITTED TO THE DEPARTMENT OF ELECTRICAL ENGINEERING AND THE
COMMITTEE ON GRADUATE STUDIES
OF STANFORD UNIVERSITY
IN PARTIAL FULFILLMENT OF THE REQUIREMENTS
FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
JENNIFER M. KNALL
JUNE 2021
© 2021 by Jennifer Maria Knall. All Rights Reserved.
Re-distributed by Stanford University under license with the author.
ii
I certify that I have read this dissertation and that, in my opinion, it is fully adequate
in scope and quality as a dissertation for the degree of Doctor of Philosophy.
I certify that I have read this dissertation and that, in my opinion, it is fully adequate
in scope and quality as a dissertation for the degree of Doctor of Philosophy.
David Miller
I certify that I have read this dissertation and that, in my opinion, it is fully adequate
in scope and quality as a dissertation for the degree of Doctor of Philosophy.
Olav Solgaard
This signature page was generated electronically upon submission of this dissertation in
electronic format. An original signed hard copy of the signature page is on file in
University Archives.
iii
Abstract
The performance of fiber lasers is often limited by waste heat. The resulting temperature variations in the
gain medium can induce undesirable fluctuations in the laser mode, frequency, and power. To date, the only
solution is the use of conventional mechanical cooling methods, which are cumbersome and sometimes
detrimental to the quality of the laser output. Optical cooling through the use of anti-Stokes fluorescence
(ASF) provides a compact solution that reduces or eliminates the degradation of laser performance. This
phenomenon was first theorized in 1929 but not realized experimentally until 1995. Since then, only a select
few gain materials have exhibited ASF cooling, and almost exclusively in bulk form and in a vacuum. All of
these materials are fairly exotic and generally unavailable in a fiber form. In the realms of lasers and
amplifiers, none of them have the scale or technological importance of silica fibers.
In this thesis, we report an extensive theoretical and experimental investigation of ASF cooling in Yb-
doped fibers at atmospheric pressure. We started by developing a model to simulate ASF cooling in a fiber
doped with a quasi-two-level laser ion. This model was instrumental in optimizing the pump and fiber
parameters to maximize cooling and, by fitting the model to our experimental data, it enabled us to infer fiber
parameters relevant to cooling. The model was later expanded to simulate radiation-balanced lasers and
amplifiers, devices in which the heat generated by the quantum defect is negated by cooling due to ASF.
These models are now available to other researchers as a valuable tool to predict the amount of cooling in a
Our experimental work started with the first-ever demonstration of ASF cooling in a fiber at atmospheric
pressure. The initial demonstrations were in Yb-doped ZBLAN fibers, but significant material-science
breakthroughs ultimately lead to cooling in Yb-doped silica fibers, an achievement that was thought to be
very unlikely due to the high levels of concentration quenching typically associated with silica. Capitalizing
on this success, we used the best silica fiber to create the first radiation-balanced fiber amplifier, which
exhibited 17 dB of gain (146 mw of output power) and no net heating. We then integrated the fiber into a
cavity formed with fiber Bragg gratings and created the first radiation-balanced fiber laser, a device that
produced 105 mW of output power and zero average temperature change along the 2.64-m fiber length. By
eliminating the need for conventional coolers, this result will enable fiber lasers with lower relative intensity
iv
noise and greater temporal coherence than possible until now. Given the prominence of silica fiber lasers in
technology, this new functionality is expected to have far-reaching benefits to many applications ranging
v
Acknowledgements
This work has been a hugely collaborative effort, and the many breakthroughs presented in this thesis
would not have been possible without the knowledge, support, and contributions from numerous people
First, I would like to thank my Ph.D. advisor, Michel Digonnet. Thank you for the generous amount of
time and knowledge you have given to this project. I have learned so much under your patient guidance.
Thank you for kind words of encouragement and your trust in my abilities.
Next, I would like to thank the collaborators that have been pivotal to the success of this project. Magnus
Engholm, John Ballato, and Wade Hawkins, thank you for providing the silica fibers that made this
breakthrough work possible. Thank you, Peter Dragic and Nanjie Yu, for characterizing these numerous
fibers. Martin Bernier and Tommy Boilard, thank you for manufacturing the fiber Bragg gratings that were
fundamental to the high-resolution temperature sensor; and for making the fiber laser pump sources. Thank
you, Samuel and Marcel Poulain, for providing the ZBLAN fibers and working hard to perfect the design for
the double clad ZBLAN fiber. I would like to thank everyone for creating such a positive, supportive, and
motivating collaboration. I am genuinely grateful for the experience your participation has fostered.
I would also like to thank all my lab mates, Arushi Arora, Matthew Grant, Behrad Habib Afshar, Therice
Morris, Jonathan Wheeler, Adele Zawada, Mina Esmaeelpour, and Pierre-Baptiste Vigneron. Thank you for
the comradery, the support, and the laughs. I would particularly like to thank Arushi for teaching me most of
what I know about optics experiments. Thank you for your seemingly endless patience and encouragement.
I would also like to thank Pierre for helping with experiments during my last two years.
I would also like to thank Carsten Langrock and the rest Martin Fejer's group for the generous amount of
Finally, I would like to thank my family for their support and patience. Thank you for creating the solid
foundation on which I felt I could try anything. The confidence and security fostered by this unconditional
support is the basis for everything I have accomplished and everything I will accomplish in the future.
vi
Contents
Abstract iv
Acknowledgements vi
Chapter 1: Introduction 1
1. Overview 1
for Mitigation 13
2.2. Absorptive Loss 16
4. References 20
1. Introduction 22
2. Model 25
vii
3. Simulations of Cooling in a Yb-doped ZBLANP Fiber 34
4. Summary 46
5. References 47
1. Introduction 49
2. Model 52
4. Summary 66
5. References 66
viii
Chapter 5: Cooling in ZBLAN Fibers 68
1. Introduction 68
2. Conventional Yb-doped ZBLAN Fibers 69
3.3. Results 84
4. Summary 86
5. References 87
1. Introduction 88
2. Background 89
ix
3.2. Summary of Tested Fibers 93
7. References 112
1. Introduction 114
2. Experimental Design 115
x
3.1. First Demonstration of a Radiation-Balanced Fiber Amplifier 120
4. Summary 124
5. References 124
5. References 133
1. Summary 134
2. Future Work 137
3. References 140
xi
List of Tables
Chapter 3
Table 1. Yb-doped fiber parameters 34
Chapter 4
Table 1. Parameters for the Yb-doped fibers, pump, and laser signals 57
Chapter 5
Table 1. Yb-doped ZBLAN fiber parameters 69
Chapter 6
Table 1. Yb-doped silica fiber parameters 93
Table 3. Measured and inferred (*) Yb-doped silica fiber parameters 104
Table 4. Inferred Yb-doped silica fiber parameters 107
Chapter 7
Table 1. Parameters for the Yb-doped silica fiber used in the radiation-balanced
Chapter 8
Table 1. Parameters for the Yb-doped silica fiber used in the radiation-balanced fiber laser 125
xii
List of Figures
Chapter 2
Fig. 1. The Stokes process dictating laser operation. First, high energy light is used to excite
electrons into the upper manifold. The electrons then thermalize according to the
Boltzmann distribution by generating phonons. Stimulated emission induces lower energy
laser light to exit the system. 11
Fig. 2. The anti-Stokes fluorescence process. First, lower energy light is used to excite electrons
into the upper manifold. The electrons then thermalize according to the Boltzmann
distribution by absorbing phonons. Spontaneous emission generates higher energy
fluorescence that escapes the system. 12
Fig. 3. Mechanism for diffusion-limited concentration quenching: an excited Yb ion transfers its
energy to an impurity, and nonradiative relaxation ensues. 14
Fig. 4. Effect of concentration quenching on the total upper-state lifetime of Yb for ZBLAN (red
curve) and silica (blue curve), plotted as a function of Yb concentration. 16
Fig. 5. Effect of absorptive loss and concentration quenching on the pump power dependent
temperature change. 17
Fig. 6. (Taken from [17]) Custom slow-light fiber Bragg grating temperature sensor used in this
work. 18
Fig. 7. Energy level diagrams for three cases: ideal lasing (left), ideal cooling (middle), and
radiation-balanced lasing (right). To create an RBL, the device needs to be pumped at an
energy that compromises between cooling and amplification, so that both processes can
occur simultaneously. 20
Chapter 3
Fig. 1. (a) Schematic of the doped fiber and definitions of key model parameters; (b) Energy-level
diagram of Yb3+ as an example of a two-level system for ASF cooling; (c) Absorption and
emission cross-section spectra of the Yb-doped ZBLAN fiber simulated in this chapter. 24
Fig. 2. Position-dependent values for N2 (thick dotted red curve), the pump power (thick solid
blue curve), and ASE (thin black curve) calculated for a 1-W pump power at 1010 nm
launched into a 2-m Yb-doped ZBLANP fiber with the parameter values listed in Table 1
and the cross-section spectra shown in Fig. 2. 36
Fig. 3. Predicted dependence on pump wavelength of the average heat extracted from a 40-cm or
80-cm Yb-doped ZBLANP fiber, with and without ASE, and of the average gain along
the 40-cm fiber at selected wavelengths (when ASE is included). 37
xiii
Fig. 4. Dependence of the maximum extracted heat on the launched pump power at 1015 nm for
a fiber 1.2-m long with and without absorptive background loss. 39
Fig. 5. Temperature distribution along a 5.5-m Yb-doped ZBLANP fiber predicted for various
launched pump powers at 1015 nm. 40
Fig. 6. (a) Dependence of the total extracted heat on the launched pump power at 1031 nm for
three fiber lengths. (b) Dependence of the total extracted heat and cooling efficiency on
pump wavelength for a fiber long enough to absorb most of the pump even at the longest
wavelength (i.e., 20 m) and the corresponding cooling efficiency. At each wavelength the
power is optimized to maximize the total extracted heat. 42
Fig. 7. (a) Maximum total heat extraction as a function of Yb concentration for a 1031-nm pump
launched into a 15-m fiber with an absorptive loss of 15 dB/km. (b) Cooling efficiency as
a function of concentration for the same fiber. 43
Fig. 8. Maximum extracted heat per unit length and maximum temperature change (when the fiber
is placed in air) predicted as a function of fiber absorptive loss for different core radii and
Yb concentrations. 44
Fig. 9. The absorptive loss required to achieve 90% of the maximum local cooling efficiency as
a function of Yb concentration, for two values of the pump absorption cross-section (sa0
is the value obtained from the spectrum in Fig. 1c). 46
Chapter 4
Fig. 1. Cross-section of the (a) double-clad fiber considered in this work, in which both the core
and the cladding are doped with Yb3+, and (b) an alternate triple-clad design that
incorporates an undoped inner cladding to pump the core. (c) To create an RBFL, the
double-clad fiber is sliced between two FBGs and bidirectionally pumped into the doped
cladding. 50
Fig. 2. (a) Signals and pump powers simulated by the RBFL model. (b) Example of the z-
dependent profiles of the cladding-pump powers, laser-signal powers, and upper-state
populations. The ASE power distributions are comparable to the signal powers except that
they start at zero power at the ends of the fiber. 51
Fig. 3. Temperature-change profile, computed with and without ASE, for an 8-m silica fiber laser
bidirectionally pumped with 200 W at 1030 nm. 60
Fig. 4. (a) Forward (solid curve) and backward laser signal powers (dashed curve) calculated as
a function of position for two 37-W lasers with unidirectional and bidirectional pumping
schemes. (b) The resulting position-dependent temperature profiles. 61
xiv
Fig. 5. a) Calculated maximum output power and associated optical-to-optical efficiency for a
silica (solid curves) and ZBLAN (dashed curves) RBFL as a function doped cladding area
(cladding diameter is labeled above). b) The difference between the highest and lowest
temperature along the fiber for the laser operating at the maximum power. 63
Fig. 6. Temperature profile as a function of normalized cavity length predicted for a conventional
cladding-pumped fiber laser pumped at 976 nm and an RBFL pumped at 1030 nm, both
with 115 W of output power at 1064 nm. 65
Chapter 5
Fig. 1. Experimental set-up used to measure the temperature change in the Yb-doped ZBLAN
fibers. 71
Fig. 2. (a) Although thermal gel creates a good thermal contact between the test fiber and the
FBG, it induces an erroneous source of heating as it absorbs the ASF escaping radially
from the doped fiber. (b) The temperature change induced by core-pumping a Yb-doped
borophosphosilicate fiber with a constant power measured for different thicknesses of gel
applied to the Yb-doped fiber. As the amount of gel is increased, the temperature also
increases, confirming the heating effect of the gel. 72
Fig. 3. (a) The absorption spectrum measured by LVF, and the emission spectrum calculated with
the McCumber relation; (b) the same absorption spectrum, adjusted at longer wavelengths
after additional measurements as described in the text, and the emission spectrum
calculated from it with the McCumber relation. 73
Fig. 4. Experimental data (blue crosses) and simulated fits (solid red curves) for the output power
as a function of launched pump power for (a) the single-mode Yb-doped ZBLAN fiber,
and (b) the multimode Yb-doped ZBLAN fiber. The inset in (b) shows the mode profile
measured at the output of the multimode fiber. 75
Fig. 5. Temperature measurement for the single-mode Yb-doped ZBLAN fiber end-pumped at
1020-nm to induce cooling via anti-Stokes fluorescence. The blue curve is the temperature
of the fiber measured over a period of 90 seconds, during which the pump was
alternatively turned on and off. The black curve is the averaged data for each pump-on
and pump-off section, and the pump-power on-off cycles are pictorially represented by
the red curve. 77
Fig. 6. Temperature measurements (crosses) and fits (solid curves) plotted as a function of the
pump power at the FBG sensor location for (a) the single-mode fiber, and (b) the
multimode fiber. 77
Fig. 7. An example measurement of the temperature evolution over time in the multimode Yb-
doped ZBLAN fiber as the pump was turned on and off. 79
xv
Fig. 8. Cooling efficiency measured as a function of pump power at the sensor location for the
single-mode Yb-doped ZBLAN fiber (upper red cross) and the multimode Yb-doped
ZBLAN fiber (lower blue crosses). 80
Fig. 9. Temperature change measured as a function of absorbed pump power per unit length at
1025 nm for a multimode Yb-doped ZBLAN fiber. The red crosses represent the
measurements performed with thermal gel between the test fiber and the FBG sensor, and
the blue squares show the data taken with the improved isopropanol-based method. 81
Fig. 10. Cross section of the ZBLAN fibers with a Yb-doped core and cladding: (a) a double-D
shaped inner cladding and (b) an octagonal inner cladding. 83
Fig. 11. Measured output pump power versus input pump power at 1025 nm for the double-D
ZBLAN fiber when the pump was launched in the core and cladding, along with the
theoretical fit (solid red curve) and the expected outcome of the measurement had the
pump energy filled the entire cladding uniformly (dashed purple curve). 84
Fig. 12. Experimental results and theoretical predictions for the double-D cladding-doped ZBLAN
fiber. (a) Temperature change as a function of the pump power at the measurement
location, with and without a mode-mixing fiber at the input. (b) Comparison between the
observed cooling and the cooling that was expected had the pump energy filled the entire
cladding uniformly. 85
Fig. 13. Experimental results for the octagonal cladding-doped ZBLAN fiber. (a) The measured
output pump power as a function of input pump power (blue crosses) and a comparison
with the model (solid red curve). (b) Temperature change in the fiber as a function of the
pump power (blue crosses) at the measurement location and the model prediction using
the filling factor of 38% inferred from the fit in (a) (sold red curve). The purple dashed
curves are the predicted behaviors had the pump filling ratio been 100%. 85
Chapter 6
Fig. 1. Temperature change as a function of (a) pump power and (b) absorbed pump power per
unit length (both at the location of the temperature measurement) for two identical fibers
that differ only in their dopant concentration. 90
Fig. 2. The experimental setup used to measure temperature changes in the core-pumped Yb-
doped silica fiber. As the temperature in the doped fiber and the slow-light FBG changes,
the spectral shift induced in the FBG is interrogated by a probe laser tuned to one of the
resonance peaks of the FBG. 93
Fig. 3. Temperature evolution (red curve) over time in the silica fiber as the pump was turned on
and off (blue curve). 95
Fig. 4. Temperature measurements (crosses) and fits (solid curves) plotted as a function of the
pump power at the FBG sensor location for the nanoparticle fibers. 95
xvi
Fig. 5. Temperature measurements (crosses) and fits (solid curve) as a function of absorbed pump
power per unit length at the sensor location for the silica and nanoparticle fibers. (a)
Expanded view; (b) same data zoomed in near the origin. 96
Fig. 6. Temperature measurements (square points) and fits (solid curves) as a function of pump
power (left) and pump power absorbed per unit length (right) for the silica fiber and fibers
with network modifiers. 97
Fig. 7. Measured temperature change as a function of absorbed pump power per unit length for
all tested fibers described in Section 3. 98
Fig. 8. Concentration of Yb and the co-dopants in the silica fiber as a function of radial distance
from the center of the core. 100
Fig. 9. Example measurement of the temperature evolution over time in the cooled Yb-doped
silica fiber as the pump is abruptly was turned on 20 seconds after the start of the
measurement. The gray sections indicate the portions of data that were averaged to
calculate the steady-state temperature drop of the fiber. 101
Fig. 10. Measured dependence of the Yb-doped silica fiber temperature on pump power absorbed
per unit length at the location of the measurement for three pump wavelengths, along with
fits from our ASF model. 103
Fig. 11. Measured temperature change as a function of pump wavelength in the Yb-doped silica
fiber (with one standard deviation of uncertainty) and the dependence predicted by the
model in Chapter 3 [12]. An optimum cooling wavelength exists around 1037 nm due to
the competition between pump absorption strength and the energy difference between the
pump and the ASF. 104
Fig. 12. Measured dependence of temperature change on pump power absorbed per unit length at
the location of the measurement (at 1030 nm) for all six fibers, along with fits using the
ASF-cooling model. 107
Fig. 13. The relationship between the absorptive loss measured at 1380 nm and the absorptive loss
at 1030 nm obtained from fitting the model of Chapter 3 [12] to the temperature
measurements of Fig. 12. 109
Fig. 14. Measured temperature change of Fiber 1 as a function of pump power at the location of
the measurement for three pump wavelengths, along with fits from the ASF model
described in Chapter 3 [12]. 111
Chapter 7
Fig. 1. Results from cut-back measurements performed on the Yb-doped fiber at 1040 nm and
1064 nm. The measured output power Pout is plotted as a function of pump power Pin
launched into the fiber for each wavelength, and the data points are fitted to a model of
saturated absorption in a fiber (equation 1). 116
xvii
Fig. 2. Temperature change as a function of 1040-nm pump power at the measurement location
for the Yb-doped silica fiber presented in this work. The data is fit to our model of ASF
cooling in a fiber [7] to infer the absorptive loss and critical quenching concentration. 118
Fig. 3. Experimental set-up used to measure the temperature change in the Yb-doped silica fiber
amplifier. 119
Fig. 4. Measured temporal trace of the temperature change in the Yb-doped silica fiber as the
1040-nm pump and 1064-nm seed are sequentially turned on, launching 1.64 W and
3 mW in the fiber core, respectively. 120
Fig. 5. Average measured temperature change (n = 3) at seven locations along a 2.74-m silica
fiber amplifier for three different pump powers at 1040 nm. The fiber is core-pumped to
create gain at 1064 nm for the 3-mW seed. 121
Fig. 6. Measured temperature change versus position along the fiber amplifier, and simulated
dependencies using the model based on [8] for (a)-(c) a 2.74-m and (d) a 4.35-m amplifier
fiber. 122
Fig. 7. Measured (red crosses) and simulated (red curve) small-signal gain at 1064 nm for a 3-mW
seed as a function of input pump power at 1040 nm into a 2.74-m silica fiber amplifier,
and the associated temperature change along the length of the amplifier as predicted by
the model based on the one described in Chapter 4 [8] (solid blue curve). 123
Chapter 8
Fig. 1. The cooled Yb-doped silica fiber laser and the experimental setup used to measure
temperature changes along the fiber. 127
Fig. 2. Temporal trace of the temperature change recorded 19 cm from the output end of the Yb-
doped silica fiber, core-pumped at four powers of 1040-nm light, represented pictorially
by the red curve. 128
Fig. 3. The laser output power measured as a function of the launched pump power, along with a
linear fit using a model of cooled fiber laser. The color gradient is a pictorial
representation of the average temperature change along the length of the gain fiber. 130
Fig. 4. Average measured temperature change at eight locations along a 2.64-m silica fiber laser
for four different output powers at 1065 nm. 131
Fig. 5. Average temperature change along the fiber laser and the optical-to-optical laser efficiency,
both measured as a function of laser output power, along with their associated fits. 132
xviii
Chapter 1: Introduction
1. Overview
Since the first laser was demonstrated in 1960, extensive research across the globe has resulted in an
enormous wealth of laser applications ranging from mundane every-day devices to highly specialized
research. Lasers with continuous-wave output powers greater than 1 MW have been demonstrated [1],
enabling countless industrial and military applications such as laser welding, laser cutting, energy research
[2], laser fusion [3], and laser-based directed-energy weapons. The use of lasers has also been widely adopted
by the medical field for cancer treatment [4], corrective eye surgeries and other ophthalmologic applications
such as retinal re-attachment [5], dental care [6], cauterization [7], and skin lesion identification [8]. Lasers
have also enabled numerous scientific discoveries such as the recent detection of gravitational waves [9].
Other scientific applications include light-based radar (LIDAR) [10], the manipulation of individual
molecules with optical tweezers [11], optical fiber sensors [12], and the determination of a molecule’s
vibrational modes through the use of Raman spectroscopy [13]. Many of these applications require the laser
to have superior beam-quality, an ultra-stable power and wavelength, and a narrow linewidth, especially
optical sensors, which often aim to detect changes in optical path lengths of the order of 10 parts per billion
of a wavelength [12].
One of the biggest challenges limiting further power scaling and improvements in laser temporal and
spatial quality is the unavoidable internal heating introduced by the laser’s quantum defect. The quantum
defect is the difference between the pump and laser photon energies, which is converted into heat through
nonradiative relaxation processes within and between electronic manifolds of the laser material. In Yb3+ for
example, which has one of the smallest quantum defects of the rare-earth ions used in solid-state lasers, it is
in the range of 4% to 8%, depending on the host [14,15]. This effect is larger in most other laser materials
[16]. Even in a relatively low-power 1-W Yb-doped fiber laser, the quantum defect is still large enough to
Internal heating in a laser often has substantial deleterious effects [17-22]. Temperature variations induce
instabilities in the laser frequency, which result in a broadening of the laser linewidth (loss of temporal
coherence) or equivalently an increase in frequency noise [18]. These two limitations have serious negative
1
impacts on many applications. LIDAR, for example, requires linewidths less than a 100 kHz [10]. Also, in
fiber sensors utilizing a resonator, such as a slow-light fiber Bragg grating [19] or a distributed feedback laser
[20], this laser-frequency jitter is converted to noise, which places a hard limit on the resolution of the sensor.
In high-power lasers, these same thermal effects also limit the output power through the onset of transverse
mode instability caused by the generation a thermal long-period index grating in the core of the active fiber,
which results in severe stochastic distortions of the shape of the laser output mode [21]. In the most extreme
case, heating will fracture or melt the laser’s gain medium, as observed for example in fiber fuses [22]. In
short, thermal effects hinder the pursuit of power scaling and the development of fiber lasers with ultra-stable
mean wavelength and/or sub-Hz linewidths, qualities of critical importance to ultra-high-precision metrology
applications.
Currently, two main techniques are used in commercial and research lasers to remove this internal heat:
1) water cooling, and 2) thermoelectric cooling. Laser systems that utilize water cooling circulate water
around the gain element, often using a primary closed cooling loop of pure water to do so, and either a
conventional water cooler or a secondary open loop of tap water to cool this primary loop. While they are
comparatively energy efficient and work generally well, they are cumbersome, they are prone to leaking,
they add significant bulk to the laser, and they induce vibrations in the gain medium that add noise to the
resonance frequencies of the laser cavity and degrade the spectral and spatial quality of the output beam.
Thermo-electric coolers, based on the Peltier effect [23], are much smaller and generally vibration-free, but
they tend to cool the fiber asymmetrically, which creates undesirable thermal gradients.
Anti-Stokes fluorescence (ASF) cooling provides a compact, vibrationless solution. To extract heat from
a rare-earth-doped sample, the active ions are optically pumped at a photon energy that is smaller than the
average photon energy of the spontaneous emission [24]. To satisfy the Boltzmann distribution, the excited
electrons must then acquire energy from the phonon bath. As the excited electrons radiatively relax down to
the ground state, they emit fluorescence that escapes radially from the sample and carries this additional
thermal energy out of the system, cooling the material. For a laser cooled with ASF, the cooling system is
integrated into the gain medium, adding no additional bulk to the laser. Also, since the system is fully optical,
there are no moving parts. This not only eliminates harmful vibrations, but also prolongs the lifetime of the
device, minimizing maintenance time and cost. While this solution results in a lower optical-to-optical
2
efficiency [25] compared to the current cooling systems [26], lasers cooled by ASF are expected to exhibit
superior frequency and power stability, and to be invaluable to applications that involve metrology to the
highest precision. ASF cooling has also been proposed for numerous other applications, including cooled
detectors for reduced thermal noise, cooled reference cavities for ultra-stable lasers, and dark current
ASF cooling was first proposed by Peter Pringsheim in 1929 [27], but many material and technological
advancements were needed before its first successful demonstration in 1995 [28]. In this experiment, a bulk
sample of Yb-doped ZBLANP was cooled from room temperature by 0.3 K. Since then, ASF cooling has
been explored in many hosts and trivalent rare-earth ions, including a number of Yb-doped hosts (ZBLAN
[29-33], CNBZn [34], BIG [31,35], YAG [36-38], YLF [39-41], and numerous other crystals [41-43]), Er-
doped hosts (CNBZn [44,45] and KPb2Cl5 [44,45]), Tm-doped hosts (ZBLANP [46-48], BYF [49], and YLF
[50]), and Ho-doped YLF [50]. Since cooling scales with doped area, most experiments were performed in
bulk samples to maximize the achievable temperature change. Most samples were also placed in a vacuum
chamber to minimize convective heat transfer between the sample and air. With these strategies, cryogenic
temperatures (down to 91 K) were achieved in 2015 using a highly purified YLiF4 crystal highly doped with
Yb3+ [51]. For Tm-doped samples, the most amount of cooling was demonstrated in ZBLANP (up to 24 K
below room temperature) [48] and 3 K of cooling has been demonstrated in BYF crystal samples [49]. Er-
doped and Ho-doped samples have exhibited the least amount of cooling: 0.7 K of cooling has been achieved
in Er-doped KPb2Cl5 [45] and 0.1 K of cooling has been achieved in Ho-doped YLF [50]. A few experiments
have been performed with fibers resulting in temperature changes as large as -65 K in short sections of Yb-
doped ZBLANP fibers [52], and -30 K in Tm-doped tellurite fibers [53], both in a vacuum. All fibers were
multimode, which makes it easier to achieve lower absolute temperatures. None of the work with fibers used
oxide materials.
At the beginning of this thesis work, only one optically cooled radiation-balanced laser (RBL) had been
reported [54]. In this demonstration, a 3x120 mm rod of highly doped Yb:YAG crystal was end-pumped with
an array of high-power Yb-doped silica fiber lasers at 1030 nm, simultaneously inducing lasing at 1050 nm
and ASF cooling. Since the extracted heat is proportional to the doped area and the gain medium had a
relatively large transverse dimension (3 mm) [54], the pump beam also had a large transverse dimension and
3
was able to extract significant heat. This laser produced 80-W of laser output power while maintaining
radiation-balanced operation (i.e., zero average temperature change along the crystal). In 2019, a second RBL
was demonstrated with a Yb:YAG disk [25]. The disk was pumped at 1030 nm to create an RBL with 1 W
of output power at 1050 nm. To power-scale the output to the kilowatt level, a scheme was proposed in which
several disks are placed in series in a single cavity. Several theoretical papers also presented models for
radiation-balanced operation in semiconductor lasers [55], fiber lasers [56-59], and fiber amplifiers [60,61].
These models were instrumental in elucidating the need for low-loss materials [59,60], proposed insightful
designs for integrating ASF cooling into the device [55-58,61], and established cooling dependencies on the
pump wavelength and geometry of the gain medium [59]. None of these simulated devices have been
demonstrated in the laboratory. For fibers devices, this was partly due to the limited heat that can be removed
from a fiber because of the small volume of the doped core [62]. Also, until recently, cooling in fibers had
been primarily limited to fluorides [52], the only fiber host known to offer both high quenching-free rare-
earth concentrations and low residual absorptive loss. Concentration quenching, as described further on and
throughout this thesis, induces internal heating and overwhelms ASF cooling. Cooling in silica, by far the
most ubiquitous and versatile fiber laser host, was thought to be highly improbable due the high level of
quenching typically associated with this host. Yet, given the commercial dominance of silica, it was critical
to achieve cooling in this host before conceiving a practical radiation-balanced fiber laser or amplifier.
In short, a majority of the prior work pertaining to ASF cooling has been performed with bulk crystals
placed in a vacuum. There had been very few experimental demonstrations of ASF cooling in fibers, the most
versatile laser material. In addition, the experimental fiber studies were limited to Yb-doped fluorozirconate
fibers. While these fibers are uniquely valuable for specific kinds of lasers such as visible and mid-IR lasers,
they do not have the immense breadth of properties and applications of silica fiber lasers and amplifiers. In
this thesis, we sought to significantly expand the work done on ASF cooling in fibers. In particular, we
focused on the exploration of ASF cooling in silica and ZBLAN fibers doped with trivalent ytterbium with
We started this work by developing a complete model of ASF cooling in a fiber [62]. This provided us
with a sound theoretical background to optimize the fiber and pump parameters for cooling and later analyze
our experimental data to infer fiber parameters that are relevant to cooling. The model is also a valuable tool
4
now available to all researchers in the field to predict the amount of cooling in a very broad range of fibers,
single-mode or multimode, doped with any ion that behaves like a quasi-two-level laser, and with any host
composition, amorphous or crystalline. The model was also expanded to simulate ASF cooling in radiation-
balanced fiber lasers (RBFLs) and amplifiers (RBAs) [63]. This theoretical backbone was instrumental in
providing us with the understanding that we needed to create these devices [64,65].
Our experimental work started with the first-ever demonstration of ASF cooling in a fiber at atmospheric
pressure [66,67]. The initial demonstrations were with single-mode and multimode Yb-doped ZBLAN fibers,
but significant material-science breakthroughs ultimately lead to cooling in Yb-doped silica fibers [68-70].
This achievement was enabled by the significant progress made by our collaborators at Clemson University
(John Ballato), University of Illinois (Peter Dragic), and Mid Sweden University (Magnus Engholm). They
were instrumental in the fabrication and characterization of the silica fibers that cooled. At the onset of this
work, we expected the first radiation-balanced devices to be made with ZBLAN fiber, but these significant
advances in the synthesis of highly doped silica fibers enabled our first RBFL [65] and RBA [64] to be made
with silica, a considerably more versatile and easier to use laser medium. This parallel development was
In Chapter 3, we start by presenting a comprehensive model that quantifies analytically and numerically
the heat that can be extracted by ASF from a fiber doped with a quasi-two-level laser ion [62]. This model is
used to investigate the effects on cooling of all relevant fiber and pump parameters, as well as amplified
spontaneous emission. Simulations of a typical Yb-doped ZBLANP single-mode fiber show that for short
enough fibers the heat extraction is relatively uniform along the fiber length. This theoretical work also
established the existence of an optimum pump wavelength and power that maximizes the heat extracted per
unit length. For a pump launched at this optimum power, the coolest point is at the fiber input end. At higher
powers, the coolest spot moves further down the fiber. The total heat extracted from a fiber, a metric
important for payload cooling, depends on the fiber absorptive loss, the pump wavelength, and the pump
power. Simple expressions are derived to predict the optimum dopant concentration that maximizes heat
extraction and the maximum tolerable absorptive fiber loss above which cooling is unobtainable. In a fiber
5
with negligible residual absorption, the cooling efficiency in a typical Yb-doped ZBLANP fiber is predicted
to be 3.7%. In the modeled fiber, it is reduced to 1.7% in part by concentration quenching, but mainly due to
the fiber absorptive loss (∼15 dB/km). Since the total extracted heat increases linearly with core radius and
dopant concentration (up to a limit determined by concentration quenching), highly doped multimode fibers
In Chapter 4, we build upon this model to simulate RBFLs in which the fiber core and cladding are both
doped with a quasi-two-level laser ion like Yb3+, and both the core and the cladding are pumped [63]. With
this model, we show that an RBFL made of silica can produce substantial output powers. Bidirectional
pumping is found to reduce the average temperature of the laser, enabling higher output powers while
maintaining radiation-balanced operation. For a large-mode-area silica fiber doped with Yb in the core and
cladding, simulations predict that output powers as large as 115 W can be achieved by bidirectionally
pumping the doped cladding. This is not only slightly higher than for a Yb-doped ZBLAN fiber with the
same dimensions, but the temperature gradient in the silica fiber is also about half as large. Since silica is the
most common host in fiber lasers, these predictions are very promising for the near-term realization of
practical RBFLs.
In Chapter 5, we begin our experimental work and report the first-ever demonstration of optical cooling
in a fiber at atmospheric pressure [66,67]. Using the specialized slow-light fiber Bragg grating temperature
sensor, -5.2 mK and -0.65 K were measured in a single-mode (1% YbF3) and multimode (3% YbF3) ZBLAN
fiber with respective cooling efficiencies of 2.2% and 0.90%, respectively. Fitting our model to the measured
temperature change dependence on the pump power per unit length validates the model and allows us to infer
the fibers’ absorptive loss and quenching lifetime, key parameters that are scarce in the literature. These
values are necessary for accurate cooling predictions and will aid in the future development of fibers for
application in optical coolers and radiation-balanced lasers. We also investigate the cooling capabilities of
the cladding-doped fibers modeled Chapter 4 [71]. The results indicate that sufficient mode-mixing within
In Chapter 6, we explore several techniques to reduce concentration quenching in silica and ultimately
succeed in finding a composition that allowed for the first-ever demonstration of cooling in a silica fiber [68-
70,72]. The fiber has a 21-μm diameter core doped with 2.06 wt.% Yb and co-doped with Al2O3 and F to
6
increase the critical quenching concentration by a factor of 16 over the largest reported values for Yb-doped
silica. Temperature changes up to -50 mK were measured with 0.33 W/m of absorbed pump power per unit
length at 1040 nm [68,70]. To maximize cooling in silica, we then investigated how fiber composition, core
size, and OH contamination influence cooling performance [69]. Six Yb-doped silica fibers were studied to
quantify the influence of these parameters. The best fiber cooled by -70 mK with only 170 mW/m of absorbed
pump power at 1040 nm, which corresponds to twice as much heat extracted per unit length compared to the
first silica fiber that we cooled [68,69]. This new fiber has an extremely low OH loss and a higher Al
concentration (2.0 wt.% Al), permitting a high Yb concentration (2.52 wt.% Yb) without incurring significant
quenching. The measured dependency of the temperature change on pump power is in excellent agreement
with the predictions of our model [62], and reflect the fiber’s groundbreaking quality for radiation-balance
fiber lasers. We also find a strong correlation between the absorptive loss responsible for heating and the loss
In Chapter 7, we capitalize on this breakthrough and create the first radiation-balanced fiber amplifier
[64]. The amplifier was core-pumped with 1040-nm light to create ASF cooling and gain in the core at
1064 nm. Temperature measurements were taken at multiple locations along the amplifier. A 4.35-m fiber
pumped with 2.62 W produced ~17 dB of gain while the average fiber temperature remained slightly below
room temperature. The temperature profiles and gains measured with different pump powers and lengths are
fundamental step toward the creation of ultra-stable lasers necessary to many applications, especially low-
In Chapter 8, we put everything together to create the first radiation-balanced silica fiber laser [65]. The
cavity was constructed by splicing high and low reflectivity FBGs to the ends of the Yb-doped silica gain
fiber. The laser was core-pumped at 1040 nm and lased at 1065 nm, resulting in a slope efficiency of 41%
and a threshold power of 1.07 W. Temperature measurements were taken at eight locations along the 2.74-m
long gain fiber and radiation-balanced operation was measured for 105 mW of output power.
7
3. References
1. J. R. Albertine, "Recent high-energy laser system tests using the MIRACL/SLBD", Proc. SPIE 1871,
229 (1993).
2. S. Atzeni, A. Schiavi,J. J. Honrubia, X. Ribeyre, G. Schurtz, Ph. Nicolaï, M. Olazabal-Loumé, C. Bellei,
R. G. Evans, and J. R. Davies, "Fast ignitor target studies for the HiPER project," Phys. Plasmas. 15,
056311 (2008)
3. H. Moltz, The physics of laser fusion, London and New York: Academic Press, 1979.
4. E. Schena, P. Saccomandi, and Y. Fong, "Laser ablation for Cancer: Past, Present and Future," J. Funct.
Biomater 8, 1 (2017).
5. J. M. Wilkinson, E. W. Cozine, and A. R. Khan, "Refractive eye Surgery: Helping Patients Make
Informed Decisions About LASIK," Am Fam Physician. 95, 637 (2017).
6. C. M David and P. Gupta, "Lasers in dentistry: A review," Int. J. Adv. Health Sci. 2, 7 (2015).
7. S. Wanga, Y. He, Y. Zhang, J. Zhang, R. Shah, G. Feng, X. Li, W. Ge, Y. Liu, Y. Guo, H. Liu, J. Tai,
and X. Ni, "CO2 laser cauterization approach to congenital pyriform sinus fistula," J. Pediatr. Surg. 57,
1313 (2018).
8. E. L.Tanzi, J. R. Lupton, and T. S. Alster, "Lasers in dermatology: Four decades of progress," J. Am.
Acad. Dermatol. 49, 1 (2003).
9. P. B. Abbott, et al., "Observation of gravitational waves from a binary black hole merger," Phys. Rev.
Lett. 116, 061102-1 (2016).
10. R. T. H. Collis, "Lidar," Apl. Opt. 9, 1792 (1970).
11. A. Ashkin, J. M. Dziedzic, J. E. Bjorkholm, and S. Chu, "Observation of a single-beam gradient force
optical trap for dielectric particles," Opt. Lett. 11, 288 (1986).
12. G. Rajan, Optical fiber sensors: Advanced techniques and applications, Boca Raton, FL: CRC Press,
2015.
13. N. Jiang, E. T. Foley, J. M. Klingsporn, M. D. Sonntag, N. A. Valley, J. A. Dieringer, T. Seideman, G.
C. Schatz, M. C. Hersam, and R. P. Van Duyne, "Observation of multiple vibrational modes in Ultrahigh
Vacuum Tip-Enhanced Raman Spectroscopy Combined with Molecular-Resolution Scanning Tunneling
Microscopy," Nano Lett. 12, 5061 (2012).
14. K. Lu and N. K. Dutta, "Spectroscopic properties of Yb-doped silica glass," J. Appl. Phys. 91, 576
(2002).
15. P.-H. Haumesser, R. Gaumé, B. Viana, E. Antic-Fidancev, and D. Vivien, "Spectroscopic and crystal-
field analysis of new Yb-doped laser materials," J. Phys.: Condens. Matter 13, 5427 (2001).
16. R. Epstein and M. Sheik-Bahae, Optical Refrigeration: science and applications of laser cooling of
solids, Weinheim, GE: Wiley-VCH, 2009, pp. 38.
17. M. K. Davis and M. J. F. Digonnet, "Thermal effects in doped fibers," J. Lightwave Technol. 6, 1013
(1998).
18. N. A. Brilliant and K. Lagonik, "Thermal effects in a dual-clad ytterbium fiber laser," Opt. Lett. 26, 1699
(2001).
19. G. Skolianos, A. Arora, M. Bernier, and M. J. F. Digonnet, "Photonics sensing at the thermodynamic
limit," Opt. Lett. 42, 2018 (2017).
20. S. B. Foster, G. A. Cranch, J. Harrison, A. E. Tikhomirov, and G. A. Miller, "Distributed feedback fiber
laser strain sensor technology," J. Light. Technol. 35, 3514 (2017).
21. C. Jauregui, C. Stihler, and J. Limpert, "Transverse mode instability," Adv. Opt. Photonics 12, 429
(2020).
22. R. Kashyap and K. J. Blow, "Observation of catastrophic self-propelled self-focusing in optical fibres,"
Electron. Lett. 24, 47 (1988).
23. T. M. Tritt, "Thermoelectric materials: Principles, structure, properties, and applications," in
Encyclopedia of Materials: Science and Technology, Elsevier (2002).
24. M. Sheik-Bahae and R. I. Epstein, "Optical refrigeration," Nat. Photon. 1, 693 (2007).
25. Z. Yang, J. Meng, A. Albrecht, and M. Sheik-Bahae, "Radiation-balanced Yb:YAG disk laser," Opt.
Express 27, 1392 (2019).
26. J. Hecht, "Photonic frontiers: High-efficiency optical pumping: ‘Going green’ cranks up the laser
power," Laser Focus World, Apr. 13, 2016.
8
27. P. Pringsheim, "Zwei bemerkungen über den unterschied von lumineszenz- und temperaturstrahlung,"
Z. Phys. 57, 739 (1929).
28. R. I. Epstein, M. I. Buchwald, B. C. Edwards, T. R. Gosnell, and C. E. Mungan "Observation of laser-
induced fluorescent cooling of a solid," Nature 377, 500 (1995).
29. B. C. Edwards, J. E. Anderson, R. I. Epstein, G. L. Mills, and A. J. Mord, "Demonstration of a solid-
state optical cooler: an approach to cryogenic refrigeration," J. Appl. Phys. 86, 6489 (1999).
30. A. Rayner, M. E. J. Friese, A. G. Truscott, N. R. Heckenberg, and H. Rubinsztein-Dunlop, "Laser cooling
of a solid from ambient temperature," J. Mod. Opt. 48, 103 (2001).
31. M T. Murtagh, G. H. Sigel Jr., J. C. Fajardo, B. C. Edwards, and R. I. Epstein, "Laser-induced fluorescent
cooling of rare-earth-doped fluoride glasses," J. Non-Cryst. Solids 253, 50 (1999).
32. A. Rayner, N. R. Heckenberg, and H. Rubinsztein-Dunlop, "Condensed-phase optical refrigeration," J.
Opt. Soc. Am. B 20, 1037 (2003).
33. B. Heeg, M. D. Stone, A. Khizhnyak, G. Rumbles, G. Mills, P. A. and DeBarber, "Experimental
demonstration of intracavity solid-state laser cooling of Yb3+:ZrF4−BaF2−LaF3−AlF3−NaF glass," Phys.
Rev. A 70, 021401 (2004).
34. J. Fernández, A. Mendioroz, A. J. Garcı́a, R. Balda, J. L. Adam, and M. A. Arriandiaga, "On the origin
of anti-Stokes laser-induced cooling of Yb3+-doped glass," Opt. Mater. 16, 173 (2001).
35. J. Fernández, A. Mendioroz, A. J. García, R. Balda, and J. L. Adam, "Anti-Stokes laser-induced internal
cooling of Yb3+- doped glasses," Phys. Rev. B 62, 3213 (2000).
36. S. R. Bowman and C. E. Mungan, "New materials for optical cooling," Appl. Phys. B 71, 807 (2000).
37. R. I. Epstein, J. J. Brown, B. C. Edwards, and A. Gibbs, "Measurements of optical refrigeration in
ytterbium-doped crystals," J. Appl. Phys. 90, 4815 (2001).
38. E. S. de Lima Filho, G. Nemova, S. Loranger and R. Kashyap, "Laser-induced cooling of a Yb:YAG
crystal in air at atmospheric pressure," Opt. Express 21, 24711 (2013).
39. S. Bigotta, "Laser cooling of solids: new results with single fluoride crystals," Nuovo Cimento B Ser.
122, 685694 (2007).
40. S. Bigotta, A. Di Lieto, D. Parisi, A. Toncelli, and M. Tonelli, "Single fluoride crystals as materials for
laser cooling applications," SPIE 6461, 64610E (2007).
41. D. V. Seletskiy, S. D. Melgaard, S. Bigotta, A. Di Lieto, M. Tonelli, and M. Sheik-Bahae, "Laser cooling
of solids to cryogenic temperatures," Nat. Photon. 4, 161 (2010).
42. R. I. Epstein, J. J. Brown, B. C. Edwards, and A. Gibbs, "Measurements of optical refrigeration in
ytterbium-doped crystals," J. Appl. Phys. 90, 4815 (2001).
43. S. Bigotta, D. Parisi, L. Bonelli, A. Toncelli, A. D. Lieto, and M. Tonelli, "Laser cooling of Yb3+-doped
BaY2F8 single crystal," Opt. Mater. 28, 1321 (2006).
44. J. Fernandez, A. J. Garcia-Adeva, and R. Balda, "Anti-Stokes laser cooling in bulk erbium-doped
materials, Phys. Rev. Lett. 97, 033001 (2006).
45. A. Garcia-Adeva, R. Balda, and J. Fernandez, "Anti-Stokes laser cooling in erbium-doped low phonon
materials, Proc. SPIE 6461, 646102 (2007).
46. C. W. Hoyt, M. Sheik-Bahae, R. I. Epstein, B. C. Edwards, and J. E. Anderson, "Observation of anti-
Stokes fluorescence cooling in thulium-doped glass," Phys. Rev. Lett. 85, 3600 (2000).
47. C. W. Hoyt, M. P. Hasselbeck, M. Sheik-Bahae, R. I. Epstein, S. Greeneld, J. Thiede, J. Distel, and J.
Valencia," Advances in laser cooling of thulium-doped glass," J. Opt. Soc. Am. B 20, 1066 (2003).
48. C. W. Hoyt, W. Patterson, M. P. Hasselbeck, M. Sheik-Bahae, R. I. Epstein, J. Thiede, and D. V.
Seletskiy, "Laser cooling thulium-doped glass by 24 K from room temperature," Trends Opt. Phot. 89,
QThL4 (2003).
49. W. Patterson, S. Bigotta, M. Sheik-Bahae, D. Parisi, M. Tonelli, and R. I. Epstein, "Anti-Stokes
luminescence cooling of Tm3+ doped BaY2F8," Opt. Express 16, 1704 (2008).
50. S. Rostami, A. R. Albercht, M. R. Ghasemkhani, S. D. Melgaard, A. Gragossian, M. Tonelli, and M.
Sheik-Bahae, "Optical refrigeration of Tm:YLF and Ho:YLF crystals," Proc. SPIE 9765, 97650P (2016).
51. S. D. Melgaard, A. R. Albrecht, M. P. Hehlen, and M. Sheik-Bahae, "Solid-state optical refrigeration to
sub-100 Kelvin regime," Sci. Rep. 6, 1 (2016).
52. T. R. Gosnell, "Laser cooling of a solid by 65 K starting from room temperature," Opt. Lett. 24, 1041
(1999).
53. G. Nemova, Ed., Progress Toward Laser Cooling for Thulium-Doped Fibers, 1st ed. Singapore: Pan
Stanford Pub. Pte. Ltd., 2017, ch. 3.
9
54. S. Bowman, S. P. O’Connor, S. Biswal, N. J. Condon, and A. Rosenberg, "Minimizing heat generation
in solid-state lasers," IEEE J. Quantum Electron. 46, 1076 (2010).
55. J. Khurgin, "Radiation-balanced tandem semiconductor/Yb3+:YLF lasers: feasibility study," J. Opt. Soc.
Amer. B 37, 1886 (2020).
56. G. Nemova and R. Kashyap, "Yb-doped fiber laser with integrated optical cooler," Proc. SPIE 7686,
768614-1 (2010).
57. X. Xia, A. Pant, E. J. Davis, and P. Pauzauskie, "Design of a radiation- balanced fiber laser via optically
active composite cladding materials," J. Opt. Soc. Amer. B 36, 3307 (2019).
58. G. Nemova and R. Kashyap, "High-power fiber lasers with integrated rare-earth optical cooler," Proc.
SPIE 7614, 761406-1 (2010).
59. M. Peysokhan, E. Mobini, A. Allahverdi, B. Abaie, and A. Mafi, "Characterization of Yb-doped ZBLAN
fiber as a platform for radiation-balanced lasers," Photon. Res. 8, 202 (2020).
60. E. Mobini, M. Peysokhan, and A. Mafi, "Heat mitigation of a core/cladding Yb-doped fiber amplifier
using anti-Stokes fluorescence cooling," J. Opt. Soc. Amer. B 36, 2167 (2019).
61. G. Nemova and R. Kashyap, "Fiber amplifier with integrated optical cooler," J. Opt. Soc. Amer. B 26,
2237 (2009).
62. J. M. Knall, M. Esmaeelpour, and M. J. F. Digonnet, "Model of anti-Stokes fluorescence cooling in a
single-mode optical fiber,” J. Lightwave Technol. 36, 4752 (2018).
63. J. M. Knall, and M. J. F. Digonnet, "Design of high-power radiation-balanced silica fiber lasers with a
doped core and cladding," J. Lightwave Technol. 39, 2497 (2021).
64. J. M. Knall, M. Engholm, T. Boilard, M. Bernier, and M. J. F. Digonnet, "A radiation-balanced silica
fiber amplifier," arXiv:2103.02698, accepted in Phys. Rev. Lett. (2021).
65. J. M. Knall, M. Engholm, T. Boilard, M. Bernier, P.-B. Vigneron, N. Yu, P. D. Dragic, J. Ballato, and
M. J. F. Digonnet, "A radiation-balanced silica fiber laser," Optica, doi: 10.1364/OPTICA.425115
(2021).
66. J. M. Knall, A. Arora, M. Bernier, S. Cozic, M. J. F. and Digonnet, "Demonstration of anti-Stokes
cooling in Yb-doped ZBLAN fibers at atmospheric pressure," Opt. Lett. 44, 2338 (2019).
67. J. M. Knall, A. Arora, M. Bernier, and M. J. F. Digonnet, "Anti-Stokes fluorescence cooling in Yb-doped
ZBLAN fibers at atmospheric pressure: experiments and near-future prospects," Proc. SPIE 10936,
109360F-1 (2019).
68. J. M. Knall, P.-B. Vigneron, M. Engholm, P. D. Dragic, N. Yu, J. Ballato, M. Bernier, and M. J. F.
Digonnet, "Laser cooling in a silica optical fiber at atmospheric pressure," Opt. Lett. 45, 1092 (2020).
69. J. M. Knall, M. Engholm, J. Ballato, P. D. Dragic, N. Yu, and M. J. F. Digonnet, "Experimental
comparison of silica fibers for laser cooling," Opt. Lett. 45, 4020 (2020).
70. J. M. Knall, P.-B. Vigneron, M. Engholm, P. D. Dragic, N. Yu, J. Ballato, M. Bernier, and M. J. F.
Digonnet, "Experimental observation of cooling in Yb-doped silica fibers," Proc. SPIE 11298, 112980F-
1 (2020).
71. P.-B. Vigneron, J. M. Knall, T. Boilard, M. Bernier, and M. J. F Digonnet, "Observation of anti-Stokes-
fluorescence cooling in a ZBLAN fiber with a Yb-doped cladding," Proc. SPIE 11702, 117020A-1
(2021).
72. J. M. Knall, A. Arora, P. D. Dragic, J. Ballato, M. Cavillon, T. Hawkins, S. Jiang, T. Luo, M. Bernier,
and M. J. F. Digonnet, "Experimental investigations of spectroscopy and anti-Stokes fluorescence
cooling in Yb-doped silicate fibers," Proc. SPIE 10936, 109360G-1 (2019).
10
Chapter 2: Principles of Anti-Stokes Fluorescence Cooling
Before we address anti-Stokes fluorescence, it is useful to first review the more commonly known Stokes
process utilized in rare-earth doped lasers. As an example, consider a gain medium doped with a two-level
laser ion such as Yb, placed in an optical resonator with a resonance at photon energy hνl, equal to the energy
difference between the bottom of the upper manifold (level N21 in Fig. 1) and the top of the lower manifold
(level N14). To induce lasing, the laser is typically pumped at a wavelength that excites the electrons from
approximately the bottom of the lower manifold to approximately the top of the upper manifold (Step 1 in
Fig. 1). On a picosecond timescale, the remaining electrons in the lower manifold and the excited electrons
in the upper manifold then thermalize according to the Boltzmann distribution (Step 2). To this end, the
electrons lose a small amount of energy that excites lattice vibrations in the host (e.g. generating phonons),
and create a population inversion between N21 and N14 at steady state. If the gain medium is pumped with
sufficiently high power so that the gain at frequency νl exceeds the resonator loss, lasing ensues at this
frequency and energy is carried out the system through the emission of light with lower energy than the pump
(Step 3). The heat generated through this process is the difference between the pump photon energy hνp and
Figure 1: The Stokes process dictating laser operation. First, high energy light is used to excite electrons into
the upper manifold. The electrons then thermalize according to the Boltzmann distribution by generating
phonons. Stimulated emission induces lower energy laser light to exit the system.
11
Anti-Stokes fluorescence is essentially the reverse of this process. Lower energy light is used to excite
electrons from approximately the top of the lower manifold to approximately the bottom of the upper
manifold (Step 1 in Fig. 2). Like before, the electrons in the lower and upper manifolds then thermalize
according to the Boltzmann distribution which, in this case, dictates that, instead of generating phonons, the
electrons absorb phonons so that they acquire the energy required to move to higher energy states within
their respective manifold (Step 2). Spontaneous emission ensues and energy is carried out of the system
through fluorescence (Step 3). The average energy of the fluorescence hνf is higher than the energy of the
pump hνp, which means that energy is extracted by ASF. This extracted energy is the difference between
Figure 2: The anti-Stokes fluorescence process. First, lower energy light is used to excite electrons into the
upper manifold. The electrons then thermalize according to the Boltzmann distribution by absorbing phonons.
Spontaneous emission generates higher energy fluorescence that escapes the system.
Since the efficiency of ASF cooling is dictated by the energy difference between the pump and
fluorescence photons, which is a small fraction of the pump energy, ASF cooling extracts relatively small
amounts of power per unit volume [1]. Consequently, any number of undesirable exothermic effects can
incidentally inject heat into the sample and either partially negate or overwhelm the heat extracted by ASF.
The most common mechanisms are absorption of pump and/or fluorescence photons by impurities [2], and
12
concentration quenching [3]. Therefore, it is critical to select materials and develop material synthesis and
fiber fabrication protocols that minimize these mechanisms. However, even if these effects are completely
eliminated, cooling is still limited by its proportional relationship to doped area [1], the simple physical
reason being that the extracted heat is proportional to the volume of gain medium that is optically pumped,
and therefore to the area of the pump beam. This is particularly relevant for ASF cooling in fibers in which
the doped area, typically limited to the fiber core, tends to have a diameter smaller than 200 μm. This results
in maximum temperature changes as low as a few mK when the fiber is single-moded (a core diameter of 5–
10 µm) and cooled at atmospheric pressure [1]. Coupled with the other temperature sensor requirements
There are two types of concentration quenching in Yb-doped materials: diffusion-limited quenching and
fast-diffusion quenching [4,5]. Quenching is detrimental because it (1) generates heat, and (2) reduces the
population inversion (and therefore the cooling efficiency). In diffusion-limited quenching, an excited Yb
ion transfers its energy to an impurity, and nonradiative relaxation ensues (see Fig. 3). This nonradiative
relaxation takes place within the impurity, and typically converts all the energy that was originally stored in
the excited Yb electron into heat. Relaxation within the impurity unfortunately has a short lifetime, since the
impurities are typically a heavy-metal ion or a hydroxyl. So, as soon as the energy is transferred to the
impurity, nonradiative relaxation and the associated generation of phonons occurs, which increases the rate
of heat production (compared to what it would be if the impurity had a long metastable lifetime for example).
The rate of quenching is then dominated by the rate of energy transfer between a Yb ion and an impurity.
This mechanism is diffusion-limited because the probability of energy transfer between two neighboring Yb
ions (i.e., a donor in the excited state donates its energy to an acceptor in the ground state) is comparable to
the probability of energy transfer between a Yb ion and a neighboring impurity, thereby limiting energy
13
Figure 3: Mechanism for diffusion-limited concentration quenching: an excited Yb ion transfers its energy to
an impurity, and nonradiative relaxation ensues.
In fast-diffusion quenching, self-generated quenching centers are formed by the interaction of two or
more Yb ions. This is due to an increase in the electron-phonon coupling associated with an increase in Yb
concentration [5]. This mechanism is labeled as fast-diffusion since the quenching step is typically less
probable than energy diffusion between Yb ions. In both cases, the quenching rate, as represented by the
inverse of the quenching lifetime τq, increases with Yb concentration. However, the contribution of each type
depends on the impurity concentration in the sample. In general, fast-diffusion quenching is only relevant for
extremely high purity samples (mostly crystals), in which diffusion-limited quenching is not dominant [4,5].
Since the Yb-doped silica and ZBLAN fibers considered in this work have measurable impurity contents,
the dominant form of quenching is diffusion-limited. In this regime, increasing the Yb concentration N0
#$& ). #
%
𝜏! (𝑁" ) = '*+ (' *+ -) 0 (1)
'() ,' /
where τrad is the radiative lifetime and τnr is the nonradiative lifetime of the excited state due to multiphonon
relaxation to the ground state. Nc is the critical quenching concentration, a material-dependent parameter for
which the average distance between dopant ions is such that the probability of nonradiative energy transfer
is the same as the probability for photon emission [4]. Defining t0 = 1/(trad-1 + tnr-1) as the total lifetime of
the Yb ions for sufficiently low Yb concentrations (for which tq-1 ≈ 0), and rearranging (1) gives an
14
'.
𝜏(𝑁" ) = 1 4 2 (2)
*(23+ 4.,
/
This dependence is plotted in Fig. 4 for nominal Yb-doped silica (Nc = 0.75x1026 m-3) [6] and ZBLAN
(Nc = 3.54x1027 m-3) [7] samples. The plot for the Yb-doped silica was generated for pure silica fabricated
into a fiber preform [6]. This represents the performance of a baseline Yb-doped silica fiber in which no co-
dopants have been added to reduce quenching. Below a certain concentration, for both materials the lifetime
remains relatively constant and is primarily radiative, a necessary condition for significant ASF cooling. As
the concentration increases, the lifetime decreases rapidly and reaches half its low-concentration value (t0/2)
for a concentration of 0.84xNc (obtained from (2)). The total lifetime is then equally due to radiative and
quenching components. The relationship between the lifetime and cooling potential for materials with long
non-radiative lifetimes (including Yb-doped silicates and Yb-doped ZBLAN) can be quantified by the
radiative quantum efficiency [8], defined as ηq = tq/(tq + trad). For convenience, and to follow the terminology
used in the field of laser cooling, in the rest of this thesis this quantity is referred to as quantum efficiency.
This equation describes the fraction of the pump power absorbed by the Yb ions that partakes in radiative
transitions (and thus cooling). At low concentrations, tq ≫ trad and the quantum efficiency is ~1. To achieve
significant ASF cooling in Yb-doped materials, the quantum efficiency must be greater than 98% [9], which
corresponds (using (2)) to a maximum dopant concentration of 0.15xNc. For ZBLAN, this concentration is
5.20x1026 m-3, while in pure silica the maximum tolerable concentration is 50 times lower (1.10x1025 m-3).
This is illustrated by Fig. 4, in which the total lifetime for silica drops to half its value at a Yb concentration
that is 50 times lower than for ZBLAN. Since cooling scales with the Yb concentration, the potential for
cooling in silica is severely constrained, and even a small amount of heating due to impurity absorption will
overwhelm the limited cooling. Therefore, to achieve significant ASF cooling in silica-based hosts, it is
critical to establish a glass composition and fabrication protocol that increases the critical quenching
15
Figure 4: Effect of concentration quenching on the total upper-state lifetime of Yb for ZBLAN (red curve)
and silica (blue curve), plotted as a function of Yb concentration.
Fiber loss originates from absorption by impurities, and scattering. Scattering loss is primarily caused
by variations in the material density, compositional fluctuations, and structural defects (mostly core-cladding
boundary irregularities) in the fiber [10]. While this type of loss leads to pump attenuation, it is small and
also not a source of heating since the photons are merely scattered out of the fiber and no phonons are
generated. Absorption by impurities, however, does generate significant heat and needs to be minimized in
fibers designed for ASF cooling. Harmful impurities are primarily composed of OH-, heavy metal ions (Fe2+,
Cu2+, etc.), and unwanted rare-earth ions [11]. These impurities directly absorb the pump, laser, and/or
spontaneous emission signals propagating in the fiber, then they decay non-radiatively, generating phonons
in the process. Impurity absorption is minimized by using high-purity precursors to fabricate the sample
[12,13]. The chemical vapor deposition methods employed to fabricate silica-based optical fibers is ideal in
this regard because they afford near-intrinsic levels of purity and, hence, very low impurity levels.
Since a majority of the impurities decay on a nanosecond timescale, their power absorption is nominally
unsaturable. As will be discussed in Chapter 3, this leads to a linear relationship between pump power and
heating due to impurity absorption. In contrast, the heat generated in the fiber from concentration quenching
is always a fixed percentage of the pump power absorbed by the Yb ions. Since the pump absorption by Yb
is saturable (due to the much longer, millisecond lifetime), so too is the heat generated by concentration
16
quenching. Figure 5 illustrates the influence of absorptive loss and concentration quenching on the
dependency between temperature change and pump power. Using the model on ASF cooling in a fiber
described in Chapter 3, the temperature change was simulated as a function of pump power for four different
cases. In the base case (blue curve), the fiber is simulated to have no absorptive loss or concentration
quenching. This results in a temperature that initially decreases linearly with pump power, but then
asymptotically approaches a certain temperature at higher pump power as the Yb absorption (and thus
cooling) saturates. The addition of quenching (yellow curve) causes the magnitude of the baseline curve to
decrease but the general shape of the dependence is still monotonic. In contrast, adding absorptive loss to the
baseline (green curve) has minimal influence for low pump powers but, once the cooling saturates, it induces
a linear increase in temperature as additional power induces heating but no more cooling. As expected,
simulating the fiber to have both absorptive loss and concentration quenching (red curve) results in a
Figure 5: Effect of absorptive loss and concentration quenching on the dependence on pump power of the
temperature change induced by ASF.
17
2.3 Custom Fiber Temperature Sensor
When a sample to be cooled by ASF is placed in air, its temperature drop can be small due to the heat
intake through the convective flow of the surrounding air. This is especially true in a fiber, which has a large
surface-to-volume ratio (less cooling, more convective heating). As a result, most ASF-cooling experiments
are performed in vacuum to minimize convective heating and achieve larger negative temperature changes
that are easier to measure [12,14-16]. From a practical standpoint, however, for ASF cooling to be broadly
adopted by the industry in commercial lasers and amplifiers, it is also important to demonstrate cooling in
air. To measure the exceedingly small temperature changes predicted for the single-mode fiber, we needed a
temperature sensor that could resolve temperature changes on the order of a few mK, yet still have a large
enough dynamic range to measure the larger temperature changes expected for the multimode fiber, and other
fibers in the future. Additionally, successful measurements required the sensor to be stable over a typical
measurement time (~1 minute) and to have a fast enough response time to track the rapidly changing
temperature of a small object such as a fiber. Also, the sensor should be transparent at the mean ASF
wavelength (~1 μm) so that it does not absorb the fluorescence and heat up. Finally, in order to resolve the
position-dependent temperature change, the sensor must have a longitudinal resolution smaller than
Figure 6: (Taken from [17]) Custom slow-light fiber Bragg grating temperature sensor used in this work.
18
The sensor used in this work is a custom fiber temperature sensor developed in our group by A. Arora
and is illustrated in Fig. 6 [17,18]. The central component of the sensor is a slow-light fiber Bragg grating
(FBG). Like a conventional FBG, it expands or contracts in response to a temperature change, inducing a
spectral shift of ~10 pm/K for a silica fiber [19]. The main difference and improvement over a conventional
FBG temperature sensor is that the FBG is fabricated using femtosecond pulses in order to induce a much
stronger periodic index modulation. As a result, the FBGs, fabricated at Université Laval, exhibit a series of
very narrow resonances located just inside the short-wavelength edge of the bandgap [18]. By interrogating
one of the grating’s slow-light resonances with a 1550-nm tunable probe laser, it is possible to resolve
extremely small wavelength shifts, and therefore very small fiber temperature changes. The FBG is placed
along the side of the cooled Yb-doped fiber, and measures are taken ensure a good thermal contact between
the fibers (see Section 2.4 in Chapter 5 for more details). An enclosure is placed over the setup to reduce
temperature drifts caused by air currents. Since the temperature changes involved are small (≪10 K), at steady
state the surface temperature of the doped fiber is almost identical to the temperature of the fiber core [20].
The sensor measures the change in the surface temperature of the doped fiber induced by ASF and therefore,
the core temperature. Since development of the sensor was not a focus of this dissertation, in-depth
elaboration has been omitted. Further details can be found in [18]. In short, this sensor fulfills the above
requirements and can resolve a temperature change as low as ~2 mK in the ~1 minute that it takes to perform
For lasers with a larger gain-medium cross-section, such as large-mode-area fiber lasers or bulk crystal
lasers, it is reasonable to spatially couple the cooling and lasing. In this configuration, the device is pumped
at an energy that compromises between cooling and amplification, so that both processes can occur
simultaneously, and the cooling cancels out the heating (see Fig. 7). For example, a typical Yb-doped silica
laser is pumped with 976-nm light to maximize absorption of the pump and gain at the signal wavelength,
taken to be 1064 nm for this example. However, to induce ASF cooling, the gain medium needs to be pumped
at a wavelength that is higher than the mean fluorescence wavelength, which is typically around 1005 nm for
19
Yb-doped silica [21]. Therefore, to create a radiation-balanced device, the laser needs to be pumped at an
optimum between 1005 nm and 1064 nm, which turns out to be around 1040 nm [21].
Figure 7: Energy level diagrams for three cases: ideal lasing (left), ideal cooling (middle), and radiation-
balanced lasing (right). To create an RBL, the device needs to be pumped at a photon energy that compromises
between cooling and amplification, so that both processes can occur simultaneously.
To cool a laser with ASF, the cooling and lasing can also occur in separate hosts that are thermally
connected. This method is particularly useful for semiconductor lasers since ASF cooling has yet to be
demonstrated in this medium: cooling can then be carried out in a separate material in thermal contact with
the semiconductor laser [22]. Fiber lasers may also benefit from this decoupling because the doped area of a
single-mode fiber is small, which limits the achievable cooling [1]. For example, cooling can be induced in
a much larger doped cladding to mitigate heating due to lasing in the thermally coupled single-mode core
[23,24,25].
4. References
20
4. F. Auzel, G. Baldacchini, L. Laversenne, and G. Boulon, "Radiation trapping and self-quenching
analysis in Yb3+, Er3+, and Ho3+ doped Y2O3," Opt. Mater. 24, 103 (2003).
5. F. Auzel, "A fundamental self-generated quenching center for lanthanide-doped high-purity solids," J.
Lumin. 100, 125 (2002).
6. P. Barua, E. H. Sekiya, and A. J. Ikushima, "Influences of Yb3+ ion concentration on the spectroscopic
properties of silica glass," J. Non-Cryst. Solids 354, 4760 (2008).
7. J. M. Knall, A. Arora, M. Bernier, S. Cozic, and M. J. F. Digonnet, "Demonstration of anti-Stokes
cooling in Yb-doped ZBLAN fibers at atmospheric pressure," Opt. Lett. 44, 2338 (2019).
8. E. Desurvire, Erbium-Doped Fiber Amplifiers: Principles and Applications, New York, Wiley, pp. 220
(1994).
9. M. Sheik-Bahae, and R. I. Epstein, "Laser cooling in solids," Laser and Photon. Rev. 3, 67 (2009).
10. S. E. Miller, E. A. J. Marcatili, and T. Li, "Research toward optical-fiber transmission systems," Proc.
of IEEE 61, 1703 (1973).
11. M. P. Hehlen, R. I. Epstein, and H. Inoue, "Model of laser cooling in Yb3+-doped fluorozirconate glass
ZBLAN," Phys. Rev. B 75, 144302–1 (2007).
12. S. D. Melgaard, A. R. Albrecht, M. P. Hehlen, and M. Sheik-Bahae, "Solid-state optical refrigeration to
sub-100 Kelvin regime," Sci. Rep. 6, 20380 (2016).
13. S. R. Nagel, J. B. MacChesney, and K. L. Walker, "An overview of the modified chemical Vapor
Deposition (MCVD) Process and Performance," IEEE Trans. Microwave Theory Tech. 30, 305 (1982).
14. T. R. Gosnell, “Laser cooling of a solid by 65 K starting from room temperature,” Opt. Lett. 24, 1041
(1999).
15. T. R. Gosnell and X. Luo, "Laser cooling of a solid by 21 K starting from room temperature," Technical
Digest—European Quantum Electronics Conference (1998).
16. A. Rayner, M. Hirsch, N. Heckenberg, and H. Rubinsztein-Dunlop, "Distributed laser refrigeration,"
Appl. Opt. 40, 5423 (2001).
17. A. Arora, M. Esmaeelpour, M. Bernier, and M. J. F. Digonnet, "High-resolution slow-light fiber Bragg
grating temperature sensor with phase-sensitive detection," Opt. Lett. 43, 3337 (2018).
18. A. Arora, "High-resolution temperature and acoustic pressure sensors utilizing slow-light fiber Bragg
gratings," Ph.D dissertation, Dept. of Elect. Eng., Stanford Univ., Stanford, CA, June 2019. Accessed
on: May 24, 2021. [Online]. Available: https://www-proquest-com.stanford.idm.oclc.org/docview/
2466049752?pq-origsite=gscholar&fromopenview=true
19. A. D. Kersey, M. A. Davis, H. J. Patrick, M. LeBlanc, K. P. Koo, C. G. Askins, M. A. Putnam, and E.
J. Friebele, "Fiber grating sensors," J. Lightwave Technol. 15, 1442 (1997).
20. M. K. Davis, M. J. F. Digonnet, and R.H. Pantell, "Thermal effects in doped fibers," J. Lightwave
Technol. 16, 1013 (1998).
21. J. M. Knall, M. Engholm, J. Ballato, P. D. Dragic, N. Yu, and M. J. F. Digonnet, "Experimental
comparison of silica fibers for laser cooling," Opt. Lett. 45, 4020 (2020).
22. J. Khurgin, "Radiation-balanced tandem semiconductor/Yb3+:YLF lasers: feasibility study," J. Opt. Soc.
Amer. B 37, 1886 (2020).
23. G. Nemova, and R. Kashyap, "High-power fiber lasers with integrated rare-earth optical cooler," Proc.
SPIE 7614, 761406-1 (2010).
24. G. Nemova, and R. Kashyap, "Fiber amplifier with integrated optical cooler," J. Opt. Soc. Amer. B 26,
2237 (2009).
25. P.-B. Vigneron, J. M. Knall, T. Boilard, M. Bernier, and M. J. F Digonnet, "Observation of anti-Stokes-
fluorescence cooling in a ZBLAN fiber with a Yb-doped cladding," Proc. SPIE 11702, 117020A-1
(2021).
21
Chapter 3: Model of Anti-Stokes Cooling in a Yb-Doped Fiber
1. Introduction
Most models of ASF cooling focus exclusively on free-space gain media [1-6]. These studies were
instrumental in establishing the fundamental requirements for ASF cooling on which this thesis work was
founded. [1] and [2] analyze some of the key material requirements and highlight three attributes of a
desirable host: low phonon energy, ultra-pure starting material, and strong absorption by the active ion. [2]
and [3] also consider the detrimental effects of radiative retrapping and how this heating mechanism can be
minimized by strategically selecting the dopant concentration in conjunction with the sample's transverse
size and shape. [4] quantitively assesses the purity requirements by establishing maximum impurity
concentrations for different operating temperatures. The results also determined that Cu2+, Fe2+, Co2+, Ni2+
and OH- are the most problematic impurities. [5] establishes basic figures of merit for cooling and applies
them to Yb-doped materials, particularly crystals and fluoride glass. Since all these studies focused on bulk
media, they justifiably neglected several effects that only become significant for longer sample lengths. In
particular, these models assume that pump power and temperature do not vary with position in the sample.
They also neglect the effects of amplified spontaneous emission (ASE), which is legitimate since the short
sample lengths prohibit the ASE from acquiring significant power. The models also primarily simulate crystal
hosts, which allow for significantly higher dopant concentrations than amorphous hosts. This means that, in
these previous studies, concentration quenching was much less of a concern. As a result, dopant
concentrations up to 30 wt.% Yb were recommended [3]. In contrast, for optical fibers drawn from
amorphous glass, concentrations in this range would induce significant quenching and therefore require
models that place a higher emphasis on this heating mechanism. Some models on cooling in optical fibers
have been published [7-10] but few have been used for parametric studies designed to predict the optimum
parameters for maximum cooling in a fiber. [7] presents a theoretical framework to predict cooling in fibers
that incorporates background loss and absorption saturation effects. In [8], a model was developed to
numerically predict the optimum core size that maximizes the temperature change in a vacuum for a lossless
fiber with a fixed ratio of core to cladding diameters. A majority of the other models were established as a
tool for experimental validation through empirical fitting [9,10]. All of the fiber models neglected the effects
22
of ASE as well as of the mode profiles of the pump and signals.
Clearly, there was a need for a more advanced model of cooling in a doped fiber. Compared with bulk
samples, a few major differences are expected when cooling a fiber. First, since in a fiber the pump is guided,
large excited-state population densities can be sustained over considerably longer lengths, possibly leading
to greater heat extraction. This fiber property also means that the pump and ASE powers have mode profiles
defined by the fiber's core size and numerical aperture. Since the spatial profile of the pump and ASE along
the fiber critically affects their rate of absorption and emission as well as saturation, and therefore the amount
of heat that can be extracted by ASF per unit length, it is critical to take the mode profiles into account to
achieve accurate cooling predictions. Second, since the fiber length may equal many pump-absorption
lengths, both the excited-state population density and the heat extraction can vary significantly along the
fiber. In addition, since the physical cross-section of a fiber is much smaller than that of a typical bulk sample,
heat conduction along the fiber is negligible, and there is no temperature equalization along the fiber. As a
result of these last two effects, significant thermal gradients are expected along a cooled fiber, including hot
and cool areas. Third, longer lengths produce higher gains and possibly sufficient ASE to affect the cooling.
These significant differences point to the need for a model that accounts for these effects to provide an
accurate prediction of the total heat that can be extracted from an optical fiber.
The analysis of ASF cooling in a fiber presented in this chapter builds on the existing models by
incorporating in them all the aforementioned effects that were previously neglected. Namely, the model
includes ASE and the dependence of the excited-state population on position, the background absorptive loss
due to impurities in the fiber, the pump and fluorescence spatial mode profiles, and concentration quenching,
which turns out to have a significantly detrimental effect on the cooling efficiency. The model simulates a
pump at a given wavelength launched into the core of a single-mode or multimode fiber doped with a two-
level laser ion, using Yb3+ as a specific example. As the pump travels along the fiber (in either the
fundamental mode in a single-mode fiber or multiple modes in a multimode fiber), the upper manifold
becomes populated and spontaneous emission ensues. Some of the spontaneous emission escapes from the
sides of the fiber and produces cooling, while some of it is guided by the core in the forward and backward
directions and is amplified, resulting in ASE that escapes from both ends of the fiber. We derive expressions
for the heat removed at all points along the fiber as a function of pump wavelength, pump power, core size,
23
dopant concentration, as well as all relevant spectroscopic parameters. These expressions can be used to
model a fiber in air (convective heating) or in a vacuum (radiative heating). Modeling of a typical single-
mode Yb-doped ZBLANP fiber shows that ASE has a negative effect on cooling that must be accounted for
in longer fibers, and that there is an optimum pump power that achieves maximum cooling at some point
along the fiber. Importantly, a simple expression based on the spectroscopic and physical fiber parameters is
derived for the approximate heat that can be extracted per unit length from a fiber when absorptive loss is
negligible. This expression turned out to be invaluable in this research to quickly assess the cooling
performance of a prospective rare-earth-doped fiber and host composition. These expressions can be readily
applied to model common rare-earth ions in single-mode and multimode fibers. While we study here a core-
pumped fiber, the model can also be used to model cladding-pumped or multimode fibers by changing the
pump and signal intensity distributions from the fundamental core modes to multimode distributions [11].
As shown in the next chapter, the model can also be applied to fiber lasers and amplifiers by taking into
account how the excited-state population is altered by the presence of either the signal being amplified or the
laser signal.
Figure 1: (a) Schematic of the doped fiber and definitions of key model parameters; (b) Energy-level diagram
of Yb3+ as an example of a two-level system for ASF cooling; (c) Absorption and emission cross-section spectra
of the Yb-doped ZBLAN fiber simulated in this chapter.
24
2. Model
Consider a fiber doped with a two-level laser ion (Fig. 1a), with a circular core of radius a, a cladding of
radius b, and a length L. The fiber coordinates (r, z) are shown in Fig. 1a. The fiber is single-mode at the
pump and ASE wavelengths, hence all physical parameters are independent of azimuthal angle f. A pump of
power Pp0 and wavelength lp is launched into the fiber at z = 0, with no input signal or laser signal. The fiber
core is doped with a laser ion concentration N0. The laser ions’ lower manifold has i sub-levels, when optically
pumped, and a total electronic population density N1 (see Fig. 1b). The upper manifold has j sub-levels and
a population density N2 = N0 – N1. Figure 1c shows the cross-section spectra for the Yb3+-doped ZBLANP
fiber modeled as an example in this chapter [12]. The upper manifold total lifetime is
t = 1/(trad-1 + tnr-1+ tq-1), where trad is the radiative lifetime, tnr is the nonradiative lifetime due to multi-
phonon relaxation within a Yb3+ ion, and tq is the concentration-quenching lifetime. Quenching is caused by
energy transfer from an excited Yb3+ ion to an impurity such as Fe2+ or OH-, producing subsequent
nonradiative relaxation, or to another Yb3+ ion. The probability of this process increases with dopant
concentration, and its effect on the upper manifold lifetime is given by [27]:
τ0
(1)
τ (N 0 ) = 2
9 ⎛ N0 ⎞
1+
2π ⎜⎝ N c ⎟⎠
t0 = 1/(trad-1 + tnr-1) is the total lifetime at sufficiently low concentrations (tq-1 ≈ 0), and Nc is the critical
concentration for which energy transfer to an impurity has the same probability as energy transfer to another
Yb3+ ion. Inserting the expression of t and t0 in (1) and solving for tq gives [13]:
2
2π / 9 ⎛ N c ⎞
τ q ( N0 ) =
τ rad −1 + τ nr −1 ⎜⎝ N 0 ⎟⎠ (2)
Radiative trapping in the fiber's axial direction, in which spontaneous photons traveling along the doped
region as a core mode are reabsorbed by ions in the ground state, is accounted for by simulating the forward
and backward ASE. Radiative trapping in the radial direction is neglected because the diameter of a single-
mode fiber core is sufficiently short that the distance traveled by photons is negligible and trapping is
25
insignificant [14].
To model the power in the ASE, the emission and absorption spectra of the laser ions are sliced into n
adjacent segments centered at wavelength li (frequency ni), each with the same narrow width dl (n = 100 in
all simulations). In spectral slice li the forward and backward ASE intensities at r and z in the fiber are
defined as Iase±(r,z,li) (see Fig. 1a), and the absorption and emission cross-sections are sa(li) and se(li),
respectively. The rate equation for the upper manifold’s total population density N2 is [15]:
ε
dN 2 (r, z) I p (r, z) p N (r, z) − n I ase (r, z, λi )
dt
=
hν p
(σ a N1 (r, z) − σ ep N 2 (r, z)) − 2
τ
+∑∑
ε =+ i=1 hν i
( ) ( )
(σ a λi N1 (r, z) − σ e λi N 2 (r, z))
(3)
where Ip(r,z), is the pump intensity at r and z, and σap and σep are the pump absorption and emission cross-
sections of the laser ions, respectively. The first term (proportional to spa) accounts for excitation of an
electron from the ground state to the excited state by absorption of a pump photon, and the second term
(proportional to spe) for de-excitation of an electron in the excited state by stimulated emission. The third
term (proportional to 1/t) represents de-excitation of excited electrons by all three relaxation mechanisms.
The final two terms account for changes in exited-state population induced by absorption and stimulated
emission of ASE photons. The total density of laser ions is Nt(r) = N1(r,z) + N2(r,z). For a core uniformly
doped at a laser-ion concentration N0 and an undoped cladding, which is the assumption made in the rest of
this chapter:
⎪⎧ N 0 ≤ r ≤ a (4)
Nt (r) = ⎨ 0
⎩⎪ 0 a < r
I p (r, z) n +
I ase (r, z, λi ) n I ase−
(r, z, λi )
+∑ +∑
I p,sat ,a i=1 I ase,sat ,a ( λi ) i=1 I ase,sat ,a ( λi )
N 2 (r, z) = N t (r)
I p (r, z) n +
I ase (r, z, λi ) n I ase
−
(r, z, λi )
1+ +∑ +∑
I p,sat I ase,sat ( λi ) I ( λi )
i=1 i=1 ase,sat (5)
where the pump and ASE saturation intensities are defined as:
hν p hν p
I p ,sat , a = a nd I p ,s at = (6a)
σ τ
p
a (σ p
a )
+ σ ep τ
26
hν i hν i
I ase,sat ,a ( λi ) = and I ase,sat ( λi ) =
σ a ( λ i )τ (σ ( λ ) + σ ( λ )) τ
a i e i
(6b)
I p (r , z ) = Pp ( z ) f p ( r ) (7)
where Pp(z) is the pump power at z, and fp(r) is the normalized transverse intensity distribution of the pump
power, assumed to be in the fundamental mode. Similarly, in each spectral slice i the ASE intensities are
related to the ASE powers Pase+(z,li) and Pase-(z,li) and to the ASE fundamental mode distribution fi(r) by:
±
I ase (r, z, λi ) = Pase
±
(z, λi ) f i (r) (8)
2π ∞ 2π ∞
∫ ∫
0 0
f p ( r )r dr dφ = ∫
0 ∫ 0
f i (r ) r dr dφ = 1 (9)
The same formulation can be used for multimode fibers, for which fp(r) can be
dI p ( r , z )
dz
( )
= I p (r , z ) σ ep N 2 ( r , z ) − σ ap N 1 ( r , z ) − α b I p ( r , z ) (10)
= I (r , z ) ((σ
p e
p
+σ p
a )N ( r , z ) − σ
2 a
p
)
N t ( r ) − α b I p (r , z )
where αb is the fiber background loss coefficient at lp, resulting in particular from absorption by impurities
(associated with the presence of the laser ions) and OH-, and to a lesser extent scattering. The first two terms
in (10) account for changes in pump intensity induced by absorption and stimulated emission of pump
photons by the laser ion. The last term (proportional to αb) accounts for pump attenuation caused by the
background loss. Substituting N2(r,z) from (5) in (10) gives an expression of dIp(r,z)/dz as a function of Ip(r,z)
and Iase±(r,z,li). Replacing these intensities by their expressions as a function of powers ((7) and (8)), then
27
⎛ P (z) τ ⎞
1 dPp (z) 2π ∞ 2π ∞
(σ p
a )
+ σ ep ⎜ p
I
⎝ p,sat ,a
f p (r) +
τ
S1 (z) f i (r)⎟
⎠
= −σ a (v p ) ∫ ∫ N t (r) f p (r)r dr dφ + ∫ ∫
rad
N t (r) f p (r)r dr dφ − α b
Pp (z) dz 0 0 0 0 Pp (z) τ
1+ f p (r) + S2 (z) f i (r)
I p,sat τ rad
(11)
n +
Pase (z, λi ) n
P − (z, λi )
S1 (z) = ∑ + ∑ ase
i=1 I ase,sat ,a ( λi ) i=1 I ase,sat ,a ( λi )
n +
Pase (z, λi ) n Pase −
(z, λi )
S2 (z) = ∑ +∑
i=1 I ase,sat ( λi ) i=1 I ase,sat ( λi )
(12)
Replacing in (11) the dopant distribution Nt(r) by (4) and integrating in f gives:
dPp (z)
= −2π Pp (z)N 0 ∫
a (
σ ap (1+ S2 (z) f i (r) ) − σ ap + σ ep S1 (z) f i (r) ) f p (r)r dr − α b Pp (z)
dz 0 Pp (z)
1+ f p (r) + S2 (z) f i (r)
I p,sat
(13)
We assume that the fiber has a step-index profile, so that its fundamental mode has an intensity
distribution closely approximated by a Gaussian [16]. Since the pump and ASE are very close in wavelength,
2 −2 r 2
/w2
f p (r ) ≈ fi (r ) = f (r ) = e (14)
π w2
Pp ( z ) S1 ( z ) P ( z ) S2 ( z )
p1 ( z ) = + a nd p 2 ( z ) = p + (15)
Pp , sat ,a Am Pp ,sat Am
where Am = pw2/2 is the mode effective area, and Pp,sat,a and Pp,sat are the pump saturation powers:
Replacing (14) in (13) and integrating in r yields the expression for the pump power evolution along the
fiber:
28
The boundary condition for the pump power at z = 0 is Pp(0) = Pp0. (17) accounts for all main mechanisms
that affect the pump power distribution along the fiber, and therefore the distribution of the excited-state
population along z, including the Yb3+ absorption and emission cross-sections, the radiative, nonradiative,
and quenching-related relaxation lifetimes, and the absorption and emission of forward and backward
fluorescence photons.
The evolution of the ASE intensities Iase±(r,z,li) in slice li and at location r and z in the fiber are
±
dI ase (r, z, λi ) cδλ
= ±γ se (r, z, λi )2hvi 2 f i (r) ± γ ase (r, z, λi )I ase
±
(r, z, λi ) ∓ α b I ase
±
(r, z, λi )
dz λi (18)
where hni is the average energy of the photons in that spectral slice and:
γ s e ( r , z , λ i ) = σ e ( λi ) N 2 ( r , z ) (19a)
The first term in (18) is the seed term, representing the spontaneous emission power emitted in all
polarizations in spectral window i; it is guided in the fundamental mode, hence the factor fi(r) in it. The
second term represents amplification of this seed. The third term accounts for fiber loss, assumed wavelength
independent.
Following the same methodology as for the pump, the differential equations (18) for the ASE powers can
be written as:
±
dPase (z, λi ) p (z) ⎛
= ± N0 1 ⎜ η +
1 (
1+ 1− η p2 (z) ⎞ ⎛ ± ) ( ) cδλ ⎞
ln ⎟ ⎜ Pase (z, λi ) σ e ( λi ) + σ a ( λi ) + 2hvi 2 σ e ( λi )⎟
dz p2 (z) ⎝ p2 (z) 1+ p2 (z) ⎠ ⎝ λi ⎠
∓σ a ( λi )N 0η Pase
±
(z, λi ) ∓ α b Pase
±
(z, λi )
(20)
The boundary conditions for the n forward ASE signals are known at z = 0:
+
Pase (0, λi ) = 0 (21a)
29
while for the backward ASE signals they are known at the forward output of the fiber (z = L):
−
Pase (L, λi ) = 0 (21b)
As in other laser problems, the 2n + 1 differential equations for the pump (17) and ASE (20) powers are
coupled through their nonlinear saturation terms p1(z) and p2(z), and they must be solved numerically. To this
end, the following algorithm was used. First, at z = 0 the pump power is set to its known input value Pp0 and
the forward ASE powers to their known values (21a), while initial values for the backward ASE powers,
which are unknown at z = 0, are guessed. The equations are solved forward along z using a fourth-order
Runge-Kutta method. The solutions for the pump power and the n backward ASE powers at z = L are then
compared to their respective boundary values (Pp(L) > 0 and (21b)). If all differences between these values
are small enough (typically 10-7 W), the convergence loop is terminated. Otherwise, the n backward ASE
powers at z = L are set to zero (as dictated by (21b)), and the equations are propagated backward, from z = L
to z = 0, using the pump power at z = L calculated at the end of the previous run. The pump and backward
ASE powers obtained at z = 0 at the end of this run are compared to their respective boundary values. If all
differences are sufficiently small, the loop is terminated. This cycle is repeated until the boundary conditions
at both ends are satisfied. The algorithm converges relatively quickly (typically less than 20 iterations). It
Various competing mechanisms remove and deposit heat into the pumped fiber. The main source of
heating is pump absorption by the laser ions, and to a smaller extent by the host impurities. Absorption of the
two ASE signals also contributes, but to a lesser extent since ASE carries less power. The spontaneous
emission that radiates out of the fiber sides cools the fiber by anti-Stokes fluorescence. The ASE signals also
contribute to cooling, since they carry power out of the fiber ends.
The fiber background loss originates from absorption by impurities, and scattering. Absorption on purely
nonradiative electronic transitions is converted into heat. Absorption on electronic transitions that are not
purely nonradiative contributes to phonon generation (heating) and fluorescence (no heating). Scattering
leads to a loss of photons without heating. To account for these three contributions, we label aba the loss
coefficient representing phonon-generating mechanisms, and abs the loss coefficient due to scattering, with
30
ab = aba + abs.
The rate of accumulation of heat in a thin ring of fiber of length dz and radial width dr located at (r,z) can
be written as:
du(r, z) +
dI (r, z) n dI ase (r, z, λi ) n dI ase
−
(r, z, λi ) ⎛ n n
⎞
=− p −∑ +∑ − α bs ⎜ I p (r, z) + ∑ I ase
±
(r, z, λi ) + ∑ I ase
−
(r, z, λi )⎟
dt dz i=1 dz i=1 dz ⎝ i=1 i=1 ⎠
N 2 (r, z) n
cδλ
− h〈v f 〉 + 4∑ σ e ( λi )N 2 (r, z)hν i 2 f (r)
τ rad i=1 λi (22)
The first term is the difference between the pump intensities entering and exiting the volume, divided by dz
to convert intensity into energy per unit volume. This term is therefore equal to the pump power lost within
the volume due to absorption by the laser ions and all loss mechanisms. Since the pump power decreases
from z to z + dz, this term is positive, and provides heat to the fiber. The second term is the difference between
the forward ASE intensities entering and exiting the volume, divided by dz. It is summed over all frequencies
ni to cover the full emission spectrum of the laser ions. Since the forward ASE generally grows from z to
z + dz, this term is generally negative, and it extracts energy from the fiber. The third term is the same as the
second term but for the backward ASE. Since the backward ASE always decreases from z to z + dz, this term
is negative, and it also extracts energy. The fourth term represents the heat removed by scattering of the pump
and ASE. Since scattering loss is already accounted for in the attenuation of the pump and ASE signals in
the first three terms, yet does not contribute to heating, this term must be subtracted, hence the minus sign in
front of it. The fifth term represents the spontaneous emission generated at all angles. However, the
spontaneous emission that falls within the acceptance angle of the core was already included in the second
and third terms. The sixth term is the correction required to correct this double counting. Stated differently,
the sum of the fifth and sixth term represents the spontaneous emission generated at angles that do not fall
within the forward and backward acceptance angle of the core and is, therefore, not significantly amplified.
The heat removed/deposited in a thin slice of fiber of thickness dz at position z is obtained by integrating
(22) across the fiber. To do so, in (22) dIp/dz is replaced by (10) and dIase±/dz by (18). In (18), the coefficient
gase is replaced by (19a) and gse by (19b) to express them as a function of N2(r,z). Finally, N2(r,z) is replaced
by (5), and the resulting equation is integrated in r and f. The result can be cast in the form:
31
Integrating this quantity along the fiber gives the total power Pcool extracted from the fiber, a useful metric
often cited in the literature to quantify ASF cooling [8,10]. The first term represents the heat input due to
pump absorption by laser ions, the second term is the heat removed by ASE, the third term is the heat removed
by fluorescence, and the fourth term is the heat added by the absorptive loss. These terms are given by:
dQ pump (z) ⎛ ( )
1+ 1− η p2 (z) ⎞
dt
= ησ ap N 0 Pp (z) −
p1 (z)
p2 (z)
( )
× Pp (z) σ ap + σ ep N 0 ⎜ η +
⎝
1
p2 (z)
ln
1+ p2 (z) ⎠
⎟
(24a)
dQase (z) − n
p (z) ⎛ − n n
cδλ ⎞
dt p (z) ⎝ ε =+ i=1
( )
= N 0η ∑ ∑ Pi ε (z,ν i )σ a (ν i ) − N 0 1 ⎜ ∑ ∑ Pi ε (z,ν i ) σ a (ν i ) + σ e (ν i ) + 4∑ σ e ( λi )hvi 2 ⎟
λi ⎠
ε =+ i=1 2 i=1
⎛
×⎜η +
1
ln
(
1+ 1− η p2 (z) ⎞)
⎟
⎝ p2 (z) p2 (z) ⎠ (24b)
⎛ n
cδλ ⎞
4h∑ σ e ( λi )vi 2 ⎟
dQ fluo (z) ⎜
= ⎜ Am
h〈v f 〉
+
i=1 λi
⎟ N 0 1 ln
(
p (z) 1+ 1− η p2 (z)) p (z) n cδλ
+ 4N 0η 1 ∑ σ e ( λi )hvi 2
dt ⎜ τ rad p2 (z) ⎟ p2 (z) 1+ p2 (z) p2 (z) i=1 λi
⎜ ⎟
⎝ ⎠ (24c)
dQloss (z) ⎡ n n
⎤
= α ba ⎢ Pp (z) + ∑ Pase
+
(z, λi ) + ∑ Pase
−
(z, λi ) ⎥
dt ⎣ ⎦
i=1 i=1
(24d)
To calculate the fiber temperature change from the net extracted heat, the thermodynamic interactions
between the fiber and its surroundings must be specified. For a fiber placed in air at atmospheric pressure,
there are three primary contributions to the head load: (1) convection of the surrounding air, (2) conduction
from the fiber supports, and (3) thermal radiation from surrounding objects. Assuming the temperature
measurement is taken sufficiently far away from the fiber supports, the overwhelmingly dominant source of
heating is convection. As a result, most reported cooling experiments were conducted in a vacuum in order
to achieve lower temperatures. In this case, the largest contribution to external heat intake is thermal radiation.
To reduce the heat load from thermal radiation, a common approach is to tightly enclose the fiber in a low-
emissivity copper "clamshell" inside the vacuum chamber, as reported for cooling bulk crystals [20,21]. The
dominant heat load on the fiber is then thermal radiation from the clamshell, given by [22]:
⎛ dQ(z) ⎞
=
(
ε f σ B 2π R f T f (z)4 − Tc4 ) (25)
⎜⎝ dt ⎟⎠ 1+ χ
rad
32
where ef is the emissivity of the fiber, sB is the Stefan-Boltzmann constant, Rf is the fiber radius, Tf (z) and Tc
Rf ε f
χ = (1− εc ) (26)
Rc ε c
Rc is the inside radius of the clamshell (typically a few millimeters) and ec is the emissivity of the clamshell.
In this arrangement, as in all others, conduction along the length is assumed to be negligible due to the small
cross-sectional area of the fiber. At thermal equilibrium, the heat removed by ASF (23) equals the radiative
dQ ( z ) 1 + χ (27)
T f ( z ) = 4 Tc4 +
dt ε f σ B 2π R f
When a core-pumped fiber is cooled by ASF, its core cools first. The core is then cooler than the cladding,
and heat flows inward from the outside of the fiber through the cladding and into the core via thermal
conduction. For temperature changes that are small compared to room temperature, it has been shown that
the steady-state temperature profile at a given location z is nearly constant across the fiber cross section, even
for a material with a relatively poor thermal conductivity such as silica glass [23]. This is the reason why no
If the fiber is placed in air instead, the fiber primarily interacts with its surrounding through convection,
⎛ dQ ( z ) ⎞
⎜⎝ dt ⎟⎠ (
= 2π R f h T f ( z) − Ts urr ) (28)
conv
where h is the temperature-dependent heat transfer coefficient between air and ZBLANP, and Tsurr is the
surrounding air temperature, commonly room temperature. At thermal equilibrium, equating heat removed
dQ ( z) 1 (29)
T f ( z ) = Ts urr +
dt 2 π R f h
For a typical fiber with a 125-µm outer diameter in air at room temperature, it takes around 3–4 s for the fiber
temperature profile to reach steady-state after the cooling pump power has been turned on [23].
33
3. Simulations of Cooling in a Yb-doped ZBLANP Fiber
This formalism was used to predict the cooling that can be achieved in a Yb-doped ZBLANP fiber, and its
fiber with parameter values obtained from [9,12,24] (see Table 1). The cross-section spectra used in the
simulations are shown in Fig. 1c. The fiber is nearly single-mode throughout the emission region of Yb3+
(~920 nm to ~1080 nm). The background loss was taken to be 20 dB/km; 75% of it was assumed to be
absorptive. To obtain the absorption and emission cross-section spectra (Fig. 1c), the cross-sections published
in [12] were digitized. However, lack of sufficient resolution prevented accurate acquisition of values for the
absorption spectrum at higher wavelengths and the emission spectrum at lower wavelengths. Therefore, at
their tail ends the spectra disobeyed the McCumber relation [25]:
⎛ hc / λi − hc / λ Z ⎞
σ a ( λi ) = σ e ( λi ) exp ⎜ ⎟⎠
⎝ k BT (30)
where lZ is the wavelength associated with the zero-line frequency (the frequency associated with the
transition between the lowest sublevels of the lower and upper manifold), kB is the Boltzmann constant, and
T is the temperature of the sample. Since physically accurate simulations require this relation to be satisfied,
we used the McCumber relation to obtain the absorption cross sections at longer wavelengths from the more
reliably digitized values of the emission at these wavelengths. Similarly, we calculated the emission spectrum
at shorter wavelengths from the absorption values in this range. Piecing together the digitized and calculated
values for the absorption and emission resulted in the two full spectra shown in Fig. 1c, which satisfy the
McCumber relation. The mean fluorescence wavelength calculated from the emission spectrum is
<lf> = 993 nm. The nonradiative relaxation rate, calculated from [19], is sufficiently small (~10-8 s-1) to be
negligible compared to the radiative relaxation rate. The concentration-quenching lifetime for a 1 wt.% Yb-
doped ZBLANP fiber (τq = 850 ms) was calculated using (2) with values from experimental ASF cooling
34
TABLE I
YB-DOPED FIBER PARAMETERS (VALUES WITH AN ASTERISK ARE CALCULATED, ALL OTHERS ARE MEASURED)
The output of a typical simulation is illustrated in Fig. 2, which plots the dependence of several quantities
of interest on position along a 2-m fiber pumped at 1010 nm. The pump power (thick solid blue curve),
injected on the left side (z = 0), decreases monotonically. However, the population N2(z) (thick dotted red
curve) is not monotonic. It grows at first because the total ASE power is high near the pump-input end of the
fiber and depletes the gain in this region. After reaching a maximum, N2(z) decreases because of the
decreasing pump power. The forward ASE power (thin solid black curve) grows from z = 0 to z ≈ 1.2 m,
exponentially at first, then sublinearly as a result of gain saturation. After z ≈ 1.2 m, it decreases because N2
is very small in this region. Since most of the ions are in the ground state, the ASE power is absorbed. The
backward ASE power (thin dashed black curve) grows monotonically in the opposite direction, from z = L to
z = 0, reaching its maximum value at z = 0. This behavior is commensurate with the well-known behavior of
35
Figure 2: Position-dependent values for N2 (thick dotted red curve), the pump power (thick solid blue curve),
and ASE (thin black curve) calculated for a 1-W pump power at 1010 nm launched into a 2-m Yb-doped
ZBLANP fiber with the parameter values listed in Table 1 and the cross-section spectra shown in Fig. 2.
The average heat extracted per unit length is plotted in Fig. 3 as a function of pump wavelength for 1 W
of power launched into a 40-cm and an 80-cm fiber (thick black curves). For any reasonable fiber lengths,
and irrespective of whether ASE is taken into account, cooling is maximum near 1015 nm. When ASE is
neglected, the upper state population is completely saturated along the entire fiber, for either length.
Therefore, the average heat extracted per unit length is independent of fiber length (thick dotted curve). In
this case, there is net heating for pump wavelengths below 995 nm, and net cooling for wavelengths above
995 nm. This threshold wavelength is slightly higher than <lf> due to quenching and absorptive loss, which
is consistent with previous findings [15,26]. Including ASE in the simulations for the 40-cm fiber (bold
dashed curve) causes (1) a slight increase in this threshold wavelength (to 996 nm), and (2) a significant
increase in heating below this wavelength. However, for wavelengths above this threshold, where cooling
occurs, ASE has a negligible impact: it only slightly decreases the average extracted heat, for example by
~0.1% at 1015 nm for the 40-cm fiber. When the length is increased to 80 cm (bold solid curve), the
wavelength at which ASE becomes negligible shifts to longer values. The conclusion is that since even for
reasonable fiber lengths this critical wavelength falls in the range often used for ASF cooling, to accurately
model cooling in fibers it is important to include ASE, especially in longer fibers. The net heating caused by
36
ASE arises from three effects. First, ASE depletes the upper level population, which reduces ASF. Second,
it produces nonradiative relaxation between sublevels since its mean wavelength <lASE> is longer than the
pump wavelength. However, since <lASE> is longer than <lf>, reabsorption of the ASE also contributes to
Figure 3 also plots the average gain coefficient along the 40-cm fiber for selected signal wavelengths
when ASE is included (thin gray dashed curves). These curves illustrate that there is a range of pump
wavelengths for which the average gain coefficient is positive and the average temperature change is
negative. In other words, there is a fairly broad range of pump wavelengths that can be used to produce both
gain and cooling. For example, at a signal wavelength of 1010 nm the gain coefficient (gray long-dashed
curve) is positive below a pump wavelength of ~1010 nm, while the extracted heat for the 40-cm fiber (bold
dashed curve) is negative above 996 nm (shaded area). As the pump wavelength is increased, the gain
coefficient decreases. This explains why the ASE is negligible for longer pump wavelengths: there is not
Figure 3: Predicted dependence on pump wavelength of the average heat extracted from a 40-cm or 80-cm Yb-
doped ZBLANP fiber, with and without ASE, and of the average gain along the 40-cm fiber at selected
wavelengths (when ASE is included).
37
3.3 Pump Power Dependence of Heat Removal
At the optimal pump wavelength (1015 nm), the pump power was swept to determine the value that
maximizes the heat extracted per unit length at some location in a 1.2-m fiber (Fig. 4). For a lossless fiber,
the maximum heat extracted per unit length increases asymptotically with increasing pump power. This is
expected since the spontaneously emitted power is proportional to the upper level population N2 and
increasing the pump power increases N2 until N2 saturates and the extracted heat levels off. For a lossy fiber,
the extracted heat per unit length increases to a maximum, flattens out over some length of fiber, then
decreases (see Fig. 4). This can be explained with (24d): as the pump power is increased, heating due to
absorptive loss increases indefinitely, while cooling induced by ASF (24c) increases but eventually levels off
(due to the saturation terms in 1 + p2(z) in the denominators of (24c)). Eventually, heating dominates and the
extracted heat decreases. We refer to the power that produces the maximum extracted heat the optimum pump
For a lossless fiber operating in a regime with insignificant effects from ASE (as justified in Fig. 3), the
optimum pump power for maximum cooling is infinite. In this case, an exact expression for the maximum
heat that can be extracted per unit length can be derived by setting the loss (24d) and ASE (24b) terms in (23)
to zero, and taking the limit as the pump power approaches infinity:
⎛ dQ ⎞ ⎛ τ ⎞ σ a π a2
= ⎜ r ad hν p − h 〈 v f 〉⎟ a p e N0 (31)
⎜⎝ dt ⎟⎠
max ⎝ τ ( No ) ⎠ σ p + σ p τ rad
where t(N0) is the concentration-dependent lifetime given by (1). (31) is similar to expressions for bulk
media [9], and it agrees with Kashyup’s widely used expression [8] for cooling a fiber in the limit of high
An exact expression can also be derived for the optimum pump power that maximizes cooling for a lossy
fiber with negligible ASE. Setting the ASE term in (23) to zero, taking the derivative with respect to pump
power, setting it to zero, and solving for the pump power gives:
σ aN ⎛ τ h 〈v f 〉 ⎞
−(2 − η ) + η 2 + 4 (1− η ) p 0 ⎜τ − 1⎟
α ba ⎝ r ad hν p ⎠
Pp ,opt = Pp , sat (32)
2 (1− η )
(31) and (32) agree well with the values predicted in Fig. 4.
38
Figure 4: Dependence of the maximum extracted heat on the launched pump power at 1015 nm for a fiber
1.2-m long with and without absorptive background loss.
Figure 5 shows the heat extracted per unit length as a function of z for a fixed fiber length (5.5 m) and
various launched pump powers Pp0 at 1015 nm. These plots further illuminate the dependence on pump power
and help predict the location of maximum cooling along a fiber. If the pump is launched at a power less than
Pp,opt (10 mW, dashed blue curve in Fig. 5), the coolest point is at the start of the fiber, and the temperature
change decreases until a majority of the pump power is absorbed and neither cooling nor heating occurs.
Increasing the launched pump power to Pp,opt (58 mW, dotted orange curve) increases the temperature change
at the start of the fiber, but the coolest point is still located at the pump-input end. However, increasing the
pump power above Pp,opt (200 mW, dot-dashed purple curve) moves the coolest point into the fiber as the
heating due to absorptive loss reduces the amount of cooling at the input end. Nevertheless, although the
coolest point moves, the temperature change at the coolest spot is still approximately equal to what it is when
launching Pp,opt. This behavior explains the flat section in Fig. 4 for the much shorter 1.2-m fiber: at the start
of the flat region (58.3 mW), the power is equal to Pp,opt (and the extracted heat is maximum) at the start of
the fiber. As the power is increased above 58.3 mW, the pump must travel a longer distance to be attenuated
to Pp,opt, and the coolest point moves further down the fiber. At some pump power (~160 mW in Fig. 4) the
coolest point reaches the far end of the 1.2-m fiber. Since the coolest point cannot move any further in z, any
further increase in pump power results in an increase in the fiber temperature, as seen in Fig. 4.
For the long 5.5-m fiber considered in Figure 5, the effect of ASE becomes significant when the pump
power is increased above 200 mW (e.g., 300 mW, solid green curve). In this case, the coolest point never
reaches the maximum possible negative temperature change. This is due to the fact that larger pump powers
39
increase the ASE power in the fiber, resulting in net heating. However, since the mean wavelength of ASE
<λASE> is longer than <λf>, ASE is able to negate some of its own heating effect as it is reabsorbed in the
more distal region of the fiber, creating ASF cooling. The circled area on the solid green curve identifies the
location where the majority of the pump power has been absorbed and the forward ASE starts to be
reabsorbed, acting as a pump with wavelength equal to <λASE>. Therefore, the rest of the fiber (after 3 m) is
being cooled by absorption of forward ASE. At this pump power, heating due to absorptive loss also becomes
more significant and surpasses the cooling due to ASF at the input end. This causes the input to experience a
positive temperature change (left end of the green curve in Fig. 5). With sufficient pump power (much higher
than 300 mW), the fiber would be heating throughout its length, because heating due to absorptive loss (which
is proportional to pump power) far exceeds ASF cooling (which saturates at large pump powers).
Figure 5. Temperature distribution along a 5.5-m Yb-doped ZBLANP fiber predicted for various launched
pump powers at 1015 nm.
For payload cooling, it is important to consider the total heat extracted from the fiber. To optimize this
quantity, which is the extracted heat per unit length integrated along the fiber, the pump power, wavelength,
and fiber length must then be optimized differently. As was illustrated in Fig. 5, when the launched pump
power is equal to the optimum value, the extracted heat is maximum at the input and decreases monotonically
40
inside the fiber. The temperature gradient along the fiber is then relatively large. When the launched pump
power is increased above Pp,opt, the temperature decreases gradually along the fiber to the coolest point (e.g.,
~1.6 m for 200 mW), then gradually increases. The temperature gradient is then smaller, and the average
temperature is lower (heat extracted higher). Consequently, maximizing the total extracted heat requires a
pump power higher than Pp,opt and a fiber long enough to absorb most of the pump (e.g., 99%). This
optimization maximizes the length of fiber along which a negative temperature change is created.
The dependence of extracted heat on pump power is plotted in Fig. 6a for a 1031-nm pump and different
fiber lengths. The competition between heating from absorptive loss and quenching on one hand, and cooling
from ASF on the other, determines the total extracted heat. For example, for the 12-m fiber (solid curve), as
the launched pump power is increased from zero to 430 mW, cooling increases more than heating (negative
slope). However, at 430 mW, cooling is maximum, and increasing the pump power further merely increases
heating from absorptive loss (positive slope): for powers above 430 mW, the total extracted heat decreases.
The three curves overlap at small powers because most of the fiber length is unpumped and only the beginning
Assuming a sufficiently long fiber and optimizing the pump power for each wavelength, the total
extracted heat was plotted as a function of pump wavelength (solid curve in Fig. 6b). The dashed curve is the
cooling efficiency, defined as the total extracted heat divided by the launched pump power. As the
wavelength is increased, the total extracted heat initially increases since the difference between the pump and
mean fluorescence wavelengths increases, which increases the ASF conversion efficiency proportionally to
(lp - <lf>) / <lf> [27]. However, at sufficiently long wavelengths, the total extracted heat decreases because
the pump absorption weakens, ASF removes too little heat per unit length, and heating due to the fiber
absorptive loss dominates. There is therefore a wavelength that maximizes the total extracted heat
(~1031 nm). This wavelength is longer than the wavelength that maximizes local cooling, and it is largely
Optimizing the pump power (430 mW) and wavelength (1031 nm) to maximize the total extracted heat
(7.15 mW), and using the minimum fiber length for which more than 99% of the pump is absorbed (20 m),
leads to a cooling efficiency of ~1.7%. The maximum possible cooling efficiency is (lp - <lf>) / <lf>, or
3.7%. This difference in efficiency is due to the background absorptive loss and quenching. However, the
41
maximum possible efficiency can be approached at low pump power, since less of the pump power is
converted into heat by the fiber background absorptive loss. This is confirmed by Fig. 6a, in which the ratio
of extracted heat to launched pump power is ~3.4% for low pump powers, as indicated by the dashed line.
The maximum efficiency can also be approached when the absorptive loss is negligible: simulations show
that the cooling efficiency is then 3.5%, nearly equal to its maximum possible value for this wavelength
Figure 6. (a) Dependence of the total extracted heat on the launched pump power at 1031 nm for three fiber
lengths. (b) Dependence of the total extracted heat and cooling efficiency on pump wavelength for a fiber long
enough to absorb most of the pump even at the longest wavelength (i.e., 20 m) and the corresponding cooling
efficiency. At each wavelength the power is optimized to maximize the total extracted heat.
The heat extracted per unit length increases with the number of interactions per unit time and unit length
between phonons and Yb3+ ion. This rate can be improved by increasing the core radius (and concomitantly
the pump power), as illustrated by (31), which shows a direct proportionality between the heat extracted per
unit length and the core area. The dopant concentration N0 can also be adjusted to maximize the heat extracted
per unit length. In (31), N0 appears in the numerator (increasing the concentration increases the ASF power
and thus cooling) but also in the total lifetime t(N0) in the denominator (increasing the concentration increases
42
quenching, which reduces t(N0) and thus reduces cooling). Therefore, an optimum concentration exists.
Inserting the expression for t(N0) (1) into (31), taking the derivative with respect to N0, and solving for the
concentration that maximizes the heat extracted per unit length in a lossless fiber gives:
π ⎛ τ 0 λp ⎞
N 0 ,opt = N c ⎜ − 1⎟ (33)
14 ⎝ τ r ad 〈λ f 〉 ⎠
For the fiber simulated in this chapter, (33) gives N0,opt = 1.9 wt.% Yb. To assess the impact of absorptive
loss on this optimum, the maximum heat extracted from a 15-m section of the studied fiber (whose absorptive
loss is 15 dB/km) is plotted as a function of dopant concentration (Fig. 7a). For each concentration, the
launched pump power is optimized to maximize the total extracted heat. The maximum extracted heat occurs
at a concentration of 1.9 wt.% Yb, as predicted by (33), confirming that this expression is a valid
Figure 7. (a) Maximum total heat extraction as a function of Yb concentration for a 1031-nm pump launched
into a 15-m fiber with an absorptive loss of 15 dB/km. (b) Cooling efficiency as a function of concentration
for the same fiber.
However, plotting the cooling efficiency as a function of dopant concentration (Fig. 7b) shows that the
optimum concentration for maximizing the extracted heat is not necessarily the same as the concentration
what maximizes efficiency. At sufficiently low concentrations, the number of interactions per unit time and
length between phonons and Yb3+ ions is relatively low, and a significant portion of the pump power is wasted
through absorptive loss. As the concentration increases, so does the probability that a pump photon is
absorbed by a Yb ion before it encounters an impurity and its energy is converted into heat. This increases
43
the cooling efficiency until quenching becomes significant (~0.5 wt.% Yb in Fig. 7b) and the efficiency starts
to decrease. For the simulated fiber, the heat extraction is maximized for a concentration of 1.9 wt.% Yb,
while a significantly lower 0.5 wt.% Yb optimizes the cooling efficiency, further demonstrating the trade-off
between extracted heat and cooling efficiency when the fiber loss is significant (15 dB/km in these
simulations).
For the concentration that maximizes the extracted heat (1.9 wt.% Yb), and in the limit of negligible fiber
loss, (31) predicts that in a large-mode-area (LMA) fiber with a core radius of 25 µm, which is approximately
the largest value achievable with current commercial technology, the maximum possible heat extraction is
161 mW/m. This number is concerningly small. The simplest way to improve it is to further increase the core
area. However, technology limitations can only improve this value by a factor of at most ~2–4. Achieving
greater heat extractions necessitates using multimode fibers. The largest negative temperature changes
reported in a fiber have indeed been observed in multimode fibers [7,9,24,28,29]. Another solution is to
switch to a different host with a higher inherent cooling efficiency, such as Tm3+.
Figure 8. Maximum extracted heat per unit length and maximum temperature change (when the fiber is placed
in air) predicted as a function of fiber absorptive loss for different core radii and Yb concentrations.
44
3.6 Loss Dependence of Heat Removal
As demonstrated in Fig. 4, the fiber absorptive loss has a significant effect on the optimal parameters and
the maximum achievable cooling. Figure 8 quantifies this dependence by graphing the maximum extracted
heat per unit length and associated temperature change as a function of absorptive loss. The curves were
calculated for different core radii and Yb concentrations, and no scattering loss. For all curves, as the
absorptive loss increases the maximum extracted heat stays relatively constant at first, then decreases rapidly
until the loss reaches a value where heat can no longer be extracted. For example, this critical loss is 1.04
dB/m for a = 3.1 µm and N0 = 1 wt.% Yb. Above this value (which cannot be shown on a log scale), the fiber
heats up for any launched pump power. This threshold is independent of the core radius (see the two solid
curves) and proportional to the Yb3+ concentration until quenching starts to dominate (see the curves for
N0 = 1 wt.% and 2 wt.%). Since the optimum pump power described by (32) must be positive, this threshold
loss can be calculated by setting the right side of (32) to be greater than zero and solving for the absorptive
loss:
⎛ τ ( N ) h 〈v f 〉 ⎞
α ba ,max (dB/m) = 4.343 ησ ap N 0 ⎜ 0
− 1⎟ (34)
τ
⎝ rad hν p ⎠
However, to achieve significant cooling, the requirement for the absorptive loss is much more stringent
than (34). Figure 9 plots the absorptive loss a90% required to achieve 90% of the maximum possible extracted
heat per unit length (which is the left end of each curve in Fig. 8). a90% increases with pump absorption cross-
section as well as Yb concentration until tq becomes comparable to trad and quenching starts to dominate
around 1.7 wt.% Yb. For our 1-wt.% Yb-doped fiber, the absorptive loss must not exceed ~2.1 dB/km. This
is lower than the typical total background loss of fibers with this level of doping, especially ZBLAN fibers.
In general, the absorptive loss is unknown. This quantitative analysis stresses the need to measure and reduce
this loss to qualify rare-earth-doped fibers for ASF cooling. The experimental work conducted for this thesis
shows that measuring the temperature of an optically-pumped fiber is an accurate method to measure this
45
Figure 9. The absorptive loss required to achieve 90% of the maximum local cooling efficiency as a function
of Yb concentration, for two values of the pump absorption cross-section (sa0 is the value obtained from the
spectrum in Fig. 1c).
4. Summary
We presented a versatile analytical model of anti-Stokes fluorescence cooling in optical fibers doped with
a quasi-two-level laser ion. This model includes important effects that have thus far not been considered in
one comprehensive model, including ASE, the dependence of the excited-state population on position along
the fiber, concentration quenching, absorptive loss, and the pump and fluorescence mode profiles. It provides
insight, quantitative predictions, and optimization tools for the design of fibers for future use in radiation-
balanced fiber lasers and amplifiers. When applied to a representative Yb-doped single-mode ZBLANP fiber,
it showed that ASE has significant negative effects on cooling for shorter pump wavelengths, higher pump
powers, and longer fibers. Therefore, to accurately predict cooling in these cases, ASE must be accounted
for. The model also provides a means to numerically predict the optimum pump wavelength, and analytically
calculate the optimum pump power and optimum dopant concentration that maximize the heat extracted per
unit length. Since heat conduction along the fiber is negligible, heat extraction is generally non-uniform along
the length, and the longitudinal temperature profile depends strongly on pump power. When the pump is
launched at the optimum power, the coolest point is at the fiber input end; increasing the power moves this
spot further inside the fiber. We also derived a simple predictive expression for the maximum absorptive loss
above which cooling is unobtainable (1.04 dB/m for N0 = 1 wt.% Yb). In practice, this loss needs to be
significantly smaller than this value in order to achieve 90% of the maximum possible local heat extraction.
46
For example, for a 1-wt.% Yb-doped fiber, this loss is 2.1 dB/km.
We also demonstrated the influence of absorptive loss and concentration quenching on the trade-off
between total extracted heat and cooling efficiency. Increasing the pump power to the value that maximizes
the total extracted heat causes the cooling efficiency to constantly decrease (from 1.7%) as more of the power
is wasted by the absorptive loss. In a fiber with negligible residual absorption and quenching, the efficiency
is predicted to be 3.7%. The dopant concentration must also be considered carefully: a 2-wt.% Yb
concentration maximizes the total extracted heat, while a 0.5-wt.% Yb concentration maximizes the cooling
efficiency. For an optimized large-core Yb-doped ZBLANP fiber, the absolute maximum heat extraction per
unit length predicted by our model is 161 mW/m. This value is primarily limited by the core area, which
stresses and quantifies the greater difficulty of achieving significant cooling in a fiber compared to a bulk
sample because of the much smaller size of a fiber core. This model can be readily used to explore solutions
5. References
47
14. S. Dai, J. Yang, L. Wen, L. Hu, and Z. Jiang, "Effect of radiative trapping on measurement of the
spectroscopic properties of Yb3+: phosphate glasses," J. Lumin. 104, 55 (2003).
15. A. A. M. Saleh, R. M. Jopson, J. D. Evankow, and J. Aspell, "Modeling of gain in erbium-doped fiber
amplifiers," Photon. Technol. Lett. 2, 714 (1990).
16. D. Marcuse, "Loss analysis of single-mode fiber splices," Bell Syst. J. 56, 703 (1977).
17. D. Marcuse, "Gaussian approximation of the fundamental modes of graded-index fibers," J. Opt. Soc.
Am. 68, 103 (1978).
18. M. J. F. Digonnet, "Broadband fiber sources," in Rare-earth-doped fiber lasers and amplifiers, Second
ed., Marcel Dekker, Inc., Ch. 6 (2001).
19. M. J. F. Digonnet, "Optical and electronic properties of rare-earth ions in glasses," in Rare-earth-doped
fiber lasers and amplifiers, Second ed., Marcel Dekker, Inc., Sec. 2.1.5 (2001).
20. S. D. Melgaard, A. R. Albrecht, M. P. Hehlen, and M. Sheik-Bahae, "Solid-state optical refrigeration to
sub-100 Kelvin regime," Sci. Rep. 6, 1 (2016).
21. D. V. Seletskiy, S. D. Melgaard, S. Bigotta, A. Di Lieto, M. Tonelli, and M. Sheik-Bahae, "Laser cooling
of solids to cryogenic temperatures," Nat. Photonics 4, 161 (2010).
22. T. L. Bergman, Fundamentals of Heat and Mass Transfer, Seventh ed., John Wiley & Sons, Ch. 13
(2011).
23. M. K. Davis, M. J. F. Digonnet, and R.H. Pantell, "Thermal effects in doped fibers," J. Lightwave
Technol. 16, 1013 (1998).
24. T. R. Gosnell, "Laser cooling of a solid by 65 K starting from room temperature," Opt. Lett. 24, 1041
(1999).
25. D. E. McCumber, "Theory of phonon-terminated optical masers," Phys. Rev. 134, A299 (1964).
26. M. Sheik-Bahae and R. I. Epstein, "Optical refrigeration," Nat. Photonics 1, 693 (2007).
27. B. C. Edwards, "Development of Los Alamos solid-state optical refrigerator," Rev. of Sci. Inst. 69, 2050
(1998).
28. J. M. Knall, A. Arora, M. Bernier, S. Cozic, and M. J. F. Digonnet, "Demonstration of anti-Stokes
cooling in Yb-doped ZBLAN fibers at atmospheric pressure," Opt. Lett. 44, 2338 (2019).
29. J. M. Knall, P.-B. Vigneron, M. Engholm, P. D. Dragic, N. Yu, J. Ballato, M. Bernier, and M. J. F.
Digonnet, "Experimental observation of cooling in Yb-doped silica fibers," Proc. SPIE 11298, 112980F-
1 (2020).
48
Chapter 4: Model of Radiation-Balanced Fiber Lasers
1. Introduction
To evaluate the potential of silica-based radiation-balanced fiber lasers (RBFLs), at the outset of this work
a model was needed to assess the expected output power and compare this performance to that of a ZBLAN
RBFL. Previously, there had been several theoretical reports on radiation-balanced solid-state and
semiconductor lasers (RBLs) [1-3] and fiber amplifiers (RBFAs) [4-6]. All the reports on solid-state RBLs
modeled Yb-doped crystals due to their high purity and low probability for concentration quenching. [1]
highlighted that the optimum pump wavelength is a trade-off between reducing the quantum defect and
reducing the laser efficiency because of pump saturation. For Yb-doped materials, the authors of [1] report
that this optimum is ~2% of the output power. [2] investigated the effect of cavity losses, output coupling,
and the radial distributions of the pump and laser modes. The report concluded that the main limitation to
power-scaling is the thermal gradients induced by radial intensity variations of the pump and signal powers.
Since successful demonstrations of ASF cooling in semiconductor remains elusive, [3] proposes an
alternative design in which the semiconductor laser is thermally coupled to ASF-cooled Yb:YLF crystal.
Since all these reports considered relatively short gain media, they justifiably neglected the position-
dependence of cooling within the cavity, an approximation that is generally invalid for a fiber. The RBFA
papers [4-6] presented and simulated several insightful designs that are later applied to RBFLs [4,7,8]. In one
design (discussed in more detail below), the core and inner cladding of a double-clad fiber are both doped.
Specifically, [5] proposes to dope the core with Yb and the cladding with Tm, pumping each separately to
induce gain in the former and cooling in the latter. [4] proposes another design in which only the core is
doped (in this case, with Yb) and the fiber is core-pumped at two wavelengths: one to induce cooling with
the Yb ions, and the other to create gain through the Raman amplification process. These studies all neglect
the absorptive loss and amplified spontaneous emission (ASE), mechanisms that we later showed (see
Chapter 3 and this chapter) need to be accounted for to predict temperature changes accurately in a fiber [9].
Also, most of these studies [4,7] focus on ZBLAN hosts. While these are critical steps toward the design of
optimum RBFLs, little work had been done to explore the vast parameter space required to maximize the
49
For bulk lasers, it is reasonable to pursue designs in which the lasing and cooling occur in the same volume,
as shown with highly doped Yb:YAG [10]. The comparatively large doped area of bulk lasers allows for
substantial cooling, enabling for example more than 4 W/m of heat extraction in Yb:YLF [11]. Since the
extracted heat scales linearly with the pumped doped area, it is much lower in single-mode fibers, typically
under 150 mW/m [12], which was predicted to severely restrict the output power of an RBFL [13]. An elegant
solution, first proposed in [4,5], consists of using a double-clad fiber in which the core and the cladding are
both doped with a rare-earth ion (Fig. 1a). The core is pumped to create gain for the signal, and the cladding
is pumped to induce cooling. The larger cladding can extract significantly more heat, enabling higher signal
powers while maintaining zero net heating. The design of Fig. 1a was expanded in [14] to include an undoped
inner cladding between the core and the doped cladding (Fig. 1b), enabling the laser to be cladding pumped
(in the inner cladding) independently from the cooling in the outer cladding and produce higher output
powers.
Figure 1: Cross-section of the (a) double-clad fiber considered in this work, in which both the core and the
cladding are doped with Yb3+, and (b) an alternate triple-clad design that incorporates an undoped inner
cladding to pump the core. (c) To create an RBFL, the double-clad fiber is sliced between two FBGs and
bidirectionally pumped into the doped cladding.
This chapter models the design of Fig. 1a to produce an RBFL with the general configuration of Fig. 1c. It
consists of a Yb-doped fiber spliced to two fiber Bragg gratings (FBGs) to form a resonator. The fiber is
pumped either unidirectionally or bidirectionally with light launched simultaneously in both the core and the
50
inner cladding to produce cooling in the cladding and gain in the core for lasing. The fiber is placed in air, so
that the heat flow in and out of the fiber is dominated by convection. Within this configuration, there is
enormous potential for customization. Design variables include the core and cladding diameters, the FBG
reflectivities, the concentration and type of rare-earth ions in each region, and the fiber length. The fiber can
be pumped at a single wavelength with one laser exciting both the core and doped cladding, or with two
lasers exciting the core and cladding separately, each with either the same wavelength or a different
wavelength to optimize the gain and the cooling rate independently. Additionally, the fiber can be
bidirectionally pumped. An informed comparison based on this model is now possible thanks to recent
cooling studies of Yb-doped silica [9,15] and ZBLAN [16,17] fibers (described in Chapters 5 and 6), which
provide firm experimental values for the model parameters that most critically control the cooling
Figure 2: (a) Signals and pump powers simulated by the RBFL model. (b) Example of the z-dependent profiles
of the cladding-pump powers, laser-signal powers, and upper-state populations. The ASE power distributions
are comparable to the signal powers except that they start at zero power at the ends of the fiber.
Ultimately, the model shows that a radiation-balanced fiber laser (RBFL) made of silica can produce
51
substantial output powers. Bidirectional pumping is found to reduce the average temperature of the laser,
enabling higher output powers while maintaining radiation-balanced operation. For a large-mode-area silica
fiber doped with Yb in the core and cladding, simulations predict that radiation-balanced output powers as
large as 115 W can be achieved by bidirectionally pumping the doped cladding. This is not only slightly
higher than the predicted powers for a Yb-doped ZBLAN fiber laser with the same dimensions, but the
temperature gradient in silica is also about half as large. Since silica is the most common host in fiber lasers,
these predictions are very promising for the near-term realization of practical RBFLs with substantial output
powers.
2. Model
The model used to simulate RBFLs was adapted from the model presented in the previous chapter, which
simulates ASF cooling in a conventional fiber with a Yb-doped core without an input signal or a resonator.
In that model, the laser rate equations were used to obtain the position-dependent population inversion of
Yb3+, and the evolution equations were used to describe the pump and ASE power propagation along the
fiber. These coupled equations were solved numerically using a Runge-Kutta method to obtain the position-
dependent powers and population inversion, which were then used to calculate the heat extracted per unit
length and associated temperature change at each position. All powers were assumed to propagate with a
Gaussian mode distribution. All the parameters relevant to cooling (absorptive and scattering losses,
concentration quenching, and radiative and nonradiative relaxation) were taken into account.
To adapt that model to cladding-doped RBFLs, four changes were implemented: (1) the pump equations
were modified to accommodate bidirectional pumping and simultaneous pumping of the core and cladding
with two different pumps, (2) laser signals were added, (3) the ASE convergence algorithm was improved to
accommodate higher pump powers and longer fibers, and (4) the pumps and signals were assumed to have
top-hat intensity distributions to simplify the closed-form solutions. Since the inner cladding is highly
multimoded, a top-hat distribution is a reasonable approximation. For the small, single-mode core,
simulations indicate that the absorption difference between a Gaussian and top-hat mode profile is small, and
it was neglected. This is particularly true for powers well above the saturation of Yb absorption, which is the
regime considered in this chapter. The absorption difference is also minimized by the presence of the doped
52
inner cladding. This can be understood by considering the case in which only the core is doped: the top-hat
mode, with all of its energy in the core, would absorb more pump power than the fiber’s actual LP01 mode,
whose energy is partly (10–30%) confined in the undoped cladding. However, in a fiber where both the core
and the cladding are doped, this difference in absorption is significantly reduced when the Yb concentration
in the core and the cladding is comparable. In the following, the dependencies on position z along the fiber
are accounted for but omitted in the notation to simplify the expressions. The symbols used to denote the
various pump and signal powers used in the model are listed in Fig. 2a. The core and cladding are doped with
Yb3+ (with possibly different concentrations) making the absorption and emission cross-sections as well as
the nonradiative and radiative lifetimes the same in both regions. This model can be applied to any other rare
The model presented here considers any combination of the following three pumps: (1) a forward cladding
( 0 (
pump 𝑃%,./ , (2) a backward cladding pump 𝑃%,./ , and (3) a forward core pump 𝑃%,.1 . The solid blue curves in
Fig. 2b show an illustrative example of the z-dependent power profiles of the cladding pumps (the profile of
the core pump has a distribution similar to the forward cladding pump). Their propagation along the fiber is
B
238,/A (
= <𝛾%,.1 − 𝛼5 @𝑃%,.1 (1b)
24
where the + sign refers to the forward pump power and the – sign to the backward pump power. 𝛾C,DE =
Γ:𝜎F <𝜆𝑝,𝑐𝑙 @𝑁H − 𝜎I <𝜆𝑝,𝑐𝑙 @𝑁J C + (1 − Γ):𝜎F <𝜆𝑝,𝑐𝑙 @𝑀H − 𝜎I <𝜆𝑝,𝑐𝑙 @𝑀J C and 𝛾%,.1 = 𝜎𝑒 <𝜆%,.1 @𝑁# − 𝜎𝑎 <𝜆%,.1 @𝑁*
are the absorptions experienced by the cladding and core pumps, respectively, and 𝛼5 is the total background
loss of the fiber, which is equal to the sum of the scattering loss 𝛼5I and the absorptive loss 𝛼5J (both assumed
to be independent of wavelength). 𝜎J and 𝜎K are the absorption and emission cross-sections at the pump
wavelengths 𝜆%,./ and 𝜆%,.1 , and Γ = 𝐴.1 /𝐴./ is the spatial overlap factor between the core area 𝐴.1 and the
cladding areas 𝐴./ . N1 and M1 are the lower-level populations in the core and cladding, and N2 and M2 are the
53
respective upper-state populations represented by:
n ε
Pp,cl +
Pp,co ( ) + +1
P n ε
Pase λi +1 ε
∑A + + ∑∑ ∑ l
N2 =
i=1
co
I
cl sat ,a (λ ) p,cl
co
Aco I sat ,a (λ ) p,co
A I (λ ) A I (λ )
ε =−1 i=1
N
co
cl sat ,a i ε =−1
co
co sat ,a l
+1 ε
P P +
P (λ ) P +1 n ε +1 ε 0
1+ ∑ p,cl
+ + ∑∑
p,co
+∑ ase i l
ε =−1 A I co
cl sat (λ ) p,cl
A I (λ )
co
co sat
A I (λ )
p,co
A I (λ )
ε =−1 i=1
co
cl sat i ε =−1
co
co sat l
(2a)
+1 ε
( )
Pp,cl +1 n ε
Pase λi
∑A + ∑∑
M2 =
ε =−1 ( ) Icl
cl sat ,a
( )λ p,cl
M
ε =−1 i=1
cl
Acl I sat ,a
λi
P+1 P (λ )ε +1 n ε 0
1+ ∑ + ∑∑ p,cl ase i
A I (λ ) A I (λ )
cl cl
ε =−1 ε =−1 i=1
cl sat p,cl cl sat i
(2b)
where the cladding pump and laser signal powers 𝑃/L at each position are summed over the propagation
L
directions (ε = + for forward and ε = – for backward), and the ASE powers 𝑃MNO (𝜆P ) are summed over
direction and over each wavelength 𝜆P in the discretized ASE spectrum (see Section 2.3). The saturation
.1/./ ST
𝐼IJQ,J (𝜆) = ' (3a)
/A//: U( (V)
.1/./ ST
𝐼IJQ (𝜆) = ' . (3b)
/A//: [U( (V)(UR (V)]
N0 and M0 are the total Yb densities in the core and cladding, respectively, where N0 = N1 + N2 and
M0 = M1 + M2. 𝜏.1 and 𝜏./ are the total upper-state lifetimes of the ions in the core and cladding, respectively,
where 1/𝜏.1/./ = 1/𝜏YZ + 1⁄𝜏ZJ2 + 1⁄𝜏!,.1/./ [19]. The radiative lifetime 𝜏ZJ2 and the nonradiative lifetime
𝜏YZ are the same in both regions, but the Yb concentration-dependent quenching lifetime 𝜏!,.1 in the core and
𝜏!,./ in the cladding differ since the concentrations are different in the two regions.
As the pump is absorbed along the fiber, the Yb ions in the core and cladding become excited (gray dashed
curves in Fig. 2b) and a z-dependent gain profile is induced, generating forward and backward ASE in both
the core and cladding. As explained in Chapter 3 [12], the ASE signals are modeled by dividing the gain
region into n = 100 wavelength bins centered at 𝜆P (1 < i < n) and of equal width δλ. The forward and
backward ASE powers in each bin are propagated along the fiber according to the following 2n coupled
differential equations:
54
± (V )
23(XR ± (𝜆 )
Y
= ±𝛾JIK (𝜆P )𝑃JIK P ± 𝛾IK (𝜆P ) (4)
24
where 𝛾JIK (𝜆P ) = Γ[𝜎K (𝜆P )𝑁# − 𝜎J (𝜆P )𝑁* ] + (1 − Γ)[𝜎K (𝜆P )𝑀# − 𝜎J (𝜆P )𝑀* ] is the wavelength-dependent
gain experienced by the ASE, and 𝛾IK (𝜆P ) = 𝑚𝜎K (𝜆P )𝑐 # ℎδλ(Γ𝑁# + (1 − Γ)𝑀# )/𝜆[P is the spontaneous
emission power per unit length that seeds the ASE signals [20]. m is the number of modes guided by the
doped inner cladding. For simplicity, m is assumed to be independent of wavelength and is set to V2/2 [21],
where NAcl is the numerical aperture of the doped cladding and dcl is its diameter [21]. The ASE powers must
0 (𝑧
𝑃JIK = 𝐿, 𝜆P ) = 0 (6b)
The main difficulty in solving these equations numerically is that the two boundary conditions (6a) and
(6b) are expressed at different locations. Consequently, one cannot start with known values of all powers at
z = 0 and solve the equations by calculating the powers iteratively from z = 0 to z = L with a Runge-Kutta
algorithm (this is all the more true when the laser is pumped bidirectionally, because the backward pump
power at z = 0 is also unknown). This difficulty was solved as follows. To find the solutions of the 2n coupled
differential equations for the ASE powers in addition to the (up to) three evolution equations for the pump
powers, the following parameters were iteratively changed one at a time: the population inversions (and
consequently, the gains 𝛾JIK (𝜆P ), 𝛾IK (𝜆P ), 𝛾%,.1 , and 𝛾%,./ ), the forward signals and pumps, and the backward
signals and pump powers. While finding these solutions, the laser signal powers were not updated -- they
were either set to zero or fixed at the solutions obtained from the algorithm described in the next section (as
dictated by the process described in Section 2.4). To start, the backward ASE signals/pump are fixed to zero
along the entire length, and the forward ASE signals/pumps at z = 0 are set according to their boundary
conditions (equal to the launched powers at z = 0 for the pumps and (6a) for the ASE signals). Using Euler’s
method to propagate (4), (1a), and (1b), the powers in the forward ASE signals/pumps are calculated along
the fiber length, and the population inversions are updated accordingly. Then the forward ASE signal/pump
55
powers and gains are fixed while the backward ASE signals/pump are set to their boundary conditions at
z = L and propagated backward according to (4) and (1a). Next, the gains are recalculated and subsequently
fixed along with the powers in the backward ASE signals/pump to update the powers in the forward ASE
signals/pumps. Finally, the gains are updated, and the process is repeated until successive iterations result in
negligible changes in the power distributions along z (typically, less than 25 iterations). This process results
in z-dependent power distributions for the pump and ASE signals. As described in the next two sections,
these solutions will need to be updated after new solutions for the laser signals are obtained.
If the launched pump power is above laser threshold, the pump-induced gain produces forward and
backward laser signals with z-dependent powers 𝑃/( and 𝑃/0 (red dash-dotted curves in Fig. 2b) that propagate
where 𝛾/ = 𝜎K (𝜆/ )𝑁# − 𝜎J (𝜆/ )𝑁* is the gain experienced by the forward and backward laser signals at
position z. These signals are constrained by the two boundary conditions at the reflectors:
where R1 and R2 are the power reflectivities of the high reflector and output coupler that establish the laser
cavity of length L, respectively. Like in the previous section, numerically solving these equations is difficult
because the boundary conditions (8a) and (8b) are expressed at different locations. In this case, an additional
challenge also arises from the fact that the boundary conditions merely define a relationship between the
forward and backward laser signal powers at z = 0 and z = L. This means that the initial powers for the laser
signals are unknown at both ends of the gain fiber. This unknown necessitates a different convergence
algorithm than the one used for the ASE signal powers. As a result, while solving for the z-dependent laser
powers, the ASE power distributions are fixed to the values obtained in Section 2.2 (the ASE powers are later
updated as described in Section 2.4). To solve for the z-dependent laser powers and appropriately update the
( ( (𝑧
pump power distributions, 𝑃%,./ (𝑧 = 0) and 𝑃%,.1 = 0) are initially set to the launched powers, and
56
0
𝑃%,./ (𝑧 = 0) is set to the value obtained from the solution of the ASE algorithm. Then, an initial value is
guessed for 𝑃/( (𝑧 = 0). 𝑃/0 (𝑧 = 0) is calculated according to the boundary condition (8a), then the signal
and pump powers are propagated along the fiber according to (7), (1a), and (1b). Based on the errors in
0
𝑃%,./ (𝑧 = 𝐿) and the relationship between 𝑃/0 (𝑧 = 𝐿) and 𝑃/( (𝑧 = 𝐿), the values for 𝑃/( (𝑧 = 0) and
0
𝑃%,./ (𝑧 = 0) are modified and the process is repeated until convergence (typically less than 150 iterations).
This process results in z-dependent power distributions for the laser signals and pumps, under the assumption
that the ASE powers are fixed at the values calculated in Section 2.2.
As described above, solutions for the laser and ASE power distributions are found using different
convergence algorithm, which means that the two methods must be implemented sequentially with an
iterative algorithm. To start, the laser signals are set to zero and solutions are found for the ASE and pump
powers (see Section 2.2). Then the ASE powers are fixed at these values while solutions are found for the
laser signals and the pump powers are updated (Section 2.3). The ASE and pump powers are then recalculated
with the laser signals fixed at these new values. This process is iterated until all power distributions have
The power distributions are then inserted into the following equation to determine the net heat extracted
per unit length and time at each location z along the fiber [12]:
dPε dPε +1 n dP ( λ )
ε
dQ +1 dP +1
⎛ +1 ε +1 +1 n
⎞
dt
= − ∑ ε p,cl − p,co − ∑ ε l − ∑ ∑ ε ase i − α bs ⎜ ∑ Pp,cl
dz dz dz ε =−1 i=1 dz ⎝ ε =−1
+ Pp,co + ∑ Pl ε + ∑ ∑ Pase ε
( λi )⎟
⎠
ε =−1 ε =−1 ε =−1 ε =−1 i=1
n
hc
− ( N 2Γ + M 2 (1− Γ ) ) Acl + ( N 2Γ + M 2 (1− Γ ) ) ∑ 2mσ e ( λi ) hc 2δλi / λi2
λ F τ rad i=1
(9)
The first four terms represent the difference between the powers entering and exiting the fiber cross-section
at location z, divided by dz to convert power into energy per unit length. These terms are therefore equal to
the power lost or gained within the cross-section due to absorption or emission by the laser ions and all loss
mechanisms. The fifth term represents the heat removed by scattering of the pump, laser, and ASE signals.
Since scattering loss contributes to optical attenuation but not to heating, this term must be subtracted to
offset its contribution to the first four terms. The sixth term (first term in the second line in the equation)
represents the heat extracted by the total spontaneous emission generated into all possible modes. However,
57
the spontaneous emission captured by the cladding modes was already included in the fourth term. The last
The temperature change along the fiber is then calculated by assuming a convective heat load on an infinite
cylinder:
2] *
∆𝑇 = 2Q $^ (10)
iS
where Df is the outer diameter of the fiber and h is the temperature-dependent heat transfer coefficient
TABLE I
PARAMETERS FOR THE YB-DOPED FIBERS, PUMP, AND LASER SIGNAL
Symbol Parameter Silica fiber ZBLAN fiber
dco core diameter 30 µm
Df outer fiber diameter 350 µm
h heat transfer coefficient (for Df = 350 µm) 81.4 W/m2/K
NAco core numerical aperture 0.07
NAcl cladding numerical aperture 0.45
Nc critical Yb3+ concentration 1.64x1027 Yb/m3 3.80x1027 Yb/m3
trad radiative lifetime 0.765 ms 1.7 ms
tnr nonradiative lifetime ~108 s [20]
aba absorptive loss 5 dB/km 25 dB/km
<lF> mean fluorescence wavelength 1004 nm 995 nm
lp pump wavelength 1030 nm
ls laser signal wavelength 1064 nm
R1 reflectivity of input FBG 100%
R2 output FBG reflectivity 1%
The parameter values of the Yb-doped fibers are listed in Table 1. For the silica fiber, the absorptive loss
(5 dB/km) and critical quenching concentration (1.64x1027 Yb/m3) are the values inferred for the highest
cooling silica fiber [9]. The radiative lifetime (0.765 ms), mean fluorescence wavelength (1004 nm), and
cross-section spectra are typical of aluminosilicate [26]. The ZBLAN fiber has a nominal absorptive loss of
25 dB/km and critical quenching concentration of 3.8x1027 Yb/m3, also inferred from cooling measurements
(see Section 2.6 in Chapter 5) [16]. The cross-section spectra and mean fluorescence wavelength (995 nm)
were obtained from [22]. The double-clad fiber is excited by up to three pumps launched into the core and
The design parameters of the radiation-balanced fiber laser modeled here are the core and cladding size,
58
dopant concentration, signal wavelength, cavity length, output coupling, and the power, wavelength, and
configuration of the pumps. Since an exhaustive exploration of this highly multi-dimensional space would
result in unintuitive results, this chapter illustrates trends for a few selected parameters while setting the other
low output reflectivity was selected (R2 = 1%), which minimizes the circulating laser power and has two
beneficial effects. First, it reduces depletion of the excited-state population by stimulated emission, which
increases cooling. Second, it reduces heating caused by absorption of the circulating power by impurities. As
in traditional cladding-pumped lasers, a larger core diameter increases the laser-signal gain, enabling the use
of shorter fibers. This choice is particularly important for RBFLs since it counteracts the need for longer
fibers imposed by pumping at weakly absorbing wavelengths. Therefore, the core diameter dcore was set to
the largest reasonable value for single-mode operation (30 µm). The pump wavelength (1030 nm) and signal
wavelength (1064 nm) were kept constant. Pumping at 1030 nm has been experimentally shown to induce
significant ASF cooling in both silica and ZBLAN fibers (see Chapters 5 and 6) [9,16], and simulations
indicate that this pump wavelength induces sufficient gain at 1064 nm for lasing. These four constant
parameter values are summarized in Table 1. The remaining parameters were swept to determine the
In previous reports on ASF cooling in fibers, the effect of ASE was found to be insignificant [12] because
the fiber had either a small doped area, a short length, and/or was pumped at low power. Also, to achieve
cooling, the pump wavelength had to be sufficiently longer than the wavelength of peak ground-state
absorption. All these features resulted in low gain and negligible ASE powers. However, RBFLs with high
output powers require large doped areas for sufficient cooling, long fibers to provide acceptable gain, and
high pump powers to create gain and cooling. ASE may then carry significant power and may have a non-
59
Figure 3: Temperature-change profile, computed with and without ASE, for an 8-m silica fiber laser
bidirectionally pumped with 200 W at 1030 nm.
To illustrate the impact of ASE on cooling, Fig. 3 shows the temperature profile of an 8-m fiber laser
bidirectionally pumped with 100 W in each direction, computed with and without the inclusion of ASE. The
core is 30 µm in diameter and doped with 2.5 wt.% Yb, and the cladding is 150 µm with 0.8 wt.% Yb (these
concentrations fall in the range of optimum values determined in later sections). When ASE is considered,
the temperature change is higher than when it is not, at all points along the fiber except near the center,
indicating that ASE significantly contributes to heating. Along the fiber, the total ASE power ranges from
2.1 W to 11.6 W, with an average of 4.5 W. ASE causes heating through two main mechanisms: (1)
absorption by impurities, and (2) the release of phonons due to the quantum defect. These effects increase
the average temperature change from 4.7 K without ASE to 6.8 K with ASE. ASE also affects the output
power. Without ASE, the laser output power is 83.6 W (an optical-to-optical efficiency of 41.8%), while with
ASE it drops to 72.0 W (36.0% efficiency). Clearly, the inclusion ASE is essential for accurate predictions
of RBFL performance.
The performance of unidirectionally and bidirectionally pumped RBFLs were simulated for a 10-m fiber
with the same specifications as in the previous section. For the bidirectional case, launching two 68-W pumps
resulted in zero average temperature change along the fiber and 37 W of output power. The z-dependent
forward (solid blue curve) and backward signal power (dashed blue curve) along the fiber are plotted in
Fig. 4a, and the temperature profile is shown in Fig. 4b. To obtain the same output power with a single pump,
60
113 W was launched at the input. The resultant signal power and temperature change profiles are shown by
the red curves in Fig. 4; the average temperature change is now 1.6 K.
Figure 4: (a) Forward (solid curve) and backward laser signal powers (dashed curve) calculated as a function
of position for two 37-W lasers with unidirectional and bidirectional pumping schemes. (b) The resulting
position-dependent temperature profiles.
In both cases, the temperature profiles (Fig. 4b) are not homogenous. Bidirectional pumping induces a
fairly symmetric profile in which the ends experience a positive temperature change and the middle is
consistently below ambient temperature. For a single pump, the temperature change is positive at the pump
input, briefly dips below ambient temperature as the signal amplification starts to decrease, then increases
and stays positive for the remainder of the fiber. The higher average temperature for the unidirectional case
can be mainly attributed to this later section of fiber. Pumping the cavity with one high-powered pump causes
a majority of the gain to be in the first half of the fiber (see solid red curve in Fig. 4a). Therefore, the forward
laser signal experiences rapid amplification at the start and maintains this high power for the remaining
length. In contrast, the bidirectional-pumping scheme distributes the gain more evenly. The forward laser
signal is amplified more gradually, which reduces the average power in the cavity. Since heating due to
61
absorptive loss is proportional to power, this configuration promotes a lower average temperature change in
the middle of the cavity, which negates the heating at the ends, and ultimately allows for higher output power
while preserving zero average temperature change. Given these significant benefits, the remainder of
When maximizing the output power of a cladding-cooled RBFL, certain considerations distinguish the
design process from that of conventional cladding-pumped fiber lasers. In an RBFL, significant pump
absorption by the laser ions occurs in the doped cladding as well as in the core. Therefore, special care must
be taken to ensure that enough power is simultaneously launched in the core for lasing and in the cladding
for cooling. In addition, significant fiber length is required in RBFLs to compensate for the reduced pump
absorption.
To estimate the maximum output power achievable in a bidirectionally pumped silica RBFL with the fiber
and pump parameters listed in Table 1, the cladding size, core and cladding Yb concentrations, cavity length,
and pump power were swept. For simplicity, the bidirectional pump powers were kept equal. The pump
beams were assumed to fill the entire cladding, i.e., as the cladding size was increased, the size of the beams
was increased proportionally. The doped-cladding diameter was limited to 300 µm since above this size re-
absorption of the radially escaping ASF becomes significant [23] and detrimental to cooling. In future work,
it may prove beneficial to expand this model to investigate larger diameters and take into account ASF re-
absorption.
The simulations showed that for a given cladding size, there are optimum Yb concentrations that maximize
the achievable output power: as the diameter is increased, the optimum cladding concentration M0 decreases
and the optimum core concentration N0 increases. Varying the cladding diameter from 145 µm to 300 µm
resulted in M0 optima between 0.9 and 0.6 wt.% Yb, and N0 optima from 2.3 to 2.5 wt.% Yb. The optimum
pump power was also found to increase slower than the rate required to maintain a linear relationship between
These trends can be understood as follows. Consider what would happen if the fiber laser was radiation
balanced and the cladding size was increased while keeping the pump intensity constant. The core diameter
62
is fixed, so the overlap of the core and cladding modes decreases. The portion of power absorbed by the
cladding then increases relative to the portion absorbed by the core. This causes the cooling in the cladding
to increase more than the heating in the core: the fiber is now overcooled. The first order correction is to
increase the core concentration (to increase the heat and output power) and decrease the cladding
concentration (to reduce the heat extraction). An additional correction is to increase the pump intensity. Since
the cooling mechanism is saturable while the lasing process is not, at high enough power, further pump-
power increase will cause proportionally more heating than cooling, not only bringing the laser back to
radiation-balanced operation, but also increasing the output power. However, the optimum increase in pump
power is not only determined by the relative increase in these heating and cooling mechanisms; it is also
highly influenced by parasitic heating associated with increasing the pump power (i.e., absorption by
impurities and higher ASE powers). Therefore, as the cladding area is increased, the increase in pump power
is limited such that the relationship between cladding area and maximum output power is sub-linear.
Figure 5: a) Calculated maximum output power and associated optical-to-optical efficiency for a silica (solid
curves) and ZBLAN (dashed curves) RBFL as a function doped cladding area (cladding diameter is labeled
above). b) The difference between the highest and lowest temperature along the fiber for the laser operating at
the maximum power.
To illustrate the effect of the cladding diameter, the core and cladding concentrations were set to the
optimum values calculated earlier for a fiber with a 300-µm cladding (2.5 and 0.6 wt.% Yb, respectively).
63
The bidirectional pump powers and cavity length were then swept to find the maximum radiation-balanced
output power for different cladding sizes. The results show that the achievable output power increases with
cladding size (solid blue curve in Fig. 5a) with a maximum output power of ~115 W. This trend makes sense,
since the cooling power increases with doped area [12]. The slower increase predicted at larger cladding
diameters can be explained by a lower pump overlap between the core and the cladding. This also explains
the decrease in efficiency (solid red curve in Fig. 5a). While the main thermal criterion for an RBFL is zero
average temperature change, it is also important to consider the temperature gradient along the fiber,
quantified by the difference between the highest and lowest temperatures. This is plotted as a function of
cladding area in Fig. 5b (solid curve). The temperature gradient increases linearly with cladding area. Since
the optimum pump power increases faster than the cladding area, the fiber ends experience larger positive
temperature changes due to impurity absorption and increased ASE power. Also, since the cooling area is
larger, the fiber mid-section experiences larger negative temperature changes, making the temperature
For comparison, these simulations were repeated for a Yb-doped ZBLAN fiber with the same core size
and a 300-µm cladding. The optimum concentrations are now 3.5 wt.% Yb (core) and 1.4 wt.% Yb
(cladding). As shown in Fig. 5a, the maximum output power is now ~110 W, which is slightly lower than for
the silica RBFL, and the temperature gradient (Fig. 5b) is about twice as large as it is with silica. Since the
optimum core and cladding concentrations are higher in ZBLAN, more pump is absorbed per unit length.
This produces more heating or cooling per unit length, thereby increasing the temperature difference between
the warmest and coolest points along the fiber. In short, silica not only enables higher output powers: it does
so with a lower temperature gradient, demonstrating that silica is (at least currently) a superior host for
RBFLs.
64
3.5 Comparison to a Conventional Cladding-Pumped Laser
To quantify the performance improvement enabled by a 115-W Yb-doped silica RBFL, its temperature
profile was compared to that of a conventional bidirectional cladding-pumped fiber laser with the same output
power. Both lasers have a 30-µm core doped with 2.5 wt.% Yb and a 300-µm cladding. The RBFL cladding
is doped with 0.6 wt.% Yb and pumped with two 400-W 1030-nm pumps, while the conventional laser is
pumped with two 64-W 976-nm pumps. The cavity lengths for the two lasers are 44 m and 1.0 m,
respectively. The other (spectroscopic) parameters for both lasers are summarized in Table 1. The resultant
temperature profiles are shown in Fig. 6. Clearly, the temperature change in the conventional laser (dashed
red curve) is consistently higher than that of the RBFL (solid blue curve), resulting in a large average
temperature change of 331 K. As discussed previously, the cooling systems required to mitigate this heating
adds bulk and instabilities to the laser. It also requires additional power – while the optical-to-optical
efficiency in this example is 90%, the external cooling power typically reduces the wall-plug efficiency to
25-50% [24]. However, this is still significantly higher than the wall-plug efficiency of the RBFL. To start,
the optical-to-optical efficiency of the RBFL is only 14%. Significant power is also required to cool the 400-
W pumps. Even if these pumps have the highest reasonable wall-plug efficiency (50%), 1600 W of wall-plug
power is needed to run the 115-W RBFL, resulting in a total wall-plug efficiency of 7.2%. Therefore, RBFLs
may not be appropriate for applications that prioritize efficiency over optical stability.
Figure 6: Temperature profile as a function of normalized cavity length predicted for a conventional cladding-
pumped fiber laser pumped at 976 nm and an RBFL pumped at 1030 nm, both with 115 W of output power at
1064 nm.
65
4. Summary
This chapter presents a versatile model of one of the most efficient RBFL configurations, which is a
cladding-cooled Yb-doped fiber laser. With the model, bidirectional pumping was shown to reduce the total
power circulating in the laser cavity, thereby decreasing the heating due to absorptive loss and enabling the
RBFL to achieve higher output powers than a unidirectionally pumped RBFL. The model predicts that 115 W
of output power can be achieved for a Yb-doped silica fiber with a 30-µm core diameter and a 300-µm doped
cladding bidirectionally pumped with two 400-W pumps. A Yb-doped ZBLAN fiber with the same
dimensions is predicted to produce a slightly lower power (110 W) and stronger temperature gradients.
Although the overall wall-plug efficiency of these RBFLs is substantially lower than that of a conventionally
cooled cladding-pumped lasers with the same output power, these predictions are promising for the future of
5. References
1. S. Bowman, "Lasers without internal heat generation," IEEE J. Quantum Electron. 35, 115 (1999).
2. M. Sheik-Bahae and Z. Yang, "Optimum operation of radiation-balanced lasers," IEEE J. Quantum
Electron. 56, 1 (2020).
3. J. Khurgin, "Radiation-balanced tandem semiconductor/Yb3+:YLF lasers: feasibility study," J. Opt. Soc.
Am. B 37, 1886 (2020).
4. G. Nemova and R. Kashyap, "High-power fiber lasers with integrated rare-earth optical cooler," Proc.
SPIE 7614, 761406-1 (2010).
5. G. Nemova and R. Kashyap, "Fiber amplifier with integrated optical cooler," J. Opt. Soc. Am. B 26, 2237
(2009).
6. E. Mobini, M. Peysokhan, and A. Mafi, "Heat mitigation of a core/cladding Yb-doped fiber amplifier
using anti-Stokes fluorescence cooling," J. Opt. Soc. Am. B 36, 2167 (2019).
7. G. Nemova and R. Kashyap, "Yb-doped fiber laser with integrated optical cooler," Proc. SPIE 7686,
768614-1 (2010).
8. X. Xia, A. Pant, E. J. Davis, and P. Pauzauskie, "Design of a radiation-balanced fiber laser via optically
active composite cladding materials," J. Opt. Soc. Am. B 36, 3307 (2019).
9. J. M. Knall, M. Engholm, J. Ballato, P. D. Dragic, N. Yu, and M. J. F. Digonnet, "Experimental
comparison of silica fibers for laser cooling," Opt. Lett. 45, 4020 (2020).
10. S. Bowman, S. P. O'Connor, S. Biswal, N. J. Condon, A. Rosenberg., "Minimizing heat generation in
solid-state lasers," IEEE J. Quantum Electron. 46, 1076 (2010).
11. S. D. Melgaard, A. R. Albrecht, M. P. Hehlen, and M. Sheik-Bahae "Solid-state optical refrigeration to
sub-100 Kelvin regime," Sci. Rep. 6, 1 (2016).
12. J. M. Knall, M. Esmaeelpour, and M. J. F. Digonnet, "Model of anti-Stokes fluorescence cooling in a
single-mode optical fiber," J. Lightwave Technol. 36, 4752 (2018).
13. M. Peysokhan, E. Mobini, A. Allahverdi, B. Abaie, and A. Mafi, "Characterization of Yb-doped ZBLAN
fiber as a platform for radiation-balanced lasers," Photon. Res. 8, 202 (2020).
14. J. M. Knall and M. J. F. Digonnet, "Anti-Stokes-fluorescence-cooled fiber-based gain element," Sept.
29, 2020. Pat. No. 10790633.
15. J. M. Knall, P.-B. Vigneron, M. Engholm, P. D. Dragic, N. Yu, J. Ballato, M. Bernier, and M. J. F.
Digonnet, "Laser cooling in a silica optical fiber at atmospheric pressure," Opt. Lett. 45, 1092 (2020).
66
16. J. M. Knall, A. Arora, M. Bernier, S. Cozic, and M. J. F. Digonnet, "Demonstration of anti-Stokes
cooling in Yb-doped ZBLAN fibers at atmospheric pressure," Opt. Lett. 44, 2338 (2019).
17. J. Thiede, J. Distel, S. R. Greenfield, and R. I. Epstein, "Cooling to 208K by optical refrigeration," Appl.
Phys. Lett. 86, 154107-1 (2005).
18. A. A. M. Saleh, R. M. Jopson, J. D. Evankow, and J. Aspell, "Modeling of gain in erbium-doped fiber
amplifiers," IEEE Photon. Technol. Lett. 2, 714 (1990).
19. W. Wang, L. Hu, W. Xua, S. Wang, L. Zhang, C. Yu, and D. Chen, "Spectroscopic properties of Ho3+
and Al3+ co-doped silica glass for 2-μm laser materials," J. Lumin. 166, 276 (2015).
20. M. J. F. Digonnet, Rare-Earth-Doped Fiber Lasers and Amplifiers, 2nd ed. Boca Raton, FL, USA: CRC
Press, 2001.
21. D. Gloge, "Weakly guiding fibers," Appl. Opt. 10, 2252 (1971).
22. J. M. Knall, A. Arora, M. Bernier, and M. J. F. Digonnet, "Anti-Stokes fluorescence cooling in Yb-doped
ZBLAN fibers at atmospheric pressure: experiments and near-future prospects," Proc. SPIE 10936,
109360F-1 (2019).
23. S. Dai, J. Yang, L. Wen, L. Hu, and Z. Jiang, "Effect of radiative trapping on measurement of the
spectroscopic properties of Yb3+:phosphate glasses," J. Lumin. 104, 55 (2002).
24. J. Hecht, "Photonic frontiers: High-efficiency optical pumping: 'Going green' cranks up the laser power,"
Laser Focus World, Apr. 13, 2016.
67
Chapter 5: Cooling in ZBLAN Fibers
1. Introduction
As discussed in Chapter 1, there have been many successful demonstrations of ASF cooling in bulk
samples placed in a vacuum [1-5], but little work on cooling in fibers. Yet, investigation of this domain is
critical for the application of ASF cooling to radiation-balanced fiber lasers. We started our investigation
with Yb-doped ZBLAN fibers because this material had already been proven, in a bulk form [6] and in a
fiber form [7], to be among the most efficient materials for ASF cooling. The model of ASF cooling presented
in Chapter 3 provided a solid understanding of cooling trends in such a fiber, and how to maximize the
amount of heat that can be extracted from a fiber [8]. Based on the knowledge provided by the model, we
designed an experiment that successfully demonstrated cooling in a single-mode and multimode fiber at
atmospheric pressure (Section 2). Specifically, we investigated cooling in two core-pumped Yb-doped
ZBLAN fibers. A maximum temperature change of -5.2 mK was measured in the single-mode fiber,
and -0.65 K in the multimode fiber. This was the first time ASF cooling was reported in a single-mode fiber,
and the first time cooling was demonstrated in any fiber at atmospheric pressure. By fitting the temperature
measurements to our model [8], we inferred values for the fibers' absorptive loss and critical quenching
concentration, two much needed parameters that most drastically limit the amount of achievable cooling.
Results from the two ZBLAN fibers not only validated our model for future development of devices that rely
on ASF cooling, but also confirmed the cooling trends predicted by the model. Specifically, we showed that
cooling is severely limited in single-mode fibers by their small doped area. The results also validated our
experimental set-up and confirmed the temperature sensor's ability to measure deviations within a few
After our success with the conventional Yb-doped ZBLAN fibers, we investigated cooling in fibers with
alternative designs (Section 3). Specifically, we explored the cooling potential of the design modeled in
Chapter 4 in which the core and cladding of a double-clad fiber are both doped with Yb3+. To make a
radiation-balanced laser or amplifier with this fiber, gain is induced in the few-moded core while cooling
occurs in the much larger doped cladding [9]. The main challenge with this geometry is that to take full
advantage of the large cladding and maximize heat extraction, the pump mode must fill the entire cladding
68
area. In this work we compared the performance of three fiber designs: the conventional large-core
multimode fiber discussed above, and two doped-cladding fibers, one with a cladding that has a double-D
shape, and the other with an octagonal cladding, both shapes designed to induce greater mode mixing. The
octagonal fiber is shown to provide significantly more uniform excitation of the doped cladding, resulting in
The two Yb-doped ZBLAN test fibers were fabricated by Le Verre Fluoré (LVF) in France [10]. Their
parameters are summarized in Table 1. The first fiber was polyacrylate coated and single-mode around 1 µm,
with a 3-µm-diameter core doped with 1 mol.% YbF3. The second one was an uncoated multimode fiber
(200-µm core diameter, 3 mol.% YbF3). Some of the parameters (Yb3+ concentrations, core sizes, and
numerical apertures) were determined through measurements at LVF. The effective mode diameter for the
single-mode fiber (4.0 µm) was calculated to be the 1/e2 intensity diameter; the nonradiative lifetime was
calculated using the known phonon energy of the fiber material [11]. The radiative lifetime was found by
measuring the decay of the fluorescence on the side of the multimode fiber. The absorption spectrum was
measured by LVF and the emission spectrum was calculated from it with the McCumber relation. The mean
fluorescence wavelength calculated from this spectrum is 995 nm, in agreement with [12,13]. More precise
values for the absorption and emission cross-sections at the pump wavelengths (1020 nm and 1025 nm), and
the effective mode area for the multimode fiber were found by fitting the model to absorption measurements,
as described in Section 2.4. Since the absorptive loss and quenching lifetime have a small effect on the pump
absorption, they could not be determined from this data. However, because these two parameters do have a
significant impact on the temperature change induced by ASF, they were determined by fitting the model to
the measured relationship between temperature change and pump power, as described in this chapter. Their
inferred values, listed in Table 1, were within reasonable bounds to permit cooling.
69
TABLE I
YB-DOPED ZBLAN FIBER PARAMETERS (VALUES WITH AN ASTERISK ARE GIVEN, ALL OTHERS ARE
FITTED OR MEASURED)
Parameter Single-mode fiber Multimode fiber
Yb3+ concentration* (N0) 1.64 1026 m-3 4.92 1026 m-3
radiative lifetime (trad) 1.87 ms 1.87 ms
nonradiative lifetime (tnr) ~108 s ~108 s
quenching lifetime (tq) >420 ms 78 ms
absorptive loss (aba) <12 dB/km 25 dB/km
core radius* (a) 1.5 µm 100 µm
effective mode radius (w*, b) 2.0 µm 26 µm
numerical aperture* (NA) 0.205 0.200
To measure the temperature change induced in the Yb-doped ZBLAN fibers, we used the custom
temperature sensor described in Section 2.3 of Chapter 2. The sensor FBG was placed along the side of the
Yb-doped fiber, and a good thermal contact was established between the two fibers, as discussed in the next
section. To demonstrate cooling, each Yb-doped fiber was core-pumped (see Fig. 1). The input end of the
single-mode fiber was polished at an 8° angle, then connectorized with angled connectors to a 10% fiber
splitter at the output of a 1020-nm fiber laser. The input end of the multimode fiber was flat cleaved, then
butt-coupled to the multimode fiber output of a 1025-nm semiconductor laser. The multimode fiber output
was sent to a thermal power meter to measure the output pump power. The pump wavelengths were chosen
to be near the optimum wavelength for cooling (1015 nm for the single-mode fiber and 1035 nm for the
multimode fiber), as determined by the model (they differ due to their dependence on Yb concentration and
core size). Deviations of these values from the optimum wavelengths were due to the availability of pump
sources. For each measurement, the pump was turned on for ~25 s, then turned off. The temperature change
was measured as a function of time as the temperatures of the two fibers equalized.
To accurately characterize the relationship between temperature change and pump power at the sensor
location, the power was first measured at the input of the single-mode fiber (using the 10% splitter) and at
the output of the multimode fiber (with the thermal power meter). Then, propagation of the pump power was
simulated (by the method discussed in Section 2.4) to calculate the absorbed power per unit length at the
location of the temperature measurement, which was 11.5 cm from the input for the single-mode fiber and
70
Figure 1: Experimental set-up used to measure the temperature change in the Yb-doped ZBLAN fibers.
2.3 Thermal Contact Between the Yb-doped Fiber and the FBG Sensor
Accurate temperature measurements of the test fiber rely on a good thermal contact between the test fiber
and the FBG so that the temperature is homogenous across the two fibers. Initially, thermal gel was placed
between the test fiber and the FBG to ensure a good thermal contact between the fibers [14-16]. However,
upon further investigation, the gel turned out to exhibit a small but excessive absorption in the wavelength
range of the ASF (~1 µm). As a result, the gel was reabsorbing the fluorescence escaping radially from the
doped fiber, as illustrated in Fig. 2a. This caused the gel to heat up a little, exposing the FBG to an extraneous
source of heat. The gel-induced heating was confirmed by performing temperature measurements on a Yb-
doped borophosphosilicate fiber with differing thicknesses of gel around the doped fiber (Fig. 2b). Keeping
all other factors constant, the positive temperature change induced in the fiber consistently increased with the
amount of gel.
In an effort eliminate the thermal gel, a solution was developed that involved the use of isopropanol. The
test fiber and FBG fiber are placed close together and a small amount of isopropanol is applied with a Q-tip.
The isopropanol wicks along the fibers and creates a capillary force that pulls the fibers together. After the
isopropanol (mostly) evaporates, the fibers are in intimate physical contact with no (or extremely little)
absorbing substance between them. This technique eliminates the erroneous source of heating introduced by
the gel and results in much more reproducible and accurate temperature measurements.
71
The isopropanol method was developed after a majority of the temperature measurements with the Yb-
doped ZBLAN fibers had been performed and published. Consequently, the results presented in Sections 2.5
through 2.8 were obtained while using thermal gel. Thus, the temperature changes reported for the single-
mode and multimode fibers are smaller than if the experiments were repeated with the isopropanol contact
method. This is verified in Section 2.9, in which the temperature change measured in the multimode ZBLAN
fiber without gel is four times larger than with the gel for the same pump absorption per unit length (-0.65 K
with gel and -2.5 K with isopropanol). Since the thermal gel essentially acts as a source of quenching, it is
likely that the values inferred for the critical quenching concentration in the ZBLAN fibers are slightly lower
Figure 2: (a) Although thermal gel creates a good thermal contact between the test fiber and the FBG, it induces
an erroneous source of heating as it absorbs the ASF escaping radially from the doped fiber. (b) The temperature
change induced by core-pumping a Yb-doped borophosphosilicate fiber with a constant power measured for
different thicknesses of gel applied to the Yb-doped fiber. As the amount of gel is increased, the temperature
also increases, confirming the heating effect of the gel.
Precise knowledge of the absorption and emission cross-sections is required to accurately model the pump
absorption in the ZBLAN fibers. To obtain the emission spectrum, it is common practice [17] to calculate it
from the absorption spectrum using the McCumber relation [18]. However, the absorption is relatively weak
at long wavelengths, making it difficult to obtain precise absorption cross-section values in this range. These
72
small inaccuracies can cause the calculated emission spectrum to diverge at long wavelengths, because in the
McCumber relation [18], σa(λ) is multiplied by a large exp(λ) term. This is illustrated in Fig. 3a, which plots
the absorption spectrum, measured by LVF, of bulk ZBLAN of a similar composition to the fibers, and the
emission spectrum calculated from it with the McCumber relation. The calculated spectrum is clearly in error
above ~1 µm, where its knowledge is most critical. To address this issue, we preformed more precise
absorption measurements at 1020 nm and 1025 nm, as described below. The absorption cross-sections
measured at these wavelengths (obtained from the measurements shown in Fig. 4a) were then used to adjust
the absorption spectrum at wavelengths greater than 1 µm. For wavelengths greater than 1040 nm, where the
calculated emission diverges more dramatically (see Fig. 3a), a correction factor was applied to ensure that
the absorption cross-section converges to zero at the longest wavelengths (above ~1080 nm) (see solid blue
curve in Fig. 3b). Now the emission cross-section spectrum calculated from this adjusted absorption cross-
section spectrum using the McCumber relation also converges to zero (see dashed orange curve in Fig. 3b),
and exhibits a shape similar to that of other reported emission spectra of Yb-doped fluorozirconate glass [19].
Figure 3: (a) The absorption spectrum measured by LVF, and the emission spectrum calculated with the
McCumber relation; (b) the same absorption spectrum, adjusted at longer wavelengths after additional
measurements as described in the text, and the emission spectrum calculated from it with the McCumber
relation.
To perform the absorption measurements, each fiber, of known length, was core-pumped with 1020-nm
light (for the single-mode fiber) or 1025-nm light (for the multimode fiber). The launching conditions for
each fiber were the same as those described for the temperature measurements (see Section 2.2). The output
73
power was measured versus input pump power. This resulted in the blue crosses shown in Fig. 4a for the
single-mode fiber and Fig. 4b for the multimode fiber. This dependency was then fitted to the following
equation, which describes the evolution of the pump power Pp(z) along a fiber with a radially symmetric
⎛ ⎞
dPp (z) a⎜ N 0σ a (v p ) ⎟
= −2π Pp (z) ∫ ⎜ − α b ⎟ f p (r)r dr
dz 0
⎜ 1+ Pp (z) f (r) ⎟
⎜⎝ I sat p ⎟⎠
(2)
where a is the core radius, σa(νp) is the absorption cross-section at the pump frequency νp, αb is the total fiber
loss due to impurity absorption and scattering, Isat is the absorption saturation intensity, equal to
hνp/(σa(νp)+σe(νp))τ, τ is the total lifetime including quenching, and σe(νp) is the pump emission cross-section.
Since both fibers have step-index profiles, fp(r) was approximated as a Gaussian for the single-mode fiber
and a top hat for the multimode fiber, with effective mode radii w and b, respectively. (2) was fitted to
minimize the χ2 between the measured and predicted output power dependencies. To find the best fit, the
parameters N0, w, a, and τrad were fixed to the values of Table 1. The total loss and quenching lifetime were
fixed at nominal values, since these parameters do not significantly affect the pump absorption or saturation.
This is due to the fact that pump attenuation is heavily dominated by Yb absorption, and hence insensitive to
fiber loss. Similarly, since the quenching lifetime is at least ten times larger than the radiative lifetime, it has
a negligible effect on pump absorption saturation. These parameters do, however, have a significant effect
on the temperature change and can, therefore, be determined by fitting the model to the temperature
measurements, as done further on. Best-fit values were found for the input coupling (the fiber is too fragile
to determine the input coupling with a cut-back technique) and small corrections were applied to scale the
cross-sections. For the multimode fiber, the effective mode radius b was also fitted, since the actual mode
The red curves in Fig. 4 show the resulting fits for the single-mode fiber (92.7 cm in length, Fig. 4a) and
the multimode fiber (29 cm in length, Fig. 4b). For the multimode fiber, the χ2 error between the data and the
fit was minimized for an input coupling of 89%, an effective mode diameter of 52 µm, and absorption and
emission cross-sections of 1.63x10-26 m2 and 2.94x10-25 m2 at 1025 nm, respectively, which corresponds to
a scaling factor of 0.70 between the measured cross-sections (Fig. 3a) and the corrected values (Fig. 3b). This
74
results in a fit that is in excellent agreement with the measured data (Fig. 4b). Since the core diameter
(200 µm) was significantly larger than that of the 1025-nm pump fiber (50 µm), the input coupling was
expected to be high and limited by (1) back-reflection at the interface between air and the ZBLAN fiber (~4%
loss), as well as misalignment and scattering on the polished fiber end (a few percent each). The fitted value
Figure 4: Experimental data (blue crosses) and simulated fits (solid red curves) for the output power as a
function of launched pump power for (a) the single-mode Yb-doped ZBLAN fiber, and (b) the multimode Yb-
doped ZBLAN fiber. The inset in (b) shows the mode profile measured at the output of the multimode fiber.
In an effort to verify the fitted effective mode diameter of 52 µm, the mode profile at the output of the
ZBLAN fiber was measured by scanning a fiber (6-µm core diameter) radially across its output and recording
the power collected by the fiber as a function of radial position. As expected, the measured profile was an
approximate top-hat distribution that filled most of the core (see inset of Fig. 4b). Its effective mode diameter
was at least 190 µm, much greater than the fitted value. Therefore, a thorough experimental investigation
was conducted in an attempt to explain the small effective mode radius (or equivalently, the apparently low
absorption saturation power of this fiber). Under the assumption that the modes in the ZBLAN fiber were
perhaps not sufficiently mixed, we applied various combinations of the following techniques: (1) static mode-
mixing of the modes in the multimode input fiber by looping it around a tight mandrel (two loops ~10 cm in
diameter), (2) dynamic mixing of the modes in the ZBLAN and in the input fiber by securing them to PZTs
driven at 20 kHz, and (3) increasing the transverse size of the excitation by butt-coupling a larger multimode
fiber (200-µm core diameter) between the 50-µm laser pigtail and the ZBLAN fiber. After each one of these
75
changes, the transmission curve of Fig. 4b was re-measured and found to be sensibly unchanged. We also
measured the spectrum at the outputs of the pump fiber and ZBLAN fiber and found nothing anomalous,
namely no lasing, and negligible power at shorter pump wavelengths (where the saturation intensity is lower).
For the single-mode fiber, the same corrective scale factor was used for the absorption cross-section at
1020 nm and the input coupling was fit to an optimum value of 25%. This fit (red curve in Fig. 4a) is also in
excellent agreement with the measured data points. From calculations based on the known mode intensity
diameters of the 1020-nm pump and the ZBLAN fiber, the input coupling was expected to be 75%. Deviation
from this value can be attributed to alignment errors and/or scattering at the fiber input due to imperfect end
polishing. The finalized cross-sections in Fig. 3b were obtained by applying the measured scale factor of 0.70
to the appropriate absorption values in Fig. 3a, and then computing the associated emission spectrum using
the McCumber relation. From this emission spectrum, the mean fluorescence wavelength <λf> for both fibers
The experimental results for the single-mode fiber can be seen in Fig. 5. The red curve is a pictorial
representation of the power from the 1020-nm laser coupled into the ZBLAN fiber, and the blue curve is the
temperature data from the sensor. The sensor data was fairly noisy because the temperature changes were
very small. Averaging the data along each pump-on and pump-off section gave the noise-free black curve,
which represents the average measured fiber temperature in each section. In spite of the noise, it is clear that
every time the pump was turned on, the fiber temperature dropped, on the order of a few millidegrees. For
12 mW of 1020-nm light at the location of the sensor, a temperature change of -5.2 mK was measured. The
limited amount of cooling can be attributed to the relatively small doped area. However, the fact that cooling
was demonstrated in a fiber with such a particularly small core is a testament to the low level of loss and
quenching in the ZBLAN fibers fabricated at Le Verre Fluoré. When published in 2019, this was the first
report of cooling of a single-mode fiber, and the first report of cooling of a fiber at atmospheric pressure.
76
Figure 5: Temperature measurement for the single-mode Yb-doped ZBLAN fiber end-pumped at 1020-nm to
induce cooling via anti-Stokes fluorescence. The blue curve is the temperature of the fiber measured over a
period of 90 seconds, during which the pump was alternatively turned on and off. The black curve is the
averaged data for each pump-on and pump-off section, and the pump-power on-off cycles are pictorially
represented by the red curve.
Figure 6: Temperature measurements (crosses) and fits (solid curves) plotted as a function of the pump power
at the FBG sensor location for (a) the single-mode fiber, and (b) the multimode fiber.
Since quenching and absorptive loss in the fiber have a significant impact on the temperature change,
these parameters can be quantified by fitting the model to the temperature measurements (using the
corrected spectra of Fig. 3b). With only one data point for the single-mode fiber (see Fig. 6a), a unique
fit was not possible. However, an upper bound for the absorptive loss can be found by fitting this point
assuming negligible quenching (τq = ∞) (solid green curve in Fig. 6a). Similarly, a lower bound for the
quenching lifetime was determined by finding the best fit when the absorptive loss was set to zero
(dashed red curve). This resulted in a maximum possible absorptive loss of 12 dB/km and a minimum
77
possible quenching lifetime of 440 ms (or, equivalently a minimum critical quenching concentration Nc
of 3.0x1027 m-3).
The very small amount of cooling predicted and observed in the single-mode fiber was limited by its
small core diameter (3 µm). Much larger temperature changes (~67 times) were measured with the multimode
fiber due to its large core. An illustrative temperature measurement for this fiber is shown in Fig. 7. As the
pump was abruptly turned on (represented pictorially by the solid red curve), the fiber temperature gradually
decreased and reached a plateau at -0.37 K. When the pump was turned off, the fiber temperature increased
back to room temperature. As expected, the temperature change is directly correlated to the pump power.
Additionally, the rise and fall time of the temperature change (3.77 s for the measurement in Fig. 7) agree
The measured dependence of the temperature drop on pump power for the multimode fiber is plotted in
Fig. 6b. Launching between 63 mW and 1.1 W of 1025-nm light resulted in temperature changes
from -51 mK to -0.65 K. Values for the absorptive loss and quenching lifetime were inferred by minimizing
the χ2 between the fit and these measured data points, which gave αba = 25 dB/km and τq = 78 ms (or
Nc = 3.8x1027 m-3). The value for the quenching lifetime is much shorter than the lower bound for the single-
mode fiber (422 ms) because the Yb3+ concentration in the multimode fiber is three times higher (see Eq. 1).
However, since the fibers have similar compositions, their inferred critical concentrations are expected to be
consistent, and they are: 3.8x1027 m-3 for the multimode fiber versus a lower bound value of 3.0x1027 m-3 for
the single-mode fiber. These values also agree with the value of 3.54x1027 m-3 calculated for the Yb-doped
ZBLAN sample in [20]. The higher absorptive loss inferred for the multimode fiber is likely due, in part, to
a higher concentration of OH- at the surface of the fiber. ZBLAN is hygroscopic [21] and since the multimode
fiber is unjacketed, it probably absorbed more OH- from the surrounding air over time than the jacketed
single-mode fiber, resulting in surface reabsorption of ASF. Extrapolating the fit for the multimode fiber
demonstrates that greater negative temperatures would have been achieved with a higher pump power (see
fit in Fig. 6b), namely -1 K for 1.6 W of pump power at the measurement location. To put this value in the
context of previous reports on ASF cooling of a multimode ZBLANP fiber in a vacuum (-65 °C) [7], the
78
model predicts that if this multimode fiber were instead cooled in a vacuum, the maximum temperature
Figure 7: An example measurement of the temperature evolution over time in the multimode Yb-doped
ZBLAN fiber as the pump was turned on and off.
To maximize the heat extracted per unit length in a fiber, it is important to investigate how it scales with
core size and Yb3+ concentration. In the limit of negligible absorptive loss, it is proportional to the doped area
and has on optimum with respect to the Yb3+ concentration [8]. Therefore, if the fiber absorptive loss was
reduced to zero, the maximum achievable temperature change in the multimode fiber using the inferred value
of Nc (3.8x1027 m-3) and assuming the pump modes fill the entire core is calculated to be 4290 times higher
than in the single-mode fiber. However, comparing the maximum values measured in the two fibers only
gives a factor of 120 increase. This difference is largely attributed to the difference in absorptive loss, which
was higher in the multimode fiber. Also, the effective mode diameter in the multimode fiber (50 µm) was
much smaller than the core diameter (200 µm), essentially reducing the size of the fiber and therefore,
reducing the effective area difference between the single-mode and multimode fibers. In short, cooling
increases with area, but optimum scaling will not be achieved unless the absorptive loss is significantly
79
2.8 Cooling Efficiency
The cooling efficiency is defined as the ratio between the power extracted and the absorbed pump power
(both per unit length). In an ideal fiber, with no loss, no quenching, and negligible non-radiative relaxation,
the cooling efficiency is equal to (λp - <λf>)/<λf> [22]. Therefore, the single-mode fiber, pumped at 1020 nm,
has an ideal cooling efficiency of 2.7%, and the multimode fiber, pumped at 1025 nm, 3.7%. However,
plotting the efficiency for each fiber as a function of pump power (Fig. 8) shows that the maximum measured
efficiency was lower for both fibers, namely 2.2% in the single-mode fiber and 0.90% in the multimode fiber.
These suboptimal efficiencies are due to the presence of loss and quenching, particularly in the multimode
fiber since it has a higher concentration, and hence higher quenching, and a potentially higher absorptive
loss. The plot also shows that the efficiency decreases with pump power, as expected, since the power
Figure 8: Cooling efficiency measured as a function of pump power at the sensor location for the single-mode
Yb-doped ZBLAN fiber (upper red cross) and the multimode Yb-doped ZBLAN fiber (lower blue crosses).
80
2.9 Cooling Record for Yb-doped ZBLAN Fiber
As stated, the previous results and analysis were performed using thermal gel to create a good thermal
contact between the ZBLAN test fiber and the FBG sensor, but the gel introduced an extraneous source of
heat. The gel was eliminated after developing a contact method involving the use of isopropanol, as described
in Section 2.3. To maximize the negative temperature change measured in a Yb-doped ZBLAN fiber, the
measurements were repeated for the multimode ZBLAN fiber using the isopropanol contact method. Like
before, temperature measurements were performed by core-pumping the fiber with the same multimode
1025-nm pump, butt-coupled to the test fiber. With the original gel-based set-up, the largest negative
temperature change induced in the ZBLAN fiber at atmospheric pressure was -0.65 K, as shown by the red
crosses in Fig. 9. With the new isopropanol method, much larger negative temperature changes were achieved
(blue squares in Fig. 9): for the same absorbed pump power per unit length of 2.3 W/m the steady-state
temperature change was four times larger (-2.5 K). Increasing the pump power further induced a temperature
change 5.4 times larger than the previous maximum. Specifically, for 3.3 W/m of absorbed pump power per
unit length at 1025 nm, a temperature change of -3.5 K was measured, setting a new record for a fiber laser-
Figure 9: Temperature change measured as a function of absorbed pump power per unit length at 1025 nm for
a multimode Yb-doped ZBLAN fiber. The red crosses represent the measurements performed with thermal gel
between the test fiber and the FBG sensor, and the blue squares show the data taken with the improved
isopropanol-based method.
81
3. ZBLAN Fibers with Yb-Doped Core and Cladding
The results from Section 2 confirmed the limited cooling achievable in conventional single-mode fibers,
and the improved cooling performance of multimode fibers. To overcome the limitation of single-mode
fibers, we experimentally investigated the cladding-doped fiber design originally proposed by Nemova and
Kashyap [9] that is modeled in Chapter 4. In this design, the large cladding is pumped and cools the fiber
independently from the size of the core (which can then support single-mode operation if need be). To assess
the cooling performance of the two cladding-doped ZBLAN test fibers, the temperature and absorption
measurements were compared to the results from the multimode Yb-doped ZBLAN fiber reported in
Section 2. As described in Section 2.4, the effective mode diameter inferred for the multimode fiber was
52 μm, significantly smaller than the 200-μm-diameter core. This inferred value equates to an effective mode
area that is 6.8% of the core area. To first order, this means that the fiber exhibited only 6.8% of the possible
cooling if 100% of the doped area had been excited. This early work made it clear that although a low filling
factor in a conventional amplifier or a laser can be offset by increasing the fiber length with minimal loss of
performance, in an ASF-cooled device this trade-off has no benefit and a low filling factor carries a significant
penalty. Therefore, the primary goal of our work on cladding-doped fibers was to establish a design that
The first fiber (Fig. 10a) was a custom double-clad fiber in which both the core and the cladding were
doped with Yb3+. For improved mode mixing, the cladding had a double-D shape. The core diameter was
10 μm, the cladding diameter was 125 μm, and the flat-to-flat distance was 115 μm. The Yb concentration
was 5 wt.% in the core and 1.4 wt.% in the cladding. The second fiber (Fig. 10b) was similar, except that its
cladding had an octagonal cross section for even better mode mixing. Its core had a 30-μm diameter, and its
cladding had a flat-to-flat diameter of 170 μm. The inner cladding was surrounded by a lower index outer
cladding made of a low-index resin with a relatively low absorption in the 1-μm region. The core was doped
with 3.5 wt.% Yb, and the inner cladding with 1.8 wt.% Yb. These concentrations were obtained from
82
the pump energy across the core and the cladding [24]. The higher Yb concentration in the core favors gain
over cooling, which turns out to be a necessary compromise, at least with this fiber design, to produce as high
a gain as possible while maintaining the average fiber temperature at room temperature. The fibers were
designed using our model and in collaboration with LVF, where the fibers were fabricated.
Figure 10: Cross section of the ZBLAN fibers with a Yb-doped core and cladding: (a) a double-D shaped inner
cladding and (b) an octagonal inner cladding.
To measure the temperature change induced in the cladding-doped ZBLAN fibers, we used a similar set-
up to the one depicted in Fig. 1 and described in Section 2.2. The pump light from the 1025-nm fiber-pigtailed
semiconductor laser (and on one noted occasion a 1030-nm high-power Yb-doped fiber laser) was butt
coupled into the ZBLAN fiber under test. In some measurements, to excite a more uniform mode distribution
across the cladding a short length of intermediary multimode fiber was placed between the pump laser fiber
pigtail and the ZBLAN fiber. In this case, one end of the intermediary fiber was butt-coupled to the laser
fiber, and the other end was butt-coupled to the ZBLAN fiber. The intermediate fiber had an octagonal-
83
3.3 Results
Using the protocol described in Section 2.4, absorption measurements were performed on the ZBLAN
fiber with the double-D shaped cladding. The experimental data points are shown by the blue crosses in
Fig. 11, and the associated fit is shown by the red curve. From this fit, the saturation power is inferred to be
220 mW. This corresponds to an effective mode area that is 6.2% of the cladding area, comparable to the
Figure 11: Measured output pump power versus input pump power at 1025 nm for the double-D ZBLAN fiber
when the pump was launched in the core and cladding, along with the theoretical fit (solid red curve) and the
expected outcome of the measurement had the pump energy filled the entire cladding uniformly (dashed purple
curve).
The insufficient mode-mixing is reflected in the poor cooling performance (blue crosses in Fig. 12a. The
maximum temperature change measured for this fiber was -78 mK for 240 mW, significantly less than the
expected cooling based on 100% filling of the doped cladding (purple dashed curve in Fig. 12b). In an effort
to increase the effective mode area, the cooling experiment was repeated after butt coupling a 38-cm length
of undoped multimode fiber (octagonal-shaped core with flat-to-flat diameter of 125 μm) between the laser
fiber pigtail and the ZBLAN fiber. This had a negligible impact on cooling, as shown by the green circles in
Fig. 12a. The procedure was then repeated for a longer ~1-m length of the same multimode fiber, which
84
Figure 12: Experimental results and theoretical predictions for the double-D cladding-doped ZBLAN fiber. (a)
Temperature change as a function of the pump power at the measurement location, with and without a mode-
mixing fiber at the input. (b) Comparison between the observed cooling and the cooling that was expected had
the pump energy filled the entire cladding uniformly.
Figure 13: Experimental results for the octagonal cladding-doped ZBLAN fiber. (a) The measured output pump
power as a function of input pump power (blue crosses) and a comparison with the model (solid red curve).
(b) Temperature change in the fiber as a function of the pump power (blue crosses) at the measurement location
and the model prediction using the filling factor of 38% inferred from the fit in (a) (sold red curve). The purple
dashed curves are the predicted behaviors had the pump filling ratio been 100%.
Absorption results for the ZBLAN fiber with the octagonal cladding are shown by the blue crosses in
Fig. 13a. Under direct excitation with the laser pigtail fiber, the mode filled a noticeably larger portion of the
cladding, resulting in a much larger inferred saturation power of 3.45 W. This corresponds to an effective
mode area that is 38% of the cladding area, a significant improvement over the ~6% inferred for the fibers
with a double-D and circular cladding shape. This improvement was reflected in the temperature
measurements, shown by the blue crosses in Fig. 13b. For this fiber, the maximum temperature change
measured with the 1025-nm pump was -1.3 K for 3 W of power at the sensor location. To improve on this
85
value, the same fiber was pumped with a more powerful 1030-nm laser (albeit with a possibly different filling
of the cladding because of a different mode distribution in the laser fiber). For 15 W at the sensor location,
the measured temperature change was -2 K (rightmost data point in Fig. 12b). The model prediction fitted to
these data points (solid red curve) using the effective area obtained from the pump-absorption measurement
is in excellent agreement. The purple dashed curve is the temperature dependence that was expected if the
pump beam had excited all the Yb ions uniformly in the cladding. At 15 W of pump power, the ratio between
the measured temperature change (-2 K) and the expected temperature change (-8.5 K) is ~23%, which is
approximately commensurate with the pump filling ratio (~38%). This analysis confirms the importance of
4. Summary
In summary, ASF cooling was demonstrated in several Yb-doped ZBLAN fibers at atmospheric pressure.
We started with two conventional fibers, one single-mode and one multimode, which led to the first
successful demonstration of cooling in a single-mode fiber, and of cooling in any fiber at atmospheric
pressure. The temperature changes measured in the multimode fiber were two orders of magnitude larger
than in single-mode fiber. This confirmed that cooling is severely limited in single-mode fibers and can be
dramatically improved by increasing the doped area. The doped area of the fiber must therefore be carefully
considered when designing fiber-optic coolers with high heat-extraction capabilities. This issue must also be
addressed in radiation-balanced fiber lasers – since cooling in a single-mode core is insufficient, it may be
necessary to implement cladding-cooled designs such the one modeled in Chapter 4. In this work, the cooling
capabilities of such designs were investigated in two different cladding-doped ZBLAN fibers. The results
indicated that sufficient mode-mixing within the cladding is required to capitalize on the increased cooling
offered by this design. For circular and double-D shaped guiding interfaces, the effective mode area was
inferred to be ~6% of the expected mode size. This dramatically reduced the measurable cooling below the
expected value for 100% mode filling. Mode-mixing was significantly improved with an octagonal cladding,
bringing the effective mode area to 38% of the cladding area. This improvement was reflected in the amount
of cooling that was achieved, namely a maximum temperature change of -2 K for 15 W at 1030 nm.
86
5. References
87
Chapter 6: Experimental Observation of Cooling in Silica Fibers
1. Introduction
Since commercial fiber lasers are predominantly made with silica, ASF cooling in this host is of the
utmost importance for the realization of practical RBLs. Until recently, however, cooling has only been
demonstrated in crystal and fluoride-based samples [1-3]. Relative to silica, fluorides suffer from fairly high
loss and brittleness, which have limited their progress and adoption -- the highest power ZBLAN laser is only
Since ASF cooling extracts relatively small amounts of power per unit volume, it is critical to minimize
incidental heating mechanisms that will either partially negate or overwhelm the heat extracted by ASF. The
two main sources of heat are (1) absorption by impurities [6], and (2) concentration quenching [7]. In silica,
the chemical vapor deposition methods employed to fabricate fibers enable near-intrinsic levels of purity and,
hence, very low losses. Therefore, impurities in rare-earth-doped silica fibers are primarily introduced
through contaminated dopant precursors [8]. Since similar concentrations of these dopant-related impurities
are also prevalent in ZBLAN, a host that has been successfully cooled, it is clear that concentration quenching
is the dominant obstacle to cooling silica. Since pure silica has a relatively high phonon energy (1100 cm-1)
[9], energy transfer between Yb ions and impurities is much more probable than in a lower phonon host such
at ZBLAN (phonon energy of 500 cm-1) [9]. This leads to a critical quenching concentration for Yb-doped
silica that is typically over a hundred times lower than for Yb-doped ZBLAN [10,11]. As a result, before this
thesis work silica could only be doped with relatively low Yb concentrations without suffering from
quenching. In the absence of quenching, the heat extracted by laser cooling is proportional to the Yb
concentration [12] (see Section 3.5 in Chapter 3). Therefore, a low dopant concentration limits the amount
of heat that can be extracted from the fiber. To achieve cooling in silica, it is crucial to increase the critical
quenching concentration. This can be achieved by ensuring that the Yb ions are uniform spaced. Eliminating
Yb clusters mitigates the larger probability of energy transfer between Yb ions facilitated by the close
In this work, we investigate the addition of either nanoparticles or network modifiers to mitigate
quenching in Yb-doped silica. In general, the test fibers were fabricated at Clemson or Mid Sweden
88
University, the spectroscopic properties were characterized at the University of Illinois at Urbana-
Champaign, and the cooling properties were characterized as part of this research effort at Stanford.
Preliminary tests (Section 3) show that the use of these techniques reduce heating by as much as 94% over a
Yb-doped silica baseline. With this knowledge, a Yb-doped aluminosilicate fiber was developed that enabled
the very first demonstration of cooling in a silica-based host (Section 4). In order to maximize cooling,
investigations were then performed to establish the influence on cooling of several fiber parameters, namely
the co-dopants concentrations, the core size, and OH- contamination (Section 5). This ultimately led to a new
cooling record in which a –70-mK temperature change was measured for 170 mW/m of absorbed pump
2. Background
Plotting the temperature change as a function of pump power at the measurement location enables
characterization of the cooling behavior of an individual fiber. However, to fairly compare the cooling
behavior of different fibers, the temperature change must be plotted as a function of pump power absorbed
per unit length at the measurement location. To illustrate this point, Fig. 1a shows the simulated temperature
change as a function of the pump power at the measurement location for two identical fibers that differ only
in their dopant concentration. The fiber with one fourth the concentration exhibits one fourth the heating.
However, this reduced heating is solely due to the fact that it absorbs one fourth the pump power: it is not a
reflection of the fiber’s ability to produce better cooling. If, instead, the temperature change is plotted against
the pump power absorbed per unit length (Fig. 1b), the curves lie on top of each other, as expected since the
fibers have the same cooling-related parameter values (in particular quenching and absorptive loss). In short,
plotting the temperature change as a function of pump power is useful for characterizing these parameters
for an individual fiber, and plotting it versus pump power absorbed per unit length is necessary for comparing
different fibers.
89
Figure 1: Temperature change as a function of (a) pump power and (b) absorbed pump power per unit length
(both at the location of the temperature measurement) for two identical fibers that differ only in their dopant
concentration.
To further understand the influence of absorptive loss and concentration quenching on temperature
change, a closed-form expression was derived for the net heat extraction dQ(z)/dt induced by ASF as a
2] V8 V
2Q
(𝑧) = 𝑃% (𝑧)𝛼5J 𝜂
⟨Vi ⟩ !
− 𝑃J5I (𝑧) k⟨V8⟩ 𝜂! − 1l (1)
i
where λp is the pump wavelength, <λf> is the mean fluorescence wavelength, Pp(z) is the pump power at z,
and Pabs(z) is the pump power absorbed per unit length at z. ηq is the quantum efficiency, equal to τq/(τq+τrad),
where τq and τrad are the quenching and radiative lifetimes, respectively. The first term in (1) accounts for the
heating due to the average absorptive loss 𝛼5J from fiber impurities at the ASF wavelengths, and the second
term accounts for the cooling due to ASF. To extract any heat from the fiber, the second term needs to be
positive, which means that the pump wavelength needs to be longer than the mean fluorescence wavelength.
However, in order to achieve net cooling, the second term cannot just be positive – it needs to be positive
enough to exceed the heating due to the absorptive loss. These factors were considered during the selection
of the pump wavelength as well at the composition of the silicate fibers, since the latter influences the mean
fluorescence wavelength.
The derivation of (1) can be understood as follows. The heat deposited in a thin longitudinal slice of fiber
is the difference between the pump power absorbed in the slice and the power escaping the slice in the form
90
of fluorescence [12]. The pump power absorbed per unit length is determined by two primary mechanisms:
(1) absorption by the Yb3+ ions, and (2) absorption by impurities. To determine the pump power absorbed
per unit length by the Yb3+ ions, the portion due to absorption by impurities abaPp(z) is subtracted from the
total pump power absorbed per unit length Pabs(z) to give Pabs(z) – abaPp(z). The absorbed pump power that
is able to participate in ASF cooling (i.e., the portion that is unquenched) is obtained by multiplying this
quantity by the quantum efficiency: ηq[Pabs(z) - abaPp(z)]. To determine the amount of power extracted by
ASF, this quantity is multiplied by the energy ratio between the pump and the fluorescence wavelength:
ηq[Pabs(z)-abaPp(z)] lp/<lf>. Finally, subtracting this value from the total pump power absorbed per unit
length gives the net heat deposited in the fiber per unit length:
2] V
(𝑧) = 𝑃J5I (𝑧) − 𝜂! <𝑃J5I (𝑧) − 𝛼5J 𝑃% (𝑧)@ 〈V8〉 (2)
2Q i
Rearranging to illustrate the influence of aba and lp gives the form presented in (1). Note that in order to
induce negative temperature changes, this expression must be negative, indicating net heat extraction.
Several techniques have been developed to reduce quenching in silica-based hosts, in particular the
addition of network modifiers or nanoparticles. In the first and most common technique, the network
modifiers (such as aluminum, phosphate, and/or boron) increase the solubility of Yb in the host matrix, which
reduces clustering and quenching [14]. The addition of fluorine serves to reduce the linear refractive index
[15] (i.e., offset the increased refractive index imposed by the Al2O3 and Yb2O3) as well as to dehydrate the
glass of OH- impurities, which otherwise promote extrinsic quenching. In the second technique, the Yb3+ ions
are sequestered in nanoparticles (e.g., LaF3, BaF2, or YbF3) evenly distributed in the fiber core. The
nanoparticles’ heavy cations (i.e., La or Ba) promote a locally low-energy phononic environment that
To assess the extent to which these potential solutions reduce quenching, we measured the ASF cooling
in six Yb-doped silicate fibers: three fibers with network modifiers (borophosphosilicate, fluorosilicate, and
aluminosilicate glasses) and three fibers doped with nanoparticles (LaF3, BaF2, or YbF3). The results were
compared to a commercial Yb-doped silica fiber from CorActive. Like most modern silica fibers, this fiber
91
was also co-doped with network modifiers (namely, Al and Ge) but in much smaller concentrations since the
primary purpose was to tailor the refractive index. While none of the six silicate fibers exhibited cooling, all
of them exhibited significantly reduced heating over the commercial silica fiber baseline. Specifically,
heating was reduced by as much as 93.9% in the BaF2 nanoparticle fiber. Comparable performances were
achieved with the borophosphosilicate (92.7%), aluminosilicate (90.5%), and LaF3 nanoparticle fibers
(89.6%). The fluorosilicate and YbF3 nanoparticle fibers did not perform as well, but it still exhibited reduced
The experimental set-up is very similar to the one used to measure temperature changes in the Yb-doped
ZBLAN fibers presented in Chapter 5 (see Fig. 2). The Yb-doped silica test fiber is placed in thermal contact
with the custom slow-light fiber Bragg grating (FBG) sensor, and the temperature change is measured by
interrogating the resultant spectral shift in the FBG (see Section 2.3 in Chapter 2). Since the isopropanol
contact method (see Section 2.3 in Chapter 5) had not yet been developed, thermal gel was used to create a
thermal contact between the FBG and the silica fibers tested in this section. Therefore, the temperatures
reported in Section 3 are slightly higher than if the experiments were to be repeated without thermal gel. The
trends, however, remain valid. For each measurement, the Yb-doped fiber was core pumped with a 1020-nm
fiber laser. The pump was turned on for ~25-60 s (long enough for the fiber temperature to reach steady
state), then turned off for the same amount of time, and the temperature change was measured continuously
as a function of time. An example measurement is shown in Fig. 3. The unabsorbed pump power at the output
of the doped fiber was measured with a thermal power meter. The unabsorbed power, along with the known
absorption values for each fiber, was then used to calculate the pump power and pump absorption per unit
length at the FBG sensor location, a known distance from the output of the fiber. These values were then
used in conjunction with the temperature measurements to create the power-dependent temperature curves
92
Figure 2: The experimental setup used to measure temperature changes in the core-pumped Yb-doped silica
fiber. As the temperature in the doped fiber and the slow-light FBG changes, the spectral shift induced in the
FBG is interrogated by a probe laser tuned to one of the resonance peaks of the FBG.
The parameters for the seven test fibers are summarized in Table 1. The aluminosilicate fiber samples
were provided by Shibin Jiang at AdValue Photonics, whereas the other specialty fibers were provided by
John Ballato from Clemson University, and characterized by Peter Dragic at the University of Illinois. The
Yb-doped silica fiber used as a baseline was purchased from CorActive. The Yb concentrations ranged from
0.5 to 6 wt.% Yb2O3. The fibers had core diameters around 6 to 10 µm, mean fluorescence wavelengths
between 990 nm and 1010 nm, and radiative lifetimes around 1 ms. The fibers were all core-pumped at
1020 nm, above their mean fluorescence wavelength, so they all had the potential to cool. The absorptive
loss, quantum efficiency, and quenching lifetime listed for most fibers were found by fitting the model
reported in Chapter 3 to experimental data. The absorptive loss in the silica and borophosphosilicate fiber
were fixed at nominal values since the data for these fibers was insufficient for a multi-parameter fit. The
aluminosilicate fiber was fabricated by the rod-in-tube technique and drawn at ~900oC. The
borophosphosilicate, silica, and nanoparticle fibers were fabricated using modified chemical vapor deposition
(MCVD) with solution doping, then drawn at ~2000oC. The fluorosilicate fiber was fabricated using the
molten-core method.
93
TABLE I
Yb-Doped Silica Fiber Parameters (Values With an Asterisk Are Fitted or Measured, All Others Are Given)
Host material Yb conc. Core dia. Mean fluor. Absorptive Quantum Radiative Quenching
(N0 ; (a ; µm) wavelength loss* efficiency* lifetime lifetime*
wt.% Yb) (λf ; nm) (αba ; dB/km) (ηq) (trad ; ms) (tq ; ms)
Silica 2.71 6.2 1010.2 15 37.8% 0.85 0.52
Borophosphosilicate 0.6 7 992.7 15 94.5% 1.4 24.1
Fluorosilicate 2.45 9.9 1001.3 570 89.2% 1.27 10.5
Aluminosilicate 6 10 1009.9 30 94.7% 0.743 13.3
BaF2 nanoparticles 0.5 10 995-1010 65-69 95.9-97.3% 0.760 17.8-27.4
YbF3 nanoparticles 1.5-2 10 995-1010 <400 82.4-83.7% 0.820 3.8-4.2
LaF3 nanoparticles 0.5 10 995-1010 130 94.9-96.3% 0.710 13.2-18.5
The measured temperature curve for the baseline Yb-doped silica fiber is shown in Fig. 3. As the
1020-nm pump was abruptly turned on to 40 mW at the sensor location (top blue curve), the fiber
temperature gradually increased according to a well-known 1 - exp(-t/t) law, where t is the thermal time
constant of the fiber in air, and reached a plateau at 6 K above room temperature (lower red curve). This
temperature increase was expected due to the presence of significant quenching in silica. When the pump
was turned off, the fiber temperature gradually decreased back to room temperature, following an
exponential law with the same thermal time constant as above (since cooling is dominated by air
convection) [16]. Additionally, the rise and fall times of the temperature change (12.7 s) agree very well
with COMSOL simulations, as discussed in [17]. To compare this temperature measurement with the
silicate fibers in the upcoming sections, the pump power at the temperature-measurement site (40 mW)
was converted into the corresponding absorbed pump power per unit length (0.31 W/m). This absorption
value was also used to fit the model [12] to the temperature measurements, which resulted in a quantum
efficiency of 37.8%. This means that 62.2% of the pump power absorbed by Yb ions is converted into
94
Figure 3: Temperature evolution (red curve) over time in the silica fiber as the pump was turned on and off
(blue curve).
For all three nanoparticle fibers, the temperature was also found to increase with pump power instead of
decrease (see Fig. 4). The temperature increased more rapidly at lower pump powers, and the rate of increase
rolled off for higher powers. This saturation behavior indicates that quenching, a saturable heating
mechanism, was still present in the nanoparticle fibers (see Section 2.2 in Chapter 2).
Figure 4: Temperature measurements (crosses) and fits (solid curves) plotted as a function of the pump power
at the FBG sensor location for the nanoparticle fibers.
To compare the cooling potential of the three nanoparticle fibers to each other as well as to the silica
baseline, the temperature was plotted as a function of the absorbed power per unit length (Fig. 5). From
95
Fig. 5a, it is clear that all three nanoparticle fibers performed significantly better than the silica fiber. To
quantify exactly how much better, Fig. 5b shows the same data zoomed in near the origin. For the absorbed
pump power indicated by the red ellipse (~0.038 W/m), the YbF3 nanoparticle fiber exhibited 73.0% less
heating, the LaF3 fiber demonstrated 89.6% less heating, and the BaF2 fiber exhibited the largest
improvement, with 93.9% less heating. Using our model, we were also able to quantify the quantum
efficiencies for the nanoparticle fibers to be between 82% and 97% – YbF3 exhibited the lowest and BaF2
exhibited the highest. This is significantly higher than the quantum efficiency exhibited by the silica fiber
(37.8%), which demonstrates that the nanoparticles effectively reduced concentration quenching compared
to the baseline.
Figure 5: Temperature measurements (crosses) and fits (solid curve) as a function of absorbed pump power per
unit length at the sensor location for the silica and nanoparticle fibers. (a) Expanded view; (b) same data zoomed
in near the origin.
The fibers with network modifiers also performed significantly better than the silica-fiber baseline (see
Fig. 6). Nevertheless, there is clearly still a significant amount of quenching in the fluorosilicate and
aluminosilicate fibers, as indicated by the concave curvature in the plot of temperature change versus pump
power (Fig. 6a). When considering the temperature dependence on pump power, the borophosphosilicate
exhibits significantly reduced cooling over the other two fibers. This can be attributed partially to its low Yb
content, which is 4 to 10 times lower than the other two fibers. When plotting against pump power absorbed
96
per unit length, the performance of the borophosphosilicate is much more comparable to that of the
aluminosilicate fiber. However, it still performs the best, and exhibits a 92.7% lower temperature change
compared with the silica fiber. The other two fibers also outperformed the silica fiber, with 90.5% less heating
in the aluminosilicate fiber and 81.8% in the fluorosilicate fiber. The improved performance is also
represented by the increase in quantum efficiency inferred from the fits in Fig. 6b, characterized as 89.2%
for the fluorosilicate fiber, 94.5% for the borophosphosilicate fiber, and 94.7% for the aluminosilicate fiber.
Figure 6: Temperature measurements (square points) and fits (solid curves) as a function of pump power (left)
and pump power absorbed per unit length (right) for the silica fiber and fibers with network modifiers.
To compare all seven fibers, the results were compiled in Fig. 7. All fibers showed significant
improvement over the commercial silica fiber, with as much as 93.9% less heating in the borophosphosilicate
fiber. The BaF2 nanoparticle fiber performed the best, with comparable performances from the
borophosphosilicate, aluminosilicate, and LaF3 nanoparticle fibers. This work emphasized the need to
dramatically reduce quenching in order to achieve negative temperatures in Yb-doped silicate fibers. In the
Yb-doped silica fiber, 6 K degrees of heating was measured for 40 mW (0.31 W/m absorption) of 1020-nm
pump power. The maximum temperature measured in the silicate fibers was an order of magnitude less than
this value, demonstrating significant progress towards ASF cooling in silicate fibers. This also confirmed the
97
Figure 7: Measured temperature change as a function of absorbed pump power per unit length for all tested
fibers described in Section 3.
With the knowledge from the previous section, we tested a Yb-doped aluminosilicate fiber that enabled
the first-ever demonstration of cooling in silica-based host. The silica fiber was designed by Magnus
Engholm at Mid Sweden University to have a high Yb concentration (2.06 wt% Yb) while exhibiting low
quenching. However, as pointed out earlier, such a high Yb concentration can have several adverse side-
effects, such as increased concentration quenching and a reduced nonradiative lifetime. These effects were
mitigated by doping the silica core with fluorine and alumina (Al2O3). The fiber also had a low loss, which
mitigated absorptive heating [18]. The addition of cerium (Ce2O3) to mitigate photodarkening [19] was not
included in this particular fiber in order to provide additional space for an even higher Yb concentration and
still ensure a low numerical aperture (NA) for the fiber to be few-moded. Care was also taken during the
preform fabrication process to ensure that all Yb ions were in their trivalent state, as Yb-doped fibers with
mixed valence states (i.e., Yb2+ and Yb3+) are known to have a higher background loss and a lower efficiency
[20]. Samples with lower efficiency require higher pump powers (and hence more parasitic heating) to
achieve saturation of the upper level, a necessary condition to maximize cooling. The fiber was fabricated
using conventional MCVD, followed by fiber drawing at about 2000°C, typical for fibers with a high silica
content. The fiber dimensions were also chosen in accordance with the current industry standard.
98
4.1 Characterization of the Yb-doped Silica Fiber
The parameters of the Yb-doped silica fiber are summarized in Table 2. The fiber had a core diameter of
21 µm and a numerical aperture of 0.070, so that it was slightly multimoded in the wavelength range of
interest (a V number of 4.6 at 1 µm.) The upper-state lifetime was obtained by end-pumping a 1-mm segment
of fiber with a 976-nm pulsed semiconductor laser (pulse width of 10 μs, pulse period of 6.5 ms, and peak
power of 300 mW) and measuring the decay of the fluorescence power at the output of the segment with an
avalanche photodiode. This resulted in an exponential decay with a single slope on the log-log scale,
confirming the absence of significant non-radiative relaxation mechanisms such as quenching. The upper-
state lifetime was taken to be the time from the end of a pulse to when the output power reached 1/e of its
original value. The fluorescence spectrum was measured by pumping the fiber continuously, collecting the
fluorescence emitted from the side of the fiber with a multimode fiber, and measuring the spectrum of the
collected light with a grating-based optical spectrum analyzer. The emission cross-section spectrum was
calculated from this spontaneous emission spectral intensity I(λ) via the well-known Füchtbauer-Ladenburg
equation [21]:
* cm
V e(V)
𝜎K (𝜆) = ' 2 n (3)
'() d$.Y ∫ 2 e(V)2V
n+
where n is the refractive index, c is the speed of light, 𝜆* and 𝜆# bound the wavelength range of the considered
transition, 𝜆̅ is the mean wavelength in this range, and τrad is the measured lifetime, assumed to be purely
radiative. The mean fluorescence wavelength was calculated from this spectrum with the following equation:
n U (V)
∫n 2 R & m 2V
V
〈𝜆\ 〉 = +
n U (V) (4)
∫n 2 R & r 2V
+ V
Cut-back measurements, like the ones described in Chapter 5, were performed to obtain absorption values
for the 1020-nm, 1030-nm, and 1040-nm pumps used for temperature measurements. For the other nine
wavelengths between 1030 nm and 1070 nm that were used to measure the temperature-change dependence
on pump wavelength, insufficient pump power was available to perform a cut-back measurement. The
absorption values at these intermediate wavelengths were either interpolated or extrapolated, which is a
reasonably accurate method since the dependence on wavelength is almost linear in this range.
99
TABLE 2
PARAMETERS FOR THE FIRST ASF-COOLED YB-DOPED SILICA FIBER
(VALUES WITH AN ASTERISK ARE FITTED, ALL OTHERS ARE MEASURED OR TAKEN FROM LITERATURE)
FIBER PARAMETER PARAMETER VALUE
Yb concentration (N0) 2.06 wt.% Yb
Al concentration (NAl) 0.86 wt.% Al
F concentration (NF) 0.88 wt.% F
Core diameter (2a) 21 µm
Numerical aperture (NA) 0.070
Mean fluorescence wavelength (<λf>) 1006.0 nm
Radiative lifetime (τrad) 0.890 ms
Nonradiative lifetime (τnr)
11 ~108 s
Quenching lifetime* (τq) 41.7 ±1.3 ms
Critical quenching concentration* (Nc) 16.7 ±0.25 wt.% Yb
Absorptive loss (aba) <5 dB/km
The concentration of Al, Yb, and F in the fiber (Fig. 8) was measured by Matthew Tuggle at Clemson
University using energy-dispersive X-ray spectroscopy (EDX) on a Hitachi SU5000 high-resolution scanning
electron microscope (HRSEM) in variable pressure mode with 50 Pa of pressure, 15 keV of accelerating
voltage, and a 10-mm working distance. Using the EDX data, the F, Al, and Yb concentrations averaged
across the area of the core (a radius of ~10 µm) are 0.88, 0.86, and 2.06 wt.%, respectively. The average Yb3+
Figure 8: Concentration of Yb and the co-dopants in the silica fiber as a function of radial distance from the
center of the core.
100
4.2 Experimental Set-Up and Measurement Procedure
The experimental set-up is shown in Fig. 2. To prepare the silica fiber for temperature measurements, a
~10-cm section of coating was stripped from the fiber in the vicinity of the measurement site. This was done
to ensure that the radially escaping ASF is not reabsorbed by the coating. The output of the pump source was
then connectorized to the fiber input end, and the stripped section was secured next to the FBG fiber.
Isopropanol was then applied between the two fibers, as described in Chapter 5. For each measurement, the
silica fiber was core-pumped at one of 12 wavelengths between 1020 nm and 1070 nm.
Figure 9: Example measurement of the temperature evolution over time in the cooled Yb-doped silica fiber as
the pump is abruptly was turned on 20 seconds after the start of the measurement. The gray sections indicate
the portions of data that were averaged to calculate the steady-state temperature drop of the fiber.
An example measurement result is shown in Fig. 9. First, the pump power was kept off for 20 s, then
abruptly turned on for 40 s, as shown by the red curve. The blue curve shows that when the pump was on,
the fiber cooled. The temperature of the cooled silica fiber equilibrated quickly (a few seconds) with the
FBG. The steady-state temperature drop is taken to be the difference between the average fiber temperature
during the first 20 seconds prior to turning the pump on (leftmost gray area in Fig. 9) and the last 20-second
period starting 20 seconds after the pump had been turned on (to make sure that the temperature had
stabilized) (rightmost gray area). For each measurement, the remaining pump power was measured at the
output of the silica fiber. The known values of the fiber saturation power, small-signal absorption, and
distance between the FBG and the fiber output were then used to calculate the absorbed pump power per unit
101
length at the location of the temperature measurement. Each measurement was performed four times to obtain
the average values shown in Figs. 10 and 11, and the standard deviation shown in Fig. 11 (omitted from
The steady-state temperature change was measured as a function of pump power at 1020 nm, 1030 nm,
and 1040 nm (see Fig. 10). Launching the 1020-nm pump induced a positive temperature change that
increased with pump power, with a maximum measured temperature change of 62 mK for 95 mW of pump
power at the measurement location, corresponding to an absorption of 0.46 W/m. Since this pump wavelength
is longer than the mean fluorescence wavelength (1006 nm), anti-Stokes fluorescence was indeed extracting
heat from the fiber. However, since the energy difference between the pump and the mean fluorescence is
relatively small, the extracted energy was insufficient to negate the small amount of heating due to
concentration quenching and pump absorption by impurities – the second term in (1) was too small.
Increasing the pump wavelength to 1030 nm increased the (𝜆% 𝜂! /⟨𝜆\ ⟩ − 1) term by ~50%, which was
sufficient to enable net cooling, as shown by the blue circles. As the pump power was increased, the
temperature change became more negative, with a maximum change of -20 mK for a pump power of 63 mW
at the measurement location (absorbed pump power of 0.29 W/m). Further increasing the pump wavelength
to 1040 nm created an even larger energy difference between the pump and the mean fluorescence
wavelengths, and a larger maximum temperature change of -50 mK. These data points, collected around
October 2019, represent the first experimental observation of laser cooling in a silica fiber.
The solid curves in Fig. 10 were fit to the experimental data using the same model of ASF cooling in a
Yb-doped fiber [12]. The saturation power and small-signal absorption of the silica fiber at each pump
wavelength were measured with a conventional cut-back method. Therefore, the only fitting parameters were
the critical quenching concentration Nc and the background loss αba. Since all of the temperature
measurements were performed on the same fiber, Nc and αba were constrained to have the same value for all
fits, and were chosen to minimize the c2 error between the measured and predicted temperature changes. The
fits yielded αba ≤ 5 dB/km and Nc = 1.28x1027 ±2% Yb/m3. This value of Nc is a factor of 17 larger than the
highest value reported for a Yb-doped silica fiber preform (7.5x1025 Yb/m3) [11], and comparable to values
102
reported for Yb-doped ZBLAN fibers (3.54x1027 Yb/m3) [12]. The critical concentration obtained for the
silica fiber corresponds to a quenching lifetime of τq = 41.7 ± 1.3 ms, one order of magnitude larger than the
radiative lifetime (890 μs). These values give a quantum efficiency ηq = τq/(τq+τrad) = 97.9 ±0.1%.
Figure 10. Measured dependence of the Yb-doped silica fiber temperature on pump power absorbed per unit
length at the location of the measurement for three pump wavelengths, along with fits from our ASF model.
Reconsidering the expression for heat extraction (1), it is clear that there are two competing parts in the
second term, 𝑃J5I (𝑧)(𝜆% 𝜂! /𝜆\ − 1). As the pump wavelength is increased, the second part, (λpηq/λf – 1),
also increases. However, since the absorption cross-section decreases with increasing wavelength after the
absorption peak at 976 nm, the first part, Pabs(z), decreases. Therefore, for a constant pump power, increasing
the wavelength above λf initially causes the temperature of the fiber to decrease, until at some wavelength
Pabs(z) is so weak (the pump wavelength is then at the upper edge of the Yb3+ absorption spectrum) that fewer
pump photons are absorbed per unit time, fewer fluorescence photons are emitted, and the temperature starts
to increase again. This trend is reflected in the temperature change measured for various pump wavelengths
with 18 ± 1 mW of pump power at the cooling site (red crosses in Fig. 11). At 1020 nm, λp/<λf> was too small
and the extracted energy was, therefore, overwhelmed by the residual heat induced by concentration
quenching and absorption by impurities. As the pump wavelength was incrementally increased above
1020 nm, the measured temperature change became negative and continued to decrease until the maximum
103
amount of cooling was measured at 1037 nm, with a maximum temperature change of –14 mK. Above this
wavelength, the magnitude started to decrease due to the reduced pump absorption. The blue curve in Fig. 11
is the prediction from the aforementioned model [12] calculated for this particular fiber using the parameter
values listed in Table 2. The excellent fit confirms the accuracy of our understanding of the cooling
mechanism, as well as the consistency of the inferred values of the critical quenching concentration and the
Figure 11. Measured temperature change as a function of pump wavelength in the Yb-doped silica fiber (with
one standard deviation of uncertainty) and the dependence predicted by the model in Chapter 3 [12]. An
optimum cooling wavelength exists around 1037 nm due to the competition between pump absorption strength
and the energy difference between the pump and the ASF.
There are many design considerations when manufacturing a fiber designed for cooling, including the Yb
concentration, the nature and concentration of co-dopants, the levels of various contaminants, the core
dimension, and the draw temperature. To maximize heat extraction in silica fibers, it is essential to develop
a quantitative understanding of how to choose these parameters such that aba is minimized and Nc is
maximized. In pursuit of this goal, we investigated six different Yb-doped silica fibers with various Yb
concentrations, core sizes, OH- concentrations, and co-dopant concentrations. By comparing their cooling
performances, valuable observations were made regarding the influence of these parameters on cooling.
Ultimately, a core composition was identified that allowed for a significantly higher Yb concentration than
104
previously possible without inducing quenching. The high Nc value inferred for this fiber led to a cooling
record of -70 mK for 170 mW/m of absorbed pump power at 1040 nm.
All six Yb-doped silica fibers were fabricated at Clemson or Mid Sweden University using MCVD. The
fibers were doped with Yb concentrations between 0.5 and 2.5 wt.% (assuming the same molar mass for all
hosts, so that 1 wt.% Yb corresponds to 7.67x1025 Yb/m3) and co-doped with various amounts of Al, F, and
Ce. They also had different core sizes and residual OH- levels, the latter determined by the absorptive loss
measured at 1380 nm, a harmonic peak of OH- absorption. Each fiber was characterized using the methods
described in Section 4.1 above; the resulting fiber parameters are summarized in Table 3. Fiber 1 has the
highest Yb concentration and the lowest OH- loss. Fiber 2 is the one reported in Section 4. Fiber 3 was drawn
from the same preform as Fiber 2 but at a higher temperature, which led to a ~20 dB/km lower OH- loss.
Fibers 4 and 5 have lower Yb concentrations, much higher OH- losses, and core diameters that are half the
size of the other fibers. Fiber 6 also has a low Yb concentration. Fibers 5 and 6 are the only fibers that are
co-doped with Ce. All fibers have a cladding with a 125-µm diameter.
TABLE 3
MEASURED AND INFERRED (*) YB-DOPED SILICA FIBER PARAMETERS
FIBER 1 FIBER 2 FIBER 3 FIBER 4 FIBER 5 FIBER 6
Yb concentration (N0) (wt.% Yb) 2.52 2.06 2.06 1.12 0.85 0.50
Al concentration (NAl) (wt.% Al) 2.00 0.86 0.86 0.72 1.33 0.74
F concentration (NF) (wt.% F) 0 0.88 0.88 0 0.59 0
Ce concentration (NCe) (wt.% Ce) 0 0 0 0 0.48 0.15
Core diameter (2a) (µm) 21 21 21 10 9 21
Mean fluorescence wavelength (<λf>) (nm) 1003.9 1006.0 1005.8 1007.2 1009.2 1007.3
Radiative lifetime (τrad) (ms) 0.765 0.890 0.895 0.790 0.780 0.805
Absorptive loss at 1.38 µm (aOH) (dB/km)* 12 100 82 352 272 114
The experimental set-up (Fig. 2) and measurement procedure are identical to those described in
Section 4.2. In short, each fiber was stripped of its coating at the location of the temperature measurement
and isopropanol was applied to create a good thermal contact between the test fiber and the FBG temperature
105
sensor. Each fiber was core-pumped at various powers of 1030-nm light, a wavelength chosen to be
sufficiently longer than the mean-fluorescence wavelengths of all fibers to be near the optimum cooling
wavelength. The silica fiber that exhibited the largest amount of cooling at 1030 nm (Fiber 1) was also core-
pumped at 1020 nm and 1040 nm to further characterize it and maximize the measured negative temperature
change. For each fiber, the temperature change induced by the 1030-nm pump is plotted as a function of
absorbed pump power per unit length at the measurement location (see Fig. 12). Plotting against this
parameter, as opposed to the pump power, enables a fair comparison of cooling performance since it removes
the variability in temperature change due to the difference in pump absorption per unit length between fibers
(see Section 2.1 above). Accordingly, each fiber is evaluated based on the percentage of absorbed pump
power that is converted into cooling. For each fiber, this data was fit to our model of ASF cooling in a fiber
(Chapter 3) to extract the absorptive loss and the critical quenching concentration (and the corresponding
quenching lifetime τq) of each fiber. The error bars were determined by treating each data point as a Gaussian
random variable with a 1 mK standard deviation and performing Monte Carlo simulations (500 trials) to find
the mean and standard deviation for the fitted aba and Nc values.
The temperature measurements for all six fibers (points) and resultant fits (solid curves) are shown in
Fig. 12, and the inferred values for Nc and aba are summarized in Table 4. Fiber 1 performed the best,
with -50 mK of cooling for 170 mW/m of absorbed pump power. Fibers 2 and 3 also exhibited significant
cooling. As expected, these two fibers have similar cooling performance since they originated from the same
preform and only differ in their draw temperature. The slightly better cooling of Fiber 3 may be attributed to
lower OH- levels. All three fibers exhibited absorptive losses less than 15 dB/km and Nc values greater than
15 wt.% Yb. The fit for Fiber 6 resulted in a similarly low absorptive loss (14 dB/km) but a much lower Nc
(3.46 wt.% Yb), which caused the fiber to heat, as shown in Fig. 12.
106
Figure 12: Measured dependence of temperature change on pump power absorbed per unit length at the location
of the measurement (at 1030 nm) for all six fibers, along with fits using the ASF-cooling model.
Fibers 4 and 5 exhibited heating and performed the worst of the six fibers. This is reflected in their high
absorptive losses (79 dB/km for Fiber 4 and 35 dB/km for Fiber 5) and low Nc values (6.78 wt.% Yb and
7.23 wt.% Yb, respectively). Their poor performance may also be attributed to their small core area (4 times
smaller than the other fibers) and lower Yb concentrations (up to 5 times lower), resulting in proportionally
smaller saturation powers and small-signal absorptions, respectively. This significantly limits the pump
power that can be absorbed before the Yb absorption saturates. Once this happens, the excess pump power is
primarily absorbed by impurities, which act as unsaturable absorbers that efficiently convert all of the pump
power that they absorb into heat. The onset of saturation is indicated by the elbow in the fits for Fibers 4, 5,
and 6 in Fig. 12. Where the curves become approximately vertical (shaded region in Fig. 12), any further
increase in pump power does not increase the (weak) cooling induced by ASF, since Yb excitation is
predominantly saturated. It does, however, increase the power absorbed by impurities, and thus heating. The
fits for Fibers 4 and 5 enter this near-vertical regime at much lower absorbed pump power than for the other
four fibers, highlighting the detrimental impact of a small core and of a low Yb concentration.
107
TABLE 4
INFERRED YB-DOPED SILICA FIBER PARAMETERS
Fiber Absorptive loss, at Nc,min for Fitted Nc Quenching lifetime
1030 nm λp = 1030 nm (wt.% Yb) τq (ms)
aba (dB/km) (wt.% Yb)
1 < 5.0 18.7 21.4 ± 1.3 37.4 ± 4.5
2 < 5.0 15.9 16.7 ± 0.25 41.7 ± 1.3
3 < 15 15.9 17.4 ± 0.40 44.6 ± 2.1
4 79 ± 3.1 8.9 6.78 ± 0.20 20.2 ± 1.2
5 35 ± 5.6 7.05 7.23 ± 0.70 39.4 ± 8.0
6 14 ± 6.7 3.92 3.46 ± 0.20 26.9 ± 3.2
To understand the influence of OH- contamination, the fitted absorptive loss at 1030 nm was plotted
against the measured absorptive loss at 1380 nm (Fig. 13), which is proportional to the OH- concentration in
the fiber core (typically, 55 dB/km of loss at 1380 nm corresponds to 1 ppm of OH- [21]). Figure 13 shows
that a higher loss at 1380 nm generally corresponds to a higher loss at 1030 nm (Pearson correlation
coefficient r = 0.926). In general, fibers with higher OH- losses also had lower critical quenching
concentrations, indicating that Nc may be a function of impurity concentration. This suggests that the
detrimental impact of OH contamination is two-fold: it not only causes heating through direct absorption of
the pump power, but it also increases the probability of quenching. More data is needed to confirm this
correlation. Fortunately, this temperature measurement and fitting procedure provides a relatively simple way
to accurately extract the Nc for a large number of samples and helps clarify this correlation in the future.
Figure 13 displays the temperature change measured in each fiber for 20 mW/m of absorbed pump power
per unit length at 1030 nm. Clearly, a lower OH- loss is strongly correlated to greater cooling. These trends
confirm the detrimental impact of OH- contamination and the need to minimize this quantity to maximize
ASF cooling.
108
Figure 13: The relationship between the absorptive loss measured at 1380 nm and the absorptive loss at
1030 nm obtained from fitting the model of Chapter 3 [12] to the temperature measurements of Fig. 12.
By correlating the cooling performance of each fiber to their composition, a few general trends were
tentatively inferred for the impacts of Al, F, Ce, and Yb on the quantum efficiency. Fluorine is used to tailor
the refractive index and to dehydrate the glass [15]. Some fibers doped with fluorine cooled whereas others
heated, suggesting that fluorine may not be not especially consequential for ASF cooling. Cerium is added
to mitigate photodarkening [19]. Though both Ce-doped fibers heated, they also exhibited higher OH- losses,
a known cause of heating. Accordingly, the effect of cerium on heating is presently inconclusive. Al is added
to reduce Yb clustering, and hence, quenching [14]. Indeed, Nc values generally increase with Al
concentration (see Tables 3 and 4), which translated into increased cooling, except for Fibers 4 and 5, which
were limited by small core size, low Yb concentration, and high OH- level, as pointed out earlier. On average,
fibers with higher Yb concentrations exhibited greater cooling, which shows that compositions that can
accommodate significant amounts of Yb without clustering are achievable. Previously, the optimum Yb
concentration for ASF cooling in silica was theorized to be less than 1 wt.% Yb [18]. Yet, Fiber 1 exhibits
significant cooling with 2.52 wt.% Yb, 2.5 times larger than this predicted optimum, and 30-40 % (in some
Glass composition influences ASF cooling in two ways: (1) it affects Nc by tailoring Yb solubility in the
host, and (2) it affects the mean fluorescence wavelength <λf> by modifying the emission spectrum.
109
Decreasing <λf> increases the average energy difference between the pump and fluorescence photons.
Therefore, for each pump photon that is converted into fluorescence, more energy is extracted. It is clear
from Table 3 that the fibers that exhibited cooling also exhibited the lowest <λf> values (under 1006 nm).
The mean fluorescence wavelength also determines the amount of quenching that a sample can tolerate before
cooling is unobtainable for a given pump wavelength λp [3]. More specifically, for a fiber with no impurity
absorption (aba = 0), heating is entirely due to concentration quenching, and, for cooling to take place, the
〈V| 〉/V8
𝑁.,gPY > 𝑁" {23 . (5)
<*0〈Vi 〉/V8 @
1
When impurity absorption is noticeable, the minimum tolerable quenching concentration must be even
higher than this threshold so that more Yb ions are available to compensate for heating due to impurities.
Therefore, it is important to identify fiber compositions that not only increase Nc, but also decrease <λf>.
Nc,min was calculated for each fiber (see Table 4). As expected, for all three fibers that exhibited cooling,
Nc > Nc,min. For Fibers 4 and 6, the fitted Nc is less than Nc,min. Therefore, even if the absorptive losses were
eliminated in these fibers, cooling would still be unobtainable due to quenching. However, this assessment
is based on the assertion that quenching and absorptive loss are completely decoupled, which may not be the
case, as noted in the previous section. If the critical quenching concentration does increase with increasing
impurity concentration, then minimizing the loss due to impurities would also increase Nc, potentially raising
the value of Nc for Fibers 4 and 6 above the minimum required for cooling.
Finally, to set a new baseline for ASF cooling in silica, a more comprehensive characterization of the best
performing fiber (Fiber 1) was carried out. The dependence of the temperature change on pump power was
measured for two additional pump wavelengths: 1020 nm and 1040 nm. The inclusion of this additional data
yielded more reliable values for the fitted absorptive loss and critical quenching concentration, since these
parameters are constrained to have the same values at all pump wavelengths. As expected, the fiber heated
when pumped at 1020 nm, and it exhibited the most cooling when pumped at 1040 nm (see Fig. 14). At
1020 nm, pump absorption is relatively large but the mean energy difference between the pump and
110
fluorescence photons is relatively small, such that the extracted energy is insufficient to negate heating due
to concentration quenching and absorptive loss. At 1040 nm, however, the trade-off between pump
absorption and energy difference is closer to the optimum and the fiber was able to cool down to -70 mK
below room temperature for 80 mW of pump power at the measurement location (170 mW/m of absorbed
pump power). The fits from the ASF cooling model (solid curves in Fig. 14) yielded αba < 5 dB/km and
Nc = 21.4 ±1.3 wt.% Yb (τq = 37.4 ±4.5 ms). This is the highest reported value for the critical quenching
concentration of Yb-doped silica, 30% higher than the previous record reported in Section 4 (16.7 wt.% Yb)
Figure 14: Measured temperature change of Fiber 1 as a function of pump power at the location of the
measurement for three pump wavelengths, along with fits from the ASF model described in Chapter 3 [12].
6. Summary
In this chapter, we reported on the first-ever demonstration of ASF cooling in a silica-based host. This
was achieved by first demonstrating that the addition of nanoparticles or network modifiers effectively
mitigates concentration quenching in Yb-doped silica. While none of the fibers in Section 3 exhibited
negative temperature changes, heating was reduced by as much as 94% over a conventional Yb-doped silica
fiber Cooling was achieved in a Yb-doped aluminosilicate fiber that exhibited significantly less quenching
and low impurity losses. This was achieved through the use of network modifiers and by minimizing the
111
presence of OH- and Yb2+. The critical quenching concentration for this fiber was 16 times higher than the
highest value ever reported in a silica host. This enabled cooling to be demonstrated in a silica fiber for the
first time, with up to -50 mK of cooling for 0.33 W/m of absorbed pump power per unit length at 1040 nm.
Next, an experimental investigation was presented to quantify the impact of several influential factors on
ASF cooling in silica fibers. This was accomplished by comparing the cooling performance of six Yb-doped
silica fibers with various network modifiers and network-modifier concentrations, OH- loss, and core
dimensions. Temperature measurements performed with a 1030-nm pump confirmed the correlation between
the loss measured at 1380 nm due to OH- absorption and the loss contributing to heating at 1030 nm,
highlighting the need to minimize OH- to maximum ASF cooling. Compositional analysis suggests that
fluorine may have a negligible effect on cooling. These results also showed that silica can accommodate
unexpectedly high Yb concentrations without suffering from quenching, in spite of its high phonon energy,
as demonstrated by cooling a fiber with 2.52 wt.% Yb. Cooling a fiber with such a high Yb concentration
was enabled by reducing the absorptive loss below 5 dB/km and co-doping the fiber with 2.0 wt.% Al,
leading to a record-breaking Nc of 21.4 ±1.3 wt.% Yb. In this fiber, a temperature change of -70 mK was
measured at atmospheric pressure for 170 mW/m of pump power absorbed at 1040 nm, a record for cooling
in a silica-based fiber. These results are instrumental to the understanding of ASF cooling in silica fibers and,
ultimately, to the future development of practical, silica-based radiation-balanced lasers and amplifiers.
7. References
1. D. V. Seletskiy, R. I. Epstein, and M. Sheik-Bahae, "Laser cooling in solids: advances and prospects,"
Reports Prog. Phys. 79, 96401 (2016).
2. C. E. Mungan, M. I. Buchwald, B. C. Edwards, R. I. Epstein, and T. R. Gosnell, "Laser cooling of a solid
by 16 K starting from room temperature," Phys. Rev. Lett. 78, 1030 (1997).
3. M. Sheik-Bahae and R. I. Epstein, “Optical refrigeration,” Nat. Photon. 1, 693 (2007).
4. Y. O. Aydin, V. Fortin, R. Vallée, and M. Bernier, "Towards power scaling of 2.8 μm fiber lasers," Opt.
Lett. 43, 4542 (2018).
5. [Online]. Available: http://www.ipgphotonics.com/en/products /lasers/high-power-cw-fiber-lasers/1-
micron/yls-sm-1-10-kw
6. P. C. Schultz, "Optical absorption of the transition elements in vitreous silica," J. Am. Ceram. Soc. 57,
309 (1974).
7. F. Auzel, G. Baldacchini, L. Laversenne, and G. Boulon, "Radiation trapping and self-quenching
analysis in Yb3+, Er3+, and Ho3+ doped Y2O3," Opt. Mater. 24, 103 (2003).
8. S. R. Nagel, J. B. MacChesney, and K. L. Walker, "An overview of the modified chemical vapor
deposition (MCVD) process and performance," IEEE Trans. Microwave Theory Tech. 30, 305 (1982).
9. M. J. F. Digonnet, Rare-Earth-Doped Fiber Lasers and Amplifiers, 2nd ed. New York, NY, USA:
Marcel Dekker, Inc., Ch. 2, Sec. 2.1.5 (2001).
10. J. M. Knall, A. Arora, M. Bernier, S. Cozic, M. J. F. and Digonnet, "Demonstration of anti-Stokes
112
cooling in Yb-doped ZBLAN fibers at atmospheric pressure," Opt. Lett. 44, 2338 (2019).
11. P. Barua, E. H. Sekiya, K. Saito, and A. J. Ikushima, "Influences of Yb3+ ion concentration on the
spectroscopic properties of silica glass," J. Non-Cryst. Solids 354, 4760 (2008).
12. J. M. Knall, M. Esmaeelpour, and M. J. F Digonnet, "Model of anti-Stokes fluorescence cooling in a
single-mode optical fiber," J. Lightwave Technol. 36, 4752 (2018).
13. F. Auzel and P. Goldner, "Towards rare-earth clustering control in doped glasses," Opt. Mater. 16, 93
(2001).
14. J. Ballato and P. Dragic, "On the clustering of rare earth dopants in fiber lasers," J. Dir. Energy 6, 175
(2017).
15. J. W. Fleming and D. L. Wood, "Refractive index dispersion and related properties in fluorine doped
silica," Appl. Opt. 22, 3102 (1983).
16. M. K. Davis, M. J. F. Digonnet, and R.H. Pantell, "Thermal effects in doped fibers," J. Lightwave
Technol. 16, 1013 (1998).
17. A. Arora, M. Esmaeelpour, M. Bernier, and M. J. F Digonnet, "High-resolution slow-light fiber Bragg
grating temperature sensor with phase-sensitive detection," Opt. Lett. 43, 3337 (2018).
18. E. Mobini, M. Peysokhan, B. Abaie, M. P. Hehlen, and A. Mafi, "Spectroscopic investigation of Yb-
doped silica glass for solid-state optical refrigeration," Phys. Rev. Appl. 11, 014066 (2019).
19. M. Engholm, P. Jelger, F. Laurell, and L. Norin, "Improved photodarkening resistivity in ytterbium-
doped fiber lasers by cerium codoping," Opt. Lett. 34, 1285 (2009).
20. J. Kirchhof, S. Unger, A. Schwuchow, S. Grimm, and V. Reichel, "Materials for high-power fiber
lasers," J. Non. Cryst. Solids 352, 2399 (2006).
21. O. Humbach, H. Fabian, U. Grzesik, U. Haken, and W. Heitmann, "Analysis of OH absorption bands in
synthetic silica," J. Non-Cryst. Solids 203, 19 (1996).
22. T. Feng, M. H. Jenkins, F. Yan, and T. K. Gaylord, "Arc fusion splicing effects in large-mode-area
single-mode ytterbium-doped fibers," App. Opt. 52, 7706 (2013).
23. J. M. Knall, P.-B. Vigneron, M. Engholm, P. D. Dragic, N. Yu, J. Ballato, M. Bernier, and M. J. F.
Digonnet, "Laser cooling in a silica optical fiber at atmospheric pressure," Opt. Lett. 45, 1092 (2020).
113
Chapter 7: Demonstration of a Radiation-Balanced Silica Fiber Amplifier
1. Introduction
After we demonstrated significant ASF cooling in a Yb-doped silica fiber, the next step was to use it to
assemble and test a laser-cooled fiber amplifier. The pursuit of radiation-balanced fiber amplifiers (RBFAs)
is critical for the realization of ultra-stable, low-noise laser light. The integration of ASF cooling into the gain
fiber will eliminate output instability caused by fluctuations in temperature and thermal gradients [1-3]. When
we started this work in 2016, the status of materials science dictated that fluoride fibers were the only
candidates for RBFAs [4,5] – cooling in silica was thought to be a highly improbable goal. The recent
material-science breakthrough that enabled cooling in silica gave the pursuit of RBFAs an even greater
motivation, since almost all fiber amplifiers in research, in R&D, and in commercial products are made of
Before we successfully demonstrated the first silica RBFA, there was a sizable amount of uncertainty as
to whether the small amount of cooling demonstrated in silica fibers (on the order of tens of mK) was
sufficient to overcome the heat generated by the addition of an amplified signal. The reason is that the Stokes
process used in lasers generates considerably more heat than the less probable anti-Stokes process used for
cooling. So, it was not at all obvious that the addition of a seed and the resulting heat induced by its
amplification would not overwhelm the weak cooling. Yet, simulations of the best performing silica fiber
(see Chapter 6) indicated that it was theoretically possible to achieve significant gain (>20 dB) while
maintaining radiation-balanced operation. The results reported in this chapter provide a definitive answer to
this question and show experimentally, for the first time, that with proper optimization of the amplifier
parameters, a fiber can provide significant amplification at the 150-mW output level with no net heating –
and much higher values are predicted after further optimization. This success paves the way toward the
creation of ultra-stable amplifiers (and lasers) that are necessary for applications involving metrology at the
highest precision.
114
2. Experimental Design
The amplifier was constructed with the best performing silica fiber reported in Chapter 6 (Fiber 1). The
fiber parameters, obtained with the procedure described in Chapter 6, are summarized in Table 1. The large
21-µm-diameter core is beneficial for ASF cooling, but also caused the fiber to be slightly multimoded
TABLE I
PARAMETERS FOR THE YB-DOPED SILICA FIBER USED IN THE
RADIATION-BALANCED FIBER AMPLIFIER
Fiber parameter Parameter value
Yb3+ concentration (N0) 2.52 wt.% Yb
Al- concentration (NAl) 2.00 wt.% Al
Small signal absorption at 1040 nm (a0,1040) 2.0 m-1
Saturation power at 1040 nm (Psat,1040) 452 mW
Small signal absorption at 1064 nm (a0,1064) 0.83 m-1
Core diameter (2a) 21 µm
Numerical aperture (NA) 0.13
Nonradiative lifetime (tnr) ~108 ms
Radiative lifetime (trad) 765 µs
Mean fluorescence wavelength (<λF>) 1003.9 nm
Quenching lifetime (tq) 38 µs
Critical quenching concentration (Nc) 21.0 wt.% Yb
Absorptive loss at 1040 nm (aba) 18 dB/km
The upper-state lifetime, fluorescence spectrum, emission cross-sections, and fiber composition were
measured using the procedure described in Section 4.1 of Chapter 6, and were already reported in that chapter.
Cut-back measurements were performed to obtain precise absorption values at 1040 nm and 1064 nm, which
were needed to carry out all simulations. Since the Yb-doped silica fiber is slightly multimoded, the
absorption parameter at each wavelength depends on the launching conditions. Therefore, to ensure that the
appropriate values were obtained, the absorption measurements were performed with the same experimental
set-up and launching conditions as used to demonstrate the radiation-balanced amplifier. For each
wavelength, the pump power was varied and the corresponding output power after 2.74 m of Yb-doped silica
115
fiber was measured with a thermal power meter. The fiber was then cut so that only 10 cm remained sliced
to the set-up. The power sweep was repeated and the output power remeasured. The red crosses in Fig. 1
represent the output power Pout measured before the fiber was cut versus the input power Pin measured after
the fiber was cut. This dependency was then fitted to the following absorption equation:
(1)
where L = 2.64 m is the length of fiber removed by cleaving. Since the fiber is multimoded, the fit assumes
a top-hat mode distribution for the power in the fiber. With a0 and Psat as fitting parameters, (1) was fitted to
minimize the χ2 between the measured and predicted output powers (blue curves in Fig. 1, which fit nicely
to the experimental data points). For the 1040-nm pump, this resulted in a0,1040 = 2.0 m-1 and
Psat,1040 = 452 mW. At 1064 nm, there was not enough power to reach Psat,1064 (calculated from measured
spectroscopic parameters to be ~420 mW). In this low-power regime, the first term in the exponential in (1)
approaches zero and the only fitting parameter left is a0,1064, which is all the data needed for this work since
the power at 1064-nm is maintained well below the saturation power. Fitting the small-signal absorption for
Figure 1: Results from cut-back measurements performed on the Yb-doped fiber at 1040 nm and 1064 nm. The
measured output power Pout is plotted as a function of pump power Pin launched into the fiber for each
wavelength, and the data points are fitted to a model of saturated absorption in a fiber (equation 1).
116
Temperature measurements were performed on this fiber in the absence of an input signal (seed) to infer
the residual absorptive loss aba due to impurities and the critical quenching concentration Nc of the Yb-doped
fiber. The FBG temperature sensor was placed in contact with a stripped section of the Yb-doped silica fiber
60-cm from the input end, and a small amount of alcohol was applied between the fibers for better thermal
contact. The doped fiber was core-pumped at 1040 nm and the induced temperature change was measured at
the FBG location for various powers up to 1.8 W (red crosses in Fig. 2). The fiber was tested with much
higher pump powers than used for the work reported in Chapter 6 (~2 W here, compared to 100 mW in
Chapter 7) because significantly higher pump powers were needed to obtain substantial gain in the fiber
amplifier. As the pump power is increased from zero, more and more Yb ions are excited, resulting in
increased ASF emission and greater cooling. In this regime, the slope of the temperature curve is largely
influenced by the quantum efficiency of the Yb ions, which is determined by Nc. At large pump powers (well
above saturation of the Yb absorption), the excitation of the Yb ions is saturated and the cooling rate reaches
a maximum. However, at large pump powers any pump power in excess of the saturation power is absorbed
by impurities, which contributes to increased heat generation. Since this absorption mechanism is essentially
unsaturable, heating increases in proportion to this additional pump power. As a result, at large pump power
the temperature trend is reversed, as observed in Fig. 2, and the fiber starts warming up again, ultimately
increasing above room temperature (this crossing point was not reached in this test for lack of sufficient pump
power). In this regime, the slope is largely determined by aba. From Fig. 2, it is clear that these cooling and
heating mechanisms create an optimum pump power that maximizes the extracted heat per unit length. For
The data was then fitted to our model of ASF cooling in a fiber [7] to minimize the χ2 between the
measured and predicted pump power dependencies (solid blue curve in Fig. 2). Since the pump absorption
coefficients were independently measured, the only fitting parameters were the critical quenching
concentration and the absorptive loss. These parameter values were inferred to be Nc = 21.0 wt.% Yb
(1.63x1027 Yb3+/m3) and aba = 18 dB/km. As discussed above, Nc predominately affects the fit below the
optimum pump power (510 mW), while aba determines the slope for powers above this optimum. Since pump
117
powers less than 100 mW were used for the temperature measurements reported for the same fiber in
Chapter 6 (Fiber 1), only the initial slope was characterized in that chapter. This enabled a reliable fit for Nc,
but a broad range of values were possible for aba. In contrast, the significantly higher powers used in this
chapter resulted in a definitive curvature (Fig. 2) that enabled a much more reliable fit for both parameters.
It follows that the newly inferred value for Nc is comparable to the one reported in Chapter 6, but that the
values for aba have a larger deviation. While the aba reported in Chapter 6 (<5 dB/km) resulted in the
minimum χ2 for the data presented in that chapter, using 18 dB/km results in a very comparable fit with an
error that is within a few percent of the minimum. This confirms the reliability and increased accuracy of the
Figure 2: Temperature change as a function of 1040-nm pump power at the measurement location for the Yb-
doped silica fiber presented in this work. The data is fit to our model of ASF cooling in a fiber [7] to infer the
absorptive loss and critical quenching concentration.
The experimental set-up for the laser-cooled silica fiber amplifier is shown in Fig. 3. The Yb-doped silica
fiber was core-pumped at 1040-nm to create both cooling and gain at 1064 nm. A 90/10 fiber splitter
combined the pump with the 1064-nm seed. The splitter output that carried 90% of the pump power and 10%
of the seed power was spliced to the input of the Yb-doped silica fiber. At the amplifier output, a long-pass
118
filter removed the residual pump power and passed the signal to a power meter. The net gain of the amplifier
was defined as the ratio of the output power at 1064 nm and the signal power launched into the fiber.
The splitter output that carried 10% of the pump power and 90% of the seed power was used as a power
tap to measure the power launched into the fiber amplifier. The ratio between the tapped power and the power
launched into the amplifier was determined after all the temperature and gain measurements were completed,
and after the cut-back measurement described in the previous section was performed. After the cut-back
measurement, only 10 cm of the doped fiber remained spliced to the splitter output. First, the 1040-nm pump
was turned on and the output power from the 10% tap and the 10-cm section of doped fiber was measured.
The 1040-nm pump was then turned off and the measurement was repeated for the 1064-nm pump. Using
the known absorption values for each wavelength, the output power measured after the doped fiber was
inserted into (1) to calculate the power launched into the fiber. The ratio between the output power of the
10% tap and the power launched in the Yb-doped fiber was then calculated.
Figure 3: Experimental set-up used to measure the temperature change in the Yb-doped silica fiber amplifier.
The induced temperature change was measured at seven locations along the Yb-doped fiber using the
custom slow-light FBG temperature sensor. At each measurement location, a short length (~10 cm) of the
Yb-doped fiber was stripped of its jacket to prevent reabsorption of the radially escaping ASF. The stripped
119
section was then placed in contact with the FBG sensor (see Fig. 3) and a small amount of isopropanol was
Figure 4 shows an exemplary temperature measurement of the fiber amplifier. At time t = 0 s, the pump
was abruptly turned on, causing the fiber to cool to ~130 mK below room temperature. After 15 s, the seed
was also turned on, causing the fiber to heat up slightly (~20 mK) to a new steady-state value of ~110 mK
below room temperature. This small amount of heating was due to the phonons released by the amplification
of the 1064-nm signal. The temperature change of the fiber amplifier was defined as the difference between
the average temperature in the first 5 s when both the pump and signal were off and the average temperature
in the last 5 s after both had been turned on. Each measurement was repeated three times and averaged.
Figure 4. Measured temporal trace of the temperature change in the Yb-doped silica fiber as the 1040-nm pump
and 1064-nm seed are sequentially turned on, launching 1.64 W and 3 mW in the fiber core, respectively.
3. Experimental Results
A 2.74-m section of the Yb-doped silica fiber was core-pumped with 1.12 W at 1040 nm and seeded
with 3 mW at 1064 nm. This resulted in negative average temperature changes at all seven measurement
120
locations (blue squares in Fig. 5) and 5.7 dB of small-signal gain (11.7 mW of output power at 1064 nm). As
explained in relation to Fig. 2, the optimum pump power that maximizes the extracted heat per unit length
for this fiber is 510 mW. In this experiment, the input pump power was 1.12 W and the residual pump power
at the output of the fiber amplifier was only 80.2 mW. It follows that the negative temperature changes are
smaller near both ends, where the pump power is either above (at the input) or below (at the output) this
optimum value (510 mW); the lowest temperature is observed near the middle of the fiber, where the pump
Figure 5. Average measured temperature change (n = 3) at seven locations along a 2.74-m silica fiber amplifier
for three different pump powers at 1040 nm. The fiber is core-pumped to create gain at 1064 nm for the 3-mW
seed.
When the launched pump power was increased to 1.38 W, the signal output increased to 26 mW (a gain
of 9.1 dB). As expected, the negative temperature change (green asterisks in Fig. 5) at the start of the fiber
was now smaller than in the first case (blue squares) because the input pump power was further above the
510-mW optimum. This increase in launched power also resulted in a higher pump power at the fiber output
end (154 mW). Since this value was now closer to the optimum, the output end of the fiber experienced a
larger negative temperature change. The trend continued when the pump power was further increased to
1.64 W (red crosses in Fig. 5): cooling decreased near the input end and increased near the output end, where
121
the residual pump power was now 247 mW. The signal amplification also increased, resulting in 11.4 dB of
To determine the degree of uncertainty in the measured temperatures (Fig. 5), Figure 6 (a-c) plots in
separate frames for each pump power the three data points measured at each location. The solid curves in
Fig. 6 are simulation results from the model of a radiation-balanced fiber laser described in Chapter 4, but
with the cavity removed to simulate a fiber amplifier. All the fiber parameter values needed for these
simulations were either measured or inferred from fits to independently measured cooling and absorption
data (see Section 2.2). Thanks to this comprehensive characterization, no fitting parameters were needed to
generate the solid curves in Fig. 6. For each pump power, the measured temperatures agree well with
simulations. From the model curves, the average temperature change is -107 mK for 1.12 W of input pump
The blue curve in Fig. 7 shows the average temperature change simulated as a function of input pump
power using the fitted and measured parameter values of this fiber. As expected, the cooling initially increases
with pump power as more of the Yb ions are excited, but eventually plateaus and starts to decrease as the
cooling saturates and the additional power induces more heating via impurity absorption (around 1.1 W).
Figure 6. Measured temperature change versus position along the fiber amplifier, and simulated dependencies
using the model based on [8] for (a)-(c) a 2.74-m and (d) a 4.35-m amplifier fiber.
122
For each pump power, the gain of the 2.74-m amplifier was measured three times and plotted as a
function of input pump power (red crosses in Fig. 7). The solid red curve was generated with the same model
as above. The excellent fit confirms the accuracy of the model and of the measured and inferred parameter
values for the Yb-doped silica fiber. As expected, the gain initially increased linearly with pump power, then
started to saturate as more power was extracted from the gain fiber. For the highest pump power launched
Figure 7. Measured (red crosses) and simulated (red curve) small-signal gain at 1064 nm for a 3-mW seed as a
function of input pump power at 1040 nm into a 2.74-m silica fiber amplifier, and the associated temperature
change along the length of the amplifier as predicted by the model based on the one described in Chapter 4 [8]
(solid blue curve).
To increase the gain, the pump power was increased to 2.62 W and the length of the amplifier fiber was
increased to the calculated optimum value for this pump power (4.35 m). This resulted in 16.9 dB of gain (or
146 mW of output power at 1064 nm) and the average temperature change along the amplifier length
remained negative (see Fig. 6d). The higher pump power resulted in a positive temperature change at the
input end, but once the power was sufficiently attenuated (after ~100 cm) the fiber cooled below room
temperature for the remaining length. The solid curve in Fig. 4d shows the simulated temperature profile
predicted by the amplifier model. It is in excellent agreement with the measured temperature changes. The
average temperature of the amplifier was -24.4 mK. Note that the average temperature could have been
123
reduced, while keeping the gain substantially the same, by splitting the pump power and pumping the fiber
bidirectionally.
4. Summary
This work reports the first radiation-balanced fiber amplifier, in a fiber made not of a fluoride glass, but
of silica, a significantly more ubiquitous, yet challenging material due to its historically low threshold for
concentration quenching. The few-moded fiber used in this work has an aluminosilicate composition tailored
to be doped with 2,500 ppm of Yb3+ while exhibiting negligible concentration quenching. The fiber was core-
pumped at 1040-nm to cool the core while amplifying a 3-mW seed at 1064 nm. With 2.62 W of launched
pump power, the 4.35-m fiber amplifier produced 16.9 dB of small-signal gain while maintaining a negative
average temperature of -24.4 mK below room temperature. A model of a radiation-balanced fiber amplifier
was developed to validate the gain and temperature profile recorded for various pump powers. This work
establishes that silica fibers can be produced with such chemical purity and a low degree of quenching that
they can be highly doped with unprecedented concentrations of ytterbium. This material-science
breakthrough enabled the coherent amplification of light with a gain approaching 20 dB and no net internal
heating. This fundamental development is ushering silica fibers into a new era where it is possible to create
5. References
1. M. K. Davis, and M. J. F Digonnet, "Thermal effects in doped fibers," J. Lightwave Technol. 6, 1013
(1998).
2. N. A. Brilliant, and K. Lagonik, "Thermal effects in a dual-clad ytterbium fiber laser," Opt. Lett. 26,
1699 (2001).
3. C. Jauregui, C. Stihler, and J. Limpert, "Transverse mode instability," Adv. Opt. Photonics 12, 429
(2020).
4. T. R. Gosnell, "Laser cooling of a solid by 65 K starting from room temperature," Opt. Lett. 24, 1041
(1999).
5. P. Barua, E. H. Sekiya, K. Saito, and A. J. Ikushima, "Influences of Yb3+ ion concentration on the
spectroscopic properties of silica glass," J. Non-Cryst. Solids 354, 4760 (2008).
6. K. Oh, and U. Paek, Silica Optical Fiber Technology for Devices and Components: Design, Fabrication,
and International Standards, 1st ed. Hoboken, NJ: Wiley, 2012.
7. J. M. Knall, M. Esmaeelpour, and M. J. F. Digonnet, "Model of Anti-Stokes fluorescence cooling in a
single-mode optical fiber," J. Lightwave Technol. 36, 4752 (2018).
8. J. M. Knall, and M. J. F. Digonnet, "Design of High-Power Radiation-Balanced Silica Fiber Lasers with
a Doped Core and Cladding," J. Lightwave Technol. 39, 2497 (2021).
124
Chapter 8: Demonstration of a Radiation-Balanced Silica Fiber Laser
1. Introduction
In Chapter 7, we demonstrated that it is possible to induce significant cooling and gain in a Yb-doped
silica fiber amplifier core-pumped at a single wavelength. In this chapter, we capitalize on this result to make
a radiation-balanced fiber laser (RBFL). By eliminating the need for conventional coolers, RBFLs have the
potential to exhibit superior output beam quality that surpasses the current standard for ultra-stable lasers in
terms in amplitude noise and frequency noise. The main challenge is the limited amount of heat extraction
offered by ASF cooling. Even in the ideal case in which absorptive loss and quenching are negligible, ASF
cooling only extracts a few percent of the absorbed pump power. Therefore, to create RBFLs with practical
output powers, it is paramount that the small amount of heat extraction is fully utilized to negate the heating
due to the quantum defect. RBFLs cannot afford to waste the achievable cooling on extraneous heating
mechanisms such as absorption by impurities or concentration quenching. With this in mind, the creation of
a practical RBFL is reliant on the development of high-purity fibers that exhibit negligible quenching when
To further maximize the amount of cooling that is used to offset the quantum defect, it is also important
to carefully consider the laser design. In particular, simulations have shown that it is beneficial to have a low-
reflectivity output coupler (see Chapter 4) since this increases the laser threshold (and therefore the gain in
the laser resonator) and decreases the circulating power in the cavity. The first effect increases the amount of
heat extraction, since cooling is proportional to the population inversion, and the second effect decreases the
heating caused by absorption of the circulating power by impurities. In this chapter, we experimentally show
that the right silica composition and laser design can produce RBFLs with practical output powers. In
particular, a maximum output power of 114 mW is demonstrated while maintaining near zero average
temperature change. This result is a pivotal step toward the development of fiber lasers with unprecedented
125
2. Experimental Design
To construct the amplifier, we used the first Yb-doped silica fiber that cooled (Fiber 1, see Section 4 in
Chapter 6). The large 21-µm-diameter core was beneficial for cooling, while the low numerical aperture
(0.070) allowed for few-moded operation (V = 4.5 at the signal wavelength of 1065 nm). The high Yb
concentration (2.06 wt.% Yb) was also beneficial for applications to the RBFL since gain and cooling scale
with the number of active ions [1]. The fiber parameters are summarized in Table 1.
TABLE I
PARAMETERS OF THE YB-DOPED SILICA FIBER USED IN THE RADIATION-BALANCED FIBER LASER
(VALUES WITH AN ASTERISK ARE FITTED; ALL OTHERS ARE MEASURED)
Fiber Parameter Parameter value
Yb concentration 2.06 wt.% Yb
Al concentration 0.86 wt.% Al
F concentration 0.88 wt.% F
Core diameter 21 µm
Numerical aperture 0.070
Radiative lifetime 860 μs
Mean fluorescence wavelength 1006 nm
Quenching lifetime* 31 ms
Critical quenching concentration* 15.1 wt.% Yb
Absorptive loss at 1040 nm* 11 dB/km
The fiber was core-pumped at 1040 nm to create both cooling and gain at 1065 nm (Fig. 1). The output
pigtail of the pump laser was spliced to a 10% fiber splitter. The 90% output was spliced to a fiber Bragg
grating (FBG) with a ~100% reflectivity at 1065 nm photo-inscribed in a Lucent HI-1060 fiber. The output
of this fiber was then spliced to a Corning SMF-28 fiber, which was spliced to the input of a 2.64-m section
of the Yb-doped silica fiber. This length was chosen to ensure that at least 85% of the launched pump power
was absorbed for all the powers utilized in this work. The SMF-28 fiber increased the optical transmission
from the HI-1060 fiber (6-µm-diameter core) to the gain fiber (21-µm diameter), since its intermediate core
size (9-µm diameter) allowed for a more gradual evolution of the propagating mode’s size. This arrangement
also reduced the amount of pump and laser power that was coupled into the cladding of the Yb-doped fiber,
126
thereby minimizing the heat generated by impurity absorption in the cladding and jacket. For the same reason,
a second SMF-28 fiber was spliced between the output of the Yb-doped fiber and the output coupler (8%
reflectively at 1065 nm), an FBG also inscribed in an HI-1060 fiber to ensure a single-mode output. The
reflectivity of the output coupler was chosen to be low to (1) reduce the circulating laser power in the cavity,
since heating due to impurity absorption is proportion to power, and to (2) increase the laser threshold power.
A higher threshold results in a higher population inversion during lasing and, therefore, more heat extracted
by ASF cooling, since it is proportional to the population inversion. At the laser output, a long-pass filter
removed the residual pump power and passed the signal to a power meter. The 10% splitter output was used
to monitor the spectrum of the pump laser and ensure that it did not lase at 1065 nm due the high-reflectivity
FBG (which could have happened given that the gain medium of the 1040-nm pump laser has a very broad
Figure 1: The cooled Yb-doped silica fiber laser and the experimental setup used to measure temperature
changes along the fiber.
To characterize the laser fiber’s temperature change dependence on output power, the laser was pumped
at four different powers. For each power, the induced temperature change was measured at eight locations
along the Yb-doped fiber using our high-resolution slow-light FBG sensor [2]. At each location, a ~10-cm
length of the Yb-doped fiber was stripped of its jacket to prevent reabsorption of the radially escaping ASF.
127
The stripped section was placed in contact with the FBG sensor (see Fig. 1). A small amount of isopropanol
was applied between the fibers to hold them together through capillary forces. When the Yb-doped fiber is
pumped, its temperature changes and the two fibers quickly (within seconds) reach thermal equilibrium [3].
Figure 2: Temporal trace of the temperature change recorded 19 cm from the output end of the Yb-doped silica
fiber, core-pumped at four powers of 1040-nm light, represented pictorially by the red curve.
Figure 2 shows an exemplary temperature measurement of the laser fiber, recorded 19 cm from the output
end of the Yb-doped fiber. Before the measurement started, the pump was turned on to the lowest power of
interest (1.22 W launched into the Yb-doped fiber) and the output of the fiber laser was monitored with a
power meter. The measurement started (t = 0 s) once the output power stabilized (~2 minutes after the pump
was turned on). At this time, the fiber temperature was -104 mK below room temperature. The pump power
was left at 1.22 W for ~25 s, then slowly turned up to the next power of interest (1.36 W). This increase
induced a slight heating of the fiber to a new steady-state value (around -80 mK). This procedure was repeated
for the remaining two pump powers (1.54 W and 1.73 W). After the fiber temperature reached a steady state
for the highest power, the pump was abruptly turned off and the fiber temperature equalized to room
temperature. The temperature change induced by each pump power was defined as the average difference
between the last 5 s when the pump was turned off and the 5 s prior to changing the pump power. Each
128
3. Experimental Results
The fiber laser output power at 1065 nm was measured as a function of launched pump power (blue
crosses in Fig. 3), which was measured upon completion of the experiments using a cut-back method. The
laser output was in the TEM00 mode, since the higher order modes were filtered out by the SMF-28 and
HI-1060 fibers spliced to the output of the gain fiber. This filtering caused only slight fluctuations in the
output power (~5%) due to variations in the amount of power coupled to the higher order modes along the
Yb-doped fiber. As expected, the output power increased linearly once the threshold pump power was
These data points were fitted to the model of a radiation-balanced fiber laser described in Chapter 4 [4].
This fit is plotted as the blue curve in Fig. 3, which is in very good agreement with the experimental data. All
the fiber parameter values needed for these simulations were either measured directly or inferred from fits to
independent cooling and absorption measurements, as described in Section 2.2 of Chapter 7. The FBG
reflectivities were also measured. The only fitting parameters were the losses of the four splices to the
SMF-28 fibers. The total round-trip loss (in dB) of the two splices at the input of the Yb-doped fiber is defined
as ain; the corresponding quantity at the output end is aout. The output loss is further broken down into a
forward propagating loss aout,fw and a backward propagating loss aout,bw, so that aout,fw + aout,bw= aout. aout,fw
directly affects the laser output power as well as the residual pump power and needs to be fitted independently
from aout,bw. Since the splices at both ends of the Yb-doped fiber are nominally identical, ain and aout were
assumed to be equal. Completing the fit resulted in fitted values of ain = aout = 4.8 dB and aout,fw = 2.4 dB.
These relatively high losses are in agreement with expectations. The calculated spatial overlap integral
between the fundamental modes of the HI-1060 fiber and the SMF-28 fiber gives 0.27 dB of single-pass loss.
Between the SMF-28 fiber and the Yb-doped fiber, this calculation gives 1.9 dB of loss. In total, the
calculated round-trip loss for the two splices is 4.4 dB (compared to the fitted value of 4.8 dB). The extra
0.4 dB of loss can be reasonably attributed to (1) coupling to higher order modes in the Yb-doped fiber, which
results in an additional loss when these modes are filtered out at the splice to the SMF-28 fiber (but not in
the other direction), and/or (2) core misalignment at the splices. The threshold pump power given by the fit
129
is 1.07 W, and the slope efficiency is 41%, about half as much as typical commercial silica fiber lasers. The
threshold and slope are in excellent agreement with the experimental data (see Fig. 3). In future iterations,
the slope efficiency and the threshold can be significantly improved by eliminating the splice losses
Figure 3: The laser output power measured as a function of the launched pump power, along with a linear fit
using a model of cooled fiber laser. The color gradient is a pictorial representation of the average temperature
change along the length of the gain fiber.
The temperature change along the 2.64-m fiber laser was measured for four different pump powers
(Fig. 4). This resulted in the same general trend for all four powers. The fiber was the warmest at the input
end and cooled monotonically along its length. For the lowest pump power (1.22 W), this resulted in a
transition to negative temperature changes ~1 m from the input end. Increasing the pump power caused this
transition to occur further along the fiber, ultimately never occurring for the highest pump power (1.73 W).
Three mechanisms contribute to heating at the input end. First, the pump power in this region exceeds the
power that produces maximum cooling, which is calculated to be ~420 mW for this fiber (see Section 3.3 of
Chapter 3 for a physical explanation). Second, the higher pump power corresponds to greater signal
amplification, which results in the generation of more Stokes phonons (more heating). Third, this is also
where the intra-cavity signal power is the greatest, and, therefore, where the depletion of the excited-state
130
population is the highest (less cooling). Further along the fiber, as the pump power is attenuated and the
signal amplification begins to saturate, these heating effects are lessened and ASF cooling dominates. The
dashed segments in Fig. 4 are interpolations of the eight temperature measurements taken for each output
power. Integrating under each curve and dividing by the length of the fiber (2.64 m) gives an approximation
for the average temperature change along the length of the fiber laser. For the 72-mW laser, this average
temperature change is -24.4 mK, indicating that more output power needs to be extracted to achieve radiation-
balanced operation. The latter was very nearly achieved with the 114-mW laser, whose average temperature
was calculated to be only 2.5 mK above room temperature (which is within the measurement error of the
temperature sensor).
Figure 4: Average measured temperature change at eight locations along a 2.64-m silica fiber laser for four
different output powers at 1065 nm.
As expected, the average temperature change continued to increase linearly with output power, resulting
in 58.8 mK for 181 mW and 113.8 mK for 264 mW. The blue crosses in Fig. 5 show these averages plotted
as a function of laser output power, along with a linear fit (solid blue curve). Figure 5 also shows the optical-
to-optical efficiency of the laser, calculated from the data points and fit in Fig. 3 by dividing the laser output
power by the launched pump power (red asterisks and curve). From the curves, the laser output power
131
associated with zero average temperature change is calculated to be 105 mW, and the associated optical-to-
Figure 5: Average temperature change along the fiber laser and the optical-to-optical laser efficiency, both
measured as a function of laser output power, along with their associated fits.
4. Summary
For the first time, a radiation-balanced fiber laser is demonstrated. The gain medium was a highly doped
Yb-silica fiber materially tailored to significantly reduce sources of non-radiative relaxation, including
concentration quenching and the parasitic effects of OH- and Yb2+ species. Pumped at 1040-nm to strike a
near-optimum compromise between cooling and gain, the 1065-nm fiber laser had a threshold power of
1.07 W and a slope of efficiency of 41%, which was limited in part by avoidable splices losses between the
Yb-doped fiber and the FBG fibers. At an output power of 114 mW, the 2.64-m Yb-doped fiber laser had an
average temperature within 3 mK of room temperature. By simply writing the FBGs directly into the Yb-
doped fiber and thereby eliminating the associated splice losses, simulations predict that the output power
for radiation-balanced operation will increase to over 200 mW. As shown in Chapter 4, further power-scaling
to the 100-W level may be possible by pumping a doped cladding. For lower-power lasers, this work
eliminates the need for conventional coolers that degrade the output beam quality. This cornerstone result is
132
5. References
133
Chapter 8: Conclusion and Future Work
1. Summary
Anti-Stokes fluorescence (ASF) cooling has been proposed for numerous applications including cooled
detectors for reduced thermal noise, cooled reference cavities for ultra-stable lasers, and dark current
reduction in space-borne IR and γ-ray sensors. The most important application, however, and the one with
the most far-reaching benefits, is the creation of radiation-balanced lasers and amplifiers, devices in which
the heat generated by the quantum defect is mitigated by cooling due to ASF. This would eliminate the need
for conventional coolers that introduce mechanical vibrations and thermal gradients into the laser system.
Since ASF cooling is compact and vibration-free, lasers cooled by ASF have the potential to exhibit
unprecedented stability in output power and frequency. Prior to this work, ASF cooling had been primarily
explored in bulk samples of crystals and fluoride glass [1-20], and a few fluoride fibers [21,22], doped with
Yb, Tm, Nd, or Er, and placed in a vacuum to reduce the thermal load due to air convection. Often, the
thermal load was further reduced by placing the sample in a small copper enclosure to minimize the radiative
heating [11,23]. Compared with glass fibers, the larger size and superior material properties (in particular in
terms of purity) of crystals enable significantly more heat extraction per sample [23,24]. In 2016, the
combination of these factors ultimately enabled a highly doped Yb:YLF crystal to be cooled down to
cryogenic temperatures [23]. This significant body of work with crystals has also led to two radiation-
balanced lasers [25,26], the first one producing 80 W of output power with no net heating [25], which is a
remarkable accomplishment.
The work reported in this thesis investigates the largely unexplored domain of ASF cooling in optical
fibers. Unlike previous ASF demonstrations, we performed all of our experiments at atmospheric pressure to
represent the environment in which the radiation-balanced devices will be used. Even more importantly, we
focused on cooling in silica since this is the most ubiquitous fiber laser host. At the start of this work, cooling
in silica was thought to be highly improbable due to the high probability of concentration quenching [27],
which injects heat in the host and can preclude the generation of net cooling. This work shows that with the
right composition and fabrication protocol, ASF cooling in silica is not only possible but sufficient to
134
demonstrate the first-ever radiation-balanced fiber laser (RBFL), with over 100 mW of output power [28], as
To accomplish these objectives, we started by developing a model on ASF cooling in a fiber [24]. The
model was meant to be used as a tool to analyze experimental data to understand the magnitude of the various
effects that compete with ASF cooling and to hopefully minimize them. To this end, this model included all
the relevant fiber and pump parameters, including amplified spontaneous emission (ASE), absorptive and
scattering loss, concentration quenching, and the mode profiles of the pump and ASE. The model enabled us
to design experiments and optimize fibers to maximize ASF cooling. With the model, we were able to infer
the optimum pump wavelength and power that maximizes cooling for a particular fiber. Fitting the model to
temperature measurements also allowed us to infer the fiber's absorptive loss and critical quenching
concentration, parameters critical for assessing the fiber's cooling potential. Later, the model was updated to
simulate a fiber amplifier and then, ultimately, a fiber laser [30]. The results from these simulations were
instrumental in determining the optimum cavity length, expected temperature profile, and optimum pump
power for a given RBFL. They also showed that alternative fiber laser designs in which the core and inner
cladding of a double-clad fiber are both doped may enable radiation-balanced operation with up to 115 W of
output power.
Our model of ASF cooling in a fiber also showed that typical temperature changes at atmospheric pressure
are on the mK scale. Since there were no suitable commercial sensors to measure such small temperature
changes, a fellow graduate student designed a custom sensor utilizing a slow-light fiber Bragg grating (FBG)
whose transmission peaks shift with changing temperature [31]. This sensor was instrumental in the success
of this program. It exhibited a record-setting resolution of 0.3 mK/√Hz and a drift of less than 20 mK/min
[31]. For our application, the accuracy of this sensor was contingent on creating a good thermal contact
between the FBG fiber and the test fiber. In addition, this contact had to be accomplished without introducing
material between them that reabsorbs the ASF emitted by the test fiber, since this reabsorption would induce
undesirable heating. Accordingly, we developed a method involving a small amount of isopropanol applied
between the two fibers [32]. The resultant capillary forces wicked the fibers together and, once the
isopropanol evaporated, the fibers were in intimate contact with negligible absorbing substance between
them.
135
To test our experimental set-up and validate our model, we started by testing two Yb-doped ZBLAN
fibers [33,34]. There were several previous reports of ASF cooling in bulk samples of Yb-doped ZBLAN
[1,35,36], but only one demonstration reported cooling in a short section of highly multimoded fiber [21]. In
this work, we tested two fibers at atmospheric pressure: one single-mode (3-µm-diameter core) and one
multimode (200-μm diameter). In the single-mode fiber we measured -5.2 mK of cooling, while in the much
larger multimode fiber we measured temperature changes up to -0.64 K, over two orders of magnitude more
cooling. These results confirmed the sensor's ability to measure temperature changes within a few mK from
room temperature and emphasized the importance of a large doped area to maximize cooling. Fitting the data
to our model of ASF cooling in a fiber [24] allowed us to infer values for the absorptive loss and critical
quenching concentration that were consistent with values reported in literature. This validated the accuracy
of the model, which was critical for future use in designing fibers for ASF cooling. In particular, the model
was instrumental in the development of Yb-doped silica fibers for applications to RBFLs.
The main limitation that precluded cooling of silica, the most important material with regards to fiber
lasers and amplifiers, was concentration quenching. In this work, two main avenues were explored to
overcome this obstacle, namely the introduction of network modifiers and nanoparticles in the host [37-39].
In general, the test fibers were fabricated at Clemson or Mid Sweden University, the spectroscopic properties
were characterized at the University of Illinois at Urbana-Champaign, and the cooling properties were
characterized as part of this research effort at Stanford. We confirmed the efficacy of these techniques to
reduce quenching by testing six fibers with various nanoparticles or network modifiers [37]. All six fibers
demonstrated significantly reduced heating compared to a commercial Yb-doped silica fiber from CorActive.
Ultimately, although nanoparticle fibers did not produce net cooling, the use of adequate network modifiers
was highly successful. A composition and fabrication protocol were established that increased the critical
quenching concentration in a fiber by a factor of 16 over previously reported values for Yb-doped silica
[32,38]. This breakthrough enabled us to demonstrate the first-ever ASF cooled silica fiber! The fiber was
highly doped with 2.06 wt.% Yb and co-doped with 0.86 wt.% Al and 0.88 wt.% F. For 330 mW/m of
absorbed pump power per unit length at 1040 nm, the fiber cooled by -50 mK [38]. After this initial
demonstration, two more Yb-silica fibers were cooled, the best cooling by -70 mK for half the absorbed pump
power at 1040 nm (170 mW/m) [39]. This best performing fiber was doped with 2.0 wt.% Al, which resulted
136
in an even higher critical quenching concentration – 30% higher than in the first cooled fiber. This
improvement enabled the fiber to be doped with an even higher Yb concentration of 2.52 wt.%. Performing
simulations with this fiber showed that it had the potential to exhibit cooling while inducing significant gain,
To construct the amplifier, the Yb-doped silica fiber was core-pumped with 1040 nm light to create ASF
cooling and gain in the core at 1064 nm [29]. We explored several fiber lengths and pump powers, and for
each case, the average temperature was found to be negative! A 4.35-m fiber pumped with 2.62 W and seeded
with 3 mW produced ~17 dB of gain (146 mW of output power) while the average fiber temperature
remained slightly below room temperature. All gain and temperature measurements were in good agreement
with the model. This cornerstone result validated the model's ability to make accurate predictions for future
Success with the RBAs established that the seemingly small amount of ASF cooling generated in a Yb-
doped silica fiber is sufficient to induce significant gain. With that in mind, we used the first silica fiber that
cooled to create a RBFL [28]. The laser cavity was defined by a high reflectively FBG spliced to the input of
the Yb-doped fiber and a low reflectively FBG spliced to the output. The 2.74-m gain fiber was core-pumped
at 1040 nm and lased at 1065 nm. Ultimately, we demonstrated a 105-mW laser with no net heating.
Alternative fiber designs were also explored in an effort to increase the achievable ASF cooling in single
to few-moded fibers [40]. Specifically, we explored cooling in double-clad Yb-doped ZBLAN fibers in which
the core and the inner cladding are both doped. We showed that cooling can be significantly increased with
this design, but that the improvement is contingent on adequate mode-mixing of the pump power within the
doped inner cladding. We showed that an octagonal cladding profile exhibits a 6-fold improvement in
effective mode area when compared with a traditional circular cladding. This design produced 28 times more
2. Future Work
This work capitalized on a sound principle and a largely unexplored field. This opportunity enabled us to
establish a solid foundation with many breakthroughs and experimental firsts. The work presented in this
thesis paves the way toward a new class of lasers and amplifiers with unprecedented stability in power and
137
frequency. Yet, there is still much to be done to practically realize this goal. Some of these advancements are
straightforward improvements to the existing set-up, but other aspects may require years of materials
research.
In the near-term, the output power of the RBA could be improved by increasing the pump or seed power.
For every RBA demonstration presented in this work, the average temperature change along the length of the
gain fiber was below room temperature [29]. This means that there was an excess of heat extracted by ASF:
the fiber was over-cooled. Heating due the quantum defect (i.e. amplification) can therefore be increased to
bring the average temperature change to zero. Simulations from our model of RBAs predict that injecting a
higher power seed will achieve zero net heating: for a 15-mW seed (instead of 3 mW), radiation-balanced
A simple improvement to the RBFL would be to eliminate the splice losses at the input and output of the
gain fiber [28]. This can be achieved by writing the cavity FBGs directly into the gain fiber or into a undoped
fiber that is mode-matched to the gain fiber. Eliminating these losses will lower the laser threshold power
Another straightforward improvement may be to strategically coil the gain fiber. In this work, we
demonstrated devices with zero average temperature, but there was still a temperature gradient along the
fiber length. Coiling the fiber so that the warmer sections are thermally coupled to the cooler sections may
Implementing these three improvements will result in valuable progress, but the achievable output powers
will still be limited to a few hundred milliwatts. Significant power-scaling is dependent on alternative fiber
designs and further improvements to the fiber's material properties. In this work, we presented a preliminary
investigation of one alternative fiber design in which the core and inner cladding of a double-clad fiber are
doped with an active ion [30,40]. This produced promising results, but further investigation is necessary to
maximize the pump's effective mode area in order to fully capitalize on the increased cooling offered by this
design. Further work is also needed to test this fiber as a laser or amplifier. Also, this work demonstrates the
feasibility of this design with a ZBLAN host [40], but future work would benefit from duplicating this design
in silica fibers.
138
Ultimately, power-scaling radiation-balanced devices to hundreds of watts will require material
advancements that significantly reduce the competing heating mechanisms within the fiber. To provide a
physical picture of the efficiency of the ASF cooling process in general, and to predict how much this
efficiency may be improved in the future, it is informative to carry out an approximate calculation of the
energy budget. How much optical power must be provided to keep the fiber radiation balanced is dictated by
the (small) percentage of energy extracted by ASF each time a Yb ion fluoresces (the cooling mechanism),
and by the two main residual heating mechanisms, namely quenching and absorption by impurities. In the
following calculation, we consider the case of the 2.74-m fiber amplifier from Chapter 6 pumped with a
The portion of pump power absorbed by the Yb-doped fiber in the presence of the seed, 1.39 W, decreases
to 1.38 W without the seed (lower ground-state population, less absorption). Essentially all this power
escapes from the fiber in the form of anti-Stokes fluorescence. For every fluorescence photon that escapes,
the net energy extracted from the fiber (ignoring quenching for now) is the difference between the pump
photon energy (a wavelength of 1040 nm) and the mean fluorescence photon energy (a wavelength of
1003.9 nm), or 6.87x10-21 J. Scaled to the total absorbed pump power (1.38 W, or 7.22x1018 photons per
second), this is P ≈ 49.6 mW of heat extracted from the fiber. The quenching lifetime inferred from fitting
our model to temperature measurements is 38 ms, compared to the Yb radiative lifetime of 765 µs. In other
words, in the optically pumped fiber, 98% of the ions in the excited state relax to the ground state radiatively
and contribute to ASF cooling; the rest relax non-radiatively and do not contribute. The value of 49.6 mW
therefore needs to be reduced by 2%, giving a total extracted heat of 48.6 mW. The remaining 2% of the Yb
population relax via quenching. Although this percentage is small, every time an Yb ion is quenched it
releases its full excitation (the energy of the pump photon that it absorbed to be excited) to the host in the
form of heat. This released heat is therefore 2% of the absorbed pump power, or 27.6 mW. The absorption
due to impurities is 18 dB/km, or 0.05-dB for the 2.74-m fiber. On average, this mechanism releases 1.15%
of the absorbed pump power, or 15.9 mW. The combination of the two residual heat-generating mechanisms,
impurity absorption and quenching, therefore inject 43.5 mW of extraneous heat into the host. The net heat
139
ASF cooling required 1.38 W of absorbed pump power to extract 5.1 mW of heat, which corresponds to
a cooling efficiency of only 0.37%. This metric is not entirely fair, since this absorbed pump power was
mostly used to induce the gain expected from the amplifier. Nevertheless, this back-of-the-envelope
calculation provides a number of useful conclusions. First, as is well known, the ASF cooling process, even
with a laser ion as well suited to this process as Yb3+, is inefficient, although it is clearly efficient enough to
be of practical use. Second, the two spurious heating mechanisms that counteract ASF cooling, quenching
and impurity absorption, as significantly reduced as they are in the fiber used in this work compared to
conventional Yb-doped silica fibers, still steal most of the energy extracted by ASF: 43.5 mW out of the
49.6 mW of expected ASF power is lost. Nearly 90% of the ASF cooling power is wasted! Third, quenching
is the worst of these two mechanisms, as it accounts for two thirds of this undesirable heat. These
considerations clearly show that the cooling efficiency would be significantly greater if quenching, and to a
lesser extent the absorptive loss, were further reduced. For example, if the impurity absorption was halved
and the quenching lifetime doubled, for the same pump power the cooling efficiency would increase five-
fold, to 2%, and the extracted heat would increase from 5.1 mW to 28 mW. Simulations show that increasing
the fiber length to 6 m, the amplifier would produce 2 W of output power (a gain of 28.2 dB) for 5.7 W of
pump power (versus the current 146 mW for 1.64 W of pump) while remaining radiation balanced. It follows
that a salient conclusion of this thesis work is that to improve both the efficiency of the cooling process and
to power-scale the devices reported in this work to hundreds of watts, it is paramount to continue material
research with the single most important goal of reducing the residual quenching.
3. References
140
8. E. S. de Lima Filho, G. Nemova, S. Loranger and R. Kashyap, "Laser-induced cooling of a Yb:YAG
crystal in air at atmospheric pressure," Opt. Express 21, 24711 (2013).
9. S. Bigotta, "Laser cooling of solids: new results with single fluoride crystals," Nuovo Cimento B Ser.
122, 685694 (2007).
10. S. Bigotta, A. Di Lieto, D. Parisi, A. Toncelli, and M. Tonelli, "Single fuoride crystals as materials for
laser cooling applications," SPIE 6461, 64610E (2007).
11. D. V. Seletskiy, S. D. Melgaard, S. Bigotta, A. Di Lieto, M. Tonelli, and M. Sheik-Bahae, "Laser cooling
of solids to cryogenic temperatures," Nat. Photon. 4, 161 (2010).
12. R. I. Epstein, J. J. Brown, B. C. Edwards, and A. Gibbs, "Measurements of optical refrigeration in
ytterbium-doped crystals," J. Appl. Phys. 90, 4815 (2001).
13. S. Bigotta, D. Parisi, L. Bonelli, A. Toncelli, A. D. Lieto, and M. Tonelli, "Laser cooling of Yb3+-doped
BaY2F8 single crystal," Opt. Mater. 28, 1321 (2006).
14. J. Fernandez, A. J. Garcia-Adeva, and R. Balda, "Anti-Stokes laser cooling in bulk erbium-doped
materials, Phys. Rev. Lett. 97, 033001 (2006).
15. A. Garcia-Adeva, R. Balda, and J. Fernandez, "Anti-Stokes laser cooling in erbium-doped low phonon
materials, Proc. SPIE 6461, 646102 (2007).
16. C. W. Hoyt, M. Sheik-Bahae, R. I. Epstein, B. C. Edwards, and J. E. Anderson, "Observation of anti-
Stokes fluorescence cooling in thulium-doped glass," Phys. Rev. Lett. 85, 3600 (2000).
17. C. W. Hoyt, M. P. Hasselbeck, M. Sheik-Bahae, R. I. Epstein, S. Greeneld, J. Thiede, J. Distel, and J.
Valencia, "Advances in laser cooling of thulium-doped glass," J. Opt. Soc. Am. B 20, 1066 (2003).
18. C. W. Hoyt, W. Patterson, M. P. Hasselbeck, M. Sheik-Bahae, R. I. Epstein, J. Thiede, and D. V.
Seletskiy, "Laser cooling thulium-doped glass by 24 K from room temperature," Trends Opt. Phot. 89,
QThL4 (2003).
19. W. Patterson, S. Bigotta, M. Sheik-Bahae, D. Parisi, M. Tonelli, and R. I. Epstein, "Anti-Stokes
luminescence cooling of Tm3+ doped BaY2F8," Opt. Express 16, 1704 (2008).
20. S. Rostami, A. R. Albercht, M. R. Ghasemkhani, S. D. Melgaard, A. Gragossian, M. Tonelli, and M.
Sheik-Bahae, "Optical refrigeration of Tm:YLF and Ho:YLF crystals," Proc. SPIE 9765, 97650P (2016).
21. T. R. Gosnell, ''Laser cooling of a solid by 65 K starting from room temperature,'' Opt. Lett. 24, 1041
(1999).
22. T. R. Gosnell and X. Luo, "Laser cooling of a solid by 21 K starting from room temperature," Technical
Digest—European Quantum Electronics Conference (1998).
23. S. D. Melgaard, A. R. Albrecht, M. P. Hehlen, and M. Sheik-Bahae ''Solid-state optical refrigeration to
sub-100 Kelvin regime,'' Sci. Rep. 6, 1 (2016).
24. J. M. Knall, M. Esmaeelpour, and M. J. F. Digonnet, ''Model of anti-Stokes fluorescence cooling in a
single-mode optical fiber,'' J. Lightwave Technol. 36, 4752 (2018).
25. S. Bowman, S. P. O’Connor, S. Biswal, N. J. Condon, and A. Rosenberg, ''Minimizing heat generation
in solid-state lasers,'' IEEE J. Quantum Electron. 46, 1076 (2010).
26. Z. Yang, J. Meng, A. Albrecht, and M. Sheik-Bahae, "Radiation-balanced Yb:YAG disk laser," Opt.
Express 27, 1392 (2019).
27. P. Barua, E. H. Sekiya, K. Saito, and A. J. Ikushima, "Influences of Yb3+ ion concentration on the
spectroscopic properties of silica glass," J. Non-Cryst. Solids 354, 4760 (2008).
28. J. M. Knall, M. Engholm, T. Boilard, M. Bernier, P.-B. Vigneron, N. Yu, P. D. Dragic, J. Ballato, and
M. J. F. Digonnet, "A radiation-balanced silica fiber laser," Optica, doi: 10.1364/OPTICA.425115
(2021).
29. J. M. Knall, M. Engholm, T. Boilard, M. Bernier, and M. J. F. Digonnet, "A radiation-balanced silica
fiber amplifier," arXiv:2103.02698 (2021).
30. J. M. Knall, and M. J. F. Digonnet, "Design of high-power radiation-balanced silica fiber lasers with a
doped core and cladding," J. Lightwave Technol. 39, 2497 (2021).
31. A. Arora, M. Esmaeelpour, M. Bernier, and M. J. F. Digonnet, "High-resolution slow-light fiber Bragg
grating temperature sensor with phase-sensitive detection," Opt. Lett. 14, 3337 (2018).
32. J. M. Knall, P.-B. Vigneron, M. Engholm, P. D. Dragic, N. Yu, J. Ballato, M. Bernier, and M. J. F.
Digonnet, ''Experimental observation of cooling in Yb-doped silica fibers,'' Proc. SPIE 11298, 112980F-
1 (2020).
33. J. M. Knall, A. Arora, M. Bernier, S. Cozic, M. J. F. and Digonnet, "Demonstration of anti-Stokes
cooling in Yb-doped ZBLAN fibers at atmospheric pressure," Opt. Lett. 44, 2338 (2019).
141
34. J. M. Knall, A. Arora, M. Bernier, and M. J. F. Digonnet, ''Anti-Stokes fluorescence cooling in Yb-doped
ZBLAN fibers at atmospheric pressure: experiments and near-future prospects,'' Proc. SPIE 10936,
109360F-1 (2019).
35. B. C. Edwards, J. E. Anderson, R. I. Epstein, G. L. Mills, and A. J. Mord, "Demonstration of a solid-
state optical cooler: an approach to cryogenic refrigeration," J. Appl. Phys. 86, 6489 (1999).
36. A. Rayner, M. E. J. Friese, A. G. Truscott, N. R. Heckenberg, and H. Rubinsztein-Dunlop, "Laser cooling
of a solid from ambient temperature," J. Mod. Opt. 48, 103 (2001).
37. J. M. Knall, A. Arora, P. D. Dragic, J. Ballato, M. Cavillon, T. Hawkins, S. Jiang, T. Luo, M. Bernier,
and M. J. F. Digonnet, "Experimental investigations of spectroscopy and anti-Stokes fluorescence
cooling in Yb-doped silicate fibers," Proc. SPIE 10936, 109360G-1 (2019).
38. J. M. Knall, P.-B. Vigneron, M. Engholm, P. D. Dragic, N. Yu, J. Ballato, M. Bernier, and M. J. F.
Digonnet, "Laser cooling in a silica optical fiber at atmospheric pressure," Opt. Lett. 45, 1092 (2020).
39. J. M. Knall, M. Engholm, J. Ballato, P. D. Dragic, N. Yu, and M. J. F. Digonnet, "Experimental
comparison of silica fibers for laser cooling," Opt. Lett. 45, 4020 (2020).
40. P.-B. Vigneron, J. M. Knall, T. Boilard, M. Bernier, and M. J. F Digonnet, "Observation of anti-Stokes-
fluorescence cooling in a ZBLAN fiber with a Yb-doped cladding," Proc. SPIE 11702, 117020A-1
(2021).
41. J. M. Knall and M. J. F. Digonnet, "Anti-Stokes-fluorescence-cooled fiber-based gain element," Sept.
29, 2020. Pat. No. 10790633.
142