IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 2, NO. 3, SEPTEMBER 1996 435
Semiconductor Saturable Absorber Mirrors
(SESAM’s) for Femtosecond to Nanosecond
Pulse Generation in Solid-State Lasers
Ursula Keller, Member, IEEE, Kurt J. Weingarten, Member, IEEE, Franz X. K¨artner, Daniel Kopf, Bernd Braun,
Isabella D. Jung, Regula Fluck, Clemens H
¨
onninger, Nicolai Matuschek, and Juerg Aus der Au
(Invited Paper)
Abstract— Intracavity semiconductor saturable absorber mir-
rors (SESAM’s) offer unique and exciting possibilities for pas-
sively pulsed solid-state laser systems, extending from
-switched
pulses in the nanosecond and picosecond regime to mode-locked
pulses from 10’s of picoseconds to sub-10 fs. This paper reviews
the design requirements of SESAM’s for stable pulse generation
in both the mode-locked and
-switched regime. The combina-
tion of device structure and material parameters for SESAM’s
provide sufficient design freedom to choose key parameters such
as recovery time, saturation intensity, and saturation fluence, in
a compact structure with low insertion loss. We have been able
to demonstrate, for example, passive modelocking (with no
-
switching) using an intracavity saturable absorber in solid-state
lasers with long upper state lifetimes (e.g., 1-
m neodymium
transitions), Kerr lens modelocking assisted with pulsewidths as
short as 6.5 fs from a Ti:sapphire laser—the shortest pulses
ever produced directly out of a laser without any external pulse
compression, and passive
-switching with pulses as short as
56 ps—the shortest pulses ever produced directly from a
-
switched solid-state laser. Diode-pumping of such lasers is leading
to practical, real-world ultrafast sources, and we will review
results on diode-pumped Cr:LiSAF, Nd:glass, Yb:YAG, Nd:YAG,
Nd:YLF, Nd:LSB, and Nd:YVO
.
I. HISTORICAL BACKGROUND AND INTRODUCTION
A. Semiconductor Saturable Absorbers for Solid-State Lasers
T
HE use of saturable absorbers in solid-state lasers is
practically as old as the solid-state laser itself [1]–[3].
However, it was believed that pure, continuous-wave (CW)
modelocking could not be achieved using saturable absorbers
with solid-state lasers such as Nd:glass, Nd:YAG, or Nd:YLF
with long upper state lifetimes (i.e.,
100 s) without -
switching or
-switched mode-locked behavior (Fig. 1). This
limitation was mostly due to the parameter ranges of available
saturable absorbers [4]. However, the advent of bandgap
engineering and modern semiconductor growth technology
has allowed for saturable absorbers with accurate control
of the device parameters such as absorption wavelength,
saturation energy, and recovery time, and we have been able to
demonstrate pure passive
-switching, pure CW modelocking
Manuscript received September 24, 1996; revised January 9, 1997.
The authors are with the Institute of Quantum Electronics, Swiss Federal
Institute of Technology (ETH), ETH-H
¨
onggerberg HPT, CH-8093 Z
¨
urich,
Switzerland.
Publisher Item Identifier S 1077-260X(96)09675-X.
Fig. 1. Different modes of operation of a laser with a saturable absorber.
CW
-switching typically occurs with much longer pulses and lower pulse
repetition rates than CW mode-locking.
or, if desired, -switched modelocking behavior [5]–[9]. In
addition, semiconductor absorbers have an intrinsic bitemporal
impulse response (Fig. 2): intraband carrier–carrier scattering
and thermalization processes which are in the order of 10
to 100 fs as well as interband trapping and recombina-
tion processes which can be in the order of picoseconds to
nanoseconds depending on the growth parameters [10], [11].
As we will discuss, the faster saturable absorption plays an
important role in stabilizing femtosecond lasers, while the
slower response is important for starting the pulse formation
process and for pulse forming in lasers with pulsewidths of
picoseconds or longer.
Many other classes of laser can be passively mode-locked
with saturable absorbers. Previously, semiconductor saturable
absorbers have been successfully used to mode-locked semi-
conductor diode lasers, where the recovery time was reduced
by damage induced either during the aging process [12], by
proton bombardment [13], or by multiple quantum wells [14].
More recently, both bulk and multiple quantum-well semi-
conductor saturable absorbers have been used to mode-lock
color center lasers [15]. In both cases, the upper state lifetime
of the laser medium is in the nanosecond regime, which
strongly reduces the tendency for self-
-switching instabilities
(discussed further in Section II). This is not the case for
1077–260X/96$5.00 1996 IEEE
436 IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 2, NO. 3, SEPTEMBER 1996
Fig. 2. A measured impulse response typical for a semiconductor saturable
absorber. The optical nonlinearity is based on absorption bleaching.
most other solid-state lasers with an upper state laser lifetime
in the microsecond to millisecond regime. First results with
SESAM’s in solid-state lasers were reported in 1990, and they
were initially used in nonlinear coupled cavities [16]–[21],
a technique termed RPM (resonant passive mode-locking).
This paper was motivated by previously demonstrated soliton
lasers [22] and APM (additive pulse mode-locking) lasers
[23]–[25], where a nonlinear phase shift in a fiber inside
a coupled cavity provided an effective saturable absorption.
Most uses of coupled cavity techniques have been supplanted
by intracavity saturable absorber techniques based on Kerr
lens mode-locking (KLM) [26] and SESAM’s [5], due to
their more inherent simplicity. In 1992, we demonstrated a
stable, purely CW-mode-locked Nd:YLF and Nd:YAG laser
using an intracavity SESAM design, referred to as the an-
tiresonant Fabry–Perot saturable absorber (A-FPSA) [5]. Since
then, many new SESAM designs have been developed (see
Section III) that provide stable pulse generation for a variety
of solid-state lasers.
B. Mode-Locking Mechanism for Solid-State Lasers: Fast-
Saturable-Absorber Mode-Locking or Soliton Mode-Locking
Passive mode-locking mechanisms are well-explained by
three fundamental models: slow saturable absorber mode-
locking with dynamic gain saturation [27], [28] [Fig. 3(a)],
fast saturable absorber mode-locking [29], [30] [Fig. 3(b)]
and soliton mode-locking [31]–[33] [Fig. 3(c)]. In the first
two cases, a short net-gain window forms and stabilizes an
ultrashort pulse. This net-gain window also forms the minimal
stability requirement, i.e., the net loss immediately before and
after the pulse defines its extent. However, in soliton mode-
locking, where the pulse formation is dominated by the balance
of group velocity dispersion (GVD) and self-phase modulation
(SPM), we have shown that the net-gain window can remain
open for more than ten times longer than the ultrashort pulse,
depending on the specific laser parameters [32]. In this case,
the slower saturable absorber only stabilizes the soliton and
starts the pulse formation process.
Until the end of the 1980’s, ultrashort pulse generation was
dominated by dye lasers, where mode-locking was based on a
balanced saturation of both gain and loss, opening a steady-
Fig. 3. The three fundamental passive mode-locking models: (a) passive
mode-locking with a slow saturable absorber and dynamic gain saturation [27],
[28], (b) fast absorber mode-locking [29], [30], and (c) soliton mode-locking
[31]–[33].
state net gain window as short as the pulse duration [Fig. 3(a)]
(the slow-absorber with dynamic gain saturation model [27],
[28]). Pulses as short as 27 fs with an average power of
10 mW were generated [34]. Shorter pulse durations to 6
fs were achieved through additional amplification and fiber-
grating pulse compression, although at much lower repetition
rates [35].
The situation changed with the development and commer-
cialization of the Ti:sapphire laser [36], which has a gain-
bandwidth large enough to support ultrashort pulse generation.
However, existing mode-locking techniques were inadequate
because of the much longer upper state lifetime and the
smaller gain cross section of this laser, which results in
negligible pulse-to-pulse dynamic gain saturation. Initially it
was assumed that a fast saturable absorber would be required
to generate ultrashort pulses [Fig. 3(b)]. Such a fast saturable
absorber was discovered [26] and its physical mechanism
described as Kerr lens mode-locking (KLM) [19], [37], [38],
where strong self-focusing of the laser beam combined with ei-
ther a hard aperture or a “soft” gain aperture is used to produce
a self amplitude modulation, i.e., an equivalent fast saturable
absorber. Since then, significant efforts have been directed
toward optimizing KLM for shorter pulse generation, with the
current results standing at around 8 fs [39]–[41] directly from
the laser. Using a broad-band intracavity SESAM device in
addition to KLM and higher order dispersion compensation
[42], [43] we recently generated pulses as short as 6.5 fs
[Fig. 12(b)] directly out of a Ti:sapphire laser with 200 mW
average output power at a pulse repetition rate of
85 MHz
[44]. External pulse compression techniques based on fiber-
grating pulse compressors have been used to further reduce
the pulse duration from a Ti:sapphire laser to
5 fs at a center
wavelength of
800 nm [45], [46]. These are currently the
shortest optical pulses ever generated.
Besides the tremendous success of KLM, there are some
significant limitations for practical or “real-world” ultrafast
lasers. First, the cavity is typically operated near one end
of its stability range, where the Kerr-lens-induced change of
the beam diameter is large enough to sustain mode-locking.
This results in a requirement for critical cavity alignment
where mirrors and laser crystal have to be positioned to an
accuracy of several hundred microns typically. Additionally,
the self-focusing required for KLM imposes limitations on
the cavity design and leads to strong space-time coupling of
the pulses in the laser crystal that results in complex laser
KELLER et al.: SEMICONDUCTOR SATURABLE ABSORBER MIRRORS 437
dynamics [47], [48]. Once the cavity is correctly aligned, KLM
can be very stable and under certain conditions even self-
starting [49], [50]. However, self-starting KLM lasers in the
sub-50-fs regime have not yet been demonstrated without any
additional starting mechanisms as for example a SESAM. This
is not surprising, since in a 10-fs Ti:sapphire laser with a 100
MHz repetition rate, the peak power changes by six orders
of magnitude when the laser switches from CW to pulsed
operation. Therefore, nonlinear effects that are still effective
in the sub-10-fs regime are typically too small to initiate mode-
locking in the CW-operation regime. In contrast, if self-starting
is optimized, KLM tends to saturate in the ultrashort pulse
regime or the large SPM will drive the laser unstable.
However, we have shown that a novel mode-locking tech-
nique, which we term soliton mode-locking [31]–[33], [51],
addresses many of these issues. In soliton mode-locking, the
pulse shaping is done solely by soliton formation, i.e., the
balance of GVD and SPM at steady state, with no additional
requirements on the cavity stability regime. An additional loss
mechanism, such as a saturable absorber [31], [33], or an
acousto-optic mode-locker [51], [52], is necessary to start the
mode-locking process and to stabilize the soliton.
This can be explained as follows. The soliton loses energy
due to gain dispersion and losses in the cavity. Gain dispersion
and losses can be treated as perturbation to the nonlinear
Schr
¨
odinger equation for which a soliton is a stable solution
[51]. This lost energy, called continuum in soliton perturbation
theory [53], is initially contained in a low intensity background
pulse, which experiences negligible bandwidth broadening
from SPM, but spreads in time due to GVD. This continuum
experiences a higher gain compared to the soliton, because it
only sees the gain at line center (while the soliton sees an ef-
fectively lower average gain due to its larger bandwidth). After
a sufficient build-up time, the continuum would actually grow
until it reaches an effective lasing threshold, destabilizing the
soliton. However, we can stabilize the soliton by introducing a
“slow” saturable absorber into the cavity. This slow absorber
adds sufficient additional loss so that the continuum no longer
reaches threshold, but with negligible increased loss for the
short soliton pulse.
Depending on the specific laser parameters such as gain
dispersion, small signal gain, and negative dispersion, a “slow”
saturable absorber can stabilize a soliton with a response
time of more than ten times longer than the steady-state
soliton pulsewidth [Fig. 3(c)]. High-dynamic range autocor-
relation measurements have shown ideal transform-limited
soliton pulses over more than six orders of magnitude, even
though the net gain window is open much longer than the pulse
duration [32], [54], [55]. Due to the slow saturable absorber,
the soliton undergoes an efficient pulse cleaning mechanism
[33]. In each round-trip, the front part of the soliton is absorbed
which delays the soliton with respect to the continuum.
In contrast to KLM, soliton mode-locking is obtained over
the full cavity stability regime, and pulses as short as 13 fs
have been generated currently with a purely soliton-mode-
locked Ti:sapphire laser using a broad-band SESAM [33],
[56]. Soliton mode-locking decouples SPM and self-amplitude
modulation, potentially allowing for independent optimization.
We justify the introduction of a new name for this mode-
locking process because previously soliton effects were only
considered to lead to a moderate additional pulsewidth re-
duction of up to a factor of 2, but the stabilization was still
achieved by a short net gain window as discussed for CPM
dye [57]–[60] and for KLM Ti:sapphire lasers [61], [62].
II. D
ESIGN CRITERIA FOR A SATURABLE ABSORBER
First we consider the basic design parameters of a general
saturable absorber. These consist of the saturation intensity
and saturation fluence , which will be seen to influ-
ence the mode-locking build-up and the pulse stability with
respect to self-
-switching. In addition, the recovery time of
the saturable absorber determines the dominant mode-locking
mechanism, which is either based on fast saturable absorber
mode-locking [Fig. 3(b)] in the positive or negative dispersion
regime, or soliton mode-locking [Fig. 3(c)], which operates
solely in the negative dispersion regime. For solid-state lasers
we can neglect slow saturable absorber mode-locking as shown
in Fig. 3(a), because no significant dynamic gain saturation is
taking place due to the long upper state lifetime of the laser.
When the recovery time of the absorber is on the order of or
even larger than the laser’s cavity round-trip time, the laser will
tend to operate in the pure CW-
-switching regime (Fig. 1).
In addition, the nonsaturable losses of a saturable absorber
need to be small, because we typically only couple a few
percent out of a CW mode-locked solid-state laser. As the
nonsaturable losses increase, the laser becomes less efficient
and operates fewer times over threshold, which increases the
tendency for instabilities [see (4) and (6) below] such as
-switched mode-locked behavior.
Fig. 4 shows the typical saturation behavior for an ab-
sorber on a mirror. Initially, the pulses are formed by noise
fluctuations in the laser, and the saturation amount at this
early stage is dominated by the CW intensity
incident on
the absorber [Fig. 4(a)]. In general, we can assume that the
saturable absorber is barely bleached (i.e.,
) at CW
intensity, because if the absorber were fully bleached at this
intensity, there would be insufficient further modulation to
drive the pulse forming process.
The saturation intensity
is given by
(1)
where is the photon energy, the absorption cross section
and
the absorber recovery time. It is important to note that
the absorption cross section is effectively a material parameter.
The absorption coefficient
of the material is then given by
(2)
where
is the density of absorber atoms or the density of
states in semiconductors, for example.
Referring again to Fig. 4(a), the slope
at around
determines the mode-locking build-up time under
certain approximations [9] can be written as
(3)
438 IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 2, NO. 3, SEPTEMBER 1996
(a)
(b)
Fig. 4. Nonlinear reflectivity change of a saturable absorber mirror due to
absorption bleaching with the (a) CW intensity and (b) short pulses.
is
the saturation intensity,
is the saturation fluence, is the CW intensity,
and
is the pulse energy density incident on the saturable absorber.
As expected, the build-up time is inversely proportional to
this slope. This follows directly from Fig. 4(a), which shows
that small intensity fluctuations will introduce a larger reflec-
tivity change of the saturable absorber if the slope is larger.
Therefore, the mode-locking build-up time decreases with
smaller saturation intensities. However, there is a tradeoff: if
the saturation intensity is too small, the laser will start to
-
switch. The condition for no
-switching is derived in [4],
[9]:
no
-switching: (4)
where
is the pump parameter that determines how many
times the laser is pumped above threshold,
is the cavity
round trip time, and
is the upper state lifetime of the laser.
The stimulated lifetime
of the upper laser level is given
by
for . The small signal
gain of the laser is given by
, where is the total loss
coefficient of the laser cavity. From (4), it then follows that
-switching can be more easily suppressed for a small slope
(i.e., a large saturation intensity), a large (i.e., a laser
that is pumped far above threshold with a large small-signal
gain
or small losses ), a large cavity round-trip period
(i.e., for example a low mode-locked pulse repetition rate).
Equation (4) also indicates that solid-state lasers with a large
upper state lifetime
will have an increased tendency for
self-
-switching instabilities.
The physical interpretation of the
-switching threshold
(4) is as follows: The left-hand side of (4) determines the
reduction in losses per cavity round-trip due to the bleaching
in the saturable absorber. This loss reduction will increase
the intensity inside the laser cavity. The right-hand side of
(4) determines how much the gain per round-trip saturates,
compensating for the reduced losses and keeping the intensity
inside the laser cavity constant. If the gain cannot respond
fast enough, the intensity continues to increase as the absorber
is bleached, leading to self-
-switching instabilities or stable
-switching.
Equations (3) and (4) give an upper and lower bound for the
saturation intensity which results in stable CW mode-locking
without self-
-switching. Of course, we can also optimize
a saturable absorber for
-switching by selecting a small
saturation intensity and a short cavity length, i.e., a short
.
This will be discussed in more detail in Section V.
If we use a fast saturable absorber with recovery time much
shorter than the cavity round-trip time (
), then the
conditions given by (3) and (4) are typically fulfilled and much
shorter pulses can be formed. But now, an additional stability
requirement has to be fulfilled to prevent
-switched mode-
locking (Fig. 1). For this further discussion, we assume that the
steady-state pulse duration
is shorter than the recovery time
of the saturable absorber, i.e., . In this case the
saturation [Fig. 4(b)] is determined by the saturation fluence
, given by
(5)
and the incident pulse energy density
on the saturable
absorber. The loss reduction per round-trip is now due to
bleaching of the saturable absorber by the short pulses, not the
CW intensity. This is a much larger effect when
.
Therefore, in analogy to (4), we can show that the condition
to prevent
-switched mode-locking is given by [9]:
no
-switched mode-locking:
(6)
We can easily fulfill this condition by choosing
[Fig. 4(b)]. This also optimizes the modulation depth, resulting
in reduced pulse duration.
However, there is also an upper limit to
, determined
by the onset of multiple pulsing [63]. Given an energy fluence
many times the saturation energy fluence
, we can see that
the reflectivity is strongly saturated and no longer a strong
function of the pulse energy. In addition, shorter pulses see
a reduced average gain, due to the limited gain bandwidth
of the laser. Beyond a certain pulse energy, two pulses with
lower power, longer duration, and narrower spectrum will
be preferred, since they see a larger increase of the average
gain but a smaller increase in the absorption. The threshold
for multiple pulsing is lower for shorter pulses, i.e., with
spectrums broad compared to the gain bandwidth of the laser.
Our experimentally determined rule of thumb for the pulse
energy density on the saturable absorber is three to five times
the saturation fluence. A more detailed description of multiple
pulsing will be given elsewhere. In general, the incident pulse
energy density on the saturable absorber can be adjusted by
the incident mode area, i.e., how strongly the cavity mode is
focused onto the saturable absorber.
Equations (3), (4), and (6) give general criteria for the
saturation intensity
(1) and saturation fluence (5) of
the saturable absorber. Normally, the saturation fluence of the
absorber material is a given, fixed parameter, and we have to
KELLER et al.: SEMICONDUCTOR SATURABLE ABSORBER MIRRORS 439
(a) (b)
Fig. 5. Measured absorption bleaching and electron trapping times (i.e., recovery time of saturable absorber) for low-temperature MBE grown InGaAs –GaAs
multiple quantum-well absorbers. The MBE growth temperature is the variable parameter used in the nonlinear reflectivity.
adjust the incident mode area to set the incident pulse energy
density onto the saturable absorber to fulfill the conditions
given by (6) and the multiple pulsing instabilities. Therefore,
the only parameter left to adjust for the saturation intensity is
the absorber recovery time
(1). However, if we want to use
the absorber as a fast saturable absorber, we have to reduce
.
Semiconductor materials are interesting in this regard, because
we can adjust
from the nanosecond to the subpicosecond
regime using different growth parameters (Section III-A). In
this case, however, it is often necessary to find another
parameter with which to adjust
rather than with .
We will show in the next section that this can be obtained
by using semiconductor saturable absorbers inside a device
structure which allows us to modify the effective absorber
cross section
(1), which is a fixed material parameter.
For cases where the cavity design is more restricted and the
incident mode area on the saturable absorber is not freely
adjustable, modifying the device structure offers an interesting
solution for adjusting the effective saturation fluence of the
SESAM device to the incident pulse energy density. This is
particularly useful for the passively
-switched monolithic
ring lasers [64] and microchip lasers [65], [66], discussed in
more detail in Section V.
III. S
EMICONDUCTOR SATURABLE
ABSORBER MIRROR (SESAM) DESIGN
A. Material and Device Parameters
Normally grown semiconductor materials have a carrier
recombination time in the nanosecond regime, which tends
to drive many solid-state lasers into
-switching instabilities
(Section II). In addition, nanosecond recovery times do not
provide a fast enough saturable absorber for CW mode-
locking. We use low-temperature grown III–V semiconductors
[5], [7], [67] which exhibit fast carrier trapping into point
defects formed by the excess group-V atoms incorporated
during the LT growth [11], [68], [69]. Fig. 5 shows typical
electron trapping times (i.e., absorber recovery times) and the
nonlinear absorption bleaching as a function of MBE growth
temperature. For growth temperatures as low as 250
C, we
still obtain a good nonlinear modulation of the saturable
absorber with recovery times as low as a few picoseconds.
The tradeoff here is that the nonsaturable absorber losses for
increase with reduced growth temperatures [8].
This tradeoff will ultimately limit the maximum thickness of
the absorber material used inside a solid-state laser cavity.
For femtosecond pulse generation, we can benefit from
the intraband thermalization processes that occur with time
constants from tens to hundreds of femtoseconds, depending
on the excitation intensity and energy [70]. A larger fem-
tosecond modulation depth can be obtained for quantum-well
structures because of the approximately constant density of
states above the bandgap. However, we can strongly reduce
the requirements on this fast recovery time if we do not
use the semiconductor saturable absorber as a fast saturable
absorber, according to Fig. 3(b), but just to start and stabilize
soliton mode-locking. In this case, no quantum-well effects are
absolutely necessary and, therefore, bulk absorber layers are
in most cases sufficient as well. The reduced requirements on
the absorber dynamics also allowed us to demonstrate 50-nm
tunability of a diode-pumped, soliton-mode-locked Cr:LiSAF
laser with a one-quantum-well low-finesse A-FPSA (Fig. 6)
[71], [72]. We would not obtain this broad tunability if the
excitonic nonlinearities in the SESAM provided the dominant
pulse formation process. In addition, in the soliton mode-
locking regime we can also obtain pulses in the 10-fs range
or below, even though the mode-locked spectrum extends
beyond the bandgap of the semiconductor saturable absorber,
for example [56].
We can further adjust the key parameters of the saturable ab-
sorber if we integrate the absorber layer into a device structure.
This allows us to modify the effective absorber cross section
(2) beyond its material value, for example. In addition,
we can obtain negative dispersion compensation by using a
Gire–Tournois mirror or chirped mirrors. In the following, we
will discuss the different device designs in more detail.
![Fig. 3. The three fundamental passive mode-locking models: (a) passive mode-locking with a slow saturable absorber and dynamic gain saturation [27], [28], (b) fast absorber mode-locking [29], [30], and (c) soliton mode-locking [31]–[33].](/figures/fig-3-the-three-fundamental-passive-mode-locking-models-a-2g0qwmmj.webp)
