Continuous Wave Operation of
a Mid-Infrared Semiconductor
Laser at Room Temperature
Mattias Beck,
1
* Daniel Hofstetter,
1
Thierry Aellen,
1
Je´roˆme Faist,
1
Ursula Oesterle,
2
Marc Ilegems,
2
Emilio Gini,
3
Hans Melchior
3
Continuous wave operation of quantum cascade lasers is reported up to a
temperature of 312 kelvin. The devices were fabricated as buried heterostructure
lasers with high-reflection coatings on both laser facets, resulting in continuous
wave operation with optical output power ranging from 17 milliwatts at 292
kelvin to 3 milliwatts at 312 kelvin, at an emission wavelength of 9.1 microme-
ters. The results demonstrate the potential of quantum cascade lasers as con-
tinuous wave mid-infrared light sources for high-resolution spectroscopy, chem-
ical sensing applications, and free-space optical communication systems.
The mid-infrared portion of the spectrum,
covering approximately the wavelength range
from3to12m, is sometimes referred to as
“underdeveloped” because of its lack of con-
venient coherent optical sources. Especially
when compared to the visible or near-infrared
spectral range, where interband semiconduc-
tor lasers are now produced very economical-
ly with continuous wave (CW) output power
of tens of milliwatts, this assertion holds true.
In the mid-infrared, a new class of semicon-
ductor lasers—intersubband quantum cas-
cade (QC) lasers (1)— has become a promis-
ing alternative to interband diode lasers (2, 3)
in the past 7 years. In these devices, photon
emission is obtained by electrons making op-
tical transitions between confined energy lev-
1
Institute of Physics, University of Neuchaˆtel, CH-
2000 Neuchaˆtel, Switzerland.
2
Institute of Micro- and
Optoelectronics, Department of Physics, Swiss Feder-
al Institute of Technology, CH-1015 Lausanne, Swit-
zerland.
3
Institute for Quantum Electronics, Depart-
ment of Physics, Swiss Federal Institute of Technolo-
gy, CH-8093 Zurich, Switzerland.
*To whom correspondence should be addressed. E-
mail: Mattias.Beck@unine.ch
Published in Science 295, no 5553, 301-305, 2002
which should be used for any reference to this work
1
els. As such, QC lasers can be fabricated
from wide-bandgap technologically mature
semiconductors, and their emission wave-
length can be tailored over a wide range by
changing only the layer thicknesses. In addi-
tion, because their main nonradiative mecha-
nism is optical phonon emission and because
of the atomic-like joint density of states of
intersubband transitions, QC lasers exhibit a
gain that is weakly temperature dependent.
As a result, QC lasers have demonstrated
high mid-infrared output powers in pulsed
operation up to temperatures above 400 K
(4 ). However, CW operation of QC lasers
based on standard designs has remained lim-
ited to cryogenic temperatures below 175 K
(5). A recent active region design enabled
CW operation up to 243 K (6 ), which is
barely high enough to be maintained by a
thermoelectric cooler.
Because the main nonradiative mecha-
nism in mid-infrared interband lasers is Au-
ger recombination, these devices exhibit a
much stronger temperature dependence of the
threshold current density, and CW operation
is limited to temperatures below 225 K (2).
Although chemical sensing based on op-
tical absorption has been successfully dem-
onstrated with pulsed QC lasers (7 ), these
systems are typically limited by the fairly
wide emission linewidth of the QC laser
(⬎500 MHz); high sensitivity can only be
achieved with the narrow linewidth of a
CW-operated device (8). Mid-infrared at-
mospheric optical communication systems
(9, 10), using QC lasers for data transmis-
sion through the two transparent atmo-
spheric windows, will potentially benefit
from light sources operating in CW mode at
noncryogenic temperatures.
The limiting factor for CW operation of
semiconductor lasers is the large heat dissi-
pation. At high duty cycles, the temperature
of the active region T
act
is much higher than
the heat sink temperature T
sink
. In a simple
model (11), the heat transport between active
region and heat sink is characterized by the
thermal conductance G
th
per unit of area of
the active region; that is, T
act
⫽ T
sink
⫹ U ⫻
J
th
/G
th
at threshold. Assuming a constant
threshold voltage U and that the temperature
dependence of the threshold current density
J
th
can be described by the phenomenological
relation J
th
⫽ J
0
exp(T
act
/T
0
), the maximum
CW operating temperature T
sink,max
ofaQC
laser is given by T
sink,max
⫽ T
0
⫻ [ln(T
0
G
th
/
J
0
U) – 1]. A high value of the characteristic
temperature T
0
is therefore an absolute neces-
sity to achieve room-temperature CW opera-
tion. The large value of T
0
achieved in recent
QC laser structures (T
0
⬎ 170 K) results in a
low temperature sensitivity of J
th
and shows
the potential of these devices for CW opera-
tion. However, the devices used in the early
attempts to reach high-temperature CW op-
eration (12, 5) had a J
0
U product of about 10
kW/cm
2
. This value, corresponding to a
threshold current density of ⬃5 kA/cm
2
at
300 K, is just too large to be evacuated from
the active region when the laser is operated at
high duty cycles, even assuming an idealized
device geometry.
In addition, the actual core temperature is
equal to T
act
⫽ T
0
⫻ ln(T
0
G
th
/J
0
U) when the
laser is operated at T
sink,max
(13), resulting in
a temperature difference between the active
region and the heat sink of T
0
. For large T
0
values, the device might fail by thermal stress
before reaching the temperature T
sink,max
. For
this reason, it is essential to (i) minimize the
threshold current density and (ii) use a device
geometry that minimizes thermal stress. To
reduce the room-temperature threshold cur-
rent density, active region designs based on a
double-phonon resonance (6 ) and a bound-
to-continuum transition (14 ) were recently
developed. In these structures, the injection
and extraction efficiency to and from the
active region were significantly improved by
means of wave function engineering. Addi-
tionally, growth conditions and the doping
concentration of the active region were opti-
mized. As a consequence, the pulsed thresh-
old current density at 300 K dropped to a
value as low as 3 kA/cm
2
(6 ).
We present QC lasers with an active re-
gion based on a double-phonon resonance
and designed for a lasing transition at an
energy of 135 meV (corresponding to a
wavelength of 9.18 m) between the upper
and lower lasing states (levels 4 and 3 in Fig.
1A). The active region is composed of four
quantum wells (QWs), which results in three
coupled lower energy states (levels 1, 2, and
3) separated from each other by one phonon
energy (15). The active region used a narrow
QW-barrier pair just after the injection barrier
[similar to the classical three-QW design
(16 )] that enhances the injection efficiency
into the upper lasing level by increasing lo-
cally the magnitude of the upper state wave
function (12). The observation of a clear res-
onant tunneling effect in structures with a
three-QW active region demonstrated the ef-
fectiveness of this approach (17 ).
The fast intersubband scattering between
the lowest subbands separated by an optical
phonon energy should lead to a high popula-
tion inversion in a three-QW active region
Fig. 1. (A) Schematic
conduction-band dia-
gram of one period of
the laser core with
moduli-squared rele-
vant wave functions
in the four-QW active
region based on a dou-
ble-phonon resonance.
The red wavy arrow in-
dicates the transition
responsible for laser
action, and green ar-
rows represent optical
phonon transitions. (B)
Computed temperature
contours around the ac-
tive lasing region of the
buried heterostructure
QC laser at room tem-
perature. The solid lines
display the geometry of
the different materials
used in the simulation,
and the dashed lines
represent the geometry
of the laser. (C) Com-
puted thermally in-
duced stress in the ac-
tive region. The maxi-
mum stress occurs at
the corner of the active
region.
2
device, even at room temperature. However, the
relatively slow measured tunneling time from
the active region of about 2 ps (18, 19) intro-
duces an effective bottleneck to the electron
transport, enhancing the lower laser state pop-
ulation. A model of the room-temperature elec-
tron kinetics in the active region (20), which
included optical phonon emission and absorp-
tion from all points in k-space of the active
region subbands, demonstrated that this bottle-
neck effect resulted in an effective lower state
lifetime as long as 0.8 ps for a typical three-QW
structure. The same computation also shows
that the presence of an additional state in the
active region (that is, level 1) allowing the
emission of two optical phonons from the lower
laser level decreases the lifetime of that level
down to 0.5 ps. As a result, the computed ratio
between the upper and lower state population
increased from 1.9 for the three-QW structure
to 2.8 in the double-phonon resonance struc-
ture, simultaneously decreasing the threshold
current density and increasing the slope effi-
ciency and maximum power, as was indeed
observed (6).
Moreover, our devices were processed in
a narrow-stripe, planarized, buried hetero-
structure geometry (21), in which the gain
region was vertically and laterally buried
within InP cladding layers, the geometry of
which provides a number of advantages. The
choice of a buried stripe greatly improves the
heat transport by allowing heat flow from all
sides of the active region. Additionally, the
narrow-stripe geometry also decreases the to-
tal amount of strain that builds up in a mate-
rial subjected to a very strong temperature
gradient. Indeed, the results from a simula-
tion, done with a commercial finite-elements
differential equation solver (22), of both ther-
mal transport and thermally induced stress
lead to the same conclusions (Fig. 1, B and
C). A thermal conductance of 820 W/Kcm
2
is
predicted for a buried, 12-m-wide, junction-
down–mounted device, as compared to the
calculated value of 510 W/Kcm
2
for a 28-
m-wide, ridge, junction-down–mounted de-
vice (6 ). Similarly, the maximum thermally
induced shear stress that builds up at the
edges of the active region (Fig. 1C) decreases
from 22 MPa in the ridge device to 3.6 MPa
for the buried structure.
Fabrication of the laser structure relied on
molecular beam epitaxy for the growth of the
lattice-matched InGaAs/InAlAs laser core on
an InP substrate. The laser core consists of 35
periods, each comprising a partially n-doped
injector region and the undoped four-QW
active region, embedded in an optical
waveguide formed on one side by the sub-
strate and a lower InGaAs waveguide layer
and on the other side by an upper InGaAs
waveguide layer and the InP top cladding,
which was grown by metalorganic vapor-
phase epitaxy (MOVPE).
The lasers were fabricated into 12-m-
wide buried stripes by wet etching and selec-
tive MOVPE regrowth of i-InP using a SiO
2
mask. The devices were then cleaved into
750-m-long lasers, soldered junction-down
onto a diamond platelet, and finally facet-
coated by a ZnSe/PbTe high-reflectivity (R ⫽
0.7) layer pair.
The CW optical output power emitted
from one facet (Fig. 2) was measured with a
calibrated thermopile detector, which was
mounted directly in front of the laser facet. At
room temperature (292 K), the laser exhibited
a threshold current of 390 mA (corresponding
to a threshold current density J
th
⫽ 4.3 kA/
cm
2
at a voltage bias U ⫽ 7.6 V) and a slope
efficiency dP/dI of 101 mW/A (where P is
optical power and I is current). This device
emitted 13 mW of optical power from a
single facet at a driving current of 550 mA,
resulting in a wall plug efficiency of 0.55%
per facet. Continuous wave operation was
observed up to 312 K (39°C). At this temper-
ature, the threshold current increased to 520
mA ( J
th
⫽ 5.8 kA/cm
2
), while still more than
1 mW of output power was emitted at 550
mA. The electrical transport characteristics of
the device (inset of Fig. 2) display the expect-
ed discontinuity of the differential resistance
at threshold (390 mA).
A laser with the same cavity length but a
slightly larger stripe width of 15 m emitted
17 mW per facet at a drive current of 600 mA
at room temperature. This device could be
operated up to 311 K, with a maximum op-
tical power of 3 mW and a threshold current
of 540 mA ( J
th
⫽ 4.8 kA/cm
2
).
The performance of these buried hetero-
structure devices demonstrates the results of
the thermoelastic predictions. Although the
junction-down–mounted, 28-m-wide, con-
ventional ridge lasers failed systematically at
4 kA/cm
2
(6 ), the 12-m-wide buried lasers
discussed in this paper supported more than 6
kA/cm
2
.
The emission frequency of a QC laser
can be tuned over a small range of a few
cm
⫺1
by changing the current and tempera-
ture. The CW spectral properties were ana-
lyzed with a Fourier transform infrared spec-
trometer. The emission spectra (Fig. 3A) col-
lected at a constant heat sink temperature of
292 K and various currents between 395 and
530 mA reveal frequency tuning from
1096.74 cm
⫺1
to 1094.54 cm
⫺1
, linear with
the electrical input power (inset of Fig. 3A).
At a fixed current of 530 mA, the emission
frequency of the laser shifts from 1094.54
cm
⫺1
at 292 K to 1092.90 cm
⫺1
at 313 K
(Fig. 3B). The measured center frequencies
are well fitted by a linear function (inset of
Fig. 3B) with a tuning coefficient of ⌬/⌬T ⫽
– 0.078 cm
⫺1
/K. Single mode emission was
observed for this particular device over the
whole investigated current and temperature
range, with a side mode suppression ratio
better than 30 dB. This rather surprising fact
can be explained by a small defect within the
laser cavity, as indicated by an intensity mod-
ulation of the subthreshold Fabry-Perot fring-
es at twice the cavity mode spacing.
Assuming that the emission frequency
is a function only of the temperature of the
active region, we can deduce a thermal
resistance R
th
of the device from the vari-
ation of the emission frequency (11) with
R
th
⫽
冉
⌬T
⌬P
冊
⫽
冉
⌬
⌬P
冊
⫻
冉
⌬
⌬T
冊
⫺1
.
Inserting the above tuning rates, we get a
thermal resistance of 19.4 K/W in the range
between 292 K and 313 K (corresponding to a
thermal conductance G
th
of 574 W/Kcm
2
).
This R
th
value is higher than the calculated one
(13.6 K/W), most likely because the thermal
interface resistance was not included in our
simulation.
The dependence of the threshold current
density J
th
on the actual core temperature T
act
of the laser is measured in pulsed mode at
low duty cycles, where heating effects are
negligible (that is, T
act
⬃ T
sink
). At 292 K, we
measured a pulsed J
th
as low as 3.1 kA/cm
2
(compared to 4.3 kA/cm
2
in CW operation)
for the 12-m-wide device, corresponding to
a threshold current I
th
⫽ 280 mA. The exper-
imental pulsed threshold current densities
(Fig. 4) can be fitted by the expression J
th
⫽
J
0
exp(T
act
/T
0
) with a T
0
⫽ 171 K and J
0
⫽
560 A/cm
2
.
The temperature dependence of the CW
threshold current density can be computed
from the data in pulsed operation, using a
slightly modified model that takes into ac-
count the change in applied voltage with
Current (A)
Optical power (mW/facet)
0 0.10.20.30.40.50.6
0
3
6
9
12
1
5
CW
T=292K
302K
312K
297K
307K
Current (A)
Diff. resistance
292K
0
5
10
15
20
25
30
0 0.2 0.4 0.6
0
3
6
9
Voltage (V)
Fig. 2. CW optical power from a single laser
facet as a function of drive current for various
heat sink temperatures. The laser is 0.75 mm
long and 12 m wide. The power was measured
with near-unity collection efficiency and a cal-
ibrated thermopile detector. (Inset) Electrical
transport characteristics of the laser at 292 K:
bias voltage as a function of injection current
and differential resistance deduced from the V-I
curve.
3
injected current (23) and uses the thermal
conductance G
th
obtained from the spectral
measurements. The calculated J
th
curve fits
well with the experimental CW J
th
values
(solid symbols in Fig. 4). The data also
show that, at 292 K, the temperature differ-
ence ⌬T ⫽ T
act
– T
sink
between laser core
and heat sink is 58 K at threshold and
increases to 87 K at the maximum injected
current. With this modified model, we cal-
culate a maximum CW operating tempera-
ture T
sink,max
⫽ 321 K and a ⌬T ⫽ 119Kat
that operating temperature.
The threshold currents of our buried het-
erostructure QC lasers scale accurately with
the laser stripe width (inset of Fig. 4). It
means that our device architecture does not
introduce additional lateral waveguide losses
or current leakage paths.
The far-field distributions in the two direc-
tions parallel and perpendicular to the grown
layers exhibited a Gaussian profile with far-
field angle of 40° full width at half maximum in
the in-plane direction and 80° perpendicular to
the layers, proving that the device oscillates in
its fundamental lateral and transverse mode.
Fundamental intersubband processes did not
limit the device performance. Assuming the
measured waveguide loss value of 10 cm
⫺1
(24) and in the limiting case of unity injection
efficiency, the threshold current density of our
QC laser should be 2.1 kA/cm
2
(25), which is
significantly lower than the measured value
(3.1 kA/cm
2
). We believe that further improve-
ments in active region design and growth con-
ditions should bring us closer to this limiting
case. In addition, a reduction of the stripe width
to5to6m should further improve G
th
(by
about 20%) while still maintaining a large value
of confinement factor. In that case, our model
predicts a maximum CW operating temperature
of T ⬎ 370 K.
References and Notes
1. J. Faist et al., Science 264, 553 (1994).
2. Z. Feit et al., Appl. Phys. Lett. 68, 738 (1996).
3. W. W. Bewley et al., Appl. Phys. Lett. 76, 256 (2000).
4. C. Gmachl et al., IEE Electron. Lett. 36, 723 (2000).
5. C. Gmachl et al., IEEE Photon. Technol. Lett. 11, 1369
(1999).
6. D. Hofstetter et al., Appl. Phys. Lett. 78, 1964 (2001).
7. K. Namjou et al., Opt. Lett. 23, 219 (1998).
8. A. A. Kosterev et al., Opt. Lett. 24, 1762 (1999).
9. R. Martini et al., IEE Electron. Lett. 37, 111 (2001).
10. S. Blaser et al., IEE Electron. Lett. 37, 778 (2001).
11. J. Faist et al., IEEE J. Quantum Electron. 34, 336
(1998).
12. J. Faist et al., Appl. Phys. Lett. 68, 3680 (1996).
13. This follows from the derivation of the equation
T
sink
⫽ T
act
– U ⫻ J
0
exp(T
act
/T
0
)/G
th
.
14. J. Faist et al., Appl. Phys. Lett. 78, 147 (2001).
15. The layer sequence of one period, in nanometers,
from left to right in Fig. 1A and starting with the
injection barrier, is as follows: 4.0/1.9/0.7/5.8/0.9/
5.7/0.9/5.0/2.2/3.4/1.4/3.3/1.3/3.2/1.5/3.1/1.9/
3.0/2.3/2.9/2.5/2.9, where InAlAs barrier layers are
in bold, InGaAs well layers are in roman, and
n-doped layers (2 ⫻ 10
17
cm
⫺3
) are underlined.
16. The three-QW active region design has widely been
used in QC devices. Its schematic band structure rough-
ly corresponds to the first three QWs, counting from
the injection barrier, of the active region in Fig. 1A. The
corresponding energy spectrum is similar to the double-
phonon resonance device but without the n ⫽ 1 state.
17. C. Sirtori et al., IEEE J. Quantum Electron. 34, 1722
(1998).
18. J. Faist et al., Phys. Rev. Lett. 76, 411 (1996).
19. S. Blaser et al., IEEE J. Quantum Electron. 37, 448
(2001).
20. The electron kinetic computation assumed that the
electrons interacted with a thermal distribution of
bulk optical phonons at 300 K. The emission (e) and
absorption (a) lifetimes from the active region states
are (in ps):
43e
⫽ 1.88,
43a
⫽ 9.3,
42e
⫽ 1.92,
42a
⫽ 8.5,
41e
⫽ 2.51,
41a
⫽ 11.5,
32e
⫽ 0.73,
32a
⫽
3.8,
31e
⫽ 0.23,
31a
⫽ 2.3,
21e
⫽ 0.28,
21a
⫽ 2.7
(n ⫽ 4 is the upper laser level). The lifetime for
intrasubband processes is
em
⫽ 0.14 ps for emission
and
abs
⫽ 0.55 ps for absorption of optical phonons.
The escape time
esc
⫽ 2 ps was the same for all the
lower (n ⫽ 1..3) states.
21. M. Beck et al., IEEE Photon. Technol. Lett. 12, 1450
(2000).
22. The simulation of the thermoelastic behavior of the
device was done with a commercial finite-elements
software package (PDease2D). Room-temperature
thermal conductivities were used. The elastic prop-
erties of the semiconductor were approximated by
Wavelength (µm)
1092
1093
1094
1095
290 300 310 320
Temperature (K)
Frequency (cm
-1
)
292 K
313 K
Emission frequency (cm
-1
)
1093 1095 1097 1099
1091
9.16 9.14 9.12 9.10
395 mA
525 mA
1094
1095
1096
1097
3.0 3.5 4.0 4.5
Electrical power (W)
Frequency (cm
-1
)
Intensity (a.u.)Intensity (a.u.)
A
B
Fig. 3. (A) High-reso-
lution (0.125 cm
⫺1
)
CW spectra as a func-
tion of injection cur-
rent. The emission spec-
tra were measured at a
constant temperature
of 292 K for various
drive currents ranging
from 395 up to 525 mA
in steps of 10 mA. The
curves are normalized
to the emission spec-
trum measured at 525
mA and plotted in lin-
ear scale. a.u., arbitrary
units. (Inset) Emitted
peak frequency in de-
pendence of the elec-
trical input power at
constant temperature.
We deduced a linear
tuning coefficient of
⌬/⌬P ⫽ –1.51 cm
⫺1
W
⫺1
.(B) Series of CW
emission spectra as a
function of temperature
at a constant drive cur-
rent of 530 mA. The
temperature varies from
292 to 313 K. The curves
are normalized to the
emission spectrum mea-
sured at 292 K. (Inset)
Measured temperature
dependence of the cen-
ter emission frequency
at a fixed injection
current.
Heat sink temperature (K)
275 300 325 350375 400 425 450 475
Threshold current density (kA/cm
2
)
3
4
5
6
7
CW
pulsed
Laser width (µm)
Threshold current (A)
0
0.2
0.4
0.6
0.8
1.0
0 10 2030
pulsed mode
Fig. 4. Threshold current density as a function of
heat sink temperature. The experimental pulsed
J
th
(open symbols) are fitted by the function J
th
⫽
J
0
⫻ exp(T/T
0
), with T
0
⫽ 171 K and J
0
⫽ 560
A/cm
2
(lower solid line). Solid symbols represent
the CW threshold current densities of the same
device. The upper solid line is the computed de-
pendence of the CW threshold current density on
the heat sink temperature, using the modified
model with a current-dependent voltage bias.
(Inset) Pulsed threshold current as a function of
laser width. The threshold currents of the facet-
coated devices (open symbols) are ⬃20% lower
as compared to the threshold currents of the
same devices but without facet coatings (sol-
id symbols).
4
isotropic elastic constants with the Young modulus
E ⫽ 84 GPa and Poisson’s ratio ⫽0.3.
23. At 292 K, the voltage-current (V-I) curve above
threshold was modeled by U ⫽ U
0
⫹ R ⫻ I, with
U
0
⫽ 5.34 V and R ⫽ 5.87 ohm.
24. M. Rochat, M. Beck, J. Faist, U. Oesterle, Appl. Phys.
Lett. 78, 1967 (2001). This value agrees with the one
computed by a Drude model.
25. Computed with the formula used in (11), with the
following parameters (at 300 K): period length L
p
⫽
60 nm; dipole matrix element z ⫽ 3 nm; intersub-
band lifetimes
32
⫽ 1.45 ps,
3
⫽ 0.52 ps,
and an effective
2
⫽ 0.5 ps; overlap factor ⌫
p
⫽
0.0183; broadening of the transition 2␥
32
⫽ 22
meV; and effective index n
eff
⫽ 3.2 and n
th
⫽ 8 ⫻
10
8
cm
⫺2
.
26. We gratefully acknowledge M. Ebno¨ther for tech-
nical assistance with the lateral InP regrowth.
Financially supported by the Swiss National Science
Foundation and the Science Foundation of the Euro-
pean Community under IST project SUPERSMILE.
5
