Optimization of Active Region of Quantum Cascade Laser (QCL) by
Coupled Calculation of Genetic Algorithm and QCL Simulator
Shigeyuki Takagi
1a
, Tsutomu Kakuno
2
, Rei Hashimoto
2
, Kei Kaneko
2
and Shinji Saito
2b
1
Department of Electrical and Electronics Engineering, School of Engineering, Tokyo University of Technology,
1404-1 Katakura, Hachioji, Tokyo, Japan
2
Corporate Manufacturing Engineering Center, Toshiba Corporation, 33 Shinisogo, Isogo, Yokohama, Kanagawa, Japan
Keywords: Quantum Cascade Lasers, QCLs, Active Region, non-Equilibrium Green’s Function, Genetic Algorithm, GA,
Optimization, Gain, Wavelength, Injector Barrier, Well.
Abstract: We applied a coupled calculation of genetic algorithm and a quantum cascade laser (QCL) simulator
(nextnano.QCL) to calculate the gain that excites laser light in the active region of the QCL. The film
thicknesses of the nine layers constituting the active region were changed simultaneously, and the film
structure with the maximum gain was determined from 1000 type of film structures. The QCL simulator
incorporating a non-equilibrium Green's function was used to calculate the gain of the QCL, and the validity
of the simulation was evaluated using the active region structure reported in the previous paper. In the coupled
calculation of the QCL simulator and genetic algorithms, we used gain as an objective function and methods
of crossing, natural selection, and mutation simulating the evolutionary process of living organisms to
optimize the thickness of nine films. As a result of the optimization calculation, the optimized structure had
gain (78.44 cm
–1
) higher than that (50.01 cm
–1
) in a structure reported in a previous paper.
1 INTRODUCTION
Quantum cascade lasers (QCLs) are n-type
semiconductor lasers in which two types of
semiconductor film are alternately stacked, and the
laser light in the infrared region can be obtained (Faist
et al., 1994). Since the wavelengths of QCLs are in
the infrared region, they are expected to be applied to
trace gas analysis and remote gas detection (Faist et
al., 2016). It is necessary to develop a laser with a
wavelength suitable for measurement. With such
trace substance detection and gas detection from a
distance, a higher sensitivity is expected by
increasing the output. Since the amount of laser
absorption is measured in the detection of trace
substances, it is necessary to propagate along a large
optical path length. To develop such high-power
lasers, it is effective to utilize a simulator that can
predict the oscillation wavelength and gain.
Figure 1 shows the band structure of a QCL. It has
an injector region that transports electrons and an
active region consisting of several sets of barrier and
a
https://orcid.org/0009-0009-6444-8748
b
https://orcid.org/0000-0002-1829-6482
well layers to excite laser light. Laser light is emitted
when electrons transition from an upper level to a
lower level in a quantum well formed in well layers.
In the current simulators, the Schrödinger equation is
solved to calculate the wave function, and the rate
equation solution method is used at the level, where
the electrons undergo transition. The amount of light
emitted is calculated semiclassically from the
lifetimes of the upper and lower levels, and the
transition probabilities between the upper and lower
levels (Lu et al., 2006). Recently, a calculation
method using the non-equilibrium Green's function
has been proposed (Grange, 2015). In the simulator,
the distribution of electron density and the transition
of electrons from the upper level to the lower level
can be calculated quantumly.
To increase the output of a QCL, it is necessary to
increase the intensity of the laser light excited in the
active region. An effective method is to use a QCL
simulator to optimize the thicknesses of the barrier
and well layers in the active region by increasing the
laser gain. However, the active region of a typical
72
Takagi, S., Kakuno, T., Hashimoto, R., Kaneko, K. and Saito, S.
Optimization of Active Region of Quantum Cascade Laser (QCL) by Coupled Calculation of Genetic Algorithm and QCL Simulator.
DOI: 10.5220/0012428200003651
Paper published under CC license (CC BY-NC-ND 4.0)
In Proceedings of the 12th International Conference on Photonics, Optics and Laser Technology (PHOTOPTICS 2024), pages 72-77
ISBN: 978-989-758-686-6; ISSN: 2184-4364
Proceedings Copyright © 2024 by SCITEPRESS Science and Technology Publications, Lda.
QCL is composed of four barrier layers and four well
layers. For this reason, the main method of
optimization thus far has been to fix some film
thicknesses and sequentially optimize the film
thickness. Otherwise, optimization was performed
under certain constraints, such as reducing the
thickness of all barrier layers by a certain percentage
(Tanimura et al., 2022). There are few reports on
methods for simultaneously optimizing all the layers
in the active region.
Figure 1: Band structure of QCL. (a) Injector region that
transports electrons and (b) active region that excites laser
light (Tanimura et al., 2022).
In this study, we developed an automatic
calculation method that combines a non-equilibrium
Green's function and a genetic algorithm (GA), and
applied the method to optimizing the film
thicknesses in the active region. Changing the
thicknesses of five types of barrier layer and four
types of well layers as parameters, we calculated the
gain and wavelength of 1000 types of the film
structures. As a result, a structure with optimized
film thickness was obtained with a gain (28.43 cm
1
) higher than that (50.01 cm
1
) in the QCL structure
reported in the paper. We also fabricated a prototype
QCL device, performed electro-luminescence (EL)
measurements, and verified the validity of the
optimized results.
2 QCL SIMULATOR
2.1 QCL Simulator Incorporating
non-Equilibrium Green's Function
We used nextnano.QCL (nextnano GmbH) as the
QCL simulator in this study (Grange, 2015). Figure 2
shows the calculation flow in the simulator. In the
Schrödinger equation, assuming that the unperturbed
Hamiltonian is H
o
and the electron scattering is the
perturbed Hamiltonian H
scatt
, the Hamiltonian H in
the Schrödinger as
H=
+

.
(1)
Using this Hamiltonian, we solved the Schrödinger
equation and calculated the electron orbit.
Figure 2: QCL simulator incorporating non-equilibrium
Green's function (Tanimura et al., 2022).
Next, we solved Poisson's equation to find the
mean-field electrostatics potential and calculated the
self-energy Σ
R
of delay and the self-energy Σ
<
of
Lesser, which describes electron scattering. The
Density of States (DOS) was obtained from the
delayed Green's function G
R
using the Dyson
equation as follows;
=

∑

.
(2)
Using the Keldysh equation shown in equation (3),
we calculated the electron density matrix from the
Lesser Green's function G
<
.
=
Σ


(3)
where G is the advanced Green's function. Current
and gain were calculated based on the basis of the
obtained electron density matrix.
From the calculation flow in Fig. 2, the relationship
between emitted energy (wavelength) and gain was
calculated. By using a non-equilibrium Green's
function, we effect such as the crystal lattice and
electron scattering. This is considered a major
advantage over conventional semiclassical
calculation methods.
2.2 Evaluation of Simulation Model
To examine the validity of this simulator, we inputted
the film structure and composition reported previously
into the QCL simulator (Evans et al., 2007). The film
structure was modeled using two sets of injector and
active regions. We assumed that the maximum gain
obtained in the simulation was the laser oscillation
wavelength and compared it with the oscillation
Optimization of Active Region of Quantum Cascade Laser (QCL) by Coupled Calculation of Genetic Algorithm and QCL Simulator
73
wavelength in the previous paper (Takagi et al., 2021).
Figure 3 shows the calculation results. Figure 3(a)
shows the gain intensity and Fig. 3(b) shows the DOS
distribution. The horizontal axes in both (a) and (b)
indicate the position of the QCL film. As shown in
Fig. 3(a), a high gain was observed at the position
corresponding to the active region. The photon
energy of this gain was 260 meV, which corresponds
to a light wavelength of 4.77 nm. The oscillation
wavelength obtained by Evans et al. in their
experiment was 4.71 nm (Evans et al., 2007), which
is in good agreement with that in our simulation.
Figure 3: Calculation results of QCL simulator. (a) Gain
intensity and (b) DOS.
In Fig. 3(b), there are two stages of electron
distribution in the active region, and the upper and
lower electron distribution regions correspond to the
upper and lower levels of laser oscillation,
respectively. In addition, electrons with almost the
same energy level are distributed in the injector
region, indicating that electrons propagated through
the injector region. Nextnano.QCL was able to
accurately model electron propagation and laser light
emission. As a result, we were able to obtain the same
calculation results for the oscillation wavelength as
the experimental results.
3 COUPLED CALCULATION OF
GENETIC ALGORITHMS AND
QCL SIMULATOR
3.1 Optimization of Parameters
We optimized the structure of the active region that
excites the laser beam shown in Fig. 4. In the active
region, the barrier layer was formed with In
0.363
AlAs,
and the well layer was formed with In
0.669
GaA. The
thicknesses of the nine layers were optimized from
the injector barrier to the fourth barrier layer. The
reference structure of the active region was set to the
same structure of the QCL reported by Evans et al.
(Evans et al., 2007).
Figure 4: Optimizing film thicknesses in active region.
Table 1 shows the ranges of film thicknesses in
the active region. The thickness of each film was
varied within a range of ±1.0 nm from the reference
film thickness. Moreover, for films prepared by
Molecular beam epitaxy (MBE), their thicknesses are
less than 1.0 nm are difficult to control. Therefore, the
minimum film thickness was set to 1.0 nm.
Table 1: Ranges of film thicknesses of barrier and well
layers.
Barrier/well Film
material
[nm]
Thickness
range
[nm]
Minimum
thickness
[nm]
Injector barrier 3.8
± 1.0 1.0
1
st
well 1.2
1
st
barrier 1.3
2
n
d
well 4.3
2
n
d
barrier 1.3
3
r
d
well 3.8
3
r
d
barrier 1.4
4
th
well 3.6
4
th
barrier 2.2
3.2 GA and Optimization Method
To optimize the active region, we developed a
calculation method that combines nextnano.QCL and
GA. GA is an evolutionary computational method for
solving optimization problems. It is one of the search
methods that can rapidly find the optimal solution, as
shown in Fig. 5 (Holland, 1975); (Holland,1992).
This algorithm reflects the evolutionary process of
PHOTOPTICS 2024 - 12th International Conference on Photonics, Optics and Laser Technology
74
living organisms in its optimization process and
consists of three distinctive methods: crossing,
natural selection, and mutation. Crossing and natural
selection can efficiently generate better parameter
combinations. Furthermore, mutation enables the
finding of the optimal solution for parameters that are
far from the initial conditions (Takagi et al., 2023).
Figure 5: Optimization flow using GA.
The optimization software modeFRONTIER
incorporating NSGA-II was used for GA calculations.
NSGA-II is a GAs proposed by Debi et al. (Debi et
al., 2022). It has excellent convergence and
computational stability. Figure 6 shows the flow of
the coupled calculation of modeFRONTIER and
nextnano.QCL. Table 2 shows the calculation method
and setting parameters. In this calculation, the gain
and wavelength were extracted from the results of
calculation using nextnano.QCL.
In the optimization, the parameter sets that
maximize gains as the objective function were
calculated. Furthermore, under the conditions where
the wavelength was not in the range from 4.0 µm to
5.0 µm, we judged the parameter sets to be
inappropriate. In the calculation using the GA, 10
parameter sets were calculated simultaneously as the
same generation, and 10 new parameter sets in the
next generation were created using the gain and
wavelength as evaluation indicators. These
calculation processes were repeated 100 times, and
1000 types of the parameter sets were calculated and
evaluated.
The actual calculation was performed as follows.
Referring to the film thicknesses in the QCL reported
previously (Evans et al., 2007), each film thickness
was determined using random numbers within the
range of initial film thicknesses ± 1.0 nm. Ten
parameter sets were generated and used for the first
generation. The parameter sets were inputted into
Figure 6: Coupled calculation of QCL and optimization
simulators.
Table 2: Calculation method and setting parameters.
Optimization method GA
(NSGA-)
Objective functions Gain, Wavelength
Number of parameter sets in
N
th
generation
10
Number of generations 100
Total number of simulations 1000
nextnano.QCL in order and the gains and
wavelengths were calculated. Among the calculation
results, parameter sets with high gains were
synthesized by taking the average film thickness
(crossing). On the other hand, parameter sets with low
gains were discarded (natural selection). In addition,
a certain film thickness was randomly changed
(mutation) in several parameter sets. Then, ten
parameter sets were generated by crossing, natural
selection, and mutation, and used as the second
generation of parameter sets. By repeating the series
of calculations up to 100
th
generation, we optimized
nine film thicknesses in the active region.
4 SIMULATION RESULTS
Figure 7 shows the modeFRONTIER setting screen
showing the flow of calculation. The upper row
shows the input parameters, the middle row shows the
nextnano.QCL calculation, and the lower row shows
the output items. The input parameters correspond to
the nine film thicknesses shown in Table 1. The nine
film thicknesses of the injector barrier and active
region were inputted into nextnano.QCL, and the gain
and wavelength for each film structure were
Optimization of Active Region of Quantum Cascade Laser (QCL) by Coupled Calculation of Genetic Algorithm and QCL Simulator
75
calculated. On the basis of the gain, the next-
generation parameter set was obtained using the GA
described in Section 3.2, and the process of
optimization proceeded.
Figure 7: Setting screen of modeFRONTIER for
optimization of active region in QCL.
To examine the effect of optimization by
experiments, we added the reference structure
reported previously (Evans et al., 2007) and the
verification structure in which all film thicknesses
were reduced by 10% to the optimization
calculations. The reason for reducing the film
thickness of each layer by –10% was that such a film
can be deposited by shortened MBE process time and
the desired film structure can be reliably produced.
Figure 8 shows the results of the coupled
calculation of nextnano.QCL and genetic algorithm.
The horizontal axis shows the number of calculation,
and the vertical axis shows the gain of the structure
corresponding to the calculation number. As the
generation of genetic algorithms increases, the gain
increases.
The reason why the gain intermittently decreases
during optimization is that the mutation conditions for
the GA were calculated and some of them showed
low gains. As a result, while the gain of the reference
structure was 50.01 cm
–1
, that of the verification
Figure 8: Progress in optimization by coupled calculation
of genetic algorithm and QCL simulator.
structure was as high as 65.00 cm
–1
. Furthermore, in
the final structure, a gain of 78.44 cm
–1
was obtained,
and it was estimated that high-power laser oscillations
can be obtained.
5 EVALUATION OF
OPTIMIZATION RESULTS
To examine the validity of the optimization
calculations, we fabricated prototype devices with the
reference and verification structures, and we
examined the EL light output. Figure 9(a) shows a
photograph of external the prototype QCL chip. The
injector and active regions, which contribute to light
emission, were stacked 33 times, and the chip had a
length of 2 mm and a ridge width of 100 µm.
(a) (b)
Figure 9: Photograph of prototype QCL. (a) QCL chip with
length of 2 mm and ridge width of 100 mm, and (b) QCL
device soldered on CuW mount.
Figure 9(b) shows a QCL device soldered on a
CuW mount. The device was cooled to 77 K and
operated at a frequency of 100 kHz and a pulse width
of 300 nm (3% duty). The emitted light was focused
by a concave mirror, and the EL intensity was
measured by an MCT (HgCdTe) detector.
Nicolet8700 (ThermoScientific Inc.) was used to
measure the EL spectra.
Figure 10 shows the EL spectra of the prototype.
The horizontal axis represents the wavelength and the
vertical axis represents the EL emission intensity. A
strong luminescence was observed for both the
reference and verification structures. The oscillation
wavelengths of the reference and verification
structures were 4.53 μm and 4.77 μm, respectively.
Here, in the simulation results in Section 4, the
wavelengths of the reference and verification
structures were 4.77 nm and 4.93 nm, respectively.
For both film structures, the experimental results
had shorter wavelengths than the simulation results.
However, in both experiments and simulations, the
wavelengths of the verification structure were longer
than those of the reference structure. The trend in the
two types of structure was consistent in experiments
PHOTOPTICS 2024 - 12th International Conference on Photonics, Optics and Laser Technology
76
Figure 10: EL emission intensity of prototype QCL device
(Tanimura et al., 2022).
and simulations. Corresponding to the gain of the
optimization calculation, the emission intensity of the
verification structure was 1.37 times higher than that
of the reference structure. From these results, it is
estimated that the structure optimized using the gain
in the optimization simulation can have a higher
emission intensity than the reference and verification
structures.
6 CONCLUSIONS
We applied a coupled calculation of genetic algorithm
and the QCLsimulator (nextnano.QCL) to calculate
the gain that excites laser light in the active region of
the QCL. The thicknesses of the nine layers
constituting the active region were changed
simultaneously, and the film structure with the
maximum gain was determined from 1000 types of
the parameter sets.
Nextnano.QCL incorporating a non-equilibrium
Green's function was used to calculate the gain of
QCL, and the validity of the simulation was evaluated
using the active region structure reported previously
(Evans et al., 2007). In the coupled calculation of
genetic algorithm and nextnano.QCL, we used gain
as an objective function and used the methods of
crossing, natural selection, and mutation simulating
the evolutionary process of living organisms to
optimize the nine film thicknesses in the active
region. As a result of the optimization calculation, the
optimized structure had a gain (78.44 cm
–1
) higher
than that (50.01 cm
–1
) in the structure reported in a
previous paper.
In addition, as a result of prototyping the QCL of
the reference and verification structures and
measuring the EL emission, the emission intensity of
1.37 higher than that of the literature structure was
obtained for the verification structure, demonstrating
the validity of the optimization.
ACKNOWLEDGEMENTS
This work was supported by Innovative Science and
Technology Initiative for Security, ATLA, Japan.
REFERENCES
Faist, J., Capasso F., Sivco D. L., Sirtori C., Hutchinson A.
L., & Cho A. Y., 1994. “Quantum cascade laser”,
Science, 264, 553–556.
Faist J., Villares G., Scalari G., Rösch M., Bonzon C., Hugi
A., & Beck M., 2016. “Quantum Cascade Laser
Frequency Combs”, Nanophotonics, 5, 272–291.
Lu S. L., Schrottke L., Teitsworth S. W., Hey R., & Grahn
H. T., 2006. “Formation of electric-field domains in
GaAsAlxGa1xAs quantum cascade laser structures”,
Phys. Rev. B, 73, 033311.
Grange, T., 2015. “Contrasting influence of charged
impurities on transport and gain in terahertz quantum
cascade lasers”, Phys. Rev. B, 92, 241306-1–5.
Tanimura H., Takagi S., Kakuno T., Hashimoto R., Kaneko
K., & Saito S., 2022. “Analyses of Optical Gains and
Oscillation Wavelengths for Quantum Cascade Lasers
Using the Nonequilibrium Green’s Function Method”,
Journal of Computer Chemistry, Japan-International
Edition, 8, 2021–0024.
Evans, A., Darvish, S. R., Slivken, S., Nguyen, J., Bai, Y.,
& Razeghi, M., 2007. “Buried heterostructure quantum
cascade lasers with high continuous-wave wall plug
efficiency”, Appl. Phys. Lett. 91, 071101.
Takagi S., Tanimura H., Kakuno T., Hashimoto R, & Saito
S., 2021. “Evaluation of Simulator Incorporating Non-
equilibrium Green’s Function and Improvement of
Quantum Cascade Lasers Output using the Simulator”,
Proceedings of PHOTOPTICS 2021, 58–63.
Holland J. H., 1975. “Adaptation in natural and artificial
systems: an introductory analysis with applications to
biology, control, and artificial intelligence”, University
of Michigan Press.
Holland J. H., 1992. “Genetic Algorithms”, Scientific
American, 267, 66–73.
Takagi S., Sekine M., Nakaegawa T., & Hsiao S.-N., 2023.
“Optimization of RF Frequencies in Dual-Frequency
Capacitively Coupled Plasma Apparatus Using Genetic
Algorithm (GA) and Plasma Simulation”, IEEE Trans.
Semicond. Manuf., 36, ,547–552.
Debi K., Pratap A., Agarwal S., & Meyaraivan T., 2022. “A
Fast and Elitist Multiobjective Genetic Algorithm:
NSGA-II”, IEEE Trans. Comput., 6, 182–197.
Optimization of Active Region of Quantum Cascade Laser (QCL) by Coupled Calculation of Genetic Algorithm and QCL Simulator
77