IEEE JOURNAL OF PHOTOVOLTAICS, VOL. 8, NO.

5, SEPTEMBER 2018 1363

Modeling Multijunction Solar Cells by Nonlocal

Tunneling and Subcell Analysis
Yiming Liu , Mehrad Ahmadpour , Jost Adam, Jakob Kjelstrup-Hansen , Horst-Günter Rubahn,
and Morten Madsen

Abstract—Multijunction solar cells have been demonstrated as and allow competitively low-cost solar modules. Multiple tan-
an efficient approach to overcome the Shockley–Queisser limit and dem structures based on currently popular photovoltaic materi-
achieve high efficiency. However, cost reduction and further devel- als have been proposed and developed [5]–[11]; however, the
opment of high-efficiency tandem solar cells require many device
optimization studies and ask for heavy theoretical assistance. In complex electrical and optical coupling within the tandem de-
this paper, we present a versatile alternative via an updated ver- vices makes the optimization almost impossible by experimental
sion of a free solar cell simulation software, wxAMPS, which in- efforts alone. Therefore, device modeling for tandem solar cells
corporates a nonlocal band-to-band tunneling model and several has been developed and employed and turned out to be an in-
specific physical mechanisms into the solver to better describe the formative approach to accelerate the optimization procedures
device behavior of tandem solar cells. The simulation results are
compared against a commercial technology computer-aided design [12], [13].
program, showing satisfactory agreement. Moreover, a useful fea- At present, there are several solar cell modeling codes avail-
ture, subcell analysis, has been developed to give better insight into able to the PV community, such as the free programs PC1D [14],
the individual subcell performance of the devices and to provide SCAPS [15], and AFORS-HET [16], and the commercial codes
guidance for current matching. The algorithm used in the subcell
ASA [17], ASPIN3 [18], and SETFOS [19]. These programs
analysis, which avoids the complexity in treating nonlocal behav-
iors of tunneling junctions, notably accelerates the tandem solar are developed based on the classic semiconductor drift-diffusion
cell simulation, and thus allows implementing batch simulation for model [20], and most have been further adapted by incorpora-
device optimization. The results of the subcell analysis have been ting thermionic emission [21] and specific tunneling mecha-
used to experimentally optimize subcell thicknesses and success- nisms (tunneling through insulator layer [16], intraband tun-
fully improved the efficiency of the tandem solar cell. neling [22], and/or trap-assisted tunneling [23]) to better treat
Index Terms—Numerical simulation, photovoltaic cells, the carrier transport at hetero-interfaces. In terms of monolith-
tunneling. ically grown multijunction devices, where subcells are inter-
connected via Esaki tunneling diodes [24], another important
I. INTRODUCTION
tunneling mechanism, band-to-band (BTB) tunneling, must be
ULTIJUNCTION solar cells have been demonstrated taken into account to correctly describe the tunneling junction
M as state-of-the-art technology for achieving high power
conversion efficiencies in photovoltaic devices. So far, the high-
behavior [12]. For example, the AMPS gradient band-gap and
drift-dependent mobility approach [25] and the trap-assisted
est solar cell efficiency has been obtained by monolithically tunneling model cannot reproduce the negative resistance phe-
stacked four-junction structures based on III–V semiconduc- nomenon that is typical of Esaki diode. However, this BTB tun-
tor materials [1]. Single-junction solar cells are approaching neling mechanism is missing in most solar cell simulation pro-
their fundamental limits [2]–[4], and tandem solar cells are be- grams. As a result, in order to simulate the current–voltage char-
lieved to be a practical solution to further improve the efficiency acteristics of multijunction devices, a majority of the modeling
investigations have to combine optical simulation with simpli-
Manuscript received November 25, 2017; revised March 15, 2018 and May fied electrical models [6], [8], [26]. Up to now, the fully coupled
16, 2018; accepted June 25, 2018. Date of publication July 11, 2018; date electrical/optical modeling of tandem solar cells is mostly car-
of current version August 20, 2018. This work was supported in part by the ried out by using commercial technology computer-aided design
European Union Seventh Framework Programme under Grant 607232 [THIN-
FACE] and in part by the SDU2020 project “Production of Next-Generation (TCAD) tools [12], [13], [27], or in-house developed programs
Energy Devices.” The research leading to these results was supported by the [28]. The lack of free and open-source software for tandem solar
RollFlex project, which was financed by Interreg Deutschland-Danmark with cell modeling limits the theoretical studies in the field.
means from the European Regional Development Fund. (Corresponding author:
Morten Madsen.) Here, we report on recent enhancements to wxAMPS, a free
The authors are with the SDU NanoSYD Center, Mads Clausen Institute, solar cell simulation program, which already includes intraband
University of Southern Denmark, Sønderborg DK-6400, Denmark (e-mail:, tunneling and trap-assisted tunneling [29], and which has al-
[email protected]; [email protected]; [email protected];
[email protected]; [email protected]; [email protected]). ready been successfully employed in modeling a variety of solar
This paper has supplementary downloadable material available at http:// cells [30]–[35]. The updated version (wxAMPS 3.0) incorpo-
ieeexplore.ieee.org. rates the BTB tunneling to meet the growing demand for multi-
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. junction device modeling. In addition, several new features, such
Digital Object Identifier 10.1109/JPHOTOV.2018.2851308 as band-gap narrowing [36], field-dependent mobilities [27],

2156-3381 © 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications standards/publications/rights/index.html for more information.

Authorized licensed use limited to: Nanyang Technological University Library. Downloaded on March 14,2024 at 15:23:29 UTC from IEEE Xplore. Restrictions apply.
1364 IEEE JOURNAL OF PHOTOVOLTAICS, VOL. 8, NO. 5, SEPTEMBER 2018

semiclassical drift-diffusion model. In local models [39], [41],

this recombination term RBTB is determined by local variables,
like local electric field, band gap, or the gradient of Fermi
levels. The scenario in nonlocal models is more complex but
reflects the realistic nonlocal phenomena of the BTB tunneling.
In nonlocal models, the tunneling probability depends on the
whole potential profile across the tunneling region, and RBTB is
determined by much more variables across the whole tunneling
region, as indicated in (1).
The local models are more easily integrated into the drift-
diffusion solver and are usually numerically robust. Neverthe-
less, imprecise physical assumptions, such as a constant elec-
tric field across the tunneling junction and equal recombination
rates for electrons and holes, are typically used to simplify the
physical description of tunneling at MOS gate interfaces. These
assumptions are not applicable to the tunneling diode of multi-
Fig. 1. Illustration of basic physical concepts. (a) Local. (b) Nonlocal BTB
junction solar cells. The comparison between local and nonlocal
tunneling models. E c , E v represent conduction band and valence band, re- tunneling models has already been carried out in previous work,
spectively. R bulk denotes the bulk recombination caused by nonradiative defect- which demonstrates that nonlocal models are better choices for
related SRH recombination or radiative recombination [42]. R BTB is linked to
the tunneling current Jt by the relation qR BTB (x) = dJt (x)/dx, where q is
correctly describing the tunneling junction behavior [12]. There-
the basic electron charge. (a) In local tunneling models, R BTB only involves fore, in the updated version of wxAMPS 3.0, the BTB tunneling
electrons and holes at the same position. As a result, the electrons and holes model is also based on the nonlocal tunneling concept.
share the same recombination rates, and the maximum of R BTB is usually lo-
cated at the middle of the tunneling junction where the n–p product is high.
The rate of RBTB between tunneling points xi and xj at the
(b) In nonlocal tunneling models, R BTB occurs between electrons and holes on energy level E is given by [43]
different sides of the tunneling barrier at the same energy. Recombination rates
of electrons and holes may be different across the tunneling region and needed  E f p ( i ) −E 
to be treated separately. R p and R n stand for R BTB of holes and electrons,
A∗ T F 1+e kT
respectively. As indicated, the R BTB profile is totally different from that in local RBTB (i, j, E) = CV Γ (i, j, E) ln E f n ( j ) −E
models. qk 1+e kT

(1)
and the transfer matrix method [37], are also integrated, in order where ACV ∗ is the effective Richardson constant, T is the tem-
to render the code more suitable to simulate/optimize III–V mul- perature, F is the electric field, k is the Boltzmann constant, and
tijunction solar cells as well as inorganic/organic hybrid tandem Ef p and Ef n are quasi-Fermi levels for electrons and holes, re-
solar cells. The new version of wxAMPS code has also ex- spectively. The BTB tunneling probability Γ(i, j, E) between xi
plicitly integrated a useful and commonly employed approach, and xj , at the energy level E, is calculated by using a two-band
separated subcell analysis, and allows such subcell analysis eas- dispersion relation
ier and faster to perform.
x
−2 xi
j √κ c2κ v dx
II. BAND-TO-BAND TUNNELING Γ (i, j, E) = e κc+κ2
v (2)
A. Physics
where the local wave numbers
The theory of BTB tunneling has been developed since the
late 1950s [38], [39], and corresponding implementations have 
been integrated into TCAD simulators since the 1990s [40]. Its κc = 2mc |Ec (x) − E|/ (3)
application in TCAD modeling was first implemented for an- 
κv = 2mv |E − Ev (x)|/ (4)
alyzing tunneling transistors, e.g., nanometer-scaled MOSFETs,
where BTB tunneling began to be an important current trans-
port mechanism and required more detailed modeling. With the are based on the Wentzel–Kramers–Brillouin approximation
modeling progress of BTB tunneling and the success in the [12]. Here, h̄ denotes the reduced Plank constant, and mc , mv
simulation of single-junction solar cells, BTB tunneling mod- are the effective masses of electrons and holes, respectively. At
els were then extended to the analysis of tandem solar cells by thermal equilibrium, Ef p (i) = Ef n (j) and, as expected, (1)
commercial TCAD software [12], [13]. equals zero. When numerically calculating (2), a finely tuned
So far, the BTB tunneling models generally fall into two mesh at the tunneling junction is required so as to evaluate the
categories: local tunneling models and nonlocal tunneling integral and tunneling probability accurately.
models (illustrated in Fig. 1). In both models, the tunneling RBTB is incorporated into the current continuity equations for
current is represented by an additional recombination term nodes at xi and xj . In the case of the energy band diagram in
(or generation term with negative sign in some references) Fig. 1(b), the nonlocal RBTB (i, j, E) is related to the holes at xi
in the electron and hole current continuity equations in the and electrons at xj . Hence, the current continuity equations for

Authorized licensed use limited to: Nanyang Technological University Library. Downloaded on March 14,2024 at 15:23:29 UTC from IEEE Xplore. Restrictions apply.
LIU et al.: MODELING MULTIJUNCTION SOLAR CELLS BY NONLOCAL 1365

carriers at position xi are

1
∇Jn (xi ) = Rbulk (xi ) − G(xi ) (5)
q
1
∇Jp (xi ) = G(xi ) − Rbulk (xi ) − RBTB (i, j, E). (6)
q
For carriers at position xj , the continuity equations read
1
∇Jn (xj ) = Rbulk (xj ) + RBTB (i, j, E) − G(xj ), (7)
q
1
∇Jp (xj ) = G(xj ) − Rbulk (xj ) (8)
q
where G is the generation rate caused by photon absorption, Fig. 2. Given Jacobian matrix J and the vector f1–n , the matrix equations
J Δd = f have to be solved in each iteration. In 1-D local models, J is a well-
and Jn and Jp are the current densities of electrons and holes, structured matrix with two bands (a 1–n and c1–n ) confined to the main diagonal
respectively. Equations (5)–(8) are only employed for the simu- band (b1–n ), and the variables Δd1–n can be solved by using a standard
lation nodes in the tunneling junction. Outside of the tunneling algorithm, such as LU decomposition [45]. In nonlocal models, as denoted
by the addition of red asterisks, J becomes a nonregular sparse matrix. These
junction region (in wxAMPS 3.0 it is judged by the strength additional derivative elements significantly increase the coding complexity and
of the local electric field, 105 V/cm currently used), the RBTB require much more complicated algorithms to solve the matrix equations.
term is eliminated and carrier transport returns to traditional
drift-diffusion descriptions. equations separately and sequentially. As indicated in [44],
Newton’s method converges quickly but is very sensitive to
B. Algorithm the initial guess. Gummel’s method is not sensitive to the initial
Adapting the current continuity equations according to guess but converges very slowly when the three equations are
(5)–(8) and discretizing these equations together with the Pois- heavily coupled, for example, for large recombination rates.
son equation by using the finite difference method [44] give In the wxAMPS solver, Gummel’s method is used first to
rise to a large system of nonlinear equations, with three key solve the nonlocal BTB tunneling model. As RBTB in tunneling
variables (local vacuum level ψ, Ef n , Ef p ) at each node. This regions can exceed 1030 cm–3 s–1 , Gummel’s method converges
system can be solved by means of numerical iterative methods, very slowly after a few iterations. Newton’s iteration is then
such as Newton’s method, used in most TCAD solvers [43]. employed to accelerate the convergence, and due to better ini-
The nonlocal property of RBTB , however, results in a signif- tial guess values provided by the previous Gummel’s iteration,
icant complexity in the Jacobian matrix when using the gen- the stability of Newton’s method is greatly improved. In both
eralized Newton method. As seen from (1)–(4), RBTB (i, j, E) methods, BTB tunneling pairs are updated every time the band
depends on values of ψ from xi to xj , and thus the Jacobian diagram is changed in each iteration. Finally, the self-consistent
matrix requires the derivatives dRBTB (i, j, E)/dψi –j , which solution for the nonlocal model is obtained when a specified
are very complex to calculate concurrently. In local tunneling numerical precision is reached.
models, such complicated derivatives are reduced to the form Similar to commercial TCAD [43], it takes significantly
of dRBTB (i)/dψi , and the Jacobian matrix becomes a regular longer time to solve the nonlocal tunneling model than the local
banded matrix whose equations can be solved by well-developed model. The algorithm in wxAMPS is still being improved so as
algorithms. This is why local tunneling models possess a good to increase the convergence speed and the code robustness.
numerical robustness. In the nonlocal case, the existence of
the elements dRBTB (i, j, E)/dψi –j , dRBTB (i, j, E)/dEf p (i), C. Esaki Diode Modeling
dRBTB (i, j, E)/dEf n (j) makes the Jacobian matrix a nonreg-
In order to validate the BTB tunneling model implemented
ular sparse matrix (illustrated in Fig. 2). It requires much more
in wxAMPS 3.0, an Esaki diode was simulated with the param-
complex sparse matrix technology to solve the nonlinear set
eters from [48] where the same Esaki diode was modeled by
of equations, and such Jacobian matrices are very close to be
a commercial TCAD program. The modeling results predicted
singular. These difficulties result in a great challenge in solving
by the two programs are compared in Fig. 3. The energy band
the nonlocal tunneling model, and so far only few commercial
diagram and current–voltage characteristics are very close to the
TCAD simulators can fulfill this task to model multijunction
values reported in [48].
solar cells.
In wxAMPS 3.0, we employ an improved algorithm, which
combines the advantages of the Newton and Gummel meth- III. MULTIJUNCTION SOLAR CELL MODELING
ods [46], [47], and have developed our own sparse matrix solver Having incorporated BTB tunneling into the solver, the code
based on LU decomposition to handle the Jacobian matrix equa- is now capable of handling three tunneling mechanisms and
tions [45]. In Newton’s method, Poisson and current continu- becomes more suitable to model tandem solar cells of differ-
ity equations are solved together for each iteration, whereas ent types. In addition to changes in the tunneling model, the
Gummel’s method solves the Poisson and current continuity revised code also integrates field-dependent mobility and the

Authorized licensed use limited to: Nanyang Technological University Library. Downloaded on March 14,2024 at 15:23:29 UTC from IEEE Xplore. Restrictions apply.
1366 IEEE JOURNAL OF PHOTOVOLTAICS, VOL. 8, NO. 5, SEPTEMBER 2018

Fig. 4. Band diagram of the simulated InGaP/GaAs tandem cell at short

circuit. Light enters from the left side. As seen from the area highlighted by
the dashed ellipse, a reverse bias voltage develops, lowering the energy band
around the space charge region of the GaAs bottom cell. This causes E f n of the
n-GaAs to be lower than E f p of the p-GaAs. The InGaP top cell is forwardly
biased to decrease its photocurrent and guarantee that the total voltage across the
entire device is 0 at short circuit. The forward bias causes E f n of the n-InGaP
to be farther away from E f p of the p-InGaP.

Fig. 3. (a) Energy band diagram at thermal equilibrium for an InGaP p–n
junction with high doping densities of 8 × 1019 and 9 × 1018 cm–3 for the
p- and n-regions, respectively. The Fermi energy E f = 0 by definition at ther-
mal equilibrium. Band gap narrowing is turned ON in the model and is important
to the fit. For example, in the p-region, the InGaP band gap is narrowed by heavy
doping from the input value of 1.42 to ∼1.34 eV. (b) Current–voltage (IV) curves
obtained from wxAMPS 3.0 (with different mesh sizes at tunneling junction)
and Silvaco. The small discrepancy in the peak current and peak position may
arise from different implementations of evaluating tunneling probability [48] or
in the mesh setup at the tunneling region. Fig. 5. Modeled optical generation rates across the device. The AM1.5 spec-
trum is used here.

band-gap narrowing. These can be turned ON or OFF to check device operation picture at short circuit, and the behavior of the
their significance with respect to the final device performance. tunneling junction can be examined. From this it can be shown
The optical model has also been updated by utilizing the trans- that the bottom cell limits the current, since it is reversely biased
fer matrix method, which considers the internal reflection and to match the higher photocurrent generated by the top cell.
transmission between internal layers. Fig. 5 shows the optical modeling results by using a transfer
A typical InGaP/GaAs dual junction solar cell was simulated matrix method. At the bottom of each subcell, a clear optical
incorporating the same device structure that was studied in [13]. interference pattern can be observed. The simulation results are
The modeling parameters are listed in Table S1 (Supplementary compared with those reported in [13], and as shown in Fig. 6,
information). The tunneling junction IV curve (similar to the the two models agree well.
one in [13] and thus not shown here) demonstrates that the The calculation of external quantum efficiency (EQE) is con-
contact resistance of the tunneling junction is around 0.3 Ωcm2 . sistent with the implementation of real EQE measurement of
Considering the tandem device is working under currents below tandem solar cells [49]. In this case, two subsequent bias lights
Jsc , 8.5 mA/cm2 in this case, the voltage drop at the tunneling with different wavelength ranges are input before scanning the
junction is expected to be around 3 mV. chopped monochromatic light with varied wavelengths to cal-
The device simulation results are shown in Figs. 4–6. From culate the EQE of the related subcell. In the first step, a long
Fig. 4, a band diagram obtained by wxAMPS 3.0, the full wavelength (700 nm is used in this case) is input as a bias
Authorized licensed use limited to: Nanyang Technological University Library. Downloaded on March 14,2024 at 15:23:29 UTC from IEEE Xplore. Restrictions apply.
LIU et al.: MODELING MULTIJUNCTION SOLAR CELLS BY NONLOCAL 1367

Fig. 7. Simulated subcell IV curves and corresponding pseudo-IV curve for the
InGaP/GaAs dual-junction cell investigated in the previous section. The subcell
analysis only consumes seconds of computation time with a common computer,
and the generated pseudo-IV curve is very close to the curve simulated without
simplifying the tunneling junction [seen in Fig. 6(a)]. Note from the red curve
that, as before, the bottom cell limits the current of the entire device. In some
simple tandem solar cell models [6], the short-circuit current Jsc of the whole
cell is chosen as the minimum of the subcell Jsc values. This is not always true.
If the currents do not match, some subcells will be positively biased and others
are negatively biased. As a result, Jsc of the entire device will be a little higher
than the subcell Jsc minimum.

solution. Taking the InGaP/GaAs device discussed in Sec-

tion III for example, it takes almost half an hour to obtain the IV
curve with the simulation running on a common-configuration
desktop computer. The extensive computing time makes it
tedious to carry out systematic modeling investigations or
implement device optimization for tandem cells. In order to
Fig. 6. (a) Simulated IV curves obtained by wxAMPS 3.0 and reported in [13]
for the same InGaP/GaAs tandem cell. In wxAMPS, the open-circuit voltage facilitate batch simulations for tandem cells, wxAMPS 3.0 has
was fitted to be ∼2.1 V by tuning defect parameters in the InGaP and GaAs been updated with a useful feature, subcell analysis.
absorber layers. (b) External quantum efficiency (EQE) predicted by wxAMPS When using subcell analysis, the tunneling junction is treated
and Silvaco for the investigated tandem cell. For the bottom cell EQE, an
interference pattern is obtained from the transfer matrix method in the optical as an ideal Ohmic contact. Each subcell is simulated separately,
model. The disagreement in the bottom cell EQE between the two models might based on the previous algorithm for modeling single junctions,
arise from differences in the complex refractive indices used in the two models. but with the optical generation rates appropriate to the full tan-
dem structure. Using the subcell IV curves and series connection
light to generate a photocurrent in the bottom cell, and then a of the devices, a “pseudo-IV curve” for the whole cell can be
monochromatic light from 400 to 900 nm with adjustable inten- obtained (seen in Fig. 7). Voc from pseudo-IV is slightly higher
sity is used to produce the EQE of the top cell. The second step (<5 mV) than that of simulated IV shown in Fig. 6(a), since
then uses a shorter wavelength bias light (480 nm) to generate a there is no voltage drop at the tunneling junction when the tun-
photocurrent in the top cell, and the monochromatic light with neling junction is assumed as Ohmic contact. This slight Voc
the same wavelength range as the first step probes the EQE of difference is consistent with the voltage drop value (∼3 mV)
the bottom cell. The weighted integral of the top and bottom discussed in the previous section.
EQE give Jsc EQE as 18.9 mA/cm2 , which is bigger than two This approach of subcell analysis avoids the complexity in
times the Jsc value in Fig. 6(a) (8.53 mA/cm2 given by wx- the nonlocal BTB tunneling model, converges fast, and the
AMPS) and implies a current mismatch in the device. The EQE simulated device outputs will be very close to those from the
also shows that the optical loss of this tandem cell is mainly due “whole-cell” analysis as long as the tunneling junction is not
to the reflection at the top surface, where antireflective coating the bottleneck of the photocurrent. The subcell analysis reveals
was not present [13]. more details for each subcell behavior and is able to locate which
one is the bottleneck for the current matching. This feature is
very helpful in the current matching analysis and device opti-
IV. SUBCELL ANALYSIS
mization, which usually requires a large number of simulation
As mentioned in Section II-B, the nonlocal BTB tunneling trials. In addition, the subcell IV curves can also be utilized to
model results in great challenges to the numerical solver and analyze four-terminal mechanically stacked tandem solar cells.
requires a long computing time before achieving the final When an optimized geometry is determined, the final result can
Authorized licensed use limited to: Nanyang Technological University Library. Downloaded on March 14,2024 at 15:23:29 UTC from IEEE Xplore. Restrictions apply.
1368 IEEE JOURNAL OF PHOTOVOLTAICS, VOL. 8, NO. 5, SEPTEMBER 2018

be checked by turning ON the complete device simulation and

any effects due to the tunnel diode can be identified.
We have employed this subcell analysis for an experimental
optimization of organic tandem solar cells. The organic tandem
stack consists of five organic subcells, each of which is based on
the system of tetraphenyldibenzoperiflanthen (DBP)/Fullerene
(C70), which is a well-investigated system in organic photo-
voltaics [50], [51]. It is almost impossible to optimize such
a tandem device by experimental efforts alone, even pick-
ing only five values for the thicknesses of each DBP and
C70 layer.
Based on the theoretical subcell analysis implemented in wx-
AMPS 3.0, we attained an optimized thickness combination as
verified by experiments shown in Fig. 8. With the thickness
optimization guided by the subcell analysis, the current match-
ing is significantly improved and the device efficiency has been
increased from 2.8% to 4.4%. More technique details (such as
optical generation rates, subcell thicknesses before/after opti-
mization) and modeling parameters of these organic tandem
cells have been described in another paper [51].
For better modeling organic solar cells, we have incorporated
additional physics relevant to organic materials. For example,
we have added exciton diffusion equation based on the work
reported in [52]. This organic solar cell model is packaged into
a dynamic-link library that can be called by wxAMPS 3.0 and
used for given subcell IVs as needed.

V. SUMMARY
wxAMPS 3.0 incorporating the nonlocal BTB tunneling
model and updating the numerical algorithm is shown to repro-
duce realistic behaviors of a variety of devices including Esaki
tunneling diodes and tandem solar cells. The simulation results
are compared with a commercial TCAD program and demon-
strate good consistency. A useful and highly demanded feature,
subcell analysis, allows quick diagnosis of current matching
and permits examination and optimization of individual compo-
nents of the multijunction efficiently. wxAMPS 3.0 provides a
freeware option for researchers to carry out fundamental studies
and to facilitate the experimental optimization for multijunction
solar cells.
Future work will further enhance the functionality of the code,
such as incorporating more physical mechanisms, like photon
recycling and upgrading the numerical algorithm to allow the use
of high-performance computing. wxAMPS 3.0 is freely shared
to the PV community, and the source code is also available ac-
cording to open-source license, which allows more researchers
to participate in the modeling development, with the possibility
of adapting the code into a versatile simulation platform for Fig. 8. (a) Simulated and experimental IV curves for a prototype or-
modern solar cell research. ganic tandem cell before optimization. Each subcell is fabricated of
MoO3 /DBP(10 nm)/C70(10 nm)/BCP(7 nm) and interconnected via a 0.5 nm
Ag recombination layer. Strong optical interference effects occur within these
ACKNOWLEDGMENT thin layers and lead to severe current mismatches between subcells. (b) Simu-
lated and experimental IV curves for the organic tandem cell with the optimized
The authors would like to acknowledge Dr. Z. Xiao from the thicknesses. The optimization is achieved by increasing the thicknesses of sub-
18th Institute of China Electronics Technology Group Corpora- cells #2 and #3 and decreasing the thicknesses of subcells #1 and #4. (c) Modeled
and experimental reflection rates before and after optimization. As shown, the
tion for advising the GUI design, and Prof. A. Rockett from the optical loss mainly arises in the top surface reflection, due to the absence of the
University of Illinois for carefully reading the manuscript. antireflective layer.

Authorized licensed use limited to: Nanyang Technological University Library. Downloaded on March 14,2024 at 15:23:29 UTC from IEEE Xplore. Restrictions apply.
LIU et al.: MODELING MULTIJUNCTION SOLAR CELLS BY NONLOCAL 1369

REFERENCES [28] P. Muralidharan, D. Vasileska, and S. M. Goodnick, “Advanced tunneling

models for solar cell applications,” in Proc. IEEE 39th Photovolt. Spec.
[1] F. Dimroth et al., “Four-junction wafer-bonded concentrator solar cells,” Conf., 2013, pp. 2113–2117.
IEEE J. Photovolt., vol. 6, no. 1, pp. 343–349, Jan. 2016. [29] Y. Liu, Y. Sun, and A. Rockett, “A new simulation software of solar
[2] W. Shockley and H. J. Queisser, “Detailed balance limit of efficiency of cells—wxAMPS,” Sol. Energy Mater. Sol. Cells, vol. 98, pp. 124–128,
p-n junction solar cells,” J. Appl. Phys., vol. 32, no. 3, pp. 510–519, 1961. Mar. 2012.
[3] M. A. Green et al., “Solar cell efficiency tables (version 49),” Prog. Pho- [30] Y. Liu, Y. Sun, W. Liu, and J. Yao, “Novel high-efficiency crystalline-
tovolt., Res. Appl., vol. 25, no. 1, pp. 3–13, Jan. 2017. silicon-based compound heterojunction solar cells: HCT (heterojunction
[4] K. Masuko et al., “Achievement of more than 25% conversion efficiency with compound thin-layer),” Phys. Chem. Chem. Phys., vol. 16, no. 29,
with crystalline silicon heterojunction solar cell,” IEEE J. Photovolt., p. 15400–15410, Jun. 2014.
vol. 4, no. 6, pp. 1433–1435, Nov. 2014. [31] T. Gokmen, O. Gunawan, and D. B. Mitzi, “Semi-empirical device model
[5] J. P. Connolly, D. Mencaraglia, C. Renard, and D. Bouchier, “Designing for Cu2ZnSn(S,Se)4 solar cells,” Appl. Phys. Lett., vol. 105, no. 3, 2014,
III-V multijunction solar cells on silicon,” Prog. Photovolt., Res. Appl., Art. no. 33903.
vol. 22, no. 7, pp. 810–820, Jul. 2014. [32] M. Chitambar, Z. Wang, Y. Liu, A. Rockett, and S. Maldonado, “Dye-
[6] C. D. Bailie et al., “Semi-transparent perovskite solar cells for tandems sensitized photocathodes: Efficient light-stimulated hole injection into
with silicon and CIGS,” Energy Environ. Sci., vol. 8, pp. 956–963, 2014. p-GaP under depletion conditions,” J. Amer. Chem. Soc., vol. 134, no. 25,
[7] J. H. Seo et al., “High efficiency inorganic/organic hybrid tandem solar pp. 10670–10681, Jun. 2012.
cells,” Adv. Mater., vol. 24, no. 33, pp. 4523–4527, Aug. 2012. [33] T. Minemoto and J. Julayhi, “Buffer-less Cu(In,Ga)Se2 solar cells by
[8] P. Loper et al., “Organic-inorganic halide perovskites: Perspectives for band offset control using novel transparent electrode,” Curr. Appl. Phys.,
silicon-based tandem solar cells,” IEEE J. Photovolt., vol. 4, no. 6, vol. 13, no. 1, pp. 103–106, Jan. 2013.
pp. 1545–1551, Nov. 2014. [34] Y. Wang et al., “Towards printed perovskite solar cells with cuprous oxide
[9] J. Werner et al., “Efficient monolithic perovskite/silicon tandem solar cell hole transporting layers: A theoretical design,” Semicond. Sci. Technol.,
with cell area >1 cm2,” J. Phys. Chem. Lett., vol. 7, no. 1, pp. 161–166, vol. 30, no. 5, May 2015, Art. no. 54004.
Jan. 2016. [35] X. Sheng et al., “Enhanced ultraviolet responses in thin-film InGaP
[10] S. P. Bremner, M. Y. Levy, and C. B. Honsberg, “Analysis of tandem solar solar cells by down-shifting,” Phys. Chem. Chem. Phys., vol. 15, no. 47,
cell efficiencies under AM1.5G spectrum using a rapid flux calculation pp. 20434–20437, Dec. 2013.
method,” Prog. Photovolt., Res. Appl., vol. 16, no. 3, pp. 225–233, May [36] A. Schenk, Advanced Physical Models for Silicon Device Simulation.
2008. New York, NY, USA: Springer, 1998.
[11] P. T. Chiu et al., “High performance 5J and 6J direct bonded (SBT) space [37] L. A. A. Pettersson, L. S. Roman, O. Inganäs, I. Introduction, and
solar cells,” in Proc. IEEE 42nd Photovolt. Spec. Conf., 2015, pp. 1–3. O. Inganäs, “Modeling photocurrent action spectra of photovoltaic devices
[12] M. Hermle, G. Létay, S. P. Philipps, and A. W. Bett, “Numerical simulation based on organic thin films,” J. Appl. Phys., vol. 86, no. 1, pp. 487–496,
of tunnel diodes for multi-junction solar cells,” Prog. Photovolt., Res. 1999.
Appl., vol. 16, no. 5, pp. 409–418, Aug. 2008. [38] E. O. Kane, “Theory of tunneling,” J. Appl. Phys., vol. 32, no. 1, pp. 83–91,
[13] M. Baudrit and C. Algora, “Modeling of GaInP/GaAs dual-junction solar 1961.
cells including tunnel junction,” in Proc. 33rd IEEE Photovolt. Spec. Conf., [39] G. A. M. Hurkx, D. B. M. Klaasen, and M. P. G. Knuvers, “New recom-
2008, pp. 1–5. bination model for device simulation including tunneling,” IEEE Trans.
[14] D. A. Clugston and P. A. Basore, “PC1D version 5: 32-bit solar cell mod- Electron Devices, vol. 39, no. 2, pp. 331–338, Feb. 1992.
eling on personal computers,” in Proc. Conf. Rec. 26th IEEE Photovolt. [40] M. K. Ieong, P. M. Solomon, S. E. Laux, H.-S. P. Wong, and D. Chi-
Spec. Conf., 1997, pp. 207–210. dambarrao, “Comparison of raised and Schottky source/drain MOSFETs
[15] M. Burgelman, K. Decock, S. Khelifi, and A. Abass, “Advanced electrical using a novel tunneling contact model,” in Proc. Int. Electron Devices
simulation of thin film solar cells,” Thin Solid Films, vol. 535, pp. 296–301, Meeting, 1998, pp. 733–736.
2013. [41] A. Schenk, “Rigorous theory and simplified model of the band-to-band
[16] R. Varache et al., “Investigation of selective junctions using a newly tunneling in silicon,” Solid State Electron., vol. 36, no. 1, pp. 19–34, 1993.
developed tunnel current model for solar cell applications,” Sol. Energy [42] S. Fonash, Solar Cell Device Physics, 2nd ed. New York, NY, USA:
Mater. Sol. Cells, vol. 141, pp. 14–23, 2015. Academic, 2010.
[17] B. Pieters, J. Krc, and M. Zeman, “Advanced numerical simulation tool [43] Synopsys, Sentaurus Device User Guide, 2009.
for solar cells - ASA5,” in Proc. IEEE 4th World Conf. Photovolt. Energy [44] Snowden, Semiconductor Device Modeling, 1989.
Conf., 2006, vol. 2, pp. 1513–1516. [45] W. H. Press, S. A. Teukolsky, W. T. Vetterling, and B. P. Flannery, Numer-
[18] M. Filipič et al., “Analysis of lateral transport through the inversion layer in ical Recipes in C: The Art of Scientific Computing, 2nd ed. Cambridge,
amorphous silicon/crystalline silicon heterojunction solar cells,” J. Appl. U.K.: Cambridge Univ. Press, 1992.
Phys., vol. 114, no. 7, Aug. 2013, Art. no. 74504. [46] Y. Liu, Y. Sun, and A. Rockett, “An improved algorithm for solving
[19] B. Ruhstaller et al., “Comprehensive simulation of light-emitting and equations for intra-band tunneling current in heterojunction solar cells,”
light-harvesting organic devices,” Proc. SPIE, vol. 7051, 2008, Art. Thin Solid Films, vol. 520, no. 15, pp. 4947–4950, May 2012.
no. 70510J. [47] D. Scharfetter and H. Gummel, “Large-signal analysis of a silicon read
[20] S. Selberherr, Analysis and Simulation of Semiconductor Devices. New diode oscillator,” IEEE Trans. Electron Devices, vol. ED-16, no. 1,
York, NY, USA: Springer, 1984. pp. 64–77, Jan. 1969.
[21] K. Yang, J. R. East, and G. I. Haddad, “Numerical modeling of abrupt [48] J. Lavery, “Quantum tunneling model of a p-n junction in Silvaco,” Thesis,
heterojunctions using a thermionic-field emission boundary condition,” Naval Postgraduate School, Monterey, CA, USA, 2008.
Solid State Electron., vol. 36, no. 3, pp. 321–330, 1993. [49] R. L. Mueller, “Spectral response measurements of two-terminal triple-
[22] J. Verschraegen and M. Burgelman, “Numerical modeling of intra-band junction a-Si solar cells,” Sol. Energy Mater. Sol. Cells, vol. 30, no. 1,
tunneling for heterojunction solar cells in SCAPS,” Thin Solid Films, pp. 37–45, Jun. 1993.
vol. 515, pp. 6276–6279, 2007. [50] B. R. Patil, Y. Liu, T. Qamar, H.-G. Rubahn, and M. Madsen, “4P-NPD
[23] M. Zeman, J. A. J. Willemen, L. L. A. L. Vosteen, G. Tao, and J. W. ultra-thin films as efficient exciton blocking layers in DBP/C7 0 based
Metselaar, “Computer modelling of current matching in a-Si: H/a-Si: H organic solar cells,” J. Phys. D, Appl. Phys., vol. 50, no. 38, 2017, Art.
tandem solar cells on textured TCO substrates,” Sol. Energy Mater. Sol. no. 385101.
Cells, vol. 46, no. 2, pp. 81–99, May 1997. [51] M. Ahmadpour, Y. Liu, H.-G. Rubahn, and M. Madsen, “Current matching
[24] L. Esaki, “New phenomenon in narrow germanium p-n junctions,” Phys. in multifold DBP/C70 organic solar cells with open-circuit voltages of up
Rev., vol. 109, no. 2, pp. 603–604, Jan. 1958. to 6.44 V,” IEEE J. Photovolt., vol. 7, no. 5, pp. 1319–1323, Sep. 2017.
[25] J. Y. Hou, J. K. Arch, S. J. Fonash, S. Wiedeman, and M. Bennett, “An [52] Y. Liu et al., “Role of the charge-transfer state in reduced langevin re-
examination of the ‘tunnel junctions’ in triple junction a-Si:H based solar combination in organic solar cells: A theoretical study,” J. Phys. Chem.
cells: Modeling and effects on performance,” in Proc. Conf. Rec. 22nd C, vol. 119, no. 47, pp. 26588–26597, Nov. 2015.
IEEE Photovolt. Spec. Conf., 1991, pp. 1260–1264.
[26] M. Filipič et al., “CH3NH3PbI3 perovskite/silicon tandem solar cells:
Characterization based optical simulations,” Opt. Express, vol. 23, no. 7,
pp. A263–A278, Apr. 2015.
[27] Z. Q. Li, Y. G. Xiao, and Z. M. S. Li, “Modeling of multi-junction solar
cells by crosslight APSYS,” Proc. SPIE, vol. 6339, 2006, Art. no. 633909. Authors’ biographies not available at the time of publication.

Authorized licensed use limited to: Nanyang Technological University Library. Downloaded on March 14,2024 at 15:23:29 UTC from IEEE Xplore. Restrictions apply.