This article has been accepted for inclusion in a future issue of this journal.

Content is final as presented, with the exception of pagination.

IEEE JOURNAL OF PHOTOVOLTAICS 1

Numerical Simulation of Silicon Heterojunction

Solar Cells Featuring Metal Oxides as
Carrier-Selective Contacts
Christoph Messmer, Martin Bivour, Jonas Schön, Stefan W. Glunz, and Martin Hermle

Abstract—The applicability of different high (low) work func- the Shockley equations
tion contact materials for the formation of alternative passivating
and hole (electron) selective contacts is currently re-explored for J = J gen − J 0 {exp [qV /(n kB T )] − 1} (1)
silicon solar cells. To assist the engineering of those contacts, which
is still in its infancy, numerical device simulations are used to im- where q is the elementary charge, n the ideality factor, and kB
prove knowledge regarding relevant heterojunction and thin film the Boltzmann constant. One important assumption for (1) is
properties with the focus on metal oxide based hole contacts. The that the voltage (V), which corresponds to the quasi-Fermi level
importance of 1) a high metal oxide work function for the induced splitting within the absorber at the edge of the pn-junction,
c-Si pn-junction is shown. It is elucidated that for an efficient hole equals the external voltage (Vexternal ) at the contacts. The
transport from this induced c-Si junction into the external elec-
trode, via the buffer and the metal oxide, 2) the metal oxide’s quasi-Fermi level splitting can also be expressed by the im-
conduction band must be below the valence band of the buffer (or plied voltage (Vim plied ). The assumption Vexternal = Vim plied
c-Si absorber) for direct band-to-band tunneling, or 3) bulk traps is typically fulfilled for a negligible gradient in the majority
near the valence band edge of the buffer (or c-Si absorber) are carrier quasi-Fermi level in the contact region. This means that
needed for trap-assisted tunneling. low-injection conditions in the contact region during operation
Index Terms—Amorphous silicon, band-to-band (B2B) tunnel- must be maintained [2], i.e., that the majority carrier density re-
ing, fundamentals, heterojunctions, inversion-layer, simulation, se- mains well above the minority carrier density (nm a j  nm in ).
lectivity, trap-assisted tunneling (TAT). Or, in a more general expression, that the conductivity σ of
majority carrier remains well above those of the minority car-
I. INTRODUCTION riers (σm a j = q μm a j nm a j  σm in = q μm in nm in , μ is the
mobility) [3]. Basically, we define contacts to be “selective” for
ASSIVATING and carrier-selective contacts [1] have the
P potential to overcome some of the intrinsic efficiency limi-
tations of the established homojunction contact schemes. For the
which Vexternal = Vim plied and hence the validity of the diode
equation is fulfilled. Accordingly, we define contacts to be “non-
selective” when having a significant voltage loss in the contact
latter, doping of the silicon surface region is applied to obtain region during operation leading to Vexternal < Vim plied .
carrier-selective device regions below the unpassivated metal Similar to classical homojunctions, for TOPCon [4] and poly-
electrodes. The high recombination in this region is linked to a Si [5] related contacts obtaining carrier selectivity is typically
high dark saturation current J0 and, therefore, results in mod- not the main design constraint. This is explained by the high
erate voltage at maximum power point (MPP) and open-circuit doping efficiency of the partly crystalline silicon thin films and
(OC) conditions. Both passivating and nonpassivating carrier- the c-Si wafer surface. It ensures that the majority carrier density
selective contacts typically obey the diode model described by in the contact region during equilibrium (nm a j,0 ) is already
much higher than the excess carrier density that will be added
Manuscript received October 13, 2017; revised December 13, 2017; accepted during operation (Δn = Δp). Accordingly
January 9, 2018. This work was supported in part by the European Project “Disc”
under the European Union’s Horizon 2020 Research and Innovation Program nm a j = nm a j,0 + Δp  nm in = nm in,0 + Δn
under Grant agreement N°727529 and in part by the German Federal Min-
istry for Economic Affairs and Energy under contract N°0324141 “Selektiv.” with Δn  nm in,0 , is typically fulfilled. For TOPCon and poly-
(Corresponding Author: Christoph Messmer.) Si related contacts, an intermediate passivating dielectric buffer
C. Messmer, M. Bivour, and M. Hermle are with the Fraunhofer Institute for (SiOx ) ensures that those contacts are not only selective but also
Solar Energy Systems, Freiburg 79110, Germany (e-mail: christoph.messmer@
ise.fraunhofer.de; [email protected]; martin.hermle@ise. passivating. Accordingly, J0 is low and the device characteristic
fraunhofer.de). and the limiting losses are well described by the diode equation.
J. Schön and S. W. Glunz are with the Fraunhofer Institute for Solar Energy The application of different high (low) work function thin
Systems, Freiburg 79110, Germany, and also with the Department of Sustain-
able Systems Engineering (INATECH), Albert-Ludwigs-Universität Freiburg, films for the formation of passivating and hole (electron) selec-
Freiburg im Breisgau 79110, Germany (e-mail: [email protected]. tive contacts is another currently explored approach for silicon
de; [email protected]). solar cells. So far, most promising candidates are transition
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. metal oxide (TMO) based heterojunctions leading to efficien-
Digital Object Identifier 10.1109/JPHOTOV.2018.2793762 cies up to 22% [6]–[10]. However, other materials are also under
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.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

2 IEEE JOURNAL OF PHOTOVOLTAICS

investigation [11]. Work in this field can be understood as a re-

naissance of the induced silicon junctions (e.g., metal-insulator-
semiconductor) from the 1970s [12] by taking advantage of 1)
novel or better studied high (low) work function contact mate-
rials mainly adapted from organic electronics [13] and 2) im-
provements of the passivating intermediate buffer layers (a-Si:H,
ultrathin dielectrics). However, knowledge regarding relevant
junction and material properties and their engineering, stability
issues, and the integration of those high (low) work function
contact materials in the c-Si solar cell device architectures is
still in its infancy. Basically, engineering toward sufficient se-
lectivity seems to be of more importance compared with devices
featuring n- and p-type doped carrier-selective silicon regions.
Within this work the operation principles, important design
parameters, and losses of contact schemes based on induced
junctions are reviewed with the help of numerical device simu-
lations. The importance of the induced c-Si dark band bending,
the voltage loss in the contact region, the band line-up, and an
efficient tunneling transport are addressed.

II. SIMULATION METHOD

All simulations in this paper are carried out using Sentau-
rus device that allows the implementation of various models
necessary for modelling hetero devices [14].

A. Simulation Setup
Since the investigated devices have no lateral distinction sim-
ulations are carried out in (quasi-)1D. The silicon bulk is n-type
and the electron contact on the rear side is idealized and the same
for all structures featuring a layer of intrinsic amorphous silicon
a-Si(i):H and n-type a-Si(n) (parameters listed in the Appendix,
Table II). However, the focus will be on the p-contact at the front
side featuring a passivating buffer layer and a high work function
wide band gap metal oxide. We will approach this final contact
system step-by-step by analyzing the following structures:
a) The simplest contact is a metal–semiconductor (MS) con-
tact with an induced pn-junction described by the Schottky
model [14]. The equilibrium band diagram of such a con-
tact is shown in Fig. 1(a) and the highly hole populated and
inverted c-Si surface can be seen. There are two important
parameters to set for this simple contact: Fig. 1. Band diagrams of the three investigated p-contacts in the dark: (a)
i) The work function of the metal-like layer Φm etal MS—Metal-Silicon; (b) MIS—with additional intrinsic amorphous Si-layer;
(c) T-SHJ—Transition metal oxide Silicon hetero junction, B2B path (case C);
determining the built-in potential of the c-Si part of (d) T-SHJ for a lower work function with TAT. (case A, A∗ ) Four different TMO
the induced junction. The latter corresponds to the trap distributions (α-δ) are schematically shown.
band bending ϕcSi in the crystalline silicon. Note
that for this simulation study we do not speak of
vacuum work functions as listed in the literature but faces so that S0m etal is much higher. However, this
always of an “effective work function” after contact- MS contact with “artificially” lowered S0m etal is still
ing, which takes Fermi-level pinning into account. useful to distinguish between phenomena arising
Still, all the simulation results can be compared with from this simple induced junction and those from
experimental data by using ϕcSi as a figure of merit the more complex structures (b) and (c).
that can be easily measured by SPV measurements. b) The next step to approach a more realistic contact is to
ii) The surface recombination velocity S0m etal that in add a passivating buffer layer—in our case 5 nm thick
this case is set to 10 cm/s mimicking a good chem- undoped amorphous silicon. We get an MIS-like structure
ical surface passivation of the metal/c-Si. Note that [see Fig. 1(b)] where the buffer decouples S0m etal from
MS contacts in reality form highly defective inter- the junction recombination. S0m etal is set to 106 cm/s for
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

MESSMER et al.: NUMERICAL SIMULATION OF SILICON HETEROJUNCTION SOLAR CELLS FEATURING METAL OXIDES 3

all the following structures. As can be seen, the induced energy level lies at E = ΦTM O + Ecut . (Note: The over-
band bending is shared by the buffer and the c-Si surface. all trap density is, therefore, getting smaller for higher
Hence, ϕcSi is reduced to some extent but the c-Si is still Ecut , whereas the trap density per energy is still equal to
highly inverted. We get a heterojunction between a-Si and that of (α)).
c-Si which usually has small defect concentrations at the (δ) Some further investigations are carried out by a single
interface, thus, we set the surface recombination velocity defect level. This level has an energy ELevel with respect
S0cSi to zero, unless stated otherwise. to the conduction band of the TMO and a density of
−3
c) Finally, we add a TMO like 10 nm thick layer with a high ρTM O 20
Level = 10 cm .
band gap of 3 eV and a variable work function ΦTM O
and bulk defect density [see Fig. 1(c)]. For simplicity and III. SIMULATION RESULTS
referring to the standard silicon hetero junction (SHJ) fea-
In this section, we aim to separate the different basic require-
turing p-type a-Si instead of the high work function TMO,
ments needed to build a hole-selective contact step-by-step. In
we will call this hole contact TMO-based SHJ (T-SHJ).
the first section, we will present a simple showcase to discuss the
The TMO/a-Si:H contact adds a second heterojunction
importance of the induced c-Si junction and to clarify what we
and in principle two tunneling paths that are essential for
mean by nonideal selectivity. We will show how it can be quan-
the contact to work: Path (I) is direct band-to-band (B2B)
tified and distinguished from other phenomena like classical
tunneling from the valence band of the buffer to the con-
losses as described by the diode model where ideal selectivity
duction band of the TMO [see Fig. 1(c)]. For path (II) traps
is linked to a proper rectifying, i.e., diode-like contact charac-
in the band gap of the TMO provide an additional trans-
teristic. Section III-B will compare the influence of the work
port path, i.e., trap-assisted tunneling [TAT, see Fig. 1(d)]
function on the structures presented in Section II-A, whereas
[14]. How we modeled the defect traps within the TMO
Section III-C will investigate the influence of the tunneling
is described in the next section.
transport for the T-SHJ structure.

B. Modeling of the Traps Within the TMO A. Importance of Induced c-Si Junction: Nonideal Selectivity
As will be shown the density and the energetic distribution Let us first consider the three different T-SHJ hole contacts
of the traps inside the TMO band gap greatly affect the TAT listed in Table I. Contact A has a sufficiently high work function
current. First some general notes on how the traps are modeled (Φ = 5.3 eV) exceeding the valance band energy of the c-Si
in this paper. absorber (5.19 eV) and no c-Si surface recombination (S0cSi =
The TMO has a thickness of 10 nm and is n-type doped in 0) and is, therefore, solely limited by Auger recombination
a way that its work function equals the TMO conduction band (ideal case). Furthermore, we constructed two cases A∗ and
energy ΦTM O = ECTM O as shown in Fig. 1(d) (n-doping con- B in a way that contact A∗ has the same high work function
centration: 1020 cm−3 ). The traps are implemented electrically but no surface passivation (i.e., high S0cSi )—Contact B is well-
neutrally (that means we put half donor traps and half accep- passivated (S0cSi = 0) but its work function is not sufficiently
tor traps in the TMO) in order to just influence the tunneling high (Φ = 5.0 eV). (For all of these three contacts tunneling is
current. Furthermore, the traps are in all cases implemented ideal, i.e., distribution (α) was used.)
spatially uniform throughout the TMO. If we look at the JVlight -curves (Fig. 2, solid line) for an
In this paper, we will consider four different distributions experimentally typical range and the external cell parameters
(α)–(δ) of the traps that we try to visualize in Fig. 1(d) and (see Table I) we notice no significant distinction of the two
describe in the following. curves of contact A∗ (blue) and B (red) although their power
(α) The energetic distribution of the traps is uniform through- losses arise from different phenomena. For a better loss analysis
out the whole bandgap with a very high density of we want to discuss three quantities, namely the implied OC
−3
ρTM O 20
Traps = 10 cm (which is equal to the n-doping con- voltage (iVo c ), the difference of the iVo c , and external Vo c (ΔV )
centration). Therefore, we will call this case “metallic” and the inversion of the c-Si surface during equilibrium that
or “metal-like” since the defect distribution within the is quantified by the induced c-Si band bending (ϕcSi ). These
TMO provides energy states through the whole band gap. “internal” cell parameters, given in Table I, are depicted in a
In this case, we can assume that the TAT is not limiting band diagram in Fig. 3(a) and can be extracted experimentally
the current of the cell. with little effort [15].
(β) This case is like (α), i.e., we still have a uniform en- Starting with ϕcSi : For contact A∗ the band bending ϕcSi
ergy distribution of the traps, but we lower the trap den- is the same as for contact A since the work function is the
sity ρTM O
Traps by orders of magnitude down to values like same (see Table I). However, the work function of contact B
−3
11
10 cm . is too low to form a proper pn-junction that results in a ϕcSi ,
(γ) To investigate the influence of the traps close to the con- which is 169 meV lower. Basically, ϕcSi increases the local hole
duction band of the TMO, starting from (α) we define density in the contact region with respect to the c-Si bulk as
additional cutoff energy Ecut [depicted in Fig. 1(d)]. Be- shown in Fig. 3(b) and (c) (pEQ , solid line) during equilibrium.
tween the conduction band of the TMO-like layer and this Accordingly, the asymmetry of the hole/electron conductivity
energy all traps are omitted, therefore, the first available in this device region is also increased that is the decisive pa-
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

4 IEEE JOURNAL OF PHOTOVOLTAICS

TABLE I
THREE DIFFERENT HOLE CONTACTS WITH THEIR EXTERNAL AND INTERNAL
CHARACTERISTICS: (A) INTRINSIC AUGER LIMIT (A∗ ) LIMITED BY HIGH
SURFACE RECOMBINATION Sc0 S i (B) LIMITED BY WORK FUNCTION Φ.

Fig. 3. (a) Band diagram of contact A∗ and B (see Table I) at OC conditions.

(b, c) Hole and electron densities at OC (p o c , n o c ) and hole density at equilib-
rium (p E Q ) for contact A∗ (blue) and B (red).

Fig. 2. JV-curves for the contacts listed in Table I. Solid lines show the charac-
teristics for one sun on an n-type bulk, dotted lines show the dark characteristics Fermi-level in the contact region and is significant for contact B
on a p-type bulk. (ΔV  0), whereas for contact A∗ it is negligible. The origin
of this voltage loss ΔV for contact B is due to p ∼ n [see
Fig. 3(c)] in the contact region that lowers the hole conductivity
rameter for obtaining the selective extraction of excess holes σh . According to jh = σh grad[EF,p ]/q [3], [16], to drive a
from the absorber [3], [16], i.e., a proper rectifying and diode- certain hole current density jh through the p-contact, a reduction
like carrier transport. Although ϕcSi describes this asymmetry of the hole conductivity σh has to come along with a higher
only in the dark it is directly related to the asymmetry dur- gradient in the hole quasi-Fermi level EF,p , i.e., a voltage loss
ing operation when excess electrons and holes are generated ΔV . In our example, this voltage loss of contact B is split
by the same amount (p/n = (p0 + Δp)/(n0 + Δn)) [3]. Ac- up into ΔV aSi = 67 mV within the amorphous silicon [visible
cordingly, the influence of ϕcSi can also be seen during oper- fermi level gradient in Fig. 3(a)] and ΔV IF = 66 mV at the
ation at MPP and OC conditions. For the latter this is shown TMO/a-Si interface (not directly visible).
in Fig. 3(b) and (c). For the high work function of contact A∗ Since voltage losses ΔV arise from nonideal selectivity, we
[see Fig. 3(b)] this asymmetry is maintained during operation will use ΔV as a figure of merit for selectivity losses. However,
since po c (dashed lines) is well above no c (dotted lines) in the it should be noted that besides ΔV another experimental method
contact region. For the low work function of contact B, this to probe insufficient selectivity of a contact is to measure the
asymmetry cannot be maintained during operation, i.e., p ∼ n dark JV-curve for the contact put on an absorber with inverse
in the contact region [see Fig. 3(c)]. The latter is accompa- doping (p-type) and a low resistive ohmic rear contact. In this
nied by a significant gradient in the majority carrier Fermi-level case, an identical and low-ohmic characteristic is obtained for
and voltage loss (ΔV) in the contact region as highlighted in high work function contacts A and A∗ , which is mainly limited
gray in Fig. 3(a). It violates one important boundary condition by the absorber resistivity (see Fig. 2, dotted JVdark ). For con-
of the standard pn-junction theory [2]. Accordingly, such con- tact B a poor hole selectivity is obtained for the p-type absorber,
tacts and their losses are not properly described by standard reflected in a diode-like characteristic.
diode equation. We conclude: While contact A∗ is limited due to classical re-
iVo c : Owing to the lack of chemical passivation the iVo c of combination losses, contact B is limited by an insufficient work
contact A∗ shows a significant drop to 578 mV, whereas the function Φ and hence insufficient asymmetry of the conductivi-
iVo c of contact B is above 700 mV. This difference can be ties leading to a selectivity loss ΔV in the contact region during
directly seen in the band diagram at OC where the iVo c equals operation. It should be noted that similar conditions are also
the splitting of the hole and electron quasi-Fermi levels in the obtained at MPP.
bulk of the absorber (see Fig. 3(a), right dashed lines). More generally, we have seen that a high iVo c is linked to
ΔVo c = iVo c − Vo c : This difference between internal and a good c-Si surface passivation and a low ΔV indicates ideal
external voltage results from the gradient in the majority carrier selectivity. For the latter, a high work function Φ is needed to get
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

MESSMER et al.: NUMERICAL SIMULATION OF SILICON HETEROJUNCTION SOLAR CELLS FEATURING METAL OXIDES 5

of the c-Si mid-gap energy the efficiency drops and then de-
creases linearly until it reaches zero. Therefore, a sufficiently
high (low) work function is the first condition for building a good
hole (electron) selective contact since it modifies the asymmetry
of the conductivity as motived before.
To analyze where these power losses come from, we will have
a closer look at the OC voltage of the cell, as shown in Fig. 4(b).
For the hole contact with a work function above the valence band
energy of c-Si (Φ > EVcSi ) the curves of iVo c (dotted blue line)
and the Vo c (solid blue line) match, therefore, ΔV = 0 which
indicates an ideal selectivity of the contact. In this range, the Vo c
is solely limited by the intrinsic limit (Auger recombination)
comparable to the case of contact A in Section III-A.
If the work function is lower than the valence band energy
of c-Si (Φ < EVcSi ), the Vo c drops while the iVo c stays above
728 mV. The latter indicates that the mimicked “passivation”
at the metal/c-Si interface is very good (S0m etal = 10 cm/s).
However, the external voltage Vo c drops for decreasing work
functions in a linear fashion [17]–[19]. A similar behavior is
obtained for fill factor FF and jsc (not shown). The difference be-
tween iVo c and Vo c per definition leads to ΔV > 0 and indicates
that the losses in voltage (and efficiency) are due to nonideal
selectivity of the contact (similar to contact B in Section III-A).
2) Metal-iLayer-Silicon: When adding 5 nm of intrinsic
amorphous silicon between the metal and the c-Si bulk, which
for real devices is needed for a good surface passivation, we get
the MIS-like structure shown in Fig. 1(b) and the characteristic
shown in Fig. 4, green.
Both the efficiency η and the OC voltages Vo c and iVo c show
the same characteristic as for the MS structure, except that they
are shifted by 0.15 eV to higher work functions. Whereas the
MS curves were symmetric around the mid-gap energy of c-Si,
the green curves are centered about the mid-gap energy of a-Si
Fig. 4. (a) Efficiency η, (b) V o c and iV o c as a function of the metal/TMO
EmaSi
idgap ≈ 4.78 eV. This raises the demand on the TMO work
work function Φ for the three different p-contact systems of Fig. 1. For MS and function Φ in order to reach a high c-Si band bending ϕcSi ,
MIS, the light-colored curves show the mirrored n-contact. The p-contact for but still a work function Φ slightly greater than the c-Si valence
T-SHJ is shown for two different surface passivations Sc0 S i (orange and pink)
and the three cases A, A∗ , and B (of Section III-A) are highlighted. band energy EVcSi is sufficient to reach a high ϕcSi and, therefore,
maximum efficiency.
Like for the MS structure the Vo c around the mid-gap energy
a high c-Si band bending ϕcSi and, thus, to maintain the asym- shows the well-known linear behavior [17], [20] with a slope of
metric p/n conductivity in the contact region during operation. 1 V/eV, which indicates that for this range the work function Φ
shows a one-to-one relation with respect to the limiting built-in
voltage of the contact and, therefore, the c-Si band-bending ϕcSi
B. Influence of the Work Function
[3], [19].
This section discusses the influence of the work function for 3) TMO Silicon Hetero Junction: In this last step, we add
the p-contacts introduced in Section II (see Fig. 1) step-by-step. the high-band gap, n-type TMO layer with a high trap density
1) Metal-Silicon: The results for the MS interface are shown inside its band gap and hence another heterojunction to obtain
in Fig. 4 (in blue). The dark blue curve in Fig. 4(a) shows the the TMO/a-Si:H(i)/c-Si p-contact introduced in Fig. 1(c) and (d).
efficiency for the hole contact introduced in Fig. 1(a) (p-contact Let us first consider the results shown in orange (see Fig. 4).
on an n-type c-Si bulk) while the light blue curve shows the We see that the T-SHJ contact in this case shows a similar
efficiency for an electron contact (n-contact on a p-type c-Si behavior with respect to efficiency and voltage (also FF and
bulk), both as a function of the work function Φ of the metal. jsc ) as the MIS-like structure, despite its higher complexity due
The two curves are symmetric around the mid gap energy to tunneling at the TMO/a-Si interface. The reason for this is
of c-Si Em idgap ≈ 4.63 eV. For the hole (electron) contact, the
cSi
that we modeled the TMO in a way that it behaves “metal-
efficiency reaches a maximum of η = 28% for a work function like” (distribution (α) in Section II-B). This means that the hole
Φ greater (smaller) than the valence (conduction) band energy extraction from the inverted c-Si region toward the external
EVcSi ≈ 5.19 eV (ECcSi ≈ 4.07 eV). Getting closer to the region electrode is based on an efficient tunneling via the TMO gap
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

6 IEEE JOURNAL OF PHOTOVOLTAICS

Fig. 5. (a) OC voltage V o c , Additionally shows the c-Si band-bending ϕ c S i and iV o c (same for all curves) and selectivity losses ΔV Φ , ΔV TAT due to
insufficient work function (blue arrow) and insufficient TAT tunneling (red arrow), respectively. (b) Short-circuit current jsc , Additionally shows the TAT current
j TAT (dash-dotted). (c) FF; Additionally shows the cutoff energies in gray. (d) efficiency η as a function of the TMO work function Φ for different trap distributions
(α–γ introduced in Section II-B); highlights the cases A [see Fig. 1(d) and Section III-A) and B (see Section III-A) and C [see Fig. 1(c)]. The MIS-like structure
and the “metal-like” T-SHJ of Fig. 4 are plotted as references (in green and light orange). (β) Shows variation of the uniformly distributed trap density as solid
lines in different shades of orange. (γ) Shows variation of Ec u t as dotted lines in orange.

states. The influence of the distribution of these gap states will characteristics for different trap concentrations are shown as a
be discussed in the next section. function of ΦTM O .
For the sake of completeness, Fig. 4 shows a T-SHJ contact Let us first concentrate on the efficiency η [see Fig. 5(d)]. The
with poor chemical surface passivation (S0cSi = 5.3 · 106 cm/s, MIS-like structure (green), the T-SHJ structure of Fig. 4 (light
pink) resulting in a lower iVo c and hence lower efficiency. How- orange, solid line), and the former discussed cases A, B, and C
ever, for higher work functions ΔV = 0 indicates ideal selectiv- [see Fig. 1(c)] are plotted as references reflecting only the power
ity despite the poor chemical passivation. The iVo c stays above losses caused by an insufficient work function. The other data
531 mV for work functions around the mid-gap of a-Si while provide information on the TMO properties namely the traps
the Vo c drops further due to insufficient c-Si band-bending ϕcSi . and their distribution. An important TMO/a-Si heterojunction
Note that the three p-contacts A, A∗ , and B (discussed in property is the position of the TMO conduction band with re-
Section III–A) are highlighted in Fig. 4 to relate back to spect to the valence band of the a-Si:H buffer that we define as
Section III-A and emphasize once more the difference between EBarrier = EVaSi − ECTM O [see Fig. 1(d)]. We plot this as a sec-
power losses due to an insufficient work function (case B) and ond x-axis in Fig. 5. It should be highlighted again that for sim-
power losses due to a poor chemical passivation (case A∗ ). plicity reasons ECTM O ≡ ΦTM O for the TMO is assumed here.
We conclude that the T-SHJ structure behaves like a MIS- All curves match for EBarrier < 0 (ECTM O > EVaSi ). This is
like structure as long as the TMO electrically behaves like a because a TMO conduction band energy greater than the a-Si
metal. However, it is important to understand what happens if valance band energy allows an efficient transport at the TMO/a-
the tunneling at the TMO interface has a limiting influence. The Si:H contact by B2B tunneling independent of the TMO traps
next section will discuss some cases. [see Fig. 1(c)].
However, if EBarrier > 0 (ECTM O < EVaSi ) the TMO conduc-
C. Transport Limitations at the TMO/a-Si Interface tion band energy is lower than the valence band energy of the
a-Si layer [shown in Fig. 1(d)], an efficient TAT is needed. Thus,
In the last section, we assumed the transport via tunneling the efficiency crucially depends on the TMO traps. When low-
at the TMO/a-Si interface to be nonlimiting. However, for real ering the density ρTM O
Traps of the homogeneously distributed traps
cells there might be complex and nonmetal-like distributions of we see that the efficiency decreases [see β in Fig. 5(d), orange
the defect traps that are subject of this section. In Fig. 5, the cell to black lines]. This effect is less pronounced for higher work
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

MESSMER et al.: NUMERICAL SIMULATION OF SILICON HETEROJUNCTION SOLAR CELLS FEATURING METAL OXIDES 7

functions Φ. Accordingly, lower trap densities come along with

higher demands on the work function. Assuming an ideal TAT
transport a work function larger than 5.2 eV is needed to reach
the maximum efficiency.
The influence of the energetic trap distribution within the
TMO band gap [see γ in Fig. 1(d)] is shown by different dotted
orange lines in Fig. 5(d): There is a general loss in efficiency, the
more we cut off energy levels close to the conduction band of the
TMO. The FF limitation for the nonmetal-like cases is depicted
in Fig. 5(c). For decreasing trap densities, we get an overall
drop of the FF except for EBarrier < 0 (Φ > EVaSi ) where B2B
tunneling is efficient. The dotted FF curves for increasing cutoff
energies Ecut match the metal-like case (solid orange line) for
low work functions. But as the work function increases the FF
reaches its maximum and drops again for higher work functions.
To understand the influence of Ecut we have to understand
the contribution of different energetic levels within the TMO
bandgap. Therefore, we placed a single trap level at energy
ELevel and varied its energetic position inside the TMO band
gap [variation δ in Fig. 1(d)]. Let us consider the green curve in
Fig. 6(a) that shows the FF of the T-SHJ structure as a function
of ELevel for a TMO work function and conduction band energy
of Φ = 5 eV and the corresponding equilibrium band diagram.
One can see that the FF is poor (<20%) and, therefore, transport
is inefficient for ELevel between zero and EBarrier , which means
for a trap between the conduction band energy of the TMO and
the valance band energy of the a-Si:H buffer. For a trap level
ELevel slightly larger than EBarrier we get efficient transport
since holes in the a-Si valence band are now able to reach the
TMO defect level. However, the further we increase ELevel
the fewer holes can reach the defect level and contribute to
the transport. This is explained by their thermal energy that is
not sufficient. For even higher ELevel , e.g., the TMO mid gap
energy, no significant transport is observed. This shows that for Fig. 6. Fill factor (FF) as a function of the energy EL e ve l of a single trap
an efficient TAT transport, the traps close to the a-Si valence band level within the TMO bandgap (variation δ in Fig. 1) for (a) Φ = 5 eV and (b)
Φ = 5.3 eV and corresponding equilibrium band diagram. The green curves cor-
are of most importance. (Note that, we only considered elastic respond to the FF of the T-SHJ structure, whereas the yellow curves correspond
TAT which does not include phonon interactions [21]. However, to the T-SHJ structure without a-Si(i) passivation layer for which the TMO is
for phonon energies greater than zero, the FF onset would be in direct contact with the c-Si. For the latter the equilibrium band diagrams are
not shown.
smoother including energies between EVaSi and ECTM O .)
However, considering the case that the a-Si buffer layer is
omitted [yellow curve in Fig. 6(a), band diagram not shown] efficient transport. This shows that the actual band alignment
the traps close the c-Si valence band are of most importance defines the requirements on the traps inside the TMO and if
and efficient tunneling is shifted by 0.45 eV to lower ELevel . traps are needed at all.
This shift corresponds to the a-Si/c-Si valence band offset. This Going back to Fig. 5(c), this finding helps to understand the
suggests that without buffer, traps closer to the TMO conduction behavior of the FF when varying the trap distribution near the
band and with buffer traps further way from ECTM O are needed TMO conduction band (γ, dotted lines). For Ebarrier > Ecut the
for an efficient TAT. The FF for around mid-gap energies is dotted curves match the metal-like TMO (solid orange curve).
hardly affected since for these energetic locations it is the mean This is because it makes no difference if we cut off traps with en-
thermal energy of the charge carriers that is decisive for the ergy E < Ebarrier , since they do not contribute to the transport.
transport (and is unchanged). However, Fig. 6(b) shows that However, if we increase the TMO work function ΦTM O and,
omitting the buffer can further change the requirements on the therefore, decrease Ebarrier to Ebarrier < Ecut , the omitted en-
provided tunneling paths: The green curve in Fig. 6(b) shows the ergy levels move into the energy region close to the a-Si valence
T-SHJ with a-Si for a higher TMO work function of Φ = 5.3 eV band energy. This leads to a FF loss. The further we increase
compared with Fig. 6(a). This lowers EBarrier by ΔΦ = 0.3 eV. Φ (i.e., decrease Ebarrier ), the fewer holes thermally reach the
It means that traps closer to the TMO conduction band are available energy levels at E > Ecut . However, for Ebarrier < 0,
needed for a high FF. Without buffer (yellow curve) EBarrier B2B tunneling sets in leading to a FF that is independent of the
becomes negative for the higher work function of the TMO traps. Accordingly, a sharp FF increase is observed for ΦTM O
that allows for B2B tunneling and traps are not needed for an (i.e., ECTM O ) close to the a-Si valence band energy of 5.62 eV.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

8 IEEE JOURNAL OF PHOTOVOLTAICS

1) Selectivity losses ΔVΦ due to insufficient work function:

As already observed for the simple induced MS junction,
an insufficient work function leads to an insufficient c-Si
band bending that violates the asymmetric conductivities
in the contact region. This leads to selectivity losses ΔVΦ
like shown in Fig. 7 (blue arrow) and highlights once
more that a sufficiently high band bending is an essential
requirement for good selectivity. Such a one-to-one rela-
tion of ΔVΦ on ϕcSi (blue shaded area with slope −1)
is due to the Schottky equation and has already been ob-
served experimentally for metal [19], doped a-Si:H [15],
[22] and metal oxide based contacts [22]–[25].
2) Selectivity losses ΔVTAT due to insufficient tunneling:
A major finding of this paper is that inefficient TAT
tunneling leads to additional selectivity losses ΔVTAT
(exemplarily shown by the red arrow in Fig. 7). Therefore,
selectivity problems (i.e., ΔV > 0) do not only arise
from insufficient band bending (i.e., work function),
Fig. 7. Selectivity loss ΔV o c = iV o c − V o c as a function of the c-Si band as described in 1), but can also arise from insufficient
bending ϕ c S i . The blue shaded area with slope −1 corresponds to selectivity
losses ΔV Φ only due to insufficient work function (exemplarily shown as blue tunneling transport due to the absence of defects. This
arrow), whereas the red shaded area with a slope smaller than −1 shows ad- should be taken into account when experimental data are
ditional selectivity losses ΔV TAT due to insufficient TAT tunneling transport evaluated, since it might give further insight in the origin
(exemplarily shown as red arrow).
of selectivity losses, especially when the slope of Fig. 7
deviates from −1 (for, e.g., for the red shaded area).
Fig. 5(b) shows that the short-circuit conditions are least
Whether or not insufficient tunneling plays a decisive role
affected by an inefficient hole transport. High short-circuit
in an actual device is subject to further investigations. A first
currents jsc are obtained while for OC and MPP there are already
comparison with experiments was drawn in [22], [26], leading
significant losses. The graph also highlights that for ECTM O <
to promising agreement of the simulations of this paper and
EVaSi the total current is driven only by TAT (jsc = jTAT )
experimental data both qualitatively and quantitatively.
while for ECTM O > EVaSi it is driven by two parallel currents
jsc = jTAT + jB 2B corresponding to the paths of TAT and B2B.
Finally, we address the OC conditions in Fig. 5(a) that shows IV. CONCLUSION
Vo c , iVo c , and c-Si band bending ϕcSi during equilibrium. First A fundamental understanding of carrier-selective contacts
of all the c-Si band bending ϕcSi is the same for all curves featuring metal oxides was generated. This supports the search
(black dotted curve) reflecting that the TMO trap distributions for suitable materials and improves the interpretation of ex-
(α)-(γ) have no influence on ϕcSi . Furthermore, the iVo c perimental data. This is especially true when using ΔV (the
(depicted as black dashed line) is similar for all curves and difference between internal and external voltage that is linked
stays above 700 mV that indicates that the losses in Vo c are not to a gradient of the majority quasi-Fermi level in the contact
due to a poor surface passivation, but due to selectivity losses region) as a figure of merit for the contacts selectivity. We mod-
reflected in ΔV > 0. Fig. 5(a) shows that there are two kinds of eled a TMO-based silicon heterojunction featuring a high band
selectivity losses ΔVΦ , ΔVTAT that are exemplarily shown by gap n-type metal oxide to form a hole-selective contact. We de-
the blue and the red arrow, respectively, and will be described duced the following guidelines to build a hole-selective contact
in the next section. (similar findings should apply for electron contacts).
1) The induced hole density in the contact region, quantified
D. Analysis of Selectivity Losses by the band bending ϕcSi , has to be sufficiently high. This
Fig. 7 represents the simulation data of Fig. 5(a) by plotting is the case for a TMO work function Φ close to the valence
the selectivity losses ΔV as a function of the c-Si band bending band energy of the absorber. However, a passivating buffer
ϕcSi , which can be measured by SPV and takes Fermi-level pin- layer (e.g., a-Si) is expected to further raise the demand
ning into account. The bottom right-hand corner corresponds to on the TMO work function:
an ideal contact characteristic featuring a high c-Si band bend-
ing and no selectivity losses. For these well-selective contacts, Φ >
∼ EVcSi ⇒ ϕcSi sufficiently high.
the established characterization via ohmic contact and standard
recombination losses as quantified by RS and J0 is a meaning-
ful procedure since the diode equation is valid as motivated in 2) A TMO conduction band energy ECTM O greater than the
Section I. However, for novel contact schemes a prior classifi- valence band energy of the buffer EVaSi (or, if omitted, the
cation in terms of Fig. 7 can be helpful for contact evaluation. absorber EVcSi ) allows for B2B tunneling of excess holes:
Furthermore, Fig. 7 helps to distinguish two different kinds of
selectivity losses: ECTM O > EVaSi ⇒ B2B.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

MESSMER et al.: NUMERICAL SIMULATION OF SILICON HETEROJUNCTION SOLAR CELLS FEATURING METAL OXIDES 9

TABLE II [2] S. M. Sze, Physics of Semiconductor Devices, 2nd ed. New York, NY,
SIMULATION PARAMETERS USA: Wiley, 1981.
[3] P. Würfel and U. Würfel, Physics of Solar Cells: From Basic Principles
to Advanced Concepts, 3rd ed. Weinheim, Germany: Wiley, 2016.
Symbol Quantity Value [4] F. Feldmann, M. Bivour, C. Reichel, M. Hermle, and S. W. Glunz, “Passi-
vated rear contacts for high-efficiency n-type Si solar cells providing high
Sentaurus
interface passivation quality and excellent transport characteristics,” Sol.
Version L-2016.03-SP2
Energy Mater. Sol. Cells, vol. 120, pp. 270–274, 2014.
Global
[5] R. Peibst et al., “Building blocks for back-junction back-contacted cells
Temperature 300 K
and modules with ion-implanted poly-Si junctions,” in Proc. 40th IEEE
Spectrum AM1.5, 1 sun
Photovolt. Spec. Conf., Denver, CO, USA, 2014, pp. 852–856.
TMO
[6] J. Geissbühler et al., “22.5% efficient silicon heterojunction solar cell with
Thickness 10 nm
molybdenum oxide hole collector,” Appl. Phys. Lett., vol. 107, no. 8, 2015,
Band gap 3 eV
Art. no. 81601.
Donor concentration 1020 cm−3
[7] X. Yang et al., “High-performance TiO2 -based electron-selective con-
ρT MO
Tra p s Trap concentration varied tacts for crystalline silicon solar cells,” Adv. Mater., vol. 28, no. 28,
Trap volume 10−6 μm³ pp. 5891–5897, 2016.
Trap capture cross section 10−12 cm² [8] M. Mews, L. Korte, and B. Rech, “Oxygen vacancies in tungsten oxide
(eXSection, hXSection) and their influence on tungsten oxide/silicon heterojunction solar cells,”
Amorphous silicon (a-Si) Sol. Energy Mater. Sol. Cells, vol. 158, pp. 77–83, 2016.
Thickness 5 nm [9] L. G. Gerling et al., “Transition metal oxides as hole-selective contacts in
Crystalline silicon (c-Si) silicon heterojunctions solar cells,” Sol. Energy Mater. Sol. Cells, vol. 145,
Thickness 150 μm pp. 109–115, 2016.
Resistivity 1 Ωcm (n-type for p-contact [10] M. Bivour, J. Temmler, H. Steinkemper, and M. Hermle, “Molybdenum
p-type for n-contact) and tungsten oxide: High work function wide band gap contact materials
Rear for hole selective contacts of silicon solar cells,” Sol. Energy Mater. Sol.
a-Si(i) thickness 5 nm Cells, vol. 142, pp. 34–41, 2015.
a-Si(doped) thickness 10 nm [11] J. Bullock et al., “Survey of dopant-free carrier-selective contacts for
a-Si(doped) doping conc. 1021 cm−3 (n-type for p-contact silicon solar cells,” in Proc. 2016 IEEE 43rd Photovolt. Spec. Conf., 2016,
p-type for n-contact) pp. 210–214.
Interfaces [12] R. Singh, M. A. Green, and K. Rajkanan, “Review of conductor-insulator-
S 0m e ta l Recombination velocity at metal 101 cm/s for MS 106 cm/s for semiconductor (CIS) solar cells,” Sol. Cells, vol. 3, no. 2, pp. 95–148,
MIS, T-SHJ 1981.
S 0c S i Recombination velocity at c-Si 0 cm/s for MIS, T-SHJ [13] L.-M. Chen, Z. Xu, Z. Hong, and Y. Yang, “Interface investigation
interface and engineering—Achieving high performance polymer photovoltaic de-
Tunneling vices,” J. Mater. Chem., vol. 20, no. 13, pp. 2575–2598, 2010.
e-, h-Tunneling masses 0.1 [14] Synopsis, “Sentaurus device user guide: Release L-2016.03-SP2,” 2016.
g factors 1 [Online]. Available: http://www.synopsys.com
TAT-Tunneling Elastic (E Phonon = 0) [15] M. Bivour et al., “Doped layer optimization for silicon heterojunctions
Models by injection-level-dependent open-circuit voltage measurements,” IEEE
Auger model Standard (“Auger”, [11]) J. Photovolt., vol. 4, no. 2, pp. 566–574, Mar. 2014.
[16] U. Würfel, A. Cuevas, and P. Würfel, “Charge carrier separation in solar
cells,” IEEE J. Photovolt., vol. 5, no. 1, pp. 461–469, Jan. 2015.
[17] W. Schottky, “Vereinfachte und erweiterte Theorie der Randschicht-
3) If 2) is not fulfilled, an efficient hole transport via TAT by gleichrichter,” Zeitschrift für Physik A Hadrons Nuclei, vol. 118, no. 9,
traps in the TMO close to the valence band energy of the pp. 539–592, 1942.
buffer (absorber) is needed: [18] H. Steinkemper, “Numerical simulation of silicon solar cells,” Ph.D. dis-
sertation, Universität Konstanz, Konstanz, Germany, 2017.
[19] M. Bivour, “Silicon heterojunction solar cells analysis and basic
ECTM O < EVaSi ⇒ TAT. understanding,” Ph.D. dissertation, Fraunhofer ISE, Albert-Ludwigs-
Universität Freiburg, Freiburg im Breisgau, Germany, 2015.
An important finding for experimental investigations is that [20] M. A. Green and J. Shewchun, “Minority carrier effects upon the small sig-
both an insufficient work function and inefficient tunneling nal and steady-state properties of Schottky diodes,” Solid-State Electron.,
vol. 16, no. 10, pp. 1141–1150, 1973.
(e.g., due to absence of traps) can cause insufficient selectivity [21] F. Jiménez-Molinos, A. Palma, F. Gamiz, J. Banqueri, and J. A. Lopez-
(i.e., ΔV > 0) resulting in power losses. Taking the guidelines Villanueva, “Physical model for trap-assisted inelastic tunneling in metal-
from above into account allows us to minimize those losses oxide-semiconductor structures,” J. Appl. Phys., vol. 90, no. 7, pp. 3396–
3404, 2001.
by engineering the contact toward better selectivity. Only after [22] M. Bivour et al., “Principles of carrier-selective contacts based on induced
that, once the external voltage matches the internal voltage, the junctions,” in Proc. 33rd Eur. PV Solar Energy Conf. Exhib., Amsterdam,
diode equation is valid and the well-established optimization of The Netherlands, Sep. 25–29, 2017, pp. 348–352.
[23] M. Bivour, B. Macco, J. Temmler, W. E.M. Kessels, and M. Hermle,
“classical” losses like ohmic contact resistivity and passivation, “Atomic layer deposited molybdenum oxide for the hole-selective con-
quantified by RS and J0 , is a meaningful procedure. tact of silicon solar cells,” Energy Procedia, vol. 92, pp. 443–449,
2016.
[24] L. Neusel, M. Bivour, and M. Hermle, “Selectivity issues of MoOx based
APPENDIX hole contacts,” Energy Procedia, vol. 124, pp. 425–434, 2017.
[25] M. Bivour, F. Zähringer, P. Ndione, and M. Hermle, “Sputter-deposited
Simulation parameters are shown in Table II. WOx and MoOx for hole selective contacts,” Energy Procedia, vol. 124,
pp. 400–405, 2017.
[26] C. Messmer, “Numerical simulation and analysis of metal oxide contact
REFERENCES properties for silicon heterojunction solar cells,” Master’s thesis, Albert-
Ludwigs-Universität Freiburg, Freiburg im Breisgau, Germany, 2017.
[1] R. M. Swanson, “Approaching the 29% limit efficiency of silicon solar
cells,” in Proc. 31st IEEE Photovolt. Spec. Conf. Lake Buena Vista, Lake
Buena Vista, FL, USA, 2005, pp. 889–894. Authors’ photographs and biographies not available at the time of publication.