Content uploaded by Pietro P. Altermatt
Author content
All content in this area was uploaded by Pietro P. Altermatt on Jan 20, 2014
Content may be subject to copyright.
BORON-DOPED PECVD SILICON OXIDES AS DIFFUSION SOURCES FOR SIMPLIFIED
HIGH-EFFICIENCY SOLAR CELL FABRICATION
Nadine Wehmeier1, Gunnar Schraps1, Hannes Wagner3, Bianca Lim1,
Nils-Peter Harder2*, and Pietro P. Altermatt3
1 Institute of Solar Energy Research Hamelin (ISFH), Am Ohrberg 1, 31860 Emmerthal, Germany
2 Institute of Electronic Materials and Devices, Univ. Hannover, Schneiderberg 32, 30167 Hannover, Germany
*Now with Total New Energies, USA; seconded to SunPower Corp., 77 Rio Robles, San Jose, California 95134, USA
3 Dep. Solar Energy, Leibniz Universität Hannover, Appelstr. 2, 30167 Hannover, Germany
ABSTRACT: We develop plasma-deposited boron-doped silicon oxide layers with multifunctional capabilities: Firstly,
they are used as local diffusion sources for attaining low saturation current densities of J0 = 84 fA/cm2 at sheet resistances
as low as 67 /□ and surface boron concentrations in the range of 4–71019 cm‐3, suitable for efficient metallization.
Secondly, they act as diffusion barriers against other dopants. Therefore a co-diffusion process consisting of a boron
drive-in and a phosphorus diffusion in a POCl3 furnace in only one high temperature step potentially allows to simplify
the fabrication of high-efficiency n-type PERT cells. Process and device simulations indicate that a co-diffusion
temperature near 900C and a co-diffusion time of about 20 minutes are optimum for reaching cell efficiencies near 21%.
Keywords: PECVD, oxide, boron, diffusion, simulation
1 INTRODUCTION
During conventional fabrication of high-efficiency
solar cells, a high number of process steps is usually
necessary for creating structured or single-side emitter
and back surface field (BSF) regions. The process steps
include at least two high-temperature furnace diffusions,
each accompanied with depositing and removing capping
layers.
Resource, cost and time efficient production processes, as
shown in figure 1, may be achieved by using oxide layers
deposited by means of plasma-enhanced chemical vapour
deposition (PECVD): the doped oxide layers are
deposited one-sided, and in only one co-diffusion
process, the dopant atoms from the oxide are driven in
while simultaneously a second doping type is diffused
from the gas phase as already demonstrated in [1].
Alternatively, PECVD oxide layers with different doping
types can be deposited on both sample sides and act as
diffusion sources in a high-temperature step, so that the
different doping profiles can be controlled and optimised
independently as shown in [2]. Multifunctional PECVD
layer stacks can also be used for surface passivation and
as anti-reflection coating as demonstrated in [3].
In this paper, we investigate how the boron and
phosphorus doping profiles are influenced and can be
controlled independently during the co-diffusion process.
This is done by applying a nitride capping layer on top of
the boron-doped PECVD oxide and by varying the
nitrogen gas-flow and thus modifying the atmosphere in
the furnace. In addition, experimental and simulated
doping profiles resulting from different diffusion times
and temperatures are investigated both by experiments
and simulations. This allows us to achieve a deeper
understanding of the co-diffusion optimisation with
regard to high efficiencies of solar cells fabricated with
doped oxides.
2 EXPERIMENTAL PRCEUDRES
Doped oxide layers, between 40 nm and 100 nm thick,
are grown on monocrystalline silicon substrates by means
of PECVD within about four minutes, using an
inductively coupled plasma (ICP) source. As precursor
gases, silane (SiH4) and nitrous oxide (N2O) are used. To
produce doped oxides, the precursor gas diborane (B2H6)
or phosphine (PH3) is added. In contrast, depositing
PECVD borosilicate glass (BSG) layers from
trimethylborate (TMB) has been investigated e.g. in [1].
In this paper, we first focus on growing boron-doped
silicon oxides on planar p-type monocrystalline silicon
substrates. Then we present boron-doped oxides on n-
type alkaline textured silicon surfaces, relevant for
fabricating n-type PERT cells.
Figure 1: Simplified fabrication of n-type passivated
emitter and rear totally diffused (PERT) solar cells. Using
boron-doped PECVD oxides allows us to form the boron
emitter and the phosphorus BSF simultaneously in a co-
diffusion process. Only one high temperature step and
one protection layer are needed.
To achieve a diffusion barrier for other dopants, some
samples are subsequently deposited with a SiNx capping
layer, having a refractive index of n = 1.9 and a thickness
of about 100 nm.
In a high-temperature step, the dopant atoms diffuse from
the PECVD oxide into the silicon substrate. This boron
drive-in step is combined with a POCl3 furnace diffusion
in a co-diffusion process, in order to form a n-type
emitter and a p-type BSF on a p-type substrate
28th European Photovoltaic Solar Energy Conference and Exhibition
1980
simultaneously. This process enables us to skip a separate
BBr3 diffusion within a quartz furnace.
Note that phosphorus atoms diffuse effectively at
temperatures of about 800-900°C, while the boron
diffusion is only effective above about 900°C. As a
consequence, the co-diffusion step within the furnace is
splitted into two segments: first, the drive-in of B at
1000°C for 30-60 min in a nitrogen atmosphere (with N2
gas flows between 3 l/min and 12 l/min) and, second, the
diffusion of P after cooling down to 852°C with POCl3
(400 ml/min), O2 (2 l/min) and N2 (9 l/min) gases for 20
min.
The resulting doped emitter and BSF regions are
characterised by their sheet resistivities and saturation
current densities that are determined by means of four
point probe (FPP) and inductively-coupled
photoconductance measurements. In addition
electrochemical capacitance voltage (ECV)
measurements are performed on both sample sides to
determine the B and P doping profiles, especially the
surface concentrations and the profile depths.
First PERT cells with a boron-doped PECVD oxide as
diffusion source are fabricated with the simplified
process-flow shown in figure 1. An n-type silicon
substrate with a specific resistance of 5.5 Ωcm and a
thickness of about 160 µm after alkaline texture is used.
After PECVD-BSG and SiNx capping layer deposition
the co-diffusion as described above is applied. Resulting
BSG and PSG layers are removed in hydrofluoric acid
and the front and rear surfaces are passivated by a 100
nm thick AlOx/SiNx stack and a SiNx layer, respectively.
For metallisation AgAl and Ag screen-printing are used
and co-fired. For the characterisation of the cells, current-
voltage (IV) measurements are performed.
3 EXPERIMENTAL RESULTS
3.1 Doped oxides on planar p-type Si wafers
The diffusion profiles are known to be strongly
influenced by the temperature and time during the drive-
in step. We observe that also the applied nitrogen flow
influences the resulting sheet resistivities: an increasing
N2 flow leads to decreasing sheet resistivities of the P
emitters as exemplarily shown in figure 2.
Figure 2: Doping profiles measured by ECV on the
sample side that was uncovered and therefore was P-
diffused during the co-diffusion process with N2 gas-
flows of either 8 l/min or 12 l/min during the 1000°C
step. The ECV measurements indicate that n-type doped
emitters with sheet resistivities of about 13-125 Ω/□ are
reached, depending on the N2 flow and the sample
position within the furnace.
Figure 3 demonstrates that by the co-diffusion
process, BSF regions with sheet resistances of about 67-
97 Ω/□ are successfully produced. Note the strong
influence of the SiNx capping layer: only oxide/nitride
stacks result in p-type doped regions. If the doped oxide
layers are exposed to the co-diffusion uncapped, a n-type
doped surface region is observed, indicating
overcompensation by P.
Figure 3: Doping profiles measured by ECV on the
sample sides that were covered with boron-doped
PECVD oxides, (a) with and (b) without SiNx capping
layer. A co-diffusion process with N2 gas
flows of either 8 l/min or 12 l/min during the 1000°C step
was performed. (a) P-type doped BSF regions with sheet
resistivities of about 67-97 Ω/□, measured by ECV and
FPP are reached, nearly unaffected by the N2 flow and
the sample position within the furnace. (b) In the case of
the uncapped oxide layers, n-type doping is measured at
the sample surface indicating that phosphorus was
diffused through the boron-doped oxide. At a gas
flow of 8 l/min, the measured doping type changes to p-
type due to boron diffusion, while at 12 l/min of N2 no p-
type doping is observed at all.
3.2 Doped oxides on textured n-type Si wafers
The doped oxides deposited on an alkaline textured
surface are investigated by means of scanning electron
microscopy (SEM) directly after PECVD deposition and
after co-diffusion and BSG removal in order to
investigate the homogeneity of the doped oxide and the
resulting boron emitter.
The SEM images presented in figure 4 show that the
PECVD of a boron-doped oxide with a deposition
thickness of 80 nm on a planar silicon surface leads to a
closed oxide layer with a thickness of about 40-65 nm.
This oxide thickness is to be expected due to the about
1.7 times higher surface of textured silicon. After the co-
diffusion a continuous p-type boron emitter is formed on
(a)
(b)
28th European Photovoltaic Solar Energy Conference and Exhibition
1981
the n-type silicon, that is clearly visible in the SEM
image due to the doping contrast.
Figure 4: SEM images of boron doped PECVD oxides
on alkaline textured n-type silicon. (a) A closed oxide
layer is formed, its thickness varies between about 65 nm
at the top and 40 nm at the bottom of a pyramid. (b) A
continuous p-type emitter with a depth of 500-900 nm is
formed during the boron drive-in at 1000°C (apparent as
bright area).
The boron concentration in the doped oxide layers is
increased by changing the gas flows of the precursor
gases. The resulting doping profiles measured by ECV
shown in figure 5 reveal high surface concentrations of
about Nsurf = 3.5-141019 cm-3 resulting in sheet
resistivities of Rsheet = 61-73 Ω/□ after 30 min of boron
drive-in. Increasing the drive-in time to 60 min results in
deeper doping profiles and thus lower sheet resistivities
of 21-53 Ω/□. Corresponding emitter saturation current
densities of J0e = 125-145 fA/cm² have been measured on
planar p-type 200 Ωcm silicon substrates, using the
intrinsic density ni = 8.6109 cm-3.
If the boron concentration in the doped oxide is high
enough to exceed the solubility limit of boron in silicon
of about 1.01020 cm-3 at the drive-in temperature of
1000°C [4], a boron rich layer (BRL) formation at the
silicon/oxide interface occurs during the high temperature
step. The presence of a BRL is indicated by the observed
hydrophilic behaviour of the sample surface after HF
etch. It makes a low temperature oxidation (LTO)
necessary to remove the BRL in order to prevent bulk
carrier lifetime degradation as observed in [5].
Figure 5: Boron doping profiles measured by ECV.
Higher surface concentrations and diffusion depths and
lower sheet resistivities are reached by increasing the
B2H6/SiH4 gas flow ratio and the boron drive-in time.
3.3 Cell results
From the IV measurements on the first fabricated n-
type PERT cells with an area of 4 cm² the cell parameters
shown in table I are extracted.
Table I: Results of the first n-type PERT cells we
fabricated in a simplified manner with boron-doped
PECVD oxide as diffusion source.
[%]
FF
[%]
pFF
[%]
Jsc
[mA/cm²]
Voc
[mV]
A
[cm²]
17.5 72.9 78.2 37.9 633 4.0
4 DISCUSSION
We attribute the influence of different N2 flows on the
sheet resistivities to the following effect: Nitrogen
displaces oxygen that is inserted in the furnace while
loading the quartz boat containing the samples. For low
N2 flows during the 1000°C step, a high concentration of
unwanted oxygen reacts with silicon on the uncoated
substrate surface leading to a thick thermal oxide layer.
For the rear position, the oxygen concentration and
therefore the oxide thickness is highest leading to sheet
resistivities of about 125 Ω/□. But for high N2 flows
and thin thermal oxide layers, the amount of P2O5 from
the furnace walls dissolved in the oxide is sufficient to
decrease the melting temperature of the SiO2- P2O5-phase
such that this mixed phase becomes liquid at diffusion
temperatures ≥850°C [6]. Thus P-diffused regions with
very low emitter sheet resistivities of about 13 Ω/□ are
formed during the 1000°C step and the subsequent POCl3
diffusion.
Boron-doped oxide layers grown by PECVD do not
act as diffusion barriers against phosphorus diffusion at
temperatures as high as 1000°C. The measured ECV
doping profiles, which reveal a change of doping type
shown in figure 3 (b), may be explained as follows:
during the 1000°C step, oxygen diffuses through the
PECVD oxide and forms a thermal oxide layer at the
silicon interface. In the case of a high nitrogen flow and a
thin oxide layer, P2O5 can react at the silicon surface so P
diffuses into the substrate. Simultaneously, boron is
diffused from the doped oxide, but due to the higher P
concentration and solubility in silicon compared to B, a
n-type doped region is formed. In the case of a lower N2
gas flow of 8 l/min, a thicker thermal oxide is grown that
retards the P diffusion, while boron is already driven in
so only the surface region becomes P doped. See figure 3
(b), co-diffusion 1.
5 PROCESS AND DEVICE SIMULATION
We use the software SENTAURUS PROCESS [7] to
reproduce the experimental boron profile shown in
figure 3 (a) (co-diffusion number 3, middle position). We
use a simulation approach comparable to [8] and adjust
the boron concentration in the 80 nm thick oxide layer,
from where the boron diffuses into silicon. We use the
same temperature and time sequence as in the experiment
and obtain the result shown in figure 6 as lines. The
experiment (symbols) is reproduced very well without the
need to adjust any other model parameters. For
comparison, we also simulate a standard boron silicate
glass (BSG) layer, shown in figure 6 as dashed line. A
(a)
(b)
28th European Photovoltaic Solar Energy Conference and Exhibition
1982
standard BSG has a higher boron concentration than our
boron-doped oxide, resulting in a higher boron peak
concentration in silicon. To achieve the same profile
depths, we need to reduce the diffusion time by about
50%. The reason is that the diffusivity of boron in Si
depends on the boron concentration.
Figure 6: Boron dopant profiles, fabricated with a boron-
doped oxide (symbols), as simulated (solid line), and a
simulation of a profile fabricated with a standard boron
silicate glass (BSG) layer (dashed line).
We use the software SENTAURUS DEVICE (former
DESSIS) with the models described in [9] to compute the
saturation current density J0 of these two profiles with the
method of Kane and Swanson [10]. Because J0 depends
sensitively on the surface recombination velocity S, we
update the data on Al2O3-passivated, boron-diffused
surfaces of Hoex et al. [11] by including the recent data
from Richter at al. [12], as shown in figure 7. The line is
a least-square fit to the data (symbols), expressed as:
(1)
similarly to [13]. Using this updated parameterisation for
S, we obtain J0 = 84 fA/cm2 for the doped-oxide profile,
compared to J0 = 183 fA/cm2 for the BSG profile (at
25C, using 8.312109 cm-3 as intrinsic density ni, given
in equation 3 of [14]). The reason for this difference in J0
is the reduced amount of Auger recombination in the
doped-oxide profile. By doping the oxide weaker, J0 may
be reduced further without reaching unfeasibly high sheet
resistivities.
Figure 7: Symbols: surface recombination velocity on
boron-diffused Al2O3-passivated surfaces, extracted by
device simulations from published dopant profiles and J0
values [8,9]. Line: the parameterisation given in equation
(1).
Finally, we model the performance of PERT cells in
two dimensions, again by using SENTAURUS DEVICE. As
p+-emitters, we use the simulated boron dopant profiles
shown in figure 8. The boron density in the oxide is
assumed to be sufficiently high, so the solid-solubility of
boron in Si is just reached at the surface. This is done
because, with higher boron concentrations, lower sheet
resistivities are attained within shorter diffusion times
and at lower temperatures, which is industrially relevant.
However, care must be taken that the solubility limit is
not exceeded, otherwise a boron rich layer is formed,
which would need to be oxidised for removal. At the rear
side of the cell, the phosphorus profiles measured and
modeled in [15] are used as BSFs, shown in figure 9. The
simulation domain is two-dimensional, as described in
[7] in detail. We add a lumped external series resistance
of 0.5 cm2 due to the metallisation, and an effective
shading loss of 3% [16] in the module, which
corresponds to a metallisation fraction of the front
surface in the range of 5-6%. The SRH lifetime in the 1
cm base is assumed to be 1 ms. The resulting efficiency
for the three different diffusion temperatures and
diffusion times is shown in figure 10. It is apparent that
the cells, having a front finger distance of 2 mm, perform
considerably worse than if the finger distance is lowered
to 1 mm with accordingly finer metal fingers. The reason
is that boron has a small diffusivity and solid-solubility
compared to phosphorus, leading to a high sheet
resistivity of the emitter. To keep the manufacturing costs
down, it is important to keep both the co-diffusion
temperature and time as low as possible. The simulations
indicate (with the assumptions outlined above) that a co-
diffusion temperature near 900C and a co-diffusion time
near 20 minutes are optimum. In practice, often higher
temperatures and longer diffusion times need to be
chosen to achieve an optimum. This discrepancy between
modeling and experiment may point to relevant non-
idealities in fabrication, such as: (i) the oxide near the Si
surface may start to act as diffusion barrier during the
diffusion; (ii) inhomogeneity is a notorious problem in
boron diffusions, especially at lower temperatures; (iii)
the boron diffusion is defect-mediated, so the defects
introduced by oxidisation or texturing may play an
important role.
Regarding the strategy for PERT cell development,
the rather high sheet resistivity of boron diffusions
implies that, in the first place, special attention should be
given to fine front finger metallisation to keep the front
finger spacing small. With improved screen printing
pastes, it is likely that a fine front finger metallisation is
cheaper than long diffusion times at high diffusion
temperatures. Secondly, care must be taken that the
recombination losses in the phosphorus-diffused BSF do
not dominate the total recombination losses. This easily
occurs because phosphorus has a higher solubility and
diffusivity in Si than boron, and phosphorus leads to
precipitation. The losses in the BSF are most likely to
dominate at co-diffusion temperatures above 900°C.
Therefore, low POCl3-flows should be used when
designing PERT process sequences. Thirdly, our
simulations also indicate that a p-type wafer resistivity
near 1 cm is optimum (not shown here). Wafer
resistivities near 5 cm, for example, lower the
efficiency by 0.3% absolute, assuming that the SRH
lifetime is independent of wafer resistivity.
28th European Photovoltaic Solar Energy Conference and Exhibition
1983
Figure 8: Boron dopant-profiles, simulated with
SENTAURUS PROCESS for the indicated diffusion
temperatures and diffusion times, using highly-doped
oxides.
Figure 9: Phosphorus dopant-profiles, modeled in [15]
based on experiments, for the indicated diffusion
temperatures and diffusion times.
Figure 10: Simulated PERT cell efficiencies in
dependence of the co-diffusion time, for the various co-
diffusion temperatures as indicated. The solid lines are
with a front metal finger distance of 1 mm, the dashed
lines with 2 mm.
6 CONCLUSIONS
Doped oxide layers have a high potential to simplify
the fabrication of high-efficiency Si cells like PERT and
interdigitated back contact (IBC) cells. High temperature
steps and protection layers and thus process time and
costs are likely to be reduced, which is necessary for an
efficient industrial production.
First n-type PERT cells with boron-doped PECVD
oxides have been fabricated and reached efficiencies of
= 17.5%. Simulations of dopant-profiles and PERT cell
efficiencies show that by optimising the co-diffusion
process concerning diffusion time and temperature
efficiencies in the range of 20-21% can be achieved.
There are some requirements to the emitter and BSF
of a high-efficiency PERT cell with reduced process
steps that we fulfil by optimising the deposition of doped
PECVD oxides and the co-diffusion process: the oxides
have to be deposited homogeneously to form a
continuous boron emitter on alkaline textured n-type
silicon. The emitter needs to have a sufficient thickness
and surface concentration to allow efficient metallisation.
But the boron concentration in the doped oxide layer
must be kept low enough to prevent BRL formation so
that no additional LTO and HF etch are needed. Also the
J0 value of a weaker boron doped oxide is further reduced
allowing for improved cell performance. We control the
doping profile of the phosphorus BSF efficiently by the
gas flows during the POCl3 co-diffusion while the boron
profile is unaffected by the atmosphere in the furnace and
is adjusted by temperature and time during the drive-in.
[1] B. Bazer-Bachi, C. Oliver, B. Semmache, Y.
Pellegrin, M. Gauthier, N. Le Quang, M. Lemiti, 26th
EU PVSEC Proceedings (2011), p. 1155.
[2] P. Rothardt, R. Keding, A. Wolf, D. Biro, Phys.
Status Solidi RPL (2013) 1-4.
[3] J. Seiffe, F. Pillath, D. Trogus, A. Brand , C. Savio,
M. Hofmann, J. Rentsch, R. Preu, IEEE Journal of
Photovoltaics, vol.3, no.1 (2013) p.224-229.
[4] R. Hull, Properties of cristalline silicon, 10.4
Solubility of B, Al, Ga, In, Tl, P, As and Sb in c-Si
(1999).
[5] M. Kessler, T. Ohrdes, B. Wolpensinger, N.-P.
Harder, Semicond. Sci. Technol. 25 (2010) 055001.
[6] P. Balk, J. M. Elderidge, Phosphosilicate Glass
Stabilization of FET Devices, Proceedings of the
IEEE, vol.57, no.9 (1969) p. 1558- 1563.
[5] Sentaurus Userguide, Synopsys, Mountain View, CA,
2013.
[8] H. Wagner et al., Proc. 37th IEEE PV Specialists
Conf. (2011).
[9] P. P. Altermatt, J. Comput. Electron. 10 (2011) 314.
[10] D.E. Kane and R.M. Swanson, 18th IEEE PV
Specialists Conf. (1985), 578.
[11] B. Hoex, et al., Appl. Phys. Lett. 91 (2007) 112107.
[12] A. Richter et al., IEEE J. PV3 (2013) 236.
[13] P.P. Altermatt et al., J. Appl. Phys. 3187 (2002) 92.
[14] K. Misiakos and D. Tsamakis, J. Appl. Phys. 74
(1993) 3293.
[15] A. Bentzen et al., J. Appl. Phys. 99 (2006) 064502.
[16] R. Woehl, et al., 23rd European PV Solar Energy
Conf. (2008) 1377.
28th European Photovoltaic Solar Energy Conference and Exhibition
1984