You are on page 1of 7

Solar Energy Materials & Solar Cells 131 (2014) 30–36

Contents lists available at ScienceDirect

Solar Energy Materials & Solar Cells


journal homepage: www.elsevier.com/locate/solmat

Implementation of Fermi–Dirac statistics and advanced models


in PC1D for precise simulations of silicon solar cells
Halvard Haug a,n, Achim Kimmerle b, Johannes Greulich b, Andreas Wolf b,
Erik Stensrud Marstein a
a
Institute for Energy Technology, Instituttveien 18, 2007 Kjeller, Norway
b
Fraunhofer Institute for Solar Energy Systems, Heidenhofstraße 2, 79110 Freiburg, Germany

art ic l e i nf o a b s t r a c t

Article history: In this work we present a modified version of the commonly used semiconductor simulation program
Received 15 April 2014 PC1D, named cmd-PC1D 6.0, which applies Fermi–Dirac (F–D) statistics, thus improving on the existing
Received in revised form Boltzmann approximation that is employed in the original program. Several up-to-date models for
4 June 2014
crystalline Si have been implemented in the new program, including injection dependent band gap
Accepted 8 June 2014
Available online 2 July 2014
narrowing, carrier mobility and Auger recombination. The results obtained using cmd-PC1D 6.0 are
verified against well-accepted and modern simulations tools for a range of both phosphorus- and boron-
Keywords: doped emitters with varying surface concentration. We find a good correspondence between the
Silicon simulation results obtained with the different programs, with a deviation of less than 2 fA/cm2 for the
Device simulation
calculated emitter saturation current density using F–D statistics and less than 0.8 % relative deviation in
Solar cell
all the main solar cell parameters under 1 sun illumination. By using the correct carrier statistics and the
Fermi–Dirac statistics
appropriate set of models it is now possible to avoid the commonly used effective models and adapted
values, thus taking a large step in bringing PC1D further towards a more general, consistent and
physically meaningful simulation tool for Si solar cells. The new program can be run from a command
line or within a recently developed, powerful graphical user interface. Both cmd-PC1D 6.0 and the
graphical user interface are freely available for download.
& 2014 Elsevier B.V. All rights reserved.

1. Introduction description of the relative contributions from surface recombination


and recombination in highly doped bulk regions below the surface,
Since electrons are Fermions, they obey Fermi–Dirac (F–D) F–D statistics in combination with a suitable actual BGN model (e.g.
statistics. For carrier concentrations in Si lower than approximately the parameterizations by Schenk [2] or by Yan and Cuevas [3])
1019 cm  3, F–D statistics can be approximated by Boltzmann should therefore be applied [1].
statistics. However if solar cells with doping concentrations larger One of the main reasons for the continued use of Boltzmann
than 1019 cm  3 are modelled with Boltzmann statistics, large errors statistics for solar cell simulations is the one-dimensional semi-
of the surface recombination and the recombination occurring in conductor device simulator PC1D, which has strongly influenced
the bulk of the emitter may occur [1]. The choice of carrier statistics the solar cell research community over decades [4–8]. PC1D is
is closely linked to the parameterization of band gap narrowing based on the Boltzmann approximation, and several PV-specific
(BGN), as the error induced by the use of Boltzmann statistics is models have been developed for use in this framework.
commonly compensated for by the use of apparent BGN models, The program is still commonly used today, and a command line
along with adjustments of the surface recombination velocity (SRV). version (cmd-PC1D) and a new, efficient graphical user interface
However, only adapting the SRV is not sufficient to recount for using have recently been developed to facilitate more efficient simula-
Boltzmann statistics in a physically sound manner and may even tions [9]. In this paper we present a new, modified version of cmd-
lead to unphysical negative SRV in some cases. Additionally this PC1D called cmd-PC1D 6.0, which applies the correct, more
adaption may lead to differences in the calculated carrier collect- general F–D statistics together with a comprehensive BGN model,
ion efficiency for structures with highly doped regions. For a correct thus giving more accurate results.
Furthermore, since the release of the last PC1D version, refined
models for crystalline silicon have become available. These rely on
n
Corresponding author. Tel.: þ 47 63806233. new data and are generally more precise and sophisticated than the
E-mail address: halvard.haug@ife.no (H. Haug). parameterizations of semiconductor properties already implemented

http://dx.doi.org/10.1016/j.solmat.2014.06.021
0927-0248/& 2014 Elsevier B.V. All rights reserved.
H. Haug et al. / Solar Energy Materials & Solar Cells 131 (2014) 30–36 31

in PC1D. In order to increase the quality of Si device simulation, we shown in Eq. (1). For very large doping or high injection condi-
have therefore implemented several advanced Si-specific models into tions, the quasi-Fermi energies approach or even merge into the
cmd-PC1D 6.0, including updated models for intrinsic carrier density, bands, and the semiconductor is said to be degenerate. In these
carrier mobility and Auger recombination. The updated models are circumstances the full F–D expressions are needed to obtain
not pre-initialized e.g. by the doping concentrations, but may be correct results. The main reason why this is often avoided in
coupled to the carrier densities throughout the device, enabling simulations is because Eq. (4) cannot be calculated analytically,
injection dependent evaluation. and a finely resolved numerical evaluation of the integral makes
The present paper is organized as follows. In Section 2, we the simulation computationally heavy. This issue can however be
discuss the implementation of F–D statistics in cmd-PC1D 6.0. solved by employing one of several numerical approximations. In
Section 3 deals with the choice and verification of the models cmd-PC1D 6.0, two different analytical expressions are used to
describing the fundamental semiconductor properties, e.g. intrin- effectively evaluate the F–D operator [12] and it's inverse function
sic charge carrier density, mobility and BGN. Simulations of the [13]. These expressions are also used in the emitter simulation tool
emitter saturation current density and important current–voltage EDNA [14] and have previously been shown to lead to less than
characteristics are verified against other well-accepted simulation 0.003% relative error in the carrier concentrations with little loss
tools in Section 4. In Section 5, the possible further work is in computation time [15].
discussed. The key results of the present work are summarized The carrier concentrations in cmd-PC1D 6.0 are then calculated
in Section 6. using the same solution variables as in the original PC1D program:
 
qψ þ ΔEc  qϕn;i þ lnðni;0 =N c Þ
n ¼ Nc F1=2 kB T
2. Implementation of Fermi–Dirac statistics   ð5Þ
 qψ þ ΔEv þ qϕp;i þ lnðni;0 =Nv Þ
p ¼ N v F1=2 kB T
;
In the original PC1D program, the carrier densities are calcu-
Note that the terms lnðni;0 =Nc Þ and lnðni;0 =N v Þ have been added
lated with respect to the intrinsic carrier density ni , using
in the numerator in the expressions for n and p. These terms
Boltzmann statistics. For a single-region simulation (no hetero-
correspond to ðEi  Ec Þ and ðEi Ev Þ, respectively, and must be
junctions), the electron and hole carrier densities are then calcu-
added to account for the change in the energy point of reference
lated as [10]
! when replacing the prefactor ni;0 in Eq. (1) with N c and N v . The
 
qψ þ ΔEc  qϕn;i  qψ þ ΔEv þqϕp;i replacement of Eq. (1) with Eq. (5) (as well as the reverse
n ¼ ni;0 exp ; p ¼ ni;0 exp ; relations) has now been done throughout the entire PC1D source
kB T kB T
code. When Boltzmann statistics are chosen by the user the F1=2
ð1Þ
function is simply changed to an exponential function (and a
respectively. Here, ni;0 is the reference intrinsic carrier density, q is natural logarithm for the inverse function). By also selecting the
the elementary charge, kB is the Boltzmann coefficient and T is the original PC1D models, this causes the results to revert back to
temperature. Furthermore, ψ is the electrostatic potential (taken those of the original program, as expected.
as zero for neutral intrinsic conditions), ΔEc and ΔEv are the shift
of the band edges due to band gap narrowing toward the center of
the bandgap (after accounting for electrostatic effects) and ϕn;i and 3. Implementation of new physical models
ϕp;i are the quasi-Fermi potentials for electrons and holes relative
to the equilibrium Fermi energy Ei : qϕn;i ¼ Ei  EFn and qϕp;i ¼ 3.1. Choice of models and parameterizations
Ei  EFp .
The chosen value for ni;0 at a given temperature has a major In contrast to many semiconductor devices, solar cells are
impact on the simulation results. ni;0 is related to the effective strongly affected by carrier recombination and thus by the excess
density of states N c and N v and the band gap Eg through the concentration of minority carriers. Both the pn product and the
relation [11] recombination processes in the device must therefore be known
pffiffiffiffiffiffiffiffiffiffiffi   very precisely for accurate results. The pn product scales as the
Eg square of the intrinsic carrier density ni , and at large doping
ni;0 ¼ N c N v exp  : ð2Þ
2kB T densities it is also significantly influenced by degeneracy and band
In the original PC1D program, the absolute values for N c , N v gap narrowing. The models used in PC1D up to now have been
and Eg do not affect the solution for single-region problems for a developed for use together with Boltzmann statistics, and should
given value of ni;0 , as only ni;0 enters in Eq. (1). For the full Fermi– therefore be re-evaluated when changing to F–D statistics. This is
Dirac calculation however, the carrier densities need to be directly particularly true for BGN, as the current models used in PC1D
related to the band edges and Nc and N v through the general represent the apparent BGN, which account for various effects,
expressions including degeneracy at high doping levels. Because of this we
    cannot apply F–D statistics together with the existing BGN models,
Ec  EFn EFp  Ev
n ¼ N c F1=2  ; p ¼ Nv F1=2  ; ð3Þ as this would overestimate the high doping effects. Instead, the
kB T kB T
BGN is calculated using the comprehensive theoretical model for
where EFn and EFp are the quasi-Fermi energies for electrons and both p- and n-type Si by Schenk [2], which is derived for the case
holes and Ec and Ev are the conduction and valence band energies. of F–D statistics from a non-self-consistent, full random phase
F1=2 is the complete Fermi integral of order 1/2, given by approximation formalism, taking both carrier–carrier and carrier–
Z 1 dopant interactions into account. As in the original PC1D program
2 t 1=2
F1=2 ðηÞ ¼ pffiffiffiffi dt: ð4Þ the total BGN is applied to one of the band edges only, according to
π 0 e η þ1
t 
the doping polarity of the material. In the space charge region the
If the quasi-Fermi energies lie within the band gap and are local equilibrium carrier densities are used as input to the Schenk
separated from the conduction and valance band edges by more model in order to avoid excessively large BGN values in this region.
than 0.05 eV, the Boltzmann approximation can be used with In addition, several updated values and parameterizations of Si
negligible error, meaning that F1=2 ðηÞ can safely be approximated properties, based on the current state-of-the-art experimental
by an exponential function, and Eq. (3) can be reduced to the form results, have been published since the last version of PC1D was
32 H. Haug et al. / Solar Energy Materials & Solar Cells 131 (2014) 30–36

Table 1
The full set of models that currently are implemented into the new cmd-PC1D 6.0 program, together with the
numerical approximations for the Fermi operator and it's reverse.

Model Symbol Ref.

Intrinsic carrier density ni;0 Altermatt et al. [17]


Intrinsic energy band gap Eg;0 Green [21]
Effective density of states, conduction band Nc Green [21]
Effective density of states, valence band Nv (adjusted to match ni;0 )
Bandgap narrowing ΔEc ; ΔEv Schenk a[2]
Carrier mobility μn ; μp Klaassen a[22,23]
Intrinsic recombination U intr Richter et al. [19]
Kerr and Cuevas [20]
Numerical approximations for:
Fermi integral u ¼ F1=2 ðηÞ Van Halen and Pulfrey [12]
Inverse Fermi integral η ¼ f 1=2 ðuÞ Antia [13]

a
As default, these models are only evaluated for equilibrium conditions in the beginning of the simulation. There
is also an option to update these models at each iteration to account for high injection effects, at the cost of a slightly
lower computation speed.

Fig. 1. (a) Intrinsic carrier density and (b) carrier mobilities as a function of doping concentration, calculated with the original PC1D (black dashed line) and the modified
cmd-PC1D 6.0 program (red solid line). The results from the previous Excel VBA implementation are shown as black symbols. Bottom part of each subfigure: relative
deviation between cmd-PC1D 6.0 and the Excel VBA implementation. (For interpretation of the references to color in this figure legend, the reader is referred to the web
version of this article.)

released in 1997. The updated cmd-PC1D 6.0 is now based on a set of BGN for highly n-type doped Si [3], in which the authors state
of models suggested by Altermatt et al., who have performed an that their experimental findings are not compatible with Schenk's
extensive work on the subject [1,16]. These new models include widely accepted BGN model. However, the framework for includ-
the updated value of ni;0 (9:65  109 cm  3 at 300 K) [17], which ing new models and specifying which models and parameters to
has a significant impact on many simulation results as compared use is now in place in cmd-PC1D 6.0 and additional options for
to the previously used value of 1  1010 cm  3 [18]. We have also models will therefore be included in future versions.
implemented latest parameterisation of Auger recombination
published by Richter et al. [19], which was derived using F–D
statistics and Schenk's parameterization for BGN and is able to 3.2. Verification against previous implementation of the models
reproduce the high lifetimes measured by Richter et al. As an
alternative we have also implemented the commonly used model In order to verify that the new models have been implemented
by Kerr and Cuevas [20] for Auger recombination. correctly, we have performed a range of different simulations and
A complete list of the new models that have been implemented compared the results against a previous and well-checked Excel
in cmd-PC1D 6.0 is presented in Table 1. The models have been VBA implementation of the models. Fig. 1(a) shows the intrinsic
selected with much care and to best knowledge of the authors. carrier density ni of both p-type and n-type Si as a function of the
However, Table 1 does not necessarily represent a final, fully doping concentration, based on F–D statistics and models from
correct and consistent set of models for simulation of Si solar Refs. [13,15,19]. Boltzmann statistics can be observed to give a
cells. For instance there are still issues with carrier mobility in good approximation of ni up to a doping concentration of
heavily doped and in heavily compensated material, as Schenk's  1  1019 cm  3, but for higher doping there is a large deviation
BGN parameterization and experimental data for compensated Si from the correct F–D statistics, with the difference being a factor of
were not available when Klaassen published his mobility para- 2 at N D ¼ 2  1020 cm  3. Fig. 1(b) shows the carrier mobilities in
meterization. Another open question is the choice of BGN. Very n-type Si as a function of doping concentration based on the
recently, Yan and Cuevas presented an empirical parameterization comprehensive work of Klaassen [22,23]. Also here the most
H. Haug et al. / Solar Energy Materials & Solar Cells 131 (2014) 30–36 33

significant difference to the original PC1D can be seen for the of the simulation tools. The same set of models and parameters
largest doping concentrations, with ND 4 1020 cm  3. was used for cmd-PC1D 6.0, Sentaurus TCAD and EDNA 2, with the
For all the simulations, the relative deviation from the Excel default models presented in Table 1 and the device structure
VBA implementation was found to be below 0.05 %, mainly caused defined in Table 2.
by rounding errors. (A somewhat larger deviation of 2–3% relative Fig. 2 shows the calculated emitter saturation current as a
was observed for the Auger recombination in the space charge function of surface doping density, using cmd-PC1D 6.0, Sentaurus
region of the pn-junction, possibly caused by the polynomial form TCAD and EDNA 2. We observe an excellent correspondence
of the Auger parameterization [16]. This difference is however between the J 0e values calculated from the programs for all the
small and has normally little or no effect on the simulation emitters, both when using Boltzmann and F–D statistics, with a
results). The simulated intrinsic carrier concentration (and thus relative deviation below 5%, except for the first point with
the BGN) has previously been cross-checked against the numerical Nsurf ¼ 1  1018 cm  3. The increase in relative deviation for the
simulation tool EDNA [14] and, together with Richter's parameter- lowest surface doping is mainly caused by low numerical values.
ization of Auger recombination, is also implemented in the Sinton For F–D statistics, the absolute deviation is less than 2 fA/cm2 for
WCT-120 lifetime tester software [24]. Note that the use of models all the investigated emitters. As expected the results for Boltz-
which are common with modern characterization equipment is an mann and F–D statistics are very similar for the lowest doping
advantage of the new cmd-PC1D 6.0 program, as it facilitates densities, but start to deviate for N surf larger than 1019 cm  3, with
comparison between experimental results and simulations. a significant difference for the largest surface doping.
Finally, the Klaassen mobility model has been cross-checked
against EDNA and the Sentaurus TCAD simulation package [25].
4.2. Comparison of current–voltage characteristics

4. Simulation results and comparisons with other cmd-PC1D 6.0 enables the user to extract the charge carrier
simulation tools profiles and to simulate the optical and electrical behaviour of
solar cells in one dimension. In order to validate these features,
4.1. Comparison of emitter saturation currents they are compared to Sentaurus TCAD [25]. We apply the same
phosphorus-doped emitters and surface recombination para-
The largest differences when changing from Boltzmann statis- meters as in Section 3.2. In Sentaurus TCAD, we implemented a
tics to F–D statistics are expected for highly doped regions. quasi-one-dimensional symmetry element. All models and para-
Therefore, we chose to test cmd-PC1D 6.0 by performing calcula- meters are set to the same as in cmd-PC1D 6.0, except for the
tions of the emitter saturation current J 0e for a series of Gaussian following three modifications. First, as in Schenk's original work
emitters with different surface doping concentration. J 0e is calcu- [2], the BGN is implemented in a simplified, doping-dependent
lated in PC1D in a simple manner by simulating a pn-junction manner, which leads to small artificial BGN in the space charge
in the dark under moderate forward bias (in this case with region. Second, the boundary conditions for the charge carriers at
V¼ 0.55 V). J 0e is then given by the contacts are defined slightly differently. In Sentaurus TCAD,
the boundary conditions are defined as
J min;SCR
J 0e ¼   ð6Þ !
exp qV SCR
1 jn  n^ ¼ q  Sn0  ðn  n0 Þ ð8Þ
kB T

where J min;SCR is the minority charge carrier current into the !


jp  n^ ¼ q  Sp0  ðp  p0 Þ ð9Þ
emitter and V SCR the splitting of the quasi-Fermi potentials at
! !
the edge of the space charge region. The calculated J 0e values are with the electron and hole current densities jn and jp at the
compared to results from Sentaurus TCAD and from the freeware contacts, the surface normal n,^ the elementary charge q, the
simulation tool EDNA 2 [14,15], which is specifically made for surface recombination velocities Sn0 and Sp0 for electrons and
simulation of emitter properties in Si solar cells. A series of holes, the electron and hole concentrations n and p, and their
Gaussian-shaped phosphorus and boron emitters with a surface equilibrium concentrations n0 and p0. In PC1D, the surface recom-
concentration between 1018 and 1021 cm  3 was chosen for the bination rate
simulations. The largest doping densities thus exceed the solubi-
pn  n2i;ef f
lity limit for P in Si [26] for temperatures of typical POCl3 Rsurf ¼  h i  h i ð10Þ
 Ei  Ei
diffusions (  800 1C) [27]. Thus, the heaviest emitters are not p þni;ef f exp  Et kT =Sn0 þ n þ ni;ef f exp Et kT =Sp0
relevant for real-life POCl3 diffused solar cell devices, but poten-
tially for ion implanted regions [28]. The full range of doping levels depends on the effective intrinsic density ni,eff and the surface trap
was also included in order to test the new program under the full energy level Et  Ei and is directly integrated into the discretised
range of conditions. The surface recombination velocity Sp0 electron and hole transport equations, defining the boundary
for phosphorus doped profiles is chosen depending on the surface
doping concentration according to the parameterization of Table 2
Altermatt et al. [1] of the data of Cuevas et al. for Si passivated Details of the device structure used to calculate the curves shown in Figs. 2–4.

with SiO2 [29]:


Emitter profile Gaussian
   
Nsurf γ p1 Nsurf γ p2
S0p ¼ Sp1 þ Sp2 ð7Þ Surface dopant density N surf 1018  1021 (cm  3)
N p1 N p2
Standard deviation xstd 0:2 (mm)
with the parameters Sp1 ¼ 200 cm/s, N p1 ¼ 1019 cm  3, γ p1 ¼ 0:2, Base doping N base 1  1016 (cm  3)
Sp2 ¼ 130 cm/s, N p2 ¼ 1019 cm  3 and γ p1 ¼ 2:4. For the boron Front SRV parameter for majority carriers S0maj;f 1  107 (cm/s)
doped profiles, we use this same parameterization also for Sn0 as Front SRV parameter for minority carriers S0 min ;f Eq. (7)
suggested by Hoex et al. [30] for Al2O3 passivation, but neglect the Rear SRV parameter for minority carriers S0 min ;r 100 (cm/s)
Rear SRV parameter for majority carriers S0maj;r 1  107 (cm/s)
influence of fixed charges for simplicity. The calculated results
Bulk SRH time constants τ0n ; τ0p 1000 (ms)
therefore do not necessarily describe the physics of boron profiles Cell thickness W 180 (mm)
passivated with Al2O3 correctly, but surely serve for a comparison
34 H. Haug et al. / Solar Energy Materials & Solar Cells 131 (2014) 30–36

Fig. 2. Emitter dark saturation current density J0e as a function of surface doping concentration, for a series of n-type (a) and p-type (b) Gaussian emitters with a constant
standard deviation of 0.2 mm. Emitter sheet resistance calculated using cmd-PC1D 6.0 are shown on the top axis. J 0e values are calculated using cmd-PC1D 6.0, EDNA 2 and
Sentaurus TCAD. A clear difference between Boltzmann and F–D statistics is observed for high surface doping density. Note that the EDNA 2 simulation did not converge for
the largest n-type doping density. Bottom part of each subfigure: Relative difference between the results from cmd-PC1D 6.0, Sentaurus TCAD and EDNA 2. Models and
simulation parameters are given in Tables 1 and 2.

conditions [10]. Due to these slightly different formulations of the


boundary conditions, different electron, hole and total currents at
the contacts may possibly occur, depending on the surface
recombination parameters, doping and injections conditions. The
third modification concerns the charge carrier generation profile,
which is implemented in a different way in Sentaurus TCAD than
in PC1D and cmd-PC1D. While the latter use the cumulative
generation rate Gcum ðzÞ, Sentaurus TCAD uses its derivative gðzÞ,
given in units of cm  ³/s. In order to use the same information in
both programs, we first calculate gðzÞ using the Sentaurus ray
tracer for an alkaline textured silicon solar cell coated with a
75 nm thin SiNx anti-reflection coating under 1 sun illumination,
resulting in a total cumulated generation of 41.8 mA/cm². Then, we
determine Gcum ðzÞ from gðzÞ using the trapezoidal rule for the
numerical integration at set this as input for the cmd-PC1D
6.0 simulations.
First, the charge carrier profiles at the highly doped emitter
without illumination at zero bias are compared. As shown in Fig. 3,
an excellent agreement is obtained. In the space charge region, the
product of electron and hole density is even smoother using cmd-
PC1D 6.0 than using Sentaurus TCAD. The scattering of this
quantity might be attributed to the implementation of doping- Fig. 3. Carrier and doping concentrations as a function of depth below the surface
dependent BGN in Sentaurus TCAD. for a Gaussian emitter, calculated at steady state conditions in the dark, with a
constant forward bias of 0.55 V. Results are calculated both using cmd-PC1D 6.0
Next, important parameters of the current–voltage (IV) char-
(solid black lines) and Sentaurus TCAD (dashed red lines) applying the same set of
acteristics of a set of model solar cells under 1 sun illumination parameters and models. The corresponding J 0e values are indicated, showing an
(1000 W/m², AM 1.5 g spectrum) are calculated with cmd-PC1D excellent correspondence between the values calculated with the two programs.
6.0 and are compared to Sentaurus TCAD. The same emitters as (For interpretation of the references to color in this figure legend, the reader is
before and the same models and parameters are applied in both referred to the web version of this article.)

tools (see Tables 1 and 2). As shown in Fig. 4, a good qualitative


and quantitative agreement is observed for all four parameters in caused by the implementation of F–D statistics. Instead, numerical
the whole range of surface doping concentrations. The deviations errors of partly Sentaurus TCAD and partly cmd-PC1D 6.0, together
of the short circuit current and the open circuit voltage are smaller with the slightly different implementations of the boundary
than 0.3%. The deviation of the fill factor is less than 0.8% relative. conditions at the contacts may probably cause the small deviations
Please note that the observed differences occur for both F–D and of the IV parameters. As the surface doping concentration of the
Boltzmann statistics in a similar way, and should therefore not be emitter is varied between 1018 cm  3 and 1021 cm  3, the energy
H. Haug et al. / Solar Energy Materials & Solar Cells 131 (2014) 30–36 35

Fig. 4. Calculated cell parameters: (a) Short circuit current, (b) open circuit voltage, (c) fill factor and (d) conversion efficiency, as a function of surface doping concentration.
A good correspondence is found between the results from cmd-PC1D 6.0 (black squares) and Sentaurus (red circles), both for Boltzmann (open symbols) and F–D statistics
(filled symbols). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

conversion efficiency varies between 23.7% and 18.3%, with a significantly affect the simulated I–V curves, but may be a
maximal deviation of the two tools of less than 0.8 % relative, motivation for including the abovementioned features in a future
corresponding to a deviation of approximately 0.2 % absolute. update and thus obtaining a fully consistent solution.
Furthermore, it should be noted that there exist several
problems with simulation of heavily doped emitters that are not
5. Further work solved by the introduction of F–D statistics alone. Most impor-
tantly, in many industrially relevant emitters formed by POCl3
There is still room for further improvement in the cmd-PC1D diffusion the phosphorus concentrations approaches or exceeds
6.0 program, which hopefully will be addressed in future updates the solubility limit in Si. P can then easily form various forms of
of the program. For example, in F–D statistics, the charge carriers precipitates, which may introduce a significant contribution from
interact with each other due to Pauli blocking. Therefore, the SRH recombination, sometimes even dominating over Auger
Einstein relation relating the diffusion coefficient to the mobility is recombination. In any case, the correct carrier statistics and a
no longer valid, and should be replaced with the following robust infrastructure for adding and modifying models are now in
expression: D=μ ¼ ðkB T=qÞ  ðF1=2 = F  1=2 Þ, where F  1=2 is the place for PC1D, and additional choices for models will be added in
Fermi integral of order—1/2 [31]. Also, an additional term should future versions of the program. Based on both the existing
be added in the transport equations for electrons and holes, in improvements and these future possibilities we expect PC1D
order to avoid any unphysical currents flowing at dark equilibrium simulations to continue to play an important role in the continued
conditions [16]. The effects of this and the abovementioned research effort within in this field.
modification of the Einstein relation are small, and have so far
not been implemented in Sentaurus TCAD. Note also that no
equilibrium current larger than 10  7 mA/cm² has been observed 6. Conclusions
in cmd-PC1D 6.0 so far. A different unphysical effect related to the
use of F–D statistics in cmd-PC1D 6.0 is observed in that the total We successfully verified the implementations of the new
current is not entirely constant in the emitter region for operating physical models for the intrinsic carrier density, the mobility,
conditions close to open circuit. This issue does not seem to Auger recombination and band gap narrowing into PC1D against
36 H. Haug et al. / Solar Energy Materials & Solar Cells 131 (2014) 30–36

several other implementations in Visual Basic, Microsoft Excel and [7] P.A. Basore D.A. Clugston, PC1D version 4 for windows: from analysis to
Sentaurus TCAD. We demonstrated the agreement of the emitter design, in: 25th IEEE Photovoltiac Specialists Conference, Washington, 1996,
pp. 377–381.
saturation currents for several emitters of the new program cmd- [8] D.A. Clugston, P.A. Basore, PC1D version 5: 32-bit solar cell modeling on
PC1D 6.0 with the established simulation tool EDNA 2 and personal computers, in: Photovoltaic Specialists Conference, 1997., Conference
Sentaurus TCAD with deviations being smaller than 5%. These Record of the Twenty-Sixth IEEE, 1997, pp. 207–210.
[9] H. Haug, B.R. Olaisen, Ø. Nordseth, E.S. Marstein, A graphical user interface for
deviations are most probably caused by numerical errors. We
multivariable analysis of silicon solar cells using scripted PC1D simulations,
calculated important parameters of the illuminated current–vol- Energy Procedia 38 (2013) 72–79.
tage characteristics of model solar cells as the short circuit current, [10] PC1D help file. [Online]. Available: 〈http://www.pvlighthouse.com.au/
open circuit voltage, fill factor and energy conversion efficiency Resources/PC1D/〉. [Accessed: 01-Apr-2014].
[11] P. Würfel, U. Würfel, Physics of Solar Cells: From Basic Principles to Advanced
with cmd-PC1D 6.0 and Sentaurus TCAD. A good agreement of Concepts, Wiley (2009) 44.
both tools is reached. The observed small relative deviations of less [12] P. Van Halen, D.L. Pulfrey, Accurate, short series approximations to Fermi–
than 0.8% relative are probably caused by slightly different Dirac integrals of order  1/2, 1/2, 1, 3/2, 2, 5/2, 3, and 7/2, J. Appl. Phys. 57 (12)
(1985) 5271.
boundary conditions and different working modes of the BGN in
[13] H.M. Antia, Rational function approximations for Fermi–Dirac integrals,
the two programs. In summary, the correct implementation of Astrophys. J. Suppl 84 (1993) 101–108.
both F–D statistics and new physical models in cmd-PC1D 6.0 has [14] K.R. McIntosh, P.P. Altermatt, A freeware 1D emitter model for silicon solar cells,
been successfully demonstrated. in: IEEE 35th Photovoltaic Specialists Conference, 2010, pp. 002188–002193.
[15] K.R. McIntosh, P.P. Altermatt, T.J. Ratcliff, K.C. Fong, L.E. Black, S.C. Baker-Finch,
The cmd-PC1D 6.0 program can be run from a command line, M.D. Abbott, An examination of three common assumptions used to simulate
which gives additional flexibility for simulations. It is also pro- recombination in heavily doped silicon, in: Proceedings of the 28th European
vided with a new and modern graphical user interface called PC1D Photovoltaic Solar Energy Conference and Exhibition, Paris, France, 2013,
pp. 1672–1679.
for Matlab, which can be run as a standalone executable. The PC1D
[16] P. Altermatt, Models for numerical device simulations of crystalline silicon
for Matlab interface gives easy access to changing between the solar cells—a review, J. Comput. Electron 10 (3) (2011) 314–330.
different models described in this work, as well as the original [17] P.P. Altermatt, A. Schenk, F. Geelhaar, G. Heiser, Reassessment of the intrinsic
PC1D models. It also has many built-in possibilities for data carrier density in crystalline silicon in view of band-gap narrowing, J. Appl.
Phys. 93 (3) (2003) 1598.
visualization and advanced analysis, including optimization and [18] A.B. Sproul, M.A. Green, Intrinsic carrier concentration and minority-carrier
fitting to experimental data. Both the cmd-PC1D 6.0 program and mobility of silicon from 77 to 300 K, J. Appl. Phys. 73 (3) (1993) 1214.
PC1D for Matlab are freely available for download and are [19] A. Richter, S.W. Glunz, F. Werner, J. Schmidt, A. Cuevas, Improved quantitative
description of Auger recombination in crystalline silicon, Phys. Rev. B:
currently hosted by PV Lighthouse (http://www.pvlighthouse.
Condens. Matter 86 (16) (2012) 165202.
com.au/). The source code of cmd-PC1D 6.0 will be available at [20] M.J. Kerr, A. Cuevas, General parameterization of Auger recombination in
www.sourceforge.net. crystalline silicon, J. Appl. Phys. 91 (4) (2002) 2473–2480.
[21] M.A. Green, Intrinsic concentration, effective densities of states, and effective
mass in silicon, J. Appl. Phys. 67 (6) (1990) 2944.
[22] D.B.M. Klaassen, A unified mobility model for device simulation—I. Model
Acknowledgements
equations and concentration dependence, Solid State Electron. 35 (7) (1992)
953–959.
Halvard Haug wants to thank the University Graduate Centre [23] D.B.M. Klaassen, A unified mobility model for device simulation—II. Tempera-
ture dependence of carrier mobility and lifetime, Solid State Electron. 35 (7)
(UNIK) and The Norwegian Research Centre for Solar Cell Technol-
(1992) 961–967.
ogy for funding. Achim Kimmerle acknowledges the scholarship [24] A.L. Blum, J.S. Swirhun, R.A. Sinton, A. Kimmerle, Updated analysis to the WCT-
support of the Reiner-Lemoine Foundation. 120 QSSPC measurement system using advanced device physics, in: Proceed-
ings of the 28th European Photovoltaic Solar Energy Conference and Exhibi-
tion, Paris, France, 2013.
References [25] Synopsis Inc, Sentaurus Device, Version H-2013.03. Mountain View, California,
USA.
[1] P.P. Altermatt, J.O. Schumacher, A. Cuevas, M.J. Kerr, S.W. Glunz, R.R. King, [26] S. Solmi, A. Parisini, R. Angelucci, A. Armigliato, D. Nobili, L. Moro, Dopant and
G. Heiser, A. Schenk, Numerical modeling of highly doped Si:P emitters based carrier concentration in Si in equilibrium with monoclinic SiP precipitates,
on Fermi–Dirac statistics and self-consistent material parameters, J. Appl. Phys. Rev. B: Condens. Matter 53 (12) (1996) 7836–7841.
Phys. 92 (6) (2002) 3187. [27] A. Kimmerle, A. Wolf, U. Belledin, D. Biro, Modelling carrier recombination in
[2] A. Schenk, Finite-temperature full random-phase approximation model of band highly phosphorus-doped industrial emitters, Energy Procedia 8 (2011)
gap narrowing for silicon device simulation, J. Appl. Phys. 84 (7) (1998) 3684. 275–281.
[3] D. Yan, A. Cuevas, Empirical determination of the energy band gap narrowing [28] J. Wu, Y. Liu, X. Wang, L. Zhang, Application of ion Implantation Emitter in
in highly doped n þ silicon, J. Appl. Phys. 114 (4) (2013) 044508. PERC Solar Cells, IEEE J. Photovoltaics 4 (1) (2014) 52–57.
[4] D.T. Rover, P.A. Basore, G.M. Thorson, Solar cell modeling on personal [29] A. Cuevas, P.A. Basore, G. Giroult-Matlakowski, C. Dubois, Surface recombina-
computers, in: Proceedings of the 18th IEEE Photovoltaic Specialists Confer- tion velocity of highly doped n-type silicon, J. Appl. Phys. 80 (6) (1996) 3370.
ence, Las Vegas, Nevada, USA, 1985, pp. 703–709. [30] B. Hoex, J. Schmidt, R. Bock, P.P. Altermatt, M.C.M. van de Sanden, W.M.
[5] P.A. Basore, D.T. Rover, A.W. Smith, PC-1D version 2: enhanced numerical solar M. Kessels, Excellent passivation of highly doped p-type Si surfaces by the
cell modelling, in: Proceedings of the 20th IEEE Photovoltaic Specialists negative-charge-dielectric Al2O3, Appl. Phys. Lett. 91 (11) (2007)
Conference, Las Vegas, Nevada, USA, 1988, pp. 389–396. 112103–112107.
[6] P.A. Basore, PC-1D Version 3: improved speed and convergence, in: IEEE [31] J.S. Blakemore, Approximations for Fermi–Dirac integrals, especially the
Photovoltaic Specialists Conference, 1991, Conference Record of the Twenty function used to describe electron density in a semiconductor, Solid State
Second, vol. 1, 1991, pp. 299–302. Electron. 25 (11) (1982) 1067–1076.

You might also like