Proceedings of the 2009 Spanish Conference on Electron Devices - Feb 11-13, 2009. Santiago de Compostela, Spain.

Parameter Extraction of a Solar Cell Compact

Model usign Genetic Algorithms
Nicolai Moldovan', Rodrigo Picos2, SM, Eugenio Garcia-Moreno2, SM
I
Universitatea Tehnica Cluj - Napoca, Cluj-Napoca, Romania,
2
Universitat de les Illes Balears, Palma, Spain, (rodrigo.picosguib.es)

photovoltaic (OPV) solar cells. One topic that appears

Abstract- We present a method to extract the model frequently is related to the difficulty encountered in extracting
parameters of illuminated solar cells based on genetic algorithms, the parameters of the junction, quality factor and reverse
The solar cell model is the well known lumped parameter current, when there are a significant series and shunt parasitic
equivalent circuit containing four components. Its mathematical
description has been made explicit using of the Lambert function.
.s
resistances. .

We have tested the ability of our method against a direct In this work we propose using genetic algorithms (GA) as
extraction method to give back some known results. The test the extraction method. This technique (GA) has been
consists in fitting I-V characteristics of an organic photovoltaic extensively used in FET modeling ([8],[9],[10]). The main
solar cell obtained from randomly generated parameters. The advantage of using it instead of other global extraction
accuracy and the reliability of the results are much higher, methods is that no assumption is made about the values of the
particularly concerning the reverse saturation current. parameters.
The time needed to perform the extraction of each set of parameters.
parameter is around 100s in a PC with a dual core processor, at
3.5GHz with 2GB RAM.
The paper iS organized in five sections. After this
introduction, Section 2 presents the solar cell model. Section 3
describes the parameter extraction procedure. Section 4
I. INTRODUCTION compares the ability of our procedure with a direct one in
Description of illuminated solar behavior, as it is required in extracting the parameters of an OPV solar cell. Finally,
simulating generation systems, can be achieved by a lumped Section 5 provides the main conclusions of the paper.
parameter equivalent circuit containing four elements: a light-
generated current source, a p-n junction diode and two
parasitic resistances [1]. Essential system performance
parameters like fill factor, short circuit current or open circuit Most semiconductor solar cells are based on single p-n
voltage can be extracted from the model for different operating junctions. Description of cell behavior can therefore be given
conditions. Thus it is important being able to extract the in terms of a very simple model based on the diode equation
parameter values. In fact, being able to perform a correct [1]. Applying superposition principle the cell current
parameter extraction is a great step in assessing the capabilities depending upon the bias voltage and the illumination
of a particular technology. conditions may be expressed as
The mathematical description of the current voltage (I-V)
characteristics in this circuit has been given traditionally by an Ic =Dk (bias) + IPh (light) (1)
implicit transcendental equation, but it can be explicitly
described ([1], [2]) using the Lambert function. However, it The light-generated current, IPh, is independent of the bias
still remains unsuitable for the purpose of extracting the model conditions. The dark current, IDk, is related to the junction
parameters directly by numerical fitting. voltage, V., by the non-ideal diode equation
Many different methods have been proposed ([4],[5],[6],[7])
to extract the model parameters from the measured I-V IDk = [exp(Vj / n VLh)-1] (2)
characteristics. Most of them represent approximate
procedures to make the I-V characteristics analytically where I, is the reverse saturation current, n the quality factor
manageable, extracting each parameter from restricted regions and Vth is the thermal voltage, Vh = kT / q . Different
of the I-V characteristics where some other parameter is
assumed to be negligible. These approaches may work as long vausothmdeprmtrsnndI.Trfreanie
as distinct regions are actually present in the characteristics, assignment of fit results corresponding to the physical origin
This iS not the case when considerable parasitic series Of one recombination mechanism is not possible.
resistance or shunt conductance is present, as in organic

978-1l -4244-283 9-7/09/$25 .00 (C)2009 IEEE 379

 Proceedings of the 2009 Spanish Conference on Electron Devices - Feb 1 1-13, 2009. Santiago de Compostela, Spain.
In real devices, a series resistance RS, and a shunt resistance V = (IPh +IO - I)RP -IRS -
RP may be present. The series resistance is composed of the I R R (IPh +IO -I) (5)
resistance of the metal grid, the contact resistance and the base - nVhW[ exp( nV (5
and emitter sheet resistances. Shunt conductive loses can arise
n
Vh nh
from imperfections on the device surface as well as leakage were "W" represents the usual short-hand notation for the
currents across the edge of the cell.
Accounting for series and shunt resistance losses, the p
illuminated I-V characteristic eq. 1 becomes
III. PARAMETER EXTRACTION
I=I-O[ V + JR V + IR(3) The above explicit representations are good computational
II I ( R) alte]atives to using iterative solution of the original
P
transcendental equation (3) in solar cell models. This is
where I and V are the current and voltage at the cell terminals. because there are efficient implementations of the Lambert
Fig. 2 shows the equivalent circuit of eq. 3, which consists function available into the mathematical libraries of nearly
of a diode with reverse saturation current IO and quality factor every major computer language. However, in spite of the fact
n, the light generated current 'Ph and the series and shunt that this is an explicit analytical expression, it still remains
resistances Rs and RP. By varying these parameters a wide unsuitable for the purpose of extracting the model parameters
range of experimentally observed I- V curves can be fitted. directly by numerical fitting.

Rs I A. Direct Methods
Most of the methods proposed ([4],[5],[8]) to extract the
model parameters from the measured current-voltage
}7AVW' nRLi characteristics, represent approximate procedures to make the
f IPh RP RL I-V characteristics analytically manageable, extracting each
I ~~~parameter from restricted regions of the I-V characteristics
l__________________ _where some other parameter are assumed to be negligible.

Fig. 1 Generic solar cell equivalent circuit including a load resistance RL. 10 Region 11

Is well known that the above implicit transcendental 8 -

equation (3) may not be solved explicitly in general for I or V
using common elementary functions. However, nowadays 6
exact explicit analytical solutions already exist [3, 4]. These 4
solutions make use of what is known as the Lambert W
function [5], a special function which is not expressible in 2
terms of elementary analytical functions. The Lambert W ,-,Il_l_l_l_l_l_l_V(V)
function is defined as the solution of the equation -0.2 0.2 0.4 0.6
W(x) exp[W(x)] =x .This function has been used for solving -2 -
basic diode and bipolar transistor circuit analysis problems [6] Fig.2 I-V characteristics of an organic solar cell under AM 1.5 illumination
as well as in device modeling formulations [7]. with 100 mWcm-2 [9]. The two distinct regions for parameter extraction are
This approach has also been recently used to study marked..
illuminated solar cell [8]. To do that, the authors propose using In Fig. 2, we show the most typical assumed regions,
the same circuit in Fig. 1, and solving eq. 3 in terms of the
Lambert function. Thus, the solutions for each variable I and V tm bee off,aRgn
' ~~~~~~~~~to and the and circuit iStheonly a current diode source assumed
and two
as an explicit function of the other and of the device model resistances. In this region we can extract the values of the
parameters are: resistances, and the photocurrent. In Region II, the diode is
assumed to be on, and it is dominating the behavior of the
I..= th[ WE l'RP circuit. Thus, from this region we get the values of the quality
R, nVIh (R5 + RP) (4) factor n and the reverse saturation current of the diode (Ia,).
exp(pPhR +IR+V)) V -('Ph + Io)Rp Note that there is no definite separation between these two
nUth(RS +RP) RS + RP regions, and they must be defined by the user.
These approaches may work as long as distinct regions are

978-1-4244-2839-7/09/$25.00 (C)2009 IEEE 3 80

 Proceedings of the 2009 Spanish Conference on Electron Devices - Feb 1 1-13, 2009. Santiago de Compostela, Spain.
actually present in the characteristics. This is not the case sets, based on a known cell, and varying each parameter inside
when considerable parasitic series resistance or shunt a + 20% range using a uniform distribution. The original
conductance is present, as in organic photovoltaic (OPV) solar parameters are listed in Table I, they correspond to an organic
cells [10]. In this case, the assumption of separated regions is photovoltaic solar cell [9], with Jsc = 1.15 mA/cm, VOC of
not valid anymore, and it is found that the parameters are 0.580 V and fill factor of 64%, giving a power conversion
strongly correlated. efficiency of 4.05%. For each parameter set Pi, we generated
the I-V characteristics, and from those characteristics, we
B. Genetic Algorithms
extracted the corresponding set of parameters
pE1.
To solve the above problem and provide accurately TABLE I
extracted parameters, in this work we propose using genetic MODEL PARAMETERS FROM A KNOWN SOLAR CELL [9], (PER CM2)
algorithms (GA) as the extraction method. This technique Parameter Value
(GA) has been extensively used as global parameter extraction
technique in FET modeling [10- 1], due to its known 'ph 11.2 (mA)
advantages (global minimum finding, ease of coding, etc ... ). Io 1.35 (uA)
The term GA includes a very broad family of methods; the n 2.5
one used in this work is known as crossing-mates. Its core idea Rs 0.5 Q
is depicted in Fig. 3. The process starts from a seed, and a set Rp 1KQ
of individuals (a population) is generated. Each of these
individuals is a list of values of the parameters that are to be
extracted, and are different for all the individuals. Then, the The extraction was performed using both the direct method
best fitted individuals, in the sense of minimizing a given described in the previous section, and the GA method. We
measure or error, are selected to form the basis a new then calculated the error (e) as the difference between the
population, and the process is repeated until error falls below a extracted and the original parameter value. Finally, to compare
given threshold. In our case, we use an error function both extraction methods, we have calculated the mean error
fi defined as the difference between the measured and the (/O) 0 and the standard deviation (cG) of the error for each
expected values of the current, in a lineal scale: parameter.

A ZL ('measured - 'ca/cu/ated )2 (6) TABLE II

measurements MEAN ERROR (6) AND STANDARD DEVIATION (CT) FOR THE
DIFFERENT PARAMETERS USING THE DIRECT EXTRACTION METHOD
Parameter E (%) 6 (%)
|Seed | Iph 0.1 0.01
+ Io 74.2 7.00
|Generate population n 6.7 1.35
from seed
Rs -0.54 0.23
Replace population R -0.1 0.01
with cross result Rp_______ ____
Seek best fitted
individual Table II shows the results obtained applying the direct
Cross process
extraction procedure. As can be seen from these results, this
Is ~ procedure is quite efficient in the extraction of the parameters
fitting good No
N

enough?
Rp, Iph, with errors less than 1%. The error in the other
variables is bigger, getting as high as a 74.2% in the case of Io.
Yes This is partially due to the fact that the parameter extraction
End procedure sensible
implies to using
small exponential
in the functions
data. that are
extremely changes input Another
Eig. 3. Elow diagram for the GA crossing-mates algorithm source of problems is that, for some sets of parameters, there
are strong correlations among them, thus tearing down the
IV. RESULTS assumption of two separate regions. In addition, using this
In oderto hec th ablit of ur ethd t acuraely procedure to extract parameters from experimental data will
probably result in bigger errors, due to the precision problems
extract the parameters of a given solar cell, we first checked itS ascaewt esrmns
capability to give back some known results. To do that, we
generated a random set of 1000 different solar cells parameter

978- 1-4244-2839-7/09/$25.00 (C)2009 IEEE 381

 Proceedings of the 2009 Spanish Conference on Electron Devices - Feb 11-13, 2009. Santiago de Compostela, Spain.
[3] A. Jain, S. Sharma, and A. Kapoor, "Solar cell array parameters using
TABLE III Lambert W-function," Solar Energy Materials and Solar Cells, Vol. 90,
MEAN ERROR (g) AND STANDARD DEVIATION (CT) FOR THE no. 1, pp. 25-31,2006.
DIFFERENT PARAMETERS USING THE GENETIC ALGORITHM METHOD [4] J. Lambert, Acta Helv. Phys. Math. Anat. Bot-Med, Vol. 3, pp. 3-128,
1758.
Parameter E (%) 6 (%) [5] A. Cheknane, H. Hilal, F. Djeffal, B. Benyoucef, and J. Charles, "An
Iph 0.01 0.001 equivalent circuit approach to organic solar cell modelling,"
Microelectronics Journal, 2008.
I, 0.2 0.006 [6] T.C. Banwell, "Bipolar transistor circuit analysis using the Lambert W-
n 0.2 0.04 function", IEEE Trans. on Circuits and Systems I, vol 47, no. 11, pp
1621-1633, 2000
Rs -0.04 0.001 [7] A. Lizaro, B. Nae, B. Ifiguez, F. Garcia, I. M. Tienda-Luna, and A.
Rp -0.1 0.001 Godoy, "A compact quantum model for fin-shaped field effect transistors
valid from dc to high frequency and noise simulations", J. Appl. Phys.
103, 084507, 2008
Table III shows the results obtained using the GA method. It [8] J. Ding and R. Radhakrishnan, "A new method to determine the optimum
can be seen that the recovering of original data is nearly load of a real solar cell using the Lambert W function," Solar Energy
Materials and Solar Cells, 2008.
flawless. The only problem is that the time needed to perform [9] C. N. Hoth, P. Schilinsky, S. A. Choulis, and C. J. Brabec, "Printing
the extraction of each set of parameter is quite high: 98s + 26s highly efficient organic solar cells," Nano Letters, vol. 8, no. 9, pp.
in a PC with a dual core processor at 3.5GHz, and 2GB RAM, 2806-2813, 2008.
[10] M. Murakawa, M. Miura-Mattausch, and T. Higuchi, "Towards
running XP. This may be not an issue in a research context, automatic parameter extraction for surface-potential-based mosfet
but may prove to be too slow if the purpose is checking out the models with the genetic algorithm," in ASP-DAC '05: Proceedings of the
quality of the fabricated cells in an industrial process. In this 2005 conference on Asia South Pacific design automation. New York,
last case, the best solution would be probably using this NY, USA: ACM, 2005, pp. 204-207.
[11] E. Garcia-Moreno, B. Iniguez and R. Picos, "Directed Genetic algorithms
procedure only to test a statistically significant sample of the for OTFT model parameter extraction", in Procs. of 9th International
whole batch, and not every cell. Conference on Solid-State and Integrated-Circuit Technology, Peking,
China, 2008, pp 424-427
V. CONCLUSION
A new method to extract the parameter values of a
photovoltaic solar cell model from the measured I-V dc
characteristics has been set up. The method, based on
crossing-mates genetic algorithm, uses an explicit formulation
of the I-V dc model characteristics through the Lambert
function. Applied to the solar cell model, our method gives
much more accurate and reliable results than the direct
extraction methods, mainly in cases where the parameters are
strongly correlated, as it occurs in low power conversion
efficiency solar cells.
The comparison has been carried out by a give back
method, i.e., in extracting the model parameter values by
fitting a set of I-V curves generated from known parameter
values. The only drawback of the technique is that it needs
around 100s to perform the extraction in a dual core processor
PC at 3.5GHz running XP. But note that, opposite to direct
extraction methods, this one does no require human
intervention.

ACKNOWLEDGMENT
This work was supported by the European FEDER Fund
and the Spanish CICYT under Project TEC06/04103.

REFERENCES
[1] A. L. Fahrenbruch, R. H. Bube, "Fundamentals of Solar Cells,"
Academic Press, New York, 1983.
[2] A. Jamn and A. Kapoor, "A new approach to study organic solar cell
using Lambert W-function," Solar Energy Materials and Solar Cells, Vol.
86, no. 2, pp. 197-205, 2005.

978-1 -4244-2839-7/09/$25.00 (C)2009 IEEE 382