Vous êtes sur la page 1sur 6

Challenges and Opportunities in Cell and Module I-V Testing

Ronald A. Sinton, Adrienne L. Blum, Justin Dinger, Wes Dobson, Cassidy Sainsbury
and Harrison W. Wilterdink

Sinton Instruments, Boulder, CO, 80301, USA

Abstract — The most basic and important characterization of a an approach of relegating details of traceability and absolute
solar cell is the power output. Within industry, the I-V test is measurement accuracy to the calibration laboratory, resulting
perhaps the only detailed characterization that is uniformly in relatively simple procedures for production line assessment.
performed on 100% of the product. At calibration laboratories,
this is the basis for benchmarking R&D with efficiency records Results from an experiment demonstrating this principle are
and the source of traceability for a ~100B US$ industry. This paper described. Second, the opportunities for implementing device
compares and contrasts methodologies used in calibration physics models into the production testing in order to quickly
laboratories with production facilities. It will highlight optimize the production with root-cause identification of the
opportunities to optimize the information available in production power losses are presented. Third, the relevance of letter-grade
in order to best serve both the requirements of accurate nameplate
ratings and provide relevant information for process control and simulator ratings to this process are briefly discussed.
continuous improvement of the crystalline silicon production.
Specific topics will include minimizing uncertainties in reported
values and advanced characterization of the parameters relevant II. CALIBRATION LABORATORY VS. PRODUCTION TESTING
to full device physics models for root-cause identification of Table I illustrates some of the key uncertainties for traceable
process variation.
measurements of the power of a cell or module at standard test
Index Terms — silicon, measurement, photovoltaic cells,
capacitance. conditions (STC). Uncertainties come from several sources,
and include both systematic and random effects. For contrast,
consider a module production line, where traceability is
I. INTRODUCTION determined essentially by a transfer of measured power from a
Traditionally, cells are measured in production in order reference module to production modules. Systematic effects on
to rate them for efficiency and bin them in order to optimize the left-hand side of Table I may impact production line
module production. Modules are primarily rated for power, measurements; however, the systematic uncertainty associated
representing the value of the production. The framework for with each effect ratios out of the total as long as the effect is
these measurement procedures has often come from calibration identical for the reference module and production modules.
laboratories, with an emphasis on traceability of measurements. This condition is generally satisfied when the reference module
The uncertainty analysis for this process is very complex, as the same product as the production modules. In this case, the
shown in Table I. In contrast, solar cell R&D has developed production line uncertainty assessment is vastly simplified:
detailed characterization techniques based on device physics total uncertainty will depend primarily on the reference module
understanding. This differs from the traceability approach in uncertainty (treated as a systematic uncertainty by the
that the measurements use detailed device physics models that production line) and random uncertainties added at the factory
are specific to the material and device designs. The results can (which can be determined by reproducibility testing on site).
include a detailed model of the devices under test, instead of a This principle of ratioing out select uncertainties is well
simple report of current and voltage at one sun conditions. known, for example, with spectral mismatch effects. In the
This paper will present three related concepts. First, we calibration laboratory, there are uncertainties that must be
discuss methods for production line uncertainty analysis, with calculated based on the spectrum of the light source, the
TABLE I: CALIBRATION OF REFERENCE MODULE IN CALIBRATION LABORATORY VS. PRODUCTION POWER RATING
Reference Calibration Laboratory Production Line Power Rating
Irradiance accuracy and traceability (e.g., WPVS reference cell) Calibration module accuracy (from calibration laboratory)
Voltage accuracy and traceability Reproducibility of testing
Current accuracy and traceability
Temperature uniformity, accuracy, and traceability
Illumination non-uniformity
Spectrum measurement accuracy
EQE reference cell and EQE device under test
Reproducibility of testing

978-1-5386-8529-7/18/$31.00 ©2018 IEEE 0009


spectral response of the reference device that is used to There are several widely-cited references [7]-[8] that assume
determine light intensity, and the spectral response of the a one-to-one relationship between the non-uniformity value and
module under test. It is well known that in the factory, use of a the associated uncertainty in module power—i.e., a non-
reference module with the same size, construction, and nominal uniformity of ±4.9%, for example, would set the minimum
spectral response as the production modules makes the spectral uncertainty in power to ±4.9%. This assumption seems to be
mismatch correction unnecessary [4]. This principle is less well widespread in the general understanding of module power
known with respect to some of the other uncertainties in measurements, but is not present in the data shown here, or
Table I, especially the illumination non-uniformity [1]-[7]. predicted in detailed modeling [1]-[6], [17].
Fig. 1 has 300 data points (100 for each non-uniformity
pattern) that can be visualized more clearly: by normalizing
III. DEMONSTRATION OF PRODUCTION TOLERANCE
each module’s measured power values to the nominal “silver”
We have encountered skepticism that the principle described power value of that module, the data can be aggregated to show
in the last section applies to illumination non-uniformity. An the differences resulting from each non-uniformity pattern. The
experiment was designed to test this hypothesis, and will be resulting frequency histogram is shown in Fig. 2. The
presented at the European Conference in September [1]. Ten distributions are nominally Gaussian and have a standard
modules were sampled from production, and the power of each deviation of 0.05%, representing the measurement
was measured using a solar simulator calibrated using power reproducibility of this simulated production line scenario. It can
match with a single reference module (“gold module”); this
provided a baseline power value for each sample (“silver
module”). One-by-one, the solar simulator was calibrated to a
silver module and all ten modules were measured again, until
each module had been used once as a silver module (100
measurements total). The key idea in this experiment is that if
systematic uncertainties ratio out, then the difference in module
power measured under each silver module calibration should be
dominated by random effects which will be statistically
insignificant under rigorous analysis. This outcome was
confirmed using an analysis of variance (ANOVA) approach.
To test the ratioing principle for illumination non-uniformity,
we repeated the same test procedure for three different non-
uniformity patterns: ±2.2%, ±3.4%, and ±4.9%. This simulates
data that one might expect on a production line where the solar
simulator non-uniformity drifts over time, or if comparing
measurements between production lines using different solar Fig. 1: Measured power for 10 modules. Each sample ID includes
simulators. Note—non-uniformity is determined by the power measurements relative to 10 different calibration modules.
Color coding differentiates the non-uniformity patterns.
maximum and minimum irradiance measured in the module
area: ሺ‫ܩ‬௠௔௫ െ ‫ܩ‬௠௜௡ ሻȀሺ‫ܩ‬௠௔௫ ൅ ‫ܩ‬௠௜௡ ሻ.
The result is shown in Fig. 1 for the ten modules. The module
power data shows expected production variability, spanning
265 W to 271 W. However, the data from each of the three non-
uniformity experiments aligns very well. The average power of
the ten modules for the three different non-uniformity cases was
268.20 W, 268.11 W and 268.11 W for the three cases of
different non-uniformities. The maximum difference between
the three cases was 1 part in 3000, despite the large range of
non-uniformity represented in this experiment.
This result agrees with a review of the literature for model
and data-based experiments on the effects of non-uniformity
and cell mismatch [1]-[6]. It is well appreciated in this literature
that there will be little effect of non-uniformity on the power
measurement for non-uniformity less than 5%. Additionally, in
this emulation of a production measurement, the non-
uniformity effect is present for both the reference module and Fig. 2: Normalized data showing the normalized spread in
measurements for the entire data-set for the three non-uniformity
the device under test, and so will ratio out. patterns. Color coding differentiates the non-uniformity patterns

978-1-5386-8529-7/18/$31.00 ©2018 IEEE 0010


be concluded that for this experiment, the “production”
measurement introduced about 0.05% uncertainty to the
measurement in addition to whatever uncertainty was present
in the golden module.

IV. DISCUSSION OF NOTABLE EXCEPTIONS


The previous section presented the advantages of using a
transfer standard of a power from a calibration reference to the
device under test in order to simplify measurement procedures
and minimize uncertainties. There are complications to this
method that are worth noting. Two will be discussed here in the
context of production cell measurements.
Silver is one of the most expensive components in modern
silicon solar cells. As a result, there is less and less of this silver
on the front of solar cells in order to minimize costs. This leads
to the necessity of extensive probe bar shading, with numerous Fig. 3: An illustration of the errors possible when ramping voltage
up or down during a flash measurement on an n-type solar cell
current and voltage probes to extract the current while
[12]. The cell has “extra” current when ramping down in voltage
maintaining the solar cell at a uniform lateral voltage. and a deficit when ramping up.
In order to have a uniform definition of cell efficiency,
procedures have been developed at calibration laboratories to simply normalize this with a correction factor to match the
measure these cells [9]-[11]. Generally, the short circuit current calibration laboratory result.
is measured with probing as close to the edge of the cell as This could be problematic in the case of modern solar cells
possible, using negligible shadowing, in order to obtain an with significant capacitance. Fig. 3 shows a modelled I-V curve
unambiguous value for the short circuit current. That is, a value for an n-type solar cell measured with a steady-state tester
that is independent of probe bar shadowing. Then, the probe (center curve) compared to measurements using a ramped
bars are put down onto the cells (creating extensive shadowing) voltage from short-circuit to open circuit (red) or Voc to Isc, blue.
but normalized out by increasing the light intensity to match the The maximum power, and FF, will be different in each case.
measured Isc. The probe bars are carefully optimized so that the The discrepancies in power are well-described as a function
measured voltage corresponds to the case of a uniform average of the charge in the cell, and the ramp rate of the junction
voltage across the cell [9], [11]. This procedure has the desired voltage. There is extra current (dQ/dt) when the cell is
effect of being unambiguous. It can be reproduced at any discharging from high voltage, or a deficit in current when the
calibration laboratory. However, it results in an upper bound of cell is charging from low voltage.
the Isc of the solar cell, since tabbing will eventually increase
݀ܳ ݀ܳ ܸ݀௝
the shadowing, and also an optimistic measure of the power, ൌ (1)
since the probing configuration eliminates the effects of lateral ݀‫ܸ݀ ݐ‬௝ ݀‫ݐ‬
series resistance in the busbars and tabbing that will be present
Where dQ/dVj is the “differential capacitance” of the device and
in the final solar cells when current is extracted only from the
dVj/dt is the junction voltage sweep rate used in the
ends of the tabbing.
measurement of a device. The differential capacitance is a well-
The problem of interest here though, is how to “transfer” this
defined function of cell thickness, substrate doping, and series
idealized result to a production tester. The production tester has
resistance; its value can be readily computed when these
significant shadowing from the probe bars, uses a single
parameters are nominally known or measured. The full
measurement instead of separate Isc and power measurements,
expression of dQ/dVj is defined as [12], [13]:
and has less ideal placement and number of pins in the probe
bars to save costs and maximize throughput. In addition, while ‫ܸݍ‬
݀ܳ ‫ݍ‬ ‫݊ݓݍ‬௜ ଶ ݁‫ ݌ݔ‬൬ ௝ൗ݇ܶ൰
calibration laboratries are often steady-state testers, production
ൌ (2)
testers are usually flash testers that measure the entire IV curve ܸ݀௝ ݇ܶ ‫ܸݍ‬௝
as rapidly as possible. ටܰ஺ǡ஽ ଶ ൅ Ͷ݊௜ ଶ ݁‫ ݌ݔ‬൬ ൗ ൰
݇ܶ
With industrial probe bars in place, the calibration of the cell
tester can be set by matching Isc of the reference calibration cell. Vj is the junction voltage of the device and is related to the
The Voc should be relatively good since this is relatively terminal voltage (V), current (J) and series resistance (Rs) by:
insensitive to the shadow, as long as the Isc is the same [9]-[12].
However, how should the power, especially the Fill Factor ܸ௝ ൌ ܸ ൅ ‫ܴܬ‬௦ (3)
(FF), be matched? Our experience is that the many factories

978-1-5386-8529-7/18/$31.00 ©2018 IEEE 0011


This is accomplished by measuring more than just the current
and voltage. If in addition the Suns-Voc curve and the substrate
doping is determined, this enables a full device physics module
to be used. The Suns-Voc, shown in Fig. 4, can be used with
the IV curve to determine the power loss at the maximum power
point, the current-series resistance product.
When the substrate doping is also determined during the cell
or module test, then the Suns-Voc curve can be interpreted in
terms of carrier density vs. effective lifetime, as is done in a
typical photoconductance measurement on wafers in and R&D
process optimization. For n-type solar cells or modules, this
data can be further analyzed to separate bulk SRH
recombination and emitter saturation current densities. Each
experimental parameter required for a high-efficiency device
physics model: voltage (V), current (J), short circuit current
(Jsc), doping (NA), carrier recombination lifetime (τeff), excess
Fig. 4: Suns-Voc and IV data measured on a production cell tester. carrier density (Δn), and shunt (Rsh) are measured and reported
at cell test and can be expressed with the following equation:
Every solar cell in a production line will have a different
transient response to the cell I-V tester, due to the dependence ݇ܶ ሺ‫ܬ‬௦௖ െ ‫ ܬ‬െ ܸȀܴ‫݄ݏ‬ሻ൫ܰ஺ ൅ ο݊ሻ߬௘௙௙
of dQ/dt on the substrate doping (which varies from cell to cell) ܸ െ ܴ௦ ‫ ܬ‬ൌ ݈݊ ቈ ቉ (4)
‫ݍ‬ ‫ܹ݊ݍ‬௜ ଶ
as well as the series resistance. So a single calibration reference
can’t be used to “normalize” a FF for a steady-state Where the effective lifetime (τeff) can be expressed with the
measurement of the reference cell, unless the production tester following equation further separating the effective lifetime into
has no errors due to capacitance. Production test errors will not its constituent components:
be normalized out correctly and could result in higher or lower ‫ܬ‬௢௘௙௥௢௡௧ ൅ ‫ܬ‬௢௘௕௔௖௞
ͳ ͳ
efficiencies relative to the correct value. The production tester ൌ ൅ ሺܰ஺ ൅ ȟ݊ሻ (5)
߬௘௙௙ ߬௕௨௟௞ ‫݊ݍ‬௜ଶ ܹ
needs to be error-free with respect to capacitance.
This then permits us to calculate the power loss due to the
IV. ADVANCED IV TESTING FOR ROOT CAUSE IDENTIFICATION various specific loss parameters; Rsh, Rs, emitter and bulk
recombination. An outline of several reference points for this
The previous sections described some key issues in the study is shown in Fig. 5 and Fig. 6. There are two relevant
production measurements of cells and modules. Another area reference points for a power loss study. The first is the ultimate
of recent interest is using the cell and module-test station to do limit for the particular silicon substrate thickness. For this, we
advanced measurements on the cells and modules in order to use the limits defined by Richter et al., based on fundamental
enable a root-cause analysis and identification of the individual Auger and radiative recombination, light trapping, and infra-red
cell or module characteristics that are limiting the efficiency parasitic absorption that competes with photogeneration to limit
[14].

Fig. 5 (left) and Fig. 6 (right). Ideal limits for solar cell performance to allow for reference points for a power loss analysis. The “Ideal Limit”
curve is shown at left and the Suns-Voc ultimate limit curve is shown at right.

978-1-5386-8529-7/18/$31.00 ©2018 IEEE 0012


The technology losses are shown relative to cell power output
in the histogram found in Fig. 7. For this technology, the Rs loss
is very low, at 2%, and very well controlled. The shunt loss is
negligible. The bulk recombination and emitter recombination
losses are more significant, at 6% of the power loss, with a bit
of a yield tail of higher losses showing process variation in this
parameter.
Fig. 8 shows the same analysis for a group of monocrystalline
PERC cells. In this case, the global parameter are a Vmp limit of
88% of the ideal limit, and a Jsc limit of 84% of ideal. Then the
local losses, are negligible for the shunt, but 4-5% for Rs, and
4-8% for the emitter and bulk recombination losses. In this case,
significant process variation is seen in both of these parameters.
Another example is shown in Fig. 9. In this case, n-type HIT
modules were measured in a study of field test degradation [16].
Fig. 7: Technology losses relative to cell power output for a For n-type modules, the excess carrier density in Equation 5 is
group of Al-BSF cells. significant compared to the doping, and the slope of the curve
in inverse effective lifetime vs. excess carrier density can be
used to separate out the sum of the front and back emitter
saturation current densities.
Fig. 9 shows the IV curves and Suns-Voc curves of a control
module and one module that has been fielded for ten years. By
using the analysis of the Suns-Voc data and Equations 4 and 5,
the likely loss mechanisms can be identified. In this case, it is
most likely that of the 11 W in degradation (out of 200W), most
of it was due to emitter degradation, with an increase in series
resistance coming in a distant second place.

IV. THE RELEVANCE OF LETTER-GRADE SIMULATOR RATINGS


Letter grade simulator ratings are often used to promote cell
and module-test simulators for sale to manufacturing
companies. Recently, these ratings are generally defined by an
Fig. 8: Technology losses relative to cell power output for a group
of monocrystalline PERC cells.

the current[15]. The “Ideal Limit” IV curve is shown for a 190


mm cell in Fig. 5. Then, there is a second limit (Fig. 6), which
is determined by the cell design and its constituent components.
In this case, the Suns-Voc curve is reference. For example, if
you reduce the bulk and emitter recombination by 30%, you
will not move 30% of the distance towards the “Richter limit”
curve. Both the IV-curve and the Suns-Voc curve will move
towards higher voltage as in eq. 4. Similarly, if you lower Rs
by 30%, you will move 30% of the distance in voltage out
toward the Suns-Voc curve. This “local” loss analysis is a
useful metric for real-time feedback into the production line.
Two examples of loss analysis will now be discussed with
supporting data. For a group of Al-BSF solar cells, the global
losses relative to the limit curves are 84% for the Vmp and 78%
for the Jsc. This Jsc loss is due to imperfect light trapping,
parasitic absorption of IR and UV in the cell coatings and the Fig. 9: Data reported for a study of HIT module degradation in the
doped silicon, and grid shadowing. The Vmp loss is due to field for 10 years [15]. A control module (dark red Suns-Voc data
and solid blue diamonds for IV data) was compared to a fielded
emitter, surface, and bulk recombination. These figures are module (light red Suns-Voc data and open blue diamonds for IV
based on the median losses for the group of cells. data).

978-1-5386-8529-7/18/$31.00 ©2018 IEEE 0013


IEC standard [18]. What is the relevance of these ratings for REFERENCES
production? It is well known that the spectral classifications [1] H. W. Wilterdink, A. L. Blum, C. Sainsbury and R. A. Sinton, "Practical
have no benefit for production testing [4], [17]. Class C Assessment of Power Rating Uncertainties for Industrial Silicon
spectrum simulators have often been shown to give superior Modules," in To be Published at 35th European Photovoltaic Solar Energy
results to “Class A+” (a marketing extension of the IEC Conference, Brussels, 2018.
standard) [17]. Inspection of the spectrum for many commercial [2] W. Herrmann and W. Wiesner, "Modelling of PV Modules -The Effects
of Non-Uniform Irradiance on Performance Measurements with Solar
simulators has shown aspects that are clearly designed to test Simulators," in 16th European Photovoltaic Solar Energy Conference,
well in the standard, but perhaps detrimental to actual Glasgow, 2000.
performance with respect to testing products [17]. [3] H. W. Wilterdink, A. L. Blum, C. Sainsbury and R. A. Sinton, "Spatial
In the first section of this paper, to be described in much non-uniformity of irradiance and uncertainty in power rating for c-Si
greater detail in a future publication [1], it is shown that for the modules," in NREL PVRW, Denver, 2016.
[4] C. Monokroussos, D. Etienne, K. Morita, V. Fakhfouri, J. Bai, C. Dreier,
range of non-uniformity from 2-5%, spanning up to the B-C U. Therhaag and W. Hermann, "Impact of Calibration Methodology into
boundary in the IEC standard, there is no detectable difference the Power Rating of c-Si PV Modules Under Industrial Conditions," in
in module power measurements. Although low non-uniformity 28th European Photovoltaic Solar Energy Conference and Exhibition,
is a good goal, and useful for calibration labs that deal with Paris, 2013.
various sizes and shapes of solar cells, there is no detectable [5] R. Evans, K. H. Kim, X. Wang, A. Sugianto, X. Chen, R. Chen and M.
A. Green, "Simplified technique for calculating mismatch loss in mass
benefit in production. The limits used for marketing are quite production," Solar Energy Materials and Solar Cells, vol. 134, pp. 236-
arbitrary. On the other hand, the actual quantification of power 243, 2015.
uncertainties on the production line is quite straightforward, [6] L. L. Bucciarelli Jr., "Power Loss in Photovoltaic Arrays Due to
and has nothing to do with letter grade standards, as shown in Mismatch in Cell Characteristics," Solar Energy, vol. 23, no. 4, pp. 277-
the second section in this paper. 288, 1979.
[7] V. Fakhfouri, W. Hermann, W. Zaaiman and L. Johnson, "Uncertainty
We find it unfortunate that it is widely believed that these assessment of PV Power measurement in industrial environments," in
ratings have value when we have been unable to find examples 26th European Photovoltaic Solar Energy Conference and Exhibition,
in the literature to match the persistent commercial claims. We Hamburg, 2011.
conclude that the use of letter grade ratings for production [8] K. Emery, "Uncertainty Analysis of Certified Photovoltaic Measurements
testing decisions is extraneous at best, and likely to be at the National Renewable Energy Laboratory," NREL, Golden, 2009.
[9] J. Hohl-Ebinger, D. Grote, B. Hund, A. Mette and W. Warta, "Contacting
detrimental to best practice. bare cells for STC measurements," in 23rd European Photovoltaic Solar
Energy Conference, Valencia, 2008.
[10] S. Raj, J. W. Ho, J. Wong and A. G. Aberle, "Impact of Non-uniform
IV. CONCLUSIONS Illumination and Probe Bar Shading on solar Cell I-V Measurement,"
This paper described three issues of current interest to the IEEE Journal of Photovoltaics, vol. 7, no. 5, pp. 1203-1208, 2017.
[11] C. N. Kruse, M. Wolf, C. Schinke, D. Hinken, R. Brendel and K. Bothe,
authors. First, the main high-level differences between "Impact of Contacting Geometries When Measuring Fill Factors of Solar
calibration laboratory measurements for absolute accuracy Cell Current–Voltage Characteristics," IEEE Journal of Photovoltaics,
were compared to the power ratings of cells and modules in vol. 7, no. 3, pp. 747-754, 2017.
production. In production, the reference calibration cells and [12] R. A. Sinton, H. W. Wilterdink and A. L. Blum, "Assessing Transient
modules are the key to low uncertainties. The production line Measurement Errors for High-Efficiency Silicon Solar Cells and
Modules," IEEE Journal of Photovoltaics, vol. 7, no. 6, pp. 1591-1595,
measurement reproducibility adds uncertainty, but this can be 2017.
determined and is generally small in comparison to the absolute [13] A. L. Blum, H. W. Wilterdink and R. A. Sinton, "Determining the
reference uncertainty. The opportunities for using IV-test Accuracy of Solar Cell and Module Measurements on High-Capacitance
equipment with device physics models in order to provide Devices," in to be published at the 7th World Conference on Photovoltaic
detailed information relevant to the process optimization was Energy Conversion, Waikoloa, 2018.
[14] A. L. Blum, R. A. Sinton, W. Dobson, H. Wilterdink and J. Dinger,
described. This can be done on every cell or module in "Lifetime and Substrate Doping Measurements of Solar Cells and
production yielding relevant data in real time. The resulting Application to In-Line Process Control," in 43rd IEEE Photovoltaic
data has also been applied to degradation studies to identify the Specialists Conference (PVSC), Portland, 2016.
root cause of power loss after degradation in the field. Finally, [15] A. Richter, M. Hermle and S. W. Gluntz, "Reassessment of the Limiting
the relevance of simulator letter-grade ratings to these issues Efficiency for Crystalline Silicon Solar Cells," IEEE Journal of
Photovoltaics, vol. 3, no. 4, pp. 1184-1191, 2013.
was discussed. The relevance of these ratings to industrial [16] D. C. Jordan, C. Deline, S. Johnston, et al., "Silicon Heterojunction
production is not supported by experimental validation. System Field Performance," IEEE Journal of Photovoltaics, vol. 8, no. 1,
pp. 177-182, 2018.
ACKNOWLEDGMENT [17] R. A. Sinton, H. Wilterdink, J. Dinger, A. L. Blum, W. Dobson and C.
Sainsbury, "Critical Evaluation of the Foundations of Solar Simulator
The authors would like to acknowledge Kris Davis Standards," in 44th IEEE Photovoltaic Specialist Conference,
(University of Central Florida), and Trina Solar for providing Washington D.C., 2018.
solar cells and modules for this work. [18] IEC 60904-9:2007 (Ed. 2).

978-1-5386-8529-7/18/$31.00 ©2018 IEEE 0014

Vous aimerez peut-être aussi