Version of Record: https://www.sciencedirect.

com/science/article/pii/S0038092X1730395X
Manuscript_301aff9268aebcf2f0c6a9b88291203d

Detailed investigation of TLM contact resistance measurements on crystalline

silicon solar cells
Siyu Guo1,2, Geoffrey Gregory1,2, Andrew M. Gabor3, Winston V. Schoenfeld1,2, Kristopher O.
Davis 1,2

1. Florida Solar Energy Center, University of Central Florida, 1679 Clearlake Road, Cocoa, FL,
USA 32922

2. c-Si Division, U.S. Photovoltaic Manufacturing Consortium, 12354 Research Parkway,

Orlando, FL 32826

3. BrightSpot Automation LLC, 5 Abbot Mill Lane 10F, Westford, MA, USA 01886

Abstract
The transmission line method (TLM) is often used in characterizing the contact resistance
of c-Si solar cells by cutting cells into strips parallel to the busbars. When applying this method
to industrial solar cells, we found various problems that have not been sufficiently explained in
prior work. In this paper, we investigate different factors that influence the accuracy of this
measurement, using both simulation and experimental methods. The following factors are shown
to influence the extracted contact resistivity and are investigated in this work: (1) strip width; (2)
edge shunting; (3) current flow through the intermediate unprobed fingers; (4) non-uniform
contact resistance; and (5) non-uniform sheet resistance. In cases where the contact resistivity
values determined from the TLM measurements and simulations were found to be inaccurate, we
introduce correction procedures and measurement guidelines that reduce error. For example,
when strip width is a factor, the measurement error of a 30mm sample is reduced from 95.5% to
4.5% using a correction procedure validated by simulation. Furthermore, the methods are also
shown to be very effective when applied to industrial solar cells. TLM measurements have an
important role to play in both cell R&D and factory quality control, and this work can serve as a
guide towards more accurate contact resistivity measurements.

1. Introduction
Optimized solar cell performance is critically dependent on the nature of the metal
contacts joined to the semiconductor absorber. Accurately measuring the contact resistivity of
the metal contacts is therefore very important. The most common method used to calculate the
contact resistivity of crystalline silicon (c-Si) solar cells is the transmission line method (TLM).
This method was originally proposed by Shockley [1] and further developed by Berger et al. [2-
4]. It is important not to confuse the transmission line method with the transfer length method, a
similar type of contact resistivity measurement. The TLM mentioned in this paper only refers to
transmission line method. In order to prepare a test structure from a solar cell, either a special
structure is fabricated with variable spacing between the contacts [5, 6], or a standard solar cell is
cut into strips parallel to the busbars [7-9]. The latter approach is more commonly applied, since

© 2017 published by Elsevier. This manuscript is made available under the Elsevier user license
https://www.elsevier.com/open-access/userlicense/1.0/
the measurement can be easily performed on a finished solar cell, which is very convenient and
achievable. By measuring the resistance between pairs of contacts with different spacing, the
TLM can be applied to calculate the contact resistivity and the sheet resistance of the underlying
semiconductor. Much of the literature focused on solar cell characterization uses this method to
measure contact resistivity. In [10], the TLM plot was modified according to the sheet resistance
under the contact. In [11], the influence from actual contact resistance values on the accuracy of
the TLM measurement was investigated.
Despite these existing works investigating the TLM method for c-Si solar cells, many
problems found with this measurement still remain unresolved. For example, our experiments
have shown that the measured contact resistivity changes significantly depending on the width of
the sample cut from a standard solar cell [12]. The measured contact resistivity is often
artificially high for sample widths that are either too large or too small. It is also found that the
transfer length extracted from the TLM plot varies significantly with the strip width, and this
phenomenon is not explained by TLM theory. In this work, we explain how and why these
problems, and others, occur and propose solutions. To do this, we introduce some basic theory
behind the TLM approach and the simulation methods used in this work. We then present the
key factors influencing the contact resistivity values measured from TLM and offer correction
procedures and measurement guidelines to reduce error. The key factors addressed in this paper
are: (1) strip width; (2) edge shunting; (3) current flow through the intermediate unprobed fingers;
(4) non-uniform contact resistance; and (5) non-uniform sheet resistance.

2. Theory and Method

2.1 Applying TLM in solar cell characterization

Basic TLM theory has been explained in many publications [3, 10, 13]. Here, we only
give a brief introduction of the method. When performing the TLM technique, the total
resistance T between two contacts with length (Z) and width (L) is measured and plotted as a
function of contact spacing (shown in Figure 1). Three parameters can be extracted from this
plot: the contact resistance ( C), the sheet resistance (Rsh), and the transfer length ( T). As
contact spacing d increases, the effect of the sheet resistance on the total resistance measurement
increases, thus creating the slope of the TLM plot (k). Therefore, the value of sheet resistance
can be extracted using:

Δ (1)
= =
Δd

Thus Rsh can be calculated by k·Z.

The transfer length indicates the average distance along L over which current transfers from the
semiconductor into the metal and vice versa. It is defined as:

= ⁄ . (2)
The value of the total resistance at the y-intercept of the plot is 2 . can be deduced
according to the potential distribution underneath the contact and is represented by:

= coth( ), (3)

where is the contact resistivity or specific contact resistance. When ≥ 1.5 , which can be
the case for c-Si solar cells with very good contact resistivity, Equation 3 can be approximated as:

= . (4)
"

Here, the effective contact area is and therefore the contact resistivity can be approximated
as effective contact resistivity:

#$%% = . (5)

Another common scenario seen in c-Si solar cells occurs when ≤ 0.5 . In this case Equation
3 can be approximated as:

= . (6)

Here, the effective contact area is the entire interfacial area of the contact and the contact
resistivity is calculated as:

#$%% = . (7)

In this study, Equation 5 is applied to calculate the contact resistivity of solar cells. When
applying this method for c-Si solar cell characterization, usually a strip is cut from a finished
solar cell without any special test structures. The voltage between variably spaced pairs of
fingers is measured one by one, and thus the total resistance between each finger pair is
calculated (R1, R2…Rn). Depending on the distance between each finger pair, a plot similar to the
one shown in Figure 1 can be constructed and the contact resistivity can be extracted.

Figure 1: (a) Typical TLM structure used for characterizing c-Si solar cells. (b) Fitted curve of
total resistance versus contact spacing.
2.2 Simulation method
In order to investigate the TLM method in detail, we use two simulation models in this
work. One method is a circuit model implemented in the open-source software LTspice [14]. The
implementation of this model is based on procedures presented in [15, 16]. The circuit model
constructed in this work represents a small strip cut from a solar cell. The model is a 2-D
network of circuit elements, and each unit element is based on the one-diode model which
includes a current source, a diode, a shunt resistance and a series resistance. Additional elements
represent the front metal grid and the contact resistance between the grid and the solar cell. The
model parameters for each unit are assigned according to their position in the TLM structure and
the model geometry. Figure 2 shows the circuit model of the simulated solar cell strip and the
structure of the intersection between two fingers.

Figure 2: Diagram of the equivalent circuit model built in this work. The intersection between
two fingers is shown.
Table 1: TCAD and Circuit Model Parameters
Parameter Value Units
Contact resistivity (ρc) 2 mΩ·cm2
Sheet resistance (Rsh) 85.0 Ω/□
Sample width (Z) 5.0 - 30.0 mm
Contact width (L) 60.0 μm
Contact spacing (d) 0.175 cm
Current level (I0) 0.01 A
Emitter thickness (t) 0.3 μm
 Another model used for studying the TLM method is a device model implemented in
Silvaco Atlas®, a device simulator. This TCAD software allows users to define the physical
structure, the physical models, and the bias conditions [17] . It has been widely used in solar cell
simulation [18]. Silvaco Atlas® requires that each structure be defined on a mesh that covers the
physical simulation domain. In order to get accurate results, it is important that the mesh is
defined properly. Therefore, a very fine mesh is defined near the metal-semiconductor interface
and then made coarse away from the contacts for the sake of computational efficiency. Table 1
summarizes the input parameters used in both the circuit model and in the TCAD model.

2.3 Experiment setup

In this work, a semi-automatic tool, the ContactSpot from BrightSpot Automation, is
used to measure the contact resistivity of solar cells. Figure 3 shows a photo of two ContactSpot
units. This unit includes a sample platform and microscope with translation and rotation controls
for easy contact alignment. Each probe head (seen in the inset of Figure 3) contains 10 sets of
current and voltage probes allowing the measurement of 10 contacts at a time. This tool can also
be used to measure the line resistance of metal contact segments. For such a measurement,
current is injected down the length of a contact and the resultant voltage drop is measured. The
resistance of the metal can then be calculated and used to find a value of resistance per unit
length.

Figure 3: Two ContactSpot tools for measuring contact resistivity, one with the cover removed.

3. Results and analysis

3.1 Strip width

3.1.1 Measurement and simulation results for samples with different strip widths
In the experiments, we use samples with different widths (5 mm to 25 mm) to measure
the contact resistivity of solar cells. The measurement results indicate strip width has a large
influence on the measured contact resistivity value. Figure 4 shows the typical relationship
between the effective contact resistivity and the sample width. As the strip width gets larger, the
measured contact resistivity and transfer length increase. Additionally, both device and circuit
simulations show the change in the measured effective contact resistivity is influenced by the
line resistance of the contacts. The circuit simulation results are also included in Figure 4. They
have the same trend with the measurement result. Since the finger line resistance used in
simulation (0.265 Ω/cm) was different from the measured sample, they didn’t match exactly with
each other.

Figure 4: Relationship between strip width and effective contact resistivity obtained from
measurement and simulation results.
Figure 5 shows the simulated TLM plots (i.e., RT versus contact spacing) for two sample
widths (5 mm and 30 mm). Ideally, with the same input material properties, if we scale the 5 mm
results by a factor of 1/6, we should get the same TLM plot. However, as indicated in Figure 5,
both the contact resistance and the transfer length obtained from the 30 mm sample (0.0048
Ω·cm2, 0.0055 cm) is much larger than that obtained from the 5 mm sample (0.0024 Ω·cm2,
0.0072 cm). The transfer length for the 30 mm sample is even larger than the finger width. This
phenomenon is not described in the conventional TLM theory.
From our investigation, this strip width effect is caused by poor conductivity of the
fingers. The conventional TLM theory assumes a pure two-dimensional model to explain the
current flow and ignored the finger line resistance. As is indicated in Figure 1(a), the TLM plot is
the relationship between total resistance and the distance between two fingers. As a consequence,
the part of the total resistance that does not change with distance is attributed to the contact
resistance. The portion of the total resistance that does change with distance is attributed to sheet
resistance. This can be well applied to devices which either have very narrow strip width or
highly conductive fingers. For other cases, the voltage drop down the length of the fingers cannot
be ignored, and this voltage drop contributes error to the measured contact resistance. Figure 6
shows the simulated voltage distribution along a probed finger for the 30 mm sample given line
resistance values of 0 (ideal case) and 0.265 Ω/cm. We observe that the voltage change along the
fingers for normal line resistance is much larger than the ideal case. The TLM plot for the ideal
case is also included in Figure 5. It’s calculated that this voltage change along the probed fingers
causes a 0.12 Ω resistance increase on the current flow path, and this explains why there is an
overall shift of (0.12 Ω) on the TLM plot for the 30 mm sample. However, the TLM plot for a 5
mm sample is very close to the ideal case since the voltage change on the fingers is much smaller.

Figure 5: TLM plot for simulated structures with 5 mm and 30 mm widths. The line resistance of
both structures are 0.265 Ω/cm. The ideal case indicates the TLM plot for the 30-mm structure
with zero finger line resistance.

Figure 6: Voltage distribution along a probed finger for the 30-mm structure with zero finger line
resistance and 0.265 Ω/cm finger line resistance.

3.1.2 Correction procedure for addressing strip width and finger line resistance
In the previous section, we showed that the influence of strip width on the measurement
results is caused by finger line resistance. Here, we make a correction to the TLM plot according
to the strip width, so that accurate values of ρc and LT can be obtained. First, we assume the
sample used in the measurement is composed of many sub-cells connected in parallel and that
each sub-cell has exactly the same property. In the ideal case (when the finger line resistance is
zero), current flow through the sheet of each sub-cell is uniform. In reality, there is a voltage
change along the probed finger, and thus the voltage drop and current flow across each sub-cell
is different. Assuming the resistance of each sub-cell is equal to R, there are n sub-cells
connected in parallel, and the voltage of the sub-cells is V1, V2.... Vn. The total current flow
through all the sub-cells is equal to:
-. -1 -
()*)+, = /
+ /
+ ⋯ /3 . (8)

Thus, the average current flow through the sub-cells is equal to:
456578 : - -1 - ∑3
>?@ ->
(+ = = ; .+ + ⋯ 3< = . (9)
9 9 / / / 9/

Current (+ is equal to the current flow through every sub-cell in the ideal case. As a consequence,
the voltage of the sub-cell with (+ current is equal to:
∑3
>?@ ->
A+ = (+ × = 9
. (10)

A+ is thus equal to the voltage drop through each sub-cell in the ideal case, and it is also
equal to the average of the voltages of all of the sub-cells in the real case. As a consequence, A+
should be the correct voltage used in calculating the resistance value in the TLM plot.
In order to examine the validity of this method, we apply the method to the simulation
result for the 30 mm sample shown in Figure 4. Instead of using the voltage at the probe, the
average voltages calculated for both fingers are used in the TLM plot. Comparing to the voltage
at the probe, the corrected voltage is reduced by 0.00121 V, and the resistance is reduced by
0.121 Ω. This is exactly equal to the difference between the initial TLM plot and the TLM plot
of the ideal case shown in Figure 5. Thus the correction method shown above is valid.
In the real application, the voltage along the probed finger cannot be easily measured. However,
it can be estimated as follows:
In the TLM measurement, current flows through the finger connected to the contact probe,
then through the emitter sheet underneath the finger. From simulation results, if the sheet
resistance of the sample is uniform, current flowing through the sheet has a uniform density. As a
consequence, the current then decreases linearly from the center of the finger to the edge (as
indicated in Figure 7). Assuming the total injection current from the probe is I0, the current flow
in each direction is I0/2 and drops to zero linearly at the edge of the finger. The voltage on any
position of the finger is a function of the distance from the probe (x):
 L
(F (F C (F (F (11)
A(C) = D E − × J K C=( C − CN) K
2 2 I 2 4I
F

where ρL is the metal line resistance, I0 is injection current, and S is the distance between the
current injection point to the edge, which is equal to the finger length Z divided by 2.
The average voltage drop along one finger can then be calculated using another integration:
P
(F (F (F I (F K
(12)
K K
AO = D ( C − C N ) C = =
I 2 4I 6 12
F

Considering both probed fingers, the total voltage after correction is then:
A *RR$ )$S = AT$+ UR$S − 2A+ (13)

Figure 7: Current flow along a probed finger in TLM measurement.

Based on Equation 12 and 13, for the 30 mm wide sample shown in Figure 4, the corrected
voltage is 0.00128 V lower than the measured voltage, which corresponds to 0.128 Ω resistance,
and this value is close to the actual value 0.121 Ω. The difference is likely caused by the
limitation of the resolution of the simulation model. In the next step, we shift the initial TLM
plot of the 30 mm sample down by 0.128 Ω. The transfer length and the effective contact
resistivity after correction for the two simulated samples are shown in Table 2. The results for
the ideal case are also included for comparison. The error of the measured ρc-eff is reduced from
95.5% to 4.5% after correction. Thus the correction method proposed in this work allows
accurate contact resistivity measurements to be obtained regardless of the chosen sample width.
Table 2: The ρc-eff and transfer length obtained from TLM plot for 5 mm and 30 mm samples
before and after correction. The ideal case is also included for comparison.
ρc-eff Error Error
Ideal case 0.00246 0.0% 0.00538 0%
30mm 0.00481 +95.5% 0.00752 +39.7%
30 mm (Corrected) 0.00235 -4.5% 0.00526 -2.2%
5mm 0.00247 +0.4% 0.00539 +18.6%
5mm (Corrected) 0.00235 -4.5% 0.00526 -2.2%

3.1.3 Applied correction method in the real measurement

Next, we implement this correction method in the actual TLM measurements. The
required input finger line resistance can obtained in one of two ways:
(1) The finger line resistance can be directly measured using the ContactSpot tool.
(2) The finger line resistance can be deduced using the measured Rc values for different strip
widths.
We first applied the Method (1) to correct the measurement results shown in Figure 4(a).
The measured finger line resistance is 0.292 Ω/cm. The initial measured ρc-eff for 20 mm and 25
mm samples are 3.2 mΩ·cm2 and 3.9 mΩ·cm2. After correction, the ρc-eff becomes 2.0 mΩ·cm2
for both strip widths. And the measured finger line resistance agrees well with the extracted data
using Method 2, which is 0.288 Ω/cm, a value similar to that obtained from measurement.
3.2 Edge shunting
In this work, we performed measurements on many different samples with different
sample widths. Generally, the measured contact resistivity should increase as the sample width
becomes larger. However, we find that in many cases, the measured contact resistivity is very
large for small sample widths (5 mm) as indicated in Figure 8. In this session, we would like to
investigate the reason behind this effect. One possibility is that laser cutting induces shunting
effect on the edges of the samples and creates additional shunting paths when the TLM structures
are created. Shunt paths lead to an increase in the measured contact resistance, and this influence
becomes more pronounced when the sample width gets smaller.

Figure 8: Relationship between ρc-eff and strip width obtained from measurement.
In order to prove this, we incorporated shunted regions in our simulation. The shunt
resistance of the cell within 0.05 cm from the edge is set to be 400 Ω·cm2 instead of 10,000
Ω·cm2. Figure 9 shows a plot of ρc-eff versus strip width for structures with and without edge
shunting. We observe that edge shunting leads to higher measured values of ρc-eff at all strip
widths, and this influence is much more significant at small strip width. Thus the influence of
shunt resistance on contact resistivity observed from measurement is proved. Figure 10(a) shows
the voltage distribution along a probed finger for a simulated structure with a 10 mm width. We
observe that the edge shunting changes the voltage distribution along the finger. The voltage
change at the end of the finger is much larger than the normal case, which indicates more current
flow from the end of the finger. Figure 10(b) shows a diagram of the current flow path for a
sample with and without edge shunting. From simulation it is found that instead of flowing
through the emitter, part of the current flows through the shunt path and the rear contact, which
makes the measured contact resistance larger. Also, the percentage of this current becomes
smaller as the strip width gets larger, which explains why its influence gets smaller for larger
strip widths.

Figure 9: Simulated relationship between ρc-eff and strip width for structures with and without
edge shunting effect.

For the case shown in our simulation, the influence from the shunting effect makes the
measured ρc-eff 49.3% higher. However, for the 30-mm sample, the measured ρc-eff is only 4.5%
larger. If we make a further correction based on the strip width according to the method in
Section 3.1, the total error can be kept within 10%. If shunts from the sample preparation method
cannot be avoided, we then recommend to use samples with widths larger than 10 mm for typical
cells and to make further corrections to minimize the measurement error. The minimum value of
strip width in order to avoid shunt effects will depend on the line resistivity of the fingers and the
severity of the shunt paths, and it is best determined empirically for any particular cell
architecture.
 (a) (b)

Figure 10: (a) Simulated voltage distribution along the finger width for 10 mm structures with
and without edge shunting effect. (b) Diagram of current flow path for a sample without and with
edge shunting.

3.3 Current flow through intermediate fingers

It has already been mentioned that when finished c-Si solar cells are used for contact
resistivity measurements, it is necessary to skip over contacts in order to measure the total
resistance RT at different contact spacings d. In the past, groups have reported seeing consistently
higher values of when using TLM structures with large numbers of unprobed contacts [19].
Thus there might be an influence from the unprobed contacts on the TLM measurement. In this
work, we find that depending on the sheet resistance and contact resistance, some current flows
through the intermediate fingers, which affects the accuracy of the measured ρc-eff.
Figure 11(a) shows a diagram of the current flow path when there is an intermediate
finger. By modeling a series of resistors connecting the finger and the sheet, we found that a
portion of the current flows through the contact resistors in the intermediate fingers and then
back into the sheet. In the simulation, there are four resistors connecting the finger and the sheet.
The current flow through each resistor of a solar cell with 85 Ω/□ sheet resistance and 0.002
Ω·cm2 contact resistivity is shown in Figure 11(b). Based on this, the current flow through the
sheet underneath the fingers is also calculated, and the total voltage drop is obtained. Compared
to the case where all current flows through the sheet, the voltage drop is reduced by 17.5%.
 In order to make a correction to the TLM plot, the finger width is modified according to
the actual voltage drop. The actual finger width 0.006 cm is then multiplied by 0.825 to be used
as the finger width in the TLM plot. The x axis of each point is then calculated by:
C (V) = (V − 1) × + (V − 2)WX × YX . (n≥2) (14)

Here, C (V) represents the input distance on the TLM plot when probing the nth finger on the
sample, is the finger pitch, WX is the correction factor and YX is the actual finger width.

(a) (b)

Figure 11: (a) The current flow path within an intermediate finger in the TLM measurement (b)
The actual current flow in each resistor obtained from the simulation.

Figure 12 shows the TLM plot for a 5 mm sample before and after correction. Before
correction, ρc-eff calculated from the TLM plot is 0.0029 Ω·cm2, and this value is reduced to
0.0025 Ω·cm2 after correction, which is much closer to the input ρc value. The correction factor
depends on the sheet resistance and the contact resistance according to our investigation. Figure
13 shows the relationship between the sheet resistance and the correction factor for the simulated
structure. As the sheet resistance gets higher, more current flows through the intermediate fingers,
and the correction factor is reduced. In the real application, the correction needs to be performed
based on the actual sheet resistance and the contact resistance.
Figure 12: The simulated TLM plot for a 5mm structure before and after correction for
intermediate fingers.

Figure 13: Relationship between the correction factor and the sheet resistance for the simulated
structure.

3.4 Inhomogeneous contact resistance

The TLM theory assumes that all contacts being measured on a structure have a uniform
value of contact resistance. However, in experimental settings this assumption is rarely true,
especially for commercial solar cells. A minimum of three fingers are required in order to create
the data points of a TLM plot. If one or more of the fingers possesses a value of contact
resistance that is different than that of its neighboring finger, the line of best fit in the TLM plot
will have a coefficient of determination (r2) that is not equal to unity when plotting three or more
points. Therefore, the values of contact resistance and sheet resistance will not be accurately
represented. TLM plots use a least square linear regression to create the line of best fit between
several data points. This type of regression analysis works to minimize the sum of the squares of
the errors made when fitting the trend line to each total resistance point in the TLM plot. As each
data point is skewed in the plot, the desired results for contact resistivity and sheet resistance are
consequently skewed.
In order to investigate how the inhomogeneity of contact resistance on a TLM structure
would affect the line of best fit in the TLM plot, we built a model. The structure shown Figure 14
was simulated with four metal fingers, each of which was 60 μm wide and 10 μm tall with a
resistivity value of 1.59 × 10-8 Ω.m. The fingers were modeled on top of a 10 mm wide and 300
nm thick silicon emitter with an 85 Ω/□ sheet resistance. All four fingers have a contact
resistance value of 0. 6Z Ω. One of the fingers in the group of four was given a non-uniform value
of contact resistance in each simulation. The non-uniform values were set to be 105, 115, 125,
and 135 percent of the neighboring contact resistance values.

Figure 14: Schematic of TLM structure built in Silvaco Atlas for contact resistance analysis.
All four fingers were given non-uniform values of contact resistance separately in order
to analyze how each data point in the TLM plot would affect the desired results if it was skewed.
Figure 15(a) shows plots of the error in the measured contact resistance of finger #1 versus the
finger number which had a different contact resistance value; each magnitude of skew (5-35%) is
plotted in the graph. The error is defined to be the simulated result minus the result from a case
of uniform contact resistance. This error reflects solely the influence from the non-uniformity of
contact resistance on the measurement accuracy and does not include any residual error in the
model. Figure 15(b) shows plots of the error in measured sheet resistance versus the finger
number which has a non-uniform value. The combination of these two graphs can help explain
how the results are skewed in this case.

(a) (b)
Figure 15: Plot of error in (a) contact resistance and (b) sheet resistance versus finger number
given a non-uniform value of contact resistance.
Figure 15(b) shows that the error in sheet resistance is close to zero when the first finger
is given a non-uniform value. Since the first finger is involved in every total resistance
measurement used to create the TLM plot, each point in the plot is skewed by the same amount
of added resistance and only the contact resistance changes. If each point is skewed by an equal
amount, the line of best fit simply translates in the y-direction without changing slope, as the red
curve shown in the schematic in Figure 16(a). Figure 15(b) also shows that the sheet resistance
is under-measured when the second finger has a skewed value of contact resistance. The red plot
in Figure 16(b) shows how the increase in only the first total resistance point affects the TLM
plot due to finger #2 being skewed. Since the slope of the plot is lower than expected, the contact
resistance is over-measured. Notice that the magnitude of the error in contact resistance is high in
this case, which is due to the proximity of the first data point to the y-intercept. The closer the
skewed data point is to the y-intercept, the larger the error in contact resistance will be.

(a) (b)

(c) (d)
Figure 16: Schematic TLM plots showing the effect of various fingers having a different value of
contact resistance from the others (red line).
 Figures 15 and 16 also show that the error of the measured contact resistance and sheet
resistance is at a minimum when the third finger has an inhomogeneous contact resistance value.
In this case, the line of best fit minimizes the sum of the squares of the errors when it is drawn
directly in between the three data points, as shown in Figure 16(c). If there are an even number
of data points in the TLM plot, and only one contact resistance value is skewed, the line of best
fit will be forced to change slope. Figures 15 and 16 also show that the contact resistance is
under-measured and the sheet resistance is over-measured when finger #4 is given a positive
skew of contact resistance. The slope of the TLM plot must increase in this case, in order to
account for the deviation of the third data point. This is shown in Figure 16(d). When performing
the TLM it is important to understand the behavior of the line of best fit as these types of non-
uniformities occur.
3.5 Inhomogeneous sheet resistance
The TLM model assumes that during a measurement the sheet resistance of the
semiconductor is uniform between the pairs of fingers. However, commercial solar cells rarely
have uniform sheet resistance and many are intentionally given differing values of sheet
resistance underneath and in between their fingers in order to maximize cell efficiency. Schroder
[13] has outlined a method of extracting the value of sheet resistance underneath and in between
the contacts of a semiconductor device using the TLM method along with other four point probe
measurements. This technique has been used on experimental structures in the past, but little has
been reported on this method in relation to solar cells. In this work, we investigate the influence
of inhomogeneous sheet resistance on the TLM measurement of crystalline silicon solar cells and
simulate the extraction of sheet resistance underneath the fingers.
A structure like the one seen in Figure 17 was simulated with four metal fingers, each
having a width of 60 μm and a height of 10 μm with a resistivity value of 1.59 × 10-8 Ω.m. The
fingers were modeled on top of a 10 mm wide and 300 nm thick silicon emitter. During each
simulation, the sheet resistance underneath the fingers was given a different value than that of the
sheet resistance in between the fingers, which was fixed at 85 Ω/□. The values of sheet resistance
underneath the fingers were: 20, 40, 60, 80, 100, and 120 Ω/□. The low values could possibly
correspond to selective emitter structures, while the higher values could correspond to a case
where the firing of the metallization paste consumed some of the emitter.

Figure 17: Schematic of a TLM structure built in Silvaco Atlas for contact resistance analysis.
The model was first simulated with a uniform sheet resistance and contact resistance
throughout the entire TLM structure in order to obtain reference results. The results were then
compared to the cases where the sheet resistance was inhomogeneous. Figure 18 (a) shows that
the calculated sheet resistance increases as the sheet resistance underneath the fingers increases.
Similarly, Figure 18 (b) shows that as the sheet resistance underneath the fingers increases, the
calculated contact resistance decreases.

Figure 18 (a): Measured sheet resistance versus sheet resistance underneath the fingers. (b):
Measured contact resistance versus sheet resistance underneath the finger.

With the effects of inhomogeneity quantified, we then attempted to perform Schroder’s

method of extraction, where both the sheet resistance in between the contacts [\ , and the sheet
resistance underneath the contacts [] are identified. We performed this method on the
inhomogeneous sheet resistance structures that were previously simulated above as well as on a
series of TLM structures with variably spaced contacts. The results of the study are shown below
in Table 3:
Table 3: The values of [] and ] obtained from TLM plot for samples modeled on solar cell
structures and on TLM structures with variably spaced contacts.
Calculated [] on TLM
Value of [] set in the Calculated [] on solar cell
structures with variably
simulation (Ω/□) structures (Ω/□)
spaced contacts (Ω/□)
40.0 36.6 91.3
60.0 54.4 92.4
80.0 77.3 92.9
100.0 97.6 95.0
120.0 117.8 95.8

The results of the simulation show that [] was only accurately calculated on the TLM
structures with variably spaced contacts. The solar cell structures yielded a relatively constant
value of [] , which was not accurate. The error in the calculation is a result of the intermediate
fingers and their inhomogeneous sheet resistance values interfering with each total resistance
point on the TLM plot. While Schroder’s method is able to account for the inhomogeneous sheet
resistance values underneath the contacts, it is sensitive to slight fluctuations in X and [\ .
Therefore, the method cannot be applied to TLM structures made from solar cells, which include
multiple fluctuations in sheet resistance along the current path.

4. Conclusion
Although the transmission line method was initially developed for structures with
variable spaced contacts, the method is often used in characterizing the contact resistance of c-Si
solar cells by cutting cells into strips parallel to the busbars. Despite the significant differences in
the geometries of such structures, little has been published on the accuracy and potential
problems of this approach. In this paper, we investigated different factors influencing the results
when applying the TLM measurement to a standard c-Si solar cell. In cases where we found
inaccuracies in the method, we developed methods of making corrections to the results based on
the structure’s dimensions (strip width) and current flow through intermediate fingers. With this
approach, we successfully explained the large measured contact resistivity seen for small strip
widths, and provided guidelines for preparing samples when there is a shunting problem. TLM
measurements have an important role to play in both cell R&D and factory quality control, and
this work can serve as a guide towards more accurate contact resistivity measurements.

Acknowledgements
This material is based upon work supported by the U.S. Department of Energy, Office of Energy
Efficiency and Renewable Energy, in the Solar Energy Technologies Program, under Award
Number DE-EE0004947.
Reference
1. Shockley, W., 1964. Research and investigation of inverse epitaxial UHF power transistors. Air
Force At. Lab. Wright-Patterson Air Force Base Ohio.
2. Berger H.H., 1972. Models for Contacts to Planar Devices. Solid State Electron 15(2),145.
3. Berger H.H., 1969. Contact resistance on diffused resistors. IEEE Solid-State Circuits
Conference , pp. 160-161.
4. Berger H.H., 1972. Contact Resistance and Contact Resistivity. J. Electrochem. Soc. 119(4):507.
5. Pysch D., Mette A., Filipovic A., Glunz S.W., 2009. Comprehensive Analysis of Advanced Solar
Cell Contacts Consisting of Printed Fine-line Seed Layers Thickened by Silver Plating. Prog.
Photovoltaics 17(2), 101-114.
6. Untila G.G., Kost T.N., Chebotareva A.B., 2017. Multi-wire metallization for solar cells: Contact
resistivity of the interface between the wires and In2O3:Sn, In2O3:F, and ZnO:Al layers. Sol.
Energy 142, 330-339.
7. Zeng F., Feng Y., Liang Z., Shen H., 2012. Specific contact resistance measurements on C-Si
solar cells by novel TLM method. 38th IEEE Photovoltaics Specialists Conference, pp. 509-513
8. Willsch B., Kumar P., Aabdin Z., Peranio N., Eibl O., 2015. Series and Contact Resistance
Measurements between 80K and Room Temperature for Industrial Solar Cells. Energy Proc. 67,
49-63.
9. Vinod P.N., 2011. Specific contact resistance measurements of the screen-printed Ag thick film
contacts in the silicon solar cells by three-point probe methodology and TLM method. J. Mater.
Sci.: Materials in Electronics 22(9), 1248-1257.
10. Reeves G.K., Harrison H.B., 1982. Obtaining the specific contact resistance from transmission
line model measurements. IEEE Electr. Device L. 3(5), 111-113.
11. Meier D.L., Schroder D.K., 1984. Contact resistance: Its measurement and relative importance to
power loss in a solar cell. IEEE T. Electron Dev. 31(5), 647-653.
12. Gabor A.M., Gregory G., Payne A.M., Janoch R., Anselmo A., Yelundur V., Davis K.O., 2016.
Dependence of Solar Cell Contact Resistivity Measurements on Sample Preparation Methods.
43rd IEEE Photovoltaics Specialists Conference.
13. Schroder D.K., 2006. Semiconductor Material and Device Characterization. John Wiley & Sons.
14. Engelhardt M., 2011. LTspice/SwitcherCAD IV: Linear Technology Corp.
15. Guo S., Ma F.J., Hoex B., Aberle A.G., Peters M., Luther J., 2012. Energy Proc., 25, 28-33.
16. Guo S., Peters M., Aberle A.G., 2012. PV Asia Pacific Conference 2012Investigating Local
Inhomogeneity Effects of Silicon Wafer Solar Cells by Circuit Modelling. Energy Proce. 33, 110-
117.
17. Atlas User’s Manual, Device Simulation Software, 2015. Silvaco International.
18. Michael S., 2005. A novel approach for the modeling of advanced photovoltaic devices using the
SILVACO/ATLAS virtual wafer fabrication tools. Sol. Energ. Mat. Sol. C. 87(1-4), 771-784.
19. Janoch R., Gabor A.M., Anselmo A., Dub C.E., 2015. Contact resistance measurement -
observations on technique and test parameters. 42nd IEEE Photovoltaic Specialist Conference, pp.
1-6.