Articles

https://doi.org/10.1038/s41560-018-0126-z

Economic viability of thin-film tandem solar

modules in the United States
Sarah E. Sofia1*, Jonathan P. Mailoa1, Dirk N. Weiss2, Billy J. Stanbery3, Tonio Buonassisi1*
and I. Marius Peters1*

Tandem solar cells are more efficient but more expensive per unit area than established single-junction (SJ) solar cells. To
understand when specific tandem architectures should be utilized, we evaluate the cost-effectiveness of different II–VI-based
thin-film tandem solar cells and compare them to the SJ subcells. Levelized cost of electricity (LCOE) and energy yield are cal-
culated for four technologies: industrial cadmium telluride and copper indium gallium selenide, and their hypothetical two-ter-
minal (series-connected subcells) and four-terminal (electrically independent subcells) tandems, assuming record SJ quality
subcells. Different climatic conditions and scales (residential and utility scale) are considered. We show that, for US residential
systems with current balance-of-system costs, the four-terminal tandem has the lowest LCOE because of its superior energy
yield, even though it has the highest US$ per watt (US$ W–1) module cost. For utility-scale systems, the lowest LCOE architec-
ture is the cadmium telluride single junction, the lowest US$ W–1 module. The two-terminal tandem requires decreased subcell
absorber costs to reach competitiveness over the four-terminal one.

P
ower-conversion efficiency is a key driver to reduce the cost the operating conditions and spectral variations12. Thus, location-
of photovoltaic (PV) electricity1. Tandem solar cells open a specific energy yield is a significantly more informative metric than
path to efficiencies above 30%, which exceeds the Shockley– STC efficiency or a focus on just one location.
Queisser limit of single-junction (SJ) devices by combining mul- In this work, we explore the economic viability of thin-film
tiple solar cell materials together2,3. However, because the higher tandems, and focus on the two thin-film materials that have been
efficiency is at least partially offset by higher fabrication costs, it is industrially manufactured at gigawatt scales: CdTe and CIGS. We
unknown whether tandems can achieve a sufficiently low levelized present a bottom-up manufacturing cost model for CdTe–CIGS
cost of electricity (LCOE) to compete with SJ devices in one-sun tandem solar modules, along with the corresponding subcells’ SJ
applications. To assess the cost-effectiveness of tandems requires modules, as well as location-specific energy-yield calculations for
coupled assessments of energy yield and cost4–9. each architecture using a published model11. To motivate future
Prior efforts to explore the economic viability of tandems6,7,10, development, we investigate tandems based on subcells with record
including cadmium telluride (CdTe)/copper indium gallium sel- efficiency quality. These tandems mark what is technologically pos-
enide (CIGS) and III–V-on-Si tandems, concentrate on standard sible but go beyond what has been achieved. We then compare the
testing condition (STC) efficiency and consider neither LCOE nor economic viability using LCOE as a figure of merit. This is done
energy yield. These studies focus on a parallel-connected, four- using current system-installation costs in the United States, as well
terminal (4T) mechanically stacked tandem architecture in which as a future scenario in which these costs are substantially reduced.
subcells are electrically independent. This precludes a detailed We use parametric cost analyses to examine changing manufactur-
comparison between the two main tandem architectures: the par- ing and system costs to reveal what changes to technical variables
allel-connected 4T tandem and the monolithic series-connected are required to make tandems more cost-effective than SJ devices,
two-terminal (2T) tandem. To determine which of those tandem the relative advantages of 2T and 4T tandems, and tandem viability
architectures is preferable is a pressing question, as each architec- in the face of declining PV balance-of-system (BOS) costs. Though
ture requires many months of research and development (R&D) we focus on industrially mature II–VI thin-film technologies, these
to perfect, and thus significant investments in equipment, time results also apply broadly to emerging thin-film materials.
and funds.
Recent energy-yield calculations for II–VI-based thin-film tan- Solar cell architectures and energy-yield calculations
dem solar cells examine the potential of several tandem architectures From the three tandem architectures explored previously11, we
based on CdTe and CIGS, and include both 2T and 4T configura- selected the two highest-performing tandems: a 4T CdTe-on-CIGS
tions11. These tandems have the advantage of leveraging known and tandem (Fig. 1c) and a 2T high-Eg (1.68 eV)-top-cell-on-CIGS
industrially mature PV materials. Furthermore, thin-film/thin-film tandem (Fig. 1d). We use a higher-bandgap top cell for this archi-
tandems combine subcells with similar manufacturing processes tecture because the series-connected subcells of a 2T tandem are
and cost structures, which favour tandems economically10. One constrained to have equal current, which causes non-ideal band-
analysis11 considers device performance in three geographical loca- gap pairings to be more detrimental for 2T than for 4T tandems.
tions that represent three distinct climates: dry (Albuquerque), tem- For comparison, we additionally consider both CdTe and CIGS SJ
perate (Rapid City) and humid (Miami). Tandem architectures and modules. The efficiency and energy yield for each device is cal-
SJ devices of different bandgaps perform differently depending on culated using published methodology11, and is given in Table 1.

1
Massachusetts Institute of Technology, Cambridge, MA, USA. 2First Solar Inc., Santa Clara, CA, USA. 3Siva Power, Santa Clara, CA, USA.
*e-mail: [email protected]; [email protected]; [email protected]

Nature Energy | VOL 3 | MAY 2018 | 387–394 | www.nature.com/natureenergy 387

© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
Articles Nature Energy

a b
ARC glass ARC glass

direction
FTO

Growth
EVA
CdTe absorber AZO

direction
Growth
CIGS buffer layer
Mo
CIGS absorber
EVA
Mo
Glass
Glass
JB
JB

c d
ARC glass
direction

ARC glass
Growth

FTO FTO

direction
Growth
CdTe absorber High Eg II–VI
ITO TJ
EVA CIGS absorber
Mo
AZO
CIGS buffer layer EVA
direction
Growth

CIGS absorber Glass

Mo JB
Glass
JB

Fig. 1 | Device architectures schematics. a–d, Schematic of the four device architectures, CdTe SJ (a), CIGS SJ (b), 4T tandem (c) and 2T tandem (d).
EVA, ethylene-vinyl acetate; ARC, antireflection coating; FTO, fluorine-doped tin oxide; AZO, aluminium-doped zinc oxide; ITO, indium-doped tin oxide;
JB, junction box; Mo, molybdenum; TJ, tunnel junction.

Methods gives more details on the device assumptions and effi- Considering LCOE
ciency calculations. Though neither tandem architecture is cost competitive on a mod-
ule level due to their significantly higher MSP than the SJ modules,
Module manufacturing cost tandems can still have advantages on a system level as the mod-
We developed a bottom-up manufacturing cost model for both the ule costs are only part of the total PV system cost. High efficiency
2T and 4T tandem modules and their comprising subcells, employed reduces the cost-per-watt of a system by outputting more power for
as SJ cells. The cost model is based on the step-by-step process flows a given system area. Thus, to compare fully the relative cost effec-
given in Supplementary Tables 2–5. The cost breakdown for each tiveness of the tandem modules, we calculate LCOE for the four
step is given in Supplementary Tables 1–5. From the manufacturing architectures in all three locations for residential- and utility-scale
cost, we calculate the minimum sustainable price (MSP) for each installed PV systems in the United Sates.
module1. Methods gives details on the development of and assump- System installation costs are calculated using utility and residen-
tions used in the cost model and MSP calculations. The financial tial data13 for the 2016 US PV system costs (Supplementary Table 7).
parameters used for MSP are given in Supplementary Table 6. To explore how LCOE changes for the different architectures if
From the cost model, manufacturing process flows, financial system costs continue to fall, we include a hypothetical future sce-
parameters and calculated efficiencies in Table 1, the US$ Wp–1 nario with substantially reduced system installation costs, based
cost and MSP for each tandem and SJ architecture are derived on a proposed US Department of Energy SunShot target scenario14
(Fig. 2 and Table 2). The CdTe SJ has the lowest US$ Wp–1 price, (Supplementary Table 7). We refer to this case as the ‘reduced sys-
US$0.30 Wp–1, whereas the two tandems are the most expensive. tem cost scenario’ (RSCS). Methods gives further details on the cal-
The 4T tandem is the most-expensive architecture in US$ Wp–1, and culation of system costs and LCOE. The total system costs are given
exceeds the cost of the 2T by about US$23.32 m–2, with only a small in Table 2.
STC efficiency advantage of 1.5% absolute. The high cost of CIGS Using the module and system costs, along with the calculated
relative to CdTe is due to the high capital expenditure (CapEx) of energy yields, we compute the LCOE for all four architectures
the CIGS absorber deposition with the current commercialized for both residential- and utility-scale systems using equation (1)
CIGS deposition process. (Methods) for Figs. 2 and 3. We first discuss the LCOE results for

Table 1 | Module performance

Energy yield (kWh m–2 yr–1)
Architecture STC efficiency (%) Dry Temperate Humid
Fixed Tracked Fixed Tracked Fixed Tracked
4T tandem 26.5 621.19 716.44 504.84 548.54 506.21 602.59
2T tandem 25.0 569.91 658.38 459.45 500.41 445.31 539.55
CdTe SJ 19.9 471.52 538.55 381.56 413.31 397.06 466.46
CIGS SJ 19.8 464.22 536.15 377.76 411.07 378.53 451.35
Efficiency under STC and calculated energy yield for each of the considered device architectures for fixed tilt (Fixed) and single-axis tracking (Tracked). Efficiencies of the tandem solar cells were calculated
using subcell EQE curves and current–voltage (I–V) curve parameters (open circuit voltage, series resistance and shunt resistance) fitted from the specified cell results. The EQE curves of the bottom cells
include the optical losses due to imperfect transmission and optical coupling of the top cell, which were calculated using the transfer-matrix method of optical modelling11. The fixed-axis yields are used for
residential-system LCOE calculations and the single-axis tracking yields are used for utility systems exclusively.

388 Nature Energy | VOL 3 | MAY 2018 | 387–394 | www.nature.com/natureenergy

© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
Nature Energy Articles
0.50
4T 2T CdTe SJ CIGS SJ
US$0.43 a b

12.5
0.40 US$0.36 US$0.36 5.5
Cost and MSP (US$ W–1)

US$0.30

LCOE (UScents kWh–1)

Maintenance and
overhead 11.5
0.30
Labour 5

Utilities

0.20 Materials 10.5

Depreciation 4.5

Estimated MSP
0.10 9.5

ry

id

ry

id
t

t
um

um
D

ra

ra
0.00

pe

pe
H

H
m

m
Te

Te
4T

2T

IS
dT

C
C

SJ
SJ

Fig. 3 | 2016 system-cost scenario LCOE values. a,b, Calculated LCOEs

Fig. 2 | Module costs. Module cost and MSP for each architecture output for residential-scale (a) and utility-scale (b) systems in each of the three
from our module manufacturing cost model. The MSP is calculated such specified dry, temperate and humid locations for each architecture, using
that the cash flow has an IRR equal to the weight average cost of capital the 2016 system and installation costs.
(WACC), assuming a WACC of 14%.
to current mismatch in realistic climates and the slightly lower effi-
the 2016 scenario (Fig. 3). The 4T tandem has an LCOE advantage ciency (−​1.5%) mean the 2T tandem is a more expensive tandem
for residential-scale systems over both the 2T and the SJ modules. configuration, on a system level, than the 4T architecture. The 2T
It has the lowest LCOE in all three locations, that is, 8.0%, 8.4% and tandem performs best in dry and temperate locations where it has
4.9% lower than the next cheapest SJ module in the dry, temperate the second-lowest LCOE, whereas it performs worst, comparatively,
and humid locations, respectively. This advantage exists due to the in the humid location, with a higher LCOE than both the 2T and
high BOS and installation costs for residential systems in the United the CdTe SJ, because water vapour causes a significant spectral
States, relative to the PV modules, which make up 72–78% of the shift that results in current mismatch, and so reduces energy yield.
total system cost for these four architectures. This creates a signifi- Furthermore, the absorption edge of CdTe is above the absorption
cant premium on maximizing the energy generated per area of a peaks of water vapour, which makes the CdTe SJ energy yield par-
system, and so causes residential systems to favour high-efficiency ticularly robust to humid climates. As a result, the added cost for the
and high-yield modules. Additionally, the 4T tandem continues to 2T module is not justified compared to a CdTe SJ.
have the lowest LCOE if the optical transmission to the bottom cell Utility-scale systems, alternatively, show no advantage for tan-
is worse than modelled. The power from the bottom cell can drop dems. The CdTe SJ module is associated with the lowest LCOE in a
by up to 30% in a humid climate and 40% in dry and temperate utility system for all three locations, though the 2T LCOE is equal
climates and still remain the lowest LCOE residential architecture. in the dry location. This is a result of the significantly lower installa-
Furthermore, the 4T tandem LCOEs are 2.0%, 2.7% and 6.0% tion and BOS costs, which diminish the value of high efficiency and
lower than the 2T tandem LCOEs for residential systems for dry, cause utility-scale systems to favour the cheapest module.
temperate and humid locations, respectively. Despite having a lower The LCOE results produced using the RSCS (Fig. 4) show that
manufacturing cost, the 2T tandem’s lower performance ratio due to reduce the non-module system costs makes tandems less favour-
able. The 4T tandem loses its advantage over the CdTe SJ for resi-
Table 2 | Total module and system costs dential systems in humid climates, with a 2% higher LCOE in this
case. Tandems remain the lowest LCOE option for dry and temper-
Architecture Module Module Total Total installation + ate locations: the 4T LCOE is still the lowest in the temperate loca-
Cost MSP installation BOS costs, Utility tion and the 2T and 4T LCOEs are almost equal in the dry location,
+ BOS costs, with the 2T providing the lowest LCOE. The relative advantage of
Residential tandems compared to SJ architectures for these locations, however,
(US$/ (US$/ (US$) (US$/ (US$) (US$/ is reduced for this RSCS to just 1.4% and 1.8% for dry and temperate
m2) m 2) Wp) Wp) locations, respectively. The utility-scale results are similar to those
4T Tandem 76.44 111.17 12,140 1.77 77,557,623 0.98
for the 2016 case, with the CdTe SJ having the lowest LCOE for all
locations; the benefit of the CdTe SJ over the tandem architectures
2T Tandem 61.13 90.16 12,019 1.76 76,617,691 0.93 has increased.
CdTe SJ 42.44 58.84 11,609 1.97 73,417,333 0.96 These results show the impact of system costs in determining
CIGS SJ 49.42 71.18 11,601 2.05 73,398,925 1.04 whether tandems are favourable. There is significant potential for
installation costs to change over time or across different locations.
Total calculated module manufacturing cost and minimum sustainable price (MSP) for each
architecture, and the total cost of installation (excluding module costs) for each architecture The costs used in the RSCS, however, are aggressive and other sce-
in a residential (area =​35.9 m2) and utility (area =​0.6 km2) system based on the 2016 system narios are possible. To have a fuller understanding of the effect of
cost scenario. Wp refers to the module power output, in watts, at peak power equal to system costs, we varied the total installation cost, the area-dependent
nameplate capacity.
cost and the per-project cost to look at how the lowest LCOE

Nature Energy | VOL 3 | MAY 2018 | 387–394 | www.nature.com/natureenergy 389

© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
Articles Nature Energy

4T 2T CdTe SJ CIGS SJ The modelled CIGS absorber price is too high for a tandem sub-cell,
given how much added energy-yield a CIGS bottom cell generates.
a b If the MSP of the CIGS absorber comes down by about 10% or if
8.5 we are overestimating the cost, tandems are much more viable. The
5 high MSP for CIGS is due to the high CapEx associated with the
CIGS absorber deposition. This is based on current commercialized
manufacturing processes and tools. CIGS has only recently been
LCOE (UScents kWh–1)

8
commercialized, so there is potential for the CapEx to come down
as the technology matures or sees innovation and so reduces the
4.5
7.5 cost and MSP of the CIGS absorber and enables a cost-competitive
tandem. Additionally, the 2T is favoured over the 4T in utility scale
because the modules are cheaper per watt, other than in the humid
7 location, because the 2T suffers considerably from spectral mis-
4 match, as discussed previously.
Generally, if the price of both subcells decreases, tandems
6.5
become cost competitive, which shows the advantage of tandems
for very-low-cost absorbers. This is because those module compo-
nents that are not doubled when going from a SJ to a tandem, such
ry

te

id

ry

te

id
um

um
D

ra

ra

as glass and the junction box, as well as many aspects of the instal-
pe

pe
H

H
m

m
Te

Te

lation costs, are a larger fraction of the total system cost. Tandems
Fig. 4 | Reduced system cost scenario LCOE values. a,b, Calculated LCOEs
then provide an increase in efficiency with only a small percent
for residential-scale (a) and utility-scale (b) systems in each of the three
increase in price.
specified locations for each architecture for the hypothetical future reduced
We then perform the same analysis of varying the absorber
system and installation cost scenario.
MSP but for the RSCS (Fig. 6b). This shows us that, for signifi-
cantly reduced system costs, whereas tandems are the lowest resi-
dential LCOE option for dry and temperate locations, the CIGS
architecture changes with each. Fig. 5 shows the stability of the absorber MSP would need to drop by about 25% for a tandem to
cheapest architecture remaining the cheapest and how much each have the lowest residential LCOE in the humid location. In this
type of cost would need to change to alter the outcome. scenario, the differences in relative energy yield between the 2T
As expected, as the system costs, especially the area- and project- and 4T tandems among locations becomes more relevant, as the
dependent costs, decrease, tandems become less favourable. In tandem architecture that has the lower LCOE varies across loca-
other words, the economic window of opportunity for tandems tions. The 2T tandem dominates the majority of the parameter
narrows as the trend of system cost reduction continues. The total space in which tandems have the lowest LCOE in the dry location
2016 system costs have to come down substantially (over 50%) for and both architectures are somewhat equally represented for the
a SJ device to have a lower LCOE than a tandem one in residential temperate location, whereas the 4T architecture is always cheaper
systems in temperate and dry locations, and the 4T architecture has than the 2T architecture in the humid location. As the lower sys-
a fairly stable financial preference over fluctuations in system costs. tem cost diminishes the premium on a high energy yield, the 2T
The SJ cannot become cheaper than the 4T tandem without substan- tandem is much more competitive in residential systems in this
tially reducing the project costs, such as permitting, interconnection scenario than in the 2016 scenario. In the residential RSCS, every-
fees and overheads. In the dry location, there is a narrow range, at where that the 2T tandem has the lowest LCOE in this parameter
46–52% total system-cost reduction, in which the 2T tandem has space, the 4T tandem has the second-lowest LCOE, lower than
the lowest LCOE. As the system costs drop more, the advantage of either SJ architecture.
the 2T tandem does not last and the CdTe SJ quickly becomes the The dry and temperate residential cases in this scenario also
lowest LCOE architecture. The 4T LCOE remains lower than CdTe illustrate how absorber cost affects the comparison of 2T and 4T
until the system costs drop by 52%. tandems. Cheap subcells favour 2T tandems, which makes them
Finally, we explore the effect of absorber MSP on the lowest more cost-effective relative to 4T, as demonstrated by the lowest
LCOE architecture, as this is one of the most impactful parameters LCOE changing from 4T to 2T moving towards the bottom-left cor-
on the viability of tandems compared to their SJ counterparts. We ner (Fig. 6b); however, in the humid location, the 2T disadvantage is
co-varied the MSP of the CdTe and CIGS absorbers and mapped too large for this effect to make 2T tandems have the lowest LCOE.
which architecture has the lowest LCOE for both the 2016 scenario For utility-scale systems, the CIGS absorber cost needs to be
(Fig. 6a) and RSCS (Fig. 6b), which allows us to explore the effect about 20% lower for tandems to have a chance of succeeding in dry
of potential future changes in absorber costs, uncertainties in our and temperate locations for this hypothetical future scenario. This
model’s absorber costs and different deposition methods. The latter seems feasible, as it is likely that module manufacturing costs will
is particularly interesting as it gives the opportunity to evaluate the come down in the time needed for system and installation costs to
value, or lack thereof, of very-low-cost deposition methods, such as drop, and thus offset some of the impact of the lower installation
solution-based coating15,16. costs on tandem viability. In the humid location, the CIGS absorber
From this analysis, for 2016 residential systems, the 4T tandem price would need to fall by about 70%, which appears improbable
has a stable advantage and remains the cheapest option over a wide given current technology and material costs.
range of absorber MSPs. For all three climates, even as subcells Finally, we expect the R&D needed to bring tandem technology
become cheap, the 4T is always preferred over the 2T tandem. This, to market to be less for 4T than for 2T technology. The 4T archi-
again, shows the value of high-energy yield for rooftop solar. tecture comprises two mature materials and can be manufactured
For 2016 utility-scale systems, as the cost of the absorbers using their currently preferred processes. The 2T tandem, on the
decrease, the 2T tandem becomes the lowest LCOE architecture in contrary, requires the development of a tunnel junction, the devel-
dry and temperate locations (Fig. 6a, red) and the 4T tandem in the opment of a high-Eg top cell and potentially optimizing the super-
humid location (Fig. 6a, blue). This shows that the price of CIGS strate deposition of CIGS. This is a vital consideration that is not
needs to drop for a tandem in a utility-scale system to be favoured. directly accounted for in these results. Additionally, the advantage

390 Nature Energy | VOL 3 | MAY 2018 | 387–394 | www.nature.com/natureenergy

© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
Nature Energy Articles
a System- b System-
cost type 4T 2T CIGS CdTe cost type 4T 2T CIGS CdTe
Project Project

location location location

location location location

Dry Temperate Humid

Dry Temperate Humid

Area Area
Total Total

Project Project
Area Area
Total Total

Project Project
Area Area
Total Total
–100 –50 0 +50 +100 –100 –50 0 +50 +100
% change (cost) % change (cost)

Fig. 5 | Analysis of the impact of system-cost variability. a,b, Maps of which architecture has the lowest LCOE as various system-cost parameters are
varied relative to their 2016 scenario reference value (given in Supplementary Table 7) for residential-scale (a) and utility-scale (b) systems in each location.
The colour in each bar represents the architecture with the lowest LCOE for the relative value given on the x axis of the specified parameter (project-
dependent, area-dependent or total system cost) given along the y axis. The blue region represents the conditions for which the 4T tandem LCOE is lowest,
the red correspondingly for the 2T tandem, the yellow for the CIGS SJ and the green for the CdTe SJ. The dotted white lines mark the values for the RSCS.

a Dry Temperate Humid

40 40 40

CIGS
30 30 30
Residential
CIGS absorber MSP (US$ m–2)

20 20 20

10 10 10

CdTe
0 0 0
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40
40 40 40

2T
30 30 30
Utility

20 20 20

4T
10 10 10

0 0 0
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40
–2
CdTe absorber MSP (US$ m )

b Dry Temperate Humid

40 40 40

CIGS
30 30 30
Residential

20 20 20
CIGS absorber MSP (US$ m–2)

CdTe

10 10 10

0 0 0
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40
40 40 40
2T

30 30 30
Utility

20 20 20
4T

10 10 10

0 0 0
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40
CdTe absorber MSP (US$ m–2)

Fig. 6 | Analysis of the impact of the absorber price. a,b, Maps of the architecture with the lowest LCOE over varied MSP for the CdTe and CIGS absorber
for residential-scale (top row) and utility-scale (bottom row) systems in the dry, temperate and humid locations assuming the 2016 scenario (a) and the
hypothetical RSCS installation costs (b). The blue region represents where the 4T tandem LCOE is lowest, the red is where the 2T tandem LCOE is lowest,
the yellow where the CIGS SJ LCOE is lowest and the green is where the CdTe SJ LCOE is lowest. The white dashed lines correspond to the absorber cost
specified by the cost model.

of 4T tandems is even more significant when considering land scar- Though some of the costs used in this model may vary or have
city; rooftop PV systems have limited space, which makes efficiency some uncertainty, because we frame our analysis as a comparison
more valuable. between architectures and focus on the relative LCOE values rather

Nature Energy | VOL 3 | MAY 2018 | 387–394 | www.nature.com/natureenergy 391

© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
Articles Nature Energy

than their magnitude, the effect of error has significantly less impact Finally, this analysis shows the importance of using energy yield
on our conclusions. This is because many of the process steps are when tandem viability is assessed, as using STC efficiency alone
used in multiple or all architectures. results in the 2T tandem having the lowest LCOE in both system
Though the efficiencies used in this work are relatively high sizes for both system cost scenarios.
compared to commercially available modules, we do not expect this
to have a significant impact on our results. High efficiencies have Methods
been shown to have diminishing returns on reducing the cost of Device efficiency and yield calculations. Due to the 2T tandem’s series-connected
system installation17. Thus, we expect lower efficiencies to show the configuration, the subcells are forced to generate equal current, a constraint
known as current matching. If there is a mismatch in current generation, the
same result, but for the relative benefit of the lowest LCOE architec- device is limited by the cell with the least current. This causes 2T devices to be
ture to increase. more sensitive to subcell bandgap pairings and can additionally cause significant
losses in the harvesting efficiency of the tandem under a naturally varying solar
Beyond CdTe–CIGS tandems spectrum27,28. As a result, a CdTe-on-CIGS 2T tandem is not viable due to the
degree of current mismatch with such a bandgap pairing. We therefore, instead,
The results demonstrated provide knowledge and have implica-
consider a high-Eg CdTe alloy, such as CdZnTe, with a bandgap of 1.68 eV (ref. 11)
tions for examining the cost-effectiveness of tandems that comprise as a top cell on a CIGS bottom cell. We assume a top-cell device performance
other materials systems. For thin-film tandems with absorber costs similar to that of current CdTe devices, despite current alloys such as CdZnTe
similar to CdTe and CIGS, the best avenue for a cost-competitive being much less mature and having not yet reached comparable efficiencies.
tandem is a 4T tandem architecture, used in residential systems in Currently, the record CdZnTe efficiency is 16.4%, and a 16.8% CdZnTe 2T tandem
on silicon has been demonstrated29.
dry and temperate locations. This has the advantage that areas with The 4T tandem configuration, however, has no current matching constraint,
significant residential solar markets tend to have primarily cooler, as the subcells are electrically independent, each with their own set of contacts and
dryer climates. To pursue a cost-competitive 2T tandem, low-cost an insulating layer between the top and bottom cell. This allows more flexibility
subcells should be used, such as solution-processed perovskite-on- in bandgap pairings of subcells and decreases the sensitivity to spectral changes.
perovskite tandems18–20 (assuming similar contact and module costs Thus, we can leverage two mature materials as subcells in the 4T tandem, CdTe
(1.48 eV bandgap) and CIGS (1.04 eV bandgap). Though the higher bandgap CdTe
can be achieved). alloy (1.68 eV bandgap) is a better bandgap pairing, the difference in absolute
Perovskite-on-silicon marks another popular tandem pair- efficiency is small: 1.6%, assuming detailed balance-efficiency limits, and 0.4% for
ing21–23. Though the fabrication process of silicon is quite different the parameters used in this work.
from those described in this paper, we expect the key trends and The efficiency and energy yield for each device is calculated using published
findings to hold. Though the cost of the two absorbers are prob- methodology11 for each location. We have updated the calculations to reflect
recent advances in device performance. The CdTe quantum-efficiency curve
ably unequal, the analysis in this paper (Fig. 6a) suggests that it used previously11 was replaced with that of the most-recent record CdTe cell from
may be possible for the 4T perovskite-on-silicon tandem to have a First Solar30, and we also used the quantum-efficiency curve from a recent CIGS
lower LCOE than either SJ module, provided a similar SJ-to-tandem solar cell, based on device results from Solar Frontier31, as this device is fabricated
efficiency boost to that of CdTe–CIGS is achieved. As system costs by sputtering and sulfurization after selenization (SAS). We then added cell-to-
come down (Fig. 6b), however, the unequal absorber costs make it module losses with increased series resistance, reflection and inactive area between
the cells (Supplementary Table 8 and Supplementary Methods give details). The
quite difficult for the tandem to have a lower LCOE than a low-cost new I–V curve parameters and the external quantum efficiency (EQE) were then
perovskite solar cell by itself. used to calculate the STC efficiency (Table 1). Our choice to use record-level
Perovskite-on-CIGS tandems have also been explored24–26. efficiency solar cell results was motivated by the aim to address the next-generation
Similar to CdTe–CIGS, a perovskite–CIGS tandem may be viable PV modules in this work and to anticipate the rapid progress of PV development.
in residential systems, even as system costs decrease, if the absorber Energy yield was calculated using a published approach11 that utilizes
time-resolved spectra for each location generated with the Simple Model for
cost of CIGS falls, again assuming similar device performances. A Atmospheric Radiative Transfer of Sunshine (SMARTS) from National Renewable
larger reduction in CIGS cost is needed for a lower-cost top cell, Energy Laboratory32,33. As previously11, the efficiency for each architecture was
however, and the 2T configuration may then be favoured. calculated using the one-diode model. The input parameters were determined by
the experimental open-circuit voltages, the EQE for each device and fitting the
series and shunt to the published I–V curves30,31. As SMARTS does not account
Conclusions for cloud coverage, thus overestimating annual insolation, the yields for each
In this paper, we evaluate the financial viability of CdTe- and CIGS- location were scaled by the ratio of the local average insolation34 and the modelled
based thin-film tandem solar cells using a detailed bottom-up cost insolation (Supplementary Table 9). The resulting energy yields for all four
model and energy-yield calculations. architectures are shown in Table 1.
We find that for residential systems, the 4T tandem, hav-
ing the highest efficiency, features the lowest LCOE given 2016 Module manufacturing cost model. In this section, we describe our bottom-up
system costs. In utility-scale systems, tandems have no LCOE manufacturing cost models for the two tandem modules and their comprising
subcells, employed as SJ cells. The assumed step-by-step manufacturing process
advantage, and the lowest LCOE is achieved by the technology flows for all four architectures are listed in Supplementary Tables 2–5. In the
with the lowest US$ W–1 cost, the CdTe SJ. As installation costs described process, CdTe is deposited in a superstrate configuration by vapour
come down, tandems have less of an advantage as the importance transport35,36 and CIGS is deposited in a substrate configuration by sputtering plus
of energy yield is reduced; however, the total system cost must SAS, as these are the predominant commercial methods of fabrication and the
come down by almost 50% for residential systems before the 4T method by which the quoted devices were deposited31.
For the 4T tandems (Fig. 1c), modification of the process flow for each subcell
tandem no longer has the lowest LCOE. In other words, the eco- is straightforward, as the superstrate top-cell and substrate bottom-cell fabrication
nomic window of opportunity for tandems shrinks when the BOS are naturally compatible with integration in the tandem structure. The top- and
costs decline. bottom-cell processes are assumed to flow in parallel until each subcell has the
Tandems become financially more favourable as the price of final contact deposited. The two subcells, each integrated onto a glass sheet, are
both subcells decreases. As absorbers become cheap, 2T tandems then laminated together, assuming similar loading and encapsulation costs to those
of standard SJ glass–glass module encapsulation. We assume no added module
become more favourable relative to 4T tandems and SJs; however, circuitry or installation cost for the 4T tandem, as subcells would connect in
whether 2T tandems have a lower LCOE than 4T tandems depends parallel voltage-matched strings37. If additional circuitry is needed, however, the 4T
on location and system costs. Furthermore, the 2T tandem currently module can cost up to an additional US$8.87 m–2, US$12.54 m–2 or US$23.06 m–2
requires significant R&D to achieve an effective tunnel junction and and remain the cheapest residential LCOE in the dry, temperate and humid
high-bandgap top cell material to enable this device architecture, locations, respectively, as is the case in Fig. 3.
Despite the significant R&D barriers for the 2T tandem (Fig. 1d), we assume
which potentially results in a more expensive fabrication process that the cost of the high-Eg II–VI top-cell absorber is equivalent to that of CdTe.
than anticipated. For tandems to succeed on the utility scale, the We consider a superstrate configuration (front-to-back process) for the 2T tandem.
absorber costs need to be substantially lower. First, the CdTe alloy top cell is deposited by vapour transport deposition on the

392 Nature Energy | VOL 3 | MAY 2018 | 387–394 | www.nature.com/natureenergy

© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
Nature Energy Articles
front contact, followed by the tunnel junction, which is deposited by sputtering, 6. Bobela, D. C., Gedvilas, L., Woodhouse, M., Horowitz, K. A. W. & Basore, P.
and, last, the rest of the CIGS absorber and back contact are deposited, front- A. Economic competitiveness of III–V on silicon tandem one-sun
to-back, on to the CdTe–tunnel junction stack. This requires four primary R&D photovoltaic solar modules in favorable future scenarios. Prog. Photovolt. Res.
challenges: the development of the high-Eg CdTe alloy top cell, a polycrystalline Appl. 15, 41–48 (2016).
tunnel junction, a process flow that prevents copper diffusion and a high-efficiency 7. Nanayakkara, S. U., Horowitz, K., Kanevce, A., Woodhouse, M. & Basore, P.
superstrate deposition method for CIGS. Evaluating the economic viability of CdTe/CIS and CIGS/CIS tandem
We have derived the cost for each step through supplier quotes, discussions photovoltaic modules. Prog. Photovolt. Res. Appl 25, 271–279 (2017).
with contacts in industry and other academic and non-academic sources35,38–43, 8. Azzopardi, B. et al. Economic assessment of solar electricity production from
conscious of recent changes in manufacturing costs. The costs associated with each organic-based photovoltaic modules in a domestic environment. Energy
process step for module manufacturing are given in Supplementary Table 1. From Environ. Sci. 4, 3741 (2011).
the process costs and manufacturing process flows, we compute the manufacturing 9. Louwen, A., Van Sark, W., Schropp, R. & Faaij, A. A cost roadmap for silicon
cost-per-area for each of the four architectures. From these costs, we then compute heterojunction solar cells. Sol. Energy Mater. Sol. Cells 147, 295–314 (2016).
the MSP using an iterative goal-seek algorithm assuming straight-line depreciation. 10. Peters, I. M., Sofia, S., Mailoa, J. & Buonassisi, T. Techno-economic analysis
The MSP is defined as the minimum price for which the internal rate of return of tandem photovoltaic systems. RSC Adv. 6, 66911–66923 (2016).
(IRR) is equal to the weighted average cost of capital (WACC), a metric used to 11. Mailoa, J. P. et al. Energy-yield prediction for II–VI-based thin-film tandem
define a price that will sustain a manufacturer1. All the financial parameters used in solar cells. Energy Environ. Sci. 9, 2644–2653 (2016).
these calculations are given in Supplementary Table 6. 12. Peters, I. M., Liu, H., Reindl, T. & Buonassisi, T. Global prediction of
photovoltaic field performance differences using open-source satellite data.
LCOE. The LCOE is computed with the equation44: Joule 2, 307–322 (2018).
13. Fu, R. et al. US Solar Photovoltaic System Cost Benchmark: Q1 2016 NREL/
total lifetime cost TP-6A20-66532 (NationalRenewable Energy Laboratory, 2016).
LCOE =
total lifetime electricity production 14. Jones-Albertus, R., Feldman, D., Fu, R., Horowitz, K. & Woodhouse, M.
I+ ∑
N OM (1) Technology advances needed for photovoltaics to achieve widespread grid
i = 0 (1 + r) i price parity. Prog. Photovolt. Res. Appl. 24, 1272–1283 (2016).
=
N E (1 − d) i 15. Cai, M. et al. Cost-performance analysis of perovskite solar modules. Adv. Sci.
∑i =0 (1 + r) i 4, 1600269 (2016).
16. Liu, D. & Kelly, T. L. Perovskite solar cells with a planar heterojunction
where I is the total initial investment to install the PV system (including cost of structure prepared using room-temperature solution processing techniques.
PV modules, racking, interconnects, labour, permits and so on), OM is the annual Nat. Photon 8, 133–138 (2014).
cost for operation and maintenance, E is the annual energy output by the system 17. Goodrich, A., James, T. & Woodhouse, M. Residential, Commercial, and
as electricity in the first year (Table 1), N is the system lifetime in years, d is the Utility-Scale Photovoltaic (PV) System Prices in the United States: Current
annual module degradation rate and r is the nominal discount rate for the customer Drivers and Cost-Reduction Opportunities NREL/TP-6A20-53347 (National
(Supplementary Table 6). We use an annual relative degradation rate of 0.5% yr–1 for Renewable Energy Laboratory, 2012).
both CdTe and CIGS45 in all locations. We assume both tandems’ subcells degrade 18. Jiang, F. et al. Two-terminal perovskite/perovskite tandem solar cell. J. Mater.
at the same rate as in SJ devices, so we use the same 0.5% yr degradation rate for Chem. A. 4, 1208–1213 (2015).
both tandems. 19. Eperon, G. E. et al. Perovskite–perovskite tandem photovoltaics with
To calculate the total installation cost of a system, we use data13 for the 2016 optimized band gaps. Science 354, 861–865 (2016).
US PV system costs for utility- and residential-scale systems. The costs are broken 20. Rajagopal, A. et al. Highly efficient perovskite–perovskite tandem solar cells
down into those that scale with system area and system power, and those that are reaching 80% of the theoretical limit in photovoltage. Adv. Mater. 29,
fixed per project17 (Supplementary Table 7). These include all costs of installing 1–10 (2017).
a PV array (excluding the PV modules), such as BOS, racking, inverters, labour, 21. Bush, K. A. et al. 23.6%-efficient monolithic perovskite/silicon tandem solar
permitting and so on. To calculate the installation costs for each type of module, cells with improved stability. Nat. Energy 2, 17009 (2017).
we assume a constant area of 35.9 m2 for residential systems and 0.6 km2 for utility- 22. Duong, T. et al. Rubidium multication perovskite with optimized bandgap for
scale systems. The total module cost to the customer is calculated from the module perovskite–silicon tandem with over 26% efficiency. Adv. Energy Mater. 7,
MSP and an installer markup (Supplementary Table 7). 1700228 (2017).
The hypothetical future RSCS with substantially reduced system installation 23. Yu, Z. J., Leilaeioun, M. & Holman, Z. Selecting tandem partners for silicon
costs is based on the US Department of Energy SunShot target scenario14 solar cells. Nat. Energy 1, 16137 (2016).
(Supplementary Table 7). We use the US$ W–1 costs from this source14 and break 24. Paetzold, U. W. et al. Scalable perovskite/CIGS thin-film solar module with
them into area-, power- and project-dependent costs based on a published power conversion efficiency of 17.8%. J. Mater. Chem. A 5, 9897–9906 (2017).
breakdown17. A few of the costs in this SunShot scenario14 were higher than the 25. Guchhait, A. et al. Over 20% efficient CIGS–perovskite tandem solar cells.
2016 costs, as they fell faster than anticipated between 2015 and 2016. In these few ACS Energy Lett. 2, 807–812 (2017).
cases, the 2016 cost13 was used instead. 26. Mantilla-Perez, P. et al. Monolithic CIGS–perovskite tandem cell for optimal
light harvesting without current matching. ACS Photonics 4, 861–867 (2017).
Code availability. The cost model and parametric analyses from this paper are 27. Faine, P., Kurtz, S. R., Riordan, C. & Olson, J. M. The influence of spectral
available online at http://pv.mit.edu/tma/ in an excel document. Additional analysis solar irradiance variations on the performance of selected single-junction and
code is available upon reasonable request. multijunction solar cells. Sol. Cells 31, 259–278 (1991).
28. Liu, H. et al. The realistic energy yield potential of GaAs-on-Si tandem solar
Data availability. The data that support the plots within this paper and other cells: a theoretical case study. Opt. Express 23, A382–A390 (2015).
findings of this study are available on pv.mit.edu or from the corresponding 29. Garland, J. W., Biegala, T., Carmody, M., Gilmore, C. & Sivananthan, S.
authors upon reasonable request. Next-generation multijunction solar cells: the promise of II–VI materials. J.
Appl. Phys. 109, 102423 (2011).
30. Green, M. A., Emery, K., Hishikawa, Y., Warta, W. & Dunlop, E. D. Solar cell
Received: 11 July 2017; Accepted: 2 March 2018;
efficiency tables (version 46). Prog. Photovolt. Res. Appl. 23, 805–812 (2015).
Published online: 30 April 2018 31. Nakamura, M. et al. In 2014 IEEE 40th Photovolt. Spec. Conf. 107–110
(IEEE, 2014).
References 32. Gueymard, C. A. SMARTS, A Simple Model of the Atmospheric Radiative
1. Powell, D. M., Winkler, M. T., Goodrich, A. & Buonassisi, T. Modeling Transfer of Sunshine: Algorithms and Performance Assessment Professional
the cost and minimum sustainable price of crystalline silicon Paper FSEC-PF-270-95 (Florida Solar Energy Center, 1995).
photovoltaic manufacturing in the United States. IEEE J. Photovolt. 3, 33. Gueymard, C. A. Parameterized transmittance model for direct beam and
662–668 (2013). circumsolar spectral irradiance. Sol. Energy 71, 325–346 (2001).
2. de Vos, A. Detailed balance limit of the efficiency of tandem solar cells. J. 34. Roberts, B. J. Photovoltaic Solar Resource of the United States (National
Phys. D. 13, 839–846 (1980). Renewable Energy Laboratory, 2012); www.nrel.gov/gis/images/eere_pv/
3. Henry, C. H. Limiting efficiencies of ideal single and multiple energy gap national_photovoltaic_2012-01.jpg
terrestrial solar cells. J. Appl. Phys. 51, 4494–4500 (1980). 35. Modules: Our Technology (First Solar, accessed 26 February 2017); www.
4. Powell, D. M. et al. Crystalline silicon photovoltaics: a cost analysis firstsolar.com/Modules/Our-Technology
framework for determining technology pathways to reach baseload electricity 36. First Solar 2015 Annual Report (First Solar, 2015); http://investor.firstsolar.
costs. Energy Environ. Sci. 5, 5874–5883 (2012). com/static-files/eb9f8191-1f74-46ac-9678-7e118cfdf41f
5. Horowitz, K. A. W., Fu, R. & Woodhouse, M. An analysis of glass–glass CIGS 37. Gee, J. M. A comparison of different module configurations for multi-band-
manufacturing costs. Sol. Energy Mater. Sol. Cells 154, 1–10 (2016). gap solar cells. Sol. Cells 24, 147–155 (1988).

Nature Energy | VOL 3 | MAY 2018 | 387–394 | www.nature.com/natureenergy 393

© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
Articles Nature Energy
38. Fthenakis, V. et al. Life Cycle Inventories and Life Cycle Assessments of conversations at BAPVC are noted. This work additionally benefitted greatly from the
Photovoltaic Systems Life Cycle Inventories and Life Cycle Assessments of prior work of D. M. Powell and S. C. Siah.
Photovoltaic Systems IEA-PVPS T12-02:2011 (International Energy Agency, The CdTe cost model was made independently, without contribution from or
Paris, 2011). corroboration by First Solar. The CIGS cost model was made independently, without
39. Sputtering Yield Rates (Semicore Equipment, Inc.); www.semicore.com/ contribution from or corroboration by Siva Power.
reference/sputtering-yields-reference
40. Indium Tin Oxide (ITO) for Deposition of Transparent Conductive Oxide Author contributions
Layers (Umicore, accessed 1 August 2016); www.thinfilmproducts.umicore. S.E.S. compiled cost data and developed the cost model and analysis tools, with cost
com/Products/TechnicalData/show_datenblatt_ito.pdf inputs and feedback contributed by D.N.W., B.J.S., I.M.P. and T.B. J.P.M. and S.E.S.
41. Richard, D. On the cutting edge: thin-film laser structuring survey. Phot. Int. performed energy-yield calculations. S.E.S. performed analysis and data visualization.
9, 172–195 (2010). I.M.P., T.B. and D.N.W. conceptualized the initial project. The manuscript was written by
42. Electric Power Monthly, Industrial, 2016 (US Energy Information S.E.S. and edited by all the co-authors. I.M.P. provided lead mentorship. All the authors
Administration); https://www.eia.gov/electricity/monthly/epm_table_grapher. reviewed and approved the manuscript.
php?t=​epmt_5_6_a
43. Siah, S.-C. Defect Engineering in Cuprous Oxide (Cu2O) Solar Cells
(Massachusetts Institute of Technology, 2015). Competing interests
44. System Advisory Model (SAM): Financial Model Documentation (National The authors declare no competing interests.
Renewable Energy Laboratory, 2010); https://sam.nrel.gov/financial
45. Jordan, D. C. & Kurtz, S. R. Photovoltaic degradation rates—an analytical
review. Prog. Photovolt. Res. Appl. 21, 12–29 (2013).
Additional information
Supplementary information is available for this paper at https://doi.org/10.1038/
s41560-018-0126-z.
Acknowledgements Reprints and permissions information is available at www.nature.com/reprints.
This works was funded in part by the National Research Foundation Singapore through
the Singapore-MIT Alliance for Research and Technology, the Bay Area Photovoltaic Correspondence and requests for materials should be addressed to S.E.S. or
Consortium (BAPVC) under Contract no. DE-EE0004946, the US Department of T.B. or I.M.P.
Energy under Award no. DE-EE0006707 and the National Science Foundation (NSF) Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in
and Department of Energy (DOE) under NSF CA No. EEC-1041895. Numerous peer published maps and institutional affiliations.

394 Nature Energy | VOL 3 | MAY 2018 | 387–394 | www.nature.com/natureenergy

© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.