Physica E 14 (2002) 65 – 70

www.elsevier.com/locate/physe

Third generation photovoltaics: solar cells for 2020 and beyond

Martin A. Green ∗
Special Research Centre for Third Generation Photovoltaics, University of New South Wales, Sydney, 2052, Australia

Abstract

Many working in the )eld of photovoltaics believe that ‘)rst generation’ silicon wafer-based solar cells sooner or later will
be replaced by a ‘second generation’ of lower cost thin-)lm technology, probably also involving a di.erent semiconductor.
Historically, CdS, a-Si, CuInSe2 , CdTe and, more recently, thin-)lm Si have been regarded as key thin-)lm candidates.
Since any mature solar cell technology is likely to evolve to the stage where costs are dominated by those of the constituent
materials, be it silicon wafers or glass sheet, it is argued that photovoltaics will evolve, in its most mature form, to a ‘third
generation’ of high-e4ciency thin-)lm technology. By high e4ciency, what is meant is energy conversion values double or
triple the 15 –20% range presently targeted. Tandem cells provide the best-known example of such high-e4ciency approaches,
where e4ciency can be increased merely by adding more cells of di.erent band gap to a stack. However, a range of other
better-integrated approaches are possible that o.er similar e4ciency to an in)nite stack of such tandem cells. ? 2002 Elsevier
Science B.V. All rights reserved.

Keywords: Photovoltaics; Solar cells; Conversion e4ciency; E4ciency limits

1. Introduction can be fabricated without unduly complicated cell

processing [1].
Most solar cells presently on the market are based For the past 15 years, a switch to a ‘second
on silicon wafers, the so-called ‘)rst generation’ tech- generation’ of thin-)lm cell technology has seemed
nology. As this technology has matured, costs have imminent. Regardless of semiconductor, thin-)lms
become increasingly dominated by material costs, o.er prospects for a major reduction in material costs
mostly those of the silicon wafer, the strengthened by eliminating the silicon wafer. Thin )lms also o.er
low-iron glass cover sheet, and those of other encap- other advantages, particularly the increase in the unit
sulants. This trend is expected to continue as the pho- of manufacturing from a silicon wafer (∼100 cm2 )
tovoltaic industry continues to mature. A 1997 study to a glass sheet (∼1 m2 ), about 100 times larger. In
[1] of costs of manufacturing in greatly increased terms of energy conversion e4ciency, this ‘second
500 MW=y production volume suggests material costs generation’ technology, with time, might be expected
in such volumes would account for over 70% of total to largely bridge the present gap between itself and
manufacturing costs. This favours high-e4ciency ‘)rst generation’ product.
processing sequences that produce solar cells with As thin-)lm ‘second generation’ technology
high energy conversion e4ciency, provided these matures, costs again progressively will become dom-
inated by those of the constituent materials, in this
∗ Tel.: +61-2-938-540-18; fax: +61-2-966-242-40. case, the top cover sheet and other encapsulants
E-mail address: [email protected] (M.A. Green). required to maintain a 30-year operating life. There

1386-9477/02/$ - see front matter ? 2002 Elsevier Science B.V. All rights reserved.
PII: S 1 3 8 6 - 9 4 7 7 ( 0 2 ) 0 0 3 6 1 - 2
66 M.A. Green / Physica E 14 (2002) 65 – 70

will be a lower limit on such costs (say, US$30=m2 )

which, when combined with likely cell e4ciency Energy
1
(15% or 150 Wp=m2 ), determines the lower limit
on photovoltaic modules costs (US$0:20=Wp) and 2
electricity generation costs (US$0:02–0:04 kWh−1 ), 3
assuming module costs are half of system costs and a
range of insolation and cost-of-money scenarios. 4'
usable
To progress further, conversion e4ciency needs to
4 (qV)
be increased substantially. The Carnot limit on the 3'
conversion of sunlight to electricity is 95% as op-
posed to the theoretical upper limit of 33% for a stan- 1'
dard solar cell. This suggests the performance of solar
cells could be improved 2–3 times if di.erent con- Fig. 1. Loss processes in a standard solar cell: (1) thermalisation
cepts were used to produce a ‘third generation’ of loss; (2) and (3) junction and contact voltage loss; (4) recombi-
nation loss.
high-performance, low-cost photovoltaic product.
Fortunately, with the likely evolution of new mate-
rials technology over the coming decades, prospects limited to 31.0% e4ciency for an optimal cell with a
for thin-)lm cells based on new concepts appear to be band gap of 1:3 eV. This is lesser than the )gure of
quite good! In the following sections, possible third 44% previously mentioned since the output voltage of
generation approaches are reviewed and implementa- the cell is less than the band gap potential, with the
tion strategies are suggested [2]. di.erence made up by voltage drops at the contact and
junction (Fig. 1).
These drops can be reduced if sunlight is focussed
2. Eciency losses in standard cell to increase the photon density striking the cell. Under
the maximum possible sunlight concentration (46,200
Losses in a standard solar cell are shown in Fig. 1. times!), the limiting e4ciency increases to 40.8%.
A key loss process is process 1, whereby the photoex- However, only direct sunlight can be focussed in this
cited pair quickly loses energy in excess of the band way. As the )gure under maximal concentration gives
gap. A low-energy red photon is just as e.ective as a the highest numerical value and also applies to the con-
much higher energy blue photon. Balancing this loss version of direct light even when unconcentrated, this
with the loss of low-energy photons passing straight e4ciency is a useful )gure in comparing the ultimate
through the device alone limits conversion e4ciency potential of any given approach. This e4ciency is also
of a cell to about 44%. more directly comparable with results from classical
Another important loss process is process 4, recom- thermodynamics.
bination of the photoexcited electron–hole pairs. This For example, the conversion e4ciency of energy
can be kept to a minimum by using material with high from a source at 6000 K with a sink temperature
lifetimes for the photogenerated carriers, ensured by of 300 K is limited by the Carnot e4ciency
eliminating all unnecessary defects. The lifetime is (1 − Tsink =Tsource ) to 95.0%. However, this value does
then determined by radiative recombination in the cell, not count the photons emitted by the cell as a waste,
the inverse to photoexcitation. As shown in 1960 [3], since it assumes they get back to the sun, helping
this symmetry between light absorption and light emis- it to maintain its temperature! Some of the schemes
sion can be used to derive quite fundamental limits to be described, however, can approach this limit
on achievable solar cell performance. This approach reasonably closely.
revisits ‘blackbody’ radiation, the topic that stimulated
the birth of quantum mechanics. 3. Tandem cells
By relating the light emitted by an ideal cell to
that emitted by a blackbody, Shockley and Queisser The key loss process 4 of Fig. 1 can be largely
showed that the performance of a standard cell was eliminated if the energy of the absorbed photon is just
 M.A. Green / Physica E 14 (2002) 65 – 70 67

Solar cells the Si:Ge:H alloy system. Modules with e4ciencies

up to 6 –7% are available incorporating double- and
triple-junction devices.

4. Multiple electron-hole pairs

If, instead of giving up their excess energy as heat,

Sunlight the high-energy electron–hole pair used it to create
additional pairs as allowed by energy conservation,
higher e4ciency would be possible. Evidence for the
Decreasing band gap creation of more than one pair by high-energy photons
Fig. 2. Tandem cell approach.
is documented [5], attributed to impact ionization by
the photoexcited carriers.
The limiting e4ciency for an idealised cell capa-
ble of taking full advantage of this impact ionisation
a little higher than the cell band gap. This leads to e.ect is calculated as 85.9% for a cell of band gap ap-
the tandem cell concept, where multiple cells are used proaching zero. This design allows, on energy grounds
each with di.erent band gaps and each converting a at least, many electron–hole pairs to be generated by
narrow range of photon energies close to its band gap. each incident photon. In reality, the measured e.ect to
Fortunately, just stacking the cells with the high- date is so weak so as to produce negligible improve-
est band gap cell uppermost as in Fig. 2 automatically ment in device performance. Competitive processes
achieves the desired )ltering. Performance increases for the relaxation of the high-energy photoexcited
as the number of cells in the stack increases, with a carriers are too e4cient.
direct sunlight conversion e4ciency of 86.8% calcu- A more recent idea is based on Raman lumines-
lated for an in)nite stack of independently operated cence [2] (Fig. 3). Raman scattering is a generic term
cells [4]. applied to the inelastic scattering of photons (scat-
Having to independently operate each cell is a com- tering that results in a change in photon energy and
plication best avoided. Usually, cells are designed with also, usually, in direction). Formally, the scattering
their current outputs matched so that they can be con- involves creation of a ‘virtual’ electron–hole pair by
nected in series. This constraint reduces performance.
More importantly, it makes the design very sensitive
to the spectral content of the sunlight. Once the output
current of one cell in a series connection drops more Eu
than about 5% below that of the next worst, the best for
overall performance is to short-circuit the low-output
cell, otherwise it will consume, rather than generate
power. hf in hf out E th
Tandem cells are now in commercial produc-
tion. Double and triple junction cells based on
GaInP=GaAs=Ge have been developed for use on
spacecraft with terrestrial e4ciencies approaching
30%. Quadruple junction devices with e4ciencies
approaching 40% are presently under development.
Tandem cells are also used to improve the perfor- EL
mance and reliability of terrestrial amorphous silicon
cells with stabilised e4ciencies up to 12% con-
)rmed for small-area triple-junction cells based on Fig. 3. Device based on Raman luminescence.
68 M.A. Green / Physica E 14 (2002) 65 – 70

the photon in a process that conserves momentum but

not necessarily energy. The virtual pair remains viable
for a short time determined by the energy imbalance.
During this period, for the Raman luminescence, the CB

virtual pair relaxes emitting a photon of an energy that 1 2 3 4 5 6 7 8

di.ers from that of the original photon by the energy VB

of the generated electron–hole pair. An analysis of the
e4ciency of cells based on the Raman luminescence
gives identical bounds to those based on impact ioni- t < 0 t = 0+ <1ps -1ps -1ns ~1µs
sation. In principle, 85.9% e4ciency is possible from Time
such cells. The di.erence may prove to be in the prac-
Fig. 4. Energy relaxation of carriers after a short, high-intensity
ticality of implementation. laser pulse at t = 0.

5. Hot carrier cells important. The distribution of electrons and holes

retain the same general shape, determined by the
When photoexcited carriers collide elastically with ambient temperature, but the number of carriers at
one another, no energy is lost. It is inelastic collisions each energy reduces until )nally reaching the levels
with the atoms of the cell material that result in an prior to the laser pulse.
energy loss (through phonon emission). In principle, if A standard cell is designed to collect the carriers
such atomic collisions can be avoided during the time before they get too far into the latter recombination
it takes a photogenerated carrier to traverse the cell, stage of this decay sequence. A hot carrier cell has to
the energy loss associated with process 1, of Fig. 1 catch them before the carrier cooling stage. Carriers
can be avoided. either have to traverse the cell very quickly or cooling
The various time constants can be appreciated by rates have to be slowed in some way. Special con-
imagining a direct band gap cell illuminated by a short tacts to prevent the contacts from cooling the carriers
pulse of monochromatic light such as from a laser. may also be required [6,7]. The limiting e4ciency
Such a pulse would create electrons in the conduction of this approach is 86.8% as for an in)nite tandem
band and holes in the valence band of distinct energy cell stack. However, to reach this limiting e4ciency,
and momentum as in Fig. 4. Collisions of these carri- carrier cooling rates would have to have been reduced
ers occur in less than a picosecond, tending to smear su4ciently or radiative recombination rates su4-
this distribution. The peaked distributions will become ciently accelerated that the latter was faster than the
broader and tend towards a Boltzmann type of distri- former. A particular band structure also is required.
bution. If carriers collide elastically only with carriers
of the same type, no energy is lost from this group. The
temperature of the ‘hot carrier’ distribution will be de- 6. Multiband cells
termined by the total number of carriers created by the
laser pulse and the total energy given to each carrier Standard cells rely on excitations between the
type. Di.erent temperatures are possible for electrons valence and conduction band. A recent analysis [8]
and holes unless e4cient at sharing their energy. has shown advantages if a third band, nominally an
In the next phase, collisions with the lattice atoms impurity band, is included in the analysis (Fig. 5).
become important. These result in energy loss (phonon This theory has been extended to an n-band cell
emission). During this phase, the number of electrons and additional implementation approaches discussed.
and the number of holes remain constant, but the aver- These include using excitations between minibands
age energy and carrier temperature decrease due to this in superlattices, if phonon relaxation processes can be
loss. The temperature of electrons and holes equalise controlled, the use of semiconductors with multiple
and both reduce towards that of the host material. narrow bands, such as that reported for I–VII and
Finally, recombination in the semiconductor becomes I3 –VI compounds, or the use of high concentrations
 M.A. Green / Physica E 14 (2002) 65 – 70 69

finite
band 1
width 2
1 2 hopping
3

Fig. 7. Multiple quantum well solar cell meeting the constraints

of three-band theory.
Fig. 5. Three-band solar cell.

impurity levels and shown that most of the bene)ts

4 also transfer to such ‘impurity photovoltaic’ devices,
4 if properly designed.
3 3

2
2
7. Thermophotovoltaic and thermophotonic devices
1
Thermophotovoltaics is a well-established branch
1
of photovoltaics where a light from a heated body
other than the sun is used as the illumination. A
Fig. 6. Four-band cell and equivalent circuit.
recent development of this approach has been dubbed
‘thermophotonics’ [2]. In this case, the exponentially
of impurities such as rare earths to form multiple enhanced light output of a device where the light is
impurity bands in wide band gap semiconductors. generated by band-to-band recombination is used to
The limiting e4ciency for an n-band cell is identical advantage.
to the 86.8% )gure for a large stack of tandem cells. Fig. 8 shows the basic arrangement which is nearly
However, the e.ective cell connections in the n-band symmetrical. Two diodes acting as solar cells=light
approach show much more redundancy than in a emitters face each other, connected by a load. Heat
series connected tandem cell (Fig. 6). This suggests is supplied to one to heat it hotter than the other
the approach may be more tolerant to spectral varia- and heat is extracted from the other to cool it. The
tions in sunlight. devices are optically coupled but thermally isolated.
Recent work [9] has also already resolved a con- The combination is able to convert heat supplied to
troversy as to whether an idealised cell incorporating the hotter device to electricity in the load with an
multiple quantum wells can exceed the e4ciency of e4ciency approaching the Carnot e4ciency, in prin-
an idealised standard cell. The structure of Fig. 7 ciple. Basically, the heated device acts as an emitter
shows a multiple quantum well cell that meets all the of narrow bandwidth light within an energy, kT , of its
requirements, in principle, to attain limiting three-band band gap energy. This near-monochromatic light can
cell performance. The previous question can now be converted very e4ciently by the cell. Moreover,
be answered in the a4rmative! In fact, a source of light emitted by the cell is recycled back to help drive
energy other than photons can be used for excitation the light-emitting diode. Since the same current Pows
3 of Fig. 7, provided it occurs through sources with in the cell and source diode, the voltage across the
)nite chemical potential (e.g., hot phonons). diode will be smaller than that across the cell when the
The author’s group has recently extended the diode is at higher temperature. This results in power
multi-band approach to devices with discrete mid-gap dissipation in the load.
70 M.A. Green / Physica E 14 (2002) 65 – 70

R 8. Conclusion

IL IC The author’s prognosis is that material costs will,

with the fullness of time, push photovoltaic evolution
in the direction of the highest possible e4ciency.
Work to date suggests there is scope for improving
solar cell performance by exploring approaches
capable of giving e4ciencies closer to thermodynamic
QL QC limits. Low-dimensional structures seem to show
P N N P some promise due to the small dimensions and new
features o.ered. A new Centre for Third Generation
Photovoltaics commenced operation at the University
of New South Wales in early 2000 supported by the
Australian Research Council (ARC), with a 9-year
programme to bring the most promising of these
approaches to the ‘proof-of-concept’ level.
TL I TC
References
Fig. 8. Thermophotonic conversion (conceptual only, any non-
radiative recombination would necessitate storage elements). [1] T.M. Bruton, G. Luthardt, K-D. Rasch, K. Roy, I.A. Dorrity,
B. Garrard, L. Teale, J. Alonso, U. Ugalde, K. Declerq,
J. Nigs, J. Szlufcik, A. Rauber, W. Wettling, A. Vallera,
A study of the manufacture at 500 MWp p.a. of crystalline
With on-going evolution in device design, both silicon photovoltaic modules, Conference Record, 14th
experimental solar cells and light emitting diodes are European Photovoltaic Solar Energy Conference, Barcelona,
approaching the stage where internal recombination June, 1997, pp. 11–16.
[2] M.A. Green, Prog. Photovoltaics 9 (2001) 123.
can be limited by radiative processes, a prerequisite [3] W. Shockley, H.J. Queisser, J. Appl. Phys. 32 (1961) 510.
for this scheme. If used to convert solar radiation in [4] A. Marti, G.L. Araujo, Sol. Energy Mater. Sol. Cells 43 (1996)
conjunction with a thermal absorber, energy conver- 203.
sion e4ciency up to 85.4% is obtainable in principle. [5] S. Kolodinski, J.H. Werner, T. Wittchen, H.J. Queisser, Sol.
Alternatively, the approach could be used for max- Energy Mater. Sol. Cells 33 (1994) 275.
[6] R.T. Ross, A.J. Nozik, J. Appl. Phys. 53 (1982) 3813.
imally e4cient conversion of fossil fuels or waste [7] P. Wurfel, Sol. Energy Mater. Sol. Cells 46 (1997) 43.
heat, particularly when the heat is available at low [8] A. Luque, A. Marti, Phys. Rev. Lett. 78 (1997) 5014.
temperature. [9] M.A. Green, Mater. Sci. Eng. B 74 (2000) 118.