2014-Green-The Emergence of Perovskite Solar Cells
2014-Green-The Emergence of Perovskite Solar Cells
2014-Green-The Emergence of Perovskite Solar Cells
net/publication/280388277
CITATIONS READS
4,936 25,199
3 authors, including:
Some of the authors of this publication are also working on these related projects:
Development of light-trapping schemes in evaporated laser-crystallised silicon thin-film solar cells on glass superstrates View project
All content following this page was uploaded by Martin Green on 26 July 2015.
R
ecent substantial reductions in the manufacturing costs of ethylammonium (CH3CH2NH3+) (RA = 0.23 nm)9 and forma-
mainstream silicon solar cell technology assure the future midinium (NH2CH=NH2+) (RA is estimated to lie in the range
large-scale use of photovoltaics, with a recent forecast antici- 0.19–0.22 nm) also give good results10–13. The anion X is a halo-
pating photovoltaics will contribute nearly a third of new electric- gen, generally iodine (RX = 0.220 nm), although Br and Cl are also
ity generation capacity worldwide between now and 20301. As in commonly used (RX = 0.196 nm and 0.181 nm), usually in a mixed
microelectronics, silicon has a combination of strengths that has halide material. For efficient cells, cation B has universally been Pb
made it difficult to displace as the favoured photovoltaic material. (RB = 0.119 nm); Sn (RB = 0.110 nm) forms similar compounds with
Opportunities exist for technologies that promise either signifi- lower, theoretically more ideal bandgaps14, but generally lower sta-
cantly higher energy conversion efficiencies or significantly lower bility (attributed to the ease of oxidation of Sn to SnI4 in the iodide
processing costs. A new generation of mixed organic–inorganic hal- perovskite; relativistic effects in the Pb counterpart are thought to
ide perovskites offers tantalizing prospects on both fronts2–6. provide greater protection against oxidation14). The archetypal com-
Some key attributes of these perovskites include ease of fabrica- pound is thus methylammonium lead triiodide (CH3NH3PbI3), with
tion, strong solar absorption and low non-radiative carrier recom- mixed halides CH3NH3PbI3−xClx and CH3NH3PbI3−xBrx also being
bination rates for such simply prepared materials, plus the ability to important. Calculated and estimated t and μ factors for a range of
capitalize on over 20 years of development of related dye-sensitized these perovskites are shown in Fig. 1b.
and organic photovoltaic cells. A reasonably high carrier mobil-
ity is an important property for some cell architectures, as is the Notable achievements to date
range of properties accessible by forming mixed compounds within Interest in organic–inorganic halide perovskites dates back over
a compatible materials system. One negative aspect of perovskites a century 15, although present results benefit from recent investi-
is the fact that lead has been a major constituent of all highly per- gations of related thin-film transistors and light-emitting diodes
forming perovskite cells to date, raising toxicity issues during device (LEDs) by Mitzi and co-workers16,17. Although these LEDs doubt-
fabrication, deployment and disposal. Also, they generally undergo less exhibited photovoltaic properties and their use in solar
degradation (sometimes quite rapid) on exposure to moisture and cells was anticipated17, this was not studied at the time, partly
ultraviolet radiation. because of Pb toxicity and the fact that more benign Sn materials
Perovskites are materials described by the formula ABX3, where were not considered to be sufficiently robust (D. Mitzi, personal
X is an anion and A and B are cations of different sizes (A being communication).
larger than B). The crystal structure of perovskites is depicted in Miyasaka and co-workers18–20 were apparently the first to report
Fig. 1a. Their crystallographic stability and probable structure can photovoltaic results for perovskites; they were attracted by the self-
be deduced by considering a tolerance factor t and an octahedral organization potential of perovskite in the nanoporous TiO2 layer of
factor μ (ref. 7); here, t is defined as the ratio of the distance A-X to dye-sensitized cells. In 2006, they reported CH3NH3PbBr3 cells with
the distance B-X in an idealized solid-sphere model (t = (RA + RX)/ an efficiency of 2.2% (ref. 18); by replacing bromine with iodine
{√2(RB + RX)}, where RA, RB and RX are the ionic radii of the corre- they were able to increase the efficiency to 3.8% in 200919 (although
sponding ions) and μ is defined as the ratio RB/RX. For halide perovs- all devices were unstable). An organic electrolyte containing lithium
kites (X = F, Cl, Br, I)7, generally 0.81 < t < 1.11 and 0.44 < μ < 0.90. halide and the corresponding halogen formed the hole-transporting
If t lies in the narrower range 0.89–1.0, the cubic structure of Fig. 1a medium (HTM), allowing positive contact. Attempts to replace this
is likely, with lower t values giving less symmetric tetragonal or HTM with a solid-state HTM were not particularly successful20.
orthorhombic structures. Despite these constraints, transitions Subsequently, Park and colleagues21 employed similar structures
between such structures on heating are common for any given per- in which perovskite was shown to be deposited as sparsely spaced
ovskite, with the high-temperature phase generally being cubic. hemispherical nanoparticles that were approximately 2.5 nm in
For the organic–inorganic halide perovskites of present inter- diameter. By applying TiO2 surface treatment prior to deposition,
est, the larger cation A is organic; it is generally methylammo- they realized an efficiency of 6.5% in 201121. Perovskite nanoparti-
nium (CH3NH3+) with RA = 0.18 nm (ref. 8), although related cles exhibited better absorption than standard N719 dye sensitizers,
1
Australian Centre for Advanced Photovoltaics (ACAP), School of Photovoltaic and Renewable Energy Engineering, University of New South Wales, Sydney
2052, Australia. 2University of Oxford, Clarendon Laboratory, Parks Road, Oxford OX1 3PU, UK. *e-mail: [email protected]
0.70
0.60
Octahedral factor, μ
A
0.50
B 0.40
Figure 1 | Perovskite crystal structure and associated tolerance and octahedral factors. a, Cubic perovskite crystal structure. For photovoltaically
interesting perovskites, the large cation A is usually the methylammonium ion (CH3NH3), the small cation B is Pb and the anion X is a halogen ion
(usually I, but both Cl and Br are also of interest). For CH3NH3PbI3, the cubic phase forms only at temperatures above 330 K due to a low t factor (0.83).
b, Calculated t and μ factors for 12 halide perovskites. The corresponding formamidinium (NH2CH=NH2) based halides are expected to have intermediate
values between those of the methylammonium (MA) and ethylammonium (EA; CH3CH2NH3) compounds shown.
but they dissolved in the electrolyte, resulting in a rapid degradation higher t factor due to the smaller ionic radius of Br (molecules in
of performance21. position A exclude full cubic symmetry 29).
This stimulated the replacement of problematic electrolytes by a Further progress was reported at the European Materials
solid-state HTM22,23. Park, Grätzel and colleagues22 introduced a spiro- Research Symposium in May 2013, with two groups reporting
MeOTAD (2,2’,7,7’-tetrakis(N,N-di-p-methoxyphenylamine)-9,9’- efficiencies above 15%. Grätzel’s group used TiO2 scaffolding and
spirobifluorene) HTM, which was developed for organic LEDs24 two-step iodide deposition, which improved the morphology 30.
but was also found to be effective in solid-state dye cells25. When They also reported the first independent measurements of effi-
dissolved in an organic solvent, spiro-MeOTAD penetrates nanopo- ciency by an independent accredited test centre, which confirmed
rous TiO2, leaving only solute molecules after solvent evaporation. an efficiency of 14.1%. (Independent measurements were found
Spiro-MeOTAD not only improved the stability, as expected, it also to be an essential quality-control measure with other photovoltaic
boosted the reported efficiency to 9.7% (ref. 22). The cell structure technologies to prevent inflated results infiltrating the literature31;
is encompassed by the more general device of Fig. 2a if the optional inexperience or overenthusiasm often result in ‘in-house’ data being
continuous perovskite layer is removed, leaving only scaffolding overestimates.) Snaith’s group reported similar results using vastly
infiltrated by perovskite (and subsequently HTM). different planar cells that did not have scaffolding. The simpler
Almost simultaneously (mid-2012), Snaith and co-workers23 structure allowed CH3NH3PbI3−xClx deposition by two-source ther-
also reported success with spiro-MeOTAD along with four addi- mal evaporation, again giving a better morphology 32 and a reported
tional developments that split the field wide open. One of these efficiency of 15.4%.
developments was the use of the mixed-halide CH3NH3PbI3−xClx, No efficiency improvements were announced at the 2013 Fall
which exhibited better stability and carrier transport than its Materials Research Society (MRS) Meeting, although a special per-
pure iodide equivalent 23,26. A second involved going beyond ear- ovskite session attracted considerable attention. At the end of 2013,
lier nanoparticle structures by coating nanoporous TiO2 surfaces Seok’s group achieved an independently confirmed efficiency of
with a thin perovskite layer and thereby forming extremely thin 16.2% by using the mixed-halide CH3NH3PbI3−xBrx (10–15% Br)
absorber (ETA) cells. A third was replacing conducting nanopo- and a poly-triarylamine HTM (S. I. Seok, personal communica-
rous TiO2 by a similar but non-conducting Al2O3 network. This tion). Both optional layers shown in Fig. 2a were included, with the
improved the open-circuit voltage (Voc), boosting the reported thickness ratio of perovskite-infiltrated TiO2 scaffolding relative to
efficiency to 10.9%; it also demonstrated that perovskites have a the continuous perovskite layer being the key to the improved effi-
broader potential than just being used as sensitizers, as they are ciency (S. I. Seok, personal communication). This was increased to
able to transport both electrons and holes between cell terminals. a confirmed efficiency of 17.9% in early 2014 (S. I. Seok, personal
The fourth development exploited such ambipolar transport by communication). An unconfirmed efficiency of 19.3% was reported
demonstrating simple planar cells with the scaffolding (Fig. 2a) at the 2014 Spring MRS Meeting 6 (some uncertainty surrounds this
completely eliminated. reported efficiency due to the absence of supporting information).
A jump to a reported efficiency of 12.0% came from the com- Significantly, the latter three results were obtained using device
bined efforts of Seok, Grätzel and colleagues using both optional structures that span all the possibilities inherent in Fig. 2a: one
layers shown in Fig. 2a, including a solid perovskite capping layer or both optional layers, three different mixed-halide perovskites
overlying the scaffolding 27 (nanoporous TiO2 infiltrated by per- and two different HTMs were used (although some ambiguities
ovskite). Of the HTMs they investigated (which included spiro- remain30,33). Fabrication simplicity combined with similarities with
MeOTAD), poly-triarylamine proved to be the best. Seok’s group dye-sensitized and organic photovoltaics has resulted in a rapid
further improved the performance to realize a reported efficiency increase in the number of researchers working in this field. Before
of 12.3% using similar structures and mixed-halide CH3NH3PbI3− 2013, only seven journal papers had been published that discussed
xBrx perovskites . A low Br content (<10%) gave the best initial photovoltaic devices based on halide perovskites, whereas by the end
28
efficiency due to a lower bandgap, but higher Br contents (>20%) of 2013, relevant publications were appearing at the rate of seven per
provided a better high-humidity stability. This was correlated with month. This spurt further increased the diversity of approaches that
a tetragonal to pseudo-cubic structural transition arising from a gave creditable performances.
Electron
Light energy Electrons
−2.1
−3.9
−4.0
Glass −4.4
Scaffold/
perovskite FTO −5.1
(optional)
Compact TiO2
Perovskite Perovskite
(optional) −5.2
−5.5
HTM
Au −7.2
Hole
energy
Au or Ag −8.1 Holes
Figure 2 | Perovskite cell structure and associated vacuum energy levels. a, General organic–inorganic halide solar cell, which includes two optional layers
that are not essential for high performance; an energy conversion efficiency of over 15% has been reported for devices that have both optional layers, with
only the scaffold layer infiltrated by the perovskite (and then by the HTM) and without scaffolding; the structure then corresponds to a simple planar thin-
film cell. b, Vacuum energy levels (in eV) for corresponding materials (CH3NH3PbI3 perovskite, conducting TiO2 scaffold).
Key recent results include demonstrations of an additional depo- the challenges in collecting photogenerated carriers. Absorption
sition process involving PbI2 deposition from solution with in situ measurements (Fig. 3a) are consistent with calculations show-
conversion to perovskite by a vapour-phase CH3NH3I reaction34. ing direct-bandgap properties for perovskites of interest 14,45. Two
Promising results have also been reported with perovskites based on strong, spin–orbit split, excitonic absorption thresholds are appar-
organic cations other than CH3NH3+. Cations with larger ionic radii, ent, as in direct-bandgap III–V semiconductors46. However, reverse
specifically ethylammonium (CH3CH2NH3+)10 and formamidinium ordering of band-edge states (specifically a p-like conduction
(NH2CH=NH2+)11–13, increase the t factor and push structures towards band)47 results in splitting in the conduction band, rather than in
the symmetrical cubic phase. Different proportions of organic cati- the valence band. Interestingly, reverse band-edge ordering also
ons, inorganic cations (Pb, Sn) and halide anions (I, Br, Cl) can be gives a bandgap that increases with increasing temperature for any
incorporated in mixed perovskites, allowing their properties to be given phase48,49, which is the opposite trend to that of tetrahedrally
fine-tuned; however, the use of mixed halides has dominated. coordinated semiconductors.
Recent work has also demonstrated the use of new HTMs and The strong excitonic absorption edge also means there is no
electron transport media (ETMs). Effective ETMs have been reported basis for the common practice of determining bandgaps using Tauc
in which the standard fluorine-doped tin oxide (FTO)/compact plots50. The absorption edge is determined by a broadened exciton
TiO2 combination is replaced by indium tin oxide as a transparent impulse response, as is the case for direct III–V semiconductors,
conducting oxide combined with a thin (25 nm) ZnO-nanoparticle with the unbroadened response described by Elliott’s theory 46,51.
layer 35; this gave a reported efficiency of 15.7% for planar cells on The relatively high exciton binding energy compared to those of
glass. Low-temperature processing also gave a creditable perfor- III–V semiconductors with a similar bandgap (37–50 meV has been
mance on flexible polyethylene terephthalate. Inorganic HTMs such reported for iodide in the low-temperature phase52 and 35–75 meV
as CuI36 and CuSCN37 also give reasonable results, as do organic- for the mixed chloride at room temperature49), not only lowers the
photovoltaic-derived organics for both ETMs and HTMs, specifi- absorption threshold, but also increases the strength of the above-
cally a (6,6)-phenyl C61-butyric acid methyl ester ETM combined bandgap absorption that generates unbound electron–hole pairs.
with a poly(2,3-dihydrothieno-1,4-dioxin)-poly(styrenesulphonate) Correspondingly, above-bandgap absorption is comparable to or
HTM38–40 (efficiencies up to 12%, refs 38,39). stronger than that in many direct-bandgap III–V semiconductors,
Such diversification increases potential applications. Flexible such as GaAs, although it is lower than that of some inorganic chal-
cells5 require low processing temperatures of less than 150 °C, rather cogenides (see Fig. 3a).
than 500 °C, which is typical for compact TiO2. Graphene nano- For devices with a continuous perovskite layer or an insulating
flakes in the normally compact TiO2 layer allow such processing 41. scaffold (Fig. 2a), transport across the perovskite is important for
An efficiency of 15.6% has been reported for a structure that includes device operation. The exciton binding energy is still sufficiently low
all optional layers and Al2O3 scaffolding (0.6% graphene/TiO2 by for photogeneration of both excitons and unbound electron–hole
weight)41, and an efficiency of 15.9% has been achieved by com- pairs, with thermal dissociation of excitons into free carriers (and
bining small TiO2 nanoparticles with a titanium diisopropoxide reassociation into excitons) expected. The respective photocurrent
bis(acetylacetonate) binder 42. In other work, the non-uniformity contributions ideally depend on the coupling between these popula-
produced by solution deposition is exploited to produce neutral- tions, as characterized by the intrinsic exciton dissociation time —
colour semi-transparent cells43. Instead of continuous perovskite, the time that would prevail if excitons were not mobile or could not
small invisible islands are grown, which should be less expensive be ionized at heterojunctions.
than laser approaches for fabricating semi-transparent amorphous- At the low-coupling extreme (dissociation times large compared
silicon cells44. to recombination times), the two populations will be independent
and relative concentrations will be determined by the spectral com-
Enabling attributes position of illumination. For sunlight, free electron–hole pair gen-
Strong optical absorption is the key to the outstanding performance eration is expected to be completely dominant, as primary exciton
of these perovskite cells, reducing both the required thickness and generation will occur only at wavelengths near band edges (at least
105 InP
Relative permittivity
50
CIGS
104 40
CIS
103 CdTe 30
102 CdS 20
CH3NH3Pbl3
101 10
CH3NH3Pbl3−xClx Imaginary
0 0
200 400 600 800 1,000 1,200 1,400 5 6 7 8 9 10 11 12 13 14 15
Wavelength (nm) Log (frequency, Hz)
Figure 3 | Absorption coefficients and relative permittivity. a, Absorption coefficient of CH3NH3PbI3 (ref. 40) (these values are about 50% higher
those obtained by Xing et al.74) and CH3NH3PBI3−xClx (ref. 96) compared to other solar cell materials (various sources). b, Real and imaginary parts of the
CH3NH3PbI3 dielectric constant at 300 K as a function of frequency. Dipolar and ionic components successively disappear as the frequency increases.
Low-frequency values (circles) are from ref. 64, mid-frequency (90 GHz) values (squares) are from ref. 65 and optical-frequency value (triangle) is
from ref. 66. The solid lines are Debye relaxation fits to mid-range values (ref. 65). The dashed lines show possible transitional values in the far infrared
(qualitative only).
for continuous perovskite layers). To contribute to the photocurrent In conventional polycrystalline semiconductors, low non-
in this low-coupling limit, excitons would need to dissociate at an radiative recombination generally requires large grain sizes, a low
interface (that with the ETM or the HTM or both), as in organic grain-boundary activity and a low density of intragranular defects.
photovoltaic devices. The present perovskites give narrow X-ray diffraction peaks that
At the opposite extreme of high coupling (rapid dissociation), are consistent with both near micrometre grain sizes and a reason-
free electron–hole and exciton concentrations will equilibrate with ably low intergranular defect density 58. Large grain sizes might be
relative concentrations theoretically determined by a mass action expected from the high ‘effective’ homologous temperatures corre-
law rather than by the illumination (this law will take the form sponding to relatively low processing temperatures due to the low
deduced by Combescot 53,54, when corrected to account for excited imputed perovskite melting points (most decompose before melt-
exciton states and scattering states associated with electron–hole ing 29). For metals, a 0.2 increase in the homologous temperature
attraction55). Excitons are expected to flow in the same direction as increases the grain size by a factor of ten59. To reduce detrimental
electrons or holes (Fig. 2b), depending on which has the larger elec- grain boundary activity, specific processing steps are required for
trochemical gradient 56. Because electron and hole currents increase polycrystalline Si, CdTe and copper–indium–gallium diselenide
in opposite directions (Fig. 2b), these gradients will probably be (CIGS) cells. These perovskites do not seem to require such process-
equal at one point in the device, where the total flux is free carriers. ing steps60. This may be because of their high effective homologous
No additional exciton dissociation features will be required in this temperatures, the tendency for the inorganic planes to align parallel
case. Transport could then be treated as free carrier with transport to substrates61 (which reduces one degree of misorientations), crys-
and recombination parameters determined by a weighted combi- tallographic flexibility (allowing more graceful accommodation of
nation of free carrier and excitonic values56. For a relatively narrow misorientations than in more rigid materials), or a combination of
coupling range lying between the low and high extremes, interme- all these factors.
diate behaviour is expected54. As well as strengthening absorption, A recent study 62 investigated the activity of intrinsic intergranular
excitons provide an additional pool of carriers that may assist carrier defects in CH3NH3PbI3 using density functional theory. Two types of
transport. Benefits are expected if the lifetime of exciton diffusion or intrinsic defects were studied: neutral Schottky defects (equal num-
recombination exceeds that of free carriers54. However, despite rela- ber of positive and negative vacancies) and Frenkel defects (equal
tively high exciton binding energies and uncertainties arising from number of vacancies and interstitials of the same ion). Schottky
attempting to separate correlated electron–hole pairs into bound and defects, such as PbI2 and CH3NH3I vacancies, were found not to
scattering states55, it seems as if exciton concentrations will be low generate defect states having energies within the perovskite band-
compared to both free-carrier concentrations except in extreme cir- gap; this was attributed to the ionic bonding of the perovskite. This
cumstances, a conclusion supported by a recently published work49. implies Schottky defects are unlikely to be effective as non-radiative
Another striking attribute of these perovskites is their low non- recombination centres. Elemental defects, such as Pb, I and CH3NH3
radiative recombination rates compared to other thin-film poly- vacancies associated with Frenkel defects, were found to form shal-
crystalline semiconductors. This property manifests itself in the low levels near band edges, again reducing the effectiveness as
relatively small difference between Voc of experimental cells and non-radiative recombination centres. However, these conclusions
their effective bandgap potential3 (Eg/q) or alternatively by their may need to be moderated by the well-known limitations of den-
high external radiative efficiency, a parameter deducible from Voc sity functional theory studies14,45. Another recent study 63 that used a
and the spectral response57. The best perovskite cells have relatively similar approach reached similar conclusions. Dominant defects in
low values for the difference Eg/q − Voc (about 450 meV, ref. 3) and the iodide are deduced to be p-type Pb vacancies and n-type meth-
relatively high calculated external radiative efficiencies (0.058%; ylammonium interstitials, with growth conditions determining the
I. Al Mansouri, personal communication). This makes perovskites final doping polarity. This study also considered anti-site substitu-
particularly interesting for high Eg cells in tandem cell stacks3, where tions, which, along with Pb interstitials, formed states near the mid-
the high Voc values of such cells give rise to substantial efficiency dle of the bandgap. However, formation energies were high, meaning
advantages. The diverse range of fabrication methods that have been that these defects are expected to form in low concentrations at the
used, which include low-temperature approaches, further enhances relatively low material deposition temperature involved. Although
the prospects of these perovskites as a tandem cell component. experimental confirmation is required, these results may help explain
a b Electron beam c
A χP
Perovskite
χP
TiO2
2 ФHTM
χT χF
7 1 4 6
5 hf
Au Perovskite FTO
HTM TiO2 Glass HTM FTO
EBIC signal
Perovskite
3
HTM
Perovskite 1 µm
Figure 4 | Electron-transfer processes in nanoparticle and bulk cells together with a bulk energy-band diagram. a, Schematic of electron-transfer
processes in a perovskite nanoparticle or ETA device71. The thick green and thin red arrows respectively indicate the processes desirable for energy
conversion and those associated with losses. hf, photon energy. b, Schematic of EBIC experiment 72. A scanned electron beam generates a cloud of
carriers, creating a position-dependent current in a short-circuiting load. c, Energy-band diagram deduced from the vacuum energy levels shown in Fig. 2b.
χP, χT and χF represent the electron affinities of the perovskite, TiO2 and FTO layers, respectively, and ΦHTM represents the work function of the HTM layer.
(The barrier at the TiO2/FTO interface has been a source of discussion in the dye-cell literature97,98, because it impedes carrier collection in the direction
shown. Some work suggests states generated at the interface under white-light illumination cause a pinning effect, reducing its significance98. Other work
suggests a failure of the Anderson rule, which is not uncommon, and that the barrier has the opposite direction at thermal equilibrium99.)
why high-performance perovskite cells can be produced by a diverse points on the surfaces of titania nanoparticles26 to conventional pla-
range of deposition approaches and a wide variety of cell structures. nar thin-film operation32.
Although the dielectric properties of perovskites have not yet In the dye-sensitized and ETA configurations, perovskites are
been shown to be important for good performance, they are so not required to have good carrier transport, as the interfacial prop-
extreme compared to those of conventional semiconductors that erties mainly determine the performance (Fig. 4a depicts the key
it is a possibility. Figure 3b shows a composite of reported data processes involved71). The desirable processes involve photoex-
obtained at 300 K for CH3NH3PbI3. At low frequencies, the die- citation in perovskite (1), electron transfer to titania (2) and hole
lectric constant is large; it has been reported to be 60.9 over the transfer to the HTM (3) (or, equivalently, electron transfer from the
20 Hz – 1 MHz range64, which is consistent with the low-frequency HTM to the perovskite). Undesirable processes are recombination
value of 60.2 obtained from fits at higher frequencies65. This low- of photogenerated species (4), back charge transfer at the interfaces
frequency value is appropriate in determining steady-state prop- of TiO2 and the HTM with the perovskite (5,6) and between TiO2
erties, with static fields consequently varying slowly with position and the HTM (7) (this may occur if perovskite is absent in some
— five times slower than in silicon under similar electrostatic dis- areas — for example, when nanoparticles or voids are present). For
turbances. The high value of the dielectric constant results from a high performance, processes (4)–(7) must operate on much slower
combination of dipolar, ionic and electronic contributions. As the timescales than charge generation and extraction (1)–(3). Time-
excitation frequency increases, the permanent dipole associated resolved transient techniques have proved useful for studying the
with the organic cation can no longer respond and a new plateau is related kinetics72.
reached with a dielectric constant of 29.7 (ref. 65). Finally, at infra- For thicker perovskite layers, photogenerated carrier transport
red frequencies, the ionic component drops out, leaving only the becomes important. Initial data suggested that the mixed halide
electronic response. The dielectric constant drops to 6.5 at optical CH3NH3PbI3−xClx has a distinct advantage over the pure iodide —
frequencies66, lower than that of inorganic semiconductors with a its carrier diffusion length (>1 μm; ref. 73) is about ten times greater
similar bandgap. The dashed lines in Fig. 3b indicate possible tran- than that (about 100 nm) of the pure iodide73,74. Subsequent work
sition values of the constants. with the iodide prepared by the vapour-phase CH3NH3I conversion
The other possibility is that CH3NH3PbI3 may possess paraelec- of PbI2 resulted in good carrier collection in 350-nm-thick iodide
tric or even ferroelectric properties at room temperature and above, films, suggesting that the diffusion lengths exceed this thickness34;
impacting device performance67,68. The main evidences for this is this is supported by recent electron-beam measurements60.
the hysteresis observed in resistivity measurements, which is not Recently, collection across a film cross-section has been directly
seen in related compounds prepared similarly, and the remnant probed using the electron beam induced current (EBIC) technique
polarization, which is apparent as a finite voltage output at zero cur- (Fig. 4b)60,72. In this technique, the electron beam generates a cloud
rent 29. The crystallographic point group (4mm) of both the room- of excited carriers (these are expected to be largely free electron–hole
temperature and high-temperature iodide phases is consistent with pairs, some of which may combine to form excitons if they have short
ferroelectric behaviour 29. Hysteresis has also been reported for association times). Because the resolution is limited by the size of this
high-efficiency devices69,70, although many other explanations are cloud, approximately 1,500-nm-thick perovskite layers were grown,
also possible for this besides ferroelectric effects69. Best performance which are appreciably thicker than those in optimal planar cells.
has been realized when hysteresis is minimized (S. I. Seok, personal The typical line scan shown in Fig. 4b has two interesting features.
communication). Carrier collection along the boundaries of micro- The first is the double peak in the EBIC signal within the perovs-
scopic ferroelectric domains has also been suggested68, although kite. This indicates the cell’s ambipolar response with carriers gener-
this presently lacks experimental support. ated both near the spiro-MeOTAD HTM and near the ETM (TiO2
plus FTO) being collected. The second interesting feature is the dip
Device operation between the two peaks, which arises because an appreciable fraction
High efficiencies have been realized using a diverse range of oper- of the generated carriers recombine on their way to their preferred
ating modes, spanning from generation and collection at sparse contact, due to the sample thickness exceeding the optimal thickness.