1
In situ observation of heat-induced degradation of
perovskite solar cells
G. Divitini*
1
, S. Cacovich
1
, F. Matteocci
2
, L. Cinà
2
, A. Di Carlo
2
, C. Ducati
1
1
Department of Materials Science & Metallurgy, University of Cambridge, 27 Charles Babbage road,
CB3 0FS Cambridge (UK)
2
C.H.O.S.E. (Centre for Hybrid and Organic Solar Energy), Department of Electronic Engineering,
University of Rome “Tor Vergata”, via del Politecnico 1, Rome 00133, Italy
* Corresponding email: gd322@cam.ac.uk
Abstract
The lack of thermal stability of perovskite solar cells is hindering the progress of this technology
towards adoption in the consumer market. Different pathways of thermal degradation are activated
at different temperatures in these complex nanostructured hybrid composites. Thus, it is essential to
explore the thermal response of the mesosuperstructrured composite device in order to engineer
materials and operating protocols. Here we produce devices according to four well established
recipes, and characterise their photovoltaic performance as they are heated within the operational
range. The devices are analysed using transmission electron microscopy as they are further heated in
situ, to monitor changes in morphology and chemical composition. We identify mechanisms for
structural and chemical changes, such as iodine and lead migration, which appear to be correlated to
the synthesis conditions. In particular, we identify a correlation between exposure of the perovskite
layer to air during processing, and elemental diffusion during thermal treatment.
Introduction
Solar cells based on lead halide perovskite composites have become increasingly popular in the last
few years due to a combination of low synthesis cost and high power conversion efficiency, with
certified values in excess of 20%
1-5
. The stability of such devices is however a concern – it is well
known that heating at or above around 85°C, a temperature close to those reached during normal
operation in full sunlight, performance degrades rapidly, and such instability is exacerbated by
exposure to moisture
6-8
; systematic thermal and ageing studies are required to understand such
degradation process
9
. Changes happen in both the organic and inorganic components of the cells;
the resilience of the perovskite layer, in particular, is expected to become a limiting factor once
different hole conducting materials (or hole-conductor-free cells) are developed
10
. In order to
overcome this limitation, it is vital to understand the degradation pathways of the structures
involved, which here are observed at nanometer-scale spatial resolution in situ, inside an analytical
scanning transmission electron microscope (STEM), while the composition is monitored with
elemental mapping through energy-dispersive X-ray analysis (EDX). The analysis of such devices is
challenging due to several factors. The spatial dimensions relevant to the fabrication and the
operation of the cells are in the 1 – 100 nm range, and the materials are easily damaged by exposure
to an electron beam in a TEM, requiring to carefully tune the electron dose. The system also includes
organic and inorganic components in an assembly with complex chemistry and morphology. Finally,
the rapid changes to the devices in air and the low degradation temperatures pose an additional
challenge to the experiment, which needs to be timed appropriately and carefully executed.
2
The monitoring of this process is made possible by combining several recent advances in TEM
technology. The use of high-brightness electron guns and detectors with large collection areas allows
the fast acquisition of high quality EDX maps with limited electron dose on the sample
11
; the signal-
to-noise ratio of the maps can be further increased by applying denoising algorithms (PCA, Principal
Components Analysis) within an open-source software suite
12-13
. The development of novel in situ
heating holders for TEM, based on micro-heaters and featuring high-stability and fast-response, was
also crucial – in particular, the holder used here allows very precise control (sub-degree) at values
just above room temperature, as well as providing fast heating and cooling (a few seconds for the
temperatures in use in this paper). The good spatial stability of the holder is crucial in acquiring EDX
maps.
Four samples, prepared with different processing routes, have been examined under TEM in order to
correlate synthesis process, morphology, photovoltaic properties and behaviour under heating. For
three samples a double-step fabrication
14
was employed – here labelled vacuum conversion (sample
A), glovebox (B) and air conversion (C, with relative humidity ~50%) – while one was manufactured
with a single-step approach (D, in glovebox). Specifically, sample B is a representative of established
synthesis inside a glove-box, and sample C is interesting due to the processing in humid air
15
. Each
synthesis process was optimised separately, and full details are reported in the methods section.
Cross-sectional specimens were extracted as lamellae for TEM analysis using focused ion beam (FIB)
milling. The procedure takes a few hours and is carried out in vacuum, hence minimal changes are
expected to happen in the samples. Devices were unsealed just before FIB processing, limiting
exposure to air to a few minutes. TEM characterisation and in situ heating were performed just after
the FIB step, ensuring that specimen were exposed to air for a maximum of 5 minutes before being
transferred into vacuum again.
A representative STEM high-angle annular dark field (HAADF, in this case taken from sample B)
cross-sectional view of a device is shown in figure 1. The cell is manufactured on FTO-coated glass
(Fluorine-doped Tin Oxide), further coated with a compact TiO
2
hole blocking layer. A mesoporous
TiO
2
layer is infiltrated and capped with methyl-ammonium lead iodide (MAPbI
3
). A spiro-OMeTAD
(2,2',7,7'-Tetrakis-(N,N-di-4-methoxyphenylamino)-9,9'-spirobifluorene) hole transporting layer and
a gold electrode complete the cell; platinum is deposited on top during FIB sample preparation as a
standard procedure in order to protect the sample during milling. EDX elemental maps are shown in
figure 1, showing well defined layers and good infiltration of lead and iodine inside the titania
scaffold.
The samples present different morphologies. The images reported in figure 2 are Z-contrast images,
where the brightness of a given pixel is proportional to the average atomic number and thickness (in
this case constant, around 150 nm, for all samples, in the region of interest). The brightest areas are
the Au electrodes, while the organic hole transporter appears black. The titania scaffold appears as a
dark feature overlapping with the brighter perovskite layer; darker areas in this region (voids) result
from incomplete infiltration of the perovskite inside the scaffold. Samples B and C present good
perovskite infiltration inside the titania, as well as a homogeneous perovskite capping layer (around
120 and 200 nm thick respectively). Sample A has a thin capping layer, and incomplete infiltration in
the scaffold can be observed as dark regions near the TiO
2
compact film. Sample D, grown with a
single step deposition, presents a very inhomogeneous perovskite distribution, leading to micron-
sized gaps and an infiltration of hole transporter material (HTM) inside the titania scaffold, as well as
the formation of voids between the HTM and the Au electrode
The perovskite layers grow differently according to the synthesis conditions – samples A and B
present fairly homogeneous layers; sample C has larger, faceted crystals. The perovskite region in
sample C also appears to be composed of two separate layers: the area directly above the titania is
3
brighter, suggesting a higher average atomic number, whereas the top layer is darker and more
clearly faceted, probably due to a higher degree of crystallinity.
In all samples, voids are visible in the spiro-OMeTAD layer across the entire depth (see
Supplementary Figure 1). With a size up to 50 nm, similar features have been previously observed by
AFM (Atomic Force Microscopy, seen in top view) and SEM (Scanning Electron Microscopy, in cross-
section) and are suspected to play a major role in the infiltration of moisture inside the cell
16
. These
voids are introduced during fabrication and are not influenced by the thermal treatments.
Photovoltaic characterisation
A characterisation of the photovoltaic properties was carried out at room temperature and upon
heating. Twin samples (4 per type), nominally identical to the ones used for TEM analysis, were
heated ex-situ up to 90°C. This is representative of the temperatures the cell would operate at under
full illumination; a reversible behaviour of the parameters is observed for heating within this
temperature range (sample B shown in Figure 3, see Supplementary Figure 2 for the other samples)
– upon cooling, the original performance is recovered in all cases except for sample D, where the
inhomogeneity of the active layer can lead to connectivity issues. In the comparison between all
sample types, reported in Figure 4, samples B and C show relatively stable power conversion
efficiency (PCE, change up to 10%). Sample A exhibits a strong hysteresis. It is known that
mesoporous TiO
2
significantly reduces hysteresis compared to planar cells
17
; in the case of sample A
we attribute such effect to the lack of a continuous perovskite capping layer, which can result in
poor connectivity of the domains infiltrating the mesoporous volume. In all samples an overall decay
of the PCE is observed for increasing temperature. This is related to a decrease in V
oc
(open-circuit
voltage – decrease of about 20% in all samples), partially compensated, in samples B and C, by an
increase in fill factor. The change in V
oc
is consistent with results in the literature and ascribed to
increased recombination
18
. The V
oc
decrease in sample C is slightly slower than in B or D; we
hypothesise that this could be caused by the presence of iodine dopants/vacancies at the interfaces
(mainly between perovskite and HTM) affecting the interfacial energy mismatch
19
. J
sc
(short-circuit
current) is moderately reduced with heating up to around 50°C in B and C; since previous work in the
literature showed a constant J
sc
for hole transporter-free perovskite cells
20
, we attribute the
observed reduction to a change in the transport properties of spiro-OMeTAD. At a temperature
corresponding to the perovskite phase transition (~60°C) the short circuit current starts moderately
increasing
21
. This effect is stronger in sample D, where the film is less constrained due to the
presence of voids in the structure. The observed fill factor increases for increasing temperature, due
to a decrease in the series resistance (see Supplementary Figures 3 and 4). This is due to the
resistivity decrease of both perovskite
21
and spiro-OMeTAD
22
. As the temperature is brought down,
the devices (A, B and C) recover the initial performance, showing the reversibility of the phase
transformation.
Above 90°C the HTM, here doped spiro-OMeTAD
23
, is known to degrade non-reversibly due to the
evaporation of the dopants (Li-TFSI - lithium bis(trifluoromethanesulfonyl)imide - and TBP - tert-
butylpyridine). TBP could evaporate after heating stress at lower temperature (~85°C) with respect
to the TBP boiling point (197°C), as already shown for solid-state dye solar cells by Bailie et al.
24
TEM in situ heating
Cross-sectional views of all samples, at different stages of heating, are shown in figure 2. Videos of
the STEM images acquired during the heating ramp for each sample are available as Supplementary
Videos 1 (sample A) to 4 (sample D). To enhance speed and reproducibility of in situ heating cycles, a
reference temperature of 50°C was chosen. Photovoltaic measurements show that devices are
stable at this temperature and the initial room temperature performance can be restored after
4
cooling. The thermal decomposition was “frozen” after each heating step by bringing the
temperature down to 50°C during the acquisition of the EDX maps, which took about 20 minutes per
map. The heating profile, reported in Supplementary Figure 5, was chosen in order to explore the
full degradation process, and includes steps at 100, 150, 175, 200, 225, 250°C. Initial steps are
prolonged for 30 minutes, while the later ones are shortened to 15 minutes due to the faster sample
dynamics.
No change in the perovskite layer is visible in the HAADF images for heating up to 150°C, even if the
light conversion properties of the devices appear to degrade irreversibly above 90°C. This suggests
that the main mechanism for performance degradation at low temperatures is related to the HTM,
and in particular to how the HTM and the additives react upon heating. The low concentration and
low atomic number of the additives are beyond the detection and mapping capabilities of STEM-
EDX.
The first noticeable change in the perovskite layer occurs on device C after heating at 150°C. Due
probably to heat-assisted electron beam damage, holes start forming in the region of the perovskite
layer close to the interface with the titania scaffold. The methylammonium lead iodide layer,
synthesised in air in this case, presents two different levels of contrast in STEM-HAADF, with a
brighter region where combined electron beam / thermal degradation happens earlier than in the
darker region. This particular behaviour, not observed in the other devices, can be caused by the
warm air flow used during the drying step of the PbI
2
layer that induces the incorporation of a low
concentration of oxygen and moisture; elemental analysis of the two perovskite regions with
different electron density shows comparable stoichiometry within the EDX accuracy and can not
account for the change in HAADF intensity, therefore suggesting a difference in crystallinity.
As the temperature increases, the perovskite layer starts degrading and small particles, which EDX
suggests to be PbI
2
, form at the interface with the FTO layer and start migrating and coalescing in
specimens A, B and D. These features are visible in Figure 2 as bright spots. This is a phenomenon
due to the intrinsic 2D nature of the specimen (a lamella about 150 nm thick), where particles can
move on the surface of the cross-section. Lead and iodine originate from the perovskite layer; the
preferential formation of PbI
2
aggregates over the FTO layer can be explained by the presence of a
rougher surface – the FTO layer, being below a mesoporous layer, presents fine roughness due to
FIB milling and the “curtaining” effect
25
. Such high mobility of atomic species on heating of MAPbI
3
has recently been observed in samples heated ex situ at 85°C for 24h while exposed to different
gases
26
. The later onset of this phenomenon observed here could be attributed to the surrounding
environment (heating here is conducted in vacuum) and to the timescale of the process (the heating
steps in this work are 30’ each). Interestingly, the aggregation of such particles is not visible in
sample C. This is ascribed to the different dynamics of the heating process, leading to elemental
diffusion of Pb and I rather than coalescence.
The temperature ramp is continued on all samples up to 250°C, where the perovskite component of
the cell disappears within a few minutes, while the overall structure is kept in place by the FTO and
metal electrodes. For reference, a sample prepared from the same device after 2 months of
exposure to air (in the dark) was compared to the different stages of the heating ramp, finding a
comparable degradation to the step at 200°C (Supplementary Figure 6).
The diffusion of heavy elements inside the spiro-OMeTAD can be seen in the HAADF signal as sample
C is heated (Figure 5). EDX mapping identifies the first migrating species to be iodine. This behaviour
has been suggested in the literature
19-27
, although it had never been directly observed in real time in
a TEM. On the pristine sample the presence of iodine extends for ~100 nm beyond the limit of the
perovskite layer under a concentration gradient; the iodine front advances towards the Au electrode
under heating at increasing diffusion speed, reaching a depth of 50 to 80 nm after the last thermal
step. Elemental migration of lead in the same direction is also observed, although it is triggered at
5
higher temperature (starting from ~175°C). Since partial elemental diffusion is observed for iodine
before heating, we suggest that this process might be happening as part of the ageing process, and
could be responsible for the long term loss of performance at room temperature for devices
fabricated without moisture control, outside a dry-box.
Analysis of the STEM-HAADF images can provide a quantitative estimate of the degradation of the
cell structure with time. As the sample thickness is constant and the TiO
2
layer doesn’t degrade at
such temperatures, the signal decays as the inorganic components of the perovskite diffuse or
evaporate. Figure 6 reports the profile of the HAADF signal for samples B and C, averaged over the
entire film, divided into the infiltrated and the capping regions. As seen in Figure 2, very little
changes until 150°C, after which a steady decay is observed. The loss in HAADF signal upon heating is
slower in the mesoporous regions compared to the capping ones, suggesting a higher stability. This
resilience towards heating combines with robustness upon light irradiation and oxygen exposure
that has been demonstrated in the literature
28
. Surprisingly, the signal decay in sample C is slower
than in sample B, even though the former was exposed to air and moisture during the synthesis. A
possible explanation is that the incorporation of low concentrations of oxygen and water in the
perovskite film, while determining local degradation, also induce pinning sites and prevent the
diffusion of elemental Pb and PbI
2
towards the FTO electrode which was observed in other samples.
Conclusions
In this work we reported the in situ TEM observation of the thermal response of four perovskite-
based solar cells heated to and above operational temperatures until full degradation of the
composite device. The fabrication routes strongly affect the perovskite coverage and scaffold
infiltration, as well as the degradation mechanisms observed. In all samples the perovskite layer is
shown to be stable for short times until 150°C. Exposure to air during fabrication results in an
inhomogeneous perovskite layer, which presents a different degradation pathway compared to
devices processed in a glovebox; the device processed in air also shows significant migration of
iodine into the hole transporting layer. These results, obtained with imaging and chemical analysis at
nanometre scale, provide new insight on the effective morphologies of perovskite-based solar cells
and on the correlations between morphology, chemical composition and thermal stability of the
meso-superstructured functional composites.
Methods
Devices details
Common to all the samples
A raster scanning laser (Nd:YVO
4
pulsed at 30 kHz average output power P=10 W) was used to etch
the FTO/glass substrates (Pilkington, 8 Ωcm
-1
, 25 mm x 25 mm). The patterned substrates were
cleaned in an ultrasonic bath, using detergent with de-ionized water, acetone and isopropanol (10
minutes for each step). A 80nm-thick patterned blocking TiO
2
layer (BL-TiO
2
) layer was deposited
onto the patterned FTO using spray pyrolysis in according with a previously reported procedure
29
. A
250 nm-thick mesoporous TiO
2
layer (18NR-T paste, Dyesol) diluted with terpineol and ethylcellulose
was screen-printed over the BL-TiO
2
surface and sintered at 480°C for 30 min. The final thickness was
measured by profilometer (Dektak Veeco 150). Different approaches for the perovskite layer
deposition were used for each sample, as discussed below.
The hole-transport layers was deposited by spin-coating a 75 mg ml
-1
solution of 2,20,7,70-tetrakis-
(N,N-dip-methoxyphenylamine)9,9’-spirobifluorene (spiro-OMeTAD) doped with 8 μl of tert-
butylpyridine and 12 μl of lithium bis(trifluoromethanesulfonyl)imide (Li-TFSI) solution (520 mg in 1
