A Review: Thermal Stability of Methylammonium Lead Halide Based Perovskite Solar Cells

: Perovskite solar cells have achieved photo-conversion efﬁciencies greater than 20%, making them a promising candidate as an emerging solar cell technology. While perovskite solar cells are expected to eventually compete with existing silicon-based solar cells on the market, their long-term stability has become a major bottleneck. In particular, perovskite ﬁlms are found to be very sensitive to external factors such as air, UV light, light soaking, thermal stress and others. Among these stressors, light, oxygen and moisture-induced degradation can be slowed by integrating barrier or interface layers within the device architecture. However, the most representative perovskite absorber material, CH 3 NH 3 PbI 3 (MAPbI 3 ), appears to be thermally unstable even in an inert environment. This poses a substantial challenge for solar cell applications because device temperatures can be over 45 ◦ C higher than ambient temperatures when operating under direct sunlight. Herein, recent advances in resolving thermal stability problems are highlighted through literature review. Moreover, the most recent and promising strategies for overcoming thermal degradation are also summarized.


Introduction
Increasing global demand for energy and continued reductions in fossil fuel-based energy sources call for the greater use of alternative renewable energy sources. Among the possible renewable sources of energy, solar energy is the most promising, as it can be converted into electrical energy and reach almost every part of the world. Commercially available solar cells are based on silicon, GaAs (gallium arsenide), CdTe (cadmium telluride) and CIGS (copper indium gallium (di)selenide) and have sufficiently established efficiencies to meet global energy demand. However, efforts have been made to reduce the cost [1] of these existing technologies by replacing them with cheaper alternative photovoltaic devices such as perovskite solar cells (PSCs). Currently, the photo-conversion efficiency (PCE) of PSCs has soared beyond 20% in less than five years of laboratory research [2].
Miyasaka, et al. first introduced organic-inorganic hybrid perovskite absorber layer in dye sensitized solar cells (DSSCs) configuration with an efficiency of only 3.81% [3]. The poor stability and efficiency of liquid electrolyte-based structures were attributed to iodine-based redox processes. The efficiency increased to 10% when the liquid electrolyte was replaced with a solid-state hole transport layer of spiro-OMeTAD and led to stability improvements as well [4]. Currently, for PSCs, a planar heterojunction structure (introduced by Snaith, et al. [4]) is widely used, and consists of a solid perovskite layer with electron and hole selective contacts. The planar heterojunction structure can be fabricated at a low temperature (<150 • C), while the mesoporous TiO 2 scaffold structure requires high temperature (<400 • C), making the former structure superior to the latter structure in terms of cost [5].

Structural Stability of Perovskite
The structural stability of the perovskites is critical when used in photovoltaic devices, since they are expected to remain stable for more than 25 years [22]. Structural stability is defined as the ability for a crystalline phase to be stable over a wide range of external factors such as heat, pressure, moisture, etc. Perovskite compounds have a general chemical formula of ABX 3 , where A and B are cations and X is an anion. Their crystal structure is similar to that of calcium titanium oxide. Usually, the A cations are larger than the B cations. Oxide perovskites have been studied extensively because of their multifunctional nature [23]. However, owing to their wide bandgap, oxide perovskites harvest only 8-20% of the solar spectrum, limiting their use in photovoltaic applications [23]. Instead, halide organic-inorganic perovskites were developed by replacing the oxygen anion of oxide perovskites with an inorganic halide (I − , Cl − , Br − ) [23]. An organic or inorganic monovalent A + cation (e.g., Rb + , Cs + , CH 3 NH 3 + , HC(NH 2 ) 2 + ) and a divalent B 2+ metal cation (e.g., Pb 2+ , Sn 2+ ) were also implemented in the frame of the perovskite structures [23]. The halide perovskites obtain a desired crystal symmetry by maintaining an allowable tolerance factor. A tolerance factor developed by Goldschmidt [24] determines the radii sizes associated with cubic symmetry, described by where R A , R B , R X are the ionic radii of A, B, X, respectively. The tolerance factor provides a rough estimate of the stability and distortion of crystal structures of a compound. In addition, it gives an idea of whether the phase is cubic (t = 1) or deviates into the tetragonal or orthorhombic phase [25]. In general, an established tolerance factor value for halide perovskites lies in the range of 0.85 < t < 1.11 [26]. Non-perovskite structures are formed when the tolerance factor is higher or lower.
In an inorganic-organic hybrid perovskite, it is difficult to calculate the absolute tolerance factor as the Appl. Sci. 2019, 9,188 3 of 25 organic cation has a non-spherical geometry [27]. However, it is possible to qualitatively analyze the transition of structure in these materials. For example, formamidinium lead iodide, HC(NH 2 ) 2 PbI 3 (FAPbI 3 ), has a larger A cation than methylammonium lead iodide, CH 3 NH 3 PbI 3 (MAPbI 3 ) and a larger cation would generally represent a higher tolerance factor [28]. Two phases can be obtained in solution-processed FAPbI 3 materials. One is the photoactive α-phase (black phase) and the other is a non-photoactive phase (yellow phase or δ-phase) [29]. Figure 1a,b shows the ideal case of cubic symmetry for perovskites. The cubic symmetry corresponds to a Pm3m space group with 12-fold coordination for the A cation, 6-fold coordination for the B cations, and BX 6 octahedra residing in the corners [23]. The deviation from cubic symmetry can be attributed to several factors including the atomic sizes of the constituents. The A cation does not directly affect the electronic properties [30][31][32]; however, the size of A cations can cause distortion of the B-X bonds, which undesirably affects the symmetry. The best electronic properties are obtained with cubic symmetry due to high ionic bonding. An octahedral tilting occurs when the tolerance factor exceeds the ideal range, and affects the electronic properties. In general, an established tolerance factor value for halide perovskites lies in the range of 0.85 < t < 1.11 [26]. Non-perovskite structures are formed when the tolerance factor is higher or lower. In an inorganic-organic hybrid perovskite, it is difficult to calculate the absolute tolerance factor as the organic cation has a non-spherical geometry [27]. However, it is possible to qualitatively analyze the transition of structure in these materials. For example, formamidinium lead iodide, HC(NH2)2PbI3 (FAPbI3), has a larger A cation than methylammonium lead iodide, CH3NH3PbI3 (MAPbI3) and a larger cation would generally represent a higher tolerance factor [28]. Two phases can be obtained in solution-processed FAPbI3 materials. One is the photoactive α-phase (black phase) and the other is a non-photoactive phase (yellow phase or δ-phase) [29]. Figure 1a,b shows the ideal case of cubic symmetry for perovskites. The cubic symmetry corresponds to a Pm3m space group with 12-fold coordination for the A cation, 6-fold coordination for the B cations, and BX6 octahedra residing in the corners [23]. The deviation from cubic symmetry can be attributed to several factors including the atomic sizes of the constituents. The A cation does not directly affect the electronic properties [30][31][32]; however, the size of A cations can cause distortion of the B-X bonds, which undesirably affects the symmetry. The best electronic properties are obtained with cubic symmetry due to high ionic bonding. An octahedral tilting occurs when the tolerance factor exceeds the ideal range, and affects the electronic properties. Another illustration of BX6 octahedral network, where B cations are surrounded by X anions. (c) A tilted BX6 octahedral structure due to non-ideal size effects and additional factors. Reprinted from [23], with permission from Elsevier.
In recent years, methylammonium lead trihalide (MAPbX3) has received increasing attention as an absorber material for perovskite solar cells. However, the stability of MAPbX3 has been found to strongly vary depending on temperature, changing phase or crystal orientation. The structural data of various methylammonium lead trihalides (MAPbX3; X = I, Cl, Br) [33,34] are summarized in Table 1. As indicated in Table 1, both MAPbBr3 and MAPbCl3 crystallize in the cubic phase at room temperature, while MAPbI3 needs to be heated to a temperature higher than 323 K to allow the transition from a tetragonal to a cubic phase. Interestingly, many studies on MAPbI3 materials report that the tetragonal phase of MAPbI3 can still exist even after heating at temperature of 373 K [4,[35][36][37]. This suggests that the tetragonal phase is surprisingly stable in the thin films, but also emphasizes an ambiguity about the exact phase transition temperature and the nature of the phase transition between tetragonal and cubic phases. In recent years, methylammonium lead trihalide (MAPbX 3 ) has received increasing attention as an absorber material for perovskite solar cells. However, the stability of MAPbX 3 has been found to strongly vary depending on temperature, changing phase or crystal orientation. The structural data of various methylammonium lead trihalides (MAPbX 3 ; X = I, Cl, Br) [33,34] are summarized in Table 1. As indicated in Table 1, both MAPbBr 3 and MAPbCl 3 crystallize in the cubic phase at room temperature, while MAPbI 3 needs to be heated to a temperature higher than 323 K to allow the transition from a tetragonal to a cubic phase. Interestingly, many studies on MAPbI 3 materials report that the tetragonal phase of MAPbI 3 can still exist even after heating at temperature of 373 K [4,[35][36][37]. This suggests that the tetragonal phase is surprisingly stable in the thin films, but also emphasizes an ambiguity about the exact phase transition temperature and the nature of the phase transition between tetragonal and cubic phases.

Impact of Thermal Stress on Perovskites
Typically, device temperatures can be over 45 • C higher than ambient temperatures when solar cells operate under direct sunlight. According to International Standards (IEC 61646 climatic chamber tests), long-term stability at 85 • C is required to compete with other solar cell technologies [38]. Therefore, the study of the thermal stability of perovskite solar devices has attracted world-wide research attention. Different degradation pathways have been proposed and the topic remains under discussion, but the material MAPbI 3 (CH 3 NH 3 PbI 3 ) clearly loses its excellent light harvesting properties over time because it evolves into PbI 2 after a nominal loss of CH 3 NH 3 I. The degradation from CH 3 NH 3 PbI 3 to PbI 2 is most likely accompanied by a release of gases via simple sublimation or assisted chemical reaction. It is proposed that the first mass loss step during the thermal degradation of CH 3 NH 3 PbI 3 and CH 3 NH 3 I under an inert atmosphere proceeds as [39]: Therefore, the determination of the composition of these released gases during the controlled thermal degradation of CH 3 NH 3 PbI 3 is expected to pinpoint the degradation pathways. Different interpretations can be found in the literature regarding the chemical nature of these released gases during thermal degradation and several studies have analyzed the effect of intrinsic degradation and thermal stress on the perovskites. Herein, an overview of the impact of thermal stress on perovskites, based on different aspects such as chemical, crystallographic, morphological and optical degradation, is described.

Chemical and Structural Degradation
Fan, et al. [40] studied the thermal degradation of MAPbI 3 by fabricating MAPbI 3 microplates. Microplates are highly crystalline and thus provide an outstanding opportunity to study the structural degradation mechanisms of perovskites. The authors investigated the crystal structure under thermal stimulation using in-situ high-resolution transmission electron microscopy (HRTEM). They found that almost 75% of the original perovskite tetragonal phase returned to trigonal PbI 2 after 100 s of heating at 85 • C. Degradation is initiated by breaking the weak Pb-I-Pb bond along the (001) direction [41], after which the PbI 2 relaxes into its energetically favorable trigonal structure ( Figure 2). During this process, CH 3 NH 2 and HI sublime into the gas phase. Their studies conducted in dry and inert gaseous environments revealed that no MAPbI 3 hydrates are generated during the degradation process [42][43][44]. The authors therefore concluded that the phase transition to PbI 2 is not initiated by the intrinsic hygroscopicity of alkylammonium cations [38,44], but instead is completely due to the thermally induced degradation. They suggested that a sequential transition occurs from tetragonal MAPbI 3 to trigonal PbI 2 based on the time-dependent studies, which showed an overlap of the MAPbI 3 and PbI 2 crystal structures in the electron diffraction patterns under thermal stress. In particular, the diffraction intensity of the perovskite tetragonal phase was correlated with a direct increase in the PbI 2 trigonal phase intensity. This correlation indicates that the structure is a stacked architecture, with PbI 2 formed on top of the underlying MAPbI 3 layers. In addition, the thickness of each layer varies with the annealing time, with the thermal transformation from MAPbI 3 to PbI 2 occurring through bulk or surface reactions. To understand the actual reaction process, they conducted density functional theory (DFT) calculations [44]. The corresponding reaction rate calculated for the minimum energy path from MAPbI 3 to PbI 2 through bulk degradation was found to be approximately 1000 times slower than surface-initiated decomposition [44]. Therefore, they concluded that surface degradation is kinetically preferred. The first layer is degraded, and the underlying layer gets exposed. The process is repeated until the entire material degrades, which supports the theory that it is a surface-dominated reaction [44]. showed an overlap of the MAPbI3 and PbI2 crystal structures in the electron diffraction patterns under thermal stress. In particular, the diffraction intensity of the perovskite tetragonal phase was correlated with a direct increase in the PbI2 trigonal phase intensity. This correlation indicates that the structure is a stacked architecture, with PbI2 formed on top of the underlying MAPbI3 layers. In addition, the thickness of each layer varies with the annealing time, with the thermal transformation from MAPbI3 to PbI2 occurring through bulk or surface reactions. To understand the actual reaction process, they conducted density functional theory (DFT) calculations [44]. The corresponding reaction rate calculated for the minimum energy path from MAPbI3 to PbI2 through bulk degradation was found to be approximately 1000 times slower than surface-initiated decomposition [44]. Therefore, they concluded that surface degradation is kinetically preferred. The first layer is degraded, and the underlying layer gets exposed. The process is repeated until the entire material degrades, which supports the theory that it is a surface-dominated reaction [44]. The thermal stability of perovskite solar cells, with a structure of ITO/TiO2/MAPbI3, was studied by Conings and co-workers [38]. In order to separate the perovskite layer degradation from degradation at other interfaces, the hole transfer layer (HTL) and the top electrode were excluded. The samples were heated at 85 °C for 24 h, under four environmental conditions: pristine film (reference), O2 environment, N2 environment and ambient conditions. The X-ray diffraction (XRD) patterns revealed the formation of PbI2 under ambient conditions, as shown in Figure 3a. Smecca, et al. [45] showed that the peak at 2θ = 14.00° (corresponding to a unique (001) plane of MAPbI3) decreased along with an increase of the peak at 2θ = 12.61° (corresponding to a (003) plane of PbI2) (Figure 3b,c). The thermal stability of perovskite solar cells, with a structure of ITO/TiO 2 /MAPbI 3, was studied by Conings and co-workers [38]. In order to separate the perovskite layer degradation from degradation at other interfaces, the hole transfer layer (HTL) and the top electrode were excluded. The samples were heated at 85 • C for 24 h, under four environmental conditions: pristine film (reference), O 2 environment, N 2 environment and ambient conditions. The X-ray diffraction (XRD) patterns revealed the formation of PbI 2 under ambient conditions, as shown in Figure 3a. Smecca, et al. [45] showed that the peak at 2θ = 14.00 • (corresponding to a unique (001) plane of MAPbI 3 ) decreased along with an increase of the peak at 2θ = 12.61 • (corresponding to a (003) plane of PbI 2 ) (Figure 3b,c).  Reprinted with permission from [45]. (d) The atomic ratio of I/Pb (red circle) and N/Pb (black diamond) at elevated temperatures analyzed from photoelectron spectroscopy measurement.
Reprinted with permission from [46]. Copyright (2015) American Chemical Society. Reprinted with permission from [45]. (d) The atomic ratio of I/Pb (red circle) and N/Pb (black diamond) at elevated temperatures analyzed from photoelectron spectroscopy measurement. Reprinted with permission from [46]. Copyright (2015) American Chemical Society.
All peaks related to the perovskite layer disappeared in the diffraction pattern during thermal annealing. On the basis of these findings, the authors concluded that the perovskite layer degraded into PbI 2 in less than 10 h. Similarly, Phillippe and co-workers investigated the effect of higher temperatures on both MAPbI 3 and MAPbI 3−x Cl x films [46], but used hard X-ray photoelectron spectroscopy (HX-PES) instead of XRD in their study. HX-PES aids the understanding of the film decomposition procedure by determining the chemical composition of the sample regardless of crystallinity. To investigate the decomposition process, the samples were prepared in an oxygenand water-free environment and were heated in an analysis chamber under ultra-high vacuum. The films were quantitatively analyzed using HX-PES by extracting the I/Pb and N/Pb atomic ratios. The reduction in these ratios represents the conversion of perovskite to PbI 2 . The atomic ratio of I/Pb (red circle) and N/Pb (black diamond) plotted at various temperatures are shown in Figure 3d. Heating at 100 • C for 20 min significantly reduced both ratios. Furthermore, upon heating at 200 • C, the two ratios decreased to 2 and 0, respectively. This indicates that the film decomposed completely into PbI 2 .

Morphological Degradation
Several reports have investigated the effect of temperature on the morphology of perovskites. As mentioned above, Conings and co-workers [38] studied the thermal stability of perovskite solar cells with an ITO/TiO 2 /MAPbI 3 structure when subjected to heat treatment at 85 • C for 24 h under various environmental conditions. High-angle annular dark field (HAADF) TEM was used to understand the effect of temperature on devices in different environmental conditions. Cross-sectional views of the HAADF images are shown in Figure 4a,b. The pristine perovskite device exhibited well-defined layers stacked evenly on top of each other, while the degraded samples exhibited structural variations. Interestingly, a large PbI 2 grain, highlighted in red, is formed in the O 2 environment (see Figure 4c). The calculated ratio of Pb/I determined by EDX revealed the highlighted oblong structure to be PbI 2 . When exposed to ambient atmosphere, more severe device degradation occurred with spot-like structures in the perovskite and delamination from the TiO 2 layer, as shown in Figure 4d. Additionally, topographical AFM images (Figure 4e-h) showed significant structural changes for samples measured under ambient conditions. Significant dark areas appearing in AFM images were associated with PbI 2 regions, which indicates a photocurrent reduction in those regions upon degradation.
Han et al. [17] tested the stability of perovskite solar devices in an environmental chamber, where the temperature was controlled in the range of −20 • C to 100 • C. The actual cell temperature inside the chamber was found to be approximately 30 • C higher than the environmental temperature, so the temperatures were denoted as "environmental temperature (actual cell temperature)", e.g., 55 • C (85 • C) [17]. The degradation mechanism of CH 3 NH 3 PbI 3 solar cells was analyzed for 500 h at 55 • C (85 • C) using cross-sectional focused ion beam-scanning electron microscopy (FIB-SEM). The cross-sectional FIB-SEM images shown in Figure 5 reveal the degradation mechanism of an encapsulated device. The direct exposure equivalent to one sun illumination clearly damaged the entire device, with the most degradation observed in the silver layer. A number of degradation features were observed in the degraded cells; including degradation of the silver layer, formation of voids in spiro-OMeTAD and the perovskite layer, and delamination of the perovskite layer from the TiO 2 layer, as shown in Figure 5a-c. Particularly, the formation of a PbI 2 layer was found, as indicated by a bright contrast in the SEM image (see Figure 5d). It is anticipated that the degradation was initiated by the reaction of HI gas and Ag in an encapsulated device structure. Therefore, the authors suggested replacing the silver contact and using highly heat-resistant encapsulating materials.

Optical Degradation
Foley et al. [47] investigated the temperature dependence of the bandgap and energy levels of MAPbI3 by employing absorbance and photoluminescence (PL) spectroscopy (Figure 6a,b). They showed that both valence band maximum (VBM) and conduction band minimum (CBM) levels were reduced by 110 meV and 77 meV, respectively, when the temperature was increased from 25 °C to 85 °C. The bandgap of MAPbI3 also increased by 33 meV, with an increase in temperature. This provided deeper insights into the relationship between the observed shift in VBM level and thermal expansion of the lattice using density functional theory (DFT) calculations. These results are significant when designing MAPbI3 solar cells at different operating temperatures. Recently, the effects of heating and cooling on the performance of MAPbI3−xClx perovskite solar cells were investigated by varying the device temperatures from room temperature to 82 °C and then returning them to room temperature [48]. For this study, temperature-dependent steady-state PL and time-resolved lifetime decay measurements were investigated using a Linkam THM S600E system. Interestingly, an irreversible optical and electrical degradation of the perovskite films was observed at 70 °C. This irreversible

Optical Degradation
Foley et al. [47] investigated the temperature dependence of the bandgap and energy levels of MAPbI 3 by employing absorbance and photoluminescence (PL) spectroscopy (Figure 6a,b). They showed that both valence band maximum (VBM) and conduction band minimum (CBM) levels were reduced by 110 meV and 77 meV, respectively, when the temperature was increased from 25 • C to 85 • C. The bandgap of MAPbI 3 also increased by 33 meV, with an increase in temperature. This provided deeper insights into the relationship between the observed shift in VBM level and thermal expansion of the lattice using density functional theory (DFT) calculations. These results are significant when designing MAPbI 3 solar cells at different operating temperatures. Recently, the effects of heating and cooling on the performance of MAPbI 3−x Cl x perovskite solar cells were investigated by varying the device temperatures from room temperature to 82 • C and then returning them to room temperature [48]. For this study, temperature-dependent steady-state PL and time-resolved lifetime decay measurements were investigated using a Linkam THM S600E system. Interestingly, an irreversible optical and electrical degradation of the perovskite films was observed at 70 • C. This irreversible degradation continued up to 82 • C and even when the device was cooled back to below 82 • C. PL measurements, shown in Figure 6c,d, demonstrate this irreversible degradation. An increase in the disordered phase of the perovskite films (observed by PL measurements) was found to result in a drastic increase in charge trapping and the development of a deeper trap depth. Temperature-dependent lifetime decay measurements were performed and the trapping and detrapping rates were calculated [48] to show a trap depth of 145 meV for temperatures in the range 40-70 • C. However, a deeper trap depth of 1.1 eV was observed at temperatures over 70 • C, indicating the formation of deep trap states due to the thermal decomposition of the perovskite layer (Figure 6e,f). Herz et al. also observed a gradual shift in the band-edge of MAPbI 3 at elevated temperature (100 • C) based on PL and transmittance measurements [49]. These results suggest that the observed intrinsic thermal decomposition of perovskite solar cells at such low temperatures may limit their commercial applications.

Impact of Thermal Stress on Various Layers and Possible Solutions
We now move the stability discussion to the other important layers of perovskite solar cells. In particular, we will discuss how the temperature affects the electron transport layer (ETL), the hole transport layer (HTL), and the metal electrode.

Electron Transport Layers
The most commonly used ETL in perovskite solar devices is titanium dioxide (TiO 2 ). However, non-stoichiometric defects, such as oxygen vacancies and titanium interstitials, can also form in this layer [50] and cause deep sub-band gap trap states. Perovskite devices with TiO 2 ETL show rapid degradation under illumination. Recently, Ahn et al. reported the degradation of the photo conversion efficiency of perovskite solar cells due to the decomposition of the perovskite film at the interface with TiO 2 [51]. As shown by the SEM images, degradation is initiated at the perovskite/TiO 2 interface in a TiO 2 -based device. The authors postulated that the trapped charges at the interface were responsible for the irreversible degradation of perovskites along grain boundaries [51]. To enhance stability, compact TiO 2 ETL can be replaced with C 60 , which shows much more stable performance when deployed in the device structure.
Wojciechowski et al. used C 60 as interface modification layer for TiO 2 and as an electron-accepting layer [52,53]. They performed steady-state PL measurements over time at the PL peak of 775 nm in the devices with TiO 2 and C 60 as ETLs (See Figure 7a,b) [53]. The C 60 layer was shown to increase the photoluminescence decay and exhibited more stable device performance under full-spectrum illumination for 500 h, as shown in Figure 7c [53]. The control device with TiO 2 as an n-type layer showed a notable reduction in the photovoltaic performance with time as shown in Figure 7d, while the C 60 based device exhibited much more stable performance when annealed at 60 • C for 500 h. To overcome the non-stoichiometric defects of TiO 2 , Pathak et al. [50] used Al-doping for their TiO 2 , and developed a sol-gel deposition process with an Al-containing precursor. Photo-thermal deflection spectroscopy (PDS) measurements indicated that Al-doping in TiO 2 reduced the number of trap states and they found that Al-doping not only passivated non-stoichiometric defects but also improved the stability of encapsulated devices.
increase the photoluminescence decay and exhibited more stable device performance under fullspectrum illumination for 500 h, as shown in Figure 7c [53]. The control device with TiO2 as an n-type layer showed a notable reduction in the photovoltaic performance with time as shown in Figure 7d, while the C60 based device exhibited much more stable performance when annealed at 60 °C for 500 h. To overcome the non-stoichiometric defects of TiO2, Pathak et al. [50] used Al-doping for their TiO2, and developed a sol-gel deposition process with an Al-containing precursor. Photo-thermal deflection spectroscopy (PDS) measurements indicated that Al-doping in TiO2 reduced the number of trap states and they found that Al-doping not only passivated non-stoichiometric defects but also improved the stability of encapsulated devices.

Hole Transport Layers
The most commonly used hole transport layer in inverted perovskite device structure is poly(3,4-ethylenedioxythiophene) polystyrene sulfonate (PEDOT:PSS). However, the prime disadvantage of the PEDOT:PSS based hole transport layer (HTL) is that it leads to the chemical instability of perovskite solar devices due to its acidity and highly hygroscopic nature [54][55][56]. Vitoratos et al. [57] investigated the thermal stability of PEDOT:PSS and found a decrease in the electrical conductivity of perovskites with aging due to the shrinkage of PEDOT conductive grains. Moreover, the XPS and UPS measurements suggested a broken conformational PEDOT:PSS structure, which results in the thermal instability of this material [57]. Despite this, a recent study [58] showed the potential for improved efficiency and stability of perovskite solar cells with enhanced PEDOT: PSS HTLs. This was done by fabricating PEDOT:PSS monolayers, using spin coating and ultrapure water rinse, and constructing ultra-thin layer of PEDOT:PSS. The water rinsed device sustained more than 50% of its stability, attributed to the less hygroscopic nature of the thin PEDOT:PSS layer [58]. Another improvement approach is to use an oxide-based hole transporting material such as NiO, which exhibits better air [59][60][61] and thermal stability [62].
Xu et al. [63] synthesized cross-linked films using an arylamine derivative (N4,N4 -Di(naphthalen-1-yl)-N4,N4 -bis(4-vinylphenyl)biphenyl-4,4 -diamine) (VNBP) [63]. A two-step process was employed to construct a thermally cross-linked hole extracting layer, as shown in Figure 8a. In the first step, a layer of VNBP is deposited and then thermally cross-linked; in the second step, an interface doping layer of MoO 3 is simply deposited on top of the HTL. The device architecture consisted of a double layer of VNBP-MoO 3 , and showed a steady-state efficiency of 16.5% and improved thermal stability compared to the control device. The device performance shown in Figure 8b,c indicates that the devices with cross-linked VNBP-MoO 3 HTL retained almost 90% of their initial performance with low hysteresis for 1 h of annealing at 110 • C. Figure 8d illustrates the optical microscopy images of the control and cross-linked device before and after annealing. No morphological deformation was found for the cross-linked device, while the control device with spiro-OMeTAD HTL exhibited the chain-like structures which lead to irreversible degradation. The thermally cross-linked devices were found to be thermally stable for 30 days during thermal annealing at 110 • C for 1 h and high humidity exposure at 70% relative humidity, as evidenced by the XRD patterns (Figure 8e). The authors suggested that the exceptional barrier properties of cross-linked organic hole transport material with metal oxide allowed the device to maintain high performance under thermal stress for a longer period than that found in previous studies.  Habisreutinger et al. investigated the effect of the HTL on the thermal degradation of perovskite solar cells [64]. A schematic illustration of the investigated device structure is shown in Figure 9a. The authors investigated thermal stability by employing the three most common and highly efficient HTLs, Spiro-OMeTAD, P3HT, and PTAA. A rapid degradation of the perovskite films was observed for all three HTLs, as evidenced by the photographs in Figure 9b. To achieve better thermal stability, organic HTLs were replaced with functionalized single-walled nanotubes (SWNT) embedded in an insulating polymer of polycarbonate (PC) and poly (methyl methacrylate) (PMMA). The absorption and XRD measurements revealed that the devices with spiro-OMeTAD degraded when heated at 80 • C for 96 h in ambient environment. In contrast, devices with PMMA and PMMA-SWNT showed less degradation in absorption spectra and no significant change in the XRD patterns, as shown in Figure 9d,e. Devices fabricated with P3HT/SWNT-PMMA HTLs did not exhibit PCE reduction when annealed at 80 • C. Interestingly, the efficiency increased from 13.3% to 14.3% with an annealing time of 96 h, but a further increase in temperature led to a reduction in PCE (see Figure 9c), possibly due to the degradation of the perovskite layer at such high temperatures.

Metal Contacts
The most commonly used electrodes for high efficiency perovskite solar devices are gold (Au) or silver (Ag) [65][66][67]. Gold is expensive and can be typically replaced by Ag as an electrode in perovskite solar cells [18,68], but both Ag and Au layers show degradation when exposed to thermal stress [69,70]. Domanski

Metal Contacts
The most commonly used electrodes for high efficiency perovskite solar devices are gold (Au) or silver (Ag) [65][66][67]. Gold is expensive and can be typically replaced by Ag as an electrode in perovskite solar cells [18,68], but both Ag and Au layers show degradation when exposed to thermal stress [69,70]. Domanski et al. reported the migration of gold through HTL into the perovskite materials at 70 • C [70]. Li et al. showed, in an inverted perovskite solar cell, that ions migrating from MAPbI 3 thin film diffused through the Phenyl-C61-butyric acid methyl ester (PCBM) ETL and accumulated at the Ag surface in N 2 at 85 • C [71]. The proposed mechanism of the Ag-electrode induced thermal degradation of the inverted PSCs is shown in Figure 10a. The loss of methylammonium iodide (MAI) occurred at the grain boundaries (GBs), and upon thermal annealing, the AgI, MA + , and I − ions migrate to reconstruct the GBs, creating more defects in the bulk grains and at the MAPbI 3 /PCBM interface. Multiple reports have investigated an alternative to Au and Ag based electrodes by employing carbon as a back-contact material [16,72]. Li et al. [16] fabricated a hole-conductor-free PSC based on a triple layer architecture with printed carbon electrode. The triple layer consists of mesoporous TiO 2 and ZrO 2 , as well as a 10 µm thick carbon layer. A schematic is shown in Figure 10b. The perovskite precursor solution was infiltrated by drop casting using a carbon counter electrode. The fabricated perovskite was (5-AVA) x (MA) 1−x PbI 3, where 5-AVA is 5-ammoniumvaleric acid iodide. The device structure displayed excellent stability and an encapsulated device showed stable performance during outdoor operation for seven consecutive days (Figure 10c) in Jeddah, Saudi Arabia [16]. Moreover, the encapsulated device remained thermally stable when annealed at 80-85 • C for 90 days, in the dark. Hence, the use of carbon as a back-contact can be beneficial to make stable PSCs at a temperature of 100 • C, representing progress toward the commercialization of PSCs [73]. Baranwal et al. [73] also fabricated a three-layer printable HTL-free MAPbI 3 PSC with a mesoporous carbon back contact and UV-curable sealant and performed thermal stability tests of over 1500 h at 100 • C. Interestingly, they found that the position of the UV-curing glue considerably affected device stability. The PCE of side-sealed devices (Figure 10e) remained almost constant during the thermal test over 1500 h (Figure 10f), while for the over-sealed device (Figure 10d), the PCE deteriorated to 20% of the initial value during the 100 • C thermal stress test after only 30 h.

Other Strategies to Improve Thermal Stability
Several techniques can be used to achieve thermally stable perovskite solar cells. Manipulating the electron and hole transport layers and metal contacts as discussed extensively in the previous section is one approach. Here, we describe complementary technologies that researchers have adopted to improve the thermal stability of perovskite solar cells.
Yun et al. [74] showed that the thermal stability of perovskite solar cells can be improved by constructing 2D/3D stacking structures. 2D perovskite materials exhibit better moisture stability than 3D perovskite materials [75][76][77][78][79], but the highest efficiency of 2D perovskite solar cells reported in literature is relatively low [80]. Yun et al. fabricated a thin layer of 2D perovskite on top of 3D perovskites in order to cover the surface and grain boundaries. Since the grain boundaries of the perovskite are defective and degradation occurs at the grain boundaries [81], covering the crystal boundaries of the 3D structure by the 2D stacking structure improves overall device stability and maintains high efficiency. They constructed 2D/3D stacking structures by reacting n-butylamine (BA) with the surface of 3D perovskites (see Figure 11a-c). Two types of solutions were used to investigate the thermal stability: BA dissolved in chlorobenzene and n-butylamine iodide (BAI) dissolved in isopropyl alcohol. The SEM images (Figure 11d-f) revealed that the BA solution spun-coated on top of the 3D perovskite layers enhanced the surface smoothness due to the gradual diffusion of BA molecules into the perovskite crystal structure, passivating surface and grain boundaries. The BAI treated film exhibited a relatively rougher surface morphology (Figure 11f), which was attributed to 3D to 2D phase transition [82]. The thermal stress studies using capacitance versus frequency measurements showed that when the device was annealed at 85 • C for 20 h, the device based on the 3D structure had a capacitance greater than that of the fresh perovskite in the low-frequency regime. This high capacitance in the low-frequency regime was attributed to ionic migration at moderately high temperature [8,83,84]. The modified perovskite film also showed better thermal stability under continuous annealing for 1000 h compared to the fresh perovskite, with the modified devices maintaining 96.5% and 88.2% of their initial efficiencies for devices with BA and BAI treatments, respectively, while the fresh device degraded to 69.8% of its initial efficiency under thermal stress [74]. days ( Figure 10c) in Jeddah, Saudi Arabia [16]. Moreover, the encapsulated device remained thermally stable when annealed at 80-85 °C for 90 days, in the dark. Hence, the use of carbon as a back-contact can be beneficial to make stable PSCs at a temperature of 100 °C, representing progress toward the commercialization of PSCs [73]. Baranwal et al. [73] also fabricated a three-layer printable HTL-free MAPbI3 PSC with a mesoporous carbon back contact and UV-curable sealant and performed thermal stability tests of over 1500 h at 100 °C. Interestingly, they found that the position of the UV-curing glue considerably affected device stability. The PCE of side-sealed devices (Figure 10e) remained almost constant during the thermal test over 1500 h (Figure 10f), while for the over-sealed device (Figure 10d), the PCE deteriorated to 20% of the initial value during the 100 °C thermal stress test after only 30 h. increasing the efficiency by more than 18% [86]. Habisreutinger et al. proposed a device architecture consisting of single-walled carbon nanotubes (SWNTs) and a polymer matrix of PMMA or PC. The SWNTs in such architecture facilitate selective charge extraction, while the polymer matrix serves as an encapsulating layer to protect the device from moisture penetration [64]. The proposed devices showed enhanced thermal stability at high temperatures of 80 °C in ambient humidity conditions, compared to the control devices. Another approach to improve thermal stability is substituting the methylammonium (MA) cation of the perovskite material, with mixed cation-based perovskites having been extensively Figure 11. (a) A schematic of 2D/3D stacking structure obtained from BA-treated perovskite film. (b,c) 2D/3D molecular junctions on the surface and at grain boundaries of 3D perovskite films with BA and BAI treatments, respectively. SEM images of (d) MAPbI 3 films, (e) BA-treated MAPbI 3 films, and (f) BAI-treated MAPbI 3 films. Capacitance−frequency spectra (g) before and (h) after thermal stress at 85 • C for 20 h of the control, BA-treated, and BAI-treated devices; (i) Time evolution of normalized PCE under thermal stress. Reprinted with permission from [74]. Copyright (2018) American Chemical Society.
Li et al. investigated the thermal stability of MAPbI 3 by modifying its surface. In that study, an additive butylphosphonic acid 4-ammoniumchloride (4-ABPACl) was spun-coated onto the surface [16]. Structural, morphological, and elemental examinations then showed that the additive acted as a crosslink between perovskite grains. The phosphonic acid ammonium additive facilitated obtaining a smooth surface layer because perovskite was incorporated into the mesoporous TiO 2 scaffold. The 4-ABPACl additive increased the absorption and almost doubled the efficiency of the devices, compared to the control device. The effect of thermal stress was investigated on an encapsulated device at 45 • C under 10 mW/cm 2 illumination. The CH 3 NH 3 PbI 3 -ABPA modified devices maintained 90% of their initial efficiency after one week. A heat stress test was performed on CH 3 NH 3 PbI 3 -ABPA device at 85 • C in the dark for 350 h to confirm if the modified device could endure the exposure to high temperature and the modified device maintained 80% of its initial PCE after 350 h of continuous annealing.
Perovskite thin films are very sensitive to oxygen and moisture, even though different cations and mixed halides have been employed in perovskite structures. Moisture ingress is known to accelerate the thermal degradation of perovskite solar cells, so encapsulation is used to protect perovskites from moisture. Matteocci et al. [85] compared five different glass-glass sealing procedures and found an optimized procedure using Kapton and glue bonded glass. They also showed that an additional edge sealing increased moisture resistance, leading to improved thermal stability with the ability to maintain initial PCE for over 1300 h of shelf-life. Furthermore, they found that the optimized sealing procedure prevents the intrinsic degradation of CH 3 NH 3 PbI 3 , although a reduction in PCE of 0.1%/h at 60 • C and 0.21% at 85 • C was also observed. They suggested that this reduction was attributed to the degradation of Spiro-OMeTAD HTL. A different sealing procedure approach can include the use of polymers such as poly(methyl methacrylate) (PMMA), polycarbonate (PC), or polystyrene to protect perovskite films from oxygen and moisture [64]. Along with protection from air, polymers act as an insulating tunneling contact and passivate defects in perovskite film by increasing the efficiency by more than 18% [86]. Habisreutinger et al. proposed a device architecture consisting of single-walled carbon nanotubes (SWNTs) and a polymer matrix of PMMA or PC. The SWNTs in such architecture facilitate selective charge extraction, while the polymer matrix serves as an encapsulating layer to protect the device from moisture penetration [64]. The proposed devices showed enhanced thermal stability at high temperatures of 80 • C in ambient humidity conditions, compared to the control devices.
Another approach to improve thermal stability is substituting the methylammonium (MA) cation of the perovskite material, with mixed cation-based perovskites having been extensively studied by researchers. Enhanced crystallinity and structural stability with improved power conversion efficiency were obtained by replacing MA with formamidinium (FA) [87][88][89], cesium (Cs) cations, [90][91][92][93] or a mixture of the two [94][95][96][97][98]. Eperon et al. reported a relatively slower degradation for FAPbI 3 compared to MAPbI 3 when annealed at 150 • C [89]. Although FAPbI 3 is more thermally stable than MAPbI 3 , the structural instability of FAPbI 3 in the presence of moisture limits its use in perovskite solar devices [99]. Replacing FAI by MAI or MABr results in FA 1−x MA x PbI 3 or FA 1−x MA x PbI 3−y Br y perovskites, with a stabilized black perovskite phase [87,88]. Binek et al. achieved 3D black FA-based perovskite by adding a smaller MA cation compared to FA, resulting in no phase transition in the 25-150 • C temperature range [98]. Such stability improvement was ascribed to the larger dipole moment of MA, which increases the Coulomb interactions within the structure [98].
However, since the unstable nature of the MA cation can hinder long-term stability, inorganic cations, such as cesium, have attracted on-going research attention due to their established sustainability at higher temperatures [90][91][92][93]. For example, CsPbI 3 based perovskite forms in a cubic structure with a band gap of 1.73 eV [89,90]. but crystallizes into a photo-inactive yellow phase when exposed to air. In contrast, CsPbBr 3 is less sensitive to moisture [92,100], but does not have a favorable bandgap for single-junction solar cell applications [101]. Since inorganic cesium lead halide perovskites (CsPbX 3 , X = Cl, Br, I) have demonstrated advantageous thermal stability, efforts have been made to develop mixed cation perovskites with simultaneous high efficiency and good thermal stability [28,[94][95][96]102]. Several groups have studied Cs/FA mixtures, showing enhanced stability results in light, moisture, and heat compared to the pure material (FAPbI 3 ). Saliba et al. have also fabricated solar cells using quadruple-cation perovskites (MA/FA/Cs/Rb), revealing improved reproducibility and stability [103]. Saliba et al. tested devices annealed for 500 h at 85 • C under continuous illumination in a nitrogen atmosphere, and the current-voltage characteristics of the best performance solar cells and the external quantum efficiency (EQE) electroluminescence (EL) are shown in Figure 12a-c, where the compounded stress test exceeds industrial standards [104]. They showed that the device retained 95% of its initial performance when annealed at 85 • C for 500 h, as shown in Figure 12d. reproducibility and stability [103]. Saliba et al. tested devices annealed for 500 h at 85 °C under continuous illumination in a nitrogen atmosphere, and the current-voltage characteristics of the best performance solar cells and the external quantum efficiency (EQE) electroluminescence (EL) are shown in Figure 12a-c, where the compounded stress test exceeds industrial standards [104]. They showed that the device retained 95% of its initial performance when annealed at 85 °C for 500 h, as shown in Figure 12d.

Conclusions
Organic-inorganic halide perovskites have been extensively used in solar cells as a promising light harvesting material due to their excellent optoelectronic properties. Although perovskite solar cells have achieved photo-conversion efficiencies above 20%, the instability of their materials at elevated temperatures is a major issue that needs to be solved for commercial use. Particularly, perovskite films tend to exhibit intrinsic thermal degradation mechanisms under thermal stress. Typically, the commonly used MAPbI 3 perovskite material is decomposed into MAI and PbI 2 at elevated temperatures. The decomposition of perovskites has been discussed here from the chemical, morphological, and optical points of view, confirming the deterioration of perovskites during thermal stress. Furthermore, since perovskite solar devices are composed of many layers, notably an electron transport layer, a hole transport layer, and electrodes, other decomposition challenges exist. Thus, to facilitate better thermal stability, the selective incorporation of electron and hole transport layers and the replacement of silver or gold electrodes is discussed, and the enhancement of the stability of the perovskite layer by encapsulation or including additives is also reviewed. Currently, perovskite solar cells have been made thermally stable for annealing of up to 500 h at 85 • C by using quadruple-cation perovskites, but with available solar products on the market (c-Si, CdTe, CIGS, GaAs) guaranteed to remain stable for up to 25 years, perovskites continue to need research attention. Recent advances in perovskite solar cell thermal stability and in determining the causes of degradation indicate that it should be possible to achieve similar longevity, and support the scalable solutions that would allow this promising technology to reach its full potential and compete in the existing solar market.
Author Contributions: Authors equally contributed to the review paper.
Funding: This research was partially supported by the National Science Foundation under projects (1355678 and 1547771) and the Virginia Microelectronic Consortium Program.