Consensus statement for stability assessment and reporting for perovskite photovoltaics based on ISOS procedures

Improving the long-term stability of perovskite solar cells is critical to the deployment of this technology. Despite the great emphasis laid on stability-related investigations, publications lack consistency in experimental procedures and parameters reported. It is therefore challenging to reproduce and compare results and thereby develop a deep understanding of degradation mechanisms. Here, we report a consensus between researchers in the field on procedures for testing perovskite solar cell stability, which are based on the International Summit on Organic Photovoltaic Stability (ISOS) protocols. We propose additional procedures to account for properties specific to PSCs such as ion redistribution under electric fields, reversible degradation and to distinguish ambient-induced degradation from other stress factors. These protocols are not intended as a replacement of the existing qualification standards, but rather they aim to unify the stability assessment and to understand failure modes. Finally, we identify key procedural information which we suggest reporting in publications to improve reproducibility and enable large data set analysis. Reliability of stability data for perovskite solar cells is undermined by a lack of consistency in the test conditions and reporting. This Consensus Statement outlines practices for testing and reporting stability tailoring ISOS protocols for perovskite devices.

T o ensure economic feasibility and competitive levelized cost of electricity, new photovoltaic (PV) technologies must offer long-term stability alongside high power conversion efficiency (PCE). For instance, the lifetime expectation for a PV module in a power plant is 20-25 years, to match the reliability of silicon-wafer-based modules. At present, the long-term stability of emerging technologies such as organic photovoltaic (OPV) cells, dye-sensitized solar cells (DSSCs) and halide perovskite solar cells (PSCs) is not meeting this target and improvements are hampered by a lack of understanding of the module failure modes.
The existing qualification tests described in the International Electrotechnical Commission (IEC) standards on terrestrial PV modules (such as IEC 61215) 1-3 are designed for the field performance of silicon panels to screen for well-understood degradation modes generally associated with issues at the module level. These tests, however, are unlikely to be well-suited to OPV cells, DSSCs and PSCs because of their fundamentally different material properties and device architectures. In fact, various reports have shown that the stability of these devices cannot be fully assessed by the procedures developed for conventional PV products [4][5][6][7][8][9][10] , which led to the publication of various studies that attempted to understand the degradation mechanisms in emerging PV systems. Unfortunately, these studies lacked consistency in the assessment and reporting procedures, which prevented data comparison and, consequently, the identification of various degradation factors and failure mechanisms.
In light of such shortcomings, in 2011, a broad consortium of researchers developed recommendations for evaluating the stability of OPV cells 11 . These standardized ageing experiments (the socalled ISOS protocols) were established at the International Summit on Organic PV Stability (ISOS) held in Roskilde, Denmark, in 2010. They outline a consensus between researchers in the OPV field on performing and reporting degradation studies in a controlled and reproducible way. These protocols are not intended to be a standard qualification test, nor are they suited for application by industry or insurance agencies; however, it is worth mentioning that tests based on the ISOS protocols were recently considered at the IEC level Existing ISOS stability protocols ISOS protocols designed for OPV cells, which are reported in detail elsewhere 11 , are grouped in terms of the applied stresses into five categories, shelf-life or dark storage testing, outdoor testing, light soaking testing, thermal cycling testing and light-humidity-thermal cycling testing, all of which are also highly relevant to PSCs as discussed below. Each category has three levels of sophistication, basic, intermediate and advanced, the goal of which is to cover different levels of laboratory infrastructure. The first, basic level (level 1) requires only commonly available equipment and provides relatively low control over the stress factors. While limited insights into the PSC degradation can be gathered from these tests, we suggest such first-level procedures as the minimum requirement for stability testing. The second and third levels of sophistication (intermediate and advanced, level 2 and 3, respectively) require more specialized tools, such as environmental chambers and maximum power point (MPP) trackers, but provide higher levels of confidence in the results and, in most cases, more stringent test conditions. The protocols can be applied to both encapsulated and unencapsulated devices provided it is clearly reported (see the detailed discussion below in 'Intrinsic stability testing (ISOS-I)'). Stress tests specific to encapsulated modules (including hail tests, potential-induced degradation, bypass diode stability and so on 1,2 ), as well as mechanical stability and special consideration for space applications, are outside the scope of this report, and indeed will likely be leveraged from the existing IEC 61215 standard.
Dark storage studies (ISOS-D) provide information on the tolerance of the solar cells to oxygen, moisture, other aggressive atmospheric components naturally present in air (for example, CO 2 , NO x , H 2 S), and elevated temperatures. In other words, ISOS-D tests estimate a cell's shelf life under ambient conditions when it is not exposed to light. The ambient atmosphere plays a crucial role in determining the lifetime of perovskite absorbers and of some of the transport layers used in PSC architectures 23,37,38 . In particular, interaction with ambient species can promote the formation of traps 39 or charge carrier barriers 40 (as a result of increased density of mobile defects or ions and electronic traps within the active layer) as well as perovskite decomposition, which quickly deteriorate device performance 22 . Atmospheric species were also shown to charge the perovskite surface, affecting the ion distribution across the device 41 . The impact of these factors is taken into consideration in the ISOS-D-1 tests, where the cell environment is monitored but not explicitly controlled (room temperature in the laboratory is assumed to be 23 ± 4 °C).
Another important stress factor is temperature. Elevated temperatures are applied to study the thermal stability of solar cells and to accelerate the degradation induced by other stressors 42 . Thermal degradation in the dark occurs in PSCs at elevated temperatures due to chemical and structural instabilities of the absorber materials 10,43 or transport layers 44,45 . Notably, some metal halide perovskites undergo phase transitions in the temperature range relevant to PV applications 46,47 . At the moment, the impact of phase transitions on the device lifetime is unclear and therefore so is the impact of different temperatures during (accelerated) ageing. As protocols should be applicable to any type of perovskites regardless of their phase transitions, ISOS protocols cannot accommodate all the possible temperature options. We are therefore suggesting maintaining the temperature settings described in the original ISOS protocols that would be above the tetragonal to cubic phase transition for MAPbI 3 . The effects of elevated temperatures on the device stability are assessed with the ISOS-D-2 test that is carried out at controlled elevated temperatures of 65 or 85 °C. While not explicitly controlled in ISOS-D-1 and D-2, we stress that monitoring and reporting the ambient relative humidity (RH) is of critical importance because dry (RH < 20%) and humid air represent dramatically different stress conditions for PSCs 38,48 . The ISOS-D-3 damp heat test specifically takes into account the impact of humidity (set at 85% RH) when devices are kept at high elevated temperatures (65 or 85 °C).
Light soaking tests (ISOS-L, 'Laboratory weathering' in the original ISOS protocols 11 ) promote ion and defect migration in PSCs [49][50][51][52][53][54] as well as phase segregation 55,56 in the perovskite photoactive layer, which reduces efficiency. Additionally, light can catalyse or accelerate harmful chemical reactions, which lead to perovskite decomposition 57 or defect formation 58 . Detrimental changes in organic charge extraction layers, material intermixing at the interfaces 59 and ion exchange between adjacent solar cell layers can also be caused by cell illumination 52,60,61 . As with OPV 62 and DSSC 63 , the spectral composition of the light source also merits special attention, especially in the UV range. UV light assists perovskite decomposition 64 and increases the non-radiative recombination rate in PSCs based on mesoporous TiO 2 (ref. 65 ,) which may thus require UV-blocking layers to become more stable. Recently, PSCs with novel transport layers have been shown to be tolerant to UV irradiation 66,67 . PSC stability may be affected differently by light in the UVA and UVB spectral ranges 68 .
In outdoor stability studies (ISOS-O), ageing occurs by illumination with natural sunlight in the ambient environment. Although these conditions are not necessarily reproducible (they depend on weather, location, season, and so on), the results of outdoor tests are the most relevant to device operation. Unlike other protocols, they can be directly applied to obtain realistic assessments of device lifetime, albeit specific to a given climate. Field tests can also determine whether the list of failure modes identified in the lab is complete and adequate for understanding reliability of the solar cell under real operation and, furthermore, can provide reference points for calculating acceleration factors that correlate lifetime under real-weather conditions to the lifetime obtained under the accelerated stress conditions. Previously, this approach was pursued to help establish IEC 61215 by correlating outdoor tests results for Si modules with results obtained from various acceleration tests [69][70][71] . To date, studies of PSC outdoor stability are scarce [72][73][74][75] , but the community has gained some critical insights with ISOS-O experiments, such as the importance of light-dark cycling 74 and the unexpectedly high opencircuit voltage at low illumination intensities 73 .
Under the ISOS-O-1 protocol, periodic measurements of J-V curves are done under illumination by a solar simulator. In ISOS-O-2, the J-V measurements are periodically acquired under illumination by natural sunlight. ISOS-O-3 requires both in situ MPP tracking under natural sunlight and periodic performance measurements under a solar simulator. The results obtained by J-V measurements and MPP tracking do not necessarily coincide in PSCs (Fig. 1a) 8 , although they generally have similar trends 4,76 . Therefore, it is crucial when characterizing PSCs to rigorously describe the load and recovery time before J-V measurements. We encourage the use of MPP tracking, whenever possible, both as the most practical electrical-bias condition for ageing and as a reliable tool for PSC performance assessment (see further discussion below in 'Checklist for PSC stability studies'). However, we indicate MPP tracking as mandatory only at the third, most advanced level of ISOS protocols. For lower levels of sophistication, we give options of exposure under open-circuit condition or using a fixed voltage bias near the MPP (instead of active MPP tracking) in line with what was suggested in the original ISOS protocols 11 .
Thermal cycling in the dark (ISOS-T) and light-humiditythermal cycling ('solar thermal humidity cycling' in the original ISOS protocols 11 ) (ISOS-LT) allow evaluation of the damage to PV c, PCE evolution of PSCs exposed to continuous (blue curve) or cycled (6/6 h, red curves) illumination by white-light-emitting diodes. d, normalized PCE changes of PSCs exposed to different forward bias in the dark. e, Light J-V curves for a fresh PSC, after 1 min at −20 mA cm -2 and after recovering for over 3 h of MPP tracking. Panel a was adapted with permission from ref. 8  devices caused by diurnal and seasonal variations in the weather in terms of solar radiation, temperature and humidity. These tests are relevant for any outdoor-dedicated PV technology (including PSCs) because they simulate realistic conditions, stimulate failure mechanisms related to delamination of layers or contacts 77 and are included in the qualification standards 78 . Particularly for PSCs, degradation under varying temperature could be more severe than that under constant extreme temperatures, which is attributed to the effect of ion accumulation at the contacts 20 . Delamination from thermal cycling was mitigated by adding a flexible polymer buffer layer around the mechanically fragile perovskite, which resulted in PSCs retaining more than 90% of their performance after 200 cycles between -40 and 85 °C 77 . Temperature and illumination cycling, which resembles weather conditions in central Europe Each test group is divided into three levels of sophistication that reflect the complexity of required equipment and the harshness of the applied stress. reported ISOS protocols are taken from ref. 11 . Proposed additional ISOS protocols are printed in bold. a V OC , V MPP , and J MPP are determined from light J-V curves measured under standard solar cell testing conditions on a fresh device. E g and q are the band gap of the active layer and elementary charge, respectively. b relative humidity is controlled at temperatures above 40 °C and is not controlled for the remainder of the cycle. Env., environmental; OC, open-circuit condition; MPP, maximum power point; rT, room temperature; rH, relative humidity.
for several representative days, was applied to PSCs under a nitrogen atmosphere 79 . This study enabled from a temperature point of view insight into the real-world operation of PSCs and emphasized the complex interplay of temperature-dependent transient effects during the day with reversible and irreversible degradation processes. Depending on the available equipment, temperature cycling varies from simple turning on and off a hotplate installed in an ambient environment to complex temperature and humidity cycles in an environmental chamber. Examples of such cycles are available elsewhere 11,78 .

Suggested specific ISOS protocols relevant for PSCs
Recently some ageing protocols for PSC were suggested 10,28 ; they mostly feature subsets of the original ISOS protocols described above, although some additional ideas, particularly concerning the application of electrical-bias 10 , were also introduced. Below we suggest extensions of the ageing procedures, summarized in Table 1, to account for the specific properties of perovskite materials and solar cells (proposed additional ISOS protocols in Table 1 are in bold). Although a major part of the original motivation for the ISOS protocols was to limit conditions for each level to facilitate comparison, data are limited on the protocols discussed below. We therefore propose a reporting framework, give some example conditions, and discuss why these might be relevant. As in the case of the existing ISOS protocols, we recommend testing the device following the basic procedure (level 1) as a minimum requirement. A similar framework is likely to be useful for OPV and other emerging PV technologies. If these protocols become widespread over the next few years, the community can adopt a more informed decision on a limited number of consensus conditions.

Light-dark cycling (ISOS-LC)
Various PSC degradation modes have repeatedly been shown to be entirely or partly reversible in the dark (often referred to as metastability) 20,36,51,58,74,[80][81][82][83] . Therefore, cycling through light-dark periods to simulate day-night cycles constitutes a significantly different stress test than applying constant illumination (ISOS-L) 4,74 . Two opposite types of dynamics are reported in the literature: reversible photoinduced PCE increase with subsequent decrease in the dark 74,82 ( Fig. 1b) and photo-induced degradation with recovery in the dark 51,58 (Fig. 1c). Reversible performance loss is attributed to cation redistribution 51 , metastable defect formation 58 and reversible chemical reactions 57 . The PCE improvement under illumination after storage in the dark is commonly attributed to the neutralization of interfacial defects by photogenerated charge carriers or to changes in the built-in electric field due to ion migration [84][85][86] . The PCE dynamics during a cycle depends on the status of cell degradation 81 . For example, PSCs have shown a 'fatigue' effect: while the PCE decreased in the dark and recovered under illumination, the rate of PCE restoration reduced with each consecutive light-dark cycle ( Fig. 1b) 82 . Such metastability is attributed to the migration of ions, which is known to be pronounced in metal halide perovskites 83 .
Reversible and irreversible degradation mechanisms may co-exist in a given PSC 79,81,87 . The ISOS protocols revised for PSCs should, therefore, include a group of light-dark cycling protocols to account for the recovery phenomena (ISOS-LC in Table 1). For the ISOS-LC experiments, we suggest exposing the cells to simulated sunlight turned on and off with cycle periods of 2, 8, or 24 h and duty cycles (light:dark) of 1:1 or 1:2. Of the suggested conditions, 24-h-long cycles (12 h light and 12 h dark or 8 h light and 16 h dark) mimic the diurnal sun cycle. However, because the interplay between degradation and recovery in realistic conditions can be complex 79 and depend on cell history, varying cycle duration and duty cycle should provide additional information on the extent of reversibility and sufficient recovery times. At the ISOS-LC-1 level, the solar cell is maintained at ambient conditions while the temperature and RH are monitored, but not controlled. At the ISOS-LC-2 level, the cell is maintained at a fixed set point temperature of 65 or 85 °C in the ambient atmosphere. At the ISOS-LC-3 level, RH is held at 50% and high temperatures.

Electrical bias in the dark (ISOS-V)
Electrical bias causes PSC degradation 25,26,88,89 by triggering ion migration 26 or charge carrier accumulation that result in thermally activated traps formation 90 or detrimental electrochemical reactions 23 . Electric fields also promote moisture-initiated perovskite degradation 89,91 because moisture ingress can lead to the formation of hydrated perovskite phases containing mobile ions, whose drift accelerates the degradation 92 . Both positive (Fig. 1d) and negative (Fig. 1e) biases are potentially harmful 26,88 and might occur during the operation of solar panels. In our view, ISOS protocols revisited for PSCs should include the ISOS-V group of testing in which the behaviour of the cell is analysed when exposed to a certain electrical bias in the dark (Table 1).
Usually, a solar cell is kept near its MPP (that is, positively biased with voltage V < V OC ); however, disconnected cells under illumination are biased at open-circuit voltages (typically, ~ 1 V for iodine single-junction PSCs). Therefore, we suggest applying a voltage equal to V MPP or V OC (as measured under AM1.5 G one sun illumination on a fresh device) as a positive bias condition. Since bias-induced degradation effects may have a threshold behavior 26 , we recommend voltages below the bandgap energy divided by the charge of the electron to avoid unnatural overstressing. In Si PVs, constant-current stress was shown to mimic MPP operation under full sunlight exposure, while testing current-induced degradation is technically easier 93 . To date, however, no similar data exist for PSCs, so such stability tests might also be useful.
For a partly shaded solar module (shaded by clouds, dirt, nearby trees, and so on), shaded solar cells can be forced to operate under reverse bias to match the current flow through the rest of the module 88,94,95 . The choice of negative-bias stressing conditions depends on the anticipated module connection scheme, particularly on the use and choice of bypass diodes. At present, experience with perovskite modules is insufficient, so it is reasonable to learn from both experiments: with a constant negative bias applied (for example, −V OC ), which is relevant for modules with bypass diodes, and with the current enforced up to −J MPP (which in the dark would Ticks and dashes denote the presence and absence of the stress, respectively. In case of temperature, '-' refers to room temperature. mean a relatively high negative bias applied to the cell). The latter condition simulates the situation of a partially shaded module in the absence of a bypass diode. Practical negative-bias conditions depend on the details of the module layout. The three sophistication levels of the ISOS-V protocols differ by the level of control over the sample temperature and atmosphere and the required equipment, which is similar to the corresponding ISOS-D protocols (Table 1) 11 . Electrical bias can redistribute charged species across the PSC, which might be reversible after stress removal 26 . Thus, we recommended tracing the solar cell recovery after ageing by storing it in the dark under open-circuit (disconnected) conditions and periodically checking its performance until it reaches saturation. For a similar reason, measuring the steady state of the J-V curve may be difficult after electrical biasing. We recommend using MPP tracking (or a stabilized current at a constant voltage close to MPP for ISOS-V-1 and ISOS-V-2) to account for possible transient effects.
It may also be informative to report the evolution of the in situ dark current (or voltage in constant current density mode) during stressing in addition to periodic J-V measurements in ISOS-V protocols.

Intrinsic stability testing (ISOS-I)
The stress factors can be divided into two groups: intrinsic and extrinsic. The intrinsic factors include light, temperature, and electrical bias (relevant regardless of the cell encapsulation or protective environment), and the extrinsic factors are governed by the cell interactions with ambient species such as oxygen and/or moisture (which are relevant assuming imperfect device encapsulation).
Generally, 'encapsulation' of devices refers to the protection of the solar cells by gas-barrier materials, which delays the contact between the cell and ambient air (especially moisture). Encapsulation can be done by glass-glass sealing, lamination of rigid or flexible gas-barrier materials, direct deposition of protective layers  or a combination of these processes. Extrinsic stability depends on the device sensitivity to air and on the properties of the barrier material (including their chemical compatibility with the device). A full understanding of degradation of encapsulated devices requires knowledge of the stability and properties of the gas barrier. Specific tools exist to characterize the gas-barrier properties and to determine the amount of moisture that has permeated orthogonally [96][97][98][99] and laterally 100,101 within the encapsulation. Any measurement of the lateral permeation from the encapsulation edge should mimic, to the extent possible, the operational encapsulation and take into account the interfacial permeation that is not considered in gas-barrier measurements of bulk materials. Control of the self-resistance of encapsulation is particularly important because ageing tests, which involve high temperature, high humidity and UV irradiation, could degrade the gas-barrier protection, leading to a dramatic drop in PV performance. Therefore, we recommend ageing encapsulating materials under the same ageing conditions as done with encapsulated devices and to determine the gas-barrier properties after ageing. Developing dedicated encapsulation procedures constitutes a separate technological challenge, especially for PSCs 77,102-105 .
In the vast majority of studies, the barrier properties of the encapsulants are unknown, which inhibits the ability to differentiate between intrinsic and extrinsic cell stability. Even if the device is nominally 'unencapsulated' , the top evaporated electrode can play the role of barrier (with unknown properties). This has motivated numerous studies to focus on intrinsic PSC stability by stressing the cells in an inert atmosphere, for example, in sealed pouches, or with equipment installed in inert atmosphere glove boxes or environmental chambers. This approach provided important insights into PSC degradation mechanisms and is helpful for differentiating between how thermal stress, light, electrical bias and the cycles thereof affect device degradation 4,106,107 .
We suggest including protocols to address the intrinsic stability of solar cells in inert atmospheres (nitrogen, argon, and so on). The protocol is labelled by the index 'I' at the end of the protocol name (Table 2) to indicate an inert atmosphere in the corresponding test, with the other parameters kept the same. For example, ISOS-L-1I stands for the intrinsic photo-stability at room temperature (similar to ISOS-L-1 except that the atmosphere is inert), ISOS-L-2I stands for the intrinsic photo-stability at elevated temperature, and so on. The latter protocol is essential because it is often used to determine the lifetime of a given PSC in research papers. Notably, the new family of protocols includes ageing experiments with a single stress factor (only heat in ISOS-D-2I; only electrical bias in ISOS-V-1I; only light in ISOS-L-1I, and so on), which simplifies the analysis of degradation modes. We also included ISOS-LC-3I and ISOS-T-3I protocols in Table 2 despite the original conditions requiring a certain relative humidity value. The advanced, level 3 protocols differ from LC-1,2 and T-1,2 in the temperature cycle and some technicalities that are reported elsewhere 11 . For instance, ISOS-T-3 goes to -40 °C, while ISOS-T-2 only down room temperature. We therefore suggest performing ISOS-LC-3I and ISOS-T-3I tests in the absence of humidity, given that I-tests are conducted in an inert atmosphere, to stress the device with a harsher temperature cycle.
Note that encapsulation not only inhibits reactions with ambient species but may also prevent out-diffusion of volatile perovskite decomposition products from the device. Recently the light-induced degradation of unencapsulated PSCs was also shown to accelerate in an ultrahigh vacuum 108 . Therefore, encapsulated PSCs may have a longer lifetime than unencapsulated PSCs, even for degradation experiments conducted in an inert atmosphere. In addition, the environment in which the PSC is encapsulated may also play a role. Thus, reporting the presence or absence and the details of encapsulation is also mandatory in I protocols.

Checklist for PSC stability studies
To compare results and ensure reproducibility, sufficient information must be reported about ageing experiments, in addition to giving a detailed description of device preparation 27 . Table 3 proposes a checklist for reporting stability data in accordance with that introduced by Nature journals for reporting PV performance data 109 . We stress that, even if a parameter is not controlled during the ageing experiment (for example, temperature or RH in 'ISOS-1' protocols), it is still important to monitor and report the parameters listed in Table 3.
In particular, we recommend specifying the number of samples studied in each ageing test. According to a critical analysis of the quality of PSC stability studies 27 , nearly half of the studies consider only a single sample of each kind, which is particularly worrisome for PSCs that are typically characterized by relatively low reproducibility. Ideally, statistics should be provided to account for sampleto-sample and batch-to-batch variations. The same work provides estimates of the desired sample size 27 .
Stability data are often reported in the form of normalized parameters as a function of ageing time, while only specifying the performance of a fresh representative device (the 'champion' or average). Thus, the reported stabilities and efficiencies may be measured on different devices and thus cannot be directly related. Any plot with normalized parameter variation should include the values to which each parameter was normalized 9 .
Due to the ongoing development of best practices for measuring J-V curves and efficiency of PSCs [28][29][30][31]110 , the procedure for making periodic measurements during ageing tests should be clearly described. Typically, measurements of the J-V curve (or parts thereof) are made with a periodicity that depends on the characteristic degradation timescale of each given device. Because J-V hysteresis is common in PSCs, steps should be taken to ensure that the measurements are taken under (quasi) steady-state conditions. This can usually be done by using a dynamic J-V approach 29,111 , which allows time at each voltage step for the current to settle (stabilize), or, alternatively, by using a very slow J-V scan in the vicinity of the MPP, usually repeated in the reverse direction to check for consistency. If suitable equipment is available, a third approach to logging the steady-state performance of the device over time is MPP tracking (MPPT) 6,30,35 . MPPT simultaneously holds the device at its normal operating voltage and measures the output. However, due to hysteresis, standard perturb and observe MPPT algorithms might be suboptimal for PSCs 112 . Modifications of a perturb and observe algorithm were suggested for effective MPPT in solar cells exhibiting hysteresis, including extended thresholds for switching the voltage sweep direction 112 and predictive MPPT algorithms with lowered settling times 113,114 .If any form of MPPT is used during an experiment, the hardware and MPPT algorithm should be clearly referenced. Similar to organic solar cells 115 , preconditioning the PSC (with light and/or electrical bias) prior to each J-V scan may affect the outcome and should be reported.
MPPT is also recommended as a bias condition for ageing experiments done under illumination (mandatory only at the third sophistication level). When MPPT is performed throughout the entire ageing test, it is recommended to periodically measure J-V curves, because such measurements provide information on the degradation mechanism that is more detailed. Additional nondestructive characterizations at the intermediate stages of PSC ageing are also encouraged, although care must be taken to account for possible cell recovery during the measurement.
Ideally, light sources with an irradiance of 800-1000 W m -² (1 sun = 1000 W m -²) should be applied, and the exact irradiance, the type of light source and its spectrum should be reported. Table 4 shows a collection of the most stable reported PSCs to date that can withstand over 1000 h of light-soaking while losing <15% of their initial PCE. Five main types of light sources were used: sulfur plasma lamp, white light-emitting diodes (LEDs), metal halide xenon lamp, solar simulator and outdoor (that is, real sunlight). The solar simulator section encompasses devices analysed under unspecified conditions of irradiation. Sulfur plasma and white LED illumination typically do not include UV light, so it is redundant to report the use (or not) of a UV filter (except for some modern LED sources with extended range, which may have components in the range 300-400 nm); in fact, reporting that "no UV filter was applied" would be misleading in this case. Metal halide and xenon arc lamps have UV light in their spectra, so any filtering used must be reported. Reporting the company and model of the solar simulator is good practice. Note that ASTM's class ' A' for simulated solar spectrum is only relevant in the range 400-1100 nm 116 . Therefore, the type of light source used and its spectrum should always be clearly specified. For example, 2 T S80 20% of PCE decay from a certain PCE value during the ageing experiment, corresponding to t = T max or t = T burn-in : from the extrapolated t = 0 value from the postburn-in decay fitting (see Fig. 2e,f). For cells with an increase in PCE, T 80 should be estimated for time at which the efficiency has dropped to 80% of the maximum PCE, with the complete time from t = 0 to this point quoted as the T 80 value.
3 η 1000 (PCE after 1000 h) In case T 80 is not reached within the timeframe of the ageing experiment, so the decrease observed over first 1000 h should be reported in addition to (optionally) an extrapolation applied to determine T 80 and/or T S80 .

T 80 analogue, corrected for the recovery processes
If the restoration process has been tested after the stress removal.

5
T 95 and T S95 Analogous to T 80 and T S80 , apart from to 95% of the t = 0, maximum, or post-burn-in back-extrapolated t = 0 PCE. reporting ' AM 1.5 G illumination' without specifying the light source, the solar simulator details and the calibration procedure, is not sufficient. Furthermore, some of the light sources described above (especially xenon lamps) may degrade significantly on the timescale of stability experiments, so it is recommended to periodically check the light intensity with a reference cell.
Several reports claim that PSCs, as well as DSSCs 117,118 and organic solar cells [119][120][121][122] , might be suitable for indoor and outdoor low-light-intensity applications 123,124 . Indoor-light illuminance is significantly lower (100 to 200 lux in a typical home, and 250 to 1000 lux in an office 125 , which corresponds to ~ 0.01 sun irradiance). The spectra of indoor-light sources also differ significantly from natural sunlight, and there is still no standard spectrum for indoor PV testing 126,127 . The share of LEDs in the indoor lighting market is expected to increase due to their high lighting efficiency 128 . For PSCs intended for low-intensity illumination, we encourage device characterization at several intensity levels (for example, 200, 500, 1000 lux) and reporting the light source (preferably LED) spectrum in accordance with the original ISOS procedures 11 .

Stability figures of merit and acceleration factors
The time required to drop to 80% of the initial efficiency is commonly denoted T 80 and often serves as a figure of merit for solar cell stability. It would, therefore, be an optimal minimum ageing test time. Extrapolation of degradation data (or readily achieved T 80 lifetimes) can be used to evaluate the cell lifetime energy yield 129 , which is an important parameter for calculating return on investment and life cycle analysis. Despite the apparent simplicity, there are several approaches 129,130 to determine T 80 for devices (see Fig. 2 and Table 5 for a summary). Figure 2 explains three approaches for calculating T 80 , which are applicable to different ageing scenarios. It is, therefore, vital to detail the metrics used when reporting stability studies. In particular, the original ISOS protocols 11 suggested the use of a 'stabilized T 80 time' (denoted T S80 ), which is the time during which the PCE decreases by 20% of its magnitude after an arbitrarily defined stabilization time (indicated in blue in Fig. 2f). This suggestion is based on the widely known shape of the 'PCE versus time' curve in OPVs, which reflects a rapid initial degradation ('burn-in') 129 followed by a stabilized region. Although similar dynamics occur in some PSCs 4,131,132 , it is not a universal trend for these devices. Figure 2 shows various examples, including the nonmonotonic evolution of PCE with time.
For the most stable PSCs, T 80 exceeds 1000 h (42 days) or even 10000 h ( > 1 year) for two dimensional (2D)-3D perovskites 133 under certain stress conditions, including illumination (see Table 4). Considering recent advancements in PSC stability, we strongly recommend that reviewers and journal editors discourage the use of the word 'stable' in the title of scientific papers in an unspecific manner (thermal, photo-, operational, and so on) and without matching the state of the art for a specific device type. Notably, such long exposure times are challenging to realize. If T 80 is not reached, it is difficult to predict the lifetime based on the observed 'PCE versus time' trend because of the variety of possible curve shapes (see Fig. 2). In this case, we suggest ageing the sample for at least 1000 h and using the PCE after 1000 h of stress (η 1000 , as a percentage of the initial PCE) as the stability figure of merit. If researchers choose to apply any type of extrapolation to determine T 80 or T S80 , it must be clearly differentiated from the measured data. We recommend limiting the extrapolation times to less than one order of magnitude beyond the actual ageing time. As PSC stability improves, it may become common to quote T 95 , which is the time required to degrade to 95% of the initial efficiency. This will also be in accordance with the IEC procedures, where the pass criterion is for the modules, after exposure to stress, to operate at > 95% of their starting performance.
The presence of reversible degradation in PSCs makes it harder to assess their stability. Although no broadly accepted figures of merit currently exist to account for partial reversibility, some procedures are currently under debate 8,74 . Recovery effects can be studied in two types of experiments: continuous ageing followed by performance tracking after stress removal, or cycled stress experiments. In the former case, it was suggested to correct T 80 to account for the restoration done during the rest period 8 . In cycling experiments (such as ISOS-LC), an analogue of the T 80 metric might be introduced for the energy output per cycle 74 .
The ISOS testing protocols do not provide direct information on the expected lifetime of solar cells under operational conditions. For such evaluations, the concept of acceleration factor (AF) has been used 134 . The AF is a constant that relates the times to failure in an accelerated stress test with that in a reference stress test. Once determined for an ageing protocol (and validated through real-world operation), the AF provides an estimate of the solar cell lifetime in a fast and reproducible manner in the laboratory. AF for each stress should, in principle, be derived from a physical model, such as the activation energy (that is, the Arrhenius factor) in thermal  Inert atmosphere  D-1I  D-2I  T -1I  T-2I  T-3I   V-1I  V-2I  L-1I  L-2I  LC-1I LC-2I, 3I Ambient humidity  42,134,135 . AFs for most ISOS tests were determined relative to outdoor conditions in northern Europe using OPV mini modules 134 . Overall, dark-storage ISOS-D tests can yield AFs between 0.45 (ISOS-D-1) and 12 (ISOS-D-3). Light soaking ISOS-L tests gave an AF of 15 and 24 in ISOS-L-1 and ISOS-L-2, respectively. To further ramp up the development of stable PSCs, some suggest using high-intensity light 5,136-139 , which has been used for degradation studies of perovskite absorbers 136 and solar cells 138 with intensities up to 100 and 10 suns, respectively. Rough estimates show that even moderate light intensities of several suns can tremendously accelerate the degradation. For example, a PV operating for 1000 h under continuous illumination with an intensity of only 5 suns under 85 °C is estimated to provide an equivalent of tens of years of outdoor testing if degradation scales linearly with light intensity 5 . Such intensities are easily achievable with commercial light sources and solar concentration.
Although high light intensities (up to hundreds of suns) can provide substantially higher AFs, such tests require careful, independent control of the solar cell temperature and illumination intensity 140,141 . Moreover, care should be taken to check whether the same degradation mechanisms occur under high light intensity and under 1 sun. Such experiments were done for OPVs 135,142,143 and predicted lifetimes in the range of tens of years 143,144 . To the best of our knowledge, no similar studies have yet been reported for PSCs.

Protocols applications and outlook
We suggest that future stability studies include at least one or, ideally, multiple ISOS tests with level 1 procedures being the minimum requirement for testing. Using the labelling we proposed would allow easy identification of the testing conditions used. We recommend reporting all the data indicated in Table 3 to foster transparency and reproducibility.
Getting a better understanding of solar cell failure modes is critical for PSCs as they are still at the early stage of technology development. The ISOS protocols we have discussed in the present article can be used to screen a variety of stressor combinations. Figure 3 summarizes the relationship between the protocols with the major currently known degradation factors for PSCs: atmosphere, temperature, electrical bias and light. For each combination of light and bias, nine types of ageing protocols might be suggested with respect to temperature (ambient, elevated or cycled) and atmosphere (inert, ambient or controlled humidity). Apart from determining device lifetime under specific conditions, the impact of the stress factors can be understood by comparing the results of different ageing procedures with each other. For example, comparing protocols D-1 and D-2 provides insight on the effect of elevated temperature, comparing protocols D-2I, D-2, and D-3 provides insights on the effect of the atmosphere, comparing protocols D-2 and L-2 provides insights on the combined effect of light and heat, and so on. The arrows in Fig. 3 schematically depict these relationships.
The blank spaces in Fig. 2 can be filled with the corresponding ageing procedures (by analogy with the additional ISOS tests we have suggested in this Consensus Statement). Other protocols can be constructed by varying the 'fixed' parameters (for example, temperature, light intensity, RH). We do not aim to cover all possibilities with the ISOS protocols nor discredit studies with a systematic variation of a particular stressor. Nevertheless, these investigations would benefit from having common 'reference points' with other studies conducted at different laboratories, with different device architectures, different perovskite materials and even different research questions. Consensual conditions, like ISOS protocols, may serve as such references.
Unified procedures for stability studies and consistency in data reporting could lead to the creation of a large machine-readable database on PSC stability. Machine learning (ML) methods [145][146][147] could potentially identify patterns in such data, detect statistically significant stress factors, correlate repeated phenomena in different studies to detect universal degradation mechanisms and stabilizing approaches, and predict lifetimes and failure modes. Information from ageing measurements under the relevant stressors can optimize the steps required for supervised learning algorithms. Ultimately, ML relies heavily on having sufficient quantity and diversity of information from ageing tests to provide an accurate prediction of device performance and degradation. Thus, ML algorithms must be trained with extensive laboratory data so that predictions can be compared with actual performance measurements. Once working, ML tools should provide knowledge extraction without the need to do all the tests in Fig. 3.
To accommodate the large number of perovskites possibly suitable for PV, a shared-knowledge repository database has been proposed 145 , where positive and negative results from stability tests are considered equally important for ML (that is, not only the champion cell but also the suboptimal or underperforming cells aged under similar conditions), because they all represent valuable training data. It is therefore critical that a trend emerges that all cells that undergo ageing tests are measured and reported for the duration, even total failures. At the laboratory scale, this information could be shared if researchers report the conditions used for device fabrication and testing through a common website. Progress in this direction might be significantly accelerated if researchers, who foresee their data being useful for ML, provide complete information (on device fabrication, performance and stability) in standardized tabulated format (a possible template is available in the Supplementary Information). A similar strategy can be extended from PSCs to other emerging PV systems.

Conclusions
We have presented our consensus on procedures for studying the stability of perovskite solar cells. The protocols we suggest primarily rely on the original ISOS standards developed for OPV cells 11 , which have proven to be highly relevant for uncovering various degradation pathways in PSCs. We have further extended the protocols with a set of testing procedures in accordance with specific stability features of PSCs, including light-dark cycling (ISOS-LC) mimicking the diurnal cycle, study of solar cell behaviour under continuously applied bias in the dark (ISOS-V) and protocols for studying intrinsic solar cell stability (indexed with 'I'). We have indicated which figures-of-merit for device stability should be used to take into account the evolution of the PSC performance over time. To improve reproducibility, we also propose a checklist for reporting results from PSC stability studies more consistently.
We hope that the guidelines for conducting and reporting stability studies described in this paper will improve comparisons between data from different laboratories and from different device architectures. The set of procedures and practices suggested here serves as an intermediate stage in perovskite solar cell technology maturation, aimed at the identification of degradation pathways and the prospects for their mitigation.

Methods
The work extends the outcomes of the round table discussion on PSC stability assessment that took place during the 11th International Summit on Organic and Hybrid Photovoltaics Stability (ISOS 11) in Suzhou, China, in October 2018 (http:// isos11.csp.escience.cn/dct/page/1) The round table was followed by drafting of the discussed procedures and manuscript text, which were circulated between the contributing authors multiple times until a consensus was reached among the authors. In cases when no single opinion was possible between all the co-authors we extended the number of options presented.