The challenge of designing accelerated indoor tests to predict the outdoor lifetime of perovskite solar cells

21 Over the past decade, perovskite solar cells have travelled an amazing way towards high efficiency. However, a major roadblock remaining is the operational stability, while achieving technological maturity and proving real-world stability is crucial to gain trust among investors. In that sense, it is of high interest to be able to predict the operational lifetime, which needs to be in 25 the range of years or decades, within an experimentally reasonable timeframe. Yet, peculiarities of 26 perovskite solar cells’ ageing behaviour lead to severe difficulties in translating the results of indoor tests to their outdoor counterpart. In particular, transient processes cause diverse results among 28 different ageing tests. Here, for the first time, we show a complete set of constant illumination indoor testing, cycled illumination indoor testing and real-world outdoor testing on equal in-house devices. Exemplarily, 31 we compare two different types of perovskite solar cells, in which only the hole-transport layer is 32 varied. Despite this small change, the devices show distinctly different transient behaviour. In either 33 case, the commonly used constant illumination experiments fail to predict the outdoor behaviour of 34 the cell. Yet, we observe a good correlation between the cycled illumination test and the outdoor 35 behaviour of one of the two solar cells, while this is not the case for the other system. This result 36 highlights the urge for further research on how to perform meaningful accelerated indoor tests to 37 predict the outdoor lifetime of perovskite solar cells.


Introduction
Perovskite solar cells (PSCs) are a rapidly developing photovoltaic technology 1   More recently, the influence of transient behaviour in PSCs was found relevant in studying the 61 long-term stability of devices 10,11 . The origin of that behaviour is the coexistence of several dynamics 62 with characteristic times spanning from timescales of seconds to hours 12  To achieve the prediction of a realistic outdoor lifetime of PSCs with an accelerated indoor test, 79 it is essential to understand the temporary changes in device power output due to weather conditions 80 and the state of the device. This section will discuss the shortcomings of existing PSCs' outdoor data  the hole-transport-layer (HTL) was varied. As one option for the HTL, NiO was used, which is the 107 "standard" for stable p-i-n devices 20 , while as other HTL the newly developed self-assembled-108 monolayer (SAM) "MeO-2PACz" 21 ([2-(3,6-dimethoxy-9H-carbazol-9-yl)ethyl]phosphonic acid) 109 was used. A variant of this molecule, the "MeO-4PACz", recently gained attention since it enabled a 110 record perovskite-silicon monolithic tandem solar cell 1,22 .

111
As with other PV technologies, PSCs' power output depends on irradiance, temperature, solar 112 light spectrum, and incidence angle. As can be seen from Figure 1 a) and c), the power output of the 113 solar cells resembles the measured irradiance, which is due to the linear dependence between 114 irradiance and MPP current density (JMPP), see Figure S3. Yet, the JMPP-to-irradiance ratio is not a 115 straight line as shown in Figure 1b. Some deviations from this linearity are caused by changes in the 116 incident sunlight spectrum. They coincide with changes in the average photon energy (APE) of the 117 incident spectrum, which is a technology-agnostic figure of merit of the spectrum (see Eq. S1 for the 118 definition) 23,24 . In essence, a blueshift (higher average photon energy) in the incident spectrum leads 119 to a marginally improved device current, while a redshift (lower energy) leads to a marginal decrease. 120 Figure 1: Impact of real-world stresses on the PSC's power output: One day of outdoor exposure of PSCs with organic self-assembled monolayers (SAM) and NiO hole transport layers. The respective power output c) measured via MPP tracking closely follows the measured irradiance a). The average photon energy (APE) in b), is calculated from spectral data measured in the plane of samples and plotted next to the ratio of Jmpp to irradiance (black dots in b)) to highlight the effect of the sun spectrum.
With outdoor data available, a logical next step to take -also for investors-would be energy 121 yield predictions and analyses 25-27 . However, we advise caution here, since the extraction of 122 parameters from outdoor data might lead to a misinterpretation in case of PSCs. For example, E. PSCs' temperature coefficients being positive in some cases when derived from outdoor data 25 , 125 meaning that device efficiency improves with temperature. While a positive temperature coefficient 126 may correlate to annealing effects or ion redistribution, it might also be entirely misleading due to 127 unaccounted transient behaviour overlapping with temperature effects in real-world conditions. Table   128 S3 summarises reported temperature coefficients obtained from either indoor or outdoor data. When  Relatively high series resistances contribute to this behaviour, possibly explaining a better 164 performance at low irradiance in the evening (see Figure S4). However, since evening values are still 165 noticeably below morning ones, we assume the impact of reversible degradation. On the contrary, 166 PSCs with SAM require several hours (strongly depending on the day of observation) to reach their 167 peak efficiency.

168
It is not always easy to judge such transient behaviour from indoor experiments, partly due to 169 frequent practice of removing the initial "stabilisation" phase from the reported PSC ageing curves. 170 We strongly recommend providing full datasets, because under the natural day-night cycle in outdoor 171 conditions, initial transient processes might have a significant contribution to the observed behaviour.  illumination test. This challenge will be discussed in detail in the next section.

183
As a first step towards predicting outdoor lifetime with indoor testing, we present a comparative 184 study with indoor and outdoor data. Three different ageing tests, indoor and outdoor, have been 185 performed (see Table 1).

186
In our study, both types of solar cells were tested at first under constant ageing conditions 187 indoors. These were in brief: 100 mW/cm 2 continuous illumination by a solar simulator (Figure S1  When analysing the constant indoor test (Figure 4a), it seems that the overall stability is very 203 similar. However, the tracks' slope is different, and the curves are crossing each other after ~420 hours 204 of testing (see the full test in Figure S8). Consequently, NiO could be considered more stable in this 205 test with a T80 (i.e. the time it takes for PCE to decrease 80% of initial value) of 835 h, while SAM's 206 T80 equals 580 h.

207
The behaviour changes significantly when the illumination is cycled (Figure 4c). Within each  an exponential shape (compare Figure S10). Since the only difference between those two indoor 246 experiments is cycling the light, the different behaviour likely originates in recovery effects 5,7,12 . It 247 seems that pausing the illumination and putting the cells to open-circuit condition harms the device more than constant illumination and MPPT conditions. It appears counterintuitive that giving the cells 249 "a rest" hurts them more than applying constant stress.

250
Interestingly, the indoor cycled PR for SAM-cells does match quite well with the outdoor 251 experiment's ones until day 10. Then the performance drops rapidly over the next days until total 252 failure on day 13. We assume that this is due to the breakdown of the encapsulation: As soon as water 253 or oxygen have passed the encapsulation, the degradation rate is strongly enhanced and also leads to 254 visual changes (see Figure S11) in the device due to perovskite decomposition 37 . Since passing the 255 encapsulation is likely a diffusion process (see Figure S12), an error function can be used to model