In order to investigate the effects of illumination-induced degradation in the perovskite solar cells, we started off investigating the commonly used triple cation Cs0.05(FA0.83MA0.17)0.95Pb(I0.83Br0.17)3 perovskite composition with a bandgap of 1.63 eV in a pin-type architecture (referred to as “83:17 TC” throughout the rest of this manuscript). The following cell architecture was used: ITO/PTAA/PFN-Br/perovskite/C60/BCP/Cu, ITO is indium tin oxide, PTAA is poly[bis(4-phenyl)(2,4,6-trimethylphenyl)amine], PFN-Br is Poly({9,9-bis[30-({N,N-dimethyl}-N-ethylammonium)-propyl]-2,7-fluorene}-alt-2,7-{9,9-di-n-octylfluorene}) dibromide, and BCP is bathocuproine.50 The cells were subjected to three different external stressors: light, heat, and electrical bias, with a focus on light-induced degradation. During the light-induced ageing, devices were exposed to 1 sun illumination at VOC and cooled such that the device temperature remained at 25⁰C. Measurements under (MPP) maximum power point conditions are discussed further below. In Fig. 1a, the stabilized JV curves (wo/ hysteresis) at slow scan speeds (10 mV/s) after different illumination times under VOC are exemplified. Figure 1b shows how the different device parameters change over time with ageing. In these cells, the increasing JSC losses over time dominate the degradation losses although FF losses are also significant, and the VOC decreases slightly as well. Based on our previous works,49 we suspect that the increased current losses observed here might be caused by an increase in the mobile ion density in the perovskite.
In order to quantify the impact of mobile ions on the light-induced performance degradation we performed fast-hysteresis (FH) JV measurements at different points in time during the ageing process. While details are presented in the Experimental methods and in ref.51, we note that the cell is initially hold slightly above the initial VOC (prebias), followed by a reverse and forward sweep with variable frequency or scan speed. The duration of the prebias was 5x longer than the total scan time of the voltage sweep. The methodology allows us to determine the efficiency in steady state and the “ion-freeze” efficiency, which refers to the condition at which the ions are effectively immobilized when the scan rate is much quicker than the diffusion rate.51 The difference between slow (steady-state) and fast-scan speeds can be directly attributed to the movement of mobile ions. In the following, these losses are referred to as mobile ion-induced PCE losses. However, it should be noted that the presence of mobile ions can still affect the PCE even when they are immobilized depending on their distribution and accumulation throughout the device and at the interfaces.51,52 This effect will be small if the ions are homogeneously distributed in the bulk at the start of the scan which is discussed further below. The results, displayed in Fig. 2a, reveal once again that the steady-state efficiency drops significantly with ageing. Conversely, the ion-freeze efficiency, determined at fast scan speeds of around 2500 V/s experiences a much smaller, almost negligible reduction. Next, we investigated the effects on each of the three JV metrics: the VOC, JSC and fill factor (FF), displayed in Fig. 2b-d. It can be seen that the illumination-induced performance losses are largely related to a drastic loss of the steady-state current output with respect to the ion-freeze JSC (even more than 10 mA cm− 2 current loss can be observed at longer aging times and in other systems). However, the VOC and the FF are also affected to some extent. Furthermore, the “peak hysteresis’” i.e. the maximum difference between the PCE determined from the forward versus the reverse scan also increases significantly with aging. Overall, the remarkable difference in the PCE between the fast and slow scan speeds (Fig. 2e and f) shows that the dominant loss mechanism upon ageing acts more significantly at the slower timescales.
To get some understanding of the chemical changes of the perovskite film during light exposure, we performed absorption spectroscopy (Figure S1), X-ray diffraction (XRD) measurements on fresh and aged perovskite layers (Figure S2), scanning electronic microscopy (SEM) and atomic-force microscopy (AFM) measurements (Figure S3). As discussed in the Supporting information, while these measurements indicate possible changes in the morphology, these changes are considered to be relatively small on the timescale of the measurements (up to 24 h of illumination).
To generalize these findings, we then investigated the mobile ion-induced ageing loss in a range of different perovskite compositions beyond the “83:17” model system. These perovskite systems include a standard 1.6 eV methyl ammonium lead iodide MAPbI3 (“MAPI”), a double cation perovskite FA0.85Cs15Pb(I77Br23)3 (“CsFA”) with a bandgap of 1.69 eV relevant for Silicon/perovskite tandems, a wide-bandgap Cs0.05(FA0.60MA0.40)0.95Pb(I0.60Br0.40)3 (“1.8 eV“) relevant for all-perovskite tandems and a high performance (22% PCE) 1.57 eV formanidinum lead iodide rich (“95:5 TH“) triple halide perovskite (Cs0.05(FA0.95MA0.05)0.95Pb(I0.95Br0.05)3 + 20 wt% MACl). The FH results on all systems are shown in Figure S4 – S7. These results demonstrate that the increase of the ionic loss (Lion = PCEfast/PCEslow) contributes significantly to total degradation loss for all systems (Fig. 3a). Moreover, we find that the reduced JSC is the most important factor contributing to the ion-induced loss in all systems (Figure S8 – S9). Figure 3b exemplifies the large reduction of the steady-state JSC after several hours of illumination (5h except for 12h in MAPI) with respect to the initial current. Figure 3c shows that the ionic losses also emerge in the much more stable 95:5 TH perovskite system which degraded to roughly 68% of its initial PCE (21%) after tracking the stabilized power output over 1630 h. The FH measurements taken initially and after the MPP tracking reveal a nearly unchanged ion-freeze PCE, meaning that the observed power output losses can be attributed to ionic losses. We note that other factors, such as shunt formation, moisture ingress or electrode corrosion can cause an even earlier device failure, however we observe that the ionic losses are very prominent if these “extrinsic” issues are mitigated.
Having established that the ionic losses dominate the overall device degradation, we now set out to try to understand the origin of these losses by further investigating the example of the 83:17 triple cation perovskite. At this point, it is important to note that the increase of the ionic loss with aging time could be related to many other factors.51 This is because the magnitude of the ionic loss depends on various device and material parameters for a given ion density. To identify the underlying mechanisms we now discuss, simulate, and analyse distinct scenarios. For the simulations, we used the software IonMonger53, while the results were crossed-checked using the software SETFOS from FLUXiM.54 In our first hypothesis, the concentration of mobile ions increases with ageing, which leads to enhanced ionic losses as shown by the simulations in Fig. 4a (see Figure S10, Figure S11 and Table S1 for the JVs and the corresponding PCE parameters and the simulation parameters, respectively). In this case, the increased mobile ion density causes enhanced screening of the internal field electric field in the absorber layer, which leads to a continuous reduction in charge extraction efficiency due to band flattening (Figure S12). The details of this loss process are discussed below in more detail. Another possible explanation for the enhanced ionic losses are degrading perovskite/transport layer interfaces with higher recombination velocities (S). As shown by the scan-rate dependent PCE in Fig. 4b, the increased recombination velocity leads to a reduced steady-state PCE and increased peak hysteresis. However, in contrast to the case of an increased ion density (Fig. 4a), the ion-freeze PCE at fast scan speeds is decreased as well. In addition, we also investigated the impact of an increased energy level offset on the FH results, as well as the impact of more bulk defects. However, we found that both (i.e. increased energetic offset and reduced bulk lifetime) lead to a parallel downshift of the scan-rate dependent PCE rather than an increase in the ionic losses, which is not consistent with the experimental results (Figure S13).
To further investigate the impact of potentially worsening interfaces or absorber layer quality on the enhanced ionic losses, photoluminescence (PL) measurements of different partial cell stacks were recorded as a function of illumination time. From the PL yield the quasi-Fermi level splitting (QFLS) was calculated following our previous methodology.50,55 Surprisingly, the results displayed in Fig. 4c, suggest that the interfaces do not significantly worsen as the QFLS of all partial device stacks increases slightly over time, while the QFLS of the neat perovskite layer is nearly unchanged. Yet, the device VOC decreases resulting in an increasing QFLS-eVOC mismatch55 upon ageing. While increased interfacial defects (leading to an enhanced S) are not consistent with the increased QFLS and can be likely ruled out as a dominant factor of the light-induced degradation in this cell, the increased QFLS-VOC mismatch is intriguing. Although this requires further investigation, the QFLS-VOC mismatch provides clues about the effect of ions at fast scan speeds. The FH simulations displayed in Figure S14 show that if the built-in field is not offset with the prebias of the FH measurement (i.e. Vpre < VBI), the VOC decreases with increasing ion density. Considering that such a VOC decrease (and QFLS-eVOC mismatch) is observed in various systems, the most likely scenario is that ions accumulate in an unfavourable position under open-circuit conditions (e.g. cations at the hole selective interface).52 This is discussed further below.
Having ruled out an increased recombination velocity at the interfaces, an increased mobile-ion density seems likely to be the root-cause of the observed ageing-induced losses. In order to directly assess whether the mobile ion density increases with degradation, we performed bias assisted charge extraction (BACE) measurements and charge extraction by linearly increasing voltage (CELIV) at the same points in time when we also performed fast hysteresis measurements. Both transient charge extraction techniques can be used to estimate the mobile ion density by integrating the external current. In BACE, the cell is held under “quasi” open-circuit conditions, where the injection current equals the short-circuit current density at which the mobile ions are distributed throughout the absorber layer before the voltage is switched to 0 V, at which point the mobile ions drift to the contact layers. However, we note that if the ionic charge is larger than the electrode charge (Qion < CVBI), the formation of zero field regions should limit the displacement of ions in excess of CVBI. Thus, the CVBI is expected to be a natural limit for the externally detectable ionic charge. Moreover, the ratio of the drift length of charges versus the film thickness (\({d}_{\text{d}\text{r}\text{i}\text{f}\text{t}}/d\)) limits the externally integrated charge. Therefore, the ions should be roughly homogenously distributed under the prebias condition to maximize the drift length and to avoid an underestimation of the ionic density.56
The results of the BACE measurements are displayed in Fig. 5a and b. It can be seen that the externally measured current is greatly enhanced the longer the illumination. Notably, the ionic “time of flight” in the fresh device at around 50 ms (Figure S15) matches nearly perfectly the scan time at the peak hysteresis observed in averaged FH on fresh devices (~50 ms). Due to the associated diffusion constant (\(D\) = 7x10−10 cm/s) and ionic mobility (\(\mu\) = 2x10−8 cm2/Vs), these results indicate that the responsible ion species are consistent with halide vacancies.42,57 Therefore, these measurements link the transient charge extraction with the scan rate dependent JV results. It is also interesting to note that with prolonged ageing time, more slower species extracted which could point to A-site vacancies.42,57 Notably, this could also explain why the peak hysteresis in the FH measurements does not shift to faster scan rates with increasing ion densities as predicted by the simulations for a given ion diffusion coefficient (Fig. 4a). As shown in Fig. 5b, by integrating the external current, a rough estimate for the ion density can be obtained. By plotting the ionic losses obtained through FH as a function of the ion density in Fig. 5c, we observe a linear dependence after an initial dwell time. It can be seen that the ion density increases significantly, even within the first hour of device ageing and reaches values over 1018 cm− 3 after 20 h of ageing. Although the reason why such large densities can be obtained in BACE are not yet clear considering the above discussed limit of the detectable ionic density which requires further investigation, the obtained ion densities make sense qualitatively. As shown in Fig. 4a, the simulations can well reproduce the observed losses at the given ion densities. To highlight the effect, an ion density of 5x1018 cm− 3 approximately halves the initial steady-state PCE due to enhanced ionic losses. We also note that we do not exclude some collection of ions from the side of the pixel as recently demonstrated, which could increase the ion density within the active device area.44 Finally, with CELIV a similar picture is obtained, where the increased mobile ion density after ageing is reflected in a strong increase in the signal on slower timescales, starting at roughly 1x10− 4 s (Figure S16). Finally, we note that an increase of the ion density with increasing ageing time was also confirmed for the other studied systems (Figure S17), consistent with the results on the 83:17 TC cells.
In order to investigate whether the devices would age similarly under real-world conditions and to better understand the dominant factor triggering the degradation, we expanded our measurements to other ageing conditions. The FH results for degradation at elevated temperatures (75°C) in the dark (Figure S18), under MPP conditions at 1 sun illumination (Figure S19), and under electrical bias (VOC) in the dark (Figure S20) can be seen in the SI. From these results, it becomes clear that the ion induced performance loss is a general degradation mechanism, that is also triggered by electrical bias. Nevertheless, the aging at elevated temperatures does not show the characteristic step-function like scan-rate dependent JSC (Figure S18), nor displays a very large increase of the ion density on these timescales (Figure S21). This might point towards a different path of degradation or a slower evolution of the ionic losses. While determining the exact rates of degradation requires a much more focused investigation that goes beyond the scope of the present work, our measurements indicate that mobile ions are easily created under different stressors with different rates, in particular in the presence of electrical charges in the active layer. The fact that the increase of ionic losses behaves qualitatively similar under VOC, regardless of the presence of light, but differs at elevated temperatures in the dark (and 0 V) is consistent with the interpretation that halide vacancies are generated by injected or photo-generated free holes.45,58 In particular, it has been previously shown that iodine atoms can be oxidized and kicked out of the lattice by photogenerated holes, creating halide vacancies via \({I}_{\text{I}}^{x}+{h}^{+}= {I}_{i}^{x}+{V}_{\text{I}}^{+}\), where \(\text{I}\) and \(i\) refer to the regular and the interstitial lattice site, respectively and \(x\) to the neutral charge.36,45,58 Another possible ion source is unreacted lead halide PbI2 (Figure S2), which can decompose into metallic lead and mobile iodine under illumination.59 Both of these mechanisms are consistent with additional X-ray photoelectron spectroscopy (XPS) measurements which reveal the presence of iodine on top of the C60 layer after illuminating the triple cation perovskite for approximately 10 h, demonstrating the possibility of iodine diffusion through fullerenes in line with previous reports (Figure S22).24, 60–62 Moreover, we could detect iodine the perovskite film on a Si substrate that was placed ~1 mm above the sample (in N2) during illumination, hence highlighting the egress of volatile iodine species (I2) (Figure S23).
To recap, the mobile-ion induced efficiency losses observed in the FH measurements dominate the early degradation loss in the here tested halide perovskite compositions and based on various measurements such as PL and BACE, we attribute these enhanced ionic losses to an increasing ion density upon aging. Figure 6a shows a breakdown of the identified degradation losses under illumination for the 83:17 triple cation system. The JSC and FF loss due to the movement of mobile ions is attributed to the field screening effect, the loss due to the QFLS-VOC mismatch which is likely also a mobile-ion-induced loss, and other, yet unspecified loss processes. Figure 6b-d schematically illustrates a possible explanation of the observed JSC and FF losses and the QFLS-VOC mismatch. The increased ion density during the ageing successively screens the internal (built-in) field, thus increasing the fraction of the active layer which is effectively under VOC (or flat band) conditions. The formation of the zero-field region under short-circuit condition leads to an increased accumulation of electronic charges (\({e}^{-}\) and \({h}^{+}\)) at the hole selective interface. This charge accumulation leads in turn to enhanced recombination at the hole transport layer and in the bulk. We note that the reason why the p-interface is more affected than the electron-selective interface is that we consider positively charged mobile ions (halide vacancies) which screen the field at the hole transport layer. Likewise, the electron-selective interface would be stronger affected in case of a dominant negative mobile ion species. To explain the observed QFLS-VOC mismatch with increased ion densities, we analyse the band diagrams at VOC conditions displayed in Figure S24 for devices with different ion densities. The graph shows that higher ion densities cause a larger electron QFL bending (\(\nabla {E}_{\text{f},\text{e}}\)) at the C60 interface. The bending of the electron QFL can be explained by a population inversion of electronic charges at the electron selective interface, i.e. enhanced hole accumulation and depletion of electrons, in case of an enhanced cation vacancy concentration (Figure S25). Considering that at VOC, \(\nabla {E}_{\text{f},\text{e}}\)depends on the electron and hole conductivities and the gradient of the hole QFL, i.e. \(\nabla {E}_{\text{f},\text{e}}=\frac{{\sigma }_{h}}{{\sigma }_{e}}\cdot \nabla {E}_{\text{f},\text{h}}\),63–65 allows us to conclude that the mobile ions are likely also responsible for the observed QFLS-VOC mismatch and the reduction of the VOC. Finally, while there might be some small bulk recombination losses at short-circuit conditions depending on the diffusion length in the absorber layer as shown in Fig. 6d, the recombination at the interfaces increases more rapidly with enhanced cation vacancies at 0 V. Moreover, the interfacial recombination current scales faster with the applied forward bias voltage than the bulk losses, thus affecting the FF more significantly (Fig. 6d).