Improving the eﬃciency and stability of perovskite solar cells using p -conjugated aromatic additives with diﬀering hydrophobicities †

,


Introduction
Organic-inorganic metal hybrid perovskites are regarded as highly promising photovoltaic materials due to their excellent optoelectronic performance and ability to enable low-cost solution processable technologies. [1][2][3][4] Their advantageous optoelectronic properties include high absorption lengths, low exciton binding energies and high carrier mobility. [5][6][7] Perovskite solar cells (PSCs) have emerged as one of the most promising 3rd generation solar cell candidates, not only because of their excellent power conversion efficiency (PCE) of 25.7% 8,9 but also their scalable preparation. 10,11 Despite extensive research and development efforts, PSCs are still faced with relatively poor stability. [12][13][14][15] Additive engineering has proven to be successful in improving both the PCE and stability of PSCs, 16 which is due in large part to the passivation of defects 17 and decreasing the migration of mobile ions. 18,19 Indeed, p-conjugated aromatic additives (CAAs) offer a range of potential advantages for PSCs because their delocalized electrons enable charge to easily pass at perovskite interfaces and their planarity favors the alignment of neighboring molecules at interfaces. 20 They can also passivate Lewis acids (such as Pb) by donating electrons [21][22][23] and act as interfacial compatibilizers between the perovskite and the hole transport matrix (HTM). 20 Whilst many studies have used CAAs in PSCs, 20 very few have included them in the precursor perovskite solution. Wei et al. included 3-phenyl-2-propen-1-amine iodide (PPEAI) as an antisolvent additive and reported that this small-molecule CAA provided conductive channels between neighboring perovskite grains. 24 Wu et al. used 2-amidinopyrimidine in a similar manner to increase PSC efficiency. 12 Several studies have used CAAs as interfacial modifiers between the perovskite and HTM. 25,26 In other cases, CAAs have been used to modify the interface between the perovskite and electron transport matrix (ETM). 22,27 In such cases the CAAs were very hydrophobic and not likely to be soluble in the polar aprotic solvents used for perovskite precursor solutions, such as DMF and DMSO. In contrast, the present study uses two coumarins as CAAs that are able to be included in the perovskite precursor solution. We use two structurally similar CAAs with differing hydrophobicities to elucidate the role of this parameter and the structural differences on PSC performance and stability. This approach has the advantages of providing a simpler, potentially more scalable, construction method that enables the additives to distribute throughout the perovskite layer as well as the bottom and top perovskite interfaces.
Coumarins belong to the benzopyrone family and contain a benzene ring fused with a pyrone ring. 28 Here, coumarin methacrylate (CMA) and coumarin hydroxyethyl (CHE) additives are investigated for planar films and PSCs ( Fig. 1a and b). We hypothesized that these CAAs would allow charge to pass more easily at interfaces. 22,29 Moreover, by introducing the methacrylate group via CMA the hydrophobicity is increased. CMA contains carbonyl (CQO) functional groups at two positions ( Fig. 1b) and such functional groups are well known to bind to Pb and passivate perovskites. 30 The other single bonded O atoms have a lone pair of electrons, which can bond with the under-coordinated Pb, in principle. In contrast, CHE contains a terminal hydroxyl (OH) group and only one CQO group (Fig. 1b) and was expected to be less hydrophobic than CMA.
There have only been two reports of coumarins used in PSCs to the best of our knowledge. Chen et al. used the coumarin dye 8GFF (Disperse Yellow 82) as an additive for PSCs and reported a best PCE of 19.16%. 31 Liu et al. included Coumarin 343 dye in their devices and achieved a champion PCE of 20.9%. 32 8GFF and Coumarin 343 have UV-visible maximum absorption values (l max ) at wavelengths of 430 33 and 446 nm, 34 respectively. In contrast, the coumarins selected for this investigation absorb in the UV region of the spectrum where photon-to-current conversion by the perovskite is negligible. This enables CHE and CMA to be used at much higher concentrations than Coumarin 343 32 without significantly contributing to parasitic absorption.
In this study we show that CMA is more hydrophobic than CHE and that this difference, together with relative strengths of Fig. 1 (a) Depiction of the method used to prepare the perovskite films. CBZ is chlorobenzene. (b) Structures of CMA and CHE. (c) UV-visible spectra of water or CBZ solutions containing CMA or CHE. Each additive was placed in CBZ/water mixtures, shaken thoroughly and the solvents allowed to phase separate prior to analysis of each layer. The calculated CBZ/water partition coefficients (P CBZ/water ) are shown. (d) Water contact angle results for perovskite films prepared using CMA or CHE. The concentrations of the latter (wt%) are shown. (e) Water contact angles from (d).
bonding interactions, strongly affect the optoelectronic and device properties of the perovskite films and PSCs. Both coumarins increase the PCE of the PSCs. The champion CMA modified device achieves a high open circuit voltage of 1.15 V and a PCE of 21.28%. The champion CHE modified device had a PCE of 21.14%. In contrast, the best additive-free device had a PCE of 19.15%. Inclusion of the CMA or CHE into the PSC strongly increased device stability, with CMA providing the greatest stability. The CMA-based device has a very high shelflife stability at room temperature with 98 days being required to reach 80% of the initial PCE (t 80 ). In contrast the t 80 values for PSCs containing CHE or no additive are 38 days and 14 days, respectively. The results of this study show that increasing the hydrophobicity of a precursor-soluble CAA such as a coumarin via substitution is beneficial for both PSC efficiency and stability.

Measurements on hydrophobic properties
A solvent partition measurement was developed to demonstrate the difference in the hydrophobicities of CMA and CHE. Accordingly, CMA or CHE were added to a solvent mixture containing equal volumes of chlorobenzene (CBZ) and water. Each mixture was shaken and then the CBZ and water phases were allowed to separate. The concentration of each additive in the CBZ and water phase was measured using UV-visible spectroscopy for each phase (Fig. 1c) as well as spectra measured (Fig. S1, ESI †) for calibration curve construction (Fig. S2, ESI †). The ratio of the additive concentrations in CBZ to that in water is the CBZ/water partition coefficient (P CBZ/water ). The P CBZ/water values for CMA and CHE are 46.6 and 5.58 (Fig. 1c), respectively. These data demonstrate that CMA is more hydrophobic than CHE. This trend is confirmed by the calculated logarithms of the octanol/water partition coefficients for CMA and CHE which are 3.08 and 1.54, respectively. 35 Fig. 1c also shows that CMA and CHE have l max values of 320 nm, which implies that these CAAs will not contribute significantly to parasitic absorption in PSCs.
Perovskite films were prepared by adding CMA or CHE to the precursor solution prior to spin coating as depicted in Fig. 1a. Fig. 1d and e show that the water contact angle on the perovskite surfaces increases from 551 to 721 when the CMA concentration increases from 0 to 1.0%. This trend indicates that CMA increases the perovskite surface hydrophobicity. In contrast, the contact angle decreases from 551 to 491 when the CHE concentration in the precursor solutions increases from 0 to 1.0%. In the latter case, the perovskite surface becomes more hydrophilic. Furthermore, CMA has a methacrylate group (Fig. 1b) which is polymerizable in principle. An 1 H NMR spectroscopy investigation (Fig. S3, ESI †) showed that CMA did not undergo polymerization and behaved as a single molecule additive under the conditions used in this study (see discussion and text in the ESI †). The 13 C NMR spectra for CMA and CHE are also shown in Fig. S4 (ESI †).

Device performance and stability
Planar n-i-p PSCs were constructed (Fig. 2a) to investigate the effect of introducing CMA or CHE in the precursor solution on performance. The J-V curves for the champion devices are shown in Fig. 2b. We investigated additive concentrations of 0.3, 0.5 and 1.0%. The champion CMA-and CHE-based devices were both prepared using 0.3% of additive and they had a PCE of 21.28% and 21.14%, respectively. In contrast, the champion control device PCE was 19.15% (The photovoltaic parameters measured are shown in Table S1, ESI †). The best PCE reported in this study is close to the average of the best PCE values reported for PSCs prepared using similar perovskite compositions as shown in Table S2 (ESI †). Our devices are not designed for maximizing the PCE, but to evaluate the effects of two structurally related CAAs on efficiency and stability. The stabilized power outputs and the short-circuit current density ( J sc ) data ( Fig. 2c) agree with those obtained from J-V measurements ( J-V curves and stabilized power outputs are shown for all of the devices in Fig. S5a and b of the ESI †). EQE spectra were measured for the devices (Fig. S5c, ESI †) and the integrated J sc values calculated from the respective EQE data agree with the J sc values from the J-V curves to within 10%. We calculated the average EQE values from the spectra (Fig. S5d, ESI †) which show maximum values for the devices containing 0.3% CMA and CHE. Furthermore, the average EQE values are highest for the CMA-based devices. Box plots for the PCE data for all of the devices are shown in Fig. 2d (Box plots for V oc , J sc and FF are shown in Fig. S6, ESI †). Overall, these data confirm that 0.3% is the optimum concentration for both CMA and CHE and that both coumarins improve the PCE.
Whilst the use of 0.3% CMA and CHE additives gives improved PCE, V oc , FF and J sc values compared to the 0% system ( Fig. 2d and Fig. S6, ESI †), the differences between the performance of the devices prepared using the two additives is difficult to distinguish at a glance when only the reverse sweep data are considered (Fig. 2d). Therefore, we show the average PCE values that include the forward and reverse data in Fig. 2e. Fig. S7 (ESI †) shows the values for V oc , FF and J sc calculated using both forward and reverse sweep data. These data show that the average PCE, V oc and J sc values are larger for the CMA devices compared to the CHE devices. This is due in part to the much higher hysteresis index (HI) values for the CHE devices as shown in Fig. 2f. The HI values are 4.0% for the 0.3% CHE device and 6.3% for the 0% control device. Whereas, the HI is only 2.0% for the 0.3% CMA device. Furthermore, the HI is a minimum of 1.6% for the 0.5% CMA device. Hence, a pronounced decrease of the mobile ion concentration occurs for PSCs containing low CMA concentrations. 36,37 We investigated the shelf-life stability of the highest performing systems (0.3% CMA and 0.3% CHE) as well as the control (see Fig. 2g). The device stability is characterized in terms of the time required for the PCE to decrease to 80% of the initial PCE (i.e., t 80 ). The values of t 80 for the 0%, 0.3% CHE and 0.3% CMA devices are 14, 38 and 98 days, respectively, indicating that both coumarins increased device stability. Notably, CMA was by far the most effective at increasing device stability. The thermal stability for these devices at 85 1C was also investigated (see Fig. S8, ESI †). The t 80 values for the 0%, 0.3% CHE and 0.3% CMA devices at 85 1C are B5, 15 and much greater than 30 days, respectively. These results demonstrate improved thermal stability using the coumarins and, especially, CMA. We also investigated the room temperature moisture stability of the perovskite films at 90% RH. Images for the films are shown in Fig. 2h and UV-visible spectra recorded for these films are shown in Fig. S9, ESI. † Both coumarins improved the moisture stability compared to the control and the CMA film had the best moisture stability. The stability of the devices was also studied using 100 mW cm À2 illumination in laboratory air without encapsulation. The data (Fig. S10, ESI †) show that the stability of the devices decreased in the order 0.3% CMA 4 0.3% CHE 4 0% (control). Indeed, the degradation rate for the 0.3% CMA device was less than half of that of the control (Fig. S10b, ESI †). Top-view SEM images for the films are shown in Fig. S11a-g (ESI †). There is a slight increase of the average grain size from 395 nm (for the additive-free, 0% film) to 417 and 435 nm, for the 0.3% CHE and 0.3% CMA films, respectively. The grain size subsequently decreases for 0.5% and 1.0% concentrations of both CMA and CHE (Fig. S12, ESI †). The grain size distributions are shown in Fig. S13 of the ESI. † Generally, the widths of the size distributions are smaller for the CMA samples than those containing CHE which may indicate that CMA was more effective in producing uniform grain growth. The perovskite grain size increases with decreasing nucleation density. 38 Therefore, the use of 0.3% CMA decreases nucleation density. The crosssectional images of PSCs prepared with different concentrations of CMA/CHE are shown in Fig. S14a-g (ESI †). The perovskite film thickness increases slightly from 490 nm for the 0% control to 503 nm and 515 nm, respectively, for the 0.3% CHE-and CMAbased systems (Fig. S15, ESI †), before decreasing again for the 1.0% systems. The perovskite films prepared using 0.3% CMA or CHE have well-defined single grains in the vertical direction that extend from the HTM to the ETM. These large grains are more prevalent for the CMA films and benefit charge transport. The perovskite layers prepared using 1.0% of CMA or CHE have noticeably smaller grains in the vertical direction and, hence, more grain boundaries. This would be expected to increase defect density and decrease device PCE as shown in Fig. 2d.  UV-visible absorption spectra of perovskite films with different concentrations of CMA or CHE additives are shown in Fig. S16a (ESI †). The absorbance measured at 733 nm increases with additive concentration (Fig. S16b, ESI †) and is highest for the 0.3% CMA system before decreasing at higher CMA concentrations. A similar trend is apparent for the CHE systems. However, the absorbance values are higher for the CMA films compared to the CHE-based systems. These trends follow from the film thickness values (Fig. S15, ESI †) and grain size (Fig. S12, ESI †). Hence, 0.3% CMA most effectively increased light harvesting due to thicker films with larger grains.
To investigate potential passivation of the perovskite by the coumarins, steady state PL spectra were recorded for the films (see Fig. 3a and Fig. S17a, ESI †). The PL intensity reached a maximum value (Fig. S17b, ESI †) and the wavelength at the maximum PL intensity (l max ) was a minimum value (Fig. S17c, ESI †) for the 0.3% CMA and 0.3% CHE films. Both of these trends indicate passivation of defects 39 occurred with CMA and CHE. As explained by Salado et al. 39 a passivating additive can decrease the shallow trap density at the grain surface which causes a blue shift. Notably, the PL maximum was highest for the 0.3% CMA system. The lifetime of the charge carriers was probed using time-resolved PL (TRPL) spectroscopy. Data are shown in Fig. 3b and Fig. S17d (ESI †). The equation used to fit the TRPL data is 40,41 where A i is the fractional amplitude for each exponential term. 42 The extracted fitting parameters obtained are listed in Table S3 (ESI †). The parameter t 1 is due to fast charge carrier quenching by Shockley-Read-Hall (SRH) recombination. 43,44 The decay time t 2 is the slow decay process from defect-assisted charge recombination at the surface and/or grain boundaries. 45 The average carrier lifetime (t avg ) is calculated using: 46 The variation of t avg with CMA or CHE concentration is shown in Fig. 3c. For each series, the maximum t avg values occur for a coumarin concentration of 0.3%. Notably, the t avg value for the CMA system (257 ns) is much higher than that for the 0.3%  CHE system (182 ns) and both values are higher than that measured for the 0% control (127 ns). The results concur with the steady state PL data (Fig. S17b, ESI †) and confirm that CMA most effectively inhibits non-radiative recombination by passivating defects in the perovskite. Furthermore, the steady-state PL and TRPL data ( Fig. 3c and Fig. S17b, ESI †) agree well with the V oc trends for the respective devices (Fig. S7a, ESI †). We indirectly assessed the ability of the CAAs to decrease the defect density at the buried interface by measuring steady-state PL spectra and TRPL data for perovskite films deposited on glass using light incident on the glass side (see Fig. S18, ESI †). The maximum PL intensity and also the values for t avg decreased in the order 0.3% CMA 4 0.3% CHE 4 0%. These data suggest that both additives also decrease the defect density at the buried perovskite interface.
To further explore the charge recombination mechanism light intensity-dependent measurements for V oc and J sc were performed for the 0%, 0.3% CMA and 0.3% CHE devices. The light intensity dependence of V oc (Fig. 3d) was analyzed using: 47 where n id is the ideality factor, k B is the Boltzmann constant, T is the temperature, q is the elementary charge and I is the normalized light intensity. The 0% device has a n id of 1.90, which decreases to 1.44 and 1.65 for the 0.3% CMA and 0.3% CHE devices, respectively. This indicates that the coumarins decrease the amount of trap-assisted recombination 48 and that this decrease is most pronounced for CMA. The power-law dependence of J sc on light intensity 49 (i.e., J sc B I a ) is shown in Fig. 3e, where a is the light intensity exponent. Gradients close to 1.0 indicate that bimolecular charge recombination and space charge effects are negligible. 50 The a values for the 0.3% CMA and 0.3% CHE devices increased to 0.94 and 0.91, respectively, compared to the value of 0.88 for the 0% control device. Hence, both the coumarins decrease biomolecular recombination.
To quantify the trap density within the films space-chargelimited current (SCLC) analysis was performed using electrononly (ITO/SnO 2 /perovskite/PC 61 BM/Au) devices. The SCLC curves (Fig. 3f) consist of the ohmic contact region (with voltage exponent, n = 1), Child's law region (n = 2), and trap-filled limit region (n 4 3). 51 Fig. 3f shows the dark current-voltage characteristics of representative devices and the trap-filled limit voltage (V TFL ) values. The trap density (N t ) is calculated using 52 where e and e 0 are the relative dielectric constant of perovskite (which is 46.9 53 ) and the vacuum dielectric constant, respectively. 54 The parameter L is the perovskite film thickness and these values were taken from the data shown in Fig. S15 (ESI †). Accordingly, the trap densities are 2.1 Â 10 16 , 9.8 Â 10 15 , and 6.4 Â 10 15 cm À3 for the 0%, 0.3% CHE, and 0.3% CMA films, respectively. These data confirm that CMA and CHE decrease defect density and enhance passivation. Furthermore, the trap density is most effectively decreased by CMA.
The electron mobility (m e ) is calculated from the n = 2 region of the data shown in Fig. 3f A m e value of 5.1 Â 10 À3 cm 2 V À1 s À1 was calculated for the 0% film using the gradient of the J 1/2 /V vs. V plot (Fig. 3g). The m e values for the 0.3% CHE and 0.3% CMA films calculated in the same manner are 8.0 Â 10 À3 and 9.9 Â 10 À3 cm 2 V À1 s À1 , respectively. Both values are higher than that measured for the 0% control. Indeed, the value for CMA is almost double the pure perovskite m e value. An increase in mobility indicates improved charge transport and less recombination. 56 To further probe the effects of the coumarins on perovskite energy levels UPS data were measured for the 0%, 0.3% CHE and 0.3% CMA films (Fig. 3h). Tauc plots for the films (Fig. S19, ESI †) provided the band gaps that were used for the calculations of the energy levels. Fig. 3i shows the inclusion of 0.3% CMA leads to the upward shift of the Fermi level (E F ) and the device becoming increasingly n-type. An increase of the valence band maximum (E vbm ) for the 0.3% CMA system also occurs indicating better energy level alignment between the perovskite and Spiro-OMeTAD, which should improve carrier extraction. Well-matched energy levels may reduce the energy offset for hole transfer, 57 which in turn, can increase V oc . 58,59 and is consistent with the V oc values shown in Fig. S6 and S7 of the ESI. † However, it is noted that such an effect is not always observed. 60 Furthermore, there is no correlation between the photocurrents for the three systems (Table S1, ESI †) and the valence band offsets between perovskite and Spiro. It follows that the energy level changes that are apparent in Fig. 3i had no significant effect on charge transfer between the perovskite/Spiro interface. This finding is consistent with the work of Belisle et al. 60

Atomistic insights into coumarin-perovskite interactions
The results above consistently show that both coumarins passivate the perovskite, which implies strong interactions between each of the coumarins and the perovskite. To gain atomistic insights into such coumarin-perovskite interactions we employed DFT calculations (see the Experimental section for computational methods). Since the precise molecular structures of these CMA and CHE coumarins are not available, we first examined different conformations for each of these additives. The lowest energy structures for CMA and CHE were determined and are shown in Fig. 4a and b, respectively. The most energetically stable CMA conformation is planar while the CHE conformation is slightly bent. More details regarding CMA are provided in Fig. S20 and Table S4 (ESI †). We found that CMA has a higher calculated dipole moment (6.9 D) compared to CHE (6.3 D).
Next, we modelled the interaction of the most favourable CMA and CHE structures on the perovskite surface. We considered five different configurations for the adsorbed CMA or CHE on two different terminations of the (001) perovskite surface: one with PbI/Br (Pb/X) termination and the second with FACs/I/Br (A/X) termination. (X represents I and/or Br; whereas, A corresponds to FA and/or Cs.) We focused on the (001) surface as it has been shown to be one of the most stable and studied perovskite halide surfaces. 61 The interface models used in this study are shown in Fig. S21 and S22 of the ESI, † and the calculated relative energies of the configurations with respect to the most favourable configuration on each surface are shown in Table S5 (ESI †).
The results show that the arrangement of CMA on the Pb/X terminated surface with the lowest energy, and hence the most favorable, is Configuration 2. Overall analysis indicates that Configuration 2 is the most favored arrangement of these CMA and CHE coumarin additives on the perovskite films and is shown in Fig. 4c and d. This result can be rationalised by the large contact area that Configuration 2 confers when CMA or CHE are in this partly planar arrangement on the perovskite surface compared to the other configurations. This degree of contact allows greater favourable interactions between the atomic species of the coumarins and those of the perovskite surfaces. We mention that the shape of the CMA additive is no longer fully planar after its adsorption on the surface due to the local strain imposed by the bonding interactions between the atomic species of the coumarin and the perovskite.
We have also computed and analyzed the charge density difference between the most stable configuration of the coumarins and the perovskite surfaces, as shown in Fig. 4. On the Pb/X terminated surface, the charge density distribution is mainly found between the oxygen atoms of the coumarins and Pb atoms of the perovskite that are close to each other ( Fig. 4e and f), indicating bond formation between these species. Indeed, the Pb-O bond distances are found to be 2.47 Å for CMA and 2.51/ 2.86 Å for CHE. On the A/X terminated surface, the charge density difference shows that the coumarins essentially interact with the A-site organic cations including weak hydrogen bonding (Fig. 4g  and h). The calculated binding energies for CMA/CHE on the Pb/ X and A/X terminated surfaces are found to be À1.89/À1.84 eV and À1.23/À0.88 eV, respectively, indicating that the coumarins bind most strongly to the Pb/X inorganic cage. The data also show that charge transfer occurs most strongly from CMA to the perovskite, which correlates with the change in the Fermi level observed experimentally (Fig. 3i). Therefore, these findings show that the preferential planar arrangement of CMA and CHE additives can indeed passivate the perovskite surface via strong Pb/X surface-coumarin interactions including Pb-O bond formation, which minimize the defect density at the surface. In addition, both CMA and CHE will likely prevent the interaction of the perovskite with air and moisture, which will confer a better resistance to the ambient environment, enhancing the film stability.

View Article Online
X-Ray photoelectron spectroscopy (XPS) characterization was used to investigate the interactions between the coumarins and perovskite. Fig. 5a shows that the under-coordinated Pb 0 signal that is evident for the 0% system is either absent or mostly absent for the 0.3% CHE and 0.3% CMA systems. (A small fraction could be buried under the tails in the peaks for CHE and CMA.) Furthermore, after CMA inclusion the peaks for Pb 4f 7/2 and Pb 4f 5/2 at 138.2 and 143.1 eV shift to lower binding energies by 0.36 eV. In the case of CHE, the shift also occurred, but was only by 0.11 eV. These results strongly imply: (1) binding of both CME and CHE to Pb and (2) a stronger binding of CMA with Pb compared to CHE. These results further support the DFT modelling results on coumarin-perovskite interactions in Fig. 4 and show that both coumarins bind to Pb. Interestingly, the peaks for I 3d 3/2 and 3d 5/2 (Fig. 5b) shifted to lower binding energy by B0.6 eV for CMA. However, no such shift occurred for CHE. This may indicate better coordination of the iodine ions by CMA compared to CHE. Taken together, these data show that CMA can better bind to the perovskite.
An interesting question concerns the distribution of the CAAs within the perovskite films. We addressed this question in two ways. Firstly, we used high angle annular dark field (HAADF) scanning transmission electron microscopy (STEM) imaging with energy dispersive X-ray spectroscopy (EDS) to probe the elemental distribution within the perovskite films. The films were prepared under dilute precursor conditions using 1.0% CMA or in the absence of added CMA. Because CMA contains five oxygen atoms (Fig. 1b) we used oxygen elemental mapping to understand the CMA distribution. The imaging of the 1.0% CMA sample show oxygen is present (Fig. S23d, ESI †) and located on the perovskite grains (Fig. S23e, ESI †). In contrast, no oxygen could be detected for the control system (Fig. S24d,  ESI †). These data support our view that CMA (and CHE) adsorbed onto the grain surfaces. Furthermore, the PL data measured for the buried interface (Fig. S18, ESI †) imply that CMA and CHE are also present at, or in the vicinity of, the buried perovskite interface. Together these data indicate that the CAA additives are distributed at the surfaces of the grains throughout the films.

Additional evidence for perovskite-coumarin binding
Crystallization rate tests were conducted to determine whether CMA or CHE could control perovskite crystal growth (see Fig. 5c). The delay in achieving the black colour shows that the crystallization time was increased from 6 s for the 0% control film to 20 s in the presence of 1.0% CHE or 1.0% CMA. Hence, both CMA and CHE decrease the perovskite crystallization rate, which implies that both coumarins bind to the growing perovskite. The X-ray scattering patterns for the asprepared films (Fig. S25, ESI †) show an increase in PbI 2 conversion when the coumarins are present. Moreover, the fullwidth at half-maximum height (FWHM) is a minimum for the 0.3% CHE and 0.3% CMA systems (Fig. 5d). This trend agrees with the grain size changes recorded from SEM (Fig. S12, ESI †). In addition, the (001) scattered intensity is highest for the 0.3% CMA-based film (Fig. S25b, ESI †). These data provide evidence for an interaction of CMA and CHE wherein the crystallization rate is decreased and the crystal size increased.
We used FTIR spectroscopy to further probe the interaction between PbI 2 and CHE as well as CMA. The spectra recorded over the whole FTIR range are shown in Fig. S26, ESI. † Fig. 5e shows that the stretching vibration of the CQO group of CMA occurs at 1714 cm À1 . When PbI 2 was mixed with CMA, the peak View Article Online moves to 1708 cm À1 . This red-shift is due to CQO binding to Pb 2+ , 62,63 and supports the simulation data (Fig. 4c). The position of the CQO band for CHE at 1682 cm À1 moves to 1702 cm À1 when mixed with PbI 2 which is an indication of a changed environment. Furthermore, the O-H stretch for CHE at 3455 cm À1 shifts to 3438 cm À1 when mixed with PbI 2 (Fig. 5f). This indicates a direct interaction between the CHE O-H group and Pb and supports the DFT findings of Fig. 4d.

Underlying mechanism explanation
The results presented have provided insights into the improved performance and stability of the PSCs containing 0.3% CMA or 0.3% CHE as well as the greater overall benefit provided by CMA. There is evidence from FTIR, XPS and DFT that the CQO groups from CMA and CHE coordinate to Pb 2+ . The benefit from this effect is optimum at 0.3% CMA or CHE. However, at concentrations of 0.5 or 1.0%, it is likely that excessive CMA or CHE deposited at the perovskite grains impede charge transport. Moreover, the above data show that CMA more strongly binds to Pb 2+ than CHE which is proposed to be due to the higher dipole moment calculated from DFT analysis. The more effective binding by CMA decreases the defect density and recombination, which increased V oc and the PCE. We applied the Student's t-test to the J sc data for the CMA and CHE based devices measured at the same concentrations. When data from the reverse and forward sweeps are compared for each system, the analysis shows that the J sc values for CMA are significantly higher than those for CHE (see discussion in ESI † and Table S6). The J sc difference contributed to the greater PCE for the CMA-based devices compared to those prepared using CHE in this work.
The contact angle data (Fig. 1e) showed that CMA provided a protective hydrophobic layer. The stability of the films and devices benefits from the enhanced defect passivation of the perovskite surface by CMA and the good moisture resistance afforded by the hydrophobic groups. The moisture stability of perovskites is dependent on access to defects because water can bind firmly to the defect surface through hydrogen bonds, thereby accelerating the degradation. 64 Notably, the device containing 0.3% hydrophilic CHE had better moisture stability than the 0% film (Fig. 2h), which is likely to benefits from the bonding of the OH groups in CHE with the perovskite as evidenced from the DFT simulation results. Moreover, the moisture-resistance of the perovskite layer for this system is likely due to the protective barrier affect provided by bound CHE as noted above. Nevertheless, CMA provided the best stability due to stronger bonding to the perovskite and greater hydrophobicity.
An important question emanating from this study is why the PCEs for the CMA devices are only slightly higher than that for those prepared using CHE devices (Fig. 2e) whereas the stabilities for the former systems are far greater than those for the latter (Fig. 2g and h). To address this question we calculated the maximum number of CHE and CMA layers on the top and bottom of a nominal cubic perovskite grain that extended from the ETM to the HTM (see discussion and Fig. S27 in the ESI †). We estimate that there are a maximum of about 6 layers for both 0.3% CMA and 0.3% CHE systems. Accordingly, the relatively insulating nature of CMA and CHE (compared to the perovskite) will tend to impede charge transport to the ETM and HTM. This may be a reason why the J sc values are not greatly different; although, J sc is larger for the CMA as discussed above. Moreover, it is the nature of the hydrophobicity differences for these CMA and CHE multilayers that provide a far greater difference to water penetration. The latter will be more difficult for CMA multilayers and provides much stronger stability enhancement for the CMA-based systems.

Conclusions
This study has compared two new CAAs, i.e., CMA and CHE, that are free of parasitic absorption, which enabled them to be added to the precursor solution (simplifying processing) at relatively high concentration compared to related studies 32 and revealed the effects of CAA hydrophobicity on PSC performance and stability.
Three key results emerge. First, the CQO functional group in CMA and CHE coordinates with Pb and passivates defects on the perovskite surface and grain boundaries throughout the perovskite films. This inhibits non-radiative recombination as determined by the PL measurements. Second, the alignment of binding energy levels also improves the extraction of charge-carriers and contributes to the increase of the V oc . These effects contributed to the PCE of 21.28% for the champion CMA-modified devices and 21.14% for the best CHE-modified devices, whereas the additive-free system had a maximum PCE of 19.15%. Third, we also successfully fabricated devices containing hydrophobic CMA with high shelf-life, moisture and thermal stabilities. This system achieved a t 80 value of 98 days when stored in ambient air. The increased hydrophobicity of CMA provided a decisive stability advantage.
Overall, our results show that hydrophobic CAAs, such as CMA are good choices for improving PSC device performance. An unexpected result is that CHE also improved device stability even though it increased the perovskite hydrophilicity; such protection was attributed to effective binding of CHE to the perovskite. The study demonstrates that the substituents of CAAs can be used to enhance the performance and stability of PSCs, with the greatest enhancement from substituents that increase the overall hydrophobicity. We believe that selecting CAAs that were soluble in the perovskite precursors enabled these benefits to be conferred throughout the perovskite films. (FAI, 99.5%), and methylammonium bromide (MABr, 99.5%) were purchased from Ossila Limited. PbI 2 (99.99%) was purchased from Tokyo Chemical Industry UK Ltd. All materials were used as received. SnO 2 (15% in H 2 O) was a colloidal dispersion supplied by Alfa Aesar. All water used was de-ionized and of ultra-high purity.

Synthesis of CHE
The synthesis of CHE (7-(2-hydroxyethoxy)-4-methylcoumarin) followed our earlier study. 65  All operations were carried out in the glove box. The precursor solution (1.4 M) was prepared by dissolving appropriate amounts of CsI (0.013 g, 0.050 mmol), FAI (0.139 g, 0.81 mmol.), MABr (0.016 g, 0.143 mmol.), PbBr 2 (0.056 g, 0.153 mmol.) and PbI 2 (0.406 g, 0.880 mmol., which was 4 mol% in excess) in DMF/DMSO (v/v 4 : 1). Different concentrations of CMA or CHE were added directly to the precursor and mixed with stirring at 55 1C for 2 h. To prepare the 0.3% solutions, CHE or CMA (4.0 mg) were mixed with the other solid precursors before adding DMF (568 mL)/DMSO (142 mL). The precursor solution was prepared and subsequently spin-coated in two steps: 1000 rpm for 10 s followed by 6000 rpm for 20 s. Then, 10 s before the end, CBZ (200 mL) was drop-coated as an antisolvent. The substrates were annealed at 110 1C for 50 min to produce perovskite films. The Spiro-OMeTAD solution was prepared by dissolving Spiro-OMeTAD (85.7 mg) and additives in CBZ (1000 mL). Li-TFSI solution (22 mL of 520 mg mL À1 solution in ACN), Co(III) TFSI salt (18 mL of 375 mg mL À1 solution in ACN), and t-BP (32 mL) were added to the Spiro solution. The HTM (B250 nm thickness) was prepared by spin-coating using 3000 rpm for 30 s. Finally, a gold layer (80 nm) was thermally evaporated on top of the HTL as an electrode.
Characterization 1 H and 13 C NMR spectra were measured using a B400 Bruker Avance III HD400-Cyan spectrometer. XPS was conducted on an ESCA2SR spectrometer (Scienta Omicron GmbH) using monochromatic Al Ka radiation (1486.6 eV, 25 mA emission at 300 W, 1 mm spot size) at a base vacuum pressure of B1 Â 10 À9 mbar. UPS was characterized by a vacuum UV source producing HeI (21.2 eV, 60 W, FOCUS GmbH) and an Argus electron energy analyzer (Scienta Omicron GmbH) with a bias voltage of 20.2 V and a pass energy of 5.0 eV. SEM images were obtained with Tescan Mira 3 SC. The morphology and chemical composition of the films were characterized by scanning transmission electron microscopy (HR-STEM) and energy dispersive X-ray spectroscopy (EDS) using a Thermo Fisher Titan STEM (G2 80-200) equipped with a Cs probe corrector (CEOS) operating at 200 kV. The TEM samples were prepared by depositing dilute perovskite precursor solutions (0.28 M) in the presence or absence of 1.0 wt% CMA spin coated onto a copper grid coated with an amorphous carbon films on 300 mesh Cu grids. XRD data were acquired with an XRD5 -PANaytical X'Pert Pro X-ray diffractometer. FTIR was recorded using a Nicolet 5700 spectrometer. FTIR samples were prepared by mixing PbI 2 , CMA or CHE with DMF/DMSO mixture and stirring at 55 1C for 2 h, and then evaporating the solvent at 110 1C to obtain a dry powder. UV-visible spectra were measured using an Agilent Cary 60 UV-Vis spectrophotometer. Steady-state PL was measured by an Edinburgh Instrument FLS980. The excitation wavelength was 470 nm. TRPL data were also measured using the FLS980 and an excitation wavelength of 405 nm. All UV-visible, PL and TRPL spectroscopy measurements were recorded using glass/ITO/SnO 2 substrates. For all sample measurements, the light was incident on the film side. Contact angle measurements were performed with a drop of water (60 mL) on the film surface using a Kruss DSA100. J-V curves were generated by a Keithley 2420 source meter simulating AM 1.5G sunlight (100 mW cm À2 ) in an Abet solar simulator. The instrument was calibrated and corrected for spectral irradiance mismatch using a certified Oriel Si reference cell. The device area is 0.079 cm 2 . Unless otherwise stated, PCE and other data discussed in this article are from reverse scans. For all stability tests, two devices were used per system. A Keithley 2420 source meter was also used for measuring the SCLC characteristics of electron only devices. A Newport QuantX-300 instrument was used for external quantum efficiency (EQE) measurements. The Long-term light soaking test was performed under full AM 1.5 sun-equivalent white LED arrays. All devices were in ambient air. PCE was measured in the same way as the J-V curves.

DFT simulation
We carried out density functional theory (DFT) calculations as implemented in the Quantum ESPRESSO package. 66,67 For the reaction energies, the Kohn-Sham wave-functions and energies are calculated with the functional GGA-PBE 68,69 for electron exchange and correlation, using a plane-wave basis, with energy and charge density cutoffs of 30 and 300 Ry, respectively. The Grimme dispersion correction DFT-D3, 70,71 was used to account for the dispersion corrections. Ultrasoft pseudopotentials are used to describe the core-valence interactions. 72 The structural relaxation is performed until the force on each atom is smaller than 0.01 eV Å À1 . For the geometry optimization, a Gamma-point and a 4 Â 4 Â 1 k-point sampling for the Brillouin zone integration are used for the interface and bulk systems, respectively, following the Monkhorst-Pack scheme. 73

Conflicts of interest
There are no conflicts to declare.