In situ growth of graphene on both sides of a Cu–Ni alloy electrode for perovskite solar cells with improved stability

The instability of rear electrodes undermines the long-term operational durability of efficient perovskite solar cells. Here, a composite electrode of copper–nickel (Cu–Ni) alloy stabilized by in situ grown bifacial graphene is designed. The alloying makes the work function of Cu suitable for regular perovskite solar cells. Cu–Ni is the ideal substrate for preparing high-quality graphene via chemical vapour deposition, which simultaneously protects the device from oxygen, water and reactions between internal components. To rivet the composite electrode with the semi-device, a thermoplastic copolymer is applied as an adhesive layer through hot pressing. The resulting devices achieve power conversion efficiencies of 24.34% and 20.76% (certified 20.86%) with aperture areas of 0.09 and 1.02 cm2, respectively. The devices show improved stability: 97% of their initial efficiency is retained after 1,440 hours of a damp-heat test at 85 °C with a relative humidity of 85%; 95% of their initial efficiency is retained after 5,000 hours at maximum power point tracking under continuous 1 sun illumination. The instability of contact layers for perovskite solar cells under operating conditions limits the deployment of the technology. Now, Lin et al. develop a Cu–Ni electrode sandwiched between in situ-grown graphene protective layers, enabling solar cells with improved stability under light, humidity and high temperature.

M etal halide perovskite solar cells (PSCs) have attracted great attention in both academia and industry owing to their excellent optoelectronic performance and low manufacturing costs [1][2][3][4][5][6] . However, for PSCs to realize commercialization, they must survive the long-term natural erosion imposed by oxygen, moisture, light and heat 7,8 . Thanks to the optimization of the perovskite materials, charge transport materials and interface layers [9][10][11][12] , the operational stability of PSCs has made great progress; but one of the key functional layers, the rear electrode, is still prone to fail, which limits the overall durability of efficient PSCs 13,14 .
Silver (Ag) and aluminium (Al) are commonly used rear electrodes; however, they tend to react with migrated halide anions from perovskite to form resistive compounds such as AgI and AlI 3 (refs. [15][16][17]. In the case of gold (Au), although the formation enthalpy of Au-I is much higher than those of Ag-I or Al-I, Au atoms can diffuse into the perovskite to form antisite defects of Au Pb with deep level, which act as efficient non-radiative recombination centres 18 . In addition, the use of noble metals comes at a high price and implies more stringent requirements for vacuum and temperature, which will significantly increase the manufacturing costs. To suppress the interaction between perovskite and metal atoms, introducing a thin buffer layer to separate them without hindering the charge transport has been reported as a successful strategy 19,20 . However, materials from the perovskite or the electrode tend to penetrate that buffer layer on a long timescale because achieving a thin but uniform, compact coverage on a large area with buffer layers that are usually formed by solution-processable small molecules or polymers is hard 21 ; thus, increased intrinsic chemical stability of rear electrodes for PSCs is needed.
Copper (Cu) is a potential candidate due to its relative inertness to migrated perovskite components, and it has been widely used in inverted PSCs 22 . However, its work function (WF) of 4.65 eV limits its application in efficient regular PSCs, which usually use 2,2′,7,7′-tet rakis(N,N-dip-methoxyphenylamine)-9,9′-spirobifluorene (spiro-MeOTAD) or poly(bis(4-phenyl)(2,4,6-trimethylphenyl) amine) (PTAA) with a highest occupied molecular orbital (HOMO) of around −5.2 eV (ref. 23 ). In addition, Cu can be oxidized by oxygen and moisture to generate products such as Cu 2 (OH) 2 CO 3 , and Cu atoms were reported to diffuse into perovskite at 85 °C (ref. 24 ). In addition to metal electrodes, intrinsically stable carbon electrodes have also been applied to PSCs; 9,000 hours of operational stability at the maximum power point (MPP) and 55 °C was recently reported with no obvious change in the power conversion efficiency (PCE) 25 . However, big challenges exist for carbon electrodes to obtain high efficiencies because of the severe potential losses 26 . Therefore, it is still an open issue to develop a low-cost rear electrode with a tunable WF that combines chemical stability and high efficiency.
Herein, we report a composite electrode that consists of (1) copper-nickel (Cu-Ni) alloy and (2) in situ grown bifacial graphene. The alloying of Cu with Ni improves its electrochemical stability without compromising the electrical properties and makes it possible to tune the WF of Cu to fit the requirements of regular PSCs by simply varying the content of Ni. Moreover, Cu-Ni alloy is an ideal substrate for preparing high-quality graphene on both sides In situ growth of graphene on both sides of a Cu-Ni alloy electrode for perovskite solar cells with improved stability Xuesong  The instability of rear electrodes undermines the long-term operational durability of efficient perovskite solar cells. Here, a composite electrode of copper-nickel (Cu-Ni) alloy stabilized by in situ grown bifacial graphene is designed. The alloying makes the work function of Cu suitable for regular perovskite solar cells. Cu-Ni is the ideal substrate for preparing high-quality graphene via chemical vapour deposition, which simultaneously protects the device from oxygen, water and reactions between internal components. To rivet the composite electrode with the semi-device, a thermoplastic copolymer is applied as an adhesive layer through hot pressing. The resulting devices achieve power conversion efficiencies of 24.34% and 20.76% (certified 20.86%) with aperture areas of 0.09 and 1.02 cm 2 , respectively. The devices show improved stability: 97% of their initial efficiency is retained after 1,440 hours of a damp-heat test at 85 °C with a relative humidity of 85%; 95% of their initial efficiency is retained after 5,000 hours at maximum power point tracking under continuous 1 sun illumination.
by chemical vapour deposition (CVD); this graphene acts as a natural barrier and substantially strengthens the stability of the Cu-Ni alloy because the outer graphene can block the penetration of water and oxygen, while the inner graphene can suppress the migration of perovskite components or metal atoms caused by light and heat. The composite electrode is hot-pressed on the semi-device with the assistance of a thermoplastic copolymer, which ensures the efficient charge collection at the interface of the semi-device and the electrode. As a result, a PCE of 24.34% was achieved for the regular device using graphene/Cu-Ni/graphene composite as the rear electrode (the CNG device) with an aperture area of 0.09 cm 2 , comparable to that of the device using a Ag electrode. After the optimization of the thermoplastic copolymer with a conductive graphene nanoflake (GN), we obtained a PCE of 20.76% (certified PCE, 20.86%) for the CNG device with an aperture area of 1.02 cm 2 . The CNG devices retained 97% of their initial PCEs after the aging test of heating at 85 °C with a relative humidity of ~85% for 1,440 hours, and the encapsulated devices maintained 95% of their initial PCEs after operating at the MPP under continuous 1 sun illumination for 5,000 hours.
Fabrication and characterization of CNG electrodes. The fabrication process of the CNG composite electrode and the assembly of the whole device ( Supplementary Fig. 1) are detailed in the Methods. The thickness of the sputtered Ni before annealing is set at 200 nm to obtain a WF for Cu-Ni (denoted as CN-2) that is compatible with the HOMO of spiro-MeOTAD or PTAA ( Supplementary  Fig. 2a-d). As shown in the cross-sectional scanning electron microscopy (SEM) image ( Supplementary Fig. 2e), the thickness of Cu-Ni is approximately equal to that of Cu foil (~25 μm). The corresponding energy dispersive spectroscopy (EDS) elemental mapping images confirm the uniform distribution of Ni in the Cu foil ( Supplementary Fig. 2f,g) after thermal annealing under a reducing atmosphere, and the semiquantitative stoichiometry of the Cu-Ni alloy is 81:19 (Cu/Ni). The exact stoichiometry of the Cu-Ni alloy is measured by an inductively coupled plasma spectrometer, and the result of 83.537:16.463 (Cu/Ni) is close to that measured by EDS. The process of thermal annealing under a reducing atmosphere also smoothes the surface of the substrate, thereby preparing it for the in situ growth of graphene, as demonstrated by atomic force microscopy (AFM; Supplementary Fig. 2h).
A uniform but thin layer of graphene is required to meet the demand of being an efficient barrier without introducing extra defects or hindering the carrier transport 27 . We focus on the characterization of one side since the quality of graphene on both sides should be identical in a CVD method. Using CH 4 as the carbon source, with respective flow rates of 5, 10 and 15 standard cubic centimetres per minute (sccm), diluted in 100 sccm Ar/H 2 , we arm both sides of the Cu-Ni alloy with graphene with varying numbers of layers and quality, denoted as CNG-5, CNG-10 and CNG-15, respectively. Their optical microscope images are provided in Fig. 1a-c. Obviously, due to the differences in the content of the carbon source, the graphene layers on CNG-5 and CNG-15 form a non-uniform monolayer and a coexisting monolayer and bilayer, respectively. By contrast, a uniform monolayer graphene is observed on CNG-10.
To further determine the quality and layer numbers of graphene, Raman spectra are measured for the three samples. The full width at half maximum values for the 2D peak at ~2,680 cm −1 are shown in Fig. 1d-f; the graphene of CNG-10 exhibits the most uniform distribution of full width at half maximum for the 2D peak with an average value of below 45 cm −1 . Coupled with the intensity ratios of the 2D peak and G peak at ~1,584 cm −1 (I 2D /I G ), as shown in Fig. 1g-i, the layer numbers of graphene on CNG-5, CNG-10 and CNG-15 can be deduced 28 . The graphene of CNG-5 contains regions of I 2D /I G > 2.5, which represents a discontinuous monolayer graphene film and is consistent with the optical microscope image in Fig. 1a. The I 2D /I G for CNG-10 almost resides between 1.5 and 2.5, indicating the formation of monolayer graphene. A further increase in the content of CH 4 leads to blue regions for CNG-15 (0.8 < I 2D /I G < 1.2), which is indicative of the formation of a non-uniform bilayer and agrees well with the full width at half maximum of 60 cm −1 in Fig. 1f. Incomplete coverage may weaken the function of being the barrier for water, oxygen and migrated components, whereas overlapped graphene not only enlarges the defect density at this interface ( Supplementary Fig. 3a-c) but also increases the series resistance (R S ) of PSCs. Therefore, CNG-10 is expected to be the most suitable composite electrode among the three samples.
Assembly and performance of CNG devices. Before the assembly of an integral device, we first demonstrate the high quality of the perovskite film used in our work, which shows a microscale grain size ( Supplementary Fig. 4a), with a thickness of ~650 nm (Fig. 2a). Featured peaks of α-NH 2 CHNH 2 PbI 3 (α-FAPbI 3 ) and PbI 2 at 13.9° and 12.7°, respectively, are observed in X-ray diffraction patterns ( Supplementary Fig. 4b), and the δ-FAPbI 3 at 11.8° is suppressed. The results from ultraviolet-visible (UV-vis) absorption spectroscopy and photoluminescence show the absorption edge of our perovskite at 820 nm ( Supplementary Fig. 4c), which is equivalent to an optical bandgap of 1.51 eV. Moreover, the time-resolved photoluminescence result of the perovskite determined from the mono-exponential fitting exhibits a long carrier lifetime of 2.50 μs (Supplementary Fig. 4d and Supplementary Table 1). Then, we investigate the effect of graphene on the WFs of the Cu-Ni alloy by ultraviolet photoelectron spectroscopy. As shown in Supplementary Fig. 5a-d, the WFs of the CNG-5, CNG-10 and CNG-15 composite electrodes are 5.15, 5.11 and 4.99 eV, respectively, which are also compatible with the HOMOs of the commonly used hole transport materials (HTMs) in regular PSCs. The integral devices are fabricated by hot pressing CNG electrodes on the semi-device with the structure of fluorine-doped tin oxide (FTO)/SnO 2 /perovskite/HTM. A thin adhesive layer (~2.9 nm) of thermoplastic ethylene vinyl acetate (EVA) copolymer with a WF of 3.69 eV and a deep HOMO level of −7.48 eV ( Supplementary Fig. 6a-c) is employed at the interface to assist the combination of HTMs and CNG electrodes and to fill the possible gaps left after hot pressing, which is beneficial for realizing efficient carrier tunnelling and thereby enhanced ohmic contact 29 . The temperature of hot pressing is set at 130 °C, which is a conventional value for EVA-based materials 30,31 . In addition, the pressure and time are optimized to be 3 bar and 60 seconds, respectively, after the measurement of the resistance of the CNG electrodes/EVA/ spiro-MeOTAD/FTO ( Supplementary Fig. 7a,b). It is also found that the resistance will increase when the time is over 120 seconds (3 bar, 130 °C), while the pressure seems to have no obvious influence on the sample resistance.
To further investigate the effect of time and pressure for hot pressing on the thermally instable spiro-MeOTAD, UV-vis absorption spectroscopy and X-ray diffraction were employed; as the time for hot pressing exceeds 120 seconds, the featured peak of spiro-MeOTAD at ~375 nm in UV-vis absorption spectroscopy starts to decrease ( Supplementary Fig. 7c). In addition, the crystallization peaks of spiro-MeOTAD appear in X-ray diffraction patterns ( Supplementary Fig. 7e), indicating the recrystallization of spiro-MeOTAD [32][33][34] . However, the increase in pressure from 0 to 9 bar doesn't have an obvious influence on spiro-MeOTAD ( Supplementary Fig. 7d,f). The results of UV-vis absorption spectroscopy and X-ray diffraction are consistent with the change in the resistance of the CNG electrodes/EVA/spiro-MeOTAD/FTO. The effect of the hot pressing (temperature, 130 °C; pressure, 3 bar; time, 60 seconds) on the perovskite and the charge transfer at the perovskite-spiro-MeOTAD interface were then investigated. As shown in Supplementary Fig. 8a,b, the perovskite films before and after hot pressing show a similar morphology with a microscale grain size. The X-ray diffraction patterns both show a strong featured peak of α-FAPbI 3 at 13.9°, and the peak at 11.8° for the photo-inactive δ-FAPbI 3 is not observed ( Supplementary  Fig. 8c), which is consistent with the results of Supplementary Fig.  4b. Supplementary Fig. 8d exhibits the time-resolved photoluminescence results of the glass/perovskite/spiro-MeOTAD before and after hot pressing, which show a similar decay with average carrier lifetimes of around 7.0 ns; the fitted results are summarized in Supplementary Table 2. All the results demonstrate that hot pressing with optimized conditions will not damage the perovskite, the spiro-MeOTAD or their interface. Figure 2a shows the cross-sectional SEM image of a CNG device; the EVA is too thin to be distinguishable so that it can realize efficient carrier tunnelling. The current density-voltage (J-V) curves of PSCs using CNG-5, CNG-10, CNG-15 and Ag electrodes (the CNG-5, CNG-10, CNG-15 and Ag devices, respectively) under a forward scan are presented in Fig. 2b. The CNG-10 device obtained a PCE of 24.34% with an aperture area of 0.09 cm 2 , which is comparable with that of the Ag device (best PCE = 24.65%). The corresponding incident photon-electron conversion efficiency is provided in Supplementary Fig. 9a, and the integrated current density of 25.77 mA cm −2 shows a minimal mismatch with the short-circuit current density (J SC ) from the J-V scan (26.16 mA cm −2 ). The steady-state PCE is measured to be 24.28% ( Supplementary  Fig. 9b), which is consistent with the PCE measured by the J-V curves. The detailed photovoltaic parameters of Fig. 2b are summarized in Supplementary Table 3. The inferior performance of the CNG-5 and CNG-15 devices can be attributed to the lower quality of graphene, leading to charge accumulation at the HTM-electrode interface. This effect can be verified by capacitance-voltage (C-V) measurements via searching the characteristic peaks in capacitance at a specific region of bias under light 35 . As shown in Supplementary  Fig. 10a-d, the C-V curve of the CNG-10 device shows no obvious peaks, indicating the fast charge transport in the device, whereas the CNG-5 device shows a peak at approximately −1.1 V and the CNG-15 device shows peaks at both approximately −0.3 V and approximately −1.1 V, indicating the charge accumulation in those devices.
Twenty devices of each batch were fabricated; the histogram of average PCE values is shown in Supplementary Fig. 11. The CNG-10 devices exhibit a narrower distribution of PCE compared with the other CNG devices and thereby a higher reproducibility due to the uniform and high-quality graphene layer, as demonstrated in Fig. 1.
To show the effect of the adhesive layer on the performance of the CNG device, we also fabricate a CNG-10 device without the EVA layer, which shows a comparatively low efficiency of 20.65%, probably due to the unfilled gaps at the interface ( Supplementary Fig. 12).
We further fabricated a CNG device with an aperture area of 1.02 cm 2 . However, a much lower PCE of 15.43% was achieved ( Supplementary Fig. 13a), probably due to the non-uniformity of hot pressing, which may cause larger gaps at the interface beyond the thickness of EVA. In this case, we slowed down the rotation speed of EVA from 6,000 to 5,500 and 5,000 r.p.m., and the R S of the corresponding devices first decreases and then suddenly increases to a  much higher value ( Supplementary Fig. 13b), which means that the choice of 5,000 r.p.m. leads to the formation of an insulating layer too thick to allow charge tunnelling. To allow for higher EVA thickness in localized areas with larger gaps, we mitigated resistive losses via addition of GN in EVA to boost the conductivity. The carbonyl group (C=O) on the branch chain of EVA can interact with GN via van der Waals forces, as confirmed by the small blueshift (~0.33 eV) of the C=O peak measured by X-ray photoelectron spectroscopy (XPS) in Supplementary Fig. 13c, which helps the formation of a conductive network of GN in the EVA matrix 36 . As is shown in Supplementary Fig. 13d, the volume resistance of EVA goes through a sudden drop when the mass ratio of GN reaches 1.6 wt%, and further increase to 2.4 wt% barely improves the conductivity, indicating the formation of a stable conductive network. The J-V curves of the devices with 0.8 wt%, 1.6 wt% and 2.4 wt% GN in EVA are shown in Fig. 2c (the 0.8%, 1.6% and 2.4% devices, respectively). Obviously, the PCE of the device is greatly improved to 20.76% (certified PCE, 20.86%; Supplementary Fig. 14) when the network is constructed (1.6% device); the corresponding incident photon-electron conversion efficiency is provided in Supplementary Fig. 13e, showing a minimal mismatch with the J SC from the J-V scan (<3%). In addition, no obvious hysteresis is observed ( Supplementary Fig. 13f), and the steady-state PCE is measured to be 20.80% ( Supplementary Fig. 15), which is consistent with the PCE measured by the J-V curves. However, a drop in the optoelectronic performance is found for the 2.4% device, especially in the open-circuit voltage (V OC ) and the fill factor. To discover the reason, we measured the transient photovoltage decay to analyse the charge carrier dynamics in the devices (Fig. 2d). The faster decay of transient photovoltage for the 2.4% device suggests a shorter charge carrier lifetime than that for the 1.6% device, which may be related to the aggregation of GN 37 . The distribution of GN in EVA is characterized by Raman spectra (Fig. 2e,f); apparent aggregation of GN represented by the red regions in the figure is found in EVA with 2.4 wt% GN, which may bring about a higher concentration of defects that cause extra non-radiative losses at the interface. A variety of conductive additives can be chosen to boost the conductivity of the copolymer. However, to guarantee the stability of the interface, stable materials with good dispersity will be the priority selection.
Stabilizing mechanism of CNG electrode. To reveal the stabilizing mechanism of CNG, we first studied its resistance to water and oxygen. As shown in Supplementary Fig. 16a-e, the contact angle with water in air increases from 74.9° for a Ag electrode to 102° for CNG-10, which should be beneficial for repelling water. Even if water is dropped on and stays on the surface of CNG-10 for 24 hours, the colour of the perovskite remains black, while the perovskite under Ag turns yellow ( Supplementary Fig. 16f,g). To exclude the effect of electrode thickness, the permeation rates of water (water vapour transmission rate) and oxygen (oxygen transmission rate) for the CNG-10 electrodes are measured and decrease to 3.32% and 5.38% of the initial values for the Cu-Ni alloy after being equipped with the air-tight and hydrophobic graphene ( Supplementary Fig. 17a,b). The reduced permeability meets the requirements for encapsulation 38 . To study this encapsulation effect of CNG electrodes on devices, in Fig. 3a, we conducted gas chromatography-mass spectrometry (GC-MS) to compare the volatile products escaping from the Ag and CNG-10 devices after aging tests at 85 °C and 85% relative humidity in air for 500 hours 39 . Barely any volatile products are found for the CNG device, except for a small quantity of NH 3 and NO. By contrast, a variety of products are detected for the Ag device, including the additive 4-tert-butylpyridine from the HTM and the CH 3 COOH hydrolysed from the passivation layer by the intrusion of moisture. The GC-MS results demonstrate the superior damp-heat stability of the CNG device.
To investigate the electrochemical corrosion of the CNG electrodes, Tafel curves for Cu, Cu-Ni and CNG-10 are measured (Fig. 3b). Compared with Cu, the corrosion current is lower at the anodic section for Cu-Ni, implying a slower corrosion rate due to alloying with Ni. For the sample of CNG-10, the self-corrosion potential (E corr ) positively shifts from −0.34 V (Cu and Cu-Ni) to −0.29 V, representing a much higher resistance to chemical corrosion, which is due to not only the blocking of water and oxygen, but also the decrease in the migration rate of ionic species that can trigger oxidation and reduction 40 .
The effect of the CNG-10 electrode on the suppression of migrated ions and metal atoms in the devices is further demonstrated. After removing the package and peeling off the electrodes of the devices that were aged at the MPP for 1,000 hours, XPS measurement revealed an I − signal inside the CNG-10 that was much lower than that inside the Ag (Supplementary Fig. 18a,b), a result that preserves the morphologies of spiro-MeOTAD and perovskite, as shown in Supplementary Fig. 18c,d. The spatial distributions of I − , Ag − , Au − and Cu − for the aged Ag, Au and CNG devices were analysed by time-of-flight secondary ion mass spectroscopy ( Supplementary Fig. 18e,f and Supplementary Fig. 19a). Obviously, the I − (Fig. 3c) and Ag − (Fig. 3e) in the aged Ag device have migrated over the whole depth of the device. Even though the migration of I − (Supplementary Fig. 19b) and metal atoms ( Supplementary  Fig. 19c) is suppressed in the aged Au device compared with the Ag device, there are still (AuI 2 ) − signals (Supplementary Fig. 19d) and Au clusters (Supplementary Fig. 19a) that are detected in the perovskite layer, which is detrimental to the long-term stability  of the Au device 41,42 . On the contrary, the I − (Fig. 3d) and Cu − (Fig. 3f) in the aged CNG-10 device are almost confined to their original layers.

Stabilizing effect of CNG electrode on devices.
Benefiting from the effect of the CNG electrode on resisting water and oxygen and suppressing component migration, CNG devices show great damp-heat and operational stabilities. The CNG-10 devices retained 97% of their initial efficiency after the aging test of heating at 85 °C with ~85% relative humidity for 1,440 hours ( Supplementary Fig. 20). For the operational stability test, another control sample using Cu-Ni alloy with bifacial sprayed graphene as the electrode (the SG device) was tested. All the encapsulated devices were measured at the MPP with continuous 1 sun illumination; the CNG-10 devices retained 95% of their initial PCEs after 5,000 hours with a small deviation across five individual cells, while the PCEs of the SG, Au and Ag devices dropped to 59%, 47% and 31% of their initial PCEs after ~2,500, 1,750 and 1,250 hours, respectively (Fig. 4). The initial PCEs of the devices are summarized in a box plot (Supplementary Fig. 21). The inferior operational stability of the SG device compared to the CNG device may be ascribed to the numerous pores in overlapped sprayed graphene, which decreases the effect of suppressing component migration. The EVA layer is also incorporated in the SG devices. Therefore, the main reason for the enhanced operational stability of CNG devices is the CNG electrode.
To further decouple the effect of the EVA layer on the stability, we coated the EVA on the perovskite and observed the degradation of the perovskite at 85% relative humidity and 85 °C using UV-vis absorption spectroscopy and X-ray diffraction ( Supplementary  Fig. 22). The EVA has a positive effect on preventing the decomposition of perovskite for the first 200 hours, whereas an obvious drop in the absorption is found after 1,000 hours, accompanied by a much stronger featured peak of PbI 2 at 12.7°. In addition, to analyse the possible ability of the EVA to suppress metal atom diffusion, we coated EVA on glass/80 nm Au or glass/CN-2 and detected the surfaces by XPS (penetration depth, 2-5 nm) before and after the samples were heated at 85 °C for 500 hours. To ensure that the signals are from the metal diffusion, we increased the concentration of the EVA dispersion from 0.025 g ml −1 to 0.250 g ml −1 . As a result, distinct peaks of the Au 4f core level and Cu 2p core level are detected for the samples of glass/80 nm Au/EVA and glass/CN-2/ EVA, respectively ( Supplementary Fig. 23). Therefore, it is hard for the original EVA to prevent metal diffusion in a CNG device. Both results indicate the limited effect of the thin EVA layer on the long-term stability of CNG devices.
Finally, the manufacturing costs of the CNG electrode were estimated to be US$149.992 m −2 by considering the raw materials, magnetron sputtering and CVD (Supplementary Table 4). This cost is about one-third of that of the evaporated Au electrode. More importantly, the size of our sample for CVD was restricted by the equipment in our lab, which increased the manufacturing costs. It is expected that the manufacturing costs of a CNG electrode will be significantly reduced after commercialization.

Conclusions
We have developed a composite electrode for PSCs that ensures efficient and stable charge collection during the long-term operation of the devices. A Cu-Ni alloy is used as the substrate to simply tune the WF, and strong barriers of graphene are in situ grown on both sides. A modified adhesive layer is also introduced at the interface of the semi-device and the CNG electrode to improve the ohmic contact. The corresponding devices retained 97% of their initial efficiencies after 1,440 hours of a damp-heat test at 85 °C with a relative humidity of 85%, and retained 95% of their initial efficiencies after 5,000 hours of tracking at the MPP under continuous 1 sun illumination. We anticipate that this work will open new avenues for designing electrodes to improve the stability of PSCs.

Materials.
All the chemicals were used without further purification, including (1) the SnO 2 electron transport material: dispersing SnO 2 , 15% in H 2 O colloidal dispersion (Alfa Aesar), and deionized water (Sigma-Aldrich); (2) the perovskite precursor: CH(NH 2 ) 2 I (98%, TCI), CH 3 NH 3 I (98%, TCI), CH 3  Synthesis of EVA-based adhesives. Different amounts of GN from 0 to 6 mg were added to 10 ml xylene and sonicated for 1 hour to form homogeneous GN dispersions. Then 0.25 g of EVA was added into the dispersions. The samples were placed in a drying oven at 80 °C to swell for 1 hour. The swelled sample was further stirred for 10 hours to form homogeneous EVA adhesive and EVA-GN adhesive.

Fabrication of CNG and SG electrodes.
For the fabrication of the CNG electrodes, the Cu foil was first chemically polished by immersion in an ammonium persulfate etchant solution in water (0.2 M) for 3 minutes 43 , which doesn't have an obvious effect on the thickness of the Cu foil. After that, the Cu foil was sonicated one by one with detergent, deionized water, ethanol, acetone and IPA for 15 minutes each. Ni layers with thicknesses of 100, 200 or 300 nm were then sputtered on the Cu foil using the magnetron sputtering method followed by thermal annealing under a 100 sccm Ar/H 2 reducing atmosphere at 1,000 °C for 1 hour to form uniform Cu-Ni alloys. By employing 5, 10 and 15 sccm of CH 4 diluted in 100 sccm Ar/H 2 as a carbon source at 1,000 °C for 30 minutes, both sides of the Cu-Ni alloy were protected by graphene with varying numbers of layers and quality, denoted as CNG-5, CNG-10 and CNG-15, respectively. For the fabrication of the SG electrode, 20 mg graphene was dispersed into 5 ml IPA through ball-milling for 12 hours to obtain a homogeneous dispersion. Then, 150 μl dispersion was sprayed onto both sides of a Cu-Ni alloy followed by annealing at 85 °C (ref. 44 ).
Sample preparation for investigating the effect of hot pressing on the perovskite. To simulate the real conditions, we fabricated the integral device with a structure of FTO/SnO 2 /perovskite/spiro-MeOTAD/EVA/CNG-10, and then peeled off the electrode and washed off the spiro-MeOTAD with chlorobenzene to obtain the perovskite sample after hot pressing (temperature, 130 °C; pressure, 3 bar; time, 60 seconds). Similarly, to prepare the perovskite sample before hot pressing, we fabricated the semi-device of FTO/SnO 2 /perovskite/spiro-MeOTAD and washed off the spiro-MeOTAD with chlorobenzene.

Fabrication of PSCs with different electrodes.
A precursor solution composed of methyl isobutyl ketone and silicon dioxide nanoparticles was brought from Shanghai Juanrou Newtech and diluted with IPA (v/v, 1:1). The diluted solution was spin-coated atop the glass side of FTO substrates at 3,000 r.p.m. for 30 seconds, followed by the thermal annealing process at 120 °C for 2 hours to remove the solvent and obtain the anti-reflection layer. The FTO substrates were etched with zinc powder and 6 M HCl for 15 seconds to obtain patterned substrates, followed by sonication with detergent, deionized water, ethanol, acetone and IPA for 15 minutes each. Before use, the substrates were cleaned with ultraviolet ozone for 20 minutes to remove the organic residues. Then a thin layer of SnO 2 was prepared by spin-coating a SnO 2 dispersion (v/v, 1:8.5 in deionized water) at 3,000 r.p.m. for 30 seconds, followed by annealing in the ambient air at 150 °C for 30 minutes. After the 20 minute treatment by ultraviolet ozone, the substrates were transferred to a glove box. Then 1.5 M PbI 2 in dimethylformamide/dimethyl sulfoxide (v/v, 9:1) was spin-coated on SnO 2 at 1,500 r.p.m. for 40 seconds and then a solution of CH(NH 2 ) 2 I/CH 3 NH 3 I/CH 3 NH 3 Cl (90:9:9 mg ml −1 ) in IPA was spin-coated on PbI 2 at a spin rate of 1,500 r.p.m. for 30 seconds, followed by thermal annealing at 150 °C for 15 minutes in the ambient air (~30% humidity). The substrates were then transferred to the nitrogen glove box and a potassium acetate (KAc) solution in IPA (1.5 mg ml −1 ) was spin-coated atop the perovskite at a spin rate of 5,000 r.p.m. for 30 seconds 45 . After that, the layer of spiro-MeOTAD or PTAA was spin-coated on the perovskite film at a spin rate of 3,000 r.p.m. for 30 seconds. This is a dynamical spin-coating; the dissolution of the KAc might occur if we load the HTM solution for a while and then start the rotation in static spin-coating, leading to an inhomogeneous morphology ( Supplementary Fig. 24). The spiro-MeOTAD was dissolved in chlorobenzene (72.3 mg ml −1 ) with additives of 17.5 µl bis(trifluoromethane)sulfonimide lithium salt/acetonitrile (520 mg ml −1 ) and 28.8 µl 4-tert-butylpyridine. The PTAA was dissolved in toluene (10 mg ml −1 ) with additives of 7.5 μl bis(trifluoromethane)sulfonimide lithium salt/acetonitrile (170 mg ml −1 ) and 4 μl 4-tert-butylpyridine. For PSCs with CNG electrodes, the EVA or EVA-GN adhesive was spin-coated on one side of the CNG at 5,000, 5,500 or 6,000 r.p.m. for 30 seconds, followed with a thermal treatment at 70 °C for 30 minutes to remove the solvent and form a thin and uniform bonding layer. The CNG electrodes were hot-pressed with the semi-device (glass/FTO/SnO 2 / perovskite/HTM) at 130 °C under a pressure of 0-9 bar for 0-180 seconds (hot pressing machine, TH-XC601-H, Shenzhen XOTECH). For PSCs with control electrodes, 80-nm-thick Ag or Au was thermally evaporated using a black metal shading mask. The 0.09 and 1.02 cm 2 masks were used to define the aperture areas during measurement.
Damp-heat and operational stability tests of the control and CNG devices. For the operational stability tests in Fig. 4, the devices were aged at the MPP based on a forward J-V scan every 36 hours. In addition, to rule out thermal degradation from the HTM layer, all the devices in Fig. 4 and Supplementary Fig. 20 used the thermally stable chlorinated graphene oxide/PTAA as the HTM 11 .
Characterization. The Raman mapping spectroscopy was performed on an inVia spectroscope (Renishaw) to assess the morphology and quality of the graphene with an excitation source of 532 nm Ar ion laser. Field emission SEM (JEOL JSM-7800F Prime) equipped with EDS was employed to observe the morphology and element distribution of the samples. The inductively coupled plasma spectrometer (iCAP7600, Thermo Scientific) was employed to confirm the exact stoichiometry of the Cu-Ni alloy. An AFM instrument (TT2-AFM) was employed to measure the roughness of the Cu-Ni substrates. The electrochemical impedance spectroscopy and C-V measurements were characterized by a multifunctional electrochemical analysis instrument (Zahner) under dark or white light (60 mW cm −2 ). The transient photovoltage measurements were carried out with a TranPVC instrument with a 640 nm laser under a light-emitting diode white light bias. The J-V curves of PSCs were measured under simulated solar illumination at 100 mW cm −2 , air mass 1.5 global (WXS-155S-10, Wacom Denso, recorded by a digital source meter (Keithley 2400)). The solar simulator was calibrated by a standard silicon reference cell, which was certified by the Calibration, Standards and Measurement Team at the Research Center for Photovoltaics in AIST, Japan. The spectral mismatch is <3%. The J-V measurements were measured under forward scan (−0.2 to 1.2 V) or reverse scan (1.2 to −0.2 V). The incident photon-electron conversion efficiency spectrum was characterized by a monochromatic incident light of 1 × 10 16 photon cm −2 (ref. 46 ). The time-of-flight secondary ion mass spectroscopy was performed with an IONTOF TOF.SIMS 5-100 instrument. A stable Cs + ion beam was used as the ion beam to peel off the PSCs with an analysis area of 100 × 100 μm 2 . The XPS and ultraviolet photoelectron spectroscopy were measured with a Kratos AXIS UltraDLD by Al Kα X-ray source. The static contact angle tests were characterized by a water droplet with a DSA100 instrument. The volume resistivity of the bonding layer (EVA or EVA-GN) on glass was measured on a high resistance meter (Agilent 4339B) to identify the percolation threshold of graphene in EVA polymer. The permeation rates of water vapour and oxygen for the Cu-Ni alloy and CNG electrodes were measured by a gas transmission rate tester followed by ISO 2556. Tafel polarization curves were obtained by an electrochemical workstation (CHI660D) with a three-electrode system. The GC-MS data were recorded by headspace GC-MS analytical methods based on a previous work 37 . A Rt-QPLOT column (30 m × 0.32 mm) or Rtx-VolatileAmine column (30 m × 0.32 mm) was connected to the programmed temperature vaporization inlet. The helium carrier gas was set to a constant flow, and the oven equilibration time was 10 minutes.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
Source data are provided with this paper. All data generated or analysed during this study are included in the published article and its Supplementary Information and Source Data files.