Metal-electrode-free Window-like Organic Solar Cells with p-Doped Carbon Nanotube Thin-film Electrodes

Organic solar cells are flexible and inexpensive, and expected to have a wide range of applications. Many transparent organic solar cells have been reported and their success hinges on full transparency and high power conversion efficiency. Recently, carbon nanotubes and graphene, which meet these criteria, have been used in transparent conductive electrodes. However, their use in top electrodes has been limited by mechanical difficulties in fabrication and doping. Here, expensive metal top electrodes were replaced with high-performance, easy-to-transfer, aerosol-synthesized carbon nanotubes to produce transparent organic solar cells. The carbon nanotubes were p-doped by two new methods: HNO3 doping via ‘sandwich transfer’, and MoOx thermal doping via ‘bridge transfer’. Although both of the doping methods improved the performance of the carbon nanotubes and the photovoltaic performance of devices, sandwich transfer, which gave a 4.1% power conversion efficiency, was slightly more effective than bridge transfer, which produced a power conversion efficiency of 3.4%. Applying a thinner carbon nanotube film with 90% transparency decreased the efficiency to 3.7%, which was still high. Overall, the transparent solar cells had an efficiency of around 50% that of non-transparent metal-based solar cells (7.8%).

However, there are few reports on SWNT films as a top electrode because SWNT lamination is difficult from above [26][27][28] . Li et al. 26 used SWNT films as the top electrode in perovskite solar cells. However, their SWNT films could not be doped, because it is difficult to dope top-laminated SWNT electrodes without damaging the device.
Here, we report SWNT-based metal-free OSCs with window-like transparency. The SWNT films were p-doped with two dopants, HNO 3 via sandwich transfer and MoO 3 via bridge transfer. The HNO 3 -doped and MoO 3 -doped 60% transparent SWNT-laminated OSCs showed PCEs of 4.1% and 3.4%, respectively. Using 90% transparent SWNT films, which produced OSCs visually more similar to a window, resulted in PCEs of 3.7% and 3.1% for HNO 3 -doped and MoO 3 -doped, respectively, whereas the reference ITO-based OSC showed a PCE of 7.8%. Our transparent OSCs, which are suitable for window applications, were fabricated by doping through direct lamination and dry lamination of SWNT films for the top electrode (Fig. 1). The double-sided light response of these transparent yet highly efficient solar cells offers advantages in many applications. We expect that the methods presented here will pave the way to future multifunctional OSCs.

Results and Discussion
Aerosol single-walled carbon nanotubes. Randomly oriented SWNT networks with high purity and long nanotube bundle lengths were synthesized by the aerosol chemical vapor deposition (CVD) method 23,29 . Floating catalyst aerosol CVD was performed in a scaled-up reaction tube with a diameter of 150 mm. The dry-deposited SWNT networks showed high purity, as confirmed by clear Van Hove peaks in the UV-vis spectra and the low intensity of the defect-derived D band in Raman spectra 30 . Furthermore, because the process required no sonication-based dispersion step, the resulting SWNT network consisted of exceptionally long SWNTs. Facile transferability is another advantage of the aerosol SWNT films. Once deposited from the aerosol, the CNTs showed strong tube-to-tube interactions and assembled into a freestanding thin film. The SWNT films were easily peeled off from a nitrocellulose film with a pair of tweezers and transferred onto other substrates for device fabrication.
Architectures of the solar cell devices. The structures of SWNT-based transparent OSCs are shown in Fig. 2. Figure 2a shows the conventional inverted OSC structure where Ag was used as an anode. This entails metal deposition and produces a non-transparent device. Figure 2b shows the same structure, except that Ag has been replaced with a highly transparent aerosol SWNT film. The light can be shone from either the ITO or SWNT side, or both sides, to generate photo-induced power. Thus, we refer to it as a window-like transparent OSC. The conductivity and transparency of SWNTs must be enhanced by doping to produce efficient solar cells. Figure 2c

Performance of SWNT-laminated transparent organic solar cells. Photoluminescence quenching
can be used to measure charge extraction ability 31 . A PTB7-based organic photoactive layer was deposited on a glass substrate and an SWNT film was top-laminated. As shown in Figure S1, the spectrum of the organic photoactive layer was suppressed substantially when the SWNT film was placed on top of it. This indicates effective charge extraction and successful lamination.
OSCs were fabricated with 90% transparent SWNT films. The PCEs measured with light shining from the SWNT side, the ITO side, and the ITO side with a mirror reflecting from behind were different (Table 1: Devices B and C). When light was shone from the ITO side, a PCE of 2% was obtained, which was approximately twice the PCE when light was shone from the SWNT side (0.9%). These values were very low compared with the non-transparent conventional reference, Device A (7.8%), because the SWNT films were not doped. The UV-vis spectra ( Figure S2a) showed that ITO was more transparent than the 90% transparent SWNT film. The difference increased when we included the whole device, including the photoactive layer, ZnO, and MoO 3 . The increased difference in transmittance is surmised to be due to the internal surface reflection between the layers. When adjacent two layers have difference in refractive index, internal surface reflection occurs. Thus, the greater the difference, the greater the transmittance loss. In other words, shining light on the ITO side [glass(1. . This results in a higher PCE for Device C than for Device B because a larger number of photons are converted to a higher short-circuit current density (J SC ) in Device C. The incident photon-to-current efficiency (IPCE) was measured to confirm this. As expected, when the light was shone from the ITO side, more charges were extracted ( Figure S2b). The same behavior was observed from the doped SWNT-based devices as well.
Compared with Device B, Device C showed not only a higher J SC but also a higher open-circuit voltage (V OC ) and fill factor (FF). This is a typical characteristic of solar cells that can be described by the Shockley equation. In principle, it is related to logarithmic scaling of V OC with light intensity 32 . Therefore, Device C with a higher J SC will exhibit a higher V OC . Equation (1) shows that FF is also affected by V OC 33 . This is especially true in real solar cell devices, which show non-ideal diode behavior. Thus, low J SC can induce low V OC and FF.
The shunt resistance (R SH ) is particularly important in transparent OSCs because the light intensity is not sufficient. At low light intensity, both the bias point and the current of the solar cell devices decrease. This causes the equivalent resistance of the solar cell devices to approach R SH 34 . If the equivalent and shunt resistances are similar, the fraction of the total current flowing through the R SH will increase, and this may lead to the recombination of charges. Therefore, it is crucial that we have a device system with a sufficiently high R SH value to avoid recombination. The current-voltage (J-V) curves in Fig. 3a,b show that 90% transparent SWNT OSCs possess sufficiently high R SH regardless of the light direction.
In conventional OSCs, metal electrodes can act as a rear reflector to direct unabsorbed light back to the photoactive layer. This provides a higher photocurrent, especially in the wavelength region below 700 nm. However, for the transparent OSCs, because no light can be reflected back and the active material is not thick enough to absorb all the sunlight, much light passes through unabsorbed. When a silver reflector (mirror) was placed on the opposite side of the light source, the J SC increased from 6.5 to 8.6 mA/cm 2 ( Table 1: Device D; Fig. 3d). However, despite the increased light intensity, V OC and FF did not increase further. This reveals that the maximum V OC obtained by using pristine 90% transparent SWNTs is limited to around 0.66 V. This could be due to imperfect interface contact between SWNTs and MoO 3 35,36 as demonstrated by the unparalleled dark J-V curves and light J-V curves at high current density. In general, V OC is controlled by the difference between the highest occupied molecular orbital (HOMO) of a donor and the lowest unoccupied molecular orbital (LUMO) of an acceptor. Furthermore, the HOMO and LUMO are affected by the interfacial layers' Fermi levels and the electrodes' work functions 37 . Therefore, poor contact between the SWNTs and MoO 3 may have been the limiting factor for the V OC . The overall PCE improvement was only 0.4%. This suggests that the double-sided light response of the transparent OSCs leads to sufficient photon excitations and that using a reflector at the cost of losing the transparency is not desirable.

Doping methodologies for SWNT-laminated transparent solar cells. Although transparent OSCs
were achieved as shown above, SWNTs should be p-doped to improve the conductivity and transmittance to boost the PCE of the OSCs. Doping top-laminated SWNTs has not been reported because of the mechanical difficulty of doping. Unlike SWNTs on a glass substrate, doping top-laminated SWNTs damages the device underneath. Hence, in this work, we devised two methods for safely doping SWNTs with HNO 3 or MoO 3 . HNO 3 (aq) acid is an effective p-dopant 38 . Nevertheless, its high acidity makes it impossible to apply directly. When a drop of HNO 3 was applied to an SWNT laminated device, it percolated through the film and completely destroyed the organic materials underneath ( Figure S3). Therefore, doping was performed on the SWNT film separately first. Figure 4a shows how the HNO 3 sandwich transfer was performed. One drop of HNO 3 was applied to an SWNT film on a glass substrate followed by heating at 80 °C for 5 min. The SWNT film turned slightly reddish as the acid dried and this signified successful doping. A decrease in the Fermi level value from − 5.0 to − 6.0 eV by photoelectron yield spectroscopy (PYS) confirmed a successful p-doping. The HNO 3 -SWNT film was sandwiched onto a MoO 3 film on a partially fabricated OSC. UV resin was applied at the edges to reinforce the adhesion. A PCE of 3.7% was achieved with the light source positioned at the ITO side (Table 1: Device E). An increase in J SC and a reduction in series resistance (R S ) confirmed the improvement of the transparency and conductivity of the HNO 3 -doped SWNT OSC. The increase in V OC meant that the interfacial contact improved. We suggest this is because of the pressure applied to HNO 3 -SWNT OSC during the sandwich transfer.
Thermal MoO x doping of SWNTs is a more stable doping method than doping with HNO 3 despite its slightly lower effectiveness 39 . The method has been used in OSCs 40 , but it is not suitable for top-laminated SWNT films, because it requires high-temperature annealing above 300 °C. Thus, we propose a bridge transfer method (Fig. 4b).
An SWNT film was transferred onto a metal holder where the film was hung like a bridge ( Figure S4). A shadow mask was placed below the SWNT film to mask the electrode contact area. MoO 3 was deposited from below by vacuum thermal evaporation. The MoO 3 -SWNT film was then annealed at 400 °C together with the holder to boost the doping effect and reduce the film to MoO x -SWNT, where x is between 2 and 3 39 . A decrease in the Fermi level value from − 5.0 to − 5.6 eV by photoelectron yield spectroscopy (PYS) confirmed a successful p-doping. MoO x -SWNT was gently laminated by using the holder on a partially fabricated device, where the MoO 3 film was not deposited because the MoO x on the SWNTs functioned as both a dopant and electron-transporting layer. A PCE of 3.1% was recorded for this device (Table 1: Device F). J SC was lower and R S was higher than those of Device E because MoO x thermal doping was less effective than HNO 3 doping. Importantly, because the SWNT film was hung precariously on the metal holder, extra caution was necessary during handling. Any small external impact or draft strong enough to crumple the SWNT film created microwrinkles, which were invisible to the naked eye, but were detected by atomic force microscopy (AFM; Figure S5). The lower V OC in this device may have been caused by the remnants of microwrinkles undermining the interface. Furthermore, pressure was not applied during the lamination of the SWNT film unlike the sandwich transfer method. Compared with the HNO 3 -SWNT sandwich transfer, the bridge transfer method had lower reproducibility because it is a sensitive process.
Despite high PCEs, both doping methods suffered from instability in the J-V sweeps (Figure 3e,f). We ascribe this to the mechanical variability of the fabrication methods, namely excess pressure applied to the SWNT film during the HNO 3 -SWNT sandwich transfer, and the sensitivity of the MoO x -SWNT bridge transfer method. However, if the processes are mechanically optimized, high efficiency and stability could be obtained.
Application of thicker SWNT films. Thicker SWNT films possess higher conductivity, although their transmittance is lower. By incorporating the thicker SWNT films (60% transparency at 550 nm wavelength), higher PCEs were obtained (Figures 5 and S6). The PCE of the HNO 3 -doped device was 4.1% (Table 1: Device G) and that of the MoO x -doped device was 3.4% (Table 1: Device H). Because of the higher conductivity of the 60% transparent SWNT films, the FF was higher than that of the 90% transparent SWNT-based devices by around 0.1. Interestingly, V OC of Device H was higher than expected. We attribute this to thicker SWNT films being less vulnerable to microwrinkle formation during the bridge transfer. Despite the lower transmittance of the films, Devices G and H displayed high J SC , because the main source of photons came from the ITO side not the SWNT side. Although it may seem obvious to use a thicker SWNT film to gain higher PCEs, it would compromise the transparency of the OSCs ( Figure S7). The improvement in PCE could be achieved only at the expense of the transparency.

Conclusions
In conclusion, undoped SWNT (90% transmittance)-based MoO 3 /PTB7:PC 71 BM:DIO/ZnO/ITO transparent OSCs showed a PCE of 1.8%. The aerosol-synthesized SWNT electrode, which was laminated from above as a top electrode, was easy to fabricate, chemically stable, electrically compatible, and mechanically resilient. Applying p-doping to the SWNT film through our novel HNO 3 sandwich transfer and MoO 3 bridge transfer methods, the  PCEs of the transparent OSCs increased to 3.7% and 3.1%, respectively. An even higher PCE of 4.1% was obtained at the expense of transparency by incorporating thicker SWNT films. By replacing the metal electrodes, these OSCs were inexpensive, had window-like transparency, and were visually glare-free. This research demonstrated the promising potential in window solar cell applications and flexible tandem OSCs.

Experimental Methods
Aerosol SWNT Preparation. SWNTs were synthesized by an aerosol (floating catalyst) CVD method based on ferrocene vapor decomposition in a CO atmosphere. The catalyst precursor was vaporized by passing room-temperature temperature CO through a cartridge filled with ferrocene powder. The flow containing the ferrocene vapor was then introduced into the high-temperature zone of a ceramic tube reactor through a water-cooled probe and mixed with additional CO. To obtain stable SWNT growth, a controlled amount of CO 2 was added together with the carbon source (CO). SWNTs were directly collected downstream of the reactor by filtering the flow through a nitrocellulose or silver membrane filter (HAWP, Millipore Corp., USA; 0.45 μ m pore diameter).
Device Fabrication. For the reference device, ITO substrates 15 × 15 mm in size with an active area of 3 × 3 mm and a sheet resistance of 9 Ω/square (Techno Print Co., Ltd.) were sonicated in cleaning surfactant (Semi Clean, M-Lo), water, acetone, and 2-isopropanol for 15 min each. The substrates were then dried with a nitrogen gun. ITO substrates were exposed to UV/O 3 for 30 min to remove any remaining organic impurities. They were transferred to a nitrogen-filled glovebox for further fabrication. ZnO sol-gel films were prepared by the method reported by Heeger et al. 41

Transfer of HNO 3 -doped SWNT Films by the Sandwich Transfer Method. The SWNT film was
transferred to a bare glass substrate. HNO 3 (70% in water) was applied dropwise and dried at 80 °C to p-dope the SWNT films. The HNO 3 -doped SWNT substrates were sandwiched onto a MoO 3 and Ag-patterned device (MoO 3 /PTB7:PC 71 BM:DIO/ZnO/ITO) and UV resin was applied at the edges to hold the two substrates and encapsulate the device.

Transfer of MoO x -doped SWNT Films by the Bridge Transfer Method.
A special holder for SWNT films was prepared. An SWNT film was transferred onto the holder so that the film was hung like a bridge. A 15-nm-thick MoO 3 layer was thermally deposited on the bridged SWNT film followed by thermal annealing at 300 °C for 3 h anaerobically to induce MoO x doping. The bridged SWNT film was transferred carefully to the Ag-patterned photoactive layer. A drop of PEDOT:PSS was applied and it was spin coated at 4500 rpm for 60 s to assist lamination. Because MoO x also functions as the hole-transporting layer, the MoO 3 step was omitted in this method. In other words, SWNT/MoO x was laminated on to PTB7:PC 71 BM:DIO/ZnO/ITO rather than MoO 3 / PTB7:PC 71 BM:DIO/ZnO/ITO. Characterization. J-V characteristics were measured by a software-controlled source meter (2400, Keithley) in the dark and under 1 sun AM 1.5G simulated sunlight irradiation (100 mW/cm 2 ) by using a solar simulator (EMS-35AAA, Ushio Spax Inc.), which was calibrated with a silicon diode (BS-520BK, Bunkokeiki). Topographic images were recorded by AFM (SPI3800N, SII) operating in tapping mode. The devices were also characterized by scanning electron microscopy (S-4800, Hitachi), Raman microscopy (inVia, Renishaw), and UV-vis-NIR spectroscopy (UV-3150, Shimadzu). Fermi levels were measured by Riken Keiki PYS-A AC-2 and kelvin probe S spectroscopy in air (ESA). They were calibrated by Au before the measurement.