Insights into chemical doping to engineer the carbon nanotube/silicon photovoltaic heterojunction interface † 2017 J. Mater. Chem. A ,

Graphene oxide/single-wall carbon nanotube (GOCNT) hybrid ﬁ lms have been used to fabricate heterojunction solar cells with silicon (Si) due to their compatibility with both aqueous and organic processing. In these cells GOCNT ﬁ lms are required to be both highly transparent and conducting. Di ﬀ erent approaches are used to improve these optoelectronic properties of the GOCNT ﬁ lms, including hybridization with silver nanowires (AgNWs) and p-type doping with CuCl 2 , AuCl 3 , SOCl 2 , HCl, H 2 SO 4 , HNO 3 and HClO 4 . UV-vis-NIR absorbance, Raman spectroscopy, and the sheet resistance of the ﬁ lms were used to evaluate the properties of the treated ﬁ lms and quantify doping. The most e ﬀ ective way to improve the optoelectronic properties of the GOCNT ﬁ lms was the incorporation of AgNWs which improved the ﬁ gure of merit (FOM, the ratio of transparency and conductivity) by over 600%. However, GOCNT/Si heterojunction photovoltaic devices with HNO 3 doped GOCNT ﬁ lms showed the highest solar photocurrent conversion e ﬃ ciency (11.38 (cid:1) 0.26%). In terms of stability, CuCl 2 and HCl doped ﬁ lms have the best electrode FOM stability, and devices made with such ﬁ lms have the most stable e ﬃ ciency as well. This report suggests that the electronegativity of the active elements in the dopants has a strong in ﬂ uence on the optoelectronic properties of the ﬁ lms as well as the solar cell performance.


Introduction
Carbon nanomaterial based transparent conducting lms are a potential alternative to replace the currently used transparent conductive oxides (such as indium doped tin oxide, ITO) which present several drawbacks 1 including scarcity of indium, 2 their brittle structure, 3 instability in acid environments and high production costs. 4 Carbon nanotubes (CNTs) are one example of a carbon nanomaterial famous for their unique optical and excellent electrical properties which enable the CNT-based thin lms to be transparent and conductive at the same time. [5][6][7][8][9][10][11] However, the hydrophobicity and the strong van der Waals interaction between individual nanotubes limit their compatibility with aqueous processing. [12][13][14][15] Different dispersing agents which have both hydrophobic and hydrophilic moieties, such as polymers, 16 porphyrins, 17 cellulose derivatives, 18 surfactants 19 and single stranded DNA sequences, 20 are commonly used to address this issue, but the liquid processing of the CNT lms prepared from these aqueous dispersions oen does not yield lms with a high gure of merit (FOM). 21 Recently, graphene oxide (GO) has been shown to be able to both help to disperse CNTs in water and enhance the aqueous compatibility of the as-prepared lm due to its amphiphilic nature. [22][23][24][25] Such hybrid lms made with GO and CNTs (GOCNTs) have been used as transparent window electrodes and applied on silicon to fabricate graphene oxide carbon nanotube/silicon heterojunction solar cells (GOCNT/Si). 26 Briey, in a typical device formed with p-type GOCNT lms as the window electrode and an n-type silicon base, the silicon is the light absorbing layer which produces excitons. The resulting excitons are then separated into charge carriers under the inuence of the built-in potential at the interface of the heterojunction, as shown in Fig. 1(a). [27][28][29][30][31] Holes are transported through the GOCNT network acting as a window electrode and collected by the front metal electrode while electrons are collected at the back metal contact via the silicon layer. 21 The exact nature of the heterojunction is still not clear although it has been discussed thoroughly in the recent literature. [32][33][34][35][36][37] It can be regarded as a Schottky barrier, a metal-insulator-semiconductor or a p-n junction. In some cases, the thin silicon oxide layer between Si and GOCNTs improves the device performance via a reduction in the reverse saturation current. 38 In these solar cells, devices made with the as-deposited CNT lms typically show mediocre performance due to high sheet resistance and low native p-doping. Chemical doping of the CNT lm is normally performed to satisfy the practical requirements of the optoelectronic properties of CNT based window electrodes. 39 The general p-type dopants include HCl, 40 HNO 3 , 41 H 2 SO 4 , 42 SOCl 2 , 43 AuCl 3 , 44 and CuCl 2 . 37 The sheet resistance (R sheet ) of CNT based networks is dependent on the conductivity of individual CNTs as well as the contact resistance between CNTs. 1 Chemical doping results in the shis of the Fermi level, which can increase the density of charge carriers and reduce the Schottky barrier height, f B , between metallic and semiconducting species, and as a result, the optical absorption peak caused by the rst and second interband transitions in SWCNTs is suppressed or completely bleached (as shown in Fig. S1 †). 45 Recently, a bilayer structure of AgNWs/ SWCNTs has been used to improve the efficiency of SWCNT/Si devices by nearly two-fold (from 4.31 to 7.89%) and the main improvement was attributed to the dramatic improvement of the optoelectronic properties with the addition of the AgNW top layer which provides highly conducting paths between CNTs. 33,46 In this report, different p-type dopants as well as AgNWs are used to improve the optoelectronic properties of the CNT-based (GOCNT) transparent conducting lms. Bath doping, as shown in Fig. 1(b), was used due to the benet of the previously developed organic-aqueous transfer process making the lms available for bath processing. 21 This approach increases the doping period compared to simply dropping the chemicals on the lms and eliminates the effect of the chemicals on the metal coated Si substrates (for example, the acid might dissolve the Cr layer) at the same time. AgNWs, due to their limited dispersity in water, could not be processed this way and they were ltered with GOCNTs on lter papers. The treated transparent lms were then deposited on n-type Si to create GOCNT/Si heterojunction solar cells. Both the efficiency and the stability of the devices were studied and compared.

Preparation of stock solutions
GO was prepared following the improved synthesis based on Hummers' method reported by Marcano et al. with minor modications as shown in the ESI. † 47 The nal concentration of the GO aqueous dispersion was kept at 1 mg mL À1 . The GOCNT suspension was fabricated by following our previous report with the ideal mass ratio of GO : CNT to be 1 : 2.25 (this ratio produces lms that have excellent optoelectronic properties which can be used as transparent conducting electrodes) 26 and the details are shown in the ESI. † AgNWs were synthesized by following a typical polyol approach with slight modications with the UV-vis spectrum and SEM images shown in Fig. S2. † 48 The nal concentration of AgNWs in water is 0.2 mg mL À1 .
The redox colloidal solution (CuCl 2 (99.0%, Sigma-Aldrich)/ Cu(OH) 2 ), (denoted as CuCl 2 in the following section since Cu 2+ in CuCl 2 has been reported to be the active part for rapid p-type doping of SWCNTs by extracting electrons and being reduced to Cu 1+ while Cu 2+ hydroxide provides long-term doping stability) 37 was prepared by adding a 1 M NaOH (98.0%, Chem-Supply) aqueous solution into a 0.1 M CuCl 2 ethanol solution gradually. The volume ratio of these two solutions was kept at 1 : 2000.

Fabrication and treatment of transparent conducting hybrid lms
The preparation of GOCNT hybrid lms follows the procedure previously reported with slight modications. 21 In detail, 200 mL of the GOCNT initial suspension was diluted to 100 mL with deionized water for vacuum ltration. A 45 mm diameter mixed cellulose ester lter membrane (MCE, 0.45 mm, HAWP, Millipore, Australia) was used to collect GO and CNTs, and beneath the MCE membrane was a nitrocellulose membrane (25 nm, VSWP, Millipore, Australia) with four circular holes (0.49 cm 2 Fig. 1 (a) Schematic energy diagram of GOCNT/Si heterojunction solar cells; (b) schematic of the bath doping process. MCE is the mixed cellulose ester filter and the substrate can be a glass slide or Si substrate coated with metal electrode. The filter paper dissolves in acetone and the GOCNT electrodes are then transferred to a dopant bath to conduct chemical doping. The GOCNT film is picked up with the substrate after being transferred to the next acetone bath. The band gap of GOCNT is estimated from the S 11 peak position of the UV-vis-NIR spectra of the GOCNT films. 32 each) so that a faster ow rate in the areas of four holes resulting from the different pore sizes in these two membranes produced four identical GOCNT hybrid lms on the MCE during one ltration. Aer being dried under ambient conditions, each individual GOCNT hybrid lm on the lter membranes was cut with a pair of scissors and placed into a bath of acetone for 3 Â 30 min to remove the MCE. For the study of the doping effect, the free-oating GOCNT lms were transferred using a homemade Teon spoon from the acetone bath to the bath of a particular dopant, containing one of CuCl 2 , AuCl 3 , SOCl 2 , HCl, HNO 3 , H 2 SO 4 or HClO 4 . The doping duration was 30 min and then the lms were transferred back to a new, fresh bath of acetone and ready to be picked up with an asfabricated Si substrate to fabricate solar cells or with a glass slide to characterize the optoelectronic properties of the hybrid electrodes, T and R sheet . For a control experiment, the GOCNT lms were picked up right aer the removal of the MCE in the acetone bath.
In terms of the devices based on AgNWs, AgNWs-GOCNT lms were fabricated by ltering a mixture of 150 mL GOCNT initial suspension and 50 mL AgNW stock solution diluted to 100 mL with deionized water. The reduced amount of GOCNT solution was needed in order to keep the transmittance at 550 nm (T) the same as that of the lm without AgNWs. Aer dissolving the MCE in the rst acetone bath, the AgNWs-GOCNT lms were then picked up as described before.

Fabrication of solar cells
Solar cells were prepared from silicon wafers doped with phosphorus (n-type, 5-10 U cm, 525 mm thick with a 100 nm thermal oxide, ABC GmbH, Germany). UV photolithography was used to dene a grid pattern of the front metal electrode (gold/ chromium) in a clean room (Class 1000). 27 A positive photoresist (AZ1518, micro resist technology GmbH, Munich, Germany) was spin-coated on the wafer surface for 30 s at 3000 rpm before so baking for 50 s at 100 C on a hot plate. Aer the photoresist-coated wafer cooled down to room temperature on a stainless steel plate, a mask aligner (EVG 610, EVG) was used to dene the grid patterns with a photo mask by exposing the Si wafer to UV light (9.1 mW cm 2 ), aer which the wafer was immersed in a developer solution (AZ 726 MIF, AZ electronic Materials, GmbH, Munich, Germany) for 20 s to develop the photoresist, rinsed with copious amounts of deionized water to remove the residue of the photoresist and dried under a nitrogen ow. Then, post baking of the wafer was performed on a hot plate at 115 C for 50 s. The line spacing of the grid pattern was 0.5 mm and the line width was 0.01 mm, as shown in Fig. S3. † These grid lines were dened in a 3 Â 3 mm square with the nal active area to be 0.087 cm 2 . 48 devices were produced on each 4 inch wafer. A sputter coater (Quorumtech, Q300T-D) was used to deposit the front metal electrode (Au/Cr 90/5 nm, 95 nm in total) with the thickness monitored using a quartz crystal microbalance. Then, the substrate was immersed in acetone for 90 min before mild rubbing with ne cotton brush to remove residual photoresist. One droplet of aqueous buffered oxide etch (6 : 1 of 40% NH 4 F and 49% hydrouoric acid (HF), Sigma Aldrich, Australia) was placed on the top of the active area with the dened Au/Cr grid to remove the thermal oxide layer in between the grid lines (until the Si surface repels the droplet). The as-prepared Si substrates were then ready to pick up the untreated GOCNT, AgNWs-GOCNT and GOCNT lms treated with different dopants from the acetone bath. Details of the GOCNT lm pick up have been described in detail in our previous report. 21 Aer drying GOCNT lms with a very gentle nitrogen ow, a diamond pen was used to scratch the underside of the silicon to remove SiO 2 and create a conducting path from Si to the gallium indium eutectic (eGaIn) back contact. The devices were then xed onto stainless steel plates.

Characterization
The device performance was measured by using a custom Labview™ virtual instrument connected with a Keithley 2400 source-measure unit. The power intensity of the collimated xenon-arc light source with a lter of AM 1.5G was calibrated to 100 mW cm À2 at the device surface using a standard silicon cell (PV Measurements, NIST-traceable certication). The solar cells were tested under both light and dark conditions to evaluate the efficiency and estimate the diode properties. Ideality and J sat were estimated from previous reports and the details are shown in the ESI. † 27,49,50 The external quantum efficiency (EQE) of the devices was measured as a function of wavelength from 300 to 1100 nm by passing the chopped light from a xenon source through a monochromator.
The transparency of the GOCNT hybrid and AgNWs-GOCNT lms at 550 nm was determined by UV-vis-NIR spectroscopy (Lambda 950 PerkinElmer) with the background subtraction of a clean glass slide. The R sheet of the GOCNT lms was measured using a four-point probe in linear conguration (Keithlink). The optoelectronic properties of the hybrid lms could then be evaluated with a gure of merit (FOM) which is calculated by using eqn (1) and rearranged to eqn (2): where m 0 is the free space permeability (4p Â 10 À7 N A À2 ); 3 0 is the free space permittivity (8.854 Â 10 À12 C 2 N À1 m À2 ); and s OP (l) and s DC are the optical and electrical conductivity, respectively. [50][51][52] Raman spectra of the lms were obtained by using an AFM-Raman system (Nanonics-Horiba) using an excitation energy of 1.96 eV. X-ray photoelectron spectroscopy (XPS) spectra of pure CNT and GOCNT lms were collected at the so X-ray beamline of the Australian synchrotron with more details shown in the ESI. † 53

Hybridization/doping effect on the optoelectronic properties of GOCNT lms
In this study, in order to improve the optoelectronic properties of GOCNT lms for application in solar cells, the inuence of 8 different types of materials either by hybridization (AgNWs) or p-type doping (CuCl 2 , AuCl 3 , SOCl 2 , HCl, H 2 SO 4 , HNO 3 and HClO 4 ) was analyzed. The working mechanisms of the dopant and hybrid materials are summarized in Table 1. XPS analysis (Fig. S4 †) shows that the GOCNT C 1s spectrum closely resembles that of the CNT starting material (peak center: 284.4 eV) meaning that the amount of GO in the lm is in fact small. GO would be readily observable with another strong C 1s peak centered at 287.5-280 eV contributed by C-OH, C]O and O] C-OH. 54 The small amount of GO in the lms means that the inuence of the p-type doping must be mainly on CNTs rather than GO.
The optoelectronic properties of the prepared and modied GOCNT lms (on microscope slides) were studied by UV-vis-NIR absorbance and sheet resistance. As shown in Fig. 2(a), all of the GOCNT lms have a very similar absorbance at 550 nm and thus the visible transmittance of these lms can be considered to be essentially the same (80%). In addition, the untreated GOCNT lm shows two broad optical absorption peaks at about 1945 nm (S 11 ) (shown in the inset of Fig. 2(a)) and 1010 nm (S 22 ), which are consistent with those of large diameter arc discharge SWCNTs. 58 The S 11 peak remains and a wide absorption feature above 1000 nm is observed aer the hybridization of GOCNTs with AgNWs, which indicates the existence of AgNWs in the electrode and suggests they may not have a p-type doping effect on CNTs. 59 In contrast, both S 11 and S 22 peaks are suppressed to various extents aer GOCNT lms are treated with p-type dopants (the degree of the S 11 and S 22 suppression: HNO 3 z H 2 SO 4 z SOCl 2 z AuCl 3 z CuCl 2 > HCl, as shown in Fig. S6 †), 60 which is a result of the shi of the Fermi level of CNTs into the valence band with electrons transferred from CNTs to dopants. 61 As shown in Fig. S1, † a lower population of electrons in the valence band aer p-doping is responsible for the suppressed S 11 peak. Further evidence of doping is obtained from the Raman spectra, as shown in Fig. 2(b), the G/D ratios of all the lms are very close, which indicates that both hybridization and p-type doping do not introduce structural defects into the CNTs. 62 The untreated lm has an evident metallic contribution for the G band on the low wavenumber side of the G peak (Breit-Wigner-Fano (BWF) peak), which is due to the strong interactions between electrons and phonons in metallic CNTs (Fig. 2(b) and S7 †). 63,64 GOCNT lms treated with p-type dopants show a reduction in the metallic contribution in the G band (the reduction of the BWF intensity is more evident in lms treated with dopants of higher electronegativity, as shown in Fig. 2(c and d), see Fig. S7 † for the peak tting of the Raman spectra) with a blue shi of the G peak position to different extents. The G band position of lms treated with dopants with active elements of higher electronegativity seems to shi further, as shown in (Fig. 2(c and e)), which infers CNT stiffening leading to higher phonon energies aer electron transfer from CNTs to dopants. 65,66 However, lms hybridized with AgNWs behave in a different way compared to the p-typed dopants. They show a broader feature in the BWF peak range and there is a red shi in the G band position with a similar BWF intensity to that of the untreated lm, which suggests that AgNWs have a different working mechanism in GOCNT lms. Table 1 Working mechanisms of the materials used for hybridization and p-type doping to improve the optoelectronic properties of GOCNT films. The sum of the electronegativity of the active atoms (bold) in these chemicals is calculated based on the Pauling scale. 55 The SEM images of the GOCNT films before and after different treatments are shown in Fig. S5 Chemical names (sum of electronegativities) Working mechanisms AgNWs (1.93) The hybridized AgNWs act as the additional charge transport channels to bridge the less conductive GOCNT network which is regarded as a mechanical support 33 CuCl 2 (8.22) Instant electron transfer from CNTs to Cu 2+ in CuCl 2 which is reduced to Cu 1+ while the less active Cu 2+ in Cu(OH) 2 provides a persistent charge transfer effect on the CNTs. 37 In addition, it is also possible that CNT-Cl is formed by the doping in which Cl attracts electrons from CNTs 44 AuCl 3 (12.02) The doping mechanism is not clear. Some reports suggest the reduction of cationic Au 3+ to Au nanoparticles causes p-type doping while some attribute the doping effect to the formation of CNT-Cl where electron transfer from CNT to Cl takes place 44 SOCl 2 (6.32) The doping effect is caused by the decomposition of the molecules, 2SOCl 2 + 4e À / S + SO 2 + 4 Cl À , which naturally withdraw electrons from CNTs 43 due to adsorption of SO 2 and possibly production of CNT-Cl H 2 SO 4 (16.54) The intercalated SO 3 À /SO 4 2À withdraws electrons from CNTs and shis the Fermi level with the formation of more holes in the valence band. In addition, it could also remove some polymeric and amphipathic dispersant and therefore results in the reduction in the contact resistance between individual CNTs 56 HNO 3 (13.46) Electrons are transferred from the surface of CNTs to the physisorbed molecules with NO 3 À groups, which causes the shi of the Fermi level to the valence band. A better contact is also created by the removal of impurities, such as metal catalysts and amorphous carbon 56,57 HCl (3.44) The mechanism is similar to those of HNO 3  The inuence of the hybridization and p-type doping on the optoelectronic properties is further studied with measured R sheet and FOM (calculated by eqn (1) and (2)) versus the sum of electronegativity of the molecules or species acting on the CNTs, as shown in Fig. 3(a and b). Among all of the lms, the untreated lm has the highest R sheet with the lowest FOM and the AgNW hybridized GOCNT lm has the lowest R sheet (120 U sq À1 (AgNWs-GOCNT hybrids) versus above 350 U sq À1 (pdoped GOCNT lms)) with the highest FOM (12 (AgNWs-GOCNT hybrids) versus less than 4 (p-doped GOCNT lms)), which indicates that the optoelectronic properties of the GOCNT lm are signicantly improved aer the hybridization of AgNWs. Compared to the FOM value (above 90) in the previous literature, our value for AgNW hybridization is much lower, which might be due to the fact that the lm in the literature had a much higher content of AgNWs (only 6 wt% SWCNT and the absorption peaks of SWCNTs were not observed in the UV-vis-NIR spectrum for the hybrid at all) 69 than the lms in this study. In terms of the p-type doped GOCNT lms, it seems that the R sheet /FOM decreases/increases exponentially with the sum of electronegativity of active elements in these dopants. Since the p-type doping is realized by electron transfer from the CNTs to adsorbed dopant molecules, it is not surprising that molecules containing more and/or higher electronegativity atoms have better ability to withdraw electrons. In order to test the validity of this relationship, we decided to use  a dopant which had not previously been used in the literature to determine if it followed a similar trend. Doping with HClO 4 (electronegativity sum ¼ 16.92) was performed and the R sheet and FOM seem to t well in the correlation found before, as shown in Fig. 3(a and b) (the star). In addition, both the bleaching of S 11 in the UV-vis-NIR and the upshiing of G band in the Raman spectrum revealed the p-doping nature of HClO 4 treatment (Fig. 2).
Overall, hybridization of GOCNTs with AgNWs is the most effective way to enhance the optoelectronic properties of the lms, and materials containing active atoms with higher electronegativity values are better dopants than those with lower values. Since the working mechanism of AgNWs (creating a more conductive metallic path between less conductive GOCNT networks, as shown in Fig. S5 †) is different from that of p-type dopants, it does not t the trend found in this study. We used a measure of electronegativity as opposed to (for example) redox potentials because in many cases there is more than one mechanism of action suggested and we used the sum of electronegativity to try to take into account all means of interaction. Fig. 4(a and b) show the J-V curves of GOCNT/Si devices with GOCNTs treated with AgNWs and different p-type dopants under AM 1.5 solar simulated light and dark conditions and the detailed performance parameters are shown in Fig. 4(c-l) with the details listed in Tables S1 and S2. † As shown in Fig. 4(c), the efficiency increases with the electronegativity for devices with ptype doped GOCNT lms. Interestingly, devices with GOCNT lms with the best optoelectronic properties (AgNWs-GOCNT) do not have the highest efficiency. For more details, the values of short circuit current density (J SC ) for all devices are very similar (the average values are within the range from 26 to 27 mA cm À2 ), which is further supported by the fact that the EQE curves all have the same shape and a similar signal intensity, as shown in Fig. 4(d and e). Both the open circuit voltage (V OC ) and ll factor (FF) generally increase with the electronegativity (Fig. 4(f and g)). Thus, the trend in efficiency might be due to the combined effect of V OC and FF. The inconsistency in the V OC of AgNWs and AuCl 3 treated devices (relatively larger error bars compared to those of other devices) is possibly caused by non-ideal short-circuits caused by the metal nanowires/nanoparticles bridging the front metal electrode (Au/Cr) to the Si layer, which has been observed previously in other reports. 21,33 The properties of the junction were further studied by analysis of dark J-V curves. The shunt resistance (R shunt ) of the interfaces increases with the electronegativity while the series resistance (R series ) has a decreasing trend, as shown in Fig. 4(h and i). In terms of the diode properties, there is no clear trend in ideality, J sat and f B due to the relatively large error bars related to the values (Fig. 4(j-l)). However, the general observation is that devices with untreated GOCNT lms have poorer diode performance (the highest ideality and J sat with the lowest f B ) while solar cells whose lms were treated with dopants with a high total electronegativity, such as HNO 3 and H 2 SO 4 , have excellent diode properties (with ideality approaching 1 and increasing f B ).

Performance of solar cells
Similarly, when investigating optoelectronic properties, in order to test the validity of the correlation between the efficiency and the sum of the active element electronegativity in the dopants, devices with HClO 4 treated top electrodes were fabricated and the results t well as shown in Fig. 4(c). In addition, most of the other parameters of the HClO 4 related devices, including J SC , V OC , FF, R shunt , R series , ideality, J sat , and f B , t the general trend as other dopants, as shown in Fig. 4. The trend in V OC , FF, ideality and f B infers that a better junction has been created between Si and the GOCNT lm aer different treatments with the improvement in the optoelectronic properties as well as the formation of a better contact.
Overall, the devices treated with dopants with higher electronegativity values for the active elements have better efficiency (HNO 3 treated devices have the highest efficiency among them, 11.38 AE 0.26%) as well as the best diode properties while the solar cells fabricated with the lms of the best optoelectronic properties (AgNWs-GOCNT) have efficiency just higher than that of control devices (8.14 AE 0.27% versus 7.11 AE 0.35%). This suggests that the performance of the GOCNT/Si heterojunction solar cells is not solely dependent on the optoelectronic properties of the transparent conducting window electrodes. Since AgNWs can signicantly enhance the charge transport through the GOCNT network while the improvement in the FOM for ptype doped GOCNT lms is mainly due to the increased density of free charge carriers, 65 it is inferred that the density of free charge carriers is the most important parameter in improving the solar cell performance. Furthermore, the role of the GOCNT network is more than that of a window electrode to transport the separated holes. More importantly, it contributes to the separation of the excitons by forming a p-n or Schottky junction between Si.

Stability of the treatment and solar cell performance
Since one of the issues for the industrial application of GOCNT lms is the instability of the effects of the treatment, normalized FOM and the efficiency of the solar cells as a function of time were studied, as shown in Fig. 5 and S8. † As shown in Fig. 5(a), the optoelectronic properties of the untreated lms are very stable over 10 days while GOCNT lms treated with AgNWs, HNO 3 and AuCl 3 are the most unstable (retaining about 70% of the original value aer 10 days). The FOM degradation of the AgNW hybridized GOCNTs is caused by the oxidation of the metal nanowire, 70 while the severe desorption of physisorbed NO 2 and HNO 3 molecules 57 and the aggregation of Au nanoparticles 71 as well as the desorption of Cl À which may react with air are the potential reasons for the FOM degradation of GOCNT lms treated with HNO 3 and AuCl 3 . 44 The doping stability of SOCl 2 is better than those of the aforementioned three but worse than that of CuCl 2 . The better stability of CuCl 2 treated GOCNT lms is due to the Cu 2+ hydroxide in the redox dopant, which can persistently withdraw electrons from CNTs over the long term. 37 Though HCl seems to be a very stable dopant as well, it has the least effective doping effect on GOCNTs. H 2 SO 4 doped GOCNT lms are quite stable for the rst few days and then degrade rapidly, which is due to the desorption of HSO 4 À and H 2 SO 4 molecules and is consistent with previous studies. 56 The degradation of the solar cell performance is shown in Fig. 5(b) and the normalized efficiency divided by FOM is plotted in Fig. 5(c) to study the inuence of electrode properties on the solar cell performance. The efficiency decay is mainly caused by a combination of the degradation of the optoelectronic properties of GOCNT lms and the growth of the insulating oxide layers (SiO x ) between Si and the GOCNT electrode. Aer 10 days, the untreated devices have an efficiency of about 75% of their original state (this is similar to previous reports), 21 which is solely due to the growth of the oxide layer. Solar cells with GOCNT lms doped with CuCl 2 , SOCl 2 and HCl (the three most stable dopants) have very similar stability to the control devices while the efficiencies of AgNW, HNO 3 , AuCl 3 and HClO 4 (the least stable four dopants) treated devices degrade to about 65% of their starting state. This suggests that the main reason for the degradation might be the growth of SiO x rather than the degradation of the GOCNT optoelectronic properties, which is further evidenced by the fact that the value of the normalized efficiency divided by FOM for most of the treated devices is very similar (about 0.8) and slightly higher than that of the untreated devices (about 0.75) aer 10 days (with the exception of H 2 SO 4 treated devices), as shown in Fig. 5(c). Interestingly, H 2 SO 4 doped devices degrade to about 35% efficiency of its starting point within 5 days but the properties of GOCNT lms are stable for the rst 5 days. In addition, the value of the normalized efficiency divided by FOM for such devices is much lower than that of untreated samples. Both facts suggest that the oxidation rate of Si might be faster for the devices with the H 2 SO 4 treated electrodes due to the presence of oxidizing species (HSO 4 À and H 2 SO 4 ). Different doping strategies have been widely explored and applied to CNT based transparent conducting lms in order to improve the optoelectronic properties. As shown in this report, the FOM of GOCNT electrodes is improved using various dopants and the resulting solar cells show improved performance compared to the control devices. The bath doping approach shows excellent compatibility in the processing in terms of dopants, doping periods and the limited adverse effect on the nal devices. Based on the results in this report, an ideal dopant would contain atoms of high electronegativity. However, the instability of the doping as well as the devices is still a concern which must be addressed in any commercial development.

Conclusion
The optoelectronic properties of GOCNT lms were improved with different doping approaches, including incorporation of AgNWs and p-type dopants (CuCl 2 , AuCl 3 , SOCl 2 , HCl, H 2 SO 4 , HNO 3 , and HClO 4 ), and these lms were applied in GOCNT/Si heterojunction solar cells to study both the performance and stability of the devices. Among these various agents, using AgNWs is the most effective approach to improve the FOM while dopants with a high total electronegativity of the active atoms, such as HNO 3 , H 2 SO 4 and HClO 4 , enhance the efficiency of the solar cells signicantly with the best performance recorded for HNO 3 doping (11.38 AE 0.26% vs. 7.11 AE 0.35% before doping). CuCl 2 and HCl have the best doping stability for the GOCNT lms and devices fabricated with such lms have the best performance stability.

Conflicts of interest
There are no conicts to declare.