Greatly enhanced hole collection of MoOx with top sub-10 nm thick silver films for gridless and flexible crystalline silicon heterojunction solar cells

Greatly enhanced hole collection of MoOx is demonstrated experimentally with a top sub-10 nm thick Ag film, allowing for an efficient dopant-free contacted crystalline silicon (c-Si) heterojunction solar cell without a front grid electrode. With the removal of shadows induced by the front grid electrode, the gridless solar cell with the MoOx/Ag hole-selective contact (HSC) shows an increment of ∼8% in its power conversion efficiency (PCE) due to the greatly improved short-circuit current density (Jsc) as well as the almost undiminished fill factor (FF) and open-circuit voltage (Voc), while the gridless solar cells with the conventional MoOx/ITO and pure MoOx HSCs exhibit ∼20% and ∼43% degradations in PCE due to the overwhelming decrease in their FF and Jsc, respectively. Through systematic characterizations and analyses, it is found that the ultrathin Ag film (more conductive than ITO) provides an additional channel for photogenerated holes to transport on MoOx, contributing to the great enhancement in the hole collection and the great suppression of the shunt loss in the gridless solar cells. A 50 μm thick gridless c-Si heterojunction solar cell with the MoOx/Ag HSC is 75% thinner but is 86% efficient compared to its 200 μm thick counterpart (while the 50 μm thick gridless solar cell with the MoOx/ITO HSC is much less efficient). It is over 82% efficient after being bent to a curvature radius as small as 4 mm, also showing superior mechanical flexibility to its counterpart with the MoOx/ITO HSC. Our MoOx/Ag double-layer HSC can be easily fabricated through thermal evaporation without breaking the vacuum, saving both the time and cost of the fabrication of the whole device. Therefore, this work provides a guide for the design of efficient HSCs for high-efficiency, low-cost, and flexible solar cells.


Introduction
Crystalline silicon (c-Si) solar cells have dominated the worldwide photovoltaic (PV) market for decades. To reduce the levelized cost of electricity, c-Si PV devices are required to constantly increase their power conversion efficiency (PCE) and simultaneously decrease their cost. Recently, remarkable progress has been made owing to advanced passivating contacts. [1][2][3] Based on polycrystalline silicon (poly-Si)/tunnel oxide and hydrogenated amorphous silicon (a-Si:H) passivating contacts, the PCE has reached and even surpassed 26%. [4][5][6] In these contacts, poly-Si and a-Si:H need to be either p-type or n-type doped to selectively collect holes or electrons. However, doping has its own drawbacks. Physically, both the short-circuit current density (J sc ) and open-circuit voltage (V oc ) will be negatively affected through dopant-induced Auger recombination, bandgap narrowing, surface/interface recombination, and freecarrier absorption. [7][8][9][10][11] Additionally, both poly-Si and a-Si:H are grown with inammable and explosive precursor gases. For poly-Si, high-temperature post-annealing (e.g., 900 C) 4 is required, which is bound to increase the fabrication cost.
In this work, a sub-10 nm thick Ag lm is employed on top of MoO x to form a double-layer HSC for efficient hole extraction from a c-Si heterojunction solar cell with LiF x /Al as the back ESC. Compared with previous reports, 36-38 the outer layer of TMO is not added because the MoO x /Ag double-layer HSC is simple and is already sufficient to demonstrate the excellent hole transport and collection ability of the metal-incorporated HSC. When the front grid electrode is removed, our solar cell with the MoO x /Ag HSC shows surprisingly improved PCE due to the enhanced J sc and almost undiminished FF and V oc ; while in great contrast, solar cells with the conventional MoO x /ITO HSC and pure MoO x HSC exhibit obvious degradations in PCE due to the overwhelming decrease in FF and J sc , respectively. Systematic characterization of the three HSCs has been conducted and analyzed. Finally, a 50 mm thick gridless c-Si heterojunction solar cell with the MoO x /Ag double-layer HSC is demonstrated, which is 75% thinner but retains 86% of the PCE of its 200 mm thick counterpart. It also demonstrates much better mechanical exibility than its counterpart with the MoO x /ITO HSC.

Experimental section
Fabrication of c-Si heterojunction solar cells N-Type (100) Czochralski double-side polished c-Si wafers (thickness: 200 mm; resistivity: 1-5 U cm) were employed to fabricate the c-Si heterojunction solar cells. Thin solar cells were fabricated on thin c-Si wafers, which were thinned from the 200 mm thick wafers via 50 wt% KOH etching at 90 C, as described in detail in our previous work. 49,53 Before device fabrication, the original or thinned wafers were thoroughly cleaned and immersed in a dilute 5 wt% hydrouoric (HF) acid aqueous solution for 60 s to remove the surface oxide generated during the cleaning processes. Immediately aer being dried in an oven at 90 C for 10 min, the wafers were transferred to the vacuum chamber of our thermal evaporator (VNANO VZZ-300S). Another thin oxide layer regrew on the surfaces of the c-Si wafer because air was in the oven. However, it was found that such a drying process was of great necessity to improve the interfacial qualities between c-Si and LiF x /MoO x for deposition, and thereby the power conversion performance of the device (Note S1, ESI †). The rear ESC with $2.1 nm thick LiF x and 200 nm thick Al was sequentially deposited without breaking the vacuum condition. All the three MoO x /Ag, MoO x /ITO, and pure MoO x HSCs included a $23 nm thick MoO x layer, which was thermally evaporated directly onto the front surface of the c-Si wafers. For the MoO x /Ag HSC, an ultrathin Ag lm was evaporated immediately aer the MoO x deposition without breaking the vacuum condition. The pressure was 1.2 Â 10 À4 Pa. Since the Ag lms to be deposited were no thicker than 10 nm, the deposition rate could not be too high, or the thickness would be difficult to control. A very low rate was also not preferred because it was unfavorable for the formation of continuous Ag lms under 10 nm in thickness. 54 Therefore, 0.7 nm s À1 was chosen for the deposition of our ultrathin Ag lms. For the MoO x /ITO HSC, a 55 nm thick ITO lm was sputtered onto the MoO x -coated c-Si wafer with a magnetron sputter (Kurt J. Lesker PVD75; DC power: 100 W; argon pressure: 3 mTorr; room temperature). Finally, 200 nm thick Ag electrodes were thermally deposited with either a grid or gridless shadow mask to produce grid or gridless solar cells with different HSCs.

Characterization
For the characterization of the c-Si solar cells, current densityvoltage (J-V) curves were recorded with a source meter (Keithley 2450) when the solar cells were illuminated with simulated AM1.5 G sunlight produced by a solar simulator (SAN-EI ELECTRIC AAA). The contact resistivity and sheet resistance of the HSCs on c-Si were measured using the TLM method, as schematically shown in the upper le inset of Fig. 2a and described in detail in Note S2, ESI. † The morphologies of the ultrathin Ag lms with different thicknesses and the crosssection of the thin c-Si wafer were inspected using SEM (Carl Zeiss Ultra 55). The transmittance spectra were recorded with an integrating sphere-based spectrometer for the HSCs deposited on glass and normalized to the spectrum of the glass. A four-probe system was employed to measure the sheet resistance of the ultrathin Ag lms with different thicknesses. Minority carrier lifetimes were measured with a quasi-steadystate photoconductance decay tester (Sinton WCT-120) and the samples were a piece of bare n-type Si wafer and three pieces of n-type Si wafers, which were covered with pure MoO x , MoO x / Ag, or MoO x /ITO HSCs on only one side of each wafer. It is a contactless method for generating carriers with light illumination and detecting the photocarriers through an induction coil embedded in the instrument under the sample. 55 The interfaces of c-Si/MoO x , MoO x /Ag and MoO x /ITO were observed with an FEI Tecnai G2 F20 S-TWIN TEM instrument. The TEM samples were fabricated by focused ion beam (FIB) milling (Carl Zeiss Quata 3D FEG). The original and etched wafer surfaces were inspected by scanning probe microscopy (SPM, MultiMode VEECO). For the mechanical exibility measurements, the thin solar cell was carefully attached to a piece of exible PET substrate, which was xed to two sample holders. One was xed and the other was moving, producing different curvature radii for the device on the PET substrate.

Results and discussion
Great carrier collection with few nanometer thick Ag lms on MoO x for gridless c-Si heterojunction solar cells  53,54 The ultrathin Ag lm is transparent (Fig. S3, ESI †), allowing sunlight to transmit through MoO x into the c-Si active layer. Importantly, the sub-10 nm Ag lm is able to greatly enhance the hole collection through the bottom MoO x lm. In order to demonstrate the great hole collection capability of the MoO x /Ag double-layer HSC, solar cells with top grid and gridless electrodes (with thicknesses of $100 nm, shown in the inset of Fig. 1a) were fabricated and compared. For further comparison, the hole collection capabilities of two other conventional HSCs, namely, the MoO x /ITO double-layer and the pure MoO x layer, were also demonstrated with their cell structures schematically shown in Fig. 1b and c, respectively. Here, the MoO x lms have the same thickness of $23 nm, while the ITO layer is $55 nm thick for antireection. For each kind of solar cell, at least three cells were fabricated to evaluate the fabrication repeatability. The light J-V curves were characterized and the averaged characteristic parameters are summarized in Table 1, where the characteristic parameters of the solar cell with the maximum power conversion efficiency (PCE) are presented below the average values. The light J-V curves of the champion solar cells with grid and gridless electrodes are plotted below the corresponding schematic diagrams, as shown in Fig. 1a-c, respectively.
From Fig. 1a-c, it can be seen that for the grid solar cells, both the MoO x /Ag and MoO x /ITO HSCs enable nearly square J-V curves with large shunt resistances and small series resistances, indicating high ll factors (FFs), in great contrast to the pure i.e., J sc,norm , V oc,norm , FF norm and PCE norm , calculated from the average values of the three kinds of solar cells without the grid electrodes to those of the counterparts with the grid electrodes. J sc,norm , V oc,norm , FF norm and PCE norm calculated from the values corresponding to the champion PCEs of the three kinds of solar cells are indicated by squares, triangles, and circles, respectively. MoO x HSC. Especially for the MoO x /Ag HSC grid cell, due to the good conductivity of the $8 nm Ag lm, the shunt resistance is maximal, leading to a maximum FF of up to 75.92%. However, due to the absorption of the ultrathin Ag lm, the transmittance of the MoO x /Ag HSC is lower than those of the MoO x /ITO and the pure MoO x HSCs (Fig. S3, ESI †). This results in a much lower short-circuit current density (J sc ) and thus in a degraded opencircuit voltage (V oc ). Therefore, the MoO x /Ag HSC grid solar cell cannot compete with the MoO x /ITO HSC grid solar cell (PCE $ 11.49%) but still performs better (PCE $ 6.16%) than the pure MoO x HSC grid solar cell (PCE $ 5.57%) because of the rather high FF.
When the inner grid is removed from the top electrode, the three solar cells behave differently ( Fig. 1a-c). In order to clearly see the changes, the characteristic parameters of the three gridless solar cells were normalized to those of the grid counterparts and are plotted in Fig. 1d. The normalized parameters for the champion solar cells are also indicated in Fig. 1d. Without the shadow of the grid, the illuminated area is increased by about 9% from 23 to 25 mm 2 . For the MoO x /Ag and MoO x /ITO HSC solar cells, the J sc values are improved by $12% and $16%, respectively; while for the pure MoO x HSC solar cell, the J sc is decreased by $55%. This indicates that the double-layer HSCs are much more effective in collecting holes, even those generated in the central illuminated area far from the electrode frame. In contrast, the great reduction in J sc for the pure MoO x HSC solar cell mainly results from the rather poor conductivity of MoO x . Despite the increased solar illumination and subsequently increased photocarriers, a large number of holes cannot be efficiently extracted by the electrode frame without the inner grid.
Due to the great reduction in J sc , the V oc is also degraded for the gridless solar cell with the pure MoO x HSC. For the solar cells with the MoO x /Ag and MoO x /ITO HSCs, because the increase in photocarriers and J sc with the absence of the optical shadows is inevitably accompanied by the increase in carrier recombination, the average V oc decreases a little ( Fig. 1d and Table 1). This can be veried by the smaller minority carrier lifetime for the MoO x /Ag-and MoO x /ITO-coated wafers than that for the pure MoO x -coated wafer (demonstrated in Fig. 3a).
Fortunately, the V oc degradation is very small for all the solar cells (less than 4%) and does not affect the PCE much. For the gridless MoO x /Ag HSC solar cell with the champion PCE, its V oc is even slightly higher than that of the counterpart with the inner grid electrode ( Fig. 1d and Table 1).
For the gridless solar cells with the MoO x /Ag and pure MoO x HSCs, the FFs remain almost unchanged compared with those of the grid cells, but the mechanisms are quite different. In the former, both the shunt and series resistances decrease when the grid is removed. This means that the conductive ultrathin Ag lm is favorable for collecting photogenerated holes, but the shunt loss increases without the help of the inner grid for conduction. The opposite behavior happens for the latter because of the less conductive MoO x without the grid, leading to increased shunt and series resistances (Table 1). For the gridless solar cell with the MoO x /ITO HSC, the balance between the hole collection and shunt loss is broken with greatly increased shunt channels. Thus, its FF is greatly reduced (Fig. 1d).
Based on all the above parameters, only the MoO x /Ag HSC can improve the cell PCE (by $8%) when the grid is removed. Great PCE degradations of $20% and $43% are observed when the conventional MoO x /ITO and MoO x HSCs are applied to the gridless solar cells. Unfortunately, the PCE of the gridless MoO x /Ag HSC solar cell is still inferior to that of the gridless MoO x /ITO HSC solar cell due to the much lower J sc , which is induced by the much lower optical transmissivity of the MoO x /Ag HSC (Fig. S3, ESI †). That being said, the MoO x /Ag HSC is more advantageous in fabrication through thermal evaporation without breaking the vacuum, enabling high-throughput and low-cost fabrication of the whole device. However, to fabricate the MoO x /ITO HSC, pumping twice is necessary for thermal evaporation and sputtering, respectively, which is time-consuming and cost-ineffective. The MoO x /Ag HSC is also more mechanically exible than the MoO x /ITO HSC, allowing for thin and highly exible c-Si solar cells as emerging wearable power sources (demonstrated below).

Comprehensive analyses of contact performance
In order to elaborate on the various above photovoltaic behaviors, we characterized the contact performances between the three different HSCs and the n-type c-Si with the transfer length method (TLM). 27 As schematically shown in the upper le inset of Fig. 2a, a series of current-voltage curves were measured through two Ag line pads on top of the HSCs with various pad spacings, which are plotted in Fig. 2a-c, respectively, for the MoO x /Ag, MoO x /ITO, and pure MoO x HSCs. All contacts exhibit very good ohmic features. In each panel, the total resistances were also plotted in the lower right inset as a function of the pad spacing and tted using a straight line, from which the sheet resistance, R sh_HSC , and the contact resistivity, r c_HSC , of the HSC were calculated and are plotted in Fig. 2d and e, respectively. Details can be found in Note S2, ESI. † Both the R sh_HSC and r c_HSC are drastically reduced when the MoO x lm is covered with an ultrathin Ag lm or an ITO lm and the MoO x / Ag HSC affords the smallest values of R sh_HSC and r c_HSC . For all the three HSCs, c-Si is in direct contact with MoO x , which has a much higher workfunction than c-Si. The Fermi level equilibration makes the conduction and valence bands of c-Si bent up near this contact interface, where electrons are blocked and holes are accumulated to form an inversion layer. Such band alignment is benecial for hole transportation and collection. Therefore, both the measured R sh_HSC and r c_HSC include the resistances of the whole HSC and the inversion layer. In order to clearly show how holes are transported from one Ag line pad to another, we present a schematic illustration in Fig. 2f, where there are three channels for hole transport at the c-Si/MoO x /Ag or c-Si/MoO x /ITO contacts. When current is injected from the le Ag line pad, part of the holes rst go vertically through the double-layers of the HSC and then ow horizontally in the inversion layer to the right Ag line pad; part of them go through the top Ag or the ITO layer and into the MoO x layer for horizontal transportation; and most holes will ow directly in the top Ag or the ITO layer because of their much better conductivity than MoO x or the inversion layer. Due to the absence of the last highly conductive channel in the pure c-Si/MoO x contact, both vertical and horizontal transportation will meet extremely high resistances. Therefore, for the pure MoO x HSC solar cell, the FF is rather low and a grid is required to collect photogenerated holes, which are otherwise likely to be lost by recombination. Since Ag is more conductive than ITO (to be compared in the next section), the photogenerated holes, which can be collected with lower resistance, can also transport a longer distance. Thus, even for the gridless MoO x /Ag HSC solar cell, the FF still remains very high and the holes are collected efficiently ( Fig. 1c and Table 1). Fig. 3a shows that the measured minority carrier lifetimes of the c-Si wafers covered with the above three HSCs are all higher than that of the individual c-Si wafer. This suggests the passivation effects of the MoO x layer, which is however weakened a little bit by the additional Ag or the ITO layer. As mentioned above, due to the large difference in workfunction, an inversion layer forms near the c-Si/MoO x interface, where holes are accumulated and electrons are blocked. Consequently, carrier recombination here is reduced, which is known as the eld-effect passivation of MoO x . 49,53 From the eldemission transmission electron microscopy (TEM) image in Fig. 3b, an interlayer can be clearly observed between MoO x and the n-type c-Si regardless of whether MoO x is covered with Ag or ITO. The interlayer thickness is $3.98 nm. As reported in our previous work, 49,53 this interlayer, containing Mo, O, and Si elements, is mainly due to the oxidization of the c-Si surface and the reduction of the MoO x lm. To conrm this point, in this work we explored the interfacial chemical state of the c-Si/MoO x interface through depth proling X-ray photoelectron The existence of the oxide interlayer at the MoO x /c-Si interface is also conrmed by the Mo 3d and Si 2p spectra. The Mo 3d spectrum shis to a lower binding energy, indicating the reduction of Mo 6+ cations on the MoO x side to Mo atoms on the c-Si side, while Si is partially oxidized and a mixture of substoichiometric species coexist in the interlayer. It is fully illustrated that the oxide interlayer is caused by the evaporated MoO x , which oxidizes the bottom c-Si surface, making it reduced. The oxide interlayer is able to chemically passivate the dangling bonds of the c-Si surface, further improving the minority carrier lifetime (Fig. 3a).
In contrast, no apparent interlayers appear at the MoO x /Ag and MoO x /ITO interfaces shown in Fig. 3c and d, respectively. The thermally evaporated $8 nm ultrathin Ag lm does not look very uniform (whose morphology is demonstrated below) and the interface seems not very at (Fig. 3c); while the interface between MoO x and the sputtered ITO appears at, and there is no apparent damage during sputtering (Fig. 3d). In order to further explore the interfacial variation at the MoO x /Ag interface, depth proling XPS was performed and is illustrated in Fig. 3f. On the Ag side, Ag-O bonds can hardly be distinguished from the O 1s spectrum. However, as the etching depth goes to the MoO x side, the Ag-O peak (529.7 eV) appears clearly, meaning that Ag is likely to be oxidized. From the Mo 3d spectra, it is seen that the main peaks move towards the low binding energy, indicating that part of the Mo 6+ ions are reduced to Mo 5+ with a signicant increase in the intensity on the MoO x side. For the Ag 3d spectra, the dominant peaks are rst related to Ag atoms in the outer layer and then become associated with Ag + cations in the inner layer. These spectral variations illustrate that the initially evaporated Ag is oxidized upon being deposited onto the MoO x lm. The remaining Ag atoms at the interface act as recombination centers, leading to a relatively lower carrier lifetime for the c-Si/MoO x /Ag sample compared to that for the c-Si/MoO x /ITO sample, as shown in Fig. 3a. Fortunately, such a lifetime decrease does not cause too much degradation in the V oc for the photovoltaic devices, as shown in Fig. 1 and Table 1.

Effects of Ag thickness on carrier collection in the MoO x /Ag HSC
As a consequence of physical vapor deposition technology, the thermal evaporation of metal lms inevitably suffers from the three-dimensional (3D) Volmer-Weber growth mode in the initial stage of deposition. 57-61 A continuous lm with low resistance cannot be achieved with thickness below 10 nm. A thick lm will however lead to rather low optical transparency. Fortunately, MoO x in this work can serve as a seed layer to mitigate the surface energy difference between the Ag to be deposited and the c-Si active layer. The 3D Volmer-Weber growth mode can thus be suppressed effectively, allowing a sub-10 nm Ag continuous lm to be achieved with both low sheet resistance and high transmittance.   . 4a shows the morphological evolution of our thermally evaporated Ag lms with different thicknesses on top of the MoO x -coated c-Si substrates, inspected with scanning electron microscopy (SEM). The thickness-dependent sheet resistances, R sh_lm , and optical transmittance spectra of identical Ag lms were characterized on MoO x -coated quartz substrates and are shown in Fig. 4b and S3, ESI, † respectively. All the MoO x coatings were $23 nm thick, the same as that applied in solar cells. As shown in Fig. 4a, dense Ag nanoparticles are distributed on the 4 nm thick lm. They are separated from each other. Therefore, the Ag lm is non-conductive and its R sh_lm is too large to measure. When the lm thickness increases to 6 nm, the small Ag nanoparticles become largely isolated nanoclusters. The R sh_lm is measurable but is quite large (Fig. 4b).
Upon light illumination, localized surface plasmons (LSPs) are excited in either nanoparticles or nanoclusters, 57-61 resulting in enhanced absorption and thus transmission dips around 650 nm (Fig. S3, ESI †). Further increasing the thickness to 8 nm and 10 nm makes the Ag nanoclusters coalesce into sparse and dense networks, respectively. Due to the increased electron transport paths, the R sh_lm drops dramatically to below 10 U sq À1 , much smaller than that of the $55 nm thick ITO (Fig. 4b).
Meanwhile, the LPR-induced transmission dips disappear with improved optical transmittance (Fig. S3, ESI †). Among all the ultrathin Ag lms, the 8 nm lm shows the best transparency (especially in the near-infrared wavelength range) with a sunlight weighted average transmittance of 55.2% but is much lower than those of the ITO or the uncoated pure MoO x lms. Fortunately, its rather low R sh_lm enables extremely good carrier collection performance in terms of the sheet resistance and contact resistivity of the HSC on c-Si, to be described below. The small error bars in the R sh_lm at various lm thicknesses show the very good repeatability of the deposition process.
As shown in Fig. 4c and d, both the sheet resistance, R sh_HSC , and contact resistivity, r c_HSC , of the MoO x /Ag HSC follow the trend of the R sh_lm (rather than the optical transparency), dependent on the Ag lm thickness (Fig. 4b). Both the R sh_HSC and r c_HSC drop dramatically when the Ag lm thickness changes from 6 to 8 nm. This strong correlation indicates the dominant role of the ultrathin Ag lm on the whole HSC regarding carrier collection. Compared with the pure MoO x HSC, the top Ag lm provides another channel to conduct holes. If it is continuous, e.g., in the 8 and 10 nm thick lms, the carrier collection can naturally be improved both vertically and horizontally; while for the discontinuous Ag lms with thicknesses below 8 nm, both the R sh_HSC and r c_HSC show no obvious improvement compared with the pure MoO x HSC ( Fig. 4c and d). Due to the large R sh_lm of ITO (Fig. 4b), the MoO x /ITO HSC shows inferior carrier collection ability in terms of its R sh_HSC (Fig. 4c) and r c_HSC (Fig. 4d) compared to the MoO x /Ag HSCs with Ag lm thicknesses of 8 and 10 nm, whose Ag lms are more conductive. It is also found that our MoO x / 8 nm Ag HSC owing to a quite low r c_HSC outperforms almost all the HSCs reported previously (Table S2, ESI †). Therefore, it is believed to be a quite good dopant-free contact for hole collection.   Table 2. In order to clearly show the inuence of the wafer thickness, the ratios of the characteristic parameters of the 50 mm thick cell to those of the 200 mm thick cell, i.e., J sc_50 /J sc_200 , V oc_50 /V oc_200 , FF 50 /FF 200 , and PCE 50 /PCE 200 , were calculated and are plotted in Fig. 5b, where the c-Si thickness ratio, t Si_50 / t Si_200 , is also indicated for comparison. When the c-Si thickness, t Si , is reduced by 75% from 200 mm to 50 mm, for the MoO x / Ag HSC solar cell, the PCE still retains 86% of the initial value of the 200 mm thick cell with degradation of merely 14%, resulting from the 7%, 4%, and 4% decreases in J sc , V oc , and FF, respectively (Fig. 5b). However, due to the larger reduction in J sc (13%) and FF (36%), the PCE decreases more dramatically by 45% for the MoO x /ITO HSC solar cell. Its absolute PCE (5.43%) was even slightly lower than that of the thin MoO x /Ag HSC solar cell (5.85%).
When the c-Si is thinned, there must be some photons not fully absorbed, especially those in the near-infrared wavelength range. Due to the highly reective LiF x /Al ESC, they will be re-ected towards the front HSC. Since the MoO x /ITO HSC is more transparent than the MoO x /Ag HSC, more light will be transmitted through the MoO x /ITO HSC into the air, leading to larger reectivity. In contrast, the MoO x /Ag HSC benets the light trapping 62 and thus the reduced reectivity of the thin cell ( Fig. S3 and S4, ESI †). Therefore, the J sc degradation for the MoO x /Ag HSC solar cell is smaller than that for the MoO x /ITO HSC solar cell (Fig. 5b).
Because of the very fast etching process, the c-Si surface roughness increased from 0.75 nm to 8.02 nm (Fig. 5c and d), which must lead to enhanced carrier recombination at the c-Si surfaces. Fortunately, owing to the good passivation effects of the HSC and ESC, as demonstrated in Fig. 3, in this work and in our previous work, 49 the V oc values of both solar cells with the  MoO x /Ag and MoO x /ITO HSCs are not affected much by the rough surface. The relatively larger V oc degradation for the MoO x /Ag HSC solar cell is likely due to the Ag atoms occurring at the MoO x /Ag interface as recombination centers, as discussed in Fig. 3. In contrast, the rough c-Si surface seems to affect the quality of the sputtered ITO signicantly, leading to a signicant increment in the series resistance and an apparent decrease in the shunt resistance of the device (Table 2). Therefore, FF for the MoO x /ITO HSC solar cell drops by 36%, much larger than the FF degradation of 4% for the MoO x /Ag HSC solar cell (Fig. 5b). This indicates the more excellent tolerance of the ultrathin Ag lm to the substrate surface to be deposited. Furthermore, a mechanical exibility test was conducted for the two kinds of 50 mm thick solar cells with the setup shown in the inset of Fig. 5e. The measured J-V curves of the solar cells under different curvature radii are plotted in Fig. S5, ESI. † To show the performance variation with the curvature radius, the characteristic parameters of the bent solar cells were normalized to those of the unbent cells and are plotted in Fig. 5e-h. It can be seen that all the characteristic parameters for the MoO x / Ag HSC solar cell are above 82% of those for the unbent counterpart, even when the bending radius is reduced to as small as 4 mm. There are no apparent cracks at the Ag surface (inset of Fig. 5g). In great contrast, all the normalized parameters for the MoO x /ITO HSC solar cell fail to remain higher than 80% but instead drop dramatically when the bending radius becomes smaller than 15 mm. We observed obvious cracks at the 55 nm thick ITO surface at a curvature radius of 10 mm (inset of Fig. 5g), which were the main reason for the failure of the MoO x / ITO HSC solar cell at small bending radii. Such a comparison fully illustrates the superior mechanical exibility of the 8 nm thick Ag lm, enabling exible c-Si solar cells as emerging wearable power sources.

Conclusions
In summary, greatly enhanced hole collection of MoO x has been demonstrated experimentally with a top sub-10 nm thick Ag lm, which enables an efficient undoped contacted c-Si heterojunction solar cell without a front grid electrode. Compared with solar cells with front grid electrodes, the gridless solar cell with the MoO x /Ag HSC shows an increment of $8% in PCE due to the shadow removal-induced improvement of the J sc as well as the almost-undiminished FF and V oc , while in great contrast, the gridless solar cells with the conventional MoO x /ITO HSC and pure MoO x HSC exhibit obvious degradations in PCE ($20% and $43%, respectively) due to the overwhelming decrease in their FF and J sc , respectively. Systematic characterizations of the three HSCs have been conducted. It is found that the more conductive ultrathin Ag lm (rather than ITO) provides an additional channel for photogenerated holes to transport more quickly on MoO x , which contributes to the great enhancement in hole collection and shunt loss suppression in the gridless solar cells. Our MoO x /Ag double-layer HSC can be easily fabricated through thermal evaporation without breaking the vacuum, saving both the time and cost of the fabrication of the whole device. Based on this HSC, a 50 mm thick gridless c-Si heterojunction solar cell is demonstrated, which is 75% thinner but retains 86% of the PCE of its 200 mm thick counterpart (while the 50 mm thick gridless solar cell with the MoO x /ITO HSC is much less efficient). It is over 82% efficient aer being bent to a curvature radius as small as 4 mm, showing much better mechanical exibility than its counterpart with the MoO x / ITO HSC. This work provides a guide for the design of highefficiency and low-cost solar cells, promising for applications as emerging wearable power sources.

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