Influence of a Gold Seed in Transparent V2Ox/Ag/V2Ox Selective Contacts for Dopant-Free Silicon Solar Cells

Dielectric/metal/dielectric structures based on vanadium pentoxide with a thin silver interlayer have been optimized to replace traditional transparent electrodes. As would be expected, there is a tradeoff in the metal thickness to achieve high transparency and low sheet resistance simultaneously. It has been demonstrated that an ultra-thin gold seed prevents the tendency of silver to form clusters. This wetting effect reduces the metal thickness needed to form a continuous film, which leads to a higher averaged transmittance and very low sheet resistance. On the other hand, vanadium pentoxide on silicon forms a high-quality hole-selective contact. Thus, these structures can be used as an all-in-one transparent electrode and selective contact for a new kind of heterojunction solar cells. This concept has been proved in a 13.3% efficient solar cell fabricated on n-type silicon wafers. Besides being dopant-free, the complete fabrication route did not require any sputtered transparent electrode.


Influence of a Gold Seed in Transparent
C ONCEPTUALLY, a solar cell consists of an absorber (semiconductor) sandwiched between two selective contacts, which act as filters for one type of charge-carrier [1]. The exclusive attribution of these filters is their high permeability for one charge-carrier type (electron/hole) while simultaneously blocking the other one. These selective contacts are usually referred to in the literature as electron transport layers and hole transport layers (HTLs), respectively. The use of selective contacts in solar cells was originally developed in emerging photovoltaic (PV) technologies, such as organic solar cells [2] or, more recently, in perovskites solar cells [3]. However, this novel approach also applies to well-established inorganic-based solar cells [4]. The application of this approach to crystalline silicon (c-Si) technology is particularly relevant since it now represents about 90% of the global market share [5]. Despite being the most developed and mature technology, efforts to tackle remaining issues are still ongoing. Carrier selectivity in silicon-based solar cells is regularly achieved by doping, which introduces technological complexity, i.e., conventional homojunction solar cells diffused at high-temperatures involve an energy-intensive fabrication process. Heterojunction solar cells can be processed at low temperatures, but demand hazardous gas precursors. On the other hand, some materials used in new PV technologies have demonstrated excellent charge-carrier selectivity. Thus, the use of such materials as substitutes for standard silicon dopants deserves particular study [6]. In this sense, the use of transition metal oxide (TMO) films on a c-Si absorber has been a success story. Thin films of different TMOs are really good HTLs; they allow easy hole transport while effectively blocking electrons. Power conversion efficiencies of 22.5% have been achieved on n-type c-Si devices with the use of molybdenum trioxide (MoO 3 ) as the hole-selective contact [7]. In addition, remarkable efficiencies that exceed 18% have been reported by using vanadium pentoxide (V 2 O 5 ) and tungsten trioxide [8]. This new kind of heterojunction solar cells still uses a transparent conductive oxide (TCO) electrode owing to the high sheet resistance of TMO layers. The high-vacuum sputtering process to deposit TCO layers tends to generally increase manufacturing costs. More importantly, commonly used tin-doped indium oxide (ITO) electrodes suffer from the high demand of the scarce indium element. Particularly, companies that manufacture high added-value flat-panel displays cause price hikes [9]. Therefore, indium-free alternatives to TCO electrodes would be desirable for PV applications.
Over the past few years, different TCO/metal/TCO multilayers have been investigated to replace ITO electrodes. These structures can achieve globally better optical and electrical properties compared with a single TCO layer [10]. Tin oxide and either intrinsic or aluminum-doped zinc oxide layers have been used as indium-free TCOs [10]- [12]. Silver is the most widely used metallic interlayer, although structures with copper 2156-3381 © 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications standards/publications/rights/index.html for more information.
and gold have also been reported [13], [14]. More recently, dielectric-metal-dielectric (DMD) stacks have been used in organic semiconductor devices [15], [16]. In DMD stacks, the TCO layers are substituted by semi-insulating films, which very frequently are TMO layers. The sheet resistance is kept at low values by the metallic layer, while the UV transparency can be even higher [17]. All these results point to the viability of using TMO-based DMD stacks in c-Si solar cells to fulfill two objectives. First, the bottom TMO layer in contact with silicon determines a strong hole-selective character. Second, the whole DMD structure would operate as an indium-free transparent electrode. This concept has already been implemented as a front MoO 3 /Ag/MoO 3 (MAM) selective contact for ntype c-Si solar cells [18]. Different V 2 O 5 /M/V 2 O 5 multilayers (M = Au, Ag, Ca) have also been incorporated in heterojunctionback-contact solar cells, although in that case the transparency was not a strong requirement [19].
In this paper, we study the optical and electrical properties of DMD structures composed of V 2 O 5 semi-insulating layers and silver as the intermediate metal. Particularly, a detailed analysis of V 2 O 5 /Ag/V 2 O 5 (VAgV) structures is presented in terms of their quality as a transparent electrode. Additionally, the beneficial effect of intercalating an ultra-thin gold seed is demonstrated Vanadium oxide was chosen over other TMO alternatives because it can provide good surface passivation with no interfacial buffer layer [20]. The study was completed with the implementation of optimized VAuAgV structures as a front hole-selective contact of complete n-type c-Si heterojunction solar cells. A detailed characterization of these devices is presented, which evidences a high potential for such a cost-effective new heterojunction structure.

II. EXPERIMENTAL
The V 2 O 5 /metal/V 2 O 5 (VMV) multilayers studied in this paper were deposited by thermal evaporation in a vacuum system with a base pressure below 10 −6 mbar. The films were grown at low deposition rates in the range of 1-4Å/s and the substrates were maintained at room temperature. The thickness of the evaporated layers was measured by means of an INFICON SQC-310 deposition controller. It must be considered that quite often very thin layers have not coalesced into homogeneous films. In that case, the thickness measured by the quartz microbalance must be interpreted just as the amount of material evaporated onto the substrate.
First, optical and electrical characterizations were done for VMV structures deposited onto borosilicate glass. These substrates had been previously cleaned in ultrasonic baths of acetone and isopropanol, followed by a thorough deionized water rinse and a nitrogen blow until dry. The VMV structures were examined by field-emission scanning electron microscopy (FE-SEM), with a focus on the morphology of ultra-thin metallic layers. The sheet resistance R sh was evaluated using a 4-point probe. The optical transmittance was measured using an UVvisible-near infrared response Shimadzu 3600 spectrophotometer. These measurements were complemented with calculations of spectral absorption in the silicon wafer, which were obtained by the transfer-matrix method (TMM) [21]. With regard to complete solar cells, these devices were fabricated on polished (100) n-type (2 Ω·cm) c-Si wafers of 280-μm thickness. The VMV structures developed in this paper were deposited on the front side to serve as a hole-selective contact as well as a transparent electrode. Since the VMV structures include ultra-thin layers, we preferred to maintain the c-Si surface as flat (non-textured). As a preliminary study, this will allow a more reliable analysis of the results. The low-temperature rear electrode consisted of a stack of aluminum and titanium oxides (Al 2 O 3 /TiO 2 ), followed by a magnesium buffer, and a final aluminum contact. Such a structure behaves as a good quality dopant-free electron selective contact, as reported in a previous work [22]. The active area of the solar cells (1 cm 2 , 4 cm 2 ) was defined by conventional photolithography and wet-etching steps. Finally, a 2-μm-thick Ag grid was thermally evaporated through a shadow mask as a front contact. The metallic grid covered 4% of the VMV active area. A schematic of the device structure is shown in Fig. 1. On complete devices, the current density versus voltage (JV) electrical characteristics were measured using a Newport solar simulator and Keithley 2400 dc source-meter. Additionally, quasi-steady-state open-circuit voltage (QSSVoc) measurements [23] were done with a system made in-house. Using this technique, the actual electrical characteristics can be compared with pseudo-JV curves with no influence of the series resistance. Finally, the external-quantumefficiency (EQE) curves of the solar cells were measured by means of a QEX10 PV measurement equipment.

III. RESULTS AND DISCUSSION
First, the influence of the metallic layer on the optical properties of VMV structures was investigated. For that purpose, different samples were deposited on glass substrates to measure their optical transmittance. The thicknesses of the bottom and top V 2 O 5 layers were fixed at 20 and 35 nm, respectively. As an intermediate metal, we started to study the use of silver layers (M = Ag) with a thickness between 6 and 12 nm. As can be seen in Fig. 2, the overall transmittance decreases with the thickness of the intermediate Ag layer. This behavior could be expected owing to the optical absorption by the metallic layer. However, the transmittance in the infrared region was unexpectedly low for very thin Ag layers. Besides, in the wavelength region of 700-900 nm, transparency was not in direct correlation to the metal thickness. The characteristic growth of very thin Ag layers might account for this phenomenon. As silver is evaporated onto a dielectric substrate, isolated islands tend to form for the first few nanometers (Volmer-Weber growth) [24]. The absorption caused by the formation of localized surface plasmons would explain the relatively low optical transmittance [25]. An equivalent behavior has been reported for other DMD structures, where silver was also used as an intermediate metal [26]. The initial formation of islands is explained because Ag atoms prefer to bond with each other rather than with the substrate. In this sense, the nominal thickness would only be a reference value for the amount of evaporated metal. A successful approach to promote the formation of a continuous film (Frank-Van der Merwe growth) is to use a seed (wetting) layer [27], [28]. In this paper, our choice was to evaporate an ultrathin (1 nm) gold seed preceding the Ag layer. The dissociation energy of the Au-Ag bond (202.9 kJ/mol) is higher than that of a diatomic Ag-Ag bond (160.3 kJ/mol) [29]. Thus, impinging Ag atoms are expected to bond tightly with the gold seed rather than migrate over the surface to form clusters. Then, a series of VAuAgV multilayers was prepared to compare their characteristics with the previously studied VAgV structures. The Au seed layer (1 nm) was combined with 5-, 7-, and 9-nm Ag layers for total metal thicknesses equivalent to those of VAgV samples (6, 8, and 10 nm). A comparison of the optical transmittance evidences the interest in using a seed layer (see Fig. 3). The VAuAgV multilayers were noticeably more transparent in the visible and infrared regions. The optical transmittance was slightly reduced only at short wavelengths (λ < 400 nm) because of the gold layer. Nevertheless, a better use of the solar spectrum by the VAuAgV multilayers is clear. The wetting effect of the Au seed could be definitely confirmed in the FESEM images (see Fig. 4). The left image corresponds to the VAgV sample with an 8-nm-thick Ag layer, where isolated clusters of silver can be clearly resolved. In contrast, the sample with an intermediate Au/Ag stack (1 nm/7 nm) shows a continuous and a much more homogeneous film.  Considering the final intend to fabricate solar cells, not only high transparency but also good electrical properties are needed. Thus, the sheet resistance of the different multilayers was measured by means of a 4-point probe. Fig. 5 shows an abrupt drop in the sheet resistance of the VAgV samples when the metal thickness is increased to 10 nm. The samples with Ag layers thinner than 8 nm behaved as insulators with a high sheet resistance (R sh ≈ 100 Ω sq ). This behavior can be directly related to the initial Volmer-Weber growth of the silver layer, which results in the earliest formation of isolated islands. Above a nominal thickness, these islands coalesce, which reduces the sheet resistance by three orders of magnitude. Similar percolation thresholds have been reported in the literature for MAM structures, although these values could shift depending on the deposition rate [30]. Among the series of VAgV samples, only those with 10-nm (87 Ω sq ) and 12-nm (30 Ω sq ) metal layers have good R sh values to fabricate solar cells. In contrast, all the VAuAgV layers evidenced rather low R sh values that start from 97 Ω sq for the thinnest Au/Ag metallic stack (1 nm/5 nm). The sheet resistance steadily decreases to values as low as 29 Ω sq (1 nm/7 nm) and 5 Ω sq (1 nm/9 nm) for thicker Ag layers. On the other hand, the samples with a gold seed presented average transmittances a 10%-15% higher compared to those of standard VAgV structures (see Fig. 5 inset). All these results outline the convenience to use an adequate seed in structures based on silver as its main metallic layer.
At this point, we must remember the prime objective to develop an all-in-one selective-contact plus a transparent electrode for silicon-based solar cells. Thus, these multilayers should be actually tuned for their best optical performance on a c-Si substrate. For this purpose, we used the TMM algorithm to calculate the wavelength-dependent absorption A of a Si wafer coated by different VAuAgV structures. The optical parameters (refractive index and extinction coefficient) of the different layers were taken from the literature [31], [32]. Then, the total photogenerated current J ph was evaluated by integrating the air-mass 1.5 global irradiance AM1.5 properly weighted by the absorptance spectra in the Si wafer as follows: Fig. 6 shows the results of this analysis for a fixed 10-nm-thick Ag interlayer, when both the top and bottom V 2 O 5 layer thicknesses were varied. In order to achieve a good hole-selective character, the bottom V 2 O 5 layer should be around 20-nm thick [20]. Thus, an adequate thickness for the top V 2 O 5 layer that minimizes reflection losses would be around 50-60 nm. This value did not shift very significantly when the same analysis (not shown) included either a thin (1 nm) Au seed or the Ag thickness was varied from 6 to 10 nm.
This study finished with the fabrication of complete solar cells that implement an optimized VAuAgV multilayer on the front side. The thickness for the bottom V 2 O 5 layer was 20 nm, which ensures a strong hole-selectivity on the Si absorber. The metallic interlayer consisted of a 1-nm-thick Au seed followed by a  7-nm-thick Ag layer. Finally, a top V 2 O 5 layer (55 nm) coated the metallic stack to reduce reflection losses. These values were chosen for the best solar cell performance, in accordance with all the studies done in this paper. The final device yielded a 13.3% conversion efficiency η under a simulated AM1.5 irradiance of 100 mW/cm 2 . The photovoltaic parameters of the JV curve were J sc = 27.1 mA/cm 2 , V oc = 618 mV, and fill factor = 79.5% (see Fig. 7). The open-circuit voltage was remarkably high compared with that of a similar device (V oc = 583 mV) with a MAM front contact [18]. Such a difference might be explained by the superior performance of silicon heterojunctions based on V 2 O 5 over other TMO alternatives [8], [20]. Nevertheless, the final V oc values could also be influenced by the particular rear contact implementation. On the other hand, the high FF value of the solar cell presented here is noteworthy. The low sheet resistance achieved with the Au/Ag stacks could be the main factor that contributes to this good result. In this regard, the QSSVoc curve of the device was measured as a function of the illumination intensity. These data were used to calculate a pseudo-JV curve in which the effect of the series resistance is eliminated (see Fig. 8). The corresponding pseudo-FF (pFF) increases to 81.6%, which points to a good-quality diode. Actually, the exponential region of the pseudo-JV curve could be fitted by considering a single-diode model as follows: with a low saturation current density J o = 4.8 pA/cm 2 and a rather good (close to one) ideality factor n = 1.1. The value of the total series resistance R S can be obtained from the pFF and the parameters of the JV curve under illumination [33] as follows: This calculation results in an R S value as low as 0.6 Ω·cm 2 , which proves the good electrical quality of optimized VAuAgV electrodes.
The moderate J sc value is the main factor that limits the efficiency of this solar cell structure. Fig. 9 shows the EQE curve of the solar cell together with its front reflectance spectra. In the inset, charge-carrier generation and different loss mechanisms are balanced for an AM1.5 irradiance. The incident photons with energy higher than the Si bandgap could generate a maximum photocurrent of 42.3 mA/cm 2 . Reflection on the front surface causes a 16.7% loss, which limits the photogenerated current to approximately 35.2 mA/cm 2 . The measured EQE curve implies a 23% internal loss, which results in a final J sc value of 27.1 mA/cm 2 . This reduction can be explained by considering the optical absorption of the VAuAgV structure, mainly at its metallic stack. The internal charge-carrier recombination also reduces the J sc value but less significantly because of the goodquality c-Si absorber and the use of a passivated rear contact. All the samples and solar cells studied in this paper were fabricated on flat substrates. Assuming that internal losses should remain similar, the use of textured substrates could lead to J sc values of about 30 mA/cm 2 . Furthermore, according to the electrical characterization (see Fig. 5), the Au/Ag stack can be thinned below 6 nm and R sh values maintained in the range of 10 2 Ω sq . In combination with an optimized metallization scheme, this change would increase the J sc value by 3-4 mA/cm 2 without a significant FF degradation.

IV. CONCLUSION
Traditional TCO electrodes can be replaced by optimized stacks of VMV layers. This approach circumvents the scarcity of indium, which could limit the viability of widely used ITO electrodes. Furthermore, the evaporation of these layers is more gentle to the underlying interface compared with the sputtering technique. The sheet resistance of VMV structures clearly improves that of standard TCO electrodes, although the averaged transmittance is slightly reduced. In this sense, a minimum Ag thickness of 10 nm is required to achieve low R sh values. However, the use of an ultra-thin Au seed (1 nm) favors that Ag layers as thin as 6 nm already coalesce into a continuous film. The reduction in the thickness of the intermediate metallic layer increases the optical transmittance by more than 15%. Given the hole-selective character of V 2 O 5 layers on silicon, the fabrication of dopant-free heterojunction solar cells is straightforward. In this paper, an optimized VAuAgV structure on n-type c-Si has yielded a conversion efficiency of 13.3%. The remarkably high fill-factor (79.5%) points out the low sheet resistance that has been achieved with this alternative transparent electrode. The open-circuit voltage (618 mV) is also rather good, more so when considering that it is likely limited by the also novel rear contact that has been implemented here. There is definitely room for improvement in the J sc value (27.1 mA/cm 2 ). In this sense, technological transfer to textured substrates could be the next step for this research line.