Graphene-Al₂O₃-silicon heterojunction solar cells on flexible silicon substrates

Two-dimensional materials have been studied extensively^1–3 since the seminal work with graphene in 2004.⁴ The appeal of this class of materials arises from their unique electronic, optical, and physical properties. For example, graphene has remarkable mechanical properties and flexibility,⁵ it is optically transparent (absorbing less than 3% of incident light per layer),⁶ and it is highly conductive electrically and thermally.⁷ One application for graphene which has garnered early attention is its use as a transparent conducting film.^8,9 For many applications, the current market standard transparent conductor is indium tin oxide (ITO) and graphene can provide unique advantages over ITO, namely, mechanical flexibility and lower cost with earth abundant materials.¹⁰ As a transparent conductor, graphene can advance energy generation in solar cell applications. To this end, graphene has been used to form a heterojunction with semiconductors such as silicon and GaAs, resulting in graphene semiconductor-Schottky barrier solar cells (GS-SBSC).^11,12 While early work gave efficiencies of less than 2% in power conversion efficiency (PCE), researchers have boosted the efficiencies of these GS-SBSC up to 15.6% by various techniques such as graphene doping, insertion of a thin insulator between the graphene and Si, anti-reflective coating, and surface passivation.^13–18 

Thin solar cells possess the advantage of reducing the material cost as well as potentially increasing the PCE.¹⁹ Additionally, thin solar cells can be made flexible, which opens a whole new range of applications such as wearable devices.^20,21 Utilizing the inherent flexibility of graphene, GS-SBSC fabricated on thinned Si body have been demonstrated with efficiencies up to 8.4%.^22,23 These flexible GS-SBSC are fabricated on bendable Si films which are produced by wet etching a bulk Si wafer. In this paper, we demonstrate GS-SBSC on bendable, thin silicon foils obtained by a kerf-less mechanical exfoliation technique. The exfoliation process does not require the strong etchants of the other processes, and the “parent” wafer can be subsequently used to generate additional thin Si films, and both aspects bolster the effort of reducing material cost. We also study the effects of an Al₂O₃ interlayer between the exfoliated Si and doped graphene, giving a graphene-insulator-silicon (GIS) solar cell. This thin insulator layer, deposited by atomic layer deposition (ALD), increases the overall performance of the device by preventing recombination and increasing charge carrier lifetime.

Multi-layer graphene (MLG) was grown by chemical vapor deposition (CVD) on Cu-Ni alloy foils in a low-pressure system. A single layer of graphene can be grown in a self-limiting process on pure copper due to the low solubility of carbon in copper.^24,25 Although other reports have shown high performance using single layer graphene (SLG),¹⁷ it has also been shown that MLG enhances the solar cell performance.^26,27 Therefore, in these devices, we incorporate MLG. By growing on a Cu-Ni alloy instead, the solubility of carbon is increased, and MLG can be formed by precipitating carbon out (in the form of graphene) during cooling, as has been shown previously.^28,29 The dissolution and precipitation process is shown schematically in Figure 1(a). Methane was flowed at 3 sccm and carried by hydrogen gas flowed at 10 sccm to the foil substrate held at 1050 °C. After the high temperature growth, the chamber was cooled also under 3 sccm flow of methane and 10 sccm flow of hydrogen. After growth and graphene transfer as described below, the MLG film is characterized by confocal Raman spectroscopy (Renishaw inVia, 532 nm wavelength laser) and atomic force microscopy (AFM), which is shown in Figures 1(b) and 1(c). The Raman characterization gives a spectrum consistent for MLG and a narrow distribution for the G band position.^30,31 The AFM scan at the edge of a MLG film shows a step height of ∼10 nm, which corresponds to 27–29 layers of graphene.

After synthesis, the MLG is charge-transfer doped p-type by spin coating a 10 mM solution of AuCl₃ in nitromethane at 2000 rpm for 1 min.^32,33 The doping effect is verified by sheet resistance and transmission measurement, as shown in Figure 1(d). The AuCl₃ dopant does adversely impact the transmission of the MLG layer, and there is a trade-off between loss in transmission due to more graphene layers and doping, and a boost in conductivity. After the doping process, the CVD graphene is transferred using a poly(methyl methacrylate) (PMMA)-assisted wet etching method.³⁴ A PMMA film is spin coated onto the graphene at 2000 rpm and baked for 1 min at 90 °C. The PMMA-graphene stack is floated on 0.5 M ammonia persulfate solution to etch away the Cu-Ni alloy substrate. The PMMA-graphene is then rinsed in deionized water to remove any etchant or byproducts from the Cu-Ni removal process. The PMMA-graphene is lifted from the water directly by the target substrate, the silicon foil. The doping was done prior to the transfer for reasons discussed below.

As we show in Section III, the tradeoff favors utilizing MLG (as opposed to SLG) and doping it. Doping the graphene (with AuCl₃) reduces the graphene sheet resistance which is measured by the van der Pauw method. Undoped and pre-doped MLG were transferred onto a SiO₂/Si bulk substrate followed by metal contact formation using silver paste for measurement. Sheet resistances of 448 and 11 Ω/◻ were obtained for undoped and doped MLG, respectively, with an encapsulating layer of PMMA on top.

Figure 2 illustrates the kerf-less exfoliation process based on spalling and the resulting exfoliated silicon film on a metal foil. Bulk Si wafers (n-type, ⟨100⟩, 1–5 Ω cm) were covered with a dual-layer of hydrogenated amorphous silicon (a-Si:H) using remote plasma chemical vapor deposition (RPCVD). 5 nm, intrinsic a-Si:H was deposited on the front side to passivate the surface, and 7 nm, n⁺ doped a-Si:H was deposited on the back side to achieve ohmic contact with backside metal.³⁵ Chromium (10 nm) and Ni (100 nm) were sequentially deposited onto the back surface by electron beam evaporation as a seed layer for electroplating 50–55 μm of Ni onto the metal seed layer, with current density fixed to 30 mA/cm². At this step, the backside of the GS-SBSC device was complete and the kerf-less exfoliation process was done.^36,37 Exfoliation was initiated with a crack at the edge of the wafer after the thermal annealing process followed by controlled spalling. The exfoliated Si thickness was controlled by the thermal cycling process (270 °C–310 °C, 10 min) which thermally induced stress at the metal/Si interface due to the difference in the coefficient of thermal expansion (CTE) in the different materials. Silicon thicknesses down to 8 μm can be successfully exfoliated with surface RMS roughness as low as 2 nm. Other parameters such as metal thickness, annealing time, current density during electroplating, and mechanical exfoliation conditions can further control the Si thickness.³⁸ 

Solar cells fabricated by this method reduce the material cost through re-use of the parent wafer for subsequent exfoliation. This method is also beneficial with respect to handling issues and preventing cracks in thin Si films because of the mechanical support provided by the electroplated Ni layer.

The exfoliated silicon film with Ni metal backing foil was first sonicated in acetone and iso-propyl alcohol, followed by SC-1 clean which further removes Si particles on the exfoliated Si surface. Silicon films with thicknesses of 35 μm were used to maximize absorption, while maintaining the flexibility of the film. Immediately before any subsequent process, the native oxide was removed from the silicon surface in diluted HF. In the case where an interlayer oxide was inserted (GIS cells), Al₂O₃ was deposited by atomic layer deposition in a Cambridge nanotech Fiji F200 ALD system, using tri-methyl aluminum (TMA) and water precursors at 200 °C. To fabricate the GS-SBSC device (or GIS cell), MLG was transferred using the method described above onto the silicon surface (with or without an intervening Al₂O₃ layer) and then allowed to dry under ambient conditions. The sample was then heated up to 110 °C for 10 min to improve graphene/substrate adhesion. To further aid adhesion and remove wrinkles in the MLG, an additional application of PMMA was made on top of the PMMA-graphene.^39,40 The PMMA was spin coated at 3000 rpm for 1 min and then baked at 90 °C for 1 min. It has been previously reported that removing the PMMA after graphene transfer can degrade the graphene quality.²² To effectively dope the graphene while retaining the graphene quality, the graphene is pre-doped prior to PMMA coating.

Silver paste was directly applied onto the PMMA/AuCl₃/graphene to form the top contact with the graphene. All remaining areas were covered with black tape to ensure precise measurements. The active area of the solar cells is 0.17–0.37 cm². Figure 3 illustrates the GS-/GIS-SBSC device fabrication and includes a photograph of a completed GIS cell.

Figure 4(a) shows the current density-voltage (J-V) characteristics of un-doped and pre-doped heterojunction GS-SBSC on thin exfoliated Si films. GS-SBSC were measured using an ABET solar simulator (Model Sun 2000) and were calibrated using a reference solar cell (ABET model #15150) prior to measurements. Results show that doping the graphene increases open circuit voltage (V_OC) from 209 to 299 mV, fill factor (FF) from 43% to 53%, and thus improving the PCE from 1.8% to 3.0%. The increase in performance after doping can be explained by analyzing the GS-SBSC dark I-V graph and band diagram. According to the Schottky-Mott rule of Schottky barrier formation, the Schottky barrier height (SBH, ϕ_SBH) between MLG and thin Si film is described based on the work function of the MLG (W_G) and the electron affinity of the Si (χ), ϕ_SBH ∼ W_G − χ. Doping the MLG with AuCl₃ is known to shift the MLG work function (⁠$ϕSBH′$⁠)³² and can be extracted from the equation

where J_S is the saturation current extrapolated at V = 0 from the linear portion of the forward current (solid lines in Figure 4(b), un-doped J_S = 8.9 μA/cm², doped J_S = 1.14 μA/cm²), A* is the Richardson constant (∼112 A/cm² K² for n-Si), T is the absolute temperature, q is the electron charge, and k is the Boltzmann constant.¹³ The SBH for un-doped and doped GS-SBSC was found to be 0.71 and 0.77 eV, respectively.

The change in saturation current results in a change of V_OC given by the following relation:

where n is the ideality factor of the graphene/Si interface, and J_ph is the light-induced photocurrent density of the solar cell.⁴¹ The ideality factor can be extracted by fitting the device response in the dark, giving n of 1.5 and 1.8, respectively, and shown in Figure 4(b). Following Eq. (2), an increase in n and decrease in J_S lead to an increase in V_OC for doped MLG solar cells. The drop in J_SC can be explained due to the decrease in transmitted photons after doping. There is a loss in transmission through the MLG due to AuCl₃ charge-transfer doping, which results in a decrease of light reaching the silicon absorber as shown in Figure 1(d). Figure 4(c) illustrates the band diagram for un-doped and p-type doped graphene on n-type silicon, describing the change at the graphene-silicon interface.

The increase in FF is due to the reduction in series resistance (R_S) after graphene doping. The R_S for un-doped and doped GS-SBSC is extracted using the slope of the dV/d(lnJ) plot, which is shown in Figure 4(d). The R_S is shown to decrease from 11.75 to 4.08 Ω cm² for doped GS-SBSC cells. Since FF is strongly related to R_S, the decrease in R_S gives the increase in FF. Overall, the GS-SBSC is improved from 1.8% PCE to 3.0% PCE upon doping with 10 mM AuCl₃ in nitromethane.

To build upon the efficiency improvements of doped GS-SBSC, a thin insulating layer of Al₂O₃ was inserted between the graphene and Si and forms a graphene-insulator-semiconductor (GIS) cell. Previous studies have shown that materials such as silicon oxide,¹⁷ graphene oxide (GO),¹⁴ and 2-D materials such as hexagonal boron nitride (h-BN)⁴² and MoS₂ (Ref. 27) increase the overall efficiencies of graphene-on-silicon (GS) solar cells. Introducing an inter-layer between graphene and Si was proven to improve the overall performance of GS heterojunction solar cells by blocking carriers from recombining at the interface, reducing the density of interface states, and increasing the carrier lifetime (τ).

Here, we introduce another inter-layer material, Al₂O₃, which was selected because it has been shown to effectively passivate c-Si by suppressing surface recombination and increase τ.^43–45 To check the passivation quality of our Al₂O₃, we deposited a thin layer on both sides of a cleaned float zone (FZ) n-type Si wafer (1–5 Ω cm) and measured the carrier lifetime by the photoconductance method (Sinton instruments WCT-120). The carrier lifetime before and after deposition was 2 and 27 μs, respectively, with 2 nm of Al₂O_3, which is comparable to the 33 μs lifetime for a cell with a graphene oxide (GO) interlayer.¹⁴ 

Figure 5(a) compares the V_OC and J_SC of GIS cells on thin silicon films with different Al₂O₃ thicknesses. The J_SC shows limited variation for the different Al₂O₃ thicknesses, while the V_OC tends to increase up to 1 nm Al₂O₃ and then saturates. This increase in V_OC upon including the Al₂O₃ layer shows that the interlayer indeed passivates the Si surface of the GIS cells.

To extract more information about the GIS device performance, we consider the diode characteristics. In the case of a device with the metal-insulator-semiconductor (MIS) structure, with the thin insulator material at the interface, the saturation current density J_S is described as

where ϕ_T is the barrier height presented by Al₂O₃, and d is the Al₂O₃ thickness.⁴⁶ Figure 5(b) illustrates the band diagram of the GIS structure. While the current density is inverse exponentially proportional to insulator thickness, changing the thickness does not drastically affect J_SC, which is consistent with previous publications.^14,17 For our GIS structure with Al₂O₃, while increasing the Al₂O₃ thickness initially increases the V_OC, a competing process of photocurrent suppression seems to occur with thicker Al₂O₃ devices, resulting in V_OC saturation. This phenomenon is also consistent with previous studies, such as with Si native oxides.¹⁷ The addition of Al₂O₃ also gives a boost to the FF compared with the GS-SBSC, as shown in Figures 4(a) and 5(a). This phenomenon may be understood by considering the interaction between MLG and its substrate. ALD of the Al₂O₃ gives a uniform and conformal high k dielectric substrate for the MLG, compared with bare Si, and it is understood to impact the mobility (and R_S) of graphene.^47–49 For the GIS cells, enhancement in hole transport in the MLG would lower the device R_S with thin Al₂O₃. As the oxide layer is made thicker, the insulating nature of the layer becomes dominant and an increase in R_S and decrease in FF are observed.¹⁷ 

Figure 6 shows J-V and external quantum efficiency (EQE) measurements for the optimum GIS device with n and R_S values extracted from dark current. The highest efficiency of the GIS solar cell on the exfoliated thin Si film was 7.4% with an Al₂O₃ interlayer thickness of 1 nm, showing V_OC of 459 mV, J_SC of 23.2 mA/cm², and FF of 70%. A low R_S leading to high FF, given by controlling the Al₂O₃ thickness, provided the path to optimize the GIS device and maximize efficiency.

Previous GS- and GIS-SBSC devices have been reported to have unstable performance over time. For example, some devices have utilized volatile dopants for the graphene layer which degraded over time⁵⁰ or utilized PMMA encapsulation to show stability for a few days.²² To check the stability of our devices, the optimum GIS sample was initially measured after fabrication and then left under ambient conditions before re-measurement. The temperature and humidity of the facility are maintained at approximately 21 °C and 50%. The device performance showed <1% degradation after 40 days, which is the longest stability of GIS cells reported so far.^14,22,50 We attribute the increase in stability of these GIS cells to a combination of high quality CVD MLG and ALD AL₂O₃ materials and encapsulation with PMMA.

We report flexible graphene-Al₂O₃-Si solar cells, with a crystalline 35 μm thick Si absorber with 7.4% PCE with stable performance over 40 days under ambient conditions. Our investigation shows that the performance of the cell is controlled by the interfacial Al₂O₃ dielectric thickness. The ability to control the ALD process, along with incorporation of other fabrication processes such as AuCl₃ doping, CVD MLG growth, and thin Si exfoliation, allowed the optimization of the GIS device. The other functional layers in the cell can potentially be optimized for further improvements. For example, the thickness of the thin Si would determine the amount of incident light absorbed.²⁶ Also, texturing the front and/or back surface of the silicon or adding metamaterials could enhance the light absorption of the device by acting as an anti-reflection layer. The GIS cell performance would depend on the type of insulator. Other 2D materials may also be compatible considering this flexible device.^27,42 By utilizing new materials and new processes, the scope of possible applications for thin and flexible solar cells can be further expanded.

This work was supported by the NSF NASCENT ERC, DOE BAPVC Program, and the NSF NNCI Program.