High-Specific-Power Flexible Transition Metal Dichalcogenide Solar Cells

Semiconducting transition metal dichalcogenides (TMDs) are promising for flexible high-specific-power photovoltaics due to their ultrahigh optical absorption coefficients, desirable band gaps and self-passivated surfaces. However, challenges such as Fermi-level pinning at the metal contact-TMD interface and the inapplicability of traditional doping schemes have prevented most TMD solar cells from exceeding 2% power conversion efficiency (PCE). In addition, fabrication on flexible substrates tends to contaminate or damage TMD interfaces, further reducing performance. Here, we address these fundamental issues by employing: 1) transparent graphene contacts to mitigate Fermi-level pinning, 2) $\rm{MoO}_\it{x}$ capping for doping, passivation and anti-reflection, and 3) a clean, non-damaging direct transfer method to realize devices on lightweight flexible polyimide substrates. These lead to record PCE of 5.1% and record specific power of $\rm{4.4\ W\,g^{-1}}$ for flexible TMD ($\rm{WSe_2}$) solar cells, the latter on par with prevailing thin-film solar technologies cadmium telluride, copper indium gallium selenide, amorphous silicon and III-Vs. We further project that TMD solar cells could achieve specific power up to $\rm{46\ W\,g^{-1}}$, creating unprecedented opportunities in a broad range of industries from aerospace to wearable and implantable electronics.

lightweight and flexible polyimide (PI) substrate. 39 The flexible TMD (WSe2) solar cells made in this fashion achieve a PCE of 5.1%, surpassing previous flexible TMD solar cells by more than an order of magnitude. 20 Furthermore, the integration on an ultrathin substrate enables a PS of 4.4 W g -1 , more than 100x higher than previous results on flexible TMD photovoltaics 20 and in the same range as champion solar cells of prevailing thin-film technologies cadmium telluride (CdTe), copper indium gallium selenide (CIGS), amorphous silicon (a-Si) and group III-V semiconductors. [40][41][42][43][44][45][46][47] In future, TMD solar cells on even thinner substrates and with higher PCEs could potentially achieve an unprecedented PS of ~46 W g -1 (as we project in this work) opening up far-reaching possibilities in a broad range of industries. 9 We  the charge neutrality level of WSe2 located at midgap. 26,51,52 This decreases the effective work function of Au and makes it a decent electron-collecting contact. We find that replacing Au with lower work function metals such as Ti and Al leads to a lower performance, most probably due to their reactive nature therefore forming poor interfaces with WSe2 (see Supplementary Information Section S2). 53 On the other hand, layered materials Gr and WSe2 experience no Fermi-level pinning at their vdW interface. 32,33 The work function of undoped graphene is ~4.6 eV (e.g. in vacuum), which increases to ~5.0 eV when graphene is exposed to air, 37,54,55 forming a Schottky junction with the "undoped" WSe2. The MoOx on top of graphene further increases its work function by doping it p-type, 37 therefore enhancing the built-in potential of the Gr-WSe2 Schottky junction. MoOx also passivates the top surface of the solar cell. 24 These lead to a higher open-circuit voltage (VOC) and short-circuit current density (JSC) in MoOx-capped WSe2 solar cells (Supplementary Information Section S3). As we will discuss later in the optical characterization section, Given the approximate location of Gr, WSe2 and Au Fermi levels, the depletion regions of Gr-WSe2 and Au-WSe2 Schottky junctions are estimated to be in the order of 1 μm and therefore expand throughout the entire depth of the ~200-nm-thick WSe2 layer, leading to fully depleted devices. the Au-WSe2 interface, but not at the van der Waals graphene (Gr)-WSe2 interface. 26,32,33,51,52 MoOx increases the Gr work function and the built-in potential of the Gr-WSe2 Schottky junction. 37 Fig. S6). Similar J-V characteristics are observed in solar cells with "undoped" tungsten disulfide (WS2) absorber layers.
However, due to the higher work function of WS2 compared to WSe2 and therefore the lower built-in potential in Gr-WS2 Schottky junctions, WS2 solar cells exhibit lower VOC, JSC, FF and hence PCE (see Supplementary Fig. S7).
Next, we measure the current density vs. voltage (J-V) characteristics of flexible WSe2 solar cells in the dark and for AM 1.5G illumination at various incident power intensities ( Fig. 2a- Fig. 2b shows a zoomed-in view of the photovoltaic region. An analysis of this data indicates that the shunt resistance decreases almost linearly with increasing incident intensity (see Supplementary Fig. S8). This phenomenon, known as photoshunting, occurs due to increased minority carrier conductivity across the device under illumination. 59,60 Improving the charge carrier selectivity of the solar cell, for example by utilizing carrier-selective metal-interlayer-semiconductor (MIS) contacts or introducing a high built-in potential p-n homojunction could reduce or eliminate photoshunting. Given the initially high shunt resistance of the device, photoshunting does not affect the shape of the J-V curve and therefore fill factor stays constant at various intensities.
By fitting a power law equation on the measured current density and incident power data (Fig. 2c), we observe that short-circuit current density versus incident power follow a linear trend (JSC = β·(Pin) α , α = 1), WSe2 solar cells demonstrate a desirable near-unity ideality factor of n = 1.18 and dark saturation current of Jo = 3.49 nA cm -2 . The near-unity ideality factor and small dark saturation current indicate low levels of charge carrier recombination and therefore a high internal quantum efficiency as confirmed later by comparing the measured JSC and JSC, max derived from absorption measurements. To accurately define the active area of the device, the photocurrent profile across the width of the device (x-axis in Fig. 3a) is plotted on a linear scale (Fig. 3b). In this specific device, MoOx is slightly misaligned with respect to Gr (see Fig. 3a). The misalignment only occurred in few devices due to lithography issues.
Most devices, such as the one shown in Fig vertical TMD solar cells. 22 The active area of the solar cells tested vary from ~10 3 to ~10 4 μm 2 .
We measure the absorption spectrum of WSe2 solar cells at different stages of fabrication, i.e. after polyimide release (Au-WSe2), after Gr transfer (Au-WSe2-Gr) and finally after MoOx deposition (Au-WSe2-Gr-MoOx), as shown in Fig. 3c. For consistency, each measurement is taken at exactly the same spot at the center of the active area of the device. The data in Fig. 3c corresponds to the device whose J-V characteristics are shown in Fig. 1f. This device has a 209-nm-thick WSe2 absorber layer, as measured by a stylus-based surface profiler.
After transferring Gr on top of WSe2, the overall absorption of the stack is slightly reduced. Optical simulations using the transfer matrix method produce a similar result ( Supplementary Fig. S10a).
Depositing 10 nm of MoOx on top of Gr increases the overall absorption of the stack. This can be either due to parasitic absorption within MoOx or its anti-reflection coating effect improving the absorption within the WSe2 absorber layer. To answer this question, we simulate absorption using the transfer matrix method.  To test the performance of devices under bending, we attach the PI substrate onto an 8-mm-diameter metal cylinder, which bends the substrate at a curvature radius of 4 mm (Fig. 4a). The flexible WSe2 solar cells show the same J-V characteristics in flat and bent states under AM 1.5G illumination (Fig. 4b), indicating consistent performance levels under bending. This is not surprising because given the PI substrate thickness of only 5 μm, the materials encounter small strain values of ~0.06% at this bending radius. 39 We have investigated similar TMD, metal and dielectric stacks in electronic devices in more detail in our recent work and found that there is no discernable change of electrical device properties. 39 For our solar cells with small exfoliated flakes on length scales that are tens of microns, there is no effect of substrate curvature on the light-coupling, as seen in Fig. 4b. In future, if the active area of the solar cell is increased e.g., by large-area synthesis of TMDs, these bending studies will become more important to quantify the effects of the bent surface on light-coupling and thus JSC, which would alter solar cell performance. By reducing the PI substrate thickness to 1 μm, similar as in some of the champion organic PV (OPV) and perovskite solar cells in Fig. 5, specific power can be further increased to 8.6 W g -1 (path #1 in Fig. 5).
According to realistic detailed balance models developed for TMD photovoltaic cells, single-junction multilayer TMDs can in principle achieve ~27% PCE with an optimized optical and electronic design. 13 Such PCE would lead to an ultrahigh specific power of 46 W g -1 (path #2 in VOC is another important area of improvement. The built-in potential and therefore VOC of these devices can be improved by employing n-type WSe2. The doping process can be performed during growth, or by means of metal oxides such as AlOx and TiOx. 69,70 Replacing Au with a lower work function metal but ensuring a similar interface quality could also improve VOC. We showed in our experiments that Al and Ti are not good candidates for this purpose (see Supplementary Information Section S2). Forming a high built-in potential p-n homojunction in the TMD absorber layer is another way to achieve a high VOC, possibly by p-type TMDs to achieve high power conversion efficiency and specific power at a low cost as well as their stability and environmentally-friendliness (in contrast to perovskites) makes them a serious candidate for nextgeneration photovoltaics, especially in high-specific-power applications.
In summary, we demonstrated flexible WSe2 solar cells with record-breaking power conversion efficiency PCE of 5.1% and power per weight PS of 4.4 W g -1 . We performed detailed optical and electrical characterizations on these solar cells to explain their superior performance and identify areas of improvement, providing practical guidelines on the optical and electronic design to enhance PCE and PS.
We also tested the flexible solar cells under bending and showed similar performance in flat and bent states.  • Section S10. Effects of Gr and MoOx on optical absorption • Section S11. Benchmarking Section S1. Detailed fabrication process including transfer procedure We follow a recently developed fabrication and transfer approach, 39 where we perform the initial fabrication steps on a rigid Si/SiO2 substrate followed by the release of patterned electrodes and the transition metal dichalcogenide (TMD) embedded into an ultrathin (~5 μm) flexible polyimide (PI) substrate (Fig. S1). This technique is advantageous for vertical device architectures as it eliminates large steps in surface topography leading to a flat surface for the subsequent transfer of transparent graphene top electrodes. The detailed process steps are described below and the summarized process sequence is schematically shown in Fig. S1. First, 90 nm of SiO2 was grown on a bulk silicon wafer by dry thermal oxidation at 1100 °C. The wafers were then manually cut into ~2 cm × 2 cm pieces. Intrinsically doped ("undoped" per manufacturer's description) WSe2 flakes were mechanically exfoliated from the bulk crystal (2D Semiconductors) onto low-residue thermal release tape (Nitto Denko REVALPHA). They were subsequently transferred from the tape onto the Si/SiO2 substrate using a WF film (Gel-Film ® , WF-20-X4).
Next, we spin-coated a lift-off layer LOL 2000 (3000 rpm, 60 s) on our substrates and baked it on a hot plate at 200 °C for 5 minutes. This was followed by spin-coating of Shipley 3612 photoresist

Section S3. Doping effects of MoOx
The doping/passivation effect of MoOx on graphene and WSe2 was analyzed by Raman spectroscopy (Table S1 and Fig. S3a-b). We find small peak shifts towards higher wavenumber for graphene after MoOx deposition, which has been previously reported to be a result of p-type doping and an increase of graphene's work function. 37 The WSe2 peaks have much smaller yet discernable shift towards lower wavenumbers, also associated with material doping as shown in. 38 Overall, the two doping effects result in an increase in the built-in potential of the Gr-WSe2 junction (more p-type doping in Gr than in WSe2), leading to increased open circuit voltage (VOC) and short-circuit current (JSC). The passivation 24 and anti-reflection coating ( Fig. S10b-d) effects of MoOx further increase the JSC, leading to significant PCE improvements (Fig. S3c).   We calculate PS (unit: W g -1 ) by dividing Pmax by the total area mass density from Table S2. With Pmax = 5.1 mW cm -2 (see Fig. 1f) we obtain PS = 4.4 W g -1 . A simple approach to reduce PS would be thinning down the substrate to ~1 μm, which is the approximate substrate thickness in some of the works with highest PS achieved so far, 62,87,88 leading to a value of ~8.6 W g -1 . A PCE of ~27% can be practically achieved in an optimized TMD single-junction as shown in, 13 further increasing the PS to ~46 W g -1 . We included these projections in Fig. 5 to emphasize the great potential of TMDs for high-specific-power photovoltaics. b, a zoom-in view of the photovoltaic region. WS2 solar cells show the same J-V characteristics as WSe2 cells, however with lower performance. This is due to the higher work function of undoped WS2 compared to undoped WSe2 (higher electron affinity and bandgap in WS2), 89 leading to a smaller built-in potential in WS2-Gr Schottky junction and therefore reduced VOC, JSC, and FF. Further doping Gr (for example by using flame-deposited MoO3) 37 is one way to achieve a similar built-in potential and therefore performance in WS2 solar cells.

Section S10. Effects of Gr and MoOx on optical absorption
Figure S10 | Effects of Gr and MoOx on optical absorption. a, Absorption reduction by graphene. The difference in total absorption between the Au-WSe2 and Au-WSe2-Gr stacks, showing that adding graphene on top slightly reduces the total absorption in the ~500-800 nm wavelength spectrum for WSe2 thicknesses around 200 nm, in agreement with experimental measurements (Fig. 3c). b-d, Anti-reflection coating effects of MoOx. b, Absorption spectrum and c, average absorption of WSe2 in the Au-WSe2-Gr-MoOx stack as a function of MoOx thickness. WSe2 thickness is 209 nm, similar to the device in Fig. 3c. d, Maximum short-circuit current density (JSC) attainable from WSe2 in the same stack, as a function of MoOx thickness, calculated by integrating Absorption(λ) × (spectral photon flux of AM1.5G spectrum at one-sun solar intensity) over the wavelength range of λ = 400-1000 nm, assuming unity internal quantum efficiency (IQE). This maximum JSC value is slightly underestimated as absorption at wavelengths below 400 nm and above 1000 nm are not included due to lack of material data. An optimal MoOx thickness of ~70 nm can increase the average absorption in WSe2 and similarly JSC by ~50%, leading to a remarkable ~80% average absorption and ~30 mA cm -2 JSC in WSe2 solar cells.

Section S11. Benchmarking
We performed an extensive literature review on flexible and light-weight solar cells with various absorber materials and calculated PS based on information provided. Some works directly stated PS or at least PCE and the areal mass density, while others did not explicitly provide information on the weight of their solar cells. In some cases, we calculated the weight based on the substrate and the solar cell layer stack if the substrate weight seemed small (sufficiently thin/low density material). In other cases, we only used the substrate weight for estimating PS if it appeared to dominate (thicker high-density material substrates) and neglected the weight of other materials. Table S3 lists all the works shown in Fig. 5 indicating how PS was obtained.