High Responsivity, Large-Area Graphene/MoS2 Flexible Photodetectors.

We present flexible photodetectors (PDs) for visible wavelengths fabricated by stacking centimeter-scale chemical vapor deposited (CVD) single layer graphene (SLG) and single layer CVD MoS2, both wet transferred onto a flexible polyethylene terephthalate substrate. The operation mechanism relies on injection of photoexcited electrons from MoS2 to the SLG channel. The external responsivity is 45.5A/W and the internal 570A/W at 642 nm. This is at least 2 orders of magnitude higher than bulk-semiconductor flexible membranes. The photoconductive gain is up to 4 × 10(5). The photocurrent is in the 0.1-100 μA range. The devices are semitransparent, with 8% absorptance at 642 nm, and are stable upon bending to a curvature of 1.4 cm. These capabilities and the low-voltage operation (<1 V) make them attractive for wearable applications.

Modern electronic and optoelectronic systems, such as smart phones, smart glasses, smart watches, wearable devices, and electronic tattoos, increasingly require ultrathin, transparent, low-cost, and energy efficient devices on flexible substrates.[1] The rising demand for flexible electronics and optoelectronics requires materials that can provide a variety of electrical and optical functionalities, with constant performance upon application of strain.[2] A wide range of optoelectronic devices on flexible substrates have been reported to date, such as photodetectors (PDs),[3,4] light emitting diodes (LEDs),[5] optical filters,[6] optical interconnects,[7,8] photovoltaic devices,[9,10] and biomedical sensors.[11,12]

Major challenges in the development of flexible optoelectronic devices stem from the limitations associated with the high stiffness of bulk semiconductors.[13,14] In the case of flexible PDs, the current approaches primarily rely on thin (μm-thick) semiconductor membranes[4,15] and compound semiconductor nanowires (NWs),[3,16−18] mainly because of their ability to absorb light throughout the whole visible range (0.4–0.7 μm) and the possibility to adapt their fabrication techniques from rigid to plastic, or deformable substrates.[1]

One of the key parameters for PDs characterization is the responsivity. This is defined as the ratio between the collected photocurrent (Iph) and the optical power. The responsivity is named external (Rext = Iph/Po)[19] or internal (Rint = Iph/Pabs),[19] whenever the incident (Po) or absorbed (Pabs) optical power is used in the denominator. Since not all incident photons are absorbed by a PD, i.e., Pabs < Po, then Rint is typically larger than Rext.[19]

In flexible PDs, Rext up to∼ 0.3 A/W was reported for crystalline semiconductor membranes (InP, Ge)[4,15] with integrated p–i–n junctions, showing photocurrent up to∼ 100 μA, with∼ 30% degradation upon bending at a radius rb ∼ 3 cm.[15] PDs made of a single semiconductor NW on flexible substrates[3,16−18] demonstrated Rext up to∼ 105 A/W, for rb down to 0.3 cm.[3] Yet, these provide limited Iph in the order of nA[3,16,18] up to <1 μA.[17] For flexible devices exploiting NW arrays by drop-casting,[3,16,18] rather than based on single NWs, Rext degrades significantly from ∼105 A/W to the mA/W range,[3,16,18] due to photocurrent loss at multiple junctions in the NW network.[3,16,18]

Graphene and related materials (GRMs) have great potential in photonics and optoelectronics.[20−23] A variety of GRM-based devices have been reported, such as flexible displays,[24] photovoltaic modules,[25,26] photodetectors,[22,27−29] optical modulators,[30] plasmonic devices,[31−35] and ultrafast lasers.[23] Heterostructures, obtained by stacking layers of different materials, were also explored,[21,22]e.g., in photovoltaic[36] and light emitting devices.[37]

Flexible PDs based on GRMs were studied for ultraviolet,[38,39] visible,[40−45] and near-infrared bands.[46,47] In these devices, different materials and heterostructures produced by mechanical exfoliation,[40,41] Chemical Vapor Deposition (CVD),[42,43,46] and liquid-phase exfoliation (LPE)[44,45,47] were employed. The flexible PDs produced by mechanical exfoliation[40,41] have a small (∼μm2) photoactive area, and they cannot be scaled up to mass production. LPE-based PDs have low (44,45] responsivity. Ref (47) showed that thick (∼μm) films of chemically modified, charge-transfer optimized, LPE-produced, MoSe2 and MoS2 polymer composites can provide ∼ A/W responsivity[47] at near-infrared bands. Nevertheless, these PDs require high (∼10 V) operation voltage and are nontransparent. Flexible PDs at 450 nm using CVD MoS2 transistors[42] and MoS2/WS2 heterostructures[43] were previously reported, and PDs at 780 nm were prepared from doped SLG p–n junctions.[38] However, these devices have responsivity in the mA/W range. CVD-based SLG/MoS2 heterostructures[48] showed good photodetection on rigid Si/SiO2 substrates, with back-gate-dependent Rint ∼ 108A/W for optical intensities <0.1pW/μm2.

Here we demonstrate a polymer electrolyte (PE) gated, CVD-based, flexible PD, for visible wavelengths, with large (∼mm2) photoactive area combined with high responsivity (∼hundreds A/W), high (>80%) transparency, gate tunability, low (<1 V) operation voltage, and stable (±12%) Iph upon multiple (>30) bending cycles. The device is assembled by stacking on a PET substrate a centimeter-scale CVD single layer graphene (SLG) on top of a CVD-grown single layer MoS2 (1L-MoS2). In this configuration, 1L-MoS2 acts as visible light absorber, while SLG is the conductive channel for photocurrent flow.[48] We show that Rext can be increased by promoting carrier injection from 1L-MoS2 to SLG using PE gating, or by increasing the source-drain voltage. This is achieved in devices with∼ 82% transparency, twice that reported for semiconductor membrane devices.[15] We get Rint ∼ 570A/W for ∼0.1nW/μm2 at 642 nm, similar to SLG/MoS2 PDs[48] on rigid substrates operating at the same optical power. This shows that SLG/MoS2 heterostructures on PET retain their photodetection capabilities. We note that the devices from ref (48) have at least 2 orders of magnitude smaller photoactive area with respect to ours, and they are not flexible, not transparent, and require tens of volts operation, unlike the <1 V of ours. Upon bending, our PDs have stable performance for rb down to ∼1.4 cm. This is comparable to rb measured in semiconductor membranes PDs,[4,15] which show lower (<0.3 A/W) responsivities.[4,15] Although our rb is 1 order of magnitude larger than for flexible single NWs,[3,16−18] the latter had at least 3 orders of magnitude smaller device areas (<5 μm2)[3,16−18] compared to our PDs (>0.2 mm2). Given the responsivity, flexibility, transparency, and low operation voltage, our PDs may be integrated in wearable, biomedical, and low-power optoelectronic applications.[11,12,17]

Results

Figure a plots a schematic drawing of our PDs. We fabricated 4 PD arrays with 10 devices each, with channel lengths of 100 μm, 200 μm, 500 μm, and 1 mm. Each device consists of a 1L-MoS2 absorber covered by a SLG channel, clamped between source and drain electrodes. We chose PET as a flexible substrate due to its ∼90% transparency in the visible range[49] and ability to withstand solvents (e.g., acetone and isopropyl alcohol)[50] commonly used in the transfer processes of layered materials grown by CVD (e.g., transfer of SLG grown on Cu).[51] A 1L-MoS2 is used as absorber in order to preserve >80% transparency, considered suitable by industry for wearable applications,[52]Figure b. The SLG/1L-MoS2 heterostructure is gated using a PE.[53,54]

Figure 1

(a) Schematic SLG/MoS2 flexible PD, side-gated with a PE. (b) Picture of a typical PD, showing transparency and flexibility. (Inset) Optical image of 4 PDs with different channel lengths and common side gate electrode. Scale bar is 200 μm.

The operation principle of our devices is depicted in Figure . For energy bands alignment, the electron affinity of 1L-MoS2 and the Dirac point of SLG are assumed to be ∼4–4.2 eV[55,56] and ∼4.6 eV,[57,58] respectively. We also assume SLG to be initially p-doped (Figure a), as reported in previous works involving SLG transferred on PET.[59,60] At zero voltage the device is in thermodynamic equilibrium with a constant Fermi level (EF) and zero current flow between the layers. During illumination and photon absorption in MoS2, part of the photogenerated electrons would be injected from the 1L-MoS2 conduction band into the lower energy states in p-doped SLG,[48] leaving behind the uncompensated charge of photogenerated holes. The latter would be trapped in 1L-MoS2 and act as an additional positive gate voltage, VGS, applied to the SLG channel, resulting in a shift of the charge neutrality point (VCNP) to more negative voltages. The injected electrons from 1L-MoS2 would occupy energy states above EF (Figure b), thus reducing the hole concentration and decreasing the hole current in the SLG channel. Electron injection can be further promoted by gating. When a negative VGS is applied, higher p-doping of the SLG channel would induce a stronger electric field at the SLG/1L-MoS2 interface,[48] thus favoring electron transfer from 1L-MoS2 (Figure b). Hence, for negative VGS, Rext is expected to increase, due to injection of more photoelectrons to SLG and consequent more pronounced current reduction. The opposite should happen for positive VGS, where the gate-induced negative charge in SLG would reduce the p-doping and shift EF toward the Dirac point. In this case, the photogenerated electrons in 1L-MoS2 would experience weaker electric fields at the SLG/1L-MoS2 interface[48] and would become less attracted by the SLG channel. Thus, we expect Rext to decrease. For high enough positive VGS, EF would cross the Dirac point, and SLG becomes n-doped (Figure c). As a result, only a weak electron injection from 1L-MoS2 would be possible, if EF in SLG remains below the 1L-MoS2 conduction band, retaining a weak electric field at the interface. In this regime, the transferred electrons would increase the free carrier concentration in the n-doped channel, hence only minor increments of Rext and Iph are expected.

Figure 2

Schematic band diagram of PE gated SLG/1L-MoS2 PD at (a) zero, (b) negative, and (c) positive VGS.

Our devices are built as follows: 1L-MoS2 is epitaxially grown by CVD on c-plane sapphire substrates,[61] while SLG is grown on a 35 μm Cu foil, following the process described in refs (51 and 62) (see Methods for details). Prior to assembling the SLG/MoS2 stack, the quality and uniformity of MoS2 on sapphire and SLG on Cu are inspected by Raman spectroscopy and photoluminescence (PL), using a Horiba Jobin Yvon HR800 spectrometer equipped with a 100× objective. The laser power is kept below 100 μW (spot size <1 μm) to avoid possible heating effects or damage. Figure a (green curve) plots the Raman spectrum of CVD MoS2 on sapphire for 514 nm excitation. The peak at ∼385 cm–1 corresponds to the in-plane (E2g1) mode,[63,64] while that at ∼404 cm–1 is the out of plane (A1g) mode,[63,64] with full width at half-maximum FWHM (E2g1) = 2.5 and FWHM(A1g) = 3.6 cm–1, respectively. The E2g1 mode softens, whereas the A1g stiffens with increasing layer thickness,[65,66] so that their frequency difference can be used to monitor the number of layers.[65] The peak position difference ∼20 cm–1 is an indicator of 1L-MoS2.[65] The peak at∼ 417 cm–1 (marked by an asterisk in Figure a) corresponds to the A1g mode of sapphire.[67]

Figure 3

(a) Raman spectra at 514 nm for 1L-MoS2 on sapphire, 1L-MoS2 on PET, and SLG/1L-MoS2. (b) Comparison at 514 nm of the Raman spectra of as-grown SLG on Cu (magenta curve) and SLG/1L-MoS2 after transfer on PET and normalized subtraction of the PET substrate signal (blue curve). (c) Raman spectra at 514 nm of PET substrate (black curve), 1L-MoS2 on PET (red curve) and SLG/1L-MoS2 on PET (blue curve).

The Raman spectrum measured at 514 nm of SLG on Cu is shown in Figure b (magenta curve). This is obtained after the removal of the background PL of Cu.[68] The two most intense features are the G and the 2D peak, with no significant D peak. The G peak corresponds to the E2g phonon at the Brillouin zone center.[69] The D peak is due to the breathing modes of sp2 rings and requires a defect for its activation by double resonance.[69−72] The 2D peak is the second order of the D peak.[69] This is always seen, even when no D peak is present, since no defects are required for the activation of two phonons with the same momentum, one backscattering from the other.[69] In our sample, the 2D peak is a single sharp Lorentzian with FWHM(2D) ∼26 cm–1, a signature of SLG.[70] Different (∼20) measurements show similar spectra, indicating uniform quality throughout the sample. The position of the G peak, Pos(G), is ∼1588 cm–1, with FWHM(G) ∼6 cm–1. The 2D peak position, Pos(2D) is ∼2705 cm–1, while the 2D to G peak intensity and area ratios, I(2D)/I(G) and A(2D)/A(G), are ∼2.6 and ∼5.8, respectively, indicating a p-doping ∼300 meV,[53,73,74] which corresponds to a carrier concentration ∼6 × 1012cm–2.

Another evidence for 1L-MoS2 comes from the PL spectrum [Figure a (green curve)], showing a peak at ∼658 nm (∼1.88 eV), due to band-to-band radiative recombination in 1L-MoS2.[75]

Figure 4

(a) PL spectrum at 514 nm (2.41 eV) of 1L-MoS2 on sapphire and 1L-MoS2 after transfer on PET. (b) PL spectra of PET substrate (black curve), 1L-MoS2 on PET (red curve), and SLG/1L-MoS2 on PET (blue curve).

Then, the MoS2 film is transferred onto a PET substrate from sapphire using a KOH-based approach.[61] The samples are first spin coated with ∼100 nm poly(methyl methacrylate) (PMMA). This is detached in a 30% KOH solution, washed in deionized (DI) water, and transferred onto PET. The PMMA is then dissolved in acetone. Subsequently, SLG is transferred on the 1L-MoS2 on PET. PMMA is spin coated on the SLG/Cu substrate and then placed in a solution of ammonium persulfate (APS) in DI water until Cu is etched.[51,76] The PMMA membrane with attached SLG is then transferred to a beaker filled with DI water for cleaning APS residuals. The membrane is subsequently lifted with the target PET substrate, having 1L-MoS2 on top. After drying, PMMA is removed in acetone leaving SLG on 1L-MoS2.

Raman and PL characterizations are performed at each step of the SLG/1L-MoS2 assembly on PET, i.e., on 1L-MoS2 transferred on PET and on SLG on 1L-MoS2. This is to confirm no degradation during the fabrication process. Figure a (red curve) plots the Raman spectrum of 1L-MoS2 on PET. The frequency difference between E2g1 and A1g and the FWHMs are preserved on PET, suggesting no degradation. The PL spectrum of 1L-MoS2 on PET is shown in Figure b (red curve). The signal from 1L-MoS2 is convolved within the background due to the PET substrate [Figure b (black curve)]. In order to reveal the underlying PL signature of 1L-MoS2, we use a point-to-point subtraction between the spectrum of 1L-MoS2 on PET [Figure b (red curve)] and the reference PET spectrum [Figure b (black curve)]. Prior to subtraction, the spectra are normalized to the intensity of the Raman peak at ∼1615 cm–1 (corresponding to the peak at ∼560 nm in Figure b), due to the stretching vibrations of benzene rings in PET.[77] As a result, the PL signal of 1L-MoS2 can be seen in Figure a (blue curve) revealing no significant changes after transfer. The subsequent transfer of SLG on 1L-MoS2 does not alter the 1L-MoS2 PL position and line shape [Figure b (blue curve)].

We then characterize the SLG transferred on 1L-MoS2/PET. The intense Raman features of the underlying PET substrate[77] [Figure c (black curve)] mask the SLG peaks. In order to reveal the Raman signatures of SLG, we first measure the reference spectrum, shown in Figure c (black curve), of a PET substrate, using identical conditions as those for SLG/1L-MoS2/PET. We then implement a point-to-point subtraction, normalized to the intensity of the PET peak at ∼1615 cm–1, of the PET reference spectrum, Figure c (black curve), from the total spectrum, Figure c (blue curve). The result is in Figure b (blue curve). The 2D peak retains its single-Lorentzian line-shape with FWHM(2D) ∼28 cm–1, validating the transfer of SLG. The negligible D peak indicates that no significant defects are induced during transfer. Pos(G) is ∼1583 cm–1, FWHM(G) ∼ 17 cm–1, Pos(2D) ∼ 2683 cm–1, and A(2D)/A(G) ∼ 4.8, indicating a p-doping ∼4 × 1012cm–2 (∼250 meV).[53,73]

We then measure the absorptance and transmittance of SLG/1L-MoS2 using a broadband (400–1300 nm) white light from a tungsten halogen lamp. The transmitted light is collected by a 10× objective lens (NA = 0.25) with a Horiba Jobin Yvon HR800 spectrometer equipped with a 300 grooves/mm grating, charged coupled device (CCD) detector and a 50 μm pinhole. Figure a plots the optical transmittance of bare PET (TPET, black line), 1L-MoS2 on PET (TMoS, red line), and the final SLG/1L-MoS2 stack on PET (THetero, blue line) measured in the 400–800 nm wavelength range. Figure b plots the absorptance of 1L-MoS2 on PET (AbsMoS, red line) and of SLG/1L-MoS2 on PET (AbsHetero, blue line), calculated as AbsMoS = (TPET – TMoS)/TPET and AbsHetero = (TPET – THetero)/TPET. The three peaks in Figure b at ∼650 nm (1.91 eV), ∼603 nm (2.06 eV), and ∼428 nm (2.90 eV) correspond to the A, B, C excitons of 1L-MoS2.[75,78] Their positions remain unchanged after SLG transfer. The absorptance difference between the two curves (red and blue) is ∼2.6%, consistent with the additional SLG absorption.[79]

Figure 5

(a) Transmittance of PET (black curve), 1L-MoS2 on PET (red curve), and SLG/1L-MoS2 on PET (blue curve). (b) Absorptance of 1L-MoS2 and SLG/1L-MoS2 as derived from the transmittance measurements. Dashed lines indicate our test wavelength.

The PD area is shaped by etching, whereby SLG extending beyond the 1L-MoS2 flake is removed in an oxygen plasma. The source, drain and gate electrodes are then defined by patterning the contacts area, followed by Cr/Au (6 nm/60 nm) evaporation and lift-off. PDs with different channels lengths (100 μm to 1 mm), 2 mm channel width, and common side-gate electrodes (1 × 0.5 cm) are built (Figure b).

Ref (48) showed that the responsivity of SLG/MoS2 PDs can be enhanced by gating. This induces a stronger electric field at the SLG/MoS2 interface and promotes charge transfer. Various gating techniques have been exploited for GRM-based devices, including conventional Si/SiO2 back-gates,[80] high-k dielectrics (Al2O3, HfO2),[81] chemical dopants,[82] ionic liquids,[83] and PEs.[53,74] In order to gate our SLG/1L-MoS2 on PET, we employ the latter due to its compatibility with flexible substrates[84] and the ability to substantially dope SLG (±0.8 eV)[53,74] using small gate voltages (up to 4 V), unlike other gating techniques, which would require considerably higher biases to reach the same doping.[80,82] We use a PE consisting of LiClO4 and poly(ethylene oxide) (POE).[53,74] We place the PE over both the SLG channel and the side-gate electrode. To evaluate the effect of PE deposition on the SLG channel doping, we use Raman analysis. We get Pos(G) ∼ 1583 cm–1, FWHM(G) ∼ 19 cm–1, Pos(2D) ∼ 2686 cm–1, and A(2D)/A(G) ∼ 5.3, consistent with a small reduction of p-doping to ∼230 meV.[53,73] For electrical measurements we apply −1 V < VGS < 1 V in order to avoid electrochemical reactions, such as hydrolysis of residual water in the electrolyte.[85,86] These may permanently modify the graphene electrode[85,86] and compromise the stability and performance of the device. To control the stability of the PE gating, we continuously monitor the gate leakage current (Igate), keeping Igate < 1 nA throughout the experiments. The devices are tested ∼30 times, showing no degradation in the leakage current over at least six months.

We then characterize the responsivity at 642 nm (∼1.93 eV), slightly above the A exciton peak, where absorption of 1L-MoS2 is maximized (Figure b). At 642 nm the SLG/1L-MoS2 heterostructure shows ∼8% absorptance (Figure b), and the device retains ∼82% transparency (Figure a).

The IDS–VGS measurements in Figure a are done at room temperature using a probe station and a parameter analyzer (Keithley 4200). The PD is illuminated at normal incidence by a collimated laser with Po ranging from 100 μW to 4 mW. At these Po and with VDS = 0.1 V we measure a positive VCNP ranging from ∼0.39 to 0.47 V, indicating an initial SLG p-doping ∼220 meV, consistent with the Raman estimate.

Figure 6

(a) Transfer characteristics as a function of Po. (b) Rext as a function of VGS and Po. Channel length and width are 100 μm and 2 mm, respectively.

Figure a shows that for −1 V < VGS < 0.5 V, where SLG transport is hole dominated, the current decreases under illumination (∼10 μA at VGS = −1 V), as anticipated from the band-diagram of Figure . For VGS > 0.5 V, where SLG is electron-doped, the PD shows a small (up to ∼0.2μA) current increase under illumination. Figure b plots Rext as a function of VGS, as derived from transconductance measurements using:[19]where Ilight and Idark are the PD current under illumination and in dark, |Ilight – Idark| = Iph is the photocurrent defined as the absolute change in the device current upon illumination, Ao is the laser spot area, APD is the PD area, and APD/Ao is a scaling factor that takes into account the fact that only a fraction of optical power impinges on the PD. As expected from the band-diagram in Figure , Rext tends to increase for more negative VGS, up to ∼5.5 A/W at VGS = −1 V, VDS = 0.1 V for Po = 100 μW. By taking into account that only 8% of light is absorbed (Pabs = 0.08 × Po), we derive Rint = Rext/0.08 = 69A/W. Figure b implies that the higher Po, the lower Rext. This can be explained considering that the more photogenerated electrons are injected into the p-doped channel, the lower the electric field at the SLG/1L-MoS2 interface, therefore a reduced injection of electrons causes Rext to decrease.

Given that Rext,Rint > 1A/W, we expect a photoconductive gain (GPD),[19,87] whereby absorption of one photon results in multiple charge carriers contributing to Iph. Our PDs act as optically gated photoconductors, where the SLG conductance is modulated by optical absorption in the 1L-MoS2. In this configuration, the presence of GPD implies that the injected electrons in SLG can recirculate multiple times between source and drain, before recombining with trapped holes in 1L-MoS2. Consequently, GPD can be estimated as the ratio of electrons recombination (τrec) and transit (ttr) times in the SLG channel: GPD = τrec/ttr.[19,21,22,87] For higher VDS, the free carriers drift velocity υd in the SLG channel increases linearly with bias (ohmic region) until it saturates, because of carriers scattering with optical phonons.[88] The linear increase in υd results in shorter ttr, with ttr = L/υd, where L is the channel length.[19,21,22,87] Therefore, GPD is also expected to grow linearly with VDS, providing higher Rext. To confirm the photoconductive nature of GPD in our devices and test the dependence of Rext on VDS, we measure IDS–VDS under illumination at Po = 100 μW for VGS = −1 V and calculate Rext using eq . The IDS–VDS characteristics of the PD show linear dependence, confirming ohmic behavior of the metal-SLG-metal channel.[89] We use VDS < 1 V to keep the device operation in the linear (ohmic) regime and minimize the effects of the nonlinear dependence of υd on VDS (such as velocity saturation) that might appear for VDS > 1 V.[88] As shown in Figure , Rext scales with VDS and reaches ∼45.5A/W (Rint ∼ 570A/W) at VDS = 1 V. This is almost 1 order of magnitude higher than at VDS = 0.1 V, consistent with the similar increase in VDS. These results are at least 2 orders of magnitude higher than semiconductor flexible membranes.[4,15] Furthermore, such a combination of high responsivity with μA range photocurrent surpasses that found in other GRM-based PDs in the visible range.[40−45,47] We also fabricate a control device with a 1L-MoS2 channel only, without SLG. This has Rext ∼ 2 mA/W, which is 4 orders of magnitude smaller than that of our SLG/1L-MoS2 heterostructure. We thus conclude that SLG/1L-MoS2 heterostructures are necessary to achieve high (hundreds A/W) responsivity, due to the presence of photoconductive gain.

Figure 7

Rext as a function of VDS for Po = 100 μW at VGS = −1 V.

To assess the photoresponse uniformity in our SLG/1L-MoS2 heterostructures, we perform photocurrent mapping using the same laser source (642 nm) as for optoelectronic characterizations. We scan areas of 80 × 140 μm (pixel size 3 × 3 μm) at different locations. At each position (pixel) the device photocurrent is measured for VDS = 0.3 V (Figure a). We also collect the backscattered light to give a reflection map (Figure b). Figure a indicates that the entire channel area confined between the source-drain electrodes is photoactive, and shows uniform photocurrent photoresponse with standard deviation ±15%. We thus conclude that interface imperfections (e.g., bubbles, polymer residuals, etc.) have minor effect on the charge transfer process from MoS2 to graphene.

Figure 8

(a) Photocurrent map of channel area, simultaneously measured with backscattered light map. A uniform signal is observed in the channel area (between the electrodes). (b) Reflection map of backscattered light from the device channel. The yellow areas, corresponding to the contact areas, show higher reflectance than the substrate (in blue).

We define GPD as the ratio between electrons recirculating in the SLG channel, thus sustaining Iph, and the initial electron concentration injected into SLG from 1L-MoS2:[48]where q is the electron charge and Δnch is the concentration per unit area and per unit time of the injected electrons. Δnch is equal to the trapped-hole concentration per unit area and per unit time in 1L-MoS2, which is related to a charge neutrality point shift ΔVGS = ΔVCNP in the transfer characteristics. To calculate Δnch, we first write the potential balance in the metal-dielectric-SLG structure. When VGS is applied, it creates a gate-to-channel potential drop (Vdiel), and it induces a local electrostatic potential in the graphene channel (Vch = EF/q):[19,53]where QG and CG are the charge concentration and the geometrical capacitance per unit area associated with the gate electrode, respectively. |QG| = |q·nch|, reflecting the charge neutrality of the gate capacitor, with nch the charge carrier concentration per unit area in the channel induced by VGS. Any variations of nch change QG and VGS. From eq 3 we get:which leads towhere CQ = dQG/dVch is the SLG quantum capacitance[90,91] that characterizes the changes of the channel potential ΔVch as a result of additional gating ΔQG, and (1/CG + 1/CQ)−1 is the total capacitance Ctot.

To calculate QG and Δnch, we first need to find CG and CQ. In PE gating, CG is associated with the electric double layer (EDL) at the SLG/electrolyte interface.[53,90,92,93] The EDL acts like a parallel-plate capacitor with a dielectric layer thickness of the order of the Debye length λD, so that CG = CEDL = ϵϵ0/ λD, where ϵ is the PE dielectric constant, and ϵ0 is the vacuum permittivity. In principle, for a monovalent electrolyte, λD can be explicitly calculated[94] if the electrolyte concentration is known. However, in the presence of a polymer matrix, the electrolyte ions can form complexes with polymer chains,[95] therefore the precise ion concentration is difficult to measure. For PE gating, different EDL thicknesses in the range ∼1–5 nm have been reported.[53,54,92,93] To estimate CEDL in our devices, we take λD ∼ 2 nm[53] and the dielectric constant of the poly(ethylene oxide) matrix to be ϵ ∼ 5,[96] as done in ref (53). As a result, we obtain CEDL = 2.2 × 10–6 F/cm2. This is the same order of magnitude as the SLG CQ.[90] Therefore, the latter cannot be neglected in eq . CQ is given by[90]where ℏ is the reduced Planck constant, υF = 1.1 × 106 m/s is the SLG Fermi velocity,[80,97] and ni is the intrinsic carrier concentration in SLG near the Dirac point induced by charge impurities, defects and local potential fluctuations in the SLG channel.[90,98−100] From our Raman and transconductance measurements we estimate ni ∼ 3 × 1012 cm–2. From eq we then get CQ = 4 × 10–6 F/cm2 at VCNP. From Figure a, and extracting ΔVCNP between the dark current and the transfer curves measured under illumination, and with eq , we get Δnch ranging from 4 to 8 × 1011 cm–2 for Po going from 100 μW to 4 mW. As a result, we obtain GPD ∼ 5 × 104 at VDS = 0.1 V for different Po as shown in Figure . As discussed previously, GPD becomes larger for higher VDS. Thus, we measure an increase of almost 1 order of magnitude (GPD ∼ 4 × 105 at Po = 100 μW) for VDS going from 0.1 to 1 V.

Figure 9

GPD as a function of Po at VGS = −1 V and VDS = 0.1 V.

Finally, we test Iph as a function of bending using a Deben Microtest three-point bending setup (Figure a). In this case, rb = [h2 + (L/2)2]/2h, where L is the chord of circumference connecting the two ends of the arc, and h is the height at the chord midpoint. The plotted values of Iph in the bent state at each rb (Iph,bend) are normalized to the values of Iph measured at rest with the sample in the flat position (Iph,rest). Figure b plots the normalized Iph,bend/Iph,rest for different rb, showing deviations within 15% for rb down to 1.4 cm. This rb is comparable to that reported for semiconductor membrane PDs,[4,15] yet the latter have 2 orders of magnitude lower (<0.3 A/W) responsivities.[4,15] Although our rb is five times larger than that reported for flexible single NW devices,[3,16−18] the area of our PDs (>40 mm2) is at least 6 orders of magnitude larger than that of the NW devices (<5 μm2). To test the device performance upon bending cycles, we first measure the photocurrent at rest (Iph,rest, flat position) and then at the smallest rb (Iph,bend, rb∼1.4 cm), repeating these measurements for 30 bending cycles. Figure c plots Iph,bend/Iph,rest as a function of bending cycles. This shows that our PDs retain stable photocurrent after multiple bending tests with a Iph,bend/Iph,rest standard deviation ±12%.

Figure 10

(a) Schematic three-point bending setup. LD = laser diode; FC= fiber collimator; (b) Iph,bend normalized to the value at rest Iph,rest as a function of rb; (c) Iph,bend normalized to the value at rest Iph,rest as a function of the number of bending cycles.

Conclusions

We reported polymer electrolyte gated, flexible photodetectors, for visible wavelengths with external responsivity up to ∼45.5 A/W, photoconductive gain ∼4 × 105, operation voltage <1 V, and optical transparency >82%. The responsivity is at least 2 orders of magnitude higher than in semiconductor flexible membranes. The devices show stable performance upon bending for radii of curvature larger than ∼1.4 cm. Owing to their responsivity, flexibility, transparency, and low operation voltage, our photodetectors can be attractive for wearable, biomedical, and low-power optoelectronic applications.[11,12,17]

Methods

1L-MoS2 is epitaxially grown by CVD on c-plane sapphire substrates.[61] These are annealed at 1000 °C in air for 1 h after consecutive cleaning by acetone/isopropyl alcohol/DI water. They are then placed face-down above a crucible containing ∼5 mg MoO3 (≥99.998% Alfa Aesar). This is loaded into a 32 mm outer diameter quartz tube placed in a split-tube three-zone furnace. A second crucible containing 350 mg sulfur (≥99.99% purity, Sigma-Aldrich) is located upstream from the growth substrates. Ultrahigh-purity Ar is used as carrier gas at atmospheric pressure. The procedure is to ramp the temperature to 300 °C with 200 sccm Ar flow, set to 300 °C for 10 min, ramp to 700 °C (50 °C/min increase temperature rate) with 10 sccm Ar flow, set at 700 °C for 10 min, cool to 570 °C with 10 sccm of Ar, increase the gas flow to 200 sccm and open the furnace for rapid cooling.[61] SLG is grown on a 35 μm Cu foil, following the process described in ref (51). The substrate is annealed in hydrogen atmosphere (H2, 20 sccm) up to 1000 °C for 30 min. Then, 5 sccm CH4 is added to initiate growth.[51,62] The sample is then cooled in vacuum (1 mTorr) to room temperature and removed from the chamber.