Planar carbon nanotube–graphene hybrid films for high-performance broadband photodetectors

Graphene has emerged as a promising material for photonic applications fuelled by its superior electronic and optical properties. However, the photoresponsivity is limited by the low absorption cross-section and ultrafast recombination rates of photoexcited carriers. Here we demonstrate a photoconductive gain of ∼105 electrons per photon in a carbon nanotube–graphene hybrid due to efficient photocarriers generation and transport within the nanostructure. A broadband photodetector (covering 400–1,550 nm) based on such hybrid films is fabricated with a high photoresponsivity of >100 A W−1 and a fast response time of ∼100 μs. The combination of ultra-broad bandwidth, high responsivities and fast operating speeds affords new opportunities for facile and scalable fabrication of all-carbon optoelectronic devices.

C ombining low-dimensional nanomaterials into hybrid nanostructures is a promising avenue to obtain enhanced material properties and to achieve nanodevices operating with novel principles 1,2 . The family of carbon allotropes, with its rich chemistry and physics, attracts a great deal of attentions in forming novel hybrid nanostructures 2 . In particular, the excellent electrical conductivities and large specific surface areas of twodimensional (2D) graphene and one dimensional (1D) carbon nanotubes (CNTs) have stimulated earlier theoretical and experimental investigations of 3D nanotube-graphene hybrid architectures for hydrogen storage 3 , supercapacitors 4 and fieldemitter devices 5 . Recently, ultrathin CNT layers have been used as a reinforcing component for centimetre-sized chemical vapour deposition (CVD) grown graphene, where hybrid films with enhanced in-plane mechanical strength, uncompromised electrical conductivity and optical transparency are formed 6 . Even more remarkably, the synthesis of covalently-bonded single-wall CNTs (SWNTs) and graphene hybrid film is recently achieved and its use as flexible transparent electrodes is demonstrated 7 . The simple and scalable route for fabricating quasi-2D all-carbon hybrid films is envisioned to offer new opportunities beyond mechanical and nanoelectronic applications 2 .
Both graphene and CNTs exhibit intriguing optical properties, such as broadband and tuneable light absorption, which make them promising materials for photodetectors 8,9 . Graphene has proved excellent for ultrafast and ultrasensitive photodetectors. However, the relatively low absorbance of a single sheet of carbon atoms adversely limits the photoresponsivities of the earlier metal-graphene-metal devices (B10 À 2 A W À 1 ) 10,11 . A number of heterogeneous schemes employing plasmonic resonance 12,13 , microcavities 14 , evanescent-wave coupling 15 and conventional semiconductor nanostructures 16 have been explored and lifted the photoresponsivity of graphene photodetectors to B8.6 A W À 1 (ref. 13). However, these approaches offer only modest responsivity enhancement and introduce fabrication steps that are not manufacturing scalable. Graphene photodetectors based on a photogating mechanism, that is, by depositing a semiconducting quantum-dots (QDs) overlayer, is by far the most superior method in terms of responsivity enhancement and ease of fabrication 17,18 . But the spectral coverage is limited by the absorption range of the QDs and enhancement is effective only at extremely low light intensities (that is, pico-Watt or 10 À 12 W) due to the use of exfoliated graphene flakes (lateral sizeB5 mm) 16 . Although an astonishing gain-bandwidth product of B10 9 has been demonstrated 16 , such systems are intrinsically slow with electrical bandwidth around 1-10 Hz, as limited by the low carrier mobility and long carrier recombination time of the QDs 17 .
SWNTs, on the other hand, are p-conjugated, 1D structures with nanometre diameters 19 . They exhibit either metallic or semiconductor features depending on the tube chiralities 20 . Both metallic and semiconducting SWNTs are effective light absorbers in a wide spectral range 21 and intrinsic mobility in semiconducting SWNTs is estimated to be as high as 10 5 cm 2 V À 1 s À 1 at room temperature 22 . Due to 1D quantum confinement, photoexcitations in SWNTs demonstrate rich physics 9 . For example, localized excitons with large binding energy is the main excitation for low-energy transitions (that is, S 11 ) in semiconducting SWNTs 23 , while metallic tubes can act as an efficient charge transport channel for SWNT ensembles to adjacent conducting media 24 . Although extensive efforts have been directed towards the realization of nanotube photodetectors, the weak photoresponse (o10 À 3 AW À 1 ) of single-tube photovoltaic devices greatly limits their practical use 25 , while nanotube bolometric detectors are known to suffer from slow response time 26 . Recently, efficient charge transfer has been identified at junctions formed by graphene and SWNTs 24 . The intimate electronic coupling between the two sp 2 -hybridized carbon allotropes, combined with the various strategies available for structural and chemical engineering of the interfacial electronic properties 27,28 , make such all-carbon hybrid an excellent candidate in enabling phototransistors with balanced and tuneable gain-bandwidth characteristics.
Here we demonstrate a proof-of-concept photodetector based on a planar atomically thin SWNT-graphene hybrid film. In our design, enhanced broadband light absorption is achieved in the hybrid film. Compared with pristine graphene photodetectors which exhibit only weak Schottky junction and high Auger recombination rate, the large built-in potential at the 1D-2D interface promotes effective separation of electron-hole pairs and reduces recombination of spatially isolated photocarriers. Furthermore, the trap-free interface enables a relatively fast operation rate. The devices exhibit a significant photoconductive gain of B10 5 , together with a high electrical bandwidth ofB10 4 Hz (response timeB100 ms) across visible to nearinfrared range (400-1,550 nm). The reported device constitutes a first implementation of large-area, quasi-2D SWNT-graphene hybrid film for optoelectronic devices and is envisaged to be important for optical communication, spectroscopy, remote sensing and high-resolution imaging applications. Equally important, we demonstrate for the first time the possibility of harvesting robust excitons widely supported in 1D systems by a 2D layered material. Our results not only open up new avenues for studying fundamental carrier transport and relaxation pathways in nanometre-scale, 1D van der Waals junctions, but also pave the way for constructing high-performance optoelectronic heterostructures by planar 1D-2D hybrid building blocks, other than the use of purely 2D layered materials.

Results
SWNT-graphene phototransistor fabrication. We fabricated proof-of-concept phototransistors using the SWNT-graphene hybrid film, as illustrated schematically in Fig. 1a. The degenerately n-doped Si substrate with 285 nm thermal oxide was used as the back-gate. Wafer-scale SWNT-graphene hybrid film is fabricated by transferring CVD grown graphene onto an ultrathin layer of SWNTs formed on the SiO 2 /Si substrate (see Methods). Figure 1b,c shows a representative tapping mode atomic force microscope (AFM) image of the hybrid film, where nanotubes form filament structures similar to the vein-like support as in ref. 6. Figure 1d shows the height profile along the red line in Fig. 1b, where unbundled SWNTs with diameters in the range of 1.0-1.6 nm are observed. This is in agreement with a mean tube diameter of B1.4 nm as inferred from the Radial Breathing Mode position of the Raman measurement (Supplementary Note 1, Supplementary Fig. 1). To investigate the physical characteristics of the junctions formed at graphene and SWNTs, a section of graphene is mechanically removed using a tape stripe (dark brown area, Fig. 1c), so that the height of the SWNT-graphene junction with respect to the substrate can be directly measured. This yields a height of B2.2 nm for the SWNT-graphene junction (Fig. 1e), while the isolated SWNT height as measured from the uncovered portion indicates a height of 1.2 nm ( Supplementary Fig. 2). The height of graphene on SiO 2 /Si substrate is measured to be B0.8 nm, indicative of single atom layer 29 . The inferred offset at the SWNT-graphene interface is B0.6 nm, which is larger than the interlayer distance of graphite (B0.335 nm), providing evidence for a long-range van der Waals interactions 30 . Figure 1f shows the optical absorption spectrum of the graphene and the SWNT-graphene hybrid film, illustrating enhanced broadband absorption due to the incorporation of SWNTs. The S 11 and S 22 bands are found to be located at B1,800 and B1,000 nm with higher SWNTs loadings ( Supplementary  Fig. 3), which agree well with the tube diameter distributions.
To probe the electrostatic doping scenario at the 1D van der Waals junctions formed by graphene and SWNTs, transfer curves of the devices before and after forming the hybrid film are compared. It is observed that the Dirac point of the hybrid transistor shifted from 2 to 17 V, indicating p-type doping of the graphene sheet by the SWNTs layer. It should be pointed out that due to the electronic inhomogeneity of the SWNTs, the electrostatic doping of graphene is expected be an overall effect from both metallic and semiconducting SWNTs. As metallic SWNT is found to form Ohmic contact with graphene 31 (that is, there is no p-n junction or Schottky barrier), photoresponse of the hybrid channel is expected to arise from semiconducting SWNTs with suitable band alignment with graphene. Figure 1h provides a representative band alignment that supports the electrostatic-doping scenario at the graphene semiconducting SWNT interface (where constituent semiconducting SWNTs have work functions larger than that of graphene), that is, photogenerated electrons transport from SWNTs to graphene under the built-in field at the SWNT-graphene junction. In addition, the relatively low contact resistance for SWNTgraphene junctions (o0.1 MO) 31  It is worth noting that in fabricating our device the SWNT layer is placed beneath graphene to ensure the formation of high quality SWNT-graphene interface, that is, to avoid degradation caused by the residual poly(methyl methacrylate) used in the graphene transfer procedures 36 . Photoresponse of the SWNT-graphene hybrid device to 650 nm visible light is measured. Figure 2a shows a set of source-drain currents (I SD ) as a function of the back-gate voltage (V G ) under different illumination levels, where continuous negative shift of the voltage for Dirac point (the charge neutrality point) is observed, confirming a photocurrent generation mechanism as depicted by Fig. 1h. Control devices with only SWNT layer as the channel are found to be insulating and exhibit no detectable photoresponse, probably due to the low density of nanotubes on the substrate (Inset of Fig. 1g). The on/off ratio is about 4, suggesting graphene is the primary conductive channel 37 . For 650 nm light (1.9 eV) with photon energies larger than the S 22 bandgap of the semiconducting SWNTs (B1.2 eV), free electrons can be directly excited from the valence band to the conduction band. Photogenerated electrons in SWNTs are transferred to the graphene channel due to the built-in electric field, while the holes are trapped in the SWNTs due to the potential barrier at the interface. The conductance of the graphene channel decreases for V BG oV D where the carrier transport is hole-dominated, but  Fig. 4a). In Fig. 2c, linear scaling of the photocurrent with source-drain bias voltage (V SD ) for different optical powers at V BG ¼ 0 is clearly observed. Figure 2b also plots the photoresponsivity as a function of the illumination power. To faithfully illustrate the performance of the device we define the responsivity using the incident power rather than the absorbed power. At an illumination power of B0.2 mW, the photodetector exhibits a drastically enhanced responsivity of B120 A W À 1 , as compared with a bare graphene device B10 -2 A W À 1 . Since the responsivity does not show sign of saturation, higher photoresponse on the order of B1,000 A W À 1 is expected at lower excitation power levels ( Supplementary  Fig. 4b).
To rule out potential contributions from thermal effects, including bolometric contribution from the SWNT layer 26 and photothermoelelectric contribution from the metal electrodes 38 , we measured the I SD -V G at temperatures from 53 to 273 K (Supplementary Fig. 5 and Supplementary Note 2). Source-drain current increases with an increasing temperature from 53 to 173 K, and a slight decrease in the range 173-273 K, as expected for SWNT response 26 . The rather small temperature dependence of I SD allows us to rule out resistance increase caused by heating of SWNTs as the mechanism for photocurrent generation. Light induced heating of source or drain metal contacts is reported to lead to a temperature gradient, resulting in a photothermoelectric contribution to the photocurrent 38 . However, we did not observe detectable photocurrent in our devices by illuminating visible light (650 nm) on either metal contact. Therefore, the photoresponse of our device is ascribed to a photogating effect, where channel resistance is modulated through capacitive coupling caused by trapped photogenerated charges.
We further verify the operation mechanism of our devices by correlating the gate dependent photocurrents with the band alignment at the SWNT-graphene junctions. As schematically illustrated in Fig. 2d, incident photons excite ground-state electrons of SWNTs into excited states, and then electron-hole pairs are formed at the SWNT-graphene interface. For V G o0, the Fermi level of graphene is lowered, which leads to a steeper upward band bending and an enhanced built-in electric field at the SWNT-graphene junction. This facilitates more photoelectrons to transfer from SWNTs to graphene, leading to an increasing photocurrent. As the magnitude of negative V G continues to increase, the trapped holes in the valence band of SWNTs begin to tunnel through the thinned barrier into the graphene channel, resulting in a decreased photocurrent. In the case of V G 40, the Fermi level of graphene is raised up, the potential offset for electrons is effectively decreased. Therefore, the photocurrent at a given positive gate voltage is smaller than that for a negative gate voltage of equal amplitude. When V G is sufficiently high, the SWNT energy band will flip to bend downwards, allowing photogenerated holes to transfer to the  graphene channel. This is the reason why the conductivity decreases for V G 440 V, and the photocurrent subsequently switches sign (Supplementary Fig. 4a).
Fast response time. Response time is another key figure of merit for photodetectors and is also relevant in revealing the physical mechanism of the device operation. Figure 3a shows the on/off source-drain current of the photodetector at an incident power of 440 mW (650 nm, V SD ¼ 1.2 V). The temporal photoresponse was measured at 173 K to suppress the outside interference including scattering centres from the substrate and charge-trapping surface states 31 . After hundreds of on-off cycles, photocurrent level is well retained, demonstrating good reliability and reversibility of our devices. At the same experimental conditions, we did not observe any detectable photocurrent in a control device with a bare graphene channel. Figure 3b gives the high-resolution temporal response (sampling interval 100 ms). To encourage photocarriers separation we apply a 5 V source-drain voltage. Sharp responses are observed from which we estimate the rise time and the fall time to be B100 ms. The operation speed of our device is 41,000 times faster than a graphene-QDs detector with similar photosensitive area 17 , and can be further reduced by scaling the device dimensions. The high mobility and fast transfer of carriers within the SWNT-graphene hybrid film are believed to account for the excellent temporal performance of our devices. The EQE (External Quantum Efficiency) was estimated using the parallel plate capacitor model 18 , QE¼ DV Â C ox /f photon , where DV is the Dirac voltage shift, C ox is the capacitive coupling of the back-gate for 285 nm thick thermal oxide (7 Â 10 10 cm À 2 V À 1 ), f photon is the photon flux. As shown in Fig. 3c, for 650 nm light our devices show a power-dependent EQE ranging from 34 to 12% for power illumination o1 mW, and an EQE of o0.7% with incident power 425 mW. The peak value for the QE is slightly higher than the 25% achieved with QDs 18 , demonstrating the high efficiency of photocarriers generation and transport in the SWNTs-graphene hybrid. From the transfer curve measured in the absence of light, it is estimated by the Kim model 39 that the field-effect mobility for electrons and holes are 3,920 and 3,663 cm 2 V À 1 s À 1 , respectively. This is superior to graphene sensitized with PbS QDs 18 . Such high mobilities confirm that graphene can be doped by 1D SWNTs, yet preserve its outstanding electronic properties, facilitating high gain and fast response time. The carrier transit time of our device is estimated to be on the order of 10 À 9 s (based on a carrier mobility of B3,920 cm 2 V À 1 s À 1 , a channel length of 70 mm and a bias of 5 V), thus the photoconductive gain G ¼ t lifetime /t transit (the ratio of the lifetime of the trapped holes over the electron transit time from source to drain) is on the order of B10 5 using a trapped carrier lifetime of B100 ms. Therefore the gainbandwidth product of our SWNT-graphene device is on the order of 1 Â 10 9 Hz (given by the product of B10 5 gain and B10 4 Hz of bandwidth), which is quantitatively similar to the PbS QD-graphene hybrid photodetector 18 .
Ultra-broadband photoresponse. Characteristics of broadband photoresponse of our device over a range of incident wavelengths is investigated by using multiple laser diodes operating from visible to the near-infrared range (405, 532, 650, 980 and 1,550 nm). The responsivities for various wavelengths are summarized in Fig. 3d, where ultra-broadband spectral coverage is observed. Higher responsivities were obtained at lower incident power levels. At a moderate illumination power of B0.3 mW, all wavelengths exhibit a responsivity 410 A W À 1 , a typical specification for commercial photodetectors. The responsivity value is It is shown that the responsivity of the metallic tube-based devices is B5% of that of (6,5) chirality tube-based devices.
higher for shorter wavelengths than for longer ones ( Supplementary Fig. 6), consistent with the absorption spectrum of the SWNTs used, which is the usual case for phototransistors based on a photogating mechanism 18 . From the wavelength dependence, even higher photoresponsivities are expected for ultraviolet wavelengths.

Discussion
The essential ingredients of our devices are the sensitivity to external electrostatic in graphene conductivity, the broadband and tuneable photon absorption in both graphene and SWNTs, and the high-speed transportation of photocarriers in the planar SWNT-graphene hybrid. To resolve the impact tube diameters and chiralities may have on the photodetection processes, we fabricated control devices using highly purified metallic tubes and (6,5) Figure 3e compares the photoresponsivities of purified metallic and semiconducting SWNTs-based devices, and the responsivity of the metallic tube-based devices is seen to be B5% of that of (6, 5) chirality tube-based devices. The results clearly identify semiconducting tubes as the main contribution for photocurrents. In the meantime, unlike nanoelectronic devices where distinct SWNT chirality is desirable, our devices will not degrade by the presence of multiple species of SWNTs, similar to SWNT-based nonlinear absorption devices 40 . The mitigation of requirements for tube chiralities also facilitates the practical and scalable fabrication of such films. The performance of the proof-of-concept photodetector based on SWNT-graphene hybrid film can be further optimized by multiple strategies including controlling the density, chiralities 41 and alignment direction of the nanotube layer and by using refined SWNT-graphene hybrid film fabrication methods 6,7 . In summary, by combining large-area CVD grown graphene with atomically thin SWNT layer, we have formed a quasi-2D allcarbon hybrid film, which exhibits strikingly enhanced photodetection capabilities superior to either graphene or SWNTs. In contrast to previous charge trap based phototransistors, such hybrid film based photodetectors exhibit not only a significant photoconductive gain of B10 5 , but also a fast response time of B100 ms and an ultra-broadband sensitivity across visible to near-infrared, covering the telecommunication band at B1.5 mm.
Benefiting from the solution processability of SWNTs, large-scale growth and transfer of graphene, as well as compatibility with standard photolithography, our devices exhibit great potential for use in photodetection applications requiring large sensing area, high photoresponsivities, video-frame-rate processing and substrate flexibility. The results demonstrated here represent a significant step towards facile and scalable fabrication of highperformance optoelectronic devices using all-carbon hybrid nanostructures and have substantial implications for fundamental investigation of the van der Waals interactions between layered materials and in-plane quantum wires, in the 1D limit.

Methods
Phototransistor fabrication. The graphene samples were grown on copper foil by CVD method, and Raman spectroscopy combined with optical microscope characterizations point to a defect-free single-layer sample. We use single-wall CNTs (SWNTs) from a commercial supplier (Carbon solutions Inc.). The phototransistors are fabricated as follows: SWNTs suspensions are produced by tipsonicating 1 mg nanotube soot in 10 ml NMP (N-methyl-2-pyrrolidone). The resultant suspensions are centrifuged with 10,000 g for 1 h before the supernatant is collected for the fabrication of SWNTs thin film on SiO 2 /Si wafer. CVD graphene is transferred on top of the SWNTs layer using the poly(methyl methacrylate) supported procedures. Electrodes are patterned by standard photolithography. Different metal composition (Ti/Au and Pd/Au) are subsequently deposited. The graphene channel is patterned by another photolithography and oxygen plasma etching.
Photoresponse characterization. For photoresponse characterization, we used 405, 532, 650, 980 and 1,550 nm laser diodes, respectively. The beam is guided through an optical fiber with a FC/PC ferrule and is subsequently incident onto the channel of the devices without focusing. The beam at the device was measured to be Gaussian-shaped with a diameter of about 300 mm at 650 nm illumination. The area of the channel is o100 Â 40 mm. Therefore, when calculating the photoresponsivity, we used only the portion of the incident light intensity that overlaps with the channel area. The electrical measurements were carried out in a closed cycle cryogenic probe station under vacuum (10 À 6 torr) at room temperature and the data were collected by a Keithley-4200 semiconductor parameter analyzer. The temporal photoresponse was measured at 173 K for suppressing the outside interference.
AFM measurements. AFM measurements were performed using an Asylum Research Cypher AFM operating at room temperature and ambient conditions.
Raman and optical absorption spectra. Raman measurements were performed on a Horiba Jobin Yvon LabRAM HR 800 system using a 514 nm excitation laser operating at 1 mW, Â 100 objective lens with about 1-mm diameter spot size, and 1,800 lines per mm grating with about 0.45 cm À 1 spectral resolution. Optical absorption spectrum was measured using a Shimadzu UV-vis-NIR UV-3600 spectrophotometer.