Integrating an electrically active colloidal quantum dot photodiode with a graphene phototransistor

The realization of low-cost photodetectors with high sensitivity, high quantum efficiency, high gain and fast photoresponse in the visible and short-wave infrared remains one of the challenges in optoelectronics. Two classes of photodetectors that have been developed are photodiodes and phototransistors, each of them with specific drawbacks. Here we merge both types into a hybrid photodetector device by integrating a colloidal quantum dot photodiode atop a graphene phototransistor. Our hybrid detector overcomes the limitations of a phototransistor in terms of speed, quantum efficiency and linear dynamic range. We report quantum efficiencies in excess of 70%, gain of 105 and linear dynamic range of 110 dB and 3 dB bandwidth of 1.5 kHz. This constitutes a demonstration of an optoelectronically active device integrated directly atop graphene and paves the way towards a generation of flexible highly performing hybrid two-dimensional (2D)/0D optoelectronics.

A vast number of applications calls for highly sensitive detectors that can sense light from the ultraviolet to the short-wave infrared (SWIR) range, covering a broad spectrum of 300-3,000 nm 1 . Ideally these detector technologies should be based on CMOS compatible platforms for monolithic integration with read-out electronics to cater for high-density, high-throughput and low-cost manufacturing. Graphene 2,3 and colloidal quantum dots (CQDs) 4,5 are two material platforms that have proved to fulfil those requirements [6][7][8][9] . Essential features for highly sensitive photodetectors is the very high quantum efficiency expressed in the number of primary photo-generated carriers collected per incident photon, the low noise and the presence of an additional amplification mechanism, the gain, which is the number of electrically circulated carriers per incident photon. The electrical output of a photodetector is expressed by its responsivity (in A W À 1 or V W À 1 ) which is proportional to both the quantum efficiency and the gain. These parameters eventually determine the sensitivity of a photodetector quantified via the specific detectivity D*. D* is defined as D Ã ¼ ffiffiffiffiffi where NEP is the noise equivalent power, R is the responsivity, A is the optically active area of the detector, f B is the noise bandwidth and S n is the noise spectral density of the detector.
Graphene's unique properties such as high charge carrier mobility, atomically thin profile and broadband absorption has made this material a very attractive platform for photodetectors 10 . Graphene has rapidly proved to be excellent for ultrafast photodetectors [11][12][13] . The high carrier mobility enables ultrafast conversion of photons or plasmons to electrical currents or voltages. Yet, graphene's atomically thin profile inhibits efficient optical absorption. Until now, most sensitive graphene-based photodetectors were phototransistors that relied on the presence of an electrically passive sensitizing layer that consisted of CQDs 14,15 , perovskites 16 , carbon nanotubes 17 or other two-dimensional materials 18,19 . The passive aspect of the sensitizing layer has thus far limited the quantum efficiency (defined as the number of transferred charges per incident photon), linear dynamic range and speed of the detector. Namely, the external quantum efficiency (EQE) of these systems is a trade-off between light absorption, which increases for thicker sensitizing layers, and charge transfer, which decreases for thicker layers as it is limited by carrier diffusion towards the graphene channel. This has limited the quantum efficiency to about 25% for an optimized layer of CQDs of thickness 80 nm 14 , the time response, which is limited by the long trapping times of the sensitizing centres 20 and the dynamic range, which is determined by the density of the trap-state sensitizing centres 21 .
In this letter, we address these three main challenges by presenting a photodetector architecture that is essentially a graphene-CQD photodiode and a high-gain phototransistor. This is realized by transforming the electrically passive sensitizing layer to an active one, where the electric field in the photodiode significantly enhances charge collection due to carrier drift instead of solely relying on diffusion. Our devices operate in the visible, near-infrared and SWIR range and show an EQE as high as 70-80% (limited by reflection), a sub-millisecond temporal response, a gain-bandwidth product on the order of 10 8 and a linear dynamic range in excess of 110 dB. This highly performing detector also exhibits very high sensitivity with experimentally measured D* of 1 Â 10 13 Jones.

Results
Device structure and essence. Our photodetector platform integrates a CQD photodiode atop a graphene transistor. The CQD photodiode consists of an indium tin oxide (ITO) top-contact acting as the cathode of the CQD photodiode, whereas graphene acts as the hole acceptor contact and the charge transport channel for the phototransistor. Figure 1a illustrates a schematic of the hybrid graphene transistor-CQD photodiode detector. Figure 1b shows a top-view optical microscopy image of a typical device and the electrodes in use. The device consists of a graphene channel with a patterned 300-nm-thick CQD layer on top, which is overcoated with a top-contact of an ITO electrode. The active area of the photodetector is formed in the overlapping region of the three layers. The CQD layer is patterned to ensure that no charge carriers can escape from it to the drain contacts omitting the graphene. The read-out of the photodetector is then based on the voltage read-out at the sides of this active area, while source and drain electrodes are biased at a constant current (four-terminal sensing). Such device structure has been employed to eliminate the contact resistance from the measurements and to isolate the measurements over the active area of the photodetector from the area not covered by the ITO electrode. The electronic circuit and diagram of the device is illustrated in the side-view schematic of Fig. 1c.
The essence of the device operation is captured in the band structure, shown in Fig. 2a, representing the operation without and with voltage applied on the ITO top-contact relative to drain electrode (V TD ). Considering first operation without V TD : on carrier photogeneration in the CQD layer, holes are transferred to the graphene via drift from the narrow depletion region formed in the PbS QD solid near the graphene-QD interface and via diffusion further away from the quasi-neutral region. Prior studies have shown that efficient charge collection in this configuration occurs at an optimum thickness of 80-90 nm of PbS QDs, limited by the short carrier diffusion length 22,23 . In the case of a passive CQD sensitizing layer, the electrons remain trapped in the CQD solid and significant gain is achieved 14 as: where t lifetime is the carrier lifetime of the trapped electrons in the sensitizing centres of PbS QDs and t transit is the transit time of holes in the graphene channel. Significant enhancement in charge collection can be foreseen by the application of an electric field by using a transparent top-contact-ITO in this case. The application of V TD bias extends the depletion region in the 300-nm-thick CQD layer, amplifying thus the zone of efficient charge collection and concomitantly quantum efficiency. This enhancement is evident in Fig. 2b, which shows a map of the measured photovoltage of the detector as a function of the back-gate bias (V BD ) and the top-contact bias (V TD ), and Fig. 2e,f which represents line traces extracted from the map. The enhancement of the photoresponse on the application of a top-contact bias exceeds a factor of 4 at BV TD ¼ 1.2 V, as compared with photoresponse at V TD ¼ 0 V. Further increase of V TD results in opposite effects and fully suppresses the photo-induced voltage change at V TD ¼ 4 V. This phenomenon is observed for any V BD value for both n-doped and p-doped graphene channel, and this is due to increased leakage current from the CQD photodiode (as discussed below). It is important to note that changing the potential of the top electrode induces a corresponding change in graphene doping; however, this change is weak and cannot account for the observed change in photocurrent. This is clear in Fig. 2c, which shows the weak dependence of graphene's doping on the applied V TD as a result of suppressed gating via the CQD layer. Figure 2d shows the transconductance of the device (qg/qn) extracted from the corresponding resistance curves that is not affected by the electric bias of the top-contact, from which we conclude that the photoresponse enhancement is not due to increasing transconductance (see Supplementary Fig. 1 for more details).
Photodiode operation. To assess the origin of the enhancement of charge collection in the QD layer, we first operate the ITO-CQD-graphene as a photodiode, which enables us to independently measure the quantum efficiency of our device. To this end, we apply a V TD bias across the PbS QD layer and measure the current I TD as illustrated in the schematic of Fig. 3a, while floating the source V þ and V À contacts of the graphene. This structure operates as a photodiode in which a Schottky junction is formed at the graphene-QD interface. The diode characteristics are shown in Fig. 3b in dark and in Fig. 3c under optical illumination and confirm the formation of a Schottky junction at the graphene-QD interface, also evidenced by the presence of a built-in potential and the sustainment of  The quantum efficiency can be then directly calculated as: where R TD is the photodiode's responsivity (Fig. 3d). In result, EQE TD is measured to be B10% without any bias applied and increases with V TD to about 75% at 1.2 V reverse bias, beyond which it saturates. The inset of Fig. 3d illustrates the corresponding band diagrams corresponding to the two regimes of operation as a function of the reverse bias V TD : for V TD ¼ 0 V the photodiode operates at short circuit conditions and charge collection is facilitated by a small part via carrier drift within the thin depletion region at the graphene-QD interface and by a large part via carrier diffusion within the large quasi-neutral region of the QD layer 24 . With progressively increasing V TD , the depletion region grows and thus enhances the efficiency of the charge collection. The EQE TD value at saturation of 75% is estimated to approach the optical reflection limited value at the front PbS QD-ITO-air interface 23,25,26 , suggesting a nearly 100% charge collection efficiency.
Phototransistor operation. To quantify the impact of the EQE TD enhancement on the performance and speed of the phototransistor operation, we measured the responsivity and time response of the phototransistor as a function of V TD , plotted in Fig. 4a,b. For the bias current I SD of 100 mA, the responsivity reaches a value of 5 Â 10 8 V W À 1 at V TD of 1.2 V following the dependence of EQE TD of the photodiode on V TD (Fig. 3d). For a photodetector with a current read-out circuit this corresponds to a responsivity of 2 Â 10 6 A W À 1 (see Methods). To corroborate our findings we also extracted the EQE of the phototransistor using the formula 27 for responsivity (in units of VW À 1 ) of photoconductive detectors: bandwidth and gain are mainly determined by the slower component. The transit time t transit of the carriers traversing the graphene channel was calculated to be on the order of 1 ns by measuring the carrier mobility of the graphene (see Methods). This yields a photoconductive gain G on the order of 10 5 and gain-bandwidth product of 41.5 Â 10 8 Hz. The extracted EQE of the phototransistor (shown in Fig. 4a) is in very good agreement with the directly measured EQE TD of the photodiode. The responsivity follows the increase of EQE at lower values of V TD and reduces at higher values of V TD . This is due to the accelerated response of the photodetector, resulting in a corresponding reduction of the gain. The hybrid photodetector achieves a concomitant demonstration of responsivity of 5 Â 10 8 V W À 1 , EQE of 75% and bandwidth of 1.5 kHz. An equally important feature of a photodetector is its dynamic range for high-contrast applications such as remote sensing and imaging. We therefore showcase here the significant impact the transformation from a passive to an active sensitizing layer has on the dynamic range of the photodetector. The power dependence of the detector's photoresponse expressed in DV is illustrated in Fig. 4c. In these measurements, we change V BD along with changing V TD to keep the value of R O at 3.7 kO, being the point of optimal photoresponse and highest DC transport mobility. Thus we keep the optimized values of (qg/qn) and R O and compensate for the slight gating effect, induced by V TD bias. Considering first the passively sensitized photodetector (at V TD ¼ 0 V the linear dynamic range is limited to 75 dB with an onset of photoresponse saturation at a power density of about 0.5 W m À 2 . With the application of V TD , we note a significant enhancement of the linear dynamic range of the detector to 110 dB at V TD of 1.2 V. The high linear dynamic range is even more evident in the flat responsivity response across the wide range of power density, as shown in Fig. 4d. This is the result of having a photodiode as the sensitizing element instead of a passive sensitizer. In the latter case, the detector operates as a photoconductor whose dynamic range depends on the density of the sensitizing centres 14,21,28 . To assess the ultimate sensitivity of our detector, we performed independent noise measurements that yield D* of 1 Â 10 13 Jones at a wavelength of 635 nm, and 1 kHz light modulation (see Supplementary Fig. 3). The accompanied enhancement of the detector responsivity as a function of illumination wavelength is illustrated in Supplementary Fig. 4, following the typical absorption spectrum of PbS QDs with a bandgap of 1.2 eV.
To showcase the power of our approach in extending the spectral coverage further into the short-wave infrared, where many applications require high sensitivity low-cost CMOS-compatible photodetectors, we report a phototransistor employing larger PbS QDs with absorption onset at 1.8 mm. The colour map in Fig. 5a shows the photoresponse of such device as a function of V TD and V BG , revealing up to factor 3 enhancement when activating the charge transfer by V TD . Maximum photoresponse in this device was found at V TD ¼ 0.3 V. Detailed characterization of this device is described in Supplementary Note 1. The temporal response of the SWIR-sensitive Irradiance, Wm -2 0 50 25 0 0.5 photodetector is faster compared with the visible-near-infrared devices described above, with a measured 3 dB bandwidth exceeding 4 kHz (Fig. 5b,c). Figure 5d illustrates the spectral photoresponse of the detector at a reverse V TD bias of 0.3 V, which shows that the responsivity of this device exceeds 1 Â 10 8 V W À 1 in the visible range and 4 Â 10 7 V W À 1 in the SWIR range. The calculated EQE is 14% for illumination wavelength of 1,600 nm and 35% for 635 nm, in both cases for the optimum V TD of 0.3 V. This is a sixfold improvement compared with V TD ¼ 0 V.

Discussion
We report a photodetector platform that integrates an electrically active colloidal QD photodiode architecture with a graphene transistor. The former allows the achievement of high charge collection efficiency and therefore EQE, as well as fast photoresponse, whereas the latter enables the simultaneous demonstration of ultra-high gain. The use of a photodiode instead of a passive sensitizing layer gives access to a much broader performance parameter space, as the time response and the gain of the phototransistor is now dependent to the temporal response of the photodiode and the carrier mobility of the graphene channel. It also allows the exploitation of additional carrier multiplication effects previously observed in QDs photodetectors and solar cells 29,30 giving rise to the possibility of higher responsivity without sacrifice of bandwidth. This hybrid architecture finally illustrates the potential of graphene and other two-dimensional materials to be successfully integrated with other optoelectronic materials and open new ways for hybrid two-dimensional/zero-dimensional optoelectronics.

Methods
Graphene FET fabrication. Metal contacts were designed on a precleaned Si/SiO 2 (285 nm) wafer using a direct laser writer lithography. Ti/Pd electrodes (2 nm/40 nm thickness) were e-beam evaporated at a base pressure 10 À 3 mbar and lifted-off in warm acetone bath. CVD graphene grown on copper foil was transferred by a poly(methyl methacrylate) (PMMA)-assisted wet process onto the substrate with metal contacts. Ammonium persulfate was used to etch the copper substrate. After the transfer, the substrate with metal contacts and PMMA/graphene film was dried overnight at room temperature. PMMA was then removed by warm acetone, followed by annealing in forming gas at 300°C to remove residues.
Laser writer lithography and Ar/O 2 plasma etching were used to pattern the graphene layer. Graphene stripes had dimensions of 70 Â 10 nm with 10 Â 10 mm area between V þ and V À electrodes.
PbS QD layer fabrication. The synthesis of PbS nanocrystals was carried out under inert conditions using a Schlenk line as previously described in literature 31,32 . The final PbS oleate-capped nanocrystals were dispersed in octane for device fabrication.
Electron beam lithography (EBL) was used to pattern 30 Â 30 mm windows in PMMA resist at the centre of the graphene Hall bars. Desired PbS CQDs were then spin-coated right on the PMMA. Lift-off in warm acetone performed directly after spin-coating removed PMMA together with the excess of CQDs. As a result, the CQDs remained only in the areas defined by the lithography. The areas were designed to ensure isolation of the graphene channel from the ITO electrode introduced next and to fully cover the central region of the graphene Hall-bar.
PbS QD layers were fabricated by layer-by-layer spin-coating. For each layer, PbS solution (45 mg ml À 1 in octane) was spin-cast onto the substrate at 2,500 r.p.m. for 15 s. Deposition of each layer was followed by spin-casting of EDT solution (2 vol.% in isopropanol) and two rinse-spin steps with octane and isopropanol. All the spin-coating steps were performed under ambient condition The spectral sensitivity is determined by the absorption spectra of the quantum dots that can be readily tuned at their synthesis. The red line is a guide to the eye.
and at room temperature. The resulting thicknesses of nine PbS-EDT layers were 300 ± 20 nm, as determined by a profilometer (Alpha-Step IQ).
ITO electrode deposition. About 150-nm-thick ITO layer was DC sputtered (2 mTorr, 60 W) to form a transparent electrode on top of the PbS layer. EBL lithography allowed to form the overlapping area of ITO, CQD and graphene precisely between the V þ and V À electrodes and avoid shortening of ITO and graphene channels. Each fabricated chip contained up to seven devices with top-electrodes.
Electrical characterization. All transport measurements were carried out at room temperature, in a shielded enclosure and vacuum (10 À 3 mbar) to improve the device stability. Device illumination was done through a viewport by a 635 nm wavelength diode laser, modulated with a mechanical chopper or turned on and off. The laser was coupled to a digital variable optical attenuator to provide output power tunability. In addition, a supercontinuum Vis/NIR laser (SuperK Select) was used to measure photoresponse spectra. Ithaco 1211 amplifier was used to measure I TD . Ithaco 1201 and Stanford SR810 amplifier were used for voltage read-out of photoresponse in combination with a mechanical light chopper with 250-3,000 Hz range.
We utilized NI DAQ 6230 as a voltage and TTI PLH250 as a current source in our measurements. For photovoltage read-out, a constant current of 100 mA was driven between the source and the drain probes, while V þ and V À probes were used to measure the differential voltage drop in the active area of the device.
Responsivity conversion. Directly measured values of responsivity in V W À 1 can be converted into A W À 1 for convenient comparison to photodetectors with current-based read-out and different bias using the formula: where R V is the responsivity in V W À 1 , E SD is the electric field, L is the length of active area, I SD is the source-drain current and R O is the resistance of our detector. E comp is the electric field value used for comparison. Values presented in Fig. 4d were calculated for R O ¼ 3.7 kO, I SD ¼ 100 mA, L ¼ 10 mm and E comp ¼ 5 kV cm À 1 .
Transconductance, mobility and s transit calculations. We calculate the transconductance of our phototransistor (qg/qn), where n is the charge carrier concentration in graphene, by taking a derivative of the R O (V BD ) curve. Capacitive coupling of the 285-nm-thick SiO 2 gate is 7 Â 10 10 cm À 2 V À 1 . Charge carrier mobility m is then calculated for the positive (electron) and negative (hole) transconductance maximums using: qs/qn ¼ e Â m, where e is elementary charge. Maximum transconductance of 2.4 S cm 2 corresponds to 1,500 cm 2 V À 1 s À 1 of electron mobility in graphene.
Transit time t transit of charge carriers in the graphene channel is calculated as: Data availability. The data that support the findings of this study are available from the corresponding authors upon request.