Thermoelectrically driven photocurrent generation in femtosecond

Jul 18, 2018 - In this work, we utilize the ultrafast laser functionalization of single-layer CVD graphene for highly desirable maskless fabrication o...
0 downloads 0 Views 3MB Size
Article pubs.acs.org/journal/apchd5

Cite This: ACS Photonics 2018, 5, 3107−3115

Thermoelectrically Driven Photocurrent Generation in Femtosecond Laser Patterned Graphene Junctions Aleksei V. Emelianov,*,† Dmitry Kireev,‡ Andreas Offenhäusser,‡ Nerea Otero,§ Pablo M. Romero,§ and Ivan I. Bobrinetskiy† †

National Research University of Electronic Technology, Zelenograd, 124498, Russia Institute of Complex Systems - Bioelectronics (ICS-8), Forschungszentrum Jülich, 52425 Jülich, Germany § Laser Applications Centre, AIMEN, Porriño 36410, Spain ACS Photonics 2018.5:3107-3115. Downloaded from pubs.acs.org by UNIV OF SOUTH DAKOTA on 08/21/18. For personal use only.



S Supporting Information *

ABSTRACT: Single and few-layer graphene photodetectors have attracted much attention in the past few years. Pristine graphene shows a very weak response to visible light; hence, fabrication of complex graphene-based detectors is a challenging task. In this work, we utilize the ultrafast laser functionalization of single-layer CVD graphene for highly desirable maskless fabrication of micro- and nanoscale devices. We investigate the optoelectronic response of pristine and functionalized devices under femtosecond and continuous wave lasers irradiation. We demonstrate that the photocurrent generation in p−p+ junctions formed in single-layer graphene is related to the photothermoelectric effect. The photoresponsivity of our laser patterned single-layer graphene junctions is shown to be as high as 100 mA/W with noise equivalent power less than 6 kW/cm2. These results open a path to a low-cost maskless technology for fabrication of graphene-based optoelectronic devices with tunable properties for spectroscopy, signal processing, and other applications. KEYWORDS: graphene, photocurrent, photodetector, femtosecond laser, photothermoelectric effect

G

in graphene, such as nanoribbons, nanomeshes, or nanodots.13,14 However, these techniques demand unconventional lithographic methods and chemical processing that increase the cost of the process and greatly alter the properties of untreated graphene.15 High energy exposure methods, such as plasma treatment, typically result in controllable defect generation in graphene lattice, resulting in an effective separation of photoinduced charge carriers.16 The fabrication still includes preliminary mask patterning of the device. Restoring the pristine graphene properties after resist based lithographic techniques is a challenging task and the use of maskless methods are preferable and economically viable. The laserbased methods provide selective local tailoring of graphene properties in situ.17 Besides using laser for patterning, it can even be used to facilitate direct transfer of graphene onto arbitrary substrates.18 The completely laser-based technology opens the road toward maskless direct writing technology for graphene-based device processing. The distinct photoresponse after continuous wave (CW) laser treatment was previously observed in exfoliated graphene.19 It was found that laser treatment can cause the electrostatic gating effect in graphene field effect transistors

raphene is one of the most promising materials for advanced electronic applications, due to its extraordinary chemical and physical properties.1,2 The interaction between graphene lattice and photons opens the way for the development of new optoelectronic and photovoltaic devices.3 The photoresponse in graphene is a very promising effect that can be important for the integrated optoelectronic devices because of the wide range of sensitivity from UV to far-infrared wavelengths,4 as well as the low power consumption and high efficiency.5 The sensing mechanism for visible light irradiation is based on the interband transition between conduction and valence bands in graphene and can involve photovoltaic, photothermoelectric and bolometric effects.6 Nevertheless, the photogenerated charge carries typically undergo ultrafast recombination7 if the built-in electric field for their separation is not enough. Additionally, inefficient light conversion is caused by the high transparency of graphene.8 There are, however, different ways of physical graphene modification9 or electrostatic substrate coupling,10 which were suggested to enhance the optical graphene sensitivity. The main challenge in the graphene-based optoelectronic devices is fabrication of graphene junctions11,12 required to facilitate photovoltaic or thermoelectric effects. One of the most promising ways of local energy profile engineering in graphene structures is the generation of quantum confinement © 2018 American Chemical Society

Received: March 18, 2018 Published: July 18, 2018 3107

DOI: 10.1021/acsphotonics.8b00350 ACS Photonics 2018, 5, 3107−3115

ACS Photonics

Article

Figure 1. (a) Device geometry (top) and schematic band structure (bottom) of each region of graphene. (b) In situ resistance change of GFET-8 upon fs-laser irradiation (shown as green area) with a photon flux of 4.6 × 1025 photons/(s·m2). Optical photographs before and after laser treatment (inset) of the GFET-8. The dashed line in the left inset illustrates the laser path. (c) Micro-Raman map of ID/IG ratio of GFET-17 treated with a flux of 4.6 × 1025 photons/(s·m2). The dashed lines show the edges of the graphene channel. (d) Raman spectra for pristine (bottom curve, black) and functionalized (top curve, red) parts of GFET-8 taken from spots that are marked in (b).

10 Hz) less than 1.3 × 10−9 A/√Hz for functionalized with oxygen species GFETs is very promising for optoelectronic application in integrated device schemes. Upon analysis of experimental data, the photothermoelectric effect is proposed as the main mechanism of photocurrent generation in fs-laser processed GFET junctions.

(GFETs) by transfer charges to trap states in the gate oxide20 or chemical doping by generating oxidative groups on graphene.19 Moreover, the sensitivity of chemically modified graphene can be increased by varying the gate voltage.21 The dominant mechanism of photocurrent generation in graphene with modified region is based on different thermoelectric properties of materials on the both sides of the junction.7,22 Thus, the photothermoelectric effect typically dominates over the effect of built-in electric field even at low bias.22 Local laser processing of a GFET’s channel can provide different junction formation, including p−n, p−p+, or n−n−. Nevertheless, the CW laser-based process is time-consuming, as it demands several minutes of continuous laser irradiation of a single point resulting only in ∼0.2 mA/W photoresponsivity.19 Nanosecond pulsed laser processing provides a scalable decrease of processing time as it was previously shown for photosensitive Schottky junction fabrication in epitaxial graphene.23 Moreover, high energy ultrafast laser pulses (pico- and femtosecond) are shown to provide selective tailoring of graphene properties with resolution down to 200 nm,24 high precision defect density control, and sharp edges of the processed areas. Ultrafast laser pulses can easily tune the chemical potential of graphene by covalently grafting the molecules to graphene.25 Thus, the ultrafast pulse laser treatment technology paves a promising way to further increase the efficiency of photocurrent generation in graphene-based devices. In this report, we describe the relatively simple route of physical and chemical local modification of CVD-grown graphene through femtosecond pulsed laser (fs-laser) treatment to increase the photoresponse in GFETs. The achieved photoresponsivity of 100 mA/W with low-frequency noise (at



RESULTS AND DISCUSSION Fabrication of p−p+ Junctions in GFETs via fs-Laser Irradiation. A total of 22 GFETs with different channel geometries were processed across the channel using 280 fs pulse laser as schematically shown in Figure 1a. We chose the pulse energy of 2 nJ and the repetition rate of 500 kHz and varied the photon flux by changing the number of pulses per μm2 (see Supporting Information for photon flux calculation and Table S1 for details). As is clearly seen from the resistance over time plot (see Figure 1b), measured during the local treatment of a GFET-8, the changes of graphene properties during laser processing are remarkable. The in situ resistance change during laser processing provides information about the linearity of graphene modification and local probing of graphene electrical properties. Resistance growth during GFETs treatment strongly depends on photon flux. The increase of the accumulated energy fluence raises the probability of partial thermal ablation of defects in graphene, such as wrinkles, bilayer islands, or atomic vacancies and causes a nonmonotonic growth of resistance during fs-laser treatment. The micro-Raman map (Figure 1c) was measured for another structure with the same treatment parameters and furthermore confirmed the local modification of the graphene lattice. Moreover, the increase of the D band intensity, the blue shift 3108

DOI: 10.1021/acsphotonics.8b00350 ACS Photonics 2018, 5, 3107−3115

ACS Photonics

Article

Figure 2. (a) AFM image of a laser irradiated GFET-8 with a photon flux of 4.6 × 1025 photons/(s·m2). Zoomed area of functionalized and attached (left inset) region of the channel and profile of processed area (right inset). (b) R−VGS curves before (bottom curve, black) and after (top curve, red) laser treatment of GFET-11 processed with a photon flux of 4.6 × 1025 photons/(s·m2) measured using liquid gating. (c) G peak position and total charge carrier concentration across the GFET-8 for pristine (black triangles) and processed (red circles) graphene channels. (d) G peak position change (black squares) and relative change of CNP (blue circles) versus photon flux and number of pulses per μm2 recorded for GFETs-1−12.

= 2/18 μm, VDS = 100 mV), the charge carrier mobilities were ∼1400 and ∼1000 cm2/(V·s) for holes and electrons, respectively. After the local functionalization through fs-laser irradiation of the same GFET-11, at the points of maximum transconductance we observed that the mobilities are almost an order of magnitude lower (∼170 and ∼90 cm2/(V·s) for holes and electrons, respectively). The calculations of transconductance and hole and electron mobilities can be found in Supporting Information (Table S3, Figure S4). As it is visible from Figure 2d, the prominent changes in graphene properties start from a photon flux of 2.5 × 1025 photons/(s·m2). If we assume that graphene absorbs only 2.3% of initial power,34 the calculated peak energy fluence for 5000 pulses per μm2, at which the functionalization becomes visible, is equal to 70 mJ/cm2. After a photon flux reaches 5 × 1025 photons/(s·m2), the number of defects in the processed area becomes significant to induce graphene disassembly.9 At this point, the energy fluence (equal to 220 mJ/cm2) slightly exceeds the theoretically predicted35 and experimentally proven36,37 graphene ablation threshold values of 185 and 200 mJ/cm2 for a single pulse, respectively. In our work, the threshold values are larger due to irradiating GFETs with a lower power density and a larger number of pulses. The threshold photon flux for our system was observed to be ∼1.5 × 1026 photons/(s·m2). We compared the position of vibrational G band as well as CNP shift (VCNP) of the processed GFETs and found the optimal parameters for processing with the photon flux of 4.6 × 1025 photons/(s·m2) or 9000 pulses per μm2. Thus, the simple and effective maskless technology was applied to produce the local chemical modification with estimated parameters in GFET channels. Photocurrent Generation in GFETs with p−p + Junctions. To investigate the optoelectronic response and the origin of the generated photocurrent in fabricated

of the G and 2D bands, and the appearance of the D′ band after fs-laser irradiation (Figure 1d, Supporting Information, Table S2) prove the two-photon local oxidation of graphene.24,26 The ID/ID′ ratio of ∼13 shows that defects in the processed area are mostly sp3 type.27,28 Furthermore, based on previous works, we can clearly state that the majority of groups, grafted to the graphene plane, are epoxy (C−O−C) and hydroxyl (C−OH).24,26 The process of fs-laser pulses interaction with the graphene lattice is rather complicated. It includes mechanical, chemical and structural modification. Notably, we observed the decrease of the channel height in the laser processed area (Figure 2a). We believe that this is related to the thermal removal of polymer residuals and trapped water between graphene and the substrate29 that leads to a closer contact, enhancing charge transfer from graphene to trapped states in the substrate during laser irradiation. The strong change in graphene properties was also observed on a rather localized level, through STM spectroscopy measurements (Supporting Information, Figure S3), which shows the non-Ohmic behavior of the I−V curve in the area of fs-laser treatment. R−VGS curve of the GFET after irradiation is drastically altered (Figure 2b) and two charge neutrality points (CNP) were occasionally observed for higher photon fluxes. These peaks represent two regions with different doping levels (CNP1 and CNP2),30,31 showing the formation of p−p+ junctions due to laser-induced oxidation. The strong blue shift of the G band (Figure 2c) in the processed area indicates an increase in hole concentration caused by fs-laser local oxidation. The shift of the G band can be quantitatively translated into a charge density.32,33 We found that fs-laser treatment of single-layer graphene increases the charge density of up to 1 × 1013 cm−2 (Figure 2c). The significant change in hole concentration at the edges of irradiated region defines sharp p−p+ junctions. For pristine graphene (GFET-11, W/L 3109

DOI: 10.1021/acsphotonics.8b00350 ACS Photonics 2018, 5, 3107−3115

ACS Photonics

Article

Figure 3. (a) Photocurrent profile of GFET-8 before (black triangles) and after (red circles) partial functionalization measured across the channel at 100 μW laser power. (b) Photocurrent as a function of incident power density measured at pristine GFET-8 (black triangles, the laser is focused to the center of the graphene channel) and functionalized GFET-8 and GFET-10 (blue squares and red circles, the laser is focused to the p−p+ junction in the partly functionalized channel), respectively. Colored background is given to show the range of possible applications in terms of their requirements for power density. (c) Photocurrent in GFET-10 at VGS = 0 V treated with a photon flux of 7.2 × 1025 photons/(s·m2) as a function of VDS upon varying 532 nm laser power from 0.1 μW (black) to 100 μW (blue) and 300 μW (red). (d) Photocurrent generation and carrier relaxation dynamics upon a photon flux of 5 × 1024 photons/(s·m2) in GFET-8. The times τ1 and τ2 are intrinsic and trapped charge relaxation times, respectively. Drain-source bias, VDS, was always set at 0.5 mV.

photodetector. For our pristine graphene, α is equal to 0.25 ± 0.02. Such a low value of the exponent can be related to the high defect density of graphene and the symmetric band profile of the device. However, in the functionalized GFETs the exponent rises with the photon flux and reaches 0.36 ± 0.01 for the flux of 7.2 × 1025 photons/(s·m2). We associate the increase of α with different relaxation dynamics of carriers in pristine and functionalized graphene. Our experimental results have a good correlation with the theoretical study of Bistritzer et al.42 who predicted that the equilibration process of the carrier temperature is faster for doped graphene. The calculated photoresponsivity, defined as Iph/Popt, for our functionalized GFETs varies from 0.3 to 100 mA/W and highly depends on power density and applied voltages VDS and VGS (see Figure 3c). These values are comparable or exceed the results from similar works for single-layer graphene photodetectors (see Supporting Information, Table S4 for a detailed overview).7,40,43 In most applications, the noise level is an important parameter, which can affect the detection limit of the device. We observed no significant increase in the noise level at higher source-drain voltages (Figure 3c). The measured lowfrequency noise (at 10 Hz) with and without CW laser irradiation is less than 4 × 10−10 and 1.3 × 10−9 A/√Hz for pristine and functionalized GFET, respectively. In general, the noise level linearly decreases with frequency increasing, which indicates the dominance of 1/f noise in our devices.44,45 The

structures we irradiated the samples with 532 nm CW laser. Photocurrent signals were measured in situ, while focused CW laser beam at 100 μW power scanned the channel. A photocurrent as large as 7 nA was measured directly at and close to the functionalized area, as it can be seen from Figure 3a. As expected, no significant photocurrent was observed for pristine graphene areas. Furthermore, the photocurrent generation area is slightly wider (∼1 μm2) than the spatial region of the junctions, which indicates that the excited carriers stayed hot across the GFET channel and contributed to the photocurrent.38,39 In order to further explore the phenomena, we varied the laser power in the range of 0.1 μW to 0.5 mW by neutral density filter and compared the laser power dependence of generated photocurrent in pristine and functionalized areas. Figure 3b shows the power dependence of the photocurrent (I ∼ Pα) for a pristine GFET and two GFETs functionalized with different photon fluxes. The photocurrent of pristine graphene channel is found to saturate at 2 × 106 W/cm2, while the saturation limit for functionalized graphene goes beyond 1 × 107 W/cm2, which is rather high.40 We suggest that the saturation starts when functionalized groups begin to detach from the graphene plane due to the temperature increasing above 150 °C upon laser irradiation. The nonunity exponent α may be a result of complex processes of electron−hole generation, trapping, and recombination,40,41 implying carrier transport and electron trapping in the processed graphene 3110

DOI: 10.1021/acsphotonics.8b00350 ACS Photonics 2018, 5, 3107−3115

ACS Photonics

Article

Figure 4. (a) R−VGS characteristics of the functionalized GFET-10 for the different power of 532 nm laser focused at the junction. (b) Photocurrent in GFET-10 obtained from the curve in (a) as a function of gate voltage. (c) Calculated thermoelectric power for pristine (black) and functionalized (red) GFET-11 as a function of VGS. (d) Temperature dependence of generated upon continuous wave laser irradiation photocurrent measured for GFET-8.

Photothermoelectric Effect in Laser Processed GFETs. Several physical mechanisms can give rise to a photocurrent, but only three can play a significant role in graphene-based photodetectors. These are the photovoltaic, the bolometric, and the photothermoelectric effects.43 In the case of the photovoltaic effect, incident photons generate a density of carriers that are separated and induce current at the electrodes when an electric field is applied. In that case, the photocurrent can be observed even in the absence of applied bias due to the presence of the built-in electric field, which separates electron− hole pairs.4 This field is small for p−p+ junctions and the direction of the generated current is opposite to the observed photocurrent,4,6,22 which indicates that the photovoltaic current is negligible in our devices. The photothermoelectric effect, on the other hand, originates from a difference in Seebeck coefficients from areas with a different density of states. The different local temperatures at the junction result in the diffusion of carriers after irradiation. As it was shown before for the case of supported graphene devices, hot-carrier dynamics generally dominate photocurrent generation because of inefficient cooling of electrons with the lattice.49 Another possible effect is bolometric, in which heating of graphene causes a resistance increase due to the enhancement of the electron-acoustic phonon scattering.50 However, the bolometric effect is also negligible here, since a very small source-drain voltage (usually below 0.5 mV) applied during the photocurrent measurements. The gate-modulated current gives clear evidence of a CNP shift upon CW laser irradiation. The shift can reach up to 30 V in back gate measurements (Figure 4a), converting the graphene channel from heavy p-doped to almost intrinsic.

noise equivalent power (NEP) can be estimated from Figure 3b as 2 kW/cm2 for pristine and 6 kW/cm2 for functionalized GFET. We calculated the linear dynamic range (LDR), as iP y LDR = 10 × log10jjj saturation zzz k NEP {

(1)

The resulted LDR for pristine graphene is 31 dB. Functionalized graphene shows larger LDR equal to 42 dB. These results are compared with literature data (see Supporting Information, Table S4). Additionally, to investigate relaxation dynamics in fabricated structures we irradiated the samples with 515 nm 280 fs-laser at photon fluxes lower than required for graphene photooxidation. Experiments with fs-laser reveal the photocurrent generation in both pristine and modified structures (Figure 3d). We observed the two-step mechanism of photocurrent dynamics after laser pump. The first step with time τ1 is related to intrinsic charge relaxation after hot carrier generation in graphene. The average time of hot carrier cooling is ∼50 ps,46 yet, in our case, we observed only the residual effect of this process since the minimal resolution of the multimeter is ∼1 ms, and the total irradiation time was ∼5 ms. The second part of the mechanism with time τ2 we associate with charge carrier trapping at the graphene−SiO2 interface.47,48 We observe that the amplitude of the photocurrent, generated in the channel during the laser treatment, is dependent on the graphene width: narrow graphene ribbons provide a higher response to laser irradiation (see Supporting Information, Figure S5). Additional data on photocurrent measurements with fs-laser can be found in Figure S6 in the Supporting Information. 3111

DOI: 10.1021/acsphotonics.8b00350 ACS Photonics 2018, 5, 3107−3115

ACS Photonics

Article

values at temperatures above 100 K.55 The difference between calculated and measured coefficient is related to enhanced inelastic scattering time of carriers in the presence of electron− electron interactions at temperatures above 100 K in nonideal graphene, which give strong deviations to the Mott relation.58 At room temperature, we expect the Seebeck coefficient to be reaching up to 100 μV/K.55 Finally, we measured the photocurrent dependency on temperature (Figure 4d). We varied the temperature in the range from 25 to 75 °C and observed a drastic increase of photocurrent. The photocurrent reaches values up to 150 nA at 30 mV when 75 °C is applied. This behavior only proves our theory, since the elevated temperatures intensify the release of charge carriers that are trapped at the graphene−SiO2 interface as well as from functionalized groups, consequently, intensifying the photocurrent. Dynamic measurements of photocurrent generation upon CW laser irradiation demonstrate fast response and rather long relaxation time about 100 s (Supporting Information, Figure S8).

Such a decrease of the hole concentration and an increase of the off-current can be related to the rise of electron concentration through reducing the energy barrier between grafted functional groups and graphene and via charge traps occupation at the graphene-SiO2 interface during the illumination.51 In pristine graphene, on the contrary, we did not observe any shift of CNP upon CW laser irradiation (Supporting Information, Figure S7). By tuning the gate voltage VGS, the majority of carriers in the processed graphene channel change from holes to electrons. The photocurrent is minimal near the charge neutrality point (Figure 4b). It rises with an increase of the |VGS−VCNP| difference for both types of carriers. When we consider the energy levels far from Fermi level, all carriers are involved in charge transport and the quantity of charges generated upon laser irradiation is minimal, hence the photocurrent tends toward zero. For these observations the photothermoelectric effect dominates as was shown previously by Xu et al.22 The physical picture of photocurrent generation due to the photothermoelectric effect can be described as follows: when the electrons are excited from the valence band to the conduction band, they relax down to the Fermi level on the time scale of femtoseconds by phonon emission and form a hot Fermion distribution.52,53 The hot free carriers tend to diffuse from the pristine part of the channel into the functionalized one due to the temperature gradient across the channel, which generates a positive current for hole doped graphene. In this case the photocurrent, I, can be formulated as I=

(S2 − S1)ΔT R



CONCLUSIONS In conclusion, we report a highly perspective maskless technique of p−p+ junctions fabrication in graphene FETs by local femtosecond laser treatment. This technique can be applied to process carbon nanomaterials at large scale with relatively high speed. In perspective, fs-laser treatment with submicrometer resolution can fully replace the lithography and plasma etching processes. The laser processed graphene photodetectors show a high photoresponsivity and photocurrent generation, which are more than 2 orders of magnitude higher than for pristine graphene. For 532 nm laser irradiation, the devices show the maximum photoresponsivity of 100 mA/ W at 9 mV source-drain bias voltage. The measured lowfrequency noise (at 10 Hz) is less than 1.3 × 10−9 A/√Hz and LDR as high as 42 dB for functionalized GFET with estimated NEP ∼ 6 kW/cm2. Such noise level is comparable with lowlevel-signal operations in integrated circuit devices. We suggest the thermoelectric effect is the main mechanism of photocurrent generation upon fs and CW laser irradiation in local p−p+ junctions in fabricated graphene FETs. Moreover, our findings propose that photoresponse of the graphene junction could be further enhanced by decreasing fs-laser processed area to submicrometer area to increase the effective charge separation. Additional photocurrent measurements at different wavelengths can give more information about detector working range. Thus, the study shows that the fully integrated photodetector with high responsibility, low noise, and high LDR is feasible by maskless patterning on graphene surfaces through ultrafast laser processing.

(2)

where S1 and S2 are the thermoelectric power for pristine and functionalized graphene devices, ΔT is the temperature difference, and R is the resistance. The thermoelectric power can be calculated by using the conductance of the sample with the Mott relation:54,55 S=−

π 2kB2 T 1 dG dVGS 3e G dVGS dE

E = EF

(3)

where kB is the Boltzmann constant, e is electron charge, T is temperature, G is conductance, and EF is the Fermi energy. When EF is away from CNP, mobility, μ is approximately a constant, and consequently, S is proportional to the density of states. Moreover, the term dVGS/dE can be rewritten as dVGS e 2 (VGS) = |ΔVGS| dE F CGSπ ℏυF

(4)

where dVGS = VGS − VCNP, ℏ is a Planck’s constant, υF = 106 m/s is the Fermi velocity, and CGS is the interface capacitance, which is assumed to be 2−3 μF/cm2 in the case of liquid gate experiments.56,57 The Seebeck coefficient dependency on VGS, calculated from the transfer I−V characteristics for pristine and processed under high photon flux GFET-11 (see Figures 2b and S4), is shown in Figure 4c. As it can be seen, the maximum value of the Seebeck coefficient for a channel with a functionalized area is higher than in the case of pristine graphene. The scattering length in functionalized graphene is drastically smaller due to the sp3 defects formation, which highly affect the heating of the graphene lattice. We would like to note here that the real Seebeck coefficient is likely much larger, since the Mott relation is known to give underestimated



METHODS Graphene FET Fabrication. Single-layer graphene was CVD-grown on 25 μm thick copper foil for 30 min at 1050− 1070 °C and methane gas was used as a carbon source. The process is performed in a quartz tube furnace in the presence of a Ar/H2/CH4 gas mixture (300 sccm, 15 sccm, and 0.5−1.0 sccm, respectively). Further, a thin layer of PMMA was spincoated on top of the graphene/copper stack that is used as a support layer during the transfer. In order to not waste a large fraction of the graphene, we used the modified, highthroughput transfer technique, described elsewhere.59 Prior to graphene transfer, the surface of Si/SiO2 wafer was treated 3112

DOI: 10.1021/acsphotonics.8b00350 ACS Photonics 2018, 5, 3107−3115

ACS Photonics

Article

device analyzer (MNIPI, Belarus) otherwise. Transfer I−V curves were measured using a back gate (p++ Si) electrode during photocurrent measurements or a liquid gate (using Ag/ AgCl pellet electrode) for accurate measurements of the CNP shift. The liquid gate measurements were carried out at 1× PBS buffer (∼150 mM, pH = 7.4). After every experiment, the chips were thoroughly washed in DI water in order to eliminate any possible salt based contamination. Continuous wave laser of Raman spectrometer with lower laser power was used to conduct the photocurrent experiments. The measurements of charge carrier relaxation dynamics were performed via the previously mentioned fs-laser at the energies below graphene photooxidation.

with oxygen plasma (0.8 mbar, 100 W, 5 min) in order to clean and increase the hydrophilicity of the surface, therefore, improving the graphene to SiO2 adhesion. The graphene was then transferred onto the wafer. When the PMMA/graphene stack was fished out by the target substrate, it was left for 24 h under ambient conditions to slowly dry out, then annealed at 150 °C for 10 min in order to reflow the PMMA and improve graphene to substrate adhesion.60 Afterward, the PMMA was dissolved in acetone (1 h in 50 °C acetone followed by 12 h in cold acetone). Finally, the structure was washed with isopropanol and DI water, dried under nitrogen flow, and annealed at 350 °C in N2 atmosphere.The graphene was then patterned via oxygen plasma etching (300 W, 200 sccm, 10 min). The 10 nm Ti and 100 nm Au metallization was deposited via e-beam assisted evaporation and further lift-off through the LOR-3B and AZ nLOF-2020 photoresists structure was performed. Photostructurable polyimide (HD8820, HD Microsystems) was used as the last step of passivation. Spin-coated at 5000 rpm, soft-baked, exposed (iline, 250 mJ/cm2), developed, and annealed with a slow ramp up to 350 °C and slow cooling down, the polyimide layer forms a perfect, pinhole-free, 3 μm thick passivation covering the metal feedlines. The passivation covers all metallic feedlines as well as partial area (