Performance enhancement of graphene/Ge near-infrared photodetector by modulating the doping level of graphene

The demand for high-performance, low-cost near-infrared (NIR) photodetectors is rapidly increasing due to the potential applications in distance sensor, medical sensor, optical communication, and night vision systems. However, most material systems used for NIR photodetectors, such as mercury cadmium telluride (HgCdTe),^1–3 indium gallium arsenide (InGaAs),⁴ germanium PIN on silicon (Ge PIN on Si),^5,6 and Ge/Si/Ge junction type devices,⁷ require complex and costly processes, such as molecular beam epitaxy (MBE) and ion implantation. On the other hand, graphene provides an opportunity for low cost, large area NIR photodetectors because of its high speed operation (∼1.5 THz) and high mobility (200 000 cm² V⁻¹ S⁻¹).^8–10 However, when a photodetector is fabricated using only graphene, the graphene monolayer absorbs only 2.3% of light, and the dark current is very high due to the zero bandgap characteristics, which is a serious limitation for practical applications.¹¹ 

To take full advantage of graphene, various hybrid type photodetectors utilizing the heterojunction of a semiconductor and graphene have been suggested.^12–15 At the graphene/semiconductor junction, the graphene behaves like a semi-metal and the current flow in this device is strongly affected by the Schottky barrier formed between graphene and the semiconductor.¹⁶ Especially for NIR applications, graphene/Ge heterojunction photodetectors have been reported because of the narrow bandgap (∼0.67 eV) of the substrate.^17–19 Zeng et al. reported that a monolayer graphene/Ge Schottky junction photodetector achieved 51.8 mA/W responsivity and 1.38 × 10¹⁰ Jones detectivity under zero bias condition (V_d = 0 V) at 1550 nm.¹⁷ Khurelbaatar and Choi also reported 0.45 A/W responsivity at the graphene/Ge PIN photodetector.¹⁸ Chang et al. reported 0.75 A/W responsivity and 2.53 × 10⁹ Jones detectivity at 1550 nm using a gate controlled graphene/Ge Schottky junction photodetector.¹⁹ Despite recent progress, the performances of these heterojunction based photodetectors are still lower than those of typical commercial photodetectors.

The most direct way to improve the performance of heterojunction devices is to find a way to increase the depletion region at the interface of two materials without applying external bias, to avoid the concurrent increase of the dark current. Thus, several ways to control the work function of graphene, such as inserting a metal,²⁰ adding a gate metal to control gate bias,^19,21 and graphene chemical doping,^22–24 have been investigated so far. Among these methods, the chemical doping process is relatively easy and advantageous in terms of cost and process temperature.

In this work, we applied chemical doping to the graphene/Ge heterojunction photodetector to control the Fermi level of graphene. With proper doping, the responsivity (R) of the graphene/Ge photodetector is improved from 0.87 to 1.27 A/W at 1550 nm. Specific detectivity (D*) is also increased to 9.6 × 10⁹ Jones, which is an ∼540% improvement compared to undoped graphene devices.

First, 90 nm SiO₂ was deposited on an n-type Ge substrate (resistivity = 2.5 Ω cm) using the plasma-enhanced chemical vapor deposition (PECVD) process. Then, an active Ge region (area = 14 × 16 µm²) was patterned by wet etching the oxide with a buffered oxide etchant (BOE). Next, the chemical vapor deposition (CVD) grown monolayer graphene was transferred to the Ge active region using the polymethyl methacrylate (PMMA) mediated transfer method²⁵ and Raman analysis was performed to verify that graphene was transferred successfully on the substrate.

Then, 30 nm of the Au hardmask layer was deposited using an e-beam evaporator and patterned with i-line lithography. The reactive ion etching (RIE) process with O₂ plasma was performed to form the graphene channel. Then, the drain contact region was patterned by etching 90 nm SiO₂, and 60 nm of silver antimony (AgSb, Ag with 1% Sb) was deposited by thermal evaporation before removing the photo-resist. After that, the lift-off process was performed to pattern the drain contact. The AgSb contact was chosen because the lowest contact resistance of 1.09 × 10⁻⁶ Ω cm² was obtained and this value is relatively low for moderately doped n-Ge. The rapid thermal annealing (RTA) process was performed to form an ohmic contact at 450 °C for 5 min in N₂ atmosphere between AgSb and the Ge substrate.²⁶ After forming the drain contact, 100 nm Au was blanket deposited on the whole substrate. Then, source and drain electrode regions were patterned by the lithography process and formed by the wet etch process. The schematic structure and the optical microscope image of the graphene/Ge photodetector is shown in Figs. 1(a) and 1(b), respectively. The drain has the Au/AgSb/Ge contact structure and the source has Au/graphene/SiO₂/Ge. Different metals for the drain and the source are used to optimize the contact resistances for Ge and graphene independently.

After the thermal annealing process, the chemical doping process was performed by dipping the graphene/Ge photodetector in a different concentration of the polyacrylic acid (PAA) dissolved ethanol solution for 3 h (0.1, 0.25, 1, and 2.5 wt. %). The thickness of the PAA layer that remained on the graphene is less than 1 nm and the transmittance of the PAA layer measured using a glass substrate is 98.5% or more in the NIR wavelength.^27,28 Then, 30 nm aluminum oxide (Al₂O₃) was deposited as a passivation layer using the atomic layer deposition (ALD) process at 130 °C. Finally, the fabricated devices were annealed at 300 °C in vacuum to eliminate residual water molecules and enhance the adhesion between the graphene and Ge. The entire fabrication process flow is shown in the supplementary material, Fig. S1.

When the carboxyl group (–COOH) in the molecule of the PAA polymer is dehydrated, it becomes –COO⁻, which induced holes in the graphene. Thus, when the graphene surface is covered with the PAA polymer, it can shift the Fermi level of graphene without generating any physical defects. The Raman analysis was performed to verify the graphene chemical doping process, as shown in Fig. 1(c). As the concentration of PAA increased, the G peak (∼1600 cm⁻¹) and 2D peaks (∼2700 cm⁻¹) shifted to the right, which confirms the p-type doping of graphene.^29–31 The D peak (∼1350 cm⁻¹) was not affected before and after doping, confirming that this process does not damage graphene.^32,33 The analysis of electrical characteristics was performed using a semiconductor parameter analyzer (Keithley 4200) at room temperature. A 1550 nm wavelength solid-state laser diode was used as a light source for photoelectric characterization. Details of the photocurrent measurement setup is shown in Fig. S2.

First, the I_d–V_d characteristics were measured under the dark condition to characterize the change in the Schottky barrier height. As shown in Fig. 2(a), the I_d–V_d curves show rectifying characteristics, i.e., the current flows well only in one direction, which confirms that the Schottky contact is formed at the interface of graphene and Ge. The reverse saturation current (V_d > 0) was decreased as the Schottky barrier height increased with increasing the concentration of the PAA solution. The Schottky barrier heights were mathematically extracted from the I–V characteristics shown in Fig. 2(b). The Schottky barrier height was extracted using the reverse saturation current of the ideal diode equation, which is expressed using the following equations:

where I_S is the reverse saturation current, A is the junction area of the graphene/Ge, A* is the effective Richardson constant for Ge, q is the electron charge, and Φ_b is the Schottky barrier height. The constants used in the calculation are listed in the supplementary material, Table S1. As the concentration of the doping solution increased, the Schottky barrier height between graphene and Ge increased from 0.39 eV at the pristine graphene device to 0.44 eV at the graphene device doped with the 2.5 wt. % PAA solution, resulting in a lower dark current under the reverse bias condition (V_d > 0 region). The band diagrams shown in Fig. 2(c) illustrate the change in the Schottky barrier height of graphene/Ge using the chemical doping. As the graphene is doped with a p-type doping agent (PAA), the Fermi level of graphene moves downward and the Schottky barrier height increases. As the Schottky barrier height increases, the carrier transport between graphene and Ge increases and the dark current decreases by more than one order of magnitude from 10⁻⁶ to 10⁻⁷ A.

Next, photoelectric characteristics were measured at 1550 nm with 52.6 µW illumination power. Figures 3(a) and 3(b) show the dark current and photocurrent of the pristine graphene device and the p-type doped graphene device (2.5 wt. %), respectively. The photocurrent (I_ph) was obtained by subtracting the dark current (I_dark) from the illuminated current (I_light). The dark current measured at V_d = 1 V decreased by 20 times after the doping and the I_ph/I_dark ratio increased up to 30 times, as shown Fig. 3(c). Responsivity (R) and detectivity (D*), which show the performance of the photodetector, were calculated using the following equations:

where I_ph is the photocurrent and P₀ is the incident light power, A is the active area (8 × 16 µm²), q is the electron charge, and I_d is the dark current. Here, the detectivity is a measure of minimum signal detection, assuming that thermal noise and shot noise are not dominant factors for the dark current.^34,35 The responsivity of the pristine graphene/Ge photodetector is 0.87 A/W. The responsivity is improved to 1.27 A/W after doping in the 2.5 wt. % PAA solution [Fig. 4(a)]. The specific detectivity of the graphene/Ge photodetector is gradually improved by 5.4 times as a function of PAA doping conditions, from 1.5 × 10⁹ Jones to 9.6 × 10⁹ Jones [Fig. 4(b)]. The responsivity shows much weaker doping dependence than the case of detectivity because the photocurrent shows weaker doping dependence as shown in Fig. 4(d). To check the stability of doping, these devices were repeatedly measured up to 29 days, and the results are shown in Figs. 4(c) and 4(d). There was no noticeable decrease in performance for 2 weeks. After 16 days, dark currents gradually increased for the heavily doped devices, but the photocurrent did not show a significant change.

Figure 4(e) shows the comparison for the responsivity and specific detectivity of the Ge photodetectors measured at 1550 nm.^13,34–36 After the graphene doping, the graphene/Ge photodetector showed about 1.5 times higher responsivity than the commercial photodetector.³⁶ The primary benefit of the graphene/Ge photodetector is that it is a simple process compared to other commercial IR detectors because it does not require semiconductor processes to form pn junctions in the Ge region. Even though the detectivity still needs further improvement, this device is quite a promising option for low cost IR detectors because further reduction of the dark current can be achieved by device optimization.

The performance of graphene/Ge heterojunction photodetectors has been improved by optimizing the Fermi level of graphene using a chemical doping process. By modulating the barrier height at the graphene/Ge junction, the dark current could be reduced by 1/20, and the responsivity and detectivity are improved by 150% and 540%, respectively. Since these improvements were obtained by the simple chemical dipping process, this process can be a promising way to achieve low cost, large area NIR sensors.

See the supplementary material for the complete device fabrication process, the setup for measuring the photocurrent characteristics, and device parameters used to extract the Schottky barrier height.

This research was supported by the Global Frontier Program through the Global Frontier Hybrid Interface Materials (GFHIM) (Grant No. 2013M3A6B1078873), the Creative Materials Discovery Program (Grant Nos. 2015M3D1A1068062 and 2017M3D1A1040828), and the Nano Materials Technology Development Program (Grant No. 2016M3A7B4909942) of the National Research Foundation of Korea (NRF), funded by the Ministry of Science, ICT and Future Planning (MOSIP), Korea.

The authors have no conflicts to disclose.

The data that support the findings of this study are available from the corresponding author upon reasonable request.