Visible and infrared dual-band imaging via Ge/MoS2 van der Waals heterostructure.

Multispectral photodetectors are emerging devices capable of detecting photons in multiple wavelength ranges, such as visible (VIS), near infrared (NIR), etc. Image data acquired with these photodetectors can be used for effective object identification and navigations owing to additional information beyond human vision, including thermal image and night vision. However, these capabilities are hindered by the structural complexity arising from the integration of multiple heterojunctions and selective absorbers. In this paper, we demonstrate a Ge/MoS2 van der Waals heterojunction photodetector for VIS- and IR-selective detection capability under near-photovoltaic and photoconductive modes. The simplified single-polarity bias operation using single pixel could considerably reduce structural complexity and minimize peripheral circuitry for multispectral selective detection. The proposed multispectral photodetector provides a potential pathway for the integration of VIS/NIR vision for application in self-driving, surveillance, computer vision, and biomedical imaging.

INTRODUCTION

The combination of visible (VIS) and infrared (IR) photodetectors is a promising technique for use in various applications such as computer vision (, ), biomedical engineering (), and tactical vision (). This multispectral image fusion can provide a more accurate understanding of the surrounding information compared to data acquisition by a single spectral band (). In particular, along with the VIS spectrum detection, the imaging of IR spectrum near 1550 nm is advantageous because of the high atmospheric transmission () and eye safety concerns. Furthermore, IR spectrum imaging attracts considerable interest for night vision (), three-dimensional (3D) ranging (), etc. However, the integration of individual VIS and IR photodetectors for multispectral detection is limited by structural complexity. The conventional side-by-side configuration of independent VIS and IR photodetectors leads to large pixel size and optical cross-talk between photodetectors. Furthermore, related external circuits require large footprint and cost for switching capabilities.

To address these challenges, vertical-stacked and multispectral photodetectors have been proposed (–). The multispectral functionality has been realized via broadband and selective photodetectors. Broadband photodetectors exhibit photonic responsivity to both VIS and NIR spectra simultaneously and can hardly resolve the VIS and NIR information (–). Photodetectors based on organic heterojunctions (), quantum dot nanomaterials (), and semiconductor multiple junctions () obtain selective band information from broadband signal. However, the structures are based on a multijunction structure, including back-to-back–connected diodes with three terminals that require dual bias polarity operations for selective detection capabilities, resulting in complex fabrication processes and nonconventional back plane and readout circuit designs (table. S1).

In this paper, we present a two-terminal spectrally selective photodetector to achieve both VIS and IR spectral perceptions. We designed a heterostructure of a small-bandgap p-type Ge and large-bandgap n-type MoS2 to achieve bias-dependent photocurrent generation (, ). Fabricated van der Waals p-Ge/n-MoS2 heterojunction photodetector responds to multiple spectrum via modulating the operation mode of near-photovoltaic (−0.5 V) and photoconductive (−3.5 V) modes. Unlike the conventional dual-band photodetectors, our multispectral photodetector is enabled by a single heterojunction and allows unprecedented single-bias polarity (reverse bias) operation for multispectral detection, which ensures miniaturized footprint and eliminates the need for additional polarity switching schemes. We believe that the simplified multispectral detection using a single junction photodetector provides a great potential for its widespread adaption in various imaging fields such as machine vision, biomedical engineering, and object recognition.

RESULTS

Device operation principle and analysis

We fabricated a p-Ge/n-MoS2 photodetector with bias-dependent wavelength selection capability. The selective detection capability can be applied to vision under adverse weather conditions such as night, fog, and dust (Fig. 1A). The vision information obtained by the different wavelength spectra achieves advanced ambient information systems. A schematic of the device is shown in Fig. 1B. The low-dimensional MoS2 was mechanically exfoliated and transferred to the device substrate (see Materials and Methods). Figure 1C shows the Raman spectra of the p-Ge/n-MoS2 photodetector obtained using a 532-nm excitation laser. The results include the bulk Ge peaks and multiple MoS2 peaks, including E12g, A1g, and 2LA (longitudinal acoustic) phonon modes (, ). The area of the p-Ge/n-MoS2 heterojunction is 28.3 μm2 (exfoliated 66-nm-thick MoS2 flake), as verified by atomic force microscopy mapping (see fig. S1). The exposed Ge layer was fully covered by the MoS2 flake. The heterojunction was also verified by cross-sectional transmission electron microscopy (TEM) images, as shown in Fig. 1D. The heterojunction is shown as a white region between the bulk Ge and MoS2 regions.

Fig. 1.

Design and operation principle of the Ge/MoS2 multispectral photodetector.

(A) Schematic of the device application. The selective dual-band detection enables vision for a harsh environment (e.g., fog, top) using VIS and IR visions simultaneously and separately (bottom). (B) Schematic of the device structure. (C) Raman spectroscopy of the Ge/MoS2 heterojunction structure. The peaks corresponding to Ge and MoS2 (E12g, A1g, and 2LA) are measured. (D) Cross-sectional TEM image. (E) XPS measurement of the Ge/MoS2 heterojunction structure. (F) Energy band diagram in equilibrium calculated by the results in (E), UPS, and EELS. a.u., arbitrary units.

Figure 1E shows x-ray photoelectron spectroscopy (XPS) results of the heterostructure device. The high-resolution spectra of the C 1s, Mo 3d3/2, Mo 3d5/2, and Ge 3d peaks were collected in steps of 0.1 eV. The measurement of the band levels before and after the heterojunction reveals the band shift indicating the energy barriers of conduction (ΔEGe) and valence (ΔEMoS2) bands of approximately 0.03 and 0.34 eV, respectively (). The full XPS survey spectra are shown in fig. S2. The apparent valance band maximum (VBM) locations are 0.14 eV for Ge and 0.89 eV for MoS2 via ultraviolet photoelectron spectroscopy (UPS) (see fig. S3), and additional electron energy loss spectroscopy (EELS) results also verify the bandgaps (fig. S4). On the basis of the apparent VBM and energy shift, the equilibrium band diagram of the p-Ge/n-MoS2 heterostructure device was calculated (Fig. 1F).

Photoresponse characterizations

We use a band heterostructure to enable multispectral detection capability. Figure 2 (A and B) shows energy band diagrams of the p-Ge/n-MoS2 heterostructure under different biases. A heterojunction with asymmetric transport is designed to enable selective spectrum detection. As shown in Fig. 2A, the transport of the generated holes in the n-MoS2 region because of the VIS absorption is blocked by the downward band bending, which are eventually recombined with the electrons. Thus, the overall transport of photogenerated carriers under VIS illumination is limited. By contrast, both generated electrons and holes under IR illumination in the p-Ge region easily pass through the heterojunction. Furthermore, the band configuration under a strong bias is harnessed to achieve a completely selective detection functionality. The mechanism of the VIS spectrum detection is illustrated in Fig. 2B. Under IR illumination, the transport from the generated electrons and holes in the p-Ge region is weak and diluted by the increased dark current. However, under VIS illumination, the strong bias alters the band structure bending downward toward the interface of n-MoS2 and Ti/Au, which elicits a higher photocurrent compared with the IR illumination. Moreover, the photogenerated holes in the n-MoS2 region were trapped in the heterojunction interface, resulting in the downward band bending of the n-MoS2 region (). As a result, these trapped holes increase the tunneling current included in the total output current.

Fig. 2.

Photoresponse characteristics.

Energy band diagrams under (A) near equilibrium (near-photovoltaic mode) and (B) strong reverse bias (photoconductive mode). (C) I-V characteristics at various wavelengths. I-V characteristics of the device at (D) 406 nm and (E) 1550 nm with various incident power densities. (F) PDRs at multiple wavelengths under the near-photovoltaic and photoconductive modes. Photocurrent (left) and responsivity (right) under (G) 1550-nm and (H) 406-nm illuminations as a function of the incident power density under biases of −0.12 and −3.5 V, respectively. (I) PDR as a function of the bias and incident power density under the 406-nm and 1550-nm illuminations.

Here, we verify the trap-assisted tunneling mechanism at the photoconductive mode. The tunneling mechanism is verified by the Fowler-Nordheim (FN) plots from the I-V characteristics (Supplementary Text and fig. S5) and by the trap-assisted photocurrent of the heterojunctions with either n-doped (unintentionally doped) or additional n+-doped MoS2 layer (fig. S6). The doping concentrations are 7.62 × 1014 cm−3 and 5.53 × 1017 cm−3, respectively, measured by the Hall effect measurement (fig. S7). The heterostructure enables combined selectivity, and the individual responsivities of each Ge and MoS2 layer are also verified by fabricating a pn photodiode and metal semiconductor Schottky photodiode, respectively (figs. S8 and S9).

The band alignment between Ge and MoS2 is theoretically investigated through the local density of states (LDOS) calculation using density functional theory (Supplementary Text). For the LDOS calculation, the Ge/MoS2 heterostructure is constructed by combining Ge and MoS2, as shown in fig. S10. Ge is 1 × 1017/cm3 p-type doped, while MoS2 is n-doped or 1 × 1017/cm3 n+-type doped. LDOS is calculated for the Ge/MoS2 heterostructures with n-doped and 1 × 1017/cm3 n+-type–doped MoS2, respectively, at equilibrium. From LDOS plots in fig. S11, the conduction band and valence band edges of Ge and MoS2 at the interface confirmed that the heterostructure exhibits (type II) staggered gap consistent with the experimental measurements in Fig. 1F.

Figure 2C shows the I-V characteristics of the p-Ge/n-MoS2 photodetectors with different wavelengths of illumination. The wavelength is in the range of 406 to 1550 nm. The power density of the incident light is fixed. The longer wavelength exhibits higher responses at −0.5 V (near-photovoltaic mode), whereas the shorter wavelength exhibits higher responses at −3.5 V (photoconductive mode). We calculated the ideality factor (n = 1.31) on the basis of the experimental dark I-V characteristics (fig. S12). Under a small positive bias (less than 0.5 V), Fig. 2C shows the exponential relationship between the current and the voltage. Under a high positive bias (more than 0.5 V), however, the relationship is no longer exponential, and further analysis has been carried out according to the additional FN plot (fig. S12). The barrier height in thermionic emission (0.31 eV) is also calculated by the temperature-dependent I-V characteristics (fig. S13).

The selective multispectral detection performance of the p-Ge/n-MoS2 photodetectors is shown in Fig. 2 (D and E). The I-V characteristics of the p-Ge/n-MoS2 photodetectors exhibit selective detection at wavelengths of 406 and 1550 nm under various incident power densities. The basic rectifying behavior of the p-Ge/n-MoS2 heterojunction is also observed in the dark I-V characteristics. Under the 406-nm illumination, a large photocurrent is generated in the photoconductive mode (VIS vision). By contrast, under the 1550-nm illumination, most of the photocurrent is generated at the near-photovoltaic mode (IR vision). The bias-dependent selective phenomena are attributed to the interplay of band energy configurations, such as band offset, trapping, and tunneling electron, and hole currents at different biases. Although the illumination spot is larger than the Ge/MoS2 junction region (fig. S14), most regions out of the junction are blocked by the metal. Therefore, negligible amount of light generates an undesirable Schottky junction effect by reaching the MoS2 out of the junction (fig. S9).

The photoresponses of the different spectra are distinguishable under the near-photovoltaic and photoconductive modes, as shown in Fig. 2F. The responses of the incident light at a bias of −0.5 V (−3.5 V) show a high (low) photocurrent-to-dark current ratio (PDR) at 1550 nm and relatively low (high) ratio at 406 nm. As shown in Fig. 2G, the slope of the incident power to the photocurrent under the 1550-nm light is approximately unity, which implies that most of the photon energy is linearly converted to electrical power. By contrast, Fig. 2H shows the photocurrent transition under the 406-nm illumination. The slope is 0.24 (0.04) at a power density lower (higher) than 0.5 mW/cm2. The nonlinear transition is attributed to the saturation of the trapped photogenerated carriers. In general, trap saturation leads to a high (low) responsivity under a low (high) incident optical power, resulting in an overall nonlinear photocurrent response to the incident power density (). The wavelength- and voltage-dependent PDRs are shown in Fig. 2I. The IR and VIS spectra responded strongly under the near-photovoltaic and photoconductive modes, respectively. A high PDR is achieved in both modes, which provides selective VIS/IR vision capability of the device.

Frequency and temporal responses

The results of the measured noise power density are shown in Fig. 3A. The noise power density of the dark current shows a dominant 1/f noise in the photoconductive mode, while the flat noise power density in the near-photovoltaic mode might be attributed to thermal and shot noises. The results include two voltage-dependent cases that provided selective wavelength detection in previous studies. The bandwidth is set to 10 kHz based on this cutoff frequency. Both wavelength and frequency dependencies of the noise equivalent power (NEP) are calculated and shown in Fig. 3 (B and C), respectively. At this defined bandwidth, the NEP in the photoconductive mode is three orders of magnitude larger than that in the near-photovoltaic mode. However, the larger responsivity under the photoconductive mode overcomes such a high noise current, as shown in the previous section. Figure 3 (D and E) shows the transient characteristics of the p-Ge/n-MoS2 photodetectors. The power density is fixed to 30 mW/cm2 with bias voltages of −0.5 and − 3.5 V. The photoresponses at the near-photovoltaic mode show a higher photocurrent when the wavelength becomes longer. By contrast, the responses in the photoconductive mode show an inverse trend compared to that under the near-photovoltaic mode, which corresponds to the previous PDR measurements.

Fig. 3.

Frequency and temporal responses.

(A) Noise power density with respect to the frequency. Dominant 1/f noise is observed in the photoconductive mode. The cutoff frequency is set to 10 kHz. Calculated NEP at various (B) wavelengths and (C) frequencies. The results are shown with respect to the near-photovoltaic and photoconductive modes. Temporal response of the device at various illumination wavelengths in the (D) near-photovoltaic and (E) photoconductive modes. The temporal responses for different wavelengths are offset for clarity. (F) Calculated rising and falling times under the near-photovoltaic and photoconductive modes at various wavelengths.

Figure 3F shows the wavelength-dependent rise and fall times of the device. The two metrics are similar, 16 ms, in the near-photovoltaic mode. In contrast, the fall time is approximately five times larger than the rise time in the photoconductive mode. These bias-dependent variations are attributed to the different absorption aspects of Ge (~0.67 eV) and MoS2 (~1.23 eV). The bulk Ge layer is covered by the exfoliated 66-nm MoS2, and the corresponding penetration depths for various VIS wavelengths are comparatively lower than the MoS2 thickness (66 nm), implying that most of the illuminated VIS are absorbed in the MoS2 region before reaching the Ge region (fig. S15). Notably, the higher bandgap of MoS2 with the heterojunction allows VIS absorption and transport of the generated carriers at a higher bias (−3.5 V). Under weak reverse bias and illumination, the electrons in Ge and holes in the MoS2 region move toward the barrier. Owing to the modulated heterojunction barrier, the electrons easily move over the barrier, while the transport of the holes is hindered by the downward bending of the MoS2 valence band. By contrast, under a strong bias, holes in the n-MoS2 region pass through the heterojunction interface via trap-assisted tunneling. The trapped hole carriers require an additional time to be fully released from the trap states when the light is off. This trap-assisted tunneling mechanism increases the photocurrent decay time, resulting in the slow switching time. Note that the built-in potential of the heterojunction is smaller than the reverse bias of the photoconductive mode and that the response time is dominated by the carrier diffusion in the bulk region at the near photovoltaic mode. Therefore, the temporal effect of the built-in potential is negligible.

VIS/IR detection under simultaneous detection and selective VIS-NIR dual imaging

The multispectral detection capability of the heterojunction photodetector is retained under simultaneous VIS and NIR illuminations (Fig. 4, A and B). A small cross-talk occurs near 0.6 and 1.6 s. However, the responsivity of the VIS spectrum is small compared to that of the NIR spectrum. We also demonstrate the dual-spectral imaging capability of the multispectral photodetector (Fig. 4C). Along with the visual information under VIS spectra (object 2), an additional visual information (object 1) is obtained via this multispectral photodetector. Figure 4D shows a schematic of the experimental setup of the dual imaging. We prepared a double-sided target object that can transmit and reflect NIR and VIS sources, respectively (see Materials and Methods). The VIS and NIR spectra were sequentially illuminated on the sample. The NIR light penetrates the backside of the sample, except in the region where the metal is not present (transmission imaging). Similarly, the VIS light toward the front side of the sample is reflected, where the metal is present (reflection imaging). As a result, the transmission imaging exhibits an angry face in the near-photovoltaic mode (Fig. 4E), while no visual information is achieved in the photoconductive mode (Fig. 4F). However, the reflection mode offers hidden visual information (smile) via the photoconductive mode (Fig. 4G) without achieving an angry image in the near-photovoltaic mode (Fig. 4H). This selective imaging capability is attributed to the specific responsivities of the different spectra via switching between the near-photovoltaic and photoconductive modes. Operating in the NIR vision, an additional NIR imaging is performed for further night vision capabilities (fig. S16).

Fig. 4.

VIS/IR detection under simultaneous detection and dual-band selective imaging.

(A) Schematic and (B) results of the dual-band detection under simultaneous illumination of VIS and IR spectra. Owing to the cross-talk between the VIS and IR spectra, a small VIS response also occurs at the near-photovoltaic mode. (C) Schematic of the dual-band imaging. By switching between the photoconductive and near-photovoltaic modes, the p-Ge/MoS2 heterostructure enables dual-band imaging capability. (D) Schematic of the dual-band selective imaging experiment. The double-side–patterned silicon target object exhibits angry and smile face images (see Materials and Methods). Results of the dual-band imaging for transmission imaging (IR illumination) operating at the (E) near-photovoltaic and (F) photoconductive modes. Results of the dual-band imaging for reflection imaging (VIS illumination) operating at the (G) photoconductive and (H) near-photovoltaic modes.

DISCUSSION

Conventional broadband photodetectors provide inseparable VIS and NIR detection capabilities, whereas the van der Waals heterojunction photodetector offers selective VIS and NIR vision. The separation of each spectrum is important to extract more synergistic information. For instance, VIS and NIR lights generally propagate simultaneously because of multiple driving forces, such as black body radiations (temperature distribution), spectral reflections, and direct illuminations from light sources. Each spectrum is thus originated from the different mechanisms, and by applying a temporal voltage pulse train to our device, separable visual information can be achieved. The separated VIS and NIR vision could thus be harnessed in various applications such as image fusion (), red-green-blue depth imaging (), and classification via image segmentation (). The heterojunction photodetector exhibits the same-polarity operation for both near photovoltaic and photoconductive mode, which would alleviate the circuitry complexity compared to the recent multispectral photodetectors (table S1). As a future direction, the use of 2D materials would provide strong light matter interactions and ease of integration with conventional silicon readout electronics as well as array-scale implementation (–). Furthermore, various heterogeneous integration methods would allow the device to be formed in arbitrary 3D geometries, which offers a wide field-of-view vision along with compact bioinspired functionalities.

We demonstrated a spectrally selective photodetector based on the heterojunction of the fabricated p-Ge/n-MoS2 photodetectors. The two-terminal heterojunction device exhibited both VIS and IR responsivities depending on the reverse bias level. The selective detection capability is attributed to the heterojunction band modulation and reverse bias level, which promotes or hinders the carrier transport. The vertical heterostructure and single polarity of the bias voltage offer miniaturization of the system and dual-vision imaging capability. We believe that the multispectral devices will be useful in the realization of neuromorphic vision systems for various advanced applications, including light detection and ranging, health care, computer vision, and in vivo biomedical imaging.

MATERIALS AND METHODS

Device fabrication

A schematic of the fabrication of the photodetector is shown in fig. S17. A 20-nm oxide (Al2O3) layer was deposited on a p-Ge substrate (5 × 1017 cm−3) at 200°C by atomic layer deposition to isolate the metal contact of MoS2. To control the junction area, a circular shape with a diameter of 6 μm was etched in 1% diluted hydrofluoric acid in the Al2O3 layer until the Ge surface was completely exposed, resulting in a sloped sidewall at the Ge/Al2O3 boundary (). To ensure air stability, the Ge substrate was passivated by a solution of HCl:H2O in a ratio of 1:1 for 1 min. Multilayer MoS2 flakes were mechanically exfoliated from bulk MoS2 crystals (Graphene Supermarket, USA) and then transferred onto the exposed Ge surface. Ti/Au (5/60 nm) was deposited as an Ohmic metal of MoS2 by electron beam evaporation using a standard liftoff technique. A 5/60-nm Ni/Pt p-contact was deposited on the back side of the p-Ge substrate by electron beam evaporation. The device was annealed for 1 hour at 150°C in vacuum using a rapid thermal processor to form a reliable junction at the Ge-MoS2 interface.

The smile and angry images were fabricated on an Si substrate with a 230-nm SiO2 antireflective coating. At the backside, a 100/300-nm-thick Ti/Au array image was deposited and patterned. On the front side, a 300-nm Al layer was deposited. Both metal layers were deposited and patterned by electron beam evaporation and a standard liftoff technique, respectively.

Device analysis

Micro-Raman spectroscopy of the Ge-MoS2 heterojunction was performed using a LabRAM HR Raman system with a spatial resolution of 1 μm and excitation wavelength of 532 nm. The sample for the lateral TEM analysis was prepared using a focused ion beam system (Quanta 3D FEG, FEI). TEM was performed using a high-resolution TEM (JEM-2100F, JEOL) with an operating voltage of 200 kV. XPS was performed using a Nexsa XPS system (Thermo Fisher Scientific) with an Al Kα source (1486.6 eV) operating at 300 W. The base pressure of the XPS system was 5.0 × 10−9 mbar. The spot size of the x-ray was 100 μm. Every scan of the data was repeated 10 times to obtain a precise peak position. The high-resolution spectra of the C 1s, Mo 3d3/2, Mo 3d5/2, and Ge 3d peaks were collected in steps of 0.1 eV with a pass energy of 20 eV.

Electrical and optoelectrical measurements

The electrical and optoelectrical measurements were carried out using a semiconductor parameter analyzer (4200A-SCS, Keithley Instruments) with various light sources of a 1550-nm laser (LSR1550NL-1W-FC, Civil Laser), four-channel fiber-coupled laser (406, 520, 638, and 850 nm, MCLS1, Thorlabs), and 940-nm IR light–emitting diode (HBL0307) under various optical power densities of 0.1, 0.3, 3, 10, and 30 mW/cm2. To determine the transient photoresponses, the pulsed optical signal was generated by the same lasers at 30 mW/cm2 with a chopper, and the resulting photocurrents were monitored using a source meter (Keithley 2614B).

Responsivity and NEP extraction

The responsivity (R) was calculated as R = Iph/Pin, where Iph is the photocurrent, and Pin is the incident optical power. The photocurrent was derived from the current-voltage characteristics, expressed by Iph = Ilight − Idark, where Ilight is the measured current under various intensities of multiple wavelengths, and Idark is the dark current. The NEP of the Ge-MoS2 heterojunction was measured in a dark environment using a semiconductor parameter analyzer (B1500A, Keysight) and calculated as NEP = , where Pin,noise is the integrated input noise power, BW is the specified bandwidth, Iin,noise is the input noise current, and Rmax is the maximum responsivity ().

The data sampled at each frequency (1 kHz, 10 kHz, 20 kHz, 100 kHz, 200 kHz, 1 MHz, 2 MHz, and 10 MHz) at biases of −0.5 and − 3.5 V were transformed to the frequency domain by the fast Fourier transform method, square rooted, and integrated by BW to convert to Iin,noise. Pin,noise was then calculated by dividing Iin,noise by Rmax at each bias. Rmax was calculated by the IR (1550 nm, 30 mW/cm2) and VIS (466 nm, 30 mW/cm2) spectra, where a high responsivity occurs at −0.5 V. On the basis of this computation, the NEP was calculated by Pin,noise divided by .