Visible to Mid-Infrared Photodetector Based on Black Phosphorous-MoS2 Van Der Waals Heterojunction

Van der Waals heterostructures of black phosphorous(BP) and Molybdenum Disulfide(MoS<sub>2</sub>) is designed and evaluated for photodetection application from visible to mid-infrared wavelength up to 4.5 μm. The device possesses a low dark current less than 0.1 μA for short wavelength light. For visible and near infrared radiation, the contribution of heat and photo energy to current enhancement is analyzed. Dark current drift by joule heat and laser heat absorbed is observed in short waveband. The difference in response speed is believed to be a result of the competition of BP and MoS<sub>2</sub> photocurrent. Gate tunable response time and on-off ratio makes the device versatile in potential application. For mid infrared radiation, BP-MoS<sub>2</sub> heterojunction with a dedicated thickness tailoring of MoS<sub>2</sub> exhibits similar negative response as an intrinsic BP. The normalized detectivity is <inline-formula><tex-math notation="LaTeX">${D}^*$</tex-math></inline-formula> of <inline-formula><tex-math notation="LaTeX">$2.02 \times {10}^8\ Jones$</tex-math></inline-formula> for the 4.5 μm mid-infrared radiation. Elaborate reduction of the noise is conducted to get the time resolvable photo-response. A low bias is necessary for a time resolved photocurrent, reducing the flicker noise to an acceptable level.


I. INTRODUCTION
B LACK phosphorous [1] is a two dimensional material with layer dependent band gap, ranging from 0.3 eV to 1.8 eV for bulk form BP to monolayer BP [2]. The narrow direct band gap of bulk BP makes it suitable for infrared detection [3]. While in previous research, lock-in method used in mid infrared measurement enhances detection performance artificially and makes it difficult to find the intrinsic physical mechanism [4], [5]. Besides, intrinsic uniform BP lacks visible and near infrared sensitivity due to low light harvesting and instant local carrier recombination. Even if asymmetric structure is employed, the photodetection suffers from a large dark current [6]. Combining BP with another two-dimensional material is a solution for Manuscript  Van der Waals heterojunction confirms its feasibility [7], [8], [9]. Geonyeop Lee fabricated BP/MoS 2 /BP heterojunction and got a gate tunable rectifying ratio [10]. Lei Ye extended the detection wavelength of BP/MoS 2 heterojunction to 1550 nm [11]. Thayer S. Walmsley uncovered that response time can be tuned through changing the gate voltage [12]. In this work, time resolvable photo-response of BP-MoS 2 heterojunction in visible, near infrared and mid infrared waveband is measured and compared with that of intrinsic BP.

II. MATERIALS AND METHODS
Au electrodes are deposited on 300 nm SiO 2 /Si substrate with a 10 nm titanium moist layer. followed by black phosphorous transferring by use of universal two-dimensional material dry transfer method [13], [14]. Multilayers of BP, is first transferred above on electrode pad, leaving the other pad open. Then MoS 2 is transferred, connecting the BP to the other pad. Necessary measurement such IV curve is carried on single materials and their heterojunction. Multilayers of MoS 2 are purchased from Six-carbon technology, Shenzhen, China. The black phosphorous is purchased from Nanjing MuKe Nanotechnology Co., Ltd.
The photocurrent is collected by a PDA source meter. For visible and near infrared light, a laser with specified wavelength is used. The light spot differs one from another. For mid infrared radiation, a narrow band filter centered at 4.5 μm is placed before SiC black body for certain mid-infrared wavelength. The FWHM of the filter is around 500 nm. The spot size for these mid-infrared filtered light is estimated as 2 cm 2 . And the time resolved response is recorded with a mechanical chopper.

III. RESULTS AND DISCUSSION
Basic measurement setup is illustrated in Fig. 1(a) where the light source resembles lasers in visible to near infrared wavelength or a black body in with a filter in mid infrared wavelength. A metallographic microscope photograph of BP-MoS 2 heterojunction is in the inset of Fig. 1(a). Raman spectrum is measured to verify the overlying of BP [15] and MoS 2, also shown in the inset of Fig. 1(a).To fabricate this device, gold electrodes are patterned on SiO 2 /Si substrate by ordinary ultraviolet lithography and thermal evaporation. BP exfoliated by scotch tape is transferred on the edge of the electrode, followed by the MoS 2 transfer connecting BP sheet and the other electrode. So the BP is covered by MoS 2 sheet. The thickness of MoS 2 is selected thin to decrease its radiation blocking influence to BP. The thickness of gold is 100 nm and the thickness of SiO 2 is 300 nm. Detailed thickness information of the heterojunction is measured using a profilometer. The thickness of gold electrode is known and subtracted from the total height data. As a result, the thickness of BP sheet is evaluated as 80 nm and the thickness of MoS 2 is around 6 nm.
IV curve of the Van Der Waals heterojunction is extracted for dark and bright condition, as shown in Fig. 1(b). Under dark condition, an apparent heterojunction behavior is observed. The current is −6.03 nA with a bias of −1 V and 41.4 nA with a positive bias of 1 V. This feature is similar to an ordinary p-n junction. While for a Van Der Waals junction, the gap at the interface needs modelling in a more subtle manner. Lee et al. modeled similar WSe 2 /MoS 2 Van Der Waals heterojunction using traditional drift-diffusion equation with a phenomenological interlayer gap [16] and got a good accordance, which is enough for the analysis of BP/MoS 2 device. Comparing to single BP or MoS 2 , the current is lower. For a BP sheet with a bias of 1 V, the resistance is several kilo-olms and for a MoS 2 sheet, the resistance is around 1 MΩ. While for our BP-MoS 2 device, the resistance is 24 MΩ in the same measurement environment. A steeper slope under small reverse bias indicates a part of charge stored in the junction. For a series circuit, voltage drops according to the ratio of resistance of each part. The voltage drop on pure BP or MoS 2 sheet is small, while most voltage is applied on the gap in the heterojunction.
There is a large difference in total current under laser incident for various wavelength as illustrated in Fig. 1(b). There is a changing of slope of the IV curve especially apparent for a 808 nm laser. Absolute value of total current rises rapidly after the bias cross 0.3 V to positive and −0.5 V to negative direction. leaving a flat zone is around zero bias. In this flat zone, even excess carriers are generated in a fast speed, the external electric field for carrier separation is not strong enough to overcome the barrier between BP and MoS 2 . As a result, a relatively low external quantum efficiency (EQE) is recorded. By applying a high bias, the barrier is overcome. Then a combination of photoconductor and photo voltage effect appears. The thorough mechanism is hard to get. But when positive biased, the device behaves more like a photoconductor. The feature can be more apparent with a close look on the photocurrent for 405 nm and 520 nm lasers. When the Van Der Waals heterojunction is positive biased, the majority carriers in the BP or MoS 2 are accumulated near the interface of the junction. The excess carriers generated by incident radiation are swept by the electric field. They must cross the space between the BP and MoS 2 to make contribution to total current. Drawing from the experiment, the probability for the carriers leaping turns high when the bias is larger than 0.3 V.
To dismantle the photocurrent, the case of 520 nm radiation is taken as an example, illustrated in Fig. 1(c). A series of power is adopted for incident light. When positively biased, the photocurrent rises smoothly and responsivity decreases accordingly as the power increases. The photocurrent for a 19.2 nW is 13.2 nA under a bias of 1 V, yielding a responsivity of 0.69 A/W. And we get a photocurrent of 31 nA for an incident light power of 540 nW and a responsivity of 57 mA/W.
The photo-response is in expectation for visible wavelength. While if the laser wavelength is turned to near infrared, the responsivity changes a lot, as plotted in Fig. 1(d) in logarithmic coordinates. The bandgap of bulk MoS 2 is reported to be 1.2 eV [17]. That is, the absorption edge is 1033 nm. For the incident light with longer wavelength, intraband absorption of MoS 2 and the influence of BP would be complex. Experimentally, we get a responsivity of 0.9 A/W for a 808 nm and 0.01 mA/W for a 980 nm laser with a bias of 2 V. However, the responsivity directly evaluated from IV curve may be overestimated because heat absorbed by the device and joule heat cannot be released enough and they can make contributions to the total current. To discuss the photo-response mechanism in depth, time dependent current is measured.
Time resolvable photocurrent is shown in Fig. 2. More information of the BP-MoS 2 heterojunction can be clarified. The photosensitive mechanism of photodetectors are usually classified into two types, i.e., photo-electronic and photo-thermal effect. These two mechanism coexist in our BP-MoS 2 photodetector. The thermal effect is contributed both by the joule heat and the incident laser absorption. In the Fig. 2(a), the drift of dark current is due to the joule heat plus 405 nm laser energy absorbed. The baseline of photocurrent keeps steady if the bias is lower than 0.5 V, as shown in Fig. 2(a) for the 405 nm case. With a bias of 1 V, the dark current starts to grow as time passes. The dark current is 40 nA when the measurement is initially conducted. While the time is larger than 15 s, the dark current is 65 nA. After the light is turned off, the dark current then starts falling.
We can clarify the joule heat from laser power induced heat through varying the laser attributes. In Fig. 2(c), the total current of the device is measured under 808 nm for different incident power with a bias of 1 V. The power density of the laser for the case of 7430 nW is the same as that of 405 nm case in Fig. 2(a). The baseline in Fig. 2(c) keeps constant, meaning that joule heat plus the 808 nm laser energy absorbed by the device cannot induce the total current increasing with time. As a conclusion, the drift of dark current is mainly due to the 405 nm laser energy absorbed in the Fig. 2(a). This can be verified through absorption coefficient modelling of the device. The absorption spectrum modeled by first-principle is shown in Fig. 2(d). In the short waveband, the heterojunction absorption coefficient is much higher than other cases.
An illustration of photocurrent under 650 nm laser with a bias ranging from −0.3 V to 0.3 V is shown in Fig. 2(b). The dark current keeps constant thorough the measurement process. At a zero bias, holes in black phosphorus electrons in MoS 2 near the interface tends to diffuse to each other, leaving a built-in electric field in individual material. This process is like the traditional p-n junction except that a phenomenological interlayer should be introduced to model the quantum tunneling effect [16]. At a reverse bias, holes in black phosphorus and electrons in MoS 2 near the interface tends to be depleted, leaving a stronger built-in electric field. When the incident light impinges on the junction, the excess carriers generated are swept by the built-in electric field. The additional current due to the drift effect of built-in field can be denoted as I pv . The total photocurrent I total consists of photoconductivity part I ph and I pv . I pv keeps its sign despite the direction of the bias, while I ph turns negative with a negative bias and its absolute value magnitude is controlled by the magnitude of the lateral electric field, which is dominated by the external bias. If a little negative bias is applied such that |I pv |>|I ph |, the total photocurrent I total = I pv − |I ph | is still positive. However, the total photocurrent I total = I ph + I pv changes to negative if the magnitude of the applied negative bias is large enough. In Fig. 2(b), this is the case of a bias of −0.1 V. If the negative bias is set to −0.3 V, the photoconductivity effect dominates the photocurrent and we get a photocurrent of −2 μA.
The photocurrent for a 808 nm is plotted in Fig. 2(c). In this setup, a relatively long time of light exposure is performed before sampling until the dark current is almost flat. The thermal contribution to the photocurrent can be 150 nA, one third of the The BP-MoS 2 heterojunction is gate tunable [12], [19]. In our experiment, response time and on-off ratio can be tuned obviously as shown in Fig. 3. To analyze the response, a combination of gate voltage and bias voltage is applied on the device. When a large bias is induced, the excess electrons and holes are drifted by the applied voltage, then the response speed increases. The relationship of drain current and gate voltage shown in Fig. 3(a) proves that the BP-MoS 2 heterojunction behaves like a n type semiconductor, the majority carrier is electron. With a negative gate voltage, the electrons are expelled from the channel, so the number of the total carriers decreases accordingly. Then an incident light induced excess carriers can play a dominant role in current contribution. As a result, a relatively high on-off ratio is achieved in this condition.
The photocurrent is constantly evaluated with a laser power ranging from 1.12 nW to 7.78 μW, gate voltage ranging from −20 V to 40 V. As shown in Fig. 3(a), the dark current and photocurrent increase when the gate voltage is swept from −20 V to 40 V. At the bottom left corner of Fig. 3(a), a minimum of dark current is got. In this setup, the dark current is 1.75 nA and the corresponding total current under 11.9 μW incident light is 117 nA, yielding an on-off ratio of 66.9. When the gate voltage is swept to 40 V, the dark current increases to 800 nA and the photocurrent is 200 nA, yielding an on-off ratio of 1.25. Fig. 3(b) shows an overall photocurrent with varying the gate voltage, keeping a bias of 3 V. The magnitude of photocurrent keeps smooth among different gate voltage cases. With a negative gate voltage, the photocurrent can keep steady. But as the gate voltage increases, the photocurrent drops as time passes. Combined with the experiment and conclusion in Fig. 2, along with the intrinsic feature reported in reference, a heat dissipation induced photocurrent decrease is inferred. There is a decay of total current with a large time constant.
The next interesting feature of the BP-MoS 2 heterojunction is the speed of the photocurrent. With the speed of the device determined, the normalized detectivity can be estimated. The typical response speed under 650 nm is relatively fast, shown in Fig. 3(c). The rise time is 34 μs and the falling time is 58 μs. The response speed for 980 nm laser is shown in Fig. 3(d). The rising time is 46 ms and the falling time is 49 ms. The incident light processes a switching speed fast enough through a calibrating work using a InGaAs commercial photodetector.
With the noise estimation approach we can calculate the detectivity for our heterojunction. The radiation power on the device is 59 nW and the photocurrent is 25 nA, thus the responsivity is 0.43 A/W. For the sampling rate could be low, we take a conservative mathematical analog rather than making Fourier transform to the signal [20]. The bandwidth is estimated thorough the equation 1/(η(t r + t f )), where t r and t f resembles the rising time and falling time and η is an phenomenological coefficient. The rms noise is evaluated from moving average method for visible and near infrared wavelength. Then we can calculate the noise equivalent power(NEP) as NEP = P SD R = 6.97 × 10 −14 W · Hz −1/2 and the corresponding normalized detectivity D * 1 = A 1 2 /N EP = 1.45 × 10 10 Jones [21]. The low detectivity here is because that we use high incident light and a low bias.
With the same experiment condition for 650 nm, a lower incident light power gives a better vision of NEP. If the incident light is switched to 3.57 nW, the photocurrent is around 7 nA. Estimation of responsivity gives 1.96 A/W and detectivity is 6.6 × 10 10 Jones.
The noise of the BP-MoS 2 heterojunction is mainly located in low frequency range, as shown in Fig. 4(a).The main source of this noise is flicker noise. Actually, this is also true for a polysilicon resistor [22], MOSFET [23], [24] and many other systems [25]. However, with correlated double sampling (CDS) technique [26], low frequency noise can be reduced effectively. And this may be the reason of the appearance of the I Dark dependent version of normalized detectivity. But in the scope of pure photodetector without discussing the read-out circuit, the normalized detectivity D * or NEP considering the flickering noise is proper to estimate the performance of the device. The photocurrent for mid infrared is in Fig. 4. Here a negative photo-response can be time resolvable. The incident light serves as heater and the black phosphorous sheet presents its metallicity. A dark current of ∼1.2 nA is set for the BP-MoS 2 heterojunction and the photocurrent is approximately equally in value but opposite in direction, shown in Fig. 4(b). If the incident radiation is set to 1.11 mW/cm 2 , the photocurrent is 1.2 nA. The radiation on the device is calculated to be 6.9 nW based on its area. The power spectral density of noise is extracted using a pwlech algorithm. It is a general spectral estimation utilizing fast Fourier transform. The result is in illustrated in Fig. 4(a). The rms noise is evaluated by integrating directly the power of signal density (PSD) over frequency bandwidth and taking the square root of the integrating result. As a result, the NEP is evaluated as NEP = 8.2 pW , the corresponding D * is 2.02 × 10 8 Jones. A total current falls around zero. We restricts bias low to decrease flicker noise, as the flicker noise can varies in order of magnitude with bias. In fact, time resolvable photocurrent cannot be extracted if a large dark current is present for both black phosphorous or BP-MoS 2 devices.
In Fig. 4(b), the relationship of mid-infrared photocurrent and the incident power for 4.5 μm light is shown. The photocurrent is negative. As the incident power is increased, it takes longer to reach the maximum photocurrent. The total current get saturable to around zero, and thorough mechanism requires more exploration.
Referring to a recent overview of black phosphorus based photodetectors [27] and some related works, a summary of photodetector based on BP/MoS 2 Van Der Waals heterojunction and other similar structures is listed in Table I. Most of them focus on the visible to near-infrared wavelength. The response time in ref [11] and ref [12] is 15 μs and 13 μs, faster than that of this work. Our device provides a overall estimation of BP/MoS 2 Van Der Waals heterojunction in a wide waveband.

IV. CONCLUSION
Heterostructures of black phosphorous(BP) and Molybdenum Disulfide(MoS 2 ) can be used for photodetection from visible to mid-infrared wavelength up to 4.5 μm. The device possesses a low dark current less than 0.1 μA for short wavelength light. The responsivity under a 520 nm is 0.69 A/W and the detectivity for a typical 650 nm laser is 6.6 × 10 10 Jones. The competition of photo-electronic and photo-thermal mechanism of photodetectors is account for the response speed difference for a visible light and near infrared radiation. The heterojunction is gate tunable, so the on-off ratio can be adjustable. Photon injection and joule heat can make hybrid enhancement on the photocurrent. Time resolvable photocurrent for mid-infrared radiation is achieved. A responsivity of 2.2 mA/W and a D * of 2.02 × 10 8 Jones is recorded for 4.5 μm md-infrared radiation. The phenomena and theory could be beneficial for the designing of photodetectors based on two-dimensional Van Der Waals heterojunction.