Solar-blind ultraviolet detection based on TiO₂ nanoparticles decorated graphene field-effect transistors

1 Introduction

Ultraviolet (UV) photodetectors are important for many applications such as space communication, flame detection, military surveillance, industrial quality control, and environmental monitoring [1], [2], [3], [4], [5], [6], [7], [8], [9], [10]. Due to the excellent electrical and optical properties, such as ultrafast carrier mobility [11], [12], high conductivity, ultra-wide spectral range (from UV to terahertz) [13], [14], [15] and tunable optical properties via electrostatic doping [16], [17], graphene has attracted enormous attention as a promising candidate for high-performance photodetectors. Especially, the pronounced photon absorption near the saddle-point singularity makes graphene well suited for UV photodetectors [18]. However, the photoresponsivity of pure graphene-based photodetectors is still limited within 1–2×10⁻² A W⁻¹ [19], [20], [21], owing to the intrinsic low optical absorption and short minority carrier lifetime of graphene [22].

A feasible way to increase the photoresponsivity of graphene-based photodetectors is to construct Schottky-diode-like graphene-semiconductor heterojunctions [23], [24], [25], [26], [27]. Using a graphene-silicon heterojunction device, a photoresponsivity about 0.1 A/W was obtained in the UV region [23]. By spin-coating a layer of titanium dioxide (TiO₂) nanoparticles (NPs, size, 3–5 nm) [24] or Ag NPs [25] on such devices for UV light absorption, the photoresponsivity could be further increased by up to ~100%. To further increase the photoresponsivity of graphene-based UV photodetectors, graphene membranes were integrated with various semiconducting light harvesters, such as transition-metal dichalcogenides [28], [29], fullerene (C₆₀) [30], organic molecules [31], and quantum-dots (QDs) [32], [33], based on the photogating effect. For example by synthesizing a layer of silicon QDs doped with boron on a back-gated graphene field-effect transistor (GFET), a photoresponsivity up to ~10⁸ A/W and a detectivity of ~10¹² Jones were obtained at 375 nm [33]. However, as most of the existing light harvesters have modest bandgaps, the corresponding graphene UV photodetectors are not solar-blind, limiting their applications to some military and civilian fields such as missile interception, biological analysis, and UV radiation monitoring below the ozone hole [26]. Furthermore, the synthesis techniques of some light harvesters and the processes of integrating them with graphene are complex and require special equipment, leading to high cost and unavailability. Therefore, sensitive, solar-blind, and cost-effective UV photodetectors are still highly desirable.

In this paper, a simple and feasible technique was demonstrated to realize highly sensitive and solar-blind UV photodetectors, by decorating a buried-gated GFET with solution-synthesized TiO₂ NPs. The buried-gate structure allows tunable optical properties via a small gate voltage, while the solution-synthesized TiO₂ NPs, which are solar-blind, nontoxic, low-cost and long-term stable [24], [34], [35], significantly increases the photoresponsivity. Under the illumination of a 325-nm laser with an intensity of 11 W/cm², a photoresponsivity of 118.3 A/W and a detectivity of 1.75×10¹¹ Jones were obtained, at the conditions of zero gate bias and a source-drain bias voltage of 0.2 V. Furthermore, the photoresponsivity and photoresponse time of the devices can be easily modulated by applying a small (≤1 V) gate bias or/and changing the source-drain bias voltages, to meet the needs in different application situations.

2 Experiments

The buried-gate GFETs were fabricated using a similar method shown in our previous work [36]. The buried chromium (Cr)/gold (Au) gate electrodes with thicknesses of 10 and 30 nm were deposited on a substrate consisted of 200-nm-thick silicon nitride (Si₃N₄), 50-nm-thick aluminum (Al), and 500-μm-thick silicon (Si), using magnetron sputtering. Next, a layer of silicon dioxide (SiO₂) with the thickness of 30 nm was deposited by plasma-enhanced chemical vapor deposition, to form the dielectric layer. Then, monolayer graphene (ACS Material, LLC) was transferred onto the SiO₂ layer and patterned using oxygen plasma etching. After this step, Cr/Au (10 nm/50 nm) source and drain electrodes were deposited on the graphene layer by electron beam evaporation and the buried-gate GFETs were obtained (Figure 1A).

Figure 1:

Titanium dioxide (TiO₂) nanoparticles (NPs) decorated buried-gate graphene field-effect transistors (GFETs).

(A) and (B) Buried-gate GFETs before and after the decoration of TiO₂ NPs on the surface of graphene. (C) Scanning electron microscopy micro-image of a 2×5 array of TiO₂ NPs decorated GFETs. (D) Zoomed-in details of the graphene/TiO₂ NPs conductive channel. (E) Atomic force microscopy micro-image of agglomerated TiO₂ NPs (marked in red box in Figure 1D). (F) Typical Raman spectra of the GFETs with (red line) and without (blue line) TiO₂ NPs.

The TiO₂ NPs were synthesized using the sol-gel method [24]. Firstly, a mixed solution of TiCl₄ (1 ml), ethanol (5 ml), and benzyl alcohol (35 ml) with a volume ratio of 2.4:12.2:85.4 was heated at 80°C for 6 h. The TiCl₄, ethanol and benzyl alcohol were purchased from Sinopharm Chemical Reagent Co., Ltd., Shanghai, China. Then, the product was washed three times with diethyl ether (Sinopharm Chemical Reagent Co., Ltd., Shanghai, China) and followed by centrifugation at 12000 rpm (9200 g). for 10 min, to obtain the white TiO₂ precipitate. Finally, the TiO₂ precipitate was dispersed in ethanol to obtain an ethanol suspension of TiO₂ NPs with a mass concentration of 4 mg/ml. The average diameter of the TiO₂ NPs synthesized using this method is about 3–5 nm [24]. By dropping the suspension of TiO₂ NPs (4 mg/ml) onto the surface of graphene of the buried-gate GFETs, and drying them in the air, TiO₂ NPs decorated GFETs were obtained (Figure 1B).

Scanning electron microscopy (SEM) and atomic force microscopy (AFM) measurements were carried out to characterize the morphological and structural properties of the buried-gate GFETs before and after the TiO₂ NPs decoration process. Raman spectroscopy and energy dispersive X-ray spectroscopy (EDS) were also performed to analyze the components of the devices. The electrical properties of the buried-gate GFETs with and without TiO₂ NPs were investigated using a semiconductor parameter analyzer (B1500A, Keysight) and a probe station (Summit 12000, Cascade) at room temperature.

In order to investigate the photoelectrical properties of the TiO₂ NPs decorated GFETs, the wafer was diced into small (5 mm×5 mm) chips, and bonded onto printed circuit boards with Au wires. In the photoelectrical experiments, a 325-nm laser beam with a spot size of ~2 μm, which was generated by a laser confocal Raman spectrometer (LabRam HR-800, Horiba JobinYvon), was illuminated to the conductive channel of the TiO₂ NPs decorated GFETs. A precision source/measure unit (B2911A, Agilent) was used to supply the source-drain bias voltage and monitor the source-drain current of the GFETs, while a source-meter (2400, Keithley) was used to supply the gate bias. All photoelectrical measurements were performed at room temperature and under ambient conditions.

3 Results and discussion

The schematics of buried-gate GFETs before and after the decoration of TiO₂ NPs are shown in Figure 1A,B, respectively. The dimension (length×width) of the graphene conductive channel is 30 μm×200 μm, while the lengths of the source, drain, and gate electrodes and the intervals between them are 10 μm. Figure 1C shows a SEM micro-image of a 2×5 array of TiO₂ NPs decorated GFETs, indicating the potential for large-scale integration of the devices, which is important in sensing and imaging areas [13], [30], [37]. From the zoomed-in SEM micro-image of a TiO₂ NPs decorated GFET (Figure 1D), the source, drain, and gate electrodes are well constructed, and the conductive channel is relatively smooth, except for several agglomerates. An EDS measurement was performed on one of the agglomerates in the center of the conductive channel (labeled by red box in Figure 1D). Significant Ti and O peaks were observed in the EDS spectroscopy (Figure S1C), indicating the agglomerates consisted of TiO₂ NPs. More EDS measurements were carried out at different positions in the conductive channel, and demonstrated that the whole channel was coated with a thin layer of TiO₂ NPs (Figure S1C–S1F).

Atomic force microscopy measurements were also performed to characterize the morphology of the TiO₂ NPs decorated graphene conductive channel, as shown in Figure 1E. It can be seen that the bottom diameter of the TiO₂ NPs agglomerate is about 2 μm and the height is about 1 μm. The average thickness of the TiO₂ NP layer far away from the agglomerate is less than 500 nm. According to the work reported by Zhu et al. [24], the optimal thickness of the TiO₂ NP layer for UV light harvesting is between 250 and 500 nm, and the agglomerates of TiO₂ NPs will decrease the UV light harvesting efficiency. Therefore, the TiO₂ NPs decoration process should be optimized carefully. We further characterized the buried-gate GFETs with and without TiO₂ NPs using the Raman spectroscopy and confirmed the presence of both graphene and TiO₂ in the TiO₂ NPs decorated GFETs (Figure 1F). For the buried-gate GFETs without TiO₂ NPs, the characteristic G and 2D peaks of graphene locate at 1594 and 2686 cm⁻¹, respectively. After the decoration of TiO₂ NPs, both G and 2D peaks of graphene are slightly up-shifted, 1 cm⁻¹ for the G peak and 5 cm⁻¹ for the 2D peak. Simultaneously, the characteristic TiO₂ peaks are observed at 153, 400, 508, and 640 cm⁻¹ [38], [39], in the TiO₂ NPs decorated GFETs.

Figure 2A shows the transfer curves of the buried-gate GFETs before and after the TiO₂ NPs decoration. A typical ambipolar characteristic can be clearly observed in the GFET without TiO₂ NPs, where the Dirac point, which is defined as the gate voltage (V_gs) at minimum conductance, locates at 9.8 V, indicating the graphene is p-type doped by H₂O, O₂, or other contaminations [40], [41]. After the TiO₂ NPs decoration, no Dirac point is observed within the gate voltage range from −2 to 20 V, and the source-drain current decreases monotonously with the increase of gate voltage. It is worth noting that the slope of the transfer curve of the GFET with TiO₂ NPs is smaller than that of the GFET without TiO₂ NPs in the p-type linear region. The source-drain current (I_ds) of the former is also smaller than that of the latter, under the same gate voltages, indicating the conductivity of the graphene degrades in the TiO₂ NPs decoration process. These phenomena were opposite with the observation reported by Zheng et al. [35], and need to be carefully investigated in the future. From the enlarged transfer characteristic curve of the TiO₂ NPs decorated GFET (Figure 2B), the gate modulation effect still exists, although it is weaker than that of the GFET without TiO₂ NPs. In order to investigate the effect of TiO₂ NPs on the carrier mobility (μ) and contact resistance (R_contact) of the device, a model proposed by Kim et al. [42] was used to analyze the resistive behavior of the devices. Detailed information about the analysis can be found in Figure S2. For the GFET without TiO₂ NPs, μ=764 cm²/(V s), and R_contact=147 Ω. After coating TiO₂ NPs, the carrier mobility of the GFET decreases to μ=690 cm²/(V s), and the contact resistance increases to R_contact=405 Ω. These phenomena can be attributed to increased disorder caused by perturbations to the graphene surface during the coating process [43], [44]. Figure 2C,D display the output characteristics of the buried-gate GFETs without and with TiO₂ NPs, respectively. Under a certain gate voltage, the source-drain current of the GFET without TiO₂ NPs increases linearly with the increase of the gate voltage (Figure 2C). Similar phenomenon is also observed in the GFET with TiO₂ NPs (Figure 2D). The source-drain current of the GFETs with and without TiO₂ NPs also changes with the gate voltage, indicating reliable gate modulation capability.

Figure 2:

Electrical properties of the GFETs before and after the TiO₂ NPs decoration process.

(A) Comparison of the transfer characteristics for GFETs with (red curve) and without (blue curve) TiO₂ NPs. (B) The transfer characteristic of the TiO₂ NPs decorated GFET. (C) and (D) Output characteristics of the GFETs without and with TiO₂ NPs, respectively.

The photoelectrical properties of the buried-gate GFETs with and without TiO₂ NPs decoration were characterized using an experimental setup shown in Figure 3A. Figure 3B shows the on/off source-drain current (I_ds) of the TiO₂ NPs decorated GFET under the illumination of a 325-nm laser with a power of 3.47 mW, at the conditions of zero gate bias (V_gs) and a source-drain bias voltage (V_ds) of 0.1 V. As mentioned above, the Dirac point of the TiO₂ NPs decorated GFET is over than 20 V (Figure 2B), the carrier transport in the graphene channel of the TiO₂ NPs decorated GFET under zero gate bias is hole-dominant. When the TiO₂ NPs decorated GFET is exposed to UV light, the ground-state electrons in TiO₂ NPs are activated into the excited state, producing a large amount of free electrons which are injected into the graphene layer (see the inset of Figure 3C). The photogenerated electrons from TiO₂ NPs and photogating-induced electrons recombine with the dominant holes in the graphene channel, leading to a decreased conductivity. Thus, the I_ds decreases immediately and significantly when the laser is turned on. Once the laser is turned off, the photogenerated electrons and holes in the TiO₂ NPs recombine with each other, resulting in that the density of the dominant holes in the graphene channel recovers to its original level, so does the I_ds. After several on-off cycles, the photocurrent, which is defined as the difference between the source-drain current under UV light illumination and the dark source-drain current, is well retained, demonstrating good reliability and reversibility of the devices. It is worth noting that there was no significant photocurrent observed under the 514-nm laser illumination (Figure S3), at the same conditions. These results indicate that the TiO₂ NPs decorated GFET photodetector is solar-blind.

Figure 3:

Photoresponse of the TiO₂ NPs decorated GFETs.

(A) Schematic of the experimental setup. (B) Temporal photocurrent response of the TiO₂ NPs decorated GFETs, under zero gate bias (V_gs) and a source-drain voltage (V_ds) of 0.1 V. (C) A single modulation cycle of temporal photocurrent response for the TiO₂ NPs decorated GFET. The inset illustrates the operation mechanism of the device. (D) The laser position dependence of the photocurrent, under the conditions of V_gs=0 V and V_ds=0.1 V. Inset: SEM micro-image of the device with 7 laser positions. A 325-nm laser was used in the experiments, the laser powers for (B) and (C) were 3.47 mW, while that for (D) was 1.74 mW.

To investigate the photoresponse speed of the TiO₂ NPs decorated GFET photodetectors, the rise/decay time is defined as the photocurrent increases/decreases from 10/90% to 90/10% of the stable photocurrent. From a zoomed-in single modulation cycle (Figure 3C), the rise time and the decay time are extracted to be 2.0 and 158.7 s, respectively. The photoresponse speed of the TiO₂ NPs decorated GFET photodetector is relatively slow, which can be attributed to the long-lived charge trapping achieved by the spatial separation of photoexcited charges across the graphene/TiO₂ interface [35], [39]. The lifetime of the trapped charges in the semiconductor light harvesters is influenced by many factors, such as defect types, trap-state density, and defect trap depths [45]. By varying the above factors, the charge trap lifetime can be significantly reduced [33]. Furthermore, the photoresponse speed of the TiO₂ NPs decorated GFET photodetector can also be tuned by the source-drain bias and gate bias, which will be discussed below.

Thanks to the small spot size of the laser (~2 μm), scanning photocurrent experiments were performed by mechanically moving the TiO₂ NPs decorated GFET photodetector step by step to change the laser beam position between the source and drain electrodes, as illustrated in the inset of Figure 3D. The experimental result (Figure 3D) demonstrates that photocurrent generation clearly arises from the graphene/TiO₂ NPs hybrid channel rather than electrode/channel junctions or other regions, indicating the photogating effect dominates the photoresponse [46]. More experiments found that the laser position on the TiO₂/graphene channel of the device affects not only the photocurrent but also the photoresponse speed, as shown in Figure S4. In the following experiments, the laser position was fixed in the middle of the channel.

Figure 4A shows temporal photoresponses of the TiO₂ NPs decorated GFET under different source-drain bias voltages (V_ds), at the conditions of an incident laser power of 1.74 mW (325 nm) and zero gate bias (V_gs=0 V). With the increase of the V_ds, the photocurrent increases, as well as the dark current. Furthermore, the rise times (τ_on) of the photoresponses under different source-drain bias voltages of 0.1, 0.2, and 0.4 V are 3.23, 3.75, and 3.91 s, respectively. This result indicates that the photoresponse speed becomes slower with the increase of the source-drain bias voltage. The photoresponsivity (R_ph) is defined as R_ph=I_ph/P, where P is total power of the incident laser. More experiments showed that the R_ph increases almost linearly with the V_ds, as shown in Figure 4B. When the V_ds increases from 0.1 to 1 V, the R_ph increases 7.2 times. Liu and Kar [47] and Liu et al. [48] also observed similar phenomena of the linear bias voltage dependence of photoresponsivity. The photoconductive gain can be calculated by using τ/t_tr, where τ is the lifetime of the trapped charges, and t_tr is the time for the transfer of carriers through the device channel [49]. As t_tr=L²/(μV_ds), where L is the length of the device channel and μ is the carrier mobility, the photoconductive gain is proportional to V_ds. The tunable photoresponsivity of photodetectors is important in some practical applications such as imaging for variable light intensity.

Figure 4:

Photoelectrical properties of the TiO₂ NPs decorated GFET photodetectors.

(A) Photoresponses at different source-drain bias voltages (V_ds), under the conditions of zero gate bias (V_gs=0 V) and an incident laser power of 1.74 mW. (B) The normalized photoresponsivity [R_ph/R_ph,0, where R_ph,0 is the photoresponsivity (R_ph) obtained at V_ds=0.1 V] as a function of the V_ds, under the conditions of an incident laser power of 1.74 mW and V_gs=0 V. (C) Photoresponses under different incident laser powers. (D) The photoresponsivity as a function of incident laser power, at V_ds=0.1 and V_gs=0 V.

The photoresponses of the TiO₂ NPs decorated GFET under different incident laser powers are plotted in Figure 4C. The source-drain bias voltage is 0.1 V, while the gate bias voltage is 0 V. It can be seen that the photocurrent rises with the increase of the incident laser power. Furthermore, the rise times (τ_on) of the photoresponses under different laser powers of 3.47 mW, 0.35 mW, and 34.7 μW are 1.1, 12.7, and 91.4 s, respectively, as shown in Table 1, indicating that higher laser power results in faster photoresponse speed. This phenomenon was also observed by Tan et al. in a black phosphorus carbide phototransistor based on the photogating effect [50].

Figure 4D shows the photoresponsivity of the TiO₂ NPs decorated GFET photodetector as a function of the incident laser power, under the conditions of zero gate bias and a source-drain bias of 0.1 V. As with most graphene-based phototransistors based on the photogating effect [47], [48], the photoresponsivity decreases at high incident photon densities due to the saturated absorption and the weakened built-in field [32]. A photoresponsivity as high as 60.3 A/W was obtained under the laser power of 0.35 μW (11.1 W/cm²). It is worth noting that the laser power mentioned here is the total incident laser power, rather than the laser power on the graphene conductive channel. By simply increasing the source-drain bias voltage from 0.1 to 0.2 V, the photoresponsivity of the TiO₂ NPs decorated GFET can be increased to 118.3 A/W (Figure S5), which is much higher than that of the GFET before coating TiO₂ NPs (0.78 μA/W, as shown in Figure S6). Furthermore, this photoresponsivity (118.3 A/W) is over 600 times higher than that of a recently reported solar-blind UV photodetector based on graphene/vertical Ga₂O₃ nanowire array heterojunction (0.185 A/W) [26]. We would like to point that the photoresponse time of our device now is in the level of several seconds to hundreds of seconds, which is much slower than that of the graphene/Ga₂O₃ photodetector (~8 ms) [26]. A major reason is that the graphene/Ga₂O₃ photodetector is a photodiode, while our device is a phototransistor. The typical photoresponse time of graphene-based phototransistors is in the level of several seconds, such as the Si quantum dot/graphene phototransistors reported by Ni et al. [33] and the graphene/MoS₂ phototransistors reported by Roy et al. [51]. Other reasons for the slow photoresponse of our device are related to the photosensitive materials and their morphologies. For example, there are a lot of trap states in TiO₂ NPs and the graphene/TiO₂ interface, which trap photogenerated electrons, resulting in a slow photoresponse [24]. In order to improve the photoresponse speed of the TiO₂ NPs decorated GFET photodetectors, gate pulses [52] may be employed to sweep out the trapped carriers of TiO₂ NPs within an appropriate time frame in the future. Another important parameter of photodetector is the specific detectivity (D*), which determines the capability of a photodetector to detect a weak light signal, can be given as D⋆=R⋅A1/22⋅e⋅Idark [53], where R is photoresponsivity, A is the effective photoactive area of detector, e is the electron charge, and I_dark is the dark current. Using the photoresponsivity of 60.3 A/W, a D*=1.24×10¹¹ jones (1 jones=1 cm Hz^1/2 W⁻¹) is obtained, which is comparable to the best performances of reported graphene-based UV photodetectors [39], [54]. Importantly, the photoresponsivity and detectivity can be further enhanced at lower incident laser powers [27], [29], [53].

Temporal photoresponses of the TiO₂ NPs decorated GFET under different gate bias voltages are plotted in Figure 5A. The magnitudes of the photocurrents under different gate bias voltages are different, so are the photoresponse speeds. The photocurrent under V_gs=–0.5 V is 52.7% larger than that under V_gs=–1 V, while the rise time of the photoresponse under V_gs=–0.5 V is 65.7% slower than that under V_gs=–1 V. More experiments showed that the photoresposivity of the TiO₂ NPs decorated GFET can be modulated by gate bias voltages, as shown in Figure 5B. With the increase of the gate voltage from 0 to 1 V, the photoresponsivity increases first and then decreases the threshold gate voltage which is about 0.5 V. Similar characteristic is also observed in the negative gate voltage range. By applying a –0.5 V voltage to the buried-gate electrode, the photoresponsivity can be increased by ~20%. These results indicate both the photoresponsivity and the photoresponse speed can be modulated by the gate bias. Table 2 summarizes the reported high-performance UV photodetectors based on graphene. Obviously, the device in this work has extremely high photoresponsivity under small bias voltages, but the photoresponse speed needs to improve in the future.

Figure 5:

The gate-voltage-dependent photoresponse of the TiO₂ NPs decorated GFET photodetectors.

(A) Temporal photocurrent responses of the TiO₂ NPs decorated GFET photodetector at different gate bias voltages (V_gs), under the conditions of a laser power of 0.347 mW and V_ds=0.1 V. (B) The gate-voltage-dependent photoresponsivity (R_ph) of the TiO₂ NPs decorated GFET photodetectors, where R_ph,0 is the R_ph at V_ds=0.1 V and V_gs=0 V. The incident laser power is 0.347 mW.

4 Conclusions

A highly sensitive, solar-blind, and cost-effective UV photodetector based on a buried-gate GFET decorated with TiO₂ NPs was demonstrated in this paper. The TiO₂ NPs significantly enhance the UV light absorption of the device. Under the illumination of a 325-nm laser with an incident power of 0.35 μW, a photoresponsivity as high as 118.3 A/W and a detectivity of ~10¹¹ Jones were obtained, at zero gate bias and a source-drain bias voltage of 0.2 V. Importantly, the photoresponsivity can be further enhanced by increasing the source-drain bias voltage or properly tuning the gate bias voltage. Furthermore, the photoresponse speed of the TiO₂ NPs decorated GFET photodetectors can also be tuned by the source-drain bias and gate bias.

Supplementary material: See Supplementary material for energy X-ray spectrum of the buried-gate TiO₂ NPs GFETs, the carrier mobility and graphene-metal contact resistance of the buried-gate GFET with and without TiO₂ NPs, the photoresponse of the device under the 514 nm illumination, the influence of the incident laser position on photoresponse time, the photoresponsivity of the device before coating TiO₂ NPs, and the temporal photoresponse of the buried-gate TiO₂ NPs GFET at V_ds=0.2 V.