Tunable Photodetectors via In Situ Thermal Conversion of TiS₃ to TiO₂

Abstract

In two-dimensional materials research, oxidation is usually considered as a common source for the degradation of electronic and optoelectronic devices or even device failure. However, in some cases a controlled oxidation can open the possibility to widely tune the band structure of 2D materials. In particular, we demonstrate the controlled oxidation of titanium trisulfide (TiS₃), a layered semicon-ductor that has attracted much attention recently thanks to its quasi-1D electronic and optoelectron-ic properties and its direct bandgap of 1.1 eV. Heating TiS₃ in air above 300 °C gradually converts it into TiO₂, a semiconductor with a wide bandgap of 3.2 eV with applications in photo-electrochemistry and catalysis. In this work, we investigate the controlled thermal oxidation of indi-vidual TiS₃ nanoribbons and its influence on the optoelectronic properties of TiS₃-based photodetec-tors. We observe a step-wise change in the cut-off wavelength from its pristine value ~1000 nm to 450 nm after subjecting the TiS₃ devices to subsequent thermal treatment cycles. Ab-initio and many-body calculations confirm an increase in the bandgap of titanium oxysulfide (TiO_2-xS_x) when in-creasing the amount of oxygen and reducing the amount of sulfur.

1. Introduction

Low-dimensional semiconductors are attracting increasing interest in the scientific community working on optoelectronic devices thanks to their outstanding optical and electronic properties combined with reduced dimensionality [1,2,3]. The large surface-to-volume ratio of two-dimensional (2D) materials benefits many applications such as gas-sensing, but it may enhance the sensitivity of these materials to oxidation compared to bulk materials. Moreover, lattice vacancies and atomic-level defect combined with the presence of light can accelerate the oxidation process [4,5,6,7,8], which is typically accompanied by a degradation of the electrical and optical properties reducing the device performance [9,10]. Furthermore, shining high intensity light on 2D materials can induce additional processes of photo-oxidation [11,12,13,14]. The overall performance reduction induced by oxidation seems to be one of the main issues to solve in developing industrial applications based on 2D materials, therefore controlling the oxidation process is a very active subject for both fundamental and applied research in the context of band engineering.

Titanium trisulfide (TiS₃) is a layered semiconductor which has attracted much attention recently thanks to its quasi-1D electronic and optoelectronic properties [15,16,17,18] and its direct bandgap of 1.1 eV [19,20,21,22,23,24,25]. Using first-principles calculations, Iyikanat et al. showed that TiS₃ can react with various forms of oxygen [26]. An experimental demonstration was given by Molina-Mendoza et al., who reported thermogravimetric analysis (TGA) of bulk TiS₃ in oxygen atmosphere showing the partial conversion of the material into TiO₂, a large bandgap (3.2 eV) insulator with a wide range of applications [19,27,28,29]. In this article we investigate the controlled thermal oxidation of individual TiS₃ nanoribbons and its influence on the optoelectronic properties of TiS₃-based photodetectors. We first study the oxidation of TiS₃ powder and single nanoribbons deposited on a glass substrate. Using Raman spectroscopy and optical analysis we can monitor the material properties as a function of time while heating at 320 °C in air. We find that an individual TiS₃ nanoribbon converts to crystalline TiO₂ in approximately 10 min. Control experiments performed on TiS₃ nanoribbons fully encapsulated between hexagonal boron nitride flakes confirm that the direct contact between TiS₃ and air is necessary for the oxidation process to happen. After establishing the change in material properties we demonstrate the controlled oxidation of a TiS₃ nanoribbon photodetector that allows tuning the cut-off wavelength and sensitivity of the device. By monitoring the change in its current–voltage characteristics and in its spectral photoresponse, we find that the cut-off wavelength is blue-shifted upon oxidation, reaching a cut-off wavelength of 450 nm (while it is ≈1100 nm for pristine TiS₃ according to previous works [19]). Various intermediate states are observed, demonstrating the tunability of the nanoribbon bandgap. Ab-initio and many-body calculations confirm an increase in the bandgap near to that of titanium oxysulfide (TiO_2-xS_x) when increasing the amount of oxygen and reducing the amount of sulfur.

2. Materials Synthesis

The starting TiS₃ material was synthetized by a solid-gas reaction using Ti powder and sulfur powder sealed into a quartz ampoule and kept at 550 °C for 20 h. Additional details about the synthesis and elemental characterization of TiS₃ can be found in the references [17,30,31]. Figure 1a shows an optical picture of TiS₃ powder while Figure 1b shows the same powder after heating it for 5 min at 350 °C in air. After heating the material, we observed a dramatic change of its appearance, with a clear color from black to white, due to the conversion of TiS₃ to TiO₂. Thanks to the layered structure of TiS₃, individual nanoribbons can be isolated by mechanical exfoliation. To study an individual TiS₃ nanoribbon, we first exfoliated the powder onto Nitto tape and then transfered part of the flakes from the tape to a viscoelastic polidimethylsiloxane (PDMS) stamp. After the identification, we transfered the chosen nanoribbon to a different substrate (such as glass, SiO₂/Si…) with an all-dry deterministic transfer method [32,33].

A typical TiS₃ nanoribbon transferred onto a glass slide is shown in Figure 1c. The microscope picture is recorded in transmission mode and the nanoribbon appears black since it is absorbing most of the white light due to the bandgap of 1.1 eV. The nanoribbon has a length of approximately 150 µm (oriented along the crystal b axis) and a width of 5 µm (a axis). Figure 1d shows the same nanoribbon after heating it for 5 min at 350 °C in air. The morphology of the nanoribbon appears intact, but the substantial change in color indicates that its absorption, which is related to the bandgap and the band-structure, has changed dramatically. The change from black to white/transparent due to a reduction of the light absorption indicates an opening of the bandgap. A higher resolution picture of the initial and final status of TiS₃ nanoribbons can be obtained using electron microscopy. Figure 1e,f shows a field emission gun scanning electron microscopy (FEGSEM) image of TiS₃ nanoribbons before and after the heat treatment. As can be seen from the images, the nanoribbons morphology is maintained after the heat treatment, although the final nanoribbons show higher roughness.

3. Experimental Results and Discussion

3.1. Thermal Oxidation Analysis

To gain deeper insight into the thermal oxidation process of TiS₃ we used TGA coupled to mass spectrometry (MS). Figure 2a shows a TGA curve of TiS₃ kept under a flux of 90 mL/min of air and heated at a rate of 10 °C/min. The graph of temperature as a function of time, shown in the top panel of Figure 2a, displays a discontinuity at approximately 30 min, indicative of an exothermic reaction occurring at 300 ± 10 °C. This reaction is accompanied by a loss of approximately 43% of the initial mass (see the bottom panel of Figure 2a) that is consistent with the difference between the mass of TiS₃ and TiO₂ (44%). These results indicate that the conversion of TiS₃ into TiO₂ takes place under atmospheric conditions starting at 300 °C. A more in-depth look can be achieved using a mass spectrometer to detect the species present during the reaction. Figure 2b shows the ionic currents at m/q = 16, 32 and 48 (m/q is the ration between the atomic mass m and the atomic charge q of the species) as a function of time recorded during the TGA experiment, corresponding to O₂ (whose cracking pattern shows two signals) and SO₂. The dip observed at 30 min in the traces of m/q = 16 and 32 indicated that the O₂ present in the atmosphere was reacting with TiS₃. At the same time, the peak in current of m/q = 48 is consistent with the liberation of sulfur atoms from TiS₃ and their successive reaction with oxygen to form gaseous SO₂.

The composition of the final product after the thermal treatment of TiS₃ was studied with x-ray diffraction measurements (XRD). The x-ray diffraction measurements were performed using a X-pert PRO diffractometer under a θ/2 θ configuration. Figure 2c shows XRD patterns of TiS₃ before and after the heating treatment with the main diffraction planes indicated. In the pristine material a single polycrystalline phase is observed, TiS₃. All the diffraction peaks in the XRD pattern can be indexed to the phase monoclinic TiS₃ in good agreement with the standard JCPDS card no. 00-036-1337. The most intense diffraction peaks at Bragg angles 10.2°, 20.4°, 30.8° and 41.4°are assigned to the (001), (002), (003), (012) and (004) planes. After the heating, the XRD pattern changed significantly. The new diffraction peaks can be indexed to polycrystalline tetragonal TiO₂, anatase, whose more intense peaks at angles 25.3°, 38.6°, 37.8°and 48.1° can be attributed to the (101), (004), (112) and (200) planes in good agreement with the standard JCPDS card no. 00-021-1272. Comparing the two spectra it can be seen that the diffraction peaks after the treatment are broader than the ones before, indicating that the produced TiO₂ has crystallites of a smaller size. Overall, the previous results support a scenario in which TiS₃ heated above 300 °C in ambient conditions undergoes the global exothermic reaction: TiS₃ + 4O₂ → TiO₂ + 3SO₂, which converts the trisulfide in anatase TiO₂.

After characterizing the thermal oxidation of bulk TiS₃, we focused on individual nanoribbon oxidation which was interesting for the fabrication of high-quality optoelectronic devices. We started by using Raman spectroscopy to study the composition of a single nanoribbon. Figure 3a shows the Raman spectra of a TiS₃ nanoribbon recorded in its pristine form (after deposition onto a SiO₂/Si substrate) and during a heating cycle with spectra taken every 2 min while heating the sample at a temperature of 320 °C. The Raman spectra were recorded in a Raman Microscope (SENTERRA II, Bruker) while illuminating the sample with a laser of 532 nm focused in a circular spot (area ∼2 µm², power 2 mW, power density 1 mW/µm²) and an integration time of 20 s. The power density that we use is lower than the threshold density for photooxidation of the TiS₃, which we estimate to be 5 mW/µm² (see Section S4 of the Supporting Information). The Raman signal of the pristine TiS₃ shows four prominent peaks due to TiS₃ and a very weak peak at 520 cm⁻¹ due to the silicon substrate. The peaks at energies 177 cm⁻¹, 302 cm⁻¹, 371 cm⁻¹, and 559 cm⁻¹ correspond to A_g Raman modes of the TiS₃ nanoribbon and are in good agreement with the modes reported for bulk TiS₃ [34]. After heating up the sample we observe a reduction in the intensity of the TiS₃ peaks and an increase in the Si peak intensity during the first two cycles (4 min) that can be attributed to an increase in the transparency of the nanoribbon. After approximately 6 min of heating (after three cycles) we observed the quenching of the TiS₃ peaks with only the 520 cm⁻¹ Si peak visible in the Raman spectrum of the sample. The spectra recorded after 8, 10 and 12 min of heating show the appearance of a new peak centered at 142 cm⁻¹. This peak is consistent with the signature of an E_g Raman mode of TiO₂ [35,36]. The evolution of the Raman spectra of the nanoribbon shows that a pristine TiS₃ nanoribbon can be converted to TiO₂ by heating at 320 °C. The TiS₃ to TiO₂ oxidation process can be readily visualized from the plot in Figure 3b in which we show the extracted intensities of the 142 cm⁻¹ (TiO₂) and 302 cm⁻¹ (TiS₃) peaks as a function of the number of heating cycle. Apart from the change in the Raman signal, we also observe a clear change in the color of the nanoribbon deposited on the SiO₂/Si surface from green to yellow during the conversion process as shown in the inset of Figure 3b.

To study the role of the environment in the oxidation process of a single nanoribbon we fabricated a hexagonal boron nitride (h-BN) encapsulated TiS₃ nanoribbon on top of a transparent indium tin oxide (ITO) substrate. The top panel of Figure 3c shows the schematic of the samples. We first transferred a flake of h-BN onto the ITO surface and then transferred two TiS₃ nanoribbons onto the h-BN surface. We finally transferred a second h-BN flake covering just one of the two TiS₃ nanoribbons. Figure 3c shows an optical picture of the fabricated stack after heating it. We recorded the Raman spectrum of each nanoribbon before and after heating up the sample at 450 °C (a temperature much larger than the threshold for oxidation of 300 °C). Figure 3d shows the Raman spectra of the two nanoribbons before heating that display very similar features and are both characterized by the four TiS₃ peaks discussed above. Notice that compared to Figure 3a the Si peak is missing since the substrate is ITO. When comparing the spectra after heating at 450 °C for 30 min we observe a large difference between the two nanoribbons. While the fully encapsulated nanoribbon (2) does not show a significant change in its spectrum, indicating that the final material is TiS₃, the unencapsulated flake (1) shows a dramatic change in its spectrum, due to the conversion from TiS₃ to TiO₂. The encapsulation (with bottom and top h-BN) prevents the oxidation of the TiS₃ nanoribbons from happening. The effect of the h-BN layers are visible also in the optical picture of Figure 3c where a big difference in the aspect of the two nanoribbons (one being oxidized and the other not) can be seen. As a side note, the Raman spectrum of the oxidized uncovered nanoribbon shows additional peaks compared to Figure 3a, located at energies of 396 cm⁻¹, 518 cm⁻¹ and 635 cm⁻¹. These peaks, associated to the TiO₂ crystalline phase, are due to anatase TiO₂ in agreement with the results from the XRD measurements of Figure 2c.

3.2. Bandgap Energy Calculation

The experiments discussed above show that a thermal oxidation process can convert TiS₃ nanoribbons into TiO₂ and that this evolution can be followed on a single ribbon level with Raman spectroscopy. We now focus on the changes in the band-structure of the system when passing from TiS₃ to TiO₂.

To calculate the electronic band structure, we have performed state-of-the-art ab-initio Density Functional Theory (DFT) calculations with a pseudo-potential plane-wave method as implemented in the PWSCF code of the Quantum-ESPRESSO suite [37,38,39]. Figure 4 shows some examples of the electronic band structure calculated with DFT for the TiO₂ in the rutile, anatase, IV and V allotropes (see a 3D representation of the different structures in Figure 5a) over the first Brillouin zone. A GW calculation most of the time opens up the band gap with a rigid shift of the conduction bands.

For both Ti and S, the electron exchange-correlation potential is evaluated within the generalized gradient approximation throughout the Perdew–Burke–Ernzerhof’s functional. For S the Martins–Troulliers’, while for Ti the Goedecker–Hartwigsen–Hutter–Teter’s pseudo-potentials are used, including semi-core states for the valence electrons. In all structures we have optimized the atomic positions with a residual force after relaxation of 0.001 a.u. and have also included van der Waals corrections. The kinetic energy cut-off for the plane wave basis set is at 220 Ry, while the cut-off for the charge density is 880 Ry. The sampling of the Brillouin zone is 6x6x6 according to the Monkhorst–Pack scheme. The parameters chosen ensure a convergence of the DFT band gap within an accuracy of around 0.01 eV. In general, DFT underestimates the band gap, however those results might be used to estimates tendencies, like an increase or decrease in the band gap under oxidation. To enable a better comparison with the experimental values, we have further performed a more refined calculation for some of the structures based on non-self-consistent GW method. This opens up the DFT gap (0.37 eV) for pristine TiS₃ to the experimental levels (about 1.2 eV). The GW band gaps have been converged within an accuracy of around 0.05 eV.

In order to construct the intermediate oxidation structures, we have started with the relaxed structures of pristine TiS₃ and have replaced three S atoms (in a unit-cell of eight atoms) with two O atoms. We have relaxed possible oxidation states by varying the position of the atom replacements. The three structures in Figure 5a correspond to the ones of the lowest total energy, which are most likely to be formed during the oxidation process. Similar techniques have been applied to find the structures for the fully oxidized states. For these structures, we have also performed a GW calculation to evaluate the band gap. Both DFT and GW show an increase in the gap at the $\Gamma$-point of the band structures of interest for these experiments, ranging from about 1eV for TiS₃ to about 3 eV for the TiO₂ in the rutile structure. For intermediate oxidation states, the evaluation of the actual atomic configuration is more difficult since one should consider exceedingly large super cells to build the possible atomic configurations, but our results show clearly an increase in the DFT band gap from 0.31 eV for TiS₃ to about 0.7 eV for the partially oxides structures, to more than 2 eV for the TiO₂ in the different allotropic forms. These results are consistent with others already present in the literature for TiO₂.

Figure 5a shows the calculated bandgap of TiS₃ (left), intermediate Ti₂S₃O₂ phases (middle) and of four different polytypes of TiO₂ (right). The different materials are ordered along the horizontal axis according to the total energy (when going from left to right the total energy decreases and the thermodynamic stability increases). Because DFT typically underestimates the band gap energy [40], we have further performed a more refined calculation for some of the structures based on non-self-consistent GW method.

Both DFT and GW calculations predict that the bandgap energy increases when the content of sulfur decreases and the oxygen increases. The stable intermediate titanium oxysulfide phases predicted by the theory are consistent with FEGSEM measurements of partially oxidized nanoribbons. The left panel of Figure 5b shows a FEGSEM image of such a nanoribbon in which the core and the edges show a different contrast. The Energy-dispersive X-ray spectroscopy (EDX) analysis of the image performed in the center and at the edges of the nanoribbon reveals a higher oxygen concentration along the edges. This indicates that the oxidation process occurs through the formation of intermediate phases TiO_2-xS_x at the ribbon surface that form a sheath around TiS₃. This phenomenon is also visible in the microscope pictures of partially oxidized nanoribbons in Figure S2, Section S1 of the Supporting Information.

3.3. Electronic and Optoelectronic Characterization

After the change in the optical appearance and vibrational properties of TiS₃ nanoribbons after high temperature treatment due to the oxidation of TiS₃ to TiO₂ was established, we then investigated electronic transport through an individual nanoribbon and monitored the change of the optoelectronic properties. We fabricated a TiS₃ photodetector by transferring an individual TiS₃ nanoribbon onto two pre-patterned platinum electrodes separated by a distance of 20 µm. The electrical measurements were carried out in atmospheric conditions using a home-built probestation equipped with a source measure unit (Keithley 2450). For the optoelectronic measurements the devices were illuminated by focusing the light of different high-power fiber-coupled LED sources with different emission wavelengths (Thorlabs), forming a circular spot (400 µm in diameter) onto the surface of the sample. The total optical power reaching the sample was measured with a silicon photodetector (Thorlabs power meter PM100D with sensor S120VC).

Figure 6a shows a schematic of the device and an optical image of the pristine TiS₃ device. The TiS₃ nanoribbon bridges the two electrodes and light can be shined on the exposed TiS₃ channel to study its photoresponse. Figure 6b shows the current–voltage characteristics (I-Vs) of the device recorded just after the fabrication (top panel) and after heating it for 12 min at 320 °C (bottom panel). From previous measurements the electrical resistivity of the exfoliated TiS₃ thin nanoribbons is ~0.1 Ω·cm [15] while measurements on macroscopic TiS₃ whiskers, reported by Gorlova et al., reach 2 Ω·cm [41]. Photographs of the device at various stages of the evolution are shown in Figure S6, Section S3 of the Supporting Information. The I-Vs were recorded in dark conditions (black curve) and under global illumination at 405 nm with power density 1 W/cm² (purple curve). Comparing the I-Vs of the pristine with those of the oxidized device we can see a difference in the shape of both I-Vs and in the magnitude of the current. The TiS₃ device is characterized by linear I-Vs and current in the range of µA while the oxidized device has non-linear I-Vs with currents in the pA (approximately six orders of magnitude lower than the pristine device). In both cases the device responds to light, evidenced by the larger current observed in the I-Vs under illumination at 405 nm in comparison to the ones recorded in the dark. Considering that the area of the channel is 100 µm² and that the incident optical power is 1 W/cm², the responsivity of the device to 405 nm at 3 V (−3 V) is 8 A/W (11 A/W) in the pristine (TiS₃) case and 0. 16 mA/W (0. 22 mA/W) in the oxidized case. In total in this work we fabricated and characterized 11 devices that have been heated at 320 °C in air, see Section S3 of the Supporting Information. In four cases out of 11 we observed the conversion of the TiS₃ photodetector in a TiO₂ one without losing the functionality (success rate 36%).

In order to study the responsivity of the device to different wavelengths and incident optical powers we measure current–time traces (at fixed bias voltage) while switching ON and OFF in time with the incident light. Figure 6c shows various photocurrents versus time traces measured with wavelengths in the range between 375 nm and 660 nm. At the beginning of the measurement the light is switched OFF and the current passing through the device has only the dark current contribution. When switching ON the illumination (approximately at time = 5 s in the plot), the current passing through the device rapidly increases thanks to the additional contribution given by the photogenerated current. By extracting the current difference between the OFF and ON illumination, we can calculate the photocurrent and responsivity of our device. From the plot in Figure 6c one can see that the responsivity of the TiS₃ pristine device decreases when increasing the wavelength of the incident radiation. Figure 6d shows the photocurrent of the pristine device at 405 nm for different values of the illumination power density showing larger values for higher incident powers. Figure 6e,f shows similar measurements to those shown in Figure 6c,d made on the oxidized nanoribbon photodetector. These measurements show that after oxidation the photocurrent decreases and the time response increases. Moreover, the photodetector responds only to light with a wavelength shorter than 405 nm. The cut-off wavelength of the photodetector (defined as the largest wavelength for which the photodetector shows a response higher than the noise level 2 × 10⁻⁶ A/W), is larger than 660 nm in the pristine case and blue-shifts to 405 nm in the oxidized state. As a control experiment we heated one device in high vacuum conditions (pressure ∼10⁻⁵ mbar) to test the effect of heat in the absence of oxygen, see Section S2 of the Supporting Information. This device maintained the photodetecting properties of TiS₃ (responsivity spectrum and cut-off wavelength) even after 30 min at 320 °C.

From current–time traces similar to the ones in Figure 6c–f we extracted the responsivity of the device, in its pristine state and after subsequent heating cycles, as a function of wavelength and incident power. Figure 7a shows the evolution of the wavelength-resolved responsivity of the device measured at an incident power of 100 nW during the heating process. The pristine device has a responsivity of approximately 1 A/W that after the first heating cycle decreases to 10⁻⁴ A/W. The dependence on the wavelength also gets modified by the heating process. While in the first three traces the device responds to all the probed wavelengths from 375 nm to 660 nm, in the last four traces the responsivity shows an abrupt decrease for wavelengths larger than 450 nm, indicating that there is a blue-shift of the cut-off wavelength of the photodetector. The wavelength-resolved responsivity reported in the initial and final stages of the device are compatible with previously reported spectra of TiS₃ and TiO₂ photodetectors [17,42].

Figure 7b shows the evolution of the responsivity at 405 nm as a function of the incident optical power. In a photodetector, the responsivity at a certain wavelength $R\left(\lambda \right)$ versus the incident optical power $P$ can be described by a power law according to the formula:

$$R\left(\lambda \right)=P^{\alpha }$$

(1)

where $\alpha$ is a dimensionless exponent, which assumes a value of −0.5 in the case of a photodetector dominated by bimolecular recombination between photoexcited carriers or at high injection levels (band-to-band) and 0 in the case of monomolecular recombination or low injection levels (trap-assisted) [43,44]. The log–log plot shown in Figure 6b reveals that our device is characterized by a negative $\alpha$ for all the heating cycles since the slope of each curve in the figure gives directly the exponent $\alpha$. Figure 7c shows the extracted cut-off wavelength and the exponent $\alpha$ of the device as a function of the heating cycle. The evolution of the cut-off wavelength (here defined again as the largest wavelength for which the photodetector shows a response higher than the noise level 2 × 10⁻⁶ A/W) is consistent with the predicted bandgap evolution shown in Figure 4a. As can be seen the exponent $\alpha$ starts from a value of −0.42 which indicates that the pristine TiS₃ device is mostly dominated by band-to-band recombination (0 heating cycle). After the first heating cycle the exponent decrease to a value of −0.7 that is the minimum value observed for $\alpha$. In the subsequent heating cycles (2–6) we observe a gradual increase in the value of the exponent $\alpha$ toward 0, signifying an increase in the density of traps and/or an effect on the lower effective light injection levels due to the lower absorption in the visible range of the spectrum by the TiO₂ (see Section S1 of the Supporting Information). The exponent $\alpha$ saturates around −0.25, a value larger than the starting value of −0.4, which indicates that the final TiO₂ material contains a larger trap density than the starting TiS₃ material that favors monomolecular recombination between photoexcited carriers [45,46].

4. Conclusions

In conclusion, we studied the thermal oxidation of TiS₃ nanoribbons with optical spectroscopy showing that this material can be gradually converted to anatase TiO₂ in a controlled way. We built photodetectors based on single TiS₃ nanoribbons and we tuned their cut-off wavelength by gradually oxidizing the nanoribbons. We observe that the oxidation also induces a decrease in the responsivity, which could have practical implications for the applicability of the TiO_2-xS_x, as a larger detector area would be required to obtain the same photocurrent. Ab-initio calculations of the band-structure of the materials are in agreement with the experiments. The shift of the cut-off wavelength in our photodetectors with a simple annealing step, demonstrates the ability to tune on-demand the bandgap of the TiO_2-xS_x for novel applications.