Femtosecond spectroscopy on MoS2 flakes from liquid exfoliation: surfactant independent exciton dynamics

Abstract. Ionic surfactants, which are widely used to stabilize nanomaterials in dispersions, can drastically alter the nanomaterial’s photophysical properties. Here, we use femtosecond optical spectroscopy to study the dynamics of excitons and charges in few-layer flakes of the two-dimensional semiconductor MoS2. We compare samples obtained via exfoliation in water with different amounts of adsorbed sodium cholate, obtained by repeated washing of the dried flakes. We find that the femtosecond dynamics is remarkably stable against the surfactant adsorption, with a slight increase of the initial exciton quenching occurring during the first few picoseconds as the only appreciable effect.

Femtosecond spectroscopy on MoS 2 flakes from liquid exfoliation: surfactant independent exciton dynamics 1 Introduction In the last five years, most of the know-how acquired by the carbon community and culminating in the discovery and exploitation of graphene 1 has been transferred to isolate and engineer transition metal dichalcogenide (TMD) layered materials. Semiconducting TMDs, such as MoS 2 , WS 2 , MoSe 2 , and WSe 2 , can also be exfoliated into two-dimensional (2-D) layers of single unit cell thickness, suitable for logic devices and optoelectronics applications. [2][3][4] A peculiarity of the semiconducting TMDs is the crossover from indirect to direct gap semiconductor when going from bulk or few-layer material to monolayers. 5 This is a result of the gradual change of the band structure as a function of the number of layers due to quantum confinement. 6 Liquid exfoliation has been proven to produce stable dispersions of flakes of 2-D TMDs via a procedure that involves ultrasonication and centrifugation. However, processing nanomaterials in liquid dispersion can crucially alter their optical response and their electronic properties. 7 For example, during continuous wave excitation of single wall carbon nanotubes (SWNTs) covered with sodium cholate, chemisorption of oxygen leads to hole doping and photoluminescence quenching. In low pH solution, protonation of the oxygen absorbed on an SWNT with a sodium dodecyl sulfate shell bleaches the optical absorption and consequently quenches the luminescence by Auger recombination. [8][9][10] Moreover, it localizes electrons on the SWNT and increases the number of holes in the valence band (VB). 11 The surfactant surrounding SWNT can induce photoluminescence blinking via trapping photoexcited charges followed by a slow recombination. This process also changes the emission spectrum via Stark effect. 12,13 Femtosecond pump-probe spectroscopy represents a versatile technique to study the electron dynamics in semiconducting nanomaterials. It has elucidated, for instance, the dynamics of charge carriers, excitons, and phonons 14,15 or the Stark effect induced by photogenerated and trapped charge carriers. 16 Here, we use femtosecond pump-probe spectroscopy to study fewlayer MoS 2 from liquid exfoliation in water and sodium cholate 17 with different amounts of adsorbed surfactant controlled by a multistep washing protocol. We exploit the sensitivity of the pump-probe signal to the change of the electronic excited state population 18 in order to investigate the effect of the surfactant on the carrier and exciton dynamics of MoS 2 . For different amounts of surfactant, we obtain the same shape and relaxation behavior of the pump-probe spectra. This work demonstrates the optical stability of MoS 2 against surfactant adsorption, which sets it apart from other semiconducting nanomaterials.
2 Experimental Details

Sample Preparation
An initial solution of 4.5 g sodium cholate and 15 g of MoS 2 in powder form (Sigma Aldrich) dissolved in 1 l of deionized (DI) water was ultrasonicated for 100 h, using an Elmasonic P ultrasonic bath at 37 kHz and maximum power. This solution, which contains the flakes of MoS 2 , was used as a starting point for producing all the samples. 4 In order to obtain films of few-layer MoS 2 flakes with high surfactant content (SC), 10 ml of ethanol was added to 10 ml of the main solution, which was then spray-deposited on a quartz substrate (kept at 150°C). In parallel, different washing steps have been done to produce samples with medium and low SC.
MoS 2 with medium SC was obtained by centrifugation of a part of the initial solution in a Cole Parmer Power Spin BX centrifuge for 60 min at 11,000 rpm, followed by the addition of clean DI water and ultrasonication in a bath. This procedure was repeated three times to remove a part of the surfactant from the solution and from the MoS 2 surface. At a later stage, ethanol was added as described above, and the mixture was sprayed on quartz. Figure 1(a) shows a transmission electron microscope image of a typical few-layer MoS 2 flake obtained by this process. To further reduce the concentration of sodium cholate on the surface of MoS 2 , we introduced an additional step in the procedure for the medium SC samples: after each centrifugation, the solution was washed with DI water and HCl (0.1 ml of 1 M HCl per 50 ml of MoS 2 dispersion). HCl reacts with sodium cholate and forms sodium chloride and cholic acid, which is soluble in ethanol. This step is expected to remove most of the sodium cholate from the flake surface. Such obtained low SC dispersion is again deposited on a quartz substrate.
As a reference sample, we used MoS 2 obtained in crystalline form (Wolfram Camp Mining Ltd.), expanded in hydrazine 19 and exfoliated, without surfactant, in the organic solvent Nmethyl-2-pyrrolidone by ultrasonication. The films were deposited on quartz substrates by a Langmuir-Blodgett process. 20

Pump-Probe Spectroscopy
The laser source was a commercial regeneratively amplified mode-locked Ti:sapphire laser (Coherent Libra), yielding pulses at 1 kHz repetition rate, with 800 nm center wavelength and 100 fs duration. The pump pulses were generated by frequency doubling the fundamental in a thin β barium borate crystal. The pump energy was 140 nJ energy per pulse, with 150 μm spot size. Broadband white-light continuum pulses generated in a sapphire plate, covering the 420 to 750 nm wavelength range, were used as probe pulses. The pump and probe beams were polarized parallel to each other. After passing through the sample, the probe beam was focused onto the entrance slit of a spectrometer equipped with a 1024-pixel back-thinned CCD sensor S7030-1006 and electronics specially designed for fast read-out times and low noise. 21 By recording the transmitted probe spectrum with (T on ) and without (T off ) the pump as a function of wavelength λ and pump-probe delay τ, we obtain the differential transmission (ΔT∕T) spectrum as E Q -T A R G E T ; t e m p : i n t r a l i n k -; e 0 0 1 ; 1 1 6 ; 3 9 8 ΔT∕Tðλ; τÞ ¼ ½T on ðλ; τÞ − T off ðλ; τÞ∕T off ðλ; τÞ: (1) The delay between pump and probe pulse was varied by moving a translation stage, which changes the length of the optical path of one beam with respect to the other.

Absorption Spectroscopy
The optical absorption spectra were recorded in transmission using a V-570 Jasco spectrometer with 0.5 nm spectral resolution, equipped with a single monochromator covering the spectral range from 190 to 2500 nm (6.52 to 0.49 eV).

Raman Spectroscopy
Raman spectroscopy of MoS 2 was performed with an NT-MDT NTEGRA SPECTRA confocal Raman microscope in backscattering geometry with spectral resolution of 0.7 cm −1 . We used the excitation line at 488 nm and a 20× objective (N.A. 0.4) to focus onto a spot size of 20 μm. The Raman signals were detected with a CCD array at −80°C. We used a laser power of 5 mW.

Results and Discussion
The normalized absorption spectra in Fig. 1(b) show the characteristic exciton resonances A to D at 1.8, 2.0, 2.8, and 3.0 eV. The resonances A and B are ascribed to excitons associated with the direct gap transition at the K point. 22,23 The splitting originates primarily from the splitting of the VB due to spin-orbit coupling. The C exciton is ascribed either to a transition from a band below the VB toward the conduction band (CB) or to a transition from the VB into a nested region 24 of the CB. The D exciton is commonly ascribed to impurities. 22 Recently, it has been shown that for surfactant dispersions of MoS 2 , the characteristics of the absorption spectrum provide a metric for the average flake size and thickness. 25 Since it has been formally proven only for aqueous dispersion, we do not extract any quantitative characterization from our absorption spectra; however, the general trends should be valid also for dried films. Specifically, the ratio between the B exciton peak and the absorption at 3.6 eV is a measure of the typical flake size, and the position of the A exciton peak is a measure of the typical flake thickness (number of layers). The samples differ very little in both respects, the most significant difference being that the reference sample, probably due to the different exfoliation protocol, seems to consist of somewhat smaller flakes. Repeated washing, on the other hand, does not appreciably affect the flake size and thickness distribution.
To evaluate the homogeneity of the films on the relevant scale of 150 microns (the spot size used for the pump-probe experiment), we show optical images of the deposited MoS 2 [Figs. 1(c)-1(f)]. The darker hue of green seen for the medium SC film, matched by a higher optical density and stronger Raman and pump-probe signals (see below), is due to a higher film thickness. For the used high concentration, our simple spraying technique gives little control over the spraying speed and hence the film thickness. A significant inhomogeneity is seen only for the high SC sample. The high surfactant concentration may favor lateral aggregation of flakes or have an influence on the drying dynamics of the film. However, the apparent islands and voids are small enough to mostly average out over the spot size of the pump-probe experiment.
To find out whether the inhomogeneity in the film thickness is also related to an inhomogeneity in the flake characteristics, we measured Raman spectra (depicted in Fig. 2) on each sample on four randomly selected spots with a distance of several millimeters and a spot size of 20 microns, much smaller than the characteristic length scale of the inhomogeneity. The distance between the peaks E 1 2g (in-plane mode) and A 1g (out-of-plane mode) is a robust measure of the flake thickness. 26 We obtain Δω ¼ 24.5 cm −1 , corresponding to an average flake thickness of four layers for the high SC sample, and Δω ¼ 25 cm −1 , corresponding to five or more layers for all the others. Given the similar electronic structure of the flakes within the same film, we can use the Raman intensity as an indication of the amount of material present. 27 The thicker sample (medium SC) shows a similar Raman intensity at all spots, while for the others, the intensity varies strongly. This confirms the observation in the optical micrographs that on a length scale of 20 microns, the thicker sample is quite homogeneous, while the others are not. However, the flake thickness and hence the electronic structure is more homogeneous, and over the spot size of the pump-probe experiment (150 microns), the inhomogeneity averages out.
The ΔT∕T spectra of the reference sample are shown in Fig. 3(a) for different pump-probe delays. As previously observed on MoS 2 , there are six salient features in the visible spectral range: 28 three photobleaching peaks of the exciton absorption resonances A 0 , B 0 , and C 0 , and three photoinduced absorption peaks A − , B − , and C − ascribed to photogenerated charges,  one for each of the excitons. The exciton bleaching features show a fast decay during the first few picoseconds, followed by a slower decay on the few-100 ps scale. The photoinduced absorption peaks grow during the first few picoseconds and then decay with a characteristic time similar to the slow exciton decay. Charge photogeneration occurs via a combination of a direct excitation mechanism (i.e., within our instrumental resolution of ∼100 fs), and via dissociation of hot excitons, on a time scale of a few hundred femtoseconds to few picoseconds. 28 This process can be seen in Fig. 3(a) as a fast decrease of the bleaching features A 0 , B 0 , and C 0 and a corresponding growth of the charge features A − , B − , and C − during the first few picoseconds. Together with the spectral change of shape-relative increase of the absorption and decrease of the bleaching features-the peaks also undergo a blueshift with increasing delay. In a more comprehensive study with different excitation fluences, it was found that the more the excited states the more the spectra are shifted to the red. 28 This has been explained as Stark effect due to photoexcited charges, 16,29 inter-excitonic interaction, 30 or band gap renormalization. 31 Additionally, at short delays, the A − and C − photoinduced absorption peaks appear artificially redshifted due to the strong overlap with the A 0 and C 0 bleaching peaks, respectively. The fast decay of the bleaching peaks diminishes this effect with increasing pump-probe delay, resulting in an additional apparent temporal blueshift of these two photoinduced absorption features. After ∼10 ps, there is no significant shift of the spectra anymore and all features decay with a time constant of several hundred picoseconds, as has already been observed 14   one finds a trend toward slightly faster decay for higher SC. Additionally, for higher SC, the C exciton bleaching is weaker. The trend toward faster initial exciton decay for higher SC observed in the ΔT∕T spectra is confirmed by the dynamics at representative probe energies, close to the respective spectral peaks depicted in Figs. 4(a)-4(f). The timescale of the delayed charge formation [Figs. 4(a), 4(c), and 4(e)] corresponds to that of the initial exciton decay [Figs. 4(b), 4(d), and 4(f)], confirming the origin of these charges from the dissociation of excitons. Both the slow (few hundred picoseconds) decay component as well as the direct and delayed charge photogeneration yields do not vary appreciably between the different samples.
In conclusion, we have shown that the exciton and charge dynamics in few-layer MoS 2 flakes is robust against adsorption of the anionic surfactant sodium cholate. This is in stark contrast to other nanomaterials such as carbon nanotubes 32 or C 60 fullerenes, 33 where anionic surfactants generally lead to strong exciton quenching. This robustness highlights the viability of surfactant-assisted liquid processing (exfoliation, deposition, printing, etc) of MoS 2 and other semiconducting TMDs.