A real-time Raman spectroscopy study of the dynamics of laser-thinning of MoS 2 flakes to monolayers

Transition metal dichalcogenides (TMDCs) in monolayer form have attracted a great deal of attention for electronic and optical applications. Compared to mechanical exfoliation and chemical synthesis, laser thinning is a novel and unique “on-demand” approach to fabricate monolayers or pattern desired shapes with high controllability and reproducibility. Its successful demonstration motivates a further exploration of the dynamic behaviour of this local thinning process. Here, we present an in-situ study of void formation by laser irradiation with the assistance of temporal Raman evolution. In the analysis of time-dependent Raman intensity, an empirical formula relating void size to laser power and exposure time is established. Void in thinner MoS2 flakes grows faster than in thicker ones as a result of reduced sublimation temperature in the two-dimensional (2D) materials. Our study provides useful insights into the laser-thinning dynamics of 2D TMDCs and guidelines for an effective control over the void for...


INTRODUCTION
Layered two-dimensional (2D) materials have gained interest for their unique properties and atomic-scale thin film structure.This family of materials has lately seen a rapid growth in number of family members, now extending far beyond graphene to also include black phosphorus, metal chalcogenides, metal carbides and transition metal oxides. 1,2Transition metal dichalcogenides (TMDCs) alone can occur in more than 40 different categories and span the entire range of electronic structures. 1,3Those that are semiconducting characterised by excellent electrical properties especially when they are thinned to monolayer thickness, are considered to be the most promising channel material for thin film transistors. 1,3,4As an example, bulk MoS 2 has an indirect band gap of 1.2∼1.3eV.][6][7] Layered MoS 2 has a broad range of potential applications in gas sensing, 8,9 energy conversion (solar cell) 10 and energy storage (ion battery). 11,12However, most of the aforementioned unique properties of 2D materials only persist when the number of repeating layers is reduced to one or close to one.Effective fabrication routes for production of monolayers of 2D materials have motivated great efforts in the 2D-material community.
Several methods have been explored for producing ultrathin 2D materials with a suitable number of layers, such as mechanical exfoliation, 13,14 liquid-phase exfoliation, 15,16 chemical vapour deposition (CVD), 17 molecular beam epitaxy, 18 electrochemical thinning 19,20 and laser thinning.Mechanical exfoliation, which can produce layers of the highest quality, is a random process with low yield.Liquid-phase exfoliation, though possible to economically produce large volumes of dispersed layers, can unavoidably introduce impurities and residual solvent and yields a low fraction of monolayers.CVD is suitable for growth of large-area thin layers, but it is a high temperature process thereby limiting the choice of growth substrates.Laser thinning is a relatively new technique for "on-demand" fabrication of monolayers and patterning of desired shapes with high controllability and reproducibility. 3,22,25This technique involves temporary illumination of a focused laser beam onto thin flakes without needing complex environment control.It presents thus an efficient and easy top-down method to thin and fabricate certain patterns via a convenient design and at a controllable laser power level.The accumulation of heat induced by laser irradiation can easily lead to sublimation of the upper few layers of the flake while the bottom-most monolayer can remain intact due to efficient heat dissipation to the substrate when the laser power, P 0 , is controlled below a certain level, e.g. 10 -20 mW for MoS 2 thinning. 22Additionally, the use of laser irradiation to thin or pattern some features in 2D flakes can avoid involving photoresist or mask, thus eliminating a source of organic contamination or change of electronic structure.The monolayer MoS 2 produced by laser thinning has shown a comparable mobility to that of pristine one. 9,22,26lthough the laser thinning technique has shown several advantages including rapid material processing, large scan area, high spatial resolution and single-step capability, its dynamic process is less understood.In the present work, confocal Raman spectroscopy is used both as the thinning laser source and as the light source for in-situ real-time monitoring of the laser thinning process.This arrangement facilitates rapid characterisation of the same area before, during and after thinning, which allows for a systematic investigation of the dynamics of the laser-induced void formation as a function of P 0 and exposure time, t.An empirical formula to describe the relationship between void size and these two laser parameters is obtained.

METHODS
MoS 2 flakes from natural crystalline MoS 2 (SPI supplies) were deposited onto a SiO 2 /Si substrate by mechanical exfoliation.Prior to exfoliation, the Si substrate was pre-cleaned using sonication in acetone, isopropyl alcohol and deionized water.The number of MoS 2 layers on the substrate was identified using an optical microscope (Keyence digital microscope VHX-600), atomic force microscopy (AFM, Dimension 3100, Veeco) and Raman spectroscopy (Renishaw inVia).Raman spectra of the MoS 2 flakes were acquired in a backscattering configuration with a 514 nm laser.After careful location and focusing of the laser spot on each specimen, the Raman scattered signal was collected with a 100× objective (NA = 0.85) and dispersed by a grating of 1800 lines/mm.The spectral resolution was ∼ 1.0 cm -1 and peak position accuracy was 0.1 cm -1 .To identify the number of layers, Raman spectra were obtained with P 0 below 1 mW and an exposure time of 10 s.However, for laser thinning process, P 0 was increased up to 20 mW and the exposure time varied from 1 to 100 s.P 0 was directly measured using a laser power meter (Sanwa Laser Power Meter LP1 Mobiken Series) and carefully adjusted prior to each laser irradiation.

RESULTS AND DISCUSSION
The samples used in our laser thinning experiments were few-layer MoS 2 flakes on SiO 2 /Si substrate exfoliated from a bulk MoS 2 crystal using a mechanical exfoliation technique.The location of the flakes was identified using optical microscope.The optical image of several MoS 2 flakes is shown in Fig. 1(a).The optical contrast of MoS 2 is found to increase with the number of layers.The layer number before thinning was determined by the frequency shift of specific vibrational modes using Raman spectroscopy with a low-power laser (<1 mW) in order to avoid heating effect.Fig. 1(b) presents two prominent Raman peaks E 2g 1 (in-plane vibration mode) and A 1g (out-of-plane vibration mode) of the MoS 2 flakes.It is known that the Raman shift difference between these two peaks increases with the number of layers, 27 as demonstrated in Fig. 1(c).This same Raman setup was also used for the MoS 2 thinning experiments, but with a higher incident P 0 ranging from 10 mW to 20 mW.The Si substrate is an efficient heat sink.Consequently, only the bottom monolayer directly in contact with the substrate can be preserved without being ablated by localised laser heating, while the upper layers are sublimed due to poor interlayer thermal coupling.The nominal temperature for sublimation of a natural MoS 2 crystal is 450 o C. 28 Under laser irradiation at 10 mW or above, the lattice temperature of the irradiated part can exceed the sublimation temperature within a few seconds. 29This transient laser heating is sufficient to break the weak interlayer van der Waals bonds and then sublimes the upper layers readily.At an appropriate P 0 , the bottom monolayer is preserved regardless of the initial number of layers.Fig. 1(d) shows the Raman spectra of thinned few-layer MoS 2 flakes after laser irradiation.The frequency difference, also depicted in Fig. 1(c), is consistently reduced to ∼ 21 cm -1 from those of pristine MoS 2 flakes of different number of layers.This frequency difference is slightly higher than that of a pristine MoS 2 monolayer and can arise from shifts of Raman modes due to the presence of traces of unremoved MoS 2 .
Laser thinning of the exfoliated MoS 2 flakes to a single monolayer has been successfully demonstrated, but its dynamics is so far unexplored.The Raman microscopy performed in situ presents an efficient method for real-time monitoring of the laser thinning process.The temporal behaviour of Raman intensity is achieved when Raman spectra are collected continuously during laser thinning.When a high-power laser is focused on the few-layer MoS 2 flakes, a small void is formed quickly with the bottom monolayer remaining intact.Fig. 2(a) is an AFM image of such a void formed in a 7-layer (7 L) MoS 2 flake.About 3.6-nm-thick MoS 2 , corresponding to 6 L, is removed altogether while the bottom monolayer stays.The particles around the void rim are likely to originate from the atmosphere due to laser-induced heating effect. 29The void is then found to laterally grow in size with t.As a result of the upper layers being etched laterally, the integrated E 2g 1 peak intensity decreases with t.Fig. 2(b) exhibits an initial rapid decay of the peak intensity followed by a slow decline.However, it is important to note that the E 2g 1 peak did not disappear completely over the duration of the experiment.Rather, it decreased to and then levelled off at a small value, indicative of thinning of the MoS 2 flake to the monolayer thickness that is resistive to further decomposition.To gain additional insights into the decay dynamics, theoretical modelling was pursued by starting from this time-dependent Raman intensity.Here, the intensity distribution of the focused laser, p(r), is assumed to be Gaussian: where r the radial distance from the laser spot centre and r 0 the laser spot size taken to be the full width at half maximum intensity given by r 0 = σ 0 (2ln2) 1/2 .To estimate the laser spot size, we used the acquired Raman intensity obtained by scanning a low-power laser spot from the SiO 2 surface over to the edge of MoS 2 flake.This intensity can be regarded as the convolution of the laser intensity profile and the topography of the MoS 2 flake.In Fig. 3(a, b), the optical image of a MoS 2 flake with a sharp edge and the relative Raman intensity along a linecut are, respectively, depicted.By applying a numerical integration based on Eq. ( 1) with the best fit to the Raman data, we obtained a laser spot size of ∼ 0.2 µm, which is in good agreement with a diffraction limited spot radius (∼ 0.3 µm).
During the laser-thinning process, a void is expected to grow in size with t once it is formed.Since the bottom-most monolayer in the void is persistently well preserved under laser irradiation at the laser-power range used here, the Raman intensity recorded during the in-situ analysis should include contributions from the bottom-most monolayer as well as the unremoved few-layers, as illustrated in Fig. 4(a).According to Eq. ( 1), the time-dependent Raman intensity, I(t), can be written as: where r H is the radius of a laser-etched void and K proportionality coefficients reflecting the Raman intensity excited by a unit P 0 .The subscripts 0 and 1 refer to monolayer and few-layer MoS 2 , respectively.In order to obtain the proportionality coefficient, exfoliated MoS 2 flakes of different numbers of layers were laser illuminated at P 0 below 1 mW.Fig. 4(b) shows the variation of integrated E 2g 1 Raman intensity with P 0 .A linear fit to the data points yields K for the various few layer films, as shown in Fig. 4(c).The parabolic-like behaviour of K with number of layers is a result of multireflection and interference of the light-beams reflected from the MoS 2 and the Si/SiO 2 interface, 27 which is also observed in graphene on SiO 2 /Si substrate. 30Fig. 4(c) also shows the K value of those thinned MoS 2 under a high P 0 of 15 mW.It is found that the K value after laser thinning is close to that of pristine monolayers, which confirms that few-layer MoS 2 are successfully reduced to monolayers by laser irradiation at the P 0 range used here.
Starting from Eq. ( 1) and (2), the power density at the void edge, p(r H ), can be correlated to I(t), as the following: In Fig. 5, p(r H ) is plotted as a function of t.In the logarithmic scale, p(r H ) is found to first decrease linearly with t and then flattens out when t further increases and the void radius approaches its maximum value.This behaviour is independent of the P 0 used and it holds true for all the few-layer cases studied here.Hence, p(r H ) is only determined by t but not by P 0 , which further confirms the validity of Eq. ( 1).According to Eq. ( 3), higher P 0 induces larger void, keeping the resultant p(r H ) constant.Taking 4 L MoS 2 as an example, Fig. 6(a) illustrates the calculated p(r H ) using Eq. ( 1) as a function of void radius at 15-mW laser irradiation.In combination with Fig. 5(c), each p(r H ) corresponds to a definite t and the thus-obtained void radius.For example, an irradiation for 5 s, 20 s and 50 s leads to a void radius of 0.12 µm, 0.17 µm and 0.20 µm, respectively.P 0 of 13 mW, 15 mW and 17 mW induces a void radius of 0.13 µm, 0.17 µm and 0.20 µm, respectively.From Fig. 5, an empirical formula correlating p(r H ) to t can be obtained from the linear fitting in the logarithmic scale, as shown in the following: Here, C and D are two empirical parameters and their variations with number of layers are given in Fig. 7.It can be seen that D is smallest for bilayer.It then increases with the number of layers.The variation of C is, however, much less pronounced.From the empirical formula in Eq. ( 4), it can be deduced that for the same p(r H ), parameter D is a major factor that determines t that is also the void growth time.Substituting Eq. ( 4) into Eq.( 1), we can obtain a relationship between r H , P 0 and t as follows: Hence, the void radius can be readily obtained with known P 0 and t.Herein, the smallest D value in a bilayer MoS 2 is equivalent to the quickest void expansion or easiest decomposition of  the flake.Thinner layers are generally characterised by higher surface area to volume ratios that are essential for their chemical and thermal stability.2][33] Thinner layers also have higher oxidization rates and lower sublimation temperatures, which explains why bilayer MoS 2 decomposes more quickly than its thicker counterparts.However, the strong thermal-coupling interaction of monolayer MoS 2 with the underlying substrate results in a relatively low temperature rise during laser irradiation.It is, therefore, immune to decomposition.The calculated r H for the different MoS 2 fewlayers is plotted in Fig. 8 as a function of t.As expected, r H initially increases rapidly for about 10 s.It then gradually grows up to the approximate laser spot size.In some cases, the presence of adsorbed particles around the void makes it difficult to ascertain the void edge or topography by AFM measurement, as shown in Fig. 9(a).In comparison, scanning electron microscopy (SEM) can clearly reveal the void profile and, hence, its radius can be reliably extracted in Fig. 9(b).
It can be seen that the experimentally determined r H is consistent with the theoretical expectation in Fig. 8.

CONCLUSIONS
In summary, a confocal Raman spectroscopy has been employed to induce void formation in MoS 2 flakes of different numbers of layers, to in-situ characterise the void formation and to study the dynamics of the laser-thinning process as a function of laser power and exposure time.Raman analysis reveals that the void size is affected by these two laser parameters and eventually approaches the laser spot size at long exposure duration.In the analysis of time-dependent Raman intensity, an empirical formula correlating the void radius to the laser power and exposure time has been developed.The calculation results are found to agree well with experimental data obtained by means of SEM analysis.Due to the higher surface area to volume ratio, thinner MoS 2 flakes especially in bilayer form, are less stable and easier to decompose under a high-power laser irradiation, leading to faster void expansion.Our finding provides an in-depth understanding of the laser-thinning process, which is essential to exploit for future fabrication of monolayer or nanostructure in other 2D TMDCs.

125329- 3 Gu
FIG. 1.(a) Optical image of exfoliated MoS 2 flakes on SiO 2 /Si substrate.(b) Raman spectra of pristine MoS 2 flakes of 1 L to 7 L in thickness.(c) Variation of the frequency difference between the A 1g and E 2g 1 phonon modes with the number of layers.(d) Raman spectra of MoS 2 monolayers laser-thinned from pristine multilayers.

125329- 4 Gu
FIG. 2. (a) AFM topographic image of a void formed in a 7 L MoS 2 flake by laser irradiation at 15 mW for 50-s.(b) Temporal evolution of the integrated E 2g 1 peak intensity upon laser irradiation.

125329- 5 Gu
FIG. 4. (a) Illustration of the distribution of P 0 density and the thinned monolayer embedded in a few-layer MoS 2 flake.Raman signals obtained in-situ include contributions from both MoS 2 monolayer and multilayer.(b) Integrated E 2g 1 intensity as a function of P 0 for pristine MoS 2 with different numbers of layers.Lines are linear fits to experimental data points.(c) Dependence of K value on number of layers.Red circles represent K values of monolayer laser-thinned from pristine multilayers.

125329- 6 Gu
FIG. 5. (a-f) Dependence of p(r H ) on t for 2 L -7 L MoS 2 flakes at different P 0 .Solid lines are linear fits in the logarithmic scale.

Fig. 6 (
FIG. 6. (a, b) Dependence of p(r H ) on r H obtained from Eq. (1).In conjunction with Fig.5, p(r H ), t and r H are all correlated.Taking 4 L as an example, i.e.Fig.5(c), irradiations for 5 s, 20 s and 50 s lead to r H equal to 0.12 µm, 0.17 µm and 0.20 µm, respectively, as highlighted with illustrations and red arrows shown in (a).For t=20 s, P 0 of 13 mW, 15 mW and 17 mW induces r H of 0.13 µm, 0.17 µm and 0.20 µm, respectively, as highlighted with similar illustrations and red arrows shown in (b).

FIG. 7 .
FIG. 7. Variation of parameters C and D in Eq. (4) with number of layers.In contrast to C, D is remarkably layer dependent.

FIG. 8 .
FIG. 8. (a-d) r H as a function of t for 2 to 5 L MoS 2 flakes.Symbols are experimental data extracted from SEM measurements.

125329- 8 Gu
FIG. 9. (a) AFM image of an etched void covered with particles.(b) SEM image of the same void in (a), where the etched profile is clearly visible.