Low-dimensional nanomaterial saturable absorbers for ultrashort-pulsed waveguide lasers

A wide range of saturable absorbers composed of novel low-dimensional nanomaterials were fabricated, and their linear and nonlinear optical properties were characterized. Furthermore, their suitability for ultrashort-pulse generation in waveguide laser operating at a wavelength of 2 microns was demonstrated and passively q-switched modelocked operation was achieved with all absorbers. The material systems that were studied in this work include nanosheet-based absorbers composed of graphene, carbon nanotubes, black phosphorus, transition-metal dichalcogenides, topological insulators and indium tin oxide. By utilizing a uniform few-layer spin coating fabrication technique and by employing a single, identical laser resonator, a direct comparison of the individual characteristics of these materials in the context of short-pulse generation in waveguide lasers was made possible. Each of the individually fabricated and characterized saturable absorbers was placed inside a thulium-doped fluoride glass waveguide chip laser cavity and the resulting output performance was analyzed and contrasted. It was further found that the few-layer spin coating approach enables fine-tuning of the absorber characteristics and that all low-dimensional nanomaterials under investigation can be utilized for ultrashort pulse generation in the 2micron wavelength range. General guidelines for the design of passively modulated shortpulsed laser oscillators are presented based on those findings. © 2018 Optical Society of America under the terms of the OSA Open Access Publishing Agreement


Introduction
Ultrashort laser pulses with correspondingly high peak power levels have found a plethora of applications in fields as diverse as spectroscopy, biomedical research and telecommunications, to only name a few.In particular laser sources that emit at a wavelength that overlaps with one of the strong absorption peaks of water close to 2 and 3 μm are of great interest for medical procedures where the precise and minimally invasive cutting of tissue is crucial, e.g. in eye surgery.The standard technique that is used to transform a free-running continuous-wave (cw) laser source into a short-pulsed one is to insert a saturable absorber (SA) into the cavity which is essentially a nonlinear gating device that features high optical losses for low intensity intra-cavity light and vice versa.Thus, by fine-tuning the nonlinear optical parameters of the SA in a specific laser resonator, passively q-switched, q-switched mode-locked (QML) or cw mode-locked operation can be realized as a consequence of the dynamic interaction between the intensity-depended loss of the SA and the intensity-depended gain of the active laser medium [1].Semiconductor saturable absorber mirrors (SESAMs) have been in widespread use since the late 1990s [2].However, SESAMs are limited in bandwidth due to their semiconductor-based energy structure and require complex and expensive cleanroom equipment for their fabrication.More recently, it has been demonstrated that low-dimensional materials like graphene [3] and carbon nanotubes (CNTs) [4] exhibit saturable optical absorption and thus, under the right conditions, can offer similar performance to SESAMs, yet without the above-mentioned limitations [5].As a consequence, a large variety of different low dimensional materials with distinct physical and chemical properties have become the subject of scientific investigation in the area of ultrafast photonics, resulting in a vast number of publications that report on the generation of ultrashort-pulsed radiation using various novel low-dimensional materials based saturable absorbers with operating wavelengths that span from the visible to the mid-infrared, employing laser geometries from fibre and bulk lasers to miniature waveguide chip lasers (see e.g [6,7].).However, this wide range of possible implementations of low-dimensional saturable absorber-based passively modulated laser systems in combination with variations in the fabrication process of the saturable absorbers themselves makes it virtually impossible to analyze and to contrast and compare the performance and suitability of all the individual materials investigated.Due to this reason, Li et al. compared the performance of graphene, MoS 2 and Bi 2 Se 3 as saturable absorbers in a single Nd:YVO 4 crystal waveguide laser platform very recently [8].
In this paper, a systematic experimental study that includes nine of the most commonly utilized low-dimensional nanomaterial saturable absorbers for ultrashort-pulsed lasers is presented.The results presented here allow for a direct comparison between a wide range of different low-dimensional materials when used in a single laser resonator, operated at a single laser wavelength.Saturable absorbers based on graphene, carbon nanotubes, transition metal dichalcogenides (TMDCs), topological insulators (TIs), black phosphorus (BP) as well as indium tin oxide (ITO) were fabricated employing a consistent approach and the nonlinear optical properties of these samples were characterized using the well-known open aperture zscan method.An extended-cavity femtosecond laser inscribed thulium-doped waveguide chip resonator was purpose-built to evaluate the performance of different SAs in this setup.Qswitched mode-locked (QML) operation at 1.9 μm was successfully demonstrated using each of the SAs and the influence of the parameters of the saturable absorbers on the laser performance was analyzed in detail.While being specifically relevant for waveguide lasers operating at 1.9 μm, the results presented in this work can provide important designguidelines for passively modulated laser systems in general.

Saturable absorber fabrication
All of the saturable absorbers except one that was based on ITO were prepared via spin coating from an appropriately prepared precursor.All raw TMDCs (MoS 2 , MoSe 2 , WS 2 and WSe 2 ) were obtained in course powder form (Alfa Aesar) and first had to be manually ground in a ceramic mortar to obtain particles of only a few micrometers in size.The ground TMDCs were then diffused in de-ionized water with concentrations of 2.5 mg/ml.The smaller sized hydrothermally exfoliated TI (bismuth telluride, Bi 2 Te 3 ) nanosheets [9] and the chemical converted multilayer graphene (CCG) powder [10] were directly diffused into de-ionized water with a concentration of 1mg/ml.All particle solutions were then ultrasonicated at room temperature for at least two hours to diffuse the raw materials into the water and to further reduce the particle size.BP nanosheets, consisting of a few layers of phosphorene, were fabrication by liquid exfoliation using the ionic liquid 1-hydroxyethyl-3-methylimidazolium trifluoromethansulfonate ([HOEMIM]-[TfO]) [11].BP crystal pieces (Manchester Nanomaterials Ltd) were added to the ionic liquid with a concentration of 3 mg/ml and then ultrasonicated for 20 hours.The resulting dispersion was filtered through a 100 nm Polytetrafluoroethylene (PTFE) membrane and the remaining [HOEMIM]-[TfO] was washed out using isopropanol.After that, any remaining isopropanol residue was removed using deionized water and the retained BP flakes on the PTFE membrane were finally diffused into deionized water and ultrasonicated at room temperature for half an hour.
Water soluble polyvinyl alcohol (PVA, Sigma-Aldrich) was utilized to stabilize all asprepared nanoparticles in the spin coating precursors to ensure homogeneity of the final spin coating layers.A PVA solution with a concentration of 100 mg/ml was prepared by dissolving PVA powder in deionized water over two hours at a temperature of 100 °C.The various SA material water dispersions were then mixed with the PVA water solution at a ratio of 2:1 and ultrasonicated for two hours to create the final spin coating precursors.The asmade precursors were finally spin coated onto a 170 μm thick soda-lime glass cover slip using a table-top spin coater with 400 rpm for 15 s and 1200 rpm for 15 s.This was followed by heating for 30 s at 80 °C.In order to fine-tune the optical properties of the final SA samples, saturable absorbers with different numbers of spin coating layers between 1 (i.e. a single layer) and 15 were fabricated.
The CNTs were fabricated using the arc discharge method and were then diffused into a dichlorobenzene (DCB) solution with a concentration of 0.2 mg/ml.Subsequently, poly[(mphenylenevinylene)-co-(2,5-dioctoxy-p-phenylenevinylene)] was used for purification and the dispersion was afterwards mixed with a polymethyl-methacrylate (PMMA) solution at a volume ratio of 1:1 and then ultrasonicated [12].Finally, the as-made precursor was spin coated onto a 170 μm thick fused silica glass substrate.
The only material that required a very different fabrication process based on physical vapor deposition (PVD) instead of spin coating was ITO.All ITO SAs were prepared by physical PVD on a 0.5 mm thick alkaline earth boro-alumnosilicate glass wafer.
The particle size and coverage of all the absorber materials on their substrates were checked under an optical microscope (Reichert-Jung 6526-0).

Optical nonlinearity characterization
To characterize the nonlinear properties of the fabricated absorbers, an open-aperture z-scan setup was utilized to measure the optical transmission T through the sample as a function of intensity I that is incident on the sample [13].This allowed a quantitative measurement of the saturation intensity I s , the modulation depth ΔT = T (I = ∞ ) -T (I = 0 ) and the non-saturable losses L non = 1 -T (I = ∞ ) of each of the individual saturable absorber samples.The laser source used in the z-scan setup was a diode-pumped, acousto-optically q-switched thulium laser emitting at 1880 nm.The q-switch repetition rate was varied between 5 kHz and 0.2 kHz and the pulse duration and average output power ranged from 70.7 ns and 49.5 mW at 5 kHz to 13.5 ns and 35.9 mW at 0.2 kHz.Focusing the beam with an anti-reflection (AR) coated lens (f = 50 mm) resulted in a minimum beam diameter on the device under test (DUT) of 74.4 μm, corresponding to a confocal parameter of the probe beam of about 4.6 mm.This value was much larger than the thickness of the samples (170 μm or 500 μm) as required by the z-scan technique [13].Under these focusing conditions, the maximum peak intensity that could be achieved on the surface of the saturable absorbers was 306 MW/cm 2 at 0.2 kHz and 10 MW/cm 2 at 5 kHz.For each absorber, two complete z-scan measurements were performed, and the obtained data sets were afterwards combined to improve measurement accuracy.It is worth mentioning that none of the investigated materials showed any signs of damage under irradiation at these intensity levels.Figure 1 shows a summary of all nonlinear transmission measurements of the as-prepared SAs.The modulation depths and saturation intensities were found by fitting the individual data sets with a simplified two-level energy model as  1 are also summarized for comparison in Fig. 2 and Table 1.As can be seen, the measured saturation intensities vary from 0.53 MW/cm 2 (Bi 2 Te 3 ) to 4.16 MW/cm 2 (ITO), while the modulation depth assumes values from 1.6% (CNTs) to 13.2% (Bi 2 Te 3 ).In order to be able to better compare the nonlinear modulation capability of different SAs, a new parameter called the "nonlinear modulation coefficient ( NLM η )" was introduced in this study and is included in Table 1.NLM η is defined as the ratio of the modulation depth ( T Δ ) of a SA and its saturation intensity ( S I ), i.e. .
A higher nonlinear modulation coefficient ( NLM η ) characterizes an absorber that introduces a larger optical modulation at a given light intensity or, equivalent, one that requires a lower light intensity to achieve a certain modulation depth.From all the samples tested, the Bi 2 Te 3 saturable absorber consisting of 15 spin-coating layers featured the largest NLM η with a value of 13.3%/MW•cm −2 .At the same time, the strong dependence of the modulation depth of Bi 2 Te 3 SAs on the number of spin-coated layers enables a wide tunability of the nonlinear transmission properties of these absorbers during fabrication which makes this material a very interesting candidate.The material with the lowest NLM η were the CNTs with a value of 1.2%/MW•cm −2 i.e. less than a tenth of the value for Bi 2 Te 3 .It is also worth noting that both, the modulation depth and the saturation intensity of all SAs can be increased by a higher concentration of the raw material in the precursors and/or a higher material density on the substrates, i.e. a larger number of spin-coated layers.However, an increase in concentration is always also accompanied by higher non-saturable losses and therefore also with a corresponding degradation of the efficiency of the laser system, i.e. with an increased laser threshold and a reduced slope efficiency as will be demonstrated in the following section.From the table it can also be seen that the saturation intensity per spin-coating layer is virtually constant for all absorbers while the modulation depth shows only a weak dependence on the number of spin-coating layers for MoSe2 and BP.It is believed that this behaviour is caused by inter-layer interactions, yet further studies would be required to fully explain this effect.

Laser performance
All of the fabricated saturable absorbers were tested in an identical laser setup that is schematically shown in Fig. 3.The gain medium was a thulium-doped ZBLAN waveguide chip that was inscribed by a long-cavity chirped-pulse oscillator [14].Details about the inscription process can be found in [15].The main advantage of guided-wave lasers is that beam divergence is effectively removed, and high optical intensities can thus be maintained over a long length, resulting in high optical gain and high efficiency.In addition, a waveguide-chip geometry enables a compact and inherently robust laser design.The laser was pumped by two polarization-multiplexed laser diodes (785 nm) with a total maximum available pump power of 700 mW.One of the end-facets of the waveguide chip was butt-coupled to a dichroic input-coupling mirror (R>99%@1.85-2.1 μm and T>90%@790 nm), while the other facet was cut at Brewster's angle to avoid Fresnel reflections, as well as to generate a linear polarization state.An anti-reflection (AR)-coated plano-convex lens (f = 40 mm) was used to collimate the laser beam behind the waveguide while an additional AR-coated plano-convex lens pair was used to create an intracavity focus position where the SAs were placed under Brewster's angle to avoid additional reflection losses and Fabry-Perot effects.A high reflective output coupler (R = 91%@1.85-2.1 μm) was used to ensure a high intra-cavity power density on the SAs.The optical cavity length of the laser was 252 mm and the laser was fundamentally mode-locked at the corresponding repetition rate of 595 MHz.As a result of the Brewster-cut, the beam waist between lenses 2 and 3 was elliptical with an axial radius of 6.9 μm and 10.2 μm, respectively, corresponding to an average intracavity intensity on the absorbers (when placed at the waist position under Brewster's angle) of about 500 kW/cm 2 for a typical average output power of 100 mW.The focusing geometry was specifically designed to result in an average intracavity intensity on the SAs in the vicinity of the saturation intensity, which is required to initiate pulsed operation originating from individual noise peaks.The resulting build-up of short pulses circulating in the cavity subsequently leads to a complete saturation of the absorber and therefore to the generation of high peak-power q-switched mode-locked (QML) pulses (see e.g [2].for further details).
It is important to point out that neither cw-mode-locked nor pure q-switched operation was observed in any of our experiments at any pump power levels and that the laser always transitioned from cw operation directly to QML operation at a certain threshold pump power.The only exception was the 2-layer WS 2 absorber where pure q-switching but no QML was observed.This can be explained by the particular properties of the gain medium, the resonator and the absorber as the pulse build-up is governed by a dynamic interplay between gain saturation/recovery and absorber saturation/recovery.A detailed analysis can be found for example in the publication by Hoenninger et al [16].In our particular case, all absorber materials are fast saturable absorbers with recovery times ranging from 25 fs for BP [17] to 3 ps for the TDMC material MoSe 2 [18], which favors passive mode-locking over pure qswitching.The longest recovery time of 5 ps has been reported for the TMDC material WS 2 [19] which explains the observed pure q-switching behavior for only this particular material.The long upper state life time of Tm:ZBLAN of 11 ms means that in order to achieve pure cw mode-locked operation (instead of QML), much higher pump power levels would be required which explains the observed QML behavior versus cw mode-locking.By utilizing gain materials with much shorter upper state life times like e.g.Yb:ZBLAN (1.8 ms [20]) and high power pump diodes, transition to cw mode-locked operation should be feasible.
Figures 4-12 show the results obtained with each of the individual types of absorbers.Within the available pump power range, with all of the SAs the threshold for QML operation was reached, as indicated in the corresponding graphs.Table 2 provides a summary of all the results obtained.

Discussion and conclusions
From our experimental study, it can be seen that all the tested low-dimensional nanomaterial saturable absorbers are suitable candidates for the development of ultrashort-pulsed lasers in the 2-micron range.Given their broadband optical response, this conclusion can also be applied more generally to a much wider range of various passively q-switched and modelocked laser systems.Moreover, utilizing a few-layer spin coating approach for fabrication is an efficient way to control the effective concentration of the saturable absorber materials which is crucial considering the trade-off between a high modulation depth (requiring a high concentration) and low non-saturable losses (requiring a low concentration).From the graphs that are shown in Fig. 13, a few crucial trends can clearly be identified: There is an approximately linear dependency of both, the modulation depth as well as the QML threshold pump power on the non-saturable loss value of the SAs (which in turn is proportional to the concentration of the absorber material or the number of spin coating layers), see Figs. 13(a) and (c).However, compared to the other materials, the SAs composed of graphene, ITO and Bi 2 Te 3 -15L (15 spin-coating layers) feature an extraordinarily high modulation depth which comes at the expense of a highly increased QML threshold pump power.On the other hand, the BP-5L (5 layers), WSe 2 and MoSe 2 SAs stand out as offering the lowest QML pump power threshold values of approximately 240 mW.This threshold value is almost exactly the value that was also measured for the CNT absorber which is remarkable considering that their non-saturable losses are approximately five times higher.This feature is also reflected in the plots for the slope efficiency and the maximum output power that are shown in Figs.13(e) and (f).In general, these plots suggest that there is a negative linear dependence of the slope efficiency and the maximum output power on the non-saturable losses.Again, the data points corresponding to the BP-5L, WSe 2 and MoSe 2 absorbers all lie significantly high above the trendline which indicates that those materials in particular are excellent candidates for laser systems that require a high slope-and wall plug efficiency.The worst performing material in that regard is ITO that leads to an unusually low slope efficiency of 4.6% and a comparably low maximum output power of only 19.6 mW despite the rather moderate non-saturable losses of <7%.
Any possible correlation between the saturation intensity of the absorbers and their nonsaturable losses is less clear defined.As mentioned before, for each individual material the saturation intensity monotonically increases if the number of spin coating layers in increased (compare Fig. 2).However, across all the different SA samples, there is virtually no correlation between the saturation intensity and the non-saturable losses.If graphene and ITO are excluded, a linear trend-line fitted in Fig. 13(b) is almost horizontal and highlights that the values for the saturation intensity are uniformly scattered around a value of about 1.25 MW/cm 2 .Graphene and ITO on the other hand exhibit unusually high saturation intensity values with 3.5 MW/cm 2 and 4.2 MW/cm 2 , respectively.
In terms of pulse duration (and thus peak intensity), the shortest q-switched pulse widths could be achieved when utilizing the CNT and the BP-10L absorbers with approximately 750 ns, while the use of the Bi2Te3-10L absorber resulted in the generation of the longest qswitched pulses with a duration of 1.25 μs.The highest repetition rate of 68 kHz was observed when the MoS 2 absorber was used, in contrast to the lowest repetition rate which as achieved with the Bi 2 Te 3 -15L SA and which was approximately half this value (36 kHz).
In conclusion, the systematic experimental investigations presented here reveal that all low-dimensional nanomaterial based saturable absorbers considered in this study are suitable candidates for the realization of ultrashort-pulsed lasers in the 2-micron range.It was shown that the choice of SA material employed mainly depends on the overall design requirements of the laser, e.g.high slope efficiency versus high repetition, high modulation depth versus low non-saturable losses, etc.In addition, we have also demonstrated that a fabrication process that is based on few-layer spin coating is a very effective way to control and to finetune the parameters of a saturable absorber via the choice of the number of individual spin coating layers.

1 α
1) with L being the thickness of the sample, 0 α the non-saturable absorption coefficient as per T (I = ∞ ) = exp (-0 ) L α and represents the saturable absorption coefficient as per T (I = 0 ) results shown in Fig.

Fig. 2 .Table 1 .
Fig. 2. Modulation depth of the fabricated saturable absorbers as function of their saturation intensity.SAs located in the top left corner feature a high modulation depth and low saturation intensity corresponding to a large nonlinear modulation coefficient NLM η .Table 1.Summary of the experimentally determined nonlinear transmission properties of all fabricated saturable absorbers.The saturation fluence F S was calculated based on their recovery time τ r according to F S = I S τ r .I S : saturation intensity, ∆T: total modulation depth of the absorber, η NLM : nonlinear modulation coefficient, L non : nonsaturable losses.The number of spin-coated layers for the individual SAs are shown in brackets.Minimum and maximum values are indicated for each property.Material I s (MW/cm 2 )

Fig. 3 .
Fig. 3. Schematic of the laser setup used to investigate the individual saturable absorbers.The diameter and length of the femtosecond laser inscribed waveguide (WG) was 50 μm and 12 mm, respectively.The focal lengths of lenses 1-3 were 40, 20 and 20 mm respectively.The distances d 1 , d 2 , d 3 and d 4 were 40, 120, 40 and 35 mm respectively.z: laser propagation direction, Brewster angle is in the (z-x) plane.

Fig. 4 .
Fig. 4. Laser performance of graphene saturable absorbers with 5, 10 and 15 spin coating layers.(a) Output power as function of pump power.(b) Pulse train shown on different times scales.(c) Q-switched repetition rate and pulse duration as function of pump power.(d) RF spectrum at maximum pump power.

Fig. 5 .
Fig. 5. Laser performance of the CNTs saturable absorber.(a) Output power as function of pump power.(b) Pulse train shown on different times scales.(c) Q-switched repetition rate and pulse duration as function of pump power.(d) RF spectrum at maximum pump power.

Fig. 6 .
Fig. 6.Laser performance of the Bi 2 Te 3 saturable absorbers with 5, 10 and 15 spin coating layers.QML operation was not observed with the 5-layer sample.(a) Output power as function of pump power.(b) Pulse train shown on different times scales.(c) Q-switched repetition rate and pulse duration as function of pump power.(d) RF spectrum at maximum pump power.

Fig. 7 .
Fig. 7. Laser performance of the TDMC MoS 2 saturable absorbers with 1, 2 and 5 spin coating layers.QML operation was only observed with the 2-layer sample.(a) Output power as function of pump power.(b) Pulse train shown on different times scales.(c) Q-switched repetition rate and pulse duration as function of pump power.(d) RF spectrum at maximum pump power.

Fig. 8 .
Fig. 8.Typical laser performance of the TDMC MoSe 2 saturable absorbers with 1, 2 5 and 10 spin coating layers.QML operation was observed with the 5-and the 10-layer sample.(a) Output power as function of pump power.(b) Pulse train shown on different times scales.(c) Q-switched repetition rate and pulse duration as function of pump power.(d) RF spectrum at maximum pump power.

Fig. 9 .
Fig. 9. Laser performance of the TDMC WS 2 saturable absorbers with 1 and 2 spin coating layers.QML operation was observed with the 1-layer sample whereas pure q-switched operation was observed with the 2-layer sample.(a) Output power as function of pump power.(b) Pulse train shown on different times scales.(c) Q-switched repetition rate and pulse duration as function of pump power.(d) RF spectrum at maximum pump power.

Fig. 10 .
Fig. 10.Laser performance of the TDMC WSe 2 saturable absorbers with 1, 2, 5 and 10 spin coating layers.QML operation was only observed with the 10-layer sample.(a) Output power as function of pump power.(b) Pulse train shown on different times scales.(c) Q-switched repetition rate and pulse duration as function of pump power.(d) RF spectrum at maximum pump power.

Fig. 11 .
Fig. 11.Laser performance of the BP saturable absorbers with 1, 2, 5 and 10 layers.QML operation was achieved using the 5 and the 10-layer sample.(a) Output power as function of pump power.(b) Pulse train shown on different times scales.(c) Q-switched repetition rate and pulse duration as function of pump power.(d) RF spectrum at maximum pump power.

Fig. 12 .
Fig. 12. Laser performance of the ITO saturable absorber.(a) Output power as function of pump power.(b) Pulse train shown on different times scales.(c) Q-switched repetition rate and pulse duration as function of pump power.(d) RF spectrum at maximum pump power.