1. Introduction

Mid-infrared lasers have been used in a range of application fields, such as plastic and glass processing [1], gas detection [2], long-range light detection and ranging (LIDAR) [3, 4], free-space optical communication [5], medical diagnostics [6], and laser surgery [7]. Until now, most mid-infrared lasers have been based on solid-state lasers, since crystal-based gain mediums are easily available at various mid-infrared wavelengths [8]. However, significant technical attention has been paid recently to the development of highly stable and compact mid-infrared lasers that use optical fiber-based platforms, because fiberized lasers can provide a range of advantages over traditional free-space optics-based lasers in terms of beam quality, reliability, and environmental stability [9].

It is well known that optical fibers doped with thulium (Tm), holmium (Ho), or erbium (Er) ions can be used to build lasers that operate at the spectral region between 1.8 and 3.5 μm, depending on the host materials [10]. The host materials include silica [11], silicate [12], tellurite [13], and fluoride glass [14]. The use of such high-performance lasers built with these fibers with high pump efficiency and good power scalability has been successfully demonstrated [10, 15, 16]. Fiber lasers can be operated in either the continuous-wave or pulse mode, depending on the application field. In order to produce pulsed output beams from laser cavities, either a Q-switching or a mode-locking component must be incorporated into the cavity. Q-switching is the modulation of the optical cavity’s quality factor (Q-factor) to ensure the cavity in a transient mode, whereas mode-locking is the locking of the relative phases of multiple lasing modes within the cavity in a steady state mode. These mechanisms for laser pulse formation can be accomplished through two basic schemes: passive and active.

One of the key components for the implementation of passively Q-switched or mode-locked lasers is the saturable absorber. The saturable absorbers most commonly used have been based on semiconductors for a long time, and in the past decade, intensive investigations have been conducted for an alternative that is based on carbon nanotubes (CNTs) [17–24]. In the past few years, graphene, which has a single layered hexagonal structure of carbon atoms, has attracted enormous technical interest for research because its gapless energy band structure is known to be capable of providing saturable absorption operation over a wide bandwidth [25–36]. The excellent saturable absorption performance of CNTs and graphene over a wide bandwidth that includes 2 μm wavelengths has been successfully demonstrated [25, 28, 37–41].

Quite recently, a new material called “topological insulator (TI)” has gained significant scientific and technical attention in the field of condensed matters because of the extraordinary charge and spin properties on the edge or surface modes of TIs [42]. Since mercury telluride (HgTe) quantum well structures were experimentally identified as the first 2-dimensional (2D) topological insulator by König et al. in 2007 [43], the bulk crystals of Bi_1-xSb_x, Bismuth selenide (Bi₂Se₃), Bismuth telluride (Bi₂Te₃), and Sb₃Te₃ were subsequently confirmed to be 3D topological insulators [44–47]. Thus, TIs have been considered as a promising material platform for spintronic and quantum computing devices. However, the potential of their unique optical properties had attracted less technical attention, until the experimental demonstration by Bernard et al. in 2012 [48]. In their work, Bernard et al. showed the feasibility of using the new Dirac materials for photonic applications as a nonlinear saturable absorber [49]. Since then, quite a few experimental investigations on the use of TI-based saturable absorbers for mode-locking or Q-switching, have been conducted [49–55], including our recent demonstration [55]. The saturable absorption of TIs occurs due to the Pauli blocking process [56], since they have an energy band structure similar to semiconductors. The combination of the small bandgap bulk (0.2 ~0.3 eV) and the gapless surface enables TIs to possess an ultra-broad bandwidth of saturable absorption operation [52], which can cover the mid-infrared spectral region. The operating bandwidth of TI-based saturable absorbers is expected to be no less than those of graphene and carbon nanotubes due to their gapless energy band structure on the surface, even if it is yet to be further investigated.

However, there has been only one experimental demonstration on the mid-infrared operating capability of TI-based saturable absorbers so far [57], to the best of the authors’ knowledge. Therefore, we believe that it would be technically meaningful to perform another systematic investigation on the applicability of TI-based saturable absorbers for the implementation of 2 μm pulsed fiber lasers, which can be covered by thulium (Tm) or holmium (Ho)-doped fiber lasers from a perspective of expanding the technological base of TIs.

In this paper, we report our recent investigation results on the use of a saturable absorber based on a bulk-structured Bi₂Te₃ TI for the generation of mode-locked femtosecond pulses in the 2 μm spectral region. Our fiberized saturable absorber was prepared by depositing a mechanically exfoliated, thick Bi₂Te₃ TI layer on top of the flat side of a side-polished optical fiber. For this experimental demonstration we chose to use the evanescent field interaction scheme, which is enabled by a side-polished fiber, since it has several advantages such as a low mode-locking threshold, all-fiber configuration, and high-power operation [58]. Unlike the previous TI-based saturable absorber demonstrations [48–54, 57], a bulk-structured, micrometer-thick TI layer was used as a saturable absorption material because of ease of fabrication. Even if the prepared, ~30 μm-thick saturable absorber does not possess ultra-thin nanosheet structures, we found that it can readily provide a modulation depth of ~20.6% at 1.95 μm, which is sufficient for mode-locking. Stable mode-locked pulses with a temporal width of ~795 fs is shown to be readily obtainable through evanescent field interaction between the oscillated beam and the Bi₂Te₃ TI layer from the laser cavity at an operating wavelength of 1935 nm.

2. Experimental laser schematic

The laser schematic used for our experimental demonstration is shown in Fig. 1(a). The gain medium was a 1 m long Tm/Ho co-doped fiber (CorActive, TH512) with absorption of 13 dB/m at a wavelength of 1550 nm. The ring cavity was constructed with the passive components of an isolator, a 90:10 coupler, a polarization controller, a 1550/2000 nm wavelength division multiplexing (WDM) coupler, and a Bi₂Te₃-deposited side-polished fiber. The pump beam from a 1550 nm semiconductor laser diode with a maximum output power of ~250 mW was coupled into the Tm/Ho co-doped fiber via the 1550/2000 nm WDM coupler. The mode-locked laser output was extracted from the ring cavity via a 10% output port of a 90:10 fiber coupler.

 

Fig. 1 (a) The laser schematic. (b) Measured SEM image of the mechanically exfoliated, thick Bi₂Te₃ layer. (c) Bi₂Te₃ layer depth measurement result using an alpha step profiler.

Download Full Size | PDF

The laser cavity was ensured to be built with a single-mode fiber (SM2000) that is optimized at a wavelength of 2 μm [59] for the purpose of optimum laser performance, except the Tm/Ho co-doped fiber. The SM2000, which has a cut-off wavelength of ~1.7 μm, is well known to be optimized for 2 μm operation [59] because of a much smaller bending loss than the conventional standard fiber of SMF28, despite almost similar propagation loss and dispersion properties. All the components within the cavity were fusion-spliced; the total length of the ring cavity was ~7 m. The dispersion of the SM2000 and Tm/Ho co-doped fiber were measured to be −0.067 (+/−0.005) ps²/m and −0.056 (+/−0.005) ps²/m, respectively, at a wavelength of 1.95 μm. The dispersion measurement method for the fibers is fully described in [41]. Using the measured dispersion values of the SM2000 and Tm/Ho co doped fiber, we roughly estimated the total cavity dispersion at about −0.458 ps² at 1.95 μm. The insertion losses of the coupler, isolator, and WDM were measured to be ~0.8, ~1.2, and ~0.5 dB, respectively, at 1.95 μm. The average power of the mode-locked laser output was measured to be ~1 mW.

The saturable absorber used in this experimental demonstration was prepared by depositing mechanically exfoliated Bi₂Te₃ particles onto the flat side of a side-polished SM2000 fiber. To prepare the side-polished fiber, one side of the SM2000 was polished while the fiber was fixed onto a V-grooved quartz block. The distance between the flat side and the top side of the fiber core was measured at ~7 μm, as shown in the inset of Fig. 1(a). The propagation loss of the side-polished fiber without the Bi₂Te₃ TI layer deposited was measured to be ~0.8 dB.

In order to obtain a Bi₂Te₃ TI layer, we used the well-known mechanical exfoliation method [60, 61]. The commercially available Bi₂Te₃ bulk single crystal (Alfa, Aesar) was used as a starting material. Bi₂Te₃ particles were repeatedly peeled from the surface of the bulk crystal using scotch tape to form a thick layer. Note that our goal was to obtain a bulk-structured layer with a thickness of micrometers, rather than a nanometer-thick, atomic layer; thus, special care to obtain nanometer-thick, atomic layers did not need to be taken during the exfoliation process. The prepared Bi₂Te₃ layer, which adhered to a small segment of scotch tape, was then transferred together and placed onto the flat side of the side polished fiber, as shown in Fig. 1(a). A small amount of index match oil was used on the flat surface to help the light coupling between the Bi₂Te₃ film and the fiber core. The beam interaction length of the Bi₂Te₃ TI-deposited side-polished fiber was ~2.5 mm.

Figure 1(b) shows the measured scanning electron microscope (SEM) image of the prepared Bi₂Te₃ layer. A clean surface without distinctive layered structures is evident from the figure, which indicates that the prepared layer was bulk-structured. The thickness of the prepared film was estimated to be ~30 μm using an alpha step profiler, as shown in Fig. 1(c). It is well known that the evanescent field interaction with a deposited film is not influenced by the film thickness in the case of $nd>>\lambda$, where $n$ and $d$ are the refractive index and the thickness of the deposited film, respectively, and $\lambda$ is the wavelength of light [62]. Considering the fact that the $nd$ value of our prepared bulk-structured Bi₂Te₃ TI-deposited side-polished fiber is much larger than the light wavelength of 1935 nm, it is believed that the impact of the film thickness variation on the saturable absorption performance is negligible. The minimum insertion loss and polarization-dependent loss (PDL) of the Bi₂Te₃ TI-deposited side-polished fiber were ~2.1 and ~2 dB at 1.95 μm, respectively.

The measured Raman spectrum of the Bi₂Te₃ layer is shown in Fig. 2(a). Four Raman optical phonon peaks were identified as E_g¹ at ~40 cm⁻¹, A_1g¹ at ~62 cm⁻¹, E_g² at 101 cm⁻¹, and A_1g² at 137 cm⁻¹. These four peaks are typical in bulk crystalline Bi₂Te₃ [63]. As expected, the A_1u² peak, which is known to be generated by the broken symmetry in atomically thin films, was not observed [64]. This indicates that our prepared, ~30 μm-thick layer is in a bulk crystalline status.

 

Fig. 2 (a) Measured Raman spectrum of the mechanically exfoliated, thick Bi₂Te₃ layer. (b) XPS scan for Bi 4f region. (c) XPS scan for Te 3d region. (d) Measured nonlinear absorption curve at 1.95 μm.

Download Full Size | PDF

To verify the chemical stoichiometry of the mechanically exfoliated layer, we measured the X-ray photoelectron spectroscopy (XPS) spectrum of the layer. Figure 2(b) shows the high resolution Bi 4f spectrum, whereas the Te 3d spectrum is shown in Fig. 2(c). The two small peaks at 162.7 and 157.3 eV in the Bi 4f region of Fig. 2(b) are consistent with the reported values of the binding energies of Bi 4f_5/2 and Bi 4f_7/2, whereas the two small peaks at 582.2 and 571.8 eV in the Te 3d region of Fig. 2(c) are consistent with those of Te 3d_3/2 and Te 3d_5/2 [65]. It should be noted that two additional strong peaks were observed at higher binding energies in each region. The additional peaks were located at 164.2 and 158.9 eV in the Bi 4f region, whereas they were at 586 and 575.7 eV in the Te 3d region. The existence of those additional peaks can be attributed to the oxidation of Bi and Te atoms on the surface [65].

Next, we measured light absorption as a function of the input optical pulse peak power in order to determine the nonlinear absorption performance of the prepared Bi₂Te₃ TI-deposited side-polished fiber. We conducted this measurement using an ~1-ps, mode-locked fiber pulse laser operating at 1.95 μm, as shown in Fig. 2(d). The saturation power was found to be ~29 W. The estimated modulation depth was ~20.6%. We believe that the nonlinear saturable absorption effect occurs mainly because of the narrow bandgap of the bulk, rather than the surface of our bulk-structured Bi₂Te₃ TI layer, unlike nanosheet-based TIs in which both the bulk and the surface states contribute to nonlinear saturable absorption. Note that the ~20.6% modulation depth is three times higher than that of our recently reported, graphene oxide-based saturable absorber that has the same side-polished fiber platform [56]. The reason for the modulation depth difference between the two materials must be investigated further.

3. Experimental femtosecond pulse generation from the cavity

Figure 3(a) shows the measured oscilloscope trace of the output pulses. The period of the output pulses was measured to be ~35.8 ns, which corresponds to a repetition rate of 27.9 MHz. This measured repetition rate coincided with the estimated fundamental frequency of the implemented ring cavity. The measured electrical spectrum of the output pulses is shown in Fig. 3(b). A strong signal peak with a peak-to-background ratio of 76 dB was clearly observed at the fundamental repetition rate of ~27.9 MHz.

 

Fig. 3 Measured (a) oscilloscope trace and (b) electrical spectrum of output pulses (Inset: wide-span view)

Download Full Size | PDF

The measured optical spectrum of the output pulses is shown in Fig. 4(a). The center wavelength and 3 dB bandwidth were measured to be ~1935 nm and ~5.64 nm, respectively. Assuming that the pulses were in a transform-limited soliton form, the output pulse-width was expected to be 698 fs using the time-bandwidth product value 0.315, of transform-limited hyperbolic secant pulses [66]. Kelly sidebands were clearly observed on the spectrum [67]. The first-order Kelly sideband position relative to the center wavelength was measured to be 8.2 nm. Using the experimental total cavity dispersion value of L|β₂| = 0.458 ps², the theoretical position of the first-order Kelly sideband relative to the center wavelength was calculated to be 9.117 nm in the case of transform-limited 698 fs pulses [66]. Figure 4(b) shows the theoretically calculated Kelly sideband position relative to the center wavelength (Δλ) as a function of the temporal width of transform-limited pulses (TFWHM) for various Kelly sideband orders. It is obvious that there exists a non-negligible difference between our measured Δλ of ~8.2 nm and the theoretical value of ~9.117 nm for the first-order Kelly sideband, which implies that the output pulses of our laser were chirped [68].

 

Fig. 4 (a) Measured optical spectrum of output pulses (b) Theoretically calculated Kelly sideband position relative to the center wavelength (Δλ) as a function of the temporal width of transform-limited pulses (T_FWHM) for various Kelly sideband orders (m).

Download Full Size | PDF

Finally, we performed an autocorrelation measurement using a second harmonic generation (SHG)-based autocorrelator (FR-103HS(XL)/IR, Femtochrome). Given that the optical power of the output pulses (average power of ~1 mW) was not high enough to induce a second harmonic signal, we used a Tm/Ho co-doped fiber amplifier just before the autocorrelator. In order not to induce any temporal and spectral distortion of the optical pulses during the amplification process, the amplifier gain was carefully adjusted to the level at which the output spectrum had no change, but could induce a second harmonic signal within the autocorrelator. The average optical power of the amplified pulses was ~20 mW. The amplified spectrum is shown in Fig. 5(a). Compared to the non-amplified pulse spectrum of Fig. 4(a), it is obvious that there is no significant spectral change. Figure 5(b) shows the measured autocorrelation trace of the amplified pulses, in which the amplifier dispersion of 0.5 ps² was deducted. The measured pulse width was estimated to be ~795 fs through the sech²() curve fitting. Assuming that the pulses were in a soliton form, the time bandwidth product value was estimated to be ~0.35.

 

Fig. 5 (a) Measured optical spectrum of amplified pulses. (b) Measured autocorrelation trace.

Download Full Size | PDF

4. Conclusion

We have experimentally demonstrated that a bulk-structured Bi₂Te₃ TI-based saturable absorber can readily be used for the generation of femtosecond mode-locked pulses in the 2 μm region. Using a bulk-structured, ~30 μm-thick Bi₂Te₃ TI layer transferred on top of the flat side of a side-polished fiber as a mode-locker, it was shown that stable ultrafast pulses with an ~795 fs temporal width could readily be generated at 1935 nm from a Tm/Ho co doped fiber laser.

Further to our recent work on the use of a bulk-structured Bi₂Te₃ TI-based saturable absorber for 1.55 μm mode-locking [55], this experimental demonstration also confirms that high crystalline quality, atomic-layered films of TI, which demand complicated and expensive material processing facilities, is not essential for ultrafast laser mode-locking applications. However, further theoretical and experimental investigations are required to precisely determine the advantages and disadvantages of bulk and nanosheet-structured TIs as saturable absorption materials for mode-locked laser applications.