Coherent laser sources emitting around 3 µm are always attracting continuous attention in biomedical applications because of their strong absorption in liquid water. Currently, this kind of laser has been extensively applied in skin^[1,2] and endodontic^[3] treatments, lithotripsy^[4], ophthalmology^[5], etc. To compliment the great successes achieved by solid-state lasers such as the ${\mathrm{Er}}^{3+}:\mathrm{YAG}$ laser^[6,7], the laser architecture based on advanced fluoride fibers with a compact arrangement, excellent beam quality, and maintenance-free design, provides us another alternative^[8–12]. As the pioneer of this class of lasers, ${\mathrm{Er}}^{3+}$-doped fluoride fiber lasers operating on the ${{}^4\mathrm{I}}_{11/2}\to {{}^4\mathrm{I}}_{13/2}$ transition have shown outstanding performance at $\sim 2.8\mathrm{\mu m}$, recording the highest continuous wave (CW) power of 42 W^[9] and slope efficiency of up to $\sim 60\%$^[13]. Based on this transition, pulse generation with µs–fs scale temporal widths based on the techniques of $Q$-switching^[14], gain-switching^[15], and mode-locking^[16] has also been reported, wherein nanosecond pulses with high energies and peak powers are particularly favored in laser surgery because of the reduction in unwanted thermal effects. To realize this kind of pulse, $Q$-switching, specifically based on an actively switched scheme implemented by externally controlled modulators such as an acousto-optic modulator (AOM)^[17], an electro-optic modulator^[18], or even a mechanical chopper wheel^[19], is a popular route because of its unique behavior for its energy storage before it releases a pulse, albeit at the sacrifice of some average power. Until now, maximum pulse energies of 560 µJ (53 ns width)^[20] and peak power of 16 kW (13 ns width)^[18] (not simultaneously) have been acquired from actively $Q$-switched ${\mathrm{Er}}^{3+}$-doped fluoride fiber oscillators. Following fiber amplifiers, the pulse energy could be even scaled to 0.75 mJ with a width of 95 ns and an estimated peak power of 7.4 kW^[21]. Nevertheless, it remains challenging when the wavelength is pushed toward 2.94 µm, the strongest water absorption peak, which has an absorption coefficient 3 times higher than at 2.8 µm. Despite the availability of efficient CW lasing at 2.94 µm using a long fiber that is needed for compensating for the long-wavelength gain^[22], the resulting high gain easily leads to the premature onset of lasing, therefore impeding the achievement of high energy and peak power. Accordingly, most actively $Q$-switched reports on ${\mathrm{Er}}^{3+}$-doped systems have been around or below 2.8 µm^[17–20,23]. As its neighbor in the spectral domain, the ${{}^5\mathrm{I}}_6\to {{}^5\mathrm{I}}_7$ transition of ${\mathrm{Ho}}^{3+}$ is faced with similar trouble, although it has a slightly red-shifted emission spectrum. The four-level nature particularly in a fiber architecture results in easy prelasing even in the low Q factor phase of $Q$-switching. As a result, the maximum reported pulse energy and peak power at 2.95 µm were 4.16 µJ and 12.5 W, respectively^[24]. At the shorter wavelength of 2.87 µm, the pulse energy and peak power were only enhanced to 19 µJ and 576 W, respectively^[25], indicating the contradictory requirements for obtaining high pulse energy and peak power, and the wavelength close to 2.94 µm. As a contrast, the three- or quasi-three-level ${\mathrm{Dy}}^{3+}$ (${{}^6\mathrm{H}}_{13/2}\to {{}^6\mathrm{H}}_{15/2}$) not only can operate around 2.94 µm in a short fiber (helpful for narrowing pulse thereby enhancing peak power) but also requires a large population inversion fraction for lasing (beneficial for energy storage hence scaling energy). Recently, Pajewski et al. employed an AOM to $Q$-switch a 1.1 µm core-pumped ${\mathrm{Dy}}^{3+}$-doped fluoride fiber oscillator at 2.912 µm, yielding a pulse width of 74 ns^[26], whereas only a low pulse energy of 17 µJ and peak power of 201 W were obtained due to the resonant pump excited state absorption (ESA) that depopulated the laser upper state^[27], detrimental for energy storage. From this view, clad pumping may be more suggested due to weaker pump intensity; however, this will lead to a significantly increased laser threshold^[28]. To address the issue, 980^[29,30] and 660 nm^[31] clad-pumped ${\mathrm{Er}}^{3+}/{\mathrm{Dy}}^{3+}$-codoped fluoride fiber oscillators have been developed recently, in which ${\mathrm{Er}}^{3+}$ aims to absorb the pump and then transfer energy to ${\mathrm{Dy}}^{3+}$ for emission, thus circumventing the potential pump ESA of ${\mathrm{Dy}}^{3+}$. Based on this platform, we presented active $Q$-switching in the region of 2.9–3.6 µm very recently^[32]. In that case, to demonstrate the wideband potential, 660 nm diode was adopted to activate both the $\sim 3\mathrm{\mu m}$ band of ${\mathrm{Dy}}^{3+}$ and $\sim 3.5\mathrm{\mu m}$ band of ${\mathrm{Er}}^{3+}$, while a moderate fiber length was selected to balance their gains for maximizing the spectral tuning range. Accordingly, the pulse energy and peak power at 2.951 µm were only 24 µJ and 55 W, respectively, with a pulse width of 404 ns^[32]. If focusing on the ${\mathrm{Dy}}^{3+}$ emission around 2.94 µm, 980 nm pumping should be more efficient, owing to smaller quantum defects^[30].

In this work, we employ a commercially available 980 nm diode as the pump, demonstrating an AOM actively $Q$-switched ${\mathrm{Er}}^{3+}/{\mathrm{Dy}}^{3+}$-codoped fluoride fiber oscillator with a short pulse width of 41 ns, in which a diffraction grating has been used to lock the wavelength at 2.943 µm. The significantly enhanced pulse energy of 108 µJ and peak power of 2.48 kW, far beyond our recent presentation^[32], set the new records for pulsed fiber oscillators around 2.94 µm.

The setup of the actively $Q$-switched ${\mathrm{Er}}^{3+}/{\mathrm{Dy}}^{3+}$-codoped fluoride fiber oscillator is depicted in Fig. 1, where pumping is provided by a commercial multimode laser diode (LD) (BWT, China) pigtailed to a silica fiber with a core diameter of 105 µm and a numerical aperture (NA) of 0.15. The gain fiber provided by FiberLabs Inc. is a segment of 4%/0.25% (mole fraction) ${\mathrm{Er}}^{3+}/{\mathrm{Dy}}^{3+}$-codoped double-clad fluoride fiber. The fiber core has a diameter of 18.8 µm and an NA of 0.13. The clad has a circular diameter of 249 µm and an NA of 0.5, with an absorption coefficient of 1–2 dB/m at 980 nm (provided by the manufacturer). In this case, the short fiber length of 1.05 m was selected in order to extract more energy (i.e., higher pulse energy) by increasing the prelasing threshold and to achieve a shorter pulse width and, thus, a higher peak power, at the sacrifice of the pump absorption ratio (i.e., $\sim 30\%$ measured). The fiber facet close to the pump was perpendicularly cleaved and butted against the front mirror (FM) (highly transmissive at 980 nm but reflective at 2.7–3.7 µm at 0°) as the cavity feedback, while the other one was 8° angle-cleaved to inhibit prelasing. The pump was coupled into the fiber clad via a pair of lenses (i.e., L1 and L2) with an estimated efficiency of 86%. The outputs from the angle-cleaved facet were collimated using the L3 before being incident on the dichroic mirror (DM) (highly transmissive at 980 nm but reflective at 2.7–3.7 µm at 45°) to remove the residual pump and then were steered into a Ge AOM with a Bragg angle. The AOM, anti-reflective coated at 2.5–5 µm, has a maximum diffraction efficiency of 80% and a rise time of $\le 140\mathrm{ns}/\mathrm{mm}$, and it was driven by a 68 MHz radio frequency (RF) source modulated by a rectangular wave from a function generator with variable duty cycle. The zeroth-order light functions as the output, and the first-order diffraction is resonant by a reflector, ensuring no feedback is left due to incomplete diffraction of the AOM when the lasing is held off, hence inhibiting prelasing^[25]. Here the reflector is a narrowband ruled diffraction grating (RDG) (Thorlabs, GR2550-30035) in a Littrow configuration, aiming to suppress the undesirable amplified spontaneous emission (ASE) components at the laser wavelength, therefore leading to a shorter pulse width and, thus, a higher peak power^[18,19], while locking the wavelength for stable $Q$-switching. The total cavity length of 2.25 m, containing a long free-space distance of 1.2 m (ensuring narrow filtering bandwidth), corresponds to a cavity round-trip time of 18 ns. The inset plots the simplified energy level diagram with some relevant transitions, where the ${{}^6\mathrm{H}}_{13/2}\to {{}^6\mathrm{H}}_{15/2}$ transition can be activated by pumping at 980 nm, based on the processes of ET1, ET2, and $\sim 2.8\mathrm{\mu m}$ emission and absorption.

The laser power was measured using a high-resolution thermal sensor with a detectable power of 100 µW to 5 W (Thorlabs, S405 C). The pulse signals were recorded using an HgCdTe detector with a $\le 3\mathrm{ns}$ time constant (VIGO, PCI-2TE-12) connecting with a 500 MHz oscilloscope. The optical spectra were collected by a spectrometer (Yokogawa, AQ6377) with a minimum resolution of 0.2 nm. Note that the above measurements were performed after a long-pass filter ($>2.9\mathrm{\mu m}$) except the spectrum.

First, we rotated the RGD to search for the position which could lock the wavelength at 2.94 µm while aligning the AOM slightly to maximize the output power. In our experiment, the maximum launched pump power was set at 24.9 W for reducing the thermal damage risk of the bare fluoride fiber facet for pump coupling, where its temperature was up to 80°C, identified by a thermal imager.

Then, the laser performance variations with the repetition rate from 100 Hz to 6 kHz were recorded as displayed in Fig. 2, where the duty cycle was adjusted at the same time to maintain stable $Q$-switching. One can observe that the pulse energy increases rapidly with the decreased repetition rate from 6 to 1 kHz as a result of longer time used for energy storage, and then tends to be saturated below $\sim 1\mathrm{kHz}$, corresponding to the scale of inverse upper-state lifetime of 650 µs. In addition, a higher pump power yields a higher pulse energy without any saturation signs, implying the possibility of energy scaling. On contrary, the pulse width decreases sharply first with the decreased repetition rate, owing to the fact that the increased pulse energy leads to a stronger temporal modulation of the net gain and, thus, a faster rise and decay of the optical power. Then it exhibits a saturation behavior as a result of the saturated pulse energy, where a higher pump power results in the earlier saturation relative to the repetition rate, similar to the previously reported actively $Q$-switched ${\mathrm{Dy}}^{3+}$-doped fluoride fiber oscillators^[26,32,33]. With the increase of the repetition rate, the average power increases and then approaches a constant value, approximately equal to the output power under the CW condition^[33]. In terms of the peak power, its evolution behavior is similar to the pulse energy as a result of the relatively moderate variation of the pulse width.

To clearly reveal the dependence of the output performance on the launched pump power, the case at the low repetition rate of 100 Hz has been exhibited as shown in Fig. 3. The average power increases almost linearly with the launched pump power, where the low slope efficiency was mainly caused by the use of short fiber and large energy loss via fluorescence at such a low repetition rate.

Of note is that the outputs include many linearly polarized components, showing a polarization extinction ratio (PER) of $\sim 10\mathrm{dB}$ since both the AOM and RDG are polarization sensitive. If replacing the RDG with a broadband reflector, the PER would decrease to 6–7 dB. To further improve the PER, a polarizer (e.g.,  Brewster window, film polarizer, and Glan laser prism) with the orientation parallel to the mounting base of the AOM (ensuring a large diffraction efficiency) can be introduced into the cavity between the DM and AOM. The pulse width decreases and then tends to be saturated as a result of $Q$-switching dynamics. At the launched pump power of 24.9 W, the shortest pulse width of 41 ns (2–3 times the cavity round-trip time) was obtained, which is, to our knowledge, the shortest of $Q$-switched and gain-switched fiber oscillators beyond 2.9 µm. By shortening the cavity length (either fiber or free-space length) under the premise of ensuring sufficient gain and a narrow filtering bandwidth, while increasing the modulation depth of the $Q$-switched element (i.e., AOM diffraction efficiency in this case), a shorter pulse width is expected. Owing to the fixed repetition rate, the pulse energy exhibits the same linear evolution behavior as the average power, thus leading to an almost linear evolution for the peak power, apart from the initial stage. At the maximum pump power, a record pulse energy of 108 µJ and peak power of 2.48 kW (calculated based on a shape factor of 0.94 for Gaussian pulses) have been achieved, which are more than 4 times^[32] and one order of magnitude^[26] higher than the previous best results from either $Q$-switched or gain-switched fiber oscillators around 2.94 µm (defined within 2.9–3 µm), respectively. Further scaling of the pulse energy and peak power was primarily limited by the pump power allowed by the bare fiber facet for pump coupling. To solve this issue, a potential way is to splice the LD pigtail with the ${\mathrm{Er}}^{3+}/{\mathrm{Dy}}^{3+}$-codoped fluoride fiber, in which a high-reflectivity fiber Bragg grating at the laser wavelength is directly written to replace the FM as the feedback. To significantly scale the output level, the use of a master oscillator power amplifier scheme is an optional solution. Figure 4 shows the corresponding pulse train and single pulse waveform with a basically Gaussian shape, where no evident temporal interference fringes are observed due mainly to the limited bandwidth of the detector, although the pulse wavelength was shifted twice by the AOM per round trip. The low amplitude fluctuation of 2% indicates great stability.

At this pump power, the optical spectrum scanning from 2.7 to 3.1 µm was also recorded as plotted in Fig. 5, where the long-pass filter was removed in order to exhibit the emission feature of the whole system. A strong laser signal centered at 2943 nm with a 3 dB bandwidth of 0.9 nm has been observed, where there are no obvious ASE components from the ${{}^6\mathrm{H}}_{13/2}\to {{}^6\mathrm{H}}_{15/2}$ transition of ${\mathrm{Dy}}^{3+}$, indicating the filtering effect of the RDG. In the spectral blue region, however, the strong parasitic lasing at $\sim 2.8\mathrm{\mu m}$ sitting on an obvious ASE pedestal from the ${{}^4\mathrm{I}}_{11/2}\to {{}^4\mathrm{I}}_{13/2}$ transition of ${\mathrm{Er}}^{3+}$ was captured. This is not surprising in such a short fiber considering the high gain for the ${\mathrm{Er}}^{3+}$ transition and weakened absorption of ${\mathrm{Dy}}^{3+}$ at $\sim 2.8\mathrm{\mu m}$. Here, the actual power proportion of the $\sim 2.8\mathrm{\mu m}$ emission was measured to be $\sim 30\%$. To inhibit the unwanted $\sim 2.8\mathrm{\mu m}$ components, appropriately increasing the ${\mathrm{Dy}}^{3+}$ doping concentration to enhance the absorption is an alternative strategy. In addition, we found that with the repetition rate increasing from 100 Hz to 1 kHz, this power proportion significantly decreased to $\sim 7\%$ due to the enhanced absorption by ${\mathrm{Dy}}^{3+}$ as a result of more ions returning to the ${{}^6\mathrm{H}}_{15/2}$ state. Of note is that the $\sim 2.8\mathrm{\mu m}$ signal was a CW instead of pulses identified by the detector because it was not resonant in the cavity closed by the FM and RDG. Nevertheless, it did not impact the stability and energy scaling of the $Q$-switched pulses. Furthermore, the calculated normalized frequency of 2.61 at this wavelength, slightly higher than the cut-off frequency of 2.405, implies the quasi-single-mode operation of the pulses with a high brightness, although the quantitative characterization needs a beam quality analyzer, which was absent in our lab. Based on a 1:1 imaging system, the optical fluence was estimated (according to the single-mode approximation) to be $36\mathrm{J}/{\mathrm{cm}}^2$, which has reached the levels required by skin treatment targeting shallow stratum corneum and epidermal layer (typical fluence of $2-20\mathrm{J}/{\mathrm{cm}}^2$ at $\sim 2.94\mathrm{\mu m}$^[1,2,34] and deeper tissue ablation with an improved demand of $25–100\mathrm{J}/{\mathrm{cm}}^2$^[2,35–37]. Of note is that even at a higher repetition rate of 1 kHz, which may be more preferred in some special medical applications^[38,39], our fluence still could reach $31\mathrm{J}/{\mathrm{cm}}^2$ [estimated according to the value in Fig. 2(c)]. These results highlight the great medical potential of our source.

In this work, we report active $Q$-switching operation of a 980 nm diode-clad-pumped ${\mathrm{Er}}^{3+}/{\mathrm{Dy}}^{3+}$-codoped fluoride fiber oscillator toward the 2.94 µm water absorption peak. Based on a short fiber length of 1.05 m and an RDG for locking the wavelength, stable nanosecond pulses with the shortest temporal width of 41 ns have been obtained at a low repetition rate of 100 Hz in the quasi-single-mode state. The achieved maximum pulse energy of 108 µJ and peak power of 2.48 kW, only limited by the damage risk of the bare fluoride fiber facet used for pump coupling, represent the highest levels of fluoride fiber oscillators around 2.94 µm (see Table 1), to the best of our knowledge. This high-performance nanosecond laser source provides a new opportunity for laser medical applications.