High-brightness laser sources generating visible ultrafast pulses are sought after for many applications, including bone surgery [1], corneal ablation [2], tattoo removal [3], blood cell imaging [4], materials processing [5,6], and laser displays [7]. They can also serve as seeds in nonlinear processes to generate ultraviolet light, which is currently in high demand [8].

The first demonstrations of femtosecond lasers operating in the visible spectrum were reported as early as in the 1970s, using mode-locked CW dye lasers [9]. Unfortunately, dye lasers suffered from rapid degradation during operation and required the handling of toxic and carcinogenic chemicals [10]. Nowadays, various options are used for the generation of ultrashort pulses of light in the visible spectrum: optical parametric oscillators seeded by solid-state lasers [11], all-fiber supercontinua [12], sources based on four-wave mixing and frequency doubling of near infrared lasers, and sources based on Cherenkov radiation wavelength conversion [13]. A useful property of those lasers is the possibility of wavelength tuning. However, they rely on hard-to-operate complex setups involving seed lasers in the near infrared, and they suffer from inherently low efficiencies. Ultrafast fiber lasers in the visible spectrum could thus represent an interesting option, owing to their ruggedness, simplicity, efficiency, reliability, lower cost, ease of use, and excellent beam quality.

The recent advent of semiconductor-based pump sources operating in the blue spectrum, such as gallium nitride (GaN) laser diodes, has provided a low-cost and efficient way to pump lanthanide-doped fibers for direct visible laser emission. Accordingly, over the last decade, significant breakthroughs have been achieved for CW visible fiber lasers. To date, the record output power reported for such lasers is 2.3 W for single-mode operation [14] and 5 W for multimode operation [15]. While these records have been established using Pr$^{3+}$-doped fluoride fibers operating in the red region of the visible spectrum, demonstrations of blue, green, orange, and deep-red lasers have also been reported based on this type of fiber [16]. Moreover, highly efficient laser operation in the green region has been demonstrated using a Ho$^{3+}$-doped fluoride fiber [17], and watt-level yellow laser operation has been demonstrated using a Dy$^{3+}$-doped fluoride fiber [18]. Demonstrations of visible fiber lasers have mostly been achieved using fluoride glass as a host medium because of its low phonon energy, which prevents nonradiative phononic decay. The electronic configuration of Dy$^{3+}$ and Sm$^{3+}$ has enabled visible laser operation within glass hosts of higher phonon energy, such as silica fibers, but such demonstrations are hindered by photodarkening issues [19–21].

Whereas an unstable mode-locked and $Q$ switched regime at 650 nm was observed in a Sm$^{3+}$-doped silica fiber laser back in 1990 [22], the first successful demonstration of a stable picosecond fiber laser operating in the visible spectrum took place in 2000, using a Pr$^{3+}$: Yb$^{3+}$-doped fluoride fiber [23]. This laser, actively mode-locked using an acousto-optic phase modulator, emitted 550 ps pulses at 239 MHz. Among the different well-established techniques currently used to obtain mode-locking in fiber lasers, the most popular rely on saturable absorber mirrors [24], nonlinear optical loop mirrors (NOLMs) [25], and nonlinear polarization evolution (NPE) of light in optical fibers [26]. These techniques have been applied successfully to generate ultrashort pulses of light in fiber lasers at various infrared wavelengths. In normal dispersion fiber lasers, the all-normal dispersion (ANDi) regime [27] yields energetic femtosecond pulses, while the dissipative soliton resonance (DSR) regime [28] generates rectangular-profile pulses of higher energies with picosecond to nanosecond durations. Mode-locked fiber lasers in the visible based on some of these concepts were recently demonstrated. A Pr$^{3+}$-doped fluoride fiber laser pumped by a solid-state laser at 850 nm through an upconversion scheme and mode-locked by a NOLM generated 96 ps pulses at 635 nm [29]. A Pr$^{3+}$: Yb$^{3+}$-doped fiber laser, pumped by a 444 nm laser diode and mode-locked by a NOLM, yielded 85 ps pulses at 635 nm [30]. In addition, 9 ps pulses with an energy of 1.1 nJ and a peak power of 124 W at 635 nm were obtained from a spatiotemporal mode-locked Pr$^{3+}$-doped fluoride multimode fiber laser based on NPE and pumped by a 443 nm laser diode [31]. Finally, a laser based on a Dy$^{3+}$-doped fluoride fiber, mode-locked by NPE, and pumped by a 453 nm laser diode, yielded 83 ps pulses with an energy of 2.4 nJ at 575 nm [32]. To sum up, all these demonstrations were done in the DSR regime, leading to narrow pulse spectra and pulse durations well above 1 ps.

In this Letter, we report a mode-locked fiber laser emitting ultrafast pulses at 635 nm. This ring laser, pumped by a 445 nm GaN laser diode and operated in the ANDi regime, is mode-locked through NPE. This single-mode fiber laser generates 1.6 ps pulses with an energy of 0.66 nJ and a repetition rate of 137 MHz. The pulses are compressed by a pair of gratings to reach a duration of 168 fs and a peak power of 0.73 kW. To our knowledge, this is the first femtosecond fiber laser operating in the visible.

The laser, based on the transition $^3$P$_0 \rightarrow ^3$F$_2$ in Pr$^{3+}$, is depicted in Fig. 1. It consists of a ring cavity made up of an optical gain fiber and a free-space section that includes quarter- and half-wave plates, an optical isolator, a polarizing beam splitter, and a bandpass filter. The gain fiber is a 90-cm-long double clad Pr$^{3+}$-doped zirconium fluoride (ZrF$_4$) fiber provided by Le Verre Fluoré. The fiber core is doped with 0.6 mol.% of Pr$^{3+}$ ($\mathrm {NA} = 0.08$) and has a diameter of 5.5 µm, which enables single-mode operation at the signal wavelength ($\lambda _{\text {cutoff}} = {0.57}\,\mathrm{\mu}\textrm{m}$). The 125 µm inner cladding has a circular shape cut by two parallel flats separated by 115 µm. The 180 µm outer cladding is fabricated of low index fluoroacrylate polymer and allows multimode propagation of the pump signal along the inner cladding ($\mathrm {NA} > 0.5$). The background losses are $\sim$1.5 dB/m in the fiber cladding at the pump wavelength and $\sim$0.2 dB/m in the fiber core at the signal wavelength [14]. The group-velocity dispersion (GVD) of the fiber is estimated at 75 fs$^2$/mm, which leads to a net cavity dispersion of 0.068 ps$^2$. Both fiber ends are cleaved at a 5° angle to avoid parasitic reflections, which are known to be detrimental to mode-locking operation. The pump signal is provided by a 445 nm highly multimode GaN laser diode (BWT K450FA5FG-20.00W) pigtailed to a standard multimode silica fiber (105/125 µm, NA = 0.22). The pump signal is coupled to the active fiber’s cladding by means of a pair of achromatic lenses (Thorlabs AC254-030-A-ML). The lens used to inject the pump signal in the Pr$^{3+}$-doped fiber also serves as a collimator for the red output signal without significant chromatic aberration. The same model of lens is used to re-inject the red laser signal into the fiber core at the distal end of the fiber. Inside the resonator, two dichroic mirrors placed at a 45° angle are used to close the ring cavity. For unpolarized light, they both provide a transmission of $\sim$95% at 445 nm and a reflection of 86% at 635 nm. To reach and maintain mode-locking operation by NPE in the active fiber, two quarter-wave plates (Thorlabs WPQ10E-633), a half-wave plate (Thorlabs WPH10E-633), a polarizing optical isolator (Thorlabs IO-5-633-VLP), and a bandpass filter (Thorlabs FLH633-5) are placed inside the resonator. The optical isolator acts as a polarizer while ensuring unidirectional laser propagation in the cavity, which is counterclockwise in Fig. 1. A Gaussian-shaped bandpass filter is used to operate the laser in the ANDi mode-locking regime [27]. It is centered at 634 nm and has a 5 nm bandwidth. This filter also ensures that the laser will be based on the 635 nm transition, and not on the 615 nm or 700 nm transitions. The output coupling is provided by a polarizing beam splitter and is controlled using a half-wave plate. The length of the free-space propagation section in the laser cavity is 82 cm. Outside the resonator, the laser output is directed using two metallic mirrors into a double-pass compression stage, which consists of two volume-phase holographic transmission gratings (Thorlabs GP1006G) separated by $\sim$12 cm and a tilted metallic mirror to generate Output 2.

 

Fig. 1. Experimental setup of the fiber laser. BD, beam dump; DM, dichroic mirror; ISO/POL, polarizing optical isolator; L, lens; M, mirror; PBS, polarizing beam splitter; TG, transmission grating; $\lambda /2$, half-wave plate; $\lambda /4$, quarter-wave plate.

Download Full Size | PDF

For an injected pump power of 14 W at 445 nm and a proper orientation of the wave plates, a stable and self-starting mode-locking regime is obtained. The laser properties are first characterized before the compression stage, i.e., at Output 1 in Fig. 1. The temporal properties of the laser are measured using a 2 GHz photodetector (Thorlabs DET025A/M). The pulse train is shown in Fig. 2(a). Figure 2(b) presents the radio frequency spectrum acquired with a resolution of 50 Hz over a 20 kHz span. The fundamental repetition rate is 137.123 MHz, in agreement with the 7.3 ns period of the pulse train, and also with the expected fundamental repetition rate for this cavity length. The signal-to-noise ratio (SNR) is 60 dB and confirms the stability of the laser.

 

Fig. 2. (a) Pulse train of the mode-locked laser measured with a 2 GHz photodetector. Interval between pulses, 7.3 ns. (b) Radio frequency spectrum of the laser: resolution, 50 Hz; central frequency $f_0$, 137.123 MHz; SNR, 60 dB.

Download Full Size | PDF

The spectral properties of the laser were measured using an optical spectrum analyzer (OSA) (Yokogawa AQ6373B). The blue curve in Fig. 3(a) shows the normalized output spectrum measured with a resolution of 0.1 nm. The laser is centered at 635 nm and has a FWHM bandwidth of 7.1 nm. A measurement of the laser output spectrum between 400 nm and 800 nm confirmed that the red output signal remained more than 25 dB higher than the signal at any other wavelength. The average output power measured for the chirped pulses is 90 mW, which leads to a pulse energy of 0.66 nJ. A measurement of the autocorrelation trace at Output 1 using a Femtochrome FR-103 MN autocorrelator has a FWHM of 2.3 ps and is shown in the inset of Fig. 3(b). Assuming a Gaussian pulse profile, the pulse duration at FWHM is estimated to be 1.6 ps, leading to a peak power of 388 W. After the compression stage, i.e., at Output 2 in Fig. 1, the average power of the laser is reduced to 19.7 mW, owing to a transmission efficiency of $\sim$70% at the first order of diffraction for each grating. The blue curve in Fig. 3(b) shows the autocorrelation trace of the dechirped pulses measured on the Femtochrome FR-103 MN autocorrelator, leading to a FWHM of 242 fs. The absence of other features in the 20-ps scan range of the autocorrelation trace, the absence of strong interference fringes in the optical spectrum, and the fact that a pulse train with the fundamental repetition rate was measured imply that a single pulse is present in the cavity. Furthermore, continuous reduction of the pump power until the single pulse disappeared did not introduce any discontinuous modification of the optical spectrum that could have indicated the disappearance of other pulses; this further confirms the previous assumption. Pulse reconstruction based on the pulse spectrum and its autocorrelation trace using the PICASO algorithm [33] leads to a pulse duration of 168 fs and a peak power of 732 W. This value of pulse duration, which is very close to the value of 171 fs obtained assuming a perfect Gaussian temporal profile, is the shortest reported to date for a visible-wavelength fiber laser. The pulses have an energy of 0.14 nJ. Gratings with a better transmission figure, 90% for instance, could compress the pulses more efficiently and yield a peak power of the order of 2 kW. A physical model and a numerical simulator were developed to study the behavior of this laser and are presented in the Supplemental document accompanying this Letter. Simulation results are shown in Fig. 3 alongside the experimental results. The simulated pulse has a relatively flattop spectrum with a slight asymmetry in favor of the short wavelengths, and a bandwidth of 7.1 nm centered at 635 nm. The pulse duration is 1.66 ps and its energy is 0.66 nJ, with a peak power of 394 W. The pulse is compressed to 176 fs. These results are in excellent agreement with the experimental results, paving the way for the following discussion of the laser dynamics.

 

Fig. 3. (a) Normalized output spectrum of the fiber laser: FWHM, 7.1 nm; OSA resolution, 0.1 nm. (b) Autocorrelation trace of the compressed pulse: FWHM, 242 fs. Inset: Autocorrelation trace before the compression stage ($\mathrm {FWHM} = 2.3$ ps). (c) Compressed pulse profile: FWHM, 168 fs. (d) Simulated power and instantaneous frequency profiles of the pulse at the laser output, before the compression stage.

Download Full Size | PDF

The output of the laser is a highly chirped pulse, as expected in the ANDi regime. Its 7.1-nm spectrum is slightly larger than the 6-nm bandwidth of the 635 nm transition of the Pr$^{3+}$ ions, owing to the nonlinear spectral broadening of the pulse, associated with self-phase modulation occurring in the fiber. The pulse is compressible by a factor $\sim$10 and its compression shows a Strehl ratio well above 90%. This tends to indicate that the pulse behaves like a similariton within the gain fiber, acquiring a pulse profile close to a parabola with a linear chirp profile, favoring efficient pulse compression by a Treacy compressor. Figure 4 shows the evolution of the pulse parameters within the gain fiber obtained in the simulation. The pulse dynamics show that the pulse energy, pulse duration, and pulse bandwidth increase linearly in the last half of the fiber, supporting the similariton interpretation. The NPE kicks in at a position of around 0.4 m, where the saturation of the peak power in the $x$ polarization is combined with the sudden increase of the energy and peak power in the $y$ polarization. In this configuration of the wave plates, the pulse energy reaches a value close to the saturation of the NPE mechanism. Increasing the pump power, which was not possible in our case, since the limit of the laser diode was reached, would probably lead to a degradation of the pulse profile or even to wave breaking, owing to the nonlinear transmission of the NPE mode-locking mechanism. In other words, the nonlinear rotation of the polarization ellipse would become too important, distorting the pulse profile and increasing its losses, which would seriously hinder efficient pulse compression and eventually lead to pulse splitting in the cavity.

 

Fig. 4. Evolution of the pulse parameters as a function of the position in the gain fiber: peak power, energy, duration, and pulse bandwidth. The latter two are defined as $2.355 \times$ the rms temporal and spectral widths, respectively. Solid lines, $x$ polarization; dotted lines, $y$ polarization.

Download Full Size | PDF

It was not a priori obvious that the ANDi regime could be achieved in such a laser cavity. Indeed, the dispersion of the fiber at 635 nm is approximately four times larger than the dispersion of Yb-doped fibers operated at 1030 nm, for instance. This could have had a detrimental impact on the nonlinear pulse evolution. Fortunately, the nonlinear parameter of this fiber is approximately four times larger than its counterpart in 10-µm-core Yb-doped fibers at 1030 nm, which compensates for the larger dispersion value and allows the nonlinear effects to broaden the spectrum and generate mode-locking, enabling the formation of a similariton in the cavity. Higher-order dispersion terms and nonlinear effects, such as Raman scattering and self-steepening, were included in the simulations. Their impact is almost negligible in the laser configuration discussed here. In fact, a portion of the slight asymmetry of the pulse spectrum could be associated with those higher-order propagation terms, but further investigation is required to pinpoint the mechanism generating this asymmetry.

The 0.6% lasing efficiency of this system is, obviously, really low. The design was, however, not optimized to yield a high efficiency, but rather to produce ultrashort pulses. A short gain fiber was used to constrain detrimental nonlinear effects, leading to a significant amount of residual pump power at the end of the fiber. The system also shows some important limitations, because it possesses a free-space section. The injection efficiency of the pump power is $\sim$70%. The signal undergoes important insertion losses at each component and during re-injection in the optical fiber. We estimate those losses to be of the order of 67%. An all-fiber setup involving a pump combiner to inject the pump power would certainly improve the overall loss figure and efficiency. A shorter gain fiber with an optimized design in terms of cladding–core geometry, doping concentration, dispersion, and nonlinear parameters could thus lead to energy scaling in the nanojoule range. These considerations are under investigation, taking into account several constraints, such as the availability of fiber-pigtailed optical components that can transmit and support several watts of pump power. The good news is that the feasibility of splicing silica fibers to fluorozirconate fibers has been demonstrated [14,34], opening the door to the use of passive silica fibers in the laser cavity, and thus to a wide variety of fiber-pigtailed components. Nonetheless, there are still different simpler avenues to explore for the optimization of our actual system [35]. For instance, using a longer fiber could lead to a higher pulse energy at the expense of a longer pulse duration. Using a smaller or a larger filter bandwidth would lead to shorter pulses with a larger spectral width or more energetic dissipative solitons [36], respectively.

In conclusion, we have demonstrated, to our knowledge, the first femtosecond fiber laser operating in the visible range of the electromagnetic spectrum. This single-mode laser is based on a praseodymium-doped fluorozirconate gain fiber and is mode-locked through NPE. It is operated in the ANDi regime and it generates 1.6 ps, 0.66 nJ, highly chirped, similariton-like pulses. Those pulses are compressed down to 168 fs using a Treacy compressor. The optimization of the cavity design and the integration of fiber-pigtailed components could eventually lead to an all-fiber version of this laser, yielding femtosecond pulses in the nanojoule energy range. Moreover, several avenues closely related to this proof of concept, such as the generation of pulses based on other visible transitions, the use of alternative ultrafast laser architectures, and the incorporation of an amplification stage, will be investigated. Overall, this demonstration paves the way for a new class of reliable, efficient, and compact ultrafast lasers operating in the visible spectrum. Several existing and upcoming applications will certainly benefit from the development of such lasers.