Visible femtosecond laser holds enormous potential for various applications in material processing, bioimaging, biomedicine, ultraviolet laser generation, and scientific research.1^–4 Particularly within the deep-red wavelengths, their superior tissue penetration and lower risk of photo-induced damage make them ideal for treating deep-seated tissue diseases, including cardiovascular, neurological, and ophthalmic conditions.5 Compared with traditional surgeries, procedures with deep-red femtosecond lasers offer greater precision, minimal invasiveness, and quicker recovery, which are favored by patients. The laser is valuable for exciting and imaging biomolecules [e.g., NAD(P)H, FAD, and keratin], enabling non-invasive diagnosis and therapeutic monitoring.3 In addition, deep-red wavelengths also exhibit great potential in two-photon microscopy (2PM), which enables label-free imaging of red blood cells6^–8 and efficient fluorescence excitations.9^,10 Currently, deep-red femtosecond lasers are mainly generated using Ti:sapphire femtosecond oscillators or nonlinear frequency conversion of near-infrared femtosecond lasers.11 Ti:sapphire systems are mature and cover the entire deep-red wavelengths but are costly and bulky. Frequency conversion of near-infrared lasers requires complex setups and compromises performance, limiting their adaptability. In comparison, passively mode-locked fiber lasers (MLFLs) are considered the ideal choice for producing femtosecond laser pulses due to their advantages in terms of miniaturization, high performance, and low cost;12^–15 however, achieving mode-locking operation at deep-red wavelengths remains arduous. Therefore, a significant drive exists to develop passively MLFLs operating in the deep-red spectral region.

Currently, the advancement of low-loss and high-performance rare-earth-doped fluoride fibers,16^–19 coupled with the progress in blue semiconductor pump laser technology, has led to the successful development of wavelength-tunable,20 continuous-wave,21^,22 or Q-switched23 deep-red fiber lasers, demonstrating the enormous potential of rare-earth-doped fluoride fibers (e.g., ${\mathrm{Pr}}^{3+}$ and ${\mathrm{Ho}}^{3+}$) as gain media for efficient deep-red laser generation. Furthermore, breakthroughs in passive mode-locking techniques in the visible region, including nonlinear amplifying loop mirror,24 nonlinear optical loop mirror,25 and nonlinear polarization rotation (NPR),26 have catalyzed the development of passively MLFLs encompassing the red,24^–29 orange,30 yellow,31 and green30^,32 regions. The achievement in the 635-nm red spectral region is particularly impressive, where fiber laser pulses with a duration as short as 168 fs27 and an output power reaching the watt-level28 have been successfully achieved. Despite notable advancements in the red wavelength range, femtosecond MLFLs still need to be expanded to other visible wavelengths (e.g., green, yellow, orange, and deep red). In addition, there is an ongoing demand for these lasers to achieve shorter pulse durations, targeting tens of femtoseconds or even few-cycle pulses. The ${\mathrm{Pr}}^{3+}$:ZBLAN fiber is a promising contender in this endeavor, showcasing an impressive gain bandwidth that extends over 25 nm in the deep-red region, coupled with lower dispersion at this wavelength, which suggests the potential for dispersion compensation. These characteristics confer substantial feasibility for achieving sub-100-fs visible ultrafast fiber lasers.

In this article, we present a deep-red passively MLFL based on the double-clad (DC) ${\mathrm{Pr}}^{3+}$:ZBLAN fiber and customized diffraction gratings. The laser achieves remarkably stable self-starting mode-locking utilizing the NPR and produces femtosecond pulses with a central wavelength of 716.6 nm. These pulses have a 3-dB bandwidth of 13.0 nm, a minimum pulse duration of 83 fs, and a repetition rate of 73.684 MHz. Manipulating the intracavity dispersion allowed us to observe multiple mode-locking states, including conventional soliton, dispersion-managed soliton, dissipative soliton, and bound-state soliton. We studied and analyzed the formation, evolution, and performance of the deep-red MLFL by solving the Ginzburg–Landau equation. The numerical results align excellently with our experimental data. To showcase the practical applications with this sub-100-fs 717-nm MLFL, 2PM imaging was demonstrated with outstanding performance regarding multicolor excitation, resolution, and high excitation efficiency. These findings represent a new paradigm in the generation of sub-100-fs visible fiber laser sources, with multiple potential applications in life sciences, biomedicine, and ultraviolet ultrafast generation.

The schematic and photograph of the deep-red MLFL are shown in Figs. 1(a) and 1(b). The laser utilizes a commercial 443-nm laser diode as its pump source with a maximum output power of 8 W. However, the laser beam exhibits asymmetric divergence, with measured $M^2$ values of 47.8 (horizontal) and 128.6 (vertical), indicating a strongly multimode spatial profile. The gain fiber employed is a 2.0-m DC ${\mathrm{Pr}}^{3+}$:ZBLAN fiber manufactured by Le Verre Fluoré, possessing the following specifications: a concentration of 6000 ppm (parts per million) for ${\mathrm{Pr}}^{3+}$, a core diameter of $5.5\mu \mathrm{m}$, a core numerical aperture (NA) of 0.08, an inner-clad profile in a double D-shaped configuration with dimensions of $115\mu \mathrm{m}\times 125\mu \mathrm{m}$, an inner-clad NA of 0.45, and an inner-clad absorption coefficient of $0.54{\mathrm{m}}^{-1}$ at 443 nm. The normalized frequency V-number of gain fiber core at 717 nm is $\sim 1.93$, ensuring single-transverse mode (${\mathrm{LP}}_{01}$) operation. In addition, it has a core group velocity dispersion (GVD) coefficient (${\beta }_2$) of $+0.0626{\mathrm{ps}}^2{\mathrm{m}}^{-1}$ (see Fig. S1 in the Supplementary Material) and a core nonlinear coefficient ($\gamma$) of $0.0048{\mathrm{W}}^{-1}{\mathrm{m}}^{-1}$ at 717 nm. The 443-nm pump laser passes through a visible dichroic mirror (DM) and then launches into the gain fiber using a collimator (Thorlabs, PAF2P-A10A, Newton, New Jersey, United States), thereby generating a deep-red light gain. The transition diagram of ${\mathrm{Pr}}^{3+}$ ions in the gain fiber is given as follows: (1) ${{}^3\mathrm{P}}_2$ level is excited by ground state absorption of the ${{}^3\mathrm{H}}_4$ level with the 443-nm laser, (2) nonradiative relaxation ensues from ${{}^3\mathrm{P}}_2$ to ${}^3{\mathrm{P}}_0$, and (3) stimulated radiative transition occurs from ${{}^3\mathrm{P}}_0$ to ${{}^3\mathrm{F}}_4$, generating emission at $\sim 717\mathrm{nm}$.21^,22 The DM is fabricated by coating ${\mathrm{SiO}}_2/{\mathrm{Ta}}_2{\mathrm{O}}_5$ dielectric films onto fused quartz substrates using an ion beam-assisted deposition system (DJ-800, Golden Vacuum). As shown in Fig. 1(c), the DM boasts a 98.6% reflectance at 717 nm and a 93.9% transmittance at 443 nm. Engineered to sustain $>95\%$ transmittance at 635 nm, it effectively suppresses red laser emission. Both ends of the gain fiber are angle-cut to prevent self-lasing of 635-nm red light caused by parasitic reflections detrimental to stable mode-locking operation. Light emitted from the gain fiber is collimated by another collimator, passed through a polarization beam splitter (PBS), and two DMs with high reflectivity at deep-red wavelengths and high transmittance at red wavelengths, which are then coupled back to the fiber by the collimator to form a ring cavity. A free space polarization-dependent isolator (PD-ISO) acts as a deep-red optical isolator to ensure unidirectional operation. A pair of transmission gratings (TGs) with high efficiency ($>90\%$ at 717 nm) and high groove density ($1398.6\text{lines}/\mathrm{mm}$) was used to manage intracavity dispersion. The TGs can provide a ${\beta }_2$ of $-2.95{\mathrm{ps}}^2{\mathrm{m}}^{-1}$ at 717 nm. Mode-locking is initiated and maintained by NPR via two half-wave plates (HWPs), a quarter-wave plate (QWP), and the PD-ISO. Output pulses extracted from the PBS are directed into the grating pair compressor. The compressor includes a D-shaped mirror (DSM), a hollow roof prism mirror (HRPM), and a pair of TGs with a groove density of $1398.6\text{lines}/\mathrm{mm}$. In addition, a DM was also used to filter out the remaining pump power (443 nm).

In alignment with the pulse shaping limit theory, achieving narrower laser pulses requires broadening the mode-locking spectrum, which can be accomplished by finely tuning the intracavity net dispersion toward near zero through control of the spacing of TGs. With the parameter optimized, the widest achievable spectrum is obtained when the spacing among the TGs is set to 40 mm, resulting in an intracavity net dispersion of $+0.005{\mathrm{ps}}^2$. Deep-red laser emission is detected in the optical spectrum analyzer once the pump power ($P_p$) surpasses 1.84 W. As the power increases, the pulse jitter becomes apparent on the oscilloscope. Upon reaching 3.11 W, self-starting mode-locking with multiple pulse states is observed. By reducing the $P_p$ to 1.96 W, stable single-pulse mode-locking is achieved. We recorded the output performance of the deep-red mode-locked laser at this time, as illustrated in Fig. 2. As shown in Fig. 2(a), the mode-locked spectrum boasts a central wavelength of 716.6 nm and a 3-dB bandwidth of 13.0 nm. In Fig. 2(b), a typical pulse train is depicted, exhibiting a uniform pulse intensity and a pulse interval of 13.6 ns. The inset of Fig. 2(b) accentuates the exceptional stability of the mode-locking operation. Figure 2(c) showcases the intensity autocorrelation trace of the output pulses, accurately fitted with a sech2 function, revealing a remarkably short pulse duration of 83 fs. In addition, the inset of Fig. 2(c) displays the intensity autocorrelation trace of the pulses extracted after PBS via a mirror, with a pulse duration of 2.43 ps. Figure 2(d) demonstrates the radio-frequency (RF) spectrum of the output laser, characterized by a fundamental frequency of 73.684 MHz and a high signal-to-noise ratio of $\sim 85\mathrm{dB}$. Notably, the inset of Fig. 2(d) underscores the stable mode-locking of the laser by revealing the absence of spectral modulation within the 3.6 GHz span RF spectrum.

To evaluate the operational stability of mode-locking, we monitored the evolution of the mode-locking spectrum and average power over time at a pump power of 1.96 W, as illustrated in Figs. 3(a) and 3(b). Throughout the testing period, both the center wavelength and spectral bandwidth exhibited remarkable stability, with no noticeable drift or variation. In addition, the standard deviation of mode-locked output power was 0.06 mW, indicating a calculated power fluctuation of 0.48%. These findings demonstrate the excellent long-term operational stability of MLFL. The $M^2$ values of the deep-red mode-locked laser were measured to be 1.20 and 1.18, respectively, indicating fundamental mode (i.e., ${\mathrm{LP}}_{01}$) operation. In addition, we recorded the output performance of the 716.6-nm mode-locked laser at various pump powers, as depicted in Figs. 3(c) and 3(d) and Fig. S3 in the Supplementary Material. Specifically, Fig. 3(c) presents the average power and pulse energy in relation to the pump power. Both the average power and pulse energy exhibit linear increases without any signs of saturation. Maximum average power of 31.7 mW was achieved at a pump power of 2.97 W, corresponding to a pulse energy of 0.43 nJ. Upon increasing the pump power from 1.96 to 2.97 W, the duration of laser pulses extracted from the PBS output narrows from 2.43 to 1.68 ps (see Fig. S3b in the Supplementary Material), whereas the corresponding dechirped pulse duration from DSM initially slightly broadens from 83 to 86 fs and then narrows to 84 fs [see Fig. 3(d)]. During this process, the peak power of laser pulses gradually increased from 1.62 to 5.12 kW.

In addition, to better understand the output performance of our deep-red MLFL, the mode-locking evolution with different dispersions is further explored, and the results are summarized in Fig. 4 and Table S1 in the Supplementary Material (see Fig. S4 in the Supplementary Material for the output performance with different pump powers). Figures 4(a) and 4(b) give a comprehensive overview of the optical spectra and autocorrelation traces of the pulses extracted after PBS via a mirror. As observed in Fig. 4(a), with the net dispersion varying from $+0.049$ to $-0.057{\mathrm{ps}}^2$, the spectral bandwidth of the mode-locked laser initially widens and subsequently narrows. It also indicates that the spectral bandwidth of the mode-locking broadens with the net cavity dispersion decreases to near zero. The widest spectral bandwidth is $\sim 13.0\mathrm{nm}$, which occurs at the net dispersion of $+0.005{\mathrm{ps}}^2$. The autocorrelation traces in Fig. 4(b) illustrate a decrease in pulse duration as the net dispersion changes from $+0.049$ to $-0.057{\mathrm{ps}}^2$, followed by a slow increase. The minimum value achieved at a net dispersion of $-0.024{\mathrm{ps}}^2$, corresponding to 614 fs. Due to the existing chirps, the durations of the mode-locked pulses outputted from the PBS vary slightly from the trend observed in the bandwidth of the mode-locking. However, a consistent trend with the mode-locked spectrum is observed in the corresponding autocorrelation traces of the dechirped pulses outputted from DSM [see Fig. 4(c)]. At a net dispersion of $+0.005{\mathrm{ps}}^2$, the minimum pulse duration is achieved, reaching as short as 83 fs. It is noteworthy that under the net dispersions of $+0.049{\mathrm{ps}}^2$ (blue curve) and $-0.057{\mathrm{ps}}^2$ (magenta curve), the mode-locked spectra exhibit typical characteristics of dissipative and conventional solitons, respectively. Collectively, these findings underscore the versatility of deep-red mode-locked laser, allowing for the customization of its output characteristics for diverse applications through precise control of net dispersion.

To gain further insights into the output performance and observe the establishment and evolution processes, we conducted a numerical simulation of the deep-red passively MLFL. The simulation model and parameters are elaborated in the Supplementary Material.33 Initially, we investigated the formation and evolution of deep-red MLFL with different net dispersions. The numerical simulation findings when the net dispersion is $+0.005{\mathrm{ps}}^2$ are presented in Figs. 5(a)–5(d) (see Figs. S5 and S6 in the Supplementary Material for the results of other net dispersions). As the roundtrip times increase, the initial pulse rapidly converges, culminating in the establishment of a stable pulse in the cavity [see Fig. 5(a)]. Due to the effects of GVD, self-phase modulation, and filtering action in the ring cavity, the pulse duration first increases and then remains nearly unchanged. The corresponding spectral evolution is depicted in Fig. 5(b). Finally, even if the roundtrip number continues increasing, the pulse, spectrum, and peak power remain almost constant, evidencing the establishment of stable mode-locking. Figure 5(c) illustrates the mode-locked pulse profile (solid) and the associated frequency chirp (dashed), revealing two different chirp features along the pulse profile and a pulse duration of 2.33 ps. The spectra, both on a logarithmic scale (solid) and a linear scale (dashed), of the 716.5 nm mode-locking are shown in Fig. 5(d). The spectrum displays a bandwidth of 13.92 nm and a Gaussian profile with steep edges, characteristic of a typical dispersion-managed soliton. Interestingly, the simulation results are in excellent agreement with the experimental results in Fig. 2, indicating the accuracy of the numerical model.

Next, we simulated the performance of mode-locking at different positions within the cavity. The numerical results when the net dispersion is $+0.005{\mathrm{ps}}^2$ are depicted in Figs. 5(e)–5(g) (see Figs. S7 and S8 in the Supplementary Material for the results of other net dispersions). Figures 5(e) and 5(f) illustrate the one-roundtrip evolution of the spectrum and pulse within the cavity, respectively. Correspondingly, Fig. 5(g) illustrates the one-roundtrip evolution of the spectral bandwidth and pulse duration. Upon closer examination of Fig. 5(g), it becomes apparent that both the spectral bandwidth and pulse duration initially decrease and then broaden within the DC ${\mathrm{Pr}}^{3+}$:ZBLAN fiber segment, mainly attributed to the ongoing compensation of the normal dispersion in the fiber. Thereafter, influenced by the saturable absorption (SA) effect, both parameters undergo a slight narrowing. As the pulse traverses the TGs, the spectral bandwidth maintains relative stability, whereas the pulse duration widens as a result of the reaccumulation of significant anomalous dispersion from the TGs. In addition, the simulations of MLFL with different net dispersions, depicted in Fig. S9 in the Supplementary Material, revealed the narrowest pulse duration at near-zero dispersion, with results closely matching the experiments, highlighting the critical role of dispersion management in achieving minimal duration.

Among multiple applications of the 717-nm MLFL, biophotonic imaging is a significant field for life sciences and biomedicine, where the imaging performance is largely determined by the characteristics of the laser source. Here, we employed this deep-red passively MLFL to perform two-photon-excited fluorescence imaging. We demonstrated great advantages of the 717-nm MLFL in 2PM imaging from three aspects, including the multicolor excitation, resolution, and high excitation efficiency.

First, the 717-nm laser was utilized to perform the three-color two-photon imaging of a mouse kidney section, stained with Alexa Fluor 488 (AF488), Alexa Fluor 568 (AF568), and DAPI (D-1306). As the 717-nm wavelength is located at the intersection of the two-photon excitation spectra of the three dyes, it can efficiently excite them simultaneously, as demonstrated in Fig. 6(a). The three-color images were obtained by isolating the emission of the three dyes through different optical filters, that is, a short-pass filter (SP500, JCOPTIX, OFE1SP-500) for DAPI, a bandpass filter (BP520, JCOPTIX, OFBH140-520) for AF488, and a long-pass filter (LP590, PHTODE) for AF568, as shown in Fig. 6(b). Figure 6(c) shows the fluorescence images of DAPI, AF488, and AF568 dyes, whereas their composite image is shown in Fig. 6(a). As can be observed, the cell nuclei are clearly marked by DAPI, and the elements of the glomeruli and convoluted tubules can be easily identified by AF488, whereas filamentous actin prevalent in the glomeruli and brush border are highlighted by AF568.

Second, we evaluated the lateral resolution of the 2PM with the 717-nm MLFL. According to the wave property of light, a shorter wavelength contributes to a smaller beam focus, resulting in a better resolution. Based on the resolution equation of the 2PM,36 the theoretical value of the lateral resolution was calculated to be 302 nm based on the objective NA of 0.9. The point spread function (PSF) of 717-nm 2PM was measured to be $359\pm 17\mathrm{nm}$ based on several measurements at different regions of the field of view​ (FOV) . A typical PSF in our experiments is illustrated in Fig. 6(d), and its full width at half maximum is 382 nm. The higher value of the experimental PSF than the theoretical one is mainly attributed to the unfilled back aperture of the objective, resulting in a slightly low effective NA. Compared with the conventional excitation wavelengths at 920 or 1064 nm, the theoretical 2PM resolution is enhanced by $\sim 22\%$ or $\sim 33\%$, respectively.

Third, the 717-nm MLFL was showcased to excite the mCherry-labeled Caenorhabditis elegans. It was found that the two-photon excitation efficiencies of a series of red-emitting fluorescent proteins (RFPs),10 including the mCherry, tdTomato, and mStrawberry, were promisingly greater or comparable compared with the conventional excitation wavelength, i.e., 1064 nm, as shown in Fig. 6(e). For mCherry, the two-photon excitation efficiency at $\sim 717\mathrm{nm}$ is nearly four times higher than that at $\sim 1064\mathrm{nm}$. Figures 6(f) and 6(g) exhibit clear images of the mCherry-labeled intestinal lumina of the C. elegans.

Here, we have successfully developed a passively MLFL operating in the deep-red spectral waveband and showcased its promising applications in 2PM.

Leveraging the NPR mechanism, the laser exhibits remarkably stable self-starting mode locking, generating laser pulses with a central wavelength of 716.6 nm, a maximum 3-dB bandwidth of 13.0 nm, a minimum pulse duration of 83 fs, and a repetition rate of 73.684 MHz. Notably, the laser system demonstrates excellent long-term stability, characterized by a low-power deviation of 0.32% and negligible wavelength drift. By meticulously adjusting the intracavity gratings, we can effortlessly transition the cavity among anomalous, near-zero, and normal dispersion regimes, enabling the generation of conventional solitons, dispersion-managed solitons, and dissipative solitons. This dispersion management not only serves as a platform for exploring the rich soliton dynamics in the deep-red region but also allows for precise control over critical laser parameters, such as pulse duration and spectral bandwidth. Furthermore, through numerical simulations using the split-step Fourier method to solve the Ginzburg–Landau equation, we have studied and analyzed the formation, evolution, and performance of the deep-red MLFL. The experimental outcomes align closely with our numerical predictions, validating the accuracy of our simulations. Our soliton MLFL represents a significant advancement in the sourcing of femtosecond laser pulses in the deep-red region, particularly notable for its impressive 83-fs pulse duration.

Although modern Ti:sapphire or optical parametric oscillator-based two-photon systems can indeed tune across $\sim 700$ to 1100 nm, those systems are complex and costly. Our work demonstrates a compact fiber laser source at 717 nm that fills a niche at the short end of this range. The 717-nm MLFL enables simultaneous multicolor imaging and greatly higher excitation efficiency for a series of RFPs in 2PM while promisingly providing a much better resolution. The three-color excitation covers a wide range of fluorescent dyes from the blue, green, to red ones, indicating the possibility for efficiently exciting abundant fluorescent dyes with this portable laser source. Furthermore, the much higher 2P excitation efficiency for multiple RFPs, especially the mCherry, will contribute to brighter images in 2PM or enable us to excite the dyes/proteins with much less laser power for minimizing the photo-blenching and photo-damage effects. Finally, the 717-nm laser inherently provides us with a better resolution in 2PM owing to its short wavelength. If a 1064-nm laser aims to achieve the same lateral resolution as the 717-nm laser, an oil-immersed objective lens of $\sim 1.4$ NA was required instead of the 0.9-NA water-immersed one, where the latter is easy and cheap to get considering the working distance and imaging quality. The outstanding and unique 2PM imaging performance confirms the versatile and reliable biophotonic applications enabled by this deep-red MLFL.

Fluorophores with significant one-photon absorption in the UV/blue range or far-red emission profiles are well-suited for 717-nm excitation. This includes ultraviolet-excited dyes such as DAPI (demonstrated in Fig. 6) and potentially blue fluorescent proteins or other dyes with strong absorption near 350 to 400 nm. Likewise, as we have shown, many RFPs and dyes have appreciable two-photon excitation cross-sections around 700 to 720 nm, making 717 nm a favorable excitation wavelength for them. Another important consideration is the imaging depth. Owing to the stronger scattering of 717-nm light in biological tissues (relative to the commonly used 800 to 1100 nm two-photon excitation range), the maximum imaging depth achievable with our deep-red laser will be inherently limited compared with longer-wavelength sources. In practice, we expect the 717-nm excitation to be most effective for superficial imaging up to a few hundred micrometers deep, depending on tissue optical properties. This trade-off accompanies the enhanced resolution and multicolor excitation capability provided by the shorter wavelength. Therefore, our 717-nm fiber laser is ideal for high-resolution, multicolor imaging of shallow tissue regions or thin specimens.

In summary, this breakthrough in ultrafast laser technology bridges a crucial gap and opens up new possibilities for scientific research, medical applications, and biological imaging. We anticipate this femtosecond fiber laser will spark further activities and inspire innovative applications in diverse fields.

The online version contains Supplementary Material available at https://doi.org/10.1117/1.AP.7.4.046009.s01.

Jinhai Zou received his PhD in electronic science and technology from Xiamen University in 2022. He is currently serving as an associate researcher at the National University of Defense Technology. His research concentrates on visible fiber lasers and their applications.

Hongsen He received his PhD in 2022 from the Department of Electrical and Electronic Engineering at the University of Hong Kong, with a major in biophotonics. He is currently an assistant professor at the School of Electronic Science and Technology, Xiamen University. His research focuses on the invention and application of imaging lasers and microscopy techniques, including the multiphoton and photoacoustic microscopy.

Zhengqian Luo received his PhD in electrical engineering from Xiamen University in 2009. From 2007 to 2009, he studied at the Nanyang Technological University as a joint PhD student. He joined Xiamen University as a faculty member in 2010. From 2016 to 2017, he was a visiting professor at the Massachusetts Institute of Technology. Since 2017, he has been a full professor at Xiamen University and currently serves as the head of the Electronic Engineering Department of Xiamen University. His research focuses on ultrafast fiber lasers and on-chip photonic devices.

Biographies of the other authors are not available.