1. Introduction

High-average-power, high-repetition-rate fiber amplification systems have attracted increasing attention in industrial applications and scientific research due to their outstanding reliability, compactness, and user-friendly operation [1,2]. Particularly, Yb³⁺-doped fibers are commonly used thanks to their high efficiency, high absorption and low defects [3,4]. Such Yb³⁺-doped fiber femtosecond laser sources show great potential in fast manufacturing and driving high-photon-flux table-top coherent extreme ultraviolet sources [5,6]. However, the pulse duration is limited to several hundreds of femtoseconds, owning to the finite gain bandwidth of Yb³⁺-doped fibers. The long pulse duration and high repetition rate will induce thermal accumulation, which is undesirable in some applications, especially for transparent material micromachining and biophotonics [7]. For example, thermal accumulation can cause photodamage to cells in biophotonics research, where the interaction process is highly thermal sensitive. So high-peak-power sub-100-fs laser pulses with a low repetition rate (< 1MHz) are highly desirable. The reason is that a shorter pulse duration with a constant peak power corresponds to less energy deposition and a smaller photodamage region [8]. Additionally, the larger pulse interval can effectively avoid thermal accumulation [9].

The pulse output from a Yb³⁺-doped fiber chirped-pulse amplification (FCPA) system is typically longer than 200 fs [10]. To obtain sub-100-fs pulses, nonlinear amplification is always adopted [11,12], but the nonlinearity accumulation in amplification degenerates the pulse quality. To solve this problem, pre-chirp control [13,14], compensation between the third-order dispersion and nonlinear phase shift [15–17] and self-similar amplification [18,19] have been utilized to obtain short pulse duration and high quality at the same time. But there are few reports on sub-100-fs fiber amplification systems with a low repetition rate.

In this paper, we demonstrate a hybrid femtosecond fiber laser amplification system with both CPA and nonlinear amplification techniques to generate 1-μJ, 24-fs transform-limited (TL) pulses at 1 MHz repetition rate. To the best of our knowledge, it is the shortest pulse with transform-limited pulse quality from fiber nonlinear amplifiers. A CPA structure is employed in the pre-amplification stage, providing enough pulse energy for the main-amplifier. Particularly, the stretched pulse in pre-amplification is immune to nonlinear phase shift accumulation, which can aggravate spectral modulation of the signal pulse in the main amplifier and result in degradation of the compressed pulse quality [20]. The main amplifier utilizes nonlinear amplification for energy scaling as well as output spectral broadening. To improve the quality of the dechirped pulse from nonlinear amplification, a pre-chirper has been frequently used to adjust the pre-chirp of the signal pulse [13,14]. Here, a much shorter pulse duration and a better pulse quality rely on a combined optimization of the signal pulse parameters in terms of pre-chirp, pulse energy and central wavelength. On the basis of this investigation, we can optimize the amplification system to generate 1-MHz, sub-ten-cycle (<30 fs) TL laser pulses over a large range of output pulse energies. In this system, all the fibers are conventional polarization-maintaining single-mode fiber (PM-SMF) or double-cladding Yb³⁺-doped fiber without any large-mode-area Yb³⁺-doped photonic crystal fiber, so that different parts can be connected by fiber fusion. It makes the system more compact and practical. To demonstrate the advantages of this hybrid fiber femtosecond laser source, we use it to study the enhanced damage effect to cells co-cultured with gold nanorods (GNRs) [21]. It is found that the shorter pulse duration significantly reduces the thermal effect and contributes little photodamage to cells, although the corresponding peak power is higher.

2. Experimental setup and results

The experimental scheme is shown in Fig. 1. The oscillator is a dissipative-soliton mode-locked Yb³⁺-doped fiber laser operating in the all-normal-dispersion regime. It generates ~3-ps laser pulses with a 49-mW average power at 48-MHz repetition rate. The spectrum and auto-correlation (AC) trace of the dechirped pulse are shown in Fig. 2(a). A 50-m-long PM SMF stretcher is utilized to broaden the pulse to ~30 ps. Due to the decreased peak power, the nonlinear phase shift accumulation is negligible in the pre-amplifiers. So the spectral bandwidth of the pulse after the second pre-amplifier stage is identical to that in the oscillator, and the temporal fidelity does not degrade much, although the spectral shape is changed a little (as shown in Fig. 2(b)). Between the two pre-amplifiers, an AOM with input/output pigtails (PM-SMF) is equipped to reduce the repetition rate from 48 MHz to 1 MHz. The output pulse of the second pre-amplifier stage is coupled directly into the pre-shaper, which can adjust the pre-chirp, central wavelength and pulse energy of the signal pulse for the following main-amplifier. The pre-shaper consists of a 1200 lines/mm blazing grating pair, 12-nm bandpass filter, half-wave plate and polarizing beam splitter. By adjusting the distance of the grating pair, the group delay dispersion of the stretcher and the materials in the pre-amplifiers can be compensated, and spacing them further apart will flip the chirp. So the chirp of the signal pulse can be tuned from negative to positive. The filter alters the central wavelength by tuning the beam incident angle. And the pulse energy can be varied by rotating the half-wave plate before the PBS. In the main-amplifier, the gain fiber is a 2.2-m polarization-maintaining large-mode-area double-cladding Yb³⁺-doped fiber (PLMA-YDF-30/250-VIII, Nufern) with a 30-μm core diameter, which is counter-directionally end-pumped by a 976-nm laser diode. To suppress the high-order modes, the gain fiber is coiled with a radius of ~10 cm. After amplification in the main amplifier, the pulse is compressed by a transmission grating pair of 1000 lines/mm, with a compression efficiency of ~74%.

 

Fig. 1 Schematic of experimental setup, QWP: quarter-wave plate, HWP: half-wave plate, BF: bandpass filter, DM: dichroic mirror, WDM: wavelength-division multiplexer, PBS: polarizing beam splitter, LD: laser diode, and AOM: acoustic-optical modulator.

Download Full Size | PDF

 

Fig. 2 (a) Spectrum and AC trace (inset) of the dechirped seed pulse from the oscillator. The dechirped pulse duration is ~200 fs. (b) Spectrum and AC trace (inset) of the dechirped pulse from the second pre-amplifier before the pre-shaper. The dechirped pulse duration is 230 fs.

Download Full Size | PDF

In the nonlinear amplification, the pulse evolution is determined by self-phase modulation, dispersion and gain together. So the pre-chirp and energy of the signal pulse play important roles in the nonlinearity of amplification [22]. By controlling these two parameters, the amplification can be optimized not only to output a broadened spectrum for supporting short dechirped pulses, but also to provide a linear chirp in most part of the pulse for high temporal quality after compression. There is an optimum negative pre-chirp of the signal pulse to achieve the best compression. However, in the spectral broadening process, the gain shaping effect due to the finite gain bandwidth induces asymmetry in the output spectrum, resulting in the degradation of dechirped pulse quality [23], especially when the input central wavelength does not correspond to the gain peak, which depends on fiber length, doping concentration and pump power. In order to match it to the gain peak, we tune the central wavelength of the signal pulse using the pre-shaper.

To demonstrate nonlinear amplification optimization by adjusting these three parameters of the signal pulse, especially the correlation of the central wavelength with the gain peak, the negative pre-chirp and energy of the signal pulse are both optimized at different input central wavelengths. In the experiment, the distance between the two transmission gratings of the compressor is tuned to compress the amplified pulse to the shortest pulse duration. Then, the AC trace is measured by an autocorrelator (APE Pulse Check). Furthermore, the temporal envelope of the compressed pulse is retrieved by the phase and intensity from correlation trace and spectrum only (PICASO) algorithm [24]. Based on the retrieved pulse intensity, we employ the Strehl ratio [25] to assess the temporal quality, which is defined as the ratio of the peak power of the compressed pulse to that of the TL pulse. A Strehl ratio approaching 1 means that the compressed pulse is close to the TL pulse, and the chirp of the uncompressed pulse is nearly linear.

The Strehl ratio curve of the dechirped pulse is shown in Fig. 3. Three typical retrieved intensity profiles are presented in the insets of Fig. 3, and compared with the corresponding TL pulses, which are inversely transformed from the measured spectra with ideally zero spectral phase. Just as analyzed above, there is an optimal central wavelength for the Strehl ratio curve. When the central wavelength of the signal pulse is tuned to ~1040 nm, we can obtain a nearly TL compressed pulse. However, the dechirped pulse qualities at other central wavelengths are worse. The reason is simply that the center of the gain spectrum of this gain fiber is ~1040 nm at this gain level. When the central wavelength of the signal pulse deviates from it, the gain shaping effect becomes more obvious.

 

Fig. 3 Strehl ratio versus central wavelength. Insets (a)~(c): PICASO-retrieved dechirped pulse profiles (colored lines) and corresponding TL pulse profiles (black lines) for different central wavelengths; inset (d): output spectra from the pre-shaper, corresponding to the compressed pulses shown in insets (a)~(c).

Download Full Size | PDF

In the experiment, the gain peak alters with the pump power [26]. So it is necessary to tune the central wavelength of the signal pulse in order to obtain a short pulse of high quality over a large range of pump powers, as shown in Fig. 4(a). The pulse duration decreases as the pump power increases, and the pulse is shorter than 30 fs when the output pulse energy is higher than 0.6 μJ. Meanwhile, with careful optimization, the Strehl ratio is higher than 0.9, which means that the temporal shape is nearly TL. It can be verified from Fig. 4(b) that the temporal intensity profiles of the pulses with 0.6-μJ and 1-μJ pulse energies are TL. However, the temporal pedestal of the 0.6-μJ pulse is smaller than that of the 1-μJ pulse, leading to a higher ratio of the main-pulse energy to the whole pulse energy. This energy ratio is 89% for the 0.6-μJ pulse and 74% for the 1-μJ pulse. The reason is that the Strehl ratio only shows the degree of approaching to a TL pulse, indicating the linearity of chirp of the uncompressed output pulse, but it does not reflect the pedestal of the TL pulse. To analyze the temporal quality comprehensively, we use the root-mean-square (RMS) width to illustrate the influence of the pedestal. When the FWHM duration is unchanged, an increasing RMS width means the pedestal becoming more and more obvious. As shown in Figs. 4(a) and (c), the RMS of the dechirped pulse increases significantly when the output pulse energy is higher than 0.6 μJ, but the FWHM duration decreases. Considering the fact that the dechirped pulses are nearly TL in this range, it means that the pedestals of the TL pulses rise and degrade the pulse quality when the output pulse energy increases. These pedestals are mainly caused by Raman scattering, which broadens the output spectrum and increases the power ratio in the spectrum beyond 1100 nm when the output pulse energy becomes higher. We can conclude that the best temporal quality of the compressed pulses can be obtained when the output pulse energy is 0.6 μJ. However, the pedestal rises up with increased pulse energy, although the pulse duration is shorter. When the output pulse energy is 1 μJ, pulses with a 24-fs pulse duration are generated, but their quality is not as good as that for pulses with 0.6-μJ energy.

 

Fig. 4 (a) Pulse duration and Strehl ratio versus pulse energy; (b) Dechirped pulse profile and TL pulse profile at 0.6 μJ (left) and 1 μJ (right); (c) RMS width versus pulse energy; (d) Dechirped pulses and spectra (inset) at different pulse energies.

Download Full Size | PDF

For many applications, the power stability and the beam quality are crucial parameters. Figure 5(a) shows the long-term power stability of the system. The output power varies only by about 0.22% RMS in one hour. The amplifier output beam is characterized by measuring the beam radius at varying distance from the waist. The best-fit M² value is 1.14, which confirms that the beam is single mode. The results indicate that this fiber amplifier is reliable in a lab level working environment.

 

Fig. 5 (a) Long-term power stability, which reaches 0.22% RMS in one hour; (b) The beam radius with respect to the distance from waist (the blue circle refers to the measured data and the red line is the fit of beam radius ω to $\omega ={\omega }_0{\left(1+M^4{\lambda }^2{z}^2/{\pi }^2{\omega }_0^4\right)}^{1/2}$, where λ is the wavelength and ω₀ is waist radius, the inset shows the beam profile at the waist.

Download Full Size | PDF

3. Investigation of photodamage to cells

By optimizing the parameters of the pre-shaper, we can obtain ~30-fs pulses with high quality and a low repetition rate. The pulse temporal duration can be tuned by changing the distance between the gratings in the compressor. With this fiber laser source, we can investigate the role of pulse duration and peak power in photodamage caused by femtosecond laser to cells via GNRs.

GNRs show low toxicity to cells and their resonance wavelength can be easily controlled by the size and shape [21]. Due to their nano-level size, GNRs can stain in cells with minimal damage and enhance the response to photons in the NIR range. In this paper, GNRs (65 nm × 10 nm, Nanopartz Inc.) with surface plasmon resonance peak at around 1045 nm are co-cultured with HeLa cells for 36 hours. The concentration of GNRs is around 100 pM. By fluorescently labeling the mitochondrial membrane potential (MMP) with Tetramethylrhodamine methylester (TMRM), a standard indicator of cell viability whose fluorescence will decrease due to the MMP depolarization when the cells are going to die, we carefully examine the MMP after co-culture and find no MMP depolarization. Therefore, the toxicity of GNRs at such a low concentration will generally have little influence on the cells.

Subsequently, cells co-cultured with GNRs are irradiated by 0.5-W femtosecond laser pulses with pulse durations of 34 fs, 134 fs and 600 fs, respectively. The diameter of the irradiation area is 1 mm. The cell damage is indicated by the TMRM fluorescence level. As shown in Fig. 6, cells irradiated by 34-fs laser pulses suffer little damage. With the pulse duration increased, cell damage was more and more severe. Laser pulses of 34 fs, corresponding to 15-MW peak power, can induce very little thermal effect. But if the pulse duration is stretched to 600 fs (the peak power < 1 MW), the MMP of the irradiated cells is significantly depolarized, suggesting great damage for cell death. The peak power here contributes little to the thermal effect mediated by GNRs. In contrast, the longer pulse duration allows more heat deposition inside a single pulse and less thermal diffusion during pulse intervals, resulting in higher thermal accumulation. Therefore, the pulse duration plays the essential role in cell damage by photo thermal effect rather than peak power [7]. It can be found that the pulse duration at around 130 fs, a level widely used in two-photon microscopy, is still moderately safe in the presence of GNRs as long as the irradiation duration is less than one minute. The long pulse duration, such as 600 fs, rather than the peak power, contributes the major part to heat the GNRs.

 

Fig. 6 Cell viability after laser irradiation with GNRs for different durations. (a) MMP fluorescence before and 60 s, 120 s, and 360 s after laser irradiation with different pulse widths. Fluorescence decrease indicates cell damage. (b) Normalized MMP fluorescence level compared with controls that are irradiated with lasers without GNRs. Bar scale in (a): 20 μm.

Download Full Size | PDF

4. Conclusion

In this paper, we have demonstrated a hybrid femtosecond fiber laser amplification system. In the pre-amplifier stage, the CPA technique is applied to alleviate nonlinearity accumulation to avoid the aggravation of spectral modulation of the signal pulse for the main amplifier. Nonlinear amplification is used in the final stage to broaden the output spectrum, supporting short pulse duration. In order to obtain ~30-fs laser pulses with high quality, a pre-shaper is employed to control the pre-chirp, the central wavelength and the pulse energy of the signal pulse in the main amplifier, thus allowing pulse evolution process to be optimized. With the aid of the pre-shaper, this fiber laser source generates ~30-fs, 1-MHz, and high-quality pulses over a large range of output pulse energies. When the output pulse energy is 1 μJ, the pulse duration is 24 fs, which, to the best of our knowledge, is the shortest pulse duration with TL temporal quality from nonlinear fiber amplifiers. Then, we use this fiber laser source to study the enhanced damage effect to cells co-cultured with GNRs. It is found that a shorter pulse duration induces less thermal accumulation corresponding to less cell photodamage, although the peak power is much higher. So the peak power plays a little role in the thermal effect mediated by GNRs. In the future, the pre-shaper can be built from fiber elements to support all-fiber femtosecond laser amplification systems, which will be more reliable and compact, and also turn-key operated.