The development of ultrafast laser technology has become one of the most cutting-edge domains and hot points in photonics and even modern science. Ultrafast laser technology is expected to make a breakthrough in numbers of fields and plays an important role in promoting the future development of science and technology. As one of the greatest inventions in the 20th century, the development of lasers has gone through nearly 60 years since the first, to the best of our knowledge, demonstration of coherent light generated from a flashlamp-pumped ruby crystal^[1]. Many other types of lasers emerged soon after, including gas lasers^[2], glass lasers^[3], and semiconductor lasers^[4]. It was not until the 1980s that stable passive mode-locking with a femtosecond (fs) pulse was demonstrated using a dye saturable absorber (SA) in a laser cavity^[5]. In the early 1990s, Ti:sapphire mode-locked lasers with the Kerr-lens effect infinitely promoted the rapid development of fs laser technology, mainly manifesting in the huge reduction of pulse durations and the sharp increase of peak powers. The pulse durations have decreased from the original 100 fs to nearly sub-10-fs and even further narrowed to the astonishing magnitude of attosecond (10⁻¹⁸ s) via high-order harmonics^[6–10]. Meanwhile, the peak power has been increased from megawatt (MW) to more than several petawatts (PW, $1\mathrm{PW}={10}^{15}\mathrm{W}$)^[11,12]. Subsequently, a large range of revolutionary applications have been demonstrated to take full advantages of ultrafast lasers (e.g.,  fs spectroscopy, optical frequency comb, ultrafast and high-resolution imaging). These applications prove that fs laser technology is of great significance to fundamental scientific research^[13–17]. Further, ultrafast laser technology has been widely used in industries, for example, lighting and displaying, remote sensing and environmental monitoring, biology and medical diagnosis, communication and national defense, and advanced manufacturing^[17–20].

Mode-locking, which usually incorporates either active or passive pulse modulators, is a commonly used technique to generate ultrashort pulses. More specifically, active mode-locking requires an external signal normally driven by an acousto-optic or electro-optic modulator. A typical external sinusoidal signal periodically modulates the intracavity loss, the modulation rate of which is integer multiples of the cavity frequency^[20]. A lithium-niobite-crystal-based electro-optic modulator can easily achieve several hundreds of gigahertz (GHz) repetition rate, which is desirable for high-capacity optical communication systems^[21–25]. However, the pulse widths of actively mode-locked lasers are usually in the picosecond region. Passive mode-locking refers to the situations that ultrashort pulses are formed through nonlinearities inside the cavity and can easily achieve fs pulses. For passive mode-locking, nonlinear optical modulators, named SAs, whose response to an entering optical pulse is intensity dependent, are used to obtain self-amplitude modulation. The earliest fs lasers based on SAs were realized in dye lasers for tens of fs pulses generation^[26–28]. However, the dye lasers suffer from complex structure and poor environmental stability. In addition, the organic dyes are poisonous and require recycling, which limit their generalization. Then, self-mode-locking (i.e., Kerr-lens mode-locking), generated from the third-order nonlinearity of the gain medium, emerges as another new mode-locking mechanism for fs pulses generation^[29–31]. Although Kerr-lens mode-locking does not require a complex modulator, it depends on external disturbance for self-starting and is extremely sensitive to misalignment. The development of semiconductor epitaxy technology and semiconductor bandgap engineering makes semiconductor SA mirrors (SESAMs) the hottest candidates for passive mode-locking^[32]. The precisely controllable fabrication of SAs is a huge leap forward in passively mode-locked lasers. Nowadays, SESAMs have been commonly adopted in commercial mode-locked laser systems due to their validity and reliability at specific wavelengths. Nevertheless, the fabrication of SESAMs requires complex procedures as well as expensive equipment. In addition, SESAMs are bulky and inflexible, and need spatial alignment, which is inconvenient for fiber laser systems. The limitations of SESAMs have triggered the development of novel low-dimensional materials (LDMs) SAs, which can be easily integrated into various configurations. As strong contenders for SESAMs, LDMs have already been widely applied in ultrafast laser systems^[33–38].

Here, we review the recently studied advances of SAs in terms of fabrication and applications. In Section 2, we mainly introduced the fundamental properties of the most widely studied semiconductor materials including SESAMs, carbon nanotubes (CNTs) and graphene, and transition metal dichalcogenides (TMDCs). This was then followed by the fabrication of the SAs and their ultrafast laser applications both in free-space and fiber laser systems. In the last section, we discussed the pros and cons of the present SA fabrication methods and ended up with perspectives on LDMs-based SAs.

Saturable absorption properties are of great significance for passively mode-locked lasers. Pulse operation initiates from the random noise sequence in the laser cavity. When an optical pulse propagates through the SA, its wings experience more loss than the central part, because high intensity more easily saturates the absorber. As a result, the loss modulation is synchronized automatically with the laser pulses. The Pauli blocking effect can explain the saturable absorption phenomenon at the micro-level. When high-intensity light interacts with a nonlinear optical material, electrons in the valence band absorb the incident photons and transit to a higher conduction band. For low-incident intensity, most of the photons are absorbed, hence, resulting in low transmission. Under high incident beam intensity, the conduction band is full with electrons and cannot accept more electrons due to the Pauli blocking effect. Consequently, most of the photons transit through the material resulting in high transmission. The basic mechanism for mode-locking can be easily understood by considering the SA as a passive optical modulator whose absorption (loss) is intensity-dependent, which can be described by a two-level model:$$\alpha (I)=\frac{{\alpha }_{\mathrm{sat}}}{1+I/I_{\mathrm{sat}}}+{\alpha }_{\mathrm{non}-\mathrm{sat}},$$(1)where ${\alpha }_{\mathrm{sat}}$ is the modulation depth (saturable loss), ${\alpha }_{\mathrm{non}-\mathrm{sat}}$ is non-saturable loss, and $I_{\mathrm{sat}}$ is the saturation intensity (required intensity to reduce the absorption by 0.5${\alpha }_{\mathrm{sat}}$). In this section, we mainly focus on the fundamental properties of the most concerned SAs (SESAMs and LDMs), especially their nonlinear optical properties.

Since 1990, Keller et al. have attempted to implement semiconductor SAs into lasers for self-starting mode-locking^[39]. Experimental reports have demonstrated that stable mode-locking can be obtained by deliberately designing macroscope parameters of SESAMs (such as carrier lifetime, saturation energy, and modulation depth)^[21]. Dynamics of the SESAM are complex, in that the pulse evolution must be treated with a modified Ginzburg–Landau equation with several nonlinear non-conservative terms^[22]. The saturable loss of a SESAM is given by$$\frac{\partial {\delta }_s(t)}{\partial t}=-\frac{{\delta }_s(t)-\mathrm{\Delta }R}{{\tau }_{\mathrm{rec}}}-\frac{{|\phi (t)|}^2}{E_{\mathrm{sat}}}{\delta }_s,$$(2)where $\mathrm{\Delta }R$ is the saturable loss of an unexcited absorber (modulation depth), ${|\phi (t)|}^2$ is the optical field intensity, ${\tau }_{\mathrm{rec}}$ is the absorption recovery time, and $E_{\mathrm{sat}}$ is the absorber saturation fluence. The solution of Eq. (2) is$${\delta }_s(t)=\frac{\mathrm{\Delta }R}{f(t)}[\frac{1}{{\tau }_{\mathrm{r}\mathrm{e}\mathrm{c}}}\int f(t)\mathrm{d}t+1],$$(3)where$$f(t)=\exp \{\int [\frac{1}{{\tau }_{\mathrm{rec}}}+\frac{{|\phi (t)|}^2}{E_{\mathrm{sat}}}\left]\mathrm{d}t\right\}.$$(4)

For practical purposes, we assume that ${\tau }_{\mathrm{pulse}}<<{\tau }_{\mathrm{rec}}$, where ${\tau }_{\mathrm{pulse}}$ is the pulse duration and the incident pulse energy $E_p$ is proportional to ${\int }_{-\infty }^{\infty }{|\phi (t)|}^2\mathrm{d}t$. Therefore, ${\delta }_s\approx \mathrm{\Delta }R{e}^{-E_p/E_{\mathrm{sat}}}$, and the effective average loss for this pulse can be simplified as ${\delta }_p=\mathrm{\Delta }R(1-e^{-E_p/E_{\mathrm{sat}}})(E_p/E_{\mathrm{sat}})$. The mirror reflectivity is $R=1-{\delta }_p-{\alpha }_0$, where ${\alpha }_0$ is non-saturable loss. SAs can be divided into slow and fast ones. Most of the real SAs are slow ones with recovery time at the nanosecond regime, while the fast SA (like Kerr lens) has a recovery time as fast as the fs regime. A previous report has shown that a slow SA can generate pulses much shorter than the recovery time of the absorber^[40]. However, the fast SA is more stable against multi-pulse in a high-power laser cavity^[41]. Reference [42] provides a comprehensive discussion on the relationship between pulse duration and recovery time based on analytical and numerical calculation methods.

Besides, saturation fluence plays an important role in the reduction of the mode-locking threshold and the avoidance of optical damage. The saturation energy of commercial SESAMs is typically centered between $50\text{\hspace{0.17em}and\hspace{0.17em}}100\mathrm{\mu }{\mathrm{J}/\mathrm{cm}}^2$. Keller et al. have presented two novel SESAM designs to decrease the saturation energy^[43]. Low saturation energy is also suitable for ultra-high repetition rate (i.e., $\gg 1\mathrm{GHz}$) operation^[43]. In addition, a multi-layer quantum well (QW) can be chosen to obtain a large modulation depth, whereas the modulation depth should not be too high to avoid the $Q$-switching operation^[44]. Moreover, several approaches have been adapted to improve the damage threshold, such as careful design of an anti-resonant cavity, expanding the input spot size, coating a protecting dielectric film on the surface of the SESAMs, or adopting quantum dots (QDs) SESAMs^[41,45,46].

The unique lattice structure of carbon-based LDMs (i.e., CNTs, graphene) endows them with superior linear and nonlinear optical properties. The explosion in research of CNTs was initiated by the group of Iijima in 1991^[47]. The angle that the sheet rolls to form a tube determines the electronic properties of the CNTs, and the band energy of the CNTs is inversely proportional to the diameter^[48,49]. Experiments have demonstrated that CNTs with semiconducting behavior own an ultrafast recovery time (∼30 ps)^[50], whereas in practice situations, nanotubes containing metallic and semiconducting behavior are entangled with each other, resulting in a recovery time of less than 1 ps^[51]. The distribution of diameters so far can be reduced to a desired range by precise control of the growth conditions. Additionally, the one-dimensional (1D) nature of the material gives rise to Van Hove singularities with a low density of states resulting in a small saturation fluence^[49]. CNTs-SA mode-locking in an erbium (Er)-doped fiber laser (EDFL) was firstly, to the best of our knowledge, reported by Set et al. in 2004^[52]. The as-prepared CNT films can be applied for wideband mode-locking with absorption wavelengths ranging from the visible to mid-infrared (MIR)^[53–56].

Graphene is a semimetal with a honeycomb lattice structure. The linear absorption of single-layer graphene is determined by the fine structure constant ($\approx 2.29\%$), which is independent of the frequency of the incident light^[57]. Nonlinear absorption can be obtained under stronger illumination that the modulation depth of graphene scales up with the number of layers^[58,59]. However, the observed modulation depth reduced with the layer number further increasing. This can be explained as follows: not all of the graphene sheets are saturated, and the SA cannot be fully bleached. Therefore, it is necessary to adopt the appropriate layer number of SAs in practical applications. In addition, the linear dispersion of Dirac electrons makes graphene an effective SA for ultrafast pulses generation, where the fast relaxation time (corresponding to carrier-carrier intraband scattering rates) is in the fs scale as well as the slow relaxation time (correlating with electron-hole interband recombination) in the picosecond scale^[60,61]. In 2009, Bao et al. firstly, to the best of our knowledge, reported graphene-SAs for ultrafast lasers^[62]. Compared with SESAMs or CNT-SAs, the most distinctive performance of graphene is the extremely broad absorption bandwidth covering from the ultraviolet to microwave^[63]. Thus far, graphene-based SAs have been used for 0.8 to 2.9 µm mode-locked lasers^[64–68].

Similar with graphene, bulk TMDCs are formed by stacking single-material layers with weak van der Waals forces. TMDCs exhibit rich physical behaviors from wideband insulators to narrowband semiconductors, semimetals, or metals, which offer evident optical responses over an extremely wide spectral range (from ultraviolet to terahertz or even microwave)^[69,70]. One of the most remarkable properties of TMDCs is that they exhibit an indirect to direct bandgap transition when decreasing the layer number. For example, the indirect bandgap of bulk ${\mathrm{MoS}}_2$ is 1.29 eV and changes to a direct bandgap (1.9 eV) when decreasing to a monolayer^[71]. Similar to ${\mathrm{MoS}}_2$, ${\mathrm{WS}}_2$ and ${\mathrm{MoSe}}_2$ also experience the transition from an indirect bandgap of 1.3 eV (1.1 eV) to a direct bandgap of 2.1 eV (1.55 eV)^[72,73]. The optical absorbance of monolayer TMDCs at their resonance wavelength is extremely large, which can reach 10% compared with that of monolayer graphene (2.3%). The strong light–matter interaction in TMDCs makes it possible for the generation of high-order susceptibility, which offers a wide platform for ultrafast photonics. Another unique property of TMDCs is the emergence of strong excitonic effects, which can absorb photon energies less than the bandgap. Photons with sub-bandgap energies can saturate the material due to Pauli blocking, enabling mode-locking at the near-infrared (NIR) wavelength^[70].

The emergence of SESAMs is considered the revolutionary improvement for ultrafast lasers. Molecular beam epitaxy (MBE) and metal-organic vapor phase epitaxy (MOVPE) are two generally used fabrication techniques with different growth pressures and temperatures. A SESAM typically consists of a high-quality distributed Bragg reflector (DBR) followed by a multi-QWs or QDs absorption layer. The design of the DBR and absorption region has been optimized for desirable optical parameters, such as the modulation depth, saturation energy, non-saturable losses, and recovery time^[74–77]. To date, most SESAMs are based on QW absorbers. Nevertheless, QD-SESAMs show great degrees of design freedom in the control of saturable parameters^[78]. The additional parameter of the dot density combined with the field enhancement makes it possible for independent control of saturation energy and modulation depth^[75]. Furthermore, the absorption wavelength can be adjusted with different QD sizes. High-temperature grown InAs QD-SESAMs working at 1.55 µm were shown in 2019^[79]. Keller et al. have demonstrated SESAM mode-locked vertical external-cavity surface-emitting lasers (VECSELs) based on QDs gain material and pushed the laser performance to a new level. In 2020, the first anti-resonant QD-SESAMs with high-temperature stability (withstand the long DBR overgrowth at 600°C) were demonstrated, which provides great opportunities for the monolithic integration into external-cavity surface-emitting laser (MIXSEL) devices^[80]. Latterly, Ko et al. have investigated the operating characteristics of the pulses with different SESAM positions. The results can give some good ideas to those who are interested in the improvement of the SESAMs-assisted mode-locked lasers^[81].

After nearly a decade of efforts, SESAM mode-locking has motivated the frontier in hundred GHz repetition rates (up to more than 160 GHz at 1 µm and 100 GHz at 1.5 µm)^[82,83]. Furthermore, SESAM mode-locked thin-disk lasers (TDLs) have pushed the pulse energies up to more than 80 µJ at a megahertz (MHz) repetition rate without additional amplification, and the average output power has been scaled to 350 W with 39 µJ pulse energies based on the $\mathrm{Yb}:{\mathrm{Y}}_3{\mathrm{Al}}_5{\mathrm{O}}_{12}$ (Yb:YAG) (~1 µm), Tm∶Ca₃Li_xNb_1.5xGa_3.5-2xO₁₂ (Tm:CLNGG) (~2 µm) thin-disk gain material (Fig. 1)^[84,85]. Figure 1(a) demonstrates the schematic of the mode-locked Yb:YAG TDL cavity. Figures 1(b) and 1(c) are beam profiles at 350 W output power in continuous wave (CW) and mode-locked operation.

SESAM mode-locked solid-state lasers have stimulated the novel ultrafast semiconductor disk laser (SDL), which bridges the gap between the solid-state laser and the solid-state TDL. VECSELs combining the advantage of semiconductor gain materials and SESAMs possess large gain cross section and are therefore ideally suited for high repetition rates, high average output powers generation. Additionally, advantages such as mass production on the wafer scale and easy integration into complex optical circuits are important for practical applications^[86,87]. After the first, to the best of our knowledge, demonstration of a passively mode-locked VECSEL in 2000, high repetition rates in the multi-GHz range and pulse widths reaching the fs level are no longer an issue^[88]. The large emitting area of the VECSEL allowed the average output power up to 5.1 W and the fundamental repetition rate scaled up to 50 GHz with the peak power over 4.3 kW^[89–93]. Figures 2(c) and 2(d) display the measured laser performance with different cavity lengths and output couplers. The shortest pulse of 196 fs was obtained with a 1.67 GHz repetition rate and a 112 mW output power^[89]. Then, a sub-100-fs (95 fs) pulse directly from a mode-locked VECSEL without compression has also been reported at a 1025 nm central wavelength^[90].

Many applications require more compact and simpler fs sources with a minimum number of components. A logical following step toward more compact SDL is the MIXSEL, which integrates both the gain and the absorber within one wafer to improve the integration level so that stable self-starting mode-locking can be achieved in a simple straight cavity. After the first, to the best of our knowledge, demonstration of optically pumped MIXSEL in 2007^[94], the MIXSEL cavity enabled repetition rate scaling up to 101.2 GHz^[95]. With the progress in QD SESAMs, the MIXSEL allowed the average output power to increase to 6.4 W^[96]. The schematic design and scanning electron microscope (SEM) image of the MIXSEL layer stack are given in Figs. 3(a) and 3(b). The straight laser cavity with only two components has generated 28 ps pulses with 2.5 GHz repetition rate at 959 nm central wavelength [Figs. 3(b)–3(e)]. Lately, a new record was obtained by optimizing the gain structure and thermal management, where the MIXSEL achieved a pulse duration of 144 fs at a center wavelength of 1033 nm with a pulse repetition rate of 2.73 GHz^[97]. Though VECSEL and MIXSEL are not expected to provide higher average output powers than mode-locked solid-state TDLs or shorter pulses than mode-locked Ti:sapphire lasers, they do provide simple solutions for compact, low-cost ultrafast lasers in the multi-GHz repetition rate region.

The wavelength flexibility and controllable fabrication make SESAMs superior to other types of SAs. However, limited by the available gain materials and saturable absorption materials, SESAMs mode-locking in the MIR or extreme ultraviolet (XUV) have not reached the same mature degree as in the NIR. In 2017, Labaye et al. reported the first intracavity high-harmonic generation (HHG) inside a TDL oscillator for XUV light generation^[98]. An illustration of the experimental setup is shown in Fig. 4. The mode-locked system was based on a diode-pumped $\mathrm{Yb}:{\mathrm{Lu}}_2{\mathrm{O}}_3$ thin disk and a SESAM. When the high-pressure gas jet was emitted into the beam focus, high harmonics of orders up to the 17th (60.8 nm) were detected with 255 fs pulse width, 64 MW intracavity peak power, and 17.35 MHz repetition rate. Due to the relatively long recovery times and the poor saturable absorption properties of GaAs-based SESAMs, GaSb-based SESAMs exhibiting better lattice matching, lower non-saturable loss, and significantly shorter recovery time after appropriate post-processing technology, have been explored for MIR region mode-locking. They have been successfully adopted both in solid-state lasers and SDLs for sub-picosecond pulses generation^[99–103]. Zhao et al. reported the first, to the best of our knowledge, sub-100-fs mode-locked Tm/Ho:CaYAlO₄ laser employing a GaSb-SESAM^[103]. As shown in Fig. 5, pulses as short as 87 fs are produced at a 2042 nm wavelength with an 80.45 MHz repetition rate. Nevertheless, mode-locking operation beyond 3 µm is hampered due to the limited absorption wavelength of GaSb-SESAMs.

More details are listed in Table 1, which collects some typical laser performances of SESAM mode-locking from visible to MIR wavelengths.

The fabrication of LDMs-based SAs is usually directly depositing or transferring nanomaterials onto optical substrates to form reflection or transmission type structures. It is worth noting that the insertion loss should be well designed due to a much lower single-pass gain in solid-state lasers. Passively mode-locked all-solid-state lasers based on CNT-SAs have been successfully investigated from 0.8 to 2.1 µm wavelengths^[121–126]. The single-walled carbon nanotube SA mirror (SWCNT-SAM) mode-locking was firstly, to the best of our knowledge, demonstrated with the gain medium of Er/Yb co-doped glass in 2005^[121]. Since then, pulse durations from picoseconds to fs have been achieved in various CNTs mode-locked lasers including Ti:sapphire (~800 nm)^[122,123], $\mathrm{Yb}:\mathrm{KLu}{({\mathrm{WO}}_4)}_2$ (Yb:KLuW), Yb:YAG (∼1 µm)^[124,125], Cr:forsterite (∼1.25 µm)^[126], Tm:Ca₃Na_xNb_1.6875Ga_3.1875O₁₂ (Tm:CNNGG), Cr:ZnS (∼2 µm, 2.4 µm)^[127,128]. The fast and slow recovery times of SWCNT-SAs are 0.25 ps and 1.16 ps, which are comparable to GaAs-based SESAMs^[129]. However, the diameters and chirality of the SWCNTs cannot support operation beyond 2100 nm due to the difficulty insynthesis. Later, other LDMs-based SAs such as graphene have also been successfully employed in all-solid-state lasers for ultrashort pulses generation. Ma et al. reported the first, to the best of our knowledge, ultra-broadband graphene-gold film SA mirror (GG-SAM) mode-locking in $\mathrm{Yb}:{\mathrm{YCa}}_4\mathrm{O}{({\mathrm{BO}}_3)}_4$ (Yb:YCOB) (∼1 µm), $\mathrm{Tm}:{\mathrm{Ca}}_3{\mathrm{Li}}_x{\mathrm{Nb}}_{1.5x}{\mathrm{Ga}}_{3.5-2x}{\mathrm{O}}_{12}$ (Tm:CLNGG) (∼2 µm), and Cr:ZnSe (∼2.4 µm) bulk lasers with the operation range exceeding 1300 nm (from 2310 nm to 2426 nm)^[130]. Similarly, graphene mode-locking was also achieved in Nd:Y₃Al₅O₁₂ (Nd:YAG) crystal lasers, realizing an ultra-high repetition rate of 11.5 GHz at 1064 nm^[131]. Remarkably, with graphene-SAs, the generation of sub-100-fs (41 fs) pulses around 2.4 µm was achieved in a Cr:ZnS laser^[132]. Apart from CNTs and graphene, other novel LDMs-based SAs have also been proved to support solid-state mode-locking, which generally operates around 1–2 µm.^[133,134] Recently, optically pumped SDLs mode-locked by CNTs and graphene-SAs have also been reported. Seger et al. reported the CNT mode-locked SDL for the first time, to the best of our knowledge^[135]. Stable mode-locking was obtained with 1.23 ps pulse width and 613 MHz repetition rate. Zaugg et al. demonstrated tunable VECSEL mode-locking with different gain chips by employing a single-layer graphene-SAM (G-SAM)^[136]. The laser cavity is sketched in Fig. 6(a), and the picture of 8/λ G-SAM is displayed in Fig. 6(b). The maximum tuning range (from 935 nm to 981 nm) is achieved with a QW VECSEL, which is larger than that of previously reported SESAM mode-locked VECSELs [Fig. 6(e)]. The mode-locked VECSEL delivered the repetition rate up to 2.48 GHz with the pulse duration minimizing to 466 fs [Figs. 6(c) and 6(d)]. In particular, Husaini et al. reported a mode-locked VECSEL with 10 W output power and 353 fs pulse duration using a graphene mirror^[137].

More details are listed in Table 2, which collects some typical laser performances of LDMs mode-locking in solid-state lasers.

Recently, the study on fiber lasers has shown explosive growth owing to their inherent advantages in the respect of flexibility, high beam quality, alignment-free format, and favorable heat sinking. LDMs can be easily integrated into various optical configurations to achieve alignment-free and all-fiber formats, which cannot be realized for SESAMs.

The solution-processing technique is widely used for high-yield few-layer nanosheets preparation, which includes either chemical exfoliation (e.g.,  lithium ion intercalation) or liquid phase exfoliation (LPE)^[139,140]. The as-prepared nanosheets can then be incorporated with fiber configurations via series methods (e.g.,  optical adsorption, spin-coating, drop-coating, vacuum filtration, or mixing with polymer)^[141–144]. A commonly used SA fabrication method is incorporating nanosheets with polyvinyl alcohol (PVA) solutions and naturally drying to form freestanding polymer composites^[139]. For example, CNT-PVA has been used in Yb-, Er-, and Tm-doped fiber lasers (TDFLs) for fs pulses generation^[145–147]. Figure 7(a) displays the SWCNT-PVA mode-locked fiber laser, and stable sub-100-fs pulses (∼74 fs) were achieved by the all-fiber stretched pulse design^[145]. As is shown in Fig. 7(c), the spectral width was measured to be 63 nm. However, the poor thermal damage threshold of the polymer limits their applications for high energy lasers. In 2007, Nicholson et al. proposed the optically driven deposition method for the fabrication of SWCNT-SAs^[148]. The deposition process arises from the gradient force of light and thermophoresis of nanomaterials. By applying SWCNTs-SAs into an Yb-doped and Er-doped laser, stable mode-locking could be obtained with pulse durations of 137 fs and 247.5 fs, respectively. Furthermore, Sobon et al. have fabricated CNT films with thickness ranging from 50 to 200 nm via a vacuum filtration technique^[149]. The shortest pulse duration was achieved by using a 100-nm-thick SWCNT layer, generating the bandwidth and pulse duration of 8.5 nm and 501 fs, respectively. Recently, Liu et al. fabricated an ${\mathrm{AuTe}}_2{\mathrm{Se}}_{4/3}–\mathrm{SA}$ by directly dripping the dispersion on the fiber end-face. Applying the ${\mathrm{AuTe}}_2{\mathrm{Se}}_{4/3}–\mathrm{SA}$ into an EDFL, ultrashort pulses (147.7 fs) operating at 1557.53 nm were realized with high stability (91 dB SNR)^[150].

Although excellent laser output performances have been obtained with fiber-end-face-integrated SAs, the mode-locking stability gradually deteriorates by increasing the pump power. Therefore, tapered-fiber (TF) and side-polished fiber (SPF)-based SA structures are proposed, in which the light–matter interaction is processed via the evanescent wave. The first, to the best of our knowledge, TF-integrated CNT mode-locked laser was reported in 2007, which delivered 694 fs/1.7 nJ pulses with a fundamental repetition rate at 13.3 MHz^[151]. Since then, numerous nanomaterials incorporated with TF/SPF have been reported for broadband mode-locked fiber lasers^[152–156]. For LDMs deposition, a frequently used method is fixing the TF in a groove and adsorbing the nanomaterials via the capillary effect or optical gradient force. However, it is challenging to exactly control the deposition area with high continuity. We have built a photonic crystal fiber (PCF)-assisted deposition system where the PCF sprinkler is controlled by a three-dimensional (3D) adjusting mount [Fig. 8(a)]^[155]. By applying the ${\mathrm{HfS}}_2–\mathrm{SA}$ into an EDFL, soliton mode-locking centered at 1562 nm was achieved with 221.7 fs ultrashort pulses. Figure 8(f) records the optical spectra of the soliton mode-locking operation for 4 h, which suggests the high stability of the laser cavity. Furthermore, ${\mathrm{ReS}}_2$ films with different thicknesses were fabricated by the vacuum filtration method to form a double-covered ${\mathrm{ReS}}_2$ microfiber (DCRM) SA^[157]. The nonlinear transmittance of the ${\mathrm{ReS}}_2$ SA at 1550 nm is shown in Fig. 9(c) with the modulation depth of 0.25%. Mode-locking operation was observed at the 1563 nm wavelength with a pulse duration of 3.8 ps. The main advantages of this type of SA devices are ease of fabrication and resistance to optical damage.

Although the LPE technique greatly facilitates the investigations on two-dimensional (2D) materials, the SA devices resulting from it suffer from the limitations of repeatability and durability. Inkjet printing provides a scalable craftsmanship that enables mask-less patterning with high resolution on flexible substrates. Indeed, 2D materials such as graphene, ${\mathrm{MoS}}_2$, black phosphorus (BP), and MXenes have already been introduced to inkjet printing systems^[158–160]. There are several challenges during the inkjet printing process that affect the quality of the devices, including stable jetting of single droplets, appropriate wetting of the substrate, and the uniform distribution of nanomaterials during droplet drying. Hasan et al. proposed a binder-free ink formulation by mixing low boiling point alcohols (e.g.,  ethanol, anhydrous isopropanol, and anhydrous 2-butanol) as solvents. Figure 10(a) is the photograph of the formulated BP ink. The as-prepared binder-free BP ink suppresses coffee ring formation by inducing recirculated Marangoni flow, which supports excellent consistency and spatial uniformity without pre-treatment of the substrates [Figs. 10(b) and 10(c)]^[161]. Figure 10(d) shows the optical micrograph of printed tracks on $\mathrm{Si}/{\mathrm{SiO}}_2$, glass, and polyethylene terephthalate (PET), confirming that the BP flakes are distributed uniformly without noticeable coffee rings. Self-starting mode-locking was achieved by applying the BP-SA into an EDFL with a fundamental frequency of 31.6 MHz. Figures 10(f)–10(h) display the operation stability of mode-locking by recording every 6 h for 30 days. Later, Zhang’s group also reported an inkjet printing BP-SA integrated ultrafast pulse fiber laser. Stretched pulses operating at 1555 nm are generated through intracavity dispersion management with pulse duration of 102 fs, which is ∼3 times shorter than the previous results^[162]. Then, Zhang et al. reported the inkjet printing ${\mathrm{Ti}}_3{\mathrm{C}}_2{\mathrm{T}}_x$ on an SPF for broadband mode-locked lasers^[163]. The minimum pulse durations are 215 ps at ∼1.06 µm and 114 fs at ∼1.5 µm, while the maximum output powers are 14.1 mW and 9.2 mW, respectively. Recently, Hassan et al. fabricated inkjet printing ${\mathrm{MoS}}_2$ optical devices in a large scale for ultrashort pulse (∼164.5 fs) generation^[158]. The laser output performances are depicted in Figs. 11(b)–11(e). In addition, 3D printing is a technique that deposits materials onto one another in layers to produce a 3D object. For photonic devices, 2D materials based on 3D printing are in the early stage but are still attractive. Lee et al. firstly, to the best of our knowledge, reported a filament-based 3D printing technique by printing graphene/polylactic acid onto an SPF for the fabrication of low-cost and wideband SAs^[164,165]. The measured modulation depths were ∼2.2% and ∼1.5% at 1.5 µm for the transverse electric (TE) and transverse magnetic (TM) modes, respectively. By applying the SA into an EDFL, stable mode-locked pulses were produced with a pulse duration of 979 fs at a 1560.2 nm central wavelength^[164]. Another SA device fabricated by the same approach has also been reported for the 1.9 µm mode-locked fiber laser with the temporal pulse width of 924 fs^[165]. Table 3 summarizes the output performances of the passively mode-locked fiber lasers based on SAs fabricated by the LPE technique. For each reference, we quote the best results.

Physical vapor deposition (PVD) is a widely used vacuum deposition technique to produce nanometer-thick 2D material films. In a typical PVD process, bulk materials can be vaporized by several techniques, such as bombardment of energic ions [i.e., magnetron sputtering deposition (MSD)], ultraviolet laser ablation [i.e., pulsed laser deposition (PLD)], or heating in a furnace [i.e., thermal evaporation deposition (TED)]. PLD and MSD are the most commonly used transfer-free methods to grow nanomaterial films directly on the target substrates at relatively low working temperature. Those remarkable advantages have motived researchers to re-focus on PVD methods to fabricate the desired SAs on various substrates.

In 2014, our group firstly, to the best of our knowledge, fabricated a novel topological isolator (TI) film on the TF by the PLD technique^[192]. The film thickness and crystalline quality can be controlled by varying the deposition conditions, such as laser power density, temperature, and deposition time. The modulation depth of ${\mathrm{Bi}}_2{\mathrm{Te}}_3–\mathrm{SA}$ is ∼6.7% with a saturation intensity of $26.7{\mathrm{MW}/\mathrm{cm}}^2$. By embedding the ${\mathrm{Bi}}_2{\mathrm{Te}}_3–\mathrm{SA}$ in to an EDFL, we achieved 286 fs nearly free-chirped soliton pulses at 1560 nm. After that, the ${\mathrm{Bi}}_2{\mathrm{Te}}_3$ 170th harmonic mode-locking (HML) also has been obtained by changing the polarization state and increasing the pump power with the maximum repetition rate of 2.95 GHz^[193]. A dissipative soliton mode-locked laser using a ${\mathrm{Bi}}_2{\mathrm{Te}}_3–\mathrm{SA}$ also has been reported, generating sub-170 fs pulses with 21 nJ pulse energy^[194]. In addition, the ${\mathrm{Bi}}_2{\mathrm{Te}}_3–\mathrm{SA}$ has also been applied in a TDFL for ultrashort pulse generation^[195]. The ${\mathrm{MoTe}}_2–\mathrm{SA}$ fabricated by the MSD technique has been proved for stable high-energy soliton mode-locking at 1.5 µm and 2 µm (Fig. 12). The generated pulse duration/pulse energy/average output power were 229 fs/2.14 nJ/57 mW at 1.5 µm and 1.3 ps/13.8 nJ /212 mW at 2 µm, respectively^[196]. The pulse energy and output power are the highest among the reported TDFLs using TMDC-SAs. We also fabricated microfiber-based ${\mathrm{WTe}}_2–\mathrm{SAs}$, where the modulation depth, non-saturable loss, and saturation intensity are 31%, 34.3%, and $7.6{\mathrm{MW}/\mathrm{cm}}^2$, respectively^[197]. Additionally, we have fabricated a novel fiber-integrated ${\mathrm{WS}}_2$ SAM by sequentially depositing the ${\mathrm{WS}}_2$ layer and Au film over the fiber end-face (Fig. 13)^[198]. The modulation depth, saturation intensity, and reflectivity were measured to be 4.48%, $138{\mathrm{MW}/\mathrm{cm}}^2$, and 98%, respectively. The schematic setup of the linear laser cavity is depicted in Fig. 13(a). By incorporating a ${\mathrm{WS}}_2$-SAM into an EDFL, self-starting mode-locking has been achieved in a wideband range [Figs. 13(b)–13(d)]. After that, high-quality ${\mathrm{WS}}_2$, ${\mathrm{MoS}}_2$, ${\mathrm{WS}}_2$, and gold films were vertically deposited on the fiber end to form a fiber-integrated heterostructure SAM and greatly enhanced the nonlinear optical properties compared with a single-material SAM (Fig. 14)^[199]. Figure 14(c) illustrates the band alignment and electron-hole recombination process of the ${\mathrm{MoS}}_2–{\mathrm{WS}}_2$ heterostructure. The large modulation depth (∼16.99 %) and small saturation intensity ($\sim 6.23{\mathrm{MW}/\mathrm{cm}}^2$) were beneficial for self-starting and pulse shaping in the laser cavity [Fig. 14(d)]. Self-starting mode-locking operation emitting at 1562 nm was realized with excellent laser performance. Table 4 summarizes the results of the passively mode-locked fiber lasers based on the SAs prepared by the PVD technique. For each reference, we quote the best results.

Chemical vapor deposition (CVD) is a versatile, scalable, and industry compatible technique for the controllable synthesis of atomic thin LDM films. Until now, a variety of LDMs including graphene, TMDs, and TIs have been synthesized in the lab^[206–218]. The first, to the best of our knowledge, CVD-graphene mode-locking was reported by Bao et al. in 2009, which demonstrated the generation of 756 fs pulses at the telecommunication band^[62]. Sobon et al. firstly, to the best of our knowledge, reported a comprehensive study on the nonlinear optical properties of graphene by changing the layer number, which demonstrated that the modulation depth scales with the number of layers^[59]. Apart from graphene, other nanomaterials (e.g.,  ${\mathrm{MoS}}_2$, ${\mathrm{WS}}_2$, ${\mathrm{Bi}}_2{\mathrm{Se}}_3$, ${\mathrm{MoSe}}_2$, ${\mathrm{WSe}}_2$, ${\mathrm{PtSe}}_2$, and ${\mathrm{ReS}}_2$) combined with different optical configurations also have been reported for ultrafast laser applications, which are shown in Table 5. CVD grown ${\mathrm{WSe}}_2$ film with high crystalline quality has been transferred onto the microfiber to form a hybrid SA for broadband mode-locking^[215]. Figures 15(b) and 15(c) are SEM images of the ${\mathrm{WSe}}_2–\mathrm{SA}$, where a large-area ${\mathrm{WSe}}_2$ film is tightly clinging to the waist of the TF. Stable soliton mode-locking has been achieved with the shortest pulse durations of 477 fs (at 1.5 µm) and 1.18 ps (at 2 µm). The obtained saturable absorption properties and laser performances with different CVD materials are summarized in Table 5.

MIR (2–10 µm) laser sources are of particular interests for various applications due to the unique fingerprint molecule spectra and transparent atmosphere windows in this region. Owing to the water vapor absorption in the atmosphere, which will induce considerable non-saturable loss, all-solid-state lasers around or after 2.8 µm have not been realized yet. Fiber lasers, hence, display their unique superiority for mode-locking in this region. Fluoride fiber lasers (FFLs) with various cavity structures, gain fibers, and SAs have been realized in the $2.8–3.5\mathrm{\mu m}$ region^[219–225]. In 2015, Jiang et al. reported an MIR mode-locked FFL based on ${\mathrm{Bi}}_2{\mathrm{Te}}_3$ nanosheets, which displays a promising application of LDMs for ultrafast lasers at 2830 nm^[220]. CW mode-locking was observed with repetition rate of 10.4 MHz and pulse width of 6 ps. In 2016, Qin et al. firstly, to the best of our knowledge, demonstrated the Er:ZBLAN mode-locked fiber laser around 2.8 µm utilizing a BP-SAM^[221]. In 2018, they also reported the first, to the best of our knowledge, 2.8 µm all-fiber mode-locking with BPs^[223]. Different from most solid-state lasers, FFLs require large modulation depths to initiate and sustain stable mode-locking operations. One effective method is to increase the sample thickness. Subsequently, Qin et al. reported a dual-wavelength pumped mode-locked Er-doped FFL at 3.5 µm with a bulk BP-SAM^[224]. The output power was 40 mW with the pulse duration estimated to be tens of picoseconds. Recently, they demonstrated pulse energy and peak-power scaling of MIR FFLs by dispersion management^[225]. The mode-locked FFL produced 215 fs pulses with pulse energy of 9.3 nJ and peak power of 43.3 kW.

As far as the reported results are concerned, the performance of SESAMs and LDM-SAs has improved steadily. The fabrication of SAs is improving toward being controllable, low cost, easy integration as well as mass production. Meanwhile, the operating bands of SAs are developing toward MIR and far-infrared wavelengths. To ensure further advance of SAs, there are several basic requirements for SAs. (1) Self-starting ability: the SA is required to achieve mode-locking immediately after being inserted into a laser cavity without introducing perturbations. (2) Durability: in order to achieve the long-term mode-locking operation under high pump power, the as-fabricated SAs should possess a higher damage threshold as well as a lower non-saturable loss. (3) Ultrafast recovery time: SAs with ultrafast recovery time are more capable of achieving shorter pulse widths. (4) Low-cost and controllable fabrication are important factors for industry compatibility.

SESAMs with good customization have been widely employed in commercial lasers. In the past few decades, many researches have been done to improve the performance of SESAMs. The development of VECSEL and MIXSEL chips has greatly simplified the cavity structures, which allow for generating GHz repetition rates and several kilowatts peak power directly from a laser cavity. However, a retaining limitation of SESAMs is inflexible, which does not meet the characteristics of all-fiber integration. In addition, SESAMs have limited bandwidth ($<100\mathrm{nm}$) and function wavelengths (800–2500 nm) due to the lack of effective materials.

LDMs with features like flexibility and ultrafast and broadband photoresponsivity are considered to meet these requirements. Figure 16 summarizes the mode-locked pulse width with different SAs in wide operating wavelength ranges. We use several colors to distinguish different SA fabricating technologies. LPE is a simple, contactless micromechanical exfoliation technique for high-yield nanosheets production. However, there are glaring problems such as poor fabrication controllability and repeatability and mode-locking performance degradation for a long-term operation, which should not be neglected. Notably, the inkjet printing method provides a new platform for the mass production of SA devices, and the long-term operation with high stability can be achieved by encapsulating SAs with a hexagonal boron nitride (h-BN) layer. Another advantage of inkjet printing is the controllable printing of desired graphic arrays on a myriad of substrates, which has already been applied in various photoelectronic devices (e.g., ultrafast lasers, photodetectors, and wearable electronic devices).

PVD is a transfer-free method to grow nanomaterials directly on the target substrates at relatively low working temperatures. The deposited films generally have good crystallinity after annealing. For fiber-end-face-integrated SAMs, selecting materials with an appropriate modulation depth is critical for self-starting mode-locking. This compact SA device with a protective gold layer can be applied for stable high-frequency (∼GHz) mode-locking. CVD fabricated SAs possess a low damage threshold due to the atomic thickness and cannot work steadily for long periods of time without encapsulation. In addition, the additional transfer processes to the target substrate are prone to introducing undesired physical damages or defects, causing higher non-saturable losses. Polymer as an auxiliary is often added during the transfer process. Rapid and residue-free transfer of CVD synthesized films is still a challenge so far.

Different atomically thin 2D materials can be readily stacked together by van der Waals forces to form a heterostructure, which offers a flexible and easy approach for bandgap engineering. In comparison with a single material, the heterostructure exhibits excellent nonlinear optical properties due to strong interlayer coupling^[226]. Take the type-II ${\mathrm{WS}}_2–{\mathrm{MoS}}_2$ heterojunction as an example; the photo-excited electrons and holes prefer to stay at the bottom of the conduction band and the top of the valence band, which can lead to an ultrafast carrier lifetime. A previous report has verified that the intralayer recombination of single-layer ${\mathrm{MoS}}_2$ is 2 ps, while it is 50 fs in a ${\mathrm{WS}}_2–{\mathrm{MoS}}_2$ heterostructure^[227]. Additionally, due to the similar crystalline structure and lattice constants, the interfaces between ${\mathrm{MoS}}_2$ and ${\mathrm{WS}}_2$ could be bonded with minimal structural defects^[228]. So far, 2D heterojunctions have been implemented in various optical and optoelectronics devices that are not limited to ultrafast lasers but are also in field-effect transistors, photodetectors, sensors, and light-emitting diodes^[229,230]. As mentioned earlier, Chen et al. have fabricated a ${\mathrm{MoS}}_2–{\mathrm{WS}}_2–{\mathrm{MoS}}_2$ heterojunction by the MSD technique on the tip of a single-mode fiber (SMF)^[199]. Liu et al. have also reported ${\mathrm{MoS}}_2–{\mathrm{WS}}_2$ heterojunctions as a useful SA for ultrafast pulses generation^[231]. The remarkable nonlinear optical properties are observed with a large modulation depth (∼19.12%) and lower saturation intensity ($1.361{\mathrm{MW}/\mathrm{cm}}^2$). Heterojunctions by mechanically stacking different CVD grown 2D materials have also been reported for ultrafast lasers^[232,233]. Mechanical stacking of 2D materials does not necessarily lead to strong interlayer coupling, which needs proper thermal annealing to reduce the interlayer distance. In general, 2D material heterojunctions with specific structure design show great potential for applications that are in their budding period.

Metasurfaces, artificial materials made of subwavelength elementary cells, have emerged as promising platforms for unexpected physical properties generation. Recently, Wang et al. demonstrated that plasmonic gold metasurfaces with excellent saturable absorption properties could behave as reliable SAs^[234]. This saturable metasurfaces material provides an idea to further increase the modulation depth and decrease the saturation intensity. Thus, saturable plasmonic metasurfaces offer a high degree of freedom to design novel ultrathin and efficient SAs for the desired optical performances.

Active manipulation of light in an optical fiber has gained research interest due to its compatibility with numerous optical systems. In 2015, Lee et al. demonstrated an electronically tunable all-fiber graphene device by fabricating graphene-based field effect transistor (FET) on the SPF^[235]. $Q$-switching and mode-locking were realized by applying the tunable graphene-SA in a laser system. This can also be extended to photodetectors, optical waveguides, and other photonic devices. In addition, a graphene-based active mode-locker has been reported by Sun’s group for electrically controlled broadband high repetition rate operation^[236]. This device provides a more compact, cost-effective, and robust active mode-locking solution. This research also provides a new idea for GHz pulse generation using 2D materials with high modulation depths (particularly, 2D heterostructures). A novel all-fiber integration electro-optic modulator was reported recently, where the graphene film was directly grown on the hollow holes and the outside wall of the PCF by the CVD technique with the growing length up to half a meter^[237]. This electro-optic modulator exhibits a strong light–matter interaction, which would inspire the generation of novel electrically tunable fiber devices such as all-fiber mode-locking lasers, nonlinear converters, and optical limiters. The exploration of actively tunable SAs is in its infancy, which is of great significance for promoting the development of active optical devices.

During the past two decades, various SAs have been adopted for FFL mode-locking. However, the output pulse durations were several to tens of picoseconds with low average powers. Mode-locking in the fs range around 2.8 µm was achieved by nonlinear polarization rotation (NPR) in 2015^[238,239]. LDMs with broadband absorption and ultrafast recovery time have been reported for CW mode-locking in the MIR region, whereas the pulse width still remains at the picosecond scale. So far, no fs mode-locked FFL at 3.5 µm has been reported. The gapless electronic structures of topological Dirac semimetals (e.g.,  ${\mathrm{Cd}}_2{\mathrm{As}}_3$) endow them with broadband saturable absorption effects, which can be used as potential SAs for MIR mode-locking^[240].

In summary, the overall development of SAs tends to compactness, controllable fabrication, and scalability as well as active controllability. It is crucial to put emphasis on SAs that are fully compatible with fiber lasers, which play an increasingly important role in industry and human life. The fabrication of reliable and effective SAs will open up new directions for mass production of next-generation lasers.