A Decade of Sub‐100‐fs Thin‐Disk Laser Oscillators

Thin‐disk lasers (TDLs) are best known for their high‐power continuous‐wave industrial applications. Nonetheless, the thin‐disk geometry is also highly attractive for ultrafast laser oscillators. The short propagation distance and large beam diameter inside the gain crystal allows for very low induced nonlinearity, low dispersion, and extreme peak powers inside the laser cavity. The path toward TDL oscillators directly delivering high average power at ultrafast pulse duration required for many scientific applications has, however, been tangled and is still ongoing. A decade ago, the first sub‐100‐fs laser oscillator is demonstrated, initiating the pursuit of even shorter pulses. Since then, many gain materials have been investigated in the thin‐disk geometry as well as various mode‐locking mechanisms for their suitability for efficient short‐pulse operation. In this review, the fast‐evolving development trends of TDL oscillators, as well as their scientific applications and prospects will be discussed.


Introduction
The thin-disk laser (TDL) concept, first demonstrated in 1993, [1] has been designed and optimized for most efficient cooling of the laser gain material while maximizing the beam size inside the gain crystal. This endowed the TDL technology with a unique suitability for high-power applications. A single thin-disk gain crystal with a ≈100-μm thickness can provide several kilowatts of average power [2] and withstand many gigawatts of peak power. [3] The small thickness of the disk directly contacted onto a very good heat sink, nowadays diamond in most systems, results in a very uniform single-dimensional heat flow inside the disk, minimizing the induced thermal lens. [1] In addition, the very short interaction length in the crystal keeps the dispersive and nonlinear response of the gain medium at a very low level, benefiting ultrafast operation.

DOI: 10.1002/lpor.202200258
The most common use of TDLs is high-power continuous-wave (CW) industrial lasers. Here, the thin-disk geometry allows to efficiently convert the high average power of low-brightness laser diodes into a high-brightness laser beam suitable for remote welding and cutting applications. [4][5][6] The vast use in industrial applications has led to the development of highly efficient multipass pumping solutions. Pumping TDL heads with up to 72-passes through the disk are nowadays commercially available, allowing for nearly 100% absorption of the pump power inside the gain crystal. Similarly, the production of high-quality industrial-grade Yb:YAG disks has been optimized to near perfection.
These developments later also benefitted pulsed operation of TDLs. Most notably, multipass and regenerative amplifiers have reached tremendous success, both industrially and scientifically. Up to 1.9 kW of average power in 1.3-ps pulses has been demonstrated from a multipass TDL amplifier [3] as well as 200 mJ in 1-ps pulses and 1 kW of average power out of a regenerative amplifier. [7] Recently, 720 mJ at 1-ps pulse duration and 1 kHz repetition rate have been demonstrated using a combination of a regenerative amplifier and two multipass amplifiers from a TDL system designed for triggering lightning by laser filamentation. [8] However, due to the narrow gain bandwidth of Yb:YAG disks, the pulse duration of these amplifier systems is typically limited to ≈200 fs. Shorter pulses can be achieved by operating TDLs as high-power mode-locked laser oscillators. Thanks to the soliton mode-locking, TDL oscillators can reach significantly shorter pulse durations than Yb-based laser amplifier systems. In addition, ultrafast TDL oscillators typically emit nearly transformlimited soliton pulses with clean sech 2 spectrum in an excellent transverse-mode TEM 00 Gaussian beam.
The development of mode-locked TDL oscillators, thus, became an important research direction aiming for ultrafast pulses from Yb-doped gain materials at tens to hundreds of watts of average power and megahertz repetition rates. In Figure 1, we show a historical overview of state-of-the-art TDL oscillators in terms of pulse duration and average output power. The color scheme together with the colored polygons categorizes the results into several time frames revealing a trade-off between achievable pulse duration and average power within each time frame. During the first decade of development indicated in red, TDL oscillators strongly increased their average power starting from initial 16 W [9] up to 141 W [18] and reached pulse durations as short Figure 1. Overview of state-of-the-art TDL oscillators operating with >1 W of average power and sub-picosecond pulse duration. The markers depict the mode-locking scheme: KLM -Kerr-lens mode-locking, SESAM -semiconductor saturable absorber mirror mode-locking, NLM -nonlinear mirror mode-locking, NPR -nonlinear polarization rotation mode-locking. The color scheme shows the historical evolution of the technology, categorizing the individual results into several time frames. A trade-off between average power and pulse duration can be observed within each time frame, which is moving over time toward the green-shaded corner indicating the desired parameter range. The numbers in the markers correspond to the references.  The underlying data used for generation of the graphs are also provided in the Appendix in Table A1.
as 240 fs. [10] In 2012, the first sub-100-fs TDL oscillator was demonstrated [21] triggering the pursuit of even shorter pulses. Within the following decade the frontier of achievable performance shown by the colored polygons has been pushed strongly into the sub-100-fs domain. This transition was allowed by using broadband Yb-doped gain materials as well as by the transition from semiconductor saturable absorber mirror (SESAM) mode-locking to Kerr-lens mode-locking (KLM). Interestingly, a very large step in this direction has been made only since 2021 as could be seen from the green markers. The strive toward short pulse duration at high-power levels is still a highly active research direction and will most likely enable many research and industrial breakthroughs. In this paper, we will discuss in detail this progress into the sub-100-fs domain and identify the current trends and prospects of the technology.
We will start with a discussion on the gain materials for ultrafast TDLs in Section 2. Initially, it appeared mandatory to switch to more broadband gain materials than Yb:YAG in order to progress into the sub-100-fs regime. [58] Numerous materials were evaluated and optimized in the thin-disk geometry, which will be discussed in Section 2. Over time it turned out that in high-power ultrafast TDLs the mode-locking method is more important than initially considered, which will be discussed in Section 3. Surprisingly, it has been shown that efficient ultrafast operation can also be achieved with Yb:YAG in KLM TDL oscillators. In Section 4, we will focus on nonlinear pulse compression, where the clean sech 2 optical spectrum and beam quality of TDL oscilla-tors provides an ideal starting point for reaching few-cycle pulses. In Section 5, we will review the methods of carrier-envelopeoffset stabilization of these lasers, which is a necessary ingredient for frequency comb and attosecond science applications. Finally, these lasers have been successfully used for further nonlinear conversion toward longer wavelengths for field-resolved midinfrared (MIR) [59] and terahertz [60][61][62][63][64] spectroscopy applications or toward shorter wavelengths through high harmonic generation (HHG). [65] The high average power from the laser oscillator has also been harnessed by single-cavity dual-comb systems. [66,67] Another very important potential of TDL oscillators is driving nonlinear processes at extremely high intensities directly inside the laser cavity such as intraoscillator HHG. [68][69][70] In Section 6 we provide a quick overview of these application areas for sub-100-fs TDLs.

Gain Materials
After the first demonstration of a mode-locked TDL oscillator in 2000 based on Yb:YAG and SESAM mode-locking, [9] one development direction soon oriented toward shorter pulses. The evolution of the shortest achieved pulse duration with respect to modelocking scheme and gain materials is depicted in Figure 2. During the first decade of development, mode-locked TDL oscillators have been enabled by SESAM technology. The initial studies showed that decreasing the pulse duration of SESAM modelocked Yb:YAG TDL oscillators is very difficult due to the narrow Laser Photonics Rev. 2023, 17, 2200258 Figure 2. Historical evolution of the shortest pulse duration achieved by TDL oscillators. The markers depict the mode-locking scheme, while the color code differentiates between Yb:YAG and other more broadband Ybdoped host materials. SESAM mode-locked lasers have achieved shortpulse operation only with broadband gain materials, whereas KLM laser can reach the short pulses also with Yb:YAG.
gain bandwidth of Yb:YAG. This can also be seen in Figure 2 from the green circle markers, which show that the pulse duration of these lasers has decreased only marginally since the first demonstration. The focus was therefore put on mode-locking studies using more broadband Yb-doped gain materials. Already in 2002, the pulse duration dropped from the initial 730 fs [9] down to 240 fs using an Yb:KYW disk. [10] The direction toward even shorter pulses was outlined in 2009 in a detailed review article highlighting the challenges and necessary steps for TDL oscillators operating in the sub-100-fs regime. [58] The first TDL oscillator which reached the 100-fs milestone was demonstrated three years later in 2012 using Yb:LuScO as gain medium and an optimized fast-recovery-time SESAM. [21] Many other gain materials have been utilized in the thin-disk geometry in order to allow for efficient short-pulse operation including Yb:KLuW, [16] Yb:CALGO, [22,25,26,29,43] Yb:LuO, [14,17,18,23,40,41,51] Yb:ScO, [30] and Yb:LuScO, [21] as shown in Figure 3. Another development direction of TDL oscillators leads toward operation at longer wavelength of 2 μm based on Ho:YAG. [45,55] The 2-μm TDLs have been recently reviewed in detail by Wang et al. [71] and will not be discussed further in the context of this review.
A great challenge for gain materials in the thin-disk geometry is not only the gain bandwidth but also equally important for the thermal properties, achievable doping concentration, and homogeneity. A high-thermal conductivity is required for efficient heat removal and reduced thermal lensing. A high homogeneity in terms of thermomechanical and optical properties over the disk is a necessary ingredient for efficient operation in TEM 00 mode. Among all tested gain materials shown in Figure 3, we will focus on the most abundant ones of Yb:YAG, Yb:LuO, and Yb:CALGO, which also form a good representative sample of the Yb-doped family in terms of performance trade-offs and technical challenges. Their parameters are summarized in Figure 4 and Table 1. The first and also by far the most used material in thin-disk geometry has been Yb:YAG. It was selected as an industrial standard for high-power CW applications and optimized to a great degree for the thin-disk geometry in terms of crystal growth quality, doping concentration, thermal properties, and bonding procedure. It features a relatively narrow gain bandwidth of 8-nm full-width at half maximum (FWHM), shown in Figure 4. The thermal conductivity amounts to 11 W m −1 K −1 , which however drops for high doping concentrations down to 7 W m −1 K −1 . [72] The crystal can be optically pumped at wavelengths ≈940 or 969 nm.
Yb:LuO was introduced to the TDL family in 2001 as the most promising representative of Yb-doped sesquioxides and a potential successor of Yb:YAG for shorter pulse durations. [73] It features a broader gain bandwidth of ≈13 nm ( Figure 4) and slightly higher thermal conductivity of 12.8 W m −1 K −1 compared to Yb:YAG. Thanks to the very close weight ratio between lutetium and ytterbium, Yb:LuO allows for higher doping concentrations than Yb:YAG without compromising the thermal conductivity of the gain crystal. [72] Initial studies characterizing Yb:LuO disks also showed unprecedented multimode slope efficiencies, exceeding 80% with >70% of optical-to-optical efficiency in CW operation. [72,74] In mode-locked operation, also very promising initial results were demonstrated, reaching 141 W of average power in 738-fs pulses and 40% optical-to-optical efficiency. [18] These first results and the remarkable properties of Yb:LuO have motivated many further mode-locking studies with this material, [14,17,18,23,40,41,51] as shown in Figure 3. However, despite the theoretical advantage of Yb:LuO, up to nowadays, these lasers have not yet outperformed the ones based on Yb:YAG, as can be clearly seen in Figure 3. This rather surprising outcome can be mostly attributed to the great difficulty of growing Yb:LuO crystals connected to its high melting point of ≈2500°C. [75] This temperature exceeds the capabilities of the commonly used iridium crucibles. Rhenium is the only suitable material sustaining such temperatures and allowing for the use of the Czochralski growth method. Unfortunately, the production of rhenium crucibles is very costly and time consuming since it involves galvanic deposition of the material. This technical difficulty has resulted in only a few growth attempts of Yb:LuO crystals for thin-disk laser applications, which have not yet allowed to reach a comparable level of optimization as for Yb:YAG.
A promising approach toward further development in this direction is based on compositions of Lu 2 O 3 with other sesquioxides such as Sc 2 O 3 and Y 2 O 3 , which have significantly lower melting temperatures. Recently, the first high-quality mixed sesquioxides crystal has been grown by conventional Czochralski method using an iridium crucible. [76] This approach might allow for a new generation of Yb-doped sesquioxide crystals with suitable properties for the thin-disk geometry in the near future. [77] Yb:CALGO is another notable gain material for TDLs. It offers a very broad gain bandwidth with >60 nm FWHM ( Figure 4). In principle, such bandwidth should support ≈20-fs pulse durations without requiring any further nonlinear spectral broadening effects. Its thermal conductivity of 6.3 W m −1 K −1 is lower than for Yb:YAG, but still sufficiently high for the use in TDLs. The first CW Yb:CALGO TDL oscillator was demonstrated in 2011 and showed a 40% slope efficiency with 32% optical-to-optical efficiency. [78] A SESAM mode-locked version followed already in 2012. [22] The laser was operated in three configurations, delivering 28    The gain cross-sections, gain , are calculated from the emission and absorption cross-sections as gain = · emission -(1 -)· absorption , for the inversion levels of = 0.1, 0.2, and 0.3. The underlying data was obtained from [58,79] 135 fs, respectively. Later studies showed even shorter pulse durations of 62 and 49 fs using SESAM mode-locking [25,29] and 30 fs using a KLM TDL oscillator. [43] All these short-pulse lasers, however, operated at rather low average powers for TDLs of <6 W.
In contrast to the TDL results, Yb:CALGO lasers have proven to be very successful in bulk geometry, where they even outperform their thin-disk counterparts. In CW operation, slope efficiencies of 73% with optical-to-optical efficiencies up to 65% have been demonstrated. [79] In mode-locked operation, up to 12.5 W has been obtained from a 94-fs laser oscillator with 20% optical-to-optical efficiency. [80] Shorter pulse durations of 22 and 31 fs at average powers of 0.7 and 1.6 W have been shown using another Yb:CALGO bulk oscillator. [81] The bottleneck hindering the efficient operation of Yb:CALGO lasers in the thindisk geometry seems to be again connected to the crystal growth quality. So far, the distorted crystalline structure of Yb:CALGO has not provided sufficiently homogeneous gain over the disk surface, [25,29,43] which prevents efficient operation in the fundamental TEM 00 mode and limits the performance in mode-locked operation.
Up to nowadays, the production of high-quality thin disks comes at a great effort and has been difficult to achieve outside of an industrial environment. It can also be noticed that several research groups have been recently returning back to Yb:YAG. [48,50,54] The commercial availability and the suitability for high average powers seem to outweigh the narrower gain bandwidth. Moreover, with the recent progress of Kerrlens mode-locked TDL oscillators, it has been shown that the narrow gain bandwidth can be compensated by the self-phase Laser Photonics Rev. 2023, 17,2200258   modulation (SPM) inside the laser, [33,40,53] as will be discussed in the next section. Recently, the shortest pulse durations achieved with Yb:LuO and Yb:CALGO TDL oscillators have been even surpassed by an Yb:YAG laser. [53] We, thus, expect that Yb:YAG will continue to dominate the realm of 1-μm TDL oscillators also in the next years.

Mode-Locking Mechanisms
Nearly every ultrafast TDL oscillator operates in the soliton mode-locked regime yielding high-contrast soliton pulses without pre-or post-temporal features, typically at very good transverse beam quality with M 2 values <1.1. TDLs are known to be power-scalable, meaning that the average power can be scaled by increasing the pumped surface of the disk and adapting the cavity to adjust the laser mode. This power-scaling principle is easily applicable in CW operation of these lasers, where the average power has been pushed toward the 10-kW levels. In mode-locked operation, the bottleneck is often the mode-locking mechanism itself, which needs to withstand the high peak and average powers, while providing sufficient self-amplitude modulation for soliton mode-locking. Various trade-offs are typically involved in the design of these lasers including the selection of the mode-locking mechanism to meet the desired parameters. The historical evolution of the highest average power with respect to pulse duration is shown in Figure 5. It can be clearly identified that the technology has been constantly evolving and that both SESAM and KLM lasers have an important role in this evolution.

SESAM Mode-Locking
SESAM mode-locked TDLs have opened up the possibility of obtaining tens to hundreds of watts directly from a laser oscillator, while also offering a favorable self-starting operation. Initial designs of SESAMs were originally inspired by saturable absorbers used for all-optical switches based on InGaAs/GaAs quantum wells grown on AlAs/GaAs dielectric Bragg reflector mirror. [82,83] But since the first demonstration, the technology has been constantly evolving and experiencing strong improvements. [41,[84][85][86][87] The semiconductor structure of SESAMs is mostly produced by molecular beam epitaxy or metal-organic vapor phase epitaxy (MOVPE). A thin layer, typically several nanometers, of a low bandgap material (for 1-μm lasers typically InGaAs) is grown within a higher bandgap Bragg reflector creating a quantum well (QW) saturable absorber. Here, photons are absorbed transferring their energy to electrons being excited from the valence band to the conduction band. This process saturates at higher fluences when the initial electronic states are getting depleted, and the electrons accumulated in the conduction band block further excitation. The absorption is generally recovered by two processes on different time scales. First, a fast intraband thermal relaxation occurs within several tens to hundreds of femtoseconds partially recovering the absorption. Later, the absorption is fully recovered through carrier recombination on a picosecond time scale.
For efficient operation inside ultrafast high-power TDL oscillators several requirements have to be met. The SESAM needs to provide sufficient modulation depth, typically ≈1%, for stable mode-locking, while featuring fast recovery time for short-pulse operation. For reaching high average power and high pulse energies inside the oscillator, low nonsaturable losses and high damage threshold as well as a reflectivity rollover shifted toward high fluences are essential. Since some residual losses always occur in the SESAM, efficient heat removal is also of high importance to prevent thermal lensing.
In 2012, general guidelines have been laid out by Saraceno et al. [85] elaborating on the design considerations for high-power SESAMs. It has been identified that a strong limitation originates from two photon absorption in the semiconductor structure causing a reflectivity rollover. It is, thus, beneficial to decrease the intensity in the SESAM by applying a top dielectric coating limiting the intensity reaching the underlying semiconductor structure. This, however, also decreases the modulation depth of the SESAM. In order to maintain a sufficiently high modulation depth several QWs can be incorporated into the SESAM. In 2016, Alfieri et al. proposed a method to increase the number of QWs by introducing an additional layer of AlGaAsP around each QW for Laser Photonics Rev. 2023, 17,2200258 www.advancedsciencenews.com www.lpr-journal.org strain compensation. [86] This allowed for up to 8-QW SESAMs, which were grown by MOVPE with high quality.
A further challenge toward high average powers is connected to the absorbed laser power and thermally induced lensing. The SESAM concept is power scalable, meaning that the power can be increased while correspondingly scaling the spot size on the SESAM while maintaining constant intensity. However, at larger spots, the requirements on surface deformation and thermal lensing become increasingly stringent. Both these effects are strongly dependent on the temperature gradient and thus the cooling properties. To improve the cooling of SESAMs, a substrate-removed concept has been proposed. [87] After growing, the original GaAs substrate can be etched away and the SESAM structure is directly bonded to a superpolished SiC heatsink resulting in excellent flatness and heat removal conditions. Another concept proposed for controlling the thermal lens of SESAMs is based on bonding a sapphire substrate to the top of the SESAM. [88] This allows for engineering of the thermal lens both in amplitude and sign through the thickness of the bonded sapphire substrate.
As these parameters have been improving, SESAM modelocked TDLs have been continuously pushing the frontier of high-power laser oscillators, as can be seen in Figure 5. The 100-W level was reached in 2010 using Yb:LuO gain material. [18] In 2012, this level was pushed to 275 W [24] and in 2019, 350 W was demonstrated using Yb:YAG. [48] On the other hand, decreasing the pulse duration of SESAM mode-locked TDLs turned out to be much more challenging. Due to the moderate modulation depth of only a few percent, SESAM mode-locked TDLs require broadband gain materials to reach short pulse durations. Further, the requirement of a short recovery time typically comes at a trade-off with other SESAM parameters required for high-power operation, such as linear absorption, damage threshold or rollover fluence. [41,86] Consequently, sub-100-fs operation of SESAM mode-locked TDLs has been so far only achieved at the cost of average power. The first sub-100fs TDL used a SESAM without a dielectric top coating to achieve a sufficient modulation depth of 3.3% in combination with a broadband gain material of Yb:LuScO. [21] The laser reached 5.1 W of average power at 96-fs pulse duration and 78 MHz repetition rate. Further sub-100-fs results have been achieved with Yb:CALGO. Shorter pulses of 62 fs at the same average power of 5.1 W were demonstrated in 2013 [25] and 49 fs was shown in 2014. [29] Nevertheless, due to the limited quality of broadband ytterbium-doped disks combined with conflicting requirements on the SESAM technology, it remains challenging for SESAM mode-locked TDLs to provide significant improvements in the sub-100-fs category.

Nonlinear Polarization Rotation Mode-Locking
The concept of nonlinear polarization rotation (NPR), commonly utilized in fiber-based laser oscillators, has been demonstrated inside a TDL oscillator in 2015. [36] Here, the nonlinear polarization rotation was induced by a phase-mismatched second harmonic generation in an LBO nonlinear crystal. The study used two nonlinear crystals at orthogonal orientation with 20-mm and 19-mm lengths, where the ef-fective propagation length corresponds to the difference between both crystals. This allowed for a long interaction length inside the crystal for strong polarization rotation while maintaining broadband properties. The mode-locked laser emitted 44 W of average power with ≈500-fs pulse duration. Although this more exotic mode-locking mechanism offers some theoretical advantages compared to the SESAM mode-locking, such as no absorptive losses and instantaneous response of the second order nonlinear process, the required propagation trough centimeter-long nonlinear crystals would likely prevent significant power-scaling or decrease of pulse duration.

Nonlinear Mirror Mode-Locking
Another mode-locking scheme aiming to reduce the pulse duration of TDL oscillators is frequency-doubling nonlinear mirror mode-locking (NLM). In this scheme, the saturable losses are formed by an intracavity second harmonic nonlinear crystal followed by a spectrally tailored output coupler, having partial reflection for the fundamental wavelength but full reflection at the second harmonic wavelength. This way, the low intensity light passes through the nonlinear crystal without significant second harmonic conversion and experiences a partial transmission on the output coupler. By contrast, the high intensity light is partially transferred to the second harmonic, which experiences complete reflection on the output coupler. On the way back through the nonlinear crystal, the second harmonic light is back-converted to the fundamental wavelength through optical parametric amplification.
The NLM concept has been first applied to a TDL oscillator in 2017. [42] Compared to the bulk oscillators operating with NLM, the TDL oscillators are much more promising candidates for this technique. Thanks to the high peak intensities inside TDL cavities, it is possible to use much shorter nonlinear crystals and thus increase the mode-locking bandwidth and reach shorter pulse durations. The first NLM mode-locked TDL oscillator has reached 323-fs pulse duration at 21 W of average power using a 0.5-mm long BBO crystal. In a later study, the authors combined the NLM with SESAM mode-locking, reaching more than three times higher average power of 66 W with 426-fs pulses in a selfstarting laser operation. [47]

Kerr-Lens Mode-Locking
The most commonly used approach for reaching shortest pulses is KLM. Accidently discovered in 1990 [89] together with the novel Ti:sapphire gain material, [90] KLM has revolutionized the ultrafast world. Conventionally, KLM lasers utilize the naturally occurring self-focusing inside the gain crystal, which leads to better pump overlap for high intensity light, so-called soft-aperture mode-locking. The same effect can be used in combination with a physical hard aperture inside the laser, so-called hard-aperture mode-locking. Here, the self-focused high-intensity light passes better through the aperture and experiences lower losses. The instantaneous effect of the Kerr lens together with high reachable modulation losses allow for the shortest pulse generation of all mode-locking mechanisms. [91] In TDLs, the large beam size inside the gain material usually does not provide sufficient self-focusing for KLM. For this purpose, an additional Kerr medium is placed close to an intracavity focus and a hard aperture is used for KLM. An oftenemphasized drawback of KLM lasers is the coupling between the self-amplitude modulation required for the mode-locking and the cavity dynamics, leading toward changing beam size between CW and mode-locked operation. This also makes it difficult to quantitatively characterize the modulation losses inside the laser. The KLM mechanism cannot be easily taken out of the cavity and independently characterized as it is in the case for SESAMs, thus hindering theoretical studies. Another difficulty originates from the requirement of a laser perturbation to initiate the modelocked operation, since KLM lasers are mostly not self-starting. Typically, this is achieved via shaking or moving a cavity mirror. Nevertheless, the short achievable pulse duration and the simple implementation make this approach very popular.
The first KLM TDL was demonstrated in 2011 [19] and reached a pulse duration of 200 fs at 17 W of average power using Yb:YAG gain material. This was by far the shortest pulse duration achieved by any Yb:YAG TDL oscillator, way below the 680 fs of SESAM mode-locked TDLs. [9] The next demonstration in 2016 has manifested the potential even clearer, showing a 155-W Yb:YAG TDL oscillator delivering 140-fs pulses. [37] The bandwidth of the demonstrated pulses has fully covered the available gain bandwidth of Yb:YAG, suggesting that the optimal operation point in terms of pulse duration for Yb:YAG has been reached.
Several attempts have been made to reach even shorter pulse durations using more broadband gain materials such as Yb:LuO or Yb:CALGO in the KLM TDL configuration. These lasers have reached very short pulse durations of 35 fs, [40] or 30 fs, [43] however, at very low average powers of 1.5 W and 150 mW, respectively. KLM TDLs typically operate at higher cavity losses compared to the SESAM mode-locked ones. This requires sufficient roundtrip gain to cover these losses while maintaining a comparably high output coupling rate in order to operate efficiently. Unfortunately, the more broadband gain materials typically provide lower gain compared to Yb:YAG. The roundtrip cavity gain can be increased by implementing several bounces over the disk in order to allow for higher output coupling rates and reach more efficient operation, which has been shown in several studies. [28,45,49,51] However, implementing several passes over the disk also increases the sensitivity of the laser cavity to the thermal lens induced by the disk. So far, the highest power of broadband Yb-doped gain material TDLs in the sub-200-fs category has been limited to ≈20 W, utilizing a double pass over an Yb:LuO disk. [51] The so far limited performance of the broadband-gain-material disks and the narrow gain bandwidth of Yb:YAG hindered power scaling in the sub-100-fs regime for a long time. In 2015, it was shown that the pulse duration of a KLM Yb:YAG TDL oscillator can be significantly decreased down to 49 fs [33] by placing several additional Brewster plates inside the cavity. The corresponding 23-nm FWHM optical bandwidth by far exceeded the 8 nm gain bandwidth of Yb:YAG. A similar result was shown in 2017, reaching 35-fs pulses using an Yb:LuO disk, exceeding the gain bandwidth by a similar factor. [40] This operation beyond the gain bandwidth was allowed by strong SPM inside the laser cavity, which generated the additional frequencies not covered by the gain material. Nevertheless, both these results showed only several watts of average power at a few percent optical-to-optical efficiency. Up to very recently, it was not believed that efficient operation is feasible with the pulse spectral bandwidth exceeding the gain bandwidth.
In spite of these assumptions, two recent studies have shown that the narrow gain bandwidth of Yb:YAG is not as strongly a limiting factor for reaching short pulse durations as originally expected. The first study investigated the limits of decreasing the pulse duration of a KLM Yb:YAG TDL oscillator operating in a strongly SPM-broadened regime. [53] It showed an overall better performance of Yb:YAG in the sub-100-fs regime compared to the more broadband Yb-doped hosts. The higher gain cross-section of Yb:YAG combined with the SPM-broadened regime has outperformed the broadband gain materials both in terms of average power, demonstrating 69 W at 84 fs, as well as in pulse duration, reaching the shortest pulses of any TDL oscillator with a duration of 27 fs at 3.3 W of average power. The second study utilizing the same laser system has pushed the frontier in the sub-100-fs regime even further by demonstrating 100 W of average power at 52-fs pulse duration. [54] These two recent studies represent a rather surprising performance improvement of TDL oscillators. Interestingly, the results do not seem to be enabled by any new technology. The enabling factors have been the use of an industrial-grade Yb:YAG disk, broadband GTI mirrors, and operation in a vacuum environment, all long recognized in the TDL community. [58] Also, the SPM-broadened regime does not represent any new modelocking scheme. [53] The pulse duration of the KLM lasers decreases linearly with the introduced group delay dispersion (GDD) according to the soliton equation. In the experience of the authors, most KLM TDL oscillators can be operated in the SPMbroadened regime without any additional requirement, apart from using broadband optical coatings, simply by decreasing the introduced GDD inside the cavity and optimizing the size of the hard aperture. However, since it was not expected that efficient operation in this regime would be possible, this option was not sufficiently investigated.
The optimization procedure of SPM-broadened oscillators is essentially the same as for oscillators operating within their gain bandwidth. The intracavity peak power strongly depends on the beam size inside the Kerr medium, its thickness, and the used material. The most common materials are sapphire, undoped YAG, quartz, and CaF 2 among which sapphire so far allowed for the highest peak power. [92] There is currently no consensus on which material properties are the most relevant, but it is likely connected to good thermal conductivity and a large bandgap. [93] The intracavity peak power typically scales linearly with decreasing thickness of the Kerr medium. A thinner Kerr medium increases the intracavity peak power, however, it also reduces the maximum usable output coupling rate. Thus, the combination of Kerr medium thickness and output coupling rate needs to be optimized for the maximum output power.
Further increase of the intracavity peak power can be achieved by increasing the beam diameter inside the Kerr medium. The Kerr medium is placed in the vicinity of an intracavity focus created by two curved mirrors. Its position can be normally tuned within a few-centimeter range where the position of the Kerr lens Laser Photonics Rev. 2023, 17, 2200258 Figure 6. Intracavity peak power for different radius of curvature of the focusing mirrors creating the Kerr section of the oscillator. The figure shows a linear increase of the intracavity peak power with respect to the mode size inside the Kerr medium. The mode size is estimated with the ray transfer matrix formalism in CW limit at the stability center. In reality, it may differ. The pulse duration for all parameter sets remained constant at Fourierlimited 330 fs. Reproduced with permission. [28] Copyright 2014, Optical Society of America.
creates the desirable saturable losses on the hard aperture (decrease of the beam diameter on the hard aperture). Since hard aperture KLM lasers allow the use of plane-parallel substrates under the Brewster's angle, the Kerr medium can be freely moved inside the cavity without misaligning the laser. This is very useful for initiating the mode-locked operation for which a position close to the intracavity focus is preferred. Once the laser is modelocked the Kerr medium can be moved out of the focus to maximize the power.
Additional increase of the beam diameter inside the Kerr medium can be achieved by scaling up the radius of curvature of the mirrors creating a focus for the Kerr medium, as proposed by Brons et al. [28] The study showed that the beam diameter inside the Kerr medium as well the intracavity peak power scales linearly with the radius of curvature of these mirrors while keeping all other laser parameters constant as shown in Figure 6. This was a very important realization, which later led to the most significant improvements of the KLM TDL oscillators. In the experience of the authors of this review, the law could be well reproduced also for KLM lasers operating in the SPM-broadened regime. Nevertheless, the beam size inside the Kerr medium should not be used as a universal determining parameter in between different systems since the intracavity peak power depends on several other parameters such as the Kerr medium thickness, introduced GDD, or pump intensity.
In the authors' experience, the challenges for power scaling of the short pulse KLM TDL oscillators are mostly linked to the high average and peak power inside the cavity. Although a fairly high intracavity peak power can be sustained in air atmosphere, nearly 200 MW at 1.3 kW of average power has been demonstrated, [28] a vacuum environment is usually preferred at high peak powers. For KLM lasers, it is not necessarily SPM, which is causing a problem, but rather the air fluctuations and self-focusing which disturb the laser and often lead to damage during the initiation of mode-locking. In the past, one of the major obstacles for power  Table 2. The cavity implements two passes over an Yb:YAG thin disk (TD) and uses two curved mirrors (F1) and (F2) to create a focus for the Kerr medium (KM). Reproduced with permission. [37] Copyright 2016, Optical Society of America. Table 2. Parameter comparison of four KLM TDL oscillators based on a similar cavity design inspired by Brons et al. [37] The cavity is shown in Figure 7. All oscillators use Yb:YAG as gain material and operate with >100 W of average power at comparable repetition rate and sub-150-fs pulse duration. OC -output coupling, RoC -radius of curvature, GDDgroup delay dispersion, n 2 -nonlinear refractive index, KM -Kerr medium.

Parameters
Unit Brons et al., 2016 [37] Fischer et al., 2021 [54] Goncharov et al., 2021 [52] Goncharov et al., 2022 [57] Output scaling used to be the thermal lens induced by the thin disk. [94] This is nowadays becoming less important thanks to diamondcontacted industrial-grade disks optimized for several kilowatts of average power. For KLM lasers, the influence of the disk thermal lens is mostly negligible compared to the Kerr lens and can be often completely disregarded. On the other hand, the linear Laser Photonics Rev. 2023, 17, 2200258 www.advancedsciencenews.com www.lpr-journal.org absorption inside GTI mirrors can become a limiting factor at high intracavity powers. This might require use of ultralow expansion substrates and optimizing the coating properties for low absorption.
In the pursuit of the shortest pulses, it has been shown that, provided broadband enough optical components, the pulse duration can be pushed to the limit, where the mode-locked spectrum extends even beyond the pump wavelength. [43,53] This is possible thanks to the TDL pumping scheme, which does not restrict reflectivity around the pump wavelength. At these rather extreme configurations, the optical-to-optical efficiency typically drops to only a few percent leading to operation at a high inversion level. Although this does not increase the FWHM gain bandwidth, it limits the reabsorption inside the gain medium allowing the frequencies generated by SPM to circulate in the cavity.
On the other hand, when not pushed to the absolute extreme, the operation in SPM-broadened regime can remain relatively efficient. An optical-to-optical efficiency of 26% has been achieved at 50-fs pulse duration and 100 W of average power using Yb:YAG, [54] which is comparable to the efficiency of TDL oscillators operating within their gain bandwidth. This also suggests that further power scaling in the sub-100-fs regime is within reach. For instance, a 220-W, 140-fs TDL oscillator operating at 24% efficiency has been recently demonstrated [52] whose pulse duration was already decreased to 115 fs. [57] It seems reasonable to assume that it could also be pushed into the sub-100-fs domain without compromising the efficiency.
To illustrate the aforementioned semiempirical scaling laws, we compare four KLM TDL oscillators [37,52,54,57] based on a similar cavity design. [37] The reference cavity design is reproduced from the original study and shown in Figure 7. All compared systems implement two bounces over an Yb:YAG disk, deliver >100 W of average power at comparable optical-to-optical efficiency of 20% to 30% and a similar repetition rate. The relevant parameters are summarized in Table 2. The most significant differences between these lasers include the Kerr medium thickness and its material, introduced GDD, radius of curvature of the focusing mirrors F1 and F2, beam size inside the Kerr medium, output-coupling rate, and pump power. Here, half of the systems operate within the available gain bandwidth of Yb:YAG and half in the SPM-broadened regime. Using the comparison table several dependencies can be identified. Most notably, the influence of introduced GDD on the pulse duration is clearly visible. The 52-fs laser uses only −2000 fs 2 , whereas the other >100-fs systems use −10 000 fs 2 to −15 000 fs 2 . Second, the peak power dependence on the beam size inside the Kerr medium as shown in Figure 6 can be further discussed. Here, the beam size inside the Kerr medium itself is clearly not a sufficient parameter since the smallest size of 130 μm corresponds to the highest peak power of 1.2 GW. However, when also the inverse proportional dependence on the thickness is considered (the 52-fs laser uses 5× thinner Kerr medium), the linear law agrees within 25% uncertainty margin. Lastly, one can observe a relatively strong variation in output coupling rate ranging from 8.5% to 19%, while all systems operate at comparable optical-to-optical efficiency. This shows that the highest output coupling rate does not strictly guarantee the highest efficiency. As suggested earlier, the ratio be-tween Kerr medium thickness and output coupling rate should be optimized for a given laser configuration.

Comparison of Mode-Locking Schemes
Each of the individually discussed mode-locking schemes has its favored parameter range and different operating conditions. The SESAM mode-locked lasers have been pushing the highest achievable average powers, whereas KLM shines at short pulse durations. For the NPR and NLM, at the current state of development, the performance cannot yet compete with the wellestablished KLM and SESAM mode-locked TDLs.
One of the crucial parameters for TDL oscillators, as a highpower laser technology, is the optical-to-optical efficiency. Operation at high efficiency is crucial for decreasing the demand on high-power pump diodes while limiting the excessive parasitic heat, which typically causes further problems such as thermal lensing or misalignment. An overview of the achieved opticalto-optical efficiency with respect to pulse duration and average power is shown in Figure 8. It can be seen that the highest efficiency of up to 40% [18] as well as highest average power has been achieved using SESAM mode-locked TDLs. This is due to the very low nonsaturable losses in the range of a few per mil for SESAMs which allows for efficiencies close to the CW operation. The high efficiency together with the high average power of SESAM mode-locked lasers has allowed for significant pulse energies. Up to 80 μJ has been demonstrated at 3 MHz repetition rate and 1-ps pulse duration. [31] However, the long pulse duration puts SESAM mode-locked TDLs in competition with Yb-based laser amplifier systems, which easily outperform them both in terms of average power and pulse energy. Although much shorter pulses have been achieved using SESAM mode-locked TDLs, extending far into the sub-100-fs domain, we assume that several of these lasers could have been SESAM-assisted soft-aperture KLM due to the presence of a Brewster plate for polarization selection and additional SPM. [19] By contrast, KLM TDLs typically operate at lower optical-tooptical efficiencies but much shorter pulse durations, which distinguish them from amplifier systems. As can be seen in Figure 5, KLM TDLs have completely taken over the sub-100-fs domain since the last decade. The optical-to-optical efficiency of these lasers has been typically ranging between 10% and 30% at >100-fs pulse durations but dropped significantly in the sub-100-fs domain as shown in Figure 8. Recently, this limitation has been overcome by the demonstration of an SPM-broadened KLM TDL oscillator delivering 52-fs pulse duration with 26% optical-to-optical efficiency. [54] Thus, it is likely that more efficient TDL oscillators will soon operate in the sub-50-fs domain, which will also allow to increase the average power in this regime.

Peak Power Scaling and Pulse Compression
One of the most important parameters for many applications is the peak power of the driving laser, which rules the efficiency of many nonlinear processes. TDL oscillators are very interesting for scaling the peak power since the thin-disk gain medium Laser Photonics Rev. 2023, 17, 2200258 Figure 8. Overview of the optical-to-optical efficiency depicted by color with respect to average power and pulse duration. The highest optical-to-optical efficiencies of ≈40% have been typically achieved by SESAM mode-locked lasers at >500-fs pulse durations. At shorter pulse durations KLM lasers typically offer higher efficiencies than SESAM mode-locked lasers.
does not pose a strong limitation in this sense. In SESAM modelocked lasers, the peak power is mainly determined by the design of the SESAM, mostly through the reflectivity rollover induced by two-photon absorption and by the gain bandwidth of the gain crystal. [41] In KLM TDL oscillators, the intracavity peak power has been shown to scale with the beam size inside the Kerr medium. [28] Toward high intracavity peak powers, a further challenge arises from the nonlinear response of the air inside the cavity. Several approaches have been proposed to mitigate this issue. The laser can be operated in a low-nonlinearity atmosphere such as helium [12] or the SPM can be canceled using a phase-mismatched second-harmonic crystal. [46] The most frequently employed method is, however, based on operating the laser in a vacuum environment, [24,31,32,37,41,48,53,54] which also prevents the gas thermal lens induced by the multipass pumping scheme. [95] Figure 9 depicts the historical evolution of the highest peak power reached by TDL oscillators. Whereas SESAM mode-locked lasers had been continuously increasing their peak power up to 66 MW demonstrated in 2014, [31] KLM lasers have very rapidly surpassed this value in 2013, [27] three years after their first demonstration, and nowadays reach values of 100 MW. [52,54] It can also be noticed that most of the recent high-peak-power systems are operated in vacuum as indicated by the bell jar symbol around the markers, shown in Figure 9.
Further increase of the peak power can be achieved by nonlinear pulse compression that relies on SPM to induce spectral broadening. Driving this process in an anomaly dis- persive medium can lead to self-compression of the pulse, whereas subsequent chirp removal is required for compression in normal-dispersive media. Both approaches have been pursued in numerous studies that are summarized in Figure 10. The first compression experiments were based on microstructured large-mode-area fibers [96,97] and reached ≈30-fs pulse duration and 10-MW-level peak power. These were followed by verylarge-mode-area rod-type gain fibers [98] generating 50-MW-level peak power and then by gas-filled hollow-core photonic crystal fibers [65,99] surpassing 100 MW peak power levering the scal-Laser Photonics Rev. 2023, 17, 2200258 Figure 10. Overview of peak power and pulse duration of state-of-the-art TDL oscillators operating at sub-picosecond pulse duration. The red hexagon markers depict the parameters after nonlinear pulse compression. [35,50,65,[96][97][98][99][100][101][102][103] The bell jar symbols indicate systems operated in vacuum. For references, [35,103] the peak power was calculated based on the compression ratio, power efficiency and compression efficiency stated in the manuscripts, which is likely an overoptimistic estimate.
ability of the compression schemes. All experiments showed that the nearly transform-limited soliton pulses and the excellent beam quality of the TDL oscillators allow for highquality pulse compression. Unfortunately, the average power generally remains limited to several tens of watts for practical application due to the damage vulnerability of the fiber tips.
More recently, multipass cells were employed for nonlinear compression, both in the normal [100] and the net-anomalous dispersive regime. [101] Their high overall transmission and their robustness against beam pointing instabilities allow for reliable operation at average powers well above 100 W. The highest demonstrated peak power is close to 170 MW [102] in ≈30-fs pulses. Compression into the few-cycle regime (≤10 fs at 1 μm) has been demonstrated as well using single-or multiplate compressors to avoid the need for broadband, dispersion-engineered multipass cell mirrors. [35,103] Overall, nonlinearly compressed TDL oscillators generate peak power on the 100-MW level, which already suffices for many nonlinear applications. Only for the efficient generation of high harmonics the peak power is still about an order of magnitude too low. In the future, this can change with the availability of oscillators directly emitting 100 MW peak power [52,54] combined with nonlinear compression in gas-filled multipass cells, which support both few-cycle pulses and high average power. [104,105] This allows envisioning GW-class sources suitable for high harmonic generation in the near future.

CEO Frequency Stabilization
An attractive application field of ultrafast oscillators is precision spectroscopy based on optical frequency combs. Here, the high average power and short pulse duration of TDL oscillators are especially attractive for efficient nonlinear conversion toward mid-infrared or extreme-ultraviolet frequencies, where directly emitting laser sources are not readily available.
The key requirement for optical frequency-comb applications is the carrier-envelope offset frequency (f CEO ) stability of the source.
The most common way to access the f CEO is using f-to-2f interferometry after coherent supercontinuum generation. [106] Here, the low-frequency part of an octave-spanning optical spectrum is frequency doubled and interfered with the high-frequency part of the fundamental spectrum. The resulting optical beat note provides access to measure the f CEO . The f CEO is affected by many parameters of the laser such as nonlinearity and dispersion providing a wide range of options for active stabilization. The most common method is modulation of the pump current. However, this method has a limitation for Yb-based lasers given by the long upper-state lifetime of the Yb-ion in the millisecond range. Correspondingly, the response of the f CEO to a pump current modulation is limited to a few kilohertz range in bandwidth. An additional challenge of TDL oscillators compared to bulk lasers is the high amount of acoustical noise introduced by turbulent water cooling of the thin disk. [35,67] The high noise level increases the Laser Photonics Rev. 2023, 17, 2200258 www.advancedsciencenews.com www.lpr-journal.org demand on the bandwidth of the control loop to cancel the noise. Many different modulation schemes have been investigated in order to stabilize the f CEO of TDL oscillators, which will be discussed in more detail in the following.

Pump Current Modulation
Regardless of the bandwidth limitation of the pump current modulation in Yb-based lasers, several studies have utilized this technique for f CEO stabilization in TDL oscillators. The first f CEO -stable TDL oscillator was demonstrated in 2013. [26] The laser was a SESAM mode-locked Yb:CALGO TDL delivering 2 W of average power in 90-fs pulses. The octave-spanning supercontinuum spectrum was obtained by optical broadening in a highly nonlinear photonic-crystal fiber and the f CEO was detected using a conventional f-to-2f interferometer. The f CEO lock was implemented using the pump current modulation and resulted in a residual in-loop integrated phase noise of 120 mrad (1 Hz-1 MHz).
Another challenge for the f CEO stabilization over the pump current is the difficulty to modulate a high-power pump diode which generally involves custom-made electronics. An approach circumventing this issue was demonstrated in 2016, introducing dual wavelength pumping of a KLM Yb:YAG TDL. [107] This concept uses a high-power 940-nm pump diode for pumping the laser combined with a low-power 969-nm diode for fast f CEO modulation. The difference in wavelength allows for applying a simple dichroic mirror to combine both pump beams. The octavespanning supercontinuum spectrum was obtained by first compressing the 250-fs pulses down to 20-25-fs pulses in a large mode area fiber followed by multiple bounces on chirped mirrors and then further optical broadening in an all-normal-dispersion fiber and the f CEO was detected using a conventional f-to-2f interferometer. The achieved residual in-loop integrated phase noise amounted to 390 mrad (1 Hz-500 kHz) at 250-fs pulse duration and 45 W of average power.
A third demonstration of f CEO stabilization by modulating the pump current has been shown using a KLM Yb:LuO TDL oscillator operating in a strongly SPM-broadened regime. [108] The laser generated 50-fs pulses at 4.4 W of average power and the f CEO current-based lock resulted in a residual in-loop integrated phase noise of 197 mrad (1 Hz-1 MHz). Also, in this result, the short pulse duration allowed to directly generate the coherent octave-spanning supercontinuum spectrum in a highly nonlinear photonic-crystal fiber to detect the f CEO using a conventional f-to-2f interferometer.

Intracavity AOM
A different approach to act on the f CEO is based on modulating the intracavity losses. This can be achieved by placing an acoustooptic modulator (AOM) directly inside the cavity of the laser, as first demonstrated in 2015. [35] This remarkable study utilized a nonlinearly compressed KLM TDL oscillator reaching a pulse duration of 7.7 fs at 6 W of average power. The AOM inside the cavity was capable of inducing up to 2% losses, withstanding the average intracavity power of 280 W. The resulting residual in-loop integrated phase noise amounted to 180 mrad (1 Hz-1 MHz). The out-of-loop measurement was also implemented in this work, revealing a value of 270 mrad, showing the importance of the outof-loop measurement. The concept has been further scaled toward 100-W level in 2019, demonstrating 40-fs nonlinearly compressed pulses at 105 W of average power and residual in-loop integrated phase noise of 90 mrad. [50]

Depletion Control
The most recently published method of f CEO stabilization of a TDL oscillator is based on controlling the depletion of the gain through bouncing part of the laser output beam back over the disk. [109] The fraction of the laser output was controlled by an AOM and bounced four times over the disk depleting part of the available gain for the laser. This method is similar in its principle to pump current modulation and a similar control bandwidth can be expected. In this first study, the high free-running noise of the laser oscillator only allowed for a comparably poor stabilization with a residual integrated phase noise amounting to 1.5 rad (100 Hz to 500 kHz).

Applications
Over the last decade TDL oscillators have not only gained significant performance, but also reached sufficiently mature state of development to become a driving source for further experiments and applications. In between other laser technologies, they offer some unique properties, which make them an attractive choice. Operations at 100-MHz repetition rates make them suitable for optical comb applications providing sufficient power per comb line. High-contrast soliton pulses without any temporal pre-or postfeatures at excellent TEM 00 beam quality are highly beneficial for pump-probe experiments. The delivered peak power directly allows for efficient nonlinear conversion toward MIR, THz or UV domains for spectroscopy applications. In addition, mode-locked TDL oscillators represent a relatively low complexity solution for obtaining ultrafast pulses. A KLM TDL oscillator practically requires a single GTI mirror, a sapphire window, and a hard aperture to reach few tens of femtoseconds pulse duration using a commercially available Yb:YAG gain material. Given so, ultrafast TDL oscillators might become a cheaper alternative to chirped pulse amplifier systems.

High Harmonic Generation
One of the first applications of TDL oscillators clearly manifesting the importance of high repetition rate and short pulse duration was nonlinear conversion toward extreme ultraviolet (XUV) through HHG in a noble gas target. Many applications such as photoelectron spectroscopy or XUV diffractive imaging have been relying on coherent XUV radiation delivered by large scale synchrotron facilities. HHG in a noble gas target is a lab-scale alternative, which is constantly growing in popularity and finding more and more applications. [110] Historically, the biggest challenge of HHG systems has been the low conversion efficiency from the driving laser to the XUV light. The efficiency strongly increases with the peak power of the driving Laser Photonics Rev. 2023, 17,2200258 www.advancedsciencenews.com www.lpr-journal.org laser, while also benefiting from short pulse durations. Thus, HHG systems have been often driven by kilohertz repetition rate Ti:sapphire laser amplifiers. Unfortunately, some experiments such as photoelectron spectroscopy suffer from spacecharge effects, meaning that when multiple photoelectrons are ejected by an XUV pulse from the sample, their trajectories are affected by their mutual coulomb interaction, which limits the measurement resolution. In this case, lower-energy pulses emitting only a few electrons at higher repetition rates are favored in order to maximize the resolution while allowing for short acquisition times. Regardless of space charge effects, the higher repetition rate is often useful for increasing the photon flux of these systems, benefiting most of the XUV applications. Compared to systems based on Ti:sapphire, Yb-based lasers are able to reach considerably higher average powers due to their lower quantum defect and well-developed high-power pump diodes. Hence, Yb-based lasers are much better suited for reaching the HHG required pulse energies at high repetition rates, enabling the transition from kilohertz-to megahertz repetition rate HHG systems.
The first TDL driven HHG source was presented in 2015, based on a 70-W, 780-fs SESAM mode-locked oscillator operating at 2.4 MHz. [65] The laser output was nonlinearly compressed in a krypton filled HC-PCF fiber to 105 fs at 105 MW of peak power. HHG was obtained from a xenon gas target and reached photon energies up to 30 eV with an extracted XUV flux in a single harmonic of 0.18 nW. These were, however, rather modest values, which have been surpassed by orders of magnitude by high-power Yb-based laser amplifier systems. [111] In recent years, a new avenue for TDL oscillators has been opening up in the XUV realm. With the increasing capabilities of nonlinear pulse compression combined with the decreasing pulse duration of high average power KLM oscillators, there is a potential that these sources will reach single-cycle pulse duration in the near future. This would allow for generation of isolated attosecond pulses at megahertz repetition rates. Such a megahertz-rate isolated attosecond pulse source would highly benefit many applications in the attosecond domain, in particular attosecond streaking experiments. [112] Recently several major steps have been made in this direction. The demonstration of CEO stabilization of a 100-W class TDL oscillator [50] followed by another 100-MW class oscillator directly emitting 50-fs pulses [54] is highly promising. Especially, in combination with the advances in multipass cell bulk compression schemes, which have been routinely reaching sub-30-fs pulses at these power levels [100,102] and most recently even 8.5 fs. [103] Combining these recent results will likely provide a CEO-stable single-cycle 100-W-class laser source capable of generating isolated attosecond pulses at megahertz repetition rates.

Intraoscillator High Harmonic Generation
A high potential of TDL oscillators is driving extreme nonlinear processes such as HHG directly inside the laser cavity since this enables the utilization of the very high intracavity peak powers. This concept is similar in its principle to femtosecond enhancement cavities. [110,113] In both concepts, the unconverted energy of the driving laser pulse is recycled inside the cavity, while increasing the available peak power by the enhancement factor of the cavity. In comparison to femtosecond enhancement cavities, the laser oscillator is a free-running system, which does not require active phase lock and coherent coupling of fspulses into the enhancement cavity, thus further simplifying the concept.
The high peak power and short pulse duration required for efficient HHG [114] make TDL oscillators ideal candidates for this task. An overview of the intracavity peak power with respect to the pulse duration of TDL oscillators including intraoscillator HHG systems (red hourglass markers) is shown in Figure 11. The first HHG inside a TDL oscillator was demonstrated in 2017. [68] The SESAM mode-locked Yb:LuO TDL operated at 320 W of average intracavity power, 255-fs pulse duration, and 17.4 MHz repetition rate, corresponding to 64 MW of intracavity peak power. The XUV average power generated in xenon was very modest in this first demonstration and amounted to only 0.5 nW in a single harmonic at 13 eV. This first demonstration was closely followed in the same year by a different system based on a KLM Yb:YAG TDL oscillator operating at 500 MW of intracavity peak power and 600-fs pulse duration at 3 MHz repetition rate. [69] The intraoscillator HHG yielded an XUV flux of 47 nW at 21 eV generated in argon. The next intraoscillator HHG system was published in 2021. [70] The KLM TDL used Yb:YAG gain material and operated at 365 MW with 105-fs pulse duration at 11 MHz. HHG was driven in argon and the generated XUV photon flux at 30 eV amounted to 0.4 μW. This value has been recently increased to 10 μW by further optimization of the laser system. [115] The connecting line in Figure 11 indicates the results obtained with the same laser systems developed for intraoscillator HHG. It can be noticed that even much higher performance has been achieved by these lasers when not driving HHG. This can be attributed to the plasma effects inside the HHG gas target, which typically limit the laser performance. [70] Without HHG, intracavity peak powers as high as 1.7 GW at 600-fs pulse duration [27] as well as 1.2 GW at 52 fs [54] have been shown, also at substantial output power of 150 and 100 W, respectively, pushing the frontier of the whole TDL oscillator domain.

THz Generation
The properties of mode-locked TDL oscillators also benefit frequency conversion toward long wavelength in the THz range. This potential has been recognized and discussed in detail in a review article in 2018. [116] Analogically to HHG, THz generation typically suffers from low conversion efficiency ranging from 10 −5 up to a few percent. Therefore, increasing the THz power available for the experiment also requires an increase in the driving laser power. In addition to the high average power, the short pulse duration of TDL oscillators provides a broad optical spectrum in the THz domain relevant for broadband spectroscopy applications.
Optical rectification in nonlinear crystals is particularly suitable for this purpose since it supports the high average powers delivered by TDLs. The first THz generation driven by a modelocked TDL was published in 2018. [60] The original demonstration used one of the simplest approaches of optical rectification Laser Photonics Rev. 2023, 17, 2200258 Figure 11. Overview of intracavity peak power with respect to pulse duration of state-of-the-art TDL oscillators operating at sub-picosecond pulse duration. The red hourglass symbols depict intraoscillator HHG systems, [68][69][70] labeled with the year and the generated XUV flux within the highest harmonic and its photon energy. The bell jar symbols indicate systems operated in vacuum. The gray connecting line indicates laser results, which were obtained with the systems developed for driving intraoscillator HHG, showing that development of these systems has also been driving the progress in the TDL oscillator technology.
in GaP in a collinear geometry. The generated THz radiation featured a broad spectrum extending up to 7 THz with an average power ≈10 μW. The same concept has been power-scaled to 300 μW in 2019 [61] and to 1.3 mW later the same year. [62] In a narrowband regime, even up to 66 mW was generated in lithium niobate driven by a TDL oscillator. [63]

Optical Parametric Oscillators
The high repetition rate of TDL oscillators in the range of tens to hundreds of megahertz makes the technology suitable for driving cavity enhanced nonlinear processes such as optical parametric oscillators (OPOs). [117,118] An interesting direction toward highly tunable THz and MIR sources has been recently proposed by Lang et al. [119] A 250-W, 10-MHz repetition rate TDL oscillator was frequency doubled and used for driving a near degenerate OPO. This provides two beams near the fundamental laser frequency, with a tunable frequency difference. The efficiency of this process is ≈40%, providing ≈50 W of average power both in the signal and the idler at 5 μJ of pulse energy. Importantly, this process nearly preserves the initial peak power of the driving laser amounting to 26 MW by shortening the pulse duration in the nonlinear process. The two output beams are highly suitable for frequency conversion toward MIR and THz through difference frequency generation, with broad tunability and high efficiency. This step still remains to be demonstrated by the authors, but it does not seem to pose strong technological challenges.

Dual-Comb Oscillators
Dual-comb spectroscopy is an extremely powerful tool for applications simultaneously requiring fast acquisition time and high spectral resolution. [120] Conventionally, the technology requires two optical frequency combs with slightly different line spacing, such as two fully stabilized mode-locked lasers operating at slightly different repetition rates. The spectroscopic information is obtained through heterodyne beating between the two combs.
A simplified alternative to fully stabilized lasers is based on generating the two optical frequency combs from the same laser cavity in a free-running operation. Here, the noise of the unstabilized f CEO and repetition rate of the two pulse trains is highly correlated and the common part of the noise cancels out in the beating process. This allows for short-term measurements within a time frame where the two pulse trains remain mutually stable, without the requirement for active stabilization. TDL oscillators are promising sources for this purpose since their high peak power allows for efficient nonlinear conversion toward MIR or UV wavelengths, which are of high interest for many spectroscopy applications. [120,121] The first steps on this path were made in 2019 [66] by the demonstration of the highest average power dual-comb laser based on a KLM TDL oscillator. The dual-comb cavity was based on splitting both end mirrors while the rest of the cavity components remained shared. The second demonstration of this concept came in 2020 based on polarization splitting. [67] The study showed that the stability of the free-running KLM TDL oscillator is sufficient for performing dual-comb spectroscopy within a 1-ms window in a proof-of-principle experiment measuring the Laser Photonics Rev. 2023, 17,2200258 www.advancedsciencenews.com www.lpr-journal.org absorption spectrum of acetylene gas. The acquisition time was limited by the uncorrelated noise in the laser mostly originating from water cooling of the disk. Longer acquisition times would likely require the development of a passive cooling of the thin disk. The suitability for high resolution measurements after nonlinear frequency conversion still needs to be demonstrated.

Field Resolved Spectroscopy
An advanced application utilizing a TDL oscillator as a driving source building upon the recent advances in this technology is a field-resolved MIR spectroscopy, [59] published in 2020. The study uses a nonlinearly compressed TDL oscillator delivering 16-fs pulses at 60 W of average and 28 MHz repetition rate for driving intrapulse difference frequency generation toward MIR. The few-cycle pulse duration of the laser enables detection of the MIR light through electro-optic sampling in the time domain, providing information directly about the electric field rather than its intensity. The measurement in the time domain allows for starting the acquisition after the passage of the MIR excitation pulse and sampling only the response of the sample. This endows the field-resolved spectroscopy with unique sensitivity and resolution and also broad tunability in the MIR range. The technology was shown to be highly suitable for fingerprinting of complex molecular ensembles such as human blood serum or in vivo transmission spectroscopy of intact strongly absorptive biological samples such as living cell cultures or even plant leaves. The unprecedented sensitivity and repeatability of the presented experiments proves the suitability of the TDL oscillators for driving demanding applications requiring short pulse durations, high average powers, and also stable long-term operation.

Conclusion
We have reviewed the recent development of ultrafast TDL oscillators. Over the last decade, we have observed a strong push toward short pulse duration. The development was mostly split in two directions. In the beginning, SESAM mode-locked TDLs represented the major research direction. SESAM technology has progressed dramatically decreasing the recovery times, pushing the rollover fluence, and optimizing the thermal lensing. Similarly, a lot of effort was focused on the development of broadband gain materials suitable for thin-disk geometry. Dozens of materials have been tested and optimized for reaching shortest pulses, while not compromising average power. Nevertheless, the progress in this direction has turned out to be much slower than originally anticipated and up to nowadays SESAM mode-locked TDLs shine rather at longer pulse duration (several hundred femtoseconds) and high average power.
Later, KLM TDL oscillators have become more promising for reaching shorter pulses mostly thanks to the instantaneous response of the Kerr effect and higher modulation depth compared to high-power SESAMs. The high amount of SPM also allows for nonlinear broadening directly inside the laser cavity enabling shortening the pulse duration even beyond the gain material bandwidth limit. This important property allowed KLM TDLs to use commercially available gain material of Yb:YAG in the sub-100-fs regime, which greatly reduces the cost and effort required for development of these oscillators. Against expectations, the industrial standard of Yb:YAG has still remained the most successful candidate even for the shortest achieved pulses. It will be very interesting to see if this trend continues in future or if YAG will be replaced by another host since the development of broadband gain materials is still ongoing. In contrast to SESAMmode-locked TDLs, in KLM it has been so far much more challenging to understand the fundamental limitations. Many of the recent results have been driven rather by empirical optimization than by means of theoretical investigation. Here, the difficulty originates from highly convoluted dependence on many parameters such as beam size inside the Kerr medium, peak power, pulse duration, thermal lensing, population inversion, losses on the hard aperture, and many others. These parameters generally vary depending on the laser operating point and are often subjected to measurement uncertainties hindering reliable quantitative theoretical studies. For further progress, it seems important to focus on the development of a concise theory describing the KLM operation of TDLs. A dedicated experimental study will likely be required for this purpose thoroughly characterizing the influence of the individual parameters and supporting the theoretical results. Gaining a deeper insight into these oscillators later promises further improvement of the laser performance. At the current state of development, KLM TDLs already offer a relatively simple solution to reach sub-100-fs pulses at tens to hundred watts of average power. Besides the commercially available TDL pumping scheme, KLM TDLs are based on easily accessible passive components such as dispersive and highly reflective mirrors, a Brewster plate, a hard aperture, and an output coupler. So far, most high-power TDL oscillators have been operated in vacuum environment, which increases the complexity and cost of these systems. This is often preferred in laboratory setups during the development phase. Nevertheless, at a later stage, the vacuum environment will likely be replaced by a helium atmosphere, or the laser can be optimized to run in ambient air reducing the overall cost. Given so, ultrafast TDL oscillators might become a simpler and cheaper alternative to chirped pulse amplifier systems for applications requiring short pulses at megahertz repetition rates and high average powers.
Hand in hand with the improving performance of TDL oscillators came the development of nonlinear pulse compression techniques suitable for the high average powers and moderate pulse energies of these lasers. Here, the short pulse duration together with the perfect TEM 00 gaussian beam and the clean soliton spectrum of TDL oscillators offers an excellent starting point to reach extremely short pulses. Nowadays few-cycle pulses are already routinely available from nonlinearly compressed TDL oscillators. With the fast development of both technologies, we will likely observe these sources entering a single-cycle regime in the near future. In combination with CEO-frequency stabilization, this will likely allow TDL oscillators to become reliable waveform-stable single-cycle sources delivering GW-level peak powers at megahertz repetition rates. Such sources will benefit many scientific applications particularly in the attosecond domain. Another important potential of ultrafast TDLs is driving extreme nonlinear processes such as HHG directly inside the laser oscillator, offering a way toward compact coherent XUV Table 3 sources based on the commercially available and well-developed TDL technology.