Sources of phase-stable few or single-cycle terahertz (THz) pulses driven by near-infrared (NIR) ultrafast lasers have enabled THz time domain spectroscopy (THz-TDS) to become an established tool for a variety of applications. Important scientific breakthroughs have recently been realized using THz-TDS, such as time resolved orbital imaging of molecules [1] or conductivity mapping of graphene films [2], and new areas continue to emerge, fueled by progress in state-of-the-art ultrafast THz sources.

However, in spite of this immense progress, the lack of powerful table-top ultrafast THz sources remains a challenge for many applications, and accelerator-based sources are the only alternative when a combination of high pulse energy and high repetition rate is required [3]. Some examples of areas that could immensely benefit from an increase in the driving power of table-top ultrafast THz sources are THz near-field imaging, remote sensing, and time-resolved spectroscopy of samples in aqueous environments. Our main motivation to develop high average power THz sources is the latter. THz-TDS allows us to study picosecond molecular dynamics of liquids and, more particularly, water as the “solvent of life”, i.e., its role in biological function of proteins or other bio-molecules [4]. At present, many well-established time-resolved spectroscopic methods are difficult or impossible to perform with low-power table-top THz setups because of the strong absorption of water at THz frequencies. This severely limits the achievable signal-to-noise ratio, which, in most experiments, either strongly limits the variety of samples that can be explored or results in measurements with very limited sensitivity, impractically long measurement times, and/or low temporal resolution when studying kinetics [5].

The low average power of commonly used ultrafast THz sources (typically $<1\mathrm{\mu W}$) directly mirrors the limited average power of most commonly used driving lasers based on Ti:sapphire technology, which are typically restricted to average powers of a few watts. Increasing the THz average power calls for ultrafast laser technologies capable of providing higher driving average power. In the last decade, diode pumped Yb-doped solid-state lasers in various optimized geometries have shown that ultrafast lasers can overcome these limitations, and systems exceeding the kW level have been demonstrated [6–8]. Among these technologies, mode-locked thin-disk lasers (TDLs) are particularly attractive because they can generate hundreds of watts of average power from a simple one-box mode-locked oscillator at MHz repetition rates with typical pulse durations in the range 0.2–1 ps [9–11]. Additionally, they show great potential for further power scaling [12] and very recently, systems with significantly shorter pulse durations (sub-50 fs) have been demonstrated, albeit at more moderate power levels [13].

Despite this impressive progress, very few attempts have so far been made to use higher excitation average power for THz generation and none exceeding the 100 W level [14]. The highest THz average power demonstrated so far at a megahertz (MHz) repetition rate is 0.3 mW, which was achieved by optical rectification (OR) in GaP crystals, employing an Yb-doped fiber amplifier system with 21 W of average power, a repetition rate of 42 MHz and 141 fs chirped pulses [15]. Lower THz powers on the μW level were demonstrated with comparable systems [16–18]. A similar THz power of 0.25 mW was obtained with a 1 MHz, 14 W, and 250 fs Yb:fiber amplifier employing the somewhat more cumbersome tilted pulse front method in LiNbO₃ [19]. Higher powers in the mW range were demonstrated using this method, but with kHz repetition rates [20,21].

In this Letter, we present to the best of our knowledge the first THz generation experiment with a laser excitation power exceeding 100 W, which is $5\times$ higher than previously reported [15]. While THz generation by OR is well understood at lower power levels ($<20\mathrm{W}$), it is not trivial to extrapolate results to this regime due to additional thermal effects that can easily be dominant in nonlinear crystals. In this first experiment, we explore the simple case of OR in GaP and show that this excitation power is viable to generate THz radiation, without having to significantly compromise on the generation efficiency. We generate THz average powers in the 100 μW range, which is close to the highest power demonstrated in the THz regime at MHz repetition rate so far (e.g., 300 μW [15]). We believe this is an important first step towards more powerful, compact THz sources for TDS, and we present paths to reach significantly higher powers in the near future.

Our driving laser is schematically depicted in Fig. 1. It is a home-built Yb:YAG mode-locked thin-disk oscillator with a comparable design to the one that achieves the highest average power from any ultrafast oscillator [9]. The Yb:YAG disk is 100 μm thick with a doping concentration of approximately 10%. It is diode-pumped with 18 passes of a Volume Bragg Grating stabilized diode laser at 969 nm, and a pump spot diameter of 2.5 mm. The single transverse mode resonator at 13.4 MHz supports up to 155 W of power in continuous wave operation, at an output coupling coefficient of 11% and a pump power of 420 W. An intracavity thin-film polarizer ensures linear polarization of the output.

 

Fig. 1. (a) Mode-locked thin-disk laser. HR: high-reflective mirror, CHR: concave high-reflective mirror. DM, dispersive mirror; TFP, thin-film polarizer; OC, output coupler. (b) Intensity autocorrelation. The inset shows the beam profile of the laser. (c) Power stability over 17 h measured with a slow detector at a typical operation power of 107.5 W.

Download Full Size | PDF

In order to achieve soliton mode-locking [22], we carefully balance the intracavity nonlinearity, mostly originating from the air inside the resonator [23] with negative dispersion. Since our laser is operated in a compact vacuum enclosure, the intracavity nonlinear phase shift is controllable and very small. In this way, we only require a small amount of negative dispersive mirrors with a total GDD of $-6000{\mathrm{fs}}^2$ per roundtrip, even at large intracavity peak powers. Starting and stabilization of mode-locking is achieved using a semiconductor saturable absorber mirror (SESAM) designed for high power operation [24] with a saturation fluence of $F_{\mathrm{sat}}=75\mathrm{\mu J}/{\mathrm{cm}}^2$, a modulation depth of $\mathrm{\Delta }R=1.5\%$, nonsaturable losses of $\mathrm{\Delta }{R}_{\mathrm{ns}}=0.4\%$, and an induced absorption coefficient $F_2=550\mathrm{mJ}/{\mathrm{cm}}^2$ [25]. In this configuration, we reach a mode-locked average power up to 120 W, resulting in a pulse energy of 9 μJ as well as diffraction limited beam quality with $M^2<1.1$. The transform-limited pulses are centered at 1030 nm. We routinely operate at a pressure of 47 mbar resulting in a pulse duration of 580 fs and a corresponding peak power of 13.6 MW. It is worth highlighting that this performance is achieved directly from an oscillator without the need for any additional amplification stages. Significant effort was dedicated to developing a turn-key “prototype-style” laser that can be used reliably for THz generation and future experiments of THz-TDS. Active alignment stabilization compensates for slow thermal drifts and ensures long term power stability, as shown in Fig. 1(c). Except for minor realigning of the piezo-driven end-mirrors, no readjustments are necessary upon starting the laser daily, and the vacuum chamber usually stays at constant pressure for weeks.

The experimental setup for THz generation by OR in a water-cooled $⟨110⟩$ cut GaP crystal is depicted in Fig. 2. Our choice of crystal was motivated by phase matching close to the pump wavelength in a collinear geometry at room temperature [26]. We focus the output of our laser to a $1/{\mathrm{e}}^2$-waist-diameter of 300 μm using a concave HR mirror. The GaP crystal is placed in front of the focus, which allows to easily set the desired spot size and intensity. We choose a beam diameter of 700 μm, resulting in a peak intensity of $6.7\mathrm{GW}/{\mathrm{cm}}^2$ on the surface of the anti-reflection coated crystal at the maximum applied average power of 106 W. The generated THz field is then collimated and refocused for detection by two $3^{\prime \prime }$ off-axis gold-coated parabolic mirrors with a focal length of 150 mm. The laser beam is transmitted through a hole in the first mirror and blocked by a beam dump. The electric field of the emitted THz pulse is fully characterized by electro-optic sampling (EOS) [27]. A small fraction of our laser output ($<1\%$) serves as the probe beam. In order to increase the sampling resolution, we temporally compress the probe pulses to below 100 fs via spectral broadening in a fused silica large-mode-area photonic crystal fiber and recompression using a pair of transmission gratings. The delay between THz pulse and optical probe pulse is generated by a combination of a motorized linear translation stage and a fast scan module set to a repetition rate of 1 Hz with a total scan range of 15 ps. This allows us to record the EOS traces in a quasi-instantaneous manner. The probe and THz pulses are overlapped in a second 3 mm thick GaP crystal, where the polarization state of the probe is changed by the presence of the THz electric field due to the electro-optic effect. The polarization is then evaluated using a standard configuration of a $\lambda /4$-plate and a Wollaston prism followed by a balanced photodiode. A lock-in amplifier is used to read out the resulting voltage. The THz power is measured by replacing the detection crystal with a commercial pyroelectric power meter (Ophir RM9-THz). Special care was taken to block off residual reflections of the pump beam from the first parabolic mirror, to ensure that no pump radiation falls on the detector. For this purpose, we employ a series of filters, consisting of a silicon wafer, a 10 mm thick piece of high density polyethylene and a sheet of black paper taped to the power meter. The transmission of each of the filters was measured by EOS and accounted for in the reported power values.

 

Fig. 2. Experimental setup consisting of the 120 W driving laser and an EOS setup. A powermeter can be used to measure the THz power.

Download Full Size | PDF

Figure 3. shows a typical THz waveform measured by EOS along with its power spectrum for a 1 mm-thick GaP crystal and an excitation power of 106 W. We obtain a clean single-cycle pulse centered at 0.8 THz. Under ambient conditions the spectrum is modulated by absorption lines, caused by the humidity of air (orange line). This effect can be largely removed by purging the setup with dry nitrogen (blue line). Our measured spectrum (blue) stands in very good agreement with our simulations [Fig. 3(b), dash-dotted line]. Our model is derived from the solution of the coupled wave equations under the undepleted pump approximation (UPA) assuming perfect phase matching and neglecting dispersion effects. These assumptions are well fulfilled in our case because of our long pulse duration and the absence of cascading effects [28], which we verified by measuring the spectrum of the transmitted pump beam with an optical spectrum analyzer. Additionally, our model considers full propagation of the THz beam through our imaging optics, including the apertures and holes of the mirrors. It is worth noting that simulating the effect of the THz beam propagation was important to reproduce our experimental spectrum. In fact, only considering the coupled wave equations, the simulated spectrum was shifted to lower frequencies (dashed line). This can be intuitively understood when considering clipping of the strongly diverging THz beam by the parabolic mirrors. Since the divergence is higher for lower frequencies, this effect is more pronounced on the left side of the spectrum. Furthermore, it leads to an increase in the focal spot size in the detection plane, such that the on-axis spectral weight of the lower frequencies decreases. These effects are particularly severe in our case where the excitation energy is relatively moderate, and the pulse duration of the driving laser is rather long, resulting in moderate peak power available for the optical rectification process and, therefore, requiring a relatively tight focusing geometry.

 

Fig. 3. (a) Electric field obtained by EOS for a 1 mm GaP crystal at an excitation power of 106 W for purged and unpurged cases. (b) Corresponding (normalized) power spectra, obtained by Fourier transformation. Calculated spectra are shown for comparison.

Download Full Size | PDF

In addition to distorting the measured spectrum, clipping of the parabolic mirrors causes significant loss. These losses are maximum at low frequencies and small spot sizes, due to the strong divergence of the THz beam, which is generated with spot sizes close to the diffraction limit. In our configuration, we calculate a total loss of 46%. This means that we generate significantly higher average power than we are able to measure in our present setup. It is straightforward to use significantly larger aperture mirrors to circumvent this problem and maximize the collected light. However, as we plan to increase the driving laser parameters towards higher pulse energy and shorter pulse durations in the near future, these difficulties will become less severe, even for standard mirror apertures.

Additionally, we tested several crystal thicknesses to optimize the generated power. We measured EOS traces and THz average power as a function of driving power for GaP thicknesses of 1, 2, and 3 mm (Fig. 4). We did not observe any significant change of the THz spectrum or waveform as compared to the ones shown in Fig. 3, which further substantiates the assumption of the UPA. A maximum power of 78 μW was measured for an optimal crystal thickness of 2 mm for 90 W of pump power. For the thicker crystals (2 and 3 mm), we avoided pumping at the maximum power due to the onset of damage on their back side. This indicates that the result obtained with our 2 mm crystal is close to the optimum efficiency for our current laser parameters with this generation technique. We believe a possible reason for the observed damage with higher powers and thicker crystals is the strong self-focusing of the beam in GaP, resulting from its large nonlinear refractive index [29] in combination with possible thermal effects. At the time of the experiment it was not possible to measure the temperature increase in the crystals; however, a more detailed investigation of the exact damage mechanisms in this unusual excitation regime as well as an exploration of thermal effects in the crystal is ongoing. In fact, we note that some previously reported experiments apply significantly higher intensity on the crystals, see for example [17]. However, in our case, the use of smaller spots is most likely prevented by the inevitable onset of thermal effects. Nevertheless, we verified, that the full power can be applied to the thicker crystals by increasing the spot size, thus at lower peak intensity.

 

Fig. 4. Power curves as a function of pump power for different crystal thicknesses, obtained under ambient conditions.

Download Full Size | PDF

The observed lower efficiency obtained with both shorter and longer crystals results from the well-known competition between THz generation and both laser and THz absorption [30,31], which leads to an optimal generation length. In our above-mentioned damage investigation, we are exploring how to reconcile these effects with thermal effects occurring in the crystal, which need to be considered in order to model this effect quantitatively.

The highest measured power of 78 μW corresponds to a conversion efficiency of approximately 10^-6, which is comparable to reported values of similar systems in GaP [15–17]. We note that this power level is still lower than the current record of 300 μW [15], obtained at only 21 W of average power. This can be attributed to the shorter pulse duration of 141 fs used in this experiment because the conversion efficiency of the OR process scales with the pulse duration as $1/{\tau }^2$ [26] at a given peak intensity. Consequently, we expect to increase the obtained THz power by more than an order of magnitude by reducing our pulse duration towards the 100 fs regime. Whereas our long-term goal is to achieve shorter pulse durations directly from the oscillator using novel gain media [13], external pulse compression [32] is also a viable route to reach the targeted parameters in the near future.

Moreover, reducing the pulse duration will relax the demands for strong focusing and reduce clipping effects of the THz beam as discussed above, as well as the thermal load on the crystals. Therefore, this will lead to THz average powers in the mW range, even in this simple generation scheme with GaP.

In conclusion, we have demonstrated, to the best of our knowledge for the first time, ultrafast THz generation at a driving average power of more than 100 W, using a compact amplifier-free mode-locked oscillator as a driving source for OR in GaP. In spite of the relatively long pulses and moderate pulse energies obtained with our laser, we obtain THz powers close to the highest values achieved so far at MHz repetition rate. We show that GaP crystals can sustain these average power levels, even in a comparatively tight focusing geometry. We believe that this is an important first step towards more practical high-power THz sources. Upcoming improvements in the available parameters of our compact laser system should enable mW levels from collinear OR in GaP. Additionally, we present important considerations for future scaling of this type of high-average power sources. These aspects will also be of importance in other more efficient generation schemes, such as tilted pulse front excitation in Lithium Niobate which bear the promise to successfully reach the Watt-level.