1. Introduction

Ultrafast lasers are the main enabling tool of THz time-domain spectroscopy (THz-TDS), which is nowadays a well-established technique in various fundamental and applied fields of science and technology. The performance of ultrafast laser driven THz sources for THz-TDS has enormously progressed in the last decades in terms of bandwidth [1], tunability [2], pulse energy and peak fields, with pulse energies approaching the millijoule regime, and corresponding field strengths exceeding 1 MV/cm [3,4]. Pushing these limits continues to open new application possibilities. However, one performance barrier which remains a critical challenge for many fields is the low average power of current THz-TDS setups. One example of an area where this has been a major constraint is the study of biologically relevant samples [5] which are naturally water-rich and thus exhibit very strong absorption in this frequency region. The resulting THz-TDS experiments are often limited by extremely poor signal-to-noise ratio (SNR) and/or impractically long measurement times [6], forcing scientists to either abandon these research lines or to operate in extremely difficult experimental conditions (for example in non-biological conditions by cooling the samples, or off-resonance at frequencies <<1 THz). Most experiments in this area are currently performed using accelerator-based sources [7], which are very restrictive and costly. For this and many other applications, laboratory-based high average power ultrafast THz sources in the 1 THz to 10 THz region would represent a breakthrough.

The limited average power of current ultrafast THz sources is mostly due to the low average power of commonly used Ti:Sa driving lasers (typically in the few watt range) in combination with low conversion efficiencies (from 10⁻⁶ up to the percent level in optimized cases). Surprisingly, whereas ultrafast diode-pumped solid-state lasers based on Yb-doped gain media have largely progressed and nowadays routinely provide hundreds of watts up to kilowatts of average power [8–11], THz generation with more than a few tens of watts of excitation power remains widely unexplored. We focus our attention on one Yb-doped technology which is particularly promising for THz generation at high-power and high-repetition rate [11]: modelocked thin-disk oscillators. These systems provide a very interesting alternative to more complex amplifier systems by providing femtosecond pulses at several 100 W average power directly from a compact, amplifier-free one-box system, operating at MHz repetition rates [12]. Compared to typical amplifier setups with similar performance, these systems offer outstanding beam quality (as required for stable modelocking), nearly transform-limited pulses and the potential to reach low noise levels, as well as a greatly reduced complexity. The reasons why these (and other Yb-doped) systems are so far not widely established in the THz community are manifold: on the one hand their rather long pulse durations, most commonly >600 fs, result in poor THz conversion efficiency and a frequency span limited to <1 THz [13]. Furthermore, most commonly used nonlinear crystals used for THz generation have not been explored at high excitation average power, where thermal effects and damage are expected to be critical issues.

Nevertheless, several attempts have been made in the direction of higher average powers in the last few years. Yb-fiber lasers were explored early on using the tilted-pulse front technique in lithium niobate (LN) [14]. In this early experiment, 14 W from a fiber amplifier operating at 1 MHz repetition rate and a central wavelength around 1 µm were used to obtain 0.25 mW of average power [15]. More recently, a 6 W Yb:YAG regenerative amplifier operated at 1 kHz repetition rate was used to obtain 1.6 mW [16]. In [17], a commercial 1.2 W average power Yb-amplifier at 1 kHz repetition rate was used to generate THz radiation in cryogenically cooled LN, reaching 45.6 mW of average power with an exceptional conversion efficiency of 3.8% [17], providing the highest average power so far achieved with ultrafast laser driven THz sources. This technique is promising to extend to much higher average powers, however only allowing to generate narrow bandwidths <1 THz due to the strong THz absorption near its phonon resonance at 4 THz.

Remarkable results have also been obtained using GaAs photoconductive antennas. Early on, power levels of 1.5 mW were achieved using GaAs large area photoconductive antennas at a repetition rate of 250 kHz with watt level driving powers [18]. In [19], 4 mW of average power were demonstrated at MHz repetition rate using only Watt-level excitation in plasmonic structures. However, these structures are typically complex to fabricate in large areas and further power scaling appears difficult.

For applications requiring very broadband spectra with frequencies >>1 THz, ultra-broadband THz generation in two-color plasma filaments is commonly used, although actual implementations for THz-TDS are still scarce [20–22]. Using this method, 1.4 mW were demonstrated with peak fields of 8 MV/cm at 1 kHz repetition rate [23]. This technique offers the obvious advantage of being damage-free, allowing for high excitation power, however, it requires mJ-class pulses, which are not yet within the parameter range of MHz repetition rate thin-disk oscillators.

Optical rectification (OR) in GaP is an excellent alternative for the state-of-the-art parameters supported by current thin-disk oscillators, supporting bandwidths up to 7 THz in a collinear geometry for 1030 nm excitation [24–26]. These advantages come at the expense of a relatively low nonlinearity (as compared to LN) and higher free-carrier absorption (FCA) [15], leading to typically low conversion efficiencies on the order of 10⁻⁵. So far, the highest power achieved with OR in GaP is 300 µW, obtained using 141 fs long chirped pulses at 21 W power from a fiber laser [27]. In a recent proof-of-principle experiment, we showed for the first time that >100 W excitation average power can be used for OR in GaP [13]. However, in this first realization, we were severely limited by the long driving pulse duration of 580 fs of our oscillator and were able to only reach 78 µW of THz power at a central frequency of 0.8 THz despite the high driving power.

Here, we overcome these limitations by nonlinearly compressing our driving source at very high efficiency to sub-100 fs, and report on a THz source based on OR in GaP delivering 1.35mW of average power. This is, to the best of our knowledge, the highest average power reported with OR in GaP, exceeding the previous state-of-the-art by a factor of four [27]. Remarkably, compared to most other milliwatt-class THz sources, our broadband source operates at a center frequency of 2 THz, with a spectrum extending up to 6 THz. This result was obtained using an amplifier-free ultrafast laser system delivering 88 fs pulses and 112 W of average power. This simple source represents a very attractive driver for future experiments of THz-TDS of aqueous samples in the linear regime with potentially very high SNR, and for other applications that are currently limited by SNR. Furthermore, upscaling of this simple and compact setup to the 10 mW level appears possible in the near future by using higher driving powers as well as optimizing the laser parameters further.

2. Experimental results

2.1 Laser source

The full experimental setup is depicted in Fig. 1. Our driving laser source (Fig. 1(a)) is a self-built semiconductor saturable absorber mirror (SESAM) mode-locked Yb:YAG thin-disk oscillator delivering an average power of up to 123 W. At this average power, the laser produces 534 fs pulses at a repetition rate of 13.4 MHz. The laser operates at a central wavelength of 1030 nm. Soliton modelocking is achieved by balancing intracavity self-phase modulation (SPM) with negative dispersion provided by several dispersive mirrors with a total group delay dispersion (GDD) of −6000 fs². The laser system is built in a vacuum chamber with a 1.3 m by 0.6 m footprint, evacuated to a residual air pressure of 35 mbar in order to reduce the intracavity nonlinearity. This allows soliton modelocking with high pulse energy and moderate amounts of dispersion [12]. A SESAM is used as one end mirror to start and stabilize soliton modelocking and an intracavity thin-film polarizer ensures linear polarization at the output. More details about this laser system can be found in [13]. As expected from a soliton modelocked laser, the pulses out of the oscillator are sech²-shaped transform-limited pulses with a full-width half maximum bandwidth of 3 nm, and the beam quality is excellent with a measured M² < 1.05.

 

Fig. 1. Complete experimental setup consisting of the modelocked thin-disk laser (a), Herriott type MPC and dispersive mirrors (b), as well as THz generation and detection setup (c).

Download Full Size | PDF

In order to compress our pulses to sub-100 fs (with the goal of increasing the conversion efficiency of the OR process as well as to access the > 1 THz frequency region), we built a nonlinear compressor based on spectral broadening by SPM in a Herriott-type multi-pass cell (MPC) [28] and subsequent compression using dispersive mirrors (Fig. 1(b)). This technique has recently been demonstrated [29] and has already shown its potential for compressing few-µJ pulses with several 100 fs duration and at hundreds of watts of average power with outstanding efficiency [30,31], which is exactly the operation regime of our compact oscillators. In such an MPC the laser beam undergoes many roundtrips through a suitable nonlinear medium (in the case of our µJ-level pulses, a fused-silica (FS) plate). A small nonlinear phase shift due to SPM is accumulated during each roundtrip, resulting in significant spectral broadening and a much shorter Fourier-limited pulse duration. Such an arrangement allows for circumventing the detrimental effects of self-focusing, resulting in negligible spatio-temporal couplings, good compressibility and excellent beam quality. Our Herriott-type MPC has a similar layout as used in [30], consisting of two 2-inch concave mirrors with a radius of curvature (ROC) of 300 mm, separated by 540 mm and was described in more detail in [32]. The beam undergoes 21 roundtrips, corresponding to 42 passes through a 12 mm thick anti-reflection (AR) coated plate of FS, placed in the center of the MPC. One of the mirrors has a GDD of −350 fs², thereby partly compensating for the material dispersion of the FS plate. After the MPC, the beam undergoes 24 reflections on dispersive mirrors with a GDD of −550 fs² per bounce, compressing the pulse nearly to its transform limit. We characterize our compressed pulses with a home-built second-harmonic generation (SHG) Frequency Resolved Optical Gating (FROG) apparatus [33]. Figures 2(a) and 2(b) show the measured and retrieved traces, which are in excellent agreement. The retrieval was carried out on a 128 × 128 grid with a final FROG error of 2.1·10⁻³. Figure 2(c) shows the unbroadened oscillator spectrum (grey) together with the retrieved (blue) and independently measured (orange) spectra, which are again in very good agreement and further confirm the validity of the FROG retrieval. The peak structure of the obtained spectra is a clear signature of SPM and thereby confirms that this is the main broadening mechanism. Furthermore, we simulate the pulse propagation inside the MPC using a 3D split-step Fourier method as used in [34], which allows us to predict the amount of spectral broadening and final pulse duration for a given MPC layout. We measure a FWHM pulse duration of 88 fs and only small deviations from the calculated Fourier limit as shown in Fig. 2(d). The inset shows the focused beam profile with an average 1/e² beam diameter of 950 µm, used for the THz generation experiment at the position of the GaP crystal.

 

Fig. 2. Compression results measured with SHG-FROG. a) and b) Measured and retrieved traces. c) Retrieved, simulated and measured spectrum and spectral phase. The unbroadened oscillator spectrum is shown for reference in grey. d) Retrieved pulse intensity and phase in the time domain. The transform limit obtained from the measured spectrum is shown for comparison. The inset shows the beam profile used for THz generation at the position of the GaP crystal.

Download Full Size | PDF

The laser shows a close-to-ideal Gaussian beam profile with an M²<1.15 measured directly after the dispersive mirror pair. The overall transmission of the compression stage reaches 94%, mainly limited by the small Fresnel losses on the FS plate. Another 3% are used for various diagnostics, making 112 W available for the THz generation experiment. The peak power of the laser is increased from 15 MW to 80 MW after the compressor. We note that the FS plate inside the MPC had to be shifted twice because of burnt dust particles. After these isolated events, no other significant deterioration has been observed in about 500 h of operation, making this a robust and alignment insensitive compressor for applications.

2.2 THz generation

We generate THz radiation by OR in AR-coated <110>-cut GaP crystals of various thicknesses from 0.5 mm to 3 mm as shown in Fig. 1(c). The different crystals are placed in water-cooled copper mounts and the beam is focused down with a 500 mm focal length lens. The beam diameter is controlled between 320 µm and 950 µm by placing the crystal out of focus. The THz field is collected and refocused by 4 off-axis aluminum mirrors with a 180 mm diameter and 160 mm focal length. A 10 mm hole in the center allows us to separate the strong pump beam from the THz radiation.

The THz electric field transients are fully characterized by an electro-optic sampling (EOS) setup using 500 µm thick GaP as the sampling crystal. The EOS consists of a standard configuration of a λ/4 plate, followed by a Wollaston prism and a balanced photo detector, while a small fraction of the excitation beam serves as the probe beam. The delay is facilitated by a retro-reflector shaking at 15 Hz, providing fast data acquisition [35] and direct visualization of the THz electric field on an oscilloscope. The power values are measured with a calibrated pyroelectric detector (Ophir RM9-THz), and the values obtained were cross checked with a second detector from a different manufacturer (Gentec THZ12D-3S-VP).

We paid special attention to avoiding any residual scattered light of the strong pump laser on the detector, that would otherwise compromise the power measurement. While the main part of the pump beam is separated from the THz beam by the large hole in the first aluminum mirror, any remaining light that hits this mirror is further scattered by the rough aluminum surface, which is not a perfect reflector for the NIR. Pump photons that still reach the detector are then blocked off by a sheet of black paper, whose transmission was characterized using the EOS. We ensured that all remaining pump photons are sufficiently blocked by comparing the transmission obtained from the EOS with that obtained from the power meter, when placing a second sheet of paper in front of it, which showed excellent agreement. The whole setup can be purged with dry nitrogen in order to reduce water absorption of the THz radiation in air.

Figure 3(a) shows EOS traces for three different crystal thicknesses under purged conditions with approximately 10% relative humidity at the full pump power of 112 W. The electric fields are to scale with respect to each other but were offset in time and amplitude for better visibility. We obtain clean single-cycle THz pulses with minimal ringing, which results from residual water absorption that manifests itself in form of absorption lines in the corresponding spectra (Fig. 3(b)). Note that in the current setup we cannot control the relative humidity precisely, which is the reason for the slight variations in the strength of the water absorption lines observed for different crystal thicknesses. The purging conditions will be improved in the future, thereby possibly also further improving the obtained THz power. We observe a decrease in bandwidth when increasing the crystal thickness, which is expected from phase-matching limitations when using thicker crystals. It is worth noting that our 88 fs pump pulses support even broader bandwidths than we generated with the thinnest crystal available at the time of the experiment. By using thinner crystals, we expect to increase the generated bandwidth up to 7 THz, however most likely at slightly lower average powers due to the decrease in generation length (see power values below). To illustrate the effect of the shorter pump pulse duration compared to our uncompressed oscillator, we added the grey curve in Fig. 3(b), which depicts the THz spectrum measured with a 580 fs long pump pulse (directly from the oscillator, i.e. without subsequent compression) in the same setup, and using a 2 mm crystal. The spectrum was significantly narrower and centered at lower frequencies around 0.6 THz, as expected from the longer pump pulse duration. As an indicative comparison with our current results, using these longer pump pulses we measured a THz average power of 182 µW. However, please note that a direct comparison between the obtained power with long and shorter excitation pulses, using the same crystal thickness is not straightforward, as the longer pump pulses were fully phase matched contrarily to our 88 fs pulses. This difference in phase-matching conditions results in an additional power-loss mechanism in the case of the shorter pulses. We refer to the discussion section below for further details.

 

Fig. 3. EOS results for different crystal thicknesses under purged conditions. a) Measured electric fields. The amplitudes are to scale with respect to each other, however offset from each other for better visibility. The inset shows the measured THz focus for the 2 mm crystal. b) Corresponding spectra. A spectrum obtained with a 580-fs driving pulse directly from the oscillator in a 2 mm crystal is shown for comparison. c) Same spectra as in b) on a logarithmic scale. A dynamic range better than 60 dB is achieved by averaging 50 traces.

Download Full Size | PDF

Figure 3(c) shows the same spectra as in Fig. 3(b) plotted on a logarithmic scale, illustrating the dynamic range and full usable bandwidth of our setup. Some additional kinks are clearly visible in the spectra (marked by the black arrows), that are not due to water absorption, but a result of a remaining phase mismatch. At the noise level, our source shows a spectral coverage of up to 6 THz, which is comparable with bandwidths obtained with state-of-the-art photoconductive antennas [36]. By averaging over 50 traces we obtain a dynamic range of better than 60 dB for all investigated crystal thicknesses. However, note that in this experiment, we focused our attention on the generation of the THz transients and did not optimize the detection system for highest dynamic range. A significant improvement can therefore be expected in the future, thus allowing to make full use of the high average power of our source in applications using the corresponding THz-TDS. In order to optimize the conversion efficiency and further understand the limitations in our setup, we measured the THz power obtained as a function of crystal thickness and pump spot diameter. Figure 4(a). shows the results of the pump spot optimization for a 1 mm thick crystal, which delivered highest powers. All power curves show clear saturation around peak intensities of 20 GW/cm², most likely due to free carrier absorption as a result of multi-photon absorption of the pump [15]. By increasing the spot size on the generation crystal, we systematically shift the onset of the saturation to our maximum available average power to achieve the highest THz power. In a next step we varied the crystal thickness at a fixed spot size of 950 µm as shown in Fig. 4(b). The highest average power of 1.35 mW is obtained for the 1 mm thick crystal, but similar values are reached for 2 mm and 3 mm, albeit with a narrower bandwidth. We believe the reason for this is the tradeoff between a longer generation length and the decrease in bandwidth due to phase matching. Only the thinner 0.5 mm thick crystal shows significantly lower power values. No damage of the crystals was observed for any of these measurements, however we avoided pumping too far into the saturation regime.

 

Fig. 4. THz power as a function of pump power. a) Optimization for different 1/e² pump-beam diameters used with the 1 mm crystal. b) Variation of the crystal thickness for a 950 µm spot size.

Download Full Size | PDF

The maximum power of 1.35 mW obtained in this experiment is approximately a factor of four higher than previously reported for GaP [27] and corresponds to 1.2·10⁻⁵ conversion efficiency. Based on measurements of the THz spot size using a microbolometer camera (Spiricon Pyrocam 3) together with the EOS and power measurements we estimate a peak electric field of 7.5 kV/cm at the center of the focused THz beam.

Next, we discuss the possible physical origin of the observed increased power levels, compared to our previous results using longer pump pulses [13]. The maximum power of 1.35 mW is about a factor of 16 higher than our previous result with 580 fs pulse duration. However, we would like to emphasize that this improvement can not only be attributed to the reduction of driving pulse duration. We additionally improved the crystal cooling geometry and use mirrors with larger aperture, thereby collecting more of the strongly diverging THz radiation (see [13] for a more detailed explanation). Based on the above measurements we made with the 580 fs pulses in the current setup resulting in 182 µW average power we can roughly estimate a factor of seven resulting solely from the reduction in pulse duration. This is in principle expected from the theory of OR, which predicts an increase in conversion efficiency for shorter pulses and correspondingly broader THz bandwidth [37]. However, this effect is at least partly reduced by the reduction in bandwidth due to phase matching, which again reduces the conversion efficiency. Furthermore, here we were able to safely apply a pump peak intensity of about 20 GW/cm², while in our previous result we were limited to roughly 4 GW/cm² before observing the onset of damage. This is possibly the strongest contribution to the increased THz output power reported in this experiment. The origin of this effect is not fully understood at this point, although a possible reason could be an improved heat dissipation together with a lower total absorbed energy per pulse due to multi-photon absorption at shorter pulse duration and constant peak intensity. Using even shorter pulses might therefore be beneficial to increase the applicable peak intensity, THz power and THz bandwidth. A more detailed study on the thermal effects, damage thresholds and carrier dynamics in GaP in this excitation regime is currently ongoing. This will provide a more complete picture of the relevant parameters and allow to further optimize pulse duration, focusing geometry, crystal thickness and crystal temperature for the highest output power and/or bandwidth. We believe that further upscaling towards the 10 mW range is feasible by using optimized parameters and higher driving powers.

3. Conclusion

We demonstrate OR in GaP of a nonlinearly compressed high-average power modelocked thin-disk oscillator delivering 88 fs pulses at 112 W of average output power and a repetition rate of 13.4MHz. Under optimized focusing conditions and for the optimal crystal thickness we reach a maximum of 1.35 mW of THz power with a spectrum peaking at 2 THz and extending up to 6 THz. To the best of our knowledge, our source provides the highest average power so far demonstrated using optical rectification in GaP. In the near future, optimization of our EOS detection setup should make this system very attractive for future linear THz-TDS of absorptive samples with very high SNR.

Scaling laws in this unusual excitation regime are currently being explored in a detailed investigation of temperature-dependent effects in GaP. However, our current room-temperature results already show the potential of GaP for further upscaling. In particular, we believe that reaching output powers >10 mW is possible by increasing the pump power, controlling the crystal temperature and optimizing the pulse duration of our source. Furthermore, in optimized phase-matching conditions and using shorter driving pulses which are readily available from the laser system [32] central frequencies extending well beyond 2 THz should be feasible with comparable average power, which would broaden the field of applications of this unique source even further.