1. Introduction

Ultrashort pulses in the deep ultraviolet (DUV) spectral range (200–300 nm) are of high interest for a wide range of applications. In ultrafast spectroscopy these pulses enable studies of excitation dynamics in small molecules [1–3] important biomolecules [4–9] and large-bandgap nanomaterials [10]. Due to the high ionization potential of most molecules [11,12] and high work function of many bulk materials [13], DUV pulses are also beneficial in photoionization experiments to ensure clean single-photon ionization conditions. Moreover, DUV laser pulses are required in seeded free-electron lasers for coherent extreme ultraviolet and X-ray pulse generation [14,15].

DUV pulses with durations of a few tens of femtoseconds are nowadays readily generated [16]. However, to access the full bandwidth of photo-induced processes and to resolve them in real time, DUV pulses of sub-10 fs duration are necessary. The generation and characterization of such ultrashort pulses is much more challenging and only a few experimental demonstrations have been reported, so far [17–25]. The application of sub-10 fs DUV pulses in spectroscopy is equally demanding. To our knowledge only three groups have reported such experiments to date [2,5,7].

The major obstacle in the experimental application of DUV pulses is the strong material dispersion at short wavelengths leading to a significant pulse elongation, even after the propagation through air. Hence, advanced dispersion control schemes are usually necessary. At the same time it has been realized that for many applications high laser repetition rates ($\nu _{\mathrm {rep}}\geq 100\,$kHz) are pivotal. Examples include photoionization imaging methods [26] and coincidence detection [27,28], where space charge effects limit the experimental resolution, thus requiring low laser intensities at high repetition rates. Furthermore, in spectroscopy high repetition rates improve the statistics while avoiding sample saturation and parasitic higher-order nonlinearities [29,30]. Yet, the generation and compression of sub-10 fs DUV pulses at high repetition rates has so far not been shown.

The majority of reported ultrashort DUV pulse generation schemes [17,19–22,24,25] do not support flexible dispersion control, which restricts their application to experiments in vacuum. Moreover, they require intense laser sources of $\sim$ mJ pulse energies which are not available at high repetition rates. An exception is the achromatic phase matching (APM) concept pioneered by the Riedle group [18], which incorporates adaptive optics for flexible dispersion control and implies the possible use of fundamental laser pulses with $\le \mathrm {\mu J}$ pulse energies. A disadvantage of this method is its complexity and the use of an expensive adaptive optical element, which has presumably limited a wide-spread application so far.

In the presented work, we considerably simplify the pulse compression in the APM approach while preserving the wavelength tunability. To this end, we modified the optical geometry, supported by numerical ray tracing simulations and introduced moderate self-phase modulation (SPM) in bulk material. We achieved compression to 10.5 fs of DUV pulses centered at 265 nm using standard optics only, without requiring adaptive or custom-designed dispersive optics. We also demonstrate $<10$-fs DUV-pulse generation for a high-repetition-rate of $\nu _{\mathrm {rep}}=200\,$kHz. This is facilitated by the recent development of Ytterbium femtosecond amplifiers and non-collinear optical parametric amplifiers (NOPAs) for high repetition rates [31]. Due to the high stability and low-noise characteristics of such systems compared to standard Ti:Sapphire-based laser sources, we achieve an excellent noise characteristic of 0.3% (rms) power fluctuations over 5 h for the DUV pulses. The demonstrated relative simplicity and robustness of our approach and the high repetition rate operation opens-up for new applications of ultrashort DUV pulses in imaging and spectroscopy experiments.

2. Experimental setup

The most straight-forward method for broadband DUV pulse production is second harmonic generation (SHG) of broadband visible (VIS) pulses. However, for spectral bandwidths supporting DUV pulses of 10 fs duration, phase mismatch must be kept small, which requires extremely thin nonlinear crystals ($\leq$10 µm for 10-fs Fourier transform limit). This severely limits the SHG efficiency. The limitation can be overcome with the help of APM by introducing an angular dispersion in the incident fundamental laser beam, which matches the type I SHG phase-matching condition inside the nonlinear crystal. With this approach phase matching over a very large spectral range can be achieved [18]. The small phase mismatch in this scheme permits the usage of a factor of $>$10 thicker SHG crystals without reducing the SHG bandwidth. The combination of a relatively thick SHG crystal and APM thus enables both broad phase-matching bandwidth and high SHG efficiency.

The optical setup for the achromatic SHG and pulse compression is shown in Fig. 1. Our laser system consists of a regenerative amplifier (Pharos, Light Conversion, Lithuania, $\lambda =1027\,$nm, 6 W, $\nu _{\mathrm {rep}}=200\,$kHz) which pumps one lab-built and one commercial (Light Conversion) NOPA. The DUV pulses are produced by SHG inside a $\beta$-barium borate (BBO) crystal (thickness $L=100\,$µm) of the output from one NOPA. The output of the second NOPA is used for the characterization of the DUV pulses (see below).

 

Fig. 1. Optical setup for UV pulse generation and compression. P1-P5: dispersion prisms; DM: deformable mirror; BBO: nonlinear BBO crystal; $\lambda$/2: retardation waveplate; $l_1$–$l_5$: tip-to-tip distances between prisms and other optical elements. $f_{1,2}$: effective focal lengths of two $90^\circ$ off-axis parabolic mirrors and one spherical mirror, respectively. Prism P4 is suspended above the incoming beam and picks out the outgoing beam.

Download Full Size | PDF

The optical setup can be understood as a four-prism compressor (prisms P1–P4) with the SHG stage in the center. Accordingly, the first half of the compressor acts in the VIS, whereas the second half acts in the UV spectral range. The first prism pair (P1, P2) induces a well-defined lateral chirp in the horizontal plane of the VIS input beam. Focusing the beam with a short focal length ($f_1$) into the BBO crystal transforms this lateral chirp into an angular dispersion in order to provide the required phase-matching condition for each wavelength component. This part of the setup forms the achromatic SHG scheme, which is complemented by the prisms P3 and P4. The latter serve to recombine the UV components into a single beam after the SHG stage. In addition, prism P3 spatially separates the UV from the fundamental light which is then conveniently blocked by a beam block.

A $4f$-compressor stage (grey area in Fig. 1) is inserted into the optical path between P3 and P4, which is used for compensation of residual temporal chirp in the UV pulses. The compressor consists of the dispersing prism P5, a spherical mirror ($f_2$) and a linear deformable mirror (active aperture = $40\,$mm, 32 electrostatic actuators, max$.$ stroke of 10 µm). The prism P5 increases the angular chirp of the UV beam, while the spherical mirror collimates the beam in the horizontal plane and focuses it vertically onto the deformable mirror. The UV beam coming from P3 first passes underneath P4 and enters the $4f$-compressor stage where a slight tilt of the deformable mirror leads to a vertical offset of the back-propagating beam. The reversed beam is then picked up by the prism P4 at the convergence point of the horizontal angular dispersion to produce a collimated UV output beam.

The achievable phase-matching bandwidth depends on the interplay between prism material, apex angle $\gamma$, prism separation length $l_1$ and focal length $f_1$. In comparison to Ref. [18], we use tighter focusing in order to account for the lower pulse energies from high-repetition-rate laser systems. The change in angular dispersion can be corrected by reducing $l_1$, leading to a more compact setup in our case. We performed ray tracing simulations of the optical setup in order to find optimum design parameters (see Supplement 1 for supporting content). The simulations reveal a strong dependence of the phase-matching bandwidth on the parameters, suggesting a high sensitivity on the optical alignment and the need of highly optimized geometry. However, in practice the UV yield and spectral bandwidth showed much less sensitivity on the parameters of the optical setup. This property is attributed to sum-frequency generation, saturation and SPM effects inside the BBO (see below), which are exploited in our setup to generate larger UV bandwidths. Moreover, the finite beam diameter ($\approx 2\,$mm) of the NOPA in combination with the tight focusing leads to a large angular spread of each spectral component which additionally relaxes the demands on phase matching (see Supplement 1 for supporting content). The surprising robustness against changes in the alignment and geometry of the setup is an important result of our work and puts less stringent requirements for alignment and optimization procedures.

A set of dimensions, which worked well in our setup, are given in Fig. 1. In particular, the prism separation $l_1$ can be changed in the range of 700–900 mm without a substantial effect on the UV yield and bandwidth. The optimal distances between optical elements in the $4f$-stage depends on the rotation angle of prism P5 – all these parameters are optimized for optimal pulse compression. Therefore, $l_3$ is minimized as much as geometrical constraints permit to reduce the dispersion of prism 5 (see below). Further details of the setup are: prisms P1, P2 (fused silica, $\gamma =68.1^\circ$), P3 (CaF$_2$, $\gamma = 68.8^\circ$), P4, P5 (CaF$_2$, $\gamma = 69.9^\circ$). All apex angles are chosen close to Brewster condition to minimize reflection losses. An achromatic $\lambda / 2$-waveplate (25 mm aperture) rotates the laser polarization (initially p-polarized to minimize reflection losses in the prisms) for the type I phase matching in the BBO crystal rotated around the vertical axis. A set of two identical $90^\circ$ off-axis parabolic mirrors ($f_1=20.3$ mm) is used for the focusing into the BBO crystal and recollimation of the UV beem afterwards.

For the optimal SHG efficiency, it is important to have compressed VIS pulses at the BBO crystal and to have a tight focus at the crystal position. We obtain a small, horizontally elongated focal spot of $\approx 20 \times 10$ µm, as measured by a beam profiler, with coinciding horizontal and vertical focus positions. The compression of the VIS pulse depends on the prism separation $l_1$. In previous implementation of achromatic SHG [18], the VIS pulses were over-compressed at the crystal position. In contrast, our VIS compressor stage is approximately 40% shorter leading to almost full compression (slight under-compression) of the VIS pulses for minimum insertion of prisms P1, P2. Under these conditions, significant SPM and self-focusing is observed for crystal positions close to the laser focus. Therefore, we slightly displace the crystal from the focus position to reach a compromise between SHG efficiency and undesirable third-order effects.

Since our VIS pulses are compressed at the SHG crystal, sum-frequency mixing processes may contribute to the UV generation leading to a change of the wavevector distribution in the UV beam. By displacing the SHG crystal from the focus position, the different spectral components are mostly spatially separated and frequency-mixing between different wavelengths is minimized. Nonetheless, for BBO crystals with thickness $L=360$ µm, we experienced a strong horizontal divergence of the UV beam at the output of the setup which we partly attribute to sum-frequency mixing inside the thicker crystal. The problem is alleviated by using a thinner BBO crystal with $L=100$ µm.

From the design of the optical setup involving prisms of different refraction index and different apex angles, one may expect sub-optimal collimation and a residual lateral chirp on the output beam. However, our ray tracing simulations show that beam distortions and lateral beam dispersion introduced by the optical setup are negligible. The simulations yield the residual lateral chirp to be of the order of $dx_0 / d\lambda \approx 1.1\,\mathrm {\mu m / nm}$ around the center wavelength, which is negligible for most applications ($x_0$ is the lateral beam position and $\lambda$ is the wavelength). Likewise, only a negligible angular dispersion of $d\theta / d \lambda \approx 0.003^\circ / \mathrm {nm}$ is found. This is in agreement with our experimental observations. We achieve a good beam quality of the UV output beam, with vertically elongated cross-section and with small spatial chirp. A slight divergence in the horizontal plane however remains.

3. Results

3.1 Tunable broad-bandwidth DUV spectra

We generated DUV pulses with very large spectral bandwidths, featuring a full-width at half maximum (FWHM) of up to 63 nm [Fig. 2(a)], which corresponds to a Fourier transform limit of 1.5 fs. Such broadband pulses are ideal for coherent nonlinear spectroscopy experiments to reveal couplings and complex dynamics [7,8,10]. For comparison, we show the calculated spectrum for SHG without using APM, which is a factor of $\approx 30$ narrower.

Another advantage of achromatic SHG is the straightforward wavelength tuning over a wide spectral range [18]. By tuning the NOPA wavelength and rotating the SHG crystal in the UV setup, we can readily produce pulses at different DUV wavelengths [Fig. 2(b)]. All DUV spectra exhibit spectral widths $>$79 THz which corresponds to Fourier transform limit well below 10 fs. Once the fundamental wavelength is set, additional alignment other than the crystal position relative to the focal point and crystal rotation is not necessary for the wavelength tuning. While these features were demonstrated previously [18], we show here that broadband spectra down to a central wavelength of 250 nm can be reached. In addition, we introduced SPM inside the BBO crystal to spectrally broaden the UV pulses. We kept the SPM at a level where the spatial mode of the UV beam is not affected and no spectral modulations occur, yielding a spectral broadening of the UV spectra in the range of $20$–$70$%. Similarly, SPM in a thin CaF$_2$ plate has been previously used to achieve spectral broadening and to improve the pulse compression by a factor of 2.4 for DUV pulses [32]. In general, the combination of APM and moderate SPM inside the SHG crystal makes the setup more robust against geometrical changes and simplifies considerably the generation of broadband, tunable DUV pulses.

 

Fig. 2. UV spectra. (a) Example data for a very broad-band UV spectrum (black) along with a theoretical spectrum obtained without APM (grey). (b) UV spectra obtained by tuning the VIS NOPA wavelength and the BBO crystal in the APM setup. Numbers indicate the FWHM of the spectra.

Download Full Size | PDF

3.2 Pulse compression using adaptive optics

As mentioned, both pulse compression and characterization are challenging in the DUV spectral domain, especially for broadband pulses in the sub-10 fs range and with low pulse energies, as in our setup. Another complication with ultrashort DUV pulses is the significant dispersion caused by the propagation through material and even through air. Therefore it is essential to pre-compress the DUV pulses in order to achieve compressed pulses at the interaction spot of the actual spectroscopic/imaging experiment. For a reliable pulse characterization, we thus measure the DUV pulses after propagating them an optical path equivalent to the one in the intended spectroscopic experiment. Originally the APM concept required a sophisticated pulse compression scheme combining a four-prism compressor (half in the visible and half in the UV part) with a pulse shaper based on a deformable mirror [18]. This approach, in principle, facilitates flexible dispersion control and pre-compression of the DUV pulses, which is an asset over most other DUV generation schemes.

For the low UV pulse energies in our setup ($\leq 50$ nJ), a robust pulse characterization with large acceptance bandwidth, minimum dispersion and high sensitivity is required. Most methods for characterization of DUV pulses [17,33,34] lack either of these features or involve complex optical setups, which are difficult to align and/or calibrate [23,35,36]. Here, we adapted two-color TG-XFROG [37] to the bandwidth and pulse length requirements of our setup. The DUV pulses are characterized via cross-correlation with a transient grating induced by near-infrared (NIR) pulses of $\approx 10$ fs duration in a very thin ($< 10$ µm) fused silica plate (Valley Design Corp., USA). In our scheme, the NIR pulses, produced by the second NOPA, are separately characterized by interferometric SHG FROG measurements (see Supplement 1 for supporting content). Since the UV and NIR pulses are of a similar duration, the pulse length uncertainty in the deconvolution of XFROG traces is minimal, and we observed robust XFROG reconstruction results.

The optimal compression of the DUV pulses requires optimization of the VIS-UV prism compressor as well as of the deformable-mirror-based $4f$-compressor stage. As previously demonstrated [23], good compensation of the second-order phase distortion can be reached with the VIS-UV prism compressor, while the third- and higher-order phase distortion is minimized with the deformable mirror. The third-order contribution can be substantial, especially since the length of the VIS-UV prism compressor is determined by the APM condition and cannot be easily adapted to the chirp of the laser pulses. Here, a shorter focal length of the parabolic mirrors leading to a shorter prism compressor is of advantage. For our geometry, we find best compression results at minimum insertion of prisms P1–P5. Fine tuning is most conveniently done by the insertion of P3. The deformable mirror is optimized with an evolutionary algorithm maximizing the amplitude of the two-color transient-grating cross-correlation traces. The algorithm typically converges within 70 iterations.

We compressed DUV pulses at various wavelengths to $< 10$ fs duration, including pulses at 255 nm. As an example, for the spectrum centered at 265 nm [Fig. 2(b)] and having 19-nm-bandwidth (6.3 fs Fourier transform limit), we obtain the pulse duration of 8.4 fs (Fig. 3). Remarkably, the obtained temporal profile is very clean with no discernible side lobes [Fig. 3(d)]. This is in contrast to previously reported sub-10 fs pulses in the DUV spectral range. A slight mismatch in the temporal width of the measured and retrieved FROG traces can be observed [Figs. 3(a) and 3(b)] as well as some artificial spectral modulation in the reconstructed pulse spectrum [Fig. 3(c)]. The spectral modulations were absent in the UV spectra directly measured with a spectrometer [see Fig. 2] and seem to be a common feature of the TG-XFROG retrieval algorithm [37]. The TG-XFROG yielded consistent results for multiple pulse measurement and characterization attempts under various experimental conditions, implying that the method is robust. We also varied the amount of SPM inside the BBO crystal and found that it does not impair compression of the UV pulses. However, by testing different geometries of the TG-XFROG setup we noticed, that the pulse characterization method becomes insensitive for shorter pulse durations, which limits the recursive pulse compression scheme with the deformable mirror. This suggests that in principle even shorter pulses should be feasible in our setup.

 

Fig. 3. TG-XFROG characterization of DUV pulses using NIR pulses of 10.4 fs duration for the transient grating. (a) Measured and (b) reconstructed FROG traces. (c) Retrieved DUV spectrum (solid) and spectral phase (dashed). (d) Reconstructed temporal intensity profile.

Download Full Size | PDF

3.3 Pulse compression with standard optics

In general, it is desirable to avoid adaptive optics in the optical setup, especially for spectroscopy and microscopy experiments with long acquisition times. Adaptive optics can cause shaping artifacts, may introduce beam distortions, or are simply not available in most laboratories [38–41]. Yet, in order to compress optical pulses to durations in the 10-fs range, simultaneous compensation of second- and third-order phase distortions is necessary, which usually requires pulse-shaping optics or tailored dispersion mirrors. Accordingly, in the APM scheme, the prism compressor was complemented by an additional compressor stage for compression to the sub-10-fs regime. However, in contrast to these considerations, we found that careful optimization of the compressor enables pulse compression very close to 10 fs even without the use of adaptive optics. We tested this concept using a flat deformable mirror membrane (all actuators set to 0 V) and alternatively replaced the deformable mirror by a flat mirror.

Since prism P5 introduces a large amount of dispersion in the UV pulse, which is amplified by the double-pass geometry, it is of paramount importance to reduce the insertion of P5 and place it as close as possible to the spot where all spectral components converge. This is achieved by choosing prisms with a small side length (15 mm) for P4 and P5 and arranging a mount for the P4 prism to suspend it upside down. Furthermore, our experimental efforts as well as ray tracing calculations revealed that the introduced chirp depends very sensitively on the prism P5 rotation angle. We thus adjust the angle of prism P5 to get a horizontal beam diameter of 11–16 mm in the Fourier plane of the $4f$-setup, which is a compromise between spectral resolution required for the third- and higher-order compensation and the chirp introduced by the prism itself. The combination of tuning the P5 rotation angle and the position of the spherical mirror in the $4f$-compressor, i.e., by varying $l_4$, $l_5$ (Fig. 1), leads to sufficient third-order compensation to achieve compression close to $10\,$fs pulse duration. The results are shown in Fig. 4 for a DUV pulse centered at 265 nm.

 

Fig. 4. Pulse characterization of a DUV pulse compressed with standard optics. Here, the deformable mirror was powered-off, effectively yielding a flat mirror surface. (a) Measured and (b) reconstructed TG-XFROG traces. (c) Retrieved DUV spectrum (solid) and spectral phase (dashed). (d) Reconstructed temporal intensity profile.

Download Full Size | PDF

The measurements are supported by the ray tracing simulations, which calculate the temporal dispersion of the entire UV generation setup (as shown in Fig. 1) and the dispersion resulting from the propagation to the pulse characterization point. The chirp of the NOPA pulses used for the achromatic SHG was estimated from the material dispersion inside the NOPA and the propagation to the UV setup (see Supplement 1 for supporting content). In the simulations, we minimize the temporal dispersion of the setup by varying the geometry of the $4f$-compressor stage as well as insertion, orientation and separation of the UV prisms (P3, P4 and P5). The amount of insertion of prisms P1 and P2 is also tuned by the optimization routine. An optimum geometry is found for $l_4=239\,$mm and $l_5=248\,$mm which is close to the dimensions in the experimental setup (Fig. 1). The ray tracing results are presented in Fig. 5. Using the measured UV spectrum, we obtain a theoretical pulse duration of 8.8 fs, which is in good agreement with the experimental data. This result implies that DUV pulses of $\approx 10$ fs duration can be achieved solely with standard optics, without the need of advanced compression techniques based on deformable mirrors, liquid-crystal displays or acousto-optic elements [42–44].

 

Fig. 5. Ray tracing simulation of the overall optical setup assuming a flat deformable mirror surface. (a) Experimental pulse spectrum (gray) corresponding to the measured pulse in Fig. 4 along with the group delay (blue dashed line) introduced by the optical setup calculated with ray tracing. (b) Corresponding temporal intensity profile with a FWHM of $8.8\,\mathrm {fs}$, obtained by a Fourier transform of the experimental pulse spectrum and including calculated group delay.

Download Full Size | PDF

We note that our optimized geometry also favors the compression scheme using the deformable mirror. Since the phase distortions are already well compensated, the compression to $<10$-fs-pulses only requires a slight bending of the deformable mirror membrane, which greatly minimizes beam distortions.

3.4 Efficiency and stability

Due to the high repetition rate of the laser system, pulse energies are naturally lower than in low-repetition-rate laser setups. As outlined above, the lower intensity is compensated in our setup by tight focusing and minimizing the pulse duration of the VIS pulses at the position of the SHG crystal. This enabled us to consistently reach SHG conversion efficiencies of 25% for 340 nJ incident pulse energy on the BBO and 440 nJ/pulse in the input beam entering the APM setup, respectively. Even for a factor of 40 lower input pulse energy (-10 nJ/pulse), we reached the same conversion efficiency and similar UV bandwidth of sub-10-fs transform limit. This indicates that ultrashort DUV pulse generation with laser repetition rates up to the 10-MHz range is feasible. We attempted to increase the conversion efficiency further, e.g., by additionally compressing the VIS pulses with chirped mirrors. The increase of SPM and self-focusing inside the BBO crystal, however, limits reaching higher efficiencies and a conversion efficiency of 28% seems to be the upper boundary in our geometry. The overall efficiency of the setup is $\approx 11$% (VIS input to compressed UV output) and we get typically $\approx 10\,$mW average UV power, which corresponds to pulse energies of 50 nJ.

Furthermore, the setup features a high robustness and stability (Fig. 6). The power fluctuations of the UV output over 5 h are typically 0.3% (rms) for broadband sub-10 fs pulses. This is a factor of $\sim 2$ larger than the fluctuations of the VIS laser power, as expected for a second order nonlinear process. Furthermore, no realignment of the setup was necessary over the course of several months. The good efficiency of the setup, its high power stability and robustness against misalignment provide ideal features for spectroscopy applications.

 

Fig. 6. UV output power stability measured over a duration of 5 h.

Download Full Size | PDF

4. Conclusions

We have demonstrated the generation of sub-10 fs DUV pulses at high pulse repetition rates of 200 kHz using a combination of achromatic SHG and adaptive pulse compression. By geometrical optimization of the setup, DUV pulses close to 10 fs have been compressed even without adaptive optics, but solely with standard optics. This finding opens up the presented generation and compression method for many more laboratories. The DUV pulses were characterized with two-color TG-XFROG using NIR reference pulses of $\approx 10$-fs duration. Despite the low pulse energies we obtain a remarkable high frequency-doubling efficiency of 25%. Our setup is insensitive to misalignment as well as exact geometry and features high power stability of 0.3% (rms) over five hours. The combination of these features makes the DUV pulse generation ideal for a wide range of applications, including ultrafast multidimensional spectroscopy and time-resolved photoelectron spectroscopy, with particular benefits for coincidence methods, and ultrafast photo-emission electron microscopy.