1. INTRODUCTION

Ultrashort pulses tunable in the deep ultraviolet (DUV) down to the transmission limit of air (slightly below 190 nm) are important for spectroscopic applications. The needed pulse energies are in the range of hundreds of nanojoules, e.g., for the probe pulse in time-resolved photoelectron spectroscopy [1]. The direct excitation of bound higher electronic states in small molecules also can readily be performed with such a modest pulse energy and even the generation of solvated electrons from neat solvents has been demonstrated in this range [2]. In all mentioned applications, extremely fast processes are investigated, typically on the time scale of skeletal motions, i.e., with periods in the range of 20 to 50 fs. The investigations therefore need pulses with a duration significantly below 50 fs.

Early work was focused on the generation of the harmonics of the broadly available Ti:sapphire amplifier in consecutive nonlinear optical crystals [3,4]. No particular effort for the management of the chirp due to the dispersive materials was undertaken. The amount of traversed material was minimized by suitable geometries and the use of very thin crystals. In this way pulses in the 100–200 fs regime could be generated from suitably short 800 nm pulses. Tunability was added by a 800 nm pumped optical parametric amplifier (OPA) [5]. The OPA also is able to add spectral width to the final pulse and therefore allows in principal the generation of a DUV pulse shorter than the original pump pulse.

Later, frequency conversion in gases was demonstrated in various schemes. These include in particular four-wave-mixing of different harmonics of the Ti:sapphire pump laser [6,7] or with the output of an OPA [8–10]. For the enhancement of the efficiency even hollow fiber guiding had to be used [9]. In all attempts, rather high pump pulse energies were needed. Besides multimillijoule short pulse length Ti:sapphire amplifiers [6,7], a 300 μJ noncollinear OPA (NOPA) pumped by 10 mJ pulses at 800 nm was central to the conversion [10].

None of the schemes renders the needed combination of pump and output parameters of a prototype spectroscopic experiment: $\sim 1\mathrm{mJ}$ pump energy of $\sim 100\mathrm{fs}$ duration and fully tunable 30 fs output pulses with well above 100 nJ in the range from 189 to 240 nm, all with high reliability and low pulse fluctuations. In this contribution we will demonstrate that these goals can readily be achieved with an all-solid-state system pumped by less than 400 μJ at the Ti:sapphire fundamental. No compressor in the UV is needed despite the operation under ambient air. Instead we use a single two-prism arrangement in the visible output of the NOPA that provides the fundamental tunability. The proper chirp management in both the second-harmonic generation (SHG) and the sum-frequency generation (SFG) stage with part of the laser fundamental is sufficient to ensure the shortest DUV pulse length. By tailoring the spectral bandwidth to the acceptance bandwidth of the crystals, we achieve a good conversion efficiency despite the low pump energy and the extremely small crystal thicknesses.

2. CONCEPT FOR THE FREQUENCY CONVERSION AND THE CHIRP MANAGEMENT

The principal idea of our generation scheme is shown in Fig. 1. A commercial Ti:sapphire amplifier system (CPA 2010; Clark-MXR) with an output wavelength of 779 nm and a pulse duration of 200 fs pumps a NOPA that provides the necessary spectral bandwidth for the complete conversion chain [11,12]. Only 280 μJ of the available 750 μJ at 1.9 kHz repetition rate are used. This allows the operation of a second NOPA and even leaves some residual power for additional demands in the spectroscopic experiment. The horizontally polarized NOPA output is overcompressed in a double-passed pair of fused silica Brewster prisms. Subsequently it is frequency-doubled in a thin BBO crystal and the resulting tunable mid-UV pulses are mixed in a further BBO crystal.

 

Fig. 1. Schematic of the multistage frequency conversion to generate tunable 30 fs pulses in the deep UV (from 189 to 240 nm).

Download Full Size | PDF

For the design of our setup we start in the DUV. From the desired tuning range of 189–240 nm we can determine the intermediate UV tuning range to 250–347 nm. The corresponding fundamental range of 500–694 nm is readily reachable with the NOPA. We can also verify by the use of the program SNLO [13] that the necessary phase matching can be achieved in BBO. The required angles for type I phase matching range from 50.4° to 74.8°. Other commercially available crystals either do not allow such a wide tuning range or have a significantly lower effective nonlinearity.

The thinnest BBO crystals that we found to be commercially available without a glass support are nominally 30 μm. The one used in our experiments was manufactured and mounted by Castech Inc. and distributed by DÖHRER Elektrooptik GmbH. Ultrathin crystals cemented onto a support can most likely not withstand the extreme intensities in the UV and are thought to be only useful for pulse characterization. Careful handling avoiding mechanical shock or any attempts to clean the crystal is needed to ensure years of usage. For the 30 μm crystal we had to ensure by our own measurements that the crystal is indeed so thin. This can be done by observing the interference fringes in a high-resolution spectral photometer.

For the measured crystal thickness of 32 μm we determine the acceptance bandwidth (FWHM; 29 THz for 198 nm generation) and set the bandwidth of the intermediate UV slightly larger. Since the bandwidth of the 779 nm light is just 3 THz, the generated bandwidth of the SFG stage will be equal to the acceptance bandwidth if sufficient UV width is available. For perfect compression, DUV pulse lengths below 20 fs are possible for the whole range from 189 to 240 nm.

Next we find that a BBO crystal in the range of 50 μm can produce this bandwidth in the SHG step. We show the calculated UV bandwidth obtainable in a 50 μm crystal in Table 1. This estimate assumes that the visible pulse supplies enough bandwidth. Numerically we find that a visible pulse with 50% more bandwidth than the SHG acceptance bandwidth only leads to a 10% decrease in the generated UV bandwidth. As this leads to some loss of conversion efficiency, we opted to match the visible bandwidth to the UV bandwidth (values shown in Table 1). Any spectral components outside the acceptance bandwidth will only be poorly converted and at large spectral distances even possess a phase shifted by $\pi$. This makes the resulting pulse practically impossible to compress. To tailor the NOPA output, one can insert a glass block of suitable group delay dispersion after the seed light generator in the NOPA. For a variation of the DUV wavelength a slight variation of the material length or glass should be considered.

Table 1. Design Parameters and Actually Measured Values for the Generation of the DUV Pulses

View Table

As a final step we perform a backward determination of the pulse chirp. We know that we want an optimally compressed pulse in the DUV. Propagation through air from the SFG crystal will already add chirp as does the SFG crystal. This is explicitly accounted for. The SHG step simply halves the linear chirp of the visible pulse [14], i.e.,

To set the proper linear chirp of the NOPA output, we increase the prism separation in the compressor sufficiently to obtain strongly down-chirped pulses. Usage of the prisms as close as possible to the apex is essential to avoid unnecessary higher-order chirp contributions. Supplementary calculations and previous experience [14] show that in the present situation the influence of the higher-order chirp does not yet influence our pulse parameters strongly. This is a consequence of the still modest demand for 30 fs pulses in the DUV. Accordingly, the spectral widths are quite small compared to what is typically achieved from a fully bandwidth optimized NOPA [16].

3. EXPERIMENTAL REALIZATION AND CHARACTERIZATION

The visible pulses out of the NOPA and the compressor are focused with an $f=250\mathrm{mm}$ spherical mirror with silver coating [see Fig. 2(b)]. The focal length is not particularly important, but the beam size has to be kept small enough to stay within the acceptance angle of the 50 μm BBO crystal of about 38 mrad. The intensity at the focus is way too high for proper SHG despite the long pulse length at this point. We find that the SHG crystal is best placed about 1 cm behind the focus, just at the point where a further approach to the focus does not increase the efficiency anymore. In this way a clean and round beam profile can be obtained. When the compression is optimized for the shortest DUV length, the position of the SHG crystal might have to be varied slightly.

 

Fig. 2. (a) Experimental setup for the simultaneous variation of the NOPA compression and the delay of the 779 nm Ti:sapphire fundamental. (b) Schematic of the nonlinear mixing scheme. The relevant pulse parameters (center wavelength, linear chirp ${\mathrm{\Phi }}^{\prime \prime }$, and pulse length) are indicated.

Download Full Size | PDF

The emerging UV beam of about 1 μJ pulses is recollimated and focused with just one $f=250\mathrm{mm}$ dielectric mirror. Again the numerical aperture is the decisive value to watch. In addition, as small as possible off-axis angles are used to avoid excessive astigmatism. The SFG crystal is placed slightly behind the UV focus and the supplementary 779 nm beam (pulse energy 80 μJ) is overlapped within the crystal. The red beam is focused with a fused silica lens, since its bandwidth is quite small. Throughout our setup no dichroics are needed, as the small angle between the UV and the red allows geometric separation of the DUV pulse. The resulting weak spectral chirp is found to be much smaller than the natural divergence [17] and should therefore not influence the use of the DUV pulses.

To optimize the compression of the DUV, the effective prism separation of the folded visible compressor is varied as shown in Fig. 2(a). This demands an exactly equal change in the path length of the red sum beam. To ensure this linkage, we place a delay line for the 779 nm beam onto the manual delay line inserted between the prisms.

With the setup we were readily able to generate pulses from 193 to 219 nm. Shorter wavelengths were not attempted, as proper steering mirrors and appropriately cut crystals were not available. Longer wavelengths than 220 nm were not of interest in our own laboratory, as they can be readily obtained by direct doubling of the NOPA. We are, however, sure that a good spectral overlap of the two schemes can be achieved. The width of the spectra [see Fig. 3(a)] allows for approximately 20 fs pulses throughout the tuning range [Fig. 3(b)]. The pulse energy [see Fig. 3(c)] was not fully optimized as the beam transport in the visible and UV was generously designed due to other experimental demands. This resulted in a rather modest SHG conversion from 15–20 μJ NOPA output (10–13 μJ after the compressor) to the typical 1 μJ intermediate UV. Still, pulse energies well above 100 nJ were measured throughout the DUV tuning range corresponding to energy conversion efficiencies from the intermediate UV to the DUV of well above 10%. At particularly carefully optimized wavelengths we achieved much higher pulse energies up to 630 nJ. We believe that mechanics with higher adjustment sensitivity will readily lead to similarly high output throughout the tuning range. The pulse energies correspond to milliwatts of average power in the DUV. Visually we do not notice any significant fluctuations and the DUV is stable over many hours. The beam shape is close to Gaussian. No degradation of either the optics or the crystals has been observed.

 

Fig. 3. (a) Spectra of fully tunable DUV pulses. (b) Fourier limit of the pulses derived from the spectral width. (c) Measured energies.

Download Full Size | PDF

To fully characterize the generation scheme and to validate the theoretical considerations, we measured the spectral width and pulse lengths with fiber-coupled spectrometers and autocorrelators of our own design [18,19]. The relevant values are summarized in Table 1. They compare very well with the predictions and thus confirm our concept. Strictly speaking, the acceptance bandwidths are calculated for the collinear case. However, they are only altered by less than 1% if the small angle used in the experiment is considered explicitly. Remarkably, any possible higher-order chirp does not prohibit the extremely simple and straight forward multistage conversion scheme.

Representative spectra and autocorrelation traces are shown in Fig. 4. Note that the visible pulse is 324 fs long and the intermediate UV pulse 119 fs. They were measured with a 100 μm thick BBO crystal for the visible SHG autocorrelation and a 50 μm thick BBO crystal for the two-photon absorption UV autocorrelation. From the design we know that at these points the pulses are down-chirped and propagation through normally dispersive material compresses them without any additional loss in energy. In this way we perform all the critical compression efforts in the visible region where the highest optical quality is available. The 198 nm pulse was characterized by two-photon absorption in a 50 μm thick sapphire disk, which overestimates the pulse length by less than 1 fs. To determine the pulse length from the autocorrelation curve we fit a Gaussian to it. This fits nicely and in addition shows that there are no detectable shoulders or satellites to the DUV pulse. This confirms our assumption that higher-order contributions to the chirp do not lead to a pulse substructure. The deconvoluted pulse length is 31 fs and about 35% above the Fourier limit of 22.9 fs calculated from the measured spectrum. Our scheme for generating compressed DUV pulses with just a compressor in the visible does indeed work quite well.

 

Fig. 4. Spectral and temporal characterization of the DUV and intermediate pulses.

Download Full Size | PDF

The FWHM spectral width of the 198 nm pulse is smaller than the predicted one. We believe this to be due to the fact that the red mixing pulse is roughly of the same length as the mid-UV pulse. We have shown previously that this leads to a decreased conversion efficiency in the spectral wings of the chirped pulse [20], which we do indeed see. To overcome this limitation, which will be even stronger for still shorter wavelengths, the red pulse could also be chirped [15,20]. As sufficient red pulse energy is available, this will not lead to a decrease of efficiency.

4. SUMMARY AND CONCLUSIONS

We have presented a straight forward and simple to implement scheme for the generation of fully tunable DUV pulses. We generate energetic visible pulses with controlled bandwidth and down-chirp from a NOPA followed by a long prism compressor. These pulses are then frequency-doubled in a 50 μm thick BBO crystal and subsequently mixed with supplementary 779 nm light in a 32 μm BBO crystal. All processes are fully phase matched. No additional compression is needed in the UV or DUV and even a thin entrance window into a vacuum chamber can be precompensated [14]. Pulse energies in the submicrojoule regime have already been demonstrated and could readily be increased into the microjoule regime.

The output energies are presently limited by our NOPA design that limits the pump energy to below 300 μJ. With a high-power NOPA [21] we would immediately expect more than an order of magnitude higher output energy, getting us into the solid microjoule regime. The present energy efficiency is roughly 1‰, as compared to values of about ${10}^{-9}$ for recently reported tunable high-harmonic generation in krypton [22] and 0.1‰ for nonresonant four-wave mixing in argon [7]. This shows that the conversion in crystals is an excellent method if pulses above 189 nm—the accepted transmission edge of BBO—and in the 30 fs regime are needed. This has also been corroborated recently at 250 kHz repetition rate where about 10 μW average power at 198 nm [23] and well above 100 nJ at 226 nm and 100 kHz [24] were reported, however without the possibility of wavelength tuning. Even shorter DUV pulses might become feasible with newly emerging crystals like ${\mathrm{KBe}}_2{\mathrm{BO}}_3{\mathrm{F}}_2$ (KBBF) [25]—provided they become generally available.

The presented setup allows for the conversion of the 779 nm pulses of our Ti:sapphire amplifier system to the DUV. In addition, we achieve a pulse shortening from 200 to 30 fs, more than a factor 6. The necessary spectral bandwidth is provided by the NOPA. A NOPA and the continuum seed source can even be pumped by still longer pulses, e.g., from ${\mathrm{Yb}}^{3+}$ based laser systems [17]. The conversion scheme can be easily transferred to the center wavelength of around 1030 nm. The NIR pump wavelength even results in a 20% larger mixing bandwidth in the SFG step. This will allow the generation of 30 fs DUV pulses with a repetition rate of 100 kHz.

The tunable DUV pulses in the 30 fs regime are believed to be of immediate use for many spectroscopic experiments. In time-resolved photoelectron spectroscopy they can render more electron kinetic energy as compared to the typical mid-UV pulses and the tunability can help to minimize the generation of detrimental one-color signal contributions. DUV pulses are also of great interest for the investigation of small polyatomic molecules to bridge the gap between the many ongoing investigations on ${\mathrm{H}}_2$ and the wide field of work on organic molecules. All this comes at very low complexity and high stability of the setup due to the fully solid-state design.