1. INTRODUCTION

Ultrashort light pulses ranging from the extreme-ultraviolet (XUV) to the hard x-ray regime are available from several x-ray free-electron laser (XFEL) facilities worldwide [1]. The latest generation of these is based on superconducting radio-frequency (SRF) technology and provides pulses at repetition rates from the few 10 Hz to 100 MHz range [2,3], thereby enabling one to probe dynamics in materials, molecules, and atoms on all relevant time, length, and energy scales with excellent sensitivity [4–6]. However, despite this progress, the intrinsic and often unknown fluctuations of the x-ray pulse properties still do not allow one to realize the full potential of the many fascinating research opportunities.

In this paper, we present a novel concept to determine critical x-ray pulse parameters such as arrival time, pulse duration, and pulse energy changes. Our approach is based on the analysis of terahertz (THz) light pulses generated by the femtosecond x-ray/XUV pulses via optical rectification inside a broadband spintronic THz emitter (STE). Details of the experimental setup are shown in Fig. 1. We show that the envelope of the x-ray/XUV pulse is encoded into the ultrashort photocurrent inside the x-ray-irradiated STE. By sampling and analyzing the full electric field of the concomitantly emitted THz pulse, we are able to retrieve information about the arrival time, duration, and energy of the original x-ray/XUV pulse. The advantage of our concept in comparison to other schemes to derive the arrival time [7–12], pulse duration [13–18], and pulse energy [19] is its simplicity, robustness, and potential to obtain all of these XUV/x-ray pulse properties from one single measurement. Our results also demonstrate optical rectification of XUV pulses with wavelengths one order of magnitude shorter than in all previous experiments [20–22].

 

Fig. 1. Experimental setup as employed at the FERMI free-electron laser. (a) Schematic showing all relevant components of the seeded free-electron laser and implementation of the THz monitor concept, (b) plot of three THz pulses emitted from three subsequent XUV pulses with an average pulse energy of 20 µJ and wavelength of 20 nm, and (c) histogram showing the observed jitter of below 13 fs (rms) for the nominally intrinsically synchronized seeded x-ray pulses from the FERMI free electron laser.

Download Full Size | PDF

2. EXPERIMENTAL SETUP

Our experiments were performed at the EIS-TIMEX beamline of the XUV free-electron laser FERMI in Trieste/Italy. As schematically shown in Fig. 1(a), the XUV pulses, generated by a high-gain harmonic-generation (HGHG) seeding scheme, were normally incident onto a purpose-built, large-area ${10} \times {10}\;{{\rm mm}^2}$ STE (TeraSpinTec GmbH) [23–25], which consists of a ${\rm Pt}({2}\;{\rm nm}){\rm |CoFeB}({1.8}\;{\rm nm}){\rm |W}({2}\;{\rm nm})$ tri-layer on a 500 µm thick $z$-cut sapphire substrate. The sample was magnetized in-plane by a constant magnetic field of 100 mT. The emitted THz pulse propagates in the forward direction through the sapphire substrate before entering free space.

The pulse energy of the XUV pulses could be adjusted up to 42 µJ per pulse. The experiments were performed with wavelengths of 20 and 29 nm and an estimated pulse duration of 30 and 35 fs, respectively. The repetition rate was 25 Hz. To operate the STE in the linear regime, the pump-beam diameter at the sample position was chosen to be 8 mm, thereby keeping the pump fluence below ${0.1}\;{{\rm mJ/cm}^2}$ [24]. The emitted THz pulses were detected by a purpose-built single-shot detection scheme. The scheme is based on electro-optic encoding of a co-propagating THz electric-field waveform onto the frequency spectrum of the chirped optical probe pulse (1.6 ps, full-width half-maximum), and subsequent decoding by decomposing the probe beam into the frequency spectrum as shown in Fig. 1(a) (see [26,27] and Supplement 1 for details). The single-shot traces were additionally denoised by spectral filtering [27].

3. RESULTS

A. Arrival Time Monitoring

Figure 1(b) shows three consecutive THz pulses emitted from the STE upon impact of three subsequent XUV pulses. They exhibit not only different amplitudes, but as can be seen in the magnified view (right panel), differences in arrival time of only a few femtoseconds can clearly be observed. A histogram of the jitter as determined by the arrival time monitor for 2500 subsequent XUV pulses is shown in Fig. 1(c). The typical arrival time jitter of FERMI observed in our measurement is below 13 fs (rms). Our results compare well with earlier jitter measurements performed at FERMI, which yielded values between 5.9 and 9.8 fs (rms) [12]. The slightly larger jitter observed in our scheme may stem from instabilities in the optical paths of our setup or from a slightly decreased precision of our measurement for lower XUV pulse intensities (see Supplement 1). Another possibility is that the jitter of the facility was simply larger during our measurements.

To characterize the robustness and reliability of our scheme, we artificially introduced periodic arrival time variations of the optical delay line between the common oscillator and seed laser optical parametric amplifier (OPA), thus altering the arrival time of the seed laser on the electron bunch [see Fig. 1(a)]. The results are presented in Fig. 2 where the determined arrival times for in total 600 subsequent pulses are plotted. A timing jitter of 10 fs can easily be detected with our arrival time monitor as can be seen in Fig. 2(b). The observed timing jitter for a delay line of 0 fs, corresponding to the intrinsic timing jitter of the FERMI FEL convolved with the precision of our arrival time monitor, is shown in Fig. 2(c). This particular measurement yields a timing jitter of below 13 fs (rms), as also shown as a histogram in Fig. 1(c).

 

Fig. 2. Benchmarking of the scheme as arrival time monitor: 2D plot of the THz electric field for 600 consecutive pulses with artificial timing jitter of 40 fs (left, color coding represents field strength); white line represents the central peak of the THz waveform. (a) Corresponding line-out of the temporal position of the central maximum of the THz waveform for an artificial jitter of approximately 40 fs, (b) line-out for approximately 10 fs, and (c) line-out for 0 fs. The light blue line indicates artificially introduced arrival time variations, and the light red line denotes the observed arrival time jitter. The blue and red lines are smoothed values of the introduced and observed time jitters. The XUV wavelength for these measurements is 20 nm, pulse duration is 30 fs, and pulse energies vary between 5 µJ and 30 µJ.

Download Full Size | PDF

B. Pulse Energy Monitoring

In the current understanding of STE operation, the absorbed FEL pulse triggers an ultrafast spin current into the nonferromagnetic layers where it is converted into a charge current by the inverse spin Hall effect. The amplitude of the emitted THz pulse [23,28] is expected to be a direct measure of the incident x-ray pulse energy for a constant pulse duration and shape. To put this idea to test, we took advantage of the sizeable pulse energy fluctuations of the FERMI FEL. Figure 3(a) correlates the amplitude of the emitted THz pulses with the XUV pulse energy measured by the FERMI XUV pulse energy monitor, which is based on photoionization of a rare gas at low particle density (similar to [19]). We observe a clear linear relationship between the two quantities, which corroborates our approach.

 

Fig. 3. THz amplitude and XUV pulse energy. (a) Correlation of the XUV pulse energy and THz amplitude over roughly 2500 pulses. R-squared of 0.784 is a statistical measure of the linear fit in gray. Correlation widths $\sigma$ are fitted to be 1.7 µJ and 1.9 µJ, respectively, indicating a strong linear correlation between XUV pulse energy and scaled THz amplitude. (b) XUV pulse energy (blue) and THz amplitude (red) as a function of pulse number (top). Plotted difference between the reading from the gas monitor and the observed THz amplitude.

Download Full Size | PDF

This conclusion is further confirmed by Fig. 3(b): the XUV pulse energy and the THz amplitude are compared for several thousand XUV pulses and follow the same evolution. The observed correlation between the THz amplitudes and measured XUV pulse energies yields a (relative) standard deviation of $\pm 2\,\,\unicode{x00B5}\rm J$. We thereby demonstrate that STEs can measure pulse energies as low as a few µJ, and it has been shown that STEs can work in the linear regime up to few mJ pulse energies when operated below a fluence of a few ${\rm mJ}/{{\rm cm}^2}$ [29]. In comparison to the FERMI XUV pulse energy monitor, STEs provide a similar pulse energy range over which they can be operated although with a larger error bar. Furthermore, while a calibration would be needed to provide absolute energies, STE-based energy monitors would provide a direct measurement of the pulse energy at the sample, as opposite to ionization monitors, which are typically separated from the experimental chambers by several optical components. Therefore, our scheme also emerges as a robust and simple alternative technology to detect pulse energy fluctuations directly in the experimental end-station.

We note that arrival time measurements were in principle possible down to pulse energies of 5 µJ, which corresponds to about 10% of the typical pulse energies available from FERMI in this wavelength range [see Fig. 3(a)]. We observe a scaling of the timing jitter from below 10 fs for pulse energies larger than 25 µJ to below 30 fs for pulse energies below 10 µJ. This increase may stem from the smaller THz signal levels, which may lead to a decreased precision of our technique (see Supplement 1).

C. Pulse Duration Monitoring

Figure 4 shows a typical THz waveform and the corresponding spectrum as detected for pump pulses with 30 fs duration and a wavelength of 20 nm. The shape of the generated THz pulse is determined by the intensity profile of the pump pulse, not the actual electric field [30] (Supplement 1, Fig. S5), which can hence be determined by monitoring the THz waveform and spectrum.

 

Fig. 4. THz waveform generated by 30 fs pulses at a wavelength of 20 nm: (a) time-domain THz wave form and (b) corresponding THz spectrum as derived from the experiment (blue), simulation neglecting the transfer functions of the sapphire substrate and the 2 mm ZnTe crystal (black), and simulation including the transfer function of the 2 mm ZnTe crystal (green).

Download Full Size | PDF

 

Fig. 5. Calculated THz waveforms for excitation by two subsequent 30 fs pulses with various pulse–pulse delays at a wavelength of 20 nm: (a) time-domain waveforms and (b) THz amplitude spectra assuming different separation of two XUV pulses.

Download Full Size | PDF

One can simulate the experimentally derived THz electric field [black curve in Fig. 4(a)] by convoluting the emitted THz electric field directly behind the metal stack for a 10 fs pump pulse, which was determined in a previous experiment, with the respective x-ray pulse envelope [28]. Note that the signal waveform measured by electro-optic sampling is affected by the frequency-dependent transfer function of the setup and the electro-optic sampling process. Once these influences are accounted for, good agreement between experiment and simulation is achieved (see Fig. 4 and Supplement 1).

This example demonstrates that the detection of the THz pulse is strongly limited by the detection system used in this experiment and that the true spectrum of the emitted THz pulse is much broader. Our simulations indicate that by using a sufficiently broadband detection setup, an XUV pulse duration of below 50 fs (see Supplement 1, Fig. S5) can be monitored by detecting the THz waveform by the STE.

For experiments exploiting the high intensity of current FELs (e.g., nonlinear studies), the structure of each individual x-ray/XUV pulse becomes significant. At the XUV FEL FLASH, it has been observed by the THz streaking technique that roughly 10% of the pulses consist of two or more distinct peaks [31]. As shown in Fig. 5, our technique should be well capable to observe such fine structure down to the level of a few 10 fs.

4. DISCUSSION

We have demonstrated experimentally that the arrival time of XUV FEL pulses can be clocked by single-shot electro-optic sampling of THz pulses generated by optical rectification in an STE, and that the relative energy of each XUV pulse can be monitored by the amplitude of the corresponding emitted THz pulse. We also showed that the emitted THz waveform is indicative of the XUV pulse duration and even its fine structure. Our scheme, therefore, provides access to many important properties of XUV/x-ray FEL pulses at once. The encoding of the XUV pulse properties into THz pulses opens up the possibility to monitor arrival time, pulse duration, and pulse energy of individual x-ray pulses at MHz repetition rates in a cost-efficient and robust way by utilizing established single-shot electro-optic detection techniques, which are already in routine operation at large-scale photon science facilities [27,32].

The precision of the arrival time measurement depends on the XUV pulse energy and on the specific design of the spectral decoding scheme (see Supplement 1). In our experiments, we demonstrate that a value of below 10 fs can be achieved with our prototype setup for XUV pulse energies above 25 µJ. This value compares well with operational arrival time monitors, e.g., at the European XFEL [33,34]. We currently investigate, whether there is a fundamental limit for the precision given by the THz generation process in the STE. Further improvement of the arrival time precision can also be expected from increasing the detection bandwidth and signal-to-noise ratio.

The potential for the determination of relative pulse energies needs to be further explored. The current implementation of our scheme will not rival the precision of the successfully routinely employed gas monitor detectors [19] but can provide convenient access to the x-ray pulse energies directly in or behind the experiment.

The capabilities of our scheme to determine pulse duration and fine structures within the pulses are currently limited by our choice of the electro-optic crystal to about 200 fs. Note that for broadband THz pulses, the electro-optic response of the detection crystal will have spectral dips with lower sensitivity, which might partially blind the detection system. Currently, detection crystals with different absorption bands are known [35,36]. Depending on the specific demand and spectral range of interest, the most suitable crystal should be chosen. Also, the detection bandwidth can be increased by improving the temporal resolution of the single-shot electro-optic detection setup. In this context, alternative implementations of electro-optic single-shot detection schemes or appropriate numerical retrieval techniques that provide a better temporal resolution should be considered (see, e.g., [37–40]). While more elaborate concepts such as THz streaking [13,14,41] still achieve superior precision and accuracy, our demonstrated methodology has the advantage of being considerably less complex with large potential for future improvements.

Fundamentally, the time resolution of our approach with respect to pulse duration is given by the response time of our STE, which is predominantly determined by the electron-spin relaxation time of the ferromagnetic layer and amounts to approximately 100 fs [28]. By a proper deconvolution procedure [42] and sufficient signal-to-noise ratio, we expect that the time resolution can be reduced to several 10 fs.

This study is also interesting from a fundamental scientific viewpoint. The applicability of our scheme has been verified at wavelengths of 20 and 29 nm. We did not observe noticeable differences in the efficiency of the spintronic THz emission process between the two wavelengths. According to our current understanding of the working principle of the STE, it is only relevant that the pump pulse deposits energy as fast as possible in the electronic system of the ferromagnet, regardless of the specific photon energy [28,30,43]. Our experiments indicate that this notion extends to even an order of magnitude larger photon energies than used previously [30,43]. It is consistent with works that did not observe a noticeable impact of the pump-photon energy from XUV to the visible [44] to the THz range [42] on ultrafast demagnetization, which has the same driving force as ultrafast spin transport [28].

We note that the penetration depth at the wavelengths utilized in our experiments is between 10 and 20 nm. Therefore, over 40% of the incident XUV energy is absorbed in the STE. As absorption coefficients generally decrease in the x-ray regime, a drop in deposited energy and, thus, in THz emission, can be expected. However, we believe that by choosing appropriate absorber materials and layer stack designs, e.g., by considering characteristic core-level absorption edges, our scheme should be applicable also in the soft and harder x-ray regime.

If the aforementioned absorption loss in metallic layers is endurable, STEs grown on SiN or SiC membranes can be envisioned as an online monitor in the direct beamline in front of the experiment if the heat management in the thin substrate is accomplished. Apart from that, STEs on solid substrates, which typically absorb all of the XUV energy, may find their application behind certain gas phase experiments that leave the XUV pulse properties largely unperturbed. Another option is to combine STEs on solid substrates with online XUV grating spectrometers such as the variable line spacing (VLS) spectrometer in operation at the FLASH FEL [45] and perform the XUV pulse characterization on a higher diffraction order.

In conclusion, our study establishes spintronic THz sources as ultrafast versatile photodiodes operative from the infrared to the XUV regime with large potential for future improvements in terms of arrival time and pump-pulse energy resolution.