Microwave soft x-ray microscopy for nanoscale magnetization dynamics in the 5–10 GHz frequency range Free

Modern science requires instrumentation that allows for investigating ever smaller spatial dimensions and increasingly fast time scales. In magnetism research, demand for such techniques has increased in the last decade due to the prediction and the realization of current induced magnetization dynamics, achieved either via spin transfer torque or spin-orbit effects such as the spin-Hall effect. However, while it is nowadays possible to map magnetism at the nanoscale and to measure magnetization dynamics at fast time scales, it is still challenging to combine high spatial and temporal resolution in the same measurement.

In the last decade, a few groups around the world have started to bridge this gap using x-rays generated at synchrotron light sources. Such x-rays are in principle capable of looking at magnetism both at the nanoscale, given their short wavelength, and at fast time scales, due to their typical temporal duration of 50-100 ps. Some work has been done on measuring ferromagnetic resonance (FMR) in extended samples,^1–10 but measurements combining the time-resolved probing of fast magnetization dynamics with nanoscale x-ray microscopy are challenging and still very rare.^11–15 Such studies have also been restricted to dynamics at frequencies lower than 3 GHz. This is an important constraint, which greatly limits the type of samples that one can study. Indeed, many of the materials and of the phenomena of interest in modern magnetism show dynamics in the 5–10 GHz range.

In this work, we describe a scanning transmission soft x-ray microscope that has been optimized for studying magnetization dynamics both in the 5–10 GHz range and with high spatial resolution. In order to achieve this, we designed microwave synchronization electronics that can be synchronized with the RF frequency of the Stanford Synchrotron Radiation Lightsource (SSRL). Our measurement technique relies on a quasi-stroboscopic detection scheme, which allows us to compensate for typical drifts associated with x-ray absorption measurements at a synchrotron. In addition, the mechanical components of the microscope and the x-ray illumination optics have been optimized for long-term stability. This allows for acquiring images with long integration times, which in turn enables the detection of extremely small transient magnetization changes. We demonstrate the capabilities of our instrument by recording the magnetization dynamics present in the spin waves emitted by a nanocontact spin torque oscillator at 6.3 GHz and the ferromagnetic resonance of ∼0.1° amplitude in a micron-sized lithographic element at 9.1 GHz.

The measurement setup comprises three parts. The first part is the scanning transmission x-ray microscopy (STXM) instrument at beamline 13-1 at the SSRL, which we illustrate in Subsection II A. The second part is a single-photon counting detector synchronized with the RF frequency of SSRL. This detector has been previously developed in our group,¹⁶ and in Subsection II B, we briefly describe its functionality and the modifications that we introduced. The third part, discussed in Subsection II C, is a microwave synchronization board that is able to produce a single frequency microwave signal phase-locked to the synchrotron frequency. This microwave signal is then applied to the sample to drive dynamics that are in turn phase-locked to the synchrotron and to the detector. Thus, our instrument resembles a setup for the “pump-probe” type of experiments, where the microwave signal acts as the “pump” and the x-rays as the “probe.”

The STXM instrument consists of a zone plate (ZP) with 30 nm outer zone-width and an order sorting aperture (OSA). For our OSA, we used a standard transmission electron microscopy PtIr(95:5) strip aperture (Hitachi 67137), with 30, 50, 70, and 200 μm apertures, while for most experiments, the 50 μm aperture is used. STXM instruments are common nowadays at synchrotron sources and commercially available. Rather than modifying a commercial microscope to accommodate the special requirements necessary for studies of magnetization dynamics, we developed a dedicated instrument at SSRL. Commercial microscopes rely on interferometer control of the sample position¹⁷ to achieve sub 10 nm resolution and stability. However, the use of an interferometer in an UHV environment and in combination with an electromagnet in close proximity to the sample is extremely challenging. For this reason, we decided to achieve the required stability by minimizing the size of the setup and keeping the mechanical path lengths between the different optical components short. The complete setup as mounted on an in-vacuum breadboard is shown in Figure 1.

Both the total length and height of the setup are approximately 10 cm and the width is 2.5 cm. The zone plate is moved along the beam direction using piezo-stepper stages with resistive encoders (ATTOCUBE ANPx101/RES), and the order sorting aperture can be moved in the plane perpendicular to the beam using similar stages (Attocube ANPx101/RES and ANPz101/RES). Each of these stages allows for 5 mm of motion, which is sufficient for alignment. The detector is mounted in a copper shield that provides electric shielding and temperature stabilization for the avalanche photodiode. The detector is also mounted on piezo-stepper motors like the OSA and ZP. The samples are typically mounted on small boards (15 mm × 40 mm) with built-in wave guides and SMA connectors. Since the SMA cabling is capable of transmitting electric RF signals up to 18 GHz and is relatively stiff, we use robust piezo-motor actuated roller bearing stages (PPX-20 and PPX-32, MICOS) to control the sample position. These also provide a larger range of motion (18 mm horizontal and 12 mm vertical), which is convenient for alignment purposes and allows us to potentially mount more than one sample at once.

The sample can be placed into different magnetic fields, using either permanent magnets or electromagnets. Permanent magnets are used to apply up to 0.8 T perpendicular to the sample surface (assuming that the x-rays are incident normal to the sample). A water cooled electromagnet with a gap of 9.0 mm can supply magnetic fields in plane up to 0.25 T. The sample can be rotated around its vertical axis to change the angle between incoming x-rays and sample magnetization, allowing the experimenter to detect different components of the magnetization. The entire microscope, which is housed in a stainless steel vacuum chamber, can be aligned with respect to the x-ray beam using six different struts (not shown). All microscope components are vacuum compatible and an ion pump allows base pressures down to 2 × 10⁻⁸ mbars during imaging experiments. Operating the microscope in vacuum is advantageous since it avoids contamination of the surface with hydrocarbons while imaging, especially for experiments requiring long averaging times. Even after several days of exposure of the sample, we have never observed a decrease in transmission of more than 50%, typically reached within a few minutes if operated in atmosphere.

The “probe” part of the instrument is a detection system able to look at the individual x-ray pulses generated by all the electron buckets present in the SSRL storage ring. The operating frequency of the SSRL storage ring f_s ≈ 476 MHz, meaning that the electron buckets (and in turn the x-ray pulses) are temporally separated by 2.1 ns. Only recently, researchers have been able to develop special photon-counting electronics necessary to detect x-ray pulses so close in time.¹⁶ Such electronics look at the signal coming from a fast avalanche photodiode (Hamamatsu S12426) and perform three operations, as we detail in the following.

First, it digitizes the APD signal using a discriminator synchronized with the synchrotron frequency f_s. This is necessary because the number of x-ray photons reaching the APD is low: at the highest intensities through a typical sample, we estimate that on average 1 of every 4 electron buckets results in one photon being transmitted through the sample.

Second, it sorts all the photons detected by the discriminator by using up to 16 different counters, which can be then accessed independently. This is achieved using field-programmable gate-array (FPGA), as described in detail in Ref. 16.

Last, the electronics can be synchronized with the period of the orbit of the storage ring f_s/N_buckets = 1.28 MHz, since N_buckets = 372 at SSRL. Using this feature, half of the counters can be used to measure the x-ray signal during odd orbits and the other half during even orbits. This allows for designing a very effective normalization scheme based on a modulation technique, where the “pump” is applied to the sample only during odd revolutions of the ring and even revolutions are reserved for the reference signal. Such capability is particularly useful in a synchrotron light source such as SSRL, where the current of the storage ring is kept constant via top-up injections of electrons. Each injection lasts for about 10 s and is repeated every 5 min. The injections appreciably offset the charge of some buckets, which in turn results in an offset of the x-ray intensity of the order of 1%. This has the potential of completely masking real signals which, as we will show below, can show downwards of 0.1% variation with respect to the background. However, by collecting the signal and the reference from an electron bucket that is almost identical after one revolution period (it takes 1/1.28 MHz⁻¹ ≈ 780 ns for one electron buckets to orbit around the ring), these offsets normalize out in an efficient way, and this allows us to reach a noise level of 10⁻⁴, enough to clearly detect the signal.

The goal of the microwave synchronization board is to generate a clean microwave signal at a single frequency, phase locked to the synchrotron frequency. The microwave signal can be used to drive excitations in the sample, and these excitations will thus be phase locked to the synchrotron frequency.

The peculiarity of our microwave synchronization board is that it is phase-locked not to an exact harmonic n of the synchrotron frequency, but to a frequency that is offset from it by a subharmonic f_S/m, where m is an integer. This is an approach similar to the one followed in Ref. 11. When this situation is realized, the measurement is no longer a pure stroboscopic type of detection, because two consecutive x-ray photons will not measure the same phase of the excitation. However, the phase observed by the photons is not random: every mth photon will always see the same phase. Our measurement is thus “quasi”-stroboscopic in the sense that a measurement does not record only one specific phase of the excitation, but a finite subset of phases in each pixel before moving to the next pixel. By collecting all the phases almost simultaneously, we completely suppress even the slightest long-term drifts, inevitably left in our scanning stages without interferometric control.

Fig. 2 shows the schematic of the microwave synchronization scheme. The synchrotron clock frequency f_s ≈ 476 MHz is split with a 3 dB power divider into two parallel transmission lines. Along the first one, a frequency comb generator is inserted in order to obtain simultaneous generation of n harmonics of f_s, with 2 ≤ n ≤ 24. In the second transmission line, a frequency divider is mounted in order to produce a single f_s/m, where m can be programmed to have a value between 2 and 17.

The two transmission lines carrying the n harmonics and one subharmonic of f_s are recombined using the frequency mixer M1 to produce modulation sidebands located at the desired f_pump = f_s(n ± 1/m). This signal is already in phase with f_s, but it also contains 23 other harmonics of f_s, some of which may act as a further pump frequency for the sample and disturb the measurement. In order to excite the sample with a clean signal at the single frequency f_pump, we introduce a low phase-noise microwave signal generator (Anritsu MG3692B) into the circuit.

The frequency f_mw of the microwave signal generator can be tuned to be close to f_pump with a beating of a few Hz. The low frequency beating signal is now a measure of the phase difference between the two signals f_pump and f_mw. Connecting the output of a PID controller (Stanford Research Systems SIM960) to the electronic frequency control (EFC) port of the microwave signal generator,¹⁸ allowed us to realize a phase locked loop (PLL), that minimizes the phase difference between the two signals. When the PLL is closed, f_mw = f_pump, and a single frequency signal locked to the synchrotron frequency f_s is provided to the sample.

The successful implementation of the synchronization scheme can be checked with a sampling oscilloscope (Tektronix TDS8200 with 80E04 sampling module) where the trigger port is connected to the synchrotron reference signal at f_s = 476 MHz and the sampling port to the output of the microwave synchronization board. For this measurement, we used n = 21 and m = 6, to get f_mw = 10.07 GHz. We chose this specific frequency to prove that our microwave synchronization board works reliably up to the higher end of the 5-10 GHz frequency range. Taking a series of single acquisition shots of the scope trace produces consecutive output traces similar to the one plotted in Fig. 3. The six phases of the signal at a frequency very close to 10 GHz are clearly distinguishable on the screen of the sampling scope, and the fact that consecutive single shots do not drift on the screen proves that there exists a well-defined phase relation between the synchrotron frequency and the output of our board.

In the actual experiment, the microwave signal is applied to the sample only during odd orbits of the storage ring and is turned off during even orbits, as mentioned in Section II A. This is achieved using a fast pin diode, a device able to switch on and off the transmission of a microwave signal within ∼10 ns, much faster than the storage ring orbit period of 780 ns.

In this section, we quantify the robustness of the synchronization scheme, both in terms of jitter and drift.

In order to quantify the in-loop jitter J, we use the output of mixer M2, which contains the beating between the signals from the microwave source and from the comb generator. This beating is a direct measure of the phase difference between the two signals, and one only needs to normalize it in the proper units to retrieve the in-loop timing jitter. This is done by measuring the amplitude of the beating with the PLL both open and closed. Then, one can use the relation

where V_closed is the rms voltage measured for the closed loop and V_open measures the amplitude of the beating oscillation in open loop. The expression is valid when the phase fluctuations are small compared to the oscillation period, i.e., sinϕ(t) ≈ δϕ(t) = J 2πf. In Table I, we list the calculated jitter for the different harmonics n and for a single value m = 6 of the subharmonics. It is clear that for all harmonics, the jitter is on the order of 1 ps and below 500 fs for frequencies up to 10 GHz.

We now turn to the estimation of the drift of the microwave synchronization board. In order to do this, we used a second board consisting of three parts: a frequency multiplier that generates the 16th harmonics of f_s at 7.62 GHz, a frequency divider that produces f_s/7 = 68 MHz, and a mixer that can be used to produce a modulation signal at 7.69 GHz. This modulation signal can then be down-mixed with the nominally identical signal generated by our microwave synchronization board for n = 16 and m = 7. Similarly to the jitter measurements, the down-mixing voltage is a measure of the phase difference between the two signals, and can be used to estimate the drift of the board.

This method actually only allows us to estimate the upper limit of the drift, because the contribution from the two boards cannot be disentangled. For simplicity, we assumed that our board was responsible for all the drift observed in the measurement. We also note that we did not implement any temperature stabilization, since such stabilization may be unpractical in a real experiment. Fig. 4 plots the drift over a 1 h measurement, showing the timing drift is bounded between ±2 ps over the hour-long time span.

Both timing jitter and drift values are much smaller than the temporal FWHM of the x-ray photons, which in the low-emittance operation mode is about 50 ps. Hence, we conclude that our microwave synchronization board does not cause any significant degradation of the temporal resolution available at the synchrotron.

In order to demonstrate the capabilities and the versatility of our setup for measurements at high-frequency and with high-spatial resolution, we show the results from two types of experiments. First, we map the large amplitude spin waves excited by a spin polarized current in a 5 nm think NiFe film. The frequency of the spin wave was about 6 GHz and the size of the excitation on the order of 100 nm. Then, we measure small angle (∼0.1°) magnetization precession due to ferromagnetic resonance in a Co microstrip (20 nm thick), driven by a microwave magnetic field at a frequency of approximately 9 GHz.

For both cases, a schematic of the experimental geometry is shown in Fig. 5 in a schematic way. A static magnetic field H_dc is applied along the plane of the film. This defines the precessional axis of the magnetization (parallel to the applied field), which results, during precession, in a finite component of the magnetization parallel to the direction of incoming x-rays direction (M_z in our notation). This is a necessary condition to detect an x-ray magnetic circular dichroism (XMCD)¹⁹ signal, which requires that the spins are parallel or anti-parallel to the helicity of the x-ray photons. The helicity of the x-rays is controlled by the EPU that delivers x-rays to beamline 13-1 at SSRL, as described in Section II A. The undulator in combination with a monochromator also allows us to finely tune the energy of the x-rays to the desired value.

The time-resolved mapping of spin waves excited in a spin torque oscillator has so far remained an elusive goal for the magnetism community. In such a device, a high-density, spin-polarized direct current is injected via a nanometer-size contact into an extended ferromagnetic film. Due to the spin transfer torque effect,^20,21 spin waves are excited in the region surrounding the nanocontact. The spin waves can also be efficiently phase-locked to an external microwave signal superimposed to the driving direct current.²² Both these aspects (the nanoscale characteristic of the excitation and the possibility of phase-locking with an external signal) make this sample the ideal testing ground for our setup. Our sample fabrication process is detailed in Ref. 23 and in our upcoming publication.²⁴ Here, we only note that the sample is grown on a SiN membrane transparent to x-rays. Such a substrate allows for detection of the x-rays transmitted through the sample via an avalanche photodiode, as required by our detection scheme. Since the spin waves are excited in a thin permalloy (Ni₈₀Fe₂₀) film, we tuned the x-ray energy to the Ni L₃ absorption edge (E = 852.7 eV) to maximize the XMCD contrast.

The spin wave dynamics are first excited via a direct current provided to the sample with a precision current source (Keithley Source Meter 2400), and the frequency of the excitation is observed in real-time using a spectrum analyzer (Rhode&Schwarz FSU13). By varying the current magnitude, one can accurately control the frequency of the spin wave and bring it close to one of the frequencies generated by our microwave synchronization board. For the data presented below, we found that a direct current I_DC = 8 mA excites spin waves at a frequency of 6.27 GHz, very close to the n = 13, m = 6 line of the microwave board.²⁵ 

The signal from the microwave board is then superimposed on the sample with the direct current to induce synchronization between the spin wave excitation and the synchrotron. When the synchronization condition is satisfied, successive circularly polarized x-ray photons will probe six subsequent projection M_z of the magnetization vector M along the x-ray propagation direction. This realizes a phase-resolved detection of the magnetization dynamics which can be used to build a real-space, dynamical map of the spin waves excited at the nanoscale, i.e., a spin wave “movie.”

The results are presented in Fig. 6. In each row, the images corresponding to the six different phases are plotted in a sequence, and the corresponding time delay relative to the first image is indicated above each image in the first row. The contrast arises because of varying x-ray transmission through the sample, with the brighter (darker) colors indicating regions of higher (lower) transmission. Each image is 0.6 μm × 0.6 μm in size and the step size for the scanning sample stage was 30 nm, close to the nominal resolution provided by our zone plates (approximately 36 nm). This resolution can be estimated using the Rayleigh criterion 1.22Δr, where Δr, the outer zone-width, is 30 nm in our case.

The first row shows the data recorded with positive helicity x-rays and with both direct current (I_DC) and microwave signal (I_mw) applied to the sample. On the top-right corner, a white contrast is observed in the first image at t = 0 ps. The contrast becomes larger in the second image at t = 27 ps, weakens in the third image, and turns black in the fourth. The black contrast undergoes a similar evolution in the fourth, fifth, and sixth images (t = 81, 108, and 135 ps, respectively) as the white contrast in the first three images. The contrast is due to magnetic dichroism; hence, the opposite contrast signifies magnetization in the sample pointing in the opposite direction. The topographic features of the sample, such as the nanocontact where the current is injected, are normalized out by the even vs odd orbit normalization scheme described in Section II A. The maximum amplitude of the signal corresponds to a variation with respect to the background signal on the order of 10⁻³, with the noise fluctuations on the order of 10⁻⁴ in each of the recorded phases.

In order to prove that we are observing a magnetic excitation, we reverse the helicity of the x-rays. In the second row of Fig. 6, we plot the images recorded when the same I_DC and I_mw were applied to the sample, but now with the x-ray helicity reversed. It is evident that the contrast has changed sign, proving that the signal is magnetic. The shift of the contrast towards the middle of the image compared to the first row is due to drifting of the microscope stages in the time between the two consecutive measurements, on the order of a couple of hours. This observation further motivates our measurement scheme, which is insensitive to such drift.

Finally, in the last row, we present the images taken when no microwave signal was applied to the sample. The spin wave was still observed in the spectrum analyzer, but it was not phase locked to the synchrotron. In absence of phase-locking, the average out-of-plane component of the magnetization is zero, and uniform gray images are expected and observed. Similarly uniform images are recorded if only the microwave signal is applied to the sample, without any I_DC sustaining the spin wave excitation.²⁴ 

FMR in multilayers composed of different materials can result in complex dynamics that cannot be easily discerned only by measuring magnetic fields. X-ray FMR (X-FMR) solves this problem by exploiting the elemental sensitivity of the XMCD effect. In this specific measurement, we aimed at addressing the upper frequency limit of our synchronization scheme, in combination with 50 ps FWHM x-ray pulses. The sample is a 20 nm thick Co microstrip (2 μm × 0.5 μm), that is deposited in the center using optical and electron beam lithography tools of an omega-shaped microresonator of several tens of micrometer diameter,²⁶ as shown in the top panel of Fig. 7. The lithography was performed also in this case on a SiN membrane that allows for x-rays to be transmitted through it.

An external magnetic field of 135 mT was applied perpendicular to the easy axis of the sample (horizontal, parallel to the short side of the strip in Fig. 7), while the sample was exposed to a RF field at 9.129 GHz. The RF field frequency and applied magnetic field were optimized after conventional ferromagnetic resonance measurements were acquired ex situ. In analogy with the spin wave measurements, the STXM images in Fig. 7 show the XMCD contrast at the Co L₃ edge (E = 778.1 eV) arising because of the time-varying magnetization component perpendicular to the plane of the image, as depicted schematically in Fig. 5. Between two consecutive images, there is a phase difference of 60°, corresponding to a time difference of 18 ps. Images acquired 180° apart show inverted contrast as expected. By acquiring two images, one with microwave and one without microwave excitation, we are able to directly measure only the change in the absorption cross section caused by the magnetization dynamics, while removing the background fluctuations common to both images.

The data show that the cross section Δμt changes by ±2 × 10⁻³. Using reference values for the XMCD, we can then estimate the out-of-plane precession angle of the magnetization to be about 0.1°, without considering the reduction of the observed contrast due to the x-ray pulse length comparable with the investigated time scale. The convolution of a 50 ps FWHM Gaussian with a sinusoidal curve at 9 GHz frequency results in a sinusoidal curve with approximately 50% reduced amplitude. Hence, we can conclude that in the studied sample, we are able to detect a precession angle on the order of 0.1°.

We have built a microwave synchronization electronics setup that, combined with a synchrotron based scanning x-ray microscope, allows for nanoscale and time-resolved measurements at frequencies of up to 10 GHz, beyond the current state-of-the-art instruments. The jitter (≈500 fs) and the drift (≈2 ps over 1 h measurement) of the board are much smaller than the temporal FWHM of the x-ray pulses (≈50 ps), and therefore, it is this FWHM which ultimately determines the temporal resolution of the setup. We have used a quasi-stroboscopic synchronization scheme to record multiple phases of the magnetization precession in parallel. This has allowed us to virtually eliminate the drifts which are otherwise intrinsically present in a scanning microscope.

The illuminating x-ray optics and the in-house designed x-ray microscope are optimized for long-term stability and minimal contamination due to its operation in ultra-high vacuum. The in-house design also allows for a flexible sample environment, which allows for including magnetic fields and electric excitations of various kinds and which is critical for state of the art experiments addressing dynamical properties at the nanoscale.

A single photon detection system combined with our microwave synchronization board allowed us to record micrometer-sized images with high spatial resolution within an acquisition time of a few hours. In this time frame, we could reliably measure dynamical signals which vary only 0.1% over the background. When six different phases were recorded, the typical noise level was of the order of 0.01%. This has allowed us to create a nanoscale dynamical map of the spin waves excited in a nanocontact spin torque oscillator, as well as of the ferromagnetic resonance in a patterned Co strip.

Our electronic design can be readily extended to frequencies up to 20 GHz and theoretically to the ∼100 GHz range given the availability of microwave components compatible with those frequencies. This is of potential interest in storage rings operating in low-alpha mode,²⁷ which can produce 1-10 ps long x-ray pulses, as well as in free electron lasers, where sub-ps pulses are available.

We are very grateful to Sergei Urazhdin at Emory University for fabricating the samples for the spin wave measurements. This work is supported by the Department of Energy, Office of Science, Basic Energy Sciences, Materials Sciences and Engineering Division, under Contract No. DE-AC02-76SF00515. Use of the Stanford Synchrotron Radiation Lightsource, SLAC National Accelerator Laboratory, is supported by the U.S. Department of Energy, Office of Science, Office of Basic Energy Sciences under Contract No. DE-AC02-76SF00515. Stefano Bonetti gratefully acknowledges support from the Knut and Alice Wallenberg Foundation.

The scanning transmission x-ray microscope uses the elliptically polarizing undulator (EPU) at beamline 13 at SSRL as an x-ray source. High energy electrons are stored in a closed orbit at the synchrotron. For this purpose, the total stored charge is split and compressed in so-called buckets. The particular layout of SSRL is compatible with 372 buckets filled with about 1 pC of charge at a kinetic energy of 3 GeV. Each electron bucket is 50 ps long (FWHM) and subsequent buckets are separated by 2.1 ns (476.2 MHz). As an electron bucket traverses the EPU, it is forced by variable and tunable magnetic fields to follow an undulating trajectory either in the vertical plane, horizontal plane, or a combination of both, which causes the generation of linear or circular polarized x-rays.^28,29 The beamline 13 EPU²⁸ produces polarized x-rays in the energy range between approximately 200 eV to 1200 eV. This energy range — commonly referred to as soft x-ray region — has been shown to be very well suited to study magnetic properties of complex magnetic nanostructures based on 3d and 4f transition metals.¹⁹ 

The diverging x-ray beam generated in the EPU is guided to the microscope using appropriate x-ray mirrors as shown schematically in Figure 8. Note that the typical angle of incidence for soft x-ray optics is a few degrees. To obtain monochromatic x-rays at the sample a spherical grating monochromator (SGM) is used in this case. First, the x-ray beam is focused on the entrance slit, about 15 μm in size, using a spherical mirror. The spherical grating produces a 2:1 magnified image of the entrance slit onto the exit slit, which is located in the energy dispersive plane so that rocking of the SGM changes the energy passing through the exit slit. The typical energy resolution is E/ΔE = 7000. In the horizontal plane, the beam is deflected first at a flat horizontal mirror that also serves as a power filter, removing higher energy x-rays from the downstream spectrum and thus reducing thermal load on the remaining optical components. Since the source is shared between different experiments at beamline 13, another horizontal mirror is used to deflect the beam towards the experiment. This mirror produces a de-magnified image of the source onto the horizontal exit slit, typically about 50 − 100 μm in size. Each slit of the horizontal exit slit assembly is electrically isolated and the difference in electron yield current is used to feedback onto the angle of the horizontally focusing mirror, so that the horizontal beam position can be kept constant independent of beam movements of the source. Once the beam passes through the exit slit, which is the common focus in the horizontal and vertical planes, it diverges again and finally illuminates the zone plate used to focus the x-rays in the STXM instrument at 26 meters from the source. The beam size at this point is 2.0 mm × 1.0 mm, while the zone plate (XRADIA, 160 μm diameter, 80 μm inner stop, 30 nm outermost zone width) is much smaller. Due to the significant overfilling of the zone plate, the setup becomes less sensitive to small fluctuations in beam position, while still preserving an acceptable photon flux onto the sample. The transmission of the zone plate is optimized for a photon energy of 800 eV, which is close to the absorption resonances of the ferromagnetic 3d transition metals Fe (706.8 eV), Co (778.1 eV), and Ni (852.7 eV). Using a calibrated x-ray photodiode (AXUV-100G, Opto Diode), we measured a photon flux through the zone plate of 2 × 10⁹ photons/s at 800 eV under normal operating conditions.