Picosecond Fresnel transmission electron microscopy

Control of magnetization on combined nanometer spatial and picosecond temporal scales is currently being investigated for next-generation information storage, communications, and data processing.^1–7 Two main approaches are being explored: an electronic-based method employing spin-transfer torque effects and laser-based methods using photothermal or all-optical switching effects. For methods based on spin-transfer torque, a spin-polarized electric current is used to change the magnetization orientation in magnetically active films. Several applications have emerged using this magnetization switching mechanism over the past decade.^8–10 Comparatively, laser-based approaches are generally thought to operate via either photothermal-assisted or all-optical switching effects.^11–17 For these methods, short laser-pulse durations, combined with near-field excitation, can be used to promote magnetization reversal on nanoscale bit sizes, as implemented in devices based on heat-assisted (i.e., photothermal-assisted) magnetic recording (HAMR).^18,19

Importantly, variations in structure, morphology, and composition—as can occur with repeated laser irradiation (for example) —can deleteriously affect predictable magnetization behavior via pinning of magnetic moments at newly formed nanocrystallites or through larger-scale crystallization, restructuring, and ablation.²⁰ Accordingly, suitable characterization techniques would ideally enable direct correlation of real-time, ultrafast magnetic responses to nanometer-to-micrometer (time-varying) structure and morphology, such that a comprehensive view of the spatiotemporal dynamics could be generated. Frequently employed imaging techniques, such as time-resolved magneto-optic Kerr-effect and X-ray microscopies, can be used to resolve sub-picosecond magnetic phenomena with diffraction-limited spatial resolutions.^21–28 However, it is challenging to directly correlate and resolve changes in film structures with these techniques. Conversely, transmission electron microscopy (TEM) can be used to visualize magnetic domains and domain walls using Foucault and Fresnel imaging, respectively,^29–33 in addition to providing access to (up to) atomic-scale structural, morphological, and chemical information. Indeed, TEM Fresnel imaging in both conventional and Lorentz-enabled microscopes has been extensively used to study the magnetization properties of thin films and nanostructures.^34–42 

Although conventional TEM is limited to studying approximately millisecond dynamics, the advent of stroboscopic, pump/probe ultrafast electron microscopy (UEM) has expanded the accessible temporal parameter space to sub-picosecond regimes.^43–45 Thus, UEM is, in principle, well-suited for elucidation and direct correlation of ultrafast structural and magnetic dynamics. Despite numerous studies focused on structural, morphological, and electronic dynamics,^43,45–47 magnetic dynamics have thus far been limited to a single, nanosecond study of a partially oxidized Fe film,⁴⁸ though more-recent work has exploited static, in situ laser-irradiation capabilities.^20,49 Here, we report picosecond-resolved UEM Fresnel imaging of photoinduced magnetic dynamics. Specifically, the in situ photoinduced demagnetization of an FePt thin film was visualized and quantified by monitoring picosecond domain-wall-contrast dynamics at large defocus values (±25 mm) with an effective 3-ps image shutter. For this particular specimen, the change in domain-wall contrast displays a single-exponential response, with a time constant of 24 ps. Owing to challenges associated with inherently low UEM beam currents, stemming from the need to operate at relatively low repetition rates for complete specimen thermal dissipation and in order to preserve picosecond time resolutions,^50,51 Fresnel-imaging experiments were enabled by optimization of photoelectron-collection efficiencies into the illumination system of the UEM.⁵² 

The specimens studied here consisted of thin FePt films on amorphous Si₃N₄ membranes (TEMWindows.com, 15-nm thick). The films were deposited directly onto the membranes by co-sputtering Fe and Pt at room temperature using an eight-target magnetron sputtering ultra-high vacuum system operated at a base pressure of at most 10⁻⁷Torr. Here, FePt was chosen for study owing to recently observed ultrafast photoinduced demagnetization dynamics and, more generally, because of its favorable properties (e.g., small, thermally and chemically stable magnetic-grain core sizes) for application in HAMR technologies.^18,53,54 The as-deposited films were 35-nm thick, displayed a saturation magnetization of 380 emu/cm³, had a composition of Fe₃₂Pt₆₈, and had an in-plane magnetization direction (see supplementary material). A Pt island (40 μm by 20 μm by 100-nm thick) was deposited onto the FePt film using a focused ion beam and a gas injection of (CH₃)₃Pt(C₅H₅CH₃). The island served as both a heat sink to mitigate photothermal accumulation and as a pinning site for magnetic domain walls.

Both static and picosecond Fresnel imaging were conducted with an FEI Tecnai Femto UEM (Thermo Fisher Scientific) and a diode-pumped, Yb:KGW [fundamental wavelength (λ) = 1,030 nm] femtosecond laser (PHAROS, Light Conversion), the layout of which is shown in Fig. 1. The Wehnelt assembly of the UEM electron gun consisted of a graphite-encircled, 50-μm flat LaB₆ source (for added photoelectron-beam stability; Applied Physics Technologies) and a 2-mm diameter Wehnelt aperture. The LaB₆ source was set 0.25 mm back from the Wehnelt aperture in order to optimize photoelectron coupling into the UEM illumination system.^51,52 For static imaging, the UEM was operated as a conventional thermionic TEM. For picosecond stroboscopic imaging, frequency-quadrupled laser pulses [λ  =  257.5 nm, 700-fs duration full-width at half-maximum (FWHM)] were used to generate discrete (probe) photoelectron packets from the LaB₆ source, while linearly polarized, frequency-doubled (pump) pulses (λ  =  515 nm, 700-fs FWHM) were used to demagnetize the FePt film in situ. A 5-kHz repetition rate was used for all pulsed-photoelectron and stroboscopic pump/probe experiments. The detector was a 4-megapixel, 14-bit, fiber-coupled CCD camera (Gatan Orius SC200B), with a phosphor conversion of four photons per incident electron. All Fresnel imaging was conducted at ±25-mm defocus values and without excitation of the UEM objective lens. In this state, a 10-mT stray field was measured at the specimen position.

Figure 2 displays a direct comparison of Fresnel images obtained in thermionic and stroboscopic UEM modes (i.e., continuous-beam and pulsed-beam modes, respectively). Light and dark regions in the defocused images correspond to specimen areas that generate convergent and divergent electron deflections, respectively, arising from the Lorentz force at magnetic domain walls. Though the signal-to-noise ratios (SNRs) are significantly lower for the stroboscopic UEM images, as compared to the thermionic images, it can be seen that the domain-wall positions and appearance are reproduced [compare Fig. 2(b) to 2(d)]. This demonstrates that, while the spatial resolution is reduced relative to the thermionic mode, the coherency and beam current in the picosecond, stroboscopic mode at 5 kHz is sufficient for imaging magnetic domain-wall contrast. It is important to also note that the particular CCD camera used here is equipped with a rather low-sensitivity phosphor and is air-cooled. As a result, the SNRs seen here are much lower than would be obtained with a more-sensitive detector (for the same beam current and image-acquisition time). This further illustrates the feasibility of the measurements and the importance of efficient coupling of photoelectrons into the UEM illumination system.^51,52

In order to quantitatively compare the domain-wall contrast for the thermionic and pulsed-photoelectron Fresnel images, the degree of electron coherence (i.e., the visibility) and the domain-wall dimensions were determined for each. A measure of the degree of electron coherence can be given as $Imax−IminImax+Imin$⁠, where I_max and I_min are the maximum and minimum intensities next to and within the magnetic domain wall in the image, respectively.⁵⁵ Here, it was found that the degree of electron coherence decreased from 0.15 for the thermionic image to 0.073 for the pulsed-photoelectron image. Further, the width of the domain-wall contrast increased from 0.51 μm to 1.0 μm (FWHM) for the pulsed-photoelectron Fresnel image. Importantly, the domain-wall width measured in the thermionic images matches well with the calculated value (0.5 μm) at the defocus conditions used here, assuming a Néel wall (see supplementary material).

To image the picosecond photoinduced response of magnetic domain walls in the specimen, a series of stroboscopic pump/probe Fresnel images were acquired (Fig. 3). Here, a pump fluence of 1 mJ/cm² was used (spot size = 140 μm FWHM, as measured ex situ with a beam profiler). The pump pulse was centered on the Pt island such that a thermal gradient was generated with the FePt film during photoexcitation. This resulted in a measurable change in the degree of coherence at the domain wall from an initial value of approximately 0.056 to 0.021. The decrease followed a single exponential decay with a time constant of 24 ps. Note that this response is slower than previous demagnetization measurements of FePt films. This could be attributed to the time required for thermal transport to occur from the Pt island to the monitored domain wall, to differences in film composition, or to differences in photothermal temperature rise. Indeed, variations in temporal response have been found to depend upon a number of factors, including film composition, electron-phonon coupling properties, and the specific experimental parameters employed.^53,56–58 It is unlikely, however, that the relatively slow response observed is due to limitations in the instrument-response time. At a 5-kHz repetition rate and with 1.2 counts per pixel per second, each probe packet contains an average of 240 photoelectrons (at the detector). For 700-fs laser-pulse durations, this corresponds to an estimated UEM instrument response time, for the particular system used here, of 3 ps FWHM.⁵¹ 

Owing to the manner in which stroboscopic UEM experiments are conducted, careful consideration must be given to identifying and controlling potential artifacts and errors.⁵⁰ Here, each image series was acquired by randomizing the time-delay points (i.e., by randomly, rather than sequentially, acquiring images). This was done in order to isolate the measured intrinsic picosecond dynamics from systematic errors present during the entire experiment scan, the total duration of which can be several hours. Once the scan is complete, the images are re-ordered into the sequential time delay, thus revealing the presence of any ultrafast dynamics. Specific to UEM Fresnel imaging, an additional control experiment that can be conducted is to change the sign of the defocus value, followed by repeating the image scan. In this case, an intrinsic domain-wall response should show the same behavior for each defocus value, the only difference being a change in the change in relative intensity (i.e., high-to-low vs. low-to-high relative-intensity change). This is because the sign of the defocus value sets the initial domain-wall contrast appearance. The results of these experiments are summarized in Fig. 4.

While the results summarized above demonstrate the feasibility of conducting picosecond Fresnel imaging with UEM, a number of practical developments would further enhance and expand the anticipated application space. For the experimental configuration described here, the highest-impact developments would result in an increase in SNR with a commensurate reduction in image acquisition times. Further, any such developments would also preserve low instrument repetition rates and ultrafast temporal resolutions. That is, resolving challenges associated with rapid specimen photothermal dissipation and photoelectron-packet coherence (spatial and temporal) would be balanced with optimization of magnification, and spatial and temporal resolutions. Accordingly, a viable advancement would be a UEM equipped with a high-brightness, nanoscale photocathode (for optimum spatial coherence) operated at or near the single-electron-per-packet regime (for optimum temporal coherence and resolution).⁵⁹ In such a configuration, however, specimen thermal dissipation would likely be a challenge, so additional incorporation of a highly sensitive detector and a dedicated Lorentz module would enable reductions in acquisition times and instrument repetition rates, with concomitant increases in spatial resolution. Indeed, additional efforts are currently underway to advance TEM capabilities with respect to the study of ultrafast magnetization dynamics.⁶⁰ 

In conclusion, picosecond Fresnel TEM has been demonstrated here on photoinduced magnetic-domain wall dynamics in FePt thin films. Importantly, this extends the UEM temporal range for studying magnetic dynamics to timescales on which intrinsic photoinduced switching occurs. Further, owing to the multi-mode capabilities inherent to TEM (e.g., imaging, diffraction, and spectroscopy), performing such experiments opens the way to direct correlation of structural, electronic, and magnetic dynamics from specific, nanoscale specimen regions of interest. As such, it is anticipated that TEM pump/probe experimental approaches in general—with the pump potentially taking on a variety of forms^28,61—will find application in the study of spintronic device dynamics and fundamental magnetic phenomena.

See supplementary material for thin-film characterization, material-property calculations, and SNR definition.

This work was supported partially by the National Science Foundation through the University of Minnesota MRSEC under Award Number DMR-1420013. This work was partially supported by C-SPIN, one of six STARnet Centers, a Semiconductor Research Corporation program, sponsored by MARCO and DARPA. Part of this work was carried out in the College of Science and Engineering Characterization Facility, University of Minnesota, which has received capital equipment funding from the NSF through the UMN MRSEC program under Award Numbers DMR-0819885 and DMR-1420013. Part of this work was carried out in the College of Science and Engineering Institute for Rock Magnetism, University of Minnesota, which is made possible in part through the Instrumentation and Facilities program of the NSF Earth Science Division. Additional support was provided by the Arnold and Mabel Beckman Foundation through a Beckman Young Investigator Award (D.J.F.).