Rotating spintronic terahertz emitter optimized for microjoule pump-pulse energies and megahertz repetition rates

Spintronic terahertz (THz) emitters (STEs) have emerged as versatile sources of ultra-broadband single-cycle terahertz pulses without spectral gaps,¹ delivering peak electric fields up to ∼1 MV/cm.^2,3 Among other advantages,^4,5 STEs offer easy and versatile polarity and polarization control via external magnetic fields,^6–8 wavelength-independent excitation⁹ without phase matching constraints, excellent beam quality and focusability,² and Fourier-limited single-cycle transients with ultra-wide bandwidth.¹ Therefore, they are of particular interest for terahertz field-driven applications such as terahertz-light wave scanning tunneling microscopy (THz-STM)^10–12 and field-resolved terahertz scanning near-field optical microscopy (THz-SNOM).^10,13 A crucial aspect for these applications is the generation of measurable terahertz-field-induced currents or scattered terahertz near-field signals, which requires operation at megahertz repetition rates and sufficiently high terahertz field strength, e.g., up to few kilovolts per cm in the case of THz-STM.^10,14,15 Modern Yb-based high-power femtosecond laser systems^16–18 combined with external pulse compressors^19–22 provide pulses with a duration down to a few 10 fs at several 10–100 W of output power and megahertz repetition rates, motivating the optimization of broadband terahertz sources such as the STE at such laser parameters.^23–27 This will further pave the way for future technological applications in broadband terahertz imaging,^28,29 sensing, and spectroscopy that will greatly benefit from high terahertz powers.

To exploit the full potential of STEs for terahertz-field-driven applications, the conversion of optical energy into emitted terahertz field (optical-to-terahertz-field conversion efficiency, FCE) as well as the propagation and focusing of the broadband terahertz pulses in the experiment must be optimized. Optimal FCE of the STE is typically achieved by exciting the STE at a pump fluence of ∼1 mJ/cm².^2,23 However, at pump powers of several 10 W and megahertz repetition rates, accumulated heating over many laser pulses can significantly reduce the FCE and even lead to irreversible degradation or immediate damage of the STE.²³ Excitation of the STE with pump spot sizes of several centimeters, as shown for multi-Watt operation at kilohertz repetition rates,^2,3 minimizes heating and results in an optimal FCE at millijoule pump-pulse energies. However, at megahertz repetition rates and few 10 μJ pulse energies, this approach will result in very low fluence and correspondingly reduced FCE. On the other hand, if the fluence is increased by decreasing the pump spot size, the elevated power density and resulting significant heating will limit STE operation to low pump powers. Thus, achieving a fluence of 1 mJ/cm² at megahertz repetition rates and multi-Watt average powers is a major challenge and requires efficient heat management. Active backside cooling of the STE was demonstrated recently,²³ but it requires operation in a backreflection geometry that is less straightforward than a transmission geometry. In addition, the maximum usable power will be limited by heat transport within the cooled STE device, and scaling to higher powers will require complex cryogenic environments.

Here, we demonstrate efficient high-power operation of a trilayer STE excited at megahertz repetition rates with up to 18 W chopped pump power and pump spot sizes of a few millimeters, resulting in a fluence of ∼1 mJ/cm² and power densities as high as several 100 W/cm². This is achieved by rotating the STE at an angular speed of 100–300 Hz, which effectively reduces the locally accumulated heat by distributing the absorbed power over an area much larger than the excitation spot. In addition to the actual terahertz generation process, achieving high terahertz electric fields in an experiment requires an optimal terahertz-beam guiding. Accordingly, we optimize the collimation, propagation, and focusing of the terahertz beam generated by the rotating STE. This procedure allows us to generate 10 kV/cm peak incident terahertz fields at the tip of a low-temperature STM, resulting in several volts peak terahertz bias between the STM tip and a metallic sample.

Figure 1 shows the experimental setup. We excite the STE with 1030 nm near-infrared (NIR) pulses of ∼35 fs duration. The NIR pulses are generated by a high-power femtosecond laser (Light Conversion Carbide-80W), whose output pulses (∼200 fs) are spectrally broadened and temporally compressed in a multi-plate compressor (MPC).^19,20,22 The system offers two operating modes, providing either 50 or 20 μJ of compressed NIR pulse energy at 1 MHz or 2 MHz repetition rate. Part of the NIR power is split off for use as sampling pulses in electro-optic sampling (EOS) or for phase-resolved terahertz waveform sampling in the STM.^11,30 A large-area STE of 2″ diameter on 500 μm sapphire (TeraSpinTec GmbH) is excited at a radial position $rp=20$ mm from its center under normal incidence. The diameter $2wp$ (1/e² intensity width) of the collimated pump beam can be varied between 1.7, 2, and 3.7 mm by using different lenses in the beam path. The NIR pump power $Pp$ is controlled by a half-wave plate and a pair of thin-film polarizers. The STE is mounted on the shaft of a direct current (DC) motor that can rotate at a maximum frequency of $frot=$ 300 Hz and is magnetized by a permanent magnet of 25 mm diameter, i.e., significantly larger than the pump spot size to ensure homogeneous magnetization of the excited region [see Fig. 2(a)]. The center axis of the magnet and, thus, the magnetization is directed toward the center of the STE so that the emitted terahertz pulses are linearly polarized along the direction perpendicular to the resulting sample magnetization.

A glass plate coated with indium-tin-oxide (ITO) is mounted closely behind the STE to separate the terahertz radiation from the residual pump beam. The reflected terahertz pulses are collected by a 90° off-axis parabolic mirror (OAPM), which is part of a telescope (1:6) to expand the terahertz beam. The distance between the STE and the first telescope mirror is minimized to ∼5 cm to minimize divergence of the generated terahertz beams. A germanium (Ge) wafer is installed at an angle of 55° to block the residual pump light. A second ITO-coated glass is used to overlap the sampling pulses collinearly with the terahertz beam. A flip mirror allows to send the terahertz beam either into the STM or to an identical beam path for EOS. In both cases, we focus the terahertz beam using a 90° OAPM with a diameter of 1″ and a focal length of 35 mm. EOS traces are recorded using a 300 μm-thick ZnTe(110) crystal and are deconvoluted with the detector response function to extract the terahertz electric field at the detector position.^31–33 The terahertz power emitted by the STE is measured by a pyroelectric detector (Gentec THZ9B-BL-DA) at the EOS position, for which the NIR beam and, thus, the terahertz beam are chopped at 5 Hz with a 50% duty cycle to allow lock-in detection. All given NIR powers are chopped values. Another Ge wafer is mounted on the terahertz power meter to block NIR background radiation. A terahertz polarizer is used to block the linearly polarized terahertz pulses to estimate the power of thermal radiation emitted by the STE. Details about the terahertz power calibration and the correction for transmission losses are explained in the supplementary material. Finally, we measure the temperature of the STE by a heat camera (Model FLIR InfraCAM SD).

First, we characterize the steady-state peak temperature $Tmax$ of the STE during stationary and rotating operation. Figure 2(b) shows a heat camera image of the stationary STE excited with $2wp=3.7$ mm at $Pp=6$ W. The STE heats up to $Tmax≈125$ °C in the center of the pump beam. Although the STE is not immediately damaged in this state, its efficiency decreases irreversibly within a few minutes, or several hours to days at lower powers in the range down to 1 W. Figure 2(c) shows the spatial distribution of $Tmax$ for the STE rotating at $frot=$ 300 Hz. Remarkably, the peak temperature decreases significantly to $Tmax≈31$ °C. Moreover, the temperature increase is now distributed over a larger area, which is given by the annulus defined by $rp$ and $2wp$ and the thermal transport in the film. Note that only a sector of the annulus is visible in Fig. 2(c) because the mount partially obscures the STE.

If we assume for simplicity that the temperature increase $ΔT$ due to linear absorption scales linearly with the absorbed power density, we expect an inverse scaling with the illuminated area. From the ratio of $(π2rp2wp)/(πwp2)≈43$ of the areas of the annulus and the pump spot, the temperature increase should be reduced by a factor of ∼43 when rotating the STE. At $Pp=6$ W, $Tmax$ increases by $ΔTstat=125 °C−22 °C=103 °C,$ and $ΔTrot=31 °C−22 °C=9$°C for the stationary and rotating STE, respectively. Therefore, the temperature increase is reduced by a factor of $∼$ 11.5 when rotating the STE, which is ∼3.7 times smaller than the increase in area. This simple estimate neglects effects such as increased air cooling during rotation, heat transfer into the substrate, temperature-dependent material constants, or different volume-to-surface ratios of the excited areas. A detailed analysis of the cooling rates and steady-state temperature profile of the STE is beyond the scope of this work.

To further characterize the thermal properties of the STE, we measure the thermal radiation power $Pth$ emitted by the STE that reaches the detector. Here, $Pth$ is corrected by a factor 2 to account for the 50% loss at the polarizer to give the thermal power that reaches the sample in future experiments without the polarizer. A detailed explanation of the power calibration is provided in the supplementary material. Figure 2(d) shows $Pth$ (right ordinate) and $Tmax$ (left ordinate) measured at $frep=$ 2 MHz. For the stationary STE, $Pth$ increases steeply and similar to $Tmax$⁠, where we limit the pump power to 6 W to avoid damaging the STE. Remarkably, $Tmax$ stays below 50 °C and $Pth$ decreases considerably over the full power range. The results in Fig. 2 clearly show that moderately fast rotation efficiently reduces the average power heating of the STE.

Next, we analyze the dependence of the power $PTHz,STE$ of the coherent terahertz pulses emitted by the STE on $Pp$ for different pump-beam sizes for the stationary and rotating STE, see Fig. 3(a). $PTHz,STE$ is corrected for thermal radiation and transmission losses through the setup (see the supplementary material). For the stationary STE (open markers), after an initial increase in the range $Pp≲1$ W, $PTHz,STE$ saturates and eventually decreases for higher pump powers. When the STE rotates, $PTHz,STE$ scales quadratically up to $Pp≈1–2$ W and continues to increase with a subquadratic dependence at higher powers. We find that the scaling starts to deviate from the expected quadratic dependence at a fluence of ∼0.1 mJ/cm². This behavior is independent of the spot size, as shown together with quadratic fits to the data in the low power range plotted in the supplementary material, Fig. S3(b).

Moreover, at a given pump power, $PTHz,STE$ is higher for smaller pump beams due to the higher fluence and correspondingly higher optical-to-terahertz power conversion efficiency (PCE), as seen in Fig. 3(b) and as expected for a second-order nonlinear-optical process. The PCE saturates to values similar to those reported in previous work,²³ and a decrease in PCE is only observed for very high pump powers and fluences above ∼1 mJ/cm² [see Fig. S4(a)]. Finally, Fig. 3(a) shows that the STE performance does not depend on $frot$ between 100 and 300 Hz. This observation implies that lateral heat transport out of the illuminated area takes longer than one rotation, i.e., longer than 10 ms.

For field-driven applications such as THz-STM, the terahertz field is a more relevant quantity than the terahertz power. Figure 3(c) shows the pump-power scaling of the terahertz peak field $ETHz,pk$ for the three pump spot sizes and pump powers up to 6 W for $frot=300$ Hz. In accordance with the quadratic terahertz power scaling, a linear scaling is observed up to a power that corresponds to a fluence of ∼0.1 mJ/cm² [see also Fig. S4(b)]. Interestingly, in contrast to the terahertz power, $ETHz,pk$ is almost independent of the pump spot size for $wp=1.7$ mm and $wp=2$ mm, and it decreases only slightly for $wp=3.7$ mm. This behavior can be understood by considering that the higher terahertz power emitted at small pump spot size (i.e., at higher fluence) is compensated by the reduced terahertz beam diameter and the resulting larger terahertz focus at the detector, as sketched in the inset of Fig. 3(c) and as qualitatively supported by a simplified calculation in the supplementary material. This finding indicates that for the smaller spot sizes, the terahertz pulses propagate through the setup without major aperture losses and that the terahertz beam is slightly clipped at optical elements for the largest pump beam, as also evident from the larger horizontal width of the terahertz focus [inset of Fig. 4(c)].

The results in Fig. 3 show that for a given pump power, smaller spot sizes and higher fluences are advantageous for achieving higher terahertz powers. However, when high terahertz fields are required at low terahertz power, such as in THz-STM, larger pump spot sizes are advantageous as long as the terahertz beam propagates through the setup without clipping. The optimal condition for achieving the highest FCE and terahertz peak fields is obtained for pump spot sizes that ensure an excitation fluence of ∼0.5–1 mJ/cm², where the PCE is maximum [Fig. S4(a)], and using a telescope with a magnification that allows the full aperture of the setup to be used. However, a critical limit is reached when the beam size becomes comparable to the wavelength of the lower terahertz frequencies within the STE spectrum. In this case, low and high terahertz frequencies propagate at drastically different divergence angles, making broadband collection and collimation of the pulses difficult (see supplementary material in Ref. 34). Together with the single-pulse saturation limit imposed by the fluence,²³ this behavior will limit the performance of the STE at smaller pump spot sizes.

Figures 4(a) and 4(b) show the normalized terahertz field waveforms $ETHzt$ and their frequency spectra obtained by deconvoluting the EOS signal with the ZnTe detector response. Figure 4(c) shows the scaling of $ETHz,pk$ (left ordinate) with pump-pulse energy $Wp$⁠. At the highest available $Wp$ of about 36 μJ at 1 MHz, we achieve a terahertz peak field of ∼10 kV/cm. Note that, at $2wp=1.7$ mm, we did not use the full available power to limit the fluence and avoid laser-pulse-induced optical damage.³⁶ 

Without saturation effects and aperture losses, we expect a constant $ETHz,pk$ for constant $Wp$ independent of the pump fluence. Figure 4(c) confirms the expected behavior in the limit of small pulse energies. For low $Wp$ up to few microjoules, $ETHz,pk$ scales linearly with $Wp$ with a conversion factor that is independent of $frep$ and $2wp$⁠. At higher pulse energies, a sub-linear scaling is observed that is slightly more pronounced at 2 MHz, which we ascribe to residual pulse-to-pulse accumulation effects. At $frot=$ 300 Hz and 2 MHz, about 6600 pulses arrive at the STE during one revolution, and the STE rotates by an arc length of $≈$17 μm between two consecutive pulses. Thus, $≈$100 (⁠ $≈$215) pulses arrive in an excited area of 1.7 mm (3.7 mm) before a completely “fresh” STE region moves into the pump spot. This condition appears to be sufficient to cause accumulation effects. Its understanding requires detailed knowledge of the heat transport within the STE, which is beyond the scope of this work. A higher rotational speed in the kilohertz range combined with a larger STE should allow for a complete elimination of accumulative heating effects by operating the STE in the megahertz single-pulse regime.

Figure 4(c) (right ordinate) shows the terahertz peak bias voltage $UTHz$ that is induced between the STM tip and a metallic sample. The calibration of $UTHz$ is described in previous work and in the supplementary material.^11,30 The rotating STE allows us to generate peak terahertz bias voltages up to $≈$ 6.2 V for the tungsten tip used in this work, and we find a conversion factor of ∼0.6 V/(kV/cm) similar to previously reported values.¹⁴ We emphasize that we achieve several volts terahertz bias in the STM at terahertz pulse energies of few 10 pJ, much lower compared to similar setups using LiNbO₃. This is advantageous for heat-sensitive field-driven applications such as THz-STM and demonstrates the high quality of the terahertz focus and terahertz beam emitted by the STE.

In conclusion, we demonstrate a rotating STE operating at high pump powers and high repetition rates, resulting in power densities of up to 925 W/cm² that would lead to irreversible damage of the STE without rotation. At the highest power density used in this work (10.5 W at 1.7 mm pump beam diameter), no degradation of the STE is observed. Stable operation with no measurable fluctuations during rotation is enabled by the large-area homogeneity of the thin-film growth used for STE fabrication. We find that rotational frequencies of a few 100 Hz are sufficient to significantly reduce the temperature increase in the pumped region of the STE. This is enabled by distributing the absorbed power over a large annulus, thereby reducing the local absorbed power density. Importantly, our approach allows to excite the STE by laser pulses of several 10 μJ energy within a beam of few millimeters diameter at megahertz repetition rates, facilitating STE operation at optimal fluences of ∼0.1–1 mJ/cm² at such laser parameters. The rotating STE is of particular interest for field-driven terahertz applications that require high repetition rates and reasonably high terahertz fields, for example, THz-STM, where it can compete with standard single-cycle terahertz sources such as LiNbO₃. Furthermore, the design is scalable, and using a larger STE and faster rotational speeds, it should allow STE operation at much higher pump powers up to the kilowatt level.