Introduction

With the advancement of ultrafast laser technology, the THz wave between microwave and infrared gets much attention1,2. The resonance frequencies of various kinds of materials are at the THz band3. THz technology has a wide range of applications due to its unique features, including wireless communications4, security5, materials characterization6,7, biomedical8,9, spectroscopy, and imaging10,11,12. The development of THz technology is closely dependent on efficient sources. Traditional THz generation techniques rely on photo-generated electrons in photoconductive antennas13,14 and the optical rectification effect15,16 of non-linear crystals, such as ZnTe17, GaP18, LiNbO319 crystals but the bandwidth of the THz waves generated through these methods is limited due to phonon absorption and phase matching. The rise of spintronics has led to a type of spintronics-based terahertz emitter called spintronic terahertz emitters (STEs)20,21,22. Typical STEs are based on ferromagnetic/nonmagnetic (FM/NM) metallic bilayers, that can convert the spin-polarized currents into transient charge currents via the inverse spin Hall effect (ISHE) or inverse Rashba-Edelstein effect (IREE) and radiate THz waves under the excitation of a femtosecond laser pulse23,24,25. STEs have the advantages of low cost, ultrathin film, broadband, and optional pumping wavelength, which make STEs promising THz sources, especially for on-chip sources. In recent years, STEs have been studied extensively aiming to obtain stronger signals and wider bandwidths, such as different material compositions and geometrical stacking of FM/NM layers with a variety of thicknesses26,27,28, interface engineering of FM/NM interface29, coupling of different mechanisms30, and special patterns31. Although various strategies have been adopted to improve the THz signal, there is still a great demand for obtaining higher intensity to satisfy the practical application of ultrafast coherent control over the electron spins32, the motion of lattice ions33 and the transport of charge carriers, even across the atomic-scale junction of scanning tunneling microscopes34.

In STEs, the spin Hall angle can quantify the conversion efficiency of spin current into charge current via ISHE. Pt, W, Ta, and other common 5d heavy metals are frequently utilized as non-magnetic layers for STEs due to their large spin Hall angle35,36,37. Other materials also have been investigated to replace heavy metals with non-magnetic layers like antiferromagnets38, semiconductors39, and topological insulators (TI)40,41 like BiSe and BiTe. Although the TI materials have large spin Hall angles, they exhibit low efficiency in THz emission due to the spin-to-charge conversion (SCC) via IREE being restricted to the surface of TI. Therefore, finding materials with larger spin Hall angles and higher SCC efficiency has become one of the effective methods to improve the emission intensity of STEs.

There have been reports on the possibility of increasing the spin Hall angle of non-magnetic metals by alloying or introducing impurities into the heavy metal materials, such as the enhancement of the spin Hall effect in AuPt42, CuPt43, PtAl44, and the enhanced spin Hall angle of Pt by introducing MgO45, TiO246 into Pt to form dirty metals. However, these studies focus on the low-frequency properties of materials using electrical methods such as ferromagnetic resonance and spin Hall measurement. In contrast, terahertz emission spectroscopy can also study these doped and alloyed metals with large spin Hall angles. Actually, the employment of enhancing spin Hall angle of heavy metal by doping can provide a great chance to obtain an enhanced spintronic terahertz emission in a simple FM/NM bilayer structure. On the other hand, T. Kampfrath et al. found interfacial skew scattering47, an interfacial spin-charge conversion mechanism. Impurities introduced not only remain in the bulk of the metal but may also exist in the interface of FM/NM layers, which likely enhances the effect of interfacial skew scattering47 and increases the SCC efficiency. Therefore, to obtain a high-intensity STE, further mechanism about the spin Hall effect in the doped heavy metals still needs to be understood clearly.

In this article, we propose a high-intensity STE with a simple FM/NM bilayer structure by introducing MgO impurities into the NM heavy metal Pt, aiming at modifying the bulk spin angle and the interface conditions and thus raising the SCC efficiency. We deposited Pt0.93(MgO)0.07/CoFeB nanofilm heterostructures and successfully achieved an enhancement of 100% relative to the standard Pt/CoFeB structure, which reached the signal strength of commercial ZnTe crystals under the same pump power. Moreover, the spin-diffusion length (2 ± 0.2 nm) of Pt0.93(MgO)0.07 and thickness-dependent relative spin Hall angle concerning Pt were obtained by fitting the experimental data. Introducing MgO intersite impurities reduces the spin diffusion length but increases the bulk spin Hall angle (originates from the increase in resistance). Furthermore, interlayer investigation on THz emission further elucidates that the underlying enhancement mechanism is a combined effect of increasing bulk spin Hall angle and interfacial skew scattering. Our finding not only achieves a high-efficient THz source but also reveals the non-negligible role of interfaces in the ISHE.

Results

Enhanced THz emission from Pt1-x(MgO)x/CoFeB heterostructure

Figure 1a is a schematic diagram of our spintronic terahertz emitter for terahertz emission measurement, with the experimental setup as shown in Supplementary Note 1. The x-ray diffraction (XRD) θ−2θ patterns of Pt1−x(MgO)x single layer are shown in Fig. 1b. The typical fcc (111) diffraction peak of Pt is observed. With the increase of MgO doping content, the Pt (111) peak becomes weaker and broader, showing the reduction of Pt atomic ratio. However, the position of the peak almost does not shift, indicating that the MgO exists in the form of interstitial impurities in Pt and does not destroy the long-range order of Pt. To investigate the valence state of Pt and the sample composed of Pt, Mg, and O in Pt1-x(MgO)x, we performed x-ray photoemission spectroscopy (XPS) measurement on Pt1-x(MgO)x. The 4f7/2 and 4f5/2 peaks of Pt are shown in Fig. 1c. Obviously, the 4f7/2 and 4f5/2 peaks in pure Pt and Pt1-x(MgO)x are almost unshifted, both at 71.1 and 74.4 eV, respectively48,49. Meanwhile, that Mg 1 s and O 1 s peaks of the sample are located at 1303.9 and 532.1 eV [Supplementary Note 2], which are in agreement with 1303.9 and 532.1 eV for MgO50,51. Therefore, we believe that Pt is not oxidized in Pt1-x(MgO)x, while MgO exists in the molecular form. Figure 1d illustrates cross-sectional energy dispersive x-ray spectroscopy (EDS) Pt, Mg, and O mapping of the stack of Pt0.93(MgO)0.07(5)/CoFeB(4)/MgO(1.5) (numbers are thickness in nanometers) under the high-angle annular dark field scanning transmission electron microscopy (HAADF-STEM) mode. We can find that MgO molecules are well dispersed in the Pt layer without obvious aggregation. A cross-sectional transmission electron microscopy (TEM) image is presented in Fig. 1e, indicating that Pt1−x(MgO)x layer has a polycrystalline and amorphous mixed texture. The atomic force microscopy (AFM) measurement and the CoFeB magnetization for these stacks with different MgO concentrations as shown in Supplementary Note 3 and Supplementary Note 4, respectively. Figure 1f illustrates the typical THz waveform of the SiO2/Pt1-x(MgO)x(3)/CoFeB(4)/ MgO(1.5) multilayer film with different MgO content (x = 0, 0.02, 0.05, 0.06, 0.07, 0.13, respectively), under the same experiment condition. The shape of the THz pulses stayed constant when different MgO concentrations were doped into Pt. In contrast, the peak amplitude of the THz signal decreased first and subsequently increased, reaching its maximum amplitude at x = 0.07. Then we grew the structure Pt0.93(MgO)0.07(3)/CoFeB(4) on a 10 × 10 mm2 SiO2 substrate and compared its THz signal with Pt(3)/CoFeB(4) stack and a commercial 500-µm-thick 10 × 10 mm2 ZnTe crystal (Fig. 1g), finding that the signal of the STE could reach 200% of Pt(3)/CoFeB(4) stack and 90% of the ZnTe crystal at the same pump fluence (720 μJ cm−2). Figure 1h shows the Fourier spectra of the THz signal, where the bandwidth is limited by the detection crystal.

Fig. 1: Sample structure and enhanced THz emission of Pt1-x(MgO)x/CoFeB heterostructures.
figure 1

a Schematic diagram of our spintronic terahertz emitter in the experiment, with femtosecond laser incident from the substrate side along the z-direction and the applied magnetic field along the x-direction. b XRD θ-2θ patterns for Pt1−x(MgO)x single layer with different x. c X-ray photoemission spectroscopy (XPS) for Pt 4 f peaks in a Pt0.93(MgO)0.07 single layer. d Cross-sectional high-angle annular dark field scanning transmission electron microscopy (HAADF-STEM) image and energy-dispersive x-ray spectroscopy (EDS) mapping of Pt, Mg, and O. The white line in the lower left corner is the scale, and the length of the white line corresponds to the actual length of 10 nm. e Cross-sectional high-resolution transmission electron microscopy image of a magnetic stack of Pt0.93(MgO)0.07(5)/ CoFeB(4)/MgO(1.5). The white line in the lower left corner is the scale, and the length of the white line corresponds to the actual length of 10 nm. f THz waveforms generated by Pt1-x(MgO)x(3)/CoFeB(4) structure with different MgO content. g Comparison between the time-domain THz signal from the Pt0.93(MgO)0.07(3)/CoFeB(4) (solid red line), Pt(3)/CoFeB(4) (deep red dotted line) and that from the ZnTe crystal (light red dash-dot line), at the same pump power. h Fourier spectra obtained from the Pt0.93(MgO)0.07(3)/CoFeB(4) (solid red line), Pt(3)/CoFeB(4) (deep red dotted line) and ZnTe crystal (light red dash-dot line). The data in f are shifted horizontally for clarity.

Full size image

For better research on the enhancement of MgO doping on THz emission, we performed further experiments for the stack of Pt0.93(MgO)0.07(3)/CoFeB(4) (doped sample) and Pt(3)/ CoFeB(4) (reference sample). The THz signal emitted from Pt0.93(MgO)0.07(3)/CoFeB(4) under different magnetic field directions is shown in Fig. 2a. The phase of the THz wave undergoes a 180° shift when the direction of the magnetic field is reversed. Furthermore, the incident direction of the pump laser also plays a role in determining the phase of THz radiation. Figure 2b shows the pump fluence dependence of THz signal amplitude. The THz signal amplitude of the reference sample and the doped sample shows the same relationship with incident pump fluence. As the pump fluence increases, the signal grows approximately linearly in the low fluence domain and nonlinearly in the high fluence domain. However, both samples approach a saturated value above the same pump fluence of ~1800 μJ cm−2, indicating that MgO doping does not affect the saturation laser threshold of the sample. The signal of the doped sample is twice as high as that of the reference sample over the entire fluence range, showing that the enhancement effect is stable and independent of the pump fluence.

Fig. 2: THz radiation from Pt0.93(MgO)0.07/CoFeB heterostructure.
figure 2

a Magnetic field directions dependence and laser incident directions dependence of Pt0.93(MgO)0.07(3)/CoFeB(4) sample. The red squares represent excitation on the side of the substrate (front) and x-direction magnetic field. Light red circles represent excitation on the side of the substrate (front) and -x-direction magnetic field. Blue up arrows represent excitation on the side of the films (back) and x-direction magnetic field. Light blue down arrows represent excitation on the side of the films (back) and -x-direction magnetic field. b Terahertz amplitudes of Pt/CoFeB (blue circles) and Pt0.93(MgO)0.07/CoFeB (red triangles) under different pump fluence. The solid lines represent fitting results. c THz amplitude as a function of the sample rotation angle γ (black circles) and magnetic field rotation angle β (red triangles). The red line is a constant fit. The black line is a curve fit proportional to |cosγ|. d Relationship between terahertz amplitude and polarizer rotation angle of Pt/CoFeB (blue circles) and Pt0.93(MgO)0.07/CoFeB (red triangles). Solid lines are curve fit proportional to |cos2α|.

Full size image

We fixed the position and angle of the doped sample while rotating the magnetic field direction. Figure 2c depicts the signal amplitude as a function of magnetic field angle (β) and the function ETHz∝cosβ best fits the experimental data. We fixed the magnetic field along the x-axis in another experiment and rotated the doped sample’s azimuth. Figure 2c shows that the THz amplitude is nearly consistent regardless of the rotation angle (γ). This isotropic behavior implies that the THz enhancement is unaffected by anisotropy in the films. Figure 2d demonstrates the variation in THz amplitude caused by adjusting the angle (α) of the polarizer behind the sample. The experimental data are well fitted by the \({E}_{{THz}}\propto {E}_{0}{\cos }^{2}\alpha\) (E0 is the THz amplitude at α = 0). The THz emission behavior exhibited by the doped samples reveals that SCC still plays a key role in THz enhancement without coupling other mechanisms.

The SCC in traditional FM/NM heterostructure is usually ascribed to the ISHE or IREE. IREE usually appears in symmetry-breaking crystals or at special interfaces of thin films52. Therefore, we can determine that ISHE dominates the SCC in our experiment. Pump laser pulses excite the FM/NM heterostructure to generate the non-equilibrium carriers and then form a spin current (jS) flowing toward the z-direction with its spin polarization parallel to the magnetization (M) direction from FM to NM. The longitudinal spin currents are transformed into the transverse transient charge currents (\({J}_{C}\propto {\theta }_{{SHE}}{j}_{s}\times {{{{{\bf{M}}}}}}\), where θSHE represents the spin Hall angle for heavy metals, M denotes the magnetization of FM) due to the strong spin-orbit coupling in NM. All of the processes result in THz emission.

Spin Hall angle and spin diffusion length of Pt1-x(MgO)x

In order to investigate the underlying physics in the MgO doping enhanced THz emission process, we grew the stacks with different layer thicknesses of Pt and Pt0.93(MgO)0.07 and fixed layer thickness of CoFeB at 4 nm (4 nm which is thicker than the dead layer and the signal from 4 nm CoFeB is negligible, the film thickness dependence of the THz signal for CoFeB is shown in Supplementary Note 5). THz pulse is obtained in all samples as shown in Fig. 3a, b. As the increase of Pt0.93(MgO)0.07 thickness, the THz amplitude experiences a considerable increase firstly and then, a smooth decrease after tPt(MgO) = 3 nm (Pt(MgO) appearing later in the article represents Pt0.93(MgO)0.07). The maximum value of the THz signal is reached at tPt(MgO) = 3 nm. The trend of the THz signal variation with the layer thickness of Pt is similar to the doped sample, but the signal of Pt(t)/CoFeB(4) stack reaches its maximum value at tPt = 5 nm. In a typical FM/NM heterostructure, such thickness dependence is determined by the spin diffusion length (λs) of NM materials.

Fig. 3: Spin diffusion length and relative spin Hall angle of Pt0.93(MgO)0.07.
figure 3

THz pulses emitted from the sample were measured under different thickness of a Pt, b Pt0.93(MgO)0.07 (t = 2, 3, 4, 5, 6, 8, 10, 14 nm). c The relation of STHz/Pabs with tNM. The grey circles represent Pt0.93(MgO)0.07/CoFeB samples and the red squares represent Pt/CoFeB samples. The solid lines represent fitting results. The error bars in c are due to the deviation of the mean values of multiple measurements. d Dependence of the enhancement ratio (yellow diamond) and relative spin Hall angle (blue circles) on the NM layer thickness. When t is greater than 4 nm, the relative spin Hall Angle is around 1.65. The error bars in d are due to the deviation of the fitting of λS of Eq. (1).

Full size image

To obtain a quantitative analysis of the effect of MgO doping on the spin diffusion length, we used a theoretical model to fit the experimental results as shown in Fig. 3c (The details of the fitting method is discussed in the Supplementary Note 6.). An approximate theoretical formula for the amplitude of the terahertz electric field can be expressed as25,53

$${E}_{{THz}}\propto \frac{{{AZ}}_{0}{P}_{{abs}}T{\theta }_{{SHE}}{\lambda }_{s}\tanh (\frac{{t}_{{NM}}}{2{\lambda }_{s}})}{({t}_{{NM}}+{t}_{{FM}})({n}_{1}+{n}_{2}+{Z}_{0}{\int }_{0}^{{t}_{{NM}}+{t}_{{FM}}}\sigma (z){dz})}$$
(1)

Where Z0 = 377 Ω, is the impedance of the vacuum, n1 = 1 and n2 = 1.98 are the refractive index of air, and the SiO2 substrate (n2 was obtained by THz transmission spectroscopy), respectively, θSHE is the spin Hall angle of NM material, λS is the spin diffusion length of NM layer, A is an insignificant parameter that depends on the excitation of the sample by the incident laser, t is the films thickness, Pabs is the absorption of the laser power by the sample [Supplementary Note 7], σ is the terahertz conductivity [Supplementary Note 8], T is the spin transparency from the FM layer to the NM layer. The doped and reference samples’ spin diffusion lengths are fitted as λPt(MgO) = 2 ± 0.2 nm and λPt = 4 ± 0.2 nm for the doped sample and reference samples, respectively. The spin diffusion length we obtained for Pt is comparable to that reported in earlier works53,54. However, the spin-diffusion length of Pt1-x(MgO)x has rarely been studied, and in most works, the spin-diffusion length was approximated by the electrical conductivity45,55. In addition, we can see from Fig. 3c that the fitted curves deviate from the experimental data in the range of thin NM layers. We attribute this phenomenon to the neglect of interfacial spin loss. Interfacial spin resistance and spin memory loss occur at the interface of FM and NM films due to resistive mismatch and spin scattering, which causes spin loss27. The reduction in the spin-diffusion length of Pt0.93(MgO)0.07 relative to Pt is attributed to the doping of MgO, which enhances the spin flip of polarized electrons. Another key parameter in Eq. (1) that affects terahertz emission is spin Hall angle, θSHE. To obtain the relative spin Hall angle of Pt0.93(MgO)0.07 with respect to Pt, we assume that the two structures have the same spin transparency T and A. We used the spin diffusion length that fit above and compared the THz signals generated by the two kinds of structures (shown in Fig. 3d by yellow diamond) with Eq. (1) to obtain the relative spin Hall angle θrela = θPt(MgO)/θPt (θPt(MgO) and θPt represent spin Hall angles for Pt(MgO) and Pt, respectively)55. The computed relative spin Hall angle is NM film thickness dependent, and the relationship between θrela and tNM is shown in Fig. 3d (blue circles). We find that Pt0.93(MgO)0.07 exhibits an almost stable and constant relative spin Hall angle of about 1.65 ± 0.20 ( ± 0.20 represents that the calculated spin Hall Angle at different Pt thicknesses fluctuates in the range of 0.20 around 1.65, as shown in Fig. 3d) in the thick NM thickness range (t å λPt), while shows a larger relative spin Hall angle in the thin NM thickness range, particularly at t = 2 nm, where the θrela of Pt0.93(MgO)0.07 approaches 3.5.

The origin of high-intensity THz signal

Considering the spin Hall effect (SHE) based on spin-orbit interactions can be divided into intrinsic and extrinsic mechanisms (side jump and skew scattering), the spin Hall resistivity and spin Hall angle can be expressed as \({\rho }_{{SH}}={\sigma }_{{SH}}^{{{{{\mathrm{int}}}}}}{\rho }_{{xx}}^{2}+{\rho }_{{SH}}^{{imp}}\) and \({\theta }_{{SH}}={\rho }_{{SH}}/{\rho }_{{xx}}\), where \({\rho }_{{SH}}\) is the total spin Hall resistivity, \({\sigma }_{{SH}}^{{{{{\mathrm{int}}}}}}\) is the intrinsic spin Hall conductivity of Pt, \({\rho }_{{xx}}\) is the longitudinal electrical resistivity, and \({\rho }_{{SH}}^{{imp}}\) is the extrinsic spin Hall resistivity induced by the MgO intersite impurities56. Typically, The \({\sigma }_{{SH}}^{{{{{\mathrm{int}}}}}}\) of Pt originates from the intrinsic SHE determined by the topology of band structure, which for simple fcc metals is only dependent on the long-range crystal structure57. Figure 1c depicts the XRD θ−2θ patterns of Pt1−x(MgO)x single layer. As x rises, Pt still exhibits robust long-range fcc order as x increases, indicating negligible modification in intrinsic \({\sigma }_{{SH}}^{{{{{\mathrm{int}}}}}}\). However, a giant variation in \({\sigma }_{{SH}}^{{{{{\mathrm{int}}}}}}\) can be brought via tuning the scattering from the finely dispersed MgO intersite impurities, which are built on the basis of high doping concentration45. The variation in the \({\sigma }_{{SH}}^{{{{{\mathrm{int}}}}}}\) of our samples is insignificant since the concentration of MgO in Pt in our study is low, as evidenced by the weak change in intensity and full width at half maximum in the XRD patterns (Fig. 1c). Supplementary Fig. 8b depicts the resistance of a single NM layer;, the room temperature resistivities of the doped and reference samples are ρPt(MgO) = 54 μΩ cm and ρPt = 36 μΩ cm, respectively. The introduction of MgO forms a large number of intersite impurities to concentrate electrons and create an internal electric field, thus increasing the resistance. Theoretically, the intrinsic contribution dominates the SHE of Pt in the dirty-metal regime (ρxx larger than ~30 μΩ·cm) while the extrinsic contributions become dominant in the ultraclean regime58. Therefore, the variation of \({\rho }_{{SH}}^{{imp}}\) should be small. On the other hand, the contribution of skew scattering should scale inversely with the residual resistivity ratio (RRR = ρxx/ρxx,0, where ρxx,0 is the low-temperature resistivity)59, the RRR of the doped and reference samples are 1.12 and 1.16, respectively, the slight change indicating this contribution is negligible in our samples. And side jump should also give rise to a linear contribution in θSHE with respect to ρxx like intrinsic mechanism43. Obviously, we can obtain that the spin Hall angle is approximately proportional to the longitudinal electrical resistivity (θSHE ∝ ρxx) in our low-MgO doping samples. The relative resistance at room temperature ρrela = ρPt(MgO)/ρPt = 1.50, in excellent agreement with the relative spin Hall angle (θrela = 1.65 ± 0.20) as shown in Fig. 3d. Although continuing to increase the doping concentration will further increase the resistance, the carrier lifetime rapidly decreases, thereby reducing the intrinsic spin Hall conductance and shortening the spin diffusion length, resulting in a weakened THz signal. By now we infer that the competing relationship of enhanced bulk spin Hall angle, which is caused by the increase in resistance, and decreased spin diffusion length can be used to explain the influence of THz signal due to low-MgO doping. According to this extrapolation, we should get a stable spin Hall angle increase rather than a NM film thickness-dependent relative spin Hall angle. And the resistance change we measured is not large enough to bring about such a large signal enhancement in the thin NM layer thickness range. Therefore, in addition to the bulk-like effects, the interfacial contributions also need to be considered in the process of THz emission from STEs47,60.

To evaluate whether the FM/NM interface plays a role in the enhancement phenomenon and investigate the relationship of θrela-tNM, we study samples with varied interlayer thicknesses using Cu as the interlayer. Cu is suitable as an interlayer due to its long spin diffusion length and negligibly small ISHE. As we can see in Fig. 4, the introduction of the Cu interlayer changes the THz intensity of all samples. Many previous works53,61 have reported that the Cu interlayer suppresses the transmission of spin current and thus reduces the THz signal, but in our experiments, the introduction of the Cu interlayer has a different effect for samples with different Pt layer thicknesses. In the doped samples, the Cu interlayer attenuates the THz signal, but in the reference sample, the THz signal is enhanced due to the introduction of Cu at tPt = 2 and 3 nm (Fig. 4a, b). This phenomenon may be caused by the variation of pump laser absorbance and the ISHE of Cu itself. However, the experimental results prove that these cases are negligible (Supplementary Note 9 and Supplementary Note 10).

Fig. 4: The effect of interlayer on the amplitude of THz signal.
figure 4

Cu thickness dependence of the THz amplitude from Pt0.93(MgO)0.07/CoFeB (blue circles) and Pt/CoFeB (red squares) with different thickness of NM a NM(2)/Cu(t)/CoFeB(4), b NM(3)/Cu(t)/CoFeB(4), c NM(5)/Cu(t)/CoFeB(4), d NM(8)/Cu(t)/CoFeB(4). The error bars in a, b, c, and d are due to deviation from the mean values of multiple measurements. e THz emission signals of Pt(3)/CoFeB(4) (red solid line), Pt0.93(MgO)0.07(3)/CoFeB(4) (red dotted line), Pt(3)/MgO(1)/CoFeB(4) (blue solid line) and Pt0.93(MgO)0.07(3)/ MgO(1)/CoFeB(4) (blue dotted line) samples.

Full size image

Then we turn our attention to the NM/FM interface, Pt/Cu(tCu) /CoFeB and Pt(MgO)/Cu(tCu)/CoFeB exhibit different Cu thickness dependence in thin NM layer thickness range (Fig. 4a, b) implying that the interface of doped sample and reference sample is different and that the bulk property of Pt(MgO) is not the only reason for terahertz signal enhancement. The EDS mapping image (Fig. 1d) shows that the MgO impurities are not only uniformly distributed in the Pt, but also contain a large amount of Mg elements within the CoFeB layer, which comes partly from the penetration of the MgO protective layer and partly from the penetration of the MgO impurities in the Pt, having a modification on the Pt(MgO)/CoFeB interface. In the thick NM layer region, Fig. 4c, d exhibit identical Cu thickness dependence, indicating that interfacial effects can be ignored. In another experiment, we replaced Cu with MgO as an interlayer. The THz signals generated from Pt(3)/MgO(1)/CoFeB (4) and Pt0.93(MgO)0.07(3)/MgO(1)/CoFeB(4) are shown in Fig. 4e. The insertion of MgO severely attenuates the spin current, causing the signal to decrease rapidly. Interestingly, we found that the signals of the doped sample and the reference sample were almost the same when a 1 nm MgO interlayer was inserted. The insertion of MgO destroys the FM/NM interface on one hand, while the diffusion of MgO from the interlayer into the NM layer causes Pt to form Pt1-x(MgO)x, both of which make the reference sample and the doped sample form similar structural compositions thus making the signals similar. Therefore, we propose the following hypothesis: The role of bulk and interface differs significantly in NM layers of varying thickness.

Figure 5a depicts the generation of charge current in a thin NM layer, where the bulk effect (Jc1) is weak due to restricted spin diffusion, whereas the interface has a large contribution (Jc2). The lower panel of Fig. 5a shows that in the doped sample, both interfacial and bulk effects are considerably amplified, increasing final charge current. The interface has been reported to contribute considerably to the ISHE, ascribed to impurity skew scattering47 at the interface. MgO molecule diffuses into the Pt(MgO)/ CoFeB interface region and the interior of CoFeB, amplifying the skew scattering of spin-polarized electrons at the interface and in CoFeB as illustrated in the lower panel of Fig. 5a, resulting in terahertz signal enhancement. This explains the doped samples’ notably high relative spin Hall angles for thin NM layers in our finding. Introducing the Cu interlayer disrupts the Pt(MgO)/CoFeB interface and decreases the impact of interfacial skew scattering, decreasing the THz signal. While the THz signal in Pt/Cu/ CoFeB is visibly enhanced compared to Pt/CoFeB, this phenomenon may be due to the characteristics of distinct interfaces, such as interface-caused enhancement of ISHE in thin Pt range62 or IREE induced at the Pt/Cu interface63. The scenario differs when the NM layer thickness is substantial, as seen in Fig. 5b. Because the NM layer thickness is greater than the spin diffusion length in this situation, the spin current may flow sufficiently into NM to provide a large charge current, and the effect produced by the interface can be ignored in terms of the bulk effect. Considering only the enhanced bulk spin Hall angle and the reduced spin diffusion length without the interface effect coincides well with the experimental data, which show that the signals of the doped sample and the reference sample converge at thick NM and have identical Cu thickness dependency. In addition, as shown in Supplementary Note 11, various enhancement phenomena arise in different Pt1-x(MgO)x/FM systems, indicating that the interface plays a non-negligible role in THz signal enhancement. This interface effect is often overlooked in much of the works based on electrical methods45,46.

Fig. 5: Separation of bulk effect and interface effect.
figure 5

Schematic diagram of terahertz enhancement by MgO doping at different NM thicknesses. a t <λ, b tå λ. jS, jC1, and jC2 represent the spin current, the charge current generated in the bulk, and the charge current generated at the interface, respectively. The spin diffusion length of Pt and Pt(MgO) is represented by λPt and λPt(MgO). The enlarged figure I represents that the bulk spin Hall angle increases from θ1 to θ2 after doping MgO. The enlarged figure II represents the interface skew scattering contribution caused by MgO molecules. Separation of interfacial (red squares) and bulk (black circles) enhancement effects in Pt0.93(MgO)0.07(t)/CoFeB c t = 2 nm, d t = 3 nm.

Full size image

The 2 nm Cu interlayer is sufficient to disrupt the NM/FM interface. To separate the bulk and interfacial contributions, we approximate Ebluk + Einterface = EPt(MgO)/CoFeB − EPt/CoFeB, Ebluk = EPt(MgO)/Cu(2)/CoFeB − EPt/CoFeB, so Einterface = EPt(MgO)/CoFeB − EPt(MgO)/Cu(2)/CoFeB (Fig. 5c, d). We find that the interfacial effect is non-negligible at thin NM layers. Here, We ignored the attenuating effect of Cu on the spin current in Pt(MgO)/Cu(2)/CoFeB and the probable enhancing effect of the Pt0.93(MgO)0.07/Cu interface, which would greatly underestimate the role of the interface in our calculations. However, more detailed studies can be done for a further understanding of the enhancement. In addition, we conducted experiments on W1-x(MgO)x and obtained similar results to Pt (Supplementary Note 12). This result shows that our method is somewhat universal.

Conclusions

In summary, we propose a high-intensity spintronic terahertz emitter with a simple FM/NM bilayer structure by doping MgO impurity into Pt as the NM material and demonstrate THz signal enhancement compared to the undoped sample. An optimized Pt0.93(MgO)0.07/CoFeB structure reaches a 200% intensity compared to Pt/CoFeB and approaches the signal of 500μm ZnTe crystal. The mechanism of THz emission from the doped sample is still ISHE. We fitted the experimental results to the spin-diffusion model and found that MgO doping shortened the spin-diffusion length. A thickness-dependent relative spin Hall angle was obtained by further analysis. We attribute the enhancement of the THz signal to increased bulk spin Hall angle and interfacial skew scattering by MgO impurities. Additionally, we separated the effect of interfacial contribution and bulk contribution using the interlayer experiment. And we find that the Cu interlayer has different effects on different thicknesses of Pt. Our findings also reveal the role of the interface in the THz emission of spintronics sources. Furthermore, stronger THz signals may be obtained by optimizing the doping concentration, changing the main NM material, changing the doping material, or combining with the NM/FM/NM trilayer or the multilayer structures, which proposes a promising route to enhance the performance of THz optoelectronic devices.

Methods

Sample fabrication

The Pt1-x(MgO)x/CoFeB/MgO stack with different MgO content was grown on a 500-µm-thick 5 × 5 mm2 SiO2 substrate by magnetron sputtering. The Pt1-x(MgO)x layer was co-sputtered from a Pt and MgO target. The speed of film growth was measured by X-ray reflection (XRR) as shown in Supplementary Note 13. The content of MgO was controlled by fixing the sputtering power of the Pt target and changing the sputtering power of the MgO target. The content of MgO and elemental valence state were determined by x-ray photoemission spectroscopy (XPS). MgO capping was used as a protective layer to prevent sample oxidation. Element content and distribution were determined by cross-sectional energy dispersive X-ray spectroscopy (EDS).

Microscopic topography experiments

The cross-sectional morphology of the stack and the surface morphology of the Pt1-x(MgO)x single layers were measured by cross-sectional transmission electron microscopy (TEM) and atomic force microscopy (AFM), respectively.

THz emission spectroscopy measurement

A static external magnetic field of 700 Oe was applied parallel to the sample surface along the x-axis to keep the CoFeB saturated magnetization. We used a Ti: Sapphire laser with a pulse width of 35 fs, a center wavelength of 800 nm, and a repetition rate of 1 kHz. The exciting beam was focused on the sample with a spot size of about 3 mm in diameter and an average pump fluence of 720 μJ cm−2. In general, the laser can be incident from the substrate side or the sample side. But in the paper, the laser was incident from the substrate side to avoid undesired THz absorption by the substrate unless otherwise stated for convenience. The THz emission was collected by a pair of parabolic mirrors and then measured using electro-optic sampling in a 500-µm-thick (110)-oriented ZnTe crystal. All the measurements were conducted at room temperature in dry air gas to prevent water vapor absorption in the ambient air. (The details of the setup are shown in Supplementary Note 4).

Vibrating sample magnetometry (VSM) measurements

The sample magnetization was measured by a superconducting quantum interference device-vibrating sample magnetometry (VSM) at room temperature.