Effect of Cu Intercalation Layer on the Enhancement of Spin-to-Charge Conversion in Py/Cu/Bi2Se3

The spin-to-charge conversion in Permalloy (Py)/Cu/Bi2Se3 is tunable by changing the Cu layer thickness. The conversion rate was studied using the spin pumping technique. The inverse Edelstein effect (IEE) length λIEE is found to increase up to ~2.7 nm when a 7 nm Cu layer is introduced. Interestingly, the maximized λIEE is obtained when the effective spin-mixing conductance (and thus Js) is decreased due to Cu insertion. The monotonic increase in λIEE with decreasing Js suggests that the IEE relaxation time (τ) is enhanced due to the additional tunnelling barrier (Cu layer) that limits the interfacial transmission rate. The results demonstrate the importance of interface engineering in the magnetic heterostructure of Py/topological insulators (TIs), the key factor in optimizing spin-to-charge conversion efficiency.


Introduction
Traditional electronic devices use electrical charge and voltage to process and read information. In addition, spin-based electronic devices use electron spins to carry information. The term "spin-to-charge conversion" essentially refers to the ability of materials to link the electrical charge for information-processing with the use of electron spins. In other words, efficient spin-to-charge conversion permits the effective exchange of spin currents into voltages, allowing electronic devices to easily read information. Spin-to-charge conversion can be achieved in Rashba systems and topological insulators (TIs) through the IEE [1,2]. Among these materials, a TI-based heterostructure is a prospective candidate for ultrahigh inverse Edelstein effect length (λ IEE ) due to its unique surface states with spin-momentum locking [3,4]. Moreover, highly efficient spin-to-charge conversion can be used in logic devices such as the magneto-electric spin-orbit (MESO) device [5,6]. Therefore, the enhancement of spin-to-charge conversion in TIs is suitable for readout schemes in many advanced spintronic devices.
Due to the spin momentum locking, the 3D spin current density J s injected onto the TI surface will produce two-dimensional (2D) charge density J 2D c on the TI surface states (SS), the so-called IEE. The IEE length (λ IEE ) is determined to be J 2D c /J s , which can be experimentally probed using the spin pumping technique [7][8][9][10][11][12]. Numerous studies have been carried out to determine the spin-to-charge conversion efficiencies in 3D Tis [3,4,7,10,12]. In particular, a spin Hall angle (SHA) as large as~0.43 has been reported in Bi 2 Se 3 , which is attributed to the enhanced spin current through surface states which is then converted into DC-voltage due to bulk inverse spin Hall effect (ISHE) [11]. However, Nanomaterials 2022, 12, 3687 2 of 9 large variations in the SHA were found, in an order of magnitude difference, which the authors attribute to the inhomogeneity of the interface quality [11].
On the other hand, a dominant role for surface states in spin-to-charge conversion was observed, despite the unavoidable conducting bulk in Bi 2 Se 3 [7]. Although the thickness of Bi 2 Se 3 varied from 2 QL to 60 QL, the effective spin-mixing conductance does not increase monotonically, suggesting that surface states dominated the mechanism [7]. Clearly, the spin pumping characteristics are an important parameter to study the spin-to-charge conversion mechanism in 3D TIs, where controlling the interfacial properties is a necessary step [4,11].
To improve the spin-injection efficiency via the ISHE, it has been proposed to use an intercalator (e.g., Cu or Ag) as a potential barrier between the ferromagnets (FM) and the heavy metal (HM) [13,14]. This is effective in tuning the interfacial spin-dependent resistivity and improving the spin injection. Cu is widely used to control the spin transmissivity in multilayer devices [13,15,16]. Du et al., demonstrated that the insertion of a Cu layer between Y 3 Fe 5 O 12 (YIG) and W substantially improved the spin current injection into W, while a similar insertion between YIG and Pt degraded the spin current [13]. The authors reported a quantitative analysis finding that the spin transport efficiency in heterostructures depends on the spin conductance of each component and their interfaces [13]. Similar results were reported by Deorani et al., where the effect of the Cu interlayer on spin-mixing conductance does depend on the materials (Pt versus Ta) [16]. The spatial mapping of spin accumulation in Cu due to the spin-pumping effect was observed using scanning transmission X-ray microscopy [17]. Recently, Cu layers have been deposited on TI films to eliminate proximity-induced ferromagnetism in spin-orbit torque (SOT) devices [18]. In particular, the deposition of Cu on TI may play a minor role in changing the surface-state conditions, such as causing an energy shift of the Dirac point and valence-band maximum, since the work function of Cu and the electron negativity of TI are of similar magnitudes [19]. Although Cu is the most commonly used spacer layer in the spintronic devices, there is still a lack of quantitative studies on the role of Cu insertion in the spin-to-charge conversion in TIs that measured based on a spin-pumping mechanism.
In this work, we fabricated a trilayer structure of Py/Cu/Bi 2 Se 3 and studied the spinpumping characteristic by varying the thickness of the Cu layer (as shown in Figure 1a,b). The Cu layer was used to protect the TI surface from exchange interaction with Py. Our results show that Cu also acts as a barrier for spin transport into the TI film. More importantly, the spin-to-charge conversion efficiency was enhanced due to the introduction of the Cu barrier. The related mechanism is discussed in this work. Bi2Se3, which is attributed to the enhanced spin current through surface states which is then converted into DC-voltage due to bulk inverse spin Hall effect (ISHE) [11]. However, large variations in the SHA were found, in an order of magnitude difference, which the authors attribute to the inhomogeneity of the interface quality [11]. On the other hand, a dominant role for surface states in spin-to-charge conversion was observed, despite the unavoidable conducting bulk in Bi2Se3 [7]. Although the thickness of Bi2Se3 varied from 2 QL to 60 QL, the effective spin-mixing conductance does not increase monotonically, suggesting that surface states dominated the mechanism [7]. Clearly, the spin pumping characteristics are an important parameter to study the spinto-charge conversion mechanism in 3D TIs, where controlling the interfacial properties is a necessary step [4,11].
To improve the spin-injection efficiency via the ISHE, it has been proposed to use an intercalator (e.g., Cu or Ag) as a potential barrier between the ferromagnets (FM) and the heavy metal (HM) [13,14]. This is effective in tuning the interfacial spin-dependent resistivity and improving the spin injection. Cu is widely used to control the spin transmissivity in multilayer devices [13,15,16]. Du et al., demonstrated that the insertion of a Cu layer between Y3Fe5O12 (YIG) and W substantially improved the spin current injection into W, while a similar insertion between YIG and Pt degraded the spin current [13]. The authors reported a quantitative analysis finding that the spin transport efficiency in heterostructures depends on the spin conductance of each component and their interfaces [13]. Similar results were reported by Deorani et al., where the effect of the Cu interlayer on spinmixing conductance does depend on the materials (Pt versus Ta) [16]. The spatial mapping of spin accumulation in Cu due to the spin-pumping effect was observed using scanning transmission X-ray microscopy [17]. Recently, Cu layers have been deposited on TI films to eliminate proximity-induced ferromagnetism in spin-orbit torque (SOT) devices [18]. In particular, the deposition of Cu on TI may play a minor role in changing the surface-state conditions, such as causing an energy shift of the Dirac point and valence-band maximum, since the work function of Cu and the electron negativity of TI are of similar magnitudes [19]. Although Cu is the most commonly used spacer layer in the spintronic devices, there is still a lack of quantitative studies on the role of Cu insertion in the spinto-charge conversion in TIs that measured based on a spin-pumping mechanism.
In this work, we fabricated a trilayer structure of Py/Cu/Bi2Se3 and studied the spinpumping characteristic by varying the thickness of the Cu layer (as shown in Figure 1a,b). The Cu layer was used to protect the TI surface from exchange interaction with Py. Our results show that Cu also acts as a barrier for spin transport into the TI film. More importantly, the spin-to-charge conversion efficiency was enhanced due to the introduction of the Cu barrier. The related mechanism is discussed in this work.

Experimental
Bi 2 Se 3 films with a thickness of 10 nm were synthesized using the molecular beam epitaxy (MBE) method [10]. The structural characterization of Bi 2 Se 3 thin films is investigated by X-ray diffraction (XRD) and atomic force microscopy (AFM) (see Section 2 of the Supplementary Information). The as-grown Bi 2 Se 3 were in situ capped with a 2 nm Se layer, which was used as a protective layer. The samples were then transferred into a pulsed laser deposition (PLD) chamber for deposition of Cu and subsequently Ni 80 Fe 20 (Py) layers at room temperature. Before depositions, the Se layer was decapped in the PLD chamber at about 180 • C for 1 h. A series of trilayer samples were prepared by varying the thickness of Cu from 3 to 11 nm. The quality of the Py/Cu/ Bi 2 Se 3 trilayer was examined by high-resolution transmission electron microscopy (HRTEM), as shown in Figure S3c in the Supplementary Information). AvPy/Bi 2 Se 3 bilayer was also prepared for comparison. The Py thickness was fixed at 17 nm. A 1 nm of Al film was deposited on Py as a capping layer. To evaluate the spin-to-charge conversion, a spin-pumping technique was utilized (Figure 1). A spin current was generated in Py via its ferromagnetic resonance (FMR) condition and injected into Bi 2 Se 3 , passing through the Cu-inserted layer (-z direction) (Figure 1b). The DC voltage was measured in the x-direction and the resulting 2D charge current can be evaluated [12]. All measurements were performed at room temperature. Figure 2a shows the spin-pumping voltage as a function of the magnetic field (H) measured at an excitation frequency of 3 GHz for the sample Py/Cu (7nm)/Bi 2 Se 3 . The results of other frequencies and magnetic field sweeps can be found in Figures S1 and S2 in the Supplementary Information. The voltage signals consisting of symmetric (V s ) and antisymmetric (V as ) parts can be isolated by fitting the measured voltage (data curve) to the form

Results and Discussion
Here H r is the FMR resonant field and ∆H is the line width of the signal. The obtained DC voltage signals consist of several components of the IEE, the inverse spin Hall effect (ISHE) [13,14,20], the shunting effect [13,21], the spin rectification effects (including the anomalous Hall effect (AHE) and the anisotropic magnetoresistance (AMR)) [21,22]. The contributions of the IEE and ISHE are related to the V s of the resonance field. However, due to the shunting effect of the Py layer and the overlapping symmetrical AMR and AHE signals in the excitation geometry, the relative weights of the contribution of the voltage generated by the IEE to the total Vs cannot be quantitatively separated [23,24]. Moreover, the voltage polarization of our results is similar to other Bi 2 Se 3 -based spin pumping [7,25,26]. Similar fitting was also done here, and V s was extracted as shown in Figure 2b. It was found that Vs is larger in the presence of a Cu layer. FMR experiments were also conducted as shown in Figure 2c,d.
The FMR linewidth (∆H pp ) of samples with different Cu thicknesses was plotted as a function of the excitation frequency for the Py/Cu/Bi 2 Se 3 samples in Figure 2d. The damping factor (α) was obtained by fitting ∆H pp to f using this formula, ∆H pp = H 0 + 4π γ α f , in which H 0 corresponds to the presence in the Py layer [13,14]. Compared with single Py, the linear fitted slopes are larger for the Py/Bi 2 Se 3 bilayer and Py/Cu/Bi 2 Se 3 trilayer samples, indicating that the injection of the spin current into Bi 2 Se 3 results in the broadening of the FMR linewidth and thus larger damping constants α. Interestingly, α Py/Cu/Bi2Se3 was found to decrease from (1.262 ± 0.05) × 10 −2 to (1.185 ± 0.05) × 10 −2 when the thickness of the Cu layer was increased to 7 nm. The resistance of the multilayer samples Rd was measured using a four-probe method.
was determined as = Ic/w = Vs/wRd, where w and Ic are the width of the sample and charge current, as shown in Figure 3a. The estimated was derived from the 2D charge current in the x-direction at the interface via the inverse Edelstein effect (IEE), the charge current induced by the ISHE of the Cu layer [27], and the current derived from the ferromagnetic transport in the Py layer. We evaluated the spin-to-charge conversion /Js. using standard analysis of spin pumping on TI [7,10,12]. The spin-mixing conductance ¯ used to account for the efficiency of generating the spin current was extracted using Equation (2): where Ms is the saturation magnetization of Py, tPy is the thickness of Py, g is the Landé factor and uB is the Bohr magneton. Ms was calculated from f vs. Hr using the Kittel formula, = + 4 , in which γ is the gyromagnetic ratio to extract the effective saturation magnetization ( ) (Figure 2c) [10,12,13]. Dα = αPy/Cu/Bi2Se3 − αPy and is determined by analyzing DHpp vs. f, as shown in Figure 2d. For the spin current densities injected through the interface due to spin pumping, Equation (3) was utilized as follows: The resistance of the multilayer samples R d was measured using a four-probe method. J 2D c was determined as J 2D c = I c /w = V s /wR d , where w and I c are the width of the sample and charge current, as shown in Figure 3a. The estimated J 2D c was derived from the 2D charge current in the x-direction at the interface via the inverse Edelstein effect (IEE), the charge current induced by the ISHE of the Cu layer [27], and the current derived from the ferromagnetic transport in the Py layer. We evaluated the spin-to-charge conversion J 2D c /J s . using standard analysis of spin pumping on TI [7,10,12]. The spin-mixing conductance G ↑↓ e f f used to account for the efficiency of generating the spin current was extracted using Equation (2): where M s is the saturation magnetization of Py, t Py is the thickness of Py, g is the Landé factor and u B is the Bohr magneton. M s was calculated from f vs. H r using the Kittel formula, f = γ 2π H r H r + 4πM e f f , in which γ is the gyromagnetic ratio to extract the effective saturation magnetization (M e f f ) (Figure 2c) [10,12,13]. ∆α = α Py/Cu/Bi2Se3 − α Py and is determined by analyzing ∆H pp vs. f, as shown in Figure 2d. For the spin current Nanomaterials 2022, 12, 3687 5 of 9 densities injected through the interface due to spin pumping, Equation (3) was utilized as follows: in which γ is the gyromagnetic ratio, ω(=2πf) is the frequency, and h rf is the amplitude of the microwave rf field . h rf = I rf 2w , [10][11][12] in which I r f is the microwave current at a frequency of 3 GHz and w is the linewidth of the coplanar waveguide, respectively. h rf is estimated to be 0.112 Oe for I r f = 0.0178 A and w = 1 mm. The calculated J s is presented in Figure 3b. By dividing J 2D c with J s , the spin-to-charge conversion efficiency J 2D c /J s (λ IEE ) can be determined.
in which γ is the gyromagnetic ratio, ω( = 2πf) is the frequency, and hrf is the amplitude of the microwave rf field. h = , [10][11][12] in which is the microwave current at a frequency of 3 GHz and is the linewidth of the coplanar waveguide, respectively. h is estimated to be 0.112 Oe for = 0.0178 A and w = 1 mm. The calculated Js is presented in Figure 3b. By dividing with Js, the spin-to-charge conversion efficiency /Js (λIEE) can be determined.  Figure 3a plots the versus tCu. There is an optimized at a thickness of 3 nm and 7 nm. In contrast, Js decreases when 3 and 7 nm Cu are added, as shown in Figure 3b. The variation trend of Js vs tCu is consistent with the change in the effective spin-mixing conductance G ¯ , where the G ¯ is found to decrease with the introduction of 3 and 7 nm Cu in Figure 4c. The variation in G ¯ is discussed below. Interestingly, a maximized /Js is observed at tCu = 7 nm, where λIEE reaches ~2.7 nm, as shown in Figure 3c. This result suggests that the optimization of /Js may be related to the reduction of Js due to the Cu insertion. In addition, the value of λIEE = 1.25 nm for the Py/Bi2Se3 without the Cu insertion sample, which is larger than that previously reported on FM/Bi2Se3 systems [7,10,11].
To investigate the possible reason for the enhancement of /Js, we plotted /Js as a function of the effective spin-mixing conductance ¯ (Py/Cu/TI) as shown in Figure  4a. Various ¯ (Py/Cu/TI) values were obtained by changing the Cu layer thickness. Larger /Js are obtained at low values of ¯ (Py/Cu/TI) (hence the minimum Js as shown in Figure 3b). We further examined vs. ¯ (Py/Cu/TI), as shown in Figure  4b.
does not increase with the increasing ¯ (Py/Cu/TI), revealing that the spin-tocharge mechanism may not be dominated by the bulk spin Hall effect (SHE) [16]. Therefore, we propose here that the spin-to-charge conversion in the Py/Cu/Bi2Se3 system arises from the IEE, where the origin is the spin-momentum locked surface states of the TI layer, as explained in other literatures [7,25,26].  Figure 3a plots the J 2D c versus t Cu . There is an optimized J 2D c at a thickness of 3 nm and 7 nm. In contrast, J s decreases when 3 and 7 nm Cu are added, as shown in Figure 3b. The variation trend of J s vs t Cu is consistent with the change in the effective spin-mixing conductance G ↑↓ eff , where the G ↑↓ eff is found to decrease with the introduction of 3 and 7 nm Cu in Figure 4c. The variation in G ↑↓ eff is discussed below. Interestingly, a maximized J 2D c /J s is observed at t Cu = 7 nm, where λ IEE reaches~2.7 nm, as shown in Figure 3c. This result suggests that the optimization of J 2D c /J s may be related to the reduction of J s due to the Cu insertion. In addition, the value of λ IEE = 1.25 nm for the Py/Bi 2 Se 3 without the Cu insertion sample, which is larger than that previously reported on FM/Bi 2 Se 3 systems [7,10,11].
To investigate the possible reason for the enhancement of J 2D c /J s , we plotted J 2D c /J s as a function of the effective spin-mixing conductance G ↑↓ e f f (Py/Cu/TI) as shown in Figure 4a. Various G ↑↓ e f f (Py/Cu/TI) values were obtained by changing the Cu layer thickness. Large r J 2D c /J s are obtained at low values of G ↑↓ e f f (Py/Cu/TI) (hence the minimum J s as shown in Figure 3b). We further examined J 2D c vs.
G ↑↓ e f f (Py/Cu/TI), as shown in Figure 4b. J 2D c does not increase with the increasing G ↑↓ e f f (Py/Cu/TI), revealing that the spin-to-charge mechanism may not be dominated by the bulk spin Hall effect (SHE) [16]. Therefore, we propose here that the spin-to-charge conversion in the Py/Cu/Bi 2 Se 3 system arises from the IEE, where the origin is the spin-momentum locked surface states of the TI layer, as explained in other literatures [7,25,26]. Low ¯ (Py/Cu/TI) indicates a strong spin backflow and spin memory loss (spin absorption) at the high SOC interface [28,29]. Both factors are relevant in this Py/Cu/TI trilayer system. If we examine ¯ (Py/Cu/TI) at various tCu, as presented in Figure 4c, except for Py/Cu (3 nm)/TI and Py/Cu (7 nm)/TI, the samples Py/TI, Py/Cu (9 nm)/TI and Py/Cu(11 nm)/TI exhibit ¯ (Py/Cu/TI) ~1.25  10 19 m −2 , which is typical for metal-metal interfaces [30,31]. As reported by Du et al., the effective spin-mixing conductance of the trilayer system (FM/Cu/NM, FM for ferromagnetic, while it is NM for nonmagnetic material) is determined by the serial contribution of the two interfaces (FM/Cu and Cu/NM) and the spin resistance of Cu [13]. Here we refer to FM as Py and NM as the TI film; the ¯ (Py/Cu/TI) can be as described by Equation (4): where / ¯ is the spin-mixing conductance of the Py/Cu interface, is the spin resistance and ⁄ is the spin conductance of Cu/TI. One of the reasons for the lower ¯ (Py/Cu/TI) compared to / ¯ may be due to the fact that the ⁄ is smaller than that of / ¯, similar to the case in Cu/Pt [13,16]. However, since ¯( Py/Cu/TI) ≈ / ¯ at tCu ≥ 9 nm, here we assume that Cu/TI and Py/Cu exhibit similar qualities to / ¯ ≈ ⁄ . Thus, by assuming that the degree of spin absorption at the Cu/TI interfaces is similar in all cases, we suggest that the reason for the lower ¯( Py/Cu/TI) of 3 nm and 7 nm Cu-based trilayer samples may be due to the strong spin accumulation at this ultrathin regime [13]. When the Cu layer is too thin, Py/Cu does not cause significant damping enhancement due to the poor spin sinking of Cu. The spin accumulation in Cu leads to a backflow into Py; therefore, ¯( Py/Cu/TI) is much smaller than / ¯. The