Arxiv:1903.12592V1 [Cond-Mat.Mes-Hall] 29 Mar 2019 Als, May Produce a Giant Interfacial SOC (ISOC) [24–27]

Boundary conditions for spin and charge diﬀusion in the presence of interfacial spin-orbit coupling

J. Borge1, ∗ and I. V. Tokatly1, 2, 3, † 1Nano-Bio Spectroscopy group, Departamento F´ısica de Materiales, Universidad del Pa´ısVasco, Av. Tolosa 72, E-20018 San Sebastián,Spain 2IKERBASQUE, Basque Foundation for Science, E48011 Bilbao, Spain 3Donostia International Physics Center (DIPC), Manuel de Lardizabal 4, E-20018 San Sebastian, Spain Breaking of the inversion symmetry at the interface between different materials may dramatically enhance spin-orbit interaction in the vicinity of the interface. We incorporate the effects of this interfacial spin-orbit coupling (ISOC) into the standard drift-diffusion theory by deriving generalized boundary conditions for diffusion equations. Our theoretical scheme is based on symmetry arguments, providing a natural classification and parametrization of all spin-charge and spin-spin conversion effects that occur due to ISOC at macroscopically isotropic interfaces between nonmagnetic materials. We illustrate our approach with specific examples of spin-charge conversion in hybrid structures. In particular, for a lateral metal-insulator structure we predict an “ISOC-gating” effect which can be used to detect spin currents in metallic films with weak bulk SOC.

Correlations between charge and spin degrees of free- been recognized that the spin-dependent scattering of the dom induced by spin-orbit coupling (SOC) in crystals bulk continuum states at the interface also contributes and nanostructures open a pathway to control spin dy- strongly to the interfacial spin-charge conversion and the namics by purely electric means, without using magnetic spin swapping [34–38]. Currently, theoretical studies of fields. Not surprisingly, spin-charge conversion phenom- spin transport in the presence of ISOC are limited to spe- ena mediated by SOC are attracting a growing atten- cific effects in specific microscopic models with simplest tion in the field of spintronics [1–4]. Among them, the geometries. Apparently this is not sufficient for the de- most known are the spin Hall effect (SHE) [5–7], and scription of realistic device structures, and it is highly the inverse spin-galvanic effect also known as the Edel- desirable to classify all effects of ISOC and consistently stein effect [8, 9] (EE). The SHE universally exists in all incorporate them into a general theoretical scheme of de- conductors without any symmetry restriction, provided vice modeling. SOC is sufficiently strong. In particular, it is responsible The spin and charge transport in a typical spintronics for the spin-charge conversion in the bulk of centrosym- device is usually well described by the drift-diffusion the- metric materials, like Pt or Au [10–12]. In contrast, the ory. Within this approach the evolution of the spin and EE, that is, the spin polarization induced by a charge charge densities is governed by diffusion equations [39], current [6, 7, 13–16], occurs only in the absence of in- supplemented with proper boundary conditions (BC) at version symmetry or, more precisely, only in gyrotropic all interfaces and boundaries. In the absence of SOC the materials/structures [17]. Usually it is discussed for two- BC reduce to the conservation of normal to the inter- dimensional (2D) electron gases in semiconductor het- face components of all currents, and relations between erostructures or in surface bands at surfaces or interfaces the currents and possible discontinuities of the densi- [7, 18–20]. SOC also leads to the spin-spin conversion via ties across the interface. The latter are usually formu- the spin swapping effect (SSE) [21–23]. lated in terms of spin-dependent interface conductances The symmetry conditions for all spin-charge conver- [40–42]. The modifications of BC by the bulk SOC in sion effects are naturally met at interfaces between dif- noncentrosymmetric materials have been intensively de- ferent materials as any interface is always locally gy- bated in the literature [43–47]. However the role of ISOC rotropic. Moreover the strong inversion symmetry break- and the ways of incorporating its effects into the BC for ing across the interface dramatically enhances manifesta- the drift-diffusion theory remain largely unexplored. Re- tions of SOC, and, depending on the nature of the materi- cently a generalization of the magnetoelectronic circuit arXiv:1903.12592v1 [cond-mat.mes-hall] 29 Mar 2019 als, may produce a giant interfacial SOC (ISOC) [24–27]. theory, which partly accounts for the ISOC via coupling This makes interfaces promising candidates for active re- to the in-plane electric field at the interfaces has been gions in spintronics devises, where the spin-charge and proposed [48, 49]. This indeed captures the interfacial spin-spin conversion occur most efficiently. In the last generation of spin current by the in-plane charge current years these effects have been measured using different [33, 50], but apparently it does not cover all physically experimental techniques for various interfaces [28–33]. expected effects of ISOC and the general form of the cor- First experiments on the spin-charge conversion due responding BC still remains unknown. to ISOC were interpreted as the inverse EE (IEE) in the The present paper is aimed at filling this gap by de- 2D Rashba-splitted interface band [20, 28]. Later it has riving the full set of BC describing all possible spin- 2 charge and spin-spin conversion effects that may occur Formally Eqs. (2)-(3) are linear relations between the at the macroscopically isotropic interface separating non- densities and their first derivatives. Such relations are magnetic materials. We do not use any specific micro- forbidden by the symmetry in the isotropic bulk, but scopic model, but rely solely on symmetry arguments, they are allowed at the interface as it provides us with which is similar to the symmetry based derivation of spin an additional polar vector nˆ. By constructing a scalar diffusion equations in the presence of bulk SOC [39, 51]. differential operator nˆ · ∇ we can compile linear relations Let us consider two nonmagnetic materials labeled by involving the densities and their derivatives, and trans- the index α = 1, 2 and separated by a flat interface forming as a scalar, Eq. (2), and a pseudovector, Eq. (3). characterized by unit normal vector nˆ. The interface These are the general BC for the scalar µ(r) and the located at the surface nˆ · r = 0 is assumed macroscopi- pseudovector S(r) densities, allowed by the interface C∞v symmetry under the requirements of the charge and spin cally isotropic with a symmetry group C∞v. In the bulk of the materials the charge and spin degrees of freedom conservation in the absence of the charge-spin mixing. are described, respectively, by the distribution of electro- In the presence of ISOC spin is not conserved and only the charge conservation (the gauge invariance) require- chemical potentials µα(r) and the spin density Sα(r). To focus on the effects of ISOC, we assume that both mate- ment remains. This allows for additional terms in the rials posses the inversion symmetry and the spin-charge BC. Let us consider first the modification of the scalar BC coupling in the bulk is negligible. In this case the charge in Eq. (3). The only additional scalar invariant that is lin- a ear in the densities and their first derivatives is (ˆn×∇)·S. jα and spin Jαi currents are given by the standard dif- D a a Therefore the most general scalar BC takes the form, fusive formulas, jα = −σα ∇µα, and Jαi = −Dα∂iSα, where σD and D are the Drude conductivity and the α α D X sc diffusion coefficient, respectively. In the steady state the σα (ˆn · ∇)µα = G∆µ + θαβDβ(ˆn × ∇) · Sβ. (4) charge and spin diffusion equations on either side of the β interface reduce to Laplace equations for µ (r), and the α Because of the gauge invariance the electrochemical po- stationary spin diffusion equations, tentials enter only as ∆µ, and there is only one charge conductance G. The second term in Eq. (4) describes 2 2 Sα(r) the spin-charge conversion via the interfacial ISHE – the ∇ µα(r) = 0 ; Dα∇ Sα(r) = , (1) τα generation of a normal charge current from in-plane spin currents at either side of the interface. This channel of where τα is the spin relaxation time. In the absence of the spin-charge conversion at hybrid interfaces has been ISOC the BC at the interface are well known and read discussed in Ref. 53 within a simple ballistic scattering model. Our symmetry arguments show that in general it D c is parametrized by four spin-charge Hall angles θsc . The σα (ˆn · ∇)µα = G0∆µ, (2) αβ cross-interface angles θsc and θsc should vanish for non- D (ˆn · ∇)S = Gs∆S, (3) 12 21 α α 0 transparent interfaces. For example, for metal-insulator interfaces there is by only one spin-charge Hall angle. c/s where ∆µ = µ1 − µ2 and ∆S = S1 − S2, and G0 is the Similarly we generalize the pseudovector BC of Eq. (3) charge/spin conductance [52]. Physically Eqs. (2) and by adding all symmetry allowed pseudovectors con- (3) relate the currents passing through the interface to structed from the densities and their derivatives [54]. It the interfacial density/potential drops. The appearance is convenient to write the resulting general BC by sepa- of the differences of the densities in the BC, and the inde- rating the normal and the parallel to the interface spin pendence of conductances on the material index α reflect components S = S⊥ + Sk, where S⊥ = ˆn(ˆn · S) and the conservation of all currents in the absence of SOC. Sk = (ˆn × S) × ˆn,

n n ¯ X n Dα(ˆn · ∇)Sα⊥ = Gα∆S⊥ + LαS⊥ + καβDβ(ˆn × ∇) × Sβk (5) β p p ¯ X p X cs D Dα(ˆn · ∇)Sαk = Gα∆Sk + LαSk + καβDβ(ˆn × ∇) × Sβ⊥ + θαβσβ (ˆn × ∇)µβ, (6) β β

where S¯ = S1 + S2. In the presence of ISOC the spin responding contributions are parametrized by the spin n/p n/p is not conserved. Therefore the right and left values of conductances Gα and the spin loss coefficients Lα , the boundary spin can independently enter BC. The cor- which in general depend on the material index α, and 3 are different for the normal (n) and the parallel (p) spin bulk states also leads to the appearance of interfacial spin components. The third term in the right hand sides in polarization [36] and interfacial currents [37], localized on Eqs. (5) and (6) describes the spin-spin conversion due to the scale of the Fermi wavelength λF . the interfacial SSE. Namely, the in-plane current of the The localized contributions can be included into the parallel (normal) spin component generates the normal drift-diffusion theory by representing the total physical current of the normal (parallel) spin component. This observables in the following form, effect of ISOC is characterized by a set of swapping coef- n/p Otot(r) = Θ(−z)O1(r) + Θ(z)O2(r) + δ(z)OI (rk), (7) ficients καβ . Finally, the last term in the right hand side in Eq. (6) is responsible for the charge-spin coupling. It where z = nˆ · r is the normal to the interface coordinate, can be interpreted as an interfacial SHE – generation of O (r) are the slow “diffusive” parts that satisfy the bulk the spin current across the interface by an in-plane charge α diffusion equations, and O (r ) is the localized part of the current. This effect has been studied recently in Refs. 33 I k observable. Within the standard linear transport theory and 50 for different hybrid structures via first principle the localized spin SI , the charge current jI , and the spin transport calculations. The corresponding transport co- a sc current JiI should be related linearly to the interfacial efficients in Eq. (6) are the charge-spin Hall angles θαβ. values of the diffusive observables µα and Sα. Formally Equations (4)-(6) generalize the standard BC of the latter act as the sources (effective driving fields) for Eqs. (4) and (3). However this is not sufficient to fully the former. The general form of such relations for SI , j , and J a can be determined from the symmetry argu- describe the physics of interfaces with ISOC. The reason I iI is that the diffusion equations and the derived BC in- ments by combining, respectively, all linearly indepen- volve only smooth “diffusive” parts of the densities that dent pseudovector, vector, and pseudotensor invariants vary slowly on the scale of the mean free path `. In constructed out of µα, Sα and their first derivatives [55]. addition, strongly localized (on the scale less than `) in- A straightforward analysis [54] leads to the following ex- terfacial charge and spin currents as well as the interfa- pressions for the localized parts of the spin polarization cial spin polarization in general appear near nontrivial and the charge current, spin-orbit active interfaces. Most obviously the localized X cs X ss SI = σα (ˆn × ∇)µα + σα (ˆn × ∇) × Sα (8) observables can be related to the interface bands, as it is α α commonly assumed to interpret experiments on the in- X sc X sc terfacial spin-charge conversion [19, 20, 28, 31]. Apart jI = σα (ˆn × Sα) + θIα(ˆn × ∇)(ˆn · Sα), (9) from that, in the presence of ISOC the spin-dependent α α interference between the incident and reflected waves for while the localized spin current takes the form,

a cs X h p i n cs i 0 i JiI = g iaknˆk∆µ + gαnˆaSα + gαδai n · Sα + θIαnˆa(ˆn × ∇)iµα + κIα(ˆn × (ˆn × ∇))aSα + κIαδai∇ · Sαk (10) α

In the last equation the spatial index i takes only in- in-plane spin current. Finally, in Eq. (10) the first term a plane values as the interface spin current JiI flows in the can be interpreted as a cross-interface SHE (the spin cur- interface plane [56]. For brevity we do not show the “triv- rent at the interface plane generated by the voltage drop a ial” terms proportional to Sα, ∇µα, and ∂iSα, allowed across the interface), the fourth term is the 2D interfa- in Eqs. (8), (9), and (10), respectively. These terms may cial SHE, the last two terms describe the 2D SSE, while describe, if necessary, the usual 2D diffisive transport in the second and third terms are responsible for the spin the interface bands. The contributions shown explicitly current produced directly by the non-equilibrium spin are those responsible for the spin-charge and the spin- polarization. spin conversion. The first term in Eq. (8) describes the interfacial EE – the local spin polarization induced by the The localized observables of Eqs. (8)-(10) were intro- in-plane charge current [36]. The second term is the induced on physical grounds. Now we show that the ap- terfacial spin generated by the spin current flowing along pearance of in-plane localized currents is also required the interface, and polarized in the direction orthogonal by the internal consistency of the theory. Let us look to that of the current. The first and the second terms on Eq. (4). The second term in the right hand side is in Eq. (9) correspond, respectively, to the interfacial IEE allowed by the symmetry and meaningful physically, but and the ISHE, i. e., the charge current at the interface it manifestly violates conservation of the charge current. induced by the non-equilibrium spin polarization and the Indeed Eq. (4) states that a part of the charge current passing through the interface is lost in the presence of 4

y FIG. 1. Streamlines of the charge current generated by the FIG. 2. Charge flow generated by the spin current Jx (x) in x spin current Jz flowing perpendicular to the interface (at z = the metallic film with a insulator (shown in gray) deposited on 0) between two conducting materials. The density plot shows its top surface. The density plot shows the current strength. the distribution of the induced electrostatic potential.

the bulk we have to solve the Laplace equation, ∇2µ = 0, in-plane spin gradients. The interfacial charge current with the condition of vanishing normal component of the jI ﬁxes this problem by providing a missing sink. In the total current at the sample boundaries at y = ±W/2 presence of j the continuity equation for the total charge I D current, after the integration across the interface, reads, − σ ∂yµ(y, z)|y=±W/2 + jyI δ(z) = 0. (13)

D D By solving this problem analytically [54] we find the spa- σ (nˆ · ∇)µ1 − σ (nˆ · ∇)µ2 = −∇ · jI . (11) 1 2 tial distribution of the charge current, and the total volt- By substituting Eqs. (4) and (9) into the left and right age drop across the sample hand sides we find that the charge continuity equation is Z W fulfilled identically if the spin-charge Hall angles θsc are αβ ∆V = µ(W/2, z) − µ(−W/2, z) dz = jyI D , (14) sc σ related to the “Edelstein conductivity” σα as follows, sc sc sc The stream lines of the induced charge flow together with σα = Dα(θ1α − θ2α). (12) the density plot for the potential are shown in Fig. 1. The Therefore there is a deep connection between the inverse current in the bulk forms a counterflow that compensates SHE described by Eq. (4) and the generation of the local the localized currents generated at the interface. Both charge current via the inverse EE in Eq. (9). Inclusion of the induced potential and the current are concentrated one effect necessarily implies the presence of the other. near the edges of the interface at a macroscopic scale of The BC Eqs. (4)-(6) together with Eqs. (8)-(10) com- the order of the sample size W . plement the standard bulk drift-diffusion equations to As a second example we consider a spin-charge con- model spintronics devices of any experimentally relevant version in a lateral hybrid structure made from a metal- geometry. It is worth noting that technically the local- lic film of thickness W , with a part of its upper surface ized currents become important in non-1D geometries covered by an insulator with large SOC, like Bi2O3, see with interfaces of a finite size. At the edges of the inter- Fig. 2. In this way we create an interface with ISOC face the total currents should be conserved, and therefore on the top boundary at z = W for x > 0, while the rest the edges act as local sources and sinks, which generates (x < 0) of the top boundary as well as the bottom bound- nontrival patterns of the charge and spin flows. The ex- ary at z = 0 remain “trivial”. We assume a diffusive amples below illustrate this point and demonstrate our spin current polarized along y, flowing in the x-direction, y y y −x/ls general phenomenological construction at work. that√ is, Jx (x) = −D∂xS (x) with S (x) ∼ e , where The first example models the spin-charge conversion ls = Dτs is the spin diffusion length. at the interface [28]. We consider a conducting bilayer of The induced potential µ(x, z), is obtained by solving a finite width W in the y-direction and separated by the the Laplace equation with two BC. On the bottom sur- D interface with ISOC at z = 0 plane, as shown in Fig. 1. A face the standard BC of σ ∂zµ|z=0 = 0 is imposed. To x x get the BC on the top surface we combine Eqs. (4) and spin current Jz (z) = −D∂zS (z), polarized along x-axis, and injected from the left, flows in z-direction, crosses the (9) in form of Eq. (11) that for the metal-insulator inter- interface, and determines via Eq. (6) the spin polarization face and the chosen spin density reads, Sx(0) at the interface. The latter, in turn, generates a D sc y σ ∂zµ|z=W = −D∂x θ (x)S (x) , (15) localized charge current in the y-direction via the IEE in sc x cs sc Eq. (9), jyI = σ S (0). To determine the distribution of were θ (x) = θ Θ(x) reflects the stepwise distribution the potential µ(r) and the charge current j = −σD∇µ in of ISOC at the top surface [57]. This problem is also 5 solvable analytically [54]. The corresponding charge flow, [3] H.-A. Engel, E. I. Rashba, and B. I. Halperin, in Handbook shown in Fig. 2, demonstrates a typical dipolar pattern of Magnetism and Advanced Magnetic Materials, Vol. V, with a local sink at the edge of the interface a distributed edited by H. Kronmüllerand S. Parkin (Wiley, Chichester, UK, 2007) pp. 2858–2877. ∼ Θ(x)e−x/ls source. As the film has a finite width this [4] G. Vignale, J. Supercond. Nov. Magn. 23, 3 (2010). dipole filed generates a lateral voltage drop: [5] M. I. Dyakonov and V. I. Perel, Phys. Lett. A 35, 459 (1971). y sc ∆V = µ(∞, z) − µ(−∞, z) = χS (0)θ ls/W, (16) [6] Y. K. Kato, R. C. Myers, A. C. Gossard, and D. D. Awschalom, Science 306, 1910 (2004). where χ = D/σD is the inverse compressibility of the [7] V. Sih, R. C. Myers, Y. K. Kato, W. H. Lau, A. C. Gos- metal. If the top “ISOC gate” has a finite length L the sard, and D. D. Awschalom, Nature Physics 1, 31 (2005). [8] Y. Lyanda-Geller and A. Aronov, JETP Lett. 50, 431 voltage drop acquires an additional factor 1−e−L/ls . No- (1989). tice that by measuring the induced lateral voltage, and [9] V. Edelstein, Solid State Communications 73, 233 (1990). using Eq. (16) we get a direct experimental access to the [10] S. Valenzuela and M. Tinkham, Nature 442, 176 (2006). sc interfacial spin-charge Hall angle θ . [11] T. Kimura, Y. Otani, T. Sato, S. Takahashi, and In conclusion, we derived a full set of additional con- S. Maekawa, Phys. Rev. Lett. 98, 156601 (2007). ditions that complement the standard drift-diffusion the- [12] T. Seki, Y. Hasegawa, S. Mitani, S. Takahashi, H. Ima- ory to model spin and charge dynamics in the presence mura, S. Maekawa, J. Nitta, and K. Takanashi, Nat Mater 7, 125 (2008). of interfaces with strong ISOC. These conditions con- [13] S. D. Ganichev, E. L. Ivchenko, V. V. Bel’kov, S. A. sist of the generalized BC describing the interfacial spin- Tarasenko, M. Sollinger, D. Weiss, W. Wegscheider, and charge and spin-spin conversion, and the expressions for W. Prettl, Nature 417, 153 (2002). the spin, the charge current, and the spin current, local- [14] I. Mihai Miron, G. Gaudin, S. Auffret, B. Rodmacq, ized at the interface within a microscopic scale (smaller A. Schuhl, S. Pizzini, J. Vogel, and P. Gambardella, Na- than `). Our construction provides a natural classifica- ture Materials 9, 230 (2010). tion and parametrization of all spin-charge and spin-spin [15] J.-i. Inoue, G. E. W. Bauer, and L. W. Molenkamp, Phys. Rev. B 67, 033104 (2003). conversion effects mediated by ISOC at macroscopically [16] C. L. Yang, H. T. He, L. Ding, L. J. Cui, Y. P. Zeng, isotropic interfaces between nonmagnetic materials. The J. N. Wang, and W. K. Ge, Phys. Rev. Lett. 186605, 96 phenomenological coefficients entering the derived BC (2006). should be determined from comparison with experiments [17] S. D. Ganichev, M. Trushin, and J. Schliemann, in Hand- or first principle calculations for specially chosen geome- book of Spin Transport and Magnetism, edited by E. Tsym- tries. To demonstrate the working power of our theory bal and I. Zutic (Chapman and Hall, 2011) Chap. 24, pp. we considered two specific examples. In particular, we 487–497. [18] B. M. Norman, C. J. Trowbridge, D. D. Awschalom, and predict a generation of a lateral voltage drop in a metal- V. Sih, Phys. Rev. Lett. 112, 056601 (2014). lic film by a spin current if an insulator with a strong [19] G. Seibold, S. Caprara, M. Grilli, and R. Raimondi, SOC is deposited on the top surface of the film. This Phys. Rev. Lett. 119, 256801 (2017). “ISOC gate” effect can be used to detect spin currents in [20] K. Shen, G. Vignale, and R. Raimondi, Phys. Rev. Lett materials with weak bulk SOC. 112, 096601 (2014). Acknowledgments. We acknowledge financial sup- [21] M. B. Lifshits and M. I. Dyakonov, Phys. Rev. Lett. 103, port from the European Research Council (ERC-2015- 186601 (2009). [22] K. Shen, R. Raimondi, and G. Vignale, Phys. Rev. B AdG-694097), Spanish grant (FIS2016-79464-P), Grupos 90, 245302 (2014). Consolidados (IT578-13), AFOSR Grant No. FA2386- [23] H. B. M. Saidaoui and A. Manchon, Phys. Rev. Lett. 15-1-0006 AOARD 144088, H2020-NMP-2014 project 117, 036601 (2016). MOSTOPHOS (GA No. 646259) and COST Action [24] C. R. Ast, J. Henk, A. Ernst, L. Moreschini, M. C. Falub, MP1306 (EUSpec). J.B. acknowledges funding from D. Pacilé, P. Bruno, K. Kern, and M. Grioni, Phys. Rev. the European Unions Horizon 2020 research and inno- Lett. 98, 186807 (2007). vation programme under the Marie Sklodowska-Curie [25] S. Mathias, A. Ruffing, F. Deicke, M. Wiesenmayer, I. Sakar, G. Bihlmayer, E. V. Chulkov, Y. M. Koroteev, Grant Agreement No. 703195. J.B. acknowledges Prof. P. M. Echenique, M. Bauer, and M. Aeschlimann, Phys. Roberto Raimondi for stimulating conversations. Rev. Lett. 104, 066802 (2010). [26] A. G. Rybkin, A. M. Shikin, V. K. Adamchuk, D. Marchenko, C. Biswas, A. Varykhalov, and O. Rader, Phys. Rev. B 82, 233403 (2010). [27] L. Moreschini, A. Bendounan, H. Bentmann, M. Assig, ∗ [email protected] K. Kern, F. Reinert, J. Henk, C. R. Ast, and M. Grioni, † [email protected] Phys. Rev. B 80, 035438 (2009). [1]I. Zutić,J.ˇ Fabian, and S. Das Sarma, Rev. Mod. Phys. [28] J. R. Sánchez, L. Vila, G. Desfonds, S. Gambarelli, J. At- 76, 323 (2004). tané,J. D. Teresa, C. Magén, and A. Fert, Nat. Commun. [2] J. Sinova, S. O. Valenzuela, J. Wunderlich, C. H. Back, 4, 2944 (2013). and T. Jungwirth, Rev. Mod. Phys. 87, 1213 (2015). [29] M. Isasa, M. C. Mart´ınez-Velarte, E. Villamor, C. Magén, 6

L. Morellón, J. M. De Teresa, M. R. Ibarra, G. Vig- [57] To arrive at this form of BC we have used the relation nale, E. V. Chulkov, E. E. Krasovskii, L. E. Hueso, and σsc = Dθsc which follows from Eq. (12) for the case of a F. Casanova, Phys. Rev. B 93, 014420 (2016). metal-insulator interface. [30] S. Karube, H. Idzuchi, K. Kondou, Y. Fukuma, and Y. Otani, Applied Physics Letters 107, 122406 (2015). [31] S. Karube, K. Kondou, and Y. Otani, Applied Physics Express 9, 033001 (2016). [32] I. M. Miron, K. Garello, G. Gaudin, P.-J. Zermatten, M. V. Costache, S. Auffret, S. Bandiera, B. Rodmacq, A. Schuhl, and P. Gambardella, Nature (London) 476, 189 (2011). [33] S.-H. C. Baek, V. P. Amin, Y.-W. Oh, G. Go, S.-J. Lee, G.-H. Lee, K.-J. Kim, M. D. Stiles, B.-G. Park, and K.-J. Lee, Nature Materials 17, 509513 (2018). [34] X. Wang, J. Xiao, A. Manchon, and S. Maekawa, Phys. Rev. B 87, 081407 (2013). [35] J. Borge, C. Gorini, G. Vignale, and R. Raimondi, Phys. Rev. B 89, 245443 (2014). [36] I. V. Tokatly, E. E. Krasovskii, and G. Vignale, Phys. Rev. B 91, 035403 (2015). [37] J. Borge and I. V. Tokatly, Phys. Rev. B 96, 115445 (2017). [38] S. Li, K. Shen, and K. Xia, arXiv:1807.03433v1. [39] M. I. Dyakonov and A. V. Khaetskii, “Spin physics in semiconductors,” (Springer, Berlin, 2017) Chap. 8, pp. 211–244, 2nd ed. [40] D. Huertas Hernando, Y. V. Nazarov, A. Brataas, and G. E. W. Bauer, Phys. Rev. B 62, 5700 (2000). [41] A. Brataas, Y. Nazarov, and G. Bauer, The European Physical Journal B - Condensed Matter and Complex Sys- tems 22, 99 (2001). [42] A. Brataas, G. E. Bauer, and P. J. Kelly, Physics Reports 427, 157 (2006). [43] A. G. Mal’shukov, L. Y. Wang, C. S. Chu, and K. A. Chao, Phys. Rev. Lett. 95, 146601 (2005). [44] I. Adagideli and G. E. W. Bauer, Phys. Rev. Lett. 95, 256602 (2005). [45] V. M. Galitski, A. A. Burkov, and S. Das Sarma, Phys. Rev. B 74, 115331 (2006). [46] O. Bleibaum, Phys. Rev. B 74, 113309 (2006). [47] Y. Tserkovnyak, B. I. Halperin, A. A. Kovalev, and A. Brataas, Phys. Rev. B 76, 085319 (2007). [48] V. P. Amin and M. D. Stiles, Phys. Rev. B 94, 104419 (2016). [49] V. P. Amin and M. D. Stiles, Phys. Rev. B 94, 104420 (2016). [50] V. P. Amin, J. Zemen, and M. D. Stiles, Phys. Rev. Lett. 121, 136805 (2018). [51] M. I. Dyakonov, Phys. Rev. Lett. 99, 126601 (2007). [52] All quantities entering the BC and labeled by the material index α = 1 and 2 are evaluated at a given point of the interface as the right and left limits, respectively, i. e. nˆ · r = ±0. Throughout this paper for brevity we omit the spatial arguments in the BC. [53] J. Linder and T. Yokoyama, Phys. Rev. Lett. 106, 237201 (2011). [54] See Supplementary Material for details. [55] The allowed by symmetry terms involving normal derivatives can always be eliminated using the BC of Eqs. (4)- (6), and therefore they do not appear in the expression for the localized observables. [56] More formally this condition is expressed in terms of the a a interface projection JiI 7→ (δik − nînˆk)JkI , which is im- plicitly assumed in Eq. (10). 1

Supplementary Material for “Boundary conditions for spin and charge diﬀusion in the presence of interfacial spin-orbit coupling”

CONSTRUCTION OF INTERFACE TENSOR INVARIANTS

In this Section we construct and classify all possible scalars, vectors, pseudovectors and second rank pseudotensors involved in the boundary conditions (BC) of Eqs. (4)-(6) and in the interfacial observables of Eqs. (8)-(10) described in the main text.

Boundary conditions

As it is explained in the main text of the manuscript, the BC can be viewed as linear relations between the interface a values of the electrochemical potentials µα, the components of the spin density Sα, and their first derivatives, ∂iµα and a ∂iSα. In the bulk the charge and spin dynamics are described by one scalar (charge diffusion) and one pseudovector (spin diffusion) second order differential equations. Therefore we need two scalar and two pseudovector BC at each interface. Formally the problem of finding a general form of BC reduces to finding a general linear function of scalar a a µ, pseudovector S , vector ∂iµ, and pseudotensor ∂iSα variables, which (i) transforms as a scalar or a pseudovector, (ii) is gauge invariant (i. e. invariant under a global shift of the electrochemical potential), and (iii) compatible with the interface symmetry. The gauge invariance implies that the electrochemical potentials themselves can enter only as difference ∆µ = µ1 − µ2. The compatibility with the interface symmetry formally means that the tensor coefficients in the above linear function should be invariant under operations of the interface symmetry group C∞v. Technically these invariant tensor coefficients can be constructed from all possible products/contractions of the vector normal to the interface nk, the second-rank tensor δij, and the third-rank pseudotensor ijk, which are the “natural” available objects preserving the required symmetry. Let us start with the scalar BC. It obviously involves the genuine scalar ∆µ. We can also generate a scalar by contracting the vector ∂jµ with available C∞v invariant vector Aj = Anj, and by contracting the second rank b b pseudotensor ∂jS with the invariant second rank pseudotensor αj = αjbknk. However, there is no way to construct b a pseudovector from the available objects (nk, δij, ijk) in order to contract it with the pseudovector S . Hence the symmetry does not allow to make a scalar out of the spin density. To summarize, all possible scalar invariants are

b b ∆µ; Aj∂jµ; αj∂jS , (S1)

b where the symmetry allowed tensor coeﬃcients, Aj and αj, read

Aj = Anj (S2) b αj = αjbknk (S3) (here and in the following we use latin letters for tensors and greek letters for pseudotensors). Taking this into account and using the vector notations we can write the most general scalar BC as follows,

D X sc σα (ˆn · ∇)µα = G∆µ + θαβDβ(ˆn × ∇) · Sβ. (S4) β

The indexes α, β indicate the side of the interface, and thus all together we have two scalar BC, as required. This equation corresponds to the Eq. (4) of the main text. The pseudovector BC involves the spin pseudovector Sa along with other pseudovector invariants that we describe a below. (i) We can make a pseudovector by contracting the vector ∂jµ a second rank pseudotensor βj . (ii) We may ab b ab also contract a second rank tensor B with the pseudovector S . (iii) The contraction of a third rank tensor Cj b with the second rank pseudotensor ∂jS will also generate a pseudovector. On the other hand, there is no way to construct a pseudovector in order to connect it with the scalar ∆µ. All the possible pseudovectors are

a a ab b ab b S ; βj ∂jµ; B S ; Cj ∂jS (S5) 2 where the tensor coefficients invariant under the operations of C∞v take the following general forms a βj = βjaknk (S6) ab B = B1(δab − nanb) + B2nanb (S7) ab Cj = C1njnanb + C2nj(δab − nanb) + C3na(δjb − njnb) + C4nb(δja − njna). (S8) Notice that two terms in Eq. (S7) correspond to the in-plane and out-of-plane projectors. As these terms can enter with different coefficients, B1 and B2, the spin densities Sα⊥ = ˆn(ˆn · Sα) and Sα|| = (ˆn × Sα) × ˆn may have different couplings. Because of this it is natural to split the pseudovector BC into two sets of equations for the normal and parallel to the interface components of the spin,

n n X n Dα(ˆn · ∇)Sα⊥ = Gα∆S⊥ + LαS⊥ + καβDβ(ˆn × ∇) × Sβ|| (S9) β p p X p X cs D Dα(ˆn · ∇)Sα|| = Gα∆S|| + LαS|| + καβDβ(ˆn × ∇) × Sβ⊥ + θαβσβ (ˆn × ∇)µβ, (S10) β β which corresponds to Eqs. (5) and (6) of the main text.

Localized interface observables

Now we describe the construction of the interfacial observables introduced in the main paper. There are three possible physical objects which can be generated at the interface: the interfacial spin density, the interfacial charge current, and the interfacial spin current. The general form of linear relations of these localized observables to the “diﬀusive” a a variables µα, Sα, and their ﬁrst derivatives, ∂iµα and ∂iSα, can be determined from the symmetry arguments. a The interfacial spin density SI is a pseudovector and all possible pseudovector invariant have been already described above in Eqs. (S5)-(S8). The interfacial charge current jIi is a vector and therefore we need to construct all allowed vector invariants. A vector can be generated by (i) multiplying the scalar ∆µ with a vector Di, (ii) contracting a second rank tensor Eij b b with the vector ∂jµ, (iii) contracting a second rank pseudotensor γi with the pseudovector S , and (iv) contracting a b b third rank pseudotensor ηij with the second rank pseudotensor ∂jS :

b b b b Di∆µ; Eij∂jµ; γi S ; ηij∂jS , (S11) where the allowed tensor coeﬃcient read

Di = Dni (S12)

Eij = E1(δij − ninj) + E2ninj (S13) b γi = γibknk (S14) b ηij = η1ijknknb + η2jbknkni + η3ibknknj. (S15) a To ﬁnd the form of the interfacial spin current JIi, which is a second rank pseudotensor, we generate all allowed second rank pseudotensors. Speciﬁcally we can construct a second rank pseudotensor (i) from the scalar ∆µ through a a a second rank pseudotensor ζi , (ii) by contracting a third rank pseudotensor, ξij, with the vector ∂jµ, (iii) by contracting ab b ab a third rank tensor Fi with the pseudovector S , and (iv) by contracting a fours rank tensor Gij with the second b rank pseudotensor ∂jS :

a a ab b ab b ζi ∆µ; ξij∂jµ; Fi S ; Gij ∂jS (S16) where the allowed tensor coeﬃcients have the following general form, a ζi = ζiaknk (S17) a ξij = ξ1ijknkna + ξ2jaknkni + ξ3iaknknj (S18) ab Fi = F1ninanb + F2ni(δab − nanb) + F3na(δib − ninb) + F4nb(δia − nina) (S19) ab Gij = G1ninjnanb + G2ninj(δab − nanb) + G3(δij − ninj)nanb + G4(δij − ninj)(δab − nanb) (S20)

+ G5ninb(δja − njna) + G6(δib − ninb)njna + G7(δib − ninb)(δja − njna)

+ G8nina(δjb − njnb) + G9(δia − nina)njnb + G10(δia − nina)(δjb − njnb). 3

One can verify that the (pseudo)tensors appearing in each of the above equations with different coefficients form a complete set of linearly independent (pseudo)tensors of a given rank and compatible with the required symmetry. For example, we may also construct fours rank tensors by contracting two different third rank pseudotensors or combining them with two vectors. However, these fours rank tensors can be written as a linear combination of the terms present ab in Gij of Eq. (S20):

ijlabl = δiaδjb − δibδja (S21)

ijlabknlnk = δiaδjb − δibδja − δianjnb − ninaδjb + δibnjna + ninbδja. (S22) Before writing the interfacial observables in a more compact way we should take into account two facts. Firstly, the index i in the current observables indicates the direction of the localized interfacial charge and spin currents. Since these currents may flow only in the interface plane we should exclude (project out) all the invariants implying the i-direction to be parallel to nˆ. Secondly, the index j indicates the direction of the diffusive charge and spin currents (the direction of the spatial derivatives of the diffusive observables). The terms corresponding to the j-direction parallel to nˆ should also be excluded as the normal derivatives (∇· ˆn)µα and (∇· ˆn)Sα can always be eliminated using the BC of Eqs. (S4, S9, S10). In particular this implies that from 10 terms in Eq. (S20) only 4 can appear in the expression for interface spin current, namely those with the coefficients G3, G4, G7, and G8. This leads to 4 possible contributions to the spin-spin conversion in Eq. (S26) below. Taking into account all above arguments we can finally write the following expressions for the interface observables:

X || X cs X ss SI|| = χI,αSα|| + σα (ˆn × ∇)µα + σα (ˆn × ∇) × Sα⊥ (S23) α α α X ⊥ X ss0 SI⊥ = χI,αSα⊥ + σα (ˆn × ∇) × Sα|| (S24) α α X D X sc X sc jI = σI,αˆn × (ˆn × ∇)µα + σα (ˆn × Sα) + θIα(ˆn × ∇)(ˆn · Sα), (S25) α α α and a X h || a ⊥ a i cs JiI = DI,α∂iS||α + DI,α∂iS⊥α + g iaknˆk∆µ α X h p i n cs i 0 i + gαnâSα + gαδai n · Sα + θIαnâ(ˆn × ∇)iµα + κIα(ˆn × (ˆn × ∇))aSα + κIαδai∇ · Sαk (S26) α These equations correspond to Eqs. (8)-(10) of the main. Notice that for brevity in the main text we did not show the “trivial” contributions described by the first terms in Eqs. (S23)-(S26), but kept only the relevant terms responcible for the spin-charge and spin-spin conversion.

SPIN-TO-CHARGE CONVERSION: TRANSPORT THROUGH THE INTERFACE

In this Section we present a solution of the boundary value problem for the spin-charge conversion in a metallic x x bilayer shown on Fig. 1 of the main text. As we explain in the main text, when a spin current Jz (z) = −D∂zS (z), polarized along x-axis, flowing in z-direction crosses and interface at z = 0 it generates a localized charge current in sc x the y-direction via the IEE jyI = σ S (0), (see Eq. (9) of the main text). We want to describe the profile of the electrochemical potential and of the charge current in the case of a finite sample of width W in the y-direction (see Fig. 1 of the main text). To do so, we have to solve the Laplace equation with the BC of vanishing total current at the boundaries of the sample y = ±W/2 ∇2µ = 0 (S27) D −σ ∂yµ(y, z)|y=±W/2 + jyI δ(z) = 0. (S28) By performing the Fourier transform with respect to the z-variable we obtain 2 2 ∂y µ(y, q) − q µ(y, q) = 0 (S29) D −σ ∂yµ(y, q)|y=±W/2 + jyI = 0, (S30) which has the following solution

µ(y, q) = C1 cosh(qy) + C2 sinh(qy), (S31) 4 with the coeﬃcients determined from the BC of Eq. (S30)

jyI C1 = 0 C2 = qW . (S32) σDq cosh( 2 ) The solution for the electrochemical potential is given by the inverse Fourie transform,

Z dq j sinh(qy) µ(y, z) = eiqz yI , (S33) 2π qW σDq cosh( 2 ) which after performing the integral using the residue theorem can be written as

2(nπ+ π )z 2(nπ+ π )z  2 π 2 π  ∞ n − 2(nπ+ 2 )y −1 n − 2(nπ+ 2 )y jyI X (−1) e W sin( ) X (−1) e W sin( ) µ(y, z) = Θ(z) W + Θ(−z) W . (S34) σD  2(nπ + π ) 2(nπ + π )  n=0 2 n=−∞ 2

After summing the series we ﬁnd the following compact representation for the electrochemical potential

jyI h π(−|z|+iy) i µ(y, z) = Im arctan(e W ) . (S35) σDπ The charge current consists of two parts, the diﬀusive part in the bulk of the metals and the interfacial one,

D jy = −σ ∂yµ(y, z) + jyI (y, z) (S36) D jz = −σ ∂zµ(y, z), (S37) which can be written as follows.   σD 1 jy = Re   jyI + jyI δ(z) (S38) W π(z−iy) cosh W   σD 1 jz = Im   jyI . (S39) W π(z−iy) cosh W

Finally, by using the integral representation (S33) we can calculate the total voltage drop across the sample:

Z W ∆V = µ(W/2, z) − µ(−W/2, z)dz = j . (S40) yI σD

SPIN-TO-CHARGE CONVERSION: TRANSPORT PARALLEL TO THE INTERFACE

Here we present a detailed analysis of our second example: the spin-charge conversion in a lateral hybrid structure made from a metallic film of thickness W , with part of its upper surface covered by an insulator with large SOC. This insulator generates an interface with ISOC on the top boundary at z = W for x > 0, while the rest (x < 0) of the top boundary as well as the bottom boundary at z = 0 remain “trivial” (see Fig. 2 in the main text). We assume a diffusive y y y −x/ls spin√ current polarized along y, flowing in the x-direction, that is, Jx (x) = −D∂xS (x) with S (x) = S0e , where ls = Dτs is the spin diffusion length of the normal metal. In order to calculate the induced electrochemical potential we have to solve the Laplace equation with two BC. One of them is the trivial condition of vanishing normal charge current at the bottom boundary, while in order to calculate the BC for the upper interface we have to combine the different BC for the diffusive and interfacial quantities (see Eqs. (4), (9), and (11) of the main text). The Laplace equation with the BC read,

∇2µ = 0 (S41) D σ ∂zµ(x, z)|z=0 = 0 (S42) D sc y σ ∂zµ(x, z)|z=W = −D∂x θ Θ(x)S (x) , (S43) 5

0.10

0.08

0.06

0.04

0.02

x -2 2 4 6 W

FIG. S1. Proﬁle of the potential µ(x, 0) at z = 0 as a function of x for ls = 0.1W .

By performing the Fourier transform with respect to the x-variable we obtain the following 1D problem

2 2 ∂z µ(q, z) − q µ(q, z) = 0 (S44) D σ ∂zµ(q, z)|z=0 = 0 (S45) θscS q2 σD∂ µ(q, z)| = D 0 , (S46) z z=W −1 + (q − ils )(q − i0 ) where inﬁnitesimal imaginary shift i0+ comes from the Θ(x) function which forces the pole at q = 0 to be evaluated at the “positive side” of the real axis. The solution for the Fourier component of the electrochemical potential takes the following form

χθscS q cosh(qz) µ(q, z) = − 0 , (S47) −1 + (q − ils )(q − i0 ) sinh(qW ) which in the real space reads,

Z dq χθscS q cosh(qz) µ(x, z) = eiqx 0 , (S48) + −1 2π (q − i0 )(q − ils ) sinh(qW )

After performing the integral using the residue theorem we ﬁnd the following series representation for the electrochemical potential,

x ( " − z ∞ n − nπx nπz # −1 n − nπx nπz ) e ls cos( ) (−1) e W cos( ) (−1) e W cos( ) sc ls X W X W µ(x, z) = χθ S0 Θ(x) + + Θ(−z) . (S49) sin( W ) nπ − W nπ − W ls n=0 ls n=−∞ ls

This series can be expressed in terms of the the Lerch transcendent function [S1],

∞ X zn Φ(z, s, a) = , (S50) (n + a)s n=0 as follows

( " − x z # ls e cos( l ) 1 π(x+iz) W π(−x+iz) W sc s − W W µ(x, z) = χθ S0 Θ(x) W + Φ −e , 1, − + Φ −e , 1, − sin( ) 2π lsπ lsπ ls 1 π(x+iz) π(x+iz) W π(x−iz) − π(x−iz) W − Θ(−z) e W Φ −e W , 1, − + e W Φ −e W , 1, − , (S51) 2π lsπ lsπ which is our ﬁnal results for the electrochemical potential in the lateral structure. 6

μ 0.10

0.08

0.06

0.04

0.02

x -2 2 4 6 W

L FIG. S2. Proﬁle of the potential µ (x, 0) at z = 0 as a function of x for ls = 0.1W for diﬀerent values of L, L = 5ls (blue), L = ls (yellow) and L = 0.5ls (green).

In Fig. S1 we show the potential µ(x, 0) at z = 0 as a function of x. We observe that the longitudinal voltage drop is formed of the scale of the order of the sample width W . To ﬁnd the value of this voltage drop explicitly we the asymptotic behavior of Eq. (S48) for |x| >> W . In this limit the electrochemical potential can be written as follows

Z dq χθscS l µ(x, z) ≈ eiqx 0 = χθscS s Θ(x), (S52) + −1 0 2π (q − i0 )ils W W so the total voltage drop reads, l ∆V = µ(∞, z) − µ(−∞, z) = χθscS s . (S53) 0 W

Lateral structure with ISOC-gate of a ﬁnite length

Finally, we analyze the same problem for the case when the “ISOC-gate” has a finite length L. If the ISOC acts within a finite region 0 < x < L on the top surface the corresponding BC is modified as follows

D L sc y σ ∂zµ |z=W = −D∂x θ [Θ(x) − Θ(x − L)]S (x) . (S54)

In this case the ﬁnal solution for the electrochemical potential takes the form

L ∞ − L ∞ µ (x, z) = µ (x, z) − e ls µ (x − L, z), (S55) where µ∞(x, z) is the potential for L → ∞, given by Eq. (S51). In Fig. S2 we show the potential µL(x, 0) at z = 0 as a function of x for ls = 0.1W for diﬀerent values of L. For the corresponding lateral voltage drop we ﬁnd,

L ∞ − L ∆V = ∆V (1 − e ls ), (S56) where ∆V ∞ is defined after Eq. (S53). y Finally in Fig. S3 we show the profile of the charge flow generated by the spin current Jx (x) in the metallic film with an insulator of length L, where we have chosen L = ls = 0.5W .

∗ [email protected] † [email protected] [S1] A. Erd´elyi,W. Magnus, F. Oberhettinger, and F. Tricomi, Higher Transcendental Function (McGraw-Hill, New York, NY, 1953), vol. V. 7

y FIG. S3. Charge ﬂow generated by the spin current Jx (x) in the metallic ﬁlm with a insulator of length L (shown in gray) deposited on its top surface. The density plot shows the current strength.