Optical spin-orbit torque in Rashba ferromagnets Junwen Li1,2 and Paul M. Haney1

arXiv:1701.03495v1 [cond-mat.mes-hall] 12 Jan 2017

1. Center for Nanoscale Science and Technology, National Institute of Standards and Technology, Gaithersburg, MD 20899 2. Maryland NanoCenter, University of Maryland, College Park, MD 20742, USA We study the optically induced torques on thin film ferromagnetic layers under excitation by circularly polarized light. We include Rashba spin-orbit coupling and assume an out-of-plane magnetization, and consider incident light with an in-plane component of angular momentum (e.g. light with an oblique angle of incidence). Solving the semiconductor Bloch equations, we find simple expressions for the torque per optical absorption rate in the limits of weak and strong spin-orbit coupling. For weak spin-orbit coupling, the torque is bounded by the optical angular momentum injection rate. For strong spin-orbit coupling, the torque can exceed the angular momentum injection rate by a factor ∆τ /¯ h, where ∆ is the magnetic exchange splitting and τ is the carrier scattering time. I.

INTRODUCTION

The interaction between light and magnetism is of fundamental and technological interest [1]. There are several mechanisms underlying optical control of magnetism. Among these include a range of thermal and quantum mechanical effects which lead to ultrafast demagnetization [2–5]. Light absorption also modifies the electron distribution function, which can change the magnetic anisotropy and lead to magnetic dynamics [6–8]. Another effect is optical spin transfer torque from absorption of circularly polarized light. Optical spin transfer torque operates on the same principle as electrical current-induced spin transfer torque [9–11]: In both cases conservation of total spin angular momentum implies that a net flux of angular momentum flow into a ferromagnet results in a torque on the magnetization [12–14]. In this simple picture of optical spin transfer torque, spinorbit coupling is considered to be negligible so that total spin is conserved. In instances where spin-orbit coupling is not negligible, spin conservation no longer applies and an excitation may induce a torque on the magnetization in which the angular momentum is supplied by the lattice [15]. The angular momentum transfer is mediated by spin-orbit coupling, and the resulting torque is known as spin-orbit torque. Spin-orbit torques have been realized and extensively studied using DC electrical excitation of heavy metal-ferromagnet bilayers [16–19], and may have advantages over spin transfer torque in terms of the efficiency of magnetic switching [18]. Here we consider the optical spin-transfer and spinorbit torques which are present in a thin film ferromagnet with Rashba spin-orbit coupling. Similar systems have been considered in recent works: Ref. [20] computes the optically induced torques for circularly polarized light with polarization perpendicular to the magnetization of a bulk material, while Ref. [21] considers the optically induced effective magnetic fields in a thin film Rashba ferromagnet with magnetization parallel to the optical circular polarization. The geometry considered in this work is shown in Fig. 1: the magnetization is aligned with the symmetry breaking direction and we consider incident

light with an in-plane component of circular polarization (e.g. with an oblique angle of incidence). The ferromagnetism introduces a spin-dependent exchange splitting of magnitude ∆, while the spin-orbit coupling acts as a kdependent effective magnetic field Bso . Upon absorption of circularly polarized light with in-plane angular momentum, the electron distribution function acquires an asymmetry in k-space due to optical selection rules. A DC charge current is developed via the circular photogalvanic effect [22]. This effect has been observed in a number of quantum well structures [23, 24], and more recently in topological insulators [25, 26] and bulk Rashba semiconductors [27]. As in the Rashba-Edelstein effect, a nonequilibrium spin accumulation also results from this distribution [28], and exerts a torque on the magnetization [29]. For the model system we study here, we show that if Bso ≫ ∆, the optically induced torque is proportional to the absorption rate times ∆τ /¯h, where τ is the carrier scattering time. The optical spin-orbit torque may exceed the absorbed angular momentum flux if ∆τ /¯h > 1. The effect we describe is applicable to any ferromagnet with broken inversion symmetry and strong spin-orbit coupling, such as magnetic topological insulators [30], or heavy metal-ferromagnet bilayers. We discuss our results in terms of experiments and real systems in Sec. IV. II. A.

MODEL

System description

Our starting point is an effective model for a Rashba semiconductors with perovskite lattice structure [31]. This is a convenient model for studying optical transitions with Rashba spin-orbit coupling, and describes recently studied mixed halide perovskite semiconductors which exhibit both exceptional optical absorption and strong spin-orbit coupling [32]. In these materials the valence band has a predominantly s-like orbital character and consists of spin S = 1/2 states, while the conduction band consists of the spin-orbit splitoff J = 1/2 states.

2

FIG. 1: (a) shows magnetization and spin of eigenstates for ∆ ≫ Bso , (b) shows the nonequilibrium spin density induced by circularly polarized light. (c) and (d) show the same in the case ∆ ≪ Bso .

This band ordering is opposite to that of commonly studies semiconductors such as GaAs (see Appendix A for more details about the electronic structure). In terms of real space atomic orbitals |px,y,z i and spin | ↑, ↓i, the Jz = +1/2 state is given by: 1 +1/2 |J1/2 i = − √ (|px , ↓i + i|py , ↓i + |pz , ↑i) . 3

(1)

Note that the expectation value of the spin is anti-parallel to J and has a magnitude of 1/6. Rashba spin-splitting arises from inversion symmetry breaking. The symmetry breaking in these materials is typically the result of displacement of the B atom from the octahedra center of the conventional ABO3 perovskite lattice structure along the +z-direction. This leads to Rashba terms acting on both J and S states of conduction and valence bands, respectively. To include ferromagnetism we add a spin-dependent exchange field of magnitude ∆ along the z-direction. Due to the spin character of the states described above, the exchange field results in a spin splitting ∆ of the S = 1/2 valence band and −∆/3 of the J = 1/2 conduction band. The Hamiltonian for the system consists of the conduction band Hc , the valence band Hv , and the conductionvalence band coupling Hc−v , given by: ¯ 2 k2 h + αc σ · (k × z) − σz (∆/2) + ǫ0 , (2) 2me ¯h2 k 2 + αv σ · (k × z) + σz (∆/6) , (3) = 2mh = iξ (k · σ) + γ1 (4 − k+ k− ) σz . (4)

Hc = Hv Hc−v

Here σ is the Pauli spin matrix, me(h) is the electron (hole) effective mass, αc(v) is the Rashba parameter for conduction (valence) band, ξ is the s-p hopping parameter, γ1 is the s-p hopping parameter associated with broken inversion symmetry, and ǫ0 is an energy offset for the conduction band. Fig. 2(a) shows the band structure for the model. We label states parallel (anti-parallel) to

FIG. 2: (a) shows the band structure of the perovskite with ∆ = 0.15 eV, Bso = 0.55 eV. Gray vertical lines indicate positions of energetically allowed transitions for h ¯ ω = 1.9 eV, while the thicker black vertical line indicates the dominant transition. u(d) label states aligned (anti-aligned) with the effective magnetic field, and a is the lattice constant. For kx > 0, the u state for conduction (valence) band corresponds to Jy < 0 (Sy < 0), denoted with a dot, while the d state is denoted with an x. (b) shows the k-resolved steady state density upon illumination by light with circular polarization along the y-direction.

the k-dependent effective magnetic field u(d). Note that the Rashba parameter of the s-like valence band αv relies on hybridization with the remote conduction band, and is therefore smaller than the Rashba parameter of the conduction band αc . As seen in Fig. 2, we use the same sign for the Rashba parameter for conduction and valence bands [33]. We define Bso as the Rashba-derived splitting at the conduction band minimum: Bso = 2α2c me /¯h. The relative magnitude of Bso and the magnetic exchange splitting ∆ is a key parameter in this system. Appendix A gives the more general form of the 4 × 4 Hamiltonian in terms of basic tight binding hopping parameters. Eqs. 2-4 are a good approximation in the limit where the band gap is greater than other energy scales. However we note that the general conclusions given in this work do not rely on the details of the Hamiltonian, as we discuss more in Sec. IV.

B.

Optical Bloch equations

In Appendix B, we derive the formula for the steady state density matrix under monotonic optical excitation E cos(ωt). The j, k component of the steady state hole density matrix ρhjk is: ρhjk (k) =

h ¯ τ

−

X 1
+ i Ejv − Ekv ℓ∈c ∗ vjℓ vkℓ

¯hω − (Ekv − Eℓc ) +

i¯ h τ

∗ vjℓ vkℓ
¯hω − Ejv − Eℓc − !

,

i¯ h τ

(5)

where the subscripts j, k refer to the u, d valence bands, the sum ℓ is over conduction band states, τ is the

3 c(v)

is the (k depencarrier scattering time, and Ej dent) j-th energy eigenvalue of conduction (valence) band. The dipole transition matrix element vjk is
The vjk = hj|v · E|ki/ Ejv − Ekc , where v = ∂H ∂k . dipole transition matrix is determined by the conductionvalence band coupling of Eq. 4, and takes a simple form when the first term in Eq. 4 is much larger than the second term: vjk ≈ iξ

hj|σ · E|ki . Ejv − Ekc

(6)

We use this approximation in the analytic results of the next section. We find it’s useful to write the optical response in terms of photon flux Φ, which is related to |E|2 via Φ = cǫ|E|2 / (2¯hω), where c is the speed of light and ǫ is the dielectric permittivity of the material. Performing the sum over crystal momentum and the trace over the diagonal components of ρh gives the steady state, nonthermalized photoexcited hole-density n. Eq. 5 can then be recast in terms of the absorption coefficient α and the photon flux as: X
Φατ = ρhuu (k) + ρhdd (k) = n. (7) k

This equation states that the steady state density of nonthermalized electron-hole pairs is equal to the absorption rate Φα multiplied by the carrier scattering time τ [34]. In our analysis, we use Eq. 7 to express the torque in terms of the absorption rate. The spin density for holes and electrons is given by:  1  (8) sh = + Tr ρh σ , 2 1 se = − Tr [ρe σ] . (9) 6 The relative sign and magnitude of electron and hole spin are derived from the spin of the J = 1/2 state, as discussed earlier. The torque on the magnetization is determined by the spin component transverse to the magnetization [35, 36]:
∆ e (10) s − sh × z . h ¯ For the numerical results, we consider a 2-dimensional system so that the sum over k in Eq. 7 is restricted to (kx , ky ). For the results given in Sec. III, ∆τ /¯h is a key dimensionless parameter; for ease of presentation we set h = 1 and present results in terms of ∆τ . ¯ Γ=

III.

RESULTS

We present analytic results using the Bloch equations with Eqs. 2-4, and also present numerical results using the full 4 × 4 Hamiltonian given in Eq. A1. We consider photon energies which are large enough to ensure transitions between all conduction and valence bands.

A.

∆ ≫ Bso : Optical spin-transfer torque

When spin-orbit coupling is negligible, conservation of total spin angular momentum applies and the optical spin transfer torque is limited by the photon absorption rate. The optical spin transfer torque was derived semiclassically in previous works for this case [9, 11]; here we derive the torque by evaluating the Bloch equations directly. For light with circular polarization in the yˆdirection, E = E (ˆ x + iˆ z). The valence and conduction band eigenstates are spinors along the z-direction;     0 1 (11) ψd = ψu = 1 0 Eq. 6 yields the following relation for dipole matrix elements: vuu,dd ∝ ±1 vud,du ∝ i.

(12)

In Appendix B, we use these dipole transition matrix elements to give the explicit form of the hole density matrix elements. In the limit where the band gap energy is larger than other energy splittings, inspection of the density matrix elements Eqs. (B7)-(B10) indicates the following relation between diagonal and off-diagonal elements:
(i − ∆τ ) ρhud − (i + ∆τ ) ρhdu = ρhuu + ρhdd = n. (13) Eq. 13 determines the hole transverse spin density sh⊥ in terms of ∆τ and the photoexcited density n:   n 1 sh⊥ = (− (∆τ ) x ˆ+y ˆ) . (14) 2 1 + (∆τ )2

A similar analysis holds for electrons, leading to the transverse electron spin density:     ∆τ 1 n se⊥ = x ˆ − y ˆ . (15) 6 1 + (∆τ /3)2 3 Using Eq. 10 for the torque, and substituting n = Φατ , we obtain the following result for the torque per absorption rate: ! 3 1 ∆τ Γx , (16) = 2 − 2 Φα 2 1 + (∆τ ) 9 + (∆τ ) ! 2 1 (∆τ ) Γy 1 = . (17) 2 + 2 Φα 2 1 + (∆τ ) 9 + (∆τ ) Both components of torque per absorption rate are strictly less than 1, as required by angular momentum conservation. In the limit ∆τ ≫ 1 we obtain that Γy = Φα. In this case the injected angular momentum is entirely transferred to the magnetization. In the limit ∆τ ≪ 1, we find that the torque is aligned primarily in the x-direction, and is given by Γx = ∆τ /3.

4 net spin in the y-direction. The θ integral for Syh is: Z
τ Syh ∝ dθ cos (θ) ρhuu (θ) − ρhdd (θ) = 2 Z h i 2 2 τ dθ cos (θ) (1 + cos (θ)) − (1 − cos (θ)) = 2πτ.

(21)

Eqs. 20-21 show that the spin polarization due to absorption of circularly polarized light is given by: FIG. 3: x and y components of the torque per absorption rate as a function of ∆τ for Bso = 0. The system parameters are me = 1.2 m∗ , mh = 2.0 m∗ , where m∗ is the bare electron mass, τ = 0.6 ps, the photon energy is 2 eV, ∆ is varied between 10−5 eV to 10−1 eV. Symbols are numerically computed values, and solid lines are Eq. 17.

In this case the injected angular momentum is mostly lost to the lattice [11]. Fig. 3 shows a comparison of the torques given by Eq. 17 (given in solid lines) and the numerical results obtained with the full Hamiltonian and k-integration (given by symbols). We find excellent agreement between the numerical and analytical results.

B.

⇒

Syh 1 = . n 4

(22)

A similar analysis for electrons reveals that Sye /n = 1/12. The resulting torque on the magnetization is along the x-direction with magnitude:   1 1 Γx = ∆n − 4 12 Γx ∆τ ⇒ = . (23) Φα 6

∆ ≪ Bso : Optical spin-orbit torque

When the spin-orbit splitting is greater than the magnetic exchange splitting, the spinors of conduction and valence bands are aligned to the k-dependent effective magnetic field, which is directed along k × z. Letting k = k (cos (θ) , sin (θ) , 0), the spinors take the following form:     1 1 1 1 √ √ ψd = (18) ψu = iθ iθ 2 −ie 2 ie For light with circular polarization in the yˆ-direction, the dipole matrix elements depend on θ as: vud,du ∝ (1 ± cos (θ)) , vuu,dd ∝ ∓i sin (θ) .

(19)

To evaluate the spin density, the sum over k for the density matrix is transformed to an integral over k and θ. Due to the spin texture of the Rashba model, the net spin polarization is determined by the θ integral. Using Eqs. 5 and 19, the θ integral for the hole density is [37]: Z h 2 2 n ∝ τ dθ (1 + cos (θ)) + (1 − cos (θ)) +  2 sin2 (θ) = 8πτ. (20)

In the case where Bso ≫ ∆, the transverse spin density is mostly aligned to the spin of the eigenstate and interband coherence can be neglected. As shown in Fig. 2(b), optical absorption is asymmetric in kx and results in a

FIG. 4: (a) Torque per absorption rate as a function of Bso /∆. For these results, τ = 0.13 ps, ∆ = 0.1 eV, and αc and αv are varied between (0 − 0.14) eV · nm and (0 − 0.06) eV · nm, respectively. (b) Torque per absorption as a function of ∆τ . For these results, αc = 0.9 eV · nm, αv = 0.05 eV · nm, ∆ is fixed to 0.1 eV, and τ is varied between 6.5 × 10−15 s and 6.5 × 10−13 s. Black line is Eq. 23, and blue dots are numerical results.

In this case the ratio of magnetization torque to incoming optical angular momentum is proportional to ∆τ . The optical excitation enables angular momentum to flow between the lattice and magnetization, and the magnitude of angular momentum flow exceeds the angular momentum injection rate if ∆τ /6 > 1. The torque is the result of the misalignment of the eigenstate spin and the magnetic exchange interaction. Fig. 4(a) shows the crossover between regimes ∆ ≫ Bso and ∆ ≪ Bso computed numerically. For small Bso /∆ and large ∆τ , the torque is along the y-direction and is bounded by the angular momentum absorption rate, as discussed in the previous section. For larger Bso /∆, the torque is along the x-direction and its magnitude exceeds the angular momentum absorption rate. Fig. 4(b) shows that torque in the x-direction as a function of ∆τ . The dots indicate Eq. 23 and the solid lines are the numerical results, and show good agreement.

5 IV.

DISCUSSION

We first use our results to estimate the required photon flux and fluence necessary to induce magnetic switching. The optically induced spin transfer torque competes with the intrinsic damping torque of the magnetic layer. For a layer of thickness t, out-of-plane anisotropy field B, and magnetization Ms , the damping rate is αd γBtMs /µB . Here αd is the magnetic damping, γ is the gyromagnetic ratio, and µB is is the Bohr magneton. Setting the optical spin-orbit torque equal to the damping torque and solving for Φ results in: Φ = αd γBt

Ms 6¯ h 1 . µB ∆τ W α

(24)

Here W is the thickness of the absorbing layer, α is the absorption coefficient, and we assume W α ≪ 1. For parameter values of αd = 0.01, Ms = 105 A/m, B = 0.1 T, ∆ = 0.1 eV, τ = 10−14 s, W = 5 nm, t = W , α = (100 nm)−1 , we obtain a value of Φ = 7 × 1029 m−2 · s−1 . Choosing an optical pulse length of 1 ns and a photon energy of ¯hω = 1.5 eV, the corresponding fluence is 17 mJ/cm2 . This can be compared to a maximum fluence of 1 mJ/cm2 used in Ref. [38]. The large fluence needed for switching is due in part to a smallness factor (W α)−1 : The torque is derived from optical absorption only by interfacial layers so that the total absorption rate is low. This could be mitigated with light trapping techniques optimized for the photon energy and angle of incidence, so that the light has multiple round trips through the interfacial layers for increased absorption. We also note that the assumption of linearity and steady state employed in this work may not apply for such short, high

[1] A. Kirilyuk, A. V. Kimel, and T. Rasing, Rev. Mod. Phys. 82, 2731 (2010). [2] G. Zhang and W. H¨ ubner, Phys. Rev. Lett. 85, 3025 (2000). [3] C.-H. Lambert, S. Mangin, B. C. S. Varaprasad, Y. Takahashi, M. Hehn, M. Cinchetti, G. Malinowski, K. Hono, Y. Fainman, M. Aeschlimann, et al., Science 345, 1337 (2014). [4] L. Guidoni, E. Beaurepaire, and J.-Y. Bigot, Phys. Rev. Lett. 89, 017401 (2002). [5] M. Battiato, K. Carva, and P. M. Oppeneer, Phys. Rev. Lett. 105, 027203 (2010). [6] N. Duong, T. Satoh, and M. Fiebig, Phys. Rev. Lett. 93, 117402 (2004). [7] Y. Hashimoto, S. Kobayashi, and H. Munekata, Phys. Rev. Lett. 100, 067202 (2008). [8] N. Tesaˇrov´ a, P. Nˇemec, E. Rozkotov´ a, J. Zemen, T. Janda, D. Butkoviˇcov´ a, F. Troj´ anek, K. Olejn´ık, V. Nov´ ak, P. Mal` y, et al., Nat. Phot. 7, 492 (2013). [9] J. Fern´ andez-Rossier, A. S. N´ un ˜ez, M. Abolfath, and A. MacDonald, arXiv preprint cond-mat/0304492 (2003). [10] J. Chovan, E. Kavousanaki, and I. Perakis, Phys. Rev.

intensity pulses. The inclusion of nonlinear effects such as Pauli blocking and time-dependent effects has been considered in other systems [10], and is the subject for future work. We next comment on the generality of our results. Our model of a magnetized perovskite is idealized for this problem, although the model presented here could be realized with a ferromagnet-perovskite heterostructure in which the perovskite is magnetized via the proximity effect. Such heterostructures have been proposed and implemented for FM-GaAs systems [39–41]. However the scaling of optical spin-orbit torque given in Eq. 23 is generally valid when the magnetization is aligned to the symmetry breaking direction and Bso ≫ ∆. The wellstudied heavy metal-ferromagnet metallic bilayer satisfies the requirements for this effect. Optical absorption in a Co-Pt bilayer was recently shown to induce a spin photovoltaic effect [42], and photocurrents induced by magnetization dynamics [38]. The primary difference between semiconductor and metallic systems in this analysis is the reduced τ associated with optical excitations in metals. However for Co-Pt, the magnitude of the proximate magnetization on the Pt of approximately 0.2 µB leads to an exchange field of 0.1 eV [43], so that ∆τ /¯h > 1 for τ > 6.5 fs. Acknowledgments

J. L. acknowledges support under the Cooperative Research Agreement between the University of Maryland and the National Institute of Standards and Technology Center for Nanoscale Science and Technology, Award 70NANB10H193, through the University of Maryland.

Lett. 96, 057402 (2006). [11] P. Nˇemec, E. Rozkotov´ a, N. Tesaˇrov´ a, F. Troj´ anek, E. De Ranieri, K. Olejn´ık, J. Zemen, V. Nov´ ak, M. Cukr, P. Mal` y, et al., Nat. Phys. 8, 411 (2012). [12] J. C. Slonczewski, J. Magn. Magn. Mat. 159, L1 (1996). [13] L. Berger, Phys. Rev. B 54, 9353 (1996). [14] D. C. Ralph and M. D. Stiles, J. Magn. Magn. Mat. 320, 1190 (2008). [15] P. M. Haney and M. D. Stiles, Phys. Rev. Lett. 105, 126602 (2010). [16] 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 476, 189 (2011). [17] K. Garello, I. M. Miron, C. O. Avci, F. Freimuth, Y. Mokrousov, S. Bl¨ ugel, S. Auffret, O. Boulle, G. Gaudin, and P. Gambardella, Nat. Nanotech. 8, 587 (2013). [18] L. Liu, C.-F. Pai, Y. Li, H. Tseng, D. Ralph, and R. Buhrman, Science 336, 555 (2012). [19] J. Kim, J. Sinha, M. Hayashi, M. Yamanouchi, S. Fukami, T. Suzuki, S. Mitani, and H. Ohno, Nat. Mat. 12, 240 (2013).

6 [20] F. Freimuth, S. Bl¨ ugel, and Y. Mokrousov, Phys. Rev. B 94, 144432 (2016). [21] A. Qaiumzadeh and M. Titov, Phys. Rev. B 94, 014425 (2016). [22] V. Asnin, A. Bakun, A. Danishevskii, E. Ivchenko, G. Pikus, and A. Rogachev, Sol. St. Comm. 30, 565 (1979). [23] S. Ganichev, E. Ivchenko, S. Danilov, J. Eroms, W. Wegscheider, D. Weiss, and W. Prettl, Phys. Rev. Lett. 86, 4358 (2001). [24] X. He, B. Shen, Y. Tang, N. Tang, C. Yin, F. Xu, Z. Yang, G. Zhang, Y. Chen, C. Tang, et al., App. Phys. Lett. 91, 071912 (2007). [25] P. Hosur, Phys. Rev. B 83, 035309 (2011). [26] C. Kastl, C. Karnetzky, H. Karl, and A. W. Holleitner, Nature communications 6 (2015). [27] N. Ogawa, M. Bahramy, Y. Kaneko, and Y. Tokura, Phys. Rev. B 90, 125122 (2014). [28] V. M. Edelstein, Sol. St. Comm. 73, 233 (1990). [29] A. Manchon and S. Zhang, Phys. Rev. B 78, 212405 (2008). [30] N. Ogawa, R. Yoshimi, K. Yasuda, A. Tsukazaki, M. Kawasaki, and Y. Tokura, Nat. Comm. 7 (2016). [31] M. Kim, J. Im, A. J. Freeman, J. Ihm, and H. Jin, Proc. Nat. Ac. Sc. 111, 6900 (2014). [32] D. Niesner, M. Wilhelm, I. Levchuk, A. Osvet, S. Shrestha, M. Batentschuk, C. Brabec, and T. Fauster, Phys. Rev. Lett. 117, 126401 (2016). [33] We find that the results are qualitatively similar (differ by less than a factor of 2) if αc and αv have different signs. [34] We note that τ is the carrier scattering time, which is generally much shorter than the electron-hole recombination time. [35] A. S. N´ un ˜ez and A. H. MacDonald, Sol. St. Comm. 139, 31 (2006). [36] P. Haney, D. Waldron, R. Duine, A. Nunez, H. Guo, and A. MacDonald, Phys. Rev. B 76, 024404 (2007). [37] In writing this relation, we assume that the band gap is c much greater than other band splittings, so that Ejk − ′ ′ v Ej ′ k′ is constant for all j, k, j , k . [38] T. Huisman, R. Mikhaylovskiy, J. Costa, F. Freimuth, E. Paz, J. Ventura, P. Freitas, S. Bl¨ ugel, Y. Mokrousov, T. Rasing, et al., Nat. Nanotech. (2016). [39] V. Korenev, Semiconductor Science and Technology 23,

114012 (2008). [40] J. McGuire, C. Ciuti, and L. Sham, Phys. Rev. B 69, 115339 (2004). [41] X. Lou, C. Adelmann, M. Furis, S. Crooker, C. Palmstrøm, and P. Crowell, Phys. Rev. Lett. 96, 176603 (2006). [42] D. Ellsworth, L. Lu, J. Lan, H. Chang, P. Li, Z. Wang, J. Hu, B. Johnson, Y. Bian, J. Xiao, et al., Nat. Phys. (2016). [43] K.-J. L. A. M. Paul M. Haney, Hyun-Woo Lee and M. D. Stiles, Phys. Rev. B 88, 214417 (2013). [44] W. Sch¨ afer and M. Wegener, Semiconductor optics and transport phenomena (Springer Science & Business Media, 2013). Appendix A: Tight-binding form of Hamiltonian

Here we discuss the specific form of the Hamiltonian describing the class of perovsite materials. Taking the example of the mixed halide perovskite CH3 NH3 PbI3 , its cubic form has direct band gap at R point [2π/a, 2π/a, 2π/a], where a is the lattice constant. The near-gap conduction band states are composed of the p orbitals of Pb while the valence bands states are derived from the Pb s orbital and I p orbitals. The energy states near Fermi level can be described by a 8 × 8 tight-binding Hamiltonian including only s and p orbitals of Pb occuping the cubic lattice sites. We consider four types of interand intra-orbital hopping parameters, such as tσss , tσpp , tπpp and tσsp . The effect of I p orbitals is implicitly taken into account through tuning the magnitude of hopping parameters. The spin-orbit coupling splits degenerate conduction band states (L = 1) into lower J = 1/2 and upper J = 3/2 bands, leading to a J = 1/2 conduction band and S = 1/2 valence band. In this study, we focus on the optical transition between valence and conduction bands so that we can truncate the 8 × 8 Hamiltonian to a 4 × 4 minimal continuum model to describe the neargap optical transition. With the basis {|S, ↑i, |S, ↓i, |J = 1/2, jz = +1/2i, |J = 1/2, jz = −1/2i}, the effective continuum Hamiltonian near the R point up to second order in k is given by

 tσss k 2 − ∆/2 0 itkz − γ1 (k+ k− − 4) itk−  itk+ −itkz + γ1 (k+ k− − 4)  0 tσss k 2 + ∆/2 , H= π 2   −itkz − γ1 (k+ k− − 4) −itk− ǫ0 + tpp k + ∆/6 iγ2 k− π 2 −itk+ itkz + γ1 (k+ k− − 4) iγ2 k+ ǫ0 + tpp k − ∆/6 

√ √ z where k± = kx ± iky , t = 2tσsp / 3, γ1 = γsp / 3,
z γ2 = 4γpp /3, ǫ0 = (ǫp − ǫs ) − 2 tσpp + 2tpp π − λ + 6tσss . ∆ is the exchange interaction and λ is the spin-orbit couz z pling. γsp and γpp are the spin-independent hoping integrals induced by inversion symmetry breaking along z direction.

(A1)

When the band gap energy is larger than all other energy scales, Eq. A1 can be projected on to separate 2 × 2 dimensional Hamilotians for conduction and valence band, as given in Eqs. 2-4 of the main text. The parameters entering these projected Hamiltonians are related to the hopping parameters given here as: αc = γ2 ,

7 αv = 2tγ1 /ǫ0 , me,h = ¯h2 /(2a2 t′s,p ). Appendix B: Derivation of steady state density matrix

Here we review the derivation of the optical Bloch equations [44]. In the eigenstate-basis of the ground state and considering only valence-conduction interband transitions, the Hamiltonian for the system including the optical excitation is:  v    Eu 0 0 0 0 0 vuu vud v 0 vdu vdd   0 Ed 0 0   0 H(k) =  + ∗ ∗ 0 0 Euc 0   vuu vdu 0 0  c ∗ ∗ 0 0 0 Ed vud vdd 0 0

c v (Eu,d ) is the k-dependent valence (conduction) Here Eu,d band energy for the u, d state. The (u, d) label corresponds to the spin direction of the eigenstate, which is parallel (u) or anti-parallel (d) to the k-dependent effective magnetic field. vjk denotes the optical field-induced coupling between state j of the valence band and state k of the conduction band. For the density matrix ρ, we denote the valence (conduction) band density matrix as ρv (ρc ), and P to denote electron-hole density matrix elements. The structure of ρ is:

ρvuu  ρv ρ =  du ∗ Puu ∗ Pud 

ρvud ρvdd ∗ Pdu ∗ Pdd

Puu Pdu ρcuu ρcdu

 Pud Pdd  . ρcud  ρcdd

(B1)

In our analysis, we switch from the conduction-valence representation of the density matrix to a electron-hole picture, where the electron density matrix if ρe = ρc , and the hole density matrix is ρh = 1 − ρv . The equation of motion for the density matrix ρ is given by: 1 ρ ∂ρ = [ρ, H] − . ∂t i¯ h τ

(B2)

Equation B2 leads to the semiconductor Bloch equations. Writing these perturbatively in E, and assuming that at t = 0, ρhuu = ρhdd = 1 while other elements of the density matrix are 0 leads to the following equation for Pjk :  
1 i Pjk (t) = −ivjk (t) . Ejv − Ekc + ∂t + ¯h τ

(B3)

Equation B3 is solved by Fourier transform techniques, and yields the following expression for the dipole density matrix element:

Pjk (t) =

vjk 2

exp (iωt)
¯hω + Ejv − Ekc −

i¯ h τ

exp (−iωt)
−¯hω + Ejv − Ekc −

+

i¯ h τ

!

, (B4)

where τ is the carrier scattering time, and the dipole matrix element v is given by: vjk =

hj|v · E|ki , Ejv − Ekc

(B5)

where the velocity operator is v = ∂H ∂k . Near a resonance condition ¯hω ≈ Ej − Ek for a pair of valence/conduction bands.
In this case, the term with denominator ¯hω − Ejv − Ekc has the maximal contribution. The equation of motion for hole-hole density matrix is, to lowest order in v:

X ∂ h ∗ ∗ ρjk (t) = i (vjℓ Pkℓ − vkℓ Pjℓ ) ∂t ℓ∈c  
1 v v ρhjk (t). − i Ej − Ek − τ The sum ℓ is over conduction band states, while the indices j and k correspond to valence bands. Letting ∂ρh /∂t = 0 and only including
terms with denominators of the form h ¯ ω − Ejv − Ekc yields the following expression for the steady state ρhjk :

ρhjk =

h ¯ τ

X 1
+ i Ejv − Ekv ℓ∈c

∗ vkℓ vjℓ v ¯hω − (Ek − Eℓc ) +

∗ vjℓ vkℓ
¯hω − Ejv − Eℓc − !

i¯ h τ

.

i¯ h τ

−

(B6)

Eq. B6 is the general form for the hole density matrix under optical excitation. The electron density matrix has a similar form. Here we give the explicit forms of the elements of the hole density matrix for incident light polarized in the y-direction. In the case where Rashba spin-orbit is negligible and eigenstates are spinors along the zdirection, vuu,dd = ±1, vud,du = i. If ǫ0 ≫ ∆, then Eud ≈ Edu ≈ Euu , Edd . These equations then simplify to:

8

= τ

1 Euu − ¯ hω +

ρhdd = τ

1 Edd − ¯ hω +

ρhuu

i τ

i τ

1 + Eud − ¯hω + 1 + Edu − ¯hω +

ρhud

−iτ = 1 + i∆τ

1 Edu − ¯ hω +

ρhdu

iτ = 1 − i∆τ

1 Euu − ¯ hω +

i τ

i τ

i τ

i τ

1 − Euu − ¯hω − 1 − Edd − ¯hω −

1 + Edd − ¯hω + 1 + Eud − ¯hω +

i τ

i τ

i τ

i τ

1 − Eud − ¯hω − 1 − Edu − ¯hω −

1 − Euu − ¯hω − 1 − Edu − ¯hω −

i τ

i τ

i τ

i τ

! !

,

(B7)

,

(B8)

1 − Eud − ¯hω − 1 − Edd − ¯hω −

i τ

!

,

(B9)

i τ

!

.

(B10)