Non-local spin valve in Van der Pauw cross geometry with four ferromagnetic electrodes. K.-V. Pham∗ 2 Laboratoire de Physique des Solides, Univ. Paris-Sud, 1 0 CNRS, UMR 8502, F-91405 Orsay Cedex, France 2 n Abstract a J We consider a non-local spin valve in a Van der Pauw cross geometry with four ferromagnetic 4 1 electrodes. Two antiparallel ferromagnets are used as (charge) source and drain while the de- ] l tector circuit involves measuring the voltage between two collinear ferromagnets with parallel or l a h antiparallel magnetizations. We find a potentially large increase of the non-local spin voltage. The - s e setupdisplaysseveraladditional interestingproperties: (i)infiniteGMRforthenon-localresistance m . (if a symmetry requirement for the device is met); (ii) ON-OFF switch effect, when the injector t a m electrodes are parallel instead of antiparallel; (iii) insensitivity to offset voltages. The device can - d additionally be used as a Direct Spin Hall Effect probe and as a reprogrammable magneto-logic n o gate implementing basic operations (NOR, NAND, inverter, AND, OR, etc). c [ 1 v 2 8 9 2 . 1 0 2 1 : v i X r a 1 I. INTRODUCTION. Pure spin manipulation is an important topic in spintronics due to possible applications for programmable logic and memory. Non-local spin valves1–6 provide an example of pure spin current generation. In the latter a pure spin current generated at a ferromagnetic - paramagnetic interface reaches a ferromagnetic probe in the absence of charge current; the spin accumulation created in the probe vicinity can then be detected as a charge voltage by virtue of Johnson-Silsbee charge-spin coupling1,7,8. Such pure spin currents have already proved useful to switch magnetizations9,10. This is opening promises for further applications in logic, sensing and memory devices but to that end it is desirable to increase the signals: the spin voltages are typically in the µV range while mV would be more suitable to ensure sufficient SNR (signal to noise ratio). While spin valves are in everyday use in hard-drives, in order to reach or even go beyond the 1 Tbit/inch2 density in hard-drives11–13, novel spin valves are required with a low RA resistance times area product (typically RA 0.1 Ωµm2)12–14 while sustaining a mA current ≤ and mV voltage; this is beyond TMR (tunneling magnetoresistance) sensors capability since they are too resistive. Metallic spin valves are therefore more appropriate. However in order for them to have a suitable SNR it is also necessary that the read-heads function with a large enough contrast12: ∆RA 5 mΩµm2. Metallic spin valves with larger GMR ratio ≥ ∆R/R are therefore required. At first sight non-local metallic spin valves are not obvious candidates for larger MR (magnetoresistance) ratios: they indeed underperform when compared to their local coun- terparts; using the same materials and dimensions a local CPP spin valve is expected to have a larger ∆RA since spin confinement is better15–17. The goal of this paper is to discuss a non-local spintronics device with potentially: enhanced spin voltage in the mV range for currents mA (so that the non-local • ∼ resistance variation ∆R is in the Ohm range) with realistic density currents j < nl 108 A/cm2 addressing the needs of industry. enhanced non-local GMR ratio: ∆R /R (up to 100% for the pessimistic ratio; or nl nl • up to infinity for the optimistic ratio), helping quite generally for better SNR and perhaps making them suitable candidates as sensors or read-heads for hard-drive areal 2 densities larger than 1 Tbit/inch2. Regarding the enhancement of the non-local signal, spin valves with tunnel junctions have been reported with non-local resistance in the Ω range but due to a polarization decreasing rapidly when the current is larger than µA , the spin voltage remains small in the usual ∼ µV range18,19. However much progress has been reported recently in pure metallic lateral spin valves ( 10 µV for nanopillars10) or lateral valves with very low resistance tunnel ∼ junctions (using a thin nm MgO layer)20–22, reaching in the latter case the 100 µV range with RA 0.2 Ωµm2 so that already ∆RA is of the order of a few 1 mΩµm2. ∼ We propose to go even further in the improvement by relying on two ideas: (i) use two injectors instead of a single one, which should at face value double the signal; (ii) enhance the spin confinement by making good use of tunnel barriers, thin enough to stay close to the metallic regime but resistive enough to hinder spin leakage. The idea of minimizing the spin relaxation volume has been expressed in particular in17,23 and explains the large signals seen in spin valves using carbon nanotubes24. Our basic setup applies these ideas by using four collinear ferromagnetic terminals. In a standard lateral spin valve2,3,25 the charge current flows from a ferromagnetic electrode to a paramagnetic drain; the injector electrode is connected by a lateral wire to another ferromagnetic electrode used as a detector. In our setup we propose to replace the param- agnetic drain by a ferromagnet antiparallel to the terminal acting as current source; the two antiparallel electrodes are connected by a paramagnetic metal with thin tunnel barriers in order to better confine spin. The two antiparallel ferromagnets act as spin sources although in terms of charge one is a source and the other a drain: this effectively doubles the spin accumulation in the lateral wire while the tunnel junctions make sure spin is confined. We further change the standard detection setup by using a ferromagnetic counter- electrode instead of a paramagnetic one. The advantage of using two ferromagnets is evidenced when the two detector electrodes are placed symmetrically with respect to the injectors (source and drain): provided they are otherwise identical terminals this implies that when their magnetizations are parallel, their voltage difference should be identically zero by symmetry. This is how we reach an infinite non-local GMR ratio. We will also address the issue of voltage offsets plaguing non-local setups1,2,4,5,15,16: while voltages generated by spin accumulation are clearly observed, some additional voltages of variousorigins arealso usually seen. These offset (or baseline) voltages have been credited to 3 charge current inhomogeneities26–30 (which impact the calculations done for non-local setups since they usually assume one dimensional drift-diffusion equations7,31,32), or to heating (notably Joule and Peltier heating33–35). They may or may not be a nuisance but at any rate they prevent observation of pure non-local voltages. The device we discuss in this paper canbemadeinsensitive totheseoffsetvoltageswhenthetwodetector electrodesareidentical and symmetric since the offsets will cancel out when the voltage difference is measured. This is an additional advantage of our device. The geometry of our device is that of a Van der Pauw cross as in the Jedema and coll. seminal experiments4. A close device within a pure lateral geometry will be discussed elsewhere36. In Section II we introduce the Van der Pauw geometry with four ferromagnetic terminals and give general expressions for the non-local voltage. The basic functionalities of the device arediscussed, andnotablywewillshowthatthedevicecanperformlogicoperations(notably as a NOR or NAND gate), be reprogrammed to perform other functions (AND, XOR and inverter gates), displays a potentially interesting ON-OFF switch effect, and when used as a standard 1-bit read-head shows an infinite GMR for the non-local resistance. Use as a Direct Spin Hall Effect probe will also be discussed. The next section III studies in detail the impact of the transparency of interfaces and of the number of ferromagnetic terminals (two or three out of four) on spin confinement, re- sulting in small or large non-local signals. The signals expected are systematically compared to those in the standard lateral geometry. The last section IV discusses the main setup with four ferromagnetic symmetric terminals since it displays the previously mentioned properties of (i) immunity to offset voltages and (ii) infinite GMR ratio for the non-local resistance. Issues pertaining to the use as a sensor are briefly touched upon. The bulk of calculations are relegated to the Appendices. Appendix B revisits the bipolar spin switch calculations by including spin leakage in the measuring electrodes. 4 II. VAN DER PAUW SETUP. A. Geometry and notations. 1. Geometry. We consider in this section a four-terminal device in a Van der Pauw geometry (see Fig. 1). The terminals are ferromagnets F1 F4 positioned as in the Figure 1; we allow for − arms of unequal lengths. In sections III-IIIB we will allow some of these terminals to be paramagnetic through a suitable choice of parameters. One-dimensional assumption. We will assume that width and thickness of all arms are much smaller than their length. Experimentally, current inhomogeneities due to depar- tures from strict one-dimensional flow can arise; however the basic functionalities of our device are for the most part independent of that assumption although quantitative predic- tions may accordingly lose accuracy. Injector electrodes. F1 injects a charge current which is collected in terminal F2. We will designate them collectively as injector electrodes; when the need to differentiate them shows up, we will say that F1 is the source or injector electrode while F2 is the drain or collector electrode. Detector electrodes. The detection sub-setup consists in terminals F3 and F4 hooked to a voltmeter (or a potentiometer or an ammeter). The latter will measure the non-local voltage as a function of the magnetization orientations of each terminal. Orientation. We define points O(x = 0; z = 0) the origin and center of the Van der Pauw cross, A(x = 0; z = L ), B(x = 0; z′ = L ) , C(x = L ; z = 0) and D(x = 0; z′ = L ) 1 2 3 4 where each arm has been for later convenience oriented away from O (axis Ox, Ox′, Oz and Oz′ ) following Jedema and coll.4. The four paramagnetic arms are: I IV (resp. OA, OB, OC, OD). − The charge current flowing through F1 I II F2 is I and flows from top to bottom c − − − (is therefore negative relative to arm I, but positive, relative to arm II). The spin accumulation is defined as: (µ µ ) ↑ ↓ ∆µ = − (1) 2e 5 Figure 1. Van der Pauw cross with four ferromagnetic terminals. The arm lengths are respectively L , L , L and L . The arrows represent the magnetization direction. When the device is used as a 1 2 3 4 basic spin-valve, only one terminal can switch its magnetization (here F3). The arms are oriented away from origin O. (where for later convenience we have divided by the electron charge e). The spin currents are oriented away from origin O (therefore are counted positive on a given arm when flowing away from O). 2. Spin parameters. Arms parameters. The central cross is a normal metal. Its parameters are its spin resistance R = ρ∗ l /A where l is the spin diffusion length, A is the cross-section and N N N N N N ρ∗ is the resistivity. We define the lengths of each arm relative to the spin diffusion length N as (for i = 1 4): − L i l = . (2) i l N Ferromagnet parameters. For each ferromagnetic terminal F1 F4 (i = 1 4), one − − defines the conductivity polarization P (= β in Valet-Fert notation31), spin resistance F,i i 6 R = ρ∗ l /A where l is the spin diffusion length, A is the cross-section and ρ∗ = F,i F,i F,i F,i F,i F,i F,i (ρ ρ )/4 . i↑+ i↓ (NB: note on terminology; we will call throughout the paper ’spin resistance’ the char- acteristic resistance found as the product of the resistivity times the spin diffusion length divided by the cross-section ). Interface parameters. Attheinterfacebetweentheferromagnetsandtheparamagnetic arms we assume there is a spin dependent interface resistance so that one can define for each interface F N (i = 1 4) a conductance polarization P (= γ in Valet-Fert notation), a i ci i − − spin resistance R = (R R )/4. For simplicity we will neglect all spin flips at interfaces ci i↑+ i↓ so that spin relaxation occurs solely in the bulk of the device. 3. Spin resistance mismatch. We define spin resistance mismatch parameters at F - N interfaces as: R +R F c X = . (3) R N Three limits can be singled out: X < 1: this corresponds to the limit of a transparent junction, which as we will see later in detail (Appendix A2e) is very leaky in terms of spin: this favors large spin currents at the cost of reduced spin accumulations in the central paramagnet. X > 1: spin confining or tunneling regime, for which spin accumulation increases but spin current decreases (in magnitude) (see Appendix A2e). For X 10, for which the ∼ contact resistance is moderate (about 10 Ω) we will say that we are in the weak tunneling limit. This is the most interesting limit in terms of applications to all-metallic read-heads or sensors. For R = 102 104 Ω which are usual values in tunnel junctions, X 102 104 c − ∼ − which we will qualify as strong tunneling limit. Although the strong tunneling regime can be described by our equations, we will focus primarily in the discussions on the transparent and weak tunneling regime where resistances are in the metallic range which interests us for sensor applications. X = 1: spin impedance matching. The naming for this border situation will be justified below in the discussion on effective spin resistance (section IIA5). 7 4. Effective polarizations. The following definition will also prove useful. We define for each terminal (i = 1 4) an − effective polarization as: PR i P = (4) eff,i δ+ i g where: PR = (P R +P R )/R (5) i Fi Fi ci ci N and: g (X +1) (X 1) δ± = i expl i − exp l . (6) i 2 i ± 2 − i The effective polarization can be rewritten as: P R +P R Fi Fi ci ci P = ; (7) eff,i R sinhl +(R +R )coshl N i Fi ci i clearly, P 1. eff,i | | ≤ Uponmagnetizationreversal oftheelectrode, theeffective polarizationisanoddfunction: P P . eff,i eff,i −→ − In the limit of short arm length l 1: i ≪ P R +P R Fi Fi ci ci P (8) eff,i −→ R +R Fi ci which is a weighted average of the electrode bulk and interface polarizations. When l the effective polarization vanishes exponentially which translates the i −→ ∞ complete spin relaxation in the arm: P R +P R Fi Fi ci ci P = 2 exp l . (9) eff,i i R +(R +R ) − N Fi ci The effective polarization therefore varies between 0 and its maximum value PFiRFi+PciRci RFi+Rci which is bounded from above by sup(P ; P ). Fi ci 5. Effective spin resistances. We also define an effective spin resistance for each arm of length l (i = 1 4) as: i − 8 δ+ R (l , X ) = R i (10) eff,i i i N δ− i X coshl +sinhl i i i = R ; (11) N X sinhl +coshl i i i R (X ) is an increasing function of spin resistance mismatch X . eff,i i i It is shown in Appendix A2b that the spin current I (O) at the cross center O on arm i s,i and the spin accumulation there are related by ∆µ(O) = R I (O) (for arms III IV; eff,i s,i − − for arms I II a more general relation taking into account the spin injection at F1 and F2 − holds). This is an analog of Ohm’s law for spin which explains our identification of R as eff,i a spin resistance. (Note that the analogy is not complete: the relation for spin is a local one (expressed here at point O), while Ohm’s law holds for a voltage difference and is therefore non-local. This results of course from the non-conservation of spin current.) We also define a total effective spin resistance for the device: 1 R = . (12) eff 1 i=1−4 Reff,i The previous expression admits obviousPgeneralization to an arbitrary number n 4 of ≥ arms. As can be seen from its definition, the total effective resistance R is related to the arms eff spin resistances R by the analog of a parallel resistance addition law. We will show later eff,i (section IIB2) that the spin voltage is proportionalto the totaleffective spin resistance R eff so that large effective resistances are desirable; this will also effectively demonstrate for our geometry the parallel addition law for spin resistances. (For a discussion of spin resistance addition law at nodes we refer the reader to37.) Since the total effective spin resistance is the sum of four resistances in parallel, whenever one is much smaller than the others, it will short the other arms: spin leakage will be stronger so that spin accumulation will be reduced. The length dependence of the effective resistance for various values of the spin resistance mismatch X is shown in Fig. 2-3 . At large distance the effective resistance converges exponentially fast to R the spin resistance of the paramagnetic arm: n R R (13) eff,i N −→ 9 which reflects the fact that the spin relaxation is dominated by the paramagnet bulk. In the limit l 0 i −→ R R +R (14) eff,i Fi ci −→ (which is sensible since the paramagnet is then too short for spin relaxation to occur). When l 1, the effective resistance remains close to R +R if the distance l obeys: i Fi ci ≪ l X2 1 < X. − (cid:12) (cid:12) (cid:12) (cid:12) When X 1, the condition becomes l 1/X which reflects a steeper exponential decrease. ≫ ≪ The effective spin resistance for a given arm is therefore comprised between R and N R +R : F,i c,i R R R +R (15) N eff,i Fi ci ≤ ≤ (or the reverse inequality if R R +R ). N Fi ci ≥ As a rule the effective spin resistance will be larger for large interface or ferromagnet spin resistance (R and R ); since the ferromagnet spin resistance R is in general much smaller c F F than the paramagnet spin resistance R due to short spin diffusion lengths, large interface N resistances R are required to achieve large effective resistances R . c eff,i ForX = 1,oneobservesthattheeffectivespinresistancedoesnotdependanymoreonthe arm length l and is equal to R . One then has (on arms III or IV) ∆µ(O) = R I (O) i N N s,i − which is the relation one would get from an infinite arm. Everything happens as if the interface had been washed away: this is the reason why we qualified the case X = 1 as corresponding to spin impedance matching in IIA3. In the transparent regime (X < 1), the effective spin resistance is larger at large distance, which is an interesting feature for the design of large non-local circuits. This advantage is circumvented by the exponential decrease of the effective polarization (the non-local resis- tance will be shown to be proportional to both in IIB2) so that in terms of large signals the transparent regime is not interesting, neither in the short-distance nor the large-distance limit. 10