Electrical Spin Injection into InN Semiconductor ... - ACS Publications

Aug 13, 2012 - Peter Grünberg Institut (PGI-9) and JARA-Fundamentals of Future Information Technology, Forschungszentrum Jülich, 52425. Jülich, Ger...
4 downloads 0 Views 4MB Size
Letter pubs.acs.org/NanoLett

Electrical Spin Injection into InN Semiconductor Nanowires S. Heedt,*,† C. Morgan,‡ K. Weis,† D. E. Bürgler,‡ R. Calarco,†,∥ H. Hardtdegen,† D. Grützmacher,† and Th. Schap̈ ers*,†,§ †

Peter Grünberg Institut (PGI-9) and JARA-Fundamentals of Future Information Technology, Forschungszentrum Jülich, 52425 Jülich, Germany ‡ Peter Grünberg Institut (PGI-6) and JARA-Fundamentals of Future Information Technology, Forschungszentrum Jülich, 52425 Jülich, Germany § II. Physikalisches Institut, RWTH Aachen University, 52056 Aachen, Germany ABSTRACT: We report on the conditions necessary for the electrical injection of spin-polarized electrons into indium nitride nanowires synthesized from the bottom up by molecular beam epitaxy. The presented results mark the first unequivocal evidence of spin injection into III-V semiconductor nanowires. Utilizing a newly developed preparation scheme, we are able to surmount shadowing effects during the metal deposition. Thus, we avoid strong local anisotropies that arise if the ferromagnetic leads are wrapping around the nanowire. Using a combination of various complementary techniques, inter alia the local Hall effect, we carried out a comprehensive investigation of the coercive fields and switching behaviors of the cobalt micromagnetic spin probes. This enables the identification of a range of aspect ratios in which the mechanism of magnetization reversal is single domain switching. Lateral nanowire spin valves were prepared. The spin relaxation length is demonstrated to be about 200 nm, which provides an incentive to pursue the route toward nanowire spin logic devices. KEYWORDS: III-V semiconductors, InN nanowires, nanowire planarization, electrical spin injection, nonlocal spin valve, Co micromagnets

A

the estimated spin relaxation length lsf of several hundred micrometers, are challenged by dissenting results8 and based exclusively on local spin valve measurements. This local measurement geometry9,10 is generally regarded to be unreliable for proving spin injection.11,12 Only nonlocal measurements of spin imbalance13 provide an unambiguous interpretation excluding certain spurious effects; that is, the impact of local stray fields from the ferromagnetic metal (FM) electrodes, the anisotropic magnetoresistance (AMR) of the injector and detector and the magneto-Coulomb effect. The nonlocal open-circuit voltage mirrors the relative magnetization between the FM source electrode and the FM detector electrode. This circumstance is known as the spin valve effect. Apart from metallic systems, only few materials have been investigated in nonlocal spin valve measurement configurations, for example, graphene,14 carbon nanotubes,12,15 GaAs,16−18 and InAs.19,20 The first convincing results on spin injection in bottom-up nanowires7 were presented for 35−50 nm-wide Si nanowires (cf. also the work on axial Si-MnSi heterostructures by Lin et al.21). However, with regard to the small spin−orbit coupling for Si, investigations on nanowires with more

part from being promising building blocks for future integrated spintronic circuitry, semiconductor nanowires fabricated in the bottom-up paradigm theoretically represent suitable channels for spin transport due to their quasi onedimensional1,2 and scalable nature. The bottom-up nanowire growth process circumvents complex lithography steps and facilitates a less costly approach for future device applications. A major goal is to harness spin control in these nanodevices to reduce energy consumption and increase switching speed. In contrast to optical means of injection and detection, a purely electrical setup is essential for integrated circuits. Owing to the high Curie temperatures of ferromagnetic metals, roomtemperature applications are feasible.3 Nanowire spintronic devices embrace possible applications based on pure spin currents4 due to carrier control and tunable spin−orbit interaction. Hence, our curiosity was drawn toward lateral spin injection into these nanodevices. The first investigations on ferromagnetic contacts to semiconductor nanowires were carried out by Zwanenburg et al.5 on InP nanowires. Although none of the samples showed a nonlocal spin valve signal, the magneto-Coulomb effect was observed. This exemplifies that due to their geometry nanowires are particularly susceptible to spurious contact-related magnetoresistance (MR) effects. Recently, two groups6,7 claimed to have achieved spin accumulation in semiconductor nanowires. However, the presented results for Ge nanowires,6 in particular © 2012 American Chemical Society

Received: March 17, 2012 Revised: July 16, 2012 Published: August 13, 2012 4437

dx.doi.org/10.1021/nl301052g | Nano Lett. 2012, 12, 4437−4443

Nano Letters

Letter

reciprocally with n3d.33 The InN nanowires we investigate in this study feature a surface electron accumulation zone with the conduction band being pinned 0.9 eV below the Fermi energy,30 making low impedance ohmic contacts to metal leads feasible. The nanowires are mechanically transferred onto a Si/Si3N4 substrate equipped with electron beam lithography marker structures. As depicted in Figure 1a, individual nanowires are

pronounced spin−orbit coupling are desirable for spintronics applications. A prominent example for a spintronic device is the Datta− Das spin field-effect transistor,22 which comprises ferromagnetic electrodes for spin injection and detection, while the spin in the semiconductor channel is manipulated by electrical means via the Rashba effect. Only recently, this concept was successfully implemented for the first time by Koo et al.19 and Chang et al.20 by injecting spin-polarized electrons into a topgated InAs two-dimensional electron gas (2DEG). There, the spin alignment in the micromagnet is well-defined along the 8μm-wide 2DEG channel. However, the situation changes for narrow nanostructures like the nanowires presented here. In this case, the narrow geometry does not allow for planar interface areas for spin injection with distinct spin orientations. Nevertheless, the diameters of the nanowires are too large with respect to the FM film thickness to be negligible. Also, Monzon et al.23 hint at the fact that relatively small steps with respect to the FM layer thickness can already generate strong local domain fluctuations. This holds if the FM electrode runs across a steep mesa structure24 or, as in our case, across a semiconductor nanowire of a diameter of typically 120 nm in comparison to the FM film thickness of 60 nm. This highlights the necessity of introducing a more elaborate contacting scheme for narrow nanostructures, as presented below. The complex magnetic domain behavior at the interface25 and contact resistance engineering impose major challenges in fabricating efficient spin injection devices. In spite of the nontrivial contact preparation, these quasi one-dimensional nanostructures are promising candidates for coherent spin channels due to diffusive transport along confining boundaries in the regime of motional narrowing.1,2,26 With the elastic mean free path of about le = 50 nm27 being comparable to the InN nanowire width, the spin relaxation is consequently expected to be significantly reduced.28 The material under investigation is the low-bandgap semiconductor indium nitride with Eg = 0.73 eV.29,30 Wurtzite InN surfaces exhibit an intrinsic electron accumulation layer and electron densities which are significantly higher than in any other III-V semiconductor. The nanowires are single crystalline, only slightly tapered, and of high crystal quality (TEM images have been published elsewhere by Gotschke et al.31). In many cases, a relatively small Eg gives rise to a strong spin−orbit interaction, which is essential for spin transistor operations. But as the spin−orbit splitting in the valence band Δso has been calculated to be only about 5 meV,32 the Rashba coefficient αR ∝ [Eg−2 − (Eg + Δso)−2] is considered to be small enough so that a sufficiently large spin relaxation length for viable device dimensions can be expected. This is particularly true as the conduction band curvature in our nanowires might be reduced due to Si-doping. Experimental Section. The investigated nanowires are prepared in a plasma-assisted molecular beam epitaxy (MBE) chamber. The nanowires are grown on a Si (111) substrate using Si as an n-type dopant. The growth parameters have been given by Richter et al.,30 who also demonstrated field-effect transistor operation on our nanowires. The majority of the wires under investigation have a diameter of 100−130 nm and a length of about 2 μm. The electron density of n3d = 1.1 × 1020 cm−3 for this epitaxy run was adjusted by doping with Si. Such a high value suggests that the devices are less prone to Hall effect contributions in the local measurement geometry that might arise as manifestations of magnetic stray fields and scale

Figure 1. (a) Illustration of the planarization process and the final device geometry. The first inset depicts an InN nanowire mechanically transferred onto the substrate. Consequently, it is covered with spincoated HSQ resist, which transforms into SiO2 upon electron-beam exposure. The third inset illustrates the nanostructure after incremental reactive dry etching (using CHF3) of the surrounding oxide, leaving the nanowire partially reexposed. After each etching step the progress is monitored by AFM and scanning electron microscopy (SEM). The main image pictures the fully contacted device comprising two nonmagnetic Ti/Au-contacts and two ferromagnetic Co leads separated from the InN nanowire by an ultrathin (0.8-nm-thick) Al2O3 tunnel barrier. Bottom: Scanning electron micrograph (b) with and (c) without nanowire planarization. In part b the interruption in the Co layer adjacent to the FM-nanowire intersection is reflected in the absence of an electric contact, not to mention a continuous magnetic domain structure in the micromagnet. The result of the more elaborate preparation method in (c) proves that the continuity of the FM electrode has been enhanced significantly.

contacted laterally in a four-terminal geometry employing several steps of electron beam lithography. First, FM and nonmagnetic leads (NM) are defined in a PMMA-resist multilayer stack, which is followed by a transfer using thermal evaporation and lift-off patterning. As can be seen in Figure 1b, thin metallic layers become discontinuous when evaporated onto the nanowire due to shadowing effects. This prevents the formation of a continuous magnetic domain and thus a welldefined injection axis. To realize planar contacts in which the magnetization reversal process is unhampered, a new preparation scheme is introduced: After mechanically spreading the nanowires onto the nitrided Si (100) substrate, HSQ (hydrogen silsesquioxane) is spin-coated onto our samples. As 4438

dx.doi.org/10.1021/nl301052g | Nano Lett. 2012, 12, 4437−4443

Nano Letters

Letter

magnetization reversal by harnessing the otherwise spurious micromagnetic stray fields. The patterning was realized by reactive dry etching with CH4/H2 process gas on a Ti hard mask. Shubnikov−de Haas measurements yield a low electron density of n2DEG = 2.6 × 1011 cm−2 in the 2DEG channel. Single Co micromagnets are placed onto the Hall crosses by means of electron beam lithography and are evaporated onto an Al2O3 layer, reproducing the above conditions for nanowire contacts. Accordingly, the magnetic stray fields emanating from the ends of the FM microbars penetrate the 2DEG beneath the Hall cross intersection (see Figure 2a). We carried out lowfrequency lock-in measurements of the local Hall effect (LHE) signal relative to a magnetic field applied along the easy axis of the micromagnets (cf. Figure 2a, inset). Hence, the magnetization hysteresis of the cobalt strips is recorded (see Figure 2b) with particular sensitivity to vortex and domain wall formation at the ends of the magnets. This does not necessarily reveal the micromagnetic behavior of the entire magnet,25 but as the switching originates at the magnet ends, the reversal mechanism can be resolved more accurately than with magnetic force microscopy.39 Thus, this technique enables a detailed engineering of the magnetic properties of the Co electrodes, allowing us to choose the desired switching field and transition behavior. All measurements reported here were performed at a temperature of 4 K. Results and Discussion. Local Hall Effect. The local Hall measurements on Co electrodes of different geometries are shown in Figure 2. As can be seen here, the magnetization switching occurs at coercive fields Hc depending exclusively on the absolute width and the length-to-width aspect ratio. The corresponding values of Hc are plotted in Figure 3. X-ray ω-2Θ diffraction scans collected on Co thin films imply no preferred crystal growth direction, excluding magnetocrystalline anisotropy. This is also nicely reflected by the agreement between the measured hysteresis and the numerical simulation of the magnetization reversal using the computer simulation framework OOMMF40 (cf. Figure 3). To this end, we assumed the K1 coefficient of the magnetocrystalline anisotropy to be zero. Furthermore, the saturation magnetization for hcp cobalt amounts to MS = 1.45 × 106 A/m,41 while the exchange stiffness is known to be A = 3.5 × 10−11 J/m.42 We observe that the mechanism of magnetization reversal changes from domain wall driven to vortex driven for an aspect ratio greater than 5. Hc rapidly increases as the narrower strips impede switching of magnetization more strongly due to shape anisotropy. In addition to the aspect ratio, the absolute size has an influence on Hc,43 as reflected by measurements on the 3-μm-long micromagnets. For narrow FM electrodes, sharp transitions between two remanent magnetic states are observed owing to a singledomain configuration. This is reflected by the magnetic stray fields depicted in the magnetic force microscopy (MFM) images in the inset of Figure 3. There, the dark and light contrast corresponds to the density of magnetic poles ρ which represents the derivative of the magnetization M⃗ according to ρ = −∇·M⃗ . The hysteretic behavior found in the LHE measurements and the imaging of the magnetic remanent configuration are in accordance with the computed results obtained by OOMMF. The central inset in Figure 3 compares a simulated pole-density distribution (top) with an MFM image of an equally shaped sample (bottom). In both cases the

a consequence, the on average 120-nm-wide nanowires are coated with a 153 ± 2 nm-thick HSQ layer (determined by ellipsometry). The thickness is controlled by dilution with methyl isobutyl ketone (MIBK). Using softbake and electron beam exposure, the layer is then transformed into a wellinsulating oxidic structure resembling a SiO2-like stoichiometry. Subsequently, the nanowires are uncovered again by reactive dry etching with CHF3 process gas, leaving enough oxide that only 30 nm of the nanowire stick out of the oxidic layer. To this end, three etching steps of 20−30 s each with intermediate oxygen plasma exposure for polymer removal were carried out. In between, atomic force microscopy (AFM) was used to monitor the etching progress. Afterward, the top of the nanowires can be contacted smoothly (cf. Figure 1a and c), which is obviously not the case if the wire is covered directly with the Co layer (cf. Figure 1b). The presented planarization procedure proved beneficial for reliable metallization of nanowires with thin films (especially useful for faceted and larger nanowires). The etching process was found to be suitable for selectively removing oxidic material as the InN nanowires are hardly affected by the CHF3 plasma, as shown by AFM after long-time exposure (10 min) of uncovered wires. Not all nanowires on one substrate could be contacted perfectly at the same time due to the nonuniformity of the nanowire diameters and variations in the HSQ thickness. Thus, for many devices out-of-plane magnetization at the interface can play a significant role so that spin valve operation is precluded. Apart from the intricate sample preparation process, also the high sensitivity of the nanowires to electrostatic discharge impedes the preparation of a large number of devices. For spin injection, 60-nm-thick Co electrodes with tunnel barriers are deposited on top of the InN nanowires. First, the native InOx surface layer of the nanowire is removed by in situ Ar+ Kaufmann sputtering (p = 8 × 10−5 mbar) in an MBE chamber with a beam current of 25 mA and at an acceleration voltage of 1 kV for 6 min to achieve a well-defined tunnel barrier thickness of controllable resistance. Before the deposition of the ferromagnetic metal, 0.8-nm-thick Al2O3 tunnel barriers are electron-beam evaporated in situ inside the MBE chamber (p = 10−8 mbar) at an evaporation rate of less than 10−2 nm/s. In a second lithography step, nonmagnetic ohmic Ti/Au contacts are deposited at each end of the nanowires and onto the Co electrodes to connect the micromagnets to the bond pads for four-terminal measurements. The ultrathin tunnel barriers are crucial for the experiment. They are introduced to avoid the conductivity mismatch34 and thus increase the injection efficiency, which is typically suppressed when spin-currents are driven directly from a metallic injector into semiconductors.35−37 A tunnel barrier, dominating the spin dependent resistance of the junction, can shift the number of spin flip events toward the semiconductor side of the junction. Thus, it prevents spin depolarization of the current from occurring almost exclusively on the FM side and maintains current spin polarization across the interface. Another fundamental argument for using sufficiently nontransparent interfaces for spin injection is to prevent backscattering of injected spins into the barrier, which would result in the loss of spin information. The cobalt micromagnets were characterized extensively employing various complementary techniques. Micro Hall crosses were prepared using GaxIn1−xAs/InP heterostructures (see Schäpers et al.38 for layer sequence) to investigate the 4439

dx.doi.org/10.1021/nl301052g | Nano Lett. 2012, 12, 4437−4443

Nano Letters

Letter

Figure 3. Dependence of the coercive field Hc on the length-to-width aspect ratio of the cobalt microbars. The insets depict MFM images of the investigated microbars (orange-black) and a simulated pole-density distribution (black and white).

micromagnets) are necessary for a significant suppression of the magnetization reversal.44 Interface Transparency. A parameter of upmost importance for efficient spin injection is the interface transparency between the nanowire and the FM metal. Due to reactive ion etching and Kaufmann sputter-cleaning, the surface is not expected to be atomically smooth. Since all regions along the interface contribute in parallel to the total tunnel current, the actual tunneling thickness (in contrast to the nominal thickness) will be determined by the narrowest regions in the barrier. We employed a three-terminal configuration to deduce the contact resistance for the device shown in Figure 4a, which was utilized for the spin valve measurements described below. The boundary resistances amount to RB,1 = 3.2 kΩ and RB,2 = 9.1 MΩ for the 540-nm-wide strip and 270-nm-wide strip, respectively. While the latter value is in accordance with estimations we made on the basis of tunneling transport through Co−Al2O3−InN junctions, the resistance in the kΩ range is presumably due to leakage through barrier pinholes. The existence of pinholes is also reflected in the observed baseline voltage in the nonlocal measurements reported below. As discussed by Johnson and Silsbee,45 a baseline voltage is suggestive of inhomogeneities in the tunnel barriers of both the injector and the detector contact. The typical contact resistance was on the order of 10 MΩ, while due to the distribution of the tunneling thickness on the atomic scale some junctions yield RB as low as several hundred Ohms. Several high-resistance junctions abruptly dropped in resistance when voltages above 1 V were applied. To find out if the above contact resistances are sufficient to yield measurable spin valve signals, we employed the approach of Fert and Jaffrès36 and applied the boundary condition of nontransparent interfaces to the nonlocal measurement geometry (as already reported by Takahashi and Maekawa46). The influence of the interface resistance is discussed below. Nonlocal Spin Valve Measurements. In the four-terminal measurements depicted in Figure 4, the AC-current is driven between the ferromagnetic electrode Co1 and the normal metal contact NM1, while the spin-related signal is probed by an FM

Figure 2. (a) Schematic of the measurement setup with an illustration of the magnetic stray field lines emanating from the micromagnet and penetrating the GaxIn1−xAs/InP heterostructure mesa. A colored SEM image of a representative sample is shown in the inset. (b) LHE measurements at 4 K on rectangular Co electrodes with different aspect ratios AR. Magnet width and aspect ratio are indicated for each of the curves, which are individually symmetric with respect to zero voltage. The Hall voltage signals have been multiplied by the ratio α = wc/w between the mesa width of the voltage probe wc and the magnet width w to be able to compare the different curves.

equilibrium distributions feature diamond-like multidomain structures. Spurred by results from TEM-Lorentz-microscopy investigations,44 we also checked for influences of the magnet end shapes. As magnetic vortex formation is initiated in the magnetic end domains, a geometrical reduction of the number and size of terminal domains is expected to raise the coercivity of the cobalt microbars. However, no impact on the switching properties can be observed for rectangular and rounded cobalt microbars of identical aspect ratios in either the LHE measurements or in the simulation results. The increase in coercitivity accounts for at most 1 mT. Because of the negligible difference, a discrimination between the two termination shapes was omitted in Figure 3, so that the aspect ratio and the absolute width are the only parameters there. Obviously, more drastic shape modifications (i.e., needle-like 4440

dx.doi.org/10.1021/nl301052g | Nano Lett. 2012, 12, 4437−4443

Nano Letters

Letter

±57 mT are expected, when considering the LHE measurements in Figure 3. The switching events observed here reproducibly match the predictions of the above LHE characterization, in spite of the distortions the nanowire imposes on the micromagnet. The asymmetries in the switching fields with respect to B = 0, depicted in Figure 2, were also found in the spin valve measurements. One way to understand these small shifts is to consider a possible exchange bias47 caused by the anti-FM/FM-interaction between the cobalt and the thin cobalt oxide layer that forms on top of the micromagnets. Additionally, magnetization pinning of domain walls at geometrical variations at the nanowire intersection48 may impact the switching. In the nonlocal spin valve measurement setup, we expect parasitic ohmic contributions to be eliminated, apart from a baseline signal.45 Hence, we attribute the observed MR change to spin transport only. The measured MR signals thus demonstrate the generation of spin accumulation in InN nanowires. The electrochemical potentials inside the semiconductor μF and μN, gauged by the ferromagnetic detector contact and the nonmagnetic contact, respectively, yield a magnetovoltage drop of46 μ − μN ΔVnl = 2 × F e ⎛ L⎞ = 4RNI × exp⎜ − ⎟ ⎝ lsf ⎠ P R ⎤ 2 ⎡ P R ∏i = 1 ⎣⎢ i 2 RB,i + b 2 RF,i ⎦⎥ 1 − Pi N 1 − Pb N × 2 RB, i 2 RF, i ⎤ 2L 2 ⎡ ∏i = 1 ⎣⎢1 + + ⎥ − exp − lsf 1 − Pi2 RN 1 − Pb2 RN ⎦

Figure 4. (a) Colored scanning electron micrograph of an InN nanowire contacted after planarization. The nonlocal measurement geometry is depicted, featuring a FM-contact separation of L = 200 nm. The magnetic field is applied along the easy axis of the magnets, each switching at individual coercivities owing to different aspect ratios. The AC drive current is sent through the contacts Co1 and NM1. Outside of the charge current path, the open-circuit voltage is probed between the contacts Co2 and NM2. (b) Nonlocal spin valve measurements on an InN nanowire at T = 4 K. A 5.5 mV offset with a slowly varying background has been subtracted, indicative of inhomogeneous lateral injection and detection, respectively.45 The magnitude of the effect can be quantified by the nonlocal resistance ΔRnl = ΔVnl/I ≈ 0.16 Ω for an injection current of Iinj = 2.5 μA and ΔRnl ≈ 0.07 Ω for Iinj = 0.5 μA.

( )

(1)

wherein RN =

spin detector by measuring the open-circuit voltage drop between Co2 and NM2. Between Co2 and NM2, the spinpolarized electrons travel diffusively through the nanowire without any net charge transport involved. The MR curves are obtained during a magnetic field sweep along the Co contacts easy axis and measuring the magnetovoltage drop at an applied AC bias current of 2.5 μA in a low-frequency (16 Hz) lock-in setup. The samples exhibit a negative MR change ΔR/R0 with respect to the initial resistance at high fields. When sweeping the magnetic field between ±300 mT, we observe pronounced changes in the nonlocal MR at certain field values. As can be inferred from Figure 3, the magnetization reversal of Co1 and Co2 occur at different coercive fields. In the regions of negative MR in Figure 4b the spins, which are injected into the nanowire channel and the spin detector Co2 are oppositely magnetized, while at high fields the second switch returns the system to the parallel state. The magnetic switching fields of the FM electrodes are taken from the extremum of the derivative of the MR signal and averaged over several magnetic field sweeps, which reproducibly exhibit these spin signals. As expected for very narrow magnetic leads, abrupt resistance changes associated with transitions between the two magnetization states of both Co electrodes occur at +41 ± 1 mT and at −52 ± 2 mT for the wider and the narrower micromagnet, respectively. Switching fields of approximately ±35 mT and

lsf σInNAN

and

RF, i =

lsfF σFAB,i

(2)

with the interface polarization of the two ferromagnets Pi and the bulk Co spin polarization Pb. Moreover, RN and RF,i describe characteristic resistances of the semiconductor and the ferromagnets, respectively, comprising the conductivities σInN and σF and the spin relaxation lengths lsf and lFsf of the nanowire and the Co lead. AB,i denote the respective FMnanowire interface cross-section, whereas AN denotes the nanowire cross-section. Our aim was to prove that, by engineering the interface resistance, spin injection into a nanowire of strong spin−orbit coupling is feasible. It has been shown49 that such circular 2DEG-devices can be utilized for spintronics applications. We are aware that extensive contact separation-dependent investigations and spin-precession measurements in a Hanle setup are still needed. While spin valve measurements thus only yield the spin relaxation length as a function of P, we want to estimate lsf by invoking polarization from literature. Zaffalon and van Wees50 report on Co contacts with thin layers of oxidized Al underneath for spin injection into the aluminum channel with an interface polarization of the FM of about P = 8%. This is in agreement with the results on similar devices by Valenzuela and Tinkham51 and Jedema et al.52 with P in the order of 10%. Münzenberg and Moodera53 discuss reasons for the drastic reduction of interface polarization with respect to 4441

dx.doi.org/10.1021/nl301052g | Nano Lett. 2012, 12, 4437−4443

Nano Letters

Letter

resistance change was observed. If the bias current in the injection circuit is too small, the nonlocal voltage change falls below the measurement resolution. Therefore, we are required to apply a current on the order of 1−3 μA in the measurements in Figure 4b to achieve a significant magnetovoltage change on the order of about 0.1 μV. If the tunnel resistance is too large, we cannot apply high currents, since the breakthrough voltage is reached and moreover the linear tunneling regime is exceeded above a certain bias voltage. We conjecture that the reduced barrier resistance due to the variation of the tunneling thickness or barrier pinholes does not preclude spin injection as long as the contact resistance exceeds a threshold resistance of about 1 kΩ and tunneling is the dominating transport mechanism. Furthermore, Fert and Jaffrès36 have shown that for local spin valve measurements, the requirement that the dwell time of the electrons in the semiconductor falls short of the spin lifetime sets an upper limit to the tunnel barrier resistance. We have as well observed a pronounced spin valve-like behavior for several devices in the ordinary local measurement geometry, but without demonstrating nonlocal spin injection as shown for the device in Figure 4. Since the signals are also in agreement with stray field induced LHE, AMR, or other spurious effects, these measurements can only give an initial indication for spindependent transport. No conclusions can be drawn unless the injection and detection circuits have been separated completely. Petersen et al.54 determined lsf from weak antilocalization measurements for undoped MBE-grown InN nanowires to be 106 nm at T = 4 K. As it was argued by the authors54 that this number is presumably an underestimation of the actual lsf in undoped InN nanowires, and since additionally the high carrier concentration for doped nanowires should reduce the Rashba effect related contribution to spin relaxation, our first estimate appears to be reasonable. Summary. In conclusion, we have successfully demonstrated spin accumulation in a III-V semiconductor nanowire for the first time. The reasoning for this claim is the reproducible observation of nonlocal spin signals, where MR changes could be associated with the micromagnetic reversal of the FM leads. These were characterized comprehensively beforehand, in particular by stray field measurements. The aspect-ratio dependence of the coercivities of the Co electrodes has been quantified accordingly. The nonlocal setup makes sure that ohmic contributions from the FM leads themselves do not impact the resistance switching, ensuring an exclusive spin effect. We have presented a circumvention of the challenges pointed out by Monzon et al.23 by introducing a novel planarization technique for metallization of nanostructures with thin ferromagnetic electrodes. It has been shown that semiconductor nanowires can be employed for the propagation of spin information across transistor channels since the spin relaxation length of around 200 nm is in the range of viable device dimensions. Still, this value is comparatively small, indicative of the strong influence of spin−orbit interaction, which is a premise for possible applications in spin-based devices. In future studies, we seek to examine the gate dependence and aim at improving control over the interface resistance.

the bulk value for ferromagnetic metals in the limit of ultrathin barriers. It is argued that this polarization reduction is further enhanced by barrier pinholes. Valenzuela and Tinkham51 could rule out the existence of pinholes by magnetic tunnel junction measurements on their devices. In all of the aforementioned publications, tunnel barriers were formed by in situ oxidation of thin Al films, while in our case, Al2O3 was evaporated in situ in an MBE chamber. As we suspect that transport through narrower regions in the tunnel barrier or pinholes (indicated by the baseline voltage45 of 5.5 mV) might reduce the interface polarization,53 we assume 8% spin current polarization as an upper limit. In conjunction with the assumed Pi and the measured RB,i, the MR change of ΔRnl = ΔVnl/I ≈ 0.16 Ω observed here yields lsf ≈ 130 nm from eq 1 as a lower threshold for the spin relaxation length at T = 4 K in our ndoped InN nanowires. Even if the assumed interface polarizations were overrated, this gives a good estimate as lsf does not exceed 200 nm for Pi down to 5%. By invoking the density of states at the Fermi energy N(EF) = 0.07me(3π2n3d)1/3/(π2ℏ2) one can employ the Einstein relation to obtain the diffusion constant D = σ/[e2N(EF)]. Hence, with D = 4.5 × 10−3 m2/s, the corresponding spin relaxation time τsf = l2sf/D is below 10 ps. These results are corroborated by the fact that the spin signal scales with the bias current in the separated injection circuit. In Figure 4 b, a second spin valve measurement is depicted, where the nonlocal bias-current was reduced to 0.5 μA. The magnetoresistance is much less pronounced, but the signal height of ΔRnl ≈ 0.07 Ω also yields lsf ≈ 100−140 nm for the aforementioned polarization range. Considering the reduced signal-to-noise ratio, this appears to be in reasonable agreement with the result from the first measurement. The obtained value of lsf together with L and Pi, as used above, can be employed to evaluate the impact of the interface on bipolar switching. The calculation for the InN nanowire device according to eq 1 is depicted in Figure 5, which shows a

Figure 5. Dependence of the nonlocal spin valve signal on the interface resistance RB = (RB,1 + RB,2)/2 between the nanowire and the FM metal according to eq 1 for conductivities σInN = 1.0 × 10−3 Ω·cm and σF = 5.9 × 10−6 Ω·cm, interface polarizations Pi = 8%, lsf = 130 nm, lFsf = 59 nm, Pb = 46%,36 and channel length L = 200 nm. The lower RB,i determines the resulting magnetovoltage drop.

significant MR occurring for interface resistances beyond the threshold of 1 kΩ. There, an average boundary resistance of RB = (RB,1 + RB,2)/2 is assumed as the only parameter. To classify the actual interfaces of our sample, RB,i has been designated for Co1 and Co2. For the purpose of spin injection, it is crucial that the FM contacts are within the required resistance range. For samples with small RB of several hundred Ohms (which did not show tunneling I−V characteristics), no distinct magneto-



AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected]; [email protected]. 4442

dx.doi.org/10.1021/nl301052g | Nano Lett. 2012, 12, 4437−4443

Nano Letters

Letter

Present Address

(25) Mennig, J.; Matthes, F.; Bürgler, D. E.; Schneider, C. M. J. Appl. Phys. 2012, 111, 07C504. (26) Lau, W. H.; Olesberg, J. T.; Flatté, M. E. Phys. Rev. B 2001, 64, 161301. (27) Richter, T.; Blömers, Ch.; Lüth, H.; Calarco, R.; Indlekofer, M.; Marso, M.; Schäpers, Th. Nano Lett. 2008, 8, 2834−2838. (28) Kiselev, A. A.; Kim, K. W. Phys. Rev. B 2000, 61, 13115−13120. (29) Davydov, V.; Klochikhin, A.; Emtsev, V.; Ivanov, S.; Vekshin, V.; Bechstedt, F.; Furthmüller, J.; Harima, H.; Mudryi, A.; Hashimoto, A.; Yamamoto, A.; Aderhold, J.; Graul, J.; Haller, E. Phys. Status Solidi B 2002, 230, R4−R6. (30) Richter, T.; Lüth, H.; Schäpers, Th.; Meijers, R.; Jeganathan, K.; Estévez Hernández, S.; Calarco, R.; Marso, M. Nanotechnology 2009, 20, 405206. (31) Gotschke, T.; Schäfer-Nolte, E. O.; Caterino, R.; Limbach, F.; Stoica, T.; Sutter, E.; Jeganathan, K.; Calarco, R. Nanotechnology 2011, 22, 125704. (32) Vurgaftman, I.; Meyer, J. R. J. Appl. Phys. 2003, 94, 3675−3696. (33) Tang, H.; Monzon, F.; Roukes, M.; Jedema, F.; Filip, A.; van Wees, B. Semiconductor Spintronics and Quantum Computation; Springer-Verlag: Berlin, 2002. (34) Schmidt, G.; Ferrand, D.; Molenkamp, L. W.; Filip, A. T.; van Wees, B. J. Phys. Rev. B 2000, 62, R4790−R4793. (35) Rashba, E. I. Phys. Rev. B 2000, 62, R16267−R16270. (36) Fert, A.; Jaffrès, H. Phys. Rev. B 2001, 64, 184420. (37) Heersche, H. B.; Schäpers, Th.; Nitta, J.; Takayanagi, H. Phys. Rev. B 2001, 64, 161307. (38) Schäpers, Th.; Guzenko, V. A.; Hardtdegen, H. Appl. Phys. Lett. 2007, 90, 122107. (39) Nitta, J.; Schäpers, Th.; Heersche, H. B.; Koga, T.; Sato, Y.; Takayanagi, H. Jpn. J. Appl. Phys. 2002, 41, 2497−2500. (40) Donahue, M. J.; Porter, D. G. OOMMF Users’s Guide, Version 1.0; Interagency Report NISTIR 6376; NIST: Gaithersburg, MD, 2002. (41) Myers, E. B.; Ralph, D. C.; Katine, J. A.; Louie, R. N.; Buhrman, R. A. Science 1999, 285, 867−870. (42) Belashchenko, K. D. J. Magn. Magn. Mater. 2004, 270, 413−424. (43) Last, T.; Hacia, S.; Wahle, M.; Fischer, S. F.; Kunze, U. J. Appl. Phys. 2004, 96, 6706−6711. (44) Preusche, D.; Schmidmeier, S.; Pallecchi, E.; Dietrich, Ch.; Hüttel, A. K.; Zweck, J.; Strunk, Ch. J. Appl. Phys. 2009, 106, 084314. (45) Johnson, M.; Silsbee, R. H. Phys. Rev. B 2007, 76, 153107. (46) Takahashi, S.; Maekawa, S. Phys. Rev. B 2003, 67, 052409. (47) Berkowitz, A.; Takano, K. J. Magn. Magn. Mater. 1999, 200, 552−570. (48) Kläui, M. J. Phys.: Condens. Matter 2008, 20, 313001. (49) Bringer, A.; Schäpers, Th. Phys. Rev. B 2011, 83, 115305. (50) Zaffalon, M.; van Wees, B. J. Phys. Rev. B 2005, 71, 125401. (51) Valenzuela, S. O.; Tinkham, M. Appl. Phys. Lett. 2004, 85, 5914−5916. (52) Jedema, F. J.; Heersche, H. B.; Filip, A. T.; Baselmans, J. J. A.; van Wees, B. J. Nature 2002, 416, 713−716. (53) Münzenberg, M.; Moodera, J. S. Phys. Rev. B 2004, 70, 060402. (54) Petersen, G.; Estévez Hernández, S.; Calarco, R.; Demarina, N.; Schäpers, Th. Phys. Rev. B 2009, 80, 125321.



Paul-Drude-Institut für Festkörperelektronik, Hausvogteiplatz 5-7, 10117 Berlin, Germany. Notes

The authors declare no competing financial interest.



ACKNOWLEDGMENTS We are grateful to Th. Jansen for MBE operation and metal deposition. J. Gerharz provided valuable advice during the sample preparation, and St. Trellenkamp performed electron beam lithography on our samples. Funding for this work is provided by the German Research Foundation (DFG) through an FOR 912 grant.



REFERENCES

(1) Holleitner, A. W.; Sih, V.; Myers, R. C.; Gossard, A. C.; Awschalom, D. D. Phys. Rev. Lett. 2006, 97, 036805. (2) Schäpers, Th.; Guzenko, V. A.; Pala, M. G.; Zülicke, U.; Governale, M.; Knobbe, J.; Hardtdegen, H. Phys. Rev. B 2006, 74, 081301. (3) Dash, S. P.; Sharma, S.; Patel, R. S.; de Jong, M. P.; Jansen, R. Nature 2009, 462, 491−494. (4) Jedema, F. J.; Filip, A. T.; van Wees, B. J. Nature 2001, 410, 345− 348. (5) Zwanenburg, F. A.; van der Mast, D. W.; Heersche, H. B.; Kouwenhoven, L. P.; Bakkers, E. P. A. M. Nano Lett. 2009, 9, 2704− 2709. (6) Liu, E.-S.; Nah, J.; Varahramyan, K. M.; Tutuc, E. Nano Lett. 2010, 10, 3297−3301. (7) Tarun, J.; Huang, S.; Fukuma, Y.; Idzuchi, H.; Otani, Y.; Fukata, N.; Ishibashi, K.; Oda, S. J. Appl. Phys. 2011, 109, 07C508. (8) Kumar, A.; Ghosh, B. ArXiv e-prints 2011, arXiv:1106.4378 [cond-mat.mes-hall]. (9) Sahoo, S.; Kontos, T.; Schönenberger, C.; Sürgers, C. Appl. Phys. Lett. 2005, 86, 112109. (10) Tsukagoshi, K.; Alphenaar, B. W.; Ago, H. Nature 1999, 401, 572−574. (11) Monzon, F. G.; Roukes, M. L. J. Magn. Magn. Mater. 1999, 198−199, 632−635. (12) Tombros, N.; van der Molen, S. J.; van Wees, B. J. Phys. Rev. B 2006, 73, 233403. (13) Johnson, M.; Silsbee, R. H. Phys. Rev. Lett. 1985, 55, 1790− 1793. (14) Tombros, N.; Jozsa, C.; Popinciuc, M.; Jonkman, H. T.; van Wees, B. J. Nature 2007, 448, 571−574. (15) Feuillet-Palma, C.; Delattre, T.; Morfin, P.; Berroir, J.-M.; Fève, G.; Glattli, D. C.; Plaçais, B.; Cottet, A.; Kontos, T. Phys. Rev. B 2010, 81, 115414. (16) Crooker, S. A.; Furis, M.; Lou, X.; Adelmann, C.; Smith, D. L.; Palmstrøm, C. J.; Crowell, P. A. Science 2005, 309, 2191−2195. (17) Lou, X.; Adelmann, C.; Crooker, S. A.; Garlid, E. S.; Zhang, J.; Reddy, K. S. M.; Flexner, S. D.; Palmstrøm, C. J. J.; Crowell, P. A. Nat. Phys. 2007, 3, 197−202. (18) Ciorga, M.; Einwanger, A.; Wurstbauer, U.; Schuh, D.; Wegscheider, W.; Weiss, D. Phys. Rev. B 2009, 79, 165321. (19) Koo, H. C.; Kwon, J. H.; Eom, J.; Chang, J.; Han, S. H.; Johnson, M. Science 2009, 325, 1515−1518. (20) Chang, J.; Koo, H. C.; Eom, J.; Han, S. H.; Johnson, M. J. Appl. Phys. 2011, 109, 102405. (21) Lin, Y.-C.; Chen, Y.; Shailos, A.; Huang, Y. Nano Lett. 2010, 10, 2281−2287. (22) Datta, S.; Das, B. Appl. Phys. Lett. 1990, 56, 665−667. (23) Monzon, F. G.; Tang, H. X.; Roukes, M. L. Phys. Rev. Lett. 2000, 84, 5022. (24) Hammar, P. R.; Bennett, B. R.; Yang, M. J.; Johnson, M. Phys. Rev. Lett. 1999, 83, 203−206. 4443

dx.doi.org/10.1021/nl301052g | Nano Lett. 2012, 12, 4437−4443