Extremely long quasiparticle spin lifetimes in superconducting aluminium using MgO tunnel spin injectors

Extremely long quasiparticle spin lifetimes in superconducting aluminium using MgO...

(Parte 1 de 3)

Extremely long quasiparticle spin lifetimes in superconducting aluminium using MgO tunnel spin injectors

There has been an intense search in recent years for long-lived spin-polarized carriers for spintronic and quantum-computing devices. Here we report that spin-polarized quasiparticles in superconducting aluminium layers have surprisingly long spin lifetimes, nearly a million times longer than in their normal state. The lifetime is determined from the suppression of the aluminium’s superconductivity resulting from the accumulation of spin-polarized carriers in the aluminium layer using tunnel spin injectors. A Hanle effect, observed in the presence of small in-plane orthogonal fields, is shown to be quantitatively consistent with the presence of long-lived spin-polarized quasiparticles. Our experiments show that the superconducting state can be significantly modified by small electric currents, much smaller than the critical current, which is potentially useful for devices involving superconducting qubits.

The interplay of magnetism and superconductivity has long been of great interest1–3. The influence of magnetism on superconductivity has been studied by the proximity effect, that is, the placement of magnetic materials adjacent to superconducting materials, on various lengths scales from the macroscopic, where magnetostatic fields are important, to the microscopic, where exchange effects become relevant, strongly suppressing superconductivity4–8. These exchange effects are mitigated by the insertion of ultrathin tunnel barriers between the ferromagnet and the superconductor9–1, allowing spin-polarized carriers to be injected into the superconductor. By adding a second such insulator–ferromagnet combination on the far side of the superconductor, spin-polarized carriers can be accumulated in the superconductor. The accumulation is predicted to take place only when the ferromagnetic layer moments are antiparallel to each other, and, if large enough, to lead to a suppression of the superconducting energy gap12. Here we show that in large-area planar double magnetic tunnel junctions formed with MgO tunnel barriers, a highly efficient spin injector13,14, the superconducting energy gap is indeed reduced in aluminium for the antiparallel configuration.Thisleadstoanoscillatoryvariationofthetunnelling magnetoresistancewithbiasvoltagefortemperatureswellbelowthe superconducting transition temperature of the Al layer. Modelling of the experimental results shows that these results can only be accounted for by a marked enhancement of the quasiparticle spin lifetime τS in the superconducting state by several orders of magnitude as compared with the normal state.

The spin relaxation time in the normal state of Al (τN) has been intensively studied with measured values ranging from 0.1 to 10ns depending on the film thickness and quality15–18. However, the experimental situation for the superconducting (SC) state is less clear with one group inferring a shorter spin lifetime in the SC state19

1IBM Almaden Research Center, 650 Harry Road, San Jose, California 95120, USA, 2Institute for Materials Research, Tohoku University, Sendai 980-8577, Japan, 3CREST, Japan Science and Technology Agency (JST), Tokyo 102-0075, Japan, 4Advanced Science Research Center, Japan Atomic Energy Agency, Tokai 319-1195, Japan. †These authors contributed equally to this work. ‡Present address: Department of Electrical and Computer Engineering, National University of Singapore, 117576 Singapore. *e-mail:parkin@almaden.ibm.com.

than in the normal state and others assuming that the spin lifetime isunchanged20,21.Theoretically,itisanticipatedthatτS shouldbeincreased because the group velocity of quasiparticles injected into the bottom of the quasiparticle conduction band just above the superconducting energy gap, is small, thereby leading to reduced interaction probabilities. This can be restated more clearly as being a result of the spin bottleneck to quasiparticle relaxation resulting from spin-charge separation22,23 at the energy gap edge, where the quasiparticles have only spin and no charge, leading to reduced spin– orbitscatteringandthereforeanenhancedspinrelaxationtime24.

Experiments on the interplay between spin accumulation and superconductivityhavebeencarriedoutbytwogroupsusingsingleelectron transistor (SET) devices consisting of small Al islands electrically connected on their top surface by closely and laterally spaced ferromagnetic Co nanowire electrodes20,25,26. However, the single-electron transistor experimental configuration has led to controversy over their interpretation because of magnetic fringing fields from the Co (refs 21,26,27). The effect of fringing fields on the superconducting properties is especially important for thicker SC Al layers28. Here we use a vertical geometry and planar tunnel junctions with large areas ∼700×700µm2 so that fringing fields from the Co70Fe30 ferromagnetic (F) electrodes are negligibly small. Large areas are also used so that the tunnel barriers, formed from

MgO, could be as thick as possible to ensure the absence of any exchange proximity effect on the superconducting Al layer from the F electrodes. The double tunnel junction (DTJ) devices were fabricated by dc magnetron sputter deposition using a series of four in situ shadow masks to form a junction, as schematically shown in Fig. 1c. The direction of the magnetization of the lower F electrode wasfixedusingexchangebiasfromanIrMnantiferromagneticlayer on which the F layer was deposited. The exchange bias field was ∼1,500Oe for the temperature range of interest (∼0.25–2.5K).

586 NATURE MATERIALS | VOL 9 | JULY 2010 | w.nature.com/naturematerials


2.5 K

2.25 K 2.25 K

2 K 2 K

1 K 1 K

0.25 K

0.5 K 0.5 K

0.25 K

TMR (%)dI/d V (a.u.)

Bias voltage (mV)

2.5 K 20%

T = 0.25 K

Bias voltage (mV) c d

Al MgO CoFe bP(AP) = 0.02 P = 0.046

AP = 0.117 PLT = 7.7 μeV

APLT = 7.6 μeV ¬60

Figure 1 | Experimental conductance and TMR data and comparison with model. a,b, Normalized conductance (a) and TMR curves (b) as a function of bias voltage for various temperatures. The structure is composed of CoFe/3.3nm MgO/4.5nm Al/3.3nm MgO/CoFe, for which ΓN is calculated to be ∼2×106 using values of RTA∼108 µm2, ρN ∼10µ cm and λN ∼1µm (refs 42,43). The blue (red) line corresponds to the conductance in the P (AP) configuration. The Tc of the Al layer is ∼2.3K. For clarity, the data at each temperature are offset from each other by 0.5 in a and 25% in b. c, Illustration of a planar DTJ structure. d, Comparison of calculated TMR versus bias voltage (solid lines) with the experimental data (open circles) at 0.25K. The magenta line corresponds to a calculation without consideration of depairing13 (ζP(AP) =0) or quasiparticle lifetime broadening28 (ΓP(AP) LT =0) effects. The dark blue line includes these effects with corresponding fitting parameters for the P(AP) states shown in the figure.

Thus, the moment of the upper F electrode could be independently oriented parallel (P) or antiparallel (AP) to the lower electrode’s momentinsmallmagneticfields(∼±300Oe)tocreatewell-defined magneticconfigurationsoftheDTJ.Thesemagneticfieldsalsoserve to negate the effect of any small perturbing orthogonal in-plane magnetic fields that might arise from, for example, inhomogeneities in the magnetization of the electrodes. Such fields could otherwise depolarize the accumulated spins especially for very long spin lifetimes (see the Methods section).

Conductance measurements were carried out in a 3He refrigerator using standard ac lock-in four-probe techniques for a series of samples in which the MgO and Al layer thicknesses were varied (see the Methods section). Typical bias voltage (V) dependencies of the conductance in the P and AP states (GP and GAP, respectively) are shown in Fig. 1a for a DTJ with a MgO thickness of ∼3.3nm and an Al layer thickness of 4.5nm. Data are shown for several temperatures varying from 0.25 to 2.5K. The conductance data at

0.25K show evidence for a high-quality superconducting tunnel junction with a well-defined superconducting energy gap. A distinctive feature of these data is that the superconducting energy gap is slightly smaller for the AP as compared with the P magnetic configuration. The difference between the SC energy gaps in the AP and P states (∆AP and ∆P, respectively) decreases with increasing temperature. The data also show a reduced quasiparticle peak intensity near V/2 ∼ ±∆AP/e for the AP state. From the difference in the conductance curves, we can define a tunnelling magnetoresistance, TMR = (GP − GAP)/GAP, which is plotted in Fig. 1b as a function of bias voltage at various temperatures. The

TMR data at 0.25K exhibit a distinctive oscillatory dependence on the bias voltage. In particular, as the bias voltage is increased from zero, for either positive or negative V, a large negative TMR is observed with a peak value ∼−5% at ∼±0.35mV. The TMR changes sign and becomes positive (∼+23% at ∼±0.58mV)

NATURE MATERIALS | VOL 9 | JULY 2010 | w.nature.com/naturematerials 587


SCF1 F2b

eV/2 =AP

eV/2 = P ¬ P

GP < GAPV = 2 AP/e

002 P2 APΔ Δ


Figure 2 | Schematic diagram of superconducting gap suppression and spin accumulation in a DTJ composed of two ferromagnetic electrodes and a superconducting middle electrode. Energy dependence of the spin-polarized density of states in F1, F2 (two ferromagnetic electrodes) and SC (a superconducting middle electrode) when the SC layer is in its superconducting state. Blue and red correspond to majority and minority spin-polarized density of states, respectively. The P and AP configurations of F1 and F2 are shown in the left and right columns, respectively. The dashed lines represent the electrochemical potential (ECP): blue and red dashed lines correspond to the ECPs of the up and down spins in SC. When a voltage V is applied between F1 and F2, spin-polarized tunnel current flows across the junctions. In the P configuration, the up (down)-spin currents at the left and right junctions are balanced with each other, and the ECPs of the up and down spins have no shift, so that there is no suppression of the energy gap. In the AP configuration, the up (down)-spin currents at the junctions are imbalanced to yield the ECP shift by ±δµ as indicated by the blue and red dashed lines, by which spin accumulation takes place as shown by quasiparticle electrons with up spins and holes with down spins, so that the gap is suppressed, that is,

∆AP <∆P. a, When eV/2=∆AP, the tunnel current in the P configuration is smaller than in the AP configuration because of ∆P >eV/2, so that GP <GAP and the TMR is negative. b, When eV/2=∆P, the tunnel current in the P configuration is larger than in the AP configuration because of eV/2>∆AP, so that GP >GAP and the TMR is positive.

As the temperature is increased, the TMR decreases to zero as the temperature approaches the superconducting transition temperatureoftheAllayer(Tc ∼ 2.3K),whichshowsthattheTMR originates from the superconducting state of the Al layer. Similar results were obtained in several devices in which the MgO thickness was varied, as discussed later.

A ferromagnet in direct contact with a superconducting layer causes suppression of the superconductivity through the exchange interaction at the interface4,6,7. When superconducting layers, thin compared with their corresponding coherence length, are sandwiched between F layers, it has been found that Tc (as inferred from resistance versus temperature measurements) is suppressed for both the P and AP configurations of the F layers, but that the suppression is slightly greater for the P configuration4,6,7. This is contrary to our results. Indeed, by inserting a leak-proof thick tunnelling barrier between the F and SC layers, as in our experiments, the proximity effect can be ruled out1.

Our results are consistent with a model in which nonequilibrium spin density is accumulated in the SC layer when spin-polarized current is injected through tunnel barriers from F electrodes in the AP configuration12. As illustrated in Fig. 2 no spin density SP is accumulated in the P configuration, but spin density SAP is formed in the SC in the AP state when the spin-relaxation time of the spin-polarized quasiparticles is sufficiently long. This is because the tunnelling is spin dependent: for CoFe/MgO/Al the tunnelling current is strongly majority spin-polarized so that for the symmetric DTJs used here, electrons can readily tunnel into and out of the SC Al layer in the P but not in the AP state (see

Fig. 2). The magnitude of SAP clearly depends sensitively on the spin relaxation time τS for a given spin injection rate. The lower the spin injectionrate,thelongerthespinrelaxationtimeneededtoestablish significant SAP. We will show in the following that the magnitude of τS must be extremely large to account for our data. The accumulated spin density S in the SC arises from an imbalance between the populations of the up-spin and down-spin quasiparticles, which corresponds to a small shift in their chemical potentials by ±δµ from their equilibrium values (see Fig. 2). In the presence of spin–orbit scattering of the quasiparticles by impurities or grain boundaries in the SC layer, the accumulated spin density S is given by (see Supplementary Information for details),

588 NATURE MATERIALS | VOL 9 | JULY 2010 | w.nature.com/naturematerials


e dc b N

Bias voltage (mV)

S (s)

b = 0.1 b = 0.02 b = 0.005

S (s) τ

(Parte 1 de 3)