All-spinel oxide Josephson junctions for high-efficiency spin filtering

Obtaining high efficiency spin filtering at room temperature using spinel ferromagnetic tunnel barriers has been hampered by the formation of antiphase boundaries due to their difference in lattice parameters between barrier and electrodes. In this work we demonstrate the use of LiTi2O4 thin films as electrodes in an all-spinel oxide CoFe2O4-based spin filter devices. These structures show nearly perfect epitaxy maintained throughout the structure and so minimise the potential for APBs formation. The LiTi2O4 in these devices is superconducting and so measurements at low temperature have been used to explore details of the tunnelling and Josephson junction behaviour.


Introduction
The broad spectrum of electronic and optical properties exhibited by functional oxides offers many opportunities for microelectronic devices. In particular, the experimental growth of epitaxial oxide heterostructures has increased the development of promising novel functionalities and device concepts [1]. However, the integration of complex oxides into multilayer structures is often challenging. Lattice mismatch, structural differences and different optimum growth conditions between the oxide layers hamper the epitaxial growth of heterostructures. Of particular relevance to this paper, ultrathin films of ferromagnetic insulators (FIs) oxides can form tunnel barriers that generate nearly 100% spin-polarised currents by selectively filtering electrons according to their spins [2]. This spin-filtering process is in contrast to the classic magnetic tunnel junctions in which a non-magnetic tunnel barrier is sandwiched between two ferromagnetic electrodes [3].
The likely reason for such low values is the formation of antiphase boundaries (APBs) in the spinel thin film barrier [18,19], which are detrimental to spin-filter efficiency as they dramatically affect magnetic behaviour and barrier height [20,21]. Such defects are formed due to spinels having a lattice parameter (a = 0.83-0.85 nm) [22], almost double that of the metallic layers (Au, Pt, LaNiO 3 , La 2/3 Sr 1/3 MnO 3 ) and substrates conventionally used in spin-filter devices. Achieving high spin-filter efficiencies at room temperature may therefore be dependent on overcoming structural and chemical defects in ultra-thin (<5 nm) epitaxial spinel ferrites films to be used in complex oxide heterostructures.
APBs can be reduced by using a spinel structure substrate (MgAl 2 O 4 ) and LiTi 2 O 4 as non-magnetic electrodes for a spin filter tunnel junction [23]. One of the few conducting spinels, LiTi 2 O 4 is a metallic and superconducting [24] spinel (critical temperature T C ~ 13 K) with a lattice parameter (a = 0.8405 nm) closely-lattice matched to those of the spinel CoFe 2 O 4 ferrite and of the spinel MgAl 2 O 4 (a = 0.8080 nm) substrate. The lattice mismatch to the latter is -3.8% while to CoFe 2 O 4 (a = 0.8392 nm) [22] is only +0.2%.
The growth of high quality single crystal oxide thin films by pulsed laser deposition (PLD) depends on the oxygen partial pressure P O2 in the chamber [23,[25][26][27]. LiTi 2 O 4 has a spinel structure with equal numbers of trivalent and quadrivalent Ti cations and for P O2 higher than 1 × 10 −5 Torr, Ti 3+ ions readily oxidise to Ti 4+ , leading to the formation of Li 4 Ti 5 O 12 , a transparent insulator phase [26]. Conversely, oxygen deficiencies are deleterious to the magnetic properties of spinel ferrite thin films [28] because the oxygen ions mediate the superexchange interaction between the magn etic ions in the spinel structure, producing the net magnetic moment in the ferrites. Thus any oxygen deficiency due to a growth at low P O2 , reduces the exchange interaction between the magnetic ions, and hence, the saturation magnetization of the CoFe 2 O 4 films. As a consequence, integrating LiTi 2 O 4 into spinel ferrite-based spin filter junctions requires a fine tuning of the growth conditions of these two materials, requiring very different oxygen partial pressures.
In this paper we demonstrate the successful growth of CoFe 2 O 4 /LiTi 2 O 4 bilayers in which LiTi 2 O 4 maintains its metallic and superconducting properties and CoFe 2 O 4 its insulating ferromagnetic characteristics. LiTi 2 O 4 /CoFe 2 O 4 / LiTi 2 O 4 trilayers were processed into all-spinel oxide symmetric superconductor-insulator-superconductor (SIS) tunnel junctions. The measured current-voltage characteristics show conclusive evidence of the tunnel nature of these junctions, proving that LiTi 2 O 4 can be used as bottom electrode in an almost APBs free tunnel junction.

Methods
LiTi 2 O 4 and CoFe 2 O 4 thin-films were grown by pulsed laser ablation of polycrystalline ceramic targets prepared from a mixture of Li 2 CO 3 (Alfa-Aesar) and TiO 2 (Alfa-Aesar), for Li 4 Ti 5 O 12 [29], and from cobalt iron oxide nanopowders (Sigma-Aldrich), for CoFe 2 O 4 . The higher Li/Ti ratio (0.8) of the Li 4 Ti 5 O 12 target was designed to compensate for the high loss of Li during the ablation process [30]. The PLD system (KrF excimer, λ = 248 nm) was operated at an energy density of 0.7 J cm −2 and at a repetition rate of 5 Hz for LiTi 2 O 4 , and 2.5 J cm −2 and 1 Hz and for CoFe 2 O 4 .
Structural analysis was done using x-ray diffraction (XRD, PANalytical high resolution x-ray diffractometer) with monochromatised CuK 1 radiation (0.154 nm). The deposition rate was determined by measuring the thickness of ultra-thin films by x-ray reflectivity analysis, allowing the controlled deposition of thicker films. The films' transport measurements were performed by four-wire method between 300 K and 4.2 K by direct Al-bonding to unpatterned films. Magnetic properties of the films were measured using a vibrating sample magnetometer (VSM) with a maximum dc magnetic field of 1 T.
The SIS trilayers were patterned into square pillars (size ranging from 2 × 2 µm 2 to 4 × 4 µm 2 ) by optical laser lithography, ion-milling and lift-off steps. The ion milling procedure was performed using a self-aligned process for junction fabrication [31] in a Nordiko 3600 ion beam deposition system [32] with an Ar + beam (current density ~340 µA cm −2 ), first at an angle with respect to the substrate of 70° down to the CoFe 2 O 4 barrier and subsequently at 40° until it penetrated the bottom LiTi 2 O 4 electrode. This ensured a barrier with steep profile and well controlled nominal size, while avoiding material re-deposition on the sidewalls [33,34]. A 100 nmthick Al 2 O 3 layer was deposited by RF sputtering for passivation and lateral insulation of the pillars. The top electrode (Au(100 nm)/Cr(10 nm)) was deposited in an Alcatel SCM450 multi-target DC magnetron sputtering system. Before the patterning process, the structure was covered with a 15 nm-thick Ta anti-reflection layer, deposited by ion beam deposition in a Nordiko 3000 system [35], to reduce specular reflections of the laser during the lithography process.
Device transport properties were measured with a fourprobe dc current-biased method in a closed-cycle helium cryostat. A differential conductance spectrum was obtained by numerically differentiating the I-V characteristic after applying a moving average window to smooth the data.

Results and discussion
Bilayer characterisation PLD-growth of LiTi 2 O 4 requires reducting conditions, and thus during film growth, the deposition chamber was evacuated to 1 × 10 −6 Torr and the substrate temperature was kept at 800 °C; this is the optimal temperature to reduce Li segregation at the surface [23]. During the subsequent growth of CoFe 2 O 4 , the temperature of the substrate was lowered to 450 °C to avoid any unfavourable oxidation of the deposited LiTi 2 O 4 layer. Thereafter high purity oxygen was injected into the chamber and the P O2 was maintained at 2.5 × 10 −4 Torr, to limit the formation of oxygen deficiencies in the magnetic layer. In this way, the chemical potential of oxygen ions was lower and the oxidation of Ti 3+ into Ti 4+ could be avoided, keeping LiTi 2 O 4 in its metallic, superconducting phase. To verify epitaxy and bulk phase purity of the deposited films, we measured out-of-plane XRD patterns for a Temperature-dependent resistivity measurement of a CoFe 2 O 4 (10 nm)/LiTi 2 O 4 (50 nm) bilayer shows metallic behaviour (figure 2). Moreover, the bilayer displays a superconducting transition at T C = 11.5 K, confirming that the bottom layer has kept its metallic-superconducting phase without undergoing any oxidation due to the growth of CoFe 2 O 4 . The T C is in good agreement with previous findings on single LiTi 2 O 4 films [23,25,26]. The width of the superconducting transition is less than 0.4 K (figure 2, inset). The Fermi liquid behaviour of the bilayer is confirmed by the variation of resistivity as T 2 from 50 to 150 K (blue-dashed line). The residual resistivity ρ 0 and the residual resistivity ratio RRR = ρ 300 K /ρ 25 K of the films were 460 µΩ cm and 1.5, respectively, in accordance with recent publications [23,[25][26][27]36]. At temperatures below 20 K the bilayer exhibits an increase in resistance, characteristic of weak localization in disordered 2D films [37].
The room temperature magnetic hysteresis loops of a CoFe 2 O 4 (60 nm)/LiTi 2 O 4 (50 nm) bilayer grown on MgAl 2 O 4 (1 1 1) substrate are shown in figure 3. The magnetic layer is ferromagnetically easy in the film plane, with a hard direction normal to the film. The in-plane magnetization ( M s ) at 1 T and the coercive field were 200 emu cm −3 (or a magnetic moment of 1.6 µ B per formula unit) and 95 mT, respectively. This magnetic moment value is lower than the maximum 3 µ B , theoretically obtained for bulk CoFe 2 O 4 with an inverse spinel structure [38].
The decreased M s is consistent with previous reports [28] on CoFe 2 O 4 films grown at low P O2 and low temperature, and was expected due to the conditions required to avoid any oxidation of the underlying LiTi 2 O 4 . In a spin filter device, the tunnelling spin currents depend exponentially on the barrier height difference between the two spins. Thus, a lower than expected exchange energy of the FI, due to the lower M s values, can still produce a high polarisation of the current.
Several other approaches were followed in order to combine LiTi 2 O 4 and CoFe 2 O 4 in a bilayer without detrimentally affect each other during growth: (i) a few capping monolayers of CoFe 2 O 4 were grown at the same reduced P O2 environment of LiTi 2 O 4 , in order to not expose the latter to oxygen during the growth of the subsequent monolayers of CoFe 2 O 4 in higher P O2 to increase the magnetic moment of latter; (ii) the bilayer was grown entirely in reduced oxygen environment and annealed at different P O2 and at different temperatures, to compensate for the oxygen deficiencies in the CoFe 2 O 4 layer; (iii) a mixture of N 2 O/O 2 instead of O 2 was used, as suggested by Hassan et al [39], to reduce the chemical potential of the oxygen ions. In all cases, though an increased M s of the CoFe 2 O 4 layer could be observed, the underlying LiTi 2 O 4 of the bilayers showed insulating behaviour indicating an     figure 4(a), inset). The dynamic conductance of a representative sample is depicted in figure 4(a): the dI/ dV spectrum exhibits a characteristic superconducting energy gap structure with a dip around the zero bias and strongly smeared coherence peaks. At temperatures approaching the T c of LiTi 2 O 4 , the gap decreases until it disappears for higher temperatures. The decrease of the conductance observed at voltages above 2∆ is most likely due to flux flow and heating in the electrodes at high current densities ~15 kA cm −2 . Similar behaviours are common in tunnel junctions based on high T C superconductors [40]. The broadening of the coherence peaks is an evidence for the smearing of the interfacial density of states due to the proximity effect of a ferromagnetic Mott insulator, which shortens the quasiparticle lifetime [41][42][43]. Another contributing factor to the smearing of the dI/ dV curves could be the possible stoichiometric inhomogeneity between two LiTi 2 O 4 electrodes as a consequence of their different growth conditions.
The form of the dI/dV spectra implies that at least one of the LiTi 2 O 4 electrodes preserves a superconducting density of states at the CoFe 2 O 4 interface. We will begin by assuming that both electrodes are superconducting and then justify this in the light of the available information.
A simplified BCS smeared superconductor-insulatornormal metal (SIN) model was employed to fit the dI/dV raw data and estimate the energy gap ∆. According to this model 1/2 , in the limit of low bias voltages and for low temperatures [42]. Here Γ is the Dynes parameter accounting for the experimentally observed broadening [41] and for large values of Γ in both electrodes this model can also model SIS quasiparticle conductance spectra if ∆ is replaced by 2∆. The fitting values of Γ are shown in figure 4(b). In figure 4(c) it is shown the fit to a dI/dV curve collected at 2.5 K with 2∆ = 2.47 meV and Γ = 6.9 meV. The peak height and the gap structure of the raw data are quite accurately reproduced by the fit. The superconducting energy gap width 2∆(T) was determined from this data. The dependence of 2∆ on the temperature (shown in figure 4(d)) fits well with BCS-type temperature dependence [44], 2∆(T) = 2∆ 0 tanh(1.74 (T c − T) /T (solid line)) confirming a superconducting behaviour. The fitting parameters are 2∆ 0 = (2.6 ± 0.1) meV, which is lower than the one reported in previous findings [27,45,46], and T C = (11.0 ± 0.3) K, in accordance with the value measured in our bilayers. Consequently, we find a 2∆ 0 /k b T c ratio of 2.8 ± 0.2, which is less than the typical values ranging between 3 and 4.5 for BCS like superconductors but in agreement with recent scanning tunnelling spectroscopy on LiTi 2 O 4 films [47] suggesting a modified superconductivity on the surface due to a non-stoichiometric surface layer. Another contributing factor to the reduced gap value is the suppression of the order parameter in the LiTi 2 O 4 electrodes due to the proximity with the CoFe 2 O 4 magnetic barrier; this is also presumably responsible for the large value of Γ. If we assumed SIN behaviour, our estimate for 2∆ would be doubled to 5 meV that is significantly larger than reported previously and so appears unreasonable.
SIS junctions would normally be expected to show a Josephson supercurrent with a maximum value of π∆/2R j where R j is the junction normal state resistance, but for strongly spin filtering barriers, this is expected to be substantially reduced because the tunnelling of conventional singlet Cooper pairs is blocked [48]. At the lowest temperatures a zero bias peak appears in low-resistance junctions (R j ∼ 0.05 kΩ) while in medium-resistance junctions (R j ∼ 0.9 kΩ) this feature is not observed-as might be expected given the experimental noise. Although this feature might be related to the flow of a Josephson supercurrent in the junction, its disappearance at temperatures well below T c is inconsistent with standard behaviour. Similarly, the dependence of the supercurrent peak on an in-plane external applied field (shown in figure 4(e)) does not show the Fraunhofer-like periodic suppression of the peak characteristic of Josephson tunnel junctions. Indeed, the appearance of the zero-bias peak may also be related to the presence of Andreev bound states [49].
The dI/dV curves collected at higher biases (figure 5) reveal an interesting midpoint state between the low bias SISstate (i.e. both electrodes are superconducting) and the state in which the electrodes are metallic (normal state) at high bias. This conductance midpoint state is related to bias voltages at which one of the LiTi 2 O 4 electrodes is superconducting while the other is metallic. The midpoint state, identified by the dashed arrow in figure 5, indicates that the electrodes are in different superconducting states. For high biases the two electrodes are in their normal state and the conductance of the junction is equal to that measured at temperatures above T C (12 K). At higher temperatures, lower biases are needed to turn the electrodes from the superconducting state to the metallic-normal state. This confirms the SIS-nature of the junctions, while the presence of two distinct conductancestates is another validation of a stoichiometric inhomogeneity between two superconducting LiTi 2 O 4 electrodes.
dI/dV spectra collected at 1.5 K at different out of plane applied magnetic fields are shown in figure 6. The closing of the peak position along with the closing of the gap and the suppression of the superconducting peak for values approaching the LiTi 2 O 4 upper critical field H c2 , are clearly visible. The scaling law follows a field quadratic-depend- 2 , as recently reported in point contact spectra [27]. The fit, shown in the inset of figure 6, gives an extracted value of H c2 at 2 K of ~10.8 T, which is consistent with previous results [45,46]. Figure 7 shows the temperature dependence of a typical LiTi 2 O 4 /CoFe 2 O 4 /LiTi 2 O 4 junction resistance with 1.5 nm CoFe 2 O 4 barrier measured by applying a 0.1 mA current. A sharp drop in resistance is seen at the LiTi 2 O 4 superconducting transition due to the disappearance of the in-series resistance of the leads. At higher temperatures the resistance is not exponentially increasing with decreasing temperature, which is the behaviour for a semiconducting non-magnetic barrier [50], but is instead continuously dropping with temperature. The temperature dependence of the resistance of the LiTi 2 O 4 bottom lead of the same junction was measured (inset (b), figure 7) to verify that the decreasing behaviour of R j is attributable to tunnelling current flowing across the tunnel junction and not across any series resistances, which would explain the decreasing behaviour. This is confirmed by difference in the order of magnitude between the resistance of junction ~10 1 Ω and the resistance of the bottom-lead ~10 2 Ω. In addition, large contributions of non-tunnelling (leakage) conductance to the dominant tunnel conductance due to shorts between the two electrodes can be also ruled out since R j is non-zero for temperature below T C , as opposed to the two LiTi 2 O 4 superconducting electrodes which show zero resistance.
Moreover, the resistance increases with decreasing temperature below T C , due to the fact that there are no available states for tunnelling at the Fermi energy level for measurements voltages much less than ∆. In this case the conductance is dominated by thermal excitation of quasi-particles across the gap and, as temperature decreases, the number of thermally excited quasi-particle states decreases exponentially, resulting in an increases of the sub-gap resistance for decreasing temper ature. These behaviours confirm that the mechanism of charge transport in the junctions is predominantly tunnelling in nature and thus, the drop in R j with decreasing temperature observed across the entire temperature range above T C may  be a consequence of the exchange splitting of the magnetic tunnel barrier, leading to a temperature dependent reduction of the barrier height of one spin (inset (a), figure 7). The T Curie of CoFe 2 O 4 is well above room temperature, so the absence of the typical change from semiconducting behaviour to metalliclike behaviour at T Curie , due to onset of spin filtering, reported in spin filtering devices of this type [8,12] is expected in our range of measurement.

Conclusions
In summary, we demonstrated the successful superconducting tunnel process in an all-spinel SIS tunnel junctions with CoFe 2 O 4 as FI barrier and LiTi 2 O 4 as electrodes grown on MgAl 2 O 4 substrates. The integration of the metallic-superconducting LiTi 2 O 4 in tunnel junctions offers new possibilities in the quest of achieving high efficiency room temperature spin filtering due to lattice match with the spinel Co-ferrite, reducing APBs. The CoFe 2 O 4 /LiTi 2 O 4 holds the potential for all-oxide magn etic tunnel junctions with efficient spin filtering properties at room temperature. An estimation of the polarisation of the cur rent could not be performed by extrapolating the temperature dependence of R j from the high temper ature (> T Curie ) regime as T Curie in this case is well above room temperature. This capability could be investigated by tunnel magnetoresistance-like experiments by replacing the top LiTi 2 O 4 electrode with a spinel ferromagnet (Fe 3 O 4 ) decoupled from the CoFe 2 O 4 by a thin insulating layer of MgAl 2 O 4 , as suggested by promising tunnelling spectroscopy study on junctions with Au electrode [13]. The perfect epitaxy and lattice match between all the layers of such