Maximizing Exchange Bias in Co/CoO Core/Shell Nanoparticles by Lattice Matching between the Shell and the Embedding Matrix

The exchange bias properties of 5 nm Co/CoO ferromagnetic/antiferromagnetic core/shell nanoparticles, highly dispersed in a CuxO matrix, have been optimized by matching the lattice parameter of the matrix with that of the CoO shell. Exchange bias and coercivity fields as large as HE = 7780 Oe and HC = 6950 Oe are linked to the presence of a Cu2O matrix (0.3% lattice mismatch with respect to the shells). The small mismatch between Cu2O and CoO plays a dual role: it (i) structurally stabilizes the CoO and (ii) favors the existence of a large amount of uncompensated moments in the shell that enhance the exchange bias effects. The results indicate that lattice matching may be a very efficient way to improve the exchange bias properties of core/shell nanoparticles, paving the way to novel approaches to tune their magnetic properties.


INTRODUCTION
Exchange bias refers to a set of different phenomena (e.g., loop shifts in the field axis, HE, coercivity enhancement or loop asymmetries, among others) which arise when two materials with significantly different magnetic anisotropy, typically a ferromagnet (FM) and an antiferromagnet (AFM), are exchanged coupled at their mutual interface. [1][2][3][4][5] Although most devices exploiting exchange bias effects rely on thin films, [6][7][8] new applications based on nanoparticles, such as coercivity enhancement for permanent magnets 9,10 , microwave absorbers 11 or novel spintronic devices, 12,13 are continuously emerging. Driven by the existing and prospective applications, there is currently a trend to try to enhance the different exchange bias-related effects. 1,2,4,5 In this respect, although archetypical exchange bias systems, e.g., AFM/FM bilayers or FM-AFM core/shell nanoparticles, typically exhibit HE values in the range of tens to hundreds of Oe; 1,2,4,5 HE values in excess of 10 kOe have also been occasionally reported. [14][15][16][17][18][19] However, apart from some possible minor-loop issues, 20,21 many of these systems are based on phase-separation. 14,16,17 This leads to illdefined FM and AFM counterparts often with FM phases of very small dimensions. Another strategy to obtain very large values for H E is using FM materials with very low magnetization. 15,18 HE has been classically expressed as μ0HEMFMV = γA, where MFM is the FM magnetization, V is the volume of the ferromagnet, γ is the interfacial coupling energy per unit surface area, and A is the associated surface area (leading to the well-known expression μ 0 H E = γ/M FM t FM for thin films). 2 Thus, reducing M FM and t FM are two obvious approaches to enhance HE. From the applied viewpoint, however, these may be rather unpractical due to the reduced magnetic moment. Improving HE using γ may be more appealing, despite being more challenging since γ is still not well understood. Although to a first approximation γ should only depend on the intrinsic properties of the FM and AFM materials and their mutual exchange interactions, many other parameters have been shown to affect γ (e.g., defects, crystalline orientation, crystallinity, thickness of the AFM, roughness or AFM domains). 2,[22][23][24][25][26] It has been also shown that the presence of defects and/or impurities in the AFM component, even far from the FM/AFM interface, can dramatically affect the amount of exchange bias. [27][28][29] While considerable effort has been devoted in thin film systems to harness the diverse mechanisms affecting HE, 2,4 the role of these effects in core/shell nanoparticles has been far less investigated. 1,3,5 Due to its appealing properties (e.g., a large interface energy, γ) one of the most studied FM/AFM core/shell systems is Co/CoO. [30][31][32][33][34][35][36][37][38][39][40][41][42] Interestingly, in this system H E values ranging from a few to thousands of Oe have been reported for relatively similar nanoparticles. Several effects have been reported to influence HE in Co/CoO: core diameter, 34,39,40 shell thickness, 34,37,39,40 crystallinity of the shell, 32,43,44 exchange interactions with neighboring particles, 32,45 strains, 34 orbital moments, 45 uncompensated spins 33,41 and the matrix they are embedded in (through lattice matching effects or antiferromagnetic proximity effects). 31,33,36,46 Concerning the role of the matrix in Co/CoO, HE values extending from 10 Oe (in non-AFMnon-lattice matched matrices, e.g., Al2O3 ) 31 to 7500 Oe (in AFMlattice matched matrices, e.g., CoO) 31,36 have been reported for rather similar nanoparticles. Despite this rather large spread of experimental H E values, there is a lack of systematic studies aimed at elucidating the role of the embedding matrix in the improvement of the exchange bias related properties in core/shell nanoparticles, i.e., at maximization of the effect.
Here we present a systematic study of the magnetic properties of Co/CoO core/shell clusters deposited in a CuxO matrix under different oxygen partial pressures, pO2. The magnetic properties are optimized for oxygen pressures yielding matri-ces with a majority Cu2O component, which provides an almost perfect lattice matching with the CoO shell. The results indicate that the lattice matching induces a structural stabilization (i.e., improved crystallinity) of the CoO and the concomitant migration of Cu atoms into the shell, enhancing the density of AFM uncompensated spins vital for large HE field relatively high exchange bias blocking temperature.

EXPERIMENTAL
The samples have been grown by a combination of a vapor phase aggregation technique (i.e., a cluster source) and conventional sputtering. 47 Pure Co nanoparticles are pre-formed in the cluster source by sputtering in an Ar+He atmosphere of about 0.1 mbar. 47 These are then injected by differential pumping into the sputtering deposition chamber where the Ar pressure is two orders of magnitude lower. Importantly, in the main chamber a controlled amount of O2 is also present (with partial pressures pO2 = 0 to 5·10 -4 mbar). The nanoparticles are thus partially oxidized during their flight towards the substrate, forming a core/shell structure with a Co core and a CoO shell. A Cu target is simultaneously sputtered as the particles arrive at the substrate. 47 This leads to the Co/CoO nanoparticles being embedded in a CuxO matrix (where the precise crystallographic composition of the matrix depends on the O2 partial pressure). The deposition rates of particles and matrix were adjusted to guarantee a high dilution degree of nanoparticles (less than 1 % by volume). Finally the oxygen gas was removed and a capping layer of pure Nb was deposited in order to prevent further (ex-situ) oxidation. 48 Transmission electron microscopy (TEM) using a FEI Tecnai G2 F20 TEM operated at 200 kV was performed on nanoparticles deposited on a TEM-grid in the absence of oxygen and without a matrix or capping layer. X-ray diffraction was carried out using a laboratory source (CuKα radiation) and Bragg-Brentano geometry. Diffraction patterns were analyzed using the Rietveld method implemented via FullProf. 49 Figure 1. Hysteresis loops (with different corrections) of samples grown in p(O2)=0, 27 and 50 ·10 -5 mbar, at 10K, after cooling from 330K in a saturating field. Black: as measured, blue: after removing substrate contribution; red: after removing the remaining high field susceptibility.
Magnetic properties were measured in a SQUID magnetometer. Hysteresis loops were obtained at fixed temperatures between 10K and 330K after field cooling from 330 K to 10 K using an applied field HFC = 5o kOe. Note that depending on the coercivity (HC) of the sample, a maximum field of either 50 kOe or 70 kOe was used to avoid minor loop effects. 20,21 The as-obtained hysteresis loops present a negative slope at high fields ( Fig. 1, black symbols). To separate the contribution of the substrate from that of the composite films, the high field susceptibility of the substrates was carefully evaluated and subtracted from the data ( Fig. 1, blue symbols). The resulting loops then show a positive slope, originating from the films. Finally, to obtain the exchange bias parameters (HE, HC and vertical shift) 50 the high field slope, χHF, was also subtracted (Fig. 1, red symbols). Fig. 2 shows a TEM image of the nanoparticles together with the corresponding particle size histogram. As can be seen the size distribution is rather narrow with more than one third of the particles with sizes between 4.5 and 5 nm. An independent estimation of the size can be obtained from the superparamagnetic behavior of the nanoparticles at high temperatures. The fit of the room temperature hysteresis loop of the sample with pO2 = 0 to a Langevin function 51 yields an average nanoparticle moment of (10200±200) µB (see inset of Fig. 3), which translates to an average nanoparticle diameter of (5.01 ± 0.03) nm (assuming a magnetic moment of 1.72µB per Co atom), in good agreement with TEM observations.  . Evolution of the moment per particle on the O2 pressure, as deduced from Langevin fits to room temperature loops, for the samples that could be satisfactorily fitted to such a Langevin function. Note that for high pO2 the loops show a weak secondary magnetic contribution, making it impossible to reliably fit then using Langevin function. The inset shows the 300K loop of the reference sample grown with no O2 in the chamber, together with its fit (red line).

RESULTS AND DISCUSSION
Note that the conditions for the nanoparticle generation in the cluster source are the same for all the samples. Consequently, the number of particles in each sample should be roughly the same for all samples. Thus, any changes in the saturation moment [ Fig. 4(d)] of the samples should primarily arise from the partial oxidation of the Co nanoparticles. As can be seen in Fig. 4, the values of both HE and HC at 10 K exhibit an analogous non-monotonic behavior with pO2, with maxima at around pO2 = 23·10 -5 mbar. Interestingly, the maximum values of H E and H C (7780 and 6850 Oe, respectively) are extraordinarily large for a Co/CoO diluted system with very thin CoO shells (tCoO=0.4-0.6 nm) and approach values typical of very dense systems (with strong shell-shell interactions) 32 or Co/CoO nanoparticles embedded in a CoO matrix. 31,36 Since the nanoparticles are grown as pure Co clusters and develop CoO shells during flight through the O2 atmosphere towards the substrate, the shell thickness (and thus the core diameter) could vary with pO2. These changes could affect the evolution of the exchange bias properties with pO2. A Langevin fit to the high temperature hysteresis loop of the different samples shows that, indeed, the average particle magnetic moment, and consequently the core diameter, decreases monotonically with increasing pO2 (see Fig. 3). Hence, as pO2 becomes higher the core diameter decreases, while the shell thickness concomitantly increases. However, a simple calculation shows that the changes are not significant: the core diameter is reduced from 5.0 nm to 4.2 nm, leading to a shell thickness increase of merely 0.6 nm (taking into account the lower density of CoO with respect to Co). Since the radius of the core (equivalent to the thickness of the ferromagnet in the well-known formula for thin films μ0HE = γ/MFMtFM) 1,2 becomes smaller as pO2 increases, HE is expected to increase monotonically due to the HE1/tFM relationship. Moreover, it is well-known that when the AFM thickness is very small virtually no bias is induced. 27,56 As the AFM thickness increases slightly, HE should also monotonically increase, 27,56 as observed for small p O2 . Notably, in some systems, in particular in thin films, for thick enough AFM layers HE can eventually decrease. 56 In fact, similar effects have been observed in Co/CoO nanoparticles. 57 However, the HE reduction with AFM thickness occurs for much thicker AFM layers than the shells formed in our nanoparticles. 56,57 Hence, although the slight changes in morphology of the Co/CoO samples with pO2 may explain the increase in HE and HC for small pO2, the decrease in HE for high O2 contents must have some other origin. Since the matrix has been shown to play an important role in the magnetism of Co/CoO nanoparticles, we next evaluate the evolution of the matrix with pO2. The matrices may contain different proportions of the different possible CuxO phases: Cu, Cu2O, Cu4O3 and CuO. From this list two interesting cases should be highlighted: (i) Cu2O which has a very good lattice match to CoO (with only ~0.3% mismatch) and thus could lead to a "structural stabilization" (i.e., a higher degree of crystalline order) of CoO resulting in improved antiferromagnetic properties, 32 and (ii) CuO which is antiferromagnetic with a Néel temperature of TN = 200 K and thus could enhance the magnetic properties due to exchange coupling. 36 As can be seen in Fig. 5, quantitative phase analysis (via the Rietveld method) of the XRD patterns shows that, with increasing pO2, there is certainly an evolution of the phases present in the matrix towards more oxidized states of Cu. Remarkably, the main improvement of the exchange bias properties occurs as the Cu2O phase appears, while the worsening of the properties for large pO2 is linked to the presence of CuO. However, an exact matching between the maximum HE and the amount of Cu2O is not observed probably because of the presence of Cu4O3 around the optimal pressure pO2=23·10 -5 mbar. Although Cu4O3 (AFM with TN = 40 K) 58 could play a role, it is likely that the evolution of the lattice parameter of Cu 2 O (and for example the presence of microstrains) with p O2 could be the dominating factor. Other factors, such as the increase in shell thickness and decrease in core diameter, will be simultaneously influencing the magnetic properties, thus one-to-one correlations between structure and magnetism are somewhat difficult. In any case, our results unambiguously show that for Co/CoO nanoparticles, the structuralstabilization of the shell is the dominant factor enhancing the magnetic properties, with exchange-stabilization playing a secondary role. Another outstanding aspect of our results is the unexpectedly large HE and HC values observed in optimal pO2 conditions. The current understanding of exchange bias is based on the existence of uncompensated spins. Two types of uncompensated spins have been identified: pinned (unaffected by the magnetic field and which give rise to H E ) and unpinned or rotatable (which are dragged by the FM component and are related to the enhancement of HC). [59][60][61][62][63]  In a hysteresis loop the pinned uncompensated spins are evidenced by a vertical shift. 50 Assuming that, once the FM magnetization is switched, the rotatable uncompensated spins can be further dragged by the applied magnetic field, then χHF should have a contribution from these uncompensated spins. [52][53][54][55] From panel (c) of Fig. 4, it is clear that the behavior of the pinned uncompensated spins (as quantified by the vertical shift) is similar to that displayed by HE. Similarly, the dependence of χHF on pO2 (Fig. S2), which should carry out information about the rotatable uncompensated spins (see Supporting Information), resembles that of HC. Interestingly, in thin films it has been shown that the dilution of the magnetic ions in the AFM layers by similar non-magnetic ions leads to a marked increase in HE and HC (e.g., CoO-Mg, FeMn-Cu, FeF2-Zn). 29,[64][65][66] Hence, in analogy to thin film systems, the large increase of HE and HC in the Co/CoO samples is due to the large increase in the number of uncompensated (pinned and unpinned) spins for intermediate O 2 pressures, presumably linked to a "dilution effect" in the CoO shell. Such effect is probably brought about by intermixing between the CoO shell and the Cu atoms of the CuxO matrix during the particlematrix co-deposition, 36,67 which is undoubtedly enhanced by the lattice matching effect between Cu2O and CoO. To rule out any effect related to the small morphological changes caused by the O 2 in the chamber (core diameter -shell thickness) and to confirm the dominating role of the matrix in the observed effects, we have fabricated a second series of samples with Ag as matrix, instead of Cu, while keeping all conditions unaffected (see Supporting Information, SI). Note that, although both Ag and Ag2O have a cubic structure, the lattice matching with CoO is considerably worse than for Cu 2 O (Agmismatch 4.2%; Ag 2 Omismatch 10.6%). As can be seen in Fig. S1, the Ag x O-matrix series also shows a maximum with pO2. However, compared to the CuxO-matrix (Fig. S2) (i) the maximum occurs at much lower pO2, (ii) the HE values are remarkably smaller, and (iii) both the vertical shift and HF are much smaller and without a clear link to HE. These results eliminate the possibility that the small pO2induced morphological changes contribute significantly to the enhanced exchange bias properties. Interestingly, as expected, the maximum in HE occurs at rather low pO2, in the region where the Ag-based cubic phases exist; between the pure Ag matrix and the Ag2O dominated matrix. 68,69 The fact that the maximum HE does not coincide with the best matched matrix (i.e., pure Ag) is probably due to the competition between the better lattice matching of the Ag matrix (which would better structurally stabilize the thin CoO shell) and the oxide character of the Ag2O matrix. Namely, although Ag has a better lattice matching, the fact that it is not an oxide probably makes it more difficult for the Ag to diffuse into the CoO shell. A combination of poor matching (hence, low structural stabilization) and the limited interdiffusion (i.e., fewer uncompensated spins) is understood to be the origin of the greatly reduced HE in the AgxO-matrix when compared to the CuxO-matrix. Moreover, it should be taken into account that when growing without O2 the CoO shell probably would be very thin (only due to the residual O2 in the chamber), which could also lead to reduced HE in the pO2 = 0 case. Nevertheless, it is worth emphasizing that although the maximum HE values for the AgxO-matrix (HE = 300 Oe) are smaller than for the CuxOmatrix, given the cubic character of the AgxO-matrix for low pO2 (and thus a moderate lattice matching), these values are much larger than for matrices with no structural matching, e.g., Al2O3 which exhibit HE values of only a few Oe. 32 Further evidence of the enhanced properties for the latticematched samples arises from the temperature dependence of HE. As can be seen in Fig. 6, the samples with optimal O2 pressure exhibit an exchange bias blocking temperature (i.e., T at which HE vanishes) of about 200 K. This value is typical of high quality isolated Co/CoO nanoparticles with substantially thicker CoO shells than in the NPs discussed here. 42 On the other hand, as the O2 pressure varies away from the optimum value, quickly decreases, as shown for the p O2 = 13·10 -5 and 50·10 -5 mbar samples in Fig. 6. In fact, the dependence of on pO2 nicely mimics the dependence of HE(pO2) (compare the inset of Fig. 6 with Fig. 4 (a)). These results indicate that the optimum pO2 not only enhances HE but it stabilizes the exchange bias of the nanoparticles. This is in contrast to what has been observed for thin films where often an increase of HE is usually accompanied by a decrease in 29,[64][65][66] and thus gives further credibility to the structural stabilization model, since it shows that despite the magnetic dilution behind the enhanced HE (which is the origin of the decreased in thin films), is actually highest for the samples with the largest HE. is understood to be linked to the magnetic anisotropy energy of the CoO shell, i.e., KCoOVCoO (where K and V are the anisotropy and the volume of the CoO grains, respectively). If we assume that VCoO does not change considerably between samples, an increase in should imply a higher KCoO, which would be expected for shells with a high degree of crystalline order. Additionally, the data allows us to rule out any leading role of exchange coupling in the enhanced properties, since for the case of pO2=23·10 -5 mbar no anomaly is seen about the Néel temperature of Cu3O4 [TN(Cu3O4)= 40 K]. Similarly, for high pO2, where the CuO matrix is dominant, is clearly below TN(CuO) = 200 K, in contrast to when a NiO matrix is used (where exchange coupling is known to be dominant). 36

CONCLUSIONS
In summary, we have studied the exchange bias properties of Co/CoO core/shell nanoparticles embedded in different CuxO matrices. The results show that HE, HC and are enhanced when the matrix is partially formed by Cu2O, which has a good lattice parameter match with that of the CoO shell. In fact, for the optimized conditions both HE and HC present very large values (7780 and 6950 Oe, respectively), which can be linked to the presence of a large number of uncompensated spins (pinned and unpinned) in the CoO shell, induced by interdiffusion of Cu ions from the matrix to the shell during the sputtering. The results highlight the importance of the surrounding matrix to improve the crystalline quality of the thin CoO shells in the Co/CoO nanoparticles. Hence we demonstrate that lattice matching may be a highly efficient strategy to improve the magnetic properties of certain core/shell nanoparticle systems.

AUTHOR INFORMATION
Corresponding Author * Email: j.a.gonzalez@uclm.es