Magnetic anisotropy engineering in onion-structured metal oxide nanoparticles combining dual exchange coupling and proximity effects

Abstract

A series of exchange-coupled magnetic nanoparticles combining several magnetic phases in an onion-type structure were synthesized by performing a three-step seed-mediated growth process. Iron and cobalt precursors were alternatively decomposed in high-boiling-temperature solvents (288–310 °C) to successively grow CoO and Fe_3−δO₄ shells (the latter in three stages) on the surface of Fe_3−δO₄ seeds. The structure and chemical composition of these nanoparticles were investigated in depth by combining a wide panel of advanced techniques, such as scanning transmission electron microscopy (STEM), electron energy-loss spectroscopy–spectrum imaging (EELS-SI), ⁵⁷Fe Mössbauer spectrometry, and X-ray circular magnetic dichroism (XMCD) techniques. The size of the nanoparticles increased progressively after each thermal decomposition step, but the crystal structure of core–shell nanoparticles was significantly modified during the growth of the second shell. Indeed, the antiferromagnetic CoO phase was progressively replaced by the CoFe₂O₄ ferrimagnet due to the concomitant processes of partial solubilization/crystallization and the interfacial cationic diffusion of iron. A much more complex chemical structure than that suggested by a simple size variation of the nanoparticles is thus proposed, namely Fe_3−δO₄@CoO-CoFe₂O₄@Fe_3−δO₄, where an intermediate Co-based layer was shown to progressively become a single, hybrid magnetic phase (attributed to proximity effects) with a reduction in the CoO amount. In turn, the dual exchange-coupling of this hybrid Co-based intermediate layer (with high anisotropy and ordering temperature) with the surrounding ferrite (core and outer shells) stabilized the particle moment well above room temperature. These effects allow for the production of Fe oxide-based magnetic nanoparticles with high effective anisotropy, thus revealing the potential of this strategy to design rare-earth-free permanent nanomagnets at room temperature.

---

Introduction

Permanent magnets are widely applied in most devices necessary for today's daily life applications, such as communication, transport, and renewable energies. However, their efficient magnetic properties require a high consumption of rare-earth (RE) and platinum group (PG) components, which are classified as critical raw materials by the European Union owing to potential supply risks.¹ In this context, magnetic nanoparticles represent a true alternative to build high-performance technological devices.² At the nanoscale, their physical properties can be efficiently modulated according to the size and shape of the objects. However, the transformation of soft RE- and PG-free magnetic nanomaterials into permanent magnets remains a huge challenge.

Superparamagnetic iron oxide (Fe₃O₄) nanoparticles are probably the most well-known nanomagnets owing to their extensive investigations for biomedical applications, among other reasons.^3,4 Their relatively low magnetic anisotropy can be significantly enhanced by growing a magnetically harder shell, which pins a softer core through interfacial exchange-coupling, thus increasing its effective anisotropy.^5,6 Beyond the wide range of structural parameters (core size, shell thickness, chemical composition, defects, doping, interfacial roughness, etc.) significantly affecting exchange coupling,^7–13 the selection of a hard phase is restricted by high crystal complementarity and large difference in anisotropy with a soft phase.^6,14 Although CoO fulfils both criteria (the cell parameter of Fe₃O₄ (8.396 Å) is nearly twice that of CoO (4.26 Å) and magnetic anisotropy constants differ by two orders of magnitude, K(Fe₃O₄)¹⁵ = 2 × 10⁴ J m⁻³ and K(CoO) = 5 × 10⁶ J m⁻³),¹⁶ its antiferromagnetic order is lost at room temperature (Néel temperature, T_N = 290 K).¹⁷ This drawback can be circumvented by embedding Co@CoO nanoparticles in a NiO matrix to exploit proximity effects.¹⁸ The doping of the Wüstite phase by divalent cations is also a potential route to enhance exchange-bias coupling with soft phases.^19,20 Ultimately, the Wüstite phase can be replaced by a high-anisotropy spinel phase, such as in some ferrites, with a ferrimagnetic order far above room temperature.^14,21,22

Synthetic processes can also favor the formation of ferrites, since high temperatures usually favor cation mobility through interfaces^20,23 and partial solubilization.^7,24 Therefore, Fe₃O₄@CoO nanoparticles can include an intermediate layer, such as a CoFe₂O₄ shell, and can be better described as a core@shell@shell structure.^22,23 Such an onion-type structure offers the possibility to generate additional soft–hard interfaces to enhance exchange coupling. Nevertheless, they have been rarely reported because multi-step synthesis with fine control of the size and shape is certainly very difficult.^25–27 Recently, we opened new perspectives into this research domain by reporting on the synthesis of Fe₃O₄@CoO@Fe₃O₄ nanoparticles displaying permanent magnetization at room temperature.²⁸ However, their chemical structure turned out to be much more complex than expected, hampering the understanding of the magnetic properties (including the exchange coupling phenomenon). Indeed, we noticed that the formation of a second Fe₃O₄ shell resulted in a remarkable modification of the nanoparticle structure, which was dominated by the concomitant partial disappearance of the CoO Wüstite phase and the appearance of the CoFe₂O₄ spinel phase. Such a significant modification of the nanoparticle structure could certainly contribute to an increase in the magnetic anisotropy energy. However, given, the limited number of samples that were synthesized, we could not conclude on this point.

Herein, we report an in-depth study of the chemical structure of Fe₃O₄@CoO@Fe₃O₄ nanoparticles to better understand their magnetic properties, including their remanent magnetization at room temperature. Their synthesis consisted of a three-step seed-mediated synthesis that involved alternating thermal decompositions of iron and cobalt precursors. Crucially, the amount of iron precursor used in the third step was modified in order to investigate systematically the chemical structure of the onion nanoparticles and its relationship with the magnetic properties. A wide range of advanced characterization techniques, such as element-specific electron loss spectroscopy-spectral high-resolution imaging (EELS-SI), ⁵⁷Fe Mössbauer spectrometry, and X-ray magnetic circular dichroism (XMCD), were used to accurately characterize the spatial distribution, site occupancy, and oxidation states of Fe and Co cations in the unexpectedly complex multilayer structures. This, in turn, enabled a greater understanding of the magnetic properties, in particular the unprecedented enhancement of the effective magnetic anisotropy in iron-rich oxide nanoparticles due to a combination of magnetic proximity and dual exchange-coupling effects.

Experimental section

Chemical compounds purchased

FeCl₂·4H₂O (99% Acros Organic), CoCl₂·6H₂O (reagent grade, Sigma), oleic acid (99%, Alfa Aesar), octyl ether (99%, Sigma), acetone (Pure, Carlo Erba), ethanol (Absolute, Carlo Erba), chloroform (pure stabilized with ethanol, Carlo Erba) were used as received.

Synthesis of the metal precursors

Iron stearate (FeSt₂) and cobalt stearate (CoSt₂) were prepared by a home-made synthesis in order to control precisely the purity and the thermal decomposition process with high reproducibility of the nanoparticle structure.²⁹

Synthesis of the nanoparticles

For synthesis of the nanoparticles, a three-step synthesis process was followed. The first step consisted of the synthesis of iron oxide nanoparticles (denoted C) by the decomposition of iron(II) stearate (FeSt₂) in dioFctyl ether (b. p. = 288 °C). The second step consisted of the decomposition of cobalt(II) stearate (CoSt₂) in order to grow a CoO shell on the surface of the iron oxide nanoparticles, thus forming core@shell (CS) nanoparticles. Finally, the third step consisted of the thermal decomposition of FeSt₂ in order to grow a second shell of iron oxide, e.g., core@shell@shell (CSS) nanoparticles. The amount of FeSt₂ was set according to the molar ratio R = n(FeSt₂ shell)/n(FeSt₂ core) = 0.5, 1, 1.5 in order to modulate the thickness of the second shell, resulting in three samples named CSSa, CSSb, and CSSc, respectively.

Core nanoparticles (C). Iron oxide nanoparticles (C) were synthesized according to the protocol described in a previous article.⁸ Briefly, 1.38 g (2.22 mmol) of home-made iron(II) stearate and 1.254 g (4.44 mmol) of oleic acid were poured in to a two-necked round-bottom flask, and 20 mL of dioctyl ether (b. p. = 288 °C) was then added. The mixture was stirred at 100 °C during 30 min to remove water residues. Then, the stirring was stopped and the mixture was heated up to reflux for 2 h at a heating ramp of 5 °C min⁻¹. The reaction medium was then cooled to 100 °C. The nanoparticles were precipitated by adding acetone and were then washed with a mixture of chloroform/hot acetone by centrifugation (14 000 rpm, 5 min). Finally, the obtained nanoparticles were stored as a colloidal suspension in chloroform.

Core@shell nanoparticles (CS). Next, 90% of the volume of the colloidal suspension of C nanoparticles was poured in to a 50 mL round-bottom flask to evaporate the solvent under vacuum. Then, 10 mL of dioctyl ether was added and the mixture was sonicated until the nanoparticles were in suspension. A solution of 1.248 g (1.99 mmol) of home-made cobalt(II) stearate and 1.119 g (3.96 mmol) of oleic acid diluted in 20 mL of octadecene were added. The solution was then heated to 100 °C for 30 min to remove the solvent and water residues and was finally brought to reflux for 2 h at a heating ramp of 1 °C min⁻¹. The reaction medium was then cooled to 100 °C. The nanoparticles were precipitated by adding acetone and were washed with a mixture of chloroform/hot acetone or chloroform/ethanol by centrifugation (14 000 rpm, 5 min). The obtained core@shell nanoparticles (CS) were finally stored as a colloidal suspension in chloroform.

Core@shell@shell nanoparticles (CSS). Here, 25% of the volume of the CS colloidal suspension was added to a 50 mL round-bottom flask before evaporating the solvent under vacuum. Next, 20 mL of dioctyl ether was added and the mixture was sonicated until all the nanoparticles were suspended. A mixture of home-made iron(II) stearate and oleic acid diluted in 20 mL of dioctyl ether was added to the solution according to Table 1. The reaction medium was heated to 120 °C for 30 min to remove the solvent and water residues and was finally brought to reflux for 2 h at a heating ramp of 1 °C min⁻¹. The resulting black solution was then cooled down to 100 °C and the nanoparticles were precipitated by adding acetone and then washed with a mixture of chloroform/acetone by centrifugation (14 000 rpm, 5 min). The obtained core@shell@shell nanoparticles (CSS) were finally stored as a colloidal suspension in chloroform.

Table 1 Amount of the iron precursor used to perform the third thermal decomposition step according to the molar ratio R = n(shell precursor)/n(core precursor). Oleic acid was added according to n(oleic acid) = 2n(shell precursor)

Sample	R ratio	m (iron(II) stearate) g	n (iron(II) stearate) mmol	m (oleic acid) g	n (oleic acid) mmol
CSSa	0.5	0.156	0.25	0.141	0.50
CSSb	1	0.311	0.50	0.282	1.00
CSSc	1.5	0.468	0.75	0.423	1.50

---

Transmission electron microscopy (TEM) was performed using a JEOL 2100 LaB6 instrument with a 0.2 nm point-to-point resolution and a 200 kV acceleration voltage. EDX was performed with a JEOL Si(Li) detector. The average size of the nanoparticles was calculated by measuring at least 300 nanoparticles from the TEM micrographs using the ImageJ software. The average shell thickness was calculated as half of the difference between the size of the nanoparticles before and after the thermal decomposition step. The size distribution was fitted by a log–normal function.

High-resolution scanning transmission electron microscopy (STEM) and electron energy-loss spectroscopy (EELS) analyses were performed using Cs aberration-corrected STEM on a NION UltraSTEM200 system coupled with a high-sensitivity EEL spectrometer. The convergence and collection semi-angles in the EELS experiments were respectively 35 mrad and 50 mrad. CS, CSSa, and CSSb were analyzed with the STEM microscope operated at a 100 kV acceleration voltage, while the CSSc experiments were performed at 60 kV.

X-Ray diffraction (XRD) was performed using a Bruker D8 Advance instrument equipped with monochromatic copper radiation (Kα = 0.154056 nm) and a Sol-X detector in the 20°–80° 2θ range with a scan step of 0.02°. High purity silicon powder (a = 0.543082 nm) was systematically used as an internal standard. The crystal sizes were calculated by Scherrer's equation and the cell parameters by Debye's law.

Fourier transform infra-red (FT-IR) spectroscopy was performed using a PerkinElmer Spectrum spectrometer in the energy range 4000–400 cm⁻¹ on samples diluted in KBr pellets.

Granulometry measurements were performed using a nano-sizer Malvern (nano ZS) zetasizer at a scattering angle of 173°. Each measurement corresponded to the average of 7 runs of 30 s.

Themogravimetry analyses (TGA) were performed using a SDTQ600 from TA instrument. Measurements were performed on dried powders under air in the temperature range of 20–600 °C at a heating rate of 5 °C min⁻¹.

X-Ray absorption (XAS) and X-ray magnetic circular dichroism (XMCD) spectra were recorded at the L_2,3 edges of Fe and Co, on the DEIMOS beamline at SOLEIL (Saclay, France).³⁰ All the spectra were recorded at 4.2 K under UHV conditions (10⁻¹⁰ mbar) and using the total electron yield (TEY) recording mode. The measurement protocol was previously detailed by Daffé et al.³¹ An external parallel magnetic field H⁺ (antiparallel H⁻, respectively) was applied on the sample while a polarized σ₊ (polarized σ₋, respectively) perpendicular beam was directed on the sample. Isotropic XAS signals were obtained by taking the mean of the σ₊ + σ₋ sum, where σ₊ = [σ_L(H⁺) + σ_R(H⁻)]/2 and σ₋ = [σ_L(H⁻) + σ_R(H⁺)]/2, with σ_L and σ_R the absorption cross-sections measured respectively with left and right circularly polarized X-rays. XMCD spectra were obtained by taking the σ₊–σ₋ dichroic signal with a ±6.5 T applied magnetic field.

At the DEIMOS beamline, the circularly polarized X-rays were provided by an Apple-II HU-52 undulator for both XAS and XMCD measurements while EMPHU65 with a polarization switching rate of 10 Hz was used to record hysteresis cycle at fixed energy.³⁰ Measurements were performed between 700 and 740 eV at the iron edge and between 770 and 800 eV at the cobalt edge with a resolution of 100 MeV and a beam size of 800 × 800 μm. Both the XMCD and isotropic XAS signals presented here were normalized by dividing the raw signal by the edge jump of the isotropic XAS. The samples consisted of drop-casting suspensions of nanoparticles in chloroform onto a silicon substrate. The substrates were then affixed on a sample holder.

⁵⁷Fe Mössbauer spectroscopy was performed at 77 K using a conventional constant acceleration transmission spectrometer with a ⁵⁷Co(Rh) source and a bath cryostat. The samples consisted of 5 mg Fe per cm² powder concentrated in a small surface due to the rather low quantities. The spectra were fitted by means of the MOSFIT program³² involving asymmetrical lines and lines with Lorentzian profiles, and α-Fe foil was used as the calibration sample. The values of isomer shifts are quoted relative to that of α-Fe at 300 K.

Magnetometry was performed using a superconducting quantum interference device (SQUID) magnetometer (Quantum Design MPMS-XL 5). Temperature-dependent zero-field-cooled (ZFC) and field-cooled (FC) magnetization curves were recorded as follows: powder samples were pressed into polycarbonate capsules and introduced in to the SQUID magnetometer at room temperature, then cooled down to 5 K with no applied magnetic field after a careful degaussing procedure. Then, a magnetic field of 7.5 mT was applied, and the ZFC magnetization curve was recorded upon heating from 5 to 400 K. The sample was then cooled down to 5 K under the same applied field, and the FC magnetization curve was recorded upon heating from 5 to 400 K. In-plane zero-field-cooled (ZFC) and field-cooled (FC, H = 70 kOe) hysteresis loops [M(H) curves] were measured at 10 and 300 K using a maximum field of 70 kOe. The coercive field (H_C) and the M_R/M_S ratio were extracted from ZFC M(H) curves. The exchange bias field (H_E) was measured in the FC M(H) curves. Magnetization saturation (M_S) was measured from the hysteresis recorded at 5 K and was determined after correcting for the mass of organic ligands obtained from the TGA experiments.

Results

Multicomponent nanoparticles based on metal oxides were synthesized by a three-step seed-mediated growth process. Iron(II) stearate (FeSt₂) and cobalt(II) stearate (CoSt₂) were decomposed alternatively in high-boiling-temperature solvents (around 300 °C) in order to form successively C, CS, and CSS nanoparticles (Fig. 1).²⁸ In the third step, the amount of FeSt₂ was increased to synthesize CSSa, CSSb, and CSSc nanoparticles (see experimental part for details).

Fig. 1 Schematic illustration of the synthesis of CSS nanoparticles by performing the three-step seed-mediated growth process based on the successive thermal decomposition of Fe(II) and Co(II) precursors.

According to the TEM micrographs (Fig. 2), the shape of the nanoparticles was nearly spherical with narrow size distributions gradually shifting to larger average diameters after each thermal decomposition step from 10.1 to 15.6 nm (Table 2). Such size variations corresponded to an additional shell thickness of 2.0 nm for CS, while it was much thinner for CSSa (0.3 nm), CSSb (0.6 nm), and CSSc (0.8 nm). These values were smaller than the cell parameters of the iron oxide spinel, which would lead to an inhomogeneous growth (incomplete coating), as indicated by the shape of the nanoparticles slightly deviating from the spheres and by the broadening of their size distribution. Indeed, their surface was made up of facets, i.e., crystalline planes, with different surface energies yielding different kinetics for the growth of the components.³³

Fig. 2 TEM images of (a) CS, (b) CSSa, (c) CSSb, and (d) CSSc and the corresponding size distributions (bottom).

Table 2 Structural and chemical characteristics of the nanoparticles

	C	CS	CSSa	CSSb	CSSc
Diameter (nm)	10.1 ± 1.1	14.0 ± 1.5	14.5 ± 1.5	15.1 ± 1.7	15.6 ± 2.3
Additional layer thickness (nm)	—	2.0	0.3	0.6	0.8
Fe : Co at. ratio by EDX	—	45 : 55	57 : 43	68 : 32	73 : 27
Hydrodynamic diameter (nm)	12	16	18	21	21
Cell parameter (Å)	8.37(9) ± 0.01	8.40(9) ± 0.01	8.39(1) ± 0.01	8.40(1) ± 0.01	8.41(2) ± 0.01
Crystal size (nm)	8.0 ± 0.1	9.1 ± 0.1	11.4 ± 0.1	12.2 ± 0.1	12.7 ± 0.1

---

The chemical composition of the nanoparticles was also studied by energy dispersive X-ray spectrometry (EDX). The Fe : Co atomic ratio increased progressively from CS to CSSa, CCSb, and CSSc (Table 2), as expected by the growth of a thicker iron oxide shell. The spatial distribution of Fe, Co, and O atoms was further investigated by performing electron energy-loss spectroscopy – spectral imaging (EELS-SI). Although Co was distributed around the entire surface of the iron oxide core in CS, some areas corresponded to higher loadings (Fig. 3a and e), indicating the preferential growth of CoO as a thicker shell on specific facets. Although this was similar in CSSa, the distribution of CoO appeared to be progressively more uniform in CSSb and CSSc, indicating the redistribution of Co cations as the amount of the Fe precursor increased.

Fig. 3 Electron energy-loss spectroscopy – spectral imaging (EELS-SI) images at the Fe L-edge (top), Co L-edge (bottom). (a and e) CS, (b and f) CSSa, (c and g) CSSb and (d and h) CSSc nanoparticles.

High-resolution STEM micrographs were recorded in dark-field mode to study the crystal structure of the nanoparticles (Fig. 4). Continuous and straight lattice fringes with no defects were observed across each nanoparticle, regardless of its composition (CS and all CSS particles). A doubling of the lattice fringes periodicity could be observed in some areas, which, according to the FFT, corresponded to the reflections of the spinel and Wüstite structures (Fig. 4). The interpenetration of the two lattice fringes was consistent with the epitaxial growth of the Wüstite shell (CoO) from the surface of the spinel core (Fe_3−δO₄), as expected from their similar cubic structures and low lattice mismatch. The cell parameter of Fe₃O₄ (8.396 Å, JCPDS card no 19-062) was almost twice that of CoO (4.26 Å, JCPDS card no 00-048-1719), giving a lattice mismatch of 1.5% between the two phases. This was also confirmed by the geometrical phase analysis (GPA), which evidenced a lattice difference of 1–2% for the CoO compared to the Fe_3−δO₄, indicating very low residual strains in the CS and CSSc nanoparticles (Fig. 4b and f).

Fig. 4 (a, c, d, and e) High-resolution dark-field STEM images with specific FFT corresponding to the selected zone in the HR STEM images. In the inset of the main panel: the corresponding EELS elemental maps of Fe (red) and Co (green) for: (a) CS, (c) CSSa, (d) CSSb, and (e) CSSc. (b and f) Geometrical phase analysis (GPA) showing the variation of the lattice mismatch of (a) and (b), respectively, along the direction of the dashed blue arrow. The periodicity of the spinel phase lattice fringes was taken as a reference for performing the GPA.

The crystal structure of the nanoparticles was also studied by X-ray diffraction (XRD) (Fig. 5). The XRD patterns displayed peaks that could be indexed to the spinel structure. Additional peaks corresponding to the Wüstite phase were also observed, such as the one around 36.5°, in the CS and CSS nanoparticle patterns. The intensity of this peak was markedly reduced in CSSa and then gradually decreased in CSSb and CSSc, compared to the (311) reflection of the spinel structure. Concurrently, the spinel peaks became narrower, which was ascribed to the increase in the crystal sizes from 8.0 nm for C up to 9.1, 11.4, 12.2, and 12.7 nm for CS, CSSa, CSSb, and CSSc, respectively. This was consistent with the increase in the nanoparticle size observed in the TEM micrographs, while also indicating a high crystalline coherence of the spinel phase in the growth of the successive layers. The lower values of the crystal sizes compared to the TEM nanoparticle size may be also explained by the 2D projection of the faceted nanoparticles on the TEM micrographs, which tends to overestimate their size. While the unit cell parameter for C (8.379 Å) was in agreement with the partial oxidation of magnetite into maghemite, the values extracted for the other samples were higher than the lattice parameter of magnetite (a = 8.396 Å, JCPDS card no 19-062).³⁴ This was ascribed to the strain induced by the lattice mismatch between the spinel and Wüstite phases (2a_CoO = 8.52 Å, JCPDS card no 78-0431) and to the higher Fe²⁺ content, as previously observed in related studies.^35,36

Fig. 5 XRD patterns of C, CS, CSSa, CSSb, and CSSc nanoparticles. The red and black shapes at the bottom show the reference reflections of Fe₃O₄ (JCPDS card no 19-062) and CoO (JCPDS card no 78-0431) phases, respectively. hkl index refers to the spinel phase.

These results were confirmed by FTIR (see ESI†), which shows that the band related to the M−O vibrational modes was shifted to shorter wavelengths (from 599 cm⁻¹ for CS to 582 cm⁻¹ for CSSc) when the amount of FeSt₂ increased. Such a shift for the CSS samples, shorter than that of CoFe₂O₄ (591 cm⁻¹), was in agreement with the increasing content of Fe₃O₄ (574 cm⁻¹) as a second shell on the surface of the nanoparticles. The bands recorded for CSSb and CSSc mostly overlapped, which was consistent with a possible concomitant growth of the CoFe₂O₄ and Fe_3−δO₄ shells. Nevertheless, CSSb and CSSc did not display similar structures according to the above-mentioned results (longer coherence length of the spinel phase, higher Fe/Co ratio).

In order to obtain additional information on the environment and valence state of iron cations, ⁵⁷Fe Mössbauer spectrometry was performed at 77 K. All the spectra consisted of magnetic sextets composed of asymmetrical and broad lines (Fig. 6). The C and CS spectra showed additional peaks corresponding to quadrupolar doublets ascribed to the presence of superparamagnetic fluctuations³⁶ of the smallest particles in these systems. The spectral refinement led to several components associated with the hyperfine parameters that could well describe the hyperfine structure. The isomer shift (IS), hyperfine field (B_hf), and quadrupole shift (2ε) describe the oxidation state, magnetic environment, and local electronic structure of ⁵⁷Fe atoms, respectively (see ESI†). Although spectra can be refined by different combinations of components, the mean values of each hyperfine parameter are invariant and independent on the fitting model.

Fig. 6 Mössbauer spectra recorded at 77 K in a zero-field condition. The red line represents the Fe²⁺ content.

The mean isomer shift values at 77 K can be compared to the theoretical values of pure magnetite (0.61 mm s⁻¹), maghemite (0.40 mm s⁻¹), and cobalt ferrite (0.45 mm s⁻¹) in order to determine the fraction of Fe₃O₄ in each type of nanoparticle.^37–39 By considering a linear relationship between the experimental and the theoretical values, the Fe₃O₄ fraction vs. the Fe²⁺-deficient phases (γ-Fe₂O₃ and CoFe₂O₄) can be calculated. The increase in the mean isomer shift from C (0.504 mm s⁻¹) to CS (0.528 mm s⁻¹) indicated a higher content of Fe²⁺ in CS, as the CoO shell hindered the oxidation of Fe²⁺ when the nanoparticles were exposed to air. Thus, the Fe₃O₄ fraction increased from 34% in C to 49% in CS. The mean B_hf also increased from C (47.5 T) to CS (48.5 T), which was attributed to the presence of Co in the vicinity of Fe atoms,^40,41i.e., due to the formation of an intermediate layer of Co-doped ferrite at the Fe_3−δO₄/CoO interface.^7,23,24,42

For CSSa nanoparticles, the decrease in the mean isomer shift (0.489 mm s⁻¹) indicated a lower relative content of Fe²⁺ compared to Fe³⁺, which was consistent with the larger fraction of CoFe₂O₄ indicated by FTIR and XRD. This confirmed the increase in the mean B_hf value above 51 T. However, the increase in the mean isomer shift of CSSb (0.510 mm s⁻¹) corresponded to a larger amount of Fe²⁺. Finally, the isomer shift was decreased for CSSc (0.485 mm s⁻¹), although the component tentatively ascribed to Fe²⁺ was stable (3% of the relative sub-spectral area) while that of Fe^2-3+ increased from 5% to 8% (see ESI†). Considering these values, Fe₃O₄ accounted for 24% of CSSa, 38% of CSSb, and 22% of CSSc. The mean B_hf values of CSSb and CSSc remained rather constant above 51 T, showing the stability of CoFe₂O₄ regardless of the amount of Fe precursor used to grow the second shell.

Soft X-ray absorption spectroscopy (XAS) and X-ray magnetic circular dichroism (XMCD) experiments were performed to discriminate the site occupancies and oxidation states of Fe and Co cations (Fig. 7). The isotropic XAS spectra recorded at the Fe L_2,3 and Co L_2,3 edges were typical of the spinel structure, which contained both cations.^43,44 The Fe²⁺/Fe³⁺ ratio in the octahedral (O_h) sites could be qualitatively determined from the intensity ratio I₁/I₂ of the peaks labeled in Fig. 7a, see Table 3,⁴³ and markedly increased from C (0.56) to CS (0.82), indicating a higher content of Fe²⁺.^23,36 The value of CS being higher than that of pure Fe₃O₄ was ascribed to Fe²⁺ in the Wüstite phase (FeO), as suggested by Mössbauer spectrometry. In contrast, the ratio calculated for CSSa (0.52) was lower than that measured in C. Then, it increased to 0.61 for CSSb and 0.65 for CSSc, which were indicative of the higher content of Fe²⁺.

Fig. 7 (a and c) XAS and (b and d) XMCD spectra recorded for C, CS, CSSa, CSSb, and CSSc nanoparticles at the (a and b) Fe L_2,3 edges and (c and d) Co L_2,3 edges.

Table 3 Intensity ratios of the peaks calculated from the XAS and XMCD spectra and the corresponding volume fractions

Sample	C	CS	CSSa	CSSb	CSSc	Fe3O4	γ-Fe2O3	CoFe2O4
I 1/I2 ratio	0.56	0.82	0.52	0.61	0.65	0.71	0.35	0.35
S ratio	0.83	0.95	0.77	0.89	0.95	1.14 (ref. 48)	0.69 (ref. 48)	—
Fe3O4 (%)/γ-Fe2O3 (%)	21/79	43/57	12/88	30/70	43/57	100/0	0/100	0/100
CoFe2O4 (%)/CoO (%)	—	23/77	42/58	54/46	68/32

---

XMCD spectra were also recorded at the Fe L_2,3 and were typical of the reverse spinel structure. The peaks indexed as S1, S2, and S3 corresponded to Fe²⁺ and Fe³⁺ in the O_h sites, Fe³⁺ in the tetrahedral sites (T_d), and Fe³⁺ in the O_h sites, respectively. The ratio S = (S1 + S2)/(S2 + S3) displays a similar behavior to that of the I₁/I₂ ratio (Table 3).⁴⁵ Considering the values ascribed to pure Fe₃O₄ (1.27)⁴⁴ and γ-Fe₂O₃ (0.69),³¹ the extracted S values indicated an Fe²⁺ deficiency, ascribed to the formation of γ-Fe₂O₃ and CoFe₂O₄. In CS, we expect that the Fe²⁺ deficient fraction (57%) could be mostly ascribed to CoFe₂O₄, which resulted from the diffusion of Co²⁺ in the O_h vacancies on the surface of the Fe_3−δO₄ nanoparticles.^7,23 Moreover, the significant decrease in the Fe²⁺ fraction in CSSa (12%) did not result from the oxidation of the core (protected by the CoO shell), but from a higher amount of CoFe₂O₄. In CSSb and CSSc, the thicker Fe_3−δO₄ shells resulted in a significant increase in the Fe²⁺ content (30% and 43%, respectively). These values were much higher than expected, because Fe²⁺ on the nanoparticle surface would be oxidized, as has been generally reported for Fe_3−δO₄ nanoparticles.⁴⁶

It is worth noting that the I₁/I₂ ratio calculated from the XAS spectra of CS was higher than that for Fe₃O₄, which could be attributed to the presence of an additional fraction of Fe²⁺ in the Wüstite phase, as observed recently.²⁴ Therefore, we expect the Wüstite shell to consist of Co_1−xFe_xO resulting from the co-crystallization of Co²⁺ and Fe²⁺, with the latter resulting from partial solubilization at the early stages of the Wüstite shell formation. As the CoO shell is also partially solubilized during thermal annealing in a liquid medium, this resulted in a strong decrease in the I₁/I₂ for CSSa. In contrast, the S value was lower than that of pure Fe₃O₄ and C due to the formation of CoFe₂O₄. Therefore, the increase in the S ratio in CSSb and CSSc agreed with the presence of higher Fe²⁺ contents in the iron oxide shell. This is particularly true because 70% of the XMCD signal came from the first 2 nm of the nanoparticle surface.

In the XAS spectra recorded at the Co L_2,3 edges, the intensities of peaks I₃ and I₄ refer to the distribution of Co²⁺ in the Wüstite and spinel phases, respectively.^31,47 Therefore, the I₄/I₃ ratio evidenced the progressive conversion of CoO to CoFe₂O₄ from CSSa to CSSc.^23,36 The XMCD spectra also recorded at the Co L_2,3 edges showed a negative peak S4, corresponding to Co²⁺ in the O_h sites of the spinel structure.^36,44 The intensity of this peak markedly increases from CS to CSSc, which also supported the increasing fraction of CoFe₂O₄, in agreement with the EDX, XRD, and FTIR data. Note that magnetically compensated spins in antiferromagnets do not contribute to the XMCD signal. Nevertheless, XMCD is sensitive to uncompensated Co spins of CoO, which are typically localized at the nanoparticle surface⁴⁸ or coupled to Fe spins at the interface.^23,24 The normalization of the XMCD spectra by the XAS signal allows determining the fraction of uncompensated spins. Here, the intensity of the S4 peak yielded 23% of the uncompensated Co spins in CS, which mainly consisted of a CoFe₂O₄ layer localized at the Fe_3−δO₄/CoO interface.²⁴ In CSS nanoparticles, the S4 peak indicated there were much higher amounts of uncompensated Co spins (42%, 54%, and 68% in CSSa, CSSb, and CSSc, respectively). This result unambiguously confirmed the increasing fraction of CoFe₂O₄ in the nanoparticles at the expense of CoO, completing a consistent structural picture of the particles, as graphically summarized in Fig. 8.

Fig. 8 Schematic illustration of the chemical structure of CSS nanoparticles. The last (rightmost) cartoon depicts the suggested hybrid ferrimagnet, which develops progressively from the CSSa stage.

Magnetic properties

As mentioned above, element-specific magnetization curves were registered to study the magnetic structure with respect to site occupancy and the oxidation states of Fe and Co cations (Fig. 9). The M(H) curves recorded at 4K displayed very similar coercive fields (H_C) when measured at the different energy edges of Fe and Co (see ESI† for the detailed values), thus demonstrating the coherent reversal and, thus, the strong exchange coupling between the Fe and Co cations in all the spinel structures. The mean H_C values were similar for all the CS and CSS nanoparticles (10.9, 10.8, 10.9, and 9.6 kOe for CS, CSSa, CSSb, and CSSc, respectively), that, at these low temperatures, the coercivity was mostly determined at the core–shell interface by exchange coupling of the core to similarly anisotropic Co-based phases across the series. Yet, the clear reduction in H_C in CSSc was consistent with the larger fraction of magnetically soft Fe_3−δO₄ in the outer shell leading to a doubly exchange-coupled “soft core/hard layer/soft layer” system.³⁶ However, the main factor behind this reduction was possibly the modification (with the progressive conversion of CoO into Co-ferrite) of the nature of the inner interface. These coercivity values were larger than those observed in many similar core/shell nanoparticles comprising magnetite and Co oxides; for instance, Fe_3−δO₄@CoO nanoaggregates of 9.9 nm (H_C = 3 kOe),⁷ magnetite-doped cobalt ferrite nanoparticles of 40.3 nm (H_C = 2.1 and 3.9 kOe at Fe and Co edges),⁴⁴ and even exceed those we reported recently for optimized Fe_3−δO₄@CoFe₂O₄@Fe_3−δO₄ (9.6 kOe).³⁶ There are few reports of similarly high coercivity in Co-ferrite-based nanoparticles of similar sizes.⁴⁹

Fig. 9 Element-specific magnetization curves recorded at 4K by XMCD between −6.5 T and +6.5 T at fixed energies corresponding to the maximum of S1, S2, S3 peaks (Fe L_2,3 edges) and to the S4 peak (Co L_2,3 edges) for (a) CS, (b) CSSa, (c) CSSb, and (d) CSSc.

The magnetic properties were also investigated by SQUID magnetometry. Fig. 10a presents the low-field magnetothermal curves recorded after zero-field-cooling (ZFC) and field-cooling (FC). The peak temperature of the ZFC curve, T_max, is often taken as the “blocking temperature” of the system in the measurement (magnetometry) timescale, which corresponds to the thermal energy comparable to the magnetic anisotropy energy barrier (kV ≈ 25 k_BT). However, the blocking temperature (T_B) is described more realistically as the median of a distribution of energy barriers, which can be extracted from the FC-ZFC difference as:⁵⁰

f(T_B) = d(M_FC − M_ZFC)/dT

Fig. 10 (a) Low-field magnetothermal field-cooled (FC) and zero-field-cooled (ZFC) curves recorded for CSSa, CSSb and CSSc. (b) Magnetization FC curves. (c) Distribution of blocking temperatures. The dashed line indicates the position of the M_FC dip.

This distribution was shifted to higher temperatures by interparticle dipolar interactions in the nanoparticle powders (Fig. 10c).^51,52 However, since the interparticle interactions were roughly similar in all samples, the observed evolution of the median blocking temperature (or T_max for the same reason) across the series mainly reflected variations in the average particle anisotropy barrier.⁵³

The pristine Fe_3−δO₄ nanoparticles displayed a T_max of 150 K and T_B of 93 K, in agreement with the reported values for iron oxide nanoparticles of 10 nm.⁴⁶ These values were strongly increased for CS (T_max = 290 K, T_B = 266 K), indicating an increase of the effective magnetic anisotropy energy (K_efV), which unequivocally reflected the exchange coupling at the soft/hard interface.⁵⁷ Remarkably, T_max in CS was roughly equal to the Néel temperature of CoO (T_N = 290 K), indicating that this shell was thick enough (at least 2 nm) to support thermally stable CoO grains up to the ordering temperature.^18,54 Although some CoO grains in CS were thermally stable up to T_N, a fraction of them started to fluctuate at ≈200 K (alternatively or concurrently, the AFM order in thinner shell regions may be lost due to size effects),⁵⁴ below which all particles were pinned: note the flat shape of the ZFC curve below that temperature, as previously described in other biased NP systems.^18,55 A similar behavior, modulated by the relative fraction of the remaining CoO phase, was also observed for CSSa. The ZFC curves of CSS remained flat for a wide range of low temperatures, which extended to higher temperatures for CSSa to CSSc. While the onset of the ZFC magnetization in CSSa was relatively gradual (the derivative in Fig. 10c shows that two overlapping mechanisms were at play), in CSSb and CSSc the onset was sharper and took place at about 290 K. The FC curves of these three samples are plotted separately in panel (b) to highlight the presence of a small but clearly visible dip for CSSa, signaling the loss of CoO-related exchange coupling at about 230 K. This feature vanished progressively with the decreasing amount of CoO in CSSb and CSSc. Yet, in these two systems the volume of CoO still amounted to nearly one-half and one-third, respectively, of the Co-containing species (see Table 3). This suggests that in CSSb and CSSc the remaining, possibly discontinuous, layer of CoO was so thin that it was effectively stabilized via proximity effects by the surrounding Co-ferrite layer, thus becoming a single magnetic object with properties in between those of CoO and Co-ferrite. This explains the vanishing dip in the FC curve (attributed to the “individual” magnetism of the CoO phase) while preserving, even enhancing, the median blocking temperature determined by the dual exchange coupling of the Fe oxide core and outer shell with the high-anisotropy hybrid Co–Fe–O shell in between them.

The magnetic response, M(H), measured at 300 K, for C and CS showed no hysteresis, thus confirming the room temperature superparamagnetism of these two samples (Fig. 11c). In contrast, CSS nanoparticles exhibited a remanent magnetization at 300 K, as expected from their higher blocking temperatures. Note that the single-loop aspect of all the hysteresis loops (no kinks, i.e., absence of a decoupled soft component, often observed as a low-field step)^56,57 showed that the Fe_3−δO₄, CoO, and CoFe₂O₄ phases were all fully-coupled and rotated coherently upon reversal as required in applications of exchange-coupled soft–hard magnetic systems.¹³

Fig. 11 Hysteresis loops measured at (a) 10 K after zero field cooling, (b) 10 K after cooling under a field of 7 T, and (c) at 300 K.

Hysteresis loops recorded at 10 K after cooling in zero-field showed a strong hysteresis in all the nanoparticles, except those measured in the simple C seeds (Fig. 11a). The coercive field of CS (16.4 kOe) was very high compared with similar Fe_3−δO₄@CoO nanoparticles reported in the literature.^22,23,58 It increased to ≈17.3 kOe for CSSa and CSSb (due to the hard Co-ferrite contribution), and decreased down to ≈15.0 kOe for CSSc, consistently with the growth of a thicker soft iron oxide shell.¹⁴ As expected, cooling down in a magnetic field of 7 T resulted in yet higher H_C values (Fig. 11b) due to the exchange coupling to uncompensated high-anisotropy (yet rotatable) spins at the soft–hard interfaces, an effect more apparent in the horizontal shift (exchange-bias field) due to the exchange-coupling with pinned uncompensated spins.^5,25 In contrast with the ZFC loops, the coercivity was higher in CS (19.8 kOe) than in the CSS samples (18.5 to 15.8 kOe). In other words, the exchange-coupling-induced H_C increase was significantly larger in the CS particles. This can be understood in terms of both the higher anisotropy of CoO (compared to CoFe₂O₄) and the large fraction of uncompensated spins (UCS) in the CoO phase. In any case, the H_C values of our CSS particles were higher than those reported earlier at low temperatures for core@multi-shell magnetic nanoparticles, such as FeO@Fe₃O₄@MnO@Mn₃O₄ (1.2 kOe)²⁵ or MnFe₂O₄@CoFe₂O₄@NiFe₂O₄ (7.7 kOe).²⁷ The M_R/M_S ratio of the ZFC curves increased from 0.24 (C) to 0.38 (CS) and further up to about 0.60 in CSS. Therefore, the hysteresis curves became squarer upon the gradual replacement of CoO by CoFe₂O₄ and the increase in the iron oxide shell thickness, concurrently with an increase in saturation magnetization, thus leading to larger energy products (BH)_max. The gradual removal of CoO (AFM) and substitution by Co-ferrite (FiM) were indeed clearly confirmed by the evolution of the saturation magnetization (M_S), which markedly increased from 41 (CS) to 72 (CSSc) emu g⁻¹ (Table 4), since the compensation of spins in the ordered fraction of the AFM structure resulted in a zero M_S.²²

Table 4 Magnetic parameters measured and calculated from the magnetization curves

	C	CS	CSSa	CSSb	CSSc
Diameter (nm)	10.1 ± 1.1	14.0 ± 1.5	14.5 ± 1.5	15.1 ± 1.7	15.6 ± 2.3
Thickness (nm)	—	2.0	0,3	0,6	0,8
H C 300 K (ZFC) (kOe)	0	0	0.5	0.3	0.5
H C 10 K (ZFC) (kOe)	0.4	16.4	17.5	17.2	15.0
H C 10 K (FC) kOe	0.4	19.8	18.5	17.9	15.8
H E 10 K (kOe)	0	5.3	2.8	0.9	0.5
T max (K)	150	290	≈400	>400	>400
T B (K)	93	266	315	343	343
M S at 5 K (ZFC) (emu g^−1)	60	41	51	55	72
M R/MS at 5 K (ZFC)	0.24	0.38	0.59	0.68	0.66

---

As anticipated above, all the FC M(H) curves were shifted to negative magnetic fields as a result of the exchange-bias phenomenon, namely the pinning of the ferrimagnetic (FiM) phase magnetization by the uncompensated spins of the AFM phase, which aligned with the cooling field, establishing a unidirectional anisotropy. This exchange-bias field (H_E), defined as the loop shift, was largest for CS (5.3 kOe) and decreased gradually down to 0.5 kOe for CSSc, consistently with the progressive substitution of the CoO antiferromagnet by CoFe₂O₄, as concluded from the structural characterization, together with a concurrent loss of magnetic anisotropy in CoO through proximity effects with the Co-ferrite in the CSSb and CSSc samples, as proposed above. Nevertheless, a fraction of CoO remained in each CSS nanoparticle, as observed by XRD and XAS/XMCD. In the CS nanoparticles, the significant vertical shift of the hysteresis loop evidenced a correspondingly large fraction of fully pinned uncompensated spins (which do not rotate with the core magnetization). These are the spins responsible for the strong biasing of the core magnetization reflected in the high H_E value. In contrast with this FiM(soft)–CoO coupling, the FiM(soft)–FiM(hard) coupling did not introduce unidirectional anisotropy (the CoFe₂O₄ component essentially lacks such pinned UCS), as evidenced by the strong decrease in H_E, but the large anisotropy of the CoFe₂O₄-CoO hybrid phase proposed above (majority in the CSS samples), combined with its high ordering temperature (in contrast with CoO), which stabilizes the overall moment beyond room temperature. Indeed, the H_E(T) curves show that H_E vanished at gradually lower T_onset values from CS (200 K) to CSSc (20 K) (Fig. 12). The pronounced reduction in both H_E (10 K) and T_onset even for CSSa (for which we estimated above a modest decrease in CoO content from 63% to 51%, see Table 1) suggested that the conversion of CoO into CoFe₂O₄ was accompanied by the loss of UCS in the remaining CoO, a likely effect of the recently reported effective annealing provided by the third decomposition step.²⁴ The same effects can be expected from a progressive hybridization of the shrinking CoO layer or interface islands with the Co-ferrite phase via proximity effects. Note that we have previously observed no H_E values in core/shell Fe₃O₄/CoFe₂O₄ nanoparticles.³⁶

Fig. 12 Temperature dependence of the coercive field (black squares) and exchange field (blue triangles) of (a) CS, (b) CSSa, (c) CSSb, and (d) CSSc.

The decreasing number of CoO UCS (both pinned and rotatable) was confirmed by the difference between the H_C values [H_C(FC) − H_C(ZFC)], which was highest for CS (3.4 kOe) and decreased markedly for the CSS particles (≤1 kOe). In short, the concurrent structural ordering of the CoO phase and its hybridization with Co-ferrite from CS to CSSc were consistent with the large reduction in H_E across the CS–CSS series (roughly an order of magnitude) while essentially preserving the high coercivity (with a mere 20% decrease in the 10 K loops).

In the CSS samples, Fig. 12 shows that H_E vanished at temperatures (T_onset) significantly lower than the FC feature signaled by the vertical line in Fig. 10, ascribed to the residual fraction of CoO, which still preserves its individual properties (c.f. the remaining CoO was hybridized with Co-ferrite via a proximity effect). Although the presence of such a residual fraction (decreasing from CSSa to CSSc) leaves a subtle fingerprint in the low-field M(T) curves (possibly at its Néel temperature), it was not capable of biasing the hysteresis loop above 100 K (less for CSSc). These relatively low T_onset values in the CSS particles indicated that the remarkable magnetic stabilization of the CSS particles up to (at least) 400 K was not due by exchange coupling to CoO (see also the strong coercivities measured well above T_onset), but was rather driven by exchange coupling to an intermediate CoO–CoFe₂O₄ hybrid shell stemming from a proximity effect from these two compounds, where, conveniently, CoO contributed a high anisotropy, and Co-ferrite a relatively large saturation magnetization. Then, the magnetic characterization of the samples gave rise to yet another scenario, as summarized in the last figure in Fig. 8.

Discussion

Although each thermal decomposition step resulted in the growth of additional layers onto the surface of the nanoparticles, a far more complex chemical structure than expected was revealed by a the results from a wide range of complementary techniques. Regarding the CS nanoparticles studied here, a number of new results were obtained with respect to previous studies on Fe_3−δO₄@CoO core–shell nanoparticles. First, we confirmed that the growth of the CoO shell partially preserved the Fe_3−δO₄ core from the oxidation of Fe²⁺ when the nanoparticles were exposed to air. Second, in the HRTEM micrographs the Wüstite phase was identified to grow epitaxially with low strain at the surface of the spinel phase, which may explain the efficient exchange coupling yielding unprecedentedly strong coercivities in this system. Third, EELS-SI showed that CoO did not grow uniformly at the surface of the Fe_3−δO₄ nanoparticles, which was attributed to the preferential growth on certain facets driven by the different surface energies. Fourth, Co²⁺ cations diffused at the surface of the Fe_3−δO₄ core to fill vacancies in the O_h sites, as shown by the Mössbauer and XMCD measurements. Therefore, in contrast with the simple core–shell scenario often assumed in the literature, i.e., well-defined core–shell interfaces, we demonstrated a complex structure involving three phases, namely Fe_3−δO₄@CoFe₂O₄@CoO with a discontinuous CoO shell.

The structure of CSS nanoparticles is yet more complex. The Fe/Co atomic ratio measured by EDX indicated the formation of Fe species at the surface of CS nanoparticles. By considering a simple spherical core–shell–shell model, hypothetical Fe_3−δO₄ outer shells with mean thicknesses similar to those obtained from TEM micrographs were calculated. In CSSa, the small size increase resulted in an effective shell thickness of 0.3 nm, less than the Fe₃O₄ lattice parameter (8.396 Å). A non-uniform growth of Fe₃O₄ resulting in a discontinuous shell was expected, although a partial solubilization of the CoO shell followed by recrystallization certainly took place (see below), as we reported very recently.²⁴ When the mean shell thickness increases, the shell is expected to be continuous. Although the EELS-SI micrographs showed that the spatial distribution of Fe and Co cations in CSSa was similar to that of CS, it became more homogeneous in CSSb and CSSc. These results suggest the progressive formation of CoFe₂O₄ at the expense of CoO. The Fe/Co signal ratio was also higher than for CSSa, confirming the formation of an Fe_3−δO₄ shell, which became thicker with increasing the amount of the Fe precursor. In fact, XRD confirmed the gradual disappearance of CoO and a concomitant increase in the spinel crystal size with increasing the amount of the Fe precursor used to grow the outer shell in the CSS system. The increase in the lattice parameter was consistent with the higher Fe²⁺ content and the strains resulting from the presence of CoO and a high content of defects in CoFe₂O₄.¹⁵ Nevertheless, CoFe₂O₄ resulted in lower strains, as shown for CSSc, in comparison with CS, which agreed with the better epitaxial growth expected between Fe_3−δO₄ and CoFe₂O₄. The shift in the FTIR band specific for metal oxides to shorter wavelengths consistently supported a significant increase in CoFe₂O₄ as well as a higher Fe²⁺ content in the outer shell. Although the shift was significant from CSSa to CSSb, the band of CSSc was centered at a very similar wavelength (582 cm⁻¹), suggesting the concomitant growth of CoFe₂O₄ (591 cm⁻¹) and Fe₃O₄ (574 cm⁻¹), with the latter compensating the former, as otherwise it would be closer to the wavelength of pure magnetite (574 cm⁻¹).

Although these results suggest the growth of CoFe₂O₄, it could not be clearly discriminated from Fe_3−δO₄ and CoO. Mössbauer spectrometry evidenced unequivocally the significant increase in CoFe₂O₄ at the expense of CoO in CSSa by the reduced mean isomer shift in comparison to CS. In contrast, the higher value calculated for CSSb indicated the formation of an Fe_3−δO₄ shell with a significant Fe²⁺ content. In CSSb, the formation of Fe_3−δO₄ predominated over that of CoFe₂O₄. The I₁/I₂ and S ratios calculated from the XAS and XMCD spectra showed the same evolution of Fe²⁺vs. Fe³⁺ as for the Mössbauer data, except for CSSc; whereby XAS and XMCD showed that the Fe²⁺ content was higher in CSSc than CSSb, while the Mössbauer results showed the opposite. This discrepancy can be explained by the higher surface sensitivity of soft X-rays, with 70% of the XAS and XMCD signal coming from a 2 nm surface layer (thereby, Fe²⁺ at the CSSc surface would be overestimated). The lower mean isomer shift in CSSc suggested a higher amount of Fe²⁺-deficient phases. According to the XAS and XMCD spectra, the increase in both the I₁/I₂ and S ratios rules out the oxidation of Fe²⁺ as a possible cause of the reduced mean isomer shift of CSSc. In CSSc, the replacement of CoO by CoFe₂O₄ predominated over the growth of the Fe_3−δO₄ outer shell. This result was confirmed by the M−O bands (FTIR spectra) of CSSc, which mostly overlapped with that of CSSb, with the increase in the Fe : Co ratio measured by EDX and the relatively homogeneous distribution of Co and Fe as shown by the EELS-SI micrographs. It is worth noting that in comparison to Fe_3−δO₄ nanoparticles, the surfaces of CSSb and CSSc seemed to contain a relatively high amount of Fe²⁺, although given the thickness of the outer shell, this should be mostly oxidized. The presence of CoFe₂O₄ in all CSS nanoparticles was also supported by their mean B_hf fields being significantly higher (about 51 T) than that of Fe_3−δO₄ nanoparticles (about 47 T). Finally, the XMCD spectra recorded at the Co L-edge unambiguously demonstrated the presence of CoFe₂O₄, which increased gradually up to 68% in CSSc at the expense of CoO. Although most of the CoO was converted into CoFe₂O₄ in CSSc, the CoO fraction still remained significant (32%), and has been shown to impact the magnetic properties of particles, even at room temperature.

As we reported elsewhere,²³ the formation of CoO in the second thermal decomposition step favored the diffusion of Co cations into the vacancies at the surface of the Fe_3−δO₄ nanoparticles, i.e., the formation of an interlayer of Co-doped ferrite at the Fe_3−δO₄ @CoO interface. Due to the diffusion process, a concentration gradient of Co²⁺ was expected from the Fe₃O₄/CoO interface. The gradual increase in the CoFe₂O₄ fraction in CSS was directly related to the experimental conditions upon performing the third thermal decomposition process. The CoO shell is highly unstable in liquid medium at such high temperatures.²⁴ Therefore, we expected that the CoFe₂O₄ phase in CSS nanoparticles resulted from the partial solubilization of the CoO shell during the third decomposition step, which was followed by the recrystallization of the solubilized Co species with the new Fe monomers (resulting from the decomposition of the Fe precursor). Here, CoO acted as a reservoir of Co monomers, enabling the growth of a CoFe₂O₄ shell before that of Fe₃O₄. This was clearer for low amounts of Fe precursor, e.g., CSSa. However, for larger amounts of Fe precursor, the increasing volume fraction of CoFe₂O₄ could not be caused solely by the solubilization of CoO, as this would imply that increasing the Fe monomers in the reaction medium enhanced the solubilization of CoO. Therefore, we expect that above a critical amount of Fe precursor, the Fe_3−δO₄ shell grows at the surface of nanoparticles. Considering the high temperatures involved (about 300 °C), Co²⁺ could be expected to diffuse at the CoO/Fe_3−δO₄ interface in order to partially replace CoO by CoFe₂O₄. In the case of CSSb and CSSc, the Fe precursor amount seemed to be sufficiently high to result in an Fe_3−δO₄ shell.

By combining the Mössbauer and XAS/XMCD results, we calculated the relative volume fraction of each phase in each nanoparticle (Table 5). To simplify our model, we considered that all the vacancies in Fe_3−δO₄ are filled by Co²⁺, resulting in a stable CoFe₂O₄ intermediate layer, given the self-limitation of cationic diffusion (Table 1). Considering the structure of CSS nanoparticles, we expect that the Fe_3−δO₄@CoFe₂O₄ structure would not be affected by the third thermal decomposition step, in contrast with the CoO shell.²⁴ The CoO content decreased gradually from 63% in CS to 26% in CSSc, while that of CoFe₂O₄ increased from 19% to 55%. Finally, the Fe₃O₄ fraction remained stable (≈18%) due to the compensation in CSSb and CSSc between the formation of Fe_3−δO₄ at the outer shell and that of CoFe₂O₄ at the inner shell/outer shell interface. In CSSa, the extremely thin outer shell could not provide such compensation, explaining the lower fraction of Fe_3−δO₄ (12%).

Table 5 Relative volume fractions (%) of Fe₃O₄, CoFe₂O₄, and CoO calculated from the Mössbauer and XAS/XMCD results obtained for the CS, CSSa, CSSb, and CSSc nanoparticles

Compound	CS	CSSa	CSSb	CSSc
Fe3O4	18	12	18	19
CoFe2O4	19	37	44	55
CoO	63	51	38	26

---

The magnetic properties were explained in correlation with the chemical composition and structure of the nanoparticles. The replacement of CoO by CoFe₂O₄ progressively removed the limitations in exchange coupling imposed by the relatively low AFM ordering temperature of CoO (the FiM order of CoFe₂O₄ remained well above room temperature, up to T_C = 790 K). In fact, the ultrathin islands of CoO remaining in CSSb and CSSc enabled magnetic proximity effects with the neighboring Co-ferrite, rendering a hybrid Co-based phase with a high anisotropy and ordering temperature. Hence, the magnetic properties of the CSS nanoparticles, particularly at room temperature, were no longer driven by a simple FiM/AFM exchange-bias coupling, where T_N (CoO) = 290 K, but by a “soft-FiM/hard–artificial FiM/soft-FiM” dual exchange-coupling in a core@hybrid–shell@shell structure. Indeed, the progressive replacement of CoO by CoFe₂O₄ in CSS resulted in blocking temperatures higher than room temperature. The concurrent increase in the saturation magnetization agreed with the estimated relative fractions of Fe_3−δO₄, CoO, and CoFe₂O₄. At low temperatures, the gradual reduction of the horizontal and vertical shifts measured in the FC hysteresis loops from CS to CSSc also reflected the progressive removal of the CoO phase. Only a small fraction of “independent” or interfacial spin disordered CoO remained, as shown by the residual exchange field observed at 10 K. Although the M(H) curves measured at 300 K showed moderate coercivity, they unambiguously verified that the magnetic moments of the CSS nanoparticles were blocked at room temperature in the timescale of the SQUID measurements.

It is worth noting that in the particular case of CSSa, where the CoO phase still corresponded to half of the nanoparticle volume but there was already a 37% content of CoFe₂O₄, two different exchange phenomena could be resolved from the thermal dependence of the magnetization (Fig. 10b and c), namely those between the core and the CoO phase (responsible for the still large exchange bias field at low temperatures) and between the core and the Co-ferrite (or initial stages of the hybridized CoO-Co-ferrite FiM), responsible for the increase in the blocking temperature up to at least 400 K.

Finally, the final magnetic scenario can be summarized as follows: (i) exchange coupling predominates in Fe_3−δO₄@CoO core–shell nanoparticles, even though a CoFe₂O₄ layer is already present at the core–shell interface; (ii) further increase in the CoFe₂O₄/CoO ratio in CSSa results in both Fe_3−δO₄-CoO and Fe_3−δO₄-CoFe₂O₄ couplings, thus two (un)blocking processes, as resolved in the double peak in d(Mfc-Mzfc)/dT curves; (iii) the significant reduction of the CoO fraction in CSSb and CSSc favors the proximity effects between CoO and CoFe₂O₄, leading us to consider the coupling between the Fe_3−δO₄ core and a hybrid Co shell; the latter combining both high anisotropy and ordering temperature, where CoO and Co-ferrite behave, as one as indicated by the fact that there was only one blocking process.

Conclusion

Small magnetic nanoparticles that were magnetically stable above 400 K were synthesized by a three-step synthesis growth process. Iron and cobalt precursors were alternately decomposed around 300 °C in order to form iron oxide particles first, onto which a CoO shell was successively grown, which then turned gradually into CoFe₂O₄ upon the further growth of a Fe_3−δO₄ outer shell. The thermal instability of the CoO shell during the third synthesis step led to its partial solubilization and Co diffusion at both the Fe_3−δO₄ core/CoO shell and CoO shell/Fe_3−δO₄ shell interfaces. The replacement of the AFM CoO phase by CoFe₂O₄ altered markedly the exchange coupling at the Fe_3−δO₄ core/CoO shell interface, although a significant but decreasing volume fraction of CoO remained present in the CSS series. The formation of a hard FiM CoFe₂O₄ inner shell resulted from the co-crystallization of the solubilized Co species and Fe monomers caused by the thermal decomposition of the Fe precursor and Co diffusion at the CoO/Fe_3−δO₄ interface. Two types of exchange coupling (core-AFM and core-FiM) were clearly observed (in the thermal dependence of the magnetization) in the only sample with a large content of both CoO and CoFe₂O₄ (CCSSa); whereas, for thinner layers of CoO (in CSSb and CSSc), a single blocking process is observed above room temperature, indicating the hybridization (magnetic proximity effects) of the two species into an artificial ferrimagnet with high anisotropy and ordering temperature, given that T_C (CoFe₂O₄) ≫ T_N (CoO). Besides growing a thicker Fe_3−δO₄ shell, increasing the amount of the Fe precursor accelerated the replacement of CoO by CoFe₂O₄. Furthermore, the concomitant formation of a CoFe₂O₄ inner shell and a thicker Fe_3−δO₄ outer shell by increasing the amount of Fe precursor resulted in dual interfacial coupling (soft core/hard shell/soft shell), preserving the coherent rotation of all the spins in the nanoparticles, thus allowing a further increase in the blocking temperature of CSSc above 400 K while increasing the saturation magnetization. This system constitutes one of the few examples of either single-phase or complex onion nanoparticles (as small as ∼15 nm) showing magnetic stability, i.e., blocked ferrimagnetism, well above room temperature. Hence, the complex structure of our iron oxide-based nanoparticles and the engineering of multiple exchange-couplings and magnetic proximity effects resulted in unprecedented magnetic stability. This strategy opens new perspectives for the design of rare-earth-free nanoparticles with tunable magnetic properties for a wide range of applications, including high density data storage, theranostics, and sensors.

Conflicts of interest

There are no conflicts to declare.

Acknowledgements

K. S. was supported by a PhD grant from the French Agence Nationale de la Recherche (ANR) under the reference ANR-11-LABX-0058-NIE within the Investissement d'Avenir program ANR-10-IDEX-0002-02 and SOLEIL synchrotron/Laboratoire Léon Brillouin fellowship. The authors are grateful to SOLEIL synchrotron for providing the access to DEIMOS beamline. J. A. D. T. and R. L.-M. acknowledge support from Junta de Comunidades de Castilla-La Mancha [SPBLY/21/180501/000226] and Ministerio de Ciencia, Innovación y Universidades [PID2022-142267NB-I00]. R. L.-M. acknowledges his predoctoral research grant co-funded by the European Social Fund [2020-PREDUCLM-16730].

References

1. S. Massari and M. Ruberti, Resour. Policy, 2013, 38, 36.
2. O. Gutfleisch, M. A. Willard, E. Brück, C. H. Chen, S. G. Sankar and J. P. Liu, Adv. Mater., 2011, 23, 821.
3. P. Tartaj, M. P. Morales, T. Gonzalez-Carreño, S. Veintemillas-Verdaguer and C. J. Serna, Adv. Mater., 2011, 23, 5243.
4. S. Bedanta and W. Kleemann, J. Phys. D: Appl. Phys., 2008, 42, 013001.
5. J. Nogues and I. K. Schuller, J. Magn. Magn. Mater., 1999, 192, 203.
6. J. Nogues, J. Sort, V. Langlais, V. Skumryev, S. Surinach, J. S. Munoz and M. D. Baro, Phys. Rep., 2005, 422, 65.
7. T. Gaudisson, R. Sayed-Hassan, N. Yaacoub, G. Franceschin, S. Nowak, J.-M. Grenèche, N. Menguy, Ph. Sainctavit and S. Ammar, CrystEngComm, 2016, 18, 3799.
8. W. Baaziz, B. P. Pichon, C. Lefevre, C. Ulhaq-Bouillet, J.-M. Greneche, M. Toumi, T. Mhiri and S. Begin-Colin, J. Phys. Chem. C, 2013, 117, 11436.
9. A. López-Ortega, E. Lottini, G. Bertoni, C. de Julián Fernández and C. Sangregorio, Chem. Mater., 2017, 29, 1279.
10. G. C. Lavorato, E. L. Jr, D. Tobia, D. Fiorani, H. E. Troiani, R. D. Zysler and E. L. Winkler, Nanotechnology, 2014, 25, 355704.
11. R. F. L. Evans, D. Bate, R. W. Chantrell, R. Yanes and O. Chubykalo-Fesenko, Phys. Rev. B, 2011, 84, 092404.
12. F. G. Silva, J. Depeyrot, Yu. L. Raikher, V. I. Stepanov, I. S. Poperechny, R. Aquino, G. Ballon, J. Geshev, E. Dubois and R. Perzynski, Sci. Rep., 2021, 11, 5474.
13. Q. Song and Z. J. Zhang, J. Am. Chem. Soc., 2012, 134, 10182.
14. A. López-Ortega, M. Estrader, G. Salazar-Alvarez, A. G. Roca and J. Nogués, Phys. Rep., 2015, 553, 1.
15. M. Pauly, B. P. Pichon, P. Panissod, S. Fleutot, P. Rodriguez, M. Drillon and S. Begin-Colin, J. Mater. Chem., 2012, 22, 6343.
16. J. B. Tracy, D. N. Weiss, D. P. Dinega and M. G. Bawendi, Phys. Rev. B: Condens. Matter Mater. Phys., 2005, 72, 064404.
17. C. N. R. Rao and B. Raveau, Transition Metal Oxides, Wiley-VCH, Germany, Weinheim, 2nd edn, 1995.
18. J. A. De Toro, D. P. Marques, P. Muñiz, V. Skumryev, J. Sort, D. Givord and J. Nogués, Phys. Rev. Lett., 2015, 115, 057201.
19. B. Muzzi, M. Albino, M. Petrecca, C. Innocenti, C. de J. Fernández, G. Bertoni, C. Marquina, M. R. Ibarra and C. Sangregorio, Small, 2022, 18, 2107426.
20. A. Lopez-Ortega, M. Estrader, G. Salazar-Alvarez, S. Estrade, I. V. Golosovsky, R. K. Dumas, D. J. Keavney, M. Vasilakaki, K. N. Trohidou, J. Sort, F. Peiro, S. Surinach, M. D. Baro and J. Nogues, Nanoscale, 2012, 4, 5138.
21. G. Salazar-Alvarez, J. Sort, A. Uheida, M. Muhammed, S. Surinach, M. D. Baro and J. Nogues, J. Mater. Chem., 2007, 17, 322.
22. X. Liu, B. P. Pichon, C. Ulhaq, C. Lefèvre, J.-M. Grenèche, D. Bégin and S. Bégin-Colin, Chem. Mater., 2015, 27, 4073.
23. K. Sartori, G. Cotin, C. Bouillet, V. Halte, S. Begin-Colin, F. Choueikani and B. P. Pichon, Nanoscale, 2019, 11, 12946.
24. K. Sartori, D. Ihiawakrim, C. Lefèvre, S. Reguer, C. Mocuta, S. Bégin-Colin, F. Choueikani and B. P. Pichon, Mater. Adv., 2022, 8716–8728.
25. G. Salazar-Alvarez, H. Lidbaum, A. López-Ortega, M. Estrader, K. Leifer, J. Sort, S. Suriñach, M. D. Baró and J. Nogués, J. Am. Chem. Soc., 2011, 133, 16738.
26. K. L. Krycka, J. A. Borchers, M. Laver, G. Salazar-Alvarez, A. Lopez-Ortega, M. Estrader, S. Surinach, M. D. Baro, J. Sort and J. Nogués, J. Appl. Phys., 2013, 113(17), 17B531.
27. V. Gavrilov-Isaac, S. Neveu, V. Dupuis, D. Taverna, A. Gloter and V. Cabuil, Small, 2015, 11, 2614.
28. K. Sartori, F. Choueikani, A. Gloter, S. Begin-Colin, D. Taverna and B. P. Pichon, J. Am. Chem. Soc., 2019, 141, 9783.
29. G. Cotin, C. Kiefer, F. Perton, M. Boero, B. Özdamar, A. Bouzid, G. Ori, C. Massobrio, D. Begin, B. P. Pichon, D. Mertz and S. Begin-Colin, ACS Appl. Nano Mater., 2018, 1, 4306.
30. P. Ohresser, E. Otero, F. Choueikani, K. Chen, S. Stanescu, F. Deschamps, T. Moreno, F. Polack, B. Lagarde, J.-P. Daguerre, F. Marteau, F. Scheurer, L. Joly, J.-P. Kappler, B. Muller, O. Bunau and Ph. Sainctavit, Rev. Sci. Instrum., 2014, 85, 013106.
31. N. Daffé, F. Choueikani, S. Neveu, M.-A. Arrio, A. Juhin, P. Ohresser, V. Dupuis and P. Sainctavit, J. Magn. Magn. Mater., 2018, 460, 243.
32. J. Teillet and F. Varret, MOSFIT Software, Université du Maine, Le Mans, France.
33. E. Lima, E. L. Winkler, D. Tobia, H. E. Troiani, R. D. Zysler, E. Agostinelli and D. Fiorani, Chem. Mater., 2012, 24, 512.
34. W. Baaziz, B. P. Pichon, Y. Liu, J.-M. Grenèche, C. Ulhaq-Bouillet, E. Terrier, N. Bergeard, V. Halté, C. Boeglin, F. Choueikani, M. Toumi, T. Mhiri and S. Begin-Colin, Chem. Mater., 2014, 26, 5063.
35. A. López-Ortega, E. Lottini, C. de J. Fernández and C. Sangregorio, Chem. Mater., 2015, 27, 4048.
36. K. Sartori, A. Musat, F. Choueikani, J.-M. Grenèche, S. Hettler, P. Bencok, S. Begin-Colin, P. Steadman, R. Arenal and B. P. Pichon, ACS Appl. Mater. Interfaces, 2021, 13, 16784.
37. P. M. A. de Bakker, E. De Grave, R. E. Vandenberghe and L. H. Bowen, Hyperfine Interact., 1990, 54, 493.
38. T. J. Daou, G. Pourroy, S. Begin-Colin, J.-M. Greneche, C. Ulhaq-Bouillet, P. Legare, P. Bernhardt, C. Leuvrey and G. Rogez, Chem. Mater., 2006, 18, 4399.
39. N. Yaacoub, H. Mortada, Z. Nehme and J.-M. Greneche, J. Nanosci. Nanotechnol., 2019, 19, 5014.
40. F. L. Deepak, M. Bañobre-López, E. Carbó-Argibay, M. F. Cerqueira, Y. Piñeiro-Redondo, J. Rivas, C. M. Thompson, S. Kamali, C. Rodríguez-Abreu, K. Kovnir and Y. V. Kolen’ko, J. Phys. Chem. C, 2015, 119, 11947.
41. M. Liu, M. Lu, L. Wang, S. Xu, J. Zhao and H. Li, J. Mater. Sci., 2016, 51, 5487.
42. G. Franceschin, T. Gaudisson, N. Menguy, B. C. Dodrill, N. Yaacoub, J.-M. Grenèche, R. Valenzuela and S. Ammar, Part. Part. Syst. Charact., 2018, 35, 1800104.
43. E. Pellegrin, M. Hagelstein, S. Doyle, H. O. Moser, J. Fuchs, D. Vollath, S. Schuppler, M. A. James, S. S. Saxena, L. Niesen, O. Rogojanu, G. A. Sawatzky, C. Ferrero, M. Borowski, O. Tjernberg and N. B. Brookes, Phys. Status Solidi B, 1999, 215, 797.
44. J. Li, N. Menguy, M.-A. Arrio, P. Sainctavit, A. Juhin, Y. Wang, H. Chen, O. Bunau, E. Otero, P. Ohresser and Y. Pan, J. R. Soc., Interface, 2016, 13, 20160355.
45. S. Brice-Profeta, M.-A. Arrio, E. Tronc, N. Menguy, I. Letard, C. Cartier dit Moulin, M. Noguès, C. Chanéac, J.-P. Jolivet and Ph. Sainctavit, J. Magn. Magn. Mater., 2005, 288, 354.
46. W. Baaziz, B. P. Pichon, S. Fleutot, Y. Liu, C. Lefevre, J.-M. Greneche, M. Toumi, T. Mhiri and S. Begin-Colin, J. Phys. Chem. C, 2014, 118, 3795.
47. M. W. Haverkort, arXiv, 2005, preprint, arXiv:cond-mat/0505214,  DOI:10.48550/arXiv.cond-mat/0505214.
48. A. G. Roca, I. V. Golosovsky, E. Winkler, A. López-Ortega, M. Estrader, R. D. Zysler, M. D. Baró and J. Nogués, Small, 2018, 14, 1703963.
49. P. Tancredi, P. C. Rivas-Rojas, O. Moscoso-Londoño, D. Muraca, M. Knobel and L. M. Socolovsky, J. Alloys Compd., 2022, 894, 162432.
50. I. J. Bruvera, P. Mendoza Zélis, M. Pilar Calatayud, G. F. Goya and F. H. Sánchez, J. Appl. Phys., 2015, 118, 184304.
51. M. Pauly, B. P. Pichon, P.-A. Albouy, S. Fleutot, C. Leuvrey, M. Trassin, J.-L. Gallani and S. Begin-Colin, J. Mater. Chem., 2011, 21, 16018.
52. J. A. De Toro, P. S. Normile, S. S. Lee, D. Salazar, J. L. Cheong, P. Muñiz, J. M. Riveiro, M. Hillenkamp, F. Tournus, A. Tamion and P. Nordblad, J. Phys. Chem. C, 2013, 117, 10213.
53. E. H. Sánchez, M. Vasilakaki, S. S. Lee, P. S. Normile, M. S. Andersson, R. Mathieu, A. López-Ortega, B. P. Pichon, D. Peddis, C. Binns, P. Nordblad, K. Trohidou, J. Nogués and J. A. De Toro, Small, 2022, 18, 2106762.
54. J. Nogues, V. Skumryev, J. Sort, S. Stoyanov and D. Givord, Phys. Rev. Lett., 2006, 97, 157203.
55. P. S. Normile, J. A. De Toro, T. Muñoz, J. A. González, J. P. Andrés, P. Muñiz, R. E. Galindo and J. M. Riveiro, Phys. Rev. B, 2007, 76, 104430.
56. E. H. Sánchez, M. Vasilakaki, S. S. Lee, P. S. Normile, G. Muscas, M. Murgia, M. S. Andersson, G. Singh, R. Mathieu, P. Nordblad, P. C. Ricci, D. Peddis, K. N. Trohidou, J. Nogués and J. A. De Toro, Chem. Mater., 2020, 32, 969.
57. R. L. Antón, J. A. González, J. P. Andrés, P. S. Normile, J. Canales-Vázquez, P. Muñiz, J. M. Riveiro and J. A. De Toro, Nanomaterials, 2017, 7, 61.
58. E. Skoropata, R. D. Desautels, C.-C. Chi, H. Ouyang, J. W. Freeland and J. van Lierop, Phys. Rev. B: Condens. Matter Mater. Phys., 2014, 89, 024410.

---

Footnote

† Electronic supplementary information (ESI) available. See DOI: https://doi.org/10.1039/d3na01108a

---