Reversible spin storage in metal oxide—fullerene heterojunctions

Hybrid interfaces form a spin capacitor for the generation and storage of spin angular momentum via optical or electronic stimuli.


This PDF file includes:
Section S1. Sample structure Section S2. Junction characterization Section S3. DFT simulations Section S4. X-ray absorption spectroscopy Section S5. Sample preparation and LE-μSR fitting details Fig. S1. Transport characteristics of junctions with and without an alumina barrier.   Table S1. Computed energy differences (eV) between the different adsorption geometries of C 60 on β-MnO 2 (110)-2x2 as a function of the extra electronic charge added to the system. Table S2. Computed Bader charges (Q) (63) for the different adsorption geometries of C 60 on β-MnO 2 (110)-2x2 as a function of the extra electronic charge added to the system. References (47-67) Section S1. Sample structure Two characteristic structures for the NEXAFS/Photovoltaic junctions used in this research are: Co(10-30 nm)/ Al 2 O 3 (1.4 -1.6 nm)/C 60 (20 nm)/MnOx(2.5-5 nm)/Al(1.2 nm) and Co(10-30 nm)/C 60 (20 nm)/ MnOx(2.5-5 nm)/Al(1.2 nm) -with Cu instead of Co in some samples. The sputtering pressure is of the order of mTorr, but sputtering and evaporation are not done simultaneously. The C 60 is evaporated at ~450 °C with a chamber pressure of ~10 -8 mTorr and an oxygen pressure of ~10 -10 mTorr. The use of a tunnel barrier determines whether there is spin conserved injection from the ferromagnetic electrode or whether spin scattering at the interface leads to a non-polarized current as well understood in the conductivity mismatch problem. (47) Whether or not there is a direct contact between metal and molecule also affects the doping of the C 60 film. Where there is a direct metal contact, the C 60 behaves as though it were significantly n-doped, making the IV characteristic more asymmetric.(21) With a tunnel barrier, the IV characteristic is more symmetric. In both cases, charge trapping is observed to occur at the MnOx/C 60 interface. Assuming the modeled MnO 2 /C 60 interface is representative of the system in the real device, and based on the DFT simulations, only the immediate metal oxide/C 60 interface has metallic states and is therefore the only conductive layer together with the Co-film.
Deeper MnO 2 atomic-layers (3 rd onwards), further than 0.5 nm (5 Å) from the interface, display a non-zero (>0.2 eV) band-gap. Based on the present DFT simulations, the role of the MnO 2 layer is to provide a source of O-atoms, whose diffusion, coupled with MnO 2-x /C 60 re-hybridization and addition of extra charge in the floating state altogether lead to: 1) appearance of half-metallic states at the MnO 2 /C 60 interface (likely spin polarization but not half-metallicity in the real, disordered MnO x system, leading to long but finite discharge times), and 2) spin-polarization of the C 60 closest to the MnO 2-x substrate.
The cobalt film is polycrystalline, formed by grains of several 10s of nm. At remanence, these grains generate a stray field (even though an ideal, uniformly magnetized cobalt film would not).
This stray field is a very common problem e.g. in our measurements of spin triplets with superconductors and muon spin spectroscopy. In the electrically-biased muon sample, we used the thickest C 60 layer of any experiment, and the field at the interface, as measured by the muon precession, was 5 Gauss. The distance between Co and interface for this sample was 55 nm, compared to 15-20 nm in NEXAFS and photovoltaics, so the field in the latter samples can be estimated to be 100-200 Gauss assuming a dipolar field (∝ r −3 ). This field is unlikely to be the only mechanism mediating the spin stabilization, which will depend as well on the spin dependent interfacial dipole and MnO 2 density of states, but it explains the connection between the Co magnetization and the interface transport properties (e.g. in the discharge time, the photocurrents measured with different Co configurations, the dependence of the LUMO* position on field etc.). By comparison, the MnOx film is only (weakly) magnetic at the atomic surface, with no magnetocrystallyne anisotropy (nominally no demagnetizing or stray field for the ideal 2D case) and therefore magnetically soft.

Section S2. Junction characterization
The interface has a resistivity orders of magnitude higher than the bulk MnOx layer, changing from 10 to 0.1 M for a typical 100×100 m 2 junction when the interface is broken by displacing oxygen with an electric field. For spectroscopy, XAS and XMCD, alumina barriers are used to prevent Co/C 60 hybridization features appearing in the K-edge and to allow the injection of spin polarized electrons. For V OC measurements, this tunnel barrier is not included. This is because the photovoltaic efficiency is much higher without a barrier, allowing us to access a V OC similar to the bias applied in XAS, and because the junction structure is invertedfigure S1. Inverting the structure allows us to excite optically the MnOx/C 60 without a protective cap, thus increasing the illumination and eliminate artefacts from the cap used. The junction is charged by exposure to light and the generation of a photocurrent; we then measure the drop of the open circuit voltage with time by using a nano-voltmeter connected to a data acquisition card with sub-ms resolution. While oxygen diffusion is commonly observed at the interface between complex oxides and C 60 , the process used to fabricate these samples shows minimal interdiffusion of oxygen in cross-sectional TEM.(48) This is vital since C 60 -oxide develops a mid-gap acceptor band which would quench the interfacial effects and compensate donor transport, significantly reducing conductivity. (49) Transport through the junctions is not uniform under low bias. A junction with an alumina tunnel barrier was probed via photo-luminescence spectroscopy during transport. Hot electron injection over the tunnel barrier creates secondary excitations and boosts the luminescent signal, highlighting areas where current density is highest.(50) This shows the importance of using optical excitation to measure the spin dependent leakage current, since charging of the interface during transport will occur preferentially at the junction edges where the stray fields are likely to behave differently to the junction center. Figures  and MnO x layers, including Xray reflectivity and diffraction (XRR/XRD), Raman spectroscopy and luminescence. Figure 2F shows the charge-trapping effects in the luminescence of C 60 . (1.69 eV) and phonon-assisted recombination (1.52 eV) in C 60 are reduced after an electric field is applied. The initial "0" state before a voltage is measured with the electrodes floating. Hg (6) Hg (5) Hg (4) Hg (3) Silicon Peak

A C
Hg (2) Hg (1) Ag (1) Hg (7) Counts (c/s) Raman shift (cm -1 ) Ag (2) Hg (8) (Table S1). This result appears to be qualitatively unaffected by the presence of additional electronic charge at the C 60 /β-MnO 2 (110)-2x2 interface model (Table S1). In all cases, the non-perfectly matched periodicity of the β-MnO 2 (110)-2x2 slab and the C 60 layers resulted in asymmetric relaxation of the 2 nd C 60 layer [C 60 (2 nd )] over the C 60 (1 st ).  Table S2. Owing to this undesirable aspect, the effects of extra charge in the C 60 /β-MnO 2 (110)-2x2 interface are analyzed only for addition of 0.6 e -(0.3 e -/C 60 ), which result in negligible erroneous charge-spillage on the 2 nd C 60 layer.    For the specialist reader we finally note that, as shown in Fig. S5, re-hybridization between the C 60 and β -MnO 2 (110), in either the chem or o-chem minimum, turns out to be critical in removing metallic states for the minority spin-channel. When re-hybridization is not present as

A B
for the vdW minimum, the β-MnO 2 (110)-2x2 slab recovers a non-zero density of states for the spin-minority channel as previously computed for bare the β-MnO 2 (110)   Keithley 428 current amplifier with the drain contact made using silver coated carbon tape to a point close to the junction. Transport contacts are made using silver paint and kapton insulated Cu wire attached to sprung contacts built into the TEY head. The TEY drain contact is made using a thin strip of conducting carbon tape which extends to 1mm away from the junction and is then coated with silver. Contact is made between this tape and the Cu backing plate which is connected to ground via a pico-ammeter. The total resistance of this channel was measured to be 400 Ohm for the samples shown in Fig 1. The drain contact resistance was on the order of 100 Ω, and this similarity in resistance caused mixing between the drain current and junction current, increasing the signal to noise ratio. When there was no active bias, the internal resistance of the source was >1 GΩ.
Following monochromation, the beam is focused by a final mirror at which a photoelectron current is recorded (i1). The beam then passes through a gold grid at which the normalization signal is recorded (i2). Both the mirror and normalization grid have a small amount of carbon contamination resulting in absorption at 285 and 290 eV. The TEY signal is then normalized to the gold grid, figures S7A-C. To avoid distortion of the signal by carbon contamination on the normalization grid, fresh gold is evaporated onto the grid at the beginning of the experiment.
Carbon contamination is a common issue so it is important to establish whether any component of the C K-edge arises from aliphatic or amorphous carbon layers rather than the C 60 film. By probing carbon K-edge on the substrate away from the C 60 , it was verified that the residual signal from carbon contaminants was much smaller than the C 60 signal, Fig. S7B. The main XMCD peak at 282 eV features above any artefacts due to normalization. Owing to metallization and ensuing broadening of empty states, we do not expect a contribution to the (TEY) NEXAFs signal from C 60 (1 st ). By comparison, the C 60 (2 nd ) layer in the simulations has a well-defined band-gap and sharp LUMO(*) energies. In addition, deeper layers of C 60 , beyond the 3 rd layer (i.e. ≳2 nm from the interface), will have an exponentially lower contribution toward the TEY signal, as the Auger electrons produced in these layers will not have enough energy to overcome the work function and contribute to the signal. Sum rules typically used in transition metals do not apply to the carbon K-edge, partly due to the lack of SOC and the orbital hybrid structure of carbon, so a priory it would be impossible to distinguish between spin or orbital momentum in C 60 via XMCD. Therefore, we must indirectly infer the origin of the signal from other methods.
Given that i. the effect emerges only when we use a magnetic electrode and a, theoretically, halfmetallic interface; ii. that the muon depolarization responds to the local spin ordering; and iii. the spherical symmetry of the C 60 cage (which rotates freely in the ps scale at room temperature in thin films, which will average out any orbital asymmetry in the C 60 ), it is almost certain that the spin momentum is responsible for the results observed.  The C 60 /MnO x interface acts as a pn junction with a weak photovoltaic response. Figure S9 shows the typical photovoltaic effect for a junction without an alumina barrier between the cobalt electron and C 60 layer. The presence of an alumina barrier does not change this dependence, but greatly reduces the magnitude of the photocurrent. The photocurrent decreases as the temperature is lowered due to increased internal resistance and trap dwell time, with the carrier density and mobility reduced -see Figure S10. where Δ = 25 ± 1 meV is the trap depth and J 0 = 390 ± 20 μA is a factor that takes into account the charge carrier density, mobility and electric field.
Once the device is charged using the photovoltaic effect, stopping the light irradiation leads to a capacitor-like discharge where the hopping time, and therefore the discharge time, are dependent on the magnetic field. Magnetic configurations of the cobalt electrode with high disorder or a complex domain structure, e.g. measurements at the in-plane coercive field, see Figs. 3D-E of the main manuscript, or at remanence after an out-of-plane field ( Fig. 4B and Fig. S11) have a decay 230-40% faster than when the electrode magnetization is uniform (in-plane saturated).

Fig. S11. Discharge for a photovoltaic-charged device.
Changing the out-of-plane magnetic field leads to different discharge times when the device is measured at remanence; the larger fields, always below the out of plane saturation, lead to higher disorder.

Section S5. Sample preparation and LE-μSR fitting details
Low energy muon spin rotation (LE-μSR) measurements were taken at the LEM beamline, at the Paul Scherrer institute in Switzerland. Unlike bulk µSR, low energies of implantation may be obtained by use of a moderation technique. (67) The technique uses a few-hundred-nanometer thick solid Ar moderator capped with ~10nm N 2 , grown on top of a 100µm thick silver foil.
Implantation energies varying between 0.5 keV to 30 keV, corresponding to depths of up to ~200 nm, can be sourced by choosing appropriate transport and sample voltage settings -with the large voltages being applied to the muon beam away from the sample to avoid damage. The energies

c.
Data was collected in a zero field (ZF) and transverse field (TF) geometry (field in the plane of the sample and transverse to the spin of the incoming muons). In the TF geometry the initial polarization of the muon spin is perpendicular to the applied field. The time dependence of the muon polarization is analyzed as an asymmetry function A s (t) and data fitted using musrfit software. In zero field and at 250 K, the fast rotation of C 60 molecules eliminates the observability of the anisotropic Mu-C 60 radical state, therefore we only observe a low frequency (less than 1.5 MHz) oscillation attributable to transitions between different hyperfine energies of an axially symmetric, anisotropic endohedral muonium state ( + @C 60 ). Past measurements of weak magnetic states in C 60 based interfaces show the coexistence of the endohedral muonium state with another that corresponds to the magnetic interfacial C 60 , where the muonium states include particles in neutral C 60 and in charged triplets, labelled as C 60 [-] in the manuscript.(9) In our system, the stray field of the cobalt electrode strongly contributes to the depolarization of the muonium states.
LE-μSR requires large area sample due to long count times. Therefore, wide area molecular junctions, with roughly 200 mm 2 active areas, were fabricated with aid of optical lithography techniques. Firstly, Ta/Au contacts were deposited onto Corning Eagle XG substrates. The ferromagnetic electrode and tunnel barrier, Co(20nm)/Al 2 O 3 (2nm), was then deposited through a shadow mask. Here, the Al 2 O 3 not only acts to protect the FM against Oxidation between fabrication steps, but will also become part of the tunnel barrier, ensuring spin conserved tunneling of charges from the Co into the molecular layer. To avoid short circuits, or current crowding due to variations in the electric field around the edge of the cobalt electrode, the active junction area was defined with an optical lithography step. A 200nm layer of PMMA A4950K was spun onto the whole device. A 200 mm 2 window was then exposed to UV and ozone. This was finally developed in acetone and exposed to an Ar plasma to etch any residual PMMA and to leave a clean Al 2 O 3 surface. The remainder of the device, starting with a second thin (1.4nm) Al 2 O 3 layer, was deposited by sublimation of C 60 and DC sputtering to yield wide area molecular junctions. A schematic of such a device is shown in fig. S12A, with the corresponding fraction of the muons stopping throughout the device. After using this fabrication processing, the DC I-V data shows a nonlinear dependence of the resistance with voltage. This is to be expected for transport dominated by tunneling. The resistance of the devices also increases at lower temperature, suggesting that after charge carriers tunnel into the C 60 LUMO, undergo the expected variable-range hopping transport, see Fig. S12B for a typical I-V characteristic in these large-scale devices.
We first model the time dependence of the muonium polarization with a relatively simple model.
A highly damped oscillation of the  + @C 60 state is chosen with a depolarization function which incorporates slowly relaxing mechanisms -including in a single function the oscillation for both neutral and charged C 60 . The resulting ZF-μSR spectra are fitted with the function A s (t) = Acos(φ + 2πv μ+@C60 t)e −λ μ+@C60 * t + A fast e −λ fast * t In this equation the frequency v μ+@C60 corresponds to hyperfine oscillations of any anisotropic endohedral muonium. For this model, it incorporates the typical frequency expected for muonium ( + -e-) and that of the muonium forming with interfacial/charged C 60 triplets. The parameter A is the muon decay asymmetry parameter and is proportional to the volume occupied by the corresponding μ+ state. The phase angle, is the angle of the muon spin at t=0 with respect to the positron detector. λ is a decay constant determined by the polarization loss mechanisms. In addition to the oscillation, a relaxing component (A fast with depolarization rate λ fast ) is observed. Even though the depolarization of the muonium state is affected by the cobalt stray field, measuring at each implantation energy in the separate voltage states allows us to fix to the average  + @C 60 asymmetry A, phase angle φ and slow depolarization rates λ μ+@C60 . This allows us to fit the average frequency of the muonium states and the corresponding depolarization rate (v μ+@C60 and λ μ+@C60 , respectively). The result of using this single oscillation model can be seen in figure S12C. This fit averages all the frequencies in the charged and neutral molecules. It demonstrates that, although the uncertainty is quite large, even averaging for all molecular states the only statistically significant change occurs at the stopping energies (10 and 12 keV) for the MnO X /C 60 , where we observe a 13±8% and 26±9% increase in frequency respectively - Fig. S12. As previously discussed, we expect to observe the onset of a second muonium frequency for muonium in spin polarized C 60 . We therefore used a two frequency model for the ZF data as b. c.

A B C
A s (t) = A fast e −λ fast * t + ∑ A i cos(φ + 2πv i t)e −λ i * t 2 i Here A i accounts for the typical anisotropic  + @C 60 state with a frequency of 0.2-0.4, and also for a second, higher frequency oscillation that we attribute to spin polarized charged C 60 ( + @C 60 [-]). The results obtained by this model are shown in the main manuscript and In a transverse magnetic field (perpendicular to the muon polarization, but parallel to the plane of the sample), we need only to model precession of the μ + in some static magnetic field, B ̅ local .
The TF-μSR data is therefore fitted with the function A s (t) = Acos(γ μ B ̅ local t + φ)e −λt With = 2 * 135.5 MHz T -1 being the muon gyromagnetic ratio. Once again accounts for depolarization mechanisms of the μ + , particularly due to inhomogeneous magnetic field distributions -see Figures S13C-D below for the electrically biased sample and Figure 4F in the manuscript for the photovoltaic sample. As it was the case at 250 K, there is an increase in the susceptibility under light irradiation and in the floating state. However, the change under light irradiation is smaller (and within uncertainty of the initial ground) than at 250 K. This could be due to the approx. 10 times smaller photocurrent flow at 50 K resulting in longer charging time.
C 60 compounds may display magnetic order -e.g. EDTA-C 60 is ferromagnetic at low temperatures, and C 60 -O may show paramagnetic or superparamagnetic behavior. Figures S15 below shows that the combination of both materials (C 60 and MnO 2 ) appears to enhance the magnetic signal, in particular when the interface has been charged by exposure to light -the halfmetallic properties of the MnO 2 surface extend to the C 60 , which is further charged by oxygen hybridization at the interface. The moments observed in Fig. S15a are very weak and therefore highly sensitive to contaminants etc. Fig. S15b shows how the moments may become larger once the full device is grown and the interface charged via light exposure, the magnetic properties are further enhanced -although now there is a large background due to the cobalt electrode and its interaction with the C 60 layer (13). Decoupling the C 60 and MnO x (sample C) or reversing the structure (samples D and E) leads to lower magnetization.