In this work, we investigate the use of post-fabrication thermal nanoconversion (TNC), using a heated scanning probe tip, to modify the magnetic properties of Ni80Fe20 elliptical nanoislands with varying aspect ratio. Despite Ni80Fe20 being unoptimized for TNC, by comparing quasistatic and dynamic magneto-optical Kerr effect microscopy measurements, we demonstrate that TNC at a contact temperature of 250 °C increases the saturation magnetization of the treated nanoislands, reaching a value close to 800 kA/m. Micromagnetic simulations of the nanoislands indicate that the TNC technique can be used to alter the remanent state, from a single domain to a vortex. These results demonstrate the opportunities afforded by TNC to modify the properties of selected areas in a thin film or a patterned sample, particularly when designing magnonic crystals and other nanomagnetic devices.
Standard methods for preparing masks for nanoscale pattern transfer, such as electron beam lithography, routinely allow the fabrication of designs for a wide range of applications.1–5 The process is, however, typically binary in nature—a pixel in the design represents either the presence or absence of material in the layer—ultimately limiting the phase space that may be explored when developing thin film devices. The ability to change the properties of a thin film, as opposed to the physical addition or removal of material, is more challenging, particularly on a nanoscopic length scale, i.e., nanoconversion. Such nanoconversion, and the continuous modification of the underlying film properties, is actively sought in a variety of systems and has been demonstrated, for example, through grayscale lithography,6 chemical conversion of functional materials,7 ion implantation,8 redox of materials,9,10 and local binding of biological materials.11
Thermal nanoconversion (TNC) is a particularly appealing approach as heat is a common route to alter the properties of a material. For example, heat is used to alter thin film magnetic properties through defect annealing, sintering, recrystallisation, demagnetization, and exchange bias setting.12–16 An emerging approach to TNC is via the rastering of a heated nanoscale tip across the surface of a thin film, modifying the film in the immediate vicinity of the contact point. Such an approach to TNC has been used to great effect in magnonics and nanomagnetism to generate curved domain walls for waveguides,17 pattern spin textures,18 and the fabrication of vortex cores in crystalline films,19 but its wider application remains underutilized.
In this work, we investigate the use of post-fabrication TNC to modify the magnetic properties of Ni80Fe20 elliptical nanoislands with varying aspect ratio (length:width). Despite Ni80Fe20 being an unoptimized material for TNC, we demonstrate that heating to contact temperature 250 °C for 1 ms leads to a clear change in magnetic properties of the treated islands, consistent with ∼15% increase in saturation magnetization, . Using time-resolved Kerr microscopy, we evidence a change in the Kittel-like uniform-mode resonance field for TNC treated islands, suggesting 8% increase in under similar TNC conditions. These results demonstrate the opportunities afforded by TNC to modify the properties of selected areas when designing magnonic crystals and other nanomagnetic devices.
To make the TNC structures, ferromagnetic nanoisland arrays, spanning 20 × 20 μm2, were fabricated on both oxide and undoped silicon substrates using thermal scanning probe lithography (tSPL). Each array consisted of individual nanoislands, 200 nm wide with varying length , either 200, 300, 400, or 800 nm, and an edge-to-edge separation of 400 nm. A schematic of the t-SPL technique is shown in Fig. 1(a). A thermally sensitive bilayer resist consisting of polymethylglutarimide (PMGI; 50 nm)/polyphthalaldehyde (PPA; 30 nm) is spin-coated onto the silicon oxide substrate. A heated tip is then scanned across the resist surface and electrostatically pulled in and out of contact with the sample. The PPA layer sublimates when heated above 150 °C, transferring the pattern mask when the tip is in contact with the surface, in this case, with a write depth of ∼35 nm.20 The PMGI layer is chemically etched in a tetramethylammonium hydroxide (TMAH):isopropyl alcohol solution (1:40 by volume, respectively), followed by oxygen plasma etch de-scum, exposing the substrate in the patterned regions. Ni80Fe20 is deposited through the mask via electron beam evaporation under ultra-high vacuum, with a thickness of 20 nm, and at a rate of 0.4 Å s−1, source purity of 99.9%, and deposition pressure ≤1 × 10−8 mbar. Finally, a liftoff step using either N-methyl-2-pyrrolidone or dimethyl sulfoxide (DMSO) is performed to remove the remaining resist stack, leaving the nanoisland arrays seen in Figs. 1(c) to 1(f).
(a) Removal of material by thermomechanical sublimation during thermal scanning probe lithography. (b) Thermal nanoconversion of material by local heat modification. (c)–(f) SEM images of produced Ni80Fe20 nanoislands, where (c) = 200, (d) = 300, (e) = 400, and (f) = 800 nm. The inset, below (e), are the t-SPL masks for each nanoisland length, scale bar for (c)–(f), and the coordinate system.
(a) Removal of material by thermomechanical sublimation during thermal scanning probe lithography. (b) Thermal nanoconversion of material by local heat modification. (c)–(f) SEM images of produced Ni80Fe20 nanoislands, where (c) = 200, (d) = 300, (e) = 400, and (f) = 800 nm. The inset, below (e), are the t-SPL masks for each nanoisland length, scale bar for (c)–(f), and the coordinate system.
We perform spatially resolved, longitudinal geometry, magneto-optical Kerr effect (MOKE) microscopy to characterize the magnetic reversal of the TNC treated and untreated arrays. Figures 2(a)–2(d) show hysteresis loops of the -component of the magnetization, , normalized to the saturation magnetization, , vs applied magnetic field, . The field ( ) is swept at a rate of 20 mT/s. Data are the averaged response across all islands within an array, with l ranging from 200 nm to 800 nm as shown in Figs. 2(a)–2(d). Displayed are data for the unheated (black data) and TNC heated (red data) arrays. Considering the untreated arrays first, the square hysteresis loops observed for higher aspect ratio islands [ 400 and 800 nm, Figs. 2(c) and 2(d)] indicate sufficient shape anisotropy to ensure a single domain remanent state and sharp, domain wall-mediated reversal. The 200 and 300 nm, conversely, show the characteristic “pinched” hysteresis loops of the magnetic vortex state.25 Indeed, the 200 nm, circular islands show the low remanent magnetization, , of a flux closure or a curling state. Moving to consider the TNC treated arrays, the hysteresis loops show a similar overall trend, from vortex to single domain as increases; however, the heated arrays remain in a flux closure or a curling state up to higher aspect ratio. Indeed, sharp, single domain switching only emerges at 800 nm and a clear difference in remanence, loop area, and saturation field are all visible in Figs. 2(b) and 2(c).
Normalized magnetization component, , vs applied field, , obtained from magneto-optical Kerr effect microscopy of 20 nm thick Ni80Fe20 nanoislands. Upper plots: solid lines represent longitudinal MOKE data, with applied parallel to . Lower plots: dotted line shows MuMAX3 simulated response, with varying . Images: simulated equilibrium state at remanence, where color and arrows represent in-plane magnetization direction [see vortex (i) for to color variation]. (a), (e), and (i) 200; (b), (f), and (j) 300; (c), (g), and (k) 400; and (d), (h), and (l) 800 nm.
Normalized magnetization component, , vs applied field, , obtained from magneto-optical Kerr effect microscopy of 20 nm thick Ni80Fe20 nanoislands. Upper plots: solid lines represent longitudinal MOKE data, with applied parallel to . Lower plots: dotted line shows MuMAX3 simulated response, with varying . Images: simulated equilibrium state at remanence, where color and arrows represent in-plane magnetization direction [see vortex (i) for to color variation]. (a), (e), and (i) 200; (b), (f), and (j) 300; (c), (g), and (k) 400; and (d), (h), and (l) 800 nm.
The evolution from the vortex state to the single domain on increasing the nanoisland aspect ratio is a well-known micromagnetic effect observed in the literature and is linked directly to the interplay between anisotropy and demagnetizing energy of the island, the latter of which is principally influenced by and island shape.26 Polycrystalline Ni80Fe20 has negligible crystalline anisotropy,27 and scanning electron microscopy images of pre- and post-TNC arrays confirm that the shape of the islands is unaffected by the TNC process. We, therefore, use micromagnetic simulations28,29 to understand the impact of varying on the hysteresis loops in Figs. 2(a)–2(d). For these simulations, we take the literature established values of Gilbert damping constant, 0.01, and exchange constant, 13 × 10−12 m for Ni80Fe20.28
Hysteresis loops ( generated from MuMax3 simulations of each nanoisland geometry are shown in Figs. 2(e)–2(h), for varying from 600 kA/m (black dashed line) to that of bulk Ni80Fe20, 880 kA/m (yellow dashed line). The simulations show good qualitative agreement with experimental data, with the expected evolution from a vortex state to single domain remanence on increasing the nanoisland aspect ratio. It is equally clear that increasing causes the vortex state to be maintained up to larger aspect ratios, as might be expected given the resulting increase in demagnetizing energy naturally favors flux closure. Figures 2(i)–2(l) show the obtained micromagnetic equilibrium state at remanence for each value for 800 kA/m. In both experiment and modeling, the reversal becomes dominated by an easy-axis response when 800 nm for all tested cases.
At this point, we reason that the principal effect of the TNC has been to locally increase in treated islands. To consider this further and to estimate the pre- and post-treatment values, in Fig. 3, we compare loop area, normalized to for direct comparison [Fig. 3(a)]; change in coercive field, normalized to the maximum value at 800 nm [Fig. 3(b)]; normalized remanence [ , Fig. 3(c)]; and saturation field [ , Fig. 3(d)], for experiment and numerical simulation. Presented numerical values are averages from five repeated simulations where small variation arises from thermal fluctuations, with error bars representing the standard error derived from these averages. From Figs. 3(a)–3(c), we see an increase in increases the hysteresis loop area as both coercivity and remanence increase, due to the expected increase in the anisotropy field. The square loop shape at large equally shows a decrease in [Fig. 3(d)], since a vortex core does not form. While there is only a minor distinction between the untreated and TNC treated arrays in , , or loop area, the remanent magnetization, , shows a clear trend, particularly a difference in the critical value above which the islands enter the single domain, high remanence regime. Comparing the change in Figs. 3(a)–3(d), including between heated and unheated arrays, with the evolution of across the simulated arrays, we estimate increases from 700 to >800 kA/m through TNC, representing a minimum change of 14%.
Analysis of experimental and simulated data for varying island size, solid lines representing experimental data, and dotted lines showing modeled data for different . (a) Area contained within the hysteresis loop, normalized by . (b) Relative change in coercivity, . (c) Remanent magnetization, (d) Saturation field, .
Analysis of experimental and simulated data for varying island size, solid lines representing experimental data, and dotted lines showing modeled data for different . (a) Area contained within the hysteresis loop, normalized by . (b) Relative change in coercivity, . (c) Remanent magnetization, (d) Saturation field, .
To further confirm TNC-induced changes in the magnetic properties of the treated islands, we performed spatio-temporally resolved MOKE (TR-MOKE) microscopy to acquire ferromagnetic resonance spectra from individual nanoislands. High spatial resolution was achieved using a ×100 oil immersion lens (N.A. 1.3) to focus probe laser pulses to a diffraction limited spot. The probe laser wavelength was 520 nm, derived from second harmonic generation of a 1040 nm mode-locked fiber laser with a nominal pulse duration of 140 fs and a repetition rate of 80 MHz. The probe average power was attenuated to <200 μW. Global excitation of magnetization dynamics in the arrays of nanoislands was achieved by placing the sample face-up onto a 50 Ω impedance matched coplanar waveguide (CPW) with center conductor width and ground plane separation of 0.5 and 0.36 mm, respectively. Nanoisland array samples were fabricated on 300 μm intrinsic Si substrates to avoid attenuation of the excitation field. A microwave current waveform from an 80 MHz comb generator, synchronized to the laser, was used to provide single frequency, phase stable excitation by narrow bandpass filtering the desired excitation frequency from the output 80 MHz comb (160 MHz to 18 GHz). The output was amplified before passing through the CPW and the transmitted waveform monitored using a sampling oscilloscope. Amplitude modulation of the microwave waveform at ∼31.4 kHz allowed for phase sensitive detection of the Kerr signal using a lock-in amplifier.
Time-resolved scans with ∼10 ps resolution were acquired at a fixed position on an individual nanoisland for a given in-plane bias magnetic field, applied along the length of the CPW. The CPW microwave field was in-plane and perpendicular to the equilibrium magnetization, while magnetization dynamics were probed using polar MOKE to sense the out-of-plane component of precession. At two time-delays, corresponding to approximately a node and antinode, the bias magnetic field was swept to acquire ferromagnetic resonance (FMR) spectra, with example scans shown in Fig. 4(b). These correspond to the real and imaginary parts of the dynamic susceptibility tensor components of the dynamic magnetization. At the resonance field and desired time delay, TR images were acquired by scanning the nanoisland arrays beneath the focused laser spot.
(a) Time resolved MOKE image taken of an 300 nm uniformly annealed array, taken at a field strength of 91 mT and 70 ps time delay. The inset shows a high resolution reflected intensity image of a single nanodisk (white) against the substrate (black), (b) Kerr effect rotation vs bias field under 8 GHz excitation field of a single l = 400 nm nanodisk with signal recorded at delays of 0–1 ps (blue) and 30–31 ps (red) with fitted response. (c) Extracted resonance field, , vs anneal contact temperature, . (d) Relative change in under TNC treatment vs contact temperature.
(a) Time resolved MOKE image taken of an 300 nm uniformly annealed array, taken at a field strength of 91 mT and 70 ps time delay. The inset shows a high resolution reflected intensity image of a single nanodisk (white) against the substrate (black), (b) Kerr effect rotation vs bias field under 8 GHz excitation field of a single l = 400 nm nanodisk with signal recorded at delays of 0–1 ps (blue) and 30–31 ps (red) with fitted response. (c) Extracted resonance field, , vs anneal contact temperature, . (d) Relative change in under TNC treatment vs contact temperature.
Both quasistatic and dynamic measurements confirm an increase in for the islands selectively treated with TNC, with respect to of the control, as-deposited nanoislands. Our estimates reach ∼ 800 kA/m post-TNC, i.e., close to the room temperature value for bulk Ni80Fe20.31 Literature examples of bulk annealing of thin Ni80Fe20 films in an inert atmosphere indicate a complex interplay between kinetics. For moderate temperatures ( 250 °C) and short anneal times (<30 min), the process of annealing tends to increase , due to removal of defects and modest improvement in the crystallinity of the films,32 reaching a peak of 800 kA/m at 250 °C. Atomic force microscopy measurements of the grain size distribution within the untreated and treated nanoislands reflects such a grain size increase, from 44.0 ± 0.6 to 49.5 ± 1 nm (12% increase), which is consistent with the measured increase in . Naturally, the reasonable as-grown value of 700 kA/m in our samples limits the percentage increase achievable via TNC; however, the final achieved is in excellent agreement with that found in bulk studies.32
Long anneal times and elevated temperatures are known to cause an increase in damping and loss of , due to migration of Ni into the Si substrate; a natural consequence of the high diffusion constant, 1.7 10−4 cm2/s and low activation barrier, 0.15 eV, for interstitial migration within the material pairing.33 Localized heating on short timescales is an advantage of TNC as it allows one to access different diffusion kinetics during the annealing process. Heating is highly localized, with the (typical) high thermal conductivity and large thermal mass of the substrate preventing substantial increase in from the localized probe. During the 1 ms anneal time, the diffusion length of Ni in Si at a temperature of 250 °C is estimated to be ∼110 nm,33 more than sufficient to substantively reduce the Ni content in the ferromagnet. We, thus, reason that during TNC the Si substrate remains close to room temperature, circumventing Ni diffusion and preventing the decrease in observed in other global annealing studies.31
To confirm this, we compare TNC treated nanoislands to samples heated in a high vacuum rapid annealer for 1 h. To test the impact of Ni diffusion, we also consider the rapid annealing of 20 nm Ni80Fe20 nanoislands grown on a 10 nm MgO barrier layer to prevent Ni diffusion (MgO/Ni80Fe20). Figure 5 quantifies the change in magnetic reversal on rapid annealing to 350 °C (data are shown for the as-grown array, black data; rapid annealed data, green; and annealed diffusion barrier samples, blue), where a weak increase in coercivity, and loop array is found on annealing the MgO/Ni80Fe20 nanoislands. This qualitatively mirrors increased and loop area following TNC treatment (Fig. 3). Comparing with samples where no diffusion barrier is present (green data), we see a stark difference: There, annealing (up to 300 °C) shows a considerable decrease in coercivity (up to 50%) and loop area, again indicating Ni diffusion plays a substantive role in the magnetic response of annealed Ni80Fe20, grown on Si substrates, and that this is mitigated during the short soak times of TNC.
Nanoisland magnetic reversal parameters vs length, , for as-grown Ni80Fe20 islands (black squares), annealed Ni80Fe20 islands ( 300 °C, green squares), and annealed islands with Ni diffusion barrier MgO/Ni80Fe20 ( 350 °C, blue triangles), obtained from longitudinal MOKE measurements. (a) Area contained within the hysteresis loop, normalized by . (b) Coercivity, , (c) normalized remanent magnetization, and (d) saturation field, .
Nanoisland magnetic reversal parameters vs length, , for as-grown Ni80Fe20 islands (black squares), annealed Ni80Fe20 islands ( 300 °C, green squares), and annealed islands with Ni diffusion barrier MgO/Ni80Fe20 ( 350 °C, blue triangles), obtained from longitudinal MOKE measurements. (a) Area contained within the hysteresis loop, normalized by . (b) Coercivity, , (c) normalized remanent magnetization, and (d) saturation field, .
Finally, we turn our attention to potential surface oxidation during TNC. X-ray photoelectron spectroscopy of 300 °C annealed and unannealed Ni80Fe20 thin films confirms film oxidation in our samples proceeds by preferential Fe oxidation, with a loss of Ni and NiO observed in annealed samples, and so is limited by Fe migration to the top surface. Considering the (low) parabolic rate constant found in the literature for this process, we estimate a sub-nanometer increase in the oxide layer for the TNC conditions applied here,34,35 again highlighting the noted robustness of Ni80Fe20 passivation and the advantages of the short anneal times achievable during TNC.
Ni80Fe20 is a common material for spintronic and magnonic devices and is used here as a demonstration for the potential of the TNC technique. However, the material is unoptimized for TNC and the reasonable as-grown values ( 700 kA/m) of our thin films leave only modest improvement on TNC application. Nevertheless, the ability to modify selected areas, varying the materials properties, in a pre-determined pattern offers substantial opportunity in the design of future magnetic and magnonic devices. We imagine opportunities to design 2D magnonic crystals whereby, through judicious choice of underlying material, one may tune , anisotropy, exchange coupling or other material parameters in a continuous manner. This has already been demonstrated with exchange bias resetting in ferromagnet/antiferromagnet coupled bilayers18 and, here, we show TNC annealing can directly improve for selected nanoislands. Despite the presented consistent trends in the optimally annealed islands, differences between dynamic and static measurements, in addition to variability in the measured change in in dynamic experiments for nanoislands within the same array point to potential inconsistent contact between tip and sample. Refining contact feedback should allow precise control of this factor and remains a promising avenue of further research for those working in the field.
In conclusion, in this work, we demonstrate the use of post-fabrication TNC to enhance in pre-patterned nanoislands, using a heated scanning probe tip. Comparing quasistatic and dynamic MOKE measurements, we determine the short anneal time and localized heating profile increases , approaching bulk 800 kA/m at 250 °C. Comparing with micromagnetic simulations demonstrates that the TNC technique can be used to alter the remanent state of the islands, from the single domain to the vortex. These results demonstrate the opportunities afforded by TNC to modify the properties of selected areas, particularly when designing magnonic crystals and other nanomagnetic devices.
SUPPLEMENTARY MATERIAL
See the supplementary material for movies of TR-MOKE image sequences acquired from an individual 300 × 200 nm2 elliptical nanoisland at different bias magnetic fields showing uniform and non-uniform modes of excitation at 6 GHz.
The work at the University of Liverpool was supported by the UK EPSRC, through Grant No. EP/V035134/1, and the University of Liverpool SEM and nanolithography shared research facilities. The work at the University of Exeter EXTREMAG facility was supported by the UK EPSRC, through Grant Nos. EP/R008809/1 and EP/V054112/1.
AUTHOR DECLARATIONS
Conflict of Interest
The authors have no conflicts to disclose.
Author Contributions
Oliver James Barker: Data curation (equal); Formal analysis (equal); Investigation (equal); Methodology (equal); Writing – original draft (equal); Writing – review & editing (equal). Armaan Mohammadi-Motlagh: Methodology (equal); Writing – review & editing (supporting). Alexander James Wright: Investigation (supporting); Methodology (equal); Writing – review & editing (equal). Robert Batty: Formal analysis (supporting); Investigation (supporting). Harry Finch: Investigation (supporting). Andrea Vezzoli: Investigation (supporting). Paul Steven Keatley: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Investigation (equal); Methodology (equal); Resources (equal); Writing – original draft (equal); Writing – review & editing (equal). Liam O'Brien: Conceptualization (lead); Formal analysis (equal); Funding acquisition (lead); Project administration (lead); Resources (lead); Supervision (lead); Writing – original draft (equal); Writing – review & editing (equal).
DATA AVAILABILITY
The data that support the findings of this study are available from the corresponding authors upon reasonable request.