Magneto-ionic control of magnetic anisotropy is an emerging voltage-controlled approach that aims to offer much lower power consumption than current-controlled manipulation of magnetization. Moreover, magneto-ionic systems are ideal candidates for non von Neumann computing architectures, such as neuromorphic and stochastic computing due to their non-volatile and analog nature. One of the key metrics to quantify the efficiency of voltage-controlled magnetic anisotropy (VCMA) is the magneto-electric voltage coefficient (). Here, we show greater than one order of magnitude improvement in this efficiency compared to existing solid-state systems using a Co/Pd multilayer heterostructure. By performing a systematic study of the Co thickness, the Pd thickness, and the number of repeat units of engineered Co/Pd multilayers, we identify a narrow bandwidth of the Co thickness from 2–2.5 Å, Pd thickness from 1.4–1.7 nm, and repeat units from 7–9, to maximize the VCMA. Compared to rivaled liquid electrolyte systems, this platform has the advantage of faster speeds and easier integration for on-chip logic and memory devices.
As the limits of complementary metal-oxide semiconductor technology are being reached, spin-based devices are being deployed to reduce power consumption and scale.1 The primary logic operation is the switching of the magnetization of one or more magnetic layers. This has typically been done using magnetic fields generated by currents, spin-polarized current-induced spin transfer torques (STT), or spin-orbit torques (SOT).2,3 However, the high current densities and undesirable Joule heating lead to high energy consumption and challenges with heat dissipation. Using an electric field to control magnetization has arisen as a promising alternative for on-chip devices due to its low-power consumption, reversibility, speed, and non-volatility.4–7 Magneto-ionic gating is an emerging voltage-controlled approach that aims to both electrically control and offer substantial modulation of several magnetic interactions and material properties.8,9 Hydrogen based devices, as opposed to oxygen,8–12 have shown the ability to modulate magnetic properties at higher speeds and over more cycles than any other ion without causing structural degradation.13
Magneto-ionic systems are ideal candidates for non von Neumann architectures due to their low power, non-volatile, and analog nature. Scaling current generation devices down would push power consumption down to 1–10 aJ/bit.14,15 Moreover, the continuous changes in magnetic properties, such as saturation magnetization and coercivity, allow for voltage-dependent learning processes that are desired in stochastic or neuromorphic architectures, magnetic random access memory, and logic devices.16,17
The experiments herein used a multilayer heterostructure consisting of Ta(3.5)/Pt(5)/Co(X)/[Pd(Y)/Co(X)]N/Pd(6)/GdOx(26)/Au(8) (thickness in nm) where the Co thickness (X), Pd thickness (Y), and number of repeating units (N) were systematically explored from 0.1–0.35 Å, 0.6–3.45 nm, and 3–11 repeat units, respectively. All of the samples were grown using DC magnetron sputtering at room temperature and a background pressure Torr on 50 nm wet thermally oxidized silicon (100) substrates. Au was grown under 3.5 mTorr Ar while the rest of the metal layers were grown under 3 mTorr Ar. The GdOx layer was deposited using radio-frequency sputtering with 3 mTorr Ar and 0.7 mTorr O2. Shadow masking was used to pattern circular Au electrodes on top of the GdOx with an approximate diameter of 100 μm while the continuous bottom metal layer served as the counter electrode.
The heterostructure is schematically shown in Fig. 1(a). In the virgin state, all films exhibited perpendicular magnetic anisotropy (PMA) measured via polar magneto-optical Kerr effect (MOKE) microscopy. The voltage-controlled hydrogen (H) gating of these films was studied using the solid-state proton pump platform introduced in Ref. 9 and used to gate Pd/Co/Pd trilayers. Exemplary hysteresis loops for a [Pd(1.5)/Co(0.25)]5 heterostructure are shown in Fig. 1(b). In the ungated virgin state (black line), the square out-of-plane (OOP) hysteresis loop indicates PMA while under a positive gate voltage (Vg) = +4 V (purple line), H is sourced from atmospheric H2O split at the Au/ambient interface,18 loaded into the heterostructure, and leads to in-plane (IP) anisotropy as evidenced by the flat OOP hysteresis loop. When a negative Vg is subsequently applied, H is unloaded from the system, and PMA returns. However, unlike the trilayer system studied in Ref. 9, during the first few gating cycles, the coercivity of the H-unloaded state progressively increases in an irreversible fashion, and only after several cycles does the coercivity of the unloaded state stabilize. This is seen in Fig. 1(b), where for each cycle Vg was applied for 30 s at +4 V followed by -1 V for 30 s. The application of higher negative biases does not allow for the recovery of the virgin state. After several cycles, the coercivity of the unloaded state begins to saturate, while still allowing the anisotropy to be toggled from OOP to IP by H loading/unloading reversibly, as seen in Fig. S1.
Irreversible coercivity enhancement due to magneto-ionic gating has previously been seen in exchange-biased Pd/Co/Pd tilayers.11 The irreversible change in that study was attributed to a phase transformation of the Co from an amorphous to a crystalline state—revealed by cross-sectional high resolution transmission electron microscopy (HR-TEM). However, cross-sectional HR-TEM for a [Co(0.2)/Pd(2)]11 multilayer heterostructure in this study shows no discernible change in the structure of the magnetic multilayers between its virgin state [Fig. 1(c)] and after several gating cycles [Fig. 1(d)]. In fact, in the virgin state, the Co/Pd multilayer is already crystalline, as seen in Fig. S2(a). Modification of the coercivity in Co/Pd multilayers has also been seen in studies using gaseous H exposure.19–22 However, these studies have conflicting results. The hydrogen increases the coercivity for some and decreases for others. Moreover, the mechanism by which the coercivity changes is claimed to be structural in some studies and electronic in others. A recent study using density functional theory provided some insight into the previous contradicting results.23 It was shown that small amounts of H that enter the bulk of the thin film can enhance the PMA while larger amounts at the interface will reduce the PMA. The enhancement of PMA can be attributed to a modification of the structural properties, such as the formation of a -Pd-H phase and subsequent lattice expansion.19 By contrast, the reduction of PMA can be attributed to a modification of the electronic properties, namely, the hybridization between 4d Pd orbitals and 3d Co orbitals.23,24 These two possible contributions would lead to a non-monotonic and potentially competing modification of the Co/Pd multilayer coercivity under H loading.
One possible structural transformation of the Co/Pd multilayer that could lead to an irreversible increase in coercivity is the H-induced phase transformation of α-Pd-H to β-Pd-H. This would be accompanied by an increase in the d111 spacing by approximately 4%–5% (2.25–2.35 Å).19 At room temperature, a solid solution of the α- and β-Pd-H forms at < 0.01 H/Pd25 which is far below the calculated 0.5 H/Pd23 needed to achieve reversible voltage-induced switching of the anisotropy, as seen in Fig. 1(b). Therefore, if the formation of the β-Pd-H phase is the rationale behind the irreversible increase in coercivity, it should be resolvable with HR-TEM. To look for evidence of such a transformation in our system, we performed fast Fourier transformations (FFTs) of cross-sectional HR-TEM images (Fig. S3), i.e., the extracted diffractograms [Figs. 2(a) and 2(b)] of a virgin device and a device after several gating cycles. A radial line scan of the diffractograms, shown in Fig. 2(c), shows no discernible shift corresponding to the formation of a β phase, i.e., the virgin and gated reflections overlap. Therefore, it is unlikely that a β-Pd-H phase is present and is the reason for the irreversible increase in coercivity. It is also possible that upon sample preparation of the cross-sectional lamellas, H escapes from the Co/Pd multilayer and causes the spontaneous phase transformation of β-Pd-H back to α-Pd-H. However, identically prepared lamellas were found to reveal the H-induced crystallization of exchange-biased Pd/Co/Pd trilayers.11 An alternative explanation to the irreversible increase in coercivity may be found in more complex electronic structure effects induced by small quantities of H that become irreversibly trapped in the lattice; however, it is beyond the scope of this work.
Despite the irreversible increase in coercivity of the unloaded state, the voltage-induced loaded state still allows for fully reversible voltage-control of magnetic anisotropy (VCMA) from OOP to IP. In order to maximize the VCMA, we performed a systematic study of the coercivity change and gateability of several heterostructures as a function of Co thickness (X), Pd thickness (Y), and the number of repeating units (N), shown in Fig. 3. Gate voltages of +3 to +5 V for 15–30 s per cycle were used in order to drive the anisotropy IP without causing breakdown of the device.
First, the Co thickness was varied in the heterostructure: Ta(3.5)/Pt(5)/Co(X)/[Pd(1.5)/Co(X)]5/Pd(6)/GdOx(26)/Au(8) (thickness in nm), shown in Fig. 3(a). The coercivity of the virgin state, after one gating cycle, and after several gating cycles was measured using polar MOKE. For low Co thicknesses, the heterostructure can be gated IP, but the coercivity is relatively low. The saturated state coercivity increases with Co thickness up to 3 Å; however, when the Co is too thick ( 3 Å), the heterostructure does not go fully IP in the H-loaded state. The heterostructures that did not go IP when a gate voltage was applied are indicated by the red region in Figs. 3(a) and 3(b). Exemplary hysteresis loops of [Co(0.35)/Pd(1.5)]5 in the unloaded and H-loaded states are shown using polar MOKE in Fig. 4(b). It is clear that the anisotropy is reduced in the H-loaded state, but it is not enough to drive the magnetic layer IP. This could be due to an increase of the interfacial anisotropy, with only minimal changes to the volume and shape anisotropy, as the Co thickness increases. This is well known for the Co/Pd multilayer system in the low Co thickness regime.26
Similar results are seen for the variation in Pd thickness using the heterostructure: Ta(3.5)/Pt(5)/Co(0.2)/[Pd(Y)/Co(0.2)]5/Pd(6)/GdOx(26)/Au(8) (thickness in nm), shown in Fig. 3(b). However, the red region occurs when the Pd is too thin ( 8 Å), indicating H cannot generate enough variation in the magnetic anisotropy to push the multilayer IP. In addition, the irreversible increase in coercivity is strongly reduced for thick Pd ( 2.5 nm). This is likely because the diffusion of H deep into the multilayers is diminished and, therefore, the lower concentration of H at the deep Co/Pd interfaces leads to a smaller change in the interfacial anisotropy.
Finally, using the heterostructure: Ta(3.5)/Pt(5)/Co(0.2)/[Pd(2)/Co(0.2)]N/Pd(6)/GdOx(26)/Au(8) (thickness in nm), the number of repeating units (N) was changed. Beyond N = 7, as seen in Fig. 3(c), there is a broad plateau and gating does not significantly impact the VCMA or the irreversible increase in the coercivity. This is likely due to two competing factors that reach an equilibrium after only a few repeats. By increasing N, the film thickness increases, decreasing the concentration of H that reaches the buried layers. However, the number of interfaces that contribute to the interfacial PMA also increases, unlike the case for increasing the Pd thickness, H can still penetrate the uppermost interfaces. If N were to increase significantly more, then the irreversible increase in coercivity would likely decrease as the proportion of H-modified Co/Pd interfaces has decreased relative to the total number of interfaces. The beginnings of this can already be seen in the downturn of the irreversible increase in coercivity at N = 11 in Fig. 3(c).
To maximize the VCMA the optimal range of Co thickness is 2–2.5 Å, Pd thickness is 1.4–1.7 nm, and N is 7–9. Figure 4(a) demonstrates this growth optimization via gated polar MOKE hysteresis loops of the heterostructure: Ta(3.5)/Pt(5)/Co(0.2)/[Pd(1.65)/Co(0.2)]5/Pd(6)/GdOx(26)/Au(8). With only a small bias voltage V, the OOP Co/Pd multilayer goes from Oe to IP in less than 15 s. The efficiency of the VCMA can be calculated using the magneto-electric voltage coefficient , giving a value of 858 Oe/V. This is more than an order of magnitude greater than any previous demonstration for a solid-state electrolyte system.7 Comparing several solid and liquid electrolyte systems, shown in Fig. 5 (with references included in Fig. S4), the VCMA effect achieved here only falls short behind that observed in high-anisotropy SmCo5.27 However, in that system, the authors did not demonstrate the ability to gate the anisotropy fully in-plane, the device took 1 h to charge, and the gating was performed using a liquid electrochemical cell, all together making this system less practical for on-chip logic or memory.
In conclusion, we demonstrate greater than one order of magnitude improvement in the magneto-electric voltage coefficient by engineering a multilayer heterostructure of Co and Pd. By increasing the number of interfaces, we can dramatically increase the coercivity of the heterostructure without needing to compensate with larger gate voltages. The efficiency of these solid-state devices rivals liquid electrolyte systems, however, have the advantage of faster speeds and easier integration for on-chip logic and memory devices. With scaling the electrolyte thickness and device size down, we anticipate dramatic improvements in device speed, making them an appealing alternative to reduce energy consumption as compared to current-controlled spintronic devices.
See the supplementary material for the methods, including the polar MOKE measurements and structural characterization, and for the additional figures regarding the coercivity evolution through cycling, HR-TEM, and Fig. 5 with references.
This work was supported by the National Science Foundation (NSF) through the Massachusetts Institute of Technology Materials Research Science and Engineering Center (MRSEC) under Award No. DMR-1419807 and Spintronic Materials for Advanced Information Technologies (SMART) (No. 2018-NE-2861), one of seven centers of nCORE, a Semiconductor Research Corporation program, sponsored by the National Institute of Standards and Technology (NIST). The authors would like to thank Dr. Jonas Zehner for helpful discussions regarding the TEM and Tina Walter for preparing the TEM-lamellas.
Conflict of Interest
The authors have no conflicts to disclose.
Alexander E Kossak: Conceptualization (lead); Data curation (lead); Formal analysis (lead); Investigation (lead); Methodology (lead); Project administration (lead); Writing – original draft (lead); Writing – review & editing (lead). Daniel Wolf: Data curation (equal); Formal analysis (equal); Investigation (equal); Writing – review & editing (equal). Geoffrey S. D. Beach: Conceptualization (equal); Funding acquisition (lead); Supervision (lead); Writing – review & editing (equal).
The data that support the findings of this study are available from the corresponding author upon reasonable request.