# Double Pinned Perpendicular-Magnetic-Tunnel-Junction Spin-Valve Providing Multi-level Resistance States

• 537 Accesses

## Abstract

A new design for high density integration greater than gigabits of perpendicular-magnetic-tunnel-junction (p-MTJ) spin-valve, called the double pinned (i.e., bottom and top pinned structures) p-MTJ spin-valve achieved a multi-level memory-cell operation exhibiting four-level resistances. Three key magnetic properties, the anisotropy exchange field (Hex) of the bottom pinned structure, the coercivity (Hc) of the double free-layer, and the Hc of the top pinned structure mainly determined four-level resistances producing tunneling-magnetoresistance (TMR) ratios of 152.6%, 33.6%, and 166.5%. The three key-design concepts are: i) the bottom pinned structure with a sufficiently large Hex to avoid a write-error, ii) the Hc of the double free-layer (i.e., ~0.1 kOe) much less than the Hc of the top pinned structure (i.e., ~1.0 kOe), and iii) the top pinned structure providing different electron spin directions.

## Introduction

Perpendicular spin-transfer-torque magnetic-random access memory (p-STT MRAM), which consists of a perpendicular magnetic tunneling junction (p-MTJ) spin-valve and a selective device, has attracted great research interest because of its possibility in various applications. Recently, in particular, the p-STT MRAM cells have been utilized as embedded memory in system-on-chip for mobile and internet-of-things applications1,2,3,4,5, a stand-alone memory as a solution to the dynamic-random-access-memory (DRAM) scaling limitations below the 10-nm node6, and spin-neuron and synaptic devices for deep learning7,8,9. The p-STT MRAM has many advantages over current memory devices, such as non-volatility, fast read/write speed (~10 ns), extremely low power consumption (<1 pJ/bit), high write endurance (>1012), and scalability10,11,12,13,14,15.

The researches on p-STT MRAM have been based on improving three device parameters of the p-MTJ spin-valves11,12,13: the tunneling magnetoresistance (TMR) ratio greater than 150% for ensuring a memory margin, the thermal stability (Δ = KuV/kBT, where Ku, V, kB, and T are the magnetic anisotropy energy, the volume of the free ferromagnetic layer, the Boltzmann constant, and the temperature, respectively) above 75 for a ten-year retention-time, and the switching current density of about 1 MA/cm2 for low power consumption. Moreover, these device parameters should be available at a back end of line (BEOL) temperature > 350 °C16,17. The conventional double MgO based p-MTJ spin-valve consist of upper and lower synthetic anti-ferromagnetic (SyAF) [Co/Pt]n multilayer separated by a Ru spacer, a Co2Fe6B2 magnetic pinned layer, a MgO tunneling barrier, and Co2Fe6B2 magnetic free layers, as shown in Fig. 1a: called a single pinned p-MTJ spin-valve18. A single pinned p-MTJ spin-valve can generate only two resistance states: the high-resistance state (HRS) from the anti-parallel spin direction, and the low-resistance state (LRS) from the parallel spin direction between the Co2Fe6B2 free layer and the Co2Fe6B2 pinned layer. Thus, these two different resistance states have been used to operate only single-bit p-STT MRAM memory cells. However, the scaling down for terabit-level integration of the p-STT MRAM cells, to compete with current DRAM19, 3-dimensional (3D) NAND flash memory, and 3-dimensional cross-point memory, would be necessary to achieve the thermal stability required for 10-year retention time20,21 and multi-level memory-cell operation like that of current 3-D NAND flash memory22,23. In our research, we designed a double pinned p-STT-MTJ spin-valve exhibiting multi-level (i.e., four) resistance states. The double pinned p-STT MRAM was composed of three main ferro-magnetic component layers: the bottom pinned structure, double free-layer, and top pinned structure (see Fig. 1b). The TMR ratio was maximized by introducing a bottom single SyAF [Co/Pt]n multilayers (Fig. 1b) because the upper [Co/Pt]3 SyAF multilayer anti-ferro-coupled with the lower [Co/Pt]6 SyAF multilayers (Fig. 1a) via a Ru spacer produced considerably high surface roughness18. The design concept of the double pinned p-STT MTJ will be explained in more detail in the following section. In addition, we investigated static magnetic properties of the double pinned p-MTJ spin-valve, tested the achievement of four-level magnetic-resistance states, and analyzed the operation mechanism of four-level resistances.

## Results

### Design of Double Pinned p-MTJ Spin-valve

The double pinned p-MTJ spin-valve was vertically stacked with the bottom electrode, bottom Co2Fe6B2 ferromagnetic pinned structure (called bottom pinned structure), double MgO based Co2Fe6B2 ferromagnetic free layer (called the double free-layer), top Co2Fe6B2 ferromagnetic pinned structure (called top pinned structure), and top electrode, as shown in Fig. 1b. The magnetic layers of the double pinned p-MTJ spin valve can be divided largely into five groups (M1, M2, M3, M4, and M5 layers). The bottom electrode was made by sputtering tungsten (W) and titanium nitride (TiN) layers on a thermally oxidized 300-mm-diameter Si wafer, followed by the chemical-mechanical-planarization (CMP). For the bottom pinned structure, the Ta buffer layer was used to intersect the f.c.c. crystalline texturing of the TiN electrode since the Ta layer had an amorphous structure. The face-centered-cubic crystalline Pt-seed-layer was used to form the L10 crystalline structure of the bottom-lower SyAF [Co (0.47 nm)/Pt(0.23 nm)]3 multilayer (M1 layer) so that its spin direction was perpendicularly upward, as shown in Fig. 1c24,25,26. These layers were anti-ferro-coupled with the single Co (0.51 nm)/ Pt(0.23 nm)/Co(0.47 nm) buffer layer via Ru spacer, called the single SyAF [Co/Pt]n layer. Simultaneously, the single Co/Pt/Co buffer-layer was ferro-coupled to the bottom Co2Fe6B2 ferromagnetic pinned layer (0.95 nm) via the W bridge layer, defined as M2 layer. The spin direction of the M2 layer was perpendicularly downward, as shown Fig. 1c. Thus, the spin directions of the M1 and M2 layers always face vertically inward towards each other. In particular, the anisotropy exchange field (Hex) of the bottom pinned structure should be sufficiently higher than the coercivity (Hc) of the top pinned structure to fix the spin direction of both the M1 and M2 layers. Then, the double Co2Fe6B2 free-layer (M3 layer) was stacked on the M2 layer, where the thicknesses of the bottom MgO tunneling-barrier, Fe insertion layer, lower Co2Fe6B2 ferromagnetic free layer, W spacer, upper Co2Fe6B2 ferromagnetic free layer, and top MgO tunneling-barrier layer were 1.15, 0.3, 1-0.05, 0.2, 1.05, and 1.0 nm, respectively. The spin direction of M3 layer was dependent of the polarity of the applied perpendicular-magnetic-field; i.e. vertically downward for a negative field and vertically upward for a positive field, as shown in Fig. 1c. In particular, the Hc of M3 layer should be considerably smaller than the Hc of the top pinned structure to make four different spin direction states between the M3 and M4 layers [i.e., anti-parallel (AP) state 1, AP state 2, parallel (P) state, and AP state 3]. For the top pinned structure, the Fe insertion layer (0.3 nm) and top Co2Fe6B2 ferromagnetic pinned layer (0.75 nm) were stacked on the M3 layer. The top-lower [Co(0.47 nm)/Pt(0.23 nm)]3 SyAF multilayer was ferro-coupled with the top Co2Fe6B2 ferro-magnetic pinned-layer (0.75 nm) via a W bridge layer and Co/Pt seed layer, which is defined as M4 layer. The spin direction of the M4 layer was dependent of the polarity of the applied magnetic-field. Simultaneously, M4 layer was always anti-ferro-coupled with the top-upper [Co(0.47 nm)/Pt(0.23 nm)]3 SyAF multilayers (M5 layer), via the Ru spacer. Thus, the spin direction of the M5 layer was always in the opposite of the M4 layers, as shown in Fig. 1(c). In particular, the number (m) of the top-lower [Co(0.47 nm)/Pt(0.23 nm)]m SyAF multilayers should be higher than that (n) of the top-upper [Co(0.47 nm)/Pt(0.23 nm)]n multilayers to produce four-different spin direction states between the M3 and M4 layers. If m is lower than n, only two different spin direction states would be generated between the M3 and M4 layers. Finally, a top Ta/Ru electrode was stacked on the M5 layer.

In summary, the design of the double pinned p-MTJ spin-valve could produce four-different spin direction states between the M3 and M4 layers: AP state 1 (perpendicularly upward spin direction for both M3 and M4 layers), AP state 2 (spin direction facing outward between M3 and M4 layers), P state (downward spin direction for both M3 and M4 layers), and AP state 3 (spin direction facing inward between the M3 and M4 layers, as shown in Fig. 1c). To form four-different spin direction states, we essentially need three key-design concepts: 1) for designing the bottom pinned structure, the spin direction of the M1 and M2 layers should face always inward toward each other, 2) for designing M3 layer, its Hc should be remarkably smaller than the Hc of the M4 layer to assure to produce four-different spin direction states between the M3 and M4 layers, and 3) for designing the top pinned structure, the Hex of the M4 layers should be sufficiently higher than the Hc of the the M3 layer to avoid a write-error. These three key-design concepts will be treated later in detail.

### Design and Static Perpendicular-Magnetic Behaviour of Bottom Pinned Structure and Double Free-Layer

In the bottom pinned structure (M1 and M2 layers) in Fig. 1b, the spin direction of the M2 layer should always face perpendicularly inward toward with that of the M1 layer, as shown in Fig. 1c. Thus, this bottom pinned-structure needs as large Hex of the M2 layer as possible. In our previous studies, the spin directions of the M1 and M2 layers are always facing perpendicularly inward against each other when the number of the bottom-upper SyAF [Co(0.4 nm)/Pt(0.2 nm)] layer (i.e., 3) is less than that of the bottom-lower SyAF [Co(0.4 nm)/Pt(0.2 nm)] layer (i.e., 6)27,28,29,30. Also, we implemented the p-MTJ spin-valves with a single SyAF [Co/Pt]n layer which showed large Hex, as shown in Supplementary 118.

Using this concept, we redesigned the bottom pinned p-MTJ spin-valve (Fig. 2a,b) with double and single SyAF [Co/Pt]n layers to match that of the double pinned p-MTJ spin-valve and investigated its magnetic properties. The arrow magnitude and direction corresponded to the relative magnetic-moment and spin direction of the magnetic layers when the applied magnetic field is changed from +6.5 kOe to −6.5 kOe, as shown in Fig. 2c,d. The inset of Fig. 2c,d shows the magnetic properties of the Co2Fe6B2 free layers (M3 layers) of the p-MTJ spin-valve with double and single SyAF [Co/Pt]n layers. The spin direction of both structures are aligned along the external field direction when external field is high in the upward direction (H > +5 kOe). When the external field becomes smaller than + 1.5 kOe, the spin direction of the M2 layers are switched opposite the external field as the field is not strong enough to overcome the anti-ferro coupling between M1 and M2 layers. The M2 layer squareness of the bottom p-MTJ spin-valve with single SyAF [Co/Pt]n layers (red box of Fig. 2d) is degraded compared to that of the spin-valve with double SyAF [Co/Pt]n layers (blue box of Fig. 2c). However, the Hex is increased from 2.35 kOe to 3.44 kOe which would mean that the M1 and M2 layers of the p-MTJ spin-valve with single SyAF [Co/Pt]n layers is more unsusceptible to switching. In addition, the peak-to-valley (ΔP-V) of the MgO tunneling barrier decreased from 2.03 nm to 1.75 nm when the thickness of the SyAF layers is reduced from 8.87 to 4.65 nm as shown in Supplementary 2. Note that the TMR ratio of the double pinned p-MTJ spin-valve is expected to increase by reducing the roughness of the MgO tunneling barrier27,28,29,30,31,32,33,34,35,36,37,38. Thus, the magnetic property of the double free-layer was not degraded; i.e., a good squareness and the magnetic moment of ~ 0.2 memu, as shown in the inset of Fig. 2d. This result indicates that a single SyAF [Co/Pt]3 multi-layers would be very suitable as a bottom pinned structure (M1 and M2 layers in Fig. 1b), since it could provide a sufficiently high Hex of 3.44 kOe and would increase the TMR ratio because of a lower surface roughness of the MgO tunneling-barrier. The spin direction schematic in Fig. 2c,d represents the change only when the magnetic field is changed from +6.5k Oe to −6.5 kOe. A detailed magnetic switching behavior of the p-MTJ spin-valve with double and single SyAF [Co/Pt]n layer under an external magnetic field sweep from −6.5 kOe to +6.5 kOe is shown in see Supplementary 3.

### Design and Static Perpendicular-Magnetic Behaviour of Top Co2Fe6B2 Ferro-Magnetic Pinned Structure

In the top pinned structure, the top Co2Fe6B2 magnetic pinned-layer was ferro-coupled with the top-lower SyAF [Co/Pt]m layer via W bridge layer (M4 layer), which was then anti-ferro-coupled with the top-upper SyAF [Co/Pt]n layer (M5 layer) via Ru spacer, as shown in Fig. 3a. In addition, the number of the top-lower [Co(0.47 nm)/Pt(0.23 nm)] layers (m) of the M4 layer should be less than the number of the top-upper [Co(0.47 nm)/Pt(0.23 nm)] layers (n) of the M5 layer, and the Hex of the top pinned structure should be as large as possible to avoid a write-error. Unlike the bottom pinned structure in Fig. 2d, the top pinned structure should be able to generate four different electron-spin states between the M3 and M4 layer, resulting in four different resistance states. Thus, within the scanning magnetic-field range less than the Hex of the top pinned structure, the spin direction of the M4 layer could be rotated from upward to downward or downward to upward when the polarity of the magnetic-field changes from positive to negative or from negative to positive. To test whether or not the spin directions of the top pinned structure were variable, we investigated the M-H loop of a basic top pinned-structure with the m:n ratio of 6:3 of the number of [Co(0.47 nm)/Pt(0.23 nm)] multilayers in the M4 and M5 layers, as shown in Fig. 3b. At the applied magnetic-field of +15 kOe, the spin direction of both the M4 and M5 layers faced perpendicularly upward. As the magnetic-field decreased from +15 to +4 kOe, the magnetic moment decreased from +0.6 to +0.3 memu, corresponding to the magnetic moment of the M5 layer (i.e., 0.3 memu) rotating the spin direction of the M5 layer from upward to downward. This occurred at the Hex of ~4.9 kOe arising from the anti-ferro coupling across the Ru spacer layer. As the magnetic-field decreased from +4 to −4 kOe, the magnetic moment changed from +0.3 to −0.3memu, responding to the magnetic moment of the M4 and M5 layers (i.e., 0.6 memu), rotating the spin direction of the M4 layer from upward to downward. Simultaneously, the spin direction of the M5 layers rotated from downward to upward to hold the anti-ferro coupling via the Ru spacer stably. As a result, the spin directions of the M4 and M5 layers facing perpendicularly inward changed to facing perpendicularly outward. As the magnetic-field increased over −4 kOe, the spin direction of the M5 layer rotated from upward to downward so that the spin directions of both the M4 and M5 layers were perpendicularly downward. In contrast, as the magnetic-field changed from negative to positive direction, the change of the spin directions of the M4 and M5 layers followed the same order as the magnetic-field changed from positive to negative direction. In particular, the spin directions of the M4 and M5 layers facing perpendicularly outward changed to facing perpendicularly inward. Thus, this top pinned structure could produce two spin directions between the M4 and M5 layers when the magnetic-field is greater than the Hc of the top pinned structure; facing perpendicularly outward for the negative magnetic-field and facing perpendicularly inward for the positive magnetic-field. Although the top-lower SyAF [Co(0.47 nm)/Pt(0.23 nm)]6 layer and top-upper SyAF [Co(0.47 nm)/Pt(0.23 nm)]3 layer and could provide variable spin directions between the M4 and M5 layers, the top-lower SyAF [Co(0.47 nm)/Pt(0.23 nm)]6 layer is too thick to maximize the TMR ratio of the p-MTJ spin-valve. The TMR ratio is strongly dependent on the coherent tunneling of the Δ1 Bloch state induced from the hybridization of the Fe-dz2 and O-pz orbitals at the MgO/CoFeB interface. Even a small defect at the interface reduces the coherent tunneling of the spin-polarized electrons31,39. The roughness increases with the number of top-lower SyAF [Co/Pt]n layers from about 150 pm to 240 pm as seen in Supplementary 4. The roughness needs to be reduced to maximize the TMR ratio to assure four levels of the double pinned p-MTJ spin-valve.

To reduce the thickness of the top pinned structure, first of all, we investigated the M-H loop of the top pinned structure with the ratio of the [Co(0.47 nm)/Pt(0.23 nm)]3 layers of the M4 and M5 layers (m:n ratio) of 3:1, as shown in Fig. 3c. Its static magnetic behaviour was similar to the top pinned structure with the ratio of m:n ratio of 6:3, showing the magnetic moment of the M4 and M5 layers, Hc, and Hex were 0.1 and 0.5memu, ~0.3 kOe, and ~12 kOe. Although this top pinned structure could produce two variable spin directions between the M4 and M5 layers, the Hc of ~0.3 kOe was too small to generate four different spin direction states between the M3 and M4 layers in the double pinned p-MTJ spin-valve (Fig. 1c) since the Hc of the top pinned structure (i.e., ~0.3 kOe) was not sufficiently higher than the Hc of the double free-layer (Fig. 2d: i.e., ~0.2 kOe) to avoid the write error. Thus, we observed the dependency of the Hc of the top pinned structure on the m:n ratio, as shown in Fig. 3d. The m:n ratios of 1:3, 2:3, and 3:3 showed Hc values of ~0.4, ~0.6, and ~1 kOe, as shown in the inset of Fig. 3d, and all m:n ratios demonstrated two variable spin directions between the M4 and M5 layers. This result indicates that the Hc of the top pinned structure increases with the number of [Co(0.47 nm)/Pt(0.23 nm)]3 layers in the M5 layer, while Hex decreases with the number of [Co(0.47 nm)/Pt(0.23 nm)]3 layers in the M5 layer. In particular, the m:n ratio of 3:3 showed a sufficient Hc (i.e., ~1 kOe), which could produce four different spin direction states between the M3 and M4 layers (Fig. 1c) since it is considerably larger than the Hc of the double free-layer (~0.1 kOe) (Fig. 2d). Recall that at the m:n ratio of 3:3 the static magnetic moment of the M4 layer was slightly larger than that of the M5 layer so that the spin direction of the top pinned structure (M4 and M5 layers) faced vertically outward for the negative applied magnetic-field and vertically inward for the positive applied magnetic-field. If n is larger than m, the spin directions of the M4 and M5 layers could not be variable. Therefore, the design of choosing a proper Hc and Hex of the top pinned structure would be a key research to stably produce four different spin direction states between M3 and M4 layers.

### Static Perpendicular-Magnetic Behaviour and Multi-level TMR ratio for Double Pinned p-MTJ Spin-Valve

By combining the top (3:3 of m:n in Fig. 3d) and bottom p-MTJ structures with the double free-layer (Fig. 2d), we fabricated the double pinned p-MTJ spin-valve shown in Fig. 1b. The M-H loop of the double pinned p-MTJ spin-valve showed the Hex of 4.9 kOe when the applied magnetic-field was scanned from −6.5 kOe to +6.5 kOe, as shown in Supplementary 5a. In addition, the resistance-vs.-magnetic-field (R-H) loop presented only three resistance states, when the applied magnetic-field was scanned from −Hex to +Hex kOe, as shown in Supplementary 5b,c. In order to produce four different resistance states, thus, the maximum scanning range of the applied magnetic-field should be sufficiently less than $$\pm$$Hex (i.e., ~4.2 kOe) of the bottom pinned p-MTJ spin-valve, but greater than ±Hex (i.e., ~1 kOe) of the top pinned structure (M4 and M5 layers); i.e., $$\pm$$ 2 kOe. Thus, four different spin directions between the M3 and M4 layers could be stably produced when the applied magnetic-field was scanned from −2 kOe to +2kOe, as shown in the M-H loop of Fig. 4a. First, the AP state 1 was produced when the applied magnetic-field was scanned from +2 kOe to +0.5 kOe, where the spin directions of both the M3 and M4 layers were vertically upward and parallel while the spin directions of the M4 and M5 layers faced vertically inward toward each other via an anti-ferro-coupling, as shown Fig. 4a,b. Recall that this result corresponds to the combination of the top pinned structure (M4 and M5 layers) in Fig. 3d and the bottom pinned structure (M1 and M2 layers) with the double free-layer (M3 layer) in Fig. 2d at the positive applied magnetic-field. Then, when the applied magnetic-field was scanned from +0.5 kOe to −0.5 kOe, the spin direction of only the M3 layer was rotated from upward to downward while the spin directions of both the M4 and M5 layer did not change, generating the AP state 2, where the spin direction of the M3 layer faced vertically outward against that of the M4 layer, as shown in i in Fig. 4a,b. Furthermore, when the applied magnetic-field was scanned from −0.5 kOe to −2.0 kOe, the spin directions between the M4 and M5 layers facing vertically inward were rotated to face vertically outward while the spin direction of the double free-layer was sustained downward, forming the P state, where the spin directions of both M3 and M4 layers were vertically downward and in parallel, as shown in ii in Fig. 4a,b. Subsequently, when the applied magnetic-field was scanned from −2.0 kOe to +0.5 kOe, the spin direction of only the M3 layer rotated from downward to upward while the spin directions of both the M4 and M5 layers remained facing vertically outward from each other, generating the AP state 3, where the spin direction of M3 layer faced vertically inward against that of the M4 layer, as shown in iii in Fig. 4a,b. Finally, the spin directions between the M4 and M5 layers facing vertically outward were rotated to face vertically inward while the spin direction of the M3 layer remained facing upward, returning to the AP 1 state, where the spin directions of both the M3 and M4 layers facing vertically inward transited to facing vertically and parallel upward, as shown in iv in Fig. 4a,b. As a result, four different spin directions between the M3 and M4 layers could be stably produced in a magnetic-field scanning range of $$\pm$$2.0 kOe.

The R-H loop corresponding to the M-H loop in Fig. 4a was shown in Fig. 4c. The resistance changed from the AP state 1, AP state 2, AP state 3, P state, and AP state 1 when the applied magnetic-field was scanned from +2.0 kOe, −2.0 kOe, and +2.0 kOe. The sequence of a higher resistance of the double pinned p-MTJ spin-valve was followed by AP state 3, AP state 1, AP state 2, and P state. The highest resistance was achieved when the M3 layers became the anti-parallel states against both the M2 and M4 layers; i.e., the AP state 3, corresponding to the sum of serial connection (RAP3) of the anti-parallel resistance between the M3 and M4 layers (Rt,AP) with the anti-parallel resistance between the M2 and M3 layers (Rb,AP), as shown in Fig. 4d. The second highest resistance was obtained, when the M3 layers became the parallel state against the M4 layers while it was in the anti-parallel state against the M2 layers; i.e., the AP state 1, responding to the sum of serial connection (RAP1) of the parallel resistance between the M3 and M4 layers (Rt,P) with the anti-parallel resistance between the M2 and M3 layers (Rb,AP). The third highest resistance was achieved, when the M3 layers became the anti-parallel state against the M4 layers while it did the parallel state against the M2 layers; i.e., the AP state 2, indicating to the sum of serial connection (RAP2) of the anti-parallel resistance between the M3 and M4 layers (Rt,AP) with the parallel resistance between the M2 and M3 layers (Rb,P). Note that the anti-parallel resistance between the M2 and M3 layers (Rb,AP) is much larger than the anti-parallel resistance between the M3 and M4 layers (Rt,AP) since the thickness of the bottom MgO tunneling barrier (1.15 nm) was greater than that of the top MgO tunneling barrier (1.0 nm). This is confirmed by the high-resolution transmission-electro-microscopy (HR-TEM) observation shown in Supplementary 3. Thus, the resistance of the AP state 1 was larger than that of the AP state 2.

Multi-level TMR ratios of the double pinned p-MTJ spin-valve were measure by CIPT (current-in-plane tunneling) measurement scanning the magnetic-field of $$\pm$$2 kOe. The TMR ratios were 152.6% for AP state 1, 33.6% for AP state 2, and 166.5% for AP state 3. These values correlated well with the four different resistance levels in the R-H loop in Fig. 4c. A higher TMR ratio was accounted for when the spin direction of the M3 layer becomes anti-parallel to that of the M2 layer (i.e., Rb,AP). This result evidently indicates that the double pinned p-MTJ-spin-valve can perform multi-level (i.e., four-level) non-volatile memory-cell operation.

## Discussion

Our proposed double pinned p-MTJ spin-valve well demonstrated four-level resistance as a multi-level p-STT MRAM-cell, resulting in the TMR ratios of 152.6, 33.6, and 166.5%. The maximum TMR ratio of the double pinned p-MTJ spin-valve (166.5%) was slightly less than that of a single pinned p-MTJ spin-valve (i.e., 180%), since the Pt atoms in the top pinned structure diffused into both top and bottom MgO tunneling barrier so that the coherent tunneling of the spin-electron would be decreased as shown Supplementary 6. Thus, research on how to avoid the Pt diffusion from the top pinned structure is necessary; i.e., research on the design of a nano-scale buffer layer preventing Pt atom diffusion. In addition, to minimize a write-error originated between four-level resistances, the differences between four-level resistances should be as constant as possible. Thus, research on choosing a proper thickness between the top and bottom MgO tunneling barriers is also necessary. Success in the above-mentioned research will enable us to fabricate a terabit-level p-STT MRAM for embedded, stand-alone, and neuromorphic devices.

## Methods

The p-MTJ spin-valves were fabricated using a 12-inch-wafer multi-chamber cluster-magnetron sputtering-system under a high vacuum (less than 1 × 10−8 torr). In particular, the conventional p-MTJ spin-valve with the top double free-layer and double SyAF [Co/Pt]n layers in Fig. 1a were fabricated by vertically stacking a 12-inch SiO2 wafer, W/TiN bottom electrode, Ta buffer layer, Pt seed layer, bottom-lower SyAF [Co(0.47 nm)/Pt(0.23 nm)]6 layers/Co(0.51 nm) (M1 layer), Ru spacer layer (0.85 nm), Co(0.51 nm)/Pt(0.23 nm)/ bottom-upper SyAF [Co(0.47 nm)/Pt(0.23 nm)]3 layers, and a Co buffer layer (0.47 nm). The W bridge layer of 0.22 nm was used to ferro-couple the bottom-upper SyAF layer to the pinned layer. The p-MTJ consisted of a Co2Fe6B2 bottom pinned layer (1.05 nm), MgO tunneling barrier (1.15 nm), Fe insertion layer (0.3 nm), Co2Fe6B2 free layers (1.05 nm), W spacer layer (0.4 nm), Co2Fe6B2 (1.05 nm), and MgO (1.0 nm)/W capping layer. The ferro-coupled bottom-upper SyAF [Co(0.47 nm)/Pt(0.23 nm)]3 layers and the Co2Fe6B2 bottom pinned layer is defined as the M2 layer. The bottom pinned structure of the double pinned p-MTJ spin-valve using a single SyAF [Co/Pt]n layer and double free-layer were fabricated wherein the ratio of the number of [Co/Pt]n layers between the upper and lower SyAF [Co/Pt]n layer was varied from 3:6 (i.e., a double SyAF [Co/Pt]n layer) to 0:3, as shown in Fig. 1b. In addition, a Co/Pt/Co buffer layer was used to bridge instead of the top-upper SyAF [Co(0.47 nm)/Pt(0.23 nm)]3 layer (compare Fig. 1a,b). The MgO capping layer of the conventional double MgO-based p-MTJ spin-valve structure was used as the top MgO tunneling barrier followed by an Fe insertion layer (0.3 nm), Co2Fe6B2 (0.75 nm) top pinned layer, W bridge layer (0.42 nm), Co(0.47 nm)/Pt(2 nm) buffer layer, and top-lower SyAF [Co(0.47 nm)/Pt(0.23 nm)]3 layer/ Co(0.51 nm). The ferro-coupled Co2Fe6B2 top pinned layer and the top-lower SyAF [Co(0.47 nm)/Pt(0.23 nm)]3 layer is defined as the top-pinned layer (M4 layer). Following the M4 layer is the Ru spacer layer (0.85 nm), Co(0.51 nm)/Pt(0.23 nm)/top-upper SyAF [Co(0.47 nm)/Pt(0.23 nm)]3 layer (M5 layer), and Ta/Ru top electrode. The spin-valves were ex-situ annealed at 350 °C for 30 min under a vacuum below 10−6 torr and a perpendicular magnetic-field of 3 tesla. The TMR ratios of the double pinned p-MTJ spin-valves fabricated on 12-inch Si wafers were estimated by using CIPT at room temperature. The wafers were cut into 1 × 1 cm2 pieces. The magnetic properties of the double pinned spin-valves were characterized by using vibrating-sample magnetometer (VSM) at room temperature. The R-H curve was measured with a p-MTJ spin-valve with the cell size of 2-μm × 2-μm. The 2-μm-scale p-MTJ spin-valves were wire-bonded to the sample holder and were installed into a home-made electrical probing system with a ~1 Tesla electromagnet using a Keithley 236 source measure unit and Agilent B2902A semiconductor parameter analyzer.

## References

1. 1.

Song, Y. J. et al. Highly Functional and Reliable 8Mb STT-MRAM Embedded in 28nm Logic. IEDM, 27.2.1–27.2.4, https://doi.org/10.1109/IEDM.2016.7838491 (2017).

2. 2.

Antonyan, A., Pyo, S., Jung, H. & Song, T. Embedded MRAM Macro for eFlash Replacement. IEEE International Symposium on Circuits and Systems (ISCAS), 2–5, https://doi.org/10.1109/ISCAS.2018.8351201 (2018).

3. 3.

Jan, G. et al. Demonstration of fully functional 8Mb perpendicular STT-MRAM chips with sub-5ns writing for non-volatile embedded memories. Dig. Tech. Pap. - Symp. VLSI Technol. 093008, 8–9, https://doi.org/10.1109/VLSIT.2014.6894357 (2014).

4. 4.

Lu, Y. et al. Fully functional perpendicular STT-MRAM macro embedded in 40 nm logic for energy-efficient IOT applications. IEEE Int. Electron Devices Meeting, IEDM, 26.1.1–26.1.4, https://doi.org/10.1109/IEDM.2015.7409770 (2016).

5. 5.

Aitken, R. et al. Device and technology implications of the Internet of Things. Symp. VLSI Technol. Dig. Tech. Papers, 1–4, https://doi.org/10.1109/VLSIT.2014.6894339 (2014).

6. 6.

Durlam, M. et al. MRAM Memory for Embedded and Stand Alone Systems, IEEE Int. Conf. Integr. Circuit Des. Technol. (ICICDT), 75–78, https://doi.org/10.1109/ICICDT.2007.4299546 (2007).

7. 7.

Grollier, J., Querlioz, D. & Stiles, M. D. Spintronic Nanodevices for Bioinspired Computing. Proc IEEE Inst Elestr Electron Eng. 104(10), 2024–2039, https://doi.org/10.1109/JPROC.2016.2597152 (2016).

8. 8.

Vincent, A. F. et al. Spin-Transfer Torque Magnetic Memory as a Stochastic Memristive Synapse for Neuromorphic Systems. IEEE Transactions on Biomedical Circuits and Systems 9(2), 166–174, https://doi.org/10.1109/TBCAS.2015.2414423 (2015).

9. 9.

Sengupta, A., Choday, S. H., Kim, Y. & Roy, K. Spin orbit torque based electronic neuron. Appl Phys. Lett. 106, 143701, https://doi.org/10.1063/1.4917011 (2015).

10. 10.

Ikeda, S. et al. A perpendicular-anisotropy CoFeB–MgO magnetic tunnel junction. Nature Materials 9, 721–724 (2010).

11. 11.

Tillie, L. B. et al. P-STT-MRAM Thermal stability and modeling of its temperature dependence. Int. Symp. VLSI Technol. Syst. Appl. 1–2, https://doi.org/10.1109/VLSI-TSA.2018.8403857 (2018).

12. 12.

Yakata, S. et al. Influence of perpendicular magnetic anisotropy on spin-transfer switching current in CoFeB/MgO/CoFeB magnetic tunnel junctions. J. Appl. Phys. 105, 07D131, https://doi.org/10.1063/1.3057974 (2009).

13. 13.

Park, J. G. et al. Challenging issues for terra-bit-level perpendicular STT-MRAM. IEEE Int. Electron Devices Meeting, IEDM, 19.2.1–19.2.4, https://doi.org/10.1109/IEDM.2014.7047081 (2015).

14. 14.

Lee, D. Y., Shim, T. H. & Park, J. G. Effects of Pt capping layer on perpendicular magnet anisotropy in pseudospin valves of Ta/CoFeB/MgO/CoFeB/Pt magnetic-tunneling junctions. Appl. Phys. Lett. 102, 212409, https://doi.org/10.1063/1.4808084 (2013).

15. 15.

Kent, A. D. & Worledge, D. C. A new spin on magnetic memories. Nature Nanotechnology 10(3), 187–191 (2015).

16. 16.

Wen, Z. et al. A 4-Fold-Symmetry Hexagonal Ruthenium for Magnetic Heterostructures Exhibiting Enhanced Perpendicular Magnetic Anisotropy and Tunnel Magnetoresistance. Adv. Mater. 26(37), 6483–6490, https://doi.org/10.1002/adma.201401959 (2014).

17. 17.

Honjo, H. et al. 10 nm ϕ perpendicular-anisotropy CoFeB-MgO magnetic tunnel junction with over 400 °C high thermal tolerance by boron diffusion control. Symp. VLSI Technol. Dig. Tech. Papers, 160–161, https://doi.org/10.1109/VLSIT.2015.7223661 (2015).

18. 18.

Choi, J. Y., Lee, D. G., Baek, J. U. & Park, J. G. Double MgO-based Perpendicular Magnetic-Tunnel-Junction Spin-valve Structure with a Top Co2Fe6B2 Free Layer using a Single SyAF [Co/Pt]n Layer. Sci. Rep. 8, 2139 (2018).

19. 19.

20. 20.

Peng, S. et al. Interfacial Perpendicular Magnetic Anisotropy in Sub-20 nm Tunnel Junctions for Large-Capacity Spin-Transfer Torque Magnetic Random-Access. Memory, IEEE Magn. Lett. 8, 3105805, https://doi.org/10.1109/LMAG.2017.2693961 (2017).

21. 21.

Perrissin, N. et al. A highly thermally stable sub-20 nm magnetic random-access memory based on perpendicular shape anisotropy. Nanoscale 10, 12187–12195, https://doi.org/10.1039/C8NR01365A (2018).

22. 22.

Jang, J. J. et al. Vertical Cell Array using TCAT(Terabit Cell Array Transistor) Technology for Ultra High Density NAND Flash Memory. Symp. VLSI Technol. 14–15 (2009).

23. 23.

Cai, Y. et al. Experimental Characterization, Optimization, and Recovery of Data Retention Errors in MLC NAND Flash Memory, Preprint at arXiv:1805.02819 (2018).

24. 24.

Leroux, C., Cadeville, M. C., Bohnes, V. P., Inden, G. & Hinz, F. Comparative investigation of structural and transport properties of L10 NiPt and CoPt phases; the role of magnetism. J. Phys. F Met. Phys. 18, 2033–2051, http://iopscience.iop.org/0305-4608/18/9/021 (1988).

25. 25.

Park, J. H. et al. Co∕Pt multilayer based magnetic tunnel junctions using perpendicular magnetic anisotropy. J. Appl. Phys. 103, 07A917, https://doi.org/10.1063/1.2838754 (2008).

26. 26.

Maat, S., Takano, K., Parkin, S. S. P. & Fullerton, E. E. Perpendicular Exchange Bias of Co/Pt Multilayers. Phys. Rev. Lett. 87, 87202-1–87202–4 (2001).

27. 27.

Lee, D. Y., Lee, S. E., Shim, T. H. & Park, J. G. Tunneling-Magnetoresistance Ratio Comparison of MgO-Based Perpendicular-Magnetic-Tunneling-Junction Spin Valve Between Top and Bottom Co2Fe6B2 Free Layer Structure. Nanoscale Res. Lett. 11, 433, https://doi.org/10.1186/s11671-016-1637-9 (2016).

28. 28.

Takemura, Y., Lee, D. Y., Lee, S. E. & Park, J. G. Dependency of tunneling magnetoresistance ratio on Pt seed-layer thickness for double MgO perpendicular magnetic tunneling junction spin-valves with a top Co2Fe6B2 free layer ex-situ annealed at 400 °C. Nanotechnology. 27, 485203 (2016).

29. 29.

Lee, D. Y., Hong, S. H., Lee, S. E. & Park, J. G. Dependency of Tunneling-Magnetoresistance Ratio on Nanoscale Spacer Thickness and Material for Double MgO Based Perpendicular-Magnetic-Tunneling-Junction. Sci. Rep. 6, 38125 (2016).

30. 30.

Lee, S. E., Shim, T. H. & Park, J. G. Perpendicular magnetic tunnel junction (p-MTJ) spin-valves designed with a top Co2Fe6B2 free layer and a nanoscale-thick tungsten bridging and capping layer. NPG Asia Materials 8, e324 (2016).

31. 31.

Shen, W. et al. Effect of film roughness in MgO-based magnetic tunnel junctions. App. Phys. Lett. 88(18), 182508, https://doi.org/10.1063/1.2201547 (2006).

32. 32.

Jeon, M. S. et al. The dependency of tunnel magnetoresistance ratio on nanoscale thicknesses of Co2Fe6B2 free and pinned layers for Co2Fe6B2/MgO-based perpendicular-magnetic-tunnel-junctions. Nanoscale 7(17), 8142–8148, https://doi.org/10.1039/C5NR01140J (2015).

33. 33.

Lee, D. Y., Seo, H. T. & Park, J. G. Effects of the radio-frequency sputtering power of an MgO tunneling barrier on the tunneling magneto-resistance ratio for Co2Fe6B2/MgO-based perpendicular-magnetic tunnel junctions. J.Mater. Chem. C 4, 135–141, https://doi.org/10.1039/C5TC03669K (2016).

34. 34.

Lee, S. E., Takemura, Y. & Park, J. G. Effect of double MgO tunneling barrier on thermal stability and TMR ratio for perpendicular MTJ spin-valve with tungsten layers. App. Phys. Lett. 109(18), 182405, https://doi.org/10.1063/1.4967172 (2016).

35. 35.

Chae, K. S., Shim, T. H. & Park, J. G. Dependency of anti-ferro-magnetic coupling strength on Ru spacer thickness of [Co/Pd]n-synthetic-anti-ferro-magnetic layer in perpendicular magnetic-tunnel-junctions fabricated on 12-inch TiN electrode wafer. J. Appl. Phys. 116(3), 033904, https://doi.org/10.1063/1.4887352 (2014).

36. 36.

Chae, K. S., Lee, D. Y., Shim, T. H., Hong, J. P. & Park, J. G. Correlation of the structural properties of a Pt seed layer with the perpendicular magnetic anisotropy features of full Heusler-based Co2FeAl/MgO/Co2Fe6B2 junctions via a 12-inch scale Si wafer process. Appl. Phys. Lett. 103, 162409, https://doi.org/10.1063/1.4824306 (2013).

37. 37.

Lee, D. Y., Shim, T. H. & Park, J. G. Effect of coupling ability between a synthetic antiferromagnetic layer and pinned layer on a bridging layer of Ta, Ti, and Pt in perpendicular-magnetic tunnel junctions. Nanotechnology 27(29), 295705 (2016).

38. 38.

Lee, S. E., Shim, T. H. & Park, J. G. Co2Fe6B2/MgO-based perpendicular spin-transfer-torque magnetic-tunnel-junction spin-valve without [Co/Pt]n lower synthetic-antiferromagnetic layer. Nanotechnology 26(47), 475705 (2015).

39. 39.

Yang, H. X. et al. First-principles investigation of very large perpendicular magnetic anisotropy at Fe|MgO and Co|MgO interfaces. Phys. Rev. B 84, 054401 (2011).

## Acknowledgements

This work was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIP) (No. 2017R1A2A1A05001285) and Brain Korea 21 PLUS Program in 2014.

## Author information

J.Y. Choi and J.G. Park conceived and designed the study. J.Y. Choi, H. Jun, K. Ashiba fabricated all patterns and carried out experiments, with the help of J.G. Park. Contributions to the measurements were made by J.Y. Choi, H. Jun, K. Ashiba, and J.U. Baek. All authors contributed to discussions regarding the research. J.Y. Choi, T.H. Shim and J.G. Park wrote the manuscript.

Correspondence to Jea-Gun Park.

## Ethics declarations

### Competing Interests

The authors declare no competing interests.

Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

## Rights and permissions

Reprints and Permissions