France



Rapid advancements in embedded memory technologies have opened the market to a wide range of low power computing applications, such as MCUs and Internet-of-Things (IoT). Innovative solutions, such as &lt;i&gt;E&lt;/i&gt;&lt;i&gt;dge computing&lt;/i&gt;, where data is stored and processed close to the application, have been proposed as a cost and energy-effective solution. STT-MRAM technology is an attractive candidate for edge computing memory owing to its high endurance (&amp;gt; 10&lt;sup&gt;12&lt;/sup&gt; cycles), negligible leakage and non-volatility [1-2]. To enable adoption of eMRAM in low power applications, low active power consumption at operating frequencies ≤ 50 MHz is desirable. Reduction in the write power consumption is possible by materials engineering the free and reference layers in conventional MTJ stacks to improve switching characteristics [3, 4] and RA engineering to reduce switching voltages (&lt;i&gt;V&lt;sub&gt;sw&lt;/sub&gt;&lt;/i&gt;) [3]. Unfortunately, &lt;i&gt;J&lt;sub&gt;sw&lt;/sub&gt;&lt;/i&gt; improvement has often come at the cost of a lower Δ, which is unsuitable for the targeted application.&lt;br/&gt;&lt;br/&gt;An alternative approach to minimizing &lt;i&gt;J&lt;sub&gt;sw&lt;/sub&gt;&lt;/i&gt; is to incorporate two reference layers (RLs) and MgO barriers on either side of the free layer (FL), also known as Double-MTJ (DMTJ) STT-MRAM [4] (Fig. 1a). A critical requirement is the anti-parallel alignment of the RLs which leads to a ‘doubling’ of the total spin torque. However, this approach suffers challenges in processing (two setting fields), TMR performance, RA control, stray field control, degraded retention times and back-end-of-line (BEOL) compatibility [5]. Thus, to demonstrate the feasibility of DMTJ STT-MRAM the following challenges must be addressed: Δ improvement, offset control (&lt;i&gt;μ&lt;sub&gt;o&lt;/sub&gt;&lt;/i&gt;&lt;i&gt;H&lt;sub&gt;offset&lt;/sub&gt;&lt;/i&gt; &amp;lt; 0.1*&lt;i&gt;μ&lt;sub&gt;o&lt;/sub&gt;&lt;/i&gt;&lt;i&gt;H&lt;sub&gt;c&lt;/sub&gt;&lt;/i&gt;), read/write optimization (TMR&amp;gt;100%, &lt;i&gt;J&lt;sub&gt;sw&lt;/sub&gt;&lt;/i&gt; &amp;lt; 3 MA/cm&lt;sup&gt;2&lt;/sup&gt;), and BEOL compatibility. In this work, we demonstrate device-level performance with full stray field compensation, best-in-class switching (&lt;i&gt;J&lt;sub&gt;sw&lt;/sub&gt;&lt;/i&gt;) and retention (Δ) exceeding 10 years.&lt;br/&gt;&lt;br/&gt;To achieve BEOL compatibility and compensated stray fields, a 400 °C SAF-SAF design was enabled. Both the bottom-pinned and top-pinned halves of the DMTJ stack employed a synthetic antiferromagnet (SAF) design on either side of the FL. Figure 1(a, b) illustrates the stack design alongside a blanket M-H loop. The challenges involved in the design and demonstration of a 400 °C-compatible top-pinned SAF will be discussed in this work. This includes optimization of the SAF material stack design for annealing and stray field control. The successful demonstration of the SAF-SAF design is shown in the device R-H loops of CD = 80 nm devices where &lt;i&gt;μ&lt;sub&gt;0&lt;/sub&gt;H&lt;sub&gt;offset&lt;/sub&gt;&lt;/i&gt; ~ 0 mT (Fig. 2a). Figure 2b shows the switching current density (&lt;i&gt;J&lt;sub&gt;sw&lt;/sub&gt;&lt;/i&gt;) trends from 125 devices as a function of pulse-width. An average &lt;i&gt;J&lt;sub&gt;sw&lt;/sub&gt;&lt;/i&gt; = 2.7 MA/cm&lt;sup&gt;2&lt;/sup&gt; at 50 ns is observed on devices with Δ = 82 k&lt;sub&gt;B&lt;/sub&gt;T and TMR ~ 100%. In addition, WER curves demonstrating 1ppm error-free switching, and a wide switching-breakdown margin of ~0.7 V&lt;sub&gt;DC&lt;/sub&gt; will also be presented. These performance metrics clearly exhibit the importance of the SAF-SAF design in achieving manufacturable DMTJ STT-MRAM devices, thus making it ready for implementation in low power applications.

**References:**

[1] K. Lee et. al, &lt;i&gt;IEEE IEDM&lt;/i&gt;, pp. 2.2.1-2.2.4 (2019).&lt;br/&gt;[2] J. G. Alzate et. al, &lt;i&gt;IEEE IEDM&lt;/i&gt;, pp. 2.4.1-2.4.4 (2019).&lt;br/&gt;[3] L. Thomas et. al., &lt;i&gt;IEEE IEDM&lt;/i&gt;, pp 18.612-18.615 (2018)&lt;br/&gt;[4] G. Hu et al., &lt;i&gt;IEEE IEDM&lt;/i&gt;, pp 19.38 – 19.41 (2019)&lt;br/&gt;[5] G. Hu et al., &lt;i&gt;IEEE IEDM&lt;/i&gt;, pp. 26.3.1-26.3.4 (2015)

**Images:**

![](https://assets.underline.io/uploads/markdown_image/1/image/33428b2144a71698382f51ed049cc787.jpg)
Figure 1: (a) Schematic representation of a double magnetic tunnel junction stack with the SAF-SAF design. Two magnetic field setting steps are required to enable anti-parallel alignment in the two reference layers (RL). The stack was annealed at 400 &lt;sup&gt;○&lt;/sup&gt;C prior to patterning into devices of varying sizes. (b) Corresponding M-H loop obtained from blanket film stacks by polar MOKE measurements.

![](https://assets.underline.io/uploads/markdown_image/1/image/e612ed7842f2ae8dd9b3c918b597e014.jpg)
Figure 2: (a) Normalized TMR loops of 80 nm-sized devices comparing the optimized SAF-SAF stack design to the reference DMTJ stack, elucidating full compensation of stray fields. (b) Switching current density (&lt;i&gt;J&lt;sub&gt;sw&lt;/sub&gt;&lt;/i&gt;) trends in a sample set of 125 devices (CD = 80 nm) measured as a function of pulse-width. Each device was switched 100 times in both directions.

INTERMAG 2021

 Demonstration of BEOL-compatible double magnetic tunnel junction pSTT-MRAM devices for low power computing

INTERMAG 2021 Welcome Video



	
<figure class="video_container">
 <iframe src="https://screencast-o-matic.com/player/crflqFVnUpZ?v=6&ff=1&title=0&controls=1" width=640 height=360 frameborder="0" allowfullscreen="true"> </iframe>
</figure>

Want to buy a ticket? 

Register for INTERMAG 2021 here

[![](https://assets.underline.io/uploads/markdown_image/1/image/17ce76e7960db633258024eef8d6da7f.png)](https://intermag.org/register)

Got your ticket, but still can’t access the event?

Contact us and our team will look into it.

Rapid advancements in embedded memory technologies have opened the market to a wide range of low power computing applications, such as MCUs and Internet-of-Things (IoT). Innovative solutions, such as Edge computing, where data is stored and processed close to the application, have been proposed as a cost and energy-effective solution. STT-MRAM technology is an attractive candidate for edge computing memory owing to its high endurance (&gt; 1012 cycles), negligible leakage and non-volatility [1-2]. To enable adoption of eMRAM in low power applications, low active power consumption at operating frequencies ≤ 50 MHz is desirable. Reduction in the write power consumption is possible by materials engineering the free and reference layers in conventional MTJ stacks to improve switching characteristics [3, 4] and RA engineering to reduce switching voltages (Vsw) [3]. Unfortunately, Jsw improvement has often come at the cost of a lower Δ, which is unsuitable for the targeted application. An alternative approach to minimizing Jsw is to incorporate two reference layers (RLs) and MgO barriers on either side of the free layer (FL), also known as Double-MTJ (DMTJ) STT-MRAM [4] (Fig. 1a). A critical requirement is the anti-parallel alignment of the RLs which leads to a ‘doubling’ of the total spin torque. However, this approach suffers challenges in processing (two setting fields), TMR performance, RA control, stray field control, degraded retention times and back-end-of-line (BEOL) compatibility [5]. Thus, to demonstrate the feasibility of DMTJ STT-MRAM the following challenges must be addressed: Δ improvement, offset control (μoHoffset &lt; 0.1*μoHc), read/write optimization (TMR&gt;100%, Jsw &lt; 3 MA/cm2), and BEOL compatibility. In this work, we demonstrate device-level performance with full stray field compensation, best-in-class switching (Jsw) and retention (Δ) exceeding 10 years. To achieve BEOL compatibility and compensated stray fields, a 400 °C SAF-SAF design was enabled. Both the bottom-pinned and top-pinned halves of the DMTJ stack employed a synthetic antiferromagnet (SAF) design on either side of the FL. Figure 1(a, b) illustrates the stack design alongside a blanket M-H loop. The challenges involved in the design and demonstration of a 400 °C-compatible top-pinned SAF will be discussed in this work. This includes optimization of the SAF material stack design for annealing and stray field control. The successful demonstration of the SAF-SAF design is shown in the device R-H loops of CD = 80 nm devices where μ0Hoffset ~ 0 mT (Fig. 2a). Figure 2b shows the switching current density (Jsw) trends from 125 devices as a function of pulse-width. An average Jsw = 2.7 MA/cm2 at 50 ns is observed on devices with Δ = 82 kBT and TMR ~ 100%. In addition, WER curves demonstrating 1ppm error-free switching, and a wide switching-breakdown margin of ~0.7 VDC will also be presented. These performance metrics clearly exhibit the importance of the SAF-SAF design in achieving manufacturable DMTJ STT-MRAM devices, thus making it ready for implementation in low power applications.

**References:**

[1] K. Lee et. al, IEEE IEDM, pp. 2.2.1-2.2.4 (2019). [2] J. G. Alzate et. al, IEEE IEDM, pp. 2.4.1-2.4.4 (2019). [3] L. Thomas et. al., IEEE IEDM, pp 18.612-18.615 (2018) [4] G. Hu et al., IEEE IEDM, pp 19.38 – 19.41 (2019) [5] G. Hu et al., IEEE IEDM, pp. 26.3.1-26.3.4 (2015)

**Images:**

![](https://assets.underline.io/uploads/markdown_image/1/image/33428b2144a71698382f51ed049cc787.jpg)
Figure 1: (a) Schematic representation of a double magnetic tunnel junction stack with the SAF-SAF design. Two magnetic field setting steps are required to enable anti-parallel alignment in the two reference layers (RL). The stack was annealed at 400 ○C prior to patterning into devices of varying sizes. (b) Corresponding M-H loop obtained from blanket film stacks by polar MOKE measurements.

![](https://assets.underline.io/uploads/markdown_image/1/image/e612ed7842f2ae8dd9b3c918b597e014.jpg)
Figure 2: (a) Normalized TMR loops of 80 nm-sized devices comparing the optimized SAF-SAF stack design to the reference DMTJ stack, elucidating full compensation of stray fields. (b) Switching current density (Jsw) trends in a sample set of 125 devices (CD = 80 nm) measured as a function of pulse-width. Each device was switched 100 times in both directions.

technical paper

Magneto resistive random-access memory (MRAM) holds the potential to be a universal memory with its non-volatility, high speed, robust endurance, and system power savings capability. In the semiconductor memory and storage space, MRAM is poised to deliver the performance of working memory with the persistence of storage. The pioneer generation of MRAM product dubbed Toggle MRAM was first launched in 2006 and was based on magnetic field switching. Driven by continuous demand for lower power and higher density, spin transfer torque MRAM (STT-MRAM) was introduced into the market. In its commercialization, standalone STT-MRAM product saw rapid evolution toward higher density in which the memory density quadrupled every 2 years between 2015-2019 [1]. The successful scaling of STT-MRAM product is attributed to a concerted effort in materials engineering and the use of advanced process technologies. Most recently in 2019-2020, we saw two momentous events in MRAM development: Everspin launched a milestone 1Gb product [2], and 3 major foundries and 2 other companies announced their STT-MRAM product offerings [1]. In this talk, we will review the fundamental aspects of STT-MRAM with a focus on read and write distributions of MRAM arrays. In order to achieve reliable high density STT-MRAM products it is critical to understand the contributions to these distributions and minimize the variation across the array. For instance, minimizing array write error rate (WER) relies much upon reducing the population of extrinsic magnetic tunnel junction (MTJ) bits in an array. A method was developed to quantify the population of extrinsic bits by calculating the switching failure rate upon doing N repeat of read-write sequence, which we call soft fail count. Figure 1(a) shows example of soft fail improvement by combining MTJ film engineering and process improvement. The soft fail count improvement is seen from both the reduction of outlier bits and tightening of main bits population to reach below the desired requirement. Figure 1(b) shows the median soft fail count plotted against wafer radius from center to edge. Such plot further elucidated the distinct contribution from MTJ film and process changes implemented. Detailed analysis shows that the MTJ film improvement reduces the extrinsic population and improved the switching reliability on the die level, while the process improvement extends the die uniformity across the wafer. These approaches were found to be additive when combined. A low operating voltage coupled with reliable switching (measured as distributions) is critical to integrating MTJ bits on silicon. Figure 2 shows the plots of normalized array operating voltage (Vop) and its sigma corresponding to the MTJ film and process improvement shown in Figure 1. Initial MTJ film change from A to B improved the Vop by 14% and significantly reduced the sigma by a factor 2.6x. Switching from process A to B using the improved film B improved the Vop by a further 3.3% with a comparable sigma level to process A. Finally, combining MTJ film C with Process B produced slight improvement in Vop and further reduction in sigma by 20%. This observation corroborated the reduction in bit error soft fail count shown in Figure 1. The improvement in bit-to-bit distribution is critical to enable reliable high density STT-MRAM products. In the talk, we will discuss the impact of these distributions on other key aspects such as data retention, switching pulse widths and endurance.

**References:**

References:: [1] S. Ikegawa, F. B. Mancoff, J. Janesky, and S. Aggarwal, “Magnetoresistive Random Access Memory: Present and Future”, IEEE. Trans. Elec. Dev. 67(4), 1407 (2020); doi: 10.1109/TED.2020.2965403. [2] S. Aggarwal, H. Almasi, M. DeHerrera, B. Hughes, S. Ikegawa, J. Janesky, H. K. Lee, H. Lu, F. B. Mancoff, K. Nagel, G. Shimon, J. J. Sun, T. Andre, and S. M. Alam, "Demonstration of a reliable 1 Gb standalone spin-transfer torque MRAM for industrial applications," in IEDM Tech. Dig., 2019, No. 2.1.

**Images:**

![](https://assets.underline.io/uploads/markdown_image/1/image/36c8c73ccf560c4e602b00ef4a3f958c.jpg)
Figure 1. (a) Soft fail count improvement with combinations of MTJ film and process changes, (b) median soft fail count plotted against wafer radius from center to edge showing improvement in bits distribution within die and across wafer uniformity.

![](https://assets.underline.io/uploads/markdown_image/1/image/a88c3d224319675d69acb324487122c0.jpg)
Figure 2. Improvement in operating voltage (Vop) and its sigma with combinations of MTJ film and process changes

 Reliable High Density STT-MRAM Products INVITED

STT-MRAM materials and technology have recently matured to the point of moving from “lab to fab” for 28nm-22nm node standalone and eNVM applications. Prior work has mostly placed MTJ devices above M2 or even higher in the BEOL stack, depending on the application and integration constraints [1-4]. However, optimizing MRAM for mobile or last-level cache applications requires significant improvements in write speed, write current, endurance, and density of both the MRAM cell in addition to the CMOS [5]. In this work, we present the first Embedded Spin-Transfer-Torque MRAM (eMRAM) technology in a 14 nm CMOS node. Using a 2 Mb eMRAM macro, we achieve an integration at tight MTJ pitch (0.0273 um2 cell size), which is aggressively placed between M1 and M2. We demonstrate read and write functionality, including write performance down to 4 ns, and show that the eMRAM process module can be added while obeying the existing CMOS logic BEOL design rules and reliability requirements. Several unit process innovations enabled this integration, including a novel sub-lithographic microstud (μ-stud) bottom electrode (BEL), fine profile control of the MTJ patterning and dielectric films, optimized BEL/MTJ metallization, and optimized post-MTJ low-k planarization across array and logic areas. The result preserves the minimum 3 added mask levels (2 critical for μ-stud and MTJ; 1 non-critical for MTJ alignment marks), without mask-adders for low-k planarization or top electrode (TEL). A shrink patterning process ensures that the μ-stud top CD is always smaller than the MTJ bottom CD, for MTJ sizes down to below 35 nm. This minimizes shunting by eliminating the BEL resputtering onto the MTJ sidewalls during patterning, and increases etch budget. The μ-stud BEL module re-uses current manufacturing processes. Figure 1 shows a TEM cross section of 41nm MTJ devices with a sub-lithographic μ-stud underneath. The alternating MTJ pitch presented a challenge for IBE overetch/cleanup without excessive dielectric gouge. Additional tests were completed showing good device scalability to 35nm, with magnetic property degradation at smaller sizes due to edge damage that could be improved with further etch optimization. Cache applications require writing at high speed with high cycling endurance, but data retention performance can be relaxed (e.g. using ECC and data scrubbing). Figure 2 shows the eMRAM write performance. The array was written to the 1 state, then to the 0 state, versus write pulse voltage and width. The curves show the write voltage for passing median bits vs. write pulse width, from 4 ns - 20 ns. Stack B incorporated materials optimizations to reduce write voltage and time for suitability towards the last level cache. We performed a series of retention bakes at 85°C on arrays of target and smaller sized devices, demonstrating projected fail rates well below 1 ppm for retention times of ≤1 min. at nominal 43 nm CD. Results also include endurance cycling tests on 128 kb arrays up to 5x106 cycles, and single MTJs up to 1010 cycles. No dielectric breakdown or resistance drift was observed in either case. Finally, an early BEOL reliability assessment of electromigration and time-dependent dielectric breakdown supports our eMRAM embeddability for high-performance logic applications.

**References:**

[1] S. Aggarwal, et al., IEEE IEDM, San Francisco, pp. 2.1.1-2.1.4, 2019. [2] K. Lee, et al., IEEE IEDM, San Francisco, pp. 2.1.1-2.1.4, 2019. [3] W. J. Gallagher, et al., Symposium on VLSI Technology, Kyoto, pp. T190-T191, 2019. [4] J. G. Alzate et al., IEEE IEDM, San Francisco, pp. 2.4.1-2.4.4, 2019. [5] S. Sakhare et al., IEEE IEDM, San Francisco, pp. 18.3.1-18.3.4.2018.

**Images:**

![](https://assets.underline.io/uploads/markdown_image/1/image/4ca2fc66e4cfa0a482eef090a40fd8bc.jpg)
Figure 1: TEM cross section of 41nm MTJ devices inserted between the 14nm node M1 and M2 levels. The mstud enabled sufficient overetch & cleanup during MTJ patterning.

![](https://assets.underline.io/uploads/markdown_image/1/image/005730caa67f8294cda72db1f933d05e.jpg)
Figure 2: Passing median bit write voltage vs write pulse width for both W0 and W1 operation. Stack B improvements to the free layer enabled higher speed 4ns switching.

 Embedded STT-MRAM for 14nm CMOS Node Cache Applications INVITED

CoFeB–MgO-based magnetic tunnel junctions (MTJs) with perpendicular easy axis have emerged as critical components to develop spin transfer torque magnetoresistance random access memories (STT-MRAMs) [1–6]. To develop a large capacity STT-MRAM, a high tunnel magnetoresistance (TMR) ratio, high thermal stability, and low switching current must simultaneously be realized under a thermal tolerance of 400 °C to ensure CMOS back-end-of-line compatibility. To satisfy these requirements, we have developed high performance MTJ stack structures through physical vapor deposition process development [7–9]. As the size of the MTJs decreases, a more stable reference layer is required to achieve reliable write operations. Furthermore, the magnitude of the stray magnetic field from the reference layer must be minimized. This magnitude significantly increases with a decrease in the MTJ size, resulting in an asymmetrical thermal stability factor between the parallel and antiparallel states [9]. Co/Pt-based synthetic anti-ferro (SyF) magnetically coupled layers with an Ru coupling layer have been widely used as a reference layer to reduce the stray magnetic field. Although the exchange coupling field (Hex) is maximized at a 1st-peak Ru thickness of about 0.4 nm, Hex degrades significantly after 400 °C annealing [10]. Therefore, 0.9 nm-thick Ru (2nd peak) layer is often used in SyF reference layer owing to their high thermal tolerance. Nevertheless, the SyF reference layer with 2nd peak Ru exhibits insufficient stability against the magnetic field and write voltage to realize smaller MTJs. To overcome these issues, we developed a reference layer with four SyF magnetically coupled layers (Quad-SyF). In the Quad-SyF, Hex is determined by the middle SyF layer sandwiched between the top and bottom SyF layers [11]. Moreover, a small stray magnetic field can be realized because each ferromagnetic layer in the Quad-SyF, especially the top layer, can be thinner than those in a conventional SyF. These aspects can help achieve reliable write operations with a high thermal tolerance. We prepared five types of stacks (A－E) as shown in the schematics in Fig. 1 (a) and Fig. 2(a). Stack A corresponds to the MTJ with conventional Co/Pt-based SyF reference layers having a 0.4 nm-thick Ru coupling layer (so-called “2L-1st”), and Stacks B and D correspond to those having a 0.9 nm thick Ru coupling layer (so-called “2L-2nd”). Stack C and E corresponds to an MTJ with Quad-SyF consisting of four anti-ferro magnetically coupled Co/Pt layers and three 0.9 nm-thick Ru coupling layers (so-called “4L-2nd”). All the MTJ stacks were deposited on a 300 mmΦ thermally oxidized Si wafer using a DC/RF magnetron sputtering system. Stack A, B and C were annealed at 300, 350, and 400 °C for 1 h. After annealing at 400 °C for 1 h, Stacks D and E were patterned into the MTJs with diameters ranging from 50–120 nm by using ArF immersion lithography and reactive ion etching techniques. The areal magnetic moment versus magnetic field (m–H) curves and TMR ratio were measured using a vibrating sample magnetometer and CIPT, respectively. Figure 1 (b) shows the out-of-plane major m–H curves for the MTJs after annealing at 400 °C for 1 h. The inset shows the minor m–H curves. In the case of the MTJ with 2L-1st, two magnetic reversals corresponding to the free and reference layers occurred. In the case of the MTJ with 2L-2nd, three magnetic reversals corresponding to the free layer, reference layer, and spin-flop of the SyF were observed. In the MTJ with 4L-2nd, four magnetic reversals corresponding to the free layer, reference layer, and spin-flop of the SyF were observed. Hex of the MTJ with 4L-2nd showed a larger value than those of the MTJs with 2L-1st and 2L-2nd. A distinct magnetization plateau region around zero magnetic field was observed in the 2L-2nd and 4L-2nd, whereas the plateau region was not observed in the 2L-1st because of the degradation of the anti-ferro magnetic coupling of the SyF. Figure 1 (c) shows the dependence of the TMR ratio on the annealing temperature for Stacks A, B, and C. For Stacks B and C, the TMR ratio increased as the annealing temperature increased up to 400 °C. In contrast, for Stack A, the TMR ratio degraded at an annealing temperature of more than 350 °C, as an ideal anti-ferromagnetic configuration could not be realized as mentioned above. Figure 2 shows the MTJ size dependence of the shift magnetic field (Hshift). The error bars indicate the standard deviations. Hshift of the MTJ with 4L-2nd showed a small value of less than 3 mT when the number of repetitions of the top, 2nd, 3rd and bottom Co/Pt layers set 1, 4, 1, and 2, respectively. The variation in Hshift and the MTJ size dependence of Hshift for the MTJ with 4L-2nd were smaller than those for the MTJ with 2L-2nd. The employment of Quad-SyF (4L-2nd) can achieve a large Hex, a small shift magnetic field, and high thermal tolerance of the reference layer. This work was supported by CIES’s Industrial Affiliation on STT-MRAM program, CIES Consortium, JST-OPERA, and CAO-SIP.

**References:**

[1] S. Ikeda et al., Nat. Mater. 9, 721 (2010). [2] H. Sato et al., Appl. Phys. Lett. 101, 022414 (2012). [3] H. Sato et al., IEEE Trans. Magn. 49, 4437 (2013). [4] Y. J. Song et al., IEDM 2018. [5] K. Lee et al., IEDM 2018. [6] Y.-C. Shih et al., VLSI 2018. [7] H. Sato et al., IEDM 2018. [8] H. Honjo et al., VLSI 2015. [9] H. Honjo et al., AIP Adv. 7, 055913 (2017). [10] K. Yakushiji et al., Appl. Phys. Express 8, 083003 (2015). [11] T. Suzuki et al., VLSI Tech. Dig., T188, 2005.

**Images:**

![](https://assets.underline.io/uploads/markdown_image/1/image/9eafb6f06ba0533c94a3aab5ab44bc15.jpg)
Fig. 1 (a) Schematic of stack structures of the MTJs. The nominal thickness for each layer in nanometers is represented by the numbers in the parentheses. (b) Out-of-plane major areal magnetic moment versus magnetic field (m-H) curves for the MTJ stacks. Inset is the minor m-H curves. (c) Annealing temperature dependence of TMR ratio.

![](https://assets.underline.io/uploads/markdown_image/1/image/0364a185bacf57f9522a7ed47cf5a2e8.jpg)
Fig. 2 (a) Schematic of the stack structures of the MTJs. (b) MTJ size dependence of the shift field Hshift for the MTJs with 2L-2nd and 4L-2nd reference layers.

 Perpendicular magnetic tunnel junctions with reference layer based on four anti-ferromagnetically coupled Co/Pt layers

Magnetic tunnel junctions (MTJ) with perpendicular magnetic anisotropy (PMA) [1] form the basis of the spin-transfer torque magnetic random-access memory (STT-MRAM), which is non-volatile, fast, dense, and has quasi-infinite write endurance and low power consumption. In the usual CoFeB|MgO-based MTJ, the cell diameter must be kept above ~30 nm [2], otherwise, the bit becomes thermally unstable due to insufficient PMA. Being able to increase the PMA of the storage layer opens a way to further downsize scaling. In magnetic recording media, this is typically done by introducing heavy metals like Pt or Pd [3], which increases the spin-orbit coupling parameter. However, in the context of STT-MRAM, this would increase the storage layer Gilbert damping and therefore the switching current [4]. To avoid this, multilayers exhibiting bulk PMA based on purely 3d metallic elements were developed [5,6] such as (Fe/Ni) or (Co/Ni) based multilayers. However, these structures are intrinsically complex to fabricate to get large PMA and often do not withstand the required annealing temperature ~400°C. Based on density functional theory (DFT) calculations, we propose a rather simple design of MTJ with greatly enhanced PMA up to several mJ/m2 in the form Fe(n)Co(m)Fe(n)|MgO [Fig. 1]. This design leverages the well-known interfacial PMA of Fe|MgO along with a strain-induced bulk PMA discovered within bcc Co [plot in Fig. 1]. This strain in bcc Co is induced by the underlying layer. The storage layer can be several nm thick and still exhibit perpendicular magnetization. We perform DFT calculations of the effective PMA (the sum of magnetocrystalline energy EMCA and demagnetizing energy Edd) for different Fe and Co thicknesses and compare with the case of pure Fe, as shown in Fig. 2(a). Unlike in Fe, where the effective PMA decreases with Fe thickness and becomes negative for thicknesses above 11 MLs, in the proposed structure, we see a steady increase of PMA vs. Co thickness. The tunneling magnetoresistance (TMR) estimated from the extended Julliere model is around 300%, comparable with that of the pure Fe|MgO case. After studying the electronic structure of strained bcc Co, we can explain why the large PMA emerges. Calculations based on the second-order perturbation theory [7] reveal that strain leads in bcc Co to a significant shift in the energies of dyz and dz2 minority-spin bands further away from the Fermi level. This is shown to cause an overall increase in PMA. Replacing the middle Fe layers with Co also decreases the negative demagnetizing energy [the magnetocrystalline and demagnetizing energies separately are shown in Fig. 2(b)]. The PMA enhancement can result in a possible downscaling of the cell area by 200% to 300% if an atomically sharp Fe/Co interface can be fabricated. In the case of interfacial mixing, the PMA is reduced by a few 10%. The impact is excessive only in the thinnest Fe2Co3Fe2/MgO structure, where the drop is about 70%. The thicker structures are far more robust against interdiffusion: in Fe3Co4Fe3/MgO, the drop is only 22%. In conclusion, we propose bcc Fe(3ML)Co(4ML)Fe(3ML) as a storage layer for MgO-based double-barrier MTJ with greatly enhanced perpendicular magnetic anisotropy.

**References:**

[1] B. Dieny and M. Chshiev, Rev. Mod. Phys., Vol. 89, p.025008 (2017) [2] S. Ikeda, K. Miura and H. Yamamoto, Nat. Mater., Vol. 9, p.721 (2010) [3] P. F. Carcia, A. D. Meinhaldt, and A. Suna, Appl. Phys. Lett., Vol. 47, p.178 (1985) [4] S. Mizukami, E. P. Sajitha and D. Watanabe, Appl. Phys. Lett., Vol. 96, p.152502 (2010) [5] K. Hotta, K. Nakamura and T. Akiyama, Phys. Rev. Lett., Vol. 110, p.267206 (2013) [6] K. Nakamura, Y. Ikeura and T. Akiyama, J. Appl. Phys., Vol. 117, p.17C731 (2015) [7] Y. Miura, S. Ozaki and Y. Kuwahara, J. Phys. Condens. Matter, Vol. 25, p.106005 (2013)

**Images:**

![](https://assets.underline.io/uploads/markdown_image/1/image/1267c7a4e6c05f78f2aa57129c3eaa08.jpg)
The proposed MTJ with enhanced PMA. FeCoFe trilayer is used as the storage layer. Replacing the middle Fe layers with Co provides a high positive contribution to perpendicular magnetocrystalline anisotropy, as shown in the plot.

![](https://assets.underline.io/uploads/markdown_image/1/image/9e6cc171bc4887cbfad3109a5b85de05.jpg)
(a) Effective PMA (EMCA + Edd) of the proposed Fe(n)Co(m)Fe(n)/MgO MTJ with different Fe and Co thicknesses. In Fe(n)Co(m)Fe(n)/MgO, the perpendicular anisotropy is preserved, while in pure Fe/MgO, it would turn in-plane above certain thickness (the Fe thickness is m+4 MLs). (b) The magnetocrystalline (EMCA) and demagnetizing energy (Edd) shown separately for structures with n=2. (c) Example of the proposed MTJ storage layer with enhanced PMA.

 Giant Perpendicular Magnetic Anisotropy Enhancement in MgO-Based Magnetic Tunnel Junction by Using Co/Fe Composite Layer

Over the past decades, there has been a significant development in spin-transfer-torque magneto-resistive random-access memory (STT-MRAM). The development was brought by CoFeB/MgO-based magnetic tunnel junctions (MTJs) with interfacial perpendicular magnetic anisotropy (IPMA) [1]. Because thermal stability factor (Δ) is proportional to effective PMA constant and the volume of the free layer in a coherent magnetic rotation model [2] and proportional to square root of effective PMA constant and the diameter of the free layer in a domain propagation model [3], Δ decreases with the decrease of MTJ diameter. To achieve great enough Δ value in a smaller MTJ, increasing the number of CoFeB/MgO interfaces has been proposed to increase PMA [4-6]. In double interface MTJ with MgO/CoFeB/insertion layer/CoFeB/MgO free layer, thin non-magnetic layer such as Ta, W or Mo was used as the insertion layer to obtain higher PMA and tunnel magnetoresistance (TMR) ratio [4,7]. When the insertion layer is too thick, the two CoFeB layers in the free layer behave magnetically independent, resulting in no benefits on Δ despite the double interface because effective free-layer volume does not increase. To obtain a benefit of double CoFeB / MgO interfaces on Δ, a great enough magnetic coupling between the two CoFeB layers is required. It is important to clarify the effect of magnetic coupling (the energy constant of unit area: Jcpl) on Δ. However, it is not clear how the magnitude of Jcpl affects the increase in Δ, although we achieved a strong Jcpl in the double interface MTJs [8]. In this study, we propose a method to calculate critical Jcpl value to maximize Δ for free layers of MgO/CoFeB/insertion layer/CoFeB/MgO in the double interface MTJs with various MTJ diameters. We introduced magnetic coupling energy between two magnetic layers in the free layer based on magnetic domain propagation model (Model A) and magnetization coherent rotation model (Model B) as shown in Fig.1. When d was larger than the critical diameter dc (= 16/π [As /Keff ]1/2 ), Model A was applied, in which a blue-colored domain with downward magnetization nucleates from an edge and propagates in each FL1 and FL2. The domain nucleation and propagation behavior from edge region assumed in this model is consistent with the results of micromagnetic simulations of the MTJs with CoFeB / MgO interfaces [9,10]. The nucleated domain area Ω1 in FL1 and Ω2 in FL2 were decided by angles φ1 and φ2. A magnetic energy E (φ1, φ2) in Model A was calculated as follows. E (φ1, φ2) = Λ(φ1) t1 σw1 + Λ(φ2) t2 σw2 + 2Jcpl |Ω(φ1) - Ω(φ2)|---(1) σw1 = 4(As Keff1)1/2, σw2 = 4(As Keff2)1/2 ----(2) Λ(φi) = d (π/2–φi) tan φi, (i = 1, 2)---(3) Ω(φi) = d2 φi/4 + d2/4 { (π/2 – φi）tan2φi – tanφi }, (i = 1, 2) ---(4) The 1st and 2nd terms in the equation (1) express magnetic domain wall energies of FL1 and FL2, respectively. The third term expresses magnetic coupling energy between FL1 and FL2. Λ is wall length and σw is wall energy per unit area. As is stiffness constant of the free layer. When d is smaller than the critical diameter dc, coherent rotation model (Model B) was applied, in which E (φ1, φ2) was calculated as follows. The 1st and 2nd terms in the equation (5) express PMA energies of FL1 and FL2. The third term expresses magnetic coupling energy between FL1 and FL2. E (φ1, φ2) =πd2/4 { Keff1t1 (1–cos2φ1) + Keff2t2 (1–cos2φ2)+ Jcpl (1–cos(φ1–φ2）} ---(5) Fig. 2 shows Jcpl dependence on Δ for the MTJs with various diameters. We used Keff1=Keff2 = Keff_film +(1-Nz-x) Ms2/(2μ0). Keff_film = 2.25x105 J/m3, t1=t2=1.2nm, Ms = 0.97 T and As = 5.4 pJ/m, which were experimental values in the double interface MTJ [8]. The calculated critical diameter dc was 23 nm. With increasing Jcpl, Δ increases and saturates over a critical Jcpl value (Jcpl_c). The saturated Δ is twice as great as that of Jcpl = 0. When Jcpl ≤ Jcpl_c, energy barrier is minimized in the paths where the magnetizations of FL1 (MFL1) and FL2 (MFL2) separately reverse (not shown in Figures). On the other hand, when the Jcpl ≥ Jcpl_c, energy barrier is minimized in the path where MFL1 and MFL1 reverse simultaneously, resulting in maximum Δ. Jcpl_c value rapidly increases with the decrease of the MTJ diameter in Model A (d ≥ dc), and gradually increases in Model B (d ≤ dc). Jcpl_c value, which is equivalent with Keff1t1 ( =Keff_filmt1 +(1-Nz-x) Ms2/(2μ0)t1) when d ≤ dc, increases with the decrease of the demagnetic field coefficient Nz-x. We have reported that Jcpl of the blanket film of the double interface MTJ with a W insertion layer thickness of 0.2 nm reaches 0.44 mJ/m2 [8]. Therefore, all the Jcpl_c values obtained for various diameters based on these models are experimentally achievable. Consequently, we successfully established a method to predict critical Jcpl values to maximize Δ for double interface MTJ elements. This work is supported by CIES’s Industrial Affiliation on STT-MRAM program, CIES Consortium, JST-OPERA, and CAO-SIP.

**References:**

[1] S. Ikeda et al., Nat. Mater. 9, 721 (2010). [2] M. P. Sharrock, IEEE Trans. Magn. 26, 193 (1990). [3] G. D. Chaves-O’Flynn et al. Phys. Rev. Applied 4, 024010 (2015). [4] H. Sato et al. Appl. Phys. Lett. 101, 022414 (2012). [5] K. Nishioka et al., IEEE Trans. Electron Devices 67, 995 (2020). [6] S. Miura et al., IEEE Trans. Electron Devices 67, 5368 (2020). [7] J. H. Kim et al., Sci. Rep. 5, 16903 (2015). [8] K. Nishioka et al., MMM 2020 conference, O2-06. [9] I. Volvach et al., MMM 2020 conference, O2-07. [10] I. Volvach et al., Appl. Phys. Lett. 116, 192408 (2020).

**Images:**

![](https://assets.underline.io/uploads/markdown_image/1/image/5ab35ffac43499d4ee18b9ab65311340.jpg)
Fig.1 Models to calculate magnetic energy surfaces of the free layer with double-interface MTJs. Model A is a magnetic domain propagation model and Model B is a magnetic coherent rotation model, introducing magnetic coupling energy between two FLs. Keff1t1 and Keff2t2 are effective perpendicular magnetic anisotropy const per unit area which MgO layers give rise to FL1 and FL2, respectively.

![](https://assets.underline.io/uploads/markdown_image/1/image/862f0cc74254d646eefaa20472e314b2.jpg)
Fig.2 Jcpl dependence on Δ for MTJs with various diameters. Jcpl_c is a Jcpl value which saturates Δ for a MTJ diameter d. (a) Model A is applied when d ≥ dc = 23 nm. (b) Model B is applied when d < dc = 23nm.

 Effect of magnetic coupling between two CoFeB layers on thermal stability in perpendicular magnetic tunnel junctions with MgO/CoFeB/insertion layer/CoFeB/MgO free layer

Spin transfer-torque Magnetic Random Access Memory (STT-MRAM) is expected to play a major role in the future development of information and communication technologies. One of the drivers of the technology is the non-volatile character of this type of memory that allows to reduce the energy consumption compared to SRAM or DRAM. [1] Globally, we are seeking to develop technologies that help reduce carbon emissions. However, the nature of the materials involved in such technologies is also a concern both in terms of environmental impact and supply risk. [2,3] MRAM contains several materials classified as critical raw materials, in particular by the European Union (EU), being of high importance to the EU economy and whose supply is classified of high risk. [4] In order to improve the stability of the reference layer whose magnetic configuration has to remain fixed during the lifetime of the memory, a synthetic antiferromagnet (SAF) is typically used. This SAF structure is normally based on Co/Pt or Co/Pd multilayers antiferromagnetically coupled through a thin Ru layer. Platinum Group Metals (PGMs) have been classified as critical materials with high supply risk. In addition, obtaining these metals in their refined form requires extensive amounts of energy and CO2 emissions due to the low initial ore concentration. In this work, an assessment of the impact of the substitution of the Pt required in STT-MRAM for the SAF layer has been made. Alternative solutions can be envisaged such as the use of multilayers based on Co/Ni exhibiting perpendicular anisotropy [5]. A second more challenging approach based on the use of perpendicular shape anisotropy (PSA) does not rely on the use of PGMs. However, this PSA based approach is more challenging from technological point of view due to the requirement of a strict control of the cell diameter at sub-20nm dimensions during the etching process. The possibility of using Co/Ni multilayers for the SAF of MRAM has been already demonstrated [6,7]. We have observed how similar properties to Co/Pt SAF can be obtained by using Co/Ni multilayers (Figure 1). Our study has the added advantage of using a seed layer free of Hf, used in some previous studies to promote the required (111) texture for the Co/Ni multilayers growth [6,7]. Hf is also a material classified as critical and whose production is strongly concentrated in two countries, France and USA. In our study a Cu seed layer is used allowing for more diverse production sources. An evaluation of the environmental (energy and global warming potential) and economic impact of the substitution has been performed. Strong improvements in both economic and environmental aspects can be achieved by the substitution. However due to the tiny amounts of PGM materials used in STT-MRAM, in comparison to the mass of the silicon wafer itself, the environmental impact and price of the silicon wafer on which these memory devices are grown is much higher as shown in Figure 2. Our conclusion is that more attention should be focused on the process of fabrication of silicon wafers or find a substitution to this type of wafers whenever possible. Nevertheless, the high supply concentration of Pt as four mining companies represent the 80% of the global production [8] makes the presence of these alternatives a key aspect to ensure the survival of the technology in case of supply disruptions.

**References:**

1. B. Dieny and I. L. Prejbeanu, ‘Magnetic Random-Access Memory’, in Introduction to Magnetic Random-Access Memory, John Wiley & Sons, Ltd, pp. 101–164. (2017) doi:10.1002/9781119079415 2. European Commission. Critical materials for strategic technologies and sectors in the EU - a foresight study. (2020) doi: 10.2873/58081 3. Ku, A. Y. Sustainable Materials and Technologies, 15, 27-32. (2018) doi:10.1016/j.susmat.2017.10.001 4. European Commission. Study on the Review of the List of Critical Raw Materials - Critical Raw Materials Factsheets European Commission. (2017) doi: 10.2873/398823 5. Nuss P, Eckelman MJ. PLoS ONE 9(7): e101298. (2014) doi: 10.1371/journal.pone.0101298 6. Kar, G. S., Kim, W., Tahmasebi, T. et al., IEEE International Electron Devices Meeting. (2014) doi:10.1109/iedm.2014.7047080 7. Tomczak, Y., Lin, T., Swerts, J., et al. IEEE Transactions on Magnetics, 52(7), 1-4. (2016) doi:10.1109/tmag.2016.2515109 8. Sinsalo, Pia, and Mari Lundström. Metals, vol. 8, no. 4, p. 203. (2018) doi:10.3390/met8040203.

**Images:**

![](https://assets.underline.io/uploads/markdown_image/1/image/c8b57764016ecc1627245e1d0345fc1e.jpg)
Fig 1. VSM measurements on a MTJ with conventional SAF based on Co/Pt multilayers (red) versus a MTJ with SAF comprising Co/Ni multilayers (green) after annealing at 300°C for 10 minutes. Sample based on Co/Pt SAF has the following composition (thickness in nm) : Ta3/Pt25/[Co0.5/Pt0.25]6/Co0.5/Ru0.9/[Co0.5/Pt0.25]3/ Co0.5/Ta0.2/FeCoB1.2/MgO0.7/Ox/MgO0.5/FeCoB1.5/Ta3/Pt5. Sample based on (Co/Ni) multilayers is Pt-free and has the following composition Ta22/FeCoB0.8/Ta3/Cu3/[Co0.2/Ni0.4]6/Co0.5/Ru0.9/Co0.5/[Co0.2/Ni0.4]3/Ta0.2/FeCoB1.2/MgO0.7/Ox/MgO0.5/FeCoB1.5/Ta3/Al2.

![](https://assets.underline.io/uploads/markdown_image/1/image/471f429389b4483ff4e1a167fcaeea84.jpg)
Fig 2. Energy per 300mm wafer, global warming potential (GWP) and price in log scale required to obtain the amount of Pt comprised in a conventional SAF, of Co in the PSA based substitution, of Ni in the case of (Co/Ni) based SAF alternative and for the production of a 300mm bare Si wafer.

 Towards Spintronic Memory without Platinum Group Metals (PGMs) for Improved Sustainability

Due to the reduced switching current and easier scalability to smaller technology nodes than its in plane magnetized counterpart, Perpendicular Spin transfer-torque Magnetic Random Access Memory (p-STT-MRAM) had a quick industrial development in the last few years. The thermal energy barrier required to meet industrial data retention requirements is provided by a large interfacial anisotropy. For the free layer, this large perpendicular magnetic anisotropy (PMA) is induced at the interface of the ferromagnetic electrode and the oxide barrier. [1] The reference layer must have a thermal stability well above that of the free layer so that the free layer can change its magnetic configuration while the reference layer remains stable. Nowadays, this is typically achieved by coupling a synthetic antiferromagnet structure (SAF) as reference layer. The high thermal stability of the SAF comes from the interfacial magnetic anisotropy provided by magnetic multilayers such as Co/Pt or Co/Pd and the antiferromagnetic RKKY interaction between the two oppositely magnetized components of the SAF. However, the use of critical metals such as Pt is often undesired due to the high price, scarcity or the high supply risk associated to its highly concentrated production. [2,3] Therefore, alternatives based on more common materials such as Ni could help mitigate possible disruptions that might appear on more precious metals and reduce the vulnerability of STT-MRAM to such disruptions. A key factor to achieve a high perpendicular magnetic anisotropy (PMA) in Co/Ni multilayers is to induce a good (111) texture [4,5]. In this study, a smooth buffer layer has been developed first wherein reduced roughness was obtained thanks to the insertion of an amorphous FeCoB layer. Then a Cu seed layer has been used to induce the desired (111) texture with the added advantage of Cu being a very common non-critical metal. A low roughness value of 0.25nm could be obtained for a 5nm Cu seed, sufficiently thick to provide the required (111) texture. The coercive field of the Co/Ni SAF increases with the thickness of the Cu seed layer correlatively with the improved (111) texture, in agreement with previous results on Co/Ni multilayers [5] and SAF [6]. However, further increasing the Cu thickness yields a roughness increase at the tunnel barrier that can be detrimental to the optimal properties of the MTJ. Therefore, a tradeoff must be found on the Cu buffer thickness. Thin film measurements of the coercive field of the MTJ with Co/Ni SAF annealed at 300°C for 10 minutes show a switching field of 3.8 kOe, sufficiently high to meet the typical reference layer requirements. In nanopatterned devices, the reference layer is stable in both parallel and antiparallel magnetic configuration. However, during annealing at 400°C, the coercive field of the reference layer drastically decreases. This decrease of reference layer coercive field has already been observed when annealing Co/Ni Ultrathin SAF p-MTJ at high temperatures [7]. Increasing the Cu seed layer thickness, thus improving the (111) texture increases the switching field of the reference layer (Fig.1a) but can be detrimental as previously mention due to an increased roughness at the barrier level. In this study, we show that increasing the thickness of the Ta layer, used as a boron getter, separating the top part of the Co/Ni SAF from the FeCoB reference layer allows to keep higher coercive field of the (Co/Ni) SAF layer even after annealing at 400°C (Fig.1b). However, if this Ta layer is too thick (as in Fig.1b), the reference and SAF become insufficiently coupled. To circumvent this problem, we proposed here to replace the single Ta layer by a double Ta layer of composition (Ta 0.3nm/Co 1.8nm/Ta0.45nm). Such composite layer ensures a strong magnetic coupling while providing the benefit of a thicker diffusion barrier. As a result, a higher switching field of the reference layer of 5kOe after annealing at 400°C could be obtained (Figure 2) for an improved stability of the reference layer.

**References:**

1. Dieny, B., Prejbeanu, I.L., Garello, K. et al. Nat Electron 3, 446–459 (2020). doi : 10.1038/s41928-020-0461-5 2. European Commission. Study on the Review of the List of Critical Raw Materials - Critical Raw Materials Factsheets European Commission. (2017) doi: 10.2873/398823 3. Ku, A. Y. Sustainable Materials and Technologies, 15, 27-32. (2018) doi:10.1016/j.susmat.2017.10.001 4. Enlong Liu, J. Swerts, T. Devolder et al. Journal of Applied Physics 121, 043905 (2017); doi : 10.1063/1.4974885 5. Justin M. Shaw, Hans T. Nembach, and T. J. Silva. Journal of Applied Physics 108, 093922 (2010); doi : 10.1063/1.3506688 6. Arora, M., Lee-Hone, N. R., Mckinnon, T. et al. J. Phys. D: Appl. Phys. 50 505003 (11pp). (2017) doi:10.1088/1361-6463/aa97fa 7. Tomczak, Y., Lin, T., Swerts, J. et al. IEEE Transactions on Magnetics, 52(7), 1-4. (2016) doi:10.1109/tmag.2016.2515109

**Images:**

![](https://assets.underline.io/uploads/markdown_image/1/image/ccef82b6b35bb013346bc9476ba5e5ab.jpg)
Figure 1. Coercive field (Hc) of the SAF structure increases with both Cu seed layer thickness and Ta getter thickness after annealing at 400°C for 10 minutes (a). MOKE results display the Hc dependence on the Ta getter thickness shown in (a) for a fixed 5nm Cu seed layer thickness (b). The full structure in nm: Ta 22/FeCoB 0.8/Ta 3/ Cu x /6x [Co 0.18/ Ni 0.55] /Co 0.5 / Ru 0.9 / Co 0.5 / 3x [Ni 0.55/ Co 0.18] / Ta Y / FeCoB 1.2 / MgO 1.2/ FeCoB 0.4 / Ta 3 /Pt 5.

![](https://assets.underline.io/uploads/markdown_image/1/image/e0640dbefefc13d33d46ca1c1340fc75.jpg)
Figure 2. Vibrating sample magnetometer (VSM) measurement of a full MTJ with double Ta composite B-getter layer (Ta 0.3nm/Co 1.8nm/Ta0.45nm) displaying the SAF switching at field as high as 5kOe. The full structure is in nm: Ta 22/FeCoB 0.8/Ta 3/ Cu 5 /6x [Co 0.18/ Ni 0.55] /Co 0.5 / Ru 0.9 / Co 0.5 / 3x [Ni 0.55/ Co 0.18] / double Ta composite B-getter layer /FeCoB 1.2 / MgO 1.2/ FeCoB 1 / Ta 0.2 /FeCoB 0.8/ MgO 1.2/ FeCoB 0.4/Ta 3 / Pt 5.

 SAF based on Co/Ni multilayers with improved annealing tolerance for sustainable Pt-free reference layer of STT-MRAM

As core device of spin-transfer torque magnetic random-access memory (STT-MRAM) and some other spintronics devices [1], the magnetic tunnel junction (MTJ) has been widely studied to improve performances such as endurance, scalability, read reliability and power consumption. The interfacial perpendicular magnetic anisotropy (PMA) in the CoFeB/MgO/CoFeB structure with a tunnel magnetoresistance (TMR) ratio of over 120% [2] proves that PMA can be used to build high-performance out-of-plane MTJ for STT-MRAM cells. Latter research found that the heavy metal (HM)/ferromagnet interface made great contribution to PMA [3]. First-principles investigations agree that both the MgO/CoFe interface and the CoFe/HM interface attribute to magnetic anisotropy energy (MAE) in the MgO/CoFe/HM structures [4, 5]. Moreover, the MgO/CoFeB/Ta/CoFeB/MgO double-interface structure was demonstrated to improve thermal stability by a factor of 1.9 while keeping comparable critical switching current with the single-interface structure Ta/CoFeB/MgO [6]. The MgO/CoFeB/W/CoFeB/MgO structure was demonstrated to possess strong PMA and large TMR ratio (249%) after annealing at a high temperature [7]. To investigate the MAE in the W-based structure, it is necessary to research the interaction within CoFeB/W/CoFeB layers on MAE. CoFeB/W/CoFeB sandwich may form ferromagnetic or antiferromagnetic coupling (FMC or AFMC), which is expressed as interlayer exchange coupling (IEC). However, it remains unclear about the interactions between MAE and IEC in the perpendicularly magnetized CoFeB/W/CoFeB structure, making the way to enhance PMA in double-interface structure not straight forward. In this work, we first investigated the MAE and IEC of the MgO/CoFe/W/CoFe/MgO structures as a function of W thickness with first-principles calculations as shown in Fig. 1 [8]. Clear oscillation of IEC can be observed in Fig. 1(a). The long and short period of IEC oscillations are 4.8 monolayers (MLs) and 3.2 MLs, respectively, according to simulation. Figure 1(b) shows the MAEs of MgO/CoFe/W/CoFe/MgO structures for FMC and for the arrangement corresponding to IEC are calculated for all thicknesses of W layers. Moreover, for the structure with W thickness of 7 MLs, a strong FMC as well as a strong PMA can be obtained simultaneously, which is promising for high-density STT-MRAM applications with strong thermal stability. Also, we observe a clear oscillation of MAE in ferromagnetically coupled MgO/CoFe/W/CoFe/MgO structure related to the thickness of W layer. To further study the effect of interface interactions on MAE, we calculate the k-space-resolved MAE according to force theorem [8]. Figure 1(c-f) shows the k-resolved MAEs of the ferromagnetic coupled CoFe/W/CoFe structures with the thickness of W layer varies from 1 ML to 4 MLs. The spin-orbit coupling (SOC) breaks the symmetry of 2D-Brillouin-zone slightly making MAE at the points around [kx,ky]= [0.457, 0.257] possessing the largest PMA in Fig. 2(c-f). These points that make huge contributions to MAE ferromagnetically coupled are called critical points. The electron states near the Fermi surface have a large influence on MAE. Therefore, we consider spin-resolved bands at critical points. The bands near Fermi energy for spin-up and spin-down states are different, which can be explained by exchange splitting. To explore the properties of the specific electron states, the band-decomposed charge densities of these specific electron states near Fermi energy are shown in Fig. 2(a-d). We can see that only spin-up electron states occupy W layers and that the alternating oscillation of the charge densities at center of W layers: antinode at 1 ML and 3 MLs W layer, while node at 2 MLs and 4 MLs W layers. It is reported that the electron wavefunction confined in thin film is modulated by an envelope function in reference [8]. For a given energy, as the thickness of the W layer in CoFe/W/CoFe structure varies, the center of film layers within the quantum well shows the alternation of node and antinode, shown in Fig. 2(a-d), indicating the presence of QWSs in W layers. The oscillation of IEC in CoFe/W/CoFe originates from QWSs [8]. Thus, these results clearly show that the QWSs correlate the oscillation of IEC and MAE in W-based double-interface structures. In summary, we investigated the MAE and IEC of the MgO/CoFe/W/CoFe/MgO structures as a function of W thickness with first-principles calculations. The oscillations of both IEC and MAE with the change of W thickness were observed in this structure. QWSs in the W layer are demonstrated to be critical for the oscillations of MAE, leading to significant enhancement of PMA by properly tuning W thickness.

**References:**

[1] S. Peng et al., Nature Electronics, vol. 3, no. 12, pp. 757-764, 2020/12/01 2020, doi: 10.1038/s41928-020-00504-6. [2] S. Ikeda et al., Nature materials, vol. 9, no. 9, p. 721, 2010. [3] S. Peng et al., Advanced Electronic Materials, vol. 5, no. 8, p. 1900134, 2019. [4] S. Peng et al., Appl. Phys. Lett., vol. 110, no. 7, p. 072403, 2017. [5] C.-H. Chang, K.-P. Dou, G.-Y. Guo, and C.-C. Kaun, NPG Asia Materials, vol. 9, no. 8, p. e424, 2017. [6] H. Sato, M. Yamanouchi, S. Ikeda, and H. Ohno, IEEE Trans. Magn., vol. 49, no. 7, pp. 4437-4440, 2013. [7] M. Wang et al., Nature communications, vol. 9, no. 1, p. 671, 2018. [8] J. Chen et al., J. Phys. D: Appl. Phys., 2020.

**Images:**

![](https://assets.underline.io/uploads/markdown_image/1/image/128137b585eeffa691bd5dc9bde797c5.jpg)
Fig. 1. (a) Interlayer exchange coupling (IEC) of the CoFe/W/CoFe structures as a function of the thickness of W monolayers (MLs). (b) Calculated magnetic anisotropy energy (MAE) of the MgO/CoFe/W/CoFe/MgO structures as a function of the thickness of W MLs for the ferromagnetic coupling (FMC) and for the arrangement corresponding to calculated IEC. The k-resolved MAE of (c)CoFe/W 1ML/CoFe, (d)CoFe/W 2MLs/CoFe, (e)CoFe/W 3MLs/CoFe, (f)CoFe-/W 4MLs/CoFe structures in 2D-Brillouin-zone. Critical k points are labeled by black rectangles.

![](https://assets.underline.io/uploads/markdown_image/1/image/9ab5158b814c60523c1152f11e0ead35.jpg)
Fig. 2. Charge densities of bands near Fermi energy of (a)CoFe/W 1ML/CoFe, (b)CoFe/W 2MLs/CoFe, (c)CoFe/W 3MLs/CoFe, (d)CoFe/W 4MLs/CoFe. The charge density of spin-up electrons of these bands at the center of structures switch within node and antinode as a function of the number of W layers.

 Correlation of interfacial perpendicular magnetic anisotropy and interlayer exchange coupling in CoFe/W/CoFe structures

The semiconductor industry is actively developing high-density, non-volatile random access memories based on perpendicular magnetic tunnel junction (pMTJ) nanopillars for consumer electronics and thus operation between room temperature and ~150 °C [1]. Such devices are also of interest for cryogenic computing systems (4 K) where a persistent, fast, low-energy consuming, and nanometer scale device is needed. It is known that thermal fluctuations play an important role in spin torque switching of magnetic tunnel junctions causing magnetization fluctuations that decrease the switching voltage but also introduce switching errors. Here we explore the effect of temperature on the spin-transfer torque switching probability of state-of-the-art perpendicular magnetic tunnel junction nanopillars (40 to 60 nm in diameter) from room temperature down to 4 K, sampling up to a million switching events (see Fig. 1). We measure both the read disturb rate (as shown in Fig. 1) and the switching voltage versus pulse duration to determine the thermal stability factor Δ, the energy barrier to magnetization reversal Eb divided by the thermal energy kBT [2]. For both methods, the junction temperature at the switching voltage—obtained from the thermally assisted spin torque switching model—is found to saturate at temperatures below about 75 K (see Fig. 2). This shows that junction heating is significant below 75 K and that spin-transfer torque switching remains stochastic down to 4 K. This will be discussed along with three-dimensional finite-element simulations (COMSOL) which capture aspects of this behavior and show that this effect is associated with the reduced thermal conductivity and heat capacity of the metals in the junction. These results are important to furthering the understanding the role of temperature in spin torque switching dynamics of pMTJs and their applications as cryogenic memory. Acknowledgements This research was supported by Spin Memory Inc.

**References:**

[1] A. D. Kent and D. C. Worledge, Nature Nanotechnology 10, 187 (2015). [2] R. Heindl et al., Journal of Applied Physics 109, 073910 (2011). [3] G. D. Chaves-O'Flynn et al., Physical Review Applied 4, 024010 (2015). [4] L. Rehm et al., Appl. Phys. Lett. 115, 182404 (2019). [5] L. Rehm et al., arXiv:2009.01743.

**Images:**

![](https://assets.underline.io/uploads/markdown_image/1/image/6871e17ad509985da344e8384bada071.jpg)
Fig. 1: Switching probability (Psw) in the low-voltage (read disturb, RDR) limit as a function of voltage pulse amplitudes of a 40 nm diameter pMTJ at various bath temperatures and zero field. The pulse duration for this measurement was fixed at 10 µs. The straight lines represent fits to the data. The empty data points at 106 events stand for no errors and were therefore excluded from the fit. The dashed lines indicate the behavior expected for a device temperature of 4 K.

![](https://assets.underline.io/uploads/markdown_image/1/image/11f21c1456388cc425f73b4fee274b20.jpg)
Fig 2: Effective sample temperature Teff from the pulsed voltage ramps and RDR measurements at various temperatures between 4 and 295 K. The dotted black lines represent the expected behavior for a thermally assisted spin torque switching model based on domain wall (DW) reversal [3] with material parameters extracted from VSM and FMR measurements.

 Spin torque switching of perpendicular magnetic tunnel junction nanopillars at cryogenic temperatures

Perpendicular magnetic tunnel junctions, pMTJs, with a composite free layer have been developed and utilized in embedded memories and magnetic sensors. Such applications rely on fast current-induced switching of the free layer. Investigating the switching dynamics of the pMTJ free layer allows researchers to predict and optimize their switching speed. Experimental and micromagnetic simulation studies have been conducted toward this goal [1-3]. However, there continue to be challenges to overcome that are significant for technological applications. One major challenge in prediction and understating the spin transfer torque switching of the free layer is that its magnetic properties, such as magnetic anisotropy and magnetic exchange interaction, depends strongly on the details of the fabrication process and the materials involved. Recent studies suggest that the composite free layer, where a non-magnetic interlayer is inserted to enhance the perpendicular magnetic anisotropy and thermal stability of the free layer, the magnetic exchange constant is significantly lower than the bulk values [1]. The question is how does the reduced exchange interaction affect the spin torque switching dynamics and the switching speed of the free layer? Here we have investigated the effect of reduced exchange interaction on the spin torque switching dynamics and switching speed of the free layer of disk-shaped pMTJs. Our results show that when undergoing spin-transfer torque switching, the exchange interaction plays an important role in determining the switching path[4]: 1) the sequence of the micromagnetic events depends on the exchange interaction strength (see figure 1), and 2) the switching speed is greatly reduced when for lower exchange constants (see figure 2). Figure 1: A 30-nm pMTJ free layer undergoes a spin-torque switching when a long duration current pulse is applied to it (current is 1.6 times the critical current for switching). Material parameters are those mentioned in reference [1]. When exchange interaction is large (A=19 pJ/m), the switching is domain wall mediated [4]. For the same size element, when the exchange constant is smaller (A=8.5 pJ/m), the center of the element switches first, and this switched region moves to and hits the edge of the disk, and a domain wall is created across the element. By further reduced the exchange interaction (A=4 pJ/m), a droplet is formed at the center of the disk (fully reversed cells at the center), then this droplet hits the edge, and one or more domain walls form across the element. If the diameter of the disk is below a critical value [5], dc=16/π √(A/Keff ) where A is the exchange constant, and Keff is the uniaxial anisotropy that includes demagnetization energy and uniaxial perpendicular anisotropy, spin-torque switching of the free layer is expected to be a uniform macrospin model like switching. When the diameter of the disc is greater than dc, one or more domain walls (where the magnetization on one side of the wall is reversed and the other side is in the initial magnetic state) can form across the element, and dynamics are more complex. The structure of this domain wall and the magnetization profile of the free layer is very complex and gets even more complex for elements with smaller exchange interaction. In this presentation, I will present our results on how the reduced exchange interaction results in different (and more complex) dynamics that lead to a delayed switching [6]. I will also show that increasing the current does not necessarily make switching more coherent. ACKNOWLEDGEMENTS This research supported by Spin Memory Inc.

**References:**

 1. J. B. Mohammadi, B. Kardasz and A. D. Kent, ACS Applied Electronic Materials 1, 2025 (2019) 2. C. Hahn, G. Wolf and A. D. Kent, Phys. Rev. B94, 214432(2016) 3. Devolder, A. Le Goff and V. Nikitin, Phys. Rev. B93, 224432 (2016) 4. J. Beik Mohammadi and A. D. Kent, arXiv:2003.13875, 2020 (2020) 5. G. D. Chaves-O’Flynn, G. Wolf and A. D. Kent, Phys. Rev. Applied4, 024010 (2015) 6. N. Statuto, J. B. Mohammadi and A. D. Kent, Phys. Rev. B 103, 014409 (2021)

**Images:**

![](https://assets.underline.io/uploads/markdown_image/1/image/f68c74c5ec5e7db532f030fdd1e876ea.jpg)
Figure 1: A 30-nm pMTJ free layer undergoes a spin-torque switching when a long duration current pulse is applied to it (current is 1.6 times the critical current for switching). Material parameters are those mentioned in reference [1]. When exchange interaction is large (A=19 pJ/m), the switching is domain wall mediated [4]. For the same size element, when the exchange constant is smaller (A=8.5 pJ/m), the center of the element switches first, and this switched region moves to and hits the edge of the disk, and a domain wall is created across the element. By further reduced the exchange interaction (A=4 pJ/m), a droplet is formed at the center of the disk (fully reversed cells at the center), then this droplet hits the edge, and one or more domain walls form across the element.

![](https://assets.underline.io/uploads/markdown_image/1/image/82849609136ca71b06e87913fb1a3f58.jpg)
Figure 2: Switching speed of a 30-nm pMTJ free layer vs. overdrive (that is applied voltage divided by the critical/minimum voltage required to switch the free layer). Different color symbols represent the simulated data for different exchange constant, and the solid lines are linear fit to each data set. As seen in this figure, the switching speed is greatly reduced as the exchange interaction is reduced. One interesting result is that although the switching sped vs. overdrive for each exchange constant value is fairly linear, as macrospin model suggests, the slope of the line is greatly reduced for elements with reduced exchange interaction (For example, compare the slope of the red line, that is for A=8.5 pJ/m, with the slope of the purple line, that represents the data with A=2 pJ/m).

Demonstration of BEOL-compatible double magnetic tunnel junction pSTT-MRAM devices for low power computing

Downloads

Next from INTERMAG 2021

Reliable High Density STT-MRAM Products INVITED

Similar lecture

A Novel Winding Switching Strategy of a Consequent-pole Ferrite-PM Hybrid-excited Machine for Electric Vehicle Application

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES

Demonstration of BEOL-compatible double magnetic tunnel junction pSTT-MRAM devices for low power computing

.css-70qvj9{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;}Downloads

Next from INTERMAG 2021

Reliable High Density STT-MRAM Products INVITED

Similar lecture

A Novel Winding Switching Strategy of a Consequent-pole Ferrite-PM Hybrid-excited Machine for Electric Vehicle Application

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES

Downloads