Material, Device Engineering, and Performance INVITED

Abstract

Spin-transfer torque magnetoresistive random access memories (STT-MRAMs) have been successfully introduced into integrated circuits owing to their non-volatility, high endurance, high-speed operation, and scalability. Since the discovery of a perpendicular easy-axis in a CoFeB/MgO material system [1], which has become the standard material system for magnetic tunnel junction (MTJ), a core building block of STT-MRAM, MTJ scaling has been accelerated while improving device performance in data retention by engineering MTJ device structure (Table 1) [2]. Here we review the MTJ scaling technology and discuss our recent results on ultra-small shape-anisotropy MTJs.Maintaining high data-retention performance, represented by thermal stability factor Δ, is essential for nonvolatile applications. Δ is proportional to total magnetic anisotropy energy density and a volume of a nanomagnet. Hence, magnetic anisotropy energy densities arising from the bulk, interface, and shape are needed to be enhanced to achieve high thermal stability as the MTJ size is scaled. Following the first demonstration of the perpendicular MTJs using a single CoFeB/MgO interface [1], those using double CoFeB/MgO interfaces were invented to enhance the interfacial anisotropy [2], resulting in high thermal stability [2] and thereby high-performance STT-MRAMs using MTJs with tens of nm in diameter for production. Once the MTJs are scaled beyond 2X nm, however, interfacial anisotropy, which is proportional to the MTJ area, becomes insufficient to achieve high thermal stability [4]. To overcome the limitation, the MTJ using quad interfaces was proposed to further enhance interfacial anisotropy for scaling down to 1X nm [5]. Still, scaling of those interfacial-anisotropy MTJs while achieving high performance is challenging for scaling beyond 10 nm.Shape anisotropy, which negatively impacts the total magnetic anisotropy in the interfacial-anisotropy MTJs, was shown to work positively in these dimensions [6]. By making a nanomagnet cylindrical with its thickness larger than its diameter, shape anisotropy turns the easy axis to the perpendicular direction instead of the in-plane easy-axis. In this way, both interfacial and shape anisotropy positively contribute to the total magnetic anisotropy, allowing one to scale MTJs beyond 10 nm. We made the MTJs with a 15-nm-thick FeB layer and achieved the MTJ scaling down to single-digit nanometers (X nm) while maintaining high Δ, yet observing STT switching [6]. Moreover, to understand the physics governing the temperature dependence of the properties of the shape-anisotropy MTJs, we studied the scaling relationship between the energy barrier E between parallel and anti-parallel states and spontaneous magnetization with respect to the temperature and compared it with that of the conventional interfacial-anisotropy MTJs. We found that E of the shape-anisotropy MTJs is less sensitive to temperature and that the shape-anisotropy MTJs have the potential to be used for wide-temperature applications at an X-nm scale [7].The shape-anisotropy MTJ is proved to be promising for scaling down beyond 10 nm, but several challenges in the shape-anisotropy MTJ remain: (1) switching current is needed to be further reduced at short pulse and (2) an upper limit of a nanomagnetic thickness for coherent switching exists. These challenges are arising from a large volume (thickness) of a nanomagnet meeting high Δ. To address the challenges, we recently proposed the shape-anisotropy MTJ using multilayered ferromagnets, instead of a single ferromagnet [8]. The newly proposed shape-anisotropy MTJs have multilayered ferromagnets separated by MgO layer(s) so that the total ferromagnetic layer thickness can be reduced owing to the enhancement of total interfacial anisotropy. With the multilayered ferromagnet, we showed that a 5-nm-thickness reduction is achieved while obtaining the same Δ as the single ferromagnet and that the MTJs are scaled down to 2.3 nm. In addition, we demonstrated high performance in the X-nm shape-anisotropy MTJs with the multilayered ferromagnets: (1) stable switching (Fig. 1a) and high thermal stability (Fig. 1b) at high temperatures, (2) switching efficiency improvement (Fig. 1c), and (3) high-speed switching down to 10 ns with the voltage below 1 V (Fig. 1d).The shape-anisotropy MTJs in our studies use the CoFeB/MgO material system without introducing any new material. This means that the device structure can be readily employed in the existing MTJ technology by making a taller and smaller nanomagnet. Note that, not limited to the CoFeB/MgO material system, the concept of the shape-anisotropy MTJs can be adopted to other material systems; indeed, the concept was proved in Co or NiFe material systems [9]. Having the results and the versatility of the concept, the shape-anisotropy MTJs holds promise in the era of the ultimate scaling.This work was supported in part by JST-OPERA JPMJOP1611, JSAP KAKENHI JP19K04486 and JP19J12926, Cooperative Research Projects of RIEC, and DIARE of Tohoku University. J.I. acknowledges financial support from GP-Spin of Tohoku University and JST-OPERA. **