Spintronics is gaining popularity because of the interesting physical phenomena that can be observed in these systems as well as due to the potential applications that can be derived from it. While the magnetic read sensors of the hard disk drive are already beneficiaries of the early development in spintronics, the magnetic random access memory (MRAM), magnetic logics, spin-transistors, and quantum computing are some of the exciting possibilities that may arise from the outcome of further spintronics research. Out of the several emerging applications, MRAM is given a stronger industrial emphasis due to the several advantages it offers over other competing memory/storage class memory. Moreover, the recent developments in the spin-transfer torque (STT) induced switching of magnetization have provided more excitement and realistic possibilities of MRAM being a strong candidate for future memory applications. This article provides an overview of MRAM with a particular emphasis on materials with perpendicular magnetic anisotropy (PMA) for this application.
The discussion begins with a brief review of history and the earlier generations of MRAM which were based on field-induced switching. The basic magnetics and spin-transport phenomena are also explained along. The discussion is focused next on the MRAM designs with STT for magnetization switching. The MRAM designs with in-plane anisotropy are discussed here. The discussion moves next to the key theme of the paper – materials with perpendicular magnetic anisotropy. The developments in this area are discussed in detail. This particular part is focused on three classes of materials: multilayers of the type Co/Pd, CoFe/Pd etc.; FePt based materials; and CoFeB layers with a perpendicular magnetic anisotropy. The discussions end with an analysis of the challenges, future directions and possibilities.
Fundamentals of MRAM and the history
The early versions of memory that exploits magnetism for random access date back to the 1960s, where magnetic cores were used to store memory. In core memory, the information was stored in small magnetic toroids, through which wires were threaded to read or write information. However, with the advent of semiconductor memory, the magnetic core memories lost their advantages and hence their market share. At a later stage, magnetic bubble memory drew attention as a candidate for memory, but only with a short-term success. In the 1990s, with the introduction of giant magnetoresistance (GMR) in hard disk drives, designs of magnetic random access memory with GMR were proposed. However, the interest in MRAM grew significantly only after the observation of spin-torque induced switching and magnetization excitation in Co/Cu/Co systems 1–5, long after a theoretical prediction by Sloncewski 6 and Berger 7. This section provides the basic principles of the MRAM using field-induced switching.
Figure 1(a) is a schematic of the MRAM based on field-induced switching. In this design a single transistor and a magnetic tunnel junction (MTJ) device are used to store one bit. For the reading, the magnetoresistance of the MTJ device is measured. In this scheme, the reading is based on the voltage obtained from GMR or MTJ devices whose resistance is high or low, depending on the relative orientation of the magnetizations of the free and reference layers. Giant magnetoresistance, an observation made independently in the late 1980s by the groups of Grünberg 8 and Fert 9 who received the Nobel prize in 2007, is a phenomenon where the resistance of the device depends on the relative orientation of magnetization of the magnetic layers. When the magnetizations of the layers are oriented in the same direction, the resistance will be lower due to the spin-dependent scattering of the minority electrons only. However, when the magnetizations of the free and reference layers are anti-parallel to each other, both the minority and majority electrons will be scattered resulting in a larger resistance state. These two states can be used to represent “1” and “0” states.
The selective writing of a particular MRAM cell is based on the application of direct current through the wires. Although the current will produce a magnetic field throughout the wire, it is not sufficient to switch any cell by itself. However, switching will happen only in the MRAM cell at the intersection point, where the two fields in orthogonal direction lower the switching field of the cell at the junction. In field-induced switching MRAM, the current required to achieve the desired field – which causes the switching of the free layer magnetization – increases with the reduction in the size of the current lines. This causes a limitation in the scalability of MRAM devices to smaller sizes. Therefore, the MRAMs based on field-induced writing have a limitation at about 90 nm.
Spin-transfer torque (STT)
The theoretical prediction and experimental observation of the STT effect gave an impetus to the MRAM again. Figure 2 shows a typical STT device with a reference layer and free layer separated from each other by a non-magnetic spacer that can be a conductor for the case of GMR or insulator for TMR. When the current of electrons flows through the tri-layer structure, for example, from left to right, it will be polarized by the reference layer with fixed magnetization direction. Thus, the polarized spin current direction is aligned with that in the reference layer. When the spin current reaches the free layer, it will interact with the local spins through exchange coupling and attempt to rotate the local spin direction along with that of the spin current. If the applied current density is high enough (above the switching threshold), the magnetization of the free layer could be reversed. STT-MRAM adopts the new mechanism to write a memory cell. By changing the writing current polarities, the magnetization of the free layer can be switched between parallel and anti-parallel directions with respect to that of the reference layer, which corresponds to high (“state 1”) and low resistance states (“state 0”) of the device. Figure 1(b) is a schematic of a MRAM based on the STT effect for switching magnetization of the free layer (writing). As the current required in STT switching can be reduced with the device size, the scalability of the STT-MRAM is not an issue. However, one has to keep in mind that the size of the device cannot be reduced without satisfying thermal stability condition as will be discussed in the next section and has been reported in Ref. 10 for the case of in-plane magnetized free layer. Therefore, there is a lot of research on STT-MRAM.
STT-MRAM design requirements
The design of a STT-MRAM cell for high density must satisfy the following five requirements (see Fig. 3): signal, low resistance, high thermal stability, low writing current, and compatibility with CMOS design.
For the first requirement, it is important to have a large signal so that the two states (low resistance and high resistance) could be well separated. Although there is a large number of reports on STT-MRAM based on GMR effect, i.e. spin-valve structure, the small resistance difference between the two magnetic states does not make them practical for MRAM application. Nevertheless, GMR structures allow the understanding of STT switching and the study of new structures, designs and materials. On the other hand, as TMR is much higher than GMR, MTJ structures with high TMR can help to achieve high signal and remain as practical candidates for MRAM application; especially MTJs based on MgO tunnel barrier which provide much higher TMR compared to AlOx based MTJ 11, 12.
The second requirement for a functional STT-MRAM is the low resistance or more specifically the product of the resistance and the junction area (RA) where the parameter A is the area of the device. It is relatively easy to achieve high TMR by increasing the tunnel barrier thickness, such as MgO barrier, but RA will increase almost exponentially. The value of RA is dictated by the value of breakdown voltage (VB) which is the critical voltage that can be applied to the device before it becomes unusable (damaged). It is known that MTJ based on MgO has larger VB than the one based on AlOx. For illustration, if VB is 1.5 V, the value of RA has to be lower than the ratio of VB and STT switching current density Jc. This means that the switching of the free layer magnetization should occur before reaching VB.
The third criterion is thermal stability, as one of the main advantages of MRAM is its non-volatility. Therefore, the information has to be stable for a period of 10 years as it is the case for hard disk drive for example. The stability factor of the bit is defined as the ratio of the anisotropy energy (Ku × V) and thermal energy kBT (Ku is the anisotropy energy, V is the volume of the soft layer, kB is Boltzmann constant and T is the temperature). This ratio should be larger than 40 for the information to be stored for 10 years. It can be seen clearly, that for better thermal stability, large Ku is needed. On the other hand, as will be discussed in section 5.2, the STT switching current Ic is also proportional to Ku which makes a limitation to thermal stability. It is also important to mention that PMA materials, to be described in the next section, provide better thermal stability and yet lower switching current compared to in-plane materials used in STT-MRAM. The main reason for this difference has been discussed through Eqs. (2) and (3) in Ref. 13.
The fourth key parameter in the STT-MRAM pentalemma is Jc which needs to be reduced to allow low power operation. Furthermore, the value of Jc (i.e. switching current) is related to the size of the transistor used. Smaller Jc helps to achieve large storage capacity for STT-MRAM, as the required transistors will also be smaller.
All the discussed conflicting requirements for STT-MRAM have to be achieved with a suitable fabrication process that does not compromise CMOS integration.
Materials with perpendicular magnetic anisotropy (PMA)
Materials with PMA have larger anisotropy energy compared to most of in-plane anisotropy materials. In addition, for the patterned device, the magnetization of PMA materials is more uniform and does not suffer from thermal instability due to magnetization curling observed at the edge of in-plane case 14. From the point of view of STT switching current, it is easier to reverse the magnetization of PMA free layer than the in-plane anisotropy one 15, 16. As choice for materials with PMA to be used in GMR or TMR devices, three main categories have been explored.
Multilayer-based spin valve and magnetic tunnel junctions
Multilayers based on transition metal (TM) like Co, Fe and CoFe and noble metal such as Pt, Pd and Au show perpendicular magnetic anisotropy for a certain range of thickness and number of periods 17–20. Although these materials have been extensively studied for magneto-optics and magnetic recording, especially bit-patterned media 21–27, they have also gained interest for spintronics applications 15, 16, 28–35. The magnetic properties of these materials such as saturation magnetization and anisotropy energy can be easily tailored changing the number of bilayers and also by adjusting the thicknesses of TM and Pt (or Pd). The commonly used multilayers so far are Co/Pd, Co/Pt and Co/Ni due to their relatively larger PMA compared to Co/Au or Co/Ag for example. Firstly, multilayers were used as part of spin valves in the soft layer and hard layer without the need of anti-ferromagnetic layer which is required for the in-plane magnetoresistive case. In an earlier work on Co/Pd multilayers, Carcia et al. demonstrated that for Co thickness less than 0.8 nm the multilayer showed PMA which has been attributed to the surface anisotropy at Co and Pd interfaces and the strain in Co thin layers 17.
For magnetoresistive structures, Joo et al. used FeMn as an antiferromagnetic (AFM) layer to induce exchange bias to one of the Co/Pd multilayers 36. Their structure was similar to what was used in conventional in-plane spin valve (SV). However, since it is easy to tailor the coercive field of the multilayer by changing the thickness of TM and Pd (or Pt) and the number of periods, the AFM layer is not required. This makes the structure simpler and sometimes called pseudo-spin-valve. Law et al. used Co and CoFe to further improve the softness of the soft layer and also to achieve a better magnetoresistance ratio (MR) 29.
Although much work has been done on spin valve, there is a growing interest in using multilayers with PMA in MTJ devices. The first work started with MTJ based on amorphous AlOx tunnel barrier, as AlOx is easy to be deposited in comparison to crystalline MgO. Actually, for the cases of MTJ with MgO barrier, coherent tunnelling which provides high TMR is possible only for the (001) phase of MgO. This is a challenging requirement from the materials point of view as there is a need for a judicious choice of all layers deposited below the tunnel barrier. Co/Pd and Co/Pt multilayers grow with (111) fcc structure on selected seedlayers such as Ta, Pd, Pt or Ru. So far, MTJ based on amorphous AlOx with Co/Pt as soft and hard layers showed a relatively low TMR of about 25% in the best case 37–40. This is mainly due to the low spin polarization of the Co/Pt multilayer compared to CoFe or CoFeB alloys as a consequence of the insertion of Pt or Pd non-magnetic layer.
The MTJ based on MgO requires a tight control of growth of the magnetic electrode deposited below. The best choice is a ferromagnetic layer that can favour the growth of MgO with (001) phase so that coherent tunnelling could be obtained 41. Unfortunately, the multilayers studied so far based on the alteration of TM and noble metal grow in fcc (111) phase. This is the main reason for low TMR obtained with these structures even with the insertion of very thin CoFeB at both MgO interfaces to help improve the growth of MgO tunnel barrier 42.
More recently, Yakushiji et al. showed that by using very thin Co and Pd (or Pt) of less than 0.2 nm, it is possible to achieve a high TMR in MgO based junctions 43. The multilayer called superlattice can be annealed at high temperature exceeding 350 °C which is important for promoting the (001) phase of MgO, essential for high TMR. In comparison, the conventional multilayers (referring to relatively thicker Co and Pd) will face degradation of their intrinsic properties at such high temperature because of the Co and Pd (or Pt) intermixing. As can be seen in Fig. 4, a TMR of 62% could be obtained in a structure where the soft layer is composed of [Co(0.2 nm)/Pd(0.2 nm)]4 and 0.8 nm thick Co60Fe20B20. The hard layer is composed of 1.3 nm CoFeB and 16 nm TbFeCo. Although Co/Pd and Co/Pt superlattices can be annealed at 350 °C without degradation of their magnetic properties, the existence of TbFeCo as part of the hard layer limits the annealing temperature. It is then expected that larger TMR could be obtained if one could find a replacement material for the hard layer.
FeCoB with PMA
As discussed earlier, STT-MRAM with PMA has a better scalability in comparison with in-plane STT MRAM in terms of low critical switching current density (Jc) and high thermal stability. In addition, the device with PMA has improved shape uniformity, which is critical to control the driving current distribution for STT induced switching.
In the early development stage, rare earth (RE) and transition metal (TM) alloys, such as TbFeCo and GdFeCo 44, 45, were adopted to obtain PMA in MTJ devices. However, the magnetic property of such materials is sensitive to oxidation during the fabrication process, especially for sub-100 nm size. This unavoidable oxidation is still a challenge to integrate RE materials with STT-MRAM. Another method to obtain PMA is to use multilayer structures, such as TM/Pd or TM/Pt, as discussed in the previous section. However, due to the strong spin–orbit coupling of Pd and Pt, the damping constant α is found to be several times larger than in regular TM and results in a large STT switching current Ic according to
where H is the applied field, Hk is the effective anisotropy field, Ms and t are the saturation magnetization and thickness of the soft (free) layer, respectively. In Eq. (1), A is the device area, e is the charge of electron, ħ = h /2π (where h is the Planck constant) and η is the spin polarization.
In 2010, first principle calculations conducted by Shimabukuro predicted that the hybridization of Fe-3d and O-2p orbitals may introduce PMA in an MgO/Fe/MgO system 46. In the same year, Ikeda et al. demonstrated PMA in CoFeB–MgO based MTJ devices as can be seen from Fig. 547. The easy axis of magnetization of a CoFeB layer is in-plane of the film for a relatively large thickness (more than 1.5 nm or so) while it is oriented in the out-of-plane direction for small thickness. The PMA energy density (Ku) of the CoFeB–MgO system increases with reducing CoFeB thickness. The achieved Ku could support a memory cell size as small as 40 nm with 10 years thermal stability. On the other hand, the STT study shows that Jc0 of the system has a fairly low value of 4 MA/cm2, despite the thermal stability factor of 46. Jc0 is the value of Jc when an STT current pulse of 1 ns is applied. Since the system is based on crystallized CoFeB–MgO structure, a TMR of more than 100% was achieved. The demonstrated perpendicular CoFeB–MgO MTJ system combines sufficient thermal stability, low switching current and relatively high TMR signal. Having these characteristics at the same time is a real breakthrough for STT-MRAM.
Besides the CoFeB thickness factor, PMA of a CoFe–MgO system also relies on stack structures and process conditions. Recently, Worledge et al. found that a Ta seed layer is essential to obtain PMA in the system 48. Structures with Ru seed layer were not favourable to obtain PMA in the same CoFeB thickness range. Hao et al. reported that the annealing process is also a critical step to build up PMA and needs to be optimized in order to achieve simultaneously good TMR, low resistance and low Jc049. Although the demonstrated PMA in thin CoFeB layer shows a strong potential application for STT-MRAM, it should be noted that this will result in a larger Jc. In addition, TMR is also degraded as the CoFeB thickness decreases. Thus, the CoFeB thickness in a perpendicular MTJ stack should be optimized in order to make a trade off among PMA, TMR and Jc.
To reduce Jc and yet avoid the decay of TMR, the interfacial PMA can be also applied to in-plane MTJ devices. For an MTJ with easy axis in the film plane, the STT current Ic is defined by
where H⊥ is the PMA component which partially cancels the effect of the out-of-plane demagnetization field. Although the easy axis of the free layer is still in the film plane, its PMA component is expected to reduce Jc. Based on this design, Amiri et al. reported a Jc reduction by a factor of 2.4, which is more than the corresponding reduction in the free layer volume alone 50.
FePt materials for ultra-high density
The cell size of STT MRAM has to keep shrinking in order to catch up with the increase of density. However, the thermal stability factor Δ of each cell, defined by KuV /kBT, should be above a value of 40 to resist thermal fluctuation and secure 10 years thermal stability 16. Since V is reduced with size shrinking, Ku has to be increased to maintain the value of Δ. For an MTJ with a diameter d of 20 nm, if the soft layer thickness is 2 nm for example, the required Ku should be larger than 2.5 Merg/cm3 to secure Δ above 40. The energy Ku will dramatically increase to 10 Merg/cm3 for a diameter of 10 nm and further break 40 Merg/cm3 for 5 nm diameter. One of the material candidates to provide such high Ku is L10-FePt. In 2006, Seki et al. integrated L10-FePt in a GMR device and achieved a Ku value of 34 Merg/cm3 at room temperature 51. However, growing FePt with L10 ordered phase requires a substrate temperature as high as 500 °C, which brings a process integration issue with CMOS transistor. In addition, since a special seed layer/substrate such as MgO is required to introduce texture growth before FePt deposition, the material and structure options are very limited. The same group further demonstrated the STT effect in an FePt/Au/FePt GMR structure, where Jc was up to 108 A/cm2 due to the high Ku value, weak spin polarization and large damping constant of FePt 51. Two years later, a lower Jc of 2.7 × 106 A/cm2 and a TMR of 60% were demonstrated in MgO based MTJ consisting of CoFeB/(Pd/Co) free layer and FePt/CoFeB reference layer 52. The relatively low Jc and high TMR signal are mainly attributed to the crystallized CoFeB–MgO system. However, FePt in the demonstrated MTJ stack only serves as the reference layer, which does not contribute to the thermal stability of the storage layer (free layer).
Challenges and future directions
As mentioned in an earlier section, the pentalemma – where five conflicting requirements have to be met simultaneously – remains a major challenge for materials with a perpendicular magnetic anisotropy. At the entry level, the CoFeB–MgO system with interfacial anisotropy may play a significant role for devices down to 30 nm size based on anisotropy energy only. For devices with smaller sizes (below 30 nm), it is necessary to use materials with higher anisotropy such as Co/Pt multilayers or FePt systems. However, radically new approaches remain keys to write data on such layers without sacrificing speed.
Rachid Sbiaa is a research scientist at the Data Storage Institute, Singapore. He obtained his Ph.D. in 1996 from Paris University. After his graduation he worked as a research fellow in the laboratory of magnetism and thin films, CNRS (France). In 2002 he joined TDK Co. (Japan) to work on spintronics and magnetoresistive read-head sensors. In June 2006 he joined the Data Storage Institute, and has been working on magnetic materials and nanostructures for magnetic recording and spintronics.
Hao Meng received his Ph.D. degree in Electrical Engineering from the University of Minnesota (Twin Cities), United States, in 2007. He joined Recording Head Operations, Seagate Technology in the same year. Since 2008, he has been working at the Data Storage Institute, Singapore. His main research interests include spin-transfer-torque magnetic random access memories (STT-MRAMs), magnetic sensors and spintronics logic devices.
S. N. Piramanayagam received his Ph.D. in 1994 from the Indian Institute of Technology, Bombay (India). After his post-doctoral stint at Shinshu University (Japan), he joined the Data Storage Institute, Singapore in 1999. Since then his research has mainly focussed on magnetic materials for recording media and spintronics applications. He is an editorial board member of IEEE Magnetics Letters and three other journals. He has recently published the book “Developments in Data Storage: Materials Perspective” (IEEE–Wiley Press).