MRAM catches an STT

MRAM catches an STT

After a month long absence, Pick of the Week returns with a look at a fantastic commentary article on spin-transfer torque magnetoresistive random access memory (STT-MRAM), “A new spin on magnetic memories”, by Andrew Kent and Daniel Worledge. The article is free to read (thanks to the kind sponsorship from Spin Transfer Technologies), and is part of an edition of Nature Nano’s March Focus that looks into possible routes for a new generation of solid state computing devices. It is also worth checking out other articles in this edition of Focus looking at resistive RAM, domain wall memory, and electric field control of magnetism to keep an open mind as to the competition STT-MRAM may face to become the universal memory to topple the current hierarchy of SRAM, DRAM, hard disk drives, and flash.


Something to torque about

STT was theoretically predicted in the mid 90’s by Slonczewski and Berger independent of one another. This was soon followed by several experimental observations of the phenomena. The principle of STT is that injecting a spin-polarised current into a magnetic layer can lead to a transfer of angular momentum between the injected spins and the local moments of the magnetic layer if their orientations are non-collinear. Such spin polarised currents have historically been generated by passing unpolarised electrical current through a ferromagnetic layer with a fixed magnetization orientation (known as a polarising layer), however recently spin currents have also been generated in non-magnetic materials owing to the spin Hall effect (SHE). A key difference between the spin filter and SHE methods is that the former yields a spin polarised electrical current, while the latter a pure spin current (no net charge transfer).


Writing the wrong

MRAM has been commercially available for around ten years now. MRAM differs from conventional RAMs used in today’s computers as information is stored in the state of magnetic layer rather than the charge of a transistor. MRAM also differs from hard disk drives, another form of magnetic memory, as no mechanical motion is needed for reading and writing operations. Indeed, the time scales for these operations in MRAM are short enough that it can be considered as a candidate for primary memory that is directly accessed by a computers CPU. At the same time, the stability offered by storing data magnetically also means that the operation of MRAM is non-volatile, i.e. data is retained when power is removed. MRAM that is currently commercially available is written by applying external magnetic fields (toggle MRAM), however this approach limits scalability to the extent that MRAM may not be able to offer a sufficient on chip density to compete with current RAMs. It is therefore vital to find a more scalable method of writing MRAM, and STT is the leading contender.


Read all about it

Kent’s commentary article introduces the basic design of an STT-MRAM cell, or bit. This consists of a ferromagnet/dielectric/ferromagnet stack, with associated bit and word line electrical connections. The ferromagnetic layers have different coercivities; one is ‘pinned’, meaning that its magnetic orientation cannot change with the magnitudes of current densities used in the device operation. The other is ‘free’, meaning that its magnetic orientation can be switched by the spin-polarised current in the device operation.  Current is applied through the bit line to write the free layer (it becomes spin-polarised by flowing through or reflecting off the pinned layer). Reading is performed by passing a lower current thorough the bit line and measuring the resistance across the stack. The resistance alters with the orientation of the free layer due to the tunnel magnetoresistance effect.


Up for the challenge?

Kent goes on to consider some of the major challenges that STT-MRAM needs to overcome before it can be a competitive memory technology:


Firstly, the large voltages that need to be applied across the stack for the write operation can lead to degradation of the dielectric tunnel barrier (usually MgO). This sets an upper limit to the magnitude of the current that can be used for writing. More promisingly, there is no current-induced atomic wear out of the magnetic layers, and the speed of reading and writing are better than DRAM.


Current processing of STT-MRAM is carried out at 300oC to avoid damage to the magnetic layers. However industrial scale device processing is carried out at 400oC, and so the thermal endurance of the magnetic components in STT-MRAM needs to be increased. Nevertheless, compared to flash memory, the simple fabrication procedure and compatibility with low voltage (and hence low cost) transistors of STT-MRAM make the necessary materials development advances for processing worth pursuing.


Kent highlights that greater data retention is brought about by a stronger magnetic anisotropy of the free layer, but that a stronger magnetic anisotropy of the free layer will mean that a larger current density needs to be applied for switching. This problem can be tackled by using ferromagnetic layers whose magnetizations are orientated out of the thin film plane. Stronger magnetic anisotropy than what is possible from in-plane shape anisotropy can be achieved by utilising materials with strong out of plane bulk magnetocrystalline anisotropy or interface anisotropy (e.g. at the CoFeB/MgO interface), both of which arise from spin-orbit coupling. These magnetic anisotropies can reduce the size of a stable bit from around 40nm to around 20nm. The writing current density also reduces for an out of plane magnetization due to energetic considerations.


Finally, write errors can be caused by thermal fluctuations. The thermal fluctuations can cause a magnetic moment to temporally change its orientation. If this orientation is collinear to the orientation of the spin polarised current, STT cannot act on the moment. Kent indicates that one route to overcome this problem is to use a spin-filter that is orthogonal to the anisotropy of the free layer, e.g. the spin-filter has an in-plane magnetic anisotropy for a free layer with out of plane magnetic anisotropy. Magnetization dynamics mean that this set-up allows STT to occur upon biasing the stack.


Invest in the best

STT-MRAM is extremely promising for leading the way for future memory devices that are fast, non-volatile, and have low energy consumption. While Kent highlights various problems with STT-MRAM, he also suggests solutions to overcome these. I firmly believe that considerable resources need to continue to be thrown at STT-MRAM (or similar types of MRAM that are written electrically) so that the enormous benefits of MRAM over current RAMs can be realised by consumers within the next 20 years or so.


Image: Daniel Dionne, some rights reserved

Kent, A., & Worledge, D. (2015). A new spin on magnetic memories Nature Nanotechnology, 10 (3), 187-191 DOI: 10.1038/nnano.2015.24