Science! – Spintronics

Representing data with the spin of an electron transformed hard drive read heads, but could it do the same for logic circuits too?

Spintronics is one of the hottest topics in science and tech, whether it concerns logic circuits based on the spin or magnetic moment of an electron or qubits in quantum computers. With all that excitement, it’s easy to forget that technology based on spintronics has been a part of the computing experience for the better part of a decade. You know it better as the read head of your hard drive.

In the mid- 1980s, two research teams, headed respectively by Albert Fert and Peter Grünberg, independently discovered the phenomenon of Giant Magneto Resistance (GMR), an effect found in thin film materials composed of alternating ferromagnetic and non-magnetic layers.

Spintronics has already been used in hard drive read heads for the best part of a decade

Fert and Grünberg observed a significant difference in the resistance across such a material, depending on the alignment of spins in adjacent layers. High resistance marks a region of antiparallel alignment, and low resistance a region of parallel alignment – the difference is significant enough to be used to read data from a hard drive. Their work won Fert and Grünberg the 2007 Nobel Prize, led to an explosion in hard drive data densities, and paved the way for a new way of thinking about storage, memory and logic.

At the heart of spintronics is the idea that we can exploit not only the charge of an electron, but also its spin, or magnetic moment. The instant we stop thinking about an electron as a moving object (part of a flow of current), and consider at as an entity in itself, it starts to become more interesting. However, getting from the lab to the PC isn’t always an easy road.

‘The first spintronics device went into production just eight or nine years after a new physical effect – giant magneto resistance – was discovered,’ says Rolf Allenspach, head of IBM’s Physical of Nanoscale systems research group in Zurich. However, the reason the technology moved so astonishingly quickly from lab to production was because it was a ‘drop-in’ replacement, he explains. Nothing needed to be redesigned; the new head just replaced the old one and did a better job.

Rolf Allenspach, head of IBM’s Physics of Nanoscales systems research group in Zurich

The road ahead for other spintronics devices might not be as smooth, and certainly seems to be longer, but still possible to navigate. One of the most active areas of research concerns something called magnetic domain walls and spin-transfer torque, which is most certainly not a ‘drop-in’ replacement for existing tech. the physics is at the heart of both new types of memory, and room-temperature spintronics logic.

Both are based on the same basic idea. ‘A conventional field effect transistor works because charge moves around, and we can break that flow of electrons to produce the logic states of one and zero. Spintronics, meanwhile, encodes the logic one or zero in the spin – either up or down – of an electron. Instead of diverting flow, you flip the spin to change the logic state,’ Allenspach explains. ‘In an ideal case, you wouldn’t have any current flow, just the flow of spin. This is what reduces heat dissipation.’

You can flip the spin with magnetic field (which is difficult to control inside devices), or with electric field (which may be easier, especially at a small scale). This can be done at a very low power density, which is a major reason why people are interested in this tech.

Significant work is being carried out on developing these discoveries even further into the holy grail: universal memory.

A reliable and non-volatile technology capable of high-density data storage, fast read and writes times would consign the standard hard drive, flash and all the familiar alphabetical RAM variants to the history books.

Racetrack memory, which is the brainchild of IBM researcher Stuart Parkin, is IBM’s stab at it. The firm first demonstrated the technology in 2008, but at the end of last year it showed off a version integrated with standard 200mm wafer CMOS technology.

Racetrack memory, the brainchild of IBM researcher Stuart Parkin, stores data as magnetised regions on nanowires that can be written, read and overwritten as they whizz past fixed read and write heads

Racetrack stores data as magnetised regions on nanowires that can be written, read and overwritten as they whizz past fixed read and write heads. Since the bits themselves move along the nanowires, you only need one transistor per racetrack, as opposed to one transistor per bit. You can also stack the wires, creating data densities that not only equal but surpass conventional hard drives. What’s more, the data is non-volatile: it persists even when the power is switched off.

You may recall from GCSE physics that in a ferromagnet, the spins of all the electrons are aligned. You can have regions within a material where spins are aligned one way, right next to areas where the spins are aligned the other way. The boundary between the two differently aligned areas, the region in which the magnetic moment of the electrons changes, is known as the magnetic domain wall. One direction of the wall is logic state one, the other direction is logic state zero.

The spin of a given region can be manipulated by a magnetic field, but this doesn’t necessarily move it along the nanowire. Instead, you can move the domains with a pulse of spin-polarised current. Generally, an electrical current isn’t polarised – it has an even mix of up and down spin-states. However, if you load the current with more electrons spin-up, for example, when in hits a magnetic domain wall, the spin – angular momentum – of the electrons is transferred, switching the polarisation of the region on either side of the domain wall.

This works because angular momentum is a conserved quantity – you can’t get more than you already have, and you have to keep what you have. So as the polarised electron in the pulse of current hits the magnetised domain and switches from 0 to 1, another electron – one in the magnetised domain – has to flip from 1 to 0.

This transfer of momentum can move a magnetised bit along a wire at an impressive 150nm per nanosecond in ideal conditions. This gives you a non-volatile form of high-density memory with access times in nanoseconds; much faster than hard drives.

It isn’t perfect thought. Allenspach acknowledges that the read-world performance is still some way below what theory suggests should be possible. ‘We found we needed much longer current pulses than we expected to move the domains,’ he says.

While we’ve had years of perfecting silicon production, making perfect ferromagnets is still a new game. Tiny defects in the crystalline structure were causing the magnetic domains to become stuck, meaning that the current had to be applied for much longer to shift them along. In a clear stretch of permalloy wire, though, the theory held, and pulses of just a few nanoseconds were enough to move the domains along.

The real world gets in the way of spintronics in logic too. This is too large a topic to cover in detail here, but basically, this work can be divided into two main avenues, each of which has promise, but also some very obvious difficulties that have yet to be overcome.

The first avenue works at room temperature and is based on the same underlying physics as racetrack: ferromagnetism. This would mean a total overhaul of computer design to get it into PCs.

The alternative approach, which is compatible with current materials, only works below 200⁰ Kelvin. That’s a bracing -73⁰C in real money, so it isn’t much cop in your average air-cooled desktop PC, and not really appropriate for the mass market.

To garner the full potential of spintronics, further fundamental advances are urgently needed.

While these kinds still exist in memory and logic research, and current technology works, the industry will be reluctant to move from current standard technology that has proven reliability. The implication is that, despite spintronics’ potential, for now, standard memory is doing enough to stay in the race.

‘Whether or not any of this becomes commercialised depends on the difficulties in the silicon industry,’ Allenspach concludes. ‘It’s hard to predict if an ingenious engineer will find a new solution for the problems facing existing technology, but we want to be ready if the standard approach fails.’