Smaller, faster, cheaper - that was the trend in computing for the past 50 years. As microprocessor technology is hitting the limits of what’s physically possible, some researchers are exploring whether a new technology, ‘spintronics’, may be the way forward.
At the rate we are going, in five years we will have produced the smallest electronic components possible; Moore’s Law, the prediction by Intel co-founder Gordon Moore that transistor density on integrated circuits would double approximately every two years, will come to a shuddering halt. But Cambridge University physicist Jason Robinson just might have a solution that would help us shrink our computers even further. His work is combining two of the hottest fields in solid state physics - superconductivity and spintronics. Until five years ago, this wasn’t thought possible, but recent breakthroughs show that taken together, these technologies might represent the future for super-fast large-scale computing.
For at least two decades, researchers have been looking into ‘spintronics’ or spin-electronics as an alternative to semiconductor CMOS (complementary metal-oxide semiconductor) technology. Spintronics exploits a property of electrons called ‘spin’. The electron isn’t really spinning on its axis, but the term describes a magnetic quantum energy state that can be either up or down and can be flipped by a magnetic field.
In certain materials, controlling electron spin could provide the basis for switching devices and storing and manipulating digital information. The electrons’ up or down spin states would represent the 1s and 0s used in computer logic. “A spintronics circuit in principle can be much smaller,” says Robinson. Spin, being the property of a single electron, could allow for greater miniaturisation than conventional charge-based electronics.
Spin is already being used to read data in magnetic hard disk drives, based on the ‘giant magnetoresistive’ (GMR) effect. This effect was discovered in the late 1980s and developed commercially by IBM. It has allowed the creation of ‘spin valves’ that read computer memory. A basic GMR structure is made up of two magnetic metal layers sandwiching a non-magnetic metal. Because of electron spin, the electrical resistance of the valve will change, depending on whether the two magnetic layers have their fields aligned or in opposite directions. When they are passed over the very tiny magnetic fields that store information or ‘bits’ on hard disks, the field of the outer magnetic layer will change its alignment and so effectively switch a current on or off.
GMR technology has evolved since the first IBM device in 1997, and now spin valves have been used to make magnetic random-access memory (MRAM). In 2012, Arizona company Everspin Technologies became the first developer of high-speed spin-transfer torque magnetic random-access memory (STT-MRAM), based on a phenomenon known as the spin-transfer torque effect. This allows spin transferred from an electric current into a magnetic layer to write or read bits.
However, Robinson says there are problems that hold back wider spintronics applications: “To generate enough spin, you need very large charge currents in your circuit inputs”.
Jacob Linder, physics professor at the Norwegian University of Science and Technology in Trondheim, has also been looking at the challenges that arise with spin-based logic. “The heating associated with operating spin-based devices is in many cases simply too large to be of efficient use,” he explains.
This is where superconductivity becomes interesting - because superconducting materials can conduct electricity without resistance and carry current indefinitely without losing energy to heat.
The idea of creating supercomputers using superconducting circuits has been around since the 1960s, and is based on using a so-called Josephson junction, named after 1973 Nobel physics laureate Brian Josephson. These are made from a thin layer of non-superconducting material, sandwiched between two superconducting layers, between which superconducting electrons can tunnel. This creates a device analogous to a transistor, but with the potential to switch hundreds of billions of times a second, considerably faster than a semiconductor.
Cooling it down
There is a caveat: to be superconducting, these materials need to be below their ‘critical temperature’, (the point at which their electrical resistance drops to zero) so a superconducting computer needs to be cooled to between 1.5 and 5 Kelvin - that’s at least minus 268 degrees Celsius. Despite modern cryocoolers being able to reach such temperatures without using liquid helium, cooling still has an energy cost.
According to Marc Manheimer, manager of the so-called Cryogenic Computing Complexity (C3) programme in the United States, this will be offset by the advantages of superconducting computing. “The basic calculations indicate that we could save probably a factor of 100 in energy relative to current supercomputers,” he says.
C3 is a pioneering five-year programme run by the US government high-risk research unit IARPA (Intelligence Advanced Research Projects Activity) within the Office of the Director of National Intelligence. Launched in December 2014, it is working with academic and industry partners to develop a prototype superconducting computer that will include new kinds of cryogenic spintronic memory.
Just designing the necessary superconducting circuits without any existing software tools is a big job, says Manheimer. “There is no such thing in the superconducting industry, so we have to bootstrap our way up to being able to design these large circuits.”
So could harnessing superconductivity solve the high energy consumption problem of conventional spintronics? Until 2010 this wasn’t thought possible, because of a fundamental incompatibility between the two technologies. In their superconducting state, electrons travel in twos - known as Cooper pairs - with opposite spins, one up and one down. This means the overall spin will be cancelled out, so it wasn’t believed feasible to create superconducting devices that can manipulate spin.
However, in the past few years several international research groups, including Robinson’s group at Cambridge’s Materials Science Department, have proven that superconducting and magnetic materials can be combined to create spin-carrying devices that are also energy-efficient. Robinson was one of the first scientists to show that electron spin could be manipulated within superconducting currents - the first step to marrying both technologies.
His solution involved carefully interfacing multiple nanometre-thick layers of a magnetic material such as cobalt and holmium, in which the magnetisation orientation of the different layers point in different directions. A layer of superconducting niobium is sandwiched between these layers. “In my superconducting devices, the Cooper pairs change the way they pair,” explains Robinson. “On injecting a spin-less supercurrent through the magnetic structure from the superconducting niobium, the output charge current gains a net spin, but maintains zero electrical resistance.” This is because as the superconducting Cooper pair travels through the magnetic structure, one of the spins is able to ‘flip’ direction to create a so-called spin triplet pair, in which the electron spins are parallel. These triplet supercurrents carry spin in addition to charge, but experience no resistance.
The devices that Robinson has designed could perform computer logic operations using magnetic fields to switch spin states, but with greater energy efficiency than conventional spintronics. He is now hoping to produce the first prototype superconducting spintronic device. Although still at an early stage, he is working with a network of international collaborators including the Hitachi Cambridge Laboratory. “Hitachi are interested in where the science is going… We are working with them, doing very low temperature measurements to try and characterise the nature of this superconducting state,” Robinson says. He is also experimenting with alternative materials to find the best combinations of superconducting and magnetic materials.
Professor Linder says that another major challenge is “to improve how well the spin flow in the superconducting hybrid structures can be controlled”.
So what kind of improvements to processing speeds could be expected for future superconducting spintronic supercomputers? Marc Manheimer’s C3 programme works on a prototype superconducting computer that could reach processing clock speeds of 10GHz. He says that’s about a factor of two higher than current conventional computers based on CMOS. Linder says that superconducting circuits have been demonstrated to operate as frequency dividers (a circuit that can change output frequency) at up to 770GHz, which stands as the world record for any digital technology. Using superconducting spintronics could provide even faster processing power, but those working in the field are reluctant to speculate which speeds they may achieve in real life.
One issue that might prevent this technology replacing today’s semiconductor electronics is the low temperatures needed to reach superconducting states. Physicist Detlef Beckmann at the Karlsruhe Institute of Technology in Germany acknowledges that “this is definitely not something that will make it into the next iPhone. It’s something that will be at best a niche application.” Beckmann has been working on superconducting aluminium, which has a critical temperature of 1.2K.
But Robinson says that cloud computing could change things. “I’m not entirely convinced low temperature is a problem,” he says. “The philosophy of computing is changing very fast. In the not too distant future your computer will just consist of a small processor and a screen, and everything will take place in a data centre. Eventually your computer might even be able to be powered by solar energy or other sources of energy from our surroundings, such as vibrations.”
Another solution to the temperature problem could be to use high temperature superconducting materials. Superconductors with abnormally high critical temperatures were discovered in the late 1980s. To date, the superconductor with the highest transition temperature is mercury barium calcium copper oxide at around 133K or minus140�C. “Of course, one would want to move in that direction,” says Robinson, “but it’s bound to be an even slower path.” At the moment, these sorts of materials are very difficult to make and not well understood. “The superconductivity itself is quite different,” he explains. “There is still no unified theory for superconductivity in oxide high temperature superconductors.”
There is another new class of materials that might provide a route to room temperature superconducting spintronics. Topological insulators were discovered in 2007 by researchers at the University of Würzburg in Germany. These are compounds based on bismuth, antimony, telluride and selenide. “The reason why everyone is so excited,” Robinson explains, “is because if you pass a charged current through a topological insulator, you naturally generate a spin current with very little energy loss”. The materials are insulating in the bulk, but at their edges they become superconducting of both charge and spin.
Topological insulators also have another property that makes them interesting. The electrons flowing over the surface have their spins locked to the physical direction in which they move - known as spin-momentum locking. This means that the electrons are immune to the resistance they experience inside an ordinary conductor, so they are superconducting even at room temperature. Jacob Linder says it could also prove useful for spintronics devices: “It means that all electrons moving in the same direction have to carry the same spin, so you have a completely spin-polarised current.”
The area is gaining a lot of interest, but Robinson says that these materials are in their infancy. It is not clear exactly how the effect could be exploited. “If you attach some other structure to it, it’s not easy to see how you could tap into that spin… There are lots of ideas floating around, but no one has actually done it,” he adds. On top of this, the materials themselves are very reactive and difficult to handle.
The other material that is being suggested for spintronics is graphene which was first identified in 2003. This two-dimensional form of carbon consists of a single layer of hexagonally arranged carbon atoms. According to Linder, “the electrons in graphene have some similarities to the electrons on the edge of a topological insulator, they feature very high mobilities”.
It was predicted that graphene should be able to carry spin-polarised currents, but so far there has been little success, says Beckmann. “Whether this is just imperfections of the present experiments or whether there is some more fundamental reason behind it is not clear.”
The future for superconducting spintronics is hard to predict. As Marc Manheimer points out, “clearly the technology hasn’t tipped yet”. His five-year C3 superconducting supercomputer project has buy-in from several large US companies, including IBM, and he adds that “there is some optimism among American companies that this will work and there will be commercialisable products coming out of this.”
Professor Linder suggests that adding spintronic logic to superconducting circuits will be even further down the line. “It’s fair to say as of today, we are still at the level of fundamental research, but there has been a significant experimental upswing in the most recent years,” he says.
In the meantime, we will have to hope that Moore’s Law will help us to eke out ever more computing power for a little while longer.