Putting a spin on photonics
Back in 2014, in the application for the Gravitation program, TU/e researchers envisioned hybrid devices that enable the direct optical reading and writing of magnetic memory. As utopian as it might have sounded then, now the group Physics of Nanostructures is well underway to create such high-density magnetic memories that function without energy-eating intermediate electronics.
Photonics is the technological answer to our ever-increasing hunger for data. As compared to electrons, information encoded in photons can be transported over larger distances, with more bandwidth, and at lower losses. But light has one major disadvantage: it is rather difficult to use it for storing data. At TU/e, the group Physics of Nanostructures uses its expertise on spin-based racetrack memories to circumvent energy-eating electronics for data storage by developing the so-called magneto-photonic memories.
Racetrack memories do not use the electrical charge of electrons, but rather their spin to build computer bits. Well-defined magnetic domains are pushed along a thin wire by an electric current. Read/write heads near the wire alter the domains, and by doing so, record patterns of bits. ‘Back in 2014, we imagined what would become possible if we would be able to directly combine such racetrack memories with photonics,’ group leader Bert Koopmans says. ‘We envisioned spintronic-photonic memories where data is transferred between a photonic waveguide and a magnetic racetrack memory without intermediate electronics. Though at that time we were still working out the basics of the racetrack memories themselves, we are now at a point where we have the first signs that such a hybrid memory is indeed feasible,’ he says with pride.
Less conversion steps
At the moment, different functionalities in ICT are enabled by different technology platforms, Koopmans explains. ‘Take current data centers. They consist of enormous racks, filled with high-frequency electronics for data processing and computing, glass fibers and photonic devices for communication, and magnetic hard disks for data storage. As a result, we are constantly converting light into electrons, electrons into magnetic bits, and vice versa. These conversions consume large amounts of energy and lead to delays and power losses. By combining spintronics and photonics you can get rid of the intermediate electronics and store and retrieve data in a more energy-efficient way.’
The idea is to grow a magnetic nanostrip on top of a photonic integrated chip so that it crosses a photonic waveguide. In the chip, a pattern of bits is sent through the waveguide. This bit pattern is written in the magnetic nanolayer above, right at the crossing point. The magnetization in this layer influences the light propagation in the waveguide, and vice versa, so that direct conversion between the magnetic and optical domain is achieved. By driving the magnetic bits in the nanolayer at high speeds, all written information is immediately carried along within the magnetic memory. This way, commercially interesting data rates can be achieved with much less overhead than in current electronic solutions. ‘Recently, we have shown that with this design there are spintronic materials that can be switched optically in an effective way and that are able to transport data with a speed of 1000 meters per second,’ Koopmans says.
“Since in our design the memory is much closer to the processor, we can develop new computing architectures other than the traditional Von Neumann architecture. This is a prerequisite for successful artificial intelligence systems.”
Bert Koopmans | Full Professor
This setup not only reduces some of the problems with the current photonic-electronic-magnetic systems but also enables entirely new applications, the physicist predicts. ‘Our spintronics layers have non-linear characteristics which enable additional operations. We could for example realize AND and OR operations before we write the bits.’
But also more futuristic applications are coming into view, Koopmans explains with ever-growing enthusiasm. ‘Since in our design the memory is much closer to the processor, we can develop new computing architectures other than the traditional Von Neumann architecture. This is a prerequisite for successful artificial intelligence systems, where the data transport between processor and memory units has to be minimized to enable real-time decision-making. We also had some preliminary discussions with Patty Stabile of the Electro-Optical Communication group about what our design could mean for photonics-based neuromorphic computing. When you look at neurons, there are some striking similarities with spintronics. Neurons first have to build up a threshold level before they fire. This is an intrinsic property of magneto-photonic writing as well: only after we have built up a threshold of absorbed laser light, all of a sudden an entire domain will be written.’
To circuit level
Over the past few years, in intense collaboration with Jos van der Tol of the Photonic Integration group, the physicists of the FNA group have proven that it is possible to use light to directly read and write bits in spintronic memories, and to transport those bits in an energy-efficient way. ‘This fundamental work has been done on sample systems. Now we want to take the next step by going to the level of actual circuits and building an actual on-chip demonstrator. Together with Martijn Heck from the Photonic Integration group we are exploring what would be the most interesting architectures for groundbreaking experiments to demonstrate the possibilities of this new hybrid technology.’
Though there are more groups around the world working on combining spintronics and photonics, TU/e has an important head start in this field, Koopmans thinks. ‘Within our Center for Integrated Photonics Eindhoven, we combine a unique set of skills. We have expertise in all necessary areas ranging from fundamental physics in spintronics and plasmonics all the way up to electrical engineering on a photonic circuit level. And we are closely connected to industrial partners who are able to actually produce these chips, which will undoubtedly open up a wealth of new possibilities we cannot even start to imagine now.’
Putting a spin on photonics
INSPIRE: Combining the best of two worlds
Light-based neural networks