Artificial, synthetic, or perhaps spintronic intelligence – what will substitute computers?

Spin electronics, also known as spintronics, is a new field of technology that differs from traditional electronics in that in the spintronic integrated circuits it is not only the charge, but also its spin, being an intrinsic form of angular momentum, that carry data. It is closely connected with magnetism, as an organised orientation of spin in materials affects its magnetic properties. To date, spin electronic devices have been used in digital information storage, constituting various kinds of elements of magnetic memory. Their advantage is that they can basically have an unlimited number of recording cycles because the only thing that changes is the direction of magnetisation, whereas the atomic structure remains unchanged. Additionally, by using spin currents, we may reduce heat emission from electronic devices, which means that we could eliminate the problem that keeps getting in the way of downsizing.

‘As was mentioned, the spin is connected with the magnetic properties of the materials. We know a few ways to generate a spin current – one of the simplest methods is to pass the current through a thin layer of ferromagnetic material. The problem is the distance, as the information about the electron spin is lost very quickly (in a matter of picoseconds or after travelling several nanometres). The first spin electronic device was a ferromagnet/regular metal/ferromagnet structure, for example, iron/chromium/iron. Currently, the most popular element is the so-called magnetic tunnel junction that has a ferromagnet/insulator/ferromagnet type of structure, which shows far better parameters in terms of sensitivity, power consumption, or downsizing capabilities’, Professor Skowroński describes.

Professor Witold Skowroński with a polished silicon wafer. Photo by AGH UST Spin Electronics Laboratory

Neuromorphic platforms

Magnetic tunnel junctions are almost everywhere; such sensors are found in our mobile phones, where they are used to read the direction of the Earth’s magnetic field; or in our computers, where they are part of the magnetic disc. Currently, we see the implementation of non-volatile random-access memory (otherwise known as magnetic random-access memory – not to be confused with the most recent Daft Punk music album), which are slowly pushing out DRAM, and in the long run might even replace SRAM. The chief advantage of this type of solutions is the non-volatile quality of the memory – maintaining its state requires no energy at all. Magnetic tunnel junctions can also be used in neuromorphic platforms, which, according to the scientists, may compete with other unconventional computational methodologies. Therefore, the AGH UST researchers plan to develop a new-class spintronic devices – memristors, p-Bits, and nano-oscillators – which will fill the gap between CMOS-based technologies and quantum computers.

‘Magnetic tunnel junctions can be used in neuromorphic platforms in a variety of ways. One such way is when the junction oscillates at a specific frequency after it has been powered by a direct current. When we couple several such oscillators into a small network, they can learn to recognise speech. Another way is to use thermal instability of the junction that is specifically designed so that its state randomly jumps from 1 to 0 in the time of micro- or milliseconds. Such an element is then called a p-Bit, that is, something between a traditional bit and a q-Bit. It has been shown that several of such bits combined can be used to perform prime factorisation, which can help solve np-Complete problems, such as the so-called travelling salesman problem’, claims the project leader.

The AGH UST scientists have also found another application in which the arrayed junctions constitute a quantised weight in a Hopfield network and can learn to recognise handwriting almost as precisely as a computer program. What is the advantage of spintronic neuromorphic platforms over a regular PC? Conventional computer architectures, in which memory and CPU are separated, face the so-called von Neumann bottleneck problem, that is, throughput limitation caused by inadequate rate of data transfer between these two components. This is something that unconventional computer architectures don’t have to deal with because they use mechanisms similar to neurons and synapses in the brain, the functioning of which is mirrored by the elements of spin electronics in neuromorphic platforms. The latter combine memory and computational ability, which is more efficient in terms of both the processing time and energy consumption.

Interdisciplinary adventures

To recreate brain activity, which is a complicated but energy-efficient organ, we must build artificial counterparts of nerve cells. This is where memristors come in handy (described briefly in our text on synthetic intelligence), the resistance of which might be managed in either a quantised or analogue way. As it turns out, elements of spin electronics make it possible to implement both scenarios. However, this requires proper optimisation of spintronic devices as well as development of new prototypes using innovative alternative heterostructures. Therefore, scientists keep looking for better combinations of materials in the form of several thin atomic layers, because the use of spin current – as has been mentioned above – is possible only on miniscule surfaces.

‘To design these materials, we use a well-developed technological line in clean rooms that belong to the AGH UST ACMiN. Subsequently, we have to characterise these elements – we do this in a well-equipped Spin Electronics Laboratory at the AGH UST Institute of Electronics. The IERU project money bought us, for example, a strong magnet that allows us to measure the prototypes in a broad spectrum of magnetic fields and frequencies. At the same time, we are working on applying fascinating new elements to transfer, manipulate, and detect the spin current in nanostructures. We do this by tapping into knowledge of physics, materials engineering, computer science, and electronics. All this can be found at the AGH UST!’, Professor Skowroński says.

Rotating sample station used to investigate spintronic devices in a magnetic field.
The electromagnet (from New Zealand) was bought from IERU project money.
Photo by AGH UST Spin Electronics Laboratory

***

The research described above has found its continuation in several scientific endeavours. Within the framework of a National Science Centre grant, the AGH UST engineers are to carry out a Polish-Chinese project named Sheng II, which is related to quantum-based oxide materials. Doctoral students who wish to join the team in the new academic year should notify Professor Witold Skowroński (skowron@agh.edu.pl).

The project was funded by a university grant within the framework of the “Excellence Initiative – Research University” project (the AGH UST 2020–2022, PRA-4).