Researchers have developed a technique that could enable the extreme miniaturization of computing components, paving the way for compact and high-performance devices.
The smaller the transistors and logic gates in a processor, the more computing power can be packed into a smaller area. But the physical constraints of silicon mean we are reaching the limits of how small these components can be.
However, a new technique, involving ultrafast switching between spin states in 2D magnets — to represent the switching between the binary states of 1 and 0 — can lead to much denser and more power-efficient components.
This technique is enabled by a new type of magnetic tunnel junction (MTJ) — a material structure that acts as a data storage device in a computing system. The scientists sandwiched chromium triiodide (a 2D insulating magnet) between layers of graphene and sent an electrical current through it to dictate the magnet's orientation within the individual chromium triiodide layers.
Harnessing these MTJs could mean packing more computing power into a chip than was previously deemed possible — while consuming much less energy during the switching process. The researchers published their findings in a new study published May 1 in the journal Nature Communications.
In the paper, the scientists demonstrated that 2D magnets can be polarized to represent binary states — the 1s and 0s of computing data — paving the way for highly energy-efficient computing.
Harnessing spintronics for faster computing
Precisely controlling the magnetic phase of 2D materials is a crucial step in spintronics (controlling an electron’s spin and the associated magnetic moment). By precisely controlling the current, the new technique can change the spin states in chromium triiodide using the current's polarity and amplitude. This is possible because the compound is ferromagnetic (it is magnetic and can attract magnets in a similar way to iron). This compound is also a semiconductor — a material that has a conductivity that falls between a metal and an insulator.
A key enabling component for spintronics is the MTJ — two ferromagnetic layers separated by an insulating barrier. Controlling an MTJ’s spin state is a technique that is already used in various computer components, such as the read heads of hard drives. But precisely controlling the thickness of its constituent layers and their quality of their interfaces with each other has proved challenging.
Related: 'Crazy idea' memory device could slash AI energy consumption by up to 2,500 times
Materials must withstand the high current densities of at least 10 million amps through an area approximately the size of fingernail — but also meet the demands of device miniaturization and energy efficiency. For comparison purposes, a typical bolt of lightning is 1,000 to 300,000 amps.
"This paper is about the fact that you can have two possible states of the tunneling current; spin-parallel and anti-parallel," Adelina Ilie, a reader in physics at the University of Bath in the U.K. specializing in 2D magnets, told LiveScience. "If there are two defined states, they can be used as logic gates in a computer."
Much greater energy efficiency for future AI systems
The scientists created the 2D van der Waals (chromium triiodide) magnets, then layered atomically thin flakes of graphene, hexagonal boron nitride and chromium triiodide on top of each other to form tunnel junction devices — which they chilled to near absolute zero. They simultaneously passed an electrical current through the material and measured it using a sourcemeter in 16-millisecond bursts.
They noted that the voltage underwent random switching between the levels, corresponding to the spin-parallel and spin-antiparallel states within chromium triiodide, with the switching direction determined by the polarity and amplitude of the current. The duration for each magnetic state was typically 10 milliseconds, while the switching time between the two states was in the order of microseconds (a microsecond is onemillionth of a second).
"These states are not exactly stable," explained Ilie. "What actually happens is that the current goes from one state to another, back and forth stochastically, but the average of time it stays more in one state or another, depending on the voltage. This gives us two states that we can select deterministically."
The two states, which can be used as logic gates, enable operation at a much smaller scale than was previously possible. Using this technology, manufacturers could create computer chips with greater processing power. But the need for near absolute-zero operating temperatures means implementing futuristic devices practically would be challenging.
"What makes this kind of work different is that it looks like the energy needed to go from one state to another is a magnitude lower than in conventional magnetic tunnel junctions," concluded Ilie. "With new technologies like generative AI, which increase power consumption tremendously, it won't be possible to keep up, so you need devices that are energy efficient."
