New ‘spin’ on high-speed processors
Artificial intelligence faces an energy crisis, stemming from a physical traffic jam inside modern computer chips.
Processors must continually shuffle data, such as the billions of parameters in complex models, between separate computing and memory nodes. This traffic jam, famously known as the “von Neumann bottleneck”, hinders the speed and energy consumption of advanced processors.
To tackle this problem, scientists are developing spintronics, which leverages the electron’s “spin”, or its intrinsic magnetic orientation, for more efficient devices. A long-sought milestone of this field is a single device, known as a “spin transistor”, that combines a magnetic bit with a semiconducting switch, allowing it to compute and store data simultaneously.
Professor of Physics Brian Zhou (Lee Pellegrini)
“The major challenge is understanding how magnetism and electrical current interact in nanoscale devices,” said Boston College Professor of Physics Brian Zhou, whose group led the study. “We developed a single-spin quantum microscope to observe magnetic states inside atomically thin devices as they actively process electrical information.”
This quantum-enabled perspective allowed the team led by Boston College researchers to introduce a conceptual shift in how magnetic transistors can be engineered. In a new study published in Physical Review Letters, the team demonstrated an architecture using the magnetic semiconductor, chromium sulfur bromide (CrSBr), that allows signals to be switched either by magnetism or voltage.
“Traditionally, these devices require joining two different materials together: a magnet and a semiconductor,” said Professor Zdeněk Sofer, a materials synthesis expert at the University of Chemistry and Technology, Prague. “By engineering a single van der Waals crystal, CrSBr, that inherently possesses both semiconducting and magnetic properties, we eliminate losses at interfaces entirely.”
Zhou’s research group fabricated the transistor using two-layer-thick CrSBr, placing laterally separated electrodes on opposite layers. Their unique construction forces current to travel both across and between the two magnetic layers. The device can be switched “on” and “off” either by changing the voltage on a nearby gate electrode or by changing the relative magnetic orientations of the two CrSBr layers, similar to a traditional complementary metal-oxide-semiconductor (CMOS) transistor, but with an added magnetic twist.
The interior of a spin transistor during switching is captured by a scanning quantum sensor developed by Boston College researchers. (Brian Zhou)
To characterize the spin transistor, the researchers combined electrical measurements with a high-resolution quantum sensing probe, scanning nitrogen-vacancy (NV) center magnetometry. This specialized imaging technique maps the local magnetic field by tracking variations in the magnetic resonance of a single atomic defect. The quantum microscope vividly revealed the correlated magnetic and electrical behavior of the device: how spatial changes in the magnetization modify the device conductance, and how gate voltage flips the magnetic layers between parallel and antiparallel states.
A key factor that enhanced the performance of the developed spin transistor was the researchers’ ability to access “space-charge-limited” conduction. Within this regime, the internal build-up and mutual repulsion of charges in the material alter the current-voltage relationship, causing it to display a power law scaling instead of the conventional linear, ohmic behavior.
“The rapid power law scaling allows us to dramatically tune the conductivity,” said graduate student Thomas K. M. Graham, the lead author of the study. “Our device achieves an electrical on/off ratio of a million percent and a magnetic on/off ratio of 3000 percent, the latter significantly higher than previous efforts.”
By merging switching logic with a nonvolatile memory bit, the developed architecture paves the way for ultra-efficient, “instant-on” processors that do not need to fetch data from memory, alongside reconfigurable computing circuits that can be reprogrammed after manufacturing.
To realize this potential, Zhou says researchers must continue to advance nanoscale imaging techniques and the electrical control of magnetic states.