Boston College Researchers Develop CrSBr Spin Transistor for Efficient AI

Modern computing is currently trapped by a physical traffic jam. In today’s standard processor architectures, data must be constantly shuttled between separate memory and computing nodes—a limitation famously known as the von Neumann bottleneck. This physical separation creates a significant energy drain and restricts processing speeds, becoming an increasingly difficult hurdle as artificial intelligence models grow to require billions of parameters.

A research team at Boston College, led by Professor of Physics Brian Zhou, has proposed a solution rooted in spintronics, a field that utilizes the electron’s spin, or its intrinsic magnetic orientation, for more efficient devices. Their findings, published in Physical Review Letters, detail a new transistor architecture that combines a magnetic bit with a semiconducting switch.

Engineering the CrSBr Spin Transistor

The team’s breakthrough centers on the use of chromium sulfur bromide (CrSBr), a material that functions as both a semiconductor and a magnet. By utilizing this single van der Waals crystal, the researchers eliminated the need to join two distinct materials at an interface, which traditionally results in performance losses.

“By engineering a single van der Waals crystal, CrSBr, that inherently possesses both semiconducting and magnetic properties, we eliminate losses at interfaces entirely.”

Professor Zdeněk Sofer, materials synthesis expert at the University of Chemistry and Technology, Prague

To fabricate the device, Zhou’s group used two-layer-thick CrSBr, placing electrodes on opposite sides to force current to travel both across and between the magnetic layers. This construction allows the transistor to be switched on and off through either voltage adjustments on a nearby gate or by altering the relative magnetic orientations of the layers.

Performance Metrics and Quantum Sensing

To verify the device’s functionality, the researchers employed a specialized high-resolution quantum sensing probe, specifically nitrogen-vacancy (NV) center magnetometry. This allowed them to map local magnetic fields by tracking variations in the magnetic resonance of a single atomic defect. The process revealed how spatial changes in magnetization modify conductance and how gate voltage successfully flips the magnetic layers.

The device’s performance was bolstered by a phenomenon known as space-charge-limited conduction. In this state, the mutual repulsion of charges within the material causes the current-voltage relationship to follow a power law scaling rather than traditional linear behavior.

“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.”

Thomas K. M. Graham, lead author and graduate student

Comparing Traditional and Quantum-Enabled Architectures

The transition from traditional hardware to this new architecture represents a fundamental shift in how computing hardware handles data. While standard CMOS transistors are limited by their reliance on external memory, the CrSBr architecture offers an integrated approach.

Future Directions for AI Hardware

The ability to merge switching logic with a nonvolatile memory bit could pave the way for reconfigurable computing circuits—processors that can be reprogrammed after manufacturing. Such hardware would not require the constant data-fetching cycles that define current computing limitations.

However, the transition from experimental device to consumer hardware remains a long-term goal. As Professor Brian Zhou noted, reaching this potential requires continued advancement in both nanoscale imaging techniques and the precision of electrical control over magnetic states.

Sources: Miragenews, newsy-today.com.