This Transistor Remembers What It Computes, Pointing Beyond Today’s Power-Hungry Chips

Modern computers spend an enormous amount of time and energy moving data between the processor and memory. A new experimental transistor suggests that those two jobs may not always need separate hardware.

The prototype reported an electrical on-off ratio of 1,000,000 percent and a magnetic ratio of 3,000 percent under experimental conditions.

The traffic jam inside every computer

Processors calculate while memory stores. That division has served computing for decades, but constantly carrying information between the two consumes energy and creates delays. The problem becomes especially visible in artificial intelligence, where chips repeatedly move billions of model parameters.

Engineers call this the processor-memory or von Neumann bottleneck. One possible escape is logic-in-memory: devices that can calculate and preserve a state in the same place.

One unusual crystal does two jobs

The Boston College-led team used chromium sulfur bromide, written CrSBr, a material that is both semiconducting and magnetic. They created a junction only one and two atomic layers thick, with electrodes arranged so current travels across and between the layers.

Voltage can change how easily the current flows, as in an ordinary transistor. Magnetic alignment between the layers supplies another control. Because the device can settle into multiple persistent conductance states, it also behaves partly like a memory element.

Using one crystal avoids some of the losses and fabrication problems that can occur when engineers join a separate magnet to a semiconductor.

A quantum microscope showed what was happening

Measuring current alone would not reveal why the device switched. The researchers therefore scanned it with a nitrogen-vacancy quantum sensor, a tiny defect in diamond whose spin responds to nearby magnetic fields.

The microscope mapped magnetic domains while electrical measurements tracked conductance. This linked changes in the current to changes in the magnetic structure and exposed how domain walls moved during switching. It gave the researchers a view inside an active transistor rather than an inference made only from its external output.

Why this could matter for AI hardware

If logic and nonvolatile memory can eventually share a device, future processors might retrieve less data from distant memory and waste less energy when idle. Persistent states could also allow instant-on circuits and hardware that can be reconfigured after manufacturing.

The work is especially relevant as AI accelerators run into power, heat and bandwidth limits. It does not offer a finished AI chip, but it explores a different physical architecture for reducing the movement that makes current systems expensive.

The real breakthrough is the view inside

The device itself is promising, but the measurement technique may be equally important. Being able to watch magnetism evolve while a transistor operates can expose failure modes and guide new designs much faster than electrical tests alone. The next boundary is not merely making a smaller switch. It is learning how computation and memory can become different behaviours of the same material.