Smaller, cooler, smarter: Rethinking memory for a low‑power future

A new kind of memory device could solve the problem of overheating and battery drain in electronics. This is by shrinking components to an extreme scale and redesigning their structure. Here, researchers have found a way to reduce energy loss instead of increasing it.

The result is a tiny memory unit that improves as it gets smaller—something once thought impossible. This could pave the way for ultra-efficient smartphones, wearables, and AI systems.

If your smartphone ever feels warm after streaming video or rapidly switching between apps, you are experiencing a fundamental limitation of modern electronics. Every calculation, every saved file, every message depends on microscopic electronic circuits that consume energy—and release heat as a by-product.

At the heart of this problem lies computer memory. From phones and laptops to the servers powering artificial intelligence, memory systems rely on moving electrical charge to represent binary information: the familiar 0s and 1s. The more energy required to switch between those states, the more power is consumed—and the more heat is generated.

Now, researchers are revisiting an idea more than half a century old to tackle this inefficiency from the ground up.

A Concept Ahead of Its Time?

In 1971, scientists proposed a new type of memory called the ferroelectric tunnel junction (FTJ). Unlike conventional memory, which stores information through charge accumulation, FTJs rely on a property called ferroelectricity.

Ferroelectric materials possess an internal electric polarization that can be switched between two states. Crucially, this switching alters how easily electrons can pass through the material. By toggling between these states, the device can encode binary information—without needing as much current flow.

In principle, this approach offers a major advantage: dramatically reduced power consumption. Less current means less heat, and greater efficiency.

But for decades, FTJs remained largely experimental. The main obstacle was scaling. As devices were made smaller—an essential step for modern electronics—the materials used for FTJs began to lose their desirable properties. Performance declined, and the concept struggled to compete with established technologies.

A Breakthrough Material

The turning point came in 2011, when researchers discovered that hafnium oxide—a material already widely used in semiconductor manufacturing—could exhibit ferroelectric behaviour when engineered at extremely small scales.

This was unexpected. Hafnium oxide had long been valued for its insulating properties in transistors, but its ability to maintain electric polarization in ultra-thin layers opened a new path for memory design.

Building on this insight, Professor Yutaka Majima and colleagues at the Institute of Science Tokyo have pushed the concept to its limits, developing a ferroelectric memory device just 25 nanometres across—around one three-thousandth the thickness of a human hair.

Rethinking the Nanoscale Problem

Shrinking electronic components to the nanoscale introduces a familiar challenge: leakage. At such small dimensions, the boundaries between microscopic crystal domains become weak points, allowing unwanted current to slip through. This leakage has historically limited further miniaturisation.

Rather than trying to eliminate these imperfections, Majima’s team took a counterintuitive approach. They made the device even smaller.

At this reduced scale, the influence of grain boundaries diminishes, and the structure behaves more like a single crystal. To enhance this effect further, the team developed a novel fabrication technique. By heating the electrodes during manufacturing, they caused them to form a smooth, semi-circular shape, improving the uniformity of the material.

The result is a structure with fewer defects and reduced leakage pathways—a solution that turns a longstanding problem into a design advantage.

When Smaller Becomes Better

The most striking outcome is not just that the device works—it is that it performs better as it shrinks.

This directly challenges one of the core assumptions in electronics: that miniaturisation inevitably leads to diminishing returns and increased instability. Instead, the researchers have demonstrated that, at least in this case, reducing size can enhance both efficiency and functionality.

Such behaviour could prove transformative. If memory can be made smaller and more energy-efficient at the same time, it may help sustain the long-term trend of improving computing performance without unsustainable increases in power consumption.

Implications for Everyday Technology

The potential applications are far-reaching. Devices such as smartphones, smartwatches and wearable sensors could operate with significantly lower energy demands. In some cases, battery life might extend from days to weeks or even months.

For large-scale computing systems—particularly those used in artificial intelligence—the benefits could be even greater. AI workloads are notoriously energy-intensive, requiring vast data centres that consume significant electricity. More efficient memory could reduce this footprint, enabling faster processing with less power.

Importantly, hafnium oxide is already compatible with current semiconductor manufacturing processes. This means the transition from laboratory research to commercial application could be relatively rapid, compared to entirely new materials systems.

Rethinking Limits

For Majima, the work is as much about challenging assumptions as it is about technological progress.

Scientific and engineering limits, he suggests, are often treated as fixed boundaries. But they may instead reflect gaps in understanding—gaps that can be overcome by rethinking the problem.

His team’s success illustrates this principle. Instead of accepting that miniaturisation would degrade performance, they asked whether shrinking even further might produce a different outcome.

The answer, it turns out, is yes.

A Low‑Energy Future for Computing?

As digital technologies continue to expand—from personal devices to global AI infrastructure—the demand for energy-efficient computing is becoming increasingly urgent. The environmental and economic costs of powering these systems are climbing rapidly.

Advances in memory design, such as ferroelectric tunnel junctions based on hafnium oxide, offer a glimpse of a different future—one where computing power grows without a corresponding surge in energy use.

There is still work to be done before such devices become widespread. Scaling production, ensuring durability and integrating with existing architectures will all present challenges.

But the principle has been established: sometimes, the path forward lies not in working around limitations, but in redefining them entirely.