Pushing the limits of miniaturisation
In a process once thought impossible, a tiny memory chip improves as it gets smaller, potentially leading to ultra-efficient smartphones and AI systems.
The circuits and memory inside electronic devices consume energy and release heat as they operate. This is why a smartphone might heat up after heavy use, or experience an unwelcome drop in battery.
At the most basic level, computer memory stores information as 0s and 1s by controlling how easily electricity can pass through a material. If scientists can design memory that requires far less electricity, this could dramatically reduce the energy demands of phones, computers and other electronics.
The quest for low-power memory
In 1971, researchers attempted to solve the problem by proposing the ferroelectric tunnel junction (FTJ). This type of memory depends on ferroelectricity, a property in which a material’s internal electric polarisation can be switched. When this polarisation changes, it affects how easily current flows, allowing the device to store data.
Despite its promise, traditional materials used for this type of memory struggled as devices were scaled down, leading performance to drop as components became smaller.
The potential of hafnium oxide
A key advance came in 2011, when scientists discovered that hafnium oxide, a widely used material, could retain its electric polarisation even when extremely thin. Building on this finding, Professor Yutaka Majima and his team at the Institute of Science Tokyo set out to develop an extremely small memory device measuring just 25 nanometres across: approximately one three-thousandth the thickness of a human hair.
Solving leakage at the nanoscale
Shrinking memory to this scale introduces a major challenge: electrical current tends to leak through the boundaries between tiny crystals in the material. This has long prevented further miniaturisation.
Instead of trying to avoid this issue, the team decided to lean into it, by making the device even smaller. This reduced the impact of those crystal boundaries.
They also developed a new fabrication method by heating the electrodes so they naturally formed a semicircular shape. This design created a structure closer to a single crystal, meaning there were fewer boundaries where leakage could occur.
As well as achieving high performance, the device demonstrated something unexpected: the memory actually performs better as it becomes smaller, overturning a long-held assumption in electronics.
Impact on future devices
If adapted for real-world use, the technology could have far-reaching effects. Devices like smartwatches could run for months on a single charge, and networks of connected sensors might operate without needing frequent battery replacements.
In artificial intelligence, this type of memory could support faster processing while using far less energy. The researchers believe that because hafnium oxide is already compatible with existing semiconductor manufacturing, integrating this new memory into everyday electronics could happen relatively quickly.
Majima likened the act of challenging what seem to be the limits of science — such as ‘we cannot make things any smaller’ or ‘they will break if we do’— to “walking in the dark”.
“It is a continuous struggle,” Majima said. “However, by questioning traditional assumptions and exploring new ways to overcome these barriers, we were able to discover an entirely new perspective.
“I would be delighted if this achievement sparks the curiosity of young people who will shape the future and helps build a better world.”
The research was published in the journal Nanoscale.
Fungi-based semiconductors for flexible electronics
Researchers have harnessed fungi-derived materials to create flexible semiconductors, advancing...
Novel transistor enhances sensing in liquids
Researchers have developed a novel transistor sensor that maintains high accuracy in liquids,...
Photonic AI chip processes data at the speed of light
Aussie researchers have built an ultra-compact nanophotonic AI chip that performs neural network...

