How antiferromagnetism – an ultrafast magnetic order – could rewrite computer memory
Scientists achieved a breakthrough in computing: a study in Science has demonstrated a non-volatile memory element that can switch states in just 40 picoseconds while using several orders of magnitude less power than comparable ferromagnetic devices. The team also showed it can be triggered by light, which hints at a promising future where photonics and electronics could work together seamlessly.
Modern computers are getting incredibly good at computing, yet computing isn’t the only thing that matters.
The processors are getting much faster, the models – way larger, and the data centers are consuming increasing amounts of electricity. A quite large share of that does not come from arithmetic operations, but from constant movement, data storage, and retrieval – that followed by the energy required to cool the hardware that performs these tasks.
This raises a question: is there a way to make memory become both ultrafast and non-volatile, while also requiring dramatically less energy to switch between states?
And that’s the part where we talk about the research on magnetism.
May 2026, a group of Japanese scientists published a study in Science that demonstrates how non-volatile switching elements are capable of moving between two information states only using electrical pulse lasting just 40 picoseconds (that is 40 trillionths of a second, imagine!). More importantly, the state remained stable even after electric pulse was switched off completely, without further power input.
At the very heart – a compound of manganese and tin called Mn₃Sn.
On antiferromagnetism and reshaping data storage
In a ferromagnetic material, magnetic moments usually align in the same direction. This creates a strong net magnetization, which makes ferromagnetic behavior both easy to detect and control – and is the basis of many memory technologies.
In an antiferromagnetic material, magnetic moments will align in opposite directions through exchange interactions, thus creating antiparallel coupling – this collinear magnetic order is what changes everything. This creates either little or no net magnetization, even though the internal magnetic order is still well organized.
A comparison of ferromagnets vs antiferromagnets:
| A ferromagnet | An antiferromagnet | |
| Magnetic moments | Usually align | Largely cancel one another |
| Net magnetization | Very strong | Nearly vanishes |
| Data storage | In overall magnetic orientation | In collective spin arrangement |
| State switching | Only slower magnetic dynamics | Extremely fast magnetic dynamics |
At first, this might seem like a problem – if there is no strong signal, it becomes much harder to use the material in conventional memory devices. In practice, this “invisibility” is exactly what makes this material so promising for future memory technology.
But what has Mn₃Sn to do with this?
This compound is pushing the idea even further – its atoms are forming a triangular lattice arrangement in which three neighboring magnetic moments are arranged at roughly 120-degree angles. This geometry is causing the overall net magnetization to almost completely vanish while maintaining an ordered magnetic structure.
What makes this useful is that this collective magnetic state can exist in different stable orientations that can be used to represent binary information – the 0 and 1 we use in computing. This way, the information is stored in overall magnetic orientation.
Ultrafast magnetism: why it can switch that quickly
The reason why Mn₃Sn can respond on such short timescales is strong antiferromagnetic interaction between atoms that’s driven by tight exchange coupling. It does not manipulate an isolated magnetic moment but resembles a tightly coupled network in which the entire magnetic structure can reorganize.
And because of this strong coupling between atoms, the entire magnetic configuration can respond extremely quickly to stimuli, thus opening new opportunities.
But speed by itself isn’t sufficient. What’s important is switching between states and retaining the information even without electrical power while allowing that information to be read without destroying it.
The study we mentioned does address all three of these requirements within a single experimental platform.
Spintronic switching and flipping magnetic states
To achieve controlled switching, the researchers have constructed a heterostructure that’s composed of Ta (or tantalum) and Mn₃Sn. To phrase this simply, they made a structure of two thin layers of atoms (also called bilayer system).
That’s important because tantalum, a heavy transition metal, has strong spin-orbit coupling.
- As the electric current is passing the thin Ta layer, it generates a transverse spin current
- The said spin current is injected into Mn₃Sn
- The same spin current then exerts a SOT (spin-orbit torque) on the antiferromagnetic order
- The collective magnetic configuration can thus be reorganized into another stable orientation
By now, the bit is written and stored.
State stability even after power switch-off
Ultrafast switching is impressive but wouldn’t be useful without stability. Memory devices must retain the information even after the stimulus is removed.
When talking about Mn₃Sn, once the electrical pulse is gone, the system just remains in that antiferromagnetic state even within a zero magnetic field, without requiring any additional energy input. This means memory devices can store the information without continuous power consumption.
This combination – ultrafast switching, memory non-volatility, and low energy requirements – is exactly what makes this result particularly significant. In the picosecond regime, researchers observed that the switching process surprisingly required several orders of magnitude less energy than comparable ferromagnetic devices, which they have attributed to the efficient transfer of momentum into the antiferromagnetic order.
This leads to an even more intriguing question: can the same system be driven not only by an electrical current, but using optical computing?
When magnetism meets photonics
In addition, the researchers also demonstrated that the same mechanism can also be triggered using light. The experimental setup involved a telecom-wavelength laser combined with a photoelectric converter, which transformed optical pulses directly into electrical signals.
These laser-generated photocurrent pulses lasted approximately 60 picoseconds and proved to switch antiferromagnetic states as effectively as direct electrical excitation. What’s important, the underlying switching mechanism remained unchanged – the signal was converted before interacting with Mn₃Sn.
The sequence can therefore be described as follows:
- Optical signal
- Photoelectric conversion
- Electrical pulse
- Spin current generation
- Spin-orbit torque exertion
- Mn₃Sn switching
This capability is particularly relevant today.
But still, optical signals must eventually be converted back into electrical form before storage, and this additional step can introduce energy overhead and delays. Memory elements that respond to an optical signal on a timescale comparable to its switching speed therefore suggest a pathway toward more seamless integration of photonic and electronics.
In brief, it brings memory closer to speed-of-light-based data transmission before slowing it down.
How do you read antiferromagnetic bits?
Diving into antiferromagnetic memory, one of the most immediate challenges is detection. If Mn₃Sn has almost no magnetization, how can its state be read at all? The answer is the Hall effect, a different physical effect. In Mn₃Sn, it does not originate from magnetization, but instead the unusual quantum geometry of its electronic structure.
It exhibits a significant Berry curvature, which influences the motion of electrons and produces a measurable transverse voltage. As the state changes, it changes as well, thereby providing a direct electrical readout.
The full memory cycle therefore becomes surprisingly straightforward:
- The state still remains after power is removed
- The state is read via the Hall effect
And despite the complexity of the underlying physics, the outcome is familiar: write, store, and read.
Bilayer systems aren’t limited to tantalum
Although tantalum does play a very important role in the mentioned study, it is not the only material that’s capable of generating spin-orbit torque. Both platinum and tungsten also exhibit strong spin Hall effects and have been used in several related experiments involving Mn₃Sn.
This suggests future implementations will not necessarily depend on a single combination. The most important requirement will be the ability to generate and transfer spin currents into the antiferromagnetic layer.
Low power may matter even more than speed
As switching becomes faster, the currents to drive it increase, along with the associated heat dissipation. This may be manageable at the device level, but becomes a constraint when scaled to billions of cells operating simultaneously.
Every operation – moving, storing, or retrieving – now carries energy cost.
No technology is used for everything
A comparison of different memory technologies:
| Main strength | Main limitation | |
| SRAM | High speed | High cost and continuous power consumption |
| DRAM | High density | Volatile storage and constant refresh requirements |
| Flash | Non-volatile storage | Low speed |
| Mn₃Sn-based | Ultrafast, low-power, non-volatile switching | At an early stage of research |
An ideal memory technology would combine high speed, non-volatile storage, high density and durability, and low energy consumption in one single system – and Mn₃Sn has shown that several of these can coexist.
Antiferromagnetic memory: very promising, but not yet ready
It’s important to still be cautious when interpreting these results. To demonstrate a working laboratory device is different from manufacturing a reliable memory technology.
A practical real-world implementation would definitely still need to prove long-term stability, high endurance over many write cycles, good compatibility with existing CMOS processes, efficient scaling, and low energy consumption. And that across systems and not one device.
Summing up, it’s far too early to suggest antiferromagnetic memory could replace established technologies. What the current research does show is that underlying physics do allow for combinations of properties that were previously thought to be very difficult to achieve.
What are your thoughts on this?
From magnetism to spintronics: a whole new kind of bit
For decades, digital electronics have relied on moving electric charge. But the broader applications of antiferromagnetic spintronic devices further extend this idea by using electron spin rather than electric charge as the data carrier – a shift toward genuinely macroscopic control of quantum-scale states.
That means that materials like Mn₃Sn are one step forward, where information isn’t stored in simple charge states or basic magnetic alignment, but complex collective configurations of many interacting spins.
This represents a shift in how we think about memory. The question is no longer how to speed up systems, but whether entirely new physical states of matter can be really used for storage and manipulation in fundamentally different ways.
The significance of the 40-picosecond result is therefore not just its speed, but the fact that it challenges the assumption that speed, non-volatile storage, and low energy consumption are conflicting.
Wrapping up
Antiferromagnetism remains an area of research, and the long path from laboratory to commercial is uncertain. Many challenges still lay ahead before practical applications become viable.
But nonetheless, the physics are pointing to an important possibility: that memory does not necessarily require conventional magnetization or large magnetic fields, but can be based on collective quantum states of matter.
If such can indeed be switched in tens of picoseconds while retaining the information without power, then future computing architectures may access an entirely new era.
Not as a replacement for existing memory technologies, at least not yet, but potentially as an additional layer in the ongoing search for more efficient computation.