A memory bit inside a magnet that cancels itself.
Ferromagnetic memory stores information in magnetic states that can be read electrically. Antiferromagnets seem less cooperative: their local atomic moments largely cancel, leaving no net magnetization. A paper published in Nature on 21 October 2026 reports a surprising exception. Researchers led by Northwestern University demonstrated a nanoscale antiferromagnetic tunnel junction that can be switched by an electric current and read by its resistance. The result is an experimental device and a new physical mechanism—not a finished memory chip.
Zero net magnetization is not zero magnetic structure.
In an ordinary ferromagnet, many atomic moments align. In an antiferromagnet, moments oppose one another, cancelling the overall magnetic moment. Yet the internal arrangement can remain highly ordered. In the noncollinear platinum–manganese alloy studied here, that structure can be represented by a cluster magnetic octupole. This is a description of the symmetry of a complex spin pattern, not an extra physical magnet hidden inside the material. Electron transport can respond to this pattern even when the total magnetization is zero.
The three-layer device.
The researchers fabricated junctions consisting of PtMn3 | MgO | PtMn3: two antiferromagnetic electrodes separated by a very thin magnesium-oxide barrier. Electrons can tunnel quantum-mechanically through that barrier. Their transmission depends on the relative magnetic states of the electrodes, producing a measurable resistance difference. The paper reports a room-temperature tunnelling magnetoresistance value of 363%. That number describes resistance contrast between states; it does not mean 363% energy efficiency, and it does not establish commercial memory density. The team examined nanoscale junction pillars, including diameters from 50 to 200 nanometres.
How can a spin-neutral current push a magnet?
Conventional spin-transfer torque uses spin-polarized current to reorient a magnetic layer. A current through an antiferromagnet can be spin-neutral overall, so a substantial corresponding torque was once considered unlikely. The new experiments show current-driven switching and name its mechanism octupole-driven spin-transfer torque, or OTT. The researchers explain it using unequal spin currents between matching and different magnetic sublattices across the junction. Those microscopic contributions can exert a net torque even when the simple sum of spin polarization vanishes. The result expands the physics available to spintronic device designers.
One experiment, two essential memory functions.
A memory device must support both writing and reading. In this system, OTT changes the magnetic configuration; tunnelling magnetoresistance provides an electrical signal that distinguishes configurations. Demonstrating both at room temperature is important because an exotic material that can only be written or only be read is not a complete memory element. Still, the experiment did not deliver a packaged array, manufacturing process, commercial controller or system-level endurance qualification. The scientific achievement is an electrically writable and readable nanoscale junction.
The less glamorous number engineers must examine.
The study reports switching current densities of the order of 10 megaamperes per square centimetre. Current density alone cannot determine switching energy. Actual energy per operation depends on junction area, voltage, pulse duration, wiring and peripheral electronics. Reliability also depends on retention time, thermal fluctuations, repeated cycling and fabrication variability. Strong resistance contrast is encouraging, but it is not a substitute for these measurements. Comparing only the 363% figure with other memory technologies would be misleading.
Why terahertz appears in the discussion.
Antiferromagnetic materials can have very fast intrinsic magnetic dynamics. That makes them candidates for future high-frequency spintronic systems and potentially densely scaled memory. The Nature paper identifies room-temperature terahertz technologies as a possible long-term direction. It does not demonstrate a terahertz memory circuit, a processor running at terahertz clock rates or a commercially competitive energy-per-bit result. Those outcomes require separate experiments. The present evidence concerns room-temperature switching, electrical readout and a theoretical explanation of the newly observed torque.
The next decisive tests.
To move toward an engineering product, researchers should report switching-energy distributions, pulse-width limits, bit-error rates, retention under heat, write endurance, device yield and tolerance to magnetic or electrical interference. They should compare against established ferromagnetic magnetic-tunnel-junction memory under matched operating conditions. The PtMn3 layers and MgO interface must also be manufactured reproducibly at scale. Independent replication will help establish which material structures preserve the octupole-driven effect and which lose it when fabrication conditions change.
The TETRIUM perspective: useful physics can hide in a cancellation.
A total that adds to zero can conceal a structured system rich with useful interactions. The magnetic moments in these antiferromagnets cancel, but their geometry still influences electrons and can be controlled electrically. This is the exciting part of the discovery: it enlarges the set of physical mechanisms available for future electronics. A careful technology roadmap would connect that new physics to measured power, reliability and manufacturability. The experiment proves an important effect; the next generation of engineering must prove its practical value.
Sources & further reading
Primary research and supporting publications behind this journal entry.