Idea 12/Materials & Chemistry/Device proposal/3 min

Altermagnetic Analog Weights

A magnet with no net magnetisation cannot leak field into its neighbour. That makes it a candidate substrate for dense analog memory.

Altermagnets decouple magnetic order from stray fields, so their domain structure can encode analog weights at densities where magnetic tunnel junctions run out of crosstalk budget.

01What changed

In late October 2026, SLAC researchers reported proof of altermagnetism, a magnetic phase with momentum-dependent spin splitting and zero net magnetisation, following confirmation of the effect deep inside a material by the Cavendish Laboratory.1 SLAC, 29 October 2026
The defining property is that spin splitting alternates in sign across the Brillouin zone, so the material orders magnetically while producing no macroscopic dipole field.2 Cavendish Laboratory, 2026

02The gap

Altermagnetism is being pursued for digital spintronics, essentially as a better MRAM bit. Nobody is using the absence of a stray field for the problem it actually solves: analog crossbar arrays, where neighbouring cells bleed into each other and cap density.

03The idea

Use the Néel vector orientation, not its magnitude, as the state variable. In a thin altermagnet film with strong magnetoelastic coupling, a piezoelectric substrate can rotate the order parameter by strain rather than by current, and partial rotation gives intermediate states.

Because the net magnetisation is zero, adjacent cells do not see each other's field. The dominant crosstalk channel disappears, and pitch is limited by lithography and thermal diffusion rather than by magnetostatics. This is a structural, not incremental, advantage over MRAM.

Read-out uses the transverse spin current or the anomalous Hall response, both of which scale with the projection of the Néel vector onto the read axis, so the analog level is a geometric quantity and should be robust to film thickness variation.

The honest risk is that strain-written states relax. The decisive measurement is retention at 400 kelvin over a thousand hours, and endurance beyond a billion switches, on a patterned 8 by 8 array at 50 nanometre pitch.

04Why now

The phase was confirmed experimentally weeks ago, and the thin-film growth recipes needed for device work are being published now. There is a narrow window before the field converges on digital applications.

05The decisive test

Do this

Pattern an 8 by 8 array of a confirmed altermagnet on a piezoelectric substrate, demonstrate at least 16 resolvable analog levels, and measure crosstalk, retention and endurance against a magnetic tunnel junction baseline at matched pitch.

Success looks like

Crosstalk at least ten-fold lower than the junction baseline at 50 nanometre pitch, with retention above a thousand hours at 400 kelvin and no more than five percent level drift.

Kills it

If strain coupling yields only two stable states at room temperature, the material is digital and analog encoding is impossible.

06What would break it

The confirmed altermagnets may not be growable as device-grade thin films.
Strain writing may be too slow for anything but inference weights.
Zero net magnetisation does not mean zero noise; thermal fluctuations still set a floor.

07If it works

Analog in-memory compute gets a substrate where density is not limited by magnetostatics, which is the main physical ceiling on today's crossbar approach.