Spin-spiral magnets, in which the spin direction rotates from site to site, are promising for spintronics without stray fields, but such states are hard to detect in two-dimensional materials. Researchers at Aalto University show that machine learning can extract the spin-spiral wave vector from ordinary lateral transport measurements.

The algorithm is trained on how the conductance depends on magnetic field and bias, and recovers the wave vector of arbitrary spin spirals, even with impurities and noise in the data. The work offers a new way to learn magnetic structures directly from transport experiments. It brings together two QMAT groups, Jose Lado's Correlated Quantum Materials group and Zhipei Sun's Photonics group, with Fedor Nigmatulin and Greta Lupi.

The preprint has not yet been peer reviewed.

Image: QMAT