A near-monochrome collage of seven laboratory photographs cut into diagonal panels: a cryostat and vacuum chamber trailing cabling and steel flanges; an optical table with posts, a glowing lamp and a circuit board; a bright ring of light seen through a dark chamber viewport; a chamber interior with a finned copper disc, translation stages and stepper motors; a rack of long cryostat probe inserts with stacked baffles; a vacuum chamber wrapped in aluminium foil with pneumatic valves; and a chamber with a brilliant glowing plume beneath a circular electrode.

Exchange bias between magnets pointing at right angles, and the domain structure behind it

The short version

Magnetic coupling across an interface is usually described between two magnets that share an axis. This work paired two whose preferred directions are perpendicular to each other, showed they still couple strongly enough to bias one another, and then imaged the magnetic structure inside the device to explain how. The direction of the effect is set by a magnetic-field pulse at operating temperature, without cooling the device through any transition.

Exchange bias is normally described between collinear magnets. Here the two are orthogonal: CrSBr is an antiferromagnet with in-plane spins, Fe3GeTe2 (FGT) a ferromagnet with out-of-plane spins, so neither can pin the other along a shared axis.

They couple regardless. A ±2.5 T out-of-plane preset field, applied at the measurement temperature rather than by cooling through an ordering temperature, sets a bias of roughly ±47 mT at 10 K whose sign follows the preset polarity, and the effect survives up to the 132 K Néel temperature of CrSBr. Setting a pinning direction without a field-cool step is the part that generalises beyond this material pair.

Cross-sectional off-axis electron holography shows what the coupling does inside the ferromagnet: CrSBr drives stripe-like flux-closure domains in FGT, with Bloch walls through the centre and Néel-type closure caps at the surfaces. Reversal then proceeds asymmetrically by domain nucleation, which a five-state model reproduces from interfacial exchange, the two competing anisotropies, and dipolar fields. Two limits are worth stating: the bias appears only on the first loop after each preset, a strong training effect, and it requires thick (≈87 nm) FGT, with no out-of-plane bias seen in the 9–12 nm regime.

Four panels. Top left: an optical micrograph of the device with coloured outlines labelling the FGT, CrSBr and h-BN flakes over blue and gold contacts. Top right: two Hall-voltage hysteresis loops against out-of-plane field between minus 400 and 400 millitesla, a blue loop shifted to negative field after a plus 2.5 tesla preset and a red loop shifted to positive field after a minus 2.5 tesla preset. Bottom left: a greyscale cross-sectional electron micrograph with a 20 nm scale bar showing a platinum capping layer over a CrSBr layer over an FGT layer, separated by dotted lines. Bottom right: a colour-coded magnetic induction map of the same cross-section with a 100 nm scale bar, the CrSBr, FGT and SiO2 layers marked by dashed yellow lines and the crystal axes and colour wheel shown alongside; within the FGT band a row of concentric closed contour loops in green, red and magenta marks circulating flux-closure domains, with the contours in the CrSBr band above running roughly parallel to the interface.

The device, the two oppositely shifted hysteresis loops that define the switchable bias, the cross-section, and the holographic induction map. The concentric loops inside the FGT layer are the circulating flux-closure domains themselves: colour gives the direction of the magnetic induction, so each closed set of contours is magnetisation rotating through a full turn in the plane of the cross-section.

Sole first author. Published in Small 21, e06284 (2025), 10.1002/smll.202506284. Analysis code: EB-in-FGT_CrSBr-vdW-heterostructure.