The most sensitive magnetic-field sensors ever built have a serious logistical problem. Superconducting quantum interference devices (SQUIDs) only work near absolute zero and need elaborate cryogenics. Atomic magnetometers of the spin-exchange relaxation-free (SERF) type demand near-perfect shielding from external magnetic interference. Both constraints push the sensor away from whatever it is trying to measure, which is bad news for anyone hoping to detect the faint magnetic signals of the human brain.
A team led by Wei Ji of Peking University, working with Dmitry Budker and colleagues in Mainz and Berkeley, has now demonstrated a different kind of magnetometer: a tiny magnet that floats in a vacuum chamber about the size of a Tupperware box and reads magnetic fields at the femtotesla scale, ten billion times weaker than Earth’s magnetic field. The work appears in Science.
The sensor magnet is less than one millimeter across, a disk roughly 410 micrometers in radius and 380 micrometers thick, held in midair by the balance of two forces. A stack of 80 disk magnets above it pulls it upward against gravity, while a plate of pyrolytic graphite below it, a diamagnetic material that generates its own opposing field, pushes it up and stabilizes it. A laser reflects off the polished surface of the floating magnet onto a position-sensitive detector, so any tilt caused by an external field is read out optically.
Because the magnet has no mechanical contact with anything, it is isolated from the vibrational noise that plagues ordinary instruments. Because it is ferromagnetic, its spins are intrinsically aligned at room temperature, with a spin density around 10^22 spin quanta per cubic centimeter, roughly seven orders of magnitude higher than the rubidium vapor used in atomic magnetometers. That dramatically suppresses the spin-projection noise that limits other designs. Gilbert damping, the mechanism by which a ferromagnet relaxes disturbances in its magnetization, averages out individual spin fluctuations within nanoseconds, so the sensor effectively measures a clean collective signal.
The engineering details reveal how demanding the femtotesla regime is. The team replaced an aluminum vacuum chamber with a glass one because eddy currents in the metal generated Johnson noise that contaminated the measurement. They ground the pyrolytic graphite stabilizer into a powder and bonded it with epoxy, confining eddy currents to small loops instead of large ones and cutting dissipation dramatically. The torsional mode of the levitated magnet resonates at about 305 hertz with a quality factor near 12,000, a remarkably low-loss oscillation.
The result is a sensitivity of 32 femtotesla per square-root hertz at resonance, at room temperature and in the presence of Earth’s field, tolerating residual fields up to the millitesla range. That matches the performance of SQUIDs and SERF magnetometers while dispensing with their requirements. The resonance frequency can also be tuned dynamically by adjusting the levitating field, which enables closed-loop operation over a wider frequency range.
The motivation is partly biomedical. Magnetic signals from the brain’s electrical activity are extraordinarily weak, and the closer a sensor can get to the scalp, the stronger the signal it sees. The levitating design reduces the sample-to-sensor distance to a few hundred micrometers, an order of magnitude closer than cryogenic systems permit. The same physics applies to cardiac signals and to fundamental physics: the group has long been interested in using precision magnetometers to search for dark matter candidates and exotic spin-dependent forces, and a levitated ferromagnet is an unusually clean platform for those searches.
There are caveats. The femtotesla-level sensitivity is achieved only at the torsional resonance; broadband sensitivity away from it is far more modest, on the order of 200 picotesla per square-root hertz at higher frequencies. The noise budget is dominated by Johnson noise from surrounding conducting materials, so the engineering of the chamber and stabilizer matters as much as the physics. At the femtotesla scale, the group notes, even a thin film of aluminum foil introduces 100 femtoteslas of noise. And while the magnetometer rivals established technologies in sensitivity, it has not yet been demonstrated on a living brain.
The resonance-based operation also shapes what the sensor can and cannot do. The magnetometer is most sensitive near its torsional resonance of roughly 305 hertz, a narrow band; brain signals, however, span a broad spectrum, much of it at lower frequencies. The team demonstrated that the resonance can be shifted from about 260 to 318 hertz by adjusting the levitating field, which points toward a closed-loop mode that tracks a signal as it drifts. Whether the approach can be extended to the low-frequency, broadband regime where most biomagnetic signals live is a key open question, and the answer will determine how quickly the device moves from physics demonstration to biomedical instrument.
Still, the combination of ambient-temperature operation, tolerance of Earth’s field, and extreme proximity to the sample makes this an unusually practical path toward wearable or bedside magnetometry. The next steps are to extend the bandwidth and demonstrate a real biological measurement. If those succeed, the levitating compass needle could give neuroscience a new window into the brain’s electrical life, without a single liter of liquid helium.
Sources: Ji, W., Xu, C., Qu, G. & Budker, D. Levitated sensor for magnetometry in ambient environment. Science 393, 607-610 (2026). DOI: 10.1126/science.adx1707. Preprint: arXiv:2504.21524. Nature News, August 6, 2026.