Last Updated on May 23, 2025 by Sushanta Barman
In a world-first demonstration, scientists have managed to keep an entangled group of atoms continuously squeezed and use them to track magnetic fields in real time—all with the help of machine learning. This milestone in quantum sensing, achieved by a collaboration between physicists in China and Denmark, could revolutionize how we detect elusive signals in physics, medicine, and beyond.
The research, published in Nature Physics on April 8, 2025, was conducted by teams from Fudan University, Shanxi University, and the Niels Bohr Institute at the University of Copenhagen.
At the heart of this innovation lies a quantum effect called spin squeezing, which manipulates the collective quantum state of billions of atoms. In typical conditions, atomic spins are randomly oriented, like tiny compass needles pointing in all directions. But in a spin-squeezed state, these quantum needles are aligned more precisely than classical physics allows, reducing noise and enhancing measurement precision.
This isn’t new in itself. What is new is keeping that delicate quantum state alive continuously—and using it to sense magnetic fields that change over time. Until now, such squeezed states could only be used in brief bursts.
The team used a cloud of about 40 billion rubidium-87 atoms, trapped in a tiny, paraffin-coated glass cell. The atoms were kept spinning in unison by a constant laser beam (optical pumping) and probed by a strobing laser beam that didn’t disturb them too much—an approach called quantum non-demolition (QND) measurement.
As the atoms precessed like spinning tops in a weak magnetic field, their alignment was subtly changed by the field. This tiny rotation was detected by measuring the modulation of the probing laser, although real-time interpretation of the resulting noisy signals remains a significant challenge.
That’s where deep learning (DL) came in. The researchers trained a neural network—similar to those used in voice recognition—to decode the optical signals and reconstruct the magnetic field causing them.
The model learned to track several types of time-varying magnetic fields, including random pulses, white noise, and even non-Gaussian processes. The AI didn’t need to know the physics behind the system—it simply learned from the data how to connect cause and effect.
To prove the quantum advantage, the researchers deliberately removed the squeezing and showed that their sensor became less sensitive. With squeezing, the sensitivity reached 27.97 fT Hz⁻¹ᐟ², beating the standard quantum limit.
Even more impressively, the squeezed state could be maintained for over 24 hours, limited only by equipment stability, not the atoms themselves.
“The achieved degree of steady-state spin squeezing was −3.23 ± 0.24 dB when conditioned upon the full measurement records and −1.63 ± 0.19 dB when conditioned only upon earlier measurements,” they report—a level good enough to show a clear quantum edge.
This fusion of quantum physics and machine learning opens doors to practical quantum sensors that operate continuously, not just in carefully timed bursts. Applications could include brain imaging, submarine detection, or even probing the subtle gravitational ripples of the universe.
The technique could also be extended to other quantum systems—such as nuclear spins, nitrogen-vacancy centers in diamond, or mechanical oscillators—broadening its impact across quantum technology.
In short, this research represents a critical leap toward real-world, always-on quantum sensing—where AI doesn’t just enhance measurement but enables it.
“We demonstrate that concurrent steady-state spin squeezing and sensing are possible,” the paper concludes. And with that, a new era of quantum metrology begins.
Read the full research paper: J. Duan et al., Nat. Phys. 21, 870–879 (2025).
I am a senior research scholar in the Department of Physics at IIT Kanpur. My work focuses on ion-beam optics and matter-wave phenomena. I am also interested in emerging matter-wave technologies.
