Quantum Magnetometry

In 1962, a 22-year-old PhD student at Cambridge named Brian Josephson predicted that a pair of Cooper pairs (superconducting electron pairs) could tunnel right through a thin insulating barrier between two superconductors, carrying current without any applied voltage. His supervisors didn't believe him. The experimentalists did. Josephson won the 1973 Nobel Prize in Physics for the prediction, and the device that exploits it is called a Josephson junction.

Two years after Josephson's paper, a team at Ford Research Labs (Jaklevic, Lambe, Silver, Mercereau) wired two Josephson junctions into a superconducting loop. The result was the first SQUID: Superconducting Quantum Interference Device. When magnetic flux threads the loop, the quantum phase of the superconducting wavefunction on each side of the junctions shifts, and the current through the device oscillates with flux.

Here's the useful part. The oscillation period is one magnetic flux quantum, h/2e, about 2.07 femtowebers. A SQUID can resolve a fraction of one flux quantum routinely. That translates to sensitivities of a few femtoteslas per root hertz in the best commercial devices. In plain numbers: a SQUID can detect a magnetic field change equal to a billionth of the Earth's field in a single second of averaging.

The catch is temperature. A SQUID needs superconductivity to work, which means a liquid-helium bath at 4 kelvin (low-temperature SQUIDs) or liquid nitrogen at 77 kelvin (high-temperature SQUIDs). That's why clinical magnetoencephalography machines look like dental chairs with a giant dewar bolted to the top. The dewar is a thermos flask the size of a car engine, holding 100 litres of liquid helium to keep the sensor array cold enough to measure your thoughts.

Where do real magnetic signals sit on the scale?

Fourteen orders of magnitude between a bar magnet and the noise floor of a SERF magnetometer. That's the span quantum sensors have to cover to be useful at the weak end. The trick isn't measuring absolute field; it's rejecting everything above the signal you care about.

Shielding handles some of that. A mu-metal room attenuates ambient fields by factors of 10,000 or more, which is how clinical MEG works. Gradiometer geometry handles the rest: if you put two sensors close together and subtract their readings, any uniform field cancels out and only the local gradient (the thing you want) survives.

SQUIDs have dominated high-sensitivity magnetometry for sixty years, but liquid helium is expensive and inconvenient. So physicists have spent decades hunting for a room-temperature sensor that matches SQUID performance.

The winner, for now, is the optically-pumped magnetometer. It uses a glass cell filled with rubidium or caesium vapour. A laser polarises the atomic spins so they all point the same way. When a magnetic field disturbs them, the spins precess, and a second laser reads the precession as a change in the atoms' transparency. No cryogenics. No exotic materials. Sensitivity comparable to SQUIDs in the right configuration.

The specific variant that beat SQUIDs on sensitivity is the Spin-Exchange Relaxation-Free (SERF) magnetometer, demonstrated by Michael Romalis and colleagues at Princeton in 2002. In a SERF regime, the alkali vapour is heated to about 180 °C so that spin-exchange collisions happen faster than the Larmor precession at low fields. That eliminates the dominant relaxation mechanism, and the sensor reaches subfemtotesla sensitivity in a device smaller than a coffee mug.

DARPA noticed. The QuASAR program (Quantum-Assisted Sensing and Readout, launched 2010) poured money into atomic magnetometers, nitrogen-vacancy diamond sensors, and related quantum-limited detection technologies. The follow-on AMBIIENT program (2013) targeted atomic magnetometers specifically for whole-body biomagnetic imaging without cryogens. NIST's John Kitching group shrank the whole assembly onto a chip.

The result: the current generation of commercial MEG systems uses optically-pumped magnetometers in a wearable helmet instead of a liquid-helium dewar. The patient can walk around. The sensor array can be reconfigured for different head shapes. That's the civilian version. The classified version is almost certainly further along.

Here's where the physics gets mean. A magnetic dipole, and every biomagnetic source is well approximated as a dipole, falls off as 1/r³. The signal from your heart at 10 centimetres is a thousand times stronger than the signal at 1 metre, which is a thousand times stronger than the signal at 10 metres.

Do the arithmetic. A healthy human heart produces about 100 pT at skin contact. At 1 m, that's down to maybe 100 fT. At 5 m, divide by another 125, and you're at roughly 800 aT (attotesla). That's below the single-shot noise floor of every commercial sensor on the market and right at the theoretical floor for the best SERF devices after extended averaging.

What can you do about it? Three things. None of them cheap.

First, build arrays. N sensors averaged coherently give you √N improvement over uncorrelated noise. A thousand-element array buys you about 30x in signal-to-noise. A million-element array (a chip-scale lattice of OPMs of the kind NIST has prototyped) buys you about a thousand. That's the difference between a sensor that's useless past 5 metres and one that works at hundreds.

Second, use quantum correlation. Entanglement between sensors can beat the standard quantum limit, in principle by another factor of √N on top of classical arrays. That's the Heisenberg limit, and it's been demonstrated in laboratory settings for specific tasks with small photon or atom numbers. Scaling it to a military sensor is an engineering problem nobody has publicly solved.

Third, go active. Instead of passively listening for the signal the target emits, transmit an interrogating field and measure the way the target modulates it. The eddy currents induced in moving tissue by a millitesla interrogation field are much stronger than the body's natural magnetic emission, and they can be phase-locked to the transmitter to reject noise. This is the path that turns a laboratory instrument into a fielded sensor. It's also the path that makes the most sense for a classified system.

Quantum magnetometers don't sit in a single niche. They sit in five.

Medicine is the most visible. Magnetoencephalography maps brain activity with millisecond precision and is now standard for epilepsy presurgical planning. Magnetocardiography catches arrhythmias that electrocardiograms miss. Fetal magnetocardiography monitors unborn heart activity through the mother's abdominal wall.

Defence is the oldest. Navy Magnetic Anomaly Detection, first deployed in World War II using fluxgate magnetometers, hunts submarines by the distortion their steel hulls create in the Earth's ambient field. Modern MAD uses optically-pumped cells for better noise floors. Airborne MAD booms on P-3 Orion and P-8 Poseidon aircraft detect submerged hulls at ranges measured in hundreds of metres.

Geophysics uses SQUID arrays for mineral exploration, paleomagnetism, and earthquake precursor studies. A dragged sensor over a prospecting field will pick out an ore body by the tiny field anomaly it produces.

Fundamental physics runs on them. Tests of CPT symmetry, searches for electric dipole moments, and the hunt for axion dark matter all depend on ultra-sensitive magnetometry. These are the experiments pushing the noise floor down to and past the attotesla level.

And then there's the classified layer. DARPA's program history (QuASAR, AMBIIENT, and follow-ons) points to biomagnetic targets in ways the unclassified reports don't fully describe. Intelligence-community contracts for "quantum sensing in contested environments" have been public since the early 2010s, with the deliverables redacted. Whatever Ghost Murmur turns out to be, it sits at the end of a well-funded research line that the US has been pursuing openly, inside programs that only describe their outputs in general terms.

The sensing physics and the confinement physics are the same physics.

A superconducting array that can measure a 100-femtotesla cardiac signal uses the same Josephson junctions, the same cryogenics, the same quantum-noise-limited readout, as a superconducting magnet stack that can confine a 20-tesla plasma. Different applications of the same technology family. The intelligence community funds both sides because the underlying research pays off in both directions.

When Ashton Forbes argues that the MH370 orbs are field-reversed configuration plasmoids produced by a classified fusion program, and separately argues that the CIA can track a human heartbeat from a mountain, he's not claiming two unrelated miracles. He's claiming one technology cluster with two faces. This 101 guide is the foundation for understanding the sensing face. The research pages on Ghost Murmur, Sonny White, and fusion confinement cover the weapons-side applications.

Read them together. The connection is what the investigation turns on.