Counting Photons by Their
Quantum Magnetic Footprints

Most detectors announce that light arrived. A more demanding instrument must say how much arrived: no photon, one, two, three, or more. That distinction matters when the particles of light are not merely illumination but the carriers of a quantum calculation.

A team at Japan’s National Institute of Advanced Industrial Science and Technology, or AIST, has now demonstrated a different route to that count. Rather than waiting for absorbed light to heat a sensor into a measurable state, the researchers observed the immediate phase motion that the photon triggered inside a current-carrying superconductor.

The paper, published September 2 in the American Physical Society journal Physical Review Applied and designated an Editors’ Suggestion, is by Takeshi Jodoi, Fuminori Hirayama, Tetsuya Tsuruta, Takahiro Kikuchi and Daiji Fukuda. Jodoi is also affiliated with the University of Tokyo’s School of Engineering. AIST identifies Fukuda as Principal Researcher and Tsuruta as Senior Researcher across its G-QuAT and physical-metrology units.

A voltage pulse with a fixed quantum inside

The active device was an eight-micrometre square on silicon: 20 nanometres of titanium layered with 10 nanometres of gold, connected by niobium leads and optically coupled to a fiber. Laser pulses at 1,526 nanometres—within the telecommunications band—arrived while the film sat below its 128-millikelvin critical temperature in a dilution refrigerator.

Absorbing a photon locally weakens the superconducting state, creating a transient hotspot-like region. In a biased film, that disturbance destabilizes the gradient of the superconducting phase. A vortex and an antivortex can nucleate, then move across the strip under the applied current.

Each traversal changes the phase by 2π. Through the Josephson relation, that phase slip produces a voltage pulse whose time integral equals one magnetic-flux quantum, Φ₀ = h/2e. The researchers found the minimum pulse integral remained close to 2.07 × 10⁻¹⁵ weber as they changed bias current and base temperature. Larger signals appeared in integer steps.

Peak voltage alone is not the invariant. It can change with the bias circuit and current. The area beneath the voltage-time curve carries the more fundamental record: one flux quantum for one resolved phase slip, integer multiples when more vortices traverse the film.

Why a staircase is not automatically a photon count

The elegant sequence of voltage levels conceals the central difficulty. One absorbed photon does not always produce exactly one traversing vortex pair. The conversion fluctuates with temperature, bias current and nonequilibrium dynamics. Quantized phase slips can therefore be visible while the incoming photon statistics remain obscured.

The team mapped those fluctuations with a second-order correlation function and chose a region in which vortex generation became statistically stable. At about 98 millikelvin and a three-microampere bias, with an average input of one photon per pulse, it analyzed 100,000 events. The reconstructed photon distribution agreed with the Poisson statistics expected from the laser. Its photon correlation value, 0.972, was consistent with that distribution.

The public release presents separated classes corresponding to zero, one, two and three absorbed photons, plus a class for four or more. This is the basis for the photon-number-resolution claim. It is not evidence that the instrument unambiguously resolves every photon number under every operating condition. The paper explicitly reports regimes where the vortex count fluctuated and failed to track photon statistics directly.

A third route between heat and resistance

Photon-number-resolving detectors are already real. Transition-edge sensors, or TESs, hold a superconducting film at the boundary between its zero-resistance and normal states. A photon’s energy becomes heat; a minute temperature rise changes the resistance. Because the change scales with deposited energy, a TES can distinguish multiple photons with exceptional precision.

AIST has spent years advancing that platform. In a separate 2026 paper, its group reported TES performance reaching 99 percent efficiency and more than 30 resolvable photons. The cost is recovery time: the sensor must shed the absorbed heat and return to its operating point before it is ready for the next event.

That work also belongs to AIST’s less glamorous but essential role as Japan’s national metrology institute. A detector cannot be called 76, 98 or 99 percent efficient merely because it produces a strong signal. Researchers need a traceable account of how many photons actually reached it. In 2025, AIST reported a wavelength-tunable standard quantum light source spanning the communications C-band from 1,530 to 1,565 nanometres. The institute described it as a “photon ruler”: optical power tied to national standards and attenuated into a known single-photon-level flux.

The new vortex experiment used 1,526-nanometre light, just outside that stated C-band range, and it is a detector-physics demonstration rather than a new calibration standard. Even so, the institutional continuity matters. Quantum hardware must connect exotic phase dynamics to ordinary, auditable quantities—wavelength, power, efficiency, uncertainty and time. A detector that cannot be calibrated reproducibly cannot reliably validate a quantum computer.

Superconducting nanowire single-photon detectors, or SNSPDs, occupy another corner of the trade space. They can combine excellent efficiency, very low noise and fast timing. A conventional single element, however, chiefly reports whether a detection occurred. Photon-number resolution generally requires spatial arrays, parallel elements or specialized readout and device geometries.

The AIST experiment calls its operating regime a vortex-based detector, or VBD. Its signal comes from discrete phase slips rather than a completed thermal relaxation or a macroscopic resistive transition. The same physical device could also operate in an electrothermal-feedback TES mode, allowing a revealing comparison: depending on temperature, its TES effective time constant was 4.82 to 29.1 microseconds, against 445 nanoseconds for the observed VBD pulse.

The crucial gap between 445 nanoseconds and 800 picoseconds

The fastest figure associated with the experiment is also the easiest to overstate. The paper estimates that a vortex crossing an eight-micrometre strip could produce an intrinsic response of about 800 picoseconds. That calculation assumes a velocity of roughly 10,000 metres per second reported previously for vortices in niobium carbide.

The velocity in this experiment’s titanium/gold bilayer is unknown. The 800-picosecond number is consequently an order-of-magnitude projection, not a stopwatch result. What the team measured was a 445-nanosecond time constant.

That slower value closely matches the roughly 458-nanosecond electrical constant produced by the readout’s 9.8-nanohenry inductance and 21.4-milliohm shunt resistance. The agreement suggests the electronics, rather than the underlying phase motion, set the observed limit. A wider-bandwidth circuit could expose a much faster response—but it must do so without erasing the tiny signal in noise.

Speed must ultimately be reported as a system, not an isolated crossing. A useful detector needs a measured rise and reset time, sustained count rate, timing jitter, false-count rate and classification error, all while preserving efficiency. Scaling also introduces SQUID bandwidth, cryogenic wiring, heat load and multiplexing constraints.

A century of learning how to hear one quantum

The intellectual path begins before quantum electronics had a name. Albert Einstein’s 1905 light-quantum argument supplied the particle-like packet of optical energy. Heike Kamerlingh Onnes discovered superconductivity in 1911. In 1962, Brian Josephson predicted the phase-voltage relation that makes a 2π slip electrically countable; the work contributed to his share of the 1973 Nobel Prize in Physics.

TES technology matured into a standard for exquisitely precise photon and radiation calorimetry. A different lineage accelerated in 2001 when researchers reported a picosecond superconducting single-photon optical detector based on a current-biased nanowire. Successive SNSPD designs pushed efficiency at 1,550 nanometres to 98 percent at the system level by 2020.

Throughout that evolution, theorists debated how hotspots, vortices, current redistribution and normal-domain formation combine to produce a nanowire detection event. Vortex-assisted models were not new. The AIST paper’s specific advance is the dynamic electrical observation of photon-induced vortex–antivortex motion as quantized phase-slip signals, followed by photon-number reconstruction from those counts.

“Direct observation” should therefore be understood precisely. The team did not make a magnetic-field photograph of each vortex. It measured the quantized voltage integral that a vortex traversal must create and followed the discrete phase dynamics in real time.

From a beautiful principle to a useful component

Large photonic quantum computers must create, route, interfere and measure enormous numbers of fragile optical states. Loss and imperfect components make error detection unavoidable. A detector that can rapidly distinguish an expected single photon from zero, two or more could support state verification, feed-forward decisions and some error-correction architectures.

That prospect explains the support from the Japan Science and Technology Agency’s Moonshot Goal 6 programs for fault-tolerant large-scale photonic quantum computers, along with Japanese government programs administered through SIP, BRIDGE, JSPS and NEDO. It does not mean the new device has already performed quantum error correction.

For deployment, the experiment’s 128-millikelvin critical temperature is consequential. It requires dilution-refrigerator conditions. Engineers must improve detection efficiency and classification fidelity, measure the promised speed, establish wavelength dependence and dynamic range, and build arrays whose cryogenic wiring and readout do not overwhelm the advantage.

Photon-number resolution must also be judged against the intended computation. The experiment reconstructed the Poisson distribution of an attenuated coherent laser pulse, a necessary and well-understood benchmark. Future systems will have to classify nonclassical states, distinguish closely spaced number peaks at higher occupancy and report errors quickly enough for feed-forward control. Average agreement with a known distribution is not the same test as assigning every individual event correctly inside a fault-tolerant machine.

Nor is a detector’s contribution to error correction automatic. Photonic architectures differ in their encodings, resource states and tolerance of loss. Some require number resolution at selected measurements; others trade detector complexity for additional optical modes or repeated trials. The new mechanism is best understood as a potentially valuable hardware option whose system-level benefit must be quantified inside a particular architecture.

The result nevertheless changes the design conversation. It demonstrates an experimentally accessible signal chain—photon, vortex–antivortex pair, phase slip, quantized voltage, photon-number inference—that is neither ordinary calorimetry nor ordinary resistive detection. It creates another set of compromises for device physicists to optimize.

Quantum machines will eventually be judged by a deceptively classical act: counting correctly. AIST’s experiment offers a new alphabet for that final readout, written not in heat but in the passage of magnetic quanta. The next test is whether those crisp integer marks survive faster circuits, larger arrays, more wavelengths and long operating hours.