A faster supply of quantum light
A research team from Sparrow Quantum and Ruhr University Bochum has reported a deterministic light source that delivered 515 million single photons per second into an optical fiber. The result appeared on arXiv on September 4 as the first version of a preprint titled "A photonic source with half-a-GHz single-photon flux." It has not yet passed peer review. The authors call its in-fiber flux the highest reported to date, based on their comparison with earlier high repetition rate quantum-dot sources.
That is a meaningful hardware result, but the number needs careful framing. The experiment measured the output of a photon source. It did not demonstrate a complete quantum processor, a useful algorithm, error correction, or quantum advantage. Photonic computing needs bright sources, yet it also needs photons that arrive alone, match one another closely, become entangled, survive routing losses and reach detectors together. The new paper improves one important input while documenting where another requirement becomes weaker.
Why single photons are difficult resources
A conventional laser emits a stream whose photon count varies. A photonic quantum computer often needs something more disciplined: one photon released on command for each clock cycle. The photon can carry quantum information in its path, arrival time or polarization. Optical circuits then make several photons interfere, route them through programmable components and measure the resulting pattern. Because photons travel well and interact weakly with their surroundings, they are attractive carriers for computation and networking.
Weak interaction is also the engineering problem. Useful operations often depend on several photons arriving in the right modes at the same time. If every source, connection and detector has some chance of losing a photon, the probability of collecting the full group falls rapidly as the group grows. A source that is both fast and efficient can generate more valid experimental events before loss removes one of the required photons. Deterministic quantum dots are promising because an excitation pulse can trigger an emitter directly, avoiding the low success probability of sources that generate photons randomly.
How the experiment reached one gigahertz
The team used an indium arsenide quantum dot embedded in a gallium arsenide photonic crystal waveguide. The dot emitted light at 933.1 nanometers and had a measured lifetime of 220 picoseconds. Two electro-optic modulators carved a continuous laser into pulses roughly 40 picoseconds wide. An FPGA controlled their timing at 80 megahertz, 500 megahertz and 1 gigahertz. At the fastest setting, the system attempted one excitation every nanosecond.
The waveguide directed the emitted light toward a single-mode fiber. The researchers deliberately collected the full emission without a spectral filter, including light that filtering might remove to improve quality. This matters because selective filtering can make reported purity and indistinguishability look better while lowering the number of usable photons. Measuring the unfiltered stream exposes more of the practical tradeoff an application would inherit.
What 515 million photons per second means
At the pulse setting used for maximum emission, the measured optical power rose from 9.57 picowatts at 80 megahertz to 57.3 picowatts at 500 megahertz and 110 picowatts at 1 gigahertz. The authors converted those measurements into fiber efficiencies of 56.2 percent, 53.8 percent and 51.5 percent, respectively. At 1 gigahertz, that last efficiency corresponds to an in-fiber single-photon flux of 515 megahertz. In everyday terms, slightly more than half of the billion excitation attempts per second produced a photon in the fiber output counted by the analysis.
The stream was bright enough to register on a commercial optical power meter. Single-photon experiments usually count events with specialized detectors, whose calibration, dead time and response can complicate efficiency estimates. Here, wavelength and measured optical power provide a more direct route to the average photon rate. That measurement approach could be useful beyond computing, including photon-flux standards and detector calibration, if later work establishes the required stability and traceability.
Brightness came with a quality cost
The paper measured two properties that matter when photons must work together. The first is the second-order correlation at zero delay, written as g(2)(0). A lower value means the source is less likely to emit unwanted extra photons in the same pulse. Without spectral filtering, the value was 3.98 percent at 80 megahertz, 3.63 percent at 500 megahertz and 11.13 percent at 1 gigahertz. The fastest operation therefore produced the largest multiphoton contamination of the three tested settings.
The second test was Hong-Ou-Mandel interference. It sends two photons into a beam splitter and measures how consistently they behave as indistinguishable particles. Higher visibility is better. The uncorrected visibility was 81.83 percent at 80 megahertz, 80.87 percent at 500 megahertz and 55.14 percent at 1 gigahertz. The team attributes the decline at high speed to overlapping emission. At a one-nanosecond interval, the quantum dot has less time to finish its decay before the next pulse. The measured background overlap reached about 4 percent at 1 gigahertz.
This is the central qualification to the brightness result. More photons leave the source each second, but a larger share are less suitable for interference at the maximum rate. The authors argue that a source with a shorter lifetime could reduce the overlap. They also show that filtering at 80 megahertz improved g(2)(0) to 2.80 percent and visibility to 92.30 percent, while losing about 12 percent of the light. Those results expose an engineering choice rather than eliminating it.
A source is not a quantum computer
Sparrow Quantum's announcement says the stream could be divided across ten channels and make experiments involving roughly 10 to 20 photons practical. The preprint describes that range as a potential application, not a completed demonstration. It reports the source, its drive electronics and its optical measurements. It does not show ten or twenty photons being demultiplexed, entangled, processed through a large interferometer and detected as a computational result.
A 2024 Nature Photonics experiment illustrates the missing system layers. Quandela's Ascella prototype combined an 80-megahertz quantum-dot source with a demultiplexer, a programmable 12-mode optical chip, detectors, control software and error-compensating compilation. It processed six photons on chip at a sampling rate of 4 hertz, even though its source ran millions of times faster. The gap reflects accumulated optical loss, synchronization requirements and the falling probability of detecting every photon in a multiphoton event. The comparison is not a direct benchmark, but it shows why raw source flux cannot be reported as computer speed.
The next evidence to demand
The preprint gives the field a useful combination of high clock rate, more than 50 percent fiber efficiency and measurements made without hiding unfiltered emission. Its 500-megahertz operating point may be especially interesting because its purity and interference visibility stayed close to the 80-megahertz results while output increased sharply. Future work should test whether that balance survives long runs, additional optical components and active demultiplexing.
Peer review and independent replication remain necessary, particularly because most authors are affiliated with Sparrow Quantum, which commercializes photon-source technology. The decisive follow-up would connect the source to a complete photonic system and publish end-to-end rates for simultaneous high-quality photons, entangled states or executed circuits. It should also report stability and total loss from generation through detection. Success there would shorten experiments and expand the quantum protocols researchers can test. Until then, the measured advance is a brighter component with a clearly measured quality ceiling, not a working large-scale quantum computer.