Skip to content

QRNG technologies compared

The main QRNG architectures are photon path selection, vacuum fluctuation measurement, photon arrival timing and phase diffusion. They differ less in whether they are genuinely quantum than in generation rate, how cleanly the entropy can be modelled, and how well the optics integrate into a manufacturable device.

Photon path selection

A single photon meets a beamsplitter and is detected in one of two paths. Conceptually the cleanest QRNG there is: the entropy model is close to a textbook description and easy to defend. The constraints are practical. Single-photon detectors are expensive, they have dead time, and generation rate is bounded by detection rate, which puts these designs in the megabit range rather than the gigabit range.

Vacuum fluctuation measurement

Homodyne detection measures the quadrature of a vacuum state against a local oscillator. Rates are high and the components are standard telecom parts. The modelling burden moves to the detection chain: the measured signal contains both vacuum fluctuation and classical electronic noise, and the entropy estimate depends on bounding the ratio between them under all operating conditions.

Photon arrival timing

Time between detections from a weak source is exponentially distributed and can be converted to bits. Simple to build with commodity parts. The difficulty is that detector dead time, afterpulsing and drift in the source rate all shape the distribution, so a careful implementation spends most of its effort on the parts that are not quantum.

Phase diffusion

Gain-switching a laser so each pulse builds from spontaneous emission, then interfering consecutive pulses. High rates from a small number of components, and the entropy model rests on a well-described process. The trade is that the model depends on drive conditions and interferometer stability, so those become quantities the device has to monitor. The architecture in detail.

How to compare them

Rate is the least interesting axis, because every approach above can be made fast enough for most applications. The questions that separate them are whether the entropy model is stated and testable, whether the quantities it depends on can be measured at runtime, and whether the optics can be manufactured at the volume and cost your deployment needs. Integration is where that last question is decided.