Skip to content

QRNG vs TRNG vs PRNG, the differences that matter in production

A PRNG expands a seed, a TRNG measures a physical process, and a QRNG measures a quantum one. The difference that matters in production is not speed but evidence: a quantum source can be described by a physical model, which makes its entropy estimate something you can argue about rather than assert.

The comparison, honestly

A correctly seeded CSPRNG is adequate for a great many production systems, and saying otherwise would be selling rather than explaining. If your threat model is a remote attacker who never sees your hardware, and your seed came from a source you trust, a modern DRBG will not be your weak point.

The cases where that stops being true are specific. Long-lived keys where a compromise is discovered years later. High-assurance environments where you must produce evidence rather than assurances. Devices manufactured in volume where every unit needs its own unpredictable secret and there is no operator present to supply one. Systems under regulatory regimes that ask where randomness comes from.

What distinguishes a TRNG from a QRNG

Both extract from physics. The distinction is how well the physics is understood. A classical TRNG typically relies on a process — ring-oscillator jitter, thermal noise in a resistor — that is hard to predict largely because it is hard to model. That is a real property, but it is difficult to quantify and it can degrade with temperature, voltage or age in ways that are not obvious from the output.

A quantum source is chosen so that the unpredictability follows from the physics rather than from complexity. In a phase-diffusion source the randomness comes from spontaneous emission, which is described by a model you can write down and test against. How that works in practice is worth reading if you want to check the argument rather than take it.

Where the comparison usually goes wrong

Two mistakes are common. The first is comparing on throughput, which is almost never the binding constraint — the number only matters once you tie it to reseeding rates. The second is treating statistical test results as evidence of unpredictability. A well-built PRNG passes every statistical suite you can run at it and is completely predictable to anyone holding the seed.

The question worth asking a vendor is not "is it random" but "what is your entropy model, and what evidence do you have that the source still matches it". That question is the subject of the entropy pillar.