Skip to content
Assurance levels

The four types of RNG,
from L1 to L4.

What each type is built from, and what each one can actually evidence — mapped to what NIST SP 800-90B and BSI AIS 20/31 require. Use it to separate two products whose marketing language reads the same.

Published with Deloitte

Quantum Random Number Generators — What’s All the Fuss About?

Quside and Deloitte wrote a joint article on quantum random number generation: why randomness is the part of cryptography nobody inspects, what a quantum source changes, and which questions to put to a supplier. It covers the same ground as this page at greater length, and it is published by Deloitte.

Read it on deloitte.comPDF
Cover of the Deloitte and Quside article “Quantum Random Number Generators — What’s All the Fuss About?”, showing the Deloitte logo above a sphere of glowing digits
How to read it

Compare generators on what is established about them.

No scheme issues a Level 4 certificate, and no standards body issues a Level 1, 2 or 3 certificate either. Quside publishes these levels because they are useful in a procurement conversation, not because they confer a rank.

The certificates Quside holds are named, with a stated scope: SP 800-90B entropy certificate E178 for Quside PCIe One and FMC One, and ENS Alta — the highest of the three categories of the Spanish scheme — for the same two products.

The maturity model

Each level answers a harder question.

Each level adds a kind of evidence rather than a kind of technology. A product moves up by being able to answer a question it previously could not.

Level 01

The illusion of randomness

Deterministic generation

Where does the seed come from?

An approved deterministic mechanism produces the output stream. The mechanism itself is well specified and well studied. What is unestablished is what seeded it: a deterministic generator started from a predictable seed produces a predictable stream, and its output looks exactly as random as one started from a good seed.

What this looks like in practice
  • A software generator seeded from a fixed value, a timestamp or a process ID
  • A DRBG whose seed source is undocumented, or shared across a production run
  • A generator reseeded rarely, or never, over a long device lifetime
Evidence available at this level

Algorithm-level validation of the deterministic mechanism, and the platform’s own documentation of how it is seeded.

What to establish beyond it

This says nothing about the physical source underneath. Approved DRBGs depend on strong seeding, correct construction and appropriate reseeding from trusted entropy sources, so the assurance question moves down a layer rather than disappearing.

Level 02

Opportunistic randomness

Physical source, assessed by output

Is there a model, or only statistics?

Entropy is collected from processes that happen to be unpredictable rather than from hardware built to be, and its quality is argued from statistical testing of the output.

What this looks like in practice
  • Keyboard and mouse timings, and other human input intervals
  • Disk seek and interrupt timings
  • Variation in network traffic arrival times
Evidence available at this level

Statistical test batteries run over captured output, plus a vendor description of the noise mechanism.

What to establish beyond it

Output statistics cannot distinguish a healthy physical source from a deterministic generator having a good day. Without a written model of the physical process there is nothing specific to measure the device against.

Level 03

Engineered chaos

Modelled source, validated once

Validated when, and under what conditions?

A dedicated component is built to produce entropy, the physical process has a documented model, and a third party evaluates the source against a defined methodology including the health tests the standard requires.

What this looks like in practice
  • Ring oscillators, using jitter between free-running circuits
  • Avalanche and Zener noise from a reverse-biased junction
  • Thermal noise in a resistor
  • A quantum source with a written model, evaluated once
Evidence available at this level

Third-party validation of the entropy source — the level at which a NIST SP 800-90B entropy certificate sits.

What to establish beyond it

The result is bounded to the configuration and conditions submitted, at the time of evaluation. Hardware sources move with temperature, voltage and age, and the certificate travels with the product for the life of the device without reporting any of that.

Level 04

Quantified uncertainty

Modelled source, verified while it runs

Is the device still inside the model it was validated against?

The device measures the physical process against its model during operation and reports a health state describing whether it remains inside its validated operating conditions.

What this looks like in practice
  • A written physical model of the source, not only a statistical one
  • On-device metrology sampling the parameters that model describes
  • A health state the host can read while the system runs
  • A per-device record an auditor can be shown after deployment
Evidence available at this level

Per-device runtime evidence, produced continuously and available to the host system for operations and compliance to inspect.

What to establish beyond it

The layers above the source: how the seed is consumed, how the deterministic mechanism is constructed and reseeded, and what the system does when the health state changes.

Mapping

Which properties formal requirements already ask for.

Most of what separates Level 2 from Level 3 is written into SP 800-90B and AIS 20/31 already. What the top of the model adds is the part those documents leave to product design: a health state a human can see, in the field, per device.

RNG assurance properties mapped to NIST SP 800-90B and BSI AIS 20/31 requirements
PropertyNIST SP 800-90BBSI AIS 20/31
A documented model of the physical processLevel 3SP 800-90B requires a description of the noise source and a justification for the entropy it claims.AIS 20/31 expects a stochastic model of the noise source for physical true random number generators.
Entropy quantified within that modelLevel 3SP 800-90B defines the estimation methodology used to assess min-entropy for the submitted source.AIS 20/31 derives the entropy claim from the stochastic model rather than from output statistics alone.
Continuous health testing during operationLevel 3 → 4SP 800-90B requires continuous health tests — a repetition count test and an adaptive proportion test — running while the source produces output.AIS 20/31 requires online testing and total-failure detection for physical generators.
A reported health state, per device, in the fieldLevel 4Beyond the required health tests, SP 800-90B does not define an operator-visible health interface.Beyond online testing, AIS 20/31 does not define how health is surfaced to an operator.
Evidence retained for audit and complianceLevel 4Out of scope for an entropy-source validation.Out of scope for a generator evaluation.

Comparing suppliers on this?

Send the requirement wording you have been given. We will tell you which properties it is asking for and how far our evidence goes.