Hardware RNG in practice

A "hardware" random number generator draws its unpredictability from a physical process — thermal noise, electronic feedback, radioactive decay, photon effects — rather than from an algorithm. This chapter sets out what that physical layer actually is, why raw physical output must be conditioned and continuously health-tested, and why certified builds rarely rely on it alone: the common design is a physical entropy source seeding a cryptographically strong deterministic generator. The honest through-line is that "hardware RNG" is a role, not a magic guarantee, and "PRNG" is not a weakness — unpredictability comes from correct design and genuine entropy, not from the label.

The opening chapter described a software generator as an internal state advanced by an update rule, and the seeding chapter followed that generator back to its root: since a deterministic algorithm manufactures no randomness of its own, its unpredictability has to enter from somewhere — an entropy source. This chapter is about that physical layer itself. It is where the word “hardware” in hardware RNG lives: a device that draws unpredictability from a physical process rather than an algorithm. The temptation is to treat that as the trump card — real randomness, therefore unbeatable. That framing is wrong in both directions, and the standards say why. A physical source supplies genuine entropy but is noisy, can drift, and can fail; a deterministic generator is reproducible by design and gets its unpredictability from the seed. Each has a role, and the certified answer is usually to combine them. Nothing here is a method; it is the physical layer, described as the primary standards actually specify it.

What a physical entropy source actually is

At the base of any hardware generator is a noise source: the physical thing whose behaviour is genuinely uncertain. NIST’s standard for entropy sources is blunt about its importance — “The noise source is the root of security for the entropy source and for the RBG as a whole. This is the component that contains the non-deterministic, entropy-providing process that is ultimately responsible for the uncertainty” in the bits produced. Physically, that process is some small-scale event that resists prediction. GLI-11 lists the gaming-relevant examples directly: a hardware-based RNG derives its randomness “from small-scale physical events such as electric circuit feedback, thermal noise, radioactive decay, photon spin, etc.” NIST draws the same line between the physical and the merely-messy: it distinguishes “Physical noise sources”, which “use dedicated hardware to generate randomness,” from “Non-physical noise sources” that lean on system data or human input. A true hardware RNG means the first kind — dedicated hardware sampling a physical process.

Why raw physical output must be conditioned

A physical process is unpredictable, but it is rarely clean. Thermal noise sampled off a circuit is typically biased — one bit value slightly more common than the other — and its successive samples can be correlated rather than independent. Left raw, that is real entropy wrapped in statistical structure an observer could exploit. So the standard inserts a conditioning component between the noise source and the output. NIST defines it plainly: the conditioning component is “a deterministic function responsible for reducing bias and/or increasing the entropy rate of the resulting output bits”. It is worth noticing that this step is itself deterministic: even inside a “hardware” generator, deterministic algorithms are doing load-bearing work. Conditioning does not add randomness — it cannot — it concentrates and cleans the randomness the noise source already supplied, so the delivered bits are closer to unbiased and independent.

Physical sources drift and fail — so they are health-tested

The distinctive risk of a physical source is that it is a physical device: it can degrade or break, and unlike a software bug it can do so gradually and quietly. NIST states the stakes with no hedging — “If the noise source fails to generate random outputs, no other component in the RBG can compensate for the lack of entropy; hence, no security guarantees can be made” for the application relying on it. That is why continuous health testing is not optional decoration but part of the design: health tests are “intended to ensure that the noise source and the entire entropy source continue to operate as expected”, and they are expected to raise an alarm when there is “a significant decrease in the entropy of the outputs,” when “noise source failures occur,” or when the hardware itself fails. NIST splits the tests by moment. Start-up tests run “after powering up, or rebooting, and before the first use of the entropy source” to check nothing has failed since last time; continuous tests then run “indefinitely on the outputs of the noise source while the noise source is operating”, so a failure that develops in service is caught as it happens.

GLI-11 turns exactly this reasoning into a gaming requirement. Its hardware-RNG section notes that “the performance of hardware-based RNGs may deteriorate over time or otherwise malfunction”, that “the failure of a hardware-based RNG could have serious consequences,” and therefore that where one is used there “shall be dynamic monitoring of the output by statistical testing.” The operational teeth are in the next sentence: that monitoring “shall disable game play when malfunction or degradation is detected”. A physical source that quietly goes bad is precisely the failure mode the standard is written to catch and stop — which is the honest answer to whether a hardware RNG can fail silently: it can degrade, so the design is required not to let that pass unnoticed.

Hardware and deterministic generators each have a role

It is tempting to read all this as “hardware good, software suspect.” That is not what the standards say, and the seeding chapter already made half the point: a deterministic generator’s reproducibility is a feature, and all of its unpredictability is inherited from the seed. A cryptographically strong deterministic generator, seeded with real entropy and kept secret, is not a soft underbelly — it is the workhorse that produces fast, well-distributed draws. What a physical source uniquely provides is the genuine, non-reproducible unpredictability that a deterministic algorithm cannot invent. The two are complementary, not ranked. NIST is careful to frame security as a property of the whole system rather than of one favoured component: it “is a system implementation issue; both the DRBG mechanism and its randomness source must be considered.” Unpredictability comes from correct design plus genuine entropy — not from the label “hardware.”

The common hybrid: entropy seeds a deterministic generator

Put those roles together and you get the arrangement most certified builds actually use. A physical entropy source produces genuine but slow, imperfect randomness; a deterministic generator produces a fast, uniform stream but only becomes unpredictable when it is seeded from real entropy. So the physical source is used to seed the deterministic one. NIST’s two recommendations are written to slot together exactly here. The deterministic-generator standard says its seed is “determined from the output of the randomness source” and “must contain sufficient entropy to provide an assurance of randomness,” and it defines what that source may be: “An approved randomness source is an entropy source that conforms to [SP 800-90B]”. In other words, the physical entropy source of this chapter is the sanctioned way to seed the deterministic generator of the last one. GLI-11 encodes the same pattern from the gaming side: it permits a generator’s state to be modified by “overwriting (re-seeding) or mixing (entropy injection) of all or a portion of the RNG state by an external event or entropy source”, with the plain condition that “the external event or entropy source shall not be able to be predicted or estimated by a player.”

“Hardware” is a role, not a magic word

The claim this chapter is careful not to make is that a hardware RNG renders a game provably unbeatable, or that a PRNG is an inherent weakness a player could exploit. Both are false in the same way. A physical source without conditioning and health testing can leak structure or fail unseen; a deterministic generator seeded from real entropy and built to a cryptographic bar is exactly what a certified game relies on for its outcomes. What actually protects the player is the combination the standards require — a genuine entropy source, conditioned and continuously monitored, feeding a strong deterministic generator whose seed and state stay secret. The reassurance is real, but it comes from that design, not from the presence of the word “hardware” on a spec sheet. For what a certificate does and does not vouch for once such a generator is built, see what a certificate asserts; for why statistical testing of the output cannot by itself settle unpredictability — the same reason NIST gives for testing the noise source differently — see what RNG tests can and cannot prove.

Common questions

Sources (3)

• NIST SP 800-90B — Recommendation for the Entropy Sources Used for Random Bit Generation (§2.2 entropy-source model: noise source, conditioning component, health tests; §4.1–4.2 start-up and continuous health tests)nvlpubs.nist.gov
• GLI-11 Gaming Devices, v3.0 (2016) — §3.4 Hardware-Based RNG (§3.4.1 physical entropy sources; §3.4.2 Dynamic Output Monitoring — disable play on malfunction or degradation) and §3.2.6 Unpredictability (re-seeding / entropy injection)gaminglabs.com
• NIST SP 800-90A Rev. 1 — Recommendation for Random Number Generation Using Deterministic RBGs (§8.6.5 Randomness Source: an approved source is an SP 800-90B entropy source; the seed is drawn from that source and must carry sufficient entropy)nvlpubs.nist.gov

Education, not advice. This chapter explains how the random number generators behind casino games are built, tested and certified so you can judge the claims made about them honestly. It is not a system, and nothing here treats gambling as a way to make money — over enough play the mathematics favours the house. 18+.