An entropy source is the part of a random-number-generation system that provides unpredictability.
That sounds simple, but it is one of the most important distinctions in cryptographic randomness. A deterministic algorithm can expand an existing secret state into a very large stream of random-looking bits, but it cannot manufacture unpredictability from nothing. Somewhere in the architecture, uncertainty has to enter the system.
That is the job of the entropy source.
NIST SP 800-90B is specifically focused on entropy sources used for random bit generation. It defines design principles and validation requirements for entropy sources and explains how those sources fit alongside deterministic random bit generator mechanisms from SP 800-90A and complete constructions from SP 800-90C.[1]
In short
A useful mental model is:
physical process → measurement → raw samples → health monitoring → entropy estimation → conditioning → random-bit generation → cryptographic application
The physical phenomenon matters, but so does every stage that follows it.
Entropy source vs random number generator
The terms entropy source and random number generator are often used as if they meant the same thing. They do not.
An entropy source provides observations that contain unpredictability. A random-bit generator is the larger mechanism that turns suitable entropy into output that applications can use.
A complete implementation can therefore contain several logically different components:
- a physical or environmental source of uncertainty;
- a measurement process that converts the source into digital samples;
- tests that check whether the source is behaving as expected;
- an entropy model and conservative estimate;
- conditioning or extraction;
- a deterministic generator or another random-bit-generation construction;
- the application that consumes the output.
This separation is important because output can look statistically clean even when the underlying entropy mechanism is weak, misunderstood, or failing.
Where does entropy come from?
Hardware entropy sources use physical processes whose observed behavior contains uncertainty.
Depending on the design, a classical hardware random number generator might derive entropy from oscillator timing variation, electronic noise, metastability, thermal effects, or another physical mechanism. A quantum random number generator uses a process in which the claimed unpredictability is associated with quantum behavior.
QRNG implementations can use different physical approaches. Some are photonic. Others aim to derive quantum entropy from electronic or semiconductor phenomena. ETSI's 2026 QRNG implementation guidance surveys a range of QRNG approaches and discusses implementation characteristics, testing, performance, entropy sources, and design trade-offs.[7]
The source physics, however, is only the beginning.
A real device must observe that physics through electronics and convert it into digital samples. The measurement chain can add bias, correlation, drift, saturation, deterministic artifacts, or classical noise. A credible entropy design therefore considers the entire path from the source to the final bits.
The noise source and the measurement chain
NIST SP 800-90B uses the concept of a noise source for the underlying non-deterministic behavior from which entropy is obtained.[1]
The measurement chain turns that behavior into data. In practice, this means the entropy source is not just an abstract physical event. It is a physical event plus the circuitry, sampling process, digitization, timing, thresholds, calibration assumptions, and operating conditions that determine what the system actually observes.
For engineers evaluating a source, useful questions include:
- What physical process is supposed to be unpredictable?
- Which part of the measured signal is credited as entropy?
- What classical or deterministic effects may also be present?
- How does temperature, supply voltage, aging, or interference affect the signal?
- What assumptions are made about independence or correlation?
- What happens when the source leaves its expected operating range?
These questions are more informative than asking only whether a final output file passes a statistical test suite.
Entropy estimation: how much unpredictability is really there?
A raw data stream can contain bits without containing one full bit of entropy per bit.
If a source is biased, correlated, or partially predictable, its entropy rate can be lower than its raw bit rate. This is why physical throughput and entropy throughput are not the same quantity.
SP 800-90B includes requirements and methods related to entropy estimation, including min-entropy.[1] NIST's IR 8427 further explains that the SP 800-90 series measures unpredictability using min-entropy.[4]
The engineering goal is conservative accounting: do not claim more unpredictable information than the source and its model can justify.
What is conditioning?
Raw physical samples are often not suitable for direct cryptographic use.
They may contain bias or correlations even when the underlying source is legitimate. Conditioning processes the raw samples so the output is closer to the properties required by the next layer of the system.
A critical principle is that deterministic conditioning does not create entropy from nothing. It can transform and compress entropy that is already present, but the amount of output claimed as unpredictable must remain supported by the entropy available at the input.
This is why a source's entropy estimate and its conditioning design belong in the same engineering conversation.
Why health tests matter
Physical systems can change.
Temperature, voltage, component aging, interference, manufacturing variation, hardware faults, or environmental conditions can affect an entropy source. A design that was characterized in one state should not simply assume that state persists forever.
SP 800-90B includes health-testing requirements for entropy sources.[1]
Health tests are not a replacement for initial characterization. Their role is to detect certain deviations or failure conditions during operation so that the system can respond instead of silently continuing to produce output under broken assumptions.
A useful evaluation question is therefore not only “what tests are implemented?” but also “what does the system do when a test fails?”
Where does a QRNG fit?
A QRNG is one type of hardware entropy architecture.
Its defining claim is that the underlying physical source is quantum. That can be important, but the word quantum does not remove the engineering requirements around measurement, characterization, entropy estimation, conditioning, health monitoring, and integration.
A stronger evaluation question than “does the output look random?” is:
What is the physical entropy model, how much entropy is conservatively justified, how is the source monitored, and what happens when its assumptions stop holding?
That question moves the conversation from random-looking data toward an auditable entropy architecture.
How the NIST SP 800-90 series separates the problem
The current SP 800-90 series intentionally separates major parts of random-bit generation:
- SP 800-90A Rev. 1 specifies deterministic random bit generator mechanisms.[2]
- SP 800-90B addresses entropy sources.[1]
- SP 800-90C specifies random-bit-generator constructions that use entropy sources and DRBG mechanisms.[3]
This is useful when thinking about QRNG integration. A hardware entropy source and a deterministic generator are not automatically competing products. They may occupy different layers of the same architecture.
What an entropy source does not prove
Calling a device an entropy source does not, by itself, prove:
- that every raw output bit contains one bit of entropy;
- that the source is secure under all environmental conditions;
- that its physical mechanism is quantum;
- that it has been validated under SP 800-90B;
- that it complies with a specific certification program;
- that statistical tests alone establish its security.
Those are separate claims that require separate evidence.
A practical evaluation checklist
When reviewing an entropy source, ask for enough information to understand:
- Source: What physical phenomenon provides unpredictability?
- Measurement: How is the phenomenon observed and digitized?
- Model: Which part of the signal is credited as entropy?
- Estimate: How much entropy is conservatively claimed per sample?
- Conditioning: How are raw samples transformed into usable output?
- Health monitoring: What conditions are checked during operation?
- Failure behavior: What happens when assumptions or tests fail?
- Integration: How is the output delivered to the cryptographic system?
- Evidence: Which claims are supported by standards, validation reports, test data, or published analysis?
Summary
An entropy source is where unpredictability enters a random-number-generation system.
For security-sensitive applications, a trustworthy design should be physically understood, conservatively characterized, monitored during operation, and connected to conditioning and random-bit-generation components without claiming more entropy than the source can support.
QRNG provides one possible physical origin for that entropy. Its value depends not only on the quantum phenomenon, but on how the complete entropy path is measured, modeled, monitored, and integrated.
Evaluating a hardware entropy source? Explore the iQrypto Evaluation Kit.
Sources and further reading
- NIST, SP 800-90B: Recommendation for the Entropy Sources Used for Random Bit Generation — https://csrc.nist.gov/pubs/sp/800/90/b/final
- NIST, SP 800-90A Rev. 1: Recommendation for Random Number Generation Using Deterministic Random Bit Generators — https://csrc.nist.gov/pubs/sp/800/90/a/r1/final
- NIST, SP 800-90C: Recommendation for Random Bit Generator (RBG) Constructions — https://csrc.nist.gov/pubs/sp/800/90/c/final
- NIST, IR 8427: Discussion on the Full Entropy Assumption of the SP 800-90 Series — https://csrc.nist.gov/pubs/ir/8427/final
- NIST, Random Bit Generation publications — https://csrc.nist.gov/Projects/random-bit-generation/publications
- QRNG.io, What Makes a QRNG Trustworthy? — https://www.qrng.io/articles/what-makes-a-qrng-trustworthy
- ETSI, TR 104 171 V1.1.1: Implementation Guidelines for Quantum Random Number Generators — https://www.etsi.org/deliver/etsi_tr/104100_104199/104171/01.01.01_60/tr_104171v010101p.pdf