LM-TRNG

Features

Entropy source

• Array of free-running inverter rings, configurable in count and length
• Every phase of every ring sampled and combined
• External noise source accepted instead, for a system supplying its own

Conditioning

• Configurable XOR parity filter with a length clamp
• One output bit per configured group of raw bits — no bit reused, none discarded

Continuous health tests

• Repetition Count Test — NIST SP 800-90B section 4.4.1
• Adaptive Proportion Test — NIST SP 800-90B section 4.4.2
• Total failure test — AIS-31 requirement PTG.2.5, watching for a stopped ring
• Frequency (monobit) and Runs tests — NIST SP 800-22 sections 2.1 and 2.3, applied continuously over configurable windows
• Power-up known-answer test of the deterministic path
• Output inhibited while any test is in alarm; every alarm latched and readable

Integration

• AMBA AXI4-Lite control and an AXI4-Stream random output
• All test thresholds are registers, so an integrator sets them from their own assessment
• Two delivery tops: complete AXI product, or bare core for a custom fabric
• ANSI C driver, UVM RAL, IP-XACT, SystemRDL register description

---

Applications

• Seeding a deterministic random bit generator — LM-DRBG is the matching block
• Key generation at power-up, before any seed exists
• Nonce and challenge generation
• Physical unclonable and device-identity schemes
• Any system requiring a non-deterministic source rather than a seeded one

---

Block diagram

---

Interfaces

The output stream has no end. A physical source produces bits for as long as it is enabled; there is no request length and no tlast. That is the structural difference from a deterministic generator, and it is visible at the interface.

---

Health and assurance

Four continuous tests plus a power-up self-test. The two SP 800-90B tests are implemented as the document writes them, including the boundary conditions each is easy to get wrong. The AIS-31 total-failure test watches the sampled ring phases directly rather than the combined noise bit, so a stopped ring is caught sooner and more specifically than a statistical test would catch it.

Every threshold is a register. Cutoffs, window sizes and the parity filter length are configuration, not constants, because an integrator validating this source on their own device and operating corner must be able to set them from their own assessment.

Output is inhibited while any test is in alarm, and every alarm is latched.

The entropy claim, stated plainly

The min-entropy per sample is not established by this deliverable, and no figure is claimed. A min-entropy estimate requires at least one million consecutive raw samples collected from the physical source on the target device at the operating corner, run through the SP 800-90B estimators. That is a measurement on hardware, and it cannot be produced by simulation, synthesis or any artefact in this package.

The default threshold values shipped in the register map are computed from an assumed min-entropy, and that assumption is labelled as one everywhere it appears. An integrator seeking any validation replaces every threshold from their own assessment — which is precisely why they are registers.

The ring oscillator array has no model and is not simulated. Everything downstream of the digitised noise bit is a deterministic state machine and is verified clock by clock against a reference model; the rings themselves are physical and are verified by measurement on silicon, which is the integrator's to perform.

---

Configuration

---

Resources and performance

Resource utilisation and maximum clock frequency for each target device are measured out of context and published per target in the measured results on this page. AMD and Efinix rows report LUTs; Altera rows report ALMs, and the two are never combined.

Output rate is the sample rate divided by the parity filter length, and both are configuration.

---

Deliverables