← Back to gallery

Gf180 Trng

ok

True random number generator on gf180mcu — 2AM Logic canary block

gf180-trng

Source: 2AMLogic/gf180-trng @ 0bb7e75314af

Specification

ratified 2026-07-31

Entropy source**N-way array of independent free-running ring oscillators, XOR-combined ahead of a single sampler**, N fixed by the jitter-budget sizing law `Q_array ≥ 1.5 × 4.0×10⁻³` at the entropy-binding corner ([DR-0007]) · stretch: metastability hybrid, scoped as a *secondary tap on the RO core* — not a free-standing source
Raw rate**> 2 kbps** sustained at the raw tap (sampler output), binding at the slowest-RO corner: `ss` / −10 % / +125 °C. Value from [DR-0011-raw-rate-at-the-measured-starved-cell-jitter-energy][DR-0011-rate] §1 (superseding [DR-0010] §1's > 500 bps, which superseded the original > 1 Mbps); measurement point, *sustained* and binding corner unchanged from [DR-0003]. The row is a floor **and the same law sets its ceiling — a maximum sample clock**: DR-0007 §2 holds only up to **~2.77 kHz raw** (2773 bps at the entropy-binding corner `ss` / +125 °C / 3.63 V with a = 7.32, `array_sizing.py --check`); an integrator must not clock the sampler above it, because past it the entropy bound the row rests on no longer holds · stretch: > 4 Mbps, same definition — [DR-0003]'s original stretch, not re-derived by either rate decision: **unattainable under the measured jitter-energy law**, far above the ~2.77 kHz sample-clock ceiling at which the shipped array still satisfies DR-0007 §2 (`array_sizing.py`)
Raw min-entropy per bit**placeholder — H₀ = 0.5 bit/sample is a design *target*, not a measurement.** Stated at the entropy-binding corner, **`ss` / +125 °C / 3.63 V** ([DR-0015]; `fs`/`sf` remain uncovered per [DR-0006]). Still owed: measured silicon ([DR-0004] Tier 3)
Qualitydesigned-for-SP 800-90B (raw access + RCT/APT + entropy-source model), plus a **simulation-derived design-stage min-entropy estimate** within the #10 claim limits; 90B validation itself deferred to measured silicon ([DR-0004]) · stretch: AIS-31 PTG.2 — same three-tier treatment (structure now, conformance deferred)
Conditioning**non-vetted** 32-bit CRC-32 LFSR compression (Galois, poly `0xEDB88320`), state cleared every block, **K = 8** — 256 raw bits in : one 32-bit word out. Creditable output entropy **0.85 bit per output bit** (SP 800-90B's non-vetted cap) for any raw stream at or above **H = 0.107 bit/sample**; a 4.70× margin under the H₀ = 0.5 target. ~0.005–0.008 mm² ([DR-0008]) · stretch: a 90B-*vetted* conditioning function — **rejected on area**: a compact serialised AES-128 is 88–124 % of the whole block budget ([DR-0008] §4). Live again only if the area budget grows
Delivered (post-conditioning) rate**`R_cond = R_raw / K` > 250 bps** at the raw-rate row's binding corner (`ss` / −10 % / +125 °C), K = 8 (re-derived at 2 kbps and retained: [K disposition](design/conditioner/README.md#k-at-the-2-kbps-raw-rate)); [DR-0003]'s original > 500 kbps at the stretch raw rate is **unattainable under the measured jitter-energy law** — it presupposes the > 4 Mbps raw stretch, itself unattainable (Raw rate row) ([DR-0003] §6, [DR-0008] §3, [DR-0011-rate])
Health testscontinuous RCT + APT on the **raw** stream, α = 2⁻⁴⁰, APT window W = 1024, cutoffs as formulas in min-entropy H (at H₀ = 0.5 → `C_RCT` = 81, `C_APT` = 824); failure latches a flag and gates the conditioned path until explicit clear + start-up test. The parameterization has a hard floor: **no valid APT cutoff exists at H ≤ 0.03** ([DR-0002])
Time-to-first-valid**≥ ~0.640 s** at 2 kbps — an arithmetic floor: 1024 consecutive raw samples for the start-up health test (0.512 s) plus 256 samples of conditioner latency (0.128 s), which do **not** overlap because the conditioner is held flushed while gated. Applies at power-on and after every alarm clear; binds at `ss` / −10 % / +125 °C (slowest sampling) ([DR-0002] §Failure behavior, [DR-0008] §7; rate from [DR-0011-rate]).
Power< 500 µW active, binding at `ff` / +10 % supply (fastest RO — max measured `f_osc` 2.30 GHz at −40 °C); < 1 µA idle, binding at `ff` / +10 % / +125 °C (max leakage)
Area< 0.05 mm²
Operating envelope−40 … +125 °C, 3.3 V ± 10 % (2.97–3.63 V). Every entropy, rate and health-test claim above holds **over this envelope and only over it**; the envelope is the security boundary, since an attacker chooses the operating point. Outside it, behavior is health-test-detected, not specified
Interfacestreaming, mode-selectable raw / conditioned (`OUT_MODE`), + register read (`DATA` conditioned, `RAW_DATA` raw); raw access always available and never gated ([DR-0001]). Instantiated as four word-addressed registers — `CTRL`, `STATUS`, `DATA`, `RAW_DATA` — plus a 32-bit valid/ready streaming port, with a health-test gate that flushes the conditioned path and **never** the raw one ([DR-0013])

Renders

Click a render to open an interactive, pan/zoom viewer of the actual GDS.

Overview render of Gf180 TrngClick to explore →
Overview
Comp render of Gf180 TrngClick to explore →
Comp
Contact render of Gf180 TrngClick to explore →
Contact
Layer 34/10 render of Gf180 TrngClick to explore →
Layer 34/10
Layer 36/10 render of Gf180 TrngClick to explore →
Layer 36/10
Metal1 render of Gf180 TrngClick to explore →
Metal1
Metal2 render of Gf180 TrngClick to explore →
Metal2
Metal3 render of Gf180 TrngClick to explore →
Metal3
Nwell render of Gf180 TrngClick to explore →
Nwell
Poly2 render of Gf180 TrngClick to explore →
Poly2
Via1 render of Gf180 TrngClick to explore →
Via1
Via2 render of Gf180 TrngClick to explore →
Via2
Zoomed crop render of Gf180 TrngClick to explore →
Zoomed crop

Metrics

Layers11
Cells9
Instances8

Downloads