NS-FL · Nanosatellite
On-board intelligence.
No OS. Under one watt. In orbit.
Neural inference that runs in the satellite, not on the ground. No operating system, no
framework, no cloud round-trip — the computation is in the logic gates. Decide on-board,
downlink only what matters.
0 OS
Software attack surface
<2 s
Cold-boot after reset
The orbital constraint
A nanosatellite has no cloud, a hard power budget, and minutes — sometimes hours — between ground-station windows. It must decide for itself, on a few watts, and survive without a maintenance technician for years.
A GPU/SoC running Linux brings the wrong trade-offs to orbit: high power, an OS that can fault, a software stack no certification body can audit line by line, and latency that varies with load. FLVH removes all of it — no OS, fixed latency, under one watt, auditable from the source.
Why FLVH belongs in space
Under 1 W — inference fits the power budget that makes on-board AI viable on a CubeSat. No OS, no driver, no framework — nothing in the software stack to crash, drift, or be exploited; the decision path is pure logic. Bit-exact & auditable — same input, same output, every time; a certification authority can verify the <5,000-line engine by hand. Model update by binary swap — push a new DDR binary on the next pass, no re-synthesis, no reflash of the fabric.
FLVH vs. SoC + Linux on-board
Criterion
SoC + Linux + framework
FLVH On-Board
Power (inference)
5–20 W
<1 W
Software attack surface
OS + drivers + stack
None
Latency
Variable (scheduler)
Fixed by construction
Auditability
Millions of lines
<5,000 lines Verilog
Recovery after upset
OS reboot, seconds–minutes
Cold boot <2 s
Model update
Reflash + redeploy
DDR binary swap
On-board mission cases
🛰
Earth-Observation Triage
Classify each frame on-board — cloud / no-cloud, target / no-target — and downlink only what matters. Saves scarce bandwidth and ground-station time; the satellite decides before the next pass.
◷
Autonomous Decisions
The satellite acts on what its sensors see, in orbit, without waiting for a command from Earth. Deterministic, low-power inference that runs on every pass, on its own — pure on-board AI.
⚠
Health & Anomaly Detection
Continuous inference on telemetry to flag anomalies before they become failures. Deterministic, low-power, always-on — no OS update can ever disable it.
▦
Debris & Proximity Awareness
Fixed-latency detection feeding the attitude/control loop. The decision arrives in a known, bounded time — exactly what a safety-critical maneuver requires.
The engine is proven on silicon
FLVH is the same inference core validated bit-exact on physical hardware — Zynq-7020 and AMD Kria KV260 (ZU5EV), the K26 SOM used across AMD's series deployments. Reference benchmarks (MNIST): 98.05% MLP, 99.06% CNN, 0.00% delta vs float64. Your mission model is trained for your sensor and your task — the engine, the determinism and the power envelope are already qualified.
Note: published benchmarks are reference workloads. Radiation-tolerance and flight-qualification levels are addressed per programme.
Evaluate FLVH on-board
A sealed, signed FLVH inference product on your KV260 / K26 bench — measure the power, latency and bit-exactness on your own hardware before flight.
Ground evaluation: $10,000 · 12-month KV260 bench license.
Flight & constellation: per-mission license, perpetual for the mission life — by discussion.
Request an evaluation →
BATEN Technologies can also propose a
not-yet-commercialized product,
based on a communication protocol developed by our team at
BATEN LABS —
CCP · Causal Chain Protocol.
Contact us to learn more →