AI Compass
Compass

AI on the device

Models on cameras, controllers and phones: what is possible there, what it costs, and why data protection changes fundamentally.

·2 min read·By Fachredaktion Technik
DETAIL
3 sections

The idea

Instead of sending an image to a server and waiting for a reply, the device computes itself. A camera recognises on site whether safety glasses are worn and reports only "yes" or "no". The image never leaves the camera.

What it is good for

  • Data protection: what is not transmitted cannot leak.
  • Latency: no network round trip, answers in milliseconds.
  • Availability: works without a connection.
  • Cost: no per-request fees.

What runs on what

DeviceTypical computeRealistic
Microcontroller with NPU0.1 to 2 TOPSKeyword spotting, simple sensing
Single-board computer with accelerator4 to 30 TOPSReal-time object detection
Current smartphone15 to 50 TOPSVision models, language models to ~3 bn in int4
Industrial PC with a small GPU50 to 200 TOPSSeveral camera streams, models to ~7 bn

The path from model to device

  1. 01

    Export

    To ONNX or a vendor format. This is where unsupported operations surface.

  2. 02

    Quantise

    Usually int8, with a calibration set drawn from real operating data.

  3. 03

    Compile

    Into the target hardware's format, optimised for its compute units.

  4. 04

    Cross-check

    Hold the device outputs against the original model on the same evaluation set. Deviations here are the rule, not the exception.

  • Always calibrate with real operating data, never synthetic.
  • Measure after each step, not only at the end.
  • Check thermal behaviour under sustained load; many devices throttle after a few minutes.

The energy budget

Battery run time

t [h] = C / ( P_idle + E_inf · f )

Run time is capacity divided by baseline draw plus the energy the inferences require.

C
battery capacity in watt hours
P_idle
baseline device draw
E_inf
energy per inference
f
inferences per hour

Worked through for a sensor camera: C = 20 Wh, P_idle = 0.15 W, E_inf = 0.0008 Wh, f = 3,600 per hour (one per second): t = 20 / (0.15 + 2.88) = 6.6 hours. With one inference every ten seconds instead: t = 20/(0.15 + 0.288) = 45.7 hours.

The trigger threshold is therefore the most important design parameter, not the model. A motion gate before the inference typically extends run time by a factor of five to twenty.

Field updates

  • Two storage slots on the device so a failed update rolls back.
  • Version, checksum and calibration state per device tracked centrally.
  • Staged rollout: one percent of devices first, then ten, then all.
  • A way to inspect a fielded device without removing it.
  • Signed model files; a model is executable code in its effects.

On-device processing does not automatically mean no personal data is processed. It is processed, just locally. What falls away are transmission and processing on behalf, two of the hardest points. What remains:

  • A legal basis for the processing itself is still required.
  • Information duties under Art. 13 GDPR remain.
  • Data subject rights apply, even if the data sits in the device only briefly.
  • As soon as a result is stored or transmitted, a new processing operation begins.

The practical difference is nevertheless large: a count without image transmission is routinely proportionate, while the same count transmitting images to a provider outside the EU routinely is not. See Data residency.

Related courses and sources

ToolFreeEN

llama.cpp

Running quantised models on ordinary hardware, down to single cards and small boards. The reference implementation for operating without a data centre.

For running without a data centre, down to single cards and small boards.

ToolFreeEN

Netron

Opens a model file and draws its structure. The fastest way to see what a delivered model actually contains.

For a quick look inside a delivered model before putting it into operation.

Was this page helpful?
AI on the device