01specs02arduino03modules04neuro studio05open source
Pre-order from £89 GitHub

Neuro N6

Vision AI on an Arduino-compatible board. Run real-time neural networks from a single sketch, with no cloud and milliwatts of power.

Pre-order from £89 View on GitHub

// if you can blink an LED, you can run a neural network

Neuro N6 Arduino-compatible STM32N6 edge AI development board

600 GOPS

Peak NPU compute

40 ms

YOLOv8n inference

~750 mW

Total power draw

Powered by the STM32N6.

An 800 MHz Arm Cortex-M55 and 1 GHz Neural-ART NPU, plus a built-in ISP, H.264 encoding and a NeoChrom GPU. A complete vision pipeline on one board.

cpu
Arm Cortex-M55 at 800 MHz
npu
Neural-ART at 1 GHz, up to 600 GOPS
isp
Cameras up to 5MP
video
Hardware MJPEG and H.264 encoding
gpu
NeoChrom, for image manipulation
memory
64MB OSPI RAM, 32MB flash
format
Arduino-compatible, Feather form factor
power
About 750 mW in total
Download datasheet
Neuro N6 sketch running in the Arduino IDE

One of the first Arduino boards with a dedicated NPU.

Camera capture, pre-processing, inference and results all live in a single Arduino sketch. Our libraries handle the STM32N6 peripheral setup, so there is no DMA configuration or register work.

Visit the wiki

From blink to pose estimation.

The same setup() and loop() you already know. Flip between the first sketch everyone writes and a YOLOv8 pose model running on the NPU.

setup()
Runs once: camera, model and post-processing
loop()
Runs forever: capture, infer, overlay, stream
27 lines
#include <OV5640_Arduino.h>
#include <PostProcess.h>

// Compile the model into the sketch from a .tflite file
#pragma neuron6 model="yolov8n_256_quant_pose_coco.tflite" name=yolov8_mpe

OV5640 camera;
// YOLOv8 multi-pose estimation
NEURON6_DECLARE_MODEL(yolov8_mpe);

void setup() {
  VisionConfig cfg;
  cfg.camera    = CAMERA_WVGA;
  cfg.fps       = FPS_60;
  // Stream to Neuro Studio
  cfg.transport = TRANSPORT_USB;

  // Confidence and IoU thresholds
  Vision.begin(&NN_Instance_yolov8_mpe,
               PostProcess_YOLOv8_MPE(0.4f, 0.5f),
               cfg);
}

void loop() {
  // Capture, infer, overlay, stream
  Vision.run();
}

Vision AI, made simple.

Up to 600 GOPS of on-device AI in an Arduino-compatible board, with a family of swappable vision modules.

Neuro Vision OV5640 camera module
Neuro Vision wireless OV5640 camera module
Neuro N6 edge AI development board
Neuro Vision global-shutter camera module
Neuro Vision radiometric thermal camera module

Pick your sensor.

Neuro Vision OV5640

  • 5MP rolling shutter with autofocus
  • 67° field of view
  • Time-of-flight distance sensor
  • Strobe LED

£29Connects directly to the back of the Neuro N6

Neuro Vision OV5640
Neuro Vision OV5640 Wireless
Neuro Vision ST Cam
Neuro Vision Thermal

What will you build?

600 GOPS in a Feather form factor, drawing milliwatts.

  • Wildlife and trail cameras

    Build low-power, battery-operated wildlife cameras using on-device AI and duty-cycled operation.

  • Autonomous robots

    Perception on-board, with no laptop or Wi-Fi tether.

  • Wearable devices

    Activity recognition on a board small enough to wear.

  • Smart cameras and security

    Local alerts, with no subscription or cloud footage.

  • Sound and keyword detection

    Always-listening audio models with millisecond inference.

  • Custom defect detection

    Train and deploy your own inspection models.

Neuro Studio.

It ships with the board and will be open source. Connect over USB or Wi-Fi and see exactly what the NPU is doing: boxes, labels, confidence, frame rate and timing, live, with no configuration.

  • Live camera preview over USB or Wi-Fi
  • Real-time inference overlay
  • Frame rate and timing metrics
  • Toggleable bounding boxes
  • Snapshot and video recording
  • macOS · Windows · Linux

Open-source hardware and software, with design files being released as development progresses.

The Neuro N6 schematics are already on GitHub, and the Arduino core and Neuro Studio will follow.

github.com/ohmlab-ltd

ohmlab-ltd/

  • Arduino CoreFinal testing

    Full library support for the STM32N6 NPU, camera and peripherals. Install it from the Arduino IDE Boards Manager.

  • Neuro StudioFinal testing

    Desktop app for live inference visualisation, debug overlays and session recording. Will be open source.

  • Hardware designPost-launch

    Schematics and PCB files for the Neuro N6 and every Neuro Vision module.

  • Model workflowIn development

    Train, quantise and deploy your own models. Docs and tooling follow as the workflow matures.

Software tooling for edge AI.

PixelKit preps and labels datasets, Neuro Studio shows your model running on real hardware, and an open Arduino core ties it into Neuro N6. We built each one because we needed it.

  1. PixelKit segmentation labels over motorway traffic

    1data

    PixelKit

    Prepare and export training data for vision models.

    Try PixelKit for free

  2. 2code

    Arduino Core

    Build with the Arduino core and libraries. The schematics are already on GitHub.

    Explore the files

  3. 3debug

    Neuro Studio

    Stream frames, detections, confidence and timing from Neuro N6 over USB or Wi-Fi.

    See Neuro Studio

Get the Neuro N6.

Pre-order now, or join the mailing list for development updates, early Neuro Studio access and Arduino library drops.

Pre-order from £89

// no spam, unsubscribe any time.