Specifications
N6 MCU
STMicro STM32N6: Arm Cortex-M55 up to 800 MHz with Helium and MVE
N6 NPU
ST Neural-ART accelerator at 1 GHz delivering 600 GOPS
N6 Memory
4.2 MB SRAM plus 64 MB PSRAM at 800 MB/s
N6 Storage
32 MB flash at 200 MB/s plus microSD slot on the underside
N6 Camera
Replaceable 1 MP colour global shutter sensor, 120 FPS at VGA; supports sensors up to 5 MP
N6 Video
Hardware-accelerated H.264 and JPEG encoders
AE3 SoC
Alif Ensemble E3: dual Arm Cortex-M55 at 400 MHz and 160 MHz
AE3 NPU
Dual Arm Ethos-U55 microNPUs — 204 GOPs at 400 MHz and 46 GOPs at 160 MHz, usable simultaneously
AE3 Memory
13.5 MB SRAM with a 2D GPU for image scaling
AE3 Storage
32 MB flash at 200 MB/s
AE3 Camera
1 MP colour global shutter sensor, 120 FPS at VGA
AE3 Sensors
8x8 time-of-flight depth sensor with 4 m range plus a 6-axis IMU
AE3 Wireless
2.4 GHz Wi-Fi 4 and Bluetooth LE 5.1
AI Throughput
YOLOv8 and YOLOv11 at 30 FPS on the N6 and approximately 20 FPS on the AE3
Power (AE3)
60 mA at 5 V running YOLO at 30 FPS (0.25 W); ~30 mA idle; under 500 uA deep sleep
Battery Life (AE3)
Over 1 day at full power, roughly 3 days idling and more than 4 months in deep sleep on three AA cells
Wake Sources
Deep-sleep wake on sound, motion and real-time clock
Expansion
Qwiic/STEMMA QT connector, 10-pin GPIO board-to-board header, 3.3 V through-hole GPIO
AE3 Dimensions
2.54 x 2.54 cm board only
Other Family Models
OpenMV Cam RT1062 (30 uA deep sleep), OpenMV Cam H7 Plus, OpenMV Cam H7 R2, Arduino Nicla Vision
Toolchain
MicroPython with the OpenMV IDE; models deployed from TensorFlow Lite for Microcontrollers
Overview
OpenMV builds microcontroller-class machine-vision cameras, and the N6 and AE3 are the generation where that class became genuinely useful for AI. Both boards run object detection, human pose tracking and facial landmark models at frame rates that previously required a Linux SoC, while drawing tens of milliamps instead of watts — OpenMV states the two boards are more than 100x faster than previous OpenMV cams for AI workloads.
The N6 is the performance flagship. It is built on STMicro's STM32N6, pairing a Cortex-M55 running at up to 800 MHz with the ST Neural-ART accelerator at 1 GHz for 600 GOPS, backed by 4.2 MB of SRAM and 64 MB of PSRAM. That budget is enough for YOLOv8 and YOLOv11 at 30 FPS, and the replaceable camera module accepts sensors up to 5 MP. Hardware H.264 and JPEG encoders let it stream compressed video without host assistance.
The AE3 is the compact, power-optimised sibling. Alif's Ensemble E3 gives it dual Cortex-M55 cores and, unusually, two Ethos-U55 microNPUs that can run simultaneously — 204 GOPs plus 46 GOPs — inside a 2.54 cm square board. It targets roughly 20 FPS on the same YOLO models but does so at 0.25 W. The power profile is the headline: over four months of deep sleep on three AA batteries, waking on sound, motion or a timer. That combination is what makes always-on vision viable in devices that cannot be wired.
Both boards ship with an 8x8 time-of-flight sensor option (AE3), Qwiic expansion for sensors, and the MicroPython OpenMV IDE toolchain. Around them OpenMV sells a ladder of more modest cams — the RT1062 for multi-year battery deployments, the H7 Plus and H7 R2 for image processing without AI, and the Arduino Nicla Vision for the smallest footprint. QS Compute quotes the full line, and positions it below the Jetson Orin Nano tier where a multi-watt Linux module is simply not viable.
Key Benefits
30 FPS YOLO inference on a microcontroller-class board removes the Linux SoC from simple vision products entirely. 0.25 W running and sub-500 uA sleeping makes multi-month battery operation real. Global shutter sensors eliminate rolling-shutter smear on moving objects. Dual simultaneous NPUs on the AE3 allow sensor fusion or cascaded model pipelines in one package. Qwiic expansion and MicroPython cut integration time to days, and replaceable camera modules up to 5 MP on the N6 protect the design as requirements grow.
Applications
Always-on smart cameras and doorbell-class vision devices; wildlife and remote sensing with multi-month battery life; predictive-maintenance vibration and vision nodes; smart-agriculture and livestock monitoring; gesture and presence detection in appliances; robotic end-effectors and small AMRs; industrial barcode and label reading without a Linux host; research and teaching platforms for embedded AI.
Request a Quote — OPENMV N6 AND AE3 AI CAMERAS — MICROPYTHON MACHINE VISION WITH ONBOARD NPU
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →