600 GOPS Dedicated Micro NPU
Powered by the Neural-ART acceleration engine, easily running mainstream vision models (YOLO, MobileNet) while freeing the Cortex-M55 core to focus purely on business logic.
Beyond high-power MPU architectures. Powered by the STM32N6 architecture — 600 GOPS Neural-ART NPU, 5MP Hardware ISP and 1080p H.264 Encoder — we provide end-to-end hardware ODM customization and co-development services, from schematic design, PCB layout and BSP/algorithm porting to PCBA volume manufacturing.
Skip the high-power Linux MPU architecture. This single-chip STM32N6 Edge AI platform combines a 600 GOPS Neural-ART NPU, 5MP hardware ISP and 1080p H.264 encoder with RTOS millisecond-level instant boot — ready to drop into your product.
Powered by the Neural-ART acceleration engine, easily running mainstream vision models (YOLO, MobileNet) while freeing the Cortex-M55 core to focus purely on business logic.
STM32N6 264-BGA with Flash-less architecture — high-speed signals (Octo-SPI, MIPI CSI-2, Gigabit Ethernet) optimized for ultra-compact SOM and COB designs.
Meticulous multi-rail power design optimizing 1.0V / 1.8V / 3.3V Buck/LDO circuits — total system power at 2~3W to support fanless, fully sealed industrial thermal enclosures.
5MP hardware ISP with AWB, Denoising and AE — a single camera sensor generates 3 simultaneous resolution streams for NPU inference, LCD display and remote H.264 streaming.
SESIP / PSA Level 3 with TrustZone architecture — model weight encrypted storage and hardware decryption on-the-fly (Model Protection) to protect your IP.
One-stop component sourcing, SMT assembly, AOI inspection and FCT fixture development — DFM/DFA checks prior to SMT to ensure high manufacturing yield.
How the STM32N6 Edge AI custom architecture stacks up against traditional Linux MPU + external NPU and low-end MCU approaches.
| Comparison Dimension | Traditional Linux MPU + External NPU | Traditional Low-End MCU | STM32N6 Edge AI Custom Architecture |
|---|---|---|---|
| System Complexity | Relies on Linux OS — slow boot (10s+), complex OS loading | RTOS / bare-metal, but lacks AI inference | RTOS ms-level instant boot with high real-time performance & stability |
| Power & Thermal | 5W~15W, requires active fan or large heatsinks | <0.5W, but cannot run complex neural networks | Low to 2~3W, enabling industrial fanless sealed enclosure |
| Vision Capabilities | Software encoding or external ISP, high overall cost | No hardware ISP, extremely low resolution (QVGA) | Integrated 5MP ISP + 1080p H.264 encoder + GPU |
| Estimated BOM Cost | Many external components, 8+ layer PCB, high cost | Low cost, but insufficient compute for AI workloads | Single-chip integration with 4–6 layer PCB, significantly lower BOM |
From Industry 4.0 to smart security, transportation and HMI — the STM32N6 Edge AI architecture adapts to your product's environment and interface needs.
Industrial AI SOM & gateway ODM with dual CAN-FD, RS485, Gigabit TSN Ethernet and 24V wide-voltage input. 5MP ISP + 600 GOPS NPU delivers 30 FPS defect detection on moving assembly lines with real-time GPIO/CAN-FD output, plus 1D-CNN predictive maintenance (PdM) on vibration and audio.
Compact vision AI camera PCB ODM with 5MP MIPI sensor, H.264 hardware encoder and local TF card / SPI NOR storage, supporting PoE or battery power. On-device person & fall detection and pose estimation (YOLOv8n-Pose @ 26 FPS) keep video data at the edge — uploading only metadata and alerts. The same STM32N6 platform powers our VEIN2 palm-vein smart lock.
Automotive-extended hardware ODM with enhanced thermal dissipation for –40°C to +125°C, vibration-proof industrial connectors and HDR camera support. Ideal for roadside & in-vehicle units: license plate recognition, vehicle counting and Driver Monitoring Systems (DMS) in high-temperature outdoor or in-cabin environments.
Vision AI + display controller ODM with on-board MIPI DSI / RGB display and capacitive touch driver. The 2.5D NeoChrom GPU renders smooth 30+ FPS UI alongside gesture recognition camera support — enabling touchless, camera-based control in sterile, wet or industrial environments.
A transparent, milestone-driven process from requirements to volume PCBA delivery.
Feasibility analysis on product functions, dimensions, interfaces and target BOM cost — delivering an ODM Technical & Cost Evaluation Report.
Schematic capture, PCB topology layout and thermal / power / EMC simulation to produce Gerber files and BOM.
Producing 5~10 prototype PCBAs with camera ISP tuning, BSP / driver porting and AI model quantization.
Production burning / test fixtures and FCT automated testing setup for small-batch trial runs and volume PCBA delivery.
Quick answers to the questions we hear most often about STM32N6 Edge AI ODM.
STM32N6 integrates a 600 GOPS Neural-ART NPU, 5MP hardware ISP and 1080p H.264 encoder into a single chip. It boots in milliseconds on RTOS (vs 10s+ for Linux), runs at 2~3W (vs 5W~15W) enabling fanless sealed enclosures, and achieves a 4–6 layer PCB with significantly lower BOM.
Mainstream vision models like YOLO and MobileNet run natively. Benchmark results include YOLOv8n-Pose @640×640 at 26 FPS, MobileNetV2 @224×224 at over 200 FPS, PicoSAM2 real-time segmentation at 13.7 ms latency, and audio keyword spotting under 10 ms.
Yes. We convert customer neural networks (PyTorch / TensorFlow / ONNX) via ST Edge AI tools with INT8 quantization to run natively on the Neural-ART NPU, including BSP & RTOS porting and multi-stream video pipeline integration.
MIPI CSI-2 camera modules adapted for 5MP sensors (OV5640 / IMX335) with engineering-grade ISP tuning. The hardware ISP generates 3 simultaneous resolution streams for NPU inference, LCD display and remote H.264 streaming.
STM32N6 provides financial-grade hardware security (SESIP / PSA Level 3) with TrustZone architecture, supporting model weight encrypted storage and hardware decryption on-the-fly (Model Protection) to protect your intellectual property.
Requirements & BOM assessment, hardware design & PCB layout, prototype fabrication & tuning (5~10 PCBAs with ISP tuning, BSP porting and AI quantization), then volume manufacturing & FCT testing for small-batch trials and volume PCBA delivery.
Talk to our engineering team about STM32N6 Edge AI hardware, joint R&D and ODM volume programs.
This website provides information about Qumtrix's IoT engineering, OEM/ODM development and manufacturing services.
The text, images, logos, graphics and technical content on this site belong to Qumtrix or the relevant rights holders and may not be copied or used commercially without authorisation.
Product specifications, technical information and other content on this website are provided for general informational purposes and may be updated without prior notice.
We are not responsible for the content or availability of third-party websites.
We make reasonable efforts to keep the information on this website accurate and up to date, but we do not warrant that all information is complete, current or error-free.
For questions regarding these Terms of Use, please contact us at sales@qumtrix.com.
When you send an inquiry or a partner application: company, your name, business e-mail, country, the products or quantities you are interested in, your message and an optional phone number. When you subscribe: your e-mail address. For anonymous statistics: the page, an approximate country, reading time and clicks.
To answer your inquiry and perform our contract (Art. 6(1)(b)), and for anonymous statistics under legitimate interests (Art. 6(1)(f)) which you can switch off at any time. Analytics details are deleted after 400 days, the full IP address is masked after 30 days, audit logs after 730 days.
Access, correction, export, deletion, objection and restriction. Write to admin@qumtrix.com — we answer within 30 days.
qumtrix.com sets no advertising or tracking cookies. Visits are counted anonymously — page, approximate country, reading time, clicks — with no cookie and nothing stored on your device for measurement.
qm_cookie_consent in your browser's local storage remembers whether you switched statistics on or off. It is strictly necessary, stays on your device and is never sent to us.
Use Cookie settings in the footer at any time; switching statistics off takes effect immediately and no further measurement is sent. Do Not Track and Global Privacy Control are always honoured.
Every unit carries a 2-year warranty from the shipment date. Defects caused by the product itself are repaired or replaced free of charge; damage caused by human factors that needs parts replaced is charged at parts cost only. Mechanical spare parts are guaranteed to be available for 15 years.
Send a written request with the order number, product details, quantity, reason and supporting documents (photos, video, test data) within 15 business days after the issue is found. No goods may be returned without our written approval and an RMA number.
Return freight, destination duties and re-assessed Chinese import duties and VAT follow who is at fault. If the total return cost exceeds 15% of the goods' residual or contract value, we may refuse the return or ask you to dispose of the goods locally, at your cost.