Nuvoton Launches NuML Toolkit: Accelerating AI Model Deployment on MCU
As AI technology increasingly finds its way into embedded systems, efficiently deploying trained AI models on resource-constrained microcontrollers has emerged as a major challenge. Nuvoton has introduced the NuML Toolkit, specifically designed for the NuMicro® M55M1 series MCU, enabling quick conversion of pre-trained AI models into executable code to simplify endpoint AI deployment.
Background
As artificial intelligence (AI) technology continues to advance, developers aim to rapidly deploy machine learning models onto embedded systems. However, the significant differences between traditional AI development environments and resource-constrained microcontroller unit (MCU) platforms make model deployment complex and time-consuming. To lower development barriers, Nuvoton has introduced the NuML Toolkit, a Microsoft Windows® tool specifically designed for the NuMicro® M55M1 series microcontrollers, enabling quick conversion of pre-trained AI models into executable code compatible with Arm® Keil®.
Core Hardware Platform: M55M1
The NuML Toolkit is optimized for the NuMicro® M55M1 microcontroller, which features:
- Arm® Cortex®-M55 core — the most AI/ML-capable core in the Cortex-M family
- Integrated Arm Ethos™-U55 NPU — a neural processing unit that significantly enhances AI inference performance
- 1.5 MB SRAM + 2 MB Flash — providing ample computational and storage resources for AI applications
This hardware combination makes the M55M1 the latest production-ready Endpoint AI platform for machine learning.
Key Capabilities of NuML Toolkit
The NuML Toolkit integrates Keil µVision®5 / Arm Compiler 6 (armc6), Make / GCC toolchains, and can convert fully INT8-quantized AI models into Keil example code. Three key advantages:
- Easy Model Conversion — Just a few commands to generate AI MCU projects, no need to write conversion code
- All-in-One Workflow — From model conversion, project generation, flashing to result validation, all steps are built-in and simple
- Flexible & Editable — Modular code design for easy customization; large models can automatically leverage SD cards or HyperRAM
Typical Application Scenarios
The NuML Toolkit has been widely adopted across various use cases:
- Object Detection — Recognition models for identifying specific objects
- Keyword Spotting (KWS) — Voice control and wake word detection
- Anomaly Detection — Industrial equipment monitoring and abnormal behavior alerts
- Posture Recognition — Sitting posture detection in consumer electronics like smart desk lamps (selected as The Best AI Awards finalist)
Developers no longer need to worry about converting PC-trained models into MCU-executable formats. They can focus on writing peripheral control code (cameras, audio processing, display outputs), significantly enhancing project development efficiency.
Learn More
If you are interested in this tool, please visit Nuvoton AI webpage www.nuvoton.com/ai and use the "Contact Us" form for more information.
Content adapted from Nuvoton official technical blog.