Hsinchu, Taiwan – As industries increasingly recognize the potential of artificial intelligence (AI), endpoint AI – where AI models are deployed directly on devices – is becoming a major trend. It is being widely applied across smart home devices, smart cities, industrial automation, interactive toys, and wearable devices. These application scenarios require real-time data processing and analysis while maintaining low power consumption and high performance to ensure long-term stable operation. This growing demand has led to the rise of next-generation microcontroller (MCU), neural processing unit (NPU), and microprocessor (MPU) solutions. However, despite the growing market interest in endpoint AI, developers face many design challenges and pain points in implementing endpoint AI functionality. Traditional workflows often require extensive programming knowledge, a deep understanding of machine learning frameworks, and model optimization skills. These challenges can slow down innovation, making it difficult for developers to quickly prototype and deploy AI solutions.
To address these challenges, Nuvoton Technology has introduced the NuEzAI-M55M1 development board, powered by the NuMicro® M55M1 microcontroller, which is based on the Arm® Cortex®-M55 core. Paired with an online model training tool, developers no longer need extensive programming skills or a deep understanding of complex algorithms. Users can easily capture data in real-time using devices like cameras, and in as little as three minutes, train an image recognition model. Once training is complete, the online model training tool outputs the model in tflite int8 format, which can be directly deployed on the NuEzAI-M55M1 development board. The M55M1 microcontroller is equipped with a 220 MHz Arm® Cortex®-M55 CPU and a 220 MHz Arm® Ethos™-U55 NPU, designed to enhance AI processing power. It supports machine learning inference and the operations of CNN and RNN algorithms. The microcontroller features 1.5 MB of SRAM and 2 MB of flash memory and can be expanded using the HYPERBUS™ interface to add HYPERRAM™ or HYPERFLASH™.
The M55M1 microcontroller operates within a low supply voltage range from 1.6V to 3.6V and can function reliably in extreme temperatures ranging from -40°C to +105°C. To further optimize system power consumption, security, and performance, the M55M1 incorporates three unique features:
- When the CPU enters low-power sleep mode, image sensors, microphones, and various other sensors can continue operating, monitoring predefined events like human presence, sounds, or vibration detection.
- It allows machine learning model data to be stored in an area accessible only by the NPU, not the CPU, preventing malicious programs from stealing model data and protecting intellectual property.
- The M55M1 also implements hardware circuits for sine and cosine calculations, defined within custom Arm instructions, allowing application software to access these functions efficiently.
The NuEzAI-M55M1 development tool significantly simplifies the endpoint AI development process. Even developers with no prior experience in endpoint AI can easily create and deploy AI models. This platform is ideal as an early-stage development evaluation kit, enabling developers to get up to speed quickly and shorten time-to-market. Nuvoton Technology invites developers, engineers, and AI enthusiasts to lead the endpoint AI revolution and explore its limitless possibilities.
For more information about NuEzAI-M55M1, please visit https://www.nuvoton.com/ai/product/.
