Nuvoton Technology introduces an endpoint AI platform based on microcontrollers, expanding the AI ecosystem into the microcontroller domain. This solution is based on Nuvoton’s newly designed microcontrollers and microprocessors, including the NuMicro MA35D1, NuMicro M467, and the NuMicro M55M1 series equipped with Arm Ethos-U55 NPU. Nuvoton provides a complete software stack and development tools to facilitate the rapid deployment of advanced machine learning and deep learning models, leveraging advantages such as low power consumption and cost-effectiveness to accelerate the adoption of AI applications, enhancing productivity, and improving human life.
As a leading microcontroller platform provider, Nuvoton not only offers advanced hardware chips but also provides developers with complete software development tools, significantly improving development efficiency. NuEdgeWise is an easy-to-use graphical Python machine learning development tool that provides rich machine learning sample code covering processes such as data collection, labeling, model training, and validation, making the machine learning development process easier.
Nuvoton’s latest microcontroller endpoint AI platform is designed to elevate a wide array of applications, including smart home, security access control, smart city, industrial automation, smart agriculture, interactive toys, fitness equipment, and wearable devices, by infusing embedded system products with advanced AI capabilities.
Endpoint AI microcontroller with Ethos-U55 NPU: NuMicro M55M1
The new NuMicro M55M1 series microcontroller is an innovative Endpoint AI solution that integrates comprehensive microcontroller features, including control, connectivity, security, and advanced machine learning inference capabilities. The M55M1 microcontroller features a 200 MHz Arm Cortex-M55 CPU and a 200 MHz Arm Ethos-U55 NPU, providing machine learning inference capability and supporting CNN and RNN operations. It includes built-in 1.5 MB SRAM and 2 MB flash memory and can expand HYPERRAM or HYPERFLASH via the HYPERBUS interface. To enhance the overall performance of application systems, the M55M1 microcontroller incorporates three unique features to optimize system performance, security, and power consumption. Firstly, it enables continuous operation of image sensors, microphones, and various sensors in a low-power sleep mode, allowing for constant monitoring of predefined events such as personnel presence or significant sound and vibration detection. Secondly, it stores machine learning model data in an area accessible only to the NPU but not to the CPU, to prevent malicious programs from stealing model data and thus protect intellectual property. Lastly, the M55M1 also implements sine and cosine hardware circuits, which are defined within Arm’s custom instructions for easy invocation by software. These unique features of M55M1 empower developers to develop endpoint AI applications that achieve performance, power efficiency, and security.
High-performance edge industrial IoT series: NuMicro MA35D1
The NuMicro MA35D1 series heterogeneous multicore microprocessor is designed to meet the high-end industrial IoT requirements, featuring dual-core Arm Cortex-A35 64-bit processors with a maximum frequency of 800 MHz and a 180 MHz Arm Cortex-M4F core. Combined with a USB camera and CNN models, MA35D1 can perform endpoint AI tasks such as object detection.
The Ethernet/Crypto MCU with excellent security and connectivity: NuMicro M467
The M467 series features a 200 MHz Arm Cortex-M4F core with a built-in DSP instruction set and a single-precision floating-point unit (FPU). With the tinyML software technology, the M467 can perform various endpoint AI applications, such as gesture recognition, equipment anomaly detection, and keyword spotting. The M467 series microcontrollers have also participated in the MLPerf Tiny Benchmark test, demonstrating excellent inference speed across four endpoint AI tasks.
Complete machine learning development tools – Accelerating the implementation of AI applications
In addition to innovative microcontroller specifications, Nuvoton also supports a complete machine-learning software development stack for developers to develop machine-learning applications. The software stack includes NuEdgeWise Python development environment and machine learning sample code, Tensorflow machine learning model training framework, Vela neural network compiler dedicated to Ethos NPU, Tensorflow Lite for microcontroller inference framework, Arm CMSIS-NN machine learning library, and Ethos-U55 NPU driver.
Nuvoton’s endpoint AI microcontrollers enhance products across a diverse range of applications – including smart home, security access control, smart city, industrial automation, smart agriculture, interactive toys, fitness equipment, and wearable devices – by seamlessly integrating AI capabilities to deliver added value.
