The new GUI tool comes with features for automated AI development and deployment of Nuvotion MCU firmware.
Nuvoton recently announced a new graphical user interface (GUI) tool to help developers build machine learning applications for its line of Nuvoton microcontrollers. NuML Studio combines a variety of endpoint AI tools into a single platform. It includes machine learning (ML) operations such as data collection, model training, and automatic project generation in Keil/VS Code. NuML Tool, the existing deployment tool, is also included.
Object recognition with Nuvoton endpoint AI MCUs.
NuML Studio operates on Windows and supports the NuMicro M55M1 series microcontrollers. The M55M1 MCUs are powered by an Arm Cortex-M55 CPU and an Ethos-U55 neural processing unit (NPU). They come with 1.5 MB of SRAM and 2 MB of flash. The MCUs are designed for embedded endpoint AI applications.
Key Features of NuML Studio
Employing AI development tools is often a multi-step process that includes installing Python and libraries along with the development environment. Nuvoton wrapped the new tool into a single-step installation that automatically handles those details. While that may not seem like a significant feature to some, it makes the onboarding process much easier for those new to the Nuvoton MCU universe.
NuML Studio manages the complete development workflow. It covers model conversion to MCU programming and firmware deployment. Much of the embedded AI development process involves converting PC-trained AI models to MCU-compatible software. NuML Studio automates the conversion of INT8 models to Arm Keil project templates. It supports both the Keil and GCC toolchains.
NuML Studio development block diagram.
The output is deployed to MCUs for real-time inference in an embedded application. The tool includes MCU programming and verification of AI inference results. It supports models within the MCU’s internal resources, on external memory chips or SD cards, or in the cloud.
Challenges Facing Endpoint AI Developers
Adding AI to embedded systems is not like changing languages or moving to a different MCU family. AI functionality requires a diverse set of tools and a broader worldview of the application. Data, data gathering, and data management play a much greater role than in non-AI applications.
A conventional embedded system takes sensor input, runs it through algorithms, and makes decisions from there. AI systems typically start with a model that may be as complex, or even more complex, than the entire non-AI application. The embedded MCU may be responsible for creating or augmenting the model using large amounts of sensor data, or it may spend its cycles processing input against the AI model. Much of this is unfamiliar territory for a typical MCU developer.
NuML Studio takes on the challenges from three perspectives.
First, it’s an all-in-one development tool installation to reduce barriers to entry and allow developers to reach productivity more quickly.
It also offers broad sensor support, automated data conversion, and cloud integration. The tool automates data collection and conversion and helps developers to place it locally or in the cloud, as is most appropriate for the application.
Finally, automated project generation reduces the developer’s workload in a complex AI environment. The tool supports the popular TensorFlow Lite Micro (TFLM) framework and quantized AI models. The tool can automatically generate optimized firmware.
Nuvoton Endpoint AI Applications
The software developed with NuML Studio is expected to serve applications such as voice keyword spotting (KWS), visual object identification, and equipment behavior monitoring. In one application noted by Nuvoton, a home blood pressure monitor automatically recognizes the person under test and stores and organizes the results along with the person’s identity. Other applications include gesture identification for operating lamps and small appliances, as well as conference room participant counting.
Along with the easy-to-install studio, Nuvoton offers the NuMaker M55M1 W400 development board. The board, combined with the development software, gives developers a fast path to code development.
All images used courtesy of Nuvoton.
Trending Products
