[Editor's Note] As the use of artificial intelligence becomes widespread across industries, the need for solutions that simplify the development of machine learning applications is growing. While large corporations typically hire data scientists to collect massive amounts of data and build AI models over several months, smaller companies face the challenge of even hiring data scientists. To address this, a new solution has emerged: NanoEdge AI Studio, a utility designed for embedded developers without data engineering expertise. We met with Hyunsoo Moon, Manager at STMicroelectronics, to learn more about NanoEdge AI Studio.

Please introduce yourself briefly.

I am in charge of MCU technical support at STMicroelectronics (ST) Korea. I am responsible for providing technical support for customers using STM8/STM32 MCU products and for artificial intelligence solutions based on STM32 MCU.

While AI is a hot topic in the industry, adoption and utilization seem challenging for companies with limited budgets or inexperience. What challenges do you see in the field?

Until now, high-performance hardware systems were required for artificial intelligence processing, so they were implemented through server or cloud environments, which required hardware complexity and high costs.

However, recently, frameworks such as TensorFlow Lite have been developed to enable pre-trained neural network processing in MCU-based embedded system environments, making it possible to easily apply neural network models to MCUs and perform inference in actual applications.

However, developing MCU-based edge AI devices presents several challenges. Some are related to technical limitations, while others are purely technical or financial. All of these factors combined make the project overly complex, potentially raising concerns about the practical application of AI in MCU-based environments or creating a high barrier to entry.

While the process of collecting high-quality data requires a lot of effort, the biggest problem is the lack of development personnel with the capabilities necessary for AI development in the existing MCU-based embedded system development environment.

What features does ST's NanoEdge AI Studio provide?

NanoEdge AI Studio is a new machine learning (ML) technology that makes it easy for developers using STM32 MCUs to achieve true innovation. NanoEdge AI Studio is an intuitive software tool that enables system designers using ARM-based low-power microcontrollers to quickly and easily apply machine learning algorithms to a variety of applications, including connected products, home appliances, and industrial equipment. Even without machine learning skills or knowledge, the GUI-based NanoEdge AI Studio allows developers to create optimal ML libraries for their projects based on minimal data in just a few steps. NanoEdge AI Studio can create four types of libraries: anomaly detection, outlier detection, classification, and regression.
■Can ST's NanoEdge AI Studio be easily used even without data expertise?

One of the great advantages of NanoEdge AI Studio is that it doesn't require specific data science skills. Any software developer using NanoEdge AI Studio can create optimal ML libraries in a user-friendly environment, completely free of artificial intelligence (AI) skills.

Because ML library creation is possible based on Arm® Cortex®-M0/M0+/M3/M4/M7, ML libraries created through NanoEdge AI Studio can be used on all STM32 MCUs. Additionally, it is possible to create highly accurate ML libraries even with very small datasets.

Developers using STM32 MCUs can automatically generate a C code-based ML library by simply entering the required dataset into NanoEdge AI Studio.

■What are some specific application cases for NanoEdge AI Studio?

Recently, the NanoEdge AI machine learning library has been increasingly used in electric vehicle platforms. Prediction and monitoring libraries are being used based on the vast amount of data loggable on EV platforms, such as torque prediction and stator winding temperature prediction in EVs. NanoEdge AI's machine learning libraries are being applied in diverse fields, such as human behavior recognition in smartphones and energy consumption prediction in the energy sector.

NanoEdge AI's machine learning algorithms are also being used in many other fields, including behavioral recognition, chemistry, energy, healthcare, industry, and smart homes.

■Why is edge AI necessary in today's world, and what are your future prospects?

Recently, MCUs have become the core of edge devices, and they are always used in systems at the edge. Therefore, the inclusion of AI capabilities in this area enables improved application implementations in various areas, such as powerful performance and increased efficiency.

By improving existing algorithmic methods, we provide users with a new user experience. Because input data, such as sensor data, can be inferred directly from edge devices containing MCUs using AI-based algorithms, the costs required for server or cloud connection are reduced. Additionally, since no network connection is required, it can maintain high stability in terms of security.

In this way, the demand for AI-based processing and its integration with IoT platforms in the edge device sector is expected to grow rapidly.

■What differentiated solutions does NanoEdge AI Studio plan to expand its reach in the future?

NanoEdge AI Studio is designed to enable developers using STM32 MCUs to create high-accuracy ML libraries without requiring additional AI knowledge or technical expertise. The process of creating these libraries is simple and straightforward. To enable the creation of high-accuracy ML libraries with small datasets, various ML algorithms will be developed and existing ones will be continuously updated.

■Finally, please say a word to e4ds readers.

ST offers NanoEdge AI Studio, which allows developers familiar with embedded development environments to easily integrate AI solutions into their products. ST plans to continue updating its AI-related solutions and demos. As the importance of edge devices grows, ST will be a satisfactory alternative for building a platform capable of on-device AI model-based inference.

thank you