“The ability to run AI models on power levels of a few milliwatts makes developing applications feasible with MCUs the most realistic approach possible.”

Hyunsoo Moon, Manager of STMicroelectronics, presented the 'STM32N6-based ST Edge AI Solution' at the '2025 e4ds Tech Day' event held on September 9, introducing technological advancements that enable the implementation of artificial intelligence algorithms even in low-power environments based on ST's flagship product line, the STM32 MCU.

ST is a global semiconductor company based in France and Switzerland that produces a variety of products, including power semiconductors, automotive MCUs, and analog and digital ICs.

In particular, the STM32 series is a globally widely used MCU product line, and has recently been attracting attention as a key platform for implementing edge AI.

Manager Moon Hyun-soo said,“Existing MCUs have been limited to simple hardware platforms for sensors and motor control, but they are now evolving into system-level intelligent platforms through AI algorithms,” he said.

He cited the core concept of AI as "machines performing human intellectual actions," and explained that algorithms capable of prediction and judgment are rapidly spreading across industries.

Edge AI is a method of performing inference on the device itself, as opposed to large-scale cloud-based computation.

ST focuses on this inference step and provides various solutions to enable AI functions to be implemented even on small platforms such as STM32 MCUs.

In particular, the STM32N6 series supports clock speeds of up to 800MHz and has a built-in NPU (Neural Processing Unit) to provide performance optimized for AI operations.

The advantages of edge AI are clear: real-time performance, no need for external transmission of sensitive data, and low-power operation.

ST is applying edge AI in three key areas.

The first is AI upgrades to existing MCU-based algorithms. For example, AI can be used to improve arc fault detection in solar power systems or to proactively identify signs of failure through predictive maintenance.

Second, the vision and voice recognition functions that were difficult to implement with existing MCUs are supplemented with the STM32N6 series.

The third area is the area of driving high-performance models that existing MCUs could not handle at all.

Actual cases were also introduced. An overseas home appliance company has commercialized a system that uses AI to determine the weight of clothes and whether they collide in a drum washing machine, implemented on an STM32G4 MCU.

Additionally, a Japanese electric bicycle manufacturer has developed a "virtual sensor" that uses motor data from an existing MCU to infer tire pressure. This is an example of expanding functionality solely through software, without separate hardware.

In 2018, ST launched STM32Cube.AI, which provides the ability to convert TensorFlow Lite or Keras-based models into C code suitable for MCUs.

In 2021, we launched NanoEdge AI Studio, enabling developers without machine learning experience to automatically create and deploy models using only a dataset.

“Implementing edge AI presents a variety of challenges, including model creation, data acquisition, performance management, security, and power consumption,” said Manager Moon Hyun-soo. “ST is helping customers easily access these challenges through hardware and software tools.”