[Editor's Note] Recently, on-device AI-based products are rapidly being built up across various fields. In this context, it is time for hardware-centric companies to utilize solutions that enable the easy and rapid implementation of AI capabilities, thereby shortening product development timelines and supporting preemptive market entry. On the 15th, the 2024 Start On-Device AI (hereinafter 2024 SODA) conference was held under the auspices of e4ds news, and insights were shared on solutions for lightweighting and optimizing AI models for installation on low-power, low-spec boards.

We spoke with Manager Hyun-soo Moon of STMicroelectronics, who gave a presentation on Edge AI at SODA 2024, about ST's Edge AI technology and development cases.

■ Introduction to ST's Edge AI Technology and Solutions

ST provides hardware and software that enable the processing of machine learning algorithms, including deep learning, even on small embedded platforms with small memory sizes and computational performance, such as MCU and MPU products.

With STM32Cube.AI, you can convert pre-trained deep learning models created in general-purpose frameworks such as TensorFlow Lite on a host PC into C-code-based models and apply them to STM32 products.

NanoEdge.AI Studio provides an AutoML-based pipeline that automatically generates data preprocessing, algorithm recommendation and training, and even code applicable to STM32 products, as long as you prepare the input dataset required for the machine learning model you wish to build, even if you have no experience or knowledge of machine learning.

Through STM32Cube.AI and NanoEdge AI Studio, customers using STM32 products can quickly and easily apply machine learning algorithms to STM32 products.

■ What are the key market/industry requirements, and what are the differentiations and unique features compared to competitor products (or existing products)?
Regarding the hardware that constitutes Edge AI, there is a wide variety of manufacturers, product types, and solutions offered. The core of Edge AI lies in lightweighting pre-trained machine learning models to minimize loss and configuring hardware capable of accelerating machine learning model computations at the edge.

ST's STM32 product family features a diverse lineup based on low power consumption, enabling edge computing with minimal power usage. In particular, the software tools of STM32Cube.AI and NanoEdge AI Studio allow users to easily and quickly apply machine learning models to STM32 products, and various options are provided for lightweighting to minimize model loss.

■ Introduction to Key Applications and Representative Use Cases Encountered in Industrial Sites and Daily Life

Edge AI is being used in small-scale applications such as wearables, home appliances, and robots, and is being applied in various fields such as industrial equipment, smart homes, smart buildings, and smart cities.

A representative example is predictive maintenance implemented by installing additional modules on motors. Its operating principle involves detecting motor vibrations using MEMS sensors in these modules, which can detect even minute abnormal vibrations that are imperceptible to humans.

There is a system that uses such minute motor vibration data to notify you of the maintenance time before a breakdown occurs. As such, aftermarket edge AI devices, which simply add an edge AI module to existing devices, require low investment costs and consume very little power.

A common home appliance is the drum washing machine. By applying Edge AI, energy efficiency can be increased by up to 40%. Although it is difficult to measure the actual weight of laundry in a drum washing machine due to the rotating drum, this is an AI system that optimizes energy efficiency by measuring the current during the motor's initial operation to predict the weight and determining the appropriate amount of detergent, water, and wash time. This results in higher efficiency, reduced energy consumption, and lower maintenance costs.

In the security field, using CCTV with an STM32N6 MCU, the device itself can independently recognize video images without using the cloud. It automatically recognizes cars, trucks, buses, motorcycles, and pedestrians, and displays them by color.

It operates at 5MP HD quality at 18 FPS, consumes 0.5 watts of power, and costs less than 1 cent per day. Introducing a CCTV system with added AI capabilities to a smart farm automatically detects the appearance of wild animals and notifies the farm owner.
■ Explain the future development direction of the solution and the target market

Edge AI is driving innovative changes across various industries and can significantly enhance the user experience in multiple aspects, such as real-time data processing, improved maintenance, and enhanced safety. ST is continuously updating its MCU capabilities to target the on-device AI market, which incorporates AI functions.

In addition, ST plans to intensively target the on-device AI market with new products featuring enhanced security performance. With security standards set to be strengthened in the U.S. and Europe starting next year, the newly launched low-power, high-performance MCU product family features enhanced security capabilities as well as power efficiency and performance. We plan to expand its applications not only in healthcare, door locks, and smart homes but also in various high-performance AI fields.

■ Status of ST's On-Device AI Ecosystem/Collaboration

While Edge AI offers many advantages, AI developers must carry out various development tasks from start to finish. Developing such Edge AI requires diverse hardware and software tools, which can be addressed through the ecosystem.

ST operates a very broad and comprehensive ecosystem that supports boards, software, drivers, libraries, and more. Since customers are best at application development, ST plays the role of providing tools to enable them to develop applications effectively.

Through this ecosystem, developers can efficiently and easily develop edge solutions in various environments and reduce development time.

■ What are your thoughts on participating in the event regarding the presentations at SODA 2024?

Through our participation in the 2024 SODA event, we were able to reaffirm the infinite growth potential of Edge AI. It was particularly meaningful as, from the perspective of a manufacturer producing hardware for Edge AI platforms, it was a valuable opportunity to directly introduce to customers the strengths of combining rapidly developing Edge AI technology with STM32 products.

ST will always spare no effort in providing the best support to ensure that customers' products can connect with the world faster and smarter through machine learning models.