Welcome Here  Shenzhen Mingjiada Electronics Co., Ltd.

With edge AI on fire, Tiny ML and NPU MCUs are following suit. As early as 2022, ST had announced that STM32N6 would be the first ST product to include NPU, and many engineers were eagerly waiting for this product.

At the STM32 Summit, the STM32N6 was officially unveiled. ‘Our definition of STM32N6 is the first high-performance STM32 MCU with AI acceleration capability,’ said Lisa DING, Microcontroller Product Marketing Manager, Microcontrollers, Digital ICs and RF Products Division (MDRF), STMicroelectronics, China, at the recent STM32 media conference. DING, Microcontroller Product Marketing Manager, MDRF, STMicroelectronics China, said.

The most powerful STM32 family ever!

The STM32N6 is the ‘most powerful’ STM32 product ever announced by STMicroelectronics (ST), designed to make AI at the edge available everywhere. ) with powerful AI acceleration capabilities, making it particularly suitable for edge AI applications.

Key features and benefits

High-performance processor: the STM32N6 is powered by the Arm v8.1 architecture Cortex-M55 core with up to 800MHz and performance of 3360 Coremark and 280 DMIPS.

Dedicated NPU: Integrated Neural-Art Accelerator provides up to 600 GOPS of AI arithmetic, equivalent to 600 times that of the STM32H7, with an excellent energy efficiency ratio of 3 TOPS/W.

Low-power design: No heat sink is required when running AI models, achieving a balance between high performance and low power consumption.

Rich interfaces and features: Includes MIPI CSI-2 camera interface, Image Signal Processing (ISP), H.264 hardware encoder, JPEG codec, etc. for a wide range of multimedia and computer vision applications.

Security features: Integrated Arm TrustZone, supports target authentication and PSA authentication to ensure the security of the device.

Application Scenarios

STM32N6 is suitable for a variety of edge AI application scenarios, including but not limited to:

Smart home: for precise control of smart devices through voice recognition and image recognition.

Industrial automation: for machine vision, predictive maintenance, and process control. 5.

Wearable devices: for smart health monitoring and personalised services.

In-vehicle systems: for autonomous driving, pedestrian detection and intelligent navigation.

Making Edge AI Available Everywhere

Generative AI, big language models, and cloud AI are the big trends now, and in the era of the Internet of Everything, how to sink AI from the cloud to the edge, making the edge with AI capabilities is the issue that must be considered now. And making IoT devices and cars simultaneously equipped with the three pillars of intelligence, security and interconnection capabilities is also a problem being studied at the moment.

Ding Xiaolei said that edge AI is more effective and has better privacy protection, and can be more efficiently and autonomously calculated and executed. Information security is an indispensable capability in the era of cloud-connected intelligence, while edge AI is capable of identifying verification and data protection to achieve security. Cloud AI, on the other hand, is able to use more computing power to implement larger language models. In this way, the two sides collaborate with each other, have different positioning, and are forming a cloud-edge architecture.

Based on the understanding of the above architecture, we will increasingly see the reasoning end of AI gradually sinking from the cloud to the edge, which can bring many advantages: first, the edge has superior real-time, can provide ultra-low latency, to meet the high demand for real-time in industrial, consumer, medical and home appliances, etc.; second, with the continuous improvement of arithmetic power and the reduction of costs, the edge can effectively reduce the need for data uploading to the cloud, greatly enhancing the ability of the cloud to provide real-time data. cloud, which greatly enhances privacy and security; finally, edge-end also has the advantages of high energy efficiency, low data transfer rate, and low power consumption, which enables more new types of operations. Therefore, edge-end reasoning shows great advantages in all aspects.

When building a world with AI arithmetic, we need to think about how to achieve a reasonable distribution between the cloud and the edge end. With the development, the cloud-centric architecture is gradually shifting to edge-centric, and different hardware and software support is required to achieve this shift. This is exactly why ST launched the STM32N6.

ST has been deeply involved in the field of edge AI for more than a decade. ‘We want edge AI to be everywhere and to prove that AI will be everywhere in the future.’ According to ABI Market Research, the edge-side TinyML MCU market is expected to grow at a CAGR of 113 per cent by 2030, Ding said. Whether in application areas such as power management, arc detection, face target recognition, or anomaly detection, such a growth rate means that more and more use cases will be progressively adopted.

As more and more traditional embedded developers enter the AI field, ST is expanding its ecosystem to provide more tools and development facilities. These developers are increasingly utilising MCUs for edge-side AI development.

According to statistics, as of Q1 2024, ST has more than 50,000 active customers worldwide using ST's edge AI development tools. At the same time, ST launched the ST Edge AI Suite edge AI development kit, which covers the complete edge AI development process from idea conception, data acquisition and training to model optimisation, deployment and validation.ST's full suite of development tools provides developers with strong support to help them carry out their development smoothly.

In addition, ST has been focusing on deeper collaboration with pioneers in AI such as NVDIA and AWS to drive the industry forward, with the ultimate goal of making edge AI available everywhere.