[Core Information] Device-side AI is ushering in its first year of explosive growth, and MCU/MPU giants are iterating new products to seize industry opportunities.

In addition to the rapid development of cloud large models,2024 is regarded as the first year of the explosion of device-side AI.Industrial automation, Internet of Things equipment, smart home appliances, and visual terminals are fully enabling lightweight deployment of AI. Device-side local computing takes into account high performance and low power consumption, which directly drives the demand for MCUs and MPUs. It also drives global chip giants to accelerate product architecture upgrades and evolve toward high computing power, low power consumption, and integrated NPUs.

Currently, head manufacturers such as ST, ADI, Renesas, and Microchip are focusing on the edge computing track, targeting high-value scenarios such as machine vision, industrial intelligence, Internet of Things, and smart cars, showing a clear trend of technology iteration.

1. STMicroelectronics ST: MPU+MCU dual-line layout to improve the edge AI ecosystem

ST withMPU+MCU combination boxingComprehensive coverage of the end-side edge computing market.

1. STM32MP2 series 64-bit MPU

   STM32MP25 has been put into mass production, STM32MP23 will be available at the end of the year, and STM32MP21 is expected to be put into mass production in the first half of 2025.Adopting 64-bit ARM architecture, STM32MP25 has a built-in NPU with floating point computing power of up to1.35TOPS; Equipped with rich interfaces such as full HD video channel, LVDS/DSI, MIPI CSI-2, Lite-ISP, etc., it is suitable for industrial HMI, gateways, smart homes, and smart city visual equipment.
2. 
   
3. STM32N6 series AI MCU with integrated NPU

   Integrated 0.6TOPS computing power NPU, equipped with MIPI camera interface, ISP, Gigabit Ethernet, oriented to edge computing and machine vision.It consumes less power and is suitable for smart home appliances. Its AI vision task performance far exceeds that of the traditional STM32H7 series, accelerating terminal chip replacement and upgrades.
4. 
   
5. Software ecological support

   Acquired edge AI company Cartesiam and relied on NanoEdge AI Studio tools to lower the development threshold and help companies quickly implement edge AI projects.
6. 
   

2. ADI Analog Devices: MAX7800X series upgrade, ultra-low power consumption CNN AI computing power enhancement

Product advantages are outstanding:Single chip ultra-low power consumption, suitable for battery-powered equipment, covering IoT cameras, medical equipment, factory robots, drones, etc.; compared to the MCU+DSP solution, it consumes less power and has a simpler design. It has a significant cost advantage compared to the GPU/FPGA solution, making it a cost-effective choice for lightweight end-side AI.

3. Renesas RZ/V2H: DRP-AI core performance jumps, computing power and energy efficiency double breakthrough

The DRP core has FPGA-like dynamic configurability features and can be flexibly adapted to multiple types of AI tasks; peripherals such as graphics and MIPI have been comprehensively enhanced to meet the complex visual recognition requirements of factory automation, giving end products greater freedom in terms of computing power, power consumption, and space heat dissipation design.

4. Microchip: Launches 64-bit RISC-V MPU to embrace new device-side AI architecture

With the advantages of open source, simplicity, and high efficiency, RISC-V has become an important route for major original manufacturers to deploy end-side AI acceleration.Microchip also launched a 64-bit high-performance + RISC-V architecture, indicating that competition in the end-side AI chip market will further intensify.

5. Summary of industry technology evolution trends

1. MPU fully moves towards 64-bit, the hardware performance ceiling is raised to support the operation of complex end-side AI tasks;
2. NPU/CNN/DRP-AI dedicated AI acceleration kernel comes standard, realizing localized computing with high computing power and low power consumption;
3. RISC-V accelerates penetration, becoming an important candidate for end-side AI chip architecture, with faster iteration speed and greater space for customization;

4. MCU upgrades to AI, built-in NPU, visual interface, Ethernet and other peripherals, directly adapted to machine vision and edge applications.

5. 
   

Industry Outlook