Cloud AI big models are in the ascendant, and end-side AI is also ushering in the "first year of explosion". 

01 Development leads to AI

With the continuous advancement of industrial intelligence and IoT technology, a large number of production line equipment, electronic cameras and household appliances will be empowered by AI. Performing AI calculations on the device side is the core means to achieve high performance and low power consumption, which will bring about the demand for a large number of processors (MPU) and controllers (MCU), and promote the evolution of these chips in a more powerful direction.

It is not difficult to find that the recent new products of several leading IC manufacturers in the industry, including ST, ADI, Renesas and Microchip, are all focused on edge computing. By seizing the high ground of visual recognition, they will radiate their influence to the fields of smart industry, Internet of Things and smart cars. From these new products, we can see several major trends in technological evolution. The continuous improvement of computing power and the enhancement of the efficiency of terminal devices in completing complex tasks will have a transformative impact on the industry and people's daily lives.

02 ST: MPU+MCU combination 

To expand market application scope

In terms of edge computing, ST's existing product is the STM32MP2 series MPU, which will be released in 2023. Among them, the STM32MP25 will be mass-produced in the first half of this year, the STM32MP23 will be available at the end of 2024, and the STM32MP21 is expected to be mass-produced in the first half of 2025.

According to official introduction, the STM32MP2 is a 64-bit processor, which provides a prerequisite for its computing power. In terms of architecture, in addition to the powerful ARM core, the STM32MP25 also integrates a neuron processor (NPU) with a floating-point computing power of 1.35TOPS. This series is positioned in machine vision applications, which can be used in various applications such as industry, smart home, smart city facilities, etc., so the interface resources provided are also quite rich, including full HD video pipelines and LVDS and DSI, MIPI CSI-2 camera interfaces and Lite-ISP interfaces to meet various applications.

In terms of MCU, ST will also bring the STM32N6 series with integrated NPU, which has a computing power of up to 0.6TOPS. It also integrates peripheral resources such as MIPI CSI camera, image ISP, Gigabit Ethernet controller, etc., which is suitable for multiple applications such as edge computing and machine vision, and the lower power consumption level is also suitable for smart home appliances. In processing specific visual edge AI tasks, the STM32N6 with NPU is even faster than the high-performance STM32H7 product in the STM series, which shows the huge potential of AI MCU, which will also promote the replacement of equipment and chips.

Through the combination of MPU and MCU, ST can meet the edge computing needs of a series of devices from industry to the Internet of Things to smart home appliances, enabling the intelligent transformation of all walks of life. For the rapid introduction of various applications, ST also provides powerful software tools. In 2021, ST acquired Cartesiam, a professional edge AI software development company. The latter's NanoEdge AI Studio tool has been organically combined with ST's products, allowing developers to solve problems in all aspects of edge AI development faster, and even without relevant background knowledge, they can quickly deploy products.

03 ADI: Updated MAX7800X series

Bringing stronger CNN performance

In recent years, ADI has also achieved edge AI innovation based on traditional MCUs. Its MAX7800X series MPUs are specially designed for AI edge applications. Based on the ARM M4F and RISC-V cores, they integrate convolutional neural network accelerators (CNN) for AI computing tasks. Compared with the MAX78000, the CNN part of the MAX78002 in the series has been strengthened, and the accelerator frequency has been increased from 50MHz to 200MHz, so that it can handle video processing tasks.

Thanks to the characteristics of the professional AI processing core, the MPU can achieve superior AI computing power under ultra-low power consumption conditions, and is particularly suitable for battery-powered devices, including various IoT cameras, medical equipment, factory robots and drones. Compared with the MCU+DSP combination solution, the single-chip MAX7800X has lower power consumption and more convenient solution design; and compared with GPU or FPGA solutions, the cost of ADI solutions is greatly reduced. Combining computing power, power consumption and cost, MAX7800X is a strong competitor for edge AI applications.

04 Renesas: RZ/V2H greatly enhanced

DRP core performance outstanding

Renesas released the RZ/V2H product line earlier this year, further expanding its RZ product family. RZ/V2H is equipped with a quad-core ARM A55 and a dual-core R8 at the general processor level, and an M33 sub-core is added, which is much stronger than the dual-core architecture of the previous generation. In terms of AI, the DRP-AI3 core (Dynamically Configurable Processor-AI3) brings 10 TOPS/W energy efficiency to RZ/V2H, which is 10 times stronger than the previous generation. In terms of peripherals, MIPI, graphics processors and other parts have been strengthened to adapt to more complex visual recognition tasks in fields such as factory automation.

The unique feature of Renesas AI processor is the self-developed DRP-AI core. Dynamic configurability (DRP) is its basic feature. Its principle is quite similar to FPGA and can be flexibly configured for various tasks. By working in conjunction with the processor core of the MPU, high computing power and low power consumption can be taken into account. The RZ/V2H equipped with the third-generation DRP-AI core has a conventional INT8 computing power of 8TOPS and a sparse computing power of 80TOPS. In addition, the excellent energy efficiency ratio of 10TOPS/W fully gives smart devices greater design freedom, while ensuring performance, saving heat dissipation design and space.

05 Microchip: Launches 64-bit MPU

Embraces RISC-V architecture

Not long ago, Microchip released the PIC64GX series of MPUs, and like several other major manufacturers, it has a 64-bit product line. In terms of architecture, PIC64GX uses a 64-bit RISC-V quad-core processor with asymmetric multi-processing (AMP) and deterministic latency, which can be used in industries such as industry, automobiles, communications, and the Internet of Things.

PIC64GX processor block diagram Source: Microchip

In addition to being a 64-bit processor,

also has a key feature, which is the use of the RISC-V core. Although RISC-V is not as good as instruction sets such as ARM and x86 in creating high-performance processors, its open source, simplicity, and high execution rate have been valued by major IC manufacturers for creating AI acceleration-related processing. Microchip's chips have made breakthroughs in both 64-bit and RISC-V, which also marks that after the first year of the explosion of edge AI, more intense market competition is imminent.

06 Multiple technology evolution

Trends can be found from new products

The above-mentioned original new products have their own unique characteristics in terms of computing power, architecture, energy efficiency, etc., and will become a "weapon" for the faster deployment of the Internet of Things and industrial intelligence. MPU is moving towards 64 bits, the performance limit can be improved, and the overall performance of future products will be improved to a higher level. Acceleration computing units such as CNN and DRP-AI can enable MPU to have professional AI capabilities, taking into account high computing power and low power consumption. RISC-V has found its place in the AI market, which also means that the original factory will be able to promote products at a faster speed in the future.

The chip original factory continues to innovate, which will accelerate the end-side AI empowerment industry chain. It is foreseeable that factories, the Internet of Things, home appliances and other fields will usher in equipment updates, and the demand for various components will also increase. Opportunities are waiting not far away. The kinetic energy of the new cycle is not only reflected in the large demand for GPU and HBM in cloud AI, but also the driving force of end-side AI is equally important.

View more at EASELINK