AI wave sweeps the world, making the Internet of Things become an integrated AIoT architecture, which is regarded as the driving force of IT industry in the next era. What does the beginning of AIoT mean for the MCU, the main force of IoT, to undertake this wave with "improvement" or "revolution"? Act tough and talk soft After many years of "evolution", MCU has already advanced into full force, and has advanced rapidly in high integration, high performance and low power consumption. But the problem is how old drivers drive new cars - the AIoT era has put forward different requirements. The change of MCU in AIoT era needs both hard and soft. "One is the processor hardware level, which requires higher processing capacity, more security components, multiple connectivity capabilities and lower power consumption; the other is the software level, where the operating system evolves from task scheduling to IoT OS platform, which greatly increases the software complexity and requires platform-level software and tools; the third is that at the ecosystem level, various cloud service companies enter the embedded system ecosystem. And with algorithm companies, pure software companies increased cooperation. MCU support AI has become a major direction. In addition to the hard power of higher computing, lower power consumption, stronger expansion and more security, MCU also needs better development tools, which can convert the algorithms processed by GPU, CPU or ASIC in the cloud to identify and process MCU. In the era of AIoT, many cloud service companies have entered the embedded field, and MCU needs to open API interfaces; moreover, MCU needs to cooperate with algorithmic companies to achieve high efficiency; and the finger-dyeing of cloud service providers at the operating system level will also lead to more competition. Correspondingly, it also brings new challenges to embedded engineers. In addition to understanding the communication protocol stack, security, remote management and firmware updates, they also need to start using machine learning algorithms and be able to adjust the algorithms according to the application requirements. Go with the tide Anyway, the tide of AI has arrived. What we need to do is to follow the trend. After all, the benefits of deploying lightweight AI on MCU are real. This is to customize key applications and optimize computing platforms. At the same time, AI modules can be run independently without cloud connection, and achieve low power consumption, so that IoT devices can work for a long time. However, the weakness of MCU in the application of AI lies in its weak computing power. Even the MCU with the highest performance at present has only 1.2 GOPS. To solve this problem, we need to use more advanced model structure, simplify model, optimize the underlying code, and use heterogeneous computing units. In addition, in the absence of modeling tools, it is necessary to use PC/Server to model and train. At the level of integration tools and standard operating environment, it is necessary for vendors to develop and generate model code and execution engine. The strategy is to find suitable applications, find suitable algorithms, or develop MCU integrated with AI IP. MCU adapting to AIoT is not only as simple as + AI IP, but also a good combination of development tools in terms of code portability and software compatibility. ST has been devoted to the development of MCU hardware and software tools adapting to AIoT, and is expected to make a new appearance next year. With the development of MCU technology for many years, the technological differences among the mainstream manufacturers may not be so clear. But in the era of AIoT, who will take advantage of the momentum and who will be upset by the waves, not only depends on the strength of MCUs at the cost-performance level, but also tests the perfection of manufacturers'ecological chain, including development tools, operating systems, cloud services, application development, etc. Whether it is breaking or standing, it is a test of the overall strength of MCU. Inheriting its "progress" Reviewing the history of MCU is basically the history of continuous "weighted" integration of discrete coprocessors. "The earliest CPU can only do fixed-point addition and subtraction. Multiplication and division is realized by software, which converts multiplication and division into addition and subtraction. Fixed-point operation corresponds to floating-point operation. At that time, floating-point operation also transformed floating-point operation into fixed-point operation, and then realized by software. Because floating-point arithmetic is widely used, an integrated circuit for floating-point arithmetic emerged. At that time, it was called coprocessor. In fact, it was the application-specific integrated circuit (ASIC). Then the CPU integrated the floating-point arithmetic coprocessor. "Such a similar situation has been repeated in the history of CPU development, such as DSP, multimedia and so on. As a branch of CPU, MCU has the same development route. The same path applies to peripheral integration. When MCU needs UART, it needs UART ASIC to cooperate with it, then MCU begins to integrate UART; when CAN communication function is needed, it needs CAN IC to cooperate with it, and then MCU integrates CAN, so it has been upgraded and evolved all the way up to now. As a result, AI must follow the same path, first there are ASIC, then MCU integration. The so-called AI IP can be said to be the expression of the special integrated circuit. When technology and market mature, AI IP may be integrated into MCU, and ultimately achieve quantitative change to qualitative change. Change has come In AIoT's ecosystem, Internet giants are flocking in. Xiaomi's "AI + IoT" strategy highlights its ambition. Jingdong also announced the ecological deployment of Xiaojingyu AIoT. What changes will the influx of these giants bring to AIoT? It is reported that the ecological goal of millet is to build a complete "three-ring" layout, the innermost ring includes the core products of millet, such as mobile phones, television, boxes, routers and smart speakers; the middle ring is the IoT ecological chain system built by millet through investment incubation; the outermost ring ring will be the first to access the millet IoT system.