In the field of embedded system, there are two different chipsets: MCU and AI chip. The former is an intelligent control chip to meet the requirements of intelligent tools (good for control); the latter is a computing chip for deep learning of intelligent machines. In the future, intelligent tools and intelligent machines will be gradually formed in the field of artificial intelligence.
Professor He Limin, an expert in the industry, thinks that “Ai chip” is worthy of attention! In view of this problem, Professor He has his own view: “Ai chip” does exist. Its concept is not only widely used, but also can not be replaced by other concepts, such as MCU, MPU and so on.
At present, the field of intelligent tools has become mature, and the field of intelligent machines is gradually developing into the field of strong artificial intelligence relying on AI chips, neural networks, deep learning and cloud interaction. The AI chip has various forms and belongs to the primary stage of development. Although some concepts are still open to discussion, the original concepts of various embedded processors can no longer be followed, and “Ai chip” may be conventional.
Taking face recognition as an example, MCU + graphics accelerator may be used for real-time face recognition of entrance guard. However, in order to recognize a specific face from a large number of people in real time, it is necessary to introduce deep learning to continuously improve its recognition ability; in order to compare with many faces, it is also necessary to interact with cloud big data, no matter how skillful MCU can not bear such a heavy task. Perhaps “deep learning” and “cloud interaction” are two important features of AI chips.
At the IEEE Association’s first symposium in Beijing, Professor Wang Feiyue said frankly that I don’t think there are so-called artificial intelligence chips now. Andreas Nurnberger, a professor at Magdeburg University in Germany, added: “I agree that there is no AI chip. Now the development of chips has accelerated the process of deep learning. There was this kind of deep learning in image before. Now you call them as artificial intelligence chips, but they are products produced for a specific purpose. I think the Internet of things can make the process more reliable and ensure that these networks and hardware respond faster, more reliably and more dynamically. I think these are the real world of smart hardware, but they are expensive because you need more complex infrastructure, more technology, which is different than before. “
At present, the competition in the field of AI chip is fierce, and it may be difficult to form a unified structure system for a moment. However, as a new recruit in the field of embedded system, AI chip and MCU complement each other in the field of artificial intelligence. It can not be replaced by each other, but also has different technology development direction. MCU and AI processor are used in different fields, both have great development potential.
Editor in charge: CT