Shulou(Shulou.com)11/24 Report--

In recent years, with the in-depth development of information and intelligence, the semiconductor industry represented by chips has attracted more and more attention. In the semiconductor industry chain, in addition to the well-known CPU, GPU and so on, there are many other types of chip supporting work, including MCU.

MCU is the "single chip microcomputer", also known as the "brain" of the electronic system. It can not only control other parts of the system according to certain procedures, but also make processing, calculation and decision-making by collecting external or internal data. It can be widely used in consumer, industrial, medical, automotive and other fields. For example, with the explosive growth of the Internet of things, intelligent cars and self-driving in recent years, MCU plays an irreplaceable role.

MCU plays a vital role in the electronic system, it can control and manage other parts of the system according to certain procedures, and it can also collect external or internal data, process and calculate, and make decisions, which can be said to be the "brain" of the electronic system.

Recently, ADI product experts accepted an exclusive interview, including CTOnews.com. It is understood that since 1995, ADI's MCU products have shipped more than 1 billion pieces. At the same time, from 2020, AID began to expand the edge MCU product line on the basis of traditional AI MCU, which can help battery-powered devices to realize artificial intelligence and Internet of things applications more easily.

ADI's edge AI MCU solution is the MAX7800X series, with two microcontroller cores integrated with Arm Cortex-M4F and RISC-V architecture responsible for data loading and startup, and the CNN convolution neural network accelerator for AI reasoning.

According to the functional application, ADI's MCU products can be divided into three categories:

Low power MCU: small size, low power consumption, large storage characteristics, can be used in industry, Internet of things, medical, wearable and other fields

Secure MCU: with secure system architecture and strong anti-attack encryption ability, it can be used on smart phones or terminals that require high security performance, such as POS machines, card readers, etc.

Artificial intelligence MCU: born from the first type of low-power MCU, it is characterized by the ability to push AI reasoning from the cloud to the edge, and can be used in smart home, face punching, voice control and other applications.

ADI (Yadno Semiconductor) was founded in 1965, headquartered in Norwood, Massachusetts, USA, with design and manufacturing bases around the world, is recognized by the industry as a leading supplier of data conversion and signal processing technology, with 60,000 customers around the world, covering all types of electronic equipment manufacturers. In the field of analog IC, ADI has always been the second highest, with sales of US $9.4 billion in 2021, an increase of 21% year-on-year and a 12.7% market share.

The following is the conversation between CTOnews.com and other media and Li Yong, senior business manager of ADI product line (with editors):

Media: the whole trend of the MCU market in 2022 is not very optimistic. How does ADI view the MCU market in 2023?

Li Yong: there may be a downturn in some consumer electronics sectors in 2022, but according to the latest forecast report, the demand for MCU will grow by 10% from 2020 to 2035, so ADI is very confident about the development prospects of the market in the future.

Media: AI MCU's AI kernel IP introduced by ADI today is self-developed by ADI or related IP based on Arm?

Li Yong: for example, there are three cores in MAX78000, one is Cortex-M4F of Arm (this is IP of Arm), RISC-V is developed by ADI, and CNN is IP of ADI.

Media: we use the kernel of Arm and the kernel of RISC-V in our design. Why should we use this design? What are their respective functions? In addition, is our RISC-V kernel self-developed or licensed?

Li Yong: the RISC-V kernel is also 32-bit. It only appeared in recent years and has been welcomed by many users. It is characterized by very low power consumption, relatively small size, relatively simple, and open, so many users like to use it. As ADI has always been user-centered, try its best to provide customers with the products they need, so it is the reason for further optimizing the power consumption of RISC-V in ADI products. For example, RISC-V in MAX78000 is an intellectual property developed by ADI itself, and more RISC-V-based products may be launched in the future.

Media: is the accelerator in our AI MCU developed by ourselves? In addition, what do you think of some of the competition brought to ADI MCU by emerging edge voice and image AI chips? why don't we directly use edge AI chips to dig into these markets, but adopt the concept of AI MCU instead?

Li Yong: AI MCU is the IP,ADI of ADI, which has been developing this kind of product since 2013, which is AI's product. CNN is ADI's own IP. The edge AI just mentioned puts more emphasis on low power consumption, size, price, and security. Math is only one aspect. But for example, to make a camera often want a CPU is enough, this CPU must have both control function, but also have CNN function, this is the current market demand, and ADI can meet it. If you use some traditional AI chips to do it, you may have to add a lot of things outside, including PMIC, that is, power management IC, as well as some memory, storage, MCU, all of which will result in high cost and high power consumption. Therefore, for the marginal intelligent market, ADI MCU will be a very suitable choice.

Media: MCU intelligence is a major trend. What new demands will the market put forward for marginal computing power in 2022? How much computing power can meet the existing demand? in this case, how to strike a balance between computing power and power consumption?

Li Yong: first, the computing power of AI MCU is mainly aimed at what kind of applications. ADI's AI MCU is for edge applications, and AI is mainly for image analysis, sound analysis, or waveform recognition. For a single application, because it is edge-oriented rather than for many sensors, for a certain sensor, the requirement for computing power will not be very high. For image analysis, sound analysis or waveform analysis, MAX78000 is fine for most applications. In addition, MAX78002 can also meet the needs of some video and streaming media. ADI's MCU has been widely used in the market, covering industrial, medical and other fields.

Media: AI training requires a large number of samples, our edge intelligence needs roughly how many samples can be achieved? Is there any speech recognition that can support more languages, such as Chinese? What is the power consumption?

Li Yong: the training end is actually done on the PC platform, because it involves some mathematical algorithms, and there are different algorithms for different applications. Generally speaking, about 500 samples I know will basically form some eigenvalues, which can achieve a resolution of more than 90%. This data is not very accurate and is for reference only. In addition, Chinese certainly supports, in fact, no matter what language, for example, to search 20 keywords, 50 keywords, or identify to open the curtains, open the door, turn on the lights, these are no problems. There is no relationship between power consumption and language, English, Russian, Chinese are not related, power consumption actually depends on the chip design, chip design, power consumption has been determined. The ADI is designed with very low power consumption, so it is suitable for battery power.

Media: in order to improve the ability of AI, RISC-V has launched the latest Vector1.0 standard, which has greatly improved the energy efficiency of AI. For ADI, will the RISC-V kernel be used to meet the needs of AI in some areas of AIoT in the future?

Li Yong: AI actually involves a lot of matrix multiplication and addition, whether RISC-V, Arm Cortex-M4, Cortex-M7 can be used. But we all know that around the microcontroller, in addition to the central accelerator, such as multipliers, adders, there are still many registers to instruct, and need to be put out after calculation, this is a relatively large system. Because CNN is purely a mathematical thing, involving a lot of multiplication and addition, it is quite complex to use a traditional processor core, no matter what standard it takes a long time. RISC-V may have lower power consumption and may be a little better than Arm Cortex-M series, but there is still a certain gap compared with hardware CNN, because ADI's hardware CNN has 64 8-bit processors, in which there are a lot of memory storing weight data. These memory are distributed around 64 processors, and there are not so many registers to be controlled. Just take values directly, calculate and then release them, such a process will save a lot of time. The operation is also very fast, it is completed by hardware, and the processing speed is much faster than that of the traditional microcontroller. I think the hardware CNN will save power than other traditional MCU controllers.

Media: please analyze the different development prospects and competitive advantages of pure MCU, MCU+DSP and MCU+Arm+RISC-V+FPGA from the perspective of application scenarios?

Li Yong: these schemes are actually suitable for different specific applications. For example, with Arm Cortex-M4, there is a DSP coprocessor, which is very general-purpose, and there are already related products. The pure MCU solution may be in some 8-bit, 16-bit, 32-bit processors, which will be relatively simple, may save some wafer space, and is suitable for some home appliances, toys, consumer goods and other applications. MCU+DSP 's scheme requires some mathematical operations, such as screen analysis, which is suitable for industrial applications. MCU+RISC-V+FPGA is suitable for some more professional and complex applications. Because this kind of solution is often less flexible, and due to the addition of too many things, the cost for some common applications can not be reduced. This complex MCU may be suitable for some professional applications. RISC-V can mainly be used as a Sensor hub. For example, in wearable devices, RISC-V can take some data, such as heart rate data, blood oxygen data or heartbeat data, when the Arm Cortex-M kernel is in sleep state. You can put these data there and process them when the Arm kernel wakes up. I think these schemes have their own characteristics and are suitable for different application scenarios. This is why many different CPU are adopted in the market, because these products are suitable for different scenarios and need to be configured on demand.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.