Chips, AI, drones... What are the technological frontiers and highlights?

In April 2019, the annual China Electronic ICT Media Forum and 2019 Industry and Technology Outlook Seminar was grandly held at the Ritz-Carlton, Shenzhen.

ICT: Information and Communication Technology, that is, information and communication technology, is an organic combination of telecommunication services, information services, IT services and applications. In today's rapid development of the Internet and communications, especially in popular fields such as the Internet of Things, Industry 4.0, new energy electric vehicles, unmanned vehicles, drones, 5G, AI, and chip design, what are the cutting-edge technologies at this year's ICT Forum? and the highlights?

The robot network has tracked and understood the whole process, and it is summarized as follows:

1. AI chip + platform

AI has always been designed as an acquired software algorithm. In the era of rapid iterative evolution of business models, in order to enhance the rapid monetization capability of AI, Xilinx (XILINX) has solidified a considerable part of the basic AI capabilities into the chip. Gradually make it into a chip, and at the same time, use Xilinx's data center strategy to achieve platformization.

In the past few years, various AI-related mainstream application scenarios are mainly video and image-related applications, and voice-based applications. The dilemma encountered by companies doing AI is that it is not that difficult to make the chip itself. Whether it is an AI chip including other chips, including various CPUs made by China itself, it is not difficult to make the chip itself, but in order to use it, it is necessary to Software, ecological environment, tool chain, various reference applications, these require longer time and more resources, and the most critical and core factor is to use the product.

Therefore, Xilinx provides a general AI solution. They do two things. The first thing is to define their own instruction level and IP at the bottom layer. These IPs are very efficient customized IPs, which are specially designed for artificial intelligence. Intelligently doing different operators, such as special programming, providing IP for directional acceleration, and orienting to corresponding instructions, but this is still a very low-level hardware development capability. At the same time, tools are developed, and interfaces are provided to customers through these tools and SDK. Therefore, customers do not need to write any line of code, and only need to call it to support applications in different industries and different scenarios. Therefore, whether it is a face, a vehicle, etc., are different CNAs, and their core operators are the same, that is, the network architecture and parameter configuration are different, and then different instructions are generated, and finally run on different hardware platforms.

At the same time, Xilinx launched its data center strategy and released the next-generation Versal computing engine.

In the face of high-performance scenarios of communication and artificial intelligence, a completely different chip architecture is defined, using 3D technology to provide high-performance high-bandwidth storage, providing two capabilities, one is computing capability, the other is storage capability, both artificial intelligence networks. It is the fast operation and repeated reading of tens of hundreds of layers and tens of millions of parameters. If the results of the data operation of each layer need to be read and written, one is long delay and the other is high power consumption. So why is it that the mainstream AI chips will gradually provide as much storage space as possible, and 3D technology is the most effective technology for providing high-performance storage. At the same time, it makes full use of the hard-core processor function to support fast computing in AI scenarios.

This is the trend of chipization + platformization in the AI industry. Not only Xilinx, but also other chip design companies are gradually developing in this direction.

2. ToF becomes a hot word

At this ICT forum, ToF has become a hot word, and Infineon, ams, etc. have all introduced ToF technology and related sensors.

ToF: Time of flight, literally translated as flight time. The time-of-flight method of 3D imaging is to continuously send light pulses to the target, and then use the sensor to receive the light returned from the object, and obtain the distance of the target object by detecting the flight (round-trip) time of the light pulse.

Infineon delivered a speech on "Breaking the Application Limits and Challenges of TOF Technology in Various Smart Scenarios".

Infineon's TOF can provide the most accurate depth map. When you receive it from the sensor through this lens, it will generate a depth message, because there will be a certain gap between the time it is sent and received at each point, so this is the so-called depth message. Through this part, all DATA will be calculated, and two different maps will be obtained. The upper part is a grayscale amplitude map, and the bottom part is a depth map. After obtaining these data, it can be transferred to the algorithm part. to do what it needs to do. Therefore, the transmission and transmission of the signal seem to be very simple, but it allows all customers to get the most accurate and reliable depth information. At the same time, TOF is currently in a highly integrated direction, integrating peripheral performance on the chip. This part has obvious advantages for manufacturers in production.

3. The "supersonic" era of Flash is coming

Zhaoyi Innovation released the SPI interface specification of 400MB/s in the speech "How SPI NOR Flash Responds to Trends and Demands in High-Performance Applications", heralding the advent of the "supersonic" era of the Flash memory market.

4. The popularity of drones continues

As a seemingly high-level industry, the global sales of consumer drones in a year is only about 1 million units, and there are even fewer industrial drones. However, I believe that most people have a dream of flying from a young age. Therefore, drones are a very eye-catching industry. It is precisely because the current market is still small, but it has been proved to be feasible, so the space of the drone market is huge, even if it grows several times to an order of magnitude, the drone industry can become a star-like existence.

At this forum, Xilinx and other companies have mentioned the technology of drones many times, especially the use of AI and other technologies to realize gesture recognition on drones as early as three years ago. AI processing is an important opportunity for the drone industry to get out of the market trough.

For details, please refer to the previous report of the robot network "AI era, how to use Xilinx FPGA architecture for rapid application in drone vision and other aspects".

5. Major industry direction

At this ICT forum, ADI delivered a speech on "Accelerating towards Industry 4.0 - Realizing Faster, Smarter and Safer Industrial Automation Solutions". ADI has always had relatively strong strength and Huge customer base.

4. ADI - Accelerating towards Industry 4.0 for faster, smarter and safer industrial automation solutions_page_12.jpg

In the speech, machine health monitoring is a very interesting topic, especially from last year to this year, it is a very hot point. At the same time, machine health monitoring is also a good implementation of the Industrial Internet of Things, and it is also being used by more and more customers. accept. This should also be welcomed by many engineers.

Hua Hong Hongli published "Electric Vehicle "Core" Opportunity. Hua Hong Hongli has always been low-key and pragmatic, and is a very powerful chip company. As of the fourth quarter of 2018, as the world's largest pure-play foundry for power chips, Hua Hong Grace has shipped more than 7 million 8-inch MOSFET wafers.