Author of Tencent "Deep Net" Ma Guanxia
“Car AI chip is the Everest in the artificial intelligence industry, and it is also the premise for autonomous driving to achieve large-scale landing. The Horizon first introduced the first car-level AI chip, which not only achieved a breakthrough in the mass production of China's car-level AI chip, but also filled the key link of the domestic auto-driving industry ecological construction. On August 30, at a press conference during the 2019 World Artificial Intelligence Conference, Horizon founder Yu Kai said this when introducing the company's new AI chip.
As the most highly valued AI chip technology company in China, Horizon released a car-level AI chip for automatic driving scenarios —— the second generation of the journey, according to the horizon, this is also the first car-level AI in China. chip.
The car-level AI chip is a necessary condition for autonomous driving to achieve large-scale landing, but the car-level chip needs to meet the technical requirements of “high safety, high reliability, high stability” and requires rigorous research and development. , manufacturing, packaging, testing and certification processes, so car-level chips are more difficult to develop and have longer development cycles than other types of chips.
What is rare is that while the horizon has made breakthroughs in technology, it has also gained some commercialization. According to the introduction of the Horizon, the second-generation AI chip has been in the direction of high-level automatic driving, assisted driving (ADAS), multi-mode interaction, etc., and has obtained the front-end fixed-point of customers from 5 countries, and the amount of the second-generation AI chip and solution. Production models will be available as early as early next year.
In the exclusive interview with "Shenzhen", Horizon founder Yu Kai revealed more information about the second-generation AI chip and its commercialization. At the same time, he also elaborated on his consideration of the company's positioning and industry competition. .
The forerunner of domestic AI chips
In 2015, Yu Kai stepped down as the head of the Baidu Deep Learning Research Institute (IDL) to identify the AI chip track when he founded the horizon. He was the first in the industry to propose the direction of the edge artificial intelligence chip, and said that the horizon would become "the robot era." Intel”.
In the design concept of AI chip, Yu Kai advocated a high degree of synergy between software and hardware from the beginning of the business, “this is related to my own background. I have been engaged in algorithm research for machine learning for more than 20 years before starting a business, so this is the background of software algorithms. When we start to design chips, our design ideas will fully reflect these thinking of software. & rdquo; Yu Kai said.
“Deep learning processor. Its essence is to serve deep learning software. Therefore, the two must be well adapted. Otherwise, the processor can not serve these software algorithms very efficiently. The efficiency of this processor is not going to come out. & rdquo; Yu Kai further explained the importance of software and hardware synergy.
In addition to software, for the hardware production of the chip industry, TSMC, which is the leader in manufacturing technology, is undoubtedly the best choice, and the horizon is the world's first AI chip company to TSMC.
However, looking back, the cooperation between the horizon and TSMC seems to be surprisingly successful. After all, TSMC is the leader in the chip foundry industry, and the horizon that just started its business and its choice of AI chip track are still in very early stages. At the stage, the amount of streamlining orders with fewer horizons may not satisfy TSMC's appetite.
“Suddenly, a startup company is going to break into such an area and say that it is to become Intel in the future era of artificial intelligence. I think TSMC is thinking about who is the next Intel. In Yu Kai's view, the common vision of artificial intelligence is the reason why the two companies have come together. “Although two different companies, one as an industry leader, the other is a pioneer in emerging industries. But because we all have the same outlook on the future, we can come together at this point. ”
Four years after the establishment of the Horizon, Yu Kai believes that the company has reached a second major milestone: the first is to become the first customer of TSMC's global AI chip, launching the first domestic edge AI chip at the end of 2017; the second is to launch the domestic The first car-level AI chip.
In the process, the core tasks considered by the horizon have also changed. “The horizon has been our core mission for the past three years. The first is to build an industry-leading software and hardware R&D team. The second is to prove to the industry that the horizon is a complete chip solution that can be from hard to soft, from soft to hard and delivered to customers. We can streamline, develop solutions and deliver mass production. ”
“When we arrived at the beginning of this year, in fact, we naturally think more about the issue of commercialization, and more consider the role of what role we play in the entire industry chain. & rdquo; Yu Kai said.
Company positioning and industry competition
"Don't break the self, self-evolution" This is the two words that Yu Kai used to mention in the company. Yu Kai revealed that the company spent more than half a year discussing the overall strategy of the horizon.
In April of this year, the Horizon proposed the AI on Horizon strategy, focusing more on the underlying core technology empowerment. In the automotive industry, Horizon is positioning itself as Tier2 (secondary supplier), Tier1 (first-tier supplier) and OEM AI enabler.
“Why should we locate Tier2? It means that we don’t compete with everyone at the bottom. We don’t do system integration. We don’t go to the application, we don’t touch customer data. & quoquo; Yu Kai explained & ldquo; we do the core of the perceptual calculation and positioning and its prediction, but do not make decision control. ”
Yu Kai believes that the autopilot AI chip is the Everest in the entire artificial intelligence industry, because its reliability, stability, and accuracy of recognition are very high. The autopilot AI chip is also the core strategy of the horizon. But considering the horizon's positioning for itself, Yu Kai said that the control decision horizon of autopilot calculations will not be done because it will compete with customers.
In 2015, the horizon was still a lone player in the domestic AI chip field. Now, with the rise of the artificial intelligence wave, the track of the AI chip is becoming more and more congested. In particular, the Horizon focused autopilot AI chip field, the competition seen by the head players is also More and more fierce, a number of domestic start-up AI chip companies have entered this market, Huawei, Baidu, Intel, NVIDIA and other technology giants have long been laid out.
In Yu Kai's view, compared with the giants, the advantages of the horizon are: the technological leadership brought by the high synergy between hardware and software, the early development of the AI chip field, the positioning of the industry enablers and the focus on the core business.
For the autopilot field, which is mainly used in the Horizon AI chip, Yu Kai believes that autopilot and advanced autopilot will not be able to break through in two or three years, but there will be a lot of imagination between 2025 and 2030. “We often overestimate the speed of technology development in the short, two or three years or three or four years, but the ten years of time often underestimate the speed of technological development. & rdquo; Yu Kai said.
“Let's have a series of road maps behind, because we have to go longer and make the functions of assisted driving more and more complete, and then from auxiliary driving to high-level automatic driving, we will have in the future. In 6 to 12 months, we have more advanced process and process car-level chips, at least two chips will be streamed. & rdquo; For the next step of the horizon, Yu Kai finally revealed.
The following is an interview with the "Deep Net" collation:
1. About the second generation AI chip
Q: What is the biggest breakthrough in the second generation of car-level chips from the horizon?
Yu Kai:The biggest breakthrough this time was our launch of the first car-level AI processor in China. According to our understanding, it is the first in China. After all, the horizon has started earlier than everyone else in the artificial intelligence processor, so we made a breakthrough in the car regulation level, which is still very difficult. Because the overall requirements of the car-level chip are very strict. Since the condition of the car is completely uncontrollable, unlike an electronic device, what is controllable. The car is originally in the north and south, with various working conditions, so its requirements, the design requirements for semiconductors are very high.
So for this, the horizon was established for 4 years and it was the second milestone. Looking back at a few milestones, one is that the entire TSMC, we know that TSMC is like Qualcomm and Nvidia behind TSMC as a foundry. We are the first customer of TSMC's global AI chip. In June 2017, we arrived. At the end of 2017, we launched the first AI edge computing AI chip in China. Today, we have reached a major milestone and launched the first auto-class AI chip in China. These are actually a major milestone for the horizon.
In addition to the car rules, this chip has another great development. Its computational efficiency is very high. That is to say, the horizon advocates the deep synergy between software and hardware. Therefore, we are relatively more aware of the core basic theory and algorithms of AI than the average chip company. We understand the architecture design of AI chips more than the average software company and algorithm company. Therefore, the joint iteration and joint design of this hardware and software makes the whole calculation efficiency, which is very efficient for the artificial intelligence neural network. For example, first of all the power consumption of the unit, we can handle more complex neural network operations. Now that the chip has 2 watts of power consumption, we can do 4 TOPS, but this is not the most fundamental. The most fundamental thing is that most of the chips are listed as peak power, but its actual effective power may be a fraction of its peak power, which may be 10% to 20%. Many chips are like this. Includes GPU.
Because our horizons make software and hardware work better together, our peak computing power is very close to the computing power in the actual application scenario. Our effective utilization rate can achieve more than 80% on average for various neural networks. In this case, our actual calculation power far exceeds other levels of the industry peers. We joked that essentially the horizon is a harder software company and a softer hardware company.
Q: Why can the horizon make software and hardware synergy better?
Yu Kai:This is also the difference in design philosophy. Of course, there are different talent reserves from the horizon, which are related to my own background. I have been working on algorithm development for machine learning for more than 20 years before starting a business, so this is the background of software algorithms. In other words, when I started to design chips, our design ideas would fully reflect these thinkings of software.
However, the essence of the deep learning processor is to serve deep learning software. Therefore, the two must be well adapted, otherwise the processor can not be very efficient to serve these software algorithms, then the efficiency of this processor is not coming out. Just like the most widely used foreign deep learning processor in the world I just talked about, the computational efficiency and computing power actually marked are like this, but in fact, we can use 10% of the peak efficiency in this world, 20 %, mostly idling. Horizon Because I can adapt software and hardware better, we can actually improve the efficiency and utilization ratio several times. I have just mentioned that the average is over 80%. If there are some network structures, we can still support more than 90%.
Q: The horizon is the earliest AI chip company in China to find TSMC. The TSMC mainly produces OEMs. When the horizon was just filmed, the demand should not be particularly large. How did it work with TSMC?
Yu Kai:Correct. So we attach great importance to this. This certificate was sent to me by Wei Zhejia, the president of TSMC, because at that time the horizon was still regarded as a strategic customer and partner. Because after all, suddenly there was a startup company that broke into such an area, saying that it was to become Intel in the future robot era. This itself, I think TSMC is thinking about who is the next Intel, I guess.
Artificial intelligence calculations At that time, when the horizon began, the chip was indeed a very early niche thing. However, TSMC's status is different after all. It has been watching industry trends, so we all have this view of the future. Although two different companies, one huge, as an industry leader, the other is a pioneer in emerging industries. But because we all have the same outlook on the future, we can come together at this point.
I remember which year it was. In 2017, we were already their customers. At that time, I was invited to go to TSMC and all the TSMC executives to exchange the future of artificial intelligence computing. At that time, everyone was very excited.
So although we are such a small start-up, we still get very favorable support. In particular, we now have a lot of communication in the development of future-oriented car-level chips.
Q: So can it be understood that the manufacturing process requirements of the car-level AI chip are higher than that of the mobile phone chip?
Yu Kai:Correct. Because the mobile phone chip out of these problems actually has little effect, but if the car's active security AI processor has a problem, then the impact is life.
The manufacturing process requirements of the car-level chip are very high. Usually even the same design chip, if you are at the car level, you will see a lot bigger, because its entire design is advanced enough for you to be in various situations, various temperatures and battery interference, it They all work fine.
Therefore, the horizon has now reached the forefront of the entire technology world. Therefore, I feel that more and more such industry leaders and giants are willing to cooperate with the horizon. Including the current major shareholders, we also feel very proud of the horizon. The world's largest semiconductor companies have two of the top three, Intel and Hynix are our shareholders.
Q: The horizon of the second generation chip was announced at the beginning of the year and it has been successfully released. Why is it officially released now?
Yu Kai:First of all, there are three stages in the industry. Generally speaking, many chip companies will say that they will release, and only one of his future products will be released. The road signs are not really streamlined. Most of the conferences are in this form.
Another is to announce the success of the film, that is, after taking back some of the basic tests from TSMC, I found that the film was successful. But this thing itself is a big shackle for many chip companies, because it is usually rare for a one-time successor. A single-shot is not successful. Usually it is a catastrophic event because it is a piece of hardware, unlike software that can be changed back.
After the film is returned, we have to do more complicated and sufficient test verification, and not to say a single chip verification test, but to put it in a system to test and verify, and to develop the system reference design and its software. Then you can prove it to your customers, you can see that this chip can be delivered to you, you can go to the development and put it into your vehicle system.
So after we finished this step, we felt that it was official, not called release, and it was launched. The horizon officially launched the second-generation AI processor journey. So the horizon, our two external conferences are actually different from many other peers in the industry. We are all demonstrating the overall system effect of the chip, so it is officially launched, so answering your question is that there are three Stages. Release concept, this is very popular in China.
Publishing concepts, streaming back, and getting more and more real. The system can then be delivered, we are going to the delivery system, this time we are delivering the system. Because we have already got the front mounting points of customers in 5 countries, the entire adaptation cycle is on the order of millions.
Q: Can you talk about the commercialization of the second generation of the journey and the overall layout and thinking of the horizon in the field of automatic driving?
Yu Kai:Our thinking is that we must follow the law of industrial development and respect the objective laws of industrial development. So the horizon didn't take so much effort to get too much involved in the unmanned thing, because we think this is a long time from mass production.
What really brings real value to consumers and customers now is not to replace the driver, but to make each driver's driving process safer, more convenient and more comfortable.
Our processor is aimed at assisted driving and is aimed at the mainstream market of Mobileye's flagship products. There is no amount of automatic driving and driverless driving. Where is the automatic driving? How many years will the future increase? Therefore, I think it is necessary to be pragmatic and gradual, and to truly create value for the market. This is the idea of the horizon.
Of course, because of the horizon, we designed this chip to allow the chip to cover the driver-assisted driving, but its chip interface design can be integrated and integrated into a more computationally efficient computing platform. Our Matrix computing platform is now being delivered on a large scale to many of the world's Robotaxi vendors. In this market we are also very competitive, with low power consumption and high efficiency. It's hard to find us in the world that power consumption is so inefficient.
Our competitors give Robotaxi typically 300 watts, 500 watts of power, and we give them less than 100 watts of power. So the difference is huge.
Q: What do 300 watts and 100 watts mean for a car tier chip?
Yu Kai:300 watts and 100 watts, it is unacceptable to drive more than 100 watts in the future. It has the entire power consumption, heat dissipation, and water cooling. Then, because of the heat dissipation, the electronic components are actually reliable. Sex and life are drastically reduced, and there are problems in it, including power supply.
Q: You mentioned that the second generation of the journey is for the front loading. What is the approximate ratio of the front and the rear loading?
Yu Kai: At present, we are in the front-loading market. Last year, we mainly started the delivery of Robotaxi. In the after-loading market, it is mainly for those who operate vehicles such as DMS and ADS. After installing this market, we should have a revenue of 100 million yuan this year. There is no income in the pre-installation this year, because the chip is only available when it is fixed. The chip is almost 2 years after you get the fixed-point test at the car level.
2. About industry competition
Q: Intel is also a shareholder of the horizon. What is the relationship between the horizon and Intel in this respect?
Yu Kai:Partnerships will also have a partial co-opetition relationship.
Q: What are the main aspects of competition?
Yu Kai:Intel's mainstream business is PC and data center processors, and we are not involved in these businesses. In part, our product form, business model and Mobileye acquired by Intel overlap, which inevitably leads to certain competition. The Horizon is currently being launched, including the Horizon's launch of this processor, which will be competitive with the current Mobileye mainstream product EyeQ4. Because we are lower than it is in terms of power consumption, and the computing power is stronger than it is, we are all in the single chip to focus on the L2 and L2 Plus markets. Combine multiple chips, like the Tesla FSD, to drive the high-level autopilot market.
But to be realistic, our product is aimed at the mainstream demand of the current large-scale production vehicles, not to pursue hundreds of watts of power consumption, hundreds of T's computing power, those are essentially server-side, no heat, no power supply There is no such requirement as a car gauge.
At present, the demand for smart mass driving in the world's mainstream mass production is actually L2 and L2 Plus. Although everyone talks about autonomous driving and driverless driving, unmanned driving and autonomous driving are actually large-scale commercial mass production. For a long time, even the L3 semi-automatic driving, the recent market, the international mainstream players, the most powerful Tier1 and the car manufacturer believe that L3 and their cognitive ratios have retreated three years later.
Q: In addition to the horizon, giants such as Huawei, Baidu and Nvidia are also doing car-level AI chips. What are the characteristics and advantages of the horizon compared with them?
Yu Kai:The advantage of the horizon First of all, it is still very important to emphasize the technical advantages brought by the high synergy between software and hardware, of course, including our first-mover advantage, we are earlier than them. In this case, we usually say that we can increase the effective power of the same power consumption and cost. This is a technical advantage.
The second advantage, our industry positioning is the enabler, we are positioned in the industry is Tier2, that is to say we are not directly doing this system integration delivered to the OEM. Their positioning is Tier1, and some people don't understand it. Therefore, the positioning of our industry's underlying empowerment should be more clear. This is the second point.
The third point is that we are more focused. This is our core business. However, I think the giants are basically everything. Software, hardware, and vehicle integration include operations. I can’t wait for my son’s direct operation to give replace to replace. So our positioning is to be more focused.
Q: The horizon is more focused on the field of autopilot chips, but the industry may face some real difficulties at present. How do you see the future of autonomous driving?
Yu Kai:I think that driving it automatically must be a future. In the short-term two or three years, or in three or four years, we often overestimate the speed of technological development, but in ten years, we often underestimate the speed of technological development, because technological development is exponential, starting from the beginning. It’s very slow, and then it’s steep, and it’s generally exponential. In particular, Moore's Law, Moore's Law is basically ten times faster than every five years, if Moore's Law is growing a hundred times. For example, what is the meaning of this one hundred times? If the starting point is the computing power of the mouse brain, if it grows a hundred times, it becomes the computing power of the human brain. Because the brain's computational power and the human brain's computing power are basically one hundred times different.
So everyone will feel so slow at first, but after ten years it is far beyond your imagination. So I personally think that autonomous driving, high-level automatic driving in 2025, is now we can see that in the current 2 or 3 years, I feel that it is difficult to make breakthrough progress in actual commercial use. But there will be a lot of room for imagination between 2025 and 2030, which is to achieve such a major breakthrough.
Q: What is the main breakthrough?
Yu Kai:I personally think that core competence is a breakthrough in computing power. So why do I choose this direction? Because I think this is the core of the core of the entire artificial intelligence industry. For example, let's look at the Alpha dog. Why do you exceed the imagination of people going out today? I remember that it was also the second game of Tencent. At that time, I and Tencent video and several chess players presided over the scene. At that time, these players were shocked and felt Alpha. How is the dog so powerful. But in fact, you have seen the basic algorithm in the Alpha dog 30 years ago, for example, the neural network inside, but why is it possible to play Go under the computer now? In fact, the biggest reason is that Moore's Law has increased by a million times in the past 30 years, making the computing power that could not be afforded in the past.
So, at the core, why do I have artificial intelligence for the horizon for so long, why do I choose such a difficult road? Rather than simply doing software algorithms, the core is that it is too important, too core.
Q: The horizon is currently focusing on the field of AIoT in addition to the car chip. Are the horizons more important in these two areas?
Yu Kai:The horizon can be said that our core strategy is this, the autonomous driving processor, it must be the Everest in the entire artificial intelligence industry, because first of all its reliability and stability requirements are very high. It also requires a very high level of accuracy for identification. In other areas such as face recognition, life safety issues do not occur even if the recognition is wrong, so we think that the autopilot processor is the most difficult in the industry, it is Everest.
But we also hope that the Know-How we acquired during the climb to Everest will be able to empower the needs of many applications in the automotive active safety field. In this case, we will also turn Decorate into a non-car-level processing for each generation of car-level chips, to empower non-auto industry partners. This is not just AIoT but also many other application areas.
3. Positioning on the horizon
Q: You mentioned in the company that you want to “break yourself and self-evolve”. Horizon launched the AI on Horizon strategy in April this year to be a chip solution enabler. This is a big shift because the Horizon business may have been slightly more fragmented before. Can you talk about some of your thoughts during this transition?
Yu Kai:In this case, in China, the first industrial division of labor has always been unclear. Unlike in foreign countries, the entire development of the industrial chain is relatively healthy. However, if there is no healthy and orderly industrial chain and industrial ecology development, the development of the entire industry is not fast.
In the past three years, we have been in the fourth year of this year. In the past three years, we have honestly said that our main core task is the first one to build an industry-leading software and hardware R&D team. . The second is to prove to the industry that the horizon is a complete chip solution that can be from hard to soft, from soft to hard and delivered to customers. We can streamline, develop solutions and deliver mass production.
When we arrived at the beginning of this year, in fact, we naturally considered the issue of commercialization more and more about the role of the role of the role in the entire industry chain. I think that in the technology development phase, relatively speaking, these things are not considered. When you prove your product capabilities, you don't think about it. But once these two levels have passed, then you must go to the customer every day, tell the customer what value I create for you, what is my position.
So at this point, we will spend more energy thinking about the entire business strategy of the horizon. Then we propose the AI on Horizon strategy. AI on Horizon means that we want to build Intel at the bottom of the era of artificial intelligence edge computing. Then we empower the bottom layer, which is the business model. But behind this business model is one of our core values, which are to make customers and walk with customers. This is why we should be positioned as Tier2, which means that we don't compete with everyone at the bottom. We don't do system integration. We don't go to the application, we don't touch customer data. This positioning, I think it is positioned for the horizon, the first time in the history of the horizon to locate this thing is very clear. But it is not only to tell our customers, but also to every employee on the horizon, what our mission is more emphasis on. Then what is our code of conduct, so this is our business model of building an internal and external integration, a clear strategy and the behaviors and values we must abide by.
Q: In the process, have there been any organizational adjustments within the company?
Yu Kai:a lot of. The company's internal light discussion strategy was discussed for half a year. Because you think about it, most of these young people who come to the horizon, all of them in the past, whether they are in colleges or old clubs, are actually the best, everyone is proud and proud, who is willing to climb the ground to serve customers. There is also a higher level in the industry chain that everyone wants to stand up. Most people naturally think about doing more than doing less.
Letting everyone's different ideas reach a consensus, we want to be low rather than high, we have to do less than do more, we have to achieve customers rather than achieve ourselves. So these are a big discussion of ideas and a process of building consensus. The process of building consensus must include some ideas of abandoning and breaking up your past, and truly becoming an industrial enabler.
This process, because after all, a background like the horizon, a group of people who have such an idea, so to establish such a process, at the same time, after the unity of ideas, it is necessary to establish an organization and establish such an organizational form. This is a process. Therefore, I think that the enterprises that can survive are actually a revolutionary team. The revolutionary team is different from the bandits. It must have a program, a mission, and autonomy. And it has been continually iterated.
Q: This strategy is mainly focused. Focusing often means that we need to cut down some of the businesses we have already made. Do you think there is some so-called "difficulty" in the process of "transition" or "strategic pain" during the strategic adjustment period?
Yu Kai:I think it is definitely there. For example, in our AI calculation, the edge calculation, we have been discussing. Because the horizon is actually in the long run, we want to be Intel in the entire robot era. When the company was not established four years ago, it was said to the media during the financing phase.
Q: Why is the horizon not Intel in the AI era, but Intel in the robot era?
Yu Kai:The era of robots. For example, we don't care much about cloud computing, we are more concerned about the calculation of edge measurement and terminal. Autopilot is actually a robot.
Q: How do you define the robot era and the AI era?
Yu Kai:In the AI era, for example, many calculations analyze commercial data stocks in the cloud, and that doesn't matter to me. It is that it can act autonomously and make decisions on its own, which can help us do a lot of things. I think that autonomous driving is the strategy of the horizon. Certainly, we hope that in the first 15 years, by 2030, we will spend our energy on autonomous driving, because I think that if autonomous driving is to become an automatic driving field, if this is achieved, then the basics. It became Intel in the robot era, because there is no more challenging and more complex robotic application than this automatic driving.
Speaking of this place, you just said that the trade-offs and pains, then we will consider, here is the automatic driving calculation of its control decision, we do not do? In terms of the division of labor in industrial development, we also made a very painful decision that we do not do. Because if we do, we will naturally compete with our customers. There is a deep logical thinking. From the end of the industry, we will definitely do what we empower. What do we do? Do really help everyone, not to grab a job with everyone.
So in response to your example, if there is a trade-off, then let's say yes, we will do the core computational calculation and positioning and its prediction, but do not make decision control. It may sound difficult to understand these things, but the core logic is that we make trade-offs, do not control and make decisions, and do not grab business with our customers.
Q: What kind of market structure do you think the future AI chip industry will be?
Yu Kai:good question. I don't think there will be many players in this game. As a chip, everyone actually sees the pattern of industrial development in the US market. In the end, it is actually a giant of several chip companies. Because its threshold is relatively high, basically if you do not have any competitive advantage, it will soon be eliminated. The opportunity for a generation of products is not given to you, and your generation of products is a huge investment.
So, in the end, it is actually a competitive game. But last time there was a statistic that there were two types of companies that were worth the last high, one was an Internet company and the other was a semiconductor company. In this type of enterprise, I think China should produce such an industry giant.
Q: The last question, how is the next step in the horizon planned?
Yu Kai:Next we are going to release it. Then we will have a series of road maps, because we are going longer,
We will make the function of assisted driving more and more complete, and then more and more, and then from assisted driving to high-level automatic driving, we will have more advanced process and craft cars in the next 6 to 12 months. At least two chips will be streamed.