Home > News content

Yu Kai: today, Horizon has achieved a breakthrough in the mass production of Chinese car gauge AI chips!

via:博客园     time:2019/8/30 12:31:52     readed:192

data-ratio=0.4444444444444444

[Introduction to Xin Zhiyuan]Just now, AI unicorn Horizon, valued at $3 billion, officially announced the mass production of China's first car gauge AI chip.

In February, Horizon announced a $600 million B round of investment valued at $3 billion, setting a record for AI chip start-ups to raise funds, including a billion-dollar strategic investment from top automotive groups, the largest in the industry. And just now,Horizon announces mass production of China's first gauge AI chip

data-ratio=0.3244729605866178

Second Generation Chip Parameters

There is a general consensus in the industry:Autopilot Processor is the Everest of AI IndustryIt is also the premise of large-scale landing of automatic driving. Being the king of autopilot processors also occupies the commanding heights of AI. Nowadays, the auto-driving track is becoming more and more crowded, and the competition among head players is becoming more and more fierce. The establishment of the four-year horizon to produce China's first auto-regulation-level AI chip is a milestone event in climbing the commanding heights of auto-driving. Just asFounder of HorizonAs stated:

By 2025, 30 million vehicles will have built-in Horizon Autopilot BPU

data-ratio=0.3846938775510204

Four years of the horizon

1. Yu Kai: Chips only make hardware, and customers get stones.

Compared with companies that simply do AI chip hardware and AI software algorithms and applications, the uniqueness of Horizon is that it has been going since its inception.Algorithm chipSoft-hard combinationRoad, now, this has become an important trend in the technology industry. Yu Kai once said:If we only do hardware instead of software, what we deliver to our customers is a stone.

data-type=png

Founder of Horizon

Adhering to this principle, Horizon provides high-performance, low-power, low-cost solutions to the industry through deep optimization of hardware and software and collaborative design.The only car gauge chip that has been successfully flowed and produced in China, and can provide

Established two and a half years of independent research and development of the chip officially streamed and released, set up four years of mass production of China's first car regulation AI chip, how to do? Let's start with Horizon's entry into AI chip industry.

Second, with high barriers and old brand new nobles competing for hegemony, why does the horizon lead the edge AI core?

In 2015, Horizon was the only AI chip company in China. Today, boosted by the wave of artificial intelligence, the AI chip industry is rapidly rising, which also attracts many powerful players to join.

On the foreign side, there are some similarities.Intel, Invida and SettlingSuch old chip giants stand in the way, and there areGoogle, Apple, QualcommAfter the breakdown of such a top technology company.

Domestically, there areHuawei, Ali and Baidusuch high-level players self-research, acquire, invest in AI chips, such as the AI chips recently released by HuaweiHorizon and CambrianSuch AI chip start-ups devote themselves to research and development.

Overall, the current industry structure is that Yingweida is the dominant company in the cloud AI chip market. Cloud computing companies such as Ali and Baidu are catching up with each other, while the terminal AI chip market is multi-player and there is no absolute head company. This is the opportunity of the horizon.

Horizon launched edge AI chips in 2017, and now mass-produced vehicle gauge AI chips, opening up a clearer and wider road.

The R&D cycle is at least 3 years, and the barriers to entry are high.

The breakthrough of vehicle regulation chip is a necessary condition for large-scale landing of automatic driving, but it is not easy to develop a vehicle regulation chip. The design of common chips is very difficult, and the design of AI chips at the gauge level is even more difficult, which requires rigorous R&D, manufacturing, packaging, testing and certification processes.

Challenges in the design of vehicle gauge-level chips are as follows:

  • High tolerance。 Unlike consumer electronics, cars are in a worse environment. For example, temperature difference changes require chips with higher temperature tolerance, such as road bumps, and chips with stronger anti-vibration and anti-impact capabilities.
  • high reliability。 Automobiles have higher reliability requirements than mobile phones and tablets. If they can not achieve zero defects, they may put life and property in danger. Therefore, it is necessary to meet the corresponding technical standards, such as AEC-Q100.
  • Long life。 The design life of general automobiles is about 1.2 million kilometers in 15 years, which is longer than the life requirement of consumer electronics products.
  • High barriers。 The related industry chain of vehicle gauge chip is very long, which requires cooperation in various fields. In addition to the product itself, the vehicle gauge chip has a high demand for comprehensive capabilities such as after-sales service. Moreover, the chip design has the advantages of large investment, long cycle, fast iteration and high risk.A single chip will cost at least tens of millions of dollars, with a research and development cycle of at least three years, and even seven years at the regulatory level.。 And in the short run, it's hard to get rewards. A failure may never catch up with an opponent.

data-ratio=0.3890020366598778

Quality Test Standard for Gauge Class (AEC-Q100)

What's hard to do, the horizon of technology flow is dead on AI core

Artificial Intelligence Chip for the Earliest Production of Abortion Tablets in ChinaAndLarge-scale commercial use

data-type=jpeg

In April 2018, Horizon released the second generation architecture of a new generation of autopilot processors. At the beginning of 2019, the second generation chip chip chip chip was successfully processed.

Yu Kai once said:Chip development cycle is very long, unlike game and software development, as long as overtime can be shortened, its production needs at least one and a half years of research and development time.

The starting horizon has won for itself a year and a half's starting edge, and has been running all the way. As a company with first-class technological background, the second generation chip of the journey is a master. The first AI chip in ChinaBPU2.0 (Brain Processing Unit), a high-performance computing architecture developed by Horizon Autonomous Innovation, can provide more than 4 TOPS equivalent power with a typical power consumption of only 2 watts.。 Specifically, the chip has:

  • High utilization ratio of computing powerThe computational efficiency of the typical algorithm model is not less than 90%.
  • High computing power efficiencyEach TOPS AI capability output is reachableMore than 10 times the GPU of the same calculation force.
  • Perceptual reliabilityThe recognition accuracy of typical target is more than 99%, and the delay is not more than 100 milliseconds.
  • Perceptual richnessMore than 60 categories of targets can be identified, and the number of targets per frame is more than 200.
  • Lower system costsThe horizon combines the tensor parallel computing characteristics of the chip, and proposes a new network structure, which not only keeps the computing power requirement at a low level, but also reduces the bandwidth utilization. The travel 2.0 chip only needs 32-bit DDR memory, which has a huge cost advantage compared with the 64-bit or even 128-bit DDR memory of the competing products.
  • Comprehensive opening upProvide a basic development environment from reference solutions to open perceptual results to chips and tool chains, and provide different levels of product delivery and services according to the different needs of customers.

data-ratio=0.6292585170340681

The success of streaming chips was announced in early 2019, but to ensure that they will eventually be delivered to customers at zero defect standards, Horizon officially launched the second generation chip today. During this period, Horizon is doing chip function and stable line testing, system software development and stability debugging, constantly grinding products based on the second generation of travel. At present, the second generation development kit is fully ready to support customers to carry out product design directly.

The first gauge AI chip stream chip and mass production are of great significance to the whole industry. Horizon has set an example for domestic AI chip companies, and also provides manufacturers with a new choice with stronger performance and better price.

followJust build weapons, not fight, and be empowers in the AI era.Held in April this year.AI on HorizonHe said,

  • Locate Tier2 suppliers, only build weapons, not fight, not touch data, not products;
  • The chip is open and energized, one way to achieve the customer;
  • Provide ultra-high price ratio, extreme power consumption and open services.

data-ratio=0.6666666666666666

Based on this strategy, in the automotive industry, Horizon positioning itself as Tier2 (secondary supplier), is Tier1 (primary supplier) and OEM AI enabler, by providing the basis

From the beginning of the business, I said I wanted to do it.

Pioneer of China's high-end chips going to sea, leading the way in commercial landing

From technical products to commercial landing is a deep gap, based on a combination of hard and soft product strategies, and positioning themselves in

2018 is the first year for Horizon products to land.AI chips have been shipped to hundreds of thousands of orders of magnitude with annual revenue of hundreds of millions of yuanIn 2005, its commercialization ability was preliminarily confirmed.

As the first AI chip company to gain heavy customers in the four major major global automotive markets of the United States, Germany, China and Japan, Horizon has been leading the way in commercial landing, with top Tier1 and automobile manufacturers at home and abroad, including Audi, Bosch, SAIC, Guangzhou Automobile, Chang'an, BYD, as well as Hesai Science and Technology, Gaoxing, and First Automobile. Technological companies such as car appointment, SK Telecom and travel service providers have reached strategic cooperation. At the same time,Horizon pioneered China's high-end chips to go to sea.

data-ratio=0.42392444910807975

Prefabrication breaks down, millions in two years, tens of millions in five years

The pre-installed fixed-point project is not only the touchstone of the core technical strength of autopilot companies, but also the hard standard for autopilot companies to enter large-scale commercialization.

After the successful streaming of vehicle gauge-level chips, the horizon has already cut the front-set points of customers in up to five countries in the direction of high-level automatic driving, assistant driving (ADAS), multi-mode interaction and so on, and is expected to be in the future.The first half of next year will get double-digit pre-loading model fixed point(a) Take the lead in carrying the Horizon Car Regulation - class AI chips and solutions.Mass-produced vehicles will be on the market as early as early as next year..

The front-end break means that the commercialization of Horizon journey chips will usher in explosive growth, Vice President Horizon.The journey chip will have millions of pre-loaded vehicles in two years and millions in five years.

In the rear loading market, the commercialization of Horizon is also accelerating. At present, it has reached cooperation with a number of domestic and foreign travel service providers, including Shouqi car, SK Telecom, and operators. Based on Horizon AI chip and algorithm, it provides a series of intelligent solutions, such as multi-mode interaction, high-precision map building and positioning, etc., and has achieved batch deployment.Tens of millions of cars can be deployed in the next two to three years.

In addition, Horizon's high-performance, low-power, low-cost AI chip and solution Matrix has been favored by domestic and foreign autopilot manufacturers and Robotaxi fleet. At present, it has been endowed with nearly 1,000 L4-level autopilot vehicles at home and abroad. Matrix has become the star product of the global L4 autopilot computing platform.Over the next two or three years, it is expected to reach 10,000-scale shipments..

data-ratio=0.4842219804134929

Horizon Matrix II

As high as 192TOPS, Matrix will be released next year. In addition to the second generation of journey, there are several highlights to be paid attention:

1. AI Chip Tool Chain Horizon Open Explorer

With the formal mass production of the second generation chip, Horizon AI chip tool chain Horizon OpenExplorer (Horizon

data-ratio=0.42785065590312815

AI Chip Tool Chain Horizon Open Explorer

2. Three generations of R & D road map

A third-generation chip with Horizon high-performance computing architecture BPU3.0, which meets AEC-Q100 and ISO 26262 standards, is expected to be officially launched next year.

data-ratio=0.41409691629955947

Road Map of SoC Chip for Autopilot

3, Matrix second generation development suite and computing platform.

Based on the second generation vehicle gauge chip,Will go public next yea。 Compared with the previous generation of Matrix, the Horizon released a new automatic driving computing platform in theThe calculation power is increased by up to 16 times.At the same time,The power consumption is only 2 times 3 of the original power consumption.。 At the same time, it can support up to 8 megapixels of video input, pedestrian detection distance as high as 100 meters, and meet the perceptual computing needs of multiple countries, different scenarios of automatic driving fleet and unmanned low-speed car.

data-ratio=0.4355329949238579

Horizon Matrix second Generation autopilot Computing platform

It is reported that the Matrix Autopilot Computing Platform based on three generations of journey will be released next year.The calculation power will be up to 192 TOPS.With the ability to support the application scenario of ASIL D system, it can promote the early realization of large-scale landing of automatic driving. Compared with Tesla Autopilot Platform, the third generation Horizon Autopilot Platform has higher computational power and lower power consumption.

data-ratio=0.3760775862068966

Horizon third generation autopilot platform versus Tesla autopilot platform

As the next generation of infrastructure chips, AI chips are expected to run out of the new chip giant players, grasp the technology and products, and lead the way in the commercialization of the horizon will become the number one player? We will wait and see.

China IT News APP

Download China IT News APP

Please rate this news

The average score will be displayed after you score.

Post comment

Do not see clearly? Click for a new code.

User comments