Horizon Yu Kai: The third generation of smart terminals is a car, and the operating system should not make money

"We believe that the operating system should not make money, it should be used as a public technical resource to promote the ecology of the operating system."

At the 8th China Electric Vehicle 100 Forum held on March 26, Yu Kai, founder and chief scientist of Horizon, put forward his own thoughts on chips and operating systems in the era of intelligent cars.

The largest vertical application of semiconductors will be smart cars

Yu Kai first reviewed the development of smart terminals in the past generations, from PCs to smart phones, which have bred many successful technology companies, such as Microsoft, Intel, Qualcomm, and Apple. The third generation of intelligent terminals is undoubtedly a smart car.

"How is this revolution different from the PC and smartphone era?" Yu Kai threw out this question. He believes that the market leadership of smart cars makes the birthplace of innovation happen in China.

At the same time, Chinese consumers' embrace of smart cars is unprecedented, which gives China an iterative technology for application scenarios. Truly competitive technology is not only developed in the laboratory, but also polished in the user's experience and feedback.

"When we talk about Windows or Android, is it a technological leader? We think it may be the leading technology at the beginning, but in the end, it is the leading application ecology. Yu Kai said that because there are so many global users and developers, they develop a variety of application software based on digital bases.

Similarly, in the semiconductor industry, the chip value of a single mobile phone is about tens of dollars, and the annual sales of more than one billion mobile phones have promoted the development of the largest industry in the entire semiconductor field - smart phone chips. By 2030, the number one vertical application area for semiconductors will be smart cars.

"In such an era of industrial change, new technology players and leaders will be born." Yu Kai believes that in the third generation of smart car terminals, Chinese technology companies will make a difference in smart chips and operating systems.

You can calculate fast without a heap of computing power

Challenges and opportunities coexist, and China has become the most active market for smart electric vehicles in the world, as well as the most competitive arena for global technology companies. For example, Yu Kai said that whether it is NVIDIA or Mobileye, Qualcomm or Horizon, the most advanced intelligent driving chips, the first mass production is not a Japanese car company, nor a German car company, but a Chinese independent brand.

From 2019 to 2020 to 2021, Horizon has maintained the rhythm of launching a generation of high-performance autonomous driving chips every year, from Journey 2 to Journey 3 to Journey 5, the computing power continues to rise, and each generation of chips is quickly and successfully mass-produced.

In this era of innovation, open collaboration is always important. For example, the cooperation with Ideal Automobile was at the end of 2020, but Ideal Automobile's products were already in mass production in May 2021, and in seven or eight months, from the start of the project to mass production, a new record was created. "The speed of innovation is extremely important, and at the same time, it is necessary to open up collaboration. You must open the middle boundary, don't always think that this is yours, this is mine, and then hide it, so that the efficiency of independent innovation becomes lower. ”

In the post-Moore's Law era, we can't go with the flow, and when others are talking about how much T computing power there is, we must think about our unique and differentiated way of playing, that is, "the new Moore's law in the AI era".

With the overall physics of Moore's Law slowing down, the density of the transistors can no longer be relied upon to improve computing performance. Therefore, the horizon proposes a new indicator that is faster than having a large number of calculations. Run the neural network algorithm, calculate more data in unit time and unit cost,

This optimization index can be decomposed into three factors: the first factor is the peak hash rate, and the peak hash rate is the TOPS of the peak that everyone constantly mentions in the old Moore's Law era. But in fact, TOPS is not the only indicator here; it is also related to the second factor, that is, the utilization rate of computing power resources, and how much of it is effectively utilized. This depends on innovative architectural design, on the operating system, on the compiler, on the dynamic runtime.

The third is the software algorithm of artificial intelligence, although the architecture is constantly evolving. Yu Kai said that in the next 20 years, we can also see that the software algorithm theory of artificial intelligence will continue to change a lot, how to make the chip architecture adapt to the changing and evolving innovative software algorithms.

Coincidentally, a 2020 science paper pointed out that in the post-Moore's Law era, there is still a huge space for the top-level design of the entire computing to improve computing performance.

So, how to improve? At the top level of computing, it includes the innovation of basic software (operating system), software algorithms, and innovative hardware architecture. So we designed a high-performance autonomous driving processor architecture, the BPU (Brain Processing Unit). This uses a large number of innovative advanced technologies, including high-performance, low-latency data concurrency processing technology, such as using pulsating arrays to improve parallel speed, using process computing so that storage and computing are no longer distinct, but closely integrated; when we realize that in fact, the IO of data, the access and transmission of data may produce more energy consumption than computing, we want to store and calculate together.

Based on the new Moore's Law design of the Journey 5 chip, compared with NVIDIA's flagship Orin chip, although it only uses half of the chip area and half of the computing resources, it can get considerable computing performance when using advanced neural networks to do calculations.

At the same time, more data calculations are done with lower power consumption. With the same power consumption, the computational efficiency of the horizon is doubled to tripled. This is very important for autonomous driving - autonomous driving is not a data center, there is unlimited energy supply, and there is a very good cooling system, on four wheels, the working conditions of autonomous driving are very harsh, and it is very important for efficient computing at low power consumption.

Does the chip have to be equipped with an operating system?

Yu Kai noted that in the past, chips and operating systems were never separated, which also had a "winter alliance", and two commercial companies formed an exclusive alliance, which was very contrary to the laws of business. Because no commercial company wants to be kidnapped by another company, it wants more options.

This is not an accidental phenomenon, in the entire history of the information industry, the most successful and extensive ecosystem of operating systems, often choose a chip architecture as a close alliance. For example, Windows and Intel, Android and ARM in the era of smartphones, the coupling of the architecture of NVIDIA's own operating system CUDA and NVIDIA's own GPU in the era of artificial intelligence.

However, developing an input method on Android is difficult to port to the Hongmeng system, mainly because only a high degree of software and hardware coordination can improve the efficiency of computing. This means that when the software is well optimized on one chip architecture, it is difficult to migrate to another chip architecture.

This is also what Yu Kai has to face: to make the Horizon Journey chip the base of China's automatic driving chip, do you want to equip it with an operating system?

Yu Kai said that the open source real-time operating system (TogetherOS) created by Horizon has been open source from the beginning, and now it has been jointly developed by many domestic OEMs.

He took ARM as an example, it is a very worthy example to learn, they have gone out of the different routes of traditional chip companies, to open their own chip architecture authorization, accelerate the speed of the entire industry iteration cycle, let professional people do professional things, and then promote the ARM+ Android in the smartphone boom ecology, but also promote the entire Internet of Things innovation ecology today.

Finally, Yu Kai summed up the prospect that the PC era has Intel, Windows, the smartphone era has ARM, Android, and the era of smart cars can not create an open and open source innovation ecosystem that can be participated in together, and will be a reflection on the digital base of the smart car era.

[Reporter] Gao Xiaoping

【Author】 Gao Xiaoping

【Source】 Southern Press Media Group South + client