With the development of the "new four modernizations" of the automotive industry, the demand for computing power in automobiles has doubled, and on-board AI chips have increasingly become one of the core components inside intelligent cars. Along with this, the spurt of development of the automotive chip industry has spawned a series of large computing platforms represented by Huawei MDC, Nvidia's Orin, Horizon Journey 5, Qualcomm Snapdragon Ride, etc.
All along, foreign brand chips have occupied the main market share. However, with the rapid development of China's intelligent automobile industry and the impact of the "lack of core" dilemma, self-developed on-board intelligent chips have gradually become a must for the development of the domestic intelligent automobile industry. And with the advancement of domestic chips, domestic vehicle-grade AI chips have gradually been recognized by the market, such as Horizon Journey 2, Journey 3 and the newly launched high-performance large-computing chip Journey 5.
Yu Yinan, vice president of Horizon and general manager of intelligent driving products, believes that smart car chips are like the "engine" of cars, and they are also the "Everest" of contemporary hard technology. In-vehicle chips will surpass mobile phone chips to become leaders in the semiconductor field. Automotive electronic and electrical architecture will evolve from traditional distributed architecture to domain controller architecture and central computing platform architecture, and intelligent automotive AI computing will also evolve from the logic-driven software 1.0 era to the data-driven software 2.0 era, and the demand for computing power will increase significantly.
Founded in 2015, Horizon is the world's first AI chip startup. In recent years, Horizon has largely maintained the annual launch of a generation of high-performance AI chips. Its Journey 2 chip released in 2019 has been used in many mainstream models; the new generation of Journey 3 chips in 2020 has also taken the lead in mass production in the domestic head car manufacturing new force ideal. Up to now, horizon journey chips have 20+ car companies fixed points, more than 50 models fixed points, last year to achieve 1 million chip shipments.
The first domestic large-scale COMPUTING AI chip with fixed-point mass production
With the development of automotive intelligence, the industry's requirements for the computing power and performance of automotive chips are rapidly improving.
In this context, domestic chip manufacturers represented by Horizon are accelerating the iteration, upgrading and optimization of their chip products in recent years. From the early Journey 2 and Journey 3 chips to the new generation of Journey 5 chips launched in July 2021, Horizon's automotive chip performance has made great progress.
(Horizon Founder and Chief Scientist Yu Kai)
Yu Kai, founder and chief scientist of Horizon, previously said that the Journey 5 chip has both large computing power and high performance, and the single chip AI computing power is 128 TOPS, which supports 16 camera perception calculations, which can support the needs of multi-sensor fusion, prediction and planning control required for automatic driving. This is also the third generation of Horizon's vehicle specification products after Journey 2 and Journey 3.
Not only that, based on the R&D concept of combining hardware and software on the horizon and the self-developed high-performance computing architecture BPU, Journey 5 has obvious advantages compared with NVIDIA's automotive chips of the same level, real-time latency, processing speed plus object detection. And the power consumption of the Journey 5 is as low as 30W, and the official power consumption of the NVIDIA ORIN series chips is 80W. At the same time, as a representative of domestic chips, the AI performance and energy efficiency of horizon journey 5 chips on some typical neural networks exceed those of benchmark chips such as NVIDIA Orin-X.
Horizon believes that due to the power consumption limitations of Moore's Law, the pursuit of pure computing power breakthroughs is not sustainable, and the computing power does not represent the "real performance" of automotive smart chips, and the calculation efficiency of chips also needs attention. In 2016, Horizon proposed the "new Moore's Law of AI computing", believing that the true AI performance of the chip is determined by the comprehensive physical computing power, the effective utilization of the chip architecture and the efficiency of the AI algorithm, and the comprehensive optimization based on this is the performance that the user can perceive.
This concept of Horizon, combined with the large computing power attribute of Journey 5 itself, greatly improves the comprehensive efficiency of Journey 5. At present, Zhengcheng 5 has won the mass production fixed point of many car companies, and will become the first large-scale domestic AI chip to achieve front-loading mass production.
The complexity of mass production of vehicle-grade chips on the ground is quite high
The head chip companies represented by Horizon have made good progress, but the phenomenon of "lack of cores" in the automotive industry since 2021 is still continuing. The research and development, mass production and launch of on-board AI chips are still common problems faced by domestic chip companies.
In Yu Yinan's view, from the chip to the project, including market analysis, example requirements and scene requirements, and then to chip development, this series of processes are very complex and difficult, which can be described as crossing thousands of mountains. In the development process of the past few years, Horizon has encountered the biggest problem is how to make the BPU computing architecture run more efficiently in the entire SOC chip architecture, including various computing IP, interdependencies between computing units, flexible handling of data, etc. In addition to basic BPU R&D, these aspects require a huge investment of resources and strength, and require top SOC architects to achieve.
In addition, after completing the chip front-end design, followed by back-end simulation and chip wiring, which all need to rely on many tools to complete. The next step of chip tape-out is a test of early design skills.
Not only that, there are still a lot of important tasks after the chip is successfully tape-out, including the development of functional safety, information security development, a large number of chip-level tests, and the optimization of chip quality. When a chip has just completed the tape-out, the chip's production line process and process still have a lot of room for optimization, so it is necessary to continue to improve, reduce the chip PPM, and optimize the process.
After that, software personnel began to appear, the algorithm, software, architecture and many partners, including a variety of drivers to do all, and finally came to the automakers to start closed-loop development. "So it looks like a chip, an autonomous driving, a software package, but the energy invested in the early and late stages is beyond imagination." Dr. Yu Yinan said: "We have made three generations of products, and each generation has invested a lot more than the previous generation. Although we have tried to reuse what we already have, after continuous iterative optimization, we have found that there are still many things that need to be reworked every time to achieve the entire architectural level upgrade. ”
Open source and sharing of technologies helps the industry reduce costs
Chip research and development work is complex and meticulous, and the cost is not affordable for all enterprises. Therefore, the industry has gradually had the call for technology sharing and chip IP open source, in order to cooperate with each other in the industry to promote industrial development. Horizon is actively practicing this industry trend.
At the 2022 China Electric Vehicle 100 Forum, Horizon proposed that on the basis of the open "chip + algorithm + toolchain + OS" platform, it will try to further open the BPU IP authorization to some automakers to help car companies enhance their differentiated competitiveness.
Yu Yinan believes that openness can always defeat closed systems. And the field of automatic driving is very different from the ecology of mobile phone APP. In the field of autonomous driving, the goals of industry players are the same, including from chip architecture to chips, to the underlying perception, positioning system, and regulation system. In this case, the industry spends a lot of resources, capabilities, different companies do the same thing, this kind of internal volume and internal friction is meaningless.
Yu Yinan said that open technology (including software openness, algorithm openness, etc.) and even chip IP openness will help optimize the cost of the entire industry. Horizon has always been in line with the best and optimal system and an open attitude, hoping to help the industry reduce costs and help the industrial ecology prosper together.