Jiwei Network reported that on March 25-27, 2022, the China Electric Vehicle 100 Forum (2022) was held at the Diaoyutai State Guesthouse in Beijing. On March 27th, the Global Intelligent Vehicle Frontier Summit with the theme of "New Strength and New Strength of Intelligent Vehicle Development" focused on the development of the domestic intelligent vehicle industry chain and carried out in-depth exchanges and discussions. Wang Ping, CEO of Cambrian Xingge (Nanjing) Technology Co., Ltd. (hereinafter referred to as Cambrian Xingge), was invited to "Cloud Attendance" and delivered a keynote speech.
Cambrian Xingge is an in-vehicle chip subsidiary of Cambrian Holdings, which is committed to building the world's leading autonomous driving chip company. Cambrian Xingge focuses on the automotive track, and since its inception, it has won the favor of auto industry capital, and has received strategic investment from leading auto companies including SAIC, NIO, AND NINGDE times in the seed round.
Autonomous driving is "a very rough but full of opportunities mountain road"
Talking about the development direction and trend of automatic driving, Wang Ping said, "We believe that the development of automatic driving is a very rugged but full of opportunities mountain road, the next five years, we see three major trends: first, the equipment rate of L2+ automatic driving system will be rapidly popularized and will exist for a long time, the overall penetration rate of L2+ and above in the next five years may exceed 50%; second, the L4 level of automatic driving solutions in restricted scenarios will gradually achieve landing, but there is still a long way to go from large-scale mass production." Third, the collaborative closed loop of the vehicle road cloud will further promote the continuous upgrading of the driving experience. ”
In recent years, the industrialization process of automatic driving technology is indeed so, the entire market has experienced from ideal to reality, from fanaticism to calm development process, the industry is now generally believed that the real automatic driving is still far away, but by the policy side, the vehicle side and the consumer side of the joint promotion, L2 level intelligent driving has achieved large-scale commercial applications.
At the high-level forum of the "2022 China Electric Vehicle 100 People's Association", the leaders of relevant ministries and commissions of the state clearly pointed out that in 2021, the proportion of domestic passenger car L2 level intelligent driving will reach 22.2%, of which the penetration rate of new energy passenger cars is far more than that of fuel vehicles, which is close to 38%. In addition, high-level autonomous driving such as L4 has also entered demonstration applications.
Source: 2022 China Electric Vehicle 100 People Will Give a Speech
At the same time, towards the era of automatic driving, the industry has gradually reached a consensus, in order to truly achieve efficient and safe transportation and travel, bicycle intelligence is far from enough, and faces many bottlenecks, bicycle intelligence + road cloud coordinated development is the future development direction, which is also the key path of China's intelligent automobile industry "lane change overtaking".
To meet the challenge, the two major trends of intelligent driving chips are universal openness and large computing power
Empowering the landing of autonomous driving, automotive chips still face many challenges. Wang Ping pointed out, "At this stage, we should see that whether it is bicycle intelligence or vehicle-road cloud collaboration, in terms of chips, the large-scale landing of intelligent driving systems will still face multiple challenges." ”
In terms of bicycle intelligence, the first problem is that the current monolithic SoC processing capacity is generally insufficient, generally need 2 or more pieces to achieve, which makes the complexity of the system exponentially increase, mass production difficulties; and multi-piece SoC will cause the domain controller power consumption is very high, must be used air cooling or even liquid cooling, undoubtedly increase the cost of the system, so that the intelligent driving system in fuel vehicles and models below 100,000 yuan is difficult to popularize; in addition, the proportion of domestic chips is still low, The overall chip supply is greatly affected by the global supply chain.
Vehicle-road cloud collaboration also mainly faces three major challenges: first of all, the closed loop of massive data needs large-scale AI cluster support, according to Tesla's data, each smart car on the road, need to increase the value of 500 US dollars of cloud AI computing resources to support, the cost pressure is huge; secondly, car companies to achieve data security and privacy protection, also need to invest a lot of resources; in addition, the cloud unified operation data model can not effectively meet the personalized needs of car owners.
"To meet the challenge, we believe that universal openness and large computing power are the two major trends in intelligent driving chips." Wang Ping said, "In the L1/L2 stage, the amount of data is very small, and many car manufacturers have accepted a closed integration scheme or a black-box scheme with strong coupling of chips and algorithms. In the L3/L4 era, the amount of data has increased significantly, and the algorithm has become more complex, so it needs a chip with large computing power to meet the demand, and in addition, the development of OTA also needs to be supported by a common and open software platform. ”
Cambrian Will launch chip products for the L2+ and L4 markets this year and next year
In order to meet the needs of the smart car market, Cambrian Xingge will launch products that fully cover different levels of intelligent driving chips. Wang Ping said, "This year and next year, we will officially release two blockbuster chips, one is the SD5223 chip for the L2+ market, and the other is the SD5226 series for the L4 market that can support end-of-car training." Of course, the follow-up will also choose the opportunity to launch chip products for other market segments. ”
Cambrian's SD5223 chip for the L2+ level market will redefine the entry-level chip for autonomous driving. Wang Ping pointed out that the current domain controller scheme generally uses two or even three SoCs to achieve L2+ level driving + parking functions, the system is complex, the power consumption is relatively high, natural heat dissipation cannot be used, and the cost is also correspondingly high, and it is difficult to promote models below 100,000 yuan.
The Cambrian Xingge L2 + xingbo integrated chip solution, using advanced technology, the maximum computing power of 16 TOPS, a single SoC can achieve the function of berthing integration, and can use natural heat dissipation, to promote the automatic driving system to 8-10 million yuan of entry-level models coverage, and this chip will be released in 2022.
Cambrian's autonomous driving solution for the L4 level will also help the industry enter the era of high-intelligence car 2.0. At present, the solution of 2 or even 4 SoCs is adopted for L4 level autonomous driving domain controllers, which brings risks and challenges such as complex systems, limited board-level bandwidth, excessive power consumption, and long mass production cycles.
The SD5226 series chip solution will further increase to more than 400 TOPS in terms of artificial intelligence computing power, the maximum CPU computing power exceeds 300K DMIPS, adopts 7nm process, independent safety island design, and takes the lead in providing L4 level autonomous driving solutions based on a single SoC. Moreover, the biggest highlight of this chip will be that it can support the training of the car side and support the self-learning architecture of the car side. This solution is also scheduled to be officially released in 2023.
Wang Ping said, "The current existing car-side chip is mainly based on the reasoning architecture, the update and training of the algorithm model needs to be completed in the cloud, Cambrian Xingge uses the end-cloud integration, training and pushing the AI processor architecture, which can support the training of the car end, so that the intelligent car truly has the ability to self-evolve and self-grow, thus entering the era of high-intelligence car 2.0, and the car becomes the real partner of mankind." ”
At the same time, for the vehicle end and the consumer end, the self-learning ability of the car end can also build three major advantages: First, the data can be distributed in a closed loop, which effectively reduces the cost of the cloud AI cluster, and the car company can carry out service operations more effectively; second, it can truly meet the personalized needs of users and achieve "thousands of cars and thousands of faces" or "thousands of faces", such as different driving habits, different car people interaction, etc.; third, it can better ensure the data security and information security of car owners. Bicycle data can be selected according to the customer's wishes whether to upload to the cloud.
In addition, it is very advantageous that Cambrian Xingge will also cooperate with Cambrian to launch a cloud-edged end car program. In the cloud, Cambrian provides high-performance training chips that can process massive data collected by the car end, train, and form an advanced automatic driving model that is pushed to the car end through OTA; at the edge end, based on the intelligent chip of cambrian edge, and partners launch a road test unit for vehicle-to-road collaboration, which can perceive farther information, push it to the car end, and form a collaborative perception; and on the car side, Cambrian's autopilot chip can support the large computing power demand of complex models of high-level automatic driving in the future. Continuous iteration of algorithmic models can also be supported.
The landing of China's autonomous driving technology is inseparable from the collaboration of the whole industry chain
Of course, the development of China's autonomous driving technology is inseparable from the collaboration of the whole industry chain. Wang Ping said, "Positioning itself as an in-vehicle chip company, Cambrian Xingge will work closely with The OEM to form a network of cooperative relations with Tier 1, sensor companies, algorithm companies, etc.; at the same time, we will also take the initiative to 'save the game' and work with partners to build mature algorithm and software solutions, provide multi-level cut shelf solution delivery, fully meet the multiple needs of car companies for quality, progress, customer experience differentiation, etc., and improve the driving experience of end users." ”
More importantly, the lack of cores in automobiles since the end of 2020 has made the industry realize that in the long run, the importance and necessity of localized chip supply will not only empower the landing of China's automatic driving technology, but also help build a stable and reliable supply chain, and domestic autonomous driving chip companies will become the "backbone". In this regard, Wang Ping appealed: "From the perspective of car companies, we hope that car companies can give domestic chip companies more opportunities, through joint development projects, traction domestic SoCs to become SoCs that are more in line with the needs of car companies, and more use of localization chips to improve supply chain security." At the same time, support and guide the construction of the ecology, encourage the strong cooperation of domestic chip companies, algorithm companies, Tier 1 and other enterprises, and finally, I hope that the brother enterprises in the semiconductor industry can realize the local landing of advanced process vehicle-level manufacturing and packaging at the manufacturing end as soon as possible, and work together to promote the healthy development of the intelligent driving chip industry. ”
(Proofreading/Holly)