With the opening of the era of intelligent networked cars, autonomous driving technology has become the focus of attention in the industry. In the automatic driving technology, the chip has the status of "central hub", which controls the operation of the automatic driving vehicle through the automatic driving platform. Driven by the international situation, national policies and market demand, the automobile chip track is very lively. Among them, the automatic driving chip is of representative significance.
At present, NAVIDIA and MOBILEYE, two international manufacturers, occupy major seats in the autonomous driving SoC chip market. Among them, NVIDIA is the "king" of big computing power chips. Since entering the field of autonomous driving in 2015, it has continuously refreshed the "ceiling" of computing power; MOBILEYE has been deeply involved in the field of assisted driving, and has a leading share in the field of automatic driving at L2 level and below for a long time. QUALCOMM, which has achieved a leading position in the field of smart cockpit chips, has now started to cut into the automatic driving track, and quickly won the fixed points of many head customers such as BMW and VOLKSWAGEN.
Automotive grade chip
At the same time, domestic enterprises are expanding their territory in this field. In recent years, domestic chip manufacturers such as HORIZON, BLACK SESAME, SEMIDRIVE and CAMBRICON have emerged one after another, and HUAWEI, a traditional ICT enterprise, LEAPMOTOR, a new car-making force, have developed self-driving SoC chips. The ban on AI chips issued by the US government a few months ago has made the trend of chip localization and substitution spread to the automotive field, bringing more opportunities to domestic self-driving SoC chips.
So, what level has the domestic self-driving SoC chip developed? Have the ability to "replace" the current international mainstream big computing power chips? What are the competitive advantages and product positioning of each domestic manufacturer? What kind of chip will car companies really choose?
Six barriers faced by domestic chips
It is understood that among domestic chip manufacturers, HORIZON established in 2015 and BLACK SESAME established in 2016 have launched single chips that can realize L2 ~ L4 automatic driving, and SEMIDRIVE will also expand the coverage of its V9 series automatic driving chip products to L4/L5 this year. According to SEMIDRIVE, L2 intelligent driving is still the most concerned and landed function of car companies and markets.
For high-level intelligent driving systems, the increase in the number of sensors and the improvement of resolution bring about the processing requirements of massive data, and the complexity of algorithm models is also greatly improved. Under the trend of E/E architecture centralization, the computing power of smart cars will be mainly realized by a few domain controllers or central computing platforms, which also puts forward higher requirements for the computing power of a single on-board chip.
Automotive grade chip
According to HORIZON data disclosure, with each increase in the level of automatic driving, the required chip computing power will increase dozens of times. Among them, the computing power demand of L2 level automatic driving is 2-2.5 TOPS, L3 level automatic driving needs 20-30TOPS, L4 level needs more than 200TOPS, and L5 level needs more than 2000TOPS.
However, great computing power also means higher cost. Therefore, in fact, self-driving SoC chips with medium computing power will be more favored by car companies, which also provides opportunities for local manufacturers who are starting to develop.
Products of domestic automotive SoC chip
In these products closely launched by chip companies, people often only pay attention to computing power data. In fact, in the process of developing mass-produced models, in addition to computing power, the OEM will also comprehensively consider the energy efficiency ratio, algorithm efficiency, software and hardware adaptability, processor architecture, IP configuration and development difficulty of the automatic driving chip, so as to compare the price and positioning of the standard models.
Energy efficiency ratio control
At present, the industry generally evaluates the theoretical peak computing power of autonomous driving chips in terms of "TOPS". However, in actual scenarios, the theoretical peak computing power of autonomous driving chips is almost unlikely to be completely released, because the effective utilization rate of computing power is also affected by chip power consumption performance and energy efficiency ratio control.
Comparison of automotive SoC chip’s performance
Compared with the above data, Auto Byte found that energy efficiency ratio control is an advantage of many domestic autonomous driving chip manufacturers. High energy efficiency ratio can not only save a lot of electricity for automobiles, but also generate less heat energy, which is conducive to the heat dissipation and stable operation of chips with high performance.
Efficiency of algorithm
If the peak computing power of the chip is high and released effectively, will it really improve the overall processing capacity of the automatic driving system? The answer is, unknown. If the matching degree between the chip architecture and the algorithm is not enough, the efficiency of the automatic driving algorithm is extremely low, and a large number of transistors in the whole chip are actually idling. No matter how big the computing power is, the actual processing power is not the best.
"Now the software algorithm of artificial intelligence is far from reaching the level of L5. Using the computing power that can support L5 to match the algorithm of L2, the occupancy rate is definitely low." Horizon believes that the computing power and algorithm efficiency should be developed in a matching way, and the computing power just eaten by the algorithm can be provided.
Previously, Huang Chang, CTO of HORIZON, once gave such an example: the peak computing power of the chip is equivalent to the horsepower of the car, and the driver or passenger can truly realize the acceleration of 100 kilometers, while the car with high horsepower does not necessarily accelerate quickly. As far as chips are concerned, users can feel the processing power and efficiency of the system, but large computing power does not necessarily mean fast computing.
Automotive grade chip
HORIZON has always advocated FPS (Frames PerSecond) calculation method, that is, it accurately identifies the frame rate every second. According to reports, HORIZON Journey 5 chip can reach 1531FPS, HUAWEI ASCEND can reach 829FPS, and NIVIDIA Orin is only 208FPS.
"Computational power is only a theoretical upper limit, and how much efficiency the chip can exert in the end depends on many factors such as software algorithm cooperation. In the whole vehicle life cycle and product sales process, the computing power of chips often cannot be fully exerted, and even half of them may not be exerted. " SEMIDRIVE also conveyed a similar view to Auto Byte.
Based on this problem, SEMIDRIVE has done two aspects of work. The first is to provide SDNN tools, which can allocate different algorithms to the IP that is most suitable for running, such as CPU or GPU; Second, UniLink bus technology is developed by ourselves, which realizes an efficient data distribution mechanism from the hardware level and reduces the data transmission time consumption between heterogeneous cores.
In addition, BLACK SESAME also talked about a phenomenon that restricts the efficiency of chip algorithm. At present, most AI hardware acceleration units in the market are mainly single acceleration units, and models of different sizes cannot be accelerated at the same time, which easily leads to long reasoning time for large models and low utilization rate for small models. To solve this problem, BLACK SESAME has developed NPU with medium and large computing power and multi-dimensional heterogeneous architecture, which makes different models have corresponding hardware acceleration units to accelerate, and the computational efficiency is higher.
Processor architecture scheme
In terms of architecture, there are three mainstream SoC architecture solutions for autonomous driving chips in the market: (1) CPU + GPU + ASIC, such as NVIDIA and TESLA; (2) CPU + ASIC, such as MOBIEYE and HORIZON; (3) CPU + FPGA, such as WAYMO and KUNLUNXIN.
It is generally believed in the industry that from the perspective of development trend, the self-driving SoC chip will develop into a heterogeneous architecture of "CPU + XPU"; In the long run, the architecture of CPU + GPU + ASIC will still be the mainstream before the automatic driving algorithm is mature and fixed. After maturity, the customized low-power and low-cost special automatic driving AI chip (ASIC) will gradually replace the high-power GPU, and the CPU + ASIC scheme will be the mainstream architecture in the future.
Adopting the horizon of "CPU + ASIC" architecture, Brain Processing Unit (BPU), a special processor architecture for Al, is independently designed and developed. On this basis, customers can develop automatic driving software and hardware systems and complete vehicles. According to reports, BPU adopts technologies such as large-scale heterogeneous computing, highly flexible and concurrent data bridge and pulsation tensor computing core to create matrix computing that meets the needs of end-side automatic driving. At present, five three-generation AI architectures have been launched: Gaussian architecture, Bernoulli 1.0 architecture, Bernoulli 2.0 architecture and Bayesian architecture. The next generation Journey 6 chip will integrate the fourth generation BPU architecture: Nash architecture.
At present, BLACK SESAME adopts a multi-core heterogeneous computing architecture. Its comprehensive computing power, power consumption and functional integrity require the use of 8-core A55 scheme on the chip processor, and the internal integration of self-developed ISP and NPU, as well as GPU, DSP and functional security and information security MCU. While providing powerful computing power, it ensures that different types of computing have corresponding optimal computing units to accelerate, so that the chip can undertake complex functions such as image mosaic rendering, sensor fusion and functional security, and can use single-core scheme to undertake driving and parking.
SEMIDRIVE, which also adopts multi-core heterogeneous architecture, told Auto Byte that this design is more balanced, taking into account the computing power requirements and capabilities of CPU, NPU, GPU and MCU, and can cope with the complete process of automatic driving/ADAS from perception to planning control. At present, the automatic driving system mainly relies on CPU for route planning and decision-making, and the execution of these decisions requires real-time/reliable control ability, that is, MCU computing power.
Heterogeneous IP configuration
At present, autonomous driving chips mostly adopt multi-core heterogeneous architecture, which requires different types of computing IP, including GPU, NPU and CPU, etc. Heterogeneous IP configuration is particularly important in chip design.
However, configuring IP is not as good as the higher the computing power. It involves the design and verification of the whole chip, and needs to take into account bandwidth, peripherals, memory and other aspects. Moreover, in the process of designing the chip, it is necessary to help customers optimize their hardware in every operation step according to the processing data flow inside the chip.
Black Sesame network algorithm platform
Among domestic manufacturers, Black Sesame has built certain advantages by self-developing two independent controllable core IPs, including NeuralIQ ISP image signal processor and DynamAI NN engine, a high-performance deep neural network algorithm platform. Among them, NeurallQ ISP can support up to 16 channels of high-definition camera access. In the case of severe automatic driving environment (rain, snow, foggy days, etc.), it can still achieve high-quality vehicle image processing requirements such as high dynamic exposure, low light noise reduction and LED flicker suppression; DynamAI NN engine can realize synchronous optimization of software and hardware, balance computing power and power consumption, support sparse acceleration and equip with automatic development tools, and can process more image data efficiently and quickly.
Adaptability of software and hardware
With the continuous development of smart cars, the hardware devices such as sensors and software operating systems are constantly updated, and the adaptability of automatic driving SoC chips is an unavoidable topic. "Intelligent driving of automobiles is a complex industrial chain, involving chips, algorithms, software, systems, hardware, protocol stacks, cloud and other levels. Every area needs to take into account the need for rapid iteration and upgrade. " SEMIDRIVE state
According to reports, SEMIDRIVE, the automatic driving platform of SEMIDRIVE, adopts general computing hardware acceleration and can be compatible with different automatic driving algorithm routes. Moreover, from algorithm to hardware, UniDrive is very open, which does not limit the definition of automatic driving by OEM. L1 to L4 can be designed based on this framework. At the bottom, UniDrive supports mainstream car OS such as QNX and RTOS and Linux; At the module level, UniDrive can be compatible with Adaptive AutoSAR, ROS, Cyber and other frameworks.
BLACK SESAME has also established an open chip product ecology, which not only supports OS/middleware/visual perception algorithm developed by BLACK SESAME, but also supports third-party software transplantation, without binding sensors/software/MCU, etc. In addition, BLACK SESAME provides ISP (Camera) debugging and adaptation service, which can be adapted to mainstream OS/middleware/visual perception algorithm third-party providers in the early stage.
HORIZON improves the adaptability of its chip products from three levels. At the media communication meeting in December last year, Huang Chang introduced in detail. First of all, HORIZON chip itself has various rich high and low speed interfaces, which can adapt to sensors, Camera, LiDAR, Radar, PMIC, different Tier2 and MCU. And other peripheral key devices; Secondly, in the underlying software, Horizon has opened up the drivers of related peripheral devices, and customers can complete the adaptation with peripheral hardware at the software and hardware level; Finally, in the algorithm application development, Horizon improves the algorithm tool chain, so that ecological partners can jointly complete the algorithm and application development and system integration adaptation under relatively open conditions.
Different from domestic chip manufacturers, NVIDIA self-driving SoC chip is developed based on its own GPU and strongly bound with CUDA operating system. Car companies need to develop self-driving software and hardware systems on this basis.
Development convenience
As a computing power infrastructure, chips should serve customers in algorithm development, and good chips should be "easy to use" chips. At present, NIVIDIA not only sells the developed complete automatic driving system to the outside world, but also allows car companies to purchase automatic driving SoC chips separately, and provides them with a variety of mature algorithms, basic software stacks and DRIVE Hyperion Developer Kit automatic driving development kits that can be called.
Mobileye has gradually changed from a closed black box solution to an open mode. In July last year, it released the software development kit EyeQ Kit, which allows cooperative car companies to independently develop EyeQ 6H and EyeQ Ultra chips, while still providing chip solutions without self-research.
However, domestic manufacturers of self-driving SoC chips have always adhered to the open and co-creation style, and constantly released more cooperation modes and development authority to car companies. "We will provide the corresponding development kit, which is equivalent to the service system of Tier1.5, and can quickly support Tier1 to get a quasi-mass production hardware." HORIZON company state.
Open technology platform
According to reports, HORIZON has built an efficient and open technology platform with "chip + tool chain" as the core, including hardware reference design, tool chain, AIDI development platform, basic middleware and rich reference algorithms. Through open and easy-to-use AI development tools and infrastructure, ecological partners can complete the development of full-stack automatic driving function from hardware to software and from perception to regulation in a short time based on Horizon chip. For example, with the help of Horizon, it took less than two months for the intelligent robot to quickly realize the mass production requirements of various sensing indicators of the whole system.
To tie in with the Huashan series of automatic driving computing chips, BLACK SESAME has successively released Shanhai Artificial Intelligence Development Platform and Hanhai Automatic Driving Middleware Platform, which support the rapid mass production of car enterprises through mature tool chain and middleware system.
It is reported that Shanhai Platform provides automatic optimization of AI compiler adapted to chip architecture, supports TensorFlow, Pytorch, ONNX, etc. It also has more than 50 AI reference model library transformation use cases, and supports dynamic heterogeneous multi-core task allocation and customer custom operator development; Hanhai Autopilot Middleware Platform is an intelligent driving platform SDK development package, including Target (SoC) SDK, X86 Host SDK and Target (MCU) SDK, which can support the development of vehicle-side, road-side and various intelligent driving and vehicle-road collaborative scenarios.
SEMIDRIVE has made many localized designs in its chip research and development, providing an agile and open product research and development platform with good compatibility. In addition, SEMIDRIVE has built a complete chip ecosystem, covering software, tool protocols, ecological vision and overall solutions, with more than 200 ecological partners.
Four cost considerations of chip application
From the perspective of car companies, when choosing a self-driving SoC chip, we will not only compare the performance of the product itself, but also pay more attention to the overall cost. In fact, to judge this index, we need to consider four aspects: chip price, energy efficiency ratio, development cost and supply guarantee cost.
Black Sesame BST A1000
The first is the cost of the chip. What will the system price be after using this chip?
The second is energy efficiency ratio, which represents the cost performance of the chip itself. "Good chip" can bring customers more than expected results. HORIZON told Auto Byte that if the 100TOPS chip realizes the automatic driving experience that the 200TOPS chip can achieve, it will be more cost-effective for car companies and consumers. "Just like high horsepower does not mean good acceleration performance, high computing power does not mean that the final automatic driving function experience is good." In terms of energy efficiency ratio, domestic chip manufacturers such as BLACK SESAME and HORIZON have launched products no less than international mainstream chips such as NVIDIA Orin and TESLAFSD.
The third is the cost of chip development, for example, the number and time invested in chip research and development, and the difficulty of research and development. "Tier1 or car companies get the chip to complete the development at the time and labor cost of 500 people/month, or at the cost of 5,000 people/month? If the chip is easy to use and develop, and can be deployed quickly, it is also a cost-effective choice for car companies. " HORIZON indicates. BLACK SESAME also takes Huashan No.2 A1000 and A1000L as examples. These two chips are pin2pin compatible, which can save more manpower and material resources in the process of product migration in different schemes.
The fourth is the cost of supply guarantee. SEMIDRIVE pointed out to Auto Byte that chips not specially designed for automobiles have a short life cycle and may not be available in 3-5 years. However, the supply of car regulations generally needs to be guaranteed for 10 years. If the follow-up supply cannot be guaranteed, the follow-up cost of the car factory will become higher, and the chip system must be redesigned to meet the continuous sales of models.
Mass production situation
At present, HORIZON Journey 2 has been put into mass production in front of CHANGAN UNI-T, CHANGAN UNI-K, CHERY, IMMOTORS, GAC AION Y and other models, but only CHERY applies this chip to automatic driving, and the rest are used in intelligent cockpit.
Journey 3 has been boarded in 2021 ideal ONE, which is used to realize L2 automatic driving and NOA navigation assisted driving functions. Journey 5 was first mass-produced in the ideal L8 Pro. Besides the ideal, the chip also won many fixed points such as BYD, IMMOTORS and FAW HONGQI.
Horizon
Huashan No.2 A1000 of BLACK SESAME has also completed all mass production certification after two years of software and hardware polishing verification. In May 2022, BLACK SESAME and ANHUI JIANGHUAI AUTOMOBILE reached a platform-level strategic cooperation, and Huashan No.2 A1000 chip will be mounted on a variety of Sihao brand mass production models.
As early as September, 2022, it was announced that it was developing CAMBRICON songs with three self-driving chips, and news of mass production also came out. It is reported that its first self-driving chip has been released, and has reached a cooperation with FAW, and will be carried on a model of FAW's own brand.
According to SEMIDRIVE, at present, its V9 series chips have also obtained several fixed-point mass products, but it will not announce the specific car companies that cooperate with the government for the time being. In addition, SEMIDRIVE mentioned that they are cooperating with Zongmu Technology to develop a designated project of a main engine factory.
Development Strategy of Chip Manufacturers under Local Advantages
Undeniably, the development momentum of domestic localized autonomous driving chip manufacturers is getting stronger and stronger, and it has also pried open a corner of the market. However, in the face of years of technology and customer reserves of international chip giants such as NVIDIA, MOBILEYE and QUALCOMM, how can domestic manufacturers keep the current "safety in a corner" and continue to open up market space?
Common advantage
For a long time, because NVIDIA and QUALCOMM have a complete development tool chain and a good upstream and downstream cooperation relationship, their technology iteration speed is very fast. In order to maintain such a competitive advantage, their chips will be universal in design and have a wider range of applications, and are not specially developed for autonomous driving.
Therefore, domestic chip manufacturers have the opportunity to catch up with each other in overall performance by focusing more on specific algorithms in the field of autonomous driving. Moreover, domestic manufacturers also have the advantage of localized service capability, which can better communicate with local OEMs, and the products defined after understanding the real needs of the domestic market can be better mass-produced. In addition, domestic chip manufacturers such as HORIZON and BLACK SESAME generally provide a relatively open ecology to meet the customization needs of car companies and improve the self-research ability of software algorithms.
For autonomous driving technology, the closed ecology is obviously not conducive to its rapid iteration. Taking MOBILEYE as an example, in the past, it was mainly delivered in black boxes, which was too closed. Although it could achieve rapid mass production, in the long run, it was difficult to meet the customization requirements of the main engine factory, and its computing power upgrade was conservative and the iteration speed was slow, so MOBIELEY lost some customers.
Positioning of each manufacturer
In addition to the common localization advantages, domestic autonomous driving chip manufacturers are gradually establishing their own market positioning. Different manufacturers choose different product routes, and finally the target models and applications are different.
At present, Horizon has three self-driving SoC chips, Journey 2, Journey 3 and Journey 5, which have been pre-installed and mass-produced. Horizon said that Journey 5 is a high-performance, high-computing car-level chip specially built for high-level automatic driving applications. The price of the models carried by Journey 2 is about 100,000-200,000 yuan, the price of the models carried by Journey 3 is about 350,000 yuan, and the price of the models carried by Journey 5 is 350,000-400,000 yuan.
Horizon Robotics Journey 5
It is reported that BLACK SESAMI product positioning is a general-purpose computing chip for automatic driving. It has two self-developed core IPs, and is committed to balancing the functions, costs and power consumption of chip hardware design. BLACK SESAME also revealed the benchmark models of Huashan No.2 series chips to Auto Byte, saying that Huashan No.2 A1000L and A1000 can support the integrated parking scheme from low-order 5V5R to high-order 10V5R in a single chip mode, covering entry-level models and high value-added models.
According to SEMIDRIVE, its product positioning is a car-level chip suitable for mass production. Its V9 series chips comprehensively consider the combination and optimization of user experience, algorithms and chips, integrate high-performance CPU, 3D GPU and AI engine, adopt high-reliability dual-core locking mechanism, and realize APA/ADAS solution with a single chip solution, aiming at balancing performance and cost.
Although China's self-driving SOC chip manufacturers started late, from the published product information and mass production situation, domestic chips have gradually shown the strength to replace overseas chip products, and even have been installed into about 400,000 luxury SUV models like the ideal L8 Pro.
However, most domestic self-driving SOC chips are still in the stage of getting fixed points, and their mass production models have not yet been put on the market, so it is still inconclusive whether they can stand the test of the market. If you want to gain a firm foothold in China and even go out to sea gradually, it will not happen overnight.
How to Develop Domestic Autopilot Chips in the Era of Intelligent Networked Vehicles-China exportsemi.com
