For autonomous driving above the L3 level, there is no central computing unit (domain controller) with excellent performance, and it is possible that all sensors and algorithms have become furnishings.
Perceiving the environment, data processing, path planning, and synchronous communication with the cloud, the ever-increasing number of sensors, in turn, requires higher performance and integrated computing units to maintain the delay-free processing of critical safety information.
As we all know, the traditional method of adding a discrete electronic control unit (ECU) for each new function will increase the complexity of the E/E architecture and slow down the communication and processing of critical safety information, thereby affecting the performance of autonomous vehicles.
In addition, in vehicles, wiring (wiring harness) is also a component that occupies a large body weight. Reducing wiring will result in weight reduction, thereby improving fuel efficiency. By reducing complexity and cost, domain control architecture and Ethernet backbone network will be important for future E/E development.
From the current architecture, it is mainly divided into three layers: the edge data processing trend of the sensor is obvious, the more advanced functions are processed by the centralized microcontroller unit (MCU), and the multi-sensor fusion data chooses the centralized type. The domain controller computing unit is responsible.
The competition around domain controllers has clearly begun to show signs of fierce competition.
Following ZF, Visteon, Aptiv, and Continental, another traditional car Tier1 Veoneer (Autoliv spin-off) today announced the development of the Zeus supercomputer-designed to meet the requirements of L4 autonomous driving and Zenuity's autonomous driving software Stack, based on NVIDIA Xavier.
Zeus is an ADAS/AD ECU that integrates data from cameras, radars and other sensors. It is built on the NVIDIA Xavier system chip. By integrating 6 different processors, it accelerates a variety of redundant algorithms, including deep learning artificial intelligence. Software to process data from up to 27 sensors.
At present, most of the L1 assisted driving in the market still use a separate ECU control. ADAS ECU is mainly oriented to L2 level automatic assisted driving and is used for the fusion processing of LDW/LKA and AEB. When it comes to L3 level automatic driving, the demand for AD ECU will begin to explode.
The market launch of domain controllers also benefited from the emergence of high-performance SoC processors, including Renesas’ R-Car H3, NXP S32V, NVIDIA DRIVE PX, etc. These high-performance processors must also meet real-time and functional safety And safety requirements.
At present, there are also high requirements for SoC chips mounted on domain controllers, such as:
The Ethernet architecture manages the high amount of transmitted data, including time-sensitive data, and reduces point-to-point wiring.
LPDDR4/4x operates at data rates as high as 3200 Mbit/s and above, which speeds up DRAM operations in automotive-grade SoCs.
MIPI standards, such as MIPI camera serial interface and display serial interface, provide high-performance connections in imaging and display applications;
PCI Express is a highly reliable processor-to-processor connectivity for 4G or future 5G and external SSDs. ï¼›
5G and IEEE standards, such as 802.11p, help provide real-time updates of maps or images to and from the cloud, as well as vehicle-to-vehicle or vehicle-to-infrastructure communication. ï¼›
The security protocol in hardware and software is used to protect data through USB, WiFi or Bluetooth;
The sensor and control subsystem unload the host processor and fuse sensor data to manage different types of sensor data provided by the sensor;
More advanced manufacturing process technology, from the traditional 90 nanometer, 65 nanometer and 40 nanometer to the more advanced 16 nanometer, 14 nanometer and even 7 nanometer FinFET process.
The processors used will be heterogeneous multi-core processors, possibly with several cores, GPUs and Gigabit Ethernet channels. For safety-critical functions, such as credibility checking, monitoring, and result verification, these additional core safety mechanisms will be integrated on the chip, or integrated into the second processor on the board.
At present, the general automatic driving research and development and testing use traditional industrial computers, but they cannot be mass-produced by car regulations. At the same time, power consumption and volume are disadvantages. (There is a subtext in the industry: to see if this autonomous driving solution can be mass-produced, you can easily judge by opening the trunk to see if it is an industrial computer.)
"We see new opportunities, from L3 level and above autonomous driving, they need multi-sensor data fusion." Visteon CEO Sachin Lawande said that the company launched its first model at the CES show earlier this year. Multi-domain controller DriveCore for autonomous driving.
DriveCore can seamlessly interface with cameras, millimeter wave radar and lidar sensors provided by any manufacturer. In addition, in order to prevent possible overheating of the controller, DriveCore is equipped with a complex liquid cooling system.
With the large-scale mass production time of L3 autonomous driving generally estimated to start in 2020, these Tier1s, which have already launched domain controller products, will compete around this market.
At present, many domestic manufacturers, including Desay SV, Jingwei Hengrun, Neusoft Ruichi, Joyson Electronics, Zhixing Technology, and Huanyu Zhixing, are developing and launching domain controller products for different autonomous driving levels.
Take the current mass-produced Tesla AutoPilot2.0 hardware platform as an example. This is a dual computing unit platform based on a custom liquid-cooled heat dissipation module (integrated with two computing units, autopilot and smart cockpit, on two different boards. , But placed in a box.)
A good domain controller architecture can bring about the reuse of hardware and software in the future, which will greatly reduce vehicle maintenance costs and user experience. This already has a very mature application on Tesla.
From a design perspective, the multi-domain controller architecture can separate sensing and processing, and the relationship between sensors and ECUs is no longer one-to-one. Especially for OEMs, the supplier of the sensor can be changed at will (in the standard protocol) On the basis of);
The second is the scalability of the platform itself. The type and number of sensors that can be docked are not fixed and can be developed according to the needs of OEMs.
Autopilot domain controllers focus on being flexible and available to meet the needs of B-end customers, so there are multiple options for specific implementation schemes. Domain controllers have reached a consensus in the industry, and both OEMs and Tier1 are working hard. In the next generation of models, more or less the concept of a part of the domain controller will be added.
Disposable Vape,600Puffs Disposable Vape,1200Puffs Disposable Vape,Airis Puff Disposable Vape Pod Flavors
Shenzhen Uscool Technology Co., Ltd , https://www.uscoolvape.com