Self-driving cars require massive data storage for digital panels in interiors and infotainment screens
The pandemic has created many challenges to the development of modern vehicles - people chose not to use ridesharing services and public transportation, and many did not leave their homes at all. The self-driving car industry is growing more slowly than expected. However, the movement in the digital interior systems market is opening up opportunities for memory manufacturers as cars continue to evolve into servers on wheels.
Chris Baxter, general manager of Micron Technology's embedded systems division, said that while expectations for the self-driving market have dropped, there is plenty of room for growth in the ADAS segment. He noted that Tesla has already backed away from its self-driving claims and focused on Tier 2 and 3 systems. “We will see development in this area - adaptive cruise control, lane keeping systems, automatic braking and driver monitoring systems. In 2021, the automotive technology market will have great prospects - but not in the field of self-driving, but in the field of digitalization of the passenger compartment. "
In the long term, Micron continues to expect self-driving vehicles to remain a priority for businesses in areas that can meet the associated costs (such as robo taxis and trucking). Automotive data volumes are expected to continue to grow. Baxter added that "we will also see a shift towards centralizing automotive computing."
It is worth noting that large volumes of computation will also occur outside of infotainment systems. The thing is that vehicles (unmanned vehicles and not only) are developing and interacting with various services (which, among other things, ensure the operation of unmanned driving functions). For example, Aceinna makes sensor solutions for a variety of industries (including automotive and aerospace), and although the company does not directly operate in the computer market, its technology relies on them.
“We are not in the business of computing devices or performance technology,” said CEO Yang Zhao. He also noted that his company is "focused on sensors." However, some sensors (including computer vision devices) require high processing power to process the data used for navigation and autonomous driving systems. According to Zhao, the car of the future, like any robot, must be smart enough to move independently, be equipped with many sensors, use modern algorithms and have high processing power.
Much of the computing in vehicles is focused on external systems such as GPS satellites and inertial navigation systems, Zhao said. “There is a unified algorithm that speeds up the entire system. With it, you can achieve absolute precision. " However, there is a possibility that your vehicle will disconnect from external navigation systems due to bad weather or other reasons. “Your lidar can go blind and break, but this will not happen with inertial navigation. The external environment will not affect it. "
Zhao believes that these systems will need their own computing power and memory to operate autonomously, and they should also be separated from other modules. “The vehicle control system must be completely closed and isolated. No nodes should have access to it. "
Segmentation of automotive systems means that memory requirements will differ and depend on the architectures used in other markets (for example, mobile devices or the Internet of Things). Western Digital operates in all of these markets, according to Hubert Verhoeven, senior vice president of the company's automotive, mobile and emerging industries. "There are many examples of such intersections in history." Many developers are moving towards developing more unified and isolated systems for vehicles with their own memory modules (for example, separate flash memory for infotainment systems). "Many are looking to create clusters with storage for consumer data."
Verhoeven believes that infotainment systems with navigation modules need an architecture that integrates computing resources and memory. This architecture will increase flash storage capacity and support multiple interfaces (such as eMMC and UFS). The evidence that cars are becoming servers on wheels goes beyond the use of SSDs in infotainment systems. Verhoeven noted that there is already a demand for NVMe in the transportation market, since the vehicles of such companies have huge logging requirements. "These fleets have cars that can be called data centers on wheels."
While some of the autonomous vehicle functions are related to real-time systems, Verhoeven noted that some use cases allow terabytes of data to accumulate over several days (for example, for fleet management). "Many of our customers prefer to just take out the drives at some point, plug them into their servers and do a full unload." In such scenarios, he said, the form factor is important, allowing the drives to be removed and ease of maintenance. "Customers love the ability to work with high-performance systems, and if something goes wrong with a drive, you can simply replace it."
Verhoeven also noted that the automotive industry often uses technologies that are considered basic in industrial, mobile and server scenarios. In some scenarios, NVMe drives are preferred due to the low latency and high volume data transfer requirements of the data center. While it makes sense to reduce the number of separate data stores in the car, there remains a need for segmentation - otherwise we can place the functions for ADAS on the same media on which the entertainment for children is recorded, Verhoeven noted. This architecture also balances cost, efficiency, security, and reliability by allowing certain subsystems to receive data from a common pool.
Also, the requirements for the amount of memory and storage characteristics will be dictated by the level of vehicle autonomy. Level 4 and 5 vehicles are expected to require exabytes of storage by 2027. "Infotainment and ADAS need most of that storage, though, and they are tier 2 and 3." Nonetheless, Western Digital is linking its business prospects to how OEMs will implement advanced Layer 4 and 5 features - these implementations will drive storage development.
However, even in cars of the 2nd and 3rd level there are many built-in subsystems that generate data, which are then displayed somewhere. These systems are part of larger clusters and are now found even in low-cost vehicles with ADAS functionality, Verhoeven said. Even if the vehicle does not have a ride automation system, terabytes of data can be used and processed. “We can see more and more large displays being built into vehicles by OEMs,” Verhoeven said. As advanced features begin to become standard in low-cost cars, storage requirements and speeds will increase, which means that soon we will need half a terabyte.
Micron's Baxter said the introduction of ride automation and digital systems in the cabin will lead to a shift from a single control unit architecture to an architecture that uses separate units that control their modules. Automotive engineers will increasingly evolve towards multi-chip devices combining memory and storage, with GDDR6 replacing low-power DRAM in some applications.
Baxter believes that a change is coming in the memory industry in response to the automotive market and its demands. Changes are needed in the design and structure of memory to make the industry solutions meet the market conditions and requirements of the automotive industry. At the same time, established technologies (such as NOR flash) are still essential for fast boot functions to help drivers avoid having to wait for on-board systems to start after turning the ignition key.
Power is a big issue because a data center on wheels can handle over 600 TOPS - which can take up to 1000W of power. It's not just about processors, but also about memory. Energy consumption is especially important for electric vehicles and poses challenges for engineers - they need to solve problems related to temperature conditions.
Micron's approach is to trade-offs between power and performance by using energy-efficient, low-power memory such as LPDDR4X. At the same time, more advanced data processing modes provide improved energy efficiency - a characteristic that is incredibly important for self-driving cars that use energy-intensive, high-performance AI applications.
Computational complexity is also an issue, and underestimation is causing delays in the release of Level 4 and 5 vehicles, said Robert Bilby, Micron's senior director of automotive systems architecture. "It is still not entirely clear what level of performance is required to implement self-driving systems."
According to the Yole Development Rapid Implementation Scenario, the demand for automotive flash drives is growing and the demand for automotive NAND devices for Tier 2 and 3 vehicles is expected to increase after 2022
Like Western Digital, Micron is focusing on a clustering approach that separates critical data from entertainment content and leverages shared pools of data to improve efficiency (an example of such a pool would be a partition that stores all cards used simultaneously in different applications). Bilby believes that this approach mimics a server architecture. "This is where the concept of virtualization comes in, which allows different applications to access a shared data pool."
He also noted that more and more attention is paid to security issues. “It is very important for the industry that the memory used in cars is certified for the appropriate use cases.” We are talking about a wide range of memory devices - from DRAM for ADAS to GDDR6 for multimedia screens in the cabin. "In general, a lot of people are switching to NVMe and SSD."
Many engineers are moving away from the old approach of gradually adding functionality and expanding features. Many people strive to work on architecture in general. "Replacing the use of multiple eMMC or UFS devices is moving towards a more consolidated and centralized approach to data storage." Bilby believes that all devices used in cars should be certified accordingly. "However, this will not be easy to achieve. I think that technologies need to mature and become available for these applications."
Tom Coughlin, president of Coughlin Associates, said automakers are just starting to switch to SSDs for content storage as prices continue to decline. Due to the fact that to create a server on wheels
a lot of memory is needed , new technologies will be considered, but they must cope with the extreme conditions of automotive use scenarios. Coughlin noted that MRAM
can withstand high temperatures and can potentially replace NOR flash in some applications.
There are many trade-offs to consider when choosing memory devices for cars, Coughlin said, and balancing capacity, endurance, and durability. Also, memory devices must be able to work with various technologies that are being introduced into modern cars. Among such technologies are the TCP protocol and Ethernet connection - they provide the network functionality of servers on wheels. It is these technologies that allow traffic to be transmitted to external environments where data processing tools and AI come into play.
Bilby believes that the speed of computing and the performance of memory devices must match the speeds of 5G - it is believed that the speed of these networks will become the standard for networked real-time automotive systems. “There is a need to collect data on millions and billions of kilometers traveled. This data will allow you to create and train models that will work in complex environments and conditions. " The training of such models relies on high-performance data centers that are in dire need of fast memory. Bilby believes that networking functions will become key components of self-driving systems - they will allow real-time traffic data, as well as allow additional training of algorithms.
In the longer term, 5G will also affect the development of cellular V2X communications - it will allow cars to interact not only with infrastructure, but also with each other. Thus, cars on the road will become like a cluster of servers working together - and they may not need a centralized cloud. "These technologies will help predict the state of the environment and understand how a car is driven through 5 or 10 vehicles."
- Russia's first serial control system for a dual-fuel engine with functional separation of controllers
- In a modern car, there are more lines of code than ...
- Free Online Courses in Automotive, Aerospace, Robotics and Engineering (50+)
- McKinsey: rethinking electronics software and architecture in automotive
About ITELMA
List of useful publications on Habré