The networking challenges facing autonomous vehicles

The networking challenges facing autonomous vehicles

What you will learn:

  • Networking methods for modern cars.
  • Challenges related to automotive network architectures for vehicles, including autonomous vehicles.
  • Potential solutions to the challenges of automotive networking.

In my previous article, I discussed how sensor fusion is paving the way for the future of autonomous vehicles (AV), allowing them to replicate human senses through the combination of data from multiple inputs, including radar, LiDAR, and sensors from the camera. By combining this data, automakers can mitigate single sensor deficiencies to get a complete picture of a vehicle’s surroundings and provide high levels of spatial awareness.

For sensor fusion to work reliably, however, sensor data must be collected, distributed, and processed at very high speeds. Just like the human brain, computers charged with driving vehicles must make split-second decisions to deal with unforeseen circumstances. Even small delays in this process can have catastrophic consequences for the vehicle and passengers.

As might be expected, integrating all the sensors needed for AVs adds complexity to the vehicle and requires a change of approach for automotive OEMs as they design, engineer and manufacture vehicles. The best way to network the myriad of electronic devices is particularly challenging. It could significantly affect the speed and efficiency with which data can be moved and processed, and presents a number of second-order challenges, such as its effect on overall vehicle weight.

The weight challenge

For those who work on automotive and AV networks, vehicle weight is a factor that cannot be ignored. More sensors mean more cables to connect them, which can lead to increased system weight and adverse effects on a vehicle’s overall speed, fuel efficiency or range in the case of electric vehicles. In fact, for many vehicles, the harness is one of the top four heaviest subsystems, weighing up to 132 pounds. in modern automobiles. In addition, complex wiring affects the vehicle’s production speed because it takes longer to complete and test.

This problem is compounded by the fact that more cars are going electric and have to lose excess weight due to the weight gain of the electric powertrain. Additionally, EV manufacturers are often the ones pushing the most for autonomous driving, which means they require more sensors and better networking at the same time while having to remove weight to account for the battery.

Fortunately, advances in components such as Mobile Industry Processor Interface (MIPI) controllers are helping to limit the need for additional cabling and, in some cases, even reduce the total cable length required. MIPI allows you to wire up to 45 feet directly from sensors to CPUs without latency or processing loss. In effect, it is as if the sensor were connected directly to the CPU and negated the need for additional gateways that would add excess wiring.

Additionally, automakers like Tesla are experimenting with new variants of core network architectures, including ring architectures. Recent patents indicate that systems using circular-looking wiring allow all components to connect to the wiring and CPU instead of being individually wired into the CPU at multiple points.

Other advances, such as wireless battery management systems, also promise to help reduce overall cabling needs in future vehicles. Furthermore, trends towards zonal architectures and virtualized electronic control units (ECUs) introduce a notion of more powerful multifunctional ECUs. These would be linked to smart sensors to reduce the overall number of ECUs, which, in turn, favorably affects the wiring.

Such systems have the potential to dramatically reduce the amount of wiring required and can facilitate the installation of wiring systems during production, thereby reducing vehicle production time.

A need for speed

As previously mentioned, speed, as it refers to the ability to move and process data, not miles per hour, is critically important for sensor fusion and AVs. Huge amounts of data must be collected, transferred, and calculated in real time to enable quick decision making, so high throughput is a rigorous requirement. This requirement applies to both data exchanges on automotive networks and data transfers in and out of memory.

Traditional automotive networks such as CAN, LIN, and FlexRay simply do not provide enough bandwidth to transfer the large amounts of data required for sensor fusion and on-board AI computing engines that use deep neural networks. To give you an idea of ​​where we’re headed, Micron estimates that 512 to 1,024 GB / s memory bandwidth is required to support Tier 3 and Tier 4 autonomous driving. At Tier 4, the vehicle is highly autonomous, but in some situations it still requires human interaction.

In 2020, most automotive systems were equipped with LPDRAM x32 components with I / O signaling rates of up to 4,266 MB / s (4,266 GB / s) per device. Achieving higher levels of battery life with a viable number of DRAM devices requires high-performance memory such as GDDR6. A GDDR6 x32 DRAM device running at 16Gb / s offers 64GB / s of bandwidth. An architecture with 16 of these GDDR6 DRAMs could reach the level 4 memory bandwidth requirement.


Many automakers are grappling with the limitations of current technology and looking for ways to achieve the performance levels needed for fully autonomous driving. Looking at the physical, electrical and computational challenges, it is clear that cutting-edge solutions are needed to put autonomous vehicles on the road.

In my next installment, we’ll cover how this transition to EV and AV has also led to another challenge for automakers in a new domain: cybersecurity. We will look at the safety concerns surrounding modern cars and different approaches OEMs are taking advantage of to protect vehicles and consumers.

Leave a Comment

Your email address will not be published.