With vehicle data network latency, bandwidth, interoperability and power consumption all critical to the safe efficient operation of an autonomous vehicle, PCI Express® (PCIe®) is the key to next-generation driver assistance platforms.
In May, Alphabet's Waymo announced its fleet of 250 robo-taxis was undertaking 50,000 paid rides per week across three US cities. These level-4 autonomous vehicles each integrate 29 cameras, six radar sensors and four LiDAR sensors to understand the world around them. This generates more than 1.2 GB of sensor data each second, which needs to be transmitted for processing with the minimum possible latency. At the same time, the company, like many others, is also investing heavily in autonomous trucks, with a key driver in this innovation being the ability to reduce the cost of transporting goods and produce across the country.
To put into context the volume of data we are talking about for autonomous vehicles, consider that two terabytes of sensor data would be generated for the average 27-minute American commute—all of which needs to be transmitted and processed as fast as possible. As a result, low-latency, high-bandwidth data networks are essential.
And while Waymo may be at the leading edge of what is currently possible, McKinsey predicts that by 2035, a quarter of new cars sold will include level-4 highway pilot, and over a third will include a level-3 urban pilot.
This is an important move. Even though the oft-quoted stat from the US National Highway Traffic Safety Administration (NHTSA), that 94% of serious accidents are caused by human error, is false, 40,000 people are killed each year on US roads. Recent research from the UK and Europe in 2022 has suggested the use of Advanced Driver Assistance Systems (ADAS) would reduce crashes by 24%. And in the US, research by the US automobile group AAA suggests the implementation of six key ADAS technologies could prevent 40% of crashes, 37% of injuries and 29% of crash-related deaths—that’s c.11,000 deaths per year in the US alone.
As we move from there being just a small number of exceptionally well-maintained fleet vehicles to millions of user-maintained cars, the need for sensor reliability (and therefore redundancy) will increase significantly.
This, coupled with improvements in sensor resolution, will vastly increase the amount of data being generated. Much of the focus has therefore been on processor development and the speeds required to enable the fusion processing necessary to interpret that data, but equally vital is how we’ll transmit this data to (and between) the central Artificial Intelligence (AI) processors.
It’s safe to say vehicle data networks are among the most safety-critical components in an autonomous vehicle, yet many of the traditional communication infrastructures and standards—Ethernet, Controller Area Network (CAN), SerDes—used in the automotive industry are reaching their limits. The demands being placed on automotive data networks mean these protocols must be complemented by something more suited to meeting the needs of the high-performance computing platforms (HPC) required to embed artificial intelligence and machine learning within ADAS and autonomous vehicles.
With the network’s latency, bandwidth, interoperability and power consumption all critical to the safe efficient operation of an autonomous vehicle, PCI Express® (PCIe®) is the key to next-generation driver assistance platforms. Indeed, the standard has been adopted not just by the industry leaders like Waymo, but also by sectors that face similar challenges—notably data centers.
The PCIe Standard
Like vehicles, AI and hyperscale data centers need to transmit huge volumes of data without being overly affected by poor latencies. The requirement is to connect not just servers, but storage, network and I/O peripherals; to achieve this, these facilities use the point-to-point bidirectional bus standard PCIe.
As such, the bus is already present in high-performance processors, with its bandwidth further enhanced by its ability to transmit over 16 parallel lanes. The standard is currently in its sixth generation, with Gen7 due for release in 2025.
Automotive data transfer demands aren’t quite at the level of AI data centers (yet) and, therefore, the roadmap for use in automotive applications is focused on PCIe Gen4 to 2030 and thereafter Gen5—both for silicon on chip (SoC) and networking manufacturers. This approach optimizes the bandwidth versus cost trade-off, while enabling the high-speed data handling required of advanced autonomous driving platforms.
Fig 1: A comparison of bandwidth, transmission distance and cost for some of the major automotive data transmission protocols. Note PCIe measured in GBps (8 bits per Byte:256 Gbps/512 Gbps)
The Ecosystem
Automotive Original Equipment Manufacturers (OEMs) are particularly constrained by SWaP-C (size, weight, power and cost) factors. As with the data center, where PCIe is used to transmit time-critical communications that require these huge bandwidths but Ethernet is used for other (less critical) data, automotive networks will also need to implement multiple transmission protocols to balance cost and performance. This means that PCIe is likely to be used alongside Ethernet, SerDes and/or CAN (Figure 2).
Fig 2: With vehicles becoming data centers on wheels, there is a need to balance SWaP-C factors, with PCIe used to handle high-volume and time-critical data over short distances between central processors and Ethernet/SerDes linking sensors to edge processors.
This architecture also enables the shift to cars becoming software defined vehicles (SDVs), where functions and features can be upgraded via over-the-air updates from the cloud and purchased after the sale. This is already being implemented by Tesla, and not only creates additional after-sales revenue streams for automotive OEMs, but has been shown to prevent at least one multi-billion dollar recall.
Automotive PCIe Implementations
Unlike Ethernet, there is no specific automotive PCIe standard, but this does not mean it cannot be applied to safety-critical applications. Likewise, there is no aerospace PCIe standard, yet the protocol is well-trusted by avionics and defense OEMs. As such, silicon vendors are developing PCIe products to specifically be used in the harsh automotive environment.
It is fully possible to attain automotive qualification for PCIe and here at Microchip, we have launched the industry’s first Gen 4 automotive-qualified PCIe switches (Switchtec™ PFX, PSX and PAX families), which enable high-speed interconnects that are tailored for distributed, real-time, safety-critical data processing in ADAS architectures.
We have also developed PCIe-based hardware such as NVMe® controllers, NVRAM drives, retimers, redrivers and timing solutions, as well as Flash-based FPGAs and SoCs.
To find out more about these PCIe technologies please visit our web page.
Daniel Leih, Dec 3, 2024
Tags/Keywords: Automotive and Transportation
Comments