Yet another reminder of the lengthy process from conceptualising a car to it rolling off the production line and into showrooms, and what this signifies with regards to cutting-edge technology…
The Latest Autonomous Technologies Are Already Outdated
July 3, 2023
James Jeffs
if Mercedes can get Level 3 certification with yesterday’s technologies today, what will its car designers be able to do with today’s technologies tomorrow?
Last year, Mercedes released its Level 3 Drive Pilot system for certified use on German roads. Those wealthy enough to purchase a new Mercedes equipped with the Drive Pilot system, which is an optional extra costing a few thousand euros, could relax as their car drove itself up to speeds of 60 kph. This year, Mercedes announced that its Level 3 Drive Pilot had been certified for use in Nevada, with delivery of L3-capable vehicles to that state expected in the second half.
Certified Level 3 is a gargantuan milestone on the way to fully autonomous vehicles. In fact, the step from L2 to L3 might be the trickiest of them all. So it might be surprising to learn that most of the cutting-edge technologies that made this step possible for Mercedes are already out of date, according to research findings from IDTechEx.
SAE levels and L3’s importance
“Level 3” here refers to the Society of Automotive Engineers’ six levels of autonomy, where Level 0 is a completely manually driven vehicle and Level 5 is a completely autonomous vehicle. As you might expect, with Level 3 being in the middle, L3 vehicles are somewhat autonomous; specifically, under some predetermined conditions, the vehicle can drive itself, with the driver free to take their concentration away from the road. This is an enormous step up from Level 2, where the driver is always in control of the vehicle. L2 vehicles typically have a combination of lane-keep assistance systems and adaptive cruise control to ease the monotony of driving. Manufacturers will always want to make those features as safe as possible, of course, but at the end of the day, if the vehicle crashes when L2 features are being used, the driver is responsible.
The six levels of autonomy defined by SAE (Source: IDTechEx)
Part of the challenge with Level 3 has been clarifying who is responsible if the vehicle gets into trouble while the L3 system is operating. Mercedes resolved the issue to some extent by saying the company would assume liability if the vehicle crashed while reporting that it was operating at Level 3. But it also requires lots of confidence in the vehicle’s autonomous system; after all, Mercedes will want to ensure that the profit from selling the system dwarfs any bills it needs to pick up if the system is found liable for a crash.
The autonomous system used in the Mercedes S-Class and EQS is indeed impressive. It features a LiDAR, five radars and six cameras, as well as an Nvidia-powered autonomous brain. All of these are based on semiconductor technologies, all of which are evolving rapidly. Each of the components that Mercedes chose when specifying the vehicle would have been cutting-edge and best-in-class, but semiconductor technologies advance extremely quickly. This means that the tech rolling off Mercedes’s production lines today is a generation behind what the semiconductor foundries can produce.
Here’s why.
How drive pilot technologies are being superseded
This situation is not unique to Mercedes. The progression from conceptualizing a car—including deciding on the desired autonomous features and the appropriate sensors to achieve them—to having one roll off the production line is a years-long process. During that long time to market, new and better sensors will be announced, so that the car manufacturer will be stuck shipping “last gen” tech when the car is finally available.
This effect is compounded when considering the speed of development of the semiconductor industry. Say Company X is designing a car and chooses a radar today. That radar was designed with the semiconductor tech available at the time of its design, maybe one or two years ago. At the pace of the semiconductor industry, that means that the radar could already be out of date—and it won’t even be on the car for another couple of years.
The LiDAR on the Mercedes S-Class and EQS is a great example of this phenomenon. Currently, the vehicles use a second-generation Scala from Valeo. Paris-based Valeo has cemented itself in the automotive LiDAR space, with its design slot in the Audi A8 in 2017 having made it one of the first to market. The second-generation Scala will no doubt be an excellent product, but a little over one year after announcing that the Scala would power its Level 3 system, Mercedes has already announced that it will use Luminar’s LiDAR in the future. In the information within the announcement, Mercedes says that the Luminar Iris will allow its autonomous Level 3 system to operate at speeds of up to 80 mph—double the 40-mph limit today.
The performance increase that Mercedes will get from switching LiDARs will be partly due to the shift from the 905-nm lasers in the Scala to the 1,550-nm lasers in Luminar’s Iris. The longer wavelength means that Luminar can use much more powerful lasers, yielding longer-range LiDARs, while maintaining eye safety. But this change in wavelength also means that different semiconductor technologies will be needed.
Semiconductor foundry smallest-node capabilities in recent years and on roadmaps for the future (Source: IDTechEx)
LiDAR announcements over recent years by wavelength (Source: IDTechEx)
For example, it is common to use silicon technologies for the detector in 905-nm–laser LiDAR because silicon is readily available, well understood due to its maturity and comparatively cheap. Above ~1,000-nm wavelengths, however, silicon stops absorbing light, so
LiDARmanufacturers will have to use something else, such as indium gallium arsenide. InGaAs provides the detection requirements for 1,550-nm LiDAR, but its production processes and supply infrastructure are not as mature as silicon’s, the minerals are rarer and the costs are therefore higher. Pioneering semiconductor startups could offer some hope here; TriEye, for example, has shown that it can build silicon-based image sensors that detect up to ~1,600 nm. Perhaps that technology can be adapted for use in a 1,550-nm LiDAR, a potentially game-changing prospect.
Mercedes’s switch to 1,550 nm is in keeping with LiDAR industry trends. The vast majority of LiDAR announcements these days are for products using 1,550-nm technologies. In select cases, companies are still pursuing 905 nm, but IDTechEx thinks that the industry is coming down from the fence on the side of 1,550 nm.
The autonomous brain is another area where it is nearly impossible to keep up with the pace of the semiconductor industry. Mercedes announced in 2020 that it would use the Nvidia Drive Orin SoC to power its autonomous system. That SoC was based on Nvidia’s Ampere architecture, which today is a generation out of date. The Ampere architecture uses a 10-nm process from Samsung and gave the Orin computational power of 255 trillion operations per second (TOPS). But before Mercedes won Level 3 certification in Nevada, Nvidia announced the Thor SoC, promising 2,000 TOPS. Thor will likely share the 4-nm TSMC process used in Nvidia’s Ada Lovelace architecture, found in its 40-series GPUs. Thor is expected to go into production vehicles by 2025, but by that point, TSMC will be capable of 2-nm processes, according to its roadmaps.
As you can see, the semiconductor industry moves so fast that a new vehicle can appear out of date by the time it rolls into showrooms. The silver lining is that this leaves ample headroom for technology improvements. The Mercedes S-Class and EQS, with their Level 3 systems, are incredibly advanced and capable machines, but by looking at emerging semiconductor technologies, from sensors to powerful SoCs, it’s clear to see that they are already a generation out of date. So if Mercedes can get Level 3 certification with yesterday’s technologies today, what will its car designers be able to do with today’s technologies tomorrow?