After an initial burst of autonomous activity, the automotive ecosystem regrouped, re-evaluated its goals, and is now ready to begin deploying new technologies made possible by modern development approaches and forward-looking vehicle architectures.
The pandemic hurt vehicle sales in 2020, but it also gave the OEMs a chance to catch their breath. Panic over announcements from other carmakers about fully autonomous vehicles in 2018 and 2019 have subsided, and reality has set in.
“People have had time to stop and think and realized they had better get started on something here, and it’s centered around trying to intersect the 2025 – 2026 model year,” observed David Fritz, senior director for Autonomous and ADAS at Siemens Digital Industries Software. “The world’s leading automotive OEMs are now preparing to adopt, if they haven’t already, next-generation architectures. This means understanding what they look like, what’s inside of them, the similarities between what smartphones and tablets have done in the past, and how that intersects with where things are headed.”
This new wave of adoption in automotive is an evolution from mechanical to electrical solutions, which includes a a tightly integrated combination of hardware and software. Most industries have struggled with this, largely because the software and hardware engineering teams operate semi-independently, using different tools and and languages. That impacts performance as well as energy efficiency, and in a vehicle where the movement and processing of data may need to be real-time, it’s critical to get this right.
“The epiphany the OEMs had is that it makes a lot more sense to spend the effort upfront to model the entire system, so they know when they put all the pieces together they will actually solve this big, complex problem,” said Fritz. “That, in turn, opens the door to having the models, starting the software earlier, finding the bugs earlier, and removing them. It launches the automotive industry into the same solution space that the semiconductor industry has been in with smartphones, digital TVs, and inkjet printers, of understanding the application first, and then building it — not building it, and trusting it will all come together.”
This year, automotive OEMs will implement modeling to make sure they have hard requirements to pass to chosen suppliers. That could be a new in-house team. OEMs have been actively building those teams, which is causing ripples across the ecosystem. Some of this activity circumvents Tier 1 suppliers and goes straight to Tier 2 or Tier 3 suppliers, because the OEM now can communicate exactly what to build.
This is apparent with service-oriented architectures (SOAs) being embraced by automakers. These are required for over-the-air (OTA) software and firmware updates, but they also sidestep the traditional automotive supply chain hierarchy.
“This now means the OEMs have the ability to implement cloud management software, and software programmability using OTA software updates to reprogram and add new services to vehicles,” said Ron DiGiuseppe, automotive IP segment manager at Synopsys.
Many carmakers already update software or firmware for infotainment or sensor calibration, but OTA is becoming much more integral to the operation of a vehicle. Tesla, for example, recently updated its software to provide full self-driving capability. DiGiuseppe noted that SOA must be designed in from the start to support this.
“Typically, that means the telematics module has to support OTA,” he said. “It has to support the software stack, such as an adaptive AUTOSAR software stack. If an RTOS supplier supports OTA, then it’s got to be directed toward the actual controller. Also, the central gateway typically has to support OTA. This whole service-oriented architecture has to be designed in, and we’re seeing a lot of activity there.”
The chips themselves are becoming more complicated, as well. So rather than separate domain controllers, such as ADAS and cockpit, many of these functions that previously were separate are now being integrated together.
“Where the industry continues to consolidate from that distributed architecture to multi-SoC complex functions, like ADAS controllers, those controllers have to perform multiple ADAS functions on a centralized module,” DiGiuseppe said. “So it’s going to need multi-core processors. It’s also going to have to support ADAS vision applications like automatic emergency braking, or lane-keeping. Automatic braking will make use of lidar and radar sensors, and there are multiple types of sensors here, multiple types of algorithms, such as AI-like, vision-based algorithms. The impact of this is much more complex processing, and therefore, in these ADAS domain controllers we’re seeing a trend toward 16nm or 14nm finFET processes because of all of the integrated applications. It’s also pushing toward 7nm and 8nm finFET processes, mostly because of multiple applications with high processor core counts, like 8 to 12 Arm 64-bit high-end cores, along with AI accelerators, which could be vision accelerators, or DSP accelerators for radar or lidar. Looking ahead, that integration will just continue when we go from distributed controllers to centralized compute platforms. Even more applications will go onto a common module, so it’s still going to be multi-chip. There’s no way you’re going to get all these applications on a single SoC. As a result, centralization will only accelerate with even more applications.”
A recently-published report by Arm highlighted the migration toward heterogeneous computing. Systems such as in-vehicle infotainment and ADAS are becoming more complex and interconnected. Arm said the diversity of requirements needs both a holistic approach to electronics design as well as more heterogeneous compute architectures.
Electric vehicles gain traction
Alongside of new chip architectures, electric vehicles are becoming more popular. EV development will accelerate in 2021, with more EV vehicles available, set against OEMs’ continuing to race toward better efficiency, said Marc Serughetti, senior director product marketing, embedded software solutions at Synopsys.
Tom Wong, director of marketing for design IP at Cadence, noted that electric vehicle sales will reach about 5% of total vehicle sales in 2021. “In the U.S., there are more than a million units already, whereas China has close to 3 million,” he said. “At the same time, battery costs are coming down to the magic number of $100 per kilowatt hour. Depending on who you talk to, we have already achieved that or close to it. This is thanks to the industry learning so much over the last 5 years, along with huge investments that have been made by German automakers, as well as General Motors, Tesla, and by EV battery makers. We’re seeing the crossover point, and this progress makes EVs very cost-competitive compared to internal combustion engines.”
Some believe a key turning point of affordable EVs is GM’s Ultium battery, which is expected to reduce the cost of an EV, also potentially making GM a viable force in producing competitively-priced EV batteries.
Around the globe, electrification continues. In Germany, the EV fleet is expected to continue to grow in 2021, presumably slowly but steadily, according to Roland Jancke, head of department design methodology at Fraunhofer IIS’ Engineering of Adaptive Systems Division. “We probably will see a broader range of assistant systems in a larger number of vehicles, and hopefully we will see successful examples of vehicle-to-vehicle or vehicle-to-infrastructure communication with commercially available cars, rather than with field test vehicles.”
In the U.S., EV growth is expected to transform America urban and rural planning, according to Willard Tu, senior director, automotive at Xilinx. “In many countries such as Japan, China and Germany, train stations are a hub for shopping and restaurants. EVs will necessitate the need for charging stations at hubs that will mirror these train stations, providing services to entertain consumers while they wait for their EV to recharge. Accompanying this, battery recycling will emerge as a business, given the need for approaches to deal with battery disposal and recycling to prevent future environmental impact.”
Many companies take environmental impact very seriously. Infineon, for example, has what it calls core beliefs for the automotive market, one of which is “Zero CO2 becomes real,” which points toward electrification, explained Joerg Schepers, vice president for powertrain microcontrollers at Infineon. “This is in line with, and confirmed by, an increasing number of countries and governments that are banning internal combustion engines in the timeframe between 2035 and 2040. While California is a prominent example, this is also seen in the European Union. It is indisputable that the era of the classical combustion engine is going to be terminated, marked by a transition phase with hybrid vehicles.”
ADAS first, autonomy second
Not surprisingly, ADAS development will be a strong focus in 2021, as well, Serughetti said. “Autonomous vehicle development will continue, but OEMs want to deliver capabilities that customers want and will pay for.”
ADAS is considered a steppingstone by most automakers toward total autonomy, which will be rolled out in limited use cases, such as on highways and in geofenced areas, long before vehicles are capable of navigating urban streets. Autonomous trucking will likely be one of the first beneficiaries of this rollout, particularly on limited access highways.
“The trucking industry is already going autonomous, and we’ll see even more activity in this space in 2021,” said Xilinx’s Tu. “We will see long-haul trucking from depot to depot, highway-only driving on special corridors that are enhanced with special 5G connected land-based sensors arrays, and emergency remote driver intervention in drone mode.”
At the same time, AI will go beyond ADAS, he said, including self-diagnosing and self-healing vehicles. That will significantly reduce the number of breakdowns. But it also can be used in student-driver mode, where new drivers are surrounded by cameras, radars and lidars to avoid accidents. Tu said that a fully autonomous vehicle is further out, but carmakers already are planning these kinds of vehicle valets.
“These systems will keep drivers and passengers safe, informed, entertained and productive,” he said. “Voice and gesture UI will intuitively understand commands. Augmented reality will aid in navigation, providing points of interest or alerting of safety issues, using VR projection to reveal what is behind pillar blind spots. These AI valets will need a unique blending of AI (voice, gesture), and as all of these AI technologies are only in their infancy, hardware systems will need to adapt as the systems improve with OTA updates maximizing the longevity of the system.”
There is already quite a bit of ADAS technology on the road today. “When you look at autonomy, we’re living with different levels of ADAS, such as ADAS Level 2,” Wong said. “We’ve talked about this for years, but hidden under the radar is the fact that there are probably 100 cars you can buy today that already have it. They don’t call it ADAS, but rather adaptive cruise control, automatic emergency braking, and lane-departure warnings. The next level is how to make these features better, which gets into sensor fusion, but progress is happening. Cameras are getting better. 3D cameras are getting deployed. You can see that a 3D camera is a half-step toward lidar at probably 1/100 of the cost, so clearly that’s moving in the right direction. With the cost of lidar coming down substantially over the last three years, the barrier to adoption is no longer cost, and this is making it possible for startups to do interesting things in this space.”
Fig. 1: Where different technologies will be used. Source: Cadence
And while Level 5 autonomous driving may still be far off for the average consumer, fully autonomous, driverless robo-taxis have started to be deployed in cities in the U.S., France and China. “Here, machine learning/AI may be able to provide statistics for the learning to guide the pathfinding for an autonomous vehicle, but that data is very specific to each geographic location. The data collected in downtown Phoenix is probably not very useful in Shanghai or in San Francisco because consumer behavior is different,” Wong said.
Gateways handle overall car connectivity
Once all of these autonomous vehicles are on the road, they will need to be connected. Most automakers either already have or will be adopting over-the-air updates for software, with more expected to be deployed in EVs than on internal combustion engine vehicles, Wong observed. “When you update software to a car, what are you updating? A lot of the electromechanical systems are really not connected for a software update. Even though you can update something to the car, it wouldn’t affect the mechanical system, so you see a lot more OTAs updates on the side of an EV.”
Then, in terms of connectivity between the car and the outside world, the most common technology concept discussed is vehicle-to-infrastructure (V2X). “What exactly are the standards of communication between car and infrastructure? DSRC (direct short range communication) is already deployed. The next step is 4G, and eventually 5G. Interestingly, 4G has been available in high-end General Motors cars for three or four years now, but there’s fairly limited usage. They’re not using it to update software. They are using it to enable you to call somebody when you get locked out of the car, and they unlock the car for you — simple things like that,” he continued.
Within the vehicle, along with a zonal architecture that is being embraced by automotive OEMs as a requirement for autonomous driving for a more centralized computing infrastructure, there will also be some distributed multi-purpose, zonal controllers that can take manage all of the sensors.
To make this possible, a high-speed, in-vehicle communication protocol is needed. That is expected to be either MIPI A-PHY or Ethernet, which is evolving toward 10 gbps, said Robert Schweiger, director, automotive solutions at Cadence. “More OEMs will adopt these high-speed communication standards because they enable so many other things. With that, you need high integration ECUs. The central compute units are really massive computers with very complex SoCs sitting in the middle, and are very powerful. They will sit alongside the zonal controllers. If these are multi-purpose, smaller computing units, they will evolve as an incarnation of the multiple consolidated ECUs that we have in the car today.”
Given the specialized conditions in which these SoCs and controllers will operate inside the vehicle, intelligent power management is a requirement, and is top of mind this year for automotive SoC developers.
“If you park your car on the side of the road, and you want to hear some music or you want to call up your friend, maybe only the infotainment system needs to be powered,” said Schweiger. “The rest of the system you don’t need. Intelligent power management plays a role here. On the other hand, if you’re driving on the Autobahn at 130, you probably won’t need the parking assistance. So intelligent power management also includes simple things like that. We will see much more of this, especially at higher levels of automation. It will become a safety issue because you need to make sure that your car is always sufficiently powered in terms of the available energy. Otherwise, the mandatory systems won’t work anymore. It’s not too far-fetched to imagine the overall power management could switch off certain systems to ensure that you can still get from point A to point B, and this now makes power management a safety issue.”
More applications mean more algorithms, which require higher-performance chips. That, in turn, drives up the power requirements.
“It’s not just about the absolute power number for automotive, because of where the chips are located and a lack of cooling or airflow,” Synopsys’ DiGiuseppe noted. “Power in automotive is more about how to remove the heat of these chips and modules rather than a longer lasting battery.”
Functional safety, security
On the development side, functional safety and security will continue to play an increasing role in system testing, impacting costs and requiring new techniques.
According to Arm’s recent survey of its ecosystem partners, functional safety was identified as the most important factor for achieving success in autonomous computing, and the biggest challenge to achieving the mass deployment of Level 4 vehicles. Arm said it will be important for the industry to re-evaluate how the technology of functional safety needs to change as the demand on automotive electronics rises.
Functional safety cannot be avoided because it is such a key element of automotive systems, even if it does add a lot of new requirements on top of existing semiconductor specifications, noted Guillaume Boillet, director of product management at Arteris IP. “Even at the network-on-chip level, there are enhancements that satisfy ASIL requirements up to ASIL D.”
Connected to safety are increasing security concerns as the connectivity of vehicles expands, Cadence’s Schweiger said. “The surface of a vehicle is constantly increasing, especially if there isn’t a driver anymore. It’s like a remote-controlled object, where you can do very bad things, so safety and security are a big deal. This translates into the design tools; whether it’s a safety-certified design flow for IC for PCB or certain verification mechanisms, how do you verify that a chip or a system is safe? It’s the same thing for security. How can you prove that from this one area to a sensitive area there is no physical or logical connection possible?”
Still, the next cyber-attack on autonomous vehicles can be avoided with investment into secure software. “In 2021, cyber-attacks on the automotive industry will pick up speed and we will see our first successful attack on an autonomous vehicle,” said Ian Ferguson, vice president of sales and marketing for Lynx Software. “The future of mobility is all digital, and thus, all vulnerable. While 2020 saw a number of attacks on the manufacturers, which saw companies like Honda and Tesla experience attacks on personal data, attacks are becoming more complex and vicious, with research showing that it can take up to 25 seconds for people to regain control of the vehicle if hit with a cyberattack. If they’re traveling at any real speed, it is likely to be fatal. Attacks on data centers and, irreparably, human life can ultimately cost the automotive industry upward of $23 billion by 2023. While the industry has seen many researchers acting in a ‘white hat’ way, 2021 will be the year manufacturers will take action and invest in a software platform that adapts to evolving hardware with security by design.”
It wasn’t that long ago that security was viewed as something nice to have, noted Kurt Shuler, vice president of marketing at Arteris IP. “I’ve got some friends who are in the security IP business and I joke with them that they’re actually in the insurance business since there’s no requirement for any of this. A smartphone may be insecure, but five years after you sold the IP or the software, you’re not likely to be sued for having an insecure phone, so who cares? Cars are different. Cars have a safety of life issue, and now there are standards because of that, and it’s something being designed in upfront. Not just market forces, but also standards make things safer for everybody.”
With all of the ecosystem adjustments, and technology integration challenges, the role of the automotive OEM continues to evolve in 2021. As systems becoming increasingly autonomous, OEMs will take on full responsibility for everything from developing and verifying the entire system and ECU designs, to the legal ramifications should a vehicle fail.
“When that happens, because the OEMs are dealing with all of their subcontractors, and also because things roll downhill, even the development interface agreements being signed now sometimes have questions about security, and security features within the IP,” Shuler said. “Those are things you didn’t see a few years ago. But this is a good thing. People are thinking ahead of time on these complicated car systems, and everyone is feeling the struggle.”
How Will Future Cars Interact With Humans?
Adapting technology to the driver can have very different results than training the driver to use technology.
IP Safe Enough To Use In Cars
A look at the different ASIL levels in safety-critical design.
Who Owns A Car’s Chip Architecture
Carmakers and their suppliers compete for dominance, creating challenges across the electronics industry.
Using ICs To Shrink Auto’s Carbon Footprint
How chips will play an increasingly vital role in improving efficiency of cars and infrastructure.