Software-Defined Vehicles Ready To Roll

Software-defined vehicles are driving a swell of activity across the automotive ecosystem, including new methodologies and technology approaches that could significantly reduce costs and shorten time to market for advanced features.

The SDV approach encompasses more than a single concept. It helps to think of it more as a modeling approach that connects EVs, driver assistance technology, and fully autonomous driving, as well as a whole bunch of other features and services. SDVs works like blueprints for how both software and hardware interoperate, while also being flexible enough to make adjustments with predictable results.

“In the past five years, a lot of things have shaken up the automotive industry,” said Suraj Gajendra, vice president for automotive products and software solutions at Arm. “Even before Covid there was a lot of discussion around autonomous driving. Then Covid hit and expectations reset. Back then, I still remember having casual conversations with people that by 2024 or 2025, we’ll have Level 4 or Level 5 autonomy on the road. In the automotive industry people realized that autonomy is just one piece of the puzzle in the evolution that’s going on in the automotive industry.”

This is where SDVs fit in. “The software-defined vehicle enables better autonomy, but at the same time it’s also important today in the cars we drive for the ability to upgrade, have new applications, the ability to monitor things,” Gajendra said. “We are in the middle of this massive evolution, and it’s happening at a pace that the automotive industry is normally not used to. A lot of different sorts of people are driving the industry. It’s not the traditional OEMs to Tier Ones to Tier Twos to IP suppliers. The whole supply chain is evolving massively. There are software developers now jumping in, offering new applications. 2023 was a stabilizing year since the reset. I expect 2024 and 2025 to be exciting in terms of this evolution increasing at a much more rapid pace, and I can see this evolution continuing for at least three or four more years. By then, the software-defined vehicle will likely be a norm, but being able to use some of the new applications in vehicle will be a norm in the 2027-2028 timeframe.”

Once implemented, SDVs can help solve the difficulties of ad-hoc, incremental change methodologies for implementing advanced automotive technologies.

“The ‘big thing’ in automotive changes on a regular basis, but right now it’s all about software defined vehicles,” said David Fritz, vice president hybrid and virtual systems at Siemens Digital Industries Software. “And just like every other fad that’s gone through over the industry, it takes a while to figure out and normalize the definitions of just about anything that’s new. What exactly is a software-defined vehicle? The automotive industry loves to latch on to these labels, and then everybody takes them in whatever direction they go. Once they coalesce, they find out this is not terribly different from what we were talking about not long ago. This is because everybody sees there’s such a wide-open opportunity with automotive, yet we’re still seeing an awful lot of failures. A lot of companies are going out of business, and we’re still seeing more consolidation. However, the automotive ecosystem is finally getting to the point where they understand exactly what they don’t know and what they need to go figure out. “

The SDV concept has been kicking around for some time, but it largely went ignored by many of the established carmakers. Danny Shapiro, vice president for automotive at NVIDIA, noted that the SDV approach has been central since his company began developing technology for cars. “It was always a computer that’s programmable that goes in the vehicle,” he said. “We’ve never been a chip supplier to the auto industry. That was the old model. People would say, ‘We need to do this thing in the car. Okay, let’s design a chip.’ You’d hard-code the features in that chip, and you’d put that in a box, and it would go to a Tier One. Then the OEM would put it in the car, and it did that one thing. You never touched it again. That’s essentially the evolution of the automobile industry for many decades. And now there are 100 of these ECUs under the hood and 60 miles of wire. However, in the same way we don’t say we have a software-defined computer or a software-defined phone, eventually we’ll be able to drop the ‘software-defined’ and just call it a car.”

That doesn’t diminish the significance of this trend, however. Electrification, driver assistance, and autonomy remain the big goals. The challenge has been how to realize them reliably and quickly, and at a reasonable cost, and SDVs help to achieve those goals.

“This is unchanged from 2023,” said Judy Curran, CTO for automotive at Ansys. “At the same time, I see an over-arching trend of affordability and efficiency for all of these. Yes, we all want to drive electrified vehicles, but at the moment consumers are not purchasing them as readily. So as much as we want that to happen, and many OEMs built battery plants and made declarations of being 100% EV by a certain year, some are slowing down because the consumer has to come along.”

Curran points to two main reasons for slowing EV adoption. “One is the affordability of EVs,” she says. “It’s still a chunk of change more than a combustion engine everywhere in the world but China. Then, of course, the interest rates haven’t helped. OEMs and suppliers are really thinking about how to make these electric vehicles more affordable, and that is in the piece price of the actual components in the vehicle, but also in the way they develop vehicles. If they can engineer them faster, more efficiently, then they don’t have to schmear the cost of the engineering across the price of the vehicle. Also, there is still concern by consumers about ‘My lifestyle is x,y,z. Will this vehicle let me drive the way I want to drive? Will there be charging stations?’ That said, there’s a lot of work to be done in the charging infrastructure.”

The role of SDVs
There also is much work that needs to be done before full autonomy is reliable and affordable enough for widespread adoption. In the near future, the real focus is on ADAS.

“Autonomy is complicated, and when you look at all the different environments, all the different scenarios and people and roads and weather and buildings, to think that you’ve covered yourself from a safety perspective, autonomy is going to go further out,” Curran said. “There’s a lot of being written about how many billions of dollars has been spent by the automakers, and some like Ford shut down their autonomous activities to save money, because, the other thing is, when you try to look at all these trends as an automaker, you can’t afford to do them all. You have to select your winners, and so people are slowing down their investment. However, because of all the learning from autonomous, we as an industry are developing incredible ADAS systems, which is very important. And people will pay for that. Will they pay for electrification? Maybe, but it’s rare that a person will say they don’t want to pay for a safety feature. I found during my years at Ford, ADAS was something people would pay for because whether it’s themselves, their children, their elderly, parents, whomever, people want safety.”

An SDV approach is a way of achieving that faster and more efficiently, but it requires an ecosystem to make it viable. “It’s important to provide a scalable semiconductor solution lineup to enable customers to make the transition to SDVs at their own pace,” said Daniel Sisco, senior director for digital system architecture at Renesas Electronics. “This means flexibility to cover all vehicle ranges and software applications is critical.”

To this point, Renesas announced a roadmap for the next generation R-Car Gen5 SoC, which is a scalable family perfect for application across models, applications, and vehicle classes.

In a similar vein, FPGAs will continue to play a role here because of their inherent flexibility. “There’s definitely FPGAs going into cars, with an estimated 8 to 10 chips in a typical high-end car,” said Geoff Tate, CEO of Flex Logix, noting that FPGAs being used for ADAS and SDV applications. “People are using FPGAs for flexibility, maybe for the software-defined approach, but as the volumes go up — since FPGAs are expensive, big, and power-hungry, and there’s lots of other chips — people will look to integrate the FPGA into an adjacent processor chip. That will lower power costs, but keep some flexibility. And with embedded FPGA, we can deliver the same performance as the FPGA chip. Over time, FPGAs going into automotive will lead to embedded FPGAs being increasingly used in automotive.”

There are other challenges with evolving automotive architectures. As automotive software in the automotive industry becomes more complex and extensive, how do semiconductor companies manage software development to reduce complexities and accelerate time to market? Renesas’ Sisco said that by using a common software framework across MCUs and SoCs, as well as across generations of high-performance digital compute products, development time can be shortened, and a high degree of reuse can be enabled. “Renesas has a strong focus on early development using virtual platforms, as well as a unified cloud platform to support development tools,” he said.

Automotive cybersecurity
Cybersecurity in automotive applications is always a concern, and the SDV approach can help here, as well.

“In 2024, the automotive industry will continue to have a sharp focus on semiconductor chip security due to increased legislative scrutiny and a growing awareness of hardware vulnerabilities,” said Andreas Kuehlmann, CEO of Cycuity. “Key regulatory initiatives, such as the EU’s Cyber Resilience Act (CRA) and the U.S. Cybersecurity and Infrastructure Security Agency’s (CISA) hardware bill of materials (HBOM) framework, are shaping global standards by emphasizing secure methodologies throughout a product’s lifecycle. Notably, the CRA imposes significant penalties for non-compliance, reaching up to €15 million or 2.5% of global turnover, underscoring the importance of integrating security into chip design. The decision by Porsche to discontinue the ICE-powered Macan in the EU in 2024 exemplifies the impact of these regulations, particularly for vehicles developed before the establishment of cybersecurity requirements.”

Guided by standards like ISO/SAE 21434, the automotive sector advocates a ‘security by design’ approach to enhance vehicle resilience against cyber threats from the early stages of design and production. “Given the nature of persistent hardware vulnerabilities, leading semiconductor vendors are calling for robust security assurance practices that go beyond traditional methods,” Kuehlmann said. “As we move into 2024, there’s an increasing expectation for collaboration among security, design, and verification teams to foster and strengthen this mindset. By anticipating the possibility of escalating threats, the industry’s commitment to addressing vulnerabilities and enhancing security defenses poised to grow, paving the way for a more resilient and secure automotive landscape.”

Bart Stevens, senior director of product marketing for Rambus’ Security Division, said there is an ongoing effort to educate automotive OEMs on security in general, as well as security implementation — how to do it in a robust and safe manner, while also reducing market risk.

“That information might be picked up by the OEM itself, or they will use that to influence the value chain down to the Tier Ones and Tier Twos,” said Stevens. “The goal is to allow someone to pick up a design and use it like a LEGO block in silicon. If it is a robust design, with a robust test environment, has good support and documentation, and now is certified to the various automotive and security standards, this certified design helps customers’ risk reduction. Traditionally, they would still have to certify their design. In most cases, they would get IP with a big promise from a sales guy saying, ‘Hey, this is good stuff. Put it in silicon, and we’ll help you go through certification.’ But now we can slap a certificate to it and say, ‘This has been certified. You can pick up where we left off. And if you don’t modify our design, then you could use this to speed up your certification, and more importantly pass and not fail when you already have silicon and would otherwise have to go back and change the mask set.’”

Regional variations can create other issues. “If you sell products worldwide, what’s happening is that the encryption algorithms that China will make you implement are different from those in the U.S.,” said Flex Logix’s Tate. “That means you might have to build two different chips, which is very expensive, takes a lot more engineering, and delays schedules. And now you have to double the inventory. What if your China sales ramp faster than the U.S.? You can’t ship them the U.S. chips. So customers in that situation — and these are laptops, tablets, and so forth — are looking for a programmable approach where they can put the security functions between processors. But processors aren’t always fast enough, so FPGA is used for that. That’ll be an issue with automotive as well in the future, because there’s a big Chinese market for cars.”

System-wide concerns
Functional safety and cybersecurity are table stakes when it comes to any automotive-related use case, and both of those are system-wide issues. “It’s not just autonomous driving or any one specific application,” Arm’s Gajendra said. “It’s anything you have to do with the car, because there’s lives at stake. The car has to make sure it performs the functions it’s expected to do, be it Level 2-plus autonomy or Level 5. It has to function properly. Also, if there is a malfunction, it needs to be able to either come to a safe state or let the driver know that there is a problem so they can take full control of it.”

Here, too, the SDV approach can help. “If you talk to somebody who’s been in the legacy automotive industry for a long time, they will say the benefit of having a distributed compute architecture is that they can make sure the braking system is completely isolated from, say, the infotainment system,” Gajendra noted. “They will say, ‘Look, my computers that will actually handle the vehicle breaking and any of the vehicle functions are completely isolated from you playing music. I will ensure that there are no problems as such in my braking system.’ But when they consolidate all these functions into a central car compute platform, if there is a glitch in the infotainment system software, and if it malfunctions for whatever reason, nobody is going to lose their life there. My favorite song doesn’t play. Fine. But if that software glitch interferes with a vehicle function, now that we’ve consolidated the architecture, then that is a problem. You don’t want to have that software glitch or interference when you’re going 70 miles per hour on a freeway, or you don’t want the car automatically upgrading software when it’s running 70 miles per hour. Another example is, say, for people that drive in London, the parking app company, now has payment gateway information that has to be passed on. There are credit card details that have to be passed on, so there are security aspects that they have to take care of, as well. For anybody to be comfortable using those features that app company is providing, they need to be assured that proper cybersecurity guidelines are being followed.”

Standards necessary, but not sufficient
With the recognition and desire of the automotive ecosystem to design, develop, and deploy SDVs, a number of groups have come together to collaborate across the automotive landscape to make that a reality. In March 2023, AUTOSAR, COVESA, Eclipse SDV, and SOAFEE realized that a collaboration was needed — not just within one consortia, but between them. That, in turn, led to the SDV Alliance’s “collaboration of collaborations,” to help clarify the contributions each consortium brings to SDV, and then to actively work together to demonstrate their interconnectedness and synergy.

According to the SDV Alliance’s announcement, the main purpose of the collaboration is to align efforts in the SDV ecosystem, and by embracing existing descriptions of SDV from each of these efforts, as well as other external organizations, the SDV Alliance will agree on a clear and unified definition of what constitutes an SDV. The Alliance then will look at the different technologies, methodologies, and standards of each organization and show how they can work together for the development of the SDV.

Recognizing that the SDV is conceptually too complex to be handled in a single industry consortium, and by looking at each of the organization’s core competencies and varied execution environments, the Alliance will pool these skills to create a joint SDV vision. The SDV Alliance also plans to showcase technical alignment between the different companies.

Fig. 1: SOAFEE-based vehicle architecture. Source: Arm

Siemens’ Fritz expects in the second half of 2024, the automotive ecosystem players that really understand the industry dynamics will become obvious. “We’re going to start seeing people who have a holistic view of what it’s going to take to make SDV a reality. I like to look back at smart phones, because so many new methodologies were implemented for smart phones, and they’re mainstream now. For capabilities like being able to run the software before you tape out the hardware, we’ve been doing that for 15 years. But that seems like black magic to the automotive companies. When you explain that software-defined vehicles means the software has to define what the hardware is going to do, they ask, ‘How does it do that when the hardware doesn’t exist?’ There’s a whole methodology for figuring that out. And then they’ll ask, ‘What if the software doesn’t exist?’”

There are tools that do hardware and software synthesis, of course. “For the automotive companies, they don’t know exactly what they need, but they are starting to understand there is a good place to start,” Fritz said. “They know it should behave like ‘this’ and behave like ‘that.’ There are development systems that will generate some hardware and some software with that behavior, so they can then collect the metrics and say, ‘No, that’s not right.’ Once they resolve that, then they can split off into the software path and the hardware path, develop more detailed models of the hardware, and go back and run a more detailed implementation of the software to see if they are still on track. That’s software-defined vehicle. That requires things like SOAFEE and an agile methodology, but without this process underneath it and the tools to support it, it’s just a pipe dream. Some companies are coming to that realization. But there are several methodology changes that have to happen here. We have to get to the point where we can dictate what the hardware is going to do based upon the workloads of our software, and not just say, ‘We have a datasheet that we got from our Tier Ones. Which one do we like the best?'”

Conclusion
It’s generally well understood that old approaches don’t work for increasingly complex vehicles. As a result, many EDA projects already are exploring the automotive space to leverage the SDV approach.

“We’re going to end up with some detailed hardware requirements and software requirements,” said Fritz. “Then the automotive company will go out and do requests for proposals, and requests for quotes, based upon the requirements driven by the OEM, which have been evolving for the past five or six years. It’s finally coming to fruition where the OEM says, ‘I have to be in control of my own destiny. I have to have the tools and the expertise in-house to drive my vendors and not let the vendors drive me.’”

Related Reading
For SDVs, Software Is The Biggest Challenge
Issues will grow exponentially as software-defined vehicles gain traction.
Autonomous Vehicles: Not Ready Yet
Much more R&D and real-world testing is needed to reach safe and secure L4/L5 technology.