Automotive OEMs Face Multiple Technology Adoption Challenges

Experts At The Table: The automotive ecosystem is in the midst of significant change. OEMs and tiered providers are grappling with how to deal with legacy technology while incorporating ever-increasing levels of autonomy, electrification, and software-defined vehicle concepts, just to name a few. Semiconductor Engineering sat down to discuss these and other related issues with Wayne Lyons, senior director for the automotive market at AMD; Judy Curran, CTO for automotive at Ansys; Amit Kumar, product management director for Cadence’s Compute Solutions Group at Cadence; Adiel Bahrouch, director of business development for silicon IP at Rambus; and Marc Serughetti, vice president, product management and applications engineering at Synopsys. Part one of this discussion can be found here. Part two is here.

L to R: AMD’s Lyons, Ansys’ Curran, Rambus’ Bahrouch; Synopsys’ Serughetti; Cadence’s Kumar.

SE: How does the automotive OEM decide what they can move into a zonal approach, and when?

Curran: If they have their 80 modules that are distributed and they want to get to a larger number of high-performance computers, how big of a jump do they want to take? The other point we need to remember is that if an OEM decides to go to a new architecture, not every vehicle that gets produced that year has a refresh. So even if you decide you are going to be one of those brave OEMs and go to a complete zonal architecture — but you just launched an F series last year, and the next F series comes in three years or four years — it takes time to go from your past architecture to the new one. So there is that complexity of how quickly they can move to that new architecture. Sometimes they do some step stones that are easier to get across the fleet. For this reason, we see OEMs at different stages of the journey, from a very distributed set of modules to larger domain modules, larger high-computing controllers. Not all OEMs go in one step. Some do partial steps.

SE: Are there certain application areas, like ADAS, that lend themselves better to a zonal architecture?

Lyons: We’re seeing integration of IVI and ADAS. They come together because they’re all part of the drive experience — and visualization around the vehicle, as well. You see it coming into a module, not necessarily into the same device that can be going to the same silicon. But there are some challenges to bringing those software pieces in because there are discrete, independent software teams, as well, developing their software. There’s not a lot of overlap between maybe Android Automotive that’s running your in-cabin system and your QNX or even SafeRTOS that’s running your forward-looking camera and surround view system. So yes, maybe there’s some commercial benefit to going to one silicon system, but there’s a lot of software complexity that needs to be addressed, and freedom from interference in order to do that. We are seeing it getting merged into a zonal unit, or zonal module, as a first step there, and there’s a lot of interest in combining these technologies together.

Curran: ADAS is one that is of interest, and infotainment, too. In general, to address the customer desire for their experiences, there’s been a lot of energy to put all the body control and infotainment together, making them comfortable together in one system.

Kumar: Besides ADAS and infotainment, given the upcoming electrification, the battery management system (BMS) must be zonal and split out because one is the power train of the car, the other is the safety aspect of the car. BMS is a relatively new technology in a vehicle. I suspect OEMs are trying to keep it isolated in a specific zonal container rather than integrating it.

Serughetti: It’s usually easier to bring in capabilities that are not safety-critical. You take less risk on that side, as well. As far as the safety-critical aspect, there’s still a lot of concern. You have to pay a lot of attention. If you have body control, that is not as critical as seats, so you can start integrating one more easily and faster.

Lyons: There are different technology disciplines running at different rates of change, as well, like the in-cabin experience. Car companies are looking to completely reimagine their in-cabin experience, because you’ve got a captive audience for maybe 20 or 30 minutes while they’re charging their car, so they want to be productive. They may want to be entertained. Battery management is a great example. Ten years ago it wasn’t a focus, but it’s huge now. ADAS, as well. Overall, there are some table stakes, like emergency braking and lane departure. But now there are advanced sensors coming in, like lidar or 4D imaging radar. And again, those technologies individually are all going at different speeds and developing at different rates. For this reason, some of them need to stay separate for a while as they’re still maturing.

Bahrouch: It’s already a challenge when it comes to the number of sensors. For example, Level 2 ADAS may be 20 sensors. Then, there may be 30 sensors from camera, radar, and lidar. So the intention of several OEMs is to leverage those sensors for multiple applications. You already can see some overlap of certain applications — infotainment, for example, or a camera that is meant to monitor the driver and make sure they are not distracted or tired, and at the same time, for ADAS, for the user experience with gestures, and the ability to change lanes with the head. You already can see a number of applications where infotainment, ADAS, and some regulatory applications come together. Therefore, it makes almost perfect sense to combine those things. But it comes with a number of challenges because safety and non-safety critical applications must come together. There must be freedom from interference. At the same time, we have technologies like chiplets that allow certain combinations to be created, not necessarily on the same silicon, but through a chiplet. So there are a number of choices and opportunities to combine and to consolidate. However, at the same time, it comes with a number of challenges. It’s not an easy answer, and it’s not an easy path toward zonal architectures with hardware and software together.

Lyons: On top of that you’ve got this whole AI movement, with AI deployed in the car, in perception, in the forward-looking camera, in the transformers for around the car, and then in the cabin, as well, for natural voice recognition. That completely changes the amount and the speed of memory you need. You also must couple very high bandwidth memory with the processing in the hardware system in order to deliver the right level of performance. In the long term, chiplets will be necessary to deliver the high performance and high bandwidth system requirements.

SE: Do you see a use for AI in dealing with legacy technology in automotive? Will there be AI systems that help the OEMs determine what needs to be saved, what should be brought forward, and keep track of it all?

Curran: One of the most obvious use cases for AI is in the pure engineering design and release process of the industry, because most suppliers do a few components, and they do the same components for many OEMs. They have quite a repetitive process of doing that component, whether it gets installed in a vehicle for VW or Hyundai or GM or Ford. So in the workflow to design, and use past designs to iterate and generate the next design, there’s a huge opportunity to reduce the engineering cost. One thing we haven’t talked about is the biggest trend that I see lately — the industry is struggling financially. With pressure from Chinese vehicle costs, everybody is asking, ‘What are we going to do to reduce the cost of our vehicle?’ That’s a big trend. Part of reducing the cost of the vehicle is, of course, reducing the price of the components of the vehicle, but also reducing how much money you spend to engineer the vehicle. One of the easier use cases would be in the engineering processes, because there is a lot of repeatability and repetitiveness in the industry. Then, also in the models themselves, whether they’re digital twins and using hybrid data, as well as physics-based models using AI techniques. These are huge opportunities, and they need to go faster to use them because now is the time to get some cost savings.

Lyons: It could be a forcing function on legacy, as well. They now have to make some difficult choices about what they continue to bring with them and what they leave behind.

SE: What are the opportunities for AI to improve the automotive engineering process and/or reduce costs?

Serughetti: There are a lot of areas. There is the collaborative type of AI technology, which we see a lot of today. But as you get toward generative AI and the type of things you can do to help the engineer, the biggest challenge will be things like when you have a development flow and you’re trying to find the right architecture. With all this complexity, you need to test more things interacting with each other. But no one can think of all the tests that are needed. These processes need to be completely optimized. You cannot go through brute force anymore. You need to start things earlier. We mentioned digital twins. You cannot wait to be on the big physical test bench. You need to do things earlier, and the AI capability that you can envision along the way will have addressed the impact on those things. Still, there are so many areas there to look at. It’s quite incredible.

Kumar: Even with the virtual prototypes, you don’t have to actually do 10 revs of prototyping before you reach the final mode. You could probably cut that down by 50%. Of course, you still have to do that prototyping, but AI can play an important role there.

Lyons: We’re seeing this in the testing phase, as well, with AI being used to pick up corner cases. When you’re running a car in field testing, you’re processing a lot of data. If you use AI in the testing phase, it can pick up some of the corner cases, and then you’re gathering the data, and you monitor it for that period of time, plus or minus when that corner case occurring.

Serughetti: Also, for things like root cause analysis and those types of problems, this is where AI will be applied. There’s this other part, too — data. We talk a lot about the AI, but data is another important piece in that automotive environment.

Curran: I experienced a lot of testing in my career, and sometimes I would think, ‘I can’t believe all our confidence is on this one hand-built vehicle, and some person does a test that passes and then we’re good to go.’ The opportunity to simulate, iterate, do edge cases, do a lot more scenarios, is going to improve the quality, and quality is money. When you look at the warranty bills and the recalls that these OEMs are going through right now, they’re on the increase. They’re not decreasing. So in the world of levers to make the industry more profitable in light of the new technologies they have to absorb in their vehicles, reducing their warranty bill is another opportunity where using some simulation tools and artificial intelligence techniques would be really valuable for the industry.

SE: How does this aspect of AI play a role in moving things forward?

Bahrouch: I agree with the previous statement in terms a lot of optimizations, a lot of accelerations in terms of testing, validation — but also finding the corners, the edge. This used to be very tough and labor intensive, and now you can speed up and solve the engineering part, the production part. But at the same time, it will also help the user experience. This is related to the quality, but the quality in terms of how the OEM will be perceived, the quality of the car, the user experience, how GenAI is used to support the driver to navigate from point A to point B. And because the profile of the driver is very well known, it comes with a number of suggestions. It can suggest a charging place, for example. So there are a number of optimizations that can strengthen the emotional binding of the vehicle and the user, as well.

SE: Are there any final thoughts on next steps for considering legacy and automotive?

Lyons: From our perspective, it’s driving hardware in terms of 10X compute from our current generation to next generation, and the implementation of additional technologies like AI. We come from a perspective of flexible or adaptable technology, so we allow our partners to cheat a little bit and implement some hardware fixes in the car, as well, in the in the flexible, adaptable fabric there. We can still house some legacy interfaces, or even bring in new interfaces ahead of time there. Two years ago, people thought this was happening overnight. Today there’s a more realistic attitude about how quickly this is going to happen. But it absolutely is happening, and it’s going to drive infrastructure requirements, testing, simulation, emulation. It’s going to drive all of these requirements moving forward.

Serughetti: There are different work streams. There’s a hardware work stream, there’s a software work stream, there’s a development methodology work stream, development tooling — all those things need to be considered. And, of course, there’s the business aspect related to this. We mentioned a marketplace and things like this. There are multiple work streams that must come together and around that. There also are organizational and collaborative aspects that must evolve, as well.

Bahrouch: If, let’s say, the E/E architecture is the right one, and we have all the underlying hardware that’s oversized in terms of computational and performance, with over-the-air update capability, over time the OEM will have the opportunity to update the legacy OS in portions. So every now and then there will be a data update until the point where, hopefully, the OS or the critical software parts are updated and maintained in a way that’s cybersecurity-compliant, performance-compliant, and so forth. It’s a combination of a number of things, but over time we will get there.

Kumar: Going from the bottom-up now, there is a role that IP vendors can play going to Tier Ones, like they do with the OEMs. Everything we are talking about is going top-down, but we also can have some responsibility in pushing that thought up and see what makes sense.

Curran: I agree that some of the heavy lifting and the engineering of a vehicle is done in the supply base. OEMs tend to be expert integrators. That’s what their role is. They’re not necessarily expert, detailed designers, and so taking in more of this broader architecture is a change for them. I agree that working from the bottom up and partnering with the OEM is a great approach, and it will be needed.