Auto Industry Relationships Re-Form, But Differently

The automotive industry is in the midst of rapid change on many fronts. OEMs are exploring new functions and features to add to their vehicles, including chiplets, electrification, autonomous features, as well as new vehicle architectures that will determine how vehicles are going to be designed from the foundation up. But all of this is dependent on the relationships between all of the ecosystem players involved. Semiconductor Engineering sat down with David Fritz, vice president of hybrid and virtual systems at Siemens EDA; Kamal Khouri, vice president and general manager for automotive at GlobalFoundries; and Frank Schirrmeister, vice president of solutions and business development at Arteris IP. What follows are excerpts of that discussion, which was held in front of a live audience at the recent AutoSens Conference in Detroit. To read part one of this discussion, click here . Part three is here.

SE: Tier 1s are now pairing up closely to the OEMs, creating what some are calling a Tier 0.5 and sometimes even spinning them out into new companies, with Foxconn as an example. What is the result of these adjustments in the automotive ecosystem?

Fritz: The biggest thing this solves is the whole concept of black-boxing, which engineers are familiar with. Generally, what happens in a traditional way of building a vehicle, everybody puts everything together into a big [hardware-in-the-loop] system. Everything plugs in, and then nothing works. Then you’ve got to figure out whose problem it is, and there’s a lot of finger pointing. But with this new changing of the ecosystem, you can have partners who have visibility into the behavior of your IP without jeopardizing your IP actually being stolen. So it starts to eliminate that whole integration storm that happens at the end, just before SLP. Chiplets play a role in that because if you now have a chiplet you’ve consolidated half a dozen or more ECUs, so you’ve got fewer vendors that you need to work with. You have a monolithic SoC but inside there’s a piece from Company A, a piece from Company B, and a piece from Company C. In the new tooling environments that are out there now you can see what all of those things are doing, and how they are interacting. You can look at it, log in remotely, see everything happening. That allows this continuous integration to happen, and that’s what resolves integration problems at the end. That’s what shortens time to market. And that’s what you have to do to address the additional complexity in the hardware and software sitting on top of that.

Schirrmeister: The OEMs are saying it’s just physics, that they cannot do the things they need to do five years from now, and that’s where the timelines come in. [Level 5 autonomy] is not happening necessarily in 2024. That’s why we have these longer timelines. They cannot do from an OEM perspective what they need to do, and it comes down to physics. To things like power consumption, they can’t afford the different technologies that need to be integrated. Certain things need to be at the most aggressive technology from a compute perspective. Think data center on wheels. In other cases, the sensors in 3nm may not be the right technology to have for that. For that type of integration, chiplets are the clear answer to that problem. You need this heterogeneous integration of all those components. That’s really the only way to get beyond these incredibly complex SoCs.

Khouri: Data center on wheels? Absolutely. Mobile phone on wheels? Absolutely, too. And this device also has to get your family and your friends from Point A to Point B and be safe and reliable. It’s not just the user interface. We talk about the user experience in the car that everyone expects. The OEMs want to give you that user experience. Today we’re at a sensing conference. It’s all about safety, autonomy. But there’s also a user experience perspective that the OEMs want to do. You want to walk up to your car, you want the car to recognize you, all your settings, your favorite radio stations, your seat settings. These are also important to the OEMs as features, and so it comes down to the software and controlling the software. The OEMs are realizing that if they want to control the software, that software runs on hardware. So they’d better understand the hardware that’s running their software, and optimize it for power, because power is a very scarce commodity in the world of EVs.

SE: In terms of the evolution of with the OEMs toward new vehicle architectures, what does that look like today and how can we improve that within the ecosystem?

Fritz: They need to explore more. At one German OEM, one of my friends there called me up and said, ‘We decided we want to go build a car that could outperform a Tesla. Our guys think we can do it just by slapping together a bunch of existing ECUs. We know it won’t work, but I can’t prove it. How do I prove it?’ So he delivered a product that could prove it before they did it. A lot of it is that we have enough understanding now about the behavior of the software stack — everything from hypervisors running multiple operating systems to hardware that’s needed underneath that — to allow you to start putting together these hypothetical configurations. There is another OEM that said, ‘We want to go with Automotive Ethernet from the sensors to a central compute all the way out to the actuators.’ So we set out to say, ‘We’re going to model that environment for you and prove that as long as you have 100 Gigabit Ethernet, you can do that.’ They said, ‘We don’t have that. We have 5 Gigabit.’ We asked what their alternatives were. Then it adjusted the software, and adjusted the hardware across the whole system. So when we’re saying OEMs need to control their own destiny, that’s where that happens. Not the Tier Ones underneath that saying, ‘I just want to give you one ECU to do this one particular job, and then I want to sell it to as many customers as I can.’ Nothing against that, but the motivations are very different.

Schirrmeister: From the integration perspective, that’s one of the tool infrastructures in which you can do these architecture analyses. The interesting bit that is challenging is we have never figured out how to resolve the architect’s dilemma, which is to deliver models of all the participants early, at the right fidelity level, so all those decisions can be made. There’s always a circular process going on. We are getting much better in modeling more of that, and doing the architectural analysis to make it happen. As an ecosystem, we need to work together. And as a sensor provider, you will be asked to provide a model of the sensor that can be put together with, say, our network-on-chip IP. Then comes the process of doing the processing, running in EDA tools to confirm that you actually do not blow up the device from a power perspective by doing this or that. That requires a sufficient number of corner cases. The tools and modeling perspective will be very important to make this work.