In Automotive, A Move From Microcontrollers To Massively Complex SoCs

Cars and custom, high-end chips. It’s a topic coming up more frequently these days. The most prominent example is Tesla’s FSD computer, described by Elon Musk as “the best chip in the world…objectively” during the company’s April Autonomy Day. When it comes to chips, Tesla is alone only when it comes to hyperbole, at least based on browsing job postings for big carmakers and suppliers, also for well-funded full-stack automated vehicle companies and of course, for tech giants like Amazon with not-too-subtle transportation ambitions. Which is to say, everyone in automotive appears to be at least interested in custom SoC design.

“[I]t will become increasingly difficult to rely on general-purpose SoCs to handle the massive amounts of data and real-time processing requirements of automated vehicles,” writes Jim McGregor, in a May 2019 research brief. “Just as the battery and electro-mechanical drives will have to optimized for each vehicle, so too will SoCs have to be customized for specific vehicles to produce the optimal system solution.”

McGregor, principal analyst with Tirias Research, appeared with Siemens’ David Fritz at the AutoSens conference last month in Detroit to talk about just what’s going to be required to design and validate all these bespoke chips. Fritz, who works on strategic alliances and is also a Qualcomm and Nvidia alum, announced the Siemens PAVE360 (pre-silicon autonomous validation environment) program in a main-stage keynote, then later joined McGregor for a discussion on the topic with the Autonocast podcast crew. Autonocaster Kirsten Korosec, also a senior reporter for TechCrunch, called it “one of my fave episodes to date.”

A day earlier Fritz spoke to an AutoSens interviewer, Carl Anthony, about the silicon-related demands of automated driving and consolidated ADAS systems. An edited version of this interview is offered below.

You spend a lot of time talking to executives at big car makers and suppliers. What keeps them up at night?

Generally, what I see is that it’s the rate of change coming towards their industry. And handling that rate of change is what concerns them the most. When they see companies like Intel, Nvidia, Samsung, Qualcomm, and others encroaching on their space with technologies that are nothing like what the automotive industry has really seen before, that’s very disconcerting. Then when you throw Tesla on top of that and what it’s doing with its Full Self-Driving (FSD) chip, they have a lot to be concerned about.

Still, why would these car makers and suppliers want to get into the business of creating chips, sensors and related technologies? It seems, at first glance that, different than their core business models.

The most important reason is they’ve looked back at other industries – inkjet printers, digital TVs or smartphones – and seen what the end results were in these kinds of transitions. Those companies that delay the ability to do a bit more vertical integration, which includes doing their own silicon and having more control over the software, fall behind in addressing new markets and finding new ways of generating revenue.

Are existing chip architectures sufficient for autonomous driving and ADAS applications today?

Semiconductor technology for automotive applications has in general been very simplistic – think microcontrollers driving actuators and relatively simple processors controlling the sensing of information. That’s no longer sufficient when you start talking about high-end, consolidated ADAS, let alone eventual Level 4 and Level 5 autonomous vehicles. The magnitude and value of the new solutions in these domains is very different. Indeed this is part of the concern from the automotive industry as it looks at the technology owned by the big chip companies that are encroaching on the automotive space.

Talk about your announcement of PAVE360. What is it?

Just think about it like this – imagine you had a very accurate virtual representation of a car driving itself in a virtual world that can be controlled by engineers. You can control the scenarios – first a cow stepping out in the road, then convert that cow into a cyclist, next change the cyclist’s clothes. All those things can happen automatically, sweeping through all those possibilities using validation technology that’s grown up in the semiconductor industry for the last couple of decades.

How does PAVE360 fit into the long-term strategy at Siemens?

PAVE360 is a combination of many years of work and couldn’t actually exist without all of Siemens recent software acquisitions – 11 billion euros worth in the last 10 years. Many of those acquisitions come together to produce PAVE360. Siemens has this concept of a digital twin, or more specifically, of a common digital thread stringing together different digital twins of varying fidelities. PAVE360 is a concrete example of applying that technology across multiple products to solve much, much larger problems than otherwise could be solved with just single-point solutions.

What are some of the biggest challenges facing autonomous driving today?

I think for the industry as a whole, it’s understanding that, like happened in other disrupted industries, there’s going to be an inevitable and relatively rapid consolidation of functionality. It seems very clear that the center of that functionality is going to be high-end ADAS and AV systems that have massive computational capability. Over time, these systems are going to commoditize. You’re going to have companies producing ever larger SoCs that incorporate functionality that today is in ECUs. I mean, it happened in smartphones. If you crack open your iPhone – and please don’t do it, just trust me – you’ll see one large chip in there and a few small ones. Well, a decade ago, that would have been a dozen chips or more. Consolidation changes who the major players are, and if you want to control your own destiny as a supplier or an OEM, you need to be involved and start now.

What’s your take on the evolution of chip design and verification technology at Siemens?

When I first joined Mentor Graphics and it was acquired by Siemens a couple of years ago, I was very excited about that. The whole concept of quality German engineering, that really applies here. Just a quick anecdote – my son hurt his finger a few months ago, and I got a phone call from my wife to run to the emergency room. And what do I see when I get there? I see Siemens MRI machines, Siemens x-ray equipment, Siemens everywhere. Later I thought – just like in healthcare, what matters in the automotive industry is safety, quality engineering, verification and validation, and everything that goes with that, and so Siemens is a good choice in automotive too. So I’m really excited about Siemens investing in this technology, seeing the potential, and using its clout, contacts and its history to actually help us promote PAVE360 into the industry.

Anything else?

Well, we have something we call the Center for Practical Autonomy. It’s a lab in Novi, close to Detroit, where we have an installation of PAVE360 on display. We’re inviting anyone who is interested to come take a look. We walk through using the platform, including how to change the virtual SOC structure, the number of cores and cache sizes. The results are empirical measurements that can help OEMs, Tier 1s and Tier 2s get their engineering jobs done, particularly when it comes to chips for ADAS and autonomous vehicles.

You can watch the full interview here:

For more information, checkout the Tech Talk “Building An Efficient Inferencing Engine In A Car” between Semiconductor Engineering editor in chief, Ed Sperling, and David Fritz.