• Skip to main content
  • Skip to search
  • Skip to footer
Cadence Home
  • This search text may be transcribed, used, stored, or accessed by our third-party service providers per our Cookie Policy and Privacy Policy.

  1. Blogs
  2. Breakfast Bytes
  3. AImotive: Shifting Gear in Automotive
Paul McLellan
Paul McLellan

Community Member

Blog Activity
Options
  • Subscribe by email
  • More
  • Cancel
Automotive
autonomous driving
ADAS

AImotive: Shifting Gear in Automotive

9 Aug 2019 • 5 minute read

 breakfast bytes logo At the recent Cadence Automotive Design Summit, Laszlo Kishonti, CEO of AImotive, presented From Walled Gardens to Collaboration: Shift in the Automated Driving Industry. AImotive is based in Budapest, and when introducing him, Cadence's Robert Schweiger said that he had his first experience of autonomous driving technology when they took him for a test drive there.

(Quick test: do you know what an OEM is in the automotive industry? If not, then read my earlier post Automotive Industry Basics.)

Laszlo started by pointing out that AImotive is not a classical self-driving company:

We have revenue and customers!

He felt that there had been a paradigm shift in 2018 when there were a few accidents and expectations slowed down. OEMs (remember, that is jargon for car companies) are putting less emphasis on far-out technology (levels 4 and 5). As a result, at CES in January of this year, everyone was announcing level 2+ using just cameras and radar, without high-definition maps or lidar. This has several desirable attributes:

  • Immediate mass-market revenue
  • Compelling product for the consumer
  • Additional safety versus today, will save lives
  • Simpler to regulate

At the same time, for non-technical reasons, the automotive market is flattening or shrinking in 2019 after growth for several years. This has led to restructuring at OEMs and lower R&D budgets, with less funding for costly experiments.

The AV challenge is too complex to be solved by a single company. The diagram above shows how many companies are already funded in some of the relevant subspaces. Clearly there will be a lot of consolidation and, for example, only a handful of lidar companies will survive.

OEMs and Tier-1s Must Collaborate on Modular Software

The short-term downturn in the automotive market and the longer term push-out of automated driving technologies mean that R&D efforts are under pressure. Many in the OEMs have either hit the wall or don't have a route to providing the expected return. However, there continues to be strong competition to deploy increasingly advanced ADAS functions. For the time being, the competition has moved to level 2+.

So the only solution is some level of collaboration between OEMs and tier-1s since almost nobody can do it on their own. For example, Volkswagen and Ford announced in July that they will collaborate in autonomous driving technology. Or in my post MWC Part Dos from what used to be called Mobile World Congress and now is just MWC Barcelona, I covered the Daimler/BMW collaboration. We are all so used to hearing about Daimler-Benz that it is easy to miss that Daimler (Mercedes) and BMW are major competitors of each other in the basic business of selling cars.

With L4/L5 delayed, self-driving companies have no business model. In the short term, they need to monetize the technology that they have developed, which means building more modular software stacks (since nobody has the whole stack). It also allows the same software stack to be scaled differently on different hardware platforms.

Simulation

Airlines are 1000X safer than road traffic due to a major investment in simulation. Pilots are trained by simulation and sometimes even never fly a real airliner until their first flight with passengers.

In automotive we need simulation for continuous integration to validate the software. In particular, the software lasts longer than the hardware and there is increasing importance in validating the software as the industry shifts to a new approach to hardware platforms. As Laszlo put it:

It's not just that you need to have a chip and that's it. You need a workload, real neural network algorithms, not just benchmarks. Some are still using lidar and not production-ready sensors (only cameras and radar are ready). There are parallel teams working on automated driving software, development and simulation, hardware IP. Those three need to be integrated.

Risk Sharing in Hardware Platform Design

Laszlo is keen on sharing the risks, and setting up an "automated-driving hardware consortium", to spread the costs and ensure future upgradability. He feels that this could be genuine competition to current manufacturers, but he doesn't think it is scalable for doing everything.

The above diagrams show two approaches to sharing the risks that Laszlo discussed. The first is the dedicated SoC that he estimates at $100-200M. The second (on the right) he estimates at $20M for just the AI accelerator, with a regular SoC containing a CPU and GPU for the host processor. Laszlo likes the right-hand option:

We should just build accelerator chips like NVIDIA delivers GPUs to the cloud market. That is up to 10 times cheaper than developing a high-performance SoC, also with less strain on the NoC and the memory.

He also pointed out that it is scalable, and can be built in 16nm or easily scaled up 4X and put in 7nm. Potentially this can even use the same pinout for plug-in upgrades.

i'm actually not convinced of this. I think that if you want to build your own silicon then you need to build silicon optimized for just what you require. Even GPUs in data centers are considered too generic compared to something like Google's TPU chips, and other cloud vendors are designing chips of their own. In the limit, if you use the same silicon in a lot of cars then the automotive market will become like the PC market for most of its existence. It really didn't matter if you purchased from Dell, Sony, Toshiba, or Acer...you got pretty much the same product, because it used the same silicon. As a result, it was really hard to differentiate except on price since the performance was all the same. While there are still ICE engines, there is differentiation there, but I've never heard anyone claim a lot of differentiation in electric traction—it seems all electric motors are pretty much the same (although there does seem to be some differentiation in battery management systems).

So his summary:

  • The auto industry is shrinking following growth in past years. The next big challenge for self-driving companies is to survive.
  • To survive they must:
    • Generate revenue in short-term
    • Enter collaborations
    • Deploy modules as part of self-driving solution
    • Rely on simulation for validation and verification
    • Support standardization efforts in automated driving
    • Create scalable hardware platforms to enable increasing automation

 

Sign up for Sunday Brunch, the weekly Breakfast Bytes email.