When we look closely at technologies offered by Mobileye and Intel in autonomous driving, what does each contribute?
The growing relationship between Mobileye and Intel in the highly automated vehicle platforms can be alarming to those who compete in, or follow, the market.
Exhibit A is the strategic partnership between BMW, Mobileye and Intel. Exhibit B is Delphi's disclosure earlier this week that the company is using Intel’s chip on its autonomous driving platform, along with Mobileye’s vision SoC.
After all, Mobileye is already the dominant player in vision processing—a key technology ingredient for ADAS and autonomous cars.
Intel has yet to announce a new automotive SoC featuring multiple Xeon cores, but the prevailing rule of thumb is never underestimate the power of the world’s largest processor company. Indeed, Intel CEO Brian Krzanich announced in his keynote at the LA Auto Show, just before Thanksgiving, a ₹1,666.56 crore ($250 million) investment aimed at making autonomous driving a reality.
When we posted our story, “Can Intel Win Auto Brain Chip Race?,” the knee-jerk reaction from some readers was the accusation that journalists are conditioned to keep giving Intel the benefit of doubt, assuming that it will succeed in the non-PC market.
One of the most astute comments, however, came from Philippe Lambinet, a former STMicroelectronics executive, now CEO of Cogito Instruments. Cogito Instruments is a company working to bring machine learning analytics to industrial applications.
Lambinet sees Mobileye’s true value in its dataset and algorithms. “They developed hardware because there was nothing around that could do the job but fundamentally hardware is not their core value. Intel, on the other side, is THE processor company,” he noted.
Calling the Mobileye-Intel combination “a great functional match,” Lambinet suggests that Intel could pursue building a “Wintel-like duopoly” in the automotive market, just as the processor giant did with Microsoft in the PC industry.
Although Lambinet doesn’t think Mobileye is eager to collude with Intel like the Wintel model of the past, this is an intriguing thought.
I decided to follow up.
Really, when we look closely at technologies offered by Mobileye and Intel in autonomous driving, what does each contribute? And what’s the division of labour between Mobileye’s EyeQ chip and Intel’s yet-to-be-announced automotive SoC?
I got in touch with Daniel Galves, Mobileye’s chief communications officer and senior vice president. Posing the example of the Delphi platform, I asked the question.
Galves told us, “To start, the division of labour is fairly simple. The Mobileye SoC will run all sensor processing software (8-camera surround view by Mobileye, radar/lidar processing by Delphi), localisation mapping by Mobileye REM and sensor fusion will run.”
He added, “On the Intel SoC, all driving policy (reinforcement learning algorithms for path strategy by Mobileye) and driving control (driving behaviour software by Delphi’s Ottomatika) will run.”
Does this also mean that even EyeQ5—whose engineering sample is slated for launch in 2018–isn’t set up to run driving control software? Will EyeQ5 need a companion chip, like the one Intel is working on now?
Mobileye’s Galves said, “The Driving Control software, which we don’t have a big role in, is very computation-intensive. So, it kind of makes sense to run on a separate chip. And makes sense for the path strategy policy that we do, to be very close to the control.” Further, he added that there are some other positives about having two different chips, namely, “functional safety-wise, redundancy-wise.”
In the story we posted, we quoted one analyst speculating that a closer tie between Intel and Mobileye could threat the current ST-Mobileye relationship.
Mobileye’s Galves insisted this is not the case. He noted that ST is “sourced as partner on both EyeQ4 and EyeQ5 in a competitive process.” He explained, “Our algorithms were built to run most efficiently on the jointly developed architecture. Delphi’s Ottomatika software was developed in an Intel environment, so it made sense to deploy on that type of architecture.”
A bigger question is the future.
Going beyond Delphi’s model, what can the industry reasonably expect? What steps will Mobileye and Intel take to profit from the autonomous vehicle platform?
Lambinet predicts a margin war. In his view, the closest thing to a monopoly in automotive is Mobileye. “So it must be immensely attractive to Intel,” Lambinet suspects. If the history is our guide, that’s why Intel wants to get closer to Mobileye. But it is not so clear if Mobileye is so willing to go along.
Intel became Intel because of the Wintel business model. Microsoft and Intel managed to lock up the PC market in a way immensely profitable to both. Since then, Intel keeps trying to reproduce this "protected" model because it can't let its margins decline. Intel failed in mobile and in TV. It might very well fail again in automotive because the PC model is the exception, not the rule.
If Mobileye’s algorithms continue to rule in the autonomous vehicle platform, Intel, as a hardware vendor, “will either buy Mobileye or get commoditised,” Lambinet opined.
It’s hard to imagine, though, that Intel will go that route without putting up a good fight. Intel is now targeting artificial intelligence, armed with Nervana Systems in training neural networks (an area now dominated by Nvidia’s graphics processors) and Movidius, whose acquisition has not yet closed, in computer vision and edge networks.
Well, here we go again. I am giving Intel the benefit of doubt.