Advance and Retract

One of the important things said during the Tesla Autonomy Investors Day this past April didn’t get a whole lot of attention, given that when the guy running the company is so expressive in proclamations that anything short of shouting out something like “Free cars for everyone!” would go overlooked.

The statement came from Pete Bannon, the vice president of Silicon Engineering at Tesla, who came to the company from Apple in early 2016. When introducing Bannon, Musk called him “the best chip and system architect I know in the world.” Which isn’t bad when your boss describes you as being the best in the world at whatever it is that you do.

Bannon said that when he started the project that was to lead to what is referred to as the “full self-driving computer” (FSDC) chip (which has two processors on one board), he asked Musk if he was willing to invest “all the money it takes for full custom system design” the response was yes. I wonder how many industry execs would give the go signal without an exhaustive anticipatory accounting.

Bannon said the timespan from the start of hiring team members to full production of the FSDC for all three of the Tesla models was just over three years. Realize that this system is not just something that is going to be running a Word program or a spreadsheet: as this is for an advanced driver assistance system (ADAS)—although Musk makes claims about self-driving robotaxis coming to a street near you sooner rather than later, which is probably more heat than light—there are absolute safety considerations that had to be taken into account.

And here’s the key line that Bannon said, referring to the speed of the project: “It speaks a lot of the advantages of having a tremendous amount of vertical integration to allow you to do concurrent engineering.”

Vertical integration. Concurrent engineering.

It seems that the former is something that OEMs once figured they weren’t all that keen on. Remember back in the mid-90s when General Motors spun off its components operation which became known as “Delphi” (which has subsequently spun off its autonomous driving group, which has become Aptiv)? Then in 2000, Ford decided that that seemed like a good idea, and even though Henry Ford completely transformed transportation the world over as a result of his comprehensive vertical integration, Visteon was created in 2000.

The objective, it seems, was to minimize legacy costs and to have component businesses capable of selling to other OEMs, thereby increasing profitability. Arguably, it was about concentrating on one’s “core competencies,” which essentially came down to controlling powertrain and assembly operations, period.

But Bannon also noted that during the development of the FSDC they were “totally focused exclusively on Tesla requirements, which made life much easier by having one customer.” The whole notion of having multiple customers may result in reduced costs as a result of scale, but it also means that it is possible that meeting specific requirements isn’t as achievable (e.g., the FSDC is designed so that it can be retrofitted into existing Tesla vehicles).

Perhaps the pendulum is swinging back. It is interesting how, for example, Cruise, the autonomous tech startup, was acquired by General Motors in 2016 and that in 2017 Cruise decided that it would be advantageous to have a lidar manufacturing capability so it bought a company called Strobe. Ford has invested in Argo AI, as well as companies including lidar market leader Velodyne. It is almost as though we are getting to some sort of incremental vertical integration.

Then there is the issue of concurrent engineering.

This is the thing that surprises—and troubles—me the most. Recently I’ve talked to more than a few designers and engineers who have mentioned that their vehicle programs were realized in a good way thanks to the fact that they worked with their colleagues from other functions during the development, as though this was a novel approach.

But here’s the part that’s disturbing. I went through some of the archives of this magazine, back to when it didn’t have this title but was simply Production magazine. And the July 1987 cover story of that publication was about simultaneous engineering, which is synonymous with concurrent engineering. It doesn’t take the processing power of an FSDC to realize that that was a long time ago. Yet here we have it talked about as though it is something that isn’t a matter of course but a competitive advantage. Yes, it is a competitive advantage, but after 30 years isn’t it not a “best practice” but just, well, a “practice”?

As the industry advances, perhaps it also needs to retract.