Addressing this bit.
In general it can’t be rewritten and recompiled. Those parts that require cycle accurate hard real time control are bound to the target architecture in a fundamental manner. Communication is done via explicit hardware features that cannot be emulated in software and retain the needed guarantees. Indeed there are some fundamental issues in synchronisation that you can prove cannot be solved without a hardware mechanism.
Another issue is caches. You can’t use caches because they break the ability to guarantee cycles. Most automotive processors don’t have caches and often have seperate regions of memory for different purposes, some with different speeds. Memory layout itself is hard coded.
If you really wanted to, you probably could design many components in a car to run off of old iPhone processors. Probably not all of them, but some. But the cost to do so would be huge in many ways, and the design cycle to reasonably do so would be far longer than the expected chip shortage is likely to last.
Interesting. That makes sense.
I’d suspect it’s hard to get an allocation even for standard parts these days. And TSMC would much rather piss off Ford than Apple.
We did low volume stuff, but we were process drivers so we were useful to them.
My understanding is that ARM-Cortex CPUs are pretty standard in automotive applications which is a pretty standard architecture.
I don’t doubt that, but ARM processors are cores that go into SoCs surrounded by custom logic. Car companies are way too smart to design their own processors.
This is correct, but the stuff along with the Cortex core(s) on the SoC matters, and it’s going to be application specific.
Maybe, but the lifecycle cost of designing your own dedicated chips and getting them manufactured is very expensive itself and if not at all agile. Automotive tends to be very slow to adopt new technology due to the extended product development cycles and general culture in management. I don’t know what the price to purchase in bulk recycled smartphone guts are but I bet it would be competitive. Retrofitting isn’t worth the cost, but when they implement whatever the next big automotive feature is the cost of designing around recycled chips and creating their own chip might shift in the other direction.
Is it that different to design your own SoC?
It’s much easier to design an SoC. What you do is to buy Intellectual Property (IP) for the various more or less standard functions you need, like interfaces. ARM makes processor IP - they don’t build their own chips. You hook it all together and add some random logic for additional functions.
I’ve worked on processor design teams in two big companies and processor design is way, way, way harder. And more expensive.
It isn’t the car companies that are the customers. The companies that matter are the likes of Bosch, Denso, AC Delco. They are the dominant three. Car manufactures get most of their electronics from these guys. In terms of size of customer, I suspect pissing off the likes of any of these three versus Apple would see Apple told to take a hike. But I doubt that is the contest, in that I doubt these guys compete for the same fab lines as Apple.
Maybe true, but the reporting in the media seems pretty confident that this is exactly what’s happening. I’m not sure how anyone could trump Apple in any negotiation.
Which means that I suspect chip design is not on the critical path of car development.
A good argument for them.
Hoo boy. Chips fail over time. It’s called a bathtub curve - you have higher failure rates early in the life of a chip, then it plateaus, and then it starts to climb. Burn in, in which you run the chip at high temperature for a time, makes weak chips fail early and moves you to the right of the curve. Recycled chips - assuming you can find them - are way to the right of the curve, and will hit the increasing failure region long before the car makers want.
Plus, I bet you’ve never looked under a modern chip. You can’t pull it out of a socket like in the good old days and reuse it. You’d probably kill a lot of chips pulling them off their boards and kill more trying to reattach them. Our chips were way expensive, and we never, ever tried to reuse returns from the field, even if they were probably good. You might the the module containing the chip, but we found that doing this increased the field failure rate a lot.
True about the fab lines, but the iPhone sold 40.8 million units in the US in 2019, versus 4.7 cars. It was close to 200 million units worldwide. And that’s one company versus the three auto electronic manufacturers.
There is more than one chip in a car, of course, but the problem seems to be not that they all are short, but that some are short, which is enough to mess up the supply chain.
Last news story I read said that the entire auto industry only represents about 10% of global chip volume.
Ouch, I really didn’t think the gap was that wide.
Heh. When I was at Intel a million parts a year was considered low volume. And that was 25 years ago.
Is that mainly from electromigration?
Very interesting and informative thread. Thanks to all for the discussion.
Heat dissipation has been covered as a key differentiating factor in the design of chips intended for automotive use. This makes sense. But what about kinetic factors? I assume that the engine and road will create sustained vibration throughout the vehicle that impacts the lifespan of sensitive components like this. How is this accommodated?
I ask because I used to work for a computer hardware manufacturer producing big rackmounts for a fairly exclusive (and confidential) client list. Occasionally we’d get an order for a cabinet with additional internal isolation. Client information was strictly need-to-know, even internally, but based on the tolerances we liked to speculate it meant the box was going on a high-performance boat or some other physically demanding environment that would be unpleasant for delicate electronics.
So I imagine that if you’ve got a chip near the wheel well to help keep track of tire pressure and brake status, for example, it won’t just need to survive the heat coming off Arizona asphalt, it’ll have to tolerate the constant background buzz inherent in the automotive frame.
Although I’d wager that the BOM (bill of materials) for semiconductors in a modern car outweighs that in an iPhone by at least an order of magnitude.
That’s a big factor. I’m not a reliability guy, but I’ve worked with them.