Are the microchips in cars fundamentally different than the microchips in smartphones and game consoles?

But iPhones are just a small part of demand. We’ve got Android devices, and even for Apple watches, Macs, Apple TV, etc. Not to mention IOT devices, TVs, computers, etc., etc. The 10% mentioned above sounds about right to me given what I’ve read.
And remember, the problem isn’t just the semiconductor companies freezing out the car manufacturers, it is that the car manufacturers gave up wafer starts because they missed on predicting demand.

One more problem with recycled iPhone chips, which I thought of as soon as I went to bed last night. You can’t test them.
Test programs for microprocessors are long and complex, and you need details of designs down to the gate level that Apple is not about to release to anyone, especially anyone trying to reycle their chips. (I bet there are contract issues also.) That means you will not have a good sense of whether the chip you have is good or not. Not something the car companies are going to put into a product which is life critical.
How about GM and Ford buying new chips? Nope. Apple would be constrained by the car companies in making design changes, and the famously secretive Apple is not about to reveal the secrets of the next generation chip to car companies. (Apple employees are not even supposed to ask questions at conferences for fear of revealing something.)
I’d guess that auto chips are traceable, in that they have some sort of embedded ID which will let you know where and when failures occurred. I bet if Apple chips have this, the information wouldn’t be available, and certainly not if the chip is recycled.
This is vitally important in high reliability applications since it lets you diagnose failures much better. I could see the test results of a chip that failed in system or in the field, when and where it was made, and if its neighbors on a wafer passed or failed.

End market demand for semiconductor chips is generally broken down as follows:

Cell phones, telecommunications - 33%
Computers - 28%
Consumer electronics IOT - 13%
Auto - 12%
Industrial automations - 12%
Government - 2%

I would have thought that automotive chips are more like microcontrollers than processors for laptops/desktops. The difference being the former is focused on reading/setting voltages on individual pins to turn things on/off, whereas the latter are more concerned with reading/writing data words to disk, RAM or video RAM. If true then I don’t think it would be easy to swap one for the other.

Some auto manufacturers do design their own chips. Tesla used to rely on Nvidia but have now have their own design for the specialised AI functions required for self driving capability. This approach is similar to Apple with their new M1 series of computers. Google do the same for their systems. This approach protects against the periodic shortages of very specialised components and gives them a lot of leverage with fabs.

Nonetheless Tesla is still dependent of regular automotive grade chips for many other systems required for car manufacturing. They seem to be have been able to manage this because their business plan did not anticipate a fall in demand for their products like all the other car manufacturers and so got the supply contracts and logistics in place to source the components they needed.

Chips shortages are not a new phenomenon, they happen from time to time in the normal course of business. Sometimes there are problems at the fab plants, especially when they transition to newer technology and there may be very few alternative sources of supply. Sometimes they can be dealt with by changing the design to accomodate the use of alternative components.

When I worked in the Oil and Gas business building oil rigs, they had specialist buyers whose job it was to secure the supply of essential components to make sure any shortage did not cause an interruption in the business of the getting the stuff out of the ground. Huge valves made by only one or two companies. These guys hold suppliers to very exacting contracts. They pay top dollar, but if a supplier fails to deliver they pile on the pressure, unlike any normal customer for of a commodity product. I had a problem with some faulty IT equipment and set these guys on the case. I don’t think our suppliers really knew what hit them when the ‘Expediters’ came to call. So there are well established ways of mitigating business risk, no matter how specialised the components.

However, the car manufacturing business is going through a very fundamental change as it embraces EVs and the development of new advanced electronic systems in cars. Many clearly do not have the systems in place to deal with semi-conductor shortages. I expect they will get their act together in time.

This is a business problem more than it is a technology problem, I think.

The M- and S-rated (aerospace grade, compliant with MIL-STD-883) microprocessors often take five years or more to develop and qualify, and they are used in such low volume (in comparison to consumer-grade microprocessors in terrestrial applications where charged particle radiation and extreme temperature transients are not a significant issue) that they often remain in new service for two decades or more. The slower processing speed means that there is more effort in making efficient firmware (for the most part, these processors are running applications on a very thin RTOS), and of course they aren’t running in graphics-intensive displays because there isn’t any need for them. When the STS “Shuttle” started running experiments requiring complex user interfaces, NASA qualified and sent up commercial hardened laptop computers for astronauts to use, but these were not in any way tied into the Orbiter Vehicle avionics suite which even at end-of-life was using early 'Nineties microprocessor technology for the ‘glass cockpit’ upgrades.

There are smallsat applications by amateurs and startup companies that use ARM-microprocessors intended for mobile devices and sometimes actually incorporating cellphones directly into their avionics so they can use the camera and accelerometer on the phone in their instrument suite. This works fine for a sat that is only intended to function for a few months or less and isn’t performing any critical business function, but as smallsat startups have discovered, isn’t really suitable for high reliability applications. Amusingly, a couple of the early smallsat manufacturers who were so gung-ho about using all commerical-off-the-shelf processors have now gone to chips that, while not fully S-rated, are tested to some subset of MIL-STD-883, because when the microprocessor running your software defined radio has a repeated single event upset every time is passes anywhere near the Southern Atlantic Anomaly, it causes major problems.

Microprocessors for automotive use don’t have to worry about charged particle radiation, of course, but they are often in performance- and safety-critical applications such as vehicle dynamic control, advanced ABS deployment, and engine management, are in a high EMI environment, experience some pretty significant thermal environments, and are generally expected to function for the life of the vehicle without replacement, which for the typical car is three or four times the lifespan of a smartphone. Given the costs of a field service campaign, insisting on “zero defects” is a financially-astute business call even if it does mean more intensive testing and rejection of functional but potentially marginal units.

The other issue, of course, is getting that steady stream of “old iPhone processors” to maintain the logistical needs of high volume production. Aside from the reliability issues of extracting and reusing chips from old phones, what happens when Apple shifts from an A9 to an A10, and then a year later to an A12? Is a manufacturer then going to qualify the new chips and refactor all of their code for the new chipset at a cost of hundreds of millions of dollars on top of other design costs? There is a reason that auto manufacturers keep using the same chipsets and try to reuse software from vehicle to vehicle; because the costs are huge, and the risks of having to rewrite firmware and missing a fatal flaw are enormous, notwithstanding that processors designed to run real time embedded firmware are very different than a processor purpose built to run a graphical user interface and communication device.

This is similar to the belief among the space enthusiast crowd that it should be possible to just cobble together a bunch of salvaged aerospace parts along with COTS hardware and fly a mission to Mars or Jupiter for just a few billion dollars instead of building and qualifying a purpose-designed vehicle to modern aerospace standards like AIAA–S-081 and SMC-S-016. As someone who has actually worked in propulsion system reuse and repurposing, I can say with some authority that it is nowhere near that easy to take old (but highly reliable) hardware and put it into a new application. Doing so does get the previously demonstrated reliability (provided you can address issues with age degradation and obsolescence) and the savings from having to set up manufacturing for new systems, but there is a hell of a lot of effort in making that work and interface with the other newer avionics, power, flight destruct, and other systems required for repurposing, and it mostly makes sense because we fly so few space launch vehicles.

The notion of building a high volume modern automobile design around salvaged microprocessors to obtain some theoretical savings is patently absurd. This is not to say that we shouldn’t make vehicles to be sustainable, with components that can be easily recycled, but they aren’t Lego kits that can just be torn down and reused as needed.

Stranger

Sometimes there are literal tsunamis that wipe out part of a supply chain. Sometime plants burn down. Some companies learn from these past mishaps.

https://www.france24.com/en/live-news/20210512-lessons-from-2011-disaster-help-toyota-ride-out-chip-shortage

And it’s not just chips. Seemingly mundane items like capacitors can be in short supply due to the requirements of automotive components. Non-automotive grade capacitors are prone to leakage when they sit in a hot car for prolonged periods of time. Not just voltage leakage, but physical substances oozing out of them.

Components?

Ah yes, who can forget the ‘Great capacitor plague’ that caused so many problems for electronics manufacturers. A cautious tale of intrigue and cover ups and dogy dealing in the capacitor business. It shortened the life of many a computer system.

Manufacturers of chips for the automotive industry are generally required to maintain a failure analysis team ready to go into action as soon as a failed part is returned from the field. Manufacturers want a report on the root cause of the failure within 24 hours of receipt of the failed part and a plan to prevent similar failures within a very short period after that. Generally the FA engineers took turns being “on-call”, and anyone else (like the chip designers) could be pulled off of whatever new project they were working on to address the problem.

I have designed chips that were automotive qualified and other ships that were space qualified. Space qualification is easier. And field returns are pretty rare…

What that means in practice is that the chip manufacturer will do burn in and test to identify the defective chips before shipping them. Their manufacturing process isn’t anything special in all likelihood- not compared to modern day PC CPUs or GPUs anyway. Nobody’s making 3nm automotive MCUs I guarantee. :slight_smile:

Cars these days use automotive-specific embedded microcontrollers- their main unique features are their long-term reliability and their optimization for use in cars. They’re not particularly fast, don’t have particularly rigorous memory capabilities, or anything like that. They don’t need to; most of what they do isn’t calculating stuff, but rather taking in sensor inputs, looking up results in lookup tables based on those inputs, and then taking the lookup table results and sending those values to the various engine components, to something working at 12,000 RPM. But they have to do it all the time, in all conditions for 10-15 years reliably.

There’s no reason to adapt modern-day general purpose CPUs for automotive tasks- they’re not optimized for it, they have specific cooling and power needs that are probably an issue in an underhood environment, and honestly, are probably quite a bit more expensive than the automotive specific embedded microcontrollers that are used.

Its not just under the hood that is environmently hostile, there are ECU’s (electronic control units) in the interior of the car that are subject to abuse, like a cars audio amplifier under a front seat. Little kids spill things on them, dry their shoes on them, spill drinks on them and spew literal crap on them.

There are camera ECU’s under your console, occupant detection sensors in you seats and airbag controllers under you floorboard. People treat cars like crap and still expect them to operate without fail, which they do 99% of the time. When something does fail they scream warranty return despite whatever they may have done. Phones are pussys compared to cars.

My company made both processors and high reliability servers, and we had major FA also. Our failed parts got retested on an IC tester, and also stuck into a system and had system test run. I have a few papers on analyzing No Trouble Found returns - parts that get returned and pass all tests.

Not yet, but the day is going to come. There has been some work done on dealing with bit flips in flip-flops. Memories already have ECCs, and these type of defects happen.
We recorded the altitude of installed systems, and in one case where we were getting lots of failures we saw a correlation with altitude. This wasn’t a processor problem, but it could happen some day.
So, drive your car up Pike’s Peak and all bets are off.
As for high reliability parts, I was wondering about the process node used. Larger feature sizes are less likely to fail due to defects. That was one of the reasons my colleagues used relatively primitive nodes. They had their own fab, and not wanting to upgrade was another.

Let me see what I can do…

Stranger

It wasn’t just that: they also rapidly adapted their systems to alternative suppliers:

In Q1, we were able to navigate through global chip supply shortage issues in part by pivoting extremely quickly to new microcontrollers, while simultaneously developing firmware for new chips made by new suppliers

Note that this wasn’t switching to phone chips or anything like that. Just that there were alternative suppliers of automotive-grade chips (microcontrollers in particular) which did not have the supply problems, and Tesla quickly rewrote their firmware in key areas so that they could use these chips.

Manufacturers which couldn’t adapt their software so quickly (due to institutional inertia or because they were dependent on third parties) weren’t stuck because no suitable chips were available; they were stuck because the exact chips they needed weren’t available.

I can see how switching to a new microcontroller wouldn’t be that hard.
Back during the '90s boom we had multiple sources for ASICs, but I suspect this isn’t very common anymore as the number of fabs (and ASIC suppliers) has gone down. You can clearly get a better deal single source, since your volume will be greater. Also, the supply chain is easier and you don’t have to qualify, and keep track of, two vendors.

It really depends on the software. Good code tends to be portable code–so if the new microcontroller is a functional superset of the old one, and has a good toolchain, then it could just be a matter of fixing up the pinouts, interrupt handlers, etc. and making sure the timing constraints and such are sufficient. It’s not going to be a drop-in replacement, but most of the code could be reused.

But there’s likely a ton of completely non-portable code out there, if not at Tesla then at other places. Some might be using 100% assembly, in which case it’ll likely need to be rewritten from scratch. Maybe not so bad for very simple things like window controllers, but more work either way.

Second sources is a slightly different thing; those are intended to be fully drop-in placements (in my observation). The 8086 for instance had a zillion pin-compatible sources aside from Intel. AMD’s entire early existence was as a second source for Fairchild, Intel, etc.

There are some Chinese clone chips that are pin-compatible with American ones, like fake FTDI serial-USB converters. But I doubt any legit manufacturer would use these intentionally.

I got slammed on the internet for writing that we preferred to have FTDI drivers disable fake FTDI adapters when identified. Because our experience with fake chips is nasty enough that we’d always discard anything identified as fake anyway.

Well, I got burned in the other direction. Bought some otherwise perfectly legit Arduino clones, and then one day I got an invisible Windows Update that installed the infamous FTDI drivers that bricked the chips. Took me hours to even figure out what was wrong. Eventually managed to unbrick them. I’d have happily paid $1 extra or whatever for legit chips, but unfortunately there’s a race to the bottom and everyone comes out a loser.

Fortunately, all that mess is largely over. The Arduino clones now use CH340 chips, which are cheaper and use different drivers. I’m not sure FTDI did themselves any favors in the long run.

It has long been a strategy for a suppliers in the computer business to encourage their customers to depend on them preference to alternatives. They like to lock-in their customers. On the other hand, when that supplier fails to deliver, it is in the customers interest to be able to use of second source suppliers quickly and easily. All the big systems companies try to do this and large, customers with slow, predictable businesses cycles find it convenient, during normal times.

Some customers are very savvy and have use designs that mitigate this risk with second sources.

There is something like that going on car manufacturing. Newcomers like Tesla have an advantage in that they can use the latest software development methods that can quickly change a design to use an different component, by changing the software. I guess the lot of the older established manufacturers cannot do this and are more exposed.

The electronic systems in car manufacturing are going through some dramatic changes.

Here is Munroe and some veteran auto engineers comparing trends in car electonics design.

As a car salesperson, a few observations from learning about this issue as it impacts my way of making a living greatly. A few things I didn’t see mentioned, so sorry if I repeat:

These foundries are basically mercenary companies that make chips for about every industry that needs them. Intel used to make all it’s own chips, but no longer, for example.

The foundries never closed down due to the pandemic because they already operate in full PPE and 500,000 square foot clean rooms. Plus, the consumer electronics sector EXPLODED during the pandemic as tablets and laptops for Zoom meets were ordered en masse, as were gaming consoles as both the PS5 and Xbobx Series X were debuting during all of this. Smartphones too.

The foundries make more money on a per unit basis from the consumer electronics sector than they do the auto industry, which seems surprising given the rigors of testing these particular chips undergo to accomplish 0% fail rates.

Making these chips apparently consumes a LOT of water on a per day basis. TSMC, the largest chip manufacturer on the planet, uses the equivalent of 60 olympic-sized swimming pools of water per day.

Half of the global supply of these chips are made in Taiwan (SURPRISE Trump supporters, your automatic reversion to the “Must be China, hurr hurr” attitude towards this is wrong, this time). Taiwan is suffering through a century-level drought due to climate change. This is a problem, as is non-stop and continuous harrassment from China, because China thinks Taiwan belongs to it. Thank you, US Navy!

Everywhere these things are made it seems there’s been a problem. Drought in Taiwan. A fire at a huge foundry in Japan that Ford gets 70% of it’s chips from that took it offline long term. Ripple effects.
4 foundries in Austin, TX went offline for at least a week (ripple effects) when the natural gas pipes froze last year.

The lead time for these chip orders is astoundingly long from the development of the reticules they are sent to the foundries with to production and everything else.

It’s true: the auto makers more or less lost their place in line, and the chip makers aren’t in any particular hurry to please them given the profitability of making them for others. It really sucks for me.

We normally have about 130-140 new cars on our Chevrolet lot, and we are a small store in quasi-rural Indiana. Right now we have 6. I cannot sell what I do not have.