Chipmakers To Carmakers: Time To Get Out of the Semiconductor Stone Age (fortune.com) 416
Long-time Slashdot reader BoredStiff shares this report from Fortune:
Moore's law of ever-increasing miniaturization seemingly never reached the automotive industry. Dozens of chips found in everything from electronic brake systems to airbag control units tend to rely on obsolete technology often well over a decade old. These employ comparatively simple transistors that can be anywhere from 45 nanometers to as much as 90 nanometers in size, far too large — and too primitive — to be suitable for today's smartphones.
When the pandemic hit, replacement demand for big-ticket items like new cars was pushed back while sales of all kinds of home devices soared. When the car market roared back months later, chipmakers had already reallocated their capacity. Now these processors are in short supply, and chipmakers are telling car companies to wake up and finally join the 2010s. "I'll make them as many Intel 16 [nanometer] chips as they want," Intel chief executive Pat Gelsinger told Fortune last week during his visit to an auto industry trade show in Germany.
Carmakers have bombarded him with requests to invest in brand-new production capacity for semiconductors featuring designs that, at best, were state of the art when the first Apple iPhone launched. "It just makes no economic or strategic sense," said Gelsinger, who came to the auto show to convince carmakers they need to let go of the distant past. "Rather than spending billions on new 'old' fabs, let's spend millions to help migrate designs to modern ones...."
Reliability plays a major concern. Most systems in cars are safety-critical and need to perform in practically every situation regardless of temperature, humidity, vibrations, and even minor road debris. With so much at stake, tried and true is better than new and improved....
If semiconductor suppliers like Intel and Qualcomm have their way, however, the days of the auto industry relying on these cheap commodity chips are numbered.
The article cites a prediction that 10% of pre-pandemic car production could be eliminated due to chip shortages — and includes this quote from a press briefing by the Volkswagen Group's head of procurement.
"Because of a 50-cent chip, we are unable to build a car that sells for $50,000."
When the pandemic hit, replacement demand for big-ticket items like new cars was pushed back while sales of all kinds of home devices soared. When the car market roared back months later, chipmakers had already reallocated their capacity. Now these processors are in short supply, and chipmakers are telling car companies to wake up and finally join the 2010s. "I'll make them as many Intel 16 [nanometer] chips as they want," Intel chief executive Pat Gelsinger told Fortune last week during his visit to an auto industry trade show in Germany.
Carmakers have bombarded him with requests to invest in brand-new production capacity for semiconductors featuring designs that, at best, were state of the art when the first Apple iPhone launched. "It just makes no economic or strategic sense," said Gelsinger, who came to the auto show to convince carmakers they need to let go of the distant past. "Rather than spending billions on new 'old' fabs, let's spend millions to help migrate designs to modern ones...."
Reliability plays a major concern. Most systems in cars are safety-critical and need to perform in practically every situation regardless of temperature, humidity, vibrations, and even minor road debris. With so much at stake, tried and true is better than new and improved....
If semiconductor suppliers like Intel and Qualcomm have their way, however, the days of the auto industry relying on these cheap commodity chips are numbered.
The article cites a prediction that 10% of pre-pandemic car production could be eliminated due to chip shortages — and includes this quote from a press briefing by the Volkswagen Group's head of procurement.
"Because of a 50-cent chip, we are unable to build a car that sells for $50,000."
REALLY? (Score:4, Insightful)
Re: (Score:2)
You are telling us to abandon tested and proven designs?
Don't forget migration to the kind of plants that are in much greater demand today from many other industries. That's definitely a good idea.
Re:REALLY? (Score:5, Informative)
That's fine -- as long as the parts from those newer production plants have comparable lifetime and reliability, particularly for automotive environments. With smaller lithography often comes lower reliability. If they switch from a creaky old EEPROM or SLC flash chip to QLC, capacity goes up but durability and retention time go down.
Re:REALLY? (Score:5, Informative)
It's not the smaller lithography, but pushing the envelope on other fronts that is the tradeoff between reliability and capability. In a lot of embedded work you have very conservative EEPROM made with modern tech with extremely low capacity but still modern. As well as maybe some high capacity TLC NAND for cost effective capacity for less critically important data.
Also, a lot of the 'reliability' in EEPROM is really self-imposed limitations on how it is used. When you are doing software development in that context, you avoid writing to it like the plague. A lot of designs with those reliable 'creaky old EEPROMs' have screwed up by writing to them and wearing them out.
Re:REALLY? (Score:4, Insightful)
It's not ONLY the smaller lithography, but feature size absolutely plays a role. Smaller elements and lower voltages mean it's easier for dopants to migrate enough to change the behavior, and easier for high-energy particles to cause an upset event.
It's possible to design chips to compensate for those effects, but what is the cost-benefit trade compared to using simpler, more standard parts wmade with older processes? Does it make sense for the chip manufacturers to build new, more rugged chips, and if so, are they doing that? It seems like this article should talk about that, but instead of just repeats talking points from Intel.
Re:REALLY? (Score:4, Insightful)
but what is the cost-benefit trade compared to using simpler, more standard parts made with older processes?
That was Intel's point. That to use 'simpler parts made with older processes' sounds nice and simple but represents billions of dollars the auto industry will never pay for.
If this was a bit more capitalist, it would sort itself out. Either some chip fab would see a viable business opportunity and meet the request of the automakers or the automakers end up recognizing that promises of maybe millions of dollars of business isn't going to make any supplier spend billions. However, the government is stepping in applying pressure and potentially subsidies to 'fix' it, and so it warrants more careful consideration.
Or maybe it's being melodramatic about a temporary shortage on the lower end fabs that will fix itself over time.
Re: (Score:3)
Intel claims it would cost billions of dollars, but they represent an industry that just seriously screwed up its pandemic planning. Intel has also been claiming that their 7 nm technology is almost ready for the last half decade. I don't know why you are so credulous when it comes to their claims.
Re:REALLY? (Score:5, Insightful)
They can do that, but it will involve engineering work to happen. You have headroom to work around potential issues with more redundancy in the same package form factor then you had originally. But again, someone has to do the chip design and other engineering to make it happen.
The auto maker's stance is they don't want to pay for that. But they also don't want to pay for fabs to be built either, but they are trying to make everyone think that it's the chipmaker's duty to build the fabs anyway. It's harder to pitch that someone external needs to do new engineering work for them than it is that someone else just 'does the same thing they were already doing, but more.'
Re:REALLY? (Score:4, Informative)
Also, a lot of the 'reliability' in EEPROM is really self-imposed limitations on how it is used. When you are doing software development in that context, you avoid writing to it like the plague. A lot of designs with those reliable 'creaky old EEPROMs' have screwed up by writing to them and wearing them out.
Writing the chip over and over again does put wear on the device. However, avoiding writes entirely is also a pathway to failure.
What you really want for high reliability is software that:
#1. Makes use of Error Correcting Codes (ECC) so that when a failure occurs, it can be detected and corrected.
#2. Periodically re-write the EEPROM to refresh the data, with said detected errors corrected.
You might think of EEPROM as kind of a highly improved DRAM that leaks charge much slower -- so much slower that it is deemed a "non-volatile" memory. There is still leakage though, which is why the EEPROM should be re-written once in a while, to refresh the data. Just don't do it constantly, as that leads to increasing leakage rates as the chip gets worn out.
Re:REALLY? (Score:4, Informative)
It's so slow that re-writing the data once a year should be plenty enough to keep the data intact and yet nowhere near the maximum rewrite limits.
Re: (Score:3)
Or is it really not possible to make, say, a 386 CPU (one that would work when plugged in an old motherboard) with modern processes?
If you throw enough money at the problem, it can be done.
But that is the wrong solution.
A better and faster solution is to redesign the motherboard to use a standard off-the-shelf CPU.
Designing a new PCB is a lot faster and easier than redesigning an IC.
Re: (Score:3)
Even if you had a replacement chip that was identical in form, fit, function, and reliability to the original (and you rarely get all of these), it would still have to be qualified as such. And that involves a non-trivial amount of engineering and testing plus any required government certification. As long as the currently designed-in chip is available at a satisfactory price it's difficult to justify the time and expense of replacing it in an existing product.
Plus I'm pretty sure that car manufacturers are
Re:Then they need to warehouse old chip, not JIT (Score:4, Insightful)
Re: (Score:3)
Everything has a maximum life, but for production work on simpler devices you use OTP (One Time Programmable).
Re: (Score:3)
Re: (Score:2)
Re: REALLY? (Score:2)
Re: (Score:3)
Ah yes, all you really need to do anything is a 741 and a 555.
Re: (Score:2)
The semiconductor industry needs to be regulated back into submission to provide the critical parts to keep the world moving.
No, they don't. Nobody has to produce something that they don't want to produce, especially when the buyer doesn't want to pay for what it actually costs. If the auto industry wants these older chips that badly, then they need to fork over a lot more than 50 cents for them. All that your brain dead idea would accomplish is making semiconductors more expensive for everybody else just because the auto industry wants to save a buck.
Every country that operated the way you think the world should work always had
Re: REALLY? (Score:5, Interesting)
Re: REALLY? (Score:5, Informative)
I'm an engineer with experience in high reliability products. Some if this stuff is safety critical.
Tried and tested is very important. Updating to newer tech has high costs, everything has to be tested again and there will always be issues that emerge. Manufacturers don't want to be issuing recalls all the time, or releasing janky "patch it later" crap.
For automotive in particular the environment itself makes more modern tech an issue. It needs to work in extreme temperatures, and in extreme temperature swings.
BGA packages are harder to inspect than older course pitch leaded ones. X-ray vs. normal camera. The industry is just now moving into no lead packages, but mostly for non critical stuff.
Re: REALLY? (Score:5, Informative)
I disagree with that. Tried and true is more reliable because it has been better tested over years and is known to work. Whenever you change the design, there are new bugs that can be introduced in the process. The older designs have been used in cars for decades, and their reliability factors are well known, because they have been used so much and for so long. Testing and proving is what makes reliable. So you have a new nanometer design that you believe is more reliable. But its not reliable until its proven reliable through massive, extensive testing in the field. Period. So to change these designs car companies would need to go through an extensive process to test the new stuff for long term durability so that they know it will last a long time and will not suffer failures.
This is not your Windows PC where if it blue screens you can reboot. These are critical life safety systems. So you say that the new chip is more reliable, but thats due to a theoretical design, for life safety, its unreliable until proven reliable in use, in the field, in real life conditions.
Re: REALLY? (Score:5, Insightful)
Former semiconductor roboticist here.
Former mass market consumer electronics engineer here. You're talking utter nonsense. Migration to smaller geometries (for a given chip function) is driven by devices-per-wafer cost considerations on the manufacturing side. Larger geometries - assuming a mature and working process - are inherently more reliable than smaller geometries, ceteris paribus. Semiconductors from the 1980s - MANUFACTURED in the 1980s, not merely "1980s tech manufactured the same way today" are still in active service at this very moment.
Re: REALLY? (Score:4, Insightful)
Re: (Score:3)
Former semiconductor roboticist here. .
Maybe there's a reason you're a former one....
Re:REALLY? (Score:4, Informative)
it's just not that simple. they stopped ordering back in March 2020, they were the only customer of the older Foundry Equipment, and when it went "cold" it became a heap of scrap metal. is it their fault if customers stop giving them money to pay for running the equipment? https://slashdot.org/comments.... [slashdot.org]
Re: (Score:2)
Exactly. The auto industry is huge, why isn't this viewed as a failure of the "free market" to produce competitive, current products suitable for integrating into automotive designs? Why is this a failure of the customers of these products?
Probably because those products are neither competitive nor current.
Re:REALLY? (Score:4, Interesting)
Auto industry chip demand is minuscule compared to total chip demand.
If ICE makers want to keep using old designs, they should build their own obsolete foundries. FREE MARKET!!!
Re:REALLY? (Score:5, Interesting)
At least historically, actually declaring a fab EOL, and having a former asset turn into a superfund site, was not something anyone wanted to see happen; so you had a supply of older fabs that was decided years to decades ahead of time(when the decision was made to build the then-cutting-edge fabs that eventually became today's obsolete ones); that had all paid off their capital costs during their period of being cutting edge and in demand and were now available on quite reasonable terms to anyone who wanted a product that would help keep the lines moving and the lights on.
All that is fine and mutually beneficial; so long as demand for semiconductors on older processes doesn't exceed the supply of 'free' depreciated fabs that would otherwise become negative value environmental hellzones. If, however, demand for current fabs becomes intense enough that it becomes preferrable to upgrade old ones rather than do every new fab as greenfield construction; or the demand for parts on obsolete processes becomes intense enough that it exceeds the supply of naturally-produced obsolete fabs there is a problem: the automakers and anyone with similar requirements can now only get the parts they want if someone builds a new fab for that process, which will mean nontrivial lead times and higher prices than just ordering parts from fabs that would otherwise be scrapped.
Re:REALLY? (Score:5, Informative)
No, that's not it entirely. There's still a global shortage of a wide range of chips. Even Apple has recently indicated that some production may be affected (not iPhones, but other devices). Videogame consoles are still hard to come by. The automobile industry made things worse for themselves by underestimating short-term demand and cancelling orders, but you surely can't blame Apple or videogame manufacturers for "using the old chips just like their grandpappy used.".
https://www.electronicdesign.c... [electronicdesign.com]
https://www.ign.com/articles/c... [ign.com]
https://www.videogameschronicl... [videogameschronicle.com]
Re: (Score:3)
Correction: Apparently yes, iPhones too.
Re: (Score:3)
If you ever look into how many small processors you'd find in a modern car you'll be surprised. Many switches and other input devices are actually not dumb switches but processor controlled devices that both reads user input and lights up indicators in the switch as well as controlling the backlight of the switch.
For some of the items in a car an 8051 is overkill.
Re: (Score:3)
Not just proven designs. Designs without floatng point bugs and security holes.
Floating point bugs and security holes don't magically appear with node size changes. Speaking of floating point bugs, they date back around to the decade of production that the car industry still use now.
As for security holes, have you considered not connecting your mission critical device to the internet? Security is not a criteria for cars, and what I mean by not a criteria is that precisely none of what car processors are doing are in any way encrypted or obfuscated.
I wouldn't trust modern processors in any mission-critical service.
Just as well you're not an engineer t
Re: (Score:3)
You've just missed the point, fingers in ears style, just like the auto industry has. You keep the proven design, but you BUILD the thing with current equipment!
I don't normally respond to ACs but this is too asinine to leave alone.
YOU CANNOT USE THE PROVEN DESIGN IF YOU SUBSTITUTE UNPROVEN COMPONENTS. And in this era, those components are going to require completely new software, too. In high reliability software, even just a die shrink of an existing part - bear in mind this is LITERALLY just the manufacturer saying "we are now making it in a cheaper way, all the old datasheets are still valid" - is 6-12 months of requal. But the stuff that's being talked about h
Sure. (Score:5, Insightful)
Let's swap the tried-and-tested stuff for something that may or not be less reliable under driving circumstances. I'm sure people won't mind their airbags going off accidently from time to time, or other systems glitching...
Re: (Score:3, Interesting)
Note that when the chips were first incorporated into the car design, they were best of breed and not 'tried and tested', and the automakers didn't blink at the 'risk' of adopting then-current technology. This facet is overstated. At least in the context of 'someone else needs to spend billions to keep us from spending maybe millions' request that's essentially coming from that industry. Currently the auto industry is leaning on a sort of indirect government bailout to avoid investing in refreshing to cur
Chip Reliability? (Score:2)
Just what is the reliability difference? Small transistor size chips have been pocket tested in mobile phones, with thousands of steps of walking. But not against car vibrations, pothole plunges, and much hotter and much colder temperatures (temperature may be a bigger problem).
Then how long do those chips last before they degrade? Can they go for 20 years? Look at the problems Commodore 64 owners have with a certain PLA failing after a few decades (it runs hot normally thus the shorter lifetime). And
Re: (Score:3)
I couldn't find it, but there was recent article about newer intel processors randomly flipping bits under stress, corrupting memory and filesystems... This is bad when your encrypted partition gets corrupted, and would be potentially really bad when you're driving.
Re: (Score:2)
But not against car vibrations, pothole plunges, and much hotter and much colder temperatures (temperature may be a bigger problem).
Don't people tend to bring their phones with them in these same cars?
Re: (Score:2)
People generally use climate control to limit temperature and humidity extremes while they are in the passenger cabin. It's very different inside the engine compartment, in the undercarriage, and when the vehicle is unoccupied. Phones inside the passenger cabin are not fixed to the chassis like vehicle electronics must be, so mechanical shock and vibration is muted relative to the stuff running the car.
Re: (Score:2)
Re: (Score:2)
Re:Sure. (Score:5, Informative)
It's not quite that trivial but it's not complicated enough to warrant spending billions on expanding capacity.
For chips in the passenger compartment, in the sun on a hot day, ambient air can be 65C/150F. That's 20C higher than Apple says an iPhone can be ever stored powered down, let alone be active. Explicitly Apple says never to leave your phone in the car, because it can't take the temperature.
Even worse, there are electronics in the engine compartment, where things can be around 95C/203F easily.
So refreshing to new technology will involve non-trivial investment, but still far less than they are asking the fabs to make or for the government to give the fabs.
Re: (Score:2)
What he said.
Moreover, there is another little catch in comparing the chips used in cars with those used in smartphones.
Cars are bigger.
Re: (Score:2)
I think you are forgetting that to use expensive chips means to put a lot of functions on a single chip. So if that chip goes south or, I know this is a rarity, the software screws up, so do all those systems connected with it. Having lots of inexpensive chips builds in a resilience and hysteresis to the vehicle.
Re: (Score:2)
Re: (Score:2)
Riding around in someone's pocket for years is certainly no more stressful to a computer chip than riding in a sealed box behind the dashboard of a car for an hour or two per day.
You're right -- riding in someone's pocket for 16 hours a day is much less stressful than being exposed to the thermal and mechanical environment of an engine compartment for an hour a day.
A rough rule of thumb is that reaction rates double when temperature goes up by 10 K. Typical commercial parts are rated to operate up to 85 degrees C (junction temperature, not ambient) whereas automotive rating is typically up to 125 degrees C. The engine's operating temperature is often right around boiling water,
Re: (Score:2)
You do realize that most "modern" chips are very rugged and reliable, right?
You do realize that the smaller the devices on a die get and the closer together, the more susceptible they are to corruption or destruction by voltage spikes - right? Automobiles represent pretty harsh electrical environments.
Riding around in someone's pocket for years is certainly no more stressful to a computer chip than riding in a sealed box behind the dashboard of a car for an hour or two per day.
Connect your phone's charging port via a suitable voltage adapter directly to your car's power bus, stick it somewhere in the engine compartment, and use the car normally during a full Northeastern winter. Then test the phone in the spring and tell me again how the environmental condi
Re: Sure. (Score:4, Informative)
Weird to make an assertion like "arduino does not have a watchdog timer", when it certainly does have one. At least the Atmega328p does. It's possible you or the original poster meant a raspberry Pi, not an Arduino. But that ALSO has a watchdog timer....
Re: (Score:2)
This does not apply. This isn't about software, but hardware.
It's not "obsolete" when there's no replacement. (Score:4, Insightful)
Re:It's not "obsolete" when there's no replacement (Score:4, Insightful)
You do know that semiconductor fabs have been binning and continue to bin mil-spec counterparts for just about every single part they produce, right? Just because the discrete FET produced today is done with a 10nm process doesn't mean that there aren't parts binned for "industrial-level durability, resistance to extreme temperatures and shocks, etc.", the same as there were literally at least 40 years ago with larger processes.
I mean, have you ever even browsed Digi-Key's website? Or looked up manufacturer datasheets for mil-spec parts? This is a problem with car companies not wanting to spend a few bucks to ensure their existing design works with an upgraded replacement, not with the semiconductor manufacturers moving to smaller processes. Gelsinger is right: it's tens of billions of USD to open up fabs for larger processes and millions for car manufacturers to get their heads in the game.
Re: (Score:2)
Re: (Score:3)
When you move an old chip to a new fab process, it is pin-compatible and you do not change the software.
The engineering changes are at a lower level than that.
It also doesn't change how complicated the chip is, or what it's capabilities are.
It does change the power use, but not in the sense you use the word "power."
And the price goes down, not up.
Re:It's not "obsolete" when there's no replacement (Score:5, Informative)
firstly: the lower geometries are, by virtue of being smaller, more susceptible to heat and cosmic rays. this has been a serious problem for NAND ICs, DRAMs and electronics in general that goes into space: the cosmic rays don't get any smaller, meaning that a ray going through an ASIC changes *more* bits rather than less.
secondly: the electronics in vehicles is incredibly simple, usually handling very large current. there is *no need* for finer geometries. it's like selecting a 0.2mm 3D printing nozzle to 3D print a 1 *METRE* cubic object. all that does is push up the price and take more time.
thirdly: the automotive industry is under serious "optimisation" pressure for economic reasons. we think, "huhn that part's $0.50, that's so stupid, why didn't they use the $2 part?" - but you've forgotten two things:
a) there's THOUSANDS of $0.50 parts in a vehicle. swapping all of them out with $2 parts increases the manufacturing cost of the vehicle by ONE THOUSAND FIVE HUNDRED dollars.
b) in mass-volume they manufacture millions of vehicles. that one thousand five hundred dollars times one million vehicles including spare parts that are BY LAW REQUIRED TO BE SUPPLIED FOR UP TO SEVEN YEARS has to be funded, and you are literally talking here about a BILLION DOLLAR increase when multiplying the $1.50 increase in part cost by 1,000 parts and 1,000,000 vehicles.
that's not to say that the Automotive Industry aren't themselves directly responsible for this situation: they stopped ordering parts around March 2020, they were the only companies in the world buying from those older Foundries (350nm, 2000 nm), and when they stopped ordering, then just like that Power Station in South Australia where the walls cracked when it cooled down on de-commissioning, the older Foundry equipment had to also be scrapped.
Foundry equipment is extremely sensitive to operating conditions: if a Foundry doesn't have any orders *they run empty wafers* just to keep it at the correct operating temperature.
then, worse than that: as i explained last time i posted on this topic, you simply can't take a 350nm design and hope it'll work on 180nm, 90nm or any other: each Node is completely different, has different layers, has completely different equipment, uses different chemicals.
it's *literally* a "redesign from scratch' job, and with many of the original designs having been done 40 years ago, some of the designs might not even be available any more (and their original designers retired or even died), or worse, actually require *reverse-engineering* through de-capping the ASIC and undergoing a painfully incredibly-expensive process of etching away layer-by-layer and photographing portions and reconstructing the Masks.
it's an extremely complex situation which, frankly and bluntly, the Automotive Industry only has itself to blame for.
Re:It's not "obsolete" when there's no replacement (Score:5, Insightful)
The typical computer geek, as evidenced here on /., thinks computer programming is HTML jockeying and iPad apps, they don't even think that car companies understand computing or programming apart from their savior, Elon Musk. They demonstrate a complete failure to understand just how massive and pervasive computing is and have no idea what is important to embedded applications which dominate computing applications overall.
This is all a game of redirecting blame. Car companies do not control the chip industry, they are consumers of it. The chip industry is failing to meet the needs of the market, it is not the market's fault.
Re: (Score:3)
Re: (Score:2)
...Car companies do not control the chip industry, they are consumers of it. The chip industry is failing to meet the needs of the market, it is not the market's fault.
On the contrary, the chip industry is meeting the needs of the market, just not this segment of it. Building automobile chips consumes resources that can more profitably be used to produce other chips.
If a consumer wants to freeze his technology, he needs to satisfy all of his external requirements in-house, so he doesn't depend on external vendors. Good examples of transportation industry consumers who have done this are the Mount Washington Cog Railway and the San Francisco cable cars.
Re: (Score:3)
This is not minor niche stuff, there are thousands of businesses that rely on chip that are not used on consumer PCs, tablets, or phones. These chips far outnumber the big package chips and is a larger market in volume.
The consumer is not freezing the technology, they have requirements that the big name chips don't provide.
I have some some companies sometimes not understand this. A medical company using Intel multi-core chips normally used in laptops, essentially only doing serial protocol transactions. T
Re:It's not "obsolete" when there's no replacement (Score:5, Insightful)
The automakers cut way back on their chip orders when the pandemic hit, leaving their suppliers twisting in the wind. Those suppliers have since found new customers who are willing to pay and it would be pretty stupid on their part to ditch those customers without someone putting up a better deal. That is how markets work.
The buyer I support (in a completely different industry) is currently going through a similar sort of learning: you can work with a regular set of partners everyday through the highs and lows of supply and demand or you can play musical chairs with whoever has the lowest price today. Both can work you just have to except with the latter that sometimes there won't be a chair for you at any price.
Re: It's not "obsolete" when there's no replacemen (Score:3)
No. By that logic, there is a opportunity in the market for a a new company to enter the market and capture all the chip business from the automakers. If that isn't happening, there is a reason. Either automakers won't pay enough, commit to large large enough orders, or something else.
Re: (Score:2)
As a firmware engineer I find it amusing that people think that bigger transistors must somehow be more rugged.
Regardless of transistor size, you can choose packaging to be more or less durable.
Carmakers just don't want to spend the design money to save on part cost! But they should be doing that every 5 years or so, anyway, because of gains in efficiency and reliability in the newer designs.
The difference isn't speed, BTW. And it doesn't actually cost a lot to move a design to a newer technology; it is v
You mean Linux won't run on my 286? (Score:3)
Cars are a very harsh enviornment (Score:5, Informative)
Re: (Score:2)
The question is whether engineering and testing to those standards with modern technology is going to cost more than the billions they are asking to be spent to open up fabs that no one but them wants. As he said, spend millions to avoid spending billions.
Re:Cars are a very harsh enviornment (Score:4, Insightful)
The question is whether engineering and testing to those standards with modern technology is going to cost more than the billions they are asking to be spent to open up fabs that no one but them wants.
No. The question is whether the semiconductor manufacturers are going to give ongoing support their customers or dump the costs of them migrating to newer technologies onto them for the manufacturers' own convenience as they dump older plants to concentrate on the cutting edge.
It's also whether the auto companies will continue with those manufacturers who shafted them when the go to their later generations, or find other vendors or solutions.
Re: (Score:3)
This is fine, so long as that's where it stops. However it seems Intel was approached as 'we need any company that can do chip fabrication to fab for us' as their current partners are unable to fulfill. Intel is not historically a fab that is known for taking on fabricating fabless companies chips, but for them to be talking about it suggests the auto makers are shopping around and their existing partners aren't jumping up and down to try to build more fabrication capacity faithful to 2003 specifications.
Re: (Score:3)
I worked on automotive electronics and they are tested to the same standards as electronics shot into space. If the electronics are in the engine bay they are tested with 1000 salt fog, temp cycling, and other extremely harsh tests. It is a bit naive to say the chips are comparable to those in an iphone. If your iphone stops working you don't crash going 80mph.
That’s the point most miss. The chip makers want to sell the latest while car manufacturers have designs that meet their needs and have proven to be reliable and thus don’t want to change since there is no compelling reason to redesign entire systems. It works is a lot better than new tech that needs the bugs worked out.
These articles are rubbish (Score:4, Interesting)
I bet it's not the safety critical chips that are in short supply.
It'll be the 28-14 nm infotainment chips that cars are getting all chocked up with these days. You know, the big whizz-bang touchscreens for those cellular connected devices that no-one is meant to be fiddling with while driving.
The simple models, with just engine management and the basic electric functions, if there is any still, will be shipping just fine.
Re: (Score:2)
By law, all cars sold in North America now must have the screen, for the backup camera.
Re: (Score:2)
I think his point was not that they can skip the chips, but the chips in question may not all be as scary as brake controllers. Of course even those may represent a service/warranty burden as they operate under harsher conditions than a tablet and still warrant attention.
But his point is off, as for example chip shortage caused models to use more gasoline, precisely because chips in fuel management systems were also in short supply. So it's not just the tablet in the dash that causes them headaches.
Re: (Score:3)
I bet it's not the safety critical chips that are in short supply.
Then you'd bet wrong. Article after article in both consumer and trade media for the past six or nine months has pointed out that it's the COMMODITY CHIPS that are in short supply. It's even stated explicitly in the summary for you: "Because of a 50-cent chip, we are unable to build a car that sells for $50,000."
But I'm sure you know better.
Re: (Score:3)
There are only four fabs that can make modern CPU-level chips (Intel, Samsung, GloFo and TSMC), and their capacity was already maxed out. Then AMD started usning TSMC instead of GloFo, and even Samsung and Intel started offloading some of their fabrication to them, not to mention GPU makers due to the crypto demand. So now you have a single fab creating half of the chips in the world, already barely able to meet demand, and then the pandemic hit. So in that case, the chip shortage is understandable, this ha
Not really a supplier (Score:2)
If semiconductor suppliers like Intel and Qualcomm have their way
It's a stretch to call Qualcomm a supplier since they are completely fabless, they can't make a single chip themselves -- they don't make money from 5 or 10 year old proven designs, they make money from new cutting edge designs, so of course that's what they want to push carmakers to.
Why do they need old fabs? (Score:2)
What makes a modern plant unable to produce 90-nm chips? Or is this just a business decision because they can create a more 7-nm chips per wafer than 90-nm chips, so the 7-nm chips are more profitable?
Re: (Score:2)
They have invested the money in equipment that can make the smaller lithography; if the die size for 90nm is 10x a modern version, that means a wafer can only produce 1/10th as many chips, which increases the necessary cost recovery up 10x per chip. That would make the total chip at least 5x more expensive.
Re: (Score:2)
A modern plant can easily make 90nm chips, but the cost is greater than 14nm, as you're paying for more raw materials (bigger dies for the same functionality) and the fab that you're paying off cost billions more.
Re: (Score:2)
Economy of scale. How many 7nm chips do you think they sell to various smartphone and electronics manufacturers? Compare that to sales to car manufacturers. From what I have found, 70-75 million new cars are sold in 2021. Smartphones has been shipping in numbers of around 1300-1470 million each year since 2014. That is a huge difference in volume.
So it not only that you get more chips per wafer, you have a much higher demand for the smaller chips. And the output per wafer is only one issue. You also have ca
Building modern AI chips themselves? (Score:2)
Like Tesla?
Oh the horror.
Solution (Score:4, Interesting)
The solution to this problem is vertical integration. There was a time when automakers had their own captive electrical/electronics subsidiaries. Then they figured out they could save $1.50 per car by spinning off the subsidiary, who then relocated to Mexico or Taiwan. The rest is history. Or herstory.
Spare parts (Score:2)
You cannot supply spare parts for the life of a car if the process for building those parts themselves is re-designed on a yearly basis (or however fast this article wants them to change).
In order to replace a part that contains electronics that were certified for that part, with a newer hotness part, that newer part will have to be certified to same level as the former part, yet still have to fit into the same form factor. And that ain't going to happen.
Re:Spare parts (Score:4, Insightful)
A USB stick created with 2021 technology on cutting edge NAND fabrication processes can be plugged in and function on a 20 year old PC. You can both refresh technology and maintain compatibly, so long as that is an engineering goal in the design. New dies do need new packages, but those packages can usually be pin compatible or, if the voltages are not amenable to that, at least can be compatible at the board level with some voltage regulation.
Again, they aren't saying it is trivial, they are saying that it will cost millions. They also aren't saying they have to chase changes every year, but they have to be able to at *some* point be able to refresh to avoid being utterly left behind. If 'must never refresh' then companies would still be demanding 'tried and true' vacuum tubes for their designs because they don't want to deal with refreshing to newer technology.
Re: (Score:2)
They also aren't saying they have to chase changes every year, but they have to be able to at *some* point be able to refresh to avoid being utterly left behind.
Except reducing product lifecycle durations means they would have to chase changes every year. Auto makers have enormously complex supply chains, and the use of third party assemblies means that some fraction of product lifecycle is consumed by the third party R&D and then including it in an auto design.
Separately, even if specs are identical, it's very risky to refresh every component in a new design. So if the Edsel Autovoiture gets a major overhaul every five years, it's probably not going to use a
Re: (Score:3)
No, they wouldn't have to chase changes every year. You secure commitments in advance and engineer around your multi-year commitments that you can/can not get. They are at the end of the level of commitment they are easily able to get of the old tech, which is a clear sign that they need to refresh. The stakes are much lower in my product but we still have forecasts and knowledge of when suppliers may expectedly yank the rug out from under us and multiple sources with contingency plans for continuing to
Re: (Score:3)
A USB stick created with 2021 technology on cutting edge NAND fabrication processes can be plugged in and function on a 20 year old PC.
Does your 2021 USB stick work with the 2000's computer because of it just following the USB spec, or is there a regulatory requirement for it to work with the older computer?
Because they are two different things.
Intel doesn't even make what they need (Score:3)
Most of the 'chips' that automakers need are microcontrollers, it's really ignorant and stupid of the Intel CEO to even suggest that someone use Intel processors in the place of a microcontroller because they don't really make products in that market space and the products they do make are significantly more expensive then other companies.
In addition switching microcontroller lines take significant time and expense because you have to retrain engineers and and redesign hardware like pcbs. It would probably take most teams at least 3 to 6 months to redesign. Then the software engineers have to familiarize themselves with a different product line.
You could use Altera fpgas, but Intel took that company and wrecked it. After they bought alterra, they made their FPGAs more expensive, less accessible and made the data sheets harder to read. So much so that some teams of engineers stopped using Altera parts. I would suggest that no one switch Intel
Re: (Score:2)
Intel has fabs and can produce things not of their own design. They also do boring designs no one ever talks about. And Intel CEO didn't just jump into the discussion out of nowhere, the auto manufacturers are trying to pressure every company with chip fab capability to make big changes. Intel is being asked to help, whether they are the best choice or not.
And again, he said they would have to spend millions, he didn't say it was easy. Just that millions of dollars of re-engineering is cheaper than billion
Fools. (Score:4, Insightful)
These business fools need to stay in their lane. They seem like the type of people that would tell NASA they need to stop using the RAD750, 110MHz PPC processor made with the 250nm process, because they don't understand why it was chosen.
I don't know about you but if my car was as reliable as a smartphone then I wouldn't drive it because it would be a death trap.
Next up ... (Score:2)
What is this stupidity? (Score:4, Interesting)
Safety tech 10 years old is in no way "obsolete". The term for it is "proven". Most engineering components are way older and it is not a problem, it is an advantage.
The stupidity that caused this shortage is lack of inventory because "just in time" supposedly saves money. Actually it does not, but you need to be able to plan over more than the next 3 months. Save a penny, lose a million.
Big Digital-centric View (Score:2)
Re: (Score:2)
Reliability (Score:3)
I'd rather have a slightly less powerful processor in my car that works under all circumstances than one state-of-the-art one that fails under real-life conditions all the time.
BTW "latest and greatest" is severely overrated: a processor used in a car doesn't need gigabytes or RAM and a gigahertz clock rate. You can do an awful lot if you program in assembly with a mere couple hundred of kilobytes.
8-bit love (Score:3)
6502 fo lyfe!
There's a reason chips like the 6502 are still in production.
Fraidy-Cat CEOs (Score:3)
“Because of a profit motive to use 20 year old proven technology, we see no need to invest in modern processes.”
What utter bullshit.
Cut the CEO's salary to $1m, because anything beyond that is wasteful and doesn’t improve the company. Take the rest of the profits and invest in the staff and the facilities, as well as testing because that’s the best way to spend the company's money - not by dumping it into stock buybacks and dividends for stock holders.
Compute while you commute (Score:3)
You do realize that you can up-cycle off-lease ("obsolete") server processors into other devices, for instance a car. If a car is still using 20-year old microcontrollers then even a Sandy Bridge microarchitecture era Intel Xeon E5-2650L would be a major step up. If the computer industry can align its eWaste dilemma with other industrial processes it would be a major win for the economy and our environment.
In-fact, if you follow this thought process all the way, you can have your car partake in edge cloud computing services anywhere it goes, the fuel source powers the processors, for instance 100kW worth of TDP. The input is fuel, the outputs are computational power and heat. So then recoup all of the heat using a sterling engine to power the wheels.
Just think of how many Apple A-series and Samsung DRAM chips are getting torn apart by an industrial shredder each day just to extract the trace amounts of gold and rare earth elements.
Re: (Score:2)
Droll. But the human body is also made up largely of common elements. After all, there are only a small number of common elements. It's how they are combined that is the trick.
"Most of the human body is made up of water, H2O, with bone cells being comprised of [sic] 31% water and the lungs 83%.1
"Therefore, it isn't surprising that most of a human body's mass is oxygen. Carbon, the basic unit for organic molecules, comes in second. 96.2% of the mass of the human body is made up of just four elements: oxygen,
Re: (Score:3)
Tesla has been agile in updating their boards and software to use the latest chips and as a result they haven't had to curtail production. (Actually, their production is increasing rapidly. Just another way they are going to put the ICEmakers out of business.)
And fortunately Tesla’s electronic systems have proven to be trouble free and reliable.
Re: (Score:3)
You're 11 times more likely to have an accident in an ICE car than a Tesla. They must be doing something right.
Tesla likes to tout its safety but that number lacks context, especially since it's their autopilot number. Much of the autopilot miles are probably on interstates and freeways, for example, where far fewer deaths per mile occur for any type of vehicle. It also uses all ICE cars, many of which are much older and probably not in top condition. That statistic sounds good until you look into it and discover Tesla is not providing a valid comparison.
I guess they want you to ignore the problems with screens t
Re:20 years parts availability Baby! (Score:4, Interesting)
Car manufacturers want the part to be available for 20 years. They don't want latest, smallest, fastest tech.
They want something that 1. works, 2. is reliable, and 3. if it goes bad, you can replace it.
And modern SOA chips do not meet numbers 2 and 3.
I have modern radios with modern chips in them that have turned to junk because the chip mfgrs stopped making needed parts.
And in an ironic twist, I just refurbished two mid 1960's Heathkit radios. 56 year old radios that ban be repaired easily, and 3 year old radios that can't be repaired.