Hacker News new | past | comments | ask | show | jobs | submit login
Auto chip aging accelerates in hot climates (semiengineering.com)
42 points by PaulHoule 4 days ago | hide | past | favorite | 69 comments





This is common knowledge among chip designers...

Yeah, this article could also be summarized as: electromigration exists, news at 11. It's literally always been a problem, it just might be exacerbated by increased thermal cycling and increased high temperature conditions of computationally intensive modern vehicles.

Does it start to have an impact faster on smaller nodes? I know automotive chips have been migrating to newer nodes (like 28nm not cutting edge) in response to the chip shortage, and I’m curious if that might impact this aging effect.

Edit: read the article, this seems to be what the article suggests as well


Yes, sort of. Electromigration is well known in the industry and there are extensive design rules around making sure you don't violate those rules and hit electromigration limits within the lifetime of the product.

Most IC designers tend to design for the 100k hours at absolute worst case conditions. What is known as PVT: worst case process, voltage, and temperature corners.

Assuming they do that, the likelihood that electromigration actually being the issue is not a concern. Designers can mess up, but generally they don't want to do this... especially in automotive. Auto manufacturers will ruin them if they determine there's a design flaw like this.


Chips should be standardized and easily replaceable. Have a replacement schedule during regular maintenance. Problem solved.

That is an all around terrible idea. Even in the best environments chip replacement is by no means certain and requires significant expertise. This is in no way sustainable or doable and obviously easily abused by manufacturers.

No, you've created many more problems. Your PCB now has to be larger, you need sockets for the ICs, increasing volume, weight, and cost. You also lose the conformal coating which offers moisture protection and you've introduced a new failure mode where chips vibrate out of the socket. Costs further increase as the manufacturer now needs to stock thousands of chips, most of which will never sell. You also introduce more manual steps in the assembly process. Boards can't be machine assembled to the same degree, which drives up costs even more.

You could go with a socket-less design, but then you need expensive equipment and a highly skilled technician to do the replacement. You still lose the conformal coating, as such coated boards are not very repairable in the general case.

For as much as you'd pay for a replacement chip and labor, a new PCB costs ten times less.

Component level repair is never an economical option in this scenario. It only works out if replacement parts are unavailable.


Trust me, you do not want your average grease monkey soldering a new bga while doing an oil change.

I trust you but don’t like the way you call my fellows workers.

What if a standardized chip could get some kind of standardized pin out so it can slot into a PCB with a harness or something? Maybe not safe enough without a solid solder joint? Though we use plug-in fuses all through cars. Are they only for non-critical systems?

PCBs are cheap to manufacture and are usually coated, to ensure longevity, requiring significant effort to replace any parts.

Do you want your ABS controller to be seated like that? Seems like a genuinely terrible idea.


No, I probably don’t, you’re right. I took a look at some of the pin outs for chips in my old car and even those are pretty massive in some cases. I’m not sure how you’d ensure they were seated and secure with such little movement. I suspect these are normally soldered and even epoxied in place.

Just sell me the whole ABS controller then.

That is actually the hard part, odds are the chips inside are not longer made. When making ABS controllers the OEM guesses how many replacements will be needed and buys that many extra to put in a environment controlled warehouse - when they guess wrong they run out and if the parts inside are no longer made the rest of us are out of luck. If the chips inside are still made the OEM will generally order more (if they run out before the expected date that implies there is a lot of demand so it is worth ordering more).

The expected date above is for normal care lifespan. If you are a collector good luck.


What do you expect manufacturers to do exactly? They all sell replacement parts, it is good business.

You obviously can't expect them to support models forever. Even right now their suppliers keep open ancient manufacturing lines for the ancient chips used in these controllers.

>If you are a collector good luck.

Good thing that is a small minority of, usually well off, people.


Chip manufactures have very different business from car part manufactures. The car parts manufacturers would love to sell me a new ABS controller, but the chips that go into it are not made anymore and they don't have the ability to get them made in many cases. Most often because they just bought some common part like a 16 bit CPU - when they have their own custom chips they can maybe find someone else to make them (though sometimes this isn't possible)

The surface area of each pin of a plug-in automotive fuse is somewhere around 100-150 mm^2. Vibration, thermal expansion/contraction, etc is easily handled with that much area for maintaining contact.

How big do you want your automotive-grade chips to be?


I have a feeling they could be larger, but I’m very naive in this field. Great point though, I see that the pin outs are fairly expansive on some of the chips in my old car, and they’re likely worse in my newer car. It would be difficult to achieve any kind of secure connection without solder and epoxy.

Trust me

And we should trust you because...? I wouldn't trust you because of the way you insult professional technicians that work on your car, and obviously know little about the profession.

That, and if you think replacing these parts would involve soldering and not a complete module replacement, I wouldn't trust your opinion on anything electronics-related, either.


This should be something that is pulled as a package and replaced, why would anyone be soldering anything? Do refurbishing off-site if that makes sense.

Component-level repairs died in the consumer market in the early 2000s. The only way to bring it back from the dead would be to release the complete service manuals (i.e. https://www.ebay.com/itm/266686900554)

Actually these are Sams independent repair manuals.

They existed because the manufacturers would not always release the factory schematics or service manuals for mass-produced consumer products.

This proliferated once consumerism & disposable-ism really took off, before that most electronics came with a schematic, but it was still not often up to the standards of places like HP or Tektronix.

So Sams took what info they could get and reverse-engineered the rest as they tore down and scrutinized new equipment to produce these fairly standardized publications that were available by subscription to independent repair operators, covering each popular model from many different manufacturers as they emerged in the marketplace.

A lot of the repair operators were very familiar with gear like HP and Tek and so was Sams.

Most of them had successfully leveraged that kind of top documentation and serviceability to have materially contributed to the winning of World War II. Slouches they were not.

There was a preference not to want to settle for less in consumer electronics so they published their own to make up the difference. If the skills are there why not use them? Especially when a failure is usually only one or a small number of the most common dirt-cheap components that need to be replaced to fix.

Which is what electronics servicing is supposed to be all about from the beginning.

I know it's not the beginning any more but still there was a time when the pros had in the back of their mind the kind of things it might take to win World War III, more so than now.

With people like that around you didn't need right-to-repair laws to be enacted, the whole country was still aware of the benefit of field repair if not improvisation.

All of the overseas Communists were certainly not going to sleep on it.


They already do this at the module level (besides the 'regular maintenance' aspect). No reason to change that.

“Our automotive customers are now at the leading edge with 5nm and 3nm chips.”

This is the problem. We know that the number of problems with silicon goes up as the feature size goes down. The older nodes are much more reliable. Not putting something on an AIC is often easier to fix, too.

The solution is to keep cars simpler so that they can use robust nodes. as a consumer, by the dumbest car that you can buy. You also want this little proprietary addons as possible so you don’t have to replace a whole module because a chip fails.

I’d like to see a whole, new, auto company funded the bed makes nice cars that are easy to repair with minimal electronics. Maybe customizable with different skins on the outside. Open interfaces in format so that they’re easy to customize on the inside.


> For a chip designed to last 30 years, high ambient temperatures reduced the life expectancy an extra 10% a year, so after one year the lifespan dropped to 26 years, Pateras said.

Doesn’t sound too bad?


If after one year, it should have 29 years of life left but instead will last 26 then it sounds like the chips won’t last 8 years (instead of 30). Sounds bad to me.

I think it does sounds really bad. After one year, 26 years lifespan. We get that by doing:

30 - (30.1 - 1) = 26.

Doing that again for the next year: 26 - (26.1 - 1) = 22.4.

So you now have 22.4 years of life vs an expected 28.

etc...


If the math is correct, you'd have an expected life of ~10.2 years. (30*(0.9)^n - n) = 0.

That said, when someone says "it reduces the lifespan by 10%", I'm a bit skeptical of the math. Either they are not properly testing it, or the temperatures are exceeding silicon parameters.


Can you please continue stated math for each year? You reach 0 faster than expected

People typically own vehicles for more than one year.

Do the math. It's bad.

Maybe we could do what we should already be doing for a high reliability system and build the vehicle so it can be:

A) redundant, a chip failure will not cause the car to brick/be un-drivable, spacecraft have been doing this for decades...

and

B) FUCKING REPAIRABLE


Aren’t cars already pretty good on repairability? I know the tools necessary for diagnosing and resetting electronics can be very expensive and are more and more necessary with newer vehicles, but I believe they can still be purchased by independent repair shops. I was under the impression this was protected by law

I spoke with a BMW mechanic who said he wouldn't feel comfortable repairing his own car at his home garage because of specialised tooling they have now.

I used to make third party scan tools. My advice is don't drive a BMW farther from a dealer than you will pay a tow back. Not that they are unreliable cars, but that our tools were far better than our competitors at BMW and I wouldn't give them a passing grade. (BMW writes their own tools). Of course I left that job in 15 years ago, maybe things have changed.

For other makes of cars third party scan tools work pretty well at diagnostics.


Commodity hardware that you can replace in the field. When they're saying 3nm-whatever node chips, they mean "We put a $250 Jetson SOM in the car, but fuck you, you can't replace it without replacing a $5000 "Ai CoNtRoL sYsTeM"

Tell that to the people buying Telsas since 8/24 that have HW4.1.

Do you have a tldr or some link I could read to know what you mean? Is this hardware braking a lot from factory so that it already needs repairing less than 6 months later? If so that seems like a warranty issue more than a repairability issue (although I’m sure it could turn into one). Or do you mean that this hardware prevents repair in some way that will be a problem once it starts breaking?

There are a lot of articles recently.

“most driver assistance features are disabled, including rearview cameras, forward collision warning and blind spot warning, GPS navigation, and range estimations”

https://jalopnik.com/tesla-s-brand-new-cars-not-drivable-bec...

https://www.notebookcheck.net/Tesla-HW4-computers-are-failin...

https://www.autoevolution.com/news/tesla-scrambles-to-fix-ai...


thanks. Sounds much more like a warranty / manufacturing quality issue rather than a repairability issue at this point. Although if Tesla somehow prevented parts from a donor vehicle from being used in a different vehicle then I would call that a repairability issue.

For mechanical parts cars are very good. For electronics not so much. It seems like you could repair a chip with an electron microscope and other similar tools - but I've never heard of anyone trying. I have heard of taking the case off to look at the inside of chips (for processes larger than current 3nm); but not repairing issues and putting it back into service.

Electronic parts tend to go out of production. If nobody makes the failed chip anymore what can you do to repair it? If you can get the correct chip we can put the software back on it (well good luck getting that, but it is possible), but all too often parts are not available at any price.


Those tools are mostly based on reverse engineering and exploiting security vulnerabilities though.

Remove “auto” from the title and it’s still true

this was also the case with the defective takata airbags, particularly in warm humid climates.

Can't read the article but it does make you wonder.

How would they know this unless they were already making them barely good enough to last beyond the warranty period in northern climates?

Or maybe one day the only vintage cars will be same ones there are now, only the ones made without chips at all might be capable of operating beyond 30 to 50 years ever again.

The people up north may have fair warning decades later when the collectors having something like a mint 2025 Mercedes in places like Texas or Florida start having theirs die for this reason.


>How would they know this unless they were already making them barely good enough to last beyond the warranty period in northern climates?

Life time estimation for parts has existed for over 100 years. It is not a novel science.

Also a fun little exercise, what do car companies sell, besides new cars? That is right, they all sell used cars. For most car companies used cars are a huge part of their businesses model, their ideal customer buys new cars often AND sells their old car to the manufacturer, so they can sell it again.

>Or maybe one day the only vintage cars will be same ones there are now, only the ones made without chips at all might be capable of operating beyond 30 to 50 years ever again.

All cars are made to die, there is no alternative. If the owner wants to avoid that, they will have to do constant maintenance. Vintage cars do not exist because rust didn't exist in the 1990s, but because they were extremely well cared for and had significant replacements.


Car makers need cars to last for a while. That someone can trade in their 3 year old car means and there is value means that someone can afford to spend more on the new car they are buying. However they also need old cars that the poorest drive to die eventually so that there is someone all the way down looking for a different car once in a while.

Many of the parts of the car are bought from a third party who gives a discount to the OEM because they can sell replacement parts later. (put my alternator in your car, and when it wears out in 5-10 years they will buy a new one from me)


It's not like classic cars are flawless immortals. It takes a lot to keep a classic car going and this will be no different.

While true, classic cars are "simple" in that all of their parts can be replaced and remade in people's home garages with commercially available tools like lathes. If a chip breaks down, I doubt even in 100 years someone can make a replacement at home. In theory there will be a manufacturer or market with replacements, but it won't be easy.

It is substantially easier to replace the entire computer system of most gasoline engines if something critical like the ECU is no longer commanding the engine to deliver fuel or spark. In the US this constitutes a "defeat device" and is prohibited by federal law

It is possible to retrofit many old engines that didn't have an ECU with a modern fuel injection system.

My son drives a 1996 Buck Park Avenue which dates from the beginning of the ODB II era. We got it for about $4k but it has been through a few difficult repairs that took my son or auto techs a while to figure out. For instance the immobilizer system failed with the result that we couldn't start the car, it took a tech a lot of investigation to realize this, but once he did it wasn't hard to disable.


What would a defeat device look like for an electric car?

I suppose that hasn't been explored in the legal system currently. Most electric cars would be considered a zero emission vehicle. If you bolted a generator to an electric car that'd change its emissions. One could argue this defeats the original emissions target, but that sounds like a stretch.

Overall if the car has some sort of devices that limits, mitigates, or eliminates emissions then any modification to it can be considered a defeat device


> It is substantially easier to replace the entire computer system of most gasoline engines

Verses a carburetor? Have you worked on engines before?


I read it as agreeing with you by adding an interesting anecdote. Ie “yes, the old systems are easier to keep on the road. It would be possible to keep newer systems on the road with custom built parts (aftermarket ECU) but unfortunately that is illegal “

Carburetors are one of the most complex parts of a classic car and thus a bad example. If there are major carburetors problems it is probably simpler to replace it with fuel injection. (but only if a community of people work together - all the different maps and such in the ECU need to be created before you can make a useful fuel injection - this is long tedious work)

I think that was "versus replacing a chip."

The ECU has fairly standardized interfaces, mostly CAN or just simple switching circuits on or off. You can just get a replacement ECU. These replacement ECUs generally don't do things like using the O2 sensor to adjust the fueling, so they can burn dirtier than the original ECU.


> mostly CAN

CAN is just the physical and framing layer. The data that goes over it is a whole different matter, and it's proprietary to every vendor, incompatible and generally undocumented.

For a lot of cars you can't even swap the exact same ECU between vehicles (due to bullshit like the VIN being written into it for no reason, and often no official way to change that other than "buy a new module" - so you have to hack around with reverse engineering & EEPROM programmers), let alone mix and match completely different parts.


> the VIN being written into it for no reason

There is good reason for this: theft protection. A long known crime model is the "chop shop" where they tow away a car, then cut it up for parts. Because the VIN is in the ECU a chop shop has less incentive to exist since there is less money in it.

It does of course make legitimate repairs harder, and probably there should be a process to write a new VIN in - but it should only be available to someone who can verify the ECU wasn't stolen (which isn't free).


EEPROM programmers and various other ways of rewriting this VIN (turnkey tools using undocumented diagnostic commands giving you unauthenticated read/write access to memory) means it's just one insignificant expense for the chop shop (as it's amortized over all their cars) vs a multi-thousand-dollar investment in money for a legitimate vehicle owner or part-time mechanic.

Only if it is done via EEPROM and not via something more complex. Though your point stands.

That's true, but it's usually pretty simple to reverse engineer, and that's been done for a lot of cars. CAN message are generally sent out at fixed time intervals, and correspond to some physical or control value, with a simple linear mapping between some physical unit and the value in the message. You collect some CAN data, change something like pressing the accelerator, see what changes, record results.

Just search for "aftermarket ECU", you'll find lots of them; generally for high-performance cars that gearheads want to soup up further, but the same basic principle applies to most ECUs.


Neither case is easy. Making something "simple" like a new control arm or connecting rod is way beyond a typical garage shop. The electronics are not a home shop item either but replacing parts with work-alike modules is already something people do with aftermarket ECUs etc. As long as there is enough demand people will find a way to make the parts (to supply aftermarket), for example I expect the first gen NSX to be repairable for decades.

I very much doubt cars with a lot of electronics and reliant of software and increasingly network connected are going to last that long anyway. The manufacturer stops updating software after 10 years if you are lucky and then its EOL and increasingly unsafe and insecure.

Guy with no experience maintaining classic cars. Good luck making carburetor diaphragms on your lathe.

While true, classic cars are "simple" in that all of their parts can be replaced and remade in people's home garages with commercially available tools like lathes.

I've an '81 VW van in my driveway that begs to differ. It's got a bad idle stabilizer. The vehicle won't start unless that electronic idle stabilizer is bypassed. It runs fine in the bypass state, and good thing, because NOS parts ran out a few years ago. There might be efforts to create a clone (haven't checked thesamba.com in a while). I'm sure a clone is doable by someone that paid more attention in electronics class than I did, but no one is going to pop one out on their garage lathe.

And that's just the electronics. It has mechanical fuel injection (fuck you, Bosch), but good luck making injectors in your home workshop. I'll not start on suspension parts whose manufacturing require a far larger hydraulic press than anyone is likely to have in their garage.


It will require different skills though. One of them may well be reverse engineering, it's not as if the owners manual for a modern car includes schematics and source code.

Because the physics is fairly well know, as I understand it. See for example these[1][2] articles from Microsemi and TI respectively, or this review paper[3] for power electronics.

[1]: https://www.microsemi.com/document-portal/doc_view/124041-ca...

[2]: https://www.ti.com/lit/an/sprabx4b/sprabx4b.pdf

[3]: https://www.mdpi.com/1996-1073/17/18/4589


>Can't read the article but it does make you wonder. How would they know this unless they were already making them barely good enough to last beyond the warranty period in northern climates?

Sorry to be a stick in the ass, but RTFA. The Arrhenius equation is mentioned and is quite interesting.


Didn't intend to hurt anybody's feelings but it did stimulate discussions when there were absolutely none.

The work of Arrhenius I do remember from 1970 myself, and it has been a consideration over these decades, not only with the electronics but the chemicals professionally since then.

But I haven't stepped into a fab lab since the early 1970's, I know they've come a long way though.

I was wondering if much theory was involved or actual lifetimes have begun to be exceeded.

I do remember cars from before, during, and after the arrival of silicon semiconductors, and then chips myself. It did not appear to be a very ideal progression, nor as logical as it could have been. That just makes it more difficult to arrive at very accurate predictions.

Did I mention that the web page is not open to the general public?

Not that I actually expected anybody to read the article to me ;)

Having lots of comments are more informative regardless, and naturally sometimes more helpful than others.

So everybody wins.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: